Encyclopedia of

Complexity and Systems Science Series

Editor-in-Chief: Robert A. Meyers

Cesar E. Silva
Alexandre I. Danilenko
Editors

Ergodic
Theory
A Volume in the Encyclopedia of
Complexity and Systems Science,
Second Edition
Encyclopedia of Complexity and
Systems Science Series

Editor-in-Chief
Robert A. Meyers, Ramtech Limited, Palm Desert, CA, USA
The Encyclopedia of Complexity and Systems Science series of topical vol-
umes provides an authoritative source for understanding and applying the
concepts of complexity theory, together with the tools and measures for
analyzing complex systems in all fields of science and engineering. Many
phenomena at all scales in science and engineering have the characteristics of
complex systems, and can be fully understood only through the transdisciplin-
ary perspectives, theories, and tools of self-organization, synergetics, dynam-
ical systems, turbulence, catastrophes, instabilities, nonlinearity, stochastic
processes, chaos, neural networks, cellular automata, adaptive systems,
genetic algorithms, and so on. Examples of near-term problems and major
unknowns that can be approached through complexity and systems science
include: The structure, history and future of the universe; the biological basis
of consciousness; the integration of genomics, proteomics and bioinformatics
as systems biology; human longevity limits; the limits of computing; sustain-
ability of human societies and life on earth; predictability, dynamics and extent
of earthquakes, hurricanes, tsunamis, and other natural disasters; the dynamics
of turbulent flows; lasers or fluids in physics, microprocessor design; macro-
molecular assembly in chemistry and biophysics; brain functions in cognitive
neuroscience; climate change; ecosystem management; traffic management;
and business cycles. All these seemingly diverse kinds of phenomena and
structure formation have a number of important features and underlying
structures in common. These deep structural similarities can be exploited to
transfer analytical methods and understanding from one field to another. This
unique work will extend the influence of complexity and system science to a
much wider audience than has been possible to date.
Cesar E. Silva • Alexandre I. Danilenko
Editors

Ergodic Theory
A Volume in the Encyclopedia of
Complexity and Systems Science,
Second Edition

With 50 Figures
Editors
Cesar E. Silva Alexandre I. Danilenko
Mathematics B.Verkin Institute for Low Temperature
Williams College Physics and Engineering of the NAS of
Williamstown, MA, USA Ukraine
Kharkiv, Ukraine

ISSN 2629-2327 ISSN 2629-2343 (electronic)

Encyclopedia of Complexity and Systems Science Series
ISBN 978-1-0716-2387-9 ISBN 978-1-0716-2388-6 (eBook)
https://doi.org/10.1007/978-1-0716-2388-6
© Springer Science+Business Media, LLC, part of Springer Nature 2023
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or
part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way,
and transmission or information storage and retrieval, electronic adaptation, computer software, or
by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Science+Business Media,
LLC, part of Springer Nature.
The registered company address is: 1 New York Plaza, New York, NY 10004, U.S.A.
Series Preface

The Encyclopedia of Complexity and System Science Series is a multivolume

authoritative source for understanding and applying the basic tenets of com-
plexity and systems theory as well as the tools and measures for analyzing
complex systems in science, engineering, and many areas of social, financial,
and business interactions. It is written for an audience of advanced university
undergraduate and graduate students, professors, and professionals in a wide
range of fields who must manage complexity on scales ranging from the
atomic and molecular to the societal and global.
Complex systems are systems that comprise many interacting parts with the
ability to generate a new quality of collective behavior through self-
organization, e.g., the spontaneous formation of temporal, spatial, or func-
tional structures. They are therefore adaptive as they evolve and may contain
self-driving feedback loops. Thus, complex systems are much more than a sum
of their parts. Complex systems are often characterized as having extreme
sensitivity to initial conditions as well as emergent behavior that are not readily
predictable or even completely deterministic. The conclusion is that a reduc-
tionist (bottom-up) approach is often an incomplete description of a phenom-
enon. This recognition that the collective behavior of the whole system cannot
be simply inferred from the understanding of the behavior of the individual
components has led to many new concepts and sophisticated mathematical and
modeling tools for application to many scientific, engineering, and societal
issues that can be adequately described only in terms of complexity and
complex systems.
Examples of Grand Scientific Challenges which can be approached through
complexity and systems science include: the structure, history, and future of
the universe; the biological basis of consciousness; the true complexity of the
genetic makeup and molecular functioning of humans (genetics and epige-
netics) and other life forms; human longevity limits; unification of the laws of
physics; the dynamics and extent of climate change and the effects of climate
change; extending the boundaries of and understanding the theoretical limits
of computing; sustainability of life on the earth; workings of the interior of the
earth; predictability, dynamics, and extent of earthquakes, tsunamis, and other
natural disasters; dynamics of turbulent flows and the motion of granular
materials; the structure of atoms as expressed in the Standard Model and the
formulation of the Standard Model and gravity into a Unified Theory; the
structure of water; control of global infectious diseases; and also evolution and

v
vi Series Preface

quantification of (ultimately) human cooperative behavior in politics, econom-

ics, business systems, and social interactions. In fact, most of these issues have
identified nonlinearities and are beginning to be addressed with nonlinear
techniques, e.g., human longevity limits, the Standard Model, climate change,
earthquake prediction, workings of the earth’s interior, natural disaster
prediction, etc.
The individual complex systems mathematical and modeling tools and
scientific and engineering applications that comprised the Encyclopedia of
Complexity and Systems Science are being completely updated and the major-
ity will be published as individual books edited by experts in each field who are
eminent university faculty members.
The topics are as follows:

Agent Based Modeling and Simulation

Applications of Physics and Mathematics to Social Science
Cellular Automata, Mathematical Basis of
Chaos and Complexity in Astrophysics
Climate Modeling, Global Warming, and Weather Prediction
Complex Networks and Graph Theory
Complexity and Nonlinearity in Autonomous Robotics
Complexity in Computational Chemistry
Complexity in Earthquakes, Tsunamis, and Volcanoes, and Forecasting and
Early Warning of Their Hazards
Computational and Theoretical Nanoscience
Control and Dynamical Systems
Data Mining and Knowledge Discovery
Ecological Complexity
Ergodic Theory
Finance and Econometrics
Fractals and Multifractals
Game Theory
Granular Computing
Intelligent Systems
Nonlinear Ordinary Differential Equations and Dynamical Systems
Nonlinear Partial Differential Equations
Percolation
Perturbation Theory
Probability and Statistics in Complex Systems
Quantum Information Science
Social Network Analysis
Soft Computing
Solitons
Statistical and Nonlinear Physics
Synergetics
System Dynamics
Systems Biology
Series Preface vii

Each entry in each of the Series books was selected and peer reviews organized
by one of our university-based book Editors with advice and consultation
provided by our eminent Board Members and the Editor-in-Chief.
This level of coordination assures that the reader can have a level of
confidence in the relevance and accuracy of the information far exceeding
than that generally found on the World Wide Web. Accessibility is also a
priority and for this reason each entry includes a glossary of important terms
and a concise definition of the subject. In addition, we are pleased that the
mathematical portions of our Encyclopedia have been selected by Math
Reviews for indexing in MathSciNet. Also, ACM, the world’s largest educa-
tional and scientific computing society, recognized our Computational Com-
plexity: Theory, Techniques, and Applications book, which contains content
taken exclusively from the Encyclopedia of Complexity and Systems Science,
with an award as one of the notable Computer Science publications. Clearly,
we have achieved prominence at a level beyond our expectations, but consis-
tent with the high quality of the content!

Palm Desert, CA, USA Robert A. Meyers

July 2023 Editor-in-Chief
Volume Preface

Ergodic theory is a branch of mathematics that studies statistical properties of

dynamical systems. By statistical properties we mean properties that appear as
time average for various functions computed along the orbits of the systems.
The system is modeled by a map or transformation defined on the set of states.
When this transformation is invertible, it defines an action of the group of
integers on the set of states. Ergodic theory focuses on groups acting on a
measure space in a measure-preserving (or measure-class preserving) way.
Sometimes one also considers semigroup actions, and some of entries in this
volume consider group and semigroup actions. The origins of ergodic theory
are in the nineteenth-century work of Boltzmann on the foundations of statis-
tical mechanics. Roughly speaking, Boltzmann hypothesized that for large
systems of interacting particles in equilibrium, the “time average” is equal to
the “space average.” He called this statement the ergodic hypothesis; this
strong form of the hypothesis was too restrictive and was later relaxed to
what lead to the formulation of the ergodic theorems. The word “ergodic” is an
amalgamation of the Greek words ergon (work) and hodos (path).
Ergodic theory lies at the intersection of many areas of mathematics,
including smooth dynamics, topological dynamics, statistical mechanics, dif-
ferential geometry, Lie theory, probability, harmonic analysis, number theory,
combinatorics, operator algebras, and group representations. Problems, tech-
niques, and results are related to many other areas of mathematics, and ergodic
theory has applications both within mathematics and to numerous other
branches of science.
This encyclopedia consists of independent surveys, written by distin-
guished researchers, on the fundamental directions and recent developments
in ergodic theory, including the asymptotic properties of measurable dynam-
ical systems, spectral theory, entropy, ergodic theorems, joinings, recurrence,
rigidity, nonsingular systems, and systems of probabilistic origin. Some sur-
veys are devoted to the interplay between ergodic theory and topological
dynamics, smooth dynamics, and operator algebras. Some entries enlighten
applications to combinatorics, number theory, Sarnak’s conjecture, translation
flows on translation surfaces, symbolic dynamics, pressure and equilibrium
states, fractal geometry, chaos, complexity, and classification of measurable
systems.

ix
x Volume Preface

This volume is for researchers and students interested in ergodic theory and
dynamics. We thank Bryna Kra who edited the first edition. We are grateful to
all the researchers who have contributed to this volume.

Kharkiv, Ukraine Alexandre I. Danilenko

Williamstown, USA Cesar E. Silva
July 2023 Editors
Contents

Introduction to Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Alexandre I. Danilenko and Cesar E. Silva
Ergodic Theory: Basic Examples and Constructions . . . . . . . . . . . 3
Matthew Nicol and Karl Petersen
Ergodicity and Mixing Properties .......................... 35
Terrence Adams and Anthony Quas
Ergodic Theory: Recurrence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Nikos Frantzikinakis and Randall McCutcheon
Ergodic Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Andrés del Junco
Spectral Theory of Dynamical Systems . . . . . . . . . . . . . . . . . . . . . . 109
Adam Kanigowski and Mariusz Lemańczyk
Joinings in Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Thierry de la Rue
Entropy in Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Jonathan L. F. King and Kyewon Koh Park
Isomorphism Theory in Ergodic Theory . . . . . . . . . . . . . . . . . . . . . 201
Christopher Hoffman
Dynamical Systems of Probabilistic Origin: Gaussian and
Poisson Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
Élise Janvresse, Emmanuel Roy, and Thierry De La Rue
Ergodic Theory: Nonsingular Transformations . . . . . . . . . . . . . . . 233
Alexandre I. Danilenko and Cesar E. Silva
Sarnak’s Conjecture from the Ergodic Theory Point of View . . . . 293
Joanna Kułaga-Przymus and Mariusz Lemańczyk
Smooth Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
Amie Wilkinson
Ergodic and Spectral Theory of Area-Preserving Flows
on Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Krzysztof Frączek and Corinna Ulcigrai

xi
xii Contents

Pressure and Equilibrium States in Ergodic Theory . . . . . . . . . . . 369

Jean-René Chazottes and Gerhard Keller
Parallels Between Topological Dynamics and Ergodic Theory . . . 389
Wen Huang, Song Shao, and Xiangdong Ye
Symbolic Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
Brian Marcus
Operator Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Guy Cohen and Michael Lin
Dynamical Systems and C-Algebras . . . . . . . . . . . . . . . . . . . . . . . . 491
T. Giordano and H.-C. Liao
The Complexity and the Structure and Classification of
Dynamical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
Matthew Foreman
Ergodic Theory: Interactions with Combinatorics and Number
Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
Tom Ward
Ergodic Theory on Homogeneous Spaces and Metric Number
Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597
Dmitry Kleinbock
Ergodic Theory: Rigidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615
Viorel Niţică
Chaos and Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633
Jérôme Buzzi
Ergodic Theory: Fractal Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 665
Jörg Schmeling
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683
About the Editor-in-Chief

Robert A. Meyers
President: RAMTECH Limited
Manager, Chemical Process Technology,
TRW Inc.
Postdoctoral Fellow: California Institute of Tech-
nology
Ph.D. Chemistry, University of California at Los
Angeles
B.A. Chemistry, California State University, San
Diego

Biography
Dr. Meyers was manager of Energy and Environ-
mental Projects at TRW (now Northrop
Grumman) in Redondo Beach, CA, and is now
president of RAMTECH Limited. He is
coinventor of the Gravimelt process for desulfur-
ization and demineralization of coal for air pollu-
tion and water pollution control and was manager
of the Department of Energy project leading to the
construction and successful operation of a first-of-
a-kind Gravimelt Process Integrated Test Plant.
Dr. Meyers is the inventor of and was project
manager for the DOE-sponsored Magnetohydro-
dynamics Seed Regeneration Project which has
resulted in the construction and successful opera-
tion of a pilot plant for production of potassium
formate, a chemical utilized for plasma electricity
generation and air pollution control. He also man-
aged TRW efforts in magnetohydrodynamics
electricity generating combustor and plasma
channel development.
Dr. Meyers managed the pilot-scale DoE pro-
ject for determining the hydrodynamics of syn-
thetic fuels. He is a coinventor of several
thermooxidative stable polymers which have

xiii
xiv About the Editor-in-Chief

achieved commercial success as the GE PEI,

Upjohn Polyimides, and Rhone-Poulenc
bismaleimide resins. He has also managed
projects for photochemistry, chemical lasers, flue
gas scrubbing, oil shale analysis and refining,
petroleum analysis and refining, global change
measurement from space satellites, analysis
and mitigation (carbon dioxide and ozone),
hydrometallurgical refining, soil and hazardous
waste remediation, novel polymers synthesis,
modeling of the economics of space transporta-
tion systems, space rigidizable structures, and
chemiluminescence-based devices.
He is a senior member of the American Insti-
tute of Chemical Engineers, member of the Amer-
ican Physical Society, and member of the
American Chemical Society and has served on
the UCLA Chemistry Department Advisory
Board. He was a member of the joint
USA-Russia working group on air pollution con-
trol and the EPA-sponsored Waste Reduction
Institute for Scientists and Engineers.
Dr. Meyers has more than 20 patents and
50 technical papers in the fields of photochemis-
try, pollution control, inorganic reactions, organic
reactions, luminescence phenomena, and poly-
mers. He has published in primary literature
journals including Science and the Journal of the
American Chemical Society, and is listed in Who’s
Who in America and Who’s Who in the World.
Dr. Meyers’ scientific achievements have been
reviewed in feature articles in the popular press in
publications such as The New York Times Science
Supplement and The Wall Street Journal as well as
more specialized publications such as Chemical
Engineering and Coal Age. A public service film
was produced by the Environmental Protection
Agency on Dr. Meyers’ chemical desulfurization
invention for air pollution control.
Dr. Meyers is the author or editor-in-chief of a
wide range of technical books including the
Handbook of Chemical Production Processes;
the Handbook of Synfuels Technology; Handbook
of Petroleum Refining Processes, now in fourth
edition; the Handbook of Petrochemical Produc-
tion Processes (McGraw-Hill), now in a second
edition; the Handbook of Energy Technology and
Economics, published by John Wiley & Sons;
About the Editor-in-Chief xv

Coal Structure, published by Academic Press; and

Coal Desulfurization as well as the Coal Hand-
book published by Marcel Dekker. He served as
chairman of the advisory board for A Guide to
Nuclear Power Technology: A Resource for Deci-
sion Making, published by John Wiley & Sons,
which won the Association of American Pub-
lishers Award as the best book in technology and
engineering. He also served as editor-in-chief of
three editions of the Elsevier Encyclopedia of
Physical Science and Technology. Most recently,
Dr. Meyers serves as editor-in-chief of the Ency-
clopedia of Analytical Chemistry as well as
Reviews in Cell Biology and Molecular Medicine
and a book series of the same name both published
by John Wiley & Sons. In addition, Dr. Meyers
currently serves as editor-in-chief of two Springer
Nature book series, Encyclopedia of Complexity
and Systems Science and Encyclopedia of Sus-
tainability Science and Technology.
About the Volume Editors

Cesar E. Silva is the Hagey Family Professor of

Mathematics at Williams College, where he has
taught since 1984, and was chair of the depart-
ment in 2008–2009 and 2010–2012. He received
his PhD from the University of Rochester under
the supervision of Dorothy Maharam, and his BS
from the Pontificia Universidad Católica del Perú.
He has held visiting positions at a number of
universities including Maryland and Toronto. In
2015, Silva was elected a fellow of the American
Mathematical Society. Silva’s research interests
are in ergodic theory and dynamical systems, in
particular the dynamics of nonsingular and infi-
nite measure-preserving transformations, rank-
one transformations, and p-adic dynamics. Silva
is the author of Invitation to Ergodic Theory and
Invitation to Real Analysis, both published by the
American Mathematical Society, and is co-editor
of two volumes of conference proceedings,
published in the AMS Contemporary Mathemat-
ics series. He was associate editor of AMS Notices
and is the author or co-author of over 50 research
articles in the general area of ergodic theory.

Dr. Alexandre I. Danilenko is currently a lead-

ing research fellow at B. Verkin Institute for Low
Temperature Physics and Engineering of the
National Academy of Sciences of Ukraine. He is
an expert in ergodic theory and dynamical sys-
tems. Dr. Danilenko conducts research in spectral
theory, orbit theory, entropy theory, theory of
joinings, nonsingular dynamics, topological
dynamics, etc. During his career, he has authored
or co-authored more than 60 research papers
published in internationally recognized mathe-
matical journals. Alexandre I. Danilenko received
xvii
xviii About the Volume Editors

his C.Sc. degree in Mathematics at Kharkiv State

University in 1991 under the supervision of V. Ya.
Golodets. More than 10 years he was employed as
assistant professor and associate professor at V. N.
Karazin Kharkiv National University (Ukraine).
He habilitated at Nicolaus Copernicus University
in Torun (Poland) in 2003. Since 2002 he has been
working at B. Verkin Institute for Low Tempera-
ture Physics and Engineering of the National
Academy of Sciences of Ukraine.
Contributors

Terrence Adams Department of Mathematics and Statistics, State University

of New York, Albany, NY, USA
Jérôme Buzzi C.N.R.S. and Université Paris-Sud, Orsay, France
Jean-René Chazottes Centre de Physique Théorique, CNRS/IP Paris,
Palaiseau, France
Guy Cohen School of Electrical Engineering, Ben-Gurion University, Beer-
Sheva, Israel
Alexandre I. Danilenko B.Verkin Institute for Low Temperature Physics and
Engineering of the NAS of Ukraine, Kharkiv, Ukraine
Thierry De La Rue Laboratoire de Mathématiques Raphaël Salem, CNRS –
Université de Rouen Normandie, Saint Étienne du Rouvray, France
Andrés del Junco Department of Mathematics, University of Toronto,
Toronto, ON, Canada
Matthew Foreman University of California at Irvine, Irvine, CA, USA
Krzysztof Frączek Faculty of Mathematics and Computer Science, Nicolaus
Copernicus University, Toruń, Poland
Nikos Frantzikinakis Department of Mathematics, University of Crete,
Heraklion, Greece
T. Giordano University of Ottawa, Ottawa, ON, Canada
Christopher Hoffman Department of Mathematics, University of Washing-
ton, Seattle, WA, USA
Wen Huang CAS Wu Wen-Tsun Key Laboratory of Mathematics, and
Department of Mathematics, University of Science and Technology of
China, Hefei, Anhui, China
Élise Janvresse Laboratoire Amiénois de Mathématique Fondamentale et
Appliquée, CNRS-UMR 7352, Université de Picardie Jules Verne, Amiens,
France

xix
xx Contributors

Adam Kanigowski Department of Mathematics, University of Maryland at

College Park, College Park, MD, USA
Gerhard Keller Department Mathematik, Universität Erlangen-Nürnberg,
Erlangen, Germany
Jonathan L. F. King University of Florida, Gainesville, FL, USA
Dmitry Kleinbock Department of Mathematics, Brandeis University, Wal-
tham, MA, USA
Joanna Kułaga-Przymus Faculty of Mathematics and Computer Science,
Nicolaus Copernicus University, Toruń, Poland
Mariusz Lemańczyk Faculty of Mathematics and Computer Science,
Nicolaus Copernicus University, Toruń, Poland
H.-C. Liao University of Ottawa, Ottawa, ON, Canada
Michael Lin Department of Mathematics, Ben-Gurion University, Beer-
Sheva, Israel
Brian Marcus Department of Mathematics, University of British Columbia,
Vancouver, BC, Canada
Randall McCutcheon Department of Mathematics, University of Memphis,
Memphis, TN, USA
Viorel Niţică West Chester University, West Chester, PA, USA
Institute of Mathematics, Bucharest, Romania
Matthew Nicol Department of Mathematics, University of Houston, Hous-
ton, TX, USA
Kyewon Koh Park Korea Institute for Advanced Study, Seoul, South Korea
Karl Petersen Department of Mathematics, University of North Carolina,
Chapel Hill, NC, USA
Anthony Quas Department of Mathematics and Statistics, University of
Victoria, Victoria, BC, Canada
Emmanuel Roy Laboratoire Analyse, Géométrie et Applications, Université
Paris 13 Institut Galilée, Villetaneuse, France
Jörg Schmeling Center for Mathematical Sciences, Lund University, Lund,
Sweden
Song Shao CAS Wu Wen-Tsun Key Laboratory of Mathematics, and Depart-
ment of Mathematics, University of Science and Technology of China, Hefei,
Anhui, China
Cesar E. Silva Department of Mathematics, Williams College,
Williamstown, MA, USA
Corinna Ulcigrai Institut für Mathematik, Universität Zürich, Zürich,
Switzerland
Contributors xxi

Tom Ward School of Mathematics, University of Leeds, Leeds, UK

Amie Wilkinson Northwestern University, Evanston, IL, USA

Xiangdong Ye CAS Wu Wen-Tsun Key Laboratory of Mathematics, and

Department of Mathematics, University of Science and Technology of
China, Hefei, Anhui, China
 Nonsingular Transformations”; ▶ “Ergodicity
Introduction to Ergodic Theory and Mixing Properties”; ▶ “Joinings in Ergodic
Theory”; ▶ “Pressure and Equilibrium States in
Alexandre I. Danilenko1 and Cesar E. Silva2 Ergodic Theory”; ▶ “Spectral Theory of Dynam-
1
B.Verkin Institute for Low Temperature Physics ical Systems”; and ▶ “Symbolic Dynamics.”
and Engineering of the NAS of Ukraine, Kharkiv, Some articles from the first edition are published
Ukraine in their original form: ▶ “Chaos and Ergodic The-
2
Department of Mathematics, Williams College, ory”; ▶ “Ergodic Theory: Fractal Geometry”;
Williamstown, MA, USA ▶ “Isomorphism Theory in Ergodic Theory”;
and ▶ “Smooth Ergodic Theory.” We are sad to
report that two authors of the original edition,
Ergodic theory traces its origins to questions in Andrés del Junco and Viorel Nitica, have passed
statistical mechanics about understanding motion away; their entries ▶ “Ergodic Theorems” and
in dynamical systems, with the motion of the ▶ “Ergodic Theory: Rigidity” are published as in
planets around the sun being one of the earliest the original edition.
examples of a dynamical system. Since that time We are also including new surveys in this edi-
ergodic theory has grown into a central area of tion. The entry ▶ “Sarnak’s Conjecture from the
mathematics and has interacted and contributed to Ergodic Theory Point of View” is devoted to
many areas of mathematics including harmonic impressive recent progress in studying the cele-
analysis, number theory, probability, operator brated Sarnak conjecture (formulated in 2010) on
algebras, geometry, and topology. Ergodic theory Möbius orthogonality, which states that each
is characterized by its use of measure-theoretic topological system of zero entropy is Möbius
and probabilistic tools to study dynamical sys- orthogonal. This problem became a bridge
tems. At the same time, topological, differentia- between analytic number theory and dynamics.
ble, and operator tools arise in a natural way and The survey covers the ergodic-theoretic aspects
this volume has articles that address topological of this area.
and differentiable dynamics and their connections Ergodic theory and topological dynamics are
with ergodic theory. For a more detailed descrip- sibling branches of the theory of dynamical sys-
tion of ergodic theory and its role in the modern tems. The terminology used in both branches is
mathematics, we refer to “Introduction to Ergodic very similar. Moreover, there are many analogous
Theory,” by Bryna Kra, from the first edition of theorems. However, despite the similarities
this collection. between the two theories, the proofs for “analo-
We owe a large debt to Bryna Kra, who was the gous” results are quite different and are not
editor of the first edition and organized the topics directly deduced from one another. This is the
and the original articles. This edition would not subject of the entry ▶ “Parallels Between Topo-
have existed if it were not for her original vision. logical Dynamics and Ergodic Theory.”
We have built on articles from the first edition, Many interesting and fundamental families
many of which have been updated: ▶ “Entropy in of examples in ergodic theory are of probabilis-
Ergodic Theory”; ▶ “Ergodic Theory: Basic tic nature. The systems arising from Gaussian
Examples and Constructions”; ▶ “Ergodic The- processes and the systems constructed from
ory: Interactions with Combinatorics and Number Poisson point processes are well studied in ergo-
Theory”; ▶ “Ergodic Theory: Recurrence”; dic theory. These two families are of different
▶ “Ergodic Theory on Homogeneous Spaces nature, and each one is treated with specific
and Metric Number Theory”; ▶ “Ergodic Theory: tools. They, however, share surprisingly many

© Springer Science+Business Media, LLC, part of Springer Nature 2023 1

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_182
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2017
https://doi.org/10.1007/978-3-642-27737-5_182-2
2 Introduction to Ergodic Theory

common features. This is discussed in the entry The entry ▶ “Operator Ergodic Theory” focuses
▶ “Dynamical Systems of Probabilistic Origin: on the asymptotic behavior of powers of a power-
Gaussian and Poisson Systems.” bounded operator in a Banach space. The asymp-
A classification of ergodic transformations via totic behavior is studied in different operator topol-
certain systems of parameters, such as spectrum ogies and in various modes of convergence. That is
and entropy, is the central classic problem of ergo- closely related to the classic topic of ergodic theo-
dic theory. ▶ “The Complexity and the Structure rems (see ▶ “Ergodic Theorems”).
and Classification of Dynamical Systems,” based Some aspects of the interplay between dynam-
mostly on examples from ergodic theory, explains ical systems and operator algebras are discussed
the methods for studying the complexity of struc- in ▶ “Dynamical Systems and C-Algebras.”
ture, classification, and anti-classification of the While this volume contains articles that
systems via descriptive set theory. address recent areas of ergodic theory, the sub-
Ergodic, spectral, and joining properties of ject continues to develop at a fast rate, and due to
translation flows on translation surfaces (and space and time constraints, we have not been
their Poincare maps, interval exchange transfor- able to cover every possible new development.
mations) and smooth area-preserving locally We want to thank all the authors who have
Hamiltonian flows are discussed in ▶ “Ergodic contributed to this volume, as well as the staff at
and Spectral Theory of Area-Preserving Flows Springer for their help in completing this edition.
on Surfaces.”
 of disjoint (mod m, i.e., up to sets of measure 0)
Ergodic Theory: Basic Examples measurable sets {P1, . . ., Pn} such that X ¼
and Constructions [ Pi (mod m). The entropy of P with respect to
m is HðP Þ ¼
i mðPi Þ ln m(Pi) (other bases
Matthew Nicol1 and Karl Petersen2 are sometimes used for the logarithm).
1
Department of Mathematics, University of • The metric (or measure-theoretic) entropy of
Houston, Houston, TX, USA T with respect to P is hm ðT, P Þ ¼
2
Department of Mathematics, University of North lim n!1 1n H P _ . . . _ T
nþ1 ðP Þ , where P _
Carolina, Chapel Hill, NC, USA
. . . _ T
nþ1 ðP Þ is the partition of X into sets of
points with the same coding with respect to P
under T i, i ¼ 0, . . ., n
1, that is, x, y are in the
Article Outline
same set of the partition P _ . . . _ T –nþ1(P ) if
and only if T i(x) and T i( y) lie in the same set of
Glossary
the partition P for i ¼ 0, . . ., n
1.
Definition of the Subject and its Importance
• The metric entropy hm(T ) of (X, ℬ, m, T ) is the
Introduction
supremum of hm(T, P ) over all finite measur-
Examples
able partitions P .
Constructions
• If T is a continuous transformation of a com-
Future Directions
pact metric space X, then the topological
Bibliography
entropy of T is the supremum of the metric
entropies hm(T ), where the supremum is taken
Glossary over all T-invariant Borel probability
measures.
• A transformation T of a measure space (X, ℬ, • A system (X, ℬ, m, T ) is loosely Bernoulli if it
m) is measure-preserving if m (T
1A) ¼ m (A) is isomorphic to the first-return system to a
for all measurable A  ℬ. subset of positive measure of an irrational rota-
• A measure-preserving transformation (X, ℬ, m, tion or a (positive or infinite entropy) Bernoulli
T ) is ergodic if T
1(A) ¼ A (mod m) implies system.
m (A) ¼ 0 or m (Ac) ¼ 0 for each measurable set • Two systems are spectrally isomorphic if the
A  ℬ. unitary operators that they induce on their L2
• A measure-preserving transformation (X, ℬ, m, spaces are unitarily equivalent.
T ) of a probability space is weak-mixing if • A smooth dynamical system consists of a dif-

i
lim n!1 1n n
1
i¼0 m T A \ B
mðAÞmðBÞ ¼ 0 for ferentiable manifold M and a differentiable
all measurable sets A, B  ℬ. map f : M ! M. The degree of differentiability
• A measure-preserving transformation (X, ℬ, m, may be specified.
T ) of a probability space is strong-mixing if • Two submanifolds S1, S2 of a manifold M
limn!1 m(T
nA \ B) ¼ m(A)m(B) for all mea- intersect transversely at p  M if Tp(S1) þ
surable sets A, B  ℬ. Tp(S2) ¼ Tp(M).
• A continuous transformation T of a compact • An (ϵ-) small Cr perturbation of a Cr map f of a
metric space X is uniquely ergodic if there is manifold M is a map g such that dCr ðf , gÞ < ϵ,
only one T-invariant Borel probability measure i.e., the distance between f and g, is less than ϵ
on X. in the Cr topology.
• Suppose (X, ℬ, m) is a probability space. • A map T of an interval I ¼ [a, b] is piecewise
A finite partition P of X is a finite collection smooth (Ck for k  1) if there is a finite set of

© Springer Science+Business Media, LLC, part of Springer Nature 2023 3

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_177
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_177-2
4 Ergodic Theory: Basic Examples and Constructions

points a ¼ x1 < x2 < . . . < xn ¼ b such Introduction

that T |(xi, xiþ1) is Ck for each i. The degree
of differentiability may be specified. In this chapter, we collect a brief list of some
• A measure m on a measure space (X, ℬ) is important examples of measure-preserving
absolutely continuous with respect to a mea- dynamical systems, which we denote typically
sure n on (X, ℬ) if n(A) ¼ 0 implies m(A) ¼ 0 for by (X, ℬ, m, T ) or (T, X, ℬ, m) or slight variations.
all measurable A  ℬ. These examples have played a formative role in
• A Borel measure m on a Riemannian manifold the development of dynamical systems theory,
M is absolutely continuous if it is absolutely either because they occur often in applications in
continuous with respect to the Riemannian one guise or another or because they have been
volume on M. useful simple models to understand certain fea-
• A measure m on a measure space (X, ℬ) is tures of dynamical systems. There is a fundamen-
equivalent to a measure n on (X, ℬ) if m is tal difference in the dynamical properties of those
absolutely continuous with respect to n and n systems which display hyperbolicity: Roughly
is absolutely continuous with respect to m. speaking, there is some exponential divergence
of nearby orbits under iteration of the transforma-
tion. In differentiable systems, this is associated
Definition of the Subject and its with the derivative of the transformation
Importance possessing eigenvalues of modulus greater than
one on a “dynamically significant” subset of phase
Measure-preserving systems are a common model space. Hyperbolicity leads to complex dynamical
of processes which evolve in time and for which behavior such as positive topological entropy,
the rules governing the time evolution do not exponential divergence of nearby orbits (“sensi-
change. For example, in Newtonian mechanics tivity to initial conditions”) often coexisting with
the planets in a solar system undergo motion a dense set of periodic orbits. If f, c are suffi-
according to Newton’s laws of motion: The ciently regular functions on the phase space X of a
planets move, but the underlying rule governing hyperbolic measure-preserving transformation (T,
the planets’ motion remains constant. The model X, m), then typically we have fast decay of corre-
adopted here is to consider the time-evolution as a lations in the sense that
transformation (either a map in discrete time or a
flow in continuous time) on a probability space or
fðT n xÞcðxÞdm
fdm cdm  CaðnÞ
more generally a measure space. This is the setting X
of the subject called ergodic theory. Applications
of this point of view include the areas of statistical where a(n) ! 0. If a(n) ! 0 at an exponential rate,
physics, classical mechanics, number theory, pop- we say that the system has exponential decay of
ulation dynamics, statistics, information theory, correlations. A theme in dynamical systems is that
and economics. The purpose of this chapter is to the time series formed by sufficiently regular
present a flavor of the diverse range of examples observations on systems with some degree of
of measure-preserving transformations which hyperbolicity often behave statistically like inde-
have played a role in the development and appli- pendent identically distributed random variables.
cation of ergodic theory and smooth dynamical At this point, it is appropriate to point out two
systems theory. We also present common con- pervasive differences between the usual probabi-
structions involving measure-preserving systems. listic setting of a stationary stochastic process
Such constructions may be considered a way of {Xn} and the (smooth) dynamical systems setting
putting “building-block” dynamical systems of a time series of observations on a measure-
together to construct examples or decomposing a preserving system {f ∘ T n}. The most crucial is
complicated system into simple “building-blocks” that for deterministic dynamical systems the time
to understand it better.
Ergodic Theory: Basic Examples and Constructions 5

series is usually not an independent process, measure-preserving systems (T, X, ℬ, m) and

which is a common assumption in the strictly (S, Y, C , n) are isomorphic if (possibly after delet-
probabilistic setting. Even if some weak-mixing ing sets of measure 0 from X and Y ) there is a one-
is assumed in the probabilistic setting, it is usually to-one onto measurable map f : X ! Y with
a mixing condition on the s-algebras F n ¼ s(X1, measurable inverse f
1 such that f ∘ T ¼ S ∘ f
. . ., Xn) generated by successive random vari- m a.e. and m (f
1(A)) ¼ n (A) for all A  C. If X, Y
ables, a condition which is not natural (and usu- are compact topological spaces, we say that T is
ally very difficult to check) for dynamical topologically conjugate to S if there exists a
systems. Mixing conditions on dynamical sys- homeomorphism f : X ! Y such that f ∘ T ¼ S
tems are given more naturally in terms of the ∘ f. In this case, we call f a conjugacy. If f is C r
mixing of the sets of the s-algebra ℬ of the prob- for some r  1, we will call f a C r-conjugacy and
ability space (X, ℬ, m) under the action of T and similarly for other degrees of regularity.
not by mixing properties of the s-algebras gener- We will consider X ¼ [0, 1] (mod 1) as a
ated by the random variables {f ∘ T n}. The other representation of the unit circle S1 ¼ {z  ℂ :
difference is that in the probabilistic setting, |z| ¼1} (under the map x ! e2πix) and similarly
although {Xn} satisfy moment conditions, usually represent the k-dimensional torus T k ¼ S1  . . . 
no regularity properties, such as the Hölder prop- S1 (k-times). If the s-algebra is clear from the
erty or smoothness, are assumed. In contrast in context, we will write (T, X, m) instead of (T, X,
dynamical systems theory, the transformation T is ℬ, m) when denoting a measure-preserving
often a smooth or piecewise smooth transforma- system.
tion of a Riemannian manifold X and the observa-
tion f : X ! ℝ is often assumed continuous or
Hölder. The regularity of the observation f turns Examples
out to play a crucial role in proving properties
such as rates of decay of correlation, central limit Rigid Rotation of a Compact Group
theorems, and so on. If G is a compact group equipped with Haar mea-
An example of a hyperbolic transformation is sure and a  G, then the transformation T(x) ¼ ax
an expanding map of the unit interval T(x) ¼ (2x) preserves Haar measure and is called a rigid rota-
(where (x) is x modulo the integers). Here the tion of G. If G is abelian and the transformation is
derivative has modulus 2 at all points in phase ergodic (in this setting, transitivity implies ergo-
space. This map preserves Lebesgue measure, has dicity), then the transformation is uniquely
positive topological entropy, Lebesgue at almost ergodic. Such systems always have zero topolog-
every point x has a dense orbit, and periodic points ical entropy.
for the map are dense in [0, 1). The simplest example of such a system is a
Nonhyperbolic systems are of course also an circle rotation. Take X ¼ [0, 1] (mod 1), with
important class of examples, and in contrast to
hyperbolic systems they tend to model systems T ðxÞ ¼ ðx þ aÞ where a  ℝ:
of “low complexity,” for example, systems
displaying quasiperiodic behavior. The simplest Then T preserves Lebesgue (Haar) measure
nontrivial example is perhaps an irrational rota- and is ergodic (in fact uniquely ergodic) if and
tion of the unit interval [0, 1] given by a map only if α is irrational. Similarly, the map
T(x) ¼ (x þ α), α  ℝ\ℚ. T preserves Lebesgue
measure, and every point has a dense orbit (there T ðx1 , . . . , xk Þ ¼ ðx1 þ a1 , . . . , xk þ ak Þ,
are no periodic orbits), yet the topological entropy where a1 , . . . , ak  ℝ,
is zero and nearby points stay the same distance
from each other under iteration under T. preserves k-dimensional Lebesgue (Haar) mea-
There is a natural notion of equivalence for sure and is ergodic (uniquely ergodic) if and
measure-preserving systems. We say that
6 Ergodic Theory: Basic Examples and Constructions

only if there are no integers m1, . . . , mk, not all 0, which the dynamics is topologically conjugate to
which satisfy m1 α1 þ . . . þ mk αk  ℤ. an adding machine (de Melo and van Strien 1993).

Adding Machines Interval Exchange Maps

Let {ki}i  ℕ be a sequence of integers with ki  2. A map T : [0, 1] ! [0, 1] is an interval exchange
Equip each cyclic group ℤki with the discrete transformation if it is defined in the following
topology and form the product space ¼ way. Suppose that π is a permutation of {1, . . . ,
1
Pi¼1 ℤki equipped with the product topology. An n} and li > 0, i ¼ 1, . . . , n, is a sequence of
adding machine corresponding to the sequence subintervals of I (open or closed) with i li ¼ 1.
{ki}i  ℕ is the topological space ¼ P1 i¼1 ℤki Define ti by li ¼ ti
ti
1 with t0 ¼ 0. Suppose also
together with the map that s is an n-vector with entries 1. T is defined
by sending the interval ti
1  x < ti
1 of length li
s: ! to the interval

defined by. lpðjÞ  x < lpðjÞ

s (k1 – 1, k2 – 1, . . .) ¼ (0, 0, . . .) if each entry pðjÞ<pðiÞ pðjÞ>pðiÞ

in the ℤki component is ki
1,

while. with orientation preserved if the i’th entry of s is
s (k1
1, k2 – 1, . . . , kn
1, x1, x2, . . .)¼ (0, 0, +1 and orientation reversed if the i’th entry of s is
. . . ( (n times) . . . , 0, x1 þ 1, x2, x3, . . .)
1. Thus on each interval li, T has the form
when x1 6¼ knþ1
1. T(x) ¼ six þ ai, where si is 1. If si ¼ 1 for
The map s may be thought of as “add one and each i, the transformation is called orientation
carry” and also as mapping each point to its suc- preserving.
cessor in a certain order. See section “Adic Trans- The transformation T has finitely many discon-
formations” for generalizations. If each ki ¼ 2, tinuities (at the end points of each li), and modulo
then the system is called the dyadic (or von this set of discontinuities is smooth. T is also
Neumann-Kakutani) adding machine or 2- invertible (neglecting the finite set of discontinu-
odometer. Adding machines give examples of ities) and preserves Lebesgue measure. These
naturally occurring minimal systems of low-orbit maps have zero topological entropy and arise
complexity in the sense that the topological naturally in studies of polygonal billiards and
entropy of an adding machine is zero. In fact, if more generally area-preserving flows. There are
f is a continuous map of an interval with zero examples of minimal but nonergodic interval
topological entropy and S is a closed, topologi- exchange maps (Keane 1977; Mañé 1987a).
cally transitive invariant set without periodic
orbits, then the restriction of f to S is topologically Full Shifts and Shifts of Finite Type
conjugate to the dyadic adding machine Given a finite set (or alphabet) A ¼ {0, . . . , d
1},
(Hasselblatt and Katok 2003a, Theorem 11.3.13). take X ¼ Ω+(A) ¼ Aℕ (or X ¼ Aℤ) the sets of one-
We say a nonempty set Λ is an attractor for a sided (two-sided) sequences, respectively, with
map T if there is an open set U containing Λ such entries from A. For example sequences in Aℕ,
that Λ ¼ \n  0T n(U ) (other definitions are found have the form x ¼ x0x1 . . . xn . . .. A cylinder set
in the literature). The dyadic adding machine is C yn1 , . . . , ynk , yni  A, of length k is a subset of
topologically conjugate to the Feigenbaum attrac- X defined by fixing k entries, for example,
tor at the limit point of period-doubling bifurca-
tions (see section “Unimodal Maps”). C yn1 , . . . , ynk ¼ x : xn1 ¼ yn1 , . . . , xnk ¼ ynk :
Furthermore, attractors for continuous unimodal
maps of the interval are either periodic orbits, We define the set Ak to consist of all cylinders
transitive cycles of intervals, or Cantor sets on C(y1, . . . , yk) determined by fixing the first
Ergodic Theory: Basic Examples and Constructions 7

k entries, i.e., an element of Ak is specified by This construction can be generalized to model

fixing the first k entries of a sequence x0 . . . xk one-step finite-state Markov stochastic processes as
by requiring xi ¼ yi, i ¼ 0, . . . , k. dynamical systems. Again let A ¼ {0, . . . , d
1},
Let p ¼ ( p0, . . . , pd–1) be a probability vector: and let p ¼ ( p0, . . . , pd–1) be a probability vector.
All pi  0 and d
1 i¼0 pi ¼ 1. For any cylinder B ¼ Let P be a d  d stochastic matrix with rows and
C(b1, . . . , bk)  Ak, define columns indexed by A. This means that all entries of
P are nonnegative, and the sum of the entries in
gk ðBÞ ¼ pb1 . . . pbk : ð1Þ each row is 1. We regard P as giving the transition
probabilities between pairs of elements of A. Now
It can be shown that these functions on Ak we define for any cylinder B ¼ C(b1, . . ., bk)  Ak
extend to a shift-invariant measure mp on Aℕ
(or Aℤ) called product measure. (See the article mp,P ðBÞ ¼ pb1 Pb1 b2 Pb2 b3 . . . Pbk
1 bk : ð2Þ
on Measure-Preserving Systems.) The space Aℕ
or Aℤ may be given a metric by defining It can be shown that mp,P extends to a measure
on the Borel s-algebra of Ω+(A), and its comple-
1 if x0 6¼ y0 ; tion. (See the article on Measure-Preserving Sys-
d ðx, yÞ ¼ 1 tems.) The resulting stochastic process is a
if xn 6¼ yn and xi ¼ yi for jij < n:
2jnj (one-step, finite-state) Markov process. If p and
P also satisfy
The shift s (.x0x1 . . . xn . . .) ¼ .x1x2 . . . xn . . . is
ergodic with respect to mp. The measure- pP ¼ p, ð3Þ
preserving system (Ω, ℬ, m, s) (with ℬ the
s-algebra of Borel subsets of Ω(A), or its comple- then the Markov process is stationary. In this case,
tion), is denoted by ℬ( p) and is called the we call the (one- or two-sided) measure-
Bernoulli shift determined by p. This system preserving system the Markov shift determined
models an infinite number of independent repeti- by p and P.
tions of an experiment with finitely many out- Aperiodic and irreducible Markov chains
comes, the i’th of which has probability pi on (those for which a power of the transition matrix
each trial. P has all entries positive) are strongly mixing and,
These systems are mixing of all orders (i.e., sn is in fact, are isomorphic to Bernoulli shifts (usually
mixing for all n  1) and have countable Lebesgue by means of a complicated measure-preserving
spectrum (hence are all spectrally isomorphic). recoding).
Kolmogorov and Sinai showed that two of them More generally, we say a set Λ  Aℤ is a
cannot be isomorphic unless they have the same subshift if it is compact and invariant under s.
entropy; Ornstein (Ornstein 1970) showed the con- A subshift Λ is said to be of finite type (SFT) if
verse. ℬ(1/2, 1/2) is isomorphic to the Lebesgue- there exists a d  d matrix M ¼ (aij) such that all
measure-preserving transformation x ! 2x mod entries are 0 or 1 and x  Λ if and only if
1 on [0, 1]; similarly, ℬ(1/3, 1/3, 1/3) is isomorphic axi xi þ1 ¼ 1 for all i  ℤ. Shifts of finite type are
to x ! 3x mod 1. Furstenberg asked whether the also called topological Markov chains. There are
only nonatomic measure invariant for both x ! 2x many invariant measures for a nontrivial shift of
mod 1 and x ! 3x mod 1 on [0, 1] is Lebesgue finite type. For example, the orbit of each periodic
measure. Lyons (1988) showed that if one of the point is the support of an invariant measure. An
actions is K, then the measure must be Lebesgue, important role in the theory, derived from motiva-
and Rudolph (Rudolph 1990a) showed the same tions of statistical mechanics, is played by equi-
thing under the weaker hypothesis that one of the librium measures (or equilibrium states) for
actions has positive entropy. For further work on continuous functions f : Λ ! ℝ, i.e., those mea-
this question, see (Host 1995; Parry 1996). sures m which maximize {hs(m) þ Λfdm} over
all shift-invariant probability measures, where
8 Ergodic Theory: Basic Examples and Constructions

hs(m) is the measure-theoretic entropy of s with Or, repeatedly apply the substitution 0 ! 01,
respect to m. The study of full shifts or shifts of 1 ! 10.
finite type has played a prominent role in the
development of the hyperbolic theory of dynam- 0
ical systems as physical systems with “chaotic” 01
dynamics “typically” possess an invariant set with
0 1 10
induced dynamics topologically conjugate to a
0 1 10 0110
shift of finite type (see the discussion by Smale
in (Smale 1980, p. 147)). Dynamical systems in ...
which there are transverse homoclinic connec-
tions are a common example (Guckenheimer and The n’th entry is the sum, mod 2, of the digits
Holmes 1990, Theorem 5.3.5). Furthermore, in in the dyadic expansion of n. Using Keane’s block
certain settings positive metric entropy implies multiplication (Keane 1968) according to which if
the existence of shifts of finite type. One result B is a block, B  0 ¼ B, B  1 ¼ B0 , and B  (o1
along these lines is a theorem of Katok (Katok . . . on) ¼ (B  o1) . . . (B  on), we may also
1980). Let htop( f ) denote the topological entropy obtain this sequence as
of a map f and hm( f ) denote metric entropy with
respect to an invariant measure m. 0  01  01  01  . . . :

Theorem 4.1 (Katok) Suppose T : M ! M The orbit closure of this sequence is uniquely
is a C1þϵ diffeomorphism of a closed manifold ergodic (there is a unique shift-invariant Borel
and m is an invariant measure with positive metric probability measure, which is then necessarily
entropy (i.e., hm (T ) > 0). Then for any 0 < ϵ < hm ergodic). It is isomorphic to a skew product (see
(T ), there exists an invariant set Λ topologically section “Skew Products”) over the von Neumann-
conjugate to a transitive shift of finite type with Kakutani adding machine, or odometer (see sec-
htop(T|Λ) > hm(T )
ϵ. tion “Adding Machines”). Generalized Morse
systems, that is, orbit closures of sequences like
More Examples of Subshifts 0  001  001  001  . . . , are also isomorphic
We consider some further examples of systems that to skew products over compact group rotations.
are given by the shift transformation on a subset of
the set of (usually doubly infinite) sequences on a Chacon System
finite alphabet, usually {0, 1}. Associated with This is the orbit closure of the sequence generated
each subshift is its language, the set of all finite by the substitution 0 ! 0010, 1 ! 1. It is uniquely
blocks seen in all sequences in the subshift. These ergodic and is one of the first systems shown to be
languages are extractive (or factorial) (every sub- weakly mixing but not strongly mixing. It is prime
word of a word in the language is also in the (has no nontrivial factors) (del Junco 1978) and in
language) and insertive (or extendable) (every fact has minimal self joinings (del Junco et al.
word in the language extends on both sides to 1980). It also has a nice description by means of
longer words in the language). In fact, these two cutting up the unit interval and stacking the
properties characterize the languages (subsets of pieces, using spacers (see section “Cutting and
the set of finite-length words on an alphabet) asso- Stacking”). This system has singular spectrum. It
ciated with subshifts. is not known whether or not its Cartesian square is
loosely Bernoulli.
Prouhet-Thue-Morse
An interesting (and often rediscovered) element of Sturmian Systems
Take the orbit closure of the sequence on ¼
f0, 1gℤþ is produced as follows. Start with 0, and
at each stage write down the opposite (00 ¼ 1, w[1
α,1)(nα), where α is irrational. This is a
uniquely ergodic system that is isomorphic to
10 ¼ 0) or mirror image of what is available so far.
rotation by α on the unit interval. These systems
Ergodic Theory: Basic Examples and Constructions 9

have minimal complexity in the sense that the Coded Systems

number of n-blocks grows as slowly as possible These are systems all of whose blocks are concat-
(n þ 1) (Coven and Hedlund 1973). enations of some (finite or infinite) list of blocks.
These are the same as the closures of increasing
Toeplitz Systems sequences of SFT’s (Krieger 2000). Alternatively,
Fill in the blanks alternately with 0’s and 1’s. If they are the closures of the images under finite
done correctly, you get a uniquely ergodic system edge-labelings of irreducible countable-state
which is isomorphic to a rotation on a compact topological Markov chains. They need not be
abelian group (Downarowicz 2005). context-free. Square-free languages are not
coded and, in fact, do not contain any coded
Sofic Systems systems of positive entropy. See (Blanchard and
These are images of SFT’s under continuous fac- Hansel 1986a, b, 1991).
tor maps (finite codes, or block maps). They cor-
respond to regular languages – languages whose Smooth Expanding Interval Maps
words are recognizable by finite automata. These Take X ¼ [0, 1] (mod 1), m  ℕ, m > 1 and
are the same as the languages defined by regular
expressions – finite expressions built up from 0 T ðxÞ ¼ ðmxÞ

(empty set), ϵ (empty word), + (union of two

languages), (all concatenations of words from Then T preserves Lebesgue measure m (recall
two languages), and (all finite concatenations that T preserves m if m(T
1A) ¼ m(A) for all
of elements). They also have the characteristic A  ℬ). Furthermore, it can be shown that T is
property that the family of all follower sets of all ergodic.
blocks seen in the system is a finite family; simi- This simple map exemplifies many of the char-
larly for predecessor sets. These are also gener- acteristics of systems with some degree of hyper-
ated by phase-structure grammars which are bolicity. It is isomorphic to a Bernoulli shift. The
linear, in the sense that every production is either map has positive topological entropy and exponen-
of the form A ! Bw or A ! w, where A and B are tial divergence of nearby orbits, and Hölder func-
variables and w is a string of terminals (symbols in tions have exponential decay of correlations and
the alphabet of the language). satisfy the central limit theorem and other strong
[A phase-structure grammar consists of alpha- statistical properties (Boyarsky and Góra 1997a).
bets V of variables and A of terminals, a set of If m ¼ 2, the system is isomorphic to a model
productions, which is a finite set of pairs of words of tossing a fair coin, which is a common example
(α, w), usually written α ! w, of words on V [ A, of randomness. To see this, let P ¼ {P0 ¼ [0, 1/2],
and a start symbol S. The associated language P1 ¼ [1/2, 1]} be a partition of [0, 1] into two
consists of all words on the alphabet A of termi- subintervals. We code the orbit under T of any
nals which can be made by starting with S and point x  [0, 1] by 0’s and 1’s by letting xk ¼ i
applying a finite sequence of productions]. if T kx  Pi, k ¼ 0, 1, 2, . . . . The map
Sofic systems typically support many invariant f : X ! {0, 1}ℕ which associates a point x to its
measures (for example, they have many periodic itinerary in this way is a measure-preserving map
points), but topologically transitive ones (those from (X, m) to {0, 1}ℕ equipped with the Bernoulli
with a dense orbit) have a unique measure of measure from p0 ¼ p1 ¼ 12. The map f satisfies f
maximal entropy. See (Lind and Marcus 1995). ∘ T ¼ s ∘ f, m a.e. and is invertible a.e., hence is an
isomorphism. Furthermore, reading the binary
Context-Free Systems expansion of x is equivalent to following the orbit
These are generated by phase-structure grammars of x under T and noting which element of the
in which all productions are of the form A ! w, partition P is entered at each time. Borel’s theorem
where A is a variable and w is a string of variables on normal numbers (base m) may be seen as a
and terminals. special case of the Birkhofff Ergodic Theorem in
this setting.
10 Ergodic Theory: Basic Examples and Constructions

1
Piecewise C2 Expanding Maps C  dm
dm
 C: Furthermore, T is ergodic with respect
The main statistical features of the examples in to m and displays the same statistical properties
section “Smooth Expanding Interval Maps” gen- listed above for the C2 expanding maps
eralize to a broader class of expanding maps of the (Boyarsky and Góra 1997a) (See the “Folklore
interval. For example: Theorem” in the article on Measure-Preserving
Let X ¼ [0, 1] and let P ¼ {I1, . . . , In} (n  2) Systems).
be a partition of X into intervals (closed, half-
open, or open) such that I i \ I j ¼ 0 if i 6¼ j. Let More Interval Maps

I oi denote the interior of Ii. Suppose T : X ! X
satisfies: Continued Fraction Map
This is the map T : [0, 1] ! [0, 1] given by Tx ¼
(a) For each i ¼ 1, . . . , n, T|Ii has a C2 extension 1/x mod 1, and it corresponds to the shift [0; a1, a2,
to the closure I i of Ii and |T 0(x)|  α > 1 for all . . .] ! [0; a2, a3, . . .] on the continued fraction
x  I oi . expansions of points in the unit interval (a map on
(b) T I j ¼ [i  Pj I i Lebesgue a.e. for some non- ℕℕ). It preserves a unique finite measure equiva-
empty subset Pj  {1, . . ., n}. lent to Lebesgue measure, the Gauss measure
(c) For each Ij, there exists nj such that T nj I j ¼ dx/(log 2)(1 þ x). It is Bernoulli with entropy
½0, 1 Lebesgue a.e. π2/6 log 2 (in fact the natural partition into inter-
vals is a weak Bernoulli generator; for definition
Then T has an invariant measure m which is and details, see (Phillips and Varadhan 1975)). By
absolutely continuous with respect to Lebesgue using the Ergodic Theorem, Khintchine and Lévy
measure m, and there exists C > 0 such that showed that

1 log k= log 2
1
ða1 . . . an Þ1=n ! 1þ 2
a:e: as n ! 1,
k¼1 k þ 2k
pn 1 p2
if ½0; a1 , . . . , an ¼ , then log qn ! a:e:;
qn n 12 log 2
1 p ðxÞ p2
log x
n ! a:e:;
n qn ð x Þ 6 log 2

and if m is Lebesgue measure (or any equivalent provide the intermediate convergents (best one-
measure) and m is Gauss measure, then for each sided) as well as the continued fraction (best two-
interval I, m(T –nI) ! m(I ), in fact exponentially sided) rational approximations to irrational num-
fast, with a best constant 0.30366 . . . See bers. See (Lagarias 1991, 1992).
(Billingsley 1978a; Mayer 1991).
F-Expansions
The Farey Map Generalizing the continued fraction map, let f :
This is the map U : [0, 1] ! [0, 1] given by Ux ¼ [0, 1] ! [0, 1] and let {In} be a finite or infinite
x/(1 – x) if 0  x  1/2, Ux ¼ (1
x)/x if partition of [0, 1] into subintervals. We study the
1/2  x  1. It is ergodic for the s-finite infinite map f by coding itineraries with respect to the
measure dx/x (Rényi/Parry). It is also ergodic for partition {In}. For many examples, absolutely
the Minkowski measure d, which is a measure of continuous (with respect to Lebesgue measure)
maximal entropy. This map corresponds to the invariant measures can be found and their
shift on the Farey tree of rational numbers which
Ergodic Theory: Basic Examples and Constructions 11

dynamical properties determined. See (Schweiger A β-shift is a shift of finite type if and only if
1995a). the β-expansion of 1 is finite. It is sofic if and only
if the expansion of 1 is eventually periodic. If β is
β-Shifts a Pisot-Vijayaragavhan number (algebraic integer
This is the special case of f-expansions when all of whose conjugates have modulus less than 1),
f(x) ¼ βx mod 1 for some fixed β > 1. This map then the β-shift is sofic. If the β-shift is sofic, then
of the interval is called the β-transformation. With β is a Perron number (algebraic integer of maxi-
a proper choice of partition, it is represented by mum modulus among its conjugates).
the shift on a certain subshift of the set of all
sequences on the alphabet D ¼ {0, 1, . . . , bβc}., Theorem 4.2 Parry (1966) Every strongly tran-
called the β-shift. A point x is expanded as an sitive ( for every nonempty open set U, [n > 0T nU ¼
infinite series in negative powers of β with coeffi- X) piecewise monotonic map on [0, 1] is topolog-
cients from this set; dβ(x)n ¼ bβf n(x)c. ically conjugate to a β-transformation.
(By convention, terminating expansions are
replaced by eventually periodic ones.) A one- Gaussian Systems
sided sequence on the alphabet D is in the β-shift Consider a real-valued stationary process {fk :
if and only if all of its shifts are lexicographically –1 < i < 1} on a probability space (Ω, F , P).
less than or equal to the expansion dβ(1) of 1 base The process (and the associated measure-
β. A one-sided sequence on the alphabet D is the preserving system consisting of the shift and a
valid expansion of 1 for some β if and only if it shift-invariant measure on ℝℤ) is called Gaussian
lexicographically dominates all its shifts. These if for each d  1, any d of the fk form an
were first studied by Bissinger, Rényi, and Parry; ℝd-valuedGaussian random variable on Ω : This
there are good summaries by Bertrand-Mathis means that with E( fk) ¼ m for all k and
(1986) and Blanchard
p
(1989).
1þ 5
For b ¼ 2 , db ð1Þ ¼ 10101010 . . . : Aij ¼ f ki
m f kj
m dP
For b ¼ 32 , d b ð1Þ ¼ 101000001 . . . (not even- O
tually periodic). ¼ C ki
kj for i, j ¼ 1, . . . , d,
Every β-shift is coded.
The topological entropy of a β-shift is log β. for each Borel set B  ℝ,
There is a unique measure of maximal entropy
log β.

P o : f k1 ðoÞ, . . . , f kd ðoÞ  B ¼
1 1
p exp
ðx
ðm, . . . , mÞÞtr A
1 ðx
ðm, . . . , mÞÞ dx1 . . . dxd :
2pd=2 det A B 2

The function C(k) is positive semidefinite and mixing. It is mixing if and only if C(k) ! 0 as
hence has an associated measure s on [0, 2 π] such |k| ! 1. If s is singular with respect to Lebesgue
that measure, then the entropy is 0; otherwise the
entropy is infinite (de la Rue 1993).
2p For more details, see (Cornfeld et al. 1982a).
CðkÞ ¼ eikt dsðtÞ:
0
Hamiltonian Systems
Theorem 4.3 The Gaussian system is ergodic if This paragraph is from the article on Measure-
and only if the “spectral measure” s is continuous Preserving Systems. Many systems that model
(i.e., nonatomic), in which case it is also weakly physical situations can be studied by means of
12 Ergodic Theory: Basic Examples and Constructions

Hamilton’s equations. The state of the entire sys- undergoes a perfectly elastic collision with the
tem at any time is specified by a vector angle of incidence equal to the angle of reflection
(q, p)  ℝ2n, the phase space, with q listing the and continues in a straight line until it next hits the
coordinates of the positions of all of the particles, boundary. It is usual to normalize and consider
and p listing the coordinates of their momenta. We unit speed, as we do in this discussion for conve-
assume there is a time-independent Hamiltonian nience. We take coordinates (x, v) given by the
function H(q, p) such that the time development of Euclidean coordinates in x  D together with a
the system satisfies Hamilton’s equations: direction vector v  Sd
1. A flow ft is defined
with respect to Lebesgue almost every (x, v) by
dqi @H dpi @H translating x a distance t defined by the direction
¼ , ¼
, i ¼ 1, . . . , n: ð4Þ
dt @pi dt @qi vector v, taking account of reflections at bound-
aries. ft preserves a measure absolutely continu-
Often in applications, the Hamiltonian func- ous with respect to Riemannian volume on (x, v)
tion is the sum of kinetic and potential energy: coordinates. The flow we have described is called
a billiard flow. The corresponding billiard map is
H ðq, pÞ ¼ K ðpÞ þ U ðqÞ: ð5Þ formed by taking the Poincaré map corresponding
to the cross-section given by the boundary @D. We
Solving these equations with initial state (q, p) will describe the planar billiard map; the higher
for the system produces a flow (q, p) ! Tt(q, p) in dimensional generalization is clear. The billiard
phase space which moves (q, p) to its position map is a map T : @D ! @D, where @D is
Tt(q, p) t units of time later. According to coordinatized by (s, θ), s  [0, L], where L is
Liouville’s formula (Mañé 1987a, Theorem 3.2), the length of @D and θ  (0, π) measures the
this flow preserves Lebesgue measure on ℝ2n. angle that inward pointing vectors make with the
Calculating dH/dt by means of the Chain Rule tangent line to @D at s. Given a point (s, θ), the
angle θ defines an oriented line l(s, θ) which
dH @H dpi @H dqi intersects @D in two points s and s0 . Reflecting
¼ þ
dt i
@pi dt @qi dt l in the tangent line to @D at the point s0 gives
another oriented line passing through s0 with angle
and using Hamilton’s equations shows that H is θ0 (measured with respect to the angular coordi-
constant on orbits of the flow, and thus each set of nate system based at s0 ). The billiard map is the
constant energy X(H0) ¼ {(q, p) : H(q, p) ¼ H0} is map T(s, θ) ¼ (s0 , θ0 ). T preserves a measure m ¼
an invariant set. There is a natural invariant mea- sin θds  dθ. The billiard flow may be modeled as
sure on a constant energy set X(H0) for the a suspension flow over the billiard map (see sec-
restricted flow, namely the measure given by tion “Suspension Flows”).
rescaling the volume element dS on X(H0) by the If the region D is a polygon in the plane
factor 1/k▽ Hk. (or polyhedron in ℝd), then @D consists of the
faces of the polyhedron. The dynamical behavior
Billiard Systems of the billiard map or flow in regions with only flat
These form an important class of examples in (noncurved) boundaries is quite different to that of
ergodic theory and dynamical systems, motivated billiard flows or maps in regions D with strictly
by natural questions in physics, particularly the convex or strictly concave boundaries. The topo-
behavior of gas models. Consider the motion of a logical entropy of a flat polygonal billiard is zero.
particle inside a bounded region D in ℝd with Research interest focuses on the existence and
piecewise smooth (C1 at least) boundaries. In the density of periodic or transitive orbits. It is
case of planar billiards, we have d ¼ 2. The known that if all the angles between sides are
particle moves in a straight line with constant rational multiples of π, then there are periodic
speed until it hits the boundary; at which point, it orbits (Boshernitzan 1992; Vorobets et al. 1992;
Masur 1986) and they are dense in the phase space
Ergodic Theory: Basic Examples and Constructions 13

(Boshernitzan et al. 1998). It is also known that a perturbing in the class of area-preserving
residual set of polygonal billiards are topologi- diffeomorphisms is an appropriate imposition in
cally transitive and ergodic (Zemljakov and many physical models. We will take the version of
Katok 1975; Kerckhoff et al. 1986). the KAM theorem as given in (Mañé 1987a, The-
On the other hand, billiard maps in which @D orem 5.1) (original references include
has strictly convex components are physical (Kolmogorov 1954; Arnol’d 1963; Moser
examples of nonuniformly hyperbolic systems 1962)). An elliptic fixed point for an area-
(with singularities). The meaning of concave or preserving diffeomorphism T of a surface M is
convex varies in the literature. We will consider a called a nondegenerate elliptic fixed point if
billiard flow inside a circle to be a system with a there is a local Cr, r  4, change of coordinates
strictly concave boundary, while a billiard flow on h so that in polar coordinates
the torus from which a circle has been excised to
be a billiard flow with strictly convex boundary. hTh
1 ðr, yÞ ¼ ðr, y þ a0 þ a1 r Þ þ F,
The class of billiards with some strictly convex
boundary components, sometimes called dispers- where all derivatives of F up to order 3 vanish,
ing billiards or Sinai billiards, was introduced by α1 6¼ 0, and α0 6¼ 0, p 2p
2 , π, 3 . A map of the form
Sinai (Sinaĭ 1970) who proved many of their
fundamental properties. Lazutkin (Lazutkin tðr, yÞ ¼ ðr, y þ a0 þ a1 r Þ,
1973) proved that planar billiards with generic
strictly concave boundary are not where α1 6¼ 0, is called a twist map. Note that a
ergodic. Nevertheless, Bunimovich (Bunimovič twist map leaves invariant the circle r ¼ k, any
1974; Bunimovich 1979) produced a large billiard constant k, and rotates each invariant curve by a
system, Bunimovich billiards, with strictly con- rigid rotation α1r, the magnitude of the rotation
cave boundary segments (perhaps with some flat depending upon r. With respect to two-
boundaries as well) which were ergodic and non- dimensional Lebesgue measure, a twist map is
uniformly hyperbolic. For more details, see certainly not ergodic.
(Chernov and Markarian 2006a; Katok et al.
1986; Liverani and Wojtkowski 1995; Theorem Suppose T is a volume-preserving
Tabachnikov 2005). We will discuss possibly the diffeomorphism of class Cr, r  4, of a surface
simplest example of a dispersing billiard, namely M. If x is a nondegenerate elliptic fixed point, then
a toral billiard with a single convex obstacle. Take for every ϵ > 0 there exists a neighborhood Uϵ of
the torus T 2 and consider a single strictly convex x and a set U0,ϵ  Uϵ with the properties:
subdomain S with C1 boundary. The domain of
the billiard map is [0, L]  (0, π), where L is the (a) U0,ϵ is a union of T-invariant simple closed
length of @S. The measure sin(θ)ds  dθ is pre- curves of class Cr–1 containing x in their
served. If the curvature of @S is everywhere non- interior.
zero, then the billiard map T has positive (b) The restriction of T to each such invariant
topological entropy, periodic points are dense, curve is topologically conjugate to an irratio-
and in fact the system is isomorphic to a Bernoulli nal rotation.
shift (Gallavotti and Ornstein 1974). (c) m(Uϵ
U0,ϵ)  ϵUϵ, where m is Lebesgue
measure on M.
KAM-Systems and Stably Nonergodic Behavior As a corollary, we have.
A celebrated theorem of Kolmogorov, Arnold,
and Moser (the KAM theorem) implies that the Corollary Let M be a compact surface without
set of ergodic area-preserving diffeomorphisms of boundary and Diff r(M ) the space of C r area-
a compact surface without boundary is not dense preserving diffeomorphisms with the C r topology.
in the Cr topology for r  4. This has important Then the set of T  Diff r(M) which are ergodic
implications, in that there are natural systems in with respect to the probability measure
which ergodicity is not generic. The constraint of
14 Ergodic Theory: Basic Examples and Constructions

determined by normalized area is not dense in structurally stable if and only if it is uniformly
Diff r(M) for r  4. hyperbolic and satisfies a technical assumption
called strong transversality (see below for details).
Smooth Uniformly Hyperbolic Suppose M is a C1 compact Riemannian man-
Diffeomorphisms and Flows ifold equipped with metric d and tangent space
Time series of measurements on deterministic TM with norm k k. Suppose also that U  Mis a
dynamical systems sometimes display limit laws nonempty open subset and T : U ! T(U ) is a C1
exhibited by independent identically distributed diffeomorphism. A compact T invariant set Λ  U
random variables, such as the central limit theo- is called a hyperbolic set if there is a splitting of
rem, and also various mixing properties. The the tangent space Tp M at each point p  Λ into
models of hyperbolicity we discuss in this section two invariant subspaces, TpM ¼ Eu( p) Es( p),
have played a key role in showing how this phe- and a number 0 < l < 1 such that for n  0
nomenon of “chaotic behavior” arises in deter-
ministic dynamical systems. Hyperbolic sets and Dp T n v  Cln kvk for v  Es ðpÞ,
their associated dynamics have also been pivotal Dp T
n v < Cln kvk for v  Eu ðpÞ:
in studies of structural stability. A smooth system
is Cr structurally stable if a small perturbation in The subspace Eu is called the unstable or
the Cr topology gives rise to a system which is expanding subspace and the subspace Es the sta-
topologically conjugate to the original. When ble or contracting subspace. The stable and unsta-
modeling a physical system, it is desirable that ble subspaces may be integrated to produce stable
slight changes in the modeling parameters do not and unstable manifolds
greatly affect the qualitative or quantitative
behavior of the ensemble of orbits considered as W s ðpÞ ¼ fy : dðT n p, T n yÞ ! 0g as n ! 1,
a whole. The orbit of a point may change drasti-
W u ðpÞ ¼ fy : dðT
n p, T
n yÞ ! 0g as n ! 1:
cally under perturbation (especially if the system
has sensitive dependence on initial conditions) but
The stable and unstable manifolds are immer-
the collection of all orbits should ideally be “sim-
sions of Euclidean spaces of the same dimension
ilar” to the original unperturbed system. In the
as Es( p) and Eu( p), respectively, and are of the
latter case, one would hope that statistical proper-
same differentiability as T. Moreover, Tp(Ws( p)) ¼
ties also vary only slightly under perturbation.
Es( p) and Tp(W u( p)) ¼ Eu( p). It is also useful to
Structural stability is one, quite strong, notion of
define local stable manifolds and local unstable
stability. The conclusion of a body of work on
manifolds by
structural stability is that a system is C1

W sϵ ðpÞ ¼ fy  W s ðpÞ : d ðT n p, T n yÞ < ϵg for all n  0,

W uϵ ðpÞ ¼ fy  W u ðpÞ : d ðT
n p, T
n yÞ < ϵg for all n  0:

Finally, we discuss the notion of strong trans- In the C r, r  1 topology, Robbin (1971), de Melo
versality. We say a point x is nonwandering if for (1973), and Robinson (1975, 1976) proved that
each open neighborhood U of x there exists an dynamical systems with the strong transversal
n > 0 such that T n ðUÞ \ U 6¼ 0. The NW set of property are structurally stable, and Robinson

nonwandering points is called the nonwandering (1973) in addition showed that strong trans-
set. We say a dynamical system has the strong versality was also necessary. Mañé (1988) showed
transversal property if Ws(x) intersects W u( y) that a C1 structurally stable diffeomorphism must
transversely for each pair of points x, y  NW. be uniformly hyperbolic, and Hayashi (1997)
Ergodic Theory: Basic Examples and Constructions 15

extended this to flows. Thus a C1 diffeomorphism Hyperbolic Dynamical Systems”), was shown to
or flow on a compact manifold is structurally be stably ergodic (Grayson et al. 1994), so the
stable if and only if it is uniformly hyperbolic geodesic flow is still playing a major role in the
and satisfies the strong transversality condition. development of ergodic theory.

Geodesic Flow on Manifold of Negative Curvature Horocycle Flow

The study of the geodesic flow on manifolds of All surfaces endowed with a Riemannian metric
negative sectional curvature by Hedlund and Hopf of constant negative curvature are quotients of the
was pivotal to the development of the ergodic upper half-plane ℋ+ ≔ {x þ iy  ℂ : y > 0} with
2 2
theory of hyperbolic systems. Suppose that M is the metric ds2 ¼ dx yþdy
2 , whose sectional curva-
a geodesically complete Riemannian manifold. ture is
1. The orientation-preserving isometries
Let γp,v(t) be the geodesic with γp,v(0) ¼ p and of this metric are exactly the linear fractional (also
g_ p,v ð0Þ ¼ v , where g_ p,v γ denotes the derivative known as Möbius) transformations:
with respect to time t. The geodesic flow is a
flow ft on the tangent bundle TM of M, ft : ℝ  a b þ az þ b þ
TM ! TM, defined by  SLð2, ℝÞ, z  ℋ 7! ℋ :
c d cz þ d

ft ðp, vÞ ¼ gp,v ðtÞ, g_ p,v ðtÞ : Since each matrix I corresponds to the iden-
tity transformation, we consider matrices inPSL
where (p, v)  TM. Since geodesics have constant (2, ℝ) ≔ SL(2, ℝ)/{I}.
speed, if kvk ¼ 1 then kγp,v(t)k ¼ 1 for all t, and thus The unit tangent bundle, Sℋ+, of the upper
the unit tangent bundle T 1M ¼ {( p, v)  TM : half-plane can be identified with PSL(2, ℝ). Then
kvk ¼ 1} is preserved under the geodesic flow. the geodesic flow corresponds to the
The geodesic flow and its restriction to the unit transformations
tangent bundle both preserve a volume form,
Liouville measure. In 1934, Hedlund (1934) pro- et 0
ved that the geodesic flow on the unit tangent t  ℝ 7!
0 e
t
bundle of a surface of strictly negative constant
sectional curvature is ergodic, and in 1939 Hopf seen as acting on PSL(2, ℝ). The unstable folia-
(1939) extended this result to manifolds of arbi- tion of an element A  PSL(2, ℝ) ffi Sℋ+ is
trary dimension and strictly negative (not neces- given by
sarily constant) curvature. Hopf’s technique of
proof of ergodicity (Hopf argument) was 1 t
extremely influential and used the foliation of A, t  ℝ,
0 1
the tangent space into stable and unstable mani-
folds. For a clear exposition of this technique, and and the flow along this foliation, given by
the property of absolute continuity of the folia-
tions into stable and unstable manifolds, see
1 t
(Liverani and Wojtkowski 1995). The geodesic t  ℝ 7! ,
0 1
flow on manifolds of constant negative sectional
curvature is an Anosov flow (see section “Anosov
is called the horocycle flow, similarly for the flow
Systems”). We remark that for surfaces sectional
induced on the unit tangent bundle of each quo-
curvature is the same as Gaussian curvature.
tient of the upper half-plane by a discrete group of
Recently, the time-one map of the geodesic flow
linear fractional transformations.
on the unit tangent bundle of a surface with con-
The geodesic and horocycle flows acting on a
stant negative curvature, which is a partially
(finite-volume) surface of constant negative
hyperbolic system (see section “Partially
16 Ergodic Theory: Basic Examples and Constructions

curvature form the fundamental example of a and 0 < l < 1, constant C such that
transverse pair of actions. The geodesic flow k|DT nvk < Clnkvk for all v  Es( p) and
often has many periodic orbits and many invariant kDT
nwk  Clnkwk for all w  Eu( p).
measures, has positive entropy, and is in fact A similar definition holds for Anosov flows
Bernoulli with respect to the natural measure f : ℝ  M ! M. A flow is Anosov if there is a
(Ornstein and Weiss 1973), while the horocycle splitting of the tangent bundle into flow-invariant
flow is often uniquely ergodic (Furstenberg 1973; subspaces Eu, Es, and Ec so Dp ft Esp ¼
Marcus 1975) and of entropy zero, although EsftðpÞ , Dp ft Eup ¼ Euft ðpÞ , and Dp ft Ecp ¼ Ecft ðpÞ , and
mixing of all orders (Marcus 1978). See at each point p  M
(Hasselblatt and Katok 2003a) for more details.
T p M ¼ Esp Eup Ecp
Markov Partitions and Coding
If (X, T, ℬ, m) is a dynamical system, then a finite Dp ft v < Clt kvk for v  Es ðpÞ
partition of X always induces a coding of the Dp f
t v < Clt kvk for v  Eu ðpÞ
orbits and a semiconjugacy with a subshift on a
symbol space (it may not of course be a full for some 0 < l < 1. The tangent to the flow
conjugacy). For hyperbolic systems, a special direction Ec( p) is a neutral direction:
class of partitions, Markov partitions, induce a
conjugacy for the invariant dynamics to a subshift Dp ft v ¼ kvk for v  Ec ðpÞ:
of finite type. A Markov partition P for an invari-
ant subset Λ of a diffeomorphism T of a compact Anosov proved that Anosov flows and
manifold M is a finite collection of sets Ri, diffeomorphisms which preserve a volume form
1  i  n called rectangles. The rectangles have are ergodic (Anosov 1967) and are also structur-
the property, for some ϵ > 0, if x, y  Ri then ally stable. Sinai (1968) constructed Markov par-
W sϵ ðxÞ \ W uϵ ðyÞ  Ri . This is sometimes described titions for Anosov diffeomorphisms and hence
as being closed under local product structure. We coded trajectories via a subshift of finite type.
let W u(x, Ri) denote W uϵ ðxÞ \ Ri and W s(x, Ri) Using ideas from statistical physics in (Sinaĭ
denote W sϵ ðxÞ \ Ri : Furthermore, we require for 1972), Sinai constructed Gibbs measures for
all i, j: Anosov systems. An SRB measure (see section
“Physically Relevant Measures and Strange
1. Each Ri is the closure of its interior. Attractors”) is a type of Gibbs measure
2. Λ  [iRi. corresponding to the potential
log j det DT jEu j
3. Ri \ Rj ¼ @Ri \ @Rj if i 6¼ j. and is characterized by the property of absolutely
4. If x  Roi and T ðxÞ  Roj , then W u(T(x), Rj)  continuous conditional measures on unstable
T(W u(x, Ri)) and W s(x, Ri)  T
1(W u(T(x), Rj)). manifolds.
The simplest examples of Anosov
Anosov Systems diffeomorphisms are perhaps the two-
An Anosov diffeomorphism (Anosov 1967) is a dimensional hyperbolic toral automorphisms (the
uniformly hyperbolic system in which the entire n > 2 generalization is clear). Suppose A is a 2  2
manifold is a hyperbolic set. Thus, an Anosov matrix with integer entries
diffeomorphism is a C1 diffeomorphism T of
M with a DT-invariant splitting (which is a con- a b
tinuous splitting) of the tangent space TM(x) at
c d
each point p into a disjoint sum
such that det(A) ¼ 1 and A has no eigenvalues of
T p M ¼ Eu ð pÞ Es ð pÞ
modulus 1. Then A defines a transformation of the
two-dimensional torus T 2 ¼ S1  S1 such that if
v  T 2,
Ergodic Theory: Basic Examples and Constructions 17

v1 Theorem (Spectral Decomposition Theorem)

v¼ , If T is Axiom A, then there is a unique decompo-
v2
sition of the nonwandering set NW of T
then
NW ¼ L1 [ . . . [ Lk
ðav1 þ bv2 Þ
Av ¼ : as a disjoint union of closed, invariant, locally
ðcv1 þ dv2 Þ
maximal hyperbolic sets Λi such that T is transi-
A preserves Lebesgue (or Haar) measure and is tive on each Λi. Furthermore, each Λi may be
ergodic. A prominent example of such a matrix is further decomposed into a disjoint union of closed
sets Lji , j ¼ 1, . . . ni such that T ni is topologically
2 1 mixing on each Lji and T cyclically permutes the
, Lji .
1 1
Horseshoe Maps This type of map was intro-
which is sometimes called the Arnold Cat Map.
duced by Steven Smale in the 1960s and has
Each point with rational coordinates ( p1/q1, p2/q2)
played a pivotal role in the development of
is periodic. There are two eigenvalues 1l < 1 <
p dynamical systems theory. It is perhaps the canon-
l ¼ 3þ2 5 with orthogonal eigenvectors, and the ical example of an Axiom A system (Smale 1980)
projections of the eigenspaces from ℝ2 to T 2 are and is conjugate to a full shift on 2 symbols. Let
the stable and unstable subspaces. S be a unit square in ℝ2, and let T be a
diffeomorphism of S onto its image such that
Axiom A Systems S \ T(S) consists of two disjoint horizontal strips
In the case of Anosov diffeomorphisms, the split- S0 and S1. Think of stretching S uniformly in the
ting into contracting and expanding bundles holds horizontal direction and contracting uniformly in
on the entire phase space M. A system T : M ! M the vertical direction to form a long thin rectangle,
is an Axiom A system if the nonwandering set NW and then bending the rectangle into the shape of a
is a hyperbolic set and periodic points are dense in horsehoe and laying the straight legs of the horse-
the nonwandering set. NW  M may have shoe back on the unit square S. This transforma-
Lebesgue measure zero. A set Λ  M is locally tion may be realized by a diffeomorphism, and we
maximal if there exists an open set U such that may also require that T restricted to T
1Si, i ¼ 0,
Λ ¼ \n  ℤT n(U ). The solenoid and horseshoe 1, acts as a linear map. The restriction of T to the
discussed below are examples of locally maximal maximal invariant set H ¼ \1 i
i¼
1 T S is a Smale
sets. Bowen (1970) constructed Markov partitions horseshoe map. H is a Cantor set, the product of a
for Axiom A diffeomorphisms. Ruelle imported Cantor set in the horizontal direction, and a Cantor
ideas from statistical physics, in particular the idea set in the vertical direction. The conjugacy with
of an equilibrium state and the variational princi- the shift on two symbols is realized by mapping
ple, to the study of Axiom A systems (see Ruelle x  H to its itinerary with respect to the sets S0
1976, 1978). This work extended the notion of and S1 under powers of T (positive and negative
Gibbs measure and other ideas from statistical powers).
mechanics, introduced by Sinai for Anosov sys-
tems (Sinaĭ 1972), into Axiom A systems. Solenoids The solenoid is defined on the solid
One achievement of the Axiom A program was torus X in ℝ3 which we coordinatize as a circle of
the Smale Decomposition Theorem, which breaks two-dimensional solid disks, so that
the dynamics of Axiom A systems into locally
maximal sets and describes the dynamics on X ¼ fðy, zÞg : y  ½0, 1 and j z j 1, z  ℂg
each (Bowen 1970, 1975; Smale 1980).
The transformation T : X ! X is given by
18 Ergodic Theory: Basic Examples and Constructions

1 1 • Ec is uniformly dominated by Eu and Es: If vc 

T ðy, zÞ ¼ 2y ðmod 1Þ, z þ e2piy
4 2 Ec(x) \ {0}, then there are numbers gðpÞ, gðpÞ
such that mðpÞ < gðpÞ  k kpvc k k  gðpÞ < MðpÞ.
D Tvc

Geometrically, the transformation stretches the

torus to twice its length, shrinks its diameter by a
The notion of partial hyperbolicity was intro-
factor of 4, then twists it, and doubles it over,
duced by Brin and Pesin (1974) who proved exis-
placing the resultant object without self-
tence and properties, including absolute
intersection back inside the original solid torus.
continuity, of invariant foliations in this setting.
T(X) intersects each disk Dc ¼ {(θ, z) : θ ¼ c} in
There has been intense recent interest in partially
two smaller disks of 14 the diameter. The transfor-
hyperbolic systems primarily because significant
mation T contracts volume by a factor of 2 upon progress has been made in establishing that cer-
each application, yet there is expansion in the θ tain volume-preserving partially hyperbolic sys-
direction (θ ! 2θ). The solenoid A ¼ \n0T n(X) tems are “stably ergodic,” that is, they are ergodic
has zero Lebesgue measure, is T-invariant, and is and under small (Cr topology) volume-preserving
(locally) topologically a line segment crossed with perturbations remain ergodic. This phenomenon
a two-dimensional Cantor set (A intersects each had hitherto been restricted to uniformly hyper-
disk Dc in a Cantor set). The set A is an attractor, bolic systems. For recent developments, and pre-
in that all points inside X limit under iteration by cise statements, on stable ergodicity of partially
T upon A. T : A ! A is an Axiom A system. hyperbolic systems see (Burns et al. 2001; Pugh
and Shub 2004), ?Burns-Wilkinson).
Partially Hyperbolic Dynamical Systems
Partially hyperbolic dynamical systems are a gen- Compact Group Extensions of Uniformly
eralization of uniformly hyperbolic systems in Hyperbolic Systems
that an invariant central direction is allowed but A natural example of a partially hyperbolic sys-
the contraction in the central direction is strictly tem is given by a compact group extension of an
weaker than the contraction in the contracting Anosov diffeomorphism. If the following terms
direction and the expansion in the central direc- are not familiar, see section “Constructions” on
tion is weaker than the expansion in the expanding standard constructions. Suppose that (T, M, m) is
direction. More precisely, suppose M is a C1 com- an Anosov diffeomorphism, G is a compact
pact (adapted) Riemannian manifold equipped connected Lie group, and h : M ! G is a differ-
with metric d and tangent space TM with norm entiable map. The skew product F : M  G ! M 
k k. A C1 diffeomorphism T of M is a partially G given by
hyperbolic diffeomorphism if there is a nontrivial
continuous DT-invariant splitting of the tangent Fðx, gÞ ¼ ðTx, hðxÞgÞ
space TpM at each point p into a disjoint sum
has a central direction in its tangent space
T p M ¼ Eu ð pÞ Ec ð pÞ Es ð pÞ corresponding to the Lie algebra LG of G (as a
group element h acts isometrically on G, there is
and continuous positive functions m, M, g, g such no expansion or contraction) and uniformly
that expanding and contracting bundles corresponding
to those of the tangent space of T : M ! M. Thus,
• Es is contracted: If vs  Es(x) \ {0}, then T(M  G) ¼ Eu LG Es.
kD p T n v s k
kvs k  mðpÞ < 1.
• Eu is expanded: If vu  Eu(x) \ {0}, then Time-One Maps of Anosov Flows
kDp Tvs k Another natural context in which partial hyper-
kvs k  MðpÞ > 1. bolicity arises is in time-one maps of uniformly
hyperbolic flows. Suppose ft : ℝ  M ! Mis an
Anosov flow. The diffeomorphism f1 : M ! M is
Ergodic Theory: Basic Examples and Constructions 19

a partially hyperbolic diffeomorphism with cen- to consider ergodicity (and other statistical prop-
tral direction given by the flow direction. There is erties) of the system with respect to Lebesgue
no expansion or contraction in the central measure. In dissipative systems, a measure equiv-
direction. alent to Lebesgue may not be invariant (for exam-
ple, the solenoid). Nevertheless, Lebesgue
Nonuniformly Hyperbolic Systems measure has a distinguished role since sampling
The assumption of uniform hyperbolicity is quite by experimenters is done with respect to Lebesgue
restrictive, and few “chaotic systems” found in measure. The idea of a physically relevant mea-
applications are likely to exhibit uniform hyper- sure m is that it determines the statistical behavior
bolicity. A natural weakening of this assumption, of a positive Lebesgue measure set of orbits, even
and one that is nontrivial and greatly extends the though the support of m may have zero Lebesgue
applicability of the theory, is to require the hyper- measure. An example of such a situation in the
bolic splitting (no longer uniform) to hold only at uniformly hyperbolic setting is the solenoid Λ,
almost every point of phase space. A systematic where the attracting set Λ has Lebesgue measure
theory was built by Pesin (1976, 1977) on the zero and is (locally) topologically the product of a
assumption that the system has nonzero Lyapunov two-dimensional Cantor set and a line segment.
exponents m almost everywhere, where m is Nevertheless, Λ determines the behavior of all
Lebesgue-equivalent invariant probability mea- points in a solid torus in ℝ3. More generally,
sure. Recall that a number l is a Lyapunov expo- suppose that T : M ! M is a diffeomorphism on
nent for p  M if kDpT nvk eln for some unit a compact Riemannian manifold and that m is a
vector v  TpM. Oseledet’s theorem (Oseledec version of Lebesgue measure on M, given by a
1968) (see also Walters 1982a, p. 232), which is smooth volume form. Although Lebesgue mea-
also called the Multiplicative Ergodic Theorem, sure m is a distinguished physically relevant mea-
implies that if T is a C1 diffeomorphism of M, then sure, m may not be invariant under T, and the
for any T-invariant ergodic measure m almost system may even be volume contracting in the
every point has well-defined Lyapunov expo- sense that m(T nA) ! 0 for all measurable sets A.
nents. One of the highlights of Pesin theory is Nevertheless, an experimenter might observe
the following structure theorem: If T : M ! M is long-term “chaotic” behavior whenever the state
a C1þϵ diffeomorphism with a T-invariant of the system gets close to some compact invariant
Lebesgue-equivalent Borel measure m such that set X which attracts a positive m-measure of orbits
T has nonzero Lyapunov exponents with respect in the sense that these orbits limit on X. Possibly
to m, then T has at most a countable number of m(X) ¼ 0, so that X is effectively invisible to the
ergodic components {Ci} on each of which the observer except through its effects on orbits not
restriction of T is either Bernoulli or Bernoulli contained in X. The dynamics of T restricted to
times a rotation (by which we mean the support X can in fact be quite complicated – maybe a full
of mi ¼ mjCi consists of a finite number ni of sets shift, or a shift of finite type, or some other com-
Si1 , . . . Sini cyclically permuted and T ni is plicated topological dynamical system. Suppose
there is a T-invariant measure m supported on
Bernoulli when restricted to each Sij Þ (Pesin X such that for all continuous functions f : M ! ℝ
1977; Young 1993). This structure theorem has
been generalized to SRB measures with nonzero n
1
1
Lyapunov exponents (Ledrappier 1984; Pesin f ∘ T k ðxÞ ! fdm, ð6Þ
1977). n k¼0 X

Physically Relevant Measures and Strange for a positive m-measure of points x  M. Then
Attractors the long-term equilibrium dynamics of an observ-
(This paragraph is from the article on Measure- able set of points x  M (i.e., a set of points of
Preserving Systems.) For Hamiltonian systems positive m measure) is described by (X, T, m). In
and other volume-preserving systems, it is natural this situation, m is described as a physical
20 Ergodic Theory: Basic Examples and Constructions

measure. There has been a great deal of research manifolds is absolutely continuous (takes sets of
on the properties of systems with attractors zero Lebesgue measure on W u to sets of zero
supporting physical measures. Lebesgue measure on W u), there is a positive
In the dissipative nonuniformly hyperbolic set- Lebesgue measure of points (namely an unstable
ting, the theory of “physically relevant” measures manifold and the union of stable manifolds
is best developed in the theory of SRB (for Sinai, through it) satisfying (Eq. 7). Thus an SRB mea-
Ruelle, and Bowen) measures. These dynamically sure with absolutely continuous holonomy maps
invariant measures may be supported on a set of along stable manifolds is a physically relevant
Lebesgue measure zero yet determine the asymp- measure. If the stable foliation possesses this
totic behavior of points in a set of positive property, it is called absolutely continuous. An
Lebesgue measure. Axiom A attractor for a C2 diffeomorphism is an
If T is a diffeomorphism of M and m is a example of an SRB attractor (Bowen 1975; Ruelle
T-invariant Borel probability measure with posi- 1976, 1978; Sinaĭ 1972). The examples we have
tive Lyapunov exponents which may be inte- given of SRB measures and attractors and mea-
grated to unstable manifolds, then we call m an sures have been uniformly hyperbolic.
SRB measure if the conditional measure m induced Recently, much progress has been made in
on the unstable manifolds is absolutely continu- understanding the statistical properties of non-
ous with respect to the Riemannian volume ele- uniformly hyperbolic systems by using a tower
ment on these manifolds. The reason for this (see section “Induced Transformations”) to con-
definition is technical but can be gleaned from struct SRB measures. We refer to Young’s original
the following observation. Suppose that the papers (Young 1998, 1999), to the book by Baladi
diffeomorphism has no zero Lyapunov exponents (2000a), and to (Young 1993) for a recent survey
with respect to m. Since T is a diffeomorphism, on SRB measures in the nonuniformly setting.
this implies T has negative Lyapunov exponents
as well as positive Lyapunov exponents and Unimodal Maps
corresponding local stable manifolds as well as Maps of an interval to itself are simple examples
local unstable manifolds. Suppose that a of non-uniformly hyperbolic systems that have
T-invariant set A consists of a union of unstable played an important role in the development of
manifolds and is the support of an ergodic SRB dynamical systems theory. Suppose I  ℝ is an
measure m and that f : M ! ℝ is a continuous interval; for simplicity, we take I ¼ [0, 1].
function. Since m has absolutely continuous con- A unimodal map is a map T : [0, 1] ! [0, 1]
ditional measures on unstable manifolds with such that there exists a point 0 < c < 1 and
respect to conditional Lebesgue measure on the
unstable manifolds, almost every point x in the • T is C2.
union of unstable manifolds U satisfies • T0 (x) > 0 for x < c, T0 (x) < 0 for x > c.
• T0 (c) ¼ 0.
n
1
1
lim f ∘ T j ðxÞ ¼ f dm ð7Þ Such a map is clearly not uniformly expanding,
n!1 n
j¼0
as |T0 (x)| < 1 for points in a neighborhood of c.
The family of maps Tm (x) ¼ mx(1 – x), 0 < m  4,
If y  W sϵ ðxÞ for such an x  U, then d(T nx,
is a family of unimodal maps with c ¼ 1/2 and
T y) ! 1 and hence (Eq. 7) implies
n
T2(1/2) ¼ 1/2, T4(1/2) ¼ 1.
n
1
We could have taken the interval I to be [
1, 1]
1 or indeed any interval with an obvious modifica-
lim f∘T j ðyÞ ¼ f dm
n!1 n
j¼0 tion of the definition above. A well-studied family
of unimodal maps in this setting is the logistic
Furthermore, if the holonomy between unsta- family fa : [
1, 1] ! [
1, 1], fa(x) ¼ 1
ax2,
ble manifolds defined by sliding along stable a  (0, 2). The families are equivalent under a
Ergodic Theory: Basic Examples and Constructions 21

smooth coordinate change, so statements about topologically conjugate to the dyadic adding
one family may be translated into statements machine coexisting with isolated repelling orbits
about the other. of period 2n, n ¼ 0, 1, 2, . . . There is a unique
Unimodal maps are studied because of the repelling orbit of period 2n for n  1 along with
insights they offer into transitions from regular two fixed points. The Cantor set is the o-limit set
or periodic to chaotic behavior as a parameter for all points that are not periodic or preimages of
(e.g., m or a) is varied, the existence of absolutely periodic orbits. C is the set of accumulation points
continuous measures, and rates of decay of corre- of periodic orbits. Despite this picture of incredi-
lations of regular observations for nonuniformly ble complexity, the topological entropy is zero for
hyperbolic systems. l  l1. For l > l1, the map Tl has positive
A result of Jakobson (1981) and Benedicks and topological entropy and infinitely many periodic
Carleson (1985) implies that in the case of the orbits whose periods are not powers of 2. For each
logistic family there is a positive Lebesgue mea- l  l1, Tl possesses an invariant Cantor set
sure set of a such that fa has an absolutely contin- which is repelling for l > l1. We say that Tl is
uous ergodic invariant measure ma. It has been hyperbolic if there is only one attracting periodic
shown by Young (1999) and Keller and Nowicki orbit and the only recurrent sets are the attracting
(1992) that if fa is mixing with respect to ma then periodic orbit, repelling periodic orbits, and pos-
the decay of correlations for Lipshitz observations sibly a repelling invariant Cantor set. It is known
on I is exponential. It is also known that the set of that the set of l  [0, 4] for which Tl is hyperbolic
a such that fa is mixing with respect to ma has is open and dense (Graczyk and Światek 1997).
positive Lebesgue measure. There is a well- Remarkably, by Jakobson’s result (Jakobson
developed theory concerning the bifurcations the 1981) there is also a positive Lebesgue measure
maps Tm undergo as m varies (Collet and Eckmann set of parameters l for which Tl has an absolutely
1980a). We briefly describe the period-doubling continuous invariant measure ml with a positive
route to chaos in the family Tl(x) ¼ lx(1
x). For Lyapunov exponent.
a good account, see (Hasselblatt and Katok
2003a). We let cl denote the fixed point l
1 l . For Intermittent Maps
p
3 < l  1 þ 6, all points in [0, 1] except for 0, Maps of the unit interval T : [0, 1] ! [0, 1] which
cl, and their preimages are attracted to a unique are expanding except at the point x ¼ 0, where
periodic orbit O( pl) of period 2. There is a mono- they are locally x x þ x1þα, α > 0, have been
tone sequence of parameter values ln (l1 ¼ 3) extensively studied both for the insights they give
such that for ln < l  lnþ1, Tl has a unique into rates of decay of correlations for non-
attracting periodic orbit O(ln) of period 2n and uniformly hyperbolic systems (hyperbolicity is
for each k ¼ 1, 2, . . . , n
1 a unique repelling lost at the point x ¼ 0, where the derivative is 1)
orbit of period 2k. All points in the interval [0, 1] and for their use as models of intermittent behav-
except for the repelling periodic orbits and their ior in turbulence (Manneville and Pomeau 1980).
preimages are attracted to the attracting periodic A fixed point where the derivative is 1 is some-
orbit of period 2n. At l ¼ ln, the periodic orbit times called an indifferent fixed point. It is a model
O(ln) undergoes a period-doubling bifurcation. of intermittency in the sense that orbits close to
Feigenbaum (1978) found that the limit d ¼ 1 will stay close for many iterates (since the
ln
ln
1
lnþ1
ln 4:699 . . . exists and that in a wide class expansion is very weak there), and hence a time
of unimodal maps this period-doubling cascade series of observations will be quite uniform for
occurs and the differences between successive long periods of time before displaying chaotic
bifurcation parameters give the same limiting type behavior after moving away from the indif-
ratio, an example of universality. At the end of ferent fixed into that part of the domain where the
the period-doubling cascade at a parameter map is uniformly expanding.
l1 3:569 . . . , T l1 has an invariant Cantor set A particularly simple model (Liverani et al.
C (the Feigenbaum attractor) which is 1999) is provided by
22 Ergodic Theory: Basic Examples and Constructions

xð1 þ 2a xa Þ if x  ½0, 1=2 ; of the extended complex plane ℂ to itself, in

T ðxÞ ¼ which the domain is ℂ completed with the point
2x
1 if x  ½1=2, 1 :
at infinity (called the Riemann sphere). Recall that
For α ¼ 0, the map is uniformly expanding and a family F of meromorphic functions is called
Lebesgue measure is invariant. In this case, the normal on a domain D if every sequence pos-
rate of decay of correlations for Hölder observa- sesses a subsequence that converges uniformly
tions is exponential. For 0 < α < 1, the map has an (in the spherical metric ℂ S2 ) on compact sub-
SRB measure mα with support the unit interval. sets of D. A family is normal at a point z  ℂ if it is
For α  1, there are no absolutely continuous normal on a neighborhood of z. The Fatou set
invariant probability measures though there are FðRÞ  ℂ of a rational map R : ℂ ! ℂ is the set
s-finite absolutely continuous measures. Upper of points z  ℂ such that the family of forward
and lower polynomial bounds on the rate of iterates {Rn}n0 is normal at z. The Julia set J(R)
decay of observations on such maps have been is the complement of the Fatou set F(R). The
given as a function of 0 < α < 1 and the regularity Fatou set is open, and hence the Julia set is a
of the observable. For details, see (Huyi 2004; closed set. Another characterization in the case
Liverani et al. 1999; Sarig 2002). d > 1 is that J(R) is the closure of the set of all
repelling periodic orbits of R : ℂ ! ℂ. Both F(R)
Hénon Diffeomorphisms and J(R) are invariant under R. The dynamics of
The Henón family of diffeomorphisms was intro- greatest interest is the restriction R : J(R) ! J(R).
duced and studied as Poincaré maps for the The Julia set often has a complicated fractal struc-
Lorenz system of equations. It is a two-parameter ture. In the case that Ra(z) ¼ z2
a, a  ℂ, the
two-dimensional family which shares many char- Mandelbrot set is defined as the set of a for which
acteristics with the logistic family and for small the orbit of the origin 0 is bounded. The topology
b > 0 may be considered a two-dimensional “per- of the Mandelbrot set has been the subject of
turbation” of the logistic family. The parametrized intense research. The study of complex dynamics
mapping is defined as is important because of the fascinating and com-
plicated dynamics displayed and also because
T a,b ðx, yÞ ¼ 1
ax2 þ y, bx , techniques and results in complex dynamics
have direct implications for the behavior of one-
so Ta,b : ℝ2 ! ℝ2 with 0 < a < 2 and b > 0. dimensional maps. For more details, see
Benedicks and Carleson (1991) showed that for a (Carleson and Gamelin 1993a).
positive-measure set of parameters (a, b), Ta,b has
a topologically transitive attractor Λa,b. Benedicks Infinite Ergodic Theory
and Young (1993) later proved that for a positive- We may also consider a measure-preserving trans-
measure set of parameters (a, b), Ta,b has a topo- formation (T, X, m) of a measure space such that m
logically transitive SRB attractor Λa,b with SRB (X) ¼ 1. For example, X could be the real line
measure ma,b and that (Ta,b, Λa,b, ma,b) is isomor- equipped with Lebesgue measure. This setting
phic to a Bernoulli shift. also arises with compact X in applications. For
example, suppose T: [0, 1] ! [0, 1] is the simple
Complex Dynamics model of intermittency given in section “Intermit-
Complex dynamics is concerned with the behav- tent Maps” and γ  (1, 2). Then T possesses an
ior of rational maps absolutely continuous invariant measure m with
support [0, 1], but m ([0, 1]) ¼ 1. The Radon-
a1 zd þ a2 zd
1 þ . . . adþ1 Nikodym derivative of m with respect to Lebesgue
b1 zd þ b2 zd
1 þ . . . bdþ1 measure m exists but is not in L1(m).
In this setting, we say a measurable set A is a
wandering set for T if fT
n Ag1
n¼0 are disjoint. Let
Ergodic Theory: Basic Examples and Constructions 23

D(T ) be the measurable union of the collection of systems which react to inputs from other systems,
wandering sets for T. The transformation T is con- and continuous time systems are often modeled as
servative with respect to m if (X \ D(T )) ¼ X (mod suspension flows over discrete-time dynamics) or
m) (for more details, see the recent survey by to reduce systems to simpler components (often a
A. Danilenko and C. Silva in this volume). It is factor system or induced system is simpler to
usually necessary to assume T conservative with study). Unless stated otherwise, in the following
respect to m to say anything interesting about its we discuss measure-preserving transformations
behavior. For example, if T(x) ¼ x þ α, α > 0 is a on Lebesgue spaces (see the article on Measure-
translation of the real line, then D(T) ¼ X. The Preserving Systems).
definition of ergodicity in this setting remains the
same: T is ergodic if A  ℬ, and T
1 A ¼ A mod Products
m implies that m(A) ¼ 0 or m(Ac) ¼ 0. However, the Given measure-preserving systems (X, ℬ, m, T )
equivalence of ergodicity of T with respect to m and (Y, C , n, S), their product consists of their
and the equality of time and space averages for completed product measure space with the trans-
L1(m) functions no longer holds. Thus, in general formation T  S : X  Y ! X  Y defined by (T 
m ergodic does not imply that S)(x, y) ¼ (Tx, Sy) for all (x, y)  X  Y. Neither
ergodicity nor transitivity is in general preserved
1
n
1 by taking products, for example, the product of an
lim f ∘ T j ðxÞ ¼ f dm m a:e: x  X irrational rotation on the unit circle with itself is
n!1 n
i¼0 X
not ergodic. For a list of which mixing properties
are preserved under the taking of products, see
for all f  L1(m). In the example of the intermit-
(Walters 1982a). Given any countable family of
tent map with γ  (1, 2), the orbit of Lebesgue
measure-preserving transformations on probabil-
almost every x  X is dense in X, yet the fraction
ity spaces, their direct product is defined similarly.
of time spent near the indifferent fixed point x ¼
0 tends to one for Lebesgue almost every x  X.
Factors
In fact, it may be shown (Aaronson 1997,
We say that a measure-preserving system (Y, C , n,
Section 2.4) that when m (x) ¼ 1 there are no
S) is a factor of a measure-preserving system (X,
constants an > 0 such that
ℬ, m, T ) if (possibly after deleting a set of measure
n
1
0 from X) there is a measurable onto map f :
1 X ! Y such that
lim f ∘ T j ðxÞ ¼ f dm m a:e: x  X
n!1 an
i¼0 X

f
1 C  ℬ,
Nevertheless, it is sometimes possible to obtain fT ¼ Sf, and ð8Þ
distributional limits, rather than almost sure
1
mT ¼ v:
limits, of Birkhoff sums under suitable normali-
zation. We refer the reader to Aaronson’s book
For Lebesgue spaces, factors of (X, ℬ, m, T )
(Aaronson 1997) for more details.
correspond perfectly with T-invariant complete
sub-s-algebras of ℬ. According to Rokhlin’s the-
ory of Lebesgue spaces (Rohlin 1952) (see the
Constructions article on Measure-Preserving Systems), factors
also correspond perfectly to certain kinds of par-
We give examples of some of the standard con- titions of X. A factor map f : X ! Y between
structions in dynamical systems. Often these con- Lebesgue spaces is an isomorphism if and only
structions appear in modeling situations (for if it has a measurable inverse, or equivalently f
1
example, skew products are often used to model C ¼ ℬ up to sets of measure 0.
24 Ergodic Theory: Basic Examples and Constructions

Skew Products a random composition of matrices from a group of

If (X, ℬ, m, T ) is a measure-preserving system, (Y, matrices (or more generally from a set of matrices
C , n) is a measure-space, and {Sx : x  X} is a which may form a group or not).
family of measure-preserving maps Y ! Y such
that the map that takes (x, y) to Sxy is jointly Induced Transformations
measurable in the two variables x and y; then we Since by the Poincaré Recurrence Theorem a
may define a skew product system consisting of measure-preserving transformation (T, X, m, ℬ)
the product measure space of X and Y equipped on a probability space is recurrent, given any set
with product measure m  n together with the B of positive measure, the return-time function
measure-preserving map T ⋉ S : X  Y ! X  Y
defined by nB ðxÞ ¼ inf fn  1 : T n x  Bg ð10Þ

ðT ⋉ S Þðx, yÞ ¼ ðTx, Sx yÞ: ð9Þ is finite m a.e. We may define the first-return map
by
The space Y is called the fiber of the skew
product and the space X the base. Sometimes in T B x ¼ T nB ðxÞ x: ð11Þ
the literature, the word skew product has a more
general meaning and refers to the structure Then (after perhaps discarding as usual a set of
(T ⋉ S)(x, y) ¼ (Tx, Sxy) (without any assumption measure 0) TB : B ! B is a measurable transfor-
of measure-preservation), where the action of the mation which preserves the probability measure
map on the fiber Y is determined or “driven” by the mB ¼ m/m (B). The system (B, ℬ \ B, mB, TB) is
map T : X ! X. called an induced, first-return, or derived trans-
Some common examples of skew products formation. If (T, X, m, ℬ) is ergodic, then
include the following: (B, ℬ \ B, mB, TB) is ergodic, but the converse
is not in general true.
Random Dynamical Systems The construction of the transformation TB
Suppose Sx is considered a (random) choice of a allows us to represent the forward orbit of points
mapping Y ! Y from the set {Sx : x  X}. We in B via a tower or skyscraper over B. For each
suppose T : X ! X to be the full shift. Then the n ¼ 1, 2, . . . , let
projection onto Y of the orbits of (Tx, Sxy) give the
orbits of a point y  Y under a random composi- Bn ¼ fx  B : nB ðxÞ ¼ ng: ð12Þ
tion of maps ST n x ∘ . . . ∘STx ∘Sx . More generally,
we could consider the choice of maps Sx that are Then {B1, B2, . . . } form a partition of B, which
composed to come from any ergodic dynamical we think of as the bottom floor or base of the
system, (T, X, m) to model the effect of perturba- tower. The next floor is made up of TB2, TB3, . . . ,
tions by a stationary ergodic “noise” process. which form a partition of TB \ B, and so on. All
these sets are disjoint. A column is a part of the
Group Extensions of Dynamical Systems tower of the form Bn [ TBn [ . . . [ T n
1Bn for
Suppose Y is a group, n is a measure on Y invariant some n ¼ 1, 2, . . . . The action of T on the entire
under a left group action, and Sxy : ¼ g(x)y is tower is pictured as mapping each x not at the top of
given by a group-valued function g : X ! Y. In its column straight up to the point Tx above it on the
this setting, g is often called a cocycle, since upon next level, and mapping each point on the top level
defining g(n)(x) by (T ⋉ S)(n)(x, y) ¼ (T nx, g(n)(x)y) to T nB x  B . An equivalent way to describe the
we have a cocycle relation, namely g(mþn)(x) ¼ transformation on the tower is to write for each
g(m)(T nx)g(n)(x). Group extensions arise often in n and j < n, T jBn as {(x, j) : x  Bn}, and then
modeling systems with symmetry (Field and the transformation F on the tower becomes
Nicol 2004). Common examples are provided by
Ergodic Theory: Basic Examples and Constructions 25

ðx, l þ 1Þ if l < nB ðxÞ
1; In this setting, if B R(x)dm < 1, we may

Fðx, lÞ ¼ define an F-invariant probability measure on Δ
T nB ðxÞ x, 0 if l ¼ nB ðxÞ
1:
as CðR,m BÞ  dl, where dl is counting measure and
If T preserves a measure m, then F preserves C(R, B) is the normalizing constant C(R, B) ¼
m  dl, where l is counting measure so that the m(B) B R(x)dm. This viewpoint is connected with
measure m  dl can be naturally lifted to the tower. the construction of systems by cutting and
Sometimes, the process of inducing yields an stacking – see section “Cutting and Stacking.”
induced map which is easier to analyze (perhaps it
has stronger hyperbolicity properties) than the Suspension Flows
original system. Sometimes also, it is possible to The tower construction has an analogue in which
“lift” ergodic or statistical properties from an the height function R takes values in ℝ rather than
induced system to the original system, so the ℕ. Such towers are commonly used to model
process of inducing plays an important role in dynamical systems with continuous time parame-
the study of statistical properties of dynamical ter. Let (T, X, m) be a measure-preserving system
systems (Melbourne and Török 2004). and R : X ! (0, 1) a measurable “ceiling” func-
It is possible to generalize the tower construc- tion on X. The set
tion and relax the condition that nB(x) is the first-
return time function. We may take a measurable XR ¼ fðx, tÞ : 0  RðxÞ < tg, ð13Þ
set B  X of positive m measure and define for
almost every point x  B a height or ceiling with measure n given locally by the product of m
function R : B ! ℕ and take a countable partition on X with Lebesgue measure m on ℝ, is a measure
{Xn} of B into the sets on which R is constant. We space in a natural way. If m is a finite measure and
define the tower as the set Δ : ¼ {(x, l) : x  B, R is integrable with respect to m, then n is a finite
0  l < R(x)} and the tower map F : Δ ! Δ by measure. We define an action of ℝ on XR by letting
each point x flow at unit speed up the vertical lines
ðx, l þ 1Þ if l < RðxÞ
1; {(x, t) : 0  t < R(x)} under the graph of R until it
Fðx, lÞ ¼ RðxÞ
T x, 0 if l ¼ RðxÞ
1: hits the ceiling, then jump to Tx, and so on. More
precisely, defining Rn(x) ¼ R(x) þ    þ R(T nx),

ðx, s þ tÞ if 0  s þ t < RðxÞ,

ðTx, s þ t
f ðxÞÞ if RðxÞ  s þ t < RðxÞ þ RðTxÞ
T s ðx, tÞ ¼ ð14Þ
...
T n x, s þ t
RðxÞ þ . . . þ R T n
1 x if Rn
1 ðxÞ  s þ t < Rn ðxÞ:

Ergodicity of (T, X, m) implies the ergodicity of is isomorphic to one constructed by cutting and
(Ts, XR, n). stacking. We could mention especially the von
Neumann-Kakutani adding machine (or 2-odom-
Cutting and Stacking eter) (section “Adding Machines”), the Chacon
Many of the most interesting examples in ergodic weakly mixing but not strongly mixing system
theory have been constructed by this method; in (section “Chacon System”), Ornstein’s mixing
fact, because of Rokhlin’s Lemma (see section rank one examples (see Nadkarni 1998a, p. 160
“Rokhlin’s Lemma”) every ergodic measure- ff.), and many more.
preserving transformation on a Lebesgue space
26 Ergodic Theory: Basic Examples and Constructions

We construct a Lebesgue measure-preserving edges into each vertex, and then X is partially
transformation T on an interval X (bounded or ordered as follows: x and y are comparable if
maybe unbounded) by defining it as a translation they agree from some point on, in which case we
on each of a pairwise disjoint countable collection say that x < y if at the last level n, where they
of subintervals. The construction proceeds by traverse different edges, the edge xn of x is smaller
stages, at each stage defining T on an additional than the edge yn of y. A map T is defined by letting
part of X, until eventually T is defined a.e. Tx be the smallest y that is larger than x, if there is
At each stage, X is represented as a tower, one. In nice situations, T is a homeomorphism
which is defined to be a disjoint union of columns. after defining it and its inverse on perhaps count-
A column is defined to be a finite disjoint union of ably many maximal and minimal elements.
intervals of equal length, which are numbered Invariant measures can sometimes be defined by
from 0, for the “floor,” to the last one, for the assigning weights to edges, which are then multi-
“roof,” and which we picture as lying each plied to define the measure of each cylinder set.
above the preceding-numbered one. T is defined This is a nice combinatorial way to present the
on each level of a column (i.e., each interval in the cutting and stacking method of constructing m.p.
column) except the roof by mapping it by transla- t.’s, allows for more convenient analysis of ques-
tion to the next higher interval in the column. tions such as orbit equivalence, and leads to the
At stage 0, we have just one column, consisting construction of many interesting examples, such
of all of X as the floor, and T is not defined as those based on the Pascal or Euler graphs
anywhere. To pass from one stage to the next, (Bailey et al. 2006; Frick and Petersen; Méla and
the columns are cut and stacked. This means that Petersen 2005). Odometers and generalizations
each column is divided, by vertical cuts, into a are natural examples of adic systems. Vershik
disjoint union of subcolumns of equal height (but showed that in fact every ergodic measure-
maybe not equal width), and then some of these preserving transformation on a Lebesgue space
subcolumns are stacked above others (of the same is isomorphic to a uniquely ergodic adic transfor-
width) so as to form a new tower. This allows the mation. See (Vershik and Livshits 1992).
definition of T to be extended to some parts of
X that were previously tops of towers, since they Rokhlin’s Lemma
now may have levels above them (Sometimes The following result is the fundamental starting
columns of height 1 are thought of as forming a point for many constructions in ergodic theory,
reservoir for “spacers” to be inserted between from representing arbitrary systems in terms of
subcolumns that are being stacked). If the measure cutting and stacking or adic systems, to
of the union of the tops of the columns tends to 0, constructing useful partitions and symbolic cod-
eventually T becomes defined a.e. This descrip- ings of abstract systems, to connecting conver-
tion in words can be made precise with cumber- gence theorems in abstract ergodic theory with
some notation, but the process can also be given a those in harmonic analysis. It allows us to picture
neater graphical description, which we sketch in arbitrarily long stretches of the action of a
the next section. measure-preserving transformation as a transla-
tion within the set of integers. In the ergodic
Adic Transformations nonatomic case, the statement follows readily
A.M. Vershik has introduced a family of models, from the construction of derivative
called adic or Bratteli-Vershik transformations, transformations.
into ergodic theory and dynamical systems. One
begins with a graph which is arranged in levels, Lemma 5.1 (Rokhlin’s Lemma) Let T : X ! X
finitely many vertices on each level, with connec- be a measure-preserving transformation on a
tions only from each level to the adjacent ones. probability space (X, ℬ, m). Suppose that (X, ℬ,
The space X consists of the set of all infinite paths m) is nonatomic and T : X ! X is ergodic, or, more
in this graph; it is a compact metric space in a generally, (T, X, ℬ, m) is aperiodic: that is to say,
natural way. We are given an order on the set of in the set {x  X : there is n  ℕ such that T nx ¼
Ergodic Theory: Basic Examples and Constructions 27

x} of periodic points has measure 0. Then given Because fji πj ¼ πi for all j  i, the p
1j ℬj are
n  ℕ and ϵ > 0, there is a measurable set B  X increasing, and so their union is an algebra. The
such that the sets B, TB, . . . , T n
1B are pairwise set function m can, with some difficulty, be shown
disjoint and m \n
1
k¼0 T B > 1
ϵ.
k
to be countably additive on this algebra: Since we
are dealing with Lebesgue spaces, by means of
Inverse Limits measure-theoretic isomorphisms it is possible to
Suppose that for each i ¼ 1, 2, . . . we have a replace the entire situation by compact metric
Lebesgue probability space (Xi, ℬi, mi) and a spaces and continuous maps, then use regularity
measure-preserving transformation Ti : Xi ! Xi. of the measures involved – see (Parthasarathy
Suppose also that for each i  j there is a factor 2005, p. 137 ff.). Thus, by Carathéodory’s Theo-
map fji : (Tj, Xj, ℬj, mj) ! (Ti, Xi, ℬi, mi,), such that rem (see the article on Measure-Preserving Sys-
each fjj is the identity on Xj and fji fkj ¼ fki tems), m extends to all of ℬ.
whenever k  j  i. Let Define T : X ! X by T(xj) ¼ (Tjxj). Then (T, X,
ℬ, m) is a measure-preserving system which has
X ¼ x  P1 all the (Tj, Xj, ℬj, mj) as factors, and any system
i¼1 X i : fji xj ¼ xi for all j  i :
that factors onto all the (Tj, Xj, ℬj, mj) also factors
ð15Þ onto (T, X, ℬ, m).

For each j, let πj : X ! Xj be the projection Natural Extension

defined by πjx ¼ xj. The natural extension is a way to produce an
Let ℬ be the smallest s-algebra of subsets of invertible system from a noninvertible system.
X which contains all the p
1
j ℬj . Define m on each The original system is a factor of its natural exten-
p
1
j ℬj by sion, and its orbit structure and ergodic properties
are captured by the natural extension, as will be
m p
1 seen from its construction. Let (T, X, ℬ, m) be a
j B ¼ mj ðBÞ for all B  ℬj : ð16Þ
measure-preserving transformation of a Lebesgue
probability space. Define

O ≔fðx0 , x1 , x2 , . . .Þ : xn ¼ T ðxnþ1 Þ, xn  X, n ¼ 0, 1, 2, . . .g

with s : Ω ! Ω defined by s ((x0, x1, x2, . . .)) ¼ invariant measure for the natural extension m on
(T(x0), x0, x1, . . .). The map s is invertible on Ω. Ω by defining it first on cylinder sets C(A0, A1, . . . ,
Given the invariant measure m, we define the Ak) by

mðCðA0 , A1 , . . . , Ak ÞÞ ¼ m T
k ðA0 Þ \ T
kþ1 ðA1 Þ . . . \ T
kþi ðAi Þ \ . . . \ Ak

and then extending it to Ω using Kolmogorov’s . . .) ¼ x0, then π ∘ sn(x0, . . ., xn, . . .) ¼ T n(x0) for
extension theorem. We think of (x0, x1, x2, . . .) as all x0 and thus the natural extension yields all
being an inverse branch of x0  X under the information about the orbits of X under T.
mapping T : X ! X. The maps s, s
1 : Ω ! Ω The natural extension is an inverse limit. Let
are ergodic with respect to m if (T, X, ℬ, m) is (X, ℬ, m) be a Lebesgue probability space and T :
ergodic (Walters 1982a). If π : Ω ! X is projec- X ! X a map such that T
1ℬ  ℬ and mT
1 ¼ m.
tion onto the first component, i.e., π (x0, . . ., xn, For each i ¼ 1, 2, . . . let (Ti, Xi, ℬi, mi) ¼ (T, X, ℬ,
28 Ergodic Theory: Basic Examples and Constructions

m), and fji ¼ T j
i for each j > i. Then the inverse properties of dynamical systems will continue to
limit T, X, ℬ, m of this system is an invertible be an active research area for the foreseeable
future.
measure-preserving system which is the natural
The directions will include, among others, the
extension of (T, X, ℬ, m). We have
following:

1
T ðx1 , x2 , . . .Þ ¼ ðx2 , x3 , . . .Þ: ð17Þ 1. Establishing statistical and ergodic properties
under weakened dependence assumptions.
The original system (T, X, ℬ, m) is a factor of There is a “hierarchy” of probabilistic limit
T, X, ℬ, m (using any πi as the factor map), and theorems including, among others, ergodicity
any factor mapping from an invertible system (akin to the strong law of large numbers for
onto (T, X, ℬ, m) consists of a factor mapping integrable observations); central limit theorem
onto T, X, ℬ, m followed by projection onto (distributional convergence to a Gaussian for
the first coordinate. the scaled Birkhoff sums of an observable with
finite second moment); law of the iterated log-
Joinings arithm (almost sure rate of growth of scaled
Given measure-preserving systems (T, X, ℬ, m) Birkhoff sums); and the almost sure invariance
and (S, Y, C , n), a joining of the two systems is a invariance principle (a strong form of approx-
T  S-invariant measure P on their product mea- imation by Brownian motion). Some observ-
surable space that projects to m and n, respectively, ables on some systems exhibit the same limit
under the projections of X  Y to X and Y, respec- laws as iid processes, for example, Hölder
tively. That is, if π1 : X  Y ! X is the projection observables on smooth hyperbolic systems sat-
onto the first component, i.e., π1(x, y) ¼ x, then isfy the almost sure invariance principle. In
P p
11 ðAÞ ¼ mðAÞ for all A  ℬ and similarly
other settings, the determinism of the system
for π2 : X  Y ! Y. plays a key role, for example, the return time
This concept is the ergodic-theoretic version of statistics of a system to a periodic orbit may
the notion in probability theory of a coupling. The best be described by a compound Poisson pro-
product measure m  n is always a joining of the cess rather than a Poisson law (Hirata 1993).
two systems. If product measure is the only join- An important strand of research is determining
ing of the two systems, then we say that they are conditions, usually dynamical and mixing con-
disjoint and write X ⊥ Y (Furstenberg 1967). If D ditions, on a system to determine the form of
is any family of systems, we write D ⊥ for statistics that observables on the system will
the family of all measure-preserving systems display. This research has also produced a rich
which are disjoint from every system in D: Exten- class of counterexamples. A good reference is
sive recent accounts of the use of joinings in (Melbourne and Török 2004).
ergodic theory are in (Glasner 2003a; Rudolph 2. The study of systems which display “anoma-
1990b; Thouvenot 1995a). lous statistics.” By “anomalous statistics” is
usually meant non-Gaussian limit laws such
as the convergence of a scaled observable to a
Future Directions stable law rather than a Gaussian. This arises
basically in two ways in a dynamical system:
The basic examples and constructions presented (1) a nonintegrable observable on a fast mixing
here are idealized, and many of the underlying system; (2) a regular (usually Hölder) observ-
assumptions (such as uniform hyperbolicity) are able on a slowly mixing dynamical system. For
seldom satisfied in applications, yet they have example, scaled Birkhoff sums of a non-
given important insights into the behavior of integrable observable such as d(x, x0)
1 on a
real-world physical systems. The ergodic rapidly mixing system such as the doubling
map will converge to a stable law as will scaled
Ergodic Theory: Basic Examples and Constructions 29

Birkhoff sums of a Hölder observable on a Lebesgue measure, sometimes occur. An influ-

slowly mixing intermittent map with non- ential work in this area is (Conze and Raugi
summable decay of correlations (Gouëzel 2007).
2004). 6. The ergodic theory of infinite-dimensional sys-
3. The study of the stability and typicality of tems and relations to partial differential equa-
ergodic behavior and mixing in dynamical sys- tions (Young 2017).
tems. This topic is often called “stable ergodic- 7. Advances in number theory (see the sections
ity.” Ergodicity and other mixing and statistical on Szemerédi and Ramsey theory); research
properties are often stable to perturbation of the into models with nonsingular rather than
underlying parameters of a dynamical system. invariant measures.
This strand of research has produced many 8. Infinite-measure systems. In this setting, ergo-
surprising examples of stability. For example, dic properties are considered with respect to a
skew-products of a hyperbolic dynamical sys- s-finite measure rather than a probability mea-
tem with a circle S1 fiber are typically stable to sure. An excellent reference for this topic is
perturbation. A good review is (Burns (Aaronson 1997).
et al. 2001).
4. The study of statistical recurrence, which is Other chapters in this Encyclopedia discuss in
closely related to extremal statistics on a more detail these and other topics in which future
dynamical system. One theme of research research directions will be active and fruitful.
into the statistical properties of mixing dynam-
ical systems is that limit laws for observables
on the system tend to resemble those of iid Bibliography
processes. However, this viewpoint breaks
down when considering, for example, return Primary Literature
times to a periodic point or the extreme value Aaronson J (1997) An introduction to infinite ergodic
theory, Mathematical Surveys and Monographs, vol.
statistics of observables maximized at a peri-
50. American Mathematical Society, Providence,
odic point. Such a situation leads to clustering MR1450400 (99d:28025)
of returns or extrema. This study is important Anosov DV (1967) Geodesic flows on closed Riemannian
for applications of dynamical systems theory manifolds of negative curvature. Trudy Mat Inst
Steklov 90:209 (Russian). MR0224110 (36 #7157)
to, for example, climate and population
Arnol’d VI (1963) Small denominators and problems of
models. A recent review is (Boyarsky and stability of motion in classical and celestial mechanics.
Góra 1997a). UspehiMat Nauk 18(114):91–192 (Russian).
5. Sequential and nonstationary dynamical sys- MR0170705 (30 #943)
Bailey S, Keane M, Petersen K, Salama IA (2006) Ergo-
tems. In physical systems, a common situation
dicity of the adic transformation on the Euler graph.
is where the rules governing the time evolution Math Proc Cambridge Philos Soc 141(2):231–238.
of the system change over time. This may be MR2265871 (2007m:37010)
modeled by a sequential dynamical system in Baladi V (2000a) Positive transfer operators and decay of
correlations, Advanced Series in Nonlinear Dynamics,
which the maps or flows applied to the system
vol 16. World Scientific Publishing Co. Inc., River
change over time. The term sequential dynam- Edge. MR1793194 (2001k:37035)
ical system was coined by Berend and Benedicks M, Carleson L (1985) On iterations of 1 – ax(X,
Bergelson (1984), but the idea predates this ℬ, m)2 on(
1, 1), Ann Math 122:1, 1–25. MR799250
(87c:58058)
work. This strand of work is important in the
Benedicks M, Carleson L (1991) The dynamics of the
study of fluid mixing. In this setting, if the Hénon map. Ann Math 133(1):73–169. MR1087346
sequential maps or flows are hyperbolic/ (92d:58116)
mixing then “self-norming” statistical proper- Benedicks M, Young L-S (1993) Sinaĭ-Bowen-Ruelle
measures for certain Hénon maps. Invent Math
ties may occur. Although there is usually not an 112(3):541–576. MR1218323 (94e:58074)
invariant measure, statistical properties with Berend D, Bergelson V (1984) Ergodic and mixing
respect to a reference measure, such as sequences of transformations. Ergodic Theory Dynam
30 Ergodic Theory: Basic Examples and Constructions

Syst 4(3):353–366. https://doi.org/10.1017/ Chernov N, Markarian R (2006a) Chaotic billiards, Math-

S0143385700002509. MR776873 ematical Surveys and Monographs, vol 127. American
Bertrand-Mathis A (1986) Développement en base θ; Mathematical Society, Providence. MR2229799
répartition modulo un de la suite (xθ (X, ℬ, m)n)n0; (2007f:37050)
langages codés et θ-shift. Bull Soc Math France 114(3): Collet P, Eckmann J-P (1980a) Iterated maps on the inter-
271–323 (French, with English summary). MR878240 val as dynamical systems, Progress in Physics,
(88e:11067) vol 1. Birkhäuser, Boston. MR613981 (82j:58078)
Billingsley P (1978a) Ergodic theory and information. Conze J-P, Raugi A (2007) Limit theorems for sequential
Robert E. Krieger Publishing Co., Huntington. Reprint expanding dynamical systems on [0, 1]. In: Ergodic
of the 1965 original. MR524567 (80b:28017) theory and related fields, Contemporary Mathematics,
Blanchard F (1989) β-expansions and symbolic dynamics. vol. 430, Am Math Soc, Providence, pp. 89–121,
Theor Comput Sci 65(2):131–141. MR1020481 https://doi.org/10.1090/conm/430/08253, (to appear in
(90j:54039) print). MR2331327
Blanchard F, Hansel G (1986a) Systèmes codés. Theor Cornfeld IP, Fomin SV, Sinaĭ YG (1982a) Ergodic theory,
Comput Sci 44(1):17–49. (French, with English sum- Grundlehren der Mathematischen Wissenschaften
mary). MR858689 (88m:68029) [Fundamental Principles of Mathematical Sciences],
Blanchard F, Hansel G (1986b) Systèmes codés et limites vol. 245, Springer-Verlag, New York. Translated from
de systèmes sofiques. C R Acad Sci Paris Sér I Math the Russian by A. B. Sosinskiĭ. MR832433 (87f:28019)
303(10):475–477. (French, with English summary). Coven EM, Hedlund GA (1973) Sequences with minimal
MR865864 (87m:94009) block growth. Math Syst Theory 7:138–153.
Blanchard F, Hansel G (1991) Sofic constant-to-one exten- MR0322838 (48 #1199)
sions of subshifts of finite type. Proc Am Math Soc de la Rue T (1993) Entropie d’un système dynamique
112(1):259–265. MR1050016 (91m:54050) gaussien: cas d’une action de Z(X, ℬ, m)d. C R Acad
Boshernitzan MD (1992) Billiards and rational periodic Sci Paris Sér I Math 317(2):191–194 (French, with
directions in polygons. Am Math Monthly 99(6): English and French summaries). MR1231420
522–529. MR1166001 (93d:51043) (94c:28022)
Boshernitzan M, Galperin G, Krüger T, Troubetzkoy de Melo W (1973) Structural stability of diffeomorphisms
S (1998) Periodic billiard orbits are dense in rational on two-manifolds. Invent Math 21:233–246.
polygons. Trans Am Math Soc 350(9):3523–3535. MR0339277 (49 #4037)
MR1458298 (98k:58179) de Melo W, van Strien S (1993) One-dimensional dynam-
Bowen R (1970) Markov partitions for Axiom ics, Ergebnisse der Mathematik und ihrer Grenzgebiete
A diffeomorphisms. Am J Math 92:725–747. (3) [Results in Mathematics and Related Areas (3)],
MR0277003 (43 #2740) vol 25. Springer-Verlag, Berlin. MR1239171
Bowen R (1975) Equilibrium states and the ergodic theory (95a:58035)
of Anosov diffeomorphisms, Lecture Notes in Mathe- del Junco A (1978) A simple measure-preserving transfor-
matics, vol 470. Springer-Verlag, Berlin. MR0442989 mation with trivial centralizer. Pacific J Math 79(2):
(56 #1364) 357–362. MR531323 (80i:28034)
Boyarsky A, Góra PL (1997a) Laws of chaos, probability del Junco A, Rahe M, Swanson L (1980) Chacon’s auto-
and its applications. Birkhäuser Boston Inc., Boston. morphism has minimal self-joinings. J Anal Math 37:
Invariant measures and dynamical systems in one 276–284. MR583640 (81j:28027)
dimension. MR1461536 (99a:58102) Downarowicz T (2005) Survey of odometers and Toeplitz
Brin MI, Pesin JB (1974) Partially hyperbolic dynamical flows, Algebraic and topological dynamics. In: Con-
systems. Izv Akad Nauk SSSR Ser Mat 38:170–212. temporary Mathematics, vol. 385, Am Math Soc, Prov-
(Russian). MR0343316 (49 #8058) idence, pp. 7–37. MR2180227 (2006f:37009)
Bunimovič LA (1974) The ergodic properties of certain Feigenbaum MJ (1978) Quantitative universality for a
billiards. Funkcional Anal i Priložen 8(3):73–74. class of nonlinear transformations. J Statist Phys
(Russian). MR0357736 (50 #10204) 19(1):25–52. MR0501179 (58 #18601)
Bunimovich LA (1979) On the ergodic properties of Field M, Nicol M (2004) Ergodic theory of equivariant
nowhere dispersing billiards. Commun Math Phys diffeomorphisms: Markov partitions and stable ergo-
65(3):295–312. MR530154 (80h:58037) dicity. Mem Am Math Soc 169(803):viii+100.
Burns K, Pugh C, Shub M, Wilkinson A (2001) Recent MR2045641 (2005g:37041)
results about stable ergodicity, Smooth ergodic theory Frick SB, Petersen K Random permutations and unique
and its applications (Seattle, WA, 1999). In: Proceed- fully supported ergodicity for the Euler adic transfor-
ings of Symposia in Pure Mathematics, vol. 69, Am mation. Ann Inst H Poincaré Prob Stat to appear
Math Soc, Providence, pp. 327–366. MR1858538 Furstenberg H (1967) Disjointness in ergodic theory, min-
(2002m:37042) imal sets, and a problem in Diophantine approximation.
Carleson L, Gamelin TW (1993a) Complex dynamics, Math Syst Theory 1:1–49. MR0213508 (35 #4369)
Universitext: Tracts in Mathematics. Springer-Verlag, Furstenberg H (1973) The unique ergodicity of the
New York. MR1230383 (94h:30033) horocycle flow. Recent advances in topological
Ergodic Theory: Basic Examples and Constructions 31

dynamics (Proc. Conf., Yale Univ., New Haven, Conn., vol. 1222, Springer-Verlag, Berlin. MR872698
1972; in honor of Gustav Arnold Hedlund), Springer, (88k:58075)
Berlin, pp. 95–115. Lecture Notes in Math, vol. 318. Keane M (1968) Generalized Morse sequences.
MR0393339 (52 #14149) Z Wahrscheinlichkeitstheorie Verw Gebiete 10:
Gallavotti G, Ornstein DS (1974) Billiards and Bernoulli 335–353. MR0239047 (39 #406)
schemes. Commun Math Phys 38:83–101. Keane M (1977) Non-ergodic interval exchange transfor-
MR0355003 (50 #7480) mations. Israel J Math 26(2):188–196. MR0435353
Glasner E (2003a) Ergodic Theory via Joinings, Mathe- (55 #8313)
matical Surveys and Monographs, vol 101. American Keller G, Nowicki T (1992) Spectral theory, zeta functions
Mathematical Society, Providence. MR1958753 and the distribution of periodic points for Collet-
(2004c:37011) Eckmann maps. Commun Math Phys 149(1):31–69.
Gouëzel S (2004) Central limit theorem and stable laws for MR1182410 (93i:58123)
intermittent maps. Probab Theory Relat Fields 128(1): Kerckhoff S, Masur H, Smillie J (1986) Ergodicity of
82–122. https://doi.org/10.1007/s00440-003-0300-4. billiard flows and quadratic differentials. Ann Math
MR2027296 124(2):293–311. MR855297 (88f:58122)
Graczyk J, Światek G (1997) Generic hyperbolicity in the Kolmogorov AN (1954) On conservation of conditionally
logistic family. Ann Math 146(1):1–52. MR1469316 periodic motions for a small change in Hamilton’s
(99b:58079) function. Dokl Akad Nauk SSSR (NS) 98:527–530.
Guckenheimer J, Holmes P (1990) Nonlinear oscillations, (Russian). MR0068687 (16,924c)
dynamical systems, and bifurcations of vector fields, Krieger W (2000) On subshifts and topological Markov
Applied Mathematical Sciences, vol. 42. Springer- chains, Numbers, information and complexity
Verlag, New York. Revised and corrected reprint of (Bielefeld, 1998). Kluwer Academic Publishers, Bos-
the 1983 original. MR1139515 (93e:58046) ton, pp. 453–472. MR1755380 (2001g:37010)
Grayson M, Pugh C, Shub M (1994) Stably ergodic Lagarias JC (1991) The Farey shift, manuscript
diffeomorphisms. Ann Math 140(2):295–329. Lagarias JC (1992) Number theory and dynamical systems.
MR1298715 (95g:58128) In: The unreasonable effectiveness of number theory
Hasselblatt B, Katok A (2003a) A first course in dynamics. (Orono, ME, 1991), Proceedings of Symposia in Pure
Cambridge University Press, New York. With a panorama Mathematics, vol. 46, American Mathematical Society,
of recent developments. MR1995704 (2004f:37001) Providence, pp. 35–72. MR1195841 (93m:11143)
Hayashi S (1997) Connecting invariant manifolds and the Lazutkin VF (1973) Existence of caustics for the billiard
solution of the C(X, ℬ, m)1 stability and Ω-stability problem in a convex domain. Izv Akad Nauk SSSR Ser
conjectures for flows. Ann Math 145(1):81–137. Mat 37:186–216. (Russian). MR0328219 (48 #6561)
MR1432037 (98b:58096) Ledrappier F (1984) Propriétés ergodiques des mesures de
Hedlund GA (1934) On the metrical transitivity of the Sinaï. Inst Hautes Études Sci Publ Math 59:163–188.
geodesics on closed surfaces of constant negative cur- (French). MR743818 (86f:58092)
vature. Ann Math. 35(4):787–808. MR1503197 Lind D, Marcus B (1995) An introduction to symbolic
Hirata M (1993) Poisson law for Axiom A diffeomorphisms. dynamics and coding. Cambridge University Press,
Ergodic Theory Dyn Syst 13(3):533–556. https://doi.org/ Cambridge. MR1369092 (97a:58050)
10.1017/S0143385700007513. MR1245828 Liverani C, Saussol B, Vaienti S (1999) A probabilistic
Hopf E (1939) Statistik der geodätischen Linien in approach to intermittency. Ergodic Theory Dyn Syst
Mannigfaltigkeiten negative Krümmung. Ber Verh 19(3):671–685. MR1695915 (2000d:37029)
Sächs Akad Wiss Leipzig 91:261–304. (German). Lyons R (1988) On measures simultaneously 2- and
MR0001464 (1,243a) 3-invariant. Israel J Math 61(2):219–224. MR941238
Host B (1995) Nombres normaux, entropie, translations. (89e:28031)
Israel J Math 91(1–3):419–428. (French, with English Liverani C, Wojtkowski MP (1995) Ergodicity in Hamil-
summary). MR1348326 (96g:11092) tonian systems, Dynamics reported. Dynam. Report.
Huyi H (2004) Decay of correlations for piecewise smooth Expositions Dynam. Systems (N.S.), vol. 4, Springer,
maps with indifferent fixed points. Ergodic Theory Dyn Berlin, pp. 130–202. MR1346498 (96g:58144)
Syst 24(2):495–524. MR2054191 (2005a:37064) Manneville P, Pomeau Y (1980) Different ways to turbu-
Jakobson MV (1981) Absolutely continuous invariant lence in dissipative dynamical systems. Phys D 1(2):
measures for one-parameter families of one- 219–226. MR581352 (81h:58041)
dimensional maps. Commun Math Phys 81(1):39–88. Mañé R (1988) A proof of the C(X, ℬ, m)1 stability
MR630331 (83j:58070) conjecture. Inst Hautes Études Sci Publ Math 66:
Katok A (1980) Lyapunov exponents, entropy and periodic 161–210. MR932138 (89e:58090)
orbits for diffeomorphisms. Inst Hautes Études Sci Publ Mañé R (1987a) Ergodic theory and differentiable dynam-
Math 51:137–173. MR573822 (81i:28022) ics. Ergebnisse der Mathematik und ihrer Grenzgebiete
Katok A, Strelcyn J-M, Ledrappier F, Przytycki F (1986) (3) [Results in Mathematics and Related Areas (3)],
Invariant manifolds, entropy and billiards; smooth vol. 8, Springer-Verlag, Berlin, Translated from the
maps with singularities, Lecture Notes in Mathematics, Portuguese by Silvio Levy. MR889254 (88c:58040)
32 Ergodic Theory: Basic Examples and Constructions

Marcus B (1978) The horocycle flow is mixing of all Phillips E, Varadhan S (eds) (1975) Ergodic theory.
degrees. Invent Math 46(3):201–209. MR0488168 Courant Institute of Mathematical Sciences New York
(58 #7731) University, New York. A seminar held at the Courant
Marcus B (1975) Unique ergodicity of the horocycle flow: Institute of Mathematical Sciences, New York Univer-
variable negative curvature case. Israel J Math sity, New York, 1973–1974; With contributions by
21(2–3):133–144. Conference on Ergodic Theory and S. Varadhan, E. Phillips, S. Alpern, N. Bitzenhofer
Topological Dynamics (Kibbutz Lavi, 1974). and R. Adler. MR0486431 (58 #6177)
MR0407902 (53 #11672) Pugh C, Shub M (2004) Stable ergodicity. Bull Am Math
Masur H (1986) Closed trajectories for quadratic differen- Soc (N.S.) 41(1):1–41 (electronic). With an appendix
tials with an application to billiards. Duke Math J 53(2): by Alexander Starkov. MR2015448 (2005f:37011)
307–314. MR850537 (87j:30107) Robbin JW (1971) A structural stability theorem. Ann
Mayer DH (1991) Continued fractions and related trans- Math 94:447–493. MR0287580 (44 #4783)
formations. Ergodic theory, symbolic dynamics, and Robinson C (1973) C(X, ℬ, m)r structural stability implies
hyperbolic spaces (Trieste, 1989), Oxford Sci. Publ., Kupka-Smale. Dynamical systems (Proc Sympos, Univ
Oxford University Press, New York, pp. 175–222. Bahia, Salvador, 1971). Academic Press, New York,
MR1130177 pp. 443–449. MR0334282 (48 #12601)
Melbourne I, Török A (2004) Statistical limit theorems for Robinson C (1975) Errata to: “Structural stability of vector
suspension flows. Israel J Math 144:191–209. fields” (Ann Math (2) 99 (1974), 154–175). Ann Math
MR2121540 (2006c:37005) 101:368. MR0365630 (51 #1882)
Moser J (1962) On invariant curves of area-preserving Robinson C (1976) Structural stability of C(X, ℬ, m)1
mappings of an annulus. Nachr Akad Wiss Göttingen diffeomorphisms. J Differ Equ 22(1):28–73.
Math-Phys Kl II 1962:1–20. MR0147741 (26 #5255) MR0474411 (57 #14051)
Méla X, Petersen K (2005) Dynamical properties of the Rohlin VA (1952) On the fundamental ideas of measure
Pascal adic transformation. Ergodic Theory Dyn Syst theory. Am Math Soc Transl 1952(71):55. MR0047744
25(1):227–256. MR2122921 (2005k:37012) (13,924e)
Nadkarni MG (1998a) Spectral theory of dynamical sys- Rudolph DJ (1990a)  2 and  3 invariant measures and
tems. Birkhäuser Advanced Texts: Basler Lehrbücher. entropy. Ergodic Theory Dyn Syst 10(2):395–406.
[Birkhäuser Advanced Texts: Basel Textbooks], MR1062766 (91g:28026)
Birkhäuser Verlag, Basel. MR1719722 (2001d:37001) Rudolph DJ (1990b) Fundamentals of measurable dynam-
Ornstein D (1970) Bernoulli shifts with the same entropy ics. Oxford Science Publications, The Clarendon Press
are isomorphic. Adv Math 4:337–352. MR0257322 Oxford University Press, New York. Ergodic theory on
(41 #1973) Lebesgue spaces. MR1086631 (92e:28006)
Ornstein DS, Weiss B (1973) Geodesic flows are Ruelle D (1978) Thermodynamic formalism, Encyclope-
Bernoullian. Israel J Math 14:184–198. MR0325926 dia of Mathematics and its Applications, vol. 5,
(48 #4272) Addison-Wesley Publishing Co., Reading. The mathe-
Oseledec VI (1968) A multiplicative ergodic theorem. matical structures of classical equilibrium statistical
Characteristic Ljapunov, exponents of dynamical sys- mechanics; With a foreword by Giovanni Gallavotti
tems. Trudy Moskov Mat Obšč 19:179–210. (Russian). and Gian-Carlo Rota. MR511655 (80g:82017)
MR0240280 (39 #1629) Ruelle D (1976) A measure associated with axiom-A
Parry W (1966) Symbolic dynamics and transformations of attractors. Am J Math 98(3):619–654. MR0415683
the unit interval. Trans Am Math Soc 122:368–378. (54 #3763)
MR0197683 (33 #5846) Sarig O (2002) Subexponential decay of correlations.
Parry W (1996) Squaring and cubing the circle— Invent Math 150(3):629–653. MR1946554
Rudolph’s theorem. Ergodic theory of Z(X, ℬ, m)d (2004e:37010)
actions (Warwick, 1993), London Math Soc Lecture Schweiger F (1995a) Ergodic theory of fibred systems and
Note Ser, vol. 228, Cambridge University Press, Cam- metric number theory. Oxford Science Publications,
bridge, pp. 177–183. MR1411219 (97h:28009) The Clarendon Press Oxford University Press,
Parthasarathy KR (2005) Probability measures on metric New York. MR1419320 (97h:11083)
spaces. AMS Chelsea Publishing, Providence. Reprint Sinaĭ JG (1970) Dynamical systems with elastic reflec-
of the 1967 original. MR2169627 (2006d:60004) tions. Ergodic properties of dispersing billiards. Uspehi
Pesin JB (1976) Families of invariant manifolds that cor- Mat Nauk 25(2):141–192 (Russian). MR0274721
respond to nonzero characteristic exponents. Izv Akad (43#481)
Nauk SSSR Ser Mat 40(6):1332–1379, 1440 (Russian). Sinaĭ JG (1968) Markov partitions and
MR0458490 (56 #16690) U-diffeomorphisms. Funkcional Anal i Priložen 2(1):
Pesin JB (1977) Characteristic Ljapunov exponents, and 64–89. (Russian). MR0233038 (38 #1361)
smooth ergodic theory. Uspehi Mat Nauk 32 no. Sinaĭ JG (1972) Gibbs measures in ergodic theory. Uspehi
4 (196), 55–112, 287 (Russian). MR0466791 Mat Nauk 27(4):21–64 (Russian). MR0399421
(57 #6667) (53 #3265)
Ergodic Theory: Basic Examples and Constructions 33

Smale S (1980) The mathematics of time. Springer-Verlag, geometric and probabilistic perspective; Mathematical
New York. Essays on dynamical systems, economic Physics, III. MR2105774 (2005g:37001)
processes, and related topics. MR607330 (83a:01068) Boyarsky A, Góra PL (1997b) Laws of chaos, Probability
Tabachnikov S (2005) Geometry and billiards, Student and its Applications. Birkhäuser Boston Inc., Boston.
Mathematical Library, vol 30. American Mathematical Invariant measures and dynamical systems in one
Society, Providence. MR2168892 (2006h:51001) dimension. MR1461536 (99a:58102)
Thouvenot JP (1995a) Some properties and applications of Brin M, Stuck G (2002) Introduction to dynamical sys-
joinings in ergodic theory. Ergodic theory and its con- tems. Cambridge University Press, Cambridge.
nections with harmonic analysis (Alexandria, 1993), MR1963683 (2003m:37001)
London Math Soc Lecture Note Ser, vol. 205. Cam- Carleson L, Gamelin TW (1993b) Complex dynamics,
bridge University Press, Cambridge, pp. 207–235. Universitext: Tracts in Mathematics. Springer-Verlag,
MR1325699 (96d:28017) New York. MR1230383 (94h:30033)
Vershik AM, Livshits AN (1992) Adic models of ergodic Chernov N, Markarian R (2006b) Chaotic billiards, Math-
transformations, spectral theory, substitutions, and ematical Surveys and Monographs, vol 127. American
related topics, Representation theory and dynamical Mathematical Society, Providence. MR2229799
systems. Adv Soviet Math, vol. 9, American Mathe- (2007f:37050)
matical Society, Providence, pp. 185–204. Collet P, Eckmann J-P (1980b) Iterated maps on the inter-
MR1166202 (93i:46131) val as dynamical systems, Progress in Physics,
Vorobets YB, Gal’perin GA, Stëpin AM (1992) Periodic vol 1. Birkhäuser, Boston. MR613981 (82j:58078)
billiard trajectories in polygons: generation mecha- Cornfeld IP, Fomin SV, Sinaĭ YG (1982b) Ergodic Theory.
nisms. Uspekhi Mat. Nauk 47(285):9–74, 207 (Russian, Grundlehren der Mathematischen Wissenschaften
with Russian summary); English transl, Russian Math [Fundamental Principles of Mathematical Sciences],
Surveys 47(3) (1992):5–80. MR1185299 (93h:58088) vol. 245, Springer-Verlag, New York. Translated from
Walters P (1982a) An Introduction to Ergodic Theory, the Russian by A. B. Sosinskiĭ. MR832433 (87f:28019)
Graduate Texts in Mathematics, vol 79. Springer- Denker M, Grillenberger C, Sigmund K (1976) Ergodic
Verlag, New York. MR648108 (84e:28017) theory on compact spaces. Springer-Verlag, Berlin.
Young L-S (1998) Statistical properties of dynamical sys- Lecture Notes in Mathematics, vol. 527. MR0457675
tems with some hyperbolicity. Ann Math 147(3): (56 #15879)
585–650. MR1637655 (99h:58140) Friedman NA (1970) Introduction to Ergodic Theory. Van
Young L-S (1999) Recurrence times and rates of mixing. Nostrand Reinhold Co., New York. Van Nostrand
Israel J Math 110:153–188. MR1750438 Reinhold Mathematical Studies, No. 29. MR0435350
(2001j:37062) (55 #8310)
Young LS (1993) Ergodic theory of chaotic dynamical Glasner E (2003b) Ergodic Theory via Joinings, Mathe-
systems. In: From Topology to Computation: Proceed- matical Surveys and Monographs, vol 101. American
ings of the Smalefest (Berkeley, CA, 1990), Springer, Mathematical Society, Providence. MR1958753
New York, pp. 201–226. MR1246120 (94i:58112) (2004c:37011)
Young L-S (2017) Generalizations of SRB measures to Halmos PR (1960) Lectures on Ergodic Theory. Chelsea
nonautonomous, random, and infinite dimensional sys- Publishing Co., New York. MR0111817 (22 #2677)
tems. J Stat Phys 166(3–4):494–515. https://doi.org/10. Hasselblatt B, Katok A (2003b) A first course in dynamics.
1007/s10955-016-1639-0. MR3607578 Cambridge University Press, New York. With a pano-
Zemljakov AN, Katok AB (1975) Topological transitivity rama of recent developments. MR1995704
of billiards in polygons. Mat Zametki 18(2):291–300. (2004f:37001)
(Russian). MR0399423 (53 #3267) Hopf E (1937) Ergodentheorie, 1st ed., Ergebnisse der
Mathematik und ihrer Grenzgebiete 5. Bd., 2. Jaden
Springer, Berlin
Books and Reviews
Jacobs K (1965) Einige neuere Ergebnisse der
Baladi V (2000b) Positive transfer operators and decay of
Ergodentheorie. Jber Deutsch Math-Verein 67(Abt.
correlations, Advanced Series in Nonlinear Dynamics,
1):143–182 (German). MR0186789 (32 #4244)
vol 16. World Scientific Publishing Co. Inc., River
Katok A, Hasselblatt B (1995) Introduction to the modern
Edge. MR1793194 (2001k:37035)
theory of dynamical systems. In: Encyclopedia of
Billingsley P (1978b) Ergodic theory and information.
mathematics and its applications, vol. 54, Cambridge
Robert E. Krieger Publishing Co., Huntington. Reprint
University Press, Cambridge. With a supplementary
of the 1965 original. MR524567 (80b:28017)
chapter by Katok and Leonardo Mendoza.
Billingsley P (1995) Probability and measure, 3rd ed.,
MR1326374 (96c:58055)
Wiley Series in Probability and Mathematical Statis-
Keller G (1998) Equilibrium States in Ergodic Theory,
tics. John Wiley & Sons Inc., New York. A Wiley-
London Mathematical Society Student Texts, vol 42.
Interscience Publication. MR1324786 (95k:60001)
Cambridge University Press, Cambridge. MR1618769
Bonatti C, Díaz LJ, Viana M (2005) Dynamics beyond
(99e:28022)
uniform hyperbolicity, Encyclopaedia of Mathematical
Sciences, vol. 102, Springer-Verlag, Berlin. A global
34 Ergodic Theory: Basic Examples and Constructions

Lucarini V, Faranda D, Moreira ACGM, de Freitas J, Petersen K (1989) Ergodic theory, Cambridge Studies in
Milhazes M, de Freitas M, Holland TK, Nicol M, Advanced Mathematics, vol 2. Cambridge University
Todd M, Vaienti S (eds) (2016) Extremes and recur- Press, Cambridge. Corrected reprint of the 1983 origi-
rence in dynamical systems, Pure and Applied Mathe- nal. MR1073173 (92c:28010)
matics (Hoboken). John Wiley & Sons, Inc., Hoboken. Royden HL (1988) Real analysis, 3rd edn. Macmillan Pub-
MR3558780 lishing Company, New York. MR1013117 (90g:00004)
Mañé R (1987b) Ergodic Theory and Differentiable Rudolph DJ (1990c) Fundamentals of measurable dynam-
Dynamics. Ergebnisse der Mathematik und ihrer ics. Oxford Science Publications, The Clarendon Press
Grenzgebiete (3) [Results in Mathematics and Related Oxford University Press, New York, Ergodic theory on
Areas (3)], vol. 8, Springer-Verlag, Berlin. Translated Lebesgue spaces. MR1086631 (92e:28006)
from the Portuguese by Silvio Levy. MR889254 Schweiger F (1995b) Ergodic theory of fibred systems and
(88c:58040) metric number theory. Oxford Science Publications,
Nadkarni MG (1998b) Spectral theory of Dynamical Sys- The Clarendon Press Oxford University Press,
tems. Birkhäuser Advanced Texts: Basler Lehrbücher New York. MR1419320 (97h:11083)
[Birkhäuser advanced texts: Basel textbooks]. Thouvenot J-P (1995b) Some properties and applications
Birkhäuser Verlag, Basel. MR1719722 (2001d:37001) of joinings in ergodic theory. Ergodic theory and its
Ornstein DS, Rudolph DJ, Weiss B (1982) Equivalence of connections with harmonic analysis (Alexandria,
measure preserving transformations. Mem Am Math 1993), London Math Soc Lecture Note Ser, vol.
Soc 37(262):xii+116. MR653094 (85e:28026) 205, Cambridge University Press, Cambridge,
Parry W, Pollicott M (1990) Zeta functions and the periodic pp. 207–235. MR1325699 (96d:28017)
orbit structure of hyperbolic dynamics. Astérisque Walters P (1982b) An Introduction to Ergodic Theory,
187–188:268 (English, with French summary). Graduate Texts in Mathematics, vol 79. Springer-
MR1085356 (92f:58141) Verlag, New York. MR648108 (84e:28017)
 Glossary
Ergodicity and Mixing Properties
Bernoulli shift Mathematical abstraction of the
Terrence Adams1 and Anthony Quas2 scenario in statistics or probability in which
1
Department of Mathematics and Statistics, State one performs repeated independent identical
University of New York, Albany, NY, USA experiments.
2
Department of Mathematics and Statistics, Markov chain A probability model describing a
University of Victoria, Victoria, BC, Canada sequence of observations made at regularly
spaced time intervals such that at each time,
the probability distribution of the subsequent
Article Outline observation depends only on the current obser-
vation and not on prior observations.
Glossary Measure-preserving transformation A map
Introduction from a measure space to itself such that for
Ergodicity each measurable subset of the space, it has
Mixing the same measure as its inverse image under
Hyperbolicity and Decay of Correlations the map.
Representations, Realizations, and Genericity Measure-theoretic entropy Non-negative (pos-
Future Directions sibly infinite) real number describing the com-
References plexity of a measure-preserving transformation.
Product transformation Given a pair of
Ergodicity is the condition under which the strong measure-preserving transformations: T of X and
law of large numbers holds for a dynamical sys- S of Y, the product transformation is the map of
tem. A case can be made that ergodicity holds in X
Y given by (T
S)(x, y) ¼ (T(x), S( y)).
situations much broader than is typically utilized Strong mixing Given two sets, over time the
in statistics, machine learning, and other scientific probability that points from one set end up in
disciplines. This is not surprising given the power the other set, converges to the product of the
of the pointwise ergodic theorem and the ability to probabilities of the two sets.
decompose a space into ergodic components. Weak mixing There exists a sequence of natural
Also, in the first half of the twentieth century, it numbers with density 1 in the set of all natural
was shown that topologically generic dynamical numbers such that the transformation is strong
systems are weak mixing, rigid, and thus, zero- mixing on that sequence.
entropy (and not i.i.d.). This chapter will give a
comprehensive account of mixing properties in
ergodic theory including Bernoulli, strong Introduction
mixing, weak mixing, mild, partial, and light
mixing. A hierarchy of mixing properties is pre- The term “ergodic” was introduced by Boltzmann
sented (Fig. 2), along with important examples, as (1871, 1909) in his work on statistical mechanics,
well as applications using the various properties. where he was studying Hamiltonian systems with
This chapter places many of the breakthroughs large numbers of particles. The system is described
from ergodic theory into a common framework at any time by a point of phase space, a subset of
with an eye toward emerging mathematical and ℝ6N where N is the number of particles. The con-
scientific areas and concludes with a collection of figuration describes the three-dimensional position
important unsolved problems. and velocity of each of the N particles. It has long

© Springer Science+Business Media, LLC, part of Springer Nature 2023 35

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_175
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_175-3
36 Ergodicity and Mixing Properties

been known that the Hamiltonian (i.e., the overall of concrete examples sitting at various points of
energy of the system) is invariant over time in these this hierarchy. Many of the mixing properties may
systems. Thus, given a starting configuration, all be characterized in terms of the Koopman opera-
future configurations as the system evolves lie on tors mentioned above (i.e., they are spectral prop-
the same energy surface as the initial one. erties), but the strongest mixing properties are not
Boltzmann’s ergodic hypothesis was that the spectral in nature. Also, this chapter highlights
trajectory of the configuration in phase space many of the connections between mixing proper-
would fill out the entire energy surface. The term ties and their spectral reformulations; however, it
“ergodic” is thus an amalgamation of the Greek is recommended the interested reader refer to the
words for work and path. This hypothesis then chapter ▶ “Spectral Theory of Dynamical Sys-
allowed Boltzmann to conclude that the long-term tems” (Lemancyzk and Kanigowski 2022) for
average of a quantity as the system evolves would detailed definitions and results focused mainly
be equal to its average value over the phase space. on the spectrum of measure preserving systems.
Subsequently, it was realized that this hypothe- This chapter will also bring to light some of the
sis is rarely satisfied. The ergodic hypothesis was connections between the range of mixing proper-
replaced in 1911 by the quasi-ergodic hypothesis ties and measure-theoretic entropy. In measure-
of the Ehrenfests (1911) which stated instead that preserving transformations that arise in practice,
each trajectory is dense in the energy surface, rather there is a correlation between strong mixing prop-
than filling out the entire energy surface. The mod- erties and positive entropy, although many of
ern notion of ergodicity (to be defined below) is these properties are logically independent.
due to Birkhoff and Smith (1924). Koopman One important issue for which many questions
(1931) suggested studying a measure-preserving remain open is that of higher-order mixing. Here,
transformation by means of the associated isometry the issue is if instead of asking that the observa-
on Hilbert space, UT : L2(X) ! L2(X) defined by tions at two times separated by a large time T be
UT( f ) ¼ f ∘ T. This point of view was used by von approximately independent, one asks whether if
Neumann (1932) in his proof of the mean ergodic one makes observations at more times, each pair
theorem. This was followed closely by Birkhoff suitably separated, the results can be expected to
(1931) proving the pointwise ergodic theorem. be approximately independent. This issue has an
An ergodic measure-preserving transformation analogue in probability theory, where it is well-
enjoys the property that Boltzmann first intended known that it is possible to have a collection of
to deduce from his hypothesis: That long-term random variables that are pairwise independent,
averages of an observable quantity coincide with but not mutually independent.
the integral of that quantity over the phase space.
These theorems allow one to deduce a form of Basics, Examples, and Highlighted
independence on the average: Given two sets of Applications
configurations A and B, one can consider the vol- In this chapter, except where otherwise stated, the
ume of the phase space consisting of points that are measure-preserving transformations under con-
in A at time 0 and in B at time t. In an ergodic sideration are defined on probability spaces.
measure-preserving transformation, if one com- More specifically, given a measurable space
putes the average of the volumes of these regions (X, ℬ) and a probability measure m defined on
over time, the ergodic theorems mentioned above ℬ, a measure-preserving transformation of
allow one to deduce that the limit is simply the (X, ℬ, m) is a ℬ-measurable map T : X ! X such
product of the volume of A and the volume of that m(T1B) ¼ m(B) for all B  ℬ.
B. This is the weakest mixing-type property. This While this definition makes sense for arbitrary
chapter outlines a rather full range of mixing prop- measures, not simply probability measures, most of
erties with ergodicity at the weakest end and the the results and definitions below only make sense in
Bernoulli property at the strongest end. the probability measure case. Sometimes it will be
This chapter sets out in some detail the various helpful to make the assumption that the underlying
mixing properties, basing the study on a number probability space is a Lebesgue space (i.e., the space
Ergodicity and Mixing Properties 37

together with its completed s-algebra agrees up to a form . . .x2x1  x0x1x2. . . (the  is a placeholder
measure-preserving bijection a.e. with the unit inter- that allows us to distinguish (for example)
val with Lebesgue measure and the usual s-algebra between the sequences . . .01010  10101. . . and
of Lebesgue measurable sets). Although this sounds . . .10101  01010. . .).
like a strong restriction, in practice it is barely a A Bernoulli shift is defined as a map (the shift
restriction at all, as almost all of the spaces that map) S on Aℕ by (S(x))n ¼ xnþ1 and define S on Aℤ
appear in the theory (and all of those that appear in by the same formula. Note that S is invertible as a
this chapter) turn out to be Lebesgue spaces. For a transformation on Aℤ but noninvertible as a trans-
detailed treatment of the theory of Lebesgue spaces, formation on Aℕ.
the reader is referred to Rudolph’s book (1990). The It is necessary to equip Aℕ and Aℤ with measures.
reader is referred also to the chapter on “Measure This is done by defining the measure of a preferred
Preserving Systems”. class of sets, checking certain consistency condi-
While many of the definitions presented are tions and appealing to the Kolmogorov extension
valid for both invertible and noninvertible theorem. Here the preferred sets are the cylinder
measure-preserving transformations, the strongest sets. Given m  n in the invertible case and a
mixing conditions are most useful in the case of sequence am . . . an, let ½am . . . an nm denote
invertible transformations. {x  Aℤ : xm ¼ am, . . ., xn ¼ an} and define
It will be helpful to present a selection of foun- m ½am . . . an nm ¼ Pam Pamþ1 . . . Pan : This is then
dational examples, relative to which ergodicity shown to uniquely define a measure m on the
and the various notions of mixing are explored. s-algebra of Aℤ generated by the cylinder sets. It
These examples and the lemmas necessary to is immediate to see that for any cylinder set C,
show that they are measure-preserving transfor- m(S1C) ¼ m(C), and it follows that S is a measure-
mations as claimed may be found in the books of preserving transformation of (Aℤ, ℬ, m). The con-
Petersen (1983), Rudolph (1990), Walters (1982), struction is exactly analogous in the noninvertible
and Silva (2008). More details on these examples case. See the chapter on “Measure Preserving Sys-
can also be found in the chapter on ▶ “Ergodic tems” or the books of Walters (1982) or Rudolph
Theory: Basic Examples and Constructions.” (1990) for more details of defining measures in these
systems.
Some Iconic Examples
Example 1 (Rotation on the circle). Let α  ℝ.
Example 4 (Markov Shift). The spaces Aℕ and
Let Rα : [0, 1) ! [0, 1) be defined by Rα(x) ¼ x þ
Aℤ are exactly as above, as is the shift map. All
α mod 1. It is straightforward to verify that Rα
that changes is the measure.
preserves the restriction of Lebesgue measure l to
To define a Markov shift, it is necessary to have
[0, 1) (it is sufficient to check that l R1
a ðJ Þ ¼ a stochastic matrix P (i.e., a matrix with non-
lðJ Þ for an interval J) negative entries whose rows sum to (1) with rows
and columns indexed by A and a left eigenvector π
Example 2 (Doubling Map). Let
for P with eigenvalue 1 with the property that the
M2 : [0, 1) ! [0, 1) be defined by M2(x) ¼ 2x
entries of π are non-negative and sum to 1. The
mod 1. Again, Lebesgue measure is invariant
existence of such an eigenvector is a consequence
under M2 (to see this, one observes that for an
of the Perron-Frobenius theory of positive matri-
interval J, M1
2 ðJ Þ consists of two intervals, each ces. Provided that the matrix P is irreducible (for
of half the length of J). This may be generalized in each a and a0 in A, there is an n > 0 such that
the obvious way to a map Mk for any integer k  2. Pna,a0 > 0), the eigenvector π is unique.
Given the pair (P, π), one defines the measure
Example 3 (Bernoulli Shift). Let A be a finite set
of a cylinder set by m ½am . . . an nm ¼
and fix a vector ( pi)i  A of positive numbers that
sum to 1. Let Aℕ denote the set of sequences of the pam Pam amþ1 . . . Pan1 an and extends m as before to
form x0x1x2. . ., where xn  A for each n  ℕ and a probability measure on Aℕ or Aℤ.
let Aℤ denote the set of bi-infinite sequences of the
38 Ergodicity and Mixing Properties

Example 5 (Interval Exchange Transformation). of equal length to obtain rn subcolumns of equal

The class of interval exchange transformations measure. For each i ¼ 0, . . ., rn – 1, place sn,i new
was introduced by Sinai (1976). An interval intervals, called spacers, above the ith sub-
exchange transformation is the map obtained by column. Then Cnþ1 is constructed by stacking
cutting the interval into a finite number of pieces each subcolumn, right over left, with its associ-
and permuting them in such a way that the ated spacers under the next subcolumn. The
resulting map is invertible, and restricted to each example below (Fig. 1) shows this in more detail.
interval is an order-preserving isometry. Thus Cnþ1 will consist of rn copies of Cn possibly
More formally, one takes a sequence of posi- separated by spacers. From the construction, it is
tive lengths ‘1, ‘2, . . ., ‘k summing to 1 and a clear that T Cnþ1 agrees with T Cn wherever T Cn is
permutation π of {1, . . ., k}. For 1  i  k, define defined. Let X be the union of the levels in the
ai ¼ j<i‘j and bi ¼ π( j ) < π(i)‘j. Note a1 ¼ 0 and columns and T ðxÞ ¼ lim n!1 T Cn: The height of
set akþ1 ¼ 1. The interval exchange transforma- T Cnþ1 is given by
tion defined by (‘1, . . ., ‘k) and π is the map
T : [0, 1) ! [0, 1) defined by Tj½ai ,aiþ1 Þ ðxÞ ¼ rn1
hnþ1 ¼ r n hn þ sn,i :
x þ ðbi ai Þ: It is straightforward to check that
i¼0
any such interval exchange transformation pre-
serves Lebesgue measure on the unit interval.
The transformation T will be finite measure-
Example 6 (Rank-one). There are multiple defi- preserving, if
nitions in the literature for rank-one transforma-
tions. For a comprehensive treatment of rank-one 1 rn 1
i¼0 sn,i
or more generally finite rank, see Friedman (1992, < 1,
hnþ1
1997); Adams et al. (2017); and Gao and Hill n¼0

(2014), and for the infinite measure case, see

Bozgan et al. (2015). otherwise, T will be infinite measure-preserving. In
Now, we give a description of geometric rank- the case that T is finite measure-preserving, we can
one transformations using the cutting and stacking choose the length of the original interval C0 such
notation (Friedman 1992; Bozgan et al. 2015). that X ¼ [0, 1) a.e. and T is an invertible Lebesgue
A column is a finite sequence of intervals of the probability preserving transformation on X.
same length, sometimes also called levels, that are
usually taken left-closed and right-open; its Example 7 (The Chacon Transformations). Con-
height is the number of intervals in the column. structions of measure-preserving transformations
A column C defines a column map TC so that each
interval is sent to the interval right above it by
translation; so TC is defined on every interval of
the column except the top. It is clear that on the
intervals where it is defined, TC is measurable and
measure-preserving with respect to Lebesgue
measure. To specify a rank-one construction, a
sequence of positive integers (rn)n0, rn  2, and
a doubly indexed sequence of nonnegative inte-
gers (sn,i), n  0, i ¼ 0, . . ., rn  1 are given. Begin
with column C0 consisting of a single finite inter- Cn → Cn+1 Cn+1
val, so its height is h0 ¼ 1. For the inductive step,
suppose column Cn of height hn are given. Then Ergodicity and Mixing Properties, Fig. 1 Construction
cut each interval of column Cn into rn subintervals of the Chacon-3 transformation
Ergodicity and Mixing Properties 39

satisfying certain properties are not always easy to fðTxÞ ¼ SðfxÞ:

come by. In [68], it is shown that generically in the
weak topology, invertible measure-preserving The map f is a measure-preserving homomor-
transformations are weak mixing and rigid phism and (X, ℬ, m, T ) is called an extension of (Y,
(which implies not strong mixing). However, for F , n, S). If the map f is a bijection a.e., then f is
some years, there was a dearth of explicit con- called an isomorphism, and the systems (X, ℬ, m,
structions that are weak mixing and not strong T ) and (Y, F , n, S) are said to be (measure theo-
mixing (Chacon 1966a, b; 1969; Maruyama retically) isomorphic. When the underlying prob-
1949).1 See Sect. 4 for a more detailed description ability spaces are clear, it may be said that T and
of strong and weak mixing, and Sect. 6.1 for a S are isomorphic.
description of the weak topology on invertible Measure-theoretic isomorphism is the basic
measure-preserving transformations. notion of “sameness” in ergodic theory. It is in
The rank-one transformations (Chacon 1966b; some sense quite weak, so that systems may be
1969), first published by Chacon as weak mixing isomorphic that feel very different. For example,
and not strong mixing, are not the same as the as discussed later, the time one map of a geodesic
transformation commonly referred to as Chacon’s flow is isomorphic to a Bernoulli shift). For com-
transformation (Friedman 1970, p. 86). While parison, the notion of sameness in topological
these transformations are all weak mixing, the dynamical systems (topological conjugacy) is far
transformation commonly referred to as Chacon’s stronger.
transformation uses a cut into three subcolumns at As an example of measure-theoretic isomor-
each stage with a single spacer added on the second phism, it may be seen that the doubling map
subcolumn (i.e., rn ¼ 3, sn,0 ¼ 0, sn,1 ¼ 1, sn,2 ¼ 0, is isomorphic to the one-sided Bernoulli shift on
hnþ1 ¼ 3hn þ 1). It is depicted in Fig. 1. This {0, 1} with p0 ¼ p1 ¼ 1/2 (the map f takes an
transformation is called Chacon-3. x  [0, 1) to the sequence of 0’s and 1’s in its
The transformation published in Chacon (1969) binary expansion; choosing the sequence ending
is referred to as Chacon-2. It is constructed by with 0’s, for example, if x is of the form p/2n).
cutting each column into two subcolumns at each Given a measure-preserving transformation
stage and adding a single spacer on the last sub- T of a probability space (X, ℬ, m), T is associated
column. Thus, rn ¼ 2, sn,0 ¼ 0, sn,1 ¼ 1, hn ¼ 2nþ1–1 to an isometry of L2(X, ℬ, m) by UT( f ) ¼ f ∘ T.
for all n. This operator is known as the Koopman Operator.
In the case where T is invertible, the operator UT is
Factors, Extensions, and Isomorphisms unitary. Two measure-preserving transformations
This chapter makes use of the concept of measure- T and S of (X, ℬ, m) and (Y, F , n) are spectrally
theoretic homomorphisms, isomorphisms, and isomorphic if there is a Hilbert space isomorphism
factor systems. Let (X, ℬ, m, T) and (Y, F , n, S) Θ from L2(X, ℬ, m) to L2(Y, F , n) such that
be measure-preserving dynamical systems on Θ ∘ UT ¼ US ∘ Θ. As demonstrated below, spec-
Lebesgue probability spaces. The system (Y, F , tral isomorphism is a strictly weaker property than
n, S) is a factor of (X, ℬ, m, T), if there exists a measure-theoretic isomorphism.
measure-preserving transformation f of X onto Since in ergodic theory, measure-theoretic iso-
Y such that for a.e. x  X, morphism is the basic notion of sameness, most
properties that are used to describe measure-
preserving systems are required to be invariant
under measure-theoretic isomorphism (i.e., if
1
From Parry quoted in Kakutani (1986), “For many years two measure-preserving transformations are
the first of Kakutani’s examples (which he had discussed measure-theoretically isomorphic, the first has a
with von Neumann) was the only other known example.
given property if and only if the second does). On
This example became widely known through the publicity
of Kakutani’s lectures. It appears for example in the other hand, we shall see that some mixing-type
Friedman’s book.” properties are invariant under spectral
40 Ergodicity and Mixing Properties

isomorphism, while others are not. If a property is T ðx0 , x1 , . . .Þ ¼ ðT ðx0 Þ, x0 , x1 , . . .Þ: The transfor-
invariant under spectral isomorphism, it is said mation T of X, ℬ, m is called the natural exten-
that it is a spectral property. sion of the transformation T of (X, ℬ, m) (see the
There are a number of mixing type properties chapter on “Ergodic Theory: Basic Examples and
that occur in the probability literature (α-mixing, Constructions” for more details). In situations
β-mixing, f-mixing, c-mixing, etc.) (see where one wants to use invertibility, it is often
Bradley’s (2005) survey for a description of possible to pass to the natural extension, work
these conditions). In the case of a stationary pro- there, and then derive conclusions about the orig-
cess, Xi for  1 < i < 1, the process can be inal noninvertible transformation.
modeled using a measure preserving transforma- Arguably, the simplest example of an exten-
tion T and measurable function f such that Xi ¼ sion is a two-point extension. Given an ergodic
f ∘ Ti1. These mixing properties from the proba- measure-preserving transformation T acting on (X,
bility literature often constrain the function f and ℬ, m), this can be viewed as selecting a switching
are not preserved under measure-theoretic iso- set A X of positive measure. The two-point
morphism. One example of this distinction is extension is composed of two copies
that finite ergodic Markov chains are measure- (X1, ℬ, m, T1) and (X2, ℬ, m, T2) with mirror
theoretically isomorphic to Bernoulli shifts; how- copies A1, A2 of the switching set A. Also, there
ever, these systems are viewed as distinct varieties is a measure-preserving bijection f : X1 ! X2
of stationary processes. For this reason, these between the two mirror copies such that A2 ¼
properties are not widely used in ergodic theory, f(A1). The two-point extension S is defined as,
although β-mixing turns out to be equivalent to
the so-called weak Bernoulli property (which T 1 ðxÞ if x  X1 ∖A1
turns out to be stronger than the Bernoulli prop- T 2 ðxÞ if x  X2 ∖A2
erty that is discussed in this chapter – see Sð x Þ ¼
T 2 ðfðxÞÞ if x  A1
Smorodinsky’s (1971) paper) and α-mixing is
T 1 f1 ðxÞ if x  A2 :
equivalent to strong mixing.
A basic construction (see the chapter on
Establishing that mixing properties such as
▶ “Ergodic Theory: Basic Examples and Con-
Bernoulli extend to a two-point extension is
structions”) that is required in what follows is
already nontrivial (Rudolph 1978; Kammeyer
the product of a pair of measure-preserving trans-
1990). These notions generalize to more general
formations: Given transformations T of (X, ℬ, m)
group extensions, as well as nonalgebraic exten-
and S of (Y, F , n), the product transformation is
sions such as the roof over a transformation.
defined as T
S: (X
Y, ℬ  F , m
n) by (T

A roof over a transformation is constructed by
S)(x, y) ¼ (Tx, Sy).
identifying a subset A X of positive measure
One issue that is faced on occasion is that it is
and extending the measure space using a measur-
sometimes convenient to deal with noninvertible
able set A0 and invertible measure preserving iso-
measure-preserving transformations. It turns out
morphism f such that A0 ¼ f(A) is disjoint from
that given a noninvertible measure-preserving
X. The set A0 is a disjoint copy of A. In this case,
transformation, there is a natural way to uniquely
the extension S is defined as,
associate an invertible measure-preserving trans-
formation sharing almost all of the ergodic prop-
erties of the original transformation. Specifically, T ðxÞ if x  X∖A
given a noninvertible measure-preserving trans- Sð x Þ ¼ fðxÞ if x  A
formation T of (X, ℬ, m), one lets X ¼ T f1 ðxÞ if x  A0 :
fðx0 , x1 , . . .Þ : xn  X and T(xn) ¼ xn1 for all n},
ℬ be the s-algebra generated by sets of the form The normalized measure m0 defined as m0(B) ¼
An ¼ x  X : xn  A , m An ¼ mðAÞ and m(B)/(1 þ m(A0)) is an invariant probability
Ergodicity and Mixing Properties 41

measure under the action of S. The first weak second representing velocity). The system is
mixing, nonstrong mixing transformation due to constrained in that its position is required to lie
Kakutani and von Neumann was constructed as a in a bounded region S of ℝM with a piecewise
carefully designed roof over an irrational rotation smooth boundary. The system evolves by moving
(see previous footnote 1). the position at a constant rate in the direction of
The notion of induced (or derived) transforma- the velocity vector until the point reaches @S, at
tion (Kakutani 1943) is the reverse notion from which time the component of the velocity parallel
the “roof” extension. Given a subset A X of to the normal to @S is reversed. This then defines
positive measure, the induced transformation TA a flow (i.e., a family of maps (Tt)t  ℝ satisfying
is defined on A as Ttþs ¼ Tt ∘ Ts) on the phase space. Since the
magnitude of the velocity is conserved, it is con-
T A ðxÞ ¼ T rA ðxÞ ðxÞ, venient to restrict to flows with speed 1. This
system is clearly the closest of the examples that
where rA(x) ¼ inf {n  1 : Tn(x)  A}. If T is we consider to the situation envisaged by
ergodic, invertible, and measure-preserving, then Boltzmann. Perhaps not surprisingly, proofs of
so is TA. Also, the entropy of an induced transfor- even the most basic properties for this system are
mation may be computed directly from the much harder than the other examples that we
entropy h(T) of T, as hðT A Þ ¼ mð1AÞ hðT Þ consider.
(Abramov 1959). Interestingly, it was established
for an ergodic transformation, there exists a dense Application 2 (Combinatorial Number Theory).
collection of sets A such that TA is weak mixing The legendary mathematician Paul Erdős was
(Chacon 1966a), and then a dense collection of known to disavow life’s material possessions
A such that TA is strongly mixing (Friedman and and instead, give complete devotion to mathemat-
Ornstein 1973). Hence, one method for obtaining ics, and of course, his own mother. He used all
a zero-entropy strong mixing transformation is to disposable cash to fund a growing number of
apply the result of Friedman-Ornstein (1973) to bounties on unsolved math problems. His first
any zero-entropy ergodic transformation. The major award went to Szemerédi for proving that
notion of an induced transformation has led to any sequence of natural numbers with positive
the theory of Kakutani equivalence and many upper density contains arbitrarily long arithmetic
important results for both transformations and progressions. The award was $1000 in 1974. Fast
flows. For example, see results of Ambrose and forward to 2004, mathematicians Green and Tao
Kakutani (1942) and Rudolph (1976). extended Szemerédi’s results to prove that the
prime numbers contain arbitrarily long arithmetic
Highlighted Applications and Connections progressions (Green and Tao 2008). A conjecture
Application 1 (Hard Sphere Gases and Bil- of Erdős continues to remain open: Given any
liards). We wish to model the behavior of a gas increasing sequence of natural numbers, ni such
1 1
in a bounded region. We make the assumption that that i¼1 ni ¼ 1, then ni contains arbitrarily
the gas consists of a large number N of identical long arithmetic progressions. A positive answer
balls which move at constant velocity until two to this question would imply the primes contain
balls collide, whereupon they elastically swap arbitrarily long arithmetic progressions.
momentum along the direction of contact. The In the meantime (circa 1977), Hillel Furstenberg
phase space for this system is a region of ℝ6N proved Szemerédi’s theorem using ergodic theory
(with N 3-dimensional position vectors and N techniques (Furstenberg 1977). Furstenberg
3-dimensional velocity vectors). More abstractly, established a transference principle that recast number
the system is equivalent to the motion of a single theoretic questions into properties of measure-pre-
point particle in a region of ℝM
ℝM (with the serving dynamical systems. Furstenberg, Bergelson,
first M-vector representing position and the and others spearheaded new research on multiple
42 Ergodicity and Mixing Properties

recurrence of dynamical systems to obtain far VC-dimension for almost all i.i.d. samples (Vapnik
reaching generalizations of the original Szemerédi’s and Chervonenkis 2015).
theorem, including to higher order group actions Using ergodic theory techniques, it was shown
(Furstenberg and Katznelson 1978; Bergelson and that Vapnik-Chervonenkis uniform convergence
Leibman 2002) and probabilistic versions for occur- holds for stationary ergodic processes (Adams
rences of arithmetic progressions based on multi- and Nobel 2010).
dimensional ergodic theorems (Bergelson 1987;
Conze and Lesigne 1984; Zhang 1996; Frantzikinakis Theorem 1 (Adams and Nobel 2010). Let w be a
and Kra 2005; Tao 2008; Austin 2010). Many of the complete separable metric space equipped with
ideas from Furstenberg’s ergodic theoretic approach its Borel measurable subsets S, and let C S be
find themselves in the ideas of Green and Tao. To any countable family of sets. If dimðC Þ < 1, then
better comprehend the impact of Furstenberg’s break- for every stationary ergodic process X ¼ X1, X2,
throughs, as well as Tao and Green’s, here is an . . . taking values in ðw, S Þ,
excerpt from Tao’s 2006 Fields Medal citation:
m
1
Of special note is his joint work with Ben Green, a Gm ðC : XÞ ¼ sup I ðX i  C Þ  P ðX  C Þ
Clay Research Fellow from 2005 through 2007. In CC m i¼1
their 2004 paper, “The primes contain arbitrarily
long arithmetic progressions,” the authors answered ! 0 wp1
in the affirmative a long-standing conjecture that ð1Þ
had resisted many attempts.

Application 3 (Statistical Learning Theory). as m tends to infinity. In other words, C is a

Statistical learning theory is the application of sta- Glivenko-Cantelli class for every stationary ergo-
tistics, probability, and mathematics to improve the dic process.
understanding of machine learning algorithms and In ergodic theory notation, this result implies
to measure the effectiveness of mathematical for a collection C with finite VC-dimension and a
models to learn from data, predict new outcomes, Lebesgue probability space (X, ℬ, m), for almost
and approximate quantitative relationships. Statisti- every x  X,
cal learning theory has experienced growing interest
due to renewed interest in machine learning, neural n1
1
networks, and in particular, deep neural networks. lim sup I C T i x  mðCÞ ¼ 0:
n!1
CC n i¼0
Deep neural networks (DNNs) are achieving new
state-of-the-art in many machine learning areas such
Note that ergodic stationary processes can be
as natural language processing, speech, and image
modeled using an ergodic invertible measure-
recognition. However, DNNs are not understood
preserving transformation T and measurable func-
well mathematically, nor is it easy to predict their
tion f such that Xi(x) ¼ f(Ti1x). It is possible to
failure modes in real-life situations.
extend this result to processes of the form f(Ti x) if
Statisticians often model sequential data using
the complexity of the class of functions that f is
random variables defined on a probability space. If
drawn from is restricted. See Vapnik (2013); Pol-
the random variables are independent and identi-
lard (1990); Adams and Nobel (2010) for further
cally distributed, there are many powerful results
details.
such as the central limit theorem or uniform conver-
gence of ergodic averages for classes of models with
Application 4 (Consistent Temporal Modeling
bounded complexity. In particular, Vapnik and
by HMMs and DNNs). Hidden Markov Models
Chervonenkis introduced VC-dimension as a mea-
(HMMs) have been studied for decades and were
sure of complexity for a collection of sets and then
commonly used for modeling data from various
showed that pointwise ergodic averages converge
domains (i.e., text, speech, video). The power of
uniformly over a class of sets with bounded
HMMs came from the ability to develop efficient
Ergodicity and Mixing Properties 43

algorithms for approximately learning transition This allows us to decompose the transformation
and emission probabilities. HMMs have been X into two pieces A and Ac and study the transfor-
studied in the context of ergodic theory under mation T separately on each. In fact the same
different frameworks (e.g., subshifts of finite situation holds if T 1A and A agree up to a set of
type, sofic measures). See Marcus et al. (2011) measure 0. For this reason, a set A is called invari-
for more details on applications of ergodic theory ant if m(T 1A Δ A) ¼ 0.
and entropy to HMMs. In McGoff et al. (2015), Returning to Boltzmann’s ergodic hypothesis,
the consistency of maximum likelihood estimates existence of an invariant set of measure between
for HMMs was established under general condi- 0 and 1 would be a bad situation as his essential
tions including observations with noise. idea was that the orbit of a single point would
More recently, the availability of big data has “see” all of X, whereas if X were decomposed in
led to a resurgence in neural networks and in this way, the most that a point in A could see
particular, deep neural networks (DNNs) for data would be all of A, and similarly the most that a
modeling and prediction. While DNNs may be point in Ac could see would be all of Ac.
successful in modeling data (examples include A measure-preserving transformation will be
Google translate, Apple Siri, Amazon Alexa), called ergodic if it has no nontrivial decomposi-
the DNNs are generally viewed as black boxes tion of this form. More formally, let T be a
with little understanding of why and when they measure-preserving transformation of a probabil-
work. This lack of understanding has spurred new ity space (X, ℬ, m). The transformation T is said to
research toward applying rigorous mathematics to be ergodic if for all invariant sets, either the set or
improve understanding of deep neural networks, its complement has measure 0.
both in the training phase and in the understanding Unlike the remaining concepts discussed in
of failure modes after a system is trained. During this chapter, this definition of ergodicity applies
2018–2019, several research papers have demon- also to infinite measure-preserving transforma-
strated how over-parameterization of neural net- tions and even to certain non-measure-preserving
works can lead to consistent modeling. A key transformations. See Aaronson’s book (Aaronson
ingredient in this approach is the use of 1997) for more information.
Hoeffding’s inequality to bound the probability The following lemma is often useful:
that the sum of bounded independent random
variables deviates from its expected value by Lemma 2 Let (X, ℬ, m) be a probability space
more than a certain amount. This is a computa- and let T: X ! X be a measure-preserving trans-
tional version of the ergodic theorem which does formation. Then T is ergodic if and only if the only
not hold in all situations. However, there is measurable functions f satisfying f ∘ T ¼ f (up to
research to extend Hoeffding’s inequality beyond sets of measure 0) are constant almost
the i.i.d. case (Impagliazzo and Kabanets 2010; everywhere.
Schmidt et al. 1993; Pelekis and Ramon 2017), in For the straightforward proof, notice that if the
some cases developing new concentration condition in the lemma holds and A is an invariant
inequalities. For more details on recent research set, then 1A ∘ T ¼ 1A almost everywhere, so that
in this area, see Du et al. (2019); Arora et al. 1A is an a.e. constant function and so A or Ac is of
(2019); Arora et al. (2018); and Raginsky measure 0. Conversely, if f is an invariant func-
et al. (2017). tion, it is seen that for each α, {x: f (x) < α} is an
invariant set and hence of measure 0 or 1. It fol-
lows that f is constant almost everywhere. Note
Ergodicity that it is sufficient to check that the bounded
measurable invariant functions are constant.
Given a measure-preserving transformation The following corollary of the lemma shows
T : X ! X, if T1A ¼ A, then T1Ac ¼ Ac also. that ergodicity is a spectral property.
44 Ergodicity and Mixing Properties

Corollary 3 Let T be a measure-preserving In the case where T is not ergodic, it is also

transformation of the probability space (X, ℬ, possible to identify the limit in the ergodic theo-
m). Then T is ergodic if and only if 1 is a simple rems: f ¼ ðf j I Þ, where I is the s-algebra of
eigenvalue of UT. T-invariant sets.
The ergodic theorems mentioned earlier due to Note that the set on which the almost every-
von Neumann and Birkhoff are the following (see where convergence in Birkhoff’s theorem takes
also the chapter on ▶ “Ergodic Theorems”). place depends on the L1 function f that one is
considering. Straight-forward considerations
Theorem 4 (von Neumann Mean Ergodic Theo- show that there is no single full measure set that
rem (von Neumann 1932)). Let T be a measure- works simultaneously for all L1 functions. In the
preserving transformation of the probability case where X is a compact metric space, it is well
space (X, ℬ, m). For f  L2(X, ℬ, m), let AN f ¼ known that C(X), the space of continuous functions
( f þ f ∘ T þ . . . þ f ∘ T N  1)/N. Then for all on X with the uniform norm, has a countable dense
f  L2 (X, ℬ, m), AN f converges in L2 to an set which can be denoted by ( fn)n1. If the dynam-
invariant function f *. ical system (X, ℬ, m, T) is measure preserving and
ergodic, then for each n, there is a set Bn of measure
Theorem 5 (Birkhoff Pointwise Ergodic Theo- 1 such that for all x  Bn, AN fn(x) ! fn dm. Letting
rem (Birkhoff 1931)). Let T be a measure- B ¼ \nBn, one obtains a full measure set such that
preserving transformation of the probability for all n and all x  B, AN fn(x) ! fn dm. A simple
space (X, ℬ, m). Let f  L1(X, ℬ, m). Let AN f be approximation argument then shows that for all
as above. Then for m-almost every x  X, (AN f x  B and all f  C(X), AN f(x) ! f dm.
(x)) is a convergent sequence. A point x with this property is said to be generic
Of these two theorems, the pointwise ergodic for (X, ℬ, m, T). The observations above show that
theorem is the deeper result, and it is straightfor- for an ergodic measure preserving system (X, ℬ, m,
ward to deduce the mean ergodic theorem from T), m{x: x is generic for (X, ℬ, m, T)} ¼ 1.
the pointwise ergodic theorem. The mean ergodic If T is ergodic, but Tn is not ergodic for some
theorem was reproved very concisely by Riesz n, then one can show that the space X splits up as
(1938) and it is this proof that is widely known A1, . . ., Ad for some d|n in such a way that T(Ai) ¼
now. Riesz’s proof is reproduced in Parry’s (1981) Aiþ1 for i < d and T(Ad) ¼ A1 with T n acting
book. There have been many different proofs ergodically on each Ai. The transformation T is
given of the pointwise ergodic theorem. Notable totally ergodic if T n is ergodic for all n  ℕ. One
among these are the argument due to Garsia can check that a noninvertible transformation T is
(1965) and a proof due to Katznelson and Weiss ergodic if and only if its natural extension is
(1982) based on work of Kamae (1982), which ergodic.
appears in a simplified form in work of Keane and The following lemma gives an alternative char-
Petersen (2006). acterization of ergodicity, which in particular
If the measure-preserving transformation T is relates it to mixing.
ergodic, then by virtue of Lemma 2, the limit
functions appearing in the ergodic theorems are Lemma 6 (Ergodicity as a Mixing Property). Let
constant. One sees that the constant is simply the T be a measure-preserving transformation of the
integral of f with respect to m, so that in this probability space (X, ℬ, m). Then T is ergodic if
situation AN f (x) converges to f dm in norm and and only if for all f and g in L2,
pointwise almost everywhere, thereby providing a
justification of Boltzmann’s original claim: For 1
N
ergodic measure-preserving transformations, hf , g∘T n i ! hf , 1ih1, gi:
N n¼0
time averages agree with spatial averages.
Ergodicity and Mixing Properties 45

In particular, if T is ergodic, then To see that any rank-one measure-preserving

n
ð1=N Þ N1n¼0 mðA \ T BÞ ! mðAÞmðBÞ all mea-
transformation is ergodic, note that each level
surable sets A and B. sweeps out the entire space under iterates of
Now, this chapter examines the ergodicity of T. In particular, if I is a level in a column Cn,
the examples presented above. First, for the rota- then m [1 i¼0 T ðI Þ ¼ 1: By approximation, every
i

tion of the circle, it is claimed that the transforma- set A of positive measure satisfies,
tion is ergodic if and only if the ‘angle’ α is m [1i¼0 T ðAÞ ¼ 1: This proves that every rank-
i

irrational. To see this, argue as follows. If α ¼ one transformation is ergodic, including

p/q, then f(x) ¼ e2πiqx is a nonconstant Rα- Chacon-2 and Chacon-3.
invariant function, and hence Rα is not For the hard sphere systems, there are no
ergodic. On the other hand, if α is irrational, results on ergodicity in full generality. Important
suppose f is a bounded measurable invariant func- special cases have been studied by Sinai (1970),
tion. Since f is bounded, it is an L2 function, and so Sinai and Chernov (1987), Krámli et al. (1991),
f may be expressed in L2 as a Fourier series: f ¼ Simányi and Szász (1999), Simányi (2004, 2003),
n  ℤcnen where en(x) ¼ e2πinx. It is seen that and Young (1998).
f ∘ Rα ¼ n  ℤe2πinαcnen. In order for f to be
equal in L2 to f ∘ Rα, they must have the same Ergodic Decomposition
Fourier coefficients, so that cn ¼ e2πinαcn for each It has already been observed that if a transforma-
n. Since α is irrational, this forces cn ¼ 0 for all tion is not ergodic, then it may be decomposed
n 6¼ 0, so that f is constant as required. into parts. Clearly if these parts are not ergodic,
The doubling map and the Bernoulli shift are they may be further decomposed. It is natural to
both ergodic, although we defer proof of this for ask whether the transformation can be
the time being, since they in fact have the strong decomposed into ergodic parts, and if so what
mixing property. A Markov chain with matrix form does the decomposition take? In fact such a
P and vector π is ergodic if and only if for all decomposition does exist, but rather than decom-
i and j in A with πi > 0 and πj > 0, there exists pose the transformation, it is necessary to decom-
an n  0 with Pnij > 0: This follows from the pose the measure into ergodic pieces. This is
ergodic theorem for Markov chains (which is known as ergodic decomposition.
derived from the Strong Law of Large Numbers) The set of invariant measures for a measurable
(see Feller (1950) for details). In particular, if the map T of a measurable space (X, ℬ) to itself forms
underlying Markov chain is irreducible, then the a simplex. General functional analytic consider-
measure is ergodic. ations (due to Choquet (1956b, a) – see also
In the case of interval exchange transforma- Phelps’s (1966) account of this theory) mean that
tions, there is a simple necessary condition on it is possible to write any member of the simplex
the permutation for irreducibility, namely, for as an integral-convex combination of the extreme
1  j  k, π{1, . . ., j} 6¼ {1, . . ., j}. Under this points. Further, the extreme points of the simplex
condition, Masur (1982) and Veech (1982) inde- may be identified as precisely the ergodic invari-
pendently showed that for almost all values of the ant measures for T. It follows that any invariant
sequence of lengths (‘i)1ik, the interval probability measure m for T may be uniquely
exchange transformation is ergodic. (In fact they expressed in the form
showed the stronger condition of unique ergodic-
ity: That the transformation has no other invariant
mðAÞ ¼ nðAÞ dmðnÞ,
measure than Lebesgue measure. This implies that Merg ðX,T Þ
Lebesgue measure is ergodic, because if there
were a nontrivial invariant set, then the restriction where Merg(X, T ) denotes the set of ergodic
of Lebesgue measure to that set would be another T-invariant measures on X and m is a measure on
invariant measure.) Merg(X, T ).
46 Ergodicity and Mixing Properties

A detailed proof of ergodic decomposition is not property that there is convergence in the strong
given here. The theorem can be proved using Césaro sense. That is, a measure-preserving trans-
the Birkhoff ergodic theorem and the Riesz Repre- formation T is weak mixing if
sentation theorem identifying the dual space of the
space of continuous functions on a compact space as 1
N1

the set of bounded signed measures on the space j mðA \ T n BÞ  mðAÞmðBÞ j! 0 as

N n¼0
(see Rudin’s (1966) book for details). For more
details on ergodic decomposition, see Rudolph’s N ! 1:
(1990) book which gives a full proof in the case
In order for T to be strong mixing, it is required
that X is a Lebesgue space based on a detailed
simply m(A \ TNB) ! m(A)m(B) as N ! 1. It is
development of the theory of these spaces and builds
clear that strong mixing implies weak mixing and
measures using conditional expectations. Oxtoby
weak mixing implies ergodicity.
(1952) also wrote a survey article containing many
If Td is not ergodic (so that T–dA ¼ A for
of the details (and much more besides).
some A of measure strictly between 0 and 1),
then jm(TndA \ A)  m(A)2 j ¼ m(A)(1  m(A)),
Theorem 7 Let X be a compact metric space, ℬ
so that T is not weak mixing.
be the Borel s-algebra, m be an invariant Borel
An alternative characterization of weak mixing
probability measure and T be a continuous
is as follows:
measure-preserving transformation of (X, ℬ, m).
Then for each x  X, there exists an invariant
Lemma 8 The measure-preserving transforma-
Borel measure mx such that:
tion T is weak mixing if and only if for every pair of
measurable sets A and B, there exists a subset J of
1. For f  L1(X, ℬ, m), f dm ¼ ( f dmx) dm(x);
ℕ of density 1 (i.e., #(J \ {1, . . ., N})/N ! 1)
2. Given f  L1(X, ℬ, m), for m-almost every
such that
x  X, one has AN f(x) ! f dmx;
3. The measure mx is ergodic for m-almost every lim mðA \ T n BÞ ¼ mðAÞmðBÞ: ð2Þ
n!1n J
x  X.
By taking a countable family of measurable
Notice that conclusion (2) shows that mx can be sets that are dense (with respect to the metric d-
understood as the distribution on the phase space (A, B) ¼ m(A Δ B)) and taking a suitable intersec-
“seen” if one starts the system in an initial condi- tion of the corresponding J sets, one shows that for
tion of x. This interpretation of the measures mx a given weak mixing measure-preserving trans-
corresponds closely with the ideas of Boltzmann formation, there is a single set J ℕ such that
and the Ehrenfests in the formulation of the ergo- (2) holds for all measurable sets A and B (see
dic and quasi-ergodic hypotheses, which can be Petersen (1983) or Walters (1982) for a proof).
seen as demanding that mx is equal to m for It is shown that an irrational rotation of the
(almost) all x. circle is not weak mixing as follows: let
α  ℝ\ℚ and let A be the interval 14 , 34 : There
is a positive proportion of n’s in the natural num-
Mixing bers (in fact proportion 1/3) with the property that
j T n 12  12 j< 16 : For these n’s mðA \ T n AÞ > 13 ,
As mentioned above, ergodicity may be seen as an so that in particular j mðA \ T n AÞ  mðAÞmðAÞ j
1
independence on average property. More specifi- > 12 : Clearly this precludes the required conver-
cally, one wants to know whether in some sense gence to 0 in the definition of weak mixing. Since
m(A \ TnB) converges to m(A)m(B) as n ! 1. Rna ¼ Rna , the earlier argument shows that Rna is
Ergodicity is the property that there is conver- ergodic, so that Rα is totally ergodic, but not weak
gence in the Césaro sense. Weak mixing is the mixing.
Ergodicity and Mixing Properties 47

On the other hand, now it is shown that any 1. T is weak mixing if and only if for every f,
Bernoulli shift is strong mixing. To see this, let g  L2 one has
A and B be arbitrary measurable sets. By standard
measure-theoretic arguments, A and B may each
N1
be approximated arbitrarily closely by a finite 1
j hf , g∘T n i  h f , 1ih1, gi j! 0 as
union of cylinder sets. Since if A0 and B0 are finite N n¼0
unions of cylinder sets, we have that m(A0 \ TnB0) N ! 1:
is equal to m(A0)m(B0) for large n, it is easy to
deduce that m(A \ T nB) ! m(A)m(B) as required. 2. T is strong mixing if and only if for every
Since the doubling map is measure-theoretically f, g  L2, one has
isomorphic to a one-sided Bernoulli shift, it fol-
lows that the doubling map is also strong mixing.
f , g∘T N ! h f , 1i h1, gi as N ! 1:
Similarly, if a Markov Chain is irreducible
(i.e., for any states i and j, there exists an n  0
Using this, one can see that both mixing con-
such that Pnij > 0) and aperiodic (there is a state
ditions are spectral properties.
i such that gcd n; Pnii > 0 ¼ 1 ), then given
any pair of cylinder sets A0 and B0 , it follows Corollary 10 Weak and strong mixing are spec-
from standard theorems of Markov chains tral properties.
m(A0 \ T nB0) ! m(A0)m(B0). The same argument This is an opportune time to highlight a couple
as above then shows that an aperiodic irreducible major results that establish mixing properties for
Markov Chain is strong mixing. On the other interval exchange transformations. In Katok
hand, if a Markov chain is periodic (1980a), Katok shows that no interval exchange
d¼ gcd n : Pnii > 0 > 0 , then letting A ¼ B ¼ transformation is strong mixing. On the other
{x : x0 ¼ i}, it follows that m(A \ T nB) ¼ 0 when- hand, if the underlying permutation n is irreduc-
ever d ∤ n. Thus, Td is not ergodic, so that T is not ible and not a rotation,2 then the interval exchange
weak mixing. Note that a mixing Markov Chain is transformation is weak mixing for almost all divi-
actually measure-theoretically isomorphic to a sions of the interval (with respect to Lebesgue
Bernoulli shift (Friedman and Ornstein 1970; measure on the k – 1-dimensional simplex). The
Adler et al. 1972; Keane and Smorodinsky 1979b). case k ¼ 3 was proved by Katok and Stepin (1967)
So far in this chapter, all examples of explicit and the general case including k > 3 was
mixing transformations have been isomorphic established by Avila and Forni (2007). (This
to Bernoulli transformations. Ornstein’s con- research is cited as a primary contributor to Avila’s
struction of rank-one mixing using random 2014 Fields Medal (https://www.mathunion.org/
spacers produced the first known examples of imu-awards/fields-medal/fields-medal-2014/fields-
zero-entropy mixing (Ornstein 1972). Since medallists-2014-awardees-brief-citations).)
these examples, tools have been developed to While on the face of it the formulation of weak
establish explicit rankone mixing transforma- mixing is considerably less natural than that of
tions (e.g., staircase transformations) (Adams strong mixing, the notion of weak mixing turns
1998; Creutz and Silva 2010). Amazingly, it out to be extremely natural from a spectral point of
has already been established by Danilenko that view. Given a measure-preserving transformation
any discrete countable infinite amenable group T, let UT be the Koopman operator described
admits a zero-entropy mixing action (Danilenko above. Since this operator is an isometry, any
2016). eigenvalue must lie on the unit circle. The constant
Both weak and strong mixing have formula- function 1 is always an eigenfunction with eigen-
tions in terms of functions: value 1. Note that if z ¼ x þ iy is in the complex

Lemma 9 Let T be a measure-preserving trans- 2

A permutation π on {1, 2, . . ., k} is a rotation if π(i) ¼ i þ 1
formation of the probability space (X, ℬ, m). mod k.
48 Ergodicity and Mixing Properties

plane, then its complex conjugate z ¼ x  iy: If from the fact that any sub-s-algebra F of
T is ergodic and g and h are eigenfunctions of UT ℬ gives rise to a factor mapping π : (X, ℬ, m) !
with eigenvalue l, then gh is an eigenfunction with (X, F , m) with π(x) ¼ x. By construction L2(X, ℬ0,
eigenvalue 1, hence invariant, so that g ¼ Kh m) is the closed linear span of the eigenfunctions
for some constant K. Notice that for ergodic trans- of T considered as a measure-preserving transfor-
formations, up to rescaling, there is at most one mation of (X, ℬ0, m). By the Discrete Spectrum
eigenfunction with any given eigenvalue. Theorem of Halmos and von Neumann (1942),
If UT has a nonconstant eigenfunction f, then T acting on (X, ℬ0, m) is measure-theoretically
one has j UnT f , f j¼ k f k2 for each n, whereas by isomorphic to a rotation on a compact group.
Cauchy-Schwartz, |h f, 1i|2 < k f k2. It follows that This allows one to split L2(X, ℬ, m) as
j U nT f , f  h f , 1ih1, f i j c for some positive L2 ðX, ℬ, mÞ L2c ðX, ℬ, mÞ, where, as mentioned
constant c, so that using Lemma 9, T is not weak above the first part is the discrete spectrum part,
mixing. spanned by eigenfunctions, and the second part is
The converse can be shown using the spectral the continuous spectrum part, consisting of func-
theorem. For a detailed proof, see section 2.6 of tions whose spectral measure is continuous. Since
Petersen (1983). Also, another chapter ▶ “Spec- L2 is split into a discrete part and a continuous
tral Theory of Dynamical Systems” (Lemancyzk part, it is natural to ask whether the underlying
and Kanigowski 2022) is devoted to a study of transformation T can be split up in some way into
spectral theory and its connections to ergodic the- a weak mixing part and a discrete spectrum
ory. We encourage the interested reader to refer to (compact group rotation) part, somewhat analo-
this chapter for many of the detailed definitions gously to the ergodic decomposition. Unfortu-
and results on spectral theory. nately, there is no-such decomposition available.
However, for some applications, for example, to
Theorem 11 The measure-preserving transfor- multiple recurrence (starting with the work of
mation T is weak mixing if and only UT has no Furstenberg (1977, 1981)), the decomposition of
nonconstant eigenfunctions. L2 (possibly into more complicated parts) plays a
Of course this also shows that weak mixing is a crucial role (see the chapters “Ergodic Theory:
spectral property. Equivalently, this says that the Recurrence” and “Ergodic Theory: Interactions
transformation T is weak mixing if and only if with Combinatorics and Number Theory”).
apart from the constant eigenfunction, the opera- For noninvertible measure-preserving transfor-
tor UT has only continuous spectrum (i.e., the mations, the transformation is weak or strong
operator has no other eigenfunctions). mixing if and only if its natural extension has
Using this theory, one can establish the that property.
following: The understanding of weak mixing in terms of
the discrete part of the spectrum of the operator
Theorem 12 also extends to total ergodicity. Tn is ergodic if and
only if T has no eigenvalues of the form e2πip/n
1. T is weak mixing if and only if T
T is ergodic; other than 1. From this, it follows that an ergodic
2. If T and S are ergodic, then T
S is ergodic if measure-preserving transformation T is totally
and only if US and UT have no common eigen- ergodic if and only if it has no rational spectrum
values other than 1. (i.e., no eigenvalues of the form e2πip/q other than
the simple eigenvalue 1).
For a measure-preserving transformation T, let An intermediate mixing condition between
K be the subspace of L2 spanned by the eigen- strong- and weak mixing is that a measure-
functions of UT. It is a remarkable fact that K may preserving transformation is mild-mixing if when-
be identified as L2(X, ℬ, m) where ℬ0 is a ever f ∘ T ni ! f for an L2 function f and a
sub-s-algebra of ℬ. The space K is called the sequence ni ! 1, then f is a.e. constant. Clearly
Kronecker factor of T. The terminology comes mild-mixing is a spectral property. If a
Ergodicity and Mixing Properties 49

transformation has an eigenfunction f, then it is then strong mixing implies topological mixing.
straightforward to find a sequence ni such that Mild mixing does not imply topological mixing.
f ∘T ni ! f , thus, if follows that mild-mixing There is a measure theoretic property which is a
implies weak mixing. To see that strong mixing close counterpart to topological mixing. The prop-
implies mild-mixing, suppose that T is strong erty is commonly referred to as light mixing and
mixing and that f ∘T ni ! f : Then f ∘T ni f ! was first introduced by Walters as intermixing
k f k2 : On the other hand, the strong mixing prop- (Walters 1972). A transformation T is lightly
erty implies that f ∘T ni f ! jh f , 1ij2 : The equal- mixing if given any two sets A and B of positive
ity of these implies that f is a.e. constant. Mild- measure,
mixing has a useful reformulation in terms of
ergodicity of general (not necessarily probability) lim inf mðA \ T n BÞ > 0:
n!1
measure-preserving transformations:
A transformation T is mild-mixing if and only if If T is lightly mixing, then T is mildly mixing
for every conservative ergodic measure- (since lim infn!1m(A \ T nAc) ¼ 0 for A in a
preserving transformation S, T
S is ergodic. See rigid factor). Also, there exist transformations
Furstenberg and Weiss’ article (1978) for further which are lightly mixing, but are not strong
information on mild-mixing. mixing. In particular, the transformation first
If there exists a sequence ni ! 1 such that for published by Chacon as a weak mixing, nonstrong
every f  L2, f ∘T ni ! f , then T is said to be rigid. mixing transformation is actually lightly mixing
The sequence ni is called a rigidity sequence for T. (Chacon 1969; Friedman and King 1991).
Mild mixing transformations are those transfor- If there exists α > 0 such that for all measurable
mations with no rigid factor. There has been much sets A and B,
new research on the nature of rigidity sequences,
spurred largely by (Bergelson et al. 2014). See lim mðA \ T n BÞ  amðAÞmðBÞ,
n!1
(Eisner and Grivaux 2011; Aaronson et al. 2014;
Adams 2015; Fayad and Kanigowski 2015;
then T is partially mixing, or more specifically
Grivaux 2013; Le 2017; Bayless and Yancey
α-mixing. It is not difficult to show that α  1
2015; Grivaux and Roginskaya 2013; Robertson
for T measure-preserving. There are transforma-
2019) for a selection of recent results on rigidity
tions which are partially mixing, but are not
and nonrecurrent sequences. If there exists α > 0
strongly mixing. Also, Chacon’s transformation
such that for every measurable set A,
from (Friedman and King 1991) is lightly mixing,
but not partially mixing. In Friedman (1989),
lim inf mðA \ T n AÞ  amðAÞ,
n!1 Friedman constructs explicit transformations
with an optimal combination of mixing and rigid-
then T is said to be partially rigid, or more specif- ity. In particular, for each α such that 0 < α < 1, a
ically α-rigid. Strong mixing transformations are transformation is constructed that is simulta-
not partially rigid for any α > 0. While mild neously α-rigid and (1 – α)-mixing.
mixing transformations are not rigid, they can be See Fig. 2 for a hierarchy of mixing properties.
partially rigid, and in particular, both Chacon The stronger more restrictive properties start at the
transformations (Chacon-2 and Chacon-3) are top and become weaker as one descends down the
partially rigid. tree. Note that there are known examples
In the case where the space X is endowed with distinguishing these properties, except in the
a topology, a transformation T is topologically case of strong mixing and mixing of all orders.
mixing if given two nonempty open sets U and The dashed line between these properties signifies
V, there exists N  ℕ such that for all n > N, the that it is an open question whether strong mixing
set V \ T nU is nonempty. Topological mixing (i.e., 2-mixing) implies mixing of all orders. Many
does not imply ergodicity (Muehlegger et al. of the branches are labeled with references to
1997). If every open set has positive measure,
50 Ergodicity and Mixing Properties

Bernoulli Transformation [24, 113] Rank-one Mixing [114, 5]

K Automorphism [139, 115] Finite Rank Mixing [145]

Countable Zero-
Lebesgue Mixing of All Orders Entropy
Spectrum Mixing
Strong Mixing [114]

Partial Mixing [56]

Light Mixing [166, 75, 52]

Mild Mixing [60]

Rank-one, rigid
Weak Mixing [95, 34] and weak mixing
are generic.
Totally Ergodic

Ergodic [27, 26]

Ergodicity and Mixing Properties, Fig. 2 Mixing hierarchy

articles where these properties were defined or invertible measure-preserving transformation has
important examples were presented. simple singular spectrum.
The strongest spectral property considered is The property of countable Lebesgue spectrum
that of having countable Lebesgue spectrum. is by definition a spectral property. Since it
A detailed discussion of spectral theory is not completely describes the transformation up to
included here, although this special case that can spectral isomorphism, there can be no stronger
be described simply. Specifically, let T be an spectral properties. The remaining properties that
invertible measure-preserving transformation. are examined are invariant under measure-
Then T has countable Lebesgue spectrum if theoretic isomorphisms only.
there is a sequence of functions f1, f2, . . . such An invertible measure-preserving transforma-
that f1g [ U nT f j : n  ℤ, j  ℕ forms an ortho- tion T of (X, ℬ, m) is said to be K (for Kolmogorov)
if there is a sub-s-algebra F of ℬ such that
normal basis for L2(X).
To see that this property is stronger than 1. \1 n
n¼1 T F is the trivial s-algebra up to sets of
strong mixing, simply observe that it implies
measure 0 (i.e., the intersection consists only of
that U tT UnT f j , U m
T f k ! 0 as t ! 1. Then by null sets and sets of full measure).
approximating f and g by their expansions with 2. _1
n¼1 T F ¼ ℬ (i.e., the smallest s-algebra
n

respect to a finite part of the basis, it can be containing T nF for all n > 0 is ℬ.
deduced that U nT f , g ! h f , 1ih1, gi as required.
Since already strong mixing is atypical from the The K property has a useful reformulation in
topological point of view, it follows that countable terms of entropy as follows: T is K if and only if
Lebesgue spectrum has to be atypical. In fact, for every nontrivial partition P of X, the entropy
Yuzvinskii (1967) showed that the typical of T with respect to the partition P is positive:
Ergodicity and Mixing Properties 51

T has completely positive entropy. See the chapter In the case of invertible Bernoulli shifts, Ornstein
on Entropy in Ergodic Theory for the relevant (1970, 1974) developed in the early 1970s a power-
definitions. The equivalence of the K property ful isomorphism theory, showing that two Bernoulli
and completely positive entropy was shown by shifts are measure-theoretically isomorphic if and
Rokhlin and Sinai (1961). For a general transfor- only if they have the same entropy. Entropy had
mation T, one can consider the collection of all already been identified as an invariant by Kolmogo-
subsets B of X such that with respect to the parti- rov and Sinai (Kolmogorov 1958; Sinai 1959), so
tion P B ¼ fB, Bc g, hðP B Þ ¼ 0: One can show this established that it was a complete invariant for
that this is a s-algebra. This s-algebra is known Bernoulli shifts. Keane and Smorodinsky (1979a)
as the Pinsker s-algebra. The above reformulation gave a proof which showed that two Bernoulli shifts
allows us to say that a transformation is K if and of the same entropy are isomorphic using a conju-
only if it has a trivial Pinsker s-algebra. gating map that is continuous almost everywhere.
The K property implies countable Lebesgue With other authors, this theory was extended to
spectrum (see Parry’s (1981) book for a proof). show that the property of being isomorphic to a
To see that K is not implied by countable Bernoulli shift applied to a surprisingly large class
Lebesgue spectrum, it can be pointed out that of measure-preserving transformations (e.g., geode-
certain transformations derived from Gaussian sic flows on manifolds of constant negative curva-
systems (see, for example, the paper of Newton ture (Ornstein and Weiss 1973), aperiodic
and Parry (1966)) have countable Lebesgue spec- irreducible Markov chains (Friedman and Ornstein
trum but zero-entropy. 1970), toral automorphisms (Katznelson 1971), and
The fact that (two-sided) Bernoulli shifts have more generally many Gibbs measures for hyperbolic
the K property follows from Kolmogorov’s 0–1 dynamical systems (see the book of Bowen (1975)).
n
law by taking F ¼ _1 n¼0 T P , where P is the Initially, it was conjectured that the properties
partition into cylinder sets (see Williams’s (1991) of being K and Bernoulli were the same, but since
book for details of the 0–1 law). then a number of measure-preserving transforma-
Although the K property is explicitly an invert- tions that are K but not Bernoulli have been iden-
ible property, it has a noninvertible counterpart, tified. The earliest was due to Ornstein (1973a).
namely, exactness. A transformation T of (X, ℬ, m) Ornstein and Shields (1973) then provided an
n
is exact if \1
n¼0 T ℬ consists entirely of null sets uncountable family of nonisomorphic
and sets of measure 1. It is not hard to see that a K automorphisms. Katok (1980b) gave an exam-
noninvertible transformation is exact if and only if ple of a smooth diffeomorphism that is K but not
its natural extension is K. Bernoulli, and Kalikow (1982) gave a very natural
The final and strongest property in our list is probabilistic example of a transformation that has
that of being measure-theoretically isomorphic this property (the T, T 1 process).
to a Bernoulli shift. If T is measure-theoretically While in systems that one regularly encounters
isomorphic to a Bernoulli shift, it can be said that there is a correlation between positive entropy and
T has the Bernoulli property. While in principle the stronger mixing properties discussed, these
this could apply to both invertible and non- properties are logically independent; for example,
invertible transformations, in practice the defini- taking the product of a Bernoulli shift and the
tion applies to a large class of invertible identity transformation gives a positive entropy
transformations, but occurs comparatively sel- transformation that fails to be ergodic; also,
dom for noninvertible transformations. For this besides rank-one mixing examples, there are
reason, we will restrict ourselves to a discussion zero-entropy Gaussian systems which are strong
of the Bernoulli property for invertible transfor- mixing and have countable Lebesgue spectrum.
mations (see however work of Hoffman and In many of the mixing criteria discussed above,
Rudolph (2002) and Heicklen and Hoffman a pair of sets A and B is considered and one asks
(2002) for work on the one-sided Bernoulli for asymptotic independence of A and B (so that
property). for large n, A and T –n B become independent).
52 Ergodicity and Mixing Properties

It is natural to ask, given a finite collection Bergelson (1987) generalized this by showing that
of sets A0, A1, . . . Ak, under what conditions weak mixing implies a polynomial version of
mðA0 \ T n1 A1 \ . . . \ T nk Ak Þ converges to weak mixing of all orders:
Pkj¼0 m Aj :.
A measure-preserving transformation T is said lim m A0 \ T p1 ðnÞ A1 \ . . . \ T pk ðnÞ Ak
n!1, n J
to be mixing of order k þ 1 if for all measurable
sets A0, . . ., Ak, k
¼ mðAi Þ
i¼0
lim mðA0 \ T n1 A1 \ . . . \ T nk Ak Þ
n1 !1, njþ1 nj !1

k whenever p1(n), . . ., pk(n) are nonconstant

¼ m Aj : integer-valued polynomials such that pi(n) –
j¼0
pj(n) is unbounded for i 6¼ j. The proofs in
Furstenberg (1977) and Bergelson (1987) make
The transformation T is said to be mixing of all
use of a Hilbert space version of the van der
orders, if T is mixing of order k þ 1 for all natural
Corput inequality of analytic number theory.
numbers k. An outstanding open question asked
Moreover, this method is instrumental in
by Rokhlin (1949) appearing already in Halmos’s
Furstenberg’s ergodic proof (Furstenberg 1977)
(1956) book is to determine whether mixing
of Szemerédi’s theorem on the existence of arbi-
(i.e., mixing of order 2) implies mixing of all
trarily long arithmetic progressions in a subset of
orders. Kalikow (1984) showed that mixing
the integers of positive density (see the chapter
implies mixing of all orders for rank-one trans-
“Ergodic Theory: Interactions with Combinator-
formations (existence of rank-one mixing trans-
ics and Number Theory” for more information
formations having been previously established by
about this direction of study).
Ornstein (1972)). Later Ryzhikov (1993) used
The conclusions that one draws here are much
joining methods to establish the result for trans-
weaker than the requirement for mixing of all
formations with finite rank, and Host (1991) also
orders. For mixing of all orders, it was required
used joining methods to establish the result for
that provided the gaps between 0, n1, . . ., nk
measure-preserving transformations with singular
diverge to infinity, one achieves asymptotic inde-
spectrum, but the general question remains open.
pendence, whereas for these weak mixing results,
It is not hard to show using martingale argu-
the gaps are increasing along prescribed
ments that K automorphisms and hence all
sequences with regular growth properties.
Bernoulli measure-preserving transformations are
It is interesting to note that the analogous ques-
mixing of all orders.
tion of whether mixing implies mixing of all orders
In Furstenberg (1981), Furstenberg defined
is known to fail in higher-dimensional actions.
weak mixing of all orders as a natural analogue
Here, rather than a ℤ action, in which there is a
to the stronger condition of mixing of all
single measure-preserving transformation (so that
orders. A measure-preserving transformation T is
the integer n acts on a point x  X by mapping it to
weak mixing of all orders, if given measurable sets
Tnx), one takes a ℤd action. For such an action, one
A0, . . ., Ak, there is a subsequence J of the integers
has d commuting transformations T1, . . ., Td and a
of density 0 such that
vector (n1, . . ., nd) acts on a point x by sending it to
T n11   T ndd x: Ledrappier (1978) studied the
lim m A0 \ T n A1 \ . . . \ T kn Ak following two-dimensional action. Let
n!1n J
2
k X ¼ x  f0, 1gℤ : xv þ xvþe1 þ xvþe2 ¼ 0 ðmod2Þ
¼ mðAi Þ: and let T i ðxÞv ¼ xvþei : Since X is a compact
i¼0
Abelian group, it has a natural measure m invariant
In Furstenberg (1977), Furstenberg proved that under the group operations (the Haar
weak mixing implies weak mixing of all orders. measure). It is not hard to show that this
Ergodicity and Mixing Properties 53

system is mixing (i.e., given any measurable examples of C1 expanding maps of the interval for
sets A and B, m A \ T n 1 n2
1 T 2 B ! mðAÞmðBÞ as which Lebesgue measure was invariant, but
k(n1, n2)k ! 1). Ledrappier showed that the respectively not ergodic and not weak mixing.
system fails to be 3-mixing. Subsequently Some of the key tools in controlling mixing in
Masser (2004) established necessary and suffi- one-dimensional expanding maps that are absent
cient conditions for similar higher-dimensional in the C1 case are bounded distortion estimates.
algebraic actions to be mixing of order k but not Here, there is a constant 1  C < 1 such that
order k þ 1 for any given k. given any interval I on which some power Tn of
T acts injectively and any subinterval J of I, one
has 1/C  (| T nJ| /| T nI| )/(| J| /| I| )  C. An early
Hyperbolicity and Decay of Correlations place in which bounded distortion estimates
appear is the work of Rényi (1957).
One class of systems in which the stronger mixing One important class of results for expanding
properties are often found is the class of smooth maps establishes an exponential decay of correla-
systems possessing uniform hyperbolicity tions. Here, one starts with a pair of smooth func-
(i.e., the tangent space to the manifold at each tions f and g and one estimates f  g ∘ T ndm 
point splits into stable and unstable subspaces f dm g dm, where m is an absolutely continuous
Es(x) and Eu(x) such that the DT a<1 invariant measure. If m is mixing, it is expected
Es ðxÞ
that this will converge to 0. In fact though, in good
for all x and DT 1  a and DT(Es(x)) ¼ cases this converges to 0 at an exponential rate for
Eu ðxÞ
Es(T(x)) and DT(Eu(x)) ¼ Eu(T(x))). In some each pair of functions f and g belonging to a
cases, similar conclusions are found in systems sufficiently smooth class. In this case, the
possessing nonuniform hyperbolicity. See Katok measure-preserving transformation T is said to
and Hasselblatt’s (1995) book for an overview of have exponential decay of correlations. See
hyperbolic dynamical systems, as well as the Liverani’s (2004) article for an introduction to a
chapter in this volume on ▶ “Smooth Ergodic method of establishing this based on cones. Expo-
Theory.” nential decay of correlations implies in particular
In the simple case of expanding piecewise con- that the natural extension is Bernoulli.
tinuous maps of the interval (i.e., maps for which Hu (2004) has studied the situation of maps of
the absolute value of the derivative is uniformly the interval for which the derivative is bigger than
bounded below by a constant greater than 1), it is 1 everywhere except at a fixed point, where the
known that if they are totally ergodic and topo- local behavior is of the form x 7! x þ x1þα for
logically transitive (i.e., the forward images of any 0 < α < 1. In this case, rather than exhibiting
interval cover the entire interval), then provided exponential decay of correlations, the map has
that the map has sufficient smoothness (e.g., the polynomial decay of correlations with a rate
map is C1 and the derivative satisfies a certain depending on α.
additional summability condition), the map has a In Young’s (1995) survey, a variety of tech-
unique absolutely continuous invariant measure niques are outlined for understanding the strong
which is exact and whose natural extension is ergodic properties of nonuniformly hyperbolic
Bernoulli (see the paper of Góra (1994) for results diffeomorphisms. In her article (Young 1998),
of this type proved under some of the mildest methods are introduced for studying many classes
hypotheses). These results were originally of nonuniformly hyperbolic systems by looking at
established for maps that were twice continuously suitably high powers of the map, for which the
differentiable, and the hypotheses were progres- power has strong hyperbolic behavior. The article
sively weakened, approaching, but never meeting, shows how to understand the ergodic behavior of
C1. Subsequent work of Quas (1996b, a) provided these systems. These methods are applied (for
54 Ergodicity and Mixing Properties

example) to billiards, one-dimensional quadratic interval exchange transformations. This is an

maps, and Hénon maps. extension of finite interval exchange transforma-
tions except the domain and range are partitioned
into a countable number of intervals. It was shown
Representations, Realizations, and in Arnoux and Weiss (1985) that every ergodic
Genericity invertible measure-preserving transformation on a
Lebesgue probability space is isomorphic to an
There are multiple representations of ergodic infinite interval exchange transformation.
invertible measure-preserving transformations. A similar but itself useful technique for
One of the most common representations is as a constructing a transformation is through cutting
shift transformation on doubly infinite sequences and stacking. It was also shown in Arnoux and
of characters from a finite alphabet. This can be Weiss (1985) that every ergodic invertible
made into a compact metric space by setting the measure-preserving transformation on a Lebesgue
distance between two infinite sequences x ¼ (xi) probability space has an isomorphic version
and y ¼ (yi) as which is obtained by cutting and stacking. It
may be that the supremum of the number of col-
d ðx, yÞ ¼ j x0  y0 j umns goes to infinity. In this case, if there is not an
1 isomorphic version with a bounded number of
þ 2i ðjxi yi j þ jxi  yi jÞ: columns, the transformation has infinite rank.
i¼1 Finally, another method for obtaining an ergo-
ð3Þ dic invertible measure-preserving transformation
is via an adic transformation on a Bratteli-Vershik
If T has zero-entropy, then there is a two set diagram (Vershik 1981; Vershik and Livshits
i
partition P ¼ {p0, p1} such that _1 i¼0 T P generates 1992). It is similar to the cutting and stacking
the sigma algebra. Pick a point x that is typical for procedure, although adic transformations encode
T, and let xi ¼ 0, if T ix  p0, otherwise set xi ¼ 1. points as infinite paths in a directed acyclic graph.
Let X be the closure of all left and right shifts of the Figure 3 displays an adic representation of the
sequence x. The left shift operator on the sequences Chacon-3 transformation.
in X is isomorphic to T. This procedure can be
extended to realize any ergodic invertible measure- Topological and Set-Theoretic Properties of
preserving transformation with finite entropy. They Classes of Transformations
can all be realized by a shift on a compact metric It is of interest to understand the behavior of a
space (Jewett 1970; Krieger et al. 1972). The shift “typical” measure-preserving transformation.
will be continuous in the topology induced by the
distance d defined in 3. Using an inverse limit of
refining partitions, it is possible to construct a
homeomorphism realization of T. Also, T can be
constructed such that it is strictly ergodic which
3
means there is only one invariant ergodic measure, 2 4
and every point is typical. 1
Later, it was shown by Lehrer (1987) that it is
3
possible to obtain an isomorphic transformation 2 4
1
that is strictly ergodic and topologically mixing.
This is further validation that topological
mixing is not a measure theoretic isomorphic
invariant.
Other representations of ergodic invertible Ergodicity and Mixing Properties, Fig. 3 Chacon-
measure-preserving transformations are as infinite 3 Adic
Ergodicity and Mixing Properties 55

There are a number of Baire category results invariants for determining when two transforma-
addressing this. In order to state them, one needs tions are isomorphic. However, it is also shown
a set of measure-preserving transformations and a that there are topologically generic classes of
topology on them. As mentioned earlier, it is effec- transformations (i.e., rank-one) that are Borel. At
tively no restriction to assume that a transformation this time, it is an open question to establish an
is a Lebesgue-measurable map on the unit interval easily verifiable method for determining when
preserving Lebesgue measure. The classical cate- two rank-one transformations are isomorphic.
gory results are then on the collection of invertible Nevertheless, there has been recent progress
Lebesgue-measure-preserving transformations of toward characterizing rank-one transformations;
the unit interval. One topology on these is the see Gao and Hill (2014, b); Adams et al. (2017);
“weak” topology, where a subbase is given and Foreman et al. (2019).
by sets of the form N(T, A, ϵ) ¼ {S; l(S(A)
ΔT(A)) < ϵ}. With respect to this topology, Halmos
(1944) showed that a residual set (i.e., a dense Gδ Future Directions
set) of invertible measure-preserving transforma-
tions is weak mixing (see also work of Alpern Problem 1 (Mixing of all orders). Does mixing
(1976)), while Rokhlin (1948) showed that the set imply mixing of all orders? Can the results of
of strong mixing transformations is meagre (i.e., a Kalikow, Ryzhikov, and Host be extended to
nowhere dense Fs set), allowing one to conclude larger classes of measure-preserving transforma-
that with respect to this topology, a typical trans- tions? Thouvenot observed that it is sufficient to
formation is weak but not strong mixing. establish the result for measure-preserving trans-
In Tikhonov (2007), Tikhonov introduces a nat- formations of entropy 0. This observation (whose
ural metric (i.e., leash topology) which makes the proof is based on the Pinsker s-algebra) was
set of mixing transformations into a complete sep- stated in Kalikow’s (1984) paper and is
arable metric space. It is shown that in this metric, a reproduced as Proposition 3.2 in recent work of
generic mixing transformation is mixing of all de la Rue (2006) on the mixing of all orders
orders and has simple singular spectrum and all problem.
its powers are disjoint. See the chapter ▶ “Spectral
Theory of Dynamical Systems” (Lemancyzk and Problem 2 (Multiple weak mixing). As men-
Kanigowski 2022) for further details on simple tioned above, Bergelson (1987) showed that if
singular spectrum, as well as the notion of T is a weak mixing transformation, then there is
disjointness. While it is shown that the conjugacy a subset J of the integers of density 0 such that
class of a generic mixing transformation is dense,
not until Bashtanov (2013), is it shown that the lim m A0 \ T p1 ðnÞ A1 \ . . . \ T pk ðnÞ Ak
n!1, n J
conjugacy class of every mixing transformation is
k
dense in the leash topology. Also, in Bashtanov
¼ mðAi Þ
(2013), it is shown that rank-one is generic in the i¼0
leash topology.
Viewing classes of ergodic measure-preserving whenever p1(n), . . ., pk(n) are nonconstant integer-
transformations from a descriptive set theoretic valued polynomials such that pi(n) – pj(n) is
viewpoint (Becker and Kechris 1996; Foreman unbounded for i 6¼ j. It is natural to ask what is
2000) has shed new light on the challenges char- the most general class of times that can replace the
acterizing broad classes of transformations. In sequences ( p1(n)), . . ., ( pk(n)). In unpublished
particular, in Foreman and Weiss (2011), it is notes, Bergelson and Håland considered as times
shown that the isomorphism relation is a complete the values taken by a family of integer-valued
analytic set and in particular, not Borel. This gives generalized polynomials (those functions of an
an obstacle to concrete (countably) verifiable integer variable that can be obtained by the
56 Ergodicity and Mixing Properties

operations of addition, multiplication, addition of machinery by Ornstein and Weiss (1975). The
or multiplication by a real constantpand taking Weak Pinsker Conjecture states that if a measure-
integer
p parts (e.g., gðnÞ ¼ ⎿ 2⎿pn⏌ þ preserving transformation T has entropy h > 0,
⎿ 3nÞ ). They conjectured necessary and suffi- then for all ϵ > 0, T may be expressed as a product
cient conditions for the analogue of Bergelson’s of a Bernoulli shift and a measure-preserving
weak mixing polynomial ergodic theorem to hold transformation with entropy less than ϵ.
and proved the conjecture in certain cases. This problem was recently solved in the affir-
In a recent paper of McCutcheon and Quas mative by T. Austin (2018). New results on mea-
(2007), the analogous question was addressed in sure concentrations were developed to prove the
the case where T is a mild-mixing transformation. weak Pinsker conjecture.

Problem 3 (Pascal adic transformation). Vershik References

(1974, 1981) introduced a family of transformations
known as the adic transformations. The underlying Aaronson J (1997) An introduction to infinite ergodic
spaces for these transformations are certain spaces theory. American Mathematical Society
Aaronson J, Hosseini M, Lemanczyk M (2014) Ip-rigidity
of paths on infinite graphs and the transformations
and eigenvalue groups. Ergodic Theory Dynam Syst
act by taking a path to its lexicographic neighbor. 34(4):105–1076
Among the adic transformations, the so-called Pas- Abramov L (1959) The entropy of a derived automor-
cal adic transformation (so-called because the under- phism. Dokl Akad Nauk SSSR 128:647–650
Adams TM (1998) Smorodinsky’s conjecture on rank-one
lying graph resembles Pascal’s triangle) has been
mixing. Proc Am Math Soc 126(3):739–744
singled out for attention in work of Petersen and Adams TM (2015) Tower multiplexing and slow weak
others (Petersen and Schmidt 1997; Méla and mixing. Colloq Math 1(138):47–71
Petersen 2005; Berend and Kolesnik 2001; Adams Adams TM, Nobel AB (2010) Uniform convergence of
vapnik-chervonenkis classes under ergodic sampling.
and Petersen 1998). In particular, it is unresolved
Ann Probab 38(4):1345–1367
whether this transformation is weak mixing with Adams TM, Petersen K (1998) Binomial coefficient mul-
respect to any of its ergodic measures. Weak mixing tiples of irrationals. Monatsh Math 125:269–278
has been shown to follow from a number-theoretic Adams TM, Ferenczi S, Petersen K (2017) Constructive
symbolic presentations of rank one measure-preserving
condition on the binomial coefficients (Berend and
systems. Colloq Math 150:243–255
Kolesnik 2001; Adams and Petersen 1998). Adler RL, Shields P, Smorodinsky M (1972) Irreducible
markov shifts. Ann Math Stat 43(3):1027–1029
Problem 4 (Weak Pinsker Conjecture - Alpern S (1976) New proofs that weak mixing is
generic. Invent Math 32:263–278
SOLVED). Pinsker (1960) conjectured that in a
Ambrose W, Kakutani S (1942) Structure and continuity of
measure-preserving transformation with positive measurable flows. Duke Math J 9(1):25–42
entropy, one could express the transformation as a Arnoux DSOP, Weiss B (1985) Cutting and stacking, inter-
product of a Bernoulli shift with a system with val exchanges and geometric models. Israel J Math 50:
160–168
zero entropy. This conjecture (now known as the
Arora S, Ge R, Neyshabur B, Zhang Y (2018) Stronger
Strong Pinsker Conjecture) was shown to be false generalization bounds for deep nets via a compression
by Ornstein (1973b, 1973c). Shields and approach. In: Krause A, Dy J (eds) 35th international
Thouvenot (1975) showed that the collection of conference on machine learning, ICML 2018. Interna-
tional Machine Learning Society (IMLS), pp 390–418
transformations that can be written as a product of
Arora S, Du S, Hu W, Li Z, Wang R (2019) Fine-grained
a zero-entropy transformation with a Bernoulli analysis of optimization and generalization for over-
shift is closed in the so-called d  metric that parameterized two-layer neural networks. In:
lies at the heart of Ornstein’s theory. Chaudhuri K, Salakhutdinov R (eds) Proceedings of
the 36th international conference on machine learning,
It is, however, the case that if T: X ! X has volume 97 of proceedings of machine learning
entropy h > 0, then for all h0  h, T has a factor research. PMLR, Long Beach, pp 322–332
S with entropy h0 (this was originally proved by Austin T (2010) On the norm convergence of non-
Sinai (1964) and reproved using the Ornstein conventional ergodic averages. Ergodic Theory
Dynam Syst 30(2):321–338
Ergodicity and Mixing Properties 57

Austin T (2018) Measure concentration and the weak de la Rue T (2006) 2-fold and 3-fold mixing: why
pinsker property. Publications mathématiques de 3-dot-type counterexamples are impossible in one
l’IHÈS 128(1):1–119 dimension. Bull Braz Math Soc (NS) 37(4):503–521
Avila A, Forni G (2007) Weak-mixing for interval Du SS, Zhai X, Poczos B, Singh A (2019) Gradient descent
exchange transformations and translation flows. Ann provably optimizes over-parameterized neural net-
Math 165:637–664 works. In: International conference on learning
Bashtanov AI (2013) Generic mixing transformations are representations 2019
rank 1. Math Notes 93:209–216 Ehrenfest P, Ehrenfest T (1911) Begriffliche Grundlage der
Bayless RL, Yancey KB (2015) Weakly mixing and rigid statistischen Auffassung in der Mechanik. Number 4 in
rank-one transformations preserving an infinite mea- Encycloplädie der mathematischen Wissenschaften.
sure. New York J Math 21:615–636 Teubner
Becker H, Kechris AS (1996) The descriptive set theory of Eisner T, Grivaux S (2011) Hilbertian Jamison sequences
polish group actions, volume 232. Cambridge Univer- and rigid dynamical systems. J Funct Anal 261(7):
sity Press 2013–2052
Berend MBD, Kolesnik G (2001) Irrational dilations of Fayad B, Kanigowski A (2015) Rigidity times for a weakly
Pascal’s triangle. Mathematika 48:159–168 mixing dynamical system which are not rigidity times
Bergelson V (1987) Weakly mixing pet. Ergodic Theory for any irrational rotation. Ergodic Theory Dynam Syst
Dynam Syst 7:337–349 35(8):2529–2534
Bergelson V, Leibman A (2002) A nilpotent roth theorem. Feller W (1950) An introduction to probability and its
Inventiones mathematicae 147(2):429–470 applications. Wiley
Bergelson V, del Junco A, Lemańczyk M, Rosenblatt Ferenczi S (1997) Systems of finite rank. Colloq Math
J (2014) Rigidity and non-recurrence along sequences. 73(1):35–65
Ergodic Theory Dynam Syst 34(5):1464–1502 Fields medalists 2014 awardees with brief citations. https://
Birkhoff GD (1931) Proof of the ergodic theorem. Proc Nat www.mathunion.org/imu-awards/fields-medal/fields-
Acad Sci 17:656–660 medal-2014/fields-medallists-2014-awardees-brief-
Birkhoff GD, Smith PA (1924) Structural analysis of sur- citations
face transformations. J Undergrad Math 7:345–379 Foreman M (2000) A descriptive view of ergodic theory.
Boltzmann L (1871) Einige allgemeine Sätze über Descriptive Set Theory Dynam Syst:87–171
Wärmegleichgewicht. Wiener Berichte 63:679–711 Foreman DJRM, Weiss B (2011) The conjugacy problem
Boltzmann L (1909) Wissenschaftliche Abhandlungen. in ergodic theory. Ann Math 173(3):1529–1586
Akad. Wissenschaften Berlin Foreman M et al (2019) Rank-one transformations, odom-
Bowen R (1975) Equilibrium states and the ergodic theory eters, and finite factors. arXiv:1910.126J5, pp 1–20
of Anosov diffeomorphisms. Springer Frantzikinakis N, Kra B (2005) Convergence of multiple
Bozgan F, Sanchez A, Silva CE, Stevens D, Wang J (2015) ergodic averages for some commuting transformations.
Subsequence bounded rational ergodicity of rank-one Ergodic Theory Dynam Syst 25(3):799–809
transformations. Dyn Syst 30(1):70–84 Friedman NA (1970) Introduction to ergodic theory, Van
Bradley RC (2005) Basic properties of strong mixing con- Nostrand Reinhold Mathematical Studies, No. 29. Van
ditions. A survey and some open questions. Probab Nostrand Reinhold Co, New York-Toronto,
Surv 2:107–144 Ont.-London
Chacon RV (1966a) Change of velocity in flows. J Math Friedman N (1989) Partial mixing, partial rigidity, and
Mechanics 16(5):417–431 factors. Contemp Math 94:141–145
Chacon RV (1966b) Transformations having continuous Friedman N (1992) Replication and stacking in ergodic
spectrum. J Math Mechanics 16(5):399–415 theory. Am Math Mon 99(1):31–41
Chacon RV (1969) Weakly mixing transformations which Friedman N, King J (1991) Rank one lightly mixing. Israel
are not strongly mixing. Proc Am Math Soc 22:559–562 J Math 73:281–288
Choquet G (1956a) Existence des représentations Friedman NA, Ornstein DS (1970) On isomorphism of
intégrales au moyen des points extrémaux dans les weak Bernoulli transformations. Adv Math 5:365–394
cônes convexes. C R Acad Sci Paris 243:699–702 Friedman NA, Ornstein DS (1973) Ergodic transforma-
Choquet G (1956b) Unicité des représentations intégrales tions induce mixing transformations. Adv Math 10(1):
au moyen de points extrémaux dans les cônes convexes 147–163
réticulés. C. R. Acad Sci Paris 243:555–557 Furstenberg H (1977) Ergodic behavior of diagonal mea-
Conze J-P, Lesigne E (1984) Théorèmes ergodiques pour sures and a theorem of Szemerédi on arithmetic pro-
des mesures diagonales. Bull de la Soc Math de France gressions. J Analyse Math 31:204–256
112:143–175 Furstenberg H (1981) Recurrence in ergodic theory and
Creutz D, Silva CE (2010) Mixing on rank-one transfor- combinatorial number theory. Princeton
mations. Stud Math 199(1):43–72 Furstenberg H, Katznelson Y (1978) An ergodic
Danilenko AI (2016) Mixing actions of zero entropy for Szemerédi theorem for commuting transformations.
countable amenable groups. In: Colloquium J d’Analyse Math 34(1):275–291
Mathematicum, volume 145. Instytut Matematyczny Furstenberg H, Weiss B (1978) The finite multipliers of
Polskiej Akademii Nauk, pp 179–186 infinite ergodic transformations. In: The structure of
58 Ergodicity and Mixing Properties

attractors in dynamical systems (Proc. Conf., North Katok A, Hasselblatt B (1995) Introduction to the modern
Dakota State Univ., Fargo, N.D., 1977). Springer theory of dynamical systems, Cambridge
Gao S, Hill A (2014) A model for rank one measure Katok A, Stepin AM (1967) Approximations in ergodic
preserving transformations. Topol Appl 174:25–40 theory. Uspekhi Mat Nauk 22:5(137):81–106
Garsia AM (1965) A simple proof of E. Hopf’s maximal Katznelson Y (1971) Ergodic automorphisms of n are
ergodic theory. J Math Mech 14:381–382 Bernoulli shifts. Israel . Math 10:186–195
Góra P (1994) Properties of invariant measures for piece- Katznelson Y, Weiss B (1982) A simple proof of some
wise expanding one-dimensional transformations with ergodic theorems. Israel J Math 42:291–296
summable oscillations of derivative. Ergodic Theory Keane MS, Petersen KE (2006) Nearly simultaneous pro-
Dynam Syst 14:475–492 ofs of the ergodic theorem and maximal ergodic theo-
Green B, Tao T (2008) The primes contain arbitrarily long rem. In: Dynamics and stochastics: Festschrift in Honor
arithmetic progressions. Ann Math 167(2):481–547 of M. S. Keane. Institute of Mathematical Statistics,
Grivaux S (2013) IP-Dirichlet measures and IP-rigid pp 248–251
dynamical systems: an approach via generalized riesz Keane M, Smorodinsky M (1979a) Bernoulli schemes of
products. Stud Math 215(3):237–259 the same entropy are finitarily isomorphic. Ann Math
Grivaux S, Roginskaya M (2013) Some new examples of 109:397–406
recurrence and non-recurrence sets for products of rota- Keane M, Smorodinsky M (1979b) Finitary isomorphisms of
tions on the unit circle. Czechoslov Math J 63(3): irreducible markov shifts. Israel J Math 34(4):281–286
603–627 Kolmogorov AN (1958) New metric invariant of transitive
Halmos PA (1944) In general a measure-preserving trans- dynamical systems and endomorphisms of Lebesgue
formation is mixing. Ann Math 45:786–792 spaces. Dokl Russ Acad Sci 119:861–864
Halmos P (1956) Lectures on ergodic theory. Chelsea Koopman BO (1931) Hamiltonian systems and Hilbert
Halmos P, von Neumann J (1942) Operator methods in space. Proc Nat Acad Sci 17:315–218
classical mechanics II. Ann. Math 43:332–350 Kramli A, Simányi N, Szász D (1991) The K-property of
Hoffman C, Heicklen D (2002) Rational maps are d-adic three billiard balls. Ann Math 133:37–72
Bernoulli. Ann Math 156:103–114 Krieger W et al (1972) On unique ergodicity. In: Proceed-
Hoffman C, Rudolph DJ (2002) Uniform endomorphisms ings of the sixth Berkeley symposium on mathematical
which are isomorphic to a Bernoulli shift. Ann Math statistics and probability, volume 2: Probability theory.
76:79–101 The Regents of the University of California
Host B (1991) Mixing of all orders and pairwise indepen- Le AN (2017) Nilsequences and multiple correlations
dent joinings of systems with singular spectrum. Israel along subsequences. Ergodic Theory Dynam Syst:1–21
J Math 76:289–298 Ledrappier F (1978) Un champ Markovien peut être
Hu H (2004) Decay of correlations for piecewise smooth d’entropie nulle et mélangeant. C R Acad Sci Paris
maps with indifferent fixed points. Ergodic Theory Sér A-B 287:561–563
Dynam Syst 24:495–524 Lehrer E (1987) Topological mixing and uniquely ergodic
Impagliazzo R, Kabanets V (2010) Constructive proofs of systems. Israel J Math 57(2):239–255
concentration bounds. In: Serna M, Shaltiel R, Lemancyzk M, Kanigowski A (2022) Spectral theory of
Jansen K, Rolim J (eds) Approximation, randomiza- dynamical systems. Encyclopedia of Complexity and
tion, and combinatorial optimization. algorithms and Systems Science
techniques. Springer, Berlin, Heidelberg, pp 617–631 Liverani C (2004) Decay of correlations. Ann Math 159:
Jewett RI (1970) The prevalence of uniquely ergodic sys- 1275–1312
tems. J Math Mechanics 19(8):717–729 Marcus B, Petersen K, Weissman T (2011) Entropy of
Kakutani S (1943) 131. Induced measure preserving trans- hidden Markov processes and connections to dynami-
formations. Proc Imperial Acad 19(10):635–641 cal systems. London Mathematical Society
Kakutani S (1986) Selected papers. Birkhauser Maruyama G (1949) The harmonic analysis of stationary
Kalikow S (1982) T, T1 transformation is not loosely stochastic processes. Memoirs of the Faculty of Sci-
Bernoulli. Ann Math 115:393–409 ence, Kyushu University. Series A Math 4(1):45–106
Kalikow S (1984) Twofold mixing implies threefold Masser DW (2004) Mixing and linear equations over
mixing for rank one transformations. Ergodic Theory groups in positive characteristic. Israel J Math 142:
Dynam Syst 2:237–259 189–204
Kamae T (1982) A simple proof of the ergodic theorem Masur H (1982) Interval exchange transformations and
using non-standard analysis. Israel J Math 42:284–290 measured foliations. Ann Math 115:169–200
Kammeyer JW (1990) A complete classification of the McCutcheon R, Quas A (2007) Generalized polynomials
two-point extensions of a multidimensional Bernoulli and mild mixing systems. Can J Math. to appear
shift. J d Analyse Math 54(1):113–163 McGoff ANK, Mukherjee S, Pillai N (2015) Consistency
Katok A (1980a) Interval exchange transformations and of maximum likelihood estimation for some dynamical
some special flows are not mixing. Israel J Math 35: systems. Ann Stat 43(1):1–29
301–310 Méla X, Petersen K (2005) Dynamical properties of the
Katok A (1980b) Smooth non-Bernoulli K-automor- Pascal adic transformation. Ergodic Theory Dynam
phisms. Invent Math 61:291–299 Syst 25:227–256
Ergodicity and Mixing Properties 59

Muehlegger E, Raich A, Silva C, Zhao W (1997) Lightly Robertson D (2019) Mild mixing of certain interval-
mixing on dense algebras. Real Anal Exchange 23: exchange transformations. Ergodic Theory Dynam
259–265 Syst 39(1):248–256
Newton D, Parry W (1966) On a factor automorphism of a Rokhlin VA (1948) A ‘general’ measure-preserving trans-
normal dynamical system. Ann Math Statist 37: formation is not mixing. Dokl Akad Nauk SSSR Ser
1528–1533 Mat 60:349–351
Ornstein DS (1970) Bernoulli shifts with the same entropy Rokhlin VA (1949) On endomorphisms of compact com-
are isomorphic. Adv Math 4:337–352 mutative groups. Izvestiya Akad Nauk SSSR Ser Mat
Ornstein DS (1972) On the root in ergodic theory. In: 13:329–340
Proceedings of the sixth berkeley symposium on math- Rokhlin VA, Sinai Y (1961) Construction and properties of
ematical statistics and probability (Univ. California, invariant measurable partitions. Dokl Akad Nauk
Berkeley, Calif., 1970/1971), Vol. II: Probability the- SSSR 141:1038–1041
ory. University California Press, pp 347–356 Rudin W (1966) Real and complex analysis. McGraw Hill
Ornstein DS (1973a) An example of a Kolmogorov auto- Rudolph D (1976) A two-valued step coding for ergodic
morphism that is not a Bernoulli shift. Adv Math 10: flows. Math Z 150(3):201–220
49–62 Rudolph DJ (1978) If a two-point extension of a Bernoulli
Ornstein DS (1973b) A K-automorphism with no square shift has an ergodic square, then it is Bernoulli. Israel
root and Pinsker’s conjecture. Adv Math 10:89–102 J Math 30(1–2):159–180
Ornstein DS (1973c) A mixing transformation for which Rudolph DJ (1990) Fundamentals of measurable dynam-
Pinsker’s conjecture fails. Adv Math 10:103–123 ics. Oxford
Ornstein DS (1974) Ergodic theory, randomness, and Ryzhikov VV (1993) Joinings and multiple mixing of the
dynamical systems. Yale University Press actions of finite rank. Funct Anal Appl 27:128–140
Ornstein DS, Shields PC (1973) An uncountable family of Schmidt JP, Siegel A, Srinivasan A (1993) Chernoff-
K-automorphisms. Adv Math 10:89–102 hoeffding bounds for applications with limited indepen-
Ornstein DS, Weiss B (1973) Geodesic flows are dence. In: Proceedings of the fourth annual ACM-SIAM
Bernoullian. Israel J Math 14:184–198 symposium on discrete algorithms, SODA 93. Society
Ornstein DS, Weiss B (1975) Unilateral codings of for Industrial and Applied Mathematics, p 331340
Bernoulli systems. Israel J Math 21:159166 Shields P, Thouvenot J-P (1975) Entropy zero
Bernoulli
Oxtoby JC (1952) Ergodic sets. Bull Amer Math Soc 58: processes are closed in the d  metric. Ann Probab 3:
116–136 732–736
Parry W (1981) Topics in ergodic theory, Cambridge Silva CE (2008) Invitation to ergodic theory. Am Math
Pelekis C, Ramon J (2017) Hoeffding’s inequality for sums of Soc 42
dependent random variables. Mediterr J Math 14(6):243 Simányi N (2003) Proof of the Boltzmann-Sinai ergodic
Petersen K (1983) Ergodic theory. Cambridge hypothesis for typical hard disk systems. Invent Math
Petersen K, Schmidt K (1997) Symmetric Gibbs measures. 154:123–178
Trans Am Math Soc 349:2775–2811 Simányi N (2004) Proof of the ergodic hypothesis for typical
Phelps R (1966) Lectures on Choquet’s theorem. Van hard ball systems. Ann Henri Poincare 5:203–233
Nostrand Simányi N, Szász D (1999) Hard ball systems are
Pinsker MS (1960) Dynamical systems with completely completely hyperbolic. Ann Math 149:35–96
positive or zero entropy. Soviet Math Dokl 1:937–938 Sinai YG (1959) On the notion of entropy of a dynamical
Pollard D (1990) Empirical processes: theory and applica- system. Dokl Russ Acad Sci 124:768–771
tions. In: NSF-CBMS regional conference series in Sinai YG (1964) On a weak isomorphism of transforma-
probability and statistics. JSTOR, pp 1–86 tions with invariant measure. Mat Sb (NS) 63:23–42
Quas A (1996a) A C1 expanding map of the circle which is Sinai YG (1970) Dynamical systems with elastic reflec-
not weak-mixing. Israel J Math 93:359–372 tions. Ergodic properties of dispersing billiards. Uspehi
Quas A (1996b) Non-ergodicity for C1 expanding maps Mat Nauk 25:141–192
and g-measures. Ergodic Theory Dynam Syst 16: Sinai Y (1976) Introduction to ergodic theory. Princeton.
531–543 translation of the 1973 Russian original
Raginsky M, Rakhlin A, Telgarsky M (2017) Non-convex Sinai YG, Chernov NI (1987) Ergodic properties of some
learning via stochastic gradient langevin dynamics: a systems of two-dimensional disks and three-
nonasymptotic analysis. In: Kale S, Shamir O (eds) dimensional balls. Uspekhi Mat Nauk 42:153–174
Proceedings of the 2017 Conference on learning theory, Smorodinsky M (1971) A partition on a Bernoulli shift
volume 65 of Proceedings of machine learning which is not weakly Bernoulli. Math Syst Th 5:
research. PMLR, Amsterdam, pp 1674–1703 201–203
Rényi A (1957) Representations for real numbers and their Tao T (2008) Norm convergence of multiple ergodic aver-
ergodic properties. Acta Math Acad Sci Hungar 8: ages for commuting transformations. Ergodic Theory
477–493 Dynam Syst 28(2):657–688
Riesz F (1938) Some mean ergodic theorems. J London Tikhonov SV (2007) A complete metric in the set of
Math Soc 13:274–278 mixing transformations. Mat Sb 198(4):135–158
60 Ergodicity and Mixing Properties

Vapnik V (2013) The nature of statistical learning theory. von Neumann J (1932) Proof of the quasi-ergodic hypoth-
Springer esis. Proc Natl Acad Sci U S A 18:70–82
Vapnik VN, Chervonenkis AY (2015) On the uniform Walters P (1972) Some invariant sigma algebras for mea-
convergence of relative frequencies of events to their sure preserving transformations. Trans Am Math Soc
probabilities. Springer, Cham, pp 11–30 163:357–368
Veech W (1982) Gauss measures for transformations on Walters P (1982) An introduction to ergodic theory.
the space on interval exchange maps. Ann Math 115: Springer
201–242 Williams D (1991) Probability with martingales. Cambridge
Vershik A (1974) A description of invariant measures for Young L-S (1995) Ergodic theory of differentiable dynam-
actions of certain infinite-dimensional groups. Soviet ical systems. In: Real and complex dynamical systems
Math Dokl 15:1396–1400 (Hillerød, 1993). Kluwer, pp 293–336
Vershik A (1981) Uniform algebraic approximation of shift Young L-S (1998) Statistical properties of dynamical sys-
and multiplication operators. Soviet Math Dokl 24: tems with some hyperbolicity. Ann Math 147:585–650
97–100 Yuzvinskii SA (1967) Metric automorphisms with a simple
Vershik A, Livshits A (1992) Adic models of ergodic spectrum. Soviet Math. Dokl. 8:243–245
transformations, spectral theory, substitutions, and Zhang Q (1996) On convergence of the averages.
related topics. Rep Theory Dynam Syst 9:185–204 Monatshefte für Mathematik 122(3):275–300
 T:  ! , defined by Tx ¼ 2x mod 1, preserves
Ergodic Theory: Recurrence Lebesgue measure, hence induces a measure
preserving system on .
Nikos Frantzikinakis1 and Randall McCutcheon2 Ergodic system Is a measure preserving system
1
Department of Mathematics, University of Crete, (X, ℬ, m, T) (finite or infinite) such that every
Heraklion, Greece A  ℬ that is T-invariant (i.e., T 1A ¼ A)
2
Department of Mathematics, University of satisfies either m(A) ¼ 0 or m(X \ A) ¼ 0. (One
Memphis, Memphis, TN, USA can check that the rotation Ra is ergodic if and
only if a is irrational and that the doubling map
is ergodic.)
Article Outline Ergodic decomposition Every measure-preserving
system (X, X, m, T) can be expressed as an integral
Glossary of ergodic systems; for example, one can write
Definition of the Subject and Its Importance m ¼ mt dl(t), where l is a probability measure on
Introduction [0, 1] and mt are T-invariant probability measures
Quantitative Poincaré Recurrence on (X, X ) such that the systems (X, X , mt, T) are
Subsequence Recurrence ergodic for t  [0, 1].
Multiple Recurrence Ergodic theorem States that if (X, ℬ, m, T) is a
Connections with Combinatorics and Number measure preserving system and f  L2(m), then
Theory lim N!1 N1 Nn¼1 T n f  P f L2 ðmÞ ¼ 0, where
Future Directions
Pf denotes the orthogonal projection of the
References
function f onto the subspace {f  L2(m): T
f ¼ f}.
Glossary Hausdorff a-measure Let (X, ℬ, m, T) be a mea-
sure preserving system endowed with a
Almost every, essentially Given a Lebesgue
m-compatible metric d. The Hausdorff a-measure
measure space (X, ℬ, m), a property P(x) pred- ℋa(X) of X is an outer measure defined for all
icated of elements of X is said to hold for subsets of X as follows: First, for A  X and
almost every x  X, if the set X \ {x: P (x) 1 a
e > 0, let ℋa,e ðAÞ ¼ inf n¼1 r i , where the
holds} has zero measure. Two sets A, B  ℬ
infimum is taken over all countable coverings of
are essentially disjoint if m(A \ B) ¼ 0.
A by sets Ui  X with diameter ri < e. Then
Conservative system Is an infinite measure pre-
define Ha(A) ¼ lim supe!0 ℋa,e(A).
serving system such that for no set A  ℬ with
Infinite measure-preserving system Same as
positive measure are A, T 1A, T 2A, . . .
measure preserving system, but m(X) ¼ 1.
pairwise essentially disjoint.
Invertible system Is a measure-preserving sys-
(cn)-Conservative system If (cn)n  ℕ is a decreas-
tem (X, ℬ, m, T) (finite or infinite), with the
ing sequence of positive real numbers, a conser-
property that there exists X0  X, with
vative ergodic measure preserving transformation
m(X\X0) ¼ 0, and such that the transformation
T is (cn)-conservative if for some nonnegative
T: X0 ! X0 is bijective, with T 1 measurable.
function f  L1(m), 1 n¼1 cn f ðT xÞ ¼ 1 a.e.
n
Measure-preserving system Is a quadruple (X,
Doubling map If  is the interval [0, 1] with its
ℬ, m, T), where X is a set, ℬ is a s-algebra of
endpoints identified and addition performed
subsets of X (i.e., ℬ is closed under countable
modulo 1, the (non-invertible) transformation

© Springer Science+Business Media, LLC, part of Springer Nature 2023 61

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_184
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2021
https://doi.org/10.1007/978-3-642-27737-5_184-3
62 Ergodic Theory: Recurrence

unions and complementation), m is a probabil- principle was first exploited by Poincaré in his
ity measure (i.e., a countably additive function 1890 King Oscar prize-winning memoir that stud-
from ℬ to [0, 1] with m(X) ¼ 1), and T: X ! X is ied planetary motion. Using the prototype of an
measurable (i.e., T1A ¼ {x  X: T x  ergodic theoretic argument, he showed that in any
A}  ℬ for A  ℬ) and m-preserving (i.e., system of point masses having fixed total energy
m(T 1A) ¼ m(A)). Moreover, throughout the that restricts its dynamics to bounded subsets of its
discussion, we assume that the measure space phase space, the typical state of motion (characterized
(X, ℬ, m) is Lebesgue (see Section 1.0 of by configurations and velocities) must recur to an
(Aaronson 1997)). arbitrary degree of approximation.
m-Compatible metric Is a separable metric on Among the recurrence principle’s more spec-
X, where (X, ℬ, m) is a probability space, hav- tacularly counterintuitive ramifications is that iso-
ing the property that open sets measurable. lated ideal gas systems that do not lose energy will
Positive definite sequence Is a complex-valued return arbitrarily closely to their initial states, even
sequence (an)n  ℤ such that for any when such a return entails a decrease in entropy
z1 , . . . , zk  ℂ, ki,j¼1 aij zi z j  0. from equilibrium, in apparent contradiction to the
Rotations on  If  is the interval [0, 1] with its second law of thermodynamics. Such concerns,
endpoints identified and addition performed previously canvassed by Poincaré himself, were
modulo 1, then for every a  ℝ, the transfor- more infamously expounded by Zermelo (1896)
mation Ra:  ! , defined by Rax ¼ x + a, in 1896. Subsequent clarifications by Boltzmann,
preserves Lebesgue measure on  and hence Maxwell, and others led to an improved under-
induces a measure preserving system on . standing of the second law’s primarily statistical
Syndetic set Is a subset E  ℤ having bounded nature. (For an interesting historical/philosophical
gaps. If G is a general discrete group, a set discussion, see Sklar (2004) and also Bergelson
E  G is syndetic if G ¼ F E for some finite (2000). For a probabilistic analysis of the likeli-
set F  G. hood of observing second law violations in small
systems over short time intervals, see Evans and
Upper density Is the number dðLÞ ¼
Searles (2002).)
lim supN!1 jL\f2Nþ1
N, ..., N gj
where L  ℤ (ass-
, These discoveries had a profound impact in
uming the limit to exist). Alternatively for mea- dynamics, and the theory of measure-preserving
surable E  ℝm, d ðEÞ ¼ lim suplðSÞ!1 mmðS\E Þ
ðSÞ , transformations (ergodic theory) evolved from
where S ranges over all cubes in ℝ and l(S)
m
these developments. Since then, the Poincaré
denotes the length of the shortest edge of S. recurrence principle has been applied to a variety
Notation The following notation will be used of different fields in mathematics, physics, and
throughout the article: Tf ¼ f  T, {x} ¼ x  [x], information theory. In this article we survey the
D-limn!1(an) ¼ a , d({n: |an  a| > e} ¼ 0 impact it has had in ergodic theory, especially as
for every e > 0. pertains to the field of ergodic Ramsey theory.
(The heavy emphasis herein on the latter reflects
authorial interest and is not intended to transmit a
Definition of the Subject and Its proportionate image of the broader landscape of
Importance research relating to recurrence in ergodic theory.)
Background information we assume in this article
The basic principle that lies behind several recur- can be found in the books (Einsiedler and Ward
rence phenomena is that the typical trajectory of a 2011; Furstenberg 1981; Glasner 2003; Host and
system with finite volume comes back infinitely Kra 2018; Petersen 1989; Walters 1982) (see also
often to any neighborhood of its initial point. This the chapter “Measure Preserving Systems” by
Ergodic Theory: Recurrence 63

K. Petersen in this volume). Related information we conclude (as in the proof of Theorem 2.1) that
can also be found on the survey articles m(B) ¼ 0. This shows that for almost every x  A,
(Bergelson 1996, 2006a, b; Frantzikinakis 2016; we have that T nx  A for some n  ℕ. Repeating
Kra 2006b, 2007, 2011). this argument for the transformation T m in place
of T for all m  ℕ, we easily deduce the adver-
tised statement.
Introduction
Next we give a variation of Poincaré recur-
rence for measure-preserving systems endowed
In this section we shall give several formulations of
with a compatible metric.
the Poincaré recurrence principle using the lan-
guage of ergodic theory. Roughly speaking, the
Theorem 2.3 (Poincaré Recurrence for Metric
principle states that in a finite (or conservative)
Systems) Let (X, ℬ, m, T) be a measure-
measure-preserving system, every set of positive
preserving system, and suppose that X is endowed
measure (or almost every point) comes back to
with a m-compatible metric. Then for almost every
itself infinitely many times under iteration. Despite
x  X, we have
the profound importance of these results, their pro-
ofs are extremely simple.
lim inf dðx, T n xÞ ¼ 0:
n!1
Theorem 2.1 (Poincaré Recurrence for Sets)
Let (X, ℬ, m, T) be a measure-preserving system
and A  ℬ with m(A) > 0. Then m(A \ T nA) > 0 The proof of this result is similar to the proof of
for infinitely many n  ℕ. Theorem 2.2 (see Furstenberg 1981, p. 61).
Applying this result to the doubling map T
Proof Since T is measure preserving, the sets A, x ¼ 2x on , we get that for almost every x  X,
T 1A, T 2A, . . . have the same measure. These every string of zeros and ones in the dyadic expan-
sets cannot be pairwise essentially disjoint, since sion of x occurs infinitely often.
then the union of finitely many of them would We remark that all three formulations of the
have measure greater than m(X) ¼ 1. Therefore, Poincaré Recurrence Theorem that we have given
there exist m, n  ℕ, with n > m, such that hold for conservative systems as well. See, e.g.,
m(T mA \ T nA) > 0. Again since T is measure Aaronson (1997) for details.
preserving, we conclude that m(A \ T k A) > 0, This article is structured as follows. In section
where k ¼ n  m > 0. Repeating this argument for “Quantitative Poincaré Recurrence,” we give a
the iterates A, T mA, T 2mA, . . ., for all m  N, few quantitative versions of the previously men-
we easily deduce that m(A \ T nA) > 0 for infi- tioned qualitative results. In sections “Subse-
nitely many n  ℕ. quence Recurrence” and “Multiple Recurrence,”
We remark that the above argument actually we give several refinements of the Poincaré recur-
shows that m(A\T nA) > 0 for some n  mð1AÞ þ 1. rence theorem, by restricting the scope of the
return time n and by considering multiple inter-
sections (for simplicity we focus on ℤ-actions). In
Theorem 2.2 (Poincaré Recurrence for
section “Connections with Combinatorics and
Points) Let (X, ℬ, m, T) be a measure-preserving
Number Theory,” we give various implications
system and A  ℬ. Then for almost every x  A,
of the recurrence results in combinatorics and
we have that T nx  A for infinitely many n  ℕ.
number theory (see also the chapter ▶ “Ergodic
Theory: Interactions with Combinatorics and
Proof Let B be the set of x  A such that T nx 2 =
Number Theory” by T. Ward in the present vol-
A for all n  ℕ. Notice that B ¼ A\\n  ℕT nA; in
ume). Lastly, in section “Future Directions,” we
particular, B is measurable. Since the iterates B,
give several open problems related to the material
T 1B, T 2B, . . . are pairwise essentially disjoint,
64 Ergodic Theory: Recurrence

presented in sections “Subsequence Recurrence,” dm(x) ¼ m(A). Furthermore, for ergodic measure-
“Multiple Recurrence,” and “Connections with preserving systems, we have d(Sx) ¼ m(A) a.e.
Combinatorics and Number Theory.”
Another question that arises naturally is, given
a set A with positive measure and an x  A, how
long should one wait till some iterate T nx of x hits
Quantitative Poincaré Recurrence
A. By considering an irrational rotation Ra on ,
1
where a is very near to, but not less than, 100 , and
Early Results
letting A ¼ 1[0,1/2], one can see that the first return
For applications it is desirable to have quantitative
time is a member of the set {1, 50, 51}. So it may
versions of the results mentioned in the previous
come as a surprise that the average first return time
section. For example one would like to know how
does not depend on the system (as long as it is
large m(A \ T nA) can be made and for how
ergodic) but only on the measure of the set A.
many n.
Theorem 3.3 (Kac 1947) Let (X, ℬ, m, T) be an
Theorem 3.1 (Khintchine 1934) Let (X, ℬ, m, T)
ergodic measure-preserving system and A  ℬ
be a measure-preserving system and A  ℬ. Then
with m(A) > 0. For x  X define RA(x) ¼ min{n 
for every e > 0, we have m(A \ T nA) > m(A)2 e
ℕ: T nx  A}. Then for x  A the expected value
for a set of n  ℕ that hasbounded gaps .
of RA(x) is 1/m(A), i.e., A RA(x) dm ¼ 1.
By considering the doubling map T x ¼ 2x on
 and letting A ¼ 1[0,1/2), it is easy to check that More Recent Results
the lower bound of the previous result cannot be As we mentioned in the previous section, if the
improved. We also remark that it is not possible to space X is endowed with a m-compatible metric d,
estimate the size of the gap by a function of m(A) then for almost every x  X, we have that lim
alone. One can see this by considering the rota- infn!1 d(x, T n x) ¼ 0. A natural question is how
tions Rk x ¼ x + 1/k for k  ℕ, defined on , and much iteration is needed to come back within a
letting A ¼ 1[0,1/3]. small distance of a given typical point. Under
Concerning the second version of the Poincaré some additional hypothesis on the metric d, we
recurrence theorem, it is natural to ask whether for have the following answer.
almost every x  X the set of return times Sx ¼ {n
 ℕ: T nx  A} has bounded gaps. This is not the Theorem 3.4 (Boshernitzan 1993) Let (X, ℬ, m,
case, as one can see by considering the doubling T) be a measure-preserving system endowed with
map T x ¼ 2x on  with the Lebesgue measure and a m-compatible metric d. Assume that the
letting A ¼ 1[0,1/2). Since Lebesgue almost every Hausdorff a-measure ℋa(X) of X is s-finite (i.e.,
x   contains arbitrarily large blocks of ones in X is a countable union of sets Xi with
its dyadic expansion, the set Sx has unbounded ℋa(Xi) < 1). Then for almost every x  X
gaps. Nevertheless, as an easy consequence of the
Birkhoff ergodic theorem (Birkhoff 1931), one 1
lim inf na  dðx, T n xÞ < 1:
n!1
has the following.

Theorem 3.2 Let (X, ℬ, m, T) be a measure- Furthermore, if ℋa(X) ¼ 0, then for almost
preserving system and A  ℬ with m(A) > 0. every x  X
Then for almost every x  X, the set Sx ¼ {n 
1
ℕ: T nx  A} has well-defined density and d(Sx) lim inf na  dðx, T n xÞ ¼ 0:
n!1
Ergodic Theory: Recurrence 65

One can see from rotations by “badly where tr (x) is the first return time of T k x in
approximable” vectors a  k that the exponent B(x, r) and the upper and lower pointwise
1/k in the previous theorem cannot be improved. dimensions
Several applications of Theorem 3.4 to billiard
flows, dyadic transformations, symbolic flows, log mðBðx, r ÞÞ
d m ðxÞ ¼ lim inf and
and interval exchange transformations are given r!0 log r
in Boshernitzan (1993). For a related result deal- log mðBðx, r ÞÞ
ing with mean values of the limits in Theorem 3.4, dm ðxÞ ¼ lim sup :
r!0 log r
see Shkredov (2002).
An interesting connection between rates of Then for almost every x  X, we have
recurrence and entropy of an ergodic measure-
preserving system was established by Ornstein RðxÞ  dm ðxÞ and RðxÞ  dm ðxÞ:
and Weiss (1993), following earlier work of
Wyner and Ziv (1989).
Roughly speaking, this theorem asserts that for
Theorem 3.5 (Ornstein and Weiss 1993) Let typical x  X and for small r, the first return time of
(X, ℬ, m, T) be an ergodic measure-preserving x in B(x, r) is at most rdm(x). Since d m ðxÞ  ℋa ðXÞ
system and P be a finite partition of X. Let Pn(x)
i for almost every x  X, we can conclude the first
be the element of the partition Vn1 i¼0 T P ¼ part of Theorem 3.4 from Theorem 3.6. For related
n1 i ðiÞ ðiÞ
\i¼0 T P : P  P , 0  i < n that contains results the interested reader should consult the sur-
x. Then for almost every x  X, the first return vey (Barreira 2005) and the bibliography therein.
time Rn(x) of x to Pn(x) is asymptotically equiva- We also remark that the previous results and
lent to eh(T,P )n, where h(T, P ) denotes the entropy related concepts have been applied to estimate the
of the system with respect to the partition P. More dimension of certain strange attractors (see Hasley
precisely, and Jensen (2004) and the references therein) and
the entropy of certain Gibbsian systems
log Rn ðxÞ (Chazottes and Ugalde 2005).
lim ¼ hðT, P Þ:
n!1 n We end this section with a result that connects
“wandering rates” of sets in infinite measure-
preserving systems with their “recurrence rates.”
An extension of the above result to some clas-
The next theorem follows easily from a result
ses of infinite measure-preserving systems was
about lower bounds on ergodic averages for
given in Galatolo et al. (2006).
measure-preserving systems due to Leibman
Another connection of recurrence rates, this
(2002); a weaker form for conservative, ergodic
time with the local dimension of an invariant
systems can be found in Aaronson (1981).
measure, is given by the next result.
Theorem 3.7 Let (X, ℬ, m, T) be an infinite
Theorem 3.6 (Barreira 2001) Let (X, ℬ, m, T) be
measure-preserving system and A  ℬ with
an ergodic measure-preserving system. Define the
m(A) < 1. Then for all N  N,
upper and lower recurrence rates
n
log tr ðxÞ m [N1
n¼0 T A
N1

RðxÞ ¼ lim inf and  mðA \ T n AÞ

r!0  log r N n¼0
log tr ðxÞ 1
RðxÞ ¼ lim sup ,   ðmðAÞÞ2 :
r!0  log r 2
66 Ergodic Theory: Recurrence

Subsequence Recurrence considering product systems, one can immedi-

ately show that any set of (topological) recurrence
In this section we discuss what restrictions we can R is partition regular, meaning that if R is
impose on the set of return times in the various partitioned into finitely many pieces then at least
versions of the Poincaré recurrence theorem. We one of these pieces must still be a set of
start with: (topological) recurrence. Using this observation,
one concludes, for example, that any union of
Definition 4.1 Let R  ℤ. Then R is a set of: finitely many lacunary sequences is not a set of
recurrence.We present now some examples of sets
(a) Recurrence if for any invertible measure- of recurrence.
preserving system (X, ℬ, m, T) and A  ℬ
with m(A) > 0, there is some nonzero n  R Theorem 4.2 The following are sets of
such that m(A \ T  nA) > 0. recurrence:
(b) Topological recurrence if for every compact (i) Any set of the form [n  ℕ{an, 2an, . . .,
metric space (X, d), continuous transforma- nan} where an  ℕ. This follows from a
tion T: X ! X and every e > 0, there are finitary version of Szemerédi’s theorem.
x  X and nonzero n  R such that d(x, (ii) Any IP-set, meaning a set that consists of
T nx) < e. all finite sums of some infinite set
(Furstenberg and Katznelson 1985).
It is easy to check that the existence of a single (iii) Any difference set S  S, meaning a set that
n  R satisfying the previous recurrence condi- consists of all possible differences of some
tions actually guarantees the existence of infinitely infinite set S.
many n  R satisfying the same conditions. More- (iv) The set {p(n), n  ℕ} where p is any
over, if R is a set of recurrence then one can see nonconstant integer polynomial with
from existence of some T-invariant measure m that p(0) ¼ 0 (Furstenberg 1981; Sárközy
R is also a set of topological recurrence. 1978). In fact we only have to assume that
A (complicated) example showing that the con- the range of the polynomial contains mul-
verse is not true was given by Kriz (1987) (see tiples of an arbitrary positive integer; this
also Forrest (1991) and McCutcheon (1995) for follows from Theorem 4.3 below.
simplified versions). (v) The set {p(n), n  S}, where p is an integer
Before giving a list of examples of sets of polynomial with p(0) ¼ 0 and S is any
(topological) recurrence, we discuss some nec- IP-set (Bergelson et al. 1996).
essary conditions: A set of topological recur- (vi) The set of values of an admissible general-
rence must contain infinitely many multiples of ized polynomial. This class contains in par-
every positive integer, as one can see by consid- ticular the smallest function algebra
ering rotations on ℤd, d  ℕ. Hence, the sets G containing all integer polynomials hav-
{2n + 1, n  ℕ}, {n2 + 1, n  ℕ}, and {p + 2, p ing zero constant term and such that if g1,
prime} are not good for (topological) recur- . . . , gk  G and c1, . . . , ck  ℝ, then
k 1
i¼1 ci gi  G , where x ¼ x þ 2 denotes
rence. If (sn)n  ℕ is a lacunary sequence
(meaning lim infn!1(sn+1/sn) ¼ r > 1), then the integer nearest to x (Bergelson et al.
one can construct an irrational number a such 2006).
that {sna}  [d, 1  d] for all large n  ℕ, (vii) The set of shifted primes {p  1, p prime},
where d > 0 depends on r (see Katznelson the set {p + 1, p prime} (Sárközy 1978),
(2001) for example). As a consequence, the and also the integer part of certain smooth
sequence (sn)n  ℕ is not good for (topological) functions evaluated at the primes
recurrence. Lastly, we mention that by (Bergelson et al. 2019).
Ergodic Theory: Recurrence 67

(viii) R ¼ {[a(1)], [a(2)], . . .}, where a(x) ¼ xc T is measure preserving and f  L1(m); (an) is
for any c > 0. This follows from Theorem positive definite, and we call s ¼ sf the spectral
4.3 below and standard exponential sum measure of f.
estimates; see also Boshernitzan et al. Let now (X, ℬ, m, T) be a measure-preserving
(2005) for a more general result regarding system and A  ℬ with m(A) > 0. Putting f ¼ 1A,
Hardy sequences. one has
(ix) The set of values of a random non-lacunary
sequence. More precisely, pick n  ℕ inde- 1
N

pendently with probability bn where 0  bn lim f ðxÞ  f ðT an xÞdm

N!1 N
 1 is decreasing and limn!1 nbn ¼ 1, n¼1
N
then the resulting set is almost surely a set 1
¼ lim e2pian t ds f ðtÞ ð1Þ
of recurrence (this follows from Bourgain  N!1 N n¼1
(1988)).
(x) Arbitrary shifts of integers with an even where we assume for simplicity that all limits
(or odd) number of prime factors and exist. For t irrational the limit inside the integral
other similar sets with arithmetic struc- is zero (by condition (i)), so the last integral can be
ture (Bergelson et al. “A structure theo- taken over the rational points in . Since the
rem for. . .”; Frantzikinakis and Host spectral measure of a function orthogonal to the
2017b). subspace

Showing that the first three sets are good for

ℋ¼ f  L2 ðmÞ : there exists k  ℕ with T k f ¼ f
recurrence is a straightforward modification of the
argument used to prove Theorem 2.1. The other ð2Þ
examples require more work.
A criterion of Kamae and Mendés-France has no rational point masses, we can easily deduce
(1978) provides a powerful tool that may be that when computing the first limit in (1), we can
used in many instances to establish that a set R is replace the function f by its orthogonal projection
a set of recurrence. We mention a variation of their g onto the subspace ℋ (g is again nonnegative
result. and g 6¼ 0). To complete the argument, we approx-
imate g by a function g0 such that T mg0 ¼ g0 for
Theorem 4.3 (Kamae and Mendés-France some appropriately chosen m and use condition
1978) Suppose that R ¼ {a1 < a2 < . . .} is a (ii) to deduce that the limit of the average (1) is
subset of ℕ such that: positive.
In order to apply Theorem 4.3, one uses the
(i) The sequence {ana}n  ℕ is uniformly distrib- standard machinery of uniform distribution.
uted in  for every irrational a. Recall Weyl’s criterion: a real valued sequence
(ii) The set Rm ¼ {n  ℕ: m|an} has positive (xn)n  ℕ is uniformly distributed mod 1 if for
upper density for every m  ℕ. every nonzero k  ℤ

N
Then R is a set of recurrence. 1
lim e2pikxn ¼ 0:
N!1 N
n¼1
We sketch a proof for this result. First, recall
Herglotz’s theorem: if (an)n  ℤ is a positive defi- This criterion becomes especially useful when
nite sequence, then there is a unique measure s on paired with van der Corput’s so-called third prin-
the torus  such that an ¼ e2pint ds(t). The case cipal property: if, for every h  ℕ, (xn+h  xn)n  ℕ
of interest to us is an ¼  f(x) f(Tnx) dm, where is uniformly distributed mod 1, then (xn)n  ℕ is
68 Ergodic Theory: Recurrence

uniformly distributed mod 1. Using the foregoing Write f ¼ g + h where g  H and h ⊥ ℋ, and
criteria and some standard (albeit nontrivial) expand the average in (3) into a sum of four aver-
exponential sum estimates, one can verify, for ages involving the functions g and h. Two of these
example, that the sets (iv) and (vii) in Theorem averages vanish because iterates of g are orthogo-
4.2 are good for recurrence. nal to iterates of h. So in order to show that the only
In light of the connection elucidated above contribution comes from the average that involves
between uniform distribution mod 1 and recur- the function g alone, it suffices to establish that
rence, it is not surprising that van der Corput’s
method has been adapted by modern ergodic the- 1
N
2
orists for use in establishing recurrence properties lim Tn h ¼ 0: ð5Þ
N!1 N
directly. n¼1 L2 ð mÞ

To show this we will apply the Hilbert space

Theorem 4.4 (Bergelson 1987a) Let (xn)n  ℕ be van der Corput lemma. For given h  ℕ, we let
a bounded sequence in a Hilbert space. If xn ¼ Tn h and compute

1 1
N
1
N
D  m!1
lim lim hxnþm , xn i ¼ 0, lim hxnþm , xn i ¼ lim
2 2
T n þ2nmþm h  T n hdm
2

N!1 N N!1 N N!1 N

n¼1 n¼1
N
1 2
then ¼ lim T 2nm T m h  hdm:
N!1 N
n¼1

N
1
lim xn ¼ 0: Applying the von Neumann ergodic theorem
N!1 N n¼1 (von Newmman 1932) to the transformation T2m
and using the fact that h⊥ℋ, we get that the last
limit is 0. This implies (5).
Let us illustrate how one uses this “van der Thus far we have shown that in order to compute
Corput trick” by showing that S ¼ {n2 : n  ℕ} the limit in (3), we can assume that f ¼ g  ℋ (g is
is a set of recurrence. We will actually establish also nonnegative and g 6¼ 0). By the definition of
the following stronger fact: If (X, ℬ, m, T) is a ℋ, given any e > 0, there exists a function f 0  ℋ
measure-preserving system and f  L1(m) is such that T k f 0 ¼ f 0 for some k  ℕ and
nonnegative and f 6¼ 0, then k f  f 0 kL2 ðmÞ  e. Then the limit in (3) is at least
1/k times the limit
N
1 2
lim inf f ðxÞ  f T n x dm > 0: ð3Þ
n!1 N 1
N
2
f ðxÞ  f T ðknÞ x dm:
n¼1
lim inf
n!1 N n¼1
Then our result follows by setting f ¼ 1A for
some A  ℬ with m(A) > 0.The main idea is one Applying the triangle inequality twice, we get
that occurs frequently in ergodic theory; split the that this is greater or equal than
function f into two components, one of which
contributes zero to the limit appearing in (3) and 1
N
2 2
the other one being much easier to handle than f. lim f 0 ðxÞ  f 0 T ðknÞ x dm  c  e ¼ ð f 0 ðxÞÞ dm  2e
N!1 N
n¼1
To do this consider the T-invariant subspace of 2
L2(X) defined by  f 0 ðxÞdm  c  e,

ℋ¼ f  L2 ðmÞ : there exists k  ℕ with T k f ¼ f : for some constant c that does not depend on e
ð4Þ (we used that Tk f 0 ¼ f 0 and the Cauchy-Schwartz
inequality). Choosing e small enough, we
Ergodic Theory: Recurrence 69

conclude that the last quantity is positive, com- components. Let K be the closure in L2 of the
pleting the proof. subspace spanned by the eigenfunctions of T, i.e.,
the functions f  L2(m) that satisfy f (T x) ¼ e2piaf
(x) for some a  ℝ. We write f ¼ g + h, where g 
Multiple Recurrence K and h ⊥ K . It can be shown that g, h  L1(m)
and g is again nonnegative with g 6¼ 0. We expand
Simultaneous multiple returns of positive measure the average in (7) into a sum of eight averages
sets to themselves were first considered by involving the functions g and h. In order to show
H. Furstenberg (1977), who gave a new proof of that the only nonzero contribution to the limit
Szemerédi’s theorem (Szemerédi 1975) on arith- comes from the term involving g alone, it suffices
metic progressions by deriving it from the follow- to establish that
ing theorem.
N
1
lim T n g  T 2n h ¼ 0, ð8Þ
Theorem 5.1 (Furstenberg 1977) Let (X, ℬ, N!1 N n¼1 L2 ðmÞ
m, T) be a measure-preserving system and A 
ℬ with m(A) > 0. Then for every k  ℕ, there is (and similarly with h and g interchanged, and
some n  ℕ such that with g ¼ h, which is similar). To establish (8),
we use the Hilbert space van der Corput lemma
m A \ T n A \    \ T kn A > 0: ð6Þ on xn ¼ T ng  T 2nh. Some routine computations
and a use of the ergodic theorem reduce the task
Furstenberg’s proof came by means of a new to showing that
structure theorem allowing one to decompose an
arbitrary measure-preserving system into compo-
nent elements exhibiting one of two extreme types D  lim hðxÞ  h T 2m x dm ¼ 0:
m!1
of behavior: compactness, characterized by regu-
lar, “almost periodic” trajectories, and weak But this is well known for h ⊥ K (e.g., in virtue
mixing, characterized by irregular, “quasi- of the fact that for h ⊥ K the spectral measure sh is
random” trajectories. On , these types of behav- continuous).We are left with the average (7) when
ior are exemplified by rotations and by the dou- f ¼ g  K . In this case f can be approximated
bling map, respectively. To see the point, imagine arbitrarily well by a linear combination of eigen-
trying to predict the initial digit of the dyadic functions, which easily implies that given e > 0,
expansion of T nx given knowledge of the initial one has kT n f  f kL2 ðmÞ  e for a set of n  ℕ with
digits of T ix, 1  i < n. We use the case k ¼ 2 to bounded gaps. Using this fact and the triangle
illustrate the basic idea. inequality, one finds that for a set of n  ℕ with
It suffices to show that if f  L1(m) is nonneg- bounded gaps,
ative and f 6¼ 0, one has
3

1
N f ðxÞ  f ðT n xÞ  f T 2n x dm  fdm ce
lim inf f ðxÞ  f ðT n xÞ  f T 2n x dm
N!1 N n¼1

> 0: ð7Þ for a constant c that is independent of e. Choosing

e small enough, we get (7).
An ergodic decomposition argument enables
us to assume that our system is ergodic. As in The new techniques developed for the proof of
the earlier case of the squares, we split f into Theorem 5.1 have led to a number of extensions,
“almost periodic” and “quasi-random” many of which have to date only ergodic proofs.
70 Ergodic Theory: Recurrence

To expedite discussion of some of these develop- “Single and multiple recurrence. . .”;
ments, we introduce a definition: Frantzikinakis 2009, 2010; Frantzikinakis and
Wierdl 2009), integer part polynomial sequences
Definition 5.2 Let R  ℤ and k  ℕ. Then R is a (Karageorgos and Koutsogiannis “Integer part
set of k-recurrence if for every invertible measure- independent. . .”; Koutsogiannis 2018a, b), random
preserving system (X, B, m, T ) and A  B with sequences (Frantzikinakis et al. 2012, 2016), and
m(A) > 0 there is some nonzero n  R such that sets of arithmetic nature (Bergelson et al. “A struc-
ture theorem for. . .”; Frantzikinakis and Host
m A \ T n A \    \ T kn A > 0: 2017a, b).
More generally, one would like to know for
The notions of k-recurrence are distinct for which sequences of integers a1(n), . . . , ak (n) it is
different values of k. An example of a difference the case that for every invertible measure-
set that is a set of 1-recurrence but not a set of preserving system (X, B, m, T) and A  B with
2-recurrence was given in (Furstenberg 1977); m(A) > 0, there is some nonzero n  ℕ such that
sets of k-recurrence that are not sets of (k + 1)-
recurrence for general k were given in m A \ T a1 ðnÞ A \    \ T ak ðnÞ A > 0: ð9Þ
Frantzikinakis et al. (2006) ðRk ¼
p
nℕ : n kþ1
2  ½1=4, 3=4 is such). Unfortunately, a criterion analogous to the one
Aside from difference sets, the sets of (1-) given in Theorem 4.3 for 1-recurrence is not yet
recurrence given in Theorem 4.2 may well be sets available for k-recurrence when k > 1. Neverthe-
of k-recurrence for every k  ℕ, though this has less, there have been some notable positive
not been verified in all cases. Let us summarize the results, such as the following:
current state of knowledge. The following are sets
of k-recurrence for every k: sets of the form
[n  ℕ{an, 2an, . . ., nan}where an  ℕ (this fol- Theorem 5.3 (Bergelson and Leibman 1996)
lows from a uniform version of Theorem 5.1 that Let (X, B, m, T) be an invertible measure-
can be found in Bergelson et al. (2000)); every preserving system and p1(n), . . . pk (n) be integer
IP-set (Furstenberg and Katznelson 1985); the set polynomials with zero constant term. Then for
{p(n), n  ℕ} where p is any nonconstant integer every A  B with m(A) > 0, there is some
polynomial with p(0) ¼ 0 (Bergelson and Leibman n  ℕ such that
1996) and, more generally, when the range of the
polynomial contains multiples of an arbitrary inte- m A \ T p1 ðnÞ A \    \ T pk ðnÞ A > 0: ð10Þ
ger (Frantzikinakis 2008); the set {p(n), n  S}
where p is an integer polynomial with p(0) ¼ 0 and
S is any IP-set (Bergelson and McCutcheon 2000); Furthermore, it has been shown that the n in
and the set of values of an admissible generalized (10) can be chosen from any IP set (Bergelson and
polynomial (Bergelson and McCutcheon 2010; McCutcheon 2000) and the polynomials p1, . . . ,
McCutcheon 2005). Moreover, it was shown in pk can be chosen to belong to the more general
Frantzikinakis et al. (2007) for k ¼ 2 and in Wooley class of admissible generalized polynomials
and Ziegler (2012) for general k  ℕ that the sets (McCutcheon 2005) or the class of intersective
of shifted primes {p  1, p prime} (or the set polynomials (Bergelson et al. 2008).
{p + 1, p prime}) are sets of k-recurrence (see An important boost in the area of multiple
also Bergelson et al. (2011), Frantzikinakis et al. recurrence was given by a breakthrough of Host
(2013), Koutsogiannis (2018a), and Sun (2015) for and Kra (2005). Building on work of Conze and
related work). Several other multiple recurrence Lesigne (1984, 1988) and Furstenberg and Weiss
results were obtained in the last 10 years, including (Furstenberg and Weiss 1996) (see also the excel-
results for Hardy sequences (Bergelson et al. lent survey of Kra (2006a), exploring close
Ergodic Theory: Recurrence 71

parallels with Green and Tao (2008), and the When the polynomials n, 2n, . . . , kn are replaced
seminal paper of Gowers (2001)), they isolated by linearly independent polynomials p1, p2, . . . ,
the structured component (or factor) of a measure- pk with zero constant term, similar lower bounds
preserving system that one needs to analyze in hold for every k  ℕ without assuming ergodicity
order to prove several multiple recurrence and (Frantzikinakis and Kra 2006). The case where the
convergence results. This allowed them, in partic- polynomials n, 2n, 3n are replaced with general
ular, to prove existence of L2 limits for the polynomials p1, p2, p3 with zero constant term is
so-called “Furstenberg ergodic averages” treated in Frantzikinakis (2008) (see also Donoso
1 N
et al. “Optimal lower bounds for multiple. . .”),
n¼1 Pi¼0 f ðT xÞ, which had been a major
k in
N
open problem since the original ergodic proof of and more general results involving Hardy field
Szemerédi’s theorem. Subsequently Ziegler in sequences and polynomials evaluated at the
Ziegler (2007) gave a new proof of the aforemen- primes are obtained in Donoso et al. “Optimal
tioned limit theorem and established minimality lower bounds for multiple. . ..”
of the factor in question. It turns out that this
minimal component admits of a purely algebraic
characterization; it is a nilsystem, i.e., a rotation on Connections with Combinatorics and
a homogeneous space of a nilpotent Lie group. Number Theory
This fact, coupled with some recent results about
nilsystems (see Leibman (2005a, b) for example), The combinatorial ramifications of ergodic-theoretic
makes the analysis of some otherwise intractable recurrence were first observed by Furstenberg, who
multiple recurrence problems much more man- perceived a correspondence between recurrence
ageable. These developments have made it possi- properties of measure-preserving systems and the
ble to obtain new multiple recurrence results, and existence of structures in sets of integers having
they also allowed us to estimate the size of the positive upper density. This gave rise to the field of
multiple intersection in (6) for k ¼ 2, 3 (the case ergodic Ramsey theory, in which problems in com-
k ¼ 1 is Theorem 3.1). binatorial number theory are treated using tech-
niques from ergodic theory. The following
Theorem 5.4 (Bergelson et al. 2005) Let (X, B, formulation is from Bergelson (1987b).
m, T) be an ergodic measure-preserving system
and A  B. Then for k ¼ 2, 3 and for every e > 0, Theorem 6.1 Let L be a subset of the integers.
There exists an invertible measure-preserving sys-
m A \ T n A \    \ T kn A tem (X, B, m, T) and a set A  B with m(A) ¼ d(L)
> mkþ1 ðAÞ  e ð11Þ such that

for a set of n  ℕ with bounded gaps. d ð L \ ð L  n1 Þ \ . . . \ ð L  nk Þ Þ

 mðA \ T n1 A \    \ T nk AÞ, ð12Þ
Based on work of Ruzsa that appears as an
appendix to the paper, it is also shown in for all k  ℕ and n1, . . . , nk  ℤ.
Bergelson et al. (2005) that a similar estimate
fails for ergodic systems (with any power of Proof The space X will be taken to be the
m(A) on the right hand side) when k  4. Moreover, sequence space {0, 1}ℤ, B is the Borel s-algebra,
when the system is nonergodic, it also fails for while Tis the shift map defined by (T x)(n) ¼ x(n + 1)
k ¼ 2, 3, as can be seen with the help of an for x  {0, 1}ℤ, and A is the set of sequences x with
example in Behrend (1946). Again considering the x(0) ¼ 1. So the only thing that depends on L is the
doubling map T x ¼ 2x and the set A ¼ [0, 1/2], one measure m which we now define. For m  ℕ set
sees that the positive results for k  3 are sharp. L0 ¼ ℤ\L and L1 ¼ L. Using a diagonal argument,
72 Ergodic Theory: Recurrence

we can find an increasing sequence of integers Theorem 6.2 (Bergelson and Leibman 1996)
(Nm)m  ℕ such that limm!1 |L \ [1, Nm]|/Nm ¼ d Let L  ℤ with d(L) > 0 and p1, . . . , pk be integer
polynomials with zero constant term. Then L con-
(L) and such that
tains infinitely many configurations of the form {x,
x + p1(n), . . . , x + pk (n)}.
j Li1  n1 \ Li2  n2 \   \ Lir  nr \ ½1,N m  j
lim
m!1 Nm The ergodic proof is the only one known for this
ð13Þ result, although very recently some special cases
were covered in Peluse and Prendiville (“Quantita-
exists for every n1, . . . , nr  ℤ, and i1, . . . , ir  tive bounds in the. . .”) and Prendiville (2017) using
{0, 1}. For n1, n2, . . . , nr  ℤ, and i1, i2, . . . , ir  more elementary (but complicated) arguments.
{0, 1}, we define the measure m of the cylinder Ergodic-theoretic contributions to the field of
set fxn1 ¼ i1 , xn2 ¼ i2 , . . . , xnr ¼ ir g to be the geometric Ramsey theory were made by Furstenberg
limit (13). Thus defined, m extends to a pre- et al. (1990), who showed that if E is a positive upper
measure on the algebra of sets generated by cyl- density subset of ℝ2, then (i) E contains points with
inder sets and hence by Carathéodory’s extension any large enough distance (see also Bourgain (1986)
theorem (Carathéodory 1968) to a probability and Falconer and Marstrand (1986)) and (ii) every
measure on B. It is easy to check that m(A) ¼ d d-neighborhood of E contains three points forming a
(L), the shift transformation T preserves the mea- triangle congruent to any given large enough dilation
sure m and (12) holds. of a given triangle (in Bourgain (1986), it is shown
that if the three points lie on a straight line, one
Using this principle for k ¼ 1, one may check cannot always find three points with this property
that any set of recurrence is intersective, that is in E itself). Moreover, a generalization of property
intersects E  E for every set E of positive (ii) to arbitrary finite configurations of ℝm was
density. Using it for n1 ¼ n, n2 ¼ 2n, . . . , nk ¼ kn, obtained by Ziegler (2006).
together with Theorem 5.1, one gets an ergodic We also mention some exciting connections of
proof of Szemerédi’s theorem (Szemerédi 1975), multiple recurrence with some structural proper-
stating that every subset of the integers with ties of the set of prime numbers. The first one is in
positive upper density contains arbitrarily long the work of Green and Tao (2008), where the
arithmetic progressions (conversely, one can eas- existence of arbitrarily long arithmetic progres-
ily deduce Theorem 5.1 from Szemerédi’s theo- sions of primes was demonstrated; the authors,
rem and that intersective sets are sets of in addition to using Szemerédi’s theorem outright,
recurrence). Making the choice n1 ¼ n2 and use several ideas from its ergodic-theoretic proofs,
using part (iv) of Theorem 4.3, we get an ergodic as appearing in Furstenberg (1977) and
proof of the surprising result of Sárközy (1978) Furstenberg et al. (1982). The second one is in
stating that every subset of the integers with the recent work of Tao and Ziegler (2008); a
positive upper density contains two elements quantitative version of Theorem 5.3 was used to
whose difference is a perfect square. More gen- prove that the primes contain arbitrarily long
erally, using Theorem 6.1, one can translate all of polynomial progressions. Furthermore, results in
the recurrence results of the previous two sec- ergodic theory related to the structure of the min-
tions to results in combinatorics. (This is not imal characteristic factors of certain multiple
straightforward for Theorem 5.4 because of the ergodic averages play an important role in the
ergodicity assumption made there. We refer the work of Green, Tao, and Ziegler (Green and Tao
reader to Bergelson et al. (2005) for the combi- 2010, 2012a, b, 2014; Green et al. 2012), where
natorial consequence of this result.) We mention they get asymptotic formulas for the number of
explicitly only the combinatorial consequence of k-term arithmetic progressions of primes up to x.
Theorem 5.3. This work verifies an interesting special case of
Ergodic Theory: Recurrence 73

the Hardy-Littlewood k-tuple conjecture pre- being occupied by a variable letter x, for x ¼ a1,
dicting the asymptotic growth rate of Na1, . . . , ak . . . , ak. (For example, in W4(A), the sets {(a1, x,
(x) ¼ the number of configurations of primes a2, x) : x  A} and {(x, x, x, x), : x  A} are
having the form {p, p + a1, . . . , p + ak } with combinatorial lines.)
p  x.
In a more recent development, the tools devel- At first glance, the uninitiated reader may not
oped in the last two decades to deal with delicate appreciate the importance of this “master” density
multiple recurrence problems have played an result, so it is instructive to derive at least one of
instrumental role in analyzing the structure of its immediate consequences. Let A ¼ {0, 1, . . . ,
measure-preserving systems naturally associated k  1} and interpret Wn(A) as integers in base
with bounded multiplicative functions. These k having at most n digits. Then a combinatorial
results were used in the last 2 years in works of line in Wn(A) is an arithmetic progression of
Tao and Teräväinen (The structure of logarith- length k – for example, the line {(a1, x, a2, x) :
mically. . .; The structure of correlations. . .) to x  A} corresponds to the progression {m, m + n,
make progress on the Chowla and Elliott con- m + 2n, m + 3n}, where m ¼ a1 + a2d2 and n ¼ d + d3.
jectures and in works of Frantzikinakis and Host This allows one to deduce Szemerédi’s theorem.
(2018, Furstenberg systems of bounded. . .) to Similarly, one can deduce from Theorem 6.3 multi-
make progress on the Möbius disjointness con- dimensional and IP extensions of Szemerédi’s the-
jecture of Sarnak. It appears that this interplay orem (Furstenberg and Katznelson 1979, 1985)
of ergodic theory and number theory will be and some related results about vector spaces over
useful for the resolution of several notoriously dif- finite fields (Furstenberg and Katznelson 1985).
ficult problems concerning higher-order correlations
of multiplicative and other number theoretic
functions.
Future Directions
Finally, we remark that in this article we have
restricted attention to multiple recurrence and
In this section we formulate a few open problems
Furstenberg correspondence for ℤ-actions, while
relating to the material in the previous three sec-
in fact there is a wealth of literature on extensions
tions. It should be noted that this selection reflects
of these results to general commutative, amena-
the authors’ interests and does not strive for com-
ble, and even non-amenable groups. For an excel-
pleteness. A more extensive list of problems
lent exposition of these and other recent
related to ergodic theory of ℤ-actions can be
developments, the reader is referred to the survey
found in Frantzikinakis (2016).
articles (Austin “Multiple recurrence and
We start with an intriguing question of
finding. . .”; Bergelson 1996, 2006a, b). Here, we
Katznelson (2001) about sets of topological recur-
give just one notable combinatorial corollary to
rence. A set S  ℕ is a set of Bohr recurrence if for
some work of this kind, a density version of the
every a1, . . . , ak  ℝ and e > 0 there exists s 
classical Hales-Jewett coloring theorem (Hales
S such that {sai}  [0, e][[1e, 1) for i ¼ 1, . . . , k.
and Jewett 1963).
Problem 1 Is every set of Bohr recurrence a set
Theorem 6.3 (Furstenberg and Katznelson
of topological recurrence?
(1991); see also Polymath (2012)) Let Wn(A)
denote the set of words of length n with letters in
Background for this problem and evidence for
the alphabet A ¼ {a1, . . . , ak }. For every e > 0,
a positive answer can be found in Katznelson
there exists N0 ¼ N0(e, k) such that if n  N0, then
(2001) and Weiss (2000). A negative answer for
any subset S of Wn(A) with |S|  ekn contains a
a related question concerning measure theoretic
combinatorial line, i.e., a set consisting of k
recurrence is given in Kriz (1987) (see also
n-letter words, having fixed letters in l positions,
Griesmer “Bohr topology and difference sets. . .”).
for some 0  l < n, the remaining n  l positions
74 Ergodic Theory: Recurrence

Problem 2 Is the set S ¼ {l!2m3n: l, m, n  ℕ} a Problem 5 If S  ℤ is a set of 2-recurrence, is it

set of recurrence? Is it a set of k-recurrence for true that S2 ¼ {s2: s  S} is a set of recurrence?
every k  ℕ?
A similar question was asked in Brown et al.
It can be shown that S is a set of Bohr recur- (1999): “If S is a set of k-recurrence for every k, is
rence. Theorem 4.3 cannot be applied since the the same true of S2?”.
uniform distribution condition fails for some irra- One would like to find a criterion that would
tional numbers a. A relevant question was asked allow one to deduce that a sequence is good for
by Bergelson in Bergelson (1996): “Is the set double (or higher-order) recurrence from some
S ¼ {2m3n : m, n  ℕ } good for single recurrence uniform distribution properties of this sequence.
for weakly mixing systems?”
We mentioned in section “Subsequence Recur- Problem 6 Find necessary conditions for double
rence” that the sets of shifted primes and the set of recurrence similar to the one given in Theorem 4.3.
fractional powers of integers are known to be sets
of k-recurrence for every k  ℕ. The following It is now well understood that such a criterion
related question remains open. should involve uniform distribution properties of
some generalized polynomials or two-step
Problem 3 Show that for every c  ℝ+ \ ℤ, the nilsequences.
set {[pc], p  P}, where P is the set of primes, is a We mentioned in section “Connections with
set of k-recurrence for every k  ℕ. Combinatorics and Number Theory” that every
positive upper density subset of ℝ2 contains
The problem is open even for k ¼ 2 and c ¼ 32. points with any large enough distance. Bourgain
We mentioned in section “Subsequence Recur- (1986) constructed a positive upper density subset
rence” that random non-lacunary sequences (see E of ℝ2, a triangle T , and numbers tn ! 1, such
definition there) are almost surely sets of that E does not contain congruent copies of all tn-
recurrence. dilations of T. But the triangle T used in this
construction is degenerate, which leaves the fol-
Problem 4 Show that random non-lacunary lowing question open.
sequences are almost surely sets of k-recurrence
for every k  ℕ. Problem 7 Is it true that every positive upper
density subset of ℝ2 contains a triangle congruent
Some progress is made in Bhattacharya et al. to any large enough dilation of a given non-
(“Upper tails large deviations. . .”), Briët et al. degenerate triangle?
(2017), Briët and Gopi (“Gaussian width
bounds. . .”), Christ (“On random multilinear For further discussion on this question, the
operator. . .”), and Frantzikinakis et al. (2012, reader can consult the survey of Graham (1994).
2016), but the problem is open even for k ¼ 2 The following question of Aaronson and
for random sequences with at most quadratic Nakada (Aaronson 1981) is related to a classical
growth. We refer the reader to the survey of question of Erdős concerning whether every
Rosenblatt and Wierdl (1995) for a nice exposi- K  ℕ such that n  K1/n ¼ 1 contains arbi-
tion of the argument used by Bourgain (Bourgain trarily long arithmetic progressions.
1988) to handle the case k ¼ 1.
It was shown in Frantzikinakis et al. (2006) that Problem 8 Suppose that (X, ℬ, m, T) is a {1/n}-
if S is a set of 2-recurrence, then the set of its conservative ergodic measure-preserving system.
squares is a set of recurrence for circle rotations. Is it true that for every A  B with m(A) > 0 and k
The same method shows that it is actually a set of  ℕ we have m(A \ T nA \ . . . \ T knA) > 0 for
Bohr recurrence. some n  ℕ?
Ergodic Theory: Recurrence 75

The answer is positive for the class of Markov References

shifts, and it is remarked in Aaronson (1981) that
if the Erdős conjecture is true, then the answer will Aaronson J (1981) The asymptotic distribution behavior of
transformations preserving infinite measures. J Anal
be positive in general. The converse is not known
Math 39:203–234
to be true. For a related result showing that mul- Aaronson J (1997) An introduction to infinite ergodic
tiple recurrence is preserved by extensions of infi- theory. Mathematical surveys and monographs,
nite measure-preserving systems, see vol 50. American Mathematical Society, Providence
Austin T. Multiple recurrence and finding patterns in dense
Meyerovitch “On multiple and polynomial. . ..”
sets. In: Badziahin D, Gorodnik A, Peyerimhoff N (eds)
Our next problem is motivated by the question Dynamics and analytic number theory. London mathe-
whether Theorem 6.3 has a polynomial version matical society lecture note series. Cambridge Univer-
(for a precise formulation of the general conjec- sity Press, Cambridge, pp 189–257
Barreira L (2001) Hausdorff dimension of measures via
ture, see Bergelson (1996)). Not even this most
Poincaré recurrence. Commun Math Phys
special consequence of it is known to hold. 219:443–463
Barreira L (2005) Poincaré recurrence: old and new. In:
Problem 9 Let e > 0. Does there exist N ¼ N (e) XIVth international congress on mathematical physics.
World Scientific Publishing, Hackensack, pp 415–422
having the property that every family P of subsets
2
Behrend F (1946) On sets of integers which contain no
of {1, . . ., N}2 satisfying j P j e2N contains a three in arithmetic progression. Proc Natl Acad Sci
configuration {A, A [ (g g)}, where A  {1, . . ., U S A 23:331–332
Bergelson V (1987a) Weakly mixing PET. Ergodic Theory
N}2 and g  {1, . . ., N} with A \ (g g) ¼ ∅?
Dyn Syst 7:337–349
Bergelson V (1987b) Ergodic Ramsey theory (ed:
A measure-preserving action of a general Simpson S, Logic and combinatorics). Contemp Math
countably infinite group G is a function g ! Tg 65:63–87
Bergelson V (1996) Ergodic Ramsey theory – an update,
from G into the space of measure-preserving
Ergodic Theory of ℤd-actions (ed: Pollicott M, Schmidt
transformations of a probability space X such K). London Math Soc Lecture Note Ser 228:1–61
that Tgh ¼ Tg Th. It is easy to show that a version Bergelson V (2000) The multifarious Poincare recurrence
of Khintchine’s recurrence theorem holds for such theorem. In: Foreman M, Kechris A, Louveau A, Weiss
B (eds) Descriptive set theory and dynamical systems.
actions: if m(A) > 0 and e > 0, then {g : m(A \ Tg
London mathematical society lecture note series, vol 277.
A) > (m(A)2  e} is syndetic. However it is Cambridge University Press, Cambridge, pp 31–57
unknown whether the following ergodic version Bergelson V (2006a) Combinatorial and Diophantine appli-
of Roth’s theorem holds. cations of Ergodic theory. Appendix A by A. Leibman
and Appendix B by A. Quas and M. Wierdl. In:
Hasselblatt B, Katok A (eds) Handbook of dynamical
Problem 10 Let (Tg) and (Sg) be measure- systems, vol. 1B. Elsevier, Amsterdam, pp 745–841
preserving G-actions of a probability space Bergelson V (2006b) Ergodic Ramsey theory: a dynamical
X that commute in the sense that Tg Sh ¼ ShTg approach to static theorems. In: Proceedings of the
international congress of mathematicians,
for all g, h  G. Is it true that for all positive
vol II. Madrid, pp 1655–1678
measure sets A, the set of g such that m(A \ Tg A \ Bergelson V, Leibman A (1996) Polynomial extensions of
Sg A) > 0 is syndetic? van der Waerden’s and Szemerédi’s theorems. J Am
Math Soc 9:725–753
Bergelson V, McCutcheon R (2000) An ergodic IP poly-
We remark that for general (possibly amenable)
nomial Szemerédi theorem. Mem Am Math Soc 146:
groups G not containing arbitrarily large finite sub- viii+106 pp
groups nor elements of infinite order, it is not Bergelson V, McCutcheon R (2007) Central sets and a
known whether one can find a single such g 6¼ e. noncommutative Roth theorem. Am J Math
129:1251–1275
On the other hand, the answer is known to be
Bergelson V, McCutcheon R (2010) Idempotent ultra-
positive for general G in case T 1 g Sg is a filters, multiple weak mixing and Szemerédi’s theorem
for generalized polynomials. J Anal Math 111:77–130
G-action (Bergelson and McCutcheon 2007); Bergelson V, Furstenberg H, McCutcheon R (1996) IP-sets
even under such strictures, however, it is unknown and polynomial recurrence. Ergodic Theory Dyn Syst
whether a triple recurrence theorem holds. 16:963–974
76 Ergodic Theory: Recurrence

Bergelson V, Host B, McCutcheon R, Parreau F (2000) Rech Math Rennes, 1987-1. University of Rennes I,
Aspects of uniformity in recurrence. Colloq Math Rennes, pp 1–31
84/85(Part 2):549–576 Donoso S, Le AN, Moreira J, Sun W. Optimal lower
Bergelson V, Host B, Kra B, with an appendix by Ruzsa bounds for multiple recurrence. To appear in Ergodic
I (2005) Multiple recurrence and nilsequences. Invent Theory Dyn Syst. arXiv:1809.06912
Math 160(2):261–303 Einsiedler M, Ward T (2011) Ergodic theory with a view
Bergelson V, Håland-Knutson I, McCutcheon R (2006) IP towards number theory. Graduate texts in mathematics,
systems, generalized polynomials and recurrence. vol 259. Springer London, London
Ergodic Theory Dyn Syst 26:999–1019 Evans D, Searles D (2002) The fluctuation theorem. Adv
Bergelson V, Leibman A, Lesigne E (2008) Intersective Phys 51:1529–1585
polynomials and the polynomial Szemerédi theorem. Falconer K, Marstrand J (1986) Plane sets with positive
Adv Math 219(1):369–388 density at infinity contain all large distances. Bull Lond
Bergelson V, Leibman A, Ziegler T (2011) The shifted Math Soc 18:471–474
primes and the multidimensional Szemerédi and poly- Forrest A (1991) The construction of a set of recurrence
nomial van der Waerden Theorems. Comptes Rendus which is not a set of strong recurrence. Israel J Math
Mathematique 349(3–4):123–125 76:215–228
Bergelson V, Kolesnik G, Son Y (2019) Uniform distribu- Frantzikinakis N (2008) Multiple ergodic averages for
tion of subpolynomial functions along primes and three polynomials and applications. Trans Am Math
applications. J Anal Math 137(1):135–187 Soc 360(10):5435–5475
Bergelson V, Kułaga-Przymus J, Lemańczyk M. A struc- Frantzikinakis N (2009) Equidistribution of sparse
ture theorem for level sets of multiplicative functions sequences on nilmanifolds. J Anal Math 109:353–395
and applications. To appear in Int Math Res Not. Frantzikinakis N (2010) Multiple recurrence and conver-
arXiv:1708.02613 gence for Hardy sequences of polynomial growth.
Bergelson V, Moreira J, Richter FK. Single and multiple J Anal Math 112:79–135
recurrence along non-polynomial sequences. Preprint, Frantzikinakis N (2016) Some open problems on multiple
arXiv:1711.05729 ergodic averages. Bull Hell Math Soc 60:41–90
Bhattacharya B, Ganguly S, Shao X, Zhao Y. Upper tails Frantzikinakis N, Host B (2017a) Higher order Fourier
large deviations for arithmetic progressions in a random analysis of multiplicative functions and applications.
set. To appear in Int Math Res Note. arXiv:1605.02994 J Am Math Soc 30:67–157
Birkhoff G (1931) A proof of the ergodic theorem. Proc Frantzikinakis N, Host B (2017b) Multiple ergodic the-
Natl Acad Sci U S A 17:656–660 orems for arithmetic sets. Trans Am Math Soc
Boshernitzan M (1993) Quantitative recurrence results. 369(10):7085–7105
Invent Math 113:617–631 Frantzikinakis N, Host B (2018) The logarithmic Sarnak
Boshernitzan M, Kolesnik G, Quas A, Wierdl M (2005) conjecture for ergodic weights. Ann Math 187:869–931
Ergodic averaging sequences. J Anal Math 95:63–103 Frantzikinakis N, Host B. Furstenberg systems of bounded
Bourgain J (1986) A Szemerédi type theorem for sets of multiplicative functions and applications. To appear in
positive density in ℝk. Israel J Math 54(3):307–316 Int Math Res Note IMRN. arXiv:1804.08556
Bourgain J (1988) On the maximal ergodic theorem for Frantzikinakis N, Kra B (2006) Ergodic averages for inde-
certain subsets of the positive integers. Israel J Math pendent polynomials and applications. J London Math
61:39–72 Soc 74(1):131–142
Briët J, Gopi S. Gaussian width bounds with applications Frantzikinakis N, Wierdl M (2009) A Hardy field extension
to arithmetic progressions in random settings. To of Szemerédi’s theorem. Adv Math 222:1–43
appear in Int Math Res Note. arXiv:1711.05624 Frantzikinakis N, Lesigne E, Wierdl M (2006) Sets of
Briët J, Dvir Z, Gopi S (2017) Outlaw distributions and k-recurrence but not (k + 1)-recurrence. Ann Inst Fou-
locally decodable codes. Proc ITCS arXiv:1609.06355 rier 56(4):839–849
Brown T, Graham R, Landman B (1999) On the set of Frantzikinakis N, Host B, Kra B (2007) Multiple recur-
common differences in van der Waerden’s theorem on rence and convergence for sets related to the primes.
arithmetic progressions. Can Math Bull 42:25–36 J Reine Angew Math 611:131–144
Carathéodory C (1968) Vorlesungen über reelle Frantzikinakis N, Lesigne E, Wierdl M (2012) Random
Funktionen, 3rd edn. Chelsea Publishing, New York sequences and pointwise convergence of multiple
Chazottes J, Ugalde E (2005) Entropy estimation and fluc- ergodic averages. Indiana Univ Math J 61:585–617
tuations of hitting and recurrence times for Gibbsian Frantzikinakis N, Host B, Kra B (2013) The polynomial
sources. Discrete Contin Dyn Syst Ser B 5(3):565–586 multidimensional Szemerédi Theorem along shifted
Christ M. On random multilinear operator inequalities. primes. Israel J Math 194:331–348
Unpublished manuscript. Available at arXiv:1108.5655 Frantzikinakis N, Lesigne E, Wierdl M (2016) Random
Conze J, Lesigne E (1984) Théorèmes ergodiques pour differences in Szemerédi’s theorem and related results.
des mesures diagonales. Bull Soc Math France J Anal Math 130:91–133
112(2):143–175 Furstenberg H (1977) Ergodic behavior of diagonal mea-
Conze J, Lesigne E (1988) Sur un théorème ergodique pour sures and a theorem of Szemerédi on arithmetic pro-
des mesures diagonales. In: Probabilités, Publ Inst gressions. J Anal Math 71:204–256
Ergodic Theory: Recurrence 77

Furstenberg H (1981) Recurrence in ergodic theory and Kac M (1947) On the notion of recurrence in discrete
combinatorial number theory. Princeton University stochastic processes. Bull Am Math Soc
Press, Princeton 53:1002–10010
Furstenberg H, Katznelson Y (1979) An ergodic Kamae T, Mendés-France M (1978) Van der Corput’s
Szemerédi theorem for commuting transformations. difference theorem. Israel J Math 31:335–342
J Anal Math 34:275–291 Karageorgos D, Koutsogiannis A. Integer part independent
Furstenberg H, Katznelson Y (1985) An ergodic polynomial averages and applications along primes. To
Szemerédi theorem for IP-systems and combinatorial appear in Studia Math. arXiv:1708.06820
theory. J Anal Math 45:117–168 Katznelson Y (2001) Chromatic numbers of Cayley graphs
Furstenberg H, Katznelson Y (1991) A density version of on ℤ and recurrence. Paul Erdös and his mathematics
the Hales-Jewett theorem. J Anal Math 57:64–119 (Budapest, 1999). Combinatorica 21(2):211–219
Furstenberg H, Weiss B (1996) A mean ergodic theorem Khintchine A (1934) Eine Verscharfung des Poincaréschen
for ð1=N Þ N 2
f ðT n xÞg T n x . Convergence in ergo- “Wiederkehrsatzes”. Comp Math 1:177–179
n¼1
Koutsogiannis A (2018a) Closest integer polynomial mul-
dic theory and probability (Columbus, OH, 1993), tiple recurrence along shifted primes. Ergodic Theory
Ohio State Univ Math Res Inst Publ, vol 5, de Gruyter, Dyn Syst 38:666–685
Berlin, pp 193–227 Koutsogiannis A (2018b) Integer part polynomial correla-
Furstenberg H, Katznelson Y, Ornstein D (1982) The ergo- tion sequences. Ergodic Theory Dyn Syst
dic theoretical proof of Szemerédi’s theorem. Bull Am 38:1525–1542
Math Soc 7(3):527–552 Kra B (2006a) The Green-Tao theorem on arithmetic pro-
Furstenberg H, Katznelson Y, Weiss B (1990) Ergodic gressions in the primes: an ergodic point of view. Bull
theory and configurations in sets of positive density. Am Math Soc 43:3–23
In: Mathematics of Ramsey theory. Algorithms combi- Kra B (2006b) From combinatorics to ergodic theory and
natorics, vol 5. Springer, Berlin, pp 184–198 back again. In: Proceedings of international congress of
Galatolo S, Kim DH, Park KK (2006) The recurrence time mathematicians, vol III. Madrid, pp 57–76
for ergodic systems with infinite invariant measures. Kra B (2007) Ergodic methods in additive combinatorics.
Nonlinearity 19:2567–2580 In: Additive combinatorics, CRM proceedings and lec-
Glasner E (2003) Ergodic theory via joinings. Mathemat- ture notes, vol 43. American Mathematical Society,
ical surveys and monographs, vol 101. American Math- Providence, pp 103–143
ematical Society, Providence Kra B (2011) Poincare recurrence and number theory:
Gowers W (2001) A new proof of Szemerédi’s theorem. thirty years later. Bull Am Math Soc 48:497–501
Geom Funct Anal 11:465–588 Kriz I (1987) Large independent sets in shift invariant
Graham RL (1994) Recent trends in Euclidean Ramsey graphs. Solution of Bergelson’s problem. Graphs
theory. Trends in discrete mathematics. Discrete Math Comb 3:145–158
136(1–3):119–127 Leibman A (2002) Lower bounds for ergodic averages.
Green B, Tao T (2008) The primes contain arbitrarily long Ergodic Theory Dyn Syst 22:863–872
arithmetic progressions. Ann Math 167:481–547 Leibman A (2005a) Pointwise convergence of ergodic
Green B, Tao T (2010) Linear equations in primes. Ann averages for polynomial sequences of rotations of a
Math 171:1753–1850 nilmanifold. Ergodic Theory Dyn Syst 25:201–213
Green B, Tao T (2012a) The quantitative behaviour of Leibman A (2005b) Pointwise convergence of ergodic aver-
polynomial orbits on nilmanifolds. Ann Math ages for polynomial actions of ℤd by translations on a
175:465–540 nilmanifold. Ergodic Theory Dyn Syst 25:215–225
Green B, Tao T (2012b) The Möbius function is strongly McCutcheon R (1995) Three results in recurrence. In:
orthogonal to nilsequences. Ann Math 175:541–566 Ergodic theory and its connections with harmonic anal-
Green B, Tao T (2014) On the quantitative distribution of ysis (Alexandria, 1993). London mathematical society
polynomial nilsequences- erratum. Ann Math lecture note series, vol 205. Cambridge University
179:1175–1183, arXiv:1311.6170v3 Press, Cambridge, pp 349–358
Green B, Tao T, Ziegler T (2012) An inverse theorem for the McCutcheon R (2005) FVIP systems and multiple recur-
Gowers U s+1[N ]-norm. Ann Math 176(2):1231–1372 rence. Israel J Math 146:157–188
Griesmer J. Bohr topology and difference sets for some Meyerovitch T. On multiple and polynomial recurrent
abelian groups. Preprint, arXiv:1608.01014 extensions of infinite measure preserving transforma-
Hales A, Jewett R (1963) Regularity and positional games. tions. Unpublished. Available at arXiv:0703914v2
Trans Am Math Soc 106:222–229 Ornstein D, Weiss B (1993) Entropy and data compression
Hasley T, Jensen M (2004) Hurricanes and butterflies. schemes. IEEE Trans Inform Theory 39:78–83
Nature 428:127–128 Peluse S, Prendiville S. Quantitative bounds in the non-
Host B, Kra B (2005) Nonconventional ergodic averages linear Roth theorem. Preprint, arXiv:1903.02592
and nilmanifolds. Ann Math 161:397–488 Petersen K (1989) Ergodic theory. Cambridge studies in
Host B, Kra B (2018) Nilpotent structures in Ergodic advanced mathematics, vol 2. Cambridge University
theory. Mathematical surveys and monographs, Press, Cambridge
vol 236. American Mathematical Society, Providence
78 Ergodic Theory: Recurrence

Poincaré H (1890) Sur le problème des trois corps et les Tao T, Teräväinen J. The structure of correlations of mul-
équations de la dynamique. Acta Math 13:1–270 tiplicative functions at almost all scales, with applica-
Polymath DHJ (2012) A new proof of the density Hales- tions to the Chowla and Elliott conjectures. Preprint,
Jewett theorem. Ann Math 175:1283–1327 arXiv:1809.02518
Prendiville S (2017) Quantitative bounds in the polynomial Tao T, Ziegler T (2008) The primes contain arbitrarily long
Szemerédi theorem: the homogeneous case. Discrete polynomial progressions. Acta Math 201:213–305
Anal 5:1–34 von Newmman J (1932) Proof of the Quasi-ergodic
Rosenblatt J, Wierdl M (1995) Pointwise ergodic theorems hypothesis. Proc Natl Acad Sci U S A 18(1):70–82
via harmonic analysis. In: Ergodic theory and its connec- Walters P (1982) An introduction to ergodic theory. Grad-
tions with harmonic analysis (Alexandria, 1993). London uate texts in mathematics, vol 79. Springer, New York/
mathematical society lecture note series, vol 205. Cam- Berlin
bridge University Press, Cambridge, pp 3–151 Weiss B (2000) Single orbit dynamics. CBMS regional
Sárközy A (1978) On difference sets of integers III. Acta conference series in mathematics, vol 95. American
Math Acad Sci Hungar 31:125–149 Mathematical Society, Providence
Shkredov I (2002) Recurrence in the mean. Mat Zametki Wooley T, Ziegler T (2012) Multiple recurrence and
72(4):625–632; translation in Math Notes convergence along the primes. Am J Math
(2002) 72(3–4):576–582 134:1705–1732
Sklar L (2004) Philosophy of statistical mechanics. In: Wyner A, Ziv J (1989) Some asymptotic properties of the
Zalta EN (ed) The Stanford encyclopedia of philosophy entropy of a stationary ergodic data source with appli-
(Summer 2004 Edition). http://plato.stanford.edu/ cations to data compression. IEEE Trans Inform The-
archives/sum2004/entries/statphys-statmech/ ory 35:1250–1258
Sun W (2015) Multiple recurrence and convergence for Zermelo E (1896) Über einen Satz der Dynamik und die
certain averages along shifted primes. Ergodic Theory mechanische Wärmetheorie. Ann Phys 57:485–494;
Dyn Syst 35(5):1592–1609 English translation (1966) On a theorem of dynamics
Szemerédi E (1975) On sets of integers containing no and the mechanical theory of heat. In: Brush SG
k elements in arithmetic progression. Acta Arith (ed) Kinetic theory, vol II. Oxford, pp 208–217
27:299–345 Ziegler T (2006) Nilfactors of ℝm-actions and configura-
Tao T, Teräväinen J. The structure of logarithmically aver- tions in sets of positive upper density in ℝm. J Anal
aged correlations of multiplicative functions, with Math 99:249–266
applications to the Chowla and Elliott conjectures. To Ziegler T (2007) Universal characteristic factors and
appear in Duke Math J. arXiv:1708.02610 Furstenberg averages. J Am Math Soc 20:53–97
 Iteration repeated applications of the map
Ergodic Theorems T above to arrive at the state of the system
after n units of time.
Andrés del Junco Maximal inequality an inequality which allows
Department of Mathematics, University of one to bound the pointwise oscillation of a
Toronto, Toronto, ON, Canada sequence of functions. An essential tool for
proving pointwise ergodic theorems.
Mean ergodic theorem an assertion that ergodic
Article Outline averages converge with respect to some norm
on a space of functions.
Glossary Operator any linear operator U on a vector
Definition of the Subject space of functions on X, for example one aris-
Introduction ing from a dynamical system T by setting Uf
Ergodic Theorems for Measure-Preserving Maps (x) ¼ f(Tx). More generally any linear transfor-
Generalizations to Continuous Time and Higher- mation on a real or complex vector space.
Dimensional Time Orbit of x the forward images x, Tx, T2X. . . of
Pointwise Ergodic Theorems for Operators x  X under iteration of T. When T is invertible
Subadditive and Multiplicative Ergodic Theorems one may consider the forward, backward or
Entropy and the Shannon–McMillan–Breiman two-sided orbit of x.
Theorem Pointwise ergodic theorem an assertion that
Amenable Groups ergodic averages An f(x) converge for some or
Subsequence and Weighted Theorems all x  X, usually for a.e. x.
Ergodic Theorems and Multiple Recurrence Positive contraction an operator T on a space of
Rates of Convergence functions endowed with a norm k  k such that
Ergodic Theorems for Non-amenable Groups T maps positive functions to positive functions
Future Directions and kTf k  j f j.
Bibliography Stationary process a sequence (X1, X2, . . .) of
random variables (real or complex-valued
Glossary measurable functions) on a probability space
whose joint distributions are invariant under
Automorphism a dynamical system T : X ! X, shifting (X1, X2, . . .) to (X2, X3, . . .).
where X is a measure space and T is an invert- Uniform distribution a sequence {xn} in [0, 1]
ible map preserving measure. is uniformly distributed if for each interval
Dynamical system in its broadest sense, any set I  [0, 1], the time it spends in I is asymptot-
X, with a map T : X ! X. The classical example ically proportional to the length of I.
is: X is a set whose points are the states of some
physical system and the state x is succeeded by
the state Tx after one unit of time. Definition of the Subject
Ergodic average if f is a function on X let
An f ðxÞ ¼ n1 n1 i
i¼0 f T x ; the average of the
Ergodic theorems are assertions about the long-
values of f over the first n points in the orbit term statistical behavior of a dynamical system.
of x. The subject arose out of Boltzmann’s ergodic
Ergodic theorem an assertion that ergodic aver- hypothesis which sought to equate the spatial
ages converge in some sense. average of a function over the set of states in a
physical system having a fixed energy with the
© Springer-Verlag 2009 79
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_176
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_176
80 Ergodic Theorems

time average of the function observed by starting A is asymptotically equal to m(A), a very satisfying
with a particular state and following its evolution justification of intuition. For example, applying
over a longtime period. this to the coin-tossing sequence one obtains the
strong law of large numbers which asserts that
almost every infinite sequence of coin tosses has
Introduction tails occurring with asymptotic frequency 12. One
also obtains Borel’s theorem on normal numbers
Suppose that (X, ℬ, m) is a measure space and which asserts that for almost all x  [0, 1] each
T : (X, ℬ, m) ! (X, ℬ, m) is a measurable and digit 0, 1, 2, . . ., 9 occurs with limiting frequency
measure-preserving transformation, that is m- 1
10. The so-called continued fraction transforma-
(T1E) ¼ m(E) for all E  ℬ. One important tion x 7! x1 mod 1 on (0, 1) has a finite
motivation for studying such maps is that a Ham- invariant measure 1þx dx
. (Throughout this article
iltonian physical system (see the article by x mod 1 denotes the fractional part of x.)
Petersen in this collection) gives rise to a one- Applying Birkhoff’s theorem then gives precise
parameter group {Tt : t  ℝ} of maps in the information about the frequency of occurrence of
phase space of the system which preserve any n  ℕ in the continued fraction expansion of
Lebesgue measure. The ergodic theorem of x, for a.e. x. See for example Billingsley
Birkhoff asserts that for f  L1(m) (Billingsley 1965).
n1 These are the classical roots of the subject of
1
f Tix ð1Þ ergodic theorems. The subject has evolved from
n i¼0 these simple origins into a vast field in its own
right, quite independent of physics or probability
converges a.e. and that if T is ergodic (to be theory. Nonetheless it still has close ties to both
defined shortly) then the limit is fdm. This may these areas and has also forged new links with
be viewed as a justification for Boltzmann’s ergo- many other areas of mathematics.
dic hypothesis that “space averages equal time Our purpose here is to give a broad overview of
averages”. See Zund (2002) for some history of the subject in a historical perspective. There are
the ergodic hypothesis. For physicists, then, the several excellent references, notably the books of
problem is reduced to showing that a given phys- Krengel (1985) and Tempelman (1992) which
ical system is ergodic, which can be very difficult. give a good picture of the state of the subject at
However ergodic systems arise in many natural the time they appeared. There has been tremen-
ways in mathematics. One example is rotation dous progress since then. The time is ripe for a
z 7! lz of the unit circle if l is a complex number much more comprehensive survey of the field
of modulus one which is not a root of unity. than is possible here.
Another is the shift transformation on a sequence Many topics are necessarily absent and many
of i.i.d. random variables, for example a coin- are only glimpsed. For example this article will
tossing sequence. Another is an automorphism not touch on random ergodic theorems. See the
of a compact Abelian group. Often a transforma- articles (Durand and Schneider 2003; Lemańczyk
tion possesses an invariant measure which is not et al. n.d.) for some references on this topic.
obvious at first sight. Knowledge of such a mea- I thank Mustafa Akcoglu, Ulrich Krengel,
sure can be a very useful tool. See Petersen’s Michael Lin, Dan Rudolph and particularly Joe
article for more examples. Rosenblatt and Vitaly Bergelson for many helpful
If (X, ℬ, m) is a probability space and T is comments and suggestions. I would like to dedi-
ergodic then Birkhoff’s ergodic theorem implies cate this article to Mustafa Akcoglu who has been
that if A is a measurable subset of X then for such an important contributor to the development
almost every x, the frequency with which x visits of ergodic theorems over the past 40 years. He has
Ergodic Theorems 81

also played a vital role in my mathematical devel- The first and most fundamental result about
opment as well as of many other mathematicians. endomorphisms is the celebrated recurrence theo-
He remains a source of inspiration to me, and a rem of Poincaré (1987).
valued friend.
Theorem 1 Suppose m is finite, A  ℬ and
m(A) > 0. Then for a.e. x  A there is an n > 0
Ergodic Theorems for Measure- such that T nx  A, in fact there are infinitely many
Preserving Maps such n.

Suppose (X, ℬ, m) is a measure space. It may be viewed as an ergodic theorem, in that

A (measure-preserving) endomorphism of it is a qualitative statement about how x behaves
(X, ℬ, m) is a measurable mapping T : X ! X under iteration of T. For a proof, observe that if
such that m(T 1E) ¼ m(E) for any measurable E  A is the measurable set of points which never
subset E  X. If T has a measurable inverse then return to A then for each n > 0 the set T nE is
one says that it is an automorphism. In this article disjoint from E. Applying T m one gets also
the unqualified terms “endomorphism” or “auto- T (nþm)E \ T mE ¼ ;. Thus E, T 1E, T 2E,
morphism” will always mean measure- . . . is an infinite sequence of disjoint sets all hav-
preserving. T is called ergodic if for all measur- ing measure m(E) so m(E) ¼ 0 since m(X) < 1.
able E one has Much later Kac (1947) formulated the following
quantitative version of Poincaré’s theorem.
T 1 E ¼ E ) mðEÞ ¼ 0 or mðEc Þ ¼ 0:
Theorem 2 Suppose that m(X) ¼ 1, T is ergodic
A very general class of examples comes from
and let rAx denote the time of first return to A, that
the notion of a stationary stochastic processin
is rA(x) is the least n > 0 such that T nx  A. Then
probability theory. A stochastic process is a
sequence of measurable functions f1, f2, . . . on a
probability space (X, ℬ, m) taking values in a mea- 1 1
r A dm ¼ , ð2Þ
surable space ðY, C Þ. The distribution of the pro- mðAÞ A mðAÞ
cess, a measure n on Y ℕ, is defined as the image of
m under the map ( f1, f2, . . .) : X ! Y ℕ. n captures that is, the expected value of the return time to A is
all the essential information about the process { fi}. m(A)1.
In effect one may view any process { fi} as a Koopman (1931) made the observation that
probability measure on Y ℕ. The process is said to associated to an automorphism T there is a unitary
be stationary if n is invariant under the left shift operator U ¼ UT defined on the Hilbert space
transformation S on Y ℕ, S( y)(i) ¼ y(i þ 1), that L2(m) by the formula Uf ¼ f ∘ T. This led von
is S is an endomorphism of the probability Neumann (von Neumann 1932) to prove his mean
space(Y ℕ, n). ergodic theorem.
From a probabilistic point of view the most
natural examples of stationary stochastic pro- Theorem 3 Suppose ℋ is a Hilbert space, U is a
cesses are independent identically distributed pro- unitary operator on ℋ and let P denote the
cesses, namely the case when n is a product orthogonal projection on the subspace of
measure lℕ for some measure l on ðY, C Þ. More U-invariant vectors. Then for any x  ℋ one has
generally one can consider a stationary Markov
process defined by transition probabilities on the n1
state space Y and an invariant probability on Y. See 1
U i x  Px ! 0 as n ! 1: ð3Þ
for example Chap. 7 in (Breiman 1968), also n i¼0
section “Pointwise Ergodic Theorems for Opera-
tors” below.
82 Ergodic Theorems

Von Neumann’s theorem is usually quoted as (closed) subspace of U-invariant vectors and I0
above but to be historically accurate he dealt with the (usually not closed) subspace of vectors of
unitary operators indexed by a continuous time the form f  Uf. It is easy to check that any vector
parameter. orthogonal to I0 must be in I, whence the subspace
Inspired by von Neumann’s theorem Birkhoff I þ I0 is dense in ℋ. For any vector of the form
very soon proved his pointwise ergodic theorem x ¼ y þ y0, y  I, y0  I0 it is clear that An x ¼
1 n1 i
i¼0 U x converges to y ¼ Px, since Any ¼ y and
(Birkhoff 1931). In spite of the later publication
n
by vonNeumann his result did come first. See if y ¼ z  Uz then the telescoping sum Any0 ¼
0

(Bergelson 2007; Zund 2002) for an interesting n1(z  Unz) converges to 0. This establishes the
discussion of the history of the two theorems and desired convergence for x  I þ I0 and it is easy to
of the interaction between Birkhoff and von extend it to the closure of I þ I0 since the operators
Neumann. An are contractions of ℋ (kAnk  1). Lorch
(1939) used a soft argument in a similar spirit to
Theorem 4 Suppose (X, ℬ, m) is a measure extend non Neumann’s theorem from the case of a
space and T is an endomorphism of (X, ℬ, m). unitary operator on a Hilbert space to that of an
Then for any f  L1 ¼ L1(m) there is a T-invariant arbitrary linear contraction on any reflexive
function g  L1 such that Banachspace. Sine (1970) gave a necessary and
sufficient condition for the strong convergence of
n1 the ergodic averages of a contraction on an arbi-
1
A n f ðx Þ ¼ f T i x ! gð x Þ a:e: ð4Þ trary Banach space.
n i¼0 Birkhoff’s theorem has the distinction of being
one of the most reproved theorems of twentieth
Moreover if m is finite then the convergence century mathematics. One approach to the
also holds with respect to the L1 norm and one pointwise convergence, parallel to the argument
has E g dm ¼ E fdm for all T-invariant subsets E. to argument just seen, is to find a dense subspace
Again this formulation of Birkhoff’s theorem E of L1 so that the convergence holds for all f  E
is not historically accurate as he dealt with a and then try to extend the convergence to all
smooth flow on a manifold. It was soon observed f  L1 by an approximation argument. The first
that the theorem, and its proof, remain valid for an step is not too hard. For simplicity assume that m is
abstract automorphism of a measure space a finite measure. As in the proof of von Neumann’s
although the realization that T need not be invert- theorem the subspace E spanned by the T-invariant
ible seems to have taken a little longer. L1functions together with functions of the form
The notation An f ¼ An(T )f as above will occur g  Tg, g  L1, is dense in L1(m). This can be
often in the sequel. Whenever T is an endomor- seen by using the Hahn–Banach theorem and the
phism one uses the notation Tf ¼ f ∘ T and with duality of L1 and L1. (Here one needs finiteness of
this notation An ðT Þ ¼ 1n n1 i
i¼0 T . When the scalars m to know that L1  L1.) The pointwise conver-
(ℝ or ℂ) for an L1 space are not specified the gence of An f for invariant f is trivial and for f ¼ g 
notation should be understood as referring to Tg it follows from telescoping of the sum and the
either possibility. In most of the theorems in this fact that n1T ng ! 0 a.e. This last can be shown by
article the complex case follows easily from the using the Borel–Cantelli lemma. The second step,
real and any indications about proofs will refer to extending pointwise convergence, as opposed to
the real case. norm convergence, for f in a dense subspace to all
Although von Neumann originally used spec- f in L1 is a delicate matter, requiring a maximal
tral theory to prove his result, there is a quick inequality.
proof, attributed to Riesz by Hopf in his 1937 Roughly speaking a maximal inequality is an
book (Hopf 1937), which uses only elementary inequality which bounds the pointwise oscillation
properties of Hilbert space. Let I denote the of An f in terms of the norm of f. The now standard
Ergodic Theorems 83

maximal inequality in the context of Birkhoff’s a.e. convergence. Moreover a simple application
theorem is the following, due to Kakutani and of Fatou’s lemma shows that the limiting function
Yosida (1939). Birkhoff’s proof of his theorem is in L1, hence finite a.e.
includes a weaker version of this result. Let Sn f ¼ There are many proofs of (5). Two of particular
nAn f, the nth partial sum of the iterates of f. interest are Garsia’s (1970), perhaps the shortest
and most mysterious, and the proof via the filling
Theorem 5 Given any real f  L1 let A ¼ scheme of Chacón and Ornstein (1960), perhaps
[n1{Sn f  0}. Then the most intuitive, which goes like this. Given a
function g  L1 write g+ ¼ max (g, 0), g ¼
g+  g and let Ug ¼ Tg+  g. Interpretation: the
f dm  0: ð5Þ region between the graph of g+ and the X-axis is a
A sheaf of vertical spaghetti sticks, the intervals
[0, g+(x)], x in X, and g is a hole. Now move
Moreover if one sets Mf ¼ sup An f then for
n1 the spaghetti (horizontally) by T and then let it
any α > 0 drop (vertically) into the hole leaving a new hole
and a new sheaf which are the negative and pos-
1 itive parts of Ug. Now let E0 ¼ [n1{Unf  0}, the
mfMf > ag  k f k1 ð6Þ
a
set of points x at which the hole is eventually filled
A distributional inequality such as (6) will be after finitely many iterations of U.
referred to as a weak L1inequality. Note that (6) The key point is that E ¼ E0. Indeed if Sn f
follows easily from (5) by applying (5) to f  α, at (x)  0 for some n, then the total linear height of
least in the case when m is finite. With the maximal sticks over x, Tx, . . .T n1x is greater than the total
inequality in hand it is straightforward to com- linear depth of holes at these points. The only way
plete the proof of Birkhoff’s theorem. For a real- that spaghetti can escape from these points is by
valued function f let first filling the hole at x, which shows x  E0.
Similar thinking shows that if x  E0 and the hole
Osc f ¼ lim sup An f  lim inf An f : ð7Þ at x is filled for the first time at time n then Sn f
(x)  0, so x  E, and that all the spaghetti that
Osc f ¼ 0 a.e. if and only if An f converges goes into the hole at x comes from points Tix
a.e. (to a possibly infinite limit). One has which belong to E0. This shows that E ¼ E0 and
lim sup An f  Mf  M j f j and by symmetry that the part of the hole lying beneath E is even-
lim inf An f  M j f j, so Osc f  2M j f j. To tually filled by spaghetti coming from E0 ¼ E.
establish the convergence of An f for a real-valued Thus the amount of spaghetti over E is no less
f  L1 let ϵ > 0 and write f ¼ g þ h with g  E than the size of the hole under E, that is
(the subspace where convergence has already E f dm  0.
been established), h  L1 and khk < ϵ. Then Most proofs of Birkhoff’s theorem use a max-
since Osc g ¼ 0 one has Osc f ¼ Osc h. Thus imal inequality in some form but a few avoid it
for any fixed α > 0, using (6) altogether, for example (Katok and Hasselblatt
a 1995; Shields 1987). It is also straightforward to
mfOsc f > ag ¼ mfOsc h > ag  m Mh >
2 deduce Birkhoff’s theorem directly from a maxi-
2khk1 2ϵ mal inequality, as Birkhoff does, without first
 < :
a a establishing convergence on a dense subspace.
ð8Þ However the technique of proving a pointwise
convergence theorem by finding an appropriate
Since ϵ > 0 was arbitrary one concludes that dense subspace and a suitable maximal inequality
m{Osc f > α} ¼ 0 and since α > 0 is arbitrary it has proved extremely useful, not only in ergodic
follows that m{Osc f > 0} ¼ 0, establishing the theory.
84 Ergodic Theorems

Indeed, in some sense maximal inequalities are There is a special setting where one has uni-
unavoidable: this is the content of the following form convergence in the ergodic theorem. Sup-
principle proved already in 1926 by Banach. The pose T is a homeomorphism of a compact metric
principle has many formulations; the following space X. By a theorem of Krylov and
one is a slight simplification of the one to be Bogoliouboff (1937) there is at least one proba-
found in (Krengel 1985). Suppose B is a Banach bility measure on the Borel s-algebra of X which
space, (X, ℬ, m) is a finite measure space and let is invariant under T. T is said to be uniquely
E denote the space of m-equivalence classes of ergodic if there is only one Borel probability
measurable real-valued functions on X. A linear measure, say m, invariant under T. It is easy to
map T : B ! E is said to be continuous in measure see that when this is the case then T is an ergodic
if for each ϵ > 0 automorphism of (X, ℬ, m). As an example, if α is
an irrational number then the rotation z 7! e2πiαz
xn  x ! 0 ) mfjTxn Txj> ϵg ! 0: is a uniquely ergodic transformation of the circle
{| z| ¼1}. Equivalently x 7! x þ α mod 1 is a
Suppose that Tn is a sequence of linear maps uniquely ergodic map on [0, 1]. A quick way to
from B to E which are continuous in measure and see this is to show that the Fourierco-efficients
let Mx ¼ supn j T n x j. Of course if Tnx converges mðnÞ of any invariant probability m are zero for
a.e. to a finite limit then Mx < 1 a.e. n 6¼ 0. The Jewett–Krieger theorem (see Jewett
(1969) and Krieger (1972)) guarantees that unique
Theorem 6 (Banach Principle) Suppose ergodicity is ubiquitous in the sense that any auto-
Mx < 1 a.e. for each x  X. Then there is a morphism of a probability space is measure-
function C(l) such that C(l) ! 0 as l ! 1 such theoretically isomorphic to a uniquely ergodic
that for all x  B and l > 0 one has homeomorphism. The following important result
is due to Oxtoby (1952)).

m Mx  l x  CðlÞ: ð9Þ Theorem 7 If T is uniquely ergodic, m is its

unique invariant probability measure and
Moreover the set of x for which Tnx converges f  C(X) then the ergodic averages An( f ) con-
a.e. is closed in B. verge uniformly to fdm.
The first chapter of Garsia’s book (1970) con-
tains a nice introduction to the Banach principle. This result can be proved along the same lines
It should be noted that, for general integrable f, as the proof given above of von Neumann’s theo-
M f need not be in L1. However if f  L p for p > 1 rem. Ina nutshell, one uses the fact that the dual of
then M f does belong to L p and one has the Cℝ(X) is the space of finite signed measures on
estimate X and the unique ergodicity to show that functions
of the form f  f ∘ T together with the invariant
p
kMf kp  k f kp : ð10Þ functions (which are just constant functions) span
p1
a dense subspace of C(X).
This can be derived from (6) using also the A sequence {xn} in the interval [0, 1] is said to
(obvious) fact that kMf k1  k f k1. Any such be uniformly distributed if 1n ni¼1 f ðxn Þ !
1
estimate on the norm of a maximal function will 0 f ðxÞdx for any f  C[0, 1] (equivalently for
be called a strong Lp inequality. It also follows any Riemann integrable f or for any f ¼ 1I where
from (6) that if m(X) is finite and f  L log L, that I is any subinterval of [0, 1]). As a simple appli-
is j f j log+ j f j dm < 1, then Mf  L1. In fact cation of Theorem 7 one obtains the linear case of
Ornstein (1971) has shown that the converse of the following result of Weyl (1916).
this last statement holds provided T is ergodic.
Ergodic Theorems 85

Theorem 8 If α is irrational and p(x) is any non- If n is a s-finite measure equivalent to m

constant polynomial with integer coefficients then which is invariant under t then the ergodic the-
the sequence {p(n)α mod 1} is uniformly distributed. ory of t can be reduced to the measure-
preserving case using s. The interesting case is
Furstenberg (1967), see also (Furstenberg when there is no such n. It was an open problem
1981), has shown that Weyl’s result, in full gen- for some time whether there is always an equiv-
erality, can be deduced from the unique ergodicity alent invariant measure. In 1960 Ornstein (1960)
of certain affine transformations of higher dimen- gave an example of a t which does not have an
sional tori. For polynomials of degree k > 1 equivalent invariant measure. It is curious that,
Weyl’s result is usually proved by inductively with hindsight, such examples were already
reducing to the case k ¼ 1, using the following known in the fifties to people studying von Neu-
important lemma of van der Corput (see for exam- mann algebras.
ple (Kuipers and Niederreiter 1974)). For f  L1 let Sn f ¼ ni¼0 T i f . t is said to be
conservative if there is no set E with m(E) > 0
Theorem 9 Suppose that for each fixed h > 0 the such that tiE, i ¼ 0, 1, . . . are pairwise disjoint,
sequence that is if the Poincaré recurrence theorem remains
valid. For example, the shift on ℤ is not conser-
vative. Hurewicz (1944) proved the following
fxnþh  xn mod 1gn
ratio ergodic theorem.
is uniformly distributed. Then {xn} is uniformly
distributed. Theorem 10 Suppose t is conservative, f, g  L1
When m is infinite and T is ergodic the limiting and g(x) > 0 a.e. Then Sn f/Sng converges a.e. to a
function in Birkhoff’s theorem is 0 a.e. In 1937 f dm
t-invariant limit h. If t is ergodic then h ¼ .
Hopf (1937) proved a generalization of Birkhoff’s gdm
theorem which is more meaningfulin the case of
an infinite invariant measure. It is a special case of In the case when m is t-invariant one has Tf ¼
a later theorem of Hurewicz (1944), which we will f ∘ t. If m is invariant and finite, taking g ¼ 1 one
discuss first. recovers Birkhoff’s theorem. If m is invariant and
Suppose that (X, ℬ, m) is a s-finite measure s-finite then Hurewicz’s theorem becomes the
space. If n is another s-finite measure on ℬ write theorem of Hopf alluded to earlier.
n  m (n is absolutely continuous relative to m) if Wiener and Wintner (1941) proved the follow-
m(E) ¼ 0 implies n(E) ¼ 0 and one writes n~m if ing variant of Birkhoff’s theorem.
n  m and m  n. Consider a non-singular auto-
morphism t : X ! X, meaning that t is measurable Theorem 11 (Wiener–Wintner) Suppose T is
with a measurable inverse and that m(E) ¼ 0 if and an automorphism of a probability space
only if m(tE) ¼ 0. In other words n ¼ m  t~m. By (X, ℬ, m). Then for any f  L1 there is a subset
the Radon–Nikodym theorem there is a function X 0  X of measure one such that for each x  X 0
r  L1(m) such that r > 0 a.e. and n(E) ¼ Er dm and l  ℂ of modulus 1 the sequence
1 n1 i i
i¼0 l T f ðxÞ converges.
for all measurable E. In order to obtain an associ-
n
ated operator T on L1 which is an(invertible) isom-
etry one defines It is an easy consequence of the ergodic theo-
Tf ðxÞ ¼ rðxÞf ðtxÞ: ð11Þ rem that one has a.e. convergence for a given f and
l but the point here is that the set on which the
The dual operator on L1 is then given by T g ¼ convergence occurs is independent of l.
g ∘ t1.
86 Ergodic Theorems

Generalizations to Continuous Time and In Wiener (1939) proved the following result
Higher-Dimensional Time for actions of G ¼ ℝk and ergodic averages over
Euclidean balls Br ¼ {x  ℝk : kxk2  r}.
A (measure-preserving) flow is a one-parameter
group {Tt, t  ℝ} of automorphisms of Theorem 12 Suppose T is an action of ℝk on
(X, ℬ, m), that is Ttþs ¼ TtTs, such that Ttx is (X, ℬ, m) and f  L1(m).
measurable as a function of (t, x). It will always
be implicitly assumed that the map (t, x) 7! Ttx
from ℝ X to X is measurable. Theorem 4 gen- (a) For f  L1 lim ABr f ¼ g exists a.e. If m is finite
r!1
eralizes to flows by replacing sums with integrals the convergence also holds with respect to the
and this generalization follows without difficulty L1-norm, g is T-invariant and Igdm ¼ I f dm
from Theorem 4. (As already observed this obser- for every T-invariant set I.
vation reverses the historical record.) Theorem 4 (b) lim ABr f ¼ f a.e.
may be viewed as a theorem about the “discrete r!0

flow” {Tn ¼ Tn : n  ℤ}. Wiener was the first to

generalize Birkhoff’s theorem to families of auto- The local aspect of Wiener’s theorem is closely
morphisms {Tg} indexed by groups more general related to the Lebesgue differentiation theorem,
than ℝ or ℤ. see for example Proposition 3.5.4 of (Krantz and
A measure-preserving flow is an action of ℝ Parks 1999), which, in its simplest form, states
while a single automorphism corresponds to an that for f  L1(ℝk, m) one has a.e. convergence of
1
action of ℤ. A (measure-preserving) action of a mðBr Þ Br f ðx þ tÞdt to f(x) as r ! 0. The local
group G is a homomorphism T : g 7! Tg from ergodic theorem implies Lebesgue’s theorem,
G into the group of automorphisms of a measure simply by considering the action of ℝk on itself
space (X, ℬ, m) (satisfying the appropriate joint by translation. In fact the local ergodic theorem
measurability condition in case G is not discrete). can also be deduced from Lebesgue’s theorem by
Suppose now that G ¼ ℝk or ℤk and T is an action a simple application of Fubini’s theorem (see for
of G on (X, ℬ, m). In the case of ℤk an action example (Krengel 1985), Chap. 1, Theorem 2.4 in
amounts to an arbitrary choice of commuting the case k ¼ 1).
maps T1, . . .Tk, Ti ¼ T(ei) where ei is the standard The key point in Wiener’s proof is the follow-
basis of ℤk. In the case of ℝk one must specify ing weak L1 maximal inequality, similar to (6).
k commuting flows.
Let m denote counting measure on G in case Theorem 13 Let Mf ¼ supr>0 j ABr f j. Then one
G ¼ ℤk and Lebesgue measurein case G ¼ ℝk. For has
any subset E of G with m(E) < 1 let
1 C
AE f ðxÞ ¼ f T g x dmðgÞ: ð12Þ mfMf > ag  k f k1 , ð13Þ
mðEÞ E a

One may then ask whether AE f converges to a where C is a constant depending only on the
limit, either in the mean or pointwise, asE varies dimension d. (In fact one may take C ¼ 3d.)
through some sequence of sets which “grow In the case when T is the action of Rk on itself
large” or, in case G ¼ ℝk, “shrink to 0”. The by translation (13) is the well-known maximal
second case is referred to as a local ergodic theo- inequality for the Hardy–Littlewood maximal
rem. In the case of ergodic theorems at infinity the function (Krantz and Parks 1999, Lemma 3.5.3).
continuous and discrete theories are rather similar Wiener proves (13) by way of the following cov-
and often the continuous analogue of a discrete ering lemma. If B is a ball in ℝk let B0 denote the
result can be deduced from the discrete result. concentric ball with three times the radius.
Ergodic Theorems 87

Theorem 14 Suppose a compact subset K of ℝk of n’s by Theorem 30 below). Nonetheless

is covered by a (finite)collection U of open balls. Dunford (1951) and independently Zygmund
Then there exist pairwise disjoint Bi  U, i ¼ (1951) showed that one does have unrestricted
1, . . . k such that [i B0i covers K. convergence if f  Lp for some p > 1, provided
m is finite. Let T1, . . ., Tk be any k (possibly non-
To find the Bi it suffices to let B1 be the largest commuting!) automorphisms of a finite measure
ball in U, then B2the largest ball which does not space (X, ℬ, m). For n ¼ (n1, n2, . . ., nk) let
intersect B1 and in general Bn the largest ball T n ¼ T n11 . . . T nk k . (This is not an action of ℤk unless
which does not intersect [n1 i¼1 Bi : Then it is not the Ti commute with each other.) As before when
hard to argue that the B0i cover K. F  ℤk is a finite subset write AF f ¼ jFj 1
n  FTnf .
In general, a covering lemma is, roughly Finally let Pj f ¼ lim An T j f .
n
speaking, an assertion that, given a cover U ofa
set K in some space (often a group), one may find Theorem 16 For f  Lp
a subcollection U 0  U which covers a substantial
part of K and is in some sense efficient in that
lim ASn f ¼ P1 . . . Pk f ð15Þ
UU 0 1U  C, C some absolute constant. Cov- n!1
ering lemmas play an important role in the proofs
of many maximal inequalities. See Sect. 3.5 of both a.e. and in Lp.
(Krantz and Parks 1999) fora discussion of several The proof of Theorem 16 uses repeated appli-
of the best-known classical covering lemmas. cations of Birkhoff’s theorem and hinges on (10).
As Wiener was likely aware, the same kind of Somewhat surprisingly Hurewicz’s theorem
covering argument easily leads to maximal was not generalized to higher dimensions until
inequalities and ergodic theorems for averages the very recent work of Feldman (2007). In fact
over “sufficiently regular” sets. For example, in the theorem fails if one considers averages over
the case of G ¼ ℤk use the standard total order ≺ the cube [0, n  1]d in ℤd. However Feldman was
on G and for n  ℕk let Sn ¼ {m  Zk : 0 ≺ m ≺ n}. able to prove a suitably formulated generalization
Let e(n) denote the maximum value of ni/nj. Then of Hurewicz’s theorem for averages over symmet-
one has the following result. ric cubes.
For f  L1(ℝ) the classical Hilbert transform is
Theorem 15 For any e > 0 and any integrable f defined for a.e. t by
1 f ðt  sÞ
Hf ðtÞ ¼ lim ds ð16Þ
lim ASn f p ϵ!0þ jsj>ϵ s
n!1,eðnÞ<e
f ðtsÞ
Let H f ðtÞ ¼ p1 sup j jsj>ϵ s ds j, the
exists a.e. and the limit is characterized by the ϵ>0
same properties as in Theorem 12. Moreover corresponding maximal function. The proof that
there is a maximal inequality the limit (16) exists a.e. is based on the maximal
inequality
C
m sup jASn f j> a  k f k1 , ð14Þ C
eðnÞ<e a mfHf > lg  k f k1 , ð17Þ
l
where C is a constant depending only on e and the where C is an absolute constant. See Loomis
dimension k. (1946) for a proof of (17) using real-variable
If one does not impose a bound on e(n) the methods. Cotlar (1955b) proved the existence of
a.e. convergence of ASn f may fail for f  L1, the following ergodic analogue of the Hilbert
(although it does hold for any increasing sequence
88 Ergodic Theorems

transform (actually for the n-dimensional Hilbert case T is the shift map on the integers. The idea of
transform). transference could already be seen in Wiener’s
proof of the ℝn ergodic theorem. Transfer princi-
Theorem 17 Suppose {Tt} is a measure- ples in various forms of have become an important
preserving flow on a probability space (X, ℬ, m) tool in the study of ergodic theorems. See Bellow
and f  L1. Then (1999) fora very readable overview.

f ðT t xÞ Pointwise Ergodic Theorems for

lim dt ð18Þ
ϵ!0 ϵ<t<ϵ1 t Operators

exists for a.a. x. Early in the history of ergodic theorems there were
Motivated by Cotlar’s result Calderón (1968) attempts to generalize the ergodic theorem to
proved a general transfer principle which allows more general linear operators on Lp spaces, that
one to transfer maximal inequalities and conver- is, operators which do not arise by composition
gence theorems for functions of a real or integer with a mapping of X. In the case p ¼ 1 the main
variable to the ergodic setting. Although stated for motivation for this comes from the theory of Mar-
functions on ℝ it applies equally well to ℝk or ℤk. kov processes.
This principle subsumes Birkhoff’s theorem, Wie- If (X, ℬ) is a measurable space a sub-
ner’s theorem and Cotlar’s result. For simplicity stochastic kernel on X is a non-negative function
only a special case of Calderón’s result, in the P on X ℬ such that
discrete case, will be stated here.
Let T be an automorphism of a probability (a) for each x  XP(x, ) ¼ Px is a measure on ℬ
space (X, ℬ, m), let m denote counting measure such that Px(X)  1 and,
on ℤ and suppose s is a probability measure on ℤ. (b) P(, A) is a measurable function for each
For g  L1(m) define s(g)(n) ¼ i  ℤs(i)g(n þ i). A  ℬ.
For f  L1(m) let s(T )f ¼ i  ℤs(i)T if, which is
easily seen to converge a.e. and in L1-norm. Given It is most intuitive to think about the stochastic
a fixed sequence sn of probabilities define Mg ¼ case, namely when each Px is a probability mea-
supn sn g and MðT Þf ¼ supn sn ðT Þf . sure. One then views P(x, A) as the probability
that the point x moves into the set A in one unit of
Theorem 18 (Calderón) Suppose there is a con- time, so one has stochastic dynamics as opposed
stant C such that to deterministic dynamics, namely the case when
Px ¼ δTx for a map T. In this case the measures Px
are called transition probabilities.
C
mfMg > lg < kgk1 for all g  L1 ðmÞ: If m is a s-finite measure on X one may define
l
the measure Pm ¼ Pxdm(x). Pn is also meaning-
ð19Þ
ful if n is a finite signed measure. The case when
Then one also has Pm ¼ m is the stochastic analogue of measure-
preserving dynamics and the case when Pm  m
C is the analogue of non-singular dynamics. It is
mfMðT Þf > lg < k f k1 for all T and
l easy to see that given any s-finite measure l
g  L1 ðmÞ: there is always a m such that l  m and Pm  m.
ð20Þ Let L1 denote the space of finite signed measures n
such that n  m, which is identified with L1 ¼
In other words, in order to prove a maximal L1(m, ℝ) via the Radon–Nikodym theorem. If
inequality for general T it suffices to prove it in Pm  m then P maps L1 ðmÞ into itself so the
Ergodic Theorems 89

restriction of P is an operator T on L1(m). T is a p  L1 such that p > 0 a.e. 1 k

k¼0 T p is infinite
positive contraction, that is kTk  1 and T maps a.e. on C and finite a.e. on D. These results are the
non-negative functions to non-negative functions. cornerstone of much of the subsequent work on L1
As proved in, for example, (Neveu 1965), every contractions.
positive contraction arises in this way from a Theorem 19 contains Birkhoff’s theorem for an
substochastic kernel under the assumption that automorphism t of (X, ℬ, m), simply by defining
(X, ℬ) is standard. This simply means that there Tf ¼ f ∘ t. In fact one can also deduce Theorem 19
is some complete metric on X for which ℬ is the from Birkhoff’s theorem (if one assumes only that
s-algebra of Borel sets. Virtually all measurable X is standard). Indeed the hypotheses of Hopf’s
spaces encountered in analysis are standard so this theorem imply that the kernel P associated to T is
should be viewed as a technicality only. See stochastic (T 1 ¼ 1) and that m is P-invariant
(Foguel 1969; Krengel 1985; Neveu 1965) for (T1 ¼ 1). Hopf’s theorem then follows by apply-
more about the relation between kernels and pos- ing Birkhoff’s theorem to the shift on the station-
itive contractions. ary Markov process associated to P and m. In fact
The case when X is finite, P is stochastic and m Kakutani (1940) (see also Doob (1938)) had
is a probability measure is classical in probability already made essentially the same observation,
theory. P and m determine a probability measure n except that his result assumes the stationary Mar-
on X ℕ characterized by its values on cylinder sets, kov process to be already given.
namely for all x1, . . ., xn  X In 1955 Dunford and Schwartz (1955) made
essential use of Hopf’s work to prove the follow-
n x  Xℕ : xðiÞ ¼ xi , 1  i  n  1 ing result.
n1 ð21Þ
¼ mðx1 Þ Pxi ðxiþ1 Þ: Theorem 20 Suppose m(X) ¼ 1 and T is a (not
i¼1
necessarily positive) contraction with respect to
The co-ordinate functions Xi(x) ¼ x(i) on the both the L1 and L1 norms. Then the conclusion of
space X ℕ endowed with the probability n form a Hopf’s theorem remains valid.
Markov process, which will be stationary if and
only if m is P-invariant. For a general X the anal- Note that the assumption that T contracts the
ogous construction is possible provided X is L1-norm is meaningful, as L1  L1. The proof of
standard. the result is reduced to the positive case by defin-
Hopf (1954) initiated the systematic study of ing a positive contraction jTj analogous to the total
positive L1 contractions and proved the following variation of a complex measure.
ergodic theorem. Then in 1960 Chacón and Ornstein (1960)
proved a definitive ratio ergodic theorem for all
Theorem 19 Suppose (X, ℬ, m) is a probability positive contractions of L1 which generalizes both
space and T is a positive contraction on L1(m) Hopf’s theorem and the Hurewicz theorem.
satisfying T1 ¼ 1 and T 1 ¼ 1. Then
Theorem 21 Suppose T is a positive contraction
lim n!1 n1 n1 k
k¼0 T f exists a.e.
of L1(m), where m is s-finite, f, g  L1 and g  0.
Then
The importance of Hopf’s article lies less in
this convergence result than in the methods he
developed. He proved that the maximal inequal- n i
i¼0 T f
ity (5) generalizes to all positive L1contractions n i ð22Þ
i¼0 T g
T and used this to obtain the decomposition of
X into its conservative and dissipative parts 1 i
converges a.e. on the set i¼0 T g >0 :
C and D, characterized by the fact that for any
90 Ergodic Theorems

In 1963 Chacón (1963a) proved a very general form to (10) the classical strong Lp inequality for
theorem for non-positive operators which automorphisms.(25) was proved by A. Ionesco–
includes the Chacón–Ornstein theorem as well Tulcea (now Bellow) (Tulcea 1964) in the case of
as the Dunford–Schwartz theorem. positive invertible isometries of Lp. It is a result of
Banach (1993), see also (Lamperti 1958), that in
Theorem 22 Suppose T is a contraction of L1 this case T arises from a non-singular automor-
and pn  0 is a sequence of measurable functions phism t of (X, ℬ, m) in the form Tf ¼ r1/pf ∘ t. By
with the property that a series of reductions Bellow was able to show
that in this case (25) can be deduced from (10).
Akcoglu’s brilliant idea was to consider a dila-
g  L1 , j g j pn )j Tg j pnþ1 : ð23Þ
tion S of T which is a positive invertible isometry
on a larger Lp space Lp ¼ Lp ðY, C , nÞ. What this
Then
means is that there is a positive isometric injection
n i
i¼0 T f D : Lp ! Lp and a positive projection P on Lp
ð24Þ
n
i¼0 pi
whose range is D(Lp) such that DT n ¼ PS nD for
all n  0. Given the existence of such an S it is not
converges a.e. to a finite limit on the set hard to deduce (25) for T from (25) for S. In fact
1
i¼0 pi > 0 :
Akcoglu constructs a dilation only in the case
If T is an L1, L1-contraction and pn ¼ 1 for all when Lp is finite dimensional and shows how to
n then the hypotheses of this theorem are satisfied, reduce the proof of (25) to this case. In the finite
so Theorem 22 reduces to the result of Dunford dimensional case the construction is very concrete
and Schwartz. See (Chacon 1963b) for a concise and P is a conditional expectation operator. Later
overview of all of the above theorems in this Akcoglu and Kopp (1977) gave a construction in
section and the relations between them. the general case. It is noteworthy that the proof of
The identification of the limit in the Chacón– Akcoglu’s theorem consists ultimately of a long
Ornstein theorem on the conservative part C of string of reductions to the classical strong Lp
X is a difficult problem. It was solved by Neveu inequality (10), which in turn is a consequence
(1961) in case C ¼ X, and in general by Chacón of (5).
(1962). Chacón (1964) has shown that there is a
non-singular automorphism t of (X, ℬ, m) such
that for the associated invertible isometry T of Subadditive and Multiplicative Ergodic
L1(m) given by (11) there is an f  L1(m) such Theorems
that lim sup An f ¼ 1 and lim inf An f ¼  1.
In 1975 Akcoglu (1975) solved a major open Consider a family {Xn,m} of real-valued random
problem when he proved the following celebrated variables on a probability space indexed by the set
theorem. of pairs (n, m)  ℤ2 such that 0  n < m. {X(n,m)}
is called a (stationary) subadditive process if
Theorem 23 Suppose T : Lp 7! Lp is a positive
contraction. Then An f ¼ 1n n1 i (a) the joint distribution of {Xn,m} is the same as
i¼0 T f converges
that of {Xnþ1, mþ1},
a.e. Moreover one has the strong Lp inequality
(b) Xn,m  Xn,l þ Xl,m whenever n < l < m.

p Denoting the index set by {n < m}  ℤ2 the

supjAn f j  k f kp : ð25Þ
n p
p1 distribution of the process isa measure m on
ℝ{n<m} which is invariant under the shift Tx
As usual, the maximal inequality (25) is the (n, m) ¼ x(n þ 1, m þ 1). Thus there is no loss
key to the convergence. Note that it is identical in of generality in assuming that there is an
Ergodic Theorems 91

underlying endomorphism T such that Xn,m ∘ T ¼ (b) Suppose that for each i, j P(i, j) is a strictly
Xnþ1,mþ1. In 1968 Kingman (1968) proved the positive function such that log P(i, j) is integra-
1
following generalization of Birkhoff’s theorem. ble. Then the limit p ¼ lim ðP0,n ði, jÞÞn exists
g ¼ inf 1n X0,n dm is called the time constant of a.e. and is independent of i and j.
the process.
Partial results generalizing Kingman’s theorem
Theorem 24 If the Xn,m are integrable and to the multiparameter case were obtained by
γ >  1 then 1n X0,n converges a.e. and in L1- Smythe (1976) and Nguyen (1979). In 1981
norm to a -invariant limit X  L1 ðmÞ satisfying Akcoglu and Krengel (1981) obtained a definitive
Xdm ¼ g. multi-parameter subadditive theorem. They con-
sider an action {Tm} of the semigroup G ¼ ℤd0
It is easy to deduce from the above that if one by endomorphisms of a measure space (X, ℬ, m).
assumes only that Xþ 1
0,1 is integrable then n X 0,n still Using the standard total ordering ≺ of G an inter-
converges a.e. to a T-invariant limit X taking val in G is any set of the form {k  G : m ≺ k ≺ n}
values in[1, 1). for any m ≺ n  G. Let I denote the set of non-
Subadditive processes first arose in the work of empty intervals. Reversing the direction of the
Hammersley and Welsh (1965) on percolation inequality, they define a superadditive process as
theory. Here is an example. Let G be the graph a collection of integrable functions FI, I  I ,
with vertex set ℤ2 and with edges joining every such that
pair of nearest neighbors. Let E denote the edge
set and let {Te : e  E} be non-negative integrable (a) FI ∘ Tm ¼ FIþm,
i.i.d. random variables. To each finite path P in (b) FI  FI1 þ . . . þ FIk whenever I is the dis-
G associate the “travel time” T(P) ¼ e  ETe. For joint union of I1, . . ., Ik and
integers m > n  0 let Xn,m be the infimum of T(P) (c) γ ¼ supI  I I|1 FIdm < 1.
over all paths P joining (0, n) to (0, m). This is a
subadditive process with 0  γ < Tedm and it is A sequence {In} of sets in I is called regular if
not hard to see that the underlying endomorphism there is an increasing sequence I 0n such that I n 
is ergodic. Thus Kingman’s theorem yields the I 0n and j I 0n j C j I n j for some constant C.
result that 1n X0,n ! g a.e.
Suppose now that T is an ergodic automor- Theorem 26 (Akcoglu–Krengel) Suppose FI is
phism of a probability space (X, ℬ, m) and P is a a superadditive process and {In} is regular. Then
1
function on X taking values in the space of d d jI n j FI n converges a.e.
real matrices. Define Pn,m ¼ P(T m1x)
P(T m2x). . .P(T nx) and let Pn,m(i, j) denote the {FI} is additive if the inequality in (b) is
i, j entry of Pn,m. Then Xn,m ¼ log (kPn,mk) (use replaced by equality. In this case FI ¼ f ∘ Tn
any matrix norm) is a subadditive process so one nI
obtains the first part of the following result of where f is an integrable function. Thus in the
Furstenberg and Kesten (1960) originally proved additive case the Akcoglu–Krengel result is a
by more elaborate methods. The second part can theorem about ordinary multi-dimensional ergo-
also be deduced from the subadditive theorem dic averages, which is in fact a special case of an
with a little more work. See Kingman (1976) for earlier result of Tempel’man (1972a) (see section
details and for some other applications of sub- “Amenable Groups” below).
additive processes. Kingman’s proof of Theorem 24 hinged on the
existence of a certain (typically non-unique)
Theorem 25 (a) Suppose log+(| P|)dm < 1. decomposition for subadditive processes.
Then kP0,nk1/n converges a.e. to a finite limit. Akcoglu and Krengel’s proof of the multi-
parameter result does not depend on a Kingman-
92 Ergodic Theorems

type decomposition, in fact they show that there is Then for each u  Fi(x)\Fi  1(x)
no such decomposition in general. They prove a
1
weak maximal inequality lim log Pn ðxÞu ¼ li ðxÞ:
n
C
m supjI n j1 FIn > l < g, ð26Þ
l
(c) The functions mi and li are T-invariant.
where C is a constant depending only on the (d) If T is ergodic, det P(x) ¼ 1 a.e. and
dimension, and show that this is sufficient to
1
prove their result. In the case d ¼ 1 the lim sup log Pn dm > 0
n n
Akcoglu–Krengel argument provides a new and
more natural proof of Kingman’s theorem, similar
then the li are constants, l1 < 0 and lr > 0.
in spirit to Wiener’s arguments.
Raghunathan (1979) gave a much shorter proof
Akcoglu and Sucheston (1978) have proved a
of Oseledec’s theorem, valid for matrices with
ratio ergodic theorem for subadditive processes
entries in a locally compact normed field. He
with respect to a positive L1 contraction, general-
showed that it could be reduced to the
izing both the Chacón–Ornstein theorem and
Furstenberg–Kesten theorem by considering the
Kingman’s theorem.
exterior powers of P. Ruelle (1982) extended
In 1968 Oseledec (1968) proved his celebrated
Oseledec’s theorem to the case where P takes
multiplicative ergodic theorem, which gives very
values in the set of bounded operators on a Hilbert
precise information about the random matrix
space. Walters (1993) has given a proof (under
products studied by Furstenberg and Kesten. His
slightly stronger hypotheses) which avoids the
theoremis an important tool for the study of
matrix calculations and tools from multilinear
Lyapunov exponents in differentiable dynamics,
algebra used in other proofs.
see notably Pesin (1977). If A is a d d matrix let
kAk ¼ sup {kAxk : kxk ¼ 1} where kxk is the
Euclidean norm on ℝn.
Entropy and the Shannon–McMillan–
Breiman Theorem
Theorem 27 Suppose T is an endomorphism of
the probability space (X, ℬ, m). Suppose P is a
The notion of entropy was introduced by Shan-
measurable function on X whose values are d d
non in his landmark work (Shannon 1948)
real matrices such that log+kPkdm < 1 and let
which laid the foundations for a mathematical
Pn(x) ¼ P(T n  1x)P(T n  2x) . . . P(x). Then there
theory of information. Suppose (X, ℬ, m) is a
is a T-invariant subset X0 of X with measure 1 such
probability space, P is a finite measurable parti-
that for x  X0 the following hold.
tion of X and T is an automorphism of (X, ℬ, m).
P(x) denotes the atom of P containing x. The
1
entropy of P is
(a) lim n!1 Pn ðxÞPn ðxÞ 2n ¼ AðxÞ exists.
(b) Let 0  exp l1(x)  exp l2(x)  . . .  exp h ð PÞ ¼  mðpÞ logðmðpÞÞ
pP
lr(x) be the distinct eigenvalues of A(x) (r ¼ ð27Þ
r(x) may depend on x and l1 may be 1) ¼ logðmðPðxÞÞ dmðxÞ  0:
with multiplicities m1(x), . . . mr(x). Let Ei(x)
be the eigenspace corresponding to exp(li(x))
and set.  log (m(A)) may be viewed as a quantitative
measure of the amount of information contained
Fi ðxÞ ¼ E1 ðxÞ þ . . . þ Ei ðxÞ: in the statement that a randomly chosen x  X
Ergodic Theorems 93

happens to belong to A. So h(P) is the expected convergence and Breiman (1957) obtained the
information if one is about to observe which atom a.e. convergence.
of P a randomly chosen point falls in. See The original proofs of a.e. convergence used
Billingsley (1965) for more motivation of this the martingale convergence theorem, were not
concept. See also the article in this collection on very intuitive and did not generalize to ℤn-actions,
entropy by J. King or any introductory book on where the martingale theorem is not available.
ergodic theory, e.g. Petersen (1989). Ornstein and Weiss (1983) found a beautiful and
If P and Q are partitions P _ Q denotes the more natural argument which bypasses the mar-
common refinement which consists of all sets tingale theorem and allows generalization to a
p \ q, p  P, q  Q. It is intuitive and not class of groups which includes ℤn.
hard to show that h(P _ Q)  h(P) þ h(Q). Now
i
let Pn0 ¼ _n1
i¼0 T P and hn ¼ h P0 . The sub-
n

additivity of entropy implies that hnþm  hn þ Amenable Groups

hm, so by a well-known elementary lemma the
limit Let G be any countable group and T ¼ {Tg} an
action of G by automorphisms of a probability
hn h
hðP, T Þ ¼ lim ¼ inf n  0 ð28Þ space. Suppose s is a complex measure on G,
n n n n
that is, {s(g)}g  G is an absolutely summable
exists. sequence. Let Tg act on functions via Tg f ¼
If one thinks of P(x) as a measurement f ∘ Tg, so Tg is an isometry of Lp for every
performed on the space X and Tx as the state 1  p  1. Let s(T ) ¼ g  Gs(g)Tg. A very
succeeding x after one second has elapsed then general framework for formulating ergodic theo-
h(P, T) is the expected information per second rems is to consider a sequence {sn} and ask
obtained by repeating the experiment every sec- whether sn(T )f converges, a.e. or in mean, for
ond for a very long time. See (Rudolph 1990) for f in some Lp space. When sn(T )f converges for
an alternative and very useful approach to h(P, T ) all actions T and all f in Lp in p-norm or a.e. then
via name-counting. one says that sn is mean or pointwise good in Lp.
The following result, which is known as the When the sn are probability measures it is natural
Shannon–McMillan–Breiman theorem, has pro- to call such results weighted ergodic theorems and
ved to be of fundamental importance in ergodic this terminology is retained for complex s as well.
theory, notably, for example, in the proof of Birkhoff’s theorem says that if G ¼ ℤ and sn is
Ornstein’s celebrated isomorphism theorem for the normalized counting measure on {0, 1, . . .
Bernoulli shifts (Ornstein 1970). n  1} then {sn} is pointwise good in L1. This
section will be concerned only with sequences
Theorem 28 If T is ergodic then {sn} such that sn is normalized counting measure
on a finite subset Fn  G so one speaks of mean or
pointwise good sequences {Fn}. A natural condi-
1 tion to require of {Fn}, which will ensure that the
lim  0 ðxÞ ¼ hðP, T Þ
log m Pn1 ð29Þ
n!1 n limiting function is invariant, is that it be asymp-
totically (left) invariant in the sense that
a.e. and in L1-norm.
In other words, the actual information obtained j gF n D Fn j
! 0 8g  G: ð30Þ
per second by observing x over time converges to j Fn j
the constant h(P, T ), namely the limiting expected
information per second. Shannon (1948) formu- Such a sequence is called a Følner sequence
lated Theorem 28 and proved convergence in and a group G is amenable if it has a Følner
probability. McMillan (1953) proved L1 sequence. As in most of this article G is restricted
to be a discrete countable group for simplicity but
94 Ergodic Theorems

most of the results to be seen actually hold for a The work on ergodic theorems for abstract
general locally compact group. locally compact groups was pioneered by
Amenability of G is equivalent to the existence Calderón (1953) who built on Wiener’s methods.
of a finitely additive left invariant probability The main result in this paper is somewhat techni-
measure on G. It is not hard to see that any Abe- cal but it already contains the germ of
lian, and more generally any solvable, group is Tempelman’s theorem. Other ergodic theorems
amenable. On the other hand the free group F2 on for amenable groups, whose main interest lies in
two generators is not amenable. See Paterson the case of continuous groups, include
(1988) for more information on amenable groups. Tempel’man (1967), Renaud (1971), Greenleaf
The Følner property by itself is enough to give a (1973) and Greenleaf and Emerson (1974). The
mean ergodic theorem. discrete versions of these results are all rather
close to Tempelman’s theorem.
Theorem 29 Any Følner sequence is mean good Among pointwise theorems for discrete groups
in Lp for 1  p < 1. Tempelman’s result was essentially the best avail-
able for a long time. It was not known whether
The proof of this result is rather similar to the every amenable group had a Følner sequence
proof of Theorem 3. In fact Theorem 29 is only a which is pointwise good for some Lp. In 1988
special case of quite general results concerning Shulman (1988) introduced the notion of a tem-
amenable semi-groups acting on abstract pered Følner sequence {Fn}, namely one for which
Banachspaces. See the book of Paterson (1988)
for more on this. [ F1
i Fn < C j Fn j, ð31Þ
i<n
Turning to pointwise theorems, the Følner con-
dition alone does not yield a pointwise theorem,
even when G ¼ ℤ and the Fn are intervals. For for some constant C. The advantage of the tem-
example Akcoglu and del Junco (1975) have pered condition is that any Følner sequence has a
p tempered subsequence, and in particular any ame-
shown that when G ¼ ℤ and Fn ¼ ½n, n þ n \
ℤ the pointwise ergodic theorem fails for any nable group has a tempered Følner sequence.
aperiodic T and for some characteristic function f. Shulman proved a maximal inequality in L2 for
See also del Junco and Rosenblatt (1979). such Fn which implies that {Fn} is pointwise good
The following pointwise result of Tempelman in L2. An account of this work may be found in
(1972a) is often quoted. A Følner sequence {Fn} Section 5.6 of Tempelman’s book (1992).
is called regular if there is a constant C such that Lindenstrauss (1999) was able to extend the
j F1 result to L1.
n Fn j C j Fn j and there is an increasing
sequence F0n such that Fn  F0n and
0
j Fn j C j Fn j. Theorem 31 Any tempered Følner sequence is
pointwise good in L1.
Theorem 30 Any regular Følner sequence is
The key new idea in his proof is to use a
pointwise good in L1.
probabilistic argument to establish a covering
lemma. In the discrete case Ornstein and Weiss
In case the Fn are intervals in ℤn this result can
(2003) have given a non-probabilistic proof of
be proved by a variant of Wiener’s covering argu-
Lindenstrauss’s covering lemma. Lindenstrauss
ment and in the general case by an abstraction
also generalizes the a.e. convergence in the Shan-
thereof. The condition j F1n Fn j C j Fn j cap-
non–McMillan–Breiman theorem to this setting.
tures the property of rectangles which is needed
L1 convergence was already established by
for the covering argument. Emerson (1974) inde-
Kieffer (1975) in 1975.
pendently proved a very similar result.
Ergodic Theorems 95

Subsequence and Weighted Theorems Later Akcoglu et al. (1996). were able to show
that for lacunary sequences {sS,n} is even strongly
In this section G and T are as in the previous sweeping out. A sequence {sn} of probability mea-
section and sn is a sequence of complex measures sures on ℤ is said to be strongly sweeping out if for
on G. any ergodic T and for all δ > 0 there is a character-
This section will be concerned with conditions istic function f with fdm < δ such that lim sup
on sn that ensure that it is mean or pointwise sn(T)f ¼ 1 a.e. It is not difficult to show that if {sn}
good. For the most part G will be ℤ. is strongly sweeping out then there are characteris-
Hopf’s ergodic theorem gives a class of exam- tic functions f such that lim inf sn(T)f ¼ 0 and
ples for free. Choose any probability measure l on lim sup sn(T)f ¼ 1. Thus for lacunary sequences
G, define the operator Tl ¼ l(T) and observe that the ergodic theorem fails in the worst possible way.
ðT l Þn ¼ T l n , where l n denotes the convolution Bellow and Losert (1985) gave the first exam-
power. Since Tl satisfies the hypotheses of Hopf’s ple of a sequence S  ℤ of density 0 which is
theorem it follows that the sequence sn ¼ universally good for pointwise convergence,
1 n1 i
n i¼0 l is pointwise good in L1. answering a question posed by Furstenberg.
Another sort of example is given by choosing a They construct an S which is pointwise good in
sequence gn in G and letting sn ¼ 1n n1
i¼0 dgn : If one
L1. This paper also contains a good overview of
has convergence for such a sequence one speaks of a the progress on weighted and subsequence ergo-
subsequence ergodic theorem. This section will dic theorems at that time.
mainly focus on the case G ¼ ℤ and sequences Weyl’s theorem on uniform distribution
which are the increasing enumeration of a subset (Theorem 9) suggests the possibility of an ergodic
S  N. Given S  ℕ write sS,n for the corresponding theorem for the sequence {n2}. It is not hard to see
probabilities sn and say S is good if sS,n is good. that {n2} is mean good in L2. In fact the spectral
Perhaps the first subsequence ergodic theorem theorem and the dominated convergence theorem
is due to Blum and Hanson (1960) who proved that show that it is enough to prove that the L1-
n2
an automorphism T of a probability space is bounded sequence of functions 1n n1 i¼0 z on the
strongly mixing if and only if 1n n1 mi
i¼0 T f con-
unit circle converges at each point z of the unit
verges in L2 norm for every f  L2 and every circle. When z is not a root of unity the sequence
increasing {mi}. Strong mixing means that m- converges to 0 by Weyl’s result and when z is a
(T nA \ B) ! m(A)m(B). In 1969 Brunel and root of unity the convergenceis trivial because
2
Keane (1969) proved the first pointwise subse- zn is periodic. In 1987 Bourgain (1987,
quence ergodic theorem. 1988d) proved his celebrated pointwise ergodic
theorem for polynomial subsequences.
Theorem 32 Suppose that T is a translation on a
compact Abelian group G with Haarmeasure l, Theorem 33 If p is any polynomial with rational
g  G, and E  G is any Borel set with l(E) > 0 coefficients taking integer values on the integers
and l(@E) ¼ 0. Let S ¼ {i > 0 : T ig  E}. Then then S ¼ {p(n)} is pointwise good in L2.
S is pointwise good in L1.
The first step in Bourgain’s argument is to
Krengel (1971) constructed the first example of reduce the problem of proving a maximal inequal-
a sequence S  ℕ which is pointwise universally ity to the case of the shift map on the integers, via
bad, in the strong sense that for any aperiodic T the Calderón’s transfer principle. Then the problem is
a.e. convergence of sS,n(T)f fails for some charac- transferred to the circle by using Fourier trans-
teristic function f. Bellow (1983) proved that any forms. At this point the problem becomes a very
lacunary sequence (meaning anþ1 > can for some delicate question about exponential sums and a
c > 1)is pointwise universally bad in L1. whole arsenal of tools is brought to bear. See
96 Ergodic Theorems

Rosenblatt and Wierdl (1995) and Quas and and also Thouvenot (1995), Glasner (2003) and
Wierdl (Bergelson 2006) (Appendix B) for nice Rudolph’s book (1990). Rudolph’s result concerns
expositions of Bourgain’s methods. the convergence of multiple averages
Bourgain subsequently improved this to all Lp,
N1 k
p > 1 and also extended it to sequences {[q(n)]} 1
f j T nj x ð32Þ
where now q is an arbitrary real polynomial and [] N n¼0 j¼1
denotes the greatest integer function. He also
announced that his methods can be used to show where each Tj is an automorphism of a probability
that the sequencepof primes is pointwise good in space (Xj, ℬj, mj) and the fj are L1 functions. The
Lp for any p > 1þ2 3. Wierdl (1988) soon extended point is that the convergence occurs whenever
the result for primes to all p > 1. each xj  X0j , sets of measure one which may be
chosen sequentially for j ¼ 1, . . . , k without
Theorem 34 The primes are pointwise good in knowing what Ti or fi are for any i > j. He actually
Lp for p > 1. proves something stronger, namely he identifies
an intrinsic property of a sequence {ai}, which he
It has remained a major open question for quite calls fully generic, such that the following hold.
some time whether any of these results hold for
p ¼ 1. In 2005 there appeared a preprint of (a) The constant sequence {1} is fully generic.
Mauldin and Buczolich (2005), which remains (b) If {ai} is fully generic then for any T and
unpublished, showing that polynomial sequences f  L1 the sequence ai f(Tix) is fully generic
are L1-universally bad. for almost all x.
Another major result of Bourgain’s is the (c) Fully generic implies pointwise good in L1.
so-called return times theorem (Bourgain 1988e).
A simplification of Bourgain’s original proof was The definition of fully generic will not be
published jointly with Furstenberg, Katznelson and quoted here as it is somewhat technical.
Ornstein as an appendix to an article (Bourgain For a proof of the basic return times theorem
1989b) of Bourgain. To state it let us agree to say using joinings see Rudolph (1994). Assani,
that a a sequence of complex numbers {a(n)}n  0 Lesigneand and Rudolph (1995) took a first step
has property P if the sequence of complex mea- towards the multiple theorem, a Wiener–Wintner
sures sn ¼ 1n n1
i¼0 aðiÞdi has property P, where δi version of the return times theorem. Also Assani
denotes the point mass at i. (2000) independently gave a proof of Rudolph’s
result in the case when all the Tj are weakly mixing.
Theorem 35 (Bourgain) Suppose T is an auto- Ornstein and Weiss (1992) have proved the
morphism of a probability space (X, ℬ, m), 1  p, following version of the return times theorem
q  1 are conjugate exponents and f  Lp(m). for abstract discrete groups. As with ℤ, let us
Then for almost all x the sequence {f(T nx)} is say that a sequence {ag}g  G of complex num-
pointwise good in Lq. bers has property P for {Fn} if the sequence
sn ¼ jF1n j g  Fn aðgÞdg of complex measures
Applying this to characteristic functions f ¼ 1E has property P.
one sees that the return time sequence
{i > 0 : Six  E} is good for pointwise conver- Theorem 36 Suppose that the increasing Følner
gence in L1. Theorem 32 is a very special case. It sequence {Fn} satisfies the Tempelman condition
is also easy to see that Theorem 35 contains the supn F1
n Fn =jFn j < 1 and [Fn ¼ G. If b  L1
Wiener–Wintner theorem. then for a.a. x the sequence {b(Tgx)} is pointwise
In 1998 Rudolph (1998) proved a far-reaching good in L1 for {Fn}.
generalization of the return times theorem using the
technique of joinings. For an introduction to join- Recently Demeter, Lacey, Tao and Thiele
ings see the article by de la Rue in this collection (2008) have proved that the return times theorem
Ergodic Theorems 97

remains valid for any 1 < p  1 and q  2. On Then there is a constant D such that
the other hand Assani, Buczolich and Mauldin
D
(2005) showed that it fails for p ¼ q ¼ 1. mfMf > lg  k f k1 for all T, f  L1 ðmÞ
l
Bellow, Jones and Rosenblatt have a series of and l > 0:
papers (Bellow et al. 1989, 1990, 1992, 1994)
studying general weighted averages associated to
a sequence sn of probability measures on ℤ, and, in
Ergodic Theorems and Multiple
some cases, more general groups. The following
Recurrence
are a few of their results (Bellow et al. 1990). is
concerned with ℤ-actions and moving block aver-
Suppose S  ℕ. The upper density of S is
ages given by sn ¼ mIn , where the In are finite
intervals and mI denotes normalized counting mea- j S \ ½1, n j
dðSÞ ¼ lim sup : ð33Þ
sure on I. They resolve the problem completely, n n
obtaining a checkable necessary and sufficient con-
dition for such a sequence to be pointwise good in and the density d(S) is the limit of the same quan-
L1. tity, if it exists. In 1975 Szemerédi (1975) proved
Bellow et al. (1992) gives sufficient conditions the following celebrated theorem, answering an
on a sequence sn for it to be pointwise good in Lp, old question of Erdősand Turán.
p > 1, via properties of the Fourier transforms sn .
A particular consequence is that if Theorem 38 Any subset of ℕ with positive upper
limn!1k  ℤ j sn(k)  sn(k  1) j ¼ 0 then density contains an arithmetic progression of
{sn} has a subsequence which is pointwise good length k for each k  1.
in Lp, p > 1. In (Bellow et al. 1994) they obtain
convergence results for sequences sn ¼ sn, the This result has a distinctly ergodic-theoretic
convolution powers of a probability measure s. flavor. Letting T denote the shift map on ℤ, it
A consequence of one of their main results is that says that for each k there is an n such that S0 ¼
if the expectation k  ℤks(k) is zero, the second \ki¼1 T in S is non-empty. In fact the result gives
moment k  ℤk2s(k) is finite and s is aperiodic more: there is an n for which d ðS0 Þ > 0. In this
(its support is not contained in any proper coset in light Szemerédi’s theorem becomes a multiple
ℤ) then sn is pointwise good in Lp for p > 1. recurrence theorem for the shift map on ℕ,
Bellow and Calderón (1999) later showed that equipped with the invariant “measure-like” quan-
this last result is valid also for p ¼ 1. This is a tity d. Of course d is not even finitely additive so it
consequence of the following sufficient condition is not a measure. d, however, is at least finitely
For a sequence T to satisfy a weak L1 inequality. additive, when defined, and d(ℕ) ¼ 1.
Given an automorphism of a probability space This point of view suggests the following mul-
(X, ℬ, m) let Mf ¼ sup j sn(T )f(x)j be the max- tiple recurrence theorem.
imal operator associated to{sn}.
Theorem 39 Suppose T is an automorphism of a
Theorem 37 (Bellow and Calderón) Suppose probability space (X, ℬ, m), m(B) > 0 and k  1.
there is an α  (0, 1] and C > 0 such that for Then there is an n > 0 such that m \ki¼1 T in B > 0:
each n > 1 one has
In 1977, Furstenberg (1977) proved the follow-
ing ergodic theorem which implies the multiple
jyja
j sn ðx þ yÞ  sn ðxÞ j C for all x, y  ℤ recurrence theorem. He also established a general
jxj1þa correspondence principle which puts the shaky
such that 0 < 2 j y jj x j analogy between the multiple recurrence theorem
98 Ergodic Theorems

and Szemerédi’s theorem on a firm footing and so only their absolute versions (i. e. relative to the
allows each to be deduced from the other. Thus he trivial s-algebra {;, X}) will be described here.
obtained an ergodic theoretic proof of T is said to be compact if for each f  L2 the
Szemerédi’s combinatorial result. orbit {T if : i  ℤ} is pre-compact in the norm
topology of L2. This turns out to be equivalent to
Theorem 40 Suppose T is an automorphism of the statement that T is a translation on a compact
a probability space (X, ℬ, m), f  L1, f  0, Abelian group endowed with its Haar measure.
fdm > 0 and k  1. Then The property of weak mixing is a fundamental
notion in ergodic theory which has many equiva-
lent definitions. The most appropriate for our pur-
N1 k
1 poses is that T is weakly mixing if it has no
lim inf T in f dm > 0: ð34Þ
N N n¼0 i¼1
compact factors. This turns out to be equivalent
to the ergodicity of T T acting on the product
Furstenberg’s result opened the door to the measure space (X, ℬ, m) (X, ℬ, m).
study of so-called ergodic Ramsey theory The verification of 37 in the case of compact T is
which has yielded a vast array of deep results in rather easy. In this case it is not hard to prove that
combinatorics, many of which have no non- for any f  L1 and ϵ > 0 the set {n  ℤ : kTnf 
ergodic proof as yet. The focus of this article is fk2 < ϵ} has bounded gaps and (34) follows easily.
not on this direction but the reader is referred to In the case when T is weakly mixing (34) is a
Furstenberg’s book (1981) for an excellent intro- consequence of the following theorem which
duction and to Bergelson (1996, 2006) for sur- Furstenberg proves in (1977) (as a warm-up for
veys of later developments. There is also the it’s much harder relative version).
article by Frantzikinakis and McCutcheon in
this collection. Theorem 41 If T is weakly mixing and f1, f2, . . .,
Furstenberg’s proof relies on a deep structure fk are L1 functions then
theorem for a general automorphism which
was also developed independently by Zimmer N1 k k
1
(1976a, b) in a more general context. A factor lim T in f i dm ¼ fi ð35Þ
N N
of T is any sub-s-algebra F  ℬ such that n¼0 i¼1 i¼1

T(F ) ¼ F . (It is more accurate to think of the

factor as the action of T on the measure space Later Bergelson (1987) showed that the result
(X, F , m| F )). The structure theorem asserts that can be obtained easily by an induction argument
there is a transfinite increasing sequence of fac- using the following Hilbert space generalization
tors {F α} of T such that the following conditions of van der Corput’s lemma.
hold.
Theorem 42 (Bergelson) Suppose {xn} is a
(a) F αþ1 is compact relative to F α. bounded sequence of vectors in Hilbert
(b) F α ¼ _β<α F β whenever αis a limit ordinal. space such that for each h > 0 one has
1 N1
(c) ℬ is weakly mixing relative to _F α. N n¼0 hxnþh , xn i ! 0 as N ! 1. Then
1 N1
(_ denotes the s-algebra generated by a col- N n¼0 xn ! 0:
lection of s-algebras.) _α F α is called the maxi-
mal distal factor of T and if _F α ¼ ℬ then T is Ryzhikov has also given a beautiful short proof
called distal. The definitions of the relative prop- of Theorem 41 using joinings (Ryzhikov 1994).
erties in (a) and (c) above are somewhat technical Bergelson’s van der Corput lemma and variants of
Ergodic Theorems 99

it have been a key tool in subsequent develop- limiting behavior of the averages is unchanged
ments in ergodic Ramsey theory and in the con- when any one of the fi’s is replaced by its condi-
vergence results to be discussed in this section. tional expectation on C . This means that the ques-
Bergelson (1987) used it to prove the following tion of convergence of these averages may be
mean ergodic theorem for weakly mixing reduced to the case when fi are all C -measurable.
automorphisms. So the problem is to find the right (smallest)
characteristic factor and prove convergence for
Theorem 43 Suppose T is weakly mixing, f1, . . ., that factor.
fk are L1 functions and p1, . . ., pk are polynomials The importance of characteristic factors was
with rational coefficients taking integer values on already apparent in Furstenberg’s original paper
the integers such that no pi  pj is constant for (1977), where he showed that the maximal distal
i 6¼ j. Then factor is characteristic for the averages (37). In
fact he showed that for a given k a k-step distal
factor is characteristic. (An automorphism is
N1 k k
1 k-step distal if it is the top rung in a k-step ladder
lim T pi ðnÞ f i  f i dm
N!1 N n¼0 i¼1 i¼1 of factors as in the Furstenberg–Zimmer structure-
theorem.) It turns out, though, that the right char-
¼ 0: ð36Þ
acteristic factor for (37) is considerably smaller. In
their seminal paper (Conze and Lesigne 1988)
Theorems 40 and 41 immediately raise the
Conze and Lesigne identified the characteristic
question of convergence of the multiple averages
1 N1 k in factor for k ¼ 3, now called the Conze–Lesigne
N n¼0 i¼1 T f i for a general T. Several authors factor. As shown in (Host and Kra 2005b; Ziegler
obtained partial results on the question of mean 2007), the characteristic factor for a general k is
convergence. It was finally resolved only recently (isomorphic to) an inverse limit of k-step nilflows.
by Host and Kra (2005b), who proved the follow- A k-step nilflow is a compact homogeneous space
ing landmark theorem. N/Γ of a k-step nilpotent Lie group N, endowed
with its unique left-invariant probability measure,
Theorem 44 Suppose f1, f2, . . ., fk  L1. Then on which T acts via left translation by an element
there is a g  L1 such that of N. Ergodic properties of nilflows have been
studied for some time in ergodic theory, for exam-
N1 k ple in Parry (1969). In this way the problem of L2-
1
lim T in f i  g ¼ 0: ð37Þ convergence of (37) is reduced to the case when
N n¼0 i¼1 2 T is a nilflow. In this case one has more: the
averages converge pointwise by a result of
Independently and somewhat later Ziegler Leibman (2005a) (See also Ziegler (2005)).
(2007) obtained the same result by somewhat There have already been a good many gener-
different methods. Furstenberg had already alizations of (37). Host and Kra (2005a),
established Theorem 44 for k ¼ 2 in Frantzikinakis and Kra (2005b, 2006), and
(Furstenberg 1977). It was proved for k ¼ 3 in Leibman (2005a) have proved results which
the case of a totally ergodic T by Conze and replace linear powers of T by polynomial powers.
Lesigne (1988) and in general by Host and Kra In increasing degrees of generality Conze and
(2001). It can also be obtained using the methods Lesigne (1988), Frankzikinakis and Kra (2005a)
developed by Furstenberg and Weiss (1996). In and Tao (2008) have obtained results which
this paper Furstenberg and Weiss proved a result replace the maps T, T2, . . ., Tk in (37) by commut-
for polynomial powers of T in the case k ¼ 2. They ing maps T1, . . ., Tk. Bergelson and Leibman
also formalized the key notion of a characteristic (2002, 2004) have obtained results, both positive
factor. A factor C of T is said to be characteristic and negative, in the case of two non-
for the averages (37) if, roughly speaking, the L2 commuting maps.
100 Ergodic Theorems

In the direction of pointwise convergence the theorem in various ways. Bishop (1967) proved
only general result is the following theorem of the following result which is purely finite and
Bourgain (1990) which asserts pointwise conver- constructive in nature and evidently implies the
gence in the case k ¼ 2. a.e. convergence in Birkhoff’s theorem. If y ¼
(y1, . . ., yn) is a finite sequence of real numbers
Theorem 45 Suppose S and T are powers of a and a < b, an upcrossing of y over [a, b] is a
single automorphism R and f, g  L1. Then minimal integer interval [k, l]  [1, n] satisfy-
1 N
n¼1 f ðT xÞgðS xÞ converges a.e.
n n ing yk < a and yl > b.
N

When T is a K-automorphism Derrien and Theorem 47 Suppose T is an automorphism of

Lesigne (1996) have proved that the averages (35) the probability space (X, ℬ, m). Let U(n, a, b, f, x)
converge pointwise to the product of the integrals, be the number of upcrossings of the sequence
even with polynomial powers of T replacing the A0 f(x), . . . , An f(x) over [a, b]. Then for every n
linear powers.
Gowers (2001) has given a new proof of kf k1
Szemerédi’s theorem by purely finite methods mfx : U ðn, a, b, f , xÞ > kg  : ð38Þ
ðb  aÞk
using only harmonic analysis on ℤn. His results
give better quantitative estimates in the statement Ivanov (1996a) has obtained the following
of finite versions of Szemerédi’s theorem. Although stronger upcrossing inequality for an arbitrary
his proof contains no ergodic theory it is to some positive measurable f, which also implies
extent guided by Furstenberg’s approach. Birkhoff’s theorem.
This section would be incomplete without
mentioning the spectacular recent result of Green Theorem 48 For any positive measurable f and
and Tao (2004) on primes in arithmetic progres- 0<a<b
sion and the subsequent extensions of the Green-
Tao theorem due to Tao (2005) and Tao and
Ziegler (2006). a k
mfx : U ðn, a, b, f , xÞ > kg  : ð39Þ
b

Note the exponential decay and the remarkable

Rates of Convergence
fact that the estimate does not depend on f. Ivanov
has also obtained the following result (Theorem 23
There are many results which say in various ways
in (Kachurovskii 1996a)) about fluctuations of An f.
that, in general, there are no estimates for the rate
An ϵ-fluctuation of a real sequence y ¼ (y1. . .yn) is
of convergence of the averages An f in Birkhoff’s
a minimal integer interval [k, l] satisfying jyk 
theorem. For example there is the following result
yl j  ϵ. If f  L1(ℝ) let F(ϵ, f, x) be the number
of Krengel (1978).
of ϵ-fluctuations of the sequencefAn f ðxÞg1 n¼0 .

Theorem 46 Suppose lim n!1 cn ¼ 1 and T is

Theorem 49
any ergodic automorphism of a probability space. 1
ðlog kÞ2
Then there is a bounded measurable f with fm ¼ mfx : Fðϵ, f , xÞ  kg  C ð40Þ
k
0 such that lim sup cnAn f ¼ 1 a.e.

See Part 1 of Derriennic (2006) for a selec- where C is a constant depending only on k f k1/
tion of other results in this direction. In spite of ϵ. If f  L1 then
these negative results one can obtain quantita-
tive estimates by reformulating the ergodic mfx : Fðϵ, f , xÞ  kg  AeBk , ð41Þ
Ergodic Theorems 101

where A and B are constants depending only Future Directions

on k f k1/ϵ.
See Kachurovskii’s survey (1996a) of results This section will be devoted to a few open prob-
on rates of convergence in the ergodic theorem lems and some questions. Many of these have
and also the article of Jones et al. (1998) for more been suggested by my colleagues acknowledged
results on oscillation type inequalities. in the introduction.
Under special assumptions on f and T it is The topic of convergence of Furstenberg’s
possible to give more precise results on speed of multiple averages has seen some remarkable
convergence. If the sequence T if is independent achievements in the last few years and will likely
then there is a vast literature in probability theory continue to be vital for some time to come. The
giving very precise results, for example the central question of pointwise convergence of multiple
limit theorem and the law of the iterated logarithm averages is completely open, beyond Bourgain’s
(see, for example, (Breiman 1968)). See the sur- result (Theorem 45). Even extending Bourgain’s
veys of Derrienic (2006) and of Merlevède, result to three different powers of R or to any two
Peligrad and Utev (2006) for results on the central commuting automorphisms S and T would be a
limit theorem for dynamical systems. very significant achievement. Another natural
question is whether one has pointwise conver-
gence of the averages of the sequence
2
Ergodic Theorems for Non-amenable f ðT n ðxÞÞg T n ðxÞ , which are mean convergent
Groups
by the result of Furstenberg and Weiss (1996)
(which is now subsumed in much more general
Guivarc’h (1969) was the first to prove an ergodic
results).
theorem for a general pair of non-commuting uni-
A long-standing open problem relating to
tary operators. Much work has been done in the
Akcoglu’s ergodic theorem (Theorem 23) for pos-
last 15 years on mean and pointwise theorems for
itive contractions of Lp is whether it can be
non-amenable groups. See the extensive survey
extended to power-bounded operators T (this
by Nevo (2006). Here is just one result of Nevo
means that kTkn is bounded). It is also an open
(1994) as a sample. Suppose G is a discrete group
question whether it extends to non-positive con-
and {sn} is a sequence of probability measures on
tractions, excepting the case p ¼ 2 where
G. Say that {sn} is mean ergodic if for any unitary
Burkholder (1962) has shown that it fails.
representation π of G on a Hilbert space ℋ and
There are many natural questions in the area of
any x  ℋ one has g  Gsn(g)π(g)x converges
subsequence and weighted ergodic theorems. For
in norm to the projection of x on the subspace of
example, which of Bourgain’s several pointwise
π-invariant vectors. Say that {sn} is pointwise
theorems can be extended to L1? Are there other
ergodic if for any measure preserving action T of
natural subsequences of an arithmetic character
G on a probability space and f  L2 one has
which have density 0 and for which an ergodic
g  Gsn(g)Tg f converges a.e. to the projection
theorem is valid, either mean or pointwise, and in
of f on the subspace of T-invariant functions.
what Lp spaces might such theorems be valid? Are
there higher dimensional analogues?
Theorem 50 Let G be the free group on
Since lacunary sequences are bad, to have any
k generators, k  1. Let tn be the normalized
sort of pointwise theorem ln ¼ log aanþ1 n
must get
counting measure on the set of elements whose
word length (in terms of the generators and their close to 0 and for simplicity let us assume that lim
inverses) is n. Let sn ¼ (tn þ tnþ1)/2 and ln ¼ 0. How fast must the convergence be in order
s0n ¼ 1n ni¼1 si . Then sn and s0n are mean and to get an ergodic theorem? Jones and Wierdl
1
(1994) have shown that if ln > ðlog nÞ2þϵ then
pointwise ergodic but tn is not mean ergodic.
102 Ergodic Theorems

the pointwise ergodic theorem fails inL2 while natural to ask for slower rates of growth. In par-
Jones, Lacey and Wierdl (1999) have shown that ticular, in any amenable group is there always a
an only slightly faster rate permits a sequence sequence {Fn} which is pointwise good and
which is pointwise good in L2. How well or grows at most exponentially? Can one do better
badly does the ergodic theorem succeed or fail either in general or in particular groups?
depending on the rate of convergence of ln? In Lindenstrauss’s theorem at least guarantees the
particular is there a (slow) rate which still guaran- existence of Følner sequences which are
tees strong sweeping out? (Jones et al. 1999) pointwise good in L1 but in particular groups
contains some interesting conjectures in this there are often natural sequences which one
1
direction. hopes might be good. For example in i  ℤℤ
There are also interesting questions concerning one may take Fn to be a cube based at 0 of side
the mean and pointwise ergodic theorems for sub- length ln and dimension dn (that is, all but the first
sequences which are chosen randomly in some dn co-ordinates are zero), where both sequences
sense. See Bourgain (1988c; Jones et al. 1999) increase to 1. What conditions on ln and dn will
for some results in this direction. Again (Jones give a good sequence? Note that no such sequence
et al. 1999) contains some interesting conjectures is regular in Tempelman’s sense. If dn ¼ n then
along these lines. {ln} must be super exponential to ensure
In a recent paper (Bergelson and Leibman Shulman’s condition. Can one do better? What
2007) Bergelson and Leibman prove some very about ln ¼ dn ¼ n?
interesting and surprising results about the distri-
bution of generalized polynomials. A generalized
polynomial is any function which can be built Bibliography
starting with polynomials in ℝ[x] using the oper-
ations of addition, multiplication and taking the Akcoglu MA (1975) A pointwise ergodic theorem in Lp-
spaces. Can J Math 27(5):1075–1082
greatest integer. As a consequence they derive a
Akcoglu MA, Chacon RV (1965) A convexity theorem for
generalization of von Neumann’s mean ergodic positive operators. Z Wahrsch Verw Gebiete 3:328–332
theorem to averages along generalized polyno- Akcoglu MA, Chacon RV (1970) A local ratio theorem.
mial sequences. The following is a special case. Can J Math 22:545–552
Akcoglu MA, del Junco A (1975) Convergence of aver-
ages of point transformations. Proc Amer Math Soc 49:
Theorem 51 Suppose p is a generalized polyno- 265–266
mial taking integer values on the integers and U is Akcoglu MA, Kopp PE (1977) Construction of dilations of
pðiÞ
a unitary operator on ℋ. Then 1n n1 i¼0 U x is positive Lp-contractions. Math Z 155(2):119–127
Akcoglu MA, Krengel U (1981) Ergodic theorems for
norm convergent for all x  ℋ. superadditive processes. J Reine Angew Math 323:
53–67
This begs the question: does one have Akcoglu MA, Sucheston L (1978) A ratio ergodic theorem
for superadditive processes. Z Wahrsch Verw Gebiete
pointwise convergence? If so, this would be a
44(4):269–278
far-reaching generalization of Bourgain’s polyno- Akcoglu M, Bellow A, Jones RL, Losert V, Reinhold-
mial ergodic theorem. Larsson K, Wierdl M (1996) The strong sweeping out
There are also lots of questions concerning the property for lacunary sequences, Riemann sums, con-
volution powers, and related matters. Ergod Theory
nature of Følner sequences {Fn} in an amenable
Dyn Sys 16(2):207–253
group which give a pointwise theorem. For exam- Akcoglu M, Jones RL, Rosenblatt JM (2000) The worst
ple Lindenstrauss (1999) has shown that in the sums in ergodic theory. Michigan Math J 47(2):
lamplighter group, a semi-direct product of ℤ 265–285
Alaoglu L, Birkhoff G (1939) General ergodic theorems.
with i  ℤℤ/2ℤ on which Z acts by the shift,
Proc Nat Acad Sci USA25:628–630
there is no sequence satisfying Tempelman’s con- Assani I (2000) Multiple return times theorems for weakly
dition and that any {Fn} satisfying the Shulman mixing systems. Ann Inst H Poincaré Probab Statist
condition must grow super-exponentially. So, it is 36(2):153–165
Ergodic Theorems 103

Assani I (2003) Wiener–Wintner ergodic theorems. World Bergelson V, Leibman A (2002) A nilpotent Roth theorem.
Scientific Publishing Co. Inc, River Edge Invent Math 147(2):429–470
Assani I, Lesigne E, Rudolph D (1995) Wiener–Wintner Bergelson V, Leibman A (2004) Failure of the Roth theo-
return-times ergodic theorem. Israel J Math rem for solvable groups of exponential growth. Ergod
92(1–3):375–395 Theory Dyn Sys 24(1):45–53
Assani I, Buczolich Z, Mauldin RD (2005) An L1 counting Bergelson V, Leibman A (2007) Distribution of values of
problem in ergodic theory. J Anal Math 95:221–241 bounded generalized polynomials. Acta Math 198(2):
Auslander L, Green L, Hahn F (1963) Flows on homoge- 155–230
neous spaces. With the assistance of Markus L, Massey Berkson E, Bourgain J, Gillespie TA (1991) On the almost
W and an appendix by Greenberg L Annals of Mathe- everywhere convergence of ergodic averages for
matics Studies, No 53. Princeton University Press, power-bounded operators on L p-subspaces. Integral
Princeton, NJ Equ Oper Theory 14(5):678–715
Banach S (1993) Théorie des opérations linéaires. Éditions Billingsley P (1965) Ergodic theory and information.
Jacques Gabay, Sceaux, reprint of the 1932 original Wiley, New York
Bellow A (1983) On “bad universal” sequences in ergodic Birkhoff GD (1931) Proof of the ergodic theorem. Proc
theory II. In: Belley JM, Dubois J, Morales P (eds) Natl Acad Sci U S A 17:656–660
Measure theory and its applications. Lecture notes in Bishop E (1966) An upcrossing inequality with applica-
math, vol 1033. Springer, Berlin, pp 74–78 tions. Michigan Math J 13:1–13
Bellow A (1999) Transference principles in ergodic theory. Bishop E (1967/1968) A constructive ergodic theorem.
In: Christ M, Kenig CE, Sadowsky C (eds) Harmonic J Math Mech 17:631–639
analysis and partial differential equations, Chicago lec- Blum JR, Hanson DL (1960) On the mean ergodic theorem
tures in math. University of Chicago Press, Chicago, for subsequences. Bull Amer Math Soc 66:308–311
pp 27–39 Bourgain J (1987) On pointwise ergodic theorems for
Bellow A, Calderón A (1999) A weak-type inequality for arithmetic sets. C R Acad Sci 305(10):397–402
convolution products. In: Christ M, Kenig CE, Bourgain J (1988a) Almost sure convergence and bounded
Sadowsky C (eds) Harmonic analysis and partial dif- entropy. Israel J Math 63(1):79–97
ferential equations, Chicago lectures in math. Univer- Bourgain J (1988b) An approach to pointwise ergodic
sity of Chicago Press, Chicago, pp 41–48 theorems. In: Lindenstrauss J, Milman VD (eds) Geo-
Bellow A, Jones R (eds) (1991) Almost everywhere con- metric aspects of functional analysis (1986/87), Lec-
vergence, II. Academic Press, Boston ture notes in math, vol 1317. Springer, Berlin,
Bellow A, Losert V (1985) The weighted pointwise ergo- pp 204–223
dic theorem and the individual ergodic theorem along Bourgain J (1988c) On the maximal ergodic theorem for
subsequences. Trans Am Math Soc 288(1):307–345 certain subsets of the integers. Israel J Undergrad Math
Bellow A, Jones R, Rosenblatt J (1989) Almost every- 61(1):39–72
where convergence of powers. In: Edgar GA, Bourgain J (1988d) On the pointwise ergodic theorem on
Sucheston L (eds) Almost everywhere convergence. L p for arithmetic sets. Israel J Math 61(1):73–84
Academic, Boston, pp 99–120 Bourgain J (1988e) Temps de retour pour les systèmes
Bellow A, Jones R, Rosenblatt J (1990) Convergence for dynamiques. C R Acad Sci 306(12):483–485
moving averages. Ergod Theory Dyn Sys 10(1):43–62 Bourgain J (1989a) Almost sure convergence in ergodic
Bellow A, Jones RL, Rosenblatt J (1992) Almost every- theory. In: Edgar GA, Sucheston L (eds) Almost every-
where convergence of weighted averages. Math Ann where convergence. Academic, Boston, pp 145–151
293(3):399–426 Bourgain J (1989b) Pointwise ergodic theorems for arith-
Bellow A, Jones R, Rosenblatt J (1994) Almost every- metic sets. Inst Hautes Études Sci Publ Math (69):5–45,
where convergence of convolution powers. Ergod The- with an appendix by the author, Furstenberg H,
ory Dyn Sys 14(3):415–432 Katznelson Y, Ornstein DS
Bergelson V (1987) Weakly mixing PET. Ergod Theory Bourgain J (1990) Double recurrence and almost sure
Dyn Sys 7(3):337–349 convergence. J Reine Angew Math 404:140–161
Bergelson V (1996) Ergodic Ramsey theory—an update. Breiman L (1957) The individual ergodic theorem of infor-
In: Pollicott M, Schmidt K (eds) Ergodic theory of Zd mation theory. Ann Math Statist 28:809–811
actions. London Math Soc Lecture Note Ser, vol 228. Breiman L (1968) Probability. Addison-Wesley, Reading
Cambridge University Press, Cambridge, pp 1–61 Brunel A, Keane M (1969) Ergodic theorems for operator
Bergelson V (2006) Combinatorial and Diophantine appli- sequences. Z Wahrsch Verw Gebiete 12:231–240
cations of ergodic theory. In: Hasselblatt B, Katok Burkholder DL (1962) Semi–Gaussian subspaces. Trans
A (eds) Handbook of dynamical systems, vol 1B, Am Math Soc 104:123–131
Appendix A by Leibman A, Appendix B by Quas A, Calderon AP (1953) A general ergodic theorem. Ann Math
Wierdl M. Elsevier, Amsterdam, pp 745–869 2(58):182–191
Bergelson V (2007) Some historical remarks and modern Calderón AP (1968) Ergodic theory and translation-
questions around the ergodic theorem. Int Math Nachr invariant operators. Proc Natl Acad Sci U S A 59:
205:1–10 349–353
104 Ergodic Theorems

Calderon AP, Zygmund A (1952) On the existence of Eberlein WF (1949) Abstract ergodic theorems and weak
certain singular integrals. Acta Math 88:85–139 almost periodic functions. Trans Am Math Soc 67:
Chacon RV (1962) Identification of the limit of operator 217–240
averages. J Math Mech 11:961–968 Edgar G, Sucheston L (eds) (1989) Almost everywhere
Chacon RV (1963a) Convergence of operator averages. In: convergence. Academic Press, Boston
Wright FB (ed) Ergodic theory. Academic, New York, Emerson WR (1974) The pointwise ergodic theorem for
pp 89–120 amenable groups. Am J Math 96:472–487
Chacon RV (1963b) Linear operators in L1. In: Wright FB Feldman J (2007) A ratio ergodic theorem for commuting,
(ed) Ergodic theory. Academic, New York, pp 75–87 conservative, invertible transformations with quasi-
Chacon RV (1964) A class of linear transformations. Proc invariant measure summed over symmetric hyper-
Am Math Soc 15:560–564 cubes. Ergod Theory Dyn Sys 27(4):1135–1142
Chacon RV, Ornstein DS (1960) A general ergodic theo- Foguel SR (1969) The ergodic theory of Markov pro-
rem. Ill J Math 4:153–160 cesses. In: Van Nostrand Mathematical Studies, No
Conze JP, Lesigne E (1984) Théorèmes ergodiques pour 21. Van Nostrand Reinhold, New York
des mesures diagonales. Bull Soc Math France 112(2): Frantzikinakis N, Kra B (2005a) Convergence of multiple
143–175 ergodic averages for some commuting transformations.
Conze JP, Lesigne E (1988) Sur un théorème ergodique Ergod Theory Dyn Sys 25(3):799–809
pour des mesures diagonales. In: Probabilités, Publ Inst Frantzikinakis N, Kra B (2005b) Polynomial averages
Rech Math Rennes, vol 1987. University Rennes I, converge to the product of integrals. Israel J Math
Rennes, pp 1–31 148:267–276
Cotlar M (1955a) On ergodic theorems. Math Notae 14: Frantzikinakis N, Kra B (2006) Ergodic averages for inde-
85–119. (1956) pendent polynomials and applications. J London Math
Cotlar M (1955b) A unified theory of Hilbert transforms and Soc 2) 74(1:131–142
ergodic theorems. Rev Mat Cuyana 1:105–167. (1956) Furstenberg H (1967) Disjointness in ergodic theory, min-
Day M (1942) Ergodic theorems for Abelian semigroups. imal sets, and a problem in Diophantine approximation.
Trans Am Math Soc 51:399–412 Math Sys Theory 1:1–49
del Junco A, Rosenblatt J (1979) Counterexamples in Furstenberg H (1977) Ergodic behavior of diagonal mea-
ergodic theory and number theory. Math Ann 245(3): sures and a theorem of Szemerédion arithmetic pro-
185–197 gressions. J Analyse Math 31:204–256
Demeter C, Lacey M, Tao T, Thiele C (2008) Breaking the Furstenberg H (1981) Recurrence in ergodic theory and
duality in the return times theorem. Duke Math combinatorial number theory. Princeton University
J 143(2):281–355 Press, Princeton, m B Porter Lectures
Derrien JM, Lesigne E (1996) Un théorème ergodique Furstenberg H, Kesten H (1960) Products of random matri-
polynomial ponctuel pour lesendomorphismes exacts ces. Ann Math Statist 31:457–469
et les K-systèmes. Ann Inst H Poincaré Probab Statist Furstenberg H, Weiss B (1996) A mean ergodic theorem
32(6):765–778 for (1/N)Nn¼1 f(Tnx)g(Tn2 x). In: Bergelson V, March-
Derriennic Y (2006) Some aspects of recent works on limit P, Rosenblatt J (eds) Convergence in ergodic theory and
theorems in ergodic theory with special emphasis on probability, Ohio State Univ Math Res Inst Publ,
the “central limit theorem”. Discrete Contin Dyn Syst vol 5. de Gruyter, Berlin, pp 193–227
15(1):143–158 Garsia AM (1970) Topics in almost everywhere conver-
Doob JL (1938) Stochastic processes with an integral- gence. In: Lectures in advanced mathematics,
valued parameter. Trans Amer MathSoc 44(1):87–150 vol 4. Markham, Chicago
Dunford N (1951) An individual ergodic theorem for non- Glasner E (2003) Ergodic theory via joinings, Mathemat-
commutative transformations. Acta Sci Math Szeged ical surveys and monographs, vol 101. American Math-
14:1–4 ematical Society, Providence
Dunford N, Schwartz J (1955) Convergence almost every- Gowers WT (2001) A new proof of Szemerédi’s theorem.
where of operator averages. Proc Natl Acad Sci U S Geom Funct Anal 11(3):465–588
A 41:229–231 Green B, Tao T (2004) The primes contain arbitrarily large
Dunford N, Schwartz JT (1988) Linear operators, part arithmetic progressions. http://arxivorg/abs/mathNT/
I. Wiley Classics Library, Wiley, New York. General 0404188
theory, with the assistance of Bade WG, Bartle RG, Greenleaf FP (1973) Ergodic theorems and the construc-
Reprint of the 1958 original, A Wiley–Interscience tion of summing sequences in amenable locally com-
Publication pact groups. Commun Pure Appl Math 26:29–46
Durand S, Schneider D (2003) Random ergodic theorems Greenleaf FP, Emerson WR (1974) Group structure and the
and regularizing random weights. Ergod Theory Dyn pointwise ergodic theorem for connected amenable
Sys 23(4):1059–1092 groups. Adv Math 14:153–172
Ergodic Theorems 105

Guivarc'h Y (1969) Généralisation d’un théorème de JonesRL RJM, Wierdl M (2003) Oscillation in ergodic
von Neumann. C R Acad Sci Paris Sér A-B 268: theory: higher dimensional results. Israel J Math 135:
A1020–A1023 1–27
Halmos PR (1946) An ergodic theorem. Proc Natl Acad Kac M (1947) On the notion of recurrence in discrete
Sci U S A 32:156–161 stochastic processes. Bull Amer Math Soc 53:
Halmos PR (1949) A nonhomogeneous ergodic theorem. 1002–1010
Trans Am Math Soc 66:284–288 Kachurovskii AG (1996a) Rates of convergence in ergodic
Halmos PR (1960) Lectures on ergodic theory. Chelsea, theorems. Uspekhi Mat Nauk 51(4(310)):73–124
New York Kachurovskii AG (1996b) Spectral measures and conver-
Hammersley JM, Welsh DJA (1965) First-passage perco- gence rates in the ergodic theorem. Dokl Akad Nauk
lation, subadditive processes, stochastic networks, and 347(5):593–596
generalized renewal theory. In: Proc Internat Res Kakutani S (1940) Ergodic theorems and the Markoff
Semin. Statist Lab, University of California, Berkeley. process with a stable distribution. Proc Imp Acad
Springer, New York, pp 61–110 Tokyo 16:49–54
Hopf E (1937) Ergodentheorie. In: Ergebnisse der Kalikow S, Weiss B (1999) Fluctuations of ergodic aver-
Mathematik und ihrer Grenzgebiete, vol 5. Springer, ages. In: Proceedings of the conference on
Berlin probability, ergodic theory, and analysis,
Hopf E (1954) The general temporally discrete Markoff vol 43, pp 480–488
process. J Ration Mech Anal 3:13–45 Kamae T (1982) A simple proof of the ergodic theorem
Host B, Kra B (2001) Convergence of Conze–Lesigne using nonstandard analysis. Israel J Math 42(4):
averages. Ergod Theory Dyn Sys 21(2):493–509 284–290
Host B, Kra B (2005a) Convergence of polynomial ergodic Katok A, Hasselblatt B (1995) Introduction to the modern
averages. Israel J Math 149:1–19 theory of dynamical systems, Encyclopedia of Mathe-
Host B, Kra B (2005b) Nonconventional ergodic averages matics and its Applications, vol 54. Cambridge Univer-
and nilmanifolds. Ann Math 161(1):397–488 sity Press, Cambridge, with a supplementary chapter by
Hurewicz W (1944) Ergodic theorem without invariant Katok A and Mendoza L
measure. Ann Math 2(45):192–206 Katznelson Y, Weiss B (1982) A simple proof of some
Ivanov VV (1996a) Geometric properties of monotone ergodic theorems. Israel J Math 42(4):291–296
functions and the probabilities of random oscillations. Kieffer JC (1975) A generalized Shannon–McMillan the-
Sibirsk Mat Zh 37(1):117–115 orem for the action of an amenable group on a proba-
Ivanov VV (1996b) Oscillations of averages in the ergodic bility space. Ann Probab 3(6):1031–1037
theorem. Dokl Akad Nauk 347(6):736–738 Kingman JFC (1968) The ergodic theory of subadditive
Jewett RI (1969/1970) The prevalence of uniquely ergodic stochastic processes. J Roy Stat Soc Ser B 30:499–510
systems. J Math Mech 19:717–729 Kingman JFC (1976) Subadditive processes. In: École
Jones RL (1987) Necessary and sufficient conditions for a d’Été de Probabilités deSaint–Flour, V–1975, Lecture
maximal ergodic theorem along subsequences. Ergod notes in math, vol 539. Springer, Berlin, pp 167–223
Theory Dyn Sys 7(2):203–210 Koopman B (1931) Hamiltonian systems and transforma-
Jones RL (1993) Ergodic averages on spheres. J Anal Math tions in Hilbert spaces. Proc Natl Acad Sci U S A 17:
61:29–45 315–318
Jones RL, Wierdl M (1994) Convergence and divergence Krantz SG, Parks HR (1999) The geometry of domains in
of ergodic averages. Ergod Theory Dyn Sys 14(3): space. Birkhäuser Advanced Texts: Basler Lehrbücher.
515–535 Birkhäuser Boston Inc, Boston
Jones RL, Olsen J, Wierdl M (1992) Subsequence ergodic Krengel U (1971) On the individual ergodic theorem for
theorems for L p contractions. Trans Am Math Soc subsequences. Ann Math Stat 42:1091–1095
331(2):837–850 Krengel U (1978/79) On the speed of convergence in the
Jones R, Rosenblatt J, Tempelman A (1994) Ergodic the- ergodic theorem. Monatsh Math 86(1):3–6
orems for convolutions of a measure on a group. Ill Krengel U (1985) Ergodic theorems. In: de Gruyter studies
J Math 38(4):521–553 in mathematics, vol 6. de Gruyter, Berlin. with a sup-
Jones RL, Kaufman R, Rosenblatt JM, Wierdl M (1998) plement by Antoine Brunel
Oscillation in ergodic theory. Ergod Theory Dyn Sys Krengel U, Lin M, Wittmann R (1990) A limit theorem for
18(4):889–935 order preserving nonexpansive operators in L1. Israel
Jones RL, Lacey M, Wierdl M (1999) Integer sequences J Undergrad Math 71(2):181–191
with big gaps and the pointwise ergodic theorem. Ergod Krieger W (1972) On unique ergodicity. In: Proceedings
Theory Dyn Sys 19(5):1295–1308 of the Sixth Berkeley Symposium on Mathematical
Jones RL, Rosenblatt JM, Wierdl M (2001) Oscillation Statistics and Probability, vol II. Probability
inequalities for rectangles. Proc Am Math Soc 129(5): Theory, University of California Press, Berkeley,
1349–1358. (electronic) pp. 327–346
106 Ergodic Theorems

Kryloff N, Bogoliouboff N (1937) La théorie générale de la Reinhold mathematical studies, no 34. Van Nostrand
mesure dans son application à l’étude des systèmes Reinhold Co., London
dynamiques de la mécanique non linéaire. Ann Math Ornstein DS (1960) On invariant measures. Bull Amer
38(1):65–113 Math Soc 66:297–300
Kuipers L, Niederreiter H (1974) Uniform distribution of Ornstein D (1970) Bernoulli shifts with the same entropy
sequences. Wiley, New York, Pure and Applied are isomorphic. Adv Math 4:337–352
Mathematics Ornstein D (1971) A remark on the Birkhoff ergodic the-
Lamperti J (1958) On the isometries of certain function- orem. Ill J Math 15:77–79
spaces. Pac J Math 8:459–466 Ornstein D, Weiss B (1983) The Shannon–McMillan–
Leibman A (2005a) Convergence of multiple ergodic aver- Breiman theorem for a class of amenable groups. Israel
ages along polynomials of several variables. Israel J Math 44(1):53–60
J Math 146:303–315 Ornstein D, Weiss B (1992) Subsequence ergodic theorems
Leibman A (2005b) Pointwise convergence of ergodic for amenable groups. Israel J Math 79(1):113–127
averages for polynomial actions of ℤd by translations Oseledec VI (1968) A multiplicative ergodic theorem.
on a nilmanifold. Ergod Theory Dyn Sys 25(1): Characteristic Ljapunov, exponents of dynamical sys-
215–225 tems. Trudy Moskov Mat Obšč 19:179–210
Lemańczyk M, Lesigne E, Parreau F, Volný D, Wierdl Oxtoby JC (1952) Ergodic sets. Bull Amer Math Soc 58:
M. n.d. 116–136
Lesigne E (1989) Théorèmes ergodiques pour une transla- Parry W (1969) Ergodic properties of affine transformations
tion sur un nilvariété. Ergod Theory Dyn Sys 9(1): and flows on nilmanifolds. Am J Math 91:757–771
115–126 Paterson A (1988) Amenability, mathematical surveys and
Lindenstrauss E (1999) Pointwise theorems for amenable monographs, vol 29. American Mathematical Society,
groups. Electron Res Announc Amer Math Soc 5: Providence
82–90 Peck JEL (1951) An ergodic theorem for a non-
Loomis LH (1946) A note on the Hilbert transform. Bull commutative semigroup of linear operators. Proc Am
Amer Math Soc 52:1082–1086 Math Soc 2:414–421
Lorch ER (1939) Means of iterated transformations in Pesin JB (1977) Characteristic Ljapunov exponents, and
reflexive vector spaces. Bull Amer Math Soc 45: smooth ergodic theory. Uspehi Mat Nauk
945–947 32(4(196)):55–112,287
Mauldin D, Buczolich Z (2005) Concepts behind divergent Petersen K (1983) Another proof of the existence of the
ergodic averages along the squares. In: Assani ergodic Hilbert transform. Proc Am Math Soc 88(1):
I (ed) Ergodic theory and related fields, Contemp 39–43
math, vol 430. American Mathematical Society, Prov- Petersen K (1989) Ergodic theory, Cambridge studies in
idence, pp 41–56 advanced mathematics, vol 2. Cambridge University
McMillan B (1953) The basic theorems of information Press, Cambridge
theory. Ann Math Stat 24:196–219 Pitt HR (1942) Some generalizations of the ergodic theo-
Merlevède F, Peligrad M, Utev S (2006) Recent advances rem. Proc Camb Philos Soc 38:325–343
in invariance principles for stationary sequences. Pro- Poincaré H (1987) Les méthodes nouvelles de la
bab Surv 3:1–36 mécanique céleste. Tome I, II, III. Les Grands
Neveu J (1961) Sur le théorème ergodique ponctuel. C R Classiques Gauthier–Villars. Librairie Scientifique et
Acad Sci Paris 252:1554–1556 Technique Albert Blanchard, Paris
Neveu J (1965) Mathematical foundations of the calculus Raghunathan MS (1979) A proof of Oseledec’s multipli-
of probability. Trans. Amiel Feinstein. Holden-Day, cative ergodic theorem. Israel J Math 32(4):356–362
San Francisco Renaud PF (1971) General ergodic theorems for locally
Nevo A (1994) Harmonic analysis and pointwise ergodic compact groups. Am J Math 93:52–64
theorems for noncommuting transformations. J Am Rosenblatt JM, Wierdl M (1995) Pointwise ergodic theo-
Math Soc 7(4):875–902 rems via harmonic analysis. In: Peterson KE, Salama
Nevo A (2006) Pointwise ergodic theorems for actions of IA (eds) Ergodic theory and its connections with har-
groups. In: Hasselblatt B, Katok A (eds) Handbook of monic analysis, London Mathematical Society lecture
dynamical systems, vol 1B. Elsevier, Amsterdam, note series, vol 205. Cambridge University Press, Cam-
pp 871–982 bridge, pp 3–151
Nevo A, Stein EM (1994) A generalization of Rudolph DJ (1990) Fundamentals of measurable dynam-
Birkhoff’s pointwise ergodic theorem. Acta Math ics. In: Fundamentals of measurable dynamics: ergodic
173(1):135–154 theory on Lebesgue spaces. Oxford Science Publica-
Nguyen XX (1979) Ergodic theorems for subadditive spa- tions, The Clarendon Press, Oxford University Press,
tial processes. Z Wahrsch Verw Gebiete 48(2):159–176 New York
Orey S (1971) Lecture notes on limit theorems for Markov Rudolph DJ (1994) A joinings proof of Bourgain’s return
chain transition probabilities. In: Van Nostrand time theorem. Ergod Theory Dyn Sys 14(1):197–203
Ergodic Theorems 107

Rudolph DJ (1998) Fully generic sequences and a Tempel'man AA (1972b) A generalization of a certain
multiple-term return-times theorem. Invent Math ergodic theorem of Hopf. Teor Verojatnost i Primenen
131(1):199–228 17:380–383
Ruelle D (1982) Characteristic exponents and invariant Thouvenot JP (1995) Some properties and applications of
manifolds in Hilbert space. Ann Math 115(2):243–290 joinings in ergodic theory. In: Peterson KE, Salama IA
Ryll-Nardzewski C (1951) On the ergodic theorems, (eds) Ergodic theory and its connections with
II. Ergodic theory of continued fractions. Stud Math harmonic analysis, London Math Soc Lecture Note
12:74–79 Ser, vol 205. Cambridge University Press, Cambridge,
Ryzhikov V (1994) Joinings, intertwining operators, fac- pp 207–235
tors and mixing properties of dynamical systems. Tulcea AI (1964) Ergodic properties of positive isometries.
Russian Acad Sci Izv Math 42:91–114 Bull AMS 70:366–371
Shah NA (1998) Invariant measures and orbit closures on von Neumann J (1932) Proof of the quasi-ergodic hypoth-
homogeneous spaces for actions of subgroups gener- esis. Proc Natl Acad Sci USA18:70–82
ated by unipotent elements. In: Dani SG (ed) Lie groups Walters P (1993) A dynamical proof of the multiplicative
and ergodic theory. Tata Inst Fund Res Stud Math, ergodic theorem. Trans Am Math Soc 335(1):245–257
vol 14. Tata Institute of Fundamental Research, Bom- Weber M (1998) Entropie métrique et convergence presque
bay, pp 229–271 partout, Travaux en Cours, vol 58. Hermann, Paris
Shalom Y (1998) Random ergodic theorems, invariant Weiss B (2003) Actions of amenable groups. In: Bezuglyi S,
means and unitary representation. In: Dani SG Kolyada S (eds) Topics in dynamics and ergodic theory,
(ed) Lie groups and ergodic theory. Tata Inst Fund London Math Soc Lecture Note Ser, vol 310. Cambridge
Res Stud Math, vol 14. Tata Institute of Fundamental University Press, Cambridge, pp 226–262
Research, Bombay, pp 273–314 Weyl H (1916) Über die Gleichverteilung von Zahlen mod
Shannon CE (1948) A mathematical theory of communi- Eins. Math Ann 77(3):313–352
cation. Bell System Tech J27(379–423):623–656 Wiener N (1939) The ergodic theorem. Duke Math J 5(1):
Shields PC (1987) The ergodic and entropy theorems 1–18
revisited. IEEE Trans Inform Theory 33(2):263–266 Wiener N, Wintner A (1941) Harmonic analysis and ergo-
Shulman A (1988) Maximal ergodic theorems on groups. dic theory. Am J Math 63:415–426
Dep Lit NIINTI No. 2184 Wierdl M (1988) Pointwise ergodic theorem along the
Sine R (1970) A mean ergodic theorem. Proc Am Math Soc prime numbers. Israel J Math 64(3):315–336. (1989)
24:438–439 Wittmann R (1995) Almost everywhere convergence of
Smythe RT (1976) Multiparameter subadditive processes. ergodic averages of nonlinear operators. J Funct Anal
Ann Probab 4(5):772–782 127(2):326–362
Szemerédi E (1975) On sets of integers containing no k Yosida K (1940a) An abstract treatment of the individual
elements in arithmetic progression. Acta Arith 27: ergodic theorem. Proc Imp Acad Tokyo 16:280–284
199–245 Yosida K (1940b) Ergodic theorems of Birkhoff–
Tao T (2005) The Gaussian primes contain arbitrarily Khintchine’s type. Jap J Math 17:31–36
shaped constellations. http://arxivorg/abs/math/ Yosida K, Kakutani S (1939) Birkhoff’s ergodic theorem
0501314 and the maximal ergodic theorem. Proc Imp Acad,
Tao T (2008) Norm convergence of multiple ergodic aver- Tokyo 15:165–168
ages for commuting transformations. Ergod Theory Ziegler T (2005) A non-conventional ergodic theorem for a
Dyn Syst 28(2):657–688 nilsystem. Ergod Theory Dyn Sys 25(4):1357–1370
Tao T, Ziegler T (2006) The primes contain arbitrarily long Ziegler T (2007) Universal characteristic factors and
polynomial progressions. http://frontmathucdavisedu/ Furstenberg averages. J Am Math Soc 20(1):53–97.
06105050 (electronic)
Tempelman A (1992) Ergodic theorems for group actions, Zimmer RJ (1976a) Ergodic actions with generalized dis-
mathematics and its applications, vol 78. Kluwer, Dor- crete spectrum. Ill J Math 20(4):555–588
drecht, informational and thermodynamical aspects, Zimmer RJ (1976b) Extensions of ergodic group actions.
Translated and revised from the 1986 Russian original Ill J Math 20(3):373–409
Tempel'man AA (1967) Ergodic theorems for general Zund JD (2002) George David Birkhoff and John von
dynamical systems. Dokl Akad Nauk SSSR 176: Neumann: a question of priority and the ergodic theo-
790–793 rems, 1931–1932. Hist Math 29(2):138–156
Tempel'man AA (1972a) Ergodic theorems for general Zygmund A (1951) An individual ergodic theorem for non-
dynamical systems. Trudy Moskov Mat Obšč 26: commutative transformations. Acta Sci Math Szeged
95–132 14:103–110
 preserving  -action T ¼ ðT a Þa   is called
Spectral Theory of Dynamical ergodic if w0  1   is a simple eigenvalue
Systems of U T . It is weakly mixing if U T has a contin-
uous spectrum on the subspace L20 ðX , B, mÞ of
Adam Kanigowski1 and Mariusz Lemańczyk2 zero mean functions. T is said to be rigid if
1
Department of Mathematics, University of there is a sequence (an) going to infinity in 
Maryland at College Park, College Park, MD, such that the sequence U T an goes to the
USA identity in the strong (or weak) operator topol-
2
Faculty of Mathematics and Computer Science, ogy; T is said to be mildly mixing if it has no
Nicolaus Copernicus University, Toruń, Poland non-trivial rigid factors. We say that T is
mixing if the operator equal to zero is the only
limit point of U T a jL2 ðX ,B,mÞ : a   in the
0
Article Outline weak operator topology.
Cocycles and group extensions If T is an
Glossary and Notation ergodic automorphism, G is an l.c.s.c.
Definition of the Subject Abelian group, and ’ : X ! G is measurable,
Introduction then the pair (T, ’) generates a cocycle
Maximal Spectral Type of a Koopman ’()() : ℤ  X ! G, where ’ðnÞ ðxÞ ¼
Representation: Alexeyev’s Theorem ’ðxÞ þ . . . þ ’ T n1 x for n > 0,
Spectral Theory of Weighted Operators
0 for n ¼ 0
The Multiplicity Function
 ’ðT n xÞ þ . . . þ ’ T 1 x for n < 0:
Rokhlin Cocycles
Rank-1 and Related Systems (That is, (’(n)) is a standard 1-cocycle in the
Spectral Theory of Dynamical Systems of
algebraic sense for the ℤ-action n(f) ¼ f ∘ T n
Probabilistic Origin
on the group of measurable functions on
Inducing and Spectral Theory
X with values in G; hence, the function ’ :
Rigid Sequences
X ! G itself is often called a cocycle.)
Spectral Theory of Parabolic Dynamical Systems
Assume additionally that G is compact.
Spectral Theory for Locally Compact Groups of
Using the cocycle ’, we define a group
Type I
extension T’ on ðX  G, B  B ðGÞ, m  lG Þ
Future Directions
(lG stands for Haar measure of G), where
Bibliography
T’(x, g) ¼ (Tx, ’(x) + g).
Induced automorphism Assume that T is an
Glossary and Notation automorphism of a standard probability Borel
space ðX , B, mÞ . Let A  B, m(A) > 0. The
AT property of an automorphism An automor-
induced automorphism TA is defined on the
phism T of a standard probability Borel space
conditional space ðX , BA , mA Þ , where BA is
ðX , B, mÞ is called approximatively transitive
the trace of B on A, mA(B) ¼ m(B)/m(A) for
(AT for short) if for every e > 0 and every finite
B  BA and T A ðxÞ ¼ T kA ðxÞ x , where kA(x) is
set f1, . . ., fn of non-negative L1-functions on
the smallest k  1 for which Tkx  A.
ðX , B, mÞ we can find f  L1 ðX , B, mÞ also non-
Kolmogorov group property An  -action T
negative such that k fi  j aij f ∘T n j kL1 < 
satisfies the Kolmogorov group property if
for all i ¼ 1, . . ., n (for some aij  0, nj  ℕ).
sUT  sUT sU T .
Ergodicity, weak mixing, mild mixing,
mixing, and rigidity of T A measure-
© Springer Science+Business Media, LLC, part of Springer Nature 2023 109
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_511
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2022
https://doi.org/10.1007/978-3-642-27737-5_511-2
110 Spectral Theory of Dynamical Systems

Koopman representation of a dynamical sys- BðℝÞjX f , and define m f as normalized

tem T Let  be an l.c.s.c. (and not compact) m  lℝ jX f : By a special flow, we mean the
Abelian group and T : a 7! T a a representation ℝ-action T f ¼ T tf under which a point
of  in the group Aut ðX , B, mÞ of (measure- tℝ
preserving) automorphisms of a standard prob- (x, s)  X moves vertically with the unit
f

ability Borel space ðX , B, mÞ . The Koopman speed, and once it reaches the graph of f, it is
representation U ¼ U T of T in L2 ðX , B, mÞ is identified with (Tx, 0).
defined as the unitary representation a 7! U T a Spectral decomposition of a unitary
 U L2 ðX , B, mÞ , where U T a ð f Þ ¼ f ∘ T a . representation If U ¼ ðU a Þa   is a continu-
ous unitary representation of a locally compact
Markov operator A linear operator J : L2
second countable (l.c.s.c.) Abelian group  in
ðX , B, mÞ ! L2 ðY , C, vÞ is called Markov if it
a separable Hilbert space H, then a decompo-
sends non-negative functions to non-negative
sition H ¼ 1 i¼1 ðxi Þ is called spectral if
functions and J1 ¼ J1 ¼ 1.
sx1 sx2 ... (such a sequence of
Maximal spectral type and the multiplicity
measures is also called spectral); here
function of U The maximal spectral type sU
ðxÞ≔spanfU a x : a  g is called the cyclic
of U is the type of sx1 in any spectral decom-
space generated by x  H, and sx stands for
position of H; the multiplicity function MU :
the spectral measure of x.
 ! f1, 2, :g [ fþ1g is defined sU -a.e. and Spectral disjointness Two  -actions S and T
1
M U ð wÞ ¼ i¼1 1Y i ðwÞ, where Y 1 ¼  and are called spectrally disjoint if the maximal
ds spectral types of their Koopman representa-
Yi ¼ supp dsxxi for i  2. A representation U is
1
tions U T and U S on the corresponding L20 -
said to have simple spectrum if H is reduced to
spaces are mutually singular.
a single cyclic space. The multiplicity is uni-
Time change Let R ¼ (Rt)t  ℝ be a flow on
form if there is only one essential value of MU .
ðX , B, mÞ and let n  L1 ðX , B, mÞ be a positive
The essential supremum of MU is called the
function. The function n determines a
maximal spectral multiplicity. U is said to have
cocycle over R given by the formula nðt, xÞ
discrete spectrum if H has an orthonormal t
basis consisting of eigenvectors of U; U has ≔ 0 nðRs xÞds:Then for a.e. x  X and all t  ℝ,
u
singular (Haar, absolutely continuous) spec- there exists a unique u ¼ u(t, x) such that 0 n
trum if the maximal spectral type of U is ðRs xÞds ¼ t: Now, we can define the flow Rt
singular with respect to (equivalent to, abso- ðxÞ≔Ruðt,xÞ ðxÞ. The new flow R ¼ Rt
lutely continuous with) a Haar measure of . tℝ

SCS property We say that a Borel measure s on has the same orbits as the original flow, and it
preserves the measure m m (hence, it is ergo-
 satisfies the strong convolution singularity
property (SCS property) if, for each n  1, in dic if R was), where dmdm
¼ n= X ndm:
the disintegration (given by the map Unitary actions on Fock spaces If H is a sepa-
ðw1 , . . . , wn Þ 7! w1  . . .  wn Þsn ¼ nw d sðnÞ ðwÞ rable Hilbert space, then by H n, we denote the

the conditional measures nw are atomic with subspace of n-tensors of Hn symmetric under
exactly n! atoms (s(n) stands for the n-th con- all permutations of coordinates, n  1; then the
volution s  . . .  s). An -action T satisfies Hilbert space FðH Þ≔ 1 n¼0 H
n
is called a sym-
the SCS property if the maximal spectral type metric Fock space. If V  U(H), then
of U T on L20 is a type of an SCS measure. FðV Þ≔ 1 n¼0 V
n
 UðFðH ÞÞ , where V n ¼
n n
Special flow Given an ergodic automorphism T on V |H .
a standard probability Borel space ðX , B, mÞ and Weighted operator Let T be an ergodic automor-
a positive integrable function f : X ! ℝ+, we put phism of ðX , B, mÞ and x : X !  be a measur-
X f ¼ fðx, t ÞX  ℝ : 0 t < f ðxÞg, Bf ¼ B  able function. The (unitary) operator V ¼ Vx,T
Spectral Theory of Dynamical Systems 111

acting on L2 ðX , B, mÞ by the formula Vx,T( f )(x) V ma ð f ÞðwÞ ¼ iðaÞðwÞ  f ðwÞ ¼ wðaÞ  f ðwÞ w   ,
¼ x(x)f(Tx) is called a weighted operator.

where i :  !  is the canonical Pontriagin iso-

Definition of the Subject morphism of  with its second dual. Then V m ¼
V ma a   is a unitary representation of . Since
Spectral theory of dynamical systems is a study C0  is dense in L2 , m , the latter space is
of special unitary representations, called separable. Therefore, also direct sums 1 mi
i¼1 V of
Koopman representations (see Section “Glos- such type representations will be unitary represen-
sary and Notation”). Invariants of such represen- tations of  in separable Hilbert spaces.
tations are called spectral invariants of measure-
preserving systems. Together with the entropy,
they constitute the most important invariants Lemma 1 (Wiener Lemma) If F  L2 , m is
used in the study of measure-theoretic intrinsic a closed V ma -invariant subspace for all a  , then
properties and classification problems of dynam- F ¼ 1Y L2 , ^ B  ^ , m for some Borel subset
ical systems as well as in applications. Spectral
theory was originated by von Neumann, Halmos, Y  .
and Koopman in the 1930s. In this article, we
will focus on recent progresses in the spectral Notice however that since  is not compact
theory of finite measure-preserving dynamical (equivalently,  is not discrete), we can find m
systems. continuous, and therefore Vm has no irreducible
(non-zero) subrepresentation. From now on, only
unitary representations of  in separable Hilbert
spaces will be considered, and we will show how
Introduction
to classify them.
A function r :  ! ℂ is called positive definite
Throughout  denotes a non-compact l.c.s.
if
c. Abelian group ( will be most often ℤ or ℝ).
N
The assumption of second countability implies r ðan  am Þzn zm  0 ð1Þ
that  is metrizable, and s-compact and the n, m¼0
space C0 ðÞ are separable. Moreover, the dual
group  is also l.c.s.c. Abelian. for each N > 0, ðan Þ   and (zn)  ℂ.
The central result about positive definite
General Unitary Representations functions is the following theorem (see, e.g.,
We are interested in unitary, that is, with values in Rudin (1962)).
the unitary group U(H) of a Hilbert space H,
(weakly) continuous representations V :  3 Theorem 1 (Bochner-Herglotz Theorem) Let
a 7! V a  U ðH Þ of such groups (the scalar valued r :  ! ℂ be continuous. Then r is positive def-
maps a 7! hVax, yi are continuous for each x, y  inite if and only if there exists (a unique)
H). s  Mþ  such that
Let H ¼ L2 , ^ B  ^ , m , where B  ^
stands for the s-algebra of Borel sets of  and m
r ð aÞ ¼ wðaÞdsðwÞ for each a  :
 Mþ  (whenever X is an l.c.s.c. space, by 

M(X), we denote the set of complex Borel mea-

sures on X, while M+(X) stands for the subset of
If now U ¼ ðU a Þa   is a representation of 
positive (finite) measures). Given a  , for f 
in H, then for each x  H, the function r(a) ≔
L2 , B  , m put
112 Spectral Theory of Dynamical Systems

hUax, xi is continuous and satisfies (1), so it is Koopman Representations

positive definite. By the Bochner-Herglotz theo- We will consider measure-preserving representa-
rem, there exists a unique measure sU,x ¼ tions of . It means that we fix a probability stan-
sx  Mþ  (called the spectral measure of x) dard Borel space ðX , B, mÞ, and by Aut ðX , B, mÞ,
we denote the group of automorphisms of this
such that
space, that is, T  Aut ðX , B, mÞ if T : X ! X is a
bimeasurable (a.e.) bijection satisfying m(A) ¼
sx ðaÞ≔ iðaÞðwÞdsx ðwÞ ¼ hU a x, xi m(TA) ¼ m(T1A) for each A  B. Consider then a

representation of  in AutðX , B, mÞ, that is, a group
homomorphism a 7! Ta  Aut ðX , B, mÞ; we
for each a  : Since the partial map
write T ¼ ðT a Þa   . Moreover, we require that
U a x 7! iðaÞ  L2 , sx is isometric and the associated Koopman representation U T is
equivariant, there exists a unique extension of it continuous. Unless explicitly stated,  -actions
to a unitary operator W : ðxÞ ! L2 , sx giv- are assumed to be free, that is, we assume that
ing rise to an isomorphism of UjðxÞ and V sx . Then for m-a.e. x  X the map a 7! Tax is 1–1. In fact,
the existence of a spectral decomposition is pro- since constant functions are obviously invariant
ved by making use of separability and a choice of for UT a , equivalently, the trivial character 1 is
maximal cyclic spaces at every step of an induc- always an eigenvalue of U T ; the Koopman
tion procedure. Moreover, a spectral decomposi- representation is considered only on the subspace
tion is unique in the following sense. L20 ðX , B, mÞ of zero mean functions. We will
restrict our attention only to ergodic dynamical
Theorem 2 (Spectral Theorem) If H ¼ systems (see Section “Glossary and Notation”). It
1
i¼1 ðxi Þ ¼
1 0
i¼1  xi are two spectral decom-
is easy to see that T is ergodic if and only if
positions of H, then sxi  sx0i for each i  1. whenever A  B and A ¼ Ta(A) (m - a. e.) for all
a  A, then m(A) equals 0 or 1. In case of ergodic
It follows that the representation U is entirely Koopman representations, all eigenvalues are
determined by the types (the sets of equivalent simple. In particular, (ergodic) Koopman repre-
measures to a given one) of a decreasing sequence sentations with discrete spectra have simple spec-
of measures, or, equivalently, U is determined by tra. The reader is referred to monographs
its maximal spectral type sU and its multiplicity mentioned above as well as to Cornfeld et al.
function MU . (1982), Petersen (1983), Rudolph (1990), Sinai
Notice that eigenvalues of U correspond to (1994), and Walters (1982) for basic facts on the
Dirac measures: w   is an eigenvalue (i.e., for theory of dynamical systems. See also survey
some kx k ¼ 1, Ua(x) ¼ w(a)x for each a  ) if articles (Lemańczyk 2009; Danilenko 2013).
and only if sU,x ¼ dw . Therefore, U has a discrete The passage T ! U T can be seen as functorial
spectrum if and only if the maximal spectral type (contravariant). In particular, a measure-theoretic
of U is a discrete measure. isomorphism of  -systems T and T 0 implies
We refer the reader to Glasner (2003), Katok spectral isomorphism of the corresponding
and Thouvenot (2006), Lemańczyk (1996), Koopman representations; hence, spectral prop-
Nadkarni (1998), and Parry (1981) for presenta- erties are measure-theoretic invariants. Since
tions of spectral theory needed in the theory of unitary representations are completely classified,
dynamical systems – such presentations are usu- one of the main questions in the spectral theory
ally given for  ¼ ℤ , but once we have the of dynamical systems is to decide which pairs
Bochner-Herglotz theorem and the Wiener ([s], M) can be realized by Koopman represen-
lemma, their extensions to the general case are tations. The most spectacular is the Banach
straightforward. problem concerning a realization, for  ¼ ℤ, of
Spectral Theory of Dynamical Systems 113

ð½l , M  1Þ; see section “Remarks on the Markov Operators, Joinings and Koopman
Banach Problem.” Another important problem Representations, Disjointness and Spectral
is to give a complete spectral classification in Disjointness, and Entropy
some classes of dynamical systems (classically, We would like to emphasize that spectral theory is
it was done in the theory of Kolmogorov and closely related to the theory joinings (see de la
Gaussian dynamical systems). We will also see Rue’s article (de la Rue 2009) for needed defini-
how spectral properties of dynamical systems tions). The elements r of the set J ðS, T Þ of
can determine their statistical (ergodic) proper- joinings of two -actions S and T are in a 1–1
ties; a culmination given by the fact that a spec- correspondence with Markov operators J ¼ Jr
tral isomorphism may imply measure-theoretic between the L2-spaces equivariant with the
similitude (discrete spectrum case, Gaussian- corresponding Koopman representations (see sec-
Kronecker case). An old conjecture is that a tion “Glossary and Notation” and de la Rue
dynamical system T either is spectrally deter- (2009)). The set of ergodic self-joinings of an
mined or there are uncountably many pairwise ergodic -action T is denoted by J e2 ðT Þ.
non-isomorphic systems spectrally isomorphic Each Koopman representation U T consists of
to T . Markov operators (indeed, U T a is clearly a Markov
We could also consider Koopman representa- operator). In fact, if U  U(L2 ðX , B, mÞ is Markov,
tions in Lp for 1 p 6¼ 2. However, whenever W : then it is of the form UR, where R  Aut ðX , B, mÞ;
LP ðX , B, mÞ ! Lp(Y, C, v) is a surjective isometry (Lemańczyk and Parreau 2012). This allows us to
and W ∘ UT a ¼ USa ∘ W for each a  , then by see Koopman representations as unitary Markov
the Lamperti theorem (e.g., Royden 1968), the representations, but since a spectral isomorphism
isometry W has to come from a non-singular does not “preserve” the set of Markov operators,
pointwise map R : Y ! X, and, by ergodicity, spectrally isomorphic systems can have drasti-
R “preserves” the measure n and hence establishes cally different sets of self-joinings.
a measure-theoretic isomorphism (Kachurovskii We will touch here only some aspects of inter-
1990) (see also Lemańczyk (1996)). Thus, spec- actions (clearly, far from completeness) between
tral classification of such Koopman representa- the spectral theory and the theory of joinings.
tions goes back to the measure-theoretic In order to see however an example of such
classification of dynamical systems, so it looks interactions, let us recall that the simplicity of
hardly interesting. This does not mean that there eigenvalues for ergodic systems yields a short
are no interesting dynamical questions for p 6¼ 2. “joining” proof of the classical isomorphism
Let us mention still open Thouvenot’s question theorem of Halmos-von Neumann in the discrete
(formulated in the 1980s) for ℤ-actions: For every spectrum case: Assume that S ¼ ðSa Þa   and
ergodic T acting on ðX , B, mÞ, does there exist f  T ¼ ðT a Þa   are ergodic  -actions on
L1 ðX , B, mÞ such that the closed linear span of ðX , B, mÞ and ðY, C , vÞ, respectively. Assume that
f ∘ T n, n  ℤ, equals L1 ðX , B, mÞ? both Koopman representations have purely dis-
Iwanik (1991, 1992) proved that if T is a sys- crete spectrum and that their group of eigenvalues
tem with positive entropy, then its Lp-multiplicity are the same. Then S and T are measure-
is 1 for all p > 1. Moreover, Iwanik and de Sam theoretically isomorphic. Indeed, each ergodic
Lazaro (1991) proved that for Gaussian systems joining of T and S is the graph of an isomorphism
(they will be considered in section “Spectral The- of these two systems (see Lemańczyk (1996); see
ory of Dynamical Systems of Probabilistic also Goodson’s proof in Goodson (1999)).
Origin”), the Lp-multiplicities are the same for Another example of such interactions appears in
all p > 1 (see also Lemańczyk and de Sam Lazaro the study Rokhlin’s multiple mixing problem and
(1997)). its relation with the pairwise independence
114 Spectral Theory of Dynamical Systems

property (PID) for joinings of higher order. We Golodets (2002) proved this theorem for groups
will not deal with this subject here, referring the even more general than those considered here:
reader to de la Rue (2009) (see however section The spectrum of U T on L2 ðX , B, mÞ  L2 ðP Þ is
“Lifting Mixing Properties”). Haar with uniform infinite multiplicity. This gen-
Following Furstenberg (1967), two -actions eral result is quite intricate and based on methods
S and T are called disjoint provided the product introduced to entropy theory by Rudolph and
measure is the only element in J ðS, T Þ (if they are Weiss (2000) with a very surprising use of Dye’s
disjoint, one of these actions has to be ergodic). It theorem on orbital equivalence of all ergodic sys-
was already noticed in Hahn and Parry (1968) that tems. For  which is not countable, the same
spectrally disjoint systems are disjoint in the result was proved in Arni (2005) in case of
Furstenberg sense; indeed, Im J r L2 ¼ f0g unimodular amenable groups which are not
0
since sT ,Jr f sS,f . increasing union of compact subgroups. It follows
Notice that whenever we take r  J e2 ðT Þ, that spectral theory of dynamical systems essen-
we obtain a new ergodic -action ðT a  T a Þa   tially reduces to the zero entropy case.
defined on the probability space (X  X, r). One
can now ask how close spectrally to T is this new
action? It turns out that except of the obvious
Maximal Spectral Type of a Koopman
fact that the marginal s-algebras are factors,
Representation: Alexeyev’s Theorem
ðT  T , rÞ can have other factors spectrally dis-
joint from T : the most striking phenomenon here
Only few general properties of maximal spectral
is a result of Smorodinsky and Thouvenot (1979)
types of Koopman representations are known.
(see also Danilenko and Park (2002)) saying that
The fact that a Koopman representation preserves
each zero entropy system is a factor of an ergodic
the space of real functions implies that its maximal
self-joining system of a fixed Bernoulli system
spectral type is the type of a symmetric (invariant
(Bernoulli systems themselves have countable
under the map w 7! w) measure.
Haar spectrum). The situation changes if r ¼ m
Recall that the Gelfand spectrum sðU Þ of
 m. In this case, for f, g  L2 ðX , B, mÞ, the
a representation U ¼ ðU a Þa   is defined as the
spectral measure of f  g is equal to sT ,f  sT ,g .
set of approximative eigenvalues of U , that is,
A consequence of this observation is that an ergo-
sðU Þ 3 w   if for a sequence (xn) bounded and
dic action T ¼ ðT a Þa   is weakly mixing (see
bounded away from zero, kUaxn  w(a)xn k !
section “Glossary and Notation”) if and only if
0 for each a  . The spectrum is a closed subset
the product measure m  m is an ergodic self-
in the topology of pointwise convergence, hence
joining of T .
in the compact-open topology of  . In case of
The entropy which is a basic measure-theoretic
 ¼ ℤ, the above set s(U) is equal to {z  ℂ :
invariant does not appear when we deal with
U  z  Id is not invertible}.
spectral properties. We will not give here any
Assume now that  is countable and discrete
formal definition of entropy for amenable group
(and Abelian). Then there exists a Fölner sequence
actions referring the reader to Ornstein and Weiss
(Bn)n  1 whose elements tile  (Ornstein and Weiss
(1987). Assume that  is countable and discrete.
1987). Take a free and ergodic action T ¼ ðT a Þa  
We always assume that  is Abelian; hence, it is
on ðX , B, mÞ. By Ornstein and Weiss (1987) for
amenable. For each dynamical system T ¼
each e > 0, we can find a set Yn  B such that the
ðT a Þa   acting on ðX , B, mÞ, we can find a largest
sets TbYn are pairwise disjoint for b  Bn and m
invariant sub - s field P  B, called the Pinsker
ð[b  Bn T b Y n Þ > 1   . For each w  , by con-
s-algebra, such that the entropy of the
sidering functions of the form fn ¼ b  Bn wðbÞ
corresponding quotient system is zero. Generaliz-
1T b Y n , we obtain that w  sðU T Þ. It follows that the
ing the classical Rokhlin-Sinai theorem (see also
topological support of the maximal spectral type of
Kamiński (1981) for ℤd-actions), Thouvenot
the Koopman representation of a free and ergodic
(unpublished) and independently Dooley and
Spectral Theory of Dynamical Systems 115

action is full (Katok and Thouvenot 2006; above theorem for  ¼ ℤ using analytic func-
Lemańczyk 1996; Nadkarni 1998). The theory of tions. Refining Alexeyev’s original proof, Frączek
Gaussian systems shows in particular that there are (1997) showed the existence of a sufficiently reg-
symmetric measures on the circle whose topologi- ular function realizing the maximal spectral type
cal support is the whole circle but which cannot be depending only on the “regularity” of the under-
maximal spectral types of Koopman lying probability space, e.g., when X is a compact
representations. metric space (compact smooth manifold), then
An open well-known question remains of one can find a continuous (smooth) function real-
whether an absolutely continuous measure r is izing the maximal spectral type.
the maximal spectral type of a Koopman repre- By the theory of systems of probabilistic origin
sentation if and only if r is equivalent to a Haar (see section “Spectral Theory of Dynamical Sys-
measure of  (this is unknown for  ¼ ℤ). tems of Probabilistic Origin”), in case of simplic-
Another interesting question was raised by ity of the spectrum, one can easily point out
A. Katok (see Katok and Lemańczyk (2009)): Is spectral measures whose types are not realized
it true that the topological supports of all mea- by (essentially) bounded functions. However, it
sures in a spectral sequence of a Koopman repre- is still an open question whether for each
sentation are full? If the answer to this question Koopman representation U T there exists a
is positive, then, for example, the essential sup- sequence ( fi)i1  L1 ðX , B, mÞ such that the
remum of MUT is the same on all balls of . sequence s f i i1 is a spectral sequence for U T .
Notice that the fact that all spectral measures in For  ¼ ℤ, it is unknown whether the maximal
a spectral sequence are symmetric means that U T spectral type of a Koopman representation can be
is isomorphic to U T 1 : A. del Junco (1981) realized by a characteristic function.
showed that generically for ℤ-actions, T and its The group Aut ðX , B, mÞ considered with
inverse are not measure-theoretically isomorphic the weak operator topology is closed in U
(in fact, he proved disjointness). L2 ðX , B, mÞ , hence becoming a Polish group
Let T be an -action on ðX , B, mÞ. One can ask (If we choose {Ai : i  1} a dense subset in
whether a “good” function can realize the maxi- ℬ (considered modulo null sets), then the
mal spectral type of U T . In particular, can we find weak operator topology is metrizable with
a function f  L1 ðX , B, mÞ that realizes the the metric dðT 1 , T 2 Þ≔ i1 21i ðmðT 1 ðAi ÞDT 2 ðAi ÞÞ
maximal spectral type? The answer is given in þm T 1 1
1 ðAi ÞDT 2 ðAi Þ ÞÞ: One can then ask
the following general theorem (see Lemańczyk what are “typical” (largeness is understood as a
and Wasieczko (2006)). residual subset) properties of an automorphism of
ðX , B, mÞ: It is classical (Halmos) that typically an
Theorem 3 (Alexeyev’s Theorem) Assume that automorphism is weakly mixing and rigid and has
U ¼ ðU a Þa   is a unitary representation of  in a simple spectrum. Some other typical properties
separable Hilbert space H. Assume that F  H is a will be discussed later. While Halmos already
dense linear subspace. Assume moreover that with noticed that in the weak operator topology mixing
some F - norm j  j – stronger than the norm k  k automorphisms form a meager set, in Tikhonov
given by the scalar product – F becomes a Fréchet (2012), S. Tikhonov considers a special (Polish)
space. Then, for each spectral measure s ð sU Þ, topology on the set of mixing automorphisms. In
there exists y  F such that sy s. In particular, fact, this topology was introduced by Alpern
there exists y  F that realizes the maximal spec- (1985) in 1985, and Tikhonov disproves a conjec-
tral type. ture by Alpern by showing that a generic mixing
transformation has simple singular spectrum and
By taking H ¼ L2 ðX , B, mÞ and F ¼ L1 is mixing of arbitrary order; moreover, all its
ðX , B, mÞ , we obtain the positive answer to the powers are disjoint. In Tikhonov (2013), the
original question. Alexeyev (1958) proved the topology is extended to mixing actions of infinite
116 Spectral Theory of Dynamical Systems

countable groups H; it is given by the metric dm, Therefore, the spectral analysis of such Koopman
where for two H-actions T i and H ∍ h 7! j h j  representations reduces to the spectral analysis of
ℕ so that h  H1/2jhj < + 1, we have weighted operators Vw∘’,T for all w  G.

1
d m ðT 1 , T 2 Þ≔ jhj
dðT 1,h , T 2,h Þ Maximal Spectral Type of Weighted Operators
hH 2
over Rotations
1 The spectral analysis of weighted operators Vx,T is
þ sup
h  Hi, j1 2
iþj especially well developed in case of rotations, i.e.,
when we additionally assume that T is an ergodic
j m T 1,h Ai \ A j  m T 2,h Ai \ A j j: rotation on a compact monothetic group X : Tx ¼ x
+ x0, where x0 is a topologically cyclic element of
Bashtanov (2013, 2016) proved that the X (and m will stand for Haar measure lX of X). In
conjugacy classes (of mixing automorphisms) this case, Helson’s analysis (Helson 1986) applies
are dense in this topology. Hence, properties like (see also Gromov (1991), Iwanik et al. (1993),
to have trivial centralizer and no (non-trivial) fac- Lemańczyk (1996), and Queffélec (1988)), lead-
tors are typical in this topology. ing to the following conclusions:

• The maximal spectral type sV x,T is either dis-

Spectral Theory of Weighted Operators crete or continuous.
• When sV x,T is continuous, it is either singular or
We now pass to the problem of possible essential Lebesgue.
values for the multiplicity function of a Koopman • The spectral multiplicity of Vx,T is uniform.
representation. However, one of the known tech-
niques is a use of cocycles, so before we tackle the We now pass to a description of some results in
multiplicity problem, we will go through some case when Tx ¼ x + a is an irrational rotation on
results concerning spectral theory of compact the additive circle X ¼ [0, 1). It was already
group extension automorphisms which in turn noticed in the original paper by Anzai (1951)
entail a study of weighted operators (see section that when x : X !  is an affine cocycle (x(x) ¼
“Glossary and Notation”). exp (nx + c), 0 6¼ n  ℤ), then Vx,T has a Lebesgue
Assume that T is an ergodic automorphism of spectrum. It was then considered by several
a standard Borel probability space ðX , B, mÞ: authors (originated by Kushnirenko (1974); see
Let x : X !  be a measurable function and let also Choe (1990) and Cornfeld et al. (1982)) to
V ¼ Vx,T be the corresponding weighted operator. which extent this property is stable when we per-
To see a connection of weighted operators with turb our cocycle. Since the topological degree of
Koopman representations of compact group affine cocycles is different from zero, when
extensions, consider a compact (metric) Abelian perturbing them, we consider smooth perturba-
group G and a cocycle ’ : X ! G. Then U T ’ tions by cocycles of degree zero.
(see section “Glossary and Notation”) acts on
L2(X  G, m  lG). But Theorem 4 Iwanik et al. (1993) Assume that Tx
¼ x + a is an irrational rotation. If x : ½0, 1Þ ! 
L2 ðX  G, m  lG Þ ¼ Lw , is of non-zero degree, absolutely continuous, with
wG the derivative of bounded variation then Vx,T has
where Lw ¼ L2 ðX, mÞ  w a Lebesgue spectrum.

is a UT ’ -invariant (clearly, closed) subspace. In the same paper, it is noticed that if we drop
Moreover, the map f  w 7! f settles a unitary the assumption on the derivative, then the maxi-
isomorphism of U T ’ L with the operator Vw∘’,T. mal spectral type of Vx,T is a Rajchman measure
w
Spectral Theory of Dynamical Systems 117

(i.e., its Fourier transform vanishes at infinity). It manifold ℝ3/ℤ3 on which we define the nilrotation
is still an open question whether one can find x S((x, y, z)  ℤ3) ¼ (a, b, 0)  (x, y, z)  ℤ3, where a,
absolutely continuous with non-zero degree and b, and 1 are rationally independent. It can be
such that Vx,T has singular spectrum. “Below” shown that S is isomorphic to the skew product
absolute continuity, topological properties of the defined on ½0, 1Þ2   by
cocycle seem to stop playing any role – in Iwanik
et al. (1993), a continuous, degree 1 cocycle x of T ’ : ðx, y, zÞ 7! x þ a, y þ b, z  e2pi’ðx,yÞ
bounded variation is constructed such that x(x) ¼
(x)/(Tx) for a measurable  : ½0, 1Þ !  (i.e., x ¼ ðx þ a, y þ b, z þ ayÞ  ℤ3 ,
is a coboundary), and therefore Vx,T has purely
discrete spectrum (it is isomorphic to UT). For where ’(x, y) ¼ ay  c(x + a, y + b) + c(x, y) with
other results about Lebesgue spectrum for Anzai c(x, y) ¼ x[y]. Since nil-cocycles can be consid-
skew products, see also Choe (1990), Frączek ered as a certain analog of affine cocycles for one-
(2000), and Iwanik (1997); in Frączek (2000), dimensional rotations, it would be nice to explain
ℤd-actions of rotations and so-called winding to what kind of perturbations the Lebesgue spec-
numbers instead of topological degree are consid- trum property is stable.
ered. For recent generalizations, see Cecchi and Yet another interesting problem which is
Tiedra de Aldecoa (2016) and Tiedra de Aldecoa related to the spectral theory of extensions given
(2015a). by so-called Rokhlin cocycles (see section
When the cocycle is still smooth but its degree “Rokhlin Cocycles”) arises, when given f : [0, 1)
is zero, the situation drastically changes. Given an ! ℝ, we want to describe spectrally the one-
absolutely continuous function f : [0, 1) ! ℝ, parameter set of weighted operators W c ≔V e2picf ,T ;
M. Herman (1979), using the Denjoy-Koksma here T is a fixed irrational rotation by a. As proved
inequality (see, e.g., Kuipers and Niederreiter by quite sophisticated arguments in Iwanik et al.
ðq Þ
(1974)), showed that f0 n ! 0 uniformly (here (1999), if we take f(x) ¼ x, then for all noninteger c
1
f0 ¼ f  0 fdl½0,1Þ and (qn) stands for the  ℝ, the spectrum of Wc is continuous and sin-
sequence of denominators of a). It follows that gular (see also Gromov (1991) and Medina
T e2pif is rigid and hence has a singular spectrum. (1994), where some special a’s are considered).
B. Fayad (2002a) shows that this result is no It has been open for some time if at all one can find
longer true if one-dimensional rotation is replaced f : [0, 1) ! ℝ such that for each c 6¼ 0, the operator
by a multi-dimensional rotation (his counterexam- Wc has a Lebesgue spectrum. The positive answer
ple is in the analytic class). See also Lemańczyk is given in Wysokinska (2004): For example, if
and Mauduit (1994) for the singularity of spec- f(x) ¼ x(2+e)(e > 0) and a has bounded partial
trum for functions f whose Fourier transform sat- quotients, then Wc has a Lebesgue spectrum for all
isfies o jnj 1
condition or to Iwanik et al. (1999), c 6¼ 0. All functions with such a property consid-
where it is shown that sufficiently small variation ered in Wysokinska (2004) are non-integrable. It
implies singularity of the spectrum. would be interesting to find an integrable f with
A natural class of weighted operators arises the above property.
when we consider Koopman operators of rota- We refer to Goodson (1999) and the references
tions on nil-manifolds. We only look at the partic- therein for further results especially for transfor-
ular example of such a rotation on a quotient of the mations of the form (x, y) 7! (x + a, 1[0,b)(x) + y)
Heisenberg group (ℝ3, ) (a general spectral the- on [0, 1)  ℤ/2ℤ. Recall however that earlier
ory of nil-actions was mainly developed by Katok and Stepin (1967) used this kind of trans-
W. Parry (1970) – these actions have countable formations to give a first counterexample to the
Lebesgue spectrum in the orthocomplement of the Kolmogorov group property (see section “Glos-
subspace of eigenfunctions), that is, take the nil- sary and Notation”) for the spectrum.
118 Spectral Theory of Dynamical Systems

The Multiplicity Problem for Weighted examples of cocycles over so-called dyadic
Operators over Rotations adding machine for which the multiplicity of the
In case of perturbations of affine cocycles, this Lebesgue component was equal to 2. In
problem was already raised by Kushnirenko Lemańczyk (1988), this was generalized to
(1974). Some significant results were obtained so-called Toeplitz ℤ/2ℤ-extensions of adding
by M. Guenais. Before we state her results, let us machines: For each even number k, we can find
recall a useful criterion to find an upper bound for a two-point extension of an adding machine so
the multiplicity: If there exist c > 0 and a that the multiplicity of the Lebesgue component is
sequence (Fn)n1 of cyclic subspaces of H such k. Moreover, it was shown that Mathew and
that for each y  H, k y k ¼ 1 we have Nadkarni’s constructions from Mathew and
lim infn!1 k projFn yk2  c, then esssup (MU) Nadkarni (1984) in fact are close to systems aris-
1/c which follows from a well-known lemma of ing from number theory (like the famous Rudin-
Chacon (1970; Cornfeld et al. 1982; King 1988; Shapiro sequence, e.g., Queffélec (1988)), relat-
Lemańczyk 1996). Using this and a technique ing the result about multiplicity of the Lebesgue
which is close to the idea of local rank one (see component to results by Kamae and Queffélec
Ferenczi (1984) and King (1988)), M. Guenais (1988). Independently of Lemańczyk (1988),
(1998) proved a series of results on multiplicity Ageev (1988) showed that one can construct
which we now list. two-point extensions of the Chacon transforma-
tion realizing (by taking powers of the extension)
Theorem 5 Assume that Tx ¼ x + a and let x : each even number as the multiplicity of the
½0, 1Þ 7!  be a cocycle. Lebesgue component. Contrary to all previous
examples, Ageev’s constructions are weakly
(i) If x(x) ¼ e2picx, then MV x,T is bounded by jc j mixing.
+ 1. Still an open question remains whether for
(ii) If x is absolutely continuous and x is of  ¼ ℤ one can find a Koopman representation
topological degree zero, then Vx,T has a sim- with the Lebesgue component of multiplicity
ple spectrum. 1 (or even odd).
(iii) If x is of bounded variation, then MV x,T In Guenais (1999), M. Guenais studies the
max ð2, 2pVarðxÞ=3Þ. problem of Lebesgue spectrum in the classical
case of Morse sequences (see Keane (1968) as
Remarks on the Banach Problem well as Kwiatkowski (1981)), where the problem
We already mentioned in section “Introduction” of spectral classification in this class is studied).
the Banach problem in ergodic theory, which is All dynamical systems arising from Morse
simply the question whether there exists a sequences have simple spectra (Kwiatkowski
Koopman representation for  ¼ ℤ with simple 1981). It follows that if a Lebesgue component
Lebesgue spectrum. In fact no example of a appears in a Morse dynamical system, it has mul-
Koopman representation with Lebesgue spectrum tiplicity 1. Guenais (1999) using a Riesz product
of finite multiplicity is known. Helson and Parry technique relates the Lebesgue spectrum problem
(1978) asked for the validity of a still weaker with the still open problem (a variation of the
version: Can one construct T such that UT has a classical Littlewood problem) of whether a con-
Lebesgue component in its spectrum whose mul- struction of so-called L1-ultraflat trigonometric
tiplicity is finite? Quite surprisingly in Helson and polynomials with coefficients 1 is possible
Parry (1978), they give a general construction (in the very recent preprint (Balister et al. 2019),
yielding for each ergodic T a cocycle ’ : X ! ℤ/ the Littlewood problem of existence of uniformly
2ℤ such that the unitary operator U T ’ has a flat trigonometric polynomials has been solved,
Lebesgue spectrum in the orthocomplement of but it is unclear whether it yields the ultraflatness
functions depending only on the X-coordinate. condition). However, it is proved in Guenais
Then Mathew and Nadkarni (1984) gave (1999) that such a construction can be carried
Spectral Theory of Dynamical Systems 119

out on some compact Abelian groups and it leads, ergodic measure (this approach was shown to the
for an Abelian countable torsion group  , to a second author by A. del Junco). In particular, if
construction of an ergodic action of  with simple T is mixing and T’ is weakly mixing, then for each
spectrum and a Haar component in its spectrum. w  G∖f1g, the maximal spectral type of Vw∘’,T is
In Prikhodko (2013), A. Prikhodko published a Rajchman.
construction of a rank one flow (see section See section “Rokhlin Cocycles” for a general-
“Rank-1 and Related Systems”) having Lebesgue ization of the lifting result to Rokhlin cocycle
spectrum. As rank one implies simple spectrum, extensions.
the result yields solution of Banach problem for
 ¼ ℝ. To carry out the construction, Prikhodko
proved the following L1- ultraflat version of the
The Multiplicity Function
Littlewood conjecture: For all 0 < a < b and n 
ð nÞ
2piw j t
1, there are polynomials Pn ðtÞ ¼ n1 j¼0 e In this section, only  ¼ ℤ is considered. For
ðnÞ
for some real numbers w j ,
p
so that other groups, even for ℝ, much less is known.
k Pn kL ð½a,bÞ = n ! 1 when n ! 1. It seems
1 Clearly, given an automorphism T, by inducing
however that some of the arguments in the paper its Koopman ℤ-representation, we obtain a one-
are written too briefly and no further clarifying parameter group (Vt)t  ℝ of unitary operators,
presentation of methods/results/ideas from which has precisely the same properties as the
Prikhodko (2013) has appeared so far. It would original one, except that we added the eigenvalues
also be extremely nice to explain the status of n  ℝ. Moreover, classically, the induced
(El Abdalaoui 2015) by H. El Abdalaoui, first Koopman representation is given by the suspen-
posted on arXiv in 2015, which states the solution sion of T, i.e., by the special flow Tf (see section
of the original Banach problem (i.e., in the con- “Glossary and Notation”), where f ¼ 1, whence it
servative infinite measure-preserving category). is also Koopman but is never weakly mixing. See
Danilenko and Lemańczyk (2013) and Danilenko
Lifting Mixing Properties and Solomko (2010), where some of the results
We now give one more example of interactions below proved for  ¼ ℤ have been extended to
between spectral theory and joinings (see section (weakly mixing) flows. See also the case of
“Introduction”) that gives rise to a quick proof of so-called product ℤd-actions (Filipowicz 1997)
the fact that r-fold mixing property of T (r  2) and (Solomko 2012) for general countable Abe-
lifts to a weakly mixing compact group extension lian group actions.
T’ (the original proof of this fact is due to Contrary to the case of maximal spectral type,
D. Rudolph (1985)). Indeed, to prove r-fold it is rather commonly believed that there are no
mixing for a mixing(¼2-mixing) transformation restrictions for the set of essential values of
S ðacting on ðY, C , nÞÞ, one has to prove that Koopman representations. In fact, if we drop the
each weak limit of off-diagonal self-joinings assumption that we consider the finite measure-
(given by powers of S, see de la Rue (2009)) of preserving case and let ourselves consider m
order r is simply the product measure vr. We s-finite and infinite, Danilenko and Ryzhikov
need also a Furstenberg’s lemma (Furstenberg (2010, 2011) proved that all subsets of {1, 2,
1981) about relative unique ergodicity (RUE) of . . .} [ {1} are Koopman realizable (in the
compact group extensions T’: If m  lG is an weak mixing and mixing class, respectively).
ergodic measure for T’, then it is the only
(ergodic) invariant measure for T’ whose projec- Cocycle Approach
tion on the first coordinate is m. Now, the result We will only concentrate on some results of the
about lifting r-fold mixing to compact group last four decades. In 1983, refining an earlier idea
extensions follows directly from the fact that of Oseledets from 1960, E.A. Robinson (1983)
whenever T’ is weakly mixing, (m  lG)r is an proved that for each n  1, there exists an ergodic
120 Spectral Theory of Dynamical Systems

transformation whose maximal spectral multiplic- proved that for a typical automorphism T, the set
ity is n. Another important result was proved in of the values of the multiplicity function for UT k
Robinson Jr (1986) (see also Katok (2003a)), equals {k, k(k  1), . . ., k!} and then he just passes
where it is shown that given a finite set M  ℕ to “natural” factors for the Cartesian products by
containing 1 and closed under the least common taking sets invariant under a fixed subgroup of
multiple, one can find (even a weakly mixing) T so permutations of coordinates. In particular, he
that the set of essential values of the multiplicity obtains all sets of the form {2, 3, . . ., n} on L20 .
function equals M. This result was then extended He also shows that such sets of multiplicities are
in Goodson et al. (1992) to infinite sets and finally realizable in the category of mixing transforma-
in Kwiatkowski Jr and Lemańczyk (1995) (see tions. See also Ryzhikov (2009) for a realization
also Ageev (2001)) to all subsets M  ℕ of sets of the form {k, l, kl}, {k, l, m, kl, km, lm,
containing 1. In fact, as we have already noticed klm}, etc.
in the previous section, the spectral theory for A further progress was done in 2009–2012,
compact Abelian group extensions is reduced to when first Katok and Lemańczyk (2009) proved
a study of weighted operators and then to compar- that each finite subset M  {1, 2, . . .} [ {1}
ing maximal spectral types for such operators. containing 2 can be realized as the set of essential
This leads to sets of the form values of an ergodic automorphism which was
then, by overcoming some algebraic difficulties,
MðG, v, H Þ ¼ extended by Danilenko (2010, 2012) (in the
# w ∘ vi : i  ℤ \ anihðH Þ : w  anihðH Þ mixing category) to all subsets containing 2.

(H  G is a closed subgroup and v is a contin- Multiplicity for Gaussian and Poissonian

uous group automorphism of G). Then an alge- Automorphisms
braic lemma has been proved in Kwiatkowski Jr We refer the reader to section “Spectral Theory of
and Lemańczyk (1995) saying that each set Dynamical Systems of Probabilistic Origin” for the
M containing 1 is of the form M(G, v, H) and the definition and basic properties of Gaussian and
techniques to construct “good” cocycles and a Poissonian automorphisms. Recall that given a
passage to “natural factors” yielded the following: Poissonian automorphism, there is a Gaussian
For each M  {1, 2, . . .} [ {1} containing 1, automorphism spectrally isomorphic to it
there exists an ergodic automorphism such that (whether the converse holds is unknown). In
the set of essential values for its Koopman repre- Gaussian case, the classical Girsanov’s theorem
sentation equals M. See also Robinson Jr (1988) from 1950 asserts that the maximal spectral multi-
for the case of non-Abelian group extensions. plicity in this case is either one or infinity (with a
A similar in spirit approach (that means, a possibility that 1 is not an essential value); see
passage to a family of factors) is present in the Kułaga and Parreau (2012) for an elegant proof of
paper of Ageev (2008) in which he first applies this theorem. What is the family of subsets
Katok’s analysis (see Katok (2003a, b)) for spec- appearing as sets of essential values of the multi-
tral multiplicities of the Koopman representation plicity functions of Koopman operators given by
associated with Cartesian products Tk for a Gaussian (and also Poissonian) automorphisms
generic transformation T. In a natural way, this was studied by Danilenko and Ryzhikov in 2011.
approach leads to study multiplicities of tensor They prove the following remarkable results:
products of unitary operators. Roughly, the multi-
plicity is computed as the number of atoms • This family contains all multiplicative sub-
(counted modulo obvious symmetries) for condi- semigroups of ℕ.
tional measures (see Katok (2003a)) of a product • This family contains other sets than multipli-
measure over its convolution. Ageev (2008) cative sub-semigroups of ℕ.
Spectral Theory of Dynamical Systems 121

The latter shows that Proposition 6.4.4 Theorem 6 (Ageev, Danilenko) For every uni-
(multiplicative nature of MT in the Gaussian tary representation U of G in a separable Hilbert
case) claimed in the book by Katok and space H, for which U e1 Lr e1 has no non-trivial fixed
Thouvenot (2006) and also in Robinson (1986) points for 1 r < n, the essential values of the
is not true. multiplicity function for U enþ1 are contained in the
In the unpublished preprint (Ryzhikov 2014), set of multiples of n. If, in addition, U e0 has a simple
Ryzhikov shows that all subsets containing 1 are spectrum, then U enþ1 has uniform multiplicity n.
Gaussian “realizable” (even in the mixing
category). It is then a certain work to show that the
assumption of the second part of the theorem is
satisfied for a typical action of the group G. Using
Rokhlin’s Uniform Multiplicity Problem
a special (C, F)-construction with all the cut-and-
The Rokhlin multiplicity problem (see the book
stack parameters explicit, Danilenko (2006) was
by Anosov (2003)) was, given n  2, to construct
also able to show that each set of the form k  M,
an ergodic transformation with uniform multiplic-
where k  1 and M is an arbitrary subset of natural
ity n on L20 . A solution for n ¼ 2 was indepen-
numbers containing 1, is realizable as the set of
dently given by Ageev (1999) and Ryzhikov
essential values of a Koopman representation.
(1999) (see also Anosov (2003) and Goodson
Tikhonov (2011) proved the existence of a
(1999)), and in fact it consists in showing that
mixing automorphism of uniform multiplicity
for some T (actually, any T with simple spectrum
n on L20 for all n  1.
for which 12 ðId þ U T Þ is in the weak operator
closure of the powers of UT will do), the multi-
plicity of T  T is uniform and equal to 2 (see also
section “Future Directions”). In Tikhonov (2011) Rokhlin Cocycles
and Ryzhikov and Troitskaya (2016), the case n ¼
2 is solved in case of mixing automorphisms We consider now a certain class of extensions
(flows). which should be viewed as a generalization of
In Ageev (2005), Ageev proposed a new the concept of compact group extensions. We
approach which consists in considering will focus on ℤ-actions only.
actions of “slightly non-Abelian” groups and Assume that T is an ergodic automorphism of
showing that for a “typical” action of such a ðX , B, mÞ: Let G be an l.c.s.c. Abelian group.
group, a fixed “direction” automorphism has a Assume that this group acts on ðY, C , nÞ, that
uniform multiplicity. Shortly after publication is, we have a G-action S ¼ Sg g  G on
of Ageev (2005), Danilenko (2006), following ðY, C , nÞ. Let ’ : X ! G be a cocycle. We then
Ageev’s approach, considerably simplified the define an automorphism T ’,S of the space
original proof. We will present Danilenko’s ðX  Y , B  C, m  vÞ by
arguments.
Fix n  1. Denote ei ¼ ð0, . . . , T ’,S ðx, yÞ ¼ Tx, S’ðxÞ ðyÞ :
1, . . . , 0Þ  ℤn , i ¼ 1, . . . , n . We define an
automorphism L of ℤn setting Lðei Þ ¼ eiþ1 , i ¼ Such an extension is called a Rokhlin cocycle
1, . . . , n  1 and Lðen Þ ¼ e1 . Using L we define extension (the map x 7! S’(x) is called a Rokhlin
a semi-direct product G ≔ ℤn ⋊ ℤ defining cocycle). Such an operation generalizes the case
multiplication as (u, k)  (w, l) ¼ (u + Lkw, k + l). of compact group extensions; indeed, when G is
Put e0 ¼ (0, 1), ei ¼ ðei , 0Þ, i ¼ 1, . . . , n compact, the action of G on itself by rotations
ðand Lei ¼ ðLei , 0ÞÞ. Moreover, denote enþ1 ¼ preserves Haar measure. (It is quite surprising,
en0 ¼ ð0, nÞ . Notice that e0  ei  e1 0 ¼ Lei for i that when only we admit G non-Abelian, then,
¼ 1, . . ., n (L(en + 1) ¼ en+1). as shown in Danilenko and Lemańczyk (2005),
122 Spectral Theory of Dynamical Systems

each ergodic extension of T has a form of a Theorem 7 (Lemańczyk and Parreau (2003,
Rokhlin cocycle extension.) Ergodic and spectral 2012))
properties of such extensions are examined in (i) sT ’,S L2 ðXY,mvÞL2 ðX,mÞ ¼ sV w∘’,T dsS :
G
several papers: Glasner (1994), Glasner and (ii) T ’,S is ergodic if and only if T is ergodic and
Weiss (1989), Lemańczyk and Lesigne (2001),
sS L’ ¼ 0.
Lemańczyk et al. (2003), Lemańczyk and Parreau
(iii) T ’,S is weakly mixing if and only if T is
(2003, 2012), Robinson Jr (1992), and Rudolph
weakly mixing and S has no eigenvalues in
(1986). Since in these papers rather joining
S’.
aspects are studied (among other things in
(iv) If T is mixing, S is mildly mixing, and ’ is
Lemańczyk and Lesigne (2001), Furstenberg’s
recurrent and not cohomologous to a
RUE lemma is generalized to this new context),
cocycle with values in a compact subgroup
we will mention here only few results, mainly
of G, then T ’,S remains mixing.
spectral, following Lemańczyk and Lesigne
(v) If T is r-fold mixing, ’ is recurrent, and T ’,S
(2001) and Lemańczyk and Parreau (2012). We
will constantly assume that G is non-compact. As is mildly mixing, then T ’,S is also r-fold
’ : X ! G is then a cocycle with values in a non- mixing.
compact group, the theory of such cocycles is (vi) If T and R are disjoint, the cocycle ’ is
much more complicated (see, e.g., Schmidt ergodic, and S is mildly mixing, then T ’,S
(1977)), and in fact the theory of Rokhlin cocycle remains disjoint from R.
extensions leads to interesting interactions Let us recall (Furstenberg and Weiss 1978;
between classical ergodic theory, the theory of Schmidt and Walters 1982) that an -action S ¼
cocycles, and the theory of non-singular actions ðSa Þa   is mildly mixing (see section “Glossary
arising from cocycles taking values in non- and Notation”) if and only if the  -action
compact groups – especially, the Mackey action ðS a  ta Þa   remains ergodic for every properly
associated with ’ plays a crucial role here (see the ergodic non-singular -action t ¼ ðta Þa   .
problem of invariant measures for T ’,S in Coming back to Smorodinsky-Thouvenot’s
Lemańczyk and Parreau (2003) and Danilenko result about factors of ergodic self-joinings of a
and Lemańczyk (2005); see also monographs by Bernoulli automorphism, we would like to
Aaronson (1979, 1997) Katok (2001, 2003a), and emphasize here that the disjointness result
Schmidt (1977)). Especially, two Borel subgroups (vi) above was used in Lemańczyk and Parreau
of G are important here: (2003) to give an example of an automorphism
which is disjoint from all weakly mixing trans-
S’ ¼ w  G : w∘’ ¼ c  x=x∘T for a measuable formations but which has an ergodic self-joining
whose associated automorphism has a non-trivial
x : X !  and c   weakly mixing factor. In a sense, this is opposed
to Smorodinsky-Thouvenot’s result as here from
and its subgroup L’ given by c ¼ 1. L’ turns out self-joinings, we produced a “more complicated”
to be the group of L1-eigenvalues of the Mackey system (namely, the weakly mixing factor) than
action (of G) associated with the cocycle ’. This the original system.
action is the quotient action of the natural action of wIt would be interesting to develop the theory of
G (by translations on the second coordinate) on spectral multiplicity for Rokhlin cocycle exten-
the space of ergodic components of the skew sions as it was done in the case of compact group
product T’ – the Mackey action is (in general) extensions. However, a difficulty is that in the
not measure-preserving; it is however non- compact group extension case, we deal with a
singular. We refer the reader to Aaronson and countable direct sum of representations of the
Nadkarni (1987), Host et al. (1991), and Nadkarni form Vw∘’,T, while in the non-compact case, we
(1998) for other properties of those subgroups. have to consider an integral of such representations.
Spectral Theory of Dynamical Systems 123

Rank-1 and Related Systems reader to Ferenczi (1997) as a good source for
basic properties of rank-1 transformations.
Although properties like mixing, weak (and mild) Similar to the rank one property, one can define
mixing, as well as ergodicity are clearly spectral finite rank automorphisms (simply by requiring
properties, they have “good” measure-theoretic that an approximation is given by a sequence of a
formulations (expressed by a certain behavior on fixed number of towers) – see, e.g., Ornstein et al.
sets). Simple spectrum property is another exam- (1982), or even a more general property, namely,
ple of a spectral property, and it was a popular the local rank one property, can be defined, just by
question in the 1980s whether simple spectrum requiring that the approximating sequence of sin-
property of a Koopman representation can be gle towers fills up a fixed fraction of the space (see
expressed in a more “measure-theoretic” way. Ferenczi (1984) and King (1988)). Local rank one
We now recall rank-l concept which can be seen property implies finite multiplicity (King 1988),
as a notion close to Katok’s and Stepin’s theory of and the maximal spectral multiplicity is always
cyclic approximation (Katok and Stepin 1967) bounded by rank. Mentzen (1988) showed that for
(see also Cornfeld et al. (1982)). each n  1, one can construct an automorphism
Assume that T is an automorphism of a stan- with simple spectrum and having rank n, and later
dard probability Borel space ðX , B, mÞ: T is said to Kwiatkowski and Lacroix (1997) showed that for
have the rank one property if there exists an each pair (m, r) with m r, one can construct a
increasing sequence of Rokhlin towers tending rank r automorphism whose maximal spectral
to the partition into points (a Rokhlin tower is a multiplicity is m. In Lemańczyk and Sikorski
family Fn , TF n , . . ., T hn 1 Fn of pairwise (1987), there is an example of a simple spectrum
disjoint sets, while “tending to the partition into automorphism which is not of local rank one.
points” means that we can approximate every set Ferenczi (1985) deals with the notion of funny
in B by unions of levels of towers in the sequence). rank one (approximating towers are Rokhlin
Hence, basically, rank one automorphism is given towers with “holes”) – the concept that has been
by two sequences of parameters: rn, n  1, which introduced by Thouvenot. Funny rank one also
is the number of subcolumns on which we divide implies simplicity of the spectrum. An example is
the nth tower given by Fn, and S n,j , n  1, j ¼ given in Ferenczi (1985) which is even not loosely
0, 1, . . . , r n  1, the sequence of spacers put Bernoulli (see section “Inducing and Spectral
over consecutive subcolumns. A “typical” auto- Theory”; we recall that local rank one property
morphism of a standard probability Borel space implies loose Bernoullicity (Ferenczi 1984)).
has the rank one property. The “typicality” of rank The notion of AT (see section “Glossary and
one is still true in the Alpern-Tikhonov topology Notation”) has been introduced by Connes and
we mentioned in section “Maximal Spectral Type Woods (1985). They proved that AT property
of a Koopman Representation: Alexeyev’s Theo- implies zero entropy. They also proved that
rem” by Bashtanov (2013). funny rank one automorphisms are AT. In Dooley
Baxter (1971) showed that the maximal spec- and Quas (2005), it is proved that the system
tral type of such a T is realized by a characteristic induced by the classical Morse-Thue system is
function. Since the cyclic space generated by the AT (it is an open question by S. Ferenczi whether
characteristic function of the base contains char- this system has funny rank one property).
acteristic functions of all levels of the tower, by A question by Dooley and Quas is whether AT
the definition of rank one, the increasing sequence implies funny rank one property. AT property
of cyclic spaces tends to the whole L2-space; implies “simplicity of the spectrum in L1” which
therefore, rank one property implies simplicity of we already considered in section “Introduction”
the spectrum for the Koopman representation. It (a “generic” proof of this fact is due to J.-P.
was a question for some time whether rank-1 is Thouvenot).
just a characterization of simplicity of the spec- A persistent question was formulated in the
trum, disproved by del Junco (1977). We refer the 1980s whether rank one itself is a spectral
124 Spectral Theory of Dynamical Systems

property. In Ferenczi and Lemańczyk (1991), the Even though it looks as if rank one construc-
authors maintained that this is not the case, based tion is not complicated, mixing in this class is
on an unpublished preprint of the first named possible; historically the first mixing construc-
author of Ferenczi and Lemańczyk (1991) in tions were given by D. Ornstein (1970) in 1970,
which there was a construction of a Gaussian- using probability type arguments for a choice of
Kronecker automorphism (see section “Spectral spacers. Once mixing was shown, the question
Theory of Dynamical Systems of Probabilistic arose whether absolutely continuous spectrum is
Origin”) having rank-l property. This latter con- also possible, as this would give automatically the
struction turned out to be false. In fact, de la Rue positive answer to the Banach problem. However,
(1998a) proved that no Gaussian automorphism Bourgain (Bourgain 1993), relating spectral mea-
can be of local rank one. Therefore, the question sures of rank one automorphisms with some clas-
whether rank one is a spectral property remains sical constructions of Riesz product measures,
one of the interesting open questions in that the- proved that a certain subclass of Ornstein’s class
ory. Downarowicz and Kwiatkowski (2000) pro- consists of automorphisms with singular spectrum
ved that rank-l is a spectral property in the class of (see also El Abdalaoui (2007) and El Abdalaoui
systems generated by generalized Morse et al. (2006)). Since in Ornstein’s class spacers are
sequences. chosen in a certain “non-constructive” way, quite
One of the most beautiful theorems about a lot of attention was devoted to the rank one
rank-1 automorphisms is the following result of automorphism defined by cutting a tower at the
J. King (1986) (for a different proof, see Ryzhikov n-th step into rn ¼ n subcolumns of equal “width”
(1992)). and placing i spacers over the i-th subcolumn. The
mixing property conjectured by M. Smorodinsky
Theorem 8 WCT If T is of rank one, then for was proved by Adams (1998) (in fact Adams
each element S of the centralizer C(T) of T, there proved a general result on mixing of a class of
exists a sequence (nk) such that U nTk ! US staircase transformations). Spectral properties of
strongly. rank-1 transformations are also studied in Klemes
and Reinhold (1997), where the authors proved
A conjecture of J. King is that in fact for rank-l that whenever 1 2
n¼1 r n ¼ þ1, then the spec-
automorphisms, each indecomposable Markov trum is automatically singular (see also the more
operator J ¼ J r r  J e2 ðT Þ is a weak limit of recent in Creutz and Silva (2010)). H. El
powers of UT (see King (2001) and also Ryzhikov Abdalaoui (2007) gives a criterion for singularity
(1992)). To which extent the WCT remains true of the spectrum of a rank one transformation; his
for actions of other groups is not clear. In Zeitz proof uses a central limit theorem. It seems that
(1993), the WCT is proved in case of rank one still the question whether rank one implies singu-
flows; however, the main argument seems to be larity of the spectrum remains the most important
based on the fact that a rank one flow has a non- question of this theory.
zero time automorphism T t0 which is of rank We have already seen in section “Spectral The-
one, which is not true. After the proof of the ory of Weighted Operators” that for a special class
WCT by Ryzhikov in Ryzhikov (1992), there is of rank one systems, namely, those with discrete
a remark that the rank one flow version of the spectra (del Junco 1976), we have a nice theory
theorem can be proved by a word-for-word rep- for weighted operators. It would be extremely
etition of the arguments. He also proves that if interesting to find a rank one automorphism with
the flow (Tt)t  ℝ is mixing, then T1 does not have continuous spectrum for which a substitute of
finite rank. On the other hand, for  ¼ ℤ2 , Helson’s analysis exists.
Downarowicz and Kwiatkowski (2002) gave a B. Fayad (2005) constructs a rank one differ-
counterexample to the WCT. But see also entiable flow, as a special flow over a two-
Janvresse et al. (2012). dimensional rotation. In Fayad (2006), he gives
Spectral Theory of Dynamical Systems 125

new constructions of smooth flows with singular given by the Thue-Morse sequence and related
spectra which are mixing (with a new criterion for (rank two systems), see El Abdalaoui et al.
a Rajchman measure to be singular). In Fayad (2016). Weak closure of powers for Chacon auto-
(2001a), a certain smooth change of time for an morphism is described in Janvresse et al. (2015).
irrational flows on the 3-torus is given, so that the
corresponding flow is partially mixing and has the
local rank one property.
Spectral Theory of Dynamical Systems of
Motivated by Sarnak’s conjecture on Möbius
Probabilistic Origin
disjointness (see Kułaga-Przymus and
Lemańczyk (2019)), a certain recent activity was
Let us just recall that when ðY n Þ1
n¼1 is a stationary
to study spectral disjointness of powers for rank
process, then its distribution m on ℝZ is invariant
one automorphisms. Let s be a probability mea-
under the shift S on ℝZ : S((xn)n  ℤ) ¼ (yn)n  ℤ,
sure on the additive circle [0, 1). Given a real
where yn ¼ xn+1, n  ℤ. In this way, we obtain an
number a > 0, we denote by sa the image of s
automorphism S defined on (ℝℤ, B(ℝℤ), m). For
under the map x 7! ax mod 1. If r  1 is an
each automorphism T, we can find f : X ! ℝ
integer, then by sr, we will denote the measure
measurable such that the smallest s-algebra mak-
which is obtained first by taking the image of s
ing the stationary process (f ∘ T n)n  ℤ measurable
under the map x 7! 1r x, i.e., the measure s1/r, and
is equal to B ; therefore, for the purpose of this
then repeating this new measure periodically in
entry, by a system of probabilistic origin, we will
intervals rj , jþ1 . The following holds (e.g., in
r mean (S, m) obtained from a stationary infinitely
the unpublished notes by H. El Abdalaoui,
divisible process (see, e.g., Maruyama (1970) and
J. Kułaga-Przymus, M. Lemańczyk, and T. de la
Sato (1999)). In particular, the theory of Gaussian
Rue.):
dynamical systems is indeed a classical part of
ergodic theory (e.g., Newton 1966; Newton and
if ðr, sÞ ¼ 1 then sr ⊥s if and only if ss ⊥r : Parry 1966; Vershik 1962a, b). If (Xn)n  ℤ is a
ð2Þ stationary real centered Gaussian process and s
denotes the spectral measure of the process, i.e.,
In Bourgain (2013), Bourgain used Riesz prod- sðnÞ ¼ EðXn  X0 Þ, n  ℤ, then by S ¼ Ss, we
uct technique to show that for the class of denote the corresponding Gaussian system on the
so-called rank one automorphisms with bounded shift space (recall also that for each symmetric
parameters (both (rn) and (sn,j) are bounded and no measure s on  , there is exactly one stationary
spacer over the last column), we have sr ⊥ ss for r real centered Gaussian process whose spectral
6¼ s prime. In view of (2), it follows that different measure is s). Notice that if s has an atom, then
prime powers are spectrally disjoint. In El in the cyclic space generated by X0, there exists an
Abdalaoui et al. (2014), a much larger class of eigenfunction Y for Ss – if now Ss were ergodic,
rank one automorphisms is considered. No jYj would be a constant function which is not
boundedness assumption on (rn) is made, but a possible by the nature of elements in ℤ(X0). In
certain bounded recurrence is required on the what follows, we assume that s is continuous.
sequence of spacers. Spectral disjointness of dif- It follows that U Ss restricted to ℤ(X0) is spec-
ferent powers (for the continuous part of the max- trally the same as V ¼ V s acting on L2 ð, sÞ, and
imal spectral type) is derived from the existence, we obtain that U Ss , L2 ℝℤ , ms can be
in the weak closure of powers, of sufficiently represented as the symmetric Fock space built
many analytic functions of the Koopman operator over H ¼ L2 ð, sÞ and U Ss ¼ FðV Þ – see section
UT. “Glossary and Notation” (H n is called the n-th
For a spectral disjointness of the continuous chaos). In other words, the spectral theory of
part of the maximal spectral type for powers of Gaussian dynamical systems is reduced to the
automorphisms like the substitutional system spectral theory of special tensor products unitary
126 Spectral Theory of Dynamical Systems

operators. Classical results (see Cornfeld et al. has been proved in Lemańczyk and
(1982)) which can be obtained from this point of Lesigne (2001).
view are, for example, the following: The Foiaş-Stratila theorem implies that when-
ever a spectral measure s is Kronecker, it has no
(A) Ergodicity implies weak mixing. realization of the form sf with f bounded. We will
(B) The multiplicity function is either 1 or is see however in section “Future Directions” that
unbounded. for some automorphisms T (having the SCS prop-
(C) The maximal spectral type of U Ss is equal to erty), the maximal spectral type sT has the prop-
exp(s); hence, Gaussian systems enjoy the erty that SsT has a simple spectrum.
Kolmogorov group property. Gaussian-Kronecker automorphisms are
examples of automorphisms with simple spectra.
However, we can also look at a Gaussian In fact, whenever s is concentrated on a set with-
system in a different way, simply by noticing out rational relations, then Ss has a simple spec-
that the variables e2pif ( f is a real variable), trum (see Cornfeld et al. (1982)). Examples of
where f  ℤ(X0) generate L2(ℝℤ, ms). Now, mixing automorphisms with simple spectra are
calculating the spectral measure of e2pif is not known (Newton 1966); however, it is still
difficult and we obtain easily (C). Moreover, unknown (Thouvenot’s question) whether the
integrals of type e2pif 0 e2pif 1 ∘s e2pif 2 ∘s dms
n nþm
Foiaş-Stratila property may hold in the mixing
can also be calculated; whence, in particular, class. F. Parreau (2000) using independent Helson
we easily obtain Leonov’s theorem on the multi- sets gave an example of mildly mixing Gaussian
ple mixing property of Gaussian systems system with the Foiaş-Stratila property.
(Leonov 1960). Joining theory of a class of Gaussian system,
One of the most beautiful parts of the theory of called GAG, is developed in Lemańczyk et al.
Gaussian systems concerns ergodic properties of (2000). A Gaussian automorphism Ss with the
Ss when s is concentrated on a thin Borel set. Gaussian space H  L20 ℝℤ , ms is called a
Recall that a closed subset K  T is said to be a GAG if for each ergodic self-joining r  J e2 ðSs Þ
Kronecker set if each f  C(K) is a uniform limit and arbitrary f, g  H the variable
of characters (restricted to K). Each Kronecker set
has no rational relations. Gaussian-Kronecker ℝℤ  ℝℤ , r 3 ðx, yÞ 7! f ðxÞ þ gðyÞ
automorphisms are, by definition, those Gaussian
systems for which the measure s (always assumed is Gaussian. For GAG systems one can describe
to be continuous) is concentrated on K [ K, K a the centralizer and factors, they turn out to be
Kronecker set. The following theorem has been objects close to the probability structure of the
proved in Foiaş and Stratila (1968) (see also system. One of the crucial observations in
Cornfeld et al. (1982)). Lemańczyk et al. (2000) was that all Gaussian
systems with simple spectrum are GAG.
Theorem 9 Foiaş-Stratila Theorem If T is an It is conjectured (J.P. Thouvenot) that in the
ergodic automorphism and f is a real-valued ele- class of zero entropy Gaussian systems, the PID
ment of L20 such that the spectral measure sf is property holds true.
concentrated on K [ K , where K is a Kronecker For the spectral theory of classical factors of a
set, then the process ( f ∘ T n)n  ℤ is Gaussian. Gaussian system, see Lemańczyk and de Sam
Lazaro (1997); also spectrally they share basic
This theorem is indeed striking as it gives spectral properties of Gaussian systems. Recall
examples of weakly mixing automorphisms that historically one of the classical factors,
which are spectrally determined (like rotations). namely, the s-algebra of sets invariant for the
A relative version of the Foiaş-Stratila theorem map
Spectral Theory of Dynamical Systems 127

ð. . . , x1 , x0 , x1 , . . .Þ 7! We refer to Cambanis et al. (1995), Janicki and

ð. . . , x1 , x0 , x1 , . . .Þ, Weron (1994), and Rosiński and Żak (1996, 1997)
for ergodic properties of systems given by sym-
was the first example with zero entropy and count- metric a-stable stationary processes, or more gen-
able Lebesgue spectrum (indeed, we need a sin- erally infinitely divisible processes. Again, they
gular measure s such that s  s is equivalent to share spectral properties similar to the Gaussian
Lebesgue measure (Newton and Parry 1966)). For case: Ergodicity implies weak mixing, while
factors obtained as functions of a stationary pro- mixing implies mixing of all orders.
cess, see Iwanik et al. (1997). In Roy (2007), E. Roy clarifies the dynamical
T. de la Rue (1998a) proved that Gaussian “status” of such systems. He uses Poisson suspen-
systems are never of local rank-l; however, his sion automorphisms and the Maruyama represen-
argument does not apply to classical factors. We tation of an infinitely divisible process mixed with
conjecture that Gaussian systems are disjoint from basic properties of automorphisms preserving
rank-l automorphisms (or even from local rank-l infinite measure (see Aaronson (1997)) to prove
systems). that as a dynamical system, a stationary infinitely
We now turn the attention to Poissonian sys- divisible process (without the Gaussian part) is a
tems (see Cornfeld et al. (1982) for more details). factor of the Poisson suspension over the Lévy
Assume that ðX , B, mÞ is a standard Borel space, measure of this process. In Roy (2005), a theory of
where m is infinite. Without entering too much into ID-joinings is developed (which should be viewed
details, the new configuration space X is taken as as an analog of the GAG theory in the Gaussian
the set of all countable subsets {xi : i  1} of X. class). Parreau and Roy (2015) study Poisson
Once a set A  B of finite measure is given, suspensions without non-trivial factors.
one can define a map N A : X ! ℕð[f1gÞ just Many natural problems still remain open here,
counting the number of elements belonging to A. for example (assuming always zero entropy of the
The measure-theoretic structure X~ , B, ~ m~ is dynamical system under consideration): Are
given so that the maps NA become random vari- Poisson suspensions disjoint from Gaussian sys-
ables with Poisson distribution of parameter m(A) tems? In Janvresse et al. (2017), there are exam-
and such that whenever A1, . . ., Ak  X are of finite ples of Poissonian systems which are disjoint
measure and are pairwise disjoint, then the vari- from all Gaussian systems. What is the spectral
ables N A1 , . . . , N Ak are independent. structure for dynamical systems generated by
Assume now that T is an automorphism of symmetric a-stable processes? Are such systems
ðX , B, mÞ: It induces a natural automorphism on disjoint whenever a1 6¼ a2? Are Poissonian sys-
the space X~ , B, ~ m~ defined by Tðfxi : i  1g tems disjoint from local rank one automorphisms
(cf. de la Rue 1998a))? In del Junco and
¼ fTxi : i  1g. The automorphism T is called
Lemańczyk (1999), it is proved that Gaussian
the Poisson suspension of T (see Cornfeld et al.
systems are disjoint from so-called simple sys-
(1982)). Such a suspension is ergodic if and only
tems (see Veech (1982a), del Junco and Rudolph
if no set of positive and finite m-measure is
(1987), and de la Rue (2009)); we will come back
T-invariant. Moreover, the ergodicity of T implies
to an extension of this result in section “Future
weak mixing. In fact the spectral structure of U is
T Directions.” It seems that flows of probabilistic
very similar to the Gaussian one: Namely, the first
origin satisfy the Kolmogorov group property for
chaos equals L2 ðX , B, mÞ (we emphasize that this
the spectrum. One can therefore ask how different
is about the whole L2 and not only L20 ) on which
are systems satisfying the Kolmogorov group
U acts as UT and the L2 X, m together with the property from systems for which the convolutions
T
action of U has the structure of the symmetric of the maximal spectral type are pairwise disjoint
T
Fock space F L2 ðX , B, mÞ (see section “Glossary (see also section “Future Directions” and the SCS
and Notation”). property).
128 Spectral Theory of Dynamical Systems

We also mention here another problem which function is not determined in that paper. Recall
will be taken up in section “Special Flows, Flows (without giving a formal definition, see Ornstein
on Surfaces, and Interval Exchange Transforma- et al. (1982)) that a zero entropy automorphism is
tions” – Is it true that flows of probabilistic origin loosely Bernoulli (LB for short) if and only if it
are disjoint from smooth flows on surfaces? can be induced from an irrational rotation (see
Yet one more (joining) property seems to be also Feldman (1976) and Katok (1977)). The LB
characteristic in the class of systems of probabi- theory shows that not all dynamical systems can
listic origin, namely, they satisfy so-called ELF be obtained by inducing from an ergodic rotation.
property (see Derriennic et al. (2008) and de la However, an open question remained whether LB
Rue’s article (de la Rue (2009))). Vershik asked systems exhaust spectrally all Koopman represen-
whether the ELF property is spectral – however, tations. An interesting question of M. Ratner
the example of a cocycle from Wysokinska (2004) (Ratner 1978) is whether from every ergodic auto-
together with Theorem 7 (i) yields a certain morphism T, one can induce an automorphism
Rokhlin extension of a rotation which is ELF which has countable Lebesgue spectrum (Ratner
and has countable Lebesgue spectrum in the in Ratner (1978) shows that this can be done if T is
orthocomplement of the eigenfunctions (see an irrational rotation).
Wysokińska (2007)); on the other hand, any affine In a deep paper (de la Rue 1996), de la Rue
extension of that rotation is spectrally the same, studies LB property in the class of Gaussian-
while it cannot have the ELF property. Kronecker automorphisms, in particular he con-
Prikhodko and Thouvenot (private communi- structs S which is not LB. Suppose now that T is
cation) have constructed weakly mixing and non- LB and for some A  B, U T A is isomorphic to US.
mixing rank one automorphisms which enjoy the Then by the Foiaş-Stratila theorem, TA is isomor-
ELF property. phic to S, and hence TA is not LB. However, an
induced automorphism from an LB automor-
phism is LB, a contradiction.
Another fruitful source of non-LB systems
Inducing and Spectral Theory
comes from taking Cartesian products of some
natural LB systems. In Ornstein et al. (1982), it
Assume that T is an ergodic automorphism of a
is proved that there exists a rank one (and hence
standard probability Borel space ðX , B, mÞ: Can
LB) system whose Cartesian square is not
“all” dynamics be obtained by inducing (see sec-
LB. Moreover, in Ratner (1979), it was shown
tion “Glossary and Notation”) from one fixed
that the square of the horocycle flow is not LB
automorphism was a natural question from the
(the horocycle flow itself being LB (Ratner
very beginning of ergodic theory. Because of
1978)). Recently, in Kanigowski and de la Rue
Abramov’s formula for entropy h(TA) ¼ h(T)/
(2019), the authors showed that there are staircase
m(A), it is clear that positive entropy transforma-
rank one transformations whose Cartesian prod-
tions cannot be obtained from inducing on a zero
uct is not LB.
entropy automorphism. However, here we are
interested in spectral questions, and thus we ask
how many spectral types we obtain when we
induce. It is proved in Friedman and Ornstein Rigid Sequences
(1973) that the family of A  B for which TA is
mixing is dense for the (pseudo) metric d(A1, A2) Recall (see section “Glossary and Notation”) that
¼ m(A1D A2). De la Rue (1998b) proves the fol- an automorphism T of a standard probability Borel
lowing result: For each ergodic transformation space ðX , B, mÞ is called rigid if there exists
T of a standard probability space ðX , B, mÞ, the a strictly increasing sequence qn ! 1 such that
set of A  B for which the maximal spectral type of mðT qn ADAÞ ! 0 as n ! 1, for each A  B .
U T A is Lebesgue is dense in B. The multiplicity (In fact, to have a global rigidity sequence, as
Spectral Theory of Dynamical Systems 129

observed by Thouvenot, we only need to know (which are obviously rigidity sequences for the
that for each A  B there is a sequence (qn,A) so that corresponding irrational rotations Tx ¼ x + a on
m(Tqn,AADA) ! 0.) Equivalently, for each f  L2 the additive circle) to show that such sequences
q
ðX , B, mÞ, U Tn f ! f in L2 ðX , B, mÞ (it is not hard are rigid for some weakly mixing automorphisms.
to see that the latter is equivalent that Let us also mention Aaronson’s result (Aaronson
1 2piqn x
0e ds f ðxÞ ! 1 for any sf representing the 1979): Given any sequence (rn) of density 0, there
maximal spectral type of T, k f k ¼ 1). We call (qn) is a sequence (qn) such that qn < rn, n  1, and
a rigidity sequence of T. Rigidity is one of (purely (qn) is rigid for some weakly mixing
spectral) the fundamental phenomena in ergodic automorphisms.
theory. Assuming that T is aperiodic, it is not hard Moreover, in Bergelson et al. (2014), two basic
to see that for any rigidity sequence (qn), we must questions have been formulated: Given any
have qn+1  qn ! 1. Typical automorphism is sequence rigid for some T with discrete spectrum,
rigid and weakly mixing, but since weak mixing must it be rigid for some weakly mixing automor-
implies U nT ! 0 weakly on L20 ðX , B, mÞ along a phism? What about the converse?
sequence of n of full density, there is no much The positive answer to the first question was
“room” left for rigidity sequences. So positive given by Adams (2015) and Fayad and Thouvenot
density sequences cannot be rigid, but beyond (2014). On the other hand, surprisingly, Fayad and
that, in the class of zero density sequences, there Kanigowski (2015) answered negatively the sec-
can be other, for example algebraic in nature, ond question: There are rigidity sequences (for
obstructions for rigidity. For example, as noticed weakly mixing automorphisms) which are not
in Bergelson et al. (2014) and Eisner and Grivaux rigidity sequences for any rotation. For a strength-
(2011), if P  ℚ[x] is any non-zero polynomial ening of this result (the existence of a rigidity
taking integer values on ℤ, then the sequence sequence which, as a subset, is dense in the Bohr
(P(n)) cannot be rigid for any ergodic automor- topology on ℤ), see Griesmer (2019).
phism. It is also easy to see that (2n) is a rigidity One can also consider a notion stronger than
sequence while (2n + 1) is not. rigidity, called IP-rigidity (see, e.g., Bergelson
A systematic study of sequences which can be et al. (2014)): (qn) is an IP-rigidity sequence for
rigidity sequences was originated in Bergelson an automorphism T acting on ðX , B, mÞ if T x ! Id
et al. (2014) and Eisner and Grivaux (2011). in the strong topology of L2 ðX , B, mÞÞ in the IP
Both papers use harmonic analysis approach to sense, that is, when x ! 1, where x ¼ qm1
construct rigid sequences (via the standard Gauss- þ   þ qmk and we require that the smallest ele-
ian functor). Other constructions are also pre- ment qm1 is going to 1. This notion is studied in
sented in Bergelson et al. (2014), rank one Aaronson et al. (2014) relating it to non-singular
constructions, weighted operators, and Poisson ergodic theory (more precisely, to groups of
suspensions, while Eisner and Grivaux (2011) so-called L1-eigenvalues of non-singular auto-
rather concentrates on so-called linear dynamical morphisms). As proved in Aaronson et al.
systems and studies rigidity for weakly mixing (2014), in this category, the answer to the second
automorphisms. One of the results in Bergelson question (above) from Bergelson et al. (2014)
et al. (2014 and Eisner and Grivaux (2011) states turns out to be positive. Moreover, the paper pro-
that if either qn+1/qn ! 1 or qn+1/qn is an integer, vides an example of a super-lacunary sequence
then (qn) is a rigidity sequence (for a weakly (which must be a rigidity sequence by Bergelson
mixing automorphism). On the other hand, Eisner et al. (2014) and Eisner and Grivaux (2011))
and Grivaux in Eisner and Grivaux (2011) give an which is not an IP-rigid.
example of a rigid sequence (qn), for a weakly In the recent preprint (Badea et al. 2018), rigid-
mixing automorphism, such that qn+1/qn ! 1. As ity sequences are compared to other classical
a matter of fact, both Bergelson et al. (2014) and notions in harmonic analysis. It is proved that
Eisner and Grivaux (2011) deal with the case of rigidity sequences (qn) are nullpotent, i.e., there
denominators (qn) of an irrational a  [0, 1) exists a topology t on ℤ making it a topological
130 Spectral Theory of Dynamical Systems

group such that qn ! 0, but they are never systems is a lack of many tools from representa-
Kazhdan. (A subset B  ℤ is called Kazhdan if tion theory, which is available in the algebraic
there exists e > 0 such that each unitary operator setting. Below, we focus on known results and
U on a separable Hilbert space H having a unit questions in spectral theory of non-algebraic par-
vector x with supn  B k Unx  x k < e has a non- abolic systems.
zero fixed point.) We find also there a rather sur-
prising result that the family of all rigidity Time-Changes of Algebraic Systems
sequences considered as a subset in ℤℕ is Borel. Perhaps the simplest class of non-algebraic para-
bolic dynamical systems is given by time-changes
(or reparametrizations) of algebraic systems. As
Spectral Theory of Parabolic Dynamical for algebraic systems, it is natural to consider
Systems separately the cases of time-changes of unipotent
systems and nilpotent systems. We do it in two
We say a system is algebraic if it is a ℤ (or ℝ) paragraphs below.
translation on a quotient of a Lie group by a Time-changes of unipotent systems In recent
lattice. Spectral theory (and mixing properties) years, we witnessed substantial development in
of algebraic systems is by now well understood. understanding the theory of time-changes of
The two main classes are actions on quotients of unipotent flows. The first (and most studied)
semi-simple and nilpotent Lie groups. In the first case is that of smooth time changes of horocycle
case, the two main examples are horocycle and flows. Recall first that M. Ratner (1987)
geodesic flows on quotients of SL(2, ℝ). More established measures and joinings rigidity phe-
generally, one can talk about quasi-unipotent and nomena for time-changes of horocycle flows that
partially hyperbolic actions. Recall that in the are analogous to Ratner’s theory in algebraic set-
setting of algebraic actions, being quasi-unipotent ting. In particular in Ratner (1986), Ratner proved
is equivalent to zero entropy, while being partially the H-property for all (sufficiently smooth time-
hyperbolic is equivalent to positive entropy. changes). It is not known if Ratner’s joinings and
It is known that in both cases, the spectrum is measure rigidity also holds for time-changes of
countable Lebesgue (we refer the reader to Katok general unipotent flows.
and Thouvenot (2006) for a nice description of Mixing for smooth time-changes of horocycle
spectral theory of horocycle and geodesic flows). flows was established by Marcus (1977), general-
Actions on quotients of nilpotent Lie groups are izing earlier work of Kushnirenko (1974) who
also known to have countable Lebesgue spectrum required additionally small derivative in the geo-
in the orthocomplement of the eigenspace desic direction. A crucial result for the theory is by
(we refer to Parry (1970) for details). Quantitative L. Flaminio and G. Forni (2003), where the
mixing (and higher-order mixing) of algebraic authors classify all invariant distributions and as
systems is also well understood (we refer the a consequence show that a typical time-change is
reader to a recent paper, Björklund et al. (2020), not a (measurable) quasi-coboundary (and hence
for general results on decay of correlations for the time-change is not trivially isomorphic to the
algebraic systems on semi-simple Lie groups). original flow). A. Katok and J.-P. Thouvenot
Much less is known in spectral theory of conjectured (Katok and Thouvenot 2006) that
parabolic systems beyond algebraic world. There every sufficiently smooth time change of the
is no strict definition for a system to be horocycle flow has countable Lebesgue spectrum.
parabolic. However, characteristic features of par- A partial answer to this conjecture was given by
abolic systems are zero entropy, polynomial G. Forni and C. Ulcigrai (2012), where the authors
orbit growth, strong mixing, and equidistribution show that the maximal spectral type of the time-
properties. We describe some classes of changed flow remains Lebesgue (see also a result
(non-algebraic) parabolic systems below. One of of R. Tiedra de Aldecoa (2012), where the abso-
the main difficulties in studying non-algebraic lute continuity of the spectrum is proven, and
Spectral Theory of Dynamical Systems 131

Tiedra de Aldecoa (2015b, 2017) for further appli- strengthened by D. Ravotti in Ravotti (2019) to
cations of the commutator method in ergodic the- quasi-Abelian nilflows and recently to all nilflows
ory). A full solution of the Katok-Thouvenot by Avila, Forni, Ravotti, and Ulcigrai (2019). It is
conjecture (i.e., countable Lebesgue spectrum) important to mention that the mixing mechanism
was recently given by G. Forni, B. Fayad, and is nonquantitative. Therefore, two questions are
A. Kanigowski (2016). Generalizing the approach natural to ask: What are mixing properties of
from Forni and Ulcigrai (2012), L. Simonelli general time-changes of nilflows and can one
(2018) showed that the spectrum of smooth obtain some quantitative mixing results? The
time-changes of general unipotent flows remains only case in which some progress has been
Lebesgue. It is not known if the multiplicity is recently made is that of time-changes of Heisen-
infinite, but it seems that the approach from Fayad berg nilflows. In Forni and Kanigowski (2017),
et al. (2016) has the potential of being applicable the authors show stretched polynomial decay of
in this setting. correlations for smooth time-changes of full mea-
Recall that Ratner’s work (Ratner 1983) allows sure set of Heisenberg nilflows (parameterized by
one to classify joinings between horocycle flows. the frequency of the Kronecker factor). In the case
Recently, there was a progress in understanding the flow is of bounded type, the authors prove
joinings for time-changes of horocycle flows. In polynomial speed of decay of correlations. More-
Kanigowski et al. (2018) (see also a result in over, in Forni and Kanigowski (2020), the authors
Flaminio and Forni (2019)), the authors show show that for time-changes of bounded type Hei-
that there is a strong dichotomy for two smooth senberg nilflows, every non-trivial time-change
time-changes of horocycle flows: Either the time- enjoys the R-property and as a consequence is
change functions are cohomologous or the mildly mixing. Moreover, in the above setting, it
resulting time-changed flows are disjoint. also follows that every mixing time-change is
Even though quantitative mixing for time- mixing of all orders.
changes of unipotent flows is now well under- Mixing and spectral properties of time-changes
stood, not much is known for quantitative of general nilflows are poorly understood. In par-
higher-order correlations: ticular, the following question seems interesting:

Question 1 Is the decay of higher correlations Question 2 Are all non-trivial smooth time-
for non-trivial time-changes of horocycle flows changes of general nilflows mixing?
(or more generally, unipotent flows) polynomial?
The mixing mechanism from Avila et al.
For trivial time-changes, i.e., for the horocycle (2011) (see also Ravotti (2019) and Avila et al.
flow, decay of higher correlations is indeed poly- (2019)) is nonquantitative. Therefore, the follow-
nomial by a recent result of M. Björklund, ing question seems to be far more challenging:
M. Einsiedler, and A. Gorodnik (2020) (in fact
this applies to general unipotent flows). Question 3 Does there exist a smooth time
Time-changes of nilpotent systems Recall change of a general nilflow with AC (Lebesgue)
that nilpotent flows are never weakly mixing spectrum?
since they always have a non-trivial Kronecker
factor. An interesting question is therefore Special Flows, Flows on Surfaces, and Interval
whether one can improve mixing properties of Exchange Transformations
the system by a time-change. The first result in In this section, we will describe spectral results for
this direction by A. Avila, G. Forni, and special flows over interval exchange transforma-
C. Ulcigrai (2011) is that there exists a dense set tions (IETs) (irrational rotations begin a particular
of smooth functions on the Heisenberg case). As described below, such special flows
nilfmanifold such that the resulting time-changed arise as representations of smooth locally Hamil-
Heisenberg flow is mixing. This result was tonian flows on surfaces.
132 Spectral Theory of Dynamical Systems

Interval Exchange Transformations space of IETs R  Dm1, Veech has proved the
To define an interval exchange transformation following fundamental theorem (Veech 1982b;
(IET) of m intervals, we need a permutation p of see also Masur (1982)) which is a generalization
{1, . . ., m} and a probability vector l ¼ (l1, . . ., of the fact that Gauss measure ln1 2 1þx
1
dx is invari-
lm) (with positive entries). Then, we define T ¼ ant for the Gauss map which sends t  (0, 1) into
Tl,p of [0, 1) by putting the fractional part of its inverse.

T l,p ðxÞ ¼ x þ bpi  bi for x  bi , biþ1 , Theorem 10 (Veech and Masur) (1982) Assume
that R is a Rauzy class. There exists a s-finite
where bi ¼ j<i l j , bi ¼
p
pj<pi b j : Obviously, measure mR on R  Dm1 which is P -invariant,
each IET preserves Lebesgue measure. One of equivalent to “Lebesgue” measure, conservative
possible approaches to study ergodic properties and ergodic.
of IET is an “a.e.” approach “seen” in the defini-
tion of Tl,p. It is based on the fundamental fact that In Veech (1982b), it is proved that a.e. (in the
the induced transformation on a subinterval of above sense) IET is then of rank one (and hence is
[0, 1) is also IET (see Cornfeld et al. (1982)). ergodic and has a simple spectrum). He also for-
This leads to a very delicate and deep mathematics mulated as an open problem whether we can
based on Rauzy induction, which is a way of achieve the weak mixing property a.e. This has
inducing on special intervals, considering only been answered in positive by A. Avila and
irreducible permutations whose set is partitioned G. Forni (2007) (for p which is not a rotation).
into orbits of some maps (any such an orbit is Katok (1980) proved that IET have no mixing
called a Rauzy class). If now R is a Rauzy class factors (in fact his proof shows more: The IET
of permutations and l lies in the standard simplex class is disjoint from the class of mixing trans-
Dm1, then the Rauzy induction together with a formations). By their nature, IET transformations
natural renormalization leads to a map P : are of finite rank (see Cornfeld et al. (1982)), so
R  Dm1 ! R  Dm1. For a better understand- they are of finite multiplicity. They need not be of
ing of the dynamics of the Rauzy map, Veech simple spectrum (see remarks in Katok and
(1982b) introduced the space of zippered rectan- Thouvenot (2006), pp. 88–90). It remains an
gles. A zippered rectangle associated with the open question whether an IET can have a non-
Rauzy class R is a quadruple (l, h, a, p), where singular spectrum. Joining properties in the class
l  ℝmþ , h  ℝþ , a  ℝþ , p  R and the vectors
m m of exchange of 3 and more intervals are studied in
h and a satisfy some equations and inequalities. Ferenczi et al. (2004, 2005). In Chaika and Eskin
Every zippered rectangle (l, h, a, p) determines a (2018), the authors show that a.e. 3-IET is not
Riemann structure on a compact connected sur- simple. This answers a special case of a question
face. Denote by O(R) the space of all zippered of Veech (1982a) whether a.e. IET is simple. The
rectangles, corresponding to a given Rauzy class case of d-IET’s with d > 3 is still widely open.
R and satisfying the condition hl, hi ¼ 1. In
Veech (1982b), Veech defined a form (P t)t  ℝ on Smooth Flows on Surfaces and Their Special
the space O(R) putting Representations
We consider a closed, connected, smooth, and
Pt ðl, h, a, pÞ ¼ ðet l, et h, et a, pÞ orientable surface S of genus g  1. Let X :
S 7! TS be a smooth vector field with finitely
and extended the Rauzy map. On so-called Veech many fixed points and such that the corresponding
moduli space of zippered rectangles, the flow (Pt) flow fXt preserves a smooth area form o. The
becomes the Teichmüller flow, and it preserves a form fXt is called a locally Hamiltonian flow; it
natural Lebesgue measure class; by passing to a is locally given by a smooth Hamiltonian H (up to
transversal and projecting the measure on the an additive constant), so that fXt is a solution to
Spectral Theory of Dynamical Systems 133

dx @H dy @H that critical points of the vector field in (3) become

¼ , ¼ : ð3Þ
dt @y dt @x singular points, Arnold’s special representation is
transformed to a special flow over the same irra-
There has recently been some progress in tional rotation however under a piecewise smooth
understanding ergodic and spectral properties of function. If the sum of jumps is not zero, then in
locally Hamiltonian flows. It follows by Arnold fact we come back to von Neumann’s class of
(1991) that fXt can be decomposed into finite special flows considered in von Neumann
number of topological discs filled with periodic (1932). Similar classes of special flows (when
orbits and finite number of connected components the roof function is of bounded variation) are
on which the flow is minimal. We are always obtained from ergodic components of flows asso-
interested in properties of fXt on the minimal ciated with billiards in convex polygons with
components. A classical way of studying locally rational angles (Katok and Zemlyakov 1975).
Hamiltonian flows is through their special repre- It is convenient to express ergodic and spectral
sentations. On each minimal component, one properties of locally Hamiltonian flows in terms
chooses a smooth transversal and represents of their special representation. We will do it in the
fXt by the first return map (the base transforma- next subsection.
tion T) and the first return time (the roof function
f ). We will denote the special representation of Special Flows over Rotations and Interval
fXt by T f. It is well-known that T is either a Exchange Transformations
rotation (in genus 1 case) or an interval exchange Special flows were introduced to ergodic theory
transformation (in higher genus) and the roof by von Neumann in his fundamental work (von
function is smooth except for finitely many points Neumann 1932) in 1932. Also in that work he
at which it explodes according to the nature of the explains how to compute eigenvalues for special
fixed points. Notice first that if fXt has no fixed flows, namely: r  ℝ is an eigenvalue of T f if and
points, then S is a two-dimensional torus only if the following functional equation
(by Gauss-Bonnet). In this case, fXt is a smooth
time-change of a linear flow on the two-torus. e2pirf ðxÞ ¼ xðxÞ=xðTxÞ
Therefore, it is never mixing (Kočergin 1972)
and spectrally disjoint from mixing flows has a measurable solution x : X ! . We recall
(Frączek and Lemańczyk 2004) and in particular also that the classical Ambrose-Kakutani theorem
has purely singular spectrum. From now on, we asserts that practically each ergodic flow has a
will therefore assume that the set of fixed points is special representation (Cornfeld et al. 1982; see
non-empty. If all fixed points are non-degenerated also Rudolph’s theorem on special representation
(i.e., Hessian of H is non-zero at every critical therein). As described in the previous subsection,
point), the roof function f has logarithmic singu- we will consider the case when the roof functions
larities, i.e., it blows up logarithmically at finitely is of bounded variation, has logarithmic singular-
many points. Logarithmic singularities can be ities (symmetric or asymmetric), or has power
either symmetric (for instance, if there are no singularities.
saddle connections) or asymmetric (in case there
is a saddle loop). Ergodic properties of fXt Bounded Variation Roof Function
depend strongly on whether the roof has symmet- Kochergin (1972) showed that special flows over
ric or asymmetric singularities. If there are irrational rotations and under bounded variation
degenerated fixed points, Kochergin (Kochergin functions are never mixing. This has been
1975) constructed smooth Hamiltonian for which strengthened in Frączek and Lemańczyk (2004)
the roof function at singularities blows up power- to the following: If T is an irrational rotation and f
like (like xg, g  (1, 0)). By changing a speed as is of bounded variation, then the special flow T f is
it is done in Frączek and Lemańczyk (2006) so spectrally disjoint from all mixing flows. In
134 Spectral Theory of Dynamical Systems

particular, all such flows have singular spectra. many frequencies can have the group of eigen-
Moreover, in Frączek and Lemańczyk (2004), it values) for special flows over irrational rotations
is proved that whenever the Fourier transform of is studied in Fayad et al. (2001), Fayad and
the roof function f is of order O 1n , then T f is Windsor (2007), and Guenais and Parreau (2005).
disjoint from all mixing flows (see also Frączek It follows from Frączek and Lemańczyk (2004)
and Lemańczyk (2003)). In fact in the papers that von Neumann’s flows have singular spec-
(Frączek and Lemańczyk 2003, 2004, 2005, trum. However, nothing is known about their
2006), the authors discuss the problem of multiplicity.
disjointness of those special flows with all ELF-
flows conjecturing that no flow of probabilistic Question 4 What is the spectral multiplicity of
origin has a smooth realization on a surface. In von Neumann flows?
Lemańczyk and Wysokińska (2007), the analytic
case is considered, leading to a “generic” result on
Symmetric Logarithmic Singularities
disjointness with the ELF class generalizing the
Kochergin (Kochergin 1976a) proved the absence
classical Shklover’s result on the weak mixing
of mixing for flows, where the roof function has
property (Shklover 1967). A. Katok (1980) pro-
finitely many singularities; however, some
ved the absence of mixing for special flows over
Diophantine restriction is put on a. In Lemańczyk
IET when the roof function is of bounded varia-
(2000), where also the absence of mixing is con-
tion (see also Ryzhikov (1994a)). Katok’s theo-
sidered for the symmetric logarithmic case, it was
rem was strengthened in Frączek and Lemańczyk
conjectured (and proved for arbitrary rotation)
(2005) to the disjointness theorem with the class
that a necessary condition for mixing of a special
of mixing flows. A. Avila and G. Forni (Avila and
flow T f (with arbitrary T and f) is the condition that
Forni 2007) proved that a.e. translation flow on a ðnÞ
the sequence of distributions f0 tends to
surface (of genus at least two) is weakly mixing  n
(which is a drastic difference with the linear flow d1 in the space of probability measures on ℝ .
case of the torus, where the spectrum is always K. Schmidt (Schmidt 2002) proved it using the
discrete). theory of cocycles and extending a result from
One important property in the class of special Aaronson and Weiss (2000) on tightness of
flows over rotations and IETs is Ratner’s property cocycles. Ulcigrai (2011) showed that for every
(R-property). This property may be viewed as a d  2 and a.e. IET of d intervals, the
particular way of divergence of orbits of close corresponding special flow T f is not mixing. How-
points; it was shown to hold for horocycle flows ever, Chaika and Wright (2019) proved existence
by M. Ratner (1983). We refer the reader to Ratner of an IET T such that T f is mixing. Notice that by
(1983) and the survey article in Thouvenot (1995) Frączek and Lemańczyk (2004), it also follows
for the formal definitions and basic consequences that for a.e. irrational rotation T f has purely sin-
of R-property. In particular, R-property implies gular spectral type. Recent result (Chaika et al.
“rigidity” of joinings and it also implies the PID 2019) shows that one can also prove singularity of
property; hence, mixing and R-property imply the spectrum for symmetric IETs in the base. This
mixing of all orders. In Frączek and Lemańczyk in particular shows that minimal flows on genus
(2006) and Frączek et al. (2007), a version of 2 surfaces (with two isometric saddles) have
R-property is shown for the class of von Neumann purely singular spectral type. In Kanigowski and
special flows (however, a is assumed to have Kułaga-Przymus (2016), the authors showed that
bounded partial quotients). This allowed one to if T is an IET of bounded type, then T f is mildly
prove there that such flows are even mildly mixing (and has the R-property). For symmetric
mixing (mixing is excluded by a Kochergin’s logarithmic singularities, the following two ques-
result). The eigenvalue problem (mainly how tions are open:
Spectral Theory of Dynamical Systems 135

Question 5 What is the maximal spectral type of obtained by B. Fayad (2002b). In Fayad et al.
T f for a general IETs? (2016), the authors considered the spectrum of
T f. They showed that if f has sufficiently strong
Nothing is known about multiplicity of the power singularity (of the form x1+ for small  >
spectrum in this setting. 0), then T f has countable Lebesgue spectrum for
a.e. irrational rotation. To the best of our knowl-
Asymmetric Logarithmic Singularities edge, this is the only result dealing with multiplic-
In this case, mixing properties are different. ity for smooth surface flows. The following
Khanin and Sinai (1992) showed that T f is mixing problems are natural (see Question 34 in Fayad
for a.e. irrational rotation T. This was strengthened and Krikorian (2018)):
by Ulcigrai to a.e. IET (Ulcigrai 2007). Moreover
(Ravotti 2017), Ravotti obtained quantitative Question 8 What is the maximal spectral type of
mixing estimates (with sub-logarithmic speed of Tf when f has power singularities? What is the
decay of correlations). Not much was known multiplicity?
about multiple mixing for T f. This changed
recently: In Fayad and Kanigowski (2016), the To answer this, one needs to consider general
authors showed that for a.e. irrational rotation, T f IETs in the base, as well as functions with weaker
enjoys a variant of the R-property and hence is power singularity than in Fayad et al. (2016). The
multiple mixing. This was strengthened to following question is still open (Question 38 in
a.e. IETs in Kanigowski et al. (2019). The follow- Fayad and Krikorian (2018)):
ing two questions seem to be natural (the first one
already stated as Questions 34, 35 in Fayad and Question 9 Are all mixing surface flows mixing
Krikorian (2018)): of all orders?

Question 6 What is the maximal spectral type of Finally, it may also be useful to show that
T f? smooth flows on surfaces are disjoint from flows
of probabilistic origin – see del Junco and
Moreover, one can ask about quantitative Lemańczyk (1999, 2005), Lemańczyk et al.
higher-order decay: (2011), Ryzhikov and Thouvenot (2006), and
Thouvenot (2000).
Question 7 Is the decay of higher-order correla- B. Fayad (2006) gives a criterion that implies
tions sub-logarithmic? singularity of the maximal spectral type for a
dynamical system on a Riemannian manifold. As
Both of the above questions are open even for an application, he gives a class of smooth mixing
rotations in the base. As in the symmetric case, flows (with singular spectra) on 3 obtained from
nothing is known about multiplicity of the linear flows by a time change (again this is a
spectrum. drastic difference with dimension two, where a
smooth time change of a linear flow leads to
non-mixing flows (Cornfeld et al. 1982)).
Power Singularities
We mention at the end that if we drop here (and
In case f has power singularities, T f was shown to
in other problems) the assumption of regularity of
be mixing by Kochergin (1975) (for any irrational
f, then the answers will be always positive because
rotation and IETs). In Fayad and Kanigowski
of the LB theory; in particular, there is a section of
(2016), the authors show that if T is a rotation of
any horocycle flow (it has the LB property (Ratner
bounded type, then T f is multiple mixing
1978)) such that in the corresponding special rep-
(it enjoys a variant of the R-property). Polynomial
resentation T f, the map T is an irrational rotation.
decay for some flows with power singularities was
136 Spectral Theory of Dynamical Systems

1
Using a Kochergin’s result (Kochergin 1976b) on ¼ n¼1 Gn G1 , we obtain a Borel structure
cohomology (see also Katok (2003a) and on G. It is called Mackey-Borel structure on G.
Rudolph (1986)), the L1-function f is By the Glimm theorem, the Mackey-Borel struc-
cohomologous to a positive function g which is ture is standard if and only if G is of type I (Glimm
even continuous; thus, T f is isomorphic to T g. 1960).
For n  ℕ [ {1}, denote by In the identity
operator on Kn . Then for each unitary representa-
Spectral Theory for Locally Compact tion p of G in H, there are a measure l, a
Groups of Type I measurable field G 3 o 7! H o of Hilbert spaces,
and a measurable field G 3 o 7! V o of irreducible
This section has been written by A. Danilenko. unitary G-representations such that Vo  o, and
Ho is the space of Vo on G and a measurable map
Groups of Type I m : G ! ℕ [ fþ1g such that
The spectral theory presented here for Abelian
group actions extends potentially to probability-
H¼ Hw  KmðwÞ dlðoÞ and
preserving actions of non-Abelian locally com- ^
G
pact groups of type I. We now provide the defini-
tion of type I. Let G be a locally compact second pðgÞ ¼ V w ðgÞ  I mðwÞ dlðoÞ:
^
G
countable group, H a separable Hilbert space, and
p : G ∍ g 7! p(g) a (weakly) continuous unitary It appears that if G is of type I, then the equiv-
representation of G in H. We say that p is of type I alence class of l is defined uniquely by p, and the
if there is a subset A  {1, 2, . . ., +1} such that p function m is defined up to a l-zero subset. We call
is unitarily equivalent to the orthogonal sum the class of l the maximal spectral type of p, and
k  AUk  Ik, where Uk is a unitary representation we call m the spectral multiplicity of p. If we have
of G with a simple spectrum and Ik is the trivial a probability-preserving action T ¼ (Tg)g  G of G,
representation in the Hilbert space of dimension k. then we can consider the corresponding Koopman
unitary representation p of G. The maximal spec-
Definition 1 If every unitary representation of tral type of p and the spectral multiplicity of p are
G is of type I, then G is called of type I. called the maximal spectral type of T and the
spectral multiplicity of T, respectively. If G is
Denote by G the unitary dual of G, i.e., the set Abelian, then these concepts coincide with their
of unitarily equivalent classes of all irreducible classic counterparts considered above in the
unitary representations of G. If G is Abelian, survey.
then every irreducible representation of G is one All compact groups, Abelian groups,
dimensional. Hence, G is identified naturally with connected semi-simple Lie groups, and nilpotent
the group of characters of G. In the general case, Lie groups (or, more generally, exponential Lie
let Irrn(G) stand for the set of all irreducible uni- groups) are of type I. Each subgroup of GL(n, ℝ)
tary representations of G in the n-dimensional determined by a system of algebraic equations is
separable Hilbert space Kn , 1 n þ1: also of type I. Solvable Lie groups can be as of
Endow it with the natural Borel structure, i.e., type I as not of type I. If G is a countable (discrete)
the smallest one in which the mapping p 7! group, then G is of type I if, and only if it is
hp(g)f, hi is Borel for every g  G and f , h  virtually Abelian, i.e., it contains an Abelian sub-
Kn. It is standard. Let Gn be the quotient of Irrn(G) group of finite index (Thoma 1964).
by the unitary equivalence relation. Endow Gn Even if we know that a non-Abelian group G is
with the quotient Borel structure. Since G of type I, it is usually not an easy problem to
Spectral Theory of Dynamical Systems 137

describe G explicitly. Kirillov introduced an orbit l1,2 ða, bÞ

L2 ðX, mÞ ¼ j¼1 ℂds1,2 ða, bÞ
method for a description of G when G is a ℝ 2

connected, simply connected nilpotent Lie group

l3 ða, bÞ
(the method was developed further for solvable j¼1 L2 ðℝ, LebÞds3 ðgÞ and

groups). He identified G with the space of orbits ℝ

for the co-adjoint G-action on the dual space g of l1,2 , ða, bÞ

U¼ pa,b ds1,2 ða, bÞ
its Lie algebra g. Though Kirillov’s method gives ℝ2
j¼1

an algorithm how to describe G , not so many

l3 ða, bÞ
groups are known for which the unitary dual is j¼1 pg ds3 ðgÞ:

described explicitly. ℝ

We now compare the spectral properties of

Spectral Properties of Heisenberg Group T with the spectral properties of the restriction
Actions T to the center of H3(ℝ). The center is the sub-
The three-dimensional Heisenberg group H3(ℝ) is group {[0, 0, t] : t  ℝ}.
perhaps one of the simplest examples of non-
Abelian nilpotent Lie groups for which the orbit Proposition 1 (Danilenko (2014)) The maximal
method leads to a very concrete description of the spectral type of the flow (T[0,0,t]))t  ℝ contains the
unitary dual (Kirillov 2004). Recall that measure s1,2(ℝ2)d0 + s3. The corresponding
spectral multiplicity is ℝ2 l1,2 ds1,2 at the point t
1 a c ¼ 0 and the infinity if t  ℝ.
H3 ðℝÞ ¼ 0 1 b : a, b, c  ℝ : Theorem 11 (Danilenko (2014)) If (T[0,0,t]))t  ℝ
0 0 1 is ergodic then:

For simplicity, we will denote the matrix 1. s1,2(ℝ2\{(0, 0)}) ¼ 0, i.e., there are no non-
1 a c trivial one-dimensional representations in the
0 1 b by [a, b, c]. The unitary dual spectral decomposition of U. The maximal
0 0 1 spectral type of T equals the maximal spectral
type of the restriction of T to the center of
H 3 ðℝÞ is identified with ℝ2 ℝ endowed with
H3(ℝ) (modulo the natural identification);
the natural standard Borel structure. Every irre-
2. T is mixing (see also Ryzhikov (1994b, 1996));
ducible unitary representation of H3(ℝ) is unitar-
3. the weak closure of the group {Tg : g  H3(ℝ)}
ily equivalent to either a one-dimensional pa,b
in Aut(X, m) is the union of {Tg : g  H3(ℝ)}
with (a, b)  ℝ2 or an infinite dimensional pg in
and the weak closure of {T[0,0,t] : t  ℝ};
L2(ℝ, Leb), with g  ℝ ≔ ℝ\{0}, such that
4. if T is rigid, then (T[0,0,t]))t  ℝ is rigid.
In Danilenko (2014), there were constructed
pa,b ½a, b, c≔e2piðaaþbbÞ , explicit examples of mixing of all orders rank
pg ½a, b, cf ðxÞ≔e2piðcþbxÞ f ðx þ aÞ, f  L2 ðℝ, LebÞ: one (and hence zero entropy) actions T of the
Heisenberg group.
Now, given a measure-preserving H3(ℝ)- The concept of simplicity for ergodic H3(ℝ)-
action on a standard probability space (X, m), let actions is defined in a similar way as for the
U denote the corresponding Koopman represen- Abelian actions. A simple H3(ℝ)-action T has
tation in L2(X, m). Then there are a probability MSJ if the centralizer of the action is {T[0,0,t] :
measure s1,2 on ℝ2, a function l1,2 : ℝ2 ! ℕ a t  ℝ}. It was shown in Danilenko (2014) that the
probability measure s3 on ℝ, a function l3 : ℝ ! examples of mixing H3(ℝ)-actions constructed
ℕ such that there satisfy also the following:
138 Spectral Theory of Dynamical Systems

1. The flow (T[0,0,t]))t  ℝ is simple and the central- 1

U¼ pa,b 1 pg :
ða, bÞ  [1 pðGn Þ 06¼g  [1 x1 ℤ
izer of it equals the group {Tg : g  H3(ℝ)}. n¼1 n¼1 Gn

2. The transformation T[0,0,1] is simple and the

centralizer of it equals {Tg : g  H3(ℝ)}.
3. T has MSJ. An analogous decomposition is found also for

the case where [1 n¼1 pðGn Þ is closed in ℤ (see
2

As a corollary, we obtain examples of mixing Danilenko and Lemańczyk (2016) for details).
Poisson and mixing Gaussian (probability pre- Thus, we see that the maximal spectral type of
serving) actions of H3(ℝ). Heisenberg odometers is purely atomic.
For a decreasing sequence G ¼ ðGn Þ1 n¼1 of
lattices in H3(ℝ), we let SðGÞ≔[n¼1 pðGn Þ and
1
Heisenberg Odometers 1
xG ≔[1 n¼1 xGn ℤ . The following theorem (except
In Danilenko and Lemańczyk (2016), the authors
for the first claim) and the below remarks demon-
isolated a special class of ergodic H3(ℝ)-actions,
strate a drastic difference between H3(ℝ)-
called the odometer actions. Namely, let G1  G2
odometers and ℤ-odometers.
    be a sequence of lattices in H3(ℝ). Then, we
can associate a sequence of homogeneous H3(ℝ)-
Theorem 13 Two Heisenberg odometers T and
spaces intertwined with H3(ℝ)-equivariant maps:
T associated with decreasing sequences of lattices
G and G0 respectively are unitarily equivalent if and
H 3 ðℝÞ=G1 H 3 ðℝÞ=G2   :
only if S(G) ¼ S(G0) and xG ¼ x0G . The direct
Denote by X the projective limit of this sequence. product T  T 0 ≔ T g  T 0g is not spectrally
gG
Then X is a compact Polish G-space. Endow each
equivalent to any Heisenberg odometer. T  T 0
space H3(ℝ)/Gn with the Haar probability measure.
is ergodic if and only if S(G) \ S(G0) ¼ {0}. T 
The projective limit of the sequence of these measures
T0 is ergodic and has discrete maximal spectral
is a H3(ℝ)-invariant probability measure m. Of course,
type if and only if S(G) \ S(G0) ¼ {0} and xG \
the H3(ℝ)-action on (X, m) is ergodic. It is called the
x0G ¼ f0g.
Heisenberg odometer associated with ðGn Þ1 n¼1 . We
consider the Heisenberg odometers as non-
It was shown in Danilenko and Lemańczyk
commutative counterparts of the ergodic ℤ-actions
(2016) that Heisenberg odometers are not iso-
and ℝ-actions with pure point rational spectrum.
spectral, i.e., the unitary equivalence, in general,
A complete spectral decomposition of the Hei-
does not imply isomorphism for the underlying
senberg odometers is found in Danilenko and
H3(ℝ)-actions. It was also shown in Danilenko
Lemańczyk (2016). Denote by p : H3(ℝ) ! ℝ2
and Lemańczyk (2016) that Heisenberg odome-
the homomorphism [a, b, c] 7! (a, b). The kernel
ters are not spectrally determined: A Heisenberg
of this homomorphism is the center of the Heisen-
odometer is constructed which is unitarily equiv-
berg group. Given a lattice G in H3(ℝ), we denote
alent to an H3(ℝ)-action which is not isomorphic
by xG a positive real such that G \ Ker p ¼ {[0, 0,
to any Heisenberg odometer.
nxG]| n  ℤ}. Of course, p(G) is a lattice in ℝ2. If
p(G) ¼ A(ℤ2) for some matrix A  GL(2, ℝ), then
we denote by p(G) the dual lattice (A)1ℤ2 in ℝ2. On the “Finitely Dimensional” Part of the
Spectrum
Theorem 12 Let U stand for the Koopman uni- Suppose that G is an arbitrary locally compact
tary representation of the Heisenberg odometer second countable group. If G is not of type I,
associated with a sequence of lattices G1  G2  then G furnished with the Mackey-Borel structure

  . If [1
n¼1 pðGn Þ is not closed in ℝ , then
2
is a “bad” (not standard) Borel space. Then a
Spectral Theory of Dynamical Systems 139

decomposition of a unitary representation of shown later in Danilenko and Lemańczyk (2016)

G into irreducibles can be done in essentially that the multiplicity of each irreducible compo-
nonunique way (see Kirillov (2004)). Neverthe- nent in UT equals the dimension of this compo-
less, this “badness” is related only to the infinite- nent. It was also proved in Danilenko and
dimensional part of the spectrum. Thus, the union Lemańczyk (2016) that the normal H3(ℤ)-
1
n¼1 Gn ¼ G∖G1 is a “good” standard Borel
odometers are isospectral. Thus, the unitary
space. Thus, given a measure-preserving equivalence of the Koopman representations
G-action, we can study “the finitely dimensional implies isomorphism of the underlying normal
part” of the spectrum of the corresponding H3(ℤ)-odometers.
Koopman unitary representation of G.
This approach was used by Mackey in Mackey
(1964), where he made an attempt to extend the Future Directions
theory of actions with pure discrete spectrum to
non-Abelian groups. For that, he isolated a class We have already seen several cases, where spec-
of ergodic G-actions T for which the Koopman tral properties interact with measure-theoretic
representation UT decomposes into a (countable) properties of a system. Let us mention a few
family of finite-dimensional irreducible represen- more cases which require further research and
tations. We note that the family is uniquely deeper understanding.
defined by T. Mackey called such T an action We recall that the weak mixing property can be
with pure point spectrum. He established a struc- understood as a property complementary to dis-
ture for these actions. He showed that T has a pure crete spectrum (more precisely to the distality
point spectrum if and only if T is isomorphic to a (Furstenberg 1981)), or similarly mild mixing
G-action by rotations on a homogeneous space of property is complementary to rigidity. This can
G by a compact subgroup. However, in general, in be phrased quite precisely by saying that T is not
contrast with the Abelian case, the G-actions with weakly (mildly) mixing if and only if it has a non-
pure point spectrum are not necessarily isospectral trivial factor with discrete spectrum (it has a non-
even in the case of finite G (see (Todd 1950; trivial rigid factor). It has been a question for quite
Lemańczyk et al. 2002) Section “Rokhlin a long time if in a sense mixing can be “built” on
Cocycles” for counterexamples). the same principle. In other words, we seek a
In Lightwood et al. (2014), Lightwood, Şahin, certain “highly” non-mixing factor. It was quite
and Ugarcovici considered certain odometer surprising when in 2005 F. Parrreau (private com-
actions of the discrete Heisenberg group H3(ℤ) munication) gave the positive answer to this
≔ {[a, b, c] : a, b, c  ℤ}. This group is not of problem.
type I (in contrast with H3(ℝ)) because each sub-
Theorem 14 (Parreau) Assume that T is an
group of finite index in H3(ℤ) is non-Abelian.
ergodic automorphism of a standard probability
Hence, H3 ðℤÞ is a “bad” Borel space. On the Borel space ðX , B, mÞ: Assume moreover that T is
other hand, the (standard Borel) subspace not mixing. Then there exists a non-trivial factor
1
n¼1 H 3 ðℤÞn of it is explicitly described in (see below) of T which is disjoint from all mixing
Indukaev (2007). Consider now an H3(ℤ)- automorphisms.
odometer T generated by a decreasing sequence
G1  G2     of normal subgroups of finite index In fact, Parreau proved that each factor of
in H3(ℤ). We call such an odometer normal. Then, T given by B1 (r) (this s-algebra is described in
T has a pure point spectrum in the sense of Lemańczyk et al. (2000)), where U nTk ! J r , is
Mackey. Moreover, the full list of the finitely disjoint from all mixing transformations. This
dimensional irreducible unitary representations proof leads to some other results of the same
that occur in UT is found in Lightwood et al. type, for example: Assume that T is an ergodic
(2014) in terms of the sequence ðGn Þ1 n¼1 . It was automorphism of a standard probability Borel
140 Spectral Theory of Dynamical Systems

space. Assume that there exists a non-trivial auto- (see also Ryzhikov (2007)). In Lemańczyk and
morphism S with a singular spectrum which is not Parreau (2007), it is proved that some special
disjoint from T. Then T has a non-trivial factor flows considered in section “Special Flows,
which is disjoint from any automorphism with a Flows on Surfaces, and Interval Exchange Trans-
Lebesgue spectrum. formations” (including the von Neumann class,
The problem of spectral multiplicity of Carte- however, with a having unbounded partial quo-
sian products for “typical” transformation studied tients) have the SCS property. It is quite plausible
by Katok (2003a) and then its solution in Ageev that the SCS property is commonly seen for
(2008) which we already considered in section smooth flows on surfaces.
“The Multiplicity Function” lead to a study of A classical open problem is whether each ergo-
those T for which dic automorphism has a smooth model. While this
problem stays open even for so-called dyadic
ðCS Þ sðmÞ ⊥sðnÞ whenever m 6¼ n, adding machine (ergodic, discrete spectrum
automorphism having roots of unity of degree
where s ¼ sT just stands for the reduced maximal 2n, n  1, as eigenvalues), A. Katok suggested
spectral type of UT (which is constantly assumed many years ago that one can construct a
to be a continuous measure); see also Stepin’s Kronecker measure so that the corresponding
article (Stepin 1986). Gaussian system (ℤ-action (!)) has a smooth rep-
Usefulness of the above property (CS) in ergo- resentation on the torus. No “written” proof of this
dic theory was already shown in del Junco and fact yet appeared.
Lemańczyk (1992), where a spectral counterex- In Vershik (2011), A.M. Vershik sketches a
ample machinery was presented using the follow- proof of the fact (claimed by himself for decades)
ing observation: If A is a T1-invariant sub - s- that Pascal adic transformation is weakly mixing.
algebra such that the maximal spectral type on Earlier, X. Méla in his PhD showed that this
L2 ðA Þ is absolutely continuous with respect to sT, transformation is ergodic and has zero entropy.
then A is contained in one of the coordinate Moreover, in Janvresse and de la Rue (2004),
sub-s-algebras B . Based on that in del Junco Janvresse and de la Rue proved the LB property
and Lemańczyk (1992), it is shown how to con- of this map. No complete proof of weak mixing of
struct two weakly isomorphic actions which are Pascal adic transformation seems to exist in the
not isomorphic or how to construct two non- published literature.
disjoint automorphisms which have no common In Vershik (2005), Vershik proposed to study a
non-trivial factors (such constructions were previ- new equivalence between measure-preserving
ously known for so-called minimal self-joining automorphisms, called quasi-similarity. Two
automorphisms (Rudolph 1979)). See also automorphisms T on ðX , B, mÞ and S on
Tikhonov (2002) for extensions of those results ðY, C , nÞ are quasi-similar if there are Markov
to ℤd-actions. operators J : L2 ðX , B, mÞ ! L2 ðY , C, vÞ and
Prikhodko and Ryzhikov (2000) proved that K : L2 ðY , C, vÞ ! L2 ðX , B, mÞ, both with
the classical Chacon transformation enjoys the dense ranges, intertwining the corresponding
(CS) property. The SCS property defined in sec- Koopman operators UT and US. This new equiva-
tion “Glossary and Notation” is stronger than the lence is strictly stronger than spectral isomor-
(CS) condition above; the SCS property implies phism but strictly weaker than (weak)
that the corresponding Gaussian system SsT has a isomorphism as shown in Frączek and
simple spectrum. Ageev (2000) shows that Lemańczyk (2010) answering one of the ques-
Chacon transformation satisfies the SCS property; tions by Vershik. However, possible invariants
moreover, in Ageev (2008), he shows that the SCS for this new equivalence are not well understood.
property is satisfied generically and he gives a For example, in Frączek and Lemańczyk (2010),
construction of a rank one mixing SCS-system it is shown that an automorphism quasi-similar to
Spectral Theory of Dynamical Systems 141

a K-automorphism must also be K, but an intrigu- Bibliography

ing question remains whether Bernoulli property
is an invariant of quasi-similarity. As weak iso- Aaronson J (1979) Rational ergodicity, bounded rational
ergodicity, and some continuous measures on the circle.
morphism for finite multiplicity automorphisms is
Isr J Math 33(3–4):181–197
in fact isomorphism, another question is whether Aaronson J (1997) An introduction to infinite ergodic
we can have two non-isomorphic quasi-similar theory. Mathematical surveys and monographs,
automorphisms with simple spectrum (rank one) vol 50. American Mathematical Society, Providence
Aaronson J, Nadkarni MG (1987) L1 eigenvalues and L2
(see Problem 4 in Frączek and Lemańczyk
spectra of nonsingular transformations. Proc Lond
(2010)). See also Haase and Moriakov (2018) Math Soc 55(3):538–570
for the problem of Markov quasi-factors in the Aaronson J, Weiss B (2000) Remarks on the tightness of
class of Abramov automorphisms. cocycles. Colloq Math 84(85):363–376
Aaronson J, Hosseini M, Lemańczyk M (2014) IP-rigidity
Not too many zero entropy classical dynamical
and eigenvalue groups. Ergodic Theory Dyn Syst
systems with purely Lebesgue spectrum are 34:1057–1076
known: non-zero time automorphisms of Adams T (1998) Smorodinsky’ s conjecture on rank one
horocycle flows and their smooth time changes systems. Proc Am Math Soc 126:739–744
Adams T (2015) Tower multiplexing and slow weak
and even factor of special Gaussian systems. Can
mixing. Colloq Math 138:47–72
we produce a Poissonian suspension over a con- Ageev ON (1988) Dynamical systems with a Lebesgue
servative infinite measure-preserving automor- component of even multiplicity in the spectrum. Math
phism having purely Lebesgue spectrum? USSR Sb (N.S.) (Russ) 136(178):307–319, 430; trans-
lation in Math USSR Sb 64 (1989), 305–317
Katok and Thouvenot (private communica-
Ageev ON (1999) On ergodic transformations with homo-
tion) considered systems called infinitely divisible geneous spectrum. J Dyn Control Syst 5:149–152
(ID). These are systems T on ðX , B, mÞ which Ageev ON (2000) On the spectrum of Cartesian powers of
have a family of factors Bo indexed by o  [1 n¼0
classical automorphisms. Mat Zametki (Russ)
68:643–647; translation in Math Notes
f0,1gn ðBe ¼ BÞ such that Bo0 ⊥Bo1 , Bo0 _ Bo1 68 (2000):547–551
¼ Bo and for each   {0, 1}ℕ, \n  ℕ B½0,n Ageev ON (2001) On the multiplicity function of generic
¼ 0, X . They showed (unpublished) that there group extensions with continuous spectrum. Ergodic


Theory Dyn Syst 21:321–338

are discrete spectrum transformations which are
Ageev ON (2005) The homogeneous spectrum problem in
ID and that there are rank one transformations ergodic theory. Invent Math 160:417–446
with continuous spectra which are also ID Ageev ON (2008) Mixing with staircase multiplicity func-
(clearly Gaussian systems are ID). In Lemańczyk tion. Ergodic Theory Dyn Syst 28:1687–1700
Alexeyev VM (1958) Existence of bounded function of the
et al. (2011), it is proved that dynamical systems
maximal spectral type. Vestnik Mosc Univ 5:13–15 and
coming from stationary ID processes are factors of Ergodic Theory Dyn Syst 2 (1982):259–261
ID automorphisms; moreover, ID automorphisms Alpern S (1985) Conjecture: in general a mixing transfor-
are disjoint from all systems having the SCS mation is not two-fold mixing. Ann Probab
13:310–313
property. It would be nice to decide whether
Anosov DV (2003) On spectral multiplicities in ergodic
Koopman representations associated with ID theory. Institut im. V.A., Steklova
automorphisms satisfy the Kolmogorov group Anzai H (1951) Ergodic skew product transformations on
property. the torus. Osaka J Math 3:83–99
Arni N (2005) Spectral and mixing properties of actions of
While some lacunary sequences have realiza-
amenable groups. Electron Res Announc AMS
tion as rigidity sequences for weakly mixing auto- 11:57–63
morphisms, it would be interesting to determine Arnold VI (1991) Topological and ergodic properties of
what happens in the non-lacunary case. An espe- closed 1-forms with incommensurable periods.
Funktsional Anal Prilozhen (Russ) 25:1–12
cially interesting case is when we consider the
Avila A, Forni G (2007) Weak mixing for interval
non-lacunary multiplicative semigroup {2i3j : i, j exchange transformations and translation flows. Ann
 0}. Is the corresponding sequence a rigidity Math 165:637–664
sequence? What about {2k + 3‘ : k, ‘  0} Avila A, Forni G, Ulcigrai C (2011) Mixing for the time-
changes of Heisenberg nilflows. J Diff Geom
(Bergelson et al. 2014).?
89:369–410
142 Spectral Theory of Dynamical Systems

Avila A, Forni G, Ravotti D, Ulcigrai C (2019) Mixing for Danilenko AI (2012) New spectral multiplicities for
smooth time-changes of general nilflows, preprint, mixing transformations. Ergodic Theory Dyn Syst
arXiv: 1905.11628 32:517–534
Badea C, Grivaux S, Matheron E (2018) Rigidity Danilenko AI (2013) A survey on spectral multiplicities of
sequences, Kazhdan sets and group topologies on the ergodic actions. Ergodic Theory Dyn Syst 33:81–117
integers, arXiv:1812.09014 Danilenko AI (2014) Mixing actions of the Heisenberg
Balister P, Bollobas B, Morris R, Suhasrabudhe J, Tiba group. Ergodic Theory Dyn Syst 34:1142–1167
M (2019) Flat Littlewood polynomials exist, arXiv Danilenko AI, Lemańczyk M (2005) A class of multipliers
1907.09464 for W⊥. Isr J Math 148:137–168
Bashtanov AI (2013) Generic mixing transformations are Danilenko AI, Lemańczyk M (2013) Spectral multiplicities
rank 1. Math Notes 93:209–216 for ergodic flows. Discrete Contin Dyn Syst
Bashtanov AI (2016) Conjugacy classes are dense in the 33:4271–4289
space of mixing Zd-actions. Math Notes 99:9–23 Danilenko AI, Lemańczyk M (2016) Odometer actions of
Baxter JR (1971) A class of ergodic transformations hav- the Heisenberg group. J Anal Math 128:107–157
ing simple spectrum. Proc Am Math Soc 27:275–279 Danilenko AI, Park KK (2002) Generators and Bernoullian
Bergelson V, del Junco A, Lemańczyk M, Rosenblatt factors for amenable actions and cocycles on their
J (2014) Rigidity and non-recurrence along sequences. orbits. Ergodic Theory Dyn Syst 22:1715–1745
Ergodic Theory Dyn Syst 34:14641502 Danilenko AI, Ryzhikov VV (2010) Spectral multiplicities
Björklund M, Einsiedler M, Gorodnik A (2020) Quantita- for infinite measure preserving transformations. Funct
tive multiple mixing. J. Eur. Math. Soc. (JEMS) 22 Anal Appl 44:161–170
(5):1475–1529 Danilenko AI, Ryzhikov VV (2011) Mixing constructions
Bourgain J (1993) On the spectral type of Ornstein’s class with infinite invariant measure and spectral multiplici-
of transformations. Isr J Math 84:53–63 ties. Ergodic Theory Dyn Syst 31:853–873
Bourgain J (2013) On the correlation of the Moebius func- Danilenko AI, Solomko AV (2010) Ergodic Abelian
tion with rank-one systems. J Anal Math 120:105–130 actions with homogenous spectrum. Contemporary
Cambanis S, Podgrski K, Weron A (1995) Chaotic behav- mathematics, vol 532. American Mathematical Society,
iour of infinitely divisible processes. Stud Math Providence, pp 137–148
115:109–127 de la Rue T (1996) Systèmes dynamiques gaussiens
Cecchi PA, Tiedra de Aldecoa R (2016) Furstenberg trans- d’entropie nulle, l^achement et non l^achement
formations on Cartesian products of infinite- Bernoulli. Ergodic Theory Dyn Syst 16:379–404
dimensional tori. Potent Anal 44:4351 de la Rue T (1998a) Rang des systèmes dynamiques
Chacon RV (1970) Approximation and spectral multiplic- gaussiens. Isr J Math 104:261–283
ity. In: Dold A, Eckmann B (eds) Contributions to de la Rue T (1998b) L’ergodicité induit un type spectral
ergodic theory and probability. Springer, Berlin, maximal équivalent à la mesure de Lebesgue. Ann Inst
pp 18–27 Henri Poincaré Probab Statist 34:249–263
Chaika J, Eskin A (2018) Self-joinings for 3-IET’s, de la Rue T (1998c) L’induction ne donne pas toutes les
arXiv:1805.11167v2 mesures spectrales. Ergodic Theory Dyn Syst
Chaika J, Wright A (2019) A smooth mixing flow on a 18:1447–1466
surface with nondegenerate fixed points. J Am Math de la Rue T (2004) An extension which is relatively two-
Soc 32:81–117 fold mixing but not threefold mixing. Colloq Math
Chaika J, Frączek K, Kanigowski A, Ulcigrai C (2019) 101:271–277
Singularity of the spectrum for smooth area-preserving de la Rue T (2009) Joinings in ergodic theory. In: Ency-
flows in genus two, preprint clopedia of complexity and system science. Springer,
Choe GH (1990) Products of operators with singular con- New York
tinuous spectra. In: Operator theory: operator algebras del Junco A (1976) Transformations with discrete spec-
and applications, Part 2 (Durham, NH, 1988). Proceed- trum are stacking transformations. Can J Math
ings of symposia in pure mathematics, vol 51. Ameri- 28:836–839
can Mathematical Society, Providence, pp 65–68 del Junco A (1977) A transformation with simple spectrum
Connes A, Woods E (1985) Approximately transitive flows which is not of rank one. Can J Math 29:655–663
and ITPFI factors. Ergodic Theory Dyn Syst 5:203–236 del Junco A (1981) Disjointness of measure-preserving
Cornfeld IP, Fomin SV, Sinai YG (1982) Ergodic theory. transformations, minimal self-joinings and category.
Springer, New York Prog Math 10:81–89
Creutz D, Silva C (2010) Mixing on rank-one transforma- del Junco A, Lemańczyk M (1992) Generic spectral prop-
tions. Stud Math 199:43–72 erties of measure-preserving maps and applications.
Danilenko AI (2006) Explicit solution of Rokhlin’s prob- Proc Am Math Soc 115:725–736
lem on homogeneous spectrum and applications. Ergo- del Junco A, Lemańczyk M (1999) Simple systems are
dic Theory Dyn Syst 26:1467–1490 disjoint with Gaussian systems. Stud Math 133:249–256
Danilenko AI (2010) On new spectral multiplicities for del Junco A, Lemańczyk M (2005) Joinings of distally
ergodic maps. Stud Math 197:57–68 simple systems, preprint
Spectral Theory of Dynamical Systems 143

del Junco A, Rudolph D (1987) On ergodic actions whose Fayad B, Krikorian R (2018) Some questions around
self-joinings are graphs. Ergodic Theory Dyn Syst quasi-periodic dynamics, arXiv:1809.10375
7:531–557 Fayad B, Thouvenot J-P (2014) On the convergence to 0 of
Derriennic Y, Frączek K, Lemańczyk M, Parreau F (2008) mnx mod 1. Acta Arith 165:327332
Ergodic automorphisms whose weak closure of off- Fayad B, Windsor A (2007) A dichotomy between discrete
diagonal measures consists of ergodic self-joinings. and continuous spectrum for a class of special flows
Colloq Math 110:81–115 over rotations. J Modern Dyn 1:107–122
Dooley A, Golodets VY (2002) The spectrum of Fayad B, Katok AB, Windsor A (2001) Mixed spectrum
completely positive entropy actions of countable ame- reparameterizations of linear flows on T2, Dedicated to
nable groups. J Funct Anal 196:1–18 the memory of I. G. Petrovskii on the occasion of his
Dooley A, Quas A (2005) Approximate transitivity for 100th anniversary. Mosc Math J 1:521–537
zero-entropy systems. Ergodic Theory Dyn Syst Fayad B, Forni G, Kanigowski A (2016) Lebesgue spec-
25:443–453 trum of countable multiplicity for conservative flows
Downarowicz T, Kwiatkowski J (2000) Spectral isomor- on the torus, submitted
phism of Morse flows. Fundam Math 163:193–213 Feldman J (1976) New K-automorphisms and a problem of
Downarowicz T, Kwiatkowski J (2002) Weak Closure Kakutani. Isr J Math 24:16–38
theorem fails for Zd-actions. Stud Math 153:115–125 Ferenczi S (1984) Systèmes localement de rang un. Ann
Eisner T, Grivaux S (2011) Hilbertian Jamison sequences and Inst Henri Poincaré Probab Statist 20:35–51
rigid dynamical systems. J Funct Anal 261:2013–2052 Ferenczi S (1985) Systèmes de rang un gauche. Ann Inst
El Abdalaoui H (2000) On the spectrum of the powers of Henri Poincaré Probab Statist 21:177–186
Ornstein transformations. Sankhya Ser A 62:291–306. Ferenczi S (1997) Systems of finite rank. Colloq Math
Ergodic theory and harmonic analysis (Mumbai, 1999) 73:35–65
El Abdalaoui H (2007) A new class of rank 1 transforma- Ferenczi S, Lemańczyk M (1991) Rank is not a spectral
tions with singular spectrum. Ergodic Theory Dyn Syst invariant. Stud Math 98:227–230
27:1541 Ferenczi S, Holton C, Zamboni LQ (2004) Structure of
El Abdalaoui H (2015) Ergodic Banach problem on simple three-interval exchange transformations III: ergodic
Lebesgue spectrum and flat polynomials, arXiv and spectral properties. J Anal Math 93:103–138
1508.06439v4 Ferenczi S, Holton C, Zamboni LQ (2005) Joinings of
El Abdalaoui H, Parreau F, Prikhodko A (2006) A new three-interval exchange transformations. Ergodic The-
class of Ornstein transformations with singular spec- ory Dyn Syst 25:483–502
trum. Ann Inst Henri Poincaré Probab Stat 42:671–681 Filipowicz I (1997) Product Zd-actions on a Lebesgue
El Abdalaoui H, Lemańczyk M, de la Rue T (2014) On space and their applications. Stud Math 122:289–298
spectral disjointness of powers for rankone transforma- Flaminio L, Forni G (2003) Invariant distributions and time
tions and Möbius orthogonality. J Funct Anal averages for horocycle flows. Duke Math
266:284–317 J 119(3):465–526
El Abdalaoui H, Kasjan S, Lemańczyk M (2016) 0-1 Flaminio L, Forni G (2019) Orthogonal powers and
sequences of the Thue-Morse type and Sarnaks conjec- Möbius conjecture for smooth time changes of
ture. Proc Am Math Soc 144:161–176 horocycle flows. Electron Res Announc Math Sci
Fayad B (2001a) Partially mixing and locally rank 1 smooth 26:16–23
transformations and flows on the torus Td, d3. J Lond Foiaş C, Stratila S (1968) Ensembles de Kronecker dans la
Math Soc 64(2):637–654 théorie ergodique. C R Acad Sci Paris, Ser A-B 267:
Fayad B (2001b) Polynomial decay of correlations for a A166–A168
class of smooth flows on the two torus. Bull Soc Math Forni G, Kanigowski A (2017) Time-changes of Heisen-
Fr 129:487–503 berg nilflows, preprint, arXiv:1711.05543
Fayad B (2002a) Skew products over translations on Td, Forni G, Kanigowski A (2020) Multiple mixing and
d2. Proc Am Math Soc 130:103–109 disjointness for time changes of bounded-type Heisen-
Fayad B (2002b) Analytic mixing reparametrizations of berg nilflows. J Éc polytech Math 7:63–91
irrational flows. Ergodic Theory Dyn Syst 22:437–468 Forni G, Ulcigrai C (2012) Time-changes of horocycle
Fayad B (2005) Rank one and mixing differentiable flows. flows. J Modern Dyn 6(2):251–273
Invent Math 160:305–340 Frączek K (1997) On a function that realizes the maximal
Fayad B (2006) Smooth mixing flows with purely singular spectral type. Stud Math 124:1–7
spectra. Duke Math J 132:371–391 Frączek K (2000) Circle extensions of Zd-rotations on the
Fayad B, Kanigowski A (2015) Rigidity times for a weakly d-dimensional torus. J Lond Math Soc 61(2):139–162
mixing dynamical system which are not rigidity times Frączek K, Lemańczyk M (2003) On symmetric logarithm
for any irrational rotation. Ergodic Theory Dyn Syst and some old examples in smooth ergodic theory.
35:2529–2534 Fundam Math 180:241–255
Fayad B, Kanigowski A (2016) Multiple mixing for a class Frączek K, Lemańczyk M (2004) A class of special flows
of conservative surface flows. Invent Math over irrational rotations which is disjoint from mixing
203(2):555–614 flows. Ergodic Theory Dyn Syst 24:1083–1095
144 Spectral Theory of Dynamical Systems

Frączek K, Lemańczyk M (2005) On disjointness proper- Hasselblatt B, Katok AB (2002) Principal Structures.
ties of some smooth flows. Fundam Math 185:117–142 Handbook of dynamical systems, vol 1A. North-
Frączek K, Lemańczyk M (2006) On mild mixing of spe- Holland, Amsterdam, pp 1–203
cial flows over irrational rotations under piecewise Helson H (1986) Cocycles on the circle. J Oper Theory
smooth functions. Ergodic Theory Dyn Syst 16:189–199
26:719–738 Helson H, Parry W (1978) Cocycles and spectra. Arkiv
Frączek K, Lemańczyk M (2010) A note on quasi- Math 16:195–206
similarity of Koopman operators. J Lond Math Soc Herman M (1979) Sur la conjugaison différentiable des
82(2):361–375 difféomorphismes du cercle à des rotations. Publ
Frączek K, Lemańczyk M, Lesigne E (2007) Mild mixing Math IHES 49:5–234
property for special flows under piecewise constant Host B (1991) Mixing of all orders and pairwise indepen-
functions. Discrete Contin Dyn Syst 19:691–710 dent joinings of systems with singular spectrum. Isr
Friedman NA, Ornstein DS (1973) Ergodic transforma- J Math 76:289–298
tions induce mixing transformations. Adv Math Host B, Méla F, Parreau F (1991) Non-singular transfor-
10:147–163 mations and spectral analysis of measures. Bull Soc
Furstenberg H (1967) Disjointness in ergodic theory, min- Math Fr 119:33–90
imal sets, and a problem of Diophantine approximation. Indukaev FK (2007) The twisted Burnside theory for the
Math Syst Theory 1:1–49 discrete Heisenberg group and for the wreath
Furstenberg H (1981) Recurrence in ergodic theory and products of some groups. Mosc Univ Math Bull
combinatorial number theory. Princeton University 62:219–227
Press, Princeton Iwanik A (1991) The problem of Lp-simple spectrum for
Furstenberg H, Weiss B (1978) The finite multipliers of ergodic group automorphisms. Bull Soc Math Fr
infinite ergodic transformations. Lect Notes Math 119:91–96
668:127–132 Iwanik A (1992) Positive entropy implies infinite Lp-
Glasner E (1994) On the class of multipliers for W⊥. multiplicity for p>1. Ergodic theory and related topics,
Ergodic Theory Dyn Syst 14:129–140 III (Gstrow, 1990). Lecture notes in mathematics,
Glasner E (2003) Ergodic theory via joinings. Mathemat- vol 1514. Springer, Berlin, pp 124–127
ical surveys and monographs, vol 101. AMS, Iwanik A (1997) Anzai skew products with Lebesgue
Providence component of infinite multiplicity. Bull Lond Math
Glasner E, Weiss B (1989) Processes disjoint from weak Soc 29:195–199
mixing. Trans Am Math Soc 316:689–703 Iwanik A, de Sam Lazaro J (1991) Sur la multiplicité Lp
Glimm JG (1960) On a certain class of operator algebras. d’un automorphisme gaussien. C R Acad Sci Paris Sér
Trans Am Math Soc 95:318–340 I 312:875–876
Goodson GR (1999) A survey of recent results in the Iwanik A, Lemańczyk M, Rudolph D (1993) Absolutely
spectral theory of ergodic dynamical systems. J Dyn continuous cocycles over irrational rotations. Isr J Math
Control Syst 5:173–226 83:73–95
Goodson GR, Kwiatkowski J, Lemańczyk M, Liardet Iwanik A, Lemańczyk M, de Sam Lazaro J, de la Rue
P (1992) On the multiplicity function of ergodic T (1997) Quelques remarques sur les facteurs des
group extensions of rotations. Stud Math 102:157–174 systèmes dynamiques gaussiens. Stud Math
Griesmer J (2019) Recurrence, rigidity, and popular differ- 125:247–254
ences. Ergodic Theory Dyn Syst 39:1299 (to appear) Iwanik A, Lemańczyk M, Mauduit C (1999) Piecewise
Gromov AL (1991) Spectral classification of some types of absolutely continuous cocycles over irrational rota-
unitary weighted shift operators. Algebra Anal (Russ) tions. J Lond Math Soc 59(2):171–187
3:62–87; translation in St. Petersburg Math Janicki A, Weron A (1994) Simulation and chaotic behav-
J 3 (1992):997–1021 ior of a-stable stochastic processes. Monographs and
Guenais M (1998) Une majoration de la multiplicité textbooks in pure and applied mathematics, vol 178.
spectrale d’opérateurs associés a des cocycles réguliers. Marcel Dekker, New York
Isr J Math 105:263–284 Janvresse E, de la Rue T (2004) The Pascal adic transfor-
Guenais M (1999) Morse cocycles and simple Lebesgue mation is loosely Bernoulli. Ann Inst Henri Poincaré
spectrum. Ergodic Theory Dyn Syst 19:437–446 Probab Statist 40:133–139
Guenais M, Parreau F (2005) Valeurs propres de trans- Janvresse E, de la Rue T, Ryzhikov VV (2012) Around
formations liées aux rotations irrationnelles et aux King’s rank one theorems: flows and Zn-actions. In:
fonctions en escalier, preprint Dynamical systems and group actions. Contemporary
Haase M, Moriakov N (2018) On systems with quasi- mathematics, vol 567. American Mathematical Society,
discrete spectrum. Stud Math 241:173–199 Providence, pp 143–161
Hahn F, Parry W (1968) Some characteristic properties of Janvresse E, Prikhod’ko AA, de la Rue T, Ryzhikov VV
dynamical systems with quasi-discrete spectra. Math (2015) Weak limits of powers of Chacon’s automor-
Syst Theory 2:179–190 phism. Ergodic Theory Dyn Syst 35:128141
Spectral Theory of Dynamical Systems 145

Janvresse E, Roy E, de la Rue T (2017) Poisson suspen- Khanin KM, Sinai YG (1992) Mixing of some classes of
sions and Sushis. Ann Sci École Norm Supnérieure special flows over rotations of the circle. Funct Anal
50(4):1301–1334 Appl 26:155–169
Kachurovskii AG (1990) A property of operators gener- King J (1986) The commutant is the weak closure of the
ated by ergodic automorphisms. Optimizatsiya (Russ) powers, for rank-1 transformations. Ergodic Theory
47(64):122–125 Dyn Syst 6:363–384
Kalikow S (1984) Two fold mixing implies three fold King JL (1988) Joining-rank and the structure of finite rank
mixing for rank one transformations. Ergodic Theory rank mixing transformations. J Anal Math 51:182–227
Dyn Syst 4:237–259 King JL (2001) Flat stacks, joining-closure and genericity,
Kamae T. Spectral properties of automata generating preprint
sequences, unpublished preprint Kirillov AA (2004) Lectures on the orbit method. Series:
Kamiński B (1981) The theory of invariant partitions for Graduate studies in mathematics, vol 64. American
Zd-actions. Bull Acad Polon Sci Ser Sci Math Mathematical Society, Providence
29:349–362 Klemes I (1996) The spectral type of the staircase transfor-
Kanigowski A, de la Rue T (2019) Product of two staircase mation. Tohoku Math J 48:247–248
rank one transformations that is not loosely Bernoulli, Klemes I, Reinhold K (1997) Rank one transformations
accepted in J. d’Analyse Math., arXiv:1812.08027 with singular spectral type. Isr J Math 98:1–14
Kanigowski A, Kułaga-Przymus J (2016) Ratner’s prop- Kočergin AV (1972) On the absence of mixing in special
erty and mild mixing for smooth flows on surfaces. flows over the rotation of a circle and in flows on a two-
Ergodic Theory Dyn Syst 36:2512–2537 dimensional torus. Dokl Akad Nauk SSSR (Russ)
Kanigowski A, Kułaga-Przymus J, Ulcigrai C (2019) Mul- 205:949–952
tiple mixing and parabolic divergence in smooth area- Kochergin AV (1975) Mixing in special flows over
preserving flows on higher genus surfaces. J Eur Math rearrangement of segments and in smooth flows on
Soc (JEMS) 21(12):3797–3855 surfaces. Math-USSR Acad Sci 25:441–469
Kanigowski A, Lemańczyk M, Ulcigrai C (2018) On Kochergin AV (1976a) Non-degenerate saddles and
disjointness properties of some parabolic flows, pre- absence of mixing. Mat Zametky (Russ) 19:453–468
print, arXiv:1810.11576v1 Kochergin AV (1976b) On the homology of function over
Katok AB (1977) Monotone equivalence in ergodic theory. dynamical systems. Dokl Akad Nauk SSSR
Izv Akad Nauk SSSR Ser Math (Russ) 41:104–157 231:795–798
Katok AB (1980) Interval exchange transformations and Kochergin AV (2002) A mixing special flow over a rotation
some special flows are not mixing. Isr J Math of the circle with an almost Lipschitz function. Sb Math
35:301–310 193:359–385
Katok AB (2001) Cocycles, cohomology and combinato- Kochergin AV (2004) Nondegenerate fixed points and
rial constructions in ergodic theory. In: Collaboration mixing in flows on a two-dimensional torus. II. Mat
with Robinson EA Jr, Proceedings of symposium on Sb (Russ) 195:15–46
pure mathematics, vol 69. Smooth ergodic theory and Koopman BO (1931) Hamiltonian systems and transfor-
its applications, Seattle, 1999. American Mathematical mations in Hilbert space. Proc Natl Acad Sci USA
Society, Providence, pp 107–173 17:315–318
Katok AB (2003a) Constructions in ergodic theory. Kuipers L, Niederreiter H (1974) Uniform distribution of
Unpublished lecture notes sequences. Wiley, New York
Katok AB (2003b) Combinatorial constructions in ergodic Kułaga J, Parreau F (2012) Disjointness properties for
theory and dynamics. University lecture series, vol 30. Cartesian products of weakly mixing systems. Colloq
American Mathematical Society, Providence Math 128:153–177
Katok A, Lemańczyk M (2009) Some new cases of reali- Kułaga-Przymus J, Lemańczyk M (2019) Sarnak’s conjec-
zation of spectral multiplicity function for ergodic ture from ergodic theory point of view. In: Encyclope-
transformations. Fundam Math 206:185–215 dia of complexity and system science
Katok AB, Stepin AM (1967) Approximations in ergodic Kushnirenko AG (1974) Spectral properties of some
theory. Uspekhi Mat Nauk (Russ) 22(137):81–106 dynamical systems with polynomial divergence of
Katok A, Thouvenot J-P (2006) Spectral properties and orbits. Vestn Mosk Univ 1–3:101–108
combinatorial constructions in Ergodic theory. In: Kwiatkowski J (1981) Spectral isomorphism of Morse
Handbook of dynamical systems, vol 1B. Elsevier dynamical systems. Bull Acad Polon Sci Sér Sci Math
B. V., Amsterdam, pp 649–743 29:105–114
Katok AB, Zemlyakov AN (1975) Topological transitivity Kwiatkowski J Jr, Lemańczyk M (1995) On the multiplic-
of billiards in polygons. Mat Zametki (Russ) ity function of ergodic group extensions. II. Stud Math
18:291–300 116:207–215
Keane M (1968) Generalized Morse sequences. Z Wahr Kwiatkowski J, Lacroix Y (1997) Multiplicity, rank pairs.
Verw Gebiete 10:335–353 J Anal Math 71:205–235
Keane M (1975) Interval exchange transformations. Math Lemańczyk M (1988) Toeplitz Z2–extensions. Ann Inst
Z 141:25–31 Henri Poincaré 24:1–43
146 Spectral Theory of Dynamical Systems

Lemańczyk M (1996) Introduction to ergodic theory from Maruyama G (1970) Infinitely divisible processes. Theory
the point of view of the spectral theory. Lecture notes of Probab Appl 15(1):1–22
the tenth KAIST mathematics workshop, Taejon, Masur H (1982) Interval exchange transformations and
pp 1–153 measured foliations. Ann Math 115:169–200
Lemańczyk M (2000) Sur l’absence de mélange pour des Mathew J, Nadkarni MG (1984) Measure-preserving trans-
flots spéciaux au dessus d’une rotation irrationnelle. formation whose spectrum has Lebesgue component of
Colloq Math 84(85):29–41 multiplicity two. Bull Lond Math Soc 16:402–406
Lemańczyk M (2009) Spectral theory of dynamical sys- Medina H (1994) Spectral types of unitary operators aris-
tems. In: Encyclopedia of complexity and system sci- ing from irrational rotations on the circle group. Mich
ence. Springer, New York, pp 8554–8575 Math J 41:39–49
Lemańczyk M, de Sam Lazaro J (1997) Spectral analysis of Mentzen MK (1988) Some examples of automorphisms
certain compact factors for Gaussian dynamical sys- with rank r and simple spectrum. Bull Pol Acad Sci
tems. Isr J Math 98:307–328 7–8:417–424
Lemańczyk M, Lesigne E (2001) Ergodicity of Rokhlin Nadkarni MG (1998) Spectral theory of dynamical sys-
cocycles. J Anal Math 85:43–86 tems. Hindustan Book Agency, New Delhi
Lemańczyk M, Mauduit C (1994) Ergodicity of a class of Newton D (1966) On Gaussian processes with simple
cocycles over irrational rotations. J Lond Math Soc spectrum. Z Wahr Verw Gebiete 5:207–209
49:124–132 Newton D, Parry W (1966) On a factor automorphism of a
Lemańczyk M, Parreau F (2003) Rokhlin extensions and normal dynamical system. Ann Math Statist
lifting disjointness. Ergodic Theory Dyn Syst 37:1528–1533
23:1525–1550 Ornstein D (1970) On the root problem in ergodic theory.
Lemańczyk M, Parreau F (2007) Special flows over irratio- In: Proceedings of 6th Berkeley symposium on mathe-
nal rotation with simple convolution property, preprint matical statistics and probability. University California
Lemańczyk M, Parreau F (2012) Lifting mixing properties Press, Berkeley, pp 348–356
by Rokhlin cocycles. Ergodic Theory Dyn Syst Ornstein D, Weiss B (1987) Entropy and isomorphism
32:763–784 theorems for actions of amenable groups. J Anal Math
Lemańczyk M, Sikorski A (1987) A class of not local rank 48:1–141
one automorphisms arising from continuous substitu- Ornstein D, Rudolph D, Weiss B (1982) Equivalence of
tions. Probab Theory Relat Fields 76:421–428 measure preserving transformations. Mem Am Math
Lemańczyk M, Wasieczko M (2006) A new proof of Soc 37(262):1–116
Alexeyev’s theorem, preprint Parreau F (2000) On the Foiaş and Stratila theorem. In:
Lemańczyk M, Wysokińska M (2007) On analytic flows Proceedings of the conference on ergodic theory, Toruń
on the torus which are disjoint from systems of proba- Parreau F, Roy E (2015) Prime Poisson suspensions. Ergo-
bilistic origin. Fundam Math 195:97–124 dic Theory Dyn Syst 35:2216–2230
Lemańczyk M, Parreau F, Thouvenot J-P (2000) Gaussian Parry W (1970) Spectral analysis of G-extensions of
automorphisms whose ergodic self–joinings are Gauss- dynamical systems. Topology 9:217–224
ian. Fundam Math 164:253–293 Parry W (1981) Topics in ergodic theory. Cambridge tracts
Lemańczyk M, Thouvenot J-P, Weiss B (2002) Relative in mathematics, vol 75. Cambridge University Press,
discrete spectrum and joinings. Monatsh Math Cambridge/New York
137:57–75 Petersen K (1983) Ergodic theory. Cambridge University
Lemańczyk M, Mentzen MK, Nakada H (2003) Semi- Press, Cambridge
simple extensions of irrational rotations. Stud Math Prikhodko AA (2013) Littlewood polynomials and their
156:31–57 applications to the spectral theory of dynamical sys-
Lemańczyk M, Parreau F, Roy E (2011) Systems with tems. Mat Sb 204:135–160, translation in Sb Math
simple convolutions, distal simplicity and disjointness 204 (2013)
with infinitely divisible systems. Proc Am Math Soc Prikhodko AA, Ryzhikov VV (2000) Disjointness of the
139:185–199 convolutions for Chacon’s automorphism. Dedicated to
Leonov VP (1960) The use of the characteristic functional the memory of Anzelm Iwanik. Colloq Math
and semi–invariants in the ergodic theory of stationary 84(85):67–74
processes. Dokl Akad Nauk SSSR 133:523–526. Sov Quefelec M (1988) Substitution dynamical systems – spec-
Math 1 (1960):878–881 tral analysis. Lecture notes in mathematics, vol 1294.
Lightwood S, Sahin A, Ugarcovici I (2014) The structure Springer, Berlin
and the spectrum of Heisenberg odometers. Proc Am Ratner M (1978) Horocycle flows are loosely Bernoulli. Isr
Math Soc 142:2429–2443 J Math 31:122–132
Mackey GW (1964) Ergodic transformation groups with a Ratner M (1979) The Cartesian square of the horocycle
pure point spectrum. Ill J Math 8:593–600 flow is not loosely Bernoulli. Isr J Math 34(1–2):72–96
Marcus B (1977) Ergodic properties of horocycle flows for Ratner M (1983) Horocycle flows, joinings and rigidity of
surfaces of negative curvature. Ann Math 105(2):81–105 products. Ann Math 118:277–313
Spectral Theory of Dynamical Systems 147

Ratner M (1986) Rigidity of time changes for horocycle Ryzhikov VV (1994a) The absence of mixing in special
flows. Acta Math 156:1–32 flows over rearrangements of segments. Mat Zametki
Ratner M (1987) Rigid reparametrizations and cohomol- (Russ) 55:146–149
ogy for horocycle flows. Invent Math 88(2):341–374 Ryzhikov VV (1994b) Skew products and multiple mixing
Rauzy G (1979) Echanges d’intervalles et transformations of dynamical systems. Russ Math Surv 49:170–171
induites. Acta Arith 34:315–328 Ryzhikov VV (1996) Stochastic intertwinings and multiple
Ravotti D (2017) Quantitative mixing for locally Hamilto- mixing of dynamical systems. J Dyn Control Syst
nian flows with saddle loops on compact surfaces. Ann 2:1–19
Henri Poincaré 18(12):3815–3861 Ryzhikov VV (1999) Transformations having homoge-
Ravotti D (2019) Mixing for suspension flows over skew- neous spectra. J Dyn Control Syst 5:145–148
translations and time-changes of quasi-abelian filiform Ryzhikov VV (2000) The Rokhlin problem on multiple
nilflows. Ergodic Theory Dyn Syst, published online at mixing in the class of actions of positive local rank.
https://doi.org/10.1017/etds.2018.19 Funktsional Anal Prilozhen (Russ) 34:90–93
Robinson EA Jr (1983) Ergodic measure preserving trans- Ryzhikov VV (2007) Weak limits of powers, simple spec-
formations with arbitrary finite spectral multiplicities. trum symmetric products and mixing rank one con-
Invent Math 72:299–314 structions. Math Sb 198:733–754
Robinson EA Jr (1986) Transformations with highly non- Ryzhikov VV (2009) Spectral multiplicities and asymp-
homogeneous spectrum of finite multiplicity. Isr J Math totic operator properties of actions with invariant mea-
56:75–88 sure. Sb Math 200:1833–1845
Robinson EA Jr (1988) Nonabelian extensions have non- Ryzhikov VV (2014) On spectral multiplicities of Gauss-
simple spectrum. Compos Math 65:155–170 ian actions, arXiv 1406.3321
Robinson EA Jr (1992) A general condition for lifting Ryzhikov VV, Thouvenot J-P (2006) Disjointness, divisi-
theorems. Trans Am Math Soc 330:725–755 bility, and quasi-simplicity of measure-preserving
Rosiński J, Żak T (1996) Simple condition for mixing of actions. Funktsional Anal Prilozhen (Russ) 40:85–89
infinitely divisible processes. Stoch Process Appl Ryzhikov VV, Troitskaya AE (2016) Mixing flows with a
61:277–288 homogeneous spectrum of multiplicity 2. Fundam Prikl
Rosinski J, Żak T (1997) The equivalence of ergodicity and Mat (Russ) 21:191–197
weak mixing for infinitely divisible processes. J Theor Sato K-I (1999) Lévy processes and infinitely divisible
Probab 10:73–86 distributions. Cambridge University Press, Cambridge
Roy E (2005) Mesures de Poisson, infinie divisibilit et Schmidt K (1977) Cocycles of ergodic transformation
proprits ergodiques. Thèse de doctorat de l’Universitè groups. Lecture notes in math, vol 1. Mac Millan of
Paris 6 India, New Delhi
Roy E (2007) Ergodic properties of Poissonian ID pro- Schmidt K (2002) Dispersing cocycles and mixing flows
cesses. Ann Probab 35:551–576 under functions. Fundam Math 173:191–199
Royden HL (1968) Real analysis. MacMillan, New York Schmidt K, Walters P (1982) Mildly mixing actions of
Rudin W (1962) Fourier analysis on groups. In: locally compact groups. Proc Lond Math Soc
Interscience tracts in pure and applied mathematics, 45(3):506–518
vol 12. Interscience Publishers (a division of John Shklover M (1967) Classical dynamical systems on the
Wiley and Sons), New York/London torus with continuous spectrum. Izv Vys Ucebn Zaved
Rudolph DJ (1979) An example of a measure-preserving Mat (Russ) 10(65):113–124
map with minimal self-joinings and applications. Simonelli LD (2018) Absolutely continuous spectrum for
J Anal Math 35:97–122 parabolic flows/maps. Discrete Contin Dyn Syst
Rudolph D (1985) k–fold mixing lifts to weakly mixing 38(1):263–292
isometric extensions. Ergodic Theory Dyn Syst Sinai YG (1994) Topics in ergodic theory. Princeton Uni-
5:445–447 versity Press, Princeton
Rudolph D (1986) Zn and Rn cocycle extension and comple- Smorodinsky M, Thouvenot J-P (1979) Bernoulli factors
mentary algebras. Ergodic Theory Dyn Syst 6:583–599 that span a transformation. Isr J Math 32:39–43
Rudolph D (1990) Fundamentals of measurable dynamics. Solomko AV (2012) New spectral multiplicities for ergodic
Oxford Science Publications, Oxford actions. Stud Math 208:229–247
Rudolph D (2004) Pointwise and L1 mixing relative to a Stepin AM (1986) Spectral properties of generic dynamical
sub-sigma algebra. Ill J Math 48:505–517 systems. Izv Akad Nauk SSSR Ser Math (Russ)
Rudolph D, Weiss B (2000) Entropy and mixing for ame- 50:801–834
nable group actions. Ann Math 151(2):1119–1150 Thoma E (1964) Über unitäre Darstellungen abzählbarer,
Ryzhikov VV (1991) Joinings of dynamical systems. diskreter Gruppen. Math Ann 153:111–138
Approximations and mixing. Uspekhi Mat Nauk Thouvenot J-P (1995) Some properties and applications of
(Russ) 46(5(281)):177–178 joinings in ergodic theory. In: Ergodic theory and its
Ryzhikov VV (1992) Mixing, rank and minimal self- connections with harmonic analysis. London Mathe-
joining of actions with invariant measure. Mat Sb matical Society. Cambridge University Press, Cam-
(Russ) 183:133–160 bridge, pp 207–235
148 Spectral Theory of Dynamical Systems

Thouvenot J-P (2000) Les systèmes simples sont disjoints Veech WA (1982a) A criterion for a process to be prime.
de ceux qui sont infiniment divisibles et plongeables Monatsh Math 94:335–341
dans un flot. Colloq Math 84(85):481–483 Veech W (1982b) Gauss measures for transformations on
Tiedra de Aldecoa R (2012) Spectral analysis of time- the space of interval exchange maps. Ann Math
changes of the horocycle flow. J Modern Dyn 115(2):201–242
6(2):275–285 Veech W (1984) The metric theory of interval exchange
Tiedra de Aldecoa R (2015a) The absolute continuous transformations. I. Generic spectral properties. Am
spectrum of skew products of compact lie groups. Isr J Math 106:1331–1359
J Math 208:323–350 Vershik AM (1962a) On the theory of normal dynamic
Tiedra de Aldecoa R (2015b) Commutator methods for the systems. Math Sov Dokl 144:625–628
spectral analysis of uniquely ergodic dynamical sys- Vershik AM (1962b) Spectral and metric isomorphism of
tems. Ergodic Theory Dyn Syst 35:944–967 some normal dynamical systems. Math Sov Dokl
Tiedra de Aldecoa R (2017) Commutator criteria for strong 144:693–696
mixing. Ergodic Theory Dyn Syst 37:308–323 Vershik AM (2005) Polymorphisms, Markov processes,
Tikhonov SV (2002) On a relation between the metric and and quasi-similarity. Discrete Contin Dyn Syst
spectral properties of Zd-actions. Fundam Prikl Mat 13:1305–1324
8:1179–1192 Vershik AM (2011) The Pascal automorphism has a con-
Tikhonov SV (2011) Mixing transformations with homo- tinuous spectrum. Funktsional Anal Prilozhen (Russ)
geneous spectrum. Mat Sb 202(8):139–160; English 45:16–33; translation in Funct Anal Appl
transl in Sb Math 202(8) (2011):1231–1252 45 (2011):173–186
Tikhonov SV (2012) Genericity of a multiple mixing. von Neumann J (1932) Zur Operatorenmethode in der
Uspekhi Mat Nauk (Russ) 67(4(406)):187–188; Klassichen Mechanik. Ann Math 33:587–642
translation in Russian Math Surv 67(4) (2012): Walters P (1982) An introduction to ergodic theory.
779–780 Springer, New York
Tikhonov SV (2013) Complete metric on mixing actions of Wysokinska M (2004) A class of real cocycles over an
general groups. J Dyn Control Syst 19:17–31 irrational rotation for which Rokhlin cocycle exten-
Todd JA (1950) On a conjecture in group theory. J Lond sions have Lebesgue component in the spectrum.
Math Soc 25:246 Topol Methods Nonlinear Anal 24:387–407
Ulcigrai C (2007) Mixing of asymmetric logarithmic sus- Wysokińska M (2007) Ergodic properties of skew products
pension flows over interval exchange transformations. and analytic special flows over rotations. PhD thesis,
Ergodic Theory Dyn Syst 27:991–1035 Toruń
Ulcigrai C (2011) Absence of mixing in area-preserving Yassawi R (2003) Multiple mixing and local rank group
flows on surfaces. Ann Math 173(3):1743–1778 actions. Ergodic Theory Dyn Syst 23:1275–1304
Veech W (1978) Interval exchange transformations. J Anal Zeitz P (1993) The centralizer of a rank one flow. Isr J Math
Math 33:222–272 84:129–145
 Marginal of a Probability Measure on a
Joinings in Ergodic Theory Product Space
Let l be a probability measure on the Cartesian
Thierry de la Rue product of a finite or countable collection of mea-
Laboratoire de Mathématiques Raphaël Salem, surable spaces i  I X i , i  I A i , and let J = j1,
CNRS – Université de Rouen, Saint Étienne du . . ., jk be a finite subset of I. The k-fold marginal of
Rouvray, France l on X j1 , . . . , X jk is the probability measure m
defined by:

Article Outline 8A1  A j1 , . . . , Ak  A jk ,

Glossary mðA1      Ak Þ≔l A1      Ak  Xi :

Definition of the Subject i  I∖J
Introduction
Joinings of Two or More Dynamical Systems
Self-Joinings Markov Intertwining
Some Applications and Future Directions Let ðX, A, m, T Þ and ðY, B, n, SÞ be two measure-
Bibliography preserving dynamical systems. We call Markov
intertwining of T and S any operator
P : L2(X, m) ! L2(Y, n) enjoying the following
properties:
Keywords

Measure · Preserving dynamical system · • PUT = USP, where UT and US are the unitary
Joining · Disjointness · Factor · Minimal self- operators on L2(X, m) and L2(Y, n) associated,
joining respectively, with T and S (i.e., UT f(x) = f(Tx),
and Usg(y) = g(Sy)).
Glossary
P1X ¼ 1Y ,
Disjoint Measure-Preserving Systems
The two measure-preserving dynamical systems • f  0 implies Pf  0, and g  0 implies P  g  0,
ðX, A, m, T Þ and ðY, B, n, SÞ are said to be disjoint where P is the adjoint operator of P.
if their only joining is the product measure
m  n.
Minimal Self-Joinings
Joining Let k  2 be an integer. The ergodic measure-
Let I be a finite or countable set, and for each preserving dynamical system T has k-fold mini-
i  I, let ðXi , A i , mi , T i Þ be a measure-preserving mal self-joinings if, for any ergodic joining l of
dynamical system. A joining of these systems is a k copies of T, we can partition the set {1, . . ., k} of
probability measure on the Cartesian product coordinates into subsets J1, . . ., J‘ such that:
∏i  IXi, which has the mis as marginals and 1. For j1 and j2 belonging to the same Ji, the
which is invariant under the product transforma- marginal of l on the coordinates j1 and j2 is
tion i  ITi. supported on the graph of T n for some integer
n (depending on j1 and j2).
2. For j1  J1, . . ., j‘  J‘, the coordinates
j1, . . ., j‘ are independent.

© Springer Science+Business Media, LLC, part of Springer Nature 2023 149

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_300
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2019
https://doi.org/10.1007/978-3-642-27737-5_300-2
150 Joinings in Ergodic Theory

We say that T has minimal self-joinings if T has supported on the graph of some S  C(T)
k-fold minimal self-joinings for every k  2. (depending on j1 and j2).
2. For j1  J1, . . ., j‘  J‘, the coordinates
Off-Diagonal Self-Joinings j1, . . ., j‘ are independent.
Let ðX, A, m, T Þ be a measure-preserving dynami- We say that T is simple if T is k-fold simple for
cal system and S be an invertible measure- every k  2.
preserving transformation of ðX, A, mÞ commuting
with T. Then the probability measure DS defined
on X  X by Definition of the Subject

DS ðA  BÞ≔m A \ S1 B ð1Þ The word joining can be considered as the coun-
terpart in ergodic theory of the notion of coupling
is a twofold self-joining of T supported on the in probability theory (see, e.g., Thorisson 2000):
graph of S. We call it an off-diagonal self-joining Given two or more processes defined on different
of T. spaces, what are the possibilities of embedding
them together in the same space? There always
Process in a Measure-Preserving Dynamical exists the solution of making them independent of
Systems each other, but interesting cases arise when we can
Let ðX, A, m, T Þ be a measure-preserving dynami- do this in other ways. The notion of joining orig-
cal system, and let ðE, B ðEÞÞ be a measurable inates in ergodic theory from pioneering works of
space (which may be a finite or countable set, or Furstenberg (1967), who introduced the funda-
ℝd, or ℂd. . .). For any E-valued random variable x mental notion of disjointness, and Rudolph, who
defined on the probability space ðX, A, mÞ, we can laid the basis of joining theory in his article on
consider the stochastic process (xi)i  ℤ defined by minimal self-joinings (Rudolph 1979). It has
today become an essential tool in the classification
xi ≔x ∘ T i : of measure-preserving dynamical systems and in
the study of their intrinsic properties.
Since T preserves the probability measure m,
(xi)i  ℤ is a stationary process: For any l and n, the
distribution of (x0, . . ., x‘) is the same as the Introduction
probability distribution of (xn, . . ., xn+‘).
A central question in ergodic theory is to tell when
Self-Joining two measure-preserving dynamical systems are
Let T be a measure-preserving dynamical system. essentially the same, i.e., when they are isomor-
A self-joining of T is a joining of a family phic. When this is not the case, a finer analysis
ðXi , A i , mi , T i Þi  I of systems where each Ti is a consists in asking what these two systems could
copy of T. If I is finite and has cardinal k, we speak share in common: For example, do there exist
of a k-fold self-joining of T. stationary processes which can be observed in
both systems? This latter question can also be
Simplicity asked in the following equivalent way: Do these
For k  2, we say that the ergodic measure- two systems have a common factor? The arithmet-
preserving dynamical system T is k-fold simple ical flavor of this question is not fortuitous: There
if, for any ergodic joining l of k copies of T, we are deep analogies between the arithmetic of inte-
can partition the set {1, . . ., k} of coordinates into gers and the classification of measure-preserving
subsets J1, . . ., J‘ such that: dynamical systems, and these analogies were at
1. For j1 and j2 belonging to the same Ji, the the starting point of the study of joinings in ergo-
marginal of l on the coordinates j1 and j2 is dic theory.
Joinings in Ergodic Theory 151

In the seminal paper (Furstenberg 1967) which turns out to be related with many important ques-
introduced the concept of joinings in ergodic the- tions in ergodic theory, and a large number of deep
ory, Furstenberg observed that two operations can results can be stated and proved inside the theory
be done with dynamical systems: We can consider of joinings. For example, the fact that the dynam-
the product of two dynamical systems, and we can ical systems S and T are isomorphic is equivalent
also take a factor of a given system. Like the to the existence of a special joining between S and
multiplication of integers, the product of dynam- T, and this can be used to give a joining proof of
ical systems is commutative and associative, it Krieger’s finite generator theorem, as well as
possesses a neutral element (the trivial single- Ornstein’s isomorphism theorem (see section
point system), and the systems S and T are both “Joinings Proofs of Ornstein’s and Krieger’s The-
factors of their product S  T. It was then natural orems”). As it already appears in Furstenberg’s
to introduce the property for two measure- article, joinings provide a powerful tool in the
preserving systems to be relatively prime. As far classification of measure-preserving dynamical
as integers are concerned, there are two equivalent systems: Many classes of systems can be charac-
ways of characterizing the relative primeness: terized in terms of their disjointness with other
First, the integers a and b are relatively prime if systems. Joinings are also strongly connected
their unique positive common factor is 1. Second, with difficult questions arising in the study of
a and b are relatively prime if, each time both almost everywhere convergence of non-
a and b are factors of an integer c, their product conventional averages (see section “Joinings and
ab is also a factor of c. It is a well-known theorem Multiple Ergodic Averages”).
in number theory that these two properties are Amazingly, a situation in which the study of
equivalent, but this was not clear for their analog joinings leads to most interesting results consists
in ergodic theory. Furstenberg reckoned that the in considering two or more identical systems. We
second way of defining relative primeness was the then speak of the self-joinings of the dynamical
most interesting property in ergodic theory and system T. Again, the study of self-joinings is
called it disjointness of measure-preserving sys- closely related to many ergodic properties of the
tems (we will discuss precisely in section “From system: its mixing properties, the structure of its
Disjointness to Isomorphy” what the correct ana- factors, the transformations which commute with
log is in the setting of ergodic theory). He also T, and so on. We already mentioned minimal self-
asked whether the nonexistence of a nontrivial joinings, and we will see in section “Minimal Self-
common factor between two systems was equiv- Joinings” how this property may be used to get
alent to their disjointness. He was able to prove many interesting examples, such as a transforma-
that disjointness implies the impossibility of a tion with no root, or a process with no nontrivial
nontrivial common factor, but not the converse. factor. In the same section, we will also discuss a
And in fact, the converse turns out to be false: In very interesting generalization of minimal self-
1979, Rudolph exhibited a counterexample in his joinings: the property of being simple.
paper introducing the important notion of minimal The range of applications of joinings in ergodic
self-joinings. The relationships between disjointness theory is very large; only some of them will be
and the lack of common factor will be presented in given in section “Some Applications and Future
details in section “Joinings and Factors.” Directions”: The use of joinings in proving
Given two measure-preserving dynamical sys- Krieger’s and Ornstein’s theorems, the links
tems S and T, the study of their disjointness natu- between joinings and some questions of pointwise
rally leads one to consider all the possible ways convergence, and the strong connections between
these two systems can be both seen as factors of a the study of self-joinings and Rohlin’s famous
third system. As we shall see, this is precisely the question on multifold mixing, which has been
study of their joinings. The concept of joining opened since 1949 (Rohlin 1949).
152 Joinings in Ergodic Theory

In parallel with the concept of joinings of Therefore, studying the joinings of a family of
measure-preserving systems, Furstenberg also measure-preserving dynamical system amounts to
introduced in Furstenberg (1967) the notion of study all the possible ways these systems can be
topological joinings, concerning topological together seen as factors in another big system.
dynamical systems (i.e., systems given by a con-
tinuous transformation of a compact metric
The Set of Joinings
space). In the present article, we have restricted
The set of all joinings of the Tis will be denoted by
ourselves to the measure-preserving setting. In
J(Ti, i  I). Before anything else, we have to
addition to Furstenberg’s paper, applications of
observe that this set is never empty. Indeed, what-
topological joinings can be found, for example,
ever the systems are, the product measure i  I
in Furstenberg et al. (1973), King (1990), and
mi always belongs to this set. Note also that any
Weiss (1998).
convex combination of joinings is a joining: J-
(Ti, i  I) is a convex set.
The set of joinings is turned into a compact
Joinings of Two or More Dynamical metrizable space, equipped with the topology
Systems defined by the following notion of convergence:
ln ! l if and only if, for all family of measur-
n!1
In the following, we are given a finite or countable able subsets ðAi Þi  I  i  I A i , finitely many of
family ðXi , A i , mi , T i Þi  I of measure-preserving them being different from Xi, we have
dynamical systems: Ti is an invertible measure-
preserving transformation of the standard Borel
probability space ðXi , A i , mi Þ: When it is not ln Ai !l Ai : (2)
n!1
ambiguous, we shall often use the symbol Ti to iI iI
denote both the transformation and the system.
A joining l of the Tis (see the definition in the We can easily construct a distance defining
“Glossary”) defines a new measure-preserving this topology by observing that it is enough to
dynamical system: The product transformation check (2) when each of the Ais is chosen in some
countable algebra C i generating the s-algebra A i
 T i : ðxi Þi  I 7! ðT i xi Þi  I . We can also point out that, when the Xis are
iI
themselves compact metric spaces, this topol-
ogy on the set of joinings is nothing but the
acting on the Cartesian product ∏i  IXi and pre-
restriction to J(Ti, i  I) of the usual weak
serving the probability measure l. We will denote
topology.
this big system by (i  ITi)l. Since all marginals
It is particularly interesting to study ergodic
of l are given by the original probabilities mi,
joinings of the Tis, whose set will be denoted by
observing only the coordinate i in the big system
Je(Ti, i  I). Since any factor of an ergodic
is the same as observing only the system Ti. Thus,
system is itself ergodic, a necessary condition
each system Ti is a factor of (i  ITi)l, via the
for Je(Ti, i  I) not to be empty is that all the
homomorphism pi which maps any point in the
Tis be themselves ergodic. Conversely, if all the
Cartesian product to its i-th coordinate.
Tis are ergodic, we can prove by considering the
Conversely, if we are given a measure-
ergodic decomposition of the product measure
preserving dynamical system ðZ, C , r, RÞ admit-
i  Imi that ergodic joinings do exist: Any ergo-
ting each Ti as a factor via some homomorphism
dic measure appearing in the ergodic decompo-
’i: Z ! Xi, then we can construct the map ’:
sition of some joining has to be itself a joining.
Z ! ∏i  IXi sending z to (’i(z))i  I. We can easily
This result can also be stated in the following
check that the image of the probability measure r
way:
is then a joining of the Tis.
Joinings in Ergodic Theory 153

Proposition 1 disjointness property with the existence of a par-

If all the Tis are ergodic, the set of their ergodic ticular factor.
joinings is the set of extremal points in the com- Both the first and the second results can be
pact convex set J(Ti, i  I). derived from the next theorem, giving a general
spectral condition in which disjointness arises.
From Disjointness to Isomorphy The proof of this theorem can be found in
In this section, as in many others in this entry, we Thouvenot (1995). It is a direct consequence of
are focusing on the case where our family of the fact that, if f and g are square-integrable func-
dynamical systems is reduced to two of them. tions in a given dynamical system, and if their
We will then rather call them S and T, standing spectral measures are mutually singular, then
for ðY, B, n, SÞ and ðX, A, m, T Þ. We are interested f and g are orthogonal in L2.
here in two extremal cases for the set of joinings
J(T, S). The first one occurs when the two systems
Theorem 2
are as far as possible from each other: They have
If the reduced maximum spectral types of T and
nothing to share in common, and therefore their
S are mutually singular, then T and S are disjoint.
set of joinings is reduced to the singleton {m  n}:
This is called the disjointness of S and T. The
second one arises when the two systems are iso- As we already said in the introduction,
morphic, and we will see how this property shows disjointness was recognized by Furstenberg as
through J(T, S). the most pertinent way to define the analog of
the arithmetic property “a and b are relatively
Disjointness prime” in the context of measure-preserving
Many situations where disjointness arises were dynamical systems. We must however point out
already given by Furstenberg in (1967). Particu- that the statement:
larly interesting is the fact that classes of dynam- (i) S and T are disjoint is, in general, strictly
ical systems can be characterized through stronger than the straightforward translation of the
disjointness properties. We list here some of the arithmetic property.
main examples of disjoint classes of measure- (ii) Each time both S and T appear as factors in
preserving systems. a third dynamical system, then their product S  T
also appears as a factor in this system.
Theorem 1 Indeed, contrary to the situation in ordinary
1. T is ergodic if and only if it is disjoint from arithmetic, there exist nontrivial dynamical sys-
every identity map. tems T which are isomorphic to T  T: For exam-
2. T is weakly mixing if and only if it is disjoint ple, this is the case when T is the product of
from any rotation on the circle. countably many copies of a single nontrivial sys-
3. T has zero entropy if and only if it is disjoint tem. Now, if T is such a system and if we take
from any Bernoulli shift. S = T, then S and T do not satisfy statement (i):
4. T is a K-system if and only if it is disjoint A nontrivial system is never disjoint from itself, as
from any zero-entropy system. we will see in the next section. However, they
obviously satisfy the statement (ii).
The first result is the easiest, but is quite impor- A correct translation of the arithmetic property
tant, in particular when it is stated in the following is the following: S and T are disjoint if and only if,
form: If l is a joining of T and S, with T ergodic, each time T and S appear as factors in some
and if l is invariant by T  Id, then l = m  n. dynamical system through the respective homo-
The second, third, and fourth results were orig- morphisms pT and pS, T  S also appears as a
inally proved by Furstenberg. They can also be factor through a homomorphism pT  S such that
seen as corollaries of the theorems presented in pX ∘ pT  S = pT and pY ∘ pT  S = pS, where pX
section “Joinings and Factors,” linking the non- and pY are the projections on the coordinates
154 Joinings in Ergodic Theory

in the Cartesian product X  Y (see the diagram

below).

Joinings in Ergodic Theory, Fig. 1 The joining D’

identifies the sets A  Y and X  ’(A)

Thus, in the case where S and T are isomorphic,

Joinings and Isomorphism we can find a special joining of S and T, which is
We first introduce some notations: For any prob- supported on the graph of an isomorphism and
ability measure l on a measurable space, let which identifies the two s-algebras generated by
l
‘A ¼ B’ stand for ‘l(ADB) = 0’. Similarly, if C the two coordinates. What is remarkable is that the
and D are s-algebras of measurable sets, we write converse is true: The existence of an isomorphism
l
‘C  D’ if, for any C  C, we can find some D  D between S and T is characterized by the existence of
l l
such that C ¼ D , and by ‘C ¼ D ’ we naturally such a joining, and we have the following theorem:
l l
mean that both C  D and D  C hold.
Let us assume now that our two systems S and Theorem 3
T are isomorphic: This means that we can find The measure-preserving dynamical systems S and
some measurable one-to-one map ’ : X ! Y, T are isomorphic if and only if there exists a
with T(m) = n, and ’ ∘ T = S ∘ ’. With such a joining l of S and T such that
’, we construct the measurable map c : X ! X  Y
l
by setting fX, 0g  B ¼ A  fY, 0g: ð4Þ


cðxÞ≔ððx, ’ðxÞÞ: When l is such a joining, it is supported on the

graph of an isomorphism of T and S, and both
Let D’ be the image measure of m by c. This systems are isomorphic to the joint system (T  S)l.
measure is supported on the graph of ’ and is also
characterized by This theorem finds nice applications in the
proof of classical isomorphism results. For exam-
8A  A, 8B  B, D’ ðA  BÞ ple, it can be used to prove that two discrete-
¼ m A \ ’1 B : ð3Þ spectrum systems which are spectrally isomorphic
are isomorphic (see Thouvenot 1995 or de la Rue
We can easily check that, ’ being an isomor- 2006b). We will also see in section “Joinings
phism of T and S, D’ is a joining of T and S. And Proofs of Ornstein’s and Krieger’s Theorems”
this joining satisfies very special properties (Fig. 1): how it can be applied in the proofs of Krieger’s
and Ornstein’s deep theorems.
D’ Consider now the case were T and S are no
• For any measurable A  X, A  Y ¼ X  ’ðAÞ. longer isomorphic, but where S is only a factor of
• Conversely, for any measurable B  Y, X  T. Then we have a factor map p : X ! Y which has
D’ the same properties as an isomorphism ’, except
B ¼ ’1 ðBÞ  Y.
that it is not one-to-one (p is only onto). The
measure Dp, constructed in the same way as D’,
Joinings in Ergodic Theory 155

is still a joining supported on the graph of p, but it

does not identify the two s-algebras generated by
the two coordinates any more: Instead of Condi-
tion (4), Dp only satisfies the weaker one:

Dp
fX, 0g  B  A  fY, 0g: ð5Þ



The existence of a joining satisfying (5) is a
criterion for S being a factor of T.
For more details on the results stated in this
section, we refer the reader to de la Rue (2006b).
Joinings in Ergodic Theory, Fig. 2 The relatively inde-
Joinings and Factors pendent joining mRn and its disintegration over z
The purpose of this section is to investigate the
relationships between the disjointness of two sys-
tems S and T and the lack of a common factor. The In other words, we have identified in the two
crucial fact which was pointed out by Furstenberg systems T and S the projections on their common
is that the existence of a common factor enables factor R. The twofold marginal of l on X  Y is
one to construct a very special joining of S and T: itself a joining of T and S, which we call the
The relatively independent joining over this factor. relatively independent joining over the common
Let us assume that our systems S and T share a factor R. This joining will be denoted by mRn.
common factor ðZ, C , r, RÞ, which means that we (Be careful: The projections pX and pY are hid-
have measurable onto maps pX : X ! Z and den in this notation, but we have to know them
pY : Y ! Z, respectively, sending m and n to r and to define this joining.) From (6), we immedi-
satisfying pX ∘ T = R ∘ pX and pY ∘ S = R ∘ pY. We ately get the formula defining mRn:
can then consider the joinings supported on their
graphs DpX  J ðT, RÞ and DpY  J ðS, RÞ, as defined 8A  A, 8B  B,
in the preceding section. Next, we construct a join-
ing l of the three systems S, T, and R. Heuristically, mR nðA  BÞ≔ DpX ½1xA jz DpY 1yB jz drðzÞ:
l is the probability distribution of the triple (x, y, z) Z
when we first pick z according to the probability ð7Þ
distribution r, then x and y according to their
conditional distribution knowing z in the respective This definition of the relatively independent
joinings DpX and DpY , but independently of each joining over a common factor can easily be
other (Fig. 2). More precisely, l is defined by extended to a finite or countable family of systems
setting, for all A  A, B  B, and C  C sharing the same common factor.
Note that mRn coincides with the product
measure mn if and only if the common factor is
lðA  B  CÞ≔ DpX ½1xA jz DpY 1yB jz drðzÞ:
C the trivial one-point system. We therefore get the
ð6Þ following result:

Observe that the twofold marginals of l on Theorem 4

X  Z and Y  Z are, respectively, DpX and DpY , If S and T have a nontrivial common factor, then
which means that we have z = pX(x) = pY(y) these systems are not disjoint.
l-almost surely.
156 Joinings in Ergodic Theory

As we already said in the introduction, dynamical systems which are stable under the
Rudolph exhibited in Rudolph (1979) a counter- operations of taking joinings and factors. We
example showing that the converse is not true. will call these properties stable properties. This
There exists however an important result, which is, e.g., the case of the zero-entropy property: We
was published in Glasner et al. (2000) and know that any factor of a zero-entropy system still
Lemańczyk et al. (2000) allowing us to derive has zero entropy and that any joining of zero-
some information on factors from the non- entropy systems also has zero entropy. In other
disjointness of two systems. words, T has zero entropy implies that any
T-factor has zero entropy. But the property of
Theorem 5 S being a K-system is precisely characterized by
If T and S are not disjoint, then S has a nontrivial the fact that any nontrivial factor of S has positive
common factor with some joining of a countable entropy. Hence, a K-system S cannot have a non-
family of copies of T. trivial T-factor if T has zero entropy and is there-
fore disjoint from T. The converse is a
This result leads to the introduction of a special consequence of Theorem 4: If S is not a
class of factors when some dynamical system T is K-system, then it possesses a nontrivial zero-
given: For any other dynamical system S, call entropy factor, and therefore there exists some
T-factor of S any common factor of S with a zero-entropy system from which it is not disjoint.
joining of countably many copies of T. If The same argument also applies to the
ðZ, C , r, RÞ is a T-factor of S and p : Y ! Z is a disjointness of discrete-spectrum systems with
factor map, we say that the s-algebra p1 ðC Þ is a weakly mixing systems, since discrete spectrum is
T-factor s-algebra of S. Another way to state a stable property, and weakly mixing systems are
Theorem 5 is then the following: If S and T are characterized by the fact that they do not have any
not disjoint, then S has a nontrivial T-factor. In discrete-spectrum factor.
fact, an even more precise result can be derived
from the proof of Theorem 5: For any joining l of Markov Intertwinings and Composition of
S and T, for any bounded measurable function f on Joinings
X, the factor s-algebra of S generated by the There is another way of defining joinings of two
function l ½ f ðxÞjy is a T-factor s-algebra of S. measure-preserving dynamical systems involving
With the notion of T-factor, Theorem 5 has been operators on L2 spaces, mainly put to light by
extended in Lesigne et al. (2003) in the following Ryzhikov (see Ryzhikov 1993b): Observe that
way, showing the existence of a special T-factor for any joining l  J(T, S), we can consider the
s-algebra of S comprising anything in S which operator Pl : L2(X, m) ! L2(Y, n) defined by
could lead to a nontrivial joining between T and S.
Pl ð f Þ≔l ½ f ðxÞjy :
Theorem 6
Given two measure-preserving dynamical systems It is easily checked that Pl is a Markov
ðX, A, m, T Þ and ðY, B, n, SÞ, there always exists a intertwining of T and S. Conversely, given any
maximum T-factor s-algebra of S, denoted by F T. Markov intertwining P of T and S, it can be
Under any joining l of T and S, the s-algebras shown that the measure lP defined on X  Y by
A  f 0, Y g and f 0, Xg  B are independent con-


ditionally to the s-algebra f 0, Xg  F T . lP ðA  BÞ≔hP1A , 1B iL2 ðY,nÞ



Theorem 5 gives a powerful tool to prove some is a joining of T and S.

important disjointness results, such as those stated This reformulation of the notion of joining is
in Theorem 1. These results involve properties of useful when joinings are studied in connection
Joinings in Ergodic Theory 157

with spectral properties of the transformations ðX, A, mÞÞ and denote it by C(T). It always con-
(see, e.g., Goodson 2000). It also provides us tains, at least, all the powers T n, n  ℤ.
with a convenient setting to introduce the compo- Each element S of C(T) gives rise to a twofold
sition of joinings: If we are given three dynamical self-joining DS supported on the graph of S. Such
systems ðX, A, m, T Þ, ðY, B, n, SÞ, and ðZ, C , r, RÞ, self-joinings are called off-diagonal self-joinings.
a joining l  J(T, S), and a joining l0  J(S, R), They also belong to J ke ðT Þ if T is ergodic.
the composition of the Markov intertwinings Pl It follows that properties of the commutant of
and Pl0 is easily seen to give a third Markov an ergodic T can be seen in its ergodic joinings. As
intertwining, which itself corresponds to a joining an example of application, we can cite Ryzhikov’s
of T and R denoted by l ∘ l0. When R = S = T, i.e., proof of King’s weak closure theorem for rank-
when we are speaking of twofold self-joinings of a one transformations (An introduction to finite-
single system T (cf. next section), this operation rank transformations can be found, e.g., in
turns J(T, T) = J2(T) into a semigroup. Ahn and Nadkarni (1998a); we also refer the reader to the
Lemańczyk (2003) have shown that the subset quite complete survey (Ferenczi 1997).). Rank-
J 2e ðT Þ of ergodic twofold self-joinings is a sub- one measure-preserving transformations form a
semigroup if and only if T is semisimple (see very important class of zero-entropy, ergodic
section “Simple Systems”). measure-preserving transformations. They have
many remarkable properties, among which is the
fact that their commutant is reduced to the weak
limits of powers of T. In other words, if T is rank
Self-Joinings
one, for any S  C(T), there exists a subsequence
of integers (nk) such that
We now turn to the case where the measure-
preserving dynamical systems we want to join
together are all copies of a single system T. For 8A  A, m T nk ADS 1 A ! 0: (8)
k!1
k  2, any joining of k copies of T is called a k-fold
self-joining of T. We denote by Jk(T) the set of all King proved this result in 1986 (King 1986),
k-fold self-joinings of T and by J ke ðT Þ the subset of using a very intricate coding argument. Observing
ergodic k-fold self-joinings. that (8) was equivalent to the convergence, in
J2(T), of DT nk to DS, Ryzhikov showed in
Ryzhikov (1992b) that King’s theorem could be
Self-Joinings and Commuting
seen as a consequence of the following general
Transformations
result concerning twofold self-joinings of rank-
As soon as T is not the trivial single-point system,
one systems:
T is never disjoint from itself: Since T is obviously
isomorphic to itself, we can always find a twofold
Theorem 7
self-joining of T which is not the product measure
Let T be a rank-one measure-preserving transfor-
by considering self-joinings supported on graphs
mation, and l  J 2e ðT Þ. Then there exist T  1/2, a
of isomorphisms (see section “Joinings and Iso-
subsequence of integers (nk), and another twofold
morphism”). The simplest of them is obtained by
self-joining l0 of T such that
taking the identity map as an isomorphism, and
we get that J2(T) always contains the diagonal 0
measure D0 ≔ DId. DT nk ! tl þ ð1  t Þl :
k!1
In general, an isomorphism of T with itself is an
invertible measure-preserving transformation S of Minimal Self-Joinings
ðX, A, mÞ which commutes with T. We call For any measure-preserving dynamical system T,
commutant of T the set of all such transformations the set of twofold self-joinings of T contains at least
(it is a subgroup of the group of automorphisms of the product measure m  m p, the off-diagonal
158 Joinings in Ergodic Theory

joinings DT n for each n  ℤ, and any convex minimal self-joinings for all k  2, we simply say
combination of these. Rudolph (1979) discovered that T has minimal self-joinings.
in 1979 that we can find systems for which there Rudolph’s construction of a system with twofold
are no other twofold self-joinings than these obvi- minimal self-joinings (Rudolph 1979) was inspired
ous ones. When this is the case, we say that T has by a famous work of Ornstein (1972), giving the
twofold minimal self-joinings, or for short: first example of a transformation with no roots. It
T  MSJ(2). It can be shown (see, e.g., Rudolph turned out that Ornstein’s example is a mixing rank-
1990) that, as soon as the underlying probability one system, and all mixing rank-one systems were
space is not atomic (which we henceforth assume), later proved by King (1988) to have twofold min-
twofold minimal self-joinings imply that T is imal self-joinings. This can also be viewed as a
weakly mixing and therefore that m  m and DT n , consequence of Ryzhikov’s Theorem 7. Indeed, in
n  ℤ, are ergodic twofold self-joinings of T. That the language of joinings, the mixing property of
is why twofold minimal self-joinings are often T translates as follows:
defined by the following:
T is mixing , DT n ! m  m: (10)
jnj!1
T  MSJð2Þ , J 2e ðT Þ
¼ fm  mg [ fDT n , n  ℤg: ð9Þ
Therefore, if in Theorem 7 we further assume
that T is mixing, then either the sequence (nk) we
Systems with twofold minimal self-joinings
get in the conclusion is bounded, and then l is
have very interesting properties. First, since for
some DT n , or it is unbounded and then l = m  m.
any S in C(T), DS belongs to J 2e ðT Þ, we immedi-
T  MSJ(k) obviously implies T  MSJ(k0)
ately see that the commutant of T is reduced to the for any 2 k0 k, but the converse is not known.
powers of T. In particular, it is impossible to find a The question whether twofold minimal self-
square root of T, i.e., a measure-preserving S such joinings implies k-fold minimal self-joinings for
that S ∘ S = T. Second, the existence of a non- all k is related to the important open problem of
trivial factor a-algebra of T would lead, via the pairwise-independent joinings (see section
relatively independent self-joining over this fac- “Pairwise-Independent Joinings”). But the latter
tor, to some ergodic twofold self-joining of problem is solved for some special classes of
T which is not in the list prescribed by (9). There- systems, in particular in the category of mixing
fore, any factor s-algebra of a system with twofold rank-one transformations. It follows that, if T is
minimal self-joinings must be either the trivial mixing and rank one, then T has minimal self-
s-algebra f 0, Xg or the whole s-algebra A . This


joinings.
has the remarkable consequence that if x is any In 1980, Del Junco, Rahe, and Swanson pro-
random variable on the underlying probability ved that Chacon’s transformation also has mini-
space which is not almost-surely constant, then mal self-joinings (del Junco et al. 1980). This
the process (x ∘ T n)n  ℤ always generates the well-known transformation is also a rank-one sys-
whole s-algebra A . This also implies that T has tem, but it is not mixing (it had been introduced by
zero entropy, since positive-entropy systems have R.V. Chacon in 1969 (Chacon 1969) as the first
many nontrivial factors. explicit example of a weakly mixing transforma-
The notion of twofold minimal self-joinings tion which is not mixing). For another example of
extends for any integer k  2 to k-fold minimal a transformation with twofold minimal self-
self-joinings, which roughly means that there are joinings, constructed as an exchange map on
no other k-fold ergodic self-joinings than the “obvi- three intervals, we refer to del Junco (1983).
ous” ones: those for which the k coordinates are The existence of a transformation with mini-
either independent or just translated by some power mal self-joinings has been used by Rudolph as a
of T (see the glossary for a more precise definition). wonderful tool to construct a large variety of
We denote in this case: T  MSJ(k). If T has k-fold striking counterexamples, such as:
Joinings in Ergodic Theory 159

• A transformation T which has no roots, while the fact that C(T) may contain other transforma-
T2 has roots of any order tions than the powers of T.) It turns out that simple
• A transformation with a cubic root but no systems may have nontrivial factors, but the struc-
square root ture of these factors can be explicitly described:
• Two measure-preserving dynamical systems They are always associated with some compact
which are weakly isomorphic (each one is a subgroup of C(T). More precisely, if K is a com-
factor of the other) but not isomorphic pact subgroup of C(T), we can consider the factor
s-algebra
Let us now sketch the argument showing that
we can find two systems with no common factor F K ≔fA  A : 8S  K, A ¼ SðAÞg,
but which are not disjoint: We start with a system
T with minimal self-joinings. Consider the direct and the corresponding factor transformation T jF K
product of T with an independent copy T 0 of itself, (called a group factor). Then Veech proved the
and take the symmetric factor S of T  T0, that is to following theorem concerning the structure of
say the factor we get if we only look at the non- factors of a twofold simple system.
ordered pair of coordinates {x, x0} in the Cartesian
product. Then S is surely not disjoint from T, since Theorem 8
the pair {x, x0} is not independent of x. However, If the dynamical system T is twofold simple, and if
if S and T had a nontrivial common factor, then F  A is a nontrivial factor s-algebra of T, then
this factor should be isomorphic to T itself there exists a compact subgroup K of the group
(because T has minimal self-joinings). Therefore C(T) such that F ¼ F K .
we could find in the direct product T  T0 a third
copy T of T, which is measurable with respect to There is a natural generalization of Veech’s
the symmetric factor. In particular, T is invariant property to the case of k-fold self-joinings,
by the flip map (x, x0) 7! (x0, x), and this prevents which has been introduced by Del Junco and
T from being measurable with respect to only one Rudolph in 1987 (del Junco and Rudolph 1987)
coordinate. Then, since T  MSJ(3), the systems (see the precise definition of simple systems in the
T, T0, and T have no choice but to be independent. glossary). In their work, important results
But this contradicts the fact that T is measurable concerning the structure of factors and joinings
with respect to the s-algebra generated by x and x0. of simple systems are proved. In particular, they
Hence, T and S have no nontrivial common factor. are able to completely describe the structure of the
We can also cite the example given by Glasner ergodic joinings between a given simple system
and Weiss (1983) of a pair of horocycle transfor- and any ergodic system (see also Glasner 2003
mations which have no nontrivial common factor, and Thouvenot 1995). Recall that, for any r  1,
yet are not disjoint. Their construction relies on the symmetric factor of T r is the system we get if
the deep work by Ratner (1983), which describes we observe the r coordinates of the point in Xr and
the structure of joinings of horocycle flows. forget their order. This is a special case of group
factor, associated with the subgroup of C(T r)
consisting of all permutations of the coordinates.
Simple Systems We denote this symmetric factor by Thri.
An important generalization of twofold minimal
self-joinings has been proposed by William Theorem 9
A. Veech in 1982 (Veech 1982). We say that the Let T be a simple system and S an ergodic system.
measure-preserving dynamical system T is two- Assume that l is an ergodic joining of T and
fold simple if it has no other ergodic twofold self- S which is different from the product measure.
joinings than the product measure {m  m} and Then there exist a compact subgroup K of C(T)
joinings supported on the graph of a transforma- and an integer r  1 such that:
tion S  C(T). (The difference with MSJ(2) lies in
160 Joinings in Ergodic Theory

• T jF K
hr i
is a factor of S. over F (recall that weak mixing is itself charac-
• l is the projection on X  Y of the relatively terized by the ergodicity of the direct product
independent joining of T r and S over their T  T).
hr i For more details on this subject, we refer the
common factor T jF K : reader to Lemańczyk et al. (2002). We also wish to
mention the generalization of simplicity called
If we further assume that the second system is semisimplicity proposed by Del Junco et al. in
also simple, then in the conclusion we can take (1995), which is precisely characterized by the
r = 1. In other words, ergodic joinings of simple fact that, for any l  J 2e ðT Þ, the system (T  T)l
systems S and T are either the product measure or is a relatively weakly mixing extension of T.
relatively independent joinings over a common
group factor. This leads to the following corollary:

Theorem 10 Some Applications and Future

Simple systems without nontrivial common factor Directions
are disjoint.
Filtering Problems
As for minimal self-joining, it is not known in Filtering problems were one of the motivations
general whether twofold simplicity implies k-fold presented by Furstenberg for the introduction of
simplicity for all k. This question is studied in the disjointness property in Furstenberg (1967),
Glasner et al. (1992), where sufficient spectral and he considered in particular the following sit-
conditions are given for this implication to hold. uation. Suppose we are given two real-valued
It is also proved that any threefold simple weakly stationary processes (Xn) and (Yn), with their
mixing transformation is simple of all order. joint distribution also stationary. We can interpret
(Xn) as a signal, perturbed by a noise (Yn). Under
which condition can we recover the original signal
Relative Properties with Respect to a Factor (Xn) from the observation of (Xn + Yn)? If this is
In fact, Veech also introduced a weaker, “relativ- possible, we say that the sequence (Xn, Yn) admits
ized” version of the twofold simplicity. If F  A a perfect filter. Furstenberg proved that a perfect
is a nontrivial factor s - algebra of T, let us denote filter exists if the two processes (Xn) and (Yn) are
by J 2 ðT, F Þ the set of twofold self-joinings of integrable and if the two measure-preserving
T which are “constructed over F ”, which means dynamical systems constructed as the shift of the
that their restriction to the product s-algebra two processes are disjoint. Furstenberg also
F  F coincides with the diagonal measure. observed that the integrability assumption can be
(The relatively independent joining over F is the removed if a stronger disjointness property is sat-
canonical example of such a joining.) For the isfied: A perfect filter exists if the system
conclusion of Theorem 8 to hold, it is enough to T generated by (Xn) is doubly disjoint from the
assume only that the ergodic elements of J 2 ðT, F Þ system S generated by (Yn), in the sense that T is
be supported on the graph of a transformation disjoint from any ergodic self-joining of S. Sev-
S  C(T). This is an important situation where eral generalizations have been studied (see
the study of J 2 ðT, F Þ gives strong informations on Bulatek et al. 2005 and Furstenberg et al. 1995),
the way F is embedded in the whole system T or, but the question whether the integrability assump-
in other words, on the relative properties of T with tion of the processes can be removed remained
respect to the factor TjF : A simple example of open for a long time. Finally, a short and clever
such a relative property is the relative weak mixing argument by Garbit (2011) proved in 2011 that
with respect to F , which is characterized by the indeed the only assumption of disjointness is
ergodicity of the relatively independent joining enough to ensure the existence of a perfect filter.
Joinings in Ergodic Theory 161

Joinings Proofs of Ornstein’s and Krieger’s isomorphism of two specific systems, but rather to
Theorems the isomorphism of one given system with some
We have already seen that joinings could be used other system which has to be found. We have
to prove isomorphisms between systems. This fact therefore to introduce a larger set of joinings:
found a nice application in the proofs of two major Given an integer n, we denote by Yn the set of
theorems in ergodic theory: Ornstein’s isomor- double-sided sequences taking values in {1,. . .,n}.
phism theorem (Ornstein 1970), stating that two We consider on Yn the shift transformation S, but
Bernoulli shifts with the same entropy are isomor- we do not determine yet the invariant measure.
phic, and Krieger’s finite generator theorem Now, for a specific measure-preserving dynamical
(Krieger 1970), which says that any dynamical system T, consider the set J(n, T) of all possible
system with finite entropy is isomorphic to the joinings of T with some system (Yn, n, S), when n
shift transformation on a finite-valued stationary ranges over all possible shift-invariant probability
process. The idea of this joining approach to the measures on Yn. J(n, T) can also be equipped with a
proofs of Krieger’s and Ornstein’s theorems was topology which turns it into a compact convex
originally due to Burton and Rothstein, who cir- metric space, and as soon as T is ergodic, the set
culated a preliminary report on the subject which Je(n, T) of ergodic elements of J(n, T) is not empty.
was never published (Burton and Rothstein 1977). In this setting, Krieger’s theorem can be stated as
The first published and fully detailed exposition of follows:
these proofs can be found in Rudolph’s book
(Rudolph 1990) (see also in Glasner’s book Theorem 12
(Glasner 2003)). (Krieger’s finite generator theorem). Let T be an
In fact, Ornstein’s theorem goes far more ergodic dynamical system with entropy h(T )
beyond the isomorphism of two given Bernoulli < log2 n. Then the set of l  J(n, T ) which are
shifts: It also gives a powerful tool for showing supported on graphs of isomorphisms between
that a specific dynamical system is isomorphic to a T and some system (Yn, n, S) forms a dense
Bernoulli shift. In particular, Ornstein introduced Gd in J e ðn, T Þ:
the property for an ergodic stationary process to
be finitely determined. We shall not give here the
precise definition of this property (for a complete Since any system of the form (Yn, n, S) obvi-
exposition of Ornstein’s theory, we refer the ously has an n-valued generating process, we
reader to Ornstein (1974)), but simply point out obtain as a corollary that T itself is generated by
that Bernoulli shifts and mixing Markov chains an n-valued process.
are examples of finitely determined processes. As another nice example of how joinings can
Rudolph’s argument to show Ornstein’s theorem be used to describe the isomorphisms between
via joinings makes use of Theorem 3 and of the two transformations, we can also cite the paper
topology of J(T, S). (Foreman et al. 2011) by Foreman, Rudolph, and
Weiss. With a clever construction of a family of
Theorem 11 transformations for which they can completely
(Ornstein’s isomorphism theorem). Let T and S be describe the joinings between T and its inverse,
two ergodic dynamical systems with the same they are able to prove that the isomorphism rela-
entropy, and both generated by finitely deter- tion between measure-preserving transformations
mined stationary processes. Then the set of join- of the unit interval is not Borel.
ings of T and S which are supported on graphs of
isomorphisms forms a dense Gd in J e ðT, SÞ: Joinings and Rohlin’s Multifold-Mixing
Question
Krieger’s theorem is not as easily stated in terms We have already seen that the property of T being
of joinings, because it does not refer to the mixing could be expressed in terms of twofold
162 Joinings in Ergodic Theory

self-joinings of T (see (10)). Rohlin proposed in A system is said to be pairwise independently

1949 (Rohlin 1949) a generalization of this prop- determined (PID) if for all k  3, the only
erty, called multifold mixing: The measure- pairwise-independent k-fold self-joining of the
preserving transformation T is said to be k-fold system is the product measure. Thus, in the class
mixing if 8A1 , A2 , . . . , Ak  A, of PID systems, twofold mixing implies threefold
mixing.
lim m A1 \ T n2 A2 \   \ T ðn2 þþnk Þ Ak In fact, non-PID systems are easy to find (see,
n2 , n3 ,...,nk !1
e.g., de la Rue 2006b), but the examples we know
k so far are either periodic transformations (which
¼ mðAi Þ: cannot be counterexamples to Rohlin’s question
i¼1
since they are not mixing) or transformations with
positive entropy. However, using an argument
Again, this definition can easily be translated
provided by Thouvenot, we can prove that, if
into the language of joinings: T is k-fold mixing
there exists a twofold mixing T which is not three-
when the sequence DT n2 ,...,T n2þþnk converges in fold mixing, then we can find such a T in the
Jk(T) to mk as n2,...,nk go to infinity, where category of zero-entropy dynamical systems
DT n2 ,...,T n2 þþnk is the obvious generalization of (see, e.g., de la Rue 2006a). Therefore, a negative
DT n to the case of k-fold self-joinings. The classi- answer to the following question would solve
cal notion of mixing corresponds in this setting to Rohlin’s multifold-mixing problem:
twofold mixing. (We must point out that Rohlin’s
original definition of k-fold mixing involved Question 1
k + 1 sets; thus, the classical mixing property Does there exist a zero-entropy, weakly mixing
was called onefold mixing. However it seems dynamical system T which is not PID?
that the convention we adopt here is now used
by most authors, and we find it more coherent
when translated in the language of multifold The problem of (non)existence of such
self-joinings.) pairwise-independent joinings is also related to
Obviously, threefold mixing is stronger than the question whether MSJ(2) implies MSJ(3) or
twofold mixing, and Rohlin asked in his article whether twofold simplicity implies threefold sim-
whether the converse is true. This question is still plicity. Indeed, any counterexample to one of
open today, even though many important works these implications would necessarily be of zero
dealt with it and supplied partial answers. Most entropy and would possess a pairwise-
of these works directly involve self-joinings via independent threefold self-joining which is not
the argument exposed in the following section. the product measure. Moreover, it is noticed in
[24, Corollary 12.22] that if we find a system
which is MSJ(2) but not MSJ(3), then this system
Pairwise-Independent Joinings would also be twofold mixing without being
Let T be a twofold mixing dynamical system. If threefold mixing.
T is not threefold mixing, DT n ,T nþm does not Question 1 has been answered by Host and
converge to the product measure as n and m go Ryzhikov for some special classes of zero-entropy
to 1. By compactness of J3(T), we can find sub- dynamical systems.
sequences (nk) and (mk) such that DT nk ,T nk þmk
converges to a cluster point l 6¼ m  m  m. How-
Host’s and Ryzhikov’s Theorems
ever, by twofold mixing, the three coordinates
The following theorem, proved in 1991 by Host
must be pairwise independent under l. We there-
(1991) (see also Glasner 2003 and Nadkarni
fore get a threefold self-joining l with the unusual
1998b), establishes a spectacular connection
property that l has pairwise-independent coordi-
between the spectral properties of a finite fam-
nates, but l is not the product measure.
ily of dynamical systems and the nonexistence
Joinings in Ergodic Theory 163

of a pairwise-independent, nonindependent and f1, . . . , fd are bounded measurable functions.

joining: A case ofspecial interestiswhenTi = Ti(i= 1, . . ., d)
for some measure-preserving transformation T,
Theorem 13 which is strongly connected to Furstenberg’s
(Host’s theorem on singular spectrum). Let ergodic proof of Szemeredi’s theorem, and more
ðXi , A i , mi , T i Þ1 i r be a finite family of measure- generally when T1,. . .,Td are commuting
preserving dynamical systems with purely singu- transformations.
lar spectrum. Then any pairwise-independent The study of convergence in L2 of (11) has a
joining l  J(T1, . . ., Tr) is the product measure long history and has led to the emergence of deep
m1      mr. and successful ideas. We can cite in particular the
work of Host and Kra (2005) who established the
L2-convergence for successive powers of a single
Corollary 1 transformation, introducing and describing the
If a dynamical system with singular spectrum is structure of a sequence of factors for which join-
twofold mixing, then it is k-fold mixing for any ings play a central role. The first proof of the
k  2. convergence in L2 of (11) in the case of
d commuting transformations has been given by
The multifold-mixing problem for rank-one Tao in (2008). While Tao’s original proof relies on
measure-preserving systems was solved in 1984 a purely combinatorial argument and does not
by Kalikow (1984), using arguments which do not make use of joinings, we would like here to
involve the theory of joinings. In 1993, Ryzhikov emphasize the alternative proof established by
(Ryzhikov 1993a) extended Kalikow’s result to Austin (2010), which strongly relies on the theory
finite-rank systems, by giving a negative answer of joinings, and how they are involved in the
to Question 1 in the category of finite-rank mixing description of the structure of the factors of a
systems: dynamical system. For an even more joining ori-
ented version of Austin’s proof, see de la
Theorem 14 Rue (2009).
(Ryzhikov’s theorem for finite-rank systems). All As far as pointwise convergence of multiple
finite-rank mixing transformations are PID. ergodic averages is concerned, still more ques-
tions remain open, and they certainly also involve
the theory of joinings. As a simple example, we
Corollary 2 present the relationships between the study of
If a finite-rank transformation is twofold mixing, pointwise convergence of (11) when d = 2, that
then it is k-fold mixing for any k  2. is, we consider two nonnecessarily commuting
transformations S and T and we study the
sequence
Joinings and Multiple Ergodic Averages
Important researches in ergodic theory are n1
1
devoted to the convergence of so-called multiple f T k x g Sk x : ð12Þ
n
ergodic averages, that is to say expressions of the k¼0 n>0
form
It turns out that disjointness of T and S is a
n1 sufficient condition for the almost-sure conver-
1
f 1 T k1 x f 2 T k2 x   f d T kd x , ð11Þ gence to hold. Indeed, let us first consider the
n k¼0 case where T and S are defined on a priori different
spaces ðX, A, mÞ and ðY, B, nÞ , respectively, and
where d  2, T1, . . ., Td are d measure-preserving consider the ergodic average in the product
transformations of the probability space ðX, A, mÞ
164 Joinings in Ergodic Theory

1
n1 since there are examples of transformations which
f T k x g Sk y , ð13Þ are weakly disjoint from themselves.
n k¼0
Recent works use joinings to study the
pointwise convergence of multiple ergodic aver-
which can be viewed as the integral of the func-
ages. We mention in particular Huang et al.
tion f  g with respect to the empirical distribution
(2014), where the case of successive powers of
an ergodic distal transformation is treated, and
n1
1 Donoso and Sun (2016) where this result is
dn ðx, yÞ≔ dðT k x,Sk yÞ :
n k¼0 extended to the case of d commuting transforma-
tions generating a distal action of ℤd, using some
We can always assume that T and S are contin- of Austin’s ideas.
uous transformations of compact metric spaces Also, let us observe that a strong connection
(indeed, any measure-preserving dynamical sys- with Question 1 on pairwise-independent joinings
tem is isomorphic to such a transformation on a has been established in Gutman et al. (2018).
compact metric space: see, e.g., Furstenberg Indeed, using a result of Bourgain according to
1981). Then the set of probability measures on which the almost-sure convergence of (11) holds
X  Y equipped with the topology of weak con- when d = 2 and T1, T2 are powers of a single
vergence is metric compact. Now, here is the transformation, the authors prove that if T is PID
crucial point where joinings appear: If T and and weakly mixing, then (11) converge almost
S are ergodic, we can easily find subsets X0  X surely for Ti = Ti, i = 1, . . ., d.
and Y0  Y with m(X0) = n(Y0) = 1, such that for
all (x, y)  X0  Y0, any cluster point of the
Joinings and Conjectures in Number Theory
sequence (dn(x, y))n>0 is automatically a joining
Joinings have recently proved to be a relevant tool
of T and S. (We just have to pick x and y among the
in the study of a conjecture by Sarnak, which lies
“good points” for the ergodic theorem in their
at the interface between number theory and
respective spaces.) When T and S are disjoint,
dynamical systems. Sarnak conjecture deals with
the only cluster point to the sequence (dn(x, y))
the famous arithmetic Möbius function m, which
is therefore m  n. This ensures that, for continu-
is defined for each integer n  0 by
ous f and g, (13) converges to the product of the
integrals of f and g as soon as (x, y) is picked in mðnÞ≔
X0  Y0. The subspace of continuous functions 0 if there exists a prime p such that p2 j n,
being dense in L2, the classical ergodic maximal k
ð1Þ if n is the product of k distinct prime numbers ðk  0Þ:
inequality (see Garsia 1970) ensures that, for any
f and g in L2(m), (13) converges for any (x, y) in a A famous heuristic called the Möbius random-
rectangle of full measure X0  Y0. ness principle states that the patterns of symbols
Coming back to the original question where the 1, 1, and 0 in the sequence m behave so chaot-
spaces on which T and S act are identified, we ically that m has no correlation with any reason-
observe that with probability one, x belongs both ably simple sequence x. Sarnak proposed a precise
to X0 and Y0, and therefore the sequence (12) interpretation of this principle in the context of
converges. dynamical systems:
The existence of a rectangle of full measure in
which the sequence of empirical distributions
(dn(x, y))n>0 always converges to some joining Conjecture 1
has been studied in Lesigne et al. (2003) as a (Sarnak’s conjecture). For any topological
natural generalization of the notion of disjointness. dynamical system (X, T) with zero topological
This property was called weak disjointness of S and entropy, any continuous real-valued function f
T, and it is indeed strictly weaker than disjointness, defined on X, and any x  X, we have
Joinings in Ergodic Theory 165

n1 n1
1 1
f T k x mðk Þ ! 0: (14) lim sup lim sup akr aks ¼ 0: ð15Þ
n n!1 n
k¼0
n!1
r, s ! 1, k¼0

r, sdifferent primes
Sarnak pointed out that this conjecture was
supported by an older conjecture of Chowla Then, for each bounded multiplicative function
which says that m has no autocorrelation of any n : ℕ ! ℂ, we have
order. More precisely, it can be stated as follows:
n1
1
lim ak  nðkÞ ¼ 0:
N!1 n
Conjecture 2 k¼0
(Chowla’s conjecture). For any integers r  0, i0,
i1, . . ., ir  0 with at least one odd ij, we have To apply the above lemma in the context of the
Sarnak conjecture, we have to consider the sequence
n1 ak = f(Tkx); hence (15) becomes an assumption on
1
mðk Þi0 mðk þ 1Þi1   mðk þ rÞir ! 0: this kind of multiple ergodic average
n k¼0
n!1

n1
1
As pointed out by Sarnak, that the Chowla f ðT r Þk x f ðT s Þk x :
n
conjecture implies the Sarnak conjecture can be k¼0
proved via ergodic theory arguments. Indeed, an
interpretation of the Chowla conjecture is that m is Now, let us assume for simplification that the
the product of m2 (the characteristic function of topological dynamical system (X, T ) has a unique
the set of square-free numbers) with a sequence n invariant probability measure m. Then arguments
p  {1, 1}ℕ behaving as a typical output of an similar to those exposed in section “Joinings and
infinite sequence of balanced coin tosses. Now, Multiple Ergodic Averages” lead to considering
with the assumptions of the statement of the joinings of different prime powers T r and T s of the
Sarnak conjecture, the sequence f(T kx)m2(k) is measure-preserving system (X, T, m). For exam-
produced by a zero-entropy system, whereas the ple, as explained in Bourgain et al. (2013),
sequence p comes from a K-system. Hence (14) disjointness of different prime powers of
can be viewed as a consequence of the T implies the validity of the Sarnak conjecture
disjointness between K-systems and systems of for the system. But Lemma 1 can also be applied
zero entropy. (See, e.g., El Abdalaoui et al. in a larger class of systems, which is defined
(2017) for details.) below and for which we can control all possible
Another reason why joinings are used in the joinings of different prime powers of T.
study of Sarnak conjecture comes from the fol- The ergodic measure-preserving system
lowing criterion introduced by Bourgain et al. (X, m, T ) is said to have Asymptotically Orthogo-
(2013), which provides a sufficient condition for nal Powers (AOP) if for each given f and g in
a bounded sequence (an) to be orthogonal to any L2(m) with Xf dm = Xg dm = 0, we have
bounded multiplicative function n. Recall that the
Möbius function is itself multiplicative, which
lim sup f  gdk ¼ 0:
means that m(mn) = m(m)m(n) whenever m, r, s ! 1, k  J e ðT r , T s Þ XX
n are coprime integers. r, sdifferent primes

Surprisingly, the class of AOP systems

Lemma 1
includes examples where all positive powers of
(Criterion of Bourgain, Sarnak, and Ziegler).
T are isomorphic. Examples of system with AOP
Assume that (an) is a bounded sequence of com-
and consequences of this property are given in
plex numbers, such that
166 Joinings in Ergodic Theory

Ferenczi et al. (2017) and references therein. In a Gaussian system (see Lemańczyk et al. 2000 and
particular, the Sarnak conjecture holds for any Thouvenot 1987), and this gives a complete
uniquely ergodic model of an AOP system, with description of the structure of their factors. This
uniform convergence of (14) with respect to x. kind of analysis is expected to be applicable to
other classes of dynamical systems. In particular,
Gaussian joinings find a nice generalization in the
Future Directions notion of infinitely divisible joinings, studied by
A lot of important open questions in ergodic the- Roy in (2009). These ID joinings concern a wider
ory involve joinings, and we already have cited class of dynamical systems of probabilistic origin,
several of them: joinings are a natural tool when among which we can also find Poisson suspen-
we want to deal with some problems of pointwise sions. The counterpart of Gaussian joinings in this
convergence involving several transformations latter class is Poisson joinings, which have been
(see section “Joinings and Multiple Ergodic Aver- introduced by Derriennic et al. As far as Poisson
ages”). It can therefore be assumed that they will suspensions are concerned, the analog of the GAG
play an important role in future progress on property in the Gaussian class can also be consid-
pointwise convergence of multiple ergodic aver- ered, and examples of Poisson suspensions for
ages. Their use is also fundamental in the study of which the only ergodic self-joinings are Poisson
Rohlin’s question on multifold mixing. As far as joinings have been given in Janvresse et al. (2017)
this latter problem is concerned, we may mention and Parreau and Roy (2007). In Derriennic et al. a
a recent approach to Question 1: start with a general joining property is described: T satisfies
transformation for which some special pairwise- the ELF property (from the French Ergodicité des
independent self-joining exists, and see what this Limites Faibles) if any joining which can be
assumption entails. In particular, we can ask under obtained as a limit of off-diagonal joinings DT nk
which conditions there exists a pairwise- is automatically ergodic. It turns out that this
independent threefold self-joining of T under property is satisfied by any system arising from
which the third coordinate is a function of the an infinitely divisible stationary process (see
two others. It has already been proved in Derriennic et al. and Roy 2009). It is proved in
Janvresse and de la Rue (2007) that if this function Derriennic et al. that the ELF property implies
is sufficiently regular (continuous for some topol- disjointness with any system which is twofold
ogy), then T is periodic or has positive entropy. simple and weakly mixing but not mixing. The
And there are strong evidences leading to the ELF property is expected to give a useful tool to
conjecture that, when T is weakly mixing, such a prove disjointness between dynamical systems of
situation can only arise when T is a Bernoulli shift probabilistic origin and other classes of systems
of entropy log n for some integer n  2. A question (see, e.g., Frączek and Lemańczyk 2004) in the
in the same spirit was raised by Ryzhikov, who case of ℝ-action for disjointness between ELF
asked in Ryzhikov (1992a) under which condi- systems and a class of special flows over irrational
tions we can find a factor of the direct product rotations).
T  T which is independent of both coordinates.
There is also a lot of work to do with joinings in
order to understand the structure of factors of
some dynamical systems and how different clas- Bibliography
ses of systems are related. An example of such a Ahn Y-H, Lemańczyk M (2003) An algebraic property of
work is given in the class of Gaussian dynamical joinings. Proc Am Math Soc 131(6):1711–1716.
systems, i.e., dynamical systems constructed from (Electronic)
the shift on a stationary Gaussian process: For Austin T (2010) Multiple recurrence and the structure of
probability-preserving systems. arXiv:1006.0491
some of them (which are called GAG, from the
Bourgain J, Sarnak P, Ziegler T (2013) Disjointness of
French Gaussien à Autocouplages Gaussiens), it Mobius from horocycle flows. In: From Fourier analy-
can be proved that any ergodic self-joining is itself sis and number theory to Radon transforms and
Joinings in Ergodic Theory 167

geometry, Developments in mathematics, vol 28. Furstenberg H, Peres Y, Weiss B (1995) Perfect filtering
Springer, New York, pp 67–83 and double disjointness. Ann Inst Henri Poincaré Pro-
Bulatek W, Lemańczyk M, Lesigne E (2005) IEEE Trans bab Stat 31(3):453–465
Inform Theory 51(10):3586–3593 Garbit R (2011) A note on Furstenberg’s filtering problem.
Burton R, Rothstein A (1977) Isomorphism theorems in Isr J Math 182:333–336
ergodic theory. Technical report, Oregon State Garsia AM (1970) Topics in almost everywhere conver-
University gence, Lectures in advanced mathematics,
Chacon RV (1969) Weakly mixing transformations which vol 4. Markham Publishing Co., Chicago
are not strongly mixing. Proc Am Math Soc Glasner E (2003) Ergodic theory via joinings, Mathemat-
22:559–562 ical surveys and monographs, vol 101. American Math-
de la Rue T (2006a) 2-fold and 3-fold mixing: why ematical Society, Providence
3-dot-type counterexamples are impossible in one Glasner S, Weiss B (1983) Minimal transformations with
dimension. Bull Braz Math Soc (N.S.) 37(4):503–521 no common factor need not be disjoint. Isr J Math
de la Rue T (2006b) An introduction to joinings in ergodic 45(1):1–8
theory. Discret Contin Dyn Syst 15(1):121–142 Glasner E, Host B, Rudolph DJ (1992) Simple systems and
de la Rue T (2009) Notes on Austin’s multiple ergodic their higher order self-joinings. Isr J Math
theorem, hal-00400975 78(1):131–142
del Junco A (1983) A family of counterexamples in ergodic Glasner E, Thouvenot J-P, Weiss B (2000) Entropy theory
theory. Israel J Math 44(2):160–188 without a past. Ergodic Theory Dyn Syst 20(5):
del Junco A, Rudolph DJ (1987) On ergodic actions whose 1355–1370
self-joinings are graphs. Ergodic Theory Dyn Syst Goodson GR (2000) Joining properties of ergodic dynam-
7(4):531–557 ical systems having simple spectrum. Sankhyā Ser
del Junco A, Rahe M, Swanson L (1980) Chacon’s auto- A 62(3):307–317. Ergodic theory and harmonic analy-
morphism has minimal self-joinings. J Anal Math sis (Mumbai, 1999)
37:276–284 Gutman Y, Huang W, Shao S, Ye X (2018) Almost sure
del Junco A, Lemańczyk M, Mentzen MK (1995) Semi- convergence of the multiple ergodic average for certain
simplicity, joinings and group extensions. Stud Math weakly mixing systems. Acta Math Sin Engl Ser
112(2):141–164 34(1):79–90
Derriennic Y, Fraczek K, Lemańczyk M, Parreau F. (2008) Host B (1991) Mixing of all orders and pairwise indepen-
Ergodic automorphisms whose weak closure of off- dent joinings of systems with singular spectrum. Isr
diagonal measures consists of ergodic self-joinings. J Math 76(3):289–298
To appear in Colloquium Mathematicum 110, No. 1, Host B, Kra B (2005) Nonconventional ergodic averages
81–115 and nil-manifolds. Ann Math 161(1):397–488
Donoso S, Sun W (2016) Pointwise convergence of some Huang W, Shao S, Ye X (2014) Pointwise convergence of
multiple ergodic averages. arXiv:1609.02529 multiple ergodic averages and strictly ergodic models.
El Abdalaoui H, Kulaga-Przymus J, Lemańczyk M, de la arXiv:1406.5930
Rue T (2017) The Chowla and the Sarnak conjectures Janvresse É, de la Rue T (2007) On a class of pairwise-
from ergodic theory point of view. Discret Continuous independent joinings, Preprint. arXiv:0704.3358v2
Dyn Syst 37(6):2899–2944 [math.PR]
Ferenczi S (1997) Systems of finite rank. Colloq Math Janvresse É, Roy E, de la Rue T (2017) Poisson suspen-
73(1):35–65 sions and Sushis. Ann Sci Éc Norm Super
Ferenczi S, Kulaga-Przymus J, Lemańczyk M (2017) 50(6):1301–1334
Sarnak’s conjecture – what’s new. arXiv:1710.04039 Kalikow SA (1984) Twofold mixing implies threefold
Foreman M, Rudolph DJ, Weiss B (2011) The conjugacy mixing for rank one transformations. Ergodic Theory
problem in ergodic theory. Ann Math 173(3):1529–1586. Dyn Syst 4(2):237–259
(English) King J (1986) The commutant is the weak closure of the
Frączek K, Lemańczyk M (2004) A class of special powers, for rank-1 transformations. Ergodic Theory
flows over irrational rotations which is disjoint from Dyn Syst 6(3):363–384
mixing flows. Ergodic Theory Dyn Syst King J (1988) Joining-rank and the structure of finite
24(4):1083–1095 rank mixing transformations. J Anal Math 51:182–227
Furstenberg H (1967) Disjointness in ergodic theory, min- King JL (1990) A map with topological minimal self-
imal sets, and a problem in Diophantine approximation. joinings in the sense of del Junco. Ergodic Theory
Math Syst Theory 1:1–49 Dyn Syst 10(4):745–761. (English)
Furstenberg H (1981) Recurrence in ergodic theory and Krieger W (1970) On entropy and generators of measure-
combinatorial number theory, M. B. Porter Lectures. preserving transformations. Trans Am Math Soc
Princeton University Press, Princeton 149:453–464
Furstenberg H, Keynes HB, Shapiro L (1973) Prime flows Lemańczyk M, Parreau F, Thouvenot J-P (2000) Gaussian
in topological dynamics. Isr J Math 14:26–38. automorphisms whose ergodic self-joinings are Gauss-
(English) ian. Fundam Math 164(3):253–293
168 Joinings in Ergodic Theory

Lemańczyk M, Thouvenot J-P, Weiss B (2002) Relative Science Publications. The Clarendon Press/Oxford
discrete spectrum and joinings. Monatsh Math University Press, New York
137(1):57–75 Ryzhikov VV (1992a) Stochastic wreath products and
Lesigne E, Rittaud B, de la Rue T (2003) Weak disjointness joinings of dynamical systems. Mat Zametki
of measure preserving dynamical systems. Ergodic 52(3):130–140. 160
Theory Dyn Syst 23(4):1173–1198 Ryzhikov VV (1992b) Mixing, rank and minimal self-
Nadkarni MG (1998a) Basic ergodic theory, Birkhäuser joining of actions with invariant measure. Mat Sb
advanced texts: Basler Lehrbücher. [Birkhäuser 183(3):133–160
advanced texts: Basel textbooks], 2nd edn. Birkhäuser Ryzhikov VV (1993a) Joinings and multiple mixing of the
Verlag, Basel actions of finite rank. Funkt- sional Anal i Prilozhen
Nadkarni MG (1998b) Spectral theory of dynamical sys- 27(2):63–78. 96
tems, Birkhäuser advanced texts: Basler Lehrbüacher. Ryzhikov VV (1993b) Joinings, wreath products, factors
[Birkhäuser advanced texts: Basel textbooks]. and mixing properties of dynamical systems. Izv Ross
Birkhäuser Verlag, Basel Akad Nauk Ser Mat 57(1):102–128
Ornstein DS (1970) Bernoulli shifts with the same entropy Sarnak P (2012) Mobius randomness and dynamics. Not S
are isomorphic. Adv Math 4:337–352. 1970 Afr Math Soc 43(2):89–97
Ornstein DS (1972) On the root problem in ergodic theory. Tao T (2008) Norm convergence of multiple ergodic aver-
In: Proceedings of the sixth Berkeley symposium on ages for commuting transformations. Ergodic Theory
mathematical statistics and probability (University of Dyn Syst 28(2):657–688
California, Berkeley, 1970/1971), vol II, probability Thorisson H (2000) Coupling, stationarity, and regenera-
theory. University of California Press, Berkeley, tion, Probability and its applications. Springer,
pp 347–356 New York
Ornstein DS (1974) Ergodic theory, randomness, and Thouvenot J-P (1987) The metrical structure of some
dynamical systems, James K. Whittemore lectures in Gaussian processes. In: Proceedings of the conference
mathematics given at Yale University, Yale mathemat- on ergodic theory and related topics, II (Georgenthal,
ical monographs, vol 5. Yale University Press, New 1986), Teubner-Texte zur mathematik, vol 94. Teubner,
Haven Leipzig, pp 195–198
Parreau F, Roy E (2007) Poisson suspensions with a min- Thouvenot J-P (1995) Some properties and applications of
imal set of self-joinings, Preprint joinings in ergodic theory. In: Ergodic theory and its
Ratner M (1983) Horocycle flows, joinings and rigidity of connections with harmonic analysis (Alexandria,
products. Ann Math 118(2):277–313 1993), London mathematical society lecture note
Rohlin VA (1949) On endomorphisms of compact com- series, vol 205. Cambridge University Press, Cam-
mutative groups. Izvestiya Akad Nauk SSSR Ser Mat bridge, pp 207–235
13:329–340 Veech WA (1982) A criterion for a process to be prime.
Roy E (2009) Poisson suspensions and infinite ergodic Monatsh Math 94(4):335–341
theory. Ergodic Theory Dyn Syst 29(2):667–683 Weiss B (1998) Multiple recurrence and doubly minimal
Rudolph DJ (1979) An example of a measure preserving systems, topological dynamics and applications.
map with minimal self-joinings, and applications. A volume in honor of Robert Ellis. In: Proceedings of
J Anal Math 35:97–122 a conference in honor of the retirement of Robert Ellis,
Rudolph DJ (1990) Fundamentals of measurable dynam- Minneapolis, 5–6 Apr 1995. American Mathematical
ics: ergodic theory on Lebesgue spaces, Oxford Society, Providence, pp 189–196. (English)
 Almost everywhere (a.e.) A measure-theoretic
Entropy in Ergodic Theory statement holds almost everywhere, abbreviated
a.e., if it holds off of a nullset. (Eugene Gutkin
Jonathan L. F. King1 and Kyewon Koh Park2 once remarked to us that the problem with Mea-
1
University of Florida, Gainesville, FL, USA sure Theory is. . . that you have to say “almost
2
Korea Institute for Advanced Study, Seoul, everywhere,” almost everywhere.) For example,
a:e:
South Korea B
A means that m(A\B) is zero. The a.e. will
usually be implicit.
Factor map A factor map c : ðX, X , m, T Þ !
Article Outline ðY, Y , n, SÞ is a measure-preserving map
c : X ! Y which intertwines the transformations,
Glossary c ∘ T ¼ S ∘ c after deleting a nullset (a mass-
Definition of the Subject zero set) in each space. And c is an isomorphism
Entropy Example: How Many Questions? if this c is a bijection and c1 is also a factor
Distribution Entropy map. If X and Y are topological spaces, then we
A Glance at Shannon’s Noisy Channel Theorem require c to be continuous and onto.
The Information Function Fonts We use the font ℋ, h, I for distribution-
Entropy of a Transformation entropy, entropy, and the information function.
Determinism and Zero-Entropy In contrast, the script font A, ℬ, C . . . will be
The Pinsker-Field and K-Automorphisms used for collections of sets; usually subfields of
Ornstein Theory X . Use ðÞ for the (conditional) expectation
Skew Product and Random Walk in Random operator.
Scenery Measure-preserving map A measure-
Topological Entropy preserving map c : ðX, X , mÞ ! ðY, Y , nÞ is a
Entropy of a Flow(ℝ-Action) map c : X ! Y such that the inverse image of
Entropy of ℤd-Actions each B  Y is in X , and m(c1(B)) ¼ n(B).
Finitely Observable Invariant A (measure-preserving) transformation is a
Recent Progress measure-preserving map T : ðX, X , mÞ !
Exodos ðX, X , mÞ. Condense this notation to
Bibliography ðX, X , m, T Þ or (X, m, T).
Measure space A measure space ðX, X , mÞ is a
Glossary set X, a field (i.e., a s-algebra) X of subsets of
X, and a countably-additive measure m : X !
This entry includes some easy proofs and auxil- ½0, 1. (We often just write (X, m), with the
iary results which illustrate how the ideas field implicit.) For a collection C  X , use
evolved in the field. This can be also useful for FldðC Þ
C for the smallest field including C .
persons not in the field of ergodic theory. How- The number “m(B) is the m-mass of B.”
ever, due to the page constraint, we jump to the Notation ℤ¼ integers, ℤ+¼ positive integers, and
brief presentation of recent developments of ℕ¼ natural numbers ¼0, 1, 2, . . .. (Some well-
entropy theory, skipping many important inter- meaning folk use ℕ for ℤ+, saying “Nothing
mediate advances. could be more natural than the positive integers.”
Some of the following definitions refer to And this is why 0  ℕ.) Use de and bc for the
the “Notation” paragraph immediately below. Use ceiling and floor functions; bc is also called the
mpt for “measure-preserving transformation.” “greatest-integer function.” For an interval

© Springer Science+Business Media, LLC, part of Springer Nature 2023 169

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_173
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2020
https://doi.org/10.1007/978-3-642-27737-5_173-2
170 Entropy in Ergodic Theory

J ≔ [a, b)  [1, + 1], let [a. . .b) denote the Clausius circa 1865 (Clausius 1864, 1867), taken
interval of integers J \ ℤ (with a similar conven- from the Greek ϵvtropιa, “a turning towards.”
tion for closed and open intervals). For example, This entry thus begins (Prolegomenon, “introduc-
(e. . .p] ¼ (e. . .p) ¼ {3}. For subsets A and B of tion”) and ends (Exdos (This is the Greek
the same space, O, use AB for inclusion and spelling.), “the path out”) in Greek.
6¼
Clausius, in his coinage, was referring to the
AB for proper inclusion. The difference set B\A
6¼ thermodynamic notion in physics. Our focus in
is fo  Bjo= 2Ag. Employ Ac for the complement this entry, however, will be the concept in mea-
O\A. Since we work in a probability space, if we surable and topological dynamics. (Entropy in
let x ≔ m(A), then a convenient convention is to differentiable dynamics would require an article
have xc denote 1  x, since then m(Ac) equals by itself. For instance, see Katok and Hasselblatt
xc. Use ADB for the symmetric difference [A\B] [ 1995; Ledrappier and Young 1985; Mane 1987;
[B\A]. For a collection C ¼ E j j of sets in O, let Pesin 1977.) Shannon’s 1948 paper (Shannon
the disjoint union jEj or ðC Þ represent the 1948) on Information Theory, then Kolmogorov’s
union [j Ej and also assert that the sets are (1958) and Sinai’s (1959) generalization to
pairwise disjoint. Use “8largen” to mean: “∃n0 dynamical systems, will be our starting point.
such that 8n > n0.” To refer to left hand side of Our discussion will be of mostly the one-
an (20), use LhS(20); do analogously for dimensional case, where the acting-group is ℤ.
RhS(20), the right hand side. We will briefly mention continuous time flow,
Partition A partition P ¼ (A1, A2, . . .) splits ℝ-action. Later we will discuss actions of group ℤd
X into pairwise disjoint subsets Ai  X so that which is a stepping stone to amenable group actions.
the disjoint union iAi is all of X. Each Ai is an My greatest concern was what to call it. I thought of
atom of P. Use jPj or #P for the number of atoms. calling it ‘information’, but the word was overly
When P partitions a probability space, then it used, so I decided to call it ‘uncertainty’. John von
! Neumann had a better idea, he told me, ‘You should
yields a probability vector v , where vj ≔ m(Aj). call it entropy, for two reasons. In the first place your
Lastly, use Phxi to denote the P-atom that owns x. uncertainty function goes by that name in statistical
Probability space A probability space is a mea- mechanics. In the second place, and more impor-
sure space (X, m) with m(X) ¼ 1; this m is a tant, nobody knows what entropy really is, so in a
debate you will always have the advantage.’
probability measure. All our maps/transforma- (Shannon as quoted in Tribus and McIrvine 1971)
tions in this entry are on probability spaces.
!
Probability vector A probability vector v ¼
ðv1 , v2 , . . .Þ is a list of nonnegative reals
whose sum is 1. We generally assume that Entropy Example: How Many Questions?
probability vectors and partitions (see below)
have finitely many components. Write “count- Imagine a dartboard, Fig. 11, split in five regions
able probability vector/partition,” when A, . . ., E with known probabilities. Blindfolded,
finitely or denumerably many components are you throw a dart at the board. What is the expected
considered.
Topological map A topological space ðX, V , T Þ 1
In this entry, unmarked logs will be to base-2. In entropy
is a compact set X with an open cover V ¼ theory, it does not matter much what base is used, but
fU i g. Each Ui is open and [Ui ¼ X. A trans- base-2 is convenient for computing entropy for messages
formation T is a continuous map from X to described in bits. When using the natural logarithm, some
people refer to the unit of information as a nat. In this entry,
itself.
we have picked bits rather than nats. This holds when each
probability p is a reciprocal power of two. For general
probabilities, the “expected number of questions” interpre-
Definition of the Subject tation holds in a weaker sense: Throw N darts indepen-
dently at N copies of the dartboard. Efficiently ask Yes/No
The word “entropy” (originally German, questions to determine where all N darts landed. Dividing by
1
Entropie) was coined by Rudolf Julius Emanuel N, then sending N ! 1, will be the p log p sum of (1).
Entropy in Ergodic Theory 171

hðxÞ≔x log ð1=xÞ, so extending by continuity gives

hð0Þ ¼ 0:
ð2Þ

An interpretation of “h(x)” is the number of

questions needed to winnow down to an event of
probability x.

Distribution Entropy
!
Given a probability vector v , define its distribu-
Entropy in Ergodic Theory, Fig. 1 This dartboard is a
probability space with a 5-set partition. The atoms have
tion entropy as
probabilities 12 , 18 , 18 , 18 , 18 . This probability distribu-
!
tion will be used later in Meshalkin’s example ℋ v ≔ hðxÞ: ð3Þ
!
x v
number Vof Yes/No questions needed to ascertain
the region in which the dart landed? This entry will use the term distropy for “dis-
Solve this by always dividing the remaining tribution entropy,” reserving the word entropy for
probability in half. “Is it A?” if Yes, then V ¼ 1. the corresponding dynamical concept, when there
Else: “Is it B or C?” – if Yes, then “Is it B?” – if No, is a notion of time involved. Getting ahead of
then the dart landed in C, and V ¼ 3 was the number ourselves, the entropy of a stationary process is
of questions. Evidently V ¼ 3 also for regions the asymptotic average value that its distropy
B, D, E. Using “log” to denote base-2 logarithm, decays to, as we look at larger and larger finite
the expected number of questions is thus portions of the process.
!
An equi-probable vector v ≔ K1 ,   , K1 evi-
K
1 1 1 1 1
ðV Þ ¼ 1þ 3þ 3þ 3þ 3 !
dently has ℋ v ¼ log ðK Þ: On a probability
2 8 8 8 8
4 space, the “distropy of partition P,” written
1 note
¼ p j log ¼ 2: ℋ(P) or ℋ(A1, A2, . . .) shall mean the distropy of
pj !
j¼0 probability vector v ¼ ðmðA1 Þ, mðA2 Þ, . . .Þ.
ð1Þ A (finite) partition necessarily has finite
distropy. A countable partition can have finite
distropy, for example, ℋ 12 , 14, 18, . . . ¼ 2.
!
Letting v ≔ 12 , 18 , 18 , 18 , 18 be the probability One could also have infinite distropy: Consider a
vector, we can write this expectation as piece B  X of mass 1/2N Splitting B into 2k many
equal-mass atoms gives an h-sum of 2k(k + N)/
 ðV Þ ¼ hðxÞ: (2k2N). Setting k ¼ kN ≔ 2N  N makes this h-sum
x v
! equal 1; so splitting the pieces of X ¼ 1 N¼1 BN ,
with mðBN Þ ¼ 21N , yields an 1-distropy partition.
Here,
Function h
h: [0, 1] ! [0, 1) is the important function The h(x) ¼ x log (1/x) function has vertical tan-
(There does not seem to be a standard name gent at x ¼ 0, maximum at 1/e and, when graphed
for this function. We use h, since an uppercase in nats slope 1 at x ¼ 1. (Curiosity: Just in this
h looks like an H, which is the letter that paragraph we compute distropy in nats, that is,
Shannon used to denote what we are calling using natural logarithm. Given a small probability
distribution-entropy.) p  [0, 1] and setting x ≔ 1/p, note that hðpÞ ¼
172 Entropy in Ergodic Theory

log ðxÞ
x  1=pðxÞ, where p(x) denotes the number of (b) ℋ(Q _ R)  ℋ(Q) + ℋ(R), with equality
prime numbers less-equal x. (This approximation is a IFF Q ⊥ R.
weak form of the Prime Number Theorem.) Is there (c) For d  0, 12 , the function d 7! ℋ(d, dc) is
any actual connection between the “approximate strictly increasing.
!
distropy” function ℋp p ≔ p  !p 1=pð1=pÞ and (d) R ≼ P implies ℋ(R)  ℋ(P), with equality
a:e:
Number Theory, other than a coincidence of growth IFF R ¼ P.
rate?)
Consider partitions P and Q on the same space Proof Use the strict concavity of h(), together
(X, m). Their join, written P _ Q, has atoms A \ B, with Jensen’s inequality (Figs. 2 and 3).
for each pair A  P and B  Q. They are inde-
pendent, written P ⊥ Q if m(A \ B) ¼ m(A)m(B) for Remark 1 Although we will not discuss it in this
each A, B pair. We write P
Q and say that “P entry, most distropy statements remain true with
refines Q,” if each P-atom is a subset of some “partition” replaced by “countable partition of
Q-atom. Consequently, each Q-atom is a union finite distropy.”
of P-atoms. Recall, for d a real number, our con-
vention that dc means 1  d, in analogy with m(Bc)
equaling 1  m(B) on a probability space. Binomial Coefficients
The dartboard gave an example where distropy
arises in a natural way. Here is a second example.
Distropy Fact For a small d > 0, one might guess that the
For partitions P, Q, R on probability space n
(X, m): binomial co-efficient grows asymptoti-
dn
cally (as n ! 1) like 2en, some small ϵ. But
(a) ℋ(P)  log (#P), with equality IFF P is an what is the correct relation between ϵ and d?
equi-mass partition. Well, Stirling’s formula n !  [n/e]n gives

Entropy in Ergodic Theory, Fig. 2 Using natural log, here are the graphs of: h(x) in solid red, ℋ(x, xc) in dashed
green, 1  x in dotted blue. Both h(x) and ℋ(x, xc) are strictly convex-down. The 1  x line is tangent to h(x) at x ¼ 1
Entropy in Ergodic Theory 173

Entropy in Ergodic
Theory, Fig. 3 Using
natural log: The graph of
ℋ(x1, x2, x3) in
barycentric coordinates; a
slice has been removed,
between z ¼ 0.745 and
z ¼ 0.821. The three arches
are copies of the distropy
curve from Fig. 2

n! nn log ð#XÞ ¼ ℋðP1 _    _ Pn Þ

c  c
½dn!½d n! ½dn½dn ½dc n½d n  ℋðP1 Þ þ . . . þ ℋðPn Þ
1  Hn,
¼ c
ðrecall d ¼ 1  dÞ:
c

ddn ½dc d n
making use of (a), (b) in “Distropy Fact.” And #X
n equals LhS(4).
1
Thus, n log  ℋðd, dc Þ . But by
dn
means of the above distropy inequalities, we A Glance at Shannon’s Noisy Channel
get an inequality true for all n, not just Theorem
asymptotically.
We can restate the Binomial lemma using the
Hamming metric on {0, 1}n,
Lemma 2 (Binomial Lemma) Fix a d  0, 12
and let H ≔ ℋ(d, dc). Then for each n  ℤ+: Distðx, yÞ≔#fi  ½1 . . . njxi 6¼ yi g

n for all x, y  {0, 1}n

 2Hn : ð4Þ
j  ½0...dn
j Use Bal (x, r) for the open radius-r ball cen-
tered at x, and

Proof Let X  {0, 1}n be the set of x with

Balðx, r Þ≔fyjDistðx, yÞ  r g
#fi  ½1 . . . njxi ¼ 1g  dn: for the closed ball. The above lemma can be
interpreted as saying that
On X, let P1,P2, . . . be the coordinate parti-
tions; for example, P7 ¼ A7 , Ac7 , where A7 ≔ {x|
1
j Balðx, dnÞ j
x7 ¼ 1}. Weighting each point by jXj , the uniform
 2ℋðd,d Þn , for each x  f0, 1gn :
c
distribution on X, gives that m(A7)  d. So
ℋ(P7)  H, by (c) in section “Distropy Fact.” ð5Þ
Finally, the join P1 _    _ Pn separates the points
of X. So 1
Corollary 3 Fix n  ℤ+ and d  0, 2 , and let
174 Entropy in Ergodic Theory

H≔ℋðd, dc Þ: think of the noise as a {0, 1}-independent pro-

cess, with Prob(1) ¼n, which is added mod 2 to
Then there is a set C  {0, 1}n, with the signal-process. Suppose we can arrange that
#C 2[1  H]n, that is strongly dn-separated. the set {F(x)) j x  {0, 1}k} of codewords is a
That is, Dist (x, y) > dn for each distinct pair strongly r-separated-set. Then
x, y  C.
The probability that a block is mis  decoded
Noisy Channel is the probability, flipping a n  coin n times
Shannon’s theorem says that a noisy channel has a that we get more than r many Heads:
channel capacity. Transmitting above this speed, ð6Þ
there is a minimum error-rate (depending how
much “above”) that no errorcorrecting code can Theorem 4 (Shannon) Fix a noise-probability
fix. Conversely, one can transmit below – but n  0, 12 and let H ≔ ℋ(n,nc). Consider a rate
arbitrarily close to-the channel capacity, and R < [1  H] and an e > 0. Then 8largen there exists
encode the data so as to make the error-rate less a k and a code F : {0, 1}k ! {0, 1}n so that: The
than any given ϵ. We use Corollary 3 to show the F-code transmits bits at faster than R bits-per-
existence of such codes, in the simplest case second and with error-rate <e.
where the noise is a binary independent-process
(a “Bernoulli” process, in the language later in this Proof Let H0 ≔ ℋ(d, dc), where d > n was
entry). (The noise-process is assumed to be inde- chosen so close to n that
pendent of the signal-process. In contrast, when
the perturbation is highly dependent on the signal, 1
d< and 1  H0 > R: ð7Þ
then it is sometimes called distortion.) 2
We have a channel which can pass one bit per
second. Alas, there is a fixed noise-probability Pick a large n for which
n  0, 12 so that a bit in the channel is perturbed
into the other value. Each perturbation is indepen- k
> R, where k≔b½1  H0 nc: ð8Þ
dent of all others. Let H ≔ ℋ(n, nc). The value n
[1  H] bits-per-second is the channel capacity of
By Corollary 3, there is a strongly dn-separated
this noise-afflicted channel.
set C  {0, 1}n with # C 2½1H n So C is big
0

enough to permit an injection F : {0, 1}k ! C.

Encoding/Decoding The probability of a decoding error is that of
Encode using an “k, n-block-code”; an injective getting more than dn many Heads in flipping a
map F : {0, 1}k ! {0, 1}n. The source text is split n-coin n times (see Courtesy Eq. 6). Since d > n,
into consecutive k-bit blocks. A block x  {0, 1}k the Weak Law of Large Numbers guarantees –
is encoded to F(x)  {0, 1}n and then sent once n is large enough – that this probability is
through the channel, where it comes out perturbed less than the given ϵ.
to a  {0, 1}n. The transmission rate is thus k/n
bits per second.
For this example, we fix a radius r > 0 to
determine the decoding map, The Information Function

Dr : f0, 1gn ! fOopsg Agree to use P ¼ (A1,. . .), Q ¼ (B1,. . .),

f0, 1gk :
R ¼ (C1,. . .) for partitions, and F , G for fields.
With ℭ a (finite or infinite) family of subfields
We set Dr(a) to z if there is a unique z with
of X, their join VG  ℭ G is the smallest field F such
FðzÞ  Balða, r Þ; else, set Dr(a) ≔ Oops. One can
Entropy in Ergodic Theory 175

that G  F, for each G  ℭ. A partition Q can be

8B  F : ðgjF Þdm ¼ gdm:
interpreted also as a field namely, the field of B B
unions of its atoms. A join of denumerably
many partitions will be interpreted as a field, Returning to distropy ideas, use m(A| F ) for the
but a join of finitely many, P1 _    _ PN, will conditional probability function; it is the condi-
be viewed as a partition or as a field, depending tional expectation ð1A jF Þ. So the conditional
on context. information function is
Conditioning a partition P on a positive-ass set
B, let P j B be the probability vector A 7! mðmA\B Þ
ðBÞ . Its 1
I PjF ðxÞ≔ log 1 ðxÞ: ð11Þ
distropy is AP
mðAjF Þ A

m ðA \ B Þ 1 Its integral
ℋðPjBÞ ¼ log :
AP
m ðB Þ m ð A \ B Þ=mðBÞ
ℋðPjF Þ≔ I PjF dm,
So conditioning P on a partition Q gives con-
ditional distropy
is the conditional distropy of P on F .
When F is the field of unions of atoms from
ℋðPjQÞ ¼ mðBÞℋðPjBÞ
BQ
some partition Q, then the number ℋ(P| F )
equals the ℋ(P| Q) from (9).
1
¼ mðA \ BÞ log : Write G j ↗F to indicate that fields G 1 
A  P, B  Q
mðA \ BÞ=mðBÞ
G 2  . . . are nested, and that Fld [1
1 G j ¼ F,
ð9Þ
a.e. The Martingale Convergence Theorem
A “dartboard” interpretation of ℋ(P| Q) is (p. 103 in (Petersen 1983)) gives (c) below.

The expected number of questions to ascertain the

P-atom that a random dart x  X fell in, given that Conditional-Distropy Fact
we are told its Q-atom. Consider partitions P, Q, R and fields F and G j .
For a set A  X, use 1A : X ! {0,1} for its Then
indicator function; 1A(x) ¼ 1 IFF x  A. The
information function of partition P, a map (a) 0  ℋ(P| F )  ℋ(P), with equality IFF
a:e:
I P : X ! [0, 1), is P  F , respectively, P ⊥ F . (We regard P as
a field of unions of atoms for the partition P)
1 (b) ℋðQ _ RjF Þ  ℋðQjF Þ þ ℋðRjF Þ:
I P≔ log 1 ðÞ: ð10Þ
mðAÞ A
AP (c) Suppose G j ↗F . Then ℋ PjG j ↘ℋðPjF Þ.
In particular, ℋðPjF Þ  ℋðPjG Þ if G is a
The information function has been defined so
subfield of F .
that its expectation is the distropy of P,
(d) ℋðQ _ RÞ ¼ ℋðQjRÞ þ ℋðRÞ:
(e) ℋðQ _ R1 jR0 Þ ¼ ℋðQjR1 _ R0 Þ þ ℋðR1 jR0 Þ:
ð I P Þ ¼ I P ðÞdm ¼ ℋðPÞ:
X
Imagining our dartboard (Fig. 1) divided by
Conditioning on a Field superimposed partitions Q and R, equality (d) can
For a subfield F  X , recall that each function interpreted as saying: “You can efficiently discover
g  1 ðmÞ has a conditional expectation where the dart landed in both partitions, by first
ðgjF Þ  1 ðmÞ. It is the unique F -measurable asking efficient questions about R, then – based on
function with where it landed in R – asking intelligent questions
about Q.”
176 Entropy in Ergodic Theory

Entropy of a Process A partition P generates (the whole field) under

1 T P¼m X. It turns out that P generates IFF
T, if V1 n
Consider a transformation (X, m, T) and partition
P ¼ (A1, A2, . . .). Each “time” n determines a P separates points. The equivalence of generating
partition Pn ≔ T nP, whose jth-atom is T n(Aj). and separating is a technical theorem, due to
The process T, P refers to how T acts on the Rokhlin. Assuming m to be Lebesgue is not much
subfield V1 0 Pn  X . (An alternative view of a of a limitation. For instance, if m is a finite measure
process is as a stationary sequence V0, V1, . . . on any Polish space, then m extends to a Lebesgue
of random variables Vn : X ! ℤ+, where Vn(x) ≔ j measure on the m-completion of the Borel sets. To
because x is in the jth-atom of Pn.) not mince words: All spaces are Lebesgue spaces
Write h(T, P) for the “entropy” of the T, P unless you are actively looking for trouble.) That is,
process. It is the limit of the conditional-distropy- after deleting a (T-invariant) nullset, distinct points
numbers of X have distinct T, P-names2.
A finite set {1, . . ., L} of integers, interpreted
cn ≔ℋðP0 jP1 _ P2 _    _ Pn1 Þ: as an alphabet, yields the shift space
X ≔ {1, . . ., L}ℤ of doubly-infinite sequences
This limit exists since ℋ(P) ¼ c1 c2    0. x ¼ (. . .x1x0x1. . .). The shift T : X ! X acts
Define the average-distropy-number 1n hn, where on X by

hn ≔ℋðP0 _ P1 _    _ Pn1 Þ T ðxÞn ≔xnþ1 :

Certainlyhn ¼cn +ℋ(P1 _  _Pn1)¼cn +hn1, The time-zero partition P separates points, under
since T is measure-preserving. Induction gives the action of the shift. This L-atom time-zero parti-
hn ¼ nj¼1 c j . So the Cesàro averages 1n hn con- tion has Phxi ¼ Phyi IFF x0 ¼ y0. So no matter what
verge to the entropy. shift-invariant measure is put on X, the time-zero
partition will generate under the action of T. It is
Theorem 5 The entropy of process (X, P, m, T) Kolmogorov who formulated measure-preserving
equals transformation ðX, X , m, T Þ into the frame of
stationary process, using a generating partition.
1
hðT , PÞ ≔ lim ℋðP0 _    _ Pn1 Þ
n!1 n Time Reversibility
n
¼ lim ℋ P0 VPj A transformation need not be isomorphic to its
n!1 1 inverse. Nonetheless, the average-distropy-
1 numbers show that h(T 1, P) ¼ h(T, P),
¼ ℋ P0 VPj :
1 although this is not obvious from the
conditioning-definition of entropy. Alternatively,
Both limits are nonincreasing. The entropy n
h(T, P) 0, with equality IFF P  V1 1 P jG  ℭ G. ℋ P0 jVPj _
1
And h(T, P)  ℋ(P), with equality IFF T, P is an ¼ ℋðP0 _    _ Pn Þ  ℋðP1 _    _ Pn Þ
independent process. ¼ ℋðPn _    _ P0 Þ  ℋðPn _    _ P1 Þ
n
¼ ℋ P0 jV Pj :
Generators 1

We henceforth only discuss invertible mpts, that (12)

is, when T 1 is itself an mpt. Viewing the atoms of
P as “letters,” then, each x  X has a T, P – name

. . . x2 x1 x0 x1 x2 x3 . . . , 2
We are now at liberty to reveal that our X has always been
a Lebesgue space, that is, measure-isomorphic to an inter-
where xn is PhTn(x)i, the P-letter owning Tn(x). val, [0, 1].
Entropy in Ergodic Theory 177

Bernoulli Processes Wolfgang Krieger later proved a converse to

!
A probability vector v ≔ðv1 , . . ., vL Þ can be the Kolmogorov-Sinai theorem (Krieger 1970).
viewed as a measure on alphabet {1, . . ., L}.
Let m!v be the resulting product measure on Theorem 7 (Krieger Generator Theorem,
X ≔ {1, . . ., L}ℤ, with T the shift on X and 1970) Suppose T ergodic. If h(T) < 1, then
P the time-zero partition. The independent process T has a finite generating partition. Indeed, letting
(T, P) is called, by ergodic theorists, a Bernoulli K be the smallest integer K > h(T), there is a K-
process. It can be considered as a generalization of atom generator. (It is an easier result, undoubtedly
coin tossing. Not necessarily consistently, we tend known much earlier, that every ergodic T has a
to refer to the underlying transformation as countable generating partition – possibly of
a Bernoulli shift. 1-distropy.)
The 12 , 12 -Bernoulli and the 13 , 13, 13 -
Bernoulli have different process-entropies, but Proof See § 7.5 in Rudolph (1990), where
perhaps their underlying transformations are iso- Krieger’s theorem is stated in terms of joinings.
morphic? Prior to the Kolmogorov-Sinai defini-
tion of entropy of a transformation, this question Entropy Is Continuous
remained unanswered. (This is sometimes called Given ordered partitions Q ¼ (B1, . . .) and Q0 ¼
measure(-theoretic) entropy or (perhaps unfortu-
B01 , . . . , extend the shorter by null-atoms until
nately) metric entropy, to distinguish it from topo-
|Q| ¼ |Q0|. Let Fat≔ j B j \ B0j ; this set should
logical entropy. Tools known prior to entropy,
have mass close to 1 if Q and Q0 are almost the
such as spectral properties, did not distinguish
same partition. Define a new partition
the two Bernoulli-shifts; see Cornfel’d et al.
(1982) and ▶ “Spectral Theory of Dynamical Sys-
tems” for the definitions.) Q∇Q0 ≔fFatg Bi \ B0j j with i 6¼ j :

(In other words, take Q _ Q0 and coalesce, into a

Entropy of a Transformation single atom, all the Bk \ B0k sets.) Topologize the
space of partitions by saying that Q(L) ! Q when
The Kolmogorov-Sinai definition of the entropy ℋ(Q ∇ Q(L)) ! 0. (On the set of ordered K-set
of an mpt is partitions (with K fixed) this convergence is the
same as: Q(L) ! Q when m(Fat (Q(L), Q)) ! 1.
hðT Þ≔supfhðT, QÞjQ a partition on Xg: An alternative approach is the Rokhlin metric, Dist
Q (P, Q) ≔ ℋ(P| Q) + ℋ(Q| P), which has the
advantage of working for unordered partitions.)
Certainly entropy is an isomorphism invariant – Then Lemma 8b says that process-entropy varies
but is it useful? After all, the supremum of continuously with varying the partition.
distropies of partitions is always infinite
(on nonatomic spaces) and one might fear that Lemma 8 Fix a mpt (X, m, T). For partitions P,
the same holds for entropies. The key observation Q, Q0, define R ≔ Q ∇ Q0 and let d ≔ ℋ(R). Then
(restated in Lemma 8c and proved below) was
this, from (Kolmogorov 1958; Sinai 1959). (a) jℋ(Q)  ℋ(Q0) j  d. (Distropy varies
continuously with the partition.)
Theorem 6 (Kolmogorov-Sinai Theorem) If (b) |h(T, Q)  h(T, Q0)|  d. (Process-entropy
P generates under T, then h(T) ¼ h(T, P). varies continuously with the partition.)
1 T P : hðT, QÞ 
i
(c) For all partitions Q  _1
Thereupon the 12 , 12 and 13 , 13, 13
hðT, PÞ.
Bernoulli-shifts are not isomorphic, since their
In particular, h T, VLL P‘ ¼ hðT, PÞ.
respective entropies are log(2) 6¼ log (3).
178 Entropy in Ergodic Theory

Proof (a) Evidently Q0 _ R ¼ Q0 _ Q ¼ Q _ R. So Indeed, the failure happens already for process-
ℋ(Q0)  ℋ(Q _ R)  ℋ(Q) + d entropy with respect to a fixed partition.
(b) As above, A Bernoulli process T, P has positive entropy.
Take mpts Sn ! T, each isomorphic to an irrational
N N N rotation. Then each h(Sn, P) is zero, as shown in
ℋ V Q0j  ℋ V Qj þ ℋ V Rj :
1 1 1 the later section “Determinism and Zero-Entropy.”

Sending N ! 1 gives h(T, Q0)  h(T, Further Results

Q) + h(T, R). Finally, h(T, R)  ℋ(R) and When F is a T invariant subfield, agree to use
so h(T, Q0)  h(T, Q) + d. “T F for T-restricted to F ,” which is a factor (see

+
(c) Let K ≔ j Qj. Then there is a sequence of section “Glossary”) of T. Transformations T and
K-set partitions Q(L) ! Q with QðLÞ ≼VLL P‘ . S are weakly isomorphic if each is isomorphic to a
By above, h(T, Q(L)) ! h(T, Q), so showing factor of the other.
that The foregoing entropy tools make short shrift
of the following.
L ?
h T , V P‘  hðT , PÞ Lemma 9 (Entropy Lemma) Consider T-
L invariant sub-fields G j and F .

will suffice. Note that (a) Suppose G j ↗F . Then hT G j ↗hðT F Þ.

+
In
particular, G  F implies that h T G 

+
N 1 L N 1þL
+

hN ≔ℋ V T n
V P‘ ¼ℋ V Pj : hðT F Þ , so entropy is an invariant of weak-
n¼0 L j¼L
isomorphism.
...  :
+

+
So (b) h T G 1 _G 2 _ jh T Gj

And h(T, Q1 _ Q2 _ . . .)  jh(T, Qj).

1 1 1 (c) For mpts (Yj, nj, Sj) : h(S1 S2   ) ¼
hN  ℋ VN0 1 Pj þ 2LℋðPÞ:
N N N jh(Sj).
(d) h(T1) ¼ h(T). More generally,
Now send N ! 1. h(Tn) ¼ j n j  h(T).
Meshalkin’s Map
Entropy Is Not Continuous In the wake of Kolmogorov’s 1958 entropy paper,
The most common topology placed on the space O for two Bernoulli-shifts to be isomorphic one now
of mpts is the coarse topology that Halmos discusses knew that they had to have equal entropies.
in his “little red book” (Halmos 1956). (i.e., Sn ! T Meshalkin provided the first nontrivial example
IFF 8A  X : m S1 n ðAÞDT ðAÞ ! 0 ; this is a
1
in 1959 (Meshalkin 1959).
metric-topology, since our probability space is Let S : Y ! Y be the Bernoulli-shift over the
countably generated. This can be restated in terms “letter” alphabet {E, D, P, N}, with probability
of the unitary operator UT on 2 ðmÞ, where distribution 14 , 14, 14, 14 . The letters E, D, P, N
UT( f ) ≔ f ∘ T. Namely, Sn ! T in the coarse stand for Even, Odd, Positive, Negative, and will
topology IFF U Sn ! UT in the strong operator be used to describe the code (isomorphism)
topology.) between the processes.
The Rokhlin lemma (see p. 33 in Petersen Use T : X ! X for the Bernoulli-shift over
1983) implies that the isomorphism-class of each “digit” alphabet {0, + 1,  1, + 2,  2}, with
ergodic mpt is dense in O, (e.g., see p. 77 in probability distribution 12 , 18, 18, 18, 18 . Both
Halmos 1956) disclosing that the S 7! h(S) map distributions 14 , 14, 14, 14 and 12 , 18, 18, 18, 18
is discontinuous. have distropy log(4).
Entropy in Ergodic Theory 179

We denote S by B 14 , 14, 14, 14 and T by the odd digit +1 or 1, as this D is linked to
1 1 1 1 1 Positive or Negative.
B 2 , 8, 8, 8, 8. After deleting invariant null-
sets from X and Y, the construction will produce
a measure-preserving isomorphism c : X ! Y so Markov Shifts
that T ∘ c ¼ c ∘ S. A Bernoulli process T, P has independence
P(1. . .0] ⊥ P1, whereas a Markov process is a
The Code bit less aloof:
In X, consider this point x: The infinite Past P(1. . .0] doesn’t provide any more
information about Tomorrow than Today did.
... 0 0 0 100 þ1 þ2  1 þ 1 0...
That is, the conditional distribution
P1 j P(1. . .0] equals P1 j P0. Equivalently,
Regard each 0 as a left-parenthesis, and each
nonzero as a right-parenthesis. Link them
ℋðP1 jP0 Þ ¼ ℋ P1 jPð1...0
according to the legal way of matching parenthe-
ses, as shown in the top row, below: note
¼ hðT, PÞ: ð13Þ

000 100 þ1 þ2 1 þ1 0 The simplest nontrivial Markov process (X,

P, m, T) is over a two-letter alphabet {a, b} and
PN N D P P D E D D ?
has transition graph Fig. 4, for some choice of
transition probabilities s and c. The graph’s Mar-
The leftmost 0 is linked to the rightmost +1, as
kov matrix is
indicated by the longest-overbar. The left/right-
parentheses form a 12 , 12  random-walk. Since
s c
this random walk is recurrent, every position in M ¼ mi,j ¼ ,
x will be linked (except for a nullset of points x).
i,j
1 0
Below each 0, write “P” or “N” as the 0 is
where s + c ¼ 1, and mi, j denotes the probability
linked to a positive or negative digit. And below
of going from state i to state j.
the other digits, write “E” or “D” as the digit is
If Today’s distribution on the two states is the
even or odd. So the upper name in X is mapped to !
the lower name, a point y  Y. probability-vector v ≔½pa pb , then Tomorrow’s is
!
This map c : X ! Y carries the upstairs the product v M . So a stationary process needs
! !
B 2 , 18, 18, 18, 18 to B 14 , 14, 14, 14 , downstairs.
1 v M ¼ v . This equation has the unique solution
1 c
pa ¼ 1þc and pb ¼ 1þc . An example of computing
It takes some arguing to show that independence
is preserved. the probability of a word or cylinder set (see
The inverse map c1 views D and E as right- section “The Caratheodory Construction” in
parentheses, and P and N as left. Above D, write “Measure-Preserving Systems”) in the process, is

Entropy in Ergodic Theory, Fig. 4 Call the transition probabilities s ≔ Prob (a ! a) for stay, and c ≔ Prob a ! b for
change. These are nonnegative reals, and s + c ¼ 1
180 Entropy in Ergodic Theory

ms ðbaaabaÞ ¼ pb mba maa maa mab mba 2n intervals. Thus, ℋ(P0 _    _ Pn1  log (2n).
c And 1n log ð2nÞ ! 0.
¼  1 s  s  c  1:
1þc Alternatively, the Shannon-McMillan-
Breiman Theorem implies, for an ergodic process
The subscript on ms indicates the dependence T, P, that the number of length-n names is approx-
on the transition probabilities; let’s also mark the imately 2h(T,P)n; this, after discarding small mass
mpt and call it Ts. Using (13), the entropy of our from the space. But the growth of n 7! 2n is linear
Markov map is and so, for our rotation, h(T, P) must be zero.

hðT s Þ ¼ pa  ℋðs, cÞ þ pb  ℋð1, 0Þ Theorem 10 (Shannon-McMillan-Breiman

¼0
Theorem (SMB-Theorem)) Set h ≔ h(T, P),
ð14Þ where (X, P, m, T) is an ergodic process. Then
1
¼  ½s log ðsÞ þ c log ðcÞ: the average information function
1þc
1 n!1
I ðxÞ ! h, for a:e: x  X: ð15Þ
n P½0...nÞ
Determinism and Zero-Entropy
The functions f n ≔ I P½0...nÞ converge to the con-
Irrational rotations have zero-entropy; let’s reveal stant function h both in the 1 -norm and in prob-
this in two different ways. ability. (In engineering circles, this is called the
Equip X ≔ [0, 1) with “length” (Lebesgue) almost-everywhere equipartition theorem.)
measure and wrap it into a circle. With “ ”
denoting addition mod 1, have T : X ! X be the Proof See the texts of Karl Petersen (p. 261 in
rotation T(x) ≔ x a, where the rotation number a Petersen 1983), or Dan Rudolph (p. 77 in Rudolph
is irrational. Pick distinct points y0, z0  X, and let 1990). See also Paul C. Shields (1996).
P be the partition whose two atoms are the intervals
[y0, z0) and [z0, y0), wrapping around the circle. Consequences
The T-orbit of each point x is dense in X. (Fix an Recall that P[0. . .n) means P0 _ P1 _    _ Pn1,
e > 0 and an N > 1/e. Points x, T(x), . . ., TN(x) where Pj ≔ T jP. As usual, P[0. . .n)(x) denotes the
have some two at distance less than N1 ; say, Dist P½0...nÞ atom owning x.
(Ti(x), Tj(x)) < e, for some 0  i < j  N. Since T is Having deleted a nullset, we can restate (15) to
an isometry, e> Dist (x, Tk(x)) > 0, where k ≔ j  i. now say that 8e, 8 x, 81argen:
So the Tk-orbit of x is ϵ-dense.) In particular, y0 has
dense orbit, so P separates points-hence generates-
? 2ðhþeÞn  m P½0...nÞ ðxÞ  2ðheÞn : ð16Þ
under T. Our goal, thus, is hðT, PÞ ¼ 0.
This has the following consequence. Fixing a
Rotations Are Deterministic number d > 0, we consider any set with m(B) d
The forward T-orbit of each point is dense and so and count the number of n-names of points in B.
is the backward T-orbit. That is, the backward T, The SMB-Thm implies
P-name of each x actually tells us which point x is.
That is, P  V1 n
1 T P, which is our definition of m
“process T, P is deterministic.” Our P being 8e, 8large n, 8B d:
finite, this determinism implies that h(T, P) is j fn  names in Bg j
zero, by Theorem 5. 2ðheÞn : ð17Þ

Counting Names in a Rotation Rank-1 Has Zero-Entropy

The P0 _    _ Pn1 partition places n translates of There are several equivalent definitions for “rank-1
points y0 and of z0, cutting the circle into at most transformtation,” several of which are discussed in
Entropy in Ergodic Theory 181

the introduction of King (1988). (See Chap. 6 in Restricting the random variables to be
Friedman (1970) as well as Ferenczi (1997) and countably-valued, how much of the example sur-
Shields (1973) for examples of stacking vives? Joint work with Kalikow (Kalikow and
constructions.) King 1994) produced a countably-valued station-
A rank-1 transformation (X, m, T) admits a ary V which is nonconsecutively independent as
generating partition P and a sequence of Rokhlin well as deterministic. (Strong determinism is ruled
stacks Sn  X, with heights going to 1, and with out, due to cardinality considerations.) A side-
m(Sn) ! 1. Moreover, each of these Rokhlin effect of the construction is that V’s time-reversal
stacks is P-monochromatic, that is, each level of n 7! Vn is not deterministic.
the stack lies entirely in some atom of P.
Taking a stack of some height 2n, let B ¼ Bn be
the union of the bottom n levels of the stack. There The Pinsker-Field and K-Automorphisms
are at most n many length-n names starting in Bn,
by monochromaticity. Finally, m(Bn) is almost 12 , Consider the collection of zero-entropy sets,
so is certainly larger than d≔ 13 . Thus, Eq. (17)
shows that our rank-l T has zero entropy. Z ¼ Z T ≔fA  X jhðT, ðA, Ac ÞÞ ¼ 0g: ð18Þ

Cautions on Determinism’s Relation to Zero- Courtesy of Lemma 9b, Z is a T-invariant field,

Entropy and
A finite-valued process T, P has zero-entropy iff
P  V1 8Q  Z : hðT, QÞ ¼ 0: ð19Þ
1 P j . Iterating gives

1 1 The Pinsker field of T is this Z. It is maximal

V Pj  V Pj , with respect to (19). (Traditionally, this called
0 1
the Pinsker algebra where, in this context,
that is, the future is measurable with respect to “algebra” is understood to mean “s-algebra.”)

+
the past. Unsurprisingly, the Pinsker factor T Z has zero
entropy, that is, hðT Z Þ ¼ 0 . A transformation
+

This was the case with the rotation, where a

point’s past uniquely identified the point, thus T is said to have completely positive entropy if
telling us its future. it has no (nontrivial) zero-entropy factors.
While determinism and zero-entropy mean the That is, its Pinsker field Z T is the trivial field,
same thing for finite-valued processes, this fails Ø ≔ {Ø, X}.
catastrophically for real-valued (i.e., continuum-
valued) processes, as shown by an example of one K-Processes
of the authors. A stationary real-valued process Kolmogorov introduced the notion of a K-process
V ¼ . . .V1 V0 V1 V2. . . is constructed in King or Kolmogorov process, in which the present
(1992) which is simultaneously becomes asymptotically independent of the dis-
tant past. The asymptotic past of the T, P process
strongly deterministic: The two values V0, V1
determine all of V, future and past. is called its tail field, where

and nonconsecutively independent. This latter 1 M

means that for each bi-infinite increasing integer Tail ðT , PÞ≔ \ V Pj :
1 M ¼1 j¼1
sequence n j j¼1 with no consecutive pair
(always 1 + nj < nj+1), then the list of random This T, P is a K-process if Tail (T, P) ¼ Ø.
variables. . . V n1 V n0 V n1 V n2 . . . is an independent This turns out to be equivalent to what we might
process. call a strong form of “sensitive dependence on
182 Entropy in Ergodic Theory

initial conditions”: For each fixed length L, the complete isomorphism-invariant of Bernoulli
distant future transformations; that is, that two independent pro-
cesses with same entropy necessarily have the
V Pj same underlying transformation. Earlier, Sinai
j  ðG...GþL
(1962) had shown that two such Bernoulli maps
were weakly isomorphic, that is, each isomorphic
becomes more and more independent of
to a factor of the other.
Ornstein introduced the notion of a process
V P j,
j  ð1...0 being finitely determined, see Ornstein (1970)
for a definition, proved that a transformation
as the gap G ! 1. T was Bernoulli IFF it had a finitely determined
A transformation T is a Kolmogorov automor- generator IFF every partition was finitely deter-
phism if it possesses a generating partition P for mined with respect to T and showed that entropy
which T, P is a K-process. completely classified the finitely determined pro-
Here is a theorem that relates the “asymptotic cesses up to isomorphism.
forgetfulness” of Tail (T, P) ¼ ∅, to the lack of Later he developed the equivalent property,
determinism implied by having no zero-entropy “very weak Bernoulli,” which is much easier to
factors (see Walters (1981), p. 113. Related results check and hence used often to prove the
appear in Berg (1975)). Bernoullicity (Ornstein and Weiss 1974). Moreover,
the definition distinguishes Bernoullicity from Kol-
Theorem 11 (Pinsker-Algebra Theorem) Sup- mogorov automorphisms. To define these proper-
pose P is a generating partition for an ergodic ties, he introduced d-metric between two processes,
T. Then Tail (T, P) equals Z T . (X, P, m, T) and (Y, Q, n, S), where jP j ¼ j Qj.
If two processes are d -close, then there exists f :
Since Z T does not depend on P, this means that X ! Y such that for all large n, the average Ham-
all generating partitions have the same tail field, ming metric between n-names of f(P) and Q are
and therefore K-ness of T can be detected from close except on a set of small measure.
any generator. This seminal result and the ideas in the proof led
Another nonevident fact follows from the to a vast machinery for proving transformations to
above. The future field of T, P is defined to be be Bernoulli, as well as classification and structure
Tail (T1, P). It is not obvious that if the present is theorems (Ornstein 1974; Shields 1973; Thouvenot
independent of the distant past, then it is automat- 1975). Showing that the class of K-automorphisms
ically independent of the distant future. (Indeed, far exceeds the Bernoulli maps, Ornstein and
the precise definitions are important; witness sec- Shields produced in Ornstein and Shields (1973)
tion “Cautions on Determinism’s Relation to an uncountable family of nonisomorphic
Zero-Entropy.”) But since the entropy of a pro- K-automorphisms all with the same entropy.
cess, h(T, (A, Ac)), equals the entropy of the time-
reversed process h(T1, (A, Ac)), it follows that
Z T equals Z T 1 . Skew Product and Random Walk in
Random Scenery

Let ðX, X , m, T Þ and ðY, Y , v, SÞ be two

Ornstein Theory
measure-preserving dynamical systems. We call
In 1970, Don Ornstein (1970) solved the long- T on X, X , m ¼ ðX Y, X Y , m nÞ a skew
standing problem of showing that entropy was a product if it is defined as follows
Entropy in Ergodic Theory 183

T ðx, yÞ ¼ Tx, SfðxÞ y very weak Bernoulli which is equivalent to the

property of finitely determined.
i
We note that ℋ Vn1 n1 i
i¼0 T PjVi¼0 T P is in the
where f is a measurable map from X ! ℤ. p
The following is easy to prove order of n ; hence, additional scenery contribu-
tion to the entropy is 0. That is, h T ¼ hðT Þ .
(a) If f(x) 1 for a.e. x, then T ¼ T S and After this first nonconstructive example, many
smooth examples are shown to be
h T ¼ hðT Þ þ hðSÞ. K-automorphisms, but not Bernoulli (Katok
(b) If f(x)dm ¼ 0, then h T ¼ hðT Þ. 1980). Recently random walks in ℤ2 with ℤ2-
sceneries were investigated in den Hollander
(c) If f(x)dm ¼ 1, then h T ¼ 1. et al. (2006) and shown to have the property of
K-automorphism but not Bernoulli.
We mention that this is a special kind of skew When we generalize a scenery space to an
product coming from two measure-preserving independent process (not necessarily B 12 , 12
f independent process), we call the process a ran-
systems. In general, if X, X , m, T !
dom walk with random scenery (RWRS). They
ðX, X , m, T Þ is a factor map, then there exists are all known to be K of same entropy. However,
Y such that T on X can be represented as a skew until recently it was not known if they are
product on X Y: isomorphic. We have the following theorem by
T. Austin (2014).
T ðx, yÞ ¼ ðTx, Sx yÞ

where Sx is a map on f1(x). (We have not Theorem 13 If (X, P, m, T) is an 12 , 12 indepen-

discussed the measure mx on the fiber f1(x).) dent process and (Y1, Q1, n1, S1) and (Y2, Q2, n2,
Ornstein and Shields have shown uncountably S2) are independent process, then T 1 on X Y1
many nonisomorphic K-processes of same entropy and T 2 on X Y2 defined in are isomorphic if and
which are not Bernoulli. Their examples are via only if ℋ(Q1) ¼ ℋ(Q2).
cutting and stacking method which aims at
the specific properties. We let both X ¼ Y This is a finer result than Ornstein and Shields
¼ {1, 1} with 12 , 12 independent probability in the sense that the uncountable family of non-
distribution. We define on T ¼ X, P, m ¼ isomorphic K-automorphisms are the mixing
ðX Y, P Q, m mÞ as follows extension of a K-automorphism (in fact, random
walk space). He proves the above theorem by
ðTx, SyÞ if x0 ¼ 1 introducing a new isomorphic invariant. Instead
T ðx, yÞ ¼ 1 :
Tx, S y if x0 ¼ 1 of explaining the invariant which is quite techni-
cal, we will describe two main ingredients in the
This is called {T, T1}-transformation. This is argument. The first is the mutual (shared) infor-
a well-known “natural” example of a mation defined as
K-automorphism, but it took some time to prove
n
that it is not Bernoulli (Kalikow 1982). ^ ¼ℋ
I T ,P V T^ P
i
^
i¼1

Theorem 12 The {T, T1}-process is not n1

þ ℋ V T^ P
i
^ ℋ
n
V T^ P
i
^
Bernoulli. i¼0 i¼n

Main idea of the proof is that {T, T1}- of the n-step past and n-step future of P . Since
transformation does not have the property of (T, P) is an independent process,
184 Entropy in Ergodic Theory

n n1 ℋðV Þ ¼ ℋtop ðV Þ≔ log ðCardðV Þ:

ℋ V T i P þ ℋ V T i P
i¼1 i¼0

n Analogous to the definitions for partitions,

ℋ V T 1 P define
i¼n

¼ 0:
V _ W≔fV \ W j V  V and W  W g;
Hence, the mutual information comes from T V≔ T 1 ðU Þj U  V
only the scenery part. This makes it possible for and V ½0...nÞ ≔V 0 _ V 1 _    _ V n1 where V i ¼ T i V 0 ; W
V,
us to approximate (recognize) the scenery entropy if each W  patch is a subset of some V  patch:
if n is sufficiently large.
The second is that for sufficiently small num- The T, V entropy is
ber d, the number of d-balls in d, Bd,n ðAÞ ¼
Bjd An1
0 , B0
n1
< d, B  n1 i
i¼0 T P also giv hðT, V Þ ¼ htop ðT, V Þ
es nice to almost the entropy. (The second is the 1 ð20Þ
generalization of Shannon-McMillan Theorem.) ≔ lim sup ℋtop V ½0...nÞ :
n!1 n

And the topological entropy of T is

Topological Entropy
htop ðT Þ ≔supV htop ðT, V Þ, ð21Þ
Adler, Konheim, and McAndrew, in 1965,
published the first definition of topological taken over all open covers V .
entropy in the eponymous article (Adler et al. Thus, htop counts, in some sense, the growth
1965). Here, T : X ! X is a continuous self-map rate or the divergence rate in the T-orbits of
of a compact topological space. The role of atoms length n as in the case of measure space. Evi-
is played by open sets. Instead of a finite partition, dently, topological entropy is an isomorphism
L invariant (Adler and Marcus 1979; Adler and
one uses a finite open-cover V ¼ U j j¼1 , that
Weiss 1967).
is, each patch Uj is open, and their union
[ðV Þ ¼ X . (Henceforth, “cover” means “open Lemma 14 (Subadditive Lemma) Consider a
cover.”) (Because only on a compact space, we sequence s ¼ ðsl Þ1 1  ½1, 1 satisfying
can omit “finite.” Some generalizations of topo-
sk + l  sk + sl, for all k, l  ℤ. Then the following
logical entropy to noncompact spaces require that
limit exists in [1, 1], and lim n!1 snn ¼
only finite open-covers be used (Hasselblatt et al.
inf n snn .
2005).)
Topological entropy, or “top-ent” for short,
Two continuous maps T : X ! X and S : Y ! Y
satisfies many of the relations of measure-entropy.
are topologically conjugate (as isomorphism is
called in this category) if there exists a homeo-
morphism c : X ! Y with cT ¼ Sc. We say Y is a Lemma 15(a) V ≼W implies
factor of X if c is an onto homomorphism.
Let Card(V ) be the minimum cardinality over ℋðV Þ  ℋðW Þ and hðT, V Þ  hðT, W Þ:
all subcovers.
(b) ℋðV _ W Þ  ℋðV Þ þ ℋðW Þ:
CardðV Þ≔Min ♯
V 0 jV 0  V and [ ðV 0 Þ ¼ X , (c) ℋðT ðV ÞÞ  ℋðV Þ, with equality if T is sur-
jective. Also, hðT, V Þ  ℋðV Þ.
and let (d) In (20), the lim n!1 1n ℋ V ½0...nÞ exists.
Entropy in Ergodic Theory 185

(e) Suppose T is a homeomorphism. Then Using a Metric

From now on, our space is a compact metric space
h T 1 , V ¼ hðT, V Þ, (X, d).
Dinaburg (1970) and Bowen (1971, 1973)
for each cover V . Consequently, htop(T1) gave alternative, equivalent, definitions of topo-
¼ htop(T). logical entropy, in the compact metric-space case,
(f) Suppose E is a collection of covers such that: that are often easier to work with than covers.
For each cover W , there exists a V  E with Bowen gave a definition also when X is not com-
V
W . Then htop(T) equals the supremum of pact (see Bowen 1973; Chap. 7 in Walters 1981).
htop ðT, V Þ, just taken over those V  E. (When X is not compact, the definitions need not
(g) For all ‘  ℕ: coincide; e. g., Hasselblatt et al. (2005). And
topologically equivalent metrics, but which are
htop T ‘ ¼ ‘htop ðT Þ not uniformly equivalent, may give the same
T different entropies (see p. 171 in Walters 1981).)

Metric Preliminaries
Proof (c) Let C ≼V be a min-cardinality sub- An ϵ-ball-cover comprises finitely many balls, all
cover. Then TC is a subcover of TV . So Card of radius ϵ. Since our space is compact, every
TV j TC j¼j C j. As for entropy, inequality cover V has a Lebesgue number ϵ > 0. I. e.,
(b) and the foregoing give ℋ V ½0...nÞ  for each z  X, the Bal (Z, e) lies entirely inside at
nℋðV Þ. least one V -patch. (In particular, there is an ϵ-ball-
(d) Set sn ≔ℋ V ½0...nÞ . Then cover which refines V .) Let LEB ðV Þ be the
supremum of the Lebesgue numbers. Courtesy
of Lemma 15f we can
skþl  sk þ ℋ T k V ½0...lÞ  sk þ sl ,

by (b) and (c), and so the Subadditive Lemma 14, Fix a“ universal” listV ð1Þ ≼V ð2Þ ≼ . . . , with V ðkÞ
applies. (g) WLOG, ‘ ¼ 3. Given V a cover, triple 1
a ball  cover:For every T : X ! X, then, the
it to V ≔V \ TV \ T 2 V ; so k
lim k h T, V ðkÞ computes htop ðT Þ:
j
V T3 V ¼ V T i ðV Þ:
j  ½0...N Þ i  ½0:::3N Þ An ϵ-Microscope Three notions are useful in
examining a metric space (X, d) at scale ϵ. Subset
Thus, ℋ T 3 , V , N ¼ ℋðT, V , 3N Þ, extend- A  X is an ϵ-separated-set, if d(z, z0) e for all
ing notation. Part (d) and sending N ! 1, gives distinct z, z0  A. Subset F  X is ϵ-spanning if
8x  X, ∃ z  F with d(x, z) < e.
h T3, V ¼ 3hðT, V Þ. Lastly, take covers such
Lastly, a cover V is ϵ-small if Diam (U) < e, for
that each U  V .

h T 3 , C ðkÞ ! htop T 3 and You Take the High Road and I’ll Take the
Low Road
h T, D ðkÞ ! htop ðT Þ, There are several routes to computing top-ent,
some via maximization, others, minimization.
as k ! 1. Define V ðkÞ ≔C ðkÞ _ D ðkÞ. Apply the Our foregoing discussion computed htop(T) by
above to V ðkÞ , then send k ! 1. a family of sizes f k ðnÞ ¼ f Tk ðnÞ, depending on a
186 Entropy in Ergodic Theory

parameter k which specifies the fineness of scale. Proof (a) Take F  X, a min-cardinality dn -
(In section “Metric Preliminaries,” this k is an ϵ-spanning set. So [z  FDz ¼ X, where
integer; in the original definition, an open cover.)
Define two numbers: note n1
Dz ≔dn  Balðz, eÞ ¼ \ T j Bal T j z, e :
j¼0
f 1
L ðkÞ≔ lim sup log f k ðnÞ and
n!1 n This D≔fDz gz is a cover, and it is dn -
ð22Þ
1 2ϵ-small. Thus, Cov ðn, 2eÞ j D j¼j F j . For
L f ðkÞ ≔ lim inf log f k ðnÞ:
n!1 n any metric, a maxima ϵ-separated-set is automat-
ically ϵ-spanning; adjoin a putative unspanned
f
Finally, let h f ðT Þ≔supk L ðkÞ . If the limit point to get a larger separated set.
exists in (22) then agree to write Lf(k) for the Let A be a max-cardinality dn - ϵ-separated set.
common value. The A-K-M definition used the Take C , a min-cardinality subcover of W ½0...nÞ . For
size f V ðnÞ≔ Card V ½0...nÞ , where each z  A, pick a C -patch Cz ∍ z. Could some
pair x, y  A pick the same C? Well,
CardðW Þ≔Minimumcardinality of write C ¼ \n1j¼0 T
j
W j , with each W j  W . For
a subcover from W : every j  [0. . .n), then,

Here are three metric-space sizes f  (n): d T j ðxÞ, T j ðyÞ  Diam W j < e:

Sepðn, eÞ≔Maximumcardinality of Hence, dn(x, y) < e; so x ¼ y. Accordingly, the

z 7! Cz map is injective, whence j A jj C j.
a dn  e  separated set:
(b) Choose a min cardinality dn - d-small cover
Spnðn, eÞ≔Minimumcardinality of C . For each C  C and j  [0. . .n), the d-Diam
a dn  e  spanning set: (TjC) < d. So there is a V -patch VC, j
Tj(C).
Covðn, eÞ≔Minimum cardinality of Hence
a dn  e  small cover:
note
V ½0...nÞ ∍ \n1
j¼0 T
j
V C,j
C:
These use a list ð dn Þ 1
n¼1 of progressively finer
metrics on X, where
Thus, V ½0...nÞ ≼C . So
dN ðx, yÞ ≔Max j  ½0...N Þ d T ðxÞ, T ðyÞ :
j j

Card V ½0...nÞ  CardðC Þ j C j¼ Covðn, dÞ:

Theorem 16 (All-Roads-Lead-to-Rome Theo-
rem) Fix ϵ and let W be any d - ϵ-small cover. Then (c) To upper-bound Cov (k + l, e) let V and W
be min-cardinality ϵ-small covers, respectively,
(a) 8n: Cov (n, 2e) Spn (n, e) Sep (n, e) for metrics dk and dl. Then V \ T l ðW Þ is a ϵ-
Card W ½0...nÞ . small for dk+l. Consequently Cov (k + l, e)
(b) Take a cover V and a d< LEB ðV Þ. Then 8n: Cov (k, e)Cov (l, e). Thus, n 7! log (Cov
Card V ½0...nÞ  Cov (n, d). (n, e)) is subadditive.
(c) The limit LCov ðeÞ ¼ lim n 1n log Covðn, eÞÞ (d) Pick a V from the list in section “Metric
exists in [0. . .1). Preliminaries,” choose some 2e< LEB ðV Þ
by defn followed by an ϵ-small W from section “Metric
(d) hSep ðT Þ ¼ hSpn ðT Þ ¼ hCov ðT Þ ¼ hCard ðT Þ ¼
Preliminaries.” Pushing n ! 1 gives
htop ðT Þ:
Entropy in Ergodic Theory 187

LCard ðV Þ  LCov ð2eÞ The resulting “K-step TMS (topological Markov

Spn Sep
shift)” is the shift on the set of doubly- 1 Q-
L ðeÞ  L ðeÞ names having no substring in F . In the above
 Spn Sep
L ðeÞ  L ðeÞ magnetic-tape example, K ¼ 576. Making it
L Card
ðW Þ: ð23Þ more realistic, suppose that some string of zeros,
574
say 00    0, is also forbidden. (Perhaps the f-bad-
Now send V and W along the list in section length, 574, is shorter than the l-bad-length
“Metric Preliminaries.” because, say, fs take less tape-space than 1s and
so – being written more densely – cause ambigu-
ity sooner.) Extending to length 577, we get
Pretension
23 ¼ 8 new disallowed words of form
Topological entropy takes its values in [0, 1]. 574
A useful corollary of (23) can be stated in terms of 00    0 b1 b2 b3 .
any Distance (, ) which topologizes [0, 1] as a We recode to a 1-step TMS (just called a TMS
compact interval. or a subshift of finite type) over the alphabet
P ≔ QK. Each outlawed Q-word w0w1  wK
For each continuous T : X ! X on a compact metric-
Sep
engenders a length-2 forbidden P word
space, the Distance LSep ðeÞ, L ðeÞ , goes to zero
(w0, . . . , wK  1)(w1, . . . , wK). The resulting
as e ↘ 0. Consequently, we can pretend that the TMS is topologically conjugate to the original
1
K-step. The allowed length-2 words can be
LSep ðeÞ ¼ lim log ðSepðn, eÞÞ ð24Þ viewed as the edges in a directed-graph and the
n!1 n

limit exists, in arguments that subsequently send set of points x  X is the set of doubly  1 paths
e ↘ 0. Ditto for LSpn(e). through the graph. Once trivialities removed, this
This will be used during the proof of the Var- X is a Cantor set and the shift T : X ! X is a
iational Principle. But first, here are two entropy homeomorphism (see Chaps. 2 and 6 in Lind and
computations which illustrate the efficacy in hav- Marcus 1996).
ing several characterizations of topological
entropy. The Golden Shift
htop (Isometry) ¼0 As the simplest example, suppose our magnetic-
Suppose (T : X, d) is a distance-preserving tape is constrained by the Markov graph, Fig. 5
map of a compact metric-space. Fixing ϵ, a set is that we studied measure-theoretically in Fig. 4.
dn - ϵ-separated IFF it is d  e-separated. Thus, We want to store the text of The Declaration of
Sep
Sep (n, ϵ) does not grow with n. So each L ðeÞ Independence on our magnetic tape. Imagining
is zero. that English is a stationary process, we’d like to
encode English into this Golden TMS as effi-
Topological Markov Shifts ciently as possible. We seek a shift-invariant
Imagine ourselves back in the days when com-
puter data is stored on large reels of fast-moving
magnetic tape. One strategy to maximize the den-
sity of binary data stored is to not put timing-
marks (which take up space) on the tape. This
has the defect that when the tape-writer writes,
say, 577 consecutive 1-bits, then the tape-reader
may erroneously count 578 copies of 1. We side-
step this flaw by first encoding our data so as to Entropy in Ergodic Theory, Fig. 5 Ignoring the labels
577 on the edges, for the moment, the Golden shift, T, acts on
avoid the 11    1 word, then writing to tape.
the space of doubly-infinite paths through this graph. The
Generalize this to a finite alphabet Q and a space can be represented as a subset XGold  {a, b}ℤ,
finite list F of disallowed Q-words. Extend each namely, the set of sequences with no two consecutive
word to a common length K + 1; now F  QK+1 b letters
188 Entropy in Ergodic Theory

measure m on XGold of maximum entropy, should Top-Ent of the Golden Shift

such exist. For a moment, let’s work more generally on an
View P ¼ {a, b} as the time-zero partition on arbitrary subshift (a closed, shift-invariant subset)
XGold; that is, name x ¼ . . .x1x0x1x2. . ., is in X  Qℤ, where Q is a finite alphabet. Here, the
atom b IFF letter x0 is “b.” Any measure m gives transformation is always the shift – but the space
conditional probabilities is varying – so agree to refer to the top-ent as
htop(X). Let NamesX(n) be the number of distinct
mðajaÞ≕s, mðbjaÞ≕c, words in the set {x|[0. . .n)| x  X}. Note that a
note
mðajbÞ ¼ 1,
note
mðbjbÞ ¼ 0: metric inducing the product-topology on Qℤ is

1
But recall, h(T) ¼ ℋ(P1| P[ 1 . . .0))  ℋ dðx, x0 Þ≔ , ð28Þ
1þ j m j
(P1| P0). So among all measures that make the
conditional distribution P j a equal (s, c), the for the smallest jmj with xm 6¼ x0m .
unique one maximizing entropy is the (s, c)-
Markov process. Its entropy, derived in (14), is Lemma 17 Consider a subshift X. Then the
1 1
f ðsÞ≔ ℋðs, 1  sÞ
2s lim
n!1 n
log ðNamesX ðnÞÞ
ð25Þ
1
¼ ½s log ðsÞ þ ð1  sÞ log ð1  sÞ:
2s exists in [0, 1], and equals htop(X).

Certainly f(0) ¼ f(1) ¼ 0, so f’s maximum Proof With e  (0, 1) fixed, two n-names are
occurs at the (it turns out) unique point s where dn - ϵ-separated IFF they are not the same name.
the derivative
p f 0 ðsÞ equals zero. This s ¼ Hence, Sep (n, e) ¼ NamesX(n).
1 þ 5 =2 . Plugging in, the maximum
entropy supportable by the Golden Shift is To compute htop(XGold), declare that a word is
p “golden” if it appears in some x  XGold. Each
2 1 þ 5 2 [n + 1]-golden word ending in a has form wa,
MaxEnt ¼ p log p
5 5 2 1 þ 5 where w is n-golden. An [n + 1]-golden word ending
p
3 5 2 in b, must end in ab and so has form wab, where w is
þ log p : [n  1]-golden. Summing up, NamesXGold ðn þ 1Þ ¼
2 3 5
NamesXGold ðnÞ þ NamesXGold ðn  1Þ. This is the
ð26Þ
Fibonacci recurrence, and indeed, these are the
Fibonacci numbers, since NamesXGold ð0Þ ¼ 1 and
Exponentiating, the number of m-typical
NamesXGold ð1Þ ¼ 2. Consequently, we have that
n-names grows like Gn, where
p
1þp 5
p
3p5
NamesXGold ðnÞ Const  ln ,
2 5 5 2 5 5
G≔ p p : ð27Þ p
1 þ 5 3 5 where l ¼ 1þ2 5 is the Golden Ratio. So the ses-
quipedalian number G from (27) is simply l, and
This expression looks unpleasant to simplify – htop(XGold) ¼ log (l). Since log(l)  0.694, each
it is not even obviously an algebraic number – and thousand bits written on tape (subject to the “no
yet topological entropy will reveal its familiar
nature. This is because the Variational Principle 3
A popular computer-algebra-system was not, at least
(proved in the next section) says that the top-ent of
under our in-expert tutelage, able to simplify this. How-
a system is the supremum of measure-entropies ever, once top-ent gave the correct answer, the software
supportable by the system3. was able to detect the equality.
Entropy in Ergodic Theory 189

bb substrings” constraint) can carry at most 1 1 0

694 bits of information. 0 0 1 :
1 1 0
Top-Ent of a General TMS
A (finite) digraph G engenders a TMS T : XG ! XG, Its j  j-largest eigenvalue is still l, as it must.
as well as a {0, 1}-valued adjacency matrix Now we make a new graph. We modify Fig. 5
A ¼ AG, where ai, j is the number of directed- by manufacturing a total of two s-edges, seven
edges from state i to j. (Here, each ai, j is 0 or 1.) c-edges, and three edges 11, 12, 13. Give these
The (i, j)-entry in power An is automatically the 2 + 7 + 3 edges twelve distinct labels. We could
number of length-n paths from i to j. Employing the compute the resulting TMS-entropy from the
matrix-norm kM k ≔ i, j jmi, jj, then, corresponding 12 12 adjacency matrix. Alter-
natively, look at the (a, b)-adjacency matrix
kAkn ¼ NamesX ðnÞ:
2 7
Happily Gelfand’s formula (see 10.13 in Rudin A≔ :
3 0
1973 or Spectral_radius in Wikipedia) applies:
For an arbitrary (square) complex matrix, The
p roots of its characteristic polynomial are
1  22 .pHence, htop of this 12-symbol TMS is
1
lim kAn kn ¼ SpecRadðAÞ: ð29Þ log 1 þ 22 .
n!1

This right hand side, the spectral radius of A, The Variational Principle
means the maximum of the absolute values of Let M ≔ M(X, d) be the set of Borel probability
A’s eigenvalues. So the top-ent of a TMS is thus measures, and M(T) ≔ M(X, d, T) the set of
the T-invariant m  M. Assign

htop ðXG Þ ¼ SpecRad ðAG Þ EntSupðT Þ≔sup hm ðT Þjm  MðT Þ

: ¼ Maxfjej j e is an eigenvalue of AG g:
ð30Þ Theorem 18 (Variational Principle
(Goodson)) EntSup (T) ¼ htop(T).
The (a, b)-adjacency matrix of Fig. 5 is This says that top-ent is the top entropy – if
there is a measure m which realizes the supremum.
1 1 There doesn’t have to be. Choose a sequence of
, metric-systems (Yk, mk, Sk) whose entropies
1 0
strictly increase htop(Sk) ↗ L to some limit in
whose eigenvalues are l and 1 (0, 1]. Let (Y1 , m1, S1) be the identity-map
l.
on a 1-point space. Define a new system (X, d,
T), where X ≔ k  [1. . .1]Yk. Have T(x) ≔ Sk(x),
Labeling Edges for the unique k with Yk ∍ x. As for the metric, on
Interpret (s, c, 1) simply as edge-labels in Fig. 5. Yk let d be a scaled version of mk, so that the
The set of doubly - 1 paths can also be viewed as d-Diam (Yk) is less than 1/2k Finally, for points
a subset YGold  {s, c, 1}ℤ, and it too is a TMS. in distinct components, x  Yk and z  Y‘ , decree
The shift on YGold is conjugate (topologically iso- that d(x, z) ≔ j 2k  2‘j. Our T is continuous,
morphic) to the shift on XGold, so they a fortiori and is a homeomorphism if each of the Sk
have the same top-ent, log(l). The (s, c, 1)- is. Certainly htop(T) ¼ L > htop(Sk), for every
adjacency matrix is k  [1. . .1].
190 Entropy in Ergodic Theory

If L is finite then there is no measure m of m-nice too, then U ðAc Þ  mðAc Þ. Thus, limLaL(A)
maximal entropy; for m must give mass to some exists and equals m(A). Because C≔A is closed, the
Yk; this pulls the entropy below L, since there are continuous functions fN ↘ 1C pointwise, where
no compensatory components with entropy
exceeding L. f N ðxÞ ≔1  Min ðN  dðx, CÞ, 1Þ:
In contrast, when L ¼ 1 then there is a
maximal-entropy measure (put mass 1/2j on By the Monotone Convergence theorem, then,
some component Y k j , where kj ↗ 1 swiftly);
indeed, there are continuum-many maximal- N
f N dm ! mðCÞ:
entropy measures. But there is no ergodic measure
of maximal entropy. (The ergodic measures are
the extreme points of M(T); call them MErg(T). And m(C) ¼ m(A), since A is nice. Fixing N,
This M(T) is the set of barycenters obtained from then, it suffices to establish U ðAÞ  f N dm. But
Borel probability measures on MErg(T) (see fN is continuous, so
Krein-Milman theorem, Choquet_theory in
Wikipedia). In this instance, what explains the f N dm ¼ lim sup f N daL
failure to have an ergodic maximal-entropy mea- L!1
sure? Let mk be an invariant ergodic measure on
Yk. These measures do converge to the one-point lim sup 1A daL ¼ U ðAÞ:
L!1
(ergodic) probability measure m1 on Y1. But the
map m 7! hm(T) is not continuous at m1.) Corollary 20 Suppose aL ! m, and partition P is
For a concrete L ¼ 1 example, let Sk be the m-nice. Then ℋaL ðPÞ ! ℋm ðPÞ.
shift on [1. . .k]ℤ (see Buzzi and Ruette 2006;
Denker 1976; Misiurewicz 1973 for measures of The diameter of partition P is MaxA  P
maximal entropy.) Diam(A).

Topology on M Proposition 21 Take m  M and e > 0. Then

Let’s arrange our tools for establishing the Varia- there exists a m-nice partition with Diam (P) < e.
tional Principle. The argument will follow Mis-
iurewicz’s proof, adapted from the presentations Proof Centered at an x, the uncountably many
in Brin and Stuck (2002) and Walters (1981). balls {Bal(x, r)| r  (0, e)} have disjoint bound-
Equip M with the weak-  topology. (This aries. So all but countably many are m-nice; pick
metrizable topology makes M compact. Always, one and call it Bx. Compactness gives a finite nice
M(T) is a nonvoid compact subset (see cover, say, {B1, . . ., B7}, at different centers. Then
“Measure-Preserving Systems”).) That is, mea- the partition P ≔ (A1, . . ., A7) is nice, where
sure aL ! m IFF fdaL ! fdm, for each contin- Ak ≔Bk ∖[k1j¼1 B j . (For any two sets B, B  X, the
0
uous f : X ! ℝ. An A  X is m-nice if its
union @B [ @B is a superset of the three boundaries
0
topological boundary @(A) is m-null. And a parti-
@(B [ B0), @(B \ B0), @(B\B0).)
tion is m-nice if each atom is.

Proposition 19 If aL ! m and A  X is m nice, Here is a consequence of Jensen’s inequality.

then aL(A) ! m(A).
Lemma 22 (Distropy-Averaging Lemma) For
Proof Define operator U ðDÞ≔ lim supL aL ðDÞ. It m, v  M, a partition R, and a number
suffices to show that U ðAÞ  mðAÞ. For since Ac is t  [0, 1],
Entropy in Ergodic Theory 191

tℋm ðRÞ þ ð1  tÞℋn ðRÞ  ℋtmþð1tÞn ðRÞ: Remark 24 The idea in the following proof is to
mostly fill interval [0 . . . L) with N-blocks,
Proof of the Variational Principle As usual we starting with a offset K  [0. . N). Averaging
divide the proof into 2 parts. EntSup(T) htop(T) over the offset will create a Cesàro average over
and EntSup(T) htop(T). each N-block. Averaging over the N-blocks will
Strategy for EntSup(T) htop(T). Choose an allow us to compute distropy with respect to the
e > 0. For L ¼ 1, 2, 3, . . ., take a maximal averaged measure, aL.
(L, e)-separated-set FL  X, then define
Proof (of (32)) Since L is fixed, agree to use ’ for
1 the ’L probability measure. Our dL - ϵ-separated
F ¼ Fe ≔ lim sup log ðjFL jÞ:
L!1 L set FL has at most one point in any given atom of
Q[0. . .L), thereupon
Let ’L() be the equi-probable measure on FL;
each point has weight 1/ j FLj. The desired invari- log ðjFL jÞ ¼ ℋ’ Q½0...LÞ :
ant measure m will come from the Cesàro averages

1
aL ≔ T ‘ ’L , Regardless of the “offset” K  [0 . . . N), we
L
‘  ½0...LÞ can always fit C≔ LN N many N-blocks into
[0. . .L). Denote by G ðK Þ≔½K . . . K þ CN Þ, this
which get more and more invariant. union of N-blocks, the good set of indices. Let
ℬðK Þ≔½0 . . . LÞ∖G ðKÞ be the bad index-set.
Lemma 23 Let m be any weak- accumulation Therefore,
point of the above faL g11 . (Automatically, m is
T-invariant.) Then hm(T) F. Indeed, if Q is any Bad ðK Þ
m-nice partition with Diam (Q) < e, then hm(T,
Q) F. ℋ’ Q½0...LÞ ℋ’ V Qj þ
jℬðK Þ

Tactics As usual, Q[0. . .N) means Good ðK Þ

Q0 _ Q1 _    _ QN1. Our goal is to show:

ℋ’ V Qj : ð33Þ
iG ðK Þ
? 1
8N : F  ℋ Q : ð31Þ
N m ½0...NÞ Certainly Bad (K)  3N log (|Q|). So
Fix N and P ≔ Q[0. . .N), and a d > 0. It suffices 1 3N
to verify: 8largeL  N, BadðK Þ  log ðjQjÞ:
NL L
K  ½0...N Þ

1 1?
log ðjFL jÞ  d þ ℋaL ðPÞ, ð32Þ This is less than d, since L is large. Applying
L N
1
NL K  ½0...N Þ to (28) now produces
since this and Corollary 20 will prove (31): Push-
ing L ! 1 along the sequence that produced m 1 1
log ðjFL jÞ  d þ Good ðK Þ: ð34Þ
essentially sends LhS(32) to F, courtesy (24). And L NL K
RhS(32) goes to d þ N1 ℋm ðPÞ, by Corollary 20,
since P is m-nice. Descending d ↘ 0, hands us the Note
needed (31).
192 Entropy in Ergodic Theory

V T j ðQ Þ ¼ V : Applied to a power T‘, this asserts that hm(T‘)

j  G ðK Þ c  ½0...CÞT KþcN ðPÞ
 2 + htop(T‘). Thus, hm ðT Þ  2‘ þ htop ðT Þ, using
Lemma 9d and using Lemma 15g. Now let
So ‘ ! 1.

GoodðK Þ  ℋ’ T KþcN P :
c
Entropy of a Flow(ℝ-Action)
This latter, by definition, equals
c ℋT KþcN
ð’Þ ðPÞ. We conclude that Historically study of dynamical systems started
with the continuous ℝ-action governed by New-
1
GoodðK Þ 
1
ℋT KþcN ’ ðPÞ
ton’s law. We denote the system by ðO, G, n, T t Þ
NL K
NL K c where T t2 ðT t1 ðxÞÞ ¼ T t2 þt1 ðxÞ. The geodesic flows

1
ℋT ‘ ’ ðPÞ, and horocycle flows in homogeneous dynamics
NL
‘  ½0...LÞ are well-known example of ℝ-actions. For each
by adjoining a few translates of P, given t0 , ðO, G, n, T t0 Þ gives rise to a ℤ-action.
1 We define h(Tt) ¼ h(T1), entropy of time 1 map. It
 ℋaL ðPÞ,
N is not hard to see that h(Tct) ¼ ch(Tt) for every
by the Distropy  Averaging Lemma 22,
c  ℝ. In particular, If c is an integer, then h-
1 ‘
(Tct) ¼ h(Tc) ¼ ch(T1) ¼ ch(Tt) (see Lemma 9d).
since aL is the average L ‘T ’ . Thus, (34) Ambrose (1941) has shown that every ergodic
implies (32), our goal. flow ðO, G, n, T t Þ can be represented as a flow
with a transformation on X and a ceiling function
Proof of EntSup(T) htop(T). Fix a T-invariant f(x), which is measurable with respect to X. When
measure m. For partition Q ¼ (B1, . . . , BK), we denote the base by ðX, X , m, T Þ, then the flow Tt
choose a compact set Ak  Bk with m(Bk\Ak) acts as follows: each point o ¼ (x, t) moves
small. (This can be done, since m is automatically straight up at unit speed until it hits (x, f(x))
a regular measure (Rudin 1973).) Letting which is identified as (Tx, 0). Then it continues
D ≔ [ iAi]c and P ≔ (D, A1, . . . , AK), we can up at unit speed. The base ðX, X , m, T Þ is called a
have made ℋ(P| Q) as small as desired. Courtesy cross section of the flow. It is clear that the flow
of Lemma 8b, then, we only need consider parti- has positive entropy IFF the base ðX, X , m, T Þ has
tions of the form that P has. positive entropy. Unpredictability of the base is
Open-cover V ¼ ðU 1 , . . ., U K Þ has patches inherited by the flow.
Uk ≔ D [ Ak. What atoms of, say, P[0,1,2], can the The flow has the following structure (Fig. 6).
intersection U9 \ T1(U2) \ T2(U5) touch? Only
the eight atoms

f (x)
ðD or A9 Þ \ T 1 ðD or A2 Þ \ T 2 ðD or A5 Þ:
(x, f (x))
Thus, ♯ P½0...nÞ  2n:♯ V ½0...nÞ . Here ♯() counts
the number of nonvoid atoms/patches.) So

1 1 (x, t)
ℋ P  1 þ log ♯ V ½0...nÞ
n m ½0...nÞ n
x (X, X , µ, T ) Tx
 1 þ 1 þ htop ðT Þ;

this last inequality, when n is large. The upshot: Entropy in Ergodic Theory, Fig. 6 In the representation
hm(T)  2 + htop(T). of ðO, G, n, T t0 Þ, (x, f(x)) is identified as (Tx, 0)
Entropy in Ergodic Theory 193

(i) O ¼ {(x, t) : x  X, 0  t  f(x)}. t0 if x  P0

(ii) The s-algebra G is the product s-algebra of f ðxÞ ¼
t1 if x  P1
X and Lebesgue on the time axis. Also the
measure n ¼ m Lebesgue measure. where {t0, t1} are irrationally related. It is known
(iii) f(x)dm ¼ 1. that the flow is a Bernoulli flow. That is, for every
t0 , T t0 is Bernoulli. Hence, by the isomorphism
Rudolph strengthened the representation theorem of Bernoulli transformation (Entropy
showing that the ceiling function f(x) can have determines a Bernoulli transformation up to iso-
only two irrationally related values, say t0 and t1 morphism), we can realize every Bernoulli trans-
where tt01 is irrational. Let Q ¼ {Q0, Q1} be the formation via a Bernoulli flow.
partition of X where Q0 ¼ {x : f(x) ¼ t0}. Let (X, P,
m, T) be the base, where P ¼ {P0, P1, . . ., Pk1} is a
generating partition. We denote R ¼ P _ Q. The
partition R has at most 2k atoms. Let R ¼ {R1, . . ., Entropy of ℤd-Actions
Rl}. Let R ¼ R1 , . . ., Rl be the partition of O,
where Ri ¼ fðx, tÞ : x  Ri , 0  t < f ðxÞg. By a ℤd-action, s, we mean commuting transfor-
!
mation s ¼ {T1, T2, . . ., Td}. An element n ¼
Theorem 25 The partition R is a generating ðn1 , n2 , . . ., nd Þ  ℤd corresponds to T n11
!
partition of O under any ergodic transformation Tg. T 2   T d ¼ s n acting on ðX, X , mÞ. Note that
n2 nd

study of ℤd-actions is a stepping stone of amena-

Proof For the simplicity of argument, we assume ble group-actions. For simplicity of notations we
that g < min {t0, t1}. Let (x1, t1) and (x2, t2) be assume d ¼ 2. (The following definitions can be
two distinct points in O. If x1 and x2 belong to easily generalized to ℤd-actions, d 2.)
different atoms of R, then so do (x1, t1) and (x2, t2) Like a transformation, we define
of R.
1
Suppose (x1, t1) and (x2, t2) is in Ri . hðs, PÞ ¼ n!1
lim ℋ V sði,jÞ P
n2 ði, jÞ  Rn
(i) If x1 6¼ x2 and t1 ¼ t2, then different R-names
of x1 and x2 give rise to different R-names of where Rn ¼ [0, n) [0, n) is a square and
(x1, t1) and (x2, t2).
(ii) If x1 ¼ x2, then (x1, t1) and (x2, t2) are in the hðsÞ ¼ sup hðs, PÞ:
P
same orbit of Tt. We assume t1 < t2 and let
d ¼ t2 t1
2 . Let It is not hard to see that

A ¼ ðx, tÞ : x  Ri0 , T 1 x  Ri1 ði0 6¼ i1 Þ, 0  t  d :

hðsÞ ¼ h s, P
Since nA > 0 and Tg is ergodic, there exists
n such that T ng ðx, t2 Þ  A . Hence, where P is a generating partition under s. That is,
P satisfies
R T ng ðx, t2 Þ  Ri0 and R T ng ðx, t1 Þ  Ri1 .
(iii) When x1 6¼ x2 and t1 6¼ t2, a little more thought lim V sði,jÞ P~ ¼ X :
leads to the above case either (i) or (ii). n!1 ni,jn

Consider the following flow of Totoki type We also have that

(Shields 1973).
Let (X, P, m, T) be an independent process and hðs, PÞ ¼ ℋðPjP Þ
194 Entropy in Ergodic Theory

where P ¼ V(i, j) ≺ (0, 0)s(i, j)P and “≺” is a Hence, h(T1) ¼ 1. Similar reasoning for topo-
lexicographic order. logical entropy gives that if htop(s) > 0, then
In more general amenable groups, the htop(s(i, j)) ¼ 1 for all (i, j)  ℤ2.
sequence of square Rn is replaced by a sequence
of Følner sets. Existence of Følner sequence Complexity of Zero Entropy Actions
guarantees the existence of invariant probability As the study of dynamical systems expands to big-
measure under the action (Ornstein and Weiss ger group actions, entropy zero systems arise more
1987). naturally with diverse growth rates of orbits. We
If h(s, P) > 0, then ℋ Vði,jÞ  Rn sði,jÞ P is in consider the following ℤ2-action s ¼ {T1, T2}
the order of n2. We have the following theorem. where h(s) ¼ 0.

Theorem 26 If h(s) > 0, then h T i1 T 2j ¼ 1. Let s ¼ {T1, T2} be an action where T1 and T2
2 are irrational rotations by a1 and a2 respec-
1 8ði, jÞ  ℤ ∖fð0, 0Þg.
tively on a circle. Then h(T1) ¼ h(T2) ¼ 0.
Proof For simplicity we will prove for the case Hence, h(s) ¼ 0.
(i, j) ¼ (1, 0). Let P be a generating partition under 2. We have the following ℤ2-subshift called three
s. We note that a generating partition P under s is dot model.
not necessarily a generating partition for Let X ¼ fðxði,jÞ : xiþ1,j þ xi,j þ xi,jþ1
T1 ¼ T(1,0). Let Pm denote the partition 2
k
0 ðmod 2Þg  f0, 1gℤ . It is not hard to see
Vmj¼k T 2j P and let X m ¼ V1 i m
1 T 1 Pm . Note that
that lim m!1 X m ↗X .
1
(a) lim n!1 2n ℋtop Vði,jÞ  Rn sði,jÞ P ¼ 1.
hðT 1 Þ ¼ h T , Pm
m Hence, htop(s) ¼ 0.
¼ lim ℋ Pm i m
1 (b) For each Pmm ,
m!1 m j V T 1 Pm
i¼1

m 1 k1 1
¼ lim ℋ T k2 Pj V T i Pm _ V T j P
m!1
k¼m
i¼1 1 m j¼m 2 ℋtop Pm i m
m j V T 1 Pm
i¼1
m 1 k1 1
ℋ Pj T k V T i m
V T j
¼ lim
m!1 2 i¼1 1 Pm _ j¼m 2 P ¼ ℋtop T 2 Pj V T i
m
Pm VPm1 ¼ 1:
k¼m i¼1 1 m m

Since P ¼ V(i, j) ≺ (0, 0)s(i, j)P is finer algebra Hence, htop(T1) ¼ 1. Likewise htop(T2) ¼ 1.
j
than T k
2 V1 i m k1
i¼1 T 1 Pm _ V j¼m T 2 P ,
by Conditional-Distropy Fact (c), we have There exists an obvious invariant measure m on
X such that each atom of V(i, j)  [0, m) [0, n)
j s(i, j)P has measure 2(m+n). With respect to m,
ℋ PjT k
2 V1 i m k1
i¼1 T 1 Pm _ V j¼m T 2 P we have h(T1) ¼ h(T2) ¼ 1. Hence, h(s) ¼ 0.
> hðs, PÞ More generally, if Y is a shift space with the left
shift T and f is a continuous map from Y to itself,
for each m. Hence then it is well known that f is a block code since f
is uniformly continuous. If f is not invertible,
hðT 1 Þ ℋ Pm 1 i m then we use the “natural extension method” to
m jVi¼1 T 1 Pm
m extend the space Y to X  {0, 1}ℤ as in the
hðs, PÞ above example (2). If Y has an invariant measure
k¼m and f is measure-preserving, then X carries an
¼ ð2m þ 1Þ hðs, PÞ for all m invariant measure.
Entropy in Ergodic Theory 195

To investigate the complexity of entropy zero noncocompact subgroup action of ℤ2. To under-
ℤ -actions, J. Milnor (1988) introduced the notion
2
stand entropy zero ℤd-actions (d > 2), we need to
! !
of directional entropy, h v for v ¼ ðx, yÞ  ℝ2 investigate the complexity of noncocompact sub-
group actions of ℤi (0  i  d-1) in addition to
as follows. Let P be a partition of X. We denote the
directions.
partition s(i, j)P by P(i, j)
In the case of ℤ2-action generated by a shift
and a Cellular Automaton map (a block code),
! 1 J. Milnor asked if the directional entropy function
h v , P ¼ lim m!1 lim t!1 ℋ V Pði,jÞ
t !
ði, jÞ  P ðm, v , tÞ is continuous. We have the following answer
(Park 1999).
!
where P m, v , t ¼ fði, jÞ: 0  j  ½ty, m  j
! !
x
 i  m þ j xyg and if y ¼ 0, then P m, v , t ¼ Theorem 27 Directional entropy h v is con-
y
!
fði, jÞ : m  j  m, 0  i  ½txg. That is, tinuous on k v k¼ 1 for the action s generated by
! a shift T and a Cellular Automaton map f.
P m, v , t is a parallelogram with width [m,
!
m] along the vector v . Both limsups in the defini- !
!
Sinai (2017) has shown that h v is upper
tion of h v , P are equal to limits. We define !
semi-continuous on k v k¼ 1. In fact, the conti-
! !
h v ¼ supP h v , P . If P is a generating parti- nuity is guaranteed for more general s. Topolog-
! ! !
tion under s, then h v ¼ h v, P since ical directional entropy htop v can be defined
!
X m ¼ V1t¼1 Vði,jÞ  P ðm,!
v ,tÞ P
ði,jÞ
↗X . analogously. It is not yet known if htop v is
Directional entropy has the following continuous in this restrictive class of ℤ -actions.
2
2
properties. Let ðX, X , m, sÞ be a subshift of f0, 1gℤ .
As was remarked via the examples mentioned
! ! above, entropy zero systems exhibit diverse
(i) h c v ¼ ch v :
behavior. For the growth rate of orbits of zero
! !
(ii) h T i1 T 2j ¼ h v where v ¼ ði, jÞ  ℤ2 entropy systems, several notions, slow entropy
! (Katok and Thouvenot 1997) and entropy dimen-
(iii) h v is not continuous in general for sion (Dou et al. 2019; Jung et al. 2017), have been
!
k v k¼ 1. introduced to understand the complexity. They
intend to measure the intermediate growth rate
We note that which is bigger than polynomial and less than
exponential. Although the motivation is mainly
! ! from the study of bigger group actions, investiga-
(a) Example (1) satisfies that h v ¼ 08 v  S1 .
tion of the growth rate of orbits in entropy zero
! !
(b) Example (2) satisfies that h v > 08 v  S1 . ℤ-actions is necessary for more complete
understanding.
However, an example with the following prop-
erties is known.
Finitely Observable Invariant
(i) h(s(1,0)) > 0.
Having given the basic properties of entropy, we
(ii) h(s(i, j)) ¼ 0 8 (i, j) 6¼ (1, 0)
would like to make a remark on its characteristic
! !
of finitely observable invariant in ergodic sys-
If there exists a direction v such that h v > tems. In a landmark paper (Ornstein and Weiss
0, then the growth rate of orbits along Rn is at least 2007), Ornstein and Weiss show that all “finitely
in the order of 2n. Directional entropy can be observable” properties of ergodic processes are
regarded as a generalization of the entropy of secretly entropy; indeed, they are continuous
196 Entropy in Ergodic Theory

functions of entropy. This was generalized by Theorem 29 (Gutman-Hochman) Suppose J()

Gutman and Hochman (2008); some of the nota- is a finitely observable invariant on one of the
tion below is from their paper. following classes:
Here is the setting. Consider an ergodic pro-
cess, on a nonatomic space, taking on only finitely (i) The Kronecker systems; the class of systems
many values in ℕ; let C be some family of such with pure point spectrum.
processes. An observation scheme is a metric (ii) The zero-entropy mild mixing processes.
space (O, d) and a sequence of functions S ¼ (iii) The zero-entropy strong mixing processes.
ð Sn Þ 1
1 , where Sn maps ℕ . n. . ℕ into O. On a
!
point x  ℕ , the scheme converges if
1
Then J() is constant.

n 7! Sn ðx1 , x2 , . . . xn Þ ð35Þ
Recent Progress
converges in O. And on a particular process X, say
!
that S converges, if S converges on a.e. x in X .
In addition to a survey of older results in measure-
A function J : C ! O is isomorphism invari-
theoretic entropy and in topological entropy, let us
ant if, whenever the underlying transformations of
end this entry with a brief discussion of a couple
two processes X , X 0  C are isomorphic, then
of recent results, chosen from many.
J ðX Þ ¼ J ðX 0 Þ . Lastly, say that S “converges to
J,” if for each X  C , scheme S converges to the
value J ðX Þ. Entropy of Actions of Nonamenable Groups
The work of David Bailey (1976) produced an Consider ðG, G Þ, a topological group and its Borel
observation scheme for entropy. The Lempel-Ziv field (s-algebra). Let ð G X Þ be the field on
algorithm (Ziv and Lempel 1977) was another G X generated by the two coordinate-subfields.
entropy observer, with practical application A map
(Shields 1996)
Ornstein and Weiss provided entropy schemes c:G X!X ð36Þ
in (Ornstein and Weiss 1993; Ornstein et al.
1990). Their recent paper “Entropy is the only is measurable if
finitely-observable invariant” (Ornstein and
Weiss 2007) gives a converse, a uniqueness result. c1 ðX Þ  G X:

Use cg(x) for c(g, x).

Theorem 28 (Ornstein-Weiss) Suppose J is a
This map in (36) is a (measure-preserving)
finitely observable function, defined on all ergodic group action if 8g, h  G : cg ∘ ch ¼ cgh, and
finite-valued processes. If J is an isomorphism each cg : X ! X is measure-preserving.
invariant, then J is a continuous function of the
This encyclopedia entry has mainly discussed
entropy. entropy for ℤ-actions, that is, when G ¼ ℤ. The
ergodic theorem, our definition of entropy, and
Finitely Observable Extension large parts of ergodic theory involve taking aver-
Extending the Ornstein-Weiss result, Yonatan ages (of some quantity of interest) over larger and
Gutman and Michael Hochman, in Gutman and larger “pieces of time.” In ℤ or ℝ, we typically use
Hochman (2008) proved that it holds even when the intervals Ik ≔ [0. . .k)  ℤ or Ik ≔ [0. . .k)  ℝ.
the isomorphism invariant, J, is well defined only When G is ℤd, we average over squares
on certain subclasses of the set of all ergodic Rk ¼ I k I k    I k .
processes. In particular they obtain the following dtimes
result on three classes of zero-entropy The amenable groups are those which possess,
transformations. in a certain sense, larger and larger averaging sets
Entropy in Ergodic Theory 197

called Følner sequence. Parts of ergodic theory 1

(i) lim n!1
have been carried over to actions of amenable mn
groups, for example, Ornstein and Weiss (1983) #fk  f0, 1, . . . mn  1g : sn ðghÞk ¼ sn ðgÞsn ðhÞkg
and Ward and Zhang (1992). Indeed, much of the ¼ 1 for all g, h  G:
entropy theory was extended to certain amenable
groups by Ornstein and Weiss (1987). It is remark-
able that many notions established for ℤ-actions 1
lim n!1 #fk  f0, 1, . . . mn  1g : sn ðgÞk 6¼ kg
played an essential role in developing the proper- (ii) mn
ties of amenable group actions, especially the ¼ 1for all g  G∖f1g:
notion of orbit equivalence (Connes et al. 1981;
Danilenko and Park 2002; Rudolph and Weiss We call  ¼ {sn} a sofic approximation.
2000). Properties (i) and (ii) say that the sequence is
The typical example of a nonamenable group asymptotically homomorphism and asymp-
is a free group (on more than one generator). totically free4.
Pursuit to generalize the entropy theory to non-
amenable groups had been halted by the simple, Theorem 31 Let G be a sofic group and let
ðX, X , m, sÞ be a dynamical system of G-action.
but unexpected example by Ornstein and Weiss:
Two Bernoulli actions are isomorphic IFF their
Let 2 be a free group of two generators. There
distribution entropies of independent partitions
exists a factor map from the Bernoulli action of
are the same.
distribution 12 , 12 to the Bernoulli action of
distribution 14 , 14 , 14 , 14 . That is, a factor has a
He extended the above theorem to countable
bigger entropy in the case of 2 -action, unlike groups.
amenable group actions. Bowen has opened the
door to the world of entropy theory of non- Theorem 32 Let G be a countable nonamenable
amenable group actions. Bowen (2010) group. If two partitions have the same distribution
succeeded in extending the definition of entropy entropy with at least 3-states, then two Bernoulli
to actions of finite-rank free groups. actions of G are isomorphic.

Theorem 30 (L. Bowen) Let G be a finite-rank B. Seward (2018) was able to eliminate “at
free group. Then two Bernoulli G-actions are least 3-states” condition. If G is amenable, the
isomorphic IFF they have the same entropy. sofic entropy agrees with classical entropy. The
sofic, measure and topological, entropies are
known to be dependent on the sofic approxima-
The paper introduces an isomorphism invari- tions, unlike amenable groups (Bowen 2017,
ant, the “f invariant,” and shows that, for Bernoulli Bowen). However, for Bernoulli shifts of G the
actions, the f invariant agrees with entropy, that is, sofic entropy is independent of the sofic
with the distropy of the independent generating approximation.
partition. D. Kerr and H. Li (2011) introduced topolog-
And it was not too long to introduce a new ical sofic entropy on a compact metric space
invariant, called sofic entropy to extend the result meaning the exponential growth rate of the num-
to a much larger class of groups. The sofic entropy ber of “approximate periodic points (pseudo-
is defined via a sofic approximation of a group periodic points).” This leads them to prove the
(Bowen 2012, 2017). following variational principle.
Let G be a countable discrete group. We say
G is sofic if there exists sn : G ! Symm 4
The notion of sofic group is due to Mikheal Gromov
({0, 1, . . ., mn  1}) into permutation groups (1999). There is yet no known example of a countable
satisfying group which is not sofic.
198 Entropy in Ergodic Theory

Theorem 33 (Kerr-Li) Let G be a countable center as a major tool in Ergodic Theory and Topolog-
nonamenable group action on (X, s). For a sofic ical Dynamics. Simply mentioning all the substantial
approximate  of G, resultsin entropytheorywoulddwarfthelengthofthis
encyclopediaentry manytimesover.Andastherecent
h∑,top ðsÞ ¼ sup h∑ ðs, mÞ results mentioned above (cherry-picked out of many)
m
show, Entropy shows no sign of fading away. . .
where m is an invariant measure on (X, s).

Bibliography
Weak Pinsker Conjecture
Pinsker conjectured that every measure-preserving Adler RL, Marcus B (1979) Topological entropy and
transformation is a product of K-automorphism and equivalence of dynamical systems. Memoirs of the
0-entropy transformation (Pinsker 1960). Pesin American Mathematical Society, vol 219. American
(1977) showed that some flows on surfaces have Mathematical Society, Providence
Adler RL, Weiss B (1967) Entropy, a complete metric
the property; Bernoulli 0-entropy (in fact, a invariant for automorphisms of the torus. Proc Natl
rotation). However, by Ornstein’s counterexample Acad Sci USA 57(6):1573
(Ornstein 1973) to the conjecture, it is shown to be Adler RL, Konheim AG, McAndrew MH (1965) Topolog-
false in general. Thouvenot (1977) proposed a ical entropy. Trans Am Math Soc 114(2):309–319
Ambrose W (1941) Representation of ergodic flows. Ann
weaker version of the conjecture, called weak Math 42:723–739
Pinsker conjecture that every measure-preserving Austin T (2014) Scenery entropy as an invariant of RWRS
transformation is relatively Bernoulli over a trans- processes. arXiv preprint arXiv:1405.1468
formation of arbitrarily small entropy. That is, Austin T (2018) Measure concentration and the weak
pinsker property. Publ Math IHÉS 128(1):1–119
given any  > 0, ðX, X , m, T Þ is isomorphic to a Bailey DH (1976) Sequential schemes for classifying and
product Bernoulli (transformation of entropy predicting ergodic processes. PhD thesis, Stanford
<  ). Austin (2018) proves the conjecture is the University
affirmative. In fact, he proves a little more. Berg KR (1975) Independence and additive entropy. Proc
Am Math Soc 51(2):366–370
Bowen R (1971) Entropy for group endomorphisms and
Theorem 34 (Austin) If ðX, X , m, T Þ has a factor homogeneous spaces. Trans Am Math Soc
ðY, Y , n, SÞ and h(T) > h(S) then for a given 153:401–414
 > 0, there exists an extension Y, Y , n, S of Bowen R (1973) Topological entropy for noncompact sets.
Trans Am Math Soc 184:125–136
ðY, Y , n, SÞ such that the following diagram holds: Bowen L (2010) A new measure conjugacy invariant for
actions of free groups. Ann Math 171(2):1387–1400
Bowen L (2012) Every countably infinite group is almost
(X; X ;µ;T ) Ornstein. In: Dynamical systems and group actions,
vol 567. American Mathematical Society, Providence,
pp 67–78
(Ỹ ; Ỹ; º̃; S̃) Bowen L (2017) Examples in the entropy theory of count-
able group actions. arXiv:1704.06349 [math.DS],
pp 1–131
(Y; Y;º;S) Brin M, Stuck G (2002) Introduction to dynamical sys-
tems. Cambridge University Press, Cambridge
Buzzi J, Ruette S (2006) Large entropy implies existence of
where a maximal entropy measure for interval maps. Discrete
(i) h S < hðSÞ þ  : Contin Dynam Syst A 14(4):673–688
Clausius R (1864) Abhandlungen über die mechanische
(ii) ðX, X , m, T Þ is isomorphic to Wärmetheorie, vol 1. F. Vieweg, Braunschweig
Bernoulli Y, Y , n, S . Clausius R (1867) Abhandlungen über die mechanische
Wärmetheorie, vol 2. F. Vieweg, Braunschweig
Exodos Connes A, Feldman J, Weiss B (1981) An amenable equiv-
alence relation is generated by a single transformation.
Ergodic Theory Dynam Syst 1(4):431–450
Ever since the pioneering work of Shannon, and of Cornfel’d I, Fomin S, Sinai YG (1982) Ergodic theory.
Kolmogorov and Sinai, entropy has been front and Grundlehren der mathematischen Wissenschaften: a
Entropy in Ergodic Theory 199

series of comprehensive studies in mathematics. Krieger W (1970) On entropy and generators of

Springer, New York measurepreserving transformations. Trans Am Math
Danilenko AI, Park KK (2002) Generators and bernoullian Soc 149(2):453–464
factors for amenable actions and cocycles on their Ledrappier F, Young L-S (1985) The metric entropy of
orbits. Ergodic Theory Dynam Syst 22(6):1715–1745 diffeomorphisms: Part I: characterization of measures
den Hollander F, Steif JE et al (2006) Random walk in satisfying Pesin’s entropy formula. Ann Math
random scenery: a survey of some recent results. In: 122:509–574
Dynamics & stochastics. Institute of mathematical Sta- Lind D, Marcus B (1996) An introduction to symbolic
tistics lecture Notes 48:53–65 dynamics and coding. Cambridge University Press,
Denker M (1976) Measures with maximal entropy. In: Cambridge
Théorie Ergodique. Springer, Berlin, pp 70–112 Mane R (1987) Ergodic theory and differentiable dynam-
Dinaburg EI (1970) A correlation between topological ics. Springer, Berlin/Heidelberg
entropy and metric entropy. Dokl Akad Nauk Meshalkin L (1959) A case of isomorphism of bernoulli
190:19–22. Russian Academy of Sciences schemes. Dokl Akad Nauk SSSR 128(1):41–44
Dou D, Huang W, Park KK (2019) Entropy dimension of Milnor J (1988) On the entropy geometry of cellular
measure preserving systems. Trans Am Math Soc automata. Complex Syst 2(3):357–385
371(10):7029–7065 Misiurewicz M (1973) Diffeomorphism without any mea-
Ferenczi S (1997) Systems of finite rank. Colloq Math sure with maximal entropy. Bull Acad Pol Sci Sér Sci
73:35–65 Math Astron Phys 21(10):903–910
Friedman N (1970) Introduction to ergodic theory. Van Ornstein DS (1970) Bernoulli shifts with the same entropy
Nostrand Reinhold mathematical studies. Van are isomorphic. Adv Math 4(3):337–352
Nostrand Reinhold, New York Ornstein DS (1973) A mixing transformation for which
Gromov M (1999) Endomorphisms of symbolic algebraic Pinsker’s conjecture fails. Adv Math 10(1):103–123
varieties. J Eur Math Soc 1(2):109–197 Ornstein DS (1974) Ergodic theory, randomness and
Gutman Y, Hochman M (2008) On processes which cannot dynamical systems, no 5. Yale University Press, New
be distinguished by finite observation. Isr J Math Haven
164(1):265–284 Ornstein DS (2013) Newton’s laws and coin tossing. Not
Halmos P (1956) Lectures on ergodic theory. Chelsea Am Math Soc 60(4):450
scientific books. Mathematical Society of Japan, Tokyo Ornstein DS, Shields PC (1973) An uncountable family of
Hasselblatt B, Nitecki Z, Propp J (2005) Topological k-automorphisms. Adv Math 10(1):63–88
entropy for non-uniformly continuous maps. arXiv pre- Ornstein DS, Weiss B (1974) Finitely determined implies
print math/0511495 very weak Bernoulli. Isr J Math 17(1):94–104
Jung U, Lee J, Park KK (2017) Topological entropy dimen- Ornstein DS, Weiss B (1983) The Shannon-McMillan-
sion and directional entropy dimension for ℤ2-sub- Breiman theorem for a class of amenable groups. Isr
shifts. Entropy 19(2):46 J Math 44(1):53–60
Kalikow SA (1982) T, t-1 transformation is not loosely Ornstein DS, Weiss B (1987) Entropy and isomorphism
Bernoulli. Ann Math 115:393–409 theorems for actions of amenable groups. J Anal Math
Kalikow S, King JL (1994) A countably-valued sleeping 48(1):1–141
stockbroker process. J Theor Probab 7(4):703–708 Ornstein DS, Weiss B (1993) Entropy and data compres-
Katok A (1980) Smooth non-Bernoulli K-automorphisms. sion schemes. IEEE Trans Inf Theory 39(1):78–83
Invent Math 61(3):291–299 Ornstein DS, Weiss B (2007) Entropy is the only finitely-
Katok A, Hasselblatt B (1995) Introduction to the modern observable invariant. J Mod Dyn 1(1):93–105
theory of dynamical systems. In: Encyclopedia of Ornstein DS, Weiss B et al (1990) How sampling reveals a
mathematics and its applications. Cambridge Univer- process. Ann Probab 18(3):905–930
sity Press, Cambridge Park KK (1999) On directional entropy functions. Isr
Katok A, Thouvenot J-P (1997) Slow entropy type invari- J Math 113(1):243–267
ants and smooth realization of commuting measure- Pesin YB (1977) Characteristic Lyapunov exponents and
preserving transformations. Ann Inst Henri Poincare smooth ergodic theory. Usp Matematicheskikh Nauk
Probab Stat 33:323–338. Elsevier 32(4):55–112
Kerr D, Li H (2011) Entropy and the variational principle Petersen K (1983) Ergodic theory. Cambridge studies in
for actions of sofic groups. Invent Math advanced mathematics. Cambridge University Press,
186(3):501–558 Cambridge
King JL (1988) Joining-rank and the structure of finite rank Pinsker MS (1960) Dynamical systems with completely
mixing transformations. J Anal Math 51(1):182–227 positive or zero entropy. Dokl Akad Nauk
King JL (1992) Dilemma of the sleeping stockbroker. Am 133:1025–1026. Russian Academy of Sciences
Math Mon 99(4):335–338 Rudin W (1973) Functional analysis. McGraw-Hill series
Kolmogorov A (1958) A new metric invariant of transitive in higher mathematics. McGraw-Hill, New York
dynamical systems and lebesgue space automorphisms. Rudolph D (1976) A two-valued step coding for ergodic
Dokl Acad Sci USSR 119:861–864 flows. Math Z 150(3):201–220
200 Entropy in Ergodic Theory

Rudolph D (1990) Fundamentals of measurable dynamics: Young L-S (1982) Dimension, entropy and lyapunov expo-
ergodic theory on Lebesgue spaces. Oxford science nents. Ergodic Theory Dynam Syst 2(1):109–124
publications. Clarendon Press, Oxford Ziv J, Lempel A (1977) A universal algorithm for sequen-
Rudolph DJ, Weiss B (2000) Entropy and mixing for tial data compression. IEEE Trans Inf Theory
amenable group actions. Ann Math 151(3):1119–1150 23(3):337–343
Seward B (2018) Bernoulli shifts with bases of equal
entropy are isomorphic. arXiv preprint
arXiv:1805.08279
Articles, Papers, and Books of Interest
Bowen L. https://warwick.ac.uk/fac/sci/maths/research/
Shannon CE (1948) A mathematical theory of communi-
events/2018-19/metc/abstracts
cation. Bell Syst Tech J 27(3):379–423
Boyle M, Downarowicz T (2004) The entropy theory of
Shields PC (1973) The theory of Bernoulli shifts. Univer-
symbolic extensions. Invent Math 156(1):119–161
sity of Chicago Press, Chicago
Downarowicz T (2011) Entropy in dynamical systems.
Shields PC (1996) The ergodic theory of discrete sample
Cambridge University Press, Cambridge
paths, vol 13. American Mathematical Society,
Downarowicz T, Serafin J (2003) Possible entropy func-
Providence
tions. Isr J Math 135(1):221–250
Sinai YG (1959) On the concept of entropy for a dynamic
Greven A, Keller G, Warnecke G (2003) Entropy.
system. Dokl Akad Nauk SSSR 124(4):768–771
Princeton series in applied mathematics. Princeton Uni-
Sinai YG (1962) A weak isomorphism of transformations
versity Press, Princeton
with invariant measure. Dokl Akad Nauk
Hassner M (1980) A non-probabilistic source and channel
147(4):177–207
coding theory. PhD thesis, UCLA
Sinai YG (2017) Topics in ergodic theory, vol 44.
Katok A (2007) Fifty years of entropy in dynamics:
Princeton University Press, Princeton
1958–2007. J Modern Dyn 1(4):545–596
Thouvenot J-P (1975) Quelques propriétés des systemes
Katok A, Sinai YG, Stepin AM (1977) Theory of dynam-
dynamiques qui se décomposent en un produit de deux
ical systems and general transformation groups with
systemes dont l’un est un schéma de Bernoulli. Isr
invariant measure. J Sov Math 7(6):974–1065
J Math 21(2–3):177–207
McMillan B (1953) The basic theorems of information
Thouvenot J-P (1977) On the stability of the weak pinsker
theory. Ann Math Statist 24(2):196–219
property. Isr J Math 27(2):150–162
Newhouse SE (1989) Continuity properties of entropy.
Tribus M, McIrvine EC (1971) Energy and information.
Ann Math 129(1):215–235
Sci Am 225(3):179–188
Parry W (1969) Entropy and generators in ergodic theory.
Walters P (1981) An introduction to ergodic theory. Grad-
Mathematics lecture note series. W. A. Benjamin,
uate texts in mathematics. Springer, New York
New York
Ward T, Zhang Q (1992) The Abramov-Rokhlin entropy
Wikipedia. http://en.wikipedia.org/wiki/Spectral_radius,
addition formula for amenable group actions. Monatsh
http://en.wikipedia.org/wiki/Information_entropy
Math 114(3):317–329
 Coupling of two measure spaces A coupling of
Isomorphism Theory in Ergodic two measure spaces (X, m, ℬ) and ðY, n, C Þ is a
Theory measure γ on X  Y such that γ(B  Y ) ¼ m(B)
for all B  ℬ and γ(X  C) ¼ n(C) for all
Christopher Hoffman C C.
Department of Mathematics, University of Ergodic measure preserving trans-
Washington, Seattle, WA, USA formation A measure preserving transforma-
tion is ergodic if the only invariant sets
(m(A △ T1(A)) ¼ 0) have measure 0 or 1.
Article Outline Ergodic theorem The pointwise ergodic theo-
rem says that for any measure preserving trans-
Glossary formation (X, ℬ, m) and T and any L1 function
Definition of the Subject n
1
f the time average lim f T i ðxÞ converges
Introduction n!1 n
i¼1
Basic Transformations a.e. If the transformation is ergodic then the
Basic Isomorphism Invariants limit is the space average, f dm a.e.
Basic Tools Geodesic A geodesic on a Riemannian manifold
Isomorphism of Bernoulli Shifts is a distance minimizing path between points.
Transformations Isomorphic to Bernoulli Shifts Horocycle A horocycle is a circle in the hyper-
Transformations Not Isomorphic to Bernoulli bolic disk which intersects the boundary of the
Shifts disk in exactly one point.
Classifying the Invariant Measures of Algebraic Invariant measure Likewise a measure m is said
Actions to be invariant with respect to (X, T ) provided
Finitary Isomorphisms that m(T1(A)) ¼ m(A) for all measurable
Flows A  ℬ.
Other Equivalence Relations Joining of two measure preserving trans-
Non-invertible Transformations formations A joining of two measure preserv-
Factors of a Transformation ing transformations (X, T) and (Y, S) is a
Actions of Amenable Groups coupling of X and Y which is invariant under
Future Directions T  S.
Bibliography Markov shift A Markov shift is a stochastic
process such that the conditional distribution
Glossary of the future outputs ({xn}n>0) of the process
conditioned on the last output (x0) is the same
Almost everywhere A property is said to hold as the distribution conditioned on all of the past
almost everywhere (a.e.) if the set on which outputs of the process ({xn}n0).
the property does not hold has measure 0. Measure preserving transformation A mea-
Bernoulli shift A Bernoulli shift is a stochastic sure preserving transformation consists of a
process such that all outputs of the process are probability space (X, T) and a measurable
independent. function T : X ! X such that m(T1(A)) ¼
Conditional measure For any measure space m(A) for all A  ℬ.
(X, ℬ, m) and s-algebra C
ℬ the conditional Measure theoretic entropy A numerical invari-
measure is a C-measurable function g such that ant of measure preserving transformations that
m(C) ¼ Cg dm for all C  C . measures the growth in complexity of

© Springer-Verlag 2009 201

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_297
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_297
202 Isomorphism Theory in Ergodic Theory

measurable partitions refined under the itera- Introduction

tion of the transformation.
Probability space A probability space X ¼ In this article we look at the problem of determin-
(X, m, ℬ) is a measure space such that ing of which measure preserving transformations
m(B) ¼ 1. are isomorphic. We look at a number of isomor-
Rational function, rational map A rational phism invariants, the most important of which is
function f(z) ¼ g(z)/h(z) is the quotient of two the Kolmogorov–Sinai entropy. The central theo-
polynomials. The degree of f(z) is the maxi- rem in this field is Ornstein’s proof that any two
mum of the degrees of g(z) and h(z). The Bernoulli shifts of the same entropy are
corresponding rational maps Tf : z ! f(z) on isomorphic. We also discuss some of the conse-
the Riemann sphere ℂ are a main object of quences of this theorem, which transformations
study in complex dynamics. are isomorphic to Bernoulli shifts as well as gen-
Stochastic process A stochastic process is a eralizations of Ornstein’s theory.
sequence of measurable functions {xn}n  ℤ
(or outputs) defined on the same measure
space, X. We refer to the value of the functions Basic Transformations
as outputs.
In this section we list some of the basic classes of
measure preserving transformations that we study
Definition of the Subject in this article.

Our main goal in this article is to consider when Bernoulli shifts Some of the most fundamental
two measure preserving transformations are in transformations are the Bernoulli shifts.
some sense different presentations of the same A probability vector is a vector fpi gni¼1
underlying object. To make this precise we say such that 2i¼1 pi ¼ 1 and pi  0 for all i. Let
two measure preserving maps (X, T ) and (Y, S) p ¼ fpi gni¼1 be a probability vector. The
are isomorphic if there exists a measurable map Bernoulli shift corresponding to p has state
f : X ! Y such that space {1, 2, . . ., n}ℤ, the shift operator
T(x)i ¼ xiþ1. To specify the measure we only
1. f is measure preserving, need to specify it on cylinder sets
2. f is invertible almost everywhere and
3. f(T(x)) ¼ S(f(x)) for almost every x. A ¼ fx  X: xi ¼ ai 8 i  fm, . . . , kgg

The main goal of the subject is to construct a for some m  k  ℤ and a sequence am, . . .,
collection of invariants of a transformation such ak  {1, . . ., n}. The measure on cylinder sets
that a necessary condition for two transforma- is defined by
mfx  X : xi ¼ ai ½2ex for all i such that m  i  kg
tions to be isomorphic is that the invariant be
k
the same for both transformations. Another goal ¼ pai :
of the subject is to solve the much more difficult i¼m

problem of constructing invariants such that the

invariants being the same for two transforma- For any d  ℕ if p = (1/d, . . ., 1/d) we refer to
tions is a sufficient condition for the transforma- Bernoullip as the Bernoulli d shift.
tion to be isomorphic. Finally we apply these Markov shifts A Markov shift on state n sym-
invariants to many natural classes of transforma- bols is defined by an n  n matrix, M, such that
tion to see which of them are (or are not) fMði, jÞgnj¼1 is a probability vector for each i.
isomorphic. The Markov shift is a measure preserving
Isomorphism Theory in Ergodic Theory 203

transformation with state space 1 0

{1, 2, . . ., n}ℤ, transformation T(x)i = xi+1 ht ðGg Þ ¼ Gg :
t 1

mfx1 ¼ a1 j x0 ¼ a0 g
Matrix actions Another natural class of actions
¼ mfx1 ¼ a1 j x0 ¼ a0 , x1 ¼ a1 , x2 ¼ a2 , .. .g is given by the action of matrices on tori. Let M
be an invertible n  n integer valued matrix.
for all choices of ai, i  1. Let m ¼ fmðiÞgni¼1 We define TM : [0, 1)n ! [0, 1)n by TM(x)i = M
be a vector such that Mm = m. Then an invari- (x)i mod 1 for all i, 1  i  n. It is easy to check
ant measure is defined by setting the that if M is surjective then Lebesgue measure is
measure on cylinder sets to be A = {x : x0 = invariant under TM. TM is a j det (M)j to one
a0, x1 = a1, . . ., xn = an} is given by map. If n = 1 then M is an integer and we refer
mðAÞ ¼ mða0 Þ ni¼1 Mðai1 , ai Þ. to the map as times M.
Shift maps More generally the shift map s is The [T, T1] transformations Let (X, T) be any
the map s : ℕℤ ! ℕℤ where s(x)i = xi+1 for all invertible measure preserving transformation.
x  ℕℤ and i  ℤ. We also let s designate the Let s be the shift operator on Y = {1, 1}ℤ.
shift map on ℕℕ. For each measure that is The space Y comes equipped with the
invariant under the shift map there is a Bernoulli (1/2, 1/2) product measure n.
corresponding measure defined on ℕℕ that is The [T, T1] transformation is a map on Y  X
invariant under the shift map. Let m be an which preserves n  m. It is defined by
invariant measure under the shift map. For
any measurable set of A
ℕℕ we define A on T, T 1 ðy, xÞ ¼ ðSðyÞ, T y0 ðxÞÞ:
ℕℤ by
Induced transformations Let (X, T) be a mea-
A ¼ f. . . , x1 , x0 , x1 , . . . : x0 , x1 , . . .  Ag:
sure preserving transformation and let A
X
with 0 < m(A) < 1. The transformation
Then it is easy to check that m defined by
induced by A, (A, TA, mA), is defined as fol-
m A ¼ mðAÞ is invariant. If the original trans- lows. For any x  A
formation was a Markov or Bernoulli shift then
refer to the resulting transformations as a one T A ðxÞ ¼ T nðxÞ ðxÞ
sided Markov shifts or one sided Bernoulli
shift respectively. where n(x) = inf {m > 0 : T m(x)  A}. For any
Rational maps of the Riemann sphere We say B
A we have that mA(B) = m(B)/m(A).
that f(z) = g(z)/h(z) is a rational function of
degree d  2 if both g(z) and h(z) are poly-
nomials with max(deg(g(z)), deg(h(z))) = d. Basic Isomorphism Invariants
Then f induces a natural action on the Riemann
sphere Tf : z ! f(z) which is a d to one map The main purpose of isomorphism theory is to
(counting with multiplicity). In section “Ratio- classify which pairs of measure preserving trans-
nal Maps” we shall see that for every rational formation are isomorphic and which are not
function f there is a canonical measure mf such isomorphic. One of the main ways that we can
that Tf is a measure preserving transformation. show that two measure preserving transformation
Horocycle flows The horocycle flow acts on SL are not isomorphic is using isomorphism invari-
(2, ℝ)/Γ where Γ is a discrete subgroup of SL ants. An isomorphism invariant is a function
(2, ℝ) such that SL(2, ℝ)/Γ has finite Haar f defined on measure preserving transformations
measure. For any g  SL(2, ℝ) and t  ℝ such that if (X, T ) is isomorphic to (Y, S) then
we define the horocycle flow by f((X, T)) ¼ f((Y, S)).
204 Isomorphism Theory in Ergodic Theory

A measure preserving action is said to be ergo- 1

n1
dic if m(A) ¼ 0 or 1 for every A with f ðxÞ ¼ lim f Tk x
n!1 n k¼0
1
m A D T ðAÞ ¼ 0:
exists for almost every x and if (X, T ) is ergodic
then the limit is f ¼ f dm (Birkhoff 1931).
A measure preserving action is said to be weak
A partition P of X is a measurable function
mixing if for every measurable A and B
defined on X. (For simplicity we often assume that
1 a partition is a function to ℤ or some subset of ℤ.)
1 We write Pi for P1(i). For any partition P of
jmðA \ T n ðBÞÞ  mðAÞmðBÞj ¼ 0:
n n¼1 (X, T ) define the partition TiP by P  Ti. Thus
for invertible T this is given by TiP(x) ¼ P(T i(x)).
A measure preserving action is said to be Then define ðPÞT ¼ V i  Z T i P. Thus (P)T is the
mixing if for every measurable A and B smallest s-algebra which contains Ti(Pj) for all
i  ℤ and j  ℕ. We say that (P)T is the
lim mðA \ T n ðBÞÞ ¼ mðAÞmðBÞ: s-algebra generated by P. A partition P is a
n!1
generator of (X, T ) if (P)T ¼ ℬ. Many measure
It is easy to show that all three of these prop- preserving transformations come equipped with a
erties are isomorphism invariants and that natural partition.
Rokhlin’s theorem For any measure preserving
ðX, T Þ is mixing ) ðX, T Þ is weak mixing
transformation (X, T ), any ϵ > 0 and any
) ðX, T Þ is ergodic: n  ℕ there exists A
X such that
Ti(A) \ Tj(A) for all 0  i < j  n and
We include one more definition before intro- m [ni¼0 T i ðAÞ > 1  ϵ. Moreover for any finite
ducing an even stronger isomorphism invariant. partition P of X we can choose A such that
The action of a group (or a semigroup) G on a m(Pi \ A) ¼ m(A)m(Pi) for all i  ℕ (Halmos
probability space (X, m, ℬ) is a family of measure 1960).
preserving transformations {fg}g  G such that for Shannon–McMillan–Breiman theorem
any g, h  G we have that fg( fh(x)) ¼ fgþh(x) for (Breiman 1957; Petersen 1989) For any mea-
almost every x  X. Thus any invertible measure sure preserving system (X, T) and any ϵ > 0
preserving transformation (X, T) induces a ℤ there exists n  ℕ and a set G with m(G) > 1  ϵ
action by fn(x) ¼ T n(x). with the following property. For any sequence
We say that a group action (X, Tg) is mixing of
g1, . . . gn  ℕ let g ¼ \ni¼1 P T i ðXÞ ¼ gi .
all orders if for every n  ℕ and every collection
Then if m(g \ G) > 313E0 then
of set A1, . . ., An  ℬ then

mðA1 \ T g2 ðA2 Þ \ T g2 þg3 ðA3 Þ . . . mðgÞ  2hððX, TÞÞϵ , 2hððX, TÞÞþϵ :

n
\T g2 þg3 þ...gn ðAn Þ ! mðAi Þ
i¼1 Krieger generator theorem If H((X, T )) < 1
then there exists a finite partition P such that
as the gi go to infinity. (P)T ¼ ℬ. Thus every measure preserving
transformation with finite entropy is isomor-
phic to a shift map on finitely many symbols
Basic Tools (Krieger 1970).
Measure-theoretic entropy Entropy was intro-
Birkhoff’s ergodic theorem states that for every duced in physics by Rudolph Clausius in 1854.
measure preserving action the limits In 1948 Claude Shannon introduced the
Isomorphism Theory in Ergodic Theory 205

concept to information theory. Consider a pro- of ℬ. We say that (Y, S) is trivial if Y consists
cess that generates a string of data of length n. of only one point. We say that a transformation
The entropy of the process is the smallest num- (X, T) has completely positive entropy if
ber h such that you can condense the data to a every non-trivial factor of (X, T ) has positive
string of zeroes and one of length hn and with entropy.
high probability you can reconstruct the origi-
nal data from the string of zeroes and ones. Isomorphism of Bernoulli Shifts
Thus the entropy of a process is the average
amount of information transmitted per symbol Kolmogorov–Sinai
of the process. A long standing open question was for which
Kolmogorov and Sinai introduced the concept p and q are Bernoullip and Bernoulliq
of entropy to ergodic theory in the following isomorphic. In particular are the Bernoulli
way (Kolmogorov 1958, 1959). They defined 2 shift and the Bernoulli 3 shift isomorphic. Both
the entropy of a partition Q is defined to be of these transformations have completely positive
entropy and all other isomorphism invariants
k which were known at the time are the same for
H ðQÞ ¼  mðQi Þ log mðQi Þ: the two transformations. The first application of
i¼1
the Kolmogorov–Sinai entropy was to show that
the answer to this question is no.
The measure-theoretic entropy of a dynamical
Fix a probability vector p. The transformation
system (X, T ) with respect to a partition
Bernoullip has Qp : x ! x0 as a generating parti-
Q : X ! {1, . . ., k} is then defined as
tion. By Sinai’s theorem
n
.
H Bernoullip ¼ H Qp ¼ pi log 2 ðpi Þ:
i¼1

Finally, the measure-theoretic entropy of a Thus the Bernoulli 2 shift (with entropy 1) is
dynamical system (X, T ) is defined as not isomorphic to the Bernoulli 3 shift (with
entropy log2(3)).
hððX, T ÞÞ ¼ sup hðX, T, QÞ Sinai also made significant progress toward
Q
showing that Bernoulli shifts with the same
where the supremum is taken over all finite entropy are isomorphic by proving the following
measurable partitions. A theorem of Sinai theorem.
showed that if Q is a generator of (X, T) then
h(T ) ¼ h(T, Q) (Sinaĭ 1959). Theorem 1 (Sinaĭ 1962) If (X, T) is a measure
This shows that for every measure preserving preserving system of entropy h and (Y, S) is a
0
function (X, T ) there is an associated entropy Bernoulli shift of entropy h  h then (Y, S) is a
h(T )  [0, 1]. It is easy to show from the factor of (X, T ).
definition that entropy is an isomorphism
invariant. This theorem implies that if p and q are prob-
We say that (Y, S) is a factor of (X, T ) if there ability vectors and H(p) ¼ H(q) then Bernoullip is
exists a map f : X ! Y such that a factor in Bernoulliq and Bernoulliq is a factor in
Bernoullip. Thus we say that Bernoullip and
1. f is measure preserving and Bernoulliq are weakly isomorphic.
2. f(T(x)) ¼ S(f(x)) for almost every x.
Explicit Isomorphisms
Each factor (Y, S) can be associated with The other early progress on proving that Bernoulli
f1 ðC Þ, which is an invariant sub s-algebra shifts with the same entropy are isomorphic came
206 Isomorphism Theory in Ergodic Theory

from Meshalkin. He considered pairs of probabil- jm x : P T i ðxÞ ¼ ai 8i ¼ 1,

ity vectors p and q with H(p) ¼ H(q) and all of the ða1 , ..., an Þ  ℤn
pi and qi are related by some algebraic relations.
...,n n y : Q T i ðyÞ ¼ ai 8i ¼ 1, . . .n j
For many such pairs he was able to prove that the
two Bernoulli shifts are isomorphic. In particular < d:
he proved the following theorem.
A process (X, T) and P are finitely deter-
mined if for every ε there exist n and δ such that
Theorem 2 (Meshalkin 1959) Let p ¼ (1/4,
if (Y, S) and Q are such that
1/4, 1/4, 1/4) and q ¼ (1/2, 1/8, 1/8, 1/8, 1/8).
Then Bernoullip and Bernoulliq are isomorphic.
1. (X, T ) and P and Yand Q are within δ in
n distributions and
Ornstein 2. jH(X, P)  H(Y, Q) j < δ
The central theorem in the study of isomorphisms then there exists a joining γ of X and Y such
of measure preserving transformations is that for all m
Ornstein’s isomorphismtheorem.
dm ðx, yÞ dgðx, yÞ < ϵ:
Theorem 3 (Ornstein 1970c) If p and q are x,y

probability vectors and H(p) ¼ H(q) then the

Bernoulli shifts Bernoullip and Bernoulliq are A transformation (X, T) is finitely determined
isomorphic. if it is finitely determined for every finite
partition P.
It is fairly straightforward to show that
To see how central this is to the field most of
Bernoulli shifts are finitely determined. Ornstein
the rest of this article is a summary of:
used this fact along with the Rokhlin lemma and
the Shannon–McMillan–Breiman theorem to
1. The proof of Ornstein’s theorem,
prove a more robust version of Theorem 1.
2. The consequences of Ornstein’s theorem,
To describe Ornstein’s proof we use the a
3. The generalizations of Ornstein’s theorem, and
description due to Rothstein. We say that for a
4. How the properties that Ornstein’s theorem
joining γ of (X, m) and (Y, n) that P
ϵ,g C if there
implies that Bernoulli shifts must have differ
exists a partition P of C such that
from the properties of every other class of
transformations.
g Pi D Pi < ϵ:
i
The key to Ornstein’s proof was the introduc-
tion of the finitely determined property. To explain If P
ϵ,g C for all ϵ < 0 then it is possible to
the finitely determined property we first define the show that there exists a partition P of C such that
Hamming distance of length n between sequences
x, y  ℕℤ by
g Pi D Pi ¼ 0
i
j fk  f1, . . . , ng : xk ¼ yk g j
dn ðx, yÞ ¼ 1  :
n
and we write P
g C . If P
g C then (X, T ) is a
Let (X, T ) and (Y, S) be a measure preserving factor of (Y, S) by the map f that sends y ! x
transformation and let P and Q be finite partitions where P T i x ¼ P Si y for all i.
of X and Y respectively. In this language Ornstein proved that if (X, T )
We say that (X, T ) and P and (Y, S) and Q are is finitely determined, P is a generating partition
within δ in n distributions if of X and h((Y, S))  h((X, T )) then for every
Isomorphism Theory in Ergodic Theory 207

ϵ > 0 the set of joinings γ such that P
g,ϵ C is an 1. (X, T ) is finitely determined,
open and dense set. Thus by the Baire category 2. (X, T ) is very weak Bernoulli and
theorem there exists γ such that P
g C . This 3. (X, T ) is isomorphic to a Bernoulli shift.
reproves Theorem 1.
Moreover if (X, T ) and (Y, S) are finitely
determined, h((X, T )) ¼ h((Y, S)) and P and
Q are generating partition of X and Y then by the Using the fact that a transformation is finitely
Baire category theorem there exists γ such that determined or very weak Bernoulli is equivalent
P
g C and Q
γ ℬ. Then the map f that sends to it being isomorphic to a Bernoulli shift we can
y ! x where P T i x ¼ P Si y for all i is an prove the following theorem.
isomorphism.
Theorem 5 (Ornstein 1970a)
Properties of Bernoullis 1. If (X, T n) is isomorphic to a Bernoulli shift
Now we define the very weak Bernoulli property then (X, T) is isomorphic to a Bernoulli shift.
which is the most effective property for showing 2. If (X, T ) is a factor of a Bernoulli shift then
that a measure preserving transformation is iso- (X, T ) is isomorphic to a Bernoulli shift.
morphic to a Bernoulli shift. 3. If (X, T ) is isomorphic to a Bernoulli shift then
Given X and a partition P define the past of x by there exists a measure preserving transforma-
tion (Y, S) such that (Y, S n) is isomorphic to
Ppast ðxÞ ¼ x0 : T i Pðx0 Þ ¼ T i PðxÞ 8i  ℕ (X, T ).

and denote the measure m conditioned on Ppast(x)

by mjPpast ðxÞ. Define Rudolph Structure Theorem
An important application of the very weak
Bernoulli condition is the following theorem of
dn,x,m ¼ inf dn ðx, x0 Þ dgðx, x0 Þ, Rudolph.
g

Theorem 6 (Rudolph 1983) Let (X, T) be iso-

where the inf is taken over all γ which are cou-
morphic to a Bernoulli shift,G be a compact Abe-
plings of mjPpast ðxÞ and m. Also define
lian group with Haar measure mG and s : X ! G
be a measurable map. Then let S : X  G ! X  G
d n,Ppast ,ðX,T Þ ¼ dn,x,m dm: be defined by

Sðx, gÞ ¼ ðT ðxÞ, g þ sðxÞÞ:

We say that (X, T ) and P are very weak
Bernoulli if for every ϵ > 0 there exists n such Then (X  G, S, m  mG) is isomorphic to a
that dn,Ppast ,ðX,T Þ < ϵ. We say that (X, T) is very Bernoulli shift.
weakly Bernoulli if there exists a generating par-
tition P such that (X, T) and P are very weak
Bernoulli. Transformations Isomorphic to Bernoulli
Ornstein and Weiss were able to show that the Shifts
very weak Bernoulli property is both necessary
and sufficient to be isomorphic to a Bernoulli One of the most important features of Ornstein’s
shift. isomorphism theory is that it can be used to check
whether specific transformations (or families of
Theorem 4 (Ornstein 1970b; Ornstein and transformations) are isomorphic to Bernoulli
Weiss 1974) For transformations (X, T ) the fol- shifts. The finitely determined property is the
lowing conditions are equivalent: key to the proof of Ornstein’s theorem and the
208 Isomorphism Theory in Ergodic Theory

proof of many of the consequences listed in sec- entropy it is natural to ask if (X, T ) is not isomor-
tion “Properties of Bernoullis”. However if one phic to a Bernoulli shift then is there any reason-
wants to show a particular transformation is iso- able condition we can put on (Y, S) that implies
morphic to a Bernoulli shift then the very weak the two transformations are isomorphic. For
Bernoulli property is more useful. There have example if (X, T2) and (Y, S2) are completely
been many classes of transformations that have positive entropy transformations which are iso-
been proven to be isomorphic to a Bernoulli shift. morphic does that necessarily imply that (X, T )
Here we mention two. and (Y, S) are isomorphic? The answer turns out
The first class are the Markov chains. Friedman to be no (Rudolph 1976). We could also ask if
and Ornstein proved that if a Markov chain is (X, T ) and (Y, S) are completely positive entropy
mixing then it is isomorphic to a Bernoulli shift transformations which are weakly isomorphic
(Friedman and Ornstein 1970). The second are does that imply that (X, T ) and (Y, S) are isomor-
automorphisms of [0, 1)n. Let M be any n  n phic? Again the answer is no (Hoffman 1999a).
matrix with integer coefficients and j det (M ) j ¼ The key insight to answering questions like this is
1. If none of the eigenvalues fli gni¼1 of M are roots due to Rudolph who showed that such questions
of unity then Katznelson proved that TM is iso- about the isomorphism of transformations can be
morphic to the Bernoulli shift with entropy reduced to questions about conjugacy of
n
i¼1 maxð0, logðli ÞÞ (Katznelson 1971). permutations.

Rudolph’s Counterexample Machine

Transformations Not Isomorphic to Given any transformation (X, T ) and any permu-
Bernoulli Shifts tation π in Sn (or Sℕ) we can define the transfor-
mation (Xn, Tπ, mn, ℬ) by
Recall that a measure preserving transformation
(X, T ) has completely positive entropy if for T p ðx1 , x2 , . . ., xn Þ ¼ T xpð1Þ , T xpð2Þ , . . . ,
every nontrivial (Y, S) which is a factor of
(X, T ) we have that H((Y, S)) > 0. It is easy to T xpðnÞ
check that Bernoulli shifts have completely posi-
tive entropy. It is natural to ask if the converse
where mn is the direct product of n copies of m.
true? We shall see that the answer is an emphatic
Rudolph introduced the concept of a transforma-
no. While the isomorphism class of Bernoulli
tion having minimal self joinings. If a transfor-
shifts is given by just one number, the situation
mation has minimal self joinings then for every π
for transformations of completely positive
it is possible to list all of the measures on Xn which
entropy is infinitely more complicated.
are invariant under Tπ.
Ornstein constructed the first example of a
If there exists an isomorphism f between T and
transformation with completely positive entropy
(Y, S) then there is a corresponding measure on
which is not isomorphic to a Bernoulli shift
X  Y which is supported on points of the form
(Ornstein 1973b). Ornstein and Shields built upon
(x, f(x)) and has marginals m and n. Thus if we
this construction to prove the following theorem.
know all of the measures on X  Y which are
invariant under T  S then we know all of the
Theorem 7 (Ornstein and Shields 1973) For
isomorphisms between (X, T ) and (Y, S). Using
every h > 0 there is an uncountable family of
this we get the following theorem.
completely positive entropy transformations
which all have entropy h but no two distinct mem-
Theorem 8 (Rudolph 1979) There exists a non-
bers of the family are isomorphic.
trivial transformation with minimal self joinings.
For any transformation (X, T ) with minimal self
Now that we see there are many isomorphic
joinings the corresponding transformation T p1 is
transformations that have completely positive
Isomorphism Theory in Ergodic Theory 209

isomorphic to T p2 if and only if the permutations transformations have little inherent interest out-
π1 and π2 are conjugate. side of their ergodic theory properties. This led
many people to search for a “natural” example of
There are two permutations on two elements, such a transformation. The most natural examples
the flip π1 ¼ (12) and the identity π2 ¼ (1)(2). For are the [T, T 1] transformation and many other
both permutations, the square of the permutation transformations derived from it. It is easy to show
is the identity. Thus there are two distinct permu- that the [T, T 1] transformation has completely
tations whose square is the same. Rudolph positive entropy (Meilijson 1974). Kalikow pro-
showed that this fact can be used to generate two ved that for many T the corresponding [T, T 1]
transformations which are mixing that are not transformation is not isomorphic to a Bernoulli
isomorphic but their squares are isomorphic. The shift.
following theorem gives more examples of the
power of this technique. Theorem 11 (Kalikow 1982) If h(T ) > 0 then
the [T, T1] transformation is not isomorphic to a
Theorem 9 (Rudolph 1979) Bernoulli shift.
1. There exists measure preserving transforma-
tions (X, T ) and (Y, S) which are weakly iso- The basic idea of Kalikow’s proof has been
morphic but not isomorphic. used by many others. Katok and Rudolph used
2. There exists measure preserving transforma- the proof to construct smooth measure preserving
tions (X, T ) and (Y, S) which are not isomor- transformations on infinite differentiable mani-
phic but (X, Tk, m) is isomorphic to (Y, Sk, n) folds which have completely positive entropy
for every k > 1, and but are not isomorphic to Bernoulli shifts (Katok
3. There exists a mixing transformation with no 1980; Rudolph 1988). Den Hollander and Steif
non trivial factors. did a thorough study of the ergodic theory prop-
erties of [T, T 1] transformations where T is sim-
ple random walk on a wide family of graphs (Den
If (X, T ) has minimal self joinings then it has Hollander and Steif 1997).
zero entropy. However Hoffman constructed a
transformation with completely positive entropy
that shares many of the properties of transforma- Classifying the Invariant Measures of
tions with minimal self joinings listed above. Algebraic Actions

Theorem 10 (Hoffman 1999a) This problem of classifying all of the invariant

1. There exist measure preserving transforma- measures like Rudolph did with his transforma-
tions (X, T ) and (Y, S) which both have tion with minimal self joinings comes up in a
completely positive entropy and are weakly number of other settings. Ratner characterized
isomorphic but not isomorphic. the invariant measures for the horocycle flow
2. There exist measure preserving transforma- and thus characterized the possible isomorphisms
tions (X, T ) and (Y, S) which both have between a large family of transformations gener-
completely positive entropy and are not iso- ated by the horocycle flow (Ratner 1978, 1982,
morphic but (X, Tk, m) is isomorphic to 1983). This work has powerful applications to
(Y, Sk, n) for every k > 1. number theory.
There has also been much interest in classify-
ing the measures on [0, 1) that are invariant under
T, T Inverse both times 2 and times 3. Furstenberg proved that
All of the transformations that have completely the only closed infinite set on the circle which is
positive entropy but are not isomorphic to a invariant under times 2 and times 3 is [0, 1) itself
Bernoulli shift described above are constructed and made the following measure theoretic
by a process called cutting and stacking. These conjecture.
210 Isomorphism Theory in Ergodic Theory

Conjecture 1 (Furstenberg 1967) The only non- Keane and Smorodinsky proved the following
atomic measure on [0, 1) which is invariant under strengthening of Ornstein’s isomorphism
times 2 and times 3 is Lebesgue measure. theorem.

Rudolph improved on the work of Lyons Theorem 13 (Keane and Smorodinsky 1979) If
(1988) to provide the following partial answer to p and q are probability vectors with H(p) ¼ H(q)
this conjecture. then the Bernoulli shifts Bernoullip and Bernoulliq
are isomorphic and there exists an isomorphism f
Theorem 12 (Rudolph 1990) The only measure such that f and f1 are both finitary.
on [0, 1) which is invariant under multiplication
by 2 and by 3 and has positive entropy under The nicest that we could hope f to be is if both
multiplication by 2 is Lebesgue measure. f and f1 are finitary and have finite expected
coding length. Schmidt proved that this happens
Johnson then proved that for all relatively only in the trivial case that p and q are
prime p and q that p and q can be substituted for rearrangements of each other.
2 and 3 in the theorem above.
This problem can be generalized to higher Theorem 14 (Schmidt 1984) If p and q are
dimensions by studying the actions of commuting probability vectors and the Bernoulli shifts
integer matrices of determinant greater than one Bernoullip and Bernoulliq are isomorphic and
on tori. Katok and Spatzier (1996) and Einsiedler there exists an isomorphism f such that f and
and Lindenstrauss (2003) obtained results similar f1 are both finitary and have finite expected
to Rudolph’s for actions of commuting matrices. coding time then p is a rearrangement of q.

The best known result about this problem is the

Finitary Isomorphisms following theorem of Harvey and Peres.

By Ornstein’s theorem we know that there exists

Theorem 15 (Harvey and Peres) If p and q are
an isomorphism between any two Bernoulli shifts
probability vectors with H(p) ¼ H(q) then
(or mixing Markov shifts) of the same entropy.
i( pi)2 log ( pi) ¼ i(qi)2 log (qi) if and only if
There has been much interest in studying how
the Bernoulli shifts Bernoullip and Bernoulliq are
“nice” the isomorphism can be. By this we mean
isomorphic and there exists an isomorphism f
can f be chosen so that the map x ! (f(x))0 is
such that f and f1 are both finitary and have
continuous and if so what is its best possible
finite one half moment.
modulus of continuity?
We say that a map f from ℕℤ to ℕℤ is finitary if
in order to determine (f(x))0 we only need to know
finitely many coordinates of x. More precisely if for Flows
almost every x there exists m(x) such that
A flow is a measure preserving action of the group
ℝ on a measure space (X, T). A cross section is
m x0 : x0i ¼ xi for alljij
any measurable set in C
X such that for almost
 mðxÞ and ðfðx0 ÞÞ0 6¼ fðxÞ0 every x

¼ 0: 0 < inf ft : T t ðxÞ  Cg < 1:

We say that m(x) has finitetth moment if For any flow (X, T) and {Tt}t  R and any cross
m(x)t dm < 1 and that f has finite expected section C we define the return time map for
coding length if the first moment of m(x) is finite. C R : C ! C as follows. For any x  C define
Isomorphism Theory in Ergodic Theory 211

tðxÞ ¼ inf ft : T t ðxÞ  Cg Other Equivalence Relations

then set R(x) ¼ Tt(x)(x). There is a standard method In this section we will discuss a number of equiv-
to project the probability measure m on X to an alence relations between transformations that are
invariant probability measure mC on C as well as weaker than isomorphism. All of these equiva-
the s-algebra ℬ on X to a s-algebra ℬC on C such lence relations have a theory that is parallel to
that (C, mC, ℬC) and R is a measure preserving Ornstein’s theory.
transformation.
First we show that there is a natural analog of Kakutani Equivalence
Bernoulli shifts for flows. We say that two transformations (X, T ) and (Y, S)
are Kakutani equivalent if there exist subsets
Theorem 16 (Ornstein 1970b) There exists a A
X and B
Y such that (TA, A, mA) and
flow (X, T ) and {Tt}t  ℝ such that for every (SB, B, nB) are isomorphic. This is equivalent to
t > 0 the map (X, T ) and Tt is isomorphic to a the existence of a flow (X, T ) and {Tt}t  ℝ with
Bernoulli shift. Moreover for any h  (0, 1] cross sections C and C0 such that the return time
there exists (X, T ) and {Tt}t  ℝ such that maps of C and C0 are isomorphic to (X, T ) and
h(T1) ¼ h. (Y, S) respectively.
Using the properties of entropy of the induced
We say that such a flow (X, {ft}t  ℝ, m, ℬ) is a map we have that if (X, T ) and (Y, S) are
Bernoulli flow. Kakutani equivalent then either h((X, T )) ¼
This next version of Ornstein’s isomorphism h((Y, S)) ¼ 0, 0 < h((X, T )), h((Y, S)) < 1 or
theorem shows that up to isomorphism and a h((X, T )) ¼ h((Y, S)) ¼ 1.
change in time (considering the flow X and {Tct} In general the answer to the question of which
instead of X and {Tt}) there are only two pairs of measure preserving transformations are
Bernoulli flows, one with positive but finite isomorphic is quite poorly understood. But if one
entropy and one with infinite entropy. of the transformations is a Bernoulli shift then
Ornstein’s theory gives a fairly complete answer
Theorem 17 (Ornstein 1970b) If (X, T ) and to the question. A similar situation exists for
{Tt}t  ℝ and (Y, S){St}t  ℝ are Bernoulli flows Kakutani equivalence. In general the answer to
and h(T1) ¼ h(S1) then they are isomorphic. the question of which pairs of measure preserving
transformation are Kakutani equivalent is also
As in the case of actions of ℤ there are many quite poorly understood. But the more specialized
natural examples of flows that are isomorphic to question of which transformations are isomorphic
the Bernoulli flow. The first is for geodesic to a Bernoulli shift has a more satisfactory answer.
flows. In the 1930s Hopf proved that geodesic Feldman constructed a transformation (X, T )
flows on compact surfaces of constant negative which has completely positive entropy but (X, T )
curvature are ergodic (Hopf 1971). Ornstein and is not Kakutani equivalent to a Bernoulli shift.
Weiss extended Hopf's proof to show that the Ornstein, Rudolph and Weiss extended Feldman’s
geodesic flow is also Bernoulli (Ornstein and work to construct a complete theory of the trans-
Weiss 1973). formations that are Kakutani equivalent to a
The second class of flows comes from billiards Bernoulli shift (Ornstein et al. 1982) for positive
on a square table with one circular bumper. The entropy transformations and a theory of the trans-
state space X consists of all positions and veloci- formations that are Kakutani equivalent to an irra-
ties for a fixed speed. The flow Tt is frictionless tional rotation (Ornstein et al. 1982) for zero
movement for time t with elastic collisions. This entropy transformations. (The zero entropy ver-
flow is also isomorphic to the Bernoulli flow sion of this theorem had been developed indepen-
(Gallavotti and Ornstein 1974). dently (and earlier) by Katok (1975).)
212 Isomorphism Theory in Ergodic Theory

They defined two class of transformations to be quite different from the same question for
called loosely Bernoulli and finitely fixed. The invertible transformations. In one sense it is easier
definitions of these properties are the same as the because of an additional isomorphism invariant.
definitions of very weak Bernoulli and finitely For any measure preserving transformation
determined except that the d metric is replaced (X, T ) the probability measure mjT 1 ðxÞ on T 1(x)
by the f metric. For x, y  ℕℤ we define is defined for almost every x  X. (If (X, T ) is
invertible then this measure is trivial as
f n ðx, yÞ ¼ 1 
k j{T 1(x)}j ¼ 1 and mjT 1 ðxÞ T 1 ðxÞ ¼ 1 for
n almost every x.) It is easy to check that if f is an
isomorphism from (X, T ) to (Y, S) then for almost
where k is the largest number such that sequences
every x and x0  T1(x) we have
1  i1 < i2 < . . . < ik  n and 1  j1 < j2 < . . . < jk  n
such that xil ¼ yjl for all j, 1  j  k. (In computer
mjT 1 ðxÞ ðx0 Þ ¼ vjS1 ðfðxÞÞ ðfðx0 ÞÞ:
science this metric is commonly referred to as the
edit distance.) Note that d n ðx, yÞ  f n ðx, yÞ.
From this we can easily see that if p ¼ fpi gni¼1
They proved the following analog of
and q ¼ fqi gmi¼1 are probability vectors then the
Theorem 5.
corresponding one sided Bernoulli shifts are iso-
morphic only if m ¼ n and there is a permutation
Theorem 18 For transformations (X, T ) with
π  Sn such that pπ(i) ¼ qi for all i. (In this case we
h((X, T )) > 0 the following conditions are
say p is a rearrangement of q.) If p is a
equivalent:
rearrangement of q then it is easy to construct an
isomorphism between the corresponding Bernoulli
1. (X, T ) is finitely fixed,
shifts. Thus the analog of Ornstein’s theorem for
2. (X, T ) is loosely Bernoulli,
Bernoulli endomorphisms is trivial. However we
3. (X, T ) is Kakutani equivalent to a Bernoulli
will see that there still is an analogous theory
shift and
classifying the class of endomorphism that are
4. There exists a Bernoulli flow Y and {Ft}t  ℝ
isomorphic to Bernoulli endomorphisms.
and a cross section C such that the return time
We say that an endomorphism is uniformly
map for C is isomorphic to (X, T ).
d to 1 if for almost every x we have that
j{T1(x)}j ¼ d and mjT 1 ðxÞ ðyÞ ¼ 1=d for all
Restricted Orbit Equivalence y  T1(x). Hoffman and Rudolph defined two
Using the d metric we got a theory of which trans- classes of noninvertible transformations called
formations are isomorphic to a Bernoulli shift. tree very week Bernoulli and tree finitely deter-
Using the f metric we got a strikingly similar theory mined and proved the following theorem.
of which transformations are Kakutani equivalent
to a Bernoulli shift. Rudolph showed that it is Theorem 19 The following three conditions are
possible to replace the d metric (or the f metric) equivalent for uniformly d to 1 endomorphisms.
with a wide number of other metrics and produce
parallel theories for other equivalence relations. 1. (X, T ) is tree very weak Bernoulli,
For instance, for each of these theories we get a 2. (X, T ) is tree finitely determined, and
version of Theorem 5. This collection of theories is 3. (X, T ) is isomorphic to the one sided Bernoulli
called restricted orbit equivalence (Rudolph 1985). d shift.

Jong extended this theorem to say that if there

Non-invertible Transformations exists a probability vector p such that for almost
every x the distribution of mjT 1 ðxÞ is the same as
The question of which noninvertible measure pre- the distribution of p then (X, T ) is isomorphic to
serving transformations are isomorphic turns out the one sided Bernoulli d shift if and only if it is
Isomorphism Theory in Ergodic Theory 213

tree finitely determined and if and only if it is tree corresponding map ((ℂ, mf, ℬ), Tf) is isomorphic
very weak Bernoulli (Jong 2003). to the one sided Bernoulli d shift.
Differences with Ornstein’s Theory
Markov Shifts Unlike Kakutani equivalence and restricted orbit
We saw that mixing Markov chains are isomor- equivalence which are very close parallels to
phic if they have the same entropy. As we have Ornstein’s theory, the theory of which endomor-
seen there are additional isomorphism invariants phisms are isomorphic to a Bernoulli endomor-
for noninvertible transformations. Ashley, Marcus phism contains some significant differences. One
and Tuncel managed to classify all one sided of the principal results of Ornstein’s isomorphism
mixing Markov chains up to isomorphism theory is that if (X, T ) is an invertible transforma-
(Ashley et al. 1997). tion and (X, T2) is isomorphic to a Bernoulli shift
then (X, T ) is also isomorphic to a Bernoulli shift.
Rational Maps There is no corresponding result for noninvertible
Rational maps are the main object of study in transformations.
complex dynamics. For every rational function
f(z) ¼ p(z)/q(z) there is a nonempty compact set Theorem 23 (Hoffman 2004) There is a uni-
Jf which is called the Julia set. Roughly speaking formly two to one endomorphism (X, T ) which is
this is the set of points for which every neighbor- not isomorphic to the one sided Bernoulli 2 shift
hood acts “chaotically” under repeated iterations but (X, T2) is isomorphic to the one sided
of f. Bernoulli 4 shift.
In order to consider rational maps as measure
preserving transformations we need to specify an
invariant measure. The following theorem of Factors of a Transformation
Gromov shows that for every rational map there
is one canonical measure to consider. In this section we study the relationship between a
transformation and its factors. There is a natural
Theorem 20 (Gromov 2003) For every way to associate a factor of (X, T ) with a sub
f rational function of degree d there exists a s-algebra of ℬ. Let (Y, S) be a factor of (X, T )
unique invariant measure mf of maximal entropy. with factor map f : X ! Y. Then the s-algebra
We have that h(mf) ¼ log2d and mf(Jf) ¼ 1. associated with (Y, S) is ℬY ¼ f1 ðC Þ. Thus the
study of factors of a transformation is the study of
The properties of this measure were studied by its sub s-algebras.
Freire, Lopes and Mañé (1983). Mañé that analy- Almost every property that we have discussed
sis to prove the following theorem. above has an analog in the study of factors of a
transformation. We give three such examples. We
Theorem 21 (Mañé 1985) For every rational say that two factors C and D of (X, T ) are rela-
function f of degree d there exists n such that tively isomorphic if there exists an isomorphism
(ℂ, f n, mf) (where f n(z) ¼ f( f( f. . .(z))) is compo- c : X ! X of (X, T ) with itself such that
sition) is isomorphic to the one sided Bernoulli dn cðC Þ ¼ D. We say that (X, T ) has relatively
shift. completely positive entropy with respect to C
if every factor D which contains D has
Heicklen and Hoffman used the tree very weak hðD Þ > hðC Þ. We say that C is relatively
Bernoulli condition to show that we can always Bernoulli if there exists a second factor D which
take n to be one. is independent of D and ℬ ¼ C _ D.
Thouvenot defined properties of factors called
Theorem 22 (Heicklen and Hoffman 2002) relatively very weak Bernoulli and relatively
For every rational function f of degree d  2 the finitely determined. Then he proved an analog
214 Isomorphism Theory in Ergodic Theory

of Theorem 3. This says that a factor being rela- results discussed above we can ask what is the
tively Bernoulli is equivalent to it being relatively largest class of groups such that an analogous
finitely determined (and also equivalent to it being result is true. It turns out that for most of the
relatively very weak Bernoulli). properties described above the right class of
The Pinsker algebra is the maximal s-algebra groups is discrete amenable groups.
P such that hðP Þ ¼ 0. The Pinsker conjecture was A Følner sequenceFn in a group G is a
that for every measure preserving transformation sequence of subsets Fn of G such that for all
(X, T ) there exists a factor C such that g  G we have that lim j gðFn Þ j=j Fn j. A count-
n!1
able group is amenable if and only if it has a
1. C is independent of the Pinsker algebra P Følner sequence.
2. ℬ ¼ C _ P and For nonamenable groups it is much more diffi-
3. ðX, C Þ has completely positive entropy. cult to generalize Birkhoff’s ergodic theorem
(Nevo 1994; Nevo and Stein 1994). Lindenstrauss
Ornstein found a counterexample to the Pinsker proved that for every discrete amenable group there
conjecture (Ornstein 1973c). After Thouvenot is an analog of the ergodic theorem (Lindenstrauss
developed the relative isomorphism theory he 2001). For every amenable group G and every
came up with the following question which is probability vector p we can define a Bernoulli
referred to as the weak Pinsker conjecture. action of G. There are also analogs of Rokhlin’s
lemma and the Shannon–McMillan–Breiman the-
Conjecture 2 For every measure preserving orem for actions of all discrete amenable groups
transformation (X, T ) and every ϵ > 0 there (Krieger 1970; Ornstein and Weiss 1983, 1987).
exist invariant s-algebras C , D
ℬ such that Thus we have all of the ingredients to prove a
version of Ornstein’s isomorphism theorem.
1. C is independent of D
2. ℬ¼C _D Theorem 24 If p and q are probability vectors
3. ðX, T, m, D Þ is isomorphic to a Bernoulli shift and H(p) ¼ H(q) then the Bernoulli actions of
4. hððX, T, m, C ÞÞ < ϵ: G corresponding to p and q are isomorphic.

Also all of the aspects of Rudolph’s theory of

There is a wide class of transformations which
restricted orbit equivalence can be carried out for
have been proven to satisfy the weak Pinsker
actions of amenable groups (Kammeyer and
conjecture. This class includes almost all measure
Rudolph 2002).
preserving transformations which have been
extensively studied.
Differences Between Actions of ℤ and Actions
of Other Groups
Actions of Amenable Groups Although generalizations of Ornstein theory and
restricted orbit equivalence carry over well to the
All of the discussion above has been about the actions of discrete amenable groups there do turn
action of a single invertible measure preserving out to be some significant differences between the
transformation (actions of ℕ and ℤ) or flows possible actions of ℤ and those of other discrete
(actions of ℝ). We now consider more general amenable groups.
group actions. If we have two actions S and T on Many of these have to do with the generaliza-
a measure space (X, m) which commute tion of Markov shifts. For actions of ℤ2 these are
(S(T(x)) ¼ T(S(x)) for almost every x) then there called Markov random fields. By the result of
is an action of ℤ2 on (X, m) given by f(n,m)(x) ¼ Friedman and Ornstein if a Markov chain is
Sn(Tm(x)). A natural question to ask is do there mixing then it has completely positive entropy
exist a version of entropy theory and Ornstein’s and it is isomorphic to a Bernoulli shift. Mixing
isomorphism theory for actions of two commuting Markov random fields can have very different
automorphisms. More generally for each of the properties. Ledrappier constructed a ℤ2 action
Isomorphism Theory in Ergodic Theory 215

which is a Markov random field and is mixing but Breiman L (1957) The individual ergodic theorem of infor-
has zero entropy (Ledrappier 1978). Even more mation theory. Ann Math Stat 28:809–811
Den Hollander F, Steif J (1997) Mixing E properties of the
surprising even though it is mixing it is not mixing generalized T,T1-process. J Anal Math 72:165–202
of all orders. The existence of a ℤ action which is Einsiedler M, Lindenstrauss E (2003) Rigidity properties
mixing but not mixing of all orders is one of the of Zd-actions on tori and solenoids. Electron Res
longest standing open questions in ergodic theory Announc Am Math Soc 9:99–110
Einsiedler M, Katok A, Lindenstrauss E (2006) Invariant
(Halmos 1950). measures and the set of exceptions to Littlewood’s
Even if we try to strengthen the hypothesis of conjecture. Ann Math (2) 164(2):513–560
Friedman and Ornstein’s theorem to assume that Feldman J (1976) New K-automorphisms and a problem of
the Markov random field has completely positive Kakutani. Isr J Math 24(1):16–38
Freire A, Lopes A, Mañé R (1983) An invariant measure
entropy we will not succeed as there exists a for rational maps. Bol Soc Brasil Mat 14(1):45–62
Markov random field which has completely pos- Friedman NA, Ornstein DS (1970) On isomorphism of
itive entropy but is not isomorphic to a Bernoulli weak Bernoulli transformations. Adv Math 5:365–394
shift (Hoffman 1999b). Furstenberg H (1967) Disjointness in ergodic theory, min-
imal sets, and a problem in Diophantine approximation.
Math Syst Theory 1:1–49
Gallavotti G, Ornstein DS (1974) Billiards and Bernoulli
Future Directions schemes. Commun Math Phys 38:83–101
Gromov M (2003) On the entropy of holomorphic maps.
In the future we can expect to see progress of Enseign Math (2) 49(3–4):217–235
isomorphism theory in a variety of different direc- Halmos PR (1950) Measure theory. D Van Nostrand,
New York
tions. One possible direction for future research is
Halmos PR (1960) Lectures on ergodic theory. Chelsea
better understand the properties of finitary iso- Publishing, New York
morphisms between various transformations and Harvey N, Peres Y. An invariant of finitary codes with
Bernoulli shifts described in section “Finitary Iso- finite expected square root coding length. Ergod The-
ory Dynam Syst, to appear
morphisms”. Another possible direction would be
Heicklen D (1998) Bernoullis are standard when entropy is
to find a theory of equivalence relations for not an obstruction. Isr J Math 107:141–155
Bernoulli endomorphisms analogous to the one Heicklen D, Hoffman C (2002) Rational maps are d-adic
for invertible Bernoulli transformations described Bernoulli. Ann Math (2) 156(1):103–114
Hoffman C (1999a) A K counterexample machine. Trans
in section “Other Equivalence Relations”.
Am Math Soc 351(10):4263–4280
As the subject matures the focus of research in Hoffman C (1999b) A Markov random field which is K but
isomorphism theory will likely shift to connections not Bernoulli. Isr J Math 112:249–269
to other fields. Already there are deep connections Hoffman C (2003) The scenery factor of the [T,T1] trans-
formation is not loosely Bernoulli. Proc Am Math Soc
between isomorphism theory and both number the-
131(12):3731–3735
ory and statistical physics. Finally one hopes to see Hoffman C (2004) An endomorphism whose square is
progress made on the two dominant outstanding Bernoulli. Ergod Theory Dynam Syst 24(2):477–494
conjectures in the field: Thouvenot weak Pinsker Hoffman C, Rudolph D (2002) Uniform endomorphisms
which are isomorphic to a Bernoulli shift. Ann Math
conjecture (Conjecture 2) and Furstenberg’s con-
(2) 156(1):79–101
jecture (Conjecture 1) about measures on the circle Hopf E (1971) Ergodic theory and the geodesic flow on
invariant under both the times 2 and times 3 maps. surfaces of constant negative curvature. Bull Am Math
Progress on either of these conjectures would Soc 77:863–877
Jong P (2003) On the isomorphism problem of
invariably lead the field in exciting new directions.
p-endomorphisms. PhD thesis, University of Toronto
Kalikow SA (1982) T,T1 transformation is not loosely
Bernoulli. Ann Math (2) 115(2):393–409
Bibliography Kammeyer JW, Rudolph DJ (2002) Restricted orbit equiv-
alence for actions of discrete amenable groups. Cam-
Ashley J, Marcus B, Tuncel S (1997) The classification of bridge tracts in mathematics, vol 146. Cambridge
one-sided Markov chains. Ergod Theory Dynam Syst University Press, Cambridge
17(2):269–295 Katok AB (1975) Time change, monotone equivalence,
Birkhoff GD (1931) Proof of the ergodic theorem. Proc and standard dynamical systems. Dokl Akad Nauk
Natl Acad Sci U S A 17:656–660 SSSR 223(4):789–792. in Russian
216 Isomorphism Theory in Ergodic Theory

Katok A (1980) Smooth non-Bernoulli K-automorphisms. Ornstein DS, Weiss B (1987) Entropy and isomorphism
Invent Math 61(3):291–299 theorems for actions of amenable groups. J Anal Math
Katok A, Spatzier RJ (1996) Invariant measures for higher- 48:1–141
rank hyperbolic abelian actions. Ergod Theory Dynam Ornstein DS, Rudolph DJ, Weiss B (1982) Equivalence of
Syst 16(4):751–778 measure preserving transformations. Mem Am Math
Katznelson Y (1971) Ergodic automorphisms of Tn are Soc 37(262) American Mathematical Society
Bernoulli shifts. Isr J Math 10:186–195 Parry W (1969) Entropy and generators in ergodic theory.
Keane M, Smorodinsky M (1979) Bernoulli schemes of the WA Benjamin, New York
same entropy are finitarily isomorphic. Ann Math Parry W (1981) Topics in ergodic theory. Cambridge tracts
(2) 109(2):397–406 in mathematics, vol 75. Cambridge University Press,
Kolmogorov AN (1958) A new metric invariant of tran- Cambridge
sient dynamical systems and automorphisms in Petersen K (1989) Ergodic theory. Cambridge studies in
Lebesgue spaces. Dokl Akad Nauk SSSR (NS) 119: advanced mathematics, vol 2. Cambridge University
861–864. in Russian Press, Cambridge
Kolmogorov AN (1959) Entropy per unit time as a metric Pinsker MS (1960) Dynamical systems with completely
invariant of automorphisms. Dokl Akad Nauk SSSR positive or zero entropy. Dokl Akad Nauk SSSR 133:
124:754–755. in Russian 1025–1026. in Russian
Krieger W (1970) On entropy and generators of measure- Ratner M (1978) Horocycle flows are loosely Bernoulli. Isr
preserving transformations. Trans Am Math Soc 149: J Math 31(2):122–132
453–464 Ratner M (1982) Rigidity of horocycle flows. Ann Math
Ledrappier F (1978) Un champ markovien peut être (2) 115(3):597–614
d’entropie nulle et mélangeant. CR Acad Sci Paris Sér Ratner M (1983) Horocycle flows, joinings and rigidity of
A–B 287(7):A561–A563. in French products. Ann Math (2) 118(2):277–313
Lindenstrauss E (2001) Pointwise theorems for amenable Ratner M (1991) On Raghunathan’s measure conjecture.
groups. Invent Math 146(2):259–295 Ann Math (2) 134(3):545–607
Lyons R (1988) On measures simultaneously 2- and Rudolph DJ (1976) Two nonisomorphic K-automorphisms
3-invariant. Isr J Math 61(2):219–224 with isomorphic squares. Isr J Math 23(3–4):274–287
Mañé R (1983) On the uniqueness of the maximizing mea- Rudolph DJ (1979) An example of a measure preserving
sure for rational maps. Bol Soc Brasil Mat 14(1):27–43 map with minimal self-joinings, and applications.
Mañé R (1985) On the Bernoulli property for rational J Anal Math 35:97–122
maps. Ergod Theory Dynam Syst 5(1):71–88 Rudolph DJ (1983) An isomorphism theory for Bernoulli
Meilijson I (1974) Mixing properties of a class of skew- free Z-skew-compact group actions. Adv Math 47(3):
products. Isr J Math 19:266–270 241–257
Meshalkin LD (1959) A case of isomorphism of Bernoulli Rudolph DJ (1985) Restricted orbit equivalence. Mem Am
schemes. Dokl Akad Nauk SSSR 128:41–44. in Russian Math Soc 54(323) American Mathematical Society
Nevo A (1994) Pointwise ergodic theorems for radial aver- Rudolph DJ (1988) Asymptotically Brownian skew prod-
ages on simple Lie groups. I. Duke Math J 76(1):113–140 ucts give non-loosely Bernoulli K-automorphisms.
Nevo A, Stein EM (1994) A generalization of Birkhoff’s Invent Math 91(1):105–128
pointwise ergodic theorem. Acta Math 173(1):135–154 Rudolph DJ (1990) 2 and 3 invariant measures and
Ornstein D (1970a) Factors of Bernoulli shifts are entropy. Ergod Theory Dynam Syst 10(2):395–406
Bernoulli shifts. Adv Math 5:349–364 Schmidt K (1984) Invariants for finitary isomorphisms
Ornstein D (1970b) Two Bernoulli shifts with infinite with finite expected code lengths. Invent Math 76(1):
entropy are isomorphic. Adv Math 5:339–348 33–40
Ornstein D (1970c) Bernoulli shifts with the same entropy Shields P (1973) The theory of Bernoulli shifts. Chicago
are isomorphic. Adv Math 4:337–352 lectures in mathematics. University of Chicago Press,
Ornstein DS (1973a) A K automorphism with no square Chicago
root and Pinsker’s conjecture. Adv Math 10:89–102 Sinaĭ J (1959) On the concept of entropy for a dynamic
Ornstein DS (1973b) An example of a Kolmogorov auto- system. Dokl Akad Nauk SSSR 124:768–771. in Russian
morphism that is not a Bernoulli shift. Adv Math 10: Sinaĭ JG (1962) A weak isomorphism of transformations
49–62 with invariant measure. Dokl Akad Nauk SSSR 147:
Ornstein DS (1973c) A mixing transformation for which 797–800. in Russian
Pinsker’s conjecture fails. Adv Math 10:103–123 Thouvenot J-P (1975a) Quelques propriétés des systèmes
Ornstein DS, Shields PC (1973) An uncountable family of dynamiques qui se décomposent en un produit de deux
K-automorphisms. Adv Math 10:63–88 systèmes dont l'un est un schéma de Bernoulli. Confer-
Ornstein DS, Weiss B (1973) Geodesic flows are ence on ergodic theory and topological dynamics, Kib-
Bernoullian. Isr J Math 14:184–198 butz, Lavi, 1974. Isr J Math 21(2–3):177–207. in
Ornstein DS, Weiss B (1974) Finitely determined implies French
very weak Bernoulli. Isr J Math 17:94–104 Thouvenot J-P (1975b) Une classe de systèmes pour
Ornstein D, Weiss B (1983) The Shannon–McMillan– lesquels la conjecture de Pinsker est vraie. Conference
Breiman theorem for a class of amenable groups. Isr on ergodic theory and topological dynamics, Kibbutz
J Math 44(1):53–60 Lavi, 1974. Isr J Math 21(2–3):208–214. in French
 Gaussian process, Gaussian space A Gaussian
Dynamical Systems of process is a family of real-valued random vari-
Probabilistic Origin: Gaussian ables defined on a probability space (Ω, ℙ),
and Poisson Systems such that any linear combination of finitely
many of these random variables is either 0 or
Élise Janvresse1, Emmanuel Roy2 and normally distributed.
Thierry De La Rue3 A real linear subspace of L2(ℙ) is a
1
Laboratoire Amiénois de Mathématique Gaussian space if any nonzero random vari-
Fondamentale et Appliquée, CNRS-UMR 7352, able it contains is normally distributed. The
Université de Picardie Jules Verne, Amiens, closure of the linear real subspace spanned by
France a Gaussian process is a Gaussian space.
2
Laboratoire Analyse, Géométrie et Applications, Infinite divisibility Let ðG, G, þÞ be a measur-
Université Paris 13 Institut Galilée, Villetaneuse, able Abelian semigroup, i.e., the addition
France ðG  G, G  GÞ 7! ðG, GÞ
3
Laboratoire de Mathématiques Raphaël Salem, ðg1 , g2 Þ 7! g1 þ g 2
CNRS – Université de Rouen Normandie, Saint is commutative and measurable. The convolu-
Étienne du Rouvray, France tion n  r of probability measures n and r on
ðG, G Þ is well defined as the image of v  r by
the addition.
Article Outline A probability measure n on ðG, G Þ is
infinitely divisible if for any k  1, there exists
Glossary a probability measure nk on ðG, G Þ such
Definition of the Subject that n ¼ (nk)k.
Introduction Kronecker subset of the unidimendional
From Probabilistic Objects to Dynamical Systems torus A subset K of the unidimendional torus
Spectral Theory  ¼ ℝ=ℤ is a Kronecker set if any continuous
Basic Ergodic Properties function f : K ! S1 is a uniform limit of
Joinings, Factors, and Centralizer characters: there exists a sequence (kn)  ℤ
GAGs and PAPs such that
Future Directions
max j f ðtÞ  ei2pkn t j ! 0:
Bibliography tK n!1
Any finite set of rationally independent ele-
Glossary ments of  is a Kronecker set, but there exist
also perfect Kronecker subsets of  (see for
Centralizer The centralizer of an invertible example (Cornfeld et al. 1982), Appendix 4]).
measure-preserving transformation T is the set Point process A point process N on ðX, A Þ is a
C(T) of all invertible measure-preserving trans- random variable taking values in the space X
formations on the same measure space which of counting measures on ðX, A Þ. It is said to be
commute with T. simple if N is almost surely a simple counting
(Simple) Counting measure A counting mea- measure.
sure on a measurable space ðX, A Þ is a mea- The measure A 7! ½N ðAÞ on ðX, A Þ is
sure of the form i  I dxi where (xi)i  I is a called the intensity of N.
countable family of elements of X. The A point process of intensity m is said to
counting measure is said to be simple if xi 6¼ xj have moment of order k  1 if, for all A  A
whenever i 6¼ j.
© Springer Science+Business Media, LLC, part of Springer Nature 2023 217
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_725
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2019
https://doi.org/10.1007/978-3-642-27737-5_725-1
218 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

with 0 < m(A) < 1, unidimendional torus, called the spectral mea-
 N ð AÞ k
< 1: sure of h, satisfying for all k  ℤ

Poisson point process Let ðX, A, mÞ be a

sh ðkÞ≔ ei2pkt dsh ðtÞ ¼ h, Uk h : ð1Þ
sigma-finite measure space. A Poisson point 
process of intensity m on X is a point process
N on X, such that When T is an invertible measure-preserving
• For any set A  A with 0 < m(A) < 1, transformation on a measure space ðX, A, mÞ,
N(A) is a Poisson random variable of parameter this applies in particular to the associated
m(A). Koopman operator UT. In the case of a square
• For any k  1, for any collection integrable stationary process (xn)n  ℤ (see defini-
(A1, . . ., Ak) in A , the random variables tion below), its distribution being shift-invariant,
N(Ai), 1 i k, are independent. the covariances E x0 xk of the process can be
A Poisson process is simple if and only if interpreted as scalar products hh, Ukhi, where
its intensity is a continuous measure. h is the projection on the 0-coordinate of the
Self-joining and associated Markov operator A process, and U is the Koopman operator associ-
self-joining of the probability-preserving ated to the shift. Thus the spectral measure s of
dynamical system ðX, A, m, T Þ is a proba- the process satisfies, for all k  ℤ,
bility measure n on the Cartesian square
ðX  X, A  A Þ such that sðkÞ ¼  x0 xk : ð2Þ
• Both marginals of n are equal to m.
• n is T  T-invariant. When n is such a In the case of a real-valued stationary process,
self-joining, it gives rise to a bigger the covariances are real and this measure s must
probability-preserving dynamical system be symmetric (invariant by t !  t).
ðX  X, A  A, n, T  T Þ in which we see
two copies of the original system as factors (via Definition of the Subject
the projections on the two coordinates).
To a self-joining n there corresponds a unique Measure-theoretic dynamical systems are systems
Markov operator Fv (i.e., Fv is a positive oper- of the form ðX, A, m, T Þ , where ðX, A Þ is a
ator on L2(m) with Fv1 ¼ 1) that commutes standard Borel space, m is a sigma-finite measure
with the Koopman operator UT (i.e., the unitary on ðX, A Þ, and T : X ! X is an invertible mea-
operator defined on L2(m) by UTh ≔ h ∘ T), and surable transformation preserving m. In many
which is characterized by the following rela- cases, m is a probability measure, and the theory
tion: for any A and B in A : of probability-preserving dynamical systems has
n ð A  BÞ ¼ h 1 A , F v 1 B i L2 ð mÞ : considerably developed since the mid-twentieth
century. In the immense zoo of examples that
Stationary process A stationary process is a
have caught the interest of mathematicians, very
sequence (xn)n  ℤ of random variables taking
interesting and fundamental families are directly
values in a set V, such that for each k  ℤ and
issued from probability theory. The purpose of
any i1 < < id  ℤ, the distribution of
this entry is to present two of them: systems aris-
xi1 þk , . . ., xid þk is the same as the distribu- ing from Gaussian processes and systems
tion of xi1 , . . ., xid . In other words, the constructed from Poisson point processes.
process is stationary if its distribution is shift- Although these two families are of different
invariant on Vℤ. nature, and each one has to be addressed with
Spectral measure Let U be a unitary operator on specific techniques, a good reason to present
a Hilbert space H. For any h  H, there exists a them in parallel is that they share striking common
finite positive measure sh on the features.
Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems 219

Introduction From Probabilistic Objects to Dynamical

Systems
As we all know, Gaussian distribution plays a prom-
inent role in probability theory and in mathematics Gaussian Systems
in general. It comes with a very rich structure, Stationary Gaussian process. We only consider
interesting objects and results, that fueled intensive here Gaussian processes (xn)n  ℤ that are cen-
studies by countless mathematicians. Therefore, it tered: ½xn  ¼ 0 for each n. A fundamental prop-
comes as no surprise that Gaussian distribution had erty of Gaussian processes is the fact that their
its declination in ergodic theory, in the form of distribution law is completely determined by the
stationary Gaussian processes, and received much covariances  xi x j . In particular, such a process
attention over the years. Their basic ergodic prop- is stationary if and only if the covariances satisfy,
erties started to be studied in the middle of the for all i, j  ℤ,  xi x j ¼  xij x0 .
twentieth century, especially by Ito (1944), In the stationary case, the law of the Gaussian
Maruyama (1949), and Fomin (1950). As we can process is therefore completely determined by its
completely control their spectral structure with a spectral measure, which is a symmetric positive
relatively simple object which is the spectral mea- finite measure s on  whose Fourier coefficients
sure of the generating process, Gaussian systems are given by (2).
have proved to be a very rich source of examples of Standard Gaussian systems. Conversely, given
measure-theoretic dynamical systems with specific a symmetric positive finite measure s on T, there
properties (see the examples given by Girsanov exists a unique (up to equality in distribution)
(1958), Totoki (1964), and Newton (1966)). centered stationary Gaussian process whose
The Poisson point process has a long history. It covariances are given by (2). (In some sense,
was first considered on the line to represent the this ability of realizing any spectral measure char-
occurrences of independent events over time. The acterizes Gaussian processes (see section “From
idea of letting the random points move to model Foias-Stratila to GAGs”). The distribution ms of
the behavior of a large number of particles this stationary Gaussian process is shift-invariant,
(Dobrushin 1956; Doob 1953) led to study the thus we can consider the probability-preserving
Poisson point process on more general spaces. dynamical system
The first formalization of Poisson suspensions,
due to Goldstein, Lebowitz, and Aizenman in Xs ≔ ℝℤ , B ℝℤ , ms , S
1975 (Goldstein et al. 1975), was also motivated
by questions of statistical physics. Simultaneously where S denotes the shift map: (xn)n  ℤ 7!
and independently, Vershik et al. (1975) consid- (xn+1)n  ℤ.
ered the action of a group of transformations on A standard Gaussian system is a probability-
configurations of random points. This initiated the preserving dynamical system ðX, B, m, T Þ iso-
systematic study of ergodic properties of these morphic to Xs for some symmetric positive finite
systems, in particular by Marchat (1978), measure s on . In other words, ðX, B, m, T Þ is
Grabinsky (1984) and Kalikow (1981). a standard Gaussian system if the sigma-algebra B
It is worth mentioning Maruyama (1970) again is generated by some Gaussian process (xn)n  ℤ
who used both Poisson processes and Gaussian with xn ¼ x0 ∘ T n for each n  ℤ. (Multiplying if
stationary processes as a tool to represent any necessary the process (xn) by a constant, we can
infinitely divisible stationary processes and always assume that  x20 ¼ 1, and in this case the
obtained results on ergodicity and mixing for spectral measure s is a probability measure.)
those processes.
220 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

The classical theory of Gaussian systems stud- symmetrized of γ, defined for each measurable
ies the above-mentioned standard Gaussian sys- subset A of  by
tems. But following (Lemańczyk et al. 2000), it is
useful to introduce also the generalized Gaussian gðAÞ þ gðAÞ
sðAÞ ¼ :
systems as probability-preserving dynamical sys- 2
tems ðX, B, m, T Þ satisfying the following
more general property: there exists a closed real Moreover, if s is continuous, (xp)p  ℤ gener-
subspace H of L2(m) such that ates the same sigma-algebra as (Bt)0 t 1, and the
measure-preserving system (C0([0, 1], ℂ), mW,
• H is a Gaussian space. Tγ) is a Gaussian dynamical system isomorphic to
• H is invariant by UT (in particular, the station- Xs.
ary process (x ∘ T n)n  ℤ is Gaussian).
• The sigma-algebra generated by H is B. Poisson Suspensions
Poisson point process. Let ðX, A Þ be a standard
Geometric interpretation. A geometric model Borel space, and let ðX , A  Þ be the canonical
for a standard Gaussian dynamical system is pro- space of point processes on ðX, A Þ, where
posed in de la Rue (1995) as a transformation of a
complex Brownian motion path. More precisely, • X  is the set of counting measures on X,
let B ¼ (Bt)0 t 1 be a complex Brownian motion • A  is the sigma-algebra generated by the maps
with B0 ¼ 0. For any given probability measure γ ðN A ÞA  A , where
on  ðidentified here with ½0, 1Þ, let us define,
for 0 s < 1
X ! ℕ [ fþ1g
NA :
yðsÞ ≔ inf fx  ½0, 1 ½: gð½0, xÞ  sg: o 7! o ðAÞ:

Now we consider a new process B ¼ Bt When m is a sigma-finite measure on ðX, A Þ, a

0 t 1 Poisson point process of intensity m on X always
given by exists. We denote by m its distribution, which is a
probability measure on ðX , A  Þ.
t
Bt ≔ ei2pyðsÞ dBs : ð3Þ Poisson suspensions. The distribution m of a
0 Poisson point process of intensity m has many nice
functorial features. One of the most important is
Then B is also a complex Brownian motion the following.
with the same law as B, and this enables us to Let m and n be two sigma-finite measures on the
define a transformation T g : B 7! B of the canon- standard Borel spaces ðX, A Þ and ðY, B Þ, respec-
ical space C0([0, 1], ℂ) of the Brownian motion, tively, and consider a measurable map ’ : X ! Y
preserving the Wiener measure mW. This transfor- such that m ∘ ’1 ¼ n. Then the map ’ X ! Y
mation is invertible (its inverse being given by a acting by
similar formula, where we replace ei2πθ(s) by
ei2πθ(s)). Then we define the real valued station-
ary process (xp)p  ℤ by ’ d xi ≔ d’ðxi Þ
iI iI

xp ≔ℜe B1 ∘ T pg : satisfies m ∘ ’1 

 ¼ n .
In particular, if T is a measure-preserving trans-
It turns out that (xp)p  ℤ is a stationary Gauss- formation on ðX, A, mÞ, then T is a probability-
ian process, whose spectral measure s is the preserving transformation of ðX , A  , m Þ. The
Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems 221

system ðX , A  , m , T  Þ is called the Poisson L2(Xn, mn) consisting of functions which are
suspension over the base ðX, A, m, T Þ. invariant by coordinate permutations.
Of course the properties of the Poisson suspen- With the convention H 0 ≔ ℂ, we can consider
sions depend on those of the base system, and this the vector space n0H n which is the formal direct
allows to make strong connections between sum whose elements are finite sums of vectors of
infinite-measure-preserving systems and H n, n  1. The space n0H n can be equipped
probability-preserving systems. with a scalar product by considering the direct sum as
orthogonal and by endowing each H n with the
scalar product h , iHn. The (Boson) Fock space
F(H) of H is the Hilbert space obtained as the com-
Spectral Theory
pletion of n  0H n with respect to the norm of the
scalar product we just set up.
Basics of Spectral Theory
Let us recall some basic notions of spectral theory
(see e.g., (Lemańczyk 2011)). Operators on a Fock Space
Let U be a unitary operator acting on a separa- Whenever F is an operator on H of norm less than
ble Hilbert space H, and let h  H. We denote by or equal to 1, it extends naturally to an operator F
sh the spectral measure of h (see (1)). Let C(h) be on the Fock space F(H) by acting on H n as F n,
the cyclic space of h under U, that is, the closure of that is
the linear span of the vectors Unh, n  ℤ. The
linear map between C(h) and L2(sh) that maps 8v  H, F ðv   vÞ ≔Fv   Fv:
Unh to ei2πn extends to an isometry and intertwins
the operator U with the unitary operator This operator F is called the second quantiza-
tion of F (see Attal n.d., Chap. 8, p. 16).
V : f 7! ei2p f , f  L2 ðsh Þ: ð4Þ The following proposition considers the sec-
ond quantization U of a unitary operator U.
The maximal spectral type of U is the equiva-
lence class of shmax for some hmax satisfying: Proposition 4.1 If U is unitary on H and smax is
in the equivalence class of its maximal spectral
8g  H, sg shmax : type, then

(Such an hmax always exists.) • The second quantization U of U is unitary on

We say that U has simple spectrum whenever F(H).
H itself is a cyclic space C(h) for some vector h. • The maximal spectral type of U is the equiva-
Whenever ðX, A, m, T Þ is a measure- lence class of n0 sn 0
max (where smax ≔d0 ).
preserving dynamical system with a sigmafinite • U has simple spectrum if and only if for all n,
measure m, we denote by UT : L2(m) ! L2(m) the U n has simple spectrum and for all n 6¼
associated Koopman operator defined by m, sn m
max ⊥smax .
UTh ≔ h ∘ T. It is a unitary operator on L2(m).

Fock Space
We describe here an algebraic construction that Application to Gaussian and Poisson Chaos
plays a crucial role in the study of our objects.
Let once again H be a Hilbert space and denote Fock Space Structure of L2 for Gaussian Dynamical
H n, the vector space of symmetric elements of the Systems and Poisson Suspensions
n-th tensor product Hn. When H is L2(X, m) for In the case of the standard Gaussian dynamical
some sigma-finite measure m on X, then H n can be system Xs, generated by the Gaussian process
identified with the subspace L2perm ðXn , mn Þ of (xn)n  ℤ, we denote by H r1 the real Gaussian
222 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

subspace of L2(ms) spanned by the random vari- In the case of the standard Gaussian system Xs,
ables xn, n  ℤ, and H1 ≔H r1 þ iH r1 the complex the action of US on the first chaos H1 corresponds to
subspace spanned by the process. Then H1 is the multiplication by ei2π in L2 ð, sÞ. In general,
isometric to L2 ð, sÞ, with an isometry the action of US on the n-th chaos Hn corresponds to
extending the correspondence xn $ ei2πn(n  ℤ). the multiplication by the function
In the case of the Poisson suspension ðt1 , . . ., tn Þ 7! ei2pðt1 þ þtn Þ in L2perm ðn , sn Þ.
ðX , A  , m , T  Þ, we denote by H1 the subspace In the case of the Poisson suspension
of L2(m) spanned by the random variables of the ðX , A  , m , T  Þ, the action of U T  on H1 cor-
form N A  mðAÞ, A  A, mðAÞ < 1. In this case responds to the action of UT on L2(m). In general,
H1 is isometric to L2(X, m), with an isometry the action of UT  on the n-th chaos Hn corresponds
extending the correspondence N A  mðAÞ $ to the action of UT n on L2perm ðXn , mn Þ.
1A ðA  A, mðAÞ < 1Þ.
In both cases, H0 denotes the subspace of con-
stant functions. Then for each n  2, we define
inductively the subspace Hn as the ortho- Basic Ergodic Properties
complement of 0 j<nHj in the space of all poly-
nomials of degree at most n in variables in H1 Ergodicity and Mixing
(note that elements of H1 always have moments of Ergodic properties as spectral properties. We list
any order). In the classical probabilist terminol- some classical ergodic properties that are
ogy, the subspace Hn is called the n-th chaos. spectral by nature. We start by the probability-
It turns out that Hn is, in the Gaussian case, preserving case where it is customary to consider
isometric to L2perm ðn , sn Þ, whereas in the Poisson UT : f 7! f ∘ T acting on L20 ðmÞ ¼ L2 ðmÞ ℂ, the
orthocomplement of the constant functions as UT
case it is isometric to L2perm ðXn , mn Þ. We therefore
acts trivially on the latter: UT(α1lX) ¼ α1lX. In
get the following description of L2 as a Fock space,
that case, the reduced maximal spectral type
which in the Gaussian case takes the form
denotes the maximal spectral type of UT acting
on L20 ðmÞ.
L2 ðms Þ ¼ H n ’ F L2 ð, sÞ ,
n0
Theorem 5.1 Let ðX, A, m, T Þ be a dynamical
and in the Poisson case system with m(X) ¼ 1, UT its Koopman operator
acting on L20 ðmÞ and r0max , a finite measure in the
L2 ðm Þ ¼ H n ’ F L2 ðX, mÞ : measure class of the reduced maximal spectral type.
n0

• ðX, A, m, T Þ is ergodic if and only if

We refer to Cornfeld et al. (1982) for more r0max ðf0gÞ ¼ 0.
details about the Fock space structure in the
• ðX, A, m, T Þ is weakly mixing if and only if
Gaussian case, and to Last and Penrose (2018)
r0max is continuous.
and Neretin (1996) in the Poisson case.
• ðX, A, m, T Þ is (strongly) mixing if and only
if r0max is Rajchman, i.e., r0max ðkÞ ! 0.
Second Quantization Operators jkj!1
2
Once the Fock space structure of L (ms) (in the
Gaussian case) and L2(m) (in the Poisson suspen-
sion case) is established, it is natural to look at the When m is not finite, nonzero constant func-
Koopman operator that is associated to the under- tions are not in L2(m) and that makes a radical
lying transformation. In both cases, this Koopman difference.
operator leaves each chaos Hn invariant, and is
nothing but the second quantization of its restriction Theorem 5.2 Let ðX, A, m, T Þ be a dynamical
to H1. More precisely, we can state the following. system with m sigma-finite, UT its Koopman
Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems 223

operator acting on L2(m) and rmax a finite mea- yields to the following characterizations (proved
sure in the measure class of the maximal spectral by Marchat (1978) and Grabinsky (1984)).
type.
Theorem 5.4 The Poisson suspension
• ðX, A, m, T Þ has no T-invariant set of finite ðX , A  , m , T  Þ is ergodic if and only if the
positive measure if and only if rmax is base system ðX, A, m, T Þ has no invariant set
continuous. of finite positive measure. In this case it is also
• ðX, A, m, T Þ is of zero type (i.e., m- weakly mixing.
(A \ T nB) ! 0 as n tends to infinity for all The Poisson suspension ðX , A  , m , T  Þ is
finite m-measure sets A and B in A) if and only (strongly) mixing if and only if the base system
if rmax is Rajchman. ðX, A, m, T Þ is of zero type.

Leonov (1960) proved that, if a Gaussian sys-

In particular, these two properties can only tem is mixing, then it is mixing of all orders (see
happen if m is infinite. Note also that ergodicity also (Totoki 1964) for a simple proof by Totoki).
is no longer a spectral property in the infinite The same result holds for a Poisson suspension
measure case. (see Grabinsky 1984; Marchat 1978).
Ergodicity and mixing of Gaussian systems
and Poisson suspensions. Ergodic and mixing Entropy, Bernoulli Properties
properties of Gaussian dynamical systems are The Kolmogorov-Sinai entropy of a Gaussian
easily characterized in terms of the spectral mea- system is given by the following result, already
sure s of the generating stationary Gaussian pro- stated by Pinsker in 1960 (Pinsker 1960) (see also
cess. Using the spectral characterizations of (Lemánczyk 1998; Newton 1971; de la Rue 1993)
ergodicity and mixing given in Theorem 5.1 and for various proofs).
the expression of the maximal spectral type of the
second quantization of a unitary operator in Prop- Theorem 5.5 (Entropy of a standard Gaussian
osition 4.1, we can obtain the following results system) Let s be a symmetric positive finite mea-
which were first proved by Maruyama (1949) and sure on  . If s is singular with respect to the
Fomin (1950). Details can also be found in Lebesgue measure, then h(Xs) ¼ 0. Otherwise,
Cornfeld et al. (1982). h(Xs) ¼ 1.

Theorem 5.3 The Gaussian dynamical system In the case where its spectral measure is the
Xs is ergodic if and only if the spectral measure Lebesgue measure l on , the stationary Gaussian
s is continuous, that is s({t}) ¼ 0 for each t  . process (xn)n  ℤ is composed of orthogonal, hence
In this case it is also weakly mixing. independent, random variables. Thus Xl is a
The system Xs is (strongly) mixing if and only if Bernoulli shift of infinite entropy. Now if s l,
s is Rajchman. we will see in Theorem 6.1 that Xs is a factor of a
Bernoulli shift. By Ornstein theory (Ornstein
In particular, by the Riemann-Lebesgue 1974), Xs is itself isomorphic to a Bernoulli
Lemma, if s is absolutely continuous, then the shift, and since it also has infinite entropy by
associated Gaussian system is mixing, which Theorem 5.5, it is in fact isomorphic to Xl.
was first proved by Ito (1944). But there also In the general case, decomposing the spectral
exist singular spectral measures whose Fourier measure as the sum of its singular and absolutely
coefficients vanish at infinity (Menchoff 1916). continuous parts and assuming that both are non-
The same kinds of arguments apply to Poisson zero, we get by Theorem 6.1 below that Xs is the
suspensions: together with Theorem 5.2, this direct product of a Bernoulli shift of infinite
224 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

entropy with a zero-entropy Gaussian system mð½i0 i1 , . . ., ik Þ ≔qi0 pi0 ,i1 Pikl , ik :
(corresponding to the singular part of the spectral
measure). In particular, any standard Gaussian sys- It has been proved by Kalikow (1981) and
tem satisfies the Pinsker property (it is a product of Grabinsky (1984) that the associated Poisson sus-
a zero-entropy system with a Bernoulli shift). pension ðX , A  , m , T  Þ is Bernoulli. Moreover,
As far as Poisson suspensions are concerned, as shown in Janvresse et al. (2010), its entropy is
anything can happen with the entropy which can given by the following formula generalizing the
be 0, positive finite or infinite. Together with the entropy for positive-recurrent Markov chains:
fact that, when the base ðX, A, m, T Þ is a prob-
ability space, the entropy of the Poisson suspen- hð T  Þ ¼  qi pi,j log pi,j :
sion T is just the same as the entropy of T, this iS jS
leads to view the entropy of the Poisson suspen-
sion T as a possible way to define the entropy of (Examples where the above entropy is finite
T when T is an infinite measure-preserving trans- are provided by Krengel (1967).)
formation, as proposed in Roy (2005). Other ways The general structure of a Poisson suspension
to define the entropy of an infinite measure- is, in general, not known. We can ask in particular
preserving transformation, also generalizing the whether there exists a Poisson suspension which
finite-measure case, had been proposed by is K but not Bernoulli.
Krengel (1967) and Parry (1969), and it is proved
in Janvresse et al. (2010) that these three notions
of entropy coincide in many cases. However, an
example of a transformation with zero Krengel
Joinings, Factors, and Centralizer
entropy, but whose Poisson suspension has posi-
Gaussian Factors and Centralizer
tive entropy, is presented in Janvresse and De La
Gaussian factors of a Gaussian system. In the
Rue (2012).
standard Gaussian system Xs generated by the
Assume that there exists in the base system
Gaussian process (xn)n  ℤ, let us take a nonzero
ðX, A, m, T Þ a wandering set A (i.e., whose
element z  H r1, and set for each n  ℤ, zn ≔z∘Sn  Hr1 :
images T nA, n  ℤ, are pairwise disjoint), and
Then (ζn)n  ℤ is a stationary Gaussian process, and
such that X ¼ [n  ℤT nA (the system is called
generates a particular factor sigma-algebra of Xs
dissipative in this case). Then, obviously, the
which we call a Gaussian factor. This factor can
base is of zero type and the Poisson suspension
be more precisely described by the following
is mixing. But thanks to the independence prop-
analysis.
erty of the Poisson process, the Poisson suspen-
Spectral theory provides an isometry between
sion is in fact Bernoulli. However, there also exist
the closed real subspace Hr1 of L2(ms) generated by
Bernoulli examples of Poisson suspensions where
the base is conservative (i.e., there is no wander- the Gaussian process (xn) and the following real
ing set of positive measure). Such examples are subspace of L2 ð, sÞ:
provided by Poisson suspensions over null-
L2sym ð, sÞ
recurrent Markov chains: we consider a countable
set S, an irreducible null-recurrent stochastic ≔ f  L2 ð, sÞ : fðtÞ ¼ fðtÞ for s  almost every t :
matrix P ¼ (pi, j)i, j  S, and let q ¼ (qi)i  S be a
nonzero measure which is stationary with respect The element ζ of H r1 corresponds to some
to P. We can then form the associated Markov function f  L2sym ð, sÞ, and the spectral measure
shift ðX, A, m, T Þ where X ¼ Sℤ, T is the s1 of the process (ζn)n  ℤ is absolutely continuous
shift transformation, and m is the shift invariant 2
with respect to s, with density dsds ¼ jfj .
1

infinite measure given on cylinder sets by the Observe conversely that any symmetric posi-
formula tive finite measure s1 s can be realized in this
Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems 225

way, for example, by taking the real-valued func- always rich. First, if in Formula (3) we replace
tion f≔ ds1
 L2sym ð, sÞ. ei2πθ(s) by ei2πuθ(s) for u  ℝ, we get a measure-
ds
The Gaussian factor generated by (ζn)n  ℤ preserving ℝ action T ug . Hence any stan-
uℝ
only depends on the equivalence class of s1, so dard Gaussian dynamical system can be embed-
we can denote it by F s1. The action of S on F s1 is ded in an ℝ-action. Moreover each T ug gives rise to
isomorphic to X s1 . a generalized Gaussian system (as the Gaussian
If s1 is equivalent to s, then F s1 coincides with subspace Hr1 is stable by T ug ).
the Borel sigma-algebra of Xs, and X s1 is isomor- Observe also that if we take two probability
phic to Xs. On the other hand, if s1 is not equiv- measures γ1 and γ2 on , then T g1 and T g2 always
alent to s, we can find another symmetric positive commute. Hence the centralizer of a Gaussian
finite measure s2 on  such that dynamical system always contains transforma-
tions isomorphic to any other standard Gaussian
• s2 is singular with respect to s1. dynamical system.
• s is equivalent to s1 + s2. The above examples all belong to the so-called
Gaussian centralizer, whose elements are
Using the fact that orthogonal subspaces of H r1 constructed as follows. We start by defining the
correspond to independent families of Gaussian multiplicative group
random variables, we then get another Gaussian
factor F s2 which is independent of F s1 , and such Gs ≔ c  L2sym ð, sÞ :jcj ¼ 1ðs  a:e:Þ :
that F s1 and F s2 together generate the same
sigma-algebra as the original Gaussian process
For each c in Gs, the multiplication by c in
(xn). This shows that any strict Gaussian factor
L2sym ð, sÞ is a unitary operator, which corre-
F s1 of Xs admits an independent complement. In
the case when Xs is ergodic, this independent sponds in H r1 to a unitary operator Uc. Moreover,
complement is itself weakly mixing, therefore the process (Uc(xn))n  ℤ has the same distribution
the corresponding extension X s ! X s1 is rela- as the generating Gaussian process (xn)n  ℤ,
tively weakly mixing. hence Uc comes from a measure-preserving trans-
We can summarize some of the above results formation, which we denote here by Tc, commut-
through the following theorem. ing with S on Xs.
Note that Uc can be naturally extended to a
Theorem 6.1 Let s, s1, s2 be symmetric posi- unitary operator on the complex space H1. The
tive finite measures on . Koopman operator associated to Tc is then noth-
ing but the second quantization of Uc.
• If s1 and s2 are equivalent, then X s1 and X s2 The set of transformations
are isomorphic.
• If s1 s, then X s1 is a factor of Xs and Cg ðSÞ≔ T c : c  Gs
corresponds to a Gaussian factor F s1 of Xs.
• If s is equivalent to s1 + s2, where s1 and s2 is a subgroup of the centralizer of S which is called
are mutually singular, then Xs is isomorphic to the Gaussian centralizer.
the direct product X s1  X s2 . Compact and classical factors of a Gaussian
dynamical system. Newton and Parry (1966) (see
also Maruyama (1967)) introduced another type
Gaussian centralizer. The geometric interpre- of factor of the Gaussian system Xs by consider-
ing the sigma-algebra of subsets of ℝℤ which
tation of a standard Gaussian dynamical system as
a transformation of the Brownian motion path are invariant by the transformation (xn)n  ℤ 7!
given in section “Gaussian Systems” allows to (xn)n  ℤ (called the even factor). We can see that
this factor is not a Gaussian factor, as the
show that the centralizer of a Gaussian system is
226 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

corresponding extension is never relatively weakly   

Y  , AjY , mjY , T jY and
mixing.
  
This example belongs to another class of fac- ðX nY Þ , AjX nY , mjX nY , T jX nY :
tors of Xs, which are obtained in the following
way. Take any compact subgroup K of Gs, which
The other way to get a Poisson suspension as a
corresponds to some compact subgroup CK of
factor of ðX , A  , m , T  Þ is to consider a
Cg(S), and consider
sigma-finite factor of the base system C  A ,
that is, a T-invariant sigma-algebra where m is
F K ≔ A  ℝℤ : TA ¼ A for each T  CK : still sigma-finite. We then define

Then F K is a factor sigma-algebra of Xs which C  ≔sfN A , A  C g:

is called a Gaussian-compact factor of Xs. Note
that Newton-Parry’s example corresponds to the We can check that the action of T on C  can be
subgroup K generated by the function identically viewed as a Poisson suspension. Indeed, we have
equal to 1 on . the isomorphism
Finally, these two types of factors can be com-
bined, and we call a classical factor of Xs any    
Gaussian-compact factor of a Gaussian factor of X=C , A =C , m=C , T =C
Xs.
’ X=C  , A =C  , m=C  , T =C  :

Poisson Factors and Centralizer In general, we call Poisson factor of the

There are two main ways to get natural factors of a suspension ðX , A  , m , T  Þ any T-invariant
Poisson suspension of base ðX, A, m, T Þ which 
sub-sigma-algebra of the form C jY , where C
are themselves Poisson suspensions. The first one is a sigma-finite factor of the base system and
is to consider a nontrivial T invariant set Y  X and Y a C -measurable T-invariant subset of X.
the sigma-algebra generated by restriction of the The centralizer of T always contains other
Poisson process to Y: Poisson suspensions, namely all transformations
of the form S when S  C(T). They form a
A jY ≔sfN A , A  A, A  Y g: subgroup of C(T): the Poisson centralizer.

The associated factor map is o 7! ojY and

sends ðX , A  , m , T  Þ to Gaussian and Poisson Self-Joinings
This section is based on notions borrowed from
   Lemańczyk et al. (2000) (for Gaussian) and Roy
Y, A jY , mjY , T jY : (2009), Derriennic et al. (2008) (for Poisson). The
inherent probabilistic nature of Gaussian and
As an immediate consequence of the indepen- Poisson systems allows to define remarkable fam-
dence properties of a Poisson process, we get the ilies of self-joinings.
following result. In the Gaussian setting, the definition is pretty
straightforward.
Proposition 6.2 Assume there exists a non trivial
T-invariant set Y  X. Then the Poisson suspen- Definition 6.3 Let n be a self-joining of a stan-
sion ðX , A  , m , T  Þ is isomorphic to the direct dard Gaussian system Xs. In the probability-
product of the two Poisson suspensions preserving dynamical system
Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems 227

ℝℤ  ℝℤ , B ℝℤ  B ℝℤ , n, S  S , ðX , A  , ðm  m1 Þ , T  Þ,
ðX  X , A   A  , m, T   T  Þ,
the two natural projections on ℝℤ provide two and
copies
ðX , A  , ðm  m2 Þ , T  Þ:
x0n ¼ x00 ∘ ðS  SÞn nℤ
and x00n On X  X, we define m as the pushforward
¼ x000 ∘ ðS  SÞn measure of ðm  m1 Þ  m  ðm  m2 Þ by the
nℤ
factor map
of the original Gaussian process. The self-joining
0 0
n is said to be a Gaussian self-joining if the real o0 , ðo1 , o2 Þ, o0 7! o0 þ o1 , o2 þ o0 :
subspace of L2(n) spanned by the x0n and the
x00n , n  ℤ, is a Gaussian space. We get a self-joining ðX  X , A   A  , m,
T   T  Þ of the original Poisson suspension.
Example 6.4 The product self-joinings, and
more generally the relatively independent prod- Definition 6.5 We call Poisson self-joining of a
uct over a Gaussian factor, are Gaussian self- Poisson suspension any self-joining which is
joinings. The graph self-joining associated to a obtained as explained above.
transformation in the Gaussian centralizer is
also a Gaussian self-joining. The ergodic com- Example 6.6 The product self-joining corre-
ponents of the relatively independent product sponds to the case where m is the null measure.
over a compact factor are Gaussian self- If S is in the centralizer of T and m ≔ Δs is the
joinings. corresponding graph measure, then m ¼ DS ,
i.e., the graph self-joining associated to S.
For the Poisson counterpart, the aforemen- If C  A is a sigma-finite factor and m≔mC m is
tioned family of self-joinings is easier to introduce the relatively independent joining over C , then
by expliciting their structure. m ¼ m C  m , i.e., the relatively independent
Let ðX , A  , m , T  Þ be a Poisson suspen- joining over the Poisson factor C .
sion. Start with a “sub-self-joining” of the base,
namely a system ðX  X, A  A, m, T  T Þ Despite the different nature of their structure, it
where m is a T  T-invariant measure whose pro- is possible to give a unified characterization of
jections satisfy Poisson and Gaussian self-joinings (see Roy 2005).

m1 ≔mð  XÞ m and m2 ≔mðX  Þ m: Proposition 6.7 A self-joining of a Gaussian sys-

tem (resp. a Poisson suspension) is Gaussian
We form the Poisson suspension (resp. Poisson) if and only if its distribution is
ðX  XÞ , ðA  A Þ , m , ðT  T Þ . Consider infinitely divisible.
the map k 7! (o1, o2) ≔ (k(  X), k (X  )), In the Gaussian case, this follows immediately
and denote by m the pushforward measure of m from the fact that a process with Gaussian entries
by this map. Then we get as a factor is Gaussian if and only if it is infinitely divisible.

ðX  X , A   A  , m, T   T  Þ, Theorem 6.8 (Derriennic et al. 2008;

Lemańczyk et al. 2000) A self-joining of a
which is nothing else than a joining between Gaussian system Xs is Gaussian if and only if
X , A  , m1 , T  and X , A  , m2 , T  . the associated Markov operator F acting on
The final step consists into “fixing the intensi- L2(ms) is the second quantization of an operator
ties” by considering the direct product of the three ’ on L2(s) of norm less than or equal to one and
systems which commutes with V (see (4)).
228 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

Similarly, a self-joining of a Poisson suspen- (which means that the spectral measure s satisfies
sion ðX , A  , m , T  Þ is Poisson if and only if the same assumptions as in Theorem 7.1). Let n be
the associated Markov operator F acting on a self-joining of Xs, and denote by x0 and x00 the
L2(m) is the second quantization of an operator two copies of the original Gaussian process with
’ on L2(m) corresponding to a sub-self-joining of spectral measure s in the probability-preserving
ðX, A, m, T Þ (a sub-Markov operator commut- dynamical system
ing with T).
ℝℤ  ℝℤ , B ℝℤ  ℝℤ , n, S  S ,
The following property is central in this theory
(see (Lemańczyk et al. 2000) for Gaussian, (Roy (as in Definition 6.3). In L2(n), let H0 and H00 be
2005) and (Derriennic et al. 2008) for Poisson). the real subspaces spanned, respectively, by x0 and
x00, and H ≔ H0 + H00. For any h  H, the spectral
Proposition 6.9 A Gaussian (resp. Poisson) self- measure of the stationary process (h ∘ (S  S)n)n  ℤ
joining of an ergodic Gaussian system (resp. is absolutely continuous with respect to s. Hence,
Poisson suspension) is ergodic. if we further assume that n is ergodic, Theorem 7.1
implies that H is a Gaussian subspace.
We formalize the above by the following def-
GAGs and PAPs inition and theorem.

From Foias-Stratila to GAGs Definition 7.2 The Gaussian system Xs is said to

Lemańczyk, Parreau, and Thouvenot introduced be a GAG when any ergodic self-joining of Xs is a
in Lemańczyk et al. (2000) a class of Gaussian Gaussian self-joining.
dynamical systems called GAGs (a French acro-
nym meaning “Gaussian systems with Gaussian Note that in this case, the system given by any
self-joinings”), on which they have developed a ergodic self-joining is a generalized Gaussian
beautiful theory. The proof of the existence of system.
GAGs relies on a striking rigidity result by Foias
and Stratila (1967), according to which there are Theorem 7.3 Any Gaussian-Kronecker system is
some spectral measures that, in ergodic systems, a GAG.
can only be realized by Gaussian processes (this
result improved a previous work of Sinai (1963)). Even if Foias-Stratila result is the keystone for
Foias-Stratila result can be formulated in the fol- the construction of GAGs, it is shown in
lowing way (see also (Cornfeld et al. 1982, Lemańczyk et al. (2000) that the class of GAGs
p. 375)). extends far beyond the Gaussian-Kronecker sys-
tems. Indeed, it is proved that any simple-
Theorem 7.1 (Foias-Stratila) Let ðO, F , ℙ, T Þ spectrum Gaussian system is GAG, and in partic-
be an ergodic probability-preserving dynamical ular there exist mixing GAGs.
system. Assume that for some nonzero real-valued Because of their special properties (see sec-
x0  L2(ℙ), the spectral measure of the stationary tion “Properties of GAGs and PAPs”), GAGs can
process (x0 ∘ T n)n  ℤ is continuous and concen- be used to produce very interesting examples of
trated on K [ (K), where K is a Kronecker probability-preserving dynamical systems. But
subset of  . Then the process (x0 ∘ T n)n  ℤ is beside the theory of GAGs, Foias-Stratila prop-
Gaussian. erty can lead to other surprising results. For
example, it is used in de la Rue (1998a) to show
The above result is used in Lemańczyk et al. the existence of some spectral measures on the
(2000) through the following argument (which torus that can never be obtained in a transforma-
was already stated by Thouvenot (1987, 1996). tion induced by an irrational rotation of the
Let us consider a Gaussian-Kronecker system Xs circle.
Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems 229

Poissonian Analog of Foias-Stratila Theorem

and PAPs mðA1   An Þ ¼ m \ T ki Ai ,
iP
Pp
Roughly speaking, an analog of Foias-Stratila
Theorem in the context of Poisson suspensions
where π is a partition of {1, . . ., n} and {ki} are
should identify a special class of infinite
integers.
measure-preserving dynamical systems
ðX, A, m, T Þ playing a role similar to the
Observe that such an ergodic component
Kronecker measures, such that any point process
gives rise to a system isomorphic to
invariant and ergodic under the dynamics of
X6¼p , A #p , m#p , T #p :
T would have a Poissonian character.
This above property is reminiscent to the con-
This invariance requirement is encompassed in
cept of minimal self-joinings in the probability-
the following notion of T-point process.
preserving case. However, we stress that, when
the measure is infinite, greater care is required
Definition 7.4 Let ðX, A, m, T Þ be a sigma-
when dealing with invariant measures on Carte-
finite measure-preserving system, and let
sian products (see (Janvresse et al. 2017a) where
ðO, , ℙ, SÞ be a probability-preserving sys-
this definition is discussed, and the construction of
tem. A point process N : Ω ! X is said to be a
an example of transformation in the family (FS) is
T-point process if
given).
We can now state the Poissonian version of
• for any A  A : N ðAÞ∘S ¼ N T 1 A
Theorem 7.1, proved in Janvresse et al. (2017b).
• N(A) ¼ 0 ℙ - a. s. ifm(A) ¼ 0
Theorem 7.5 (Poissonian Analog of Foias-
The T-point process N is T-free if its distribu- Stratila Theorem) Let N be a T-point process
tion is concentrated on the set of simple counting with moments of all orders defined on the ergodic
measures of the form i  I dxi such that for each probability-preserving dynamical system
i 6¼ j, xi and xj are not in the same T-orbit. ðO, F , ℙ, SÞ. If T is in the family (FS) and if
N is T-free, then N is a Poisson point process of
We assume now that X is a complete and sep- intensity αm for some α > 0.
arable metric space, and that m is an infinite and
boundedly finite measure on ðX, A Þ , that is, a By analogy with GAG, we can formulate the
measure giving finite mass to any bounded mea- following definition.
surable set. (This definition depends on the choice
of the metric, and our assumption implies in par- Definition 7.6 The Poisson suspension
ticular that the metric is unbounded.) ðX , A  , m , T  Þ is said to be a PAP when all
We consider the family (FS) of m-preserving its ergodic self-joinings are Poisson self-joinings.
invertible transformations T of X satisfying:
The analog of Theorem 7.3 in the context of
• For each n  1, ðXn , A n , mn , T n Þ is
Poisson suspensions has been established in
ergodic (T is said to have infinite ergodic
Janvresse et al. (2017b).
index).
• For each n  1, if n is a boundedly finite Tn- Theorem 7.7 If T is in the family (FS), then
invariant measure on ðXn , A n Þ ðX , A  , m , T  Þ is a PAP.
whose marginals are absolutely continuous
with respect to m, then ðXn , A n , n, T n Þ is Properties of GAGs and PAPs
conservative and any of its ergodic component GAGs and PAPs have the great advantage that we
m is of the form can quite well describe the structure of their self-
230 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

joinings. Since these self-joinings can tell a lot on Theorem 7.11 (Factors of a PAP (Janvresse
the factors and centralizer of the corresponding et al. 2017b)) Any nontrivial factor of the PAP
transformation, it is no surprise that we are able to ðX , A  , m , T  Þ contains a nontrivial Poisson
control the factors and centralizer of GAGs factor. Any factor over which T is relatively
and PAPs. weakly mixing is a Poisson factor.
We have seen in section “Gaussian Factors and
Centralizer” that in the centralizer of any Gaussian In particular, when T is in the family (FS), T
system we can find transformations isomorphic to has no nontrivial factor, as T itself has no non-
any other Gaussian system. It turns out that for trivial factor (Janvresse et al. 2018). Note that the
GAGs, there is no other element in the centralizer. primeness and the triviality of the centralizer of T
in this case are striking differences compared to
Theorem 7.8 (Centralizer of a GAG Gaussian systems, which always possess a lot of
(Lemańczyk et al. 2000)) Let Xs be a GAG, factors and a large centralizer. The dissemblance
and let T  C(S). Then for each ζ in the real can be pushed even further as it is proved (see
Gaussian subspace H r1 of L2(ms) spanned by the (Janvresse et al. 2017b) and (Janvresse et al.
generating Gaussian process, we have z∘T  H r1 . 2020)) that if T is in the family (FS), T is in fact
In particular the system ℝℤ , B ℝℤ , ms , T is disjoint from any standard Gaussian system.
a generalized Gaussian system.

We also identified in section “Gaussian Factors Future Directions

and Centralizer” a family of factors appearing in
any standard Gaussian system. Again, in the case A lot of questions remain open concerning the
of GAGs, there is no other possible factor. intersections of the classes of Gaussian systems
or Poisson suspensions with other families of
Theorem 7.9 (Factors of a GAG (Lemańczyk dynamical systems. In positive entropy, the situa-
et al. 2000)) Let Xs be a GAG. Any factor of Xs is tion is quite clear concerning the Gaussian case,
a classical factor. Any factor over which S is rel- but as already mentioned in section “Entropy,
atively weakly mixing is a Gaussian factor. Bernoulli Properties” it is not known whether we
can build a Poisson suspension which is K but not
The above theorem is used in Iwanik et al. Bernoulli. In the zero-entropy setting, some
(1997): in the case of a GAG system, the factor results have been obtained for Gaussian systems:
generated by a function of finitely many coordi- some are loosely Bernoulli, some are not (de la
nates of the Gaussian process can only be the Rue 1996), and we know that they cannot be of
whole system or the even factor (and the latter finite rank (de la Rue 1998b). Similar issues for
occurs only when the function is even). Poisson suspensions have not been studied yet.
Here are the corresponding results for Poisson A general problem that arises is the question of
suspensions. disjointness of these systems of probabilistic ori-
gin with other classes such as finite-rank systems.
Theorem 7.10 (Centralizer of a PAP (Janvresse One can hope for a unified answer taking advan-
et al. 2017b)) Let ðX , A  , m , T  Þ be a PAP, tage of the common features shared by Gaussian
and let R  C(T). Then there exists S  C(T) systems and Poisson suspensions, and inspired by
such that R ¼ S. ideas such as those developed in Derriennic et al.
(2008) or Lemańczyk et al. (2011), where it is
In particular, when T is in the family (FS), C(T) proved that both Gaussian systems and Poisson
is reduced to the powers of T(Janvresse et al. suspensions are disjoint from the class of distally
2018), thus CðT  Þ ¼ T n : n  ℤ . simple systems.
Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems 231

For Poisson suspensions, the entropy theory is de la Rue T (1995) Mouvement moyen et système
much richer and the situation where all three dynamique gaussien. Probab Theory Relat Fields
102(1):45–56. (French)
notions of entropy of a sigma-finite measure- de la Rue T (1996) Systèmes dynamiques gaussiens
preserving system do not coincide remains rather d’entropie nulle, lâchement et non lâchement Bernoulli.
mysterious. For example, is it true that the Poisson Ergodic Theory Dyn Syst 16(2):379–404. (French)
entropy always dominates the Krengel entropy of de la Rue T (1998a) L’ induction ne donne pas toutes les
mesures spectrales. Ergodic Theory Dyn Syst
the system? 18(6):1447–1466. (French)
It is noteworthy to mention that Poisson sus- de la Rue T (1998b) Rang des systèmes dynamiques
pensions can also be defined over non-singular gaussiens. Isr J Math 104:261–283. (French)
transformations, provided some integrability con- Derriennic Y, Fraczek K, Lemańczyk M, Parreau F (2008)
Ergodic automorphisms whose weak closure of off-
dition is satisfied. This is a potentially rich area to diagonal measures consists of ergodic self-joinings.
explore. (There is an ongoing work on this topic Colloq Math 110(1):81–115. (English)
by A. Danilenko, Z. Kosloff, and the second Dobrushin RL (1956) On the Poisson law for the distribu-
author of this entry.) tion of particles in space. Ukr Mat Zh 8:127–134.
(Russian)
Finally, we point out that this presentation Doob JL (1953) Stochastic processes. Wiley, New York.
focused on ℤ-actions, even though the notions of 654 S. 1953 (English)
Gaussian systems and Poisson suspensions Foias C, Stratila S (1967) Ensembles de Kronecker dans la
extend naturally to more general group actions. The’ orie ergodique. C R Acad Sci Paris 267:166–168
Fomin S (1950) On dynamical systems in a space of
For flows (ℝ-actions), Gaussian systems and functions. Ukr Mat Zh 2(2):25–47. (Russian)
Poisson suspensions provide interesting examples Fraczek K, Lemańczyk M (2009) On the self-similarity
of flows for which the self-similarity set problem for ergodic flows. Proc Lond Math Soc
99(3):658–696. (English)
Fraczek K, Kulaga J, Lemańczyk M (2013) On the self-
I ðT Þ≔ s  ℝ : ðT st Þt  ℝ is isomorphic to ðT t Þt  ℝ similarity problem for Gaussian-Kronecker flows. Proc
Am Math Soc 141(12):4275–4291. (English)
can be fully described (see (Danilenko and Girsanov IV (1958) Spectra of dynamical systems gener-
ated by stationary Gaussian processes. Dokl Akad
Ryzhikov 2012; Fraczek and Lemańczyk 2009;
Nauk SSSR 119:851–853. (Russian)
Fraczek et al. 2013)). For other groups, the gener- Goldstein S, Lebowitz JL, Aizenman M (1975) Ergodic
alization of some topics presented in the present properties of infinite systems. Dyn Syst Theory Appl
survey is not always obvious. The entropy of Battelle Seattle 1974 Renc, Lect Notes Phys 38:112–143
Gaussian actions of Abelian groups is proved to Grabinsky G (1984) Poisson process over s -finite Markov
chains. Pac J Math 111:301–315. (English)
be 0 or 1 in Lemánczyk (1998); however, it is Ito K (1944) On the ergodicity of a certain stationary
unknown if the same holds for countable amena- process. Proc Imp Acad Tokyo 20:54–55. (English)
ble groups. We can also ask for which group Iwanik A, Lemańczyk M, de la Rue T, de Sam Lazaro
J (1997) Quelques remarques sur les facteurs des
actions we have a Foias-Stratila theory.
systèmes dynamiques gaussiens. Stud Math
125(3):247–254. (French)
Janvresse É, De La Rue T (2012) Zero Krengel entropy
Bibliography does not kill Poisson entropy. Ann Inst Henri Poincaré,
Probab Stat 48(2):368–376. (English)
Attal S. Lectures in quantum noise theory. http://math. Janvresse É, Meyerovitch T, Roy E, de la Rue T (2010)
univ-lyon1.fr/attal/chapters.html Poisson suspensions and entropy for infinite transfor-
Cornfeld IP, Fomin SV, Sinai YG (1982) Ergodic theory. mations. Trans Am Math Soc 362(6):3069–3094.
Grundlehren der Mathematischen Wissenschaften (English)
[Fundamental principles of mathematical sciences], Janvresse É, Roy E, de la Rue T (2017a) Nearly finite
vol 245. Springer, New York chacon transformation, hal-01586869. Annales Henri
Danilenko AI, Ryzhikov VV (2012) On self-similarities of Lebesgue 2(2019):369–414
ergodic flows. Proc Lond Math Soc 104(3):431–454. Janvresse É, Roy E, de la Rue T (2017b) Poisson suspensions
(English) and SuShis. Ann Scient Éc Norm Sup 50(6):1301–1334
de la Rue T (1993) Entropie d’ un système dynamique Janvresse É, Roy E, de la Rue T (2018) Invariant measures
gaussien: cas d’ une action de ℤd. C R Acad Sci Paris for Cartesian powers of Chacon infinite transformation.
Sér I 317(2):191–194. (French) Israel J Math 224:1–37
232 Dynamical Systems of Probabilistic Origin: Gaussian and Poisson Systems

Janvresse É, Roy E, de la Rue T (2020) Ergodic poisson Newton D (1966) On Gaussian processes with simple
splittings, preprint. Ann Probab 48(3):1266–1285 spectrum. Z Wahrscheinlichkeitstheor Verw Geb
Kalikow S (1981) A Poisson random walk is Bernoulli. 5:207–209. (English)
Commun Math Phys 81:495–499. (English) Newton D (1971) Coalescence and spectrum of automor-
Krengel U (1967) Entropy of conservative transforma- phisms of a Lebesgue space. Z Wahrschein-
tions. Z Wahrscheinlichkeitstheorie Verw Gebiete lichkeitstheor Verw Geb 19:117–122. (English)
7:161–181. MR MR0218522 (36 #1608) Newton D, Parry W (1966) On a factor automorphism of a
Last G, Penrose M (2018) Lectures on the Poisson process, normal dynamical system. Ann Math Stat
vol 7. Cambridge University Press, Cambridge. 37:1528–1533. (English)
(English) Ornstein DS (1974) Ergodic theory, randomness, and
Lemánczyk M (1998) Entropy of Gaussian actions for dynamical systems. Yale University Press, New
countable Abelian groups. Fundam Math Haven. (English)
157(2–3):277–286. (English) Parry W (1969) Entropy and generators in ergodic theory.
Lemańczyk M (2011) Spectral theory of dynamical sys- W. A. Benjamin, Inc., New York-Amsterdam. MR
tems. Springer New York, New York, pp 1618–1638 MR0262464 (41 #7071)
Lemańczyk M, Parreau F, Thouvenot J-P (2000) Gaussian Pinsker MS (1960) Dynamical systems with completely
automorphisms whose ergodic self-joinings are Gauss- positive or zero entropy. Sov Math Dokl 1:937–938.
ian. Fundam Math 164(3):253–293 (English)
Lemańczyk M, Parreau F, Roy E (2011) Joining primeness Roy E (2005) Mesures de poisson, infinite divisibilité et
and disjointness from infinitely divisible systems. Proc propriétés ergodiques. PhD thesis
Am Math Soc 139(1):185–199. (English) Roy E (2009) Poisson suspensions and infinite ergodic
Leonov VP (1960) The use of the characteristic functional theory. Ergodic Theory Dyn Syst 29(2):667–683
and semi-invariants in the ergodic theory of stationary Sinai YG (1963) On higher order spectral measures of
processes. Sov Math Dokl 1:878–881. (English) ergodic stationary processes. Theory Probab Appl
Marchat FA (1978) A class of measure-preserving trans- 8:429–436. (English)
formations arising by the Poisson process, Ph. D. The- Thouvenot J-P (1987) The metrical structure of some
sis, Berkeley, Dec 1978 Gaussian processes, Ergodic theory and related
Maruyama G (1949) The harmonic analysis of stationary topics II. In: Proceedings of the conference on
stochastic processes. Mem Fac Sci Kyūsyū Univ Ser Georgenthal/GDR 1986, Teubner-Texte Math
A 4:45–106 vol 94, pp 195–198
Maruyama G (1967) A singular flow with countable Thouvenot J-P (1996) Utilisation des processus gaussiens
Lebesgue spectrum. J Math Soc Japan 19:359–365. en théorie ergodique., Hommage à P. A. Meyer et
(English) J. Neveu. Société Mathématique de France, Paris,
Maruyama G (1970) Infinitely divisible processes. Theory pp 303–308. (French)
Probab Appl 15(1):1–22 Totoki H (1964) The mixing property of Gaussian flows.
Menchoff D (1916) Sur l’ unicité du développement tri- Mem Fac Sci Kyushu Univ Ser A 18:136–139.
gonométrique. C R Acad Sci Paris 163:433–436. (French) (English)
Neretin YA (1996) Categories of symmetries and infinite- Vershik AM, Gel’fand IM, Graev MI (1975) Representa-
dimensional groups. Transl. from the Russian by G. G. tions of the group of diffeomorphisms. Russ Math Surv
Gould. Clarendon Press, Oxford. (English) 30(6):1–50. (English)
 Equivalently, every Borel function f : X ! ℝ
Ergodic Theory: Nonsingular such that f ∘ T = f is constant a.e.
Transformations Nonsingular dynamical system Let (X, ℬ, m)
be a standard Borel space equipped with a s-
Alexandre I. Danilenko1 and Cesar E. Silva2 finite measure. A Borel map T : X ! X is a
1
B.Verkin Institute for Low Temperature nonsingular transformation of X if for any
Physics and Engineering of the NAS of Ukraine, N  ℬ, m(T 1N) = 0 if and only if m(N) = 0.
Kharkiv, Ukraine In this case, the measure m is called quasi-
2
Department of Mathematics, Williams College, invariant for T, and the quadruple
Williamstown, MA, USA (X, ℬ, m, T) is called a nonsingular dynamical
system. If m(A) = m(T1A) for all A  ℬ, then m
is said to be invariant under T or, equivalently,
Article Outline T is measure-preserving.
Types II, II1, II1, and III Suppose that m is
Glossary nonatomic and T is invertible and ergodic
Definition of the Subject (and hence conservative). If there exists a s-
Introduction and Basic Results finite measure n on ℬ which is equivalent to m
Panorama of Examples and invariant under T then T is said to be of type
Topological Groups ΑUΤ(X, m), II. It is easy to see that n is unique up to scaling.
ΑUΤ2(X, m) and AUT1(X, m) If n is finite, then T is of type II1. If n is infinite,
Orbit Theory then T is of type II1. If T is not of type II, then T
Mixing Notions and Multiple Recurrence is said to be of type III.
Dynamical Properties of IDPFT Systems
Dynamical Properties of Nonsingular Bernoulli
and Markov Shifts Definition of the Subject
Dynamical Properties of Nonsingular Poisson
Suspensions and Nonsingular Gaussian An abstract measurable dynamical system con-
Transformations sists of a set X (phase space) with a transforma-
Spectral Theory for Nonsingular Systems tion T : X ! X (evolution law or time) and a finite
Entropy and Other Invariants or s-finite measure m on X that specifies a class of
Nonsingular Joinings and Factors negligible subsets. Nonsingular ergodic theory
Smooth Nonsingular Transformations studies systems where T respects m in a weak
Miscellaneous Topics sense: The transformation preserves only the
Applications. Connections with Other Fields class of negligible subsets, but it may not pre-
Further Directions serve m. This survey is about dynamics and
Bibliography invariants of nonsingular systems. Such systems
model “non-equilibrium” situations in which
Glossary events that are impossible at some time remain
impossible at any other time. Of course, the first
Conservativity T is conservative if for all sets A question that arises is whether it is possible to
of positive measure there exists an integer find an equivalent invariant measure, i.e., pass to
n > 0 such that m(A \ TnA) > 0. a hidden equilibrium without changing the neg-
Ergodicity T is ergodic if every measurable sub- ligible subsets. It turns out that there exist sys-
set A of X that is invariant under T (i.e., tems which do not admit an equivalent invariant
T1A = A) is either m-null or m-conull. finite or even s-finite measure. They are of our
© Springer Science+Business Media, LLC, part of Springer Nature 2023 233
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_183
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2020
https://doi.org/10.1007/978-3-642-27737-5_183-2
234 Ergodic Theory: Nonsingular Transformations

primary interest here. In a way (Baire category), this survey. Many new results related to non-
most of systems are like that. singular dynamical systems have appeared since
Nonsingular dynamical systems arise natu- the release of the first edition. The second edition
rally in various fields of mathematics: topologi- is enlarged essentially to cover (partially) this
cal and smooth dynamics, probability theory, progress. In particular, we added new Sections
random walks, theory of numbers, von Neumann 7 and 9 and totally rewrote Section 8. More than
algebras, unitary representations of groups, 100 new references have been added. We are
mathematical physics, and so on. They also can grateful to J. Aaronson, N. Avraham-Re’em,
appear in the study of probability preserving J. Hawkins, and Z. Kosloff for their remarks to
systems: some criteria of mild mixing and the second edition of the survey.
distality, a problem of Furstenberg on
disjointness, etc. We briefly discuss this in §15.
Nonsingular ergodic theory studies all of them Introduction and Basic Results
from a general point of view:
This section includes the basic results involving
• What is the qualitative nature of the dynamics? conservativity and ergodicity as well as some
• What are the orbits? direct nonsingular counterparts of the basic
• Which properties are typical within a class of machinery from classic ergodic theory: mean
systems? and pointwise ergodic theorems, Rokhlin lemma,
• How do we find computable invariants to com- ergodic decomposition, generators, Glimm-Effros
pare or distinguish various systems? theorem, and special representation of non-
singular flows. The historically first example of a
Typically there are two kinds of results: Some transformation of type III (due to Ornstein) is also
are extensions to nonsingular systems of theorems given here with full proof.
for finite measure-preserving transformations (for
instance, §2, § 4, § 12), and the other are about Nonsingular Transformations
new properly “nonsingular” phenomena (see §5 In this survey, we will consider mainly invertible
§ 9). Philosophically speaking, the dynamics of nonsingular transformations, i.e., those which are
nonsingular systems is more diverse compara- bijections when restricted to an invariant Borel
tively with their finite measure-preserving coun- subset of full measure. Thus when we refer to a
terparts. That is why it is usually easier to nonsingular dynamical system (X, ℬ, m, T ), we
construct counterexamples than to develop a gen- shall assume that T is an invertible nonsingular
eral theory. While infinite measure-preserving transformation (unless the contrary is specified
transformations are not the main subject of this explicitly). Of course, each measure n on ℬ
survey, we cover them partially as they are also which is equivalent to m, i.e., m and n have the
nonsingular systems and arise often as natural same null sets, is also quasi-invariant under T.
examples or counterexamples in the nonsingular In particular, since m is s-finite, T admits an equiv-
setting. Because of shortage of space, we concen- alent quasi-invariant probability measure. For
trate mainly on invertible transformations, and we each i  ℤ, we denote by omi or oi the Radon-
have not included as many references as we had Nikodym derivative d(m ∘ T i)/dm  L1(X, m). The
wished. General group or semigroup actions are derivatives satisfy the cocycle equation oiþj(x) ¼
practically not considered here (with some oi(x)oj(T ix) for a.e. x and all i, j  ℤ.
exceptions in §15 devoted to applications).
A number of open problems are scattered through Basic Properties of Conservativity and
the entire text. Ergodicity
We thank J. Aaronson, J. R. Choksi, V. Ya. A measurable set W is said to be wandering if for
Golodets, M. Lemańczyk, F. Parreau, and all i, j  0 with i 6¼ j, T iW \ T jW ¼ ;. Clearly, if
E. Roy for useful remarks to the first edition of T has a wandering set of positive measure, then it
Ergodic Theory: Nonsingular Transformations 235

cannot be conservative. A nonsingular transfor- 1 on the group ℤ furnished with the counting
mation T is incompressible if whenever T1C  C, measure is an example of an ergodic non-
then m(C \T 1C) ¼ 0. conservative (infinite measure-preserving)
transformation.
Proposition 2.1 (see Krengel 1985) Let
(X, ℬ, m, T ) be a nonsingular dynamical system.
Proposition 2.3 Let (X, ℬ, m, T ) be a non-
The following are equivalent:
singular dynamical system. The following are
equivalent:
(i) T is conservative.
(ii) For every measurable set
n
(i) T is conservative and ergodic.
A, m A∖[1 n¼1 T A ¼ 0.
(ii) For every set A of positive measure,
(iii) T is incompressible. n
m X∖[1 n¼1 T A ¼ 0. (In this case, we will
(iv) Every wandering set for T is null.
þ1 say A sweeps out.)
(v) i¼0 oi ðxÞ ¼ 1 at a.e. x (provided that (iii) For every measurable set A of positive mea-
m(X) < 1). sure and for a.e. x  X, there exists an
integer n > 0 such that T nx  A.
Since any finite measure-preserving transfor- (iv) For all sets A and B of positive measure,
mation is incompressible, we deduce that it is there exists an integer n > 0 such that
conservative. This is the statement of the classical m(T nA \ B) > 0.
Poincaré recurrence lemma. If T is a conservative (v) If A is such that T 1A  A, then m(A) ¼ 0 or
nonsingular transformation of (X, ℬ, m) and m(Ac) ¼ 0.
A  ℬ a subset of positive measure, we can define
an induced transformation TA of the space A set W of positive measure is said to be weakly
(A, ℬ \ A, m A) by setting TAx ≔ T nx if n ¼ n(x)
+

wandering if there is a sequence ni ! 1 such that

is the smallest natural number such that T nx  A. TA T ni W \ T nj W ¼ ; for all i 6¼ j. Clearly, a finite
is also conservative. As shown in Silva and measure-preserving transformation cannot have a
Thieullen (1991, 5.2), if m(X) ¼ 1 and T is conser- weakly wandering set. Hajian and Kakutani
nðxÞ1
vative and ergodic, A i¼0 oi ðxÞdmðxÞ ¼ 1, (1964) showed that a nonsingular transformation
which is a nonsingular version of the well-known T is of type II1 if and only if T does not have a
Kac’s formula. weakly wandering set. This entry is mainly about
systems of type III. For some time, it was not quite
Theorem 2.2 (Hopf Decomposition, see obvious whether such systems exist at all. The
Aaronson 1997). Let T be a nonsingular transfor- historically first example was constructed by
mation. Then there exist disjoint invariant sets C, Ornstein in 1960.
D  ℬ such that X ¼ C t D, T restricted to C is
conservative, and D ¼ t1 n
n¼1 T W, where W is a Example 2.4 (Ornstein 1960) Let An ¼
wandering set. If f  L (X, m), f > 0, then
1

þ1
{0, 1, . . ., n}, nn(0) ¼ 0.5 and nn(i) ¼ 1/(2n) for
C¼ x: i¼0
f T i x oi ðxÞ ¼ 1 a:e: and D ¼ 0 < i  n and all n  ℕ. Denote by (X, m) the infinite
x: þ1
i¼0 f T x oi ðxÞ < 1 a:e: .
i product probability space 1 n¼1 ðAn , nn Þ: Of course,
The set C is called the conservative part of T, m is nonatomic. A point of X is an infinite sequence
and D is called the dissipative part of T. If D is of x ¼ ðx n Þ1 n¼1 with xn  An for all n. Given a1  A1,
positive measure, we call T dissipative. If D is of . . ., an  An, we denote the cylinder
full measure, we call T totally dissipative. x ¼ ðxi Þ1 i¼1  X : x1 ¼ a1 , . . . , xn ¼ an g by
If T is ergodic and m is nonatomic, then T is [a1, . . ., an]. Define a Borel map T : X ! X by
automatically conservative. The translation by setting
236 Ergodic Theory: Nonsingular Transformations

0, if i < lðxÞ We note that UT preserves the cone of positive

ðTxÞi ¼ xi þ 1, if i ¼ lðxÞ ð1Þ functions L2þ ðX, mÞ: Conversely, every positive
unitary operator in L2(X, m) that preserves
xi , if i > lðxÞ,
L2þ ðX, mÞ equals UT for a m-nonsingular transfor-
mation T. We call UT the Koopman operator gen-
where l(x) is the smallest number l such that xl 6¼ l.
erated by T.
It is easy to verify that T is a nonsingular transfor-
mation of (X, m) and
Theorem 2.5 (von Neumann mean Ergodic The-
1 orem, see Aaronson 1997). T has no m-absolutely
nn ðTxÞn
om1 ðxÞ ¼ continuous T-invariant probability if and only if
nn ð x n Þ
n¼1 n1 n1i¼0 U T ! 0
i
in the strong operator
ðlðxÞ  1Þ!=lðxÞ, if xlðxÞ ¼ 0 topology.
¼ Proof. Let P denote the orthogonal projector in
ðlðxÞ  1Þ!, if xlðxÞ 6¼ 0:
L2(X, m) onto the subspace of UT -invariant vectors.
By the well-known fact from the theory of Hilbert
We prove that T is of type III by contradiction.
Suppose that there exists a T-invariant s-finite spaces, n1 n1 i¼0 U T ! P in the strong operator
i

measure n equivalent to m. Let ’ ≔ dm/dn. Then topology. Then P 6¼ 0 if and only if there is f  L2(-
X, m) such that f 6¼ 0 and UT f ¼ f. Of course,
1 UT j f j ¼ j f j. We now define a nontrivial finite
omi ðxÞ ¼ ’ðxÞ’ T i x for a:a: x  X and all i  ℤ: ð2Þ
measure l ≺ m by setting dm dl
≔j f j2 . It is straight-
Fix a real C > 1 such that the set forward to verify that l is invariant under T.
EC ≔ ’1([C1, C])  X is of positive measure. Denote by I the sub-s-algebra of T-invariant
By a standard approximation argument, for each sets. Let m ½j I  stand for the conditional expec-
sufficiently large n, there is a cylinder such that tation with respect to I . Note that if T is ergodic,
m(EC \ [a1, . . ., an]) > 0.9m([a1, . . ., an]). Since then m ½ f j I  ¼ fdm: Now we state a non-
nnþ1(0) ¼ 0.5, it follows that m(EC \ [a1, . . ., singular analogue of Birkhoff’s pointwise ergodic
an, 0]) > 0.8m([a1, . . ., an, 0]). Moreover, by the theorem, due to Hurewicz (1944) and in the form
pigeon hole principle there is 0 < i  n þ 1 with stated by Halmos (1946).
m(EC \ [a1, . . ., an, i]) > 0.8m([a1, . . ., an, i]).
Theorem 2.6 (Hurewicz pointwise Ergodic The-
Find Nn > 0 such that T Nn ½a1 , . . . , an , 0 ¼
orem). If T is conservative, m(X) ¼ 1, f, g  L1(X,
½a1 , . . . , an , i: Since omNn is constant on [a1, . . .,
m) and g > 0, then
an, 0], there is a subset E0  EC \ [a1, . . ., an, 0]
of positive measure such that n1
m
T Nn E0  EC \ ½a1 , . . . , an , i: Moreover, oN n ðxÞ ¼ f T i x oi ðxÞ
i¼0 m ½ f j I ðxÞ
! as
nnþ1 ðiÞ=nnþ1 ð0Þ ¼ ðn þ 1Þ1 for a.a. x  [a1, . . ., n1 m ½gj I ðxÞ
an, 0]. On the other hand, we deduce from (2) that g T i x oi ðxÞ
i¼0
omNn ðxÞ  C2 for all x  E0, a contradiction. n ! 1 for a:e: x:

Mean and Pointwise Ergodic Theorems. There is a nonsingular version (Silva and
Rokhlin Lemma Thieullen 1991) of the subadditive ergodic theo-
Let (X, ℬ, m, T ) be a nonsingular dynamical sys- rem of Kingman. Let T be a nonsingular transfor-
tem. Define a unitary operator UT of L2(X, m) by mation, and let (on) be its sequence of Radon-
setting Nikodym derivatives.
p A sequence of functions ( fn) is said to be
UT f ≔ o1  f ∘ T: ð3Þ subadditive if fnþm  fm þ fn ∘ T mom for all n,
Ergodic Theory: Nonsingular Transformations 237

m  0. Since ( fn) is a subadditive, one can verify Theorem 2.10 (Alpern’s lemma (Alpern and
that the following limit Prasad 1990)). Let T be an aperiodic nonsingular
transformation of a standard probability space
1 (X, m). Let π ¼ (π1, π2, . . .) be a probability vector
m ½ð f n ÞðxÞ≔n!1
lim m ½ f n j I ðxÞ
n such that {k| πk > 0} is a relatively prime set of
integers. Then there is a measurable partition
exists almost everywhere. P ¼ {Pk,i| k > 0, i ¼ 1, . . ., k} of X satisfying

Theorem 2.7 (Nonsingular subadditive Ergodic (a) TPk,i ¼ Pk, iþ1 for each k and every i < k.
Theorem). If T is conservative, m(X) ¼ 1, ( fn) is a k
(b) i¼1 mðPk,i Þ ¼ pk for each k.
subadditive sequence of integrable functions,
g  L1(X, m) and g > 0, then

f n ðxÞ m ½ð f n ÞðxÞ
! as Ergodic Decomposition
n1
i¼0 g T x oi ð x Þ
i m ½gj I ðxÞ
A proof of the following theorem may be found in
n ! 1 for a:e: x: Aaronson (1997, 2.2.8) and Aaronson (1987, §6).

Of course, Theorem 2.6 follows from Theorem Theorem 2.11 (Ergodic Decomposition Theo-
2.7 if we set f n ðxÞ≔ n1
i¼0 f T x oi ðxÞ for all n > 0
i
rem). Let T be a conservative nonsingular trans-
and x  X. formation on a standard probability space
A transformation T is aperiodic if the T-orbit of (X, ℬ, m). There exists a standard probability
a.e. point from X is infinite. The following classi- space ðY, n, A Þ and a family of probability mea-
cal statement can be deduced easily from Propo- sures my on (X, ℬ), for y  Y, such that
sition 2.1.
(i) For each A  ℬ, the map y 7! my(A) is Borel
Lemma 2.8 (Rokhlin’s lemma (Friedman 1970)). and for each A  ℬ
Let T be an aperiodic nonsingular transformation
of a standard probability space (X, m). For each
ε > 0 and integer N > 1, there exists a measurable mðAÞ ¼ my ðAÞdnðyÞ:
set A such that the sets A, TA, . . ., T N1A are
disjoint and m(A [ TA [    [ T N1A) > 1  ε.
(ii) For y, y0  Y, the measures my and my0 are
This lemma was refined later (for ergodic
mutually singular.
transformations) by Lehrer and Weiss as follows.
(iii) For each y  Y, the transformation T is
nonsingular and conservative, ergodic on
Theorem 2.9 (ϵ-free Rokhlin lemma (Lehrer and (X, ℬ, my).
m
Weiss 1982)). Let T be ergodic and m (iv) For each y  Y, o1 y ¼ om1 my  a:e:
nonatomic. Then for a subset B  X and any (v) (Uniqueness) If there exists another proba-
kN
N for which [1 k¼0 T ðX∖BÞ ¼ X, there is a set bility space ðY 0 , n0 , A 0 Þ and a family of prob-
A such that the sets A, TA, . . ., TN1A are disjoint ability measures m0y0 on (X, ℬ), for y0  Y0,
and A [ TA [    [ TN1A B. satisfying (i)–(iv), then there exists a
kN
The condition [1 k¼0 T ðX∖BÞ ¼ X holds of measure-preserving isomorphism θ : Y ! Y0
course for each B 6¼ X if T is totally ergodic, i.e., such that my ¼ m0yy for n-a.e. y.
T p is ergodic for any p, or if N is prime. We now
state a nonsingular version of Alpern’s lemma It follows that if T preserves an equivalent
which is a generalization of Lemma 2.8. s-finite measure, then the system (X, ℬ, my, T ) is
238 Ergodic Theory: Nonsingular Transformations

of type II for v-a.a. y. The space ðY, n, A Þ is called turns out that there is a wealth of such measures.
the space of T-ergodic components. To state a corresponding result, we first write an
important definition.
Generators
A s-algebra F is called a generator for a non- Definition 2.13 Two nonsingular systems
singular transformation T on a standard probabil- (X, ℬ, m, T ) and (X, ℬ0, m0, T0) are called orbit
ity space (X, ℬ, m), if _1 n¼1 T F ¼ ℬ . It was
n
equivalent if there is a one-to-one bimeasurable
shown in Rokhlin (1965) and Parry (1966) that map ’ : X ! X with m0 ∘ ’ m and such that f
T has a countable generator, i.e., a countable maps the T-orbit of x onto the T0-orbit of ’(x) for
partition P of X so that the s-algebra of a.a. x  X.
P-measurable sets is a generator for T. It was The following theorem was proved in
refined by Krengel (1970): If T is of type II1 or Katznelson and Weiss (1972), Krieger (1976a),
III, then there exists P consisting of two sets only. and Schmidt (1977b).
Moreover, given a sub-s-algebra F  ℬ such that
F  TF and [k>0TkF ¼ ℬ, the set {A  F j (A, Theorem 2.14 Let (X, T ) be as in Theorem 2.12.
X \ A) is a generator of T} is dense in F . It follows, Then for each ergodic dynamical system
in particular, that T is isomorphic to the shift on ðY, C , n, SÞ of type II1 or III, there exist
{0, 1}ℤ equipped with a quasi-invariant probabil- uncountably many mutually disjoint Borel mea-
ity measure. For a version of the Krengel 2-sets sures m on X such that (X, T, ℬ, m) is orbit equiv-
generator theorem in the Borel category, we refer alent to ðY, C , n, SÞ.
to Hochman (2019). On the other hand, T may not have any finite
invariant measure. The first such example
The Glimm-Effros Theorem appeared in Eigen et al. (1998b). We present a
The classical Bogolyubov-Krylov theorem states simpler one.
that each homeomorphism of a compact space
admits an ergodic invariant probability measure Example 2.15 Let T be an irrational rotation on
(Cornfeld et al. 1982). The following statement by the circle , and let K be a nowhere dense closed
Glimm (1961) and Effros (1965) is a “non- subset of  of positive Lebesgue measure. Let
singular” analogue of that theorem. (We consider X be the complement of the T-orbit [n  ℤ TnK of
here only a particular case of ℤ-actions.) K. Then X is a T-invariant Gδ-Subset of zero
Lebesgue measure. Hence X is Polish in the
+

Theorem 2.12 Let X be a Polish space and induced topology and T X is an aperiodic
T : X ! X an aperiodic homeomorphism. Then homeomorphism of X. Since T is minimal, X is
dense in  and the (T X)-orbit of each point
+

the following are equivalent:

of X is dense in X. Hence every point is recurrent.
(i) T has a recurrent point x, i.e., x ¼ By Theorem 2.10, there exists a continuous ergo-
lim n!1 T ni x for a sequence n1 < n2 <   . dic nonsingular probability Borel measure l on
+

(ii) There is an orbit of T which is not locally X. If it is invariant under T X, then l can be
closed. considered also as a finite T-invariant measure on
(iii) There is no Borel set which intersects each . Since T is uniquely ergodic, l is the Lebesgue
orbit of T exactly once. measure. However X is of zero Lebesgue mea-
(iv) There is a continuous probability Borel mea- sure, a contradiction.
sure m on X such that (X, m, T ) is an ergodic Let T be an aperiodic Borel transformation of a
nonsingular system. standard Borel space X. Denote by M(T ) the set
of all ergodic T-nonsingular continuous measures
A natural question arises: Under the conditions on X. Given m  M(T ), let N(m) denote the
of the theorem, how many such m can exist? It family of all Borel m-null subsets. Shelah and
Ergodic Theory: Nonsingular Transformations 239

Weiss showed (Shelah and Weiss 1982) that be the restriction of the product measure m Leb
\m  M(T )N(m) coincides with the collection of all on X ℝ to X f and define, for t  0,
Borel T-wandering sets.
n1

Minimal Radon Uniquely Ergodic Models for Sftðx,sÞ ≔ T n x, s þ t  f Tix ,

i¼0
Infinite Measure-Preserving Transformations
We first note that there is only one, up to a homeo-
where n is the unique integer that satisfies
morphism, locally compact noncompact Cantor
(i.e., zero-dimensional, perfect, and metrizable)
n1 n
set. Denote it by C. We recall that a Borel measure f Tix < s þ t  f Tix :
on C is called Radon if it is finite on every compact i¼0 i¼0
subset of C. The following is an infinite version of
the well known Jewett-Krieger theorem. A similar definition applies when t < 0. In
particular, when 0 < s þ t < ’ðxÞ, Sft ðx, sÞ ¼
Theorem 2.16 (Yuasa 2013). Let T be an ergodic (x, s þ t), so that the flow moves the point (x, s)
measure-preserving transformation of the standard up t units, and when it reaches (x, ’(x)), it is sent
infinite s -finite measure space (X, m). Then there to (Tx, 0). It can be shown that Sf ¼ Sft is a
tℝ
exists a minimal homeomorphism R of C that free m f-nonsingular flow and that it preserves m f if
admits a unique, up to scaling, R-invariant Radon and only if T preserves m (Nadkarni 1998). It is
measure v such that (X, m, T) is isomorphic to called the flow built under the function f with the
(C, n, R). base transformation T. Of course, S f is conserva-
Yuasa proved a relative version of this theorem tive or ergodic if and only if so is T.
in Yuasa (2020). Similar results on strictly ergodic Two flows S ¼ (St)t  ℝ on (X, ℬ, m) and V ¼
models for type III ergodic systems are not known (Vt)t  ℝ on ðY, C , nÞ are said to be isomorphic if
yet. We guess that the Radon property of the there exist invariant conull sets X 0  X and Y 0  Y
measure in this case should be replaced with and an invertible nonsingular map r : X 0 ! Y 0 that
another property, related to the Radon-Nikodym interwines the actions of the flows: r ∘ St ¼ Vt ∘ r
derivative of the original transformation. on X 0 for all t. The following nonsingular version
of Ambrose–Kakutani representation theorem
Special Representations of Ergodic Flows was proved by Krengel (1969) and Kubo (1969).
Nonsingular flows (¼ℝ-actions) appear naturally
in the study of orbit equivalence for systems of type Theorem 2.17 Let S be a free nonsingular flow.
III (see section “Orbit Theory”). Here we record Then it is isomorphic to a flow built under a function.
some basic notions related to nonsingular flows. Rudolph showed that in the Ambrose-
Let (X, ℬ, m) be a standard Borel space with a Kakutani theorem one can choose the function ’
s-finite measure m on ℬ. A nonsingular flow on to take two values. Krengel (1976) showed that
(X, m) is a Borel map S : X ℝ 3 (x, t) 7! Stx  X this can also be assumed in the nonsingular case.
such that StSs ¼ Stþs for all s, t  ℝ and each St
is a nonsingular transformation of (X, m).
Conservativity and ergodicity for flows are defined Panorama of Examples
in a similar way as for transformations.
A very useful example of a flow is a flow built This section is devoted entirely to examples of
under a function. Let (X, ℬ, m, T) be a non- nonsingular systems. We describe here the most
singular dynamical system and f a positive Borel popular (and simple) constructions of nonsingular
1
i¼0 f T x ¼
i
function on X such that systems: product odometers, nonsingular Markov
1 i
i¼0 f T x ¼ 1 for all x  X. Set odometers, tower transformations, rank-one finite
X f ≔ {(x, s) : x  X, 0  s < f(x)}. Define m f to rank systems, nonsingular Bernoulli and Markov
240 Ergodic Theory: Nonsingular Transformations

shifts, nonsingular Poisson suspensions and Gauss- (iv) A partial order on E so that e, e0  E are
ian transformations, IDPFT systems and natural comparable if and only if e, e0  E(n) for
extensions of nonsingular endomorphisms. some n and rn(e) ¼ rn(e0).

Nonsingular Product Odometers A Bratteli compactum XB of the diagram B is

Given a sequence mn of natural numbers, we the space of infinite paths
let An ≔ {0, 1, . . ., mn  1}. Let nn be a proba-
bility on An and nn(a) > 0 for all x¼ ðxn Þn>0 jxn  EðnÞ and r ðxn Þ ¼ sðxnþ1 Þ
a  An. Consider now the infinite product pro-
bability space ðX, mÞ≔1 n¼1 ðAn , nn Þ: Assume that
1 on B. XB is equipped with the natural topology
max fn ð aÞja  A g ¼ 0: Then m is non-
induced by the product topology on ∏n>0E(n). We
n¼1 n n

atomic. Given a1  A1, . . ., an  An, we denote will assume always that the diagram is essentially
by [a1, . . ., an] the cylinder x ¼ (xi)i > 0 j x1 ¼ a1, simple, i.e., there is only one infinite path xmax ¼
. . ., xn ¼ an. If x 6¼ (0, 0, . . .), we let l(x) be the (xn)n>0 with xn maximal for all n and only one xmin ¼
smallest number l such that the l-th coordinate of (xn)n>0 with xn minimal for all n. The Bratteli-
x is not ml  1. We define a Borel map T : X ! X Vershik map TB : XB ! XB is defined as follows:
by (1) if x 6¼ (m1, m2, . . .) and put Tx ≔ (0, 0, . . .) Txmax ¼ xmin. If x ¼ (xn)n>0 6¼ xmax, then let k be the
if x ¼ (m1, m2, . . .). Of course, T is isomorphic to smallest number such that xk is not maximal. Let yk
a rotation on a compact monothetic totally discon- be a successor of xk. Let (y1, . . ., yk) be the unique
nected Abelian group. It is easy to check that T is path such that y1, . . ., yk1 are all minimal. Then we
m-nonsingular and let TBx ≔ (y1, . . ., yk, xk þ 1, xkþ2, . . .). It is easy to
1
see that TB is a homeomorphism of XB. Suppose that
nn ðTxÞn ðnÞ
om1 ðxÞ ¼ we are given a sequence PðnÞ ¼ Pðv,eÞ  V n1 EðnÞ
n¼1
nn ð x n Þ
of stochastic matrices, i.e.,
lðxÞ1
nlðxÞ xlðxÞ þ 1 n n ð 0Þ
¼
nlðxÞ xlðxÞ nn ðmn  1Þ (i) Pðv,e
nÞ
> 0 if and only if v ¼ sn(e).
n¼1
ðnÞ
(ii) fe  EðnÞ jsn ðeÞ¼vg Pv,e ¼ 1 for each v 
for a.a. x ¼ (xn)n>0  X. It is also easy to verify V(n1).
that T is ergodic. It is called the nonsingular
product odometer associated to ðmn , nn Þ1n¼1 : We For e1  E(1), . . ., en  E(n), let [e1, . . ., en]
note that Ornstein’s transformation (Example 2.4) denote the cylinder {x ¼ (xj)j>0| x1 ¼ e1, . . ., xn ¼
is a nonsingular product odometer. en}. Then we define a Markov measure on XB by
setting
Markov Odometers
We define Markov odometers as in Dooley and mP ð½e1 , . . . , en Þ ¼ P1s1 ðe1 Þ,e1 P2s2 ðe2 Þ,e2   Pnsn ðen Þ,en
Hamachi (2003a). An ordered Bratteli diagram
B (Herman et al. 1992) consists of
for each cylinder [e1, . . ., en]. The dynamical sys-
tem (XB, mP, TB) is called a Markov odometer. It is
(i) A vertex set V which is a disjoint union of
easy to see that every nonsingular product odom-
finite sets V(n), n  0, V0 is a singleton.
eter is a Markov odometer where the
(ii) An edge set E which is a disjoint union of
corresponding V(n) are all singletons.
finite sets E(n), n > 0;
(iii) Source mappings sn : E(n) ! V(n1) and
Tower Transformations
range mappings rn : E(n) ! V(n) such that
This construction is a discrete analogue of flow
s1
n ðvÞ 6¼ ; for all v  V
(n1)
and r 1
n ðvÞ 6¼ ; under a function. Given a nonsingular dynamical
for all v  V , n > 0;
(n)
system (X, m, T ) and a measurable map f : X ! ℕ,
Ergodic Theory: Nonsingular Transformations 241

we define a new dynamical system (X f, m f, T f) by I n,0 ½ j, . . . I n,hn 1 ½ j followed by the spacers
setting Sn,0 ½ j, . . . Sn,sn ðjÞ1 ½ j:

Xf ≔fðx, iÞ  X ℤþ j0  i < f ðxÞg, Finally, each subcolumn is stacked from left to

dmf ðx, iÞ ≔dmðxÞ and right so that the top subinterval in subcolumn j is
ðx, i þ 1Þ, if i þ 1 < f ðxÞ sent to the bottom subinterval in subcolumn j þ 1,
T f ðx, iÞ ≔ for j ¼ 0, . . ., rn  2 (by the unique orientation-
ðTx, 0Þ, otherwise: preserving affine map between the intervals). For
example, Sn,sn ð0Þ1 ½0 is sent to In,0[1]. This defines
Then T f is m f-nonsingular and (dm f ∘ T f/dm f) a new column Cnþ1 and new column map T Cnþ1 ,
(x, i) ¼ (dm ∘ T/dm)(x) for a.a. (x, i)  X f. This which remains undefined on its top subinterval.
transformation is called the (Kakutani) tower Let X be the union of all intervals in all columns,
over T with height function f. It is easy to check and let m be Lebesgue measure restricted to X. We
that T f is conservative if and only if T is conser- assume that as n ! 1 the maximal length of the
vative; T f is ergodic if and only if T is ergodic; T f intervals in Cn converges to 0, so we may define a
is of type III if and only if T is of type III. transformation T of (X, m) by Tx ≔ lim n!1 T Cn x:
Moreover, the induced transformation (T f)X {0} One can verify that T is well-defined a.e. and that
is isomorphic to T. Given a subset A  X of it is nonsingular and ergodic. T is said to be the
positive measure, T is the tower over the induced rank-one transformation associated with
transformation TA with the first return time to A as ðrn , wn , sn Þ1
n¼1 : If all the probability vectors wn
the height function.
are uniform, the resulting transformation is
measure-preserving. The measure is infinite
Rank-One Transformations. Chacón Maps. (s-finite) if and only if the total mass of the
Finite Rank spacers is infinite. In the case rn ¼ 3 and sn(0) ¼
The definition uses the process of “cutting and sn(2) ¼ 0, sn(1) ¼ 1 for all n  0, the associated
stacking.” We construct by induction a sequence rank-one transformation is called a nonsingular
of columns Cn. A column Cn consists of a finite Chacón map.
sequence of bounded intervals (left-closed, right- It is easy to see that every nonsingular product
open) Cn ¼ fI n,0 , . . . , I n,hn 1 g of height hn. odometer is of rank-one (the corresponding maps
A column Cn determines a column map T Cn that sn are all trivial). Each rank-one map T is a tower
sends each interval In,i to the interval above it In,iþ1 over a nonsingular product odometer (to obtain
by the unique orientation-preserving affine map such an odometer, reduce T to a column Cn).
between the intervals. T Cn remains undefined on A rank N transformation is defined in a similar
the top interval I n,hn 1 : Set C0 ¼ {[0, 1)} and let way. A nonsingular transformation T is said to be of
{rn  2} be a sequence of positive integers, let {sn} rank N or less if at each stage of its construction there
be a sequence of functions sn : {0, . . ., rn  exit N disjoint columns, the levels of the columns
1} ! ℕ0, and let {wn} be a sequence of probability generate the s-algebra, and the Radon-Nikodym
vectors on {0, . . ., rn  1}. If Cn has been defined, derivative of T is constant on each nontop level of
column Cnþ1 is defined as follows. First “cut” every column. T is said to be of rank N if it is of rank
(i.e., subdivide) each interval In,i in Cn into rn sub- N or less and not of rank N  1 or less. A rank
intervals In,i[j], j ¼ 0, . . ., rn  1, whose lengths are N transformation, N  2, need not be ergodic.
in the proportions wn(0) : wn(1) :    : wn(rn  1).
Next place, for each j ¼ 0, . . .rn  1, sn( j), new Nonsingular Bernoulli Shifts
subintervals above I n,hn 1 ½ j, all of the same length A nonsingular Bernoulli transformation is a trans-
as I n,hn 1 ½ j: Denote these intervals, called spacers, formation T such that there exists a generating
by Sn,0 ½ j, . . . Sn,sn ðjÞ1 ½ j: This yields, for each s-algebra P for T (see §2.5) such that the
j  {0, . . ., rn  1}, rn subcolumns each s-algebras, T n P , n  ℤ, are mutually independent.
consisting of the subintervals Thus, we may think that T is the left shift on the
242 Ergodic Theory: Nonsingular Transformations

probability space (X, m) ¼ (Aℤ, n  ℤ mn), where M(a, b) ¼ 1, n  ℤ} > 0, a convenient

A is a standard Borel space and (mn)n  ℤ is a (computable) criterion for nonsingularity was
sequence of probability measures on A. We will found in Avraham-Re’em (2022, Theorem 5.2).
always assume that m is nonatomic. It follows By a standard computation,
from Kakutani’s criterion for equivalence of infi-
nite product measures (Kakutani 1948) that m is pn1 ðxn Þ
n
Pj1 xj , xjþ1
om1 ðxÞ ¼ lim :
nonsingular if and only if mn mnþ1 for each n and n!þ1 pn ðxn Þ Pj xj , xjþ1
j¼n

H 2 mn , mnþ1 < 1, ð4Þ We also mention here so-called infinite Markov

nℤ
shifts, i.e., Markov transformations preserving an
infinite s - finite measure (see Kakutani and Parry
where H2(m, n) denotes the Hellinger distance
(1963) and §4.5 from Aaronson (1997)). Let A ¼
defined by H 2 ðm, nÞ≔1  X
dm dn
dx dxdx, where x ℤ, and let P ¼ (P(a, b))a,b  A be a row stochastic
is a probability such that m ≺ x and n ≺ x. If (4) is A A-matrix. Suppose that P is irreducible, i.e.,
satisfied, then for a.e. x ¼ (xn)n  ℤ  X, for each pair a, b  A, there is n > 0 with
Pn(a, b) > 0. Suppose that there is a strictly pos-
mn1 ðxn Þ itive function p : A ! ℝþ such that a  Aπ(a) ¼
om1 ðxÞ ¼ :
mn ð x n Þ
nℤ 1 and a  Aπ(a)P(a, b) ¼ π(b) for all b  A. Let
T denote the restriction of the left shift to XP.
Nonsingular Markov Shifts Define a measure m on XP by setting
Let A be a finite set and let M ¼ (M(a, b))a,b  A be
a 0–1-valued A A-matrix. We let XM ≔ {x ¼ m ½aji ¼ pðai ÞPðai , aiþ1 Þ  P aj1 , aj :
(xi)i  ℤ  Aℤ| M(xi, xiþ1) > 0 for all i  ℤ}. Let
T denote the restriction of the left shift to XM. Then
Then m is infinite and s-finite and T preserves
T is a shift of finite type. Given two integers i  j,
m. By Kakutani and Parry (1963), if
and a finite sequence a ¼ ðal Þjl¼i of elements from 1
n¼1 P ð0, 0Þ ¼ 1 , then T is ergodic. We call
k
A such that M(al, alþ1) ¼ 1 for l ¼ i, . . ., j  1, we
the system (XP , m, T) the infinite Markov shift
define by ½aji the cylinder {x  XM | xl ¼ al for all
associated with (P, π).
l ¼ i, . . ., j}. Suppose that there is a sequence of
probability measures (πn)n  ℤ on A and a sequence Nonsingular Poisson Suspensions
(Pn)n  ℤ of row-stochastic A A-matrices such Let (X, ℬ, m) be a s-finite Lebesgue space with an
that πnPn ¼ πnþ1 and Pn(a, b) > 0 if and only if infinite nonatomic measure. Let ℬ0  ℬ be the
M(a, b) > 0 for each n  ℤ. Then there is a collection of subsets of finite measure. Denote by
unique probability measure m on XM such that X the space of measures o of the form o ¼
for every cylinder ½aji in XM.
i  I dxi , where I is a countable set. Endow X
with the smallest s-algebra ℬ such that the maps
m ½aji ¼ pi ðai ÞPi ðai , aiþ1 Þ  Pj1 aj1 , aj :
N A : X 3 o 7! oðAÞ  ℤþ t f1g
It is called Markov measure on XM generated
by (πn, Pn)n  ℤ. By analogy with Kakutani are all measurable, A  ℬ0. Let m be the (only)
(1948), using Kabanov et al. (1977) one can find probability measure on ℬ such that
necessary and sufficient conditions for T to be
m-nonsingular. Then the system (XM, T, m) is • m ∘N 1
A has the Poisson distribution with
called a nonsingular Markov shift. It is a natural parameter m(A), A  ℬ0.
generalization of nonsingular Bernoulli shifts. In • If A, B  ℬ0 and A \ B ¼ ;, then the maps NA
the case where M is irreducible and inf{Pn(a, b)| and NB are independent.
Ergodic Theory: Nonsingular Transformations 243

Let T be a nonsingular invertible transforma- dm∘S1 1 2 1

ðxÞ ¼ e2kyk2 þ
y x y
tion of (X, ℬ, m). Define a transformation T : dm
n¼1 n n at
X ! X by setting: T o ≔ o ∘ T1. If T is m -
nonsingular, then (X , ℬ , m , T ) is called the m  a:e: x ¼ ðxn Þ1 1 2
n¼1  X and y ¼ ðyn Þn¼1  ‘ :

nonsingular Poisson suspension of (X, ℬ, m, T )

(Danilenko et al. 2022a). It is easy to see that Sy is totally dissipative. It is
straightforward to verify that T O Sy T 1 O ¼ SOy :
Denote the affine group ℋ⋊O of ℋ by Aff ℋ.
Theorem 3.1 (Danilenko et al. 2022a). T is m - Then for each (h, O)  Aff ℋ, consider a trans-
nonsingular if and only if dm∘T
 1  L2 ðmÞ. If formation Gh,O ≔ ShTO of (X, m). It is called a
dm
nonsingular Gaussian transformation. Thus, each
dm∘T
 1  L1 ðmÞ then T is m -nonsingular and
dm nonsingular Gaussian transformation is a compo-
sition of a classical probability preserving Gauss-
 ðdm∘T ian transformation and a nonsingular totally
dm ∘T dm 1Þdm dm∘T
ðoÞ ¼ e X ðxÞ dissipative “rotation.” It is obvious that
dm dm
oðfxgÞ¼1
dm∘G1 dm∘S1 2 1
ðxÞ ¼ ekhk2 þ
h,O
ðxÞ ¼ h x h
n¼1 n n at
dm dm
at a.e. o  X .
m  a:e: x ¼ ðxn Þ1
n¼1  X:

Nonsingular Gaussian Transformations

IDPFT Transformations
Let γ stand for the normalized Gaussian measure
t2
Definition 3.2 (Danilenko and Lemańczyk
on ℝ : dgðtÞ ¼ p12p e 2 dt: Let (X, m) ≔ (ℝ, γ)ℕ. 2019). Let Tn be an ergodic nonsingular invertible
Denote by O the group of orthogonal operators in transformation of a standard probability space
a separable infinite dimensional real Hilbert space ðXn , Bn , mn Þ for each n  ℕ. Denote by T the
ℋ. Fix an orthonormal base in ℋ. Then we can infinite direct product of Tn, n  ℕ, acting on the
identify ℋ with ‘2(ℕ) and every operator from O infinite product space ðX, B, mÞ≔n  ℕ ðXn , Bn , mn Þ:
with an infinite ℕ ℕ-matrix. Every O  O deter- If T is m-nonsingular and each Tn is of finite type,
mines a Borel transformation TO : X ! X by the i.e., there exists a mn-equivalent probability mea-
formula sure nn which is invariant under Tn for each
n  ℕ, then T is called an infinite direct product
1
of finite types (IDPFT).
T O x≔ ðT O xÞn nℕ
, where ðT O xÞn ≔ On,m xm , n  ℕ:
m¼1 Let ’n ≔ dmdnn . It follows from the Kakutani crite-
n

rion (Kakutani 1948) that T is m-nonsingular if and

p
More precisely, TO is defined m-almost every- only if 1 n¼1 Xn ’n  ’n ∘T n > 0: Moreover, m is
where and it is invertible (mod 0). Moreover, TO mutually singular with the T-invariant probability n
p
preserves m. The set fT O j O  O g is exactly the ≔ n  ℤnn if and only if 1 n¼1 Xn ’n dnn ¼ 0. If
family of classical (probability preserving) Gauss- T is m-nonsingular, then
ian transformations, which are well studied in
1
ergodic theory. A wider class of nonsingular dm ∘ T dmn ∘ T n
ðxÞ ¼ ðxn Þ at
Gaussian transformations is defined by Arano, dm dmn
n¼1
Isono, and Marrakchi in Arano et al. (2021) (see
also Danilenko and Lemańczyk 2022 for a differ- a:e: x ¼ ðxn Þ1
n¼1  X:
ent but equivalent presentation of their concepts).
For each y  X, consider a transformation Natural Extensions of Nonsingular
Sy : X 3 x 7! x þ y  X. By the Cameron- Endomorphisms
Martin theorem, Sy is m-nonsingular if and only Let (X, ℬ, m) be a s-finite standard measure
if y  ‘2(ℕ) and space. A nonsingular endomorphism is a
244 Ergodic Theory: Nonsingular Transformations

measurable map R : X ! X such that m(A) ¼ 0 if (Eigen 1982). Ryzhikov showed (Ryzhikov 1993)
and only if m(R1A) ¼ 0. Suppose that m is s-finite that every element of this group is a product of
on R1ℬ. We define the Radon-Nikodym deriva- three involutions (i.e., transformations of order 2).
tive om1 of R by setting om1 ¼ dm∘R
dm
1 ∘ R . It was Moreover, a nonsingular transformation is a prod-
shown in (Silva 1988; Silva and Thieullen 1995) uct of two involutions if and only if it is conjugate
that there exists a s-finite standard measure space to its inverse by an involution.
(X , ℬ , m ), an invertible m -nonsingular trans- Inspired by (Halmos 1956), Ionescu Tulcea
formation R , and a Borel map π : X ! X such (Ionescu Tulcea 1965) and Chacon and Friedman
that the following hold: m ∘p1 ¼ m, pR ¼ (Chacon and Friedman 1965) introduced the weak
Rp, om1 is π1(ℬ)-measurable and and the uniform topologies, respectively, on Aut
n 1
_n>0R π (ℬ) ¼ ℬ . The dynamical system (X, m). The weak one – we denote it by dw – is
(X , ℬ , R , m ) is defined uniquely (up to a nat- induced from the weak operator topology on the
ural isomorphism) and called the natural exten- group of unitary operators in L2(X, m) by the embed-
sion of R. It coincides with the standard Rokhlin ding T 7! UT (see §2.3). Then (Aut(X, m), dw) is a
definition of the natural extension in the case Polish topological group and Aut0(X, n) is a closed
where R preserves m and m is finite. subgroup of Aut(X, m). This topology will not be
affected if we replace m with any equivalent mea-
Theorem 3.3 R is conservative if and only if R is sure. We note that Tn weakly converges to T if and
m-recurrent, i.e., only if m T 1 1
n ADT A ! 0 for each A  ℬ and
d(m ∘ Tn)/dm ! d(m ∘ T)/dm in L1(X, m). For each
i1
p  1, one can also embed Aut(X, m) into the
h ∘ Ri oi ¼ þ1 a:e:, where oi ¼ om1 ∘ Rj
i0 j¼0
isometry group of L p(X, m) via a formula similar
to (3) but with another power of the Radon-
for each integrable function h > 0. Moreover, if Nikodym derivative in it. The strong operator topol-
R is m-recurrent, then R is ergodic if and only if ogy on the isometry group induces the very same
R is ergodic (Silva 1988; Silva and Thieullen weak topology on Aut(X, m) for all p  1 (Choksi
1995). and Kakutani 1979). Danilenko showed in
Let R be a nonsingular one-sided Bernoulli Danilenko (1995) that (Aut(X, m), dw) is contract-
shift ðX, mÞ ¼ 1 n¼1 ðA, mn Þ: Then the natural ible. It follows easily from the Rokhlin lemma that
extension of R is isomorphic to the two-sided periodic transformations are dense in Aut(X, m).
nonsingular Bernoulli shift T on ðX , m Þ ¼ It is natural to ask which properties of non-
1 singular transformations are typical in the sense
n¼1 A, mn , where mn ¼ mn if n > 0 and
mn ¼ m1 if n  0. The corresponding projection of Baire category. The following technical lemma
(see Friedman 1970; Choksi and Kakutani 1979)
π : X ! X is the natural projection, i.e.,
is an indispensable tool when considering such
π(. . ., a1, a0, a1, a2, . . .) ≔ (a1, a2, . . .).
problems.

Topological Groups ΑUΤ(X, m), Lemma 4.1 The conjugacy class of each aperi-
ΑUΤ2(X, m) and AUT1(X, m) odic transformation T is dense in Aut(X, m) endo-
wed with the weak topology.
4.1. Let (X, ℬ, m) be a standard probability space, It follows that Aut(X, m) has the Rokhlin prop-
and let Aut(X, m) denote the group of all non- erty, i.e., there is an element of this group whose
singular transformations of X. Let v be a finite or conjugacy class is dense in the group. Using
s-finite measure equivalent to m; the subgroup of Lemma 4.1 and the Hurewicz ergodic theorem,
the v-preserving transformations is denoted by Choksi and Kakutani (Choksi and Kakutani 1979)
Aut0(X, n). Then Aut(X, m) is a simple group proved that the ergodic transformations form a
(Eigen 1981), and it has no outer automorphisms dense Gδ in Aut(X, m). The same holds for the
Ergodic Theory: Nonsingular Transformations 245

subgroup Aut0(X, n) (Sachdeva 1971; Choksi and course, C(T) is a closed subgroup of Aut(X, m)
Kakutani 1979). Combined with (Ionescu Tulcea and C(T ) {Tn| n  ℤ}. In a similar way, if
1965), the above implies that the ergodic transfor- T  Aut0(X, n), the measure-preserving central-
mation of type III is a dense Gδ in Aut(X, m). For izer C0(T ) ≔ Aut0(X, n) \ C(T ) of T is a weakly
further refinement of this statement, we refer to closed subgroup of Aut0(X, n). The following
section “Orbit Theory”. problems solved (by several authors) for
Since the map T 7! T    T ( p times) from probability-preserving systems are still open for
Aut(X, m) to Aut(X p, mp) is continuous for each the nonsingular case. The properties are:
p > 0, we deduce that the set E 1 of transformations
with infinite ergodic index (which means that T (i) T has square root.
   T ( p times) is ergodic for each p > 0) is a Gδ (ii) T embeds into a flow.
in Aut(X, m). It is nonempty by (Kakutani and Parry (iii) T has nontrivial invariant sub-s-algebra.
1963). Since this E 1 is invariant under conjugacy, it (iv) C(T ) contains a torus of arbitrary dimension.
is dense in Aut(X, m) by Lemma 4.1. Thus, we
obtain that E 1 is a dense Gδ. In a similar way, one typical (residual) in Aut(X, m) or Aut0(X, n)?
can show that E 1 \ Aut0(X, n) is a dense Gδ in The uniform topology on Aut(X, m), finer than
Aut0(X, n) (see also (Sachdeva 1971; Choksi and dw, is defined by the metric
Kakutani 1979; Choksi and Nadkarni 2000) for
original proofs of these claims). du ðT, SÞ ¼ mðfx : Tx 6¼ SxgÞ
A nonsingular transformation T is called rigid
þm x : T 1 x 6¼ S1 x :
if T ni ! Id weakly for some sequence nk ! 1.
The rigid transformations form a dense Gδ in
This topology is also complete metric. It
Aut(X, m). It follows that the set of multiply recur-
depends only on the measure class of m. However,
rent nonsingular transformations is residual
the uniform topology is not separable, and that is
(Ageev and Silva 2001). A finer result was
why it is of less importance in ergodic theory. We
established in Danilenko and Silva (2004): The
refer to (Chacon and Friedman 1965; Friedman
set of polynomially recurrent transformations in
1970; Choksi and Kakutani 1979; Choksi and
Aut0(X, n) is residual in Aut0(X, n). For the defi-
Prasad 1983) for the properties of du.
nition of multiple and polynomial recurrence, we
4.2. Suppose now that m(X) ¼ 1 but m is
refer to §6.5 below.
s-finite. We now let
Given T  Aut(X, m), we denote the central-
izer {S  Aut(X, m) | ST ¼ TS} of T by C(T ). Of

dm∘T
Aut2 ðX, mÞ ≔ T  AutðX, mÞ j  1  L2 ðmÞ and
dm
dm∘T
Aut1 ðX, mÞ ≔ T  AutðX, mÞ j  1  L1 ðmÞ :
dm

These groups appear naturally in the study of converges to T in d1 if Tn ! T weakly and

dm ∘ T n
nonsingular Poisson suspensions (see (3.7) and
dm  dmdm∘ T ! 0 as n ! 1.
§9). Define a topology d2 on Aut2(X, m) by setting 1

that a sequence ðT n Þ1
n¼1 converges to T in d2 if
Theorem 4.2 (Danilenko et al. 2022a)
dm ∘ T n dm ∘ T
Tn ! T weakly and dm  dm !0
2
as n ! 1. In a similar way, one can define a • (Aut2(X, m), d2) is a Polish group.
topology d1 on Aut1(X, m): A sequence ðT n Þ1
n¼1
• (Aut1(X, m), d1) is a Polish group.
246 Ergodic Theory: Nonsingular Transformations

• The mapping w : Aut1 ðX, mÞ 3 T 7! X dmdm∘ T  1 dm  ℝ transformations S  Aut(X, m) such that

is a continuous onto homomorphism. Sx  OrbT(x) for a.a. x. If T is ergodic, then [T]
• The short exact sequence is topologically simple (or even algebraically sim-
ple if T is not of type II1) (Eigen 1981). It is easy
to see that [T] endowed with the uniform topology
w
f1g ! kerw ! Aut 1 ðX, mÞ ! ℝ ! f1g du is a Polish group. If T is ergodic, then ([T], du)
is contractible (Danilenko 1995).
splits. The ratio set r(T ) of T was defined by Krieger
[Kr70], and as we shall see below it is the key
• Aut1(X, m) is a dense and meager subgroup of concept in the orbit classification (see Definition
(Aut2(X, m), d2). 2.13). The ratio set is a subset of [0, +1) defined
• w is not d2-continuous. as follows: t  r(T) if and only if for every A  ℬ
• The group {T | T  Aut2(X, m)} is weakly of positive measure and each ϵ > 0 there is a
closed in Aut(X , m ). subset B  A of positive measure and an integer
• The groups (Aut2(X, m), d2) and (ker w, d1) k 6¼ 0 such that TkB  A and jomk ðxÞ  tj < ϵ for all
have the Rokhlin property. x  B. It is easy to verify that r(T ) depends only
• The group (Aut1(X, m), d1) does not have the on the equivalence class of m and not on m itself.
Rokhlin property. A basic fact is that 1  r(T) if and only if T is
conservative. Assume now T to be conservative
and ergodic. Then r(T ) \ (0, +1) is a closed
Orbit Theory subgroup of the multiplicative group (0, +1).
Hence r(T ) is one of the following sets:
Orbit theory is, in a sense, the most complete part
of nonsingular ergodic theory. We present here the (i) {1}.
seminal Krieger’s theorem on orbit classification (ii) {0, 1}; in this case, we say that T is of type
of ergodic nonsingular transformations in terms of III0.
ratio sets and associated flows. Examples of trans- (iii) {ln| n  ℤ} [ {0} for 0 < l < 1; then we
formations of various types IIIl, 0  l  1 are say that T is of type IIIl.
given. “Almost continuous” refinement of the (iv) [0, +1); then we say that T is of type III1.
orbit equivalence is also considered here. Next,
we consider the outer conjugacy problem for auto- Krieger showed that r(T) ¼ {1} if and only if
morphisms of the orbit equivalence relations. This T is of type II. Hence we obtain a further subdivi-
problem is solved in terms of a simple complete sion of type III into subtypes III0, IIIl, 0 < l < 1,
system of invariants. We discuss also a general and III1. The subset of transformations of type III1
theory of cocycles (of nonsingular systems) tak- is a dense Gδ in Aut (X, m) (Choksi et al 1987;
ing values in locally compact Polish groups and Parthasarathy and Schmidt 1977).
present an important orbit classification theorem
for cocycles. This theorem is an analogue of the Example 5.1
aforementioned result of Krieger. We complete (i) Fix l  (0, 1). Let nn(0) ≔ 1/(1 þ l) and
the section by considering ITPFI-systems and nn(1) ≔ l/(1 þ l) for all n ¼ 1, 2, . . .. Let
their relation to AT-flows. T be the nonsingular product odometer associ-
ated with the sequence ð2, nn Þ1 n¼1 ð see §3.1).
Full Groups. Ratio Set and Types IIIl, 0 ≤ l ≤ 1 We claim that T is of type IIIl. Indeed, the
Let T be a nonsingular transformation of a stan- group S of finite permutations of ℕ acts on
dard probability space (X, ℬ, m). Denote by X by ðsxÞn ¼ xs1 ðnÞ, for all n  ℕ, s  S, and
OrbT(x) the T-orbit of x, i.e., OrbT(x) ¼ {T nx| x ¼ ðxn Þ1
n¼1  X: This action preserves m.
n  ℤ}. The full group [T] of T consists of all Moreover, it is ergodic by the Hewitt-Savage
Ergodic Theory: Nonsingular Transformations 247

l on the cylinder [0] which is of positive Proposition 5.2

measure. (i) T is of type II if and only if its associated flow is
(ii) Fix positive reals r1 and r2 such that logr1 the translation on ℝ, i.e., x 7! x þ t, x, t  ℝ,
and logr2 are rationally independent. Let (ii) T is of type IIIl, 0  l < 1 if and only if its
nn(0) ≔ 1/(1 þ r1 þ r2), nn(1) ≔ r1/(1 þ r1 þ associated flow is the periodic flow on the
r2), and nn(2) ≔ r2/(1 þ r1 þ r2) for all n ¼ 1, interval [0, log l), i.e., x 7! x þ
2,.... Then the nonsingular product odometer t mod (log l).
associated with the sequence ð3, nn Þ1 n¼1 is of (iii) T is of type III1 if and only if its associated
type III1. This can be shown in a similar way flow is the trivial flow on a singleton or,
as (i). equivalently, T is ergodic.
(iii) Partition ℕ into two infinite subsets A and B. (iv) T is of type III0 if and only if its associated
Fix a sequence (ϵ n)n  B of positive reals such flow is nontransitive (Hamachi and Osikawa
that ϵn < 0.4 for all n  B and n  Bϵn < 1. 1981).
We now let nn(0) ≔ 0.5 if n  A and mn(0)
≔ 1  ϵ n if n  B. Then the nonsingular
product odometer associated with the
sequence ð2, nn Þ1n¼1 is of type II1. This fol- Example 5.3 Let An ¼ 0, 1, . . . , 22
n
and
lows from (Moore 1967). 2n
nn(0) ¼ 0.5 and nn ðiÞ ¼ 0:5  2 for all
0 < i  2n. Let T be the nonsingular product
1
odometer associated with 22 þ 1, nn n¼0 : It is
n
Nonsingular product odometer of type III0 will
be constructed in Example 5.3 below. straightforward that the associated flow of T is the
flow built under the constant function 1 with the
Maharam Extension, Associated Flow, and probability preserving 2-adic product odometer
Orbit Classification of Type III Systems (associated with ð2, kn Þ1n¼1 , kn ð0Þ ¼ kn ð1Þ ¼ 0:5Þ
On X ℝ with the s-finite measure m k, where as the base transformation. In particular, T is of
dk( y) ¼ exp ( y)dy, consider the transformation type III0.
A natural problem arises: to compute
dm∘T Krieger’s type (or the ratio set) for the non-
T ðx, yÞ≔ Tx, y log ðxÞ :
dm singular product odometers – the simplest class
of nonsingular systems. Some partial progress
We call it the Maharam extension of T (see was achieved in (Araki and Woods 1968; Dooley
(Maharam 1964), where these transformations et al. 1998; Moore 1967; Osikawa 1988),
were introduced). It is measure-preserving, and it etc. However, in the general setting this problem
commutes with the flow St(x, y) ≔ (x, y þ t), remains open.
t  ℝ. It is conservative if and only if T is con- The map C : AutðX, mÞ 3 T 7! T  AutðX ℝ, m kÞ
servative (Maharam 1964). However T is not is a continuous group homomorphism. Since the
necessarily ergodic when T is ergodic. Let (Z, n) set E of ergodic transformations on X ℝ is a Gδ
denote the space of T-ergodic components. Then in Aut(X ℝ, m k) (See § 4), the subset C1(E)
(St)t  ℝ acts nonsingularly on this space. The of type III1 ergodic transformations on X is also
restriction of (St)t  ℝ to (Z, n) is called the associ- Gδ. The latter subset is nonempty in view of
ated flow of T. The associated flow is ergodic if Example 5.1(ii). Since it is invariant under
and only if T is ergodic. It is easy to verify that the conjugacy, we deduce from Lemma 4.1 that the
isomorphism class of the associated flow is an set of ergodic transformations of type III1 is a
invariant of the orbit equivalence of the underly- dense Gδ in (Aut(X, m), dw) (Choksi et al. 1987;
ing system. Parthasarathy and Schmidt 1977).
248 Ergodic Theory: Nonsingular Transformations

Now we state the main result of this section – Theorem 5.5 Every nontransitive ergodic flow is
Krieger’s theorem on orbit classification for ergo- an associated flow of an ergodic transformation
dic transformations of type III. It is a far-reaching of type III0.
generalization of the basic result by H. Dye: Any In Krieger (1976b), Krieger introduced a map
two ergodic probability-preserving transforma- F as follows. Let T be an ergodic transformation
tions are orbit equivalent (Dye 1959). of type III0. Then the associated flow of T is a flow
built under function with a base transformation
Theorem 5.4 (Orbit equivalence for type III sys- F(T ). We note that the orbit equivalence class of
tems (Krieger 1969, 1970, 1972, 1976a, b)). Two F(T ) is well defined by the orbit equivalent class
ergodic transformations of type III are orbit of T. If Fn(T ) fails to be of type III0 for some
equivalent if and only if their associated flows 1  n < 1, then T is said to belong to Krieger’s
are isomorphic. In particular, for a fixed hierarchy. For instance, the transformation
0 < l  1, any two ergodic transformations of constructed in Example 5.3 belongs to Krieger’s
type IIIl are orbit equivalent. hierarchy. Connes gave in Connes (1975) an
The original proof of this theorem is rather example of T such that F(T) is orbit equivalent
complicated. Simpler treatment of it can be to T (see also Hamachi and Osikawa (1981;
found in Hamachi and Osikawa (1981) and Giordano and Skandalis 1985a). Hence T is not
Katznelson and Weiss (1991). in Krieger’s hierarchy.
We also note that every free ergodic flow can be
realized as the associated flow of a type III0 trans- Almost Continuous Orbit Equivalence
formation. However it is somewhat easier to con- In this subsection, by a dynamical system we
struct a ℤ2-action of type III0 whose associated flow mean a quadruple (X, t, m, T ), where (X, t) is a
is the given one. For this, we take an ergodic Polish space, m is a nonatomic Borel measure of
nonsingular transformation Q on a probability full support, and T is a nonsingular ergodic
space (Z, ℬ, l) and a measure-preserving transfor- homeomorphism of X such that the function
mation R of an infinite s-finite measure space o1 : X ! ℝ is continuous (has a continuous
(Y, F , n) such that there is a continuous homomor- version).
phism π : ℝ ! C(R) with (dn ∘ π(t)/dn)( y) ¼ exp (t)
for a.a. y (for instance, take a type III1 transforma- Definition 5.6 Two dynamical systems
tion T and put R ≔ T and π(t) ≔ St). Let ’ : Z ! ℝ (X, t, m, T) and (X 0, t0, m0, T0) are almost contin-
be a Borel map with infZ ’ > 0. Define two trans- uously orbit equivalent if there are dense invariant
formations R0 and Q0 of (Z Y, l n) by setting: Gδ subsets X0  X and X00  X0 of full measure
and a homeomorphism ’ : X0 ! X00 such that
R0 ðx, yÞ ≔ ðx, RyÞ, Q0 ðx, yÞ ¼ ðQx, Ux yÞ,
• ’({Tnx| n  ℤ}) ¼ {(T0)n’(x) | n  ℤ} at
where Ux ¼ π(’(x)  log (dm ∘ Q/dm)(x)). Notice every x  X0.
that R0 and Q0 commute. The corresponding ℤ2- • m ∘ ’1 m0 and the Radon-Nikodym deriva-
action generated by these transformations is 1
tive dm∘’
dm0 is (can be chosen) continuous.
ergodic. Take any transformation V  Aut(Z
• letting S ≔ ’1T0’, we have Tx ¼ Sn(x)x and
Y, l n) whose orbits coincide with the orbits of
Sx ¼ T m(x)x, where n and m are continuous on
the ℤ2-action. (According to Connes et al. (1981),
X0.
any ergodic nonsingular action of any countable
amenable group is orbit equivalent to a single
transformation.) It is now easy to verify that the We note that in the case where X and X 0 are
associated flow of V is the special flow built under infinite product spaces, T and T0 preserve m and m0,
’ ∘ Q1 with the base transformation Q1. Then respectively, and we omit the requirement that X0
V is of type III0. Since Q and ’ are arbitrary, we and X0 are Gδ then the above definition of ’ is
deduce the following from Theorem 2.17. equivalent to the “finitary” equivalence from the
Ergodic Theory: Nonsingular Transformations 249

celebrated work of Keane and Smorodinsky iðRÞ  Aut RT , mRT by setting i(R)(x, y) ≔ (Rx,
(1979). It was shown by del Junco and Şahin Ry). Then the map R 7! i(R) is an embedding of
(2009) that any two ergodic probability- N[T] into Aut RT , mRT . Denote by t the topol-
preserving homeomorphisms of Polish spaces ogy on N[T] induced by the weak topology on
are almost continuously orbit equivalent. The Aut RT , mRT via i (Danilenko 1995). Then (N
same is true for any ergodic homeomorphisms [T], t) is a Polish group. A sequence Rn con-
preserving infinite s-finite local measures (del verges to R in (N[T], t) if Rn ! R weakly
Junco and Şahin 2009). In Danilenko and del (in Aut(X, m)) and Rn TR1 n ! RTR
1
uniformly
Junco (2011), a topological analogue rtop(T ) of (in [T]).
r(T) was introduced. It is a closed subgroup of ℝ Given R  N[T], denote by R the Maharam
which contains r(T ), and it is invariant under the extension of R. Then R  N T and it commutes
almost continuous orbit equivalence. In with (St)t  ℝ. Hence it defines a nonsingular trans-
Danilenko and del Junco (2011), two-type III formation mod R on the space (Z, n) of the asso-
homeomorphisms were constructed which are ciated flow W ¼ (Wt)t  ℝ of T. Moreover, mod
measure-theoretically orbit equivalent but not R belongs to the centralizer C(W) of W in Aut(Z,
almost continuously orbit equivalent (their rtop- n). Note that C(W) is a closed subgroup of
invariants are different). (Aut(Z, n), dw).
Let T be of type II1, and let m0 be the invariant
Theorem 5.7 Let (X, t, m, T ) and (X 0, t0, m0, T0) s-finite measure equivalent to m. If R  N[T], then
be ergodic nonsingular homeomorphisms of Pol- it is easy to see that the Radon-Nikodym deriva-
ish spaces. If the two systems are either tive dm0 ∘ R/dm0 is invariant under T. Hence it is
(Danilenko and del Junco 2011) constant, say c. Then mod R ¼ log c.

(i) Of type IIIl with 0 < l < 1 and rtop(T ) ¼ Theorem 5.8 If T is of type III, then the map
rtop(T 0) ¼ log l  ℤ or mod : N[T] ! C(W ) is a continuous onto homo-
(ii) Of type III1, morphism. The kernel of this homomorphism is
the t-closure of [T]. Hence the quotient group
then they are almost continuously orbit t
N ½T =½T  is (topologically) isomorphic to C(W).
equivalent. t
In particular, ½T  is cocompact in N[T] if and
Characterization of almost continuous orbit only if W is a finite measure-preserving flow with
equivalence for homeomorphisms of type III0 a pure point spectrum (Hamachi and Osikawa
remains an open problem. 1981; Hamachi 1981a).
The following theorem describes the homo-
Normalizer of the Full Group. Outer topical structure of normalizers.
Conjugacy Problem
Let Theorem 5.9 Let T be of type II or III, 0  l < 1.
t
The group ½T  is contractible. N[T] is homo-
N ½T  ¼ R  AutðX, mÞjR½T R1 ¼ ½T  ,
topically equivalent to C(W). In particular, N[T] is
contractible if T is of type II. If T is of type IIIl with
i.e., N[T] is the normalizer of the full group
0 < l < 1, then π1(N[T]) ¼ ℤ (Danilenko 1995).
[T] in Aut(X, m). We note that a transformation
The outer period p(R) of R  N[T] is the
R belongs to N[T] if and only if R(OrbT(x)) ¼
smallest positive integer n such that Rn  [T]. We
OrbT(Rx) for a.a. x. To define a topology on N[T],
write p(R) ¼ 0 if no such n exists.
consider the T-orbit equivalence relation
Two transformations R and R0 in N[T] are
RT  X X and a s-finite measure mR on RT
called outer conjugate if there are transformations
given by mRT ¼ X y  OrbT ðxÞ dðx,yÞ dmðxÞ: For
V  N[T] and S  [T] such that V RV1 ¼ R0S.
R  N[T], we define a transformation The following theorem provides convenient (for
250 Ergodic Theory: Nonsingular Transformations

verification) necessary and sufficient conditions extension T’ of T acting on (X G, m lG) by

for the outer conjugacy. setting T’(x, g) ≔ (Tx, ’(x)g). Thus, Maharam
extension is (isomorphic to) the Radon-Nikodym
Theorem 5.10 (Connes and Krieger 1977) for cocycle-skew product extension. We now specify
type II and (Bezuglyi and Golodets 1985) for some basic classes of cocycles (Schmidt 1977a;
type III). Transformations R, R0  N[T] are Bezuglyi and Golodets 1991; Golodets and
outer conjugate if and only if p(R) ¼ p(R0) and Sinel’shchikov 1994; Danilenko 1998):
mod R is conjugate to modR0 in the centralizer of
the associated flow of T. (i) ’ is called transient if T’ is totally disipative.
We note that in the case T is of type II, the (ii) ’ is called recurrent if T’ is conservative
second condition in the theorem is just modR ¼ (equivalently, ’ is not transient).
mod R0. It is always satisfied when T is of type II1. (iii) ’ has dense range in G if T’ is ergodic.
(iv) ’ is called regular if ’ cobounds with dense
Cocycles of Dynamical Systems. Weak range into a closed subgroup H of G (then
Equivalence of Cocycles H is defined up to conjugacy).
Let G be a locally compact Polish group and lG a
left Haar measure on G. A Borel map ’ : X ! G is These properties are invariant under the coho-
called a cocycle of T. Two cocycles ’ and ’0 are mology and the weak equivalence. The Radon-
cohomologous if there is a Borel map b : X ! G Nikodym cocycle o1 is a coboundary if and only
such that if T is of type II. It is regular if and only if T is of
type II or IIIl, 0 < l  1. It has dense range (in the
’0 ðxÞ ¼ bðTxÞ1 ’ðxÞbðxÞ multiplicative group ℝþ) if and only if T is of type
III1. Notice that o1 is never transient (since T is
for a.a. x  X. A cocycle cohomologous to the conservative).
trivial one is called a coboundary. Given a dense In case G is Abelian, Schmidt introduced in
subgroup G0  G, every cocycle is cohomologous Schmidt (1984) an invariant R(’) ≔ {g  G j ’  g
to a cocycle with values in G0 (Golodets and is recurrent}. He showed in particular that
Sinel’shchikov 1994). Each cocycle ’ extends to
a (unique) map α’ : RT ! G such that α’(Tx, x) ¼
(i) R(’) is a cohomology invariant.
’(x) for a.a. x and α’(x, y)α’(y, z) ¼ α’(x, z) for
(ii) R(’) is a Borel set in G.
a.a. (x, y), (y, z)  RT. α’ is called the cocycle of
(iii) R(logo1) ¼ {0} for each aperiodic
RT generated by ’. Moreover, ’ and ’0 are
conservative T.
cohomologous via b as above if and only if α’
(iv) There are cocycles ’ such that R(’) and
and a’0 are cohomologous via b, i.e., a’ ðx, yÞ ¼ G\R(’) are dense in G.
bðxÞ1 a’0 ðx, yÞbðyÞ for mRT -a.a. (x, y)  RT. The (v) If m(X) ¼ 1, m ∘ T ¼ m and ’ : X ! ℝ is
following notion was introduced by Golodets and integrable, then R(’) ¼ { ’dm}.
Sinelshchikov (Golodets and Sinel’shchikov
1983, 1994): Two cocycles ’ and ’0 are weakly
We note that (v) follows from Atkinson theo-
equivalent if there is a transformation R  N[T]
rem (Atkinson 1976). A nonsingular version of
such that the cocycles α’ and a0’ ∘ðR RÞ of RT
this theorem was established in Ullman (1987): If
are cohomologous. Let M(X, G) denote the set of
T is ergodic and m-nonsingular and f  L1(m), then
Borel maps from X to G. It is a Polish group when
endowed with the topology of convergence in
n1
measure. Since T is ergodic, it is easy to deduce lim inf j f T j x oj ðxÞj ¼ 0 for a:a: x
from Rokhlin’s lemma that the cohomology class n!1
j¼0
of any cocycle is dense in M(X, G). Given
’  M(X, G), we define the ’-skew product if and only if fdm ¼ 0.
Ergodic Theory: Nonsingular Transformations 251

Since T’ commutes with the action of G on Theorem 5.13 Let T be an ergodic nonsingular
X G by inverted right translations along the transformation. If G is a compactly generated
second coordinate, this action induces an ergodic [FIA]-group, then there is a bounded ergodic
G-action W’ ¼ (W’(g))g  G on the space (Z, n) of cocycle ’ of T with values in G (Danilenko 2021).
T’-ergodic components. It is called the Mackey We recall that a locally compact Polish group
range (or Poincaré flow) of ’ (Mackey 1966; G is called a [FIA]-group if the group of inner
Feldman and Moore 1977; Schmidt 1977a). automorphisms of G is relatively compact in the
We note that ’ is regular (and cobounds with group of all automorphisms of G furnished with
dense range into H  G) if and only if W’ is the natural topology (Grosser and Moskowitz
transitive (and H is the stabilizer of a point 1971). Of course, each Abelian group is [FIA].
z  Z, i.e., H ¼ {g  G | W’(g)z ¼ z}). Hence The cocycle ’ is bounded if there is a compact
every cocycle taking values in a compact group is subset K in G such that ’ takes values in K.
regular. Theorem 5.5 is a particular case of the follow-
It is often useful to consider the double cocycle ing result.
’0 ≔ ’ o1 instead of ’. It takes values in the
group G ℝþ . Since T ’0 is exactly the Maharam Theorem 5.14 Let G be amenable. Let V be an
extension of T’, it follows from Maharam (1964) ergodic nonsingular action of G ℝþ : Then
that ’0 is transient or recurrent if and only if ’ is there is an ergodic nonsingular transformation
transient or recurrent, respectively. T and a recurrent cocycle ’ of T with values in
G such that V is isomorphic to the Mackey range
Theorem 5.11 (Orbit classification of cocycles of the double cocycle ’0 (Golodets and
(Golodets and Sinel’shchikov 1994)). Let ’, Sinel’shchikov 1990; Fedorov 1985; Adams
’0 : X ! G be two recurrent cocycles of an ergodic et al. 1994).
transformation T. They are weakly equivalent if Given a cocycle ’  M(X, G) of T, we say that
and only if their Mackey ranges W ’0 and W ’00 are a transformation R  N[T] is compatible with ’ if
isomorphic. the cocycles α’ and α’ ∘ (R R) of RT are
Another proof of this theorem was presented in cohomologous. Denote by D(T, ’) the group of
Fedorov (1985). all such R. It has a natural Polish topology which
is stronger than t (Danilenko and Golodets 1996).
Theorem 5.12 Let T be an ergodic nonsingular Since [T] is a normal subgroup in D(T, ’), one can
transformation. Then there is a cocycle of T with consider the outer conjugacy equivalence relation
dense range in G if and only if G is amenable. inside D(T, ’). It is called ’-outer conjugacy. Sup-
It follows that if G is amenable then the subset pose that G is Abelian. Then an analogue of Theo-
of cocycles of T with dense range in G is a dense rem 5.10 for the ’-outer conjugacy is established in
Gδ in M(X, G) (just adapt the argument following Danilenko and Golodets (1996). Also, the cocycles
Example 5.3). The “only if ” part of Theorem 5.12 ’ with D(T, ’) ¼ N[T] are described there.
was established in Zimmer (1978). The “if” part
was considered by many authors in particular ITPFI Transformations and AT-Flows
cases: G is compact Zimmer (1977), G is solvable A nonsingular transformation T is called ITPFI
or amenable almost connected Golodets and (This abbreviates “infinite tensor product of fac-
Sinel’shchikov (1985), etc. The general case was tors of type I” (came from the theory of von
proved in Golodets and Sinel’shchikov (1983) Neumann algebras).) if it is orbit equivalent to a
and Herman (1979a) (see also a refinement in nonsingular product odometer (associated to a
(Aaronson and Weiss 2004)). sequence ðmn , nn Þ1
n¼1 , see § 3.1). If the sequence
We note that the “if” part in Theorem 5.12 can mn can be chosen bounded, then T is called ITPFI
be refined in the case where G is a compactly of bounded type. If mn ¼ 2 for all n, then T is called
generated Abelian group. ITPFI2. By Giordano and Skandalis (1985b),
252 Ergodic Theory: Nonsingular Transformations

every ITPFI-transformation of bounded type is transformation which does not satisfy this prop-
ITPFI2. In view of Theorem 5.4 and Example erty (Krieger 1972). Hence this transformation is
5.1, every ergodic transformation of type II or not ITPFI. Though not every ergodic transforma-
IIIl with 0 < l  1 is ITPFI2. tion is orbit equivalent to a nonsingular product
A remarkable characterization of ITPFI trans- odometer, a “weaker” form of this statement
formations in terms of their associated flows was holds.
obtained by Connes and Woods (1985). We first
single out a class of ergodic flows. A nonsingular Theorem 5.16 Each ergodic nonsingular trans-
flow V ¼ (Vt)t  ℝ on a space (Ω, n) is called formation is orbit equivalent to a Markov odom-
approximate transitive (AT) if given ϵ > 0 and eter (see §3.2) (Dooley and Hamachi 2003a).
f 1 , . . . , f n  L1þ ðX, mÞ, there exists f  L1þ ðX, mÞ In Dooley and Hamachi (2003b), an explicit
and l1 , . . . , ln  L1þ ðℝ, dt Þ such that example of a non-ITPFI ergodic Markov odome-
ter (not satisfying property A) was constructed.
dn∘V t Later Munteanu in Munteanu (2012) exhibited an
fj  f ∘V t l ðtÞdt <ϵ
ℝ dn j 1
ergodic non-ITPFI transformation satisfying
property A. In Joita and Munteanu (2014), it was
for all 1  j  n. A flow built under a constant constructed an explicit example of a non-AT non-
ceiling function with a funny rank-one (Ferenczi singular flow W built under a function and over a
1985) probability preserving base transformation nonsingular product odometer. Hence every non-
is AT (Connes and Woods 1985). In particular, singular ergodic transformation whose associated
each ergodic finite measure-preserving flow with flow is isomorphic to W is non-ITPFI.
a pure point spectrum is AT.

Theorem 5.15 An ergodic nonsingular transfor- Mixing Notions and Multiple Recurrence
mation is ITPFI if and only if its associated flow is
AT (Connes and Woods 1985). The study of mixing and multiple recurrence are
The original proof of this theorem was given in central topics in classical ergodic theory (Cornfeld
the framework of von Neumann algebras theory. et al. 1982; Furstenberg 1981). Unfortunately,
A simpler, purely measure-theoretical proof was these notions are considerably less “smooth” in
given later in Hawkins (1990b) (the “only if ” part) the world of nonsingular systems. The very con-
and Hamachi (1992) (the “if ” part). It follows cepts of any kind of mixing and multiple recur-
from Theorem 5.15 that every ergodic flow with rence are not well understood in view of their
pure point spectrum is the associated flow of an ambiguity. Below we discuss nonsingular systems
ITPFI transformation. This was refined recently in possessing a surprising diversity of such proper-
Berendschot and Vaes (2022b): Every ergodic ties that seem equivalent but are different indeed.
flow with a pure point spectrum is the associated
flow of an ITPFI2 transformation. This fact was Weak Mixing
proved earlier in Hamachi and Osikawa (1986) Let T be an ergodic conservative nonsingular
only for flows whose spectrum is θΓ, where Γ is a transformation. A number l  ℂ is an L1-eigen-
subgroup of ℚ and θ  ℝ. The existence of ITPFI value for T if there exists a nonzero f  L1 so that
transformations which are not of bounded type f ∘ T ¼ lf a.e. It follows that jl j ¼ 1 and f has
was shown in Krieger (1972). constant modulus, which we assume to be
Krieger introduced an invariant for the orbit 1. Denote by e(T ) the set of all L1-eigenvalues
equivalence, called property A, and showed that of T. T is said to be weakly mixing if e(T ) ¼ {1}.
each product odometer satisfies property A. He We refer to (Aaronson 1997, Theorem 2.7.1) for
also constructed an ergodic nonsingular proof of the following Keane’s ergodic multiplier
Ergodic Theory: Nonsingular Transformations 253

theorem: Given an ergodic probability-preserving (i) Power weakly mixing rank-one

transformation S, the product transformation T transformations
S is ergodic if and only if sS(e(T)) ¼ 0, where sS (ii) Nonpower weakly mixing rank-one transfor-
denotes the measure of (reduced) maximal spec- mations with infinite ergodic index
tral type of the unitary US(see (3)). It follows that (iii) Nonpower weakly mixing rank-one transfor-
T is weakly mixing if and if only T S is ergodic mations with infinite ergodic index and such
for every ergodic probability-preserving S. While that T k1    T kr are all conservative, k1,
in the finite measure-preserving case this implies . . ., kr  ℤ,
that T T is ergodic, it was shown in (Aaronson
et al. 1979) that there exits a weakly mixing non- of types II1 and III (and various subtypes of III,
singular T with T T not conservative, hence not see section “Orbit Theory”). Thus, we have the
ergodic. In (Adams et al. 1997), a weakly mixing following scale of properties (equivalent to weak
T was constructed with T T conservative but not mixing in the probability-preserving case), where
ergodic. A nonsingular transformation T is said to every next property is strictly stronger than the
be weakly doubly ergodic (originally called dou- previous ones:
bly ergodic in (Bowles et al. 2001)) if for all sets of T is weakly mixing
positive measure A and B there exists an integer
( T is weakly doubly ergodic
n > 0 such that m(A \ T nA) > 0 and
m(A \ T nB) > 0. Furstenberg (Furstenberg (T T is ergodic
1981) showed that for finite measure-preserving (T T T is ergodic
transformations weak double ergodicity is equiva- (    ( T has infinite ergodic index
lent to weak mixing. In (Adams et al. 1997; Bowles ( T is power weakly mixing:
et al. 2001), it is shown that for nonsingular trans-
formations weak mixing does not imply weak dou- There is a rank-one weakly doubly ergodic
ble ergodicity and weak double ergodicity does not T such that T T is nonconservative (Bowles
imply that T T is ergodic. et al. 2001), and there is a rank-one weakly doubly
T is said to have ergodic index k if the Carte- ergodic T such that T T is conservative but not
sian product of k copies of T is ergodic but the ergodic (Loh and Silva 2017). We also mention an
product of k þ 1 copies of T is not ergodic. If all example of a power weakly mixing transforma-
finite Cartesian products of T are ergodic, then tion of type II1 which embeds into a rank-one
T is said to have infinite ergodic index. In a flow (Danilenko and Solomko 2009). This result
similar way, one can define the conservative was sharpened in (Danilenko and Park 2011):
index of T. Parry and Kakutani (Kakutani and There is an infinite measure-preserving rank-one
Parry 1963) constructed for each k  ℕ [ {1} flow (Rt)t  ℝ such that for each t 6¼ 0, the transfor-
an infinite Markov shift of ergodic index k. We mation Tt has infinite ergodic index. Several of
note that for each infinite Markov shift, the ergo- these notions have been studied in the context of
dic index coincides with the conservative index. nonsingular actions of locally compact groups by
We constructed in (Adams and Silva 2015; Glasner and Weiss (2016); we mention one con-
Danilenko 2016a) infinite measure-preserving dition that has not yet been discussed though only
rank-one transformations of an arbitrary ergodic in the context of transformations. A nonsingular
index k and infinite conservative index. transformation T on a probability space is said to
A stronger property is power weak mixing, be ergodic with isometric coefficients if every
which requires that for all nonzero integers k1, factor map onto an isometry of a (separable) met-
. . ., kr the product T k1    T kr is ergodic (Day ric space is constant a.e. Glasner and Weiss show
et al. 1999). The following examples were that if T T is ergodic (i.e., T is doubly ergodic),
constructed in (Adams et al. 2001; Danilenko then T is ergodic with isometric coefficients, and
2001a, 2004; Day et al. 1999): that if T is ergodic with isometric coefficients,
254 Ergodic Theory: Nonsingular Transformations

then it is weakly mixing. In (Loh and Silva 1

n1

2017), it is shown that if T is weakly doubly lim

n!1 an ðFÞ
jm A \ T k B  mðAÞmðBÞuk ðFÞj ¼ 0,
k¼0
ergodic, then it is ergodic with isometric coeffi-
ð5Þ
cients. In (Glasner and Weiss 2016), there is an
example of a T that is ergodic with isometric
where uk(F) and an(F) are defined as above. When
coefficients but T T is not conservative,
m(X) ¼ 1 and we let F ¼ X, then an(F) ¼ n and (5)
hence not ergodic. In (Haddock et al. 2022), it
becomes the condition equivalent to the weak
is shown that isometric coefficients do not imply
mixing property for a finite measure-preserving
weak doubly ergodic. Further conditions related
transformation. In infinite measure, however, the
to weak mixing (in the case of infinite measure-
rational weak mixing condition is not equivalent to
preserving transformations) are discussed in the
weak mixing as we shall see. If in (5) we drop the
survey (Adams and Silva 2018).
absolute values, then this condition defines the
notion of weak rational ergodicity (Aaronson
Rational Ergodicity and Rational Weak Mixing 1979). Then Theorem 6.1 claims that rational ergo-
Let T be a conservative, ergodic measure- dicity implies weak rational ergodicity. If in (5) the
preserving transformation of a s-finite measure sequence (n) is replaced by a subsequence (ni), we
space (X, ℬ, m). For a function f : X ! X, let say T is subsequence rational weak mixing. Subse-
Sn ð f Þ ¼ n1
k¼0 f ∘T : T be called rationally ergo-
k
quence weak rationally ergodic is defined in a
dic (Aaronson 1977a) if there is a subset F  ℬ, similar way. The transformation T is boundedly
0 < m(F) < 1, satisfying a Renyi inequality, i.e., rationally ergodic (Aaronson 1979) if
there exists a constant M > 0 such that for all
n  1, a n ð FÞ
sup < 1,
n1 Sn ðF Þ 1
2
ðSn ðF ÞÞ2 dm  M Sn ðF Þdm :
F F The notions of subsequence boundedly ratio-
nally ergodic and subsequence rationally ergodic
For a related concept of p-rational ergodicity are defined when the sequence (n) is replaced by a
(i.e., the rational ergodicity in L p), see Aaronson subsequence (ni). It can be seen from the defini-
and Weiss (2018). We now set tion that bounded rational ergodicity implies ratio-
nal ergodicity (and similarly for the subsequence
m F \ T k F n1 versions). Aaronson (Aaronson 1979) showed
uk ðFÞ ≔ and an ðFÞ ≔ uk ðFÞ:
mðFÞ2 k¼0
that rational ergodicity does not imply bounded
rational ergodicity, and more recently it was
shown by Adams and Silva (Adams and Silva
Theorem 6.1 If T is rationally ergodic and
2016) that weak rational ergodicity does not
F satisfies the Renyi inequality, then for all mea-
imply rational ergodicity.
surable sets A and B contained in F (Aaronson
The following theorem was proved in
1977a),
(Aaronson 1979) for the weakly rationally ergodic
n1 transformations. The same argument works in the
1
lim m A \ T k B ¼ mðAÞmðBÞ: more general case.
n!1 an ðFÞ
k¼0
Theorem 6.2 Each subsequence weakly ratio-
An ergodic conservative transformation T is nally ergodic transformation T of (X, m) is non-
rationally weakly mixing (Aaronson 2013) if squashable, i.e., each nonsingular transformation
there exists a measurable set F of positive finite commuting with T preserves m.
measure such that for all measurable sets A and There are several examples of rationally ergo-
B contained in F we have dic transformations which are infinite Markov
Ergodic Theory: Nonsingular Transformations 255

shifts; see Aaronson (1997). More recently, it was transformation that is not rationally weakly
shown in Dai et al. (2015) and Bozgan et al. mixing. We also mention an example of a weakly
(2015) that rank-one (infinite measure-preserving) mixing, rationally ergodic, and Koopman mixing
transformations are subsequence boundedly ratio- (or zero type, see §6.3 for the definition) transfor-
nally ergodic. The first version of Bozgan et al. mation that is not subsequence rationally weakly
(2015) has a proof that the rank-one transforma- mixing (Aaronson 2016).
tions are subsequence weakly rationally ergodic; a The set of transformations that are subse-
simpler proof was found in Danilenko (2016b), quence rationally weakly mixing is residual
where this property is also established for the class (Aaronson 2013), while the set of rationally
of funny rank one transformations and the class of weakly mixing transformations is meagre
ergodic transformations of balanced finite rank. (Aaronson 2013). Since the set of power weakly
(A transformation is called of balanced finite rank mixing rank-one transformations is residual, and
if it is of finite rank and the bases of the Rokhlin the rank-one transformations are subsequence
towers on the n-th step of the cutting-and-stacking boundedly rationally ergodic, there exist rank
inductive construction have asymptotically com- one transformations that are power weakly mixing
parable measures as n ! 1.) Therefore all these and subsequence boundedly rationally ergodic but
transformations are nonsquashable in view of not rationally weakly mixing.
Theorem 6.2.
The rank-one transformations for which the Mixing, Zero-Type
sequence of cuts ðr n Þ1
n¼1 is bounded are boundedly
We now consider several attempts to define
rationally ergodic (Aaronson et al. 2017; Dai et al. (strong) mixing for nonsingular maps. Probably
2015; Bozgan et al. 2015). As for the examples of the first notion of mixing for infinite measure-
rationally weakly mixing transformations, preserving systems was proposed by Hopf in
Aaronson (2013) shows that Markov shifts with (1937). The idea was to show an asymptotic rate
certain conditions on their associated renewal for the sequence m(A \ TnB) for a large class of
sequences are rationally weakly mixing, and Dai finite measure sets A, B. More precisely, a trans-
et al. (2015) give rank-one examples. Subse- formation T is mixing for a ring R (called now
quence rational weak mixing and rational weak Krickeberg mixing), where R is a ring of sets of
mixing for products of powers have been studied finite measure that is invariant under T and gener-
in Aaronson (2013) and Adams (2015). ates the entire s-algebra, if there is a sequence
We have the following implications for rational ðrn Þ1
n¼1 such that for all A, B  R we have

weak mixing.
lim r mðA \ T n BÞ ¼ mðAÞmðBÞ:
n!1 n
Theorem 6.3 If a transformation is sequentially
rationally weakly mixing, then it is weakly mixing Hopf proved such a property for an infinite
(Aaronson 2013). measure-preserving transformation defined on
ℝ+ [0, 1] that is now called an infinite random
Theorem 6.4 If a transformation is rationally walk; with R being the ring of Riemann measur-
weakly mixing, then it is weakly doubly ergodic able subsets. If R is the ring of all subsets of finite
(Bozgan et al. 2015). measure, then there are no Krickerberg R - mixing
It is an open problem whether weak double transformations because of the existence of
ergodicity implies rational weak mixing. weakly wandering sets. We note that the above
Aaronson (2013) asked if weak rational ergodicity (purely measure theoretical) definition R - mixing
and weak mixing imply rational weak mixing. is due to Friedman (1978) who extended
This was answered in negative in (Dai et al. Krickeberg’s one (Krickeberg 1967) given for
2015), where was constructed an example of a continuous transformations of topological spaces
weakly mixing rationally ergodic rank-one endowed with a measure. Recently, there have
256 Ergodic Theory: Nonsingular Transformations

been several works showing this version of subsets A1, . . ., Ak of finite measure whenever
mixing and computing mixing rates for several |ni  nj| ! 1 if i 6¼ j.
transformations. Melbourne and Terhesiu For 0  α  1, Kakutani suggested a related
(Melbourne and Terhesiu 2012) have verified definition of α-type: An infinite measure-
mixing for a large class of maps including AFN preserving transformation is of α-type if
maps with indifferent fixed points; these methods lim supn!1 m(A \ TnA) ¼ αm(A) for every subset
were extended to invertible transformations by A of finite measure. In (Osikawa and Hamachi
Melbourne (2015) and to additional maps by 1971), examples of ergodic transformations of
Gouëzel (2011). Recently, Dolgopyat and any α-type and a transformation of not any type
Nándori (2019) have shown Krickerberg mixing were constructed.
for a class of special flows; other recent work It was shown in Danilenko (2016a) and Loh
appeared in Bruin et al. (2019). et al. (2018) that for each pair k  n, there exists
Another approach to mixing was proposed by a mixing rank-one infinite measure-preserving
Krengel and Sucheston (1969) for nonsingular transformation of ergodic index k and conservative
maps. Given a sequence of measurable sets index n. Rigid infinite measure-preserving rank-
{An}, let sk({An}) denote the s-algebra generated one transformations of arbitrary ergodic index
by Ak, Akþ1, . . .. A sequence {An} is said to be were constructed in Danilenko (2016a). Of course,
remotely trivial if \1 k¼0 sk ðfAn gÞ ¼ f;, X g modm, rigidity implies infinite conservative index.
and it is semiremotely trivial if every subsequence We now isolate an important class of concrete
contains a further subsequence that is remotely rank-one transformations and examine mixing
trivial. A nonsingular transformation T of a properties within this class. Let T be a rank-one
s-finite measure space is called mixing if for transformation associated with a sequence
every set A of finite measure the sequence ðr n , wn , sn Þ1
n¼1. If wn(0) ¼ wn(1) ¼    ¼ wn(rn  1)
{T nA} is semiremotely trivial, and completely and sn( j) ¼ zn þ j for j ¼ 0, . . ., rn1, then T is
mixing if {T nA} is semiremotely trivial for all called a high staircase (called also tower staircase
measurable sets A. Krengel and Sucheston show in Bowles et al. (2001). It was shown in Bowles
that T is completely mixing if and only if it is type et al. (2001) that each high staircase is weakly
II1 and mixing for the equivalent finite invariant doubly ergodic and hence weak mixing. However,
measure. Thus, there are no type III and II1 there exist high staircases whose Cartesian square
completely mixing nonsingular transformations is not ergodic (Bowles et al. 2001). As for the
on probability spaces. We note that this definition mixing of the high staircases, the following theo-
of mixing in infinite measure spaces depends on rem was proved in (Danilenko and Ryzhikov
the choice of measure inside the equivalence class 2011). It is an infinite analogue of the Adams
(but it is independent if we replace the measure by solution (Adams 1998) of the Smorodinsky
an equivalent measure with the same collection of conjecture.
sets of finite measure).
Hajian and Kakutani showed (1964) that an r2
Theorem 6.5 If lim n!1 r1 rnn1 ¼ 0 and
ergodic infinite measure-preserving transforma- 1 zn
¼ 1, then the associated high staircase
tion Tis either of zero type: limn!1 m(T nA \ A) ¼ n¼1 hn

0 for all sets A of finite measure, or of positive is infinite measure-preserving and mixing.
type: lim supn!1 m(TnA \ A) > 0 for all subsets Mixing high staircases which are power
A of finite positive measure. It appears that T is weakly mixing were constructed in (Danilenko
mixing if and only if it is of zero type (Krengel and and Ryzhikov 2011).
Sucheston 1969). We note that in infinite measure, We note that mixing (zero type) does not imply
mixing implies mixing of all orders, i.e., if a either ergodicity or conservativity in the category
measure-preserving T is of zero type, then of infinite measure-preserving transformations.
mðT n1 A1 \    \ T nk Ak Þ ! 0 for each k and all Indeed, a translation on ℝ endowed with the
Ergodic Theory: Nonsingular Transformations 257

Lebesgue measure is nonergodic, totally dissipa- maps that is motivated by statistical mechanics
tive but of zero type. It may seem that mixing and and uses global observables. The definition is
ergodicity together are stronger than any kind of with respect to a collection of sets, global observ-
nonsingular weak mixing considered above. ables, and local observables. We choose a family
However, it is not the case: If T is a weakly mixing V of measurable sets of finite measure so that it
infinite measure-preserving transformation of contains sets V1  V2     such that [iVi ¼ X. We
zero type and S is an ergodic probability- also have a subspace G of L1 functions (called
preserving transformation, then T S is ergodic global observables), and a subspace ℒ of L1 func-
and of zero type. On the other hand, the L1- tions (called local observables). There is also a
spectrum e(T S) is nontrivial, i.e., T S is not condition on the growth rate of the measure of V -
weakly mixing, whenever S is not weakly mixing. elements under iteration by T. Then Lenci defines
We also note that there exist rank-one infinite an infinite volume average for elements F of G by
measure-preserving transformations T of zero
type such that T T is not conservative (hence
mðFÞ ¼ lim F dm:
not ergodic) (Adams et al. 1997). In contrast to V!X V
that, if T is of positive type, then all of its finite
Cartesian products are conservative (Aaronson By this limit, we mean that for every neighbor-
and Nakada 2000). Another result that suggests hood of mðFÞ, there is a number M > 0 so that
that there is no good definition of mixing in the when m(V ) > M for a set V in V , then VFdm is in
infinite measure-preserving case was proved in the neighborhood. He shows that under the above
(James et al. 2008). It is shown there that while conditions, mðF∘T n Þ ¼ mðFÞ: Then he defines
the mixing finite measure-preserving transforma- several notions of what he calls infinite volume
tions are measurably sensitive, there exists no mixing (Lenci 2010); we mention three here. The
infinite measure-preserving system that is measur- transformation T is said to be global-local mixing-
ably sensitive. (Measurable sensitivity is a mea- 1 if for all F in G and all g in ℒ with gdm ¼ 0, we
surable version of the strong sensitive dependence have
on initial conditions – a concept from topological
theory of chaos.)
lim ðF∘T n Þg dm ¼ 0:
The Krengel-Sucheston concept of mixing n!1

(or the Hajian-Kakutani zero type) considered

above for infinite measure-preserving systems The transformation T is said to be global-local
extends naturally to nonsingular transformations mixing-2 if for all F in G and all g in ℒ we have
T of (X, ℬ, m) without finite absolutely continu-
ous invariant measure in the following way: We lim ðF∘T n Þg dm ¼ mðFÞ g dm:
n!1
say that T is Koopman mixing (or of zero type) if
the maximal spectral type of the Koopman oper-
ator UT generated by T (see (3)) is a Rajchman The transformation is said to be global-global
measure, i.e., X f  U nT fdm ! 0 for each mixing if for all F, G in G we have
f  L2(X, m). It is easy to see that this definition
of mixing will not be affected if we replace m with lim mðF ∘ T n GÞdm ¼ mðFÞmðGÞ
n!1
an equivalent measure. Examples of Koopman
mixing rank-one transformations of type III were Lenci proves in (Lenci 2013) that if T is an
constructed in (Danilenko 2001a). Nonsingular infinite measure-preserving K-automorphism,
Bernoulli transformations of type III are always then T is global-local mixing-1 for any choice of
Koopman mixing (Kosloff 2013). V satisfying the measure growth condition,
More recently, Lenci (2010) introduced a new for ℒ ¼ L1, and for C that is the closure in L1 of
notion of mixing for infinite measure-preserving T nF , where F is as in the definition of the
258 Ergodic Theory: Nonsingular Transformations

K-automorphism (see subsection “K-automor- Multiple and Polynomial Recurrence

phism”). Infinite mixing has been shown for Let p be a positive integer. A nonsingular trans-
other examples (Lenci 2010), in particular for formation T is called p-recurrent if for every sub-
uniformly expanding maps of the interval (Lenci set B of positive measure there exists a positive
2017), and for one-dimensional maps with an integer k such that
indifferent fixed point (Bonanno et al. 2018).
m B \ T k B \    \ T kp B > 0:
K-property
A nonsingular transformation T of (X, ℬ, m) is If T is p-recurrent for any p > 0, then it is called
called K-automorphism (Silva and Thieullen 1995) multiply recurrent. It is easy to see that T is
if there exists a sub-s-algebra F  ℬ such that 1-recurrent if and only if it is conservative.
T 1 F  F , \k0 T k F ¼ f;, Xg, _þ1 k¼0 T F ¼ ℬ
k
Clearly, if T is rigid, then it is multiply recurrent.
and the Radon-Nikodym derivative om1 is Furnstenberg showed (Furstenberg 1981) that
F -measurable (see also Parry 1965) for the case every finite measure-preserving transformation is
when T is of type II1; the authors in (Silva and multiply recurrent. In contrast to that, Eigen,
Thieullen 1995) required T to be conservative). If Hajian, and Halverson (Eigen et al. 1998a)
R is a nonsingular endomorphism on (X, ℬ, m), constructed for any p  ℕ [ {1}, a nonsingular
then the natural extension of R is a K-automorphism product odometer of type II1 which is p-recurrent
n
if and only if R is exact, i.e., ^1 n¼1 R ℬ ¼ but not ( p þ 1)-recurrent. Aaronson and Nakada
f;, Xg mod0: It follows from the Kolmogorov showed in (2000) that an infinite measure-
0–1 theorem that a nonsingular Bernoulli shift preserving Markov shift T is p-recurrent if and
from the generalized Krengel class (see § 3.5) is a only if the product T    T ( p times) is
K-automorphism. Parry (Parry 1965) showed that a conservative. It follows from this and (Aaronson
type II1 K-automorphism is either dissipative or et al. 1979) that in the class of ergodic Markov
ergodic. Krengel (1970) proved the same for the shifts, infinite ergodic index implies multiple
Krengel class of Bernoulli nonsingular shifts, and recurrence. However, in general this is not true.
finally Silva and Thieullen extended this result to It was shown in (Adams et al. 2001; Gruher et al.
the nonsingular K-automorphisms (Silva and 2003; Danilenko and Silva 2004) that for each
Thieullen 1995). It is also shown in (Silva and p  ℕ [ {1} there exist
Thieullen 1995) that if T is a nonsingular
K-automorphism, for any ergodic nonsingular (i) Power weakly mixing rank-one
transformation S, if S T is conservative, then it transformations
is ergodic. This property is called sharply weak (ii) Nonpower weakly mixing rank-one transfor-
mixing in (Danilenko and Lemańczyk 2022). It mations with infinite ergodic index
follows that a conservative nonsingular
K-automorphism is weakly mixing. However, it which are p-recurrent but not ( p þ 1)-recurrent
does not necessarily have infinite ergodic index (the latter holds when p 6¼ 1, of course).
(Kakutani and Parry 1963). Krengel and Sucheston A subset A is called p-wandering if
(1969) showed that an infinite measure-preserving m(A \ TkA \    \ T pkA) ¼ 0 for each k. Aaronson
conservative K-automorphism is mixing. In (Elyze and Nakada established in (Aaronson and Nakada
et al. 2018), it is shown that if T is a rank-one 2000) a p-analogue of Hopf decomposition (see
transformation with (rn) bounded and sn1(rn1  1) Theorem 2.2).
 hn/2 for all n  1 (so the space is of infinite
measure, and T is the infinite Chacón map), there Proposition 6.6 If (X, ℬ, m, T ) is conservative
exists a rank-one transformation S such that T S aperiodic nonsingular dynamical system and
is conservative but not ergodic, so it is not sharply p  ℕ, then X ¼ Cp t Dp, where Cp and Dp are
weak mixing. T-invariant disjoint subsets, Dp is a countable
Ergodic Theory: Nonsingular Transformations 259

+
union of p-wandering sets, T Cp. is p-recurrent, equivalent invariant probability measure for each
and 1 k
k¼1 m B \ T B \    \ T
dk
B ¼ 1 for n  ℕ. Assume that the transformation T ¼
every B  Cp. 1n¼1 T n is nonsingular with respect to the infinite
Let T be an infinite measure-preserving trans- product measure m ≔ 1 n¼1 mn : In other words,
formation and let F be a s-finite factor (X, m, T ) is an IDPFT dynamical system.
(i.e., invariant subalgebra) of T. Inoue (Inoue
2004) showed that for each p > 0, if T F is Theorem 7.1 The transformation T is either con-

+
p-recurrent, then so is T provided that the exten- servative or totally dissipative. Moreover, if S is
sion T ! T F is isometric. It is unknown yet an ergodic conservative nonsingular transforma-
+

whether the latter assumption can be dropped. tion, then the direct product T S is either con-
However, partial progress was achieved in servative or totally dissipative (Danilenko and
(Meyerovitch 2007): If T F is multiply recur- Lemańczyk 2019).
+

rent, then so is T.
Let P ≔ fq  ℚ½t j qðℤÞ  ℤ and q(0) ¼ 0}. Theorem 7.2 Let (Xn, nn, Tn) be mildly mixing
An ergodic conservative nonsingular transforma- (see §15.1 for the definition) for each n > 0. If T is
tion T is called p-polynomially recurrent if for m-conservative, then T is sharply weak mixing
every q1 , . . . , qp  P and every subset B of posi- (Danilenko and Lemańczyk 2019).
tive measure there exists k  ℕ with Examples of rigid ergodic but not weakly
mixing IDPFT transformations of Krieger type
IIIl, for each l  (0, 1) were constructed in
m B \ T q1 ðkÞ B \    \ T qp ðkÞ B > 0:
(Danilenko and Lemańczyk 2019). Some families
of 0-type IDPFT transformations of type III1
If T is p-polynomially recurrent for every appeared in (Danilenko and Lemańczyk 2019)
p  ℕ, then it is called polynomially recurrent. and of all possible Krieger types in (Danilenko
Furstenberg’s theorem on multiple recurrence was and Kosloff 2022).
significantly strengthened in (Bergelson and
Leibman 1996), where it was shown that every Theorem 7.3 Let K  {IIIl | 0  l  1} t {II1}.
finite measure-preserving transformation is poly- Then there is a 0-type weakly mixing IDPFT trans-
nomially recurrent. However, Danilenko and formation of type K (Danilenko and Kosloff 2022).
Silva (Danilenko and Silva 2004) constructed

(i) Type II1 transformations T which are Dynamical Properties of Nonsingular

p-polynomially recurrent but not ( p þ 1)- Bernoulli and Markov Shifts
polynomially recurrent (for each fixed p  ℕ)
(ii) Polynomially recurrent transformations T of We will use below the notation introduced in §3.5.
type II1, Thus, T stands for the nonsingular Bernoulli shift on
(iii) Rigid (and hence multiply recurrent) type the space (X, m) ¼ (Aℤ, n  ℤ mn) associated with
II1 transformations T which are not poly- a sequence of probability measures (mn)n  ℤ on A.
nomially recurrent
Krengel Class
Moreover, such T can be chosen inside the Nonsingular Bernoulli shifts appeared originally
class of rank-one transformations with infinite in Krengel’s work (Krengel 1970). He introduced
ergodic index. there a class of nonsingular shifts for which A ¼
{0, 1} and mn is the equidistribution on {0, 1} for
all n  1. We will call it the Krengel class. Krengel
Dynamical Properties of IDPFT Systems showed that this class contains totally dissipative
transformations. He also used an inductive proce-
Let Tn be an ergodic transformation of a standard dure to construct the sequence ðmn Þ1 n¼1 in such a
probability space (Xn, mn, Tn), and let nn be a mn - way that the corresponding Bernoulli shift is
260 Ergodic Theory: Nonsingular Transformations

ergodic conservative and not of type II1. Krengel class is never of type IIIl, 0  l < 1. In (Vaes
conjectured that the shift is of type III indeed. In and Wahl 2018), Vaes and Wahl, answering a
(Hamachi 1981b), Hamachi showed that question from (Danilenko and Lemańczyk
Krengel’s class contains ergodic conservative 2019), found a convenient condition for a non-
nonsingular Bernoulli shifts of type III. This was singular Bernoulli shift from the generalized
further refined by Kosloff who constructed type Krengel class to be conservative. Utilizing that
III1 ergodic conservative shifts belonging to condition, they constructed, for each l  (0, 1),
Krengel’s class (Kosloff 2011). In (Kosloff an explicit example of power weakly mixing non-
2013), Kosloff constructed a nonsingular singular Bernoulli shift of type III1 with m1(0) ¼ l.
Bernoulli shift of type III1 (and belonging to the We note that the previously known Bernoulli
Krengel class) which is power weakly mixing. shifts of type III1 were constructed via involved
Weiss asked about possible Krieger’s types for inductive procedures. Vaes and Wahl also pro-
the nonsingular Bernoulli shifts. Answering his vided in (Vaes and Wahl 2018) a family of type
question, Kosloff proved in a subsequent paper III1-examples of Bernoulli shifts that contains
(Kosloff 2014) that each conservative Bernoulli examples with finite ergodic index (less
shift from the Krengel class is ergodic and either than 73). Analyzing that family more thoroughly,
of type II1 or of type III1. In particular, the non-- Kosloff and Soo showed that it contains type III1
type-II1 conservative Bernoulli shifts constructed nonsingular Bernoulli shifts of arbitrary ergodic
in (Krengel 1970; Hamachi 1981b; Kosloff index.
2011)) are all of type III1 indeed.
Theorem 8.2 Let A ¼ {0, 1} and c > 0. Let
Generalized Krengel Class mcn ð0Þ≔ 12 þ pcn 1fn  ℕ jpn>2cg for each n  ℤ.
Kosloff’s result from (Kosloff 2014) was further
There exists D > 16 such that the Bernoulli shift
extended in (Danilenko and Lemańczyk 2019).
T on n  ℤ(A, mn) is ergodic of type III1 for all
We say that a nonsingular Bernoulli shift belongs
c < D and totally dissipative for all c > D. In
to the generalized Krengel class if A ¼ {0, 1} and
addition, if k  ℕ and pkþ1
D
 c < pDk then T is of
mn ¼ m1 for each n  0. We note that these trans-
ergodic index k (Kosloff and Soo 2022).
formations are the natural extension of the one-
sided nonsingular Bernoulli shifts defined on
(Aℕ, n>0 mn). Every shift from the generalized General Nonsingular Bernoulli Shifts
Krengel class is a K-automorphism. The studying of general nonsingular Bernoulli
shifts was initiated by Kosloff in (2013).
Theorem 8.1 (On types of nonsingular Bernoulli
shifts from the generalized Krengel class (Kosloff Theorem 8.3 (Mixing of nonsingular Bernoulli
2014; Danilenko and Lemańczyk 2019)). Let shifts (Kosloff 2013)). If #A ¼ 2 and (mn)n  ℤ is a
A ¼ {0, 1} and let T be a nonsingular Bernoulli sequence of probabilities on A such that (4) holds,
shift on (Aℤ, n  ℤ mn) from the generalized then T is either of type II1 and mixing (with respect
Krengel class. to the equivalent invariant probability measure)
or of zero type.
(i) If n>0(mn(0)  m1(0))2 < 1, then m is equiv- In (Kosloff 2019), Kosloff noticed that under
alent to n  ℤm1 and hence T is of type II1. some natural conditions, conservativity of
(ii) If n>0(mn(0)  m1(0))2 ¼ 1 and T is con- Bernoulli shifts implies ergodicity. His proof was
servative, then T is ergodic of type III1. More- based essentially on the Hurewicz ergodic theorem
over, the Maharam extension of T is a weakly and properties of the tail equivalence relation on
mixing K-automorphism. Aℤ. Danilenko (Danilenko 2019a) refined his
results by exploiting the interplay between T and
Thus, Krieger’s type of each nonsingular the measurable equivalence relation on Aℤ gener-
Bernoulli shift from the generalized Krengel ated by the finite permutations of coordinates.
Ergodic Theory: Nonsingular Transformations 261

Theorem 8.4 (Weak mixing of conservative each l  (0, 1), an example of type IIIl Bernoulli
nonsingular Bernoulli shifts). Let A be finite. shift was constructed with mn Leb for each n.
Examples of nonsingular Bernoulli shifts of each
(i) If infn  ℤ mina  Amn(a) > 0 and T is conser- possible Krieger’s type were given in a later paper
vative, then T is weakly mixing (see Kosloff (Berendschot and Vaes 2022a). An alternative
2019; Danilenko 2019a). proof of this result appeared in a recent work
(ii) If #A ¼ 2 and inf n  ℤ min a  A logjmmn ðaðaÞ Þj > (Danilenko and Kosloff 2022).
nþ1

0 and T is conservative, then T is weakly

mixing (Danilenko 2019a). Theorem 8.6 Let K  {IIIl | 0  l  1} t {II1}.
(iii) Under condition (i) or (ii), if T  T Then there is a sequence (mn)n  ℤ of probability
(p times) is conservative for some p  1, then measures on [0, 1] such that mn Leb for each
T  T (p times) is weakly mixing n  ℕ, the Bernoulli shift T on ([0, 1]ℤ, n  ℤmn)
(Danilenko 2019a). is weakly mixing, IDPFT, and of Krieger type K
(Danilenko and Kosloff 2022).
The techniques from (Kosloff 2019; Danilenko We also note that if there is C > 1 such that
2019a) were further elaborated in (Björklund et al. C1  dm
dm  C for all n  ℤ, then T can be neither
n
0
2021) to prove the following refinement of
of type III0 nor of type II1 (Berendschot and Vaes
Theorem 8.4.
2022a).
The following result is an analog of Theorem
Theorem 8.5 Let A ¼ {0, 1} and let T be a
8.5 for Bernoulli shifts with general state spaces.
nonsingular Bernoulli shift on (Aℤ, n  ℤ mn)
and let m ¼ n  ℤ mn be nonatomic. If T is not
totally dissipative, then T is weakly mixing and its Theorem 8.7 Let (mn)n  ℤ be a sequence of
type is given as follows. mutually equivalent probabilities on a stan-
dard Borel space A and let T be the Bernoulli
(i) If there is l  (0, 1) such that n  ℤ(mn(0)  shift on (X, m) ¼ n  ℤ(A, mn). If T is non-
l)2 < þ 1, then T is of type II1. singular and conservative, then T is weakly
(ii) If there is limn!1mn(0) ¼ l  (0, 1) and mixing and the following holds (Berendschot
n  ℤ(mn(0)  l)2 ¼ þ 1, then T is of type and Vaes 2022b).
III1.
(iii) If either limn!1mn(0) ¼ 0 or limn!1mn(0) ¼ (i) T of type II1 if and only if there exists a
1, then T is of type III. probability measure n m0 such that nZ m.
(iv) If the sequence (mn(0))n does not converge as (ii) T of type II1 if and only if there exists a
n ! 1, then T is of type III1. s-finite measure n m0 and Borel subsets
Bn  A such that n(Bn) < 1 for each
It follows, in particular, that the nonsingular n  ℤ and
Bernoulli shifts on {0, 1}ℤ are never of type II1.
It is still an open problem which subtypes of III are
mn ðA∖Bn Þ < 1, nn ðA∖Bn Þ ¼ 1,
realized in the case (iii). An example of T of type
nℤ nℤ
III1 with limn!1mn(0) ¼ 0 was constructed in
H 2 mn , nðBn Þ1 n Bn
+

(Berendschot and Vaes 2022a). Is there T of type

III0 satisfying (iii)? nℤ

Parts (i) and (ii) of Theorem 8.5 were general-

ized to any finite set A in Avraham-Re’em (2022, (iii) T is of type III in all other cases.
Remark 8.7).
The situation is different if we consider A ¼ [0, The next very natural question to ask is: Which
1] and the probability measures (mn)n  ℤ on A with ergodic flows arise as the associated flow of non-
infinite support. In (Kosloff and Soo 2021), for singular Bernoulli shifts of type III0?
262 Ergodic Theory: Nonsingular Transformations

Definition 8.8 Let V ¼ (Vt)t  ℝ be a nonsingular • Whether each Poisson flow is an AT-flow asso-
flow on a standard probability space (X, n). Given ciated with an ITPFI2-transformation.
n > 1, consider two mutually commuting non- • Whether each infinitely divisible flow is
singular flows U ¼ (Ut)t  ℝ and D ¼ (Dt)t  ℝ on Poisson, whether each infinitely divisible
the product space (X, n)n: AT-flow is Poisson.

U t ðx1 , . . . , xn Þ ≔ ðV t x1 , x2 , . . . , xn Þ, Bernoulli Factors of Nonsingular Bernoulli

Dt ðx1 , . . . , xn Þ ≔ ðV t x1 , V t x2 , . . . , V t xn Þ, Shifts
Definition 8.10 Let T be a nonsingular Bernoulli
for each t  ℝ and (x1, . . ., xn)  X n. The restric- shift on (X, m) ≔ n  ℤ(A, mn). Suppose that there
tion of U to the s-algebra of D-invariant subsets in is another sequence (nn)n  ℤ of probabilities on
X n is a well-defined nonsingular flow. It is called A such that T is nonsingular on the product space
the joint flow of n copies of V. The flow V is called (X, n) ≔ n  ℤ(A, nn). If there is a map
infinitely divisible if for each n > 1, there is a flow π : Aℤ ! Aℤ such that πT ¼ Tπ and the measure
W such that V is isomorphic to the joint flow of m ∘ π1 is equivalent to v, then (X, v, T ) is called a
n copies of W. factor of (X, m, T ).
If V is an AT-flow associated with an ITPFI2- We assume that the factors are nontrivial. What
transformation (see section “ITPFI Transforma- is the relation between Krieger’s types of (X, v, T )
tions and AT-Flows”), then V is infinitely divisible. and (X, m, T)?

Theorem 8.11 (Bernoulli factors of type II1

Theorem 8.9
(Kosloff and Soo 2022) Let A be finite and
(i) Let (mn)n  ℤ be a sequence of mutually equiv-
alent probabilities on a standard Borel
1 > inf min mn ðaÞ > 0:
space A, and let T be the Bernoulli shift on nℤ aA
n  ℤ(A, mn). If T is nonsingular and conser-
vative, then the associated flow of T is infinitely Then there is a probability measure r on
divisible. For each ergodic probability- A such that (X, rℤ, T ) is a factor of (X, m, T ).
preserving transformation S, the product T
S is ergodic and the associated flows of T and Theorem 8.12 Let l, l0  (0, 1]. There exists a
T S are isomorphic (Berendschot and Vaes type IIIl Bernoulli shift which has a type III l0
2022b). Bernoulli shift as a factor in each of the two
(ii) Conversely, let V be an AT-flow associated cases (Kosloff and Soo 2022):
with an ITPFI2-transformation. Then there is
a weakly mixing nonsingular Bernoulli shift (i) l0 < l ¼ 1,
whose associated flow is isomorphic to V. (ii) l < l0 :

We note that formally Berendschot and Vaes

proved in (Berendschot and Vaes 2022b) a stron- Nonsingular Markov Shifts
ger claim than (ii). They introduced a concept of An analog of Theorem 8.4(i) holds also for non-
Poisson flow as the tail boundary of certain non- singular Markov shifts (see (Kosloff 2019;
homogeneous random walks on ℝ (see §15.7 Danilenko 2019a)).
below). Every Poisson flow is infinitely divisible.
Every AT-flow associated with an ITPFI2- Theorem 8.13 (Weak mixing of conservative
transformation is Poisson. It is shown in nonsingular Markov shifts (Kosloff 2019;
(Berendschot and Vaes 2022b) that Theorem 8.9 Danilenko 2019a)). Let A be finite, and let M ¼
(ii) holds for each Poisson flow V. However, the (M(a, b))a, b  A be a 0–1-valued A A-matrix.
following problems remain open: Suppose that M is primitive, i.e., there is n > 0
Ergodic Theory: Nonsingular Transformations 263

such that all the entries of Mn are strictly positive. Theorem 8.15 Let A be finite and let M ¼
Let (XM, T, m) be a nonsingular Markov shift, and (M(a, b))a,b  A be a primitive 0–1-valued A A-
let m be generated by a sequence (πn, Pn)n  ℤ as in matrix. Let a measure m on XM be generated by a
§3.6. Suppose that m is nonatomic and that πn is sequence (πn, Pn)n  ℤ as in §3.6 and inf
fully supported on A for each n. If inf {Pn(a, b) | n  ℤ, M(a, b) ¼ 1} > 0. Let the
{Pn(a, b) | n  ℤ, M(a, b) ¼ 1} > 0 and T is Markov shift (XM, T, m) be nonsingular and con-
conservative, then T is weakly mixing. servative (Avraham-Re’em 2022).
We isolate a class of nonsingular Markov shifts
for which Pn ¼ P1 and πn ¼ π1 for all n  0 and (i) If limn!1Pn does not exist, then T is of type
call it the Markov-Krengel class. Each shift from III1.
this class is the natural extension of the (ii) If there exists the limit P+ ≔ limn!þ1 Pn and
corresponding one-sided nonsingular Markov P ≔ limn!1Pn, then P+ ¼ P.
shift (Danilenko and Lemańczyk 2019). There is (iii) If A ¼ {0, 1} and there exists the limit
an analog of Theorem 8.1 for the Markov-Krengel Q ≔ limn!1Pn, then T is either of type II1
shifts. or III1. More precisely, T is of type II1 if and
only if
Theorem 8.14 (Danilenko and Lemańczyk
1 1 2
2019). Let M ¼ , Pn be a bistochastic
1 1 Pn ða, bÞPn ða0 , b0 Þ  Qða, bÞQða0 , b0 Þ
n1 a, b, a0 , b0  A
0:5 0:5
matrix for each n  ℤ, Pk ¼ and < 1:
0:5 0:5
πk ¼ (0.5,0.5) for each k  0. Let the The corresponding equivalent invariant prob-
corresponding Markov-Krengel shift ðXM , T, mÞ ability measure (if exists) is the Markov measure
be nonsingular and conservative. Then either T is defined by Q and the distribution l on A satisfying
of type II1 (if n>0 j Pn(0, 0)  0.5 j < 1 ) or lQ ¼ l.
III1 (otherwise). In the latter case, the Maharam It was also shown in (Avraham-Re’em 2022)
extension of T is a weakly mixing that an analogue of Theorem 8.15(iii) holds also
K-automorphism. Moreover, if m is equivalent to for the golden mean Markov shift for which A ¼
a Bernoulli (i.e., infinite product) measure, then 1 0 1
T is of type II1.
{0, 1, 2} and M ¼ 1 0 1 :
Concrete examples of Markov-Krengel shifts
(XM, m, T ) of type III1 such that m is not equiva- 0 1 0
lent to a Bernoulli measure were constructed in
(Danilenko and Lemańczyk 2019; Kosloff 2021).
Recently, Avraham-Re’em extended and Dynamical Properties of Nonsingular
refined the aforementioned results on Markov Poisson Suspensions and Nonsingular
shifts (Avraham-Re’em 2022). To state his results, Gaussian Transformations
we introduce some notation. Given n  ℤ and a,
b  A, we let Nonsingular Poisson Suspensions
Let (X, ℬ, m) be a s-finite infinite standard mea-
pn1 ðbÞ sure space and let T  Aut2(X, m). Then the
P ðb, aÞ if pn ðaÞ 6¼ 0, Poisson suspension (X , m , T ) is well defined
Pn ða, bÞ ¼ pn ðaÞ n1
0 otherwise: by Theorem 3.1. The first problem to consider is
to find out when T admits an absolutely continu-
For a stochastic A A-matrix Q, let l be the ous invariant probability measure. A satisfactory
distribution on A such that lQ ¼ l. We let solution of this problem is obtained in (Danilenko
lðbÞ et al. 2022a).
Qða, bÞ ≔ lðaÞ Qðb, aÞ:
264 Ergodic Theory: Nonsingular Transformations

Theorem 9.1 The following are equivalent: Theorem 9.4 The set

• There exists a T -invariant probability measure fT  Aut 2 ðX, mÞ j Tand T are both ergodic and of type III 1 g
r≺m ,
is a dense Gδ in (Aut2(X, m), d2). The set
dm∘T n
supn  ℤ 1 < 1,
dm 2 fT  kerw j Tand T are both ergodic and of type III 1 g

• There is a measure k ≺ m such that dk

 is a dense Gδ in (kerw, d1). (Danilenko et al.
dm
2022b).
1  L2 ðmÞ and k ∘ T ¼ k. It is easy to see that if T  Aut2(X, m), then
T  Aut2(X, t  m) for each t > 0. Hence T is
It follows that there exists “properly” non- (t  m) -nonsingular for each t > 0. However the
singular Poisson suspensions, i.e., suspensions dynamical properties of the systems
that do not admit an absolutely continuous invari- (X , T , (t  m) ) depend heavily on the choice of
ant measure. The next problem is to find out when t. This is illustrated by a concrete example
T is dissipative and when it is conservative. constructed in (Danilenko et al. 2022b).
2
dm∘T n
2 1
2 < 1,
1 Example 9.5 (Phase transition). There is a totally
Theorem 9.2 (i) If n  ℤe
dm
dissipative transformation T  Aut2(X, m) and
then T is totally dissipative
t0 > 0 such that the dynamical system
(Danilenko et al. 2022a).
(X , T , (t  m) ) is ergodic of type III1 for each
(ii) Let T  Aut1(X, m), w(T ) ¼ 0 and
2 t  (0, t0) and and totally dissipative for each t > t0.
dm
dm∘T n  1  L1 ðX, mÞ for each n  ℕ. If The point t0 can be interpreted as a “bifurcation
there is a sequence ðbn Þ1
n¼1 of positive reals
point.”
such that 1 b ¼ 1 but We note that if T  Aut2(X, m) and T is totally
n¼1 n
dissipative, then T is isomorphic to a nonsingular
Bernoulli shift. Therefore, the following theorem
1 2
ðdm∘Tdmn Þ 1 dm
was proved in (Danilenko and Kosloff 2022)
b2n e X < 1, simultaneously with Theorem 8.6.
n¼1
Theorem 9.6 Let K  {IIIl | 0  l  1} t {II1}.
then T is conservative (Danilenko et al. 2022b). Then there is a totally dissipative transformation
We also note that if T  Aut1(X, m) and T  Aut2(X, m) such that (X , T , m ) is weakly
w(T ) 6¼ 0, then T is not conservative (Danilenko mixing of type K.
et al. 2022a). Moreover, T is totally dissipative This theorem is further refined in type III0 as
(Danilenko et al. 2022b). follows (cf. Theorem 8.9).

Theorem 9.3 If T is of 0-type and there is no Theorem 9.7 Let V be an AT-flow associated
with an ITPFI2-transformation. Then there is a
T-invariant measure k ≺ m such that dk

dm totally dissipative transformation T  Aut2(X, m)
2
1  L ðmÞ, then T is of 0-type (Danilenko et al. such that T is weakly mixing and the associated
2022a). flow of T is isomorphic to V (Berendschot and
The following theorem is proved via Baire Vaes 2022b).
category tools.
Ergodic Theory: Nonsingular Transformations 265

Nonsingular Gaussian Transformations 2dh,O  ydiss  2 2dh,O :

Let ℋ be a separable infinite dimensional real
Hilbert space. Let (h, O)  Aff ℋ. For n  ℤ, This result can be strengthened under addi-
define a vector h(n)  ℋ by setting (h, O)n ¼ tional assumptions on O (Arano et al. 2021).
(h(n), On). We first consider when Gh, O admits an
equivalent invariant probability measure. Theorem 9.12 Let (h, O)  Aff ℋ. Suppose that
f is not an O-coboundary and the m - preserving
Theorem 9.8 Let (X, m) denote the space of Gh,O. transformation G0,O is mildly mixing. Then for
The following are equivalent (Arano et al. 2021; each θ < θdiss, the nonsingular Gaussian trans-
Danilenko and Lemańczyk 2022): formation Gθh,O is weakly mixing, IDPFT, and of
Krieger type III1. (Arano et al. 2021; Danilenko
• There exists a Gh,O-invariant probability mea- and Lemańczyk 2022).
sure r m, This theorem is generalized in a further work
• h is an O-coboundary, i.e., there is a  ℋ (Marrakchi and Vaes 2022) for some cases in which
such that h ¼ a  Oa, G0,O is weakly mixing but not mildly mixing. In
• The affine operator (h, O)  Aff ℋ on ℋ has these cases, G0,O has “mixing subsequences” along
a fixed point. which the cocycle (h(n))n is “proper.” In particular,
• The sequence (h(n))n  ℤ is bounded in ℋ. an example of rigid weakly mixing Gaussian trans-
formation G0,O is constructed such that the non-
The following dichotomy for nonsingular singular Gaussian transformation Gh,O is ergodic of
Gaussian transformations was established in type III1 for some h  ℋ.
(Arano et al. 2021) (see also (Danilenko and Currently, the only known examples of ergodic
Lemańczyk 2022)). nonsingular Gaussian transformations are either
of type III1 or II1. Therefore, though there is an
Theorem 9.9 If O has no nontrivial fixed vectors obvious analogy between the theory of non-
in ℋ then Gh,O is either conservative or totally singular Gaussian transformations and the theory
dissipative. of nonsingular Poisson suspensions, the latter the-
In (Arano et al. 2021), for each ory looks more diverse. We illustrate this by the
(h, O)  Aff ℋ, the authors consider a one- following theorem.
parametric family of nonsingular Gaussian trans-
formations Gθh, O, θ  (0, +1). Theorem 9.13 For each K  {IIIl| 0  l  1}
t {II1}, there is transformation T  Aut2(Y, n)
Definition 9.10 The Poincare exponent of (h, O) such that the nonsingular Poisson suspension T
is is 0-type, weakly mixing, and of Krieger type K,
while the nonsingular Gaussian transformation
1 2 Gpdn∘T1,U is 0-type, weakly mixing, and of
eakh k < 1  ½0, þ1:
ðnÞ
dh,O ≔ inf a > 0 j dn T

n¼1 Krieger type III1. The unitary Koopman operators

of the two nonsingular transformations are uni-
The following theorem demonstrates a phase tarily equivalent (Danilenko 2022).
transition from conservativity to total
dissipativity.
Spectral Theory for Nonsingular Systems
Theorem 9.11 Let (h, O)  Aff ℋ. There exists
While the spectral theory for probability preserv-
θdiss  [0 þ 1] such that Gθh,O is conservative
ing systems is developed in depth, the spectral
for all θ < θdiss and totally dissipative for all
theory of nonsingular systems is still in its infancy.
θ > θdiss. Moreover,
We discuss below some problems related to L1 -
266 Ergodic Theory: Nonsingular Transformations

spectrum which may be regarded as an analogue cl ∘ T ¼ lcl for each l. Moreover, e(T) is of
of the discrete spectrum. We also include results Lebesgue measure 0 and it can have an arbitrary
on computation of the maximal spectral type of Hausdorff dimension (Aaronson 1997, 1983;
the “nonsingular” Koopman operator for rank-one Moore and Schmidt 1980).
nonsingular transformations. A proper Borel subgroup E of  is called

L1 - spectrum and Groups of Quasi-invariance (i) Weak Dirichlet if lim supn!1 lðnÞ ¼ 1 for
Let T be an ergodic nonsingular transformation of each finite complex measure l supported on E
(X, ℬ, m). A number l   belongs to the L1-
(ii) Saturated if lim supn!1 j lðnÞ jj lðEÞ j
spectrum e(T) of T if there is a function
for each finite complex measure l on ,
f  L1(X, m) with f ∘ T ¼ lf. f is called an L1-
eigenfunction of T corresponding to l. Denote by
where lðnÞ denote the n-th Fourier coefficient of
E(T) the group of all L1-eigenfunctions of abso-
l. Every countable subgroup of  is saturated.
lute value 1. It is a Polish group when endowed
with the topology of converges in measure. If T is
Theorem 10.3 e(T) is s-compact in the usual
of type I I1, then the L1-eigenfunctions are
topology on  (Host et al. 1991) and saturated
L2(m0)-eigenfuctions of T, where m0 is an equiva-
(Méla 1983; Host et al. 1991).
lent invariant probability measure. Hence e(T ) is
It follows that e(T ) is weak Dirichlet (this fact
countable. Osikawa constructed in (Osikawa
was established earlier in (Schmidt 1982)).
1977) the first examples of ergodic nonsingular
It is not known if every Polish group continu-
transformations with uncountable e(T ).
ously embedded in  as a s-compact saturated
We state now a nonsingular version of the von
group is the eigenvalue group of some ergodic
Neumann-Halmos discrete spectrum theorem. Let
nonsingular transformation. This is the case for
Q   be a countable infinite subgroup. Let K be
the so-called H2-groups and the groups of quasi-
a compact dual of Qd, where Qd denotes Q with
invariance of measures on  (see below). Given a
the discrete topology. Let k0  K be the element
sequence nj of positive integers and a sequence
defined by k0(q) ¼ q for all q  Q. Let R : K ! K
aj  0, the set of all z   such that
be defined by Rk ¼ k þ k0. The system (K, R) is 1 nj 2
called a compact group rotation. The following j¼1 aj j1  z j < 1 is a group. It is called an
theorem was proved in (Aaronson and Nadkarni H2-group. Every H2-group is Polish in an intrinsic
1987). topology stronger than the usual circle topology.

Theorem 10.4
Theorem 10.1 Assume that the L1-
(i) Every H2-group is a saturated (and hence
eigenfunctions of T generate the entire s-algebra
weak Dirichlet) s-compact subset of  (Host
ℬ. Then T is isomorphic to a compact group
et al. 1991).
rotation equipped with an ergodic quasi-invariant 1
measure. (ii) If j¼0 aj ¼ þ1, then the corresponding

A natural question arises: Which subgroups of H2-group is a proper subgroup of .

2
 can appear as e(T ) for an ergodic T? (iii) If 1 j¼0 aj nj =njþ1 < 1, then the corres-
ponding H2-group is uncountable.
Theorem 10.2 e(T ) is a Borel subset of  and (iv) Any H2-group is e(T) for an ergodic non-
carries a unique Polish topology which is stron- singular compact group rotation T.
ger than the usual topology on : The Borel
structure of e(T) under this topology agrees with It is an open problem whether every eigenvalue
the Borel structure inherited from . There is a group e(T) is an H2-group. It is known however
Borel map c : e(T ) 3 l 7! cl  E(T ) such that that e(T) is close “to be an H2-group”: If a compact
Ergodic Theory: Nonsingular Transformations 267

subset L   is disjoint from e(T), then there is an admit a continuous homomorphism c : e(S) ! E(S)
H2-group containing e(T) and disjoint from L. with cl ∘ T ¼ lcl for all l  e(S). Hence
Example 10.5 ((Aaronson and Nadkarni e(S) 6¼ H(m) for any measure m satisfying the
1987), see also (Osikawa 1977)). Let (X, m, T ) conditions of Theorem 10.6.
be the nonsingular product odometer associated Assume that T is an ergodic nonsingular com-
1
to a sequence 2, nj j¼1 . Let nj be a sequence of pact group rotation. Let ℬ0 be the s-algebra gen-
positive integers such that nj > i < jni for all j. erated by a subcollection of eigenfunctions. Then
For x  X, we put h(x) ≔ nl(x)  j < l(x)nj. Then ℬ0 is invariant under T and hence a factor (see
h is a Borel map from X to the positive integers. §12) of T. It is not known if every factor of T is of
Let S be the tower over T with height function this form. It is not even known whether every
h (see §3.3). Then e(S) is the H2-group of all z   factor of T must have nontrivial eigenvalues.
with 1 nj 2
j¼1 nj ð0Þnj ð1Þj1  z j < 1:
It was later shown in (Host et al. 1991) that if Koopman Unitary Operator for a Nonsingular
1 2 System
j¼1 nj ð0Þnj ð1Þ nj =njþ1 < 1, then the L1-
Let (X, ℬ, m, T ) be a nonsingular dynamical sys-
eigenfunctions of S generate the entire s-algebra, tem. In this subsection, we consider spectral prop-
i.e., S is isomorphic (measure theoretically) to a erties of the Koopman operator UT defined by (3).
nonsingular compact group rotation. First, we note that the spectrum of T is the entire
Let m be a finite measure on : Let H(m) circle  (Nadkarni 1979). Next, if UT has an
≔ {z  ℤ| δz m m}, where * means the eigenvector, then T is of type II1. Indeed, if there
convolution of measures. Then Hm is a group are l   and 0 6¼ f  L2(X, m) with UT f ¼ lf,
called the group of quasi-invariance of m. It then the measure n, dn(x) ≔ | f(x)|2dm(x), is finite,
has a Polish topology whose Borel sets T-invariant, and equivalent to m. Hence if T is of
agree with the Borel sets which H(m) inherits type III or II1, then the maximal spectral type sT
from , and the injection map of H(m) into  is of UT is continuous. Another “restriction” on sT
continuous. This topology is induced by the was found in (Roy 2009): No Foïaş-Strătilă mea-
weak operator topology on the unitary group in sure is absolutely continuous with respect to sT if
the Hilbert space L2 ð, mÞ via the map T is of type II1. We recall that a symmetric mea-
HðmÞ 3 z 7! U z , ðU z f ÞðxÞ ¼ ðd ðdz mÞ=dmÞðxÞf ðxzÞ for sure s on  possesses Foïaş-Strătilă property if
f  L2 ð, mÞ: Moreover, H(m) is saturated (Host for each ergodic probability-preserving system (Y,
et al. 1991). If m(H(m)) > 0, then either H(m) v, S) and f  L2(Y, n), if s is the spectral measure
is countable or m is equivalent to l (Mandrekar of f, then f is a Gaussian random variable
and Nadkarni 1969). (Lemańczyk et al. 2000). For instance, measures
supported on Kronecker sets possess this
Theorem 10.6 Let m be an ergodic with respect property.
to the H(m)-action by translations on  . Then As we have noted in §6, mixing (0-type) is an
there is a compact group rotation (K, R) and a L2-spectral property for nonsingular transforma-
finite measure on K quasi-invariant and ergodic tions. Also, if T is infinite measure-preserving,
under R such that e(R) ¼ H(m). Moreover, there is then T is mixing if and only if n1 n1 i¼0 U T ! 0
ki

a continuous one-to-one homomorphism c : e(R) in the strong operator topology for each strictly
! E(R) such that cl ∘ R ¼ lcl for all l  e(R). increasing sequence k1 < k2 <    (Krengel and
(Aaronson and Nadkarni 1987). Sucheston 1969). This generalizes a well-known
It was shown by Aaronson and Nadkarni theorem of Blum and Hanson for probability-
(1987) that if n1 ¼ 1 and nj ¼ ajaj  1  a1 for preserving maps. For comparison, we note that
positive integers aj  2 with 1 1
j¼1 aj < 1, then ergodicity is not an L2-spectral property of infinite
the transformation S from Example 10.5 does not measure-preserving systems.
268 Ergodic Theory: Nonsingular Transformations

Now let T be a rank-one nonsingular transfor- conservative infinite measure-preserving transfor-

mation associated with a sequence ðr n , wn , sn Þ1
n¼1 mation T such that the set of essential values of the
as in §3.4. multiplicity function of UT is E. In a subsequent
paper, Danilenko and Ryzhikov (2011)) sharp-
Theorem 10.7 The spectral multiplicity of UT is ened this result: For each subset E  ℕ [ {1},
1, and the maximal spectral type sT of UT (up to a there is a mixing ergodic conservative infinite
discrete measure in the case T is of type I1) is the measure-preserving transformation T such that
weak limit of the measures rk defined as follows the set of essential values of the multiplicity func-
(Host et al. 1991; Choksi and Nadkarni 1994): tion of UT is E. We note that the analogous reali-
zation problem for spectral multiplicities of
k ergodic probability-preserving transformations is
drk ðzÞ ¼ wj ð0ÞjPj ðzÞj2 dz, still open (Danilenko 2013).
j¼1
In (Danilenko and Ryzhikov 2010), a mixing
rank-one infinite measure-preserving transforma-
where P j ðzÞ ≔ 1 þ wj ð1Þ=wj ð0ÞzR1, j þ    þ tion T was constructed such that the measures sT,
sT sT, sT sT sT, . . . on  are mutually
wj mj  1 =wj ð0ÞzRrj 1, j , z  , Ri, j ≔ ihj1 þ
disjoint. Hence the unitary operator U T U 2 T
sj ð0Þ þ    þ sj ðiÞ, 1  i  r k  1 and hj is the U 3    has a simple spectrum.
T
hight of the j-th column. El Abdalaoui and Nadkarni constructed an
Thus, the maximal spectral type of UT is given ergodic nonsingular transformation whose spec-
by a so-called generalized Riesz product. We refer trum has Lebesgue component of multiplicity one
the reader to (Host et al. 1986, 1991; Choksi and (El Abdalaoui and Nadkarni 2016). The problem
Nadkarni 1994; Nadkarni 1998) for a detailed of existence of an ergodic nonsingular transfor-
study of Riesz products: their convergence, mation with a simple Lebesgue spectrum is still
mutual singularity, singularity to l , etc. open.
It was shown in (Aaronson and Nadkarni
1987) that H(sT) e(T ) for any ergodic non- Koopman Unitary Operators Associated with
singular transformation T. Moreover, sT is ergodic Nonsingular Poisson Transformations
under the action of e(T) by translations if T is We recall a very important structural feature of
isomorphic to an ergodic nonsingular compact Poisson spaces (see Neretin 1996): There is a
group rotation. However it is not known: canonical isometry between L2(m ) and the sym-
metric Fock space F(L2(m)) over L2(m). We recall
(i) Whether H(sT) ¼ e(T ) for all ergodic T that F L2 ðmÞ ≔ 1 2 n 0
n¼0 L ðmÞ , where L (m) ¼
2
(ii) Whether ergodicity of sT under e(T ) implies n
ℂ and each factor L (m) is equipped with the
2
that T is an ergodic compact group rotation
normalized scalar product n!h:, :iL2 ðmÞn : There is
an exponential map E : L2(m) ! F(L2(m)), given
The first claim of Theorem 10.7 extends to the
by
rank N nonsingular systems as follows: If T is an
ergodic nonsingular transformation of rank N, 1 n
1
then the spectral multiplicity of UT is bounded E ðf Þ≔ n¼0 f :
n!
by N (as in the finite measure-preserving case). It
is not known whether this claim is true for a more The family {E( f ) | f  L2(m)} is total in
general class of transformations which are defined F(L2(m)). Let Affℝ(L2(m)) be the subgroup of
as rank N but without the assumption that the invertible affine operators in L2(m) that
Radon-Nikodym cocycle is constant on the preserve invariantly the ℝ - subspace L2ℝ ðmÞ:
tower levels. Then Aff ℝ L2 ðmÞ ¼ L2ℝ ðmÞ⋊U ℝ L2 ðmÞ , where
Danilenko and Ryzhikov showed in (2010) U ℝ L2 ðmÞ is the group of unitary operators that
that for each subset E  ℕ, there is an ergodic preserve invariant L2ℝ ðmÞ: The Weyl unitary
Ergodic Theory: Nonsingular Transformations 269

representation W of Affℝ(L2(m)) is well defined in  P  P mðPÞ log mðPÞ: In the study of measure-
F(L2(m)) via the formula preserving systems, the classical (Kolmogorov-
Sinai) entropy proved to be a very useful invariant
1
W ð f , V ÞE ðhÞ ≔ e2k f k2 h f ,Vhi E ðf þ VhÞ, f  L2 ðmÞ: for isomorphism (Cornfeld et al. 1982). The key
fact of the theory is that if m ∘ T ¼ m then the limit
Given T  Aut2(X, m), we have that lim n!1 n1 H ni¼1 T i P exists for every P :
dm∘T
 1, U T  Aff ℝ L2 ðmÞ : However, if T does not preserve m, the limit may
dm
no longer exist. Some efforts have been made to
extend the use of entropy and similar invariants to
Theorem 10.8 (Koopman operator associated
the nonsingular domain. These include Krengel’s
with T (Danilenko et al. 2022a)). If T  Aut2(X, m),
entropy of conservative measure-preserving maps
then under the canonical identification of L2(m )
and its extension to nonsingular maps, Parry’s
and F(L2(m)),
entropy and Parry’s nonsingular version of
Shannon-McMillan-Breiman theorem, Poisson
dm∘T
UT ¼ W  1, U T : entropy, critical dimension by Mortiss and
dm
Dooley, etc. Unfortunately, these invariants are
less informative than their classical counterparts
Koopman Unitary Operators Associated with and they are more difficult to compute.
Nonsingular Gaussian Transformations
Let ℋ be a separable infinite dimensional real Krengel and Parry’s Entropies
Hilbert space. Denote by (X, m) the probability Let S be a conservative measure-preserving trans-
space where the nonsingular Gaussian transfor- formation of a s-finite measure space (Y, E, n). The
mations Gh,O for all (h, O)  Aff ℋ are defined Krengel entropy (Krengel 1967) of S is defined by
(see §3.8). It is well known that there is a canon-
ical isometry between L2(X, m) and the symmetric hKr ðSÞ ¼ supfnðEÞhðSE Þ j 0 < nðEÞ < þ1g,
Fock space F(ℋ).
where h(SE) is the Kolmogorov-Sinai entropy of
Theorem 10.9 (Koopman operator associated SE. It follows from Abramov’s formula for the
with Gh,O (Danilenko and Lemańczyk 2022)). If entropy of induced transformation that hKr(S) ¼
(h, O)  Aff ℋ, then under the canonical iden- m(E)h(SE) whenever E sweeps out, i.e.,
tification of L2(X, m) and F(ℋ), [i0SiE ¼ X. A generic transformation from
Aut0(X, m) has entropy 0. Krengel raised a ques-
1 tion in (Krengel 1967): Do there exist a zero
U Gh,O ¼ W h, O :
2 entropy infinite measure-preserving S and a zero
entropy finite measure-preserving R such that
It follows from Theorems 10.8 and 10.9 that hKr(S R) > 0? This problem was solved in
each nonsingular Poisson transformation T is (Danilenko and Rudolph 2009) (a special case
spectrally equivalent to the nonsingular Gaussian was announced by Silva and Thieullen in an
transformation G dm∘T
: October 1995 AMS conference (unpublished)):
dm 1

(i) If hKr(S) ¼ 0 and R is distal, then hKr(S

R) ¼ 0;
Entropy and Other Invariants (ii) If R is not distal, then there is a rank-one
transformation S with hKr(S R) ¼ 1.
Let T be an ergodic conservative nonsingular
transformation of a standard probability space We also note that if a conservative
(X, ℬ, m). If P is a finite partition of X, we define S  Aut0(X, m) is squashable, i.e., it commutes
the entropy H ðP Þ of P as HðP Þ ¼
270 Ergodic Theory: Nonsingular Transformations

with another transformation R such that n ∘ R ¼ cn LLB. This result was extended in (Janvresse et al.
for a constant c 6¼ 1, then hKr(S) is either 0 or 1 2010) in the following way.
(Silva and Thieullen 1995).
Now let T be a type III ergodic transformation Theorem 11.1 Let T be an ergodic quasi-finite
of (X, ℬ, m). Silva and Thieullen define an transformation. Then either there is the Krengel-
entropy h (T ) of T by setting h ðT Þ ≔ hKr T , Pinsker factor of T which is also the Parry-Pinsker
and the Poisson-Pinsker (see the next subsection
where T is the Maharam extension of T (see §5.2).
below) factor of Tor T is remotely infinite, i.e., there
Since T commutes with transformations which
exists a sub-s-algebra F  ℬ such that T1F  F ,
“multiply” T -invariant measure, it follows that
_n>0T nF ¼ F and the subalgebra ^n>0TnF
h (T ) is either 0 or 1.
does not contain subsets of positive finite measure.
Let T be the standard IIIl-odometer from Exam-
ple 5.1 (i). Then h (T) ¼ 0. The same is true for a
Poisson Entropy
so-called ternary product odometer associated with
Poisson entropy for infinite measure-preserving
the sequence ð3, nn Þ1n¼1 , where nn(0) ¼ nn(2) ¼ transformations was introduced in (Roy 2005).
l/(1 þ 2l) and nn(1) ¼ l/(1 þ l) (Silva and
Let (X, m) be an infinite s-finite space, and let
Thieullen 1995). It is not known however whether
T be a m-preserving invertible transformation of
every ergodic nonsingular product odometer has
X. The Poisson suspension T of T is well defined
zero entropy. On the other hand, it was shown in
on a probability space (X , m ) and m ∘ T ¼ m
(Silva and Thieullen 1995) that h (T) ¼ 1 for
(see §3.7). It is ergodic if and only if T has no
every K-automorphism.
invariant sets of finite positive measure. It follows
The Parry entropy (Parry 1969) of S is defined
from Theorem 10.8 that U T is the “exponent” of
by
UT. Hence, the maximal spectral type of U T is
n  0(n!)1(sT) n, where sT is a measure of the
hPa ðSÞ ≔ H S1 FjF jF is a s  finite maximal spectral type of UT.
subalgebra of B such that F  S1 Fg: Now the Poisson entropy hPo(T ) of T is h(T ).
The main question is: hPo(T ) coincides with
Parry showed (1969) that hPa(S)  hKr(S). It is hPa(T ) or hkr(T )? It was shown in (Janvresse
still an open question whether the two entropies et al. 2010) that hPa(T )  hPo(T ). If T is quasifinite
coincide. This is the case when S is of rank one or rank one, then the three entropies of T coincide
(since hKr(S) ¼ 0) and when S is quasi-finite (Parry (Janvresse et al. 2010). If T is the infinite Markov
1969). The transformation S is called quasi-finite if shift associated with a pair (P, π) for recurrent and
there exists a subset of finite measure A  Y such irreducible P (see §3.6), then
that the first return time partition (An)n>0 of A has
finite entropy. We recall that x  An , n is the hKr ðT Þ ¼ hPa ðT Þ
smallest positive integer such that Tnx  A. An ¼ hPo ðT Þ ¼  pðaÞ Pða,bÞ log Pða, bÞ:
example of non-quasi-finite ergodic infinite a A b A

measure-preserving transformation was

constructed in (Aaronson and Park 2009). If sT is singular or UT has finite multiplicity,
A natural question is about existence of the maxi- then hPo(T ) ¼ 0 (Janvresse et al. 2010). It was
mal invariant s-finite subalgebra of zero (Krengel also shown in (Janvresse et al. 2010) that given
or Parry) entropy. Such an algebra is called the a nontrivial invariant s-finite algebra F of ℬ,
Krengel-Pinsker or the Parry-Pinsker factor of T, the natural F -relative version of Poisson
respectively. Existence of the Krengel-Pinsker fac- entropy coincides with the relative (Krengel)
tors was proved in (Aaronson and Park 2009) for a entropy defined in (Danilenko and Rudolph
special class of quasi-finite transformations called 2009). Hence, if Krengel’s and the Poisson
Ergodic Theory: Nonsingular Transformations 271

entropies coinside on T F for some F , then contains x. We put o1 ¼ 0. Parry shows in (Parry

+
hKr(T ) ¼ hPo(T ). On the other hand, Janvresse 1963) that
and de la Rue constructed an ergodic conserva-
n
tive infinite measure-preserving transformation j¼0 log m Cnj T j x oj ðxÞ  oj1 ðxÞ
T such that hKr(T ) ¼ 0 but hPo(T ) > 0 (Janvresse n
i¼0 oj ðxÞ
and de la Rue 2012).
1
! H P j _ T 1 P
i¼1
Definition 11.2 An ergodic measure-preserving
transformation T of a s-finite measure space dm∘T 1 i
 log E j _ T P dm
(X, ℬ, m) is said to have totally positive Poisson X dm i¼0
entropy if for each s-finite T-invariant
sub-s-algebra F  ℬ, the Poisson entropy of the for a.a. x. Parry also shows that under the afore-
system (X, F , m F , T ) is strictly positive.
+

mentioned conditions on T,
We note that the Poisson suspension of the
system (X, F , m F , T ) from the above definition
+

n n1
1 j jþ1
is canonically a factor of X, ℬ, m, T : Such fac- H _ T j P  H _ T j P
n j¼0
i¼0
j¼0
i¼1
tors of T are called Poissonian. Roy showed in
1
(Roy 2010) that if T has totally positive Poisson ! H P j _ T i P
i¼1
entropy, then T is of zero type.

Theorem 11.3 (Existence of the Poisson-Pinsker Critical Dimension

factor (Roy 2010)). Let T be an ergodic measure- The critical dimension introduced by Mortiss
preserving transformation of an infinite s-finite (2003) measures the order of growth for sums of
measure space (X, ℬ, m). Then either T has Radon-Nikodym derivatives. Let (X, ℬ, m, T ) be
totally positive entropy and T is CPE or there is an ergodic nonsingular dynamical system. Given
a s-finite T-invariant sub-s-algebra E  ℬ such δ > 0, let
that the Poisson suspension of (X, F , m F , T ) is
+

the Pinsker factor of T: n1

i¼0 oi ðxÞ
If T has totally positive entropy, then the max- Xd ≔ x  Xj lim inf >0 and
n!1 nd
imal spectral type of T is Lebesgue countable. If
hPo(T) > 0 and T possesses a Poisson-Pinsker ð6Þ
factor, then the maximal spectral type of T in the
n1
orthocomplement to the Poisson-Pinsker factor is i¼0 oi ðxÞ
Xd ≔ x  Xj lim inf ¼0 : ð7Þ
Lebesgue countable (Roy 2010). n!1 nd

Parry’s Generalization of Shannon-MacMillan-

Then Xδ and X δ are T-invariant subsets.
Breiman Theorem
Let T be an ergodic transformation of a standard
Definition 11.4 The lower critical dimension
nonatomic probability space (X, ℬ, m). Suppose
α(T) of T is sup{δ | m(Xδ) ¼ 1}. The upper critical
that f ∘ T  L1(X, m) if and only if f  L1(X, m).
dimension β(T) of T is inf{δ | m(X δ) ¼ 1}.
This means that there is K > 0 such that K 1 <
(Mortiss 2003; Dooley and Mortiss 2009).
dm ∘ T
dm ðxÞ < K for a.a. x. Let P be a finite partition It was shown in (Dooley and Mortiss 2009)
of X. Denote by Cn(x) the atom of _ni¼0 T i P which that the lower and upper critical dimensions are
272 Ergodic Theory: Nonsingular Transformations

invariants for isomorphism of nonsingular sys- coordinate). It also follows from Theorem 11.6
tems. Notice also that that if T is a product odometer of bounded type,
then α(T 1) ¼ α(T) and β(T 1) ¼ β(T). In
n
i¼1 oi ðxÞ
log (Dooley and Mortiss 2006), Theorem 11.6 was
aðT Þ ¼ lim inf and bðT Þ extended to a subclass of Markov odometers.
n!1 log n
n Those results were further extended to so-called
log i¼1 oi ðxÞ
¼ lim sup : G-measures on product spaces (Mansfield and
n!1 log n
Dooley 2017) and a class of Bratteli-Vershik sys-
tems with multiple edges (Dooley and Hagihara
Moreover, 0  α(T)  β(T)  1. If T is of type
2012). The critical dimensions for nonsingular
II1, then α(T ) ¼ β(T ) ¼ 1. If T is the standard IIIl-
Bernoulli shifts (see §3.5) were investigated in
odometer from Example 5.1, then aðT Þ ¼
(Dooley and Mortiss 2007):
bðT Þ ¼ logð1 þ lÞ  1þl
l
log l:
Theorem 11.7 For any ϵ > 0, there exists a
Theorem 11.5 (i) For every l  [0, 1] and every nonsingular Bernoulli shift S from the Krengel
c  [0, 1], there exists a nonsingular product class with α(S) < ϵ and β(S) > 1  ϵ.
odometer of type IIIl with critical dimension
equal to c (Mortiss 2002). Nonsingular Restricted Orbit Equivalence
(ii) For every c  [0, 1], there exists a non- In (Mortiss 2000), Mortiss initiated study of a
singular product odometer of type II1 with criti- nonsingular version of Rudolph’s restricted orbit
cal dimension equal to c (Dooley and Mortiss equivalence (Rudolph 1985). This work is still in
2009). its early stages and does not yet deal with any
form of entropy. However she introduced non-
Let T be the nonsingular product odometer singular orderings of orbits, defined sizes, and
associated with a sequence ðmn , nn Þ1 n¼1 : Let showed that much of the basic machinery still
s(n) ¼ m1  mn and let H ðP n Þ denote the entropy works in the nonsingular setting.
of the partition of the first n coordinates with
respect to m. We now state a nonsingular version
of Shannon-MacMillan-Breiman theorem for Nonsingular Joinings and Factors
T from (Dooley and Mortiss 2009).
The theory of joinings is a powerful tool to study
Theorem 11.6 Let mi be bounded from above. probability-preserving systems and to construct
Then striking counterexamples. It is interesting to
study what part of this machinery can be extended
n
 log mi ðxi Þ to the nonsingular case. However, there are some
(i) aðT Þ ¼ lim inf n!1 inf i¼1
log sðnÞ principal obstacles for such extensions:
H ðP n Þ
¼ lim inf n!1 log sðnÞ and
n • There are too many quasi-invariant measures
 log mi ðxi Þ
(ii) bðT Þ ¼ lim supn!1 inf i¼1
in view of the Glimm-Effros theorem (see The-
log sðnÞ
H ðP n Þ
orem 2.12).
¼ lim supn!1 log sðnÞ • Ergodic components of a nonergodic joining
need not be joinings of the original systems.
for a.a. x ¼ (xi)i1  X.
It follows that in the case when α(T) ¼ β(T ), There are several ways to bypass these obsta-
the critical dimension coincides with cles. The principal idea is to select always an
H ðP n Þ
log sðnÞ : In (Mortiss 2002), this expression
lim n!1 appropriate (rather narrow) class of quasiinvariant
(when it exists) was called AC-entropy (average measures under consideration or impose some
Ergodic Theory: Nonsingular Transformations 273

restrictions on the structure of joinings. This dm ∘ T dn∘S dmS’ðxÞ ∘ T

approach led to some progress in understanding ðxÞ ¼ ð’ðxÞÞ ðxÞ ð8Þ
dm dn dm’ðxÞ
twofold joinings and constructing prime systems
of any Krieger type. As far as we know, the for a.e. x  X. Define now the relative product
higher-fold nonsingular joinings have not been m ¼ m ’ m on X X by setting m ¼ my
considered so far. It turned out however that an my dnðyÞ: Then it is easy to deduce from (8) that
alternative coding technique, predating joinings in
m is a nonsingular self-joining of T.
studying the centralizer and factors of the classical
We note however that the above definition of
measure-preserving Chacón maps, can be used as
joining is too general to be satisfactory (as we
well to classify factors of Cartesian products of
noted in the introduction to this section). It does
some nonsingular Chacón maps.
not reduce to the classical definition when we
consider probability-preserving systems. Indeed,
Joinings, Nonsingular MSJ, and Simplicity
the following result was proved in (Rudolph and
In this subsection, all measures are probability
Silva 1989).
measures. A nonsingular joining of two non-
singular systems (X1, ℬ1, m1, T1) and
(X2, ℬ2, m2, T2) is a measure m on the product Theorem 12.1 Let (X1, ℬ1, m1, T1) and
ℬ1 ℬ2 that is nonsingular for T1 T2 and (X2, ℬ2, m2, T2) be two finite measure-preserving
satisfies: mðA X2 Þ ¼ m1 ðAÞ and mðX1 BÞ ¼ systems such that T1 T2 is ergodic. Then for
m2 ðBÞ for all A  ℬ1 and B  ℬ2. Clearly, the every l, 0 < l < 1, there exists a nonsingular
product m1 m2 is a nonsingular joining. Given a joining m of m1 and m2 such that ðT 1 T 2 , mÞ is
transformation S  C(T ), the measure mS given ergodic and of type IIIl.
by mS(A B) ≔ m(A \ S1B) is a nonsingular It is not known however if the nonsingular
joining of (X, m, T ) and (X, m ∘ S1, T ). It is joining m can be chosen in every orbit equivalence
called a graph-joining since it is supported on class. In view of the above, Rudolph and Silva
the graph of S. Another important kind of joinings (1989) isolate an important subclass of joining. It
that we are going to define now is related to is used in the definition of a nonsingular version of
factors of dynamical systems. Recall that given a minimal self-joinings.
nonsingular system (X, ℬ, m, T ), a sub-s-algebra
A of ℬ such that T 1 ðA Þ ¼ A mod m is called a Definition 12.2 (i) A nonsingular joining m of
factor of T. There is another, equivalent, defini- (X1, m1, T1) and (X2, m2, T2) is rational if there
tion. A nonsingular dynamical system ðY, C , n, SÞ exit measurable functions c1 : X1 ! ℝ+ and
is called a factor of T if there exists a measure- c2 : X2 ! ℝ+ such that
preserving map ’ : X ! Y, called a factor map,
with ’T ¼ S’ a.e. (If ’ is only nonsingular, v may om1 ðx1 , x2 Þ ¼ o1 1 ðx1 Þo1 2 ðx2 Þc1 ðx1 Þ
m m
be replaced with the equivalent measure m ∘ ’1,
¼ o1 1 ðx1 Þo1 2 ðx2 Þc2 ðx2 Þ
m m
for which ’ is measure-preserving.) Indeed, the m a:e:
sub-s-algebra ’1 ðC Þ  ℬ is T-invariant and,
conversely, any T-invariant sub-s-algebra of ℬ (ii) A nonsingular dynamical system
defines a factor map by immanent properties of (X, ℬ, m, T ) has minimal self-joinings
standard probability spaces; see (Aaronson 1997). (MSJ) over a class M of probability measures
If ’ is a factor map as above, then m has a disin- equivalent to m, if for every m1, m2  M, for
tegration with respect to ’, i.e., m ¼ mydn( y) for every rational joining m of m1, m2, a.e. ergodic
a measurable map y 7! my from Y to the proba- component of m is either the product of its
bility measures on X so that my(’1( y)) ¼ 1, the marginals or is the graph-joining supported
measure mS’(x) ∘ T is equivalent to m’(x) and on T j for some j  ℤ.
274 Ergodic Theory: Nonsingular Transformations

Clearly, product measure, graph-joinings, and ones); however, these transformations are not
the relative products are all rational joinings. prime.
Moreover, a rational joining of finite measure- A more general notion than MSJ, called graph
preserving systems is measure-preserving and a self-joinings (GSJ), was introduced (Silva and
rational joining of type II1’s is of type II1 Witte 1992): Just replace the the words “on T j
(Rudolph and Silva 1989). Thus we obtain the for some j  ℤ” in Definition 12.2(ii) with “on
finite measure-preserving theory as a special S for some element S  C(T ).” For finite measure-
case. As for the definition of MSJ, it depends on preserving transformations, GSJ over {m} is the
a class M of equivalent measures. In the finite same as the usual twofold simplicity (del Junco
measure-preserving case, M ¼ {m}. However, in and Rudolph 1987). The famous Veech theorem
the nonsingular case no particular measure is dis- on factors of twofold simple maps (see del Junco
tinguished. We note also that Definition 12.2 and Rudolph 1987) was extended to nonsingular
(ii) involves some restrictions on all rational join- systems in (Silva and Witte 1992) as follows: If a
ings and not only ergodic ones as in the finite system (X, ℬ, m, T ) has GSJ, then for every non-
measure-preserving case. The reason is that an trivial factor A of T there exists a locally compact
ergodic component of a nonsingular joining subgroup H in C(T ) (equipped with the weak
needs not be a joining of measures equivalent to topology) which acts smoothly (i.e., the partition
the original ones (Aaronson 1987). For finite into H-orbits is measurable) and such that A ¼
measure-preserving transformations, MSJ over fB  ℬ j mðhBDBÞ ¼ 0 for all h  H}. It follows
that there is a cocycle ’ from ðX, A, m A Þ to

+
{m} is the same as the usual twofold MSJ (del
Junco and Rudolph 1987). H such that T is isomorphic to the ’-skew product
A nonsingular transformation T on (X, ℬ, m) is
+
extension (T A)’ (see §6.4). Of course, the ergo-
called prime if its only factors are ℬ and {X, dic nonsingular product odometers and, more
;} mod m. A (nonempty) class M of probability generally, ergodic nonsingular compact group
measures equivalent to m is said to be centralizer rotation (see § 10.1) have GSJ. However, except
stable if for each S  C(T ) and m1  M, the for this trivial case (the Cartesian square is non-
measure m1 ∘ S is in M. ergodic) plus the systems with MSJ from
(Rudolph and Silva 1989), no examples of type
Theorem 12.3 Let (X, ℬ, m, T ) be a ergodic III systems with GSJ are known. In particular, no
nonatomic dynamical system such that T has smooth examples have been constructed so far.
MSJ over a class M that is centralizer stable. This is in sharp contrast with the finite measure-
Then T is prime and the centralizer of T consists preserving case where abundance of simple
of the powers of T (Rudolph and Silva 1989). (or close to simple) systems are known (see del
A question that arises is whether such a non- Junco and Rudolph 1987; Thouvenot 1995;
singular dynamical system (not of type II1) exist. Danilenko 2007).
Expanding on Ornstein’s original construction
from (Ornstein 1972), Rudolph and Silva con- Nonsingular Coding and Factors of Cartesian
struct in (Rudolph and Silva 1989), for each Products of Nonsingular Maps
0  l  1, a nonsingular rank-one transformation As we have already noticed above, the non-
Tl that is of type IIIl and that has MSJ over a class singular MSJ theory was developed in (Rudolph
M that is centralizer stable. Type II1 examples and Silva 1989) only for twofold self-joinings.
with analogues properties were also constructed The reasons for this were technical problems
there. In this connection, it is worth to mention the with extending the notion of rational joinings
example by Aaronson and Nadkarni (Aaronson from twofold to n-fold self-joinings. However,
and Nadkarni 1987) of II1 ergodic transforma- while the twofold nonsingular MSJ or GSJ prop-
tions that have no factor algebras on which the erties of T are sufficient to control the centralizer
invariant measure is s-finite (except for the entire and the factors of T, it is not clear whether it
Ergodic Theory: Nonsingular Transformations 275

implies anything about the factors or centralizer of (i) If l1 6¼ l, then F is equal mod 0 to one of the
T T. Indeed, to control them one needs to know four algebras ℬ  ℬ, ℬ  N , N  ℬ, or
the fourfold joinings of T. However, even in the N  N , where N ¼ f;, Xg:
finite measure-preserving case, it is a long- (ii) If l1 ¼ l, then F is equal mod 0 to one of the
standing open question whether twofold MSJ following algebras ℬ  ℬ, ℬ  N , N 
implies n-fold MSJ. That is why del Junco and ℬ, N  N , or (Tm Id)ℬ2 for some
Silva (2003) apply alternative – nonsingular cod- integer m.
ing – techniques to classify the factors of Carte-
sian products of nonsingular Chacón maps. The It is not hard to obtain type III1 examples of
techniques were originally used in (del Junco Chacón maps for which the previous two theo-
1978) to show that the classical Chacón map is rems hold. However the construction of type II1
prime and has trivial centralizer. They were and type III0 nonsingular Chacón transformations
extended to nonsingular systems in (del Junco is more subtle as it needs the choice of on to vary
and Silva 1995). with n. In (Hamachi and Silva 2000), Hamachi
For each 0 < l < 1, we denote by Tl the and Silva construct type III0 and type II1 exam-
Chacón map (see §3.4) corresponding the ples; however, the only property proved for these
sequence of probability vectors wn ¼ (l/(1 þ maps is ergodicity of their Cartesian square. More
2l), 1/(1 þ 2l), l/(1 þ 2l)) for all n > 0. One recently, Danilenko (2004) has shown that all of
can verify that the maps Tl are of type IIIl. (The them (in fact, a wider class of nonsingular Chacón
classical Chacón map corresponds to l ¼ 1.) All maps of all types) are power weakly mixing.
of these transformations are defined on the same In (Choksi et al. 1989), Choksi, Eigen, and
standard Borel space (X, ℬ). These transforma- Prasad asked whether there exists a zero entropy,
tions were shown to be power weakly mixing in finite measure-preserving mixing automorphism
(Adams et al. 2001). The centralizer of any finite S, and a nonsingular type III automorphism T,
Cartesian product of nonsingular Chacón maps is such that T S has no Bernoulli factors. Theorem
computed in the following theorem. 12.5 provides a partial answer (with a mildly
mixing only instead of mixing) to this question:
Theorem 12.4 Let 0 < l1 < . . . < lk  1 and n1, If S is the finite measure-preserving Chacón map
. . ., nk be positive integers. Then the centralizer of and T is a nonsingular Chacón map as above, the
factors of T S are only the trivial ones, so T S
the Cartesian product T n l1
1
. . . T n
lk
k
is gen-
has no Bernoulli factors.
erated by maps of the form U1 . . . Uk, where
each Ui, acting on the ni-dimensional product Joinings and MSJ for Infinite Measure-
space Xni , is a Cartesian product of powers of Preserving Systems
T li or a coordinate permutation on Xni : (del Junco Adams, Friedman, and Silva introduced in
and Silva 2003). (Adams et al. 1997) an infinite version of Chacón
Let π denote the permutation on X X defined map T as a rank-one transformation associated
by π(x, y) ¼ (y, x), and let ℬ2 denote the sym- with ðr n , on , sn Þ1
n¼1 such that rn ¼ 3, on(0) ¼
metric factor, i.e., ℬ2 ¼ {A  ℬ  ℬ| π(A) ¼
on(1) ¼ on(2), sn(0) ¼ 0, sn(1) ¼ 1 and sn(2) ¼
A}. The following theorem classifies the factors of
3hn þ 1 for each n > 0. This is called the infinite
the Cartesian product of any two nonsingular type
Chacón transformation. Let (X, m) be the space of
IIIl, 0 < l < 1, or type II1 Chacón maps.
T. Of course, m(X) ¼ 1. This transformation has
infinite ergodic index (Adams et al. 1997), is not
Theorem 12.5 Let T l1 and T l2 be two non- power weakly mixing and not multiply recurrent
singular Chacón systems. Let F be a factor alge- (Gruher et al. 2003), and has trivial centralizer
bra of T l1 T l2 : (del Junco and Silva 2003). (Janvresse et al. 2018). For each d > 0, Janvresse,
276 Ergodic Theory: Nonsingular Transformations

de la Rue, and Roy investigated T d-invariant T-invariant measure m on X. A d-fold Radon

measures on Xd which are boundedly finite. This self-joining of T is a Radon measure on Xd
means that for each d levels of every tower of the whose marginals (which may be nonsigma-
inductive construction, the measure of the Carte- finite) are equivalent to m. We consider only
sian product of these levels is finite. The product Radon invariant measures and define Radon
dn¼1 m and graph-joinings, i.e., measures of the d-fold MSJ and Radon disjointness. Of course,
form ðA1 , . . . , Ad Þ 7! m S1 1
1 A1 ,    \ Sd A d for each ergodic component of nonergodic Radon
some transformations S1, . . ., Sd  C(T ), are joinning is Radon. However, it need not be a
boundedly finite. Moreover, T itself is uniquely joining. Then the (C, F)-construction (see
ergodic in the sense that there is only one (up to (Danilenko 2001a, 2007)) is used to produce a
scaling) boundedly finite T-invariant measure. It number of rank-one homeomorphisms of
was shown in (Janvresse et al. 2018) that each X whose ergodic joinings are explicitly
ergodic T d-invariant boundedly finite measure described. The weird measures from (Janvresse
is a direct product of so-called diagonal measures. et al. 2018) appear now as quasigraph Radon
Unlike the finite measure-preserving case, the measures, i.e., they are graphs of equivariant
class of diagonal measures does not reduce to maps whose domain and range are meager (and
the graphjoinings (with S1, . . ., Sd being the pow- of zero measure) subsets of X. It is constructed an
ers of T ). It contains so-called weird measures uncountable family of pairwise Radon disjoint
whose marginals are singular to m. As a corollary, infinite Chacon like Radon uniquely ergodic
it was proved that C(T ) ¼ {Tn| n  ℤ} (Janvresse homeomorphisms with Radon MSJ. Moreover,
et al. 2018). Some of the weird measures are every transformation of this family is Radon
totally dissipative (supported on a single orbit), disjoint with its inverse (Danilenko 2018).
and some of them are conservative. Danilenko
showed in (2018) that there is a conservative T
T-invariant boundedly finite measure with abso- Smooth Nonsingular Transformations
lutely continuous marginals whose ergodic com-
ponents are all weird. This phenomenon is Diffeomorphisms of smooth manifolds equipped
impossible for another infinite version T of with smooth measures are commonly considered
Chacon map constructed in (Janvresse et al. as physically natural examples of dynamical sys-
2019). Its construction mimics the construction tems. Therefore the construction of smooth
of the classical Chacon map so much that it models for various dynamical properties is a
gives a m-conull subset X1 such that for each well-established problem of the modern
d  1, each ergodic T d-invariant measure (probability-preserving) ergodic theory. Unfortu-
supported on Xd1 is the direct product of several nately, the corresponding “nonsingular” counter-
copies of m and the graph-joinings generated by part of this problem is almost unexplored. We
powers of T. As a corollary, we obtain that each survey here several interesting facts related to
boundedly finite d-fold self-joining of T (the mar- the topic.
ginals of a joining are absolutely continuous) is a For r  ℕ [ {1}, denote by Diff rþ ðÞ the
convex combination of countably many ergodic group of orientation-preserving Cr-
joinings. diffeomorphisms of the circle . Endow this set
In (Danilenko 2018), the problems studied in with the natural Polish topology. Fix
(Janvresse et al. 2018) are considered from a T  Diff rþ ðÞ: Since  ¼ ℝ=ℤ, there exists a
different point of view. Let T be a homeomor- C1 - function f : ℝ ! ℝ such that T(x þ ℤ) ¼
phism of a locally compact Cantor space X. We f(x) þ ℤ for all x  ℝ. The rotation number r(T )
assume that T is Radon uniquely ergodic, i.e.,
there is only one (up to scaling) Radon of T is the limit lim n!1 f ∘  ∘f ðxÞ ðmod1Þ:
n times
Ergodic Theory: Nonsingular Transformations 277

The limit exists and does not depend on the choice all C1-orientation-preserving diffeomorphisms
of x and f. It is obvious that T is nonsingular with with irrational rotation numbers. In contrast to
respect to Lebesgue measure l : Moreover, if that, Hawkins constructs in (Hawkins 1982) a
T  Diff rþ ðÞ and r(T ) is irrational, then the type III0 C1-diffeomorphism of the
dynamical system ð, l , T Þ is ergodic (Cornfeld 4-dimensional torus which is not ITPFI.
et al. 1982). It is interesting to ask: which Examples of n-to-1 conservative ergodic non-
Krieger’s type can such systems have? singular C1-endomorphisms on the 2-torus, not
Katznelson showed in (Katznelson 1977) that admitting an equivalent s-finite invariant mea-
the subset of type III C1-diffeomorphisms and the sure, were constructed in (Hawkins and Silva
subset of type II1C1-diffeomorphisms are dense 1991). In (Avila and Bochi 2007), it is shown
in Diff 1þ ðÞ: Hawkins and Schmidt refined the
that a C1 generic expanding map of  has no
idea of Katznelson from (Katznelson 1977) to absolutely continuous s-finite invariant measure.
construct, for every irrational number α  [0, 1) Kosloff in (2021) showed that 2 admits a C1
which is not of constant type (i.e., in whose con- Anosov diffeomorphism of type III1 with respect
tinued fraction expansion the denominators are to Lebesgue measure. We recall that this phenom-
not bounded) a transformation T  Diff 2þ ðÞ enon is impossible in the class of conservative
which is of type III1 and r(T ) ¼ α (Hawkins and C1þα Anosov diffeomorphisms because by a the-
Schmidt 1982). It should be mentioned that class orem of Gurevich and Oseledets, every such trans-
C2 in the construction is essential, since it follows formation is of type II1 (with respect to Lebesgue
from a remarkable result of Herman that if measure). In a later work (Kosloff 2018), he
T  Diff 3þ ðÞ , then under some condition on α extended this result to d for every d > 3. The
(which determines a set of full Lebesgue mea- case d ¼ 3 remains open.
sure), T is measure theoretically (and topologi-
cally) conjugate to a rotation by r(T ) (Herman
1979b). Hence T is of type II1. Miscellaneous Topics
In (Hawkins 1983), Hawkins shows that every
smooth paracompact manifold of dimension 3 Let T be an ergodic measure-preserving transfor-
admits a type IIIl diffeomorphism for every mation of an infinite s-finite nonatomic measure
l  [0, 1]. This extends a result of Herman space (X, ℬ, m).
(Herman 1979a) on the existence of type III1
diffeomorphisms in the same circumstances. On Normalizing Constants for Ergodic
It is also of interest to ask: Which free ergodic Theorem
flows are associated with smooth dynamical sys- Replacing m with an equivalent probability mea-
tems of type III0? Hawkins proved that any free sure, one can deduce from the Hurewicz ergodic
ergodic C1-flow on a smooth, connected, para- theorem that the average 1n n1 i
i¼0 f T x converges
compact manifold is the associated flow for a C1- to 0 a.e. for each function f  L (X, m). In view of
1

diffeomorphism on another manifold (of higher that, a natural question arises: Is there a sequence
dimension) (Hawkins 1990a). of positive numbers ðan Þ1 n¼1 such that
A nice result was obtained in (Katznelson
1979; Hawkins and Woods 1984). If n1
1
T  Diff 2þ ðÞ and the rotation number of T has f T i x ! fdm a:e: ð9Þ
an i¼0
unbounded continued fraction coefficients, then
ð, l , T Þ is ITPFI. Moreover, a converse also for each f  L1(X, m)? Aaronson answered this
holds: Given a nonsingular product odometer R, question negatively in (Aaronson 1977b). He
the set of orientation-preserving C1- showed that if there is a sequence of positive num-
diffeomorphisms of the circle which are orbit bers ðan Þ1
n¼1 and a single integrable function f  0
equivalent to R is C1-dense in the Polish set of with fdm > 0 such that (9) holds, then m(X) < 1.
278 Ergodic Theory: Nonsingular Transformations

Thus, no normalizing constants in the ergodic theo- transformation of rank one has s-finite MSJ.
rem for infinite measure-preserving transformations Since the rank-one transformations are non-
exist. Since then, other forms of convergence of squashable, it follows that the centralizer of each
ergodic averages for a given sequence have been zero-type rank-one transformation is just its
studied, for which the reader may refer to (Aaronson powers.
1981, 1997; Aaronson and Zweimüller 2014; Thaler
1998; Thaler and Zweimüller 2006) and the refer- Definition 14.2 Let T be a rank-one
ences therein. The situation is somewhat different in transformation associated with ðr n , on , sn Þ1 n¼1 :
the case of symmetric Birkhoff sums, i.e., when we It is called partially bounded if there is L > 0
replace the average n1 i¼0 in (9) with jij<n. It is such that r n  L, on ð0Þ ¼    ¼ on ðr n 1Þ,
shown in (Aaronson et al. 2017) that though (9) with max 0i<j<rn 1 jsn ðiÞ  sn ðjÞj < L, sn ðr n 1Þ ¼ 0
jij<n instead of n1i¼0 does not hold for all f  L (-
1
and min 0i<rn 1 sn ðiÞ  hn for each n > 0
X, m), there are examples of ergodic conservative (Gaebler et al. 2018).
infinite measure-preserving transformations It was shown in (Gaebler et al. 2018) that for
ðX, B, m, T Þ admitting sequences (an)n with an ! þ each partially bounded transformation, the cen-
1 and such that tralizer consists of just the powers. Of course,
the family of partially bounded transformations
1 does not intersect the set of zero-type rank-
lim sup an f T i x ¼ fdm and
n 2 one maps.
jij<n

1 1
lim inf an f Tix ¼ fdm a:e: Asymmetry and Bergelson’s Question
n 2 2
jij<n
We say that T is asymmetric if T is not isomorphic
to T1. Explicit examples of asymmetric infinite
for each f  L1þ ðX, mÞ: See also (Kosloff 2017).
rank-one transformations are constructed in
(Danilenko and Ryzhikov 2012; Ryzhikov 2014)
Around King’s Weak Closure Theorem (see there for asymmetric maps which embed into
We recall that if S is probability preserving rank- a flow). It was shown in (Gaebler et al. 2018) that
one map, then C(S) is the weak closure of the set if T is a partially bounded rank-one transforma-
{Sn| n  ℤ} (King theorem, (King 1986)). It is tion, then T is isomorphic to T 1 if and only if
still unclear whether this theorem extends to the sn(i) ¼ sn(rn  2  i) for all i ¼ 0, . . ., rn  2
infinite measure-preserving rank-one transforma- eventually in n.
tions. However, there are some classes of infinite Bergelson asked: Is there T of infinite ergodic
rank-one maps for which it is true: zero-type maps index such that T T 1 is not ergodic? Of course,
and partially bounded maps. such a T is asymmetric. The question is still open.
However some partial progress was achieved in
Definition 14.1 A s-finite self-joining (of order (Clancy et al. 2016; Danilenko 2016a). In (Clancy
2) of T is a s-finite T T-invariant measure l on et al. 2016), an example of a rank-one T was
(X X, ℬ  ℬ) such that l(A X) ¼ l(X A) ¼ constructed such that T T is ergodic but T
l(A) for all A  ℬ of finite measure. If for each T1 is not. Similar examples appeared also in
ergodic s-finite self-joining l of T, there is n  ℤ (Danilenko 2016a). However, they do not answer
such that l(A B) ¼ m(A TnB), then T is said Bergelson’s question because T has ergodic index
to have minimal s-finite self-joinings (of order 2). 2 in these examples. It was also shown in
The above concept of MSJ permits to control (Danilenko 2016a) that within the class of infinite
the m-preserving centralizer C0(T ) of T: If T has Markov shifts, the answer on Bergelson’s ques-
MSJ, then C0(T ) ¼ {Tn| n  ℤ}. It was shown by tion is negative. As for the rank-one transforma-
Ryzhikov and Thouvenot (Ryzhikov and tions, it was shown in (Clancy et al. 2016) that
Thouvenot 2015) that each zero-type T T1 is always conservative.
Ergodic Theory: Nonsingular Transformations 279

Ergodicity of Powers is to describe the set of those g  G such that Tg is

Let T be a rank-one infinite measure-preserving recurrent. Since Tg is recurrent if and only if Tng is
transformation associated with ðr n , on , sn Þ1 n¼1 : recurrent for every n  ℤ\{0}, it make sense to
(Hence on(0) ¼    ¼ on(rn  1).) For each talk about recurrent directions in ℤd or rational
n > 0, we denote by Cn the set of bottom levels of recurrent directions in the group ℝd containg ℤd
all copies of (n  1)-th tower in the n-th tower. as a standard lattice. Following Milnor’s general
Thus, formally, Cn ≔f0g [ jhn1 þ j1 i¼0 sn ðiÞjj ¼ idea of directional dynamics (Milnor 1988), John-
1, . . . , r n  1g: The following was proved in son and Şahin introduced in (2015) a concept of
(Danilenko 2019b). directional recurrence of T along an arbitrary
Theorem 14.3 (Ergodicity of powers of rank- (including irrational) direction in ℝd. They
one transformations). showed that the set R(T ) of recurrent directions
is a Gδ-subset, produced examples of T with trivial
(i) If Td is ergodic, then for each divisor p of d and nontrivial R(T), and asked about a description
there are infinitely many n such that some of all possible R(T) when T runs through the
c  Cn is not divisible by p. ergodic ℤd-action. Some partial answers were
(ii) If ðr n Þ1
n¼1 is bounded and for each divisor
obtained in (Danilenko 2017): Given a Gδ-subset
p of a positive integer d, there are infinitely Δ of the real projective space P(ℝd) and a count-
many n such that p does not divide some able subset D  Δ, there is a rank-one action
c  Cn, then Td is ergodic. T with D  R(T)  Δ. However, in general, the
(iii) If the sequence ðr n Þ1
n¼1 is bounded, then T is
problem remains open.
totally ergodic if and only if for each d > 1,
there are infinitely many n > 0 such that
some element c  Cn is not divisible by d. Applications. Connections with Other
Fields
Rigidity Sequences
Adams proved in (2015) that if S is an ergodic In this – final – section, we shed light on numerous
probability-preserving transformation such that mathematical sources of nonsingular systems.
Snk ! Id weakly for some sequence ðnk Þ1 k¼1 of
They come from the theory of stochastic pro-
positive integers, then there exists a power ratio- cesses, random walks, locally compact Cantor
nally weakly mixing infinite measure-preserving systems, horocycle flows on hyperbolic surfaces,
transformation T such that T nk ! Id weakly. The von Neumann algebras, statistical mechanics, rep-
converse assertion is obvious – just pass to the resentation theory for groups and anti-
Poisson suspension. (An infinite measure- commutation relations, etc. We also note that
preserving T is called power rationally weakly nonsingular systems can appear in the context of
mixing if for each finite sequence of nonzero probability-preserving dynamics, for example, in
integers l1, . . ., lk, the product transformation §11.1 some criteria for distality are given in terms
T l1    T lk is rationally weakly mixing.) of the Krengel entropy. See also §15.1 and §15.3
Thus the class of rigidity sequences for the ergo- below.
dic probability-preserving transformations equals
to the class of rigidity sequences for the ergodic Mild Mixing
infinite measure-preserving ones. An ergodic finite measure-preserving dynamical
system (X, ℬ, m, T ) is called mildly mixing if for
Directional Recurrence each nontrivial T-invariant s-algebra A  ℬ, the
restriction T A is not rigid. For equivalent defini-
+

Given an ergodic infinite measure-preserving ℤd -

action T ¼ T g g  ℤd , it seems natural to study tions and extensions to actions of locally compact
groups, we refer to (Aaronson 1997) and (Schmidt
the dynamics of individual transformations Tg,
and Walters 1982). There is an interesting crite-
g  ℤd. For instance, one of the natural questions
rion of the mild mixing that involves nonsingular
280 Ergodic Theory: Nonsingular Transformations

systems: T is mildly mixing if and only if for each Furstenberg in (1967) initiated studying the class
ergodic nonsingular transformation S, the product W ⊥ of transformations disjoint from all weakly
T S is ergodic (Furstenberg and Weiss 1978). mixing ones. Let D denote the class of distal
Furthermore, T is mildly mixing if and only if for transformations and MðW ⊥ Þ the class of multi-
each ergodic nonsingular transformation S, the pliers of W ⊥ (for the definitions, see (Glasner
product T S is orbit equivalent to S (Hawkins 1994)). Then D  MðW ⊥ Þ  W ⊥ : In
and Silva 1997/98). In particular, the associated (Danilenko and Lemańczyk 2005; Lemańczyk
flows of S and T S are isomorphic. Moreover, if and Parreau 2003), it was shown by constructing
R is a nonsingular transformation such that R S explicit examples that these inclusions are strict.
is ergodic for any ergodic nonsingular S, then R is We record this fact here because nonsingular ergo-
of type II1 (and mildly mixing) (Schmidt and dic theory was the key ingredient of the arguments
Walters 1982). In this context, we note that for in the two papers pertaining to the theory of
every ergodic infinite measure-preserving trans- probability-preserving systems. The examples
formation T there is an ergodic Markov shift are of the form T’,S(x, y) ¼ (Tx, S’(x)y), where
S such that T S is not conservative, hence not T is an ergodic rotation on (X, m), (Sg)g  G a
ergodic (Aaronson et al. 1979); also that S can be mildly mixing action of a locally compact group
chosen to be rank-one and rigid (Elyze G on Y and ’ : X ! G a measurable map. Let W’
et al. 2018). denote the Mackey action of G associated with ’,
and let (Z, k) be the space of this action. The key
Ergodicity of Gaussian Cocycles observation is that there exists an affine isomor-
Let T be an ergodic (equivalently, weakly mixing) phism of the simplex of T’,S-invariant probability
measure-preserving Gaussian transformation on a measures whose pullback on X is m and the sim-
standard probability space ðX, B, mÞ, and let H be plex of W’ S quasi-invariant probability mea-
the corresponding invariant Gaussian subspace of sures whose pullback on Z is k and whose Radon-
the real Hilbert space L20 ðX, mÞ≔L2 ðX, mÞ  ℝ: Nikodym cocycle is measurable with respect to Z.
The following conjecture was stated in This is a far-reaching generalization of
(Lemańczyk et al. 2001): For each function Furstenberg theorem on relative unique ergodicity
f  H, either f is a T-coboundary (equivalently, a of ergodic compact group extensions.
Gaussian coboundary, i.e., the transfer function
belongs to H ), or the skew product transformation Symmetric Stable and Infinitely Divisible
Tf acting on the product space (X ℝ, m Leb) is Stationary Processes
ergodic. We note that Tf preserves the infinite Rosinsky in (1995) established a remarkable con-
measure m Leb. The affirmative answer was nection between structural studies of stationary
obtained in (Danilenko and Lemańczyk 2022) stochastic processes and ergodic theory of non-
(and independently in (Marrakchi and Vaes singular transformations (and flows). For simplic-
2022)) under an assumption that T is mildly ity, we consider only real processes in discrete
mixing. Some examples of ergodic Tf with rigid time. Let X ¼ (Xn)n  ℤ be a measurable stationary
weakly mixing T were constructed in (Marrakchi symmetric α-stable (SαS) process, 0 < α < 2.
and Vaes 2022). However, in the general setting, This means that any linear combination
n
the problem remains open. k¼1 ak X jk , jk  ℤ, ak  ℝ has an SαS-
distribution. (The case α ¼ 2 corresponds to
Disjointness and Furstenberg’s Class W ⊥ Gaussian processes.) Then the process admits a
Two probability-preserving systems (X, m, T ) and spectral representation
(Y, v, S) are called disjoint if m n is the only T
S-invariant probability measure on X Y whose
Xn ¼ f n ðyÞMðdyÞ, n  ℤ, ð10Þ
coordinate projections are m and v, respectively. Y
Ergodic Theory: Nonsingular Transformations 281

where fn  Lα(Y, m) for a standard s-finite mea- Poisson Suspensions of Infinite Measure-
sure space (Y, ℬ, m) and M is an independently Preserving Transformations
scattered random measure on ℬ such that Poisson suspensions over infinite measure-
E exp (iuM(A)) ¼ exp (|u|αm(A)) for every preserving transformations (see (Roy 2005,
A  ℬ of finite measure. By (Rosinsky 1995), 2009)) are widely used in statistical mechanics
one can choose the kernel ( fn)n  ℤ in a special to model ideal gas, Lorentz gas, etc. (see
way: There are a m - nonsingular transformation (Cornfeld et al. 1982)). Being a particular case of
T and measurable maps ’ : X ! {1, 1} and the Poisson suspensions over nonsingular trans-
f  Lα(Y, m) such that fn ¼ Unf, n  ℤ, where formations (see §3.7), they are always probability
U is the isometry of Lα(X, m) given by Ug ¼ preserving. Together with the Gaussian dynamical
’  (dm ∘ T/dm)1/α  g ∘ T. If, in addition, the systems, they are also an important source of
smallest T-invariant s-algebra containing examples and counterexamples in ergodic theory.
f1(ℬℝ) coincides with ℬ and Supp Due to a close similarity with the well-studied
{f ∘ Tn : n  ℤ} ¼ Y, then the pair (T, ’) is called Gaussian systems, a natural question arises: Are
minimal. It turns out that minimal pairs always there ergodic Poisson suspensions whose ergodic
exist. Moreover, two minimal pairs (T, ’) and self-joinings are Poisson? Such suspensions are
(T0, ’0) representing the same SαS process are called PAP. They are analogue of GAG in the
equivalent in some natural sense (Rosinsky theory of Gaussian systems (Lemańczyk et al.
1995). Then one can relate ergodic-theoretical 2000). Janvresse, de la Rue, and Roy constructed
properties of (T, ’) to probabilistic properties of PAP suspension in (Janvresse et al. 2017) (see also
(Xn)n  ℤ. For instance, let Y ¼ C t D be the Hopf (Parreau and Roy 2008)). The example of an infi-
decomposition of Y (see Theorem 2.2). We let nite measure-preserving T with “minimal self-
XDn ≔ D f n ðyÞMðdyÞ and XCn ≔ C f n ðyÞMðdyÞ: joinnings” from (Janvresse et al. 2019) plays a
Then we obtain a unique (in distribution) decom- crucial role in their construction. We also mention
position of X into the sum XD þ XC of two inde- a result of Meyerovitch (Meyerovitch 2013) related
pendent stationary SαS-processes. to weak mixing of infinite measure-preserving sys-
Another kind of decomposition was consid- tems and Poisson suspensions: If T is a conservative
ered in (Samorodnitsky 2005). Let P be the largest ergodic infinite measure-preserving transformation,
invariant subset of Y such that T P has a finite
+

then the direct product of T with the Poisson sus-

invariant measure. Partitioning Y into P and pension T of T is ergodic.
N ≔ Y\N and restricting the integration in (10) to
P and N, we obtain a unique (in distribution) Recurrence of Random Walks with
decomposition of X into the sum XP þ XN of two Nonstationary Increments
independent stationary SαS-processes. Notice that Using nonsingular ergodic theory, one can intro-
the process X is ergodic if and only if m(P) ¼ 0. duce the notion of recurrence for random walks
Roy considered a more general class of infi- obtained from certain nonstationary processes.
nitely divisible (ID) stationary processes (Roy Let T be an ergodic nonsingular transformation
2007). Using Maruyama’s representation of the of a standard probability space (X, ℬ, m), and let
characteristic function of an ID process f : X ! ℝn be a measurable function. Define for
X without Gaussian part, he singled out the Lévy m  1, Ym : X ! ℝn by Y m ≔ m1 n¼0 f ∘ T : In
n

measure Q of X. Then Q is a shift invariant s-finite other words, (Ym)m1 is the random walk associ-
measure on ℝℤ. Decomposing the dynamical sys- ated with the (nonstationary) process ( f ∘ T n)n0.
tem (ℝℤ, t, Q) in various natural ways (Hopf Let us call this random walk recurrent if the
decomposition, 0-type and positive type, cocycle f of T is recurrent (see §5.5). It was
so-called “rigidity free” part, and its complement), shown in (Schmidt 1984) that in the case m ∘ T ¼
he obtains corresponding decompositions for the m, i.e., the process is stationary, this definition is
process X. Here t stands for the shift on ℝℤ. equivalent to the standard one.
282 Ergodic Theory: Nonsingular Transformations

Boundaries of Random Walks of homogeneous random walks, this definition is

Boundaries of random walks on groups retain equivalent to the initial one. Connes and Woods
valuable information on the underlying groups showed (Connes and Woods 1989) that the tail
(amenability, entropy, etc.) and enable one to boundary is always amenable and AT. It is
obtain integral representation for harmonic func- unknown whether the converse holds for general
tions of the random walk (Zimmer 1977, 1978; G. However, it is true for G ¼ ℝ and G ¼ ℤ: the
Kaimanovich and Vershik 1983). Let G be a class of AT-flows coincides with the class of tail
locally compact group and v a probability measure boundaries of the random walks on ℝ, and a
on G. Let T denote the (one-sided) shift on the similar statement holds for ℤ (Connes and
probability space ðX, ℬX , mÞ≔ðG, ℬG , nÞℤþ and Woods 1989). Jaworski showed (Jaworski 1994)
’ : X ! G a measurable map defined by that if G is countable and a random walk is homo-
(y0, y1, . . .) 7! y0. Let T’ be the ’ - skew product geneous, then the tail boundary of the random
extension of T acting on the space (X G, m walk possesses a so-called SAT-property (which
lG) (for noninvertible transformations, the skew is stronger than AT).
product extension is defined in the very same way
as for invertible ones; see §5.5). Then T’ is iso- Stationary Actions
morphic to the homogeneous random walk on Let T ¼ (Tg)g  G be a continuous action of a count-
G with jump probability v. Let I (T’) denote the able group G on a compact metrizable space X. By
sub-s-algebra of T’-invariant sets, and let Markov-Kakutani theorem, if G is amenable, then
F T ’ ≔ \n>0 T n ’ ðℬX  ℬG Þ: The former is there is an invariant probability Borel measure on
called the Poisson boundary of T’, and the latter K. If G is nonamenable, such a measure does not
one is called the tail boundary of T’. Notice that a necessarily exist. However, if k is a probability
nonsingular action of G by inverted right trans- measure on G whose support generates G as a
lations along the second coordinate is well defined semigroup, then there is always a T-quasiinvariant
on each of the two boundaries. The two bound- probability measure m on X such that
aries (or, more precisely, the G-actions on them)
are ergodic. The Poisson boundary is the Mackey dm∘T 1
g
k ð gÞ ðxÞ ¼ 1 for a:e: x  X:
range of ’ (as a cocycle of T ). Hence the Poisson dm
gG
boundary is amenable (Zimmer 1978). If the sup-
port of v generates a dense subgroup of G, then the m is called a k-stationary measure. For a deep
corresponding Poisson boundary is weakly theory of stationary actions and its applications,
mixing (Aaronson and Lemańczyk 2005). As for we refer to (Furman 2002; Furstenberg and
the tail boundary, we first note that it can be Glasner 2010) and references therein. However,
defined for a wider family of nonhomogeneous if G is Abelian or, more generally, nilpotent, then
random walks. This means that the jump proba- each k-stationary action is invariant under T.
bility v is no longer fixed and a sequence (nn)n>0 Thus, there are no stationary ℤ-actions except
of probability measures on G is considered for the probability-preserving ones. That is why
instead. Now let (X, ℬX, m) ≔ ∏n>0(G, ℬG, n). we do not discuss them in this survey.
The one-sided shift on X may not be nonsingular
now. Instead of it, we consider the tail equivalence Classifying s-finite Ergodic Invariant
relation R on X and a cocycle α : R ! G given by Measures
aðx, yÞ ¼ x1   xn y1
n   y1 , where x ¼ (xi)i > 0 and The description of ergodic finite invariant mea-
y ¼ (yi)i>0 are R-equivalent and n in the smallest sures for topological (or, more generally, standard
integer such that xi ¼ yi for all i > n. The tail Borel) systems is a well-established problem in
boundary of the random walk on G with time- the classical ergodic theory (Cornfeld et al. 1982).
dependent jump probabilities (nn)n>0 is the On the other hand, it seems impossible to obtain
Mackey G-action associated with α. In the case any useful information about the system by
Ergodic Theory: Nonsingular Transformations 283

analyzing the set of all ergodic quasi-invariant on M. According to Ratner, h has no finite invari-
(or just s-finite invariant) measures because this ant measures on M if G is infinite (except for
set is wildly huge (see §2.6). The situation measures supported on closed orbits). However
changes if we impose some restrictions on the there are infinite invariant Radon measures, for
measures. For instance, if the system under ques- instance, the volume measure. In the case when
tion is a homeomorphism (or a topological flow) G is free Abelian and Γ0 is cocompact, every
defined on a locally compact Polish space, then it homomorphism ’ : G ! ℝ determines a unique
is natural to consider the class of (s-finite) invari- up to scaling ergodic invariant Radon measure
ant Radon measures, i.e., measures taking finite (e.i.r.m.) m on T1(M) such that m ∘ dD ¼ exp
values on the compact subsets. We give two (’(D))m for all D  G (Babillot and Ledrappier
examples. 1998) and every e.i.r.m. arises this way (Sarig
First, the seminal results of Giordano, Putnam, 2004). Moreover, all these measures are quasi-
and Skau on the topological orbit equivalence of invariant under g. In the general case, an interest-
compact Cantor minimal systems were extended ing bijection is established in (Ledrappier and
to locally compact Cantor minimal (l.c.c.m.) sys- Sarig 2007) between the e.i.r.m. which are quasi-
tems in (Danilenko 2001b; Matui 2002). Given a invariant under g and the “non-trivial minimal”
l.c.c.m. system X, we denote by M(X) and M1(X) positive eigenfunctions of the hyperbolic
the set of invariant Radon measures and the set of Laplacian on M. This result was extended in the
invariant probability measures on X. Notice that recent work (Landesberg and Lindenstrauss
M1(X) may be empty (Danilenko 2001b). It was 2022).
shown in (Matui 2002) that two systems X and X 0
are topologically orbit equivalent if and only if Von Neumann Algebras
there is a homeomorphism of X onto X 0 which There is a fascinating and productive interplay
maps bijectively M(X) onto M(X 0) and M1(X) between nonsingular ergodic theory and von Neu-
onto M1(X 0). Thus M(X) retains an important mann algebras. The two theories alternately
information on the system – it is “responsible” influenced development of each other. Let
for the topological orbit equivalence of the under- (X, ℬ, m, T ) be a nonsingular dynamical system.
lying systems. Uniquely ergodic l.c.c.m. systems Given ’  L1(X, m) and j  ℤ, we define
(with unique up to scaling infinite invariant Radon operators A’ and Uj on the Hilbert space L2(Z
measure) were constructed in (Danilenko 2001b). ℤ, m n) by setting
The second example is related to study of the
smooth horocycle flows on tangent bundles of A’ f ðx, iÞ≔’ T i x f ðx, iÞ, U j f ðx, iÞ≔f ðx, i jÞ
hyperbolic surfaces. Let  be the open disk
equipped with the hyperbolic metric jdz j/(1  Then U j A’ U j ¼ A’∘T : Denote by M the
j

|z|2), and let MöbðÞ denote the group of Möbius von Neumann algebra (i.e., the weak closure of
transformations of . A hyperbolic surface can be the *-algebra) generated by A’, ’  L1(X, m)
written in the form M ≔ G∖MöbðÞ, where Γ is a and Uj, j  ℤ. If T is ergodic and aperiodic, then
torsion-free discrete subgroup of MöbðÞ: Sup- M is a factor, i.e., M \ M0 ¼ ℂ1, where M0
pose that Γ is a nontrivial normal subgroup of a denotes the algebra of bounded operators com-
lattice Γ0 in MöbðÞ: Then M is a regular cover of muting with M. It is called a Krieger’s factor.
the finite volume surface M0 ≔ G0 ∖MöbðÞ: The Murray-von Neumann-Connes’ type of M is
group of deck transformations G ¼ Γ0/Γ is finitely exactly the Krieger’s type of T. The flow of
generated. The horocycle flow (ht)t  ℝ and the weights of M is isomorphic to the associated
geodesic flow (gt)t  ℝ defined on the unit tangent flow of T. Two Krieger’s factors are isomorphic
bundle T 1 ðÞ descend naturally to flows, say if and only if the underlying dynamical systems
h and g, on T1(M). We consider the problem of are orbit equivalent (Krieger 1976b). Moreover, a
classification of the h-invariant Radon measures number of important problems in the theory of
284 Ergodic Theory: Nonsingular Transformations

von Newmann algebras such as classification of ℋ, and a Γ-cocycle (Ck)k1. Moreover, using
subfactors, computation of the flow of weights nonsingular ergodic theory Golodets (Golodets
and Connes’ invariants, outer conjugacy for auto- 1969) constructed for each k ¼ 2, 3, . . ., 1, an
morphisms, etc. are intimately related to the irreducible representation of CAR such that
corresponding problems in nonsingular orbit the- dim ℋ ¼ k. This answered a question of Gårding
ory. We refer to (Moore 1982; Feldman and and Wightman (Gårding and Wightman 1954)
Moore 1977; Giordano and Skandalis 1985a, b; who considered only the case k ¼ 1.
Hamachi and Kosaki 1993; Danilenko and
Hamachi 2000) for details. Unitary Representations of Locally Compact
Groups
Representations of CAR Nonsingular actions appear in a systematic way in
Representations of canonical anticommutation the theory of unitary representations of groups.
relations (CAR) is one of the most elegant and Let G be a locally compact second countable
useful chapters of mathematical physics, provid- group and H a closed normal subgroup of G.
ing a natural language for many body quantum Suppose that H is commutative (or, more gener-
physics and quantum field theory. By a represen- ally, of type I, see (Dixmier 1969)). Then the
tation of CAR, we mean a sequence of bounded natural action of G by conjugation on H induces
linear operators a1, a2, . . . in a separable Hilbert a Borel G-action, say α, on the dual space H – the
space K such that ajak þ akaj ¼ 0 and aj ak þ set of unitarily equivalent classes of irreducible
ak aj ¼ dj,k : unitary representations of H. If now U ¼ (Ug)g  G
Consider {0, 1} as a group with addition mod is a unitary representation of G in a separable
2. Then X ¼ {0, 1}ℕ is a compact Abelian group. Hilbert space, then by applying Stone decompo-
Let Γ ≔ {x ¼ (x1, x2, . . .) : limn!1xn ¼ 0}. Then sition theorem to U H one can deduce that α is
+
Γ is a dense countable subgroup of X. It is gener- nonsingular with respect to a measure m of the
ated by elements γk whose k-coordinate is 1 and “maximal spectral type” for U H on H: More-
+
the other ones are 0. Γ acts on X by translations. over, if U is irreducible, then α is
Let m be an ergodic Γ-quasi-invariant measure on ergodic. Whenever m is fixed, we obtain a one-
X. Let (Ck)k1 be Borel maps from X to the group to-one correspondence between the set of coho-
of unitary operators in a Hilbert space ℋ satisfy- mology classes of irreducible cocycles for α with
ing Ck ðxÞ ¼ Ck ðx þ dk Þ, Ck ðxÞCl ðx þ dl Þ ¼ values in the unitary group on a Hilbert space ℋ
Cl ðxÞCk ðx þ dk Þ, k 6¼ l for a.a. x. In other words, and the subset of G consisting of classes of those
(Ck)k1 defines a cocycle of the Γ-action. We now unitary representations V for which the measure
+

put ℋ≔L2 ðX, mÞ  ℋ and define operators ak in associated to V H is equivalent to m. This corre-
ℋ by setting spondence is used in both directions. From infor-
mation about cocycles, we can deduce facts about
representations and vise versa (Kirillov 1978;
ðak f ÞðxÞ ¼ ð1Þx1 þþxk1 ð1 xk ÞCk ðxÞ
Dixmier 1969).
dm∘dk
ðxÞf ðx þ dk Þ,
dm
Further Directions
where f : X ! ℋ is an element of ℋ and x ¼
(x1, x2, . . .)  X. It is easy to verify that a1, a2, . . . While some of the results that we have cited for
is a representation of CAR. The converse was nonsingular ℤ-actions extend to actions of locally
established in (Gårding and Wightman 1954; compact Polish groups (or subclasses of Abelian
Golodets 1969): Every factor-representation (this or amenable ones), many natural questions remain
means that the von Neumann algebra generated by open in the general setting. For instance, what is
all ak is a factor) of CAR can be represented as the Rokhlin lemma, or the pointwise ergodic the-
above for some ergodic measure m, Hilbert space orem (for some obstacles toward extension of the
Ergodic Theory: Nonsingular Transformations 285

ratio ergodic theorem to nonsingular actions of Aaronson J, Nadkarni M (1987) L1 eigenvalues and L2
arbitrary amenable groups, see (Hochman 2013); spectra of nonsingular transformations. Proc Lond
Math Soc 55(3):538–570
a weak version of this theorem was proved Aaronson J, Nakada H (2000) Multiple recurrence of Mar-
recently in (Danilenko 2019a)), or the definition kov shifts and other infinite measure-preserving trans-
of entropy for nonsingular actions of general formations. Israel J Math 117:285–310
countable amenable groups? The theory of Aaronson J, Park KK (2009) Predictability, entropy and
information of infinite transformations, Fundam Math
abstract nonsingular equivalence relations 206:1–21
(Feldman and Moore 1977) or, more generally, Aaronson J, Weiss B (2004) On Herman’s theorem for
nonsingular groupoids (Ramsay 1971) and poly- ergodic, amenable group extensions of endomor-
morphisms (Vershik 1983) is also a beautiful part phisms. Ergodic Theory Dyn Syst 24(5):1283–1293
Aaronson J, Weiss B (2018) Distributional limits of posi-
of nonsingular ergodic theory that has nice appli- tive, ergodic stationary processes and infinite ergodic
cations: description of semifinite traces of transformations. Ann Inst Henri Poincaré Probab Stat
AF-algebras, classification of factor representa- 54:879–906
tions of the infinite symmetric group (Vershik and Aaronson J, Zweimüller R (2014) Limit theory for some
positive stationary processes with infinite mean,
Kerov 1985), path groups (Albeverio et al. 1983), English, with English and French summaries. Ann
etc. Nonsingular ergodic theory needs different Inst Henri Poincaré Probab Stat 50(1):256–284
tools for the most part when we pass from ℤ- Aaronson J, Lin M, Weiss B (1979) Mixing properties of
actions to noninvertible endomorphisms or, more Markov operators and ergodic transformations, and
ergodicity of Cartesian products. Israel J Math 33:
generally, semigroup actions. Several concrete 198–224
open problems are mentioned throughout the entry. Aaronson J, Kosloff Z, Weiss B (2017) Symmetric
Birkhoff sums in infinite ergodic theory. Ergodic The-
ory Dyn Syst 37:2394–2416
Abdalaoui EH el, Nadkarni MG (2016) A non-singular
Bibliography transformation whose spectrum has Lebesgue compo-
nent of multiplicity one. Ergodic Theory Dyn Syst 36:
Aaronson J (1977a) Rational ergodicity and a metric 671–681
invariant for Markov shifts. Israel J Math 27:93–123 Adams TM (1998) Smorodinsky’s conjecture on rank-one
Aaronson J (1977b) On the ergodic theory of non- mixing. Proc Am Math Soc 126:739–744
integrable functions and infinite measure spaces. Israel Adams TM (2015) Rigidity sequences of power rationally
J Math 27(2):163–173 weakly mixing transformations, preprint,
Aaronson J (1979) Rational ergodicity, bounded rational arXiv:1503.05806v1
ergodicity and some continuous measures on the circle. Adams T, Silva CE (2015) On infinite transformations with
Israel J Math 33:181–197 maximal control of ergodic two-fold product powers.
Aaronson J (1981) The asymptotic distributional behav- Israel J Math 209:929–948
iour of transformations preserving infinite measures. Adams T, Silva CE (2016) Weak rational ergodicity does
J Analyse Math 39:203–234 not imply rational ergodicity. Israel J Math 214:
Aaronson J (1983) The eigenvalues of nonsingular trans- 491–506
formations. Israel J Math 45:297–312 Adams T, Silva CE (2018) Weak mixing for infinite mea-
Aaronson J (1987) The intrinsic normalizing constants of sure invertible transformations. In: Ergodic theory and
transformations preserving infinite measures. dynamical systems in their interactions with arithmetics
J. Analyse Math. 49:239–270 and combinatorics, Lecture notes in Mathematics,
Aaronson J (1997) An introduction to infinite ergodic 2213, Springer, Cham, pp 327–349
theory. American Mathematical Society, Providence, Adams S, Elliott GA, Giordano T (1994) Amenable actions
RI, USA of groups. Trans Am Math Soc 344:803–822
Aaronson J (2013) Rational weak mixing in infinite mea- Adams T, Friedman N, Silva CE (1997) Rank-one weak
sure spaces. Ergodic Theory Dyn Syst 33:1611–1643 mixing for nonsingular transformations. Israel J Math
Aaronson J (2016) Conditions for rational weak mixing. 102:269–281
Stoch Dyn 16:1660004 Adams T, Friedman N, Silva CE (2001) Rank one power
Aaronson J, Lemańczyk M (2005) Exactness of Rokhlin weak mixing for nonsingular transformations. Ergodic
endomorphisms and weak mixing of Poisson bound- Theory Dyn Syst 21:1321–1332
aries, algebraic and topological dynamics. In: Contem- Ageev ON, Silva CE (2001) Genericity of rigid and mul-
porary mathematics, 385. American Mathematical tiply recurrent infinite measure-preserving and non-
Society, Providence, pp 77–88 singular transformations, Proceedings of the 16th
286 Ergodic Theory: Nonsingular Transformations

summer conference on general topology and its appli- Choksi JR, Kakutani S (1979) Residuality of ergodic mea-
cations. Topology Proc 26(2):357–365, 2001/02 surable transformations and of ergodic transformations
Albeverio S, Hoegh-Krohn R, Testard D, Vershik AM which preserve an infinite measure. Ind Univ Math
(1983) Factorial representations of Path groups. J 28:453–469
J Funct Anal 51:115–231 Choksi JR, Nadkarni MG (1994) The maximal spectral
Alpern S, Prasad VS (1990) Return times for nonsingular type of a rank one transformation. Can Math Bull
measurable transformations. J Math Anal Appl 152: 37(1):29–36
470–487 Choksi JR, Nadkarni MG (2000) Genericity of nonsingular
Araki H, Woods EJ (1968) A classification of factors. Pub transformations with infinite ergodic index. Colloq
RIMS, Ser A 3:51–130 Math 84/85:195–201
Arano Y, Isono Y, Marrakchi M (2021) Ergodic theory of Choksi JR, Prasad VS (1983) Approximation and Baire
affine isometric actions on Hilbert spaces. Geom Funct category theorems in ergodic theory. In: Measure theory
Anal 31(5):1013–1094 and its applications (Sherbrooke, Que., 1982), Lecture
Atkinson G (1976) Recurrence of co-cycles and random notes in mathematics 1033. Springer, Berlin, pp 94–113
walks. J Lond Math Soc 13:486–488 Choksi JR, Hawkins JM, Prasad VS (1987) Abelian
Avila A, Bochi J (2007) Generic expanding maps without cocylces for nonsingular ergodic transformations and
absolutely continuous invariant s-finite measure. Math the genericity of type III1 transformations. Monat fur
Res Lett 14(5):721–730 Math 103:187–205
Avraham-Re’em N (2022) On absolutely continuous Choksi J, Eigen S, Prasad V (1989) Ergodic theory on
invariant measures and Krieger-type of Markov sub- homogeneous measure algebras revisited, Measure
shifts. J d’Anal Math 147:201–253 and measurable dynamics (Rochester, NY, 1987). In:
Babillot M, Ledrappier F (1998) Geodesic paths and Contemporary Mathematics 94. American Mathemati-
horocycle flow on abelian covers. In: Lie groups and cal Society, Providence, pp 73–85
ergodic theory (Mumbai, 1996), Tata Institute of Fun- Clancy J, Friedberg R, Kasmalkar I, Loh I, Padurariu T,
damental Research studies in Mathematics, 14, Tata Silva CE, Vasudevan S (2016) Ergodicity and
Institute of Fundamental Research, Bombay, pp 1–32 conservativity of products of infinite transformations
Berendschot T, Vaes S (2022a) Nonsingular Bernoulli actions and their inverses. Colloq Math 143:271–291
of arbitrary Krieger type. Anal PDE 15:1313–1373 Connes A (1975) On the hierarchy of W. Krieger. Ill J Math
Berendschot T, Vaes S (2022b) Bernoulli actions of type 19:428–432
III0 with prescribed associated flow. Journal of the Connes A, Krieger W (1977) Measure space automor-
Institute of Mathematics of Jussieu, to appear phisms, the normalizers of their full groups, and
Bergelson V, Leibman A (1996) Polynomial extensions of approximate finiteness. J Funct Anal 24(4):336–352
van der Waerden’s and Semerédi’s theorems. J Am Connes A, Woods EJ (1985) Approximately transitive
Math Soc 9:725–753 flows and ITPFI factors. Ergodic Theory Dyn Syst
Bezuglyi SI, Golodets VY (1985) Groups of measure space 5(2):203–236
transformations and invariants of outer conjugation for Connes A, Woods EJ (1989) Hyperfinite von Newmann
automorphisms from normalizers of type III full algebras and Poisson boundaries of time dependent
groups. J Funct Anal 60(3):341–369 random walks. Pacific J Math 37:225–243
Bezuglyi SI, Golodets VY (1991) Weak equivalence and Connes A, Feldman J, Weiss B (1981) An amenable equiv-
the structures of cocycles of an ergodic automorphism. alence relation is generated by a single transformation.
Publ Res Inst Math Sci 27(4):577–625 Ergodic Theory Dyn Syst 1:431–450
Björklund M, Kosloff Z, Vaes S (2021) Ergodicity and type Cornfeld IP, Fomin SV, Sinaĭ YG (1982) Ergodic theory,
of nonsingular Bernoulli actions. Invent Math 224(2): Grundlehren der Mathematischen Wissenschaften,
573–625 245. Springer–Verlag, New York
Bonanno C, Giulietti P, Lenci M (2018) Infinite mixing for Dai I, Garcia X, Padurariu T, Silva CE (2015) On rationally
one-dimensional maps with an indifferent fixed point. ergodic and rationally weakly mixing rank-one trans-
Nonlinearity 31:5180–5213 formations. J Ergodic Theory Dyn Syst 35:1141–1164
Bowles A, Fidkowski L, Marinello A, Silva CE Danilenko AI (1995) The topological structure of Polish
(2001) Double ergodicity of nonsingular transforma- groups and groupoids of measure space transforma-
tions and infinite measure-preserving staircase trans- tions. Publ Res Inst Math Sci 31(5):913–940
formations. Ill J Math 45(3):999–1019 Danilenko AI (1998) Quasinormal subrelations of ergodic
Bozgan F, Sanchez A, Silva CE, Stevens D, Wang J (2015) equivalence relations. Proc Am Math Soc 126(11):
Subsequence bounded rational ergodicity of rank-one 3361–3370
transformations. Dynamical Syst 30:70–84 Danilenko AI (2001a) Funny rank one weak mixing for
Bruin H, Melbourne I, Terhesiu D (2019) Rates of mixing nonsingular Abelian actions. Israel J Math 121:29–54
for non-Markov infinite measure semi-flows. Trans Am Danilenko AI (2001b) Strong orbit equivalence of locally
Math Soc 371:7343–7386 compact Cantor minimal systems. Int J Math 12:113–123
Chacon RV, Friedman NA (1965) Approximation and Danilenko AI (2004) Infinite rank one actions and non-
invariant measures. Z Wahrscheinlichketstheorie und singular Chacon transformations. Ill J Math 48(3):
Verw Gebiete 3:286–295 769–786
Ergodic Theory: Nonsingular Transformations 287

Danilenko AI (2007) (C, F)-actions in ergodic theory, in Danilenko AI, Ryzhikov VV (2010) Spectral multiplicities
“Geometry and Dynamics of Groups and Spaces”. of infinite measure preserving transformations. Funct
Progr Math 265:325–351 Anal Appl 44:161–170
Danilenko AI (2013) A survey on spectral multiplicities Danilenko AI, Ryzhikov VV (2011) Mixing constructions
of ergodic actions. Ergodic Theory Dyn Syst 33: with infinite invariant measure and spectral multiplici-
81–117 ties. Ergodic Theory Dyn Syst 31:853–873
Danilenko AI (2016a) Finite ergodic index and asymmetry Danilenko AI, Ryzhikov VV (2012) On self-similarities of
for infinite measure preserving actions. Proc Am Math ergodic flows. Proc Lond Math Soc 104:431–454
Soc 144:2521–2532 Danilenko AI, Silva CE (2004) Multiple and polynomial
Danilenko AI (2016b) Actions of finite rank: weak rational recurrence for Abelian actions in infinite measure.
ergodicity and partial rigidity. J Ergodic Theory Dyn J Lond Math Soc 69(2):183–200
Syst 36:2138–2171 Danilenko AI, Solomko AV (2009) Infinite measure pre-
Danilenko AI (2017) Directional recurrence and direc- serving flows with infinite ergodic index. Colloq Math
tional rigidity for infinite measure preserving actions 115:13–19
of nilpotent lattices. Ergodic Theory Dyn Syst 37: Danilenko AI, Kosloff Z, Roy E (2022a) Nonsingular
1841–1861 Poisson suspensions. J d’Anal Math, 146:741–790
Danilenko AI (2018) Infinite measure preserving transfor- Danilenko AI, Kosloff Z, Roy E (2022b) Generic non-
mations with Radon MSJ. Israel J Math 228:21–51 singular Poisson suspension is of type III1. Ergodic
Danilenko AI (2019a) Weak mixing for nonsingular Theory Dyn Syst 42(4):1415–1445
Bernoulli actions of countable amenable groups. Proc Day S, Grivna B, McCartney E, Silva CE (1999) Power
Am Math Soc 147:4439–4450 weakly mixing infinite transformations. New York
Danilenko AI (2019b) Rank-one actions, their (C,F)- J Math 5:17–24
models and constructions with bounded parameters. del Junco A (1978) A simple measure-preserving transfor-
J d’Anal Math 139(2):697–749 mation with trivial centralizer. Pac J Math 79:357–362
Danilenko AI (2021) On the bounded cohomology for del Junco A, Rudolph DJ (1987) On ergodic actions whose
ergodic nonsingular actions of amenable groups. Israel self-joinings are graphs. Ergodic Theory Dyn Syst 7:
J Math 243(1):421–436 531–557
Danilenko AI (2022) Krieger’s type for ergodic non- del Junco A, Şahin A (2009) Dyesś theorem in the almost
singular Poisson actions of non-(T) locally compact continuous category. Israel J Math 173:235–251
groups. Ergodic Theory Dyn Systems, to appear del Junco A, Silva CE (1995) Prime type IIIl automor-
Danilenko AI, del Junco A (2011) Almost continuous orbit phisms: an instance of coding techniques applied to
equivalence for non-singular homeomorphisms. Israel nonsingular maps. In: Takahashi Y (ed) Fractals and
J Math 183:165–188 dynamics (Okayama/Kyoto, 1992). Plenum, New York,
Danilenko AI, Golodets VY (1996) On extension of pp 101–115
cocycles to normalizer elements, outer conjugacy, and del Junco A, Silva CE (2003) On factors of nonsingular
related problems. Trans Am Math Soc 348(12): Cartesian products. Ergodic Theory Dyn Syst 23(5):
4857–4882 1445–1465
Danilenko AI, Hamachi T (2000) On measure theoretical Derriennic Y, Frączek K, Lemańczyk M, Parreau F (2008)
analogues of the Takesaki structure theorem for type III Ergodic automorphisms whose weak closure of off-
factors. Colloq Math 84/85:485–493 diagonal measures consists of ergodic self-joinings.
Danilenko AI, Kosloff Z (2022) Krieger’s type of non- Colloq Math 110:81–115
singular Poisson suspensions and IDPFT systems. Dixmier J (1969) Les C*-algebres et leurs représentations.
Proc Am Math Soc 150(4):1541–1557 Gauthier–Villars Editeur, Paris
Danilenko AI, Lemańczyk M (2005) A class of multipliers Dolgopyat D, Nándori P (2019) Infinite measure renewal
for W ⊥. Isr J Undergrad Math 148:137–168 theorem and related results. Bull Lond Math Soc 51:
Danilenko AI, Lemańczyk M (2019) K-property for 145–167
Maharam extensions of non-singular Bernoulli and Dooley AN, Hagihara R (2012) Computing the critical
Markov shifts. Ergodic Theory Dyn Syst 39(12): dimensions of Bratteli-Vershik systems with multiple
3292–3321 edges. Ergodic Theory Dyn Syst 32:103–117
Danilenko AI, Lemańczyk M (2022) Ergodic cocycles of Dooley AH, Hamachi T (2003a) Nonsingular dynamical
IDPFT systems and non-singular Gaussian actions. systems, Bratteli diagrams and Markov odometers.
Ergodic Theory Dyn Syst 42(5):1624–1654 Israel J Math 138:93–123
Danilenko AI, Park KK (2011) Rank-one flows of trans- Dooley AH, Hamachi T (2003b) Markov odometer actions
formations with infinite ergodic index. Proc Am Math not of product type. Ergodic Theory Dyn Syst 23:
Soc 139:201–207 813–829
Danilenko AI, Rudolph DJ (2009) Conditional entropy Dooley AH, Mortiss G (2006) On the critical dimension
theory in infinite measure and a question of Krengel. and AC entropy for Markov odometers. Monatsh Math
Israel J Math 172:93–117 149:193–213
288 Ergodic Theory: Nonsingular Transformations

Dooley AH, Mortiss G (2007) The critical dimensions of Gaebler J, Kastner A, Silva CE, Xu X, Zhou Z (2018)
Hamachi shifts. Tohoku Math J. 59(2):57–66 Partially bounded transformations have trivial central-
Dooley AH, Mortiss G (2009) On the critical dimension of izers. Proc Am Math Soc 146:5113–5127
product odometers. Ergodic Theory Dyn Syst 29: Gårding L, Wightman AS (1954) Representation of anti-
475–485 commutation relations. Proc Natl Acad Sci U S A 40:
Dooley AH, Klemes I, Quas AN (1998) Product and Mar- 617–621
kov measures of type III. J Aust Math Soc Ser A 65(1): Giordano T, Skandalis G (1985a) Krieger factors isomor-
84–110 phic to their tensor square and pure point spectrum
Dye H (1959) On groups of measure-preserving transfor- flows. J Funct Anal 64(2):209–226
mations I. Am J Math 81:119–159, and II, Am J Math Giordano T, Skandalis G (1985b) On infinite tensor prod-
85 (1963), 551–576 ucts of factors of type I2. Ergodic Theory Dyn Syst 5:
Effros EG (1965) Transformation groups and C*-algebras. 565–586
Ann Math 81(2):38–55 Glasner E (1994) On the multipliers of W ⊥ . Ergodic
Eigen SJ (1981) On the simplicity of the full group of Theory Dyn Syst 14:129–140
ergodic transformations. Israel J Math Glasner E, Weiss B (2016) Weak mixing properties for
40(3–4):345–349 non-singular actions. Ergodic Theory Dyn Syst 36:
Eigen SJ (1982) The group of measure preserving trans- 2203–2217
formations of [0,1] has no outer automorphisms. Math Glimm J (1961) Locally compact transformation groups.
Ann 259:259–270 Trans Am Math Soc 101:124–138
Eigen S, Hajian A, Halverson K (1998a) Multiple recur- Golodets VY (1969) A description of the representations of
rence and infinite measure preserving odometers. Israel anticommutation relations. Uspehi Matemat Nauk
J Math 108:37–44 24(4):43–64
Eigen S, Hajian A, Weiss B (1998b) Borel automorphisms Golodets VY, Sinel’shchikov SD (1983) Existence and
with no finite invariant measure. Proc Am Math Soc uniqueness of cocycles of ergodic automorphism with
126:3619–3623 dense range in amenable groups. Preprint FTINT AN
Elyze M, Kastner A, Ortiz Rhoton J, Semenov V, Silva CE USSR, pp 19–83
(2018) On conservative sequences and their application Golodets VY, Sinel’shchikov SD (1985) Locally compact
to ergodic multiplier problems. Colloq Math 151: groups appearing as ranges of cocycles of ergodic
123–145 Z-actions. Ergodic Theory Dyn Syst 5:47–57
Fedorov A (1985) Krieger’s theorem for cocycles, preprint Golodets VY, Sinel’shchikov SD (1990) Amenable ergodic
Feldman J, Moore CC (1977) Ergodic equivalence rela- actions of groups and images of cocycles. (Russian).
tions, cohomology, and von Neumann algebras. Dokl Akad Nauk SSSR 312(6):1296–1299
I. Trans Am Math Soc 234:289–324 Golodets VY, Sinel’shchikov SD (1994) Classification and
Ferenczi S (1985) Systèmes de rang un gauche. Ann Inst structure of cocycles of amenable ergodic equivalence
H Poincare’ Probab Statist 21:177–186 relations. J Funct Anal 121:455–485
Friedman NA (1970) Introduction to Ergodic Theory. Van Gouëzel S (2011) Correlation asymptotics from large devi-
Nostrand ations in dynamical systems with infinite measure.
Friedman NA (1978) Mixing transformations in an infinite Colloq Math 125:193–212
measure space, Studies in probability and ergodic the- Grosser S, Moskowitz M (1971) Compactness conditions
ory. Adv Math Suppl Stud 2:167–184 in topological groups. J Reine Angew Math 246:1–40
Furman A (2002) Random walks on groups and random Gruher K, Hines F, Patel D, Silva CE, Waelder R (2003)
transformations. In: Handbook of dynamical systems, Power weak mixing does not imply multiple recurrence
vol 1A. North-Holland, Amsterdam, pp 931–1014 in infinite measure and other counterexamples.
Furstenberg H (1967) Disjointness in ergodic theory, min- New York J Math 9:1–22
imal sets and diophantine approximation. Math Syst Haddock B, Leng J, Silva CE (2022) Nonsingular trans-
Theory 1:1–49 formations that are ergodic with isometric coefficients
Furstenberg H (1981) Recurrence in ergodic theory and and not weakly doubly ergodic. Indagationes
combinatorial number theory. Princeton University Mathematicae 33: 1297–1311
Press, Princeton Hajian AB, Kakutani S (1964) Weakly wandering sets and
Furstenberg H, Glasner E (2010) Stationary dynamical invariant measures. Trans Am Math Soc 110:136–151
systems. In: Dynamical numbers interplay between Halmos PR (1946) An ergodic theorem. Proc Natl Acad
dynamical systems and number theory, Contemporary Sci U S A 32:156–161
mathematics, vol 532. American Mathematical Society, Halmos PR (1956) Lectures on ergodic theory. Publication
pp 1–28 of the Mathematical Society of Japan 3, Tokyo.
Furstenberg H, Weiss B (1978) The finite multipliers of Reprinted Chelsea Publishing Co., New York, 1960
infinite ergodic transformations. In: The structure of Hamachi T (1981a) The normalizer group of an ergodic
attractors in dynamical systems, Lecture notes in math- automorphism of type III and the commutant of an
ematics 668. Springer, Berlin, pp 127–132 ergodic flow. J Funct Anal 40:387–403
Ergodic Theory: Nonsingular Transformations 289

Hamachi T (1981b) On a Bernoulli shift with nonidentical Host B, Méla J-F, Parreau F (1991) Nonsingular transfor-
factor measures. Ergodic Theory Dyn Syst 1:273–283 mations and spectral analysis of measures. Bull Soc
Hamachi T (1992) A measure theoretical proof of the Math France 119:33–90
Connes-Woods theorem on AT-flows. Pac J Math 154: Hurewicz W (1944) Ergodic theorem without invariant
67–85 measure. Ann Math 45:192–206
Hamachi T, Kosaki H (1993) Orbital factor map. Ergodic Inoue K (2004) Isometric extensions and multiple recur-
Theory Dyn Syst 13:515–532 rence of infinite measure preserving systems. Israel
Hamachi T, Osikawa M (1981) Ergodic groups of auto- J Math 140:245–252
morphisms and Krieger’s theorems. Seminar on Math- Ionescu Tulcea A (1965) On the category of certain classes
ematical Science, Keio University, 3 of transformations in ergodic theory. Trans Am Math
Hamachi T, Osikawa M (1986) Computation of the asso- Soc 114:261–279
ciated flows of ITPFI2 factors of type III0, Geometric James J, Koberda T, Lindsey K, Silva CE, Speh P (2008)
methods in operator algebras (Kyoto, 1983), Pitman Measurable sensitivity. Proc Am Math Soc 136:
research notes in mathematics series 123. Longman 3549–3559
Scientific and Technical, Harlow, pp 196–210 Janvresse E, de la Rue T (2012) Zero Krengel entropy does
Hamachi T, Silva CE (2000) On nonsingular Chacon trans- not kill Poisson entropy. Annales de l’I.H.P, Pro-
formations. Ill J Math 44:868–883 babilit’es et statistiques 48:368–376
Hawkins JM (1982) Non-ITPFI diffeomorphisms. Israel Janvresse E, Meyerovitch T, de la Rue T, Roy E (2010)
J Math 42:117–131 Poisson suspensions and entropy of infinite transfor-
Hawkins JM (1983) Smooth type III diffeomorphisms of mations. Trans Am Math Soc 362:3069–3094
manifolds. Trans Am Math Soc 276:625–643 Janvresse E, de la Rue T, Roy E (2017) Poisson suspen-
Hawkins JM (1990a) Diffeomorphisms of manifolds with sions and SuShis. Annales Scientifiques de l’ École
nonsingular Poincare flows. J Math Anal Appl 145(2): Normale Supérieure 50:1301–1334
419–430 Janvresse E, de la Rue T, Roy E (2018) Invariant measures
Hawkins JM (1990b) Properties of ergodic flows associ- for Cartesian powers of Chacon infinite transformation.
ated to product odometers. Pac J Math 141:287–294 Israel J Math 224:1–37
Hawkins J, Schmidt K (1982) On C2-diffeomorphisms of Janvresse E, de la Rue T, Roy E (2019) Nearly finite
the circle which are of type III1. Invent Math 66(3): Chacon transformation. Ann H Lebesgue 2:369–414
511–518 Jaworski W (1994) Strongly approximatively transitive
Hawkins J, Silva CE (1991) Noninvertible transformations actions, the Choquet-Deny theorem, and polynomial
admitting no absolutely continuous s-finite invariant growth. Pac J Math 165:115–129
measure. Proc Am Math Soc 111(2):455–463 Johnson ASA, Şahin AA (2015) Directional recurrence for
Hawkins J, Silva CE (1997/98) Characterizing mildly infinite measure preserving Zd-actions. Ergodic Theory
mixing actions by orbit equivalence of products. Dyn Syst 35:2138–2150
New York J Math, 3A. Proceedings of the New York Joita M, Munteanu RB (2014) A property of ergodic flows.
Journal of Mathematics Conference, June 9–13, Stud Math 225:249–258
(1997), 99–115 Kabanov JM, Lipcer RS, Sirjaev AN (1977) On the ques-
Hawkins J, Woods EJ (1984) Approximately transitive tion of absolute continuity and singularity of probabil-
diffeomorphisms of the circle. Proc Am Math Soc ity measures. Math USSR Sbornik 33:203–221
90(2):258–262 Kaimanovich VA, Vershik AM (1983) Random walks on
Herman M (1979a) Construction de difféomorphismes groups: boundary and entropy. Ann Probab 11:
ergodiques, preprint 457–490
Herman M-R (1979b) Sur la conjugaison differentiable des Kakutani S (1948) On equivalence of infinite product mea-
diffeomorphismes du cercle a des rotations. (French). sures. Ann Math 49:214–224
Institute Hautes Etudes Science Publication Mathemat- Kakutani S, Parry W (1963) Infinite measure preserving
ics, No. 49, 5–233 transformations with “mixing”. Bull Am Math Soc 69:
Herman RH, Putnam IF, Skau CF (1992) Ordered Bratteli 752–756
diagrams, dimension groups and topological dynamics. Katznelson Y (1977) Sigma-finite invariant measures for
Int J Math 3(6):827–864 smooth mappings of the circle. J Analyse Math 31:
Hochman M (2013) On the ratio ergodic theorem for group 1–18
actions. J Lond Math Soc 88:465–482 Katznelson Y (1979) The action of diffeomorphism of the
Hochman M (2019) Every Borel automorphism without circle on the Lebesgue measure. J Analyse Math 36:
finite invariant measures admits a two-set generator. 156–166
J Eur Math Soc (JEMS) 21:271–317 Katznelson Y, Weiss B (1972) The construction of quasi-
Hopf E (1937) Ergodentheorie, Ergebnisse der Mathematik invariant measures. Israel J Math 12:1–4
und ihrer Grenzgebiete, vol 3. Springer, Berlin, p 5 Katznelson Y, Weiss B (1991) The classification of non-
Host B, Méla J-F, Parreau J-F (1986) Analyse harmonique singular actions, revisited. Ergodic Theory Dyn Syst
des mesures. Astérisque, No. 135–136 11:333–348
290 Ergodic Theory: Nonsingular Transformations

Keane M, Smorodinsky M (1979) Bernoulli schemes of the (Proceedings of the conference on Ohio State Univer-
same entropy are finitarily isomorphic. Ann Math 109: sity, Columbus, 1970). Lecture notes in mathematics,
397–406 vol 160. Springer, Berlin, pp 158–177
King JL (1986) The commutant is the weak closure of the Krieger W (1972) On the infinite product construction of
powers, for rank-1 transformations. Ergodic Theory nonsingular transformations of a measure space. Invent
Dyn Syst 6:363–384 Math 15:144–163; and erratum in 26 (1974), 323–328
Kirillov AA (1978) Elements of the theory of representa- Krieger W (1976a) On Borel automorphisms and their
tions. Nauka, Moscow quasi-invariant measures. Math Z 151:19–24
Kosloff Z (2011) On a type III1 Bernoulli shift. Ergodic Krieger W (1976b) On ergodic flows and isomorphism of
Theory Dyn Syst 31:1727–1743 factors. Math Ann 223:19–70
Kosloff Z (2013) The zero-type property and mixing of Kubo I (1969) Quasi-flows. Nagoya Math J 35:1–30
Bernoulli shifts. Ergodic Theory Dyn Syst 33:549–559 Landesberg O, Lindenstrauss E (2022) On Radon measures
Kosloff Z (2014) On the K property for Maharam exten- invariant under horospherical flows on geometrically
sions of Bernoulli shifts and a question of Krengel. infinite quotients. Int Math Res Not IMRN:11602–11641
Israel J Math 199:485–506 Ledrappier F, Sarig O (2007) Invariant measures for the
Kosloff Z (2017) A universal divergence rate for symmet- horocycle flow on periodic hyperbolic surfaces. Israel
ric Birkhoff sums in infinite ergodic theory. Trans Am J Math 160:281–315
Math Soc 369:6373–6388 Lehrer E, Weiss B (1982) An ϵ-free Rokhlin lemma. Ergo-
Kosloff Z (2018) On manifolds admitting stable type III1 dic Theory Dyn Syst 2:45–48
Anosov diffeomorphisms. J Modern Dyn 13:251–270 Lemańczyk M, Parreau F (2003) Rokhlin extensions and
Kosloff Z (2019) Proving ergodicity via divergence of lifting disjointness. Ergodic Theory Dyn Syst 23:
ergodic sums. Stud Math 248:191–215 1525–1550
Kosloff Z (2021) Conservative Anosov diffeomorphisms of Lemańczyk M, Parreau F, Thouvenot J-P (2000) Gaussian
the two torus without an absolutely continuous invariant automorphisms whose ergodic self-joinings are Gauss-
measure. Ann Sci Ec Norm Supér 54(4):69–131 ian. Fundam Math 164:253–293
Kosloff Z, Soo T (2021) The orbital equivalence of Lemańczyk M, Lesigne E, Skrenty D (2001) Multiplicative
Bernoulli actions and their Sinai factors. J Mod Dyn Gaussian cocycles. Aequationes Math 61:162–178
17:145–182 Lenci M (2010) On infinite-volume mixing. Commun
Kosloff Z, Soo T (2022) Some factors of nonsingular Math Phys 298:485–514
Bernoulli shifts. Stud Math 262(1):23–43 Lenci M (2013) Exactness, K-property and infinite mixing.
Krengel U (1967) Entropy of conservative transforma- Publ Mat Urug 14:159–170
tions. Z Wahrscheinlichkeitstheorie und Verw Gebiete Lenci M (2017) Uniformly expanding Markov maps of the
7:161–181 real line: exactness and infinite mixing. Discrete Contin
Krengel U (1969) Darstellungssätze für Strömungen und Dyn Syst 37:3867–3903
Halbströmungen. II. Math Ann 182:1–39 Loh I, Silva CE (2017) Strict doubly ergodic infinite trans-
Krengel U (1970) Transformations without finite invariant formations. Dyn Syst 32(4):519–543
measure have finite strong generators. In: Contributions Loh I, Silva CE, Athiwaratkun B (2018) Infinite symmetric
to Ergodic theory and probability (Proceedings of the ergodic index and related examples in infinite measure.
Conference on Ohio State University, Columbus, Stud Math 243:101–115
1970). Springer, Berlin, pp 133–157 Mackey GW (1966) Ergodic theory and virtual group.
Krengel U (1976) On Rudolph’s representation of aperiodic Math Ann 166:187–207
flows. Ann Inst H Poincaré Sect B (NS) 12(4):319–338 Maharam D (1964) Incompressible transformations. Fund
Krengel U (1985) Ergodic theorems. de Gruyter Studies in Math LVI:35–50
Mathematics, Berlin Mandrekar V, Nadkarni M (1969) On ergodic quasi-
Krengel U, Sucheston L (1969) On mixing in infinite invariant measures on the circle group. J Funct Anal
measure spaces. Z. Wahrscheinlichkeitstheorie und 3:157–163
Verw. Gebiete 13:150–164 Mansfield DF, Dooley AN (2017) The critical dimension
Krickeberg K (1967) Strong mixing properties of Markov for G-measures. Ergodic Theory Dyn Syst 37:824–836
chains with infinite invariant measure. In: Proceedings Marrakchi A, Vaes S (2022) Nonsingular Gaussian actions:
of the Fifth Berkeley symposium on mathematical sta- beyond the mixing case. Adv Math 397:108190, 62 p
tistics and probability (Berkeley, 1965/66), Vol. II: Matui H (2002) Topological orbit equivalence of locally
Contributions to probability theory, Part 2. University compact Cantor minimal systems. Ergodic Theory Dyn
of California Press, Berkeley, pp 431–446 Syst 22:1871–1903
Krieger W (1969) On nonsingular transformations of a Méla J-F (1983) Groupes de valeurs propres des systèmes
measure space, I, II. Z Wahrscheinlichkeitstheorie und dynamiques et sous-groupes saturés du cercle. C R
Verw Gebiete 11:83–119 Acad Sci Paris Se’r I Math 296(10):419–422
Krieger W (1970) On the Araki-Woods asymptotic ratio set Melbourne I (2015) Mixing for invertible dynamical sys-
and nonsingular transformations of a measure space. In: tems with infinite measure. Stoch Dyn 15(2):
Contributions to Ergodic theory and probability 1550012, 25
Ergodic Theory: Nonsingular Transformations 291

Melbourne I, Terhesiu D (2012) Operator renewal theory Parry W (1966) Generators and strong generators in ergo-
and mixing rates for dynamical systems with infinite dic theory. Bull Am Math Soc 72:294–296
measure. Invent Math 189:61–110 Parry W (1969) Entropy and generators in ergodic theory.
Meyerovitch T (2007) On multiple and polynomial recur- W. A. Benjamin, Inc., New York/Amsterdam
rent extensions of infinite measure preserving transfor- Parthasarathy KR, Schmidt K (1977) On the cohomology
mations, preprint. ArXiv: http://arxiv.org/abs/math/ of a hyperfinite action. Monatsh Math 84(1):37–48
0703914 Ramsay A (1971) Virtual groups and group actions. Adv
Meyerovitch T (2013) Ergodicity of Poisson products and Math 6:243–322
applications. Ann Probab 41:3181–3200 Rokhlin VA (1965) Generators in ergodic theory. II.
Milnor J (1988) On the entropy geometry of cellular (Russian. English summary). Vestnik Leningrad Univ
automata. Complex Syst 2:357–385 20(13):68–72
Moore CC (1967) Invariant measures on product spaces. Rosinsky J (1995) On the structure of stationary stable
In: Proceedings of the Fifth Berkeley symposium, processes. Ann Probab 23:1163–1187
pp 447–459 Roy E (2005) Mesures de Poisson, infinie divisibilité et
Moore CC (1982) Ergodic theory and von Neumann alge- propriétés ergodiques, Thèse de doctorat de
bras. In: Proceedings of the symposium on pure math- l’Université Paris 6
ematics, 38. American Mathematical Society, Roy E (2007) Ergodic properties of Poissonian ID pro-
Providence, pp 179–226 cesses. Ann Probab 35:551–576
Moore CC, Schmidt K (1980) Coboundaries and homo- Roy E (2009) Poisson suspensions and infinite ergodic
morphisms for nonsingular actions and a problem of theory. Ergodic Theory Dyn Syst 29:667–683
H. Helson. Proc Lond Math Soc 40(3):443–475 Roy E (2010) Poisson-Pinsker factor and infinite measure
Mortiss G (2000) A non-singular inverse Vitali lemma with preserving group actions. Proc Am Math Soc 138:
applications. Ergodic Theory Dyn Syst 20:1215–1229 2087–2094
Mortiss G (2002) Average co-ordinate entropy. J Aust Rudolph DJ (1985) Restricted orbit equivalence. Mem Am
Math Soc 73:171–186 Math Soc 323
Mortiss G (2003) An invariant for nonsingular isomor- Rudolph D, Silva CE (1989) Minimal self-joinings for
phism. Ergodic Theory Dyn Syst 23:885–893 nonsingular transformations. Ergodic Theory Dyn
Munteanu RB (2012) A non-product type non-singular Syst 9:759–800
transformation which satisfies Krieger’s Property Ryzhikov VV (1993) Factorization of an automorphism of
A. Israel J Math 190:307–324 a full Boolean algebra into the product of three involu-
Nadkarni MG (1979) On spectra of nonsingular transfor- tions. (Russian). Mat Zametki 54(2):79–84, 159; trans-
mations and flows. Sankhya Ser A 41(1–2):59–66 lation in Math. Notes 54 (1993), no. 1–2,
Nadkarni MG (1998) Spectral theory of dynamical sys- 821–824 (1994)
tems, Birkhäuser Advanced Texts: Basler Lehrbücher. Ryzhikov VV (2014) On the asymmetry of multiple
Birkhäuser Verlag, Basel asymptotic properties of ergodic actions. Math Notes
Neretin YA (1996) Categories of symmetries and infinite- 96:416–422
dimensional groups. Oxford University Press, New York Ryzhikov VV, Thouvenot J-P (2015) On the centralizer of
Ornstein D (1960) On invariant measures. Bull Am Math an infinite mixing rank-one transformation. Funct Anal
Soc 66:297–300 Appl 49:230–233
Ornstein D (1972) On the root problem in ergodic theory. Sachdeva U (1971) On category of mixing in infinite
In: Proceedings of the Sixth Berkeley symposium on measure spaces. Math Syst Theory 5:319–330
mathematical statistics and probability. University of Samorodnitsky G (2005) Null flows, positive flows and the
California Press, pp 347–356 structure of stationary symmetric stable processes. Ann
Osikawa M (1977/78) Point spectra of nonsingular flows. Probab 33:1782–1803
Publ Res Inst Math Sci 13:167–172 Sarig O (2004) Invariant measures for the horocycle flows
Osikawa M (1988) Ergodic properties of product type on Abelian covers. Invent Math 157:519–551
odometers. Springer Lect Notes Math 1299:404–414 Schmidt K (1977a) Cocycles on ergodic transformation
Osikawa M, Hamachi T (1971) On zero type and positive groups. Macmillan lectures in mathematics,
type transformations with infinite invariant measures. vol 1. Macmillan Company of India, Ltd., Delhi
Mem Facul Sci, Kyushu Univ 25:280–295 Schmidt K (1977b) Infinite invariant measures in the circle.
Parreau F, Roy E (2008) Poisson joinings of Poisson sus- Symp Math 21:37–43
pensions, preprint Schmidt K (1982) Spectra of ergodic group actions. Israel
Parry W (1963) An ergodic theorem of information theory J Math 41(1–2):151–153
without invariant measure. Proc Lond Math Soc 13(3): Schmidt K (1984) On recurrence. Z Wahrscheinlich-
605–612 keitstheorie verw Gebiete 68:75–95
Parry W (1965) Ergodic and spectral analysis of certain Schmidt K, Walters P (1982) Mildly mixing actions of
infinite measure preserving transformations. Proc Am locally compact groups. Proc Lond Math Soc 45:
Math Soc 16:960–966 506–518
292 Ergodic Theory: Nonsingular Transformations

Shelah S, Weiss B (1982) Measurable recurrence and Series, 205. Cambridge University Press, Cambridge,
quasi-invariant measures. Israel J Math 43:154–160 pp 207–235
Silva CE (1988) On m-recurrent nonsingular endomor- Ullman D (1987) A generalization of a theorem of Atkinson
phisms. Israel J Math 61:1–13 to non-invariant measures. Pac J Math 130:187–193
Silva CE, Thieullen P (1991) The subadditive ergodic Vaes S, Wahl J (2018) Bernoulli actions of type III1 and L2-
theorem and recurrence properties of Markovian trans- cohomology. Geom Funct Anal 28:518–562
formations. J Math Anal Appl 154(1):83–99 Vershik AM (1983) Many valued mappings with invariant
Silva CE, Thieullen P (1995) A skew product entropy for measure (polymorphisms) and Markov processes.
nonsingular transformations. J Lond Math Soc 52(2): J Sov Math 23:2243–2266
497–516 Vershik AM, Kerov SV (1985) Locally semisimple alge-
Silva CE, Witte D (1992) On quotients of nonsingular actions bras. Combinatorial theory and K0-functor. Modern
whose self-joinings are graphs. Int J Math 5:219–237 Prob Math 26:3–56
Thaler M (1998) The Dynkin-Lamperti arc-sine laws for Yuasa H (2013) Uniform sets for infinite measure-
measure preserving transformations. Trans Am Math preserving systems. J d’Anal Math 120:333–356
Soc 350(11):4593–4607 Yuasa H (2020) A relative, strictly ergodic model theorem
Thaler M, Zweimüller R (2006) Distributional limit theo- for infinite measure-preserving systems. J d’Anal Math
rems in infinite ergodic theory. Probab Theory Relat 140:591–616
Fields 135(1):15–52 Zimmer RJ (1977) Random walks on compact groups and
Thouvenot J-P (1995) Some properties and applications of the existence of cocycles. Israel J Math 26:84–90
joinings in ergodic theory. In: Ergodic theory and its Zimmer RJ (1978) Amenable ergodic group actions and an
connections with harmonic analysis (Alexandria, application to Poisson boundaries of random walks.
1993), London Mathematical Society, Lecture Note J Funct Anal 27:350–372
 (see Appendix C in (Matomäki et al. 2015)). In
Sarnak’s Conjecture from the particular, m and l are strongly aperiodic.
Ergodic Theory Point of View Arithmetic function A sequence of complex
numbers is usually denoted by u ¼ (un). But
Joanna Kułaga-Przymus and Mariusz Lemańczyk if such a sequence is, in some sense, important
Faculty of Mathematics and Computer Science, from number theory point of view, one speaks
Nicolaus Copernicus University, Toruń, Poland about an arithmetic function and rather writes
u : ℕ ! ℂ, u ¼ (u(n)).
An arithmetic function u is said to be multipli-
Article Outline cative, whenever u(1) ¼ 1 and u(m  n) ¼ u-
(m)  u(n) for any choice of coprime m, n  ℕ.
Glossary The prominent examples of multiplicative
Definition of the Subject functions are the Möbius function m and the
Introduction Liouville function l. The Möbius function
Chowla Conjecture m : ℕ ! {1, 0, 1} is defined by
Sarnak’s Conjecture m(p1. . .pk) ¼ (1)k for different prime num-
Arithmetic Properties of the Möbius Function bers p1, . . . , pk (in what follows, the set of
Future Directions primes is denoted by ℙ), m(1) ¼ 1 and m(n) ¼ 0
Bibliography for all non-square-free numbers. The Liouville
function l : ℕ ! {1, 1} is given by
Glossary l ðnÞ ¼ ð1Þi1 þ...þik for n ¼ pi11 . . . pikk with
p1, . . . , pk  ℙ and i1, . . . , ik  ℕ. Clearly
Aperiodicity We say that u : ℕ ! ℂ is aperiodic m ¼ l  m2, where m2 is nothing but the
whenever u has a mean, equal to zero, characteristic function of the set S of square-
along each arithmetic progression: free numbers. In fact, l is completely multipli-
lim N!1 N1 nN uðan þ bÞ ¼ 0. Many classi- cative, that is, l(m  n) ¼ l(m)  l(n) for any
cal multiplicative functions are aperiodic, choice of m, n  ℕ. We extend both, m and l,
including m and l. to negative coordinates symmetrically.
For a parameter N  1, a distance between Completely deterministic point We say that
u, v : ℕ !  is defined as point x  X is completely deterministic
1=2
1 Re ðuðpÞvðpÞÞ
(Weiss 1971) (see also (Kamae 1973)) if for
ðu, v; N Þ≔ p : any n  V (x), we have hðT, X, B ðXÞ, nÞ ¼ 0.
p  ℙ, pN
By the variational principle, htop(T, X) ¼ 0 if
and only if all points of X are completely
We say that u : ℕ !  is strongly aperiodic deterministic.
(Matomäki et al. 2015), whenever Entropy There are two basic notions of entropy:
2
Mðu  w; N Þ≔ min jtjN ðu  w, n ; N Þ ! 1 as
it
topological and measure-theoretic. We skip the
N ! 1 for every Dirichlet character w (i.e., for definitions and refer the reader, for example, to
every periodic, completely multiplicative func- (Downarowicz 2011). For a topological
tion). Strong aperiodicity implies aperiodicity. dynamical system (T, X) its topological entropy
The converse is not in general true (see Theorem will be denoted by htop(T, X) and for a measure-
B.1 in (Matomäki et al. 2015)), but it is true for theoretic dynamical system ðT, X, B, nÞ the
(bounded) real valued multiplicative functions corresponding measure-theoretic entropy will

© Springer Science+Business Media, LLC, part of Springer Nature 2023 293

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_735
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_735-1
294 Sarnak’s Conjecture from the Ergodic Theory Point of View

be denoted by hðT, X, B, nÞ. The basic connec- T-invariant measures is denoted by M (T, X).
tion between them is the variational principle: The subset of ergodic measures (which is
htop ðT, XÞ ¼ sup hðT, X, B ðXÞ, nÞ ¼ always non-empty) is denoted by Me(T, X).
n  MðT, XÞ
Each n  M(T, X) gives rise to a measure-
sup hðT, X, B ðXÞ, nÞ: theoretic dynamical system ðT, X, B ðXÞ, nÞ,
n  Me ðT, XÞ
where B ðXÞ stands for the s-algebra of Borel
Furstenberg system Let u  ℤ : For each
subsets of X. With the weak--topology, M (T, X)
n  V(u), the system ðS, Xu , B ðXu Þ, nÞ is called becomes a compact metrizable space. (T, X) is
a Furstenberg system of u. For each n  Vlog(u), called uniquely ergodic if |M (T, X)| ¼ 1.
the system ðS, Xu , B ðXu Þ, nÞ is called a loga-
Joinings of measure-theoretic dynamical
rithmic Furstenberg system of u. systems Assume that ðRi , Zi , C i , ki Þ is a
For each jujℤ , one can consider juj[0,1]ℤ. measure-theoretic dynamical system, i ¼ 1,
Then (S, X|u|) is a topological factor of (S, Xu) (the
2. Each R1  R2-invariant measure r on
map p : Xu ! Xjuj given by p(x) ¼ jxj, under- C 1 C 2 projecting on k1 and k2, respectively,
stood coordinatewise, is equivariant with S). For
is called a joining of the automorphisms R1 and
n  V (u), we have p (n)  V (|u|), where p (n)
R2. The set of joinings between R1 and R2 is
stands for the image of n via p. Moreover, denoted by J(R1, R2) and each r  J(R1, R2)
the Furstenberg system ðS,Xu ,B ðXu Þ, nÞ is an
yields a (new) measure-theoretic dynamical sys-
extension of S, X juj ,B Xjuj , p ðnÞ . In partic-
tem ðR1  R2 , Z 1  Z2 , C 1 C 2 , rÞ: When R1
ular, if V (|u|) is a singleton (|u| is a generic point), and R2 are both ergodic, then the set Je(R1, R2) of
then all Furstenberg systems of u have
ergodic joinings between R1 and R2 is non-empty.
S, X juj ,B Xjuj , n (where n is the unique If J(R1, R2) ¼ {k1 k2}, then R1 and R2 are
member of V (|u|)) as their factor. called disjoint (in the sense of Furstenberg).
Generic point Assume that (T, X) is a topologi- Let (T, X) be a topological dynamical system. Fix
cal dynamical system. We say that x  X is a x  X and let u be an arithmetic function
generic point for a Borel measure n on X, bounded by 1, that is, u : ℕ ! : Each accumu-
whenever the ergodic theorem holds for T at lation point r of N1 nN dðT x,S uÞ is a (T  S)-
n n

x for any continuous function f  C (X), that ℤ

invariant measure on X   . Obviously, it is a
is, N1 nN f ðT n xÞ ! f dn. In other words, the joining of its projections onto both coordinates.
empirical measures N1 nN dT n x converge to n Hence, r is a joining of ðT, X, B ðXÞ, rjX Þ with a
in the weak--topology. If this convergence Furstenberg system of u.
holds along a subsequence, we say that x is Measure-theoretic dynamical system Let
quasi-generic for n. We denote by V (x) the set ðZ, C , kÞ be a standard probability Borel
of all n  M(T, X) for which x is quasi-generic space. Assume R: X ! X is invertible (a.e.),
ð 0 6¼ V ðxÞ  M ðT , X Þ by compactness). bi-measurable, and k-preserving. Then R is


We say that x  X is logarithmically ge- called an automorphism or (invertible)

neric for n, whenever L1N nN 1n dT n x , with measure-preserving transformation and the
LN ¼ nN1/n, converges to n. No harm quadruple ðR, Z, C , kÞ is a measure-theoretic
arises if in what follows we replace LN by dynamical system.
log N. Finally, we say that x  X is logarithmi- Möbius orthogonality We say that a
cally quasi-generic for n whenever this conver- (topological) dynamical system (T, X) is Möbius
gence holds along a subsequence and we denote orthogonal if
the set of all measures for which x is logarith-
mically quasi-generic by V log(x) 1
lim f ðT n xÞmðnÞ ¼ 0 ð3Þ
0 6¼ V log ðxÞ  M ðT , X Þ . N!1 N


nN
Invariant measure Given a topological dynam-
ical system (T, X), the set of probability Borel for each f  C (X) and x  X.
Sarnak’s Conjecture from the Ergodic Theory Point of View 295

Nilrotation Let G be a connected, simply and the corresponding dynamical system is

connected nilpotent Lie group and G  G a (S, X). For each u  ℤ , we define a subshift
lattice (a discrete, cocompact subgroup). For Xu  ℤ as the orbit closure of u under S.
any g0  G we define Tg0 (gG) ≔ g0gG. Then Similarly, when u  ℕ (ℕ ¼ {1, 2, . . .}), we
the topological system (Tg0 , G/G) is called a can extend u symmetrically, by setting
nilrotation. u(n) :¼ u(n) (and u(0) ¼ 0) for each n  1
Orthogonality of sequences Suppose that one and define again the corresponding subshift Xu.
of sequences (un), (vn)  ℂ is of zero mean. Finally, we will denote by F the continuous
We say that (un), (vn) are orthogonal, function on X  ℤ , defined by F(x) ¼ x0.
whenever lim N!1 N1 nN un vn ¼ 0. We say Topological dynamical system Let X be a com-
that sequences (un), (vn)  ℂ are orthogonal on pact metric space and let T be a homeomor-
short intervals, whenever lim K!1 b1K kK phism of X. Then (T, X) is called a topological
dynamical system.
bk n<bkþ1 un vn ¼ 0 for any sequence (bk)
Uniquely ergodic model of a measure-theoretic
of natural numbers with bk+1  bk ! 1. dynamical system Given a measure-theoretic
Clearly, orthogonality on short intervals (ergodic) dynamical system ðR, Z, C , kÞ and a
implies orthogonality. We can consider un uniquely ergodic topological system (T, X)
being an element of a Banach space, while in with M(T, X) ¼ {n} (n is necessarily ergodic),
the dynamical context one often takes un ¼ one says that (T, X) is a uniquely ergodic model
f(Tnxk), whenever n  [bk, bk+1) (with xk  X of the automorphism R if the measure-theoretic
and f  C (X)). Moreover, often (vn) will be in systems ðR, Z, C , kÞ and ðT, X, B ðXÞ, nÞ are
fact a multiplicative function (and then we measure-theoretically isomorphic. Each ergo-
rather write (v(n))). dic automorphism R has a uniquely ergodic
Quasi-genericity versus logarithmic quasi- model.
genericity Obviously, if |V (x)| ¼ 1, then
V (x) ¼ V log(x), but in general the sets V (x)
and V log(x) can be even disjoint. However, as
shown in (Gomilko et al. 2018), Definition of the Subject

V log ðxÞ \ Me ðT, XÞ  V ðxÞ: ð1Þ Sarnak in 2010 formulated a now celebrated con-
jecture on the Möbius orthogonality. It states that
Moreover, using an idea from Tao (n.d.), it has each topological dynamical system (T, X) of zero
been proved in Gomilko et al. (2020) that entropy is Möbius orthogonal, that is, whenever
htop(T, X) ¼ 0,
if V log ðxÞ ¼ fvg and v is ergodic, then
1
1 lim f ðT n xÞmðnÞ ¼ 0
lim k!1 d n ¼ n, for a subset
nN k T x
N!1 N ð4Þ
Nk ð2Þ nN

fN k : k  1g whose for each f  C ðXÞ and x  X:

logarithmic density is 1:
It has become a bridge between analytic number
theory and dynamics. In particular, it keeps stimu-
Subshift For any closed subset   lating a quick development of disjointness theory
≔fz  ℂ :jzj 1g (most often  will be in dynamics originated by Furstenberg in 1967,
finite), let S : ℤ ! ℤ be the left shift, that with potential applications, for example, in number
is, for each x ¼ ðxn Þn  ℤ  ℤ , Sx ¼ y, where theory. In this entry, we will focus on ergodic-
yn ¼ xn+1 for n  ℤ. Each closed and theoretic aspects of recent progress in the area.
S-invariant subset X  ℤ is called a subshift Unless we need this for a historical reason, we
rather avoid describing numerous particular classes
296 Sarnak’s Conjecture from the Ergodic Theory Point of View

of topological systems in which Möbius the point m  f1,0, 1gℤ is generic for the
disjointness has been shown. We concentrate on _
relatively independent extension nS of nS ,
main ideas, and describe some general results. ð5Þ
via the natural map
f1,0, 1gℤ 3 ðxn Þn1 7! x2n n1
 f0, 1gℤ :

_
Introduction The measure nS is given by the following
condition: for each block C over the alphabet
{1, 0, 1}, we have nS ðCÞ ¼ 2k nS C2 , where
_
The entry is organized as follows. In sections
“Chowla Conjecture” and “Sarnak’s Conjecture,” 2
C is obtained from C by taking the square (or,
respectively, we discuss Chowla and Sarnak’s equivalently, absolute value) at each coordinate
conjectures from the ergodic-theoretic viewpoint, and k is the cardinality of the support of C. To see
including dynamical interpretations of purely this, we use the following:
number-theoretic statements and the main strate-
gies used to attack Sarnak’s conjecture. Remark 4.1 The span of the
Section “Arithmetic Properties of the Möbius family of continuous functions
Function” is a survey of results on Sarnak’s con- F j0 ∘Sk0 F j1 ∘Sk1  . . .  F j‘ ∘Sk‘ : ji  0, ki  ℤg
jecture, arranged with respect to the properties of forms an algebra that distinguishes points. It fol-
the Möbius function that come into play in the lows directly from the Stone-Weierstrass theorem
proof. Finally, in section “Future Directions,” we that the values of integrals Fk1 ∘Sr1  Fk2 ∘Sr2 
state some open problems. . . .  Fk‘ ∘ Sr‘ dk determine a measure k.

Now, it suffices to look at measures k  V (m)

Chowla Conjecture and compare the value of the integrals
Fk1 ∘ Sr1  Fk2 ∘Sr2  . . .  Fk‘ ∘ Sr‘ dk with the
(C) corresponding values of Fk1 ∘Sr1  Fk2 ∘Sr2  . . . 
The Chowla conjecture deals with higher order Fk‘ ∘ Sr‘ dnS . We will also use the fact that
correlations of the Möbius function and asserts that
nS  Me S, Xm : ð6Þ
1
lim m j0 ðn þ k 0 Þ m j1 ðn þ k 1 Þ . . .
N!1 N
nN ðCÞ The simplest instance of Chowla conjecture,
that is, lim N!1 N1 nN m ðn Þ ¼ 0, is known to
m jr ðn þ kr Þ ¼ 0
hold and it is equivalent to the Prime Number
whenever 0  k0 < . . . < kr, js  {1, 2} not all Theorem, as shown by Landau. Moreover, if
equal to 2, r  0 (in Chowla (1965), the Chowla exactly one of the exponents ji is odd, then (C)
conjecture is formulated for the Liouville func- also holds (in Ram Murty and Vatwani (2018) a
tion. We follow Sarnak (n.d.). For a discussion on quantitative version of this fact has been proved).
an equivalence of the Chowla conjecture with m We also have the following conditional result of
and l, see (Ramaré 2018).). The above statement Frantzikinakis on (C):
can be translated into dynamical language. To see
this, consider first jmj ¼ m2, that is, the charac- Theorem 4.2 (Frantzikinakis 2017) Assume that
teristic function of the set S of square-free num- V(l) ¼ {k} and k is ergodic. Then (C) holds for l
bers. As a member of {0, 1}ℤ, it is well-known to (i.e., k equals the 12 -Bernoulli measure on
be a generic point for an ergodic measure, namely {1, 1}ℤ. Cf. Theorem 4.4).
for the so-called Mirsky measure nS (also denoted
as nm2 ), given by the ordinary average frequencies (C) vs. (Clog)
of blocks, see, for example, El Abdalaoui et al. Analogously, one can study the logarithmic
(2017a). (C) is equivalent to the following: Chowla conjecture (Tao 2017), asserting that
Sarnak’s Conjecture from the Ergodic Theory Point of View 297

1 1 j0 This, in turn, has the following interpretation in

lim m ðn þ k0 Þm j1 ðn þ k1 Þ . . .
N!1 log N nN n terms of spectral measures:

Clog Corollary 4.7 (Ferenczi et al. 2018)

m jr ðn þ kr Þ ¼ 0, For each logarithmic Furstenberg system
S, Xm , B Xm , k of m, the spectral measure sF
with all parameters as before. Again, this can be of F is Lebesgue. The same holds for k  Vlog(l).
translated to the dynamical language, using the
notion of a logarithmically generic point. Thus, Moreover, we have the following:
using (6), (1), and (2), we have the following:
Theorem 4.8 (Tao and Teräväinen 2018, 2019b)
Corollary 4.3 (cf. Gomilko et al. (2018); Tao (Clog) holds for odd order correlations, that is,
(n.d.)). (Clog) implies (C) along a subsequence for all even values of r.
of full logarithmic density.
Again, we want to interpret this from the
More on (Clog) dynamical viewpoint:
We begin this section with a conditional result of
Frantzikinakis on (Clog) (cf. Theorem 4.2): Corollary 4.9 For each logarithmic Furstenberg
system S, Xm , B Xm , k , the element Ry :¼ –y
Theorem 4.4 (Frantzikinakis 2017) Assume that preserves measure k and commutes with the shift
k  Vlog(l) is an ergodic measure. Then (Clog) S, hence R belongs to the centralizer of S.
holds for l along the same subsequence for which
l is quasi-generic for k.
To see that the above statement is true, it suf-
The above result is already formulated in the fices to take k  Vlog(m) and check that
language of (logarithmically) quasi-generic we also have k  Vlog(m). For this, one uses
points. We pass now to number-theoretic results, the family of continuous functions
with their ergodic-theoretic consequences. F j0 ∘ Sk0  F j1 ∘Sk1  . . .  F j‘ ∘Sk‘ : ji  0, ki  ℤ
. Then, for even correlations the desired equalities
Theorem 4.5 (Tao 2016) are obvious and for odd correlations one applies
(Clog) holds for r ¼ 1 (for l in place of m), that Theorem 4.8.
is, for each 0 6¼ h  ℤ, we have Finally, as a consequence of Corollary 4.3, we
have the following:
1 lðnÞlðn þ hÞ
lim ¼ 0:
N!1 log N nN
n Corollary 4.10 (Gomilko et al. 2018).
Suppose that (Clog) holds. Then nm2  V ðmÞ.
In fact, Tao in (Tao 2016) proves a stronger
result, being an instance of logarithmically aver-
Averaged (C)
aged Elliott conjecture. One of the consequences
Also, an averaged version of Chowla conjecture is
is that the analogue of Theorem 4.5 holds for m:
present in the literature. As we will see later, this
Corollary 4.6 (Tao 2016) form of averaging will play a special role from the
(Clog) holds for r ¼ 1, that is, for each 0 6¼ point of view of Sarnak’s program, under the
h  ℤ, we have name of convergence on short intervals.
Matomäki, Radziwiłł, and Tao showed the follow-
1 mðnÞmðn þ hÞ ing result on the order two correlations of m:
lim ¼ 0:
N!1 log N nN
n
298 Sarnak’s Conjecture from the Ergodic Theory Point of View

Theorem 4.11 (Matomäki et al. 2015) We have Sarnak’s Conjecture

lim 1
HM mðmÞmðm þ hÞ ¼ 0: (S)
M, H!1 hH mM
with H¼oðMÞ Sarnak’s Conjecture from 2010 states that

each ðtopologicalÞ zero entropy system

As a consequence of the above, we have ðT, XÞ is Möbius orthogonal, that is, ðSÞ
(cf. Corollary 4.7):
satisfies ð3Þfor all f 2 CðXÞ and x 2 X:
Corollary 4.12 (Ferenczi et al. 2018)
As zero entropy expresses the fact that the
For each Furstenberg system S, Xm , B Xm , k ,
system is “deterministic” (or of “low complex-
the spectral measure sF of F is continuous. ity”), Sarnak’s conjecture captures our expecta-
tion that prime numbers behave globally as a
(C) vs. Other Conjectures random sequence, or, more precisely, that they
Chowla conjecture is thought of as a multiplica- cannot be predicted by a low-complexity object.
tive analogue of the twin primes problem. Indeed, One can relax the entropy assumptions on T in
Twin Primes Conjecture in its quantitative form Sarnak’s conjecture in the following way:
expects that nN LðnÞLðn þ 2Þ ¼
1 Theorem 5.1 (El Abdalaoui et al. 2017a) (S) is
2 ℙ3p3 1  ðp1Þ2
N þ oðN Þ, where L is
equivalent to Möbius orthogonality for each topo-
the von Mangoldt function: L(pk) ¼ log p for logical dynamical system (T, X), each continuous
each prime p and k  1, and L vanishes at all other function f  C(X) and each completely determin-
values of n. This function is a good approximation istic point x  X (i.e. (3) holds at x for all
of 1ℙ and it is not a multiplicative function. f  C(X)).
Moreover, Chowla conjecture is a special
instance of Elliott conjecture on correlations of
multiplicative functions. Let u0, . . . , uk : ℕ !  Remark 5.2 The difficulty in Sarnak’s conjec-
be multiplicative. Elliott conjecture, in a corrected ture comes from the requirement “for all x  X.”
form given in (Matomäki et al. 2015), asserts that An a.e. version of Möbius orthogonality is true for
all dynamical systems. The proof makes use of
1 Davenport’s estimate (7) (below), see El
lim u0 ðn þ k0 Þu1 ðn þ k1 Þ . . . uk ðn þ kr Þ Abdalaoui et al. (2017a); Sarnak (n.d.).
N!1 N nN

¼ 0,
Even though Sarnak’s Conjecture is defined in
terms of topological dynamics, it can be translated
if for some 0  j  k, uj is strongly aperiodic. This
to ergodic-theoretic language. Namely
conjecture was stated first in Elliott (1992) and
Elliott (1994), and in its original version turned
1 1
out to be false, see (Matomäki et al. 2015) for f ðT n xÞmðnÞ ¼ f Fd d n n :
N nN XXm N nN ðT x,S mÞ
details. Also, a logarithmically averaged version
of Elliott conjecture appears in the literature. For
the details, see Tao and Teräväinen (2019a, b) and Thus, we need to study the properties of join-
the references therein. ings given by the limit points of
Sarnak’s Conjecture from the Ergodic Theory Point of View 299

1 (S) vs. (C)

d n n :
nN ðT x,S mÞ
N Sarnak’s conjecture was originally mainly moti-
vated by Chowla conjecture; we have the follow-
The simplest case where (S) is known to ing result:
hold is the one-point dynamical system:
lim N!1 N1 nN mðnÞ ¼ 0 is equivalent to the Theorem 5.4 (C) implies (S).
Prime Number Theorem. (S) for rotations on finite
groups is equivalent to Dirichlet’s Prime Number Theorem 5.4 is already stated in (Sarnak n.d.).
Theorem. For irrational rotations, (S) follows In fact, it is a purely ergodic theory claim: we have
from an old (quantitative) result of Davenport already noticed that both conjectures have their
(Davenport 1937): for an arbitrary A > 0, ergodic theory reformulation and a joining proof
of Theorem 5.4 can be found in El Abdalaoui et al.
N (2017a). The main idea is the following: suppose
max e2pint mðnÞ  CA that N1k nNk dðT n x,Sn mÞ ! r: The projection of
t
nN log A N ð7Þ
for some CA > 0 and all N  2: this joining onto X is a zero entropy measure k,
whereas the projection onto Xm equals nS by
As we will see later, an important role in the Chowla conjecture. Moreover, S, Xm , nS has
research around Sarnak’s conjecture is played by the property of being relative Kolmogorov
nil-systems. Green and Tao obtained the follow- with respect to its factor S, Xm2 , nS : On the
ing quantitative version of (S): other hand, the restriction of r to X  Xm2 is
of relative zero entropy over Xm2 :This yields rel-
Theorem 5.3 (Green and Tao 2012) Let G be a ative disjointness of systems S, Xm , nS and
simply-connected nilpotent Lie group with a dis- ðT  S, X  Xm2 , rjXX 2 Þ over their common fac-
m

crete and cocompact subgroup G. Let p : ℤ ! G tor S, Xm2 , nS . To complete the proof, we use the
be any its polynomial sequence: pðnÞ ¼ orthogonality of F to L2 Xm2 , nS .
p ð nÞ p ðnÞ
a1 1 . . . ak k ,
where pj : ℕ ! ℕ is a polynomial,
Remark 5.5 It still remains open whether (S)
j ¼ 1, . . . , k and f : G/G ! ℝ a Lipschitz function.
implies (C), see however Remark 5.16.
Then
In Huang et al. (2019a), Möbius orthogonal-
N ity for low complexity systems is discussed.
f ðpðnÞGÞmðnÞ ¼ O f ,G,G,A
nN log A N Following (Ferenczi 1997), we say that the
measure-complexity of m  M(T, X) is weaker
for all A > 0. than a ¼ (an)n1 if

In particular, all nilrotations are Möbius

orthogonal.

min m  1 : m [nj¼1 Bdn ðxi , eÞ > 1  e for some x1 , . . . , xm  X

lim inf ¼0
n!1 an

for each e > 0 (here dn ðy, zÞ ¼ Theorem 5.6 (Huang et al., 2019a) Suppose that
1 n
T j y, T j z ). (C) holds for correlations of order 2 (i.e., for r ¼ 1).
n j¼1 d
Then (T, X) is Möbius orthogonal whenever all
300 Sarnak’s Conjecture from the Ergodic Theory Point of View

invariant measures for (T, X) are of complexity MOMO property was introduced in El Abdalaoui
weaker than n. et al. (2018) to deal with Möbius orthogonality of
uniquely ergodic models of a given measure-
To obtain a non-conditional result, Huang, theoretic dynamical system. Moreover, we have:
Wang, and Ye used a difficult estimate of
Matomäki, Radziwiłł, and Tao (namely, “Trun- Theorem 5.9 (El Abdalaoui et al. 2018) The
cated Elliott on the average,” applied to m) from following conditions are equivalent:
(Matomäki et al. 2015). The cost to be paid is a
further strengthening of the assumptions on the (i) All zero entropy systems are Möbius orthog-
complexity of (T, X). onal, that is, Sarnak’s conjecture holds.
(ii) For each zero entropy system (T, X), we
Theorem 5.7 (Huang et al. 2019a) Suppose that have lim N!1 N1 nN f ðT n xÞmðnÞ ¼ 0
all invariant measures of (T, X) are of sub-
polynomial complexity, that is, their complexity when N ! 1, for each f  C(X), uniformly
is weaker than (nt)n1 for each t > 0. Then (T, in x  X, that is, uniform Sarnak’s conjec-
X) is Möbius orthogonal. ture holds.
(iii) All zero entropy systems enjoy the strong
See Huang et al. (n.d.) for the most recent MOMO property.
application of this result.
Finally, let us point out a consequence of the By taking f ¼ 1, we obtain that strong u-OMO
result on correlations of m of order 2. Directly implies the following:
from Corollary 4.12, we have:
1
uðnÞ ! 0 ð9Þ
Corollary 5.8 All topological dynamical sys- bK k<K bk n<bk þ1
K!1
tems whose all invariant measures yield systems
with discrete spectrum are Möbius orthogonal. for every sequence 0 ¼ b0 < b1 < b2 <    with
bk+1  bk ! 1. In particular, N1 nN uðnÞ ! 0.
In the uniquely ergodic case, an earlier and N!1

independent proof of this fact was given by In a similar way (by considering finite rotations),
Huang, Wang, and Zhang (Huang et al. 2019b) one can deduce N1 nN uðan þ bÞ ! 0. Thus, (8)
N!1
(for the totally uniquely ergodic case, see El can be seen as a form of aperiodicity. A further
Abdalaoui et al. (2017b)). The result also follows analysis reveals that, in fact, we deal with a special
from Huang et al. (2019a). behaviour of u on a typical short interval. All
strongly aperiodic multiplicative functions satisfy
(9) (this follows from Theorem A.1 (Matomäki
Strong MOMO Property et al. 2015)), hence condition (9) is satisfied both
Given an arithmetic function u, following El for m and l, cf. section “Behaviour on Short
Abdalaoui et al. (2018), we say that (T, X) satisfies Intervals.”
the strong u-OMO property if, for any increasing Recently, in Gomilko et al. (2020), the strong
sequence of integers 0 ¼ b0 < b1 < b2 <    with u-OMO property was rephrased in the language of
bk + 1  bk ! 1, for any sequence (xk) of points in functional analysis; and it is equivalent to
X, and any f  C(X), we have

1
1 lim uðnÞf ∘T n
f T nbk xk uðnÞ ! 0: ð8Þ K!1 bKþ1
kK bk n<bkþ1
bK k<K K!1 Cð X Þ
bk n<bkþ1
¼ 0 for all f  CðXÞ, ðbk Þ as above:
If u ¼ m, we speak about the strong MOMO
property. (The acronym MOMO stands for Usefulness of the strong MOMO concept is
Möbius Orthogonality of Moving Orbits.) Strong seen in the following result:
Sarnak’s Conjecture from the Ergodic Theory Point of View 301

Proposition 5.10 (El Abdalaoui et al. 2018) If orthogonal, and in particular, why zero entropy
(R, Z, D, k) is an ergodic (measure-theoretic) should play a special role. As we will see, how-
dynamical system and (T, X) is its uniquely ergo- ever, positive entropy systems are not expected to
dic model satisfying the strong MOMO property, enjoy the strong MOMO property, cf. Theorem
then all uniquely ergodic models of (R, Z, D, k) 5.9. Indeed, the following has been proved in El
are Möbius orthogonal. In fact, the strong MOMO Abdalaoui et al. (2018) (the result has been proved
holds in all of them. in El Abdalaoui et al. (2018) for the Liouville
function but it can also be proved for m):
Möbius orthogonality of Positive Entropy
Theorem 5.13 (El Abdalaoui et al. 2018) Let
Systems
u  {1, 0, 1}ℤ be a generic point for the mea-
If we take a positive entropy system (T, X), it is
sure nm2 : Then the following conditions are
natural to expect that it is not Möbius orthogonal.
equivalent:
Indeed, trivially, the full shift on {0, 1}ℤ is not,
and more generally subshifts of finite type are not,
(i) (T, X) satisfies strong u-OMO property.
see Karagulyan (2017). One can also show that
(ii) (T, X) is of zero entropy.
the subshift S, Xm2 (which is of positive entropy,
see Peckner (2015)) is not Möbius orthogonal,
As an immediate consequence, we have the
despite the fact that m2 itself is a completely
following:
deterministic point and Möbius orthogonality
holds at it: lim N!1 N1 nN f ðSn m2 ÞmðnÞ ¼ 0
Corollary 5.14 If Chowla conjecture holds, then
for each f  C Xm2 , see Ferenczi et al. (2018).
a system (T, X) has the strong MOMO property if
However, Sarnak’s conjecture does not exclude a
and only if it has zero entropy.
possibility that some positive entropy system is
also Möbius orthogonal. (It is mentioned in
Sarnak (n.d.) that Bourgain (unpublished) had (Slog) vs. (Clog)
such a construction.) Downarowicz and Serafin The logarithmic version of Sarnak’s conjecture
proved the following general result: was formulated in Tao (2016) along with (Clog)
and it postulates that
Theorem 5.11 (Downarowicz and Serafin
2019a) Fix an integer N  2. Let u be any 1 1
lim f ðT n xÞmðnÞ ¼ 0 Slog
bounded, real, aperiodic sequence. Then, there N!1 log N nN n
exists a subshift (S, X) over N symbols of entropy
arbitrarily close to log N, uncorrelated to u: (with all parameters as in (4)). In Tao (2017), Tao
lim N!1 N1 nN f ðSn xÞuðnÞ ¼ 0 for each showed the following:
f  C(X) and x  X.
Theorem 5.15 (Slog) is equivalent to (Clog).
Even more surprisingly, they proved a uniform
version of the above result: Remark 5.16 Combining Theorem 5.15 with
Corollary 4.3, we obtain that (Slog) implies (C)
Theorem 5.12 (Downarowicz and Serafin along a subsequence of logarithmic density 1. In
2019b) Under the same assumption on u, given particular, (S) implies (C) along a subsequence of
N  2, there exists a strictly ergodic subshift over full logarithmic density.
N symbols, of entropy arbitrarily close to log N,
uniformly uncorrelated to u. Let us here recall one more “logarithmic
conjecture” from (Tao 2017) which confirms a
Realizing that, one might be anxious what special role played by nil-systems in dynamics.
finally is the class of systems which are Möbius Let (T g0 , G/G) be a nilrotation. Let f  C(G/G) be
302 Sarnak’s Conjecture from the Ergodic Theory Point of View

Lipschitz continuous and x0  G. Then (for namely so-called logarithmic strong MOMO
H  N with H ! 1) property (cf. section “Strong MOMO Property”):

supg  G hH f g ðx0 GÞ mðn þ hÞ

T hþn 1 mðnÞ
lim f ∘T n ¼ 0:
n K!1 log bKþ1 kK bk nbkþ1
n
nN CðXÞ

¼ o ðH log N Þ: Snil
log Equivalently, for all increasing sequences
(bk)  ℕ with bk+1  bk ! 1, all (xk)  X and
Theorem 5.17 (Tao 2017) Snil is equivalent f  C(X),
log
to (Slog) (and (Clog)).
1 1
lim f T nbk xk mðnÞ
K!1 log bKþ1 kK n
Finally, as a consequence of the result on log- bk n<bkþ1
arithmic correlations of m of order 2 (using Cor- ¼ 0:
ollary 4.7), we obtain:
Theorem 5.20 (Gomilko et al. 2020) Assume
Corollary 5.18 All topological dynamical sys- that a topological system (T, X) satisfies the loga-
tems whose all invariant measures yield systems rithmic strong MOMO property. Then there exists
with singular spectrum are logarithmically a sequence A ¼ A(T, X)  ℕ with full logarithmic
Möbius orthogonal. density such that, for each f  C(X),

In general, we do not know if we can replace 1

“all invariant measures” with “all ergodic invari- lim mðnÞf ∘T n ¼ 0:
A∍N!1 N nN
ant measures” in the above corollary (the same CðXÞ

applies to Corollary 5.8). This replacement is

possible, however, when there are only count- In particular, Möbius orthogonality holds
ably many ergodic invariant measures, cf. the along a subsequence of full logarithmic density.
discussion on Frantzikinakis and Host (2018) in
section “Logarithmic Furstenberg Systems.” Remark 5.21 In Gomilko et al. (2020), using
Frantzikinakis and Host (2018), it is proved that
each system (T, X) for which Me(T, X) is countable
(S) vs. (Slog) satisfies the logarithmic strong MOMO property,
Clearly, (S) implies (Slog). As for the other direc- hence, for each such system Sarnak’s conjecture
tion, we have the following: holds in (logarithmic) density, cf. Theorem 6.10.

Theorem 5.19 (Gomilko et al. 2020) Suppose Strategies

that (Slog) holds. Then there exists a sequence of The first years of activity around Sarnak’s conjec-
logarithmic density 1, along which (S) holds for ture were devoted to proving Möbius orthogonal-
all zero entropy topological dynamical systems. ity in selected classes of zero entropy dynamical
systems. While this proved fruitful, and often
The idea of the proof is to use Theorem 5.15, some brilliant arguments were found ad hoc,
Remark 5.16, and then repeat the arguments from with a strong dependence on the class under con-
the proof of Theorem 5.4. sideration, it quickly became clear that it won’t be
Notice that the sequence of logarithmic density sufficient. Two main strategies to attack (S) arose:
1 in the above result is universal for all zero
entropy systems. In Gomilko et al. (2020), one A. The first strategy is to look for some additional
more version of Möbius orthogonality is studied, intrinsic structure in zero entropy systems that
Sarnak’s Conjecture from the Ergodic Theory Point of View 303

could be used to prove orthogonality from m, A basic method to prove orthogonality with a
namely, internal disjointness. Here, a priori, multiplicative function comes from the multipli-
one does not use any other property of m than cative orthogonality criterion (MOC):
multiplicativity and boundedness.
B. As we have seen, (S) is intimately related to Theorem 6.1 (Bourgain et al. 2013; Daboussi,
(C), and therefore one cannot expect to con- Kátai 1986) Assume that (fn)  ℂ is a bounded
firm (S) without using further number- sequence. Assume that for all (sufficiently large)
theoretic properties of m. This directs atten- prime numbers p 6¼ q,
tion to aperiodicity and behavior on the
so-called short intervals. It extends further 1
lim f pn f qn ¼ 0: ð10Þ
to studying Furstenberg systems of m N!1 N
nN
(including the logarithmic ones) and trying to
interpret arithmetic properties of m as ergodic Then, for each bounded multiplicative function
properties of the corresponding dynamical u, we have lim N!1 N1 nN f n uðnÞ ¼ 0. In par-
systems. One can hope finally to deduce ticular, (fn) is Möbius orthogonal.
(some kind of) Furstenberg (!) disjointness of
Furstenberg systems of m with a wide subclass Remark 6.2 Notice that Theorem 6.1 does not
of zero entropy systems (hopefully, with all require from u anything but multiplicativity and
such systems). boundedness.

As we will see, these two approaches often In the dynamical context (T, X) the simplest
intertwine, proving once again that number theory way to use Theorem 6.1 is to take fn ¼ f(Tnx). In
and ergodic theory should not be studied sepa- this form, MOC appeared for the first time in
rately from each other. Bourgain et al. (2013) and was used to prove
that the horocycle flows are Möbius orthogonal.
To see how MOC is used and how it is related to
Arithmetic Properties of the Möbius Furstenberg’s disjointness theory (Furstenberg
Function 1967), assume that M(T, X) ¼ {m}, X fdm ¼ 0,
and the corresponding measure-theoretic system
Multiplicativity is totally ergodic. Then, any measure r  V((x, x))
Internal disjointness. Joinings (introduced in a (considered in the topological dynamical system
seminal paper of Furstenberg (Furstenberg 1967)) (Tp  Tq, X  X)) is a joining of T p and T q. If we
have been present in ergodic theory for over now assume that (Tp, X, m) and (Tq, X, m) are dis-
50 years. Disjointness (absence of nontrivial join- joint for sufficiently large primes p 6¼ q, then
ings), as a form of an extremal non-isomorphism r ¼ m m and, as a result, the limit in (10) equals
and a measure-theoretic invariant, has always XX f f dr ¼ 0, that is, the assumptions of MOC
played a crucial role in classification problems. are satisfied.
(Recall also that different powers for a typical In general, a use of MOC is not that simple.
automorphism of a standard Borel space are Consider an irrational rotation Tx ¼ x + a on the
pairwise disjoint (del Junco 1981). See also circle X ¼ ℝ/ℤ. To see that (10) holds for all
more recent Kanigowski et al. (2020).) It characters, one uses the Weyl criterion on uniform
appeared, however, in many other contexts, distribution. However, there are continuous zero
including homogenous dynamics, with applica- mean functions for which (10) fails (Kułaga-
tions in number theory. Sarnak’s conjecture gave Przymus and Lemańczyk 2015), which shows
yet a new impetus, in particular for studying clearly, that in general we can only expect (10)
(approximate) disjointness for different to hold for a linearly dense set of continuous
sub-actions. functions.
304 Sarnak’s Conjecture from the Ergodic Theory Point of View

In some cases, MOC cannot be applied directly properties of m (other than multiplicativity)
(e.g., when the systems under consideration fail to begin to play a significant role.
be weakly mixing) and the spectral approach can Zero entropy continuous interval maps. In
help. Examples can be found in (El Abdalaoui Karagulyan (2015), (S) for zero entropy continu-
et al. (2014), (2016), and Bourgain (2013). ous intervals maps and orientation-preserving cir-
AOP property. The following ergodic counter- cle homeomorphisms is established. The starting
part of MOC was developed in El Abdalaoui et al. point for developing the main tools is the result of
(2017b): an ergodic automorphism ðT, X, B, mÞ is Davenport (7), which shows clearly that the
said to have asymptotically orthogonal powers examples under consideration are indeed “rela-
(AOP) if for each f , g  L20 ðX, B, mÞ, we have tives” of irrational rotations. Additionally, in
order to treat the case of interval maps one studies
o-limit sets and it turns out that, in fact, one deals
lim sup f g dk
ℙ 3p, q!1, p6¼qk  J e ðT p , T q Þ XX with an odometer.
Synchronized automata. In Deshouillers
¼ 0: ð11Þ et al. (2015), Deshouillers, Drmota, and Müllner
prove that (S) is true for automatic sequences
Clearly, if the powers of T are pairwise disjoint, generated by synchronizing automata (the inputs
then T enjoys the AOP property. However, this are read with the most significant digit first). In
condition is not necessary, the powers of T having fact, they prove orthogonality of such sequences
AOP property may even be isomorphic. Moreover, from any bounded function u that is aperiodic.
AOP implies total ergodicity and zero entropy Almost periodic sequences. We say that a
(El Abdalaoui et al., 2017b). The relation between sequence is Weyl rationally almost periodic
strong MOMO and AOP properties is described by (WRAP) whenever it can be approximated arbi-
the following result: trarily well by periodic sequences in Weyl
pseudo-metric dW given by dW ðx, yÞ ¼
Theorem 6.3 (El Abdalaoui et al. 2017a, b) Let lim supN!1 supl1 N1 j fl  n  l þ N : xðnÞ 6¼
u be a bounded multiplicative function. Suppose
yðnÞg j . It is proved in Bergelson et al. (2019)
that ðR, Z, C, kÞ satisfies AOP. Then the following
that each subshift (S, Xx) given by a Weyl almost
are equivalent:
periodic sequence is Möbius orthogonal (in fact,
we have orthogonality to any bounded aperiodic
• u satisfies (9);
arithmetic function u).
• The strong u-OMO property is satisfied in each
uniquely ergodic model (T, X) of R.
Behaviour on Short Intervals
In particular, if the above holds, for each During the last four years, an enormous progress
f  C(X), we have concerning the short interval behavior of strongly
aperiodic multiplicative functions has been made
1 due to the breakthrough result of Matomäki and
f ðT n xÞuðnÞ ! 0 uniformly on X: Radziwiłł. The main result of Matomäki and
N nN N!1
Radziwiłł (2016), for m, in its simplified form
can be written as
Aperiodicity
As all periodic sequences are orthogonal to m, one 1 1
lim mðhÞ ¼ 0:
M 1mM H
can expect that sequences with some properties M, H!1 mh<mþH
with H¼oðMÞ
similar to periodicity will also be Möbius orthog-
onal. Notice also that Möbius orthogonality of
periodic sequences ((S) for rotations on finite ð12Þ
groups) corresponds to m being aperiodic. This This gave an impetus to study convergence on
is the simplest situation where some additional short intervals in ergodic theory and it has become
Sarnak’s Conjecture from the Ergodic Theory Point of View 305

a new, crucial player from the point of view of Moreover, it is shown that all synchronized
Sarnak’s conjecture. Condition (12) can be also automata yield WRAP sequences. Thus, the
reformulated in the following way: for each above theorem strengthens the aforementioned
(bn)  ℕ with bn+1  bn ! 1, we have result by Deshouillers, Drmota, and Müllner in
Deshouillers et al. (2015).
1 Rigid systems. In Kanigowski et al. (n.d.-a),
lim mðnÞ ¼ 0, ð13Þ Kanigowski, Lemańczyk, and Radziwiłł study
K!1 bKþ1
kK bk n<bkþ1
rigid systems. (A measure-theoretic system
ðR, Z, C, kÞ is rigid if, for some increasing
cf. Section “Strong MOMO Property” 5.3. sequence (qn) of natural numbers, we have
Almost periodic sequences. In Bergelson f ∘ Rqn ! f in L2(Z, k) for each f  L2(Z, k).
et al. (2019), in case of WRAP x, the authors Rigid systems are of zero entropy. Moreover, the
also ask about the behavior of averages of the typical measure-theoretic automorphism is rigid
form and weakly mixing.) To formulate their results,
we need some definitions and facts. Given a nat-
1
f Sh z mðnÞ ð14Þ ural number q, the sum ℙ 3 pjq1/p is called the
H mh<mþH
prime volume of q. The prime volume grows
slowly with q:
(where z  Xx) for large values of H and arbitrary
m  ℕ. Under (C), convergence to zero uni- 1
 log log log q þ Oð1Þ:
formly in m does not take place; however, it is p
pjq
shown in Bergelson et al. (2019) that for a “typi-
cal” m  ℕ the averages in (14) are small. The
However “most” of the time, the prime volume
key argument in the proof comes from a result of
of q stays bounded: if we set
Matomäki, Radziwiłł, and Tao:

Theorem 6.4 (Matomäki et al. 2015) For each 1

D j≔ q  ℕ : <j ,
periodic sequence a(n), we have p
pjq

1 1 then d D j ! 1 when j ! 1. A topological sys-

lim aðhÞmðhÞ ¼ 0:
H, M!1 M Mm<2M H mh<mþH tem (T, X) is said to be good if for every n  M(T, X)
H¼oðMÞ at least one of the following conditions holds:

ð15Þ • (BPV rigidity): ðT, X, B, nÞ is rigid along a

sequence (qn)n1 with bounded prime vol-
ume, that is, there exists j such that ðqn Þn1 
As a consequence, we have: D j;
• (PR rigidity): ðT, X, B, nÞ has polynomial
Theorem 6.5 (Bergelson et al. 2019) Suppose rate of rigidity, that is, there exists a linearly
that x  ℤ is WRAP. Then for all f  C(Xx) and dense (in C(X)) set F  C(X) such that for each
z  Xx, f  F we can find d > 0 and a sequence
(qn)n1 satisfying
1 1
lim f Sh z mðhÞ ¼ 0: qdn
H, M!1 M Mm<2M H mh<mþH 2
H¼oðMÞ f ∘T j qn
f L2 ðnÞ
! 0:
j¼qdn

ð16Þ
306 Sarnak’s Conjecture from the Ergodic Theory Point of View

They prove the following: Ferenczi and Mauduit 2018; Karagulyan 2020).
Other applications are given for C2+eAnzai skew
Theorem 6.6 (a) Assume the (T, X) is a good products and for some so-called Rokhlin exten-
topological system. Then (T, X) is Möbius sions of rotations.
orthogonal.
(b) Suppose that each ergodic invariant mea- One more of the consequences is the following
sure of (T, X) yields either BPV rigidity or PR result:
rigidity and Me(T, X) is countable then (T, X) is
Möbius orthogonal. Corollary 6.9 (Kanigowski et al. n.d.-a) No
Furstenberg system of the Möbius function m is
A key tool here is a strengthening of the main either BPV or PR rigid. The same holds for the
result of Matomäki and Radziwiłł (Matomäki & Liouville function l.
Radziwiłł 2016) (cf. (12)) to short interval behav-
iour along arithmetic progressions:
Logarithmic Furstenberg Systems
Frantzikinakis and Host study logarithmic
Theorem 6.7 (Kanigowski et al. n.d.-a) For
Furstenberg systems associated to m (and l).
each e > 0, there exists L0 such that for each
They prove the following remarkable result:
L  L0 and q  1 satisfying pjq1/p
 (1  e)p  L1/p we can find M0 ¼ M0(q, L)
Theorem 6.10 (Frantzikinakis and Host 2018)
such that for all M  M0, we have
Each zero entropy topological system (T, X) with
only countably many ergodic measures is loga-
rithmically Möbius orthogonal.
M=Lq q1
mðmÞ In particular, uniquely ergodic systems of zero
j¼0 a¼0 m  ½z þ jLq, z þ ð j þ 1ÞLq topological entropy satisfy (Slog). The key argu-
m a mod q ment in the proof of Theorem 6.10 is the following
structural result on the logarithmic Furstenberg
< eM
systems of m and l:
for some 0  z < Lq.
Theorem 6.11 (Frantzikinakis and Host 2018)
Each logarithmic Furstenberg system of m or l
Remark 6.8 Despite the fact that PR rigidity
is a factor of a system that:
does not seem to be stable under different
(uniquely ergodic) models of a measure-
• has no irrational spectrum,
preserving transformation, assuming (T, X) is
• has ergodic components isomorphic to direct
uniquely ergodic, it is proved that (T, X) satisfies
products of infinite-step nilsystems and
the strong MOMO property whenever its unique
Bernoulli systems.
invariant measure yields either BPV or PR rigid-
ity. Via Proposition 5.10, we obtain that if in a
model PR rigidity holds, all of the models are
Möbius orthogonal. This, in particular, applies to The starting point for the proof of the above
all ergodic transformations with discrete spec- theorem, resulting in a reduction of the problem to
trum. Moreover, it is shown in Kanigowski et al. purely ergodic context, is an identity of Tao
(n.d.-a) that for a.e. IET (of d  3 intervals) BPV (implicit in (Tao 2016)) showing that self-
rigidity holds, so a.e. IET (and all their uniquely correlations of m (and l) are averages of its
ergodic models) is Möbius orthogonal. This is to dilated self-correlations with prime dilates.
be compared with previously known results for Frantzikinakis and Host also prove that logarith-
3-IETs (Bourgain 2013; Chaika and Eskin 2019; mic Furstenberg systems of m (and l) are “almost
Sarnak’s Conjecture from the Ergodic Theory Point of View 307

determined” by strongly stationary processes Future Directions

(introduced by Furstenberg and Katznelson in the
1990s). The structure of (measure-theoretic) Detecting Zero Entropy
dynamical systems given by strongly stationary As shown in Theorem 5.13 and Corollary 5.14,
processes has been described by Jenvey (Jenvey under the Chowla conjecture, the Möbius function
1997) who proved that ergodicity implies Bernoulli is a sequence that:
(cf. Theorems 4.2 and 4.4) and Frantzikinakis
(Frantzikinakis 2004) who described the ergodic • is strong MOMO orthogonal to all zero entropy
decomposition in the non-ergodic case. systems,
The above results are extended in • is never strong MOMO orthogonal to any pos-
Frantzikinakis and Host (2019) to strongly aperi- itive entropy system.
odic multiplicative functions. Moreover, the fol-
lowing multidimensional result is proved: In view of this, it is natural to study the follow-
ing problem:
Theorem 6.12 Let f1, . . . , fl: ℕ !  be multipli-
cative functions. Let (R, Y) be a topological Question 1 Which numerical sequences distin-
dynamical system and let y  Y be a logarithmi- guish between zero and positive entropy systems?
cally generic point for a measure n with zero
entropy and having at most countably many ergo- Note that in view of the results of Down-
dic components, all of which are totally arowicz and Serafin (Theorems 5.11 and 5.12),
ergodic. Then for every g  C(Y) that is orthog- “usual orthogonality” or even its uniform version
onal in L2(n) to all R-invariant functions we have is insufficient for these needs.

1 gðRn yÞPl j¼1 f j n þ h j

lim Proving the Strong MOMO Property
N!1 log N nN n It was already asked in Ferenczi et al. (2018)
whether whenever we have a zero entropy system
(T, X) for which we can prove Möbius orthogo-
¼0 ð17Þ
nality, then we can prove the strong MOMO prop-
erty. Recently, Lemanczyk and Müllner (n.d.)
for all h1, . . . , hl  ℤ.
proved the strong MOMO property for
(primitive) automatic sequences (previously
The unweighted version of (with g ¼ 1) is
known to be Möbius orthogonal by Müllner
expected to hold if the shifts are distinct and at
(2017)), answering a question from Ferenczi
least one of the multiplicative functions is
et al. (2018). Yet another question persists:
strongly aperiodic. This is the logarithmically
averaged version of Elliott conjecture (Elliott
Question 2 Do horocycle flows satisfy the strong
1990, 1994; Matomäki et al. 2015). In the special
MOMO property?
case of irrational rotations, with l ¼ 1, Theorem
6.12 is the logarithmically averaged variant of a
We do not even know whether Möbius orthog-
classical result of Daboussi (Collectif 1975;
onality takes place in all uniquely ergodic models
Daboussi and Delange 1974, 1982). Already in
of horocycle flows.
case l ¼ 2 it is completely new.

Mixing Properties of Furstenberg Systems

Corollary 6.9 induces the following problem:

Question 3 Are Furstenberg systems of l mildly

mixing?
308 Sarnak’s Conjecture from the Ergodic Theory Point of View

For m we need to take into account that its del Junco A (1981) Disjointness of measure-preserving
Furstenberg systems have the discrete spectrum transformations, minimal selfjoinings and category.
In: Ergodic theory and dynamical systems, I (College
factor given by the Mirsky measure of m2. Park, Md., 1979–80), volume 10 of Progress in math-
ematics, pp 81–89. Birkhäuser, Boston
Deshouillers J-M, Drmota M, Müllner C (2015) Automatic
Furstenberg Disjointness in Non-ergodic Case sequences generated by synchronizing automata fulfill
As Host and Frantzikinakis’ analysis shows, if the the Sarnak conjecture. Studia Math 231(1):83–95
(potential) logarithmic Furstenberg systems of l or Downarowicz T (2011) Entropy in dynamical systems,
volume 18 of New Mathematical Monographs. Cam-
m are non-ergodic, then they are very non-ergodic.
bridge University Press, Cambridge
One of open questions by Frantzikinakis is whether Downarowicz T, Serafin J (2019a) Almost full entropy
the system 2 3 ðx, yÞ 7! ðx, x þ yÞ  2 consid- subshifts uncorrelated to the Möbius function. Int
ered with Lebesgue measure can be a Furstenberg Math Res Not 2019(11):3459–3472
Downarowicz T, Serafin J (2019b) A strictly ergodic, pos-
system of l. Of course this example is a measure-
itive entropy subshift uniformly uncorrelated to the
theoretic system which is Furstenberg disjoint Möbius function. Studia Math 251(2020):195–206.
from all ergodic systems. It seems to be a prob- Published online (Online First)
lem of independent interest to fully understand El Abdalaoui EH, Lemańczyk M, de la Rue T (2014) On
spectral disjointness of powers for rank-one transfor-
the class of transformations disjoint from all
mations and Möbius orthogonality. J Funct Anal
ergodic transformations. 266(1):284–317
El Abdalaoui EH, Kasjan S, Lemańczyk M (2016) 0-1
sequences of the Thue-Morse type and Sarnak’s con-
jecture. Proc Am Math Soc 144(1):161–176
Bibliography El Abdalaoui EH, Kułaga-Przymus J, Lemańczyk M, de la
Rue T (2017a) The Chowla and the Sarnak conjectures
Bergelson V, Kułaga-Przymus J, Lemańczyk M, Richter from ergodic theory point of view. Discrete Contin Dyn
FK (2019) Rationally almost periodic sequences, poly- Syst 37(6):2899–2944
nomial multiple recurrence and symbolic dynamics. El Abdalaoui EH, Lemańczyk M, de la Rue T (2017b)
Ergodic Theory Dyn Syst 39(9):2332–2383 Automorphisms with Quasidiscrete spectrum, multipli-
Bourgain J (2013) On the correlation of the Moebius func- cative functions and average orthogonality along short
tion with rank-one systems. J. Anal. Math. intervals. Int Math Res Not 2017(14):4350–4368
120:105–130 El Abdalaoui EH, Kułaga-Przymus J, Lemańczyk M, de la
J. Bourgain, P. Sarnak, and T. Ziegler. Disjointness of Rue T (2018) Möbius disjointness for models of an
Möbius from horocycle flows. In From Fourier analy- ergodic system and beyond. Israel J Math
sis and number theory to Radon transforms and geom- 228(2):707–751
etry, volume 28 of Dev. Math., pages 67–83. Springer, Elliott PDTA (1990) Multiplicative functions jg j  1 and
New York, 2013. their convolutions: an overview. In: Séminaire de
Chaika J, Eskin A (2019) Möbius disjointness for interval Théorie des Nombres, Paris 1987–88, volume 81 of
exchange transformations on three intervals. Journal of Progress in mathematics, pp 63–75. Birkhäuser Boston,
Modern Dynamics 14(1):55–86 Boston
Chowla S (1965) The Riemann hypothesis and Hilbert’s Elliott PDTA (1992) On the correlation of multiplicative
tenth problem. Mathematics and its applications, functions. Notas Soc Mat Chile 11(1):1–11
vol 4. Gordon and Breach Science Publishers, Elliott PDTA (1994) On the correlation of multiplicative
New York and the sum of additive arithmetic functions. Mem Am
Collectif (1975) Fonctions multiplicatives presque péri- Math Soc 112(538):viii+88
odiques B. In Journées arithmétiques de Bordeaux, Ferenczi S (1997) Measure-theoretic complexity of ergodic
number 24–25 in Astérisque, pp 321–324. Société systems. Isr J Math 100(1):189–207
mathématique de France S. Ferenczi, J. Kułaga-Przymus, and M. Lemańczyk.
Daboussi H, Delange H (1974) Quelques propriétés des Sarnak’s conjecture: what’s new. In S. Ferenczi,
fonctions multiplicatives de module au plus égal à 1. C J. Kułaga-Przymus, and M. Lemańczyk, Ergodic the-
R Acad Sci Paris Sér A 278:657–660 ory and dynamical systems in their interactions with
Daboussi H, Delange H (1982) On multiplicative arithmet- arithmetics and combinatorics, volume 2213 of Lecture
ical functions whose modulus does not exceed one. Notes in Math., 163–235. Springer, Cham, 2018.
J Lond Math Soc 26(2):245–264 Ferenczi S, Mauduit C (2018) On Sarnak’s conjecture and
Davenport H (1937) On some infinite series involving Veech’s question for interval exchanges. J Anal Math
arithmetical functions. II. Quart J Math Oxford 134(2):545–573
8:313–320
Sarnak’s Conjecture from the Ergodic Theory Point of View 309

Frantzikinakis N (2004) The structure of strongly station- Kułaga-Przymus J, Lemańczyk M (2015) The Möbius
ary systems. J Anal Math 93(1):359–388 function and continuous extensions of rotations.
Frantzikinakis N (2017) Ergodicity of the Liouville system Monatsh Math 178(4):553–582
implies the Chowla conjecture. Discrete Anal, Paper Lemańczyk M, Müllner (n.d.) C. Automatic sequences are
no. 19, 41pp orthogonal to aperiodic multiplicative functions.
Frantzikinakis N, Host B (2018) The logarithmic Sarnak Published online in Discrete Continuous Dynamical
conjecture for ergodic weights. Ann of Math Systems (2020). https://doi.org/10.3934/dcds.2020260
187(3):869–931 Matomäki K, Radziwiłł M (2016) Multiplicative functions
Frantzikinakis N, Host B (2019) Furstenberg systems of in short intervals. Ann Math 183(3):1015–1056
bounded multiplicative functions and applications. Int Matomäki K, Radziwiłł M, Tao T (2015) An averaged form
Math Res Not. https://doi.org/10.1093/imrn/rnz037 of Chowla’s conjecture. Algebra Number Theory
Furstenberg H (1967) Disjointness in ergodic theory, min- 9(9):2167–2196
imal sets, and a problem in Diophantine approximation. Müllner C (2017) Automatic sequences fulfill the Sarnak
Math Syst Theory 1:1–49 conjecture. Duke Math J 166(17):3219–3290
Gomilko A, Kwietniak D, Lemańczyk M (2018) Sarnak’s Peckner R (2015) Uniqueness of the measure of maximal
conjecture implies the Chowla conjecture along a sub- entropy for the squarefree flow. Isr J Math
sequence. In: Ferenczi S, Kułaga-Przymus J, 210(1):335–357
Lemańczyk M (eds) Ergodic theory and dynamical Ram Murty M, Vatwani A (2018) A remark on a conjecture
systems in their interactions with arithmetics and com- of Chowla. J Ramanujan Math Soc 33(2):111–123
binatorics, volume 2213 of Lecture Notes in Mathe- Ramaré O (2018) Chowla’s conjecture: from the Liouville
matics. Springer, Cham, pp 237–247 function to the Moebius function. In: Ferenczi S,
Gomilko A, Lemańczyk M, de la Rue T (2020) Möbius Kułaga-Przymus J, Lemańczyk M (eds) Ergodic the-
orthogonality in density for zero entropy dynamical ory and dynamical systems in their interactions with
systems. Pure and Applied Functional Analysis arithmetics and combinatorics, volume 2213 of Lec-
5:1357–1376 ture notes in mathematics. Springer, Cham,
Green B, Tao T (2012) The Möbius function is strongly pp 317–323
orthogonal to nilsequences. Ann Math 175(2):541–566 Sarnak P. Three lectures on the Möbius function, random-
Huang W, Wang Z, Ye X (2019a) Measure complexity and ness and dynamics. Webpage. http://publications.ias.
Möbius disjointness. Adv Math 347:827–858 edu/sarnak/
Huang W, Wang Z, Zhang G (2019b) Möbius disjointness Tao T (2016) The logarithmically averaged Chowla and
for topological models of ergodic systems with discrete Elliott conjectures for two-point correlations. Forum
spectrum. J Mod Dyn 14:277–290 Math Pi 4:e8, 36
Huang W, Liu J, Wang K. (n.d.) Möbius disjointness for Tao T (2017) Equivalence of the logarithmically averaged
skew products on a circle and a nilmanifold. Preprint. Chowla and Sarnak conjectures. In: Number theory –
https://arxiv.org/abs/1907.01735 Diophantine problems, uniform distribution and appli-
Jenvey E (1997) Strong stationarity and de Finetti’s theo- cations. Springer, Cham, pp 391–421
rem. J Anal Math 73(1):1–18 Tao T (n.d.) The logarithmically averaged and non-
Kamae T (1973) Subsequences of normal sequences. Isr logarithmically averaged Chowla conjectures.
J Math 16:121–149 Webpage, https://terrytao.wordpress.com/2017/10/20/
Kanigowski A, Lemańczyk M, Radziwiłł M. (n.d.-a) Rigidity thelogarithmically-averaged-and-non-logarithmically-
in dynamics and Möbius disjointness, to appear in averaged-chowlaconjectures/
Fundamenta Math. https://arxiv.org/abs/1905.13256 Tao T, Teräväinen J (2018) Odd order cases of the loga-
Kanigowski A, Lemańczyk M, Ulcigrai C. (2020) On rithmically averaged Chowla conjecture. J Théor
disjointness properties of some parabolic flows. Invent Nombres Bordeaux 30(3):997–1015
Math 221(1):1–111. https://arxiv.org/abs/1810.11576 Tao T, Teräväinen J (2019a) The structure of correlations of
Karagulyan D (2015) On Möbius orthogonality for interval multiplicative functions at almost all scales, with appli-
maps of zero entropy and orientation-preserving circle cations to the Chowla and Elliott conjectures. Algebra
homeomorphisms. Ark Mat 53(2):317–327 Number Theory 13(9):2103–2150
Karagulyan D (2017) On Möbius orthogonality for sub- Tao T, Teräväinen J (2019b) The structure of logarithmi-
shifts of finite type with positive topological entropy. cally averaged correlations of multiplicative functions,
Stud Math 237(3):277–282 with applications to the Chowla and Elliott conjectures.
Karagulyan D (2020) Hausdorff dimension of a class of Duke Math J 168(11):1977–2027
three-interval exchange maps. Discrete and Contin Weiss B (1971) Normal sequences as collectives. In: Pro-
Dynam Syst 40(3):1257–1281 ceedings of the symposium on topological dynamics
Kátai I (1986) A remark on a theorem of H. Daboussi. Acta and ergodic theory. University of Kentucky
Math Hungar 47(1–2):223–225
310 Sarnak’s Conjecture from the Ergodic Theory Point of View

Further Reading ergodic affine unipotent diffeomorphisms on

Baake M, Huck C (2015) Ergodic properties of visible nilmanifolds. Studia Math 244(1):43–97
lattice points. Proceedings of the Steklov Institute of Forni G, Kanigowski A (2020) Mutliple mixing and
Mathematics 288(1):165–188 disjointness for time changes of bounded-type Heisen-
Bergelson V, Ruzsa I (2002) Squarefree numbers, IP sets and berg nilflows. ournal de l’École polytechnique –
ergodic theory. In Paul Erdos and his mathematics, I Mathématiques 7:63–91
(Budapest, 1999), volume 11 of Bolyai Soc. Math. Frantzikinakis N (2018) An averaged Chowla and Elliott
Stud., pages 147–160. János Bolyai Math. Soc., Buda- conjecture along independent polynomials. Int Math
pest, 2002 Res Not 2018(12):3721–3743
Bourgain J (2013) Möbius-Walsh correlation bounds and Green B (2012) On (not) computing the Möbius function
an estimate of Mauduit and Rivat. J Anal Math using bounded depth circuits. Combin Probab Comput
119:147–163 21(6):942–951
de la Bretèche R, Tenenbaum G (2019) A remark on He X, Wang Z. Möbius disjointness for nilsequences along
Sarnak’s conjecture. Q J Math 70(1):371–378 short intervals. Preprint, https://arxiv.org/abs/1905.
Cellarosi F, Vinogradov I (2013) Ergodic properties of k- 02864
free integers in number fields. J Mod Dyn 7(3):461–488 Huang W, Lian Z, Shao S, Ye X (2017) Sequences from
Dartyge C, Tenenbaum G (2005) Sommes des chiffres de zero entropy noncommutative toral automorphisms and
multiples d’entiers. Ann Inst Fourier (Grenoble) 55 Sarnak Conjecture. J Differential Equations 263
(7):2423–2474 (1):779–810
Downarowicz T, Kasjan S (2015) Odometers and Toeplitz Huck C (2018) On the logarithmic probability that a ran-
systems revisited in the context of Sarnak’s conjecture. dom integral ideal is A-free. In Ferenczi S, Kułaga-
Studia Math 229(1):45–72 Przymus J, Lemanczyk M (eds) Ergodic theory and
Drmota M (2014) Subsequences of automatic sequences dynamical systems in their interactions with arithmetics
and uniform distribution. In Uniform distribution and and combinatorics, volume 2213 of Lecture notes in
quasi-Monte Carlo methods, volume 15 of Radon mathematics. Springer, Cham, pp 249–258
Ser. Comput. Appl. Math. De Gruyter, Berlin, pp Kasjan S, Keller G, Lemanczyk M (2019) Dynamics of B-
87–104 free sets: a view through the window. Int Math Res Not
Drmota M, Müllner C, Spiegelhofer L (2018) Möbius 2019(9):2690–2734
orthogonality for the Zeckendorf sum-of-digits func- Keller G. Generalized heredity in B-free systems. Preprint,
tion. Proc Amer Math Soc 146(9):3679–3691 https://arxiv.org/abs/1704.04079
Dymek A (2017) Automorphisms of Toeplitz B-free sys- Konieczny J (2020) Möbius orthogonality for q-semi-
tems. Bull Pol Acad Sci Math 65(2):139–152 multiplicative sequences. Monatshefte Math 192
Dymek A, Kasjan S, Kułaga-Przymus J, Lemanczyk M (4):853–882
(2018) B-free sets and dynamics. Trans Amer Math Soc Konieczny J, Kupsa M, Kwietniak D (2018) Arcwise con-
370(8):5425–5489 nectedness of the set of ergodic measures of hereditary
Eisner T (2015) A polynomial version of Sarnak’s conjec- shifts. Proc Amer Math Soc 146(8):3425–3438. https://
ture. C. R. Math Acad Sci Paris 353(7):569–572 arxiv.org/abs/1610.00672
El Abdalaoui EH. Oscillating sequences, Gowers norms Kułaga Przymus J, Lemanczyk M (2019) Möbius
and Sarnak’s conjecture. Preprint, https://arxiv.org/abs/ disjointness along ergodic sequences for uniquely ergo-
1704.07243 dic actions. Ergodic Theory Dynam Systems 39
El Abdalaoui EH, Askri G, Marzougui H. (2018) Möbius (10):2793–2826
disjointness conjecture for local dendrite maps. Non- Kułaga-Przymus J, Lemanczyk M, Weiss B (2015) On
linearity 32(1):285–300 invariant measures for B-free systems. Proc Lond
El Abdalaoui EH, Lemanczyk M, de la Rue T (2015) A Math Soc (3), 110(6):1435–1474
dynamical point of view on the set of B-free integers. Kułaga-Przymus J, Lemanczyk M, Weiss B (2016) Hered-
Int Math Res Not 2015(16):7258–7286 itary subshifts whose simplex of invariant measures is
Fan A-H, Jiang Y (2018) Oscillating sequences, MMA and Poulsen. In Ergodic theory, dynamical systems, and the
MMLS flows and Sarnak’s conjecture. Ergodic Theory continuing influence of John C. Oxtoby, volume 678 of
Dynam. Systems 38(5):1709–1744 Contemp. Math. Amer. Math. Soc. Providence, RI,
Ferenczi S, Kułaga-Przymus J, Lemanczyk M, Mauduit C pp 245–253
(2016) Substitutions and Möbius disjointness. In Ergo- Liu J, Sarnak P (2015) The Möbius function and distal
dic Theory, Dynamical Systems, and the Continuing flows. Duke Math J 164(7):1353–1399
Influence of John C. Oxtoby, volume 678 of Contemp. Matomäki K, Radziwiłł M, Tao T (2016) Sign patterns of
Math. Amer. Math. Soc., Providence, RI, pp 151–173 the Liouville and Möbius functions. Forum of Mathe-
Flaminio L, Fraczek K, Kułaga-Przymus J, Lemanczyk M matics, Sigma, 4
(2019) Approximate orthogonality of powers for
Sarnak’s Conjecture from the Ergodic Theory Point of View 311

Mauduit C, Rivat J (2015) Prime numbers along Rudin– Sarnak P (2012) Mobius randomness and dynamics. Not S
Shapiro sequences. J Eur Math Soc (JEMS) 17 Afr Math Soc 43(2):89–97
(10):2595–2642 Sawin W. Dynamical models for Liouville and obstruc-
McNamara R. Sarnak’s conjecture for sequences of almost tions to further progress on sign patterns. Preprint,
quadratic word growth. Preprint, https://arxiv.org/abs/ https://arxiv.org/abs/1809.03280
1901.06460 Soundararajan K (2017) The Liouville function in short
Mentzen MK (2017) Automorphisms of subshifts defined intervals. Astérisque, 2015/2016(390):Exp. No. 1119,
by B-free sets of integers. Colloq Math 147(1):87–94 453–479. Séminaire Bourbaki. Vol. 2015/2016.
Mirsky L (1947) Note on an asymptotic formula connected Exposés 1104–1119
with r-free integers. Quart J Math Oxford Ser 18:178–182 Sun W. Sarnak’s conjecture for nilsequences on arbitrary
Mirsky L (1949) Arithmetical pattern problems relating to number fields and applications. Preprint, https://arxiv.
divisibility by rth powers. Proc London Math Soc (2) org/abs/1902.09712
50:497–508 Tao T. Furstenberg limits of the Liouville function.
Peckner R (2018) Möbius disjointness for homogeneous Webpage, https://terrytao.wordpress.com/2017/03/05/
dynamics. Duke Math J 167(14):2745–2792 furstenberg-limits-of-the-liouville-function/
Pleasants PAB, Huck C (2013) Entropy and diffraction of Teräväinen J (2018) On binary correlations of multiplica-
the k-free points in n-dimensional lattices. Discrete tive functions. Forum Math Sigma 6:e10, 41
Comput Geom 50(1):39–68 Veech WA (2017) Möbius orthogonality for generalized
Ryzhikov VV (2013) Bounded ergodic constructions, Morse-Kakutani flows. Am J Math 139(5):1157–1203
disjointness, and weak limits of powers. Trans Moscow Wang Z (2017) Möbius disjointness for analytic skew
Math Soc 165–171 products. Invent Math 209(1):175–196
 measures the distortion of volume under f in
Smooth Ergodic Theory that metric.
Hopf argument A technique developed by
Amie Wilkinson Eberhard Hopf for proving that a conservative
Northwestern University, Evanston, IL, USA diffeomorphism or flow is ergodic. The argu-
ment relies on the Ergodic Theorem for invert-
ible transformations, the density of continuous
Article Outline functions among integrable functions, and the
existence of stable and unstable foliations for
Glossary the system. The argument has been used, with
Definition of the Subject various modifications, to establish ergodicity
Introduction for hyperbolic, partially hyperbolic and non-
The Volume Class uniformly hyperbolic systems.
The Fundamental Questions Hyperbolic A compact invariant set Λ
M for a
Lebesgue Measure and Local Properties of diffeomorphism f : M ! M is hyperbolic if, at
Volume every point in Λ, the tangent space splits into
Ergodicity of the Basic Examples two subspaces, one that is uniformly
Hyperbolic Systems contracted by the derivative of f, and another
Beyond Uniform Hyperbolicity that is uniformly expanded. Expanding maps
The Presence of Critical Points and Other and Anosov diffeomorphisms are examples
Singularities of globally hyperbolic maps. Hyperbolic
Future Directions diffeomorphisms and flows are the archetypi-
Bibliography cal smooth systems displaying chaotic behav-
ior, and their dynamical properties are well-
Glossary understood. Nonuniform hyperbolicity and
partial hyperbolicity are two generalizations
Conservative, dissipative Conservative dynam- of hyperbolicity that encompass a broader
ical systems (on a compact phase space) are class of systems and display many of the cha-
those that preserve a finite measure equivalent otic features of hyperbolic systems.
to volume. Hamiltonian dynamical systems are Sinai–Ruelle–Bowen (SRB) measure The con-
important examples of conservative systems. cept of SRB measure is a rigorous formulation
Systems that are not conservative are called of what it means for an invariant measure to be
dissipative. Finding physically meaningful “physically meaningful”. An SRB measure
invariant measures for dissipative maps is a attracts a large set of orbits into its support,
central object of study in smooth ergodic and its statistical features are reflected in the
theory. behavior of these attracted orbits.
Distortion estimate A key technique in smooth
ergodic theory, a distortion estimate for a
smooth map f gives a bound on the variation Definition of the Subject
of the jacobian of f n in a given region, for
n arbitrarily large. The jacobian of a smooth Smooth ergodic theory is the study of the statisti-
map at a point x is the absolute value of the cal and geometric properties of measures invariant
determinant of derivative at x, measured in a under a smooth transformation or flow. The study
fixed Riemannian metric. The jacobian of smooth ergodic theory is as old as the study of

© Springer-Verlag 2009 313

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_484
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_484
314 Smooth Ergodic Theory

abstract ergodic theory, having its origins in discussion of many of the recent developments in
Bolzmann’s Ergodic Hypothesis in the late nine- the field of smooth ergodic theory is (Bonatti
teenth Century. As a response to Boltzmann’s et al. 2005).
hypothesis, which was formulated in the context This entry assumes knowledge of the basic
of Hamiltonian Mechanics, Birkhoff and von concepts in ergodic theory and of basic differen-
Neumann defined ergodicity in the 1930s and tial topology. The texts (Cornfeld et al. 1982) and
proved their foundational ergodic theorems. The (Hirsch 1979) contain the necessary background.
study of ergodic properties of smooth systems saw
an advance in the work of Hadamard and E. Hopf
in the 1930s their study of geodesic flows for The Volume Class
negatively curved surfaces. Beginning in the
1950s, Kolmogorov, Arnold and Moser devel- For simplicity, assume that M is a compact,
oped a perturbative theory producing obstructions boundaryless C1 Riemannian manifold, and that
to ergodicity in Hamiltonian systems, known as f : M ! M is an orientation-preserving, C1 map
Kolmogorov–Arnold–Moser (KAM) Theory. satisfying m(Dx f ) > 0, for all x  M, where
Beginning in the 1960s with the work of Anosov
mðDx f Þ ¼ inf kDx f ðvÞk :
and Sinai on hyperbolic systems, the study of v  T x M, kvk¼1
smooth ergodic theory has seen intense activity.
This activity continues today, as the ergodic prop- If f is a diffeomorphism, then this assumption
erties of systems displaying weak forms of hyper- is automatically satisfied, since in that case
bolicity are further understood, and KAM theory m(Dx f ) ¼ kDf(x) f 1k1 > 0. For non-invertible
is applied in increasingly broader contexts. maps, this assumption is essential in much of the
following discussion. The Inverse Function The-
orem implies that any map f satisfying these
Introduction hypotheses is a covering map of positive degree
d  1.
This entry focuses on the basic arguments and These assumptions will avoid the issues of
principles in smooth ergodic theory, illustrating infinite measures and the behavior of f near crit-
with simple and straightforward examples. The ical points and singularities of the derivative. For
classic texts (Arnol’d and Avez 1986; Mañé most results discussed in this entry, this assump-
1987) are a good supplement. tion is not too restrictive. The existence of crit-
The discussion here sidesteps the topic of Kol- ical points and other singularities is, however, a
mogorov–Arnold–Moser (KAM) Theory, which complication that cannot be avoided in many
has played an important role in the development important applications. The ergodic-theoretic
of smooth ergodic theory. For reasons of space, analysis of such examples can be considerably
detailed discussion of several active areas in more involved, but contains many of the ele-
smooth ergodic theory is omitted, including: ments discussed in this entry. The discussion in
higher mixing properties (Kolmogorov, Bernoulli, section “Beyond Uniform Hyperbolicity” indi-
etc.), finer statistical properties (fast decay of cor- cates how some of these additional technicalities
relations, Central Limit Theorem, large devia- arise and can be overcome. For simplicity, the
tions), smooth thermodynamic formalism discussion here is confined almost exclusively to
(transfer operators, pressure, dynamical zeta func- discrete time evolution. Many, though not all, of
tions, etc.), the smooth ergodic theory of random the results mentioned here carryover to flows
dynamical systems, as well as any mention of and semi flows using, for example, a cross-
infinite invariant measures. The text (Baladi section construction (see Chap. 1 in (Mañé
2000) covers many of these topics, and the texts 1987)).
(Kifer 1986, 1988; Liu and Qian 1995) treat ran- Every smooth map f : M ! M satisfying these
dom smooth ergodic theory in depth. An excellent hypotheses preserves a natural measure class, the
Smooth Ergodic Theory 315

measure class of a finite, smooth Riemannian dissipative map. A map f is called dissipative if
volume on M. Fix such a volume n on M. Then every f-invariant measure with full support has a
there exists a continuous, positive jacobian func- singular part with respect to volume. As was just
tion x 7! jacx f on M, with the property that for seen, if a diffeomorphism f has a periodic sink,
every sufficiently small ball B
M, and every then f is dissipative; more generally, if a
measurable set A
B one has: diffeomorphism f has a periodic point p of period
k such that jacp f k 6¼ 1, then f is dissipative.
n ð f ð AÞ Þ ¼ jacx f dnðxÞ :
B
The Fundamental Questions
The jacobian of f at x is none other than the
absolute value of the determinant of the derivative For a given smooth map f : M ! M, there are the
Dx f (measured in the given Riemannian metric). following fundamental questions.
To see that the measure class of n is preserved by f,
observe that the Radon–Nikodym derivative 1. Is f conservative? That is, does there exist an
1
df  n invariant measure in the class of volume? If so,
dn ðxÞ at x is equal to y  f 1 ðxÞ jacy f > 0.
is it unique?
Hence f n is equivalent to n, and f preserves the
2. When f is conservative, what are its statistical
measure class of n.
properties? Is it ergodic, mixing, a K-system,
In many contexts, the map f has a natural
Bernoulli, etc.? Does it obey a Central Limit
invariant measure in the measure class of volume.
Theorem, fast decay of correlations, large devi-
In this case, f is said to be conservative. One
ations estimates, etc.?
setting in which a natural invariant smooth mea-
3. If f is dissipative, does there exist an invariant
sure appears is Hamiltonian dynamics. Any solu-
measure, not in the class of volume, but
tion to Hamilton’s equations preserves a smooth
(in some sense) natural with respect to volume?
volume called the Liouville measure. Further-
What are the statistical properties of such a
more, along the invariant, constant energy hyper
measure, if it exists?
manifolds of a Hamiltonian flow, the Liouville
measure decomposes smoothly into invariant
There are several plausible ways to “answer”
measures, each of which is equivalent to the
these questions. One might fix a given map f of
induced Riemannian volume. In this way, many
interest and ask these questions for that specific f.
systems of physical or geometric origin, such as
What tends to happen in the analysis of a single
billiards, geodesic flows, hard sphere gases, and
map f is that either:
evolution of the n-body problem give rise to
smooth conservative dynamical systems. See
• the question can be answered using “soft”
Dynamics of Hamiltonian Systems.
methods, and so the answer applies not only
Note that even though f preserves a smooth
to f but to perturbations of f, or even to generic
measure class, it might not preserve any measure
or typical f inside a class of maps; or,
in that measure class. Consider, for example, a
• the proof requires “hard” analysis or precise
diffeomorphism f : S1 ! S1 of the circle with
asymptotic information and cannot possibly be
exactly two fixed points, p and q, f 0( p) > 1 > f 0
answered for a specific f, but can be answered
(q) > 0. Let m be an f-invariant probability
for a large set of ft in a typical (or given)
measure. Let I be a neighborhood of p.
1 parametrized family {ft}t  (1, 1) of smooth
Then \ f n ðI Þ ¼ fpg, but on the other hand, maps containing f ¼ f0.
n¼1
m( fn(I )) ¼ m(I) > 0, for all n. This implies that
m({p}) > 0, and so m does not lie in the measure Both types of results appear in the discussion
class of volume. This is an example of a that follows.
316 Smooth Ergodic Theory

Lebesgue Measure and Local Properties of S1 ∖ A by pairwise disjoint elements {Ji} of the
of Volume union [1n¼1 P n with the properties:

Locally, any measure in the measure class of

lðJ 1 Þ  lðJ 2 Þ     , and
volume is, after a smooth change of coordinates,
equivalent to Lebesgue measure in ℝn. In fact, 1
1
more is true: Moser’s Theorem implies that l [ Ji ¼ lð J i Þ < 1 :
locally any Riemannian volume is, after a smooth i¼1
i¼1
change of coordinates, equal to Lebesgue mea-
sure in Rn. Hence to study many of the local For n  ℕ, let Un be the union of all the
properties of volume, it suffices to study the intervals Ji that are contained in P n , and let
same properties for Lebesgue measure. V n ¼ [ni¼1 U n . This defines an increasing
One of the basic properties of Lebesgue mea- sequence of natural numbers i1 ¼ 1 < i2 < i3 <   
inþ1 1 nþ1 1
sure is that every set of positive Lebesgue measure such that U n ¼ [i¼i n
J n and V n ¼ [ii¼1 Jn.
can be approximated arbitrarily well in measure For each n, the interval In will be chosen to be an
from the outside by an open set, and from the element P n , disjoint from Vn, in which the density
inside by a compact set. A consequence of this of [1 i¼nþ1 U i is very small (approaching 0 as
property, of fundamental importance in smooth n ! 1). Since (S1 ∖ A) \ In is contained in
ergodic theory, is the following statement. [1i¼nþ1 U i , this choice of In will ensure that the
Fundamental Principle # 1: Two disjoint, density of A in In is large (approaching 1 as n ! 1).
positive Lebesgue measure sets cannot mix To make this choice of In, note first that the
together uniformly at all scales. density of [1 i¼nþ1 U i inside of S ∖ Vn is:
1

As an illustration of this principle, consider the

following elementary exercise in measure theory. 1
l [1
i¼nþ1 U i i¼inþ1 lðJ i Þ
First, some notation. If n is a measure and A and ¼ ¼ an :
l S1 ∖V n 1 inþ1 1
i¼1 lðJ i Þ
B are n-measurable sets with n(B) > 0, the density
of A in B is defined by:
Note that, since 1 i¼1 lðJ i Þ1, one has an ! 0 as
n ! 1. Since the density of [1 i¼nþ1 U i inside of
n ð A \ BÞ
n ð A : BÞ ¼ : S ∖ Vn is at most an, there is an interval In in P n ,
1
n ð BÞ
disjoint from Vn, such that the density of [1 i¼nþ1 U i
inside of In is at most an. Then
Proposition 1 Let P 1 , P 2 , . . . be sequence of
(mod 0) finite partitions of the circle S1 into open
lim lðA : I n Þ  lim 1  an ¼ 1 :
intervals, with the properties: (a) any element of n!1 n!1
P n is a (mod 0) union of elements of P nþ1 , □
and (b) the maximum diameter of elements of P n
tends to 0 as n ! 1. In smooth ergodic theory, it is often useful to
use a variation on Proposition 1 (generally, in
higher dimensions) in which the partitions P n are
Let A be any set of positive Lebesgue measure nested, dynamically-defined partitions. A simple
in S1. Then there exists a sequence of intervals I1, application of this method can be used to prove
I2, . . ., with I n  P n such that: that the doubling map on the circle is ergodic with
respect to Lebesgue measure, which is done in
lim lðA : I n Þ ¼ 1 : section “Lebesgue Measure and Local Properties
n!1
of Volume”.
Proof Assume that Lebesgue measure has been Notice that this proposition does not claim that
normalized so that l(S1) ¼ 1. Fix a (mod 0) cover the intervals In are nested. If one imposes stronger
Smooth Ergodic Theory 317

conditions on the partitions P n , then one can draw implies that there exists an interval I
S1 such
stronger conclusions. that the density of A in I is large: l(A : I ) > 1  ε.
A very useful theorem in this respect is the Similarly, one may choose an interval J such that
Lebesgue Density Theorem. A point x  M is l(Ac : J) > 1  ε. Without loss of generality, one
a Lebesgue density point of a measurable set may choose I and J to have the same length. Since
X  M if a is irrational, Rα has a dense orbit, which meets
the interval I. Since Rα is an isometry, this implies
lim mðX : Br ðxÞÞ ¼ 1 , that there is an integer n such that
r!0
l Rna ðI Þ D J < elðI Þ. Since l(I) ¼ l(J ), this read-
where Br(x) is the Riemannian ball of radius ily implies that j l A : Rna ðI Þ  lðA : J Þ j< e.
r centered at x. Notice that the notion of Lebesgue Also, since A is invariant, and Rα is invertible
density point depends only on the smooth struc- and preserves measure, one has:
ture of M, because any two Riemannian metrics
have the same Lebesgue density points. The l A : Rna ðI Þ ¼ l Rna ðAÞ : Rna ðI Þ ¼ lðA : I Þ
Lebesgue Density Theorem states that if A is a >1e :
measurable set and A is the set of Lebesgue den-
sity points of A, then m A D A ¼ 0. But for ε sufficiently small, this contradicts the
facts that l(A : J) ¼ 1  l(Ac : J ) < ε and
j l A : Rna ðI Þ  lðA : J Þ j< e. □
Ergodicity of the Basic Examples Note that this is not a proof of the strongest
possible statement about Rα (namely, minimality
This section contains proofs of the ergodicity of and unique ergodicity). The point here is to show
two basic examples of conservative smooth maps: how “soft” arguments are often sufficient to estab-
irrational rotations on the circle and the doubling lish ergodicity; this proof uses no more about Rα
map on the circle. See ▶ Ergodic Theory: Basic than the fact that it is a transitive isometry. Hence
Examples and Constructions for a more detailed the same argument shows:
description of these maps. These proofs serve as
an elementary illustration of some of the funda- Theorem 1 Let f : M ! M be a transitive isom-
mental techniques and principles in smooth ergo- etry of a Riemannian manifold M. Then f is ergo-
dic theory. dic with respect to Riemannian volume.
Rotations on the circle. Denote by S1 the
circle ℝ/ℤ, which is an additive group, and by l One can isolate from this proof a useful
normalized Lebesgue-Haar measure on S1. Fix a principle:
real number α  ℝ. The rotation Rα : S1 ! S1 is Fundamental Principle # 2: Isometries pre-
the translation defined by Rα(x) ¼ x þ α. Since serve Lebesgue density at all scales, for arbitrarily
translations preserve Lebesgue-Haar measure, the many iterates.
map Rα is conservative. Note that Rα is a This principle implies, for example, that a
diffeomorphism and an isometry with respect to smooth action by a compact Lie group on M is
the canonical flat metric (length) on S1. ergodic along typical (nonsingular) orbits. This
principle is also useful in studying area-preserving
Proposition 2 If α ℚ, then the rotation flows on surfaces and, in a refined form, unipotent
Rα : S1 ! S1 is ergodic with respect to Lebesgue flows on homogeneous spaces. In the case of
measure. surface flows, ergodicity questions can be reduced
to a study of interval exchange transformations.
Proof Let A be an Rα-invariant set in S1, and See the entry ▶ Ergodic Theory: Basic Examples
suppose that 0 < l(A) < 1. Denote by Ac the and Constructions for a detailed discussion of
complement of Ain S1. Fix ε > 0. Proposition 1 interval exchange transformations and flows on
318 Smooth Ergodic Theory

surfaces. ▶ Introduction to Ergodic Theory con- In this proof, the facts that the intervals in
tains detailed information on unipotent flows. P n have constant length 2n and that the jaco-
Doubling map on the circle. Let T2 : S1 ! S1 bian of T2n restricted to such an interval is
be the doubling map defined by T2(x) ¼ 2x. Then constant and equal to 2n are not essential.
T2 is a degree-2 covering map and endomorphism The key fact really used in this proof is the
of S1 with constant jacobian jacxT2  2. Since assertion that the ratio:
dðT 2 Þ l
dl ¼ 12 þ 12 ¼ 1, T2 preserves Lebesgue-Haar
measure. The doubling map is the simplest exam- l T n2 ðA \ J Þ : T n2 ðJ Þ
ple of a hyperbolic dynamical system, a topic lð A : J Þ
treated in depth in the next section.
As with the previous example, the focus here is is bounded, independently of n. In this case, the
on the property of ergodicity. It is again possible ratio is 1 for all n because T2 has constant
to prove much stronger results about T2, such as jacobian.
Bernoulli city, by other methods. Instead, here is a It is tempting to try to extend this proof to
soft proof of ergodicity that will generalize readily other expanding maps on the circle, for example,
to other contexts. a C1, l-preserving map f : S1 ! S1 with
dC1 ð f , T 2 , Þ small. Many of the aspects of this
Proposition 3 The doubling map T2 : S1 ! S1 is proof carry through mutatismutandis for such an
ergodic with respect to Lebesgue measure. f, save for one. A C1-small perturbation of T2 will
in general no longer have constant jacobian, and
Proof Let A be a T2-invariant set in S1 with the variation of the jacobian of f n on a small
l(A) > 0. Let p  S1 be the fixed point of T2, so interval can be (and often is) unbounded. The
that T2( p) ¼ p. For each n  ℕ, the preimages of reason for this unboundedness is a lack of control
p under T n
2 define a (mod 0) partition P n into 2
n
of the modulus of continuity of f ’. Hence this
open intervals of length 2 ; the elements of P n are
-n
argument can fail for C1 perturbations of T2. On
the connected components of S1 ∖T n2 ðfpgÞ. Note the other hand, the argument still works for C2
that the sequence of partitions P 1 , P 2 , . . . is perturbations of T2, even when the jacobian is not
nested, in the sense of Proposition 1. Restricted constant.
to any interval J  P n , the mapT2n is a The principle behind this fact can be loosely
diffeomorphism onto S1 ∖ {p} with constant jaco- summarized:
0
bian jacx T n2 ¼ T n2 ðxÞ ¼ 2n . Fundamental Principle # 3: On controlled
scales, iterates of C2 expanding maps distort
Since A is invariant, it follows that Lebesgue density in a controlled way.
T n
2 ðAÞ ¼ A. Fix ε > 0. Proposition 1 implies This principle requires further explanation and
that there exists an n  ℕ and an interval J  P n justification, which will come in the following
such that l(A : J ) > 1  ε. Note that section. The C2 hypothesis in this principle
T n2 ðA \ J Þ
A. But then accounts for the fact that almost all results in
smooth ergodic theory assume a C2 hypothesis
lðAÞ  l T n2 ðA \ J Þ (or something slightly weaker).

¼ jacx T n2 dlðxÞ
A\ J Hyperbolic Systems
¼ 2n lð A \ J Þ
¼ 2n lðA : J ÞlðJ Þ One of the most developed areas of smooth ergo-
> 2n ð1  eÞlðJ Þ ¼ 1  e : dic theory is in the study of hyperbolic maps and
attractors. This section defines hyperbolic maps
Since ε was arbitrary, one obtains that and attractors, provides examples, and investi-
l(A) ¼ 1. □ gates their ergodic properties. See (Robinson
Smooth Ergodic Theory 319

1995; Katok and Hasselblatt 1995) and Hyper- kDx f n ðvÞk  Cmn kvk ,
bolic Dynamical Systems for a thorough discus-
sion of the topological and smooth properties of with respect to some (any) Riemannian metric on
hyperbolic systems. M. An expanding map is clearly hyperbolic, with
A hyperbolic structure on a compact U ¼ M, Es the trivial bundle, and Eu ¼ TM. Any
f-invariant set Λ
M is given by a Df-invariant disk in M is a local unstable manifold for f.
splitting TΛM ¼ Eu Es of the tangent bundle
over Λ and constants C, m > 1 such that, for every Anosov Diffeomorphisms
x  Λ and n  ℕ: A diffeomorphism f : M ! M is called Anosov if
the tangent bundle splits as a direct sum TM ¼
v  Eu ðxÞ ) kDx f n ðvÞk  C1 mn kvk , Eu Es of two Df-invariant subbundles, such that
Eu is uniformly expanded and Es is uniformly
and v  Es ðxÞ ) kDx f n ðvÞk Cmn kvk :
contracted by Df. Similarly, a flow ’t : M ! M
is called Anosov if the tangent bundle splits as
A hyperbolic attractor for a map f : M ! M is
a direct sum TM ¼ Eu E0 Es of three
given by an open set U
M such that: f ðU Þ
U,
D’t-invariant subbundles, such that E0 is gener-
and such that the set Λ ¼ \n  0 f n(U ) carries a
ated by ’,_ Eu is uniformly expanded and Es is
hyperbolic structure. The set Λ is called the
uniformly contracted by D’t. Like expanding
attractor, and U is an attracting region. A map
maps, an Anosov diffeomorphism is an Anosov
f : M ! M is hyperbolic if M decomposes (mod 0)
attractor with Λ ¼ U ¼ M.
into a finite union of attracting regions for hyper-
A simple example of a conservative Anosov
bolic attractors. Typically one assumes as well
diffeomorphism is a hyperbolic linear automor-
that the restriction of f to each attractor Λi is
phism of the torus. Any matrix A  SL(n, ℤ)
topologically transitive.
induces an automorphism of Rn preserving the
Every point p in a hyperbolic set Λ has smooth
integer lattice Zn, and so descends to an automor-
stable manifold W s ðpÞ and unstable manifold
phism fA : T n ! T n of the n-torus T n ¼ ℝn/ℤn.
W u ðpÞ, tangent, respectively, to the subspaces
Since the determinant of A is 1, the
Es( p) and Eu( p). The set W s ðpÞ is precisely the
diffeomorphism fA preserves Lebesgue-Haar mea-
set of q  M such that d( f n( p), f n(q)) tends to
sure on T n. In the case where none of the eigen-
0 as n ! 1, and it follows that f ðW sðpÞÞ ¼
values of A have modulus 1, the resulting
W s ð f ðpÞÞ. When f is a diffeomorphism, the unsta-
diffeomorphism fA is Anosov. The stable bundle
ble manifold W u ðpÞ is uniquely defined and is the
Es at x  T n is the parallel translate to x of the sum
stable manifold of f 1. When f is not invertible,
of the contracted generalized eigenspaces of A,
local unstable manifolds exist, but generally are
and the unstable bundle Eu at x is the translated
not unique. If Λ is a transitive hyperbolic attractor,
sum of expanded eigenspaces.
then every unstable manifold of every point
In general, the invariant subbundles Eu and Es
p  Λ is dense in Λ.
of an Anosov diffeomorphism are integrable and
tangent to a transverse pair of foliations W u and
Examples of Hyperbolic Maps and Attractors W s , respectively (see, e.g (Hirsch et al. 1977). for
a proof of this). The leaves of W s are uniformly
Expanding Maps contracted by f, and the leaves of W u are uni-
The previous section mentioned briefly the Cr formly contracted by f 1. The leaves of these
perturbations of the doubling map T2. Such per- foliations are as smooth as f, but the tangent bun-
turbations (as well as T2 itself) are examples of dles to the leaves do not vary smoothly in the
expanding maps. A map f : M ! M is expanding if manifold. The regularity properties of these folia-
there exist constants m > 1 and C > 0 such that, for tions play an important role in the ergodic prop-
every x  M, and every nonzero vector v  TxM : erties of Anosov diffeomorphisms.
320 Smooth Ergodic Theory

The first Anosov flows to be studied exten- the assertion in Fundamental Principle # 3 in this
sively were the geodesic flows for manifolds of context.
negative sectional curvatures. As these flows are
Hamiltonian, they are conservative. Eberhard Lemma There exists a constant K  1 such that,
Hopf showed in the 1930s that such geodesic for all n  ℕ, and for all x, y  fn(I), one has:
flows for surfaces are ergodic with respect to
Liouville measure (Hopf 1939); it was not until ðf n Þ0 ðxÞ
the 1960s that ergodicity of all such flows was K 1 K:
ðf n Þ0 ðy Þ
proved by Anosov (Anosov 1967). The next sec-
tion describes, in the context of Anosov Proof Since f is C2 and f’ is bounded away from
diffeomorphisms, Hopf’s method and important 0, the function α(x) ¼ log ( f 0(x)) is C1. In partic-
refinements due to Anosov and Sinai. ular, a is Lipschitz continuous: there exists a
constant L > 0 such that, for all x, y  S1, jα(x) 
DA Attractors α( y) j < Ld(x, y). For n  0, let αn(x) ¼ log (( f n)0(x)).
A simple way to produce a non-Anosov hyper- The Chain Rule implies that
bolic attractor on the torus is to start with an an ðxÞ ¼ n1 a f i
ðx Þ .
i¼0
Anosov diffeomorphism, such as a linear hyper-
bolic automorphism, and deform it in a neighbor-
The expanding hypothesis on f implies that
hood of a fixed point, turning a saddle fixed point
for all x, y  fn(I ) and for i ¼ 0, . . ., n, one has
into a source, while preserving the stable foliation.
d( f i(x), f i( y)) C1mind( f n(x), f n( y))
If this procedure is carried out carefully enough, 1 i  n
C m . Hence
the resulting diffeomorphism is a dissipative
hyperbolic diffeomorphism, called a derived n1
from Anosov (DA) attractor. Other examples of j an ð x Þ  an ð y Þ j j a f i ðxÞ  a f i ðyÞ j
hyperbolic attractors are the Plykin attractor and i¼0
the solenoid. See (Robinson 1995). n1
L d f i ðxÞ, f i ðyÞ
i¼0
Distortion Estimates
n1
Before describing the ergodic properties of hyper-
L C1 min
bolic systems, it is useful to pause for a brief i¼0
discussion of distortion estimates. Distortion esti- 1
< LC 1 m1 1  m1 :
mates are behind almost every result in smooth
ergodic theory. In the hyperbolic setting, distor-
Setting K ¼ exp (LC1m1(1  m1)1), one
tion estimates are applied to the action of f on
now sees that ( f n)0(x)/( f n)0( y) lies in the interval
unstable manifolds to show that the volume dis-
[K1. K], proving the claim. □
tortion of f along unstable manifolds can be con-
In this distortion estimate, the function
trolled for arbitrarily many iterates.
α : M ! ℝ is called a cocycle. The same argument
The example mentioned at the end of the pre-
applies to any Lipschitz continuous (or even
vious section illustrates the ideas in a distortion
Hölder continuous) cocycle.
estimate. Suppose that f : S1 ! S1 is a C2
expanding map, such as a C2 small perturbation
Ergodicity of Expanding Maps
of T2. Then there exist constants m > 1 and C > 0
The ergodic properties of C2 expanding maps are
such that ( f n)0(x) > Cmn for all x and n.
completely understood. In particular, every con-
Let d be the degree of f. If I is a sufficiently
servative expanding map is ergodic, and every
small open interval in S1, then for each n, fn(I) is
expanding map is conservative. The proofs of
a union of d disjoint intervals. Furthermore, each
these facts use Fundamental Principles # 1 and
of these intervals has diameter at most C1mn
3 in a fairly direct way.
times the diameter of I. It is now possible to justify
Smooth Ergodic Theory 321

Every C2 conservative expanding map is ergo- f-invariant measure absolutely continuous with
dic with respect to volume. The proof is a straight- respect to l. With more work, one can show that
forward adaptation of the proof of Proposition 3 n is exact. See (Mañé 1987) for details.
(see, e.g (Mañé 1987)). Here is a description of the
proof for M ¼ S1. As remarked earlier, the proof of Ergodicity of Conservative Anosov
Proposition 3 adapts easily to a general expanding Diffeomorphisms
map f : S1 ! S1 once one shows that for every Like conservative C2 expanding maps, conserva-
f-invariant set A, and every connected component tive C2 Anosov diffeomorphisms are ergodic.
J of fn(S1 ∖ {p}), the quantity This subsection outlines a proof of this fact.
Unlike expanding maps, however, Anosov
lð f n ð A \ J Þ : f n ð J Þ Þ diffeomorphisms need not be conservative. The
lð A : J Þ subsection following this one describe a type of
invariant measure that is “natural” with respect to
is bounded independently of n. This is a fairly volume, called a Sinai-Ruelle-Bowen (or SRB)
direct consequence of the distortion estimate in measure. The central result for hyperbolic systems
Lemma 6.1 and is left as an exercise. states that every hyperbolic attractor carries an
The same distortion estimates show that every SRB measure.
2
C expanding map is conservative, preserving a
probability measure v in the measure class of The Hopf Argument
volume. Here is a sketch of the proof for the case In the 1930s Hopf (Hopf 1939) proved that the
M ¼ S1. To prove this, consider the push-forward geodesic flow for a compact, negatively-curved
ln ¼ f n l. Then ln is equivalent to Lebesgue, and surface is ergodic. His method was to study the
its Radon–Nikodym derivative dln ∖ dl is the Birkhoff averages of continuous functions along
density function leaves of the stable and unstable foliations of the
flow. This type of argument has been used since
1 then in increasingly general contexts, and has
rn ðxÞ ¼ :
jacy f n come to be known as the Hopf Argument.
y  f n ðxÞ
The core of the Hopf Argument is very simple.
Since f n l is a probability measure, it follows To any f : M ! M one can associate the stable
that S1 rn dl ¼ 1. A simple argument using the equivalence relation ~s, where x~s y iff limn!1
distortion estimate above (and summing up over d( f n(x), f n( y)) ¼ 0. Denote by W s(x) the stable
all dn branches of f-n at x) shows that there exists a equivalence class containing x. When f is invert-
constant c  1 such that for all x, y  S1, ible, one defines the unstable equivalence relation
to be the stable equivalence relation for f 1, and
rn ðxÞ one denotes by W u(x) the unstable equivalence
c1 c: class containing x.
rn ð y Þ
The first step in the Hopf Argument is to show
Since the integral of rn is 1, the functions rn are that Birkhoff averages for continuous functions
uniformly bounded away from 0 and 8. It is easy are constant along stable and unstable equivalence
to see that the measure nn ¼ 1n ni¼1 f i l has density classes. Let f : M ! ℝ be an integrable function,
1 n and let
n i¼1 ri . Let n be any subsequential weak* limit
of nn; then n is absolutely continuous, with density n
r bounded away from 0 and 8. With a little more 1
f ¼ lim sup f fi : ð1Þ
care, one can show that r is actually Lipschitz n!1 n i¼1
continuous.
As a passing comment, the ergodicity of n and Observe that if f is continuous, then for every
positivity of r imply that n is the unique x  M and x0  Ws(x), limn ! 1 j f( f n(x)) 
322 Smooth Ergodic Theory

f( f i(x0)) j ¼ 0. It follows immediately that be C1 foliations (a hypotheses satisfied in the

ff ðxÞ ¼ ff ðx0 Þ. In particular, if the limit in examples he considered, due to low-
(1) exists at x, then it exists and is constant on dimensionality. See also (Sinai 1961), where a
Ws(x). pinching condition on the curvature, rather than
Fundamental Principle # 4: Birkhoff aver- low dimensionality, implies this C1 condition on
ages of continuous functions are constant along the foliations.)
stable equivalence classes.
The next step of Hopf’s argument confines Absolute Continuity
itself to the situation where f is conservative and For a general Anosov diffeomorphism or flow, the
Anosov. In this case, f is invertible, the stable stable and unstable foliations are not C1, and so
equivalence classes are precisely the leaves of the final step in Hopf’s original argument does not
the stable foliation W s , and the unstable equiva- apply. The fundamental advance of Anosov and
lence classes are the leaves of the unstable folia- Anosov-Sinai was to prove that the stable and
tion W u . Since f is conservative, the Ergodic unstable foliations of an Anosov diffeomorphism
Theorem implies that for every L2 function f, (conservative or not) satisfy a weaker condition
the function ff is equal (mod 0) to the projection than smoothness, called absolute continuity. For
of f onto the f-invariant functions in L2. Since this conservative systems, absolute continuity is
projection is continuous, and the continuous func- enough to finish Hopf’s argument, proving that
tions are dense in L2, to prove that f is ergodic, it every C2 conservative Anosov diffeomorphism is
suffices to show that the projection of any contin- ergodic (Anosov 1967; Anosov and Sinai 1967).
uous function is trivial. That is, it suffices to show For a definition and careful discussion of abso-
that for every continuous f, the function ff is lute continuity of a foliation F , see (Brin and
constant (a.e.). Stuck 2002). Two consequences of the absolute
To this end, let f : M ! ℝ be continuous. Since continuity of F are:
the f-invariant functions coincide with the f1-
invariant functions, one obtains that ff ¼ ff 1 1. (AC1) If A
M is any measurable set, then
a.e. The previous argument shows ff is con-
stant along W s -leaves and ff 1 is constant along lðAÞ ¼ 0 , lF ðxÞ ðAÞ ¼ 0 ,
W u -leaves. The desired conclusion is that ff is for l  a:e:x  M ,
a.e. constant. It suffices to show this in a local
chart, since the manifold M is connected. In a local where lF (x) denotes the induced Riemannian
chart, after a smooth change of coordinates, one volume on the leaf of F through x.
obtains a pair of transverse foliations F 1, F 2 of 2. (AC2) If t is any small, smooth disk transverse
the cube [1, 1]n by disks, and a measurable to a local leaf of F , and T
t is a 0-set in t
function c : [1, 1]n ! ℝ that is constant along (with respect to the induced Riemannian vol-
the leaves of F 1 and constant along the leaves of ume on t), then the union of the F leaves
F 2. through points in T has Lebesgue measure
When the foliations F 1 and F 2 are smooth 0 in M.
(at least C1), one can perform a further smooth
change of coordinates so that F 1 and F 2 are The proof that W s and W u are absolutely
transverse coordinate subspace foliations. In this continuous has a similar flavor to the proof that
case, Fubini’s theorem implies that any measur- an expanding map has a unique absolutely con-
able function that is constant along two transverse tinuous invariant measure (although the cocycles
coordinate foliations is a.e. constant. This com- involved are Hölder continuous, rather than
pletes the proof in the case that the foliations W s Lipschitz), and the facts are intimately related.
and W u are smooth. In Hopf’s original argument, With absolute continuity of the stable and
the stable and unstable foliations were assumed to unstable foliations in hand, one can now prove:
Smooth Ergodic Theory 323

Theorem 2 (Anosov) Let f be a C2, conservative physically observable measures for hyperbolic
Anosov diffeomorphism. Then f is ergodic. attractors (Sinai 1972; Ruelle 1976; Bowen
1975). Such measures are now known as Sinai-
Proof By the Hopf Argument, it suffices to show Ruelle-Bowen (SRB) measures, and have been
that if cs and cu are L2 functions with the follow- shown to exist for non-hyperbolic maps with
ing properties: some hyperbolic features. This subsection
describes the construction of SRB measures for
1. cs is constant along leaves of W s , hyperbolic attractors.
2. cu is constant along leaves of W u , and An f-invariant probability measure m is called
3. cs ¼ cu a.e., an SRB (or physical) measure if there exists an
open set U
M containing the support of m such
then cs (and so cu as well) is constant a.e. that, for every continuous function f : M ! ℝ and
This is proved using the absolute continuity of l-a.e. x  U,
W u and W s . Since M is connected, one may argue
this locally. Let G be the full measure set of p  M 1
n

such that cs ¼ cu. Absolute continuity of W s lim f f i ðxÞ ¼ f dm :

n!1 n
i¼1 M
(more precisely, consequence (AC1) of absolute
continuity described above) implies that for The maximal open set U with this property is
almost every p  M, G has full measure in called the basin of f. To exclude the possibility
W s ðpÞ. Pick such a p. Then for almost every that the SRB measure is supported on a periodic
q  W s ðpÞ, cs(q) ¼ cu( p); defining G’ to be the sink, one often adds the condition that at least one
union over all q  W s ðpÞ \ G of W u ðqÞ, one of the Lyapunov exponents of f with respect m is
obtains that cs is constant on G \ G0. But now, positive. Other definitions of SRB measure have
since W s ðpÞ \ G has full measure in W s ðpÞ, the been proposed (see (Young 2002)). Note that
absolute continuity of W u (consequence (AC2) every ergodic absolutely continuous invariant
above) implies that G’ has full measure in a neigh- measure with positive density in an open set is
borhood of p. Hence cs is a.e. constant in a neigh- an SRB measure. Note also that an SRB measure
borhood of p, completing the proof. □ for f is not in general an SRB measure for f 1,
unless f preserves an ergodic absolutely continu-
SRB Measures ous invariant measure.
In the absence of a smooth invariant measure, it is Every transitive Anosov diffeomorphism
still possible for a map to have an invariant mea- carries a unique SRB measure. To prove this,
sure that behaves naturally with respect to vol- one defines a sequence of probability measures
ume. In computer simulations one observes such nn on M as follows. Fix a point p  M, and define
measures when one picks a point x at random and n0 to be the normalized restriction of Riemannian
plots many iterates of x; in many systems, the volume to a ball Bu in W u ðpÞ. Set
resulting picture is surprisingly insensitive to the nn ¼ 1n ni¼1 f i n0 . Distortion estimates show that
initial choice of x. What appears to be happening the density of nn on its support inside W u is
in these systems is that the trajectory of almost bounded, above and below, independently of n.
every x in an open set U is converging to the Passing to a subsequential weak* limit, one
support of a singular invariant probability mea- obtains a probability measure n on M with
sure m. Furthermore, for any open set V, the pro- bounded densities on W u -leaves.
portion of forward iterates of x spent in V appears To show that n is an SRB measure, choose a
to converge to m(V ) as the number of iterates tends point q  M in the support of n. Since n has
to 8. positive density on unstable manifolds, almost
In the 1960s and 70s, Sinai, Ruelle and Bowen every point in a neighborhood of q in W u ðqÞ is a
rigorously established the existence of these regular point for f (that is, a point where the
324 Smooth Ergodic Theory

forward Birkhoff averages of every continuous Partial Hyperbolicity

function exist). A variation on the Hopf Argu- Brin and Pesin (1974) and independently Pugh
ment, using the absolute continuity of W s , and Shub (1972) first examined the ergodic prop-
shows that n is an ergodic SRB measure. erties of partially hyperbolic systems soon after
A similar argument shows that every transitive the work of Anosov and Sinai on hyperbolic sys-
hyperbolic attractor admits an ergodic SRB mea- tems. One says that a diffeomorphism f : M ! M
sure. In fact this SRB measure has much stronger of a compact manifold M is partially hyperbolic if
mixing properties, namely, it is Bernoulli. To there is a nontrivial, continuous splitting of the
prove this, one first constructs a Markov partition tangent bundle, TM ¼ Es Ec Eu, invariant
(Bowen 1970) conjugating the action of f to a under Df, such that E is uniformly contracted, Eu
s

Bernoulli shift. This map sends the SRB measure is uniformly expanded, and Ec is dominated,
to a Gibbs state for a mixing Markov shift (see meaning that for some n  1 and for all x  M:
▶ Pressure and Equilibrium States in Ergodic
Theory). A result that subsumes all of the results kDx f n jEs k < m Dx f n jEc kDx f n jEc k < mðDx f n jEu Þ :
in this section is:
Partial hyperbolicity is a C1-open condition:
Theorem 3 (Sinai, Ruelle, Bowen) Let Λ
M any diffeomorphism sufficiently C1-close to a
be a transitive hyperbolic attractor for a C2 map partially hyperbolic diffeomorphism is itself par-
f : M ! M. Then f has an ergodic SRB measure m tially hyperbolic. For an extensive discussion of
supported on Λ. Moreover: the disintegration of m examples of partially hyperbolic dynamical sys-
along unstable manifolds of Λ is equivalent to the tems, see the survey article (Burns et al. 2001) and
induced Riemannian volume, the Lyapunov expo- the book (Pesin 2004). Among these examples
nents of m are all positive, and m is Bernoulli. are: the time-1 map of an Anosov flow, the
frame flow for a compact manifold of negative
sectional curvature, and many affine transforma-
Beyond Uniform Hyperbolicity tions of compact homogeneous spaces. All of
these examples preserve the volume induced by
The methods developed in the smooth ergodic a Riemannian metric on M.
theory of hyperbolic maps have been extended As in the Anosov case, the stable and unstable
beyond the hyperbolic context. Two natural gen- bundles Es and Eu of a partially hyperbolic
eralizations of hyperbolicity are: diffeomorphism are tangent to foliations, again
denoted by W s and W u respectively (Brin and
• partial hyperbolicity, which requires uniform Pesin 1974). Brin-Pesin and Pugh-Shub proved
expansion of Eu and uniform contraction of Es, that these foliations are absolutely continuous.
but allows central directions at each point, in A partially hyperbolic diffeomorphism
which the expansion and contraction is domi- f : M ! M is accessible if any point in M can be
nated by the behavior in the hyperbolic direc- reached from any other along an su-path, which is
tions; and, a concatenation of finitely many subpaths, each of
• nonuniform hyperbolicity, which requires which lies entirely in a single leaf of W s or a
hyperbolicity along almost every orbit, but single leaf of W u . Accessibility is a global, topo-
allows the expansion of Eu and the contraction logical property of the foliations W u and W s that
of Es to weaken near the exceptional set where is the analogue of transversality of W u and W s for
there is no hyperbolicity. Anosov diffeomorphisms. In fact, the trans-
versality of these foliations in the Anosov case
This section discusses both generalizations. immediately implies that every Anosov
Smooth Ergodic Theory 325

diffeomorphism is accessible. Fundamental Prin- hyperbolic diffeomorphisms, but the theory is

ciple # 4 suggests that accessibility might be less developed than in the conservative case.
related to ergodicity for conservative systems. Using the same construction as for Anosov
diffeomorphisms, one can construct invariant
Conservative Partially Hyperbolic probability measures that are smooth along the
Diffeomorphisms W u foliation (Pesin and Sinai 1983). Such mea-
Motivated by a breakthrough result with Grayson sures are referred to as u-Gibbs measures. Since
(Grayson et al. 1994), Pugh and Shub conjectured the stable bundle Es is not transverse to the unsta-
that accessibility implies ergodicity, for a C2, par- ble bundle Eu, the Anosov argument cannot be
tially hyperbolic conservative diffeomorphism carried through to show that u-Gibbs measures are
(Pugh and Shub 1996). This conjecture has been SRB measures.
proved under the hypothesis of center bunching Nonetheless, there are conditions that imply
(Burns and Wilkinson n.d.), which is a mild spec- that a u-Gibbs measure is an SRB measure: for
tral condition on the restriction of Df to the center example, a u-Gibbs measure is SRB if it is the
bundle Ec. Center bunching is satisfied by most unique u-Gibbs measure (Dolgopyat 2004), if
examples of interest, including all partially hyper- the bundle Es Ec is nonuniformly contracted
bolic diffeomorphisms with dim(Ec) ¼ 1. The (Bonatti and Viana 2000), or if the bundle Eu Ec
proof in (Burns and Wilkinson n.d.) is a modifi- is nonuniformly expanded (Alves et al. 2000).
cation of the Hopf Argument using Lebesgue The proofs of the latter two results use Pesin
density points and a delicate analysis of the geo- Theory, which is explained in the next
metric and measure-theoretic properties of the subsection.
stable and unstable foliations. SRB measures have also been constructed in
In the same article, Pugh and Shub also systems where Ec is nonuniformly hyperbolic
conjectured that accessibility is a widespread phe- (Burns et al. n.d.), and in (noninvertible) partially
nomenon, holding for an open and dense set hyperbolic covering maps where Ec is
(in the Cr topology) of partially hyperbolic 1-dimensional (Tsujii 2005). It is not known
diffeomorphisms. This conjecture has been pro- whether accessibility plays a role in the existence
ved completely for r ¼ 1 (Dolgopyat and of SRB measures for dissipative, non-Anosov
Wilkinson 2003), and for all r, with the additional partially hyperbolic diffeomorphisms.
assumption that the central bundle Ec is one
dimensional (Rodríguez et al. 2008). Nonuniform Hyperbolicity
Together, these two conjectures imply the The concept of Lyapunov exponents gives a nat-
third, central conjecture: in (Pugh and Shub ural way to extend the notion of hyperbolicity to
1996): systems that behave hyperbolically, but in a non-
uniform manner. The fundamental principles
Conjecture 1 (Pugh–Shub) For any r  2, the described above, suitably modified, apply to
C r, conservative partially hyperbolic diffeo- these nonuniformly hyperbolic systems and
morphisms contain a C r open and dense set of allow for the development of a smooth ergodic
ergodic diffeomorphisms. theory for these systems. This program was ini-
tially proposed and carried out by Yakov Pesin in
The validity of this conjecture in the absence of the 1970s (Pesin 1977) and has come to be known
center bunching is currently an open question. as Pesin theory.
Oseledec’s Theorem implies that if a smooth
Dissipative Partially Hyperbolic map f satisfying the condition m(Dx f ) > 0 pre-
Diffeomorphisms serves a probability measure v, then for v-a.e.
There has been some progress in constructing x  M and every nonzero vector v  TxM, the
SRB-type measures for dissipative partially limit
326 Smooth Ergodic Theory

n has arbitrarily small measure. The precise formu-

1
lðx, vÞ ¼ lim log Dx f i ðvÞ lation of this statement is omitted, but here are
n!1 n
i¼1
some of its salient features.
If m is a hyperbolic measure for a C2
exists. The number l(x, v) is called the Lyapunov
diffeomorphism, then attached to m-a.e. point
exponent at x in the direction of v. For each such x,
x  M are transverse, smooth stable and unstable
there are finitely many possible values for the
manifolds for f. The collection of all stable mani-
exponent l(x, v), and the function x 7! l(x, ) is
folds is called the stable lamination for f, and the
measurable. See the discussion of Oseledec’s The-
collection of all unstable manifolds is called the
orem in ▶ Ergodic Theorems.
unstable lamination for f. The stable lamination is
Let f be a smooth map. An f-invariant proba-
invariant under f, meaning that f sends the stable
bility measure m is hyperbolic if the Lyapunov
manifold at x into the stable manifold for f(x). The
exponents of m-a.e. point are all nonzero. Observe
stable manifold through x is contracted uniformly
that any invariant measure of a hyperbolic map is
by all positive iterates of f in a neighborhood of x.
a hyperbolic measure.
Analogous statements hold for the unstable man-
A conservative diffeomorphism f : M ! M is
ifold of x, with f replaced by f 1.
nonuniformly hyperbolic if the invariant measure
The following quantities vary measurably in
equivalent to volume is hyperbolic. The term
x  M:
“nonuniform” is a bit misleading, as uniformly
hyperbolic conservative systems are also non-
• the (inner) radii of the stable and unstable
uniformly hyperbolic. Unlike uniform hyper-
manifolds through x,
bolicity, however, nonuniform hyperbolicity
• the angle between stable and unstable mani-
allows for the possibility of different strengths of
folds at x, and,
hyperbolicity along different orbits.
• the rates of contraction in these manifolds.
Nonuniformly hyperbolic diffeomorphisms
exist on all manifolds (Katok 1979; Dolgopyat
and Pesin 2002), and there are C1-open sets of The stable and unstable laminations of a non-
nonuniformly hyperbolic diffeomorphisms that uniformly hyperbolic system are absolutely con-
are not Anosov diffeomorphisms (Shub and tinuous. The precise definition of absolute
Wilkinson 2000). In general, it is a very difficult continuity here is slightly different than in the
problem to establish whether a given map carries a uniformly and partially hyperbolic setting, but
hyperbolic measure that is nonsingular with the consequences (AC1) and (AC2) of absolute
respect to volume. continuity continue to hold.

Hyperbolic Blocks Ergodic Properties of Nonuniformly Hyperbolic

As mentioned above, the derivative of f along Diffeomorphisms
almost every orbit of a nonuniformly hyperbolic Since the stable and unstable laminations are
system looks like the derivative down the orbit of absolutely continuous, the Hopf Argument can
a uniformly hyperbolic system; the nonuniformity be applied in this setting to show:
can be detected only by examining a positive
measure set of orbits. Recall that Lusin’s Theorem Theorem 4 (Pesin) Let f be C 2, conservative
in measure theory states that every Borel measur- and nonuniformly hyperbolic. Then there exists a
able function is continuous on the complement of (mod 0) partition P of M into countably many
an arbitrarily small measure set. A sort of ana- f-invariant sets of positive volume such that the
logue of Lusin’s theorem holds for nonuniformly restriction of f to each P  P is ergodic.
hyperbolic maps: every C2, nonuniformly hyper-
bolic diffeomorphism is uniformly hyperbolic on The proof of this theorem is also exposited in
a (noninvariant) compact set whose complement (Pugh and Shub 1989). The countable partition
Smooth Ergodic Theory 327

can in examples be countably infinite; nonuniform of Ledrappier and Young (1985a, 1985b), and the
hyperbolicity alone does not imply ergodicity. proof by Barreira-Pesin-Schmeling that hyper-
bolic measures have a well-defined dimension
The Dissipative Case (Barreira et al. 1999). Hyperbolic Dynamical Sys-
As mentioned above, establishing the existence of tems contains a discussion of these results; see this
a nonsingular hyperbolic measure is a difficult entry there for further information.
problem in general. In systems with some global
form of hyperbolicity, such as partial hyper-
bolicity, it is sometimes possible to “borrow” the The Presence of Critical Points and Other
expansion from the unstable direction and lend it Singularities
to the central direction, via a small perturbation.
Nonuniformly hyperbolic attractors have been Now for a discussion of the aforementioned tech-
constructed in this way (Viana 1997). This nical difficulties that arise in the presence of sin-
method is also behind the construction of a C1 gularities and critical points for the derivative.
open set of nonuniformly hyperbolic Singularities, that is, points where Df (or even
diffeomorphisms in (Shub and Wilkinson 2000). f ) fails to be defined, arise naturally in the study of
For a given system of interest, it is sometimes billiards and hard sphere gases. The first subsec-
possible to prove that a given invariant measure is tion discusses some progress made in smooth
hyperbolic by establishing an approximate form ergodic theory in the presence of singularities.
of hyperbolicity. The idea, due to Wojtkowski and Critical points, that is, points where Df fails to
called the cone method, is to isolate a measurable be invertible, appear inescapably in the study of
bundle of cones in TM defined over the support of noninvertible maps. This type of complication
the measure, such that the cone at a point x is already shows up for noninvertible maps in
mapped by Dx f into the cone at f(x). Intersecting dimension 1, in the study of unimodal maps of
the images of these cones under all iterates of Df, the interval. The second subsection discusses the
one obtains an invariant subbundle of TM over the technique of parameter exclusion, developed by
support of f that is nonuniformly expanded. Jakobson, which allows for an ergodic analysis of
Lai-Sang Young has developed a very general a parametrized family of maps with criticalities.
method (Young 1998) for proving the existence of The technical advances used to overcome these
SRB measures with strong mixing properties in issues in the interval have turned out to have
systems that display “some hyperbolicity”. The applications to dissipative, nonhyperbolic,
idea is to isolate a region X in the manifold diffeomorphisms in higher dimension, where the
where the first return map is hyperbolic and dis- derivative is “nearly critical” in places. The last
tortion estimates hold. If this can be done, then the subsection describes extensions of the parameter
map carries a mixing, hyperbolic SRB measure. exclusion technique to these near-critical maps.
The precise rate of mixing is determined by the
properties of the return-time function to X; the Hyperbolic Billiards and Hard Sphere Gases
longer the return times, the slower the rate of In the 1870s the physicist Ludwig Boltzmann
mixing. hypothesized that in a mechanical system with
More results on the existence of hyperbolic many interacting particles, physical measure-
measures are discussed in the next section. ments (observables), averaged over time, will
An important subject in smooth ergodic theory converge to their expected value as time
is the relationship between entropy, Lyapunov approaches infinity. The underlying dynamical
exponents, and dimension of invariant measures system in this statement is a Hamiltonian system
of a smooth map. Significant results in this with many degrees of freedom, and the “expected
area include the Pesin entropy formula (Pesin value” is with respect to Liouville measure.
1976), the Ruelle entropy inequality (Ruelle Loosely phrased in modern terms, Boltzmann’s
1978), the entropy-exponents-dimension formula hypothesis states that a generic Hamiltonian
328 Smooth Ergodic Theory

system of this form will be ergodic on constant Bunimovich (1974). Sinai and Bunimovich pro-
energy submanifolds. Reasoning that the time ved that these billiards are ergodic and non-
scales involved in measurement of an observable uniformly hyperbolic. For the Boltzmann-Sinai
in such a system are much larger than the rate of problem with N  3, the relevant associated
evolution of the system, Boltzmann’s hypothesis dynamical system is a higher dimensional billiard
allowed him to assume that physical quantities table in Euclidean space, with circular arcs
associated to such a system behave like constants. replaced by cylindrical boundary components.
In 1963, Sinai revived and formalized this In a planar billiard table with circular/flat
ergodic hypothesis, stating it in a concrete formu- boundary, the behavior of vectors encountering a
lation known as the Boltzmann-Sinai Ergodic flat segment of boundary is easily understood, as
Hypothesis. In Sinai’s formulation, the particles is the behavior of vectors meeting a circular seg-
were replaced by N hard, elastic spheres, and to ment in a neighborhood of the normal vector. If
compactify the problem, he situated the spheres the billiard map is ergodic, however, every open
on a k-torus, k ¼ 2, 3. The Boltzmann-Sinai Ergo- set of vectors will meet the singularities in the
dic Hypothesis is the conjecture that the induced table infinitely many times. To establish the non-
Hamiltonian system on the 2kN-dimensional con- uniform hyperbolicity of such billiard tables via
figuration space is ergodic on constant energy conefieds, it is therefore necessary to understand
manifolds, for any N  2. precisely the fraction of time orbits spend near
Sinai verified this conjecture for N ¼ 2 by these singularities. Furthermore, to use the Hopf
reducing the problem to a billiard map in the argument to establish ergodicity, one must avoid
plane. As background for Sinai’s result, a brief the singularities in the second derivative, where
discussion of planar billiard maps follows. distortion estimates break down. The techniques
Let D
ℝs be a connected region whose for overcoming these obstacles involve imposing
boundary @D is a collection of closed, piecewise restrictions on the geometry of the table (even
smooth simple curves the plane. The billiard map more so for higher dimensional tables), and are
is a map defined (almost everywhere) on @D well beyond the scope of this paper.
[π, π]. To define this map, one identifies each The study of hyperbolic billiards and hard
point (x, θ)  @D [π, π] with an inward- sphere gases has a long and involved history.
pointing tangent vector at x in the plane, so that See the articles (Szász 2000) and (Chernov and
the normal vector to @D at x corresponds to the Markarian 2003) for a survey of some of the
pair (x, π/2). This can be done in a unique way on results and techniques in the area. A discussion
the smooth components of @D. Then f(x, θ) is of methods in singular smooth ergodic theory,
obtained by following the ray originating at with particular applications to the Lorentz attrac-
(x, θ) until it strikes the boundary @D for the first tor, can be found in (Araújo and Pacifico 2007).
time at (x0, θ 0). Reflecting this vector about the Another, more classical, reference is (Katok et al.
normal at x’, define f(x, θ) ¼ (x0, π  θ 0). 1986), which contains a formulation of properties
It is not hard to see that the billiard map is on a critical set, due to Katok–Strelcyn, that are
conservative. The billiard map is piecewise useful in establishing ergodicity of systems with
smooth, but not in general smooth: the degree of singularities.
smoothness of f is one less than the degree of
smoothness of @D. In addition to singularities Interval Maps and Parameter Exclusion
arising from the corners of the table, there are The logistic family of maps ft : x 7! tx(1  x)
singularities arising in the second derivative of defined on the interval [0, 1] is very simple to
f at the tangent vectors to the boundary. define but exhibits an astonishing variety of
In the billiards studied by Sinai, the boundary dynamical features as the parameter t varies. For
@D consists of a union of concave circular arcs small positive values of t, almost every point in I is
and straight line segments. Similar billiards, but attracted under the map ft to the sink at 1. For
with convex circular arcs, were first studied by values of t > 4, the map has a repelling hyperbolic
Smooth Ergodic Theory 329

Cantor set. As the value of t increases between continuous invariant measure (Graczyk and
0 and 4, the map ft undergoes a cascade of period- Swiatek 1997; Lyubich 1999). Jakobson’s method
doubling bifurcations, in which a periodic sink applies not only to the logistic family but to a very
of period 2n becomes repelling and a new sink of general class of C3 one-parameter families of
period 2n þ 1 is born. At the accumulation of maps on the interval.
period doubling at t ≈ 3.57, a periodic point
of period 3 appears, forcing the existence of peri- Near-Critical Diffeomorphisms
odic points of all periods. The dynamics of ft for Jakobson’s method in one dimension proved to
t close to 4 has been the subject of intense inquiry extend to certain highly dissipative diffeomo-
in the last 20 years. rphisms. The seminal paper in this extension is
The map ft, for t close to 4, shares some of the due to Benedicks and Carleson; the method has
features of the doubling map T2; it is 2-to-1, since been extended in a series of papers (Mora
except at the critical point 12, and it is uniformly and Viana 1993; Benedicks and Young 1993;
expanding in the complement of a neighborhood Benedicks and Viana 2001) and has been formu-
of this critical point. Because this neighborhood lated in an abstract setting (Wang and Young
of the critical point is not invariant, however, the 2008).
only invariant sets on which ft is uniformly hyper- This extension turns out to be highly non-
bolic have measure zero. Furthermore, the second trivial, but it is possible to describe informally
derivative of ft vanishes at the critical point, mak- the similarities between the logistic family and
ing it impossible to control distortion for orbits higher-dimensional “near critical” diffeo-
that spend too much time near the critical point. morphisms. The diffeomorphisms to which this
Despite these serious obstacles, Michael method applies are crudely hyperbolic with a one
Jakobson (1981) found a method for constructing dimensional unstable direction. Roughly this
absolutely continuous invariant measures for means that in some invariant region of the mani-
maps in the logistic family. The method has fold, the image of a small ball under f will be
come to be known as parameter exclusion and stretched significantly in one direction and shrunk
has seen application far beyond the logistic fam- in all other directions. The directions of stretching
ily. As with billiards, it is possible to formulate and contraction are transverse in a large propor-
geometric conditions on the map ft that control tion of the invariant region, but there are isolated
both expansion (hyperbolicity) and distortion on a “near critical” subregions where expanding and
positive measure set. As these conditions involve contracting directions are nearly tangent.
understanding infinitely many iterates of ft, they The dynamics of such a diffeomorphism are
are impossible to verify for a given parameter very close to 1-dimensional if the contraction is
value t. strong enough, and the diffeomorphism resembles
Using an inductive formulation of this condi- an interval map with isolated critical points, the
tion, Jakobson showed that the set of parameters critical points corresponding to the critical regions
t near 4 that fail to satisfy the condition at iterate where stable and unstable directions are tangent.
n have exponentially small measure (in n). He An illustration of this type of dynamics is the
thereby showed that for a positive Lebesgue mea- Hénon family of maps fa, b : (x, y) 7! (1  ax2 þ
sure set of parameter values t, the map ft has an by, x), the original object of study in Benedicks-
absolutely continuous invariant measure Carlesson’s work. When the parameter b is set to
(Jakobson 1981). This measure is ergodic 0, the map fa, b is no longer a diffeomorphism, and
(mixing) and has a positive Lyapunov exponent. indeed is precisely a projection composed with the
The delicacy of Jakobson’s approach is confirmed logistic map. For small values of b and appropri-
by the fact that for an open and dense set of ate values of a, the Hénon map is strongly dissi-
parameter values, almost every orbit is attracted pative and displays the near critical behavior
to a periodic sink, and so ft has no absolutely described in the previous paragraph. In analogy
330 Smooth Ergodic Theory

to Jakobson’s result, there is a positive measure set interest, including the infinite dimensional sys-
of parameters near b ¼ 0 where fa, b has a mixing, tems that arise in the study of partial differen-
hyperbolic SRB measure. tial equations.
See (Viana and Lutstsatto 2003) for a detailed • Carry the methods of smooth ergodic theory
exposition of the parameter exclusion method for further into the study of smooth actions of
Hénon-like maps. discrete groups (other than the integers) on
manifolds. When do such actions admit
(possibly non-invariant) “physical” measures?
Future Directions
There are other interesting open areas of future
In addition to the open problems discussed in the inquiry, but this gives a good sample of the range
previous sections, there are several general ques- of possibilities.
tions and problems worth mentioning:

• What can be said about systems with every- Bibliography

where vanishing Lyapunov exponents? Open
sets of such systems exist in arbitrary dimen- Alves JF, Bonatti C, Viana M (2000) SRB measures for
sion. Pesin theory carries into the non- partially hyperbolic systems whose central direction is
mostly expanding. Invent Math 140(2):351–398
uniformly hyperbolic setting the basic Anosov DV (1967) Geodesic flows on closed Riemann
principles from uniformly hyperbolic theory manifolds with negative curvature. Proceedings of the
(in particular, Fundamental Principles #3 and Steklov Institute of Mathematics, No. 90. Translated
4 above). To what extent do properties of iso- from the Russian by S. Feder. American Mathematical
Society, Providence
metric and unipotent systems (for example, Anosov DV, Sinai JG (1967) Certain smooth ergodic sys-
Fundamental Principle #2) extend to conserva- tems. (Russian) Uspehi Mat Nauk 22 no. 5(137):107–172
tive systems all of whose Lyaponov exponents Araújo V, Pacifico MJ (2007) Three Dimensional Flows.
vanish? Publicações Matemáticas do IMPA. [IMPA Mathemat-
ical Publications] 26 Col’quioquio Brasileiro de
• Can one establish the existence of and analyze Matemática. [26th Brazilian Mathematics Colloquium]
in general the conservative systems on surfaces Instituto de Matemática Pura e Aplicada (IMPA), Rio
that have two positive measure regimes: one de Janeiro
where Lyapunov exponents vanish, and the Arnol’d VI, Avez A (1986) Ergodic problems of classical
mechanics. Translated from the French by Avez,
other where they are nonzero? Such systems A. Benjamin WA, Inc., New York/Amsterdam
are conjectured exist in the presence of KAM Baladi V (2000) Positive transfer operators and decay of
phenomena surrounding elliptic periodic correlations. Advanced Series in Nonlinear Dynamics,
points. 16. World Scientific Publishing Co., Inc., River Edge
Barreira L, Pesin Y, Schmeling J (1999) Dimension and
• On a related note, how common are conserva- product structure of hyperbolic measures. Ann Math
tive systems whose Lyapunov exponents are 149(2):755–783
nonvanishing on a positive measure set? See Benedicks M, Carleson L (1991) The dynamics of the
(Bochi and Viana 2004) for a discussion. Hénon map. Ann Math 133(2):73–169
Benedicks M, Viana M (2001) Solution of the basin problem
• Find a broad description of those dissipative for Hénon-like attractors. Invent Math 143(2):375–434
systems that admit finitely (or countably) many Benedicks M, Young LS (1993) Sinai–Bowen–Ruelle
physical measures. Are such systems dense measures for certain Hénon maps. Invent Math
among all dissipative systems, or possibly 112(3):541–576
Bochi J, Viana M (2004) Lyapunov exponents: how fre-
generic among a restricted class of systems? quently are dynamical systems hyperbolic? Modern
See (Shub and Wilkinson 2000; Palis 2005) for dynamical systems and applications. Cambridge Uni-
several questions and conjectures related to versity Press, Cambridge, pp. 271–297
this problem. Bonatti C, Viana M (2000) SRB measures for partially
hyperbolic systems whose central direction is mostly
• Extend the methods in the study of systems contracting. Israel J Math 115:157–193
with singularities to other specific systems of
Smooth Ergodic Theory 331

Bonatti C, Dìaz LJ, Viana M (2005) Dynamics beyond Hopf E (1939) Statistik der geodätischen Linien in
uniform hyperbolicity. A global geometric and proba- Mannigfaltigkeiten negativer Krümmung. Ber Verh
bilistic perspective. Encyclopaedia of mathematical Sachs Akad Wiss Leipzig 91:261–304. (German)
sciences, 102. Mathematical Physics, III. Springer, Jakobson MV (1981) Absolutely continuous invariant
Berlin measures for one-parameter families of one-
Bowen R (1970) Markov partitions for Axiom dimensional maps. Commun Math Phys 81(1):39–88
A diffeomorphisms. Am J Math 92:725–747 Katok A (1979) Bernoulli diffeomorphisms on surfaces.
Bowen R (1975) Equilibrium states and the ergodic theory Ann Math 110(2):529–547
of Anosov diffeomorphisms. Lecture Notes in Mathe- Katok A, Hasselblatt B (1995) Introduction to the modern
matics, vol 470. Springer, Berlin/New York theory of dynamical systems. With a supplementary
Brin MI, Pesin JB (1974) Partially hyperbolic dynamical chapter by Katok and Leonardo Mendoza. Encyclope-
systems. Izv Akad Nauk SSSR Ser Mat 38:170–212. dia of Mathematics and its Applications, vol 54. Cam-
(Russian) bridge University Press, Cambridge
Brin M, Stuck G (2002) Introduction to dynamical sys- Katok A, Strelcyn JM, Ledrappier F, Przytycki F (1986)
tems. Cambridge University Press, Cambridge Invariant manifolds, entropy and billiards; smooth
Bunimovic LA (1974) The ergodic properties of certain maps with singularities. Lecture notes in mathematics,
billiards. (Russian) Funkcional Anal i Priložen 8(3): 1222. Springer, Berlin
73–74 Kifer Y (1986) Ergodic theory of random transformations.
Burns K, Wilkinson A, On the ergodicity of partially Progress in probability and statistics, vol 10.
hyperbolic systems. Ann of Math. To appear Birkhäuser, Boston
Burns K, Pugh C, Shub M, Wilkinson A (2001) Recent Kifer Y (1988) Random perturbations of dynamical sys-
results about stable ergodicity. Smooth ergodic theory tems. Progress in probability and statistics, 16.
and its applications. ( Seattle, 1999), 327–366. Proc Birkhäuser, Boston
Sympos Pure Math, 69, Amer Math Soc, Providence Ledrappier F, Young LS (1985a) The metric entropy of
Burns K, Dolgopyat D, Pesin Y, Pollicott, M. Stable ergo- diffeomorphisms. I. Characterization of measures sat-
dicity for partially hyperbolic attractors with negative isfying Pesin’s entropy formula. Ann Math 122(2):
central exponents. Preprint 509–539
Chernov N, Markarian R (2003) Introduction to the ergo- Ledrappier F, Young LS (1985b) The metric entropy of
dic theory of chaotic billiards. Second edition. diffeomorphisms. II. Relations between entropy, expo-
Publicações Matemáticas do IMPA. IMPA Mathemati- nents and dimension. Ann Math 122(2):540–574
cal Publications 24 Col’quio Brasileiro de Liu PD, Qian M (1995) Smooth ergodic theory of random
Matemática. 26th Brazilian Mathematics Colloquium dynamical systems, Lecture notes in mathematics,
Instituto de Matemática Pura e Aplicada (IMPA), Rio vol 1606. Springer, Berlin
de Janeiro Lyubich M (1999) Feigenbaum–Coullet–Tresser univer-
Cornfeld IP, Fomin SV, Sinai YG (1982) Ergodic theory. sality and Milnor’s hairiness conjecture. Ann Math
Translated from the Russian by A. B. Sosinskii. 149(2):319–420
Grundlehren der mathematischen Wissenschaften Mañé R (1987) Ergodic theory and differentiable dynam-
[Fundamental Principles of Mathematical Sciences], ics. Translated from the Portuguese by Silvio
vol 245. Springer, New York L. Ergebnisse der Mathematik und ihrer Grenzgebiete
Dolgopyat D (2004) On differentiability of SRB states for (3) [Results in Mathematics and Related Areas (3)],
partially hyperbolic systems. Invent Math 155(2): vol 8. Springer, Berlin
389–449 Mora L, Viana M (1993) Abundance of strange attractors.
Dolgopyat D, Pesin Y (2002) Every compact manifold Acta Math 171(1):1–71
carries a completely hyperbolic diffeomorphism. Palis J (2005) A global perspective for non-conservative
Ergod Theor Dynam Syst 22(2):409–435 dynamics. Ann Inst H Poincaré Anal Non Linéaire
Dolgopyat D, Wilkinson A (2003) Stable accessibility is 22(4):485–507
C1 dense. Geometric methods in dynamics. Pesin JB (1976) Characteristic Ljapunov exponents, and
II. Astérisque No. 287 ergodic properties of smooth dynamical systems with
Graczyk J, Swiatek G (1997) Generic hyperbolicity in the invariant measure. Dokl Akad Nauk SSSR 226. (Russian)
logistic family. Ann Math 146(2):1–52 Pesin JB (1977) Characteristic Ljapunov exponents, and
Grayson M, Pugh C, Shub M (1994) Stably ergodic smooth ergodic theory. Uspehi Mat Nauk 32(196):
diffeomorphisms. Ann of Math 140(2):295–329 55–112. 287 (Russian)
Hirsch MW (1979) Differential topology. Graduate Texts Pesin YB (2004) Lectures on partial hyperbolicity and
in Mathematics, No. 33. Springer, New York/ stable ergodicity. Zurich lectures in advanced mathe-
Heidelberg matics. European Mathematical Society (EMS), Zürich
Hirsch MW, Pugh CC, Shub M (1977) Invariant manifolds. Pesin YB, Sinai YG (1983) Gibbs measures for partially
Lecture Notes in Mathematics, vol 583. Springer, hyperbolic attractors. Ergod Theor Dynam Syst
Berlin/New York 2(3–4):417–438
332 Smooth Ergodic Theory

Pugh C, Shub M (1972) Ergodicity of Anosov actions. Sinai JG (1972) Gibbs measures in ergodic theory.
Invent Math 15:1–23 (Russian) Uspehi Mat Nauk 27 no. 4(166):21–64
Pugh C, Shub M (1989) Ergodic attractors. Trans Am Math Szász D (2000) Boltzmann’s ergodic hypothesis, a conjec-
Soc 312(1):1–54 ture for centuries? Hard ball systems and the Lorentz
Pugh C, Shub M (1996) Stable ergodicity and partial gas, Encycl Math Sci, vol 101. Springer, Berlin,
hyperbolicity. International conference on dynamical pp 421–448
systems (Montevideo, 1995), 182–187, Pitman Res Tsujii M (2005) Physical measures for partially hyperbolic
Notes Math Ser, 362, Longman, Harlow surface endomorphisms. Acta Math 194(1):37–132
Robinson C (1995) Dynamical systems. Stability, sym- Viana M (1997) Multidimensional nonhyperbolic
bolic dynamics, and chaos. Studies in Advanced Math- attractors. Inst Hautes Études Sci Publ Math 85:
ematics. CRC Press, Boca Raton 63–96
Rodríguez HA, Rodríguez HF, Ures R (2008) Partially Viana M, Lutstsatto S (2003) Exclusions of parameter
hyperbolic systems with 1D-center bundle. Invent values in Hénon-type systems. (Russian) Uspekhi Mat
Math 172(2) Nauk 58 (2003), no. 6(354), 3–44; translation in
Ruelle D (1976) A measure associated with axiom-A Russian Math. Surveys 58(6):1053–1092
attractors. Am J Math 98(3):619–654 Wang QD, Young LS (2008) Toward a theory of rank one
Ruelle D (1978) An inequality for the entropy of differen- attractors. Ann Math 167(2)
tiable maps. Bol Soc Brasil Mat 9(1):83–87 Young LS (1998) Statistical properties of dynamical sys-
Shub M, Wilkinson A (2000) Pathological foliations and tems with some hyperbolicity. Ann Math 147(2):
removable zero exponents. Invent Math 139(3):495–508 585–650
Sinai JG (1961) Geodesic flows on compact surfaces of Young LS (2002) What are SRB measures, and which
negative curvature. Dokl Akad Nauk SSSR 136: dynamical systems have them? Dedicated to David
549–552 (Russian); translated as Soviet Math Dokl 2: Ruelle and Yasha Sinai on the occasion of their 65th
106–109 birthdays. J Stat Phys 108(5–6):733–754
 preserving smooth flow. Such flows are also
Ergodic and Spectral Theory called multi-valued or locally Hamiltonian.
of Area-Preserving Flows on We say that two flows ’ℝ on (X, m) and fℝ on
Surfaces (Y, v) and are isomorphic as measure-preserving
flows if there exists an isomorphism F : X ! Y, i.e.,
Krzysztof Frączek1 and Corinna Ulcigrai2 a bimeasurable map which transports the measure
1
Faculty of Mathematics and Computer Science, m to v (i.e., m(F1(A)) ¼ n(A) for any Borel set
Nicolaus Copernicus University, Toruń, Poland A  Y) and commutes with the dynamics, i.e.,
2
Institut für Mathematik, Universität Zürich, F(ft(x)) ¼ ct(F(x)) for m-almost every (a.e.)
Zürich, Switzerland x  X. If additionally ’ℝ, fℝ and F are smooth,
then the flows ’ℝ and fℝ are smoothly
isomorphic.
Article Outline By a joining between two flows ’ℝ on (X, m)
and fℝ on (Y, v) we mean any probability (’t 
Glossary ft)t  ℝ– invariant measure on X  Y whose pro-
Definition of the Subject jections on X and Y are equal to m and v, respec-
Introduction tively. Two flows ’ℝ and fℝ are called disjoint
Examples (in the sense of Furstenberg) if their only common
Locally Hamiltonian flows joining is the product joining m  n.
Background and Tools A m-preserving flow ’ℝ is ergodic if any
Invariant Measures and (Unique) Ergodicity invariant Borel set A  X (i.e., m(’tAΔA) ¼ 0 for
Behavior of Ergodic Averages all t  ℝ) has measure zero or its compliment X\A
Mixing Properties has measure zero.
Spectral Properties A m-preserving flow ’ℝ is weakly mixing if
Disjointness Results for any pair f, g  L20 ðX, mÞ L20 ðX, mÞ is the sub-
Open Directions space of zero mean functions),
Bibliography
T
1
Glossary lim j ð f ∘’t Þ  g dmjdt ¼ 0:
T!1 T 0 X

Borel flow is a family ’ℝ ¼ (’t)t  ℝ of Borel maps A m-preserving flow ’ℝ is mixing if for any pair
on a topological space X such that (t, x) 7! ’tx is f, g  L20 ðX, mÞ,
also Borel, ’0 ¼ IdX and ’t1 þt2 ¼ ’t1 ∘ ’t2 for all
t1, t2  ℝ. The flow preserves a Borel probability lim ð f ∘’t Þ  g dm ¼ 0:
measure m on X if m(’t(A)) ¼ m(A) for any Borel t!1 X

set A and any t  ℝ. Then we say that ’ℝ is a

A m - preserving flow ’ℝ is rigid if for any
measure-preserving flow of (X, m).
f  L2(X, m),
If S is compact smooth manifold and the map
(t, x) 7! ’tx is smooth, then ’ℝ is called a
lim inf k f  f ∘’t kL2 ðX,mÞ ¼ 0:
smooth flow. If additionally S is a surface t!þ1
(dimension two manifold) and an invariant mea-
sure m is smooth and positive (has a smooth pos- A m-preserving flow ’ℝ is mildly mixing if for
itive density), then ’ℝ is called an area- any non-zero f  L2(X, m),

© Springer Science+Business Media, LLC, part of Springer Nature 2023 333

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_775
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2021
https://doi.org/10.1007/978-3-642-27737-5_775-1
334 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

lim inf k f  f ∘’t kL2 ðX,mÞ > 0: state physics, or statistical mechanics models.
t!þ1
Despite being low-dimensional systems of zero
topological entropy, they present a rich display
A m-preserving flow ’ℝ is mixing of all orders if
of fine chaotic properties.
for any n  2 and any n-tuple A0, . . ., An1 of
We define below two classes of area-
Borel sets,
preserving flows, linear flows on translation sur-
faces and locally Hamiltonian flows, whose ergo-
m A0 \ ’t1 ðA1 Þ \    \ ’t1 þþtn1 ðAn1 Þ
dic and spectral properties have been the object of
t1 , t2 ..., tn1 !1
! mðA0 Þ  mðAn1 Þ: intense research activity in the last decades. Area-
preserving flows on surfaces also provide one of
A Borel measure sf on ℝ is the spectral measure the fundamental classes of parabolic, or slowly
of f  L2(X, m) if chaotic, dynamical systems (see also the survey
Ulcigrai (2021). Contrary to hyperbolic systems,
which display sensitive dependence on initial con-
eits dsf ðtÞ ¼ ðf ∘’s Þ  f dm for all s  ℝ:
ℝ X ditions (i.e., divergence in time of nearby initial
conditions, the so-called butterfly effect) which
The spectral type of a m-preserving flow ’ℝ is an happens (infinitesimally) at exponential speed,
equivalence class of a Borel measure s on ℝ such parabolic systems display a slow form of sensitive
that for every f  L20 ðX, mÞ, sf is absolutely con- divergence, with speed of divergence which is
tinuous with respect to s and there exists sub-exponential (and actually polynomial or sub-
f 0  L20 ðX, mÞ such that sf 0 ¼ s: polynomial in all known examples).

Brief historical remarks Early results on flows

Definition of the Subject on the ergodic theory of flows on surfaces date
back to the 1970s (two articles Aranson and
We consider in this entry flows on closed, Grines (1973) and Katok (1973) can be seen as
orientable smooth surfaces of genus g  1 an impetus for development). Interval exchange
which preserve an invariant measure of full sup- transformations, discrete maps which naturally
port and survey what is known of their ergodic, appear as Poincaré sections of flows on surfaces,
mixing, spectral, and joining properties. We were introduced and studied earlier (see, e.g., the
focus in particular on compact surfaces and on work by Oseledec (1966)) independently, as an
two classes of area-preserving flows, namely interesting example to study spectral aspects in
translation flows on translation surfaces (and ergodic theory. The study of the ergodic theory
their Poincaré maps, interval exchange transfor- of linear flows on translation surfaces has taken
mations (IETs)) and smooth area-preserving off in the 1980s in connection with the study of
flows known as locally Hamiltonian or multi- interval exchange transformations and billiards in
valued Hamiltonian flows. rational polygons, with seminal works by Masur
and Veech laying the grounds for the connection
with Teichmüller dynamics, a research area which
Introduction has since then benefited from the contribution of
several Fields medalists (including Avila,
Flows on surfaces are one of the most basic and Kontsevich, McMullen, Mirzakhani, and
fundamental examples of dynamical systems, Yoccoz).
studied since the birth of dynamical systems as a The study of locally Hamiltonian flows has
research field with the work of Poincaré (1987) at seen a revival of interest in the 1990s, with the
the end of the nineteenth century. Many models of seminal work of Zorich (1984) in connection with
systems of physical origin are described by flows the Novikov problem. Novikov indeed underlined
on surfaces, starting from celestial mechanics, the interest in the study of multi-valued Hamilto-
polygonal billiard dynamics, and up to solid- nians and the associated flows (Novikov 1982) in
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 335

connection with problems arising in solid-state ðx0 ðtÞ, y0 ðtÞÞ ¼ ðcos y, sin yÞ y  S1 , ð1Þ
physics as well as in pseudo-periodic topology
(see, e.g., the survey by Zorich (1999)). Indeed, which moves points at unit speed along (the image
Novikov (1982) and his school in the 1990s advo- by the projection p : ℝ2 ! ℝ2/ℤ2 of) Euclidean
cated the study of locally Hamiltonian flows as lines in direction θ (i.e., lines making an angle θ
model to describe the motion of an electron in a with the horizontal axes); see Fig. 1a.
metal under a magnetic field in the semi-classical
approximation (the surface appears here as Fermi
energy level surface). Novikov made some con- Linear flows on translation surface Linear
jectures (known as Novikov problem) on the flows (also called translation flows) can be
asymptotic behavior of trajectories of electrons. defined more in general on translation surfaces,
At the same time, Arnold (1991) made a conjec- namely surfaces which are locally Euclidean out-
ture on mixing for the flows we call today Arnold side a finite number of conical singularities.
flows. This conjecture has been the motivation for A translation surface can be defined as a quotient
a lot of the work on the mixing properties of
locally Hamiltonian flows. S≔P1 [ P2 [    [ Pn = 
The current century has seen a lot of advances
where Pi  ℝ2 for 1  i  n are disjoint polygons
in our understanding of the chaotic properties of
in ℝ2, with clockwise oriented boundary, with the
smooth area-preserving flows (a class which
property that their edges can be paired into cou-
includes locally Hamiltonian flows), in particular
ples (e, e0) where e and e0 are parallel and isomet-
exploiting Teichmüller dynamics tools, but also
ric and have opposite orientation (see an example
under the influence of the work of Marina Ratner
in Fig. 2) and the equivalence relation ~ identifies
in homogeneous dynamics. The study of linear
the edges of the pair (e, e0) by the (unique) trans-
flows on infinite translation surfaces has only
lation of ℝ2 which maps e to e0. The request that
begun in the last decade, in particular motivated
e and e0 have opposite orientations is needed to
by the study of the periodic Ehrenfest model and it
guarantee that, after gluing, one obtains a transla-
is still a widely open research direction.
tion surface, in which all sides are identified by
translations, and not a half-translation surface,
Examples where identifications by a central symmetry com-
posed with a translation are allowed. Translation
Linear flows on the torus The basic example of surfaces can be equivalently defined as the datum
an area-preserving flow is the linear flow on the of an Abelian differential on a Riemann surface,
torus 2 ≔ℝ2 =ℤ2 (see Fig. 1) given by solutions while half-translation surfaces correspond to qua-
ðxðtÞ, yðtÞÞ  2 to dratic differentials.

Ergodic and Spectral

Theory of
Area-Preserving Flows
on Surfaces, Fig. 1 A
linear flow on the torus
2 ≔ℝ2 =ℤ2 :
336 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

Ergodic and Spectral

Theory of
Area-Preserving Flows
on Surfaces,
Fig. 2 Translation surface

The resulting space S is a compact, oriented follows. Let o be a fixed smooth area form
topological surface which is endowed with a flat, (locally given in coordinates (x, y) by f(x, y)dx ^ dy
Euclidean metric outside a finite set S ¼ S(S) the where f is a smooth positive function). Thus,
(image after gluings) of the set of vertices of the equivalently, the pair (S, o) is a two-dimensional
polygons. Points in S are known as singularities; symplectic manifold. We consider smooth flows
each of these points has a neighborhood where the ’ℝ on S which preserve a measure m given by
metric has a conical singularity with a total cone integrating a smooth density with respect to o. We
angle 2πk, k  ℕ (i.e., it is locally of the form assume that the area is normalized so that m(S) ¼
ds2 ¼ dr2 þ (k r dθ)2). Translation surfaces can be 1. It turns out that such smooth area-preserving
equivalently defined as a pair (X, o) where X is a flows on S are in one-to-one correspondence with
compact Riemann surface and o is an Abelian smooth closed real-valued differential 1-forms as
differential; we refer to the surveys (Masur and follows. Given a smooth, closed, real-valued dif-
Tabachnikov 2002; Yoccoz 2010; Forni and ferential 1-form , let X be the vector field deter-
Matheus 2014). mined by  ¼ iXo where iX denotes the
On a translation surface S, the notion of a contraction operator, i.e., iXo ¼ o(X, ) and con-
direction θ  S1 (in particular the notion of hor- sider the flow ’ℝ on S given by X. Since  is
izontal and vertical direction) is well defined not closed, the transformations ’t, t  ℝ, are area-
only locally, in each polygon, but globally (since preserving. Conversely, every smooth area-
the identifications are performed by translations). preserving flow can be obtained in this way.
Thus, for each θ  S1, we can globally define a The flow ’ℝ is known as the multi-valued
flow ’yℝ ≔ ’yt t  ℝ given locally by solutions Hamiltonian flow associated to . Indeed, the
to (1). As in the case of the torus, ’yℝ moves points flow ’ℝ is locally Hamiltonian, i.e., locally one
along the (quotient to S) of lines in direction θ can find coordinates (x, y) on S in which ’ℝ is
with unit speed; see Fig. 2. Notice that at a singu- given by the solution to the equations
larity v  S with cone angle 2πk, there is not a
unique but k lines in direction θ and v is a saddle x_ ¼ @H=@y,
point of the linear flow ’yℝ with 2 k-prongs, of the y_ ¼ @H=@x
type shown in Fig. 4c for k ¼ 3. Notice further-
more that these flows preserve a Euclidean area, for some smooth real-valued Hamiltonian func-
but are discontinuous flows, since singularities are tion H. A global Hamiltonian H cannot be in
reached in finite time. general defined (see Nikolaev and Zhuzhoma
(1999), Section 1.3.4), but one can think of ’ℝ
as globally given by a multi-valued Hamiltonian
Locally Hamiltonian flows function.
Locally Hamiltonian flows necessarily have
Smooth (in particular continuous) flows on fixed points or singularities. Singularities, as
S preserving a smooth measure can be defined as shown in Fig. 4, can be either centers (Fig. 4a),
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 337

Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces, Fig. 3 Examples of flows

Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces, Fig. 4 Type of singularities

simple saddles (Fig. 4b) or multi-saddles

(i.e., saddles with 2 k pronges, k  2; see Fig. 4c
for k ¼ 3). Examples of flow trajectories are shown
in Fig. 3. For g ¼ 1, i.e., on a torus, if there is a
singularity, then there has to be another one.

Arnold flows The simplest examples of locally

Hamiltonian flows with singularities on a torus,
i.e., flows with one center and one simple saddle
(see Fig. 3b), were studied by Arnold (1991) and are Ergodic and Spectral Theory of Area-Preserving
nowadays often called Arnold flows. Any Arnold Flows on Surfaces, Fig. 5 Blokhin flows
flow is the restriction of such locally Hamiltonian
flow to a minimal component obtained by removing construction is based on a linear ergodic flow on
the center and the disk filled by periodic orbits the torus (Fig. 5b) and a flow on the annulus,
around it (called island), which, as Arnold shows which has two simple saddles connected by
in (Arnold 1991), is always bounded by a saddle separatrices (Fig. 5a). Two identical circles are
loop. cut in the torus so that their centers lie in the
same orbit of the line flow. The annulus is glued
Blokhin examples A class of special examples in into the cut-out circles so that the separatrices
higher genus which has been studied much earlier connect to the orbit passing through the center of
than the general case are Blokhin flows; see the removed circles. The resulting flow on the
Blohin (1972). These are flows on surfaces of genus two surface has a special representation
higher genus g  2 obtained essentially by gluing over an irrational rotation. We can perform the
together via a suitable surgery flows on genus one same pasting procedure several times increasing
components. More precisely, Blokhin’s the genus of the surface.
338 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

Time-changes New flows can be obtained by by cohomologous cocycles are isomorphic (see,
perturbation starting from the examples in the e.g., Avila et al. (2021a), Lemma 2.1, or Katok
previous section. The simplest perturbation is a (2001), §9, for related results). The time-change
time-change, also known as a time- fℝ given by the cocycle t is trivial if (2) with t1 ¼
reparametrization, which produces flows that t and t2 ¼ id admits a solution. In this case, fℝ
move points along the same orbits, but with dif- and fℝ are conjugated by the conjugacy
ferent speed. The flow fℝ is a time-change x 7! fuðxÞ x and hence the time-change is isomor-
(or reparametrization) of a flow fℝ on S if there phic to the original flow. Equation (2) is an exam-
exists a measurable function t : S  ℝ ! ℝ such ple of a cohomological equation. Thus, trivial
that for every x  S the map t 7! t(x, t) is time-changes are described by solutions to the
continuous and strictly increasing, and for all so-called cohomological eq. A key feature of
x  S and t  ℝ we have ftðx,tÞ ðxÞ ¼ ft ðxÞ: area-preserving flows is the existence of obstruc-
Since fℝ is assumed to be a flow, the function t tions (invariant distributions) to solve cohomo-
is an additive cocycle over the flow fℝ, that is, it logical equations; see Forni (1997). As a
satisfies the cocycle identity: consequence, among smooth time-changes,
smoothly trivial time-changes are rare (i.e., form
tðx, s þ tÞ ¼ tðfs ðxÞ, tÞ þ tðx, sÞ a finite or countable codimension subspace) and
time-changes can display essentially different
for all x  S and all s, t  ℝ. If fℝ is a smooth spectral and ergodic properties compared to the
flow, we will say that fℝ is a smooth original flow. Nevertheless, certain ergodic prop-
reparametrization if the cocycle t is a smooth erties, like ergodicity and cohomological proper-
function such that the maps t(x, ) are homeomor- ties, which only depend on the orbit structure and
phisms. Even these simple perturbations can pro- hence are independent of the time-change, persist
duce a genuinely new flow, i.e., a flow which is in a time-change.
not measurably conjugated to the unperturbed
flow (see below). More drastic (non-smooth)
time-changes, when t is allowed to have singu- Background and Tools
larities where it blows up, can introduce singular-
ities in the time-changed flow. The simplest type Minimality and minimal components From the
of singularity is a stopping point (also known as topological dynamics point of view, one is inter-
fake saddle point), i.e., a fixed point which has a ested in the qualitative behavior of every trajec-
neighborhood foliated by flow trajectories. tory. An example where all orbits can be
There is an obvious way to produce measurably understood are linear flows on tori (e.g., given
(or even smoothly) conjugated flows using by (1)), which satisfy a well-known dichotomy:
reparametrizations, associated to solutions to the either all orbits are periodic, namely there exists a
so-called cohomological equation as follows. t0 > 0 such that ’t0 þt ðxÞ ¼ ’t ðxÞ for every x  S
Two additive cocycles t1(x, t), t2(x, t) are said to and t  ℝ (this happens exactly when the slope θ
be measurably (respectively smoothly) cohomo- in (1) is rational), or ’ℝ is minimal, i.e., every
logous if their difference t1  t2 is a measurable (forward) trajectory {’t(x), t  0}, for any x  S,
(respectively smooth) coboundary, i.e., if there is dense.
exists a measurable (respectively smooth) function In presence of fixed points (see Fig. 4), which
u : M ! ℝ, called the transfer function, such that are unavoidable in higher genus, the definition of
minimality should be adjusted as follows. We say
t1 ðx, tÞ  t2 ðx, tÞ ¼ uðxÞ  u∘ft ðxÞ ð2Þ that the trajectory of x  S is regular if the whole
orbit {’t(x), t  ℝ} is well defined and the limits
for all (x, t)  S  ℝ. An elementary, but funda- limt ! 1’t(x) do not exist. A flow ’ℝ : S ! S is
mental, result establishes that time-changes given called quasi-minimal (or simply minimal, by
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 339

abusing the terminology) if every regular trajec- connections, called an island of periodic orbits;
tory is dense. If ’ℝ has a fixed point that is a center see Fig. 6a. Hence, in presence of centers, the flow
(Fig. 4a), then since a neighborhood of the center ’ℝ is never minimal (since orbits in the comple-
is foliated by periodic orbits, ’ℝ cannot be ment of the island avoid the island and hence
(quasi-)minimal. cannot be dense).
Let us say that a segment of a trajectory, of the Mayer (1943), Levitt (1983), and Zorich
form {’t(x), a < t < b} (where possibly a ¼  1 (1999) proved independently that each smooth
or b ¼ þ 1 ) is a saddle connection if it starts and area-preserving flow can be decomposed into a
ends in a saddle point (when a or b is 1, by this finite number of subsurfaces with boundary Si,
we mean that limt ! 1’t(x) are saddle points). i ¼ 1, . . ., N such that for each i the restriction of
A saddle connection is called a saddle loops if the ’ℝ to Si is a periodic component, i.e., the interior
initial and final saddles are the same (see Fig. 6a). of Si if foliated into closed orbits of ’ℝ, or Si is
A classical result on flows on surfaces (see, such that the restriction of ’ℝ to Si is (quasi-)
e.g., Yoccoz (2010)) is that if ’ℝ has no saddle minimal; the latter are called minimal compo-
connections, then it is (quasi-)minimal. The nents. Periodic components are either islands
corresponding result at the level of IETs, proved (as in Fig. 6a) or cylinders filled by periodic orbits
by Keane (1975), is that an IET T with an irreduc- and bounded by saddle loops, as in Fig. 6b. Min-
ible combinatorial datum π such that the orbits imal components (see an example in Fig. 7a), by
discontinuities of T are all infinite and distinct topological reasons, cannot be more than g (the
(a condition called I.D.O.C. by Keane (1975) genus of S). The flows in Fig. 3, for example, can
and nowadays known as Keane’s condition in be decomposed, in the case of 3a, into three peri-
the literature; see, e.g., Yoccoz (2010)) is minimal. odic components, two islands and one cylinder
Notice that if ’ℝ has a fixed point which is a filled by closed orbits, and two minimal compo-
center, one can show that the center is contained in nents (one of genus one and one of genus two),
a disk filled with closed (i.e., periodic) trajectories while, in the case of the flow on the torus in
and bounded by a saddle loop or a union of saddle Fig. 3b, there is one island and one minimal com-
ponent (the so-called Arnold flow).
One can show (see below) that minimal com-
ponents of a locally Hamiltonian flow (and in
particular minimal such flows, for which S is in
itself a minimal component) are time-
reparametrization (see the previous section) of
linear flows on translation surfaces (although
they time-change in this case is singular), so in
Ergodic and Spectral Theory of Area-Preserving particular, they have the same orbits and the same
Flows on Surfaces, Fig. 6 Periodic components topological behavior as linear flows (see, e.g.,

Ergodic and Spectral

Theory of
Area-Preserving Flows
on Surfaces, Fig. 7 A
minimal component
340 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

Zorich (1999). This was in part one of the original δi  [1, 1] such that T(x) ¼ x þ δi for any
motivations (in addition to the unfolding of ratio- x  Ii. Notice that a rotation Rα can be seen as a
nal billiards in the West) that sparked the interest 2-IET, which exchanges the intervals [0, 1α)
of mathematicians such as Zorich in the ergodic and [1α, 1). Thus IETs provide a natural gener-
theory of linear flows. alization of circle rotations and play for linear
flows in higher genus an analogous role to rota-
Interval exchange maps and Poincaré tions for linear flows on tori. In general, the
sections A central idea introduced by Poincaré images T(I1), . . ., T(Id) are exchanged intervals
was that the study of a surface flow can be often which form a new partition of [0, 1). A d-IET is
reduced to the study of a one-dimensional discrete completely determined by two data, namely a
dynamical system, by taking what we nowadays combinatorial datum π which determines the
call a Poincaré section and considering the order of the exchanged intervals in [0, 1) (this
Poincaré first return map of the flow to the section can be a permutation of {1, . . ., d} or a pair of
(when and where it is defined). permutations of an alphabet of cardinality d – see
In genus one, if we consider the linear flow (1) Yoccoz (2010) – a convention which is often
on 2 and take the horizontal side ½0, 1  f0g  useful to study fine chaotic properties of IETs)
2 , the Poincaré map is the (rigid) rotation and a length vector l ¼ (l1, . . ., ld) which
Rα : [0, 1] ! [0, 1] given by x 7! Rα(x) ¼ x þ belongs to the simplex
α mod 1 (where α ¼ cot θ, see (1)). More
generally, if we start from a flow ’ℝ ≔ (’t)t  ℝ Dd ≔ l  ℝdþ , li  0 for all 1  i  d, ð3Þ
on a torus, i.e., on a compact, orientable surface
S of genus one, and assume that it does not have d
fixed points or closed orbits (or more generally l
i¼1 i
¼1 :
Reeb components; see Nikolaev and Zhuzhoma
(1999)), there is a (global) section given by a Then the corresponding IET T ¼ T(π,l) exchanges
closed transverse curve and the Poincaré first the intervals
return map to it is a diffeomorphism f : S1 ! S1
of the circle S1 ffi ℝ/ℤ.
I i ¼ ½li , r i Þ ¼ lj , lj for i ¼ 1, . . . , d
Interval exchange transformations (IETs) As in 0j<i 0ji
the case of genus one, an essential tool to study a
higher genus flow is to consider a (local) trans- according to the permutation π (Fig. 8b). We say
versal I  S to the flow and the Poincaré first that the combinatorial datum is irreducible if the
return map T of the flow on I (when it is defined, indexes {1, . . ., d} of subintervals cannot be
for example, almost everywhere when the flow decomposed into two subsets {1, . . ., k} and
preserves a finite measure with full support; see {k þ 1, d} with 1  k < d which are exchanged
more generally (Nikolaev and Zhuzhoma 1999)). by themselves.
Consider first a linear flow on a translation More generally, given any (smooth) flow ’ℝ
surface. In this case, Poincaré maps are piecewise on S, not necessarily preserving a smooth invari-
isometries of an interval, known as interval ant measure, first return maps T : I ! I are one-to-
exchange transformations (or for short, IETs): a one piecewise diffeomorphisms known as gener-
one-to-one map T : I ! I of I ¼ [0, 1) is a alized interval exchange transformations: a map
(standard) interval exchange transformations of T : I ! I is a generalized interval exchange trans-
d  2 intervals, or d-IET for short, if one can formations or, for short, a GIET (Fig. 8a), if one
partition I into intervals I1, . . ., Id so that the can partition I into intervals I1, . . ., Id (finitely
restriction Ti of T to Ii, for each 1  i  d, is an many since we are assuming that ’ℝ has finitely
translation. Thus, for any Ii there exists many fixed points) so that the restriction Ti of T to
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 341

Ergodic and Spectral

Theory of
Area-Preserving Flows
on Surfaces, Fig. 8 A
generalized IET (GIET) and
a (standard) IET with d ¼ 4

Ii, for each 1  i  d, is a (smooth)

diffeomorphism onto its image which extends to
a diffeo of the closure I i (see, e.g., Marmi et al.
(2012)), as shown in Fig. 8. Thus, IETs are a
special case of GIETs (from which the use of the
adjective generalized) and provide natural linear
model of a GIET. Under the assumption that ’ℝ
preserves a smooth (non-atomic) finite invariant
measure, one can always choose suitable coordi-
nates in which the Poincaré section is a standard
Ergodic and Spectral Theory of Area-Preserving
IET, i.e., there is a (smooth) reparametrization of Flows on Surfaces, Fig. 9 Special representation of a
the section such that the GIET Poincaré section linear flow over IET with d ¼ 4
becomes a standard IET. In general, the lineariza-
tion problem of characterizing the GIETs which over T under the roof function r is the flow
are (smoothly) conjugated to standard IET is an
ℝ ≔ T t t  ℝ acting on
fT,r r
open research area; see Ulcigrai (2022) for a
survey. I r ≔fðx, sÞ  I  ℝ : 0  s < r ðxÞg,

ðnÞ
Special flows representations Linear flows and t ðx, sÞ ¼ x, s þ t  r ðxÞ , where
so that fT,r
(n)
minimal components of area-preserving flows can r (x) denotes the Birkhoff sums cocycle, i.e.,
be conveniently described using special flows. the additive cocycle defined by
These representations provide a concrete tool to
work with them and study their fine ergodic and r ðnÞ ðxÞ≔ r Tkx if n  0,
0k<n
spectral properties. Given a Poincaré map of a
flow to a section and the additional information r ðnÞ ðxÞ≔  r T k ðxÞ if n < 0,
of the return time to the section, special flows nk<0

allow reconstructing (a measurable conjugated associated to r and n is the unique integer number
version of) the flow. We recall the construction with r(n)(x)  s þ t < r(nþ1)(x). It describes the
for our special setting. motion of a point in (x, s)  Ir  I  ℝ along
Let T : I ! I be an (ergodic) IET and let vertical trajectories, modulo the identification of
r : I ! ℝ>0 [ {+1} be an integrable function each point (x, r(x)), x  I, with the point (Tx, 0);
such that r ¼ inf x  I r ðxÞ > 0: The special flow see Fig. 9.
342 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

Special representations of linear flows Linear mixing) was Arnold. In Arnold (1991) he states
flows on translation surfaces can be represented that what we nowadays call Arnold flow, i.e., a
as special flows over IETs, under roof functions minimal component of a locally Hamiltonian flow
which are piecewise constant: more precisely, the on a torus with a center and simple saddle, can be
return time r : I ! ℝ>0 of a linear flow on a represented as a special flow over a rotation
translation surface S to a transverse section I  S Rα : [0, 1] ! [0, 1] under a roof function of the
is constant on each continuity interval Ii of the IET form
T ¼ T(π,l) : I ! I induced as Poincaré map; see
Fig. 9. Then the roof function can be identified r ðxÞ ¼ C j logðxÞ j þ2C j logð1 xÞ j ð4Þ
with a vector r  ℝd>0 : As noted by Veech (1982b),
the vector r is in a 2 g-dimensional subspace for some C > 0. This function is said to have
H(π)  ℝd, where g is the genus of the surface S. logarithmic singularities at the endpoints of
Smooth time-changes of the linear flow pro- [0, 1]; see Fig. 10b. The singularities are further-
duce flows which can still be represented as spe- more called asymmetric, since the constants
cial flows over T, under a roof r which is piecewise 2C 6¼ C are different. The singularities at x ¼
smooth on each Ii and extends smoothly to its 0 and x ¼ 1 correspond to the trajectory which is
closure; see Fig. 10a. a separatrix, i.e., ends in the saddle. Because of the
nature of the Hamiltonian local parametrization of
Special representation of Arnold flows It is well a saddle, it takes infinite time to reach the saddle
known that (minimal components of) locally and the motion along trajectories which are close
Hamiltonian flows can be represented as special to the separatrix is slowed down logarithmically;
flows over IETs, but in this case the function r is see, e.g., (Frączek and Ulcigrai 2012), [Appendix
singular, i.e., blows up at (some) endpoints of A] for a calculation. The fact that the constant 2C
continuity intervals (see, e.g., Ulcigrai (2011), is exactly the double of the other constant C can be
Ravotti (2017), Conze and Frączek (2011), and explained because trajectories on one side of the
Frączek and Ulcigrai (2012)). The nature of the saddle loop pass twice near the saddle, while on
singularities of r turns out to depend crucially on the other side only once.
the nature and type of fixed points of the surface
flow. The first to remark on the importance of such Special representation of locally Hamiltonian
representation and the nature of the singularities flows with simple saddles Singularities which
for the study of chaotic properties (in particular are simple saddles (standard saddles with

Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces, Fig. 10 Special representation of flows
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 343

4-prongs, as in Fig. 4b) produce singularities of the singularities. The number of exchanged intervals
roof function as several singularities which, like in is d ¼ 2g þ s  1 in the case when ’ℝ is minimal
the case of Arnold flows, have logarithmic nature and s is the number of simple saddles, or, for a
in the following sense. We say that a function minimal component S0, d ¼ 2g0 þ s0  1, where g0
r : I ! ℝ for an IET T(π,l) has logarithmic singu- is the genus of S0 and s0 is the number of (simple)
larities if there exist constants Cþ 
i , Ci  ℝ, i ¼ saddles in the closure of S0.
1, . . . , d, and a function g’ absolutely continuous We say that the logarithmic singularities are of
on the interior of each interval Ii, i ¼ 1, . . ., d (i.- geometric type if at least one among C d and
e., with the notation that we will introduce later, a C 1
p ðd Þ is zero and at least one among C þ
1 or
function gr  AC di¼1 I i such that Cþ
p1 ð1Þ is zero (as shown in the examples in
Fig. 11). We denote by LG di¼1 I i the space of
d
r ðxÞ ¼  Cþ functions with logarithmic singularities of geo-
i logðjIjfðx  li Þ=jIjgÞ
i¼1 metric type. One can furthermore show that the
d roof functions arising from suitably chosen spe-
 C
i logðjIjfðr i  xÞ=jIjgÞ þ gr ðxÞ cial representations (namely when the section is
i¼1 standard, i.e., both endpoints belong to saddle
ð5Þ separatrices) are of geometric type. This notion
plays a crucial role in some results on locally
See Fig. 11 for an example. Consider now either a Hamiltonian flows; see, e.g., (Frączek and
minimal locally Hamiltonian flow ’ℝ on S or the Ulcigrai 2012, 2021).
restriction of a locally Hamiltonian flow on S to a
minimal component S0  S. Let  be the associated Symmetric and asymmetric logarithmic
closed 1-form and assume that  is Morse (the singularities When all the saddles are simple,
corresponding local Hamiltonian is a Morse func- and hence the roof has logarithmic singularities,
tion), so all saddles of the flow ’ℝ are simple. it is crucial, for understanding finer chaotic fea-
Then ’ℝ can be shown to be (measure theoreti- tures, to distinguish between symmetric or asym-
cally) isomorphic to a special flow Tr : Ir ! Ir over metric singularities, in the following sense. Let
an interval exchange transformation T : I ! I of LSG tdi¼1 I i be the subspace of functions with
d  1 intervals and under a roof r with logarithmic geometric type logarithmic singularities
LG tdi¼1 I i which in addition satisfy the symme-
try condition

d d
C
i  Cþ
i ¼ 0: ð6Þ
i¼1 i¼1

When the locally Hamiltonian flow is minimal on

S and has only simple saddles, the representation
leads to a symmetric roof (see, e.g., Kočergin
(1976) and Ulcigrai (2011)). On the other hand,
when the flow ’ℝ has centers or more in general
saddle loops homologous to zero (i.e., which dis-
connect the surface, as the saddle loop surround-
ing a periodic island in Fig. 4a), minimal
Ergodic and Spectral Theory of Area-Preserving components lead to representations with asym-
Flows on Surfaces, Fig. 11 Example of roof with geo- metric roofs. The prime example of an asymmet-
metric logarithmic singularities ric special flow is the roof (4) of the special
344 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

representation of an Arnold flow, while minimal base, hence the name. Von Neumann flows also
flows with only nondegenerate saddles give roofs admit an interpretation as special representations
with symmetric logarithmic singularities; cf. also of area-preserving surface flows. Consider a
Blokhin flows. locally Hamiltonian flow which is non-minimal:
assume that there are saddle connections which
Degenerate saddles and power-singularities - separate the surface into subsurfaces with bound-
Consider now degenerate saddles (i.e., saddles ary, each of which is a periodic or a minimal
with 2 k pronges, k>2; see Fig. 4c for k ¼ 3). component. Note that due to the presence of
These saddles produce stronger, power-type sin- fixed points, the transition times for the flow
gularities of the roof function; see Kočergin along the saddle connections are infinite. Using a
(1975). We say that li or ri is a power-like singu- reparametrization of the flow that results in sig-
larity of power-type 0 < γ < 1 if there exists a nificant acceleration around fixed points,
constant Ci such that described in Frączek and Lemańczyk (2009b),
we obtain a flow for which all transition times
lim ðx  li Þg r ðx  li Þ ¼ Cþ along saddle connections are finite. Then, each
i ;
x!lþ minimal component of such reparametrization
i
ð7Þ
lim ðr i  xÞg r ðr i  xÞ ¼ C
i : admits a representation as a von Neumann flow.
x!ri
Furthermore, the sum of jumps of the roof is
related to the sum of transition times (summed
Thus r behaves like the functions Ci =xg in a
according to orientation) along the saddle connec-
neighborhood of li or ri. As a degenerate special
tions forming the boundary of the minimal com-
case, a stopping point (fake saddle) introduces a
ponent (cf. Conze and Frączek (2011)).
single symmetric power-singularity of the roof
function.
Open sets and typical behavior In order to
Time-changes through special representa- describe the dynamical behavior of a generic or
tions One can show that two special flows over typical (area-preserving) surface flow, one can
the same transformation T under two different introduce the following topologies and measures
roof functions are one a time-change of the or measure classes (see below) on the space of
other. Thus, comparing the special representations IETs, linear flows, and locally Hamiltonian flows
results recalled in this section for linear flows and and distinguish between open sets with different
minimal (components of) locally Hamiltonian typical dynamical properties.
flows, one can show that the latter are time-
changes of translation flows via a singular Almost every IET and almost every
reparametrization (which amount to changing an linear flow Translation surfaces and their linear
r which is piecewise constant to one which has flows, as well as their Poincaré sections, IETs, are
singularities). described by finite-dimensional spaces. We say
that a result holds for almost every d-IET on a
Von Neumann flows Another class of special unit interval if it holds for all irreducible data π on
flows over IETs which has been studied in the d-symbols and almost every length vector l in the
literature are so-called von Neumann flows. The simplex Δd; see (3). Since linear flows can be
name is used to denote special flows over rotations represented as special flows over IETs under
or IETs under a function r which is piecewise piecewise constant roof functions, it is sufficient
absolutely continuous (and continuous on each to add 2 g data (related to the values of the return
continuity interval of the IET) with non-zero time on each continuity interval). A result holds
sum of jumps, i.e., r0(x)dx 6¼ 0; see Frączek and for almost every linear flow if it holds for a.e. IET
Lemańczyk (2009a). These were first studied by in the base and a.e. choice of r  ℝd>0 \ H ðpÞ:
von Neumann (1932) in the case of rotations in the Translation surfaces (up to cut and paste
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 345

equivalence) can be locally parametrized by M  F of Morse closed 1-forms, i.e., forms

period coordinates, namely by the (holonomy) which are locally the differential of a Morse func-
vectors associated to pairs of identified sides (see tion, is open and dense in F with respect to this
also, e.g., Zorich (2006), Yoccoz (2010), and topology (see, e.g., Ravotti (2017)). Thus, a
Viana (n.d.)) for a more intrinsic definition using generic locally Hamiltonian flow (in the sense of
relative homology). Baire) has only nondegenerate fixed points, i.e.,
These finite-dimensional coordinates allow centers and simple saddles (see Fig. 4a, b), as
defining the so-called Masur-Veech measure. Per- opposed to degenerate multi-saddles (as in
haps the simplest way to define the Masur-Veech Fig. 4c).
measure is to consider a presentation of a translation
surface S as a polygon with 2 N sides with pairs of
Katok fundamental class and measure class Let
parallel, congruent sides v1 , v01 , . . . , vN , v0N
us fix an open set A s,c of closed 1-forms with
identified by gluings. Then the vectors c centers and s (simple) saddles. A measure-
(v1, . . ., vN)  ℝ2N give local coordinates for an theoretical notion of typical on A s,c can be defined
open set U of translation surfaces around S and the on each A s,c by using the Katok fundamental
Masur-Veech measure is just the Lebesgue measure class, introduced by Katok (who was motivated,
on U  ℝ2N. More formally, the Masur-Veech mea- as he explains in Katok (1973), by previous work
sure is the pull-back of the Lebesgue on ℝ2n by the of Aranson and Grines, who had previously intro-
period map which maps S to ℝ2n where n is the duced in Aranson and Grines (1973) a homotopic
dimension of the relative homology group topological invariant for flows on oriented sur-
H1(S, S, ℝ). Since the vertical flow on a translation faces) in Katok (1973). Let S be the set of zeros
surface can be represented as a special flow over an (fixed points) of . Let γ1, . . ., γn be a base of the
IET under a piecewise constant roof, a result which relative homology H1(S, S, ℝ), where n ¼ 2g þ
holds for the vertical flow on a.e. translation surface s þ c  1. The image of  by the period map Per is
with respect to this measure holds for a.e. linear flow
in the sense above. Notice though that to reconstruct PerðÞ ¼ g1 , ..., gn   ℝn : The map Per is
the geometry of the translation surface, in addition well defined in a neighborhood of  in A s,c and
to the IET and the roof function, one needs also one can show (using Moser’s so-called homotopy
additional data (see, e.g., the zippered rectangle trick; see Katok (1973) and also Prop. 2.7 in
construction introduced by Veech (1982b); see also Ravotti (2017)) that it is a complete isotopy
(Yoccoz 2010)). invariant. The pull-back Per Leb of the Lebesgue
measure class (i.e., of the equivalence class of
Genericity notions for locally Hamiltonian measures which have the same sets of measure
flows Let us denote by F the set of smooth closed zero as the Lebesgue measure) by the period map
1-forms on S (i.e., locally Hamiltonian flows) with gives the desired measure class on closed 1-forms
isolated zeros. One can define a topology on F by in A s,c : When we use the expression typical below
considering perturbations of closed smooth (or typical in U min or U : min) we mean full measure
1-forms by (small) closed smooth 1-forms: we in each A s,c with respect to this measure class on
say that a smooth, closed 1-form 0 is an ϵ-pertur- each A s,c (or on each open subset of A s,c contained
bation of the smooth closed 1-forms  if for any in the union U min or U : min). If one can show that a
x  S there exist coordinates on a simply result holds for almost linear flow (or for the
connected neighborhood U of x, such that |U ¼ vertical flow on almost every translation surface
dH and (0  )|U ¼ dh where k hkC1 < ϵ k with respect to the Masur-Veech measure), then
HkC1 : We then say that a condition is generic it follows that it holds for a full measure set
(in the sense of Baire) if it holds for flows of locally Hamiltonian flows with respect to the
described by an open and dense set of forms Katok fundamental class (see, e.g., Ravotti
with respect to this topology. The subset (2017)).
346 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

The open sets U min and U : min To classify symmetry condition (6). For a typical flow in
chaotic behavior in locally Hamiltonian flows it U : min , the restriction of the flow on each minimal
is crucial to distinguish between two component admits a representation as a special
(complementary, up to measure zero) open sets. flow over a minimal IET under a roof with asym-
Remark that if the flow ’ℝ given by a closed metric logarithmic singularities (like in the case of
1-form  has a saddle loop homologous to zero the Arnold flow; see (4)). See, for example,
(i.e., the saddle loop is a separating curve on the (Ravotti 2017).
surface), then the saddle loop is persistent under
small perturbations (see Sect. 2.1 in Zorich (1999)
or Lemma 2.4 in Ravotti (2017)). In particular, the Invariant Measures and (Unique)
set of locally Hamiltonian flows which have at Ergodicity
least one saddle loop is open and gives the set
denoted U : min above. Flows in the open set Let us discuss the existence and finiteness of
U : min admit a non-trivial decomposition into ergodic measures, (unique) ergodicity, and non-
periodic components and minimal components. uniquely ergodic examples.
The second open set, which we call U min , is
given by the interior (which one can show to be
Invariant measures: existence and
non-empty) of the complement of U : min , i.e., the
finiteness Given a (topological) flow ’ℝ on S,
set of locally Hamiltonian flows without saddle
since we are assuming that S is compact, the
loops homologous to zero. One can show that
existence of a finite (probability) invariant mea-
saddle loops non-homologous to zero (and saddle
sure is guaranteed by the Krylov-Bogolybov
connections) vanish after arbitrarily small pertur-
theorem. Katok shows in (Katok 1973) that any
bations and neither the set of 1-forms with saddle
topologically transitive surface flow with non-
loops non-homologous to zero (or saddle connec-
degenerate (Morse) saddles has a probability
tions) nor its complement is open (see Ravotti
invariant measures with non-trivial support and
(2017) for details).
no atoms, positive on open sets.
When S is endowed with a reference smooth
Full measure of minimality In view of Keane’s
area form o and ’ℝ is assumed to be smooth
criterium for minimality (Keane 1975), one can
(at least outside the singularity set), it is natural
show that the set of non-minimal IETs has mea-
to ask about the existence of a (finite) invariant
sure zero and, furthermore, Hausdorff
measure which is absolutely continuous with
codimension 1. In view of the definition of mea-
respect to o and in particular of measures which
sure class on locally Hamiltonian flows, we there-
have a smooth (or at least differentiable) density.
fore deduce that:
If the flow can be linearized, i.e., conjugated to a
linear flow, via a differentiable or smooth
Theorem 1 In U min , the typical flow is minimal
conjugacy, such measures can be obtained by
(in particular there are no centers and there is a
pull-back of the Lebesgue measure via the
unique minimal component) and the typical flow
conjugacy.
on U : min is minimal when restricted to each
component which is not a periodic component
(bounded by saddle loops homologous to zero). Linearization questions and results Questions
about linearization and regularity of the
Special flows representations in U min and U : min conjugacy are hard and largely open. Marmi
et al. showed in Marmi et al. (2012) that for almost
The typical locally Hamiltonian flow in U min every linear flow ’ℝ on S, for any r  2, among
admits a representation as a special flow over a C rþ3 -perturbations supported outside the singu-
(minimal) IET under a roof with logarithmic sin- larity set S  S, those which are linearizable with
gularities (see (5)) and the roof satisfies the a C r conjugacy form a submanifold of finite
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 347

codimension (g  1)(2r þ 1) þ s, where g denotes A sharper bound was proved by Katok (1973),
the genus of the surface and s denotes the number who showed that the number of invariant mea-
of singularities. In Marmi et al. (2012), it is sures is at most g  [d/2], where g is the genus of
conjectured that for r ¼ 1 those which are C1- S (or of the minimal component S0). The proof of
linearizable form a submanifold of codimension the dimension bound uses the idea of Katok fun-
3g þ s  3. This was proved by Ghazouani (2021) damental class and the symplectic structure on the
for a measure-zero special case (hyperbolic absolute homology H1(S, ℝ) given by the inter-
periodic-type IETs). section form; see also Forni (2002) and McMullen
In very recent work, Ghazouani and Ulcigrai (2020).
(2021) have proved a rigidity theorem for folia-
tions associated to smooth flows with Morse sad- First non-uniquely ergodic examples Katok’s
dles on surfaces (at the moment only of genus 2): upper bound is optimal, as was shown by Sataev
for almost all (a.a.) smooth flows with respect to (1975), who constructed examples of flows on
the Katok fundamental class, if the corresponding S with a cone of invariant measure of arbitrary
foliation is topologically conjugate to an dimension less or equal to the genus g. An exam-
orientable measured foliation of that fundamental ple of a minimal non-uniquely ergodic IET with
class, then it is C1 conjugate. This result confirms d ¼ 4 was discovered by Keane (1977) as a count-
(in genus 2) another conjecture of Marmi erexample of an early form of the so-called Keane
et al. (2012). conjecture (see below) which stated that all min-
imal interval exchanges are (uniquely)
Bounds on the number of ergodic invariant ergodic. Earlier examples also appeared in the
measures It is a general fact that finite (resp. work of Katok and Stepin (1966) and an example
probability) invariant measures form a cone with d ¼ 5 was built by Keynes and Newton
(resp. a simplex) generated by ergodic measures. (1976) building on Veech (1969). A modern
In our setting, the number of independent ergodic revisitation of Keane’s counterexample (which
probability measures turns out to be finite: indeed, was produced with an ad-hoc self-induction pro-
as observed by Oseledec (1966), the number of cedure) can be obtained using the tools given by
independent ergodic invariant measures for a Rauzy-Veech induction (which were developed
d-interval exchange transformation T is bounded only later). Exploiting this point of view, in the
d (see, e.g., Yoccoz (2010) for a proof). In fact, lecture notes (Yoccoz 2010), examples of minimal
Oseledets estimated the maximal spectral multi- non-uniquely ergodic IETs are produced in any
plicity of any d-IET from above by d. This can be Rauzy class.
seen by showing that a d-IET has rank d. Another question which has been investigated
Let us assume that ’ℝ preserves a smooth recently for non-uniquely ergodic IETs is
invariant probability measure and is minimal genericity, i.e., the existence of an (a fortiori
(or restrict it to a minimal component S0  S). non-uniquely ergodic) IET T and a point x  I
By considering the (minimal) interval exchange whose orbit under T equidistributes for a measure
transformations which appear as Poincaré sec- that is not ergodic. A d-IET with these properties
tions of ’ℝ to sections (after linearizing the has been constructed by Chaika and
GIET exploiting the invariant measure induced Masur (2015).
by the smooth invariant measure for ’ℝ), one
can deduce from the IETs upper bound that the Keane conjecture and typical unique
number of independent ergodic invariant proba- ergodicity Keane conjectured in Keane (1977)
bility measures for ’ℝ is also at most d, where d is that almost every IET is uniquely ergodic. Keane
the number of exchanged intervals and hence can conjecture (as it became known) was proved in
be chosen (by taking the section to be normal, i.e., 1982 at the same time by Masur (1982) and Veech
with endpoints on separatrices) to be d ¼ 2g þ (1982b). Both proofs exploit renormalization and
s  1 where g is the genus of S and s is the introduced seminal ideas and techniques and are
cardinality of fixed points. considered early milestones of the successful
348 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

application of Teichmüller dynamics to the study Theorem 3 Almost every locally Hamiltonian
of IETs and translation surfaces; see, e.g., the flow in U min is not only minimal, but also
surveys Chaika and Weiss (2022) or Zorich ergodic. For almost every flow in U : min , the
(2006). While Veech works directly on IETs restriction of the flow on each minimal component
developing an induction algorithm known as is ergodic.
Rauzy-Veech induction, Masur shows that almost Due to the presence of saddle fixed points,
every linear flow (and hence a.e. IET) is uniquely locally Hamiltonian flows (their restrictions to
ergodic by using geometric arguments and the minimal components) on higher genus surfaces
Teichmüller geodesic flow on the moduli space are never uniquely ergodic.
of translation surfaces. Later in 1992, Masur pro-
ves in Masur (1992) the celebrated Masur’s crite- Hausdorff dimension of non-uniquely ergodic
rium for unique ergodicity, which shows that directions Both Masur (1992) and Masur and
unique ergodicity follows from the assumption Smillie (1991) study the exceptional set NU ðSÞ of
that the Teichmüller geodesic starting from the directions on a given translation surface S which
given translation surface is not divergent in mod- are minimal but fail to be uniquely ergodic. They
uli space. show that not only this set has measure zero by the
proof of Keane’s conjecture, but it is also small in
Ergodicity everywhere, in almost every the Hausdorff dimension point of view:
direction An important strengthening of Masur
and Veech results was later proved by Kerckhoff Theorem 4 For every translation surface S, the
et al. (1986): Hausdorff dimension of the set NU ðSÞ of non-
uniquely ergodic directions does not exceed 1/2.
Theorem 2 For every translation surface, the Furthermore, for any connected component C in
linear flow in almost every direction in uniquely the moduli space of translation surfaces with
ergodic. g > 1 there is a constant c > 0, called the
This result covers in particular billiards in Masur-Smillie constant of component C, such
rational polygons, which can be unfolded that for almost every translation surface S  C
(according to the Zemljakov & Katok construc- the Hausdorff dimension of NU ðSÞ is exactly c.
tion in Zemljakov and Katok (1975); see also Fox Sporadic examples of non-uniquely ergodic
and Kershner (1936)) to linear flows on interval were found, as already mentioned, by sev-
(a measure zero set of) translation surfaces. eral authors before the proof of Keane’s conjecture
Keane (1977); Keynes and Newton (1976).
Consequences for locally Hamiltonian
flows The first examples of smooth flows on sur- The slit-tori g 5 2 example A much studied
faces which are ergodic were the so-called example of a translation surface where one can
Blokhin examples (see Blohin (1972)). There are produce and study the exceptional set NU of non-
measure zero examples since they are essentially uniquely ergodic directions is the surface of genus
glued out of genus one subsurfaces. Since ergo- two obtained by gluing two identical flat tori
dicity of a (minimal component of a) locally Ham- along a slit; see Fig. 12. This surface, introduced
iltonian flow is equivalent to ergodicity of (and by Masur and Smillie, can be used to give a
hence any) interval exchange transformation geometric presentation related to previous work
which appears as the Poincaré map, result on of Veech (1969) on skew products over rotations.
ergodicity of typical locally Hamiltonian flow Cheung showed in Cheung (2003) that the upper
can be deduced from the proof of Keane’s conjec- bound of 1/2 for the Hausdorff dimension of NU
ture by Masur and Veech and the relation between is achieved in these examples. A full dichotomy
the two notions of full measure. We obtain: was later proved by Cheung et al. (2011) that
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 349

also Hausdorff codimension 1/2). The result about

IETs is used in Chaika and Masur (2020) to prove
that the Masur-Smillie constant is 1/2 for hyper-
elliptic components in the moduli space.

Behavior of Ergodic Averages

Given an ergodic area-preserving flow ’ℝ (or its

restriction to an ergodic minimal component
S0  S) and a real-valued observable f, let
IT( f, x) denote the ergodic integral
Ergodic and Spectral Theory of Area-Preserving
Flows on Surfaces, Fig. 12 Example a slit torus T
I T ðf, xÞ≔ f ð’t ðxÞÞdt: ð8Þ
0
showed that the Hausdorff dimension HDðNU Þ is
either 0 or 1/2 (according to a Diophantine-like
By (unique) ergodicity, if f has zero-mean,
property on the length of the slit). Athreya and 1
Chaika (2015) showed that in the stratum of trans- T I T ðf, xÞ decay to zero for a.e. x  S (and all
x with an infinite trajectory if f is continuous and
lation surfaces with one singular point of multi-
f is zero in a neighborhood of the fixed points). If
plicity 2 the Masur-Smillie constant is 1/2.
S is a torus, for a.e. linear flow (as well as
a.e. minimal Hamiltonian flow) IT( f, x) is
Hausdorff dimension estimates in parameter
bounded and f is a coboundary (see below). Two
space Instead of fixing a translation surface
distinctive features of these ergodic integrals for
S and studying the set NU ðSÞ, one can also
higher genus surfaces, i.e., if g  2, is that there
study the set NU ðd, pÞ IETs of d  2 intervals
are obstructions for IT( f, x) to be a coboundary
rearranged by π which are non-uniquely
and that the oscillations are of polynomial nature.
ergodic. Notice that for d ¼ 2, 3 (since all irratio-
nal rotations are uniquely ergodic by Kronecker
theorem) this set is countable and coincides with Obstructions and cohomological equations -
the set of non-minimal IETs. Understanding when the integrals IT( f, x) are
Athreya and Chaika (2015) considered 4-IETs bounded can be translated in a problem of solu-
(with irreducible combinatorics π) and proved that tions of cohomological equations.
the subset NU ð4Þ of 4-IETs which are not
uniquely ergodic has Hausdorff dimension 5/2. The cohomological equation for smooth
The construction of a rich subset of non-uniquely flows Let X be infinitesimal generator of the
ergodic IETs in Δ4 is a far-reaching generalization smooth flow ’ℝ preserving the smooth area m.
of Keane’s non-uniquely ergodic 4-IET original Let us recall that a mean zero observable
example. This result was generalized by Chaika f : S ! ℝ is said to be a coboundary for ’ℝ if
and Masur (2020) to any d  4: they showed that one can find a solution u : S ! ℝ of the cohomo-
if π belongs to the Rauzy class of the hyperelliptic logical equation Xu ¼ f. When u is bounded, this
permutation, then NU ðd, pÞ has Hausdorff dimen- forces the ergodic integrals IT(f, x) to be bounded.
sion d3/2. Since the simplex Δd in (3) has When g ¼ 1 and f is smooth, for a.e. linear flow f is
dimension d  1, this shows that the Hausdorff a coboundary. In a seminal paper Forni (1997),
codimension of NU ðd, pÞ is exactly 1/2 for any Forni discovered the existence, for g  2, of
d  4 (a result to be compared with the Hausdorff finitely many distributional obstructions for a reg-
dimension 1/2 of the set NU ðSÞ of non-uniquely ular f (in a suitable Sobolev space) to be a
ergodic directions on a given S, which in S1 has coboundary, given by g invariant distributions,
350 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

i.e., linear functionals D i , 1  i  g on Sobolev display polynomial oscillations, i.e., for almost
observables. By obstruction one means here that every initial point |IT( f, x)|  O(Tn) in the sense
if D i ðf Þ 6¼ 0 for some 1  i  g, then f cannot be a that
coboundary. The value of the first distribution is
logjI T ðf , xÞj
simply the integral of S fdm, so in genus g ¼ 1 this lim sup ¼n ð10Þ
T!1 log T
is the only condition and is automatically satisfied
for mean-zero observables. Furthermore, in Forni
(2021) he shows that for a.e. smooth minimal for some exponent 0 < n < 1. This phenomenon,
area-preserving flow when f belongs to the kernel known as polynomial deviations of ergodic aver-
of all D i , 1  i  g, f is indeed a coboundary. ages, was soon after explained in seminal work by
Kontsevich (1997) and Zorich (1997) relating
The cohomological equation for IETs Inspired power deviations to Lyapunov exponents of
by the work of Forni (1997) and Marmi et al. renormalization and proposed a finer description
(2005) considered the cohomological equation of polynomial deviations which became known as
over an IET T, rediscovered the obstructions in Kontsevich-Zorich conjecture.
this setting, and described a full measure condi-
tion (that they called Roth-type condition) that The Kontsevich-Zorich conjecture Kontsevich
guarantee that after removing the obstructions, & Zorich conjectured in Kontsevich and Zorich
the cohomological equation can be solved. More (1997) that the phenomenon of polynomial devia-
precisely, they show that for a Roth-type IET tions holds for the ergodic integrals IT( f, x) of any
T : I ! I, given an observable v : I ! ℝ, absolutely smooth observable (and not only those coming
continuous on each continuity interval Ii and with from cohomology classes, which can be reduced
the derivative of bounded variation and Iv0(x) to the study of Birkhoff sums over the IET obtained
as Poincaré section of the flow for an observable
dx ¼ 0 (zero sum of jumps), one can find
that is a characteristic function of a continuity inter-
w : I ! ℝ piecewise constant on each Ii and a
val, i.e., the setting originally considered in Zorich
continuous u : I ! ℝ such that
(1997)). They conjectured furthermore that ergodic
u ∘ T  u ¼ v  w: ð9Þ integrals display a power spectrum of oscillatory
behaviors, i.e., there are exactly g positive expo-
The function w, which belongs to the nents 0 < ng      n2 < n1 ≔ 1 (which corre-
d-dimensional space, provides the obstruction spond to the positive Lyapunov exponents of
for v to be a coboundary. In Marmi and Yoccoz’s renormalization) and, for each, a subspace of finite
(2016), it was later shown that u is Hölder contin- codimension of smooth observables that present
uous. Further generalizations of this result were polynomial deviations as above with exponent n ¼
proved by Marmi et al. (2020) (who show how to ni. Forni (2002) could prove the bulk of this con-
modify the proof to assume a weaker condition, jecture for linear flows on translation surfaces and
called absolute Roth-type); Forni et al. (2017) for integrals of sufficiently regular functions, by
(who deal with the case of some zero Lyapunov showing there are indeed g positive exponents.
exponents); and Lanneau et al. (2021) (who con- His results also apply to locally Hamiltonian
sidered the case of T linear involution). flows in U min and to sufficiently regular observ-
ables which vanish at the singularities. The simplic-
Power deviations of ergodic averages Zorich in ity of the spectrum, namely that the g exponents are
the 1990s discovered experimentally (by consid- all distinct, was later proved in Avila and
ering IETs and Birkhoff sums of characteristic Viana (2007).
functions of continuity intervals) that for almost
every linear flow in higher genus (g  2), the Bufetov functionals and limit shapes A finer
integrals IT( f, x) (in the special case of observ- analysis of the behavior of Birkhoff sums or inte-
ables corresponding to cohomology classes) grals, beyond the size of oscillations, appears in
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 351

Bufetov (2014). Bufetov (2014) shows in partic- the case of degenerate saddles and discovered the
ular that (for typical translation flows and suffi- existence of a new type of power deviations asso-
ciently regular observables) the asymptotic ciated to smooth observables that do not vanish at
behavior of ergodic integrals can be described in some degenerate saddles. The new exponents are
terms of g (where g is the genus of the surface) determined by the jet of the observable at the
cocycles Fi(t, x), 1  i  g (also called Bufetov degenerate singularity.
functionals): each Fi : ℝ  S ! ℝ is a cocycle
over the flow ’ℝ (in the sense that Fi(t þ s, x) ¼
Fi(t, x) þ Fi(s, ’t(x)) for any x  S and t  ℝ), Mixing Properties
F1(T, x) T and each Fi has power deviations
jFi ðT, xÞj  OðT ni Þ with exponent ni. Together, Finer chaotic properties such as mixing or weak
the cocycles encode the asymptotic behavior of mixing (as well as, more generally, spectral prop-
the ergodic integrals up to sub-polynomial behav- erties), crucially depend not only on the orbits of
ior, in the sense that for some constants ci ¼ ci( f ), the flow, but also on the speed of motion along the
orbits (i.e., the time-parametrization). In particu-
T lar, they are very different for linear flows and
f ð’t ðxÞÞdt ¼ c1 T þ c2 F2 ðT, xÞ þ . . . their time-changes and locally Hamiltonian
0
þcg Fg ðT, xÞ þ Errðf, T, xÞ, flows. For the latter, mixing or its absence
depends crucially on the type of singularities.
ð11Þ
Weak mixing in linear flows and IETs Weak
where for every x  S with regular orbit, the
mixing is an important and much investigated
quantity Err( f, T, x) is an error term which
question for linear flows and IETs, in view of the
grows subpolynomially, i.e., for any ϵ > 0 there
lack of mixing in this setting.
exists Cϵ > 0 such that
Absence of mixing Katok proved already in the
j Errð f , T, xÞ j Cϵ T ϵ :
1980s Katok (1980) that interval exchange trans-
formations are never mixing (generalizing earlier
Using these results, Bufetov could also prove
work with Stepin; see Katok and Stepin (1967),
some limit theorems for translation flows, in par-
[Remark 8.1] in the special case of 3-IETs). In the
ticular the convergence along (exponentially
proof, Katok shows (using a simple combinatorial
sparse) subsequences of the distribution of ergo-
argument which exploits that the induced map of
dic integrals of regular observables, when the
any d-IET on a subinterval is again an IET of at
initial point x  S is randomized (see Bufetov
most d þ 2-subintervals) that any d-IETs is par-
(2014) for statements).
tially rigid, i.e., there exists 0 < α < 1 (depending
on d only) and a diverging sequence of times (nk)k
Deviations phenomena due to singulari- such that for every measurable set A  I,
ties Frączek and Ulcigrai (2021) gave new proof
of the existence of a power deviation spectrum and LebðA \ T nk AÞ  a LebðAÞ, 8k  ℕ: ð12Þ
asymptotic cocycles for smooth observables over
locally Hamiltonian flows with Morse singularities From this it follows also that IETs have no mixing
which extends Bufetov-Forni results to smooth factor. In Katok (1980) it is furthermore shown,
observables which do not vanish at singularities exploiting partial rigidity of IETs, that any special
as well as flows in U : min: Their approach, inspired flow over an IET under a roof of bounded varia-
by Marmi et al. (2005), provides also a description tion is not mixing. This implies in particular that
of the full measure set of locally Hamiltonian flows linear flow, nor any smooth time-change of a
in terms of a Diophantine-like condition. Frączek linear flow, can be mixing. In the special case of
and Kim (2021) pushed a similar approach to treat special flows over rotations under smooth roofs
352 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

(or equivalently smooth time-changes of flows on IET, was given 10 years later by Avila and Forni
2 ) absence of mixing was proved earlier, in (2007). It exploits deep results from Teichmüller
Kočergin (1972). dynamics, in particular positivity of the second
Lyapunov exponent of the Teichmüller flow
Full measure of weak mixing Recall that the (Forni 2002) and a parameter exclusion argument,
ergodic theoretic property of weak mixing, by a which is very delicate for the case of IETs, but
classical result of Halmos (1960), is topologically much simpler in the case of linear flows (see the
generic while mixing is not. In Veech (1984) it Appendix of Avila and Forni (2007)) on
was conjectured that almost every IET should be a.e. translation surface.
weak mixing, unless it is essentially a rotation
(since rotations have eigenvalues and a discrete Weak mixing explicit examples The proof by
spectrum), i.e., if the combinatorial datum π is not Avila and Forni (2007) is not constructive.
that of a rotation with fake singularities. These Explicit examples of weak mixing IET
IETs are called of rotation type. (of periodic type, i.e., self similar) were
Veech (1984) introduced a condition on the constructed in Sinai and Ulcigrai (2005) and
combinatorial datum π of an IET (called W-con- Ferenczi and Zamboni (2011). Linearly recurrent
dition) under which he could prove full measure (or bounded-type) IETs are also explicit (and
of weak mixing. As an example, when d is odd, include in particular periodic-type IETs); there-
the π which exchanges the order, i.e., maps fore also linearly recurrent IETs with combinato-
i 7! d  i þ 1 for 1  i  d is of type W. Geo- rial datum π of type W provide explicit examples
metrically, the type W condition is essentially of weak mixing IETs by Boshernitzan and
equivalent to asking that the IET is a Poincaré Nogueira (2004).
section of a flow on a (minimal component of a) Translation surfaces obtained by unfolding of a
surface which has more than one saddle point. rational billiard (which constitute a measure zero
Boshernitzan and Nogueira (2004), [Theorem subset) on which the linear flow is weak mixing
5.3] showed that all linearly recurrent IETs for a.e. direction were constructed by Avila and
(or equivalently, bounded-type IETs, which have Delecroix (2016) (in the class of Veech surfaces).
measure zero) with combinatorial datum of type
W are weakly mixing. Notice that when π is not of Hausdorff dimension of non-weak
type W, bounded-type IETs do not have to be weak mixing Once it has been shown that weak mixing
mixing; see, for example, the examples of 4-IETs IETs or linear flows have measure zero, one can
produced in Hmili (2010): these examples have π study the Hausdorff dimension of the comple-
which is not type W and are not weak mixing, ment. The parameter exclusion argument in
since they have a nonconstant eigenfunction. Avila and Forni (2007) shows some estimates on
One can verify furthermore that they are linearly the Hausdorff dimension of translation surfaces
recurrent when the eigenvalue is badly for which the vertical linear flow flows fail to be
approximable. weak mixing. Avila and Leguil (2018) studied the
Nogueira and Rudolph could prove already in Hausdorff dimension of the set of non-weakly
the 1990s that almost every IET with π not of mixing IETs. They showed that its Hausdorff
rotation type is topologically weak mixing, i.e., dimension is strictly less than d  1, where d is
has no nonconstant continuous eigenfunction (see the number of exchanged intervals. It follows
Nogueira and Rudolph 1997) (in contrast with from Chaika and Masur (2020) that the Hausdorff
weak mixing, which is equivalent to the absence dimension is at least d  32 : Combining the work
of nonconstant measurable eigenfunctions). An of Chaika and Masur (2020) and Al-Saqban et al.
unrestricted proof of almost everywhere weak (2021), one can show that the set of non-weakly
mixing for IETs not of rotation type, both for mixing IETs with permutation of type W has
almost every linear flow and for almost every dimension d  32 (see Chaika and Masur (2020)).
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 353

Mild mixing in IETs and von Neumann roof function over an irrational rotation with
flows Mild mixing, which can be defined as the bounded partial quotients to be mild mixing.
absence of rigid factors, is an intermediate prop- Using this criterion, Frączek (2009) has shown
erty between weak mixing and mixing. Thus, it is that when the genus is at least two, the set of
a natural property to explore systems which are Abelian differentials for which the vertical flow
known to be weak mixing but not mixing. is mildly mixing is dense in every stratum of
moduli space. He used an approximation of any
Mild mixing IETs As shown by Veech (1984), a translation surface by surfaces having many ver-
typical IET is rigid, so it cannot be mildly mixing. tical saddle connections. Vertical flows on such
The existence of mild mixing IETs has been first surfaces have special representations over
shown among 3-IETs: Ferenczi et al. (2005), rotations.
[Theorem 4.1] showed that linearly recurrent
3-IETs are mild mixing (since they have a prop- Mixing properties of smooth flows In smooth
erty known as minimal self-joinings; see area-preserving flows, due to the presence of
Kanigowski and Lemańczyk (2020)). Robertson Hamiltonian saddles, the closer a trajectory is to
(2019) has considered IETs with combinatorics π a saddle singularity, the more motion along the
of type W (the condition under which Veech first trajectory is slowed down. This generates a shear-
proved weak mixing) and shown that when line- ing phenomenon that can create mixing. The pres-
arly recurrent (a measure zero but full Hausdorff ence of absence of mixing depends both on the
dimension condition), they are mild mixing. strength of the singularity (in particular if it degen-
erates or nondegenerates) and on the presence of
Von Neumann flows Von Neumann flows in traps (or more generally saddle loops homologous
genus one (i.e., special flows over rotations or to zero), i.e., the symmetry or asymmetry of the
IETs under a piecewise absolutely continuous roof in case of logarithmic singularities. For
function with non-zero sum of jumps) were intro- generic flows given by Morse forms one has
duced by von Neumann (with rotations as the indeed the following dichotomy:
base) in 1936 to produce examples of weak
mixing systems. He proved that such flows are Theorem 5 Inside the open set U min in which the
weakly mixing for each irrational rotation if the typical flow is minimal, almost every locally Ham-
roof function is piecewise C1. The weak mixing iltonian flow is weakly mixing, but it is not
property was then proved for the von Neumann mixing. On the other hand, for a full measure set
class of functions but over ergodic interval of flows in U : min , the restriction to each of min-
exchange transformations by Katok (2001). imal components is mixing.
Frączek and Lemańczyk (2006) showed that von This statement summarizes a number of results
Neumann flows over rotations (with a roof with (in particular Ulcigrai (2009, 2011) and Ravotti
non-zero sum of jumps) can be furthermore mild (2017)) which we now describe in detail.
mixing (but for rotation numbers α of bounded
type, which form a measure zero set). This result Diophantine-type conditions Results on mixing
was extended in Frączek et al. (2007) to roof properties require the introduction of
functions with zero sum of jumps. Both results Diophantine-like conditions, which describe the
exploit versions of the Ratner properties (see full measure set of locally Hamiltonian flows for
below). which the results hold. For g ¼ 1 these can be
expressed as properties of the entries of the con-
Mildly mixing linear flows As for IETs, the tinued fraction expansion of the rotation number.
Veech result (Veech 1984) says that the linear In higher genus, these are expressed in terms of
flow on a typical translation surface is rigid, so it the renormalization given by the Teichmüller flow
also cannot be mildly mixing. Frączek et al. or renormalization algorithms for IETs such as
(2007) gave a criterion for a piecewise constant Rauzy-Veech induction. For a survey of these
354 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

Diophantine-like conditions, we refer to the ICM motivated the question of mixing in locally Ham-
proceedings by Ulcigrai (2022). iltonian flows with only nondegenerate saddles
(i.e., simple saddles). Arnold in the 1990s noticed
Mixing of Kocergin flows and related a geometric phenomenon which could produce
results The first examples of mixing smooth mixing in the ergodic minimal components of
flows on higher genus surfaces were given already locally Hamiltonian flows on the torus known as
in the 1970s (in Kočergin (1975)) exploiting Arnold flows (Arnold 1991). His intuition was
degenerate saddles (i.e., multi-saddles with proved to be correct shortly after by Sinai and
k  6 prongs, as in Fig. 4c) since in this case the Khanin (1992), who showed mixing under an
saddles have a much stronger slowing down arithmetic condition of full measure on the rota-
effect. Kochergin showed that special flows over tion number α. The full measure assumption on
ergodic IETs under roofs with polynomial singu- the rotation number α is that the entries an of the
larities (see (7)) are mixing for all ergodic IETs on continued fraction expansion of α do not grow too
the base (in particular for all irrational rotation α in fast, namely there exists a power 1 < t < 2 and
the base in the g ¼ 1 case). This implies that a C > 0 such that |an|  Cnt. The condition was later
typical locally Hamiltonian flow is mixing if it has improved in Kochergin (2004).
degenerate saddles (as in Fig. 4c) which all give The question of whether mixing is typical also
rise to polynomial singularities of the roof func- for flows on higher genus is more delicate and
tion (e.g., for local Hamiltonians given by ℑ (zk) requires a generalization of the Diophantine-like
for k  2). condition in higher genus, which was introduced
As a special case, Kočergin (1975) result in Ulcigrai (2007) and proved to be of full mea-
applies to flows on 2 with a stopping point sure by Avila et al. (2006). Ulcigrai proved
(which can be seen as a fake degenerate saddle) (Ulcigrai 2007) that for a.e. IET, the special flow
that admit a representation as special flows over under an asymmetric roof with one logarithmic
rotations under a roof with a symmetric power singularities is mixing. Later, Ravotti (2017)
singularity. For this special case, polynomial showed that the same full measure condition can
mixing estimates were proven in Fayad (2001) also be used to extend the result to roofs with
for special observables when the power singular- several asymmetric logarithmic singularities, as
ity is sufficiently strong. in (4). This implies (see Ravotti (2017)) that in
While smooth time-changes of linear flows on the open set U : min , the restriction of the typical
2
 are never mixing by Katok (1980), a mechanism locally Hamiltonian flow ’ℝ on each of its mini-
similar to the one exploited by Kočergin (1975) mal components is mixing.
was used by Fayad (2002) to show that there exist In Ravotti (2017), quantitative results on the
smooth (actually analytic) time-changes of a flow speed of mixing are also proved: for a typical ’ℝ
on 3 which are mixing. Time-changes of flows on in U : min , restricted to a minimal component, the
3 can be seen as special flows over a two- speed of decay of correlations (also sometimes
dimensional rotation ℝa : 2 ! 2 , given by called speed of mixing) is sub-polynomial and
ðx1 , x2 Þ 7! ðx1 þa1 , x2 þ a2 Þ  2 : For mixing actually logarithmic, namely for every pair f,
smooth time-changes to exist, a should be a g of smooth observables there exist constants
Liouville vector and the frequencies α1, α2 should c > 0, α > 0 such that
have sequences of continued fraction denominators
that are intercalated with respect to each other. ðf ∘’t Þ  gdm  fdm  gdm  clogt a :
Notice that these are measure zero examples.

Arnold conjecture and mixing in the The role of shearing A crucial ingredient in the
asymmetric case Since the presence of a degen- proofs of these mixing results (and more generally
erate saddle is not generic, Kocergin work of mixing in parabolic flows) is played by a
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 355

geometric shearing phenomenon: if we consider a shearing on one side and hence, in this case, the
small arc γ transversal to the trajectories of the accumulation of shearing predominantly in one
flow ’ℝ, so that when flowing it, ’t(γ) passes direction produces global shearing.
nearby a saddle separatrix without hitting the sad-
dle point (as shown in Fig. 13), the different Ratner’s properties Striking consequences of
deceleration rates of points cause ’t(γ) to shear shearing (such as measure and joining rigidity)
in the direction of the flow (see Fig. 13a). Shear- were proved for another famous class of parabolic
ing allows to deduce mixing from ergodicity: by a flows, namely horocycle flows on hyperbolic sur-
Fubini-type argument, given measurable set faces and their time-changes, by exploiting a
A  X, for every large t one can cover an arbi- quantitative shearing property introduced by
trarily large proportion At  A of with a collection Marina Ratner and nowadays known as Ratner
of short transversal segments γ, each of which, property (or RP). The difficulty in proving Ratner-
after time t, shadows a long trajectory of ’ℝ and, type properties for smooth flows on higher genus
therefore, by (unique) ergodicity is (close to) surfaces is given by the presence of singularities
equidistributed. Furthermore, the speed of mixing (which are unavoidable when g  2), which intro-
depends on the speed of shearing. duce discontinuities and destroy the slow form of
The shearing accumulated can be later divergence a la Ratner: as soon as two nearby
destroyed when the segment ’t(γ) passes near trajectories are separated by hitting a saddle,
the other side of a saddle (see Fig. 13b). The indeed, one drastically loses control of the diver-
presence of a saddle loop, though, (as in gence. A first generalization of the Ratner prop-
Fig. 14a) typically creates an asymmetry (this erty (the finite Ratner property) was introduced by
was the key intuition of Arnold that had motivated Frączek and Lemańczyk (2006) and proved for a
his conjecture on mixing) by producing stronger special class of von Neumann flows (special under
a piecewise absolutely continuous roof with non-
zero sum of jumps over an irrational rotation of
bounded type). The finite Ratner property was
also proved for flows when the sum of jumps is
zero by Frączek et al. (2007). Notice that von
Neumann flows are not (globally) smooth. For
other examples of this type of Ratner property,
see also Frączek and Lemańczyk (2010).
A version of the Ratner property is also a key
ingredient in the proof in Kanigowski (2015).
The Ratner property in its classical form
Ergodic and Spectral Theory of Area-Preserving as well as the weaker versions defined in Frączek
Flows on Surfaces, Fig. 13 Shearing mechanism and Lemańczyk (2006, 2010) is expected to fail
for smooth area-preserving flows with non-
degenerate fixed points. The failure of the classi-
cal Ratner property was formally proved in a
special case in Fayad and Kanigowski (2016),
[Theorem 1] (for a class of Kochergin flows),
and this result gives reasons to believe that, for
similar reasons, the classical Ratner property
should indeed always fail in presence of
singularities.
A new variant of the RP which has the same
Ergodic and Spectral Theory of Area-Preserving dynamical consequences, called Switchable
Flows on Surfaces, Fig. 14 Mixing mechanism Ratner Property (or SRP for short), was
356 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

introduced by Fayad and Kanigowski (2016) and asymmetry which produces global shearing, one
proved to hold for almost every (minimal compo- can expect that the effect of shearing on two differ-
nent of) an Arnold flow in U : min when g ¼ 1 ent sides of the same saddles compensates and
(as well as for a measure zero set of Kochergin cancels out.
flows). According to this variation, it is sufficient
to see the Ratner divergence of orbits for most Symmetric logarithms over rotations Consider a
pairs of initial conditions (x, y) either in the future special flow over an irrational rotation Rα under a
(for t > 0) or in the past (for t < 0), depending on roof with a symmetric logarithmic singularity, i.e.,
the pair of initial points. Thus, if one pair of
nearby trajectories is separated by hitting a singu- r ðxÞ ¼ C j logðxÞ j þC j logð1 xÞ j : ð13Þ
larity, and hence their distance explodes in an
uncontrolled manner, one can still hope to switch Already in the 1970s, Kočergin (1976) proved
the direction of time (from which the name of that these flows are not mixing for a full measure
switchable Ratner property) and be able to prove set of rotation numbers α and, much more
the Ratner slow form of divergence when flowing recently, extended this result to all irrational rota-
backward in time. Kanigowski et al. (2019) gen- tion numbers (Kochergin 2007). The criterion for
eralize the result by Fayad and Kanigowski to the absence of mixing exploited in Kočergin
higher genus and show that for almost every (1976) can be seen as a generalization of Kočergin
’ℝ  U : min , the restriction of ’ℝ to any minimal (1972) absence of mixing result and shows that
component satisfies the Switchable Ratner Prop- (at least for typical) locally Hamiltonian flows
erty for any genus g  1. This also covers as mixing via shearing is essentially the only possi-
special case minimal components of typical ble way of achieving mixing. Indeed it was
flows in g ¼ 1 with more than one simple saddle. conjectured in Lemańczyk (2000) and proved by
In Fayad and Kanigowski (2016) the case of more Schmidt (2002) that mixing in special flows over a
saddles is also considered, but under a condition rotation is only possible if the distribution of the
on the relative position of the saddles which has sequence of Birkhoff sums of the roof function is
measure zero. not tight, i.e., Leb({x; |r(n)(x)| > K}) tends to 1 as
K tends to infinity.
Multiple mixing Applying joining techniques
introduced by Host (1991) and developed by Symmetric logarithms over IETs While the key
Ryzhikov and Tuveno (2006), the Switchable example studied by Kochergin (special flows over
Ratner property can be used in particular to rotations under a symmetric logarithmic singular-
show that mixing can be upgraded to mixing of ity) constitutes a prototype result for the absence
all orders. In particular, the results in Fayad and of mixing, these special flows do not arise as
Kanigowski (2016), combined with Sinai and representations of typical locally Hamiltonian
Khanin (1992), imply that typical Arnold flows flows, since special representations of typical
are mixing of all orders (as well as a measure zero minimal flows with only simple saddles always
set of Kochergin flows in genus one). Similarly, yield special flows under roofs with symmetric
the result of Kanigowski et al. (2019) combined logarithmic singularities over IETs with d  4.
with Ravotti (2017) shows that for a full measure Ulcigrai (2011) proved the absence of mixing for
set of locally Hamiltonian flows in U : min , each such flows for almost all interval exchange trans-
restriction to a minimal component is mixing of formations in the base. Thus, a.e. flow in U min is
all orders. not mixing. A special case of the absence of
mixing result for surfaces with g ¼ 2 and two
Absence of mixing in the symmetric case For isometric saddles was proved in Scheglov
minimal locally Hamiltonian flows in U min , which (2009); see also the recent work by Chaika et al.
have only simple saddles, since there are no saddle (2021) for a more geometric proof. The proof in
loops homologous to zero and hence no Scheglov (2009) exploits a combinatorial analysis
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 357

of the substitutions induced by Rauzy-Veech logarithmic singularities. Thus, a.e. flow in U min
induction to show that orbits satisfy a certain is weakly mixing but not mixing.
symmetry. In Chaika et al. (2021), this symmetry
is deduced geometrically from the existence of a
hyperelliptic involution. Spectral Properties
We remark that in U min there exist nevertheless
exceptional mixing flows (contrary to the case of We move now to the properties of the spectrum of
the torus in view of Kočergin (2007)), as shown in Koopman unitary operators and their spectral
Chaika and Wright (2019), where the authors measures. We recall that the Koopman unitary
could produce sporadic mixing examples in g ¼ operator associated to a dynamical system is
5. Also for this example, shearing is still at the given by f 7! f ∘ T for f  L2(I, dx) when T is
base of mixing, but it is not produced by asym- an IET or f 7! f ∘ ’t for f  L2(S, m) when ’ℝ is a
metry of the singularities, but rather by an asym- surface flow preserving m. We refer the reader to
metric equidistribution, so that trajectories, at the entry by Kanigowski and Lemańczyk (2020)
different time scales, spend much more time on or Katok and Thouvenot (2006) for a survey of
one side of a saddle than another. The IETs con- spectral theory of dynamical systems.
sidered are finite covers of rotations (for which
Diophantine-type conditions can be given in Spectral measures of IETs and linear
terms of continued fraction entries) which are flows The first general spectral result about
only barely uniquely ergodic and for which orbits IETs is Oseledets theorem in Oseledec (1966)
equidistribute very slowly and in a very stating that the maximal spectral multiplicity of
asymmetric way. any d-interval exchange transformation T is
More recently, Kanigowski and Kułaga- bounded by d. In 1984, Veech proved in Veech
Przymus (2016) showed (exploiting the former (1984), [Theorem 1.4] the a.e. IET is rigid which
work of Kułaga (2012)) that special flows with implies by standard arguments (see, e.g., Katok
symmetric logarithmic singularities over IETs of and Thouvenot (2006)) that T has singular spec-
bounded type are mild mixing (the assumption on trum. It follows furthermore from Avila and Forni
the IET used in Kanigowski and Kułaga-Przymus (2007) work on weak mixing that a.e. IET, as well
(2016) is not explicitly that the IET is of bounded- as a.e. linear flow, has continuous spectrum.
type, but a condition on orbits of discontinuities, Exotic examples of IETs with eigenvalues with
which can a posteriori be shown to be equivalent various properties were constructed by Ferenczi
to bounded type or linearly recurrent). The main and Zamboni (2011)); see also Hmili (2010).
component of the proof is to show that the flow Let w ¼ wA be the characteristic function of a
satisfies the Switchable Ratner Property. continuity interval A ¼ Ii of T (or, more generally,
of a floor of a cutting and stacking presentation of
Weak mixing via logarithmic singulari- T, as those given by Rauzy-Veech induction).
ties Frączek and Lemańczyk (2003) showed A condition for the spectral measure sw to be
that flows over all irrational rotations under any continuous at a point o  S1 was given by
roof functions with one symmetric logarithm are Bufetov et al. (2006) and used to produce explicit
weakly mixing. Generalizing this result, Ulcigrai examples of (periodic-type) IETs with continuous
(2009) also proved weak mixing for special flows spectrum in Sinai and Ulcigrai (2005).
with logarithmic singularities (not necessarily Sinai raised the question: to find modulus of
symmetric, although in the asymmetric case continuity for the spectral measures of translation
mixing is already known; see Ulcigrai (2007) flows. In Bufetov and Solomyak (2018, 2020),
and Ravotti (2017)) for almost every IET on the Bufetov and Solomyak developed an approach
base. The proof exploits the partial rigidity of IETs to this problem and succeeded in obtaining Hölder
proved by Katok (1980) as well as the presence of estimates for spectral measures in the case of
358 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

surfaces of genus 2. The final result for surfaces of surfaces of any genus g  2 with singular contin-
higher genus, independently proved in Bufetov uous spectrum (see Frączek and Lemańczyk
and Solomyak (2021) and Forni (2022), says (2003), [Theorem 1]) using Blohin (1972) con-
that for every stratum in the moduli space there struction. Since these are essentially built gluing
are γ > 0 and β > 0 such that for a.e. translation genus one flows and the resulting flow has a lot of
surface in the stratum there exists C > 0 such that saddle connections, they are highly non-typical.
for every f : S ! ℝ smooth enough
Lebesgue spectrum via a stopping point A per-
1þ j l j
b haps the most surprising result for minimal flows
sf ð½l  r, l þ r Þ  C k f k rg ð14Þ
jlj on the torus with one stopping point or fake sin-
gularity (see Fig. 15) was proved recently by
for every l 6¼ 0 and r > 0. A similar result for Forni, Fayad, and Kanigowski. In Fayad et al.
a.e. IETs is harder to prove; Avila, Forni, and (2021), they proved that if the degenerate singu-
Safaee estimated moduli of continuity of spectral larity is sufficiently strong, the spectrum is count-
measures for a.e. IETs in Avila et al. (2021b). able Lebesgue. In Kočergin (1975), Kocergin
flows admit a special flow representation where
Quantitative weak mixing The main tool leading the roof has power-type singularities. In genus
to the proof (14) is quantitative estimates of one, for a flow with one degenerate singularity,
so-called twisted ergodic integrals of the form one has a special flow over a rotation, under the
roof r(x) ¼ c0/xγ þ c1/(1  x)γ for some 0 < γ < 1.
1 T
1þ j l j b The assumption in Fayad et al. (2021) is that γ is
e2pilt f ð’t xÞdt  C k f k T g sufficiently close to 1. The absolute continuity of
T 0 jlj
the spectrum, taking as a starting point an idea
proved (in different forms) in Forni (2022), from Forni and Ulcigrai (2012) is proved from
Bufetov and Solomyak (2021), and Avila et al. estimate on the (polynomial) speed of decay for
(2021b). On the other hand, quantitative estima- natural observables, by showing that decay is
tions of twisted intergrals lead to an efficient ver- square-summable. Forni and Ulcigrai exploited
sion of weak mixing square-summable estimates on the decay of cor-
relations to prove in Forni and Ulcigrai (2012) that
T smooth time-changes of horocycle flows have the
1
jhf ð’t xÞ, gij2 dt  C k f k k g k T g Lebesgue spectrum, as conjectured by Katok and
T 0 Thouvenot (2006). An extra difficulty in this set-
ting is given by the symmetry of the power singu-
for f, g smooth enough zero mean functions; see
larity, which creates cancellations and parts of
Forni (2022) (and Avila et al. (2021b) for IETs).
space where there is no shearing. The authors
Spectra of locally Hamiltonian flows While the
classification of mixing properties of locally Ham-
iltonian flows is essentially complete (see Theo-
rem 5), there are few results on their spectral
properties for g  2.

First examples with singular spectrum Frączek

and Lemańczyk (2003) showed singularity of the
spectrum for special flows under a symmetric
logarithmic roof over a full measure set of rota-
tions (this follows from spectral disjointness from
all mixing flows). They also use this result to build Ergodic and Spectral Theory of Area-Preserving
examples of locally Hamiltonian flows on Flows on Surfaces, Fig. 15 g ¼ 1, stopping point
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 359

also introduce a criterium to show that the spec- are not isomorphic and, moreover, their factors
trum has countable multiplicity suited for para- are also not isomorphic. We refer to the survey
bolic settings. The criterium was also used in de la Rue (2020).
Fayad et al. (2021) to complete the proof of the
Katok-Thouvenot conjecture on smooth time- Disjointness from mixing systems For all sys-
changes of horocycle flows, by showing count- tems for which absence of mixing is known, one
able multiplicity of the Lebesgue spectrum. can try to strengthen the result by proving
disjointness from all mixing flows. The proof of
Singular spectrum in genus 2 In the opposite that no IET is mixing in Katok (1980) already
direction, Chaika et al. proved in Chaika et al. implies (even if this is not explicitly stated) that
(2021) that a typical minimal locally Hamiltonian IETs are disjoint from all mixing systems; see also
flow on a genus two surface with two isomorphic Ryzhikov (1994). In fact, partial rigidity (see (12))
simple saddles (Fig. 16) has purely singular spec- implies disjointness from mixing systems. The
trum. The result, inspired by the techniques used to proof by Katok also covers the case of special
prove the singularity result (for special flows over flows over IETs under piecewise constant roofs,
rotations) in Frączek and Lemańczyk (2003, 2004), which are also partially rigid. This strategy does
deduces singularity of the spectrum from absence not work for all roof functions of bounded varia-
of mixing and rigidity, exploiting the geometric tion. Frączek and Lemańczyk (2009a), [Appendix
symmetries given by the hyperelliptic involution. A] showed that von Neumann flows over IETs are
It also provides an independent proof of absence of never partially rigid.
mixing for typical flows in the same class (g ¼ 2, Frączek and Lemańczyk strengthened
two isomorphic saddles, see Fig. 16) proved in Kočergin results in Kočergin (1972, 1976) show-
Scheglov (2009). As in Scheglov (2009), the ing absence of mixing for flows over rotations (for
assumption that the saddles are isometric is crucial smooth roofs and symmetric logarithmic singular-
to guarantee that the underlying surface has an ities respectively) by proving in Frączek and
inner symmetry, which plays a crucial role in the Lemańczyk (2004) that for every irrational α, the
proof. More precisely, the linear flow of which the special flows over Rα under a smooth roof either
locally Hamiltonian flow is a time-change is a flow has bounded variation or whose Fourier coeffi-
on a translation surface S which admits a hyper- cients decay as O(1/n) (so in particular for a roof
elliptic involution, i.e., an affine automorphism with one symmetric logarithmic singularity) are
F : S ! S which is an involution, i.e., F2 ¼ Id. disjoint from all mixing flows. Spectral
disjointness was also proved in Frączek and
Lemańczyk (2003) for a.e. rotation and the roof
Disjointness Results function with one symmetric logarithmic singu-
larity. All these results are based on tightness of
The notion of disjointness was introduced in the the distribution of Birkhoff sums of the roof func-
1960s by Furstenberg (see in particular tion, and the study of their weak limits of joinings
Furstenberg (1967)). Recall that disjoint flows (via Markov operators) determined the graphs of

Ergodic and Spectral

Theory of
Area-Preserving Flows
on Surfaces, Fig. 16 g ¼
2, isomorphic saddles
360 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

’t in S  S. This approach was strengthened in rare property. As shown by del Junco (1981) (for
Frączek and Lemańczyk (2005), where automorphisms) and by Danilenko and Ryzhikov
disjointness from all mixing systems was proved (2012) (for flows), typically we have even
also for all special flows over ergodic IETs and disjointness with the inverse action. Although
under bounded variation roofs (in particular for all non-reversibility is a typical property, it is not
von Neumann flows over IETs) and for special easy to prove in concrete examples, especially
flows over a.e. rotation and under any roof func- for flows on surfaces. In general, proving the
tion with symmetric logarithmic singularities non-isomorphism of zero entropy and spectrally
(in particular for Blokhin flows in any genus). equivalent systems (a flow is always spectrally
Notice that the tightness proved by Ulcigrai isomorphic with its inverse) is a difficult task.
(2011) is sufficient to apply the techniques devel- One of the standard techniques here is
oped in Frączek and Lemańczyk (2005). Combin- disjointness, for which proving techniques are
ing these two results one has that a.e. locally better developed.
Hamiltonian flow in U min is disjoint from all The first result for surface flows was proved in
mixing flows. Frączek and Lemańczyk (2009a), where, using a
shearing mechanism and Ratner’s techniques, the
Joining properties of IETs As we have already authors showed that for every rotation Rα of
mentioned, every IET is partially rigid, so is dis- bounded type, every von Neumann flow is dis-
joint from all mixing systems. Moreover, a.e. IET joint from its inverse. This result was expended to
is rigid; see Veech (1984). Chaika (2012) proved a a.a. rotations in Frączek et al. (2014) and to
much stronger result about rigidity of a.e. IET, a.e. IET in Berk and Frączek (2015), but in both
saying that if (qn)n  ℕ is any increasing sequence papers only the non-isomorphism of the von Neu-
of natural numbers, then a.e. IET is rigid along a mann flow and its inverse is proved. The methods
subsequence of (qn)n  ℕ. This implies an amazing developed in Frączek et al. (2014) and Berk and
result that any ergodic system is disjoint from Frączek (2015) (do not rely on Ratner’s tech-
a.e. IET; see Chaika (2012), [Theorem 1]. niques) are inspired by Ryzhikov’s results and
Veech asked, already in the 1980s (see Veech involve the study of weak limits of graph
(1982a)), whether a.e. IET is simple, i.e., the only 3-joinings. They were further creatively devel-
(ergodic) self-jonings are graphs and the product oped in Berk et al. (2020), where they were allo-
measure. This question was motivated by the arti- wed to prove disjointness of von Neumann flows
cle by del Junco (1983), where he found a one- with their inverse for a.a. IETs.
parameter family of simple 3-IETs. Further exam- The problem of reversibility in a class of linear
ples of simple d-IETs with d  3 were produced in flows on translation surfaces is more complicated.
Ferenczi and Zamboni (2011). A surprisingly neg- Here the answer depends on connected compo-
ative response to Veech’s question was given by nent C in the moduli space. If the connected
Chaika and Eskin (2021) where they studied the component C is hyperelliptic, then every linear
set of joinings of a.a. 3-IETs and proved that it is flow on any translation surfaces in C is isomorphic
rather large. The set of self-joinings is a Poulsen to its inverse via the hyperelliptic involution. On
simplex, i.e., the set of ergodic joinings is dense. the other hand, for a typical (in the topological
sense) translation surface in any non-hyperelliptic
Non-reversibility A measure-preserving flow component C, the linear flow is disjoint from its
’ℝ ¼ (’t)t  ℝ is reversible if it is isomorphic to inverse; see Berk et al. (2020).
its inverse flow ’1
ℝ ¼ ð’t Þt  ℝ : It is also some-
times assumed that the map giving the isomor- Disjointness of rescalings A property which
phism is an involution (its second iteration is the seems to be common among parabolic flows
identity). Reversibility is often observed for (apart from some well-known exceptions in the
dynamical systems of physical origin. In contrast, homogeneous world) is disjointness of rescalings,
from the point of view of ergodic theory, this is a defined as follows. Given a real number k > 0, by
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 361

the k-rescaling of ’ℝ ¼ (’t)t  ℝ, we simply mean Rescalings of von Neumann flows Using simple
the flow ’kℝ ≔ ð’kt Þt  ℝ (in which the time is joining arguments Frączek and Lemańczyk (2009a)
rescaled by the factor k). Thus, a rescaling is a showed that every von Neumann flow over any
special type of time-reparametrization of a flow, ergodic IET is not isomorphic to its any rescaling.
given by a linear time-change. We say that ’ℝ has Disjointness of rescalings for von Neumann flows
disjoint rescalings if for all (or all but finitely was studied by Berk and Kanigowski (2021). They
many) p, q > 0, the rescalings ’pℝ and ’qℝ , considered von Neumann flows over IETs and
where p, q > 0 and p 6¼ q, are disjoint (in the showed disjointness (even spectral disjointness) of
sense of Furstenberg). Disjointness of rescalings rational rescalings, i.e., rescalings ’kℝ where
has played a key role in proving some of the first k  ℚ, for a.e. IET. A similar proof also allows
instances of Sarnak’s conjecture on the Moebius dealing with special flows over an IET T under
function (see the survey Kułaga-Przymus and piecewise constant function with a singularity in
Lemańczyk (2020)). the interior of a continuity interval of the IET.
Spectral disjointness of rational rescalings is also
Rescalings of Arnold flows Kanigowski, proved in Berk and Kanigowski (2021) for a.e. IET.
Lemańczyk, and Ulcigrai recently proved in
Kanigowski et al. (2020) that disjointness of Von Neumann flows over rotations The new
rescalings is typical among Arnold flows (see criterion for disjointness introduced by
Fig. 3b): Kanigowski et al. (2020) has been also used by
Dong and Kanigowski (2020) to study von Neu-
mann flows in genus one. They proved that for
Theorem 6 Let ’ℝ be the (Arnold) special flow any irrational rotation Rα, any real rescaling of a
over a.e. rotation Rα under a roof with asymmetric special flow ’ℝ over Rα under a piecewise abso-
logarithmic singularities with constants C0 6¼ C1 lutely continuous function with only one discon-
given by r(x) ¼ C0 j log x j þ C1 j log (1  x) j þ tinuity is disjoint from ’ℝ.
h(x) where h is smooth. Then there exist only two
values of the form q, 1/q such that ’ℝ and ’pℝ are Symmetric log over rotations Berk and
disjoint for any positive p {1, q, 1/q}, where Kanigowski (2021) also studied the case of a roof
q ¼ C0/C1. with one symmetric logarithmic singularity over
The proof exploits a new criterium for rotations (Kochergin prototype example of the
disjointness based on the switchable Ratner prop- absence of mixing) and also proved that in this
erty. The criterion was devised and formulated so case one has (spectral) disjointness of rational
that it can be applied to prove disjointness of two rescalings for a.e. rotation number. The techniques
flows which both have the switchable Ratner in this symmetric setting (where no form of Ratner
property when in both one can observe a con- property is known) are very different and based on
trolled form of divergence of nearby trajectories a refinement of the techniques used to prove the
(e.g., polynomial divergence), but the speed of absence of mixing and disjointness from mixing
divergence for the two flows is different (e.g., flows in Frączek and Lemańczyk (2004).
for one flow it is linear, and in the other,
quadratic).
Open Directions
Disjointness from other parabolic flows The cri-
terion is used in Kanigowski et al. (2020) also to Despite the surge of activity and many results on
show that a typical Arnold flow is disjoint from ergodic, mixing, and spectral properties which
any smooth time-change of the horocycle flow have been proved in the last decades on area-
(and in particular from the classical horocycle preserving flows on surfaces, both on IETs and
flow itself), thus showing that these two classes linear flows exploiting Teichmüller dynamics and
of parabolic flows are truly distinct. on smooth or locally Hamiltonian flows, many
362 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

questions remain open. We indicate here only further investigated. While the typical flow in
some questions and directions which arise natu- U min is weakly mixing but not mixing (see Theo-
rally from the results surveyed in this entry. rem 5), we do not know if weak mixing holds
everywhere when g  1 and no other examples
Linear flows and IETs Even though the ergodic other than the rather special examples of mixing
theory of typical IETs and translation flows is flow in g ¼ 5 in Chaika and Wright (2019) were
much studied and well understood, some ques- constructed. Within U : min , mixing holds a.e. on
tions remain open. Weak mixing is known for every minimal component (see Theorem 5), but
a.e. IET and a.e. linear flow but is not explicit one can wonder if there exist exceptional exam-
and while explicit examples of weak mixing ples without mixing for g  2. They are known
(and its lack) were constructed, a classification not to exist in g ¼ 1 in view of Kochergin work
of all translation surfaces on which linear flows Kochergin (2004), where mixing is proved for
are weak mixing for a.e. direction is still out of every irrational rotation and the roof function
reach. Another open question is the exact compu- with logarithmic singularities of strongly
tation of Hausdorff dimension of the set of non- asymmetric type.
weakly mixing IETs (see Question 1.6 in Chaika As we discussed, a.e. flow in U : min has not
and Masur (2020)). only minimal components which are mixing, but
Another Hausdorff dimension computation also mixing of all orders. Proving that any mixing
question concerns non-uniquely ergodic IETs, minimal component of a flow in U : min , as well as
namely what is the Hausdorff dimension of any mixing Kochergin flow, is also mixing of all
d-IETs with precisely k ergodic invariant proba- orders is an open problem (see Question 38 in
bility measures, for a given 1 < k  g  d/2 (listed Fayad and Krikorian (2018) ICM proceedings).
as Question 1.8 in Chaika and Masur (2020)). Showing it would confirm the famous (still open)
While a.e. IET has singular spectrum, whether Rokhlin conjecture (i.e., that mixing implies
there can exist exceptional IETs with an abso- mixing of all orders; see Quas (2017)) in the
lutely continuous component in the spectrum is setting of area-preserving surface flows.
not known. Finally, all examples where mild mixing is
The simplicity conjecture in Veech (1982a) on known (for special flows over rotations or IETs
simplicity of d-IETs for d  4 is still wide open, under symmetric logarithmic singularities, but
together with an understanding of joinings and also for von Neumann flows and for IETs them-
self-joinings. It also remains open whether almost selves) are over IETs which are linearly recurrent
every 3-IET is prime; it was a question suggested (or bounded-type) rotations or IETs. Perhaps mild
by Veech. A measure-preserving system is called mixing is not possible otherwise.
prime if it has no non-trivial factors.
While upper bounds on the Hausdorff dimen- Nature of the spectrum Spectral property of
sion of non-uniquely ergodic IETs are now locally Hamiltonian flows is a natural question,
known, a finer understanding of the set of non- which has been lingering for decades (see, e.g.,
uniquely ergodic directions NU ðSÞ on a given Katok and Thouvenot (2006), [Section 5]).
surface S, in the spirit of the full dichotomy for The recent result by Fayad et al. (2021) for
the slit-tori example, is probably out of reach. flows in g ¼ 1 with stopping points suggests that
Even if there exists a surface for which the it may be possible to prove that the spectrum is
Hausdorff dimension of NU ðSÞ is strictly between countable Lebesgue also in higher genus when in
0 and 1/2 is not known (see also Question 1.9 in presence of degenerate, sufficiently strong (multi-
Chaika and Masur (2020)). saddle) singularities. It is not clear what to expect
when the degenerate singularity is not sufficiently
Locally Hamiltonian Flows strong. In Fayad et al. (2021), the power γ of the
Exceptional mixing results Nevertheless, many singularity of the roof of the special flow repre-
questions about exceptional behavior can be sentation (see (7)) is close to 1. One might hope
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 363

that absolute continuity could hold for all powers linear flows, as well as infinite extensions of
γ  1/2, but this is out of reach with the current locally Hamiltonian flows, has been an active
techniques. If γ < 1/2, the approach which uses recent area of research. The latter extensions, as
square-integrability of the decay of correlations well as the study of translation surfaces which are
estimates fails completely. Abelian covers of compact translation surfaces, is
In the nondegenerate case, the singularity related to the study of their Poincaré maps, which
result proved in Chaika et al. (2021) for typical are skew-product extensions of IETs, which pro-
flows in g ¼ 2 with simple (isomorphic) saddles vide important infinite ergodic theory examples
indicates that the spectrum could be purely singu- and generalize the much studied theory of piece-
lar also for a.e. flow in U min : In the open set wise smooth cocycles over rotations. A skew-
U : min , which consists of flows with non- product extension of T : I ! I over a group
degenerate singularities that are not minimal, but G given by the cocycle ’ : I ! G is the map
have several minimal components, the nature of (x, g) 7! (Tx, g þ ’(x)). Such skew products
the spectrum (for the restriction of a typical flow where G ¼ ℤd and ’ is piecewise constant appear
to a minimal component) is unclear. These flows as Poincaré maps of Abelian covers of translation
are indeed mixing, but with sub-polynomial rate surfaces; ℝ-extensions of locally Hamiltonian
(see Ravotti (2017), which provides logarithmic flows provide cocycles ’ which are piecewise
upper bounds), and it is not clear whether to continuous or have (logarithmic or power-like)
expect singularity or absolute continuity of the singularities.
spectrum. The nature of the spectrum is an open Many results on the ergodicity of such infinite
problem even in g ¼ 1, for Arnold flows. measure-preserving flows and maps were proved
for special families of examples; see, e.g., Hubert
Disjointness of rescalings We believe and Weiss (2013), Hooper (2015), Málaga
disjointness of rescalings should also hold for Sabogal and Troubetzkoy (2016), Ralston and
typical locally Hamiltonian flows in higher Troubetzkoy (2017), or Chaika and Robertson
genus, but this is currently an open problem. Pre- (2019) for linear flows and IETs with infinite
liminary work seems to indicate that, despite some invariant measures and Fayad and Lemańczyk
technical additional difficulties, the techniques (2006), Conze and Frączek (2011), or Frączek
used to prove Theorem 6 should allow proving and Ulcigrai (2012, 2021) for extensions of
disjointness of rescalings for all minimal compo- locally Hamiltonian flows. Some typical non-
nents of typical flows in the open set U : min : ergodicity results were proved as well; see
Berk and Kanigowski (2021) proof of Frączek and Ulcigrai (2014) and Frączek and
disjointness of (rational) rescalings for symmetric Hubert (2018). The problem of ergodicity is nev-
logarithmic singularity over a full measure set of ertheless still widely open, both in the context of
rotations (Kochergin prototype example of absence linear flows on Abelian covers of translation sur-
of mixing) gives a good indication that disjointness faces and for extensions of locally Hamiltonian
of rescalings could also hold for typical flows flows. Other possible directions of investigation
(under a suitable full measure Diophantine-type are the study of Radon invariant measures and the
condition) in the complementary set U min : existence of limit theorems.

Non-compact surfaces We restricted in this sur- Acknowledgments K. F. acknowledges the supporf of

vey to the discussion of the ergodic theory of the Narodowe Centrum Nauki Grant number 2017/27/B/
flows on surfaces which are compact. Even ST1/00078. C. U. is part of SwissMAP (The Mathematics
of Physics National Centre for Compentence in Research)
though it is just at its beginning, the study of
and is currently supported by a SNSF (Swiss National
infinite translation surfaces (i.e., translation sur- Science Foundation) Grant number 200021_188617/1.
faces of infinite area or infinite genus) and their Both are acknowledged for their support.
364 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

Bibliography Society Translations: Series 2, vol 217. American

Mathematical Society, Providence, RI, pp 23–35
Al-Saqban H, Apisa P, Erchenko A, Khalil O, Mirzadeh S, Bufetov AI (2014) Limit theorems for translation flows.
Uyanik C (2021) Exceptional directions for the Ann Math 179(2):431–499
Teichmüller geodesic flow and Hausdorff dimension. Bufetov AI, Solomyak B (2018) The Hölder property for
J Eur Math Soc (JEMS) 23(5):1423–1476 the spectrum of translation flows in genus two. Israel
Aranson SH, Grines VZ (1973) Certain invariants of J Math 223(1):205–259
dynamical systems on two-dimensional manifolds Bufetov AI, Solomyak B (2020) A spectral cocycle for
(necessary and sufficient conditions for the topological substitution systems and translation flows. J Anal
equivalence of transitive systems). Mat Sb Math 141(1):165–205
(NS) 90(132):372–402, 479, Bufetov AI, Solomyak B (2021) Hölder regularity for the
Arnold VI (1991) Topological and ergodic properties of spectrum of translation flows. J Éc Polytech Math 8:
closed 1-forms with incommensurable periods. 279–310
Funktsional’nyi Analiz i Ego Prilozheniya 25(2): Chaika J (2012) Every ergodic transformation is disjoint
1–12. (Translated in: Funct Anal Appl 25(2):81–90, from almost every interval exchange transformation.
1991) Ann Math (2) 175(1):237–253
Athreya JS, Chaika J (2015) The Hausdorff dimension of Chaika J, Eskin A (2021) Self-joinings for 3-IETs. J Eur
non-uniquely ergodic directions in H(2) is almost Math Soc (JEMS) 23(8):2707–2731
everywhere 12 : Geom Topol 19(6):3537–3563 Chaika J, Frączek K, Kanigowski A, Ulcigrai C (2021)
Singularity of the spectrum for smooth area-preserving
Avila A, Delecroix V (2016) Weak mixing directions in
flows in genus two and translation surfaces well
non-arithmetic Veech surfaces. J Am Math Soc 29(4):
approximated by cylinders. Commun Math Phys
1167–1208
381(3):1369–1407
Avila A, Forni G (2007) Weak mixing for interval
Chaika J, Masur H (2015) There exists an interval
exchange transformations and translation flows. Ann
exchange with a non-ergodic generic measure. J Mod
Math 165(2):637–664
Dyn 9:289–304
Avila A, Forni G, Ravotti D, Ulcigrai C (2021a) Mixing for
Chaika J, Masur H (2020) The set of non-uniquely ergodic
smooth time-changes of general nilflows. Adv Math
d-IETs has Hausdorff codimension 1/2. Invent Math
385. Paper No. 107759:65
222(3):749–832
Avila A, Forni G, Safaee P (2021b) Quantitative weak
Chaika J, Robertson D (2019) Ergodicity of skew products
mixing for interval exchange transformations. https://
over linearly recurrent IETs. J Lond Math Soc 100(1):
arxiv.org/abs/2105.10547
223–248
Avila A, Gouëzel S, Yoccoz J-C (2006) Exponential
Chaika J, Weiss B (2022) The horocycle flow on the
mixing for the Teichmüller flow. Publ Math Inst Hautes
moduli space of translation surfaces, ICM 2022 Pro-
Études Sci 104:143–211
ceedings. EMS Press
Avila A, Leguil M (2018) Weak mixing properties of
Chaika J, Wright A (2019) A smooth mixing flow on a
interval exchange transformations & translation flows.
surface with nondegenerate fixed points. J Am Math
Bull Soc Math France 146(2):391–426
Soc 32(1):81–117
Avila A, Viana M (2007) Simplicity of Lyapunov spectra:
Cheung Y (2003) Hausdorff dimension of the set of non-
proof of the Zorich-Kontsevich conjecture. Acta Math
ergodic directions. Ann Math 158(2):661–678. With an
198(1):1–56
appendix by M. Boshernitzan
Berk P, Frączek K (2015) On special flows over IETs that
Cheung Y, Hubert P, Masur H (2011) Dichotomy for the
are not isomorphic to their inverses. Discrete Contin
Hausdorff dimension of the set of nonergodic direc-
Dyn Syst 35(3):829–855
tions. Invent Math 183(2):337–383
Berk P, Frączek K, de la Rue T (2020) On typicality of
Conze J-P, Frączek K (2011) Cocycles over interval
translation flows which are disjoint with their inverse.
exchange transformations and multivalued Hamilto-
J Inst Math Jussieu 19(5):1677–1737
nian flows. Adv Math 226(5):4373–4428
Berk P, Kanigowski A (2021) Spectral disjointness of
Danilenko AI, Ryzhikov VV (2012) On self-similarities of
rescalings of some surface flows. J Lond Math Soc
ergodic flows. Proc Lond Math Soc (3) 104(3):431–454
(2) 103(3):901–942
de la Rue T (2020) Joinings in ergodic theory. Springer,
Blohin AA (1972) Smooth ergodic flows on surfaces.
Berlin Heidelberg, Berlin, Heidelberg, pp 1–20
Trudy Moskov Mat Obšč 27:113–128
del Junco A (1981) Disjointness of measure-preserving
Boshernitzan M, Nogueira A (2004) Generalized eigen-
transformations, minimal self-joinings and category.
functions of interval exchange maps. Ergod Theory
In: Ergodic theory and dynamical systems, I (College
Dyn Syst 24(3):697–705
Park, Md., 1979–80), Progress in Mathematics, vol 10.
Bufetov A, Sinai YG, Ulcigrai C (2006) A condition for
Birkhäuser, Boston, MA, pp 81–89
continuous spectrum of an interval exchange transfor-
del Junco A (1983) A family of counterexamples in ergodic
mation. In: Representation theory, dynamical systems,
theory. Israel J Math 44(2):160–188
and asymptotic combinatorics, American Mathematical
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 365

Dong C, Kanigowski A (2020) Rigidity of a class of Frączek K, Lemańczyk M (2005) On disjointness proper-
smooth singular flows on 2 J Mod Dyn 16:37–57 ties of some smooth flows. Fundam Math 185(2):
Fayad B (2001) Polynomial decay of correlations for a 117–142
class of smooth flows on the two torus. Bull Soc Math Frączek K, Lemańczyk M (2006) On mild mixing of spe-
France 129(4):487–503 cial flows over irrational rotations under piecewise
Fayad B, Forni G, Kanigowski A (2021) Lebesgue spec- smooth functions. Ergod Theory Dyn Syst 26(3):
trum of countable multiplicity for conservative flows 719–738
on the torus. J Am Math Soc 34(3):747–813 Frączek K, Lemańczyk M (2010) Ratner’s property and
Fayad B, Kanigowski A (2016) Multiple mixing for a class mild mixing for special flows over two-dimensional
of conservative surface flows. Invent Math 203(2): rotations. J Mod Dyn 4:609–635
555–614 Frączek K (2009) Density of mild mixing property for
Fayad B, Krikorian R (2018) Some questions around qua- vertical flows of abelian differentials. Proc Am Math
siperiodic dynamics. In: Proceedings of the Interna- Soc 137(12):4229–4142
tional Congress of Mathematicians—Rio de Janeiro Frączek K, Hubert P (2018) Recurrence and non-ergodicity
2018, Invited lectures, vol III. World Sci. Publ, Hack- in generalized wind-tree models. Math Nachr
ensack, NJ, pp 1909–1932 291(11–12):1686–1711
Fayad B, Lemańczyk M (2006) On the ergodicity of cylin- Frączek K, Kim M (2021) New phenomena in deviation of
drical transformations given by the logarithm. Mosc Birkhoff integrals for locally Hamiltonian flows.
Math J 6(4):657–672, 771–772 arXiv:2112.13030
Fayad BR (2002) Analytic mixing reparametrizations of Frączek K, Kułaga-Przymus J, Lemańczyk M (2014) Non-
irrational flows. Ergod Theory Dyn Syst 22(2):437–468 reversibility and self-joinings of higher orders for ergo-
Ferenczi S, Holton C, Zamboni LQ (2005) Joinings of dic flows. J Anal Math 122:163–227
three-interval exchange transformations. Ergod Theory Frączek K, Lemańczyk M (2009a) On the self-similarity
Dyn Syst 25(2):483–502 problem for ergodic flows. Proc Lond Math Soc 99(3):
Ferenczi S, Zamboni LQ (2011) Eigenvalues and simplic- 658–696
ity of interval exchange transformations. Ann Sci Éc Frączek K, Lemańczyk M (2009b) Smooth singular flows
Norm Supér (4) 44(3):361–392 in dimension 2 with the minimal self-joining property.
Forni G (1997) Solutions of the cohomological equation Monatsh Math 156(1):11–45
for area-preserving flows on compact surfaces of higher Frączek K, Lemańczyk M, Lesigne E (2007) Mild mixing
genus. Ann Math 146(2):295–344 property for special flows under piecewise constant
Forni G (2002) Deviation of ergodic averages for area- functions. Discrete Contin Dyn Syst 19(4):691–710
preserving flows on surfaces of higher genus. Ann Frączek K, Ulcigrai C (2012) Ergodic properties of infinite
Math 155(1):1–103 extensions of area-preserving flows. Math Ann 354(4):
Forni G (2021) Sobolev regularity of solutions of the 1289–1367
cohomological equation. Ergod Theory Dyn Syst Frączek K, Ulcigrai C (2014) Non-ergodic ℤ-periodic bil-
41(3):685–789 liards and infinite translation surfaces. Invent Math
Forni G (2022) Twisted translation flows and effective 197(2):241–298
weak mixing. J Eur Math Soc (JEMS) 24(2022), no. Frączek K, Ulcigrai C (2021) On the asymptotic growth of
12, 4225–4276 Birkhoff integrals for locally Hamiltonian flows and
Forni G, Marmi S, Matheus C (2017) Cohomological ergodicity of their extensions. arXiv:2112.05939
equation and local conjugacy class of diophantine inter- Furstenberg H (1967) Disjointness in ergodic theory, min-
val exchange maps to appear in Proc. Amer. Math. imal sets, and a problem in Diophantine approximation.
Soc., https://doi.org/10.1090/proc/14538 Math Syst Theory 1:1–49
Forni G, Matheus C (2014) Introduction to Teichmüller Ghazouani S (2021) Local rigidity for periodic generalised
theory and its applications to dynamics of interval interval exchange transformations. Invent Math 226(2):
exchange transformations, flows on surfaces and bil- 467–520
liards. J Mod Dyn 8(3–4):271–436 Ghazouani S, Ulcigrai C (2021) A priori bounds for GIETs,
Forni G, Ulcigrai C (2012) Time-changes of horocycle affine shadows and rigidity of foliations in genus 2. Pre-
flows. J Mod Dyn 6(2):251–273 print arXiv:2106.03529
Fox RH, Kershner RB (1936) Concerning the transitive Halmos PR (1960) Lectures on ergodic theory. Chelsea
properties of geodesics on a rational polyhedron. Duke Publishing Co., New York
Math J 2(1):147–150 Hmili H (2010) Non topologically weakly mixing interval
Frączek K, Lemańczyk M (2003) On symmetric logarithm exchanges. Discrete Contin Dyn Syst 27(3):1079–1091
and some old examples in smooth ergodic theory. Hooper WP (2015) The invariant measures of some infinite
Fundam Math 180(3):241–255 interval exchange maps. Geom Topol 19(4):1895–2038
Frączek K, Lemańczyk M (2004) A class of special flows Host B (1991) Mixing of all orders and pairwise indepen-
over irrational rotations which is disjoint from mixing dent joinings of systems with singular spectrum. Israel
flows. Ergod Theory Dyn Syst 24:1083–1095 J Math 76(3):289–298
366 Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces

Hubert P, Weiss B (2013) Ergodicity for infinite periodic Kochergin AV (2007) Nondegenerate saddles, and absence
translation surfaces. Compos Math 149(8):1364–1380 of mixing. II. Mat Zametki 81(1):145–148
Kanigowski A (2015) Ratner’s property for special flows Kontsevich M (1997) Lyapunov exponents and Hodge
over irrational rotations under functions of bounded theory. In: The mathematical beauty of physics
variation. Ergod Theory Dyn Syst 35(3):915–934 (Saclay, 1996), Advanced Series in Mathematical
Kanigowski A, Kułaga-Przymus J (2016) Ratner’s prop- Physics, vol 24. World Scientific Publishing, River
erty and mild mixing for smooth flows on surfaces. Edge, NJ, pp 318–332
Ergod Theory Dyn Syst 36(8):2512–2537 Kontsevich M, Zorich A (1997) Lyapunov exponents and
Kanigowski A, Kułaga-Przymus J, Ulcigrai C (2019) Mul- Hodge theory. arXiv:hep-th/9701164
tiple mixing and parabolic divergence in smooth area- Kočergin AV (1976) Nondegenerate saddles, and the
preserving flows on higher genus surfaces. J Eur Math absence of mixing. Mat Zametki 19(3):453–468
Soc (JEMS) 21(12):3797–3855 Kułaga J (2012) On the self-similarity problem for smooth
Kanigowski A, Lemańczyk M (2020) Spectral theory of flows on orientable surfaces. Ergod Theory Dyn Syst
dynamical systems. Springer Berlin Heidelberg, Berlin, 32(5):1615–1660
Heidelberg, pp 1–40 Kułaga-Przymus J, Lemańczyk M (2020) Sarnak’s conjec-
Kanigowski A, Lemańczyk M, Ulcigrai C (2020) On ture from the ergodic theory point of view. Springer
disjointness properties of some parabolic flows. Invent Berlin Heidelberg, Berlin, Heidelberg, pp 1–19
Math 221(1):1–111 Lanneau E, Marmi S, Skripchenko A (2021) Cohomolog-
Katok A (2001) Cocycles, cohomology and combinatorial ical equations for linear involutions. Dyn Syst 36(2):
constructions in ergodic theory. In: Smooth ergodic 292–304
theory and its applications (Seattle, WA, 1999), Pro- Lemańczyk M (2000) Sur l’absence de mélange pour des
ceedings of Symposia in Pure Mathematics, vol 69. flots spéciaux au-dessus d’une rotation irrationnelle.
American Mathematical Society, Providence, RI, Colloq Math 84–85:29–41
pp 107–173. In collaboration with E. A. Robinson, Jr Levitt G (1983) Feuillettages des surfaces. Thèse
Katok A, Thouvenot J-P (2006) Spectral properties and Málaga Sabogal A, Troubetzkoy S (2016) Ergodicity of the
combinatorial constructions in ergodic theory. In: Ehrenfest wind-tree model. C R Math Acad Sci Paris
Handbook of dynamical systems, vol 1B. Elsevier 354(10):1032–1036
B. V., Amsterdam, pp 649–743 Marmi S, Moussa P, Yoccoz J-C (2005) The cohomologi-
Katok AB (1973) Invariant measures of flows on oriented cal equation for Roth-type interval exchange maps.
surfaces. Sov Math Dokl 14:1104–1108 J Am Math Soc 18(4):823–872. (electronic)
Katok AB (1980) Interval exchange transformations and Marmi S, Moussa P, Yoccoz J-C (2012) Linearization of
some special flows are not mixing. Israel J Math 35(4): generalized interval exchange maps. Ann Math
301–310 (2) 176(3):1583–1646
Katok AB, Stepin AM (1966) Approximation of ergodic Marmi S, Ulcigrai C, Yoccoz J-C (2020) On Roth type
dynamical systems by periodic transformations. Dokl conditions, duality and central Birkhoff sums for
Akad Nauk SSSR 171:1268–1271 I.E.M. Astérisque, (416, Quelques aspects de la théorie
Katok AB, Stepin AM (1967) Approximations in ergodic des systèmes dynamiques: un hommage à Jean-
theory. Uspehi Mat Nauk 137(5):81–106 Christophe Yoccoz.II), pp. 65–132
Keane M (1975) Interval exchange transformations. Math Marmi S, Yoccoz J-C (2016) Hölder regularity of the
Z 141:25–31 solutions of the cohomological equation for Roth type
Keane M (1977) Non-ergodic interval exchange transfor- interval exchange maps. Comm Math Phys 344(1):
mations. Israel J Math 26(2):188–196 117–139
Kerckhoff S, Masur H, Smillie J (1986) Ergodicity of Masur H (1982) Interval exchange transformations and
billiard flows and quadratic differentials. Ann Math measured foliations. Ann Math 115:169–200
(2) 124((2):293–311 Masur H (1992) Hausdorff dimension of the set of non-
Keynes HB, Newton D (1976) A “minimal”, nonuniquely ergodic foliations of a quadratic differential. Duke
ergodic interval exchange transformation. Math Math J 66(3):387–442
Z 148(2):101–105 Masur H, Smillie J (1991) Hausdorff dimension of sets of
Kočergin AV (1972) The absence of mixing in special flows nonergodic measured foliations. Ann Math (2) 134(3):
over a rotation of the circle and in flows on a two-dimen- 455–543
sional torus. Dokl Akad Nauk SSSR 205:512–518. Masur H, Tabachnikov S (2002) Rational billiards and flat
(Translated in: Sov Math Dokl 13:949–952, 1972) structures. In: Handbook of dynamical systems, vol 1A.
Kočergin AV (1975) Mixing in special flows over a shifting North-Holland, Amsterdam, pp 1015–1089
of segments and in smooth flows on surfaces. Mat Sb Mayer A (1943) Trajectories on the closed orientable sur-
96(138):471–502 faces. Rec Math [Mat Sbornik] NS 12(54):71–84
Kochergin AV (2004) Some generalizations of theorems on McMullen CT (2020) Teichmüller dynamics and unique
mixing flows with nondegenerate saddles on a two- ergodicity via currents and Hodge theory. J Reine
dimensional torus. Mat Sb 195(9):19–36 Angew Math 768:39–54
Ergodic and Spectral Theory of Area-Preserving Flows on Surfaces 367

Nikolaev I, Zhuzhoma E (1999) Flows on 2-dimensional Ulcigrai C (2022) Dynamics and ‘arithmetics’ of higher
manifolds. In: Lecture notes in mathematics, vol 1705. genus surface flows. EMS Press, ICM 2022
Springer Proceedings
Nogueira A, Rudolph D (1997) Topological weak-mixing Ulcigrai C (2007) Mixing of asymmetric logarithmic sus-
of interval exchange maps. Ergod Theory Dyn Syst pension flows over interval exchange transformations.
17(5):1183–1209 Ergod Theory Dyn Syst 27(3):991–1035
Novikov SP (1982) The Hamiltonian formalism and a Ulcigrai C (2009) Weak mixing for logarithmic flows over
multivalued analogue of Morse theory. (Russian) interval exchange transformations. J Mod Dyn 3(1):
Uspekhi Matematicheskikh Nauk 37(5):3–49. 35–49
(Translated in: Russ Math Surv 37(5):1–56, 1982) Ulcigrai C (2011) Absence of mixing in area-preserving
Oseledec VI (1966) The spectrum of ergodic automor- flows on surfaces. Ann Math (2) 173(3):1743–1778
phisms. Dokl Akad Nauk SSSR 168:1009–1011 Ulcigrai C (2021) Slow chaos in surface flows. Boll
Poincaré H (1987) Les méthodes nouvelles de la Unione Mat Ital 14(1):231–255
mécanique céleste. In: Les Grands Classiques Veech WA (1969) Strict ergodicity in zero dimensional
Gauthier-Villars, Librairie Scientifique et Technique dynamical systems and the Kronecker-Weyl theorem
Albert Blanchard, Paris mod 2. Trans Am Math Soc 140:1–33
Quas A (2017) Ergodicity and mixing properties. Springer Veech WA (1982a) A criterion for a process to be prime.
Berlin Heidelberg, Berlin, Heidelberg, pp 1–20 Monatsh Math 94(4):335–341
Ralston D, Troubetzkoy S (2017) Residual generic ergo- Veech WA (1982b) Gauss measures for transformations on
dicity of periodic group extensions over translation the space of interval exchange maps. Ann Math 115:
surfaces. Geom Dedicata 187:219–239 201–242
Ravotti D (2017) Quantitative mixing for locally Hamilto- Veech WA (1984) The metric theory of interval exchange
nian flows with saddle loops on compact surfaces. Ann transformations I. generic spectral properties. Am
Henri Poincaré 18(12):3815–3861 J Math 107(6):1331–1359
Robertson D (2019) Mild mixing of certain interval- Viana M (n.d.) Dynamics of interval exchange transforma-
exchange transformations. Ergod Theory Dyn Syst tions and Teichmüller flows. Available from http://w3.
39(1):248–256 impa.br/viana. Lecture Notes
Ryzhikov VV (1994) The absence of mixing in special von Neumann J (1932) Zur Operatorenmethode in der
flows over rearrangements of segments. Mat Zametki klassischen Mechanik. Ann Math (2) 33(3):587–642
55(6):146–149. (Translated in: Math Notes Yoccoz J-C (2010) Interval exchange maps and translation
55(5–6):648–650, 1994) surfaces. In: Homogeneous flows, moduli spaces and
Ryzhikov VV, Tuveno Z-P (2006) Disjointness, divisibil- arithmetic, Clay Mathematics Proceedings, vol 10. Amer-
ity, and quasi-simplicity of measure-preserving actions. ican Mathematical Society, Providence, RI, pp 1–69
Funktsional Anal i Prilozhen 40(3):85–89 Zemljakov AN, Katok AB (1975) Topological transitivity
Sataev EA (1975) The number of invariant measures for of billiards in polygons. Mat Zametki 18(2):291–300
flows on orientable surfaces. Izv Akad Nauk SSSR Ser Zorich A (1984) S. P. Novikov’s problem of the semiclas-
Mat 39(4):860–878 sical motion of an electron in a homogeneous magnetic
Scheglov D (2009) Absence of mixing for smooth flows on field that is close to rational. Uspekhi Mat Nauk
genus two surfaces. J Mod Dyn 3(1):13–34 39(5(239)):235–236
Schmidt K (2002) Dispersing cocycles and mixing flows Zorich A (1997) Deviation for interval exchange transfor-
under functions. Fundam Math 173(2):191–199 mations. Ergod Theory Dyn Syst 17(6):1477–1499
Sinai YG, Khanin KM (1992) Mixing for some classes of Zorich A (1999) How do the leaves of a closed 1-form
special flows over rotations of the circle. wind around a surface? In: Pseudoperiodic topology,
Funktsional’nyi Analiz i Ego Prilozheniya 26(3): American Mathematical Society Translations: Series 2,
1–21. (Translated in: Funct Anal Appl 26(3)155–169, vol 197. American Mathematical Society, Providence,
1992) RI, pp 135–178
Sinai YG, Ulcigrai C (2005) Weak mixing in interval Zorich A (2006) Flat surfaces. In: Frontiers in number
exchange transformations of periodic type. Lett Math theory, physics, and geometry, I. Springer, Berlin,
Phys 74(2):111–133 pp 437–583
 f : X ! ℝ (“potential”), those invariant mea-
Pressure and Equilibrium States sures that maximize a functional of the form
in Ergodic Theory F(m) ¼ h(m) þ hf, mi are called “equilibrium
states” for f.
Jean-René Chazottes1 and Gerhard Keller2 Ergodic theory Ergodic theory is the mathemat-
1
Centre de Physique Théorique, CNRS/IP Paris, ical theory of measure-preserving dynamical
Palaiseau, France systems.
2
Department Mathematik, Universität Erlangen- Gibbs state In many cases, equilibrium states
Nürnberg, Erlangen, Germany have a local structure that is determined by
the local properties of the potential f. They
are called “Gibbs states.”
Article Outline Invariant measure In this entry: a probability
measure m on X which is invariant under the
Glossary transformation T, that is, for which hf ∘ T, mi ¼
Definition of the Subject hf, mi for each continuous f : X ! ℝ. Here
Introduction hf, mi is a short-hand notation for X f dm.
Warming Up: Thermodynamic Formalism for The triple (X, T, m) is called a measure-
Finite Systems preserving dynamical system.
Shift Spaces, Invariant Measures, and Entropy Pressure The maximum of the functional F(m)
The Variational Principle: A Global is denoted by P(f) and called the “topologi-
Characterization of Equilibrium cal pressure” of f, or simply the “pressure”
The Gibbs Property: A Local Characterization of of f.
Equilibrium Sinai-Ruelle-Bowen measure Special equili
Examples on Shift Spaces brium or Gibbs states that describe the statisti
Examples from Differentiable Dynamics cs of the attractor of certain smooth dynamical
Nonequilibrium Steady States and Entropy systems.
Production
Some Ongoing Developments and Future
Directions Definition of the Subject
Bibliography
Gibbs and equilibrium states of one-dimensional
Glossary lattice models in statistical physics play a promi-
nent role in the statistical theory of chaotic
Dynamical system In this entry: a continuous dynamics. They first appear in the ergodic theory
transformation T of a compact metric space X. of certain differentiable dynamical systems, called
For each x  X, the transformation T generates “uniformly hyperbolic systems,” mainly Anosov
a trajectory (x, Tx, T2x, . . .). and Axiom A diffeomorphisms (and flows). The
Entropy In this entry: the maximal rate of infor- central idea is to “code” the orbits of these systems
mation gain per time that can be achieved by into (infinite) symbolic sequences of symbols by
coarse-grained observations on a measure- following their history on a finite partition of their
preserving dynamical system. This quantity is phase space. This defines a nice shift dynamical
often denoted h(m). system called a subshift of finite type or a topo-
Equilibrium state In general, a given dynamical logical Markov chain. Then the construction of
system T : X ! X admits a huge number of their “natural” invariant measures and the study of
invariant measures. Given some continuous their properties are carried out at the symbolic
© Springer Science+Business Media, LLC, part of Springer Nature 2023 369
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_414
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_414-3
370 Pressure and Equilibrium States in Ergodic Theory

level by constructing certain equilibrium states in interdependence structure by means of a suitable

the sense of statistical mechanics which turn out to class of conditional probabilities built up according
be also Gibbs states. The study of uniformly to the Maxwell-Boltzmann-Gibbs formula (Georgii
hyperbolic systems brought out several ideas and 1988). Notice that Gibbs states are often called
techniques which turned out to be extremely fruit- “DLR states” in honor of Dobrushin, Lanford, and
ful for the study of more general systems. Let us Ruelle. The remarkable aspect of this construction is
mention the concept of Markov partition and its the fact that a Gibbs state for a given type of inter-
avatars, the very important notion of SRB mea- action may fail to be unique. In physical terms, this
sure (after Sinai, Ruelle, and Bowen) and transfer means that a system with this interaction can take
operators. Recently, there was a revival of interest several distinct equilibria. The phenomenon of non-
in Axiom A systems as models to understand uniqueness of a Gibbs measure can thus be
nonequilibrium statistical mechanics. interpreted as a phase transition. Therefore, the con-
ditions under which an interaction leads to a unique
or to several Gibbs measures turn out to be of central
Introduction importance. While Gibbs states are defined locally
by specifying certain conditional probabilities, equi-
Our goal is to present the basic results on one- librium states are defined globally by a variational
dimensional Gibbs and equilibrium states viewed principle: they maximize the entropy of the system
as special invariant measures on symbolic dynam- under the (linear) constraint that the mean energy is
ical systems, and then to describe without techni- fixed. Gibbs states are always equilibrium states, but
calities a sample of results they allow to obtain for the two notions do not coincide in general. How-
certain differentiable dynamical systems. We ever, for a class of sufficiently regular interactions,
hope that this contribution will illustrate the sym- equilibrium states are also Gibbs states.
biotic relationship between ergodic theory and In the effort of trying to understand phase tran-
statistical mechanics, and also information theory. sitions, simplified mathematical models were pro-
We start by putting Gibbs and equilibrium states posed, the most famous one being undoubtedly the
in a general perspective. The theory of Gibbs states Ising model. This is an example of a lattice model.
and equilibrium states, or Thermodynamic Formal- The set of configurations of a lattice model is
X≔Aℤ , where A is a finite set, which is invariant
d
ism, is a branch of rigorous Statistical Physics. The
notion of a Gibbs state dates back to R.L. Dobrushin by “spatial” translations. For the physical interpre-
(1968–1969) (Dobrushin 1968a, b, c, 1969) and tation, X can be thought, for instance, as the set of
O.E. Lanford and D. Ruelle (1969) who proposed infinite configurations of a system of spins on a
it as a mathematical idealization of an equilibrium crystal lattice ℤd and one may take A ¼ {1, +1},
state of a physical system which consists of a very that is, spins can take two orientations, “up” and
large number of interacting components. For a finite “down.” The Ising model is defined by specifying
number of components, the foundations of statisti- an interaction (or potential) between spins and
cal mechanics were already laid in the nineteenth studying the corresponding (translation-invariant)
century. There was the well-known Maxwell- Gibbs states. The striking phenomenon is that for
Boltzmann-Gibbs formula for the equilibrium dis- d ¼ 1 there is a unique Gibbs state (in fact a
tribution of a physical system with given energy Markov measure) whereas if d  2, there may be
function. From the mathematical point of view, the several Gibbs states, although the interaction is
intrinsic properties of very large objects can be very simple (Georgii 1988).
made manifest by performing suitable limiting pro- Equilibrium states and Gibbs states of one-
cedures. Indeed, the crucial step made in the 1960s dimensional lattice models (d ¼ 1) played a prom-
was to define the notion of a Gibbs measure or inent role in understanding the ergodic properties
Gibbs state for a system with an infinite number of of certain types of differentiable dynamical sys-
interacting components. This was done by the tems, namely uniformly hyperbolic systems,
familiar probabilistic idea of specifying the Axiom A diffeomorphisms in particular. The link
Pressure and Equilibrium States in Ergodic Theory 371

between one-dimensional lattice systems and present a selection of important examples: mea-
dynamical systems is made by symbolic dynamics. sure of maximal entropy, Markov measures and
Informally, symbolic dynamics consists of Hofbauer’s example of nonuniqueness of equilib-
replacing the orbits of the original system by its rium state; uniformly expanding Markov maps of
history on a finite partition of its phase space the interval, interval maps with an indifferent
labeled by the elements of the “alphabet” A. There- fixed point, Anosov diffeomorphisms and
fore, each orbit of the original system is replaced by Axiom A attractors with Sinai-Ruelle-Bowen
an infinite sequence of symbols, that is, by an measures, and Bowen’s formula for the Hausdorff
element of the set Aℤ or Aℕ, depending on whether dimension of conformal repellers. As we shall see,
the map describing the dynamics is invertible or Sinai-Ruelle-Bowen measures are the only phys-
not. The action of the map on an initial condition is ically observable measures and they appear natu-
then easily seen to correspond to the translation rally in the context of nonuniformly hyperbolic
(or shift) of its associated symbolic sequence. In diffeomorphisms (Young 2002).
general there is no reason to get all sequences of Aℤ A revival of the interest to Anosov and Axiom
or Aℕ. Instead one gets a closed invariant subset A systems occurred in statistical mechanics in the
X (a subshift) which can be very complicated. For a 1990s. Several physical phenomena of non-
certain class of dynamical systems the partition can equilibrium origin, like entropy production and
be successfully chosen so as to form a Markov chaotic scattering, were modeled with the help of
partition. In this case, the dynamical system those systems (by G. Gallavotti, P. Gaspard,
under consideration can be coded by a subshift of D. Ruelle, and others). This new interest led to
finite type (also called a topological Markov chain) new results about old Anosov and Axiom
which is a very nice symbolic dynamical system. A systems, see, for example, Chernov (2002) for
Then one can play the game of statistical physics: a survey and references. In section “Non-
for a given continuous, real-valued function equilibrium Steady States and Entropy Produc-
(a “potential”) on X, construct the corresponding tion,” we give a very brief account of entropy
Gibbs states and equilibrium states. If the potential production in the context of Anosov systems
is regular enough, one expects uniqueness of the which highlights the role of relative entropy.
Gibbs state and that it is also the unique equilibrium This entry is a little introduction to a vast subject
state for this potential. This circle of ideas – ranging in which we have tried to put forward some aspects
from Gibbs states on finite systems over invariant not previously described in other expository texts.
measures on symbolic systems and their For readers willing to deepen their understanding
(Shannon-)entropy with a digression to of equilibrium and Gibbs states, there are the clas-
Kolmogorov-Chaitin complexity to equilibrium sic monographs by Bowen (2017) and by Ruelle
states and Gibbs states on subshifts of finite (2004), the monograph by one of us (Keller 1998),
type – is presented in the next four sections. and the survey article by Chernov (2002) (where
At this point it should be remembered that the Anosov and Axiom A flows are reviewed). Those
objects which can actually be observed are not texts are really complementary.
equilibrium states (they are measures on X) but
individual symbol sequences in X, which reflect
more or less the statistical properties of an equi- Warming Up: Thermodynamic
librium state. Indeed, most sequences reflect these Formalism for Finite Systems
properties very well, but there are also rare
sequences that look quite different. Their proper- We introduce the thermodynamic formalism in an
ties are described by large deviations principles elementary context, following Jaynes (1989). In
which are not discussed in the present article. We this view, entropy, in the sense of information
shall indicate some references along the way. theory, is the central concept.
In sections “Examples on Shift Spaces” and Incomplete knowledge about a system is con-
“Examples from Differentiable Dynamics” we veniently described in terms of probability
372 Pressure and Equilibrium States in Ergodic Theory

distributions on the set of its possible states. This distribution m provided the value E can be attained
is particularly simple if the set of states, call it X, is at all by some hU, ni. In order to derive an explicit
finite. Then the equidistribution on X describes formula for this m we introduce a Lagrange mul-
complete lack of knowledge, whereas a probabil- tiplier β  ℝ and study, for each β, the
ity vector that assigns probability 1 to one single unconstrained problem
state and probability 0 to all others represents
maximal information about the system. A well- H mb þ bU , mb ¼ pðbU Þ≔ max ðH ðnÞ þ hbU , niÞ:
n
established measure of the amount of uncer-
ð2Þ
tainty represented by a probability distribution
n ¼ (n(x))x  X is its entropy
In analogy to the convention in ergodic theory
we call p(βU) the pressure of βU and the maxi-
H ðnÞ≔  nðxÞ log nðxÞ, mizer mβ the corresponding equilibrium distribu-
xX
tion (synonymously equilibrium state).
The equilibrium distribution mβ satisfies
which is zero if the probability is concentrated in
one state and which attains its maximum value
log|X| if n is the equidistribution on X, that is, if mb ðxÞ ¼ exp ðpðbU Þ þ bU ðxÞÞ for all x  X
n(x) ¼ |X|1 for all x  X. In this completely ð3Þ
elementary context we will explore two concepts
whose generalizations are central to the theory of as an elementary calculation using Jensen’s inequal-
equilibrium states in ergodic theory: ity for the strictly convex function t 7!  t log t
shows:
• Equilibrium distributions – defined in terms of
a variational problem ebU ðxÞ
H ðnÞ þ hbU , ni ¼ nðxÞ log
• The Gibbs property of equilibrium xX
nð x Þ
distributions
ebU ðxÞ
 log nð x Þ
xX
nðxÞ
The only mathematical prerequisite for this
section are calculus and some elements from ¼ log ebU ðxÞ ,
probability theory. xX

Equilibrium Distributions and the Gibbs with equality if and only if eβU is a constant
Property multiple of n. The observation that n ¼ mβ is a
Suppose that a finite system can be observed maximizer proves at the same time that p(βU) ¼
through a function U : X ! ℝ (an “observable”), log x  XeβU(x).
and that we are looking for a probability distribu- The equality expressed in (3) is called the Gibbs
tion m which maximizes entropy among all distri- property of mβ, and we say that mβ is a Gibbs
butions n with a prescribed expected value distribution if we want to stress this property.
hU, ni ≔ x  Xn(x)U(x) for the observable U. In order to solve the constrained problem (1) it
This means we have to solve a variational problem remains to show that there is a unique multiplier
under constraints: β ¼ β(E) such that hU, mβi ¼ E. This follows from
the fact that the map β 7! hU, mβi maps the real
H ðmÞ ¼ max fH ðnÞ : hU, ni ¼ Eg: ð1Þ line monotonically onto the interval
(minU, maxU ) which, in turn, is a direct conse-
As the function n 7! H(n) is strictly concave, quence of the formulas for the first and second
there is a unique maximizing probability derivative of p(βU) w.r.t β:
Pressure and Equilibrium States in Ergodic Theory 373

dp d2 p which are invariant under cyclic translations which

¼ U, mb ,
db db2 yields
2
¼ U 2 , mb  U, mb : ð4Þ PðbfÞ ¼ max n1 H ðnÞ þ hbf, ni
n
ð6Þ
As the second derivative is nothing but the ¼ n1 H mb þ bf, mb :
variance of U under mβ, it is strictly positive
(except when U is a constant function), so that If the local observable f depends only on one
β 7! hU, mβi is indeed strictly increasing. coordinate, mβ turns out to be a product measure:
Observe also that dp/dβ is indeed the directional
derivative of p : ℝ|A| ! ℝ in direction U. Hence n1

the first identity in (4) can be rephrased as: mβ is mb ða0 . . . an1 Þ ¼ exp ðPðbfÞ þ bfðai ÞÞ:
i¼0
the gradient at βU of the function p.
A similar analysis can be performed for an ℝd- Indeed, comparison with (3) shows that mβ is
valued observable U. In that case a vector β  ℝd the n-fold product of the probability distribution
of Lagrange multipliers is needed to satisfy the
b on A that maximizes H(n) þ βn(f) among all
mloc
d linear constraints.
distributions n on A. It follows that n1 H mb ¼
H mlocb so that (6) implies P(βf) ¼ p(βf) for
Systems on a Finite Lattice
observables f that depend only on one coordinate.
We now assume that the system has a lattice
structure, modeling its extension in space, for
instance. The system can be in different states at Shift Spaces, Invariant Measures, and
different positions. More specifically, let n ¼ Entropy
f0, 1, . . . , n  1g be a set of n positions in
space, let A be a finite set of states that can be We now turn to shift dynamical systems over a
attained by the system at each of its sites, and finite alphabet A.
denote by X≔An the set of all configurations of
states from A at positions of n . It is helpful to Symbolic Dynamics
think of X as the set of all words of length n over We start by fixing some notation. Let ℕ denote the
the alphabet A. We focus on observables Un which set {0, 1, 2, . . .}. In the sequel we need
are sums of many local contributions in the sense
that Un ða0 . . . an1 Þ ¼ n1 • A finite set A (the “alphabet”),
i¼0 fðai . . . aiþr1 Þ for
some “local observable” f : Ar ! ℝ. (The index • The set Aℕ of all infinite sequences over A, that
i þ r  1 has to be taken modulo n.) In terms of f is, the set of all x ¼ x0 x1 . . . with xn  A for all
the maximizing measure can be written as n  ℕ,
• The translation (or shift) s : Aℕ ! Aℕ,
mb ða0 . . . an1 Þ ðsxÞn ¼ xnþ1 , for all n  ℕ,
n1 • A shift invariant subset X ¼ s(X) of Aℕ. With a
¼ exp nPðbfÞ þ b fðai . . . aiþr1 Þ , slight abuse of notation we denote the restric-
i¼0 tion of s to X by s again.
ð5Þ
We mention two interpretations of the dynamics
where P(βf) ≔ n1p(βUn). A first immediate con- of s: it can describe the evolution of a system with
sequence of (5) is the invariance of mβ under a cyclic state space X in discrete time steps (this is the
shift of its argument, namely mβ(a1. . .an1a0) ¼ prevalent interpretation if s : X ! X is obtained
mβ(a0. . .an1). Therefore, we can restrict the maxi- as a symbolic representation of another dynamical
mizations in (1) and (2) to probability distributions n system), or it can be the spatial translation of the
374 Pressure and Equilibrium States in Ergodic Theory

configuration of a system on an infinite lattice (gen- probability distributions is compact in the weak
eralizing the point of view from subsection “Sys- topology, the coarsest topology on M ðXÞ for
tems on a Finite Lattice” above). In the latter case which n 7! hf, ni is continuous for all f  C(X),
one usually looks at the shift on the two-sided shift “Measure Preserving Systems”, subsection “Exis-
space Aℤ, for which the theory is nearly identical. tence of Invariant Measures.” (Note that in func-
On Aℕ one can define a metric d by tional analysis this is called the weak-* topology.)
Henceforth we will use both terms, “measure” and
d x, y ≔2N ðx,yÞ “distribution,” if we talk about probability
distributions.
where N x, y ≔ min fk  ℕ : xk 6¼ yk g: A measure n on X is invariant if expectations of
observables are unchanged under the shift, that is, if
ð7Þ
h f ∘s, ni ¼ h f , ni
Hence dðx, yÞ ¼ 1 if and only if x0 6¼ y0, and
for all bounded measurable f : X ! ℝ:
dðx, xÞ ¼ 0 upon agreeing that N ðx, xÞ ¼ 1 and
21 ¼ 0. Equipped with this metric, Aℕ becomes
The set of all invariant measures is denoted by
a compact metric space and s is easily seen to be a
M s ðXÞ. As a closed subset of M ðXÞ it is compact
continuous surjection of Aℕ. Finally, if X is a
in the weak topology. Of special importance
closed subset of Aℕ, we call the restriction
among all invariant measures n are the ergodic
s : X ! X, which is again a continuous surjection,
ones which can be characterized by the property
a shift dynamical system. We remark that
that, for all bounded measurable f : X ! ℝ,
d generates on Aℕ the product topology of the
discrete topology on A, just as many variants of n1
d do. For more details ▶ “Symbolic Dynamics”. 1
lim f sk x ¼ h f , n i
As usual, C(X) denotes the space of real-valued n!1 n k¼0
ð8Þ
continuous functions on X equipped with the for n  a:e: ðalmosteveryÞ x,
supremum norm kk1.
that is, for a set of x of n-measure one. They are the
Invariant Measures indecomposable “building blocks” of all other
A probability distribution n (or simply distribu- measures in M s ðXÞ , “Measure Preserving Sys-
tion) on X is a Borel probability measure on X. It is tems” or ▶ “Ergodic Theorems.” The almost
unambiguously specified by its values n[a0. . .an1] everywhere convergence in (8) is Birkhoff’s ergo-
(n  ℕ, ai  A) on cylinder sets dic theorem ▶ “Ergodic Theorems,” the constant
limit characterizes the ergodicity of n.
½a0 . . . an1 
≔fx  X : xi ¼ ai for all i ¼ 0, . . . , n  1g: Entropy of Invariant Measures
We give a brief account of the definition and basic
Any bounded and measurable f : X ! ℝ properties of the entropy of an invariant measure
(in particular any f  C(X)) can be integrated by n. For details and the generalization of this con-
any distribution n. To stress the linearity of the cept to general dynamical systems, we refer to
integral in both, the integrand and the integrator, ▶ “Entropy in Ergodic Theory” or Katok and
we use the notation Hasselblatt (1995), and to Katok (2007) for an
historical account.
Let n  M s ðXÞ . For each n > 0 the cylinder
h f , ni≔ f dn:
X probabilities n[a0. . .an1] give rise to a probabil-
ity distribution on the finite set An , see section
In probabilistic terms, hf, ni is the expectation “Warming Up: Thermodynamic Formalism for
of the observable f under n. The set M ðXÞ of all Finite Systems,” so
Pressure and Equilibrium States in Ergodic Theory 375

H n ðnÞ≔ A Short Digression on Complexity

Kolmogorov (1983) and Chaitin (1987) intro-
 n½a0 . . . an1  log n½a0 . . . an1  duced the concept of complexity of an infinite
a0 , ..., an1  A
sequence of symbols. Very roughly it is defined
as follows: First, the complexity K(x0. . .xn1) of
is well defined. Invariance of n guarantees that the
a finite word in An is defined as the bit length of
sequence (Hn(n))n>0 is subadditive, that is, Hkþn
the shortest program that causes a suitable gen-
(n)  Hk(n) þ Hn(n), and an elementary argument
eral purpose computer (say a PC or, for the
shows that the limit
mathematically minded reader, a Turing
1 machine) to print out this word. Then the com-
hðnÞ≔ lim H ðnÞ  ½0, log jAj ð9Þ plexity of an infinite sequence is defined as
n!1 n n
K ðxÞ≔ lim supn!1 1n K ðx0 . . . xn1 Þ . Of course,
exists and equals the infimum of the sequence. We the definition of K(x0. . .xn1) depends on the
simply call it the entropy of n. (Note that for particular computer, but as any two general pur-
general subshifts X many of the cylinder sets pose computers can be programmed to simulate
[a0. . .an1]  X are empty. But, because of the each other (by some finite piece of software), the
continuity of the function t 7! t log t at t ¼ 0, we limit K ðxÞ is machine independent. It is the
may set 0 log 0 ¼ 0, and, hence, this does not optimal compression factor for long initial
affect the definition of Hn(n).) pieces of a sequence x that still allows complete
The entropy h(n) of an ergodic measure n can reconstruction of x by an algorithm. Brudno
be observed along a “typical” trajectory. That is (1983) showed:
the content of the following theorem, sometimes If X  Aℕ and n  M s ðXÞ is ergodic, then
called the “ergodic theorem of information the- K ðxÞ ¼ log1 2 hðnÞ for n-a. e. x  X.
ory” (▶ “Entropy in Ergodic Theory”).
Entropy as a Function of the Measure
Theorem (Shannon-McMillan-Breiman Theorem)
An important technical remark for the further devel-
1 opment of the theory is that the entropy function
lim log n½x0 . . . xn1 
n!1 n h : M s ðXÞ ! ½0, 1Þ is upper semicontinuous.
¼ hðnÞ for n  a: e: x: ð10Þ This means that all sets {n : h(n)  t} with t  ℝ
are closed and hence compact. In particular, upper
Observe that (9) is just the integrated version of semicontinuous functions attain their supremum.
this statement. A slightly weaker reformulation of Indeed, suppose a sequence nk  M s ðXÞ converges
this theorem (again for ergodic n) is known as the weakly to some n  M s ðXÞ and h(nk)  t for all k so
“asymptotic equipartition property.” that also 1n Hn ðnk Þ  t for all n and k. As Hn(n) is an
expression that depends continuously on the prob-
Asymptotic Equipartition Property Given abilities of the finitely many cylinders [a0. . .an1]
(arbitrarily small) ϵ > 0 and α > 0, one can, for and as the indicator functions of these sets are
each sufficiently large n, partition the set An into a continuous, 1n H n ðnÞ ¼ lim k!1 1n H n ðnk Þ  t ,
set T n of typical words and a set E n of exceptional hence h(n)  t in the limit n ! 1.
words such that each a0 . . . an1  T n satisfies A word of caution seems in order: the entropy
function is rarely continuous. For example, on
enðhðnÞþaÞ  n½a0 . . . an1   enðhðnÞaÞ ð11Þ the full shift X ¼ Aℕ each invariant measure,
whatever its entropy is, can be approximated in
the weak topology by equidistributions on peri-
and the total probability a0 ...an1  E n n½a0 . . . an1  odic orbits. But all these equidistributions have
of the exceptional words is at most ε. entropy zero.
376 Pressure and Equilibrium States in Ergodic Theory

The Variational Principle: A Global entropy, and that means by the macroscopic
Characterization of Equilibrium state of the system. (In contrast, the word “state”
was used in the above section on finite systems to
Usually, a dynamical systems model of a “phys- designate microscopic states.)
ical” system consists of a state space and a map As, for each n  M s ðXÞ, the functional
(or a differential equation) describing the f 7! h(n) þ hf, ni is affine on C(X), the pres-
dynamics. An invariant measure for the system sure functional P : C(X) ! ℝ, which, by defini-
is rarely given a priori. Indeed, many (if not tion, is the pointwise supremum of these
most) dynamical systems arising in this way functionals, is convex. It is therefore instructive
have uncountably many ergodic invariant mea- to fit equilibrium states into the abstract frame-
sures. This limits considerably the “practical work of convex analysis (Israel 1979; Keller
value” of Birkhoff’s ergodic theorem (8) or the 1998; Moulin-Ollagnier 1985; Walters 1992).
Shannon-McMillan-Breiman theorem (10): not To this end recall the identities in (4) that iden-
only do the limits in these theorems depend on tify, for finite systems, equilibrium states as gra-
the invariant measure n, but also the sets of dients of the pressure function p : ℝ|A| ! ℝ and
points for which the theorems guarantee almost guarantee that p is twice differentiable and
everywhere convergence are practically disjoint strictly convex. In the present setting where
for different n and n0 in M s ðXÞ . Therefore, a P is defined on the Banach space C(X), differen-
choice of n has to be made which reflects the tiability and strict convexity are no more
original modeling intentions. We will argue in guaranteed, but one can show:
this and the next sections that a variational prin-
ciple with a judiciously chosen “observable” Equilibrium States as (Sub)-gradients
may be a useful guideline – generalizing the f  M s ðXÞ is an equilibrium state for f if and
observations for finite systems collected in the only if m is a subgradient (or tangent functional)
corresponding section above. As announced ear- for P at f, i.e., if P(f þ c)  P(f)  hc, mi for all
lier we restrict again to shift dynamical systems, c  C(X). In particular, f has a unique equilib-
because they are rather universal models for rium state m if and only if P is differentiable at f
many other systems. with gradient m, i.e., if,

Equilibrium States 1
lim t!0 ðPðf þ tcÞ  PðfÞÞ
We define the pressure of an observable f  C(X) t
as ¼ hc, mi for all c  CðXÞ: ð13Þ

PðfÞ≔supfhðnÞ þ hf, ni : n  M s ðXÞg: ð12Þ Let us see how equilibrium states on X ¼ Aℕ can
directly be obtained from the corresponding equilib-
Since M s ðXÞ is compact and the functional rium distributions on finite sets An introduced in
n 7! h(n) þ hf, ni is upper semicontinuous, the subsection “Systems on a Finite Lattice.” Define
supremum is attained – not necessarily at a unique f(n) : An ! ℝ by f(n)(a0. . .an1) ≔ f(a0. . .an1
measure as we will see (which is remarkably a0. . .an1. . .), denote by Un the corresponding
different from what happens in finite systems). global observable on An, and let mn be the equilib-
Each measure n for which the supremum is rium distribution on An that maximizes H(m) þ
attained is called an equilibrium state for f. Here hUn, mi. Then all weak limit points of the “approx-
the word “state” is used synonymously with “dis- imative equilibrium distributions” mn on An are equi-
tribution” or “measure” – a reflection of the fact librium states on Aℕ.
that in “well-behaved cases,” as we will see in the This can be seen as follows: Let the measure m
next section, this measure is uniquely determined on Aℕ be any weak limit point of the mn. Then,
by the constraint(s) under which it maximizes given ϵ > 0 there exists k  ℕ such that
Pressure and Equilibrium States in Ergodic Theory 377

1 Theorem (Variational Principle for the Pressure)

hðmÞ þ hf, mi  H k ðmÞ þ hf, mi  ϵ
k
1 1
 H k ðmn Þ þ fðnÞ , mn  2ϵ PðfÞ ¼ lim sup Pn ðfÞ
k n!1 n ð14Þ
S n fð z Þ
where Pn ðfÞ≔ log z  En e :
for arbitrarily large n, because kf  f(n)k1 ! 0
as n ! 1 by construction of the f(n). As the mn
To prove the “” direction of this identity we just
are invariant under cyclic coordinate shifts (see
have to show that 1n H n ðnÞ þ hf, ni  1n Pn ðfÞ for
subsection “Systems on a Finite Lattice”), it
each n  M s ðXÞ or, after multiplying by n, Hn(n) þ
follows from the subadditivity of the entropy
hSnf, ni  Pn(f). But Jensen’s inequality implies:
that
Hn ðnÞ þ hSn f, ni
1
hðmÞ þ hf, mi  ðHn ðmn Þ þ hU n , mn iÞ  2ϵ  n½a0 . . . an1 
n
k a0 , ..., an1  A
 log jAj:
n sup eSn fðxÞ : x  ½a0 . . . an1 
log
Hence, for each n  M s ðXÞ, n½a0 . . . an1 

1  log sup eSn fðxÞ : x  ½a0 . . . an1 

hðmÞ þ hf, mi  ðH n ðnÞ þ hUn , niÞ  2ϵ a0 , ..., an1  A
n
k
 log jAj ! hðnÞ þ hf, ni  2ϵ ¼ log z  En eSn fðzÞ ¼ Pn ðfÞ:
n
For the reverse inequality consider the discrete
as n ! 1, and we see that m is indeed an equilib-
Gibbs distributions
rium state on Aℕ.

pn ≔z  En dz exp ðPn ðfÞ þ Sn fðzÞÞ

The Variational Principle
In subsection “Equilibrium Distributions and
the Gibbs Property,” the pressure of a finite on the finite sets En, where dz denotes the unit
system was defined as a certain supremum and point mass in z. One might be tempted to think
then identified as the logarithm of the normaliz- that all weak limit points of the measures πn are
ing constant for the Gibbsian representation of already equilibrium states. But this need not be the
the corresponding equilibrium distribution. We case because there is no good reason that these
are now going to approximate equilibrium states limits are shift invariant. Therefore, one forces
by suitable Gibbs distributions on finite subsets invariance of the limits by passing to measures
of X. As a by-product the pressure P(f) is char- mn defined by h f , mn i≔ 1n n1 k¼0 f ∘s , pn . Weak
k

acterized in terms of the logarithms of the nor- limits of these measures are obviously shift invari-
malizing constants of these approximating ant, and a more involved estimate we do not
distributions. Let Sn fðxÞ≔fðxÞ þ fðsxÞ þ present here shows that each such weak limit m
. . . þ fðsn1 xÞ: From each cylinder set satisfies h(m) þ hf, mi  P(f).
[a0. . .an1] we can pick a point z such that We note that the same arguments work for any
Sn fðzÞ is the maximal value of Snf on this set. other sequence of sets En which contain exactly one
We denote the collection of the |A|n points we point from each cylinder. So there are many ways to
obtain in this way by En. Observe that En is not approximate equilibrium states, and if there are
unambiguously defined, but any choice we more than one equilibrium state, there is generally
make will do. no guarantee that the limit is always the same.
378 Pressure and Equilibrium States in Ergodic Theory

Nonuniqueness of Equilibrium States: An and hence equilibrium states need not exist, that
Example is, the supremum in (12) need not be attained by
Before we turn to sufficient conditions for the any invariant measure. A well-known sufficient
uniqueness of equilibrium states in the next sec- property that guarantees the upper semicontinuity
tion, we present one of the simplest nontrivial of the entropy function is the expansiveness of the
examples for nonuniqueness of equilibrium states. system, see, for example, Ruelle (1973): a contin-
Motivated by the so-called Fisher-Felderhof drop- uous transformation T of a compact metric space
let model of condensation in statistical mechanics is positively expansive, if there is a constant γ > 0
(Fisher and Felderhof 1970; Fisher 1967), such that for any two points x and y from the space
Hofbauer (1977) studies an observable f on X ¼ there is some n  ℕ such that Tnx and Tny are at
{0, 1}ℕ defined as follows: Let (ak) be a sequence least a distance γ apart. If T is a homeomorphism
of negative real numbers with limk!1ak ¼ 0. Set one says it is expansive, if the same holds for some
sk ≔ a0 þ . . . þ ak. For k  1 denote n  ℤ. The previous results carry over without
Mk ≔fx  X : x0 ¼ . . . ¼ xk1 ¼ 1, xk ¼ 0g and changes (although at the expense of more compli-
M0 ≔fx  X : x0 ¼ 0g, and define cated proofs) to general expansive systems. The
variational principle (14) holds in the very general
fðxÞ≔ak for x  Mk and fð11 . . .Þ ¼ 0: context where T is a continuous action of ℤdþ on a
compact Hausdorff space X. This was proved in
Then f : X ! ℝ is continuous, so that there Misiurewicz (1976) in a simple and elegant way.
exists at least one equilibrium state for f. In the monograph (Moulin-Ollagnier 1985) it is
Hofbauer proves that there is more than one equi- extended to amenable group actions.
1 sk
librium state if and only if k¼0 e ¼ 1 and
1
k¼0 ð k þ 1 Þe sk
< 1. In that case P(f) ¼ 0, so
one of these equilibrium states is the unit mass The Gibbs Property: A Local
δ11. . ., and we denote the other equilibrium state Characterization of Equilibrium
by m1, so h(m1) þ hf, m1i ¼ 0. In view of (13) the
pressure function is not differentiable at f. In this section we are going to see that, for a
What does the pressure function β 7! P(βf) sufficiently regular potential f on a topologically
look like? As h(δ11. . .) þ hβf, δ11. . .i ¼ 0 for all β, mixing subshift of finite type, one has a unique
P(βf)  0 for all β. Observe now that fðxÞ  0 equilibrium state which has the “Gibbs property.”
with equality only for x ¼ 11 . . . This implies that This property generalizes formula (5) that we
hf, mi < 0 for all m  M s ðXÞ different from δ11. . .. derived for finite lattices. Subshifts of finite type
From this we can conclude: are the symbolic models for Axiom
A diffeomorphisms, as we shall see later on.
• P(βf)  P(f) ¼ 0 for β > 1, so
P(βf) ¼ 0 for β  1. Subshifts of Finite Type
• P(βf)  h(m1) þ hβf, m1i ¼ h(m1) þ hf, m1i  We start by recalling what is a subshift of finite
(1  β)hf, m1i ¼  (1  β)hf, m1i. type and refer the reader to ▶ “Symbolic Dynam-
ics” or Lind and Marcus (1995) for more details.
It follows that, at β ¼ 1, the derivative from the Given a “transition matrix” M ¼ (Mab)a,b  A
right of P(βf) is zero, whereas the derivative from whose entries are 0’s or 1’s, one can define a
the left is at most hf, m1i < 0. subshift XM as the set of all sequences x  Aℕ
such that Mxi xiþ1 ¼ 1 for all i  ℕ. This is called
More on Equilibrium States a subshift of finite type or a topological Markov
In more general dynamical systems the entropy chain. We assume that there exists some integer p0
function is not necessarily upper semicontinuous such that M p has strictly positive entries for all
Pressure and Equilibrium States in Ergodic Theory 379

p  p0. This means that M is irreducible and We now make several comments on this
aperiodic. This property is equivalent to the prop- theorem.
erty that the subshift of finite type is topologically
mixing. A general subshift of finite type admits a • The Gibbs property (16) gives a uniform control
decomposition into a finite union of transitive of the measure of all cylinders in terms of their
sets, each of which being a union of cyclically “energy.” This strengthens considerably the
permuted sets on which the appropriate iterate is asymptotic equipartition property (11) that we
topologically mixing. In other words, topologi- recover if we restrict (16) to the set of mf measure
cally mixing subshifts of finite type are the build- 1 where Birkhoff’s ergodic Theorem (8) applies,
ing blocks of subshifts of finite type. and use the identity hf, mfi  P(f) ¼  h(mf).
• Gibbs measures on topologically mixing sub-
The Gibbs Property for a Class of Regular shifts of finite type are ergodic (and actually
Potentials mixing in a strong sense) as can be inferred
The class of regular potentials we consider is that from Ruelle’s Perron-Frobenius Theorem see
of “summable variations.” We denote by vark(f) the next subsection.
the modulus of continuity of f on cylinders of • Suppose that there is another invariant measure
length k  1, that is, m0 satisfying (16), possibly with a constant C0
different from C. It is easy to verify that m0 ¼ fm
vark ðfÞ≔sup fðxÞ  f y : x  ½y0 . . . yk1  : for some m-integrable function f by using (16)
and the Radon-Nikodym Theorem. Shift invari-
ance imposes that, m-a. e., f ¼ f ∘ s. Then the
If vark(f) ! 0 as k ! 1, this means that f is
ergodicity of m implies that f is a constant m-a. e.,
(uniformly) continuous with respect to the dis-
thus m0 ¼ m; see Bowen (2017).
tance (7). We impose the stronger condition
• One could define a Gibbs state by saying that it
1 is an invariant measure m satisfying (16) for a
vark ðfÞ < 1: ð15Þ given continuous potential f. If one does so, it
k¼1 is simple to verify that such a m must also be an
equilibrium state. Indeed, using (16), one can
We can now state the main result of this
deduce that hf, mi þ h(m)  P(f). The con-
section.
verse need not be true in general, see subsec-
The Gibbs state of a summable potential Let
tion “More on Hofbauer’s Example” below.
XM be a topologically mixing subshift of finite
But the summability condition (15) is indeed
type. Given a potential f : XM ! ℝ satisfying
sufficient for the coincidence of Gibbs and
the summability condition (15), there is a
equilibrium states. A proof of this fact can be
(probability) measure mf supported on XM, that
found in Ruelle (2004) or Keller (1998).
we call a Gibbs state. It is the unique s-invariant
measure which satisfies the following property:
Ruelle’s Perron-Frobenius Theorem
There exists a constant C > 0 such that, for all
The powerful tool behind the theorem in the pre-
x  XM and for all n  1,
vious subsection is a far-reaching generalization
of the classical Perron-Frobenius theorem for irre-
mf ½x0 . . . xn1 
C1  ducible matrices. Instead of a matrix, one intro-
exp ðSn fðxÞ  nPðfÞÞ
duces the so-called transfer operator, also called
 C: “ Gibbs property” ð16Þ the “Perron-Frobenius operator” or “Ruelle’s
operator,” which acts on a suitable Banach space
Moreover, the Gibbs state mf is ergodic and is of observables. It is D. Ruelle (1968) who first
also the unique equilibrium state of f, that is, the introduced this operator in the context of one-
unique invariant measure for which the supremum dimensional lattice gases with exponentially
in (12) is attained. decaying interactions. In our context, this
380 Pressure and Equilibrium States in Ergodic Theory

corresponds to Hölder continuous potentials: Hn njmf

these are potentials satisfying vark(f)  cθk for n½a0 . . . an1 
some c > 0 and θ  (0, 1). A proof of “Ruelle’s ≔ n½a0 . . . an1  log
a0 , ..., an1  A
mf ½a0 . . . an1 
Perron-Frobenius Theorem” can be found in
Bowen (1974b, 2017). It was then extended to ð17Þ
potentials with summable variations in Walters
(1975). We refer to the book of V. Baladi (2000) with the convention 0 log (0/0) ¼ 0. Now the
for a comprehensive account on transfer operators relative entropy of n given mf is defined as
in dynamical systems.
We content ourselves to define the transfer 1
h njmf ≔ lim sup H n njmf :
operator and state Ruelle’s Perron-Frobenius The- n!1 n

orem. Let ℒ : C(XM) ! C(XM) be defined by

(By applying Jensen’s inequality, one verifies that
h(n|mf)  0.) In fact the limit exists and can be
ðℒf ÞðxÞ≔ efðyÞ f y
computed quite easily using (16):
y  s1 x

¼ efðaxÞ f ðaxÞ: h njmf ¼ PðfÞ  hf, ni  hðnÞ: ð18Þ

a  A:Mða, x0 Þ¼1

To prove this formula, we first make the fol-

(Obviously, ax≔ax0 x1 . . .) lowing observation. It can be easily verified that
Theorem (Ruelle’s Perron-Frobenius Theo- the inequalities in (16) remain the same when Snf
rem) Let X M be a topologically mixing subshift is replaced by the “locally averaged” energy
of finite type. Let f satisfy condition (15). There fn ≔ðn½x0 . . . xn1 Þ1 ½x0 ...xn1  Sn fðyÞdnðyÞ for any
exist a number l > 0, h  C(XM), and n  M ðXÞ
cylinder with n[x0. . .xn  1] > 0. Cylinders with n
such that h > 0, hh, ni ¼ 1, ℒh ¼ lh, ℒn ¼ ln,
measure zero do not contribute to the sum in (17).
where ℒ is the dual of ℒ. Moreover, for all
We can now write that
f  C(XM),
1
kln ℒn f  h f , ni  hk1 ! 0, as n ! 1:  log C
n
1 1 1
By using this theorem, one can show that   H n njmf þ PðfÞ  hSn f, ni  Hn ðnÞ
n n n
mf ≔ hn satisfies (16) and l ¼ eP(f). 1
 log C:
Let us remark that for potentials which are such n
that fðxÞ ¼ fðx0 , x1 Þ (i.e., potentials constant on
cylinders of length 2), ℒ can be identified with a To finish we use that hSnf, ni ¼ nhf, ni
|A| |A| matrix and the previous theorem boils (by the invariance of n) and we apply (9) to obtain
down to the classical Perron-Frobenius theorem
1 1
for irreducible aperiodic matrices (Seneta lim H njmf ¼ PðfÞ  hf, ni  lim H n ð nÞ
n!1 n n n!1 n
2006). The corresponding Gibbs states are noth-
¼ PðfÞ  hf, ni  hðnÞ
ing but Markov chains with state space
A (Chap. 3 in Georgii 1988). We shall take which proves (18).
another point of view below (subsection “Mar- The variational principle revisited We can
kov Chains Over Finite Alphabets”). reformulate the variational principle in the case
of a potential satisfying the summability
Relative Entropy condition (15):
We now define the relative entropy of an invariant
measure n  M s ðXM Þ given a Gibbs state mf as
h njmf ¼ 0 if and only if n ¼ mf , ð19Þ
follows. We first define
Pressure and Equilibrium States in Ergodic Theory 381

that is, given mf, the relative entropy h(|mf), as a It is not difficult to verify that the total number
function on M s ðXM Þ, attains its minimum only at of periodic sequences of period n equals the trace
mf. of the matrix Mn, that is, we have the formula
Indeed, by (18) we have h(n|mf) ¼ P(f) 
hf, ni  h(n). We now use (12) and the fact that m

mf is the unique equilibrium state of f to Cardfx  XM : sn x ¼ xg ¼ trðMn Þ ¼ lni ,

i¼1
conclude.
where l1, . . ., lm are all the eigenvalues
More Properties of Gibbs States of M. Asymptotically, of course, card
Gibbs states enjoy very good statistical properties. fx  XM : sn x ¼ xg ¼ enPð0Þ þ O jl0 j , where
n
Let us mention only a few. They satisfy the
l0 is the second largest (in absolute value) eigen-
“Bernoulli property,” a very strong qualitative
value of M.
mixing condition (Bowen 1974b, 2017; Walters
The measure of maximal entropy, call it m0,
1975). The sequence of random variables ( f ∘ sn)n
describes the distribution of periodic points in
satisfies the central limit theorem (Chernov 2002;
XM: one can prove (Bowen 1974a; Katok and
Coelho and Parry 1990; Pollicott 2000) and a large
Hasselblatt 1995) that for any cylinder B XM
deviation principle if f is Hölder continuous
(Eizenberg et al. 1994; Keller 1998; Kifer 1990;
Cardfx  B : sn x ¼ xg
Young 1990). Let us emphasize the central role lim ¼ m0 ðBÞ:
n!1 Cardfx  XM : sn x ¼ xg
played by relative entropy in large deviations. (The
deep link between thermodynamics and large devi- In other words, the finite atomic measures that
ations is described in Lewis and Pfister (1995) in a assign equal weights 1=Cardfx  XM : sn x ¼ xg
much more general context.) Finally, the so-called to each periodic point in XM with period
multifractal analysis can be performed for Gibbs n weakly converges to m0, as n ! 1. Each such
states, see, for example, Pesin and Weiss (1997). measure has zero entropy while h(m0) ¼ P(0) > 0,
so the entropy is not continuous on the space of
invariant measures. It is, however, upper-
Examples on Shift Spaces semicontinuous (see subsection “Entropy as a
Function of the Measure”).
Measure of Maximal Entropy and Periodic In fact, it is possible to approximate any Gibbs
Points state mf on XM in a similar way by finite atomic
If the observable f is constant zero, an equilibrium measures on periodic orbits, if one assigns
state simply maximizes the entropy. It is called weights properly (see, e.g., Theorem 20.3.7 in
measure of maximal entropy. The quantity Pð0Þ ¼ Katok and Hasselblatt (1995)).
supfhðnÞ : n  M s ðXÞg is called the topological
entropy of the subshift s : X ! X. When X is a Markov Chains Over Finite Alphabets
subshift of finite type XM with irreducible and ape- Let Q ¼ (qa,b)a, b  A be an irreducible stochastic
riodic transition matrix M, there is a unique measure matrix over the finite alphabet A. It is well known
of maximal entropy, see, for example, Lind and (see, e.g., Seneta 2006) that there exists a unique
Marcus (1995). As a Gibbs state it satisfies (16). probability vector π on A that defines a stationary
By summing over all cylinders [x0. . .xn  1] allowed Markov measure nQ on X ¼ Aℕ by nQ ½a0 . . .
by M, it is easy to see that the topological entropy an1  ¼ pa0 qa0 a1 . . . qan2 an1 : We are going to
P(0) is the asymptotic exponential growth rate of the
identify nQ as the unique Gibbs distribution
number of sequences of length n that can occur as
m  M s ðX Þ that maximizes entropy under
initial segments of points in XM. This is obviously
the constraints m[ab] ¼ m[a]qab, that is, hfab, mi ¼
identical to the logarithm of the largest eigenvalue of
0 (a, b  A), where .
the transition matrix M.
382 Pressure and Equilibrium States in Ergodic Theory

Indeed, as m is a Gibbs measure, there are equilibrium state, one of them being δ11. . ., which
βab  ℝ (a, b  A) and constants P  ℝ, cannot be a Gibbs state for any continuous f.
C > 0 such that

m½x0 . . . xn1 
C1  Examples from Differentiable Dynamics
a,b  A bab fn ðxÞ  nP
ab
exp
In this section we present a number of examples to
C ð20Þ which the general theory developed above does not
apply directly but only after a transfer of the theory
for all x  Aℕ and all n  ℕ. Let from a symbolic space to a manifold. We restrict to
r ab ≔ exp bab  b0  A bab0 qab0  P . Then the examples where the results can be transferred
denominator in (20) equals r x0 x1 . . . r xn2 xn1 , and it because those aspects of the smooth dynamics we
follows that m is equivalent to the stationary Mar- focus on can be studied as well on a shift dynamical
kov measure defined by the (nonstochastic) system that is obtained from the original one via
matrix (rab)a,b  A. As m is ergodic, m is this Mar- symbolic coding. (We do not discuss the coding
kov measure, and as m satisfies the linear con- process itself which is sometimes far from trivial,
straints m[ab] ¼ m[a]qab, we conclude that m ¼ nQ. but we focus on the application of the Gibbs and
equilibrium theory.) There are alternative
The Ising Chain approaches where instead of the results the concepts
Here the task is to characterize all “spin chains” in and (partly) the strategies of proofs are transferred
x  f1, þ1gℕ (or, more commonly, {1, +1}ℤ) to the smooth dynamical systems. This has led both
which are as random as possible with the con- to an extension of the range of possible applications
straint that two adjacent spins have a prescribed of the theory and to a number of refined results
probability p 6¼ 12 to be identical. With fðxÞ≔x0 x1 (because some special features of smooth systems
this is equivalent to requiring that x is typical for a necessarily get lost by transferring the analysis to a
Gibbs distribution mβf where β ¼ β( p) is such that completely disconnected metric space).
hf, mβfi ¼ 2p  1. It follows that there is a In the following examples, T denotes a
constant C > 0 such that for each n  ℕ and (possibly piecewise) differentiable map of a
any two “spin patterns” a ¼ a0 . . . an1 and b ¼ compact smooth manifold M. Points on the man-
b0 . . . bn1 ifold are denoted by u and v. In all examples there
is a Hölder continuous coding map π : X ! M
mbf ½a0 . . . an1  from a subshift of finite type X onto the manifold
log  b Na  Nb  C,
mbf ½b0 . . . bn1  which respects the dynamics, that is, T ∘ π ¼
π ∘ s. This factor map π is “nearly” invertible in
where N a and N b are the numbers of identical the sense that the set of points in M with more
adjacent spins in a and b, respectively. than one preimage under π has measure zero for
all T-invariant measures we are interested
More on Hofbauer’s Example in. Hence such measures m on M correspond
We come back to the example described in sub- unambiguously to shift invariant measures m ¼
section “Nonuniqueness of Equilibrium States: m∘p1 . Similarly observables f on M and f ¼
An Example.” It is easy to verify that in that f∘p on X are related.
example varkþ1(f) ¼ |ak|. For instance, if
ak ¼  1/(k þ 1)2 there is a unique Gibbs/equi- Uniformly Expanding Markov Maps of the
librium state. If ak ¼  3 log ((k þ 1)/k) for k  1 Interval
and a0 ¼  log 1j¼1 j3 , then from Hofbauer A transformation T on M ≔ [0, 1] is called a
(1977) we know that f admits more than one Markov map, if there are 0 ¼ u0 < u1 < . . . < uN ¼ 1
Pressure and Equilibrium States in Ergodic Theory 383

such that each restriction Tjðui1 ,ui Þ is strictly mono- and minimizing this relative entropy just amounts to
tone, C1þr for some r > 0, and maps (ui1, ui) onto maximizing h(m) þ hf, mi for f ¼  log |T0| ∘ π.
a union of some of these N monotonicity intervals. As the results on Gibbs distributions from section
It is called uniformly expanding if there is some “The Gibbs Property: A Local Characterization of
k  ℕ such that l ≔ infx|(Tk)0(x)| > 1. It is not Equilibrium” apply, we conclude that
difficult to verify that the symbolic coding of such a
system leads to a topological Markov chain over m½a0 . . . an1 
C1  C
the alphabet A ¼ {1, . . ., N}. To simplify the dis- jI a0 . . . an1 j
cussion we assume that the transition matrix M of
this topological Markov chain is irreducible and for some C > 0. So the unique T-invariant mea-
aperiodic. sure m that minimizes the relative entropy hðmjmÞ
Our goal is to find a T-invariant measure m is equivalent to Lebesgue measure m. (The exis-
represented by m  M s ðXM Þ which minimizes tence of an invariant probability measure equiva-
the relative entropy to Lebesgue measure on [0, 1] lent to m is well known, also without invoking
entropy theory. It is guaranteed by a “Folklore
1 Theorem” (Jakobson 2002).)
hðmjmÞ≔ lim m½a0 . . . an1 
n!1 n
a0 , ..., an1  f1, ..., N g
Interval Maps with an Indifferent Fixed Point
m½a0 . . . an1  The presence of just one point x  [0, 1] such that
log ,
nn ½a0 . . . an1  T0(x) ¼ 1 dramatically changes the properties of the
system. A canonical example is the map
where nn ½a0 . . . an1 ≔jI a0 ...an1 j . (Recall that, Tα : x 7! x(1 þ 2αxα) if x  [0, 1/2[ and x 7! 2x  1
without insisting on invariance, this would just be if x  [1/2, 1]. We have T0(0) ¼ 1, that is, 0 is an
the Lebesgue measure itself.) The existence of the indifferent fixed point. For α  [0, 1[ this map
limit will be justified below – observe that m is not admits an absolutely continuous invariant proba-
a Gibbs state as v is in Eq. (17). The argument rests bility measure dm(x) ¼ h(x)dx, where h(x)~xα
on the simple observation (implied by the uniform when x ! 0 (Thaler 1980). In the physics literature,
expansion and the piecewise Hölder-continuity of this type of map is known as the “Manneville-
T0) that T has bounded distortion, that is, that there Pomeau” map. It was introduced as a model of
is a constant C > 0 such that for all n  ℕ, transition from laminar to intermittent behavior
a0. . .an  1  {1, . . ., N}n and u  I a0 ...an1 holds (Pomeau and Manneville 1980). In Gaspard and
Wang (1988) the authors construct a piecewise
C1  jI a0 ...an1 j  ðT n Þ0 ðuÞ  C, or, equivalently, affine version of this map to study the complexity
jI a0 ...an1 j of trajectories (in the sense of subsection “A Short
C1   C,
exp Sn fðuÞ Digression on Complexity”). This gives rise to a
countable state Markov chain. In Wang (1989) the
ð21Þ close connection to the Fisher-Felderhof model and
Hofbauer’s example (see subsection “Non-
where fðuÞ≔  log jT 0 ðuÞj. (Observe the similar- uniqueness of Equilibrium States: An Example”)
ity between this property and the Gibbs prop- was realized. We refer to Sarig (2001) for recent
erty (16).) Assuming bounded distortion we have developments and a list of references.
at once
Axiom A Diffeomorphisms, Anosov
n1
1 Diffeomorphisms, Sinai-Ruelle-Bowen
hðmjmÞ ¼ lim H n ðmÞ  f∘sk , m
n!1 n Measures
k¼0
The first spectacular application of the theory of
¼ hðmÞ  hf, mi,
Gibbs measures to differentiable dynamical
384 Pressure and Equilibrium States in Ergodic Theory

systems was Sinai’s approach to Anosov 2. The union of all stable manifolds through
diffeomorphisms via Markov partitions (Sinai points of Ω is a subset of M with positive
1968) that allowed one to code the dynamics of volume.
these maps into a subshift of finite type and to 3. The pressure PTjO fðuÞ ¼ 0.
study their invariant measures by methods from
equilibrium statistical mechanics (Sinai 1972) that
had been developed previously by Dobrushin, In this case the unique equilibrium and Gibbs
Lanford, and Ruelle (Dobrushin 1968a, b, c, state m+ of T|Ω is called the Sinai-Ruelle-Bowen
1969; Lanford and Ruelle 1969). Not much later (SRB) measure of T|Ω. It is uniquely characterized
this approach was extended by Bowen (1970) to by the identity hTjO ðmþ Þ ¼  fðuÞ , mþ . (For all
Smale’s Axiom A diffeomorphisms (and to other T-invariant measures on Ω one has “<”
Axiom A flows by Bowen and Ruelle (1975)); instead of “¼.”)
see also Ruelle (1976). The interested reader can Further properties of SRB measures Suppose
consult, for example, Young (2002) for a survey, PTjO fðuÞ ¼ 0 and let m+ be the SRB measure.
and either (Bowen 2017) or (Chernov 2002) for
details.
Both types of diffeomorphisms act on a 1. For a set of points u  M of positive volume
smooth compact Riemannian manifold M and we have:
are characterized by the existence of a compact
n1
T-invariant hyperbolic set Λ  M. Their basic 1
lim f T k u ¼ h f , mþ i:
properties are described in detail in the contri- n!1 n k¼0
bution ▶ “Ergodic Theory: Basic Examples and
Constructions.” Very briefly, the tangent bundle (Indeed, because of (ii) of the above character-
over Λ splits into two invariant subbundles – a ization, this holds for almost all points of the
stable one and an unstable one. Correspond- union of the stable manifolds through points of
ingly, through each point of Λ there passes a Ω.)
local stable and a local unstable manifold 2. Conditioned on unstable manifolds, m+ is abso-
which are both tangent to the respective sub- lutely continuous to the volume measure on
spaces of the local tangent space. The unstable unstable manifolds.
derivative of T, that is, the derivative DT
restricted to the unstable subbundle, is uniformly In the special case of transitive Anosov
expanding. Its Jacobian determinant, denoted by diffeomorphisms, the whole manifold is a hyper-
J(u), is Hölder continuous as a function on Λ. bolic set and Ω ¼ M. Because of transitivity,
Hence the observable f(u) ≔  log |J(u)| ∘ π is property (ii) from the characterization of attractors
Hölder continuous, and the Gibbs and equilib- is trivially satisfied, so there is always a unique
rium theory apply (via the symbolic coding) SRB measure m+. As T1 is an Anosov
to the diffeomorphism T (modulo possibly diffeomorphism as well – only the roles of stable
a decomposition of the hyperbolic set into and unstable manifolds are interchanged – T1
irreducible and aperiodic components, called has a unique SRB measure m which is the unique
basic sets, that can be modeled by topologically equilibrium state of T1 (and hence also of T ) for
mixing subshifts of finite type). The main results f(s) ≔ log |J(s)|. One can show:
are: SRB measures for Anosov diffeomorphisms
Characterization of attractors The following The following assertions are equivalent:
assertions are equivalent for a basic set Ω  Λ:
1. m+ ¼ m.
1. Ω is an attractor, that is, there are arbitrarily 2. m+ or m is absolutely continuous w.r.t the
small neighborhoods U  M of Ω such that volume measure on M.
TU U. 3. For each periodic point u ¼ Tnu  M, |J(u)| ¼ 1,
where J denotes the determinant of DT.
Pressure and Equilibrium States in Ergodic Theory 385

We remark that, similarly to the case of Mar- (hyperbolic) Julia set, such that (J, ℂ, T ) is a
kov interval maps, the unstable Jacobian of Tn at conformal repeller.
u is asymptotically equivalent to the volume of the Conformal repellers J are in general fractal
“n-cylinder” of the Markov partition around u. So sets and one can measure their “degree of
the maximization of h(m) þ hf(u), mi by the SRB fractality” by means of their Hausdorff dimen-
measure m+ can again be interpreted as the mini- sion, dimH(J ). Roughly speaking, one computes
mization of the relative entropy of invariant mea- this dimension by covering the set J by balls with
sures with respect to the normalized volume, and radius less than or equal to δ. If Nδ(J ) denotes the
the fact that P(f(u)) ¼ 0 in the Anosov (or more cardinality of the smallest such covering, then we
generally attractor) case means that m+ is as close expect that
to being absolutely continuous as it is possible for
a singular measure. This is reflected by the above ~
N d ðJ ÞddimH ðJÞ , as d ! 0:
properties (a) and (b).
We emphasize the meaning of property We refer the reader to ▶ “Ergodic Theory:
(a) above: it tells us that the SRB measure m+ is Fractal Geometry” or Falconer (2003), Pesin
the only physically observable measure. Indeed, (1997) for a rigorous definition (based on
in numerical experiments with physical models, Carathéodory’s construction) and for more infor-
one picks an initial point u  M “at random” mation on fractal geometry.
(i.e., with respect to the volume or Lebesgue mea- Bowen’s formula relates dimH(J) to the unique
sure) and follows its orbit Tku, k  0.
zero of the pressure function b 7! P bf where
f≔  ð log jT 0 jÞjJ . It is not difficult to see that
Bowen’s Formula for the Hausdorff indeed this map has a unique zero for some pos-
Dimension of Conformal Repellers itive β.
Just as nearby orbits converge towards an attrac-
By property (1), Sn f  const  n log a, which
tor, they diverge away from a repeller. Conformal
implies (by (13)) that db d
P bf ¼ f, mb 
repellers form a nice class of systems which can
 log a < 0 . As P(0) equals the topological
be coded by a subshift of finite type. The con-
entropy of J, that is, the logarithm of the largest
struction of their Markov partitions is much sim-
eigenvalue of the matrix M associated to the Mar-
pler than that of Anosov diffeomorphisms, see, for
kov partition, P(0) is strictly positive. Therefore,
example, Zinsmeister (2000).
(recall that the pressure function is continuous)
Let us recall the definition of a conformal
there exists a unique number β0 > 0 such that
repeller before giving a fundamental example.
Given a holomorphic map T : V ! ℂ where P b0 f ¼ 0.
V ℂ is open and J a compact subset of ℂ, one It turns out that this unique zero is precisely
says that (J, V, T ) is a conformal repeller if dimH(J ):
Bowen’s formula The Hausdorff dimension of
1. There exist C > 0, α > 1 such that |(Tn)0 (z)|  J is the unique solution of the equation
Cαn for all z  J, n  1. P bf ¼ 0, β  ℝ; in particular
2. J ¼ \ n  1Tn (V).
3. for any open set U such that U \ J 6¼ 0, there


exists n such that Tn(U \ J ) J. P dimH ðJ Þf ¼ 0 :

From the definition it follows that T(J ) ¼ J and This formula was proven in Ruelle (1982) for
T1(J) ¼ J. a general class of conformal repellers after the
A fundamental example is the map T : z ! z2 þ seminal paper (Bowen 1979). The main tool is a
c, c  ℂ being a parameter. It can be shown that distortion estimate very similar to (21). A simple
for jcj < 14 there exists a compact set J, called a exposition can be found in Zinsmeister (2000).
386 Pressure and Equilibrium States in Ergodic Theory

Nonequilibrium Steady States and Some Ongoing Developments and

Entropy Production Future Directions

SRB measures for Anosov diffeomorphisms and As we saw, many dynamical systems with uni-
Axiom A attractors have been accepted recently form hyperbolic structure (e.g., Anosov maps,
as conceptual models for nonequilibrium steady axiom A diffeomorphisms) can be modeled by
states in nonequilibrium statistical mechanics. Let subshifts of finite type over a finite alphabet. We
us point out that the word “equilibrium” is used in already mentioned in subsection “Interval Maps
physics in a much more restricted sense than in with an Indifferent Fixed Point” the typical exam-
ergodic theory. Only diffeomorphisms preserving ple of a map of the interval with an indifferent
the natural volume of the manifold (or a measure fixed point, whose symbolic model is still a sub-
equivalent to the volume) would be considered as shift of finite type, but with a countable alphabet.
appropriate toy models of physical equilibrium The thermodynamic formalism for such systems
situations. In the case of Anosov diffeomorphisms is by now well developed (Fiebig et al. 2002;
this is precisely the case if the “forward” and Gurevich and Savchenko 1998; Sarig 1999,
“backward” SRB measures m+ and m coincide. 2001, 2003) and used, for example, for multi-
Otherwise, the diffeomorphism models a situation dimensional piecewise expanding maps (Buzzi
out of equilibrium, and the difference between m+ and Sarig 2003). An active line of research is
and m can be related to entropy production and related to systems admitting representations by
irreversibility. symbolic models called “towers” constructed by
Gallavotti and Cohen (1995; Gallavotti 1996) using “inducing schemes.” The fundamental
introduced SRB measures as idealized models of example is the class of one-dimensional unimodal
nonequilibrium steady states around 1995. In maps satisfying the “Collet-Eckmann condition.”
order to have as firm a mathematical basis as A first attempt to develop thermodynamic formal-
possible they made the “chaotic hypothesis” that ism for such systems was made in Bruin and
the systems they studied behave like transitive Keller (1998) where existence and uniqueness of
Anosov systems. Ruelle (1996) extended his equilibrium measures for the potential function
approach to more general (even nonuniformly) fb ðuÞ ¼ b log jT 0 ðuÞj with β close to 1 was
hyperbolic dynamics; see also his reviews established. Very recently, new developments in
(Ruelle 1998, 2003) for more recent accounts this direction appeared, see, for example, (Bruin
discussing also a number of related problems; and Todd 2008, 2009; Pesin and Senti 2008).
see by Rondoni and Mejía-Monasterio (2007), A largely open field of research concerns a new
too. The importance of the Gibbs property of branch of nonequilibrium statistical mechanics,
SRB measures for the discussion of entropy pro- the so-called chaotic scattering theory, namely
duction was also highlighted in Jiang et al. (2000), the analysis of chaotic systems with various open-
where it is shown that for transitive Anosov ings or holes in phase space, and the
diffeomorphisms the relative entropy h(m+|m) corresponding repellers on which interesting
equals the average entropy production rate invariant measures exist. We refer the reader to
hlog|J|, m+i of m+ where J denotes again the Chernov (2002) for a brief account and references
Jacobian determinant of the diffeomorphism. to the physics literature. The existence of
In particular, the entropy production rate is (generalized) steady states on repellers and the
zero if, and only if, h(m+|m) ¼ 0, that is, using so-called escape rate formula have been observed
coding and (19), if, and only if, m+ ¼ m. According numerically in a number of models. So far, little
to subsection “Axiom A Diffeomorphisms, Anosov has been proven mathematically, except for
Diffeomorphisms, Sinai-Ruelle-Bowen Measures,” Anosov diffeomorphisms with special holes
this is also equivalent to m+ or m being absolutely (Chernov 2002) and for certain nonuniformly
continuous with respect to the volume measure. hyperbolic systems (Bruin et al. 2010).
Pressure and Equilibrium States in Ergodic Theory 387

Bibliography Dobrushin RL (1968c) The problem of uniqueness of a

Gibbsian random field and the problem of phase tran-
Baladi V (2000) Positive transfer operators and decay of sitions. Funct Anal Appl 2:302–312
correlations. Advanced series in nonlinear dynamics, Dobrushin RL (1969) Gibbsian random fields. The general
vol 16. World Scientific, Singapore case. Funct Anal Appl 3:22–28
Bowen R (1970) Markov partitions for Axiom Eizenberg A, Kifer Y, Weiss B (1994) Large deviations for
A diffeomorphisms. Am J Math 92:725–747 ℤd-actions. Commun Math Phys 164(3):433–454
Bowen R (1974a/1975) Some systems with unique equi- Falconer K (2003) Fractal geometry. Mathematical founda-
librium states. Math Syst Theory 8:193–202 tions and applications, 2nd edn. Wiley, San Francisco
Bowen R (1974b/75) Bernoulli equilibrium states for Fiebig D, Fiebig U-R, Yuri M (2002) Pressure and equilib-
Axiom A diffeomorphisms. Math Syst Theory 8: rium states for countable state Markov shifts. Israel
289–294 J Math 131:221–257
Bowen R (1979) Hausdorff dimension of quasicircles. Inst Fisher ME (1967) The theory of condensation and the
Hautes Études Sci Publ Math 50:11–25 critical point. Physics 3:255–283
Bowen R (2017) Equilibrium states and the ergodic theory Fisher ME, Felderhof BU (1970) Phase transition in one-
of Anosov diffeomorphisms. Lecture notes in mathe- dimensional clusterinteraction fluids: IA. Thermody-
matics, vol 470, 2nd edn (1st edn 1975). Springer, namics, IB. Critical behavior. II. Simple logarithmic
Berlin model. Ann Phys 58:177–280
Bowen R, Ruelle D (1975) The ergodic theory of Axiom Gallavotti G (1996) Chaotic hypothesis: Onsager reciproc-
A flows. Invent Math 29:181–202 ity and fluctuation-dissipation theorem. J Stat Phys 84:
Brudno AA (1983) Entropy and the complexity of the 899–925
trajectories of a dynamical system. Trans Mosc Math Gallavotti G, Cohen EGD (1995) Dynamical ensembles in
Soc 2:127–151 stationary states. J Stat Phys 80:931–970
Bruin H, Keller G (1998) Equilibrium states for Gaspard P, Wang X-J (1988) Sporadicity: between periodic
S-unimodal maps. Ergod Theory Dyn Syst 18(4): and chaotic dynamical behaviors. Proc Natl Acad Sci
765–789 USA 85:4591–4595
Bruin H, Todd M (2008) Equilibrium states for potentials Georgii H-O (1988) Gibbs measures and phase transitions.
with sup ’  inf ’ < htop( f ). Commun Math Phys In: de Gruyter studies in mathematics. de Gruyter, Berlin,
283(3):579–611. https://doi.org/10.1007/s00220-0- p9
008-0596-0 Gurevich BM, Savchenko SV (1998) Thermodynamic for-
Bruin H, Todd M (2009) Equilibrium states for interval maps: malism for symbolic Markov chains with a countable
the potential t log |Df|. Ann Sci Éc Norm Sup (4) 42(4): number of states. Russ Math Surv 53(2):245–344
559–600. https://doi.org/10.24033/asens.2103 Hofbauer F (1977) Examples for the nonuniqueness of the
Bruin H, Demers M, Melbourne I (2010) Existence and equilibrium state. Trans Am Math Soc 228:223–241
convergence properties of physical measures for certain Israel R (1979) Convexity in the theory of lattice gases.
dynamical systems with holes. Ergod Theory Dyn Syst Princeton series in physics. Princeton University Press
30(3):687–728. https://doi.org/10.1017/ Jakobson M, Świątek (2002) One-dimensional maps. In:
S0143385709000200 Handbook of dynamical systems, vol 1A. North-
Buzzi J, Sarig O (2003) Uniqueness of equilibrium mea- Holland, Amsterdam, pp 321–407
sures for countable Markov shifts and multi- Jaynes ET (1989) Papers on probability, statistics and
dimensional piecewise expanding maps. Ergod statistical physics. Kluwer, Dordrecht
Theory Dyn Syst 23(5):1383–1400 Jiang D, Qian M, Qian M-P (2000) Entropy production and
Chaitin GJ (1987) Information, randomness & incomplete- information gain in Axiom A systems. Commun Math
ness. Papers on algorithmic information theory. World Phys 214:389–409
Scientific series in computer science, vol 8. World Sci- Katok A (2007) Fifty years of entropy in dynamics:
entific, Singapore 1958–2007. J Mod Dyn 1(4):545–596
Chernov N (2002) Invariant measures for hyperbolic Katok A, Hasselblatt B (1995) Introduction to the modern
dynamical systems. In: Handbook of dynamical sys- theory of dynamical systems. Encyclopaedia of mathe-
tems, vol 1A. North-Holland, pp 321–407 matics and its applications, vol 54. Cambridge Univer-
Coelho Z, Parry W (1990) Central limit asymptotics for sity Press, Cambridge
shifts of finite type. Israel J Math 69(2):235–249 Keller G (1998) Equilibrium states in ergodic theory. In:
Dobrushin RL (1968a) The description of a random London Mathematical Society student texts, vol 42.
field by means of conditional probabilities and con- Cambridge University Press, Cambridge
ditions of its regularity. Theory Probab Appl 13: Kifer Y (1990) Large deviations in dynamical systems and
197–224 stochastic processes. Trans Am Math Soc 321:505–524
Dobrushin RL (1968b) Gibbsian random fields for lattice Kolmogorov AN (1983) Combinatorial foundations of
systems with pairwise interactions. Funct Anal Appl 2: information theory and the calculus of probabilities.
292–301 Uspekhi Mat Nauk 38:27–36
388 Pressure and Equilibrium States in Ergodic Theory

Lanford OE, Ruelle D (1969) Observables at infinity and Ruelle D (1996) Positivity of entropy production in non-
states with short range correlations in statistical equilibrium statistical mechanics. J Stat Phys 85:1–23
mechanics. Commun Math Phys 13:194–215 Ruelle D (1998) Smooth dynamics and new theoretical
Lewis JT, Pfister C-E (1995) Thermodynamic probability ideas in nonequilibrium statistical mechanics. J Stat
theory: some aspects of large deviations. Russ Math Phys 95:393–468
Surv 50:279–317 Ruelle D (2003) Extending the definition of entropy to
Lind D, Marcus B (1995) An introduction to symbolic nonequilibrium steady states. Proc Nat Acad Sci USA
dynamics and coding. Cambridge University Press, 100(6):3054–3058
Cambridge Ruelle D (2004) Thermodynamic formalism: the mathe-
Misiurewicz M (1976) A short proof of the variational matical structures of equilibrium statistical mechanics,
principle for a ℤNþ action on a compact space. In: Cambridge Mathematical Library, 2nd edn. Cambridge
International conference on dynamical systems in University Press, Cambridge
mathematical physics (Rennes, 1975), Astérisque, Sarig O (1999) Thermodynamic formalism for countable
No. 40, Soc Math France, pp 147–157 Markov shifts. Ergod Theory Dyn Syst 19(6):
Moulin-Ollagnier J (1985) Ergodic theory and statistical 1565–1593
mechanics. In: Lecture notes in mathematics, vol 1115. Sarig O (2001) Phase transitions for countable Markov
Springer, Berlin shifts. Commun Math Phys 217(3):555–577
Pesin Y (1997) Dimension theory in dynamical systems. Sarig O (2003) Existence of Gibbs measures for countable
Contemporary views and applications. University of Markov shifts. Proc Am Math Soc 131(6):1751–1758
Chicago Press, Chicago Seneta E (2006) Non-negative matrices and Markov
Pesin Y, Senti S (2008) Equilibrium measures for maps chains. In: Springer series in statistics. Springer
with inducing schemes. J Mod Dyn 2(3):397–430. Sinai JG (1968) Markov partitions and
https://doi.org/10.3934/jmd.2008.2.397 C-diffeomorphisms. Funct Anal Appl 2:61–82
Pesin Y, Weiss (1997) The multifractal analysis of Gibbs Sinai JG (1972) Gibbs measures in ergodic theory. Russ
measures: motivation, mathematical foundation, and Math Surv 27(4):21–69
examples. Chaos 7(1):89–106 Thaler M (1980) Estimates of the invariant densities of
Pollicott M (2000) Rates of mixing for potentials of sum- endomorphisms with indifferent fixed points. Israel
mable variation. Trans Am Math Soc 352(2):843–853 J Math 37(4):303–314
Pomeau Y, Manneville P (1980) Intermittent transition to Walters P (1975) Ruelle’s operator theorem and
turbulence in dissipative dynamical systems. Commun g-measures. Trans Am Math Soc 214:375–387
Math Phys 74(2):189–197 Walters P (1992) Differentiability properties of the pres-
Rondoni L, Mejía-Monasterio C (2007) Fluctuations in sure of a continuous transformation on a compact met-
nonequilibrium statistical mechanics: models, mathe- ric space. J Lond Math Soc (2) 46(3):471–481
matical theory, physical mechanisms. Nonlinearity Wang X-J (1989) Statistical physics of temporal intermit-
20(10):R1–R37 tency. Phys Rev A 40(11):6647–6661
Ruelle D (1968) Statistical mechanics of a one-dimensional Young L-S (1990) Large deviations in dynamical systems.
lattice gas. Commun Math Phys 9:267–278 Trans Am Math Soc 318:525–543
Ruelle D (1973) Statistical mechanics on a compact set Young L-S (2002) What are SRB measures, and which
with ℤn action satisfying expansiveness and specifica- dynamical systems have them? Dedicated to David
tion. Trans Am Math Soc 185:237–251 Ruelle and Yasha Sinai on the occasion of their 65th
Ruelle D (1976) A measure associated with Axiom birthdays. J Stat Phys 108(5–6):733–754
A attractors. Am J Math 98:619–654 Zinsmeister M (2000) Thermodynamic formalism and
Ruelle D (1982) Repellers for real analytic maps. Ergod holomorphic dynamical systems. SMF/AMS texts and
Theory Dyn Syst 2(1):99–107 monographs, vol 2. American Mathematical Society
 valued (or complex-valued) function on X.
Parallels Between Topological Then AN ð f , xÞ ¼ N1 N1
n¼0 f ðT xÞ is called the
n

Dynamics and Ergodic Theory ergodic average. The multiple ergodic aver-
ages (or called “nonconventional averages”)
Wen Huang, Song Shao and Xiangdong Ye are the following ones
CAS Wu Wen-Tsun Key Laboratory of N1
p ðnÞ p ð nÞ p ðnÞ
Mathematics, and Department of Mathematics,
1
N f 1 T 11 x f 2 T 22 x . . . f d T 2d x ,
n¼0
University of Science and Technology of China,
Hefei, Anhui, China where T1, T2, . . ., Td are invertible and act on
a probability space ðX, X , mÞ , f1, . . ., fd are
functions on X, and p1, . . ., pd are integral
Article Outline polynomials.
Recurrent point A point is recurrent if it is in its
Glossary own future.
Definition of the Subject Topological dynamics The study of the asymp-
Introduction and History totic behaviors of a homeomorphism from a
Recurrence and Other Dynamical Properties topological space to itself. Such a self-
Entropy Theory homeomorphism of a topological space is
Structure Theorems and Multiple Ergodic called a topological dynamical system.
Averages Transitive point A point is a transitive point
Further Directions when every point is in its future.
References Transitivity and minimality A system is transi-
tive when it contains at least one transitive
Glossary point. A system is minimal when every point
is a transitive point.
Ergodicity A measure-preserving system is ergo-
dic if it is essentially indecomposable, in the
sense that given any invariant measurable set,
Definition of the Subject
either the set or its complement has measure 0.
Measure-preserving transformation A map
By topological dynamics, it is “the study of trans-
from a measure space to itself such that for
formation groups with respect to those topological
each measurable subset of the space, it has
properties whose prototype occurred in classical
the same measure as its inverse image under
dynamics” (Gottschalk and Hedlund 1955). If in
the map. Such a measure-preserving map on a
this definition the adjective “topological” is replaced
measure space is called a measure-preserving
by “measure-theoretic,” then one obtains a descrip-
system.
tion of measurable dynamics, also called ergodic
Measure-theoretic and topological entropy A
theory.
nonnegative (possibly infinite) real number
In this entry we try to present theorems illus-
which describes the complexity of a measure-
trating the analogy between topological and mea-
preserving transformation and topological
surable dynamics, and theorems showing drastic
dynamics.
contrast between them.
(Multiple) Ergodic average Let ðX, X , m, T Þ be
a measure-preserving system and f be a real-

© Springer Science+Business Media, LLC, part of Springer Nature 2023 389

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_748
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2020
https://doi.org/10.1007/978-3-642-27737-5_748-1
390 Parallels Between Topological Dynamics and Ergodic Theory

Introduction and History Topological dynamics owes its origin to the

classic work of Poincaré and Birkhoff (1927). It
In its broadest interpretation a dynamical system is was Poincaré who first formulated and solved
the study of the qualitative properties of actions of problems of dynamics as problems in topology.
groups on spaces. The space has some structure An autonomous system of ordinary differential
(topological, measurable, smooth manifold, etc.) equations satisfying conditions that ensure
and the action preserves the structure (so acts as a uniqueness and extendibility of solutions deter-
continuous map or a homeomorphism, a measure mines a flow, that is, a one-parameter transforma-
preserving transformation, a differential map, etc.). tion group. Birkhoff was the first to note that many
The dynamical systems are then divided into notions and results obtained in the theory of
branches including topological dynamics, ergodic autonomous differential equations can be
theory, differentiable dynamical systems, etc., extended to the case of flows in abstract spaces.
according to the spaces and the actions. In this Introducing the very important concepts of mini-
entry we mainly present some aspects of the brother mal sets and of recurrent and central motions, he
branches: topological dynamics and ergodic theory. laid the foundation of the general theory of flows
The terminology used in both branches is or, it is usually called, the topological theory of
almost the same. For example, one talks about dynamical systems. This theory was further devel-
transitivity or minimality, topological weak/mild/ oped by lots of mathematicians (Nemytskiǐ 1949;
strong mixing, distality, rigidity, disjointness in Nemytskiǐ and Stepanov 1949). At the end of
topological dynamics. At the same time, one 1940s Gottschalk and Hedlund and many others
speaks ergodicity, measurable weak/mild/strong proposed a broad and very natural generalization
mixing, distality, rigidity, disjointness in ergodic of classical dynamical systems, namely, the
theory. These have led to the formulation of anal- notion of a topological transformation group
ogous theorems. However, they have up until now (Nemytskiǐ and Stepanov 1949; Gottschalk and
remained analogies; the proofs involved being Hedlund 1955; Ellis 1969). The part of the theory
entirely different and neither directly deducible of transformation groups devoted to the study of
from the other, for example, the proofs of the those notions whose prototype occurred previ-
structure theorems in both theories. This, of ously in classical dynamics, such as various
course, is not surprising since the methods used types of motions, minimal and limit set, recur-
in one are topological and in the other measure rence, and stability, is called topological
theoretic. We remark that some topological state- dynamics.
ments are proved first by using the measure theo- Ergodic theory is not one of the classical math-
retical methods and then one finds the topological ematical disciplines and its name, in contrast to,
arguments, for example, the fact that the a mini- for example, measure theory and number theory,
mal system with a trivial regionally proximal rela- does not indicate its subject. However, its origin
tion of order d is a d-step pro-nilpotent system can be described quite precisely. Ergodic theory
(Host et al. 2010; Cai and Shao 2019; Candela has its roots in Maxwell’s and Boltzmann’s kinetic
2017a, b; Gutman et al. 2019a, b, 2020). More- theory of gases. The first important result was
over, some topological statements are proved by found by Poincaré by the end of the nineteenth
using the measure theoretical methods, and one century, which now is called Poincaré recurrence
has no topological proofs until now, for example, theorem. Ergodic theory was born as a mathemat-
the statement that positive entropy implies the ical theory around 1930 by the ground breaking
existence of a proper asymptotic pair (Blanchard works of von Neumann and Birkhoff. Two other
et al. 2002b). major contributions should be mentioned. One is
This kind research arises from the study of the the introduction of the notion of entropy by Kol-
ordinary differential equation. We will skip the mogorov and Sinai at the end of the 1950s.
description and briefly mention the histories of Another is Ornstein’s theorem saying that the
topological dynamics and ergodic theory. entropy is a complete invariant for Bernoulli
Parallels Between Topological Dynamics and Ergodic Theory 391

shifts. In the 1970s, Furstenberg showed how to Let (X,T) be a t.d.s. By Krylov-Bogolioubov
translate questions in combinatorial number the- theorem, there is at least one invariant probability
ory into ergodic theory. This inspired a new line of measure m on the Borel s-algebra B X . In such a
research, which ultimately led to stunning recent way, ðX, B X , T, mÞ can be viewed as an
results in combinatorial number theory. m.p.s. which also explains why some topological
For simplicity in this article we only consider results can be proved via ergodic theory.
ℤ-actions. Thus, by a topological dynamics On the other hand, for any m.p.s. ðX, X , m, T Þ
(t.d.s. for short) we mean a pair (X,T), where X is one can find some t.d.s. X, T and an invariant
a compact metrizable space with metric r and
measure m on the Borel s-algebra B such that
T : X ! X is a homeomorphism. By a measure- X
preserving system (m.p.s. for short) we mean a ðX, X , m, T Þ is isomorphic to X, B , m, T
X
quadruple ðX, X , m, T Þ, where ðX, X , mÞ is a Borel (Furstenberg 1981). X, T is called a topological
probability space and T, T–1 : X ! X are both model of ðX, X , m, T Þ. This makes the possibility
measurable and measure preserving, that is, to use topological methods in the study of ergodic
T 1 X ¼ X ¼ T X and m(A) ¼ m(T1A) for each theory.
A  X. We try our best to exhibit the similarity or
There are two types of problems in topological parallels between two theories, from the notations
dynamics and ergodic theory. The first type can be to the results. There is a very nice survey by
viewed as the internal problems which concern Glasner and Weiss (2005) on the same subject,
with understanding the homeomorphisms or mea- here we will review the subject in some new
sure preserving transformations and trying to angles and present some new progress.
decide when two of them are conjugate or Due to our personnel interest, knowledge, and
isomorphic. The second type is the applications the space available we can only choose some
the theories to other branches of mathematics or in aspects to do so. Thus, the materials we choose
physics. For the internal problems, the usual way are restricted, and are the ones we are familiar
is to look for conjugacy or isomorphic invariants. with. In the first part of the entry we will discuss
Such invariants could be a property (e.g., mini- dynamical properties where most of results are
mality, ergodicity), or is an assignment of some classical. In the second part we focus on the
object (e.g., a number like entropy, a group or a study of entropy, particularly on the so-called
structure). For more details, see (Walters 1982). local entropy theory which is relatively new. In
Let us now introduce the notions of conjugacy the third part we will consider the structure theo-
and isomorphism. Let (X,T) and (Y,S) be two rems and their applications to the combinatorial
t.d.s. A continuous map p : X ! Y is called a number theory which are the hot points of current
homomorphism or factor map between (X,T) and research and developing very fast. We hope in
(Y,S) if it is onto and p ∘ T ¼ S ∘ p. In this case we such a way we could cover certain old and new
say (X,T) is an extension of (Y,S) or (Y,S) is a factor results, and some materials from the internal prob-
of (X,T). When p is a homeomorphism, we then lems to the applications of the theories.
say that (X,T) and (Y,S) are conjugate, so in this The authors would like to thank B. Kra,
case we review the two systems are the same. B. Host, and M. Lemańczyk for valuable com-
To define the same notion in ergodic theory, we ments which improved the writing of the survey
keep in mind that a null set is negligible. Suppose significantly.
ðXi , X i , mi Þ is a probability space and Ti : Xi ! Xi is
measure preserving, i ¼ 1, 2. We say that T2 is a
factor of T1 or T1 is an extension of T2 if there exists Recurrence and Other Dynamical
Mi  X i with mi(Mi) ¼ 1 and Ti(Mi)  Mi(i ¼ 1, 2) Properties
and there exists a measure preserving transforma-
tion f : M1 ! M2 with fT1(x) ¼ T2f(x) for any Given a dynamical system, we are interested in
x  M1. If f is 1-1 and f1 is also measure those points that can be observed repeatedly. This
preserving, then we say f is an isomorphism. leads to the notion of recurrence. The basic fact is
392 Parallels Between Topological Dynamics and Ergodic Theory

that any t.d.s. has a recurrent point and for any Minimality Versus Ergodicity
measurable set with positive measure there are
points returning to the original set infinitely Minimality and Transitivity
many times. We are particularly interested in The first recurrence property we will discuss is the
the systems that are “non-decomposable” in minimality. Recall that we only consider homeo-
some sense. Such systems are minimal or transi- morphisms. A t.d.s. is minimal if the orbit of any
tive systems in topological dynamics and ergo- point is dense in the space, where the orbit of x is
dic systems in ergodic theory. Below we will see the set {T nx : n  ℤ+}. Given a t.d.s. (X, T), we
that when studying dynamical properties, some say (Y, T) is a subsystem if Y is a closed subset of
interesting subsets of ℕ, ℤ+ ¼ ℕ [ {0} or ℤ will X and is invariant under T, that is, TY ¼ Y. The
be involved. This indicates that a dynamical minimality is equivalent to the statement that there
system has a close relation with the combinato- is no non-empty proper invariant closed subset.
rial number theory. We will find many evidence We say x is a periodic point if there is n  ℕ such
in the sequel. that T n(x) ¼ x. If x is a periodic point, then it is
Since many statements of the paper are better easy to see that the orbit of x is minimal. Basic
stated using the notion of a family, we now give examples of minimal t.d.s can be found in books
the definition. See Akin (1997), Furstenberg (Auslander 1988; Ellis 1969; Kurka 2003; de
(1967, 1981) for more details. Vries 1993), etc.
We define family on ℤ+, and it works for ℕ, ℤ, Now we consider a weaker notion than mini-
etc. A collection F of subsets of ℤ+ is a family if mality, namely transitivity. A t.d.s. (X,T) is tran-
it is hereditary upward, that is, F1  F2 and sitive (for a survey, see Kolyada and Snoha
F1  F imply F2  F. A family F is called proper (1997)) if there is some point x with dense orbit.
if it is neither empty nor the entire power set of We have
ℤ+, or, equivalently if ℤþ  F and 0 2 = F . Any


nonempty collection A of subsets of ℤ+ generates Theorem 2.1 Let (X,T) be a t.d.s. Then the fol-
a family lowing statements are equivalent:

F ðA Þ≔fF  ℤþ : F  A for some A  A g: 1. (X,T) is transitive.

2. For each non-empty open set U of X,
n
For a family F its dual is the family [1n¼1 T U is dense in X.
3. For every non-empty open sets U,V of X, there
F  ¼ F  ℤþ : F \ F 0 6¼ 0 for all F 0  F exists n > 0 with U \ T n V 6¼ 0.



¼ fF  ℤþ : ℤþ nF 2 = Fg
Let (X,T) be a t.d.s. x  X is a (positively)
recurrence point if there is a sequence ni ! + 1
It is not hard to see that if F is a family, then such that T ni x ! x, i ! 1. So, if x is recurrent,
then we can observe its motion near x infinitely
ðF  Þ ¼ F : many times. There are several ways to show the
following basic fact.
If a family F is closed under finite intersections
and is proper, then it is called a filter. A family F Theorem 2.2 (Birkhoff recurrence theorem).
has the Ramsey property if A ¼ A1 [ A2  F Any t.d.s. has a recurrence point.
implies that A1  F or A2  F . It is well known This theorem has an important generalization,
that a proper family has the Ramsey property if namely the multiple topological recurrence theo-
and only if its dual F  is a filter (Furstenberg rem (Furstenberg 1981). We mention that it is
1981). equivalent to the well-known van der Waerden’s
Parallels Between Topological Dynamics and Ergodic Theory 393

theorem (van der Waerden 1927; Furstenberg for every i  ℤ+; S is thick if it contains arbitrarily
1981). long runs of positive integers, that is, for every
When x is recurrent, the restriction of T to its n  ℕ there exists some an  ℤ+ such that {an,
orbit closure is onto and thus this orbit closure is a an + 1, . . ., an + n}  S. It is easy to show that the
transitive subsystem. A point is a transitive point dual family of syndetic subsets is the family of
if the orbit closure is dense in X. We say x is a thick subsets, and vice versa. The following fact
minimal point if the orbit closure of x is a minimal was due to Gottschalk (n.d.) (see also Gottschalk
subset. We have and Hedlund (1955)).

Theorem 2.3 Let (X,T) be a transitive t.d.s. Then Theorem 2.5 For any t.d.s. (X,T), x  X is a
the set of transitive points is a Gd subset of X. minimal point if and only if N(x, U) is syndetic for
If (X,T) is transitive and not minimal, then the set each neighborhood U of x.
of non-transitive points is dense in X. The fact that any t.d.s. has a minimal subset
We remark that there exist transitive follows from the Zorn’s lemma, and one can find a
t.d.s. whose non-transitive points are periodic constructive proof in Weiss (2000).
points (Downarowicz and Ye 2002). The minimality of some systems is important
The following well-known fact due to in the applications. For example the minimality of
Furstenberg (1981) tells us how recurrence is (Nd(X), hT  T  . . .  T, T  T2  . . .  Tdi) has
related to an IP-set. Let fpi g1
i¼1 be a sequence in an application to the simple proof of van der
ℕ. Define Waerden’s theorem (Glasner 2003), where
hT  T  . . .  T, T  T2  . . .  Tdi be the
k group generated by T  T  . . .  T,
FS ðfpi gÞ ¼ pi j : 1  i1 < . . . < ik , k  ℕ : T  T2  . . .  Td and Nd(X) is the orbit closure
j¼1
of (x,. . .,x) under the two actions (see section
“Structure Theorems and Multiple Ergodic Aver-
A subset F  ℕ is called an IP-set if it contains
ages”). The minimality of (x,. . ., x) under the face
some FS fpi g1 i¼1 : Denote the family of all group actions (see section “Structure Theorems
IP-sets by F ip. The well-known Hindman’s theo- and Multiple Ergodic Averages”) has an applica-
rem (Hindman 1974) states that if N1 [ . . . [ Nk is tion to the proof of the structure theorem of a
a partition of ℕ, then one of the cell contains an minimal t.d.s. (Shao and Ye 2012).
IP-set. It is equivalent to say that F ip has the Related to recurrence we may define a recur-
Ramsey property, and thus the dual family F ip is rence set. A  ℕ is a recurrence set if for any
a filter. Any set in F ip is called an IP*-set. t.d.s. (X, T) there is a point x  X and {ni}  A,
For a t.d.s. (X, T), x  X and U  X let i ! 1 such that T ni x ! x. Equivalently, a set A is
recurrent if and only if for any minimal t.d.s (X, T)
N ðx, U Þ ¼ fn  ℤþ : T n x  Ug: and non-empty open set U there is some n  A
such that U \ T n U 6¼ 0. We have the following


Theorem 2.4 Let (X, T) be a t.d.s. Then x is a fact (Furstenberg 1981).

recurrent point if and only if for each neighbor-
hood U of x, N(x, U) is an IP set. Theorem 2.6 A is a recurrence set if and only if
We remark that for each IP-set R, there is some for any syndetic set S, A \ ðS  SÞ 6¼ 0.


t.d.s. (X, T) such that there is a recurrent point A deep open question related to recurrence is
x  X and an open neighborhood U with that: If A is a recurrence set for rotation on a torus
R [ {0}  N(x, U) (Furstenberg 1981). of arbitrary dimension, is it a recurrence set? For
The minimality property is related the syndetic the related research see (Weiss 2000; Ellis and
subsets. A subset S of ℤ+ is syndetic if it has a Keynes 1972; Følner 1954; Katznelson 2001;
bound on the size of the gaps, that is, there is Huang and Ye 2012; Huang et al. 2016; Glasner
N  ℕ such that fi, i þ 1, . . . , i þ N g \ S 6¼ 0


1998; Host et al. 2016).

394 Parallels Between Topological Dynamics and Ergodic Theory

Ergodicity Theorem 2.8 (Poincaré recurrence theorem).

Here we will discuss the counterpart of minimality Let ðX, X , m, T Þ be an m.p.s. Then for any A  X
or transitivity in ergodic theory, namely with m(A) > 0, there is n  ℕ such that
ergodicity. m(A \ TnA) > 0.
We say an m.p.s. ðX, X , m, T Þ or m is ergodic if The above theorem also has an important gen-
B  X with T1B ¼ B, then m(B) ¼ 0 or m(X\B) eralization, namely the multiple ergodic recur-
¼ 0. For example, the irrational rotation on the rence theorem (Furstenberg 1981). The
circle is ergodic with respect to the Lebesgue generalized form and its relation to the well-
measure; the Gauss transformation is ergodic known Szemerédi’s theorem will be discussed
with respect the Gauss measure (recall that the later.
Gauss transformation T: [0, 1) ! [0, 1) is given This theorem allows us to define Poincaré
by T(x) ¼ 0 if x ¼ 0, and T(x) ¼ {1/x} if x 6¼ 0, sequences. A subset A  ℕ is a Poincaré sequence
where {y} denotes the fractional part of a positive if for any m.p.s. ðX, X , m, T Þ and for any B  X
real number y. And the Gauss measure m is given with m(B) > 0 there is n  A such that
by mðAÞ ¼ log1 2 A 1þx1
dx for any Borel set m(B \ TnB) > 0. The upper Banach density of
A  [0, 1)). Basic examples and constructions of a subset S of ℤ or ℤ+ is defined by
ergodic m.p.s. can be found in the survey paper
(Nicol and Petersen 2012), or books (Billingsley jS \ I j
BD ðSÞ ¼ lim sup ,
1965; Cornfeld et al. 1982; Katok and Hasselblatt jI j!1 jI j
1995). Those examples include translations on the
torus, symbolic dynamical systems, interval where I ranges over intervals of ℤ or ℤ+. We have
exchange transformations, Billiards, Gauss sys- Furstenberg (1981).
tems, Hamiltonian systems, horocycle flows,
etc. The ergodicity of some systems is important Theorem 2.9 A is a Poincaré sequence if and
in the applications. For example, the ergodicity of only if for any subset S of ℕ with positive upper
the Gauss transformation gives precise statistical Banach density, A \ (S  S) contains nonzero
information about the continued fraction digits of integers.
almost every real number. The ergodicity of the Thus, a Poincaré sequence is a recurrence set,
Liouville system implies the Chowla conjecture and the converse does not hold (Frantzikinakis
(Frantzikinakis 2017). and McCutcheon 2012). Examples of a Poincaré
The ergodicity of an m.p.s. has many equiva- sequence include IP-sets, p(ℤ), where p is a non-
lent statements. We list a few of them. constant polynomial with p(ℤ)  ℤ and p(0) ¼ 0,
see for example (Furstenberg 1981).
Theorem 2.7 Let ðX, X , m, T Þ be a m.p.s. Then Recurrence sets and Poincaré sequences have
the following statements are equivalent: higher order generalizations, we refer to
(Frantzikinakis and McCutcheon 2012;
1. T is ergodic. Furstenberg and Weiss 1978; Huang et al. 2016)
n
2. For A  X with m(A) > 0, m [1n¼1 T A ¼ 1. for a systematics study of the subject and the
3. For every A, B  X with m(A)m(B) > 0 there references therein.
exists n > 0 with m(A \ TnB) > 0.
4. For every A, B  X ,
Ergodic Decomposition
n1
Let ðX, X , m, T Þ be an m.p.s. Then m can be
1 disintegrated as m ¼ X mxdm(x), where mx are ergodic
lim mðA \ T n BÞ ¼ mðAÞmðBÞ:
n!1 n i¼0 measures. This disintegration is known as the ergodic
decomposition of m. The disintegration is character-
Similar to topological case we have the follow- ized by the following properties: a) for every
ing basic fact related to recurrence. f  L1 ðX, X , mÞ, f  L1 ðX, X , mx Þ for m-a.e. x  X,
Parallels Between Topological Dynamics and Ergodic Theory 395

and the map such that x 7! fdmx is in L1 ðX, X , mÞ; Theorem 2.10 (Halmos-von Neumann Theo-
b) for every f  L1 ðX, X , mÞ, m ð f jI ðT ÞÞ  rem). (Halmos and von Neumann 1942) An ergo-
ðxÞ ¼ fdmx for m-a.e. x  X, where I ðT Þ ¼ dic system has discrete spectrum if and only if it is
B  B X : m T 1 BDB ¼ 0 is the s-algebra of isomorphic to a rotation of a compact monothetic
T-invariant sets (see Glasner (2003)). group.
We note that in the topological setting we do
not have such a decomposition except for distal Equicontinuity
transformations defined below. For some discus- Let (X, T) be a t.d.s. and let C(X) be the vector
sion about this topic see Glasner (1994). space of all continuous ℂ-valued functions on X.
Then (C(X), kk1) is a complex Banach space.
Equicontinuity Versus Measurable Kronecker The Koopman operator fT: C(X) ! C(X) is
defined by fT(f) ¼ f ∘ T. We say 0 6¼ f  C(x) is
Kronecker Systems an eigenfunction for T if there exists l  ℂ such
For each m.p.s. ðX, X , m, T Þ we associate a that fTf ¼ lf. Call l the eigenvalue for
Koopman operator U: L2(m) ! L2(m) such that T corresponding to the eigenfunction f.
U f (x) ¼ f (Tx) for each f  L2(m). A complex We say (X, T) has topological discrete spectrum
number l is called an eigenvalue of T if there is if the smallest closed linear subspace of C(X)
0 6¼ f  L2(m) such that U f ¼ lf. The function f is containing the eigenfunctions of T is C(X), that is,
called an eigenfunction of T corresponding to the the eigenfunctions span C(X). A transitive t.d.s. (X,
eigenvalue l. It is immediate that T is ergodic if T) has topological discrete spectrum if and only if it
and only if 1 is a simple eigenvalue of T. is minimal and equicontinuous, that is, for each
An ergodic m.p.s. has discrete spectrum if e > 0, there is d > 0 such that r(x, y) < d implies
there is an orthonormal basis for L2(m) which that r(T nx, T ny) < e for any n  ℤ (Walters 1982).
consists of eigenfunctions of T. Each ergodic Note that the property of being equicontinuous
m.p.s. admits a maximal factor with discrete spec- does not depend on the choice of the metric d of
trum, called the Kronecker factor. Its higher order X. An important fact is that each minimal
generalization involving the rotation on t.d.s. admits a maximal equicontinuous factor
nilmanifold will be discussed in the later sections. (Xeq, Teq) (Ellis 1969; Ellis and Gottschalk 1960).
If f  L2(m) is an eigenfunction, then cl A generalization of this fact to the rotation on
{U f : n  ℤ} is a compact subset of L2(m).
n nilmanifolds will be discussed in the later section.
Generally, we say f almost periodic if cl By the following theorem, an equicontinuous
{Unf : n  ℤ} is compact in L2(m). It is well system can be viewed as the simplest t.d.s.
known that the set of all almost periodic functions
(denoted by Hc) is spanned by the set of eigen- Theorem 2.11 (Halmos-von Neumann Theo-
functions, and there exists a T-invariant rem). (Walters 1982) Let (X, T) be a t.d.s. Then
sub-s-algebra K m of B such that Hc ¼ (X, T) is minimal and equicontinuous if and only if
L2 X, K m , m (see (Hulse 1982; Zimmer 1976a)). (X, T) is topologically conjugate to a minimal
We call K m the Kronecker algebra of ðX, X , m, T Þ, rotation on a compact abelian metric group.
and X, K m , m, T is the Kronecker factor of In the study of topological dynamics, one of
ðX, X , m, T Þ (Kronecker n.d.). Thus T has discrete the first problems was to find the smallest closed
spectrum if H c ¼ L2 ðX, X , mÞ or equivalently invariant equivalence relation R(X) on (X, T) such
K m ¼ X. that (X/R(X), T) is equicontinuous. A natural can-
By the following theorem, an ergodic didate for R(X) is the so-called regionally proxi-
m.p.s. with discrete spectrum can be viewed as mal relation RP(X) introduced by Ellis and
the simplest ergodic system. Gottschalk (1960). It is a difficult problem to
396 Parallels Between Topological Dynamics and Ergodic Theory

find conditions under which RP(X) is an equiva- Topological Weak Mixing Versus Measurable
lence relation. Starting with Veech (1968), various Weak Mixing
authors, including MacMahon (1978), Ellis and
Keynes (1971), Bronstein (1979), etc., came up Topological Weak Mixing
with various sufficient conditions for RP(X) to be Transitivity and strong mixing are two basic
an equivalence relation. The generalization of notions of recurrence. There is a notion between
regionally proximal relation to higher orders will them which was introduced by Furstenberg and
be discussed in section “Structure Theorems and important to understand the dynamical properties.
Multiple Ergodic Averages.” We say a t.d.s. (X, T) is weakly mixing if
(X  X, T  T) is transitive. Furstenberg (1967)
showed that (X, T) is weakly mixing if and only if
Topological Mixing Versus Measurable Mixing
{N(U, V): U, V  X non-empty open subsets} is a
filter if and only if N(U, V) is thick for each open
Measurable Mixing
non-empty subsets U, V of X.
We say an m.p.s. ðX, X , m, T Þ is strongly mixing if
For a minimal system we can say more. Define
for any A, B  X, limn ! 1m(A \ TnB) ¼ m(A)m(B).
the upper density of A by
The stronger notions are the K-system (which
will be discussed later) and Bernoulli system
jA \ f0, 1, . . . , N  1gj
which we will not touch, see (Glasner 2003). dðAÞ ¼ lim sup :
N!1 N
We remark that a long open question is the
Rohlin’s conjecture: if ðX, X , m, T Þ is strongly
mixing, is it true that for all k  ℕ and A0, A1, Similarly we define dðAÞ. If dðAÞ ¼ d ðAÞ ¼ a,
. . ., Ak  X we then say that the density of A is a.

lim Theorem 2.12 Let (X, T) be a minimal t.d.s. The

n1 , ..., nk !1 following statements are equivalent.
n
mð A0 \ T 1 A1 \ T n1 n2 A2 \ . . . \ T n1 ...nk Ak Þ
¼ mðA0 Þ . . . mðAk Þ? 1. (X, T) is weakly mixing.
2. N(U, V) has density 1 for all U, V  X non-
For the progress on the conjecture, see empty open subsets (Huang and Ye 2004).
(Kalikow 1984; Ryzhikov 1993; Host 1991; 3. The maximal equicontinuous factor is trivial
Fayad and Kanigowski 2016; Kanigowski (Auslander 1988).
et al. 2017).
A nice fact due to Glasner (1994) is that if (X,
T) is minimal and weakly mixing, then there is a
Topological Mixing dense Gd set X0, such that the orbit closure of
A t.d.s. (X, T) is (topologically) strongly mixing if (x, . . ., x) is dense in Xd under the action of
N(U, V) is cofinite each pair of non-empty open T  T2  . . .  Td for each x  X0 (for a simple
subsets U and V of X, where proof see (Moothathu 2010; Kwietniak and
Oprocha 2012)). Recently, it was shown by
N ðU, V Þ ¼ fn  ℤþ : U \ T n V 6¼ 0g:


Huang et al. (2019c) that if (X, T) is minimal and

weakly mixing, then there is x  X such that
To get a corresponding notion of K-system in 2
T n x : n  ℤ is dense in X.
topological dynamics, Blanchard (1993) intro-
duced the notion of entropy pair which we will
discuss below. We would like to mention that a Measurable Weak Mixing
minimal topological K-system (in the definition of Similar to the topological case, there is a notion of
(Huang et al. 2005)) is strongly mixing. weak mixing between ergodicity and strong
Parallels Between Topological Dynamics and Ergodic Theory 397

mixing. An m.p.s. ðX, X , m, T Þ is weakly mixing if Topological Mild Mixing

ðX  X, X  X , m  m, T  T Þ is ergodic. For some time, people were looking for a suitable
notion of mild mixing in topological dynamics. In
Theorem 2.13 Let ðX, X , m, T Þ be an ergodic the independent works by Huang-Ye and Galsner-
m.p.s. The following statements are equivalent: Weiss in Huang and Ye (2004) and Glasner and
Weiss (2005), the authors defined topological
1. ðX, X , m, T Þ is weakly mixing. mildly mixing as follows: a t.d.s. (X, T) is topo-
2. For every A, B  X , limn m(A \ TnB) ¼ logical mildly mixing if its product with any tran-
m(A)m(B) outside a set for density zero, that sitive system is transitive. The following theorem
is, for each e < 0, {n  ℤ+ : |m(A \ TnB)  indicates that topological mildly mixing is also
m(A)m(B)| < e} has density 1. related to IP*-sets.
3. The constants are the only eigenfunctions for T,
that is, the Kronecker algebra Km is trivial. Theorem 2.15 Let (X, T) be a minimal t.d.s. Then
it is mildly mixing if and only if N(U, V) is an IP*-
Topological Mild Mixing Versus Measurable
set for each pair of non-empty open subsets U and
Mild Mixing
V of X.
There are several rigid notions in topological
Measurable Mild Mixing
dynamics. A t.d.s. (X, T) is (positively) rigid if
We now discuss a recurrence property between
there is ni ! + 1 such that T ni x ! x, i ! 1 for
weak mixing and strong mixing. It is well
each x  X, and it is uniformly rigid if T ni ! Id,
known that an m.p.s. ðX, X , m, T Þ is weakly mixing
i ! 1 in the uniform topology. We remark that
if and only if the product system
there exists a minimal weakly mixing rigid
ðX  Y, X  Y , m  n, T  SÞ is ergodic for every
t.d.s. which is not uniformly rigid (Korner 1987;
ergodic m.p.s. Y ¼ ðY, Y , n, SÞ with n(Y) ¼ 1. An
Glasner and Maon 1989). We also remark that
m.p.s. ðX, X , m, T Þ is called mildly mixing if the
each ergodic measurable rigid system has a topo-
product system ðX  Y, X  Y , m  n, T  SÞ is
logical model which is uniformly rigid (Donoso
ergodic for every ergodic m.p.s. Y ¼ ðY, Y , n, SÞ
and Shao 2017).
with n(Y)  1 (Furstenberg and Weiss 1977). It is
It was shown (Huang and Ye 2004) that if a
easy to see that strong mixing implies mild mixing,
minimal system is mildly mixing then it has no
and mild mixing implies weak mixing. The follow-
nontrivial uniformly rigid factor. It is an open
ing theorem shows that mildly mixing is related to
interesting question if the converse is true.
IP*-sets.
For a given n  ℕ, we say (X, T) is (positively)
n-rigid if (x1, . . ., xn) is a (positively) recurrent
Theorem 2.14 (Furstenberg 1982) An m.p.s.
point for T      T (n-times) and each
ðX, X , m, T Þ is mildly mixing if and only if for
(x1, . . ., xn)  Xn. A uniform rigid system has
every A, B  X and for each e < 0, {n  ℤ+ :
zero topological entropy. It was shown by Weiss
|m(A \ TnB)  m(A)m(B)| < e} is an IP*-set.
(1995) that positively 2-rigidity implies zero
We say a measurable function f  L2(X) is
entropy and it is an open question if positive or
rigid for T if for some sequence nk ↗ 1,
negative 2-rigidity implies zero entropy, see
T nk f ! f , k ! 1 in L2(X). An m.p.s. ðX, X , m, T Þ
(Weiss 1995; Hochman 2012).
is rigid if for some sequence nk ↗ 1, and every
f  L2(m), T nk f ! f, k ! 1 in L2(m). An m.d.s. is Ellis Semigroup
mildly mixing if and only if it has no nontrivial The enveloping semigroup was introduced by
rigid factor if and only if there are no nonconstant Ellis, and it is a basic tool in the study of dynam-
rigid functions in L2(X) (Furstenberg 1982). ical systems.
398 Parallels Between Topological Dynamics and Ergodic Theory

An Ellis semigroup E is a semigroup equipped Definition 2 Let (X, T) be a t.d.s. A point x  X is

with a compact Hausdorff topology such that for called a distal point if x itself is the only point in its
every p  E the map Rp : E ! E defined by Rp orbit closure which is proximal to x. A minimal
(q) ¼ qp is continuous. Let (X, T) be a t.d.s. The t.d.s. (X, T) is called point distal if it contains a
enveloping semigroup of (X, T) is the closure of distal point.
{T n : n  ℤ} in XX which is an Ellis semigroup, By Auslander-Ellis theorem, a distal point is
denoted by E(X, T). Note that in E(X, T),p 7! T p minimal. If a minimal t.d.s. (X, T) is point distal,
is continuous, but usually for p  E(X, T) the map then the set of distal point is a dense Gd subset of
Lp : E(X, T) ! E(X, T), q 7! p q is not continu- X (Ellis 1973). If in addition, all points are distal,
ous. An element n in a semigroup E is an idempo- then (X, T) is distal. If a minimal system is not
tent if n2 ¼ n. equicontinuous but it is an almost one-to-one
extension of an equicontinuous minimal system
Theorem 2.16 (Ellis-Namakura). (Ellis 1969) If (this kind of system is called almost auto-
E is an Ellis semigroup, then there exists an idem- morphic), then it is point distal but not distal. We
potent n in E. will introduce the structure theorem of distal and
Let (X, T) be a t.d.s. and (x, y)  X2. It is a point distal minimal t.d.s. in section “Structure
proximal pair if Theorems and Multiple Ergodic Averages.”
Using Theorem 2.17 Furstenberg introduced
lim inf r ðT n x, T n yÞ ¼ 0; the notion of central sets. The notion of a central
n!1
set plays an important role in topological dynam-
and it is a distal pair if it is not proximal. ics, and it has very rich combinatorial properties
(Furstenberg 1981). A subset S of ℤ+ is a central
Definition 1 A t.d.s (X, T) is distal if for all set if there is a t.d.s. (X, T), x  X, a minimal point
x 6¼ y  X, (x, y) is a distal pair, that is, lim y and a neighborhood U of y such that (x, y) is
infn!1r (T nx, T ny) > 0. proximal and S  N(x, U). Some important facts:
It is easy to see every equicontinuous system a central set is piecewise syndetic, and contains an
IP-set. Note that we say a subset is piecewise
is distal. Let T : 2 ! 2, (x, y) 7! (x + a, x + y)
syndetic if it is the intersection a syndetic subset
for an irrational a. Then this system is minimal
with a thick subset.
distal but not equicontinuous (Furstenberg
A subset A of ℤ+ is called a dynamical syndetic
1963). Using Ellis semigroup, Ellis showed that
set, if there exist a minimal system (Y, S), y  Y
a t.d.s (X, T) is distal if and only if E(X, T) is a
and an open neighborhood Vy of y such that
group; and (X, T) is equicontinuous if and only if
A  NS(y, Vy). Denote the set of all dynamical
E(X, T) is a group of homeomorphisms of X (Ellis
syndetic sets by F ds. Let
1969).
For a t.d.s. (X, T) and x  X, the proximal cell
F dps ¼ fA \ B : A  F t , B  F ds g
of x is the set of points y such that (x, y) is
proximal. The following fact shows that there
where F t is the family of thick sets. It was shown
always is some minimal point in proximal cell
in Huang et al. (2019e) that
(Auslander 1988; Ellis 1969).
Theorem 2.18 F dps is equal to the family of
Theorem 2.17 [Auslander-Ellis] Let (X, T) be a
t.d.s. Then for any x  X there is a minimal pointy central sets.
in the orbit closure of x such that (x, y) is proximal. Thus, a central set can be viewed as a dynam-
Akin and Kolyada showed that if (X, T) is a ical piecewise syndetic set. The following result
weakly mixing t.d.s, then for each x  X, the was shown by Furstenberg (1981).
proximal cell of x is a dense Gd subset of
X (Akin and Kolyada 2003). For some related Theorem 2.19 Let (X, T) be a t.d.s. and x  X. The
study, see also (Li et al. 2015b). following statements are equivalent:
Parallels Between Topological Dynamics and Ergodic Theory 399

1. x is distal. f : ðX, X , m, T Þ ! ðY, Y , n, SÞ be a homomorphism,

2. (x, y) is recurrent for any recurrent pointy of a then the so-called graph joining gr(m, f) is
system (Y, S). defined by gr(m, f)(A  B) ¼ m(A \ f1B). One
3. (x, y) is minimal for each minimal pointy of a may find more examples of joinings in Glasner
system (Y, S). (2003, Chap. 6).
4. N (x, U) is an IP* set for each neighborhood Furstenberg showed (Furstenberg 1967) that if
U of x. two m.p.s. are disjoint then one of them has zero
entropy and he was able to characterize the sys-
Inspired by the above fact, one can define and tems as follows:
study various weak product recurrence properties.
For example, one can ask the property of x which Theorem 2.20 An m.p.s. is measurable K-system
satisfies that (x, y) is recurrence for any transitive if and only if it is disjoint from every zero entropy
point y from an M-system (i.e., transitive and with m.p.s. An ergodic system is weakly mixing if and
a dense set of minimal points). It was shown that only if it is disjoint from every ergodic discrete
x is then minimal (Dong et al. 2012) and distal spectrum system.
(Oprocha and Zhang 2013). Moreover, one can One may find definitions of K-system and
ask the property of x which satisfies that (x, y) is u.p.e. in the next section. From this theorem one
recurrence for any minimal point y. It is easy to see sees that two disjoint systems should have differ-
that such an x is not necessarily minimal, and even ent properties. Nowadays the notion of joining is a
when x is minimal it is not necessarily distal, see very useful tool in the study of ergodic theory.
(Glasner and Weiss 2015; Auslander and It is easy to see that two disjoint systems do
Furstenberg 1994; Haddad and Ott 2008; Dong not have common nontrivial factors. It is a ques-
et al. 2012; Huang and Ye 2005). tion for a while if the converse is true. The first
In the measure theoretical setup, Ellis and counterexample was provided by Rudolph in
Nerurkar introduced Enveloping semigroup for Rudolph (1979). The negative answer to
an m.p.s. and showed that it plays an important Furstenbergs question and the consequent
role in ergodic theory (Ellis and Nerurkar 1988, works on joinings and disjointness show that in
1989). We refer to Ellis and Nerurkar (1988, order to study the relationship between two
1989) and Glasner (2003) for details. dynamical systems it is necessary to know all
the possible joinings of the two systems and to
understand the nature of these joinings. We refer
Disjointness and Weak Disjointness to Glasner’s book (Glasner 2003) for the
In number theory one talks about the coprime of discussion.
two natural numbers. In his seminal paper Now we turn to the topological case.
(Furstenberg 1967) Furstenberg introduced the
notion of disjointness both in topological dynam- Definition 4 If (X, T) and (Y, S) are two t.d.s., we
ics and ergodic theory in the same spirit. First we say J  X  Y is a joining of X and Y if J is a non-
discuss disjointness in ergodic theory. empty closed T  S - invariant set and is projected
onto X and Y. If the only joining is equal to X  Y,
Definition 3 Assume ðX, X , m, T Þ and ðY, Y , n, SÞ we then say that (X, T) and (Y, S) are disjoint or
be two m.p.s. We say an invariant measure l on (X, T) ⊥ (Y, S) or X ⊥ Y.
the product system of X and Y is a joining if it The examples of two minimal t.d.s. which have
projects to m and n, respectively. If the only joining no common nontrivial factor and are not disjoint
is equal to m  n, we then say that the two were found by Glasner and Weiss (1983) and
m.p.s. are disjoint. Lindenstrauss (1995).
Similarly we can define joinings on a collec- Compared with Theorem 2.20, we have that a
tion of m.p.s. fðXi , X i , mi , T i Þgi  I . There are cer- minimal u.p.e. system is disjoint from minimal
tain natural joinings. For example, let zero entropy systems (Blanchard 1993) and a
400 Parallels Between Topological Dynamics and Ergodic Theory

topologically weakly mixing system is disjoint In the measure-theoretical setup, it is easy to

from minimal equicontinuous systems. check that an m.p.s. is weakly mixing if and only
Let M be the collection of all minimal systems. if it is weakly disjoint from every ergodic m.p.s. if
Furstenberg (1967) showed that if two systems are and only if it is weakly disjoint from every weakly
disjoint then one of them is minimal (the other one mixing m.p.s.
must have dense recurrent points (Huang and Ye In the topological situation, we say a t.d.s. is
2005)). Thus the most natural question is scattering if it is weakly disjoint from every min-
Question E in Furstenberg (1967): Character- imal t.d.s. Originally the notion of scattering was
ize M ⊥ . defined by using the complexity of an open cover,
Through years of efforts of many researchers the equivalence to the above definition was give in
this question was answered finally in Huang et al. Blanchard et al. (2000). Using weak disjointness
(2019e). Before we state the result we first one can define many other scattering properties,
describe some sufficient and necessary condi- see (Huang and Ye 2002b, 2004).
tions. We remark that Furstenberg showed For more details about disjointness, we refer to
(Furstenberg 1967) that a weakly mixing Glasner (2003), Rudolph (1990), and de la
t.d.s. with dense periodic points is in M ⊥ . Rue (2012).
Recall that a t.d.s. is an M-system if it is tran-
sitive and the set of minimal points is dense. It is Chaos and Complexity: Topological Versus
easy to show that a t.d.s. (X, T) is an M-system if Measurable
and only if for any transitive point x and its neigh- Chaos and complexity theory in both of topolog-
borhood U, N(x, U) is piecewise syndetic. First we ical dynamics and ergodic play an important role
state the following result. in the study. Below we only pick some of them to
present the similarity in both theories.
Theorem 2.21 Let (X, T) be a transitive t.d.s. If First we consider the notion of sensitivity. We
ðX, T Þ⊥M , then (X, T) is a weakly mixing say a t.d.s. (X, T) is sensitive if there is d > 0 such
M-system (Huang and Ye 2005). Each weakly that for each x  X and e > 0 there are y with
mixing system with a dense set of distal points is r(x, y) < e and n  ℕ with r(T nx, T ny) > d. We
in M ⊥ (Oprocha 2010; Dong et al. 2012). have
Now we present an answer to the question in
the transitive case, for the general case see (Huang Theorem 2.23 (Akin et al. 1993; Auslander and
et al. 2019e). We remark that the sufficient condi- Yorke 1980) A minimal t.d.s. is either sensitive or
tion was proved in Oprocha (2019). equicontinuous.
Let (X, T) be a t.d.s., B X be the Borel s - algebra
Theorem 2.22 Let (X, T) be a transitive and m be an invariant measure for T. Then T is said
t.d.s. Then ðX, T Þ⊥M if and only if (X, T) is to be sensitive for m, if there is d > 0 such that for
weakly mixing and there is a countable dense any A  B X with m(A) > 0 we can find x, y  A
subset D of X such that for each non-empty open and n 0 with r(T nx, T ny) > d. For a similar
subset U of X and each central set S ¼ A \ B, we notion, see (Cadre and Jacob 2005; James et al.
can find x ¼ x(B)  D \ U with N T ðx,U Þ \ S 6¼ 0,


2008). We have:
where A is thick and B is dynamical syndetic.
Now we turn to the notion of weak Theorem 2.24 (Huang et al. 2011a) Let (X, T) be
disjointness. We say two systems are weakly dis- a t.d.s. and p be an ergodic measure on (X, T).
joint if the product system is transitive in the Then the ergodic m.p.s. ðX, B X , m, T Þ is either
topological settings or ergodic in the measure sensitive for m or has discrete spectrum.
theoretical setup. Weak disjointness is very useful Sensitivity is only one aspect of chaotic behav-
to define new dynamical properties. We have iors. Nowadays, there are many definitions of
defined mild mixing in the previous sections in chaos, for example, Li-Yorke’s chaos (Li and
such a way. Yorke 1975), Devaney’s chaos (Devaney 1989),
Parallels Between Topological Dynamics and Ergodic Theory 401

positive entropy, distributional chaos (Schweizer 1

n1
and Smital 1994). For the papers dealing with the rn ðx, yÞ ¼ r T i x, T i y
n i¼0
relationship among the definitions, see (Blanchard
et al. 2002a; Huang et al. 2014; Downarowicz
We may define the complexity with respect to
2014; Huang and Ye 2002a; Li and Ye 2016).
the mean metric. The bounded complexity is
Now we consider complexity. There are many
related to mean equicontinuity and m-mean
ways to define complexity. Below we only choose
equicontinuity. For related papers, see (Huang
some of them which relates some new distances
et al. 2019a; Li et al. 2015a; García-Ramos
induced from the original one. Let (X, T) be a
2017; Qiu and Zhao 2020; Yu 2019). We remark
t.d.s. For n  ℕ and x, y  X, define the Bowen’s
that the complexity for an invariant measure can
metric
be used to the study of Sarnak’s conjecture
(Huang et al. 2019f) (see section “Sarnak
rn ðx, yÞ ¼ max r T i x, T i y : i ¼ 0, 1, . . . , n  1 :
Conjecture”).
We say that W (e, n) - spans B if for any x  B
there is y  W such that rn(x, y) < e, where e > 0 Meet Together
and n  ℕ. A subset of X is said to be a (T, e, n)- Let (X, T) be a t.d.s. and B X be the Borel s - algebra.
span if it (e, n)- spans X. Let span(e, n) denote the Denote by M ðXÞ the set of all Borel probability
smallest cardinalities of all (T, e, n) - spans. We measures on M ðX, T Þ the set of T-invariant mea-
say that X has bounded complexity with respect to sures, and M e ðX, T Þ the set of ergodic measures.
{rn} if for every e > 0, {span(n, e)}n is a bounded Note that the ergodic measures are characterized as
sequence. the extreme points of the Choquet simplex
M ðX, T Þ . Then both M ðXÞ and M ðX, T Þ are
Theorem 2.25 Let (X, T) be a t.d.s. Then X has convex, compact metric spaces endowed with the
bounded complexity with respect to {rn} if and weak*-topology. A useful fact is that for any
only if it is equicontinuous. x  X, any weak*-limit point of 1n n1 i¼0 dT i x is an
Let (X, T) be a t.d.s. and m be an invariant invariant measure for T. A Borel probability mea-
measure on (X, T). For n  ℕ and e > 0, let sure m  M ðX, T Þ induces an m.p.s. ðX, B X , m, T Þ.
spanm (n, e) be A t.d.s. (X, T) is uniquely ergodic if there is only
one invariant measure for T, that is, M ðX, T Þ con-
sists of a single element. A uniquely ergodic min-
min #ðFÞ : F  X and m [ Brn ðx, eÞ >1e : imal t.d.s. (X, T) is also called strictly ergodic.
xF
Let ðX, X , m, T Þ be an ergodic m.p.s. We say
Recall that this is the same notion defined in that X, T is a topological model (or just a model)
Katok (1980) by Katok. We say that m has for ðX, X , m, T Þ if X, T is a t.d.s. and there exists
bounded complexity with respect to {rn} if for an invariant probability measure m on the Borel
every e > 0, {spanm(n, e)}n is a bounded s-algebra B such that the systems ðX, X , m, T Þ
X
sequence. We say T is m-equicontinuous if for and X, B , m, T are measured theoretically
X
each e > 0 there is a compact subset K with m isomorphic. It is well known that each m.p.s. has
(K) > 1  e and T |K is equicontinuous. We have: a topological model (Furstenberg 1981). Weiss
(1989) showed the following nice result: there
Theorem 2.26 (Huang et al. 2019a) Let (X, T) be exists a minimal t.d.s. (X, T) with the property
a t.d.s. and m be an invariant measure on (X, T). that for every aperiodic ergodic m.p.s. ðY, Y , n, SÞ
Then m has bounded complexity with respect to there exists m  M ðX, T Þ such that ðY, Y , n, SÞ and
{rn} if and only if T is m-equicontinuous. ðX, B X , m, T Þ are isomorphic. At the same time,
Let (X, T) be a t.d.s. For n  ℕ and x, y,  X, there is no universal topological model in the class
we can define the so-called mean metric of entropy zero systems (Serafin 2013).
402 Parallels Between Topological Dynamics and Ergodic Theory

The pioneering work on topological model was tuples in the product space and it can be used to
done by Jewett (1969/1970) and Krieger (1970/ investigate the global properties of the systems.
1971). That is For example, by using the properties of entropy
pairs, one obtains the existence of the maximal
Theorem 2.27 (Jewett-Krieger). Let ðX, X , T, mÞ zero entropy factor of any t.d.s. (topological
be an ergodic m.p.s. Then there is a minimal Pinsker factor). For related survey, see (Glasner
uniquely ergodic t.d.s. (Y, S) with invariant mea- and Ye 2009; Oprocha and Zhang 2014).
sure n such that ðX, X , m, T Þ is isomorphic to
ðY, B Y , n, SÞ.
Entropy: Topological Versus Measurable
We note that one can add some additional prop-
In 1958 Kolmogorov firstly introduced the
erties to the topological model. For example, in
entropy of an m.p.s. by generating partition
Lehrer (1987) Lehrer showed that the strictly ergodic
(Kolmogorov 1958, 1959). Sinai subsequently
model can be required to be a topological (strongly)
found a natural way to make this notion better-
mixing system in addition. An ergodic system has a
behaved by observing that generators maximize
doubly minimal (i.e., the orbit of any point
entropy relative to a partition among all parti-
(x, y)  X2 is dense provided they are not in the
tions with finite entropy, and formulated the def-
same orbit) model if and only if it has zero entropy
inition that has become the standard one (Sinai
(Weiss 1995); and an ergodic system has a strictly
1959). Later the topological entropy was intro-
ergodic, UPE (uniform positive entropy) model if
duced by Adler et al. who formulated it in terms
and only if it has positive entropy (Glasner and
of open covers (Adler et al. 1965). Equivalent
Weiss 1994). Note that not any dynamical properties
definition based on separated and spanning set
can be added in the uniquely ergodic models. For
was independently given by Bowen (1971) and
example, Lindernstrauss showed that every ergodic
Dinaburg (1971).
measurable distal system ðX, X , m, T Þ has a minimal
Now we define entropy. Let (X,T) be a t.d.s. A
topologically distal model (Lindenstrauss 1999).
cover of X is a family of subsets of X, whose union
This topological model need not, in general, be
is X. A partition of X is a cover of X whose
uniquely ergodic. Weiss (1985) generalized the the-
elements are pairwise disjoint. Given two covers
orem of Jewett-Krieger to the relative case which
U, V of X, U is said to be finer than V (denoted by
will be discussed below. For more information we
U
V or V ≼U ) if each element of U is
refer to Glasner and Weiss (2005).
contained in some element of V ; set
U _ V ¼ fU \ V : U  U, V  V g and T i U ¼
T i U : U  U for i  ℤ+. Denote by N ðU Þ
Entropy Theory
the minimal cardinality among all cardinalities
of subcovers of U.
Entropy is one of the measurements of the com-
plexity of dynamical systems. It is an important
Definition 5 Let (X,T) be a t.d.s. and U be a finite
conjugacy or isomorphic invariant. The concept
open cover of X. The topological entropy of U is
of entropy was invented by Clausius in 1854, and
defined by
Shannon introduced it into information theory in
1948. In 1958 Kolmogorov carried it over to
1 i
ergodic theory. One can find a brief history of htop ðT, U Þ ¼ lim log N _n1
i¼0 T U :
n!þ1 n
entropy in the recent book (Downarowicz 2011).
There are lots of excellent books introducing the The topological entropy of (X,T) is htop ðT Þ ¼
classical theory of entropy such as (Downarowicz supU htop ðT, U Þ , where supremum is taken over
2011; Glasner 2003; Rudolph 1990; Walters
all finite open covers of X.
1982). In this entry we mainly focus on the i 1
Note that log N _n1 i¼0 T U is a sub-
so-called local entropy theory, starting from the n¼1

pioneer work of Blanchard in the early 1990s. In additive sequence and hence htop ðT, U Þ is well
such a theory one studies the local properties of defined.
Parallels Between Topological Dynamics and Ergodic Theory 403

Let ðX, X , m, T Þ be an m.p.s. and P X be the set earlier regularity notions for random processes:
of finite measurable partitions of X. Suppose present becomes asymptotically independent of
x  P X and an s-algebra A  X . The entropy of x all sufficiently long past.
(with respect to A), written Hm(x) (resp. Hm ðxjA Þ),
is defined by the formula Definition 7 An m.p.s. ðX, X , m, T Þ is called a
Kolmogorov system, or a K-system, if there is a
H m ð xÞ ¼  mðAÞ log mðAÞ sub s-algebra K  X with the following
Ax properties:

and H m ðxjA Þ ¼ SA  x X  m ð1A jA Þ log m ð1A jA Þ 1. K  TK ,

dm, where m ð1A jA Þ is the expectation of 1A with 2. _1
n¼0 T K ¼ X (mod m),
n

respect to A. 3. \n¼0 T n K ¼ fX, 0g (mod m).

1


Definition 6 Let ðX, X , m, T Þ be an m.p.s. and A K-system completely differs from zero
x  P X . The entropy of x is defined by entropy systems, see Theorem 2.20. It can be
viewed as the most complicated systems in the
1 i language of entropy. A fundamental result is that
hm ðT, xÞ ¼ lim sup H m _n1
i¼0 T x :
n!þ1 n an m.p.s. is K-system if and only if it has
completely positive entropy (each nontrivial factor
And the entropy of ðX, X , T, mÞ is has positive entropy) if and only if every measur-
able partition by two nontrivial elements has pos-
hm ðT Þ ¼ sup hm ðT, xÞ: itive entropy if and only if every measurable
a  PX
partition by finite nontrivial elements has positive
entropy (Pinsker 1960; Rohlin 1967).
It is well known that htop(T) and hm(T) are
Now we consider t.d.s. Using the first two
conjugacy and isomorphism invariants respec-
conditions above Blanchard (1992) introduced
tively, and they have been the most successful
the notion of c.p.e. and u.p.e. in t.d.s. as an ana-
invariant so far in t.d.s. and m.p.s. In 1958 Kol-
logue of the K-system in ergodic theory.
mogorov asked if entropy is a complete isomor-
phic invariant on the collection of Bernoulli shifts, Definition 8 A t.d.s. (X,T) has completely posi-
see (Glasner 2003). This was answered affirma- tive entropy (c.p.e. for short) if each nontrivial
tively by Ornstein in 1970 (Ornstein 1970). factor has positive entropy and uniform positive
The basic relationship between topological entropy (u.p.e. for short) if every cover by two
entropy and measure-theoretic entropy is given nontrivial open sets has positive entropy.
by the variational principle (Goodwyn 1969; Blanchard (1992) showed that u.p.e. implies
Goodman 1971; Misiurewicz 1976). weak mixing and c.p.e. implies the existence of
an invariant measure with full support. The topic
Theorem 3.1 (The variational principle). Let (X, on the relative notion of c.p.e and u.p.e. can be
T) be a t.d.s. Then found in Glasner and Weiss (1995b) and Huang
et al. (2007).
htop ðT Þ ¼ sup hm ðT Þ : m  M ðX, T Þ
¼ sup hm ðT Þ : m  M e ðX, T Þ : Definition 9 (Huang and Ye 2006) Let (X,T) be a
t.d.s. (X,T) has uniform positive entropy of order n
Measurable and Topological K-Systems (u.p.e. of order n, for short), if any cover of X by
In this subsection we discuss an important class of n non-dense open sets has positive topological
m.p.s. in ergodic theory: K-systems, and its topo- entropy. We say (X,T) has u.p.e. of all orders or
logical analogy. Introduced by Kolmogorov it is topological K if it has u.p.e. of order n for
(1958), it is an isomorphism invariant version of every n 2.
404 Parallels Between Topological Dynamics and Ergodic Theory

Clearly, u.p.e. of order 2 is just u.p.e. It is Definition 10 (Glasner and Weiss 1994; Huang
shown that a u.p.e. system is mildly mixing and Ye 2006). Let (X,T) be a t.d.s. An n-tuple
(Huang and Ye 2006); and any minimal topolog- ðxi Þn1  Xn , n 2, is called an entropy n-tuple if
ical K is strongly mixing (Huang et al. 2005). for some 1  i 6¼ j  n, xi 6¼ x j , and for any
Huang and Ye (2006) answered several open admissible open cover U with respect to ðxi Þn1 ,
questions concerning the nature of u.p.e. and htop ðT, U Þ > 0. Entropy 2-tuples are called entropy
c.p.e. Namely, they showed that u.p.e. of order pairs.
n does not imply u.p.e. of order n + 1 for each We denote by En(X,T) the set of entropy
n 2 (answering a question by Host (Glasner n-tuples. Then following the ideas of Blanchard
and Weiss 1995b)); there is a transitive diagonal (1993) we have
system which does not have u.p.e. (of order 2)
(Blanchard 1993, Question 1); there is a 1. If U ¼ fU 1 ,   , U n g is an open cover of X with
u.p.e. (of order 2) system having no ergodic htop ðT, U Þ > 0 , then for all 1  i  n there
measure with full support (Blanchard 1992, exists xi  U ci such that ðxi Þn1 is an entropy
Question 2). n-tuple.
In Glasner and Weiss (1994) Glasner and 2. En(X, T) [ Dn(X) is a closed T(n)-invariant sub-
Weiss showed that if a t.d.s. admits a K-measure set of Xn.
with full support, then it has u.p.e; and there is a 3. Let p : (X, T) ! (Y, S) be a factor map. Then
minimal u.p.e. system that is universal for any p(n) (En(X, T) [ Dn(X)) ¼ En(Y, S) [ Dn(Y).
ergodic m.p.s. with positive entropy. Above
results were extended to u.p.e. of all orders in It follows from (1) that (X,T) has positive
Huang and Ye (2006). Particularly, the authors entropy if and only if E2 ðX ,T Þ 6¼ 0 ; (X,T) is


proved that if a t.d.s. admits an invariant u.p.e. of order n if and only if for every point
K-measure with full support, then it has u.p.e. of ðxi Þn1  Xn not on the diagonal Dn(X) is entropy
all orders, and a t.d.s. (X,T) has u.p.e. of all orders n-tuple.
if and only if there is an invariant measure
m  M ðX, T Þ such that for each partition a of Definition 11 Let ðX, X , m, T Þ be an
X by finite non-dense Borel sets one has m.p.s. The Pinsker s-algebra is P m ¼
hm(T, a) > 0. A  X : hm ðT, fA, X∖AgÞ0 : The corresponding
Measurable and Topological Entropy Tuples
factor is called the Pinsker factor of ðX, X , m, T Þ.
Now we consider entropy tuples which began
The Pinsker factor of ðX, X , m, T Þ is the largest
with Blanchard’s work for entropy pairs
zero entropy factor. By Rohlin-Sinai’s theorem,
(Blanchard 1993).
an m.p.s. ðX, X , m, T Þ is K-system if and only if
the Pinsker factor is trivial, that is, P m ¼ fX, 0g


Entropy N-Tuples (mod m) (Rohlin and Sinai 1961). Now we define

Given a t.d.s. (X,T) and an integer n 2, the n-th the topological Pinsker factor.
product system is the t.d.s. (Xn,T(n)), where
T(n) ¼ T  . . .  T (n times). Let B X be the Borel Definition 12 Let (X,T) be a t.d.s. The topologi-
ðnÞ
s-algebra and B X be the Borel s-algebra of cal Pinsker factor is the maximal factor with zero
X . The diagonal of Xn is denoted by
n
topological entropy.
n
Dn ðX Þ ¼ ðxi Þn1  X n : x1 ¼    ¼ xn . Let ðxi Þ1  Xn. Blanchard and Lacroix (1993) showed that for
A finite cover of X, U, is said to be an admissible a t.d.s. (X,T), the smallest closed invariant equiv-
cover with respect to ðxi Þn1 , if all the points xi do alence relation containing E2(X,T) induces the
not belong to the closure of the same element U of topological Pinsker factor. For the relative ver-
U . Analogously we define admissible partitions sion, see (Park and Siemaszko 2001).
with respect to ðxi Þn1 .
Parallels Between Topological Dynamics and Ergodic Theory 405

In Blanchard et al. (1995) the authors intro- Theorem 3.3 (Blanchard et al. 1997) For a given
duced the notion of entropy pairs for a measure t.d.s. (X,T) and a finite open cover U of X there is
and their notion cannot be directly generalized for an invariant measure m  MðX ,T Þ with inf a hm
n-tuples when n > 2. Here we will give a defini- ðT, aÞ htop ðT, U Þ , where the infimum is taken
tion of entropy n-tuples for a measure, which was over all finite Borel partitions of X which are finer
introduced by Huang and Ye in Huang and Ye than U.
(2006) and is the same as the notion of entropy In Huang and Ye (2006) Huang and Ye showed
pairs for a measure when n ¼ 2. that, if m  M ðX, T Þ and hm(T, a) > 0 for each finite
Borel partition a of X which is finer than U, then
Definition 13 Let (X,T) be a t.d.s. and infa hm(T, a) > 0 and htop ðT, U Þ > 0 providing
m  M ðX, T Þ . An n-tuple ðxi Þn1  XðnÞ , n 2, is some kind of converse statement of Theorem 3.3.
called an entropy n-tuple for m if for some for To study the question whether inf a hm ðT, aÞ ¼
some 1  i 6¼ j  n, xi 6¼ x j , and for any admis- htop ðT, U Þ for a given finite open cover U,
sible Borel partition a with respect to ðxi Þn1 , Romagnoli (2003) introduced the following entropy
hm(T, a) > 0. for covers
Let m  M ðX, T Þ and P m be the Pinsker
s-algebra of ðX, B X , m, T Þ. Define the conditional hþ
m ðT, U Þ ¼ inf hm ðT, aÞ and
a
U
independent joining ln(m) on XðnÞ , B nX , T ðnÞ by
1
hm ðT, U Þ ¼ lim inf H m ðaÞ,
n n n!þ1 n a
_n1 T i U
i¼0
ln ð m Þ Ai ¼  1Ai jP m dm,
i¼1 X i¼1
where a is a finite Borel partition of X. In fact they
are the same: hþ m ðT, U Þ ¼ hm ðT, U Þ (Huang
where Ai  B X , i ¼ 1,   , n. The following result
et al. 2006).
shows that the set of entropy n-tuples for an
invariant measure is in the support of ln(m) and
Theorem 3.4 (The local variational principle).
we remark that the case n ¼ 2 was proved in
(Romagnoli 2003; Glasner and Weiss 2005) Let
Glasner (1997) and the general case was proved
(X,T) be a t.d.s. and U be a finite open cover of
in Huang and Ye (2006).
X. Then
Theorem 3.2 Let (X,T) be a t.d.s., m  M ðX, T Þ
max hþ
m ðT, U Þ ¼ max hm ðT, U Þ
and n 2. Then m  M ðX, T Þ m  M ðX, T Þ

¼ htop ðT, U Þ:
Emn ðX, T Þ ¼ suppðln ðmÞÞ∖Dn ðXÞ:
Note that the classical variational principle fol-
By Theorem 3.2 it is clear that hm(T) ¼ 0 if and
lows from the local ones by note that hm ðT Þ ¼
only if Em2 ðX, T Þ ¼ 0 ; Emn ðX, T Þ [ Dn ðXÞ is a
supU hm ðT, U Þ, where supremum is taken over all


closed and T(n)-invariant subset of Xn.

finite open covers of X.
Topological entropy tuples are closely related
Local Variational Principles – The Connection of to measurable ones. For each invariant measure
Two Kinds of Tuples the set of entropy pairs for a measure is contained
To study the relationship between the two kinds of in the set of entropy pairs (Blanchard et al. 1995);
entropy pairs or tuples, one needs a local version the converse is also valid (Blanchard et al. 1997).
of the variational principle. Blanchard, Glasner, A variational relation of entropy n-tuple is as
and Host obtained the following: follows:
406 Parallels Between Topological Dynamics and Ergodic Theory

Theorem 3.5 (Huang and Ye 2006) Let (X,T) be 1. for m-a.e. x  X, Emn x ðX, T Þ Emn ðX, T Þ for
a t.d.s. If m  M ðX, T Þ, then En ðX, T Þ Emn ðX, T Þ each n 2.
for each n 2 and there exists m  M ðX, T Þ such 2. ðxi Þn1  Emn ðX, T Þ, then for every neighborhood
that En ðX, T Þ ¼ Emn ðX, T Þ for each n 2. V of ðxi Þn1 ,
We remark that Blanchard et al. (1997)
constructed a t.d.s. and an entropy pair for that m x  X : V \ Emn x ðX, T Þ 6¼ 0 > 0:


system, which is not a metric entropy pair for any
ergodic measure. Also the property that the prod-
Thus we can choose X0  B X such that
uct of u.p.e, of order n (resp. of all orders) systems
m(X0) ¼ 1 and [ Emn x ðX, T Þ : x  X0
is again u.p.e. of order n (resp. of all orders) was
∖Dn ðX Þ ¼ E mn ðX , T Þ
proved by Glasner (1997) for n ¼ 2 and by Huang
and Ye (2006) for the general case.
Weak Horseshoe
In Huang and Ye (2006) Huang and Ye obtained
The Ergodic Decomposition
the following equivalent characterization of topo-
Let (X,T) be a t.d.s. Let m  M ðX, T Þ and
logical entropy n-tuples. Let (X,T) be a t.d.s. and
m ¼ Xmxdm(x) be its ergodic decomposition. The
n 2. Then (x1,   , xn)  En(X, T) if and only if
ergodic decomposition of m also gives an ergodic
for any neighborhood U1      Un of
decomposition of the m-entropy of a a  P X :
(x1,   , xn), there exists a positive density subset
hm ðT, aÞ ¼ X hmx ðT, aÞdmðxÞ (Denker et al.
S ¼ {s1 < s2 <   } of ℤ+ such that
1976). This property also holds for hm ðT, U Þ for \1 si
UtðiÞ 6¼ 0 for any t  {1, 2,   , n}S
i¼1 T
any finite Borel cover U of X. 
(see also (Kerr and Li 2007, 2009)).
Motivated by this, let J be a subset of ℤ+ we
Theorem 3.6 (Huang and Ye 2006) Let (X,T) be
say that (X,T) has a weak horseshoe with an inter-
a t.d.s., m  M ðX, T Þ and U be a finite Borel cover
polating set J if there exist two disjoint closed
of X. If m ¼ Xmxdm(x) is the ergodic decomposi-
subsets U0,U1 of X such that for any
tion of m then
t  {0, 1}J, \ j  J T j U tð jÞ 6¼ 0 , that is, there

exists xt  X such that Tj(xt)  Ut(j) for any
hm ðT, U Þ ¼ hmx ðT, U ÞdmðxÞ j  J (Huang and Lu 2017) . If the subset J has
X
positive density, then we say that (X,T) has a weak
horseshoe.
Given a finite open cover U of a t.d.s. (X,T), we
know that there exists n  M e ðX, T Þ with
hn ðT, U Þ ¼ htop ðT, U Þ by Theorem 3.6 and the Theorem 3.8 (Huang and Ye 2006) A t.d.s. has
local variational principle. Moreover, it was showed positive topological entropy if and only if it has a
that the entropy map m  M ðX, T Þ 7! hm ðT, U Þ is weak horseshoe.
upper semicontinuous (Huang et al. 2011b). This result has many applications. Note that the
The following result discloses the relation of result was proved by Glasner and Weiss in
entropy tuples for an invariant measure and Glasner and Weiss (1995a) for symbolic dynam-
entropy tuples for ergodic measures in its ergodic ics, and was extended to countable amenable
decomposition, which was proven for n ¼ 2 in group actions by Kerr and Li in Kerr and Li
Blanchard et al. (1997) and in general (Huang and (2007). Recently, in Huang and Lu (2017)
Ye 2006). Huang and Lu studied the complicated dynamics
of infinite dimensional random dynamical sys-
Theorem 3.7 Let (X,T) be a t.d.s. and tems and showed that in this setting positive topo-
m  M ðX, T Þ with m ¼ Xmxdm(x) the ergodic logical entropy also implies the existence of weak
decomposition of m. Then horseshoes.
Parallels Between Topological Dynamics and Ergodic Theory 407

Sequence Entropy: Topological Versus gives an example with hStop ðT Þ ¼ log 2 but
Measurable supm  M ðX,T Þ hSm ðT Þ ¼ 0 (Goodman 1974).
The sequence entropy. of an m.p.s. for a sequence Note that for an m.p.s. ðX, X , m, T Þ, if hm(T) > 0,
of ℤ+ was introduced by Kushnirenko in 1967 then for all S  F inf , hSm ðT Þ > 0 (Saleski 1977).
(Kushnirenko 1967). Later, the topological Similarly for a t.d.s. (X,T), if htop(T) > 0, then for
sequence entropy was investigated by Goodman all S  F inf , hStop ðT Þ > 0 (Huang et al. 2005). If
in 1974 (Goodman 1974). X is a countable compact metric space, it is known
Now we define sequence entropy. Denote by that htop(T) ¼ 0 for any t.d.s. (X,T), and this is not
F inf the set of all increasing sequences of ℤ+. Let the case for the topological sequence entropy
S ¼ f0  t1  t2    g  F inf and U be a finite (Ye and Zhang 2008).
open cover of X. The topological sequence Similar to the entropy tuples, one can define
entropy of U with respect to (X,T) along S is sequence entropy n-tuple and sequence entropy
defined by n-tuple for m. They have lots of similar properties
with entropy tuples. But since there is no varia-
1
hStop ðT, U Þ ¼ lim sup log N _ni¼1 T ti U : tional principle for sequence entropy, we do not
n!þ1 n have local variational principles for sequence
entropy tuples. We will not discuss sequence
The topological sequence entropy of (X,T) entropy tuples in details, and we refer to Huang
along sequence S is hStop ðT Þ ¼ supU hStop ðT, U Þ, et al. (2003, 2004), Huang and Ye (2009), Kerr
where supremum is taken over all finite open and Li (2007), and Maass and Shao (2007), etc. to
covers of X. If S ¼ ℤ+, we recover standard interested readers.
topological entropy. In this case we omit the It is found that to characterize different mixing
superscript ℤ+. properties using sequence entropy related to some
For an m.p.s. ðX, X , T, mÞ, a sequence S  ℤ+ sequences is effective and fruitful. Here we just
and a partition x  P X , the sequence entropy give an example to make the point. For more
hSm ðT, xÞ and hSm ðT Þ can be defined similarly. about this topic, we refer to Coronel et al.
hStop ðT Þ and hSm ðT Þ are conjugacy and isomorphism (2009), Hulse (1982, 1986), Huang et al. (2005),
invariants respectively. Saleski (1977), and Zhang (1992, 1993).
Let ðX, X , m, T Þ be an m.p.s. If hm(T) > 0, then
hSm ðT Þ ¼ K ðSÞhm ðT Þ, where K(S) is a number and Theorem 3.9 (Huang et al. 2005) Let ðX, X , m, T Þ
does not dependent on T (Krug and Newton 1972). be an m.p.s. Then the following statements are
This result implies that sequence entropy is equivalent:
uninteresting as a new invariant in case T has pos-
itive entropy. However, little is known in case 1. ðX, X , m, T Þ is mildly mixing;
hm(T) ¼ 0. 2. For any a a  P X by two nontrivial elements
For a t.d.s. (X,T), Goodman (1974) showed (with and IP-set F, there exists an infinite sequence
a restriction that can be removed (Eberlein 1975), A F such that hAm ðT, aÞ > 0;
see also (Huang and Ye 2009)) that for any S  F inf 3. For any a a  P X by finite nontrivial elements
and IP-set F, there exists an infinite sequence
hStop ðT Þ sup hSm ðT Þ, A F such that hAm ðT, aÞ > 0.
m  M ðX, T Þ
In the above theorem a finite measurable set
with an equality in case htop(T) > 0. If htop(T) ¼ 0, C  X is nontrivial if 0 < m(C) < 1. We have
then the variational principle for topological similar result for weakly mixing systems. In the
sequence entropy needs not to hold. Goodman topological case we have:
408 Parallels Between Topological Dynamics and Ergodic Theory

Theorem 3.10 (Huang et al. 2005) Let (X,T) be a Theorem 3.12 (Huang et al. 2003) If a minimal t.
t.d.s. Then the following statements are d.s. (X,T) is null, then it is an almost one-to-one
equivalent: extension of its maximal equicontinuous factor
(Xeq,Teq). Moreover, it is uniquely ergodic and
1. (X,T) is topologically mildly mixing. has a discrete spectrum with respect to the unique
2. For any cover U of X by two non-dense open measure.
sets and IP-set F, there exists an infinite Recently, this result was improved by Fuhrmann
sequence A F such that hAtop ðT, U Þ > 0. et al. (2018). They showed that if a minimal t.d.s. (X,
3. For any cover U of X by finite non-dense open T) is null, then it is a regular extension of (Xeq,Teq),
sets and IP-set F, there exists an infinite that is, if p : (X, T) ! (Xeq, Teq) is the factor map
sequence A F such that hAtop ðT, U Þ > 0. and m  M ðX, T Þ be the unique measure on X, then
m({x  X:| p1(p(x))| ¼1}) ¼ 1. Note that for a
We also have a similar result for topologically minimal system, whether p : X ! Xeq is regular can
weakly mixing systems. be interpreted by the so-called diam mean
equicontinuity (García-Ramos et al. 2019).
Note that the nullness in non-minimal sys-
Null Systems: Measurable Versus Topological
tems was studied in Qiu and Zhao (n.d.). More-
An m.p.s. ðX, X , m, T Þ is called null, if hSm ðT Þ ¼ 0
over, the so-called tame system was extensively
for any S  F inf : The following Kushnirenko’s studied, see (Fuhrmann et al. 2018; Glasner
theorem gives the characterization of discrete 2018; Huang 2006; Kerr and Li 2007). Note
spectrum via sequence entropy. that a minimal null systems is tame (Kerr and
Li 2007).
Theorem 3.11 (Kushnirenko 1967) An m.p.s. It is an open problem (see (Huang et al. 2003)):
ðX, X , m, T Þ has discrete spectrum if and only if it Does a transitive non-minimal null system exist?
is null.
In fact, for an m.p.s. ðX, X , m, T Þ and a  P X
Maximal Pattern Entropy
max hSm ðT, aÞ ¼ H m ajK m , The notion of maximal pattern entropy. was intro-
S  F inf
duced in Huang and Ye (2009). For a t.d.s. (X,T),
where K m is the Kronecker algebra (Huang et al. n  ℕ and a finite open cover U let
2004). As shown in Huang et al. (2004, 2005), for
B  B and R  F inf , cl({Un1B : n  R}) is a pX,U ðnÞ ¼ max N _ni¼1 T ti U :
ðt1 <t2 <<tn Þ  ℤnþ
compact set of L2(m) if and only if
hSm ðT, fB, X∖BgÞ ¼ 0 for each infinite sequence The maximal pattern entropy of T with respect
S R. In particular, B  K m if and only if to U is defined by
hSm ðT, fB, X∖BgÞ ¼ 0 for any S  F inf .
A t.d.s. (X,T) is null if hStop ðT Þ ¼ 0 for any 1
htop ðT, U Þ ¼ lim log pX,U ðnÞ:
n!þ1 n
S  F inf : Note that it is easy to show that an
equicontinuous system is null. It is natural to
The maximal pattern entropy of (X,T) is
conjecture that if a minimal system is null then it
htop ðT Þ
¼ supU htop ðT, U Þ, where supremum is
is also equicontinuous. Unfortunately this is not
the case, see for example (Goodman 1974). But taken over all finite open covers of X.
we have that for a minimal system Kushnirenko’s Analogously, given an m.p.s. ðX, X , m, T Þ we
statement remains true modulo an almost one-to- can define the maximal pattern entropy hm ðT Þ. We
one extension. Note that p : (X, T) ! (Y, S) is remark that one of equivalent definitions is that
almost one to one if {x  X:| p1(p(x))| ¼1} is a the maximal pattern entropy is the supremum of
dense Gd subset. sequence entropies, that is,
Parallels Between Topological Dynamics and Ergodic Theory 409

Theorem 3.13 (Huang and Ye 2009) For a t.d.s. in Cánovas (2004), Tan et al. (2010), and Tan
(X,T) we have that htop ðT Þ ¼ supS  F inf hStop ðT Þ (2011) that S(X) ¼ {0, log2, 1} when X is a finite
and for an m.p.s. ðX, X , m, T Þ we have hm ðT Þ ¼ graph. Moreover, it was showed that S(X) ¼
supS  Finf hSm ðT Þ. {0, log2, log3, . . .} [ {1} when X is a
zerodimensional space with infinite derived sets
Thus a t.d.s. (X,T) is null if and only if
(Tan et al. 2010). Recently, Snoha et al. (2018)
htop ðT Þ ¼ 0 ; an m.p.s. ðX, X , m, T Þ has discrete
obtained the following:
spectrum if and only if hm ðT Þ ¼ 0. The maximal
pattern entropy has some interesting properties. Theorem 3.15 For every set {0} A {0, log2,
For example, they are conjugacy or isomorphism log3, . . .} [ {1} there exists a one-dimensional
invariant. Moreover, k  ℕ\{0}, htop T k ¼ continuum XA ℝ3 with S(XA) ¼ A.
htop ðT Þ, and hm T k ¼ hm ðT Þ. An interesting result For the related work, see (Cánovas 2004; Tan
about the value of the maximal pattern entropy is et al. 2010; Tan 2011).
as follows:
Other Tuples
Theorem 3.14 We want to mention that the idea to study the local
properties of tuples can be applied in many other
1. For a t.d.s. (X,T), htop ðT Þ  f log k : k  ℕg [ situations. For example, the notion of weakly
f1g (Huang and Ye 2009). mixing tuples can be defined and used in the
2. For an ergodic m.p.s. ðX, X , m, T Þ, study of weakly mixing systems and null systems
hm ðT Þ  f log k : k  ℕg [ f1g (Saleski 1977). (Huang et al. 2003). Tame systems are an impor-
tant class of zero entropy systems. The notion of
In fact, by introducing sequence entropy IT-tuples is a main tool to study the properties
tuples, one can show that for a t.d.s. (X,T), (Huang 2006; Kerr and Li 2007). The notion of
htop ðT Þ is logk for some k  ℕ [ {1} with sensitive tuples is also a useful tool to study sen-
k the maximal length of intrinsic sequence entropy sitivity both in topological dynamics and ergodic
tuples. For an m.p.s. ðX, X , m, T Þ, if hm ðT Þ > theory (Huang et al. 2011a). The notion of F fip -
log ðn  1Þ for n 2, then there is an intrinsic pairs and its application can be found in Dong
m-sequence entropy tuple of length n; conversely, et al. (2013). Moreover, the notion of entropy
if there is an intrinsic m-sequence entropy tuple of tuples can be generalized to entropy points,
length n and m is ergodic, then hm ðT Þ log n. See sequence, and sets (Dou et al. 2006; Ye and
(Huang and Ye 2009) for details. Zhang 2007).
Theorem 3.12 says that if a minimal system (X,T) We believe that such ideas can be applied to
is null, that is, htop ðT Þ ¼ 0 then (X,T) is an almost more situations.
one-to-one extension of its maximal equicontinuous
factor and is uniquely ergodic. There is a natural
question: if a minimal system (X,T) is bounded, that Structure Theorems and Multiple
is, htop ðT Þ < 1, then what will happen? Recently Ergodic Averages
Huang et al. showed that each bounded minimal
system is an almost finite to one extension of its To answer the question if a minimal distal t.d.s. is
maximal equicontinuous factor, and it has only an equicontinuous one, Furstenberg obtained the
finitely many ergodic measures (Huang et al. structure theorem for a minimal distal system.
2019b). Then the structure theorem for the general mini-
For a compact metric space X, let S(X) be the mal t.d.s. was built. Almost at the same time,
set of the values htop ðT Þ for all continuous self- Furstenberg and Zimmer obtained the structure
maps T on X. It is clear that {0} S(X) {0, log2, theorem for an ergodic m.p.s. It should be noted
log3, . . .} [ {1} by Theorem 3.14. It was shown that the theorem has an important role to get an
410 Parallels Between Topological Dynamics and Ergodic Theory

ergodic proof of the well-known Szemerédi’s the- 2. for every ι <  there exists an extension
orem. To solve the problem if the multiple ergodic fι : Wι+1 ! Wι which is either proximal or
averages converge in L2-norm, Host-Kra obtained equicontinuous,
a structure theorem of an ergodic m.p.s. involving 3. for a limit ordinal n <  the system Wn is the
nilsystems. The corresponding structure theorem inverse limit of the systems {Wι}ι<n,
of a minimal t.d.s. involving nilsystems was built
by Host-Kra-Maass and Shao-Ye. Here we will
W  ¼ X:
review those results. Moreover, we will show how
topological methods can be used to prove the
We say that (X,T) is a PI-system if there exists a
pointwise converges of multiple ergodic averages
for distal systems. strictly PI system X and a proximal extension y :
X ! X.
If in the definition of PI-systems, we replace
Structure Theorems in Topological Dynamics proximal extensions by almost one-to-one exten-
The structure theory of minimal systems. origi- sions we get the notion of HPI-systems. If we
nated in Furstenberg’s seminal work (Furstenberg replace the proximal extensions by trivial exten-
1963). It was mainly developed for group actions sions (i.e., we do not allow proximal extensions at
on Hausdorff spaces. Here we assume for the rest all), we have I-systems.
of the paper that T is a homeomorphism, and we In this terminology Furstenberg’s structure the-
will also restrict to metrizable systems. We refer orem for a distal system and the Veech-Ellis struc-
the reader to Auslander (1988), Bronstein (1979), ture theorem for a point distal system can be stated
Ellis (1969), Glasner (1976), Veech (1977), and as follows:
de Vries (1993) for details.
We first recall definitions of extensions. Let Theorem 4.1 (Furstenberg 1963, Furstenberg) A
p : (X, T) ! (Y, S) be an extension of t.d.s. Write minimal t.d.s. is distal if and only if it is an
I-system.
Rp ¼ ðx1 , x2 Þ  X2 : pðx1 Þ ¼ pðx2 Þ :
Theorem 4.2 (Veech 1970; Ellis 1973, Veech-
We say that p is proximal if Rp  P(X, T), Ellis) A minimal t.d.s. is point distal if and only
where P(X,T) is the set of all proximal pairs. An if it is an HPI-system.
extension p is an equicontinuous extension if for Finally we state the structure theorem for the
every e > 0 there is d > 0 such that (x, y)  Rp general minimal systems, which was built in Ellis
and r(x, y) < d imply r(T nx, T ny) < e, for every et al. (1975), McMahon (1976), Veech (1977),
n  ℤ. An equicontinuous extension is also called and Glasner (1976). Roughly speaking, the class
an isometric or almost periodic extension. An of minimal flows is the smallest class of flows
extension p is a (topologically) weakly mixing containing the trivial flow and closed under
extension if (Rp, T  T) as a subsystem of the (a) homomorphisms, (b) inverse limits, and
product system (X  X, T  T) is transitive. (c) three “building blocks” which are
equicontinuous extensions, proximal extensions,
Definition 14 We say that a minimal system (X, and topologically weakly mixing extensions. To
T) is a strictly PI system (PI means proximal- be precise, we have the following structure theo-
isometric) if there is an ordinal  (which is count- rem for minimal systems.
able since X is metrizable) and a family of systems
{(Wι, wι)}ι such that Theorem 4.3 (Structure theorem for minimal
systems). Let (X,T) be a minimal t.d.s.
1. W0 is the trivial system, Then we have the following diagram:
Parallels Between Topological Dynamics and Ergodic Theory 411

X X1 Definition 15 (Parry 1967) Let ðX, X , m, T Þ be an

y
ergodic m.p.s. A sequence A1  A2  . . . of sets in
# p1 X with m(An) > 0 and m(An) ! 0, n ! 1, is called
Y1 a separating sieve if there exists a subset X0  X
with m(X0) ¼ 1 such that for every x; x0  X0 the
where X1 is a proximal extension of X and an condition “for every n  ℕ there exists k  ℤ
RIC topologically weakly mixing extension of the with T kx, T kx0  An” implies x ¼ x0. We say that
strictly PI-system Y1. The extension p1 is an an ergodic m.p.s. ðX, X , m, T Þ is measurable distal
isomorphism (so that X1 ¼ Y1 if and only if if either X is finite or there exists a separating
X is a PI-system). sieve.
In the theorem, that the extension p1 is a Parry (1967) showed that if a minimal t.d.s.
relatively incontractible (RIC) extension means (X, T) is distal, then for any m  M ðX, T Þ,
that it is open and for every n 1 the minimal ðX, B X , m, T Þ is measurable distal. Also any mea-
points are dense in the relation surable distal m.p.s. has zero entropy, and hence
by the variational principle any distal t.d.s. has
Rnp ¼ ðx1 , . . . , xn Þ  Xn : p x j , 81  i  j  n : zero topological entropy.
Let p : ðX, X , m, T Þ ! ðX, Y , n, T Þ be a factor
Note that R1p ¼ X and R2p ¼ Rp . For applica- map between two m.p.s. and m ¼ Y mydn(y) the
tions of the structure theorem for minimal sys- disintegration of m relative to n. A function
tems, we refer (Akin et al. 2010; Auslander f  L2 ðX, X , mÞ is almost periodic over Y if for
1988; Bronstein 1979; Ellis 1969; Glasner 1994, every e > 0 there exist g1 , . . . , gl  L2 ðX, X , mÞ
1996, 2003; Shao and Ye 2012); etc. such that for all n  ℤ

Structure Theorems in Ergodic Theory min T n f  g j <e

1jl L 2 ð my Þ
Inspired by Furstenberg’s structure theorem for
distal systems, Parry in 1967 suggested an intrin-
for n almost every y  Y. Let K(X|Y, T) be the
sic definition of measurable distality (Parry 1967).
closed subspace of L2(X) spanned by the almost
In Zimmer (1976a, b), Zimmer developed the
periodic functions over Y . When Y is trivial, K(X|
theory of distal systems (for a general locally
Y, T) is H c ¼ L2 X, K m , m , the closed subspace
compact acting group) and lead to a structure
spanned by eigenfunctions of T.
theorem for the general ergodic system. Indepen-
We say that X is an isometric extensions of Y if
dently, and at about the same time, Furstenberg
K(X|Y, T) ¼ L2(X) and it is a (relatively) weak
proved the same theorem, which was used as the
mixing extension of Y if K(X|Y, T) ¼ L2(Y).
main tool for his proof of Szemerédi theorem on
An ergodic system X is an isometric extension
arithmetical progressions (Furstenberg 1977,
of (Y, n, S) if and only if X is isomorphic to a skew
1981). Roughly speaking, Furstenberg-Zimmer
product X0 ¼ Y  M, where M ¼ G/H, G is a
structure theorem says that the class of ergodic
compact metric space, H a closed subgroup and
systems is the smallest class of flows containing
m0 ¼ n  mM with mM is the Haar measure of M,
the trivial flow and closed under
that is the unique probability measure on
(a) homomorphisms, (b) inverse limits, and
M invariant under the action of G on M by left
(c) two “building blocks” which are isometric
translations. Moreover, the action of T0 on X0 is
extensions and weakly mixing extensions.
given by
First we give the definition of measurable
distality, and then we state the Furstenberg-
T 0 ðy, gH Þ ¼ ðSy, aðyÞgH Þ,
Zimmer structure theorem.
412 Parallels Between Topological Dynamics and Ergodic Theory

where a : Y ! G is a measurable map, called the A subset C of X is called scrambled if any two
cocycle defining the extension. We denote X0 by distinct points x, y  C form a scrambled pair.
YaG/H, and T0 by Ta. A t.d.s. (X, T) is said to be Li-Yorke chaotic if
there is an uncountable scrambled set in X. Using
Theorem 4.4 (Zimmer). (Zimmer 1976b) Let Furstenberg-Zimmer structure theorem, Blanchard
ðX, X , m, T Þ be an ergodic m.p.s. Then et al. (2002a) showed that positive entropy implies
ðX, X , m, T Þ is measurable distal if and only if Li-Yorke chaos. In fact, they showed that if for a
there exists a countable ordinal  and a directed t.d.s. (X, T) there is some ergodic measure m such
family of factors ðXy , X y , my , T Þ, y   such that that ðX, B X , m, T Þ is not measurable distal, then (X,
T) is Li-Yorke chaotic. Note that Kerr and Li gave
1. X0 ¼ {pt} is the trivial system and X ¼ X. an alternative proof of this fact using the combina-
2. For y <  the extension py : Xy+1 ! Xy is torial argument (Kerr and Li 2007). Using the
isometric and nontrivial (i.e., not an structure of minimal systems, Akin et al. showed
isomorphism). that if a minimal system (X, T) is not PI, then it is
3. For a limit ordinal l  , Xl ¼ Li-Yorke chaotic (Akin et al. 2010). It is known that
if each pair not in the diagonal is (positively)
lim Xy ði:e:X l ¼ _X y Þ.
y<l Li-Yorke, then it has zero entropy (Blanchard
et al. 2002b). It is an open question if this is still
Here is the Furstenberg-Zimmer structure the- true if we replace positively Li-Yorke by positively
orem for the action ℤ. For the general locally or negatively Li-Yorke.
compact acting group actions, we refer to We refer the readers to Blanchard and Huang
Furstenberg (1981), Glasner (2003), Zimmer (2008), Downarowicz (2014), Huang (2008),
(1976a, b). Huang and Jin (2016), Huang et al. (2014, 2015,
2017a), Huang and Lu (2017), Li and Qiao
Theorem 4.5 (Furstenberg-Zimmer’s structure (2018), Wang et al. (2019), Zhang (2006), Zhou
theorem). Each ergodic m.p.s. ðX, X , m, T Þ is a et al. (2017), and Liu (2019) for more study on
weakly mixing extension of an ergodic distal sys- chaotic phenomenon appearing in positive
tem X. entropy systems by using the variational principle
and the structure theorems.
p0 p1 pn1 pn pnþ1 p
X0 X1  Xn Xnþ1  X X:
distalfactor
Multiple Ergodic Averages
Ramsey theory is best defined by example and the
Structure theorems are very useful in the study classic example of a Ramsey type theorem is the
of combinatorial number theory, we refer to result of van der Waerden (1927) in the 1920s: if
Bergelson (2006), Bergelson and Leibman (1996), the integers are partitioned into finitely many sub-
Furstenberg (1977, 1981, 1991), Frantzikinakis sets, at least one of the subsets contains arbitrarily
(2017), and Kra (2006), etc. for more details. long arithmetic progressions. Erdös and Turán
Here we mention an example of the applications (1936) conjectured that a weaker assumption suf-
of the structure theorems in the theory of chaos. fices: if E is a set of integers whose upper density
The notion of Li-Yorke chaos was introduced in Li dðEÞ is positive, then E contains arbitrarily long
and Yorke (1975). Let (X, T) be a t.d.s. A pair arithmetic progressions. This conjecture was ver-
(x, y)  X  X is called a Li-Yorke pair if ified by Szemerédi in 1975 (Szemerédi 1975), and
was reproved by Furstenberg using ergodic theo-
lim inf rðT n x, T n yÞ ¼ 0 and lim suprðT n x, T n yÞ retic methods in 1976 (Furstenberg 1977). Green
n!þ1 n!þ1 and Tao showed that the primes contain arbitrarily
> 0: long arithmetic progressions (Green and Tao
Parallels Between Topological Dynamics and Ergodic Theory 413

2008). Now there are lots of excellent surveys 1

N1
p ðnÞ p ðnÞ p ðnÞ
concerning this topic, see for instance lim f 1 T 11 x f 2 T 22 x . . . f d T 2d x
N!1 N
n¼0
(Bergelson 2006; Frantzikinakis and McCutcheon
2012; Furstenberg 1988, 1991, 2010; Kra 2006; ð3Þ
Ward 2012), etc. The recent book by Host and Kra
exists in the L2(m) norm? Almost everywhere?
(2018) is an excellent source for these topics.
It was shown in Bergelson and Leibman (2002)
To prove Szemerédi theorem, Furstenberg
that this multiple ergodic average may not con-
(1977) showed that if ðX, X , m, T Þ is a m.p.s.,
verge in general when the group generated by
f  L1 ðX, X , mÞ, f 0 and not identically 0, then
T1, . . ., Td is only assumed to be solvable.
for any k,
For an m.p.s. ðX, X , m, T Þ, there were numerous
N1 partial results, including the works of Conze and
1 Lesigne (1984, 1987, 1988) for the average of an
lim inf
N!1 N n¼0 arithmetic progression with three terms, the work
of Furstenberg and Weiss (1996) for a polynomial
 f ðxÞf ðT n xÞ . . . f T ðk1Þn x dmðxÞ average with two terms, as well as other partial
convergence results (Host and Kra 2001; Host and
> 0: ð1Þ
Kra 2002). Mean convergence for arithmetic pro-
gressions was proven by Host and Kra (2005) (for
As AN ð f , xÞ ¼ N1 SN1
n¼0 f ðT xÞ is called the
n
another proof, see (Ziegler 2007)), and their proof
ergodic average, the averages above are called motivated the new structural description of an
multiple ergodic averages (or “nonconventional m.p.s. We will discuss their proof later.
averages” (Furstenberg 1988)). More generally, Walsh (2012) proved the convergence for the
we consider the averages multiple ergodic averages where the transforma-
tions are taken from a nilpotent group of trans-
1 formations and the exponents are arbitrary integer
N valued polynomials.
N1
p ðnÞ p ðnÞ p ðnÞ
 f 1 T 11 x f 2 T 22 x . . . f d T dd x
n¼0
Theorem 4.7 (Walsh). (Walsh 2012) Let G be a
nilpotent group of measure preserving transfor-
ð2Þ
mations of a probability space ðX, X , mÞ. Then, for
every T1, . . ., Td  G, the averages
where T1, T2, . . ., Td are invertible measure pre-
serving transformations of a probability space
N l
ðX, X , mÞ, and p1, . . ., pd are integer valued poly- 1 p ðnÞ p ðnÞ
T 11,j . . . T dd,j fj
nomials. The need of a polynomial character is not N n¼1 j¼1
artificial, it is meaningful for combinatorial and
diophantine applications (Bergelson 2006; always converge in L2 ðX, X , mÞ for every
Furstenberg 2010). f 1 , . . . , f d  L1 ðX, X , mÞ and every set of integer
The basic question related to the multiple ergo- valued polynomials pi, j.
dic averages is as follows: Walsh’s proof is not ergodic in nature, but
rather combinatorial, following previous work of
Question 4.6 Let T1, T2, . . ., Td be measure pre- Tao (2008). This new method sufficed to prove
serving transformations of a probability space convergence results, but it lacked the structural
ðX, X , mÞ generating a nilpotent group description of the averages, the associated
(of transformations). Is it true that for any integer description of the limit, and the combinatorial
valued polynomials pi(n) and f i  L1 ðX, X , mÞ , theorems that can be derived from this structure.
i ¼ 1, 2, . . ., d, As mentioned in Host and Kra (2018), an
414 Parallels Between Topological Dynamics and Ergodic Theory

interesting, but very difficult, problem is to find a characteristic factor Z, meaning that Z is charac-
way to give structural results that control these teristic and that every proper factor of Z is not
more general averages. Such results are not yet characteristic. To describe this minimal character-
available in the general setting of nilpotent groups istic factor, we need to introduce some notions.
of transformations, nor even just for commuting
transformations. We refer to Furstenberg (2010), Definition 16 Let G be a group. For A, B  G, we
Kra (2006), and Host and Kra (2018) for more write [A, B] for the subgroup spanned by {[a,
details on this topic. b] ¼ aba1b1 : a  A, b  B}. The commutator
subgroups Gj, j 1, are defined inductively by
setting G1 ¼ G and Gj+1 ¼ [Gj, G]. Let d 1 be an
Characteristic Factors in Ergodic Theory integer. We say that G is d-step nilpotent if Gd + 1
In the study of multiple ergodic averages, the idea is the trivial subgroup.
of characteristic factors plays a very important Let G be a d-step nilpotent Lie group and G be
role. This idea was suggested by Furstenberg in a discrete cocompact subgroup of G. The compact
Furstenberg (1977), and the notion of “character- manifold X ¼ G/G is called a d-step nilmanifold.
istic factors” was first introduced in a paper by The group G acts on X by left translations and we
Furstenberg and Weiss (1996). Let ðX, X , m, T Þ be write this action as (g, x) 7! gx. The Haar mea-
an m.p.s. and ðY, Y , m, T Þ be a factor of X. Let sure m of X is the unique probability measure on
{p1, . . ., pd} be a family of integer valued poly- X invariant under this action. Let t  G and T be
nomials, d  ℕ. We say that Y is a characteristic the transformation x 7! tx of X. Then (X, m, T) is
factor of X for the scheme {p1, . . ., pd} if for all called a d-step nilsystem.
f 1 , . . . , f d  L1 ðX, X , mÞ, Conze and Lesigne had shown (Conze and
Lesigne 1984, 1987, 1988) that an inverse limit
N1
1
lim T p1 ðnÞ f 1 T p2 ðnÞ f 2 . . . T pd ðnÞ f d of 2-step nilsystems is the characteristic factor
N!1 N n¼0
for the 3-term multiple ergodic averages. The
N1
1 general case was confirmed by constructing
 T p1 ðnÞ ð f 1 jY ÞT p2 ðnÞ ð f 2 jY Þ . . . T pd ðnÞ ð f d jY Þ ! 0:
N n¼0 L2 such factors in Host and Kra (2005) (see also
(Ziegler 2007)).
Finding a characteristic factor often gives a Now we outline Host-Kra’s methods and results,
reduction of the problem of evaluating limit and we need some notations first. Let X be a set, and
behavior of multiple averages to special systems. let d 1 be an integer. We view element in {0, 1}d
In the rest of the entry, we mainly consider the as a sequence e ¼ e1. . . ed of 0’s and 1’s, and let
scheme {n, 2n, . . ., dn}. That is, we consider the
je j ¼ e1 + e2 + . . . ed. We denote X2 by X[d]. A point
d

following multiple ergodic averages:

x  X can be written as x ¼ (xe : e  {0, 1}d).
[d]

N1
A point x  X[d] can be decomposed as x ¼ (x0, x00)
1 withx0,x00  X[d  1],wherex0 ¼ (xe0 :e  {0, 1}d  1)
f 1 ðT n xÞ f 2 T 2n x . . . f d T dn x : ð4Þ
N n¼0 and x00(xe1 : e  {0, 1}d1), which induces a natural
identification of X[d] and X[d  1]  X[d1].
To study (4), it suffices to consider ergodic As examples, points in X[2] are like (x00, x10,
systems and that we will restrict to this case x01, x11).
below. Furstenberg (1977) proved that the distal Let ðX1 , X 1 , m1 , T 1 Þ , ðX2 , X 2 , m2 , T 2 Þ be two
factor is a characteristic factor for (4), and this m.p.s. and let ðY, Y , n, SÞ be a common factor
observation was a main step in the proof of the with pi : Xi ! Y for i ¼ 1, 2 the factor maps. Let
ergodic version of Szemerédi theorem, see (1). mi ¼ mi,y dn(y) represent the disintegration of mi
But this fact is not enough to show the mean with respect to Y. Let m1 Y m2 denote the measure
convergence of (4). It always exists a minimal defined by
Parallels Between Topological Dynamics and Ergodic Theory 415

1=2k
m1 Y m2 ðAÞ ¼ m1,y  m2,y dnðyÞ,
Y j k f j kk ¼ Cjej f ðxe Þdm½k ðxÞ :
X ½k k
e  f0, 1g
for all A  X 1  X 2: The system
ðX1  X2 , X 1  X 2 , m1 Y m2 , T 1  T 2 Þ is called That jk| kk is a seminorm can be proved as in
the relative product of X1 and X2 with respect to Host and Kra (2005), and it is called the Host-Kra
Y and is denoted X1 Y X2 . m1 Y m2 is also called seminorm. The equation below follows immedi-
relatively independent joining of X1 and X2 over Y. ately from the definition of the measures and the
For an m.p.s. ðX, X , m, T Þ we write I ðT Þ for the Ergodic Theorem, which can be considered as an
s-algebra A  X : T 1 A ¼ A of invariant sets. alternate definition of the seminorms. For every
Let ðX, X , m, T Þ be an ergodic system and k  ℕ. integer k 0 and every f  L1(m), one has
We define a measure m[k] on X[k] invariant under
T[k] ¼ T  T  . . .  T (2k times), by m½1 ¼ 1=2kþ1
N1
mI ðT Þ m ¼ m  m; and for k 1, 1 2k
j k f jkkþ1 ¼ lim j f  Tnf j k
:
N!1 N
n¼0
m½kþ1 ¼ m½k  m½k :
I ðT ½k Þ Then by the Host-Kra seminorm we can define
factors ðZd1 , Z d1 , md1 , T Þ.
For an integer k 1, let (Ok, Pk) be the system
corresponding to the s-algebra I ½k and let Definition 17 Let ðX, X , m, T Þ be an ergodic
m.p.s. For d  ℕ, there exists a T-invariant
s-algebra Z d1 of X such that for f  L1(m),
m½k ¼ m½ok dPk ðoÞ
Ok
j k f jkd ¼ 0 if and only if ð f jZ d1 Þ ¼ 0:
[k]
denote the ergodic decomposition of m under
T[k]. Then by definition Let ðZd1 , Z d1 , md1 , T Þ be the factor of
X associated to the sub s-algebra Z d1 .
If X ¼ Zd–1, then X is called a system of order
m½kþ1 ¼ m½ok  m½ok dPk ðoÞ: d – 1.
Ok
Using van der Corput Lemma, Host and Kra
If (X, m, T) is weakly mixing, then by induc- showed that the original averages along arithmetic
tion I T ½k is trivial and m[k] is the 2k Cartesian progressions is controlled by the seminorms.
power m2 of m for k 1.
k

Theorem 4.8 (Host and Kra 2005) Let

The measure m[k] is invariant under T[k] and ðX, X , m, T Þ be an ergodic m.p.s. and d  ℕ. For
each of the 2k natural projections of m[k] on X is f1, . . ., fd  L1(X, m) with kf1k1, . . ., kfdk1  1,
equal to m. Letting C : ℂ ! ℂ denote the one has that
conjugacy map z 7! z, we have that for a bounded
function f on X, the integral N1
1
lim sup T n f 1 T 2n f 2 . . . T dn f d
N!1 N n¼0 L2
Cjej f ðxe Þdm½k ðxÞ
X ½k
e  f0, 1gk  min jj f j j d
1jd

is real and nonnegative. Therefore, we can define This theorem states that the factor Zd  1 is
a seminorm jk| kk on L1(m) by characteristic for the average (4). The bulk of the
416 Parallels Between Topological Dynamics and Ergodic Theory

work, and also the most technical portion, is given by Glasner in Glasner (1994). Let (X, T) be a
devoted to the description of these factors. The t.d.s. and d  ℕ. Let sd ¼ T  T2  . . .  Td.
structure theorem states the system (Y, T) is said to be an topological characteristic
ðZd1 , Z d1 , md1 , T Þ is a (measure theoretic) factor of order d if there exists a dense Gd set O of
inverse limit of d – 1-step nilsystems. X such that for each x  O the orbit closure L ¼
We isolate the first coordinate, writing X½d ¼ O xd , sd is p  . . .  p (d times) saturated, where
X2 1 and then writing a point x  X [d]as x ¼
d
xd ¼ (x, . . ., x) (d times) and p : X ! Y is the
(x0, x*), where x ¼ ðxe : e6¼0Þ  X½d and corresponding factor map. That is,
0 ¼ 00 . . . 0  {0, 1}d. (x1, x2, . . ., xd)  L if and only if
x01 , x02 , . . . , x0d  L whenever for all 1  i  d,
Theorem 4.9 (Host-Kra). (Host and Kra 2005) pðxi Þ ¼ p x0i . In Glasner (1994), it was shown
Let ðX, X , m, T Þ be an ergodic system and d  ℕ. that if (X, T) is a distal minimal system, then its
Then the following properties are equivalent: largest d-step distal factor (in the Furstenberg’s
tower of a minimal distal system) is a topological
1. X is a system of order d – 1, that is, characteristic factor of order d; if (X, T) is a weakly
ðX, X , m, T Þ ¼ ðZd1 , Z d1 , md1 , T Þ. mixing system (X, T), then the trivial system is its
2. The system ðX, X , m, T Þ is a (measure theoretic) topological characteristic factor. It is a deep open
inverse limit of d – 1-step nilsystems. problem whether for a minimal distal system one
3. jk| kd is a norm on L1(m), equivalently, jkf | can replace the largest d-step distal factor in
kd ¼ 0 implies that f ¼ 0. Glasner’s theorem by the maximal d-step
4. There exists a measurable map J : X½d ! X pro-nilfactor.
such that x0 ¼ J xe : 0 6¼ e  f0, 1gd for m[d] The second way was given by Host, Kra, and
almost every x ¼ (xe : e  {0, 1}d)  X[d]. Maass in Host et al. (2010). They obtained a
topological structure theory involving nilsystems
Z0 is the trivial factor, and Z1 is the Kronecker for all minimal distal systems, which can be
factor ði:e:Z 1 ¼ K mÞ and more generally, Zk is a viewed as an analog of the purely ergodic struc-
compact abelian group extension of Zk–1. Further- ture theory of (Host and Kra 2005) and the refine-
more, the sequence of factors is increasing ment of the Furstenberg’s structure theorem for
minimal distal systems. In Host et al. (2010), a
fpt g ¼ Z 0 Z1  Zn Z nþ1  certain generalization of the regionally proximal
X: relation, namely RP[d] (the regionally proximal
relation of order d), was introduced and used to
and if T is weakly mixing, then Zk is the trivial produce the maximal pro-nilfactors, which can be
factor for all k. seen as the characteristic factor of the minimal
Convergence of the linear multiple ergodic system (X, T). In the following we will give
average then follows easily from the general prop- some details of this approach.
erties of nilmanifolds proved by Leibman (2005) If n ¼ (n1, . . ., nd)  ℤd and e  {0, 1}d, we
(which was proved by Lesigne for connected define n  e ¼ di¼1 ni ei . Let (X, T) be a t.d.s. and
groups earlier (Lesigne 1989)). See also let d 1 be an integer. We define Q[d](X) to be the
(Bergelson et al. 2005; Ziegler 2005). For more closure in X[d] of elements of the form
details on this topic, we refer to the recent book
(Host and Kra 2018). T ne x ¼ T n1 e1 þ...þnd ed x : e ¼ e1 e2 . . . ed  f0, 1gd ,

Characteristic Factors in Topological where n ¼ (n1, . . ., nd)  ℤd and x  X. When

Dynamics there is no ambiguity, we write Q[d] instead of
There are several ways to study the counterpart of Q[d](X). An element of Q[d](X) is called a
characteristic factors in a t.d.s. The first way was (dynamical) parallelepiped of dimension d. As
Parallels Between Topological Dynamics and Ergodic Theory 417

examples, Q[2] is the closure in X[2] ¼ X4 of the set 2. If x, y  Q[d] have 2d – 1 coordinates in
{(x, Tmx, T nx, T n+mx) : x  X, m, n  ℤ}. Q[d] common, then x ¼ y.
may be viewed as the topological correspondence 3. If x, y  X are such that (x, y, . . ., y)  Q[d],
of m[d]. then x ¼ y.
½d
Face transformations T j : X½d ! X½d , j ¼
1, . . . , d are defined as follows: for every x ¼ A minimal system satisfying one of the equiva-
lent properties above is called a (topological)
ðxe Þe  f0,1gd  X½d
system of order (d – 1) or a (topological)
(d – 1)-step pro-nilsystem.
½d Txe if e j ¼ 1; Note that any system of order d is isomorphic
Tj x ¼
e xe if e j ¼ 0: in the measure theoretic sense to a topological
system of order d (Host et al. 2010). Let (X, T)
The face group of dimension d is the group be a system of order d, then the maximal measur-
F ½d ðXÞ of transformations of X[d] spanned by the able and topological factors of order j coincide,
face transformations. The cube group or parallel- where j  d (Dong et al. 2013).
epiped group of dimension d is the group G ½d ðXÞ
spanned by the diagonal transformation and the Definition 18 Let (X, T) be a t.d.s. and let d  ℕ.
face transformations. We often write F ½d and G ½d The points x, y  X are said to be regionally
instead of F ½d ðXÞ and G ½d ðXÞ, respectively. It is proximal of order d if for any d > 0, there exist
easy to verify that Q[d] is the closure in X[d] of x0, y0  X and a vector n ¼ (n1, . . ., nd)  ℤd such
Sx½d : S  F ½d , x  X . If x is a transitive point of that r(x, x0) < d, r(y, y0) < d, and
X, then Q[d] is the orbit closure of x[d] under the
group G ½d : For x  X, we write rðT ne x0 , T ne y0 Þ < d for any e  f0, 1gd ∖f0g:
x[d] ¼ (x, x, . . ., x)  X[d].
The set of regionally proximal pairs of order
Theorem 4.10 If (X, T) is minimal and d  ℕ, d is denoted by RP[d] (or by RP[d](X, T) in case of
then Q½d , G ½d is minimal (Host et al. 2010). And ambiguity), and is called the regionally proximal
for all x  X, O x½d , F ½d , F ½d is minimal relation of order d.
(Shao and Ye 2012). The above definition was introduced in Host
Theorem 4.10 can be viewed as a topological and Maass (2007) and Host et al. (2010). The
analogue of the following ergodic theorem. authors showed (Host et al. 2010) that if a system
is minimal and distal, then RP[d] is an equivalence
Theorem 4.11 (Host and Kra 2005) If relation, and a very deep result stating that
ðX, X , m, T Þ is an ergodic m.p.s. and d  ℕ, then (X/RP[d], T) is the maximal d-step pro-nilfactor
X½d , m½d , G ½d is ergodic. And (Od, Pd) is ergodic of the system. Using the theory of Ellis semi-
under the action of the group F ½d . group, Shao and Ye (2012) showed that all these
Compared with Theorem 4.9, the following results in fact hold for arbitrarily minimal systems
structure theorem characterizes the inverse limits of abelian group actions. In a recent paper by
of nilsystems using dynamical parallelepipeds. Glasner et al. (2018), the same question is consid-
ered for a general group action, and similar results
Theorem 4.12 (Host-Kra-Maass). (Host et al. are proved.
2010) Assume that (X, T) is a minimal t.d.s. and
let d 2 be an integer. The following properties Theorem 4.13 (Shao and Ye 2012; Host et al.
are equivalent: 2010) Let (X, T) be a minimal t.d.s. and d  ℕ.
Then
1. X is an (topologically) inverse limit of (d – 1)-
step minimal nilsystems. 1. RP[d](X) is an equivalence relation.
418 Parallels Between Topological Dynamics and Ergodic Theory

2. (Xd ¼ X/RP[d], T) is the maximal d-step pro- group htd, sdi. We remark that if (X, T) is mini-
nilfactor of (X, T). mal, then all Nd(X, x) coincide, which will be
denoted by Nd(X). It was shown by Glasner
Note that RP[1] is the classical regionally prox- (1994) that if (X, T) is minimal, then
imal relation and X1 ¼ Xeq is the maximal (Nd(X), htd, sdi) is minimal. Hence if
equicontinuous factor of the system. Furthermore, (Nd(X), htd, sdi) is uniquely ergodic, then it is
the sequence of factors is increasing strictly ergodic. In fact, we can give a nice
model for any ergodic m.p.s.:
fpt g ¼ X0 X1  Xn Xnþ1 
X: Theorem 4.14 (Huang et al. 2019d) Let
ðX, X , m, T Þ be an ergodic m.p.s. and d  ℕ.
and if (X, T) is topologically weakly mixing, then Then it has a strictly ergodic model X, T such
Xn is the trivial factor for all n. that N d X , htd , sd i is strictly ergodic.
There are other ways to study the counterpart Note that for each uniquely ergodic t.d.s. it has
of characteristic factors in a t.d.s. For example, in very nice convergence property: for a t.d.s. (X, G)
Cai and Shao (2019), the topological characteris- with G ¼ ℤd, then (X, G) is uniquely ergodic if
tic factors along cubes of minimal systems are and only if for every continuous function
studied; and in Glasner et al. (2019), the authors f  C(X) the sequence of functions
1
considered the regionally proximal relation of Nd g  ½0,N1 d f ðgxÞ converges uniformly to a con-
order d along arithmetic progressions. stant function. Thus as an application of Theorem
Further development of the theory of cubes in 4.14 we can show the following:
an abstract setting, calling these structures
nilspaces, were given by Host and Kra (2008), Theorem 4.15 (Huang et al. 2019d) Let
Camarena and Szegedy (Cai and Shao 2019; ðX, X , m, T Þ be an ergodic m.p.s. and d  ℕ.
Szegedy 2012), Gutman et al. (2019a, b, 2020), Then for f1, . . ., fd  L1(m) the averages
Candela (2017a, b), etc.
1
f 1 ðT n xÞ f 2 ðT nþm xÞ . . . f d T nþðd1Þm x
N2
ðn, mÞ  j0, N1j2
Topological Methods in the Study of the
Multiple Ergodic Averages
converge to a constant m - a. e.
Though results on the convergence of multiple ergo-
To prove Theorem 4.14, we found that not
dic averages in L2 norm are very rich now, there are a
every strictly ergodic model is the one we need,
few ones about almost pointwise convergence. In
and Jewett-Krieger Theorem is not enough for our
1989 Bourgain (1989) showed that
pðnÞ purpose. Fortunately, we find that Weiss’s Theo-
lim N!1 N1 N1 n¼0 f T x exists a.e. for all
0 rem is a right tool.
p(n)  ℤ[n] and f  L ðX, X , mÞ with p > 1. Note
p

that this result may fail when p ¼ 1 (Buczolich and We say that p : X ! Y is a topological model for
Mauldin 2010). And in 1990 Bourgain (1990) pro- a factor map p : ðX, X , m, T Þ ! ðY, Y , n, SÞ if p is a
ved that lim N!1 N1 N1 topological factor map and there exist measure the-
n¼0 f 1 ðT x Þ f 2 ðT a 2 n x Þ
a1 n
1
exists a.e. for all f1, f2 in L ðX, X , mÞ. Recently the oretical isomorphisms f and c such that the
authors of the current paper find that one may use diagram
topological methods to study almost pointwise con- f
vergence of multiple ergodic averages. X!X
Let (X, T) be a t.d.s. and d  ℕ. Set p ## p
td ¼ T  . . .  T (d times) and c
Y!Y
sd ¼ T  T2  . . .  Td. Let htd, sdi be the
group generated by td, sd. For any x  X, let is commutative, that is, pf ¼ cp. Weiss (1985)
N d ðX, xÞ ¼ O ððx, . . . , xÞ, htd , sd iÞ, the orbit clo- generalized the theorem of Jewett-Krieger to the
sure of (x, . . ., x) (d times) under the action of the relative case. Namely, he proved that
Parallels Between Topological Dynamics and Ergodic Theory 419

Theorem 4.16 (Weiss). (Weiss 1985) If p : (2018). A joining l of ðXi , X i , mi , T i Þ, 1  i  d, is

ðX, X , m, T Þ ! ðY, Y , n, SÞ is a factor map with pairwise independent if its projection on Xi  Xj is
ðX, X , m, T Þ ergodic and Y, B , n, S is a uniquely equal to mi  mj for all i ¼ j  f1, 2, . . . , d g, and it


Y is independent if it is the product measure.
ergodic model for ðY, Y , n, T Þ , then there is a A system ðX, X , m, T Þ is said to be pairwise inde-
uniquely ergodic model X, B , m, T for pendently determined (PID) if all pairwise inde-
X
ðX, X , m, T Þ and a factor map p : X ! Ywhich is a pendent d-self joinings (d 3) are independent.
model for p : X ! Y. Each weakly mixing m.p.s. with spectral type
For d 3 let pd2 : X ! Zd  2 be the factor singular w.r.t Lebesgue measure is PID (Host
map from X to its d – 2-step nilfactor Zd–2. By the 1991); and so is each finite-rank mixing transfor-
results of Host-Kra-Maass in Host et al. (2010), mation (Ryzhikov 1993).
Zd–2 may be regarded as a topological system in Using topological model, one can show the
the natural way. Using Weiss’s Theorem there is a following result.
uniquely ergodic model X, X , m, T for
ðX, X , m, T Þ and a factor map pd2 : X ! Zd2 Theorem 4.18 (Gutman et al. 2018) Let
which is a model for pd2 : X ! Zd2. We then ðX, X , m, T Þ be a weakly mixing and pairwise
show that X, T is what we need, that is, independently determined (PID) m.p.s. Then for
all d  ℕ and all f 1 , . . . , f d  L1 ðX, X , mÞ, (4)
N d X , htd , sd i is uniquely ergodic. exists a.s. Moreover, for a generic m.p.t. (4) exists
Using some results developed when proving a.s.
Theorem 4.15, one has Now we discuss the convergence along the
cube groups. Host and Kra showed the following
Theorem 4.17 (Huang et al. 2019d) Let convergence along the cube groups.
ðX, X , m, T Þ be an ergodic distal system, and
d  ℕ. Then for all f1, . . ., fd  L1(m)
Theorem 4.19 (Host and Kra 2005) Let
ðX, X , m, T Þ be an m.p.s., and d  ℕ. Then for
N1
1 functions fe  L1(m), e  {0, 1}d, e 6¼ (0, . . ., 0),
f 1 ðT n xÞ . . . f d T dn x
N n¼0 the averages

d
converge mm a.e. 1
: f e ðT ne xÞ
Note that by Furstenberg-Zimmer’s structure the- N i  Mi
i¼1 n  ½M1 , N 1 Þ...½Md , N d Þð0, ..., 0Þ6¼e  f0, 1gd
orem and Theorem 4.17 the open question (if (4)
converges pointwisely) is reduced to deal with the ð5Þ
weakly mixing extensions. Even for weakly mixing
systems, the question on the pointwise convergence converge in L2(X) as N1 – M1, N2  M2, . . .,
of (4) still remains open. A partial answer to this Nd  Md tend to +1.
question was obtained by Assani (1998), who Let ðX, X , m, T Þ be an ergodic m.p.s. and
showed that if ðX, X , m, T Þ is a weakly mixing d  ℕ. Using Weiss’s theorem we can prove
system such that the restriction of X to its Pinsker that it has a strictly ergodic model ðX, TÞ such
algebra has spectral type singular w.r.t. the Lebesgue that ðQ½d ðXÞ, G ½d Þ is strictly ergodic. Then we can
measure, then the limit of (4) exists a.e. Also in reprove that (5) converge m a.e. (Huang et al.
Derrien and Lesigne (1996), it was shown that the 2017b). This result for the averages along [0,
limit of (4) exists a.e. for a K-system. N – 1]d was first built by Assani (2010), and Chu
Now we describe some results for a class of and Franzikinakis (2012) extended the result to a
weakly mixing m.p.s. obtained in Gutman et al. very general case.
420 Parallels Between Topological Dynamics and Ergodic Theory

It should be noted that these methods were satisfy the equation. The case where the polyno-
further developed in Donoso and Sun (2016, mial p is linear was completely solved by Rado
2018a, b). (1933). A notorious old question of Erdös and
Graham (1980) is that whether the equation
x2 + y2 ¼ z2 is partition regular?
Further Directions Hindman-Graham in 1979 asked the following
question: for any finite coloring of ℕ, does there
Pointwise Convergence of the Multiple exist x, y  ℕ such that {x, y, xy, x + y} is
Averages monochromatic? Moreira (2017) solved a weaker
In contrast to the progress on mean convergence, form by showing that: for any finite coloring of ℕ,
the problem of pointwise convergence for multi- there exist x, y  ℕ such that {x, xy, x + y} is
ple averages along arithmetic progressions monochromatic.
remains open. At present for a single transforma-
tion, the most well-known results are due to Sarnak Conjecture
Bourgain (1990) and Huang et al. (2019d). The Möbius function m: ℕ ! {1, 0, 1} is
Donoso and Sun extended the results in Huang defined by m(1) ¼ 1 and
et al. (2019d) to commuting transformations in
Donoso and Sun (2018a).
ð1Þk if n is a product of k
mðnÞ ¼ distinct primes;
Analogue Results
It is always an interesting question to get analogue 0 otherwise:
results in topological dynamics and ergodic the-
ory. We mention two of them: Let (X, T) be a t.d.s. We say a sequence x is
We say (X, T) is totally minimal if (X, T n) is realized in (X, T) if there is an f  C(X) and an
minimal for and n 6¼ 0. A deep open question is x  X such that x(n) ¼ f(T nx) for any n  ℕ.
the so-called “odd recurrent problem” (for the A sequence x is called deterministic if it is realized
ergodic version, see (Frantzikinakis 2004)). Let in a system with zero topological entropy. Here is
(X, T) be totally minimal. For given d  ℕ and the conjecture by Sarnak (2009):
j  ℤ+ with 0  j  d  1, does there exist a The Möbius function m is linearly disjoint from
sequence {ni} with ni (mod d) ¼ j and x  X such any deterministic sequence x. That is,
that
N
1
lim mðnÞxðnÞ ¼ 0:
T ni x ! x, T 2ni x ! x, . . . , T dni x ! x, i ! 1? N!1 N
n¼1

A related question is that if (X, T) is totally Though there are many papers which solved
minimal, is it true that there is x  X such that some special classes of zero entropy systems (see
2
T n x : n  ℤ is dense in X? For a partial solu- (Green and Tao 2010; Ferenczi et al. 2018;
tion and the ergodic versions, see (Huang et al. Kułaga-Przymus and Lemańczyk 2019) and the
2019c; Furstenberg 1981; Bourgain 1989). references therein), it seems that there is a long
way to go to settle the conjecture completely.
Ramsey Type Theorems
An important question in Ramsey theory is to
determine which algebraic equations, or systems References
of equations, are partition regular over the natural
numbers. Partition regularity of the equation p- Adler RL, Konheim AG, McAndrew MH (1965) Topolog-
ical entropy. Trans Am Math Soc 114:309–319
(x1, x2, . . ., xn) ¼ 0(n 2) amounts to saying that, Akin E (1997) Recurrence in topological dynamics,
for any partition of ℕ into finitely many cells, Furstenberg families and Ellis actions. Plenum Press,
some cell contains distinct x1, x2, . . ., xn that New York/London
Parallels Between Topological Dynamics and Ergodic Theory 421

Akin E, Kolyada S (2003) Li-Yorke sensitivity. Non- Blanchard F, Glasner E, Kolyada S, Maass A (2002a) On
linearity 16:1421–1433 Li-Yorke pairs. J Reine Angew Math 547:51–68
Akin E, Auslander J, Berg K (1993) When is a transitive Blanchard F, Host B, Ruette S (2002b) Asymptotic pairs in
map chaotic? In: Convergence in ergodic theory and positive-entropy systems. Ergodic Theory Dyn Syst
probability (Columbus, OH, 1993), 25C40. Ohio State 22:671–686
University Mathematical Research Institute Publica- Bourgain J (1989) Pointwise ergodic theorems for arith-
tions, 5. de Gruyter, Berlin, 1996 metic sets. With an appendix by the author, Harry
Akin E, Glasner E, Huang W, Shao S, Ye X (2010) Suffi- Furstenberg, Yitzhak Katznelson and Donald
cient conditions under which a transitive system is S. Ornstein. Inst Hautes Études Sci Publ Math 69:5–45
chaotic. Ergodic Theory Dyn Syst 30:1277–1310 Bourgain J (1990) Double recurrence and almost sure
Assani I (1998) Multiple recurrence and almost sure con- convergence. J Reine Angew Math 404:140–161
vergence for weakly mixing dynamical systems. Israel Bowen R (1971) Entropy for group endomorphisms and
J Math 103:111–124 homogeneous spaces. Trans Am Math Soc
Assani I (2010) Pointwise convergence of ergodic averages 153:401–414
along cubes. J Anal Math 110:241–269 Bronstein IU (1979) Extensions of minimal transformation
Auslander J (1988) Minimal flows and their extensions. groups. Martinus Nijhoff Publications, The Hague
North-Holland mathematics studies, vol 153. North- Buczolich Z, Mauldin RD (2010) Divergent square aver-
Holland, Amsterdam ages. Ann of Math 171(2):1479–1530
Auslander J, Furstenberg H (1994) Product recurrence and Cadre B, Jacob P (2005) On pairwise sensitivity. J Math
distal points. Trans Am Math Soc 343:221–232 Anal Appl 309:375–382
Auslander J, Yorke J (1980) Interval maps, factors of maps, Cai F, Shao S (2019) Topological characteristic factors
and chaos. Thoku Math J 32(2):177–188 along cubes of minimal systems. Discrete Contin Dyn
Bergelson V (2006) Combinatorial and diophantine appli- Syst 39(9):5301–5317
cations of ergodic theory. Appendix A by A. Leibman Candela P (2017a) Notes on nilspaces: algebraic aspects.
and Appendix B by Anthony Quas and Máté Wierdl. Discrete Anal 15
In: Handbook of dynamical systems, vol 1B. Elsevier Candela P (2017b) Notes on compact nilspaces. Discrete
B. V., Amsterdam, pp 745–869 Anal 16
Bergelson V, Leibman A (1996) Polynomial extensions of Cánovas JS (2004) Topological sequence entropy of inter-
van der Waerden’s and Szemeredi’s theorems. J AMS val maps. Nonlinearity 17(1):49–56
9:725–753 Chu Q, Frantzikinakis N (2012) Pointwise convergence for
Bergelson V, Leibman A (2002) A nilpotent Roth theorem. cubic and polynomial ergodic averages of non-
Invent Math 147:429–470 commuting transformations. Ergodic Theory Dyn
Bergelson V, Host B, Kra B (2005) Multiple recurrence and Syst 32:877–897
nilsequences.With an appendix by Imre Ruzsa. Invent Conze J, Lesigne E (1984) Théorèmes ergodiques pour des
Math 160:261–303 mesures diagonales. (French) [Ergodic theorems for
Billingsley P (1965) Ergodic theory and information. diagonal measures]. Bull Soc Math France 112:143–175
Wiley, New York/London/Sydney Conze J, Lesigne E (1987) Sur un théorème ergodique pour
Birkhoff GD (1927) Dynamical systems, vol 9. American desmesures diagonales. (French) [On an ergodic theo-
Mathematical Society Colloquium Publications, rem for diagonal measures]. Probabilits, 1–31, Publ.
New York Inst. Rech. Math. Rennes, 1987-1, Université
Blanchard F (1992) Fully positive topological entropy and Rennes I, Rennes, 1988
topological mixing. Symbolic dynamics and its appli- Conze J, Lesigne E (1988) Sur un théorème ergodique pour
cations. AMS Contemp Math 135:95–105 des mesures diagonales. (French) [On an ergodic theo-
Blanchard F (1993) A disjointness theorem involving topo- rem for diagonal measures]. C R Acad Sci Paris Sr
logical entropy. Bull Soc Math France 121:465–478 I Math 306:491–493
Blanchard F, Huang W (2008) Entropy sets, weakly mixing Cornfeld IP, Fomin SV, Sinaĭ YG (1982) Ergodic theory.
sets and entropy capacity. Discrete Contin Dyn Syst Fundamental principles of mathematical sciences,
20(2):275–311 vol 245. Springer, New York, p x,486
Blanchard F, Lacroix Y (1993) Zero-entropy factors of Coronel A, Maass A, Shao S (2009) Sequence entropy and
topological flows. Proc Am Math Soc 119:985–992 rigid salgebras. Stud Math 194:207–230
Blanchard F, Host B, Maass A, Martinez S, Rudolph DJ de la Rue T (2012) Joinings in ergodic theory. Mathematics
(1995) Entropy pairs for a measure. Ergodic Theory of complexity and dynamical systems, vol 1–3.
Dyn Syst 15(4):621–632 Springer, New York, pp 796–809
Blanchard F, Glasner E, Host B (1997) A variation on the de Vries J (1993) Elements of topological dynamics.
variational principle and applications to entropy pairs. Kluwer, Dordrecht
Ergodic Theory Dyn Syst 17:29–43 Denker M, Grillenberger C, Sigmund K (1976) Ergodic
Blanchard F, Host B, Maass A (2000) Topological com- theory on compact spaces. Springer lecture notes in
plexity. Ergodic Theory Dyn Syst 20:641–662 mathematics, vol 527. Springer, Berlin
422 Parallels Between Topological Dynamics and Ergodic Theory

Derrien J, Lesigne E (1996) Un théorème ergodique poly- Erdös P, Graham RL (1980) Old and new problems and
nomial ponctuel pour les endomorphismes exacts et les results in combinatorial number theory, Monographies
K-systèmes. (French) [A pointwise polynomial ergodic de LEnseignement Mathématique [Monographs of
theorem for exact endomorphisms and K-systems]. LEnseignement Mathématique], vol 28. Université de
Ann Inst H Poincar Probab Stat 32:765–778 Genève, LEnseignement Mathématique, Geneva
Devaney RL (1989) An introduction to chaotic dynamical Erdös P, Turán P (1936) On some sequences of integers.
systems. Addison-Wesley Publishing Company J Lond Math Soc 11:261–264
Advanced Book Program, Redwood City Fayad B, Kanigowski A (2016) Multiple mixing for a class
Dinaburg EI (1971) A connection between various entropy of conservative surface flows. Invent Math
characterizations of dynamical systems. (Russian) Izv 203(2):555–614
Akad Nauk SSSR Ser Mat 35:324–366 Ferenczi S, Kulaga-Przymus J, Lemanczyk M (2018)
Dong P, Shao S, Ye X (2012) Product recurrent properties, Sarnak’s conjecture: what’s new. Lecture Notes Math
disjointness and weak disjointness. Israel J Math 2213:163–235
188:463–507 Følner E (1954) Generalization of a theorem of
Dong P, Donoso S, Maass A, Shao S, Ye X (2013) Infinite- Bogoliuboff to topological abelian groups. Math
step nilsystems, independence and complexity. Ergodic Scand 2:5–19
Theory Dyn Syst 33:118–143 Frantzikinakis N (2004) The structure of strongly station-
Donoso S, Shao S (2017) Uniformly rigid models for rigid ary systems. J Anal Math 93:359C388
actions. Stud Math 236:13–31 Frantzikinakis N (2017) Ergodicity of the Liouville system
Donoso S, Sun W (2016) A pointwise cubic average for implies the Chowla conjecture. Discrete analysis paper
two commuting transformations. Israel J Math no. 19, 41 pp
216:657–678 Frantzikinakis N, McCutcheon R (2012) Ergodic theory:
Donoso S, Sun W (2018a) Pointwise multiple averages for recurrence. Mathematics of complexity and dynamical
systems with two commuting transformations. Ergodic systems, vol 1–3. Springer, New York, pp 357–368
Theory Dyn Syst 38(6):2132–2157 Fuhrmann G, Glasner E, Jäger T, Oertel C (2018) Irregular
Donoso S, Sun W (2018b) Pointwise convergence of some model sets and tame dynamics. arXiv:1811.06283
multiple ergodic averages. Adv Math 330:946–996 Furstenberg H (1963) The structure of distal flows. Am
Dou D, Ye X, Zhang G (2006) Entropy sequences and J Math 85:477–515
maximal entropy sets. Nonlinearity 19:53–74 Furstenberg H (1967) Disjointness in ergodic theory, min-
Downarowicz T (2011) Entropy in dynamical systems. imal sets, and a problem in Diophantine approximation.
New mathematical monographs, vol 18. Cambridge Math Syst Theory 1:1–49
University Press, Cambridge Furstenberg H (1977) Ergodic behavior of diagonal mea-
Downarowicz T (2014) Positive topological entropy sures and a theorem of Szemerédi on arithmetic pro-
implies chaos DC2. Proc Am Math Soc gressions. J Anal Math 31:204–256
142(1):137–149 Furstenberg H (1981) Recurrence in ergodic theory and
Downarowicz T, Ye X (2002) When every point is either combinatorial number theory. M. B. Porter lectures.
transitive or periodic. Colloq Math 93:137–150 Princeton University Press, Princeton
Eberlein E (1975) On topological entropy of semigroups of Furstenberg H (1982) IP-systems in ergodic theory. Con-
commuting transformations. In: International confer- ference in modern analysis and probability (New
ence on dynamical systems in mathematical physics, Haven, Conn., 1982). Contemporary Mathematics
Rennes, 1975, pp 17–62 26, American Mathematical Society, Providence,
Ellis R (1969) Lectures on topological dynamics. W. A. 1984, pp 131–148
Benjamin, Inc., New York Furstenberg H (1988) Nonconventional ergodic averages.
Ellis R (1973) The Veech structure theorem. Trans Am The legacy of John von Neumann (Hempstead, NY,
Math Soc 186:203–218 1988). In: Proceedings of Symposia in Pure Mathemat-
Ellis R, Gottschalk W (1960) Homomorphisms of trans- ics, vol 50, American Mathematical Society, Provi-
formation groups. Trans Am Math Soc 94:258–271 dence, 1990, pp 43–56
Ellis R, Keynes H (1971) A characterization of the Furstenberg H (1991) Recurrent ergodic structures and
equicontinuous structure relation. Trans Am Math Soc Ramsey theory. In: Proceedings of the International
161:171–181 Congress of Mathematicians, vol I, II (Kyoto, 1990).
Ellis R, Keynes H (1972) Bohr compactifications and a Mathematical Society of Japan, Tokyo, 1991,
result of Følner. Israel J Math 12:314–330 pp 1057–1069
Ellis R, Nerurkar M (1988) Enveloping semigroup in ergo- Furstenberg H (2010) Ergodic structures and non-
dic theory and a proof of Moore’s ergodicity theorem. conventional ergodic theorems. International Congress
Dynamical systems (College Park, MD, 1986–87). of Mathematicians
Lecture notes in mathematics, 1342. Springer, Berlin, Furstenberg H, Weiss B (1977) The finite multipliers of
pp 172–179 infinite ergodic transformations. The structure of
Ellis R, Nerurkar M (1989) Weakly almost periodic flows. attractors in dynamical systems. In: Proceedings of
Trans Am Math Soc 313:103–119 the conference on North Dakota State University,
Ellis R, Glasner S, Shapiro L (1975) Proximal-isometric Fargo, N.D., 1977). Lecture notes in mathematics,
flows. Adv Math 17:213–260 668. Springer, Berlin, 1978, pp 127–132
Parallels Between Topological Dynamics and Ergodic Theory 423

Furstenberg H, Weiss B (1978) Topological dynamics and Glasner E, Gutman Y, Ye X (2018) Higher order regionally
combinatorial number theory. J Anal Math 34:61–85 proximal equivalence relations for general minimal
Furstenberg H, Weiss B (1996) A mean ergodic theorem group actions. Adv Math 333:1004–1041
2
for N1 N Glasner E, Huang W, Shao S, Ye X (2019) Regionally
n¼1 f ðT xÞg
n
T n x . In: Convergence in ergodic
proximal relation of order d along arithmetic progres-
theory and probability (Columbus, OH, 1993). Ohio sions and nilsystems, in prepare
State University Mathematical Research Institute Pub- Goodman TNT (1971) Relating topological entropy and
lications, 5. de Gruyter, Berlin, pp 193–227 measure entropy. Bull Lond Math Soc 3:176–180
García-Ramos F (2017) Weak forms of topological and Goodman TNT (1974) Topological sequence entropy. Proc
measure theoretical equicontinuity: relationships with Lond Math Soc 29:331–350
discrete spectrum and sequence entropy. Ergodic The- Goodwyn L (1969) Topological entropy bounds measure-
ory Dyn Syst 37:1211–1237 theoretic entropy. Proc Am Math Soc 23:679–688
García-Ramos F, Jäger T, Ye X (2019) Mean Gottschalk W (n.d.) Orbit closure decompositions and
equicontinuity, almost automorphy and regularity, almost periodic properties. Bull Am Math Soc
preprint 50:915–919
Glasner S (1976) Proximal flows. Lecture notes in mathe- Gottschalk W, Hedlund G (1955) Topological dynamics.
matics, vol 517. Springer, Berlin/New York American Mathematical Society Colloquium publica-
Glasner E (1994) Topological ergodic decompositions and tions, vol 36. American Mathematical Society,
applications to products of powers of a minimal trans- Providence
formation. J Anal Math 64:241–262 Green B, Tao T (2008) The primes contain arbitrarily long
Glasner E (1996) Structure theory as a tool in topological arithmetic progressions. Ann Math 167(2):481–547
dynamics. In: Descriptive set theory and dynamical Green B, Tao T (2010) The Möbius function is strongly
systems (Marseille-Luminy, 1996), 173C209. London orthogonal to nilsequences. Ann Math 175(2):541–566
Mathematical Society lecture note series, vol 277. Gutman Y, Huang Y, Shao S, Ye X (2018) Almost sure
Cambridge University Press, Cambridge, 2000 convergence of the multiple ergodic average for certain
Glasner E (1997) A simple characterization of the set of weakly mixing systems. Acta Math Sinica 34:79–90
m-entropy pairs and applications. Israel J Math Gutman Y, Manners F, Varjú P (2019a) The structure
102:13–27 theory of Nilspaces III: inverse limit representations
Glasner E (1998) On minimal actions of Polish groups. and topological dynamics. http://arxiv.org/abs/1605.
Topology Appl 117:259–272 08950
Glasner E (2003) Ergodic theory via Joinings. Mathemat- Gutman Y, Manners F, Varjú P (2019b) The structure
ical surveys and monographs, 101. American Mathe- theory of Nilspaces II: representation as nilmanifolds.
matical Society, Providence Trans Am Math Soc 371:4951–4992
Glasner E (2018) The structure of tame minimal dynamical Gutman Y, Manners F, Varjú P (2020) The structure theory
systems for general groups. Invent Math of Nilspaces I. J Anal Math 140:299–369
211(1):213–244 Haddad K, Ott W (2008) Recurrence in pairs. Ergodic
Glasner S, Maon D (1989) Rigidity in topological dynam- Theory Dyn Syst 28:1135–1143
ics. Ergodic Theory Dyn Syst 9:309–320 Halmos PR (1960) Lectures on Ergodic theory. Chelsea
Glasner S, Weiss B (1983) Minimal transformations with Publishing Co, New York, p vii, 101
no common factor need not be disjoint. Israel J Math Halmos P, von Neumann J (1942) Operator methods in
45:1–8 classical mechanics II. Ann Math 43:332–350
Glasner E, Weiss B (1994) Strictly ergodic, uniform posi- Hindman N (1974) Finite sums from sequences within
tive entropy models. Bull Soc Math France cells of a partition of ℕ. J Combin Theory Ser
122(3):399–412 A 17:1–11
Glasner E, Weiss B (1995a) Quasi-factors of zero-entropy Hochman M (2012) On notions of determinism in topolog-
systems. J Am Math Soc 8(3):665–686 ical dynamics. Ergodic Theory Dyn Syst 32:119–140
Glasner E, Weiss B (1995b) Topological entropy of exten- Host B (1991) Mixing of all orders and pairwise indepen-
sions. In: Ergodic theory and its connections with har- dent joinings of systems with singular spectrum. Israel
monic analysis (Alexandria, 1993). London J Math 76:289–298
Mathematical Society lecture note series, vol 205. Host B, Kra B (2001) Convergence of Conze-Lesigne
Cambridge University Press, Cambridge, pp 299–307 averages. Ergodic Theory Dyn Syst 21:493–509
Glasner E, Weiss B (2005) On the interplay between mea- Host B, Kra B (2002) An odd Furstenberg-Szemerdi theo-
surable and topological dynamics. In: Hasselblatt B, rem and quasi-affine systems. J Anal Math 86:183–220
Katok A (eds) Handbook of dynamical systems, Host B, Kra B (2005) Nonconventional averages and
vol 1B. North-Holland, Amsterdam, pp 597–648 nilmanifolds. Ann Math 161:398–488
Glasner E,Weiss B (2015) On doubly minimal systems and Host B, Kra B (2008) Parallelepipeds, nilpotent groups and
a question regarding product recurrence. Gowers norms. Bull Soc Math France 136(3):405–437
arXiv:1508.02817 Host B, Kra B (2018) Nilpotent structures in Ergodic
Glasner E, Ye X (2009) Local entropy theory. Ergodic theory. Mathematical surveys and monographs, 236.
Theory Dyn Syst 29:321–356 American Mathematical Society, Providence
424 Parallels Between Topological Dynamics and Ergodic Theory

Host B, Maass A (2007) Nilsystèmes d’ordre deux et Huang W, Xu L, Yi YF (2015) Asymptotic pairs, stable sets
parallélépipédes. Bull Soc Math France 135:367–405 and chaos in positive entropy systems. J Funct Anal
Host B, Kra B, Maass A (2010) Nilsequences and a struc- 268(4):824–846
ture theory for topological dynamical systems. Adv Huang W, Shao S, Ye X (2016) Nil Bohr-sets and almost
Math 224:103–129 automorphy of higher order. Memoirs of the American
Host B, Kra B, Maass A (2016) Variations on topological Mathematical Society. American Mathematical Soci-
recurrence. Monatsh Math 179:57–89 ety, Providence
Huang W (2006) Tame systems and scrambled pairs under Huang W, Li J, Ye X, Zhou XY (2017a) Positive topolog-
an abelian group action. Ergodic Theory Dyn Syst ical entropy and D-weakly mixing sets. Adv Math
26(5):1549–1567 306:653–683
Huang W (2008) Stable sets and e-stable sets in positive- Huang W, Shao S, Ye X (2017b) Strictly ergodic models
entropy systems. Commun Math Phys 279(2):535–557 and pointwise ergodic averages for cubes. Comm Math
Huang W, Jin L (2016) Stable sets and mean Li-Yorke Stat 5:93–122
chaos in positive entropy actions of bi-orderable ame- Huang W, Li J, Thouvenot J, Xu L, Ye X (2019a) Bounded
nable groups. Ergodic Theory Dyn Syst complexity, mean equicontinuity and discrete spec-
36(8):2482–2497 trum. Ergodic. Theory Dyn Syst. arXiv:1806.02980
Huang W, Lu K (2017) Entropy, chaos, and weak horse- Huang W, Lian Z, Shao S, Ye X (2019b) Minimal systems
shoe for infinite-dimensional random dynamical sys- with finitely many ergodic measures, preprint
tems. Commun Pure Appl Math 70(10):1987–2036 Huang W, Shao S, Ye X (2019c) Topological correspon-
Huang W, Ye X (2002a) Devaney’s chaos or 2-scattering dence of multiple ergodic averages of nilpotent group
implies Li-Yorke’s chaos. Topology Appl 117:259–272 actions. J Anal Math. arXiv:1604.07113
Huang W, Ye X (2002b) An explicit scattering, non-weakly Huang W, Shao S, Ye X (2019d) Pointwise convergence of
mixing example and weak disjointness. Nonlinearity multiple ergodic averages and strictly ergodic models.
15:1–14 J Anal Math. arXiv:1406.5930
Huang W, Ye X (2004) Topological complexity, return Huang W, Shao S, Ye X (2019e) An answer to Furstenberg’
times and weak disjointness. Ergodic Theory Dyn problem on topological disjointness. Ergodic Theory
Syst 24(3):825–846 Dyn Syst. arXiv:1807.10155
Huang W, Ye X (2005) Dynamical systems disjoint from Huang W, Wang Z, Ye X (2019f) Measure complexity and
any minimal system. Trans Am Math Soc Möbius disjointness. Adv Math 347:827–858
357(2):669–694 Hulse P (1982) Sequence entropy and subsequence gener-
Huang W, Ye X (2006) A local variational relation and ators. J Lond Math Soc 26:441–450
applications. Israel J Math 151:237–280 Hulse P (1986) Sequence entropy relative to an invariant
Huang W, Ye X (2009) Combinatorial lemmas and appli- s-algebra. J Lond Math Soc 33:59–72
cations to dynamics. Adv Math 220(6):1689–1716 James J, Koberda T, Lindsey K, Silva CE, Speh P (2008)
Huang W, Ye X (2012) Generic eigenvalues, generic Measuable sensitivity. Proc Am Math Soc
homomorphisms and weak disjointness. Contemp 136:3549–3559
Math 567:119–143 Jewett RI (1969/1970) The prevalence of uniquely ergodic
Huang W, Li S, Shao S, Ye X (2003) Null systems and systems. J Math Mech 19:717–729
sequence entropy pairs. Ergodic Theory Dyn Syst Kalikow S (1984) Twofold mixing implies threefold
23(5):1505–1523 mixing for rank one transformations. Ergodic Theory
Huang W, Maass A, Ye X (2004) Sequence entropy pairs Dyn Syst 4:237–259
and complexity pairs for a measure. Ann Inst Fourier Kanigowski A, Kułaga-Przymus J, Ulcigrai C (2017) Mul-
(Grenoble) 54(4):1005–1028 tiple mixing and parabolic divergence in smooth area-
Huang W, Shao S, Ye X (2005) Mixing via sequence preserving flows on higher genus surfaces. J Eur Math
entropy. Algebraic and topological dynamics. Contemp Soc. arXiv: 1606.09189
Math 385:101–122 Katok A (1980) Lyapunov exponents, entropy and the
Huang W, Ye X, Zhang G (2006) A local variational periodic orbits for diffeomorphisms. Inst Hautes tudes
principle for conditional entropy. Ergodic Theory Dyn Sci Publ Math 51:137–173
Syst 26(1):219–245 Katok A, Hasselblatt B (1995) Introduction to the modern
Huang W, Ye X, Zhang G (2007) Relative entropy tuples, theory of dynamical systems. In: Encyclopedia of
relative U.P.E. and C.P.E. extensions. Israel J Math mathematics and its applications, vol 54. Cambridge
158:249–283 University Press, Cambridge, p xviii,802
Huang W, Lu P, Ye X (2011a) Measure-theoretical sensi- Katznelson Y (2001) Chromatic numbers of Cayley graphs
tivity and equicontinuity. Israel J Math 183:233–283 on ℤ and recurrence. Combinatorica 21:211–219
Huang W, Ye X, Zhang G (2011b) Local entropy theory for Kerr D, Li H (2007) Independence in topological and C*-
a countable discrete amenable group action. J Funct dynamics. Math Ann 338(4):869–926
Anal 261(4):1028–1082 Kerr D, Li H (2009) Combinatorial independence in mea-
Huang W, Li J, Ye X (2014) Stable sets and mean Li-Yorke surable dynamics. J Funct Anal 256(5):1341–1386
chaos in positive entropy systems. J Funct Anal Kolmogorov AN (1958) A new metric invariant of tran-
266(6):3377–3394 sient dynamical systems and automorphisms in
Parallels Between Topological Dynamics and Ergodic Theory 425

Lebesgue spaces. (Russian) Dokl Akad Nauk SSSR Maass A, Shao S (2007) Structure of bounded topological-
(NS) 119:861–864 sequence-entropy minimal systems. J Lond Math Soc
Kolmogorov AN (1959) Entropy per unit time as a metric 76(3):702–718
invariant of automorphisms. (Russian) Dokl Akad McMahon DC (1976) Weak mixing and a note on a struc-
Nauk SSSR 124:754–755 ture theorem for minimal transformation groups. Ill
Kolyada S, Snoha L (1997) Some aspects of topological J Math 20:186–197
transitivity ł a survey. In: Iteration theory (ECIT 94) McMahon DC (1978) Relativized weak disjointness and
(Opava), Grazer mathematische Berichte, vol 334. relatively invariant measures. Trans Am Math Soc
Karl-Franzens-Universität Graz, Graz, p 3C35 236:225–237
Korner TW (1987) Recurrence without uniform recur- Misiurewicz M (1976) A short proof of the variational
rence. Ergodic Theory Dyn Syst 7:559–566 principle for a ℤnþ action on a compact space.
Kra B (2006) From combinatorics to ergodic theory and back Astérisque 40:227–262
again. International Congress of Mathematicians, vol III. Moothathu TKS (2010) Diagonal points having dense
European Mathematical Society, Zürich, pp 57–76 orbit. Colloq Math 120:127–138
Krieger W (1970/1971) On unique ergodicity. In: Proceed- Moreira J (2017) Monochromatic sums and products in ℕ.
ings of the sixth Berkeley symposium on mathematical Ann Math 185:1069–1090
statistics and probability, vol II: probability theory. Uni- Nemytskiǐ VV (1949) Topological problems in the theory
versity of California Press, Berkeley, 1972, pp 327–346 of dynamical systems, Uspekhi Matematicheskikh
Kronecker L. (n.d.) Naherrungsweise ganzzahlige Nauk, 5. American Mathematical Society translation,
Auflosunglinear Gleichungen, S.-B. Preuss, Akad. no. 103, 1954
Wiss., 1179-93, 1271-99. Werke III (1), pp 47–109 Nemytskiǐ VV, Stepanov VV (1949) Qualitative theory of
Krug E, Newton D (1972) On sequence entropy of automor- differential equations. GITTL Moscow (Russian).
phisms of a Lebesgue space. Z Wahrscheinlicht- Princeton University Press, Princeton, 1960
keitstheorie Verw Gebiete 24:211–214 Nicol M, Petersen K (2012) Ergodic theory: basic exam-
Kułaga-Przymus J, Lemańczyk M (2019) Sarnak’s conjec- ples and constructions. Mathematics of complexity and
ture from the ergodic theory point of view. This volume dynamical systems, vol 1–3. Springer, New York,
Kurka P (2003) Topological and symbolic dynamics. pp 264–287
Cours Spécialisés specialized courses, 11. Société Oprocha P (2010) Weak mixing and product recurrence.
Mathématique de France, Paris Ann Inst Fourier (Grenoble) 60(4):1233–1257
Kushnirenko AG (1967) On metric invariants of entropy Oprocha P (2019) Double minimality, entropy and
type. Russ Math Surv 22:53–61 disjointness with all minimal systems. Discrete Contin
Kwietniak D, Oprocha P (2012) On weak mixing, mini- Dyn Syst 39:263–275
mality and weak disjointness of all iterates. Ergodic Oprocha P, Zhang G (2013) On weak product recurrence and
Theory Dyn Syst 32:1661–1672 synchronization of return times. Adv Math 244:395–412
Lehrer E (1987) Topological mixing and uniquely ergodic Oprocha P, Zhang G (2014) Topological aspects of dynam-
systems. Israel J Math 57:239–255 ics of pairs, tuples and sets. In: Recent progress in
Leibman A (2005) Pointwise convergence of ergodic aver- general topology. III. Atlantis Press, Paris, pp 665–709
ages for polynomial sequences of translations on a Ornstein D (1970) Bernoulli shifts with the same entropy
nilmanifold. Ergodic Theory Dyn Syst 25:201–213 are isomorphic. Adv Math 4:337–352
Lesigne E (1989) Théeorèemes ergodiques pour une trans- Park KK, Siemaszko A (2001) Relative topological Pinsker
lation sur un nilvariété. (French) [Ergodic theorems for factors and entropy pairs. Monatsh Math 134:67–79
a translation on a nilmanifold]. Ergodic Theory Dyn Parry W (1967) Zero entropy of distal and related trans-
Syst 9:115–126 formations. In: Auslander J, Gottschalk W (eds) Topo-
Li J, Qiao YX (2018) Mean Li-Yorke chaos along some logical dynamics. Benjamin, New York
good sequences. Monatsh Math 186(1):153–173 Pinsker MS (1960) Dynamical systems with completely
Li J, Ye X (2016) Recent development of chaos theory in positive or zero entropy. Dokl. Akad. Nauk SSSR
topological dynamics. Acta Math Sinica 32:83–114 133:1025–1026 (Russian); translated as Soviet Math.
Li TY, Yorke J (1975) Period three implies chaos. Am Math Dokl. 1:937–938
Mon 82:985–992 Qiu J, Zhao J (2020) A note on mean equicontinuity. J Dyn
Li J, Tu S, Ye X (2015a) Mean equicontinuity and mean Diff Equat 32:101–116
sensitivity. Ergodic Theory Dyn Syst 35:2587–2612 Qiu J, Zhao J. (n.d.) Null systems in the non-minimal case.
Li J, Yan K, Ye X (2015b) Recurrence properties and Ergodic Theory Dyn Syst. https://www.cambridge.org/
disjointness on the induced spaces. Discrete Contin core/journals/ergodic-theory-and-dynamical-systems/
Dyn Syst 35:1059–1073 article/null-systems-in-the-nonminimal-case/F8AD15
Lindenstrauss E (1995) Lowering topological entropy. 10DBAE43CA94CC0CB5A3DD22DD
J Anal Math 67:231–267 Rado R (1933) Studien zur Kombinatorik (German). Math
Lindenstrauss E (1999) Measurable distal and topological Z 36:424C470
distal systems. Ergodic Theory Dyn Syst 19:1063–1076 Rohlin, VA (1967) Lectures on the entropy theory of trans-
Liu K (2019) D-weakly mixing subset in positive entropy formations with invariant measure. (Russian) Uspehi
actions of a nilpotent group. J Differ Equ 267:525–546 Mat. Nauk 22. 137(5):3–56
426 Parallels Between Topological Dynamics and Ergodic Theory

Rohlin VA, Sinai YG (1961) Construction and properties Veech WA (1970) Point-distal flows. Am J Math
of invariant measurable partitions. Dokl Akad Nauk 92:205–242
SSSR 141:1038–1041. [In Russian] Veech WA (1977) Topological systems. Bull Am Math Soc
Romagnoli PP (2003) A local variational principle for the 83:775–830
topological entropy. Ergodic Theory Dyn Syst Walsh M (2012) Norm convergence of nilpotent ergodic
23:1601–1610 averages. Ann Math 175:1667–1688
Rudolph D (1979) An example of a measure preserving Walters P (1982) An introduction to ergodic theory. Grad-
map with minimal self-joinings, and applications. uate texts in mathematics, 79. Springer, New York
J Anal Math 35:97–122 Wang YP, Chen EC, Zhou XY (2019) Mean Li-Yorke
Rudolph DJ (1990) Fundamentals of measurable dynam- chaos for random dynamical systems. J Diff Equat
ics. Oxford Science publications. The Clarendon Press 267(4):2239–2260
Oxford University Press, New York, p x,168 Ward T (2012) Ergodic theory: interactions with combina-
Ryzhikov VV (1993) Joinings and multiple mixing of the torics and number theory. Mathematics of complexity
actions of finite rank. (Russian) Funktsional. Anal i and dynamical systems, vol 1–3. Springer, New York,
Prilozhen 27: 63–78, 96; translation in Funct. Anal pp 313–326
Appl 27:128–140 Weiss B (1985) Strictly ergodic models for dynamical
Saleski A (1977) Sequence entropy and mixing. J Math systems. Bull Amer Math Soc (NS) 13:143–146
Anal Appl 60:58–66 Weiss B (1989) Countable generators in dynamics – uni-
Sarnak P (2009) Three lectures on the Mobius function versal minimal models. Contemp Math 94:321–326
randomness and dynamics. http://publications.ias.edu/ Weiss B (1995) Multiple recurrence and doubly minimal
sarnak/paper/512 systems. Topological dynamics and applications
Schweizer B, Smital J (1994) Measures of chaos and a (Minneapolis, MN, 1995). Contemporary mathematics,
spectral decomposition of dynamical systems on the 215. American Mathematical Society, Providence,
interval. Trans Am Math Soc 344:737–754 199, pp 189–196
Serafin J (2013) Non-existence of a universal zero-entropy Weiss B (2000) Single orbit dynamics. CBMS regional
system. Israel J Math 194:349–358 conference series in mathematics, 95. American Math-
Shao S, Ye X (2012) Regionally proximal relation of order ematical Society, Providence
d is an equivalence one for minimal systems and a Ye X, Zhang G (2007) Entropy points and applications.
combinatorial consequence. Adv Math 231:1786–1817 Trans Am Math Soc 359:6167–6186
Sinai YJ (1959) On the concept of entropy for a dynamic Ye X, Zhang R (2008) Countable compacta admitting
system. (Russian) Dokl Akad Nauk SSSR homeomorphisms with positive sequence entropy.
124:768–771 J Dyn Diff Equat 20(4):867–882
Snoha L, Ye X, Zhang R (2018) Topology and topological Yu T (2019) Measure-theoretic mean equicontinuity and
sequence entropy. Sci China Math. arXiv:1810.00497 bounded complexity. J Diff Equat 267:6152–6170
Szegedy B (2012) On higher order Fourier analysis. arXiv: Zhang Q (1992) Sequence entropy and mild mixing. Can
1203.2260 J Math 44:215–224
Szemerédi E (1975) On sets of integers containing no Zhang Q (1993) Conditional sequence entropy and mild
k elements in arithmetic progression. Acta Arith mixing extensions. Can J Math 45:429–448
27:199–245 Zhang G (2006) Relative entropy, asymptotic pairs and
Tan F (2011) The set of sequence entropies for graph maps. chaos. J Lond Math Soc 73(1):157–172
Topology Appl 158(3):533–541 Zhou XM, Chen ER, Zhou XY (2017) Relative entropy
Tan F, Ye X, Zhang R (2010) The set of sequence entropies and mean Li-Yorke chaos. Adv Math (China)
for a given space. Nonlinearity 23(1):159–178 46:407–414
Tao T (2008) Norm convergence of multiple ergodic aver- Ziegler T (2005) A non-conventional ergodic theorem for a
ages for commuting transformations. Ergodic Theory nilsystem. Ergodic Theory Dyn Syst 25:1357–1370
Dyn Syst 28:657–688 Ziegler T (2007) Universal characteristic factors and
van derWaerden BL (1927) Beweis einer Baudetschen Furstenberg averages. J Am Math Soc 20:53–97
Vermutung. Nieuw Arch Wisk 15:212–216 Zimmer RJ (1976a) Extensions of ergodic group actions.
Veech WA (1968) The equicontinuous structure relation Ill J Math 20:373–409
for minimal Abelian transformation groups. Am J Math Zimmer RJ (1976b) Ergodic actions with generalized dis-
90:723–732 crete spectrum. Ill J Math 20:555–588
 automorphism, captures many invariants of
Symbolic Dynamics topological conjugacy for shifts of finite type.
Embedding (Section “Shift Spaces and Sliding
Brian Marcus Block Codes”) A one-to-one sliding block
Department of Mathematics, University of British code from one shift space to another; equiva-
Columbia, Vancouver, BC, Canada lently, a one-to-one continuous shift-
commuting mapping from one shift space to
another.
Article Outline Factor map (Section “Shift Spaces and Sliding
Block Codes”) An onto sliding block code
Glossary from one shift space to another; equivalently,
Definition of the Subject an onto continuous shift-commuting mapping
Introduction from one shift space to another. Sometimes
Origins of Symbolic Dynamics: Modeling of called Factor Code.
Dynamical Systems Finite equivalence (Section “Other Coding
Shift Spaces and Sliding Block Codes Problems”) A common extension of two shift
Shifts of Finite Type and Sofic Shifts spaces given by finite-to-one factor codes.
Entropy and Periodic Points Full shift (Section “Shift Spaces and Sliding
The Conjugacy Problem Block Codes”) The set of all bi-infinite
Other Coding Problems sequences over an alphabet (together with the
Coding for Data Recording Channels shift mapping). Typically, the alphabet is finite.
Connections with Information Theory and Higher dimensional shift space (Section
Ergodic Theory “Higher Dimensional Shift Spaces”) A set of
Higher Dimensional Shift Spaces bi-infinite arrays of a given dimension, deter-
Future Directions mined by a collection of finite forbidden
Addendum to the Second Edition arrays. Typically, the alphabet is finite.
Bibliography Markov partition (Section “Origins of Sym-
bolic Dynamics: Modeling of Dynamical Sys-
Glossary tems”) A finite cover of the underlying phase
space of a dynamical system, which allows the
Almost conjugacy (Section “Other Coding system to be modeled by a shift of finite type.
Problems”) A common extension of two shift The elements of the cover are closed sets,
spaces given by factor codes that are one-to- which are allowed to intersect only on their
one almost everywhere. boundaries.
Automorphism (Section “The Conjugacy Prob- Measure of maximal entropy (Section “Connec-
lem”) An invertible sliding block code from a tions with Information Theory and Ergodic The-
shift space to itself; equivalently, a shift- ory”) A shift-invariant measure of maximal
commuting homeomorphism from a shift measure-theoretic entropy on a shift space. Its
space to itself; equivalently, a topological measure-theoretic entropy coincides with the
conjugacy from a shift space to itself. topological entropy of the shift space.
Dimension group (Section “The Conjugacy Road problem (Section “Other Coding Prob-
Problem”) A particular group associated to a lems”) A recently-solved classical problem in
shift of finite type. This group, together with a symbolic dynamics, graph theory and autom-
distinguished sub-semigroup and an ata theory.

© Springer Science+Business Media, LLC, part of Springer Nature 2023 427

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_531
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2020
https://doi.org/10.1007/978-3-642-27737-5_531-2
428 Symbolic Dynamics

Run-length limited shift (Section “Coding for equivalently, a shift-commuting homeomor-

Data Recording Channels”) The set of all phism from one shift space to another. Some-
bi-infinite binary sequences whose runs of times called conjugacy.
zeros, between two successive ones, are Topological entropy (Section “Entropy and
bounded below and above by specific Periodic Points”) The asymptotic growth rate
numbers. of the number of finite sequences of given
Shift equivalence (Section “The Conjugacy length in a shift space (as the length goes to
Problem”) An equivalence relation on defining infinity).
matrices for shifts of finite type. This relation Zeta function (Section “Entropy and Periodic
characterizes the corresponding shifts of finite Points”) An expression for the number of peri-
type, up to an eventual notion of topological odic points of each given period in a shift
conjugacy. space.
Shift space (Section “Shift Spaces and Sliding
Block Codes”) A set of bi-infinite sequences
determined by a collection of finite forbidden Definition of the Subject
words; equivalently, a closed shift-invariant
subset of a full shift. Symbolic dynamics is the study of shift spaces,
Shift of finite type (Section “Shifts of Finite which consist of infinite or bi-infinite sequences
Type and Sofic Shifts”) A set of bi-infinite defined by a shift-invariant constraint on the
sequences determined by a finite collection of finite-length subwords. Mappings between two
finite forbidden words. such spaces can be regarded as codes or
Sliding block code (Section “Shift Spaces and encodings. Shift spaces are classified, up to vari-
Sliding Block Codes”) A mapping from one ous kinds of invertible encodings, by combinato-
shift space to another determined by a finite rial, algebraic, topological, and measure-theoretic
sliding block window; equivalently, a continu- invariants.
ous shift-commuting mapping from one shift The subject is intimately related to many other
space to another. areas of research, including dynamical systems,
Sofic shift (Section “Shifts of Finite Type and ergodic theory, automata theory and information
Sofic Shifts”) A shift space which is a factor theory. Shift spaces and their associated shift map-
of a shift of finite type; equivalently, a set of pings are used to model a rich and important class
bi-infinite sequences determined by a finite of smooth dynamical systems and ergodic
directed labeled graph. measure-preserving transformations. These
State splitting (Section “The Conjugacy Prob- models have provided a valuable tool for classi-
lem”) A splitting of states in a finite directed fying and understanding fundamental properties
graph that creates a new graph, whose vertices of dynamical systems. In addition, techniques
are the split states. The operation that creates from symbolic dynamics have had profound
the new graph from the original graph is a basic applications for data recording applications, such
building block for all topological conjugacies as algorithms and analysis of invertible
between shifts of finite type. encodings, and problems in matrix theory, such
Strong shift equivalence (Section “The as characterization of the set of eigenvalues of a
Conjugacy Problem”) An equivalence relation nonnegative matrix.
on defining matrices for shifts of finite type. In
principle, this relation characterizes the
corresponding shifts of finite type, up to topo- Introduction
logical conjugacy.
Topological conjugacy (Section “Shift Spaces This article is intended to give a picture of major
and Sliding Block Codes”) A bijective sliding topics in symbolic dynamics. Section “Origins of
block code from one shift space to another; Symbolic Dynamics: Modeling of Dynamical
Symbolic Dynamics 429

Systems” reviews the roots of symbolic dynamics The book (1985) by Berstel and Perrin treats the
in modeling of dynamical systems. Section “Shift subject of variable length codes and contains
Spaces and Sliding Block Codes” lays the foun- many ideas related to symbolic dynamics. Finally,
dation by defining the kinds of spaces and map- there is an excellent recent survey by Boyle on
pings considered in the subject. Section “Shifts of Open Problems in Symbolic Dynamics (Boyle
Finite Type and Sofic Shifts” focuses on distin- 2007).
guished special classes of spaces, known as shifts
of finite type and sofic shifts. Section “Entropy
and Periodic Points” introduces the most funda- Origins of Symbolic Dynamics: Modeling
mental invariants, periodic points and topological of Dynamical Systems
entropy. Sections “The Conjugacy Problem” and
“Other Coding Problems” survey progress on the Symbolic dynamics began as an effort to model
conjugacy problem and other classification/cod- dynamical systems using sequences of symbols.
ing problems for shifts of finite type and sofic A dynamical system is a pair (X, T ) where X is a
shifts. In section “Coding for Data Recording set and T is a transformation from X to itself. For
Channels,” we present applications to coding for definiteness, we assume that T is invertible,
data recording. Section “Connections with Infor- although this is not a necessary restriction. Since
mation Theory and Ergodic Theory” provides a T maps X to itself, we can iterate T : T2 ¼ T ∘ T,
link with information theory and ergodic theory. T3 ¼ T ∘ T ∘ T, etc. The orbit of a point x  X is
Finally, section “Higher Dimensional Shift the sequence of points: . . ., T 2(x), T 1(x), x,
Spaces” treats higher dimensional symbolic T(x), T2(x), . . . In the theory of dynamical sys-
dynamics. tems, one asks questions about orbits such as the
While this article covers many of the most following: Are there periodic orbits (i.e., x such
important topics in the subject, others have been that Tn(x) ¼ x for some n > 0)? Are there dense
omitted or treated lightly, due to space limitations. orbits (the orbit of x is dense if for any point y in X,
These include one-sided shift spaces, countable Tn(x) is “close” to y for some n)? How does the
state symbolic systems, orbit equivalence, flow behavior of an orbit vary with x? How can we
equivalence, the automorphism group, cellular describe the collection of all orbits of the dynam-
automata, and substitution systems. References ical system? When is the dynamical system “cha-
to work in these subareas can be found in the otic”? For more information on dynamical
sources mentioned below. systems, we refer the reader to (Blanchard et al.
For introductory reading on symbolic dynam- 2004; Devaney 1987; Hassellblatt and Katok
ics and its applications, beyond this article, one 1995).
can consult the textbooks Kitchens (1998) and The subject of dynamical systems has its roots
Lind and Marcus (1995). There are also excellent in Classical Mechanics; in that setting, X is the set
introductory survey articles, such as Boyle of all possible states of a system (e.g., the posi-
(1993), Lind and Schmidt (2002), and tions, momenta of all particles in a physical sys-
S. Williams (2004a). In addition, there are very tem), and the transformation T is the time
good expositions which focus on other aspects of evolution map, which maps the state of the system
the subject. These include Beal (1993), which at one time to the state of the system at one time
focuses on connections between symbolic dynam- unit later.
ics and automata theory, the lecture notes Marcus- Symbolic dynamics provides a model for the
Roth-Siegel (1998), which focuses on constrained orbits of a dynamical system (X, T) via a space of
coding applications, and Immink (2004a), which sequences. This is done by “quantizing” X into
focuses on applications to data storage. There are cells, associating symbols to the cells and
also several excellent collections of articles on representing points as bi-infinite sequences of
special areas of the subject, such as Blanchard symbols. For instance, in Fig. 1, X is a square,
et al. (2000), Walters (1992), Williams (2004b). and T is some transformation of the square.
430 Symbolic Dynamics

how X is quantized into cells. Also, Y is typically a

set of sequences constrained by certain rules, such
as a certain symbol may only be followed by
certain other symbols.
In this way, one can use symbolic dynamics to
study dynamical systems. Properties of orbits of
the original dynamical system are reflected in
properties of the resulting sequences. For
instance, a point whose orbit is periodic becomes
a periodic sequence, and the distribution of the
orbit of a point x in X is reflected in the distribution
of finite strings within s(x). Beginning with
Hadamard (1898) in 1898 and followed by
Symbolic Dynamics, Fig. 1 Representing points Hedlund, Morse and others in the 1920s, 1930s,
symbolically and 1940s (Hedlund 1939, 1944; Morse and
Hedlund 1938, 1940), this method was used to
We have drawn a portion of the orbit of a point prove the existence of periodic, almost periodic
x  X for the dynamical system (X, T). We have and other interesting motions in classical dynam-
also quantized X into two cells: the left half, called ical systems, such as geodesic flows on surfaces of
“0,” and the right half, called “1.” Then the point negative curvature; this was done by finding inter-
x is represented by the bi-infinite sequence s- esting sequences satisfying the constraints defined
(x) ¼ . . .s2s1. s0s1s2. . . where sn is the label of by the corresponding symbolic dynamical system.
the cell to which Tn(x) belongs (here, we use the Later on, this was extended to general hyperbolic
decimal point to separate coordinates si, i < 0 from systems, where the symbolic dynamics is
si, i  0). So, for x as given in Fig. 1, we see that constructed using a Markov Partition, which is a
disjoint collection of open sets whose closures
sðxÞ ¼ . . . 11:001 . . . cover X, each of which looks like a “rectangle”,
with vertical (resp., horizontal) fibers contracted
for instance, s0 ¼ 0 because x belongs to the left (resp., expanded) by T. Markov Partitions were
half of the square, and s2 ¼ 1 because T2(x) developed by Adler and Weiss (1970), Sinai
belongs to the right half of the square. Now, if (1968), and Bowen (1970, 1973); see also Bed-
x is represented by the sequence s(x), then T(x) is ford (1986) and Sect. 6.5 in Lind and
represented by the shift of s(x): Marcus (1995).
In more recent years, symbolic dynamics has
sðT ðxÞÞ ¼ . . . 110:01 . . . : been used as a tool in classification problems for
dynamical systems. Here, the problem of deter-
So, T is represented ”symbolically” as the shift mining when one dynamical system is “‘equiva-
transformation. lent” to another becomes, via symbolic dynamics,
By representing all points of X as bi-infinite a coding problem. Roughly speaking, two dynam-
sequences, we obtain a symbolic dynamical sys- ical systems, (X1, T1) and (X2, T2), are equivalent
tem (Y, s) where Y is a set of sequences if there is an invertible mapping from X1 to X2
(representing X) and s is the shift transformation which makes T1 “look like” T2. If the
(representing T). The “symbolic” refers to the corresponding symbolic dynamical systems are
symbols, and the “dynamical” refers to the action denoted (Y1, s) and (Y2, s), then an equivalence
of the shift transformation. between (X1, T1) and (X2, T2) becomes a time-
For this representation to be faithful, distinct invariant, invertible encoding from Y1 to Y2
points should be represented by distinct (time-invariant because the shift transformation
sequences, and this imposes extra conditions on represents the dynamics). Thus, the classification
Symbolic Dynamics 431

problem in dynamical systems leads to a coding We are interested in sets that can be specified
problem between constrained sets of sequences. by a list (finite or infinite) of forbidden blocks.
We have described dynamical systems as the Namely, given a collection F of “forbidden
discrete-time iteration of a single mapping. How- blocks” over A , the subset X consisting of all
ever, continuous-time iterations have been studied sequences in A Z, none of whose subwords belong
since the inception of the subject. These are to F , is called a shift space (or simply shift), and
known as continuous-time flows, with the main we write X ¼ XF . When a shift space X is
example being the set of solutions to a system of contained in a shift space Y, we say that X is a
ordinary differential equations. Indeed, the work subshift of Y.
of Hedlund and Morse mentioned above was done
in this context.
Example 1 X is the set of all binary sequences
with no two 1’s next to each other. Here X ¼ XF ,
where F ¼ {11}. This shift is called the golden
Shift Spaces and Sliding Block Codes
mean shift, for reasons that will become apparent
later.
Let A be an alphabet of symbols, which we
assume to be finite. The principal objects of
study in symbolic dynamics are certain kinds of Example 2 X is the set of all binary sequences so
collections of sequences of symbols from A. Typ- that between any two 1’s there are an even number
ically, these sequences are infinite x ¼ x0 x1x2. . ., of 0’s. We can take for F the collection
but it is often more convenient to deal with
bi-infinite sequences x ¼ . . . x2x1x0 x1x2. . . 102nþ1 1 : n  0 :
For some problems, the results are similar in the
infinite and bi-infinite categories, while for other
This example is naturally called the even shift.
problems, they are quite different. In this article,
we focus on the bi-infinite setting.
The symbol xi is the ith coordinate of x. When Example 3 X is the set of all binary sequences
writing a specific sequence, we need to specify such that between any two successive 1’s, number
which is the 0th coordinate. As suggested in sec- of 0’s is prime. We can take for F the collection
tion “Origins of Symbolic Dynamics: Modeling
of Dynamical Systems,” this is done with a deci- f10n 1 : n is compositeg :
mal point to separate the xi with i  0 from those
with i < 0: x ¼ . . . x2x1. x0 x1x2. . . . A block or This example is naturally called the prime shift.
word over A is a finite sequence of symbols from Alternatively (and equivalently), shift spaces
A . A block of length N is called an N-block. For can be defined as closed, shift-invariant subsets of
blocks u, v, the block uv is the concatenation of full shifts. Here, “closed” means with respect to a
u and v, and for a block w, the concatenation of metric, for which two points are close if they agree
N copies of w is denoted wN. in a large “central block”; one such metric is
The full A -shift A Z is the set of all bi-infinite r(x, y) ¼ 2k if x 6¼ y, with k maximal such that
sequences of symbols from A . The full r-shift is x[k, k] ¼ y[k, k] (with the conventions that
the full shift over the alphabet {0, 1, . . ., r  1}. r(x, y) ¼ 0 if x ¼ y and r(x, y) ¼ 2 if x0 6¼ y0).
The shift map s on a full shift maps a point x to the Let X be a subset of a full shift, and let ℬN(X)
point y ¼ s(x) whose ith coordinate is yi ¼ xi+1. denote the set of all N-blocks that occur in ele-
The orbit of a point in a full shift is its orbit ments of X. The language of X is
under the shift map. The full shift contains many ℬ(X) ¼ [NℬN(X). It can be shown that the lan-
different types of orbits. For instance, it contains a guage of a shift space determines the shift space
dense orbit (namely, any sequence which contains uniquely, and so we can equally well describe a
every block in the alphabet) and periodic orbits shift space by specifying the “occurring” or “allo-
(namely, any sequence which is periodic). wed” blocks, rather than the forbidden blocks. For
432 Symbolic Dynamics

example, the golden mean shift is specified by the ðbN ðxÞÞi ¼ x½i,iþN1 :
language of blocks in which 1’s are isolated.
This establishes a connection with automata Then the Nth higher block shift or higher block
theory (Aho et al. 1974; Béal 1993), which studies presentation of a shift space X is the image
collections of blocks, rather than infinite or X[N] ¼ bN(X) in the full shift over A ðNÞ.
bi-infinite sequences. The languages that occur Similarly, define the Nth higher power code
in symbolic dynamics, i.e., as ℬ(X) for some gN : X ! A ðNÞ by
Z

shift space X, are simply those sets ℒ of blocks

that satisfy two simple properties: every sub-block ðgN ðxÞÞi ¼ x½iN ,iN þN1 :
of an element of ℒ belongs to ℒ, and every
element w  ℒ is extendable to a larger block The Nth higher power shift XN of X is the image
awb  ℒ, with a, b  A. X ¼ gN(X) of X. The difference between X[N] and
N
Some examples are best described by allowed XN is that the former is constructed by considering
blocks. One example is the famous Morse shift. overlapping blocks and the latter by non-
overlapping blocks.
Example 4 Let A0 ¼ 0 and inductively define Next, we turn to mappings between shift
blocks Anþ1 ¼ An An , where An denotes bitwise spaces. Suppose that x ¼ . . .x1. x0x1. . . is a
complement. The shift space whose allowed sequence in a shift space X over A . We can
blocks are the sub-locks of the An is called the transform x into a new sequence y ¼ . . .y1.
Morse shift. y0y1. . . over another alphabet C as follows. Fix
integers m and n with m  n. To compute the ith
This shift space has an alternative description coordinate yi of the transformed sequence, we
as follows. Since each An is a prefix of An+1, the use a function F that depends on the “window”
sequence of blocks An determines a unique right- of coordinates of x from position i  m to posi-
infinite sequence x+ (with each An as a prefix). If tion i + n. Here F : ℬmþnþ1 ðXÞ ! C is a fixed
we denote x as the left-infinite sequence obtained block map, called an (m + n + 1)-block map from
by writing x+ backwards, then the Morse shift is allowed (m + n + 1)-blocks in X to symbols in C ,
the closure of the orbit of the point x. x+. The and so
sequence x+ is known as the Prouhet-Thue-Morse
(PTM) sequence, since Prouhet and Thue intro- yi ¼ Fðxim ximþ1 . . . xiþn Þ ¼ F x½im,iþn :
duced it earlier, but for different purposes.
While it is not immediately obvious, it can be
This is illustrated in Fig. 2.
shown that the PTM sequence is not periodic;
moreover, the Morse shift is a minimal shift,
which means that all its orbits are dense (Morse
1921). In contrast, while the full, golden mean,
even and prime shifts all have dense orbits, each
also has a dense set of periodic orbits.
There are two simple, but very important, con-
structions in symbolic dynamics that construct
from a given shift space a new version which in
some sense looks deeper into the space at the cost
of a larger alphabet and more complex description.
Let X be a shift space over the alphabet A ,
and A ðNÞ ¼ ℬN ðXÞ. We can consider A ðNÞ as an
alphabet in its own right, and form the full shift
Z
A ðNÞ . Define the Nth higher block code by Symbolic Dynamics, Fig. 2 Sliding block code
Symbolic Dynamics 433

Let X be a shift space over A , and F : one shift space X to another Y, we say that X and
ℬmþnþ1 ðXÞ ! C be a block map. Then the map Y are conjugate, denoted X ffi Y.
f : X ! C Z defined by y ¼ f(x), with yi given by As an example, the higher block map bN is a
F above, is called the sliding block code with conjugacy between a shift space X and its higher
memory m and anticipation n induced by F. We block shift X[N]. Via this code, we can “re-code”
will denote the formation of f from F by f ¼ any sliding block code as a 1-block code (though
F½1
m,n
, or more simply by f ¼ F1 if the memory typically a conjugacy and its inverse cannot, by
and anticipation of f are understood. If not spec- this artifice, be simultaneously re-coded to
ified, the memory is taken to be 0. If Y is a shift 1-block codes).
space contained in C Z and f(X)  Y, we write In this section, we have given examples of
f : X ! Y. relatively simple sliding block codes. But the typ-
In analogy with the characterization of shift ical conjugacy, as well as factor code and embed-
spaces as closed shift-invariant sets, sliding ding, can be much more complicated.
block codes can be characterized in a topological
manner: namely, as the maps between shift spaces
that are continuous and commute with the shift.
Shifts of Finite Type and Sofic Shifts
This result is known as the Curtis-Hedlund-
Lyndon theorem (Hedlund 1969).
A shift of finite type (SFT) is a shift space that can
be described by a finite set of forbidden blocks,
Example 5 Let A ¼ f0, 1g ¼ C, X ¼ A Z, m ¼ 0, i.e., a shift space X having the form XF for some
n ¼ 1, and F(a0a1) ¼ a0 + a1 (mod2). Let finite set F of blocks. The terminology shift of
f ¼ F1 : X ! X. finite type (or subshift of finite type) comes from
dynamical systems (Smale 1967).
Example 6 The sliding block code, generated by An SFT is M-step (or has memory M) if it can
F(00) ¼ 1, F(01) ¼ 0 ¼ F(10), maps the golden be described by a collection of forbidden blocks
mean shift onto the even shift. all of which have length M + 1. It is easy to see that
any SFT is M-step for some M.
Example 7 There is a trivial sliding block code Since any shift space can be defined by many
from the full 2-shift into the full 3-shift, generated different collections of forbidden blocks, it is use-
by F(0) ¼ 0, F(1) ¼ 1. ful to have the following equivalent condition
expressed in terms of allowed blocks: an SFT is
If a sliding block code f : X ! Y is onto, then f M-step if and only if whenever u is an allowed
is called a factor code or factor map, and Y is a block of length at least M, u0 is the suffix of u with
factor of X. If f : X ! Y is one-to-one, then f is length M and a is a symbol, then ua is allowed if
called an embedding of X into Y. The sliding block and only if u0a is allowed. In other words, in order
code in Example 7 is an embedding but not a to tell whether a symbol can be allowably
factor code, while the codes in Examples 5 and 6 concatenated to the end of an allowed word u,
are factor maps, but not embeddings. one need only look at the last M symbols of u.
A major (and unrealistic) goal of symbolic This is analogous to the “finite memory” property
dynamics is to classify in an explicit way shift of M-step Markov chains.
spaces up to the following natural notion of equiv- The golden mean shift X is a 1-step SFT, since
alence. A sliding block code f : X ! Y is a it was defined by a forbidden list consisting of
conjugacy (or topological conjugacy) if it is exactly one block: F ¼ {11}. Equivalently, it is
invertible with sliding block inverse. Equiva- only the last symbol of an allowed block that
lently, a conjugacy is a bijective sliding block determines whether a given symbol can be
code and therefore simultaneously a factor code concatenated at the end. In contrast, the even
and an embedding. If there is a conjugacy from shift is not an SFT: for any M, the symbol 1 can
434 Symbolic Dynamics

be concatenated to the end of exactly one of the “connecting” block v such that uvw is allowed.
(allowed) words 10M and 10M+1. While shift spaces do not always decompose into
Recall that the higher block code bM is a disjoint unions of irreducible shifts, every SFT can
conjugacy from X to X[M]. Via this code, any be written as a finite disjoint union of irreducible
SFT can be recoded to a 1-step SFT. And so, any SFT’s Xi together with “transient” one-way con-
sliding block code on a shift space can be recoded nections from one Xi to another. And irreducible
to a 1-block code on a 1-step SFT. It is useful to edge shifts can be characterized in a particularly
have a concrete description of 1-step SFT’s. In concrete form: namely, XG is irreducible if and
fact, these are precisely the shift spaces consisting only if G is irreducible, i.e., for every ordered pair
of all bi-infinite sequences of vertices along paths of vertices I and J there is a path in G starting at
on a finite directed graph. These are called vertex I and ending at J.
shifts. We find it more convenient to work with There is a stronger notion which is defined by a
sequences of edges instead. To be precise: uniformity condition on the length of the
Let G be a finite directed graph (or simply connecting block. A shift space is mixing if when-
graph) with vertices (or states) V ¼ V ðGÞ and ever u and w are allowed blocks, there is an N,
edges E ¼ E ðGÞ. For an edge e, i(e) denotes the possibly depending on u and w, such that for all
initial state and t(e) the terminal state. A path in n  N, there is block v of length n such that uvw is
G is a finite sequence of edges in G such that the allowed. And an edge shift XG is mixing if and
terminal state of an edge coincides with the initial only if G is primitive, i.e., there is an integer
state of the following edge; a cycle in G is path N such that for any n  N and any ordered pair
that begins and ends at the same state. We will of vertices I and J, there is a path in G of length
assume that G is essential, i.e., that every state has n starting at I and terminating at J. It follows that
at least one outgoing edge and one incoming edge. for SFT’s in the definition of mixing, the uniform
The adjacency matrix A ¼ A(G) is the matrix connecting length N can be chosen independent of
indexed by V with AIJ equal to the number of the allowed blocks u and w.
It can be shown that, in some sense, any irre-
edges in G with initial state I and terminal state J.
ducible SFT X can be broken down into a union of
Since a graph and its adjacency matrix essentially
disjoint maximal mixing shifts; namely, X can be
determine the same information, we will fre-
written as the disjoint union of finitely many sets
quently associate a graph G with its adjacency
Xi, i ¼ 0, . . ., p  1 such that s(Xi) ¼ Xi+1 mod p
matrix A and a nonnegative integer matrix A with
and for each i, sp restricted to Xi can be regarded
a graph G.
as a mixing SFT. This is a consequence of Perron-
The edge shift XG or XA is the shift space over
Frobenius theory, upon which symbolic dynamics
the alphabet A ¼ E defined by
relies heavily; see Seneta (1980) for an introduc-
tion to this theory.
XG ¼ XA
A sofic shift is the set of bi-infinite sequences
¼ x ¼ ðxi Þi  Z  E Z : each xiþ1 follows xi : obtained from a finite labeled directed graph
G ¼ ðG, ℒÞ ; here, G is a finite directed graph
It can be readily verified that edge shifts are and ℒ is a labeling of the edges of G. The labeled
1-step SFT’s. While edge shifts do not include all graph is often called a presentation of the sofic
1-step SFT’s, any 1-step SFT can be recoded to an shift. The golden mean shift and even shift are
edge shift, and, compared with vertex shifts, edge sofic, with presentations given in Figs. 3 and 4.
shifts offer the advantage of a more compact SFT’s are sofic because any M-step SFT can be
description. presented by a graph whose states are allowed
For many purposes, one can study a general M-blocks. Note also that any sofic shift is a factor
shift space X by breaking it into smaller, more of an SFT, namely, via a (1-block) factor code ℒ1
well-behaved pieces. A shift space is irreducible on the edge shift XG based on a presentation
if whenever u and w are allowed blocks, there is a (G, ℒ). In fact, the converse is true, and so, the
Symbolic Dynamics 435

the definition of right resolving, and replace “ini-

tial” with “terminal” in the definition of right
closing).
While the class of SFT’s is defined by a “finite-
Symbolic Dynamics, Fig. 3 Presentation of golden
mean shift
memory” property, the more general class of sofic
shifts is defined by a “finite-state” property, in that
the possible symbols that can occur at time 0 are
determined by the past via one of finitely many
states. In a presentation, the vertices can be
viewed as state information which connects
Symbolic Dynamics, Fig. 4 Presentation of even shift sequences in the past with sequences in the future.
It is not difficult to show that neither the prime
sofic shifts are precisely the shift spaces that are shift (Example 3) nor the Morse shift (Example 4)
factors of SFT’s. This was the original definition are sofic and therefore also not SFT. For the prime
of sofic shifts given by Weiss (1973). shift, this can be done by an application of the
Typically, the labeling is right resolving, which pumping lemma from automata theory (Aho et al.
means that at any given state, all outgoing edges 1974) (or p. 68 of (Lind and Marcus 1995)).
have distinct labels (as in Figs. 3 and 4). There are uncountably many shift spaces, but
only countably many sofic shifts. So, it is not
Theorem 1. surprising that the behavior of sofic shifts is very
1. Any sofic shift has a right resolving special. However, they are very useful in model-
presentation. ing smooth dynamical systems (Section “Origins
2. Any irreducible sofic shift has a unique mini- of Symbolic Dynamics: Modeling of Dynamical
mal right resolving presentation. Systems”) and in information theory and applica-
Part (a) is a direct consequence of the subset tions to data recording (Section “Coding for Data
construction in automata theory (Aho et al. 1974; Recording Channels”), where they arise as
Béal 1993) which constructs a right resolving constrained systems, although this term is usually
presentation from an arbitrary presentation; see reserved for the set of (finite) blocks obtained
also Coven and Paul (1975, 1977). Part from a finite directed labeled graph (Marcus
(b) makes use of the state-minimization algorithm et al. 1998).
from automata theory (Aho et al. 1974; Béal
1993), but requires an idea beyond that found in
automata theory (Fischer 1975a, b). The unique Entropy and Periodic Points
presentation in part (b) may be regarded as a
canonical presentation. We remark that for an An invariant of conjugacy is an object associated
irreducible (resp. mixing) sofic shift, the underly- to a shift space that is preserved under conjugacy.
ing graph of the unique minimal right resolving It can be shown that many of the concepts that we
presentation is irreducible (resp., primitive). have already introduced are invariants: irreduc-
Sometimes, it is useful to weaken the concept ibility, mixing, as well as the properties of being
of right resolving to right closing, which means a shift of finite type or sofic shift. Beyond these
“right resolving with delay”; more precisely a qualitative invariants, there are many quantitative
labeling is right closing, with delay D if all paths invariants that can, in many cases, be computed
of length D + 1 with the same initial state and the explicitly. Foremost among these is topological
same label have the same initial edge. Also, we entropy.
sometimes consider left resolving and left closing The (topological) entropy (or simply entropy)
labelings (replace “outgoing” with “incoming” in of X is:
436 Symbolic Dynamics

log j ℬN ðXÞ j decomposition into primitive, and then irreduc-

hðXÞ ¼ lim ;
N!1 N ible, graphs.
To extend this to a sofic shift Y, one uses a right
here, j  j denotes cardinality, and for definiteness resolving presentation (G, ℒ) of Y. Since every
the log is taken to mean log2 but any base will block of Y is the label of at most V ðGÞ paths in G,
do. A subadditivity argument shows that the limit it follows that:
does indeed exist (Walters 1982). It should be
evident that h(X) is a measure of the “size” or Theorem 3 Let G ¼ ðG, ℒÞ be a right-resolving
“complexity” of X, as it is simply the asymptotic labeled graph presenting a sofic shift Y. Then
growth rate of the number of blocks that occur in h(Y) ¼ h(XG).
X. Topological entropy for continuous dynamical
systems was defined by Adler, Konheim, and From Fig. 3, we see that the golden mean shift
McAndrew (1965), in analogy with measure- is obtained as a right resolving presentation of the
theoretic entropy. graph with adjacency matrix:
Since a k-block sliding block code from a shift
space X to a shift space Y maps ℬN+k1(X) into 1 1
A¼ :
ℬN(Y), it is not hard to see that entropy is an 1 0
invariant of conjugacy and that it cannot increase
under factors and cannot decrease under A computation shows that lA is the golden
embeddings. mean, and so the entropy of the golden mean
In many cases, one can explicitly compute shift is the log of the golden mean; this is one
entropy. For example, for the full r-shift X, jℬN(X) j explanation of the meaning of the term golden
¼ rN, and so h(X) ¼ log r. And from the defining mean shift. From Fig. 4, we see that the even
sequence of the Morse shift, we see that the number shift has the same entropy.
of distinct 2N-blocks is at most 4(2N), it follows that One cannot overstate the importance of
the growth rate cannot be exponential and so the entropy as an invariant. Yet, it is somewhat
entropy of the Morse shift is zero. crude; it is perhaps not surprising that a single
For SFT’s and more generally for sofic shifts, numerical invariant would not be sufficient to
entropy can be computed explicitly. The key to completely capture the many intricacies of shift
this computation is the following result, which is spaces, even of sofic shifts or SFT’s.
based on the Perron-Frobenius theorem. An invariant finer than entropy is the zeta func-
tion, which combines information regarding the
Theorem 2 Parry (1964), Shannon (1948) For numbers of periodic sequences of all periods,
any graph G, h(XG) ¼ log lA(G), where lA(G) is the described as follows.
largest eigenvalue of A(G). For a shift space X, let pn(X) denote the number
of points in X of period n (i.e., the number of
The rough idea is that the number of N-blocks x  X such that sn(x) ¼ x). It is straightforward
in XG is the number of paths of length N and thus to show that each pn(X) is an invariant. Since
also the sum of the entries of A(G)N, which is distinct periodic sequences define distinct blocks,
controlled by the largest eigenvalue. This is it follows that
proven first for primitive graphs, whose adjacency
matrices have a unique eigenvalue of maximum 1
lim sup log pn ðXÞ  hðXÞ :
modulus and this eigenvalue is positive; in fact, n!1 n
for a primitive graph and a pair of states I, J, the
number of paths of length N from I to J grows like The inequality can be strict; for example, there
lNAðGÞ ¼ 2NhðXG Þ . For general graphs, one uses the are shift spaces (such as the direct product of the
Symbolic Dynamics 437

Morse shift with the full 2-shift) with positive Lanford 1970). As an example, the zeta function
entropy but no periodic points at all. of the golden mean shift is 1/(1  t  t2).
However, for irreducible SFT’s and sofic shifts, The discussion above applies equally well to
the entropy h(X) can be recovered from the sequence SFT’s since they are conjugate to edge shifts. For
pn(X). The key to understanding this is the fact that a sofic shift, the zeta function turns out to be a
for an edge shift X ¼ XA, pn(X) ¼ tr(An) and thus is rational function, i.e., quotient of two polyno-
the sum of the nth powers of the (non-zero) eigen- mials. This can be shown by analyzing properties
values of A; in the case that A is primitive, the largest of a right resolving presentation of the sofic shift.
eigenvalue lA strictly dominates the other eigen- From this it turns out that the zeta function of the
values, and thus for large n, even shift is (1  t)/(1  t  t2). The technique for
computing zeta functions of sofic shifts was
log pn ðXÞ ¼ log trðAn Þ~ ~ ðX Þ :
n log lA nh developed by Manning (1971) (actually, Manning
developed the technique to compute zeta func-
This shows that for a mixing SFT, the entropy tions of hyperbolic dynamical systems).
equals the growth rate of numbers of periodic So, for sofic shifts, all of the periodic point
points. In fact, this result applies to all SFT’s and information is determined by a finite collection
sofic shifts. of complex numbers, namely, the zeros and
poles of the zeta function.
Theorem 4 For a sofic shift X, Finally, we mention another simple invariant
obtained from the periodic points. The period,
1 per(X), of a shift space X is the gcd of lengths of
lim sup log pn ðXÞ ¼ hðXÞ :
n!1 n periodic points in X, i.e., the gcd of the set of
n such that pn(X) 6¼ 0. If X ¼ XG is an edge shift
and G is irreducible, then per(XG) ¼ per(G),
The limsup turns out be a limit in the case that which is defined to be the gcd of cycle lengths in
X is a mixing sofic shift. G and coincides with the gcd of the lengths of all
Most of what we have stated for pn(X) here cycles in G based at any given state. For an irre-
applies equally well to qn(X), the number of points ducible graph with period p and any state I, the
of least period n in X. This follows from the fact number of cycles of length N, a multiple of p, at
that “most” periodic points of period n have least I grows like lNAðGÞ ¼ 2NhðXG Þ . Also, an irreducible
period n. graph G is primitive if per(G) ¼ 1.
The periodic point information can be conve-
niently combined into a single invariant, known as
the zeta function. For a shift space X,
The Conjugacy Problem
1
pn ð X Þ n
zX ðtÞ ¼ exp t : The conjugacy problem for SFT’s and sofic shifts
n¼1
n is a major open problem. After much effort, it
remains unsolved today. Much of what we know
For an edge shift XA, one computes the zeta goes back to R. Williams (1973/74) in the 1970s.
function to be the reciprocal of a polynomial: One of Williams’ main results was that any
conjugacy can be decomposed into simple build-
1 1 ing blocks, as follows.
zXA ðtÞ ¼ ¼ ,
tr wA ðt1 Þ detðI  tAÞ Let A and B be nonnegative integral matrices,
with associated graphs G and H. An elementary
which is completely determined by the non-zero equivalence from A to B is a pair (R, S) of rectan-
eigenvalues (with multiplicity) of A (Bowen and gular nonnegative integral matrices satisfying
438 Symbolic Dynamics

A ¼ RS, B ¼ SR: ð1Þ conjugacies defined by elementary equivalences.

This can be interpreted in a way that shows that XA
In this case, we write (R, S) : A B. A strong and XB are conjugate if and only if we can pass
shift equivalence of lag ‘ from A to B is a sequence from G to H by a sequence of state splittings and
of ‘ elementary equivalences amalgamations, defined as follows.
Let G be a graph with states V and edges E .
ð R1 , S 1 Þ : A ¼ A 0 A1 , For each I  V , partition the outgoing edges from
mðI Þ
ð R2 , S 2 Þ : A1 A2 , I into disjoint nonempty sets E 1I , E 2I , . . . , E I .
Let P denote the resulting partition of E, and let P I
... ,
denote the partition P restricted to E I . The out-
ðR‘ , S‘ Þ : A‘1 A‘ ¼ B: split graph H formed from G using P has states I1,
I2, . . ., Im(I), where I ranges over the states in V ,
In this case, we write A B(lag ‘). We say that
and edges e j, where e is any edge in E and
A is strong shift equivalent to B (and write A B)
1  j  m(t(e)). If e  E goes from I to J, then
if there is a strong shift equivalence of some lag
e  E iI for some i, and we define the e j to have
from A to B.
initial state Ii and terminal state J j, The 2-block
Via the matrix Eq. (1), one creates a graph
code generated by F(ef ) ¼ e j, where f  E tjðeÞ,
K whose state set is the disjoint union of V ðGÞ
defines a conjugacy, called an out-splitting code,
and V ðHÞ; for each I  VG and J  VH, the graph
from XG to XH. Similarly, one defines an
has RIJ edges from I to J and SJI edges from J to I.
in-splitting code. Inverses of these conjugacies
The equation A ¼ RS allows one to associate each
are called out-amalgamation and
edge e of G with a unique path r(e)s(e) of length
in-amalgamation codes.
two running from V ðGÞ to V ðHÞ to V ðGÞ. Sim-
Figure 5 depicts an out-splitting. The graph
ilarly, the equation B ¼ SR allows one to associate
(a) on the left has three states I, J, K and the
each edge e of H with a path s(e)r(e) of length two
partition elements that define the splitting are
running from V ðH Þ to V ðGÞ to V ðH Þ. One can
E 1I ¼ fag, E 2I ¼ fb, cg,E 1J ¼ fd g,E 1K ¼ feg,E 2K ¼ f f g;
show that the 2-block sliding block code defined
the graph (b) is the resulting split graph.
by F(ef) ¼ s(e)r(f) defines a conjugacy from XA to
By interpreting state splitting and amalgam-
XB, and so whenever A B, XA ffi XB. Williams
ation in terms of adjacency matrices, one can
proved this and its converse:
show that such operations generate elementary
Theorem 5 R. Williams (1973/74) The edge equivalences. And one can decompose elemen-
shifts XA and XB are conjugate if and only if tary equivalences into splittings and amalgam-
A and B are strong shift equivalent. ations. It follows that:

In fact, Williams showed that any conjugacy Theorem 6 R. Williams (1973/74) Every
can be decomposed into a composition of conjugacy from one edge shift to another is the

Symbolic Dynamics,
Fig. 5 A state splitting
Symbolic Dynamics 439

composition of splitting codes and amalgamation equivalence is a shift equivalence of lag 1 and
codes. that shift equivalence is an equivalence relation.
It follows that:
This classification for edge shifts naturally
extends to SFT’s since every SFT is conjugate to Theorem 8 Williams (1973/74) Strong shift
an edge shift. It also extends to sofic shifts, and we equivalence implies shift equivalence. More pre-
describe this in the context of irreducible sofic cisely, if A B(lag ‘), then A~B(lag ‘).
shifts.
Recall that an irreducible sofic shift has a Recall from section “Entropy and Periodic
unique minimal right resolving presentation. Points” that for an edge shift XA, the set of
Any labeled graph can be completely described nonzero eigenvalues, with multiplicity, of
by a symbolic adjacency matrix, which records A determines the zeta function and hence this
the transitions (edges) in the underlying physical set is an invariant of conjugacy. Using the shift
graph, as well as the labels of the edges. Namely, equivalence equations, one can show more: the
the symbolic adjacency matrix is indexed by the entire Jordan form corresponding to the nonzero
states of the underlying graph and the (I, J)-entry eigenvalues is an invariant. This information
is the formal sum of the labels of edges from I to J. depends only on properties of the adjacency
It turns out that the notions of elementary equiv- matrix considered as a linear transformation
alence, and hence strong shift equivalence, can be (over R or Q). However, A is a nonnegative,
extended to more general categories, in particular integral matrix and both nonnegativity and inte-
to symbolic adjacency matrices. grality provide substantially more information.
One such invariant that follows from shift equiv-
Theorem 7 Krieger (1984), Nasu (1986) Let alence and makes use of integrality is the
X and Y be irreducible sofic shifts. Let A and Bowen-Franks group (Bowen and Franks
B be the symbolic adjacency matrices of the min- 1977), BF(A) ¼ Zr/Zr(I  A).
imal right-resolving presentations of X and Y, Until recently all of the information contained in
respectively. Then X and Y are conjugate if and known conjugacy invariants, such as those above,
only if A and B are strong shift equivalent. was subsumed in shift equivalence. And Kim and
Roush showed that shift equivalence is decidable
The classification, provided by these results, (Kim and Roush 1979, 1988), meaning that there is
would be of limited use if the story ended here. a finite decision procedure via a Turing machine
Fortunately, Williams showed that strong shift that decides whether two given edge shifts, and
equivalence yields a strong, delicate, and some- therefore two given SFT’s, are conjugate (they
what computable necessary condition for also showed that a notion of shift equivalence for
conjugacy. sofic shifts, formulated by Boyle and Krieger
Let A and B be nonnegative integral matrices (1986), is decidable (Kim and Roush 1990)). So, a
and ‘  1. A shift equivalence of lag ‘ is a pair central focus of the subject was the question: is shift
(R, S) of rectangular nonnegative integral matri- equivalence a complete invariant of conjugacy?
ces satisfying the shift equivalence equations The answer turns out to be No, as proven by Kim
and Roush (1999); see also the survey article Wag-
AR ¼ RB, SA ¼ BS, A‘ ¼ RS, B‘ ¼ SR: oner (1992). However, it is a complete invariant of a
weaker form of conjugacy; we say that XA and XB
We denote this situation by (R, S) : A~B(lag l). are eventually conjugate if all sufficiently large
We say that A is shift equivalent to B, written A~B, powers are conjugate. It is not hard to show that if
if there is a shift equivalence from A to B of some A~B, then XA and XB are eventually conjugate, and
lag. It is not hard to see that an elementary the converse is true as well:
440 Symbolic Dynamics

Theorem 9 Kim and Roush (1979), Williams In many areas of mathematics, objects are stud-
(1973/74) Edge shifts XA and XB are eventually ied by means of their symmetries. This holds true
conjugate if and only if A and B are shift in symbolic dynamics, where symmetries are
equivalent. expressed by automorphisms. An automorphism
of a shift space X is a conjugacy from X to itself.
Also the Kim-Roush counterexamples and The set of all automorphisms of a shift space X is a
subsequent work do not bear on a special case of group under composition and is naturally called
the conjugacy problem: if A is shift equivalent to the automorphism group, denoted aut(X).
the 1 1 matrix [n], is XA conjugate to the full The goals are to understand aut(X) as a group
n-shift? This question, known as the little shift (What kinds of subgroups does it contain? How
equivalence problem, is particularly intriguing “big” is it?) and how it acts on X, e.g., given shift-
because in this case the condition A~[n] is simply invariant subsets U, V, such as finite sets of peri-
the statement that A has exactly one non-zero odic points, when is there an automorphism of
eigenvalue, namely, n. X that maps U to V?. One might hope that the
Shift equivalence can be characterized in automorphism group would shed new light on the
another way that has turned out to be very useful. conjugacy problem for SFT’s. Indeed, tools devel-
Let A be an r r integral matrix. Let RA denote the oped to study the automorphism group eventually
real eventual range of A, i.e., RA ¼ RrAr. The paved the way for Kim and Roush to find exam-
dimension group of A is defined: ples of shift equivalent matrices that are not strong
shift equivalent. On the other hand, the automor-
DA ¼ v  RA : vAk  Zr for some k  0 : phism group cannot tell the entire story. For
instance, aut(XA) and autðXA⊤ Þ are isomorphic,
The dimension group automorphism dA of A is since any automorphism read backwards can be
the restriction of A to DA, so that dA(v) ¼ vA for viewed as an automorphism of the transposed
v  DA. The dimension pair of A is (DA, dA). shift, yet XA and XA⊤ may fail to be conjugate
If A is also nonnegative, then we define the (for an example due to Kollmer, see p. 81 of Parry
dimension semigroup of A to be and Tuncel (1982)). It is not even known if the
automorphism groups of the full 2-shift and the
Dþ þ r full 3-shift are isomorphic.
A ¼ v  RA : vA  ðZ Þ for some k  0 :
k
A good deal of our understanding of the action
The dimension triple of A is DA , Dþ of the automorphism group on an SFT comes from
A , dA .
understanding its induced representation as an
It can be shown that the dimension triple
action of the dimension group; this action is
completely characterizes shift equivalence, i.e.,
known as the dimension representation. For a
two nonnegative integral matrices are shift equiv-
much more thorough exposition on aut(X), we
alent if and only if their dimension groups are
refer the reader to Wagoner (1992, 2004).
isomorphic by an isomorphism that preserves the
dimension semigroup and intertwines the dimen-
sion group automorphisms. Also, by associating
equivalence classes of certain subsets of the shift Other Coding Problems
space XA to elements of DþA , one can interpret the
dimension triple in terms of the action of the shift The difficulties encountered in attempts to solve
map on XA (Boyle et al. 1987; Krieger 1980a). the conjugacy problem motivated the formulation
And the dimension triple arises prominently in the and study of weaker, but meaningful, notions of
study of the automorphism group of an SFT, equivalence. For instance, we might say that two
which we now briefly describe. The dimension shift spaces are equivalent if one can be invertibly
group for SFT’s was developed by Krieger encoded to the other by some kind of “finite-state
(1980a, b). machine.” A precise version of this is as follows.
Symbolic Dynamics 441

Shift spaces X and Y are finitely equivalent if almost invertible factor codes fX : W ! X,
there is an SFT W together with finite-to-one fY : W ! Y. We call (W, fX, fY) an almost
factor codes fX : W ! X and fY : W ! Y. We conjugacy between X and Y.
call W a common extension, and fX, fY the legs. For irreducible sofic shifts, it can be shown that
Here, by “finite-to-one” we mean merely that any almost invertible factor code is finite-to-one,
each point has a finite number of inverse images. and so almost conjugacy implies finite equiva-
It can be shown that any finite-to-one factor code lence. Thus, entropy is again an invariant, and
from one shift space to another must preserve together with a second very mild invariant, it is
entropy, and so entropy is an invariant of finite complete:
equivalence. For irreducible sofic shifts, entropy
is a complete invariant: Theorem 11 Adler-Marcus (1979) Let X and
Y be irreducible sofic shifts with minimal right
Theorem 10 Parry (1977) Two irreducible sofic resolving presentations (G, ℒ) and (H, M).
shifts are finitely equivalent if and only if they Then X and Y are almost conjugate if and only if
have the same entropy. h(X) ¼ h(Y) and per(G) ¼ per(H).

Note that from this result and the fact that In particular, if X and Y are mixing, then per(-
finite-to-one codes between general shift spaces G) ¼ 1 ¼ per(H) and entropy itself is a complete
preserve entropy, for irreducible sofic shifts, we invariant.
could have just as well-defined finite equivalence Thus, with an almost conjugacy, one can
with the common extension W merely being a invertibly encode most sequences in X to those
shift space. However, if W is an SFT, we get a of Y without the need for auxiliary state informa-
more concrete coding interpretation as follows. tion. Moreover, if X and Y are almost conjugate,
First, we recode W to an edge shift XG and recode then there is an almost conjugacy of X and Y in
the legs, fX ¼ (FX)1 and fY ¼ (FY)1, to one- which one leg is right-resolving and the other leg
block codes, and (with a bit more argument) we is left-resolving (and the common extension is
can assume that G is irreducible. In this set-up, the irreducible (Adler and Marcus 1979)). This gives
finite-to-one condition translates to the so-called an even more concrete interpretation to the
“no-diamond” condition, which means that for encoding.
any given pair of states I, J and finite sequence The proofs of Theorems 10 and 11 are actually
w, there is at most one path from I to J with label quite constructive. For illustration, we consider a
w (Coven and Paul 1975, 1977). Since, for any very special, but historically important, case.
fixed state I, the number of cycles of length n, a Let G be a graph with constant out-degree n.
multiple of p ¼ per(G), at I grows like 2nh(W), we A road coloring F is a labeling of G such that at
have, in this set-up a means to invertibly encode a each state of G, each symbol 0, . . ., n  1 appears
“large” set of allowed blocks in X to allowed exactly once as the label of an outgoing edge. An
blocks in Y: namely, fix state I, and a large n, n-ary word w is synchronizing if all paths that are
which is a multiple of p; for any cycle g of length labeled w end at the same state. Figure 6 gives
n at state I encode the FX -label of g to the FY examples of road-colorings, with n ¼ 2.
-label of g. For a road-coloring, a binary word may be
For encoding and decoding, one can dispense viewed as a sequence of instructions given to
with state information if the legs are “almost drivers starting at each of the states.
invertible.” A factor code f is almost invertible A synchronizing word is a word that drives every-
if it is one-to-one on sequences that are typical in body to the same state. For instance, in Fig. 6a, the
the following sense: x is typical if every allowed word 11 drives everybody to the state in the lower-
block appears infinitely often in x both to the left left corner. But Fig. 6b does not have a synchro-
and the right. We then say that shift spaces X and nizing word because whenever a driver takes a “0”
Y are almost conjugate if there is an SFT W and road he stays where he is and whenever a driver
442 Symbolic Dynamics

Symbolic Dynamics, Fig. 6 Some road-colorings

takes a “1” road he rotates by 120 . The road- Theorem 12 Road Theorem (Trachtman 2007)
coloring in Fig. 6c is essentially the only road- If G is a finite directed primitive graph with con-
coloring of its underlying graph, and there is no stant out-degree n, there a road-coloring of
synchronizing word because drivers must always G which has a synchronizing word.
oscillate between the two states.
Now, let X be the full n-shift and Y be an Trachtman’s approach relies heavily on earlier
irreducible SFT with entropy logn. Suppose that work of Friedman (1990) and Kari (2001).
we could find a presentation (G, ℒ) of Y with The primitivity assumption above is close to
G having constant out-degree n and ℒ1 finite- necessary. Clearly some kind of connectivity is
to-one. Then, define F to be any road coloring of required and in the presence of irreducibility,
G. Then F1 would be a finite-to-one (in fact, right primitivity would be necessary since otherwise
resolving!) factor code from XG to X. And we there would be a “phase” introduced in the graph
would obtain a finite equivalence with X ¼ XG, that would never allow a word to synchronize, as
fY ¼ ℒ1 and fX ¼ F1. If, moreover, we could in Fig. 6c.
choose ℒ and F such that fY and fX are almost So far, we have focused on equivalences
invertible, then we would have an almost between symbolic systems. There has also been
conjugacy. considerable attention paid to problems of embed-
It turned that this could be arranged for fY via a ding one system into another and factoring one
construction related to state splitting (Adler et al. onto another.
1977). And if Y were mixing, then G could be One of the most striking results of this type is
chosen to be primitive and fY almost invertible. In the Krieger embedding theorem. It is not hard to
this setting, fX ¼ F1 would be almost invertible show that any proper subshift of an irreducible
iff F has a synchronizing word; the sufficiency SFT must have strictly smaller entropy. Thus, a
follows from the fact that every bi-infinite binary necessary condition for a proper embedding of a
sequence which contains w infinitely often to the shift space into an irreducible SFT is that it have
left would be the label of exactly one bi-infinite strictly smaller entropy. This condition, together
sequence of edges (to see this, use the synchro- with a trivially necessary condition on periodic
nizing and road-coloring properties). points, turns out to be sufficient. Recall that qn(X)
The construction of such a labeling F became denotes the number of points of least period
known as the Road Problem, which remained n in X.
open for 30 years. In the meantime, a weaker
version of the road problem was solved and Theorem 13 Embedding Theorem (Krieger
applied to yield this special case of Theorem 11 1982) Let X and Y be irreducible shifts of finite
(see Adler et al. 1977). Nevertheless, the problem type. Then there is a proper embedding of X into
remained an important problem in graph/automata Y if and only if h(X) < h(Y) and for each n  1,
theory and was only recently solved: qn(X)  qn(Y).
Symbolic Dynamics 443

In fact, Krieger’s theorem shows that these of the Perron-Frobenius theorem for primitive
conditions are necessary and sufficient for a matrices; and the Net Trace condition assures
proper embedding of any shift space into a mixing that the number of periodic points of least period
shift space. The analogous problems for embed- n would be nonnegative.
ding into irreducible or mixing sofic shifts are still We now turn from embeddings to factors. One
open, though there are partial results (Boyle special case, which is somewhat related to the
1983). Road Problem above and also important for data
Using the embedding theorem and other tools recording applications (Section “Coding for Data
from symbolic dynamics, Boyle and Handelman Recording Channels”) is:
(1991) obtained a stunning application to linear
algebra: namely, a complete characterization of Theorem 15 Adler et al. (1983a), Marcus
the non-zero spectra of primitive matrices over (1979)) An SFT X factors onto the full n-shift
R. In fact, they obtained characterizations of iff h(X)  log (n).
non-zero spectra for primitive matrices over
many other subrings of R. While they did not While this special case treats both the equal
obtain a complete characterization over Z, they entropy case (h(X) ¼ log (n)) and unequal entropy
formulated a conjecture for Z and obtained many case (h(X) > log (n)), in general, the factor prob-
partial results towards that conjecture, which was lem naturally divides into two cases: lower
later proven using other tools. The result, stated entropy factors and equal entropy factors. In either
below, shows that three simple necessary condi- case, a trivial necessary condition for Y to be a
tions on a set of nonzero complex numbers are factor of X is that whenever qn(X) 6¼ 0, there exists
actually sufficient. In order to state these condi- a d/n such that qd(Y) 6¼ 0. This condition is
tions, we need the following notation: denoted P(X) ↘ P(Y). Building on ideas from
Let L ¼ {l1, . . ., lk} be a list of Krieger’s embedding theorem, this necessary con-
nonzero complex numbers (with multiplicity). dition was shown to be sufficient.
Let f L ðtÞ ¼ ki¼1 ðt  li Þ, and trn ðLÞ ¼
k Theorem 16 Lower Entropy Factor Theorem
d=n mðn=d Þ
k
i¼1 li , with m being the Mobius
(Boyle 1983) Let X and Y be irreducible SFT’s
Inversion function.
with h(X) > h(Y). Then there is a factor code from
X to Y if and only if P(X) ↘ P(Y).
1. Integrality Condition: fL(t) is a monic polyno-
mial (with integer coefficients).
As with the embedding theorem, the lower
2. Perron Condition: There is a positive entry in
entropy factors problem for irreducible sofic shifts
L, occurring just once, that strictly dominates
is still open.
in absolute value all other entries. We denote
The equal entropy factors problem for SFT’s is
this entry by lL.
quite different. Clearly, P(X) ↘ P(Y) is a neces-
3. Net Trace Condition: trn(L)  0 for all n  1.
sary condition. A second necessary condition
involves the dimension group and is simplest to
Theorem 14 Kim-Ormes-Roush (2000) Let L be state in the case of mixing edge shifts XA and XB.
a list of nonzero complex numbers satisfying the We say that a subgroup D of the dimension
Integrality, Perron, and Net Trace Conditions. group DA is pure if whenever an integer multiple
Then there is a primitive integral matrix A for of an element v  DA is in D, then so is v;
which L is the non-zero spectrum of A. intuitively D does not have any “rational holes”
in DA. The condition is that there is a pure dA
These conditions are all indeed necessary for L -invariant subgroup D of DA such that (DB, dB) is
to be the non-zero spectrum of a primitive integral a quotient of (D, dA|D).
matrix. The integrality condition is that L forms a In the equal entropy case, this condition and
complete set of algebraic conjugates; the Perron the trivial periodic point condition, P(X) ↘ P(Y),
condition states that L must satisfy the conditions subsume all known necessary conditions for the
444 Symbolic Dynamics

existence of a factor code from one mixing edge

shift to another. It is not known if these two
conditions are sufficient. For references on this
problem, see (Ashley 1991; Boyle et al. 1987;
Kitchens et al. 1991).
Symbolic Dynamics, Fig. 7 X(1, 3)

Coding for Data Recording Channels

In magnetic recording, within any given clock

cell, a “1” is represented as a change in magnetic
polarity, while a “0” is represented as an absence
of such a change. Two successive 1’s (separated
by some number, m  0, of 0’s) are read as a
Symbolic Dynamics, Fig. 8 Encoder
voltage peak followed by a voltage trough
(or vice versa). If the peak and trough occur too
close together, intersymbol interference can depicted in Fig. 8. It maps arbitrary binary data
occur: the amplitudes of the peak and trough are sequences, grouped into blocks of length p (called
degraded, and the positions at which they occur p-blocks), into constrained sequences, grouped
are distorted. In order to control intersymbol inter- into blocks of length q (called q-blocks). The
ference, it is desirable that 1’s not be too close encoded q-block is a function of the p-block as
together, or equivalently that runs of 0’s not be too well as an internal state. When concatenated
short. On the other hand, for timing control, it is together, the sequence of encoded q-blocks must
desirable that runs of 0’s not be too long; this is a satisfy the given constraint. Also, the encoded
consequence of the fact that only 1’s are observed: sequences should be decodable, meaning that
the length of a run of 0’s is inferred by connection given the initial encoder state, a string of p-blocks
to a clock via a feedback loop, and a long run of can be recovered from its encoded string of
0’s could cause the clock to drift more than one q-blocks, possibly allowing a fixed delay in time.
clock cell. If the constrained sequences satisfy the con-
This gives rise to run-length-limited shift straints of a sofic shift X, we say that such a code is
spaces, X(d, k), where runs of 0’s are constrained a rate p:q finite-state code into X. In terms of
to be bounded below by some positive integer symbolic dynamics, such a code consists of an
d and bounded above by some positive k  d. edge shift XG, a right resolving factor code f1
More precisely, X(d, k) is defined by the con- from XG onto the full 2p-shift and a right closing
straints that 1’s occur infinitely often in each sliding block code f2 into Xq (the right closing
direction, and there are at least d 0’s, but no condition expresses the decodability condition).
more than k 0’s, between successive 1’s. Note In most applications, it is important that a
that X(d, k) is an SFT with forbidden list stronger decoding condition be imposed. Namely,
F ¼ {0k + 1, 10i1, 0  i < d}. Figure 7 depicts a a sliding block decodable rate p:q finite-state code
labeled graph presentation of X(1, 3). The SFT’s into X consists of a finite-state code given by
X(1, 3), X(2, 7), and X(2, 10) have been used in (XG, f1, f2) and a sliding block code c : Xq !
floppy disks, hard disks, and the compact audio X2p such that c ∘ f2 ¼ f1. This means that the
disk, respectively. decoded p-block depends only upon the local
Now in order to record completely arbitrary context of the received q-block – that is, decoding
information, we need to build an encoder which is accomplished by applying a time-invariant
encodes arbitrary binary sequences into function to a window consisting of a bounded
sequences that satisfy a given constraint (such as amount of memory and/or anticipation, but other-
X(d, k)). The encoder is a finite-state machine, as wise is state-independent (see Fig. 9, which
Symbolic Dynamics 445

Theorems 17 and 18 are stated only for rate 1:

1 codes but in the context of arbitrary finite alpha-
bets (e.g., see Chap. 5 in (Lind and Marcus
1995)), this easily extends to the general p:
q case by passing to powers.
The ACH paper was the beginning of a rigor-
ous theory of constrained coding. Theorem 18 has
been extended to a large class of sofic shifts, and
bounds have been established on such figures of
merit as number of encoder states as well as the
size of the decoding window of the sliding block
decoder. While the state-splitting algorithm does
construct codes with “relatively small” decoding
Symbolic Dynamics, Fig. 9 Sliding block decoder windows, it is not yet understood how to construct
codes with the smallest such windows. This is of
substantial engineering interest since the smaller
depicts the situation where the memory is 1 and the decoding window, the smaller the error prop-
the anticipation is 2). The point is that whenever agation. There is now a substantial literature on
the window of the decoder passes beyond a raw the construction of these types of codes, including
channel error, that error cannot possibly affect the state-splitting algorithm as well as many other
future decoding; thus, sliding block decoders con- methods of encoder/decoder design and a wealth
trol error propagation. of examples that go well beyond the run length
Symbolic dynamics has played an important limited constraints. See, for example, the exposi-
role in providing a framework for constructing tions (Béal 1993; Immink 2004a; Marcus et al.
such codes as well as for establishing bounds on 1998) as well as the papers (Ashley 1988, 1996;
various figures of merit for such codes. Specifi- Béal 1990, 2003; Cidecyian et al. 2001; Franaszek
cally, by modifying the constructions used in the 1968, 1982, 1989; Hollmann 1995; Karabed et al.
proofs of Theorems 10 and 11, in the early 1980s, 1999).
Adler, Coppersmith, and Hassner (ACH)
established the following results:
Connections with Information Theory
Theorem 17 Finite-State Coding Theorem and Ergodic Theory
(Adler et al. 1983a) Let X be a sofic shift and p,
q be positive integers. Then there is a rate p:q The concept of entropy was developed by Shan-
finite-state code into X if and only if p/q  h(X). non (1948) in information theory in the 1940s and
was adapted to ergodic theory in the 1950s and to
Theorem 18 Sliding Block Decoding Theorem dynamical systems, and in particular symbolic
(Adler et al. 1983a) Let X be an SFT and p, q be dynamics, in the 1960s. Shannon focused on
positive integers. Then there is a rate p:q sliding- entropy for random variables and finite sequences
block decodable finite-state code into X if and of random variables, but the concept naturally
only if p/q  h(X). extends to stationary stochastic processes, in par-
ticular stationary Markov chains. Roughly speak-
We have described all of this in the context of ing, for a stationary process m, the entropy h(m) is
the binary alphabet for data sequences. In fact, it the asymptotic growth rate of the number of allo-
works just as well for any finite alphabet, and the wed sequences, weighted by the joint stationary
method used to prove Theorem 18 solved, at the probabilities of m (for background on entropy and
same time, a special case of the factor problem for information theory see Cover and Thomas
SFT’s: namely, Theorem 15 above. Sometimes, (1991)).
446 Symbolic Dynamics

A Markov chain on a graph G is defined by 1. Any stationary process m on G satisfies

assigning transition probabilities to the edges. Let h(m)  log l.
P denote the stochastic matrix indexed by states of 2. The Markov chain mG is the unique stationary
G, with PIJ equal to the sum of the transition process on G such that h(mG) ¼ log l.
probabilities of all edges from I to J. This, together
with a stationary vector for P, completely defines The construction of mG is effectively due to
the (joint distributions of the) Markov chain. Shannon (1948) and uniqueness is due to Parry
So, a graph itself can be viewed as specifying (1964). The unique entropy-maximizing station-
only which transitions are possible. For that rea- ary process on irreducible edge shifts naturally
son, one could view G as an “intrinsic Markov extends to irreducible SFT’s and irreducible sofic
chain,” and Parry (1964) used this terminology shifts (this process will be M-step Markov for an
when he introduced SFT’s based on graphs and M-step SFT).
matrices. Many results in symbolic dynamics were orig-
More generally, a stationary process on a shift inally proved using mG. One example is the fact
space X is any stationary process which assigns that any factor code from one irreducible SFT to
positive probability only to allowed blocks in X. another of the same entropy must be finite-to-one
For an irreducible graph G with strictly posi- (Coven and Paul 1974). This result, and many
tive transition probabilities on all edges, by others, were later proven using methods that rely
Perron-Frobenius theory, P will always have a only on the basic combinatorial structure of G and
unique stationary vector. And there is a particular XG. Nevertheless, mG provides much motivation
such Markov chain mG on G that in some sense and insight into symbolic dynamics problems, and
distributes probabilities on paths as uniformly as it illustrates a connection with ergodic theory,
possible. This Markov chain is the most “random” which we now discuss.
possible stationary process (not just among Mar- For background on ergodic theory (such as the
kov chains) on XG in the sense that it has maximal concepts of measure-preserving transformations,
entropy; moreover, its entropy coincides with the homomorphisms, isomorphisms, ergodicity,
topological entropy of XG. mixing, and measure-theoretic entropy), we refer
The Markov chain mG is defined as follows: let the reader to Petersen (1989), Rudolph (1990),
l denote the largest eigenvalue of A(G) and w, and Walters (1982). Observe that a stationary
v denote corresponding left, right eigenvectors, process on a shift space X can be viewed as a
normalized such that w  v ¼ 1; for any path g of measure-preserving transformation: the transfor-
length n from state I to state J, we define the mation is the shift mapping and the measure on
stationary probability of g: “cylinder sets,” consisting of x  X with pre-
scribed coordinate values xi ¼ ai, . . ., xj ¼ aj, is
w I vJ defined as the probability of the word ai. . .aj; the
mG ð g Þ ¼ :
ln stationarity of the process translates directly into
preservation of the measure. The symbolic
It is clear from the formula that this distribution
dynamical concepts of irreducibility and mixing
is fairly uniform, since all paths with the same
correspond naturally to the concepts of ergodicity
initial state, terminal state, and length have the
and (measure-theoretic) mixing in ergodic theory.
same stationary probability.
It is well-known (Petersen 1989; Walters 1982)
This defines a joint stationary distribution on
that the measure-preserving transformation
allowed blocks of XG, and it is not hard to show
(MPT) defined by a stationary Markov chain m
that it is consistent and Markov, given by
on an irreducible (resp., primitive) graph G is
assigning transition probability vJ/(vIl) to each
ergodic (resp., mixing) if m assigns strictly posi-
edge from I to J. To summarize:
tive conditional probabilities to all edges of G.
Now, suppose that G and H are irreducible
Theorem 19 Let G be an irreducible graph.
graphs and f : XG ! XH is a factor code. For a
Symbolic Dynamics 447

stationary measure m on G, we define a stationary block x[n(x),n(x)], where the function n(x) has
measure n ¼ f(m) on XH by transporting m to XH: finite expectation (Adler and Marcus 1979). In
for a measurable set A in XH, define particular, by Theorem 11, whenever XG and XH
are mixing edge shifts with the same entropy, the
nðAÞ ¼ m f1 ðAÞ : measure-preserving transformations defined by
mG and mH are isomorphic via an isomorphism
Then f defines a measure-preserving homo- with finite expected coding length.
morphism from the MPT defined by m to the The notions of conjugacy, finite equivalence,
MPT defined by n. Since measure-preserving almost conjugacy, embedding, factor code, and so
homomorphisms between MPT’s cannot reduce on can all be generalized to the context of station-
measure-theoretic entropy, we have h(n)  h(m). ary measures, in particular to stationary Markov
Suppose that now f is actually a conjugacy. chains. For instance, a conjugacy between two
Then it defines a measure-preserving isomor- stationary measures is a map that is simulta-
phism, and so h(m) ¼ h(n). If m ¼ mG, then by neously a conjugacy of the underlying shift spaces
uniqueness, we have n ¼ mH. Thus, f defines a and an isomorphism of the associated measure-
measure-theoretic isomorphism between the MPT preserving transformations. Many results in sym-
defined by mG and the MPT defined by mH. In fact, bolic dynamics have been generalized to the con-
this holds whenever f is merely an almost invert- text of stationary Markov chains. There is a
ible factor code. This establishes the following substantial literature on this, in particular on
result: finitary isomorphisms with finite expected coding
time, e.g., (Krieger 1983; Marcus and Tuncel
Theorem 20 Let G, H be irreducible graphs, and 1990; Mouat and Tuncel 2002; Parry 1979;
let mG, mH be the stationary Markov chains of Schmidt 1984). The expositions (Parry 1991;
maximal entropy on G, H. If XG, XH are almost Parry and Tuncel 1982) give a nice introduction
conjugate (in particular, if they are conjugate), to the subject of strong finitary codings between
then the measure-preserving transformations stationary Markov chains. See also the research
defined by mG and mH are isomorphic. papers (Gomez 2003; Marcus and Tuncel 1991,
1993; Parry and Schmidt 1984; Parry and Tuncel
Hence, conjugacies and almost conjugacies 1981; Tuncel 1981, 1983).
yield isomorphisms between measure-preserving
transformations defined by stationary Markov
chains of maximal entropy. In fact, the isomor- Higher Dimensional Shift Spaces
phisms obtained in this way have some very desir-
able properties compared to the run-of-the-mill In this section, we introduce higher dimensional
isomorphism. For instance, an isomorphism f shift spaces. For a more thorough introduction, we
between stationary processes typically has an infi- refer the reader to Lind (2004). For the related
nite window; i.e., to know f(x)0, you typically subject of tiling systems, see Robinson (2004),
need to know all of x, not just a central block Radin (1996), and Mozes (1989, 1992).
x[n, n] (these are the kinds of isomorphisms that The d-dimensional full A -shift is defined to be
Zd
appear in general ergodic theory and in particular A . Ordinarily, A is a finite alphabet, and here we
in Ornstein’s celebrated isomorphism theory restrict ourselves to this case. An element x of the
(Ornstein 1970)). In contrast, by definition, a full shift may be regarded as a function x : Zd ! A,
conjugacy always has a finite window of uniform or, more informally, as a “configuration” of alphabet
size. It turns out that an isomorphism obtained choices at the sites of the integer lattice Zd.
d
from an almost conjugacy, as well as its inverse, For x  A Z and F  Zd, let xF denote the
has finite expected coding length in the sense that restriction of x to F. The usual metric on the one-
to know f(x)0, you need to know only a central dimensional full shift naturally generalizes to a
448 Symbolic Dynamics

metric on A Z given by r(x, y) ¼ 2k, where k is

d
translate of F, belongs to ℒ. Of course, any pattern
the largest integer such that x½k,kd ¼ y½k,kd (with which occurs in X is ℒ-admissible. But an
the usual conventions when x ¼ y and x0 6¼ y0). In ℒ-admissible pattern need not occur in X.
analogy with one dimension, according to this The analogue of vertex shift (or 1-step shift of
definition, two points are “close” if they agree on finite type) in higher dimensions is defined by a
a large cube [k, k]d. collection of d transition matrices A1, . . ., Ad all
We define higher dimensional shift spaces, indexed by the same set of symbols A ¼
with the following terminology. A shape is a finite f1, . . . , mg. We set
subset F of Zd, and a pattern f on a shape F is a
function f : F ! A . We say that X is a d-dimen- OðA1 , . . . , Ad Þ
sional shift space (or d-dimensional shift) if it can d
¼ x  f1, . . . , mgZ : Ai ðxn , xnþei Þ ¼ 1 for all n, i ,
be represented by a list F (finite or infinite) of
“forbidden” patterns
where ei is as usual the ith standard basis vector and
d Ai(a, b) denotes the (a, b)-entry of Ai. Such a shift
X ¼ XF ¼ fx  A Z : sn ðxÞF2
=F space is called a matrix subshift. When d ¼ 2, this
for all n  Zd and all shapes Fg : amounts to a pair of transition matrices A1 and A2
with identical vertex sets. The matrix A1 controls
Just as in one dimension, we can equivalently transitions in the horizontal direction and the matrix
define a shift space to be a closed (with respect to the A2 controls transitions in the vertical direction. Note
d
metric r) translation-invariant subset of A Z . Here that any matrix subshift is a shift of finite type and,
“translation-invariance” means that sn(X) ¼ X for in particular, can be specified by a list ℒ of patterns
all n  Zd, where sn is the translation in direction on the unit cube F ¼ {(a1, . . ., an) : ai  {0, 1}};
n defined by (sn(x))m ¼ xm+n. specifically, OðA1 , . . . , An Þ ¼ Xℒc where ℒ is the
We say that a pattern f on a shape F occurs in a set of all patterns f : F ! {1, . . ., m} such that if n,
shift space X if there is an x  X such that xF ¼ f. n + ei  F, then Ai(f(n), f(n + ei)) ¼ 1. When we
Hence the analogue of the language of a shift speak of admissible patterns for a matrix subshift,
space is the set of all occurring patterns. we mean ℒ-admissible patterns with this particular
A d-dimensional shift of finite type X is a subset ℒ. Just as in one dimension, we can recode any shift
d
of A Z defined by a finite list F of forbidden of finite type to a matrix subshift.
patterns. Just as in one dimension, a Higher dimensional SFT’s can behave very
d-dimensional shift of finite type X can also be differently from one dimension. For example,
defined by specifying allowed patterns instead of there is a simple method to determine if a one-
forbidden patterns, and there is no loss of gener- dimensional edge shift, and therefore, a one-
ality in requiring the shapes of the patterns to be dimensional shift of finite type, is nonempty, and
the same. Thus, we can specify a finite list ℒ of there is an algorithm to tell, for a given finite list
patterns on a fixed shape F, and set ℒ, whether a given block occurs in X ¼ Xℒc . The
corresponding problems in higher dimensions,
X ¼ Xℒc called the nonemptiness problem and the exten-
d
sion problem, turn out to be undecidable (Berger
¼ x  A Z : for all n  Zd , sn ðxÞF  ℒ : 1966; Robinson 1971); see also Kitchens and
Schmidt (1988). Even for two-dimensional matrix
In fact, there is no loss in generality in assum- subshifts X, these decision problems are
ing that F is a d-dimensional cube F ¼ [0, k]d. undecidable. On the other hand, there are some
Given a finite list ℒ of patterns on a shape F, special classes where these problems are decid-
we say that a pattern f0 on a shape F0 is able. This class includes any two-dimensional
ℒ-admissible (in the shift of finite type X ¼ Xℒc ) matrix subshift such that A1 commutes with A2
if each of its sub-patterns, whose shape is a and A⊤2 . For this class, any admissible pattern on a
Symbolic Dynamics 449

cube must occur, and so the nonemptiness and 1 1

extension problems are decidable; see Markley A1 ¼ A 2 ¼
1 0
and Paul (1981a, b).
A point x in a d-dimensional shift X is peri- an explicit formula for the entropy is not known.
odic if its orbit {sn(x) : n  Zd} is finite. However, there are good numerical approxima-
Observe that this reduces to the usual notion of tions to the entropy of some matrix subshifts
periodic point in one dimension. Now an ordi- (e.g., Calkin and Wilf 1998; Nagy and Zeger
nary (one-dimensional) nonempty shift of finite 2000). And recently, the set of numbers that can
type is conjugate to an edge shift XG, where occur as entropies of SFT’s in higher dimensions
G has at least one cycle. Hence a one- has been characterized (Hochman 2009; Hochman
dimensional shift of finite type is nonempty if and Meyerovitch 2007); this characterization turns
and only if it has a periodic point. This turns out out to be remarkably different from the analogous
to be false in higher dimensions (Berger 1966; characterization in one dimension (Lind 1984).
Robinson 1971), and this fact is intimately One of the few two-dimensional SFT’s for
related to the undecidability results mentioned which entropy has been computed is the domino
above. While one can formulate a notion of zeta tiling system (see (Kastelyn 1961), Chap. 5 in
function for keeping track of numbers of peri- (Schmidt 1990)), which consists of all possible
odic points, the zeta function is hard to compute, tilings of the plane using the 1 2 and 2 1
even for very special and explicit matrix sub- dominoes. This can be translated into an SFT with
shifts, and, even in this setting, it is not a rational four symbols L, R, T, B, subject to the constraints:
function (Lind 1996).
In higher dimensions, the entropy of a shift is xi,j ¼ L ) xiþ1,j ¼ R,
defined as the asymptotic growth rate of the num-
ber of occurring patterns in arbitrarily large cubes. xi,j ¼ R ) xi1,j ¼ L,
In particular, for two-dimensional shifts it is
defined by xi,j ¼ T ) xi,j1 ¼ B,

1 xi,j ¼ B ) xi,jþ1 ¼ T:
hðXÞ ¼ lim log j X½0,n1 ½0,n1 j,
n!1 n2

The entropy of this SFT is given by a remark-

where X[0,n1] [0,n1] denotes the set of occur-
able integral formula:
ring patterns on the square
1 1
½0, n  1 ½0, n  1 1
ð4  2 cos ð2psÞ  2 cos ð2pt ÞÞdsdt
4 0 0
that occur in X. Recall from section “Entropy and
Periodic Points” that it is easy to compute the Further work along these lines can be found in
entropy of a (one-dimensional) shift of finite Kenyon (2008) and Schwartz and Bruck (2008).
type using linear algebra. But in higher dimen- One can formulate notions of irreducibility and
sions, there is no analogous formula and, in fact, mixing for higher dimensional shift spaces. It
other than the group shifts mentioned below, the turns out that for SFT’s there are several notions
entropies of only a very few higher dimensional of mixing that all coincide in one dimension, but
shifts of finite type have been computed explicitly. are vastly different in higher dimensions. For
Even for the two-dimensional “golden mean” instance, a higher dimensional shift space is
matrix subshift defined by the horizontal and ver- strongly irreducible if there is an integer R > 0
tical transition matrices such that for any two shapes F, F0 of distance at
450 Symbolic Dynamics

least R, any occurring configurations on F and F0 no points that are periodic in any single direction.
can be combined to form an occurring configura- And some other results on entropy of proper sub-
tion on F [ F0 (Burton and Steif 1994; Ward shifts of strongly irreducible SFT’s carry over
1994). For one-dimensional SFT’s, this is equiv- from one dimension to higher dimensions
alent to mixing, but much stronger than mixing for (Pavlov 2011; Quas and Trow 2000). But in
two-dimensional SFT’s. many cases where versions of the result carry
Just as in one dimension, we have the notion of over, the proofs are much different from those in
sliding block code for higher dimensional shifts. one dimension.
For finite alphabets A, ℬ , a cube F Zd and a Measures of maximal entropy for two-
function F : A F ! ℬ, the mapping f ¼ F1 : dimensional SFT’s behave very differently from
d d
A Z ! ℬZ defined by the one-dimensional case. For instance, even with
very strong mixing properties, such as strong irre-
F1 ðxÞn ¼ FðxnþF Þ ducibility, there can be more than one measure of
maximal entropy (Burton and Steif 1994), and the
is called a sliding block code. By restriction, we relationships among entropy-preserving, finite-to-
have the notion of sliding block code from one one, and almost invertibility for factor codes
d-dimensional shift space to another. As expected, discussed in section “Other Coding Problems” can
for d-dimensional shifts X and Y, the sliding block be very different in higher dimensions (Meester and
codes f : X ! Y coincide exactly with the con- Steif 2001). Other differences with respect to
tinuous translation-commuting maps from X to Y, entropy can be found in Quas and Sahin (2003).
i.e., the maps which are continuous with respect to There is also the natural notion of higher
the metric r, defined above, and which satisfy dimensional sofic shifts, which can be defined as
f ∘ sn ¼ sn ∘ f for all n  Zd. Thus, it makes those shift spaces that are factors of SFT’s. Recall
sense to consider the various coding problems, in that every one-dimensional sofic shift is a right-
particular the conjugacy, factor, and embedding resolving, and hence entropy-preserving, factor of
problems, in the higher dimensional setting, but an SFT. It is not known if there is an analogue to
these are very difficult. this fact in higher dimensions, although recently
Even the question of determining when a there has been some progress: every sofic shift Y is
higher dimensional SFT of entropy at least logn a factor of an SFT with entropy arbitrarily close to
factors onto the full n-shift seems very difficult h(Y) (Desai 2006).
(in contrast to Theorem 15). However, there are Finally, there is a subclass of d-dimensional
some positive results for strongly irreducible SFT’s which is somewhat tractable, namely,
SFT’s. For instance, it is known that any strongly d-dimensional shifts with group structure in the
irreducible SFT of entropy strictly larger than logn following sense. Let A be a (finite) group. Then
factors onto the full n-shift (Desai 2009; Johnson the full d-dimensional shift over A is also a group
and Madden 2005). In fact, that result requires with respect to the coordinate-wise group struc-
only a weaker assumption than strong irreducibil- ture. A (higher-dimensional) group shift is a sub-
d
ity, but still much stronger than mixing; recent shift of A Z which is also a subgroup. For a survey
examples (Boyle et al. 2010) show, among other on results for this class, we refer the reader to Lind
things, that one cannot weaken that assumption to and Schmidt (2002).
mere mixing.
There are results on other coding problems as
well. For instance, a version of the Embedding Future Directions
theorem in one dimension (Theorem 13) holds for
SFT’s in two dimensions with a strong mixing The future directions of the subject will likely be
property (Lightwood 2003/04); however, it is determined by progress on solutions to open prob-
required that the shift to be embedded contains lems. In the course of describing topics in this
Symbolic Dynamics 451

entry, we have mentioned many open problems Problem,” is one of the most fundamental prob-
along the way. For a much more complete list on a lems in the subject: given two SFTs X and Y, how
wealth of subareas of symbolic dynamics, we can you tell whether X and Y are conjugate? While
refer the reader to the article (Boyle 2007). many sensitive invariants of conjugacy are
While it is difficult to single out the most impor- known, it is still not known if this problem is
tant challenges, certainly the problem of under- even decidable by a finite algorithm.
standing multidimensional shift spaces, especially As described in the “The Conjugacy Problem”
of finite type, is one of the most important. section, conjugacy between a pair of SFTs is
equivalent to strong shift equivalence (SSE) of
the corresponding adjacency matrices. SSE
Addendum to the Second Edition implies shift equivalence (SE) which is known to
be decidable and, for a long time, was conjectured
Since the original publication of this article, there to be equivalent to SSE. But as mentioned in the
has been an enormous amount of activity and section, Kim and Roush (1999) showed that this is
progress in symbolic dynamics. While some false even for irreducible matrices. Since then
major problems remain unsolved, with little or there has been very little progress on the
no progress, others have been solved, partially or conjugacy problem per se.
completely, and vast new areas of exploration SSE and SE are purely algebraic notions for
have been carved out. It would be impossible in matrices over the semi-ring ℤ+. However, they
this brief update to summarize the developments can be defined as meaningful relations over other
in symbolic dynamics within the past 10 years. categories, in particular rings. It is known that
Here, we focus mainly on developments within over ℤ, SSE and SE are equivalent. But there are
subareas of symbolic dynamics most closely rings over which SSE and SE are not equivalent.
related to the content of the original article. We Boyle and Schmedling showed that the extent to
have also included some major results known at which they differ is captured by invariants from
the time of the original article but not included in algebraic K-theory (Boyle and Schmieding 2019).
the article. The section on “Other Coding Problems”
The symbolic modeling approach, introduced began with notions of equivalence of SFTs weaker
in the section on the “Origins of Symbolic than conjugacy, namely, finite equivalence and
Dynamics: Modelling of Dynamical Systems,” almost conjugacy, for which one can effectively
continues to be an essential tool in modelling decide equivalence by simple invariants, namely,
smooth dynamical systems. Markov partitions, entropy and period (Theorems 10 and 11).
such as those for the main examples, hyperbolic A version of Theorem 11 has been extended to
toral automorphisms and Smale’s horsehoe, have the setting of a certain class of shift spaces over
finitely many “rectangles” corresponding to a countably infinite alphabets by Boyle, Buzzi, and
finite alphabet, upon which a shift space model Garcia (2006).
is built. One can also consider shift spaces over Theorems 10 and 11 are intimately related to
countably infinite alphabets instead. Recently finite-to-one factor maps (also called finite-to-one
Markov partitions with countably many rectan- factor codes). Such a map has a “degree,” defined
gles were constructed by Sarig (2013) to model as the “typical” number of preimages of any given
surface diffeomorphisms by shift spaces over point. Allahbakhshi and Quas (2013) introduced
countably infinite alphabets. This enabled him to the analogous notion of class degree for infinite-
prove a fundamental conjecture of Katok [(2007), to-one factor maps. They used this notion to
problem 7.4] on the growth rate of periodic points obtain bounds on the number of maximal relative
for arbitrary surface diffeomorphisms with posi- entropy ergodic lifts of a given ergodic measure
tive entropy. on the range. The class degree has also turned out
The conjugacy problem for shifts of finite type to be key to understanding the structure of infinite-
(SFTs), considered in the section “The Conjugacy to-one factor codes. This was further developed
452 Symbolic Dynamics

by Allahbakhshi, Hong, and Jung (2015) who irreducible SFTs up to a notion of flow conjugacy
established dynamical properties for infinite-to- by simple invariants (Franks 1984). This classifi-
one factor maps analogous to those of the fibers cation has been extended to a large class of irre-
of a finite-to-one factor map. ducible sofic shifts by Boyle, Carlsen, and
The Embedding Theorem (Theorem 13) and Eilers (2018).
Lower Entropy Factor Theorem (Theorem 16) for The state-splitting algorithm for constructing
irreducible SFTs have been generalized by sliding block decodable encoders, introduced in
Thomsen (2004) for special classes of mixing the section on “Coding for Data Recording Chan-
sofic shifts. He also developed structure to explain nels,” was used widely in practice in the 1980s
why his results cannot hold for all mixing sofic and 1990s. Variants of the algorithm were used to
shifts. A recent preprint of Krieger (2018) stated a aid in the construction of codes which became
complete necessary, sufficient, and decidable set standards in the recording industry, including the
of conditions for the Lower Entropy Factor The- (1,7) code (Adler et al. 1983b) for hard disk
orem for all mixing sofic shifts. drives, EFMPlus for the DVD (Immink 1994)
Lind (1984) characterized the set of numbers (see also [(Immink 2004b), section 11.5.2]) and
that can occur as entropies of SFTs in terms of a the code used in the first generation of Linear Tape
set of algebraic integers, known as Perron num- Open (Ashley et al. 1999). Today, the algorithm is
bers; specifically, the entropies of SFTs are used mainly as a proof of concept in new schemes
exactly the logs of roots of Perron numbers. such as codes for radio frequency identification
These are positive, in particular real, algebraic (Barbero et al. 2014), weakly constrained codes
integers. Thurston (1990) introduced a generali- (Elishco et al. 2016), and parity-preserving codes
zation known as complex Perron numbers. As (Roth and Siegel 2018).
shown by Kenyon (1996), these complex num- A major development over the past two decades
bers are precisely the expansion factors l for a has been the extension of symbolic dynamics to the
collection of self-similar tilings of the plane; these multidimensional setting. The set-up was described
are certain tilings by translates of a finite set of in the section on “Higher Dimensional Shift
basic tiles s.t. for each copy T of a basic tile in the Spaces.” Here, bi-infinite sequences are replaced
tiling, lT is tiled by translates of the same set of by arrays of symbols on the points of the integer
basic tiles. For further work in this area, see lattice ℤd. As mentioned in that section, Hochman
Thurston [(2014), especially Theorem 1.3]. and Meyerovitch (2007) characterized the set of
The characterization of the collection of zeta numbers which can occur as the entropy of a ℤd
functions of SFTs (equivalently, numbers of peri- SFT, d  2, as the set of all right recursively
odic orbits of all periods) given in Theorem 14 enumerable numbers (this is very different from
was conjectured by Boyle and Handelman (1991) Lind’s characterization in d ¼ 1 (Lind 1984)).
and proven by Kim, Ormes, and Roush (Kim et al. This result led to a development that has strongly
2000). In particular, this result characterized the tied symbolic dynamics to computability theory.
possible multi-sets of non-zero eigenvalues of Many of the central results in this development
primitive square matrices over ℤ. One can ask are summarized in Jeandel (2016).
for similar characterizations if one replaces ℤ by One central concept in this development is the
other rings. A more general spectral conjecture, notion of an effective shift space X, which means a
which pertains to a wide class of subrings S of ℝ, shift space for which there exists a list F of for-
remains open (Boyle 2007), Conjecture 6.1]. The bidden patterns that defines X and can be pro-
case S ¼ ℝ was already obtained in Boyle and duced by a Turing machine. Of course, any ℤd-
Handelman (1991). For more recent work on this SFT or sofic shift is effective. But even for d ¼ 1,
subject, see Boyle and Schmieding (2016). there are many effective shifts that are not
To every SFT, there is an associated continuous sofic. Nevertheless, any effective ℤd-shift space
suspension flow, and Franks completely classified X can be “simulated” by a ℤd+1-SFT Y in the sense
Symbolic Dynamics 453

that X is a ℤd-subaction of a factor of Y (Aubrun intimately related to the theory of Gibbs states in
and Sablik 2013). statistical mechanical systems (Ruelle 2004).
Using essentially the same argument to show As in the original article, for a list of major
that the entropy of a ℤd-SFT is rre, one can show open problems in Symbolic Dynamics and pro-
that the entropy of an effective ℤd-shift space is gress on their solutions, we refer the reader to
rre. Since the collection of sofic shift spaces lies in Boyle (2007).
between the SFTs and the effective shift spaces, it
follows that the entropies of the sofic shift spaces
are exactly the rre numbers. So, for every ℤd sofic Bibliography
shift there is a ℤd SFT with the same entropy. As
mentioned in the original article, a major open Adler R, Marcus B (1979) Topological entropy and equiv-
problem is whether for every ℤd sofic shift alence of dynamical systems. Memoirs of the American
X there is a ℤd SFT which factors onto X and has Mathematical Society, vol 219. AMS, Providence
Adler R, Weiss B (1970) Similarity of automorphisms of
the same entropy. This problem remains unsolved. the torus.
The extension from ℤ to ℤd symbolic dynam- Memoirs of the American Mathematical Society,
ics can be carried further. Namely, for an arbitrary vol 98. AMS, Providence
countable group G and a finite alphabet A , a Adler R, Konheim A, McAndrew M (1965) Topological
entropy. Trans Am Math Soc 114:309–319
G-shift space is a closed subset of the full shift Adler RL, Goodwyn LW, Weiss B (1977) Equivalence of
space AG that is invariant under the action of G on topological Markov shifts. Isr J Math 27:48–63
A G: for g  G, define g: A G ! A G by: Adler RL, Coppersmith D, Hassner M (1983a) Algorithms
for sliding block codes – an application, of symbolic
dynamics to information theory. Trans IEEE Inf Theory
ðgðxÞÞh ¼ xhg 29:5–22
Adler RL, Hassner M, Moussouris JP (1983b, November
Much of the classical theory for Z and Zd 1) Method and apparatus for generating a noiseless
sliding block code for a (1,7) channel with rate 2/3.
symbolic dynamics has been generalized to large
United States patent 4,413,251
classes of amenable groups G. And more recently Aho AV, Hopcroft JE, Ullman JD (1974) The design and
a notion of sofic entropy for dynamics of certain analysis of computer algorithms. Addison-Wesley,
non-amenable groups has played a prominent role Reading
Allahbakhshi M, Quas A (2013) Class degree and relative
(Bowen 2012, 2018).
maximal entropy. Trans Am Math Soc 365:1347–1368
There are many other subareas of symbolic Allahbakhshi M, Hong S, Jung U (2015) Class closing
dynamics that are very active and of great impor- factor codes and constant-to-one factor codes from
tance today. For instance, in algebraic symbolic shifts of finite type. Dyn Syst 30:485–500
Ashley J (1988) A linear bound for sliding block decoder
dynamics one endows the shift space with a group
window size. Trans IEEE Inf Theory 34:389–399
structure such that the shift map is an automor- Ashley J (1991) Resolving factor maps for shifts of finite
phism. One can characterize many symbolic type with equal entropy. Ergod Theory Dyn Syst
dynamical properties of such shift spaces in 11:219–240
Ashley J (1996) A linear bound for sliding block decoder
terms of algebraic objects (Schmidt 2012). In
window size, II. Trans IEEE Inf Theory 42:1913–1924
cellular automata, one studies the iteration of slid- Ashley J, Jaquette G Marcus B, Seger P (1999) Runlength
ing block codes from one shift space to another as limited encoding/decoding with robust resync. US pat-
dynamical systems in their own right. These ent 5,969,649
Aubrun N, Sablik M (2013) Simulation of effective sub-
dynamical systems exhibit a rich variety of
shifts by two-dimensional subshifts of finite type. Acta
dynamical behavior (Ceccjerini-Silberstein and Appl Math 126:3563
Coornaert 2010). Thermodynamic formalism Barbero A, Rosnes E, Yang G, Ytrehus O (2014) Near-field
studies the notion of topological pressure, a gen- passive RFID communication: channel model and code
design. IEEE Trans Commun 62:1716–1726
eralization of topological entropy and equilibrium
Béal M-P (1990) The method of poles: a coding method for
states, and a generalization of measures of maxi- constrained channels. Trans IEEE Inf Theory
mal entropy (Keller 1998; Sarig 2009). This is 36:763–772
454 Symbolic Dynamics

Béal M-P (1993) Codage symbolique. Masson, Paris of the American Mathematical Society, vol 377. AMS,
Béal M-P (2003) Extensions of the method of poles for Providence
code construction. Trans IEEE Inf Theory Boyle M, Buzzi J, Gomez-Aiza R (2006) Almost isomor-
49:1516–1523 phism for countable state Markov shifts. J Reine
Bedford T (1986) Generating special Markov partitions for Angew Math 592:2347
hyperbolic toral automorphisms using fractals. Ergod Boyle M, Pavlov R, Schraudner M (2010) Multi-
Theory Dyn Syst 6:325–333 dimensional sofic shifts without separation and their
Berger R (1966) The undecidability of the domino prob- factors. Trans Amer Math Soc 362:4617–4653
lem. Memoirs of the American Mathematical Society. Boyle M, Carlsen TM, Eilers S (2018) Flow equivalence of
AMS, Providence sofic shifts. Isr J Math 225:111–146
Berstel J, Perrin D (1985) Theory of codes. Academic, Burton R, Steif J (1994) Nonuniqueness of measures of
New York maximal entropy for subshifts of finite type. Ergod
Blanchard F, Maass A, Nogueira A (2000) In: Blanchard F, Theory Dyn Syst 14(2):213–235
Maass A, Nogueira A (eds) Topics in symbolic dynam- Calkin N, Wilf H (1998) The number of independent sets in
ics and applications. LMS lecture notes, vol 279. Cam- a grid graph. SIAM J Discret Math 11:54–60
bridge University Press, Cambridge Ceccjerini-Silberstein, T. and Coornaert, M. Cellular
Blanchard P, Devaney R, Keen L (2004) Complex dynam- automata and groups (2010), Springer Berlin
ics and symbolic dynamics. In: Williams S (ed) Sym- Cidecyian R, Evangelos E, Marcus B, Modha M (2001)
bolic dynamics and its applications. Proceedings of Maximum transition run codes for generalized partial
symposia in applied mathematics. AMS, Providence, response channels. IEEE J Sel Area Commun
pp 37–59 19:619–634
Bowen R (1970) Markov partitions for axiom Coven E, Paul M (1974) Endomorphisms of irreducible
A diffeomorphisms. Am J Math 92:725–747 shifts of finite type. Math Syst Theory 8:167–175
Bowen R (1973) Symbolic dynamics for hyperbolic flows. Coven E, Paul M (1975) Sofic systems. Isr J Math
Am J Math 95:429–460 20:165–177
Bowen L (2012) Sofic entropy and amenable groups. Coven E, Paul M (1977) Finite procedures for sofic sys-
Ergod Theory Dyn Syst 32:427–466 tems. Monats Math 83:265–278
Bowen L (2018) A brief introduction to sofic entropy Cover T, Thomas J (1991) Elements of information theory.
theory, arXiv: 1711.02062v2 Wiley, New York
Bowen R, Franks J (1977) Homology for zero-dimensional Desai A (2006) Subsystem entropy for Zd sofic shifts.
basic sets. Ann Math 106:73–92 Indag Math 17:353–360
Bowen R, Lanford OE (1970) Zeta functions of restrictions Desai A (2009) A class of Zd -subshifts which factor onto
of the shift transformation. Proceedings of symposia in lower entropy full shifts. Proc Am Math Soc 137:2613–
pure mathematics, vol 14. AMS, Providence, pp 43–50 2621
Boyle M (1983) Lower entropy factors of sofic systems. Devaney R (1987) An introduction to chaotic dynamical
Ergod Theory Dyn Syst 3:541–557 systems. Addison-Wesley, Reading
Boyle M (1993) Symbolic dynamics and matrices. In: Elishco O, Meyerovitch T, Schwartz M (2016) Semicon-
Brualdi R et al (eds) Combinatorial and graph theoretic strained systems. IEEE Trans Inf Theory 62:1688–1702
problems in linear algebra. IMA volumes in mathemat- Fischer R (1975a) Sofic systems and graphs. Monats Math
ics and its applications, vol 50. Springer New York, 80:179–186
New York, pp 1–38 Fischer R (1975b) Graphs and symbolic dynamics. Colloq
Boyle M (2007) Open problems in symbolic dynamics. Math Soc János Bólyai: Top Inf Theory 16:229–243
Contemponary Math; for solutions and updates, see Franaszek PA (1968) Sequence-state coding for digital
http://www.math.umd.edu/mmb transmission. Bell Syst Tech J 47:143–155
Boyle M, Handelman D (1991) The spectra of nonnegative Franaszek PA (1982) Construction of bounded delay codes
matrices via symbolic dynamics. Ann Math for discrete noiseless channels. J IBM Res Dev
133:249–316 26:506–514
Boyle M, Krieger W (1986) Almost Markov and shift Franaszek PA (1989) Coding for constrained channels: a
equivalent sofic systems. In: Aleixander comparison of two approaches. J IBM Res Dev
J (ed) Proceedings of Maryland special year in dynam- 33:602–607
ics 1986–87. Lecture notes in mathematics, vol 1342. Franks J (1984) Flow equivalence of subshifts of finite
Springer, Berlin, pp 33–93 type. Ergod Theory Dyn Syst 4:5366
Boyle M, Schmieding S (2016) Strong shift equivalence Friedman J (1990) On the road coloring problem. Proc Am
and the generalized spectral conjecture for nonnegative Math Soc 110:1133–1135
matrices. Linear Algebra Appl 498:231–243 Gomez R (2003) Positive K-theory for finitary iso-
Boyle M, Schmieding S (2019) Strong shift equivalence and moprhisms of Markov chains. Ergod Theory Dyn
algebraic K-theory. J Reine Angew Math 752:63–104 Syst 23:1485–1504
Boyle M, Marcus BH, Trow P (1987) Resolving maps and Hadamard J (1898) Les surfaces a courbures opposées et
the dimension group for shifts of finite type. Memoirs leurs lignes geodesiques. J Math Pure Appl 4:27–73
Symbolic Dynamics 455

Hassellblatt B, Katok A (1995) Introduction to the modern Kim KH, Ormes N, Roush F (2000) The spectra of non-
theory of dynamical systems. Cambridge University negative integer matrices via formal power series. Am
Press, Cambridge J Math Soc 13:773–806
Hedlund GA (1939) The dynamics of geodesic flows. Bull Kitchens B (1998) Symbolic dynamics: one-sided,
Am Math Soc 45:241–260 two-sided and countable state Markov chains. Springer,
Hedlund GA (1944) Sturmian minimal sets. Am J Math Berlin
66:605–620 Kitchens B, Schmidt K (1988) Periodic points, decidability
Hedlund GA (1969) Endomorphisms and automorphisms and Markov subgroups. In: Alexander JC (ed) Dynamical
of the shift dynamical system. Math Syst Theory systems: proceedings of the special year. Springer lecture
3:320–375 notes in mathematics, vol 1342. Springer Berlin Heidel-
Hochman M (2009) On the dynamics and recursive prop- berg, Berlin/Heidelberg, pp 440–454
erties of multidimensional symbolic systems. Invent Kitchens B, Marcus B, Trow P (1991) Eventual factor
Math 176:131–167 maps and compositions of closing maps. Ergod Theory
Hochman M, Meyerovitch T (2007) A characterization of Dyn Syst 11:85–113
the entropies of multidimensional shifts of finite type. Krieger W (1980a) On a dimension for a class of homeo-
Ann Math 171(2010):2011–2038 morphism groups. Math Ann 252:87–95
Hollmann HDL (1995) On the construction of bounded- Krieger W (1980b) On dimension functions and topologi-
delay encodable codes for constrained systems. Trans cal Markov chains. Invent Math 56:239–250
IEEE Inf Theory 41:1354–1378 Krieger W (1982) On the subsystems of topological Mar-
Immink KAS (1994) EFMPlus, 8–16 modulation code. US kov chains. Ergod Theory Dyn Syst 2:195–202
patent 5,696,505 Krieger W (1983) On the finitary isomorphisms of Markov
Immink KAS (2004a) Codes for mass data storage, shifts that have finite expected coding time. Wahrsch
2nd edn. Shannon Foundation Press, Eindhoven Z 65:323–328
Immink KAS (2004b) Codes for mass data storage sys- Krieger W (1984) On sofic systems I. Isr J Math
tems, 2nd edn. Shannon Foundation Publishers, 48:305–330
Eindhoven Krieger W (2018) On images of sofic systems,
Jeandel E (2016) Computability in symbolic dynamics. arXiv:1101.1750
CiE, Paris, pp 124–131 Lightwood S (2003/04) Morphisms form non-periodic Z2
Johnson A, Madden K (2005) Factoring higher- subshifts I and II. Ergod Theory Dyn Syst 23:587–609,
dimensional shifts of finite type onto full shifts. Ergod 24:1227–1260
Theory Dyn Syst 25:811–822 Lind D (1984) The entropies of topological Markov shifts
Karabed R, Siegel P, Soljanin E (1999) Constrained coding and a related class of algebraic integers. Ergod Theory
for binary channels with high intersymbol Dyn Syst 4:283–300
intereference. Trans IEEE Inf Theory 45:1777–1797 Lind D (1989) Perturbations of shifts of finite type. SIAM
Kari J (2001) Synchronizing finite automata on Eulerian J Discret Math 2:350–365
digraphs. Springer Lect Notes Comput Sci Lind D (1996) A zeta function for Zd -actions. In:
2136:432–438 Pollicott M, Schmidt K (eds) Proceedings of Warwick
Kastelyn PW (1961) The statistics of dimers on a lattice. symposium on Zd -actions. LMS lecture notes, vol 228.
Physica A 27:1209–1225 Cambridge University Press, Cambridge, pp 433–450
Katok A (2007) Fifty years of entropy in dynamics Lind D (2004) Multi-dimensional symbolic dynamics. In:
1958–2007. J Modern Dyn 4:545–596 Williams S (ed) Symbolic dynamics and its applica-
Keller G (1998) Equilibrium states in ergodic theory. Cam- tions. Proceedings of symposia in applied mathematics,
bridge University Press, Cambridge vol 60. AMS, Providence, pp 81–120
Kenyon R (1996) The construction of self-similar tilings. Lind D, Marcus B (1995) An introduction to symbolic
Geom Funct Anal 6:471–488 dynamics and coding. Cambridge University Press,
Kenyon R (2008) Lectures on dimers. http://www.math. Cambridge
brown.edu/~rkenyon/papers/dimerlecturenotes.pdf Lind D, Schmidt K (2002) Symbolic and algebraic dynam-
Kim KH, Roush FW (1979) Some results on decidability of ical systems. In: Hasselblatt B, Katok A (eds) Hand-
shift equivalence. J Comb Inf Syst Sci 4:123–146 book of dynamics systems. Elsevier, Amsterdam,
Kim KH, Roush FW (1988) Decidability of shift equiva- pp 765–812
lence. In: Alexander J (ed) Proceedings of Maryland Manning A (1971) Axiom A diffeomorphisms have ratio-
special year in dynamics 1986–87. Lecture notes in nal zeta functions. Bull Lond Math Soc 3:215–220
mathematics, vol 1342. Springer, Berlin, pp 374–424 Marcus B (1979) Factors and extensions of full shifts.
Kim KH, Roush FW (1990) An algorithm for sofic shift Monats Math 88:239–247
equivalence. Ergod Theory Dyn Syst 10:381–393 Marcus BH, Roth RM (1991) Bounds on the number of
Kim KH, Roush FW (1999) Williams conjecture is false states in encoder graphs for input-constrained channels.
for irreducible subshifts. Ann Math 149:545–558 Trans IEEE Inf Theory 37:742–758
456 Symbolic Dynamics

Marcus B, Tuncel S (1990) Entropy at a weight-per- Parry W, Tuncel S (1981) On the classification of Markov
symbol and embeddings of Markov chains. Invent chains by finite equivalence. Ergod Theory Dyn Syst
Math 102:235–266 1:303–335
Marcus B, Tuncel S (1991) The weight-per-symbol poly- Parry W, Tuncel S (1982) Classification problems in ergo-
tope and scaffolds of invariants associated with Markov dic theory. LMS lecture notes, vol 67. Cambridge Uni-
chains. Ergod Theory Dyn Syst 11:129–180 versity Press, Cambridge
Marcus B, Tuncel S (1993) Matrices of polynomials, pos- Pavlov R (2011) Perturbations of multidimensional shifts
itivity, and finite equivalence of Markov chains. J Am of finite type. Ergod Theory Dyn Syst 31:483–526
Math Soc 6:131–147 Petersen K (1989) Ergodic theory. Cambridge University
Marcus BH, Roth RM, Siegel PH (1998) Constrained sys- Press, Cambridge
tems and coding for recording chapter. In: Brualdi R, Quas A, Sahin A (2003) Entropy gaps and locally maximal
Huffman C, Pless V (eds) Handbook on coding theory. entropy in Zd -subshifts. Ergod Theory Dyn Syst
Elsevier, New York; updated version at http://www. 23:1227–1245
math.ubc.ca/~marcus/Handbook/ Quas A, Trow P (2000) Subshifts of multidimensional shifts
Markley N, Paul M (1981a) Matrix subshifts for Zv sym- of finite type. Ergod Theory Dyn Syst 20:859–874
bolic dynamics. Proc Lond Math Soc 43:251–272 Radin C (1996) Miles of tiles. In: Pollicott M, Schmidt
Markley N, Paul M (1981b) Maximal measures and K (eds) Ergodic theory of Zd -actions. LMS lecture
entropy for Zv subshifts of finite type. In: Devaney R, notes, vol 228. Cambridge University Press, Cam-
Nitecki Z (eds) Classical mechanics and dynamical bridge, pp 237–258
systems. Dekker notes, vol 70. Dekker, New York, Robinson RM (1971) Undecidability and nonperiodicity
pp 135–157 for tilings of the plane. Invent Math 12:177–209
Meester R, Steif J (2001) Higher-dimensional subshifts of Robinson EA (2004) Symbolic dynamics and tilings of Rd.
finite type, factor maps and measures of maximal In: Williams S (ed) Symbolic dynamics and its appli-
entropy. Pac Math J 200:497–510 cations. Proceedings of symposia in applied mathemat-
Morse M (1921) Recurrent geodesics on a surface of neg- ics, vol 60. AMS, Providence, pp 81–120
ative curvature. Trans Am Math Soc 22:84–100 Roth R, Siegel P (2018) On parity-preserving constrained
Morse M, Hedlund GA (1938) Symbolic dynamics. Am coding. In: Proceedings of international symposium on
J Math 60:815–866 information theory, pp 1804–1808
Morse M, Hedlund GA (1940) Symbolic dynamics II, Rudolph D (1990) Fundamentals of measurable dynamics.
Sturmian trajectories. Am J Math 62:1–42 Oxford University Press, Oxford
Mouat R, Tuncel S (2002) Constructing finitary isomor- Ruelle D (2004) Thermodynamic formalism, 2nd edn.
phisms with finite expected coding time. Isr J Math Cambridge University Press, Cambridge
132:359–372 Sarig O (2009) Lecture notes on thermodynamic formalism
Mozes S (1989) Tilings, substitutions and the dynam- for topological Markov shifts. http://www.weizmann.ac.
ical systems generated by them. J Anal Math il/math/sarigo/sites/math.sarigo/files/uploads/tdfnotes.pdf
53:139–186 Sarig O (2013) Symbolic dynamics for surface
Mozes S (1992) A zero entropy, mixing of all orders tiling diffeomorphisms with positive entropy. J Am Math
system. In: Walters P (ed) Symbolic dynamics and its Soc 26:341–426
applications. Contemporary mathematics, vol 135. Schmidt K (1984) Invariants for finitary isomorphisms with
AMS, Providence, pp 319–326 finite expected code lengths. Invent Math 76:33–40
Nagy Z, Zeger K (2000) Capacity bounds for the three- Schmidt K (1990) Algebraic ideas in ergodic theory. AMS-
dimensional (0, 1) run length limited channel. Trans CBMS reg conference, vol 76. AMS, Providence
IEEE Inf Theory 46:1030–1033 Schmidt K (1995) Dynamical systems of algebraic origin.
Nasu M (1986) Topological conjugacy for sofic systems. Birkhauser, Basel
Ergod Theory Dyn Syst 6:265–280 Schmidt K (2012) Dynamical systems of algebraic origin.
Ornstein D (1970) Bernoulli shifts with the same entropy Birkhauser (reprint of 1995 edition)
are isomorphic. Adv Math 4:337–352 Schwartz M, Bruck S (2008) Constrained codeds as net-
Parry W (1964) Intrinsic Markov chains. Trans Am Math works of relations. IEEE Trans Inf Theory 54:2179–2195
Soc 112:55–66 Seneta E (1980) Non-negative matrices and Markov
Parry W (1977) A finitary classification of topological chains, 2nd edn. Springer, Berlin
Markov chains and sofic systems. Bull Lond Math Shannon C (1948) A mathematical theory of communica-
Soc 9:86–92 tion. Bell Syst Tech J 27:379–423, 623–656
Parry W (1979) Finitary isomorphisms with finite expected Sinai YG (1968) Markov partitions and
code-lengths. Bull Lond Math Soc 11:170–176 C-diffeomorphisms. Funct Anal Appl 2:64–89
Parry W (1991) Notes on coding problems for finite state Smale S (1967) Differentiable dynamical systems. Bull
processes. Bull Lond Math Soc 23:1–33 Am Math Soc 73:747–817
Parry W, Schmidt K (1984) Natural coefficients and invari- Thomsen K (2004) On the structure of a sofic shift space.
ants for Markov shifts. Invent Math 76:15–32 Am Math Soc Trans 356:3557–3619
Symbolic Dynamics 457

Thurston WP (1990) Groups, tilings, and finite state Walters P (1982) An introduction to ergodic theory.
automata. Lecture notes, AMS colloquium lectures, Springer graduate texts in mathematics, vol 79.
American Mathematical Society Springer, Berlin
Thurston WP (2014) Entropy in dimension one. Frontiers Walters P (1992) In: Walter P (ed) Symbolic dynamics and
in complex dynamics, Princeton mathematical series, its applications. Contemporary mathematics, vol 135.
vol 51. Princeton University Press, Princeton AMS, Providence
Trachtman A (2007) The road coloring problem. Israel Ward T (1994) Automorphisms of Zd -subshifts of finite
J Math 172:51–60 type. Indag Math 5:495–504
Tuncel S (1981) Conditional pressure and coding. Isr Weiss B (1973) Subshifts of finite type and sofic systems.
J Math 39:101–112 Monats Math 77:462–474
Tuncel S (1983) A dimension, dimension modules and Williams RF (1973/74) Classification of subshifts of finite
Markov chains. Proc Lond Math Soc 46:100–116 type. Ann Math 98:120–153; Erratum: Ann Math
Wagoner J (1992) Classification of subshifts of finite type 99:380–381
revisited. In: Walters P (ed) Symbolic dynamics and its Williams S (2004a) Introduction to symbolic dynamics.
applications. Contemporary mathematics, vol 135. In: Williams S (ed) Symbolic dynamics and its
AMS, Providence, pp 423–444 applications. Proceedings of symposia in applied
Wagoner J (2004) Strong shift equivalence theory. In: mathematics, vol 60. AMS, Providence, pp 1–12
Walters P (ed) Symbolic dynamics and its applications. Williams S (2004b) In: Williams S (ed) Symbolic dynam-
Proceedings of symposia in applied mathematics, ics and its applications. Proceedings of symposia in
vol 60. AMS, Providence, pp 121–154 applied mathematics, vol 60. AMS, Providence
 Ergodic measure-preserving trans-
Operator Ergodic Theory formation A mpt such that the only sets A
with θ1A = A a.e are null sets or complements
Guy Cohen1 and Michael Lin2 of null sets.
1
School of Electrical Engineering, Ben-Gurion Infinitesimal generator of a C0-semigroup The
University, Beer-Sheva, Israel operator Ax ≔ limt!0+ t1(T(t)x  x) with
2
Department of Mathematics, Ben-Gurion domain D ðAÞ≔fx  X : Ax exists}.
University, Beer-Sheva, Israel Koopman operator The operator on Lp(Ω, S,
m), induced by a measure-preserving transfor-
mation θ, defined by Tf = f ∘ θ.
Article Outline Markov-Feller operator A Markov operator on
bounded measurable functions of a compact
Glossary Hausdorff space which preserves continuity.
Definition of the Subject Markov operator The operator induced on
Introduction bounded measurable functions on (Ω, S) by
The Mean Ergodic Theorem a transition probability P(t, A), defined by
Rates of Convergence Pf(t) = f(s)P(t, ds).
Uniform Ergodic Theorems Measure (probability)-preserving transforma-
Strong Cesàro Convergence tion (mpt) A measurable mapping θ of a
Weak Stability and Mixing measure (probability) space (Ω, S, m) satisfy-
Stability ing m(θ1A) = m(A) for any A  S.
Averaging Along Subsequences Minimal topological dynamical system A
General Averaging Methods topological dynamical system (K, t) such that
Modulated Ergodic Theorems for every s  K the orbit {tns} is dense.
Resolvent Conditions and Growth of Powers Positive operator (on a Banach lattice) An
Continuous Time (C0-semigroups) operator which maps the positive cone of a
Bibliography Banach lattice into itself.
Power-bounded operator An operator T with
Glossary supnkT nk < 1.
Operator A bounded linear operator on a (real or
Blum-Hanson property Norm convergence to complex) Banach space X.
zero of all averages 1n nj¼1 T kj x along any Operator semigroup A family S of operators
increasing {kj}  ℕ whenever T nx ! 0 weakly. such that TS  S for T, S  S.
Cesàro bounded operator An operator T with Orbit of a vector under an operator semigroup
supn 1n n1 k
< 1: S The set fTx : T  S g.
k¼0 T
Coboundary (of an operator) A vector (Semi)flow A family of transformations
y  (I  T)X, i.e., such that y = x  Tx for {θt}t  ℝ({θt}t0) such that θt+s = θt ∘ θs, with
some x  X. θ0 the identity map.
Contraction An operator T with kT k  1. Stolz region The closed convex hull (in ℂ) of the
C0-semigroup/one-parameter semigroup A point 1 and a disk of radius r < 1.
family of bounded operators {T(t)}t  0 Transition probability A function P(t, A) on
defined on X such that T(t + s) = T(t)T(s) for t, Ω  S (where (Ω, S) is a measurable space),
s  0, T(0) = I, and T(t)x is continuous on such that P(t, ) is a probability on S for t  Ω
[0, 1) for every x  X. and P(, A) is measurable for every A  S.

© Springer Science+Business Media, LLC, part of Springer Nature 2023 459

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_776
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_776-1
460 Operator Ergodic Theory

Transformation A mapping of a set into itself. Halmos (1949a): “It was very quickly recognized
Topological dynamical system A compact that the proper general framework for von
Hausdorff space K with a continuous map t Neumann’s mean ergodic theorem lay in the direc-
of K to itself. tion of Hilbert spaces and Banach spaces,
Uniform ergodicity Convergence of the aver- whereas the extent of generality suitable to
ages 1n n1 k
k¼0 T in the operator norm (uniformly
Birkhoff’s theorem was to be found in the concept
on the unit ball of the Banach space). of a measure space.” Thus, the origin of operator
Uniquely ergodic topological dynamical ergodic theory is in the mean ergodic theorem,
system A topological dynamical system although the uniform distribution theorem of
(K, t) with a unique probability measure m Weyl (1916) can be restated as an ergodic theo-
satisfying m ∘ t1 = m. rem. The ergodic theory of measure-preserving
(Weak) mean ergodic theorem (weak) conver- transformations and pointwise limit theorems for
gence of the averages 1n n1 k some classes of positive operators are discussed in
k¼0 T x for every x in
other chapters of this volume.
the Banach space X on which the operator T is
Birkhoff and von Neumann were both moti-
defined.
vated by a problem in statistical mechanics,
(Weak) stability (Weak) convergence of T nx for
Boltzmann’s “ergodic hypothesis,” so accord-
every x  X.
ingly their theorems were formulated and proved
(Weakly) almost periodic operator An opera-
in continuous time, i.e., for a flow of invertible
tor T such that for every x  X the orbit {T nx}
measure-preserving transformations rather than
is (weakly) conditionally compact.
for the iterates of one such transformation
(Weakly) quasi-compact operator An operator
(discrete time). See Petersen (1996) for compari-
T such that for some n > 0 there exists a (weakly)
son of the significance in scientific applications of
compact operator Q satisfying kT n  Qk < 1.
these two theorems.
Operator ergodic theory has applications in the
study of measure-preserving transformations,
Definition of the Subject
topological dynamics, and Markov operators, but
it is also inspired by results first obtained in these
Operator ergodic theory deals with the asymptotic
areas. Results for C0-semigroups (continuous
behavior of the powers of a power-bounded oper-
time) have applications also in partial differential
ator, or a bounded one-parameter semigroup of
equations.
operators, defined on a (real or complex) Banach
space. We will focus on the theory for powers of a
single operator, and with the growth of the norms
of certain operators which are only Cesàro Introduction
bounded. The asymptotic behavior is studied in
different operator topologies and in various Let ≔fz  ℂ :jzj¼ 1g be the unit circle in the
modes of convergence. However, almost every- complex plane, and let α  [0, 1) be irrational.
where convergence of positive operators on Lp The transformation yz≔e2pia z, z   corresponds
spaces is not discussed methodically in this chap- to a rotation of the circle by the angle 2πα. In
ter, since this is done in other chapters of this 1884, Kronecker proved that for each z the orbit
volume. {θnz} is dense in . Weyl (1910) extended it in his
The topic of ergodic theorems emerged from famous uniform distribution theorem, which,
the mean ergodic theorem for unitary operators in using Weyl (1916), can be stated as a mean ergo-
complex Hilbert spaces of von Neumann (1932) dic theorem in CðÞ.
and the pointwise ergodic theorem of
G.D. Birkhoff (1931) for measure-preserving Theorem 1 Let α  [0, 1) be irrational, and for f
transformations. We quote from the survey of continuous on  define Tf(z) ¼ f(e2πiαz). Then
Operator Ergodic Theory 461

1 n
k¼1 T f ðzÞ converges uniformly to the constant
k In Hopf (1937), F. Riesz proved elementarily
n
f dm, where m is the normalized Lebesgue mea- that when T is a contraction of a Hilbert space,
sure on . Tx ¼ x if and only if T x ¼ x, and extended
The result is immediate for trigonometric poly- Theorem 3 to contractions. Visser (1938) proved
nomials, which by Fejér’s theorem are dense in weak mean ergodicity of power-bounded opera-
CðÞ; this proves the theorem. Moreover, trigo- tors in Hilbert space. G. Birkhoff (1939) proved
nometric polynomials are dense in L2 ð, mÞ, and the mean ergodic theorem for contractions in uni-
we obtain the following. formly convex spaces, extending the Lp result of
F. Riesz (1938).
In their study of Markov chains, Kryloff and
Corollary 2 Let α  [0, 1) be irrational, and let
Bogoliouboff (1937) presented the following.
Tf(z) ¼ f(e2πiαz) be the Koopman operator on
L2 ð, mÞ: Then for every f  L2 ð, mÞ we have
1 n Theorem 4 Let P(t, A) be a transition probabil-
k¼1 T f  f dm 2 ! 0:
k
n ity on (Ω, S) and T the Markov operator defined
Now let θ be any invertible measure- on the space of bounded measurable functions by
preserving transformation of a probability space
Tf(t) ¼ f(s)P(t, ds). If there exist n  1 and a
(Ω, S, m). Then the Koopman operator Tf ¼ f ∘ θ
compact operator Q, such that kT n  Qk < 1,
is unitary on L2(Ω, S, m), an observation which
then T is uniformly ergodic – the averages
led von Neumann (1932) to use the spectral theo- 1 n k
rem to prove the following mean ergodic theorem. n k¼1 T converge in operator norm.

Theorem 3 Let T be a unitary operator on a The Mean Ergodic Theorem

complex Hilbert space ℋ, and let E be the
orthogonal projection on the space F (T ) of An operator T on a (real or complex) Banach space
T-fixed vectors. Then for any x  ℋ we have X is called (weakly) mean ergodic if for every
1 n
k¼1 T x  Ex ! 0:
k
n x  X the averages Mn x≔ 1n n1 k
k¼0 T x converge
Koopman and von Neumann (1932) proved (weakly); the limit operator E is a projection on
that if T in Theorem 3 has no eigenvalues except the subspace F (T ) ≔ {y  X : Ty ¼ y} of fixed
1, then points of T. A (weakly) mean ergodic operator
must be Cesàro bounded (supnkMnk < 1) and,
Ex ¼ 0 ) since T nx ¼ (n þ 1)M(n þ 1)x  nMnx, must
1
n1 ð1Þ satisfy 1n T n x ! 0 (weakly) for every x.
j T k x, y j ! 0 for every y  ℋ,
n
Lemma 5 If 1n T n x ! 0 weakly and for some
k¼0

but T nx need not converge weakly to 0. Hopf subsequence {nj} we have Mnj x ! y weakly,
(1932) characterized (1) when T is the Koopman then Ty ¼ y.
operator of an ergodic probability-preserving
transformation t, by ergodicity of (t  t) Proposition 6 Let T be Cesàro bounded on
(u, v) ≔ (tu, tv) on (Ω  Ω, m  m); in Hopf X satisfying 1n T n x ! 0 for every x  X. Then
(1937), he called such t weak mixing, while t the following are equivalent for x  X:
was called mixing if T nx ! Ex weakly.
By a strange coincidence, Banach’s book (i) kMnxk ! 0.
(1932) was published in the same year, and it (ii) Mnj x ! 0 weakly for some increasing sub-
was realized that von Neumann’s theorem can be sequence {nj}.
generalized “to more general Banach spaces and (iii) f(x) ¼ 0 for every functional
to more general classes of operators,” quoting f  F (T )  X .
Kakutani (1950). (iv) x  ðI  T ÞX:
462 Operator Ergodic Theory

The proof of (iii) implies (iv) uses the Hahn- the semigroup and x  X. Eberlein proved a general
Banach theorem. (i) ) (ii) ) (iii) are obvious, ergodic theorem on the convergence of Mαx, from
and (iv) ) (i) is easy. Proposition 6 is part of the which he deduced Theorem 7.
proof in Yosida (1938) of the following mean The following was proved by Krotkov and
ergodic theorem, proved independently in Halperin (1953).
Kakutani (1938).
An operator T on X is called weakly almost
Proposition 9 Let T be Cesàro bounded on a
periodic if for every x  X the orbit {T nx}n is
reflexive Banach X. If 1n T n x ! 0 (weakly) for
weakly conditionally compact.
every x  X, then T is (weakly) mean ergodic.
Theorem 7 Let T be a weakly almost periodic Combining Lemma 5 and Proposition 6, we
operator on X. Then T is mean ergodic. obtain that a power-bounded operator is mean
Since the Krein-Šmulian theorem appeared ergodic if (and only if) it is weakly mean
only in 1940, Yosida assumed that T is power- ergodic. However, this is false in general;
bounded and that {Mnx} is weakly compact. Both Derriennic (2000) has an example of a weakly
assumptions follow from weak almost periodicity. mean ergodic operator on a Hilbert space which
The operator in Theorem 1 is weakly almost is not mean ergodic (so lim sup 1n kT n k > 0Þ:
periodic on CðÞ. Weak almost periodicity is not A general method for obtaining examples like
necessary for mean ergodicity of a power- Derriennic’s was given by Tomilov and
bounded operator; Sine (1976) has a mean ergodic Zemánek (2004).
positive contraction T on C(K ) such that T2 is not Hille (1945) constructed, on L1[0, 1], the first
mean ergodic, hence T is not weakly almost example of a mean ergodic operator which is not
periodic. Several other examples are presented power-bounded, with kT nk O(n1/4). Kornfeld
by Gerlach and Glück (2019). However, if T is a and Kosek (2003) constructed, for any γ  (0, 1),
positive contraction on a Banach lattice with order a positive mean ergodic operator on L1[0, 1]
continuous norm, mean ergodicity of T implies satisfying limn nγ1kT nk ¼ 1. Kosek (2011)
that of T2, by Derriennic and Krengel (1981). constructed a mean ergodic operator on L1 with
Komorník (1993) proved that a positive contrac- lim sup n1kT nk > 0. Cesàro boundeness alone
tion on L1 is weakly almost periodic if (and only does not imply mean ergodicity, even if X is finite-
if) it is mean ergodic. dimensional, by a simple example of Assani
(1986). However, Émilion (1985) proved that
Corollary 8 Every power-bounded operator on any Cesàro-bounded positive operator on a reflex-
a reflexive Banach space is mean ergodic. ive Banach lattice is mean ergodic (n1T n ! 0
Corollary 8 was proved by Lorch (1939), inde- strongly is not assumed, but follows).
pendently of Kakutani and of Yosida. Kakutani Lemma 5 and Proposition 6 yield the
and Yosida both saw that for proving Corollary following:
8 you really need only weak almost periodicity of
the operator. The results of von Neumann, Riesz,
Visser, and Birkhoff mentioned in the introduc- Theorem 10 Let T be weakly mean ergodic on
tion are special cases of Corollary 8. X. Then
Eberlein (1949) extended the mean ergodic the-
orem to general operator semigroups S, replacing X ¼ F ðT Þ ðI  T ÞX, ð2Þ
the Cesàro averages by what he called invariant
integrals – a family {Mα} of operators on the and the limit operator E is the projection on F (T )
space X, indexed by α in a directed set, such that corresponding to (2).
for any x  X, Mαx is in the closed convex hull of Conversely, if T is Cesàro bounded and
1 n
the orbit fTx : T  S g of x, supαkMαk < 1, and n T x ! 0 (weakly) for every x  X and (2)
limα Mα(I  T)x ¼ limα(I  T)Mαx ¼ 0 for any T in holds, then T is (weakly) mean ergodic.
Operator Ergodic Theory 463

For T as in Proposition 6, it is easy to show that (or L1); Anosov (1973) proved that there exists
the set of points x  X such that (Mnx) converges f  CðÞ such that f ¼ g  g ∘ θ with g non-
is a closed subspace, which equals the right-hand integrable. Gottschalk and Hedlund (1955) proved
side of (2). that if (K, t) is a minimal topological dynamical
Using the decomposition (2), we obtain an system and Tf ¼ f ∘ t on C(K), then f  (I  T)
n1 k
easy proof, due to Satō (1979), of the following C(K) if (and only if) supn k¼0 T f 1 < 1:
theorem of Sine (1970), originally proved for Inspired by this result, Browder (1958) proved the
contractions. following:

Theorem 11 Let T be Cesàro bounded with Theorem 14 Let T be power-bounded on a

1 n
nT x ! 0 for every x  X. Then T is mean reflexive Banach space X. Then
ergodic if and only if the fixed points of
T separate the fixed points of T . n1

Proposition 6 yields easily that for T as in y  ðI T ÞX if and only if sup T k y < 1:
n
k¼0
Theorem 11, the fixed points of T  always sepa-
ð3Þ
rate those of T, so if X is reflexive, we obtain
Corollary 8 from Theorem 11.
Put Sn ≔ n1k¼0 T : The seemingly weaker con-
k

Corollary 12 Let (K, t) be a uniquely ergodic dition supN N n¼1 kSn yk2 < 1 also implies
1 N

topological dynamical system. Then the operator y  (I  T )X. This condition was shown in
Tf ¼ f ∘ t on C(K ) is mean ergodic. Robinson (1960) to imply y  (I  T )X when
The corollary is due to Oxtoby (1952). Theo- T is unitary; see also Kozma and Lev (2011).
rem 1 follows from Theorem 7 or from Cuny and Weber (2018) and Volný (2018) proved
Corollary 12. that if T is a contraction on a Hilbert space, then
1 2
k¼0 k T k y < 1 implies y  (I  T )X. Itera-
Theorem 13 Let 0  rn ! 0, and let T be mean tive methods for solving Poisson’s equation
ergodic, with Ex ¼ lim Mnx. If for every x  X (I  T )x ¼ y were studied by Reich (1973).
there is cx such that kMnx  Exk  cxrn, then Lin and Sine (1983) extended Browder’s result
kMn  Ek ! 0 (T is uniformly ergodic). to dual operators, proving (3) when X ¼ Y  and
Theorem 13 is a consequence of the uniform T ¼ S , with S power-bounded on Y; see also
boundedness principle. When T acts on a complex Devys (2012). They also proved (3) for contrac-
Banach space, the decomposition (2) yields that tions on L1. See Aaronson and Weiss (2000) for
when T is uniformly ergodic, 1 is an isolated point related results. Kornfeld and Lin (1997) proved (3)
in the spectrum of T (see the section Uniform for irreducible Markov-Feller operators on C(K ),
ergodic theorems). Since, by A. Ionescu Tulcea thus extending the topological dynamics result of
(1963) and Foiaş (1964), the spectrum of the L2 Gottschalk and Hedlund (1955). Lin and Suciu
Koopman operator T of an invertible ergodic (2015) studied the equation (I  T )x ¼ y when
probability-preserving transformation of a T is weakly mean ergodic.
Lebesgue space is the whole unit circle , T has Square functions were originally used in the
no uniform rate of convergence in the mean ergo- study of a.e. convergence of orthogonal expan-
dic theorem. A constructive proof for this was sions; see the survey of Stein (1982). Motivated
given by Krengel (1978); see also Kakutani and by a square function inequality by Jones,
Petersen (1981). Ostrovskii, and Rosenblatt (1996) for contractions
Motivated by Euler’s formal approach to Fourier in Hilbert space, and by its extension by Jones
series, Wintner (1945) studied the existence of sol- et al. (1998) to certain isometries in Lp, Avigad
utions g  L1 to the coboundary equation f ¼ and Rute (2015) obtained the following
g  g ∘ θ, where θ is as in Theorem 1 and f  L2 p-variation theorem in uniformly convex spaces.
464 Operator Ergodic Theory

Theorem 15 Let X be isomorphic to a uniformly A nonreflexive Banach space with all contrac-
convex Banach space. Then there exists p  2 tions means ergodic was exhibited by Fonf
such that for every power-bounded T satisfying et al. (2010).
infnkT nxk  B kxk for some B > 0 and every For any power-bounded T we have (using
x  X, there exists C (depending on X, p, and T) Proposition 6 for the second inclusion)
such that for every increasing sequence {nk} we
have n1
ðI T ÞX  x  X : sup Tkx < 1
n
1 k¼0
p
Mnkþ1  Mnk x  C kxkp 8x  X:  ðI  T ÞX, ð4Þ
k¼1

Kohlenbach and Leuştean (2009) used proof Theorem 14 yields equality in the first inclu-
theory to obtain a quantitative proof of the mean sion when X is reflexive. Fonf et al. (2011) proved
ergodic theorem of G. Birkhoff (1939). that if X has a basis and for every power-bounded
Let T be invertible on X reflexive with T there is equality in the first inclusion of (4), then
1 ‘þn k X is reflexive.
supn  ℤkT nk<1. Then n k¼‘ T x  Ex ! 0 Bermúdez et al. (2000) studied operators such
for any ‘  ℤ. Motivated by Cotlar’s (1955) that T nx/n ! 0 implies that Mn(T )x converges
a.e. convergence of the ergodic Hilbert transform (a “local” mean ergodicity) and proved that if T k
of a probability-preserving transformation and by has this property, so does T. Radjavi et al. (2003)
Petersen (1983), Campbell (1986) used the spec- studied local ergodic properties for T which is not
tral theorem and the unitary dilation theorem to Cesàro bounded. Their main result is:
prove the following.
Theorem 17 Let T be a compact operator on a
Theorem 16 Let T be a contraction on a complex (real or complex) Banach space X, and let x  X.
Hilbert space ℋ. Then for every x  ℋ, the limit (i) If n1kT nxk ! 0 and supnkMnxk < 1, then
Hx ¼ lim n!1 nk¼1 1k T k x T k x exists, and Mnx converges in norm to a fixed point of T. (ii) If
H is a bounded linear operator on ℋ for some subsequence supk kT nk xk < 1, then Mnx
Berkson and Gillespie (1987) proved that if converges.
T is invertible on Lp, 1 < p < 1, with
n 1 k
supn  ℤkT nk < 1, then k
k¼1 k T x T x
converges for every x  Lp. Rates of Convergence
Sucheston (1976) asked whether a Banach
space, in which every contraction (or every Let T be mean ergodic. When T is uniformly
power-bounded operator) is mean ergodic, must ergodic, (2) yields that S, the restirction of T to
be reflexive. Building upon a partial result of Y≔ðI  T ÞX, satisfies kMn(S)k ! 0. For large n,
Zaharopol (1986), Emel’yanov (1997) proved we have that I  Mn(S) is invertible on Y, so I – S is
that a Banach lattice is reflexive if (and only if) invertible, i.e., I – T is invertible on ðI  T ÞX, so
every power-bounded operator is mean (I-T)X is closed, and kMn Ek ¼ O 1n : When T is
ergodic. Fonf et al. (2001) proved that if a not uniformly ergodic, by Theorem 13 there is no
Banach space has a basis and all power- uniform rate of convergence, so we may find
bounded operators are mean ergodic, then the x  X with kMnx  Exk ! 0 arbitrarily slowly.
space must be reflexive; additional equivalent However, Butzer and Westphal (1971) observed
conditions were obtained by Farkas and Kreidler that kMn xk ¼ o 1n implies x ¼ 0, so the fastest
(2021). Since a basis is assumed, this result does possible rate for x  ðI  T ÞX is O 1n , when
not include Emel’yanov’s. An analogue for C0- x  (I  T )X; by Theorem 14, if X is reflexive,
semigroups was proved by Mugnolo (2004). this rate is attained only when x  (I  T )X.
Operator Ergodic Theory 465

For a unitary operator U on a complex Hilbert ðI  T Þa X ¼ ðI  T ÞX, but if T is not uniformly

space ℋ, the spectral theorem entails a strong ergodic, the linear manifolds (I  T )αX are all
connection between the von Neumann ergodic different, and proved their following
theorem and the spectral measure of U. This con- characterization.
nection was used to study conditions on sf, the
spectral measure of f  ℋ with respect to U, Theorem 19 Let T be a mean ergodic contrac-
which yield rates for the convergence of Mn f. In tion on a Banach space X and 0 < α < 1. Then
this context, we mention the works of Gaposhkin y  (I  T )αX if and only if the series 1 1 j
j¼1 j1a T y
(1998) and of Kachurovskiǐ and Podvigin (2016),
converges strongly. Thus kMnyk ¼ o(1/nα) for
which contain references to earlier works. We
y  (I  T )αX.
present a representative result from Kachurovskiǐ
They concluded that y  (I  T )αX for some
and Sedalishchev (2010) and Kachurovskiǐ and
α  (0, 1) implies convergence of the one-sided
Podvigin (2016).
ergodic Hilbert transform Hy ≔ 1
j

j¼1 j : Cuny
Ty

Theorem 18 Let f  ℋ with spectral measure sf (2009) proved for positive contractions of a fixed
defined on(π, π], and assume Ef ¼ 0. Then for Lp, p > 1, that norm convergence of Hf implies its
any a  (0, π] and for any n, we have almost everywhere convergence. Additional
properties of H and its domain of definition were
2 given by Cohen et al. (2010) and by Haase and
sinða=2Þ a a
kM n f k2  sf  , , Tomilov (2010). Using a functional calculus,
a=2 n n
Gomilko et al. (2011) proved:
where the constant is best possible. Put Sk ¼
sf  ak, ak and ΔSk ¼ Sk  Skþ1; then Theorem 20 If T is power-bounded and the
series 1
j

j¼1 j converges weakly, then kMnyk ¼

Ty

4 1
n1 o(1/ log n); this rate is optimal.
S  kM n f k2  Sn þ ðk þ 1Þ2 DSk : Halmos (1949b) proved that if T is the
p2 n n k¼1
Koopman operator of an invertible ergodic mea-
sure preserving transformation of a nonatomic
Let α  [0, 2). Then the rate kMn f k2  Bnα
holds if and only if for some A > 0 we have probability space, there exists f  ðI  T ÞL2 for
1 Tj f
sf(δ, δ]  Aδα for every δ  (0, π]. which j¼1 j does not converge in L2-norm.
Butzer and Westphal (1972) obtained for T Cuny (2009) proved that if that series converges
power-bounded mean ergodic, but not uniformly in norm (even without invertibility of the transfor-
ergodic, the following approximation estimate of mation), then it converges a.e.
Let T be invertible on L p, 1 < p < 1, with
Mny for y  ðI  T ÞX : There exists B > 0 such that
supn  ℤkT nk < 1; Cuny (2010) used the spectral
1
kMn yk  B inf x  ðIT ÞX k y  x k þ nþ1 ðI  T Þ1 x :
integration of Berkson and Gillespie (1987) to
Nasri-Roudsari et al. (1995) proved the optimality obtain some conditions for the Lp convergence
of this type of estimates. of 1 1
j¼1 j1a T y, 0  a < 1.
j
Let T be power-bounded on a Banach space
X (so T is a contraction in the equivalent norm
jkxkj ≔ supn0kT nx k). For 0 < α < 1, Derriennic
Uniform Ergodic Theorems
and Lin (2001) used the power series ð1  tÞa ¼
ðaÞ
1 1 j¼1 aj t , 0  t  1, to define the fractional
j The convergence of the averages of a mean ergo-
ðaÞ j
power ðI  T Þa ¼ I  1 j¼1 aj T (fractional dic operator need not be in operator norm, even in
powers of generators of continuous one-parameter reflexive spaces; this is the case of the L2
semigroups had been defined differently more Koopman operator of an ergodic invertible
than 40 years earlier). They proved that probability-preserving transformation on [0, 1]
466 Operator Ergodic Theory

with Lebesgue measure. When the averages with TTj ¼ TjT ¼ ljTj, TjTi ¼ 0 for i 6¼ j, and TjS ¼
Mn(T) converge in operator norm, we say that STj ¼ 0.
T is uniformly ergodic. When T is as in Theorem 21, it is implied in the
Fonf et al. (2001) proved that if X is a Banach proof that its peripheral spectrum sðT Þ \  con-
space with basis such that every power-bounded sists only of (finitely many) eigenvalues; this fol-
operator is uniformly ergodic, then X is finite- lows from Theorem 22 below, since lT is quasi-
dimensional. Thus conditions for uniform ergo- compact, hence uniformly ergodic, for every
dicity of an operator, with or without prior l  .
assumption of mean ergodicity, are of interest. Brunel and Revuz (1974) observed that T is
We have already noted that if T is uniformly quasi-compact if and only if T ¼ S þ Q, with
ergodic, then Y ≔ (I  T )X is closed and I  T is spectral radius r(S) < 1 and Q with finite-
invertible on Y. When X is over ℂ, this yields that dimensional range.
for l near 1 the restriction of lI  T to Y is It follows from the work of Eberlein (1949)
invertible. The decomposition (2) yields that that if T is power-bounded and a convex power-
those l are in the resolvent r(T ); thus, if series S ≔ 1 k¼1 ak T ak  0,
k
k ak ¼ 1, is
1  s(T ), then it is isolated in the spectrum. (weakly) quasi-compact, then T is mean ergodic,
See Theorem 22 below for additional uniformly ergodic when S is quasi-compact; this
information. extends the results of Yosida and Kakutani (1938,
In their study of Markov chains, Kryloff and 1941).
Bogoliouboff (1937) introduced a new concept, When T is uniformly ergodic, (I  T)X is
which applies also to general operators. An oper- closed, so when T is power-bounded we have
ator T on a Banach space is called quasi-compact kMn  Ek ¼ O(1/n). Yoshimoto (1993) studied
if there exist n  1 and a compact operator Q, such the uniform rate of convergence when T is quasi-
that kT n  Qk < 1. Yosida and Kakutani (1938) compact but not power-bounded, and usual aver-
proved that a power-bounded weakly quasi- ages are replaced by C(α) averages; in Yoshimoto
compact operator (Q weakly compact) is mean (1996), he studied the power-bounded case.
ergodic. Yosida and Kakutani (1941) observed As an application of his operational calculus,
that T is (weakly) quasi-compact if and only if Dunford (1943a) obtained the following.
there exists a sequence of (weakly) compact oper-
ators Qn such that kT n  Qnk ! 0; they showed Theorem 22 Let T on a complex Banach space
that the condition of Doeblin (1937) on Markov X satisfy n1kT nk ! 0. Then the following are
operators implies quasi-compactness and proved equivalent:
the following theorem, of which Theorem 4 is a
special case. (i) T is uniformly ergodic.
(ii) Either 1  r(T ), or 1 is a simple pole of the
Theorem 21 Let T be a power-bounded quasi- resolvent R(l, T ) ≔ (lI  T)1.
compact operator on a complex Banach space (iii) (I  T)2X is closed.
X. Then T is uniformly ergodic. More precisely,
T has only finitely many eigenvalues of modulus Note that even for T power-bounded, 1 isolated
one (if any), say lj, j ¼ 1, . . ., k, each of finite in s(T ) does not imply that 1 is a pole, e.g.,
multiplicity, and there exist projections Tj, j ¼ 1, T ¼ I  V on L2[0, 1] (V the Volterra operator),
. . ., k and a quasi-compact S with kSn k  ð1þe M which is power-bounded by Allan (1997).
Þn
Burlando (1997) proved that given j  ℕ, if
for some ε > 0, such that T on a complex X satisfies njkT nk ! 0 and 1 is
k
a pole of the resolvent, then nj n1k¼0 T
k
con-
Tn ¼ lnj T j þ Sn n ¼ 1, 2, . . . verges in operator norm, and identified the limit;
j¼1 other conditions are also presented.
Operator Ergodic Theory 467

If T is uniformly ergodic on X real or complex, Theorem 25 Let T be an operator on a Banach

then n1kT nk ! 0 and (I  T )X is necessarily space X. Then kT nk ! 0 if and only if for some
closed. Lin (1974a) proved the following (which (every) p  ½1, 1Þ, 1 n p
n¼0 kT xk < 1 for every
is easy, using (2), when T is already mean ergodic, x  X.
e.g., X is reflexive and T power-bounded). Recall that over ℂ we have kT nk ! 0 if and
only if T has spectral radius r(T ) < 1. Glück
Theorem 23 Let T on a (real or complex) Banach (2015) obtained the following improvement of
space X satisfy n1kT nk ! 0. Then T is uniformly Weiss’ main result (in which p is constant).
ergodic if and only if (I  T )X is closed.
As a corollary, it is shown that also in the real Theorem 26 Let T be a bounded operator on a
case, a power-bounded quasi-compact operator is complex Banach space X. Assume that for every
uniformly ergodic. Uniform ergodicity then x  X and ’  X  there exists p  [1, 1) such
implies that I  T is invertible on (I  T )X, that 1 p
n¼0 j’ðT xÞj < 1: Then T has spectral
n

hence (I  T )kX ¼ (I  T )X is closed for any radius r(T ) < 1.

k  1. Mbekhta and Zemánek (1993) showed that When a power-bounded T is uniformly ergo-
when X is complex, one can replace condition (iii) dic, we have equality in both inclusions of (4). It
in Theorem 22 by (iii)’: (I  T)kX is closed for was observed in Fonf et al. (1996) that a power-
some k  1. Grabiner and Zemánek (2002) proved bounded T is uniformly ergodic if and only if
that if T is Cesàro bounded and (I  T )kX is closed n1 k
x  X : supn k¼0 T x <1 is closed (and
for some k  2, then X ¼ F (T ) (I  T )X.
Becker (2011) proved that if T is uniformly then equals ðI  T ÞXÞ: Since power-bounded
ergodic, then the one-sided ergodic Hilbert trans- operators on reflexive spaces need not be uni-
form Hx ≔ 1 1 k
k¼1 k T x converges for every x 
formly ergodic, the equality
(I  T )X. In Becker (2005), she proved that when
n1
kT nk ¼ O(nα) for some 0  α < 1, uniform
ðI T ÞX ¼ x  X : sup Tkx < 1 ð5Þ
ergodicity is equivalent to the set of convergence n
k¼0
of H being closed.
Extending some results of Hille (1945), does not imply uniform ergodicity. Fonf et al.
Yoshimoto (1998) proved the following. (1996) studied the equality (5) and proved that if
X is a separable real Banach space which does
Theorem 24 Let T on a complex Banach not contain an infinite-dimensional closed sub-
space X satisfy n 1 kT n k ! 0. Then T is uni- space isomorphic to a dual Banach space (e.g.,
formly ergodic with limit E if and only if c0, or more generally X  ¼ ‘1), then (5) implies
k(l  1)R(l, T )  E k ! 0 as l ! 1 + . uniform ergodicity.
Improving upon some results of Koliha Lin (1978) proved that a uniformly ergodic
(1974), Luecke (1977) proved: (i) T n converges positive operator on a Banach lattice, with F (T )
in norm if and only if T is uniformly ergodic and finite-dimensional, is quasi-compact; this is not
kT n(I  T )k ! 0. (ii) when X is complex, T n true in general (e.g., T ¼  I ).
converges in norm if and only if T is uniformly Lotz (1968) proved that when T is positive
ergodic and sðT Þ \   f1g: When T n converges power-bounded on a complex Banach lattice, if
uniformly, the rate is exponential: kT n  Ek  Mrn l  sðT Þ \ , then all powers lk are in s(T ).
for some M > 0 and r  (0, 1). Hence if T is a positive power-bounded quasi-
Weiss (1989) proved the following discrete- compact operator with r(T ) ¼ 1, there exists
time version of the Datko (1970) and Pazy d  1 such that every l  sðT Þ \  (which is an
(1972) theorem about exponential stability of eigenvalue by Theorem 21) is a dth root of unity.
C0-semigroups. We thus obtain the following.
468 Operator Ergodic Theory

Proposition 27 A positive power-bounded T on a transformation is weakly mixing if hT nf, gi is

complex Banach lattice is quasi-compact if and strong Cesàro convergent to hEf, gi for every f,
only if for some integer d  1 and a finite- g  L2(m). If T nf ! Ef weakly for every f  L2,
dimensional projection E we have kT nd  Ek ! 0. the transformation is called mixing. The category
Proposition 27 was proved for contractions by theorems of Halmos (1944) and Rokhlin (1948)
Bartoszek (1988). show that there are weakly mixing transforma-
Lotz (1981) studied uniformly ergodic tions which are not mixing; operator theoretic
Markov-Feller operators on C(K ), and in Lotz analogues were proved by Eisner and Serény
(1985) he studied uniformly ergodic positive (2008). Chacon (1969) was the first to construct
operators on L1. Groh (1983) studied the periph- a weakly mixing transformation which is not
eral spectrum of uniformly ergodic positive oper- mixing.
ators on C-algebras. L. Jones and Lin (1976) proved the following:

Theorem 28 Let T be power-bounded on a (real

Strong Cesàro Convergence or complex) Banach space X. The following are
equivalent for x  X:
Convergence of the averages 1n nk¼1 ak ! a of a
sequence {ak}k1 can sometimes be strengthened (i) 1n nk¼1 j T k x, ’ j ! 0 for every ’  X .
to 1n nk¼1 jak  aj ! 0 (strong Cesàro conver- (ii) sup’  X ,k’k1 1n nk¼1 j T k x, ’ j ! 0:
gence). It seems to have first appeared in Hardy 1 n
and Littlewood (1913). Koopman and von Neu- (iii) n
kj
j¼1 T x ! 0 for every increasing kj
mann (1932) essentially proved that a bounded with sup kj/j < 1.
sequence {ak} is strongly Cesàro convergent to a
if and only if there exists a set J  ℕ of density They proved that if T ni x ! 0 weakly for some
zero, i.e., satisfying n1 j J \ {1, 2, . . ., n} j ! 0, {ni}, then the conditions of Theorem 28 hold,
such that lim J∌k!1 ak ¼ a: and when X  is separable (but not in general)
A restatement of the theorem of Koopman and such convergence is also necessary. The relation-
von Neumann (1932) is: Let T be a unitary oper- ship of (i) and (iii) was discussed earlier in
ator on a complex Hilbert space ℋ. Then L. Jones (1971) and Nagel (1974). Additional
conditions were studied by Hiai (1978). When
1 n T is power-bounded on a reflexive space X,
j hx, yi j! 0 8y  ℋ
n k¼1 (i) implies the existence of a set J  ℕ of density
if and only if hx, zi ¼ 0 ð6Þ zero such that weak limit of fT n xgJ∌ n!1 is zero:
whenever T  z ¼ lz, j l j¼ 1: By (iii), we can assume X separable (spanned by
the orbit of x), and then by reflexivity X  is
It follows from the work of Foguel (1963) that separable.
if T is a contraction of a complex Hilbert space ℋ Zsidó (2007) proved that if we replace {T kx}
and 0 6¼ x  ℋ satisfies limn|hT nx, xi| ¼ k xk2, by an arbitrary-bounded sequence {xk}  X,
then Tx ¼ lx for some l   . This result was (ii) and (iii) of Theorem 28 are still equivalent,
explicitly stated and proved (differently) for con- but are not implied by (i). Condition (iii) for
traction C0-semigroups in ℋ by Goldstein sequences in Hilbert space was studied by Berend
(1993), and extended to Banach spaces by Gold- and Bergelson (1986). Kunszenti-Kovács (2015)
stein and Nagy (1995). showed that when T is a contraction on a Hilbert
An ergodic probability-preserving transforma- space ℋ and x  ℋ satisfies (i) of Theorem 28,
tion whose Koopman operator T on the complex then for every nonconstant polynomial q mapping
L2 has no unimodular eigenvalues different from ℕ into itself there is a set J of density zero such
1 is called weakly mixing; equivalently, by (6), the that weak- limJ∌j!1Tq( j )x ¼ 0.
Operator Ergodic Theory 469

When X is a complex Banach space, we would noted there that for general contractions the “if”
like to extend the characterization of condition part of (8) need not hold.
(i) of Theorem 28, given in (6) for unitary operators In connection with Theorem 28(iii), Eisner and
on Hilbert spaces. Extending the work of Jacobs Müller (2021) studied particular subsequences
(1957), deLeeuw and Glicksberg (1961) studied along which the averages of any weakly almost
weakly almost periodic semigroups of operators periodic operator, which has no unimodular
(the image of every vector under the semigroup is eigenvalues, converge in norm.
weakly conditionally compact), which are neces- DeLaubenfels and Vũ (1996) proved the fol-
sarily bounded. A very special case of their results lowing (see also Theorem 45 below).
is the following decomposition.
Theorem 32 Let T be power-bounded on X over ℂ,
Theorem 29 Let T be weakly almost periodic on with sðT Þ \  countable. Then kT nxk ! 0 if (and
a complex Banach space X. Then only if) hx, ci ¼ 0 whenever T c ¼ lc, j l j ¼ 1.
Extending the result of Krengel (1972) for
X ¼ spanfy : Ty¼ ly, jlj¼ 1g unitary operators, Müller and Tomilov
(2010) proved:
z : 0  weak  closurefT n zgn0 : ð7Þ
Theorem 33 Let T be power-bounded on a com-
Moreover, 0  weak - closure {T nx}n  0 if and plex Hilbert space ℋ with sðT Þ \  infinite and
only if hx, ’i ¼ 0 for every ’  X  with T ’ ¼ no unimodular eigenvalues. Then there exists a
l’, j l j ¼ 1. dense set of vectors x  ℋ such that for some
Since, for T power-bounded, kT nzk ! 0 if (and increasing {nk} we have hT nk x, T nj xi ¼ 0 when
only if) kT nk zk ! 0 along some subsequence j 6¼ k.
{nk}, we obtain: When T is power-bounded on X separable
(over ℂ), the set of unimodular eigenvalues
Corollary 30 Let T be almost periodic on a com- sp ðT Þ \  is countable, by Jamison (1965), but
plex Banach space X. Then sp(T ) may be uncountable (as the left shift on ‘2
shows).
X ¼ spanfy : Ty¼ ly, jlj¼ 1g

fz : kT n zk ! 0g:
Weak Stability and Mixing
Proposition 31 Let T be weakly almost periodic
Recall that an ergodic probability-preserving
on a complex Banach space X. Then
transformation is called mixing if its Koopman
n operator T on L2 satisfies T n ! E in the weak
1
sup j T k x, ’ j ! 0 operator topology. We therefore call a power-
k’k1 n k¼1 bounded T on a Banach space X weakly stable if
ð8Þ
if and only if hx, ci ¼ 0 for some E (necessarily a projection) T n ! E in
whenever T  c ¼ lc, j l j¼ 1: the weak operator topology. Weak stability
implies power-boundedness and mean ergodicity,
L. Jones and Lin (1980) proved (8) for so when X is over ℂ the spectral radius of T does
T power-bounded on complex X, assuming only not exceed 1.
that {T nx} is weak-* sequentially compact in X
(e.g., X does not contain an isomorphic copy of Theorem 34 (Foguel (1963)). Let T be a con-
‘1). Their proof uses the result for isometries in traction on a Hilbert space ℋ and x  ℋ. Then
Hilbert spaces, proved in Krengel (1972), but does T nx ! 0 weakly if and only if hT nx, xi ! 0. Hence
not use the DeLeeuw-Glicksberg theory. It is also T nx ! 0 weakly if and only if T nx ! 0 weakly.
470 Operator Ergodic Theory

The following result of Foguel (1963) shows every increasing {kj}; the converse implication
that weak stability of contractions in ℋ is a prob- always holds. Theorem 37 says that any contrac-
lem only for unitary operators. tion on a Hilbert space has the BH property.
A weakly stable T on X, with T n ! E in the
Theorem 35 Let T be a contraction on a Hilbert weak operator topology, has the BH property if
space ℋ and let K ≔fy  ℋ: kT nyk ¼ kT nyk ¼ 1 n kj
 Ex ! 0 for every increasing {kj}
j¼1 T x
k y k 8 n  1}. Then K is invariant under T and T ,
n

the restriction T jK is unitary, and limhT nz, zi ¼ 0 and every x  X. Theorem 37 raises the question
for any z ⊥ K . whether any weakly stable operator on a Banach
Horowitz (1969) proved that if in Theorem 35 space has the BH property.
T is normal, then kT nzk ! 0 for z ⊥ K . Akcoglu and Sucheston (1972) proved that
In the following, Badea and Müller (2009) weakly stable contractions on L1 have the BH
showed that in general the convergence to 0 of property. Millet (1976) proved that a weakly
hT nx, xi can be arbitrarily slow. stable positive operator on L1 has the BH prop-
erty. On the other hand, Akcoglu et al. (1974)
Theorem 36 Let T on a complex Hilbert space constructed a topological dynamical system
ℋ satisfy T nx ! 0 weakly for every x  ℋ (so T (K, t) such that its Koopman operator T on
is power-bounded), and assume 1  s(T ). Then C(K ) is weakly stable, but for some
for every positive sequence rn ! 0 with supn x  C(K ) (9) fails. Lefèvre and Matheron
rn < 1 there exists x  ℋ with kx k ¼ 1 such (2016) proved that when K is a compact metric
that RehT nx, xi > rn. space, all contractions have the BH property if
The following theorem, proved independently and only if K has only finitely many accumula-
by Jones and Kuftinec (1971) and by Akcoglu and tion points.
Sucheston (1972), was first proved by Blum and Akcoglu and Sucheston (1975), and then Bel-
Hanson (1960) for the L2 Koopman operator of a low (1975), proved that any weakly stable posi-
mixing probability-preserving transformation. tive contraction on Lp, 1 < p < 1, has the BH
property. Müller and Tomilov (2007) proved that
Theorem 37 Let T be a contraction in a Hilbert when T is a contraction of ‘p, 1  p < 1, the
space ℋ and x  ℋ. Then equivalence (9) holds; they also showed that a
power-bounded T on a Hilbert space may be
weakly stable without possessing the BH prop-
T n x ! 0 weakly if and only if
erty. Satō (1980) proved that if T is positive and
1 n
T kj x ! 0 ð9Þ weakly stable on Lp of a finite measure space,
n j¼1
1 < p < 1, and for some p < p1 < 1 T is also
for every increasing kj :
power-bounded in Lp1 , then T has the BH property
on Lp.
Krengel (1971) and Conze (1973) constructed
Lefèvre et al. (2016) studied the BH property
increasing sequences {kj} such that for every
in terms of general sequences {xn}  X converg-
ergodic mpt θ there exists f  L1 for which
1 n kj
ing weakly to zero. They have additional exam-
n j¼1 f∘y does not converge a.e. ples of real Banach spaces in which every
Berend and Bergelson (1986) abstracted The- contraction has the BH property; see also
orem 37, by proving that for {xn}  ℋ bounded, Grivaux (2019). Azizov and Chilin (2017) pro-
1 n
xn ! 0 weakly if and only if n j¼1 xkj ! 0 for ved the BH property for contractions on certain
separable Banach lattices.
every increasing {kj}.
Eisner (2010) proved the following discrete
We say that a power-bounded T on a Banach
analogue of a result of Chill and Tomilov (2007);
space X has the Blum-Hanson (BH) property if
when its condition (10) holds for any x  X and
1 n
Tnx ! 0 weakly implies n
kj
j¼1 T x ! 0 for ’  X , we obtain weak stability.
Operator Ergodic Theory 471

Theorem 38 Let T on a complex Banach space (i) limn ! 1kT nxk

have spectral radius r(T )  1. If x  X and ¼ sup {| hx, ’i| : ’  \n  1 T nB}.
’  X  satisfy (ii) lim n!1 1n nk¼1 T k x
¼ supfjhx, ’ij: ’  B \ F ðT  Þg:
2p
2
lim ðr 1Þ j’ R reit , T x jdt ¼ 0, ð10Þ
r!1þ 0 Hence a mean ergodic contraction is stable if
and only if \n1T nB  F (T ).
then ’(T nx) ! 0. Rokhlin (1961) called a probability-preserving
Leonov (1961), independently, proved transformation θ on (Ω, S, m) exact if
Browder’s Theorem 14 for weakly stable isometries \n1{θn(A) : A  S} is trivial modulo m. Theo-
in Hilbert space and showed that if T is the rem 40 yields that θ is exact if and only if the dual
Koopman operator of an ergodic probability- of its Koopman operator is stable in L2, with limit
preserving transformation on (Ω, S, m), then for Ef ¼ f dm.
any nontrivial A  S the function f ≔ 1A  m(A)
n
k¼1 T f 2 ! 1
k
satisfies (so f is not a Lemma 41 A mean ergodic operator T is
coboundary); see El Abdalaoui et al. (2010) for a (weakly) stable if and only if it is power-bounded
related result. and T n(I  T )x ! 0 (weakly) for every x  X.
Lemma 41 is a direct consequence of (2). Here is
an example for Lemma 41: Let S be power-bounded
Stability on a (real or complex) Banach space, and for
0 < α < 1 put T ¼ (1  α)I þ αS. Since I  T ¼
A power-bounded operator T on a Banach space
α(I  S), by Foguel and Weiss (1973) we have
X is called stable if T n converges in the strong p
kT n ðI T Þk  K= n ! 0; thus, T is stable
operator topology, necessarily to a projection on
when S is mean ergodic. In the complex case, any
F (T ). A weakly stable T is stable if and only if it is
S with r(S) ¼ 1 is the uniform limit of the uniformly
almost periodic. Corollary 30 yields:
stable operators Sn ≔ 1 1n S:
Katznelson and Tzafriri (1986) proved the fol-
Proposition 39 An operator T on a complex
lowing deep result.
Banach space is stable if and only if it is almost
periodic with no unimodular eigenvalues different
Theorem 42 Let T be a power-bounded
from 1.
operator on a complex Banach space. Then
Since Koopman operators of probability-
kT n(I  T )k ! 0 if and only if sðT Þ \   f1g:
preserving transformations are isometries in Lp,
The case s(T ) ¼ {1} was proved by Esterle
1  p < 1, they are never stable, unless T ¼ I.
(1983). Allan et al. (1987) showed that sðT Þ \
However, in the noninvertible case the dual T  of
  f1g implies the norm convergence of n0 ζ
a Koopman operator may be stable, e.g., n
(T n  T nþ1) for any 1 6¼ z   . Allan and
T induced by θt ¼ 2t mod 1 on [0, 1), by Horo-
Ransford (1989) proved that Theorem 42
witz (1968); this example shows that a contraction
is equivalent to an old result of Gelfand
(i.e., T ) on ℋ may be stable while its dual is not.
(1941), which says that if T is invertible with
A simpler example is the shift S(a1, a2, . . .) ¼
supn  ℤkT nk < 1 and s(T ) ¼ {1}, then T ¼ I.
(a2, a3, . . .) on ‘2(ℕ), which is stable, while S is
A different proof of Theorem 42 was given by Vũ
not, being an isometry.
(1992a). For additional historical information, we
In his study of stability of Markov operators,
refer to Zemánek (1994). While Gelfand’s result is
Derriennic (1976) proved the following.
valid also if Mn(T ) and Mn(T 1) are bounded, by
Drissi and Zemánek (2000), Tomilov and
Theorem 40 Let T be a contraction of a Banach
Zemánek (2004) showed that Theorem 42 fails if
space X, and let B be the unit ball of the dual
power-boundedness is replaced by Cesàro
space X. Then for any x  X we have:
472 Operator Ergodic Theory

boundedness (or even mean ergodicity); Léka 35 and 34. In Sect. II.6 of their book, Sz-Nagy and
(2010) has an example with s(T ) ¼ {1}. For Foiaş (1967) proved the following.
additional information and new proofs see the
survey of Batty and Seifert (2022). Theorem 44 Let T be a completely nonunitary
Allan et al. (1987) proved the following. contraction on a complex Hilbert space ℋ. If
sðT Þ \  has zero Haar measure (on  ), then
Theorem 43 Let f ðzÞ ¼ 1 k¼0 ak z
k
with kT nxk ! 0 and kT nxk ! 0 for any x  ℋ.
k0 k | ak | < 1, and let T be power-bounded Following Kérchy and van Neerven (1997),
on a complex Banach space X. Then kT nf(T)k ! 0 Mustafayev (2014) studied conditions on the
if and only if f(l) ¼ 0 for every l  sðT Þ \ . local spectrum which imply local stability. His
The conclusion of Theorem 43 was proved by work yields a partial extension of Theorem 44 to
Esterle et al. (1990) for every f in the disk algebra power-bounded operators, by restricting the null-
when T is a contraction of a Hilbert space. When sets containing sðT Þ \ .
X is a Hilbert space, Léka (2009) extended Theo- For the general case, Arendt and Batty (1988)
rem 43 to f with k|ak| < 1. Huang (1995) proved the following; alternate proofs are indi-
proved that if kT nf(T)k ! 0 for some f(z) with cated in Batty (1994). Lyubich and Vũ (1988)
k|ak| < 1, then sðT Þ \  has measure zero. proved the continuous case, and their proof was
A continuous time version of the Katznelson- adapted to the discrete case in Eisner (2010). See
Tzafriri theorem, for bounded C0-semigroups, also Batty and Vũ (1990).
was proved by Esterle et al. (1992) and by
Vũ (1992b). Theorem 45 Let T be a power-bounded operator
Following Kalton et al. (2004), see Theorem on a complex Banach space X. If sðT Þ \  is
67 below, Malinen et al. (2007) proved that for countable and T  has no unimodular eigenvalues,
any T, either lim infn(n þ 1)kT n(I  T)k ¼ 0, or then kT nxk ! 0 for x  X.
lim infn(n þ 1)kT n(I  T )k  e1. Dungey Eisner (2010) proved a discrete analogue of a
(2008) gave several necessary and sufficient con- resolvent condition of Tomilov (2001) which
ditions for kT n(I  T )k ¼ O(n1/2) in Theorem yields stability, and obtained as a corollary the
42. The rate 1/n is related to the Ritt resolvent following.
condition; see the section Resolvent conditions
Theorem 46 Let T be power-bounded on a com-
and growth of powers below. Seifert (2016)
plex Hilbert space ℋ and x  ℋ. Then kT nxk
obtained a rate in Theorem 42 using the growth
! 0 if and only if
of kR(eit, T)k as t ! 0. Cohen and Lin (2016)
studied the rate 1/nα, 0 < α  1, when X is a 2p
Hilbert space. Badea and Seifert (2017) obtained 2
lim R reit , T x dt ¼ 0
a condition on the numerical range which implies r!1þ 0

sðT Þ \   f1g . Ng and Seifert (2020) got the

full correspondence, when X is a Hilbert space, Lin (1998) used Theorem 42 to prove the
between the rate of growth of kR(eit, T )k as t ! 0 following.
and the rate of kT n(I  T )k ! 0.
Baskakov (2015) obtained a representation of Theorem 47 Let T be a positive power-bounded
T when sðT Þ \  is finite, an immediate corol-
n operator on a Banach lattice. Then kT n(I  T)k ! 0
p
lary of which is Theorem 42. if and only if inf n kT n ðI T Þk < 3:
A contraction T on a Hilbert space ℋ is called For contractions on Lp, 2 6¼ p  (1, 1), the
p p
completely nonunitary if its unitary subspace K , constant 3 can be replaced by a constant 3 <
defined in Theorem 35, is trivial. In that case, cp < 2, computed in Berend (1992). For L1, the
T nx ! 0 weakly for every x  ℋ, by Theorems constant is 2, by Foguel (1976).
Operator Ergodic Theory 473

Following Lasota et al. (1984), Sine (1991) contraction on a (real or complex) Hilbert space
defined a power-bounded operator T on a Banach ℋ and x  ℋ if and only if for every l   the
space X to be constrictive if there exists a compact averages 1n nj¼1 lkj converge.
set K  X such that dist(T nx, K ) ! 0 for every Motivated by a result of Brunel and Keane
kx k  1. When K ¼ {0}, we have stability. By the (1969), Bourgain (1988b) and Bourgain et al.
definition, if T is constrictive, then for each x  X (1989) proved the “return times theorem,” which
the orbit {T nx} is precompact, so T is almost provides increasing sequences {kj}, with lim
periodic, hence mean ergodic. Sine (1991) proved kj/j < 1, such that averaging any Koopman oper-
that a contraction T is constrictive if and only if ator on any Lp along {kj} converges a.e., hence also
T is strongly almost periodic and, in the decom- in Lp norm (1  p < 1). Applying it to all rotations
position (7), spanfy : Ty¼ ly, jlj¼ 1g is finite- of , we obtain, by Proposition 49, that for these
dimensional (and on its complement T is stable). sequences 1n nj¼1 T kj x converges in norm for any
Extending the result of Lasota et al. (1984) for L1, contraction T on a Hilbert space ℋ and x  ℋ.
Bartoszek (1988) proved the following. Extending a result of Furstenberg (1981) for
unitary operators (for which Bergelson (1987)
Theorem 48 Let T be a constrictive positive con- gave a nonspectral proof), Kunszenti-Kovács
traction of a Banach lattice. Then there exist r et al. (2011) proved the following.
positive unit vectors y1, . . ., yr and r positive
functionals ’1, . . ., ’r such that T permutes the Theorem 50 Let q be a nonconstant real polyno-
r mial with integer coefficients which maps ℕ to
j¼1 ’j ðxÞyj ! 0 for every
n
yj, and T x
itself. Then the limit cq ðlÞ≔ lim n!1 1n k¼1 lqðkÞ
x. Hence, for some 1  d  r !, T d is stable.
exists for l   and is zero for l not a root of unity.
For every contraction T on a Hilbert space ℋ
Averaging Along Subsequences and x  ℋ, the averages 1n nk¼1 T qðkÞ x converge
in norm.
The Blum-Hanson Theorem 37 and Theorem 28 If ℋ is over ℂ, then the limit is
l:∃jlj ¼1 cq ðlÞEl x, where El x≔ lim n n k¼0 lT x
1 n1 k
(iii) raise the question for which increasing sub-
sequences of positive integers {kj} the averages is the orthogonal projection on the eigenspace
1 n kj
n j¼1 T converge strongly for contractions, or corresponding to l  .
power-bounded operators, on a Hilbert space ℋ. Given a contraction T on ℋ, the limit is a
Blum and Eisenberg (1974) proved a “uniform projection for every q with q(0) ¼ 0 if and only if
distribution” criterion on a sequence {kj} for the roots of unity different from 1 are not eigen-
strong convergence of 1n nj¼1 T kj to the ergodic values of T.
limit E(T) for every unitary operator; it yields also Proposition 49 does not apply to power-
convergence for any contraction on ℋ, by the bounded operators. Using a different method, ter
unitary dilation theorem. For contractions on ℋ, Elst and Müller (2017) proved:
the following general criterion, implied in
Furstenberg (1981) for unitary operators, was pre- Theorem 51 Let q be a nonconstant real polyno-
sented for the complex case in Rosenblatt (1994), mial with rational coefficients which maps ℕ to
in Rosenblatt and Wierdl (1995), and in Berend itself. If T is a power-bounded operator on Hilbert
et al. (2002). For the real case, one applies the space ℋ over ℂ such that roots of unity are not
complex case in the complexification of ℋ, as in eigenvalues, then 1n nk¼1 T qðkÞ x ! 0 for every
Michal and Wyman (1941). x  ℋ.
Eisner and Müller (2021) extended Theorem
Proposition 49 Let {kj} be an increasing 51 by proving that for a general real polynomial
sequence of positive integers. The averages q with [q(k)]  0 for k  ℕ, we have convergence
1 n
of 1n nk¼1 T ½qðkÞ x for every x  ℋ.
kj
n j¼1 T x converge in norm for every
474 Operator Ergodic Theory

Bourgain (1989) proved that for T the preservation of the limit (Silverman-Toeplitz the-
Koopman operator on Lp of an invertible orem). Such a matrix is called a regular summa-
probability-preserving transformation, we have bility matrix, or in short a regular matrix. The
a.e. convergence of 1n nk¼1 T qðkÞ f for f  Lp, Cesàro averaging corresponds to the matrix with
1 < p < 1, and q as in Theorem 51. This yields an,j ¼ 1/(n þ 1) for j  n and 0 for j > n. The first
Lp-norm convergence, also in L1. For general real to consider averaging of powers of a power-
polynomials, he proved a.e. convergence of bounded operator by regular matrices was
1 n ½qðkÞ
f for f  L1. L. Cohen (1940), who proved:
n k¼1 T
Let {pj} be the sequence of prime numbers in
increasing order. Extending earlier results of Theorem 53 Let A ¼ (an,j)n,j 0 be a regular
Bourgain (1988a) and of Wierdl (1988), Nair matrix such that
(1993) proved that given q a nonconstant polyno-
1
mial mapping ℕ to itself, for the Koopman oper-
lim jan,jþ1  an,j j ¼ 0 uniformly in n:
ator as above on Lp, 1 < p < 1, we have k!1
j¼k
a.e. convergence of 1 n T qðpj Þ f for f  L ,
n j¼1 p
hence in Lp norm. Applying Nair’s result to all If T is a power-bounded operator on a Banach
rotations of  and using Proposition 49, we obtain space X such that for every x  X the sequence
the following result, proved directly in Eisner and Bn x ≔ 1 j
j¼0 an,j T x is weakly conditionally com-
Lin (2018). pact, then for every x the sequence Bnx converges
to a fixed point.
Theorem 52 Let {pj} be the sequence of primes In particular, when all the elements of A above
in increasing order. Given a nonconstant polyno- are nonnegative with 1 j¼0 an,j ¼ 1 for every n,
mial q mapping ℕ to itself, for every contraction then the above convergence holds for every
on a (real or complex) Hilbert space ℋ and weakly almost periodic T.
x  ℋ, the averages 1 n T qðpj Þ x converge in
n j¼1
Eberlein (1984) looked at the Bn in Theorem 53
norm. as invariant integrals, in the sense of
Boshernitzan and Wierdl (1996) studied con- Eberlein (1949).
ditions on certain real sequences a(t), which tend Note that when the regular matrix A is triangu-
to +1 as t ! 1, such that for every Koopman lar (i.e., an,j ¼ 0 for j > n), the series
operator T of an invertible probability-preserving Bn x≔ 1 j¼0 an,j T x is a finite sum, so is defined
j

transformation, the averages 1n nk¼1 T ½aðkÞ f con- for any operator T, and one can consider the con-
verge in L2-norm. vergence of Bnx even for T not power-bounded.
Let {wk}k0 be a sequence of nonnegative
numbers with w0 > 0 and 1 k¼0 wk ¼ 1 , and
General Averaging Methods put W n ¼ nk¼0 wk. The matrix A defined by an,j ¼
wj/Wn for j  n and zero for j > n is regular; it
Since the averages of a convergent sequence con- yields the weighted averages with weights {wk}.
verge to the same limit, it is natural to apply more Lin et al. (1999) proved:
general averaging (summability) methods which
preserve convergence and limits, when conver- Theorem 54 Let {wk} be nonnegative with
gence of the arithmetic means fails. w0 > 0 and 1 k¼0 wk ¼ 1: The weighted aver-
A matrix A ¼ (an, j)n, j0 maps ‘1 into itself, by 1 n
ages W n k¼0 wk T k x !Ex for every mean ergodic
ðAbÞn ¼ 1 j¼0 an,j bj for b ≔ (b0, b1, . . .)  ‘1, T on a Banach space X and x  X if and only if
if and only if supn 1 j¼0 jan,j j < 1; if, in addition,
lim n!1 W1n wn þ n1
k¼0 jwkþ1  wk j ! 0:
the matrix A satisfies limn!1 an,j ¼ 0 for every
j and lim n!1 1 j¼0 an,j ¼ 1, then it maps the
The latter condition is satisfied when {wk} is
space c of convergent sequences into itself, with nondecreasing and wn/Wn tends to 0.
Operator Ergodic Theory 475

For α > 0, the Cesàro-α averages of an Hille (1945) applied his results to the operator
operator T, denoted Can ðT Þ, are defined as norm topology. The converse of the first statement
follows: For β >  1, let Ab0 ¼ 1 and in Theorem 55 was proved for the operator norm
Abn ≔ðb þ 1Þ  ðb þ nÞ=n! for n > 0, and let topology by Badiozzaman and Thorpe (1992) and
n reproved by Ed-dari (2003); put together, the fol-
Can ðT Þ ¼ j¼0 Anj =An
a1 a
T j . The corresponding
lowing was obtained (proved by Yoshimoto
matrix an,j ≔Aa1
nj =An for
a
0  j  n and 0 for (1998) for 0 < α  1).
j > n is regular and satisfies the conditions of
Theorem 53. An operator T on X is called (weakly)
Theorem 56 Let α > 0. The operator T is uni-
(C, α) ergodic if Can ðT Þx converges (weakly) as
formly (C, α) ergodic if and only if it is uniformly
n ! 1, for every x  X. When supn Can ðT Þ <
Abel ergodic and nαkT nk ! 0.
1, we say that T is (C, α) bounded. Note that (C, 1)
An analogous result for the strong operator
(weak) ergodicity is (weak) mean ergodicity.
topology was proved in Ed-dari (2004).
Li et al. (2008) proved that for every 0 < α < 1
Lin et al. (2015) proved that T is uniformly Abel
there exists a positive T on some L1 such that T is
ergodic if and only if for some (all) 0 < r < 1,
(C, β) bounded for β > α, but is not (C, α)
(Ar(T ))n converges uniformly as n ! 1;
bounded.
Kozitsky et al. (2013) gave a spectral condition
Let T satisfy lim supn!1kT nk1/n  1 (spectral
for this property.
radius r(T )  1 when X is over ℂ). Then for
Li et al. (2008) showed that in any infinite-
0 < r < 1, the series 1 k k
k¼0 r T converges in dimensional Banach space, Abel boundedness
operator norm, and we call Ar ¼
does not imply Cesàro boundedness. However,
Ar ðT Þ≔ð1 r Þ 1 k¼0 r k k
T the Abel averages of T. Émilion (1985) proved that any Abel bounded
For any sequence 0 < rn " 1, the matrix an,j ¼ positive operator on a Banach lattice is Cesàro
ð1 r n Þr jn is regular and satisfies the assumptions bounded, and when the lattice is reflexive it is
of Theorem 53. An operator T as above is called mean ergodic.
Abel bounded if sup0<r<1kAr(T )k < 1, and Abel Improving the extension of Gelfand’s theorem
ergodic if Ex≔ lim r!1 Ar ðT Þx exists for every x; by Mbekhta and Zemánek (1993), Grobler and
in the latter case Ex  F (T ), and we have X¼ Huijsmans (1995) proved that if s(T ) ¼ {1} and
F ðT Þ ðI  T ÞX: If for T as above and l > 1, we both T and T 1 are Abel bounded, then T ¼ I.
denote Rðl, T Þ≔ 1 k¼0 l
ðkþ1Þ k
T (the resolvent at A regular matrix A is called uniformly regular
l in the complex case), then we have if limn!1supj|an,j| ¼ 0. Fong and Sucheston
lim r!1 Ar ðT Þx ¼ lim l!1þ ðl  1ÞRðl, T Þx. (1974), extending an earlier result of Hanson and
Hille (1945) studied (C, α) and Abel averaging Pledger (1969), proved the following extension of
of general sequences in a Banach space. The fol- the Blum-Hanson Theorem 37.
lowing is his application of his general results.
Theorem 57 Let T be a contraction on a (real or
Theorem 55 Let α > 0. If T is a (C, α) ergodic complex) Hilbert space ℋ and let x  ℋ. Then
operator on a Banach space X, then kT nxk/nα T nx converges weakly if and only if for every
converges to zero for any x  X, and T is Abel
uniformly regular matrix A the averages Bn x ¼
ergodic, with lim r!1 Ar ðT Þx ¼ lim n!1 Can ðT Þx. 1 j
If T is power-bounded and Abel ergodic, then it j¼0 an,j T xconverge in norm.
is (C, α) ergodic. They also proved that when T is a contraction
If T is (C, α) bounded and Abel ergodic, it is of L1, the convergence of T n in the weak operator
(C, β) ergodic for any β > α. topology is equivalent to the convergence in the
Analogous results for C0-semigroups were strong operator topology of 1 j
j¼0 an,j T for every
proved by Kendall and Reuter (1956), using the uniformly regular matrix A. Akcoglu and
abstract ergodic theorems of Eberlein (1949). Sucheston (1975) proved the same equivalence
476 Operator Ergodic Theory

for positive contractions of Lp, 1 < p < 1. Lin l   . By looking at the product of θ with the
and Weber (2007) used Theorem 57 to obtain: rotation by l, acting on ðO  Þ, it can be deduced
from Birkhoff’s theorem that 1n n1 k k
k¼0 l T f con-
Theorem 58 Let wj  0 with w0 > 0 and verges a.e. for any f  L1(m). Wiener and Wintner
W n ≔ nj¼0 wj ! 1: Then for every contraction (1941) proved that when θ is ergodic, for
T on a Hilbert space with T n ! 0 weakly, the f  L1(m) there is a null set A  S such that for
weighted averages W1n nj¼0 wj T j converge o A, 1n n1 k¼0 l T f ðoÞ converges for every
k k

1 n 2
strongly if and only if W 2 j¼0 wj ! 0: l  : The Wiener-Wintner theorem follows by
n
Dunford and Schwartz (1956) extended the applying the extension by Rudolph (1994) of the
pointwise ergodic theorem, proving that if T is a return times theorem to all rotations of ; see the
contraction of L1(m) of a s-finite measure, such discussion following Proposition 65. See also
that kTfk1  k fk1 for f  L1 \ L1, then for Assani (2003). The Wiener-Wintner theorem
yields that if a  ‘1 can be uniformly approxi-
every f  L1 the averages 1n n1 k
k¼0 T f converge
mated by trigonometric polynomials, then for
a.e. Akcoglu (1975) proved that if T is a positive
every Koopman operator T and every f  L1(m)
contraction of Lp(m), 1 < p < 1, then 1n n1 k
k¼0 T f the averages 1n nk¼0 ak T k f converge a.e.
converges a.e. for every f  Lp. By classical
Ryll-Nardzewski (1975) noted that if k ¼
summability theory for numerical sequences
{kj}j  0 is an increasing sequence in ℕ with limn
(extended in Hille (1945) to vector sequences),
kn/n ¼ d > 0, then averaging along k is equivalent
in both the above theorems lim r!1 Ar ðT Þf exists
to the modulated average with the zero-one
a.e. This raises the question if, when m is finite,
weights am ¼ 1 if and only if m  {kj}.
Can ðT Þf , 0 < a < 1 converges a.e. for every
For the study of sequences a ¼ {ak}k  0 such
f  Lp, in either theorem. Irmisch (1980) proved
that the modulated averages 1n nk¼0 ak T k f con-
that in Akcoglu’s theorem, the convergence holds
verge almost everywhere for every Koopman
when α > 1/p, but may fail when α ¼ 1/p. Déniel
operator of an ergodic mpt and every f  Lp
(1989) proved Irmisch’s result for the Koopman
(some fixed 1  p < 1), we refer the reader to
operator of a probability-preserving transforma-
Bellow and Losert (1985), Bourgain et al. (1989),
tion and gave an example that even for such a
Rudolph (1994), and to the book of Assani
Koopman operator the a.e. convergence may fail
(2003). For some recent results, see Fan (2019).
when α ¼ 1/p; the norm convergence holds for
Let T be mean ergodic on a complex Banach
every 0 < α < 1 by Theorem 55.
space. A complex sequence a ≔ {ak}k0 will be
called a good modulating sequence for T if the
modulated averages Mn ðT, aÞ≔ 1n n1 k
k¼0 ak T con-
Modulated Ergodic Theorems
verge in the strong operator topology; the conver-
Let T be a weakly almost periodic operator on a gence is called a modulated ergodic theorem (the
complex Banach space. Then for any l   the ak need not be positive, so we avoid the often used
operator lT is also weakly almost periodic, so by terminology of “weighted” ergodic theorems). If
Theorem 7 1n n1 k k a is a good modulating sequence for the L2-
k¼0 l T converges in the strong
operator topology. It follows that if Koopman operators of all rotations of , then
k
a ≔ {ak}k0  ‘1 can be approximated uni- we must have that cðlÞ≔ lim n 1n k¼0 ak l exists
formly in k by sequences p ¼ {p(k)} of the form for every l   ; such a sequence is called a
pðkÞ≔ m Hartman sequence, and by Kahane (1961) the
j¼1 cj lj with cj  ℂ and lj   (called
k
set fl   : cðlÞ 6¼ 0g is at most countable.
trigonomtric polynomial sequences), then the
Berend et al. (2002) showed that a second neces-
averages 1n nk¼0 ak T k converge in the strong oper-
sary condition for modulating all rotations is that
ator topology.
supn supl   j1n n1
k¼0 ak l j < 1, and these two
k
Let T be the Koopman operator of a
conditions together imply that a is a good
probability-preserving θ on (Ω, S, m), and fix
Operator Ergodic Theory 477

modulating sequence for every contraction T of a For mean ergodic operators, Lin et al. (1999)
complex Hilbert space. The limit was identified in showed:
Lin et al. (1999):
Theorem 63 A sequence a ¼ {ak} is a good
n1 modulating sequence for every mean ergodic
1
lim ak T k x ¼ cðlÞE l, T x, x  ℋ, power-bounded operator if and only if it is
n!1 n k¼0 jlj¼1 Hartman, a  W1, and 1n n1 k¼0 jakþ1  ak j ! 0:
A bounded sequence a is called (weakly)
where E(l, T ) is the orthogonal projection on the almost periodic if its orbit in ‘1 under the shift
eigenspace of l (note that also S({bk}) ≔ {bkþ1} is conditionally (weakly) com-
fl   : Eðl, T Þx 6¼ 0g is countable).
pact. Then the restriction of S to span Sk a is
For 1  p < 1, denote W p ≔ a :k akW p ≔
1=p (weakly) almost periodic. Eisner (2013) used the
lim supn 1n n1
k¼0 jak j
p
< 1g: Then k akW p Jacobs-deLeeuw-Glicksberg decomposition
is a seminorm, and Wp is complete. We put (Theorem 29) to obtain the following.
W1 ≔ ‘1. Çömez et al. (1998) proved:
Theorem 64 Let S be a weakly almost periodic
operator on a complex Banach space Y, and fix
Theorem 59 Let a  Wp, 1 < p < 1 be
y  Y and f  Y . Then the sequence
Hartman. Then a is a good modulating sequence
a ≔ {f(Sky)}k0 is Hartman (and obviously
for every weakly almost periodic operator on a
bounded). Hence a weakly almost periodic
complex Banach space.
sequence is Hartman.
By Berend et al. (2002), Theorem 59 may fail
Lin et al. (1999) used the Wiener-Wintner the-
when a  W1. However, the necessity of a  W1
orem to generate Hartman sequences.
follows from the next result, due to Lin
et al. (1999).
Proposition 65 Let t be an ergodic probability-
Theorem 60 A sequence a is a good modulating preserving transformation on O, S, n , and fix
sequence for every almost periodic operator on a g  Lp(n), 1  p  1. Then for a.e. o  O the
complex Banach space if and only if it is Hartman sequence a≔ g tk o k0 is Hartman and in Wp.
and a  W1. By Berend et al. (2002), the Hartman
Eisner and Lin (2018) proved the following. sequences a  Wp defined in Proposition 65 are
good modulating sequences for every contraction
Theorem 61 Let a  W1. Then for every power- on a Hilbert space.
bounded T on a Banach space X and x  X with Rudolph (1994) extended Bourgain’s return
T nx ! 0 (weakly), we have 1n n1 k¼0 ak T x ! 0
k times theorem, showing that for p  1, the
(weakly). Hartman sequences a  Wp defined in Proposi-
By Berend et al. (2002) there exists a Hartman tion 65 yield a.e. convergence of 1n n1 k
k¼0 ak f y o
sequence a W1 which satisfies for every probability-preserving θ on (Ω, S, m)
1 n1
and f  Lq(m), q ¼ p/( p  1). Demeter et al.
supn supl   jn k¼0 ak l j < 1, so this a is a
k

good modulating sequence for all contractions in (2008) proved that when p > 1, the above
a Hilbert space, but not for all almost periodic sequences a  Wp yield a.e. convergence of
operators. Eisner and Lin (2018) proved: 1 n1 k
n k¼0 ak f y o for every probability preserving
θ on (Ω, S, m) and f  L2(m); Demeter (2012)
Proposition 62 Let a  W1. For every contrac- refined it to f  Lq , q1 < 32  p1 :
tion T on a Hilbert space ℋ and x  ℋ with A sequence a which for some 1  p < 1 is in
T nx ! 0 weakly, we have 1n nk¼1 ak T k x ! 0 if the Wp closure of the trigonometric polynomial
and only if an ¼ o(n). sequences is called a p-Besicovitch sequence. The
478 Operator Ergodic Theory

1-Besicovitch sequences are called Besicovitch Theorem 66 An operator T on a complex

sequences. Besicovitch sequences were first used Banach space satisfies (11) if and only if
by Tempelman (1974) and by Ryll-Nardzewski supn{kT nk þ nkT n(I  T )k} < 1.
(1975) for obtaining pointwise modulated ergodic Operators satisfying (11) are now called Ritt
theorems for Koopman operators. Besicovitch operators; it was also shown by Nagy and Zemánek
sequences are Hartman, see Bellow and Losert (1999) that the spectrum of a Ritt operator is
(1985), and the space of 2-Besicovitch sequences contained in a Stolz region (the closed convex hull
is an inner product space. Parseval equality in this of 1 and a disk of radius r < 1). By Theorem 66, Ritt
space, see Besicovitch (1954), yields uniqueness operators on reflexive Banach spaces are stable (see
of the coefficient function c(l), i.e., a has c(l) 0 Lemma 41). A unitary T 6¼ I is never Ritt. El-Fallah
if and only if 1n nk¼0 jak j2 ! 0: By Bellow and and Ransford (2002) proved that (11) implies
Losert (1985), all the sequences generated in The- kT nk  C2, which for large C was improved by
orem 65 by t invertible and g  L1 are Bakaev (2003) to kT nk  AC log (AC) for some
Besicovitch if and only if t has discrete spectrum A > 0. Lyubich (2001) proved the existence of a Ritt
(i.e., L2 has an orthogonal basis of eigen- operator T 6¼ I on Lp with spectrum {1}. Kalton et al.
functions). Bellow and Losert (1985) showed (2004) proved:
that the sequence of Fourier-Stieltjes coefficients
fnðkÞgk0 of a complex Borel measure on  is a Theorem 67 Let T on a complex Banach space
bounded Besicovitch sequence. satisfy lim supn!1nkT n  T nþ1k < e1; then T is
By Lin et al. (1999), Besicovitch sequences are power-bounded, hence a Ritt operator.
good modulating sequences for every weakly Moreover, they showed that with equality to
almost periodic operator on a complex Banach e1 the result fails in any infinite-dimensional
space (even if they are not p-Besicovitch for any Banach space. Malinen et al. (2009) proved that
p > 1), since they are limits in W1 of bounded if kR(l, T) k  C/(l  1) for 1 < l < 1 þ ε and
Hartman sequences. supn1 nkT n  T nþ1k < 1, then T is power-
bounded, hence a Ritt operator. Chen and Shaw
(2009) proved that an Abel bounded T satisfying
Resolvent Conditions and Growth of sup nkT n(I  T)k < 1 is power-bounded, hence
Powers Ritt.
Le Merdy and Xu (2012) proved the following:
The theorems of Koopman and von Neumann
(1932), Dunford (1943b), and Katznelson and Theorem 68 Let 1 < p < 1 and let T be a
Tzafriri (1986) show that certain properties of positive contraction of Lp(Ω, m) of a s-finite mea-
the peripheral spectrum sðT Þ \  of a power- sure space. If T is Ritt, then T nf converges a.e. for
bounded operator T imply certain ergodic every f  Lp(Ω, m).
theorems. Examples for Theorem 68 were given in
The Ritt condition. Ritt (1953) introduced the Cohen et al. (2014).
following spectral condition for T on a complex Dungey (2011) proved that for every
Banach space with spectral radius r(T )  1: 0 < α < 1, and every power-bounded T, the
ðaÞ
operator S ¼ I  ðI  T Þa ¼ 1 j
j¼1 aj T (see the
C
k Rðl, T Þ k for j l j> 1: ð11Þ section Rates of convergence), which is a convex
jl1j
combination of the powers of T, is a Ritt operator.
Ritt proved that (11) implies n1kT nk ! 0; the Gomilko and Tomilov (2018) proved that if T is
question of whether it implies power-bounded- Ritt, then any convex power series 1 n¼0 cn T ,
n

ness was studied for many years, until Lyubich with cn  0 and n0 cn ¼ 1, is Ritt.
(1999) and Nagy and Zemánek (1999) proved The Kreiss condition. Kreiss (1962) pre-
(independently) the following. sented the following resolvent condition (Kreiss
Operator Ergodic Theory 479

resolvent condition) for T on a complex Banach (2016) is that when supn Cðn2Þ ðT Þ < 1, then
space with r(T )  1: kR(l, T ) k  C|l  1|2/(| l| 1)3, j l j > 1.
Earlier, Kreiss had given a resolvent condition
C for the generator of a C0-semigroup, inspired by
k Rðl, T Þ k for j l j> 1: ð12Þ
j l j 1 the Hille-Yosida theorem, which in finite-
dimensional spaces yields boundedness of the
Ritt’s condition (11) clearly implies (12). Oper- semigroup; however, in contrast to the discrete
ators satisfying (12) are often called Kreiss time, Eisner and Zwart (2006) constructed a C0-
bounded operators. Power-bounded operators semigroup with exponential growth whose gener-
satisfy (12). Kreiss proved that in finite- ator satisfies Kreiss’ condition.
dimensional spaces (12) implies power- Van Castren (1980) proved that if T is power-
boundedness. Lubich and Nevanlinna (1991) pro- bounded invertible on a complex Hilbert space
ved that (12) implies kT nk ¼ O(n); by Shields with sðT Þ  , and T 1 satisfies (12) (which is
(1978) or Nevanlinna (2001), this is the best esti- equivalent to condition (ii) in van Casteren’s the-
mate. However, Nevanlinna (2001) showed that if orem), then also T 1 is power-bounded. This
T satisfies (12) and its peripheral spectrum extended results of Gokhberg and Krein (1967)
sðT Þ \  has arc-length (Lebesgue) measure and of Stampfli (1972).
zero, then kT nk ¼ o(n). Independently, Cohen McCarthy (1971) gave an example of
et al. (2020) and Bonilla and Müller (2021) pro- T invertible on ‘2(ℤ) which satisfies the stronger
ved that in Hilbert spaces, (12) implies kT nk ¼ condition (strong Kreiss resolvent condition,
O(n/ log n). Cuny (2020) proved that a Kreiss sometimes called iterated Kreiss condition):
bounded T on Lp(Ω, m), 1 < p < 1, satisfies
p
kT n k ¼ Oðn= log nÞ; he obtained the estimate C
Rk ðl, T Þ 
kT nk ¼ O(n/(logn)1/s), for some s  2, when T is ðjlj1Þk ð13Þ
Kreiss bounded, defined on a space in a subclass whenever j l j> 1, k ¼ 1, 2, . . .
of the spaces which are uniformly convex in an
equivalent norm (i.e., UMD spaces, see Cuny
but is not power-bounded; in the example also T1
(2020)).
satisfies (13). By McCarthy (1971), condition (13)
Strikwerda and Wade (1997) proved that (12) p
implies that kT n k ¼ Oð nÞ: This estimate was
is equivalent to supn supjgj¼1 Cðn2Þ ðgT Þ < 1, proved also by Lubich and Nevanlinna (1991),
where Cðn2Þ is the Cesàro average of order 2. This who showed it is the best possible in general
characterization was extended by Aleman and Banach spaces. Cohen et al. (2020) showed that
Suciu (2016), who proved that T satisfies (12) if in Hilbert space (13) implies kT nk ¼ O((logn)k)
and only if for some integer r  2 we have for some k > 0 (which depends on T ). Lyubich
supn supjgj¼1 CðnrÞ ðgT Þ < 1: Gomilko and (2010) obtained a family of examples in L p[0, 1]
Zemánek (2008) proved that T satisfies (12) if satisfying (12) but not (13). Nevanlinna (1997,
and only if T m satisfies (12) for some (all) 2001) proved that T satisfies (13) if and only if
m > 1. Suciu and Zemánek (2013) proved that if for some M we have
T satisfies (12), then Cðn2Þ ðT Þx converges strongly
for x  F ðT Þ ðI  T ÞX; if in addition sðT Þ \ ezT  Mejzj 8z  ℂ: ð14Þ
 ¼ f1g, then kMnþ1(T )  Mn(T)k ! 0. Abadias
and Bonilla (2019) proved that if T satisfies (12), Gomilko and Zemánek (2013) proved that
then for every α  2 we have T satisfies (13) if and only if T m satisfies it for
ðaÞ
Cnþ1 ðT Þ  CðnaÞ ðT Þ ! 0 (with no assumption some (all) integers m > 1.
Montes-Rodríguez et al. (2005) defined the
on the spectrum). A special case of Suciu uniform Kreiss resolvent condition by
480 Operator Ergodic Theory

n
Tk C boundedness) and the strong Kreiss resolvent con-
sup  whenever dition (13) are independent. They proved also that
n1 lkþ1 j l j 1
k¼0 if T is absolutely Cesàro bounded, then kT nk ¼
j l j> 1: ð15Þ O(n1ε) for some ε > 0 (which depends on T ).
Cuny (2020) proved that when the Banach space
They showed that (15) does not imply (13) and X has type 1 < p  2 (e.g., Lr spaces, 1 < r < 1,
proved that (15) holds if and only if there exists and more generally spaces which are uniformly
C > 0 such that convex in an equivalent norm), an absolutely
Cesàro bounded T satisfies kTnk ¼ O(n1/p).
1
n Cohen et al. (2020) proved that a positive
sup ðgT Þk  C 8 j g j¼ 1: ð16Þ Cesàro bounded operator on a complex Banach
n n k¼1
lattice satisfies the uniform Kreiss resolvent
condition.
The proof that (15) implies (12) is immediate.
Strikwerda and Wade (1997) showed that (12)
does not imply (16). Gomilko and Zemánek Continuous Time (C0-semigroups)
(2008) proved that (13) implies (15), hence (16);
together with the McCarthy-Lubich-Nevanlinna Birkhoff and von Neumann were motivated by a
p
estimate kT n k ¼ Oð nÞ under (13), Proposition problem in statistical mechanics and therefore
9 yields: proved their ergodic theorems in continuous
time. Von Neumann (1932) used in his proof
Theorem 69 Let T on a reflexive Banach space the spectral theorem for unitary representations
satisfy the strong Kreiss resolvent condition (13). of ℝ.
Then γT is mean ergodic for every g  . A C0-semigroup (or one-parameter semi-
If T is power-bounded, then (13) holds (in an group, strongly continuous semigroup) is a fam-
equivalent norm, T is a contraction and C ¼ 1 ily {T(t)}t0 of bounded linear operators on a
in (12)). Bermúdez et al. (2020) proved Banach space X such that T(0) ¼ I, T(t þ s) ¼
T(t)T(s) for t, s  0, and T(t)x is continuous on
Theorem 70 Let T on a Hilbert space satisfy the [0, 1) for every x  X. The limit
uniform Kreiss resolvent condition (15). Then o0 ≔ limt!1 log k T(t) k /t < 1 (exists by
kT nk ¼ o(n). Hence γT is mean ergodic for subadditivity) is the type (or growth bound)
every g  . of the semigroup, and for any δ > o0 there
Theorem 70 was extended by Cuny (2020) to exists Mδ such that kT(t) k  Mδeδt for t  0.
the reflexive Lp spaces (in fact, to the above men- The generator (infinitesimal generator) of a
tioned UMD spaces). Bonilla and Müller (2021) C0 - semigroup {T(t)}t0 is the operator
obtained examples of mean ergodic operators on a Ax ≔ limt!0+t1(T(t)x  x) with domain
complex Hilbert space which do not satisfy the D ðAÞ≔fx  X : Ax exists}. For properties of the
Kreiss resolvent condition (12). Bermúdez et al. generator, we refer the reader to Section VIII.1
(2020) used a strict strengthening of (16) for the of the book by Dunford and Schwartz (1958):
following.
(i) The domain of the generator is dense in X,
Theorem 71 Let T on a Banach space satisfy and A is a closed operator.
supn n1 n1
k¼0 T x  C k x k for some C > 0.
k
(ii) For x  D ðAÞ and t > 0, T ðtÞx  D ðAÞ and
Then kT k ¼ o(n). When X is reflexive, γT is mean
n
dt T ðtÞx ¼ AT ðtÞx ¼ T ðtÞAx:
d

ergodic for every g  . (iii) The generator is a bounded operator if and

Cohen et al. (2020) showed that the condition only if the semigroup is continuous in the
of Theorem 71 (called absolute Cesàro uniform (operator norm) topology.
Operator Ergodic Theory 481

(iv) If Re l > o0, then l  r(A), and A C0-semigroup {T(t)}t0 is called mean
1
Rðl, AÞx ¼ 0 elt T ðtÞxdt for every x  X. (uniformly) ergodic if the averages
Ms x≔ 1s 0 T ðtÞxdt
s
converge in the strong
Hille (1952) introduced the following abstract (uniform) operator topology, as s ! 1. The
generalization (the abstract Cauchy problem) of limit is a projection on the common fixed points
the (homogenous) initial value problem of partial of {T(t)}t0.
differential equations. Let A be a linear operator
(not necessarily bounded) mapping a subspace D Theorem 72 A bounded C0-semigroup on a
(not necessarily closed) of a complex Banach reflexive Banach space is mean ergodic.
space X into X, and consider the problem of find- This continuous version of Lorch’s Theorem
ing a differentiable function y(t), defined for t > 0 (Corollary 8) follows from the abstract ergodic
with values in D, such that dtd yðtÞ ¼ AyðtÞ for theorems of Eberlein (1949). Hille (see
t > 0, and y(t) ! y0 as t ! 0+ for a given y0. Section 18.6 of Hille and Phillips (1957),
When D ¼ X and A is bounded, the operator Section V.4 of Engel and Nagel (2000)) proved
etA is a well-defined bounded operator, and the following.
y(t) ≔ etAy0 solves the abstract Cauchy problem.
Hille noted that if A is the generator of a C0- Theorem 73 Let {T(t)}t0 be a mean ergodic C0-
semigroup T(t), then y(t) ≔ T(t)y0 solves abstract semigroup on X, with generator A. Then X ¼
Cauchy problem when y0  D ðAÞ: kerðAÞ RangeðAÞ, and Ex ≔ lims!1 Msx is
Ergodic theorems for a C0-semigroup {T(t)} the projection on ker(A) corresponding to this
deal with the asymptotic behavior of T(t)x as decomposition.
t ! 1, in different topologies and different Lin (1974b) proved the continuous analogue of
modes of convergence. These yield information Theorem 23.
on the asymptotic behavior of the solutions of the
abstract Cauchy problem when A is the generator
Theorem 74 A C0-semigroup with generator
of a C0-semigroup; from this perspective, it is
A is uniformly ergodic if and only if
important to use assumptions only on the genera-
limt!1 t1 k T(t) k ¼ 0, and the range of A is
tor of the semigroup, which is the given datum of
closed.
the problem. The book of van Neerven (1996)
Krengel and Lin (1984) proved the continuous
studies the different spectral properties of the gen-
analogue of Browder’s Theorem 14.
erator (which is usually only an unbounded closed
operator), and their connections to the asymptotic Theorem 75 Let {T(t)}t0 be a bounded C0-
behavior of the semigroup. The paper of semigroup on a reflexive Banach space, with gen-
Rozendaal and Veraar (2018), which uses growth erator A. Then y is in the range of A if and only if
rates of the resolvent of the generator for s
sups>0 0 T ðtÞxdt < 1:
obtaining growth rates of the semigroup, contains
Eisner (2010) emphasizes the similarities of
references to papers which appeared after publi-
the discrete and continuous cases and proves in
cation of van Neerven’s book.
parallel results for discrete time and their ana-
A nonspectral approach is in the book of
logues for continuous time. However, not all
Emel’yanov (2007), which emphasizes C0-semi-
results in the discrete case have analogues in con-
groups of positive operators on Banach lattices, in
tinuous time, e.g., Eisner and Zwart (2006), and
particular C0-semigroups of positive operators on
some results for semigroups have no analogue in
L1 spaces.
discrete time, e.g., Gerlach (2013), where it is
Many results on the asymptotic behavior of C0-
shown that a convergence result for certain C0-
semigroups have analogues for discrete time, and
semigroups fails in discrete time under the analo-
sometimes the continuous time version was pro-
gous assumptions, due to some periodic behavior.
ved first (see Theorems 25, 38, and 46).
482 Operator Ergodic Theory

Bibliography Azizov AN, Chilin VI (2017) The Blum-Hanson ergodic

theorem in Banach lattices of sequences (Russian).
The references below are the primary sources cited in this Vladikavkaz Mat Zh 19:3–10
article. Most of them contain additional relevant results Badea C, Müller V (2009) On weak orbits of operators.
and references. Many of the listed authors have addi- Topology Appl 156:1381–1385
tional related papers, not cited in the present article Badea C, Seifert D (2017) Quantified asymptotic behav-
All books on ergodic theory have a discussion of the mean iour of Banach space operators and applications to
ergodic theorem. We list below only books not cited in iterative projection methods. Pure Appl Funct Anal 2:
the article, which have chapters containing more oper- 585–598
ator theoretic, including asymptotic behavior of C0- Badiozzaman AJ, Thorpe B (1992) A uniform ergodic
semigroups theorem and summability. Bull Lond Math Soc 24:
351–360
Bakaev NY (2003) Constant size control in stability esti-
References mates under some resolvent conditions. Numerical
Aaronson J, Weiss B (2000) Remarks on the tightness of Methods Program 4:348–357
cocycles. Colloq Math 84/85(part 2):363–376 Banach S (1932) Théorie des Opérations Linéaires. Mono-
Abadias L, Bonilla A (2019) Growth orders and ergodicity grafje Mathematyczne, Warsaw
for absolutely Cesàro bounded operators. Linear Alge- Bartoszek W (1988) Asymptotic periodicity of the iterates
bra Appl 561:253–267 of positive contractions on Banach lattices. Stud Math
Akcoglu MA (1975) A pointwise ergodic theorem in Lp- 91:179–188
Spaces. Canadian J Math 27:1075–1082 Baskakov AG (2015) Harmonic and spectral analysis of
Akcoglu M, Sucheston L (1972) On operator convergence power bounded operators and bounded semigroups of
in Hilbert space and in Lebesgue space. Period Math operators on a Banach space (Russian). Mat Zametki
Hung 2:235–244 97:174–190; English translation: Math Notes
Akcoglu M, Sucheston L (1975) Weak convergence of 97 (2015), 164–178
positive contractions implies strong convergence of Batty CJK (1994) Asymptotic behaviour of semigroups of
averages. ZWahrscheinlichkeitstheorie Verw Gebiete operators. In: Functional analysis and operator theory,
32:139–145 Banach Center Publication 30. Polish Academy of Sci-
Akcoglu MA, Huneke JP, Rost H (1974) A counter exam- ences Institute of Mathematics, Warsaw, pp 35–52
ple to the Blum Hanson theorem in general spaces. Pac Batty CJK, Vũ QP (1990) Stability of individual elements
J Math 50:305–308 under one-parameter semigroups. Trans Am Math Soc
Aleman A, Suciu L (2016) On ergodic operator means in 322:805–818
Banach spaces. Integral Eq Operator Theory 85: Batty CJK, Seifert D (2022) Some developments around
259–287 the Katznelson-Tzafriri theorem. Acta Sci Math (Sze-
Allan GR (1997) Power-bounded elements and radical ged) 88:53–84
Banach algebras. In: Linear operators, vol 38. Banach Becker ME (2005) A condition equivalent to uniform
Center Publications, Polish Academy of Sciences, ergodicity. Stud Math 167:215–218
Institute of Mathematics, Warsaw, pp 9–16 Becker ME (2011) Ergodic properties of linear operators.
Allan GR, Ransford T (1989) Power-dominated elements Rev Un Mat Argentina 52:23–25
in a Banach algebra. Stud Math 94:63–79 Bellow A (1975) An Lp-inequality with application to
Allan GR, O’Farrell AG, Ransford TJ (1987) A Tauberian ergodic theory. Houst J Math 1:153–159
theorem arising in operator theory. Bull Lond Math Soc Bellow A, Losert V (1985) The weighted pointwise ergo-
19:537–545 dic theorem and the individual ergodic theorem along
Anosov D (1973) The additive functional homology equa- subsequences. Trans Am Math Soc 288:307–345
tion that is connected with an ergodic rotation of the Berend D (1992) A note on the L p analogue of the “zero-
circle, Izv. Akad Nauk SSSR Ser Mat 37:1257–1271. two” law. Proc Am Math Soc 114:95–97
(Russian; English transl. in Math. USSR Izvestya Berend D, Bergelson V (1986) Mixing sequences in Hil-
7 (1973)) bert spaces. Proc Am Math Soc 98:239–246
Arendt W, Batty C (1988) Tauberian theorems and stability Berend D, Lin M, Rosenblatt J, Tempelman A (2002)
of one-parameter semigroups. Trans Am Math Soc 306: Modulated and subsequential ergodic theorems in Hil-
837–852 bert and Banach spaces. Ergodic Theory Dyn Sys
Assani I (1986) Sur les opérateurs à puissances bornées et 22:1653–1665
le théorème ergodique ponctuel dans L p[0, 1] 1<p< Bergelson V (1987) Weakly mixing PET. Ergodic Theory
+1. Can J Math 38:937–946 Dyn Syst 7:337–349
Assani I (2003) Wiener Wintner ergodic theorems. World Berkson E, Gillespie TA (1987) Stečkin’s theorem, trans-
Scientific, River Edge ference, and spectral decompositions. J Funct Anal 70:
Avigad J, Rute J (2015) Oscillation and the mean ergodic 140–170
theorem for uniformly convex Banach spaces. Ergodic Bermúdez T, González M, Mbekhta M (2000) Operators
Theory Dynam Syst 35:1009–1027 with an ergodic power. Stud Math 141:201–208
Operator Ergodic Theory 483

Bermúdez T, Bonilla A, Müller V, Peris A (2020) Cesàro theory, Banach Center Publications, vol 75. Polish
bounded operators in Banach spaces. J Anal Math 140: Academy of Science, Warsaw, pp 71–109
187–206 Cohen LW (1940) On the mean ergodic theorem. Ann
Besicovitch AS (1954) Almost periodic functions. Dover Math 41(2):505–509
Publications, New York Cohen G, Lin M (2016) Remarks on rates of convergence
Birkhoff GD (1931) Proof of the ergodic theorem. Proc of powers of contractions. J Math Anal Appl 436:
Natl Acad Sci U S A 17:656–660 1196–1213
Birkhoff G (1939) The mean ergodic theorem. Duke Math Cohen G, Cuny C, Lin M (2010) The one-sided ergodic
J 5:19–20 Hilbert transform in Banach spaces. Stud Math 196:
Blum JR, Eisenberg B (1974) Generalized summing 251–263
sequences and the mean ergodic theorem. Proc Am Cohen G, Cuny C, Lin M (2014) Almost everywhere
Math Soc 42:423–429 convergence of powers of some positive Lp contrac-
Blum JR, Hanson DL (1960) On the mean ergodic theorem tions. J Math Anal Appl 420:1129–1153
for subsequences. Bull Am Math Soc 66:308–311 Cohen G, Cuny C, Eisner T, Lin M (2020) Resolvent
Bonilla A, Müller V (2021) Kreiss bounded and uniformly conditions and growth of powers of operators. J Math
Kreiss bounded operators. Rev Mat Complut 34: Anal Appl 487:124035, 24 p
469–487 Çömez D, Lin M, Olsen J (1998) Weighted ergodic theo-
Boshernitzan M, Wierdl M (1996) Ergodic theorems along rems for mean ergodic L1-contractions. Trans Am Math
sequences and Hardy fields. Proc Natl Acad Sci U S Soc 350:101–117
A 93:8205–8207 Conze J-P (1973) Convergence des moyennes ergodiques
Bourgain J (1988a) An approach to pointwise ergodic pour des sous-suites. In: Contributions au calcul des
theorems. In: Geometric aspects of functional analysis probabilités, Bulletin Society Mathematics, France,
(1986/87), Lecture notes in Mathematics 1317. Mém No. 35, Society Mathematics, France, pp 7–15
Springer, Berlin, pp 204–223 Cotlar M (1955) A unified theory of Hilbert transforms and
Bourgain J (1988b) Temps de retour pour les systèmes ergodic theorems. Rev Mat Cuyana 1:105–167
dynamiques. C R Acad Sci 306(12):483–485 Cuny C (2009) On the a.s. convergence of the one-sided
Bourgain J (1989) Pointwise ergodic theorems for arith- ergodic Hilbert transform. Ergodic Theory Dyn Syst
metic sets. Inst Hautes Études Sci Publ Math 69:5–45 29:1781–1788
Bourgain J, Furstenberg H, Katznelson Y, Ornstein DS Cuny C (2010) Norm convergence of some power series of
(1989) On return-time sequences, Appendix to Bourgain operators in L p with applications in ergodic theory.
(1989). Inst Hautes Études Sci Publ Math No 69:42–45 Stud Math 200:1–29
Browder F (1958) On the iteration of transformations in Cuny C (2020) Resolvent conditions and growth of powers
noncompact minimal dynamical systems. Proc Am of operators on L p spaces. Pure Appl Funct Anal 5:
Math Soc 9:773–780 1025–1038
Brunel A, Keane M (1969) Ergodic theorems for operator Cuny C, Weber M (2018) On Nörlund summation and
sequences. Z Wahrscheinlichkeitstheorie und Verw ergodic theory, with applications to power series of
Gebiete 12:231–240 Hilbert contractions. Bull Pol Acad Sci Math 66:69–85
Brunel A, Revuz D (1974) Quelques applications pro- Datko R (1970) Extending a theorem of A. M. Liapunov to
babilistes de la quasi-compacité. Ann Inst H Poincaré Hilbert space. J Math Anal Appl 32:610–616
Sect B (NS) 10:301–337 deLaubenfels R, Vũ Q-P (1996) The discrete Hille-Yosida
Burlando L (1997) A generalization of the uniform ergodic space and the asymptotic behaviour of individual orbits
theorem to poles of arbitrary order. Stud Math 122:75–98 of linear operators. J Funct Anal 142:539–548
Butzer PL, Westphal U (1971) The mean ergodic theorem deLeeuw K, Glicksberg I (1961) Applications of almost
and saturation. Indiana Univ Math J 20:1163–1174 periodic compactifications. Acta Math 105:63–97
Butzer PL, Westphal U (1972) Ein Operatorenkalkül für Demeter C (2012) Improved range in the return times
das approximationstheoretische Verhalten des theorem. Can Math Bull 55:708–722
Ergodensatzes im Mittel. In: Linear operators and Demeter C, Lacey MT, Tao T, Thiele C (2008) Breaking
approximation, International series of numerical math- the duality in the return times theorem. Duke Math
ematics, vol 20. Birkhäuser, Basel, pp 102–114 J 143:281–355
Campbell JT (1986) Spectral analysis of the ergodic Hil- Déniel Y (1989) On the a.s. Cesàro-α convergence for
bert transform. Indiana Univ Math J 35:379–390 stationary or orthogonal random variables. J Theor Pro-
Chacon RV (1969) Weakly mixing transformations which bab 2:475–485
are not strongly mixing. Proc Am Math Soc 22:559–562 Derriennic Y (1976) Lois “zéro ou deux” pour les processus
Chen J-C, Shaw S-Y (2009) Growth order and stability of de Markov. Applications aux marches aléatoires. Ann
discrete semigroups. Nonlinear Anal 71(12):e2879– Inst H Poincaré Sect B (NS) 12:111–129
e2882 Derriennic Y (2000) On the mean ergodic theorem for
Chill R, Tomilov Y (2007) Stability of operator semi- Cesàro bounded operators. Colloq Math 84/85(part
groups: ideas and results. In: Perspectives in operator 2):443–455
484 Operator Ergodic Theory

Derriennic Y, Krengel U (1981) Subadditive mean ergodic Emel’yanov EY (1997) Banach lattices on which every
theorems. Ergodic Theory Dyn Syst 1:33–48 power-bounded operator is mean ergodic. Positivity 1:
Derriennic Y, Lin M (2001) Fractional Poisson equations 291–295
and ergodic theorems for fractional coboundaries. Émilion R (1985) Mean-bounded operators and mean
Israel J Math 123:93–130 ergodic theorems. J Funct Anal 61:114
Devys O (2012) Localisation spectrale à l’aide des poly- Esterle J (1983) Quasimultipliers, representations of H1,
nômes de Faber et équation de cobord. PhD thesis, and the closed ideal problem for commutative Banach
Lille. https://pepite-depot.univ-lille.fr/LIBRE/EDSPI/ algebras. In: Radical Banach algebras and automatic
2012/50376-2012-Devys.pdf continuity, Lecture Notes in Mathematics 975.
Doeblin W (1937) Sur les propriétés asymptotiques de Springer, Berlin, pp 66–162
mouvements régis par certains types de chaînes sim- Esterle J, Strouse E, Zouakia F (1990) Theorems of
ples. Bull Soc Math Roumaine 39–1:57–115; Katznelson-Tzafriri type for contractions. J Funct
39–2 (1938), 3–61 Anal 94:273–287
Drissi D, Zemánek J (2000) Gelfand-Hille theorems for Esterle J, Strouse E, Zouakia F (1992) Stabilité asymptotique
Cesàro means. Quaest Math 23:375–381 de certains semi-groupes d’opérateurs et idéaux
Dunford N (1943a) Spectral theory, I. Convergence to primaires de L1(R+). J Operator Theory 28:203–227
projections. Trans Am Math Soc 54:185–217 Fan A-H (2019) Weighted Birkhoff ergodic theorem with
Dunford N (1943b) Spectral theory. Bull Am Math Soc 49: oscillating weights. Ergodic Theory Dyn Syst 39:
637–651 1275–1289
Dunford N, Schwartz JT (1956) Convergence almost Farkas B, Kreidler H (2021) Relative compactness of orbits
everywhere of operator averages. J Rational Mech and geometry of Banach spaces. J Math Anal Appl
Anal 5:129–178 495:124660
Dungey N (2008) On time regularity and related conditions Foguel S (1963) Powers of a contraction in Hilbert space.
for power-bounded operators. Proc Lond Math Soc Pac J Math 13:551–562
97(3):97–116 Foguel S (1976) More on the “zero-two” law. Proc Am
Dungey N (2011) Subordinated discrete semigroups of Math Soc 61:262–264
operators. Trans Am Math Soc 363:1721–1741 Foguel S, Weiss B (1973) On convex power series of a
Eberlein WF (1949) Abstract ergodic theorems and weak conservative Markov operator. Proc Am Math Soc 38:
almost periodic functions. Trans Am Math Soc 67: 325–330
217–240 Foiaş C (1964) Sur les mesures spectrales qui interviennent
Eberlein WF (1984) On retrogression in mean ergodic dans la théorie ergodique. J Math Mech 13:639–658
theory. J Approx Theory 42:293–298 Fonf VP, Lin M, Rubinov A (1996) On the uniform ergodic
Ed-dari E (2003) On the (C,α) uniform ergodic theorem. theorem in Banach spaces that do not contain duals.
Stud Math 156:3–13 Stud Math 121:67–85
Ed-dari E (2004) On the (C,α) Cesàro bounded operators. Fonf VP, Lin M, Wojtaszczyk P (2001) Ergodic character-
Stud Math 161:163–175 izations of reflexivity of Banach spaces. J Funct Anal
Eisner T (2010) Stability of operators and operator semi- 187:146–162
groups, Operator theory: advances and applications, Fonf VP, Lin M, Wojtaszczyk P (2010) A non-reflexive
vol 209. Birkhäuser, Basel Banach space with all contractions mean ergodic. Israel
Eisner T (2013) Linear sequences and weighted ergodic J Math 179:479–491
theorems. Abstr Appl Anal:815726, 5 p Fonf VP, Lin M, Wojtaszczyk P (2011) Poisson’s equation
Eisner T, Lin M (2018) On modulated ergodic theorems. and characterizations of reflexivity of Banach spaces.
J Nonlinear Var Anal 2:131–154 Colloq Math 124:225–235
Eisner T, Müller V (2021) Power bounded operators and Fong H, Sucheston L (1974) On a mixing property of
the mean ergodic theorem for subsequences. J Math operators in Lp spaces. Z. Wahrscheinlichkeitstheorie
Anal Appl 493:124523, 25 p und Verw. Gebiete 28:165–171
Eisner T, Serény A (2008) Category theorems for stable Furstenberg H (1981) Recurrence in ergodic theory and
operators on Hilbert spaces. Acta Sci Math (Szeged) combinatorial number theory. Princeton University
74:259–270 Press, Princeton
Eisner T, Zwart H (2006) Continuous-time Kreiss resol- Gaposhkin VF (1998) On the rate of decrease of the prob-
vent condition on infinite-dimensional spaces. Math abilities of  deviations for means of stationary pro-
Comput 75(256):1971–1985 cesses. Mat Zametki 64(3):366–372; English
El Abdalaoui EH, El Machkouri M, Nogueira A (2010) translation in Math Notes 64 (1998), 316–321
A criterion of weak mixing property. Séminaires et Gelfand I (1941) Zur Theorie der Charaktere der Abelschen
Congrès 20:105–111 topologischen Gruppen. Mat Sbornik 9:49–50
El-Fallah O, Ransford T (2002) Extremal growth of powers Gerlach M (2013) On the peripheral point spectrum and the
of operators satisfying resolvent conditions of Kreiss- asymptotic behavior of irreducible semigroups of Har-
Ritt type. J Funct Anal 196:135–154 ris operators. Positivity 17:875–898
Operator Ergodic Theory 485

Gerlach M, Glück J (2019) Mean ergodicity vs weak Hopf E (1932) Proof of Gibbs’ hypothesis on the tendency
almost periodicity. Stud Math 248:45–56 toward statistical equilibrium. Proc Nat Acad Sci U S
Glück J (2015) On weak decay rates and uniform stability A 18:33–340
of bounded linear operators. Arch Math 104:347–356 Hopf E (1937) Ergodentheorie, Ergebnisse der Mathematik
Gokhberg I, Krein MG (1967) Description of contraction und ihrer Grenzgebiete 5, No. 2. Julius Springer
operators which are similar to unitary operators, Func- Horowitz S (1968) Some limit theorems for Markov pro-
tional. Anal Appl 1:33–52 cesses. Israel J Math 6:107–118
Goldstein JA (1993) Extremal properties of contraction Horowitz S (1969) Strong ergodic theorems for Markov
semigroups on Hilbert and Banach spaces. Bull Lond processes. Proc Am Math Soc 23:328–334
Math Soc 25:369–376 Huang SZ (1995) Stability properties characterizing the
Goldstein JA, Nagy B (1995) An extremal property of spectra of operators on Banach spaces. J Funct Anal
contraction semigroups in Banach spaces. Ill J Math 132:361–382
39:441–449 Ionescu Tulcea A (1963) Random series and spectra of
Gomilko A, Tomilov Y (2018) On discrete subordination measure-preserving transformations. In: Ergodic the-
of power bounded and Ritt operators. Indiana Univ ory, Proceedings of the International Symposium,
Math J 67:781–829 Tulane University, New Orleans. Academic Press, Bos-
Gomilko A, Zemánek J (2008) On the uniform Kreiss ton, pp 273–292
resolvent condition. Funct Anal Appl 42:230–233 Irmisch R (1980) Punktweise Ergodensätze für (c,α)-
Gomilko A, Zemánek J (2013) On the strong Kreiss resol- Verfahren, 0<α<1. PhD thesis, Darmstadt
vent condition. Complex Anal Oper Theory 7:421–435 Jacobs K (1957) Fastperiodizitätseigenschaften
Gomilko A, Haase M, Tomilov Y (2011) On rates in mean allgemeiner Halbgruppen in Banach-Räumen. Math
ergodic theorems. Math Res Lett 18:201–213 Z 67:83–92
Gottschalk W, Hedlund GA (1955) Topological dynamics, Jamison B (1965) Eigenvalues of modulus 1. Proc Am
American Mathematical Society Colloquium Publica- Math Soc 16:375–377
tions, vol 36. American Mathematical Society, Jones L (1971) A mean ergodic theorem for weakly mixing
Providence operators. Adv Math 7:211–216
Grabiner S, Zemánek J (2002) Ascent, descent, and ergodic Jones L, Kuftinec V (1971) A note on the Blum-Hanson
properties of linear operators. J Operator Theory 48: theorem. Proc Am Math Soc 30:202–203
69–81 Jones L, Lin M (1976) Ergodic theorems of weak mixing
Grivaux S (2019) The Blum-Hanson property. Concr Oper type. Proc Am Math Soc 57:50–52
6:92–105 Jones L, Lin M (1980) Unimodular eigenvalues and weak
Grobler JJ, Huijsmans CB (1995) Doubly Abel bounded mixing. J Funct Anal 35:42–48
operators with single spectrum. Quaest Math 18: Jones RL, Ostrovskii IV, Rosenblatt JM (1996) Square
397–406 functions in ergodic theory. Ergodic Theory Dyn Syst
Groh U (1983) On the peripheral spectrum of uniformly 16:267–305
ergodic positive operators on C*-algebras. J Operator Jones RL, Kaufman R, Rosenblatt JM, Wierdl M (1998)
Theory 10:31–37 Oscillation in ergodic theory. Ergodic Theory Dyn Syst
Haase M, Tomilov Y (2010) Domain characterizations of 18:889–935
certain functions of power-bounded operators. Stud Kachurovskiǐ AG, Podvigin IV (2016) Estimates of the rate
Math 196:265–288 of convergence in the von Neumann and Birkhoff ergo-
Halmos PR (1944) In general a measure preserving trans- dic theorems. Trans Moscow Math Soc:1–53
formation is mixing. Ann Math 45(2):786–792 Kachurovskiǐ AG, Sedalishchev VV (2010) On the con-
Halmos PR (1949a) Measurable transformations. Bull Am stants in the estimates for the rate of convergence in von
Math Soc 55:1015–1034 Neumann’s ergodic theorem. Mat Zametki 87:
Halmos PR (1949b) A nonhomogeneous ergodic theorem. 756–763, (Russian; English translation in Math. Notes
Trans Am Math Soc 66:284–288 87 (2010), 720–727)
Hanson DL, Pledger G (1969) On the mean ergodic theorem Kahane J-P (1961) Sur les coefficients de Fourier-Bohr.
for weighted averages. Z Wahrscheinlichkeitstheorie und Stud Math 21:103–106
Verw Gebiete 13:141–149 Kakutani S (1938) Iteration of linear operations in complex
Hardy GH, Littlewood JE (1913) Sur la série de Fourier Banach spaces. Proc Imp Acad Tokyo 14:295–300
d’une fonction à carré sommable. C. R. Acad Sci Paris Kakutani S (1950) Ergodic theory. In: Proceedings of the
156:1307–1309 1950 International Congress of Mathematicians, Cam-
Hiai F (1978) Weakly mixing properties of semigroups of bridge, pp 128–142
linear operators. Kodai Math J 1:376–393 Kakutani S, Petersen K (1981) The speed of convergence
Hille E (1945) Remarks on ergodic theorems. Trans Am in the ergodic theorem. Monatsh Math 91:11–18
Math Soc 57:246–269 Kalton N, Montgomery-Smith S, Oleszkiewicz K, Tomilov
Hille E (1952) Une généralisation du problème de Cauchy. Y (2004) Power-bounded operators and related norm
Ann Inst Fourier (Grenoble) 4:31–48 estimates. J Lond Math Soc 70:463–478
486 Operator Ergodic Theory

Katznelson Y, Tzafriri L (1986) On power bounded oper- Kunszenti-Kovács D, Nittka R, Sauter M (2011) On the
ators. J Funct Anal 68:313–328 limits of Cesàro means of polynomial powers. Math
Kendall DG, Reuter GEH (1956) Some ergodic theorems Z 268:771–776
for one-parameter semigroups of operators. Philos Lasota A, Li TY, Yorke JA (1984) Asymptotic periodicity
Trans R Soc Lond Ser A 249:151–177 of the iterates of Markov operators. Trans Am Math Soc
Kérchy L, van Neerven J (1997) Polynomially bounded 286:751–764
operators whose spectrum on the unit circle has mea- Le Merdy C, Xu Q (2012) Strong q-variation inequalities
sure zero. Acta Sci Math (Szeged) 63:551–562 for analytic semigroups. Ann Inst Fourier 62:
Kohlenbach U, Leuştean L (2009) A quantitative mean 2069–2097
ergodic theorem for uniformly convex Banach spaces. Lefèvre P, Matheron É (2016) The Blum-Hanson property
Ergodic Theory Dyn Syst 29:1907–1915. Erratum: for C(K ) spaces. Pac J Math 282:203–212
p 1995 Lefèvre P, Matheron É, Primot A (2016) Smoothness,
Koliha JJ (1974) Power convergence and pseudoinverses asymptotic smoothness and the Blum-Hanson property.
of operators in Banach spaces. J Math Anal Appl 48: Israel J Math 211:271–309
446–469 Léka Z (2009) A Katznelson-Tzafriri type theorem in Hil-
Komorník J (1993) Asymptotic periodicity of Markov and bert spaces. Proc Am Math Soc 137:3763–3768
related operators. In: Dynamics reported, Dynamical Léka Z (2010) A note on the powers of Cesàro bounded
report expositions dynamics systems (N.S.), 2, operators. Czechoslov Math J 60(135):1091–1100
Springer, Berlin, pp 31–68 Leonov VP (1961) On the dispersion of time means-
Koopman BO, von Neumann J (1932) Dynamical systems dependent of a stationary stochastic process. Teor
of continuous spectra. Proc Natl Acad Sci USA 18: Veroyatn Primen 6:93–101, in Russian; English transl
255–263 in Theory Probab Appl 6 (1961), 87–93
Kornfeld I, Kosek W (2003) Positive L1 operators associ- Li Y-C, Satō R, Shaw S-Y (2008) Boundedness and growth
ated with nonsingular mappings and an example of orders of means of discrete and continuous semigroups
E. Hille. Colloq Math 98:63–77 of operators. Stud Math 187:1–35
Kornfeld I, Lin M (1997) Coboundaries of irreducible Lin M (1974a) On the uniform ergodic theorem. Proc Am
Markov operators on C(K ). Israel J Math. 97:189–202 Math Soc 43:337–340
Kosek W (2011) Example of a mean ergodic L1 operator Lin M (1974b) On the uniform ergodic theorem II. Proc
with the linear rate of growth. Colloq Math 124(1): Am Math Soc 46:217–225
15–22 Lin M (1978) Quasi-compactness and uniform ergodicity
Kozitsky Y, Shoikhet D, Zemánek J (2013) Power conver- of positive operators. Israel J Math 29:309–311
gence of Abel averages. Arch Math 100:539–549 Lin M (1998) The uniform zero-two law for positive oper-
Kozma G, Lev N (2011) Exponential Riesz bases, discrep- ators in Banach lattices. Studia Math 131:149–153
ancy of irrational rotations and BMO. J Fourier Anal Lin M, Sine R (1983) Ergodic theory and the functional
Appl 17:879–898 equation (I–T )x¼y. J Operator Theory 10:153–166
Krein MG, Šmulian V (1940) On regularly convex sets in Lin M, Suciu L (2015) Poisson’s equation for mean ergodic
the space conjugate to a Banach space. Ann Math operators. In: Infinite products of operators and their
41(2):556–583 applications, Contemporary mathematics 636, Proceed-
Kreiss H-O (1962) ber die Stabilitätsdefinition für ings of the American Mathematical Society, Provi-
Differenzengleichungen die partielle dence, pp 141–148
Differentialgleichungen approximieren. BIT 2: Lin M, Weber M (2007) Weighted ergodic theorems and
153–181 strong laws of large numbers. Ergodic Theory Dyn Syst
Krengel U (1971) On the individual ergodic theorem for 27:511–543
subsequences. Ann Math Statist 42:1091–1095 Lin M, Olsen J, Tempelman A (1999) On modulated ergo-
Krengel U (1972) Weakly wandering vectors and weakly dic theorems for Dunford-Schwartz operators. Ill
independent partitions. Trans Am Math Soc 164: J Math 43:542–567
199–226 Lin M, Shoikhet D, Suciu L (2015) Remarks on uniform
Krengel U (1978) On the speed of convergence in the ergodic theorems. Acta Sci Math (Szeged) 81:251–283
ergodic theorem. Monatsh Math 86:3–6 Lorch ER (1939) Means of iterated transformations in
Krengel U, Lin M (1984) On the range of the generator of a reflexive vector spaces. Bull Am Math Soc 45:945–947
Markovian semigroup. Math Z 185:553–565 Lotz H (1968) Über das Spektrum positiver Operatoren.
Krotkov V, Halperin I (1953) The ergodic theorem for Math Z 108:15–32
Banach spaces with convex-compactness. Trans Lotz H (1981) Uniform ergodic theorems for Markov
R Soc Canada Sect III 47:17–20 operators on C(X). Math Z 178:145–156
Kryloff N, Bogoliouboff N (1937) Sur les propriétés en Lotz H (1985) Uniform convergence of operators on L1
chaîne. C R Acad Sci Paris 204:1386–1388 and similar spaces. Math Z 190:207–220
Kunszenti-Kovács D (2015) Almost weak polynomial sta- Lubich C, Nevanlinna O (1991) On resolvent conditions
bility of operators. Houst J Math 41:901–913 and stability estimates. BIT 31:293–313
Operator Ergodic Theory 487

Luecke G (1977) Norm convergence of Tn. Canadian operators, Banach Center Publications, vol 38.
J Math 29:1340–1344 IMPAN, Warsaw, pp 247–264
Lyubich Y (1999) Spectral localization, power bounded- Nevanlinna O (2001) Resolvent conditions and powers of
ness and invariant subspaces under Ritt’s type condi- operators. Stud Math 145:113–134
tion. Stud Math 134:153167 Ng ACS, Seifert D (2020) Optimal rates of decay in the
Lyubich Y (2001) The single-point spectrum operators Katznelson-Tzafriri theorem for operators on Hilbert
satisfying Ritt’s resolvent condition. Stud Math 145: spaces. J Funct Anal 279(12):108799, 21 p
135–142 Oxtoby J (1952) Ergodic sets. Bull Am Math Soc 58:
Lyubich Y (2010) The power boundedness and resolvent 116–136
conditions for functions of the classical Volterra oper- Pazy A (1972) On the applicability of Lyapunov’s theorem
ator. Stud Math 196:41–63 in Hilbert space. SIAM J Math Anal 3:291–294
Lyubich YI, Vũ QP (1988) Asymptotic stability of linear Petersen K (1983) Another proof of the existence of the
differential equations in Banach spaces. Stud Math 88: ergodic Hilbert transform. Proc Am Math Soc 88:
37–42 39–43
Malinen J, Nevanlinna O, Turunen V, Yuan Z (2007) Petersen K (1996) Ergodic theorems and the basis of sci-
A lower bound for the difference of powers of linear ence. Synthese 108:171–183
operators. Acta Math Sin (Engl Ser) 23:745–748 Radjavi H, Tam P-K, Tan K-K (2003) Mean ergodicity for
Malinen J, Nevanlinna O, Yuan Z (2009) On a Tauberian compact operators. Stud Math 158:207–217
condition for bounded linear operators. Math Proc R Ir Reich S (1973) Iterative solution of linear operator equa-
Acad 109:101–108 tions in Banach spaces. Atti Accad Naz Lincei Rend Cl
Mbekhta M, Zemánek J (1993) Sur le théorème ergodique Sci Fis Mat Nat 54(8):551–554
uniforme et le spectre. C R Acad Sci Paris Sr I Math Riesz F (1938) Some mean ergodic theorems. J Lond Math
317(12):1155–1158 Soc 13:274–278
McCarthy C (1971) A strong resolvent condition does not Ritt RK (1953) A condition that limn!1n1Tn¼0. Proc
imply power-boundedness. Chalmers Inst Technol Am Math Soc 4:898–899
Univ Gothenburg preprint 15 Robinson EA (1960) Sums of stationary random variables.
Michal AD, Wyman M (1941) Characterization of com- Proc Am Math Soc 11:7779
plex couple spaces. Ann Math 42(2):247–250 Rokhlin VA (1948) A “general” measure-preserving trans-
Millet A (1976) Un théorème ergodique en moyenne. C. R. formation is not mixing (Russian). Doklady Akad Nauk
Acad Sci Paris Sér A-B 283(16):A1103–A1106 SSSR (NS) 60:349–351
Montes-Rodríguez A, Sánchez-Álvarez J, Zemánek Rokhlin VA (1961) Exact endomorphisms of Lebesgue
J (2005) Uniform Abel-Kreiss boundedness and the spaces (Russian), Izv. Akad Nauk SSSR Ser Mat 25:
extremal behaviour of the Volterra operator. Proc 499–530; English translation: Am Math Soc Transl
Lond Math Soc 91(3):761–788 39(2), (1964), 1–36
Mugnolo D (2004) A semigroup analogue of the Fonf-Lin- Rosenblatt JM (1994) Norm convergence in ergodic theory
Wojtaszczyk ergodic characterization of reflexive and the behavior of Fourier Transforms. Canadian
Banach spaces with a basis. Stud Math 164:243–251 J Math 46:184–199
Müller V, Tomilov Y (2007) Quasisimilarity of power Rosenblatt JM, Wierdl M (1995) Pointwise ergodic theo-
bounded operators and Blum-Hanson property. rems via harmonic analysis. In: Ergodic theory and its
J Funct Anal 246:385–399 connections with harmonic analysis, London Mathe-
Müller V, Tomilov Y (2010) Weakly wandering vectors matical Society, Lecture notes series, 205. Cambridge
and interpolation theorems for power bounded opera- University Press, Cambridge, UK, pp 3–151
tors. Indiana Univ Math J 59:1121–1144 Rozendaal J, Veraar M (2018) Sharp growth rates for semi-
Mustafayev HS (2014) The behavior of the orbits of power groups using resolvent bounds. J Evol Equ 18:1721–
bounded operators. Oper Matrices 8:975–997 1744
Nagel R (1974) Ergodic and mixing properties of linear Rudolph DJ (1994) A joinings proof of Bourgain’s return
operators. Proc R Ir Acad 74:245–261 time theorem. Ergodic Theory Dyn Syst 14:197–203
Nagy B, Zemánek J (1999) A resolvent condition implying Ryll-Nardzewski C (1975) Topics in ergodic theory. In
power boundedness. Stud Math 134:143–151 Probability winter school, Karpacz 1975, Lecture notes
Nair R (1993) On polynomials in primes and J. Bourgain’s in Mathematics, vol 472. Springer, Berlin, pp 131–156
circle method approach to ergodic theorems, II. Stud Satō R (1979) The Hahn-Banach theorem implies Sine’s
Math 105:207–233 mean ergodic theorem. Proc Am Math Soc 77:426
Nasri-Roudsari D, Nessel RJ, Zeler R (1995) Resonance Satō R (1980) On the Blum-Hanson theorem in Lp. Math
principles with applications to mean ergodic theorems J Okayama Univ 22:27–32
and projection operators. Acta Math Hungar 68: Seifert D (2016) Rates of decay in the classical Katznelson-
269–285 Tzafriri theorem. J Anal Math 130:329–354
Nevanlinna O (1997) On the growth of the resolvent oper- Shields A (1978) On Möbius bounded operators. Acta Sci
ators for power bounded operators. In: Linear math (Szeged) 40:371–374
488 Operator Ergodic Theory

Sine R (1970) A mean ergodic theorem. Proc Am Math Soc Vũ QP (1992b) Theorems of Katznelson-Tzafriri type for
24:438–439 semigroups of operators. J Funct Anal 103:74–84
Sine R (1976) A note on the ergodic properties of homeo- Weiss G (1989) Weakly ‘p-stable linear operators are
morphisms. Proc Am Math Soc 57:169–172 power stable. Int J Systems Sci 20:2323–2328
Sine R (1991) Constricted systems. Rocky Mountain Weyl H (1910) Über die Gibbssche Erscheinung and
J Math 21:1373–1383 verwandte Konvergenzphänomene. Rend Circ Mat
Stampfli J (1972) A local spectral theory for operators III. Palermo 30:377–407; reprinted In: Weyl, Gesammelte
Trans Am Math Soc 168:133–151 Abhandlungen, Band I, Springer-Verlag, Berlin/
Stein EM (1982) The development of square functions in Heidelberg/New York, 1968, pp 321–353
the work of A. Zygmund. Bull Am Math Soc (NS) 7: Weyl H (1916) Über die Gleichverteilung von Zahlen mod.
359–376 Eins, Math Ann 77:313–352; reprinted In: Weyl,
Strikwerda J, Wade B (1997) A survey of the Kreiss matrix Gesanunelte Abhandhmgen, Band I, Springer-Verlag,
theorem for power bounded families of matrices and its Berlin/Heidelberg/New York, 1968, 563–599
extensions. In: Linear operators, Banach Center Publi- Wiener N, Wintner A (1941) Harmonic analysis and ergo-
cation, vol 38. IMPAN, Warsaw, pp 339–360 dic theory. Am J Math 63:415–426
Sucheston L (1976) Problems. In: Probability in Banach Wierdl M (1988) Pointwise ergodic theorem along the
spaces (Oberwolfach 1975), Springer lecture notes in prime numbers. Israel J Math 64:315–336
Mathematics, vol 526. Springer, Berlin/New York, Wintner A (1945) The linear difference equation of first
pp 285–290 order for angular variables. Duke Math J 12:445–449
Suciu L (2016) Estimations of the operator resolvent by Yoshimoto T (1993) On the speed of convergence in the
higher order Cesàro means. RM 69:457–475 (C,α) uniform ergodic theorem for quasi-compact oper-
Suciu L, Zemánek J (2013) Growth conditions and Cesàro ators. J Math Anal Appl 176:413–422
means of higher order. Acta Sci Math (Szeged) 79: Yoshimoto T (1996) On the convergence rate in the uni-
545–581 form ergodic theorem. J Math Anal Appl 200:149–161
Sz-Nagy B, Foiaş C (1967) Analyse harmonique des Yoshimoto T (1998) Uniform and strong ergodic theorems
opérateurs de l’espace de Hilbert, Masson, Paris; in Banach spaces. Ill J Math 42:525–543
Akadémiai Kiadó, Budapest. English translation: Har- Yosida K (1938) Mean ergodic theorem in Banach spaces.
monic analysis of operators on Hilbert space, North- Proc Imp Acad Tokyo 14:292–294
Holland, Amsterdam and Akadémiai Kiadó, Budapest, Yosida K, Kakutani S (1938) Applications of mean ergodic
1970. Second edition (with coauthors Bercovici and theorems to the problems of Markoff process. Proc Imp
Kérchy), Universitext, Springer, New York, 2010 Acad Tokyo 14:333–339
Tempelman A (1974) Ergodic theorems for amplitude Yosida K, Kakutani S (1941) Operator-theoretical treat-
modulated random fields. Litovsk Mat Sbor 14: ment of Markoff process and mean ergodic theorem.
221–229, (in Russian). Engl. transl.: Lithuan. Math. Ann Math 42:188–228
Transl 14 (1975), 698–704 Zaharopol R (1986) Mean ergodicity of power-bounded
ter Elst AFM, Müller V (2017) A van der Corput-type operators in countably order complete Banach lattices.
lemma for power bounded operators. Math Z 285: Math Z 192:81–88
143–158 Zemánek J (1994) On the Gelfand-Hille theorems. In:
Tomilov Y (2001) A resolvent approach to stability of Functional analysis and operator theory, Banach Center
operator semigroups. J Operator Th 46:63–98 Publication 30, Polish Academy of Sciences, Institute
Tomilov Y, Zemánek J (2004) A new way of constructing Mathematics, Warsaw, pp 369–385
examples in operator ergodic theory. Math Proc Camb Zsidó L (2007) Weak mixing properties of vector
Philos Soc 137:209–225 sequences. In: The extended field of operator theory,
Van Castren J (1980) A problem of Sz. Nagy. Acta Sci Operator theory advances and application, vol.
math (Szeged) 42:189–194 171, Birkhäuser, Basel, pp 361–388
van Neerven J (1996) The asymptotic behaviour of semi-
groups of linear operators, Operator theory: advances Books
and applications, vol 88. Birkhäuser, Basel Dunford N, Schwartz JT (1958) Linear operators, part I,
Visser C (1938) On the iteration of linear operations in a Interscience. Wiley, New York
Hilbert space. Neder Akad Wetensch 41:487–495 Eisner T, Farkas B, Haase M, Nagel R (2015) Operator
Volný D (2018) Martingale-coboundary representation for theoretic aspects of ergodic theory, Graduate texts in
stationary random fields. Stoch Dyn 18(2):1850011. mathematics, vol 272. Springer, Cham
18 pp Emel’yanov E (2007) Non-spectral asymptotic analysis of
von Neumann J (1932) Proof of the quasi-ergodic hypoth- one-parameter operator semigroups, Operator theory:
esis. Proc Natl Acad Sci U S A 18:70–82 advances and applications, vol 173. Birkhäuser, Basel
Vũ QP (1992a) A short proof of the Y. Katznelson’s and Engel KJ, Nagel R (2000) One-parameter semigroups for
L. Tzafriri’s theorem. Proc Am Math Soc 115: linear evolution equations, Graduate texts in mathemat-
1023–1024 ics, vol 194. Springer, New York
Operator Ergodic Theory 489

Foguel SR (1969) The ergodic theory of Markov pro- Krengel U (1985) Ergodic theorems. de Gruyter, Berlin
cesses. Van-Nostrand, New York Riesz F, Sz-Nagy B (1955) Leçons d’analyse
Goldstein J (1985) Semigroups of linear operators and fonctionnelle (French), 3rd edn. Gauthier-Villars,
applications, Oxford mathematical monographs. The Paris; Akadémiai Kiadó, Budapest. English transla-
Clarendon Press, Oxford University Press, New York. tion: Functional analysis, Dover Publications, Inc.,
Second ed. Dover Publications, Mineola, NY, 2017. New York, 1990
Halmos PR (1960) Lectures on ergodic theory. Chelsea Schaefer HH (1974) Banach lattices and positive operators,
Publishing Co., New York Die Grundlehren der mathematischen Wissenschaften,
Hille E, Phillips RS (1957) Functional analysis and semi- Band 215. Springer, New York/Heidelberg
groups, rev edn., vol 31. American Mathematical Soci- Yosida K (1980) Functional analysis, Sixth edition,
ety Colloquium Publications, American Mathematical Grundlehren der Mathematischen Wissenschaften,
Society, Providence vol 123. Springer, Berlin/New York
 Conjugacy Two actions G ↷ X1 and G ↷ X2 are
Dynamical Systems and C
-Algebras conjugate if there is a homeomorphism
F : X1 ! X2 such that F(gx) ¼ gF(x) for
T. Giordano and H.-C. Liao every x  X1; and g  G.
University of Ottawa, Ottawa, ON, Canada Continuous orbit equivalence Two group
actions are continuously orbit equivalent if
they are topologically orbit equivalent and the
Article Outline associated orbit cocycles are continuous.
Countable amenable group A countable group
Glossary that admits finite subsets that are almost invari-
Definition of the Subject ant under translation.
Introduction Dimension group An ordered abelian group that
Topological Orbit Equivalence is an inductive limit of simplicial groups. It is a
Mean Dimension, Small Boundary Property, and complete invariant for AF algebras.
Classification of C
-Algebras Étale equivalence relation An equivalence rela-
Boundary Actions and C
-Simplicity tion on a topological space with a topology
C
-Simplicity and Unique Trace Property such that the map (x, y) 7! y is a local homeo-
References morphism. For an action G ↷ X the associated
equivalence relation is given by x  y , gx ¼ y
Glossary for some g  G.
Free action An action G ↷ X whose stabilizer
AF algebra An inductive limit of finite- subgroup {g  G : gx ¼ x} is trivial for every
dimensional C
-algebras. x  X.
AF relation An inductive limit of finite equiva- G-boundary A topological space X equipped
lence relations. with a minimal G-action such that the induced
Affable equivalence relation An étale equiva- action of G on the space of regular Borel
lence relation that is orbit equivalent to an AF G-invariant probability measures on X is
relation. proximal.
Almost finite actions An action that admits dis- Isomorphism Two actions G1 ↷ X1 and G2 ↷ X2
joint towers such that each tower has a shape are isomorphic if there is a homeomorphism
that is almost invariant under translations and F : X1 ! X2 and a group isomorphism
the union of the towers almost covers the entire α : G1 ! G2 such that F(gx) ¼ α(g)F(x) for
space. every x  X1.
Bratteli diagram An infinite graph with a Minimal action An action G ↷ X where every
sequence of finite sets Vn of vertices and a orbit {gx : g  G} is dense in X.
sequence of finite sets En of edges connecting Proximal action An action G ↷ X where for
the vertices of Vn1 and Vn. every pair x1, x2 in X there is a net ti in
Bratteli–Vershik transformation A homeo- G such that limi tix1 ¼ limi tix2.
morphism of the path space of a Bratteli dia- Reduced group C
-algebras A C
-algebra
gram that can be viewed as a generalized constructed from the left regular representation
odometer. of a group.


C -algebra Algebra of operators on a Hilbert Shape The finite set {g1, . . ., gn} associated to a
space that is closed under adjoint and closed tower [ni¼1 gi V:
in the norm topology.

© Springer Science+Business Media, LLC, part of Springer Nature 2023 491

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_774
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_774-1
492 Dynamical Systems and C
-Algebras

Simple C
-algebra A C
-algebra that has no Introduction

nontrivial norm-closed ideals.
Strong orbit equivalence Two group actions are The appearance of operators on a Hilbert space in
strongly orbit equivalent if they are topologi- the study of dynamical systems can be traced back
cally orbit equivalent and the associated orbit to von Neumann’s mean ergodic theorem (von
cocycles have at most one point of Neumann 1932). It is perhaps not surprising that
discontinuity. von Neumann also laid the foundation of operator
Topological dynamical system A continuous algebras. In the seminal paper (Murray and von
group homomorphism from a topological Neumann 1936), the authors introduced the group
group G into the homeomorphism group of a measure space construction, which associates a
topological space X. It is also called a G-action (von Neumann) algebra to a measure-preserving
on X, denoted by G ↷ X, and X is called a group action on a measure space. Since then the
G-space. field of dynamical systems has never ceased to
Topological orbit equivalence for actions Two provide interesting and important examples in
group actions are topologically orbit equiva- operator algebras. Meanwhile, the study of orbit
lent if there is a homeomorphism F between equivalence in dynamics, initiated by Dye (1959),
the underlying topological spaces which maps was largely inspired by Murray and von
bijectively orbits onto the orbits. Neumann’s classification of hyperfinite II1 factors
Topological orbit equivalence Two étale equiv- (Murray and von Neumann 1943). This connec-
alence relations R1 and R2 are topologically tion extends to the much broader setting of ame-
orbit equivalent if there is a homeomorphism nable operator algebras and amenable equivalence
F between the underlying topological spaces relations, as seen in the Connes–Haagerup classi-
such that F  F(R1) ¼ R2. fication of amenable simple von Neumann alge-
Topologically free action An action where the bras (Connes 1976; Haagerup 1987) and the
set of points x for which the stabilizer subgroup Connes–Feldman–Weiss theorem for countable
is trivial is dense in X. amenable Borel equivalence relations (Connes
Tower A finite disjoint union [ni¼1 gi V of trans- et al. 1982). Beyond the amenable case, the
late of a set V in a G-space. study of von Neumann algebras has led to remark-
Unique trace property A C
-algebra has the able results in orbit equivalence rigidity (see, e.g.,
unique trace property if it has exactly one Ioana (2013, 2018) for surveys on this topic).
tracial state. Another striking analogy between the two
fields is the interplay between measure and topol-
ogy. Needless to say, measurable dynamics and
Definition of the Subject topological dynamics are two branches in dynam-
ical systems that are intimately related. On the
In (1936), Murrary and von Neumann showed that operator algebras side, von Neumann algebras
factors, the building blocks of von Neumann alge- can be seen as noncommutative measure-theoretic
bras, are of three different types. Defining the objects, while C
-algebras have a more topologi-
group measure space construction, which associ- cal flavor. From this viewpoint, it is only natural
ates a von Neumann algebra to a measurable that the research on linkage between topological
dynamical system, they constructed examples of dynamics and C
-algebras has seen a lot of pro-
factors of type II and III. Since then the interplay gress in the past decades, and this is what the
between von Neumann and C
-algebras and the present survey focuses on.
theory of measurable and of topological dynami- In the section “Topological Orbit Equivalence”
cal systems has continued to grow. we review the theory of topological orbit equiva-
In this survey we review some recent develop- lence for actions of finitely generated abelian
ments between C
-algebras and topological groups on the Cantor set. Similar to the measur-
dynamics. able case, the language of equivalence relations is
Dynamical Systems and C
-Algebras 493

pivotal in classifying such actions. Moreover, the orbit equivalent to a ℤ-action. This conjecture
classification of AF (approximately finite) rela- was proved by Ornstein and Weiss (1980). The
tions, which provides the base for classifying most general case was proved by Connes,
more general actions, relies on the classification Feldman, and Weiss (1982) by establishing that
of AF C
-algebras. This again parallels the influ- an amenable nonsingular countable equivalence
ence von Neumann algebras had on the study of relation R can be generated by a single transfor-
orbit equivalence in the measurable case. In this mation, or equivalently, is hyperfinite, that is, R
section, we also describe some of the recent is, up to a null set, a countable increasing union of
results on continuous orbit equivalence rigidity. finite equivalence relations.
In the section “Mean Dimension, Small For the Borel case, Weiss proved that actions
Boundary Property, and Classification of C
Alge- of ℤn are (orbit equivalent to) hyperfinite Borel
bras,” we discuss recent developments in classifi- equivalence relations, whose classification was
cation of amenable C
-algebras and the notion of obtained by Dougherty, Jackson, and Kechris
“regularity.” The main theme here is to connect (1994). It is not yet known if an arbitrary Borel
regularity at the dynamical level to the C
-alge- action of a discrete amenable group is orbit
braic level via the crossed product construction. It equivalent to a ℤ-action. More generally, hyper-
turns out that certain natural and purely dynamical finiteness was proved for Borel actions of any
questions, such as the existence of Kakutani– finitely generated group with polynomial growth
Rokhlin type decomposition for actions of more by Jackson–Kechris–Louveau (2002) and of any
general groups, have a direct impact on the study countable abelian group by Gao and Jackson
of the associated C
-algebras. Dynamical ideas (2015). In a recent preprint (Conley et al.
such as mean dimension and comparing sets by 2020), this result was extended to Borel actions
invariant measures are also crucial in this type of of polycyclic groups and for free actions of a
connections. At the same time, the study of regu- large class of solvable groups including the
larity of crossed product C
-algebras also led to Baumslag–Solitar group BS(1,2) and the lamp-
interesting dynamical results, such as a new char- lighter group.
acterization of the small boundary property in In this chapter, we present some of the devel-
terms of certain kind of tower decomposition. opments in topological orbit equivalence,
Finally, the section “Boundary Actions and C
- reviewing in particular the classification up to
Simplicity” is devoted to some of the recent (topological) orbit equivalence of minimal actions
breakthroughs that apply boundary actions to the of finitely generated abelian groups on the Cantor
study of group C
-algebras. More precisely we set. For connected spaces, using a result of
review the notion of Furstenberg boundary, dis- Sierpinski (see Kuratowski (1968), Ch V,
cuss how it can be realized by an operator alge- 47, III), any orbit equivalence is also an isomor-
braic construction (the so-called Hamana phism. Therefore, we consider only spaces that
boundary), and outline how this insight led to are totally disconnected. The strategy in the topo-
solutions to many long outstanding questions logical case follows the one used both in the
about group C
-algebras. measurable and in the Borel case:

1. Provide an invariant of (topological) orbit

Topological Orbit Equivalence equivalence for topological dynamical sys-
tems, which include actions of countable
In 1959, H. Dye (1959) introduced the notion of groups and étale equivalence relations. The
orbit equivalence and proved that any two ergodic definitions of étale equivalence relations and
finite measure-preserving transformations on a of the invariant, which is an ordered abelian
Lebesgue space are orbit equivalent. In Dye group, in the subsections “Étale Equivalence
(1963), he had also conjectured that an arbitrary Relations” and “Invariants of Étale Equiva-
ergodic action of a discrete amenable group is lence Relations.”
494 Dynamical Systems and C
-Algebras

2. Consider the rich and tractable class of AF review the notion of continuous orbit equivalence
(approximately finite) equivalence relations which is in a one-to-one correspondence with a
on the Cantor set. In the subsection “AF- strong notion of isomorphism of the associated
Equivalence Relations and Bratteli–Vershik C
-algebras. This domain of research is very
Transformations,” we recall their definition active and we hope our description, even if it is
and that they can always be realized as tail very short, will show its importance. In “ℤd-
equivalence on a Bratteli diagram. In the sub- odometers” and “Cohomology of ℤd-odometers”,
section “The Bratteli–Vershik Model,” we give we review the results on ℤd-odometers.
the definition of the Bratteli–Vershik system In the measurable case, full groups were intro-
introduced by R. Herman, I. Putnam, and duced in 1963 by H. Dye in his study of orbit
C. Skau (1992) and state their remarkable the- equivalence (1959, 1963). Several different full
orem: any Cantor minimal system is conjugate groups can be associated to a topological dynam-
to a Bratteli–Vershik system. There exist now ical system (on the Cantor set). In particular, the
several detailed presentations and surveys of topological full groups, introduced in Giordano
this very important result (see Remark 4). In et al. (1999) and Tomiyama (1996) for ℤ-actions,
the subsection “The Classification up to Iso- have been intensively studied and outstanding
morphisms of AF-Equivalence Relations: The results obtained. We will not present them here,
Bratteli–Elliott–Krieger Theorem,” we recall as several remarkable surveys have recently been
the classification up to isomorphism of mini- written (de Cornulier 2014; Matui 2017;
mal AF relations on the Cantor set. Katzlinger 2019).
In “Orbit Equivalence of AF-Equivalence Let Y be a compact metric space and G be an
Relations”, we classify, up to orbit equiva- infinite countable group represented on Y as a
lence, minimal AF-equivalence relations. group of self-homeomorphisms. Recall that the
A key technical result for this step is the topological dynamical system (Y, G) is minimal
absorption theorem for minimal AF-relations. if any of the following equivalent conditions are
This theorem, which we describe in “The satisfied:
Absorption Theorem”, states that a “small”
extension of a minimal AF relation is orbit (a) The empty set ; and Y are the only G-invariant
equivalent to the original AF relation. closed subsets of Y.
3. Prove that equivalence relations arising from (b) For every y  Y, its G-orbit OrbG{y} ¼ {gy|
minimal actions of finitely generated abelian g  G} is dense in Y.
groups are affable, that is, orbit equivalent to (c) For every nonempty open set U  Y, there
an AF-equivalence relation. We describe these exists a finite subset
results in “Orbit Equivalence of Minimal k
Actions of a Finitely Generated Abelian f g1 , g2 , . . . , gk g  G with [ gj U ¼ Y:
j¼1
Group”.
The following example is a key model of min-
In “A Topological Krieger Theorem: The imal homeomorphism of the Cantor set.
Notion of Strong Orbit Equivalence”, we describe
links between Cantor minimal systems and the Example 1 Let a ¼ ðan Þn1 , with an  ℕ,
C
-algebras associated to them. We introduce the an  2, and X be the compact abelian group
notion of strong orbit equivalence and describe a f0, 1, . . . , an  1g (under the operation of
topological Krieger theorem. We mention also n1
Jewett–Krieger-type realization results proved by addition mod an at the n-th coordinate, with
Ormes (1997). carry over to the right). Then the a-adic adding
In “Continuous Orbit Equivalence” and “Con- machine, given by ’(x) ¼ x þ (1,0,0, . . .), x  X,
tinuous Orbit Equivalence and C
-algebra”, we is a minimal homeomorphism of the Cantor set.
Dynamical Systems and C
-Algebras 495

Cantor minimal systems form a very rich class structure, (R, t) is a locally compact (principal)
of topological dynamical systems. For example, groupoid, cf. Renault (1980).
recall that Ellis (1960) proved, for any countable
(infinite), discrete group, the existence of a free, Definition 1 (Étale equivalence relation). The
action on a compact metric space and therefore on locally compact principal groupoid (R, t) on the
the Cantor set by the following G-equivariant compact metric space X is an étale equivalence
version of Alexander–Urysohn result (see relation if r : R ! X is a local homeomorphism,
Giordano and de la Harpe (1997) for a proof of that is, for all (x, y)  R, there exists an open
this result). neighborhood U(x, y)  t of (x, y) so that r(U(x,
y)) is open in X and r : U(x, y) ! r(U(x, y)) is a
Proposition 1 Let α be a continuous action of a homeomorphism. In particular, r is an open map.
countable group G on a metrizable compact
space Y. Example 2 Let G be a countable discrete group
Then there exists a continuous action a of G on acting freely on a locally compact metric space X.
the Cantor set X and a factor map w : X ! Y Let
(i.e., an equivariant continuous surjective map).
If, moreover, Y is an infinite set and α is a RG ¼ fðx, gxÞjx  X,jg  G  X  Xg
minimal action, then a can be chosen minimal.
that is, the RG-equivalence classes are simply the
Étale Equivalence Relations G-orbits. Topologize RG by transferring the prod-
We first recall the definition and the first properties uct topology on X  G to RG via the bijection
of étale equivalence relations (for more details (x, g) ! (x, gx). As G is discrete, RG is an étale
see, e.g., Renault (1980); Paterson (1999); Put- equivalence relation. (If G does not act freely, we
nam (2010)) and recall the notions of isomor- get a bijection between RG and a closed subset of
phism and orbit equivalence of étale equivalence X  G  X by the map (x, gx) ! (x, g, gx) and we
relations. transfer the product topology on X  G  X to
Let X be a compact, metric space and R be a RG.)
countable equivalence relation on X, that is,
R  X  X is an equivalence relation so that for Remark 1
all x  X, each equivalence class R[x] ¼ {y  X| 1) This definition of étaleness is equivalent
(x, y)  R} is at most countable. to the various definitions of an étale
Recall that R has a natural (principal) groupoid (or r-discrete) locally compact groupoid that
structure, with unit space equal to the diagonal can be found in the literature (e.g., Renault
Δ ¼ {(x, x)  R| x  X}. More specifically, if (1980), Definition 2.6 and Proposition 2.8
(x, y), (y, z)  R, then (x, y)(y, z) ¼ (x, z) is the and Paterson (1999), Definitions 2.2.1 and
product of this composable pair and the inverse of 2.2.3). The existence of an (essential) unique
(x, y)  R is the pair (y, x). The unit space Δ of R Haar system consisting of counting measures
is by definition {(x, y)(x, y)1| (x, y)  R} ¼ follows from our definition (see Paterson
{(x, x)| x  X}. The range map r : R ! X and (1999), Proposition 2.2.5).
the source map s : R ! X are defined by r(x, y) ¼ 2) A countable equivalence relation on the Cantor
y and s(x, y) ¼ x, respectively, where (x, y)  R set may be given nonequivalent étale topolo-
and both maps are surjective. gies ti, i ¼ 1, 2, contrasting with the situation
Assume that R is given a Hausdorff locally in the countable (standard) Borel equivalence
compact, second countable (hence metrizable) relation setting, where the Borel structure is
topology t, so that the product of composable uniquely determined by the inclusion of R in
pairs (with the topology inherited from the prod- X  X.
uct topology on R  R) is continuous. Also, the 3) A countable equivalence relation R on a com-
inverse map on R is a homeomorphism. With this pact metric space X cannot always be endowed
496 Dynamical Systems and C
-Algebras

with an étale topology. For example, the equiv- Definition 2 (Isomorphism and Orbit Equiva-
alence relation on X ¼ [0, 1] given by lence). Let (X1, R1, t1) and (X2, R2, t2) be two
étale equivalence relations on compact,
R ¼ D [ fð0, 1Þ, ð1, 0Þg
metrizable spaces X1 and X2. Then
cannot be endowed with an étale topology t.
1) (X1, R1, t1) and (X2, R2, t2) are orbit equiv-
If t were an étale topology on R, then there
alent if there exists a homeomorphism
would exist an open neighborhood U of (0, 1)
h : X1 ! X2, called an orbit equivalence,
in R such that r : U ! r(U ) is a local homeo-
such that
morphism. As the unit space Δ is closed,
{(0, 1), (1, 0)} is open in R, and therefore ðx, yÞ  R1 if and only if ðhðxÞ, hðyÞÞ  R2 :
U \ {(0, 1), (1, 0)} is also open and the
restriction of r to U \ {(0, 1), (1, 0)} is a 2) If moreover h : X1 ! X2 can be chosen such
local homeomorphism. Thus, either r|{(0,1)} that h  h : (R1, t1) ! (R2, t2) is a homeo-
or r|{(0,1),(1,0)} is open in [0, 1]. morphism, (X1, R1, t1) and (X2, R2, t2) are
Contradiction! isomorphic.

Example 3 Let X be a compact metrizable space Proposition 3 Let (X1, G1) and (X2, G2) be two
and (R, t) be a compact étale equivalence relation free actions of countable discrete groups G1 and
on X. Then (see Giordano et al. (2004), Proposi- G2 on compact metrizable spaces X1 and X2. Then
tion 3.2),
1) (X1, G1) is orbit equivalent to (X2, G2) if and
1) t is the relative topology trel from R  X  X. only if RG1 and RG2 are orbit equivalent.
2) There exists M  1 such that, for all x  X, #R 2) If (X1, G1) is isomorphic to (X2, G2) (i.e., there
[x]  M, where #R[x] is the cardinality of exist a homeomorphism h : X1 ! X2 and a
R[x]. group isomorphism α : G1 ! G2 such that
h(gx) ¼ α(g)(hx), for all g  G1 and
Therefore, if (R, t) is an étale equivalence x  X1), then RG1 and RG2 are isomorphic.
relation and R contains an infinite equivalence
class, then t is not the relative topology from Remark 2 The converse of the implication
R  X  X. (2) does not hold as shown by H. Dahl (2008),
where she constructs free minimal actions of non-
The following proposition generalizes the isomorphic locally finite groups which give iso-
Feldman–Moore characterization of countable morphic equivalence relation.
(standard) Borel equivalence relations given in
Feldman and Moore (1977), Theorem 1): By a theorem of Sierpinski (Kuratowski
(1968), Theorem 6, Ch. V, §47, III), orbit equiv-
Proposition 2 (Giordano et al. (2004), Proposi- alence implies isomorphism in the following case:
tion 2.3). Let (R, t) be an equivalence relation on
a zero-dimensional set X. Then there exists a Theorem 1 Let (X1, G1) and (X2, G2) be two free
countable group G of homeomorphisms of actions of countable discrete groups G1 and G2 on
X such that R ¼ RG. compact metrizable spaces X1 and X2.
If the two actions are orbit equivalent and if X1
Note that it is not always possible to find a is connected, then they are isomorphic.
group G acting freely on X and such that R ¼
RG as shown by Hjorth and Molberg (2006). Let R  X  X be an étale equivalence relation.
There are two natural notions of equivalence Then as defined by J. Renault (1980), a Borel
between étale equivalence relations. measure m on X is R-invariant if mðr ðU ÞÞ ¼
Dynamical Systems and C
-Algebras 497

mðsðU ÞÞ, for every bisection U  R. Recall that U (1) B(X, R) is the subgroup of C(X, ℤ) generated
is a bisection of an étale equivalence relation (R, t) by all functions wrðUÞ  wsðU Þ , for a compact
if it is open and the restriction of the range and the bisection U  R:
source map to U are homeomorphisms. The set of (2) Bm(X, R) ¼ { f  C(X, ℤ)| f dm ¼ 0,
all bisections forms a basis for the topology t (see, 8m  M(X, R)}.
e.g., Renault (1980)). Let us denote by M(X, R) the
compact convex cone of R-invariant probability Remark 3
measures on X. Using that R ¼ RG for some (a) If MðX, RÞ ¼ 0, then Bm(X, R) ¼ C(X, ℤ).


countable group G, the following proposition fol- (b) If m  M(X, R), then by definition
lows from a straightforward adaptation of the proof mðr ðU ÞÞ ¼ mðsðU ÞÞ for any compact bisec-
for Cantor minimal systems (see Giordano et al. tion U  R: Hence, B(X, R) is a subgroup
(1995), p. 80). A direct and detailed proof for étale of Bm(X, R).
equivalence relations can be found in Putnam
(2010, Theorem 2.8). Definition 4 Let (X, R) be as above. We define
the two preordered abelian groups:
Proposition 4 Let (X1, R1) and (X2, R2) be two
orbit equivalent étale equivalence relations on (1) D(X, R) ¼ C(X, ℤ)/B(X, R), where
compact, metrizable spaces X1 and X2. Then
there exists a homeomorphism h : X1 ! X2 D(X, R)+ ¼ {[ f]j ∃ g  C(X, ℤ), g  0 such
which implements a bijection between that f  g  B(X, R)}.
M(X1, R1) and M(X2, R2).
(2) Dm(X, R) ¼ C(X, ℤ)/Bm(X, R), where
Invariants of Étale Equivalence Relations
To an étale equivalence relation (X, R) on a Dm ðX, RÞþ ¼ ½ f m j∃g  CðX, ℤÞ, g  0
totally disconnected, compact, Hausdorff space,
we associate two preordered abelian groups such that ð f  gÞ dm ¼ 0, 8m  MðX, RÞ :
D(X, R) and Dm(X, R) which are invariants of,
respectively, isomorphism and orbit equivalence. If (X, R) is an étale equivalence relation on a
By a preordered group we mean a countable totally disconnected compact metrizable space X,
abelian group G with a subsemigroup then by Remark 3 (b), there is a canonical positive
G+containing 0, called the positive cone, such that homomorphism πm from D(X, R) onto Dm(X, R).
By integration, any m  M(X, R) defines a
ðiÞ Gþ þ Gþ  Gþ , ðiiÞ G ¼ Gþ  Gþ : state Im on D(X, R), that is, a positive homomor-
phism from D(X, R) to ℝ such that Im([1]) ¼ 1,
If moreover G+ \ (G+) ¼ {0}, then G is an
and as any state on D(X, R) is of this form, we
ordered group.
have
An order unit for (G, G+) is an element u  G+
such that for all g  G, there is n  ℕ with nu 
Proposition 5 Let X be a totally disconnected,
g  G+.
compact metrizable space and R be an étale
Recall that M(X, R) denotes the convex set of
equivalence relation on X.
all R-invariant probability measures on X.
Then there is a bijective correspondence
between the set M(X, R) of R-invariant probabil-
Definition 3 Let (X, R) be an étale equivalence ity measures on X and the set S(D(X, R), [1X]) of
relation on a totally disconnected compact states on (D(X, R), D(X, R)+, [1X]).
metrizable space X, and C(X, ℤ) be the countable
abelian group of continuous functions from X to By Remark 3 (a), if MðX, RÞ ¼ 0, then Dm(X,


ℤ. Then R) ¼ 0, and as the infinitesimal subgroup Inf(D

498 Dynamical Systems and C
-Algebras

(X, R)) of D(X, R) is {[ f]  D(X, R)j t([ f]) ¼ has torsion (see Gähler et al. (2013) or Matui
0 for all t  S (D(X, R), D(X, R)+, [1X])}, we get (2008b)). Hence, it is not a torsion free group.

Proposition 6 Let (X, R) be as above. Then AF-Equivalence Relations and Bratteli–Vershik

Transformations
(1) The infinitesimal subgroup of D(X, R) is The AF-equivalence relations (Renault 2003;
equal to Bm(X, R)/B(X, R) Giordano et al. 2004) form a rich, but also tracta-
(2) D(X, R)/Inf (D(X, R)) is canonically order ble class of étale equivalence relations. The ter-
isomorphic to Dm(X, R) minology AF comes from C
-algebra theory and
means approximately finite.
Then we get (see, e.g., Giordano et al. (2018),
Proposition 20.4.36 and 20.4.40]) Definition 5 An étale equivalence relation R on
a space X is an AF-equivalence relation if X is a
Proposition 7 If h : X ! X 0 is an orbit map totally disconnected compact metrizable space
between two étale equivalence relations (X, R) and if R is the union of an increasing sequence
and (X 0, R0) on totally disconnected compact of compact open subequivalence relations.
metrizable spaces X and X 0, then it induces a
unital order isomorphism from (Dm(X, R), [1X]m) For a totally disconnected, compact, Hausdorff
onto Dm ðX0 , R0 Þ, ½1X0 m : space, the above definition of an AF-equivalence
Moreover if h implements an isomorphism relation is equivalent to the definition of an
between (X, R) and (X 0, R0), then it induces an approximately proper (AP) equivalence relation
order isomorphism from D(X, R) onto D(X 0, R0) introduced by Renault (2003).
preserving the order units. Let G be a countable group acting minimally
and freely on the Cantor set X. If G is locally finite,
Example 4 then it is the union of an increasing sequence of
finite groups (Gn)n1. The associated étale equiv-
1. Let ’ be a minimal homeomorphism of the alence relation RG can be identified with X  G
Cantor space X and R’ ¼ {(x, ’n(x))| x  X, and is equal to the inductive lim RGn ¼
!
n  ℤ} be the corresponding étale equivalence lim ðX  Gn Þ: As for n  1,RGn is clearly a com-
!
relation. Then one checks that B’ ¼ { f 
pact étale equivalence relation and RGn  RGnþ1 ,
f ∘ ’| f  C(X, ℤ)} and therefore
it follows that RG is an AF-equivalence relation.
D X, R’ ¼ CðX, ℤÞ=f f  f ∘’jf  CðX, ℤÞg: The converse is also true.

Putnam (1989) in his study of the C
-algebra Theorem 2 (Giordano et al. (2004), Theorem
associated to (X, ’) showed that D(X, R’) is a 3.8). Let G be a countable group acting minimally
dimension group (see below for the precise and freely on the Cantor set X. Then RG is AF if
definition), and Poon (1989) showed that and only if G is locally finite.
D(X, R’) is an ordered group if ’ is topologi-
cally transitive. In a recent prepublication, AF-equivalence relations satisfy the following
D. Handelman and M. Boyle show that if ’ is stability properties:
a homeomorphism of a compact metrizable
zero-dimensional space X, then ’ is chain Proposition 8 (Giordano et al. (2004), Proposi-
recurrent if and only if D(X, R’) is an tion 3.12).
unperforated abelian group.
2. For N  2, there are examples of minimal actions 1) An inductive limit of a sequence of
’ of ℤN on the Cantor set, such that D(X, R’) AF-equivalence relations on X is AF.
Dynamical Systems and C
-Algebras 499

2) Let (R, t) be an AF-equivalence relation and Endowed with the relative topology of X(V,E) 
R0  R be an open subequivalence relation. X(V,E), then RN is a compact étale equivalence
Then R0 , tjR0 is AF. relation (hence each equivalence class is finite),
and RN is an open subset of RNþ1, for all N  0.
Let us now describe the fundamental example Let R ¼ [N0RN, and give R the inductive
of an AF-equivalence relation. We begin with a limit topology. This means that a sequence
Bratteli diagram (see Herman et al. 1992; Effros {(xn, yn)} in R converges to (x, y) in R if and
1981). It is a locally finite, infinite directed graph only if {xn} converges to x, {yn} converges to
which consists of a vertex set V ¼ 1 n¼0 V n and an
y (in X), and, for some N, (xn, yn) is in RN for all
edge set E ¼ 1 E , where the Vn and the En’s
’s but finitely many n. Then R is an étale equiva-
n¼1 n
are finite disjoint sets. lence relation which we denote by AF(V, E).
The edges in En connect vertices in Vn1 with If (V 0, E0) is a Bratteli diagram obtained from
vertices in Vn. If e connects v  Vn1 with (V, E) by telescoping, then the étale equivalence
u  Vn, we write s(e) ¼ v and r(e) ¼ u, where relations AF(V, E) and AF(V 0, E0) are naturally
s : En ! Vn1 and r : En ! Vn are the source and isomorphic.
range maps, respectively. We will assume that The Bratteli diagram (V, E) is simple if there
s1(v) 6¼ ; for all v  V and that r1(v) 6¼ ; for exists a telescoping of (V, E) such that the
all v  V \V0. resulting Bratteli diagram (V 0, E0) has full connec-
Given a Bratteli diagram (V, E) and a sequence tion between all consecutive levels. Then (V, E) is
0 ¼ m0 < m1 < m2 < in ℤ+, the telescoping of simple if and only if every AF(V, E)-equivalence
(V, E) to {mn} is the Bratteli diagram (V 0, E0), class is dense in X(V,E).
where V 0n ¼ V mn and E0n ¼ Emn1 þ1 ∘ ∘Emn , The above example is in fact the general case.
and the source and the range maps are as above. Indeed, we have
To a Bratteli diagram (V, E), we associate its
infinite path space X(V,E) given by Theorem 3 (Giordano et al. (2004), Theorem
3.9). Let R be an AF-equivalence relation on a
totally disconnected compact metrizable space X.
ðe1 , e2 , . . .Þjei  Ei , r ðei Þ ¼ sðeiþ1 Þ8i  1 :
Then, there exists a Bratteli diagram (V, E)
such that R is isomorphic to the AF-equivalence
1
Clearly XðV,EÞ n¼1 En , and endowed with relation AF(V, E) on X(V,E).
the relative topology, X(V,E) is a compact, Moreover, (X, R) is minimal if and only if the
metrizable, totally disconnected space. Bratteli diagram (V, E) is simple.
To a finite path p ¼ (e1, e2, . . ., en) starting at
v0  V0 is associated the cylinder set U( p) ¼ Recall that the dimension group G(V, E) asso-
{( f1, f2, . . .)  X(V,E)| fi ¼ ei, i ¼ 1, 2, . . ., n}. ciated to a Bratteli diagram (V, E) is the inductive
The cylinder sets are clopen sets and form a basis limit of the sequence of the simplicial groups
for the topology of X(V,E). If (V, E) is simple (see jV j
ℤjV n j , ðℤþ Þ n : Dimension groups were intro-
below for the definition), then X(V,E) has no iso-
lated points, and so X(V,E) is a Cantor set. (We will duced by G. A. Elliott (1976) in his classification
in the sequel disregard the trivial case where the of AF C
-algebras to denote any ordered group
path space X(V, E) is a finite set.) isomorphic to an inductive limit of a sequence of
For each N  0, let simplicial groups. The following result of Effros,
Handelman, and Shen (1980) provides an abstract
characterization of dimension groups.
RN ¼ ðe, f Þ  XðV,EÞ  XðV,EÞ jen
Theorem 4 A countable, ordered abelian group
¼ f n for all n > N : (G, G+) is a dimension group if and only if
500 Dynamical Systems and C
-Algebras

1) it is unperforated, i.e. for every g  G and Then they proved the following remarkable
every positive integer n, if ng  G+, then result:
g  G+, and
2) It has the Riesz interpolation property, that is, Theorem 6 (Herman et al. 1992). Let (X, ’) be a
for any a1, a2, b1, b2  G with a1, a2  b1, b2, Cantor minimal system. Then there exists a properly
there exists c  G with ai  c  bj, for i, ordered Bratteli diagram (V, E, ) such that the
j ¼ 1, 2. associated Bratteli–Vershik system (X(V,E), ’(V,E))
is conjugate to (X, ’).
Note that a dimension group being
unperforated is torsion free. Recall that an order Remark 4
ideal in an ordered group (G, G+) is a subgroup I a) Vershik (1981, 1982) used sequences of refin-
such that I ¼ I +  I +, where I + ¼ I \ G+, and ing measurable partitions of a measurable
if 0  a  b, with a  G and b  I , then a  I , space to realize ergodic automorphisms of a
and that an ordered group G is simple if its only measured space.
order ideals are {0} and G. b) There are several detailed presentations and
We then have (see Herman et al. 1992): surveys of the results of Herman et al. (1992),
see, for example, Skau (2000), Durand (2010);
Theorem 5 For a Bratteli diagram (V, E) and the Putnam (2010); Bezuglyi and Karpel (2016);
associated AF-equivalence relation AF(V, E) on Putnam (2018) and the monograph (Durand
X(V,E), the group D(X(V,E), AF(V, E)) is the dimen- and Perrin 2022).
sion group G(V, E) associated to (V, E). It is simple c) The case of aperiodic Cantor systems was stud-
if and only if AF(V, E) is minimal. ied in Bezuglyi et al. (2005) and
Medynets (2006).

The Bratteli–Vershik Model Let (X(V,E), ’(V,E)) be the Bratteli–Vershik sys-

Let (V, E) be a Bratteli diagram with V0 ¼ {v0}. In tem associated to a properly ordered Bratteli dia-
(Herman et al. 1992), Herman, Putnam and Skau gram (V, E, ), and let AF(V, E) be the associated
introduce the notion of an ordered Bratteli dia- AF-equivalence relation on X(V,E). By construc-
gram by defining a partial order  on E so that tion AF ðV, EÞ  R’ðV,EÞ and by the techniques
edges e, e0  E are comparable if and only if
developed in Putnam (2010), we have
r(e) ¼ r(e0). They then show that any simple
Bratteli diagram can be properly ordered, that
Theorem 7 (Herman et al. (1992), Cor. 5.4). Let
is, there exists a partial order on E with a unique
(X(V,E), ’(V,E)) be the Bratteli–Vershik system
min path xmin ¼ (e1, e2, . . .) and a unique max
associated to a properly ordered Bratteli diagram
path xmax ¼ ( f1, f2, . . .) in X(V,E) (i.e., ei
(V, E, ), and let AF(V, E) be the associated
is a minimal edge and fi a maximal one for all
AF-equivalence relation on X(V, E).
i ¼ 1, 2, . . .).
Then D(X, ’) is order isomorphic to
As a properly ordered Bratteli diagram is sim-
D(X(V,E), AF(V, E)) with distinguished order units.
ple, its path space X(V,E) is a Cantor set. Then in
(Herman et al. 1992), they defined a generalized
Using now the Effros–Handelman–Shen The-
odometer on X(V,E) as follows:
orem (Effros et al. 1980), we obtain

Definition 6 Let X(V,E) be the path space of a Theorem 8 (Herman et al. (1992), Cor. 6.3). Let
properly ordered Bratteli diagram. Then the (X, ’) be a Cantor minimal system. Then
Bratteli–Vershik transformation ’(V,E) on X(V,E) (D(X, ’), D(X, ’)+, [1X]) is a unital, simple, acy-
is defined by: if x ¼ (e1, e2, . . .)  X(V,E) and clic (i.e., not order isomorphic to ℤ) dimension
x 6¼ xmax, then ’(V,E)(x) is the successor of x in the group. Furthermore, if (G, G+) is any simple, acy-
lexicographic ordering and ’(V,E)(xmax) ¼ xmin. clic dimension group with distinguished order unit
Dynamical Systems and C
-Algebras 501

u, there exists a Cantor minimal system (X, ’) so 2. R-thin if m(Y ) ¼ 0 for all R-invariant proba-
that (G, G+, u) ffi (D(X, ’), D(X, ’)+, [1X]). bility measures m on X.

The Classification up to Isomorphisms of Remark 5

AF-Equivalence Relations: The Bratteli–Elliott– 1) In Lemma 2.5 of Phillips (2005), the author
Krieger Theorem gives a measure-free definition of an R-
For AF-equivalence relations, the invariants intro- thin set.
duced in Definition 4 have not only a well-known 2) If B ¼ (V, E) is a Bratteli diagram and C ¼
structure, but they also form complete sets of (W, F) a subdiagram (i.e., W  V, F  E, and
invariants of AF-equivalence relations on the W0 ¼ V0 ¼ {v0}), then the closed subset XC of
Cantor set up to isomorphism and in the minimal XB is an étale subset, and Theorem 3.11 of
case up to orbit equivalence. This result is based Giordano et al. (2004) shows that any R-étale
on the famous Bratteli–Elliott classification of AF subset of an AF-equivalence relation R is of
C
-algebras (Bratteli 1972; Elliott 1976) and on this form.
Krieger’s dynamical version (Krieger 1979).
Theorem 10 (Matui (2008a), Theorem 3.2). Let
Theorem 9 Let (X1, R1) and (X2, R2) be two (X, R) be a minimal AF-equivalence relation on
AF-equivalence relations on the Cantor set. the Cantor set X and let Y  X be a closed, R-étale
Then they are isomorphic if and only if there exists and R-thin subset. Suppose that an
an unital order group isomorphism from AF-equivalence relation Q  Y  Y is an étale
D(X1, R1) onto D(X2, R2). extension of R|Y (i.e., R|Y  Q and the inclusion
map is continuous).
The Absorption Theorem
If R ¼ R _ Q denotes the equivalence relation
The absorption theorem is the key tool used to
on X generated by R and Q, then there exists a
classify up to orbit equivalence minimal
homeomorphism h : X ! X such that
AF-equivalence relations and étale equivalence
relations associated to minimal actions on the
Cantor set of finitely generated abelian groups. (i) ðh  hÞ R ¼ R. In particular, R is orbit
This result shows that a “small” extension of a equivalent to R.
minimal AF-equivalence relation is orbit equiva- (ii) h(Y) is a closed, R-étale, and R-thin subset.
lent to the original one. (iii) h|Y  h|Y : Y  Y ! h(Y )  h(Y ) is a
There are three different versions of the absorp- homeomorphism from Q to R|h(Y ).
tion theorem. The first version proved in Giordano
et al. (2004) had to be generalized in Giordano et al. In the three versions of the absorption theorem,
(2008) for the classification up to orbit equivalence the strategy of the proof consists of constructing
of free minimal actions of ℤ2. It had to be strength- countable disjoint replicas of R|Y and Q, and then
ened again in order to generalize the results for using the following strong version of Bratteli–
minimal ℤ2-actions to minimal ℤd-actions. This Elliott–Krieger’s result.
final version below is due to H. Matui (2008a).
To state the theorem, let us introduce the ter-
minology we will need to define the notion of a Lemma 1 (Giordano et al. (2004), Lemma 4.15).
“small extension.” If (X, R) is an étale equiva- For i ¼ 1, 2, let (Xi, Ri) be two isomorphic min-
lence relation on the Cantor set X, then a closed imal AF-equivalence relations on the Cantor sets,
subset Y of X is and Zi  Xi be Ri-thin, and Ri-étale closed
subsets.
1. R-étale if the restriction R \ (Y  Y ) of R to Y, Assume that there exists a homeomorphism
denoted by R|Y, is an étale equivalence relation α : Z1 ! Z2, which implements an isomorphism
in the relative topology; between R1 jZ1 and R2 jZ2 :
502 Dynamical Systems and C
-Algebras

Then there exists an extension a : X1 ! X2 of As Ry _ Q ¼ R’, by the absorption theorem Ry

α which implements an isomorphism between R1 and R’ are orbit equivalent, and therefore we have
and R2.
Theorem 11 Any minimal ℤ-action on the Can-
Definition 7 (Giordano et al. (2004), Definition tor set is affable.
4.1). Let R be a countable equivalence relation on
Orbit Equivalence of AF-Equivalence Relations
the Cantor set X. We say that R is affable if it may
The classification up to orbit equivalence of min-
be given a topology t so that (R, t) is an
imal AF-equivalence relation will be a corollary
AF-equivalence relation or equivalently if R is
of the following theorem proved in 2010 by
orbit equivalent to an AF-equivalence relation.
I. Putnam (2010).
Remark 6 Keeping the notation of Theorem 10, Theorem 12 Let (X, R) be a minimal
and calling the extension R of the minimal AF-equivalence relation on the Cantor set X.
AF-equivalence relation R a “small” extension, Then there exists a minimal AF-equivalence
then the absorption theorem states in particular relation R on X, containing R and such that
that a “small” extension of a minimal
AF-equivalence is not only affable, but orbit
(1) R is orbit equivalent to R.
equivalent to R.
(2) the two unital dimension groups D X, R
Example 5 Let us use the absorption theorem to and Dm(X, R) are order isomorphic.
prove, for any Cantor minimal system (X, ’), the
affability of its associated étale equivalence Sketch of the main steps of the proof of the
relation theorem:

R’ ¼ x, ’k ðxÞ , x  X and k  ℤ : (1) By the Effros–Handelman–Shen Theorem 4,

there exists a Bratteli diagram (W, F) such that
Let (Un)n1 be a decreasing sequence of clopen GðW, FÞ ¼ DðX, RÞ=Inf ðDðX, RÞÞ ¼ Dm ðX, RÞ:
subsets of X, whose intersection is a single point y,
and for n  1, let Rn denote the equivalence (2) The key technical part (and a real “tour de
relation on X generated by {(x, ’(x))| x  X force”) of Ian Putnam’s proof is the construc-
\Un}. As ’ is minimal and as the first return map tion of a “nice” Bratteli diagram (V, E) and of
of ’ on Un is continuous, we have that Rn is a graph homomorphism: h : (V, E) ! (W, F)
compact and open. As (Un)n1 forms a decreasing with the following properties:
sequence of clopen sets, the sequence of the (a) G(V, E) ¼ D(XE, RE) ¼ D(X, R).
equivalence relations (Rn)n1 is increasing and (b) h induces a homeomorphism h : XE ! XF
their union Ry is an AF-relation. Every Ry-class such that h  h : RE ! RF is continuous
is also a ’-orbit, except for the orbit of the point and open.
y and R’ ¼ Ry _ {(y, ’( y))}. (c) There exists a subdiagram Δ of the
Then Y ¼ {y, ’( y)} is evidently a closed and Bratteli diagram (V, E) such that Y ¼ XΔ
Ry-thin subset of X. As (y, ’( y)) Ry, we is a closed, RE-étale, RE-thin subset of XE
see that and an AF-equivalence relation Q R|Y
on Y  Y such that R ¼ R _ Q:
Ry Y
¼ DY ¼ fðy, yÞ, ð’ðyÞ, ’ðyÞÞg  Y  Y
By the absorption theorem (Theorem 10),
and Y is an Ry-étale subset of X. Let Q ¼ Y  Y.
R is orbit equivalent to R and h  h R ¼
Clearly Q is an AF-equivalence relation on
Y containing Ry|Y . RF : By transferring the topology of RF via
Dynamical Systems and C
-Algebras 503

ðh  hÞ1 , R becomes an AF-equivalence actions are affable, that is, orbit equivalent to an
relation containing R and whose unital AF-equivalence relation. The key tool for this step
dimension group is order isomorphic to is the absorption theorem for minimal AF rela-
Dm(XE, R). tions (Theorem 10).
We first describe the case of minimal ℤ-actions
As a consequence of Theorem 12 and Theorem studied in Giordano et al. (1995). Notice however
9, we obtain the classification up to orbit equiva- that the proof of the classification up to orbit
lence of AF-equivalence relations. equivalence of minimal ℤ-actions given in
Giordano et al. (1995) did not use the absorption
Corollary 1 (Putnam (2010), Cor. 3.2). For a theorem (see Remark 7).
minimal AF-equivalence relation (X, R) on the We then discuss the affability of minimal
Cantor set X, its unital dimension group actions of ℤd on the Cantor set. The case d ¼ 2
was proved in Giordano et al. (2008) and the
ðDm ðX, RÞ, Dm ðX, RÞþ , ½1X Þ general case in Giordano et al. (2010). This
proof of the affability is extended to minimal
is a complete invariant of orbit equivalence. actions of finitely generated abelian groups by
using a result that ;. Johansen proved in his thesis
Remark 7 (Johansen 1998).
1) The classification of AF-equivalence relations By Theorem 14, any minimal ℤ-action on the
was originally given in Giordano et al. (1995), Cantor set is affable. Therefore, we get
as a consequence of the classification of min-
imal Cantor systems. The proof in Giordano Theorem 13 (Giordano et al. (1995), Theorem
et al. (1995) relies on nontrivial results in 2.2). For i ¼ 1, 2, let (Xi, Ri) be the two minimal
homological algebra. The new proof given by equivalence relations on the Cantor set, which are
Ian Putnam is certainly the “right” one, as it either AF or induced by a Cantor minimal system.
avoids completely the use of homological Then the following statements are equivalent:
algebra and is very dynamical in nature.
2) In Putnam (2010), the following generalized (1) (X1, R1) and (X2, R2) are orbit equivalent.
version of Theorem 12 is in fact proved: (2) The dimension groups Dm(X1, R1) and
Let (X, R) be a minimal AF-equivalence rela- Dm(X2, R2) are order isomorphic by a map
tion and H be a subgroup of Inf (D(X, R)) such preserving the distinguished order units.
that D(X, R)/H is torsion free. (3) There exists a homeomorphism F : X1 ! X2
Then there exists a minimal AF-equivalence carrying M(X1, R1) onto M(X2, R2).
relation R on X, containing R and orbit equiv-
alent to R and such that The equivalence between (1) and (2) is given by
Corollary 1, and the implication from (3) to (2)
follows directly from the definition of Bm(Xi, Ri)
D X, R ffi DðX, RÞ=H: (Definition 3). By Proposition 4, (1) implies (3).
By Proposition 1.16 and Corollary 4.2 of
Orbit Equivalence of Minimal Actions of a Effros (1981), we then get
Finitely Generated Abelian Group
The complete classification of minimal Corollary 2 (Giordano et al. (1995), Cor. 2.1).
AF-equivalence relations on the Cantor set up to Let (X1, ’1) and (X2, ’2) be two uniquely ergodic
orbit equivalence was given in Corollary 1. In this Cantor minimal systems. Then they are orbit
section, we extend this complete classification to equivalent if and only if {m1(E)| E  X1, clopen} ¼
minimal actions of finitely generated abelian {m2(E)| E  X2, clopen}, where mi is the unique
groups on the Cantor set by showing that these ’i-invariant measure on Xi for i ¼ 1, 2.
504 Dynamical Systems and C
-Algebras

Recall (see Giordano et al. (1995), p. 63 and This work was continued in a 2018 paper by
Putnam et al. (1986) for a survey) that a Denjoy Alvin, Ash, and Ormes (2018), with the additional
homeomorphism is an aperiodic homeomorphism assumption that the speedup function is bounded.
of 1 , which is not conjugate to a rigid rotation. They showed that a minimal bounded speedup of
By a Denjoy system, we mean a Denjoy homeo- an odometer is conjugate to an odometer.
morphism restricted to its unique invariant Cantor We now discuss the general case of actions of
set. A Denjoy system is uniquely ergodic. finitely generated abelian groups. Let ’ be a free
Let m be the invariant probability measure of a minimal action of ℤd on the Cantor set X and let
uniquely ergodic Cantor minimal system (X, ’), R’ be the associated étale equivalence relation.
and G be the subgroup {m(E); E  X, clopen} þ The affability of R’ is proved by constructing
ℤ of ℝ. By Corollary 2, sub-equivalence relations of R’:

Dm X, R’ , ½1X ffi ðG, 1Þ: R0  R1  . . . Rd ¼ R’

If G  ℚ, then it is the dimension group of an such that R0 is a minimal AF-equivalence relation

odometer. If G \ (R\ℚ) is nonempty, then there with the relative topology from R’, and each Ri is
exists a Denjoy transformation ’ with Dm(X, ’) ¼ a “small” extension of Ri1 for i ¼ 1, , d.
G (see Putnam et al. (1986)). Hence we have Then by applying d times the absorption theo-
rem, we get that Ri is orbit equivalent to Ri1, for
Corollary 3 (Giordano et al. (1995), Cor. 2.2). 1  i  d, and therefore R’ is orbit equivalent to
Let (X, ’) be a uniquely ergodic Cantor minimal R0, hence affable.
system. Then (X, ’) is orbit equivalent either to an The details of the construction of the minimal
odometer system or to a Denjoy system. AF subrelation R0 of R’ are in Giordano et al.
(2008) for d ¼ 2 and in Giordano et al. (2010) for
Remark 8 The study of speedups of ergodic, d  3. We then have
finite measure-preserving transformations started
in the late sixties. In Arnoux et al. (1985), they Theorem 14 Any free minimal action of ℤd on
showed that if T and S are any two ergodic, the Cantor set is affable.
measure-preserving automorphisms, there is a
speedup of T that is measurably conjugate to S, Let ’ be a minimal action on the Cantor set of a
or equivalently using Dye’s result (1959) that if finitely generated abelian group. By generalizing
T and S are two orbit equivalent, ergodic, a result of ;. Johansen (1998), R’ is orbit equiv-
measure-preserving transformations, then S is alent to an action of ℤd, with d  1. Then by
measurably conjugate to a speedup of T. Theorem 1.46 we get

In 2016, Ash studied in (2016) speedups of Theorem 15 Any minimal action of a finitely
Cantor minimal systems and showed that if generated abelian group on the Cantor set is
(Xi, ’i) are two minimal Cantor systems, affable.
(X2, ’2) is a speedup of (X1, ’1) if and only if
there is a unital surjective homomorphism F from As a consequence of Theorems 13 and 15, we
Dm(X2, ’2) onto Dm(X1, ’1) such that then have

F Dm ðX2 , ’2 Þþ ¼ Dm ðX1 , ’1 Þþ Theorem 16 Let (X, R) and (X 0, R0) be two min-

imal equivalence relations on Cantor sets which
and this is in turn equivalent to the existence of a are either AF-relations or arise from actions of
homeomorphism F : X1 ! X2 which induces an finitely generated abelian groups.
injection from M X1 , R’1 into M X2 , R’2 : Then they are orbit equivalent if and only if
Dynamical Systems and C
-Algebras 505

ðDm ðX, RÞ, Dm ðX, RÞþ , ½1X Þ Let (X, ’) be a Cantor minimal system and
C
(X, ’) be the associated C
-crossed product
þ
ffi Dm ðX0 , R0 Þ, Dm ðX0 , R0 Þ , ½1X0 : C(X)⋊’ℤ. Recall (see, e.g., Giordano et al.
(1995), p. 60 for the precise references) that
Then as a corollary of Theorem 8, we have
1. C
(X, ’) is a simple, unital, AT-algebra of real
Corollary 4 For any simple dimension groups rank zero and stable rank one, hence is classi-
G with a trivial infinitesimal subgroup, there fied by its Elliott invariant;
exists a Cantor minimal system (X, ’) such that 2. If Ki(C
(X, ’)) denotes, for i ¼ 0, 1, the
Dm(X, ’) ¼ G. K-groups of the crossed product, then
K0(C
(X, ’)) is a simple dimension group
Remark 9 Let d  2. By Theorem 16 and Cor-
order isomorphic to D(X, ’), and K1(C
(X,
ollary 4, for any free minimal action of
’)) ¼ ℤ (see [?]).
ℤd , Dm ðX, Rℤd Þ is a simple, acyclic dimension
group with no nontrivial infinitesimal elements, Hence, we have
but an exact description of the range of the invari-
ant Dm is not yet known.
Theorem 18 A complete isomorphism invariant
However, using results of M.-I. Cortez and for the family of C
(X, ’), where (X, ’) is a Can-
S. Petite (2014), S.-M. Høynes (2016) showed tor minimal system, is the simple dimension group
that it contains all unital simple dimension groups (D(X, ’), D(X, ’)+) with distinguished order unit
(G, u), with no nontrivial infinitesimal elements, [1X].
and a noncyclic rational subgroup ℚ(G, u).
Remark 10
A Topological Krieger Theorem: The Notion of
(1) Boyle and Handelman (1994) showed that all
Strong Orbit Equivalence
possible (topological) entropies can be real-
In this subsection, (X, ’) will always denote a
ized in the class of Cantor minimal system
Cantor minimal system. The motivation of
(X, ’), such that (D(X, ’), [1]) is order iso-
Giordano et al. (1995) was to obtain a topological
morphic to ℤ 12 , 1 : Therefore,
version of the famous Krieger theorem.

Theorem 17 (Krieger 1976). Let (X1, m1, T1) C
ðX 1 , ’ 1 Þ ffi C
ðX 2 , ’ 2 Þ

and (X2, m2, T2) be two ergodic nonsingular sys- )
= ðX 1 , ’ Þ conjugate to ðX 2 , ’ Þ:
1 2
tems and M1 ¼ W
(X1, m1, T1) and M2 ¼ W
(X2,
m2, T2) their associated von Neumann–Krieger Sugisaki (1998, 2003) generalized Boyle–
factors. Then the following are equivalent: Handelman’s result to the class of Cantor minimal
system (X, ’) such that (D(X, ’), D(X, ’)+, [1X])
1. The dynamical systems are orbit equivalent, is a fixed simple dimension group (G, G+, u).
that is, there exists a bimeasurable bijection
F : X1 ! X2 preserving the measure class, such (2) Let (X, ’) be a Cantor minimal system. By
that, for almost all x  X1 , FðOrbT 1 ðxÞÞ ¼ Proposition 6 and Theorem 13, the unital sim-
OrbT 2 ðFðxÞÞ: ple dimension group (D(X, ’)/Inf(D(X, ’)),
2. The von Neumann–Krieger factors M1 and M2 [1X]) is a complete invariant of orbit
are isomorphic. equivalence.
3. the Krieger associated flow of (X1, m1, T1) and Consequently, the orbit equivalence of two
of (X2, m2, T2) are conjugate or equivalently Cantor minimal systems does not imply in
the flow of weights of M1 and of M2 are general the isomorphism of the corresponding
isomorphic. crossed products.
506 Dynamical Systems and C
-Algebras

Let (X1, ’1) and (X2, ’2) be two orbit equiva- by a Cantor minimal system in a prescribed orbit
lent Cantor minimal systems and F : X1 ! X2 be equivalence class. More precisely, he proves in
an orbit equivalence. Then F defines two maps m, (Ormes 1997, Theorem 7.2):
n : X1 ! ℤ given for x  X, by
Theorem 20 Let (Y, n, T ) be an ergodic proba-
nðxÞ mðxÞ bility measure-preserving dynamical system,
Fð’1 ðxÞÞ ¼ ’2 ðFðxÞÞ and F ’1 ðxÞ
(X, ’) a Cantor minimal system, and m  M(X, ’)
¼ ’2 ðFðxÞÞ: an ergodic measure.
Then there is a Cantor minimal system (X, c),
We will call n and m the orbit cocycles associ- orbit equivalent to (X, ’) such that (Y, n, T ) is
ated to F. measurably conjugate to (X, c, m).

Definition 8 Two Cantor minimal systems Moreover in Ormes (1997, Theorem 27), he
(X1, ’1) and (X2, ’2) are strongly orbit equivalent gives necessary and sufficient spectral conditions
(SOE) if there exists an orbit equivalence for a Jewett–Krieger-type realization of an ergo-
F : X1 ! X2 such that each of the associated dic dynamical system by a Cantor minimal sys-
orbit cocycles m, n : X1 ! ℤ has at most one tem in a prescribed strong orbit equivalence
point of discontinuity. class.

Strong orbit equivalence being an equivalence Continuous Orbit Equivalence

relation will follow from the next theorem. In this section, a topological dynamical system (X,
G) will denote a countable discrete group G acting
Theorem 19 (Giordano et al. (1995), Theorem on a compact, second countable, Hausdorff space
2.1). Let (X1, ’1) and (X2, ’2) be two orbit equiv- X. Recall that (X, G) is topologically free if for
alent Cantor minimal systems. The following every g  G, g 6¼ e, the set {x  X | gx 6¼ x} is
statements are equivalent: dense in X or equivalently if the subset of elements
x  X for which the stabilizer subgroup Gx ¼
(1) (X1, ’1) and (X2, ’2) are strongly orbit {g  G| gx ¼ x} is trivial is dense in X.
equivalent.
(2) DðX1 , ’1 Þ, DðX1 , ’1 Þþ , ½1X1 and (D(X2, ’2), Definition 9 Two topological dynamical systems
D(X2, ’2)+, ½1X2 Þ are order isomorphic. (X1, G1) and (X2, G2) are continuously orbit
(3) C
(X1, ’1) ffi C
(X2, ’2). equivalent if there exist a homeomorphism
F : X1 ! X2 and two continuous maps α1 : X1 
Note that G1 ! G2 and α2 : X2  G2 ! G1 such that

a) The equivalence between the statements (1) F(g1x1) ¼ α1(x1, g1)F(x1), x1  X1, g1  G1
(2) and (3) follows from Theorem 18 (2) F1(g2x2) ¼ α2(x2, g2)F1(x2), x2  X2,
b) If (X, ’1) and (X, ’2) are strongly orbit equiv- g2  G2
alent, then the coboundary subgroups B(X, ’1)
and B(X, ’2) are equal and (2) follows Remark 12 If for i ¼ 1 and 2, (Xi, Gi) is topo-
logically free, then α1 and α2 are uniquely deter-
Remark 11 Recall that the famous Jewett– mined by the conditions (1) and (2) of Definition 9
Krieger theorem (see, e.g., Glasner (2003)) states and are continuous one-cocycles (see, e.g., Li
that any ergodic transformation of a nonatomic (2018a)).
Lebesgue space has a uniquely ergodic minimal
Cantor model. For two topological dynamical systems
Ormes (1997) presents Jewett–Krieger-type (Xi, Gi), let us recall that they are isomorphic if
realization results of an ergodic dynamical system there is a homeomorphism F : X1 ! X2 and a
Dynamical Systems and C
-Algebras 507

group isomorphism γ : G1 ! G2 such that for all continuous orbit equivalent to (X, G) is iso-
g  G1, x  X1, morphic to it. Then he gives several examples
of topologically free dynamical systems which
FðgxÞ ¼ gðgÞFðxÞ: are continuous orbit equivalence rigid.
3) Medynets, Sauer and Thom, in their study of
The notion of continuous orbit equivalence Cantor systems and quasi-isometry of groups
was initially studied by M. Boyle in his thesis (Medynets et al. 2017), prove that two finitely
(Boyle 1983), where he proved: generated non-amenable groups are quasi-
isometric if and only if they admit continuous
Proposition 9 For i ¼ 1, 2, let ’i be a topolog- orbit equivalent Cantor minimal actions. In
ically transitive homeomorphism of a compact particular, free groups of different rank admit
metrizable space Xi. If ’1 and ’2 are continuously continuous orbit equivalent Cantor minimal
orbit equivalent, then ’1 and ’2 are flip conju- actions, unlike in the measurable setting.
gate, that is, ’1 is conjugate to ’2 or to ’1
2 or the
ℤ-dynamical systems (X1, ’1) and (X2, ’2) are Li (2018b), by studying a weaker form of con-
isomorphic. tinuous orbit equivalence, extends Medynets,
Sauer, and Thom’s result and proves in particular
This proposition is the analogue in the ergodic that two finitely generated groups are quasi-
probability measure-preserving case to isometric if and only if there exist Kakutani equiv-
Belinskaya’s result: the integrability of one of alent topologically free systems of these groups
the associated orbit cocycles implies flip on compact spaces.
conjugacy (in the measurable category).
Boyle and Tomiyama (1998) generalized this Continuous Orbit Equivalence and C
-algebra
result as follows: Let (X, G) be a topological dynamical system and
X ⋊ G be the associated topological transforma-
Theorem 21 For i ¼ 1, 2, let ’i be a topologically tion groupoid. Then it is an étale groupoid and is a
free homeomorphism of a compact Hausdorff principal groupoid (i.e., X ⋊ G ¼ RG) if the action
space Xi. If ’1 and ’2 are continuously orbit equiv- is free, and X ⋊ G is essentially principal if the
alent, then there exist open partitions X1 ¼ A1 A2 action is topologically free. Recall that the
and X2 ¼ B1 B2 such that ’1 jA1 is conjugate to reduced C
-algebra C
r ðX⋊GÞ of the transforma-
’2 jB1 and ’1 jA2 is conjugate to ’12 B2 : tion groupoid X ⋊ G is canonically isomorphic to
the C
-crossed product C0(X) ⋊ G (see, e.g., Li
Remark 13 (2018a) or Matui (2017)).
1) In measurable dynamics, orbit equivalence
rigidity is an important domain of research Remark 14 If (X, G) is topologically free, then
and several impressive results have been pro- the pair (C0(X)⋊G, C0(X)) is a Cartan pair as
ved (see, e.g., Furman (1999); Ioana (2011); defined by J. Renault (2008).
Monod and Shalom (2006); Popa (2007)). In
topological dynamics, M. Boyle’s result was Generalizing a classical result of I. Singer
the first proved positive rigidity result. Li (1955) on the interplay between measured dynam-
(2018a) obtains both positive and negative ical systems and the Murray–von Neumann group
rigidity results. measure space construction, X. Li (2018a) proves:
2) In Li (2018a), the author defines a topologi-
cally free dynamical system (X, G) to be con- Theorem 22 Let (Xi, Gi) be two topologically
tinuous orbit equivalence rigid if any free dynamical systems as above. Then the follow-
topologically free dynamical system ing are equivalent:
508 Dynamical Systems and C
-Algebras

(1) (X1, G1) and (X2, G2) are continuously orbit If G1 ¼ G2 ¼ G, and if there exists a homeo-
equivalent. morphism F : X1 ! X2 such that F ∘ ’1(g) ¼
(2) The transformation groupoids X1 ⋊ G1 and ’2(g) ∘ F for all g  G, we will say that (X1, ’1)
X2 ⋊ G2 are isomorphic. and (X2, ’2) are conjugate.
(3) There is a C
-isomorphism F from C0(X1) ⋊ For two topological dynamical systems, the
G1 onto C0(X2) ⋊ G2 such that F(C0(X1)) ¼ following implications are clear: conjugacy )
C0(X2). isomorphism ) continuous orbit equivalence )
orbit equivalence.
Remark 15 In this section, we characterize (algebraically)
1) The equivalence of (2) and (3) is due to these four types of equivalences for ℤ- and ℤ2-
J. Renault (2008). odometers. Let us first recall the definition of ℤd-
2) J. Tomyiama proved the equivalence of (1) and odometers.
(3) for topologically free ℤ-actions in Let G be a (nontrivial) decreasing sequence
Tomiyama (1996). (Gn)n1 of finite index subgroups of ℤd. For
3) Theorem 22 is extended to more general n  1, let qn : ℤd/Gnþ1 ! ℤd/Gn be the quotient
groupoids in a recent publication of Carlsen, map and let XG be the corresponding inverse limit.
Ruiz, Sims, and Tomforde (2021). Let fG be the profinite action of ℤd on the XG
given by
For ℤ-actions, using Theorem 23, M. Boyle
and J. Tomiyama (1998) proved the following fG ðvÞðx1 , x2 , . . .Þ ¼ ðx1 þ v, x2 þ v, . . .Þ v  ℤd :
rigidity result:
Then following Cortez (2006), a ℤd-odometer
Theorem 23 For i ¼ 1, 2, let ’i be a topologically is any ℤd-Cantor minimal system conjugate to
free homeomorphism of a compact Hausdorff XG , fG for some G as above. Note that
space Xi. Then the following are equivalent:

(1) There is a C
-isomorphism F : C(X1) (1) XG , fG is free if and only if \1

n¼1 Gn ¼ f0g
⋊’1 ℤ ! CðX2 Þ⋊’2 ℤ such that F(C(X1)) ¼ (2) The free ℤd-Cantor minimal system
C(X2).
XG , fG is equicontinuous
(2) There exist open partitions X1 ¼ A1 A2 and
X2 ¼ B1 B2 such that ’1 jA1 is conjugate to
Remark 17 G-odometers can be defined for any
’2 jB1 and ’1 jA2 is conjugate to ’1
2 B2 :
residually finite group G (see Cortez and Petite
(2014) or Downarowicz (2005)). Moreover in
If ’1 and ’2 are topologically transitive or the Cortez and Medynets (2016), the authors show
spaces are connected, then these conditions hold that if a dynamical system (X, H ) is a free minimal
if and only if ’1 and ’2 are flip conjugate. equicontinuous system, then H is necessarily
residually finite and (X, H ) is conjugate to a H-
Remark 16 Theorem 21 was proved for Cantor odometer.
minimal systems in Giordano et al. (1995) using
Proposition 9. Following Giordano et al. (2019), let us recall
another presentation of ℤd-odometers using
ℤd-odometers Pontryagin duality. Let H be a subgroup of ℚd
We have already introduced the notions of iso- containing ℤd. Then we have
morphism, continuous orbit equivalence, and
(topological) orbit equivalence between two topo-
H=ℤd  ℚd =ℤd  ℝd =ℤd ffi d :
logical dynamical systems (X1, G1) and (X2, G2).
Before adding a strengthening of isomorphism, let
Let r denote the canonical inclusion of H/ℤd in
us denote by ’i the action of Gi on Xi, for i ¼ 1, 2.
 : By duality, let r be the homomorphism from
d
Dynamical Systems and C
-Algebras 509

ℤd to YH, the Pontryagin dual H=ℤd of H/ℤd, (1) If d, d 0  2, then (YH, cH) and ðY H0 , cH0 Þ are
which is a compact, totally disconnected space. conjugate if and only if d ¼ d 0 and H ¼ H 0.
Then let (YH, cH) be the ℤd-dynamical system (2) If d, d 0  2, then (YH, cH) and ðY H0 , cH0 Þ are
where cH denotes the action of ℤd on YH given by isomorphic if and only if d ¼ d 0 and there
exists α  GLd(ℚ) with det(α) ¼  1 such
cnH ðxÞ ¼ x þ rðnÞ, for n  ℤd , x  Y H : that αH ¼ H 0.
(3) If d, d 0  2, then (YH, cH) and ðY H0 , cH0 Þ are
Note that continuously orbit equivalent if and only if
d ¼ d 0 and there exists α  GLd(ℤ) such
(1) If ℤd  H  ℚd, then (YH, cH) is free if and that αH ¼ H 0.
0
only if H is dense in ℚd (4) The ℤd-action (YH, cH) and the ℤd -action
(2) If ℤd  H1  H2  ℚd, then there is a natural ðY H0 , cH0 Þ are orbit equivalent if and only if
factor map from Y H2 , cH2 to Y H1 , cH1 the superindex [[H : ℤd]] of ℤd in H is equal to
0
(3) If ℤd  H1  H2  ℚd, then the inverse limit superindex H 0 : ℤd :
of the ℤ -dynamical systems Y Hn , cHn n1 is
d

conjugate to (YH, cH), where H ¼ [n1Hn Remark 18

1) As any two orbit equivalent ℤ-odometers are
To describe the link between the two construc- conjugate (see, e.g., Giordano et al. (2019),
tions, recall that a finite-index subgroup G of ℤd is Cor. 5.9), the four conditions of Theorem 24
a lattice of ℝd, and that its dual lattice G
is a are equivalent for d ¼ d 0 ¼ 1.
finite-index extension of ℤd. The situation is remarkably different for ℤ2-
Then (see Giordano et al. (2019), Theorems 27 odometers as the four conditions are all
and 28) one has that inequivalent (see Cortez (2006) and Li
(2018a) or Giordano et al. (2019), Example
a) The ℤd-odometer XG , fG , given by a 5.10).
2) Cortez and Medynets (2016) study the notion
(nontrivial) decreasing sequence G ¼ ðGn Þn1
of continuous orbit equivalence for
of finite index subgroups of ℤd is conjugate to
G-odometers, where G is a residually finite
(YH, cH), where H ¼ [n1 G
n , and conversely
group G. They study in particular the topolog-
b) If (Hn)n1 is an increasing sequence of finite
ical full group associated to a free G-odometer.
index extensions of ℤd and H ¼ [n  1Hn  ℚd,
3) We reviewed in Remark 8 the research in topo-
then G ¼ H
n n1 is a decreasing sequence logical speedups of Cantor minimal systems.
G ¼ ðGn Þn1 of finite index subgroups of ℤd, Johnson and McClendon (2022) study these
and (YH, cH) and XG , fG are conjugate topological notions in the context of actions
of ℤd, and in particular for ℤd-odometers.
Before stating the main result of Giordano et al.
(2019), let us recall (Giordano et al. (2019), Def. Cohomology of ℤd-odometers
1.4) that if H is a subgroup ℤd  H  ℚd, the To any dense subgroup ℤd  H  ℚd, we have
superindex of ℤd in H is defined as associated a ℤd-action (YH, cH) on the Cantor
set YH. In this subsection, we recall the defini-
H : ℤd ¼ H 0 : ℤd jℤd  H 0  H, tion of the integer-valued cohomology of an
action of ℤd on the Cantor set and show that
H 0 : ℤd < 1 : H1(YH, cH) ffi H.
Let (X, ’) denote a ℤd-action on the Cantor set
X. Recall that C(X, ℤ), the abelian group of con-
Theorem 24 (Giordano et al. (2019), Theorem
0 0 tinuous ℤ-valued functions on X, is also a ℤd-
1.5). Let ℤd  H  ℚd and ℤd  H 0  ℚd be module (for n  ℤd and f  C(X, ℤ), n f(x) ¼
0
two dense subgroups of ℚd and, respectively, ℚd : f(’(n)x), x  X).
510 Dynamical Systems and C
-Algebras

The integer-valued cohomology of (X, ’) can Remark 20 The homology groups H
ðG Þ of an

be defined as the group cohomology of ℤd with étale groupoid were first introduced in Crainic and
coefficients in the module C(X, ℤ), or as the Čech Moerdijk (2000). When the unit space of the
cohomology of the mapping torus associated to groupoid G is totally disconnected, H. Matui
(X, ’) (see, e.g., Forrest and Hunton (1999), (2012, 2015, 2016) explored the connections
Hunton (2015)). between the homology and the dynamical proper-
ties of G. He studied in particular the interplay
Remark 19 If Γ is a discrete countable group and between the homology groups, the topological
(X, ’) is a Γ-action on the Cantor set, let G ’ full group, and also the K-theory of C
r ðG Þ, the
denote the étale transformation groupoid. Recall reduced groupoid C
-algebra of G (see, e.g.,
that G ’ ¼ Gd  X and the groupoid structure is Matui’s review paper (2017)).
defined as follows: (γ, x) and (γ0, y) are
composable if and only if x ¼ ’(γ0)( y), and As above, let (X, ’) be a free minimal ℤd-
Cantor system. Let us first consider the first
group of cohomology of (X, ’). Recall that a
ðg, ’ðg0 ÞðyÞÞðg0 , yÞ ¼ ðg g0 , yÞ and ðg, xÞ1
continuous function θ : X  ℤd ! ℤ is a
¼ g1 , ’ðgÞðxÞ : 1-cocycle if and only if

Then, the cohomology groups H
(Γ, C(X, ℤ)) yðx, m þ nÞ ¼ yðx, mÞ þ yð’ðmÞðxÞ, nÞ for all
are canonically isomorphic to the cohomology
x  X and m, n  ℤd :
groups H
G ’ associated to the étale groupoid
G ’ : The cohomology H
ðG Þ for any étale A cocycle θ is a coboundary if there is
groupoid G were introduced by J. Renault h  C(X, ℤ) such that θ(x, m) ¼ h(’(m)(x)) 
(1980) and has been widely studied. h(x) for all x  X and m  ℤd.
Let (YH, cH) be the ℤd-odometer associated to
a dense subgroup ℤd  H  ℚd. Then using
Let (X, ’) be a free minimal ℤd-Cantor system.
Giordano et al. (2019), Theorem 2.5, and a simple
Then the following holds:
case of Shapiro’s Lemma, we have

Lemma 1 Theorem 25 (Giordano et al. (2019), Theorem

4.3). Let ℤd  H  ℚd. Then H1(YH, cH) ffi H.
(1) H d(X, ’) is the group of co-invariants, that is,
For 1  d  2, a precise description can be
H d ðX, ’Þ ¼ DðX, ’Þ:
given for such an isomorphism. Note first that if m
is an invariant probability measure of a ℤd-Cantor
(2) H 0(X, ’) ¼ { f  C(X, ℤ)| f ∘ ’(n) ¼ f, for
system (X, ’) and θ is a 1-cocycle of (X, ’), then
all n  ℤd} ¼ ℤ
the map t1m ðyÞ : ℤd ! ℝ defined for n  ℤd by
(3) By Poincaré duality, there is a natural iso-
morphism between Hn(X, ’) and the homol-
ogy group Hdn(X, ’) ≔ Hdn(ℤ; C(X, ℤ)). t1m ðyÞðnÞ ¼ yðx, nÞdmðxÞ,
X

and as continuous orbit equivalence of (X, ’)

is a group homomorphism. As t1m vanishes on
corresponds to isomorphism of the transformation
groupoid, we get coboundaries, it defines a group homomorphism
t1m : H 1 ðX, ’Þ ! Hom ℤd , ℝ :
Using the canonical basis (en)1nd of ℤd to
Proposition 10 The cohomology H
G ’ is an identify Hom(ℤd, ℝ) with ℝd, we define t1m :
invariant of continuous orbit equivalence. H 1 ðX, ’Þ ! ℝ for any 1-cocycle θ by
Dynamical Systems and C
-Algebras 511

t1m ð½y Þ ¼ t1m ðyÞðe1 Þ, , t1m ðyÞðed Þ : We use the following notation and terminolo-
gies throughout this chapter. Let G be a discrete
group and X be a compact Hausdorff space.
We then get A G-action is a group homomorphism α from
G into the homeomorphism group of X. In this
Theorem 26 (Giordano et al. (2019), Theorem case, we also write α : G ↷ X for the action and say
4.4). Let 1  d  2. (YH, cH) be the ℤd-odometer that X is a G-space.
associated to a dense subgroup ℤd  H  ℚd, and Given a G-action α : G ↷ X, there is an induced
m be its unique invariant probability measure. action of G on the C
-algebra C(X) of continuous
Then functions on X given by

t1m : H1 ðX, ’Þ ! H s:f ðxÞ≔f s1 x

is an isomorphism. for s  G and f  C(X), x  X. We write C(X)

⋊rG for the reduced crossed product C
-algebras.
If we need to specify the action, then we write
Mean Dimension, Small Boundary C(X)⋊αG and C(X)⋊α,rG.
Property, and Classification of C
-
Algebras Example 6 Let G ↷ X be a free minimal action of
a countable discrete amenable group G on a com-
The theory of mean dimension for topological pact Hausdorff space X. The reduced crossed
dynamical systems was suggested by Gromov product C(X)⋊rG can be realized in the following
and largely developed by Lindenstrauss and way: let m be a G-invariant probability measure on
Weiss (2000). In many ways, it parallels the the- X (which exists because G is amenable) and H be
ory of covering dimension for topological spaces. the Hilbert space L2(X, m)  ‘2(G). We define a
The small boundary property, on the other hand, representation π : C(X) ! B(H ) by
can be viewed as a dynamical version of total
disconnectedness (see Definition 20). These two pð f Þð’  ds Þ≔ s1 :f ’  ds
closely related notions have profound implica-
tions to several long-standing open questions in for f  C(X), ’  L2(X, m) and s  G. Let l :
topological dynamics (see, e.g., Gutman et al. G ! U ‘2 ðGÞ denote the left regular represen-
(2016); Lindenstrauss (2000); Lindenstrauss and tation, that is, lg(δs) ¼ δgs. The reduced crossed
Weiss (2000)). product C(X)⋊rG is the C
-algebra generated by
In the last decade, these purely dynamical the images π(C(X)) and 1  l(G) in B(H ).
notions have found deep connections with the We refer the readers to, for example, (Brown
classification of simple nuclear C
-algebras, with and Ozawa (2008), Chap. 4) for the general con-
the central theme being the study of “regularity” structions and basic properties of these algebras.
for crossed products. As we will see, ideas and
techniques that originate from the Kakutani– Mean Dimension and Radius of Comparison
Rokhlin tower decomposition, or more generally, We start by recalling the definition of mean
the Ornstein–Weiss tiling theorem in ergodic the- dimension for a single homeomorphism from
ory (Ornstein and Weiss 1987), are fundamental in Lindenstrauss and Weiss (2000). Let X be a com-
many of these connections. Interestingly, the pact metrizable space and T : X ! X be a homeo-
attempt to understand the structure of crossed morphism. For two open covers α and β of X, we
products also led to the discovery of some purely write β  α (or α ≺ β) if β is an open refinement of
dynamical results, such as a new characterization α, and let α _ β denote the join of α and β. Define
of the small boundary property. the order of an open cover α by
512 Dynamical Systems and C
-Algebras

ℤ
ordðaÞ≔ max 1U ð x Þ  1 , mdim ½0, 1 d ,s ¼ dim ½0, 1 d ¼ d:
xX
Ua

It was a long-standing open problem whether

where 1U is the characteristic function of U, and
every minimal system can be embedded into the
set
two-sided shift ([0, 1]ℤ, s). It is not hard to see that
the mean dimension cannot increase when passing
D ðaÞ≔ minfordðbÞ : b  ag:
to closed invariant subsets, so for a dynamical
system (X, T) to be embeddable into the shift
Definition 10 Let X be a compact metrizable
(([0, 1]d)ℤ, s), it is necessary to have mdim(X, T)
space and T be a homeomorphism T of X. Its
 d. Lindenstrauss and Weiss (2000) constructed a
mean dimension mdim (X, T ) is given by
minimal subshift of (([0, 1]2)ℤ, s) with mean
D ða_T 1 a_ _T ðn1Þ aÞ
supa lim n!1 n , dimension strictly larger than one, thereby resolv-
ing the question in the negative.
where the supremum is taken over all open
covers α of X. Theorem 27 (Lindenstrauss and Weiss (2000),
Proposition 3.5). There is a minimal invariant
This definition should be compared to the closed subset X of ([0, 1]2)ℤ, s) such that
Lebesgue covering dimension (see, e.g., Pears mdim(X, s|X) > 1.
(1975), Chap. 3), which is given by
Using a similar technique Giol and Kerr (2010)
dimðXÞ≔ sup DðaÞ: constructed a minimal subshift (X, T ) whose
a
crossed product C
-algebras C(X)⋊rℤ have non-
Although we only gave the definition in the zero radius of comparison (see Definition 12) and
case of a single homeomorphism, the definition thereby gave the first example of a simple crossed
can be generalized to actions of countable amena- product of the form C(X)⋊rℤ that does not fall
ble groups (e.g., see Coornaert (2015)) and to within Elliott’s classification program (see section
actions of the even larger class of countable sofic “Classification of Simple Nuclear C
-Algebras”
groups (Li 2013). for more details). In some sense, both the mean
dimension and the radius of comparison are
Example 7 If X has finite Lebesgue covering invariants that are designed for highly compli-
dimension, then the mean dimension of any cated structure.
homeomorphism T is necessarily zero. Moreover, The relationship between these two notions
by Lindenstrauss and Weiss (2000, Theorem 5.4) seems to go far beyond the apparent analogy.
every system (X, T ) with finite topological The main objective of this section is to introduce
entropy has mean dimension zero. a conjecture that relates these two invariants in a
definitive, quantitative way (Conjecture 1).
Example 8 Let K be a compact metric space of In what follows, we briefly recall the definition
finite covering dimension and consider the shift of the radius of comparison and fix our notations.
homeomorphism s : Kℤ ! Kℤ. By Lindenstrauss For a C
-algebra A, let A+ ¼ {a
a : a  A} be the
and Weiss (2000, Proposition 3.1) we have set of positive elements in A.
The following notion of subequivalence was
mdim K ℤ , s  dimðK Þ: introduced by Cuntz (1978), which is reminiscent
of the Murray–von Neumann subequivalence on
Moreover, if K ¼ [0, 1]d for some positive integer projections in a von Neumann algebra. It can be
d, then the equality holds (see Lindenstrauss and shown that  is an equivalence relation on A+
Weiss (2000), Proposition 3.3), that is, (see, e.g., Giordano et al. (2018, Proposition
Dynamical Systems and C
-Algebras 513

2.0.1)). We refer the reader to Giordano et al. tracial state tm on the reduced crossed product
(2018, Part I) and Ara et al. (2011) for more on C(X)⋊rG (see Giordano et al. (2018), Example
the Cuntz subequivalence relation. 10.1.31). In fact, if the action is free, then every
tracial state on C(X)⋊rG has this form (see, e.g.,
Definition 11 For two positive elements a, b in Giordano et al. (2018), Proposition 11.1.21).
A+, we say a is Cuntz subequivalent to b, denoted
a ≾ b, if there exists a sequence (rn)n in A such that Given a tracial state t  T(A), define the rank
lim n!1 r n br
n  a ¼ 0: If a ≾ b and b ≾ a, function dt : A+ ! [0, 1] by
then we write a  b.
1
d t ðaÞ≔ lim t an
n!1
Example 9 Let A be the algebra Mn ðÞ of n  n
matrices. For two positive elements a and b in A,
(see Blackadar and Handelman (1982) for more
we have a ≾ b if and only if rank(a)  rank (b).
details).
Example 10 Let X be a compact Hausdorff space
Example 13 If A ¼ C(X) and t is the tracial state
and let f, g  C(X)+. Then f ≾ g if and only if
induced by a Borel probability measure m on X,
supp( f ) supp (g), where supp( f) ≔ f 1(0, +1)
then dt( f ) ¼ m( f 1(0, 1)) for any f in C(X)+.
(see, e.g., Ara et al. (2011), p. 6, Proposition 2.5).
One can show that if a ≾ b, then dt(a)  dt(b)
Example 11 Let G ↷ X be an action of a count-
for every t  T(A) (see Blackadar and
able discrete group G on a Cantor space X, and let
Handelman (1982), Theorem II.2.2). The con-
A and B be two clopen sets. Suppose there is a
verse is not necessarily true, and the radius of
clopen partition A ¼ C1 t C2 t Cn and group
comparison, as introduced by Toms (2006), is
elements s1, s2, . . ., sn such that the sets s1C1, s2C2,
designed to measure the failure of this “compari-
. . ., snCn are pairwise disjoint subsets of B, then
son property.”
the characteristic function wA is Cuntz sub-
In the next definition, we use the notation
equivalent to the characteristic function wB in the
M1 ðAÞ≔[1 n¼1 Mn ðAÞ, where each Mn(A) is
crossed product C(X)⋊rG.
embedded into Mnþ1(A) as the upper-left corner.
To avoid technicalities, from now on we only
consider the class of unital stably finite exact C
- Definition 12 Let A be a unital stably finite exact
algebras. It includes all the crossed products that C
-algebra. The radius of comparison rc(A) of A is
we are interested in (see, e.g., Blackadar (2006) the infimun of the strictly positive real numbers r
for relevant C
-algebraic definitions). such that for any a, b  M1(A)þ, if dt(a) þ r <
Recall that a tracial state on a C
-algebra A is a dt(b), for all t  T(A), then a≾b.
state (i.e., a positive linear functional with norm
one) ’ that satisfies ’(ab) ¼ ’(ba) for all a and The case rc(A) ¼ 0, also known as strict com-
b in A. Let T(A) denote the set of tracial states on parison, is especially important for the classifica-
A. By combining a result of Blackadar and tion of simple nuclear C
- algebras (see section
Handelman (1982) and a result of Haagerup “Classification of Simple Nuclear C
-algebras”).
(2014), we deduce that the set T(A) is nonempty
for any C
-algebra in the class we consider. Example 14 Let X be a finite CW complex. Then
by Toms (2006) and Elliott and Niu (2013),
Example 12 Let G ↷ X be an action of a count- rc(C(X)) is essentially half of the covering dimen-
able discrete amenable group G on a compact sion of X. More precisely, if dim X denotes the
metrizable space X, and let m be a G-invariant covering dimension of X, we have the following
Borel probability measure. Then m induces a by Elliott and Niu (2013, Corollary 1.2):
514 Dynamical Systems and C
-Algebras

• If dim X is odd, then In a recent breakthrough (Niu 2022), Niu

established one inequality in full generality for
minimal homeomorphisms.
dim X  1
rcðCðXÞÞ ¼ max 0, 1 :
2 Theorem 28 (Niu 2022, Corollary 7.10). Let α
be a minimal homeomorphism on a separable
• If dim X is even, then compact Hausdorff space X. Then

1
dim X dim X rcðCðXÞ⋊r,a ℤÞ  mdimðX, aÞ:
 2  rcðCðXÞÞ   1: 2
2 2

Very roughly, the proof is divided in two steps.

The following conjecture, due to Phillips and The first one is to show that for an action α of a
Toms (see, e.g., the introduction of Niu (2022)), countable discrete amenable group G, the conclu-
predicts a precise quantitative relationship sion of Theorem 28 holds if the following two
between the mean dimension of a topological conditions are met:
dynamical system and the radius of comparison
of the associated crossed product C
-algebra. (1) α : G ↷ X has the uniform Rokhlin property.
(2) C(X)⋊r,αG has Cuntz comparison for open
Conjecture 1 For a free minimal action α : G ↷ X sets.
of a countable discrete amenable group G on a
compact metrizable space X, Then, as the second step, it is proved that these
two conditions are satisfied when G ¼ ℤ. In the
1 preprint (Niu 2019), Niu proved that these two
rcðCðXÞ⋊r GÞ ¼ mdimðX, a, GÞ:
2 conditions also hold for free minimal ℤd-actions.
In what follows, we give the precise definitions
Example 15 Let (X, T ) be a Giol–Kerr minimal of these two conditions. The uniform Rokhlin prop-
subshift constructed in Giol and Kerr (2010). It erty, as the name suggests, is reminiscent to the
was proved in Hines (2015) that classical Rokhlin lemma in ergodic theory. Recall
that in the Rokhlin lemma for a single transforma-
1 tion T, a collection U, T(U), . . ., T n(U) of disjoint
rcðCðXÞ⋊r ℤÞ  mdimðX, T Þ:
2 subsets is often called a tower. Essentially the uni-
form Rokhlin property says that the space can be
Example 16 Elliott and Niu (2017) proved that if
almost covered by open disjoint Rokhlin towers, and
mdim(X, T ) ¼ 0, then rc(C(X)⋊rℤ) ¼ 0. In fact
the remainder is uniformly small for all G-invariant
they showed that the small boundary property,
measures. In the section entitled “Almost Finiteness,
which is equivalent to mean dimension zero for
Comparison, and the Small Boundary Property”, we
ℤ-actions, implies Z-stability of the crossed prod-
will remark that the uniform Rokhlin property is very
uct, which in turn implies strict comparison. See
similar to the notion of almost finiteness in measure.
section “Classification of Simple Nuclear
We now recall the definition of castles given in Kerr
C
-algebras” for more details.
(2020, Definition 5.7).
Example 17 In (Phillips 2016), Phillips showed Definition 13 Let G be a countable discrete
that for a general minimal homeomorphism one group and let X be a compact metrizable
always has G-space. A collection {(Vi, Si)}i  I is a castle if
each Vi is a subset of X, each Si is a finite subset
rcðCðXÞ⋊r ℤÞ  1 þ 36 mdimðX, T Þ: of G, and the collection
Dynamical Systems and C
-Algebras 515

fsV i : s  Si , i  I g ð1Þ mðEÞ < lmðFÞ

is pairwise disjoint. for all G-invariant Borel probability measure m,

In this case, the finite sets Si are called the we have
shapes and the subsets sVi are called the levels.
The following definition, due to Z. Niu in Niu ’F
(2022), was originally given in terms of orbit ’E ≾ ⋱
capacity. The statement below is equivalent
’F mm
thanks to Proposition 3.3 of Kerr and Szabó
(2020).
in M1(C(X)⋊rG).
Definition 14 (Niu 2022, Definition 3.1). Let
Theorem 29 (Niu 2022, Theorem 4.7). Let
G be a discrete amenable group and let X be a
α : G ↷ X be a free minimal action of a discrete
separable compact Hausdorff G-space. We say
amenable group on a separable compact
the action G ↷ X has the uniform Rokhlin prop-
Hausdorff space that has (URP) and (l, m)
erty (URP) if for every finite subset K G, and
Cuntz comparison for open sets for some
every δ > 0, there exist:
l  (0, 1] and m  . Then
(1) A castle {(Vi, Si)}i  I such that I is finite, each
1
level is open, and each shape is (K, δ)-invari- rcðCðXÞ⋊r GÞ  mdimðX, a, GÞ:
2
ant (i.e., |gSiΔSi|/|Si| < δ for all g  K and all
i  I)
Classification of Simple Nuclear C
-algebras
(2) supm(X \ i  ISiVi) < δ, where the supremum
This subsection contains a very brief introduction
is taken over all G-invariant Borel probability
to the classification of simple nuclear C
-algebras,
measures m.
or the so-called Elliott’s program. Since the main
purpose is to give more context to the preceding
Example 18 Every free minimal homeomor-
section on “Mean Dimension and Radius of Com-
phism on a Cantor set has the uniform Rokhlin
parison” and the next section on “Almost finite-
property (Niu 2022, Lemma 3.5). As mentioned
ness, Comparison, and the Small Boundary
in Niu (2022), this essentially follows from
Property”, the discussion will be rather informal.
Lindenstrauss’ Rokhlin-type lemma for homeo-
The aim of Elliott’s program is to classify the
morphisms (Lindenstrauss (2000), Corollary
class of (unital) separable simple nuclear C
-alge-
3.4).
bras, which includes any crossed product C(X)
⋊rG arising from a free minimal action of a count-
Meanwhile, Cuntz comparison for open sets
able discrete amenable group on a compact
basically means that C(X)⋊rG has strict comparison
metrizable space. The algebraic invariant is called
for elements in the subalgebra C(X). We give the
the Elliott invariant, which, roughly speaking, con-
precise definition here. For each open subset E of X,
sists of K-theory and traces. However, due to the
choose a continuous function ’ : X ! [0, +1) such
presence of examples such as Villadsen (1998) and
that ’1(0, +1) ¼ E. Note that by Example 10 the
Rørdam (2003), and the aforementioned crossed
Cuntz equivalence class of ’ depends only on the
products constructed by Giol and Kerr, we have
open set E.
to content ourselves with classifying “regular” C
-
algebras. One way to capture this regularity is
Definition 15 (Niu 2022, Definition 4.1). Let
through a noncommutative covering dimension,
G be a discrete group and let X be a compact
known as the nuclear dimension, introduced by
Hausdorff G-space. For l  (0, 1] and m   ,
Winter and Zacharias (2010). Then the regularity
we say the action has (l, m)Cuntz comparison for
appears in the finiteness of the nuclear dimension.
open sets if for any open subsets E, F of X with
516 Dynamical Systems and C
-Algebras

The following classification theorem is due to It is natural to ask how these properties are
many hands, including Gong et al. (2020a and related to each other. The following conjecture is
2020b); Kirchberg (1995); Phillips (2000); and due to Toms and Winter.
Tikuisis et al. (2017) (which in turn depend on a
large body of work). Conjecture 1 For an infinite-dimensional unital
separable simple nuclear C
-algebra, the follow-
Theorem 30 Unital separable simple nuclear ing three conditions are equivalent:
infinite-dimensional C
-algebras that have finite
nuclear dimension and satisfy the UCT are clas- (i) Finite nuclear dimension
sified by the Elliott invariant. (UCT stands for the (ii) Z -stability
Universal Coefficient Theorem of Rosenberg and (iii) Strict comparison.
Schochet (1987)).
The implication (i) ) (ii) is proved by Winter
In Tu (1999), the author proves that if G is (2012) and (ii) ) (iii) is due to Rørdam (2004).
an amenable group, then every crossed product For the upward implications, many results were
C(X)⋊rG satisfies the UCT. It is an open question obtained based on the groundbreaking work of
whether every nuclear C
-algebra satisfies the Matui and Sato (2012, 2014). In another recent
UCT. breakthrough (Castillejos et al. 2021), the impor-
tant notion of uniform property Γ was identified
Example 19 Let θ be an irrational number in and the equivalence between (i) and (ii) is now
0, 12 and ay :  !  be the rotation by θ on the fully established. Combining all the known impli-
cations we arrive at the following theorem:
circle. The crossed product CðÞ⋊ay ,r ℤ is known
as the irrational rotation algebra and often
Theorem 31 Let A be a unital separable simple
denoted by A y : It can be shown that A y has nuclear
nuclear infinite-dimensional C
-algebra. The fol-
dimension one (see Winter and Zacharias (2010),
lowing are equivalent:
Example 6.1) and belongs to the class of algebras
in Theorem 30. Thanks to the work of Pimsner (i) A has finite nuclear dimension.
and Voiculescu (1980) and Rieffel (1981), A y and (ii) A is Z -stable:
A y0 are isomorphic if and only if θ ¼ θ0. (iii) A has strict comparison and uniform prop-
erty Γ.
Examples and results show that regularity also
assumes other forms. One of them is the property of In particular, the Toms–Winter conjecture
strict comparison (i.e., zero radius of comparison) holds for any unital separable simple nuclear
which already appeared in the section “Mean C
-algebra with uniform property Γ.
Dimension and Radius of Comparison” (see Defi-
nition 12). Another one is known as Z -stability or At the time of writing, it is an open question
Jiang-Su stability, which says that the C
-algebra whether every unital separable nuclear C
-algebra
must remain unchanged (up to isomorphism) after has uniform property Γ.
tensoring with the specific C
-algebra Z, called the
Jiang-Su algebra (Jiang and Su 1999). One of the Example 20 Let X be an infinite compact
important features of Z is that although it is infinite metrizable space, and let T : X ! X be a minimal
dimensional as a linear space, it has the same Elliott homeomorphism. Toms and Winter proved that if
invariant as the algebra of complex numbers . X has finite covering dimension, then the reduced
Therefore, in some sense the Elliott invariant does crossed product C(X)⋊Tℤ has finite nuclear
not “see” the algebra Z, whence cannot distinguish dimension (and hence is Z -stable) (Toms and
a C
-algebra A from the tensor product A  Z. Winter 2013). Later it was shown by Elliott and
Dynamical Systems and C
-Algebras 517

Niu that C(X)⋊Tℤ is Z -stable whenever (X, T ) Basically, comparison is a property that asserts
has mean dimension zero (Elliott and Niu 2017). the converse.

Almost Finiteness, Comparison, and the Small Definition 17 Let G be a discrete group and X be
Boundary Property a G-space. We say the action G ↷ X has
A central problem at the interface of C
-algebras (dynamical) comparison if m(A) < m(B) for all
and topological dynamics is to understand when a G-invariant measures m implies A ≺ B for all
crossed product C(X)⋊rG is regular, meaning that open sets A and B.
it satisfies one of the properties appearing in the
Toms–Winter conjecture (Conjecture 1). Several Example 21 By Glasner and Weiss (1995,
dimension theories for topological dynamics, Lemma 2.5), every Cantor minimal system has
including Rokhlin dimension (e.g., Hirshberg comparison.
et al. (2015); Szabó (2015); Szabó et al. (2019)), It’s worth noting that this result essentially
dynamic asymptotic dimension (Guentner et al. follows from the existence of Kakutani–Rokhlin
2017), and tower dimension (Kerr 2020), have tower decomposition for Cantor minimal systems
been introduced and they have direct applications (see the proof of Putnam (1989), Lemma 3.1),
in establishing the finiteness of nuclear dimension. which gives us a hint to a result below that says
Note that the dynamic asymptotic dimension was almost finiteness implies comparison, since
originally motivated by the Baum-Connes conjec- almost finiteness (see Definition 19 below) can
ture. In addition to applications to C
-algebras, be viewed as a generalization of the Kakutani–
they are also interesting dynamical properties in Rokhlin decomposition.
their own rights. We apologize for not being able
Example 22 More generally, let G be a countable
to discuss these important developments due to
discrete group of subexponential growth. A result
length, but to the interested readers we recommend
of Downarowicz and Zhang shows that every free
the notes by Sims et al. (2020, Part III). In this
action of G on any zero-dimensional compact
subsection, we focus on the dynamical analogue of
metrizable space has comparison.
Z-stability and strict comparison and discuss their
relationship with the small boundary property and Remark 21 To the authors’ best knowledge, it is
the Toms–Winter conjecture. an open question whether every free action of a
We start with a dynamical version of strict countable discrete amenable group on any com-
comparison. The basic idea of comparing sets by pact metrizable space has comparison.
their measures appeared in the work of Glasner We now turn to almost finiteness, which can be
and Weiss (1995) (and the type of subequivalence viewed as a topological version of the Ornstein–
relation used there dates back to Hopf (1932)). Weiss tiling theorem and, at the same time, a
The following definitions first appeared in talks dynamical analogue of Z -stability:
given by Wilhelm Winter.
With the notion of castles introduced in Defi-
Definition 16 Let G be a discrete group and X be nition 13, we can now define almost finiteness.
a G-space. For two open subsets A and B of X, we
say A is subequivalent to B, written as A ≺ B, if for Definition 18 Let G ↷ X be a free action of a
every closed subset C of A there are group ele- countable discrete group G on a compact metric
ments s1, s2, . . ., sn in G and open sets U1, U2, . . ., space X. The action G ↷ X is almost finite if for
Un such that C [ni¼1 U i , the sets siUi are every finite subset K G, and every δ > 0, there
pairwise disjoint, and tni¼1 si U i B: exist:

It is clear that if A ≺ B, then m(A)  m(B) for (1) A castle {(Vi, Si)}i  I such that each level is
every G-invariant probability measure m. open with diameter at most δ, and each shape
518 Dynamical Systems and C
-Algebras

is (K, δ)-invariant (i.e., |gSiΔSi|/|Si| < δ for all Theorem 32 (Kerr 2020, Theorem 9.2 and The-
g  K and all i  I ) orem 12.4). Let α : G ↷ X be a free minimal action
(2) A set S0i Si for each i  I such that S0i < of a countable discrete amenable group on a
djSi j and compact metrizable space.

X∖ Si V i ≺ S0i V i ð2Þ (1) If the action α is almost finite, then it has

iI iI comparison.
(2) If the action α is almost finite and G is infinite,
then C(X)⋊rG is Z -stable:
This is reminiscent of the celebrated Ornstein–
Weiss tower decomposition (Ornstein and Weiss Example 25 Since almost finiteness implies that
1987) from ergodic theory. Essentially, almost the crossed product C(X)⋊rG is Z -stable, the
finiteness means that except for a small part, the minimal actions constructed by Giol and Kerr
space can be covered by disjoint open towers whose (2010) are not almost finite.
levels have small diameter. Here the “smallness” of
the remainder is captured by the notion of dynami-
Let us explicitly state the following corollary,
cal subequivalence, introduced in Definition 16.
which can be deduced from Theorem 32, Theo-
Note that if an action of a group G on a space
rem 31, and Theorem 30. It shows that
X is almost finite, then G is amenable by the
establishing almost finiteness, a purely dynamical
(K, δ)-invariance requirement in condition (i).
property, directly contributes to the realm of C
-
Therefore, in some sense almost finiteness can
algebras by providing more examples of classifi-
be viewed as a dynamical analogue of the con-
able algebras.
junction of nuclearity and Z -stability:

Remark 22 The term “almost finiteness” was Corollary 5 Let α : G ↷ X be a free minimal
first introduced by Matui. In Matui (2012), action of a countable discrete amenable group on
Matui defined almost finiteness for étale a compact metrizable space. If the action is almost
groupoids whose unit space is compact and totally finite, then the crossed product C(X)⋊rG is clas-
disconnected and studied the homology groups of sified by its Elliott Invariant.
these groupoids. It was shown in Kerr (2020) that
this definition is equivalent to Definition 18 for Another very natural way to quantify condition
transformation groupoids arising from groups act- (ii) in Definition 18 is through invariant measures.
ing on compact totally disconnected spaces. This leads to the following version of almost
finiteness, given in Kerr and Szabó (2020).
Example 23 The Kakutani-Rokhlin tower decom-
position shows that every Cantor minimal system is Definition 19 (Kerr and Szabó G (2020) Defini-
almost finite. Matui proved in Matui (2012) that the tion 3.5 and Proposition 3.3) Let G be a discrete
same holds for every free action of ℤn on a compact group and let X be a compact metrizable G-space.
metrizable totally disconnected space. Then, an action G ↷ X is almost finite in measure
if it satisfies the condition (i) in Definition 18 and
Example 24 By Kerr and Szabó (2020, Theorem the following condition (ii’):
B), the results from the previous example can be
extended to actions of ℤn on any finite-
dimensional compact metrizable spaces. m X∖ Si V i <d
iI
The next theorem of Kerr relates almost finite-
ness to comparison and Z -stability of the associ- where the supremum is taken over all G-invariant
ated crossed product. Borel probability measures m.
Dynamical Systems and C
-Algebras 519

If we allow, in the above definition, the levels the case of ℤ-actions was first established in
to have arbitrary diameter then we obtain pre- Lindenstrauss (2000)). It is a major open problem
cisely the uniform Rokhlin property discussed in whether the equivalence of these two properties
the section “Mean Dimension and Radius of still holds for actions of countable amenable
Comparison.” groups.
Intuitively, almost finiteness implies almost The following theorem is due to Kerr and
finiteness in measure, and comparison allows Szabó.
going backward. This turns out to be true (see
the proof of Kerr and Szabó (2020), Theorem Theorem 33 (Kerr and Szabó (2020), Theorem
6.1), so by Theorem 32 comparison together 5.6). Let G ↷ X be a free action of a countable
with almost finiteness in measure implies that discrete amenable group on a compact metrizable
the crossed product is Z -stable. To state another space. Then the action is almost finite in measure
characterization of almost finiteness in measure, if and only if it has the small boundary property.
let us recall the definition of the small boundary
property introduced by Lindenstrauss and Weiss As a consequence of Theorem 33 and Theorem
(2000). 32, the following are equivalent for a free action
of a countable amenable group on a compact
Definition 20 Let G be a discrete amenable group metrizable space:
and X be a G-space. An action G ↷ X has the small
boundary property if for every point x  X and (1) The action is almost finite.
every open neighborhood U of x, there is an open (2) The action has comparison and the small
set V such that x  V U and m V∖V ¼ 0 for boundary property.
every G-invariant probability measure m.
It might be interesting to compare this result
The definition of the small boundary property to the equivalence between (ii) and (iii) in
was first given by Linden-strauss and Weiss using Theorem 31.
the notion of orbit capacity (Lindenstrauss and Returning to the connection to the classifica-
Weiss (2000), Definition 5.1) and was proved to tion of nuclear C
-algebras, we recall that by
be equivalent to the above definition in Proposi- Theorem 31 the Toms–Winter conjecture holds
tion 3.3 of Kerr and Szabó (2020). for any simple crossed product C(X)⋊rG that
has uniform property Γ. The next result establishes
Remark 23 Recall that a topological space a formal link between the small boundary prop-
X has zero (small) inductive dimension if for erty and uniform property Γ.
every x  X and every open set U containing x,
there exists an open set V such that x  V U Theorem 34 (Kerr and Szabó (2020), Theorem
and V has empty boundary (see, e.g., Pears 9.4). Let G be an infinite countable discrete ame-
(1975), Chap. 4). Therefore, the small boundary nable group and let X be a compact metrizable
property can be viewed as a dynamical analogue G-space. If the action G ↷ X has the small bound-
of a zero-dimensional space where the smallness ary property, then the crossed product C(X)⋊rG
of the boundary is captured by the G-invariant has uniform property Γ.
measures.
As a consequence, if mean dimension zero of
It was shown in Lindenstrauss and Weiss the action turns out to imply strict comparison for
(2000) that the small boundary property implies the crossed product (note that this is a very special
mean dimension zero. The converse is known to case of Conjecture 1), then every free minimal
be true for ℤn-actions (see Gutman et al. (2016); action G ↷ X with the small boundary property
520 Dynamical Systems and C
-Algebras

gives rise to a Z -stable, and hence classifiable The Furstenberg Boundary

crossed product. Note that the same conclusion Let X be a compact Hausdorff space and G be a
can also be reached if every such action has com- discrete group. Let C(X) denote the C
-algebra of
parison, which is another open question (see all continuous functions on X, and P(X) the set of
Remark 21). Along these lines, one can ask the probability measures on X, which is a compact
following question: convex subset of C(X)
(in the weak
topology).
If X is a G-space, then P(X) carries a natural
Question 1 Let α : G ↷ X be a free minimal G-space structure. Recall that a G-space X is prox-
action of a countable discrete infinite amenable imal if for every pair x1, x2 in X, there is a net ti in
group on a compact metrizable space. Does G such that limitix1 ¼ limitix2. We say that X is
mdim(X, α, G) ¼ 0 imply that C(X)⋊rG is strongly proximal if P(X) is proximal as a G-space
Z -stable? (see Glasner (1976)). A compact G-space X is
called a G-boundary if it is minimal and strongly
Note that the question has an affirmative proximal, or equivalently if X is the unique min-
answer for ℤn-actions, by a combination of The- imal G-invariant closed subspace of M(X).
orem 28, Gutman et al. (2016, Corollary 5.4), and By a result of Furstenberg (Glasner 1976, The-
Theorem 34. orem III.2.3), if G acts by affine transformations
on a locally convex topological vector space and
K is a G-invariant compact convex subset, then
Boundary Actions and C
-Simplicity K contains a G-boundary. This shows that bound-
ary actions are ubiquitous.
In 1967, Dixmier asked if every simple C
-algebra
is generated by its projections. Kadison suggested Theorem 35 (Furstenberg 1973). Every discrete
that the reduced C
-algebra C
r ð2 Þ of the free group G admits a unique universal G-boundary
group on two generators should provide an exam- @ FG, called the Furstenberg boundary of G, in the
ple of a simple C
-algebra with no nontrivial pro- sense that for every G-boundary X there is a
jections. Powers (1975) proved that C
r ð2 Þ is surjective G-map from @ FG onto X.
simple and has exactly one tracial state. Since
then more examples of groups with these two Note that by the universal property of @ FG and
properties have been found, and it became a the paragraph before Theorem 35, for any com-
major problem to characterize the groups with pact G-space X, there exists a G-equivariant con-
either property. In particular, for a discrete group tinuous map b : @ FG ! P(X), called a boundary
G, it was open for a long time whether simplicity map.
of C
r ðGÞ is equivalent to the unique trace prop- Let X and Y be compact G-spaces. There is a
erty, that is, there is exactly one tracial state on one-to-one correspondence between continuous
C
r ðGÞ: G-equivariant maps a : Y ! P(X) and G-equivariant
Kalantar and Kennedy (2017) identified the unital positive maps a : CðXÞ ! CðY Þ given by
Furstenberg boundary of a discrete group G,
which is a purely dynamical notion, with an oper- að f ÞðyÞ ¼ haðyÞ, f i,
ator algebraic notion called the Hamana boundary.
Using this identification, they characterized the where h , i is the canonical pairing between P(X)
simplicity of C
r ðGÞ in terms of the action of and C(X). Applying this correspondence to the
G on its Furstenberg boundary. Building upon boundary map b : @ FG ! P(βG), where βG is
this new insight, Breuillard, Kalantar, Kennedy, the Stone–Čech compactification of G, we obtain
and Ozawa gave solutions to many long-standing a G-equivariant unital positive map b : CðbGÞ !
questions surrounding C
-simplicity and the Cð@ F GÞ: For a point x  @ FG, we write evx :
unique trace property (Breuillard et al. 2017). Cð@ F GÞ !  for the evaluation map. One can
Dynamical Systems and C
-Algebras 521

show that the composition evx ∘ b restricts to an said to be completely isometric if its matrix ampli-
invariant mean for the stabilizer subgroup Gx (see fication ’ðnÞ : Mn ðS Þ ! Mn ðT Þ is an isometry for
Breuillard et al. (2017, Proposition 2.7)). It fol- every n  
lows that for every x  @ FG, the stabilizer sub-
group Gx is amenable. Definition 22 (Hamana 1985). Let S be an oper-
Day proved that every group has an amenable ator system and let G be a discrete group.
normal subgroup Ra(G) that contains all the other
amenable normal subgroups of G (Day (1957), (1) A G-action on S is a group homomorphism
Lemma 4.1). The group Ra(G) is called the ame- from G into the group of complete order iso-
nable radical of G. The previous paragraph, morphisms on S. In this case, S is called a
together with strong proximality and minimality G-operator system.
of @ FG, yields the next theorem. (2) A map ’ : S ! T between two G-operator
Recall that an action of G on X is a group systems is a G-equivariant map if
homomorphism from G into Homeo(X). The ker-
’ðgaÞ ¼ g’ðaÞ
nel of the action is by definition the kernel of this
homomorphism.
for all a  S and g  G.
Theorem 36 (Furman (2003); see also Breuillard
Definition 23 (Hamana 1985). An operator sys-
et al. (2017), Proposition 2.8). The kernel of the
tem U is said to be injective if for any inclusion
action G ↷ @ FG is precisely the amenable radical
S T of operator systems and any u.c.p. map
Ra(G).
’ : S ! U, there exists a u.c.p. map ’ : T ! U
such that ’ðaÞ ¼ ’ðaÞ for all a  S (i.e., ’ is an
In particular, if G is amenable, then @ FG is a
extension of ’).
one-point space. In general, @ FG is extremally
disconnected, that is, every open subset of @ FG
has an open closure (Kalantar and Kennedy Example 26
(2017), Remark 3.16; see also Breuillard et al.
(2017), Proposition 2.4). (1) Arveson’s extension theorem (see, e.g.,
Paulsen (2002), Chap. 7) shows that B(H ) is
The Hamana Boundary injective for any Hilbert space H.
Definition 21 A unital self-adjoint subspace of a (2) If V U is an inclusion of operator systems
unital C
-algebra is called an operator system. such that U is injective and there is a u.c.p.
projection from U onto V , then V is necessar-
Recall that a linear map ’ : A ! B between two ily injective. It follows that an operator system


C -algebras is positive if ’(A+) B+, and V BðH Þ is injective if and only if there exists
completely positive (c.p.) if for every n  N , the a u.c.p. projection from B(H) onto V
matrix amplification ’(n) : Mn(A) ! Mn(B),
’(n)([aij]) ¼ [’(aij)] is positive (see, e.g., Paulsen Definition 24 Let G be a discrete group.
(2002)). Suppose A is a unital C
-algebra and A G-operator system U is said to be G-injective
S A is an operator system. Then for every if for any inclusion S T of G-operator systems
n  , the operator system Mn ðS Þ inherits a norm and any G-equivariant u.c.p. map ’ : S ! U,
and order structure from Mn(A). Therefore, we can there exists a G-equivariant u.c.p. map ’ : T !
define c.p. maps between operator systems. An U such that ’ðaÞ ¼ ’ðaÞ for all a  S (i.e., ’ is an
invertible and unital completely positive (u.c.p.) extension of ’).
map between two operator systems whose inverse
is also c.p. is called a complete order isomorphism Example 27 Let V be an injective operator sys-
(see Blackadar (2006), II.6.9.16). Moreover, a tem and let G be a discrete group. The operator
map ’ : S ! T between two operator systems is system ‘1 ðG, V Þ admits a G-action defined by
522 Dynamical Systems and C
-Algebras

s:f ðgÞ ¼ f s1 g ð f  ‘1 ðG, V Þ, s, g  GÞ I G ðÞ ! ‘1 ðGÞ: By G-essentiality of I G ðÞ, the

map ’ is a complete isometry, so we identify
By Hamana (1985, Lemma 2.2), ‘1 ðG, V Þ is I G ðÞ with its image in ‘1(G). Since I G ðÞ is
G-injective. G-injective, the identity map on I G ðÞ extends to
Given a G-operator system S, we are interested a (G-equivariant) u.c.p. projection
in finding a minimal G-injective operator system
that contains S. c : ‘1 ðGÞ ! I G ðÞ:

Definition 25 Let G be a discrete group and S be It follows that I G ðÞ is an injective operator
a G-operator system. (Kalantar and Kennedy system.
2017, p. 5). By Choi and Effros (1977, Theorem 3.1), the
injective envelope I G ðÞ becomes a C
-algebra
(1) A G-extension of S is a pair ðT , kÞ of a when equipped with the Choi-Effros product
G-operator system T and a G-equivariant
unital complete isometry k : S ! T . a ∘ b≔cðabÞ:
(2) A G-extension ðT , kÞ is said to be
• G-injective if T is G-injective; It is clear that the product is commutative,
• G-essential if for every G-equivariant hence by the Gelfand–Naimark theorem I G ðÞ is
u.c.p. map ’ : T ! W , ’ is a complete isomorphic to the algebra of continuous functions
isometry whenever ’ ∘ k is on some compact Hausdorff space.

Example 28 Let S BðH Þ be a G-operator sys- Definition 27 Let G be a discrete group. The
tem. Consider the map k : S ! ‘1 ðG, BðHÞÞ Hamana boundary @ HG is the compact Hausdorff
defined by space that satisfies Cð@ H GÞ ¼ I G ðÞ:

½kðsÞ ðgÞ≔g1 s ðs  S, g  GÞ: By the Gelfand–Naimark correspondence, the

action of G on I G ðÞ induces a G-action on @ HG.

Then the pair (‘1(G, B(H )), k) is a Proposition 11 (Kalantar and Kennedy (2017),
G-extension of S, and by examples 26 and 27, Proposition 3.4 and Proposition 3.7). The
this extension is G-injective. G-action on @ HG is minimal and strongly proxi-
mal, that is, @ HG is a G-boundary.
Definition 26 Let S be a G-operator system.
A G-extension of S that is both G-injective and Theorem 38 (Kalantar and Kennedy (2017),
G-essential is called a G-injective envelope of S. Theorem 3.11). @ HG ¼ @ FG.

Theorem 37 (Hamana (1985), Theorem 2.5).

Every G-operator system has a G-injective enve- C
-Simplicity and Unique Trace Property
lope, which is unique up to G-equivariant com-
plete order isomorphism. Let G be a discrete group. The left regular repre-
sentation l : G ! B(‘2(G)) is defined by lg(δs) ¼
Following Hamana (1985), we write I G ðS Þ for δgs, where δg is the point mass. The linear span of
the G-injective envelope of a G-operator system S {lg : g  G} forms a
-subalgebra of B(‘2(G)). Its
We now focus on the case S ¼  (equipped norm closure is called the reduced group C
-alge-
with the trivial action of G). Since ‘1 ðGÞ ¼ bra of G and will be denoted by C
r ðGÞ: We say
‘1 ðG, Þ is G-injective, the inclusion map  ! G is C
-simple if C
r ðGÞ is a simple C
-algebra.
‘1 ðGÞ extends to a G-equivariant u.c.p. map ’ : Note that C
r ðGÞ has a canonical tracial state
Dynamical Systems and C
-Algebras 523

tl(a) ¼ haδg, δgi. We say G has the unique trace Corollary 6 Every discrete C
-simple group has
property if tl is the only tracial state on C
r ðGÞ: the unique trace property. (The converse was later
The following groundbreaking result was first shown to be false by Le Boudec (2017).)
proved by Kalantar and Kennedy (2017, Theorem
6.2) and later reproved by Breuillard, Kalantar, Theorem 39 also led to algebraic criterions for
Kennedy, and Ozawa (2017, Theorem 3.1). C
-simplicity. Following Breuillard et al. (2017)
we say a subgroup H of G is normalish if for any
Theorem 39 Let G be a discrete group. Then the n  1 and t1, . . ., tn  G, the intersection \i ti Ht 1
i
following are equivalent: is infinite.

(1) G is C
-simple. Theorem 41 (Breuillard et al. (2017), Theorem

(2) G acts topologically freely on its Furstenburg 6.2). A discrete group G is C
-simple if it has no
boundary @ FG. nontrivial finite normal subgroups and no amena-
(3) There exists a topologically free G-boundary. ble normalish subgroups.

Using Theorem 41, Breuillard et al. (2017) not

Theorem 39 allows one to deduce C
-simplic-
only recovered a large class of previously known
ity of a group by exhibiting a topologically free
C
-simple groups but also exhibited many new
boundary action.
ones. For example, if a countable group
G admits a free probability measure-preserving
Example 29 Let Γ be a word hyperbolic group
action with cost strictly larger than 1 (in the
(such as the free group 2 ). The action of Γ on its
sense of Gaboriau (2000)) and has no nontrivial
Gromov boundary @Γ can be shown to be a topo-
finite normal subgroups, then G is C
-simple (see
logically free boundary action (Ghys and de la
Breuillard et al. (2017, Corollary 6.8)).
Harpe 1988; Laca and Spielberg 1996), whence
Kennedy (2020) obtained an intrinsic charac-
by Theorem 39 the group Γ is C
-simple.
terization of C
-simplicity in terms of uniformly
recurrent subgroups. The notion of uniformly
Although we will not discuss the proof of Theo-
recurrent subgroups is introduced by Glasner
rem 39, it is worth mentioning that the crossed
and Weiss (2015). For a discrete group G, let
product C(@ FG)⋊rG played an important role in
S ðGÞ be the set of all subgroups of G. We equip
both Kalantar and Kennedy (2017) and Breuillard
S ðGÞ with the Chabauty topology, which in this
et al. (2017). The techniques developed in Breuillard
case coincides with the subspace topology
et al. (2017) also led to the following theorem,
inherited from the product space {0, 1}G. The
which answers a long-standing open question.
space S ðGÞ has a G-space structure via the conju-
gation action H 7! gHg1. A uniformly recurrent
Theorem 40 (Breuillard et al. (2017), Corollary
subgroup is a minimal G-space X S ðGÞ:
4.3). A discrete group G has the unique trace
property if and only if its amenable radical Theorem 42 (Kennedy (2020), Theorem 4.1).
Ra(G) is trivial. (Haagerup (2017) gave another A discrete group G is C
-simple if and only if it
proof of this result.) has no nontrivial amenable uniformly recurrent
subgroups.
Combining Theorem 40, Theorem 39, and
Theorem 36 yields the following corollary and Using this result, one can obtain a more alge-
hence resolves another long-standing open ques- braic characterization. Following Kennedy (2020),
tion. It should be noted that one can also deduce we say a subgroup H of G is residually normal if
the triviality of Ra(G) directly from simplicity of there is a finite subset F G\{e} such that the
the C
-algebra C
r ðGÞ (see de la Harpe (2007), intersection F \ gHg1 is nonempty for every
Proposition 3). g  G.
524 Dynamical Systems and C
-Algebras

Theorem 43 (Kennedy (2020), Theorem 5.1). A Boyle M, Handelman D (1994) Entropy versus orbit equiv-
discrete group G is C
-simple if and only if it has alence for minimal homeomorphisms. Pac J Math
164(1):1–13
no amenable residually normal subgroups. Boyle M, Tomiyama J (1998) Bounded topological orbit
equivalence and C
-algebras. J Math Soc Japan 50(2):
The results and techniques obtained in 317–329
Kalantar and Kennedy (2017) and Breuillard Bratteli O (1972) Inductive limits of finite dimensional C
-
algebras. Trans Am Math Soc 171:195–234
et al. (2017) have led to many further develop- Breuillard E, Kalantar M, Kennedy M, Ozawa N (2017)
ments and research directions, including, but not C
-simplicity and the unique trace property for discrete
limited to, the study of ideal structure for crossed groups. Publ Math Inst Hautes Études Sci 126:35–71
products and groupoid C
-algebras (Borys 2019; Brown NP, Ozawa N (2008) C
-algebras and finite-
dimensional approximations. Graduate studies in math-
Bryder 2017; Kalantar and Scarparo 2021; ematics, vol 88. American Mathematical Society,
Kawabe 2017; Kennedy et al. 2021; Kennedy Providence
and Schafhauser 2019), amenability of Bryder RS (2017) Injective envelopes and the intersection
Thompson’s groups (Le Boudec and Bon 2018), property. Preprint. arXiv:1704.02723
Carlsen TM, Ruiz E, Sims A, Tomforde M (2021) Recon-
and stationary C
-dynamical systems (Hartman struction of groupoids and C
-rigidity of dynamical
and Kalantar 2017). systems. Advanced Mathematics, 390: Paper
No. 107923, 55
Castillejos J, Evington S, Tikuisis A, White S, Winter
W (2021) Nuclear dimension of simple C
-algebras.
References Invent Math 224(1):245–290
Choi MD, Effros EG (1977) Injectivity and operator
Alvin L, Ash DD, Ormes NS (2018) Bounded topological spaces. J Funct Anal 24(2):156–209
speedups. Dyn Syst 33(2):303–331 Conley C, Jackson S, Marks A, Seward B, Tucker-Drob
Ara P, Lledó F, Perera F, eds (2011) Aspects of operator R (2020) Borel asymptotic dimension and hyperfinite
algebras and applications, volume 534 of Contempo- equivalence relations. Preprint. arXiv:2009.06721
rary Mathematics. American Mathematical Society, Connes A (1976) Classification of injective factors. Cases
Providence; Real Sociedad Matemática Española, II1, II1, IIIl, l 6¼ 1. Ann Math 104(1):73–115
Madrid. Papers from the UIMP-RSME Lluís Connes A, Feldman J , Weiss B (1982) An amenable
A. Santaló Summer School in Mathematics held at the equivalence relation is generated by a single transfor-
Universidad Internacional Menéndez Pelayo, Santan- mation. Ergodic Theory Dynam Syst, 1(4):431–450
der, July 21–25, 2008 Coornaert M (2015) Topological dimension and dynamical
Arnoux P, Ornstein DS, Weiss B (1985) Cutting and systems. Universitext. Springer, Cham. Translated and
stacking, interval exchanges and geometric models. revised from the 2005 French original
Israel J Math 50(1–2):160–168 Cortez MI (2006) ℤd Toeplitz arrays. Discrete Contin Dyn
Ash DD (2016) Topological speedups. ProQuest LLC, Syst 15(3):859–881
Ann Arbor, MI, Thesis (Ph.D.), University of Denver Cortez MI, Medynets K (2016) Orbit equivalence rigidity
Bezuglyi S, Karpel O (2016) Bratteli diagrams: structure, of equicontinuous systems. J Lond Math Soc 94(2):
measures, dynamics. In: Dynamics and numbers, 545–556
volume 669 of Contemporary Mathematics. American Cortez MI, Petite S (2014) Invariant measures and orbit
Mathematical Society, Providence, pp 1–36 equivalence for generalized Toeplitz subshifts. Groups
Bezuglyi S, Dooley AH, Medynets K (2005) The Rokhlin Geom Dyn 8(4):1007–1045
lemma for homeomorphisms of a Cantor set. Proc Am Crainic M, Moerdijk I (2000) A homology theory for étale
Math Soc 133(10):2957–2964 groupoids. J Reine Angew Math 521:25–46
Blackadar B (2006) Operator algebras, volume 122 of Cuntz J (1978) Dimension functions on simple C
-alge-
Encyclopaedia of mathematical sciences. Springer, bras. Math Ann 233(2):145–153
Berlin. Theory of C
-algebras and von Neumann alge- Dahl H (2008) AF equivalence relations associated to
bras, Operator Algebras and Non-commutative locally finite groups. J Ramanujan Math Soc 23(1):
Geometry, III 77–95
Blackadar B, Handelman D (1982) Dimension functions Day MM (1957) Amenable semigroups. Ill J Math 1:
and traces on C
-algebras. J Funct Anal 45(3):297–340 509–544
Borys C (2019) The Furstenberg boundary of a groupoid. de Cornulier Y (2014) Groupes pleins-topologiques
Preprint. arXiv:1904.10062 (d’après Matui, Juschenko, Monod, . . .). Astérisque,
Boyle MM (1983) Topological orbit equivalence and fac- (361):Exp. No. 1064, viii, pp 183–223
tor maps in symbolic dynamics. ProQuest LLC, Ann de la Harpe P (2007) On simplicity of reduced C
-algebras
Arbor, MI. Thesis (Ph.D.), University of Washington of groups. Bull Lond Math Soc 39(1):1–26
Dynamical Systems and C
-Algebras 525

Dougherty R, Jackson S, Kechris AS (1994) The structure Gähler F, Hunton J, Kellendonk J (2013) Integral coho-
of hyperfinite Borel equivalence relations. Trans Am mology of rational projection method patterns. Algebr
Math Soc 341(1):193–225 Geom Topol 13(3):1661–1708
Downarowicz T (2005) Survey of odometers and Toeplitz Gao S, Jackson S (2015) Countable abelian group actions
flows. In: Algebraic and topological dynamics, and hyperfinite equivalence relations. Invent Math
volume 385 of Contemporary Mathematics. American 201(1):309–383
Mathematical Society, Providence, pp 7–37 Ghys É, de la Harpe P, eds (1988) Sur les groupes hyper-
Downarowicz T, Zhang G (2017) The comparison property boliques d’après Mikhael Gromov, volume 83 of Pro-
of amenable groups. arXiv:1901.01457 gress in Mathematics. Birkhäuser Boston, Inc., Boston,
Downarowicz T, Zhang G (2019) Symbolic extensions of MA, 1990. Papers from the Swiss Seminar on Hyper-
amenable group actions and the comparison property. bolic Groups held in Bern
arXiv:1712.05129 Giol J, Kerr D (2010) Subshifts and perforation. J Reine
Durand F (2010) Combinatorics on Bratteli diagrams and Angew Math 639:107–119
dynamical systems. In Combinatorics, automata and Giordano T, de la Harpe P (1997) Moyennabilité des
number theory, volume 135 of Encyclopedia Math. groupes dénombrables et actions sur les espaces de
Appl. Cambridge University Press, Cambridge, Cantor. C R Acad Sci 324(11):1255–1258
pp 324–372 Giordano T, Putnam IF, Skau CF (1995) Topological orbit
Durand F, Perrin D (2022) Dimension groups and dynam- equivalence and C
-crossed products. J Reine Angew
ical systems—substitutions, Bratteli diagrams and Can- Math 469:51–111
tor systems, volume 196 of Cambridge Studies in Giordano T, Putnam IF, Skau CF (1999) Full groups of
Advanced Mathematics. Cambridge University Press, Cantor minimal systems. Israel J Math 111:285–320
Cambridge Giordano T, Putnam I, Skau C (2004) Affable equivalence
Dye HA (1959) On groups of measure preserving trans- relations and orbit structure of Cantor dynamical sys-
formations. I. Am J Math 81:119–159 tems. Ergodic Theory Dynam Syst 24(2):441–475
Dye HA (1963) On groups of measure preserving trans- Giordano T, Matui H, Putnam IF, Skau CF (2008) Orbit
formations. II. Am J Math 85:551–576 equivalence for Cantor minimal ℤ2-systems. J Am
Effros EG (1981) Dimensions and C
-algebras, Math Soc 21(3):863–892
volume 46 of CBMS Regional conference series in Giordano T, Matui H, Putnam IF, Skau CF (2010) Orbit
Mathematics. Conference Board of the Mathematical equivalence for Cantor minimal ℤd-systems. Invent
Sciences, Washington, DC Math 179(1):119–158
Effros EG, Handelman DE, Shen CL (1980) Dimension Giordano T, Kerr D, Phillips NC, Toms A (2018) Crossed
groups and their affine representations. Am J Math products of C
-algebras, topological dynamics, and
102(2):385–407 classification. Advanced Courses in Mathematics.
Elliott GA (1976) On the classification of inductive limits CRM Barcelona. Birkhäuser/Springer, Cham. Lecture
of sequences of semisimple finite-dimensional alge- notes based on the course held at the Centre de Recerca
bras. J Algebra 38(1):29–44 Matemàtica (CRM) Barcelona, June 14–23, 2011
Elliott GA, Niu Z (2013) On the radius of comparison of a Edited by Francesc Perera
commutative C
-algebra. Can Math Bull 56(4):737–744 Giordano T, Putnam IF, Skau CF (2019) ℤd-odometers and
Elliott GA, Niu Z (2017) The C
-algebra of a minimal cohomology. Groups Geom Dyn 13(3):909–938
homeomorphism of zero mean dimension. Duke Math Glasner S (1976) Proximal flows. Lecture notes in Mathe-
J 166(18):3569–3594 matics, vol 517. Springer, Berlin/New York
Ellis R (1960) Universal minimal sets. Proc Am Math Soc Glasner E (2003) Ergodic theory via joinings,
11:540–543 volume 101 of Mathematical surveys and monographs.
Feldman J, Moore CC (1977) Ergodic equivalence rela- American Mathematical Society, Providence
tions, cohomology, and von Neumann algebras. Glasner E, Weiss B (1995) Weak orbit equivalence of
I. Trans Am Math Soc 234(2):289–324 Cantor minimal systems. Int J Math 6(4):559–579
Forrest AH, Hunton J (1999) The cohomology and Glasner E, Weiss B (2015) Uniformly recurrent subgroups.
K-theory of commuting homeomorphisms of the Can- In: Recent trends in ergodic theory and dynamical
tor set. Ergodic Theory Dynam Syst 19(3):611–625 systems, volume 631 of Contemp. Math. American
Furman A (1999) Orbit equivalence rigidity. Ann Math Mathematical Society, Providence, pp 63–75
150(3):1083–1108 Gong G, Lin H, Niu Z (2020a) A classification of finite
Furman A (2003) On minimal strongly proximal actions of simple amenable Z -stable C
-algebras, I: C
-algebras
locally compact groups. Israel J Math 136:173–187 with generalized tracial rank one. C R Math Acad Sci
Furstenberg H (1973) Boundary theory and stochastic pro- Soc R Can 42(3):63–450
cesses on homogeneous spaces. In Harmonic analysis Gong G, Lin H, Niu Z (2020b) A classification of finite
on homogeneous spaces (Proc. Sympos. Pure Math., simple amenable Z -stable C
-algebras, II: C
-algebras
Vol. XXVI, Williams Coll., Williamstown, MA., with rational generalized tracial rank one. C R Math
1972), pp 193–229 Acad Sci Soc R Can 42(4):451–539
Gaboriau D (2000) Coût des relations d’équivalence et des Guentner E, Willett R, Guoliang Y (2017) Dynamic
groupes. Invent Math 139(1):41–98 asymptotic dimension: relation to dynamics, topology,
526 Dynamical Systems and C
-Algebras

coarse geometry, and C
-algebras. Math Ann Kalantar M, Kennedy M (2017) Boundaries of reduced C
-
367(1–2):785–829 algebras of discrete groups. J Reine Angew Math 727:
Gutman Y, Lindenstrauss E, Tsukamoto M (2016) Mean 247–267
dimension of ℤk-actions. Geom Funct Anal 26(3): Kalantar M, Scarparo E (2021) Boundary maps and covar-
778–817 iant representations. Preprint. arXiv:2106.06382
Haagerup U (1987) Connes’ bicentralizer problem and Katzlinger L (2019) Topological full groups. Preprint.
uniqueness of the injective factor of type III1. Acta arXiv:1907.07424
Math 158(1–2):95–148 Kawabe T (2017) Uniformly recurrent subgroups and the
Haagerup U (2014) Quasitraces on exact C
-algebras are ideal structure of reduced crossed products. Preprint.
traces. C R Math Acad Sci Soc R Can 36(2-3):67–92 arXiv:1701.03413
Haagerup U (2017) A new look at C
-simplicity and the Kennedy M (2020) An intrinsic characterization of C
-
unique trace property of a group. In Operator algebras simplicity. Ann Sci Éc Norm Supér 53(5):1105–1119
and applications—the Abel Symposium 2015, Kennedy M, Schafhauser C (2019) Noncommutative
volume 12 of Abel Symp. Springer, Cham, pp 167–176 boundaries and the ideal structure of reduced crossed
Hamana M (1985) Injective envelopes of C
-dynamical products. Duke Math J 168(17):3215–3260
systems. Tohoku Math J 37(4):463–487 Kennedy M, Kim S-J, Li X, Raum S, Ursu D (2021) The
Hartman Y, Kalantar M (2017) Stationary C
-dynamical ideal intersection property for essential groupoid C
-
systems. To appear in J. Eur. Math. Soc. (includes an algebras. Preprint. arXiv:2107.03980
appendix by Uri Bader, Yair Hartman, and Mehrdad Kerr D (2020) Dimension, comparison, and almost finite-
Kalantar). arXiv:2107.03980 ness. J Eur Math Soc (JEMS) 22(11):3697–3745
Herman RH, Putnam IF, Skau CF (1992) Ordered Bratteli Kerr D, Szabó G (2020) Almost finiteness and the small
diagrams, dimension groups and topological dynamics. boundary property. Commun Math Phys 374(1):1–31
Int J Math 3(6):827–864 Kirchberg E (1995) Exact C
-algebras, tensor products,
Hines T (2015) The radius of comparison and mean dimen- and the classification of purely infinite algebras. In
sion. ProQuest LLC, Ann Arbor, MI, Thesis (Ph.D.)– Proceedings of the International Congress of Mathema-
Purdue University ticians, vol. 1, 2 (Zürich, 1994), pp 943–954.
Hirshberg I, Winter W, Zacharias J (2015) Rokhlin dimen- Birkhäuser, Basel
sion and C
-dynamics. Commun Math Phys 335(2): Krieger W (1976) On ergodic flows and the isomorphism
637–670 of factors. Math Ann 223(1):19–70
Hjorth G, Molberg M (2006) Free continuous actions on Krieger W (1979/80) On a dimension for a class of homeo-
zero-dimensional spaces. Topology Appl 153(7): morphism groups. Math Ann, 252(2):87–95
1116–1131 Kuratowski K (1968) Topology. Vol. II. Academic Press,
Hopf E (1932) Theory of measure and invariant integrals. New York-London; Państwowe Wydawnictwo
Trans Am Math Soc 34(2):373–393 Naukowe [Polish Scientific Publishers], Warsaw. New
Høynes S-M (2016) Toeplitz flows and their ordered edition, revised and augmented, Translated from the
K-theory. Ergodic Theory Dynam Syst 36(6): French by A. Kirkor
1892–1921 Laca M, Spielberg J (1996) Purely infinite C
-algebras
Hunton J (2015) Spaces of projection method patterns and from boundary actions of discrete groups. J Reine
their cohomology. In: Mathematics of aperiodic order, Angew Math 480:125–139
volume 309 of Progr. Math. Birkhäuser/Springer, Le Boudec A (2017) C
-simplicity and the amenable rad-
Basel, pp 105–135 ical. Invent Math 209(1):159–174
Ioana A (2011) W
-superrigidity for Bernoulli actions of Le Boudec A, Bon NM (2018) Subgroup dynamics and C
-
property (T) groups. J Am Math Soc 24(4):1175–1226 simplicity of groups of homeomorphisms. Ann Sci Éc
Ioana A (2013) Classification and rigidity for von Neu- Norm Supér 51(3):557–602
mann algebras. In: European Congress of Mathematics. Li H (2013) Sofic mean dimension. Adv Math 244:
European Mathematical Society, Zürich, pp 601–625 570–604
Ioana A (2018) Rigidity for von Neumann algebras. In: Li X (2018a) Continuous orbit equivalence rigidity. Ergo-
Proceedings of the International Congress of Mathema- dic Theory Dynam Syst 38(4):1543–1563
ticians—Rio de Janeiro 2018. Vol. III. Invited lectures. Li X (2018b) Dynamic characterizations of quasi-isometry
World Sci. Publ., Hackensack, pp 1639–1672 and applications to cohomology. Algebr Geom Topol
Jackson S, Kechris AS, Louveau A (2002) Countable 18(6):3477–3535
Borel equivalence relations. J Math Log 2(1):1–80 Lindenstrauss E (2000) Mean dimension, small entropy
Jiang X, Su H (1999) On a simple unital projectionless C
- factors and an embedding theorem. Inst Hautes Études
algebra. Am J Math 121(2):359–413 Sci Publ Math (89):227–262. 1999
Johansen Ø (1998) Ordered K-theory and Bratteli dia- Lindenstrauss E, Weiss B (2000) Mean topological dimen-
grams: implications for Cantor minimal systems. Ph. sion. Israel J Math 115:1–24
D. thesis, NTNU Matui H (2008a) An absorption theorem for minimal AF
Johnson ASA, McClendon DM (2022) Topological equivalence relations on Cantor sets. J Math Soc Japan
speedups of ℤd-actions. Dyn Syst 37(2):222–261 60(4):1171–1185
Dynamical Systems and C
-Algebras 527

Matui H (2008b) Torsion in coinvariants of certain Cantor Phillips NC (2016) The C
-algebra of a minimal homeo-
minimal ℤ2-systems. Trans Am Math Soc 360(9): morphism with finite mean dimension has finite radius
4913–4928 of comparison. Preprint. arXiv:1605.07976
Matui H (2012) Homology and topological full groups of Pimsner M, Voiculescu D (1980) Imbedding the irrational
étale groupoids on totally disconnected spaces. Proc rotation C
-algebra into an AF-algebra. J Operator The-
Lond Math Soc 104(1):27–56 ory 4(2):201–210
Matui H (2015) Topological full groups of one-sided shifts Poon YT (1989) A K-theoretic invariant for dynamical
of finite type. J. Reine Angew. Math. 705:35–84 systems. Trans Am Math Soc 311(2):515–533
Matui H (2016) Étale groupoids arising from products of Popa S (2007) Cocycle and orbit equivalence superrigidity
shifts of finite type. Adv Math 303:502–548 for malleable actions of w-rigid groups. Invent Math
Matui H (2017) Topological full groups of étale groupoids. 170(2):243–295
In: Operator algebras and applications—the Abel Sym- Powers RT (1975) Simplicity of the C
-algebra associated
posium 2015, volume 12 of Abel Symposium. with the free group on two generators. Duke Math.
Springer, Cham, pp 203–230 J. 42:151–156
Matui H, Sato Y (2012) Strict comparison and Putnam IF (1989) The C
-algebras associated with mini-
Z -absorption of nuclear C
-algebras. Acta Math. mal homeomorphisms of the Cantor set. Pac J Math
209(1):179–196 136(2):329–353
Matui H, Sato Y (2014) Decomposition rank of UHF- Putnam IF (2010) Orbit equivalence of Cantor minimal
absorbing C
-algebras. Duke Math J 163(14):2687–2708 systems: a survey and a new proof. Expo Math 28(2):
Medynets K (2006) Cantor aperiodic systems and Bratteli 101–131
diagrams. C R Math Acad Sci Paris 342(1):43–46 Putnam IF (2018) Cantor minimal systems, volume 70 of
Medynets K, Sauer R, Thom A (2017) Cantor systems and University lecture series. American Mathematical Soci-
quasi-isometry of groups. Bull Lond Math Soc 49(4): ety, Providence
709–724 Putnam I, Schmidt K, Skau C (1986) C
-algebras associ-
Monod N, Shalom Y (2006) Orbit equivalence rigidity and ated with Denjoy homeomorphisms of the circle.
bounded cohomology. Ann Math 164(3):825–878 J Oper Theory 16(1):99–126
Murray FJ, von Neumann J (1936) On rings of operators. Renault J (1980) A groupoid approach to C
-algebras,
Ann Math 37(1):116–229 volume 793 of Lecture notes in Mathematics. Springer,
Murray FJ, von Neumann J (1943) On rings of operators. Berlin
IV. Ann Math 44:716–808 Renault J (2003) AF equivalence relations and their
Niu Z (2019) Comparison radius and mean topological cocycles. In: Operator algebras and mathematical phys-
dimension: ℤd-actions. Preprint. arXiv:1906.09172 ics (Constantza, 2001). Theta, Bucharest, pp 365–377
Niu Z (2022) Comparison radius and mean topological Renault J (2008) Cartan subalgebras in C
-algebras. Irish
dimension: Rokhlin property, comparison of open Math Soc Bull 61:29–63
sets, and subhomogeneous C
-algebras. J Anal Math Rieffel MA (1981) C
-algebras associated with irrational
146(2):595–672 rotations. Pac J Math 93(2):415–429
Ormes NS (1997) Strong orbit realization for minimal Rørdam M (2003) A simple C
-algebra with a finite and an
homeomorphisms. J Anal Math 71:103–133 infinite projection. Acta Math 191(1):109–142
Ornstein DS, Weiss B (1980) Ergodic theory of amenable Rørdam M (2004) The stable and the real rank of
group actions. I. The Rohlin lemma. Bull Amer Math Z -absorbing C
-algebras. Int J Math 15(10):1065–1084
Soc (NS) 2(1):161–164 Rosenberg J, Schochet C (1987) The Künneth theorem and
Ornstein DS, Weiss B (1987) Entropy and isomorphism the universal coefficient theorem for Kasparov’s gener-
theorems for actions of amenable groups. J Analyse alized K-functor. Duke Math J 55(2):431–474
Math 48:1–141 Sims A, Szabó G, Williams D (2020) Operator algebras
Paterson ALT (1999) Groupoids, inverse semigroups, and and dynamics: groupoids, crossed products, and
their operator algebras, volume 170 of Progress in Rokhlin dimension. Advanced courses in Mathematics.
Mathematics. Birkhäuser Boston, Inc, Boston, MA CRM Barcelona. Birkhäuser/Springer, Cham, © 2020.
Paulsen V (2002) Completely bounded maps and operator Lecture notes from the Advanced Course held at Centre
algebras, volume 78 of Cambridge Studies in de Recerca Matemàtica (CRM) Barcelona, March
Advanced Mathematics. Cambridge University Press, 13–17, 2017, Edited by Francesc Perera.
Cambridge Singer IM (1955) Automorphisms of finite factors. Am
Pears AR (1975) Dimension theory of general spaces. J Math 77:117–133
Cambridge University Press, Cambridge, UK Skau C (2000) Ordered K-theory and minimal symbolic
Phillips NC (2000) A classification theorem for nuclear dynamical systems. Colloq Math 84/85:203–227. Ded-
purely infinite simple C
-algebras. Doc Math 5:49–114 icated to the memory of Anzelm Iwanik
Phillips NC (2005) Crossed products of the Cantor set by Sugisaki F (1998) The relationship between entropy and
free minimal actions of ℤd. Commun Math Phys strong orbit equivalence for the minimal homeomor-
256(1):1–42 phisms. II. Tokyo J Math 21(2):311–351
528 Dynamical Systems and C
-Algebras

Sugisaki F (2003) The relationship between entropy and Tu J-L (1999) La conjecture de Baum-Connes pour les
strong orbit equivalence for the minimal homeomor- feuilletages moyennables. K-Theory 17(3):215–264
phisms. I. Internat J Math 14(7):735–772 Vershik AM (1981) Uniform algebraic approximation of
Szabó G (2015) The Rokhlin dimension of topological ℤm- shift and multiplication operators. Dokl Akad Nauk
actions. Proc Lond Math Soc 110(3):673–694 SSSR 259(3):526–529
Szabó G, Wu J, Zacharias J (2019) Rokhlin dimension for Vershik AM (1982) A theorem on Markov periodic
actions of residually finite groups. Ergodic Theory approximation in ergodic theory. Zap Nauchn Sem
Dynam Syst 39(8):2248–2304 Leningrad Otdel Mat Inst Steklov (LOMI) 115:72–82.
Tikuisis A, White S, Winter W (2017) Quasidiagonality of 306. Boundary value problems of mathematical phys-
nuclear C
-algebras. Ann Math 185(1):229–284 ics and related questions in the theory of functions, 14
Tomiyama J (1996) Topological full groups and structure Villadsen J (1998) Simple C
-algebras with perforation.
of normalizers in transformation group C
-algebras. J Funct Anal 154(1):110–116
Pacific J Undergrad Math 173(2):571–583 von Neumann J (1932) Proof of the quasi-ergodic hypoth-
Toms AS (2006) Flat dimension growth for C
-algebras. esis. Proc Natl Acad Sci 18(1):70–82
J Funct Anal 238(2):678–708 Winter W (2012) Nuclear dimension and Z-stablity of pure
Toms AS, Winter W (2013) Minimal dynamics and C
-algebras. Invent Math 187(2):259–342
K-theoretic rigidity: Elliott’s conjecture. Geom Funct Winter W, Zacharias J (2010) The nuclear dimension of C
-
Anal 23(1):467–481 algebras. Adv Math 224(2):461–498
 Descriptive complexity This term is used for the
The Complexity and the placement of a structure or a classification prob-
Structure and Classification lem among the benchmarks of reducibility.
of Dynamical Systems Distal Topologically distal is definition 20, mea-
sure distal is definition 27.
Matthew Foreman =+ equivalence relation Introduced and
University of California at Irvine, Irvine, CA, discussed in section “=+ and the Friedman-
USA Stanley jump operator.”
E0 equivalence relation Definition 32.
Factor map: measurable Let X, Y be standard
Article Outline measure spaces and T : X ! X, S : Y ! Y be
measure-preserving transformations. Then S is
Glossary a factor of T if there is a (not necessarily invert-
Introduction: What Is a Dynamical System? What ible) measure-preserving π : X ! Y such that
Is Structure? What Is a Classification? π ∘ T = S ∘ π almost everywhere. The map π is
Examples of Structure and Classification Results the factor map.
in Dynamical Systems Factor map: topological Let X, Y be topological
Descriptive Complexity spaces and T : X ! X, S : Y ! Y be homeo-
Complexity in Structure Theory morphisms. Then S is a factor of T if there is a
Complexity in Classification Theory continuous surjective map π : X ! Y such that
Standard Mathematical Objects in Each Region π ∘ T = S ∘ π. The map π is the factor map.
Placing Dynamical Systems in Each Region Ill-founded, Well-founded If T
X < ℕ is a tree,
Bibliography then T is well-founded if and only if there is no
function f : ℕ ! X such that for all n, f {0,1,2,

+
Glossary . . .n}  T. Equivalently, T has no infinite
paths. If T is not well-founded, then T is ill-
Analytic set A subset A of a Polish space X is founded.
analytic if there is a Polish space Y and a Borel K -automorphisms Definition 10.
set B
X  Y such that A = {x: for some y  Y, Kakutani equivalence Definition 60.
(x, y)  B}. Equivalently A is the projection of Minimal homeomorphism A homeomorphism
B to the X-axis. h : X ! X is minimal if for every x  X,
Anosov diffeomorphism Definition 13. {hn(x) : n  ℕ} is dense.
Benchmarking An informal term describing the Measure conjugacy Let S, T be measure-pre-
location of a set or an equivalence relation in serving transformations defined on standard
terms of reducibility to other sets or equiva- measure spaces (X, ℬ, m) and (Y, C , n). Then
lence relations. S, T are measure conjugate if there is an invert-
Bernoulli shift Definition 7. ible measure-preserving transformations
Borel hierarchy Definition 70 f : X ! Y such that f ∘ T = S ∘ f almost
Co-analytic set A subset C of a Polish space X is everywhere.
co-analytic if there is a Polish space Y and a Morse-Smale diffeomorphism Definition 14.
Borel set B
X  Y such that C = {x for all P11 -norms Definition 75.
y  Y, (x, y) B. Equivalently, the comple- f(m) Let X, Y be Polish spaces, m a measure on
ment of C is analytic. X and f : X ! Y be a measurable map. Then f

© Springer Science+Business Media, LLC, part of Springer Nature 2023 529

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_726
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © This is a U.S. Government work and not under
copyright protection in the US; foreign copyright protection may apply 2022
https://doi.org/10.1007/978-3-642-27737-5_726-1
530 The Complexity and the Structure and Classification of Dynamical Systems

and m induce a measure f(m) on the Borel well-ordering  of {s  T : s  t} such for s1,
subsets of Y by setting f(m)(B) = m(f1(B). s2  t, s2Ts1 if and only if s1s2. A path
Polish space A Polish space is a topological through a tree is a one-to-one function p from an
space (X, t) such that there is some complete ordinal α whose range is t-upwards closed and
separable metric d on X inducing the topology if β < γ < α, p(γ)T p(β). A branch through T is
t. The topology t is called a Polish Topology. a maximal path through T. A tree is well-
Polish group action A Polish group is a topo- founded if and only if it has no infinite paths.
logical group with a Polish topology. A Polish Trees of finite sequences Fix a set X. Let X<ℕ be
group action is a Polish group acting G acting the collection of finite sequences of elements of
jointly continuously on a Polish space X. See X. Order X <ℕ by setting s  t if and only if s is
section “Polish Group Actions.” an initial segment of t. A set T
X <ℕ is a tree
Reduction If A and B are subsets of Polish of finite sequences of elements of X if and only
spaces X and Y, then a one dimensional reduc- if whenever s  T and t is an initial segment of
tion of A to B is a function f : X ! Y such that s then t  T. Trees of finite sequences can be
for all x, x  A if and only if f(x)  B. If <ℕ
viewed as elements of f0, 1gX : Putting the
A
X  X and B
Y  Y, then a two dimen- discrete topology on {0, 1} and the product
sional reduction is a function f : X ! Y such <ℕ
topology on f0, 1gX , the space of trees is a
that for all (x1, x2)  X  X, (x1, x2)  A if and compact topological space. If X is countable
only if (f(x1), f(x2))  B. The reduction is then the space of trees on X is homeomorphic
continuous or Borel if the function is continu- to the Cantor set. Trees are discussed in section
ous of Borel. One- and two-dimensional Borel “Reducing Ill-Founded Trees.”
reductions are notated as ≼1ℬ and ≼2ℬ : (section Topologically transitive If X is a metric space, a
“What Is a Reduction?”). homeomorphism h : X ! X is topologically
Rotation number Definition 2 transitive if for some x  X, {hn(x) : n  ℕ} is
Separable and complete measure spaces A dense.
measure space (X, ℬ, m) is separable if there Turbulent Turbulence is a property of a contin-
is a countable subset A
ℬ such that for all uous action of a Polish group on a Polish space.
S  ℬ and ϵ > 0, there is an A  A such that It gives a method for showing that an equiva-
m(AΔS) < ϵ. A measure space is complete if lence relation is not the result of an S1-action.
whenever A  ℬ has m(A) = 0 and B
A, then It is discussed in section “Turbulence” where
B  ℬ. the notion is formally defined.
Smooth equivalence relation Definition 31. Well-founded, Ill-founded If T
X<ℕ is a tree,
Smooth manifold A manifold with a Ck struc- then T is well-founded if and only if there is no
ture for some k  1. function f : ℕ ! X such that for all n, f {0,1,2,
+

Smooth transformation A Ckmap f : M ! M . . .n}  T. Equivalently, if T has no infinite

where M has a Ck-structure for some k  1. paths, T is not well-founded, then T is ill-
Standard measure space A measure space founded.
(X, ℬ, m) is standard if it is separable and
complete.
The group S1 The group S1 is the group of Introduction: What Is a Dynamical
permutations of the natural numbers (section System? What Is Structure? What Is a
“S1- Actions”). Classification?
Topological conjugacy Let X, Y be topological
spaces and f : X ! X, g : Y ! Y. Then f and g are This article is a survey of the complexity of struc-
topologically conjugate if there is a homeo- ture, anti-structure, classification, and anti-
morphism f : X ! Y such that f ∘ f = g ∘ f. classification results in dynamical systems. The
Trees A (downward branching) tree is a partial entry focusses primarily on ergodic theory, with
ordering (T, T) such that for all t  T, there is a excursions into topological dynamical systems.
The Complexity and the Structure and Classification of Dynamical Systems 531

Part of the intention is to suggest methods and The upshot is that in any given paragraph in the
problems in related areas. It is not meant to be a entry, one of the audiences will be questioning the
comprehensive study in any sense and makes no point of what is being presented. I ask the reader’s
attempt to place results in a historical context. patience in understanding the dual mission.
Rather it is intended to suggest classification and What Is the Difference Between a Structure
anti-classification results that are likely to be com- Theory and a Classification? The setting for all
mon in many areas of dynamical systems. Hence, of these results is a Polish space X, and we study
the focus is on giving particular examples where classes C
X. In some cases C ¼ X, in which
descriptive set theory techniques have been suc- case the structure question is less cogent.
cessful, and mention related work in passing to The distinction between structure and classifica-
suggest that the examples given here are not iso- tion is a bit vague. Very roughly, a structure theo-
lated (Apologies for leaving out important results, rem for C gives criteria for membership of elements
this reflects my ignorance and space limitations.). that belong to X to be in C . The structure theorem
The target audience for this entry is twofold. In gives more information than the definition of C .
no particular order, one audience consists of Often this additional information simply gives
researchers whose primary interest is in some a very explicit test for belonging to C that may use
area of dynamics who may not be aware of, or ideas slightly different than the definition.
who want to learn more about the methods for Another form of a structure theorem is a method
studying the complexity of structure and classifi- for building an element of C. The form of this type
cation offered by descriptive set theory. For these of classification is:
people, the attempt is to give familiar examples
from dynamical systems where the complexity is xC
understood relative to known benchmarks. Logi- if and only if
cal technicalities are avoided as much as possible,
x can be built using a construction with properties 1Þ-nÞ
and the background necessary is given in the
appendix. This is naive descriptive set theory – where properties 1-n) are concrete and explicit.
with the rough meaning that it omits the techni- While a structure theory involves one object at
calities of quantifiers. A self-contained source on a time and answers the question whether that
basic descriptive set theory, requiring very mini- object is in C
X, a classification theory involves
mal background is Foreman (2010). There are two objects known to be in C and asks whether
many excellent, more advanced sources for this they are equivalent with respect to an equivalence
material (Kechris 1995; Marker 2002; relation in question. It is usually concerned with
Moschovakis 2009). The author makes no attempt assigning invariants to the equivalence classes.
at historical attributions of the theorems in Examples of common equivalence relations
descriptive set theory proper, these are well- can include being in the same orbit by a group
covered in Kechris (1995); Moschovakis (2009). action, or being isomorphic, or have some other
The second audience are those descriptive set property in common.
theorists who are likely to be very familiar with This entry is concerned with two cogent ques-
the language of reductions and quantifier com- tions. The first is
plexity. For these people, the attempt is to give a
Can the structure or classification theory be done
very high-level overview of some of the known
with inherently countable information (is it Borel)?
results specifically in dynamical systems. The
choices of examples reflect the authors taste and The second question is:
background, but include extremely famous results Where does its complexity sit relative to existing
and the problems left open by them. There is a list benchmarks? For example: Are there numerical
of open problems at the end of the entry and more invariants? Can one assign countable structures to
elements of the class so that isomorphism is the
will appear in a future arXiv submission with
invariant?
several co-authors, particular F. G. Ramos.
532 The Complexity and the Structure and Classification of Dynamical Systems

Many classical structure theorems were proved in successes with the groups ℤ and ℝ, so that will
the 1960s and 1970s. They fit well into the rubric be the focus in this entry. In general the theory gets
suggested by this entry. progressively more difficult as the groups move
What is anti-classification? Reductions are from ℤ to ℤd to general amenable groups, to free
tools that allow the establishment of lower bounds groups, and then to general non-amenable groups.
on the complexity of equivalence relations. These Since part of the story being told here is that
lower bounds are often the established bench- classifications can be “impossible,” showing this
marks discussed in this entry. The complexity of in the simplest, most concrete situation illustrates
some of some benchmarks is extremely high– the point most dramatically.
often they are not even Borel. The roots of the theory of much of dynamical
Why is this important? An underlying thesis of systems can be traced to the study of vector fields
this entry is that a general solution to a question on smooth manifolds, naturally linked to smooth
about a Polish space that is not Borel cannot be ℝ-actions. (See Smale 1963). As argued by Smale
solved with inherently countable information in (1967) these actions have smooth cross sections
(This is not original to this paper). If an equiva- that give significant information about the
lence relation is not Borel a general question corresponding solutions to ordinary differential
whether x1  x2 requires some uncountable equations. The smooth cross-sections are ℤ-
resource – usually an application of the actions by diffeomorphisms of the manifold. In
uncountable Axiom of Choice. Thus, a general short, ℝ-actions give rise to interesting ℤ-actions.
theme of this entry consists of determining when In ergodic theory, a class of ℤ-actions are induced
a question can be answered using inherently by ℝ-actions using the method of first returns.
countable information: it focusses on the Borel/ This is studied via Kakutani equivalence and
non-Borel distinction. discussed in section “Kakutani Equivalence.”
This language can be confusing: for example, Turning this around, ℤ-actions can induce ℝ-
if an equivalence relation E can be “classified by actions using the method of suspensions. This
countable structures,” it sounds like it is “classifi- technique allows lifting the complexity results
able.” However even for countable structures, the from ℤ-actions to ℝ-actions. The upshot is that
isomorphism relation may not be Borel. An exam- ℤ and ℝ-actions are closely related. For this rea-
ple of this is the equivalence relation of isomor- son Smale (1967) and others argue for studying ℤ-
phism for countable groups (See section “S1- actions. Since the ℤ-actions are determined by
Actions”). their generator, this amounts to studying single
Dynamical Systems Broadly speaking a transformations.
dynamical system is any group action f : G  In summary, this entry will focus on ℤ-actions,
X ! X. However, to be interesting, the group which are determined by the generating transfor-
actions are taken to preserve some structure on mation – in effect we are studying single
X. Commonly studied types of structure include transformations.
topological, measure theoretic, smooth, and What structure and equivalence relations
complex. will be considered? While we will give examples
There is a deep and well-developed theory for of difficulties with structure theory in other con-
general groups, both amenable and non- texts, the structure and classification theory of
amenable, discrete, topological, and carrying a transformations described in this entry breaks
differentiable or complex structure. Accordingly, very roughly into the quantitative theory and the
the actions can be discrete, continuous, or smooth qualitative theory. The ergodic theorem gives a
actions. framework for studying a function by repeated
Typical groups include ℤn, ℝn, more general sampling along an orbit of a transformation T.
Lie groups, free groups n , and groups arising as As the number of samples grow, the averages con-
automorphism groups of natural structures. How- verge to the average of the function over the whole
ever, much of the theory is modeled on initial space provided the given transformation is
The Complexity and the Structure and Classification of Dynamical Systems 533

ergodic. Hence, it can be viewed as the quantitative diffeomorphisms, for example, this is the motiva-
theory. The appropriate equivalence relation is tion for focusing on single transformations, taking
conjugacy by measure-preserving transformations. the manifold to be as simple as possible (say the
In (Smale 1967), Smale describes the study of 2-torus) and the diffeomorphisms to be C1. In
transformations (in particular diffeomorphisms of many situations, the results are more general, but
manifolds) up to topological conjugacy as the that is not the goal.
qualitative theory. Symbolic shifts are also stud- Homeomorphisms Often the class being
ied up to homeomorphisms (and up to their auto- given a structure theory consists of homeomor-
morphism groups). phisms with additional properties. In addition to
The understanding of the complexity of classi- smooth structure, one can consider minimal
fications in the quantitative theory is better devel- homeomorphisms, or topologically transitive
oped than the understanding of the complexity of homeomorphisms. An important example in sec-
classifications in the qualitative theory. Ergodic tion “Natural Classes That Are Not Borel” is the
theory classifications are discussed in section collection of topologically distal transformations.
“Measure Isomorphism.” The nascent connec- We define the first two classes as they arise in
tions with the qualitative theory is discussed in many contexts. Let X be a topological space and
section “Topological Conjugacy”, along with the h : X ! X be a homeomorphism. Then h is min-
many open problems. imal if every forward h-orbit of an x  X is dense.
To repeat, the two equivalence relations this A weaker property is topological transitivity,
survey will focus on are: which means that for some x  X the forward
h-orbit of x is dense.
• Measure conjugacy What a Classification isn’t. To make sense of
• Topological conjugacy what a classification is, it might best be motivated
by giving an example what a classification isn’t.
The first equivalence relation involves transfor- For concreteness, let us consider the space of
mations that act on probability measure spaces X, Y, measure-preserving transformations of the unit
perhaps with other structural restrictions. (For interval with the equivalence relation of measure
example, volume preserving diffeomorphisms of isomorphism (conjugacy).
a compact manifold.) Two such transformations Because there are only continuum many
S, T are measure conjugate (or measure isomor- measure-preserving transformations, there are at
phic) if there is a measure isomorphism f : X ! Y most continuum many equivalence classes [T]. By
such that S ∘ f ¼ f ∘ T almost everywhere. the Axiom of Choice, one can build a one-to-one
The second equivalence relation involves function
homeomorphisms acting on topological spaces
X, Y, perhaps with other structural restrictions F : f½T : T  MPTg ! ℝ
(S and T might be required to preserve smooth
structures on X and Y.) A pair S, T are topologically where MPT is the collection of ergodic
conjugate (or topologically equivalent) if there is measure preserving transformations. By letting
a homeomorphism h : X ! Y such that S ∘ h ¼ C(T ) ¼ F([T]), we get a map giving complete
h ∘ T. numerical invariants to the equivalence relation
Choices of examples A theme of the survey is of measure isomorphism.
that many natural questions have intractable com- This is of course NOT what is meant by a
plexity, such as being infeasible using inherently classification, since simply evoking the Axiom of
countable information. For this to be most con- Choice gives no useful information whatsoever.
vincing, the classes should be taken to be as clear Requirements to be a structure theory or a
and concrete as possible in each context. In the classification A structure theory or a classification
context of the qualitative behavior of must be effective or computable in some sense.
534 The Complexity and the Structure and Classification of Dynamical Systems

There are at least three common versions of Examples of Structure and Classification
effectiveness: Results in Dynamical Systems

• Computable feasible algorithms To try to clarify the notions we begin by giving

• Computable using inherently finite techniques examples from dynamical systems (Much of this
• “Computable” using inherently countable section is modeled on the introductory chapter of
techniques K. Peterson’s book (Peterson 1983).).

We will say nothing about the first notion. Hamiltonian Dynamics

Structure and Classification results that use inher- These flows were developed by Hamilton
ently finite techniques are called computable or to describe the evolution of physical systems (See
recursive classifications. This topic is well- Wikipedia Hamiltonian system). A Hamiltonian
covered in other literature (Braverman and system is described by a twice differentiable (C2)
Yampolsky 2009; Weihrauch 2000), and we are function H(q, p) from ℝ6N to ℝ, giving the energy
content with giving a couple of examples. Struc- of the system. The system is described by
ture and Classification that uses inherently count- Hamilton’s Equations:
able techniques are Borel. Techniques to show a
classification is Borel, or not Borel, are the main dp @H
¼ ,
topic of this entry. dt @q
We note that showing a class does not have a dq @H
¼þ :
Borel structure or does not have a Borel classifi- dt @p
cation is a stronger result than showing that there
is no such computable theory. For this reason, this The variables p, q  ℝ3N are interpreted as the
entry focusses on Borel classifications. There is generalized position and momentum variables,
much earlier work showing that various kinds of and the solution r(t) is viewed as the trajectory
classifications are impossible using finitary tech- of a point in an initial position r(0)  ℝ6N. Putting
niques. These appear in places such as Mitin this the context of group actions, the solution is a
(1998); Ryzhikov (1985). one-parameter flow
Benchmarking The complexity of the
existing classifications can be measured by T : ℝ ! ℝ6N
comparison with benchmark equivalence rela-
tions coming from outside dynamical systems. What makes ergodic theory relevant to this
This benchmarking process can also give a rig- example is Liouville’s Theorem that the Hamilto-
orous method of saying that one classification is nian system {Tt}t preserves Lebesgue measure.
more complex than a second one. This For many fixed energies E the surface H ¼ E is a
benchmarking is called the theory of analytic compact manifold and thus, on this manifold, the
equivalence relations and is discussed at length flow {Tt}t preserves an invariant probability mea-
in section “Standard Mathematical Objects in sure (Khinchin 1949).
Each Region.” A very brief introduction can The Birkhoff ergodic theorem suggests the
also be found in Foreman (2018) (several of appropriate equivalence relation for studying sys-
the diagrams in this entry were motivated by tems with invariant measures:
very similar diagrams in Foreman (2018)).
There is an extensive literature about this Theorem 1. Suppose that (X, ℬ, m, T ) is a
topic, many of the results can be found in notes measure-preserving system and f  L1(m). Then
and books such as Gao (n.d.); Gao (2009); the averages N1 N1
0 f ðT n ðxÞÞ converge almost
Hjorth (2000); Kechris (1995). everywhere to a T-invariant function f with
The Complexity and the Structure and Classification of Dynamical Systems 535

• r( f ) exists and is independent of x,

f dm ¼ f dm:
• If r is defined by Eq. 1 using a different lift F
of f, then for all x, [r( f )]1 ¼ [r( f )]1.
In particular, if T is ergodic then the averages
converge to the constant value f dm. Granting these facts, the rotation number is a
well-defined member of [0, 1) depending only on
An informal interpretation of this theorem is f. We will denote this number by r( f ) rather than
that if one repeatedly samples an ergodic station- [r( f )]1.
ary system and then averages the results, the limit The equivalence relation on homeomorphisms
is the mean of the system. This suggests that the of the circle is conjugacy by a homeomorphism.
natural equivalence relation is conjugacy by a So f  g if there is a homeomorphism h such that
measure-preserving transformation, because this
equivalence relation takes repeated sampling h ∘ f ¼ g ∘ h:
along trajectories to repeated sampling along tra-
jectories. In different terminology, because mea- If f is an arbitrary homeomorphism of the cir-
sure isomorphism take averages to averages, it is cle, then it is conjugate to a rotation preserving
the natural equivalence relation to study in this homeomorphism, so we focus on those.
context. One can check that:

Rotations of the Unit Circle Proposition 3. If f, g are orientation preserving

In this example, the class C is the collection of homeomorphisms of S1 that are conjugate by an
orientation preserving homeomorphisms of the orientation preserving homeomorphism, then
unit circle. The structure is evident (as C itself r( f ) ¼ r(g).
forms the relevant Polish space X).
Let S1 be the unit circle. It is convenient to We will refer to statements like Proposition 3 as
view S1 as the unit interval [0, 1] with 0 and saying that r( f ) is a numerical invariant for the
1 identified. (An equivalent approach is to view equivalence relation of orientation preserving
S1 as [0, 2π] with the endpoints identified). Let topological conjugacy. Indeed having numerical
π : ℝ ! S1 be the map x 7! [x]1 where [x]1 is the invariants is one of the key notions of classifiability.
positive fractional part of x. The classification question for r can then be
If f : S1 ! S1 is an orientation preserving phrased as:
homeomorphism, then a lift of f is an increasing If r( f ) a complete invariant?
function F : ℝ ! ℝ where [F(x)] ¼ h([x]). Using Explicitly: can inequivalent f and g have
the notion of lift we can define the rotation num- r( f ) ¼ r(g)?
ber of f (This discussion is largely based on Clearly the rotation of the circle by 2πθ
(Barreira and Valls 2013), where the reader can radians, Rθ, has rotation number θ. So, the ques-
see details). tion of whether the rotation number is a complete
invariant can be restated as asking whether
Definition 2. Let
r( f ) ¼ θ implies that f is conjugate to Rθ.
The following results show that r( f ) IS a com-
Fn ðxÞ  x
rðf Þ ¼ lim : ð1Þ plete numerical invariant for functions with dense
n!1 n
orbits. The first is due to Poincaré.

Then [r( f )]1 is called the rotation number of f Theorem 4. (Poincaré) Suppose that f is a
homeomorphism of S1 with a dense orbit. Then
The following facts are necessary for this to f is topologically conjugate to Rθ for an
make sense: irrational θ.
536 The Complexity and the Structure and Classification of Dynamical Systems

Denjoy showed that, if f  C2, then having an is a homeomorphism. A closed, shift-invariant

irrational rotation number implies having a dense 
Sℤ is called a subshift.
orbit, however this is not true for general homeo- The structure and classification theories for
morphisms (See (Wikipedia Rotation number)). symbolic shifts splits naturally into topological
Thus, homeomorphisms of the circle with a classifications and, when the symbolic shift has a
dense orbit is a story of a successful classification specific measure associated with it, the classifica-
of the strongest kind: it is a classification by a tion up to measure isomorphism. The latter is
numerical invariant. Moreover, Eq. 1 shows that discussed in the next section.
with a sufficiently computable f, the numerical While many classes of symbolic shifts are
invariant can itself be computed and that the rota- studied, perhaps the most common are those shifts
tion number is a continuous function from X to in a finite alphabet. Among the classes studied is
[0, 1). (See Katok and Hasselblatt 1995). the collection of Shifts of Finite Type. This is the
The next theorem gives information about the class of subshifts that are determined by a fixed
situation when f has a rational rotation number: finite collection of words w1, w2, . . .wk  S<ℕ.
Given this collection the subshift consists of all
f  Sℤ such that for no interval is f [n, n þ m)

+
Theorem 5. Let f be an orientation preserving
homeomorphism from S1 to S1. Then: equal to some wi.
It is straightforward to verify that the collection
1. f has a rational rotation number if and only if f of subshifts of finite type is a countable subset of
has a periodic point. the compact subsets of Sℤ. Hence the structure
2. If r( f ) ¼ p/q and ( p, q) ¼ 1 then f has periodic aspect is easy to understand.
points and each periodic point has period q. Each shift of finite type is determined by finite
information (the forbidden words) and hence it
For homeomorphisms with period points the makes sense to ask whether it is possible to deter-
issue relates to the classification of order preserv- mine whether two subshifts are topologically con-
ing homeomorphisms of the unit interval, which is jugate using inherently finite information. Various
studied in Hjorth (2000). invariants exist which are derivable from adja-
After this entry was written Joint work of cency matrices and two subshifts of finite type
the author and Gorodetski shows that even for are conjugate by a homeomorphism if and only
diffeomorphisms of S1, the equivalence relation if their adjacency matrices are strong shift equiv-
of conjugacy-by-homeomorphisms turns out to alent. This is an example of a recursive reduction.
be a maximal among equivalence classes reduc- It reduces the equivalence relation of conjugacy
ible to S1-actions and thus, quite complex. The by homeomorphism to the equivalence relation on
complexity lies in diffeomorphisms with fixed the collection of adjacency matrices of being
points. strong shift invariant (See section “What Is a
Reduction?”).
Symbolic Shifts However, the question of whether there is an
Let S be a finite or countable alphabet (Several algorithm to determine conjugacy by homeomor-
authors refer to an alphabet as a “language,” a phisms for shifts of finite type remains an open
practice dating to Tarski). We let Sℤ stand for problem (see Open Problem 2).
the collection of bi-infinite countable sequences While other invariants exist for aperiodic sub-
hf(n) : n  ℤi written in the alphabet S. shifts such as topological entropy, these are not
The space Sℤ carries the natural product topol- complete invariants for conjugacy by homeomor-
ogy. With respect to this topology, the shift map phism. In Camerlo and Gao (2001), it is shown
sh : Sℤ ! Sℤ defined by: that the classification of homeomorphisms of the
Cantor set up to conjugacy has no Borel invari-
shðf ÞðnÞ ¼ f ðn þ 1Þ ants – in fact it is maximal for S1-actions (See
The Complexity and the Structure and Classification of Dynamical Systems 537

section “An Example of a Maximal S1-Action Two measure-preserving systems (X, ℬ, m, T )

from Dynamical Systems”). and ðY, C , n, SÞ are (depending on the author)
In fact, Clemens (2009) was able to show that isomorphic or conjugate if there is an invertible
there are no Borel computable complete numeri- measure-preserving transformation f : X ! Y
cal invariants for conjugacy of subshifts. Clemens whose range has n-measure one such that for
determined the exact Borel complexity of m-almost all x  X
conjugacy of subshifts.
S ∘ fðxÞ ¼ f ∘ T ðxÞ: ð2Þ
Measure Structure
There is an important and very well-developed If f is not invertible but still satisfies Eq. 2
study of transformations that preserve Borel mea- almost-everywhere, then f is a factor map and S is
sures that are not probability measures (Aaronson a factor of T.
1997), but to illustrate the connections with If M is a compact metrizable space and
descriptive set theory, we consider probability T : M ! M is a homeomorphism, then the Krylov-
measures. Bogoliubov Theorem shows that there is a
We will discuss two types of questions about T-invariant probability measure m on the Borel
the structure of measure-preserving transforma- subsets of M. The space PT of T-invariant mea-
tions: understanding a single transformation, and sures on M is convex and compact with the weak
trying to understand the structure of factors and topology. Hence, the Krein-Milman theorem says
isomorphisms. In section “Measure Isomor- that it is spanned by its extreme points (The inter-
phism”, we discuss the classification problem for ested reader can find explanatory discussions of
measure-preserving systems and smooth these topics in Walters (1982) and Mirzakhani and
measure-preserving systems. Kechris discusses Feng (2014).).
more general results for countable amenable A measure-preserving transformation is ergodic
groups in (Kechris 2010). if the only invariant measurable sets either have
A basic object in ergodic theory is (X, ℬ, m, T ) measure zero or measure one. The extreme points
where (X, ℬ, m) is a standard probability measure of the simplex of T-invariant measures on M are
space and T : X ! X is an invertible measure- exactly the measure for which T is ergodic. (We will
preserving transformation. Following Furstenberg call these the ergodic measures when T is clear
(1981), we will call these measure-preserving sys- from context.) The Ergodic Decomposition Theo-
tems (and T a measure-preserving transformation) rem says that if m is a T-invariant measure then m
and reserve that term for probability measure- can be written as a direct integral of the ergodic
preserving systems. (We will only be considering measures. For the statement of the theorem, we let
invertible measure preserving systems so we drop M(X) to be the space of invariant probability mea-
the adjective invertible.) sures on X with the weak topology.
While there are many models for the space of
measure-preserving transformations, one is the Theorem 6. Let T be a measure-preserving
collection of Lebesgue measure-preserving trans- transformation of a standard probability space
formations of the unit interval equipped with (X, ℬ, m). Then there is a measure space
Lebesgue measure. These are identified if they ðO, C , nÞ and a C  measurable map m : C !
agree almost everywhere, and have the weak MðXÞ such that:
topology (the Halmos topology), which carries a
complete separable metric (Halmos 1944, 1956). • For all w 6¼ w0, m(w) and m(w0) are mutually
For a standard probability space, we will call the singular and T-invariant.
group of measure-preserving transformations • For all Y  ℬ, m(Y) ¼ (m(w)(Y ))dn(w).
MPT(X). With the weak topology, it is a Polish • For almost all w  Ω, (X, ℬ, m(w), T ) is
group, with the operation of composition. ergodic.
538 The Complexity and the Structure and Classification of Dynamical Systems

Moreover, the measure n is unique up to sets of Theorem 8. (Feldman) The collection of

measure zero. measure-preserving transformations on the unit
interval that are isomorphic to a Bernoulli shift on
Thus, every invariant measure on M can be a finite alphabet is a Borel set.
written as a direct integral of ergodic measures.
We interpret this as saying that the ergodic mea- Being isomorphic to a Bernoulli shift is an
sures are the building blocks of the invariant mea- analytic condition, as we sketch in example 73.
sures and focus the classification problem on the Feldman showed that the collection of very weak
ergodic measures, or equivalently, on the ergodic Bernoulli transformations is co-analytic. The the-
measure-preserving transformations. orem of Suslin (Corollary 72) says that a collec-
Notation We use the notation EMPT(X) for the tion which is both analytic and co-analytic is
collection of ergodic measure-preserving transfor- Borel. Thus, the collection of Bernoulli shifts is
mations on (X, ℬ, m). Borel.
The property of being ergodic is a conjugacy Classification theory for Bernoulli shifts
invariant so it invariant under the conjugacy Kolmogorov and Sinai, building on the work of
action of MPT(X). Moreover, Halmos showed Shannon defined a notion of entropy for measure-
that the ergodic measure-preserving transforma- preserving transformations. Viewed as a function
tions are a dense G d set in the weak topology on defined on the ergodic measure-preserving trans-
MPT(X) (See, e.g., (Halmos 1956)). formations, EMPT, entropy is a lower-
semicontinuous. Kolmogorov showed that iso-
Rolling Dice: Bernoulli Shifts morphic Bernoulli shifts have the same entropy.
Two classes of measure-preserving systems pro- This sets the stage for the following result of
vide very nice examples of both structure and a Ornstein, which is exposited in (Ornstein 1970a,
classification. One is the full shift on a finite b, 1974):
alphabet with the Bernoulli measure. The other
is Haar measure on a compact group. The latter is Theorem 9. (Ornstein) Two Bernoulli shifts are
discussed in the next section. isomorphic if and only if they have the same
entropy.
Definition 7. Let S be a finite alphabet and n be a
measure on (S, P(S)). Let m be the product mea- Thus, the Bernoulli shifts are a paradigm of
sure nℤ. Then m is a shift-invariant measure on the both a successful structure and a successful clas-
full shift Sℤ. The system (Sℤ, ℬ, m, sh) is called a sification theory:
Bernoulli Shift.
1. (Structure Theory) Bernoulli shifts can be
Informally: the coordinates of the Bernoulli identified in a Borel manner (they are the
shift are independent, identically distributed finitely determined transformations)
(iid). In particular the “time-zero” behavior is 2. (Classification Theory) Entropy is a complete
independent of the past. invariant for isomorphism.
Structure theory for Bernoulli shifts There is
a successful structure theory for Bernoulli shifts. Among the various differences between ℤ-
Ornstein showed (Ornstein and Weiss 1974) actions and the actions of other groups is that the
showed that being finitely determined is equiva- question of whether Bernoulli shifts based on F2-
lent to being isomorphic to a Bernoulli shift and actions forms a Borel set among the F2-actions is
that being very weak Bernoulli implies being an open problem. (See Open Problem 3)
finitely determined. Ornstein and Weiss (1974) A notion related to Bernoulli shifts is that of a
proved the converse: being finitely determined Kolmogorov Automorphism, or for short, a K 
implies being very weak Bernoulli. Feldman automorphism: Informally: Bernoulli Shifts are
(1974) used this characterization to show: characterized by the present being independent
The Complexity and the Structure and Classification of Dynamical Systems 539

from one moment earlier. K automorphisms are Rokhlin (1967) proved that there is always a
characterized by the present being asymptotically countable generating set, and Krieger (1970,
independent of the past. 1972a) proved that if the entropy of T is finite,
then there is a finite generating set.
Definition 10. A K automorphism is a Krieger (1972b) showed the stronger result that
measure-preserving transformation T of a stan- if the original transformation T is ergodic and
dard probability space (X, ℬ, m) such that there K
A ℤ is the support of n, then (K, sh) is
is a sub-s-algebra K
ℬ with: uniquely ergodic. We have outlined the proof of
the following theorem
1. K
TK ,
2. the union of the algebras T n K , for n  0 Theorem 11. Let (X, ℬ, m) be an ergodic
generates ℬ as a s - algebra, measure-preserving system. Then there is a count-
n
3. \1n¼1 T K ¼ f;, X g: able alphabet A and a closed shift-invariant set

A ℤ carrying a unique shift-invariant mea-
There is a perfect set of non-isomorphic K  sure n such that ð, C , n, shÞ is isomorphic to
automorphisms, as was shown by Ornstein and (X, ℬ, m, T ). If (X, ℬ, m, T) has finite entropy
Shields in (1973). In section “Reducing E0 to then the alphabet A can be taken to be finite.
Dynamical Systems”, it is shown that the very
elegant classification theory for Bernoulli shifts Koopman Operators
does NOT extend to the K automorphisms and Let (X, ℬ, m, T ) be a measure-preserving system.
little is known about classifying the K  Define a map
automorphisms: For example, it is not known if
the measure-isomorphism relation restricted to the U T : L2 ðXÞ ! L2 ðXÞ
K automorphisms is Borel (See Open Problem 4.).
by setting UT ([ f ]) ¼ [ f ∘ T ]. Then UT is a well-
Symbolic Systems as Models for Measure- defined unitary operator on L2(X), and moreover,
Preserving Transformations if S and T are isomorphic measure-preserving
Let (X, ℬ, m) be a standard probability space, and transformations, then US and UT are unitarily
T : X ! X be an ergodic measure-preserving equivalent.
transformation. Then a set A
ℬ is a generating Thus, it is possible to assign to each measure-
set if ℬ is the smallest T-invariant s-algebra preserving transformation a unitary operator in a
containing A (where we identify sets if their way that sends conjugate transformations to uni-
symmetric difference is zero). tarily equivalent operators. These operators are
If A is a countable or finite generating set then called the Koopman Operators. The equivalence
we can make the elements of A disjoint and keep it classes of Koopman operators are thus a collec-
a generating set. For this reason, the terminology tion of invariants, but except for very special cases
generating partitions is frequently used. If A ¼ (such as in the next subsection) they are not com-
hAn : n  ℕi is a partition of a set of measure one, plete invariants.
then almost every x in X determines a ℤ-sequence
of names hAnk : k  ℕi where T k ðxÞ  Ank : The Translations on Compact Groups
sequence is called the A  name of x. If A gener- We describe the Halmos-von Neumann Theorem.
ates, then there is a set S of measure one such that Let G be a compact group. Then G carries a Haar
for all x 6¼ y  S the A  name of x is different probability measure that is invariant under left
from the A  name of y. multiplication. Suppose that g  G and Haar
Hence, on S there is a one-to-one map f from measure is ergodic for the map Tg : G ! G
X to A ℤ given by sending x to its A  name: given by Tg(h) ¼ gh. Then G must be the closure
Letting n ¼ f(m) (so copying the measure on of the powers of g : G is a monothetic abelian
X over to a measure on A ℤ), we get an isomorphic group (The material presented in this example is
copy of X as a shift space A ℤ , n : explained in (Foreman 2000)).
540 The Complexity and the Structure and Classification of Dynamical Systems

An ergodic transformation T : X ! X is said to presented with measures on different spaces can

have discrete spectrum if L2(U ) is spanned by the have different simplexes of invariant measures.
eigenfunctions of UT. The eigenfunctions can be Another example is Theorem 11 that shows that
taken to have constant modulus one and are closed every measure-preserving transformation can be
under multiplication. Since T is measure- presented as a symbolic shift.
preserving the eigenvalues have modulus one Striking examples include Bernoulli Shifts:
and form a subgroup of the unit circle. Hence, there are Bernoulli shifts that are presented as
by Pontryagin duality their dual is a compact real-analytic measure-preserving transformations
monothetic abelian group G. The identity map of the torus (with the Poulsen simplex of invariant
from the set of eigenvalues into the unit circle is measures). Bernoulli shifts can also be presented
a generator of this group. If g is the identity as uniquely ergodic symbolic systems. The issue
character and Tg is translation by g, then there is of presentation is quite different from the issue of
a unitary operator conjugating U T g with UT that isomorphism, although topological conjugacy is
preserves multiplication of bounded functions in quite sensitive to presentation.
L2(G). From this one can compute a measure Presentations can affect intrinsic measure the-
isomorphism between (G, Tg) and (X, T ). oretic properties. For example, Kushnirenko
Again by Pontryagin duality, if G is a compact showed that a diffeomorphism of a compact man-
group, then L2(G) is spanned by the characters of ifold preserving a smooth volume element has
G. An immediate calculation shows that if G is finite entropy. One of the most prominent open
monothetic, with generator g and T is translation problems in the area is the converse: can every
by g, then the characters of G are eigenvalues – finite entropy measure-preserving transformation
hence T has discrete spectrum. be presented as a C1, transformation on a com-
The Halmos-von Neumann theorem (Halmos pact manifold that preserves a volume form? (See
and von Neumann 1942) gives both a structure Open Problem 8.)
theorem and a classification theorem.
Smooth Dynamics
Theorem 12. (Halmos-von Neumann) Let T be In the context of this entry, smooth dynamics will
an ergodic transformation. Then: mean studying diffeomorphisms of compact man-
ifolds. The equivalence relation is that of
conjugacy by homeomorphisms. We repeat the
1. (Structure) T has discrete spectrum if and only
definitions from the introduction.
if it is isomorphic to an ergodic translation on
Explicitly, we let M be a smooth manifold,
a compact group with Haar measure.
usually taken to be compact. We will be given a
2. (Classification) If S, T have discrete spectrum,
class C of diffeomorphisms of M described with a
then S and T are conjugate by a measure-
structure theory. Two elements f , g  C will be
preserving transformation if and only if the
taken to be equivalent if there is a homeomor-
Koopman operators associated to S and
phism h : M ! M such that
T have the same eigenvalues.
f ∘ h ¼ h ∘ g: ð3Þ
For a proof of this theorem and more on this
topic see Walters (1982). Smale (1967) calls this equivalence relation the
qualitative theory of diffeomorphisms. He says
Presentations that “the same phenomena and problems of the
Often transformations with various relevant prop- qualitative theory of ordinary differential equa-
erties one wants to understand can be presented in tions [are] present in the simplest form in the
different contexts. The context in which they are diffeomorphism problem.” He argues that we
presented can be important. For example, two should study the equivalence relation given
isomorphic measure-preserving transformations, when the conjugacy map h is a homeomorphism,
The Complexity and the Structure and Classification of Dynamical Systems 541

rather than being a diffeomorphism, because “dif- 1. Pi is a hyperbolic periodic point for i ¼ 1, . . .l,
ferential conjugacy is too fine” to reveal the rele- 2. [li¼1 W s ðPi Þ ¼ M,
vant properties such as structural stability. 3. [li¼1 W u ðPi Þ ¼ M,
Some examples of classes of 4. Wu(Pi) and Ws(Pj) are transversal for all i, j.
diffeomorphisms There are many well-studied
classes of diffeomorphisms. For the purposes of
this entry, we will assume that they are C1 and 1. (Structure) Both the Anosov and the Morse-
usually assume that the manifolds M are compact. Smale diffeomorphisms are examples of struc-
We start with two of the well-known classes of turally stable diffeomorphisms. The structur-
diffeomorphisms. ally stable diffeomorphisms are those f for
Suppose f : M ! M is a diffeomorphism and which there is a C1-neighborhood U such that
that Λ
M is invariant. Then Λ is hyperbolic for every g  U is topologically conjugate to f.
f if there are constants C > 0 and 0 < l < 1 and a Thus
decomposition of the tangent bundle restricted to (a) The Anosov diffeomorphisms form an
Λ as: open set
(b) The Morse-Smale diffeomorphisms form
TM L ¼ Es Eu an open set
Tf ðE Þu
¼ E u
(c) The structurally stable diffeomorphisms
Tf ðEs Þ ¼ Es form an open set
It follows that there are only countably
and for v  Es, n > 0 many classes of Anosov and Morse-Smale
diffeomorphisms.
kTf n ðxÞvk  Cln k v k It would be a nice picture if the collection of
structurally stable diffeomorphisms formed an
and for v  Eu, n  0 open and dense subset of the diffeomorphisms,
but this was refuted by Newhouse (Newhouse
kTf n ðxÞvk  Cln k v k 1970, 1974).
2. (Classification) Each of the three classes form
Definition 13. A diffeomorphism f : M ! M is an open subset of the space of
Anosov if the whole manifold M is hyperbolic diffeomorphisms. Each of the open sets
for f. decompose into a union of open subsets
corresponding to each of the equivalence clas-
Suppose that x is a periodic point of order ses. Thus for each class, there is a single count-
m and x, f(x), . . . f m1(x) are hyperbolic points. able list of representatives and radii that
Then the stable and unstable manifolds of x are: capture all of the members of the class. The
representatives can be taken to be rational
Ws ¼ y  Mjf mk ðyÞ ! x as k ! 1 (in the appropriate sense) and the radii can be
take to be of the form 1/n. Hence the countable
Wu ¼ y  Mjf mk ðyÞ ! x as k ! 1
lists can be viewed as members of a Polish
space.
These are immersed Euclidean spaces of
The upshot is that for each of the three
dimension corresponding to the number of eigen-
classes, one can assign complete numerical
values of the differential D f m(x) that are greater
invariants in a Borel (even continuous) way.
or less than one in absolute value.
The equivalence relation is Borel reducible to
the equality relation on a Polish space.
Definition 14. A diffeomorphism f : M ! M is
Morse-Smale if and only if there are a finite col- In dimension 2, more intuitive invariants can
lection of periodic orbits P1, . . . Pl such that: be assigned in a computable way. For Anosov
542 The Complexity and the Structure and Classification of Dynamical Systems

diffeomorphisms on the 2-torus these are hyper- A structure theorem is concerned with a single
bolic elements of SL2(ℤ) and for Morse-Smale transformation, and classification theory is
diffeomorphisms, they are graphs with colored concerned with pairs of transformations.
edges (Oshemkov and Sharko 1998; Peixoto
1973). Similar results hold for Morse-Smale
diffeomorphisms in dimension 3 (Bonatti Descriptive Complexity
et al. 2019).
An alert reader will notice a problem with both For a structure theory or classification theory to be
the structure and classification results just stated. useful, it is important that it be computable in
The program is to classify equivalence classes some sense. As the example given in section
of diffeomorphisms, and the definition of “Introduction: What Is a Dynamical System?
Anosov, Morse-Smale, and structurally stable What Is Structure? What Is a Classification?”
diffeomorphisms are not invariant under topolog- that used the Axiom of Choice illustrated, classi-
ical conjugacy. Ideally, given a diffeomorphism fications or structure theorems that are not in some
f that is topologically conjugate to an element of sense computable are likely to be meaningless.
one of these classes, there would be a way of The structure theoretical levels of complexity
constructing a g equivalent to f that actually are easier to describe than the complexity bench-
belongs to the class. This seems to be an open marks for classification, as it is a series of yes-no
problem (See Open Problem 9). questions. Recall from the discussion in the intro-
duction the three basic questions about how effec-
More Detailed Structure Theory tive a structure theory can be are:
So far the term “structure theory” has been used
solely for the purpose of determining membership 1. Is the structure theory computable with realis-
in a class. However, in several cases a structure tic resources assumptions?
theory gives more information – for example, 2. Can it be carried out with inherently finite
about the factor structure of a member of a class. information?
The Furstenberg structure theorem for topologi- 3. Can it be carried out with inherently countable
cally distal transformations and the Furstenberg- information?
Zimmer theory for measure-preserving transfor-
mations are examples of this. These are discussed We will ignore the first question and say little
in section “Natural Classes That Are Not Borel.” about the second. The main point of the entry is
that there are natural examples of dynamical sys-
Summary tems fail even the third question. The third ques-
The previous sections are intended to give a con- tion can be rephrased as asking whether the
text for structure theory and classification theory. structure theory shows that the class C is Borel
These are somewhat vague terms, but the inten- (Of course, once a structure theory has been
tion is the following: shown to be Borel, the question arise to determine
what level in the Borel hierarchy it lies in. There is
• Structure theory takes place in the context of an such a literature, but it is not addressed in this
ambient Polish space and gives a method for survey.).
determining membership in a class C : It can Remark. In this survey, we only consider
also give information about the factors of a equivalence relations that are analytic. These
transformation or explicit information about include all orbit equivalence relations of Polish
the way the transformation moves elements in groups. (See example 73.) Some natural equiva-
the space it acts on. lence relations are not analytic (such the equiva-
• Classification theory considers a class C and an lence relation on distal flows of having the same
equivalence relation E on that class. It is distal height). The theory is not nearly as well
concerned with determining when two ele- developed for these so they are not covered in
ments of C are E-equivalent. this survey.
The Complexity and the Structure and Classification of Dynamical Systems 543

What Is a Reduction? Indeed this has a long history: the

There are many jokes about mathematicians unsolvability of the word problem for finitely
where the punchline is that the way they solved presented groups was shown by reducing an
problem A was to “reduce it to a problem B they incomputable set to the word problem using a
already knew how to solve.” (A google search in computable function.
2018 gave over 115 million hits to the search The focus of this survey is the notion of Borel
“reduce it to a problem you can solve joke.”) Reducibility (This idea originated in the late 1980s
The idea behind the joke is that to solve a in the work of Friedman and Stanley
problem A you (somewhat constructively) reduce (1989) related to model theory and independently
it to a problem B you already know how to solve. Harrington, Kechris, and Louveau who used in it
One can state this mathematically in the the context of analysis.), with some very short
following way. excursions into continuous reducibility. To repeat
the official definition, we let X and Y be Polish
Reductions of sets spaces. A set
Let X and Y be Polish spaces and A
X, B
Y.
Then A is A
X is Borel reducible to B
Y
_________- reducible
to B provided that there is a function f : X ! Y if and only if there is a Borel measurable function
such that for all x  X f : X ! Y such that for all x  X

x  A if and only if f ðxÞ  B: x  A if and only if f ðxÞ  B:

Fill in the blank As the example given earlier Thus the inverse image of B under f is A.
showed, without some restrictions on the function Because the definition of Borel measurable
f this notion is not meaningful. If B and Y\B are implies that the inverse image of a Borel set is
non-empty then every set A X is reducible to B: Borel, it follows that
fix b0  B and b1  Y\B and define f(x) ¼ b0 if
x  A and f(x) ¼ b1 if x  X \A. Then f reduces if A is not Borel, then B is not Borel:
A to B. Thus, one must fill in the blank with some
condition on the concreteness of f. Notation: We write A≼1ℬ B if A is Borel reduc-
In the complexity theory world of computer ible to B. In this notation, the previous remarks
science, the blank is often filled in by some state- can be written for the record as:
ment “linearly,” “polynomially,” “exponentially,”
or “recursively.” In the context of this entry, we Remark 15. Let X, Y be Polish spaces. Let A
X
focus on Borel functions f. and B
Y, with B Borel. If A≼1ℬ B then A is Borel.
Starting in the 1950s, in the subject then known
as Recursion Theory (Soare 1987) and now called
Reductions of Equivalence Relations
Computability Theory, the joke was reversed:
For the classification problem a two dimensional
To show a problem B is not solvable, you find an version is more useful. Let X and Y be Polish
unsolvable problem A and reduce it to B by a
_________ function.
spaces and E
X  X, F
Y  Y be equivalence
relations. Then
If B were solvable, and A is reduced to B, A would
also be solvable, yielding a contradiction. To make E is Borel reducible to F
the definition complete, one has to fill in the blank,
usually from one of the three types of functions: if and only if there is a (unary) Borel function
feasibly computable, recursively computable or f : X ! Y such that for all x1, x2  X
Borel.
544 The Complexity and the Structure and Classification of Dynamical Systems

x1 Ex2 if and only if f ðx1 ÞFf ðx2 Þ: analogue to the Cantor-Schroeder-Bernstein the-
orem fails.) We use the notation AℬB (or C
We use the notation E≼2ℬ F: or...) to mean the equivalence relation of
It is important to recognize how the two defi- bi-reducibility: A≼ℬB and B≼ℬA.
nitions differ. In the second, two-dimensional def- Let ~ be the relation A≼ℬB and B≼ℬA. Then
inition, f has domain X, not X  X. So, in essence: ≼B gives a partial ordering of ~  classes of sets or
f assigns an F-equivalence class to each x in X in a
equivalence relations depending on the dimension
manner that two elements of X get assigned the one is working in.
same F-equivalence class if and only if they are in Caveat! There are several diagrams in this entry
the same E-equivalence class. that give the relationship of reducibility between
Viewing the classes of F as invariants, this is various classes of objects. The regions signify
paraphrasing of the statement that f computes the structures that are reducible to the type of object
invariants assigned to members of X. used to label the region. So, for example, an object
Thus for ≼2ℬ , f is a function that assigns values is in the region Polish group actions if it can be
to pairs (x1, x2) by considering each xi separately. reduced to a Polish group action. In particular, this
However viewed as a function from the Polish applies to Figs. 2, 4, 5, and 10. In the diagrams, a
space W ¼ (X  X) to the Polish space Z ¼ line segment going upwards means Borel Reduc-
(Y  Y ) it is also a one-dimensional reduction. ible and a pointed arrow means that the downwards
reduction does not hold. Not having an arrow
Remark 16. Suppose that E
X  X and F
Y  means that bi-reducibility is open. An arrow at
Y, then E≼2ℬ F implies E≼1ℬ F: both ends indicates that bi-reducibility does hold.
Universality Given a class C of either sets
(in the one-dimensional case) or equivalence rela-
Continuous Reductions
tions (in the two dimensional case) an object
An important special case of Borel reducibility is
B  C is universal for C if every A  C is reducible
when the function f is continuous. This is stronger
to it. The interpretation of this is that B is the most
than being Borel reducible. Essentially, every
complex element of C . This is used for all of the
general statement below applies to continuous
notions: Borel reductions, continuous reductions,
reductions as well as Borel reductions. If A is
and computable reductions.
continuously reducible to B we write A≼1C B and
Terminology warning: Maximal is often used
if E is continuously reducible to F we write
as a synonym of universal. Less frequently com-
E≼2C F: If continuity is clear from context we
plete is also used. We will use maximal when
omit the subscript C and if the number of variables
talking about the ≼2ℬ relation on equivalence
is clear from context we will omit the superscript
relations, the two dimensional case. We will use
in both cases.
complete for ≼1ℬ restricted to sets, the one-
The Pre-ordering dimensional case.
For both Borel and continuous reductions and Because of the transitivity of the notions of
both the one- and two-dimensional cases, the rela- reducibility, if B, C  C and B is universal for C
tion of ≼ℬ and ≼C is transitive. To see this we note and B≼ℬC, then C is universal for C . If follows
that if f : X ! Y is a reduction (in the appropriate that all universal sets are bi-reducible to each
sense) and g : Y ! Z is a reduction (in the same other.
sense), then g ∘ f is a reduction in the same sense. The upshot We can view the notions ≼ℬ and
As with all pre-orderings, we can have ≼C as measures of complexity. If A≼ℬB, then B is
bi-reducible sets and bi-reducible equivalence at least as complex as A. In a given class C there
relations. There are examples where A≼ℬB and may be ≼-maximal elements. They represent the
B≼ℬA but there is no bijection f : X ! Y such that equivalence class of the most complex elements
f reduces A to B, and f1 reduces B to A. (The of C .
The Complexity and the Structure and Classification of Dynamical Systems 545

Reducing Ill-Founded Trees branch. However, in this context the trees in

There is a basic method of establishing a set is not T rees are allowed to be infinitely branching.
Borel. Because it arises in the proofs of several Remark. In alternate language, not used in
theorems, we include it here rather than in the descriptive set theory, the space of trees is the
appendix A. It gives a canonical example of a space of rooted, labeled acyclic graphs
complete analytic set that is combinatorially con- (allowing countably infinite valence), and the col-
venient to work with. lection of ill-founded trees are those graphs with
The space of Trees For applications, a com- non-trivial ends.
mon complete analytic (and so nonBorel), set is Put the discrete topology on ℕ and consider ℕℕ
the collection of ill-founded trees. Let ℕ<ℕ be the with the product topology. This yields a Polish
collection of finite sequences of natural numbers, space called the Baire space. Then, B ¼
viewed as being of the form s ¼ hn0, n1, . . .nl1i. ðT , f Þ : f  ℕℕ , T  T rees and f is a branch
The number l is the length of s. The Polish space of through T g is a closed subset of T rees  ℕℕ .
trees is defined as follows: Because

1. A tree is a set T
ℕ<ℕ that is closed under fill  founded treesg

initial segments. So if s ¼ hn0 , n1 , . . . ¼ T : for some f  ℕℕ ðT , f Þ  B ,
nl1 i  T and k < l then t ¼ hn0 , n1 , . . .
nk1 i  T : (It is important to note that the it follows that the collection of ill-founded trees is
trees are not required to be finitely branching. an analytic subset of T rees:
Every tree has the empty sequence in it.) A tree is well-founded if it is not ill-founded.
2. The space T rees is the collection of trees Hence, the well-founded trees are the complement
T
ℕ<ℕ . of the ill-founded trees and thus co-analytic.
3. A basic open set in the topology on the space of Fundamental results about the collection of ill-
trees is given by the finite sequence s and founded trees include:
hsi ¼ fT : s  T g:
Theorem 18. Let T rees be the space of trees and
Recasting the definition, putting the discrete
IFT
T rees be the collection of ill-founded
topology on {0, 1}, and the product topology on
<ℕ trees. Then:
f0, 1gℕ gives a Polish space X homeomorphic to
the Cantor set. Each tree T can be identified with its 1. If X is an arbitrary Polish space and A
X is
characteristic function XT : ℕ<ℕ ! f0, 1g, and analytic then A is Borel reducible to IFT.
hence, an element of X. It is not difficult to verify 2. IFT is an analytic set that is not Borel.
that the induced topology on fwT : T  T reesg

X gives a space homeomorphic to T rees
Corollary 19. Let WF
T rees be the collection
The first condition in the definition of a tree
of well-founded trees. Then
gives a countable sequence of requirements
corresponding to open sets, showing that
1. If X is an arbitrary Polish space and C
X is
fX T : T  T reesg is a closed subset of X and
co-analytic, then C is Borel reducible to WF.
hence a Polish space.
2. WF is a co-analytic set that is not Borel.
Definition 17. Let T  T rees and f : ℕ ! ℕ. Then
f is an infinite branch through T if for all l  ℕ the In the language of section “What Is a Reduc-
sequence s ¼ h f ð0Þ, f ð1Þ, . . . f ðl 1Þi  T . tion?”, the collection of ill-founded trees, as a
subset of T rees, is a complete analytic set. It
The König infinity lemma implies that every also follows from Theorem 18 that there are
infinite, finitely branching tree has an infinite non-Borel analytic sets.
546 The Complexity and the Structure and Classification of Dynamical Systems

Basic Technique for showing a set is analytic or • Can you determine whether an arbitrary x  X
co-analytic but not Borel. The basic technique belongs to C ?
for showing that a set C is not Borel is reducing a • How complicated is a given method for build-
known nonBorel set to C . To use this effectively, ing elements of C ?
one needs a starting point. The sets IFT or WF
play this role for analytic and co-analytic sets. The complexity of the class The basic ques-
Let X be a Polish space and C
X be tion is what level of descriptive complexity the
analytic. Then to show that C is complete analytic, class lies in relative to the basic benchmarks,
and so not Borel, it suffices to build a reduction given above.
An early example: weak mixing vs strong
f : T rees ! X mixing The first example of the use of descriptive
set theory to solve a natural structure problem in
such that dynamical systems is due to Halmos and Rokhlin.
A prominent open problem in the 1940s was
T is ill-founded if and only if f ðT Þ  C : whether the property of a transformation being
weakly mixing implied the property of being
By Theorem 18, it follows that C is not Borel. strongly mixing.
Similarly the basic technique for showing that In 1944, in a paper called In general a
a co-analytic set C is complete co-analytic is to measure-preserving transformation is mixing
build a reduction (Halmos 1944), Halmos proved that the collection
of weakly mixing transformations is a dense G d
f : T rees ! X set in the weak topology. In a paper published in
1948 with a title translated as A general measure-
such that preserving transformation is not mixing (Rohlin
T is well-founded if and only if f ðT Þ  C . 1948), Rokhlin showed that the strong mixing
The most basic Benchmarks The first collec- transformations is meager (first category). The
tion of benchmarks is determining whether a combination of the two papers show that there
classification is computable, Borel or analytic- are weakly mixing transformations that are not
and-not-Borel. These are the basic levels of strongly mixing.
descriptive complexity. If an example is Borel, A proof that explicitly uses complexity shows
one can also ask what level it lives in in the Borel that the collection of strongly mixing transforma-
hierarchy. tions is not P02 , and hence cannot be the same
Paraphrasing this informally, the question class as the weakly mixing transformations.
is whether every classification can be done (In fact the class is P03 .) In particular, there is a
using inherently finite techniques, inherently weakly mixing transformation that is not strongly
countable techniques, or whether it is impossible mixing, as it is complete S02 .
to do the classification using inherently count- An explicit example of a weakly mixing trans-
able techniques. If a classification is complete formation that is not strongly mixing was given
analytic or co-analytic, then there is no possible using Gaussian systems was provided by
classification that is feasible with countable Maruyama in 1949 (Theorem 11 of (Maruyama
resources. 1949)). A very natural example of such a transfor-
mation is due to Chacon (1969).
Complexity in Structure Theory
Examples at Three Levels of Complexity
For our purposes, structure theories for classes We now give examples of each level beyond
C X are concerned with two questions: recursively computable.
The Complexity and the Structure and Classification of Dynamical Systems 547

Non-recursive Classes • {(X, T ): T is minimal and X is Polish}

Example. (Braverman, Yampolsky) (Braverman
and Yampolsky 2009) There is a computable com- Proofs of these results (and later calculations of
plex number s such that membership in the Julia an upper bound on the complexity of topologi-
set associated with cally distal transformations) were novel in the
early 1990s and appear in Beleznay and Foreman
f ðzÞ ¼ z2 þ s (1995) (Lemma 3.3). The second item is due to
Kechris.
is not computable. Borel Classes of diffeomeomorphisms We
have already seen that the collection of structur-
Borel Classes ally stable diffeomorphisms is Borel – in fact it is
Borel Classes of measure-preserving transfor- open as are the collections of Anosov and Morse-
mations As mentioned above, the weakly mixing Smale diffeomorphisms.
and strongly mixing classes of measure- To give a flavor of tools used for more compli-
preserving transformations of the unit interval cated classes, we use the following results in
are Borel. The multiple mixing counterparts are Foreman and Gorodetski (2022):
as well. Denote by Fix( f ), the set of fixed points of the
In the topological category, the collections of map f : M ! M, i.e.,
uniformly continuous maps from the open unit
interval to the open unit interval are easily seen to Fixðf Þ ¼ fx  Mjf ðxÞ ¼ xg,
be P02 in the space of continuous transformations.
Indeed most standard classes of transforma- and by Pern( f ) the set of periodic points of period
tions are Borel. As explained in the discussion n, i.e.,
above the collection of Bernoulli shifts is Borel.
However, proving this uses Suslin’s Theorem, Pern ðf Þ ¼ x  Mjf n ðxÞ ¼ x and for all m
which involves analytic sets.
Other well-known classes of measure-
< n, f m ðxÞ 6¼ x :
preserving transformations are Borel and the
proof uses a significant amount of work. For
example, the class of rank-one-rigid transforma- Proposition. Let M be a compact manifold.
tions (King 1986). For any r ¼ 0,1,2, . . ., 1 the following statements
Borel Classes of homeomorphisms of com- hold:
pact spaces A potential problem with this collec-
tion is that there are a variety of natural spaces in 1. The set U 0 ¼ ff  Diff r ðMÞ : Fixð f Þ ¼ ;g is
which to set the various questions. For example, open.
every compact ℤ-flow can be viewed as a compact 2. For each m  ℕ the set U m ¼
subset of the ℍℤ or as a compact subset of ℍ2 f f  Diff r ðMÞ :jFixð f Þj¼ mg is Borel.
(where ℍ is the Hilbert cube) or as an appropriate 3. For each m, n  ℕ the set U nm ¼
subset of ℕℕ. All of the known natural codings of f f  Diff r ðMÞ : jPern ð f Þj ¼ mg is Borel.
the collection of homeomorphisms of compact 4. For any A : ℕ ! ℕ the set
metric spaces agree on what collections of homeo-
morphisms are Borel. (This is discussed in
UA ¼ f  Diff r ðMÞ : jPern ð f Þj ¼ AðnÞ,
(Beleznay and Foreman 1995).)
Examples include:
n ¼ 0,1,2, . . .
• {(X, T): T is topologically transitive and X is
Polish} is Borel.
548 The Complexity and the Structure and Classification of Dynamical Systems

5. The set 1
Gð f Þ ¼ lim sup logjPern ð f Þj
n!1 n

ðA, f Þ : jPern ðf Þj ¼ AðnÞ, n ¼ 0,1,2, . . . , is Borel. In particular, the set of Artin-Mazur

diffeomorphisms (defined as { f : G( f ) < 1}) is
where A : ℕ!ℕ Borel.

in the space ℕℕ  Diff r(M) is Borel. Natural Classes that Are Not Borel
In the topological setting A classical collection,
We give the proof that appeared in Foreman due to Hilbert (see Zippin 2013) is defined as
and Gorodetski (2022) for items 1–3. The first follows.
item is immediate: since M is compact there is Definition 20. Let (X, d) be a metric space and
a minimum distance between x and f(x). The T : X ! X be a homeomorphism. Then (X, T ) is
set of diffeomorphisms whose minimum dis- (topologically) distal if for all distinct x, y  X,
tance is bigger than 1/n is open, hence the set infn  ℤ{d(T nx, T ny)} > 0.
of diffeomorphisms with no fixed points
is open. Clearly, isometries are examples of distal trans-
To see the second item: To see that U 1 is Borel, formations. Moreover, the distal transformations
let us choose a countable base for the topology on are closed under taking skew products with com-
M, {Vk}k  ℕ, such that pact metric groups. Using these two facts one can
generate a wide class of distal transformations.
diam V 1  diam V 2  diam V 3  . . . Fact. The collection of distal transformations is
a co-analytic set. This is exposited in example 74.
and diam Vk ! 0 as k ! 1. The set
Definition 21. Let (X, d, T) and (Y, d 0, S) be a
V k ¼ ff  Diff r ðMÞjf has no fixed points outside of V k g compact separable metric spaces and π : X ! Y
be a factor map. Then X is an isometric extension
is open, and hence the set U 1 ¼ of Y if there is a real valued function r(x1, x2)
\K1 [1 defined for all pairs (x1, x2) in the same fibre of
k¼K V k ∖U 0 is Borel.
X (i.e., whenever π(x1) ¼ π(x2)) and such that
To see that U 2 is Borel, notice that each of the
sets 1. r(x1, x2) is continuous as a function on the
subset of X  X defined by the condition
V k1 ,k2 ¼ ff  Diff r ðMÞj f has no fixed points outside of V k1 [ V k2 g:
π(x1) ¼ π(x2).
2. For each y  Y, r(x1, x2) defines a metric on
is open, and hence the fiber π1( y) under which this fiber is iso-
metric to a fixed homogeneous metric space.
U2 ¼ \ [ V k1 ,k2 ∖ðU 0 [ U 1 Þ 3. r(Tx1, Tx2) ¼ r(x1, x2) for all x1, x2 in the
K1k1 , k2 K same fibre of X over Y.

is Borel. It is clear how to continue for n fixed The next definition captures the process of
points by induction. iterating the isometric extensions transfinitely:
Example. Artin-Mazur Diffeomorphisms It
follows easily from the proposition that the function Definition 22. (X, T ) is a quasi-isometric exten-
giving the exponential rate of growth of the number sion of (Y, T) if (Y, T) is a factor of (X, T), and there
of periodic points G : Diff r(M) ! ℝ [ { 1} is an ordinal  and a factor (Xx, T) of (X, T) for
defined by each x   such that:
The Complexity and the Structure and Classification of Dynamical Systems 549

1. (X0, T ) ¼ (Y, T); (X, T) ¼ (X, T ). norm on the collection of distal flows. It follows
2. if x < x0 then (Xx, T ) is a factor of Xx0 , T . that for all countable ordinals  there is a distal
3. (Xxþ1, T) is an isometric extension of (Xx, T). flow of distal height  and that the collection of
4. if x is a limit ordinal then (Xx, T) is the inverse minimal distal homeomorphisms of distal height
limit of Xx0 , T : x0 < x : less than  is a Borel set.
In the setting of measure-preserving trans-
The Furstenberg Structure Theorem formations Separate work of Furstenberg and
(Furstenberg 1963) for topologically distal flows Zimmer (Furstenberg 1981; Zimmer 1976) pro-
says: duced analogues of the topological Furstenberg
Structure Theorem that hold for measure-
Theorem 23. (Furstenberg) Let (X, d, T) be a preserving transformations. We describe it here
minimal, topologically distal flow. Then (X, d, T) without giving details and with very slight
is a quasi-isometric extension of the trivial flow. inaccuracies.
The category we are working in consists of
It is natural to ask what ordinals arise as  in the measure-preserving transformations of a standard
Furstenberg Structure Theorem. Given a distal measure space, which we can take to be the unit
flow (X, d, T ) define the distal height of the flow interval with Lebesgue measure. By Halmos’ the-
as the least ordinal  such that (X, d, T ) is topo- orem, the ergodic diffeomorphisms form a dense
logically equivalent to a quasi-isometric extension G d set. Hence, the weak topology makes the
of the trivial flow represented as a tower of exten- ergodic transformations a Polish space. We take
sions of height . our ambient Polish space EMPT to be the set of
It was shown in Beleznay and Foreman (1995) ergodic measure-preserving transformations of
that the collection of distal heights of minimal the unit interval.
distal flows on compact metric spaces is exactly Let ðY, C , n, SÞ be an ergodic measure-
the collection of countable ordinals. The follow- preserving transformation, and G a compact
ing theorem (Beleznay and Foreman 1995) sum- group. Let H be a compact subgroup of G. Then
marizes these results: the quotient space G/H carries a left-invariant
Haar measure. A compact co-cycle extension
Theorem 24. (Beleznay-Foreman) Let M be the
Polish space of minimal flows on compact metric Sf : Y  G=H ! Y  G=H
spaces. Then:
of Y is determined by a measurable function
1. The collection of distal flows is a complete f : Y ! G/H and given by the formula:
co-analytic set; in particular, it is non-Borel.
2. The natural function that associates to each Sf y, ½g H ¼ SðyÞ, ½fðyÞg H :
distal (X, d, T), its distal height is a P11 -norm.
Then Sf preserves the product of n and Haar
The first item is proved by showing that there is measure on G/H.
a Borel reduction from the set of codes for well-
orderings, WO to the set of distal flows as a subset Definition 25. (See, e.g., Furstenberg 1981). Let
of the space of minimal ℤ-flows. (X, ℬ, m, T ) and ðY, C , n, SÞ be ergodic measure-
If (X0, d0, T0), (X1, d1, T2) are minimal distal preserving systems. Then:
flows on compact spaces, set (X0, d0, T0)D(X1,
d1, T2) if the distal height of (X0, d0, T0) is less 1. X is a compact extension of Y if X is measure
than or equal to the distal height of (X1, d1, T2). equivalent to a compact co-cycle extension
Then the theorem shows that D is a true P11 - of Y.
550 The Complexity and the Structure and Classification of Dynamical Systems

2. Let π : X ! Y be a measure-preserving exten- (X, ℬ, m, T ) can be written satisfying 1–4 of the

sion. Let XYX be the measure space which theorem.
has domain {(x1, x2) : π(x1) ¼ π(x2)} and the The results are now analogous to the topolog-
canonical measure projecting to n that makes ical case (see Beleznay and Foreman 1996):
the two copies of X relatively independent over
ðY, C , nÞ: Let T be the measure-preserving sys- Theorem 28. (Beleznay-Foreman) Let EMPT
tem taking (x0, x1) to (T(x0), T(x1)). Then ([0, 1]) be the Polish space of ergodic measure-
Y canonically a factor of XYX and X is a preserving transformations of the unit interval.
weakly mixing extension of Y if XYX is Then:
ergodic.
1. The collection of measure distal flows is a
The theorem proved independently by complete co-analytic set; in particular, it is
Furstenberg and Zimmer is: non-Borel.
2. The natural function that associates to each
Theorem 26. Let (X, ℬ, m, T) be an ergodic measure distal (X, ℬ, m, T ) its distal height is
measure-preserving system. Then there is a count- a P11 -norm.
able ordinal  and a system of measure-preserving
transformations h(Xα, ℬα, mα, Tα) : α   þ 1i It follows from the second item that if  is a
such that fixed countable ordinal then the collection of mea-
sure distal homeomorphisms of distal height less
1. (X0, ℬ0, m0, T0) is the trivial flow. than  is a Borel set.
2. For each α < , Xαþ1 is a compact extension of An open problem in both settings The fol-
Xα. lowing problems are analogous and involve over-
3. If α is a limit ordinal then Xα is the inverse limit spill. (See Open Problem 7)
of hXβ : β < αi.
4. X þ 1 is either: • Let X be the collection of minimal homeomor-
• a trivial extension of X (so X þ 1 ffi X), or phisms on compact metric spaces and D be the
• a weakly-mixing extension of X. collection of topologically distal transforma-
tions. Is there a Borel set B
X  X such that
Note that in each case, for α ¼ 1, the transfor- B \ ðD  D Þ is the relation of topological
mation Xα has to be a translation on a compact conjugacy?
quotient G/H. In this case, G must be abelian so • Let MD be the collection of measure distal
G/H is also an abelian group. Hence, monothetic transformations. Is there a Borel set
and an example of the form due to Halmos and B
EMPT  EMPT such that B \
von Neumann. ðMD  MD Þ is the relation of measure
In analogy to the situation for topologically conjugacy?
distal transformations, we give the following
definition: In unpublished work, Foreman showed that for
the collection of measure distal transformations
Definition 27. An ergodic measure-preserving with height bounded by a countable ordinal α, the
system (X, ℬ, m, T ) is measure distal if it can be answer to the second question is yes.
written in the form of the previous theorem with
Xþ1 ¼ X.
Complexity in Classification Theory
As in the topological case, we define the distal
height of a measure distal transformation In this section, we list various benchmarks for
(X, ℬ, m, T ) to be the least  such that complexity of classifications of equivalence
The Complexity and the Structure and Classification of Dynamical Systems 551

relations E that have roots outside of dynamical with countable classes are referred to as countable
systems. The most sensitive measures of com- equivalence relations.
plexity are the two dimensional questions about The orbit equivalence relation of any countable
reductions between equivalence relations. The group action has countable classes, so there is a
general regions used to classify equivalence rela- plethora of examples.
tions are given in Fig. 2. Let hfn : n  ℕi be a collection of Borel
A basic one-dimensional benchmark If X is a bijections of a Polish space to itself. Let Γ be the
Polish space then X  X is also Polish. For any group generated by hfn : n  ℕi. Then each orbit
equivalence relation, E
X  X, one can ask the of Γ is countable and the orbit equivalence relation
one-parameter question: is Borel. Conversely, a theorem of Feldman and
Moore (1977) states:
• Is E Borel as a subset of X  X?
Theorem 29. (Feldman-Moore) Suppose that
E
X  X is a countable Borel equivalence rela-
If not, then determining whether a pair
tion on a Polish space X. Then there is a countable
(x0, x1)  X  X is in the relation E cannot be
group Γ and a Borel action of Γ on X such that E is
accomplished with inherently countable
the orbit equivalence relation of the action.
resources.
Many important equivalence relations arise out A ≼2B maximal countable Borel equivalence
of group actions. If G acts on X then the equiva- relation Let G be a countable discrete group and
lence relation x  y if and only if there is a g  G f : G  X ! X be a Borel action on a Polish space
such that gx ¼ y is called the orbit equivalence X. Then the orbit equivalence relation is a count-
relation. able Borel equivalence relation on X.
A Polish group action is an action of a Polish Following (Jackson et al. 2002) we let sG be the
group G on a Polish space X that is jointly contin- shift action of G on the space XG, consisting of
uous in each variable. Section “S1-Actions” is functions f : G ! X. Let E(G, X) be the orbit
concerned with actions of the group of permuta- equivalence relation of this action. Dougherty
tions of the natural numbers, S1. When we say G et al. (1994) showed that if G is the free group
acts on X where G and X are both Polish, we will on two generators, F2, then E(F2, X) is a maximal
always be assuming that the action is a Polish Borel countable equivalence relation. The follow-
group action. ing is immediate from the Dougherty-Jackson-
Remark. When we introduce a class of equiv- Kechris theorem.
alence relations (such as the S1-actions defined
below), we really mean to take ≼2ℬ downwards Theorem 30. Let E1 be any countable equiva-
closure. So the class we label “S1-actions” lence relation bi-reducible to E(F2, X) for any
means any equivalence relation reducible to an Polish space X. Then every countable Borel equiv-
S1-action. alence relation is reducible to E1.

We show more about countable equivalence

Equivalence Relations with Countable
relations in section “Equivalence Relations with
Classes
Countable Classes.”
The equivalence relations that have finite or
countable classes are, perhaps, the simplest equiv- The Glimm-Effros Dichotomy: E0
alence relations. There is an extensive literature We now turn to equivalence relations with
on this, see, for example, Jackson et al. (2002); uncountable classes. Because complete numerical
Kechris and Miller (2004). Equivalence relations invariants are so useful, they are the baseline
552 The Complexity and the Structure and Classification of Dynamical Systems

benchmark of complexity. If Y is a Polish space, Theorem 34. (Harrington et al. 1990) Let X be a
we let ¼Y
Y  Y be the equivalence relation of Polish space and E a Borel equivalence relation
equality on Y. on X. Then either:
If Y is a Polish space then there is a Borel
injection g : Y ! ℝ (See Foreman 2010). Let 1. E is smooth or
f be a Borel reduction of any equivalence relation 2. E0 ≼2ℬ E:
E
X  X to (Y,¼Y). Composing f with g we see
that g ∘ f is a reduction of (X, E) to (ℝ,¼ℝ).
Moreover g ∘ f is Borel if f is. Thus, we can Corollary 35. ¼ ≼2ℬ E0 but ¼. Thus E0 is
assume that Borel reductions to any ¼Y can be strictly more complicated than the identity equiv-
changed to Borel reductions to equality on the real alence relation.
numbers.
This theorem suggests an orderly stair-step
Definition 31. Let E be an equivalence relation pattern to the benchmarks, but this is misleading,
on a Polish space X. Then E is smooth if E≼2ℬ ¼ℝ : as we shall see.
We prove the easy direction of Theorem 34,
In 1990, Harrington et al. (1990), extending because it is most useful in dynamical systems.
earlier work of Glimm (1961) and Effros (1965),
showed a very general dichotomy theorem. 1. Put a measure on {0, 1} by weighting each
element 1/2. The product measure m on
Definition 32. Let E0 be the equivalence relation {0, 1}ℕ is often called the coin-flipping
on the Cantor set {0, 1}ℕ defined by setting f E0 g measure.
if and only if there is an N for all m > N, f(m) ¼ 2. Let G ¼ Sn  ℕℤ2 be the direct sum of infinitely
g(m). many copies of ℤ2 (with finite support).
3. The coordinatewise action of G on
Proposition 33. The equivalence relation ¼ on ({0, 1}ℕ, ℬ, m) preserves m.
{0, 1}ℕ is continuously reducible to E0. 4. A standard 0 – 1 law for this action says that if
A
{0, 1}ℕ is invariant under the G-action
‘ Fix a bijection h, i : ℕ  ℕ ! ℕ. Define then A either has m-measure 0 or m-measure 1.
R : {0, 1}ℕ ! ℕ by setting (This simple, elegant
proof was taken from notes from a lecture of Su Theorem 36. Suppose that E is an equivalence
Gao. See citation in the bibliography) relation on an uncountable Polish space X and
E0 ≼2ℬ E. Then E is not smooth.
Rðf Þðhm, niÞ ¼ f ðnÞ:
‘ Toward a contradiction, assume that E is
Since R is well-defined, f ¼ g implies R( f ) Borel reducible to the identity equivalence rela-
E0R(g). On the other hand suppose for all tion on ℝ by a map S : X ! ℝ. Let R : {0, 1}ℕ ! X
m > N, R( f )(m) ¼ R(g)(m). Let n  ℕ and choose be a reduction of E0 to E. By composing, we see
an M such that hM, ni > N. Then that S ∘ R is a reduction of E0 to the identity
relation on [0, 1]. It suffices to show that there
f ð nÞ ¼ Rðf ÞðhM, niÞ can be no such reduction from E0 to the identity
¼ RðgÞðhM, niÞ relation ¼ on [0, 1].
¼ gðnÞ: a Suppose that S is such a reduction. Because
G preserves E0, for each rational number r  ℝ,
The Harrington-Kechris-Louveau dichotomy Lr ¼ {f : S( f ) < r} is G invariant, and so has
is the following theorem. measure zero or one. Let
The Complexity and the Structure and Classification of Dynamical Systems 553

x ¼ supfr : Lr has measure oneg: operation” to this set at each stage also gives a
proper hierarchy in ≼2ℬ :
Then I ¼ {f : S( f ) ¼ x} has measure one. The following result is proved with a similar
Hence there are f, g  I such that f E0 g, a approach to remark 36. See (Gao 2022) for a


contradiction. simple elegant proof of the first reduction.

5+ and the Friedman-Stanley Jump Operator Theorem 37. The following relations hold:
Let S1 be the group of permutations of the natural
numbers (Some authors reserve the notation S1
for the group of permutations of ℕ that move only
finitely many points and use S1 for the whole In particular,
group of permutations. The convention in this
paper is different and is becoming universal.). S1- Actions
Then S1 is a G d subset of ℕℕ and so carries a We now discuss S1-actions, where S1 is the
Polish topology. This topology is compatible with Polish group of all permutations of ℕ that was
the group structure and so we view S1 as a Polish defined in the previous section. The S1-actions
group. are a fundamental benchmark because their
The following operation on analytic equiva- actions are ubiquitous: every non-Archimedean
lence relations was defined by Friedman and Stan- Polish group is isomorphic to a closed subgroup
ley (Friedman and Stanley 1989). of S1 (See Becker and Kechris 1996).
Let E
X  X be an analytic equivalence Fix an infinite perfect Polish space X and con-
relation on a Polish space. Define E+to be the sider P
Xℕ defined as the collection of infinite
equivalence on Xℕ given by setting sequences hxn : n  ℕi  Xℕ such that
h[xn]E : n  ℕiE+-equivalent to h[yn]E : n  ℕi
if and only if for all n 6¼ m, xn 6¼ xm :

for all n there is an m xnEym Then P is a perfect closed subset of Xℕ, S1 acts
! !
and on P coordinatewise and for x , y  P we have
for all m there is an n ymExn.
! ! ! !
x ¼þ y if and only if there is a f  S1 , f x ¼ y :
Because the collection of Borel sets (and
respectively analytic sets) are closed under count- Thus the equivalence relation on P induced by
able intersections and unions, the form of the ¼+is the orbit equivalence relation of an S1-
definition makes it clear that if E is Borel, then action.
E+is Borel (and similarly for E being analytic). If
X and Y are perfect Polish spaces, then ¼þX and ¼Y
þ
Remark 38. The equivalence relation ¼+restricted
are the Borel bi-reducible. Moreover starting with to P is a Borel equivalence relation. Thus there are
the identity relation ¼ iterating “+” operation any faithful S1-actions inducing Borel equivalence
countable ordinal number α of times leads to a relations.
sequence of equivalence hEβ : β < αi such that
! !
β < β0 implies Eb ≼ℬ Eb0 but (Friedman Remark 38 is immediate: for x and y in
! !
and Stanley 1989). Thus this is a strict hierarchy P, x ¼þ y if and only if for all n there is an
of height o1. m with xn ¼ ym and for all m there is an n with
!
If E is Borel then x  Xℕ : ½xn E 6¼ ½xm E for y m ¼ x n.
Notation The only example of the use of the
all n 6¼ m} is a Borel set. Restricting the “+- “+-operation” in this entry will be when E is the
554 The Complexity and the Structure and Classification of Dynamical Systems

identity equivalence relation, ¼, on the unit circle. 1. (Mekler 1981) F≼2ℬ Egroups
We will restrict ¼+to one-to-one sequences. In an 2. (Friedman, Stanley, see (Gao 2009))
abuse of notation, we will continue to refer to this F≼2ℬ Egraphs
relation as ¼+.
Classification by countable structures We Thus both Egroups and Egraphs are ≼2ℬ -maximal
now give an example to illustrate the importance among S1-actions.
of S1-actions for creating invariants for equiva-
lence relations. These examples illustrate why equivalence
Let (X, E) be an analytic equivalence relation relations that are reducible to S1-actions are
that has complete invariants consisting of count- referred to as those that can be classified by count-
able groups. This means that a countable group is able structures, the name for them given in
associated to each element of X by a Borel func- (Hjorth 2000). We note that there is varying ter-
tion in such a way that xEy just in case the group minology for equivalence relations that are ≼2ℬ
associated with x is isomorphic to the group asso-
bi-reducible to maximal S1-actions. They are
ciated with y. In the language we are using here,
called maximal or universal S1-actions and also
the equivalence relation E is being reduced to the
Borel Complete equivalence relations.
equivalence relation on pairs of countable groups
To illustrate the large collection of equivalence
given by isomorphism. The next example makes
relations that arise from S1-actions and are ≼2ℬ
this precise.
maximal among S1-actions we cite results of
Example. Countable groups are determined
Camerlo and Gao (2001).
by the characteristic function of their multiplica-
Recall that almost finite commutative C-
tion. Explicitly, for G an infinite countable group
algebras (commutative “AF” C-algebras) are
we can assume that its domain is ℕ. Define
those that are inverse limits of finite dimensional
commutative C algebras. They have a classifica-
1 if l G m ¼ n
wG ðl, m, nÞ ¼ tion in terms of Bratelli diagrams (See Camerlo
0 otherwise and Gao 2001).
3
Then each wG  f0, 1gℕ and {wG : G is a Theorem 40. (Camerlo-Gao) The isomorphism
3
countable group} is a G d subset of f0, 1gℕ , and relation among commutative AF C-algebras is
hence form a Polish space. Call it CG. ≼2ℬ -maximal among S1-actions
Let S1 act on CG by setting (f  wG)(l, m, n) ¼
wG(f(l), f(m), f(n)). If G, H are isomorphic, let A maximal S1-action is NOT Borel One
f : G ! H be the isomorphism. Then f takes the property that makes equivalence relations such
multiplication table of G to the multiplication as Egraphs and Egroups very useful is that maximal
table of H. Hence f  wG ¼ wH. Similarly if f  wG ¼ S1 relations are not Borel. We record this here:
wH, then f : G ! H is an isomorphism.
Fact 41. S1-actions that are ≼2ℬ -maximal
This example is clearly not specific to groups– have complete analytic orbit equivalence rela-
it can be adapted to any countable structures. In tions. Hence, they are not Borel.
particular, it applies to countable graphs viewed as In particular, isomorphism of countable
2
elements of f0, 1gℕ : Let Egroups be the equiva- groups, isomorphism of countable graphs, and
lence relation of isomorphism of countable groups isomorphism of AFC-algebras are not Borel
and Egraphs be the equivalence relation of isomor- equivalence relations.
phism of countable graphs. For some applications in dynamical systems
the particular case of “isomorphism of countable
Theorem 39. Let S1 act on an arbitrary Polish graphs” is a particularly convenient example of an
space X and F be the resulting orbit equivalence ≼2ℬ -maximal S1 orbit equivalence relation.
relation. Then:
The Complexity and the Structure and Classification of Dynamical Systems 555

A caution We do note that not all non-trivial 1. Every orbit is dense.

“classifications by algebraic structures” give 2. Every orbit is meager.
equivalence relations that are ≼2ℬ -maximal S1 3. For all x, y  X, V X, U G open, with
actions. An example, due to Friedman and Stanley x  V, 1  U, there exists y0  [y]G and
(Friedman and Stanley 1989) is the equivalence hgiii  ℕ U, hxiii  ℕ V with
relation of isomorphism of countable abelian tor- (a) x0 ¼ x,
sion groups. This is a complete analytic equiva- (b) xiþ1 ¼ gixi
lence relation that determined by an S1-action, (c) There is a subsequence in such that xin
but is not ≼2ℬ -maximal among such actions. converges to y0.
Benchmarks to this point The equivalence
relations are diagramed in Fig. 1:
Hjorth proved the following result:
Turbulence
To establish that an equivalence relation is reduc- Theorem 43. Suppose that a continuous action
ible to an S1-action, one exhibits a reduction. of G on X is turbulent. Then the orbit equivalence
Hjorth (2000) devised a tool for showing that an relation cannot be reduced to an S1-action.
equivalence relation is not reducible to an S1-
action (The tool was discovered before there Further results of Hjorth show that turbulence
were widespread interactions between descriptive almost exactly captures the property of not being
set theory and dynamical systems, so the term reducible to an S1-action (See Hjorth 2000).
turbulance does not signify any direct connection Remark. The property of being turbulent is
with chaotic behavior in the sense of dynamics). generic in the sense that if Y X is a G d comeager
set that is invariant under the G-action then the
Definition 42. Let G be a Polish Group acting orbit equivalence relation induced by G on Y is
continuously on a Polish Space X. Then the action also turbulent.
is turbulent iff:
Polish Group Actions
The next major benchmark is to ask whether a
given equivalence relation is reducible to the
equivalence relation given by a Polish Group
action. Because S1 is a Polish group, this is a
more general collection of equivalence relations.
Because many natural equivalence relations
are given by Polish group actions, this is an enor-
mous collection. Here are some facts that help
navigate the subject. An excellent book on the
subject is due to Becker and Kechris (1996),
although there have been many developments
since the book was published.

Theorem 44. Fix a Polish group G.

1. There is a ≼2ℬ maximal equivalence relation

given by Polish G-actions.
2. There is a ≼2ℬ maximal equivalence relation
The Complexity and the Structure and Classification given among all Polish group actions.
of Dynamical Systems, Fig. 1 Benchmarks
556 The Complexity and the Structure and Classification of Dynamical Systems

3. If H
G is a closed subgroup, then the max- we don’t expand on systems with numerical
imal Polish H-action is continuously reducible invariants or on equivalence relations with E0
to the maximal Polish G-action. embedded in them, though we give examples of
each for dynamical systems. For S1 we have
Items 1 and 2 are due to Becker and Kechris. already given an example of a Borel action
Item 3 is essentially due to Mackey in the Borel (Remark 38) and ≼2ℬ -maximal S1-actions
case and proved by Hjorth in the continuous case. (Theorems 39 and 40).
Uspenskiĭ proved that the group of homeomor-
phisms of the Hilbert cube is a universal Polish Countable Equivalence Relations
group: every Polish group is isomorphic to a Recall that an equivalence relation is said to be
closed subgroup of this group. Thus, item 2 fol- countable if it has countable classes (see section
lows by applying items 1 and 3 to the group “Equivalence Relations with Countable Classes”)
Uspenskiĭ used (Gao 2009; Uspenskiĭ 1986) See In section “Countable Equivalence Relations”,
Fig. 2 for a diagram of the regions discussed here. we saw that Dougherty et al. (1994) showed that
there is a ≼2ℬ -maximal countable equivalence
relation they call E1. So, understanding count-
Standard Mathematical Objects in Each able equivalence relations is, in part, understand-
Region ing their relationship to E1. Rather than try to
cover a large literature, we focus on work of
The classes with classifications in the lower Thomas (2003, 2011) that is related to dynamical
regions of the diagram are extremely plentiful so systems and gives natural example of a ≼2B -

The Complexity and the Structure and Classification of Dynamical Systems, Fig. 2 Basic regions of complexity
The Complexity and the Structure and Classification of Dynamical Systems 557

maximal equivalence relation with countable

classes.
Note that the following sets are naturally
viewed as Polish spaces endowed with countable
equivalence relations:

• The space of finitely generated groups, with the

equivalence relation of isomorphism denoted
ffifin.
• The space of torsion-free abelian groups of
rank n, with the equivalence relation of isomor-
phism denoted ffin,
• The space of torsion-free, p-local abelian
groups of rank n, with the equivalence relation
of isomorphism related ffipn (for p a prime
number).

Using Zimmer’s notion of super-rigidity, and

the Kurosh-Mal’cev p-adic localization tech-
nique, Thomas (2003, 2011) proved the following
results except 1 which is due to Baer. The fifth
item is joint work of Thomas and Velickovic The Complexity and the Structure and Classification
(1999) (Hjorth showed that for n > 1, ffin 62ℬ ffi1 of Dynamical Systems, Fig. 3 The Thomas diagram
using elementary techniques. (See Hjorth 1999.)):
normal operators on a Hilbert space can be
1. ffi1ℬE0, reduced to mutual absolute continuity of measures
2. For n  1, ffin is strictly ≼2ℬ below ffinþ1. In on the unit circle. The latter can be checked to be
symbols: ffin ≼2ℬ ffinþ1 , Borel using the Suslin Theorem (This was
3. For distinct primes p, q and n>2, the relations extended by Ding and Gao (2014) to the
ffipn , ffiqn are ≺2B -incomparable, conjugacy action of all bounded linear operators
4. For each prime p and each n  1, ffipn ≼2ℬ ffinþ1 on the unitary operators.).
in the strict sense, However, it was proved by Kechris and
5. If ffi1 is the relation of isomorphism for Sofronides that this equivalence relation is not
finitely generated groups, then E1ℬ(ffi1). reducible to an S1-action. They use the turbu-
lence of the ‘2-action on ℝℕ. (See (Kechris and
These results can be best summarized in Sofronidis 2001).)
Fig. 3. A related example is the maximal unitary
group action. This was shown by Pestov and
General Polish Group Actions Gao (Gao and Pestov 2003) to lie ≼2ℬ above the
We first give an example of a Polish group action maximal ‘1-action, which lies above all Abelian
that is not reducible to an S1-action but is still Polish group actions. In particular, it is not reduc-
Borel. ible to any ‘1-action. Kanovei (2008) showed that
Example. Let U be the collection of unitary the maximal unitary group action is not reducible
operators on a separable Hilbert Space H and N to any ‘1-action.
be the collection of normal operators on H. Let The collection of Koopman operators that arise
E be the equivalence relation on U given by from measure-preserving transformations on a
conjugacy of N . The Spectral Theorem shows measure space (X, ℬ, m) is a closed subgroup of
that equivalence relation of unitary conjugacy of the group of unitary transformations on L2(X).
558 The Complexity and the Structure and Classification of Dynamical Systems

Thus, it follows from the Mackey-Hjorth theory any Polish group action and is, moreover, ≼2ℬ -
(Theorem 44) that the isomorphism action of the minimal with these properties.
measure-preserving transformations on the ergo- View {0, 1}ℕℕ as the collection of infinite
dic transformations is reducible to the maximal sequences hxn : n  ℕi where xn  {0, 1}ℕ. Set
unitary group action. hxninE1hymim if and only if there is an N for all
Choquet Simplexes An example of a natural n > N, xn ¼ yn. Note that this is an analogue of E0
equivalence relation of maximal ≺2B -complexity where elements of {0, 1} have been replaced by
among Polish group actions is due to Sabok. elements of {0, 1}ℕ and the equivalence relation
Recall that a Chouqet simplex is a compact, is eventual equality.
separable, affine space X such that any point
x  X is the integral of a unique measure concen- Theorem 46. (Kechris-Louveau, (Kechris and
trated at the extreme points of X. There is a Louveau 1997)) E1 is Borel and not reducible to
Choquet simplex, the Poulsen simplex, which is a Polish group action.
universal in the sense that every Choquet simplex
can be embedded as a face. This puts a Polish We now give another example arising in topol-
topology on the collection of Choquet simplexes. ogy. Let X be the collection of compact separable
Let ðC , tÞ be this Polish space. metric spaces with ℕ as a dense subset. Then X
Set two Choquet simplexes S1, S2 equivalent if carries a natural Polish topology.
there is an affine homeomorphism taking S1 to S2. We define the notion of bi-Lipschitz equiva-
Let AFF(Poulsen) be the group of affine homeo- lence. For compact separable metric K 1 , K 2 set
morphisms of the Poulsen simplex. Then K 1 RK 2 if and only if there is a bijection f :
AFF(Poulsen) is a Polish group and its natural K 1 ! K 2 such that both f, f1 are Lipschitz. The
action on C maps faces to faces, so Aff(Poulsen) following theorem appears in Rosendal (2005).
acts on C by a Polish action.
Lindenstrauss et al. (1978) showed that any Theorem 47. (Rosendal) The equivalence rela-
affine homeomorphism between two proper sub- tion R
X  X is Borel but not reducible to a
faces of the Poulsen extends to an affine homeo- Polish group action.
morphism of the whole simplex. It follows that the
action of AFF(Poulsen) on C coincides with the The Rosendal theorem is proved by reducing
relation of being affinely homeomorphic. Thus, E1 to bi-Lipschitz equivalence and applying The-
we can view the equivalence relation of being orem (Kechris and Louveau 1997). It is
affinely homeomorphic as the orbit equivalence conjectured that E1 is ≼1ℬ -below every Borel
relation of a Polish group action. equivalence relation that is not reducible to a
M. Sabok (2016) showed that affine homeo- Polish group action. This is Open Problem 16.
morphism of Choquet simplices is ≼2ℬ maximal
among equivalence relations induced by Polish The Maximal Analytic Equivalence Relation
group actions: While it was a folklore result that there is a max-
imal equivalence relation among all analytic
equivalence relation, the construction was quite
Theorem 45. (Sabok) Every orbit equivalence
abstract. In Ferenczi et al. (2009), Ferenczi,
relation of a Polish group action is ≼2ℬ -reducible
Louveau, and Rosendal gave a natural example.
to the equivalence relation of affine homeomor-
It is an equivalence relation on the Polish space of
phism of Choquet simplices.
separable Banach spaces, which can be viewed as
the collection of closed subspaces of ℂ([0, 1])
Borel Equivalence Relations that Don’t Arise endowed with the Effros Borel structure.
from a Polish Group Actions
Kechris and Louveau defined a Borel equivalence Theorem 48. (Ferenczi, Louveau, Rosendal)
relation E1 on {0, 1}ℕℕ that is not reducible to The relation of isomorphism between separable
The Complexity and the Structure and Classification of Dynamical Systems 559

Banach spaces is maximal among all analytic than the relationship between topological
equivalence relations. conjugacy and descriptive set theory.
Work of Mitin (1998) and Ryzhikov (1985)
Since there are equivalence relations that are not studies measure-preserving transformations and
reducible to Polish group actions, and the Ferenczi, show that they encode the theory of Peano
Louveau, Rosendal example is maximal, it follows Arithmetic. As a result, many questions about
that the Ferenczi, Louveau, Rosendal example them are recursively undecidable. In this section,
does not arise as a Polish group action. we will focus on Borel reducibility.
We can now put these examples into the basic
benchmark regions (Fig. 4): Systems with Numerical Invariants
The clearest example here is the class of
Bernoulli Shifts. As noted in section “Rolling
Placing Dynamical Systems in Each Region Dice: Bernoulli Shifts”, the collection of
measure-preserving transformations isomorphic
Measure Isomorphism to a Bernoulli shift has a structure theory, and
The relationship between measure isomorphism Ornstein’s Theorem (Theorem 9) shows that
and descriptive complexity is better understood entropy is a complete numerical invariant.

The Complexity and the Structure and Classification of Dynamical Systems, Fig. 4 Classes placed in regions of
complexity
560 The Complexity and the Structure and Classification of Dynamical Systems

Reducing E0 to Dynamical Systems spectrum measure-preserving transformation.

A remarkable theorem of Feldman from a paper The upshot is the following, which appeared in
published in 1974 (Feldman 1974) shows the Foreman (2000):
following result:
Theorem 50. (Foreman-Louveau) Isomorphism
Theorem 49. The equivalence relation E0 is con- of discrete spectrum measure-preserving transfor-
tinuously reducible to isomorphism of K  mations is Borel bi-reducible to ¼+.
automorphisms:
Isomorphism Is Not Reducible to Any S1-Action.
Feldman’s paper predated the language and Hjorth (2001) showed that there is a turbulent
definitions given in this entry; however, his equivalence relation that is ≼2ℬ reducible to iso-
proof is easily translated into the proof given in morphism for ergodic height 2 measure distal
section “The Glimm-Effros Dichotomy: E0” after transformations (see section “Turbulence” for a
theorem 34. discussion of turbulence). It follows that:
Surprisingly, it is not known if the conjugacy
relation for K -automorphisms is Borel (See Open Theorem 51. (Hjorth) Isomorphism for ergodic
Problem 4). Nor is it known if it can be reduced to measure-preserving transformations is not reduc-
an S1-action. ible to an S1-action.

A Class Bi-reducible to ¼+ Foreman and Weiss (2004) improved Hjorth’s

In theorem 12, Halmos and von Neumann showed result by showing that the relation of isomorphism
that two discrete spectrum measurepreserving of ergodic measure transformations is itself a tur-
transformations are isomorphic if and only if bulent equivalence relation.
their associated Koopman operators have the
same eigenvalues. In tracing the proof (see Theorem 52. (Foreman, Weiss) Let MPT ([0, 1])
Foreman 2000), one sees that this is a reduction be the group of measure-preserving actions on [0,
of isomorphism to ¼+on the complex unit circle. 1] with the usual Polish topology and EMPT ([0,
The reduction in the other direction also uses 1]) be the collection of ergodic transformations.
the Halmos-von Neuman theory. First, observe Then the conjugacy action of MPT ([0, 1]) on
that there is a dense G d subset I of the complex MPT ([0, 1]) is turbulent.
unit circle S1 such that any x1, x2, . . . xn which
belong to I are algebraically independent. From the remark following Theorem 43, that
!
To each x ¼ hxn : n  ℕi with xn  I, let G! generic classes of measure-preserving transforma-
x
tions, such as the ergodic or weakly mixing trans-
be the multiplicative subgroup of S1 generated by
! formations, are also turbulent. Hence,
x and consider its Pontryagin dual G! consisting isomorphism is not reducible to an S1-action.
x
of all characters of the group G! : Let X be the
x
identity map from G! to the unit circle. Then G! Corollary 53. Let C be a dense G d collection of
x x
is a monothetic abelian group generated by w. measure-preserving transformations in [0, 1].
Multiplication by w in G! preserves Haar measure Then the isomorphism relation on C is not reduc-
x
and is an ergodic measure-preserving transforma- ible to an S1-action.
tion T ! . The Koopman operator of T ! has G! as
x x x
its group of eigenvalues. Isomorphism Is Not Borel
!
Summarizing: the map that takes x to In Hjorth (2001), Hjorth showed that the isomor-
phism relation between measure-preserving trans-
G! , ℬ, Haar, T ! is a reduction of the equi-
x x formations is not Borel. However, his proof used
valence relation ¼þ
I to isomorphism of discrete non-ergodic transformations in an essential way.
The Complexity and the Structure and Classification of Dynamical Systems 561

Foreman, Rudolph, and Weiss showed the fol- ≼2ℬ below isomorphism of ergodic measure-
lowing results (Foreman et al. 2011). preserving transformations. By Theorem 54, the
relation is strict.
Theorem 54. (Foreman, Rudolph, Weiss) Let
G be the group of measure-preserving transfor- Shortly after this entry was written, B. Weiss
mations of [0,1] with the standard Polish topology was able to show the following result, as yet
and EMPT be the ergodic transformations in unpublished:
G. The conjugacy action on G is complete Theorem. (Weiss) Let G be a countable group.
analytic. In particular, it is not Borel. Let X be the space of homomorphisms of G to
MPT([0, 1]) such that the action is free and
Using the method of suspensions one can lift ergodic. If G of the form H  ℤ, then there is a
these results to ℝ-actions (Foreman et al. 2011). Borel reduction from the isomorphism relation on
X to the isomorphism relation on ℤ-actions.
Corollary 55. The isomorphism relation Corollary. If G ¼ ℤd for d  1 then there is a
between ergodic measure-preserving ℝ-flows is Borel reduction from free measure-preserving
complete analytic, and hence not Borel. G-actions to measure-preserving ℤ-actions.

In unpublished work, Foreman was able to Smooth Measure-Preserving Transformations

establish the following: Much of the motivation for studying measure
isomorphism is to classify the quantitative behav-
Theorem 56. Isomorphism of countable graphs ior of solutions to ordinary differential equations.
is reducible to isomorphism for measure- The reductions built in Theorems 54 and 56 take
preserving transformations. their ranges in the collection of ergodic measure-
preserving transformations with non-trivial
Combining these results with Corollary 53, we odometer factors. There are no known measure-
get: preserving transformations on compact manifolds
that have odometer factors. This is a special case
Corollary 57. The equivalence relation of iso- of the classical Realization Problem (Open Prob-
morphism of measure-preserving transformations lem 8). A statement of the problem appears in
is strictly above S1-actions in ≼2ℬ . 1975 in a paper of Ornstein (1975), and the prob-
lem is attributed to Katok by Hasselblatt in
Theorem 56 also provides a second proof of the Hasselblatt (2007).
Kechris-Sofronidis result that the maximal unitary The natural question to ask is whether the
action is strictly above the maximal S1-action. measure isomorphism relation between volume
One can ask for more information in placing preserving diffeomorphisms on a compact mani-
the relation of measure isomorphism on EMPT in fold is Borel. Does the additional structure pro-
the ≼2ℬ relation. It was proved by Kechris and vided by being C1 give enough information
Tucker-Drob (2013) that the relation of unitary about measure-preserving transformations to
equivalence of normal operators is reducible to determine isomorphism?
measure isomorphism on EMPT. Since the unitary The answer is no:
equivalence is Borel and isomorphism of ergodic
measures is not, unitary equivalence is strictly ≼2ℬ Theorem 59. (Foreman, Weiss) The isomor-
below isomorphism of ergodic measures. phism relation between Lebesgue measure-
preserving C1 transformations on the 2-torus
Theorem 58. (Kechris, Tucker-Drob) The rela- 2 is complete analytic. In particular, it is not
tion of unitary equivalence of normal operators is Borel.
562 The Complexity and the Structure and Classification of Dynamical Systems

Remark. The techniques of Theorem 56 adapt A  ℬ be a set of positive measure and define
exactly to prove the analogous result for smooth mA : P(A) \ ℬ ! [0, 1] by setting
measure-preserving transformations. However,
because the isomorphism relation on smooth m ð BÞ
m A ð BÞ ¼ :
measure-preserving transformations is not a m ð AÞ
group action, the technique of turbulence is not
directly applicable. Thus it is possible that the By Poincare recurrence, for almost all x  A,
measure isomorphism relation on C1 transfor- there is an n > 0 such that T n(x)  A. Let n(x) be
mations on the 2-torus is ≼2ℬ bi-reducible with the least such x and TA(x) ¼ T n(x)(x). Then
the maximal S1-action. This is the content of TA : A ! A and is a measure-preserving transfor-
Open Problem 11. mation. If T is ergodic then TA is ergodic.
The proof of Theorem 59 involves three steps:
Definition 60. Two ergodic transformations
1. Identify a class of symbolic shifts (the circular (X, ℬ, m, T ) and ðY, C , n, SÞ are said to be
systems) that are realizable using the Anosov- Kakutani equivalent if there are positive measure
Katok ABC method. sets A  ℬ and B  C such that TA is isomorphic
2. Show that the class of measure-preserving to SB.
transformations containing an odometer has
the same factor and isomorphism structure as Gerber and Kunde (2021) proved the following
the class of circular systems. They are “isomor- theorem, following the general steps of the proof
phic” categories by an isomorphism F. of Theorem 59, but with significant
3. Build a reduction from the space of ill-founded improvements.
trees to the odometer systems; use the isomor-
phism F to transfer the range to the circular Theorem 61. The equivalence relation of
systems and then the ABC method to realize Kakutani equivalence on ergodic measure-
the results as diffeomorphisms. preserving diffeomorphisms of the 2-torus is com-
plete analytic. Hence, it is not a Borel equivalence
The proofs appear in the three papers: Foreman relation.
and Weiss (2019a); Foreman and Weiss (2019b);
Foreman and Weiss (2022). Because Kakutani equivalence for measure-
We note that these results give no upper bound preserving diffeomorphisms is ≼2ℬ reducible to
on the complexity of the isomorphism relation for Kakutani equivalence for general measure-
ergodic measure-preserving transformations, preserving transformations, the following is
other than the obvious one-that of the maximal immediate (Chronologically, the corollary pre-
Polish group action. Hence, we ask the question: ceded the theorem):
Is the maximal Polish group action reducible to
isomorphism of ergodic measure-preserving Corollary 62. The equivalence relation of
transformation? Kakutani equivalence on the space of ergodic
measure-preserving transformations on [0, 1] is
This is Open Problem 12.
complete analytic and hence, not Borel.
Kakutani Equivalence
Gerber and Kunde also proved an ℝ-flow ver-
An important relation in ergodic theory is
sion of Corollary 62.
Kakutani equivalence (see Thouvenot 2002) for
a nice exposition).
Definition 63. Let F ¼ {ft : t  ℝ} be a measure-
Let T be a measure-preserving transformation
preserving ℝ-flow on (X, m). A reparametrization
on a standard measure space (X, ℬ, m). Let
of F is a jointly measurable function t : X 
The Complexity and the Structure and Classification of Dynamical Systems 563

ℝ ! ℝ such that for almost every x the map Smooth Transformations with Numerical
t 7! t(x, t) is a homeomorphism of ℝ and the Invariants
map (x, t) 7! ft(x, t)(x) is a measure-preserving As discussed in section “Smooth Dynamics”, the
flow on X. collection of smooth transformations that is struc-
turally stable has complete numerical invariants.
If G ¼ {gt : t  ℝ} is another measure- These include the classes of Anosov and Morse-
preserving flow on X, then F and G are isomor- Smale diffeomorphisms. It is not clear how far, if
phic up to a homeomorphic time change if there is at all, these classes can be extended and still have
a reparameterization t of F and a measure- complete numerical invariants.
preserving transformation f : X ! X such that
for all t and almost every x: Diffeomorphisms of the Torus Above E0.
Foreman and Gorodetski showed that there are
gt ðfðxÞÞ ¼ f f tðx,tÞ ðxÞ : diffeomorphisms of any manifold of dimension
at least 2 that are ≼2ℬ above E0. It follows that
topological conjugacy for general diffeomorphisms
The Gerber-Kunde techniques extend to show:
is not numerically classifiable.
Corollary 64. The equivalence relation of iso- Theorem 65. (Foreman, Gorodetski) Let M be a
morphism up to a homeomorphic time change Ck manifold of dimension at least 2 and 1  k  1 .
for C1 measure-preserving flows is complete Then, there is a continuous reduction R from E0 to
analytic and hence not Borel. the collection of Ck diffeomorphisms with the
relation of topological conjugacy.
The proofs of the Gerber-Kunde theorems do
not give information about ≼2ℬ reducibility to any Remark. In fact, the construction creates a
of the benchmark analytic equivalence relations. reduction that takes values on the boundary of
(This is Open Problem 13.) the Morse-Smale diffeomorphisms of the torus.
We give here a nearly honest proof, which is
The Summary Diagram relatively easy to illustrate with a picture.
Figure 5 summarizes this section. The K  Let M be a manifold of dimension at least
automorphisms are not on the diagram because, 2. We take it to have dimension 2 for convenience
other than being above E0, their position is not (see Fig. 6). We work inside a neighborhood
known. For the same reason, the Kakutani equiv- diffeomorphic to ℝ2. Choose two points (the left
alence relation is not on the diagram. star and the right star) and four sequences of disks.
The disks’ diameters go to zero at the same rela-
tively fast rate, and two of them converge to the
Topological Conjugacy left point and two of them converge to the right
In this section, we discuss the qualitative behavior point. There are two lines that separate the region
as proposed by Smale. We do this in both the zero- into four parts. Each part contains exactly one of
dimensional and the smooth contexts. “Equiva- the sequences of disk. One of the lines passes
lence” in this context means conjugate by a through the left and right points, giving a notion
homeomorphism. of upper and lower sequences.
After this entry was finished, in joint work with
Gorodetski, the author improved the results for 1. The support of the diffeomorphism is the clo-
smooth transformations on manifolds with the sure of the interiors of the disks.
equivalence relation of topological conjugacy to 2. The diffeomorphism rotates each interior circle
all dimensions at least one. These will appear in a around the center of each disk and each disk
future papers. contains a subdisk that the diffeomorphism
 564
The Complexity and the Structure and Classification of Dynamical Systems

The Complexity and the Structure and Classification of Dynamical Systems, Fig. 5 Measure-preserving systems
The Complexity and the Structure and Classification of Dynamical Systems 565

The Complexity and the

Structure and
Classification
of Dynamical Systems,
Fig. 6 How to embed E0

rotates by a constant amount. (The subdisks are swap finitely many of the lower disks. Thus, R is a
indicated by the shading.) reduction of E0 to topological conjugacy.a
3. Each of the sequence of disks above the hori-
zontal line is identified with a fixed sequence of This example is, in many ways, unsatisfying. It
rotation numbers tending to zero and these are is the identity on large portions of the space. This
different for the two sequences. raises myriad questions. A simple one to frame is
4. There is a fixed sequence of different rotation the following (See Open Problem 14.)
numbers with |rn| < 1/2 and hrn : n  ℕi
tending to zero, with infinitely many positive Does E0 reduce to the collection of topologically
minimal diffeomorphisms of the 2-torus with the
and infinitely many negative. relation of topological conjugacy?
5. The lower subdisks have rotation numbers
h rn : n  ℕi.
6. The lower subdisks then code an infinite A Dynamical Class that Is ≼2ℬ Maximal for
sequence of 0’s and 1’s depending on whether Countable Equivalence Relations
they agree on the sign of the rotation in the nth A well-studied class of dynamical systems is the
subdisk. collection of finite subshifts. These are closed,
shift-invariant subsets of Sℤ for a finite alphabet S.
The reduction R maps {0, 1}ℕ to Let C be the class of finite subshifts. If S is the
diffeomorphisms built with the two points and alphabet then a finite code or block code is a
the sequences of disks. Given f : ℕ ! {0, 1}, function c : Sl ! S for some l  ℕ. Then
R( f ) is the diffeomorphism that c determines a mapping c : Sℤ ! Sℤ, where
c( f ) is achieved by “starting at zero and sliding
• Rotates the n-th subdisc of lower right c along f in both directions.” At each location
sequence in the positive direction and lower f restricted to the l-sized window is input into
left sequence in the negative direction if f(n) ¼ c which outputs a value in S.
1. If  and 0 are compact subshifts and conju-
• Rotates the nth subdisk of the lower left gate by a homeomorphism h, then that homeo-
sequence in the positive direction and the nth morphism is given by a pair of finite codes, one
subdisk of the lower right sequence in the for h and one for h1 (See Lind and Marcus 1995).
negative direction if f(n) ¼ 0. Since there are only countably many finite codes,
it follows that the equivalence relation has only
By construction, on the range of R, each con- countable classes.
jugating homeomorphism of ℝ2 must take the Clemens (see Clemens 2009) showed that gen-
sequences of upper disks to themselves, and only eral problem of homeomorphism is ≼2ℬ 
566 The Complexity and the Structure and Classification of Dynamical Systems

maximal among equivalence relations with count- Corollary 69. Let M be a smooth manifold of
able classes. dimension at least five. Then the equivalence rela-
tion of topological conjugacy on the space of
Theorem 66. (Clemens) The relation of topolog- diffeomorphisms of M is complete analytic. In
ical conjugacy between subshifts of Sℤ (with S particular it is not Borel.
finite) is ≼2ℬ maximal among analytic equivalence
relations with countable classes. The reduction used in the proof of Theorem 68
can be explained with pictures (although a rigor-
An Example of a Maximal S1-Action from ous proof requires some work).
Dynamical Systems Fix a graph G with vertices ℕ. Consider the
Since every perfect Polish space contains a solid unit ball in ℝ3 and choose a set I ¼
homeomorphic copy of the Cantor set, it is natural hpn : n  ℕi of countably many algebraically
to ask what the complexity is of the collection of independent points tending to the South Pole.
homeomorphisms of the Cantor set with the rela- Draw line segments between each pair of points
tion E of topological conjugacy. This was and between each point and the South Pole. The
answered by Camerlo and Gao (2001). result is a complete graph on countable many
points: the independent points together with the
Theorem 67. (Camerlo-Gao) Topological South Pole.
conjugacy of homeomorphisms of the Cantor set Let G ¼ (ℕ, E) be an arbitrary countable
is bi-reducible with a maximal S1-action. graph. Start the reduction by identifying ℕ with
the countably many points, but not the South Pole
As in the case of the smooth examples that (Fig. 7).
exhibit high complexity (Theorems 65 and 68) The coding device is to put disjoint five dimen-
the transformations in the range of the reduction sional “cigars” around each of the lines between
are far from being minimal. Thus, we have the the points in I. Given a graph G with vertices ℕ,
corresponding question: is topological conjugacy the support of the diffeomorphism fG that will be
of minimal homeomorphisms of the Cantor set the image of G under that reduction, will be the
bi-reducible with the maximal S1-action? What closure of the cigars. The diffeomorphism fG
about topologically transitive? (See Open Prob- either flows toward the center of the cigar if the
lem 15.) points connected by the cigar are connected in
G or flows away from the center if they are not
Topological Conjugacy of Diffeomorphisms Is Not connected (Fig. 8).
Borel
It might be expected that for objects as concrete as
a Ck diffeomorphism of a compact manifold, topo-
logical conjugacy would be Borel. However, in
dimensions 5 and above, it is known not to be
the case.

Theorem 68. (Foreman, Gorodetski) Let M be a

Ck smooth manifold of dimension at least five and
1  k  1. Then the relation of Graph Isomor-
phism on countable graphs is reducible to topo-
logical conjugacy of Ck diffeomorphisms of M.

Because Graph Isomorphism is a ≼2ℬ  The Complexity and the Structure and Classification
maximal S1-action, it is complete analytic. Hence, of Dynamical Systems, Fig. 7 The unit ball with edges
the following corollary is immediate. through it connecting independent points
The Complexity and the Structure and Classification of Dynamical Systems 567

To show this is a reduction one must show that homeomorphism that mimics f’s behavior on the
if we are given two graphs G ¼ (ℕ, E) and H ¼ countable set I and fixes the south pole of the ball
(ℕ, F), then the reduction produces diffeo- in ℝ3. For a single transposition, this is easy in four
morphisms fG and fH such that dimensions (See Fig. 9). However even two trans-
positions can interfere with each other. Having five
• G is isomorphic to H dimensions gives sufficient room to prevent this.
if and only if
• fG is topologically equivalent to fH. The Summary Diagram
Figure 10 summarizes the situation for topologi-
If fG is topologically equivalent to fH then cal conjugacy of diffeomorphisms, as of July 1,
the witnessing homeomorphism fixes the south 2022. The section above describes what was
pole and induces a permutation of the points in I. known on February 1, 2022 and hence does not
This, in turn, gives a permutation of the natural exactly correspond with the diagram.
numbers that is an isomorphism between the
graphs G and H. A Descriptive Set Theory Facts
The difficult direction is the opposite. Suppose We give a very quick review of descriptive set
that we are given a bijection f : ℕ ! ℕ that is an theoretic facts used in this article. For a survey that
isomorphism between G and H. We need to find a includes complete proofs of the facts stated here,

The Complexity and the

Structure and
Classification
of Dynamical Systems,
Fig. 8 Flowing toward and
away from the centerline

The Complexity and the

Structure and
Classification
of Dynamical Systems,
Fig. 9 Using the fourth
dimension to swap two
elements of the unit ball in ℝ3
 568
The Complexity and the Structure and Classification of Dynamical Systems

The Complexity and the Structure and Classification of Dynamical Systems, Fig. 10 Conjugacy by homeomorphisms
The Complexity and the Structure and Classification of Dynamical Systems 569

see the article Naive Descriptive Set Theory the use of the uncountable Axiom of Choice. For
(Foreman 2010). That article does not assume example, the Polish spaces contain the hierarchy
any familiarity with logic. of Borel Sets, the smallest s-algebra of subsets
We assume the reader is familiar with the ordi- X containing the open sets. This algebra is built by
nals (see Halmos’ Naive Set Theory), which are induction on the ordinals in a hierarchy of length
canonical representatives for any well-ordering. o1. Heuristically, it contains all sets that can be
Ordinality is concerned with order, as opposed to built with inherently countable information.
cardinality, which is concerned with size as deter- For classifying dynamical systems, Borel sets
mined by bijections. The start with the natural are not sufficient. One needs analytic and
numbers 0, 1, 2, . . . which form the ordinal o, co-analytic sets. These are extensions of the Borel
and continue by putting a point on top to get o þ sets that are nonetheless universally measurable.
1: (Fig. 11) We now give an inductive construction of the
After this a second copy of o to get o þ o ¼ Borel sets that begins with the open sets. To keep
o  2. Every ordinal α has an immediate succes- track of the level of complexity of the set, we
sor, α þ 1, and every collection of ordinals has a introduce the following “logical” notation:
supremum. Every ordinal is either a successor
ordinal or a limit ordinal. Informally, ordinals are Definition 70. Let (X, t) be a Polish topological
defined to be the set of smaller ordinals. So, 0 is space. The levels of the Borel hierarchy are as
the empty set, 1 ¼ {0} and so forth. (See (Halmos follows:
1974; Jech 2003) or (Levy 2002) for explanation.)
We will use this convention here: α ¼ {β : β is an 1. S01 sets are the open subsets of X.
ordinal and β < α}. 2. P01 sets are the closed sets.
In ZFC, the ordinals give canonical represen- 3. Suppose the hierarchy has been defined up to
tatives of the isomorphism types of every well- (but not including) the ordinal level α < o1.
ordering. The Axiom of Choice is equivalent to
Then A  S0a if and only if there is a sequence
the statement that every set can be well-ordered. It
hBi| i  oi of subsets of X with Bi  P0bi such
follows that there are ordinals of every cardinality.
that for each i  o, βi < α and
The smallest uncountable ordinal is called o1.
Because the collection of real numbers is
uncountable, it has cardinality at least the cardi- A ¼ [ Bi :
io
nality of o1 and Cantor’s Continuum Hypothesis
says that the two cardinalities are the same.
Descriptive set theory is largely concerned 4. B  P0a if and only if there is a subset of
with Polish Spaces X – these are topological X, A  S0a such that B ¼ X \A.
spaces whose topology is compatible with a sep-
arable complete metric. A Polish space is perfect We write D0a ¼ S0a \ P0a :
if it contains no isolated points. A measure on a
Polish space is standard if open sets are measur- The second and fourth items imply that the
able; it is non-atomic, complete, and separable. collection defined this way is closed under com-
Very roughly, the main topic of descriptive set plements and the third item says, in particular, that
theory is understanding what can be done without S0a sets are constructed by taking arbitrary

The Complexity and the Structure and Classification of Dynamical Systems, Fig. 11 A few ordinals
570 The Complexity and the Structure and Classification of Dynamical Systems

countable unions of sets of lower complexity. is called a co-analytic set. Note that we could get
Thus, S0a sets are closed under countable unions an equivalent definition of coanalytic by taking a
and P0a sets are closed under countable intersec- Borel set B and setting
tions. But none of the S0a ’s are closed under
countable intersections, nor are the P0a ’s closed C ¼ fx : ðfor all yÞðx, yÞ  B g: ð6Þ
under unions. Thus these definitions organize the
Borel sets in a hierarchy of length o1. Because o1 To be sure, the definition is clear – the analytic
has uncountable cofinality, [a<o1 S0a ¼ [a<o1 P0a sets are all sets of the form of A in Eq. 4 with B a
forms a s-algebra. Borel set. The co-analytic sets are all sets of the
Here are some facts: form in Eq. 5, or equivalently all sets of the form
Fact. Let X be an uncountable Polish space. in Eq. 6 with B a Borel set. For purposes of
Then for all countable α: showing that a set is analytic or co-analytic, it is
useful to use logical notation: “for some y” can be
S0aþ1 6¼ P0aþ1 : be written ∃y and “for all y” can be written 8y.
Informally analytic sets are “(∃y)Borel” and
• Both S0a and P0a are proper subsets of D0aþ1 : co-analytic sets are “(8y)Borel.” Because prod-
• In turn, each D0a is a proper subset of each ucts of Polish spaces are Polish, a set that can be
written “(∃y)(∃z)Borel” is analytic and “(8y)(8z)
S0aþ1 and each P0aþ1 .
Borel” is co-analytic.
Notation The “S11 sets” are the analytic sets
These facts are summarized in Fig. 12.
and the “P11 -sets” are the co-analytic sets.
Analytic and co-Analytic sets We now turn to Fact. Every perfect Polish space contains a
the next type higher. We define two collections of non-Borel analytic set. Moreover, the analytic
universally measurable sets using projections. sets are closed under countable intersections
Let X and Y be Polish spaces and B
X  Y be and unions. Hence, the coanalytic sets are also
a Borel set. The set closed under unions and intersections.
In many ways, the analytic and co-analytic sets
A ¼ fx : ðfor some yÞðx, yÞ  Bg ð4Þ function similarly to the open and closed sets.
Here is a remarkable theorem about the rela-
is called an analytic set. It’s complement tionship between analytic and co-analytic sets due
to Lusin with a very important corollary due to
C ¼ fx : ðfor all yÞðx, yÞ Bg ð5Þ Suslin. (See (Lusin 1933), or section 3.5 of

The Complexity and the Structure and Classification of Dynamical Systems, Fig. 12 The construction of the
Borel Sets
The Complexity and the Structure and Classification of Dynamical Systems 571

(Foreman 2010). Reference (Graham and Kantor ðT, x, yÞ : x 6¼ y andðfor some qÞðfor all nÞ,
2009) gives a popular history account.)
d ðT n ðxÞ, T n ðyÞÞ > 1=q :
Theorem 71. (Lusin’s Separation Theorem) Sup-
pose that X is a Polish space and A, B are disjoint
is Borel,
analytic subsets of X. Then there is a Borel set
C with A C and C \ B ¼ 0 . (So C separates


A from B.) T : ðfor all x 6¼ yÞfor some q for all n,

dðT n ðxÞ, T n ðyÞÞ > 1=q :

The next corollary is immediate.

Corollary 72. (Suslin’s Theorem) Let A
X. If is in the form of Eq. 6. It follows that the collection
both A and X \ A are analytic, then A is Borel. of distal homeomorphisms of X is co-analytic.

Example 73. For Borel group actions, the orbit Reductions and Hierarchies
equivalence relation is always analytic. If G is Fact. Let X, Y be Polish spaces and A
X, B
Y.
acting on X then Let f : X ! Y reduce A to B.

B ¼ fðx0 , x1 , gÞ : gx0 ¼ x1 g • If f is continuous and B  S0a then A  S0a :

• If f is continuous and B  P0a then A  P0a :
is a Borel set. The pair (x0, x1) lies in the same • If f is Borel and B is analytic, then A is analytic.
G-orbit if and only if (x0, x1)  A where • If f is Borel and B is co-analytic, then A is
co-analytic.
A ¼ fðx0 , x1 Þ : ðfor some gÞðx0 , x1 , gÞ  Bg:
Orderings and pre-orderings Let (I, I) be a
Thus being in the same orbit is analytic. linear ordering of a set of cardinality less than or
Similarly if C and D are Polish spaces of equal to the cardinality of the real numbers and X a
measure-preserving transformations and E
D perfect Polish space. Let C
X and π : C ! I be a
is a Borel set then surjective function. Then I is coded by the set of
A ¼ fT  C : (for some measure isomorphism pairs π ¼ {(x, y) : π(x)  Iπ( y)} in the sense that
f and some S  E) f ∘ T ¼ S ∘ f (a.e.)} is the isomorphism type of I can be recovered from
analytic. π. Note that π
X  X, and hence can be
studied as a subset of a Polish space, with no direct
From this one can see that the collection of mention of I itself.
measure-preserving transformations of [0, 1] that Of particular interest is the special case where
are isomorphic to a Bernoulli shift is an analytic I is an ordinal α and I is the well-ordering of the
set (details in Foreman 2000). These results are ordinals less than α. In this case π itself can be
weaker than Feldman’s paper where he shows recovered by π. When I is a well-ordering, the
that the collection of measure-preserving trans- relation π is called a pre-well-ordering because it
formations isomorphic to a Bernoulli shift is Borel is a pre-ordering and taking the quotient by the
(Feldman 1974). This is discussed in section equivalence relation
“Rolling Dice: Bernoulli Shifts.”
x  y if and only if ðx p y and y p xÞ
Example 74. Let X be a compact metric space.
Then the collection of distal homeomorphisms of results in a well-ordering of the  equivalence
X is a co-analytic set. classes that is isomorphic to α. When I is a well-
Because ordering the function π is called a norm.
572 The Complexity and the Structure and Classification of Dynamical Systems

Example. Let WO
{0, 1}ℕℕ be the set of A0 , B0  P11

characteristic functions of well-orderings of ℕ.
Then WO is a complete P11 -set. A0 \ B0 ¼ ;
The set WO is frequently called the collection
of codes for well-orderings. A0 [ B0 ¼ A [ B:
For notational convenience when π : C ! α is a
norm, it can be extended to all of X by adding a
point, denoted 1 to the range of π and extending π
by setting π(x) ¼ 1 for every element of X \ C. Open Problems
The following definition is difficult to digest, The author has made no attempt to make a com-
but the technicalities turn out to be essential. plete list of open problems – there are dozens. Nor
have they attempted correct attributions. These
Definition 75. Let X be a Polish space, α an are not even the most important problems,
ordinal, A a P11 -set and π : X! α [ {1} be although some, such as Open Problem 8, are
such that x  A iff π(x) < 1. Then π is a P11 - very classical and well studied. However, they
norm iff the relations p and <p are both P11 ask questions raised by this survey that may not
where: appear in other places.
Open Problem 1. Are there complete numeri-
1. xp y iff x  A and (π(x)  π( y)), cal invariants for orientation preserving
2. x<p y iff x  A and (π(x) < π( y)). diffeomorphisms of the circle with irrational rota-
tion number up to conjugation by orientation
A motivated reader might ask why P11 ? The preserving diffeomorphisms?
answer is that an analogous π on a S11 set can be Comment: After the first draft of this entry
shown not to exist. was written, Kunde solved this negatively by
The next proposition explains, in part, the showing that the equivalence relation E0 is reduc-
importance of P11 -norms. ible to diffeomorphisms with Liouvillean rotation
number under the equivalence relation of conju-
Proposition 76. Suppose that C
X is a P11 -set, gation by diffeomorphisms.
with X Polish and π : C ! o1 is a P11-norm. Then Open Problem 2. Is there an algorithm for
B
C is Borel implies that there is a countable determining whether two subshifts of finite type
ordinal α such that π[B]
α are conjugate by homeomorphisms?
Open Problem 3. Is the analogue of Theorem
A corollary of this proposition is that if C is a 8 true for groups besides ℤ. For example: let X be
P11 subset of a Polish space X, π : X ! o1 [ {1} the space of Lebesgue measure-preserving
is a P11-norm on C, and A
X is a Borel set with π actions of F2, the free group on two generators,
mapping a subset of A to a cofinal subset of o1, on [0, 1]. Is the collection of actions measure
then A ⊈ C. This phenomenon is called overspill. isomorphic to Bernoulli F2-actions a Borel set?
The reduction property of co-analytic sets Open Problem 4. Is measure isomorphism
The reduction property for P11-sets is related to P11 restricted to K automorphisms a Borel equiva-
-norms. It can be used to establish Lusin’s lence relation? Can it be reduced to an S1-
action?
theorem.
What about measure isomorphism for weakly
mixing transformations? Or other classes of
Theorem 77. (The reduction property for P11 - transformations?
sets.) Let X be a perfect Polish space. Suppose Open Problem 5. Consider various classes of
that A, B
X are P11-sets. Then there are A0
A, ergodic probability measure-preserving transfor-
B0
B with: mations such as weakly mixing, mixing,
The Complexity and the Structure and Classification of Dynamical Systems 573

0-entropy, and K automorphisms: Is there a ≼2ℬ Similarly in classes where there are existing
relationship between any of these classes, or invariants (such as the schemes that are used in
between one of these classes and the class of dimension 3 in (Bonatti et al. 2019)) one can ask
arbitrary ergodic probability measure-preserving whether those invariants are themselves Borel.
transformations. Open Problem 10. Is the relation of topolog-
Open Problem 6. Is isomorphism for ergodic ical conjugacy Borel when restricted to Axiom
measure-preserving transformations on s-finite A diffeomorphisms of a compact surface? Can
measure spaces Borel? E0 be embedded into topological conjugacy for
What is the ≼2ℬ relationship between isomor- Axiom A transformations?
phisms of ergodic probability measure-preserving Open Problem 11. Let M be a smooth mani-
transformations and isomorphism of ergodic fold with a smooth volume element n. Let SE be the
s-finite measure-preserving transformations? collection of smooth, ergodic, n-measure-
Open Problem 7. Do the structure theorems preserving transformations of M. Is the measure
for distality give Borel criteria for isomorphism? isomorphism relation on SE bi-reducible with the
More precisely: maximal S1-action?
Open Problem 12. (Sabok’s Conjecture) Is the
A. Let X be the collection of minimal homeomor- measure isomorphism relation on ergodic
phisms on compact metric spaces and D be the measure-preserving transformations of the unit
collection of topologically distal transforma- interval ≼2ℬ bi-reducible with the maximal Polish
tions. Is there a Borel set B
X  X such that group action?
B \ ðD  D Þ is the relation of topological Open Problem 13. Where does the Kakutani
conjugacy? equivalence relation on ergodic measure-
B. Let MD be the collection of ergodic measure preserving transformations sit among the analytic
distal transformations. Is there a Borel set equivalence relations in ≼2ℬ ? Is it reducible to an
B
EMPT  EMPT such that B \ S1-action? It is known to be a complete analytic
ðMD  MD Þ is the relation of measure set, hence not Borel (see Theorem 61).
conjugacy? In particular, what is the ≼2ℬ relationship
between the equivalence relation of Kakutani
Open Problem 8. Suppose that T : X ! X is an equivalence and isomorphism between ergodic
ergodic finite entropy transformation. Is there a probability measure-preserving transformations?
compact manifold M with a smooth volume ele- Between Kakutani equivalence and isomorphism
ment n and a measure-preserving diffeomorphism of ergodic measure-preserving transformations
S such that (X, ℬ, m, T ) is measure isomorphic to on s-finite measure spaces?
ðM, C , n, SÞ? Open Problem 14. Does E0 reduce to the
Open Problem 9. Let C be the collection of collection of topologically minimal
diffeomorphisms of a compact manifold M that diffeomorphisms of the 2-torus with the relation
are topologically conjugate to a structurally sta- of topological conjugacy? What about topologi-
ble diffeomorphism. Let E be the relation of topo- cally transitive diffeomorphisms?
logical conjugacy restricted to C : Is E Borel Open Problem 15. Is topological conjugacy of
reducible to ¼? In other words, does E have minimal homeomorphisms of the Cantor set
Borel computable, complete numerical bi-reducible with the maximal S1-action? What
invariants? about topologically transitive homeomorphisms?
This can also be asked for any standard class Open Problem 16. Suppose that F is a Borel
of structurally stable transformations, such as the equivalence relation on a Polish space that is not
Morse-Smale transformations. Is the conjugacy Borel reducible to the orbit equivalence relation
relation Borel? of a Polish group action. Is E1 ≼2ℬ F?
574 The Complexity and the Structure and Classification of Dynamical Systems

Acknowledgments The author received help from many Effros EG (1965) Transformation groups and C*-algebras.
people in writing and editing this entry. He particularly Ann Math 81(2):38–55
wants to acknowledge Filippo Calderoni, Marlies Gerber, Feldman J (1974) Borel structures and invariants for mea-
Anton Gorodetski, Philipp Kunde, Andrew Marks, Ronnie surable transformations. Proc Am Math Soc 46:
Pavlov. Alexander Kechris made invaluable corrections on 383–394
an early text and correspondence with Su Gao was essential Feldman J, Moore CC (1977) Ergodic equivalence rela-
in the completion of the text. As always, Benjamin Weiss tions, cohomology, and von Neumann algebras.
provided important suggestions and comments and pro- I. Trans Am Math Soc 234(2):289–324
vided very helpful references. Ferenczi V, Louveau A, Rosendal C (2009) The complex-
Expository Books and Articles on Analytic Equiva- ity of classifying separable Banach spaces up to iso-
lence Relations morphism. J Lond Math Soc 79(2):323–345
1. The book by Gao (2009), Foreman M (2000) A descriptive view of ergodic theory,
2. The books by Kechris (1995), Kechris–Miller Descriptive set theory and dynamical systems
(2004) and Becker–Kechris (1996), (Marseille-Luminy, 1996), London Mathematical Soci-
3. The paper by Friedman and Stanley (1989), ety lecture note series, vol 277. Cambridge Univ. Press,
4. Dave Marker’s website (Marker 2002). Cambridge, pp 87–171
The author would like to acknowledge partial support from Foreman M (2010) Naive descriptive set theory, arXiv:
NSF grant DMS-2100367. 2110. 08881, 1–53.
Foreman M (2018) What is a Borel reduction? Not Am
Math Soc 65(10):1263–1268
Foreman M, Weiss B (2019a) From odometers to circular
systems: a global structure theorem. J Mod Dyn 15:
Bibliography 345–423
Foreman M, Weiss B (2019b) A symbolic representation
Aaronson J (1997) An introduction to infinite ergodic for Anosov-Katok systems. J Anal Math 137(2):
theory, Mathematical surveys and monographs, 603–661
vol 50. American Mathematical Society, Providence, Foreman M, Weiss B (2022) Measure preserving
RI diffeomorphisms of the torus are unclassifiable. J Eur
Barreira L, Valls C (2013) Dynamical systems, an intro- Math Soc (JEMS), 1–80.
duction. Universitext, Springer, London Foreman M, Gorodetski A (2022) Anti-classification
Becker H, Kechris AS (1996) The descriptive set theory of results for smooth dynamical systems, arXiv: 2206.
polish group actions. London Mathematical society 09322, 1–56.
lecture note series. Cambridge University Press, Foreman M, Weiss B (2004) An anti-classification theorem
Cambridge for ergodic measure preserving transformations. J Eur
Beleznay F, Foreman M (1995) The collection of distal Math Soc (JEMS) 6(3):277–292
flows is not Borel. Am J Math 117(1):203–239 Foreman M, Rudolph DJ, Weiss B (2011) The conjugacy
Beleznay F, Foreman M (1996) The complexity of the problem in ergodic theory. Ann Math 173(3):
collection of measure-distal transformations. Ergodic 1529–1586
Theory Dyn Syst 16(5):929–962 Friedman H, Stanley L (1989) A Borel reducibility theory
Bonatti C, Grines V, Pochinka O (2019) Topological clas- for classes of countable structures. J Symb Log 54(3):
sification of Morse-Smale diffeomorphisms on 894–914
3-manifolds. Duke Math J 168(13):2507–2558 Furstenberg H (1963) The structure of distal flows. Am
Braverman M, Yampolsky M (2009) Computability of J Math 85:477–515
Julia sets, Algorithms and computation in mathematics, Furstenberg H (1981) Recurrence in ergodic theory and
vol 23. Springer-Verlag, Berlin combinatorial number theory, Porter lectures. Princeton
Camerlo R, Gao S (2001) The completeness of the isomor- University Press, Princeton
phism relation for countable Boolean algebras. Trans Gao S (2009) Invariant descriptive set theory, A series of
Am Math Soc 353(2):491–518 monographs and textbooks, vol 293. Taylor and
Chacon RV (1969) Weakly mixing transformations which Francis, Boca Raton
are not strongly mixing. Proc Am Math Soc 22: Gao S (2022) Dynamical systems and countable structures,
559–562 Lecture notes: https://www.birs.ca/workshops/2022/
Clemens JD (2009) Isomorphism of subshifts is a universal 22w5134/files/Su%20Gao/banff%20lecture%20gao%
countable Borel equivalence relation. Israel J. Math. 20nopause.pdf
170:113–123 Gao S, Pestov V (2003) On a universality property of some
Ding L, Gao S (2014) Is there a spectral theory for all abelian Polish groups. Fundam Math 179(1):1–15
bounded linear operators? Not Am Math Soc 61(7): Gerber M, Kunde P (2021) Anti-classification results for
730–735 the Kakutani equivalence relation, https://arxiv.org/
Dougherty R, Jackson S, Kechris AS (1994) The structure abs/2109.06086v1, 1–72.
of hyperfinite Borel equivalence relations. Trans Am Glimm J (1961) Locally compact transformation groups.
Math Soc 341(1):193–225 TAMS 101:124–138
The Complexity and the Structure and Classification of Dynamical Systems 575

Graham L, Kantor J-M (2009) Naming infinity. Belknap Khinchin AI (1949) Mathematical foundations of statisti-
Press, Cambridge cal mechanics. Dover Publications, Inc., New York.
Halmos PR (1944) In general a measure preserving trans- Translated by G. Gamow
formation is mixing. Ann Math 45(2):786–792 King J (1986) The commutant is the weak closure of the
Halmos PR (1956) Lectures on ergodic theory. Chelsea powers, for rank-1 transformations. Ergodic Theory
Publishing Company, New York Dyn Syst 6(3):363–384
Halmos PR (1974) Naive set theory, Undergraduate texts Krieger W (1970) On entropy and generators of measure-
in mathematics. Springer-Verlag, New York- preserving transformations. Trans Am Math Soc 149:
Heidelberg 453–464
Halmos PR and von Neumann J (1942) Operator methods Krieger W (1972a) Erratum to: “On entropy and generators
in classical mechanics. II, Ann of Math. 43(2):332–350 of measure-preserving transformations”. Trans Am
Harrington LA, Kechris AS, Louveau A (1990) A Glimm- Math Soc 168:519
Effros dichotomy for Borel equivalence relations. J Am Krieger W (1972b) On unique ergodicity. In: Proceedings
Math Soc 3(4):903–928 of the sixth Berkeley symposium on mathematical sta-
Hasselblatt B (2007) Problems in dynamical systems and tistics and probability, University of California, Berke-
related topics, Dynamics, ergodic theory, and geome- ley, 1970/1971, vol II: Probability theory, pp 327–346
try, Mathematical Sciences Research Institute Publica- Levy A (2002) Basic set theory. Dover Publications, Inc.,
tions, vol 54. Cambridge University Press, Cambridge, Mineola, Reprint of the 1979 original [Springer, Berlin;
pp 273–324 MR0533962 (80k:04001)].
Hjorth G (1999) Around nonclassifiability for countable Lind D, Marcus B (1995) An introduction to symbolic
torsion free abelian groups. In: Eklof PC, Göbel R (eds) dynamics and coding. Cambridge University Press,
Abelian Groups and Modules: International conference Cambridge
in Dublin, August 10–14, 1998. Birkhäuser, Basel, Lindenstrass J, Olsen G, Sternfeld Y (1978) The Poulsen
pp 269–292 simplex. Annales de l’institute Fourier 28(1):91–114
Hjorth G (2000) Classification and orbit equivalence rela- Lusin N (1933) Sur les classes des constituantes des
tions, Mathematical surveys and monographs, vol 75. complémentaires analytiques. Ann Scuola Norm
American Mathematical Society, Providence, RI Super Pisa Cl Sci 2(3):269–282
Hjorth G (2001) On invariants for measure preserving Marker D (2002) Descriptive set theory, http://homepages.
transformations. Fundam Math 169(1):51–84 math.uic.edu/~marker/math512/dst.pdf, 1–56.
Jackson S, Kechris AS, Louveau A (2002) Countable Maruyama G (1949) The harmonic analysis of stationary
Borel equivalence relations. J Math Log 2(1):1–80 stochastic processes. Mem Fac Sci Kyūsyū Univ A 4:
Jech TJ (2003) Set theory: The third millenium edition. 45–106
Springer, Berlin Mekler AH (1981) Stability of nilpotent groups of class
Kanovei V (2008) Borel equivalence relations: structure 2 and prime exponent. J Symb Log 46(4):781–788
and classification, University lecture series, vol 14. Mirzakhani M, Feng T (2014) Introduction to ergodic
American Mathematical Society, Providence, RI theory, https://www.mit.edu/~fengt/ergodic_
Katok A, Hasselblatt B (1995) Introduction to the modern theory.pdf, 1–56.
theory of dynamical systems, Encyclopedia of Mathe- Mitin AV (1998) Undecidability of the elementary theory
matics and its applications, vol 54. Cambridge Univer- of groups of measure-preserving transformations. Mat
sity Press, Cambridge Zametki 63(3):414–420
Kechris A (1995) Classical descriptive set theory, Graduate Moschovakis YN (2009) Descriptive set theory: second
texts in mathematics, 1st edn. Springer, New York edition, Mathematical surveys and monographs,
Kechris A (2010) Global aspects of Ergodic Group actions, no. 155, 2nd edn. American Mathematical Society,
Mathematical surveys and monographs. Am Math Providence, RI
Soc 160 Newhouse SE (1970) Nondensity of axiom A(a) on S2,
Kechris AS, Louveau A (1997) The classification of global analysis. In: Proceedings of the Symposium on
Hypersmooth Borel equivalence relations. J Am Math- Pure Mathematics, vol XIV, Berkeley, 1968, American
ematical Soc 10(1):215–242 Mathematical Society, Providence, pp 191–202.
Kechris A, Miller BD (2004) Topics in orbit equivalence, Newhouse SE (1974) Diffeomorphisms with infinitely
Lecture notes in mathematics. Springer, Berlin/ many sinks. Topology 13:9–18
Heidelberg Ornstein D (1970a) Bernoulli shifts with the same entropy
Kechris AS, Sofronidis NE (2001) A strong generic ergo- are isomorphic. Adv Math 4:337–352
dicity property of unitary and self-adjoint operators. Ornstein D (1970b) The isomorphism problem for
Ergodic Theory Dyn Syst 21(5):1459–1479 measure-preserving transformations. In: Proceedings
Kechris AS, Tucker-Drob RD (2013) The complexity of of the symposium on functional analysis, Monterey,
classification problems in ergodic theory, Appalachian 1969. Academic Press, New York, pp 71–74. MR
set theory 2006–2012, London Mathematical Society 0262463
lecture note series, vol 406. Cambridge University Ornstein D (1974) Ergodic theory, randomness, and
Press, Cambridge, pp 265–299 dynamical systems, Yale mathematical monographs,
576 The Complexity and the Structure and Classification of Dynamical Systems

No. 5. Yale University Press, New Haven/London, Proceedings of the International Congress on Mathe-
James K. Whittemore. Lectures in Mathematics given maticians, Stockholm, 1962. Institute Mittag-Leffler,
at Yale University Djursholm, pp 49–496
Ornstein D (1975) Some open problems in Ergodic theory. Smale S (1967) Differentiable dynamical systems. Bull
Publications mathématiques et informatique de Am Math Soc 73:747–817
Rennes, no. S4 Soare R (1987) Recursively enumerable sets and degrees,
Ornstein DS, Shields PC (1973) An uncountable family of perspectives in mathematical logic. Springer, Berlin/
K-automorphisms. Adv Math 10:63–88 Heidelberg
Ornstein DS, Weiss B (1974) Finitely determined implies Thomas S (2003) The classification problem for torsion-
very weak Bernoulli. Israel J Math 17:94–104 free abelian groups of finite rank. J Am Math Soc 16(1):
Oshemkov AA, Sharko VV (1998) On the classification of 233–258
Morse-Smale flows on two-dimensional manifolds. Thomas S (2011) The classification problem for S-local
Mat Sb 189(8):93–140 torsion-free abelian groups of finite rank. Adv Math
Peixoto MM (1973) On the classification of flows on 226(4):3699–3723
2-manifolds. In: Proceedings of the Symposium Thomas S, Velickovic B (1999) On the complexity of the
dynamical systems, University of Bahia, Salvador, isomorphism relation for finitely generated groups.
1971, pp 389–419 J Algebra 217(1):352–373
Petersen KE (1983) Ergodic theory, Cambridge studies in Thouvenot J-P (2002) Entropy, isomorphism and equiva-
advanced mathematics. Cambridge University Press, lence in ergodic theory, Handbook of dynamical sys-
Cambridge tems, vol 1A. North-Holland, Amsterdam, pp 205–238
Rohlin V (1948) A “general” a measure-preserving trans- Uspenskiĭ VV (1986) A universal topological group with a
formation is not mixing. Doklady Akad Nauk SSSR 60: countable basis. Funktsional Anal i Prilozhen 20(2):
349–351 86–87
Rohlin V (1967) Lectures on the entropy theory of trans- Walters P (1982) An introduction to ergodic theory, Grad-
formations with invariant measure. Uspehi Mat Nauk uate texts in mathematics. Springer, New York
22(5):3–56, 137. Weihrauch K (2000) Computable analysis, Texts in theo-
Rosendal C (2005) Cofinal families of Borel equivalence retical computer science. An EATCS series. Springer,
relations and quasiorders. J Symb Log 70(4):1325–1340 Berlin, An introduction
Ryzhikov VV (1985) Representation of transformations Wikipedia. Hamiltonian system. https://en.wikipedia.org/
preserving the Lebesgue measure, in the form of a wiki/Hamiltonian_system
product of periodic transformations. Mat. Zametki Wikipedia. Rotation number. https://en.wikipedia.org/
38(6):860–865, 957. wiki/Rotation_number
Sabok M (2016) Completeness of the isomorphism prob- Zimmer RJ (1976) Ergodic actions with generalized dis-
lem for separable C*-algebras. Invent Math 204(3): crete spectrum. Ill J Math 20(4):555–588
833–868 Zippin L (2013) Transformation groups, Lectures in Topol-
Smale S (1963) Dynamical systems and the topological ogy; the University of Michigan Conference of 1940.
conjugacy problem for diffeomorphisms. In: University of Michigan Press, Ann Arbor
 Hausdorff space containing ℕ characterizes
Ergodic Theory: Interactions bℕ up to homeomorphism.
with Combinatorics and Curvature An intrinsic measure of the curvature
Number Theory of a Riemannian manifold depending only on
the Riemannian metric; in the case of a surface
Tom Ward it determines whether the surface is locally
School of Mathematics, University of Leeds, convex (positive curvature), locally saddle-
Leeds, UK shaped (negative), or locally flat (zero).
Diophantine approximation Theory of the
approximation of real numbers by rational
Article Outline numbers: how small can the distance from a
given irrational real number to a rational num-
Glossary ber be made in terms of the denominator of the
Definition of the Subject rational?
Introduction Diophantine problem A system of equations for
Ergodic Theory which the set of integer solutions is sought.
Frequency of Returns The description of the set of solutions may be
Ergodic Ramsey Theory and Recurrence qualitative (asserting that the set is finite),
Orbit-Counting as an Analogous Development quantitative (bounding the number of solutions
Diophantine Analysis as a Toolbox in terms of parameters like degrees), or effec-
Future Directions tive (bounding the height of possible solutions,
Bibliography and as a result bounding both the number of
solutions and their location).
Glossary Dynamical zeta function For a map with fn
points of period n, the dynamical zeta function
Almost everywhere (abbreviated a.e.) A prop- is formally defined to be exp nP1 fnn zn for a
erty that makes sense for each point x in a complex variable z. Convergence in a disc of
measure space (X, ℬ, m) is said to hold almost positive radius is associated with exponential
everywhere (or a.e.) if the set N
X on which it bounds on the growth in the number of peri-
does not hold satisfies N  ℬ and m(N) = 0. odic points.
Baker’s theorem A lower bound for the absolute Equidistributed A sequence is equidistributed if
value of linear combinations of logarithms of the asymptotic proportion of time it spends in
algebraic numbers; this is a fundamental result an interval is proportional to the length of the
in transcendental number theory. interval.
Čech–Stone compactification of ℕ, bℕ A com- Ergodic A measure-preserving transformation is
pact Hausdorff space containing ℕ as a dense ergodic if the only invariant functions are equal
subset with the property that any map from ℕ to a constant almost everywhere (a.e.); equiv-
to a compact Hausdorff space K extends alently if the transformation exhibits the con-
uniquely to a continuous map bℕ ! K. This vergence in the quasi-ergodic hypothesis.
property and the fact that bℕ is a compact Ergodic theory The study of statistical proper-
ties of orbits in abstract models of dynamical
This entry gives a brief overview of some of the ways in
which number theory and combinatorics interact with
systems; more generally properties of
ergodic theory. The main themes are illustrated by measure-preserving (semi-)group actions on
examples related to recurrence, mixing, orbit counting, and measure spaces.
Diophantine analysis.

© Springer Science+Business Media, LLC, part of Springer Nature 2023 577

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_181
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_181-3
578 Ergodic Theory: Interactions with Combinatorics and Number Theory

Geodesic (flow) The shortest path between two infinite families of solutions arising from
points on a Riemannian manifold; such a geo- vanishing sub-sums).
desic path is uniquely determined by a starting Topological entropy A numerical invariant of
point and the initial tangent vector to the path topological dynamical systems that measures
(that is, a point in the unit tangent bundle). The the asymptotic growth in the complexity of
transformation on the unit tangent bundle orbits under iteration. The variational princi-
defined by flowing along the geodesic defines ple states that the topological entropy of a
the geodesic flow. topological dynamical system is the supremum
Haar measure (on a compact group) If G is a over all invariant measures of the measure-
compact topological group, the unique mea- theoretic entropies of the dynamical systems
sure m defined on the Borel sets of G with the viewed as measurable dynamical systems.
property that m(A + g) = m(A) for all g  G and
m(G) = 1.
Measure-theoretic entropy A numerical invari- Definition of the Subject
ant of measure-preserving systems that reflects
the asymptotic growth in complexity of measur- Number theory is a branch of pure mathematics
able partitions refined under iteration of the map. concerned with the properties of numbers in gen-
Mixing A measure-preserving system is mixing eral, and integers in particular. The areas of most
if measurable sets (events) become asymptoti- relevance to this entry are Diophantine analysis (the
cally independent as they are moved apart in study of how real numbers may be approximated by
time (under iteration). rational numbers, and the consequences for solu-
Orbit Dirichlet series For a map with on closed tions of equations in integers); analytic number
orbits of length n, the Dirichlet series nP1 onns theory, and in particular asymptotic estimates for
for a complex variable s. Convergence on a the number of primes smaller than X as a function of
right half-plane is associated with polynomial X; equidistribution, and questions about how the
bounds on orbit growth. digits of real numbers are distributed. Combinator-
Pólya-Carlson theorem If a complex power ics is concerned with identifying structures in dis-
series with integer coefficients has radius of crete objects; of most interest here is that part of
convergence 1, then either it defines a rational combinatorics connected with Ramsey theory,
function or it admits a natural boundary at its asserting that large subsets of highly structured
radius of convergence. objects must automatically contain large replicas
(Quasi) Ergodic hypothesis The assumption of that structure. Ergodic theory is the study of
that, in a dynamical system evolving in time asymptotic behavior of group actions preserving a
and preserving a natural measure, there are probability measure; it has proved to be a powerful
some reasonable conditions under which the part of dynamical systems with wide applications.
“time average” along orbits of an observable
(that is, the average value of a function defined
on the phase space) will converge to the “space Introduction
average” (that is, the integral of the function
with respect to the preserved measure). Ergodic theory, part of the mathematical study of
Recurrence Return of an orbit in a dynamical dynamical systems, has pervasive connections
system close to its starting point infinitely often. with number theory and combinatorics. This
S-unit theorems A circle of results stating that entry briefly surveys how these arise through a
linear equations in fields of zero characteristic small sample of results. Unsurprisingly, many
have only finitely many solutions taken from details are suppressed, and of course the selection
finitely-generated multiplicative subgroups of of topics reflects the author’s interests far more
the multiplicative group of the field (apart from than it does the full extent of the flow of ideas
Ergodic Theory: Interactions with Combinatorics and Number Theory 579

between ergodic theory and number theory. In this entry. The ideas and conditions surrounding
addition, the selection of topics has been chosen the quasi-ergodic hypothesis were eventually
in part to be complementary to those in related placed on a firm mathematical footing by devel-
entries in the Encyclopedia. A particularly enor- opments starting in 1890. For a single measure-
mous lacuna is the theory of a class of systems preserving transformation T : X ! X of a proba-
called “arithmetic dynamical systems” itself; bility space (X, ℬ, m), Poincaré (1890) showed a
these are rational functions that map a variety to recurrence theorem: if E  ℬ is any measurable
itself. The goal of the theory is to understand the set, then for a.e. x  E there is an infinite set of
arithmetic and geometry of orbits of points under return times, 0 < n1 < n2 <    with T nj ðxÞ  E
iteration, and (depending on the field over which (of course Poincaré noted this in a specific set-
the variety is defined) it has strong connections to ting, concerned with a natural invariant measure
algebraic and arithmetic geometry. The mono- for the “three-body” problem in planetary
graph by Silverman (2007) gives a comprehensive motion).
overview. Poincaré’s qualitative result was made quanti-
More sophisticated aspects of this connection – tative in the 1930s, when von Neumann (1932)
in particular the connections between ergodic the- used the approach of Koopman (1931) to show the
ory on homogeneous spaces and Diophantine mean ergodic theorem: if f  L2(m) then there is
analysis – are covered in the entries ▶ “Ergodic some f  L2 ðmÞ for which
Theory on Homogeneous Spaces and Metric
Number Theory” by Kleinbock and ▶ “Ergodic 1
N1

Theory: Rigidity” by Niţică; more sophisticated f ∘T n  f ! 0 as N ! 1;

N n¼0
overviews of the connections with combinatorics 2

may be found in the entry ▶ “Ergodic Theory:

clearly f then has the property that k f 
Recurrence” by Frantzikinakis and McCutcheon.
f ∘ Tk2 ¼ 0 and f dm ¼ f dm. Around the
same time, Birkhoff (1931) showed the more del-
icate pointwise ergodic theorem: for any
Ergodic Theory g  L1(m) there is some g  L1 ðmÞ for which

While the early origins of ergodic theory lie in the N1

1
quasi-ergodic hypothesis of classical Hamiltonian gðT n xÞ ! gðxÞ a:e:;
dynamics, the mathematical study of ergodic the- N n¼0
ory concerns various properties of group actions
on measure spaces, including but not limited to again it is then clear that gðTxÞ ¼ gðxÞ a.e. and
several special branches: g dm ¼ g dm.
The map T is called ergodic if the invariance
1. The classical study of single measure- condition forces the function ( f or g) to be equal
preserving transformations to a constant a.e. Thus an ergodic map has the
2. Measure-preserving actions of ℤd; more gen- property that the time or ergodic average
erally of countable amenable groups ð1=N Þ N1
n¼0 f ∘T converges to the space average
j
3. Measure-preserving actions of ℝd and more dm. An overview of ergodic theorems and their
general amenable groups, called flows many extensions may be found in the article
4. Measure-preserving actions of lattices in Lie “Ergodic Theorems” by del Junco (2008).
groups Thus ergodic theory at its most basic level
5. Measure-preserving actions of Lie groups makes strong statements about the asymptotic
We refer to the monograph of Einsiedler and behavior of orbits of a dynamical system as seen
the author (Einsiedler and Ward 2011) for an by observables (measurable functions on the
overview of the ergodic theory most relevant to space X ). Applying the ergodic theorem to the
580 Ergodic Theory: Interactions with Combinatorics and Number Theory

indicator function of a measurable set A shows This was eventually proved by Kuz’min
that ergodicity guarantees that a.e. orbit spends an (1928) and Lévy (1929), and the probability dis-
asymptotic proportion of time in A equal to the tribution of the digits is the Gauss–Kuz’min law.
volume m(A) of that set (as measured by the invari- Khinchin (1964) developed this further, showing
ant measure). This points to the start of the perva- for example that
sive connections between ergodic theory and
number theory – but as this and other articles lim ða1 ðxÞa2 ðxÞ . . . an ðxÞÞ1=n
n!1
relate, the connections extend far beyond this.
1 log n= log 2
ð n þ 1Þ 2
¼
n¼1
nðn þ 2Þ
Frequency of Returns
¼ 2:68545    for a:e: x:
In this section, we illustrate the way in which a
dynamical point of view may unify, explain, and Lévy (1936) showed that the denominator
extend quite disparate results from number qn(x) of the nth convergent qpn ððxxÞÞ (the rational
n

theory. obtained by truncating the continued fraction

expansion of x at the nth term) grows at a specific
Normal Numbers exponential rate,
Borel (1909) showed (as a consequence of what
became the Borel–Cantelli Lemma in probabil- 1 p2
lim log qn ðxÞ ¼ for a:e: x:
ity) that a.e. real number (with respect to n!1 n 12 log 2
Lebesgue measure) is normal to every base: that
is, has the property that any block of k digits in the There are many other expansions for real num-
base-r expansion appears with asymptotic fre- bers using modified continued fractions; we refer
quency rk. to the monograph of Schweiger (1995) for more
on this. There have also been several different
attempts to define multidimensional continued
Continued Fraction Digits
fractions, some aimed at very difficult problems
Analogs of normality results for the continued
concerned with simultaneous Diophantine
fraction expansion of real numbers were found
approximation. In contrast to the classical one-
by Khinchin, Kuz0 min, Lévy, and others. Any
dimensional (or single variable) case, it is less
irrational x  [0, 1] has a unique expansion as a
clear that there is one canonical algorithm that
continued fraction
emerges naturally. There is an extensive account
of many of the multidimensional algorithms in
1
x¼ another monograph of Schweiger (2000). Khanin
a1 ðxÞ þ a ðxÞþ1 1
2 a3 ðxÞþ et al. (2007), building on work of Lagarias (1994),
described a simple multidimensional continued
and, just as in the case of the familiar base-r fraction algorithm which is related to dynamics
expansion, it turns out that the digits (an(x)) obey on the homogeneous space SLd(ℝ)/SLd(ℤ). This
precise statistical rules for a.e. x. Gauss does give best possible approximations to an irra-
conjectured that the appearance of individual tional vector and is used to give theorems of KAM
digits would obey the law type with quasi-periodic behavior involving sev-
eral different frequencies (it is well-known that
1 the classical continued fraction expansion is inti-
j fk : 1OkON, ak ðxÞ ¼ jg j
N mately related to the geodesic flow on the space
2 log ð1 þ jÞ  log j  log ð2 þ jÞ SL2(ℝ)/SL2(ℤ); we refer to Einsiedler and Ward
! : ð1Þ
log 2 (2011) for the details.
Ergodic Theory: Interactions with Combinatorics and Number Theory 581

First Digits The Ergodic Context

The astronomer Newcomb (1881) noted that the All the results of this section are manifestations of
first digits of large collections of numerical data various kinds of convergence of ergodic averages.
that are not dimensionless have a specific and Borel’s theorem on normal numbers is an imme-
nonuniform distribution: diate consequence of the fact that Lebesgue mea-
The law of probability of the occurrence of numbers sure on [0, 1) is invariant and ergodic for the map
is such that all mantissæ of their logarithms are x 7! bx modulo 1 with b ⩾ 2. The asymptotic
equally probable. properties of continued fraction digits are all a
consequence of the fact that the Gauss measure
This is now known as Benford’s Law, following
his popularization and possible rediscovery of the defined by
phenomenon (Benford 1938). In both cases, this
1 dx
was an empirical observation eventually made mðAÞ ¼ for A
½0, 1
log 2 A þx
1
rigorous by Hill (1995). Arnold (Arnol’d and
Avez 1968, App. 12) pointed out the dynamics
is invariant and ergodic for the Gauss map
behind this phenomena in certain cases, best illus-
x 7! 1x , and the orbit of an irrational number
trated by the statistical behavior of the sequence 1,
2, 4, 8, 1, 3, 6, 1, . . . of first digits of powers of under the Gauss map determine the digits
2. Empirically, the digit 1 appears about 30% of appearing in the continued fraction expansion
the time, while the digit 9 appears about 5% of the much as the orbit under the map x 7! bx (mod1)
time. determines the digits in the base b expansion.
The results on equidistribution and the fre-
quency of first digits are related to ergodic
Equidistribution averaging of a different sort. For example,
Weyl (1910) (and, separately, Bohl (1909) and writing Ra(t) = t + a modulo 1 for the circle
Sierpiński (1910)) found an important instance rotation by a, the first digit of 2n is the digit j if
of equidistribution. Writing {} for the fractional and only if
part, a sequence (an) of real numbers is said to be
equidistributed modulo 1 if, for any interval log 10 jOR log 10 ð2Þ ð0Þ < log 10 ð j þ 1Þ:
[a, b]
[0, 1),
Thus the asymptotic frequency of appearance
1 concerns the orbit of a specific point. In order to
j fk : 1OkON, fak g  ½a, bg j
N see what this means, consider a continuous map
! ðb  aÞ as N ! 1; T : X ! X of a compact metric space (X, d). The
space M(T) of Borel probability measures on the
equivalently if Borel s-algebra of (X, d) is a non-empty compact
convex set in the weak -topology, each extreme
N 1
1 point is an ergodic measure for T, and these ergo-
f ð ak Þ ! f ðtÞ dt as N ! 1 dic measures are mutually singular. If M(T) is not
N k¼1 0
a singleton and m1, m2  M(T) are distinct ergo-
for all continuous functions f. Weyl showed that dic measures, then for a continuous function f with
the sequence {na} is equidistributed if and only if X f dm1 6¼ X f dm2, it is clear that the ergodic

a is irrational. This result was refined and averages N1 N1n¼0 f ðT xÞ must converge to X f dm1
n

extended in many directions; for example, a.e. with respect to m1 and to X f dm2 a.e. with
Hlawka (1964) and others found rates for the respect to m2. Thus the presence of many invariant
convergence in terms of the discrepancy of the measures for a continuous map means that ergodic
sequence, Weyl (1916) proved equidistribution averages along the orbits of specific points need
for {n2a}, and Vinogradov for {pna} where pn is not converge to the space average with respect to a
the nth prime. chosen invariant measure.
582 Ergodic Theory: Interactions with Combinatorics and Number Theory

In the extreme situation of unique ergodicity ℕ ¼ C1 t C2 t . . . t Cr , ð2Þ

(a single invariant measure, which is necessarily
an extreme point of M(T) and hence ergodic), the then there must be one set Cj that contains arbi-
convergence of ergodic averages is much more trarily long arithmetic progressions. That is, there
uniform. Indeed, if T is uniquely ergodic with is some j  {1, . . ., r} such that for any k ⩾ 1
M(T) = {m} then, for any continuous function there are a ⩾ 1 and n ⩾ 1 with
f : X ! ℝ,
a, a þ n, a þ 2n, . . . , a þ ðk  1Þn  Cj :
N1
1
f ðT n x Þ ! f dm uniformly in x The original proof appears in van der
N n¼0 X
Waerden’s paper (van der Waerden 1927), and
there is a discussion of how he found the proof
(see Oxtoby (1952)). The circle rotation Ra is
in van der Waerden (1971).
uniquely ergodic for irrational a, leading to the
Work of Furstenberg and Weiss (1978) and
equidistribution results.
others placed the theorem of van der Waerden in
The ergodic viewpoint on equidistribution also
the context of topological dynamics, giving alter-
places equidistribution results in a wider context.
native proofs. Specifically, van der Waerden’s the-
Weyl’s result that {n2a} (indeed, the fractional
orem is a consequence of topological multiple
part of any polynomial with at least one irrational
recurrence: the return of points under iteration in
coefficient) is equidistributed for irrational a was
a topological dynamical system close to their
given another proof by Furstenberg (1961) using
starting point along finite sequences of times. The
the notion of unique ergodicity. These methods
same approach readily gives dynamical proofs of
were then used in the study of nilsystems
Rado’s extension (Rado 1933) of van der
(translations on quotients of nilpotent Lie groups)
Waerden’s theorem, and of Hindman’s theorem
by Auslander et al. (1963) and Parry (1969), and
(Hindman 1974). The theorems of Rado and
these nilsystems play an essential role for polyno-
Hindman introduce a new theme: given a set
mial (and other non-conventional) ergodic aver-
A = {n1, n2, . . .} of natural numbers, write FS(A)
aging (see Host and Kra (2005a) and Leibman
for the set of numbers obtained as finite sums
(2005) in the polynomial case; Host and Kra
ni1 þ    þ nij with i1 < i2 <    < ij. Rado showed
(2005b) in the multiple linear case). Remarkably,
nilsystems are starting to play a role within that for any large n there is some Cs containing
combinatorics – an example is the work on the some FS(A) for a set A of cardinality n. Hindman
asymptotic number of 4-step arithmetic progres- showed that there is some Cs containing some
sions in the primes by Green and Tao (2010). FS(A) for an infinite set A.
Pointwise ergodic theorems have also been In the theorem of van der Waerden, it is clear
found along sequences other than ℕ, notably for that for any reasonable notion of “proportion” or
integer-valued polynomials and along the primes “density,” one of the sets Cj must occupy a posi-
for L2 functions by Bourgain (1988a, b). For more tive proportion of ℕ. A set A
ℕ is said to have
details, see the survey paper of del Junco (2008) positive upper density if there are sequences (Mi)
on ergodic theorems. and (Ni) with Ni  Mi ! 1 as i ! 1 such that

1
lim jfa  A : Mi < a < N i gj > 0:
i!1 N i  Mi
Ergodic Ramsey Theory and Recurrence
Erdős and Turán (1936) conjectured the stron-
In 1927, van der Waerden proved a conjecture ger statement that any subset of ℕ with positive
attributed to Baudet: if the natural numbers are upper density must contain arbitrary long arith-
written as a disjoint union of finitely many sets, metic progressions. This statement was shown for
Ergodic Theory: Interactions with Combinatorics and Number Theory 583

arithmetic progressions of length 3 by Roth Poincaré recurrence is the statement that ℕ is a set
(1952) in 1952, then for length 4 by Szemerédi of recurrence. Furstenberg and Katznelson (1978)
(1969) in 1969. The general result was eventually showed that if T1,   , Tk form a family of com-
proved by Szemerédi (1975) in 1975 in a lengthy muting measure-preserving transformations and
and extremely difficult argument. A is a set of positive measure, then
Furstenberg saw that Szemerédi’s Theorem
would follow from a deep extension of the 1
N1

Poincaré recurrence phenomena described in sec- liminf m T n n

1 A \    \ T k A > 0:
N!1 N
tion “Ergodic Theory,” and proved that extension n¼0

(Furstenberg 1977) (see also the survey article by

This remarkable multiple recurrence implies a
Furstenberg et al. (1982)). The multiple recur-
multidimensional form of Szemerédi’s theorem.
rence result of Furstenberg says that for any
Later, Gowers found a non-ergodic proof of this
measure-preserving system (X, ℬ, m, T) and set
(Gowers 2007).
A  ℬ with m(A) > 0, and for any k  ℕ,
Furstenberg also gave an ergodic proof of
N
Sárközy’s theorem (Sárközy 1978): if p  ℚ[t]
1 is a polynomial with p(ℤ)
ℤ and p(0) = 0, then
liminf
NM!1 N  M þ 1 {p(n)}n>0 is a set of recurrence. This was
n¼M
n 2n
m A\T A\T A \ . . . \ T kn A > 0: extended to multiple polynomial recurrence by
Bergelson and Leibman (1996).
An immediate consequence is that in the same
setting there must be some n ⩾ 1 for which Topology and Coloring Theorems
The existence of idempotent ultrafilters in the
m A \ T n A \ T 2n A \ . . . \ T kn A > 0: ð3Þ Čech–Stone compactification bℕ gives rise to an
algebraic approach to many questions in topolog-
A general correspondence principle, due to ical dynamics (this notion has its origins in the
Furstenberg, shows that statements in combina- work of Ellis (1969)). Using these methods,
torics like Szemerédi’s Theorem are equivalent to results like Hindman’s finite sums theorem find
statements in ergodic theory like (3). elegant proofs, and many new results in combina-
This opened up a significant new field of ergo- torics have been found. For example, in the parti-
dic Ramsey theory, in which methods from tion (2), there must be one set Cj containing a
dynamical systems and ergodic theory are used triple x, y, z solving x  y = z2.
to produce new results in infinite combinatorics. A deeper application is to improve a strength-
For an overview, see the entries ▶ “Ergodic The- ening of Kronecker’s theorem. To explain this,
ory on Homogeneous Spaces and Metric Number recall that a set S is called IP if there is a sequence
Theory” by Kleinbock, ▶ “Ergodic Theory: (ni) of natural numbers (which do not need to be
Rigidity” by Niţică, ▶ “Ergodic Theory: Recur- distinct) with the property that S contains all the
rence” by Frantzikinakis and McCutcheon, and terms of the sequence and all finite sums of terms
the survey articles of Bergelson (1996, 2000, of the sequence with distinct indices. A set S is
2006). The field is too large to give an overview called IP if it has non-empty intersection with
here, but a few examples will give a flavor of some every IP set, and a set S is called IPþ if there is
of the themes. some t  ℤ for which S  t is IP. Thus being IP
Call a set R
ℤ a set of recurrence if, for any (or IPþ ) is an extreme form of “fatness” for a set.
finite measure-preserving invertible transforma- Now let 1, a1, . . ., ak be numbers that are linearly
tion T of a finite measure space (X, ℬ, m) and independent over the rationals, and for any d  ℕ
any set A  ℬ with m(A) > 0, there are infinitely and kd non-empty intervals Iij
[0, 1]
many n  R for which m(A \ T nA) > 0. Thus, (1 O i O d, 1 O j O k), let
584 Ergodic Theory: Interactions with Combinatorics and Number Theory

D ¼ n  ℕ : ni aj  I ij for all i, j : (na)a  F taking values in an abelian group is called

an IP-sequence if na [ b = na + nb whenever
Kronecker showed that if d = 1 then D is non- a \ b = ∅. An IP-ring is a set of the form F (1) =
empty; Hardy and Littlewood showed that D is {[i  bai : b  F } where a1 < a2 <    is a
infinite, and Weyl showed that D has positive sequence in F , and a<b means a < b for all a
density. Bergelson (2003) uses these algebraic  a, b  b; write F m < for the set of m-tuples
methods to improve the result by showing that (a1,   , am) from F with ai < aj for i < j. Write
m

D is an IPþ set. PE(m, d) for the collection of all expressions of

ðbÞ
pi naj
r 1ObOk,1OjOm
the form T ða1 ,   , am Þ ¼ i¼1 T i ,
Polynomialization and IP-sets
ða1 ,   , am Þ  ðF [ ∅Þm < , where each p i is a poly-
As mentioned above, Bergelson and Leibman
nomial in a k  m matrix of variables with integer
(1996) extended multiple recurrence to a polyno-
coefficients and zero constant term with degree
mial setting. For example, let
Od. Then for every m, t  ℕ, there is an IP-ring
F (1), and an a = a(A, m, t, d) > 0, such that, for
pi,j : 1OiOk, 1OjOt
every set of polynomial expressions {S0,   , St}

PE(m, d),
be a collection of polynomials with rational coef-
ficients and pi, j(ℤ)
ℤ, pi,j(0) = 0. Then if
t
m(A) > 0, we have IP  lim m m \ S i ða1 ,  , am Þ1 A > a:
ða1 ,, am Þ  ðF ð1Þ Þ i¼0
<

N k t 1
1 pi,j ðnÞ
liminf m Tj A > 0: There are a large number of deep combinatorial
N !1 N
n¼1 i¼1 j¼1 consequences of this result, not all of which seem
accessible by other means.
Using the Furstenberg correspondence princi-
ple, this gives a multi-dimensional polynomial Sets of Primes
Szemerédi theorem: If P : ℤr ! ℤ‘ is a polynomial In a remarkable development, Szemerédi’s theo-
mapping with the property that P(0) = 0, and rem and some of the ideas behind ergodic Ramsey
F
ℤr is a finite configuration, then any set theory joined results of Goldston et al. (2009) in
S
ℤ‘ of positive upper Banach density contains playing a part in Green and Tao’s proof (Green
a set of the form u + P(nF) for some u  ℤ‘ and and Tao 2008) that the set of primes contains
n  ℕ. arbitrarily long arithmetic progressions. This pro-
In a different direction, motivated in part by found result is surveyed from an ergodic point of
Hindman’s theorem, the multiple recurrence view in the article of Kra (2006). As with
results generalize to IP-sets. Furstenberg and Szemerédi’s theorem itself, this result has been
Katznelson (1985) proved a linear IP-multiple extended to a polynomial setting by Tao and
recurrence theorem in which the recurrence is Ziegler (2008). Given integer-valued polynomials
guaranteed to occur along an IP-set. f1, . . ., fk  ℤ[t] with
A combinatorial proof of this result has been
found by Nagle et al. (2006). Bergelson and f 1 ð 0Þ ¼    ¼ f k ð 0Þ ¼ 0
McCutcheon (2000) extended these results by
proving a polynomial IP-multiple recurrence the- and any e > 0, Tao and Ziegler proved that there
orem. To formulate this, make the following def- are infinitely many integers x, m with 1 O m O xe
initions. Write F for the family of non-empty for which x + f1(m), . . ., x + fk(m) are primes. We
finite subsets of ℕ, so that a sequence indexed refer to a survey of Ziegler for an overview of
by F is an IP-set. More generally, an F -sequence these developments (Ziegler 2014).
Ergodic Theory: Interactions with Combinatorics and Number Theory 585

Orbit-Counting as an Analogous were symmetric about ℜðsÞ ¼ 12 . The Riemann

Development hypothesis asserts that zeros in the region
0 < ℜ(s) < 1 all lie on the line ℜðsÞ ¼ 12 , and
Some of the connections between number theory this remains open.
and ergodic theory arise through developments Analytic properties of the Riemann zeta func-
that are analogous but not directly related. tion were used by Hadamard and de la Vallée
A remarkable instance of this concerns the long Poussin to prove (4), the Prime Number Theorem,
history of attempts to count prime numbers laid in 1896. Tauberian methods developed by Wiener
alongside the problem of counting closed orbits in and Ikehara (Wiener 1932) later gave different
dynamical systems. approaches to the Prime Number Theorem.
These ideas initiated the widespread use of
Counting Orbits and Geodesics zeta functions in several parts of mathematics,
Consider first the fundamental arithmetic function but it was not until the middle of the twentieth
p(X) = j{p O X : p is prime}j. Tables of primes century that Selberg (1956) introduced a zeta
prepared by Felkel and Vega in the eighteenth function dealing directly with quantities arising
century led Legendre to suggest that p(X) is in dynamical systems: the lengths of closed geo-
approximately x/(log(X)  1.08). Gauss, using desics on surfaces of constant curvature 1. The
both computational evidence and a heuristic argu- geodesic flow acts on the unit tangent bundle to
ment, suggested that p(X) is approximated by the surface by moving a point and unit tangent
vector at that point along the unique geodesic
X
dt they define at unit speed. Closed geodesics are
liðXÞ ¼ : then in one-to-one correspondence with periodic
2 log t
orbits of the associated geodesic flow on the unit
Both of these suggestions imply the well- tangent bundle, and it is in this sense that the
known asymptotic formula quantities are dynamical. The function defined
by Selberg takes the form
X
pðXÞ  : ð4Þ 1
log X
Z ðsÞ ¼ 1  eðsþkÞjtj ,
t k¼0
Riemann brought the analytic ideas of
Dirichlet and Chebyshev (who used the zeta func- in which t runs over all the closed geodesics, and
tion to find a weaker version of (4) with upper and jtj denotes the length of the geodesic. In a direct
lower bounds for the quantity pðXÞ log
X
ðX Þ
) to bear by echo of the Riemann zeta function, Selberg found
proposing that the zeta function an analytic continuation to the complex plane and
showed that the zeros of Z lie on the real axis or on
1
1 the line ℜðsÞ ¼ 12 (the analogue of the Riemann
zð s Þ ¼ ¼ ð1  ps Þ1 , ð5Þ
n¼1
ns p hypothesis for Z; see also the paper of Hejhal
(1976)). The zeros of Z are closely connected to
already studied by Euler, would connect proper- the eigenvalues for the Laplace–Beltrami operator,
ties of the primes to analytic methods. An essen- and thus give information about the lengths of closed
tial step in these developments, due to Riemann, is geodesics via Selberg’s trace formula in the same
the meromorphic extension of ζ from the region paper. Huber (1959) and others used this approach to
ℜ(s) > 1 in (5) to the whole complex plane and a give an analogue of the prime number theorem for
functional equation relating the value of the exten- closed geodesics – a prime orbit theorem.
sion at s to the value at 1  s. Moreover, Riemann Sinaĭ (1966) considered closed geodesics on a
showed that the extension has readily understood manifold M with negative curvature bounded
real zeros, and that all the other zeros he could find between R2 and r2, and found the bounds
586 Ergodic Theory: Interactions with Combinatorics and Number Theory

log pðT Þ constructed Markov partitions, allowing the

ðdimðMÞ  1ÞrO lim inf
T!1 T dynamics of the toral automorphism to be
log pðT Þ modeled by a topological Markov shift, and used
O lim sup OðdimðMÞ  1Þ R this to determine when two such automorphisms
T!1 T
are measurably isomorphic. Sinaĭ (1968), Ratner
for the number p(T) of closed geodesics of multi- (1973), Bowen (1970, 1973), and others devel-
plicity one with length less than T, analogous to oped the construction of Markov partitions in gen-
Chebyshev’s result. eral for Anosov diffeomorphisms and flows.
The essential dynamical feature behind the Around the same time, Smale (1967) intro-
geodesic flow on a manifold of negative curva- duced a more permissive hyperbolicity axiom for
ture is that it is an example of an Anosov flow diffeomorphisms, Axiom A. Maps satisfying
(Anosov 1967). These are smooth dynamical Axiom A are diffeomorphisms satisfying the
ℝ-actions (equivalently, first-order differential same hypothesis as that of Anosov
equations on Riemannian manifolds) with the diffeomorphisms, but only on the set of points
property that the tangent bundle has a continu- that return arbitrarily close under the action of
ously varying splitting into a direct sum the flow (or iteration of the map).
Eu Es Eo and the action of the differential Thus Markov partitions, and with them associ-
of the flow acts on Eu as an exponential expan- ated transfer operators, became a substitute for the
sion, on Es as an exponential contraction, Eo is geometrical Laplace–Beltrami operators of the
the one-dimensional bundle of vectors that are setting considered by Selberg. Bowen (1972)
tangent to orbits, and the expansion and contrac- extended the uniform distribution result of
tion factors are bounded. In the setting of Anosov Margulis to this setting and found an analogue of
flows, the natural orbit counting function is Chebychev’s theorem for closed orbits. Parry
p(X) = j {t : t a closed orbit of length |t| (1983) (in a restricted case) and Parry and
OX} j. Margulis (1969, 2004) generalized the Pollicott (1983) went on to prove the prime orbit
picture to weak-mixing Anosov flows by show- theorem in this more general setting. The methods
ing a prime orbit theorem of the form are an adaptation of the Ikehara–Wiener
Tauberian approach to the prime number theorem.
ehtop X Thus many facets of the prime number theorem
pðXÞ  ð6Þ story find their echoes in the study of closed orbits
htop X
for hyperbolic flows: the role played by meromor-
for the counting function p(X) = j{t : t a phic extensions of suitable zeta functions,
closed orbit of length |t| OX}j where as before Tauberian methods, and so on. Moreover, related
htop denotes the topological entropy of the flow. results from number theory have analogues in
Integral to Margulis’ work is a result on the spatial dynamics, for example, Mertens’ theorem
distribution of the closed geodesics reflected in a (Mertens 1874) in the work of Sharp (1991) and
flow-invariant probability measure, now called Noorani (1999) and Dirichlet’s theorem in work
the Margulis measure. of Parry (1984).
Anosov also studied discrete dynamical sys- The “elementary” proof (not using analytic
tems with similar properties: diffeomorphisms of methods) of the prime number theorem by Erdös
compact manifolds with a similar splitting of the (1949) and Selberg (1949) (see the survey by
tangent space (though in this setting Eo disap- Goldfeld (2004) for the background to the results,
pears). The archetypal Anosov diffeomorphism and the unfortunate priority dispute) has an echo
is a hyperbolic toral automorphism of the sort in some approaches to orbit-counting problems
considered in section “Orbit Growth and Conver- from an elementary (non-Tauberian) perspective,
gence”; for such automorphisms of the 2-torus including work of Lalley (1988) on special flows
Adler and Weiss (Adler and Weiss 1970) and Everest et al. (2007) in the algebraic setting.
Ergodic Theory: Interactions with Combinatorics and Number Theory 587

In a different direction, Lalley (1987) found Counting Orbits for Group Endomorphisms
orbit asymptotics for closed orbits satisfying con- A prism through which to view some of the deeper
straints in the Axiom A setting without using issues that arise in section “Counting Orbits and
Tauberian theorems. His more direct approach is Geodesics” is provided by group endomorphisms.
still analytic, using complex transfer operators The price paid for having simple closed formulas
(the same objects used by Parry and Pollicott to for all the quantities involved is of course a severe
study the dynamical zeta function at complex loss of generality, but the diversity of examples
values) and indeed somewhat parallels a Tauberian illustrates many of the phenomena that may be
argument. expected in more general settings when hyper-
Further resonances with number theory arise bolicity is lost.
here. For example, there are results on the distri- Consider an endomorphism T : X ! X of a
bution of closed orbits for group extensions compact group with the property that
(analogous to Chebotarev’s theorem) and for
orbits with homological constraints (see Sharp Fn ðT Þ ¼j fx  X : T n x ¼ xg j< 1
(1993), Katsuda and Sunada (1990)).
Of course the great diversity of dynamical sys- for all n ⩾ 1. The number of closed orbits of
tems subsumed in the phrase “prime orbit theo- length n under T is then
rem” creates new problems and challenges, and in
particular if there is not much geometry to work 1
O n ðT Þ ¼ mðn=dÞFd ðT Þ: ð7Þ
with then the reliance on Markov partitions and n
djn
transfer operators makes it difficult to find higher-
order asymptotics. In simple situations (hyperbolic toral automor-
Dolgopyat (1998) has nonetheless managed to phisms for example), it is straightforward to show
push the Markov methods to obtain uniform that
bounds on iterates of the associated transfer oper-
ators to the region ℜ(s) > s0 with s0 < 1. This pT ðX Þ ¼
result has wide implications; an example most j ft : t a closed orbit under T of lengthOX g j
relevant to the analogy with number theory is the
eðX þ1Þhtop ðT Þ
work of Pollicott and Sharp (1998) in which  :
X
Dolgopyat’s result is used to show that for certain
(8)
geodesic flows there is a two-term prime orbit
theorem of the form Waddington (1991) considered quasi-
hyperbolic toral automorphisms, showing that
pðXÞ ¼ li ehtop X þ O ecX the asymptotic (8) in this case is multiplied by an
explicit almost-periodic function bounded away
for some c < htop. from zero and infinity.
For non-positive curvature manifolds less is This result has been extended further into non-
known: Knieper (1997) finds upper and lower hyperbolic territory, which is most easily seen via
bounds for the function counting closed geodesics the so-called connected S-integer dynamical sys-
on rank-1 manifolds of non-positive curvature of tems introduced by Chothi et al. (1997). Fix an
the form algebraic number field  with set of places PðÞ
and set of infinite places P1 ðÞ , an element of
ehX infinite multiplicative order x   , and a finite set
A OpðXÞOBehX S
PðÞ∖P1 ðÞ with the property that |x|w O 1
X
for all w 2= S [ P1 ðÞ . The associated ring of
for constants A, B > 0. S-integers is
588 Ergodic Theory: Interactions with Combinatorics and Number Theory

RS ¼ x   : jxjw O1 for all w 2

= S [ P1 ð  Þ : it is shown in Chothi et al. (1997) that for an
ergodic S-integer map T with  ¼ ℚ and S finite,
Let X be the compact character group of RS, there are constants kT  ℚ and CT such that
and define the endomorphism T : X ! X to be the
dual of the map x 7! xx on RS. Following Weil MT ðN Þ ¼ kT log N þ CT þ Oð1=N Þ:
(1967), write w for the completion at w, and for
w finite, write rw for the maximal compact sub- Without the restriction that  ¼ ℚ, it is shown
ring of w . Notice that if S = P then RS ¼  and that there are constants kT  ℚ, CT , and d > 0 with
Fn(T) = 1 for all n ⩾ 1 by the product formula for
 -fields. As the set S shrinks, more and more MT ðN Þ ¼ kT log N þ CT þ O N d :
periodic orbits come into being, and if S is as
small as possible (given x) then the resulting The case of S co-finite (that is, containing all
system is (more or less) hyperbolic or quasi- but finitely many primes) gives rise to a large class
hyperbolic. of group automorphisms with a polynomial bound
For S finite, it turns out that there are still on the growth of the number of closed periodic
sufficiently many periodic orbits to have the orbits. The natural analytical tool here is the orbit
growth rate result (10), but the asymptotic (8) is Dirichlet series
modified in much the same way as Waddington
observed for quasi-hyperbolic toral OT ðnÞ
dT ð z Þ ¼ ,
automorphisms: nP1
nz

XpT ðXÞ and a suitable form of ‘Dirichlet rationality’ and

lim inf >0 ð9Þ
X!1 eðXþ1Þhtop ðT Þ relations between its analytical behaviour near its
abscissa of convergence and orbit growth proper-
and there is an associated pair (X , aT), where X  ties are shown in work of Everest et al. (2010).
is a compact group and aT  X , with the prop-
erty that if aT j converges in X  as j ! 1, then
N

there is convergence in (9). Exotic Orbit Growth

A simple special case will illustrate this. Tak- The results above raise the question of what the
ing  ¼ ℚ, x = 2, and S = {3} gives a compact possible orbit growth phenomena are for compact
group endomorphism T with group automorphisms. A consequence of Linnik’s
bounds for first appearance of primes in arithmetic
Fn ðT Þ ¼ ð2n  1Þj2n  1j3 : progressions (Linnik 1944) can be used (Ward
2005) to prove that for any C  [0, 1] there is
For this example, the results of Chothi et al. a compact group automorphism T with the prop-
(1997) are sharper: The expression in (9) con- erty that
verges along (Xj) if and only if 2Xj converges in
the ring of 3-adic integers ℤ3, the expression has 1
log Fn ðT Þ ! C
n
uncountably many limit points, and the upper and
lower limits are transcendental. as n ! 1. This construction was further refined
Similarly, the dynamical analogue of Mertens’ by Haynes and White (2014) to exhibit other
theorem found by Sharp may be found for possible growth rates. The number-theoretic ques-
S-integer systems with S finite. Writing tions thrown up here are subtle ones, and the
constructions are unsatisfying in that the underly-
1
M T ðN Þ ¼ hðT Þjtj
, ing compact group is totally disconnected and the
jtjON
e group automorphism is not ergodic with respect to
Ergodic Theory: Interactions with Combinatorics and Number Theory 589

Haar measure. On connected groups, the S-integer theory in a direct way; we illustrate this by
construction has been used by Baier et al. (2013) describing a selection of dynamical problems
to show that the following “exotic” orbit growth that call on particular parts of number theory in
properties occur. an essential way. The example of mixing in sec-
• For any k  (0, 1), there is an ergodic com- tion “Mixing and Additive Relations in Fields” is
pact connected group automorphism T : X ! X particularly striking for two reasons: the results
with MT (N )  k log N. needed from number theory are relatively recent,
• For any r  ℕ and k > 0, there is an ergodic and the ergodic application directly motivated a
compact connected group automorphism further development in number theory.
T : X ! X with MT (N)  k(log log N )r.
• For any d  (0, 1) and k > 0, there is an
Orbit Growth and Convergence
ergodic compact connected group automorphism
The analysis of periodic orbits – how their number
T : X ! X with MT (N)  k(logN )d.
grows as the length grows and how they spread
We refer to the survey of Miles et al. (2015) for
out through space – is of central importance in
an overview.
dynamics (see Katok (1980) for example). An
instance of this is that for many simple kinds of
Pólya–Carlson Dichotomy dynamical systems T : X ! X (where T is a con-
Everest et al. (2005) noticed that the very simplest tinuous map of a compact metric space ðX, dÞ), the
example of a nontrivial S-integer system, namely, logarithmic growth rate of the number of periodic
the automorphism dual to x 7! 2x on ℤ 16 , had points exists and coincides with the topological
the property that its dynamical zeta function entropy h(T ) (an invariant giving a quantitative
admitted a natural boundary at its circle of con- measure of the average rate of growth in orbit
vergence. This phenomenon later emerged as a complexity under T ). That is,
pervasive feature of group automorphisms, and
Bell et al. (2014) formulated the conjecture that 1
log Fn ðT Þ ! htop ðT Þ ð10Þ
compact group automorphisms exhibit a Pólya– n
Carlson dichotomy: their dynamical zeta function
for many of the simplest dynamical systems. For
is either rational or admits a natural boundary.
example, if X ¼ r is the r-torus and T = TA is the
This has been proved in many cases with S or its
complement finite, but in general remains open. automorphism of the torus corresponding to a
matrix A in GLr(ℤ), then TA is ergodic with respect
This property certainly does not hold for maps in
to Lebesgue measure if and only if no eigenvalue
general, as it is easy to find examples of maps
whose dynamical zeta function is a nonrational of A is a root of unity. Under this assumption, we
have
function satisfying an algebraic identity.
Similar questions arise for endomorphisms of r
abelian varieties, and in positive characteristic, the Fn ðT A Þ ¼ j lni  1 j
relationship between rationality, algebraicity, and i¼1
the existence of natural boundaries for the dynam-
ical zeta function has been studied by Byszewski and
and Cornelissen (2018), where once again a
r
dichotomy of Pólya–Carlson type is found.
htop ðT A Þ ¼ log max f1, jli jg ð11Þ
i¼1

Diophantine Analysis as a Toolbox where l1, . . ., lr are the eigenvalues of A. It

follows that the convergence in (10) is clear
Many problems in ergodic theory and dynamical under the assumption that TA is hyperbolic (that
system exploit ideas and results from number is, no eigenvalue has modulus one). Without this
590 Ergodic Theory: Interactions with Combinatorics and Number Theory

assumption, the convergence is less clear: for raised the (still open) question of whether any
r ⩾ 4 the automorphism TA may be ergodic with- measure-preserving transformation can be mixing
out being hyperbolic. That is, while no eigen- without being mixing of all orders.
values are unit roots some may have unit A class of group actions that are particularly
modulus. As pointed out by Lind (1982) in his easy to understand are the algebraic dynamical
study of these quasihyperbolic automorphisms, systems studied systematically by Schmidt
the convergence (10) does still hold for these (1995): here X is a compact abelian group, each
systems, but this requires a significant Tg is a continuous automorphism of X, and m is the
Diophantine result (the theorem of Gel’fond Haar measure on X. Schmidt (1989) related mixing
(1960) suffices; one may also use Baker’s theorem properties of algebraic dynamical systems with
(Baker 1990)). Even further from hyperbolicity lie G = ℤd to statements in arithmetic and showed
the family of S-integer systems (Chothi et al. that a mixing action on a connected group could
1997; Ward 1998); their orbit-growth properties only fail to mix in a certain way. Later Schmidt and
are intimately tied up with Artin’s conjecture on the author (Schmidt and Ward 1993) showed that
primitive roots and prime divisors of linear recur- for X connected, mixing implies mixing of all
rence sequences. orders. The proof proceeds by showing that the
result is exactly equivalent to the following state-
ment: if  is a field of characteristic zero, and G is a
Mixing and Additive Relations in Fields
finitely generated subgroup of the multiplicative
The problem of higher-order mixing for commut-
group  , then the equation
ing group automorphisms provides a striking
example of the dialogue between ergodic theory
a1 x 1 þ    þ an x n ¼ 1 ð13Þ
and number theory, in which deep results from
number theory have been used to solve problems
for fixed a1 , . . . , an   has a finite number of
in ergodic theory, and questions arising in ergodic
solutions x1, . . ., xn  G for which no subsum
theory have motivated further developments in
i  Iaixi with I ⊊ {1, . . ., n} vanishes. The
number theory.
bound on the number of solutions to (13) follows
An action T of a countable group G on a prob-
from the profound extensions to W. Schmidt’s
ability space (X, ℬ, m) is called k-fold mixing or
subspace theorem in Diophantine geometry
mixing on (k + 1) sets if
(Schmidt 1972) by Evertse and Schlickewei (see
Evertse and Schlickewei (1999), van der Poorten
m A0 \ T g1 A1 \ . . . \ T gk Ak ! mðA0 Þ  mðAk Þ
and Schlickewei (1991), and Schlickewei (1990))
ð12Þ for the details).
The argument in Schmidt and Ward (1993)
as may be cast as follows: failure of k-fold mixing
in a connected algebraic dynamical system
gi g1
j ! 1 for i 6¼ j implies (via duality) an infinite set of solutions
to an equation of the shape (13) in some field of
with the convention that g0 = 1G, for any sets characteristic zero. The S-unit theorem means that
A0, . . ., Ak  ℬ; gn ! 1 in G means that for this can only happen if there is some proper sub-
any finite set F
G there is an N with n > N ) gn 2
= F. sum that vanishes infinitely often. This infinite
For k = 1 the property is called simply mixing. family of solutions to a homogeneous form of
This notion for single transformations goes back to (13) with fewer terms can then be translated
the foundational work of Rohlin (1949), where he back via duality to show that the system fails to
showed that ergodic group endomorphisms are mix for some strictly lower order, proving that
mixing of all orders (and so the notion is not mixing implies mixing of all orders by induction.
useful for distinguishing between group endomor- Stronger Diophantine results allow these argu-
phisms as measurable dynamical systems). He ments to be extended in various directions. Miles
Ergodic Theory: Interactions with Combinatorics and Number Theory 591

and Ward (2006) used explicit bounds on the x  A \ T ð2k ,0Þ A ) x  T ð0,2k Þ A,
number of solutions of S-integer systems due to
Evertse et al. (2002) to show that mixing implies so
mixing of all orders for actions of ℚd by automor-
phisms of a compact connected abelian group. A \ T ð2k ,0Þ A \ T ð0,2k Þ ðA þ x Þ ¼ ∅
Mixing properties for algebraic dynamical sys-
tems without the assumption of connectedness are
quite different, and in particular, it is possible to for all k ⩾ 1, which shows that T cannot be mixing
have mixing actions that are not mixing of all on three sets. Arenas-Carmona et al. (2008)
orders. This is a simple consequence of the fact showed that if the dilates of the specific shape
that the constituents of a disconnected algebraic {(0, 0), (1, 0), (0, 1)} are avoided, then
dynamical system are associated with fields of Ledrappier’s system exhibits mixing of all orders.
positive characteristic, where the presence of the The full picture of higher-order mixing proper-
Frobenius automorphism can prevent higher- ties on disconnected groups is rather involved; see
order mixing. Ledrappier (1978) pointed this out Schmidt’s monograph (Schmidt 1995). A simple
via examples of the following shape. Let illustration is the construction by Einsiedler and
the author (Einsiedler and Ward 2003) of systems
with any prescribed order of mixing. When such
X¼
systems fail to be mixing of all orders, they fail in a
2
xℤ2 : xðaþ1,bÞ þ xða,bÞ þ xða,bþ1Þ ¼ 0 ðmod 2Þ very specific way – along dilates of a specific shape
(a finite subset of ℤd). One insight into the range of
and define the ℤ2-action T to be the natural shift possible behaviors for algebraic dynamical systems
action, of this is seen in the following result. Call a shape
“admissable” if it does not lie on a line in ℤd, does
T ðn,mÞ x ¼ xðaþn,bþmÞ : contain 0, and has the property that for any k > 1 the
ða,bÞ
set 1k S contains nonintegral points. Then it is shown
in Ward (1997) that if S and T are admissable
It is readily seen that this action is mixing
shapes, then there is an algebraic ℤd-action that is
with respect to the Haar measure. The condition
mixing on S and not mixing on T unless a translate
x(a+1,b) + x(a,b) + x(a,b+1) = 0 (mod2) implies that,
of T is a subset of S. In Ledrappier’s example above,
for any k ⩾ 1,
the shape that fails to mix is {(0, 0), (1, 0), (0, 1)}.
This gives an order of mixing as detected by shapes;
2k
2k computing this is in principle an algebraic problem.
xð0,2k Þ ¼ xð j,0Þ
j¼0 j On the other hand, there is a more natural definition
of the order of mixing, namely, the largest k for
¼ xð0,0Þ þ xð2k ,0Þ ðmod 2Þ ð14Þ
which (12) holds; computing this is in principle a
Diophantine problem. A conjecture emerged
since every entry in the 2kth row of Pascal’s trian- (formulated explicitly by Schmidt (2001)) that for
gle is even apart from the first and the last. Now let any algebraic dynamical system, if every set of
A = {x  X : x(0,0) = 0} and let x  X be any cardinality r ⩾ 2 is a mixing shape, then the system
element with x(0,0) = 1. Then X is the disjoint is mixing on r sets.
union of A and A + x, so This question motivated Masser (2004) to prove
an appropriate analogue of the S-unit theorem on
1 the number of solutions to (13) in positive charac-
m ð A Þ ¼ mð A þ x  Þ ¼ :
2 teristic as follows. Let H be a multiplicative group
and fix n  ℕ. An infinite subset A
Hn is called
However, (14) shows that
broad if it has both of the following properties:
592 Ergodic Theory: Interactions with Combinatorics and Number Theory

• if h  H and 1 O j O n, then there are at most

finitely many (a1, . . ., an) in A with aj = h; f 1 ðT n1 xÞ f k ðT nk xÞdmðxÞ  f 1 dm  f k dm
X X X
• if n ⩾ 2, h  H and 1 O i < j O n then there k
are at most finitely many (a1, . . ., an)  H with OFk min k ni  nj k k f i kS
1Oi<jOk
ai a1
j ¼ h. i¼1

Then Masser’s theorem says the following. Let

 be a field of characteristic p > 0, let G be a for all f 1 , . . . , f k  S? Mixing of all orders means
finitely generated subgroup of  and suppose there is such a function satisfying Fk(n) ! 0 as
that the equation n ! 1 for S ¼ L1 ðXÞ, and a “rate of mixing”
result is any improvement on this, typically for a
a1 x 1 þ    þ an x n ¼ 1 much more restricted class of functions. Miles and
the author (Waldschmidt 1993) used Baker’s the-
has a broad set of solutions (x1, . . ., xn)  Gn for orem and its p-adic variants to establish a very
some constants a1 , . . . , an   . Then there is an weak general result of this shape under the
m O n, constants b1 , . . . , bm   and some assumption that X has finite topological dimen-
(g1, . . ., gm)  Gm with the following properties: sion, but in a form that makes it extremely difficult
• gj 6¼ 1 for 1 O j O m; to be explicit about Fk even in concrete examples.
• gi g1 Using more delicate Diophantine estimates for
j 6¼ 1 for 1 O i < j O m;
linear forms in logarithms of algebraic numbers
• there are infinitely many k for which
due to Waldschmidt (1993), Gorodnik and
Spatzier (2015) found explicit rates of mixing
b1 gk1 þ b2 gk2 þ    þ bm gkm ¼ 1:
for k = 3 in the setting of nilmanifolds, a new
result even in the toral case. To state their main
The proof that shapes detect the order of
result, let T be a ℤd-action by automorphisms of a
mixing in algebraic dynamics then proceeds
compact nilmanifold X = G/G with the property
much as in the connected case.
that each transformation Tn for n 6¼ 0 is ergodic
The extent to which all this behavior is a spe-
with respect to the measure m induced by Haar
cial feature of algebraic dynamical systems is
measure on G. Then they construct a function
illustrated by a construction in Ward (1997) of a
y 7! (y) > 0 such that if Cy(X) denotes the
measure-preserving ℤd-action T on a probability
class of Hölder functions on X with exponent y
space (X, ℬ, m) for any d ⩾ 2 with the following
and Hölder norm k  ky, then there is a constant
two properties:
C > 0 such that
• T is rigid, meaning that there is a sequence
(nj) going to infinity in ℤd as j ! 1 with the
property that m A \ T nj A ! mðAÞ as j ! 1 for f ðT l xÞg ðT m xÞhðT n xÞdmðxÞ
all A  ℬ (an extreme form of non-mixing);
• Every shape S
ℤd is mixing for T. f dm gdm hdm
The construction uses ideas of Ferenczi and
Kamiński (1995) and is built as a Gaussian OC ðexpðminfk l  m k , k l  n k , k n  m kgÞÞðyÞ
dynamical system. k f ky k gky khky
A natural question for mixing algebraic
dynamical systems that are mixing concerns the for all f, g, h  Cy(X).
rate at which the mixing occurs. That is, for a
suitable class S of functions X ! ℂ with a natural
norm k  kS on a compact group X carrying a ℤd- Future Directions
action T that is known to be mixing of all orders,
what can be said for each k  ℕ about functions The interaction between ergodic theory, number
Fk satisfying theory, and combinatorics continues to expand
Ergodic Theory: Interactions with Combinatorics and Number Theory 593

rapidly, and many future directions of research are Anosov DV (1967) Geodesic flows on closed Riemannian
discussed in the entries ▶ “Ergodic Theory on manifolds of negative curvature. Trudy Mat Inst
Steklov 90:209
Homogeneous Spaces and Metric Number The- Arenas-Carmona L, Berend D, Bergelson V (2008)
ory” by Kleinbock, ▶ “Ergodic Theory: Rigidity” Ledrappier’s system is almost mixing of all orders.
by Niţică, and ▶ “Ergodic Theory: Recurrence” Ergodic Theory Dynam Systems 28(2):339–365
by Frantzikinakis and McCutcheon. Some of Auslander L, Green L, Hahn F (1963) Flows on homoge-
neous spaces. With the assistance of L. Markus and
the directions most relevant to the examples W. Massey, and an appendix by L. Greenberg. Annals
discussed in this entry include the following. of Mathematics Studies, No. 53. Princeton University
The recent developments mentioned in section Press, Princeton
“Sets of Primes” clearly open many exciting pros- Baier S, Jaidee S, Stevens S, Ward T (2013) Automor-
phisms with exotic orbit growth. Acta Arith 158(2):
pects involving finding new structures in arith- 173–197
metically significant sets (like the primes). The Baker A (1990) Transcendental number theory, 2nd edn.
original conjecture of Erdős and Turán (1936) Cambridge mathematical library. Cambridge Univer-
asked if a  A
ℕ 1a ¼ 1 is sufficient to force the sity Press, Cambridge
Bell J, Miles R, Ward T (2014) Towards a Pólya–Carlson
set A to contain arbitrary long arithmetic progres- dichotomy for algebraic dynamics. Indag Math
sions and remains open. This would of course (NS) 25(4):652–668
imply both Szemerédi’s theorem (Szemerédi Benford F (1938) The law of anomalous numbers. Proc
Am Philos Soc 78:551–572
1975) and the result of Green and Tao (2008) on
Bergelson V (2003) Minimal idempotents and ergodic
arithmetic progressions in the primes. More gen- Ramsey theory. In: Topics in dynamics and ergodic
erally, it is clear that there is still much to come theory, volume 310 of London Math. Soc. Lecture
from the dialogue subsuming the four parallel Note Ser. Cambridge University Press, Cambridge,
pp 8–39
proofs of Szemerédi’s: one by purely combinato-
Bergelson V, Leibman A (1996) Polynomial extensions of
rial methods, one by ergodic theory, one by hyper- van der Waerden’s and Szemerédi’s theorems. J Am
graph theory, and one by Fourier analysis and Math Soc 9(3):725–753
additive combinatorics. For an overview, see the Bergelson V, McCutcheon R (2000) An ergodic IP poly-
nomial Szemerédi theorem. Mem Amer Math Soc
survey papers of Tao (2006, 2007a, b).
146(695):viii+106
In the context of the orbit-counting results in Birkhoff GD (1931) Proof of the ergodic theorem. Proc
section “Orbit-Counting as an Analogous Devel- Natl Acad Sci U S A 17:656–660
opment,” a natural problem is to, on the one hand, Bohl P (1909) Über ein in der Theorie der säkularen
Störungen vorkommendes Problem. J Undergrad
obtain finer asymptotics with better control of the
Math 135:189–283
error terms and, on the other hand, to extend the Borel E (1909) Les probabilités denombrables et leurs
situations that can be handled. In particular, applications arithmetiques. Rend Circ Math Palermo
relaxing the hypotheses related to hyperbolicity 27:247–271
Bourgain J (1988a) An approach to pointwise ergodic
(or negative curvature) is a constant challenge.
theorems. In: Geometric aspects of functional analysis
The rate of mixing result at the end of section (1986/87), volume 1317 of Lecture Notes in Math.
“Mixing and Additive Relations in Fields” points Springer, Berlin, pp 204–223
towards a large array of natural conjectures Bourgain J (1988b) On the maximal ergodic theorem for
certain subsets of the integers. Israel J Math 61(1):
concerned with rate of mixing, central limit theorem
39–72
phenomena, and rigidity of cocycles for ℤd-actions. Bowen R (1970) Markov partitions for axiom a
diffeomorphisms. Am J Math 92:725–747
Bowen R (1972) The equidistribution of closed geodesics.
Am J Math 94:413–423
Bibliography Bowen R (1973) Symbolic dynamics for hyperbolic flows.
Am J Math 95:429–460
Byszewski J, Cornelissen G (2018) Dynamics on abelian
Primary Literature varieties in positive characteristic. Algebra Number
Adler RL, Weiss B (1970) Similarity of automorphisms of
Theory 12(9):2185–2235
the torus. Memoirs of the American Mathematical
Chothi V, Everest G, Ward T (1997) S-integer dynamical
Society, No. 98. American Mathematical Society,
systems: periodic points. J Reine Angew Math (489):
Providence
99–132
594 Ergodic Theory: Interactions with Combinatorics and Number Theory

Dolgopyat D (1998) On decay of correlations in Anosov Green B, Tao T (2010) Linear equations in primes. Ann
flows. Ann Math (2) 147(2):357–390 Math 171(3):1753–1850
Einsiedler M, Ward T (2003) Asymptotic geometry of non- Haynes A, White CJ (2014) Group automorphisms with
mixing sequences. Ergod Theor Dyn Syst 23(1):75–85 prescribed growth of periodic points, and small primes
Erdös P (1949) On a new method in elementary number in arithmetic progressions in intervals. Adv Math
theory which leads to an elementary proof of the prime 252:572–585
number theorem. Proc Nat Acad Sci U S A 35:374–384 Hill TP (1995) Base-invariance implies Benford’s law.
Erdős P, Turán P (1936) On some sequences of integers. Proc Am Math Soc 123(3):887–895
J Lond Math Soc 11:261–264 Hindman N (1974) Finite sums from sequences within
Everest G, Stangoe V, Ward T (2005) Orbit counting with cells of a partition of ℕ. J Combin Theory Ser A
an isometric direction. In: Algebraic and topological 17:1–11
dynamics, volume 385 of Contemp. Math. American Hlawka E (1964) Discrepancy and uniform distribution of
Mathematical Society, Providence, pp 293–302 sequences. Compos Math 16(83–91):1964
Everest G, Miles R, Stevens S, Ward T (2007) Orbit- Host B, Kra B (2005a) Convergence of polynomial ergodic
counting in non-hyperbolic dynamical systems. averages. Israel J Math 149:1–19. Probability in
J Reine Angew Math 608:155–182 mathematics
Everest G, Miles R, Stevens S, Ward T (2010) Dirichlet Host B, Kra B (2005b) Nonconventional ergodic averages
series for finite combinatorial rank dynamics. Trans and nilmanifolds. Ann Math (2) 161(1):397–488
Am Math Soc 362(1):199–227 Huber H (1959) Zur analytischen Theorie hyperbolischen
Evertse JH, Schlickewei HP (1999) The absolute subspace Raumformen und Bewegungsgruppen. Math Ann
theorem and linear equations with unknowns from a 138:1–26
multiplicative group. In: Number theory in progress, Katok A (1980) Lyapunov exponents, entropy and periodic
Vol. 1 (Zakopane-Kościelisko, 1997). de Gruyter, Ber- orbits for diffeomorphisms. Inst Hautes Études Sci Publ
lin, pp 121–142 Math 51:137–173
Evertse J-H, Schlickewei HP, Schmidt WM (2002) Linear Katsuda A, Sunada T (1990) Closed orbits in homology
equations in variables which lie in a multiplicative classes. Inst Hautes Études Sci Publ Math 71:5–32
group. Ann Math (2) 155(3):807–836 Khanin K, Lopes Dias J, Marklof J (2007) Multi-
Ferenczi S, Kamiński B (1995) Zero entropy and direc- dimensional continued fractions, dynamical
tional Bernoullicity of a Gaussian Z2-action. Proc Am renormalization and KAM theory. Commun Math
Math Soc 123(10):3079–3083 Phys 270(1):197–231
Furstenberg H (1961) Strict ergodicity and transformation Khinchin AI (1964) Continued fractions. The University of
of the torus. Am J Math 83:573–601 Chicago Press, Chicago/London
Furstenberg H (1977) Ergodic behavior of diagonal mea- Knieper G (1997) On the asymptotic geometry of non-
sures and a theorem of Szemerédi on arithmetic pro- positively curved manifolds. Geom Funct Anal 7(4):
gressions. J Analyse Math 31:204–256 755–782
Furstenberg H, Katznelson Y (1978) An ergodic Koopman B (1931) Hamiltonian systems and transforma-
Szemerédi theorem for commuting transformations. tions in Hilbert spaces. Proc Natl Acad Sci U S A
J Analyse Math 34:275–291. (1979 17:315–318
Furstenberg H, Katznelson Y (1985) An ergodic Kuz’min RO (1928) A problem of Gauss. Dokl Akad Nauk
Szemerédi theorem for IP-systems and combinatorial 1928:375–380
theory. J Analyse Math 45:117–168 Lagarias JC (1994) Geodesic multidimensional continued
Furstenberg H, Weiss B (1978) Topological dynamics and fractions. Proc Lond Math Soc (3) 69(3):464–488
combinatorial number theory. J Analyse Math Lalley SP (1987) Distribution of periodic orbits of sym-
34:61–85. (1979 bolic and axiom a flows. Adv Appl Math 8(2):154–193
Gel’fond AO (1960) Transcendental and algebraic num- Lalley SP (1988) The “prime number theorem” for the
bers. Translated from the first Russian edition by Leo periodic orbits of a Bernoulli flow. Am Math Mon
F. Boron. Dover, New York 95(5):385–398
Goldfeld D (2004) The elementary proof of the prime Ledrappier F (1978) Un champ markovien peut etre
number theorem: an historical perspective. In: Number d’entropie nulle et mélangeant. C R Acad Sci Paris
theory (New York, 2003). Springer, New York, Ser A-B 287(7):A561–A563
pp 179–192 Leibman A (2005) Convergence of multiple ergodic aver-
Goldston DA, Pintz J, Yildirim CY (2009) Primes in ages along polynomials of several variables. Israel
tuples. I. Ann Math (2) 170(2):819–862 J Math 146:303–315
Gorodnik A, Spatzier R (2015) Mixing properties of com- Levy P (1929) Sur les lois de probabilité dont dependent
muting nilmanifold automorphisms. Acta Math les quotients complets et incomplets d’une fraction
215(1):127–159 continue. Bull Soc Math France 57:178–194
Gowers WT (2007) Hypergraph regularity and the multi- Lévy P (1936) Sur quelques points de la théorie des pro-
dimensional Szemerédi theorem. Ann Math (2) babilités dénombrables. Ann Inst H Poincaré 6(2):
166(3):897–946 153–184
Green B, Tao T (2008) The primes contain arbitrarily long Lind DA (1982) Dynamical properties of quasihyperbolic
arithmetic progressions. Ann Math (2) 167(2):481–547 toral automorphisms. Ergod Theor Dyn Syst 2(1): 49–68
Ergodic Theory: Interactions with Combinatorics and Number Theory 595

Linnik UV (1944) On the least prime in an arithmetic Schlickewei HP (1990) S-unit equations over number
progression. I. The basic theorem. Rec Math [Mat fields. Invent Math 102(1):95–107
Sbornik] NS 15(57):139–178 Schmidt WM (1972) Norm form equations. Ann Math
Margulis GA (1969) Certain applications of ergodic theory 96:526–551
to the investigation of manifolds of negative curvature. Schmidt K (1989) Mixing automorphisms of compact
Funkcional Anal i Priložen 3(4):89–90 groups and a theorem by Kurt Mahler. Pac J Math
Margulis GA (2004) On some aspects of the theory of 137(2):371–385
Anosov systems. Springer Monographs in Mathemat- Schmidt K, Ward T (1993) Mixing automorphisms of
ics. Springer, Berlin. With a survey by Richard Sharp: compact groups and a theorem of Schlickewei. Invent
Periodic orbits of hyperbolic flows, Translated from the Math 111(1):69–76
Russian by Valentina Vladimirovna Szulikowska Selberg A (1949) An elementary proof of the prime-
Masser DW (2004) Mixing and linear equations over number theorem. Ann Math 50(2):305–313
groups in positive characteristic. Israel J Math 142: Selberg A (1956) Harmonic analysis and discontinuous
189–204 groups in weakly symmetric Riemannian spaces with
Mertens F (1874) Ein Beitrag zur analytischen applications to Dirichlet series. J Indian Math Soc (NS)
Zahlentheorie. J Reine Angew Math 78:46–62 20:47–87
Miles R, Ward T (2006) Mixing actions of the rationals. Sharp R (1991) An analogue of Mertens’ theorem for
Ergod Theor Dyn Syst 26(6):1905–1911 closed orbits of axiom a flows. Bol Soc Brasil Mat
Miles R, Ward T (2011) A directional uniformity of peri- (NS) 21(2):205–229
odic point distribution and mixing. Discrete Contin Sharp R (1993) Closed orbits in homology classes for
Dyn Syst 30(4):1181–1189 Anosov flows. Ergod Theor Dyn Syst 13(2):387–408
Nagle B, Rödl V, Schacht M (2006) The counting lemma Sierpiński W (1910) Sur la valeur asymptotique d’une
for regular k-uniform hypergraphs. Random Struct certaine somme. Bull Int Acad Pol Sci Lett (Cracovie)
Algoritm 28(2):113–179 1910:9–11
Newcomb S (1881) Note on the frequency of the use of Sinaĭ JG (1966) Asymptotic behavior of closed geodesics
digits in natural numbers. Am J Math 4(1):39–40 on compact manifolds with negative curvature. Izv
Noorani MSM (1999) Mertens’ theorem and closed orbits Akad Nauk SSSR Ser Mat 30:1275–1296
of ergodic toral automorphisms. Bull Malaysian Math Sinaĭ JG (1968) Construction of Markov partitions.
Soc (2) 22(2):127–133 Funkcional Anal i Priložen 2(3):70–80
Oxtoby JC (1952) Ergodic sets. Bull Am Math Soc 58: Smale S (1967) Differentiable dynamical systems. Bull
116–136 Am Math Soc 73:747–817
Parry W (1969) Ergodic properties of affine transformations Szemerédi E (1969) On sets of integers containing no four
and flows on nilmanifolds. Am J Math 91: 757–771 elements in arithmetic progression. Acta Math Acad
Parry W (1983) An analogue of the prime number theorem Sci Hungar 20:89–104
for closed orbits of shifts of finite type and their sus- Szemerédi E (1975) On sets of integers containing no
pensions. Israel J Math 45(1):41–52 k elements in arithmetic progression. Acta Arith
Parry W (1984) Bowen’s equidistribution theory and the 27:199–245
Dirichlet density theorem. Ergod Theor Dyn Syst 4(1): Tao T, Ziegler T (2008) The primes contain arbitrarily long
117–134 polynomial progressions. Acta Math 201(2):213–305
Parry W, Pollicott M (1983) An analogue of the prime van der Poorten AJ, Schlickewei HP (1991) Additive rela-
number theorem for closed orbits of axiom a flows. tions in fields. J Austral Math Soc Ser A 51(1):154–170
Ann Math 118(3):573–591 van der Waerden BL (1927) Beweis einer Baudet’schen
Poincaré H (1890) Sur le problème des trois corps et les Vermutung. Nieuw Arch Wisk 15:212–216
equations de la Dynamique. Acta Math 13:1–270 von Neumann J (1932) Proof of the quasi-ergodic hypoth-
Pollicott M, Sharp R (1998) Exponential error terms for esis. Proc Natl Acad Sci U S A 18:70–82
growth functions on negatively curved surfaces. Am Waddington S (1991) The prime orbit theorem for quasi-
J Math 120(5):1019–1042 hyperbolic toral automorphisms. Monatsh Math
Rado R (1933) Studien zur Kombinatorik. Math Z 36(1): 112(3):235–248
424–470 Waldschmidt M (1993) Minorations de combinaisons
Ratner M (1973) Markov partitions for Anosov flows on linéaires de logarithmes de nombres algébriques. Can
n-dimensional manifolds. Israel J Math 15:92–114 J Math 45(1):176–224
Rohlin VA (1949) On endomorphisms of compact com- Ward T (1997/98) Three results on mixing shapes. N Y
mutative groups. Izvestiya Akad Nauk SSSR Ser Mat J Math 3A (Proceedings of the New York Journal of
13:329–340 Mathematics Conference, June 9–13, 1997):1–10
Roth K (1952) Sur quelques ensembles d’entiers. C R Acad Ward T (1998) Almost all S-integer dynamical systems
Sci Paris 234:388–390 have many periodic points. Ergod Theor Dyn Syst
Sárközy A (1978) On difference sets of sequences of 18(2):471–486
integers. III. Acta Math Acad Sci Hungar 31(3–4): Ward T (2005) Group automorphisms with few and with
355–386 many periodic points. Proc Am Math Soc 133(1): 91–96
596 Ergodic Theory: Interactions with Combinatorics and Number Theory

Weyl H (1910) Über die Gibbssche Erscheinung und University Press, Cambridge. With a supplementary
verwandte Konvergenzphänomene. Rendiconti del chapter by Anatole Katok and Leonardo Mendoza
Circolo Matematico di Palermo 30:377–407 Kra B (2006) The Green-Tao theorem on arithmetic pro-
Weyl H (1916) Uber die Gleichverteilung von Zahlen mod gressions in the primes: an ergodic point of view. Bull
Eins. Math Ann 77:313–352 Am Math Soc (NS) 43(1):3–23 (electronic)
Wiener N (1932) Tauberian theorems. Ann Math 33(1): Krengel U (1985) Ergodic theorems, volume 6 of de
1–100 Gruyter studies in mathematics. Walter de Gruyter,
Berlin. With a supplement by Antoine Brunel
McCutcheon R (1999) Elemental methods in ergodic Ram-
Books and Reviews sey theory, volume 1722 of Lecture notes in mathemat-
Arnol’d VI, Avez A (1968) Ergodic problems of classical ics. Springer, Berlin
mechanics. Translated from the French by A. Avez. Miles R, Staines M, Ward T (2015) Dynamical invariants for
W. A. Benjamin, New York/Amsterdam group automorphisms. In: Recent trends in ergodic the-
Bergelson V (1996) Ergodic Ramsey theory – an update. In: ory and dynamical systems, volume 631 of Contemp.
Ergodic theory of Zd actions (Warwick, 1993–1994), Math. Amer. Math. Soc., Providence, pp 231–258
volume 228 of London Math. Soc. Lecture Note Ser. Petersen K (1989) Ergodic theory, volume 2 of Cambridge
Cambridge University Press, Cambridge, pp 1–61 studies in advanced mathematics. Cambridge Univer-
Bergelson V (2000) Ergodic theory and Diophantine prob- sity Press, Cambridge. Corrected reprint of the 1983
lems. In: Topics in symbolic dynamics and applications original
(Temuco, 1997), volume 279 of London Math. Schmidt K (1995) Dynamical systems of algebraic origin,
Soc. Lecture Note Ser. Cambridge University Press, volume 128 of Progress in Mathematics. Birkhäuser,
Cambridge, pp 167–205 Basel
Bergelson V (2006) Combinatorial and Diophantine appli- Schmidt K (2001) The dynamics of algebraic ℤd-actions.
cations of ergodic theory. In: Handbook of dynamical In: European congress of mathematics, Vol. I (Barce-
systems, vol 1B. Elsevier B. V., Amsterdam, lona, 2000), volume 201 of Progr. Math. Birkhäuser,
pp 745–869. Appendix A by A. Leibman and Appen- Basel, pp 543–553
dix B by Anthony Quas and Máté Wierdl Schweiger F (1995) Ergodic theory of fibred systems and
Cornfeld IP, Fomin SV, Sinaĭ YG (1982) Ergodic theory. metric number theory. Oxford Science Publications. The
Springer, New York Clarendon Press/Oxford University Press, New York
Dajani K, Kraaikamp C (2002) Ergodic theory of numbers, Schweiger F (2000) Multidimensional continued fractions.
volume 29 of Carus mathematical monographs. Math- Oxford Science Publications. Oxford University Press,
ematical Association of America, Washington, DC Oxford
del Junco A (2008) Ergodic theorems. In: Encyclopedia of Silverman JH (2007) The arithmetic of dynamical systems,
complexity and systems science. Springer volume 241 of Graduate texts in mathematics. Springer,
Denker M, Grillenberger C, Sigmund K (1976) Ergodic New York
theory on compact spaces. Lecture notes in mathemat- Tao T (2006) Arithmetic progressions and the primes.
ics, vol 527. Springer, Berlin Collect. Math., vol. Extra:37–88
Einsiedler M, Ward T (2011) Ergodic theory with a view Tao T (2007a) The dichotomy between structure and ran-
towards number theory, volume 259 of Graduate texts domness, arithmetic progressions, and the primes. In:
in mathematics. Springer, London International congress of mathematicians, vol I. Euro-
Ellis R (1969) Lectures on topological dynamics. W. A. pean Mathematical Society, Zürich, pp 581–608
Benjamin, New York Tao T (2007b) What is good mathematics. Bull Am Math
Furstenberg H (1981) Recurrence in ergodic theory and Soc (NS) 44(4):623–634
combinatorial number theory. Princeton University Totoki H (1969) Ergodic theory. Lecture notes series, No.14.
Press, Princeton. M. B. Porter Lectures Matematisk Institut, Aarhus Universitet, Aarhus
Furstenberg H, Katznelson Y, Ornstein D (1982) The ergo- van der Waerden BL (1971) How the proof of Baude’s
dic theoretical proof of Szemerédi’s theorem. Bull Am conjecture was found. In: Studies in pure mathematics
Math Soc (NS) 7(3):527–552 (presented to Richard Rado). Academic, London,
Glasner E (2003) Ergodic theory via joinings, pp 251–260
volume 101 of Mathematical surveys and monographs. Walters P (1982) An introduction to ergodic theory,
American Mathematical Society, Providence volume 79 of Graduate texts in mathematics. Springer,
Hejhal DA (1976) The Selberg trace formula and the Rie- New York
mann zeta function. Duke Math J 43(3):441–482 Weil A (1967) Basic number theory. Die Grundlehren der
Iosifescu M, Kraaikamp C (2002) Metrical theory of con- mathematischen Wissenschaften, vol 144. Springer,
tinued fractions, volume 547 of Mathematics and its New York
applications. Kluwer Academic, Dordrecht Ziegler T (2014) Linear equations in primes and dynamics
Katok A, Hasselblatt B (1995) Introduction to the modern of nilmanifolds. In: Proceedings of the international
theory of dynamical systems, volume 54 of Encyclo- congress of mathematicians—Seoul 2014, vol II.
pedia of mathematics and its applications. Cambridge Kyung Moon Sa, Seoul, pp 569–589
 Metric number theory Metric number theory
Ergodic Theory on (or, specifically, metric Diophantine approxi-
Homogeneous Spaces mation) refers to the study of sets of real num-
and Metric Number Theory bers or vectors with prescribed Diophantine
approximation properties.
Dmitry Kleinbock
Department of Mathematics, Brandeis University,
Waltham, MA, USA Definition of the Subject

The theory of Diophantine approximation, named

Article Outline after Diophantus of Alexandria, in its simplest set-
up deals with the approximation of real numbers
Glossary by rational numbers. Various higher-dimensional
Definition of the Subject generalizations involve studying values of linear
Introduction or polynomial maps at integer points. Often a
Basic Facts certain “approximation property” is fixed, and
Further Results one wants to characterize the set of numbers
Further Results (vectors, matrices) which share this property, by
Future Directions means of certain measures (Lebesgue, or
Bibliography Hausdorff, or some other interesting measures).
This is usually referred to as metric Diophantine
Glossary approximation.
The starting point for the theory is an elemen-
Diophantine approximation Diophantine tary fact that ℚ, the set of rational numbers, is
approximation refers to approximation of real dense in ℝ, the reals. In other words, every real
numbers by rational numbers or, more gener- number can be approximated by rationals: for any
ally, finding integer points at which some y  ℝ and any ε > 0, there exists p/q  ℚ with
(possibly vector-valued) functions attain
values close to integers. j y  p=q j < e: ð1Þ
Ergodic theory The study of statistical proper-
ties of orbits in abstract models of dynamical To answer questions like “how well can vari-
systems. ous real numbers be approximated by rational
Hausdorff dimension A nonnegative number numbers? i.e., how small can ε in (1) be chosen
attached to a metric space and extending the for varying p/q  ℚ?”, a natural approach
notion of topological dimension of “suffi- has been to compare the accuracy of the approx-
ciently regular” sets, such as smooth sub- imation of y by p/q to the “complexity” of the
manifolds of real Euclidean spaces. latter, which can be measured by the size of its
Homogeneous spaces A homogeneous space denominator q in its reduced form. This seem-
G/Γ of a group G by its subgroup Γ is the ingly simple set-up has led to introducing many
space of cosets {gΓ}. When G is a Lie group important Diophantine approximation properties
and Γ is a discrete subgroup, the space G/Γ is a of numbers/vectors/matrices, which show up in
smooth manifold and locally looks like G itself. various fields of mathematics and physics, such
Lattice; unimodular lattice A lattice in a Lie as differential equations, Kolmogorov-Arnold-
group is a discrete subgroup of finite covolume; Moser (KAM) theory, and transcendental number
unimodular stands for covolume equal to 1. theory.
© Springer Science+Business Media, LLC, part of Springer Nature 2023 597
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_180
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer Science+Business Media LLC 2020
https://doi.org/10.1007/978-3-642-27737-5_180-3
598 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

Introduction Y  W m,n ðcÞÞ if there are infinitely many

q  ℤn such that
As the first example of refining the statement
about the density of ℚ in ℝ, consider a theorem k Yq þ p k  cðk q kÞ ð4Þ
by Kronecker stating that for any y  ℝ and any
c > 0, there exist infinitely many q  ℤ such that for some p  ℤm. Here k  k is the supremum
norm on ℝk given by kyk ¼ max1ik |yi|. (This
j y  p=q j < c= j q j i:e: j qy  p j < c ð2Þ definition is slightly different from the one used in
(Kleinbock and Margulis 1999), which involved
for some p  ℤ. A comparison of (1) and (2) powers of norms.)
shows that it makes sense to multiply both sides Traditionally, one of the main goals of metric
of (1) by q, since in the right-hand side of (2) one Diophantine approximation has been to understand
would still be able to get very small numbers. In how big the sets W m,n ðcÞ are for fixed m, n, and
other words, approximation of y by p/q translates various functions c. Of course, (4) is not the only
into approximating integers by integer multiples interesting condition that can be studied; various
of y. modifications of the approximation properties can
Also, if y is irrational, ( p, q) can be chosen to also be considered. For example, the Oppenheim
be relatively prime, that is, one gets infinitely Conjecture, now a theorem of Margulis (1989) and
many different rational numbers p/q satisfying (2). a basis for many important developments (Dani
However, if y  ℚ, the latter is no longer true for and Margulis 1993; Eskin et al. 1998, 2005), states
small enough c. Thus, it seems to be more conve- that indefinite irrational quadratic forms can take
nient to talk about pairs ( p, q) rather than arbitrary small values at integer points;
p/q  ℚ, avoiding a necessity to consider the Littlewood’s conjecture, see (18) below, deals
two cases separately. At this point, it is convenient with a similar statement about products of linear
to introduce the following central definition: if c forms. See the article Ergodic Theory: Rigidity by
is a function ℕ ! ℝ+ and y  ℝ, say that y is Nitica and surveys (Einsiedler and Lindenstrauss
c-approximable (notation: y  W ðcÞ ) if there 2006; Margulis 1997) for details.
exist infinitely many q  ℕ such that We remark that the standard tool for studying
Diophantine approximation properties of real
j qy  p j < cðqÞ ð3Þ numbers (m ¼ n ¼ 1) is the continued fraction
expansion or, equivalently, the Gauss map
for some p  ℤ. Because of Kronecker’s Theo- x 7! 1/x mod 1 of the unit interval, see
rem, it is natural to assume that c(x) ! 0 as (Khintchine 1963). However, the emphasis of
x ! 1. Often c will be assumed nonincreasing, this survey lies in higher-dimensional theory,
although many results do not require monotonic- and the dynamical system described below can
ity of c. be thought of as a replacement for the continued
One can similarly consider a higher- fraction technique applicable in the one-
dimensional version of the above set-up. Note dimensional case. Additional details about inter-
that y  ℝ in the above formulas plays the role actions between ergodic theory and number the-
of a linear map from ℝ to another copy of ℝ, and ory can be found in the article by Nitica
one asks how close values of this map at integers mentioned above, in Ergodic Theory: Recur-
are from integers. It is natural to generalize it by rence by Frantzikinakis and McCutcheon and
taking a linear operator Y from ℝn to ℝm for fixed Ergodic Theory: Interactions with Combinator-
m, n  ℕ, that is, an m  n - matrix (interpreted as ics and Number Theory by Ward, as well as in the
a system of m linear forms Yi on ℝn). We will survey papers (Einsiedler and Lindenstrauss
denote by Mm,n the space of m  n matrices with 2006; Eskin 1998; Kleinbock 2001; Kleinbock
real coefficients. For c as above, one says that et al. 2002; Lindenstrauss 2007; Margulis 1997,
Y  Mm,n is c-approximable (notation: 2002).
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 599

Here is a brief outline of the rest of the article. In enough T. The former is sometimes referred to as
the next section, we survey basic results, some clas- asymptotic approximation and amounts to study-
sical, some obtained relatively recently, in metric ing sets W m,n ðcÞ mentioned in the previous sec-
Diophantine approximation. Section “Connection tion. The latter, less studied set-up of uniform
with Dynamics on the Space of Lattices” is devoted approximation, has attracted some attention in
to a description of the connection between recent years. Both set-ups can be approached
Diophantine approximation and dynamics, specifi- using tools from dynamics on the space of lattices,
cally flows on the space of lattices. In sections as we shall see below.
“Diophantine Approximation with Dependent The simplest choice for functions c happens to
Quantities: The Set-Up” and “Further Results,” we be the following: let us denote
specialize to the set-up of Diophantine approxima-
tion on manifolds or, more generally, approximation cc,v ðxÞ ¼ cxv :
properties of vectors with respect to measures satis-
fying some natural conditions and show how appli- From Dirichlet’s Theorem, it easily follows
cations of homogeneous dynamics contributed to that W m,n c1,n=m ¼ Mm,n . The constant c ¼ 1
important recent developments in the field.
is not optimal: the smallest
p value of c for which
Section “Future Directions” mentions several open
W 1,1 cc,1 ¼ ℝ is 1= 5 , and the optimal con-
questions and directions for further investigation.
stants are not known in higher dimensions,
although some estimates can be given, see
Basic Facts (Schmidt 1980).
Systems of linear forms which do not belong to
General references for this section: (Cassels 1957; W m,n cc,n=m for some positive c are called badly
Schmidt 1980). The starting point for all the approximable; that is, we set
explorations in metric number theory is
Dirichlet’s Theorem (1842), stating that for any def
BAm,n ¼ Mm,n ∖[c>0 W m,n cc,n=m :
Y  Mm,n and for any T > 1, there exist q ¼
(q1, . . ., qn)  ℤn\{0} and p ¼ ( p1, . . ., pm)  ℤm
satisfying the following system of inequalities: Their existence in arbitrary dimensions was
shown by Perron. Note that a real number y (m ¼
n ¼ 1) is badly approximable if and only if its
k Yq  p k < T n=m and k q k  T: ð5Þ
continued fraction coefficients are uniformly
bounded. It was proved by Jarnik (1929) in the
In fact, it is this paper of Dirichlet which gave
case m ¼ n ¼ 1 and by Schmidt in the general case
rise to his box principle. Later another proof of the
same result was given by Minkowski. (Schmidt 1969) that badly approximable matrices
form a set of full Hausdorff dimension: that is, dim
A natural question to ask is whether one can
(BAm,n) ¼ mn.
improve (5) by replacing T n/m by a smaller func-
tion, that is, consider the following system of On the other hand, it can be shown that each of
inequalities: the sets W m,n cc,n=m for any c > 0 has full
Lebesgue measure, and hence, the complement
k Yq  p k < cðT Þ and k q k  T, ð5aÞ BAm, n to their intersection has measure zero.
This is a special case of a theorem due to
where c is a positive, continuous, decreasing Khintchine (1924) in the case n ¼ 1 and to
function which decays to zero at infinity. Histori- Groshev (1938) in full generality, which gives
cally, there have been two directions to pursue in the precise condition on the function c under
this regard: looking for solvability of (5a) for an which the set of c-approximable matrices has
unbounded set of T > 0 versus for all large full measure. Namely, if c is nonincreasing (this
600 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

assumption can be removed in higher dimensions VWAm,n has full Hausdorff dimension. Matrices
but not for n ¼ 1, see (Duffin and Schaeffer contained in the intersection
1941)), then l-almost no (resp., l-almost every)
Y  Mm,n is c-approximable, provided the sum \W
v>0
m,n c1,v ¼ fY  Mm,n : oðY Þ ¼ 1g
1
kn1 cðkÞm ð6Þ are called Liouville and form a set of Hausdorff
k¼1 dimension (n – 1)m, that is, to the dimension of
Y for which Yq  ℤ for some q  ℤn\{0} (the
converges (resp., diverges). (Here and hereafter l
latter belongs to W m,n ðcÞ for any positive c).
stands for Lebesgue measure.) This statement is
Metric theory of uniform approximation was
usually referred to as the Khintchine–Groshev
initiated by Davenport and Schmidt in (1969/
Theorem. The convergence case of this theorem
1970, 1970). Let us say that Y  Mm,n is c-
follows in a straightforward manner from the
Dirichlet (notation: Y  D m,n ðcÞ) if the system of
Borel–Cantelli Lemma, but the divergence case
inequalities (5a) has solutions in (p, q)  ℤm 
is harder. It was reproved and sharpened in 1960
(ℤn\{0}) for all sufficiently large T. (Again, this
by Schmidt (1960), who showed that if the
definition is slightly different from the one used in
sum (6) diverges, then for almost all Y, the number
several papers on the subject (Kleinbock and
of solutions to (4) with kq k  T is asymptotic to
Wadleigh 2018, 2019; Kleinbock et al. 2021a),
the partial sum of the series (6) (up to a constant)
where powers of norms were considered.)
and also gave an estimate for the error term.
Dirichlet’s Theorem asserts that D m,n c1,n=m ¼
A special case of the convergence part of the
Mm,n : However, the constant 1 is now optimal: it
theorem shows that W m,n c1,v has measure
was shown in (Davenport and Schmidt 1970) that
zero whenever v > n/m. Y is said to be very well the Lebesgue measure of D m,n cc,n=m is zero for
approximable if it belongs to W m,n c1,v for some any c < 1. This phenomenon can be easily
v > n/m. That is, explained by means of dynamics on homoge-
neous spaces, see (Dani 1985; Kleinbock and
def
VWAm,n ¼ [v>n=m W m,n c1,v : Weiss 2008). Y is said to be Dirichlet-Improvable
def
if it belongs to the null set DIm,n ¼
More specifically, let us define the Diophantine [c<1 D m,n cc,n=m , and singular if it belongs to
exponent o(Y ) of Y (sometimes called “the exact def
order” of Y) to be the supremum of v > 0 for SINGm,n ¼ \c>0 D m,n cc,n=m : Note that Schmidt
which Y  W m,n c1,v . Then o(Y ) is always not also showed that DIm,n contains BAm,n, hence is a
less than n/m and is equal to n/m for Lebesgue-a.e. set of full Hausdorff dimension.
Y; in fact, VWAm,n ¼ {Y  Mm,n : o(Y ) > n/m}. Note also that all the aforementioned proper-
The Hausdorff dimension of the null sets ties behave nicely with respect to transposition;
W m,n c1,v was computed independently by this is described by the so-called Khintchine’s
Besicovitch (1929) and Jarnik (1928) in the one- Transference Principle (Chapter V in (Cassels
dimensional case and by Dodson (1992) in gen- 1957)). For example, Y  BAm,n if and only if
eral: when v > n/m, one has YT  BAn,m, and Y  VWAm,n if and only if
YT  VWAn,m. In particular, many problems
mþn related to approximation properties of vectors
dim W m,n c1,v ¼ ðn 1Þm þ : ð7Þ
vþ1 (n ¼ 1) and linear forms (m ¼ 1) reduce to one
another.
See (Dodson 1993) for a nice exposition of We refer the readers to (Beresnevich et al.
ideas involved in the proof of both the aforemen- 2006; Harman 1998) for very detailed and com-
tioned formula and the Khintchine–Groshev The- prehensive accounts of various further aspects of
orem. Note that it follows from (7) that the null set the theory.
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 601

Connection with Dynamics on the Space of Y 1 ðqÞ þ p1 , . . . , Y m ðqÞ þ pm and q1 , . . . , qn ,

Lattices
General references for this section: (Bekka and and consider the collection of vectors
Mayer 2000; Starkov 2000).
Interactions between Diophantine approxima- Yq þ p
tion and the theory of dynamical systems has a p  ℤm , q  ℤn ¼ LY ℤ k
q
long history. Already in Kronecker’s Theorem
one can see a connection. Indeed, the statement where k ¼ m þ n and
of the theorem can be rephrased as follows: the
points on the orbit of 0 under the rotation of the def Im Y
circle ℝ/ℤ by y approach the initial point 0 arbi- LY ¼ , Y  Mm,n : ð8Þ
0 In
trarily closely. This is a special case of the
Poincare Recurrence Theorem in measurable This collection is a unimodular lattice in ℝk,
dynamics. And, likewise, all the aforementioned that is, a discrete subgroup of ℝk with
properties of Y  Mm,n can be restated in terms of covolume 1. Our goal is to keep track of vectors
recurrence properties of the ℤn-action on the in such a lattice having small projections onto the
m-dimensional torus ℝm/ℤm given by first m components of ℝk and big projections onto
x 7! Yx mod ℤm. In other words, fixing Y gives the last n components. This is where dynamics
rise to a dynamical system in which approxima- comes into the picture. Denote by gt the one-
tion properties of Y show up. parameter subgroup of SLk(ℝ) given by
However, the theme of this section is a differ-
ent dynamical system, whose phase space is
(essentially) the space of parameters Y and which gt ¼ diag et=m , . . . , et=m , et=n , . . . , et=n :
can be used to read the properties of Y from the
mtimes ntimes
behavior of the associated trajectory.
It has been known for a long time (see ð9Þ
(Sheingorn 1993) for a historical account) that
The vectors in the lattice LYℤk are moved by
Diophantine properties of real numbers can be
the action of gt, t > 0, and a special role is played
coded by the behavior of geodesics on the quo-
by the moment t when the “small” and “big” pro-
tient of the hyperbolic plane by SL2(ℤ). In fact,
jections equalize. That is, one is led to consider a
the latter flow can be viewed as the suspension
new dynamical system. Its phase space is the
flow of the Gauss map mentioned at the end of
space of unimodular lattices in ℝk, which can be
section “Introduction.” There have been many
naturally identified with the homogeneous space
attempts to construct a higher-dimensional ana-
logue of the Gauss map so that it captures all the
def
features of simultaneous approximation, see Ok ¼ G=G, where
ð10Þ
(Khanin et al. 2007; Kontsevich and Suhov G ¼ SLk ðℝÞ and G ¼ SLk ðℤÞ,
1999; Lagarias 1994) and references therein. On
the other hand, it seems to be more natural and and the action is given by left multiplication by
efficient to generalize the suspension flow itself, elements of the subgroup (9) of G or perhaps other
and this is where one needs higher rank homoge- subgroups H  G. Study of such systems has a
neous dynamics. rich history; for example, they are known to be
As was mentioned above, in the basic set-up of ergodic and mixing whenever H is unbounded
simultaneous Diophantine approximation, one (Moore 1966). What is important in this particular
takes a system of m linear forms Y1, . . ., Ym on case is that the space Ωk happens to be non-
ℝn and looks at the values of |Yi(q) þ pi|, pi  ℤ, compact, and its structure at infinity is described
when q ¼ (q1, . . ., qn)  ℤn is far from 0. The via Mahler’s Compactness Criterion, see
trick is to put together Chapter V in (Bekka and Mayer 2000): a sequence
602 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

of lattices giℤk goes to infinity in Ωk , there exists This correspondence allows one to link various
a sequence {vi  ℤk\{0}} such that gi(vi) ! 0 as Diophantine and dynamical phenomena. For
i ! 1. Equivalently, for ε > 0 consider a subset example, from the results of (Kleinbock and
Kε of Ωk consisting of lattices with no nonzero Margulis 1996) on abundance of bounded orbits
vectors of norm less than ε; then all the sets Kε are on homogeneous spaces, one can deduce the
compact, and every compact subset of Ωk is aforementioned theorem of Schmidt (1969): the
contained in one of them. Moreover, one can set BAm,n has full Hausdorff dimension. See also
choose a metric on Ωk such that dist(Λ, ℤk) is, up Kleinbock and Weiss (2013) where a winning
to a uniform multiplicative constant, equal to property of BAm,n is established by dynamical
 log minv  Λ\{0} kvk (see (Ding 1994)), then the methods. A dynamical Borel–Cantelli Lemma
length of the smallest nonzero vector in a lattice ˄ established in (Kleinbock and Margulis 1999)
will determine how far away is this lattice in the can be used for an alternative proof of the
“cusp” of Ωk. Note that Minkowski’s Convex Body Khintchine–Groshev Theorem; see also Sullivan
Theorem shows that Kε ¼ ; when ε > 1. On the (1982) for an earlier geometric approach. Note
other hand, when ε < 1, each of the sets Kε has that establishing a uniform approximation ana-
nonempty interior, and K1 ¼ \ε<1Kε is nonempty, logue of the Khintchine–Groshev Theorem, that
but has empty interior and zero Haar measure. It is is, finding a criterion for sets D m,n ðcÞ to have zero
called the critical locus for the supremum norm, or full measure, is still an open problem; it was
and its structure is described by the Hajós– solved in (Kleinbock and Wadleigh 2018) for
Minkowski Theorem. m ¼ n ¼ 1 using continued fractions, and a partial
Using Mahler’s Criterion, it is not hard to show solution for the general case was developed in
that Y  BAm,n if and only if the trajectory Kleinbock et al. (2021a). It is worthwhile to
point out that all the aforementioned proofs are
gt LY ℤk : t  ℝþ ð11Þ based on the following two properties of the gt-
action: effective mixing, which forces points to
is bounded in Ωk. This was proved by Dani return to compact subsets and makes preimages of
(1985) in 1985 and later generalized in cusp neighborhoods quasi-independent, and
(Kleinbock and Margulis 1999) to produce a hyperbolicity, which implies that the behavior of
criterion for Y to be c-approximable for any points on unstable leaves is generic. The latter is
nonincreasing function c. Namely, given c one important since the orbits of the group
can in a unique way define the function r(t) such {LY : Y  Mm,n} are precisely the unstable leaves
that Y  W m,n ðcÞ if and only if gtLYℤk Kr(t) for with respect to the gt-action.
an unbounded set of t > 0. An important special
case is a criterion for a system of linear forms to k Yq  p k < cðT Þ and k q k  T, ð11aÞ
be very well approximable: Y  VWAm,n if and
only if the trajectory (11) has linear growth, We note that other types of Diophantine prob-
that is, there exists a positive γ such that lems, such as conjectures of Oppenheim and
dist(gtLYℤk, ℤk) > γt for an unbounded set of Littlewood mentioned in the previous section,
t > 0. Similarly, one can produce a uniform can be reduced to statements involving actions
version, that is, a criterion for Y to be on Ωk by means of the same principle: Mahler’s
c-Dirichlet: Y  D m,n ðcÞ if and only if Criterion is used to relate small values of some
g tLYℤ k Kr(t) for all large enough t > 0. In function at integer points to excursions to infinity
particular, Y is singular if and only if {gtLYℤk} in Ωk of orbit of the stabilizer of this function. In
is divergent and is Dirichlet-Improvable if and particular, Littlewood’s Conjecture deals with
only if gtLYℤk is not in Kε for some ε > 0; that is, multiplicative approximation and is related to a
the gt trajectory of LYℤk eventually stays away multi-parameter action by a certain semigroup of
from the critical locus K1. diagonal matrices, see the section “Further
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 603

Results” part of section “Diophantine Approxi- behavior of the trajectory {gtLYℤk}. One of its
mation with Dependent Quantities: The Set-Up.” applications is a better understanding of uniform
Other important and useful recent applications Diophantine exponents of matrices; another one is
of homogeneous dynamics to metric Diophantine showing that the set DIm,n\(BAm,n [ SINGm,n) is
approximation are related to the circle of ideas uncountable, see (Beresnevich et al. 2020).
roughly called “Diophantine approximation with In another development, the correspondence
dependent quantities” (terminology borrowed between approximation and dynamics can be
from (Sprindžuk 1979)), to be surveyed in the extended to inhomogeneous approximation, that
next two sections. is, studying values Yq þ z þ p of systems of affine
forms (here Y  Mm,n and z  ℝm) at integer
points (q, p). This way one obtains a zero-infinity
Further Results law for improvement of Dirichlet’s theorem, see
(Kleinbock and Wadleigh 2019) and also (Kim
Overall there have been numerous developments and Kim 2022) for an alternative proof and a
utilizing the dynamical approach to Diophantine version for Hausdorff measures.
problems. For example, the Hausdorff dimension For recent results utilizing the correspondence
of the null set SINGm,n of singular systems of between Diophantine approximation and homoge-
linear forms was computed for m ¼ 2, n ¼ 1 by neous dynamics, see (Athreya et al. 2015;
Cheung (2011) and then for n ¼ 1 and arbitrary Dolgopyat et al. 2017; Björklund and Gorodnik
m by Cheung and Chevallier (2016). In the gen- 2019; Shapira and Weiss 2022) among many other
eral case, the sharp upper estimate was obtained in papers.
(Kadyrov et al. 2017), where dynamics was used
via the method of integral inequalities for height Diophantine Approximation with Dependent
functions on the space of lattices originally devel- Quantities: The Set-Up
oped in (Eskin et al. 1998); those are also known General references for this section: (Bernik and
as Margulis functions, see (Eskin and Mozes Dodson 1999; Sprindžuk 1979).
2022) for a recent survey. A refinement of the Here we restrict ourselves to Diophantine prop-
technique from (Kadyrov et al. 2017), combined erties of vectors in ℝn. In particular, we will look
with the exponential mixing of the gt-action on more closely at the set of very well approximable
Ωk, produces a proof of the Dimension Drop Con- vectors, which we will simply denote by VWA,
jecture (Kleinbock and Mirzadeh 2020); its dropping the subscripts. In many cases, it does not
byproduct is a sharpening of the Davenport– matter whether one works with row or column
Schmidt result, namely, establishing that vectors, in view of the duality remark made at the
dim D m,n cc,n=m < mn for any c < 1. end of section “Basic Facts.” Note however that
during recent years there has been a substantial
The complimentary lower estimate for the
progress toward establishing analogues of results
dimension of the set of singular matrices, produc-
from this section to the space of matrices (this was
ing the equality
mentioned as work in progress in [Kleinbock and
mn Margulis 1998]), see (Kleinbock et al. 2010;
dim ðSINGm,n Þ ¼ mn  , Beresnevich et al. 2015; Aka et al. 2018; Das et al.
mþn
2018; Yang 2019).
was established in (Das et al. 2019) using the We begin with a non-example of an application
method of parametric geometry of numbers, stem- of dynamics to Diophantine approximation: a cel-
ming from the work of Schmidt–Summerer (2013) ebrated and difficult theorem which currently, to
and Roy (2015). The latter method is a powerful the best of the author’s knowledge, has no dynam-
tool that gives a strong quantitative way to relate ical proof. Suppose that y ¼ (y1, . . ., yn)  ℝn is
Diophantine properties of matrices Y to the such that each yi is algebraic and 1, y1, . . ., yn are
604 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

linearly independent over ℚ. It was established by very algebraic.” At about the same time, Schmidt
Roth for n ¼ 1 (Roth 1955) and then generalized (1964) proved the extremality of fl when
to arbitrary n by Schmidt (1972) that y as above f : I ! ℝ2, I  ℝ, is C3 and satisfies
necessarily belongs to the complement of VWA.
In other words, vectors with very special algebraic f 01 ðxÞ f 02 ðxÞ
6¼ 0 for l  a:e: x  I;
properties happen to follow the behavior of a f 001 ðxÞ f 002 ðxÞ
generic vector in ℝn.
We would like to view the above example as a in other words, the curve parametrized by f has
special case of a general class of problems. nonzero curvature at almost all points. Since then,
Namely, suppose we are given a Radon measure a lot of attention has been devoted to showing that
m on ℝn. Let us say that m is extremal (Sprindžuk measures fl are extremal for other smooth
1980) if m-a.e. y  ℝn is not very well maps f.
approximable. Further, define the Diophantine To describe a broader class of examples, recall
exponent o(m) of m to be the m-essential the following definition. Let x  ℝd and let f ¼
supremum of the function o(); in other words, ( f1, . . ., fn) be a Ck map from a neighborhood of
x to ℝn. Say that f is nondegenerate at x if ℝn is
def
oðmÞ ¼ sup vjm W c1,v >0 : spanned by partial derivatives of f at x up to some
order. Say that f is nondegenerate if it is non-
Clearly it only depends on the measure class of degenerate at l-a.e. x. It was conjectured by
m. If m is naturally associated with a subset M of Sprindzuk (1979) in 1980 that fl for real analytic
ℝn supporting m (e.g., if M is a smooth sub- nondegenerate f are extremal. Many special cases
manifold of ℝn and m is the measure class of the were established since then (see (Bernik and
Riemannian volume on M, or, equivalently, the Dodson 1999) for a detailed exposition of the
pushforward fl of l by a smooth map theory and many related results), but the general
f parametrizing M), one defines the Diophantine case stood open until the mid-1990s (Kleinbock
exponent o(M) of M to be equal to that of m and and Margulis 1998), when Sprindzuk’s conjecture
says that M is extremal if f(x) is not very well was proved using the dynamical approach (later
approximable for m-a.e. x. Beresnevich (2002) succeeded in establishing and
Then o(m)  n for any m, and o(l) ¼ o(ℝn) is extending this result without use of dynamics).
equal to n. The latter justifies the use of the word The proof in (Kleinbock and Margulis 1998)
“extremal”: m is extremal if o(m) is equal to n, that uses the correspondence outlined in the previous
is, attains the smallest possible value. The afore- section plus a measure estimate for flows on the
mentioned results of Roth and Schmidt then can space of lattices which is described below.
be interpreted as the extremality of atomic mea- In the subsequent work, the method of
sures supported on algebraic vectors without (Kleinbock and Margulis 1998) was adapted to a
rational dependence relations. much broader class of measures. To define it we
Historically, the first measure (other than l) to need to introduce some more notation and defini-
be considered in the set-up described above was tions. If x  ℝd and r > 0, denote by B(x, r) the
the pushforward of l by the map open ball of radius r centered at x. If B ¼ B(x, r)
and c > 0, cB will denote the ball B(x, cr). For
f ðxÞ ¼ x, x2 , . . . , xn : ð12Þ B  ℝd and a real-valued function f on B, let

def
The extremality of fl for f as above was k f kB ¼ sup j f ðxÞ j:
conjectured in 1932 by K. Mahler (1932) and xB

proved in 1964 by Sprindzuk (1964, 1969). It

was important for Mahler’s study of transcenden- If v is a measure on ℝd such that n(B) > 0,
def
tal numbers: this result, roughly speaking, says define k f kn,B ¼ k f kB\supp n ; this is the same as
that almost all transcendental numbers are “not the L1(n)-norm of f |B if f is continuous and B is
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 605

open. If D > 0 and U  ℝd is an open subset, let us Another relevant notion is the nonplanarity of
say that v is D-Federer on U if for any ball B  U (f, v). Namely, (f, v) is said to be nonplanar if
centered at supp v, one has nnðð3BÞ
BÞ < D whenever whenever B is a ball with n(B) > 0, the restric-
3B  U. This condition is often called “doubling” tions of 1, f1, . . ., fn to B \ supp n are linearly
in the literature. See (Kleinbock et al. 2004; independent over ℝ; in other words, f-
Mauldin and Urbański 1996) for examples and (B \ supp n) is not contained in any proper
references. v is called Federer if for v-a.e. affine subspace of ℝn. Note that absolutely
x  ℝd, there exist a neighborhood U of x and good implies both good and nonplanar, but the
D > 0 such that v is D-Federer on U. converse is in general not true.
Given C, α > 0, open U  ℝd, and a measure Many examples of (absolutely) good and
v on U, a function f : U ! ℝ is called (C, α)-good nonplanar pairs (f, v) can be found in the litera-
on U with respect to v if for any ball B  U ture. Already the case n ¼ d and f ¼ Id is very
centered in supp v and any ε > 0, one has interesting. A measure m on ℝn is said to be
friendly (respectively, absolutely friendly) if
nðfx  B : j f ðxÞj< egÞ and only if it is Federer and the pair (Id, m) is
a good and nonplanar (respectively, absolutely
e
C nðBÞ: ð13Þ good). See (Kleinbock et al. 2004; Stratmann
k f kn,B
and Urbanski 2006; Urbanski 2005) for many
This condition was formally introduced in examples. An important class of measures is
(Kleinbock and Margulis 1998) for v being given by limit measures of irreducible system
Lebesgue measure and in (Kleinbock et al. 2004) of self-similar or self-conformal contractions
for arbitrary v. A basic example is given by poly- satisfying the Open Set Condition (Hutchinson
nomials, and the upshot of the above definition is 1981); those are shown to be absolutely friendly
the formalization of a property needed for the in (Kleinbock et al. 2004). The prime example is
proof of several basic facts (Dani 1979, 1986; the middle-third Cantor set on the real line. The
Margulis 1975) about polynomial maps into the term “friendly” was cooked up as a loose abbre-
space of lattices. In (Kleinbock et al. 2004) a viation for “Federer, nonplanar and decaying”
strengthening of this property was considered: and later proved to be particularly friendly in
f was called absolutely (C, α)-good on U with dealing with problems arising in metric number
respect to v if for B and ε as above one has theory, see, for example, (Fishman 2006).
Also let us say that a pair (f, v) is non-
a degenerate if f is nondegenerate at v-a.e. x.
e
nðfx  B :j f ðxÞj< egÞ  C nðBÞ: ð14Þ When v is Lebesgue measure on ℝd, it is proved
k f kB
in Proposition 3.4 in (Kleinbock and Margulis
There is no difference between (13) and (14) 1998) that a nondegenerate (f, v) is good and
when v has full support, but it turns out to be nonplanar. The same conclusion is derived in
useful for describing measures supported on Proposition 7.3 in (Kleinbock et al. 2004),
proper (e.g., fractal) subsets of ℝd. assuming that v is absolutely friendly. Thus,
Now suppose that we are given a measure v on volume measures on smooth nondegenerate
ℝd, an open U  ℝd with n(U ) > 0, and a map f ¼ manifolds are friendly, but not absolutely
( f1, . . ., fn) : ℝd ! ℝn. Following (Kleinbock and friendly.
Weiss 2008), say that a pair (f, v) is (absolutely) It turns out that all the aforementioned exam-
good on U if any linear combination of 1, f1, . . ., fn ples of measures can be proved to be extremal by a
is (absolutely) (C, α)-good on U with respect to v. generalization of the argument from (Kleinbock
If for v-a.e. X, there exists a neighborhood U of and Margulis 1998). Specifically, let v be a
X and C, α > 0 such that v is (absolutely) (C, α)- Federer measure on ℝd, U an open subset of ℝd,
good on U, we will say that the pair (f, v) is and f : U ! ℝn a continuous map such that the pair
(absolutely) good. (f, v) is good and nonplanar, then fn is extremal.
606 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

This can be derived from the Borel–Cantelli from (16). For more detail on the proof, see the
Lemma, the correspondence described in the pre- lecture notes (Kleinbock 2010).
vious section, and the following measure esti-
mate: if v, U, and f are as above, then for v-a.e.
x0  U, there exists a ball B  U centered at x0 Further Results
and C, a > 0 such that for any t  ℝ+ and any
ε > 0, The approach to metric Diophantine approxima-
tion using quantitative nondivergence, that is, the
n x  B : gt Lf ðxÞ ℤnþ1 Ke < Cea : ð15Þ implication (i) þ (ii) ) (iii), is not omnipotent. In
particular, it is difficult to use when more precise
Here gt is as in (9) with m ¼ 1 (assuming that results are needed, such as computing/estimating
the row vector viewpoint is adopted). This is a the Hausdorff dimension of the set of
quantitative way of saying that for fixed t, the c1,v-approximable vectors on a manifold. See
“flow” x 7! gtLf(x)ℤnþ1, B ! Ωnþ1 cannot (Beresnevich et al. 2007; Beresnevich and Velani
diverge and in fact must spend a big 2007) for such results. On the other hand, the
(uniformly in t) proportion of time inside com- dynamical approach can often treat much more
pact sets Kε. general objects than its classical counterpart, and
The inequality (15) is derived from a general also can be perturbed in a lot of directions, produc-
“quantitative non-divergence” estimate, which ing many generalizations and modifications of the
can be thought of a substantial generalization of main theorems from the preceding section. It has
theorems of Margulis and Dani (Dani 1979, numerous applications, most of which show up in a
1986; Margulis 1975) on nondivergence of recent survey (Beresnevich and Kleinbock 2022).
unipotent flows on homogeneous spaces. One One of the most important of them is the
of its most general versions (Kleinbock et al. so-called multiplicative version of the set-up of
2004) deals with a measure v on ℝd, a continuous section “Diophantine Approximation with
map h : B ! G, where B is a ball in ℝd centered Dependent Quantities: The Set-Up.” Namely,
def def
at supp v and G is as in (10). To describe the define functions PðxÞ ¼ i jxi j and Pþ ðxÞ ¼
assumptions on h, one needs to employ the com- i maxðjxi j, 1Þ Then, given a function
binatorial structure of lattices in ℝk, and it will be c : ℕ ! ℝ+, one says that Y  Mm, n is multipli-
convenient to use the following notation: if V is a catively c-approximable (notation: Y  W  m,n ðcÞÞ
nonzero rational subspace of ℝk and g  G, if there are infinitely many q  ℤn such that
define ‘V(g) to be the covolume of g(V \ ℤk) in
gV. Then, given positive constants C, D, α, there PðY q þ pÞ1=m  c Pþ ðqÞ1=n ð17Þ
exists C1 ¼ C1(d, k, C, α, D) > 0 with the fol-
lowing property. Suppose v is D-Federer on
for some p  ℤm. Since P(x)  P+(x)  kxkk
B, 0 < r  1, and h is such that for each rational
for x  ℝk, any c-approximable Y is multipli-
V  ℝk‘V ∘ h is (C, α)-good on B with respect to
catively c-approximable; but the converse is in
v, and k‘V ∘ hkn,B  r, where B ¼ 3ðk1Þ B. Then
general not true, see, for example, (Gallagher
for any positive ε  r, one has
1962). However if one, as before, considers the
family {c1,v}, the critical parameter for which
n x  B : hðxÞℤk K e
the drop from full measure to measure zero
 C1 ðerÞa nðBÞ: ð16Þ
occurs is again n/m. That is, if one defines the
multiplicative Diophantine exponent o(Y ) of
Taking h(x) ¼ gtLf(x) and unwinding the defi-
Y by o ðY Þ ¼ sup v : Y  W 
def
nitions of good and nonplanar pairs, one can show m,n c1,v , then
that (i) and (ii) can be verified for some balls clearly o(Y )  o(Y ) for all Y, and yet o(Y ) ¼
B centered at v-almost every point, and derive (15) n/m for l-a.e. Y  Mm,n.
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 607

Now specialize to ℝn (by the same duality prin- the group of diagonal matrices on the space of
ciple as before, it does not matter whether to think lattices. See (Cassels and Swinnerton-Dyer 1955)
in terms of row or column vectors, but we will for an implicit description of this correspondence
adopt the row vector set-up), and define the and (Einsiedler and Lindenstrauss 2006;
multiplicative exponent o(m) of a measure m on Lindenstrauss 2007; Margulis 1997) for more
ℝn by o ðmÞ ¼ sup vjm W  c1,v > 0 ; then
def detail.
 The dynamical approach also turned out to be
o (l) ¼ n. Following Sprindzuk (1980), say that m
is strongly extremal if o(m) ¼ n. It turns out that fruitful in studying Diophantine properties of
all the results mentioned in the previous section pairs (f, v) for which the nonplanarity condition
have their multiplicative analogues; that is, the fails. Note that obvious examples of non-extremal
measures described there happen to be strongly measures are provided by proper affine subspaces
extremal. This was conjectured by A. Baker of ℝn whose coefficients are rational or are well
(1975) for the curve (12) and then by Sprindzuk enough approximable by rational numbers. On the
in 1980 (Sprindžuk 1980) for analytic non- other hand, it is clear from a Fubini argument that
degenerate manifolds. (We remark that only very almost all translates of any given subspace are
few results in this set-up can be obtained by the extremal. In (Kleinbock 2003) the method of
standard methods, see, e.g., (Beresnevich and (Kleinbock and Margulis 1998) was pushed fur-
Velani 2007)). The proof of this stronger statement ther to produce criteria for the extremality, as well
is based on using the multi-parameter action of as the strong extremality, of arbitrary affine sub-
spaces ℒ of ℝn. Further, it was shown that if ℒ is
gt ¼ diagðet1 þþtn , et1 , . . . , etn Þ, where extremal (resp. strongly extremal), then so is any
smooth submanifold of ℒ which is nondegenerate
t ¼ ðt1 , . . . , tn Þ
in ℒ at a.e. point. (The latter property is a straight-
forward generalization of the definition of non-
instead of gt considered in the previous section.
degeneracy in ℝn: a map f is nondegenerate in ℒ at
One can show that the choice h(x) ¼ gtLf(x) allows
x if the linear part of ℒ is spanned by partial
one to verify (i) and (ii) uniformly in t  ℝnþ , and
derivatives of f at x.) In other words, extremality
the proof is finished by applying a multi- and strong extremality pass from affine subspaces
parameter version of the correspondence to their nondegenerate submanifolds.
described in section “Connection with Dynamics A more precise analysis makes it possible to
on the Space of Lattices.” Namely, one can show study Diophantine exponents of measures with
that y  VWA 1,n if and only if the trajectory supports contained in arbitrary proper affine sub-
gt Ly ℤk : t  ℝnþ grows linearly, that is, for spaces of ℝn. Namely, in (Kleinbock 2008) it is
some γ > 0, one has dist(gtLYℤnþ1, ℤnþ1) > γ k tk shown how to compute o(ℒ) for any ℒ and fur-
for an unbounded set of t  ℝnþ . A similar corre- thermore proved that if m is a Federer measure on
spondence was used in (Einsiedler et al. 2006) to ℝd, U an open subset of ℝd, and f : U ! ℝn a
prove that the (conjecturally empty) set of excep- continuous map such that the pair (f, v) is good
tions to Littlewood’s Conjecture, which, using the and nonplanar in ℒ, then o(fn) ¼ o(ℒ). Here we
terminology introduced above, can be called say, generalizing the definition from section
badly multiplicatively approximable vectors: “Diophantine Approximation with Dependent
Quantities: The Set-Up,” that (f, v) is nonplanar
BA ¼ ℝn ∖ [ W 
def
n,1 n,1 cc,1=n in ℒ if for any ball B with n(B) > 0, the f-image of
c>0
B \ supp n is not contained in any proper affine
¼ y: inf jqjPðqy  pÞ > 0 , subspace of ℒ. (It is easy to see that for a smooth
q  ℤ∖f0g, p  ℤn
map f : U ! ℒ, (f, l) is good and nonplanar in ℒ
ð18Þ whenever f is nondegenerate in ℒ at a.e. point.) It
is worthwhile to point out that these new applica-
has Hausdorff dimension zero. This was done
tions require a strengthening of the measure esti-
using a measure rigidity result for the action of
mate described at the end of section “Diophantine
608 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

Approximation with Dependent Quantities: The assuming some invariance properties of the fractal
Set-Up”: it was shown in (Kleinbock 2008) that subset, one can aim at divergence Khintchine-type
(i) and (ii) would still imply (iii) if r in (ii) is theorems, such as proving that the intersection of
replaced by rdim V. See (Huang 2022) for the BA with the fractal is null with respect to some
most up-to-date account of what is known about natural measure supported on the fractal. See
Diophantine exponents of affine subspaces. (Simmons and Weiss 2019; Khalil and Luethi
Another application concerns badly 2021) for several recent results in this direction.
approximable vectors. Using the dynamical The argument in these papers involves studying
description of the set BA  ℝn due to Dani nondivergence of random walks on the space of
(1985), it turns out to be possible to find badly lattices and is based on the work of Benoist and
approximable vectors inside supports of certain Quint.
measures on ℝn. Namely, if a subset K of ℝn The quantitative nondivergence method also
supports an absolutely friendly measure, then has applications to the problem of improving
BA \ K has Hausdorff dimension not less than Dirichlet’s Theorem (uniform approximation).
the Hausdorff dimension of this measure. In par- For example, in (Kleinbock and Weiss 2008)
ticular, it proves that limit measures of irreducible almost every point of any nondegenerate smooth
system of self-similar/self-conformal contractions manifold is proved not to lie in D ce,n=m for
satisfying the Open Set Condition, such as the
small enough ε depending only on the manifold.
middle-third Cantor set on the real line, contain
Some earlier work was done in (Baker 1976;
subsets of full Hausdorff dimension consisting of
Bugeaud 2002; Davenport and Schmidt 1970)
badly approximable vectors. This was established
and also in (Kleinbock and Weiss 2005b) for the
in (Kleinbock and Weiss 2005a) and later inde-
set of singular vectors. However, a different tech-
pendently in (Kristensen et al. 2006) using a dif-
nique based on Ratner’s Theorem and the Dani–
ferent approach. See also Fishman’s work
Margulis linearization method (Dani and
(Fishman 2006) for a stronger result involving
Margulis 1993) enabled Shah (2009) to prove a
winning sets of Schmidt games (the winning prop-
much stronger result, namely, the equidistribution
erty, due to the work of Schmidt (1966), implies
of gt - translates of orbits {Lf(x)ℤnþ1} for a real
full Hausdorff dimension and is stable with
analytic nondegenerate f. This was later extended
respect to countable intersections). Fishman’s
to multiplicative approximation (Shah 2010),
result has been significantly generalized in 2012
smooth nondegenerate maps (Shah and Yang
(Broderick et al. 2012) by showing that BA has
2022b), affine subspaces (Shah and Yang 2022a;
the hyperplane absolute winning (HAW) prop-
Kleinbock et al. 2021b; Chow and Yang 2019),
erty, which is stronger than winning and is
and submanifollds of Mm,n with min(m, n) > 1
inherited by supports of absolutely friendly mea-
(Yang 2016, 2020; Shah and Yang 2020).
sures. An even more striking development
It is also worthwhile to mention that a general-
occurred several years later, when Beresnevich
ization of the measure estimate discussed in section
(2015), Yang (2019), and then Beresnevich–
“Diophantine Approximation with Dependent
Nesharim–Yang (2022) showed that any non-
Quantities: The Set-Up” was used in (Bernik
degenerate real analytic curve in ℝn contains a
et al. 2001) to estimate the measure of the set of
winning set of badly approximable vectors. More-
points x in a ball B  ℝd for which the system
over, they did it in the context of approximation
with weights, thereby giving another proof of
j f ðxÞ  q þ p j< e
Scmidt’s conjecture fron (Schmidt 1982), origi-
nally proved by Badziahin, Pollington, and Velani j f 0 ð xÞ  q j < d
(2011). Note that quantitative nondivergence esti- jqi j < Qi , i ¼ 1, . . . , n,
mates play a crucial role in the argument of
(Beresnevich et al. 2022). where f is a smooth nondegenerate map B ! ℝn,
It is important to mention that for some special has a nonzero integer solution. For that, Lf(x) as
classes of fractals much more can be said: namely, in (15) has to be replaced by the matrix
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 609

1 0 f ð xÞ Finally let us mention another incarnation of

0 1 f 0 ð xÞ , Diophantine approximation with dependent quan-
tities, in which the correspondence between
0 0 In
approximation and dynamics proved to be help-
ful: intrinsic approximation on manifolds; that is,
and therefore (i) and (ii) turn into more compli-
approximating points on the manifold by rational
cated conditions, which nevertheless can be
points lying on the same manifold (as opposed to
checked when f is smooth and nondegenerate
ambient approximation described in the next sec-
and v is Lebesgue measure. This has resulted in
tion). See (Fishman et al. 2018) for some general
proving the convergence case of Khintchine–
results and (Kleinbock and Merrill 2015; Fishman
Groshev Theorem for nondegenerate manifolds
et al. 2022; Kleinbock and de Saxcé 2018) for a
(Bernik et al. 2001), in both standard and multi-
complete theory when the manifold is a quadric
plicative versions. The aforementioned estimate
hypersurface; in the latter case, dynamics takes
was also used in (Beresnevich et al. 2002) for the
place on the space of lattices in the light cone of
proof of the divergence case and in (Ghosh
the corresponding quadratic form. See also some
2005, 2006) for establishing the convergence
earlier work by Druţu (2005).
case of Khintchine–Groshev Theorem for affine
hyperplanes and their nondegenerate sub-
manifolds. This generalized results obtained by
Future Directions
standard methods for the curve (12) by Bernik
and Beresnevich (Beresnevich 2002; Bernik
Interactions between ergodic theory and number
1984).
theory have been rapidly expanding during the
Note also that in many of the problems men-
last two decades, and the author has no doubts
tioned above, the ground field ℝ can be replaced
that new applications of dynamics to Diophantine
by ℚp, and in fact several fields can be taken
approximation will emerge in the near future.
simultaneously, thus giving rise to the
Specializing to the topics discussed in the present
S-arithmetic setting where S ¼ {p1, . . ., ps} is a
paper, it is fair to say that the list of “further
finite set of normalized valuations of ℚ, which
results” contained in the previous section is by
may or may not include the infinite valuation
no means complete, and many even further results
(cf. [Sprindžuk 1969; Želudeviĉ 1986]). The
are currently in preparation. A lot should be
space of lattices in ℝnþ1 is replaced there by the
expected from recent developments in Parametric
space of lattices in ℚnþ1
S , where ℚS is the product Geometry of Numbers, that is, studying statistics
of the fields ℝ and ℚp1 , . . . , ℚps : This is the
of higher minima of lattices gtLYℤk (Das et al.
subject of the paper (Kleinbock and Tomanov
2019; Solan 2021). Let us mention two more
2007), where S-arithmetic analogues of many
directions which have been quite actively pursued
results reviewed in section “Diophantine Approx-
in recent years: the problem of effectivization of
imation with Dependent Quantities: The Set-Up”
non-effective results in Diophantine approxima-
have been established. Similarly one can consider
tion, such as Margulis’ solution of the Oppenheim
versions of the above theorems over local fields of
Conjecture (see, e.g., [Lindenstrauss and Margulis
positive characteristic (Ghosh 2007). See also
2014; Strömbergsson and Vishe 2020]), and
Kleinbock (2004) where Sprindzuk’s solution
replacing the supremum norm with arbitrary
(Sprindžuk 1969) of the complex case of Mahler’s
norms in uniform approximation problems (the
Conjecture has been generalized (the latter
latter hinges on a careful study of critical loci for
involves studying small values of linear forms
those norms, see [Andersen and Duke 2021;
with coefficients in ℂ at real integer points), and
Kleinbock and Rao 2022a, b, c]). Several interest-
(Einsiedler and Kleinbock 2007) which estab-
ing open directions are listed in (Gorodnik 2007),
lishes a p-adic analogue of the result of
Section 9, in the final sections of papers
(Einsiedler et al. 2006) on the set of exceptions
(Beresnevich et al. 2002; Kleinbock et al. 2004),
to Littlewood’s Conjecture.
in the book (Bugeaud 2002), and in surveys by
610 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

Frantzikinakis–McCutcheon, Nitica, and Ward in Dani SG (1985) Divergent trajectories of flows on homo-
this volume. geneous spaces and diophantine approximation.
J Reine Angew Math 359:55–89
Dani SG (1986) On orbits of unipotent flows on homoge-
Acknowledgments The work on this paper was neous spaces. II. Ergodic Theory Dyn Syst 6:167–182
supported in part by NSF Grants DMS-0239463 and Dani SG, Margulis GA (1993) Limit distributions of orbits
DMS-1900560. of unipotent flows and values of quadratic forms. In: IM
Gelfand Seminar. American Mathematical Society,
Providence, pp 91–137
Davenport H, Schmidt WM (1969/1970) Dirichlet’s theo-
Bibliography rem on diophantine approximation. II. Acta Arith 16:
413–424
Baker A (1975) Transcendental number theory. Cambridge Davenport H, Schmidt WM (1970) Dirichlet’s theorem on
University Press, London diophantine approximation. In: Symposia
Baker RC (1976) Metric diophantine approximation on Mathematica. INDAM, Rome, pp 113–132
manifolds. J Lond Math Soc 14:43–48 Ding J (1994) A proof of a conjecture of C. L. Siegel.
Baker RC (1978) Dirichlet’s theorem on diophantine J Number Theory 46:1–11
approximation. Math Proc Camb Philos Soc 83:37–59 Dodson MM (1992) Hausdorff dimension, lower order and
Bekka M, Mayer M (2000) Ergodic theory and topological Khintchine’s theorem in metric Diophantine approxi-
dynamics of group actions on homogeneous spaces. mation. J Reine Angew Math 432:69–76
Cambridge University Press, Cambridge Dodson MM (1993) Geometric and probabilistic ideas in
Beresnevich V (1999) On approximation of real numbers the metric theory of Diophantine approximations.
by real algebraic numbers. Acta Arith 90:97–112 Uspekhi Mat Nauk 48:77–106
Beresnevich V (2002) A Groshev type theorem for con- Druţu C (2005) Diophantine approximation on rational
vergence on manifolds. Acta Math Hungar 94:99–130 quadrics. Ann Math 333:405–469
Beresnevich V, Velani S (2007) A note on simultaneous Duffin RJ, Schaeffer AC (1941) Khintchine’s problem in
Diophantine approximation on planar curves. Ann metric Diophantine approximation. Duke Math J 8:
Math 337:769–796 243–255
Beresnevich V, Bernik VI, Kleinbock D, Margulis GA Einsiedler M, Kleinbock D (2007) Measure rigidity and
(2002) Metric Diophantine approximation: the p-adic Littlewoodtype problems. Compos Math 143:
Khintchine-Groshev theorem for nondegenerate mani- 689–702
folds. Moscow Math J 2:203–225 Einsiedler M, Lindenstrauss E (2006) Diagonalizable
Beresnevich V, Dickinson H, Velani S (2006) Measure flows on locally homogeneous spaces and number the-
theoretic laws for lim sup sets. Memoirs Am Math ory. In: Proceedings of the international congress of
Soc 179:1–91 mathematicians. Eur Math Soc, Zurich, pp 1731–1759
Beresnevich V, Dickinson H, Velani S (2007) Diophantine Einsiedler M, Katok A, Lindenstrauss E (2006) Invariant
approximation on planar curves and the distribution of measures and the set of exceptions to Littlewood’s
rational points. Ann Math 166:367–426 conjecture. Ann Math 164:513–560
Bernik VI (1984) A proof of Baker’s conjecture in the Eskin A (1998) Counting problems and semisimple
metric theory of transcendental numbers. Dokl Akad groups. In: Proceedings of the international congress
Nauk SSSR 277:1036–1039 of mathematicians. Doc Math, Berlin, pp 539–552
Bernik VI, Dodson MM (1999) Metric Diophantine Eskin A, Margulis GA, Mozes S (1998) Upper bounds and
approximation on manifolds. Cambridge University asymptotics in a quantitative version of the Oppenheim
Press, Cambridge conjecture. Ann Math 147:93–141
Bernik VI, Kleinbock D, Margulis GA (2001) Khintchine- Eskin A, Margulis GA, Mozes S (2005) Quadratic forms of
type theorems on manifolds: convergence case for stan- signature (2,2) and eigenvalue spacings on rectangular
dard and multiplicative versions. Int Math Res Notices 2-tori. Ann Math 161:679–725
2001:453–486 Fishman L (2006) Schmidt’s games on certain fractals.
Besicovitch AS (1929) On linear sets of points of fractional Israel S Math (to appear)
dimensions. Ann Math 101:161–193 Gallagher P (1962) Metric simultaneous diophantine
Bugeaud Y (2002) Approximation by algebraic integers and approximation. J Lond Math Soc 37:387–390
Hausdorff dimension. J Lond Math Soc 65:547–559 Ghosh A (2005) A Khintchine type theorem for hyper-
Bugeaud Y (2004) Approximation by algebraic numbers. planes. J Lond Math Soc 72:293–304
Cambridge University Press, Cambridge Ghosh A (2006) Dynamics on homogeneous spaces and
Cassels JWS (1957) An introduction to Diophantine Diophantine approximation on manifolds. PhD thesis,
approximation. Cambridge University Press, New York Brandeis University, Waltham
Cassels JWS, Swinnerton-Dyer H (1955) On the product of Ghosh A (2007) Metric Diophantine approximation over a
three homogeneous linear forms and the indefinite ternary local field of positive characteristic. J Number Theory
quadratic forms. Philos Trans R Soc Lond 248:73–96 124:454–469
Dani SG (1979) On invariant measures, minimal sets and a Gorodnik A (2007) Open problems in dynamics and
lemma of Margulis. Invent Math 51:239–260 related fields. J Mod Dyn 1:1–35
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 611

Groshev AV (1938) Une théoréme sur les systémes des Kleinbock D, Lindenstrauss E, Weiss B (2004) On fractal
formes linéaires. Dokl Akad Nauk SSSR 9:151–152 measures and diophantine approximation. Selecta Math
Harman G (1998) Metric number theory. Clarendon Press/ 10:479–523
Oxford University Press, New York Kontsevich M, Suhov Y (1999) Statistics of Klein polyhe-
Hutchinson JE (1981) Fractals and self-similarity. Indiana dra and multidimensional continued fractions. In:
Univ Math J 30:713–747 Pseudoperiodic topology. American Mathematical
Jarnik V (1928–9) Zur metrischen Theorie der Society, Providence, pp 9–27
diophantischen Approximationen. Prace Mat-Fiz 36: Kristensen S, Thorn R, Velani S (2006) Diophantine
91–106 approximation and badly approximable sets. Adv
Jarnik V (1929) Diophantischen Approximationen und Math 203:132–169
Hausdorffsches Mass. Mat Sb 36:371–382 Lagarias JC (1994) Geodesic multidimensional continued
Khanin K, Lopes-Dias L, Marklof J (2007) Multi- fractions. Proc Lond Math Soc 69:464–488
dimensional continued fractions, dynamical Lindenstrauss E (2007) Some examples how to use mea-
renormalization and KAM theory. Commun Math sure classification in number theory. In:
Phys 270:197–231 Equidistribution in number theory, an introduction.
Khintchine A (1924) Einige Satze uber Kettenbruche, mit Springer, Dordrecht, pp 261–303
Anwendungen auf die Theorie der Diophantischen Mahler K (1932) Uber das Mass der Menge aller S-Zahlen.
Approximationen. Math Ann 92:115–125 Math Ann 106:131–139
Khintchine A (1963) Continued fractions. P Noordhoff Margulis GA (1975) On the action of unipotent groups in
Ltd, Groningen the space of lattices. In: Lie groups and their represen-
Kleinbock D (2001) Some applications of homogeneous tations (Budapest, 1971). Halsted, New York,
dynamics to number theory. In: Smooth ergodic theory pp 365–370
and its applications. American Mathematical Society, Margulis GA (1989) Discrete subgroups and ergodic the-
Providence, pp 639–660 ory. In: Number theory, trace formulas and discrete
Kleinbock D (2003) Extremal subspaces and their sub- groups (Oslo, 1987). Academic, Boston, pp 377–398
manifolds. Geom Funct Anal 13:437–466 Margulis GA (1997) Oppenheim conjecture. In: Fields
Kleinbock D (2004) Baker-Sprindžuk conjectures for com- medalists’ lectures. World Sci Publishing, River Edge,
plex analytic manifolds. In: Algebraic groups and pp 272–327
arithmetic. Tata Inst Fund Res, Mumbai, pp 539–553 Margulis GA (2002) Diophantine approximation, lattices
Kleinbock D (2008) An extension of quantitative non- and flows on homogeneous spaces. In: A panorama of
divergence and applications to Diophantine exponents. number theory or the view from Baker’s garden. Cam-
Trans AMS 360:6497–6523 bridge University Press, Cambridge, pp 280–310
Kleinbock D, Margulis GA (1996) Bounded orbits of non- Mauldin D, Urbański M (1996) Dimensions and measures
quasiunipotent flows on homogeneous spaces. In: in infinite iterated function systems. Proc Lond Math
Sina?‘s Moscow seminar on dynamical systems. Amer- Soc 73:105–154
ican Mathematical Society, Providence, pp 141–172 Mohammadi A, Salehi Golsefidy A (2008) S-arithmetic
Kleinbock D, Margulis GA (1998) Flows on homogeneous Khintchine-type theorem. Preprint
spaces and Diophantine approximation on manifolds. Moore CC (1966) Ergodicity of flows on homogeneous
Ann Math 148:339–360 spaces. Am J Math 88:154–178
Kleinbock D, Margulis GA (1999) Logarithm laws for Roth KF (1955) Rational approximations to algebraic
flows on homogeneous spaces. Invent Math 138: numbers. Mathematika 2:1–20
451–494 Schmidt WM (1960) A metrical theorem in diophantine
Kleinbock D, Tomanov G (2007) Flows on S? Arithmetic approximation. Can J Math 12:619–631
homogeneous spaces and applications to metric Schmidt WM (1964) Metrische Satze uber simultane
Diophantine approximation. Comment Math Helv 82: Approximation abhangiger Grosen. Monatsh Math
519–581 63:154–166
Kleinbock D, Weiss B (2005a) Badly approximable vec- Schmidt WM (1969) Badly approximable systems of linear
tors on fractals. Israel J Math 149:137–170 forms. J Number Theory 1:139–154
Kleinbock D, Weiss B (2005b) Friendly measures, homo- Schmidt WM (1972) Norm form equations. Ann Math 96:
geneous flows and singular vectors. In: Algebraic and 526–551
topological dynamics. American Mathematical Soci- Schmidt WM (1980) Diophantine approximation.
ety, Providence, pp 281–292 Springer, Berlin
Kleinbock D, Weiss B (2008) Dirichlet’s theorem on Sheingorn M (1993) Continued fractions and congruence
diophantine approximation and homogeneous flows. subgroup geodesics. In: Number theory with an empha-
J Mod Dyn 2:43–62 sis on the Markoff spectrum (Provo, UT, 1991). Dek-
Kleinbock D, Shah N, Starkov A (2002) Dynamics of ker, New York, pp 239–254
subgroup actions on homogeneous spaces of Lie Sprindžuk VG (1964) More on Mahler’s conjecture. Dokl
groups and applications to number theory. In: Hand- Akad Nauk SSSR 155:54–56
book on dynamical systems, vol 1A. Elsevier Science, Sprindžuk VG (1969) Mahler’s problem in metric number
North Holland, Amsterdam, pp 813–930 theory. American Mathematical Society, Providence
612 Ergodic Theory on Homogeneous Spaces and Metric Number Theory

Sprindžuk VG (1979) Metric theory of Diophantine Chow S, Yang L (2019) An effective Ratner
approximations. VH Winston & Sons, Washington, DC equidistribution theorem for multiplicative
Sprindžuk VG (1980) Achievements and problems of the Diophantine approximation on planar lines. Preprint
theory of Diophantine approximations. Uspekhi Mat arXiv:1902.06081
Nauk 35:3–68 Das T, Fishman L, Simmons D, Urbański M (2018)
Starkov A (2000) Dynamical systems on homogeneous Extremality and dynamically defined measures, part I:
spaces. American Mathematical Society, Providence Diophantine properties of quasi-decaying measures.
Stratmann B, Urbanski M (2006) Diophantine extremality Selecta Math 24:2165–2206
of the Patterson measure. Math Proc Camb Philos Soc Das T, Fishman L, Simmons D, Urbański M (2019)
140:297–304 A variational principle in the parametric geometry of
Sullivan D (1982) Disjoint spheres, approximation by numbers. Preprint arXiv:1901.06602
imaginary quadratic numbers, and the logarithm law Dolgopyat D, Fayad B, Vinogradov I (2017) Central limit
for geodesics. Acta Math 149:215–237 theorems for simultaneous Diophantine approxima-
Urbanski M (2005) Diophantine approximation of self- tions. J Éc polytech Math 4:1–36
conformal measures. J Number Theory 110:219–235 Eskin A, Mozes S (2022) Margulis functions and their
Želudeviĉ F (1986) Simultane diophantische applications. In: Dynamics, geometry, number theory:
Approximationen abhangiger Grossen in mehreren the impact of Margulis on modern mathematics. Uni-
Metriken. Acta Arith 46:285–296 versity of Chicago Press, Chicago, pp 342–361
Fishman L, Kleinbock D, Merrill K, Simmons D (2018)
Intrinsic Diophantine approximation on manifolds:
New List
general theory. TAMS 370:577–599
Aka M, Breuillard E, Rosenzweig L, de Saxcé N (2018)
Fishman L, Kleinbock D, Merrill K, Simmons D (2022)
Diophantine approximation of matrices and Lie groups.
Intrinsic Diophantine approximation on quadric hyper-
Geom Funct Anal 28:1–57
surfaces. JEMS 24:1045–1101
Andersen N, Duke W (2021) On a theorem of Davenport
Huang JJ (2022) Extremal affine subspaces and
and Schmidt. Acta Arith 198:37–75
Khintchine-Jarník type theorems. Preprint,
Athreya J, Ghosh A, Tseng J (2015) Spiraling of approxi-
arXiv:2208.04255
mations and spherical averages of Siegel transforms.
Kadyrov S, Kleinbock D, Lindenstrauss E, Margulis GA
J Lond Math Soc 91:383–404
(2017) Singular systems of linear forms and non-escape
Badziahin D, Pollington A, Velani S (2011) On a problem
of mass in the space of lattices. J Anal Math 133:
in simultaneous Diophantine approximation: Schmidt’s
253–277
conjecture. Ann Math 174:1837–1883
Khalil O, Luethi M (2021) Random walks, spectral gaps,
Beresnevich V (2015) Badly approximable points on mani-
and Khintchine’s theorem on fractals. Preprint
folds. Inv Math 202:1199–1240
arXiv:2101.05797
Beresnevich V, Kleinbock D (2022) Quantitative non-
Kim T, Kim W (2022) Hausdorff measure of sets of
divergence and Diophantine approximation on mani-
Dirichlet non-improvable affine forms. Adv Math
folds. In: Dynamics, geometry, number theory: the
403:Paper No. 108353
impact of Margulis on modern mathematics. University
Kleinbock D (2010) Quantitative nondivergence and its
of Chicago Press, Chicago, pp 303–341
Diophantine applications. Lecture notes for Clay Insti-
Beresnevich V, Kleinbock D, Margulis GA (2015) Non-
tute Summer School (Pisa 2007). In: Homogeneous
planarity and metric Diophantine approximation for
flows, moduli spaces and arithmetic, Clay Math Proc,
systems of linear forms. J Number Theory Bordeaux
pp 131–153
27:1–31
Kleinbock D, de Saxcé N (2018) Rational approximation
Beresnevich V, Guan L, Marnat A, Ramirez F, Velani
on quadrics: a simplex lemma and its consequences.
S (2020) Dirichlet is not just bad and singular. Preprint
Enseign Math 64:459–476
arXiv:2008.04043
Kleinbock D, Merrill K (2015) Rational approximation on
Beresnevich V, Nesharim E, Yang L (2022) Winning prop-
spheres. Israel J Math 209:293–322
erty of badly approximable points on curves. Duke
Kleinbock D, Mirzadeh S (2020) On the dimension drop
Math J, to appear
conjecture for diagonal flows on the space of lattices.
Björklund M, Gorodnik A (2019) Central limit theorems
Adv Math, to appear
for Diophantine approximants. Math Ann 374:
Kleinbock D, Rao A (2022a) A zero-one law for uniform
1371–1437
Diophantine approximation in Euclidean norm. Int
Broderick R, Kleinbock D, Fishman L, Reich A, Weiss
Math Res Not 2022:5617–5657
B (2012) The set of badly approximable vectors is
Kleinbock D, Rao A (2022b) Abundance of Dirichlet-
strongly C1 incompressible. Math Proc Camb Philos
improvable pairs with respect to arbitrary norms.
Soc 153:319–339
Mosc J Comb Number Theory 11:97–114
Cheung Y (2011) Hausdorff dimension of the set of singu-
Kleinbock D, Rao A (2022c) Weighted uniform
lar pairs. Ann Math 173:127–167
Diophantine approximation of systems of linear
Cheung Y, Chevallier N (2016) Hausdorff dimension of
forms. Pure Appl Math Q 18:1095–1112
singular vectors. Duke Math J 165:2273–2329
Ergodic Theory on Homogeneous Spaces and Metric Number Theory 613

Kleinbock D, Wadleigh N (2018) A zero-one law for Shah N (2010) Expanding translates of curves and
improvements to Dirichlet’s theorem. PAMS 146: Dirichlet-Minkowski theorem on linear forms. JAMS
1833–1844 23:563–589
Kleinbock D, Wadleigh N (2019) An inhomogeneous Shah N, Yang L (2020) Equidistribution of curves in
Dirichlet theorem via shrinking targets. Compos Math homogeneous spaces and Dirichlet’s approximation
155:1402–1423 theorem for matrices. Discrete Contin Dyn Syst 40:
Kleinbock D, Weiss B (2013) Modified Schmidt games 5247–5287
and a conjecture of Margulis. J Mod Dyn 7:429–460 Shah N, Yang P (2022a) Equidistribution of expanding
Kleinbock D, Margulis GA, Wang J (2010) Metric degenerate manifolds in the space of lattices. Preprint
Diophantine approximation for systems of linear arXiv:2112.13952
forms via dynamics. Int J Number Theory 6:1139–1168 Shah N, Yang P (2022b) Equidistribution of non-uniformly
Kleinbock D, Strombergsson A, Yu S (2021a) Measure stretching translates of shrinking smooth curves and
estimates and improvements to Dirichlet’s theorem. weighted Dirichlet approximation. Preprint
Proc. London Math Soc, to appear arXiv:2204.03194
Kleinbock D, de Saxcé N, Shah N, Yang P (2021b) Shapira U, Weiss B (2022) Geometric and arithmetic aspects
Equidistribution in the space of 3-lattices and of approximation vectors. Preprint arXiv:2206.05329
Dirichlet-improvable vectors on planar lines. Ann Sc Simmons D, Weiss B (2019) Random walks on homoge-
Norm Super Pisa, to appear neous spaces and Diophantine approximation on frac-
Lindenstrauss E, Margulis G (2014) Effective estimates on tals. Invent Math 216:337–394
indefinite ternary forms. Israel J Math 203:445–499 Solan O (2021) Parametric geometry of numbers with
Roy D (2015) On Schmidt and Summerer parametric general flow. Preprint, arXiv:2106.01707
geometry of numbers. Ann Math 182:739–786 Strömbergsson A, Vishe P (2020) An effective
Schmidt WM (1966) On badly approximable numbers and equidistribution result for SL(2,R)⋉(R2) k and appli-
certain games. TAMS 123:78–199 cation to inhomogeneous quadratic forms. J Lond Math
Schmidt WM (1982) Open problems in Diophantine Soc 102:143–204
approximation. In: Diophantine approximations and Yang L (2016) Equidistribution of expanding curves in
transcendental numbers. Progr. Math., 31, Birkhäuser homogeneous spaces and Diophantine approximation
Boston, Boston, MA, pp 271–287 on square matrices. PAMS 144:5291–5308
Schmidt WM, Summerer L (2013) Diophantine approxi- Yang L (2019) Badly approximable points on manifolds
mation and parametric geometry of numbers. Monatsh and unipotent orbits in homogeneous spaces. Geom
Math 169:51–104 Funct Anal 29:1194–1234
Shah N (2009) Equidistribution of expanding translates of Yang P (2020) Equidistribution of expanding translates of
curves and Dirichlet’s theorem on Diophantine approx- curves and Diophantine approximation on matrices.
imation. Invent Math 177:509–532 Invent Math 220:909–948
 Measure rigidity Measure rigidity refers to the
Ergodic Theory: Rigidity study of invariant measures for actions of abe-
lian groups and semigroups.
Viorel Niţică
West Chester University, West Chester, PA, USA
Institute of Mathematics, Bucharest, Romania Definition of the Subject

As one can see from this volume, chaotic behavior

Article Outline of complex dynamical systems is prevalent in
nature and in large classes of transformations.
Glossary Rigidity theory can be viewed as the counterpart
Definition of the Subject to the generic theory of dynamical systems which
Introduction often investigates chaotic dynamics for a typical
Basic Definitions and Examples transformation belonging to a large class. In rigid-
Differentiable Rigidity ity one is interested in finding obstructions to
Local Rigidity chaotic, or generic, behavior. This often leads to
Global Rigidity rather unexpected classification results. As such,
Measure Rigidity rigidity in dynamics and ergodic theory is difficult
Future Directions to define precisely and the best approach to this
Bibliography subject is to study various results and themes that
developed so far. A classification is offered below
Glossary in local, global, differentiable and measurable
rigidity. One should note that all branches are
Ck Conjugacy Two diffeomorphisms f1, f2 act- strongly intertwined and, at this stage of the
ing on the same manifold M are said to be Ck- development of the subject, it is difficult to
conjugated if there exists a Ck diffeomorphism separate them.
h of M such that f1 ¼ h1  f2  h. The Rigidity is a well developed and prominent
diffeomorphism h is called Ck conjugacy. topic in modern mathematics. Historically, rigid-
Conjugacy Two elements g1, g2 in a group G are ity has two main origins, one coming from the
said to be conjugated if there exists an element study of lattices in semi-simple Lie groups, and
h  G such that g1 ¼ h1g2h. The element h is one coming from the theory of hyperbolic dynam-
called conjugacy. ical systems. From the start, ergodic theory was an
Differentiable rigidity Differentiable rigidity important tool used to prove rigidity results, and a
refers to finding invariants to the differentiable strong interdependence developed between these
conjugacy of dynamical systems, and, more fields. Many times a result in rigidity is obtained
general, group actions. by combining techniques from the theory of lat-
Global rigidity Global rigidity refers to the clas- tices in Lie groups with techniques from hyper-
sification of all group actions on manifolds bolic dynamical systems and ergodic theory.
satisfying certain conditions. Among other mathematical disciplines using
Lattice A lattice in a Lie group is a discrete results and contributing to this field one can men-
subgroup of finite covolume. tion representation theory, smooth, continuous
Local rigidity Local rigidity refers to the and measurable dynamics, harmonic and spectral
study of perturbations of homomorphisms analysis, partial differential equations, differential
from discrete or continuous groups into geometry, and number theory. Additional details
diffeomorphism groups. about the appearance of rigidity in ergodic theory
© Springer-Verlag 2009 615
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_185
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_185
616 Ergodic Theory: Rigidity

as well as definitions for some terminology used Zimmer (1987). The goal of the program was to
in the sequel can be found in the articles ▶ “Ergo- classify the smooth actions of higher rank semi-
dic Theory on Homogeneous Spaces and Metric simple Lie groups and of their (irreducible) lat-
Number Theory” by Kleinbock, ▶ “Ergodic The- tices on compact manifolds. It was expected that
ory: Recurrence” by Frantzikinakis, McCutcheon, any such lattice action that preserves a smooth
and ▶ “Ergodic Theory: Interactions with Combi- volume form and is ergodic can be reduced to
natorics and Number Theory” by Ward. The the- one of the following standard models: isometric
ory of hyperbolic dynamics is presented in the actions, linear actions on infranil manifolds, and
entry “Hyperbolic Dynamical Systems” by left translations on compact homogeneous spaces.
Viana and in the entry ▶ “Smooth Ergodic The- This original conjecture was disproved by Katok,
ory” by Wilkinson. Lewis (see (Katok and Lewis 1996)): by blowing
up a linear nilmanifold-action at some fixed points
they exhibit real-analytic, volume preserving,
Introduction ergodic lattice actions on manifolds with compli-
cated topology.
The first results about classification of lattices in Nevertheless, imposing extra assumptions on a
semi-simple Lie groups were local, aimed at try- higher rank action, for example the existence of
ing to understand the space of small perturbations some hyperbolicity, allows local and global clas-
of a given linear representation. A major contrib- sification results. The concept of Anosov action,
utor was Weil (1960, 1962, 1964), who proved that is, an action that contains at least one Anosov
local rigidity of linear representations for large diffeomorphism, was introduced for general
classes of groups, in particular lattices. Another groups by Pugh, Shub (1972). The significant
breakthrough was the contribution of Kazhdan differences between the classical ℤ and ℝ cases
(1967), who introduced property (T), allowing to and those of higher rank lattices, or at a more basic
show that large classes of lattices are finitely gen- level, of higher rank abelian groups, went
erated. Rigidity theory matured due to the remark- unnoticed for a while. The surge of activity in
able global rigidity results obtained by Mostow the 80s allowed these differences to surface: for
(1973) and Margulis (1991), leading to a complete lattices in the work of Hurder, Katok, Lewis,
classification of lattices in large classes of semi- Zimmer (see (Hurder 1992; Katok and Lewis
simple Lie groups. 1991, 1996; Katok et al. 1996)); and for higher
Briefly, a hyperbolic (Anosov) dynamical sys- rank abelian groups in the work of Katok, Lewis
tem is one that exhibits strong expansion and con- (1991) and Katok, Spatzier (1994a). As observed
traction along complementary directions. An early in these papers, local and global rigidity are typ-
contribution introducing this class of objects is the ical for such Anosov actions. This generated addi-
paper of Smale (1967), in which basic examples tional research which is summarized in section
and techniques are introduced. A breakthrough “Local Rigidity” and “Global Rigidity”.
came with the results of Anosov (1967), who pro- Differentiable rigidity is covered in section
ved structural stability of the Anosov systems and “Differentiable Rigidity”. An interesting problem
ergodicity of the geodesic flow on a manifold of is to find moduli for the Ck conjugacy, k  1,
negative curvature. Motivated by questions arising of Anosov diffeomorphisms and flows. This
in mathematical physics, chaos theory and other was tackled so far only for low dimensional
areas, hyperbolic dynamics emerged as one of the cases (n ¼ 2 for diffeomorphisms and n ¼ 3 for
major fields of contemporary mathematics. flows). Another direction that can be included
From the beginning, a major unsolved problem in here refers to finding obstructions for higher trans-
the field was the classification of Anosov verse regularity of the stable/unstable foliation of
diffeomorphisms and flows. a hyperbolic system. A spin-off of the research
In the 80s a change in philosophy occurred, done so far, which is of high interest by itself, and
partially motivated by a program introduced by has applications to local and global rigidity,
Ergodic Theory: Rigidity 617

consists of results lifting the regularity of solu- (2005), Niţică, Török (2002), and Spatzier (1995).
tions of cohomological equations over hyperbolic Among these, (Fisher 2006) concentrates mostly
systems. In turn, these results motivated a more on local and global rigidity, (Niţică and Török
careful study of analytic questions about lifting 2002) on differentiable rigidity, (Lindenstrauss
the regularity of real valued continuous functions 2005) on measure rigidity, and (Spatzier 1995)
that enjoy higher regularity along webs of folia- gives a general overview.
tions. We also include in this section rigidity Here is a word of caution for the reader. Many
results for cocycles over higher rank abelian times, instead of the most general results, we
actions. These are crucial to the proof of local present an example that contains the essence of
rigidity of higher rank abelian group actions. what is available. Also, several important facts
A more detailed presentation of the material rele- that should have been included, are left out. This
vant to differentiable rigidity can be found in the is because stating complete results would require
forthcoming monograph (Katok and Niţică). more space than allocated to this material. The
Measure rigidity refers to the study of invariant limited knowledge of the author also plays a role
measures under actions of abelian groups and semi- here. He apologizes for any obvious omissions
groups. If the actions are hyperbolic, higher-rank, and hopes that the bibliography will help fill
and satisfy natural algebraic and irreducibility the gaps.
assumptions, one expects the invariant measures
to be rare. This direction was started by a question
of Furstenberg, asking if any nonatomic probability Basic Definitions and Examples
measure on the circle, invariant and ergodic under
multiplications by 2 and 3, is the Lebesgue mea- A detailed introduction to the theory of Anosov
sure. An early contribution is that of Rudolph systems and hyperbolic dynamics is given in the
(1990), who answered positively if the action has monograph (Katok and Hasselblatt 1995). The
an element of strictly positive entropy. Katok, proofs of the basic results for diffeomorphisms
Spatzier (1996) extended the question to more stated below can be found there. The proofs for
general higher rank abelian actions, such as actions flows are similar. Surveys about hyperbolic
by linear automorphisms of tori and Weyl chamber dynamics in this volume are the article “Hyperbolic
flows. A related direction is the study of the invari- Dynamical Systems” by Viana and the entry
ant sets and measures under the action of horocycle ▶ “Smooth Ergodic Theory” by Wilkinson.
flows, where important progress was made by Consider a compact differentiable manifold
Ratner (1991a, b) and earlier by Margulis (1989, M and f : M ! M a C1 diffeomorphism. Let TM
1997; Dani and Margulis 1990). An application of be the tangent bundle of M, and Df : TM ! TM be
these results present in the last papers is the proof of the derivative of f. The map f is said to be an Anosov
the long standing Oppenheim’s conjecture, about diffeomorphism if there is a smooth Riemannian
the density of the values of the quadratic forms at metric k  k on M, which induces a metric dM called
integer points. Recent developments due to and adapted, a number l  (0, 1), and a continuous Df
Einsiedler, Katok, Lindenstrauss (2006) give a par- invariant splitting TM ¼ Es  Eu such that
tial answer to another outstanding conjecture in
number theory, Littlewood’s conjecture, and k Df v k  l k v k , v  Es , k Df 1 v k  l k v k , v  Eu ,
emphasize measure rigidity as one of a more prom-
ising directions in rigidity. More details are shown For each x  M there is a pair of embedded C1
in section “Measure Rigidity”. discs W sloc ðxÞ, W uloc ðxÞ, called the local stable
Four other recent surveys of rigidity theory, manifold and the local unstable manifold at x,
each one with a fair amount of overlap but also respectively, such that:
complementary in part to the present one, that
discuss various aspects of the field and its signif- 1. T x W sloc ðxÞ ¼ Es ðxÞ, T x W uloc ðxÞ ¼ Eu ðxÞ;
icance are written by Fisher (2006), Lindenstrauss 2. f W sloc ðxÞ  W sloc ðfxÞ, f 1 W uloc ðxÞ  W uloc f 1 x ;
618 Ergodic Theory: Rigidity

3. For any m  (l, 1), there exists a constant distributions Ecs ¼ Ec  Es and Ecu ¼ Ec  Es.
C > 0 such that for all n  ℕ, In general, all these foliations are only continuous,
but their leaves are differentiable.
d M ðf n x, f n yÞ  Cmn dM ðx, yÞ, for y  W sloc ðxÞ, Any Anosov diffeomorphism is structurally sta-
d M ðf n x, f n yÞ  Cmn dM ðx, yÞ, for y  W uloc ðxÞ: ble, that is, any C1 diffeomorphism that is C1 close
to an Anosov diffeomorphism is topologically con-
The local stable (unstable) manifolds can be jugate to the unperturbed one via a Hölder homeo-
extended to global stable (unstable) manifolds morphism. An Anosov flow is structurally stable in
W s(x) and W u(x) which are well defined and the orbit equivalence sense: any C1 small perturba-
smoothly injectively immersed. These global tion of an Anosov flow has the orbit foliation topo-
manifolds are the leaves of global foliations W s logically conjugate via a Hölder homeomorphism
and Wu of M. In general, these foliations are only to the orbit foliation of the unperturbed flow.
continuous, but their leaves are differentiable. Let SL(n, ℝ) be the group of all n-dimensional
Let f : ℝ M ! M be a C1 flow. The flow f is square matrices with real valued entries of determi-
said to be an Anosov flow if there is a Riemannian nant 1. Let SL(n, ℤ)  SL(n, ℝ) be the subgroup
metric k  k on M, a constant 0 < l < 1, and a with integer entries. Basic examples of Anosov
continuous Df invariant splitting TM ¼ diffeomorphisms are automorphisms of the n-torus
Es  E0  Eu such that for all x  M and t > 0: n ¼ ℝn =ℤn induced by hyperbolic matrices in
SL(n, ℤ). A hyperbolic matrix is one that has only
1. dtd t¼0 ft  Ecx ∖f0g, dim Ecx ¼ 1, nonzero eigenvalues, all away in absolute value from
2. kDftv k  lt k v k, v  Es, 1. A specific example of such matrix in SL(2, ℤ) is
3. kDftv k  lt k v k, v  Eu.
2 1
:
For each x  M there is a pair of embedded C 1 1 1
discs W sloc ðxÞ, W uloc ðxÞ, called the local (strong)
stable manifold and the local (strong) unstable Basic examples of Anosov flows are given by
manifold at x, respectively, such that: the geodesic flows of surfaces of constant nega-
tive curvature. The unitary bundle of such a sur-
1. T x W sloc ðxÞ ¼ Es ðxÞ, T x W uloc ðxÞ ¼ Eu ðxÞ; face can be realized as M ¼ Γ\PSL(2, ℝ), where
PSL(2, ℝ) ¼ SL(2, ℝ)/{ 1} and Γ is a cocompact
ft W s ðxÞ  W sloc ðft xÞ,
2. t locu lattice in PSL(2, ℝ). The action of the geodesic
f W loc ðxÞ  W uloc ðft xÞ for t > 0; flow on M is induced by right multiplication with
3. For any m  (l, 1), there exists a constant elements in the diagonal one-parameter subgroup
C > 0 such that for all n  ℕ,
et=2 0
dM ðft x, ft yÞ  Cmt d M ðx, yÞ, ,tℝ :
0 et=2
for y  W sloc ðxÞ, t > 0,
t t
dM ðf x, f yÞ  Cmt d M ðx, yÞ, A related transformation, which is not hyper-
for y  W uloc ðxÞ, t > 0: bolic, but will be of interest in this presentation, is
the horocycle flowinduced by right multiplication
The local stable (unstable) manifolds can be on M by elements in the one parameter subgroup
extended to global stable (unstable) manifolds
W s(x) and W u(x). These global manifolds are the 1 t
leaves of global foliations W s and W u of M. One ,tℝ :
0 1
can also define weak stable and weak unstable folia-
tions with leaves given by W cs(x) ¼ [t  ℝ(W s(x)) Of interest in this survey are also actions of more
and W cu(x) ¼ [t  ℝ(Wu(x)), which have as tangent general groups than ℤ and ℝ. Typical examples of
Ergodic Theory: Rigidity 619

higher rank ℤk Anosov actions are constructed on translations on Γ\G descends to an A-action on
tori using groups of units in number fields. See N ≔ Γ\G/M. This action is called a Weyl chamber
(Katok et al. 2002) for more details about this flow. Any Weyl chamber flow is an Anosov
construction. A particular example of Anosov ℤ2- action, that is, has an element that acts hyperbol-
action on 3 is induced by the hyperbolic matrices: ically transversally to the orbit foliation of A. Note
that all maximal connected ℝ diagonalizable sub-
0 1 0 2 1 0 groups of G are conjugate and their common
A¼ 0 0 1 , B¼ 0 2 1 : dimension is called the ℝ-rank of G. If the
ℝ-rank k of G is higher than 2, then the Weyl
1 8 2 1 8 4
chamber flow is a higher rank hyperbolic ℝk-
action.
One can check, by looking at the eigenvalues,
An example of semi-simple Lie group is
that A and B are not multiples of the same matrix.
SL(n, ℝ). Let A be the diagonal subgroup of matri-
Moreover, A and B commute.
ces with positive entries in SL(n, ℝ). An example
Typical examples of higher rank Anosov ℝk-
of Weyl chamber flow that will be discussed in the
actions are given by Weyl chamber flows, which
sequel is the action of A by right translations on
we now describe using some notions from the
Γ\SL(n, ℝ), where Γ is a cocompact lattice. In this
theory of Lie groups. A good reference for the
case the centralizer M is trivial. The rank of this
background in Lie group theory necessary here is
action is n  1. The picture of the Weyl chambers
the book of Helgason (1978). Note that for a
for n ¼ 3 is shown in Fig. 1. The signs that appear
hyperbolic element of such an action the center
in each chamber are the signs of half of the
distribution is k-dimensional and coincides with
Lyapunov exponents of a regular element from
the tangent distribution to the orbit foliation of ℝk.
the chamber with respect to a certain fixed basis.
Let G be a semi-simple connected real Lie
For this action, the Lyapunov exponents appear in
group of the noncompact type, with Lie algebra
pairs of opposite signs.
g: Let K  G be a maximal compact subgroup that
An example of higher rank lattice Anosov
gives a Cartan decomposition g ¼ k þ p, where k
action that will be discussed in the sequel is the
is the Lie algebra of K and p is the orthogonal
standard action of SL(n, ℤ) on the torus
complement of k with respect to the Killing form
n , ðA, xÞ 7! Ax, A  SLðn, ℤÞ, x  n : SLðn, ℤÞ is
of g. Let a  p be a maximal abelian subalgebra
a (noncocompact!) lattice in SL(n, ℝ). As shown
and A ¼ exp a be the corresponding subgroup.
in (Katok and Lewis 1991), this action is gener-
The simultaneous diagonalization of ad g ðaÞ gives
ated by Anosov diffeomorphisms.
the decomposition

g¼gþ gl , g0 ¼ a þ m,
lL

where Λ is the set of restricted roots. The spaces

gl are called root spaces. A point H  a is called
regular if l(H ) 6¼ 0 for all l  Λ. Otherwise it is
called singular. The set of regular elements con-
sists of the complement of a union of finitely
many hyperplanes. Its components are cones in a
called Weyl chambers. The faces of the Weyl
chambers are called Weyl chamber walls.
Let M be the centralizer of A in K. Suppose Γ is
an irreducible torsion-free cocompact lattice in G. Ergodic Theory: Rigidity, Fig. 1 Weyl chambers for
Since A commutes with M, the action of A by right SL(3, ℝ)
620 Ergodic Theory: Rigidity

We describe now a class of dynamical systems Differentiable Rigidity

more general than the hyperbolic one. A C1
diffeomorphism f of a compact differentiable We start by reviewing cohomological results. Sev-
manifold M is called partially hyperbolic if there eral basic notions are already defined in section
exists a continuous invariant splitting of the tan- “Basic Definitions and Examples”. In this section
gent bundle TM ¼ Es  E0  Eu such the deriv- we assume that the cocycles are at least continu-
ative of f expands Eu much more than E0, and ous. A cocycle β : G M ! Γ over an action
contracts Es much more than E0. See (Brin and (g, x) 7! gx, g  G, x  M, is said to satisfy
Pesin 1974; Burns et al. 2001; Hirsch et al. 1977) closing conditions if for any g  G and x  M
for the theory of partially hyperbolic such that gx ¼ x, one has β(g, x) ¼ 1Γ. Note that
diffeomorphisms. Es and Eu are called stable, closing conditions are necessary in order for a
respectively unstable distributions, and are inte- cocycle to be cohomologous to the trivialone.
grable. E0 is called center distribution and, in Since a ℤ-cocycle is determined by a function
general, it is not integrable. A structural stability β : M ! Γ, β(x) ≔ β(1, x), the closing conditions
result proved by Hirsch, Pugh, Shub (Hirsch et al. become
1977), that is a frequently used tool in rigidity,
shows that, if E0 is integrable to a smooth folia- f n x ¼ x implies b f n1 x . . . bðxÞ
tion, then any perturbation f of f is partially hyper- ¼ 1G , ð3Þ
bolic and has an integrable center foliation.
Moreover, the center foliations of f of f are where f : M ! M is the function that implements
mapped one into the other by a homeomorphism the ℤ-action.
that conjugates the maps induced on the factor The first cohomological results for hyperbolic
spaces of the center foliations by f of systems were obtained by Livshits (1971, 1972).
f respectively. Let M be a compact Riemannian manifold with a
We review now basic facts about cocycles. ℤ-action implemented by a topologically transi-
These basic definitions refer to several regularity tive Anosov C1 diffeomorphism f. Then an
classes: measurable, continuous, or differentiable. α-Hölder function β : M ! ℝ determines a
Let G be a group acting on a set M, and denote the cocycle cohomologous to a trivial cocycle if and
action G M ! M by (g, x) 7! gx; thus (g1g2) only if β satisfies the closing conditions (3). The
x ¼ g1(g2x). Let Γ be a group with unit 1Γ. M is transfer map is α-Hölder. Moreover, for each
usually endowed with a measurable, continuous, Hölder class α and each finite dimensional Lie
or differentiable structure. A cocycle β over the group Γ, there is a neighborhood U of the identity
action is a function β : G M ! Γ such that in Γ such that an α-Hölder function β : M ! Γ
determines a cocycle cohomologous to the trivial
bðg1 g2 , xÞ ¼ bðg1 , g2 xÞbðg2 , xÞ, ð1Þ cocycle if and only if β satisfies the closing con-
ditions (3). The transfer map is again α-Hölder.
for all g1, g2  G, x  M. Note that any group Similar results are true for Anosov flows.
representation π : G ! Γ defines a cocycle called Using Fourier analysis, Veech (1986) extended
constant cocycle. The trivial representation Livshits’s result to real valued cocycles over
defines the trivial cocycle. ℤ-actions induced by ergodic endomorphisms of
A natural equivalence relation for the set of an n-dimensional torus, not necessarily
cocycles is given by cohomology. Two cocycles hyperbolic. For cocycles with values in abelian
β1, β2 : G M ! Γ are cohomologous if there groups, the question of two arbitrary cocycles
exists a map P : M ! Γ, called transfer map, such being cohomologous reduces to the question of
that an arbitrary cocycle being cohomologous to a
trivial one. This is not the case for cocycles with
b1 ðg, xÞ ¼ PðgxÞb2 ðg, xÞPðxÞ1 , ð2Þ
values in non-abelian groups. Parry (1999)
extended Livshits’s criteria to one for
for all g  G, x  M.
Ergodic Theory: Rigidity 621

cohomology of two arbitrary cocycles with values assumption is essential, as follows from a counter
in compact Lie groups. Parry’s result was gener- example found by de la Llave’s (1992).
alized by Schmidt (1999) to cocycles with values Useful tools in this development have been
in Lie groups that, in addition, satisfy a center results from analysis that lift the regularity of a
bunching condition. Niţică, Török (1995) continuous real valued function which is assumed
extended Livshits’s result to cocycles with values to have higher regularity along pairs of transverse
in the group Diff k(M ) of C k diffeomorphism of a Hölder foliations. Many times the foliations are
compact manifold M with stably trivial bundle, the stable and unstable ones associated to a hyper-
k  3. Examples of such manifolds are the tori and bolic system. Journé (1988) proved the Cn,α regu-
the spheres. In this case, the transfer map takes larity of a continuous function that is Cn,α along
values in Diff k3(M), and it is Hölder with respect two transverse continuous foliations with Cn,α
to a natural metric on Diff k3(M ). In (Niţică and leaves. If one is interested only in C1 regularity,
Török 1995) one can also find a generalization of a convenient alternative is a result of Hurder,
Livshits’s result to generic Anosov actions, that is, Katok (1990). This has a simpler proof and can
actions generated by families of Anosov be applied to the more general situation in which
diffeomorphisms that do not interchange the sta- the function is regular along a web of transverse
ble and unstable directions of elements in the foliations. A real analytic regularity result along
family. An example of such an action is the stan- these lines belongs to de la Llave (1997). In cer-
dard action of SL(n, ℤ) on the n-dimensional tain problems, for example when working with
torus. Weyl chamber flows, it is difficult to control the
A question of interest is the following: if two regularity in enough directions to span the whole
Ck cocycles, 1  k  o, over a hyperbolic action, tangent space. Nevertheless, the tangent space can
are cohomologous through a continuous/measur- be generated if one consider higher brackets of
able transfer map P, what can be said about the good directions. A C1 regularity result for this
higher regularity of P? For real valued cocycles case belongs to Katok, Spatzier (1994b). In order
the question can be reduced to one about to apply this result, the foliations need to be C1
cohomologically trivial cocycles. Livshits not only along the leaves, but also transversally.
showed that for a real valued C1 cocycle An application of the above regularity results is
cohomologous to a constant the transfer map is to questions about transverse regularity of the
C1. He also obtained C1 regularity results if the stable and unstable foliations of the geodesic
action is given by hyperbolic automorphisms of a flow on certain C1 surfaces of nonpositive curva-
torus. After preliminary results by Guillemin and ture. For compact negatively curved C1 surfaces,
Kazhdan for geodesic flows on surfaces of nega- E. Hopf showed that these foliations are C1, and it
tive curvature, for general hyperbolic systems the follows from the work of Anosov that individual
question was answered positively by de la Llave, leaves are C1. Hurder, Katok (1990) showed that
Marco, Moriyon (1986) in the C1 case and by de once the weak-stable and weak-unstable foliations
la Llave (1997) in the real analytic case. Niţică, of a volume-preserving Anosov flow on a com-
Török (1998) considered the lift of regularity for a pact 3-manifold are C2, they are C1.
transfer map between two cohomologous Another application of the regularity results is
cocycles with values in a Lie group or a to the study of invariants for Ck conjugacy of
diffeomorphism group. In contrast to the case of hyperbolic systems. By structural stability, a
cocycles cohomologous to trivial ones, here one small C1 perturbation of a hyperbolic system is
needs to require for the transfer map a certain C0 conjugate to the unperturbed one. The
amount of Hölder regularity that depends on the conjugacy, in general, is only Hölder. If the
ratio between the expansion/contraction that conjugacy is C1 then it preserves the eigenvalues
appears in the base and the expansion/contraction of the derivative at the periodic orbit. The follow-
introduced by the cocycle in the fiber. This ing two results describe the invariants of smooth
622 Ergodic Theory: Rigidity

and real analytic conjugacy of low dimensional 2003) which contains rigidity results for cocycles
hyperbolic systems. They are proved in a series of over (TNS) actions with values in compact Lie
papers written in various combinations by de la groups. In this situation the number of cohomol-
Llave, Marco, Moryion (de la Llave 1987, 1997; ogy classes is finite. An example of (TNS) action
de la Llave and Moriyon 1988; Marco and is given by the action of a maximal diagonalizable
Moriyon 1987a, b). subgroup of SL(n, ℤ) on n :
Let X, Y be two C1(Co) transitive Anosov Recently Damjanović, Katok (2005) devel-
vector fields on a compact three-dimensional oped a new method that was applied to the action
manifold. If they are C0 conjugate and the eigen- of the matrix diagonal group on Γ\SL(n, ℝ). They
values of the derivative at the corresponding peri- use techniques from (Katok and Kononenko
odic orbits are the same, then the conjugating 1996), where one finds cohomology invariants
homeomorphism is C1(Co). In particular, any for cocycles over partially hyperbolic actions
C1 conjugacy is C1(Co). that satisfy accessibility property. Accessibility
Assume now that f, g are two C1(Co) Anosov means that one can connect any two points from
diffeomorphisms on a compact two dimensional the manifold supporting the partially hyperbolic
manifold. If they are C0 conjugate and the eigen- dynamical system by transverse piecewise
values of the derivative at the corresponding peri- smooth paths included in stable/unstable leaves.
odic orbits are the same, then the conjugating This notion was introduced by Brin, Pesin (1974)
diffeomorphism is C1(Co). In particular, any C1 and it is playing a crucial role in the recent surge of
conjugacy is C1(Co). activity in the field of partially hyperbolic
An important direction was initiated by Katok, diffeomorphisms. See (Burns et al. 2001) for a
Spatzier (1994a) who studied cohomological recent survey of the subject. The cohomology
results over hyperbolic ℤk or ℝk-actions, k  2. invariants described in (Katok and Kononenko
They show that real valued smooth/Hölder 1996) are heights of the cocycle over cycles
cocycles over typical classes of hyperbolic ℤk or constructed in the base out of pieces inside sta-
ℝk, k  2, actions are smoothly/Hölder ble/unstable leaves. They provide a complete set
cohomologous to constants. These results cover, of obstructions for solving the cohomology equa-
in particular, actions by hyperbolic automor- tion. A new tool introduced in (Damianović and
phisms of a torus, and Weyl chamber flows. The Katok 2005) is algebraic K-theory (Milnor 1971).
proofs rely on harmonic analysis techniques, such The method can be extended to cocycles with
as Fourier transform and group representations for non-abelian range. In (Katok and Niţică 2007)
semi-simple Lie groups. one finds related results for small cocycles with
A geometric method for cocycle rigidity was values in a Lie group or the diffeomorphism group
developed in (Katok et al. 2000). One constructs a of a compact manifold.
differentiable form using invariant structures The equivalent of the Livshits theorem in the
along stable/unstable foliations, and the commu- higher-rank setting appears to be a description of
tativity of the action. The form is exact if and only the highest cohomology rather than the first coho-
if the cocycle is cohomologous to a constant one. mology. Indeed, for higher rank partially hyper-
The method covers actions on nilmanifolds satis- bolic actions of the torus, the intermediate
fying a condition called TNS (TotallyNon- cohomologies are trivial, while for the highest
Symplectic). This condition means that the action one the closing conditions characterize the coho-
is higher rank abelian hyperbolic, and that the mology classes. This behavior provides a gener-
tangent space is a direct sum of invariant distribu- alization of Veech cohomological result and of
tions, with each pair of these included in the stable Katok, Spatzier cohomological result for toral
distribution of a hyperbolic element of the action. automorphisms, and was discovered by
The method was also applied to small (i.e. close to A. Katok, S. Katok (1995, 2005).
identity on a set of generators) Lie group valued Flaminio, Forni (2003) studied the cohomolog-
cocycles. A related paper is (Niţică and Török ical equation over the horocycle flow. It is shown
Ergodic Theory: Rigidity 623

that there are infinitely many obstructions to the approach inspired Mostow (1973) to use the
existence of a smooth solution. Moreover, if these boundaries at infinity in his proof of strong rigid-
obstructions vanish, then one can solve the coho- ity of lattices, which in turn was crucial to the
mological equation. In (Forni 1997) it is shown a development of superrigidity due to Margulis
similar result for cocycles over area preserving (1991). See section “Global Rigidity” for more
flows on compact higher-genus surfaces under details.
certain assumptions that hold generically. Recall that hyperbolic dynamical systems are
Mieczkowski (2007) extended these techniques structurally stable. Thus they are, in a certain
and studied the cohomology of parabolic higher sense, locally rigid. We introduce now a precise
rank abelian actions. All these results rely on definition of local rigidity in the infinite-
noncommutative Fourier analysis, more specifi- dimensional setup. The fact that for general
cally representation theory of SL(2, ℝ) and group actions one needs to consider different reg-
SL(2, ℂ). ularities for the actions, perturbations and
conjugacies is apparent from the description of
structural stability for Anosov systems.
Local Rigidity A Ck action α of a finitely generated discrete
group Γ on a manifold M, that is, a homomor-
Let Γ be a finitely (or compactly) generated group, phism α : Γ ! Diff k(M ), is said to be Ck,l,p,r
G a topological group, and π : Γ ! G a homo- locally rigid if any Cl perturbation a which is C p
morphism. The target of local rigidity theory is to close to α on a family of generators, is Cr conju-
understand the space of perturbations of various gate to α, i.e. there exists a Cr diffeomorphism
homomorphisms π. Trivial perturbations of a H : M ! M which conjugates a to α, that is,
homomorphism arise from conjugation by an H ∘ aðgÞ ¼ aðgÞ ∘ H for all g  Γ. Note that for
arbitrary element of G. In order to rule them out, Anosov ℤ-actions, C1,1,1,0 rigidity is known as
one says that π is locally rigid if any nearby structural stability. One can also introduce the
homomorphism π0, (that is, π0, close to π on a finite notion of deformation rigidity if the initial action
or compact set of generators of Γ), is conjugate to and the perturbation are conjugate by a continuous
π by an element g  G, that is, π(γ) ¼ gπ0(γ)g1 path of diffeomorphisms that has an end coincid-
for all γ  Γ. If G is path-wise connected, one can ing to the identity.
also consider deformation rigidity, meaning that A weaker notion of local rigidity can be defined
any nearby continuous path of homomorphisms is in the presence of invariant foliations for the initial
conjugated to the initial one via a continuous path group action and for the perturbation. The map
of elements in G that has an end in the identity. H is now required to preserve the leaves of the
Initial results on local rigidity are about embed- foliations and to conjugate only after factorization
dings of lattices into semi-simple Lie groups. The by the invariant foliations. The importance of this
main results belong to Weil (1960, 1962, 1964). notion is apparent from the leaf wise conjugacy
He showed that if G is a semi-simple Lie group structural stability theorem of Hirsch, Pugh, Shub
that is not locally isomorphic to SL(2, ℝ) and if (1977). See section “Basic Definitions and Exam-
Γ  G is an irreducible cocompact lattice, then the ples”. Moreover, for Anosov flows this is the
natural embedding of Γ into G is locally rigid. natural notion of structural stability, and appears
Earlier results were obtained by Selberg (1960), by taking the invariant foliation to be the one-
Calabi, Vesentini (1960), and Calabi (1961). dimensional orbit foliation. For more general
Selberg proved the local rigidity of the natural actions, of lattices or higher rank abelian groups,
embedding of cocompact lattices into SL(n, ℝ). this property is often used in combination to
His proof used the dynamics of iterates of matri- cocycle rigidity in order to show local rigidity.
ces, in particular the existence of singular direc- We discuss more about this when we review
tions, or walls of Weyl chambers, in the maximal local rigidity results for partially hyperbolic
diagonalizable subgroups of SL(n, ℝ). Selberg’s actions.
624 Ergodic Theory: Rigidity

We summarize now several developments in (Niţică and Török 2001) shows that the action r is
local rigidity that emerged in the 80s. Initial C1,1,2,K1 locally rigid. Ingredients in the proof
results (Lewis 1991; Zimmer 1990) were about are two rigidity results, one about TNS actions,
infinitesimal rigidity, that is, a weaker version of and one about actions of property (T) groups.
local rigidity suitable for discrete groups repre- A locally compact group has property (T) if the
sentations in infinite-dimensional spaces of trivial representation is isolated in the Fell topol-
smooth vector fields. Then Hurder (1992) proved ogy. This means that if G acts on a Hilbert space
C1,1,1,1 deformation rigidity and Katok, Lewis, unitarily and it has almost invariant vectors, then it
Zimmer (Katok and Lewis 1991, 1996; Katok has invariant vectors. Hirsch–Pugh–Shub theorem
et al. 1996) proved C1,1,1,1 local rigidity of the implies that perturbations of abelian partially
standard action of SL(n, ℤ), n  3, on the hyperbolic actions of product type are conjugated
n-dimensional torus. In these results crucial use to skew-products of abelian Anosov actions via
was made of the presence of an Anosov element in cocycles with values in diffeomorphism groups. In
the action. Due to the uniqueness of the conjugacy addition, the TNS property implies that the sum of
coming from structural stability, one has a contin- the stable and unstable distributions of any regular
uous candidate for the conjugacy between the element of the perturbation is integrable. The
actions. Margulis, Qian (2001) used the existence leaves of the integral foliation are closed, covering
of a spanning family of directions that are hyper- the base simply. Thus one obtains a conjugacy
bolic for certain elements of the action to show between the perturbation and a product action.
local rigidity of partially hyperbolic actions that Property (T) is used to show that the conjugacy
are not hyperbolic. Another important tool present reduces the perturbed action to a family of pertur-
in many proofs is Margulis and Zimmer super- bations of hyperbolic actions. But the last ones are
rigidity. These results allow one to produce a already known to be conjugate to the hyperbolic
measurable conjugacy for the perturbation. Then action in the base.
one shows that the conjugacy has higher regular- Recent important progress in the question of
ity using the presence of hyperbolicity. Having local rigidity of lattice actions was made by
enough directions to span the whole tangent Fisher, Margulis (2003, 2004, 2005). Their proofs
space is essential to lift the regularity. A cocycle are modeled along the proof of Weil’s local rigid-
to which superrigidity can be applied is the deriv- ity result (Weil 1964) and use an analog of
ative cocycle. Hamilton’s (1982) hard implicit function theorem.
The study of local rigidity of partially hyper- Let G be a connected semi-simple Lie group with
bolic actions that contain a compact factor all simple factors of rank at least two, and Γ  G a
was initiated by Niţică, Török (1995, 2001). Let lattice. The main result shows that a volume pre-
n  3 and d  1. Let r be the action serving affine action r of G or Γ on a compact
of SLðn, ℤÞ on nþd ¼ n d given by smooth manifold X is C1,1,1,1 locally rigid.
rðAÞðx, yÞ ¼ ðAx, yÞ, x  n , y  d , A  SLðn, ℤÞ: Then, Lower regularity results are also available.
for K  1, (Niţică and Török 1995) shows that r is A component of the proof shows that if Γ is a
C1,1,5,K1 deformation rigid. The proof is based group with property (T), X a compact smooth
on three results in hyperbolic dynamics: the gen- manifold, and r a smooth action of Γ on X by
eralization of Livshits’s cohomological results to Riemannian isometries, then r is C1,1,1,1
cocycles with values in diffeomorphism groups, locally rigid. An earlier local rigidity result for
the extension of Livshits’s result to general this type of actions by cocompact lattices was
Anosov actions, and a version of the Hirsch, obtained by Benveniste (2000).
Pugh, Shub structural stability theorem improving Many lattices act naturally on “boundaries” of
the regularity of the conjugacy. type G/P, where G is a semi-simple algebraic Lie
Assume now n  3 and K  1. If r is the group and P is a parabolic subgroup. An example
action of SLðn, ℤÞ on nþ1 ¼ n  given by is given by G ¼ SL(2, ℝ) and P the subgroup in
rðAÞðx, yÞ ¼ ðAx, yÞ, x  n , y  , A  SLðn, ℤÞ, G consisting of upper triangular matrices. Local
Ergodic Theory: Rigidity 625

rigidity results for this type of actions were found Global Rigidity
by Ghys (1985), Kanai (1996) and Katok,
Spatzier (1997). The first remarkable result in global rigidity
Starting with the work of Katok and Lewis, a belongs to Mostow (1973). For G a connected
related direction was the study of local rigidity for non-compact semi-simple Lie group not locally
higher rank abelian actions. They prove in (Katok isomorphic to SL(2, ℝ), and two irreducible
and Lewis 1991) the C1,1,1,1 local rigidity of the cocompact lattices Γ1, Γ2  G, Mostow showed
action of a ℤn maximal diagonalizable (over ℝ) that any isomorphism θ from Γ1 into Γ2 extends to
subgroup of SL(n þ 1, ℤ), n  2, acting on the an isomorphism of G into itself. G has an involu-
torus nþ1 : These type of results were later pushed tion s whose fixed set is a maximal compact
forward by Katok, Spatzier (1997). Using the subgroup K. One constructs the symmetric Rie-
theory of nonstationary normal forms developed mannian space X ¼ G/K. To each chamber of
in (Guysinsky 2002; Guysinsky and Katok 1998) X corresponds a parabolic group and these para-
by Katok, Guysinsky, they proved several local bolic groups are endowed with a Tits geometry
rigidity results. The first one assumes that G is a similar to the projective geometry of lines, planes
semi simple Lie group with all simple factors of etc. formed in the classical case when G ¼
rank atleast two, Γ a lattice in G, N a nilpotent Lie PGL(n, ℝ). The proof of Mostow’s result starts
group and Λ a lattice in N. Then any Anosov by building a θ-equivariant pseudo-isometric map
affine action of Γ on N/Λ is C1,1,1,1 locally f : G/K1 ! G/K2. The map f induces an inci-
rigid. Second, let ℤd be a group of affine trans- dence preserving θ-equivariant isomorphism
formations of N/Λ for which the derivatives are f0 of the Tits geometries. By Tits’ generalized
simultaneously diagonalizable over ℝ with no fundamental theorem of projective geometry,
eigenvalues on the unit circle. Then the ℤd-action f0 is induced by an isomorphism of G. Finally,
on N/Λ is C1,1,1,1 locally rigid. A related result yðgÞ ¼ f0  g  f1
0 gives the desired conclusion.
for continuous groups is the C1,1,1,1 local rigid- The next remarkable result in global rigidity is
ity (after factorization by the orbit foliation) of Margulis’ superrigidity theorem. An account of
the action of a maximal abelian ℝ-split subgroup this development can be found in the monograph
in an ℝ-split semi-simple Lie group of real rank (Margulis 1991). For large classes of irreducible
at least two on G/Λ, where Λ is a cocompact lattices in semi-simple Lie groups, this result clas-
lattice in G. sifies all finite dimensional representations. Let
One can also study rigidity of higher rank abe- G be a semi-simple simply connected Lie group
lian partially hyperbolic actions that are not of rank higher than two and Γ < G an irreducible
hyperbolic. Natural examples appear as automor- lattice. Then any linear representation π of Γ is
phisms of tori and as variants of Weyl chamber almost the restriction of a linear representation of
flows. For the case of ergodic actions by automor- G. That is, there exists a linear representation π1 of
phisms of a torus, this was investigated using a G and a bounded image representation π2 of Γ
version of the KAM (Kolmogorov, Arnold, such that π ¼ π1π2. The possible representations
Moser) method by Damianović, Katok (2007). As π2 are also classified by Margulis up to some facts
usual in the KAM method, one starts with a line- concerning finite image representations. As in the
arization of the conjugacy equation. At each step of case of Mostow’s result, the proof involved the
the iterative KAM scheme, some twisted cohomo- analysis of a map defined on the boundary at
logical equations are solved. The existence of the infinity. In this case the map is studied using
solutions is forced by the ergodicity of the action deep results from dynamics like the multi-
and the higher rank assumptions. Diophantine con- plicativity ergodic theorem of Oseledec (1968)
ditions present in this case allow to control the or the theory of random walks on groups devel-
fixed loss of regularity which is necessary for the oped by Furstenberg (1963). An important conse-
convergence of these solutions to a conjugacy. quence of Margulis superrigidity result is the
626 Ergodic Theory: Rigidity

arithmeticity of irreducible lattices in connected A topological rigidity theorem has been proved
semi-simple Lie groups of rank higher than two. by Farrell, Jones (1989). They showed that if N is
A basic example of arithmetic lattice can be a complete connected Riemannian manifold
obtained by taking the integer points in a semi- whose sectional curvature lies in a closed interval
simple Lie group that is a matrix group, like included in(1, 0], and M is a topological man-
taking SL(n, ℤ) inside SL(n, ℝ). Special cases of ifold of dimension greater than 5, then any proper
superrigidity theorems were proved by Corlette homotopy equivalence f : M ! N is properly
(1992) and Gromov, Schoen (1992) for the rank homotopic to a homeomorphism. In particular, if
one groups Sp(1, n) and respectively F4 using the M and N are both compact connected negatively
theory of harmonic maps. A consequence is the curved Riemannian manifolds with isomorphic
arithmeticity of lattices in these groups. Some of fundamental groups, then M and N are
these results are put into differential geometric homeomorphic.
setting in (Mok et al. 1993). Likewise to the case of local rigidity, a source
Margulis supper rigidity result was extended to of inspiration for results in global rigidity was the
cocycles by Zimmer. A detailed exposition, theory of hyperbolic systems, in particular their
including a self contained presentation of several classification. The only known examples of
rigidity results of Margulis, can be found in the Anosov diffeomorphisms are hyperbolic auto-
monograph (Zimmer 1984). We mention here a morphisms of infranilmanifolds. Moreover, any
version of this result that can be found in (Fisher Anosov diffeomorphism on an infranilmanifold
and Margulis 2003). Let M be a compact mani- is topologically conjugate to a hyperbolic auto-
fold, Ha matrix group, P ¼ M H, and Γ a lattice morphism (Franks 1970; Manning 1974). It is
in a simply connected, semi-simple Lie group conjectured that any Anosov diffeomorphism is
with all factors of rank higher that two. Assume topologically conjugate to a hyperbolic automor-
that Γ acts on M and H in a way that makes the phism of an infranilmanifold. Partial results are
projection from P to M equivariant. Moreover, the obtained in (Newhouse 1970), where the conjec-
action of Γ on P is measure preserving and ture is proved for Anosov diffeomorphisms with
ergodic. Then there exists a measurable map codimension one stable/unstable foliation. The
s : M ! H, a representation π : G ! H, a compact proof of the general conjecture eluded efforts
subgroup K < H which commute with π(G) and a done so far. It is not even known if any Anosov
measurable map k : Γ M ! K such that diffeomorphism is topologically transitive, that is,
γ  s(m) ¼ k(γ, m)  π(γ)  s(γ  m). One can easily if it has a dense orbit. A few positive results are
check from the last equation that k is a cocycle. available. Let M be a C1 compact manifold endo-
So, up to a measurable change of coordinates wed with a C1 affine connection. Let f be a
given by the map s, the action of Γ on P is a topologically transitive Anosov diffeomorphism
compact extension via a cocycle of a linear repre- preserving the connection and such that the stable
sentation of G. and unstable distributions are C1. Then Benoist,
Developing further the method of Mostow for Labourie (1993) proved that f is C1 conjugate to a
studying the Tits building associated to a symmet- hyperbolic automorphism of an infranilmanifold.
ric space of non-positive curvature led Ballman, The situation for Anosov flows is somehow
Brin, Eberlein, Spatzier (1985a, b) to a number of different. As shown in (Franks and Williams
characterizations of symmetric spaces. In particu- 1980), there exist Anosov flows that are not topo-
lar, they showed that if M is a complete Riemann- logically transitive, so a general analog of the
ian manifold of non-positive curvature, finite conjecture is false. Nevertheless, for the case of
volume, with simply connected cover, irreducible codimension one stable or unstable foliation, it is
and of rank at least two, then M is isometric to a conjectured in (Verjovsky 1974) that any Anosov
symmetric space with the connected component flow on a manifold of dimension greater than
of Isom(M) having no compact factors. three admits a global cross-section. This would
Ergodic Theory: Rigidity 627

imply that the flow is topologically conjugate to proportional to a rational form. The proof of the
the suspension of a linear automorphism of a conjecture is based on the study of the orbits for
torus. unipotent flows acting by translations on the
For actions of groups larger than ℤ, or ℝ, homogenous space SL(n, ℤ)\SL(n, ℝ). All these
global classification results are more abundant. results were special cases of the Raghunathan
A useful strategy in these type of results, which conjecture about the structure of the orbits of the
are quite technical, is to start by obtaining a mea- actions of unipotent flows on homogenous spaces.
surable description of the action, most of the time Raghunathan’s conjecture was proved in full
using Margulis–Zimmer superrigidity results, and generality by Ratner (1991a, b). Borel, Prasad
then use extra assumptions on the action, such as (1992) raised the question of an analog of
the presence of a hyperbolic element, or the pres- Raghunathan’s conjecture for S-algebraic groups.
ence of an invariant geometric structure, or both, S-algebraic groups are products of real and p-adic
in order to show that the measurable model is algebraic groups. This was answered indepen-
actually continuous or even differentiable. For dently by Margulis, Tomanov (1994) and
actions of higher rank Lie groups and their lattices Ratner (1995).
some representative papers are by Katok, Lewis, A basic example of higher rank abelian hyper-
Zimmer (1996) and Goetze, Spatzier (1999). For bolic action is given by the action of Sm,n, the
actions of higher rank abelian groups see Kalinin, multiplicative semigroup of endomorphisms gen-
Spatzier (2007). erated by the multiplication by m and n, two
nontrivial integers, on the one dimensional torus
1 : Ina pioneering paper (Furstenberg 1967)
Measure Rigidity Furstenberg showed that for m, n that are not
powers of the same integer the action of Sm,n has
Measure rigidity is the study of invariants mea- a unique closed, infinite invariant set, namely 1
sures for actions of one parameter and multi itself. Since there are many closed, infinite invari-
parameter abelian groups and semigroups acting ant sets for multiplication by m, and by n, this
on manifolds. Typical situations when interesting result shows a remarkable rigidity property of the
rigidity phenomena appear are for one parameter joint action. Furstenberg’s result was generalized
unipotent actions and higher rank hyperbolic later by Berend for other group actions on higher
actions, discrete or continuous. dimensional tori and on other compact abelian
A unipotent matrix is one all of whose eigen- groups in (Berend 1983, 1984).
values are one. An important case where the Furstenberg further opened the field by raising
action of a unipotent flow appears is that of the the following question:
horocycle flow. The invariant measures for it were
studied by Furstenberg (1973), who showed that Conjecture 1 Let m be a Sm,n-invariant and ergo-
the horocycle flow on a compact surface is dic probability measure on 1 : Then m is either an
uniquely ergodic, that is, the ergodic measure is atomic measure supported on a finite union of
unique. Dani and Smillie (1984) extended this (rational) periodic orbits or m is the Lebesgue
result to the case of non-compact surfaces, with measure.
the only other ergodic measures appearing being
those supported on compact horocycles. An While the statement appears to be simple,
important breakthrough is the work of Margulis proving it has been elusive. The first partial result
(1989), who solved a long standing question in was given by Lyons (1988) under the strong addi-
number theory, Oppenheim’s conjecture. The tional assumption that the measure makes one of
conjecture is about the density properties of the the endomorphisms generating the action exact.
values of an indefinite quadratic form in three or Later Rudolph (1990) and Johnson (1992)
more variables, provided the form is not weaken the exactness assumption and proved
628 Ergodic Theory: Rigidity

that m must be the Lebesgue measure provided es 0 0

that multiplication by m (or multiplication by n) H¼ 0 e t
0 : s, t  ℝ
has positive entropy with respect to m. Their st
0 0 e
results have been proved again using slightly dif-
ferent methods by Feldman (1993). A further
acts naturally on X by right translations. It was
extension was given by Host (1995).
conjectured by Margulis (1997) that any compact
Katok proposed another example for which
H-invariant subset of X is a union of compact
measure rigidity can be fruitfully tested, the ℤ2-
H-orbits. A positive solution to this conjecture
action induced by two commuting hyperbolic
implies a long standing conjecture of Littlewood:
automorphisms on the torus 3 : An example of
such action is shown in section “Basic Definitions
Conjecture 2 Let kxk denote the distance
and Examples”. One can also consider the action
between the real number x and the closest integer.
induced by a ℤn1 maximal abelian group of
Then
hyperbolic automorphisms acting on the torus
n : Katok, Spatzier (1996) developed a more
lim inf n k na kk nb k ¼ 0 ð4Þ
geometric technique allowing to prove measure n!1
rigidity for these actions if they have an element
acting with positive entropy. Their technique can for any real numbers α and β.
be applied in higher generality if the action is
irreducible in a strong sense and, in addition, it A partial result was obtained by Einsiedler,
has individual ergodic elements or it is TNS. See Katok, Lindenstrauss (2006) who proved that the
also (Kalinin and Katok 2002). This method is set of pairs (α, β)  ℝ2 for which (4) is not
based on the study of conditional measures satisfied has Hausdorff dimension zero. Applica-
induced by the invariant measure on certain tions of these techniques to questions in quantum
invariant foliations that appear naturally in the ergodicity were found by Lindenstrauss (2006).
presence of a hyperbolic action. Besides stable A current direction in measure rigidity is
and unstable foliations, one can also consider attempting to classify the invariant measures
various intersections of them. Einsiedler, under rather weak assumptions about the higher
Lindenstrauss (2003) were able to eliminate the rank abelian action, like the homotopical data for
ergodicity and TNS assumptions. the action. Kalinin and Katok (2007) proved that
Yet another interesting example of higher rank any ℤk, k  2, smooth action α on a k þ 1-
abelian action is given by Weyl chamber flows. dimensional torus whose elements are homotopic
These do not satisfy the TNS condition. to the corresponding elements of an action by
Einsiedler, Katok (2003) proved that if G is hyperbolic automorphisms preserves an abso-
SL(n, ℝ), Γ  G is a lattice, H is the subgroup of lutely continuous measure.
positive diagonal matrices in G, and m a
H-invariant and ergodic measure on G/Γ such
that the entropy of m with respect to all one param- Future Directions
eter subgroups of H is positive, then m is the
G invariant measure on G/Γ. An important open problem in differential rigidity
These results are useful in the investigation of is to find invariants for the Ck conjugacy of the
several deep questions in number theory. Define perturbations of higher dimensional hyperbolic
X ¼ SL(3, ℤ)\SL(3, ℝ) the diagonal subgroup of systems. For Anosov diffeomorphisms, de la
matrices with positive entries in SL(3, ℝ). This Llave counter example (de la Llave 1992) shows
space is not compact but is endowed with a unique that this extension is not possible for a four dimen-
translation invariant probability measure. The sional example that appears as a direct product
diagonal subgroup of two dimensional Anosov diffeomorphisms.
Ergodic Theory: Rigidity 629

Indeed, there are C1 perturbations of the product Bibliography

that are only Ck conjugate to the unperturbed
system for any k  1. In the positive direction, Primary Literature
Katok conjectured that generalizations are possi- Anosov DV (1967) Geodesic flows on closed Riemannian
manifolds with negative curvature. Proc Stek Inst 90:
ble for the diffeomorphism induced by an irreduc-
1–235
ible hyperbolic automorphism of a torus. One can Ballman W, Brin M, Eberlein P (1985a) Structure of mani-
also investigate this question for Anosov flows. folds of non-negative curvature. I. Ann Math 122:
Examples of higher rank Lie groups can be 171–203
Ballman W, Brin M, Spatzier R (1985b) Structure of mani-
obtained by taking products of rank one Lie
folds ofnon-negative curvature. II. Ann Math 122:
groups. Many actions of irreducible lattices in 205–235
these type of groups are believed to be locally Benoist Y, Labourie F (1993) Flots d’Anosov á
rigid, but the techniques available so far cannot distribuitions stable et instable différentiables. Invent
Math 111:285–308
be applied. A related problem is to study local
Benveniste EJ (2000) Rigidity of isometric lattice actions
rigidity in low regularity classes, for example the on compact Riemannian manifolds. Geom Funct Anal
local rigidity of homomorphisms from higher 10:516–542
rank lattices into homeomorphism groups. More Berend D (1983) Multi-invariant sets on tori. Trans AMS
280:509–532
problems related to local rigidity can be found in
Berend D (1984) Multi-invariant sets on compact abelian
(Fisher 2006). groups. Trans AMS 286:505–535
An important problem in global rigidity, Borel A, Prasad G (1992) Values of isotropic
emphasized by Katok and Spatzier, is to show quadratic forms at S-integral points. Compos Math 83:
347–372
that, up to differentiable conjugacy, any higher
Brin MI, Pesin YA (1974) Partially hyperbolic dynamical
rank Anosov or partially hyperbolic action is alge- systems. Izvestia 38:170–212
braic under the assumption that it is sufficiently Burns K, Pugh C, Shub M, Wilkinson A (2001) Recent
irreducible. The irreducibility assumption is results about stable ergodicity. In: Katok A, Pesin Y, de
la Llave R, Weiss H (eds) Smooth ergodic theory and its
needed in order to exclude actions obtained by
applications, Seattle, 1999. Proc Symp Pure Math,
successive application of products, extensions, vol 69. AMS, Providence, pp 327–366
restrictions and time changes from basic ingredi- Calabi E (1961) On compact Riemannian manifolds with
ents which include some rank one actions. constant curvature. I. Proc Symp Pure Math,
vol 3. AMS, Providence, pp 155–180
Another problem of current research in mea-
Calabi E, Vesentini E (1960) On compact, locally symmet-
sure rigidity is to develop a counterpart of ric Kähler manifolds. Ann Math 17:472–507
Ratner’s theory for the case of actions by hyper- Corlette K (1992) Archimedian superrigidity and hyper-
bolic higher rank abelian groups on homogenous bolic geometry. Ann Math 135:165–182
Damianović D, Katok A (2005) Periodic cycle
spaces. It was conjectured by Katok, Spatzier
functionals and cocycle rigidity for certain partially
(1996) that the invariant measures for such actions hyperbolic actions. Discrete Contin Dynam Syst 13:
given by toral automorphisms or Weyl chamber 985–1005
flows are essentially algebraic, that is, supported Damianović D, Katok A (2007) Local rigidity of partially
hyperbolic actions KAM, I. Method and Zk-actions on
on closed orbits of connected subgroups.
the torus. Preprint available at http://www.math.psu.
Margulis in (Margulis 2000) extended this con- edu/katok_a
jecture to a rather general setup addressing both Dani SG, Margulis GA (1990) Values of quadratic forms at
the topological and measurable aspects of the integer points: an elementary approach. Enseign Math
36:143–174
problem. More details about actions on homoge-
Dani SG, Smillie J (1984) Uniform distributions of
nous spaces, as well as connections to diophantine horocycle orbits for Fuchsian groups. Duke Math
approximation, can be found in the article ▶ Ergo- J 51:185–194
dic Theory on Homogeneous Spaces and Metric de la Llave R (1987) Invariants for smooth conjugacy of
hyperbolic dynamical systems. I. Commun Math Phys
Number Theory by Kleinbock.
109:369–378
de la Llave R (1992) Smooth conjugacy and S-R-B mea-
Acknowledgment This research was supported in part by sures for uniformly and non-uniformly hyperbolic
NSF Grant DMS-0500832. dynamical systems. Commun Math Phys 150:289–320
630 Ergodic Theory: Rigidity

de la Llave R (1997) Analytic regularity of solutions of Ghys E (1985) Actions localement libres du groupe affine.
Livshits’s cohomology equation and some applications Invent Math 82:479–526
to analytic conjugacy of hyperbolic dynamical systems. Goetze E, Spatzier R (1999) Smooth classification of
Ergodic Theory Dynam Syst 17:649–662 Cartan actions of higher rank semi-simple Lie groups
de la Llave R, Moriyon R (1988) Invariants for smooth and their lattices. Ann Math 150:743–773
conjugacy of hyperbolic dynamical systems. Gromov M, Schoen R (1992) Harmonic maps into singular
IV. Commun Math Phys 116:185–192 spaces and p-adic superrigidity for lattices in groups of
de la Llave R, Marco JM, Moriyon R (1986) Canonical rank one. Publ Math IHES 76:165–246
perturbation theory of Anosov systems and regularity Guysinsky M (2002) The theory of nonstationary normal
results for the Livšic cohomology equation. Ann Math forms. Ergodic Theory Dynam Syst 22:845–862
123:537–611 Guysinsky M, Katok A (1998) Normal forms and
Einsiedler M, Katok A (2003) Invariant measures on G/Γ invariant geometric structures for dynamical systems
for split simple Lie groups. Commun Pure Appl Math with invariant contracting foliations. Math Res Lett 5:
56:1184–1221 149–163
Einsiedler M, Lindenstrauss E (2003) Rigidity properties Hamilton R (1982) The inverse function theorem of Nash
of Zd-actions on tori and solenoids. Electron Res and Moser. Bull AMS 7:65–222
Announc AMS 9:99–110 Helgason S (1978) Differential geometry, Lie groups and
Einsiedler M, Katok A, Lindenstrauss E (2006) Invariant symmetric spaces. Academic, New York
measures and the set of exceptions to Littlewood con- Hirsch M, Pugh C, Shub M (1977) Invariant manifolds.
jecture. Ann Math 164:513–560 Lecture notes in mathematics, vol 583. Springer, Berlin
Farrell FT, Jones LE (1989) A topological analog of Host B (1995) Nombres normaux, entropie, translations.
Mostow’s rigidity theorem. J AMS 2:237–370 Israel J Math 91:419–428
Feldman J (1993) A generalization of a result of R Lyons Hurder S (1992) Rigidity of Anosov actions of higher rank
about measures on [0,1]. Israel J Math 81:281–287 lattices. Ann Math 135:361–410
Fisher D (2006) Local rigidity of group actions: past, Hurder S, Katok A (1990) Differentiability, rigidity and
present, future. In: Dynamics, ergodic theory and Godbillon-Veyclasses for Anosov flows. Publ Math
geometry (2007). Cambridge University Press IHES 72:5–61
Fisher D, Margulis GA (2003) Local rigidity for cocycles. Johnson AS (1992) Measures on the circle invariant under
In: Surv Diff Geom VIII. International Press, Cam- multiplication bya nonlacunary subsemigroup of inte-
bridge, pp 191–234 gers. Israel J Math 77:211–240
Fisher D, Margulis GA (2004) Local rigidity of affine Journé JL (1988) A regularity lemma for functions of
actions of higher rank Lie groups and their several variables. Rev Mat Iberoam 4:187–193
lattices. 2003 Kalinin B, Katok A (2002) Measurable rigidity and
Fisher D, Margulis GA (2005) Almost isometric actions, disjointness for Zk-actions by toral automorphisms.
property T, and local rigidity. Invent Math 162:19–80 Ergodic Theory Dynam Syst 22:507–523
Flaminio L, Forni G (2003) Invariant distributions and time Kalinin B, Katok A (2007) Measure rigidity beyond uni-
averages for horocycle flows. Duke Math J 119: form hyperbolicity: invariant measures for Cartan
465–526 actions on tori. J Modern Dyn 1:123–146
Forni G (1997) Solutions of the cohomological equation Kalinin B, Spatzier R (2007) On the classification of Cartan
for area-preserving flows on compact surfaces of higher actions. Geom Funct Anal 17:468–490
genus. Ann Math 146:295–344 Kanai M (1996) A new approach to the rigidity of discrete
Franks J (1970) Anosov diffeomorphisms. In: Chern SS, group actions. Geom Funct Anal 6:943–1056
Smale S (eds) Global analysis (Proc Symp Pure Math, Katok A, Hasselblatt B (1995) Introduction to the modern
XIV, Berkeley 1968). AMS, Providence, pp 61–93 theory of dynamical systems. Encyclopedia of mathe-
Franks J, Williams R (1980) Anomalous anosov flows. In: matics and its applications, vol 54. Cambridge Univer-
Global theory of dynamical systems, Proc Inter Conf sity Press, Cambridge
Evanston, 1979. Lecture notes in mathematics, vol 819. Katok A, Katok S (1995) Higher cohomology for abelian
Springer, Berlin, pp 158–174 groups of toral automorphisms. Ergodic Theory
Furstenberg H (1963) A Poisson formula for semi-simple Dynam Syst 15:569–592
Lie groups. Ann Math 77:335–386 Katok A, Katok S (2005) Higher cohomology for abelian
Furstenberg H (1967) Disjointness in ergodic theory, min- groups of toral automorphisms. II. The partially hyper-
imal sets, anda problem in Diophantine approximation. bolic case, and corrigendum. Ergodic Theory Dynam
Math Syst Theory 1:1–49 Syst 25:1909–1917
Furstenberg H (1973) The unique ergodicity of the Katok A, Kononenko A (1996) Cocycles’ stability
horocycle flow. In: Recent advances in topological for partially hyperbolic systems. Math Res Lett 3:
dynamics, Proc Conf Yale Univ, New Haven 1972. 191–210
Lecture notes in mathematics, vol 318. Springer, Ber- Katok A, Lewis J (1991) Local rigidity for certain groups
lin, pp 95–115 of toral automorphisms. Israel J Math 75:203–241
Ergodic Theory: Rigidity 631

Katok A, Lewis J (1996) Global rigidity results for lattice Margulis GA (1997) Oppenheim conjecture. In: Fields
actions on toriand new examples of vol preserving medalists lectures, vol 5.World Sci Ser 20th Century
actions. Israel J Math 93:253–280 Math. World Sci Publ, River Edge, pp 272–327
Katok A, Niţică V (2007) Rigidity of higher rank abelian Margulis GA (2000) Problems and conjectures in rigidity
cocycles with values in diffeomorphism groups. Geom theory. In: Mathematics: frontiers and perspectives.
Dedicata 124:109–131 AMS, Providence, pp 161–174
Katok A, Niţică V. Differentiable rigidity of abelian group Margulis GA, Qian N (2001) Local rigidity of weakly
actions. Cambridge University Press (to appear) hyperbolic actions of higher rank real Lie groups and
Katok A, Spatzier R (1994a) First cohomology of Anosov their lattices. Ergodic Theory Dynam Syst 21:
actions of higher rank abelian groups and applications 121–164
to rigidity. Publ Math IHES 79:131–156 Margulis GA, Tomanov G (1994) Invariant measures for
Katok A, Spatzier R (1994b) Subelliptic estimates of poly- actions of unipotent groups over local fields of homog-
nomial differential operators and applications to rigid- enous spaces. Invent Math 116:347–392
ity of abelian actions. Math Res Lett 1:193–202 Mieczkowski D (2007) The first cohomology of parabolic
Katok A, Spatzier R (1996) Invariant measures for higher- actions for some higher-rank abelian groups and repre-
rank abelian actions. Ergodic Theory Dynam Syst 16: sentation theory. J Modern Dyn 1:61–92
751–778; Katok A, Spatzier R (1998) Corrections to: Milnor J (1971) Introduction to algebraic K-theory.
invariant measures for higher-rank abelian actions.; Princeton University Press, Princeton
(1996) Ergodic Theory Dynam Syst 16:751–778; Ergo- Mok N, Siu YT, Yeung SK (1993) Geometric superrigidity.
dic Theory Dynam Syst 18:503–507 Invent Math 113:57–83
Katok A, Spatzier R (1997) Differential rigidity of Mostow GD (1973) Strong rigidity of locally symmetric
Anosov actions of higher rank abelian groups and spaces. Ann Math studies, vol 78. Princeton University
algebraic lattice actions. Trudy Mat Inst Stek 216: Press, Princeton
292–319 Newhouse SE (1970) On codimension one Anosov
Katok A, Lewis J, Zimmer R (1996) Cocycle superrigidity diffeomorphisms. Am J Math 92:761–770
and rigidity for lattice actions on tori. Topology 35:27–38 Niţică V, Török A (1995) Cohomology of dynamical sys-
Katok A, Niţică V, Török A (2000) Non-abelian cohomol- tems and rigidity of partially hyperbolic actions of
ogy of abelian Anosov actions. Ergodic Theory Dynam higher rank lattices. Duke Math J 79:751–810
Syst 2:259–288 Niţică V, Török A (1998) Regularity of the transfer map for
Katok A, Katok S, Schmidt K (2002) Rigidity of measur- cohomologous cocycles. Ergodic Theory Dynam Syst
able structure for Zd-actions by automorphisms of a 18:1187–1209
torus. Comment Math Helv 77:718–745 Niţică V, Török A (2001) Local rigidity of certain partially
Kazhdan DA (1967) On the connection of a dual space of a hyperbolic actions of product type. Ergodic Theory
group with the structure of its closed subgroups. Funkc Dynam Syst 21:1213–1237
Anal Prilozen 1:71–74 Niţică V, Török A (2002) On the cohomology of Anosov
Lewis J (1991) Infinitezimal rigidity for the action of SLn(- actions. In: Rigidity in dynamics and geometry, Cam-
Z) on Tn. Trans AMS 324:421–445 bridge, 2000. Springer, Berlin, pp 345–361
Lindenstrauss E (2005) Rigidity of multiparameter actions. Niţică V, Török A (2003) Cocycles over abelian TNS
Israel Math J 149:199–226 actions. Geom Dedicata 102:65–90
Lindenstrauss E (2006) Invariant measures and arithmetic Oseledec VI (1968) A multiplicative ergodic theorem.
quantum unique ergodicity. Ann Math 163:165–219 Characteristic Lyapunov, exponents of dynamical sys-
Livshits A (1971) Homology properties of Y-systems. tems. Trudy Mosk Mat Obsc 19:179–210
Math Zametki 10:758–763 Parry W (1999) The Livšic periodic point theorem for non-
Livshits A (1972) Cohomology of dynamical systems. abelian cocycles. Ergodic Theory Dynam Syst 19:
Izvestia 6:1278–1301 the Livšic cohomology equation. 687–701
Ann Math 123:537–611 Pugh C, Shub M (1972) Ergodicity of Anosov actions.
Lyons R (1988) On measures simultaneously 2- and Invent Math 15:1–23
3-invariant. Israel J Math 61:219–224 Ratner M (1991a) On Ragunathan’s measure conjecture.
Manning A (1974) There are no new Anosov Ann Math 134:545–607
diffeomorphisms on tori. Am J Math 96:422–429 Ratner M (1991b) Ragunathan’s topological conjecture
Marco JM, Moriyon R (1987a) Invariants for smooth and distributions of unipotent flows. Duke Math J 63:
conjugacy of hyperbolic dynamical systems. 235–280
I. Commun Math Phys 109:681–689 Ratner M (1995) Raghunathan’s conjecture for Cartesians
Marco JM, Moriyon R (1987b) Invariants for smooth products of real and p-adic Lie groups. Duke Math J 77:
conjugacy of hyperbolic dynamical systems. III. 275–382
Commun Math Phys 112:317–333 Rudolph D (1990) 2 and 3 invariant measures
Margulis GA (1989) Discrete subgroups and ergodic the- andentropy. Ergodic Theory Dynam Syst 10:395–406
ory. In: Number theory, trace formulas and discrete Schmidt K (1999) Remarks on Livšic’ theory for nonabelian
groups, Oslo, 1987. Academic, Boston, pp 277–298 cocycles. Ergodic Theory Dynam Syst 19:703–721
Margulis GA (1991) Discrete subgroups of semi-simple Selberg A (1960) On discontinuous groups in higher-
Lie groups. Springer, Berlin dimensional symmetric spaces. In: Contributions to
632 Ergodic Theory: Rigidity

function theory. Inter colloq function theory, Bombay. Books and Reviews
Tata Institute of Fundamental Research, pp 147–164 de la Harpe P, Valette A (1989) La propriété (T) de
Smale S (1967) Differentiable dynamical systems. Bull Kazhdan pour les groupes localement compacts.
AMS 73:747–817 Astérisque 175
Spatzier R (1995) Harmonic analysis in rigidity theory. In: Feres R (1998) Dynamical systems and semi-simple
Ergodic theory and its connections with harmonic analy- groups: an introduction. Cambridge tracts in
sis. Alexandria, 1993. London Math Soc Lect Notes Ser, mathematics, vol 126. Cambridge University Press,
vol 205. Cambridge University Press, Cambridge, Cambridge
pp 153–205 Feres R, Katok A (2002) Ergodic theory and dynamics of
Veech WA (1986) Periodic points and invariant pseudo- G-spaces. In: Handbook in dynamical systems, vol 1A.
measures for toral endomorphisms. Ergodic Theory Elsevier, Amsterdam, pp 665–763
Dynam Syst 6:449–473 Gromov M (1988) Rigid transformation groups. In:
Verjovsky A (1974) Codimension one Anosov flows. Bull Bernard D, Choquet-Bruhat Y (eds) Géométrie
Soc Math Mex 19:49–77 Différentielle (Paris, 1986). Hermann, Paris,
Weil A (1960) On discrete subgroups of Lie groups. I. Ann pp 65–139. Travaux en Cours 33
Math 72:369–384 Kleinbock D, Shah N, Starkov A (2002) Dynamics of
Weil A (1962) On discrete subgroups of Lie groups. II. Ann subgroup actions on homogeneous spaces of Lie
Math 75:578–602 groups and applications to number theory. In: Hand-
Weil A (1964) Remarks on the cohomology of groups. Ann book in dynamical systems, vol 1A. Elsevier, Amster-
Math 80:149–157 dam, pp 813–930
Zimmer R (1984) Ergodic theory and semi-simple groups. Knapp A (2002) Lie groups beyond an introduction,
Birhhäuser, Boston 2nd edn. Progress in mathematics, 140. Birkhäuser,
Zimmer R (1987) Actions of semi-simple groups and dis- Boston
crete subgroups. In: Proc Inter Congress of Math Raghunathan MS (1972) Discrete subgroups of Lie groups.
(1986). AMS, Providence, pp 1247–1258 Springer, Berlin
Zimmer R (1990) Infinitesimal rigidity of smooth actions of Witte MD (2005) Ratner’s theorems on unipotent flows.
discrete subgroups of Lie groups. J Differ Geom 31: Chicago lectures in mathematics. University of
301–322 Chicago Press, Chicago
 Hyperbolicity A measure is hyperbolic in the
Chaos and Ergodic Theory sense of Pesin if at almost every point no
Lyapunov exponent is zero. See ▶ “Smooth
Jérôme Buzzi Ergodic Theory”.
C.N.R.S. and Université Paris-Sud, Orsay, France Kolmogorov typicality A property is typical in
the sense of Kolmogorov for a topological
space F of parametrized families f ¼ ( ft)t  U,
Article Outline U being an open subset of ℝd for some d  1, if
it holds for ft for Lebesgue almost every t and
Glossary topologically generic f  F .
Definition of the Subject Lyapunov exponents The Lyapunov exponents
Introduction (▶ Smooth Ergodic Theory) are the limits,
Picking an Invariant Probability Measure when they exist, lim 1 log ðT n Þ0 ðxÞ : v
Tractable Chaotic Dynamics n!1 n

Statistical Properties where x  M and v is a non zero tangent vector

Orbit Complexity to M at x. The Lyapunov exponents of an
Stability ergodic measure is the set of Lyapunov expo-
Untreated Topics nents obtained at almost every point with
Future Directions respect to that measure for all non-zero tangent
Bibliography vectors.
Markov shift (topological, countable state) It
Glossary is the set of all infinite or bi-infinite paths on
some countable directed graph endowed with
For simplicity, definitions are given for a contin- the left-shift, which just translates these
uous or smooth self-map or diffeomorphism T of a sequences.
compact manifold M. Maximum entropy measure It is a measure m
Entropy, measure-theoretic (or: metric which maximizes the measured entropy and,
entropy) For an ergodic invariant probability by the variational principle, realized the topo-
measure m, it is the smallest exponential growth logical entropy.
rates of the number of orbit segments of given Physical measure It is a measure m whose basin,
length, with respect to that length, after restric- fx  M : 8f : M ! ℝ continuous lim n!1 1n
n
1
tion to a set of positive measure. We denote it by k¼0 ff k x ¼ f dmg has nonzero volume.
h(T, m). See ▶ “Entropy in Ergodic Theory” and Prevalence A property is prevalent in some
section “Local Complexity” below. complete metric, separable vector space X if it
Entropy, topological It is the exponential growth holds outside of a set N such that, for some
rates of the number of orbit segments of given Borel probability measure m on X: m(A þ v) ¼ 0
length, with respect to that length. We denote it for all v  X. See (Christensen 1972; Hunt
by htop( f ). See ▶ “Entropy in Ergodic Theory” et al. 1992; Tsujii 1992).
and section “Local Complexity” below. Sensitivity on initial conditions The sensitivity
Ergodicity A measure is ergodic with respect to to initial conditions on X 0  X if there exists a
a map T if given any measurable subset constant r > 0 such that for every x  X 0, there
S which is invariant, i.e., such that T
1S ¼ S, exists y  X, arbitrarily close to x, and n  0
either S or its complement has zero measure. such that d(T ny, T nx) > r.

© Springer-Verlag 2009 633

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_64
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_64
634 Chaos and Ergodic Theory

Sinai–Ruelle–Bowen measures It is an invari- chaotic dynamics can be difficult to interpret and

ant probability measure which is absolutely to plan, even misleading, and the tools and ideas
continuous along the unstable foliation of mathematical dynamical system theory are
(defined using the unstable manifolds of indispensable. Arguably the most powerful set of
almost every x  M, which are the sets such tools is ergodic theory which provides a
W u(x), of points y such that statistical description of the dynamics by
1
n
n
lim n!1 n log dðT y, T xÞ < 0). attaching relevant probability measures. In oppo-
Statistical stability T is statistically stable if the sition to single orbits, the statistical properties of
physical measures of nearby deterministic sys- chaotic systems often have good stability proper-
tems are arbitrarily close to the convex hull of ties. In many cases, this allows an understanding
the physical measures of T. of the complexity of the dynamical system and
Stochastic stability T is stochastically stable if even precise and quantitative statistical predic-
the invariant measures of the Markov chains tions of its behavior. In fact, chaotic behavior of
obtained from T by adding a suitable, smooth single orbits often yields global stability
noise with size ϵ ! 0 are arbitrarily close to the properties.
convex hull of the physical measures of T.
Structural stability T is structurally stable if any
S close enough to T is topologically the same as Introduction
T: there exists a homeomorphism h : M ! M
such that h  T ¼ S  h (orbits are sent to The word chaos, from the ancient Greek wαos,
orbits). “shapeless void” (Hesiod 1987) and “raw con-
Subshift of finite type It is a closed subset SF of fused mass” (Ovid 2005), has been used [wαos
S ¼ A ℤ or S ¼ A ℕ where A is a finite set also inspired Van Helmont to create the word
satisfying: SF ¼ {x  S : 8 k < ‘ : xkxkþ1. . . “gas” in the seventeenth century and this other
x‘ F} for some finite set F. thread leads to the molecular chaos of
Topological genericity Let X be a Baire space, Boltzmann in the nineteenth century and there-
e.g., a complete metric space. A property is fore to ergodic theory itself.] since a celebrated
(topologically) generic in a space X (or holds paper of Li and Yorke (1975) to describe evolu-
for the (topologically) generic element of X) if tions which however deterministic and defined
it holds on a nonmeager set (or set of second by rather simple rules, exhibit unpredictable or
Baire category), i.e., on a dense Gδ subset. complex behavior.

Attempts at Definition
Definition of the Subject We note that, like many ideas (Stewart 1964), this
is not captured by a single mathematical defini-
Chaotic dynamical systems are those which pre- tion, despite several attempts (see, e.g., Blanchard
sent unpredictable and/or complex behaviors. The et al. 2002; Glasner and Weiss 1993; Kolyada
existence and importance of such systems has 2004; Ruette 2003b) for some discussions as
been known at least since Hadamard (1898) and well as the monographs on chaotic dynamics
Poincaré (1892), however it became well-known (Arnol’d 1988; Arnol’d and Avez 1968; Brain
only in the sixties. We refer to (Bergelson 2006; and Berger 2001; Brin and Stuck 2002; Collet
Collet et al. 2005; Fiedler 2002; Guckenheimer and Eckmann 2006; Eckmann and Ruelle 1985;
1979; Gutzwiller 1990; Hasselblatt and Katok Guckenheimer and Holmes 1990; Hasselblatt and
2002; Murray 2002; Puu 2000; Saari 2005; Katok 2003; Robinson 2004; Viana 1997b; Young
Starkov 2000) for the relevance of such dynamics 1995). Let us give some of the most well-known
in other fields, mathematical or not (see also definitions, which have been given mostly from
▶ “Ergodic Theory: Interactions with Combina- the topological point of view, i.e., in the setting of
torics and Number Theory”, ▶ “Ergodic Theory: a self-map T : X ! X on a compact metric space
Fractal Geometry”). The numerical simulations of whose distance is denoted by d:
Chaos and Ergodic Theory 635

T has sensitivity to initial conditions on X 0  X 1. Ergodic (cannot be split) and aperiodic (not
if there exists a constant r > 0 such that for every carried by a periodic orbit);
x  X 0, there exists y  X, arbitrarily close to 2. Hyperbolic (nearby orbits converge or diverge
x with a finite separating time: at a definite exponential rate);
3. Sinai–Ruelle–Bowen (as smooth as it is
∃n  0 such that d ðT n y, T n xÞ > r: possible).

In other words, any uncertainty on the exact (For precise definitions we refer to ▶ “Smooth
value of the initial condition x makes T n(x) Ergodic Theory” or to the discussions below.) In
completely unknown for n large enough. If X is a particular such a situation implies nonzero entropy
manifold, then sensitivity to initial conditions in and sensitivity to initial condition of a set of non-
the sense of Guckenheimer (1979) means that the zero Lebesgue measure (i.e., positive volume).
previous phenomenon occurs for a set X 0 with Before starting our survey in earnest, we shall
nonzero volume. describe an elementary and classical example, the
T is chaotic in the sense of Devaney (1989) if it full tent map, on which the basic phenomena can
admits a dense orbit and if the periodic points are be analyzed in a very elementary way. Then, in
dense in X. It implies sensitivity to initial condi- section “Picking an Invariant Probability Mea-
tions on X. sure”, we shall give some motivations for intro-
T is chaotic in the sense of Li and Yorke (1975) ducing probability theory in the description of
if there exists an uncountable subset X 0  X of chaotic but deterministic systems, in particular
points, such that, for all x 6¼ y  X 0, the unpredictability of their individual orbits. We
define two of the most relevant classes of invariant
lim inf dðT n x, T n yÞ ¼ 0 and measures: the physical measures and those maxi-
n!1
lim sup d ðT n x, T n yÞ > 0: mizing entropy. It is unknown in which generality
n!1 these measures exist and can be analyzed but we
describe in section “Tractable Chaotic Dynamics”
T has generic chaos in the sense of Lasota the major classes of dynamics for which this has
(Piorek 1985) if the set been done. In section “Statistical Properties” we
describe some of the finer statistical properties
ðx, yÞ  X  X : lim inf d ðT n x, T n yÞ ¼ 0
n!1
that have been obtained for such good chaotic
systems: sums of observables along orbits are
< lim sup dðT n x, T n yÞg
n!1 statistically undistinguishable from sums of inde-
pendent and identically distributed random vari-
is topologically generic (see “Glossary”) in X  X. ables. Section “Orbit Complexity” is devoted to
Topological chaos is also sometimes character- the other side of chaos: the complexity of these
ized by nonzero topological entropy (▶ Entropy dynamics and how, again, this complexity can be
in Ergodic Theory): there exist exponentially analyzed, and sometimes classified, using ergodic
many orbit segments of a given length. This theory. Section “Stability” describes perhaps the
implies chaos in the sense of Li and Yorke by most striking aspect of chaotic dynamics: the
(Blanchard et al. 2002). unstability of individual orbit is linked to various
As we shall see ergodic theory describes a forms of stability of the global dynamics.
number of chaotic properties, many of them Finally we conclude by mentioning some of the
implying some or all of the above topological most important topics that we could not address
ones. The main such property for a smooth and we list some possible future directions.
dynamical system, say a C1þα-diffeomorphism Caveat. The subject-matter of this article is
of a compact manifold, is the existence of an somewhat fuzzy and we have taken advantage of
invariant probability measure which is: this to steer our path towards some of our favorite
636 Chaos and Ergodic Theory

theorems and to avoid the parts we know less mass constrained to remain on a surface). At that
(some of which are listed below). We make no time, such an unpredictability was considered a
pretense at exhaustivity neither in the topics nor in purely mathematical pathology, necessarily
the selected results and we hope that our col- devoid of any physical meaning [Duhem qualified
leagues will excuse our shortcomings. Hadamard’s result as “an example of a mathe-
matical deduction which can never be used by
Remark 1 In this article we only consider com- physics” (see pp. 206–211 in (Duhem 1906))!].
pact, smooth and finite-dimensional dynamical Returning to out tent map, we can be more
systems in discrete time, i.e., defined by self- quantitative. At any point x  [0, 1] whose orbit
maps. In particular, we have omitted the natural never visits 1/2, the Lyapunov exponent
and important variants applying to flows, e.g., lim n!1 1n log j ðT n Þ0 ðxÞ j is log2. (See the “Glos-
evolutions defined by ordinary differential equa- sary”.) Such a positive Lyapunov exponent corre-
tions but we refer to the textbooks (see, e.g., sponds to infinitesimally close orbits getting
Arnol’d 1988; Hasselblatt and Katok 2002; separated exponentially fast. This can be observed
Katok and Hasselblatt 1995) for these. in Fig. 2. Note how this exponential speed creates
a rather sharp transition.
Elementary Chaos: A Simple Example It follows in particular that experimental or
We start with a toy model: the full tent map T of numerical errors can grow very quickly to size 1,
Fig. 1. Observe that for any point x  [0, 1], T
n [For simple precision arithmetic the uncertainty is
(x) ¼ {(s(k, n)x þ k)  2
n : k ¼ 0, 1, . . ., 2n
1}, 10
16 which grows to size 1 in 38 iterations of T.]
where s(k, n) ¼  1. Hence [n0T
n(x) is dense i.e., the approximate orbit may contain after a
in [0, 1]. It easily follows that T exhibits sensitive while no information about the true orbit. This
dependence to initial conditions. Even worse in casts a doubt on the reliability of simulations.
this example, the qualitative asymptotic behavior Indeed, a simulation of T on most computers will
can be completely changed by this arbitrarily suggest that all orbits quickly converge to 0,
small perturbation: x may have a dense orbit which is completely false [Such a collapse to
whereas y is eventually mapped to a fixed point! 0 does really occurs but only for a countable
This is Devaney chaos (Devaney 1989). subset of initial conditions in [0, 1] whereas the
This kind of unstability was first discovered by points with dense orbit make a subset of [0, 1]
J. Hadamard (1898) in his study of the geodesic
flow (i.e., the frictionless movement of a point

Chaos and Ergodic Theory, Fig. 2 jT n(x)
T n( y)j for

T(x) ¼ 1
j 1
2xj, jx
y j ¼ 10
12 and 0 n 100.
Chaos and Ergodic Theory, Fig. 1 The graph of the full The vertical red line is at n ¼ 28 and shows when jT nx

tent map f(x) ¼ 1
j 1
2xj over [0, 1] T ny j  0.5  10
4 for the first time
Chaos and Ergodic Theory 637

with full Lebesgue measure (see below). This topological entropy (▶ Entropy in Ergodic The-
artefact comes from the way numbers are ory) of T is htop(T) ¼ log 2. [For the coincidence
represented – and approximated – on the com- of the entropy and the Lyapunov exponent see
puter: multiplication by even integers tends to below.] The positivity of the topological entropy
“simplify” binary representations. Thus the com- can be considered as the signature of the complex-
putations involved in drawing Fig. 2 cannot be ity of the dynamics and considered as the defini-
performed too naively.]. Though somewhat atyp- tion, or at least the stamp, of a topologically
ical in its dramatic character, this failure illustrates chaotic dynamics.
the unpredictability and unstability of individual Let us move on to a probabilistic point of view.
orbits in chaotic systems. Pick x  [0, 1] randomly according to, say, the
Does this mean that all quantitative predictions uniform law in [0, 1]. It is then routine to check
about orbits of T are to be forfeited? Not at all, if that i(x) follows the (1/2, 1/2)-Bernoulli law: the
we are ready to change our point of view and look probability that, for any given k, ik(x) ¼ 0 is 1/2
beyond a single orbit. This can be seen easily in and the iks are independent. Thus i(x), seen as a
this case. Let us start with such a global analysis sequence of random 0 and 1 when x is subject to
from the topological point of view. Associate to the uniform law on [0, 1], is statistically
x  [0, 1], a sequence i(x) ¼ i ¼ i0i1i2. . . of 0s and undistinguishable from coin tossing! This impor-
1s according to ik ¼ 0 if Tkx 1/2, ik ¼ 1 other- tant remark leads to quantitative predictions. For
wise. One can check that [Up to a countable set of instance, the strong law of large numbers implies
exceptions.] {i(x) : x  [0, 1]} is the set that, for Lebesgue-almost every x  [0, 1]
S2 ≔ {0, 1}ℕ of all infinite sequences of 0s and (i.e., for all x  [0, 1] except in a set of Lebesgue
1s and that at most one x  [0, 1] can realize a measure zero), the fraction of the time spent in any
given sequence as i(x). dyadic interval I ¼ [k  2
N, ‘  2
N]  [0, 1], k, ‘,
Notice how the transformation f becomes triv- N  ℕ, by the orbit of x,
ial in this representation:
1
lim # 0 k < n : Tkx  I ð1Þ
iðf ðxÞÞ ¼ i1 i2 i3 . . . if iðxÞ ¼ i0 i1 i2 i3 . . . n!1 n

Thus f is represented by the simple and univer- exists and is equal to the length, 2
N, of that
sal “left-shift” on sequences, which is denoted by interval. [Eq. (1) in fact holds for any interval I.
s. This representation of a rather general dynami- This implies that the orbit of almost every
cal system by the left-shift on a space of sequences x  [0, 1] visits all subintervals of [0, 1], i.e.,
is called symbolic dynamics (▶ Symbolic Dynam- the orbit is dense: in complete contradiction with
ics), (Lind and Marcus 1995). the above mentioned numerical simulation!]
This can be a very powerful tool. Observe for More generally, we shall see that, if
instance how here it makes obvious that we have f : [0, 1] ! ℝ is any continuous function, then,
complete combinatorial freedom over the orbits of for Lebesgue almost every-x,
T: one can easily build orbits with various asymp-
totic behaviors: if a sequence of S2 contains all the 1
n
1
lim f Tkx exists and is equal to fðxÞ dx:
finite sequences of 0s and 1s, then the n!1 n
k¼0
corresponding point has a dense orbit; if the ð2Þ
sequence is periodic, then the corresponding
point is itself periodic, to give two examples of Using strong mixing properties (▶ Ergodicity
the richness of the dynamics. and Mixing Properties) of the Lebesgue measure
More quantitatively, the number of distinct under T, one can prove further properties, e.g.,
subsequences of length n appearing in sequences sensitivity on initial conditions in the sense of
i(x), x  [0, 1], is 2n. It follows that the Guckenheimer [The Lebesgue measure is weak-
638 Chaos and Ergodic Theory

mixing: Lebesgue-almost all couples of points By the following fundamental theorem

(x, y)  [0, 1]2 get separated. Note that it is not ▶ Ergodic Theorems, each such measure is the
true of every couple off the diagonal: Counter- statistical description of some orbit:
examples can be found among couples (x, y) with
T nx ¼ T ny arbitrarily close to 1.] and study the Birkhoff Pointwise Ergodic Theorem
fluctuations of the averages 1n n
1k¼0 fðTxÞ by the Let (X, F , m) be a space with a s-field and a
way of limit theorems. probability measure. Let f : X ! X be a measure-
The above analysis relied on the very special preserving map, i.e., f
1(F )  F and m  f
1 ¼ m.
structure of T but, as we shall explain, the ergodic Assume ergodicity of m (See the “Glossary”) and
theory of differentiable dynamical systems shows (absolute) integrability of f : X ! ℝ with respect
that all of the above (and much more) holds in to m. Then for m-almost every x  X,
some form for rather general classes of chaotic
n
1
systems. The different chaotic properties are inde- 1
pendent in general (e.g., one may have topological lim f f kx exists and is f dm:
n!1 n
k¼0
chaos whereas the asymptotic behavior of almost
all orbits is periodic) and the proofs can become This theorem can be interpreted as saying that
much more difficult. We shall nonetheless be “time averages” coincide almost surely with
rewarded for our efforts by the discovery of unex- “ensemble averages” (or “phase space average”),
pected links between chaos and stability, com- i.e., that Boltzmann’s Ergodic Hypothesis of sta-
plexity and simplicity as we shall see. tistical mechanics (Gallavotti 1999) holds for
dynamical systems that cannot be split in a mea-
surable and non trivial way. [This
Picking an Invariant Probability Measure indecomposability is however often difficult to
establish. For instance, for the hard ball model
One could think that dynamical systems, such as of a gas it is known only under some generic
those defined by self-maps of manifolds, being assumption (see (Szász 2000) and the references
completely deterministic, have nothing to do therein).] We refer to (Krengel 1983) for
with probability theory. There are in fact several background.
motivations for introducing various invariant
probability measures. Remark 2 One should observe that the existence
of the above limit is not at all obvious. In fact it
Statistical Descriptions often fails from other points of view. One can
An abstract goal might be to enrich the structure: a show that for the full tent map T(x) ¼ 1
j 1
2xj
smooth self-map is a particular case of a Borel analyzed above and many functions f, the set of
self-map, hence one can canonically attach to this points for which it fails is large both from the
map its set of all invariant Borel probability mea- topological point of view (it contains a dense Gδ
sures [From now on all measures will be Borel set) and from the dimension point of view (it has
probability measures except if it is explicitly stated Hausdorff dimension 1 (Barreira and Schmeling
otherwise.], or just the set of ergodic [A measure is 2000)). This is an important point: the introduc-
ergodic if all measurable invariant subsets have tion of invariant measures allows one to avoid
measure 0 or 1. Note that arbitrary invariant some of the wilder pathologies.
measures are averages of ergodic ones, so many
questions about invariant measures can be To illustrate this let us consider the full tent
reduced to ergodic ones.] ones. By the Krylov– map T(x) ¼ 1
j 1
2xj again and the two
Bogoliubov theorem (see, e.g., Katok and ergodic invariant measures: δ0 (the Dirac measure
Hasselblatt 1995), this set is non-empty for any concentrated at the fixed point 0) and the
continuous self-map of a compact space. Lebesgue measure dx. In the first case, we obtain
Chaos and Ergodic Theory 639

a complex proof of the obvious fact that the time observed in most of the situations (see however
average at x ¼ 0 (some set of full measure!) and the caveat in the discussion of the full tent map).
the ensemble average with respect to δ0 are both The existence of a finite statistical description
equal to f(0). In the second case, we obtain a very (or even of a physical measure) is, as we shall see,
general proof of the above Eq. (2). not automatic nor routine to prove. Attracting
Another type of example is provided by the periodic points as in the above silly example pro-
contracting map S : [0, 1] ! [0, 1], S(x) ¼ x/2. vide a first type of physical measures. Birkhoff
S has a unique invariant probability measure, δ0. ergodic theorem asserts that absolutely continu-
For Birkhoff theorem the situation is the same as ous ergodic invariant measures, usually obtained
that of T and δ0: it asserts only that the orbit of 0 is from some expansion property, give another class
described by δ0. of physical measures. These contracting and
One can understand Birkhoff theorem as a (first expanding types can be combined in the class of
and rather weak) stability result: the time averages Sinai–Ruelle–Bowen measures (Ledrappier 1984)
are independent of the initial condition, almost which are the invariant measures absolutely con-
surely with respect to m. tinuous “along expanding directions” (see for the
precise but technical definition ▶ Smooth Ergodic
Physical Measures Theory). Any ergodic Sinai–Ruelle–Bowen mea-
In the above silly example S, much more is true sure which is ergodic and without zero Lyapunov
than the conclusion of Birkhoff Theorem: all exponent [That is, the set of points x  M such
points of [0, 1] are described by δ0. This leads to that lim n!1 1n log ðf n Þ0 ðxÞ:v ¼ 0 for some
the definition of the basin of a probability measure
v  TxM has zero measure.] is a physical mea-
m for a self-map f of a space M:
sure. Conversely, “most” physical measures [For
counter-examples see (Hofbauer and Keller
ℬðmÞ≔fx  M : 8f :
1990).] are of this type (Tsujii 2005; Vasquez
n
1
1 2007).
M ! ℝ continuous lim f f k x ¼ f dm :
n!1 n
k¼0
Measures of Maximum Entropy
If M is a manifold, then there is a notion of For all parameters t  [3.96, 4], the quadratic
volume and one can make the following defini- maps Qt(x) ¼ tx(1
x), Qt : [0, 1] ! [0, 1],
tion. A physical measure is a probability measure have nonzero topological entropy (de Melo and
whose basin has nonzero volume in M. Say that a van Strien 1993) and exponentially many periodic
dynamical system f : M ! M on a manifold has a points (Hofbauer 1985):
finite statistical description if there exists finitely
many invariant probability measures m1, . . ., mn # x  ½0, 1 : Qnt ðxÞ ¼ x
lim ¼ 1:
the union of whose basins is the whole of M, up to n!1 enhtop ðQt Þ
a set of zero Lebesgue measure.
Physical measures are among the main subject On the other hand, by a deep theorem (Graczyk
of interest as they are expected to be exactly those and Świątek 1997; Lyubich 1997) there is an open
that are “experimentally visible”. Indeed, if and dense subset of t  [0, 4], such that Qt has a
x0  ℬ(m) and ϵ0 > 0 is small enough, then, by unique physical measure concentrated on a peri-
Lebesgue density theorem, a point x picked odic orbit! Thus the physical measures can
according to, say, the uniform law in the ball completely miss the topological complexity (and
B(x0, ϵ0) of center x0 and radius ϵ0, will be in in particular the distribution of the periodic
ℬ(m) with probability almost 1 and therefore its points). Hence one must look at other measures
ergodic averages will be described by m. Hence to get a statistical description of the complexity of
“experiments” can be expected to follow the phys- such Qt. Such a description is often provided by
ical measures and this is what is numerically measures of maximum entropy mM whose
640 Chaos and Ergodic Theory

measured entropy [The usual phrases are • The equidistribution of periodic points with
“measure-theoretic entropy”, “metric entropy”.] respect to some maximum entropy measure
(▶ Entropy in Ergodic Theory) satisfies: mM :
1
mM ¼ lim #fx  X:x¼f n
xg dx :
hðf , mM Þ ¼ sup hðf , mÞ¼1 htop ðf Þ :
n!1
x¼f n x
m  Mðf Þ • The holonomy invariance which can be loosely
interpreted by saying that the past and the future
M( f ) is the set of all invariant measures. [One are independent conditionally on the present.
can restrict this to the ergodic invariant measures
without changing the value of the supremum Other Points of View
(▶ Entropy in Ergodic Theory).] Equality Many other invariant measures are of interest in
1 above is the variational principle: it holds for various contexts and we have made no attempt at
all continuous self-maps of compact metric completeness: for instance, invariant measures
spaces. One can say that the ergodic complexity maximizing dimension (Gatzouras and Peres
(the complexity of f as seen by its invariant mea- 1997; Pesin 1997), or pressure in the sense of
sures) captures the full topological complexity the thermodynamical formalism (Katok and
(defined by counting all orbits). Hasselblatt 1995; Ruelle 2004), or some energy
(Anantharaman 2004; Contreras et al. 2001;
Remark 3 The variational principle implies the Jenkinson 2007), or quasi-physical measures
existence of “complicated invariant measures” as describing the dynamics around saddle-type
soon as the topological entropy is nonzero (see invariant sets (Eckmann and Ruelle 1985) or in
(Bonano and Collet 2006) for a setting in which systems with holes (Chernov et al. 2000).
this is of interest).
Tractable Chaotic Dynamics
Maximum entropy measures do not always
exist. However, if f is C1 smooth, then maximum The Palis Conjecture
entropy measures exist by a theorem of Newhouse There is, at this point, no general theory allowing
(1989) and they indeed describe the topological the analysis of all dynamical systems or even of
complexity in the following sense. Consider the most of them despite many recent and exciting
probability measures: developments in the theory of generic C1-
diffeomorphisms (Bonatti et al. 2005; Crovisier
1
n
1 2006a). In particular, the question of the generality
mn,ϵ ≔ df k x in which physical measures exist remains open.
n k¼0 x  Eðn, ϵÞ
One would like generic systems to have a finite
statistical description (see section “Physical Mea-
where E(n, ϵ) is an arbitrary (ϵ, n)-separated subset sures”). This fails in some examples but these look
[See ▶ Entropy in Ergodic Theory: 8x, y  exceptional and the following question is asked by
E(n, ϵ) x 6¼ y ) ∃ 0 k < n d(T kx, T ky)  ϵ.] Palis (2000):
of M with maximum cardinality. Then accumula- Is it true that any dynamical system defined by a Cr-
tion points for the weak star topology on the space diffeomorphism on a compact manifold can be
of probability measures on M of mn,ϵ when n ! 1 transformed by an arbitrarily small Cr-perturbation
and then ϵ ! 0 are maximum entropy measures to another dynamical system having a finite statis-
tical description?
(Misiurewicz 1976a).
Let us quote two important additional proper- This is completely open though widely
ties, discovered by Margulis (2004), that often believed [Observe, however, that such a statement
hold for the maximum entropy measures: is false for conservative diffeomorphisms with
Chaos and Ergodic Theory 641

high order smoothness as KAM theory implies Remark 4 Mañé Stability Theorem (see below)
stable existence of invariant tori foliating a subset shows that uniform hyperbolicity is a very natural
of positive volume.]. Note that such a good notion. One can also understand on a more tech-
description is not possible for all systems (see, nical level uniform hyperbolicity as what is
e.g., Hofbauer and Keller 1990; Newhouse needed to apply an implicit function theorem in
1974). Note that one would really like to ask some functional space (see, e.g., Shub 1987).
about unperturbed “typical” [The choice of the
notion of typicality is a delicate issue. The The existence of a finite statistical description
Newhouse phenomenon shows that among C2- for such systems has been proved since the 1970s
diffeomorphisms of multidimensional compact by Bowen, Ruelle and Sinai (Bowen and Ruelle
manifolds, one cannot use topological genericity 1975; Ruelle 1976; Ya 1972) (the expanding case
and get a positive answer. Popular notions are is much simpler (Krzyżewski and Szlenk 1969)).
prevalence and Kolmogorov genericity – see the”
Glossary”.] dynamical systems in a suitable sense, Theorem 1 Let f : M ! M be a C1þα map of a
but of course this is even harder. compact manifold. Assume f to be (i) a uniformly
One is therefore led to make simplifying expanding map on M or (ii) a uniformly hyper-
assumptions: typically of small dimension, uni- bolic diffeomorphism.
form expansion/contraction or geometry.
• f admits a finite statistical description by ergo-
Uniformly Expanding/Hyperbolic Systems dic and hyperbolic Sinai–Ruelle–Bowen mea-
The most easily analyzed systems are those with sures (absolutely continuous in case (i)).
uniform expansion and/or contraction, namely the • f has finitely many ergodic maximum entropy
uniformly expanding maps and uniformly hyper- measures, each of which makes f isomorphic to
bolic diffeomorphisms, see ▶ “Smooth Ergodic a finite state Markov chain. The periodic points
Theory”. [We require uniform hyperbolicity on are uniformly distributed according to some
the so-called chain recurrent set. This is equiva- canonical average of these ergodic maximum
lent to the usual Axiom-A and no-cycle condition.] entropy measures.
An important class of example is obtained as • f is topologically conjugate [Up to some negli-
follows. Consider A : ℝd ! ℝd, a linear map gible subset.] to a subshift of finite type (See
preserving ℤd (i.e., A is a matrix with integer the “Glossary”.)
coefficients in the canonical basis) so that it
defines a map A : d ! d on the torus. If there The construction of absolutely continuous
is a constant Λ > 1 such that for all v  ℝd, kA. invariant measures for a uniformly expanding
vk  Λkvk then A is a uniformly expanding map. map f can be done in a rather direct way by
If A has determinant 1 and no eigenvalue on the considering the pushed forward measures
unit circle, then A is a uniformly hyperbolic 1 n
1 k
n k¼0 f Leb and taking weak star limits while
diffeomorphism (▶ Smooth Ergodic Theory) (see preventing the appearance of singularities, by,
also Brin and Stuck 2002; Katok and Hasselblatt e.g., bounding some Hölder norm of the density
1995; Robinson 2004; Shub 1987). Moreover all using expansion and distortion of f.
C1-perturbations of the previous examples are The classical approach to the uniformly hyper-
again uniformly expanding or uniformly bolic dynamics (Bowen 1975; Ruelle 2004; Shub
hyperbolic. [One can define uniform hyperbolicity 1987) is through symbolic dynamics and coding.
for flows and an important class of examples is Under the above hypothesis one can build a finite
provided by the geodesic flow on compact mani- partition of M which is tailored to the dynamics
folds with negative sectional curvature (Katok (a Markov partition) so that the corresponding
and Hasselblatt 1995).] These uniform systems symbolic dynamics has a very simple structure:
are sometimes called “strongly chaotic”. it is a full shift {1, . . ., d}ℤ, like in the example of
642 Chaos and Ergodic Theory

the full tent map, or a subshifts of finite type. The In this full generality, one already obtains sig-
above problems can then be solved using the nificant results:
thermodynamical formalism inspired from the
statistical mechanics of one-dimensional ferro- • The entropy is bounded by the expansion:
magnets (Ruelle 1968): ergodic properties are d

obtained through the spectral properties of a suit- hðf , mÞ lþ

i ðmÞ (Ruelle 1978)
i¼1
able transfer operator acting on some space of • At almost every point x, there are strong stable
regular functions, e.g., the Hölder-continuous resp. unstable manifolds W ss(x), resp. W uu(x),
functions defined over the symbolic dynamics coinciding with the sets of points y such that
with respect to the distance d ðx, yÞ≔ n  ℤ d(T nx, T ny) ! 0 exponentially fast when
2
n 1xn 6¼yn where 1s6¼t is 1 if s 6¼ t, 0 otherwise. n ! 1, resp. n !
1. The corresponding
A recent development (Baladi and Tsujii 2007; holonomies are absolutely continuous (see,
Blank et al. 2002; Gouëzel and Liverani 2006) has e.g., Brin 2001) like in the uniform case. This
been to find suitable Banach spaces to apply the allows Ledrappier’s definition of Sinai–
transfer operator technique directly in the smooth Ruelle–Bowen measures (Ledrappier 1984) in
setting, which not only avoids the complication of that setting.
coding (or rather replace them with functional • Equality in the above formula holds if and only
analytic preliminaries) but allows the use of the if m is a Sinai–Ruelle–Bowen measure
smoothness beyond Hölder-continuity which is (Ledrappier and Young 1985). More generally
important for finer ergodic properties. the entropy can be computed as
Uniform expansion or hyperbolicity can easily d þ
i¼1 gi ðmÞli ðmÞ where the γi are some fractal
be obstructed in a given system: a “bad” point dimensions related to the exponents.
(a critical point or a periodic point with multiplier
with an eigenvalue on the unit circle) is enough.
Under the only assumption of hyperbolicity
This leads to the study of other systems and has
(i.e., no zero Lyapunov exponent almost every-
motivated many works devoted to relaxing the
where), one gets further properties:
uniform hyperbolicity assumptions (Bonatti
et al. 2005). • Existence of an hyperbolic measure which is
not periodic forces htop( f ) > 0 [However
Pesin Theory h( f, m) can be zero.] by (Katok 1980).
The most general such approach is Pesin theory. • m is exact dimensional (Barreira et al. 1999;
Let f be a C1þα-diffeomorphism [It is an important Young 1982): the limit limr!0 log m(B(x, r))/
open problem to determine to which extent Pesin log r exists m-almost everywhere and is equal
theory could be generalized to the C1setting.] to the Hausdorff dimension of m (the infimum
f with an ergodic invariant measure m. By of the Hausdorff dimension of the sets with full
Oseledets Theorem ▶ Smooth Ergodic Theory, m-measure). This is deduced from a more tech-
for almost every x with respect to any invariant nical “asymptotic product structure” property
measure, the behavior of the differential Tx f n for of any such measure.
n large is described by the Lyapunov exponents
l1, . . ., ld, at x. Pesin is able to build charts around For hyperbolic Sinai–Ruelle–Bowen measures
almost every orbit in which this asymptotic linear m, one can then prove, e.g.,:
behavior describes that of f at the first iteration.
That is, there are diffeomorphisms • local ergodicity (Pesin 1976): m has at most
Fx : Ux  M ! Vx  ℝd with a “reasonable countably many ergodic components and
dependence on x” such that the differential of m-almost every point has a neighborhood
Ff n x  f n  F
1x at any point where it is defined is whose Lebesgue-almost every point are
close to a diagonal matrix with entries contained in the basin of an ergodic component
eðl1 ϵÞn , eðl2 ϵÞn , . . . , eðld ϵÞn . of m.
Chaos and Ergodic Theory 643

• Bernoulli (Ornstein and Weiss 1988): Each Conjecture 2 Let f be a C1þϵ-diffeomorphism of

ergodic component of m is conjugate in a a compact surface. If the topological entropy of
measure-preserving way, up to a period, to a f is nonzero then f has at most countably many
Bernoulli shift, that is, a full shift {1, . . ., N}ℤ ergodic invariant measures maximizing entropy.
equipped with a product measure. This in par-
ticular implies mixing and sensitivity on initial The analogue of this conjecture has been pro-
conditions for a set of positive Lebesgue ved for C1þϵ interval maps (Buzzi 1997, 2000b,
measure. Buzzi). In the above setting a classical result of
Katok shows the existence of uniformly hyper-
However, establishing even such a weak form bolic compact invariant subsets with topological
of hyperbolicity is rather difficult. The fragility of entropy arbitrarily close to that of f implying the
this condition can be illustrated by the result existence of many periodic points:
(Bochi 2002; Bochi and Viana 2005) that the
1
topologically generic area-preserving surface C1- lim sup log#fx  M : f n ðxÞ ¼ xg  htop ðf Þ:
n!1 n
diffeomorphism is either uniformly hyperbolic or
has Lebesgue almost everywhere vanishing
The previous conjecture would follow from the
Lyapunov exponents, hence is never non-
following one:
uniformly hyperbolic! (but this is believed to be
very specific to the very weak C1 topology).
Conjecture 3 Let f be a C1þϵ-diffeomorphism of
Moreover, such weak hyperbolicity is not enough,
a compact manifold. There exists an invariant
with the current techniques, to build Sinai–
subset X  M, carrying all ergodic measures
Ruelle–Bowen measures or analyze maximum
with maximum entropy, such that the restriction
entropy measures only assuming non-zero
f j X is conjugate to a countable state topological
Lyapunov exponents. Let us though quote two
Markov shift (See the “Glossary”).
conjectures. The first one is from (Viana 1998)
[We slightly strengthened Viana’s statement for Systems with Discontinuities
expository reasons.]: We now consider stronger assumptions to be able
to build the relevant measures.
Conjecture 1 Let f be a C1þϵ-diffeomorphism of The simplest step beyond uniformity is to allow
a compact manifold. If Lebesgue-almost every discontinuities, considering piecewise expanding
point x has well-defined Lyapunov exponents in maps. The discontinuities break the rigidity of the
every direction and none of these exponents is uniformly expanding situation. For instance, their
zero, then there exists an absolutely continuous symbolic dynamics are usually no longer subshifts
invariant s-finite positive measure. of finite type though they still retain some “sim-
plicity” in good cases (see Buzzi 2005).
The analogue of this conjecture has been pro- To understand the problem in constructing the
ved for C3 interval maps with unique critical point absolutely continuous invariant measures, it is
and negative Schwarzian derivative by Keller instructive to consider the pushed forwards of a
(1990), but only partial results are available for smooth measure. Expansion tends to keep the mea-
diffeomorphisms (Leplaideur 2004). sure smooth whereas discontinuities may pile it up,
We turn to measures of maximum entropy. As creating non-absolute continuity in the limit. One
we said, C1 smoothness is enough to ensure their thus has to check that expansion wins. In dimen-
existence but this is through a functional-analytic sion 1, a simple fact resolves the argument: under a
argument (allowed by Yomdin theory (Yomdin high enough iterate, one can make the expansion
1987)) which says nothing about their structure. arbitrarily large everywhere, whereas a small inter-
Indeed, the following problem is open: val can be chopped into at most two pieces.
644 Chaos and Ergodic Theory

Lasota and Yorke (1973) found a suitable shown that, for any r > 1, an open and dense
framework. They considered C2 interval maps subset of piecewise Cr and expanding maps have
with jf 0(x) j  const > 1 except at finitely many a finite statistical description.
points. They used the Ruelle transfer operator Piecewise hyperbolic diffeomorphisms are
directly on the interval. Namely they studied more difficult to analyze though several results
(conditioned on technical assumptions that can
fðyÞ be checked in many cases) are available (Baladi
ðLfÞðxÞ ¼
j T 0 ðyÞ j and Gouëzel 2008; Chernov 1999; Sataev 1992;
y  T
1 x
Young 1985).
acting on functions f : [0, 1] ! ℝ with bounded
variation and obtained the invariant density as the Interval Maps with Critical Points
eigenfunction associated to the eigenvalue 1. One A more natural but also more difficult situation is
can then prove a Lasota–Yorke inequality (which a map for which the uniformity of the expansion
might more accurately be called Doeblin–Fortet fails because of the existence of critical points.
since it was introduced in the theory of Markov [Note that, by a theorem of Mañé a circle map
chains much earlier): without critical points or indifferent periodic
point is either conjugate to a rotation or uniformly
kLfkBV akfkBV þ bkfk1 ð3Þ expanding (Mañé 1985).]
A class which has been completely analyzed at
where kkBV, kk1 are a strong and a weak norm, the level of the above conjecture is that of real-
respectively and α < 1 and β < 1. One can then analytic families of maps of the interval ft : [0, 1]
apply general theorems (Ionescu Tulcea and ! [0, 1], t  I, with a unique critical point, the
Marinescu 1950) or (Nussbaum 1970) (see main example being the quadratic family
(Baladi 2000a) for a detailed presentation of this Qt(x) ¼ tx(1
x) for 0 t 4.
approach and its variants). Here α can essentially It is not very difficult to find quadratic maps
be taken as 2 (reflecting the locally simple discon- with the following two types of behavior:
tinuities) divided by the minimum expansion: so
α < 1, perhaps after replacing T with an iterate. In (stable) the orbit of Lebesgue-almost every
particular, the existence of a finite statistical x  [0, 1] tends to an attracting periodic orbit;
description then follows (see Broise (1996a) for (chaotic) there is an absolutely continuous invari-
various generalizations and strengthenings of this ant probability measure m whose basin con-
result on the interval). tains Lebesgue-almost every x  [0, 1].
The situation in higher dimension is more com-
plex for the reason explained above. One can To realize the first it is enough to arrange the
obtain inequalities such as (3) on suitable if less critical point to be periodic. One can easily prove
simple functional spaces (see, e.g., Saussol 2000) that this stable behavior occurs on an open set of
but proving α < 1 is another matter: discontinu- parameters –thus it is stable with respect to the
ities can get arbitrarily complex under iteration. parameter or the dynamical system. The second
(Buzzi 2001; Tsujii 2000b) show that indeed, in occurs for Q4 with m ¼ dx= pxð1
xÞ. It is
dimension 2 and higher, piecewise uniform much more difficult to show that this chaotic
expansion (with a finite number of pieces) is not behavior occurs for a set of parameters of positive
enough to ensure a finite statistical description if Lebesgue measure. This is a theorem of Jakobson
the pieces of the map have only finite smoothness. (1981) for the quadratic family (see for a recent
In dimension 2, resp. 3 or more, piecewise real- variant (Young and Wang 2006)). Let us sketch
analytic, resp. piecewise affine, is enough to two main ingredients of the various proofs of this
exclude such examples (Buzzi 2000a; Tsujii theorem. The first is inducing: around Lebesgue-
2000a), resp. (Tsujii 2001). (Cowieson 2002) has almost every point x  [0, 1] one tries to find a
Chaos and Ergodic Theory 645

time t(x) and an interval J(x) such that ft(x) : • The set of t such that ft is stable is open and
J(x) ! ft(x)(J(x)) is a map with good expansion dense;
and distortion properties. This powerful idea • The remaining set of parameters has zero
appears in many disguises in the non-uniform Lebesgue measure. [This set of “strange
hyperbolic theory (see for instance (Hofbauer parameters” of zero Lebesgue measure has
1979; Young 1998)). The second ingredient is however positive Hausdorff dimension
parameter exclusion: one removes the parameters according to work of Avila and Moreira. In
at which a good inducing scheme cannot be built. particular each of the following situations is
More precisely one proceeds inductively, realized on a set of parameters t of positive
performing simultaneously the inducing and the Hausdorff dimension: non-existence of the
exclusion, the good properties of the early stage Birkhoff limit at Lebesgue-almost every point,
of the inducing allowing one to control the measure the physical measure is δp for p a repelling fixed
of the parameters that need to be excluded to con- point, the physical measure is non-ergodic.]
tinue (Benedicks and Carleson 1985; Jakobson
1981). Indeed, the expansion established at a
given stage allows to transfer estimates in the We note that the theory underlying the above
dynamical space to the parameter space. theorem yields much more results, including a
Using methods from complex analysis and very paradoxical rigidity of typical analytic fam-
renormalization theory one can go much further ilies as above. See (Avila and Moreira 2005).
and prove the following difficult theorems
(actually the product of the work of many people, Non-uniform Expansion/Contraction
including Avila, Graczyk, Kozlovski, Lyubich, Beyond the dimension 1, only partial results are
de Melo, Moreira, Shen, van Strien, Swiatek), available. The most general of those assume uni-
which in particular solves Palis conjecture in form contraction or expansion along some direc-
this setting: tion, restricting the non-uniform behavior to an
invariant sub-bundle often one-dimensional, or
Theorem 2 (Graczyk and Świątek 1997; “one-dimensional-like”.
Kozlovski et al. 2007; Lyubich 1997) Stable A first, simpler situation is when there is a
maps (that is, such that Lebesgue almost every dominated decomposition with a uniformly
orbit converges to one of finitely many periodic expanding term: there is a continuous and invari-
orbits) form an open and dense set among C r ant splitting of the tangent bundle,
interval maps, for any r  2. [In fact this is even TΛM ¼ Euu Ecs, over some Λ an attracting set:
true for polynomials.]. for all unit vectors vu  Euu, vc  Ecs,

The picture has been completed in the ðf n Þ0 ðxÞ:vu  Cln and

unimodal case (that is, with a unique critical
point): ðf n Þ0 ðxÞ:vc Cmn ðf n Þ0 ðxÞ:vu :

Theorem 3 (Avila and Moreira 2005; Avila Standard techniques (pushing the Riemannian
et al. 2003; Graczyk and Świątek 1997; volume of a piece of unstable leaf and taking
Kozlovski 2003; Lyubich 1997) Let ft : [0, 1] limits) allow the construction of Gibbs u-states
! [0, 1], t  [t0, t1], be a real-analytic family of as introduced by (Pesin and Sinaĭ 1982).
unimodal maps. Assume that it is not degenerate
[f t0 and f t1 are not conjugate]. Then: Theorem 4 (Alves–Bonatti–Viana (2000)) [A
slightly different result is obtained in (Bonatti
• The set of t such that ft is chaotic in the above and Viana 2000).] Let f : M ! M be a C2
sense has positive Lebesgue measure; diffeomorphism with an invariant compact subset
646 Chaos and Ergodic Theory

Λ. Assume that there is a dominated splitting Λ  M on which the tangent bundle has an invari-
TΛM ¼ Eu Ecs such that, for some c > 0, ant continuous decomposition TΛM ¼ E+ E

such that all vectors of E+ ∖ {0}, resp. E
∖ {0},
1
n
1 have positive, resp. negative, Lyapunov expo-
lim sup log f 0 ðf x Þ j Ecs
c < 0 nents. Then any zero-noise limit measure m of f is
n!1 n k¼0
a Sinai–Ruelle–Bowen measure and therefore, if it
is ergodic and hyperbolic, a physical measure.
on a subset of Λ of positive Lebesgue measure.
Then this subset is contained, up to a set of zero
One can hope, that typically, the latter ergodic-
Lebesgue measure, in the union of the basins of
ity and hyperbolicity assumptions are satisfied
finitely many ergodic and hyperbolic Sinai –
(see, e.g., Baraviera and Bonatti 2003).
Ruelle – Bowen measures.
By pushing classical techniques and introduc-
ing new ideas for generic maps with one
The non-invertible, purely expansive version
expanding and one weakly contracting direction,
of the above theorem can be applied in particular
Tsujii has been able to prove the following generic
to the following maps of the cylinder (d  16, a is
result (which can be viewed as a 2-dimensional
properly chosen close to 2 and α is small):
extension of some of the one-dimensional results
above: one adds a uniformly expanding
f ðy, xÞ ¼ dx mod 2p, a
x2 þ ϵ sinðyÞ
direction):

which are natural examples of maps with multi-

dimensional expansion and critical lines consid- Theorem 6 (Tsujii 2005) Let M be a compact
ered by Viana (1997a). A series of works have surface. Consider the space of C20 self-maps
shown that the above maps fit in the above non- f : M ! M which admits directions that are uni-
uniformly expanding setting with a proper control formly expanded [More precisely, there exists a
of the critical set and hence can be thoroughly continuous, forward invariant cone field which is
analyzed through variants of the above theorem uniformly expanded under the differential of f.]
(Alves 2000; Viana 1997a) and the references in
(Alves 2006). For a, b properly chosen close to 2 Then existence of a finite statistical description
and small ϵ, the following maps should be even is both topologically generic and prevalent in this
more natural examples: space of maps.

f ðx, yÞ ¼ a
x2 þ ϵy, b
y2 þ ϵx : ð4Þ Hénon-Like Maps and Rank One Attractors

In 1976, Hénon (1976) observed that the
However the inexistence of a dominated split- diffeomorphism of the plane
ting has prevented the analysis of their physical
measures. See (Buzzi 2000b; Buzzi) for their Ha,b ðx, yÞ ¼ 1
ax2 þ y, bx
maximum entropy measures.
Cowieson and Young have used completely seemed to present a “strange attractor” for a ¼ 1.4
different techniques (thermodynamical formal- and b ¼ 0.3, that is, the points of a numerically
ism, Ledrappier–Young formula and Yomdin the- simulated orbit seemed to draw a set looking
ory on entropy and smoothness) to prove the locally like the product of a segment with a Cantor
following result (see (Araujo and Tahzibi 2005) set. This attractor seems to be supported by the
for related work): unstable manifold W u(P) of the hyperbolic fixed
point P with positive absciss. On the other hand,
Theorem 5 (Cowieson–Young 2005) Let the Plykin classification (Plykin 2002) excluded
f : M ! M be a C1 diffeomorphism of a compact the existence of a uniformly hyperbolic attractor
manifold. Assume that f admits an attractor for a dissipative surface diffeomorphism.
Chaos and Ergodic Theory 647

For almost twenty years the question of the • Misiurewicz condition: d( f nc, d) > c > 0 for
existence of such an attractor (by opposition to all n  1 and all critical points c, d.
an attracting periodic orbit with a very long
period) remained open. Indeed, one knew since Assume the following transversality condition
Newhouse that, for many such maps, there exist on f at a ¼ 0: for every critical point c,
infinitely many such periodic orbits which are (d/da)( fa(ca)
pa) 6¼ 0 if ca is the critical point
very difficult to distinguish numerically. But in of fa near c and pa is the point having the same
1991 Benedicks and Carleson succeeded in pro- itinerary under fa as f(c) under c. Assume the
posing an argument refining (with considerable following non-degeneracy of T : f 00 ðcÞ ¼ 0 )
difficulties) their earlier proof of Jakobson one- @T 00 ðc, 0Þ=@y 6¼ 0.
dimensional theorem and established the first part
of the following theorem: • Tab restricted to a neighborhood of S1  {0}
has a finite statistical description by a number
Theorem 7 (Benedicks–Carleson 1991) For of hyperbolic Sinai–Ruelle–Bowen measures
any ϵ > 0, for jbj small enough, there is a set bounded by the number of critical points of f;
A with Leb(A) > 0 satisfying: for all a  A, there • There is exponential decay of correlations and
exists z  W u(P) such that a Central Limit Theorem (see below) – except,
in an obvious way, if there is a periodic interval
• The orbit of z is dense in W u(P); with period >1;
• lim inf n!1 1n log ðf n Þ0 ðzÞ > 0. • There is a natural coding of the orbits that
remains for ever close to S1  {0} by a closed
invariant subset of a full shift.
Further properties were then established, espe-
cially by Benedicks, Viana, Wang, Young
Very importantly, the above dynamical situa-
(Benedicks and Viana 2001; Benedicks and
tion has been shown to occur near typical homo-
Young 1993, 2000; Young and Wang 2001). Let
clinic tangencies: (Mora and Viana 1993)
us quote the following theorem of Wang and
proved that there is an open and dense subset
Young which includes the previous results:
of the set of all C3 families of diffeomorphisms
unfolding a first homoclinic tangency such that
Theorem 8 (Young and Wang 2001) Let
the above holds. However (Palis and Yoccoz
Tab : S1  [
1, 1] ! S1  [
1, 1] be such that
2001) shows that the set of parameters with a
Henon-like attractor has zero Lebesgue density
• Ta0(S1  [
1, 1])  S1  {0};
at the bifurcation itself, at least under an
• For b > 0, Tab is a diffeomorphism on its image
assumption on the so-called stable and unstable
with
Hausdorff dimensions. (Diaz et al. 1996) estab-
c
1b j det Tab(x, y) j cb
lishes positive density for another type of bifur-
for some c > 1 and all (x, y)  S1  [
1, 1]
cation. Furthermore (Moreira et al. 2001) has
and all (a, b).
related the Hausdorff dimensions to the abun-
dance of uniformly hyperbolic dynamics near
Let fa : S1 ! S1 be the restriction of Ta0. the tangency.
Assume that f ¼ f0 satisfies: Viana (1993) is able to treat situations with
more than one contracting direction. More
• Non-degenerate critical points: f 0(c) ¼ 0 ) recently (Young and Wang 2008) has proposed a
f 00(c) 6¼ 0; rather general framework, with easily checkable
• Negative Schwarzian derivative: for all assumptions in order to establish the existence of
x  S1 non-critical, f 0 0 0(x)/f 0(x)
3/2( f 00(x)/ such dynamics in various applications. See also
f 0(x))2 < 0; (Guckenheimer et al. 2006; Wang and Young
• No indifferent or attracting periodic point, i.e., 2003) for applications.
x such that f n(x) ¼ x and j( f )0(x) j 1;
648 Chaos and Ergodic Theory

Statistical Properties 4. The map T has the “big image property”:

infa  αm(Ta) > 0.
The ergodic theorem asserts that time averages of
integrable functions converge to phase space aver- Some piecewise expanding and C2 maps are
ages for any ergodic system. The speed of con- obviously Gibbs–Markov but the realpoint is that
vergence is quite arbitrary in that generality many dynamics can be reduced to that class by the
(Krengel 1983) (only upcrossing inequalities use of inducing and tower constructions as in
seem to be available (Bishop 1967/1968; (Young 1998), in particular. This includes possibly
Hochman 2006)), however many results are avail- piecewise uniformly hyperbolic diffeomorphisms,
able under very natural hypothesis as we are going Collet–Eckmann maps of the interval (Baladi
to explain in this section. The underlying idea is 2000a) (typical chaotic maps in the quadratic fam-
that for sufficiently chaotic dynamics T and rea- ily), billiards with convexscatterers (Young 1998),
sonably smooth observables f, the time averages the stadium billiard (Bálint and Gouëzel 2006),
Hénon-like mappings (Young and Wang 2008).
1
n
1 We note that in many cases one is led to first
An fðxÞ≔ f  T k ðxÞ analyze mixing properties through decay of cor-
n k¼0
relations, i.e., to prove inequalities of the type
(Baladi 2000a):
should behave as averages of independent and
identically distributed random variables and
therefore satisfy the classical limit theorems of f  c  T n dm
f dm c dm
probability theory. X X X

The dynamical systems which are amenable to f  kc kw  an ð5Þ

current technology are in a large part [But other
approaches are possible. Let us quote the work
where (an)n1 is some sequence converging to
(Dolgopyat 2004a) on partially hyperbolic sys-
zero, e.g., an ¼ e
ln, 1/nα, . . . and k  k, kkw a
tems, for instance.] those that can be reduced to
strong and a weak norm (e.g., the variation norm
the following type:
and the L1 norm). These rates of decay are often
linked with return times statistics (Young
Definition 1 Let T : X ! X be a nonsingular map
1999a). Rather general schemes have been
on a probability, metric space (X, ℬ, m, d) with
developed to deduce various limit theorems
bounded diameter, preserving the probability
such as those presented below from sufficiently
measure m. This map is said to be Gibbs–Markov
quick decay of correlations (see notably
if there exists a countable (measurable) partition α
(Liverani 1996) based on a dynamical variant
of X such that:
of (Gordin 1969)).
1. For all a  α, T is injective on a and T(a) is a
union of elements of α. Probabilistic Limit Theorems
2. There exists l > 1 such that, for all a  α, for The foremost limit property is the following:
all points x, y  a, d(Tx, Ty)  ld(x, y).
3. Let Jac be the inverse of the Jacobian of T. Definition 2 A class C of functions f : X ! ℝ is
There exists C > 0 such that, for all a  α, for said to satisfy the Central Limit Theorem if the
all points x, y  a, |1
Jac(x)/ following holds: for all f  C , there is a number
Jac( y)| Cd(Tx, Ty). s ¼ s(f) > 0 such that:
Chaos and Ergodic Theory 649

for some δ > 0. It holds with δ ¼ 1 in the classical,

An fðxÞ
f dm
probabilistic setting.
lim m xX : t
n!1 sn
1=2 The Local Limit Theorem looks at a finer scale,
asserting that for any finite interval [a, b], any
t
2 dx t  ℝ,
=2s2
¼ e
x p

1 2ps p p
lim nm fx  X : nAn fðxÞ  ½a, b þ nt
ð6Þ n!1
2 2
e
t =2s
þn f dmg ¼j b
a j p :
except for the degenerate case when 2ps
f(x) ¼ c(Tx)
f(x) þ const.
Both the Berry–Esseen inequality and the local
The Central Limit Theorem can be seen in many
limit theorem have been shown to hold for non--
cases as essentially a by-product of fast decay of
uniformly expanding maps (Gouëzel 2005) (also
correlations (Liverani 1996), i.e., if n0 an < 1
Broise 1996b; Rousseau-Egele 1983).
in the notations of Eq. (5). It has been established
for Hölder-continuous observables for many sys-
tems together with their natural invariant measures Almost Sure Results
including: uniformly hyperbolic attractors, piece- It is very natural to try and describe the statis-
wise expanding maps of the interval (Liverani tical properties of the averages An(x) for almost
1995b), Collet–Eckmann unimodal maps on the every x, instead of the weaker above statements
interval (Keller and Nowicki 1992; Young 1992), in probability over x. An important such prop-
piecewise hyperbolic maps (Chernov 1999), bil- erty is the almost sure invariance principle. It
liards with convex scatterers (Szász 2000), asks for the discrete random walk defined by
Hénon-like maps (Benedicks and Young 2000). the increments of f  T n(x) to converge, after a
suitable renormalization, to a Brownian motion.
Remark 5 The classical Central Limit Theorem This has been proved for systems with various
holds for square-integrable random variables degrees of hyperbolicity (Denker and Philipp
(Nagaev 1957). For maps exhibiting intermittency 1984; Dolgopyat 2004a; Hofbauer and Keller
(e.g, interval maps like f(x) ¼ x þ x1þα mod 1 1982; Melbourne and Nicol 2005). Another one
with an indifferent fixed point at 0) the invariant is the almost sure Central Limit Theorem. In
density has singularities and the integrability con- the independent case (e.g., if X1, X2, . . . are
dition is non longer automatic for smooth func- independent and identically distributed random
tions. One can then observe convergence to stable variables in L2 with zero average and unit var-
laws, instead of the normal law (Gouëzel 2004). iance), the almost sure Central Limit Theorem
states that, almost surely:
A natural question is the speed of the conver-
gence in (6). The Berry–Esseen inequality: n
1 1 p
d k
1
Xj = k
log n k j¼0
An fðxÞ
f dm t
k¼1
2
=2s2 dx
m t
e
x p
sn
1=2
1 2ps
converges in law to the normal distribution. This
C implies, that, almost surely, for any t  ℝ:
nd=2
650 Chaos and Ergodic Theory

1
n
1 htop ðf Þ ¼ sup hðf , mÞ
lim d p m  M ðf Þ
n!1 log n k k
1
X= k t
k¼1 j¼0 j

t where M( f ) is the set of all invariant probability

2
=2s2 dx
¼ e
x p :

1 2ps measures.
This is all the more striking in light of the fact
compare to (6). that for many systems, the set of points which are
A general approach is developed in (Chazottes typical from the point of view of ergodic
and Gouëzel 2007), covering Gibbs–Markov theory [for instance, those x such that
maps and those that can be reduced to it. They lim 1n n
1 k
k¼0 f T x exists for all continuous func-
n!1
show in particular that the dynamical properties tions f.] is topologically negligible [A subset of a
needed for the classical Central Limit Theorem in countable union of closed sets with empty interior,
fact suffice to prove the above almost invariance that is, meager or first Baire category.].
principle and even the almost sure version of the
Central Limit Theorem (using general probabilis- Strict Inequality
tic results, see (Berkes and Csáki 2001; Yoshihara For a fixed invariant measure, one can only assert
2004)). that h( f, m) htop( f ). One should be aware that
this inequality may be strict even for a measure
Other Statistical Properties with full support. For instance, it is not difficult to
Essentially all the statistical properties of sums check that the full tent map with htop( f ) ¼ log 2,
of independent identically distributed random admits ergodic invariant measures with full sup-
variables can be established for tractable sys- port and zero entropy.
tems. Thus one can also prove large deviations There are also examples of dynamical systems
(Kifer 1990; Liu et al. 2003; Young 1990), iter- preserving Lebesgue measure which have simul-
ated law of the logarithm, etc. We note that the taneously positive topological entropy and zero
monograph (Collet and Eckmann 2006) contains entropy with respect to Lebesgue measure. That
a nice introduction to the current work in this occurs for C1 surface diffeomorphisms pre-
this area. serving area is a simple consequence of a theorem
of Bochi (2002) according to which, generically
in the C1 topology, such a diffeomorphism is
either uniformly hyperbolic or with Lyapunov
Orbit Complexity
exponents Lebesgue-almost everywhere zero.
[Indeed, it is easy to build such a diffeomorphism
The orbit complexity of a dynamical system
having both a uniformly hyperbolic compact
f : M ! M is measured by its topological and
invariant subset which will have robustly positive
measured entropies. We refer to ▶ “Entropy in
topological entropy and a non-degenerate elliptic
Ergodic Theory” for detailed definitions.
fixed point which will prevent uniform hyper-
bolicity and therefore force all Lyapunov expo-
The Variational Principle nents to be zero. But Ruelle–Margulis inequality
Bowen–Dinaburg and Katok formulae can be then implies that the entropy with respect to
interpreted as meaning that the topological Lebesgue measure is zero.] Smooth examples
entropy counts the number of arbitrary orbits also exist (Bolsinov and Taimanov 2000).
whereas the measured entropy counts the number
of orbits relevant for the given measure. In most Remark 6 Algorithmic complexity (Li and
situations, and in particular for continuous self- Vitanyi 1997) suggests another way to look at
map of compact metric spaces, the following orbit complexity. One obtains in fact in this way
variational principle holds: another formula for the entropy. However this
Chaos and Ergodic Theory 651

point of view becomes interesting in some set- 1

htop ðf , ϵÞ≔n!1
lim log sðd, n, XÞ
tings, like extended systems defined by partial n
differential equations in unbounded space.
Recently, (Bonano and Collet 2006) has used where s(δ, n, E) is the maximum cardinality of a
this approach to build interesting invariant subset S of E such that:
measures.
x 6¼ y ) ∃0 k < n d f k x, f k y  ϵ
Orbit Complexity on the Set of Measures
We have considered the entropies of each invari-
ant measure separately, sharing only the common (see Bowen’s formula of the topological entropy
roof of topological entropy. One may ask how ▶ Entropy in Ergodic Theory). Likewise, the
these different complexity sit together. A first measure-theoretic entropy h(T, m) of an ergodic
answer is given by the following theorem in the invariant probability measure m is limϵ!0
symbolic and continuous setting. Let. h( f, m, ϵ) where:

Theorem 9 (Downarowicz–Serafin 2003) Let 1

hðT, ϵ≔n!1
lim log r ðd, n, mÞ
K be a Choquet simplex and H : K ! ℝ be a n
convex function.
Say that H is realized by a self-map f : X ! X where r(δ, n, m) is the minimum cardinality of
and its set M( f ) of f-invariant probability mea- C  X such that
sures equipped with the weak star topology if the
following holds. There exists an affine homeomor- mðf x  X : ∃y  C such that
phism C : M( f ) ! K such that, if h : M( f ) !
80 k < n d f k x, f k y < ϵgÞ > 1=2:
[0, 1] is the entropy function, H ¼ h  C.
Then
One can ask at which scales does entropy arise
for a given dynamical system?, i.e., how the above
• H is realized by some continuous self-map of a
quantities h(T, ϵ), h(T, m, ϵ) converge when
compact space if and only if it is an increasing
ϵ ! 0.
limit of upper semicontinuous and affine
An answer is provided by the local entropy.
functions.
[This quantity was introduced by Misiurewicz
• H is realized by some subshift on a finite alpha-
(1976b) under the name conditional topological
bet, i.e., by the left shift on a closed invariant
entropy and is called tail entropyby Downarowicz
subset S of {1, 2, . . ., N}ℤ for some N < 1, if
(2005).] For a continuous map f of a compact
and only if it is upper semi-continuous
metric space X, it is defined as:

hloc ðf Þ≔ lim hloc ðf , ϵÞ with

Thus, in both the symbolic and continuous ϵ!0

settings it is possible to have a unique invariant hloc ðf , ϵÞ≔sup hloc ðf , ϵ, xÞ and

xX
measure with any prescribed entropy. This stands 1
in contrast to surface C1þϵ-diffeomorphisms for hloc ðf , ϵ, xÞ≔ lim lim sup log sðd, n, fy  X :
d!0 n!1 n
which the set of the entropies of ergodic invariant
8k  0 d f k y, f k xÞ < ϵg
measures is always the interval [0, htop( f )] as a
consequence of (Katok 1980).
Clearly from the above formulas:
Local Complexity
htop ðf Þ htop ðf , dÞ þ hloc ðf , dÞ and
Recall that the topological entropy can be com-
puted as: htop ðf Þ ¼ lim htop ðf , ϵÞ where: hð f , m Þ hðf , m, dÞ þ hloc ðf , dÞ:
ϵ!0
652 Chaos and Ergodic Theory

Thus the local entropy bounds the defect in Global Simplicity

uniformity with respect to the measure of the One can marvel at the power of mathematical anal-
pointwise limit hðf , mÞ ¼ lim d!0 hðf , m, dÞ. An ysis to analyze such complex evolutions. Of course
exercise in topology shows that the local entropy another way to look at this is to remark that this
therefore also bounds the defect in upper semi- analysis is possible once this evolution has been
continuity of m 7! h( f, m). In fact, by a result of fitted in a simple setting: one had to move focus
Downarowicz (2005) (extended by David away from an individual, unpredictable orbit, of,
Burguet to the non-invertible case), there is a say, the full tent map to the set of all the orbits of
local variational principle: that map, which is essentially the set of all infinite
sequences over two symbols: a very simple set
hloc ðf Þ ¼ lim suphðf , mÞ
hðf , m, dÞ indeed corresponding to full combinatorial free-
d!0 m dom [(Weiss 2002) describes a weakening of this
¼ sup limsuphðf , nÞ
hðf , mÞ which holds for all positive entropy symbolic
m n!m
dynamics.]. The complete description of a given
typical orbit requires an infinite amount of infor-
for any continuous self-map f of a compact metric
mation, whereas the set of all orbits has a finite and
space.
The local entropy is easily bounded for smooth very tractable definition. The complexity of the
maps using Yomdin’s theory: individual orbits is seen now as coming from
purely random choices inside a simple structure.
Theorem 10 (Buzzi 1997) For any Cr map f of a The classical systems, namely uniformly
expanding maps or hyperbolic diffeomorphisms
compact manifold, hloc ðf Þ d
r log sup f 0 ðxÞ . In of compact spaces, have a simple symbolic
x
particular, hloc( f ) ¼ 0 if r ¼ 1. dynamics. It is not necessarily a full shift like for
the tent map, but it is a subshift of finite type, i.e., a
Thus C1 smoothness implies the existence of a subshift obtained from a full shift by forbidding
maximum entropy measure (this was proved first finitely many finite subwords. What happens out-
by Newhouse) and the existence of symbolic side of the uniform setting?
extension: a subshift over a finite alphabet A fundamental example is provided by piece-
s : S ! S and a continuous and onto map wise monotone maps, i.e., interval maps with
π : S ! M such that π  s ¼ f  s. More precisely, finitely many critical points or discontinuities.
The partition cut by these points defines a symbolic
Theorem 11 (Boyle, Fiebig, Fiebig (2002)) dynamics. This subshift is usually not of finite type.
Given a homeomorphism f of a compact metric Indeed, the topological entropy taking arbitrary
space X, there exists a principal symbolic exten- finite nonnegative values [For instance, the topo-
sion s : S ! S, i.e., a symbolic extension such logical entropy of the β-transformation,
that, for every s-invariant probability measure n, x 7! βx mod 1, is log β for β  1.], a representation
h(s, n) ¼ h( f, n  π
1), if and only if hloc( f ) ¼ 0. that respects it has to use an uncountable class of
models. In particular models defined by finite data,
We refer to (Boyle and Downarowicz 2004; like the subshifts of finite type, cannot be generally
Downarowicz and Newhouse 2005) for further adequate. However there are tractable “almost
results, including a realization theorem showing finite representations” in the following senses:
that the continuity properties of the measured Most symbolic dynamics S(T ) of piecewise
entropy are responsible for the properties of sym- monotone maps T can be defined by finitely
bolic extensions and also results in finite many infinite sequences, the kneading invariants
smoothness. of Milnor and Thurston: kþ
þ
0 , k1 , k1 , . . . ,
Chaos and Ergodic Theory 653

ℕ
k
dþ1  f0, . . . , d g if d is the number of critical/ Stability
discontinuity points. [The kneading invariants are
the suitably defined (left and right) itineraries of By definition, chaotic dynamical systems have
the critical/discontinuity points and endpoints.] orbits which are unstable and numerically
Namely, unpredictable. It is all the more surprising that,
once one accepts to consider their dynamics glob-
ally, they exhibit very good stability properties.
SðT Þ ¼ a  f0, . . . , dgℕ : 8n  0
Structural Stability
kþ

an ≺ s a ≺ kan þ1
n
A simple form of stability is structural stability: a
system f : M ! M is structurally Cr-stable if any
where ≺ is a total order on {0, . . ., d}ℕ making system g sufficiently Cr-close to f is topologically
the coding x 7! α non-decreasing. Observe how the same as f, formally: g is topologically conju-
the kneading invariants determine S(T ) in an gate, i.e., there is some homeomorphism [If h were
effective way: knowing their first n symbols is C1, the conjugacy would imply, among other
enough to know the sequences of length n which things, that for every p-periodic point: det(( f p)0
begin sequences of S(T ). We refer to (de Melo and (x)) ¼ det ((g p)0(h( p))), a much too strong
van Strien 1993) for the wealth of information that requirement.] h : M ! M mapping the orbits of
can be extracted from these kneading invariants f to those of g, i.e., g  h ¼ h  f.
following Milnor and Thurston (1988). Andronov and Pontryaguin argued in the
This form of global simplicity can be extended 1930’s that only such structurally stable systems
to other classes of non-uniformly expanding are physically relevant. Their idea was that the
maps, including those like Eq. (4) using the model of a physical system is always known
notions of subshifts and puzzles of quasi-finite only to some degree of approximation, hence
type (Buzzi 2005, 2006). This leads to the notion mathematical model whose structure depends on
and analysis of entropy-expanding maps, a new arbitrarily small changes should be irrelevant.
open class of non-uniformly expanding maps A first question is: What are these structurally
admitting critical hypersurfaces, defined purely stable systems? The answer is quite striking:
in terms of entropies including the otherwise
untractable examples of Eq. (4). Theorem 12 (Mañé 1988) Let f : M ! M be a C1
A generalization of the representation of uni- diffeomorphism of a compact manifold.
form systems by subshifts of finite type is pro- f is structurally stable among C1-
vided by strongly positive recurrent countable diffeomorphisms of M if and only if f is uniformly
state Markov shifts, a subclass of Markov shifts hyperbolic on its chain recurrent set. [A point x is
(See “Glossary”) which shares many properties chain recurrent if, for all ϵ > 0, there exists a finite
with the subshifts of finite type (Boyle et al. sequence x0, x1, . . ., xn such that x0 ¼ xn ¼ x and
2006; Gurevic 1996; Gurevic and Savchenko d( f(xk), xkþ1) < ϵ. The chain recurrent set is the
1998; Ruette 2003a; Sarig 2001). set of all chain recurrent points.]
These “simple” systems admit a classification
result which in particular identifies their measures A basic idea in the proof of the theorem is that
with entropy close to the maximum (Boyle et al. failure of uniform hyperbolicity gives the oppor-
2006). Such a classification generalizes (Ladler tunity to make an arbitrarily small perturbation
and Marcus 1979). The “ideology” here is that contradicting the structural stability. In higher
complexity of individual orbits in a simple setting smoothness the required perturbation lemmas
must come from randomness, but purely random (e.g., the closing lemma (Arnaud 1998; Katok
systems are classified by their entropy according and Hasselblatt 1995; Mañé 1982)) are not
to Ornstein (1970). available.
654 Chaos and Ergodic Theory

We note that uniform hyperbolicity without which topology should be put on the rather wild
invertibility does not imply C1-stability set of topological conjugacy classes. It is perhaps
(Przytycki 1977). more natural to associate to the system a topolog-
A second question is: are these stable systems ical invariant taking value in a more manageable
dense? (So that one could offer structurally stable set and ask whether the resulting map is
models for all physical situations). A deep discov- continuous.
ery around 1970 is that this is not the case: A first possibility is Zeeman’s Tolerance Sta-
bility Conjecture. He associated to each
Theorem 13 (Abraham–Smale, Simon diffeomorphism the set of all the closures of all
(Abraham and Smale 1970; Simon 1972)) of its orbits and he asked whether the resulting
For any r  1 and any compact manifold M of map is continuous on a dense Gδ subset of the
dimension 3, the set of uniformly hyperbolic class of Cr-diffeomorphisms for any r  0. This
diffeomorphisms is not dense in the space of C r conjecture remains open, we refer to (Crovisier
diffeomorphisms of M. [They use the phenomenon 2006b) for a discussion and related progress.
called “heterodimensional homoclinic A simpler possibility is to consider our favorite
intersections”.] topological invariant, the topological entropy, and
thus ask whether the dynamical complexity as
Theorem 14 (Newhouse 1974) For any r  2, measured by the entropy is a stable phenomenon.
for any compact manifold M of dimension 2, the f 7! htop( f ) is lower semicontinuous for f among
set of uniformly hyperbolic diffeomorphisms is not C0 maps of the interval [On the set of interval
dense in the space of C r diffeomorphisms of M. maps with a bounded number of critical points,
More precisely, there exists a non-empty open the entropy is continuous (Misiurewicz 1995).
subset in this space which contains a dense Gδ Also t 7! htop(Qt) is non-decreasing by complex
subset of diffeomorphisms with infinitely many arguments (de Melo and van Strien 1993), though
periodic sinks. [So these diffeomorphisms have it is a non-smooth function.] (Misiurewicz 1979)
no finite statistical description.] and for f among C1þϵ-diffeomorphisms of a com-
pact surface (Katok 1980). [It is an important
Observe that it is possible that uniform hyper- open question whether this actually holds for
bolicity could be dense among surface C1- C1-diffeomorphisms. It fails for homeomorphisms
diffeomorphisms (this is the case for C1 circle (Rees 1981).] In both cases, one shows the exis-
maps by a theorem of Jakobson (1981)). tence of structurally stable invariant uniformly
In light of Mañé C1-stability theorem this expanding or hyperbolic subsets with topological
implies that structurally stable systems are not entropy close to that of the whole dynamics. On
dense, thus one can robustly see behaviors that the other hand f 7! htop( f ) is upper semi-
are topologically modified by arbitrarily small continuous for C1 maps (Newhouse 1989;
perturbations (at least in the C1-topology)! So Yomdin 1987).
one needs to look beyond these and face that
topological properties of relevant dynamical sys- Statistical Stability
tem are not determined from “finite data”. It is Statistical stability is the property that determin-
natural to ask whether the dynamics is almost istic perturbations of the dynamical system cause
determined by “sufficient data”. only small changes in the physical measures, usu-
ally with respect to the weak star topology on the
Continuity Properties of the Topological space of measures. When the physical measure mg
Dynamics is uniquely defined for all systems g near f, statis-
Structural stability asks the topological dynamics tic stability is the continuity of the map g 7! mg
to remain unchanged by a small perturbation. It is thus defined.
probably at least as interesting to ask it to change Statistical stability is known in the uniform
continuously. This raises the delicate question of setting and also in the piecewise uniform case,
Chaos and Ergodic Theory 655

provided the expansion is strong enough (Baladi y
f ðx Þ dy

Pϵ ðx, AÞ ¼ c :
and Young 1993; Blank and Keller 1997; Keller A ϵ ϵd
and Liverani 1999) (otherwise there are counter-
examples, even in the one-dimensional case). The evolution of measures is given by:
It also holds in some non-uniform settings ( fϵm)(A) ¼ MPϵ(x, A) dm. Under rather weak
without critical behavior, in particular for maps irreducibility assumptions on f, fϵ has a unique
with dominated splitting satisfying robustly a sep- invariant measure mϵ (contrarily to f ) and mϵ is
aration condition between the positive and nega- absolutely continuous. When f has a unique phys-
tive Lyapunov exponents (Vasquez 2007) (see ical measure m, it is said to be stochastically stable
also (Alves and Viana 2002)). if limϵ!0mϵ ¼ m in the appropriate topology (the
However statistical unstability occurs in weak star topology unless otherwise specified).
dynamics with critical behaviors (Avila, 2007, It turns out that stochastic stability is a rather
personal communication): common property of Sinai–Ruelle–Bowen mea-
sures. It holds not only for uniformly expanding
Theorem 15 Consider the quadratic family maps or hyperbolic diffeomorphisms (Kifer 1977;
Qt(x) ¼ tx(1
x), t  [0, 4]. Let I  [0, 4] be Young 1986), but also for most interval maps
the full measure subset of [0, 4] of good parame- (Baladi and Young 1993), for partially hyperbolic
ters t such that in particular Qt admits a physical systems of the type Eu Ecs or Hénon-like
measure (necessarily unique). For t  I, let mt be diffeomorphisms (Alves and Araujo 2003; Araujo
this physical measure. and Tahzibi 2005; Benedicks and Viana 2006;
Blank 1989). We refer to the monographs
Lebesgue almost every t  I such that mt is not (Baladi 2000a; Blank 1997) for more background
carried by a periodic orbit [At such parameters, and results.
statistical stability is easily proved.], is a discon-
tinuity point of the map t 7! mt [M([0, 1]) is
equipped with the vague topology.]. However, Untreated Topics
for Lebesgue almost every t, there is a subset It  I
for which t is a Lebesgue density point [t is a For reasons of space and time, many important
Lebesgue density point if for all r > 0, the topics have been left out of this article. Let us list
Lebesgue measure m(It \ [t
r, t þ r]) > 0 and some of them.
lim mðI t \ ½t
ϵ, t þ ϵ Þ=2ϵ ¼ 1.] and such that Other phenomena related to chaotic dynamics
ϵ!0
have been studied: entrance times (Collet 1996;
m : It ! M([0, 1]) is continuous at t.
Collet and Galves 1995), spectral properties and
dynamical zeta functions (Baladi 2000a), escape
Stochastic Stability rates (Eckmann and Ruelle 1985), dimension
A physically motivated and a technically easier (Pesin 1997), differentiability of physical mea-
approach is to study stability of the physical mea- sures with respect to parameters (Dolgopyat
sure under stochastic perturbations. For simplic- 2004b; Ruelle 2005), entropies and volume or
ity let us consider a diffeomorphism of a compact homological growth rates (Gromov 1987; Hua
subset of ℝd allowing for a direct definition of et al. 2006; Yomdin 1987).
additive noise. Let c(x)dx be an absolutely con- As far as the structure of the setting is
tinuous probability law with compact support. concerned, one can go beyond maps or
[Sometimes additional properties are required of diffeomorphisms or flows and study: more gen-
the density, e.g., c(x) ¼ f(x)1B where 1B is the eral group actions (Hasselblatt and Katok 2002);
characteristic function of the unit ball and holomorphic and meromorphic structures
C
1 f(x) C for some C < 1.] For ϵ > 0, (Carleson and Gamelin 1993) and the references
consider the Markov chain fϵ with state space therein; symplectic or volume-preserving (de la
M and transition probabilities: Llave 2001; Hofer and Zehnder 1994) and in
656 Chaos and Ergodic Theory

particular the Pugh–Shub program around stable background) seems restricted to the C1 topology
ergodicity of partially hyperbolic systems (Pugh because of the lack of fundamental tools (e.g.,
and Shub 1999); random iterations (Arnold 1998; closing lemmas) in higher smoothness. But Pesin
Kifer 1986, 1988). theory requires higher smoothness at least techni-
A number of important problems have moti- cally. This is not only a hard technical issue but
vated the study of special forms of chaotic dynam- generic properties of physical measures, when
ics: equidistribution in number theory (Eskin and they have been analyzed are often completely
McMullen 1993; Furstenberg 1981; Host and Kra different between the C1 case and higher smooth-
2005) and geometry (Starkov 2000); quantum ness (Bochi and Viana 2005).
chaos (Anantharaman and Nonnenmacher 2007;
Gutzwiller 1990); chaotic control (Saperstone and Physical Measures
Yorke 1971); analysis of algorithms (Vallée 2006). In higher dimensions, Benedicks and Carleson
We have also omitted the important problem of analysis of the Hénon map has given rise to a
applying the above results. Perhaps because of the rather general theory of Hénon-like maps and
lack of a general theory, this can often be a chal- more generally of the dynamical phenomena asso-
lenge (see for instance (Tucker 1999) for the ciated to homoclinic tangencies. However, the
already complex problem of verifying uniform proofs are extremely technical. Could they be
hyperbolicity for a singular flow). Liverani has simplified? Current attempts like (Young and
shown how theoretical results can lead to precise Wang 2008) center on the introduction of a sim-
and efficient estimates for the toy model of piece- pler notion of critical points, possibly a non-
wise expanding interval maps (Liverani 2001). inductive one (Pujals and Rodriguez-Hertz 2007).
Ergodic theory implies that, in some settings at Can this Hénon theory be extended to the
least, adding noise may make some estimates weakly dissipative situation? to the conservative
more precise (see Kifer 1997). We refer to situation (for which the standard map is a well-
(Fiedler 2001) and the reference therein. known example defying analysis)? In the strongly
dissipative setting, what are the typical phenom-
ena on the complement of the Benedicks–
Future Directions Carleson set of parameters?
From a global perspective one of the main
We conclude this article by a (very partial) selec- questions is the following:
tion of open problems. Can infinitely many sinks coexist for a large set of
parameters in a typical family or is Newhouse phe-
General Theory nomenon atypical in the Kolmogorov or prevalent
In dimension 1, we have seen that the analogue of sense?
the Palis conjecture (see above) is established This seems rather unlikely (see however
(Theorem 2). However the description of the typ- (Araujo 2001)).
ical dynamics in Kolmogorov sense is only Away from such “critical dynamics”, there are
known in the unimodal, non-degenerate case by many results about systems with dominated split-
Theorem 3. Indeed, results like (Bruin et al. 1996) ting satisfying additional conditions. Can these
suggest that the multimodal picture could be more conditions be weakened so they would be satisfied
complex. by typical systems satisfying some natural condi-
In higher dimensions, our understanding is tions (like robust transitivity)? For instance:
much more limited. As far as a general theory is
Could one analyze the physical measures of
concerned, a deep problem is the paucity of results volume-hyperbolic systems?
on the generic dynamics in Cr smoothness with
r > 1. The remarkable current progress in generic A more specific question is whether Tsujii’s
dynamics (culminating in the proof of the weak striking analysis of surface maps with one uni-
Palis conjecture, see (Crovisier 2006a), the refer- formly expanding direction can be extended to
ences therein and (Bonatti et al. 2005) for higher dimensions? can one weaken the
Chaos and Ergodic Theory 657

uniformity of the expansion? The same questions expansion condition [For instance building on
for the corresponding invertible situation is con- our “entropy-hyperbolicity”.] of (Buzzi 2000b,
sidered in (Cowieson and Young 2005). Buzzi), that would be satisfied by a large subset
of the diffeomorphisms?
Maximum Entropy Measures and Topological Another possible approach is illustrated by the
Complexity pruning front conjecture of (Cvitanović et al.
As we explained, C1 smoothness, by a Newhouse 1988) (see also (de Carvalho and Hall 2002;
theorem, ensures the existence of maximum Ishii 1997)). It is an attempt to build a combina-
entropy measures, making the situation a little torial description by trying to generalize the way
simpler than with respect to physical measures. that, for interval maps, kneading invariants deter-
This existence results allow in particular an easy mine the symbolic dynamics by considering the
formulation of the problem of the typicality of bifurcations from a trivial dynamics to an
hyperbolicity: arbitrary one.
Are maximum entropy ergodic measures of systems We hope that our reader has shared in our
with positive entropy hyperbolic for most systems? fascination with this subject, the many surprising
and even paradoxical discoveries that have been
A more difficult problem is that of the finite
made and the exciting current progress, despite
multiplicity of the maximum entropy measures.
the very real difficulties both in the analysis of
For instance: such non-uniform systems as the Henon map and
Do typical systems possess finitely many maximum in the attemps to establish a general (and practi-
entropy ergodic measures? cal) ergodic theory of chaotic dynamical
More specifically, can one prove intrinsic ergo- systems.
dicity (i.e, uniqueness of the measure of maxi-
mum entropy) for an isolated homoclinic class of Acknowledgments I am grateful for the advice and/or
comments of the following colleagues: P. Collet, J.-R.
some diffeomorphisms (perhaps C1-generic)?
Chazottes, and especially S. Ruette. I am also indebted to
Can a generic C1-diffeomorphism carry an infinite the anonymous referee.
number of homoclinic classes, each with topolog-
ical entropy bounded away from zero?
A perhaps more tractable question, given the Bibliography
recent progress in this area: Is a C1-generic par-
tially hyperbolic diffeomorphisms, perhaps with Primary Literature
central dimension 1 or 2, intrinsically ergodic? Aaronson J, Denker M (2001) Local limit theorems for
We have seen how uniform systems have sim- partial sums of stationary sequences generated by
Gibbs–Markov maps. Stochastics Dyn 1:193–237
ple symbolic dynamics, i.e., subshifts of finite
Abdenur F, Bonatti C, Crovisier S (2006) Global domi-
type, and how interval maps and more generally nated splittings and the C1 newhouse phenomenon.
entropy-expanding maps keep some of this sim- Proc Am Math Soc 134(8):2229–2237
plicity, defining subshifts or puzzles of quasi- Abraham R, Smale S (1970) Nongenericity of Ω-stability.
In: Global analysis, vol XIV. Proc Sympos Pure Math.
finite type (Buzzi 2005; Buzzi 2006). (Young
American Mathematical Society, Providence
and Wang 2001) have defined symbolic dynamics Alves JF (2000) SRB measures for non-hyperbolic systems
for topological Hénon-like map which seems with multidimensional expansion. Ann Sci Ecole Norm
close to that of that of a one-dimensional system. Sup 33(4):1–32
Alves JF (2006) A survey of recent results on some statis-
Can one describe Wang and Young symbolic tical features of non-uniformly expanding maps. Dis-
dynamics of Hénon-like attractors and fit it in a crete Contin Dyn Syst 15:1–20
class in which uniqueness of the maximum entropy Alves JF, Araujo V (2003) Random perturbations of non-
measure could be proved? uniformly expanding maps. Geometric methods in
dynamics. I. Astérisque 286:25–62
More generally, can one define nice combina- Alves JF, Viana M (2002) Statistical stability for robust
torial descriptions, for surface diffeomorphisms? classes of maps with non-uniform expansion. Ergodic
Can one formulate variants of the entropy- Theory Dynam Syst 22:1–32
658 Chaos and Ergodic Theory

Alves JF, Bonatti C, Viana M (2000) SRB measures for Barreira L, Pesin Y, Schmeling J (1999) Dimension and
partially hyperbolic systems whose central direction is product structure of hyperbolic measures. Ann Math
mostly expanding. Invent Math 140:351–398 149(2):755–783
Anantharaman N (2004) On the zero-temperature or Benedicks M, Carleson L (1985) On iterations of 1
ax2 on
vanishing viscosity limit for certain Markov processes (
1,1). Ann Math (2) 122(1):1–25
arising from Lagrangian dynamics. J Eur Math Soc Benedicks M, Carleson L (1991) The dynamics of the
(JEMS) 6:207–276 Hénon map. Ann Math 133(2):73–169
Anantharaman N, Nonnenmacher S (2007) Half delocali- Benedicks M, Viana M (2001) Solution of the basin problem
zation of the eigenfunctions for the Laplacian on an for Hénon-like attractors. Invent Math 143:375–434
Anosov manifold. Ann Inst Fourier 57:2465–2523 Benedicks M, Viana M (2006) Random perturbations and
Araujo V (2001) Infinitely many stochastically stable statistical properties of Hénon-like maps. Ann Inst
attractors. Nonlinearity 14:583–596 H Poincaré Anal Non Linéaire 23:713–752
Araujo V, Pacifico MJ (2006) Large deviations for non- Benedicks M, Young LS (1993) Sinaĭ–Bowen–Ruelle
uniformly expanding maps. J Stat Phys 125:415–457 measures for certain Hénon maps. Invent Math 112:
Araujo V, Tahzibi A (2005) Stochastic stability at the 541–576
boundary of expanding maps. Nonlinearity 18: Benedicks M, Young LS (2000) Markov extensions and
939–958 decay of correlations for certain Hénon maps.
Arnaud MC (1998) Le “closing lemma” en topologie C1. Géométrie complexe et systèmes dynamiques, Orsay,
Mem Soc Math Fr (NS) 74 1995. Astérisque 261:13–56
Arnol’d VI (1988) Geometrical methods in the theory of Bergelson V (2006) Ergodic Ramsey theory: a dynamical
ordinary differential equations. Grundlehren der approach to static theorems. In: International congress
Mathematischen Wissenschaften, vol 250, 2nd edn. of mathematicians, vol II. European Mathematical
Springer, New York Society, Zürich, pp 1655–1678
Arnol’d VI, Avez A (1968) Ergodic problems of classical Berkes I, Csáki E (2001) A universal result in almost sure
mechanics. W.A. Benjamin, New York central limit theory. Stochastic Process Appl 94(1):
Arnold L (1998) Random dynamical systems. Springer 105–134
monographs in mathematics. Springer, Berlin Bishop E (1967/1968) A constructive ergodic theorem.
Avila A, Moreira CG (2005) Statistical properties of J Math Mech 17:631–639
unimodal maps: the quadratic family. Ann Math Blanchard F, Glasner E, Kolyada S, Maass A (2002) On
161(2):831–881 Li-Yorke pairs, English summary. J Reine Angew Math
Avila A, Lyubich M, de Melo W (2003) Regular or sto- 547:51–68
chastic dynamics in real analytic families of unimodal Blank M (1989) Small perturbations of chaotic dynamical
maps. Invent Math 154:451–550 systems. (Russian). Uspekhi Mat Nauk 44:3–28, 203.
Baladi V (2000a) Positive transfer operators and decay of Translation in: Russ Math Surv 44:1–33
correlations. In: Advanced series in nonlinear dynam- Blank M (1997) Discreteness and continuity in problems of
ics, vol 16. World Scientific Publishing, River Edge chaotic dynamics. In: Translations of mathematical
Baladi V, Gouëzel S (2008) Good Banach spaces for piece- monographs, vol 161. American Mathematical Society,
wise hyperbolic maps via interpolation. preprint Providence
arXiv:0711.1960. Available from http://www.arxiv.org Blank M, Keller G (1997) Stochastic stability versus local-
Baladi V, Ruelle D (1996) Sharp determinants. Invent ization in one-dimensional chaotic dynamical systems.
Math 123:553–574 Nonlinearity 10:81–107
Baladi V, Tsujii M (2007) Anisotropic Hölder and Sobolev Blank M, Keller G, Liverani C (2002) Ruelle–Perron–
spaces for hyperbolic diffeomorphisms. Ann Inst Fou- Frobenius spectrum for Anosov maps. Nonlinearity
rier (Grenoble) 57:127–154 15:1905–1973
Baladi V, Young LS (1993) On the spectra of randomly Bochi J (2002) Genericity of zero Lyapunov exponents.
perturbed expanding maps. Commun Math Phys 156: Ergodic Theory Dynam Syst 22:1667–1696
355–385. Erratum: Comm Math Phys 166(1): Bochi J, Viana M (2005) The Lyapunov exponents of
219–220, 1994 generic volume-preserving and symplectic maps. Ann
Baladi V, Kondah A, Schmitt B (1996) Random correla- Math 161(2):1423–1485
tions for small perturbations of expanding maps, Bolsinov AV, Taimanov IA (2000) Integrable geodesic
English summary. Random Comput Dynam 4:179–204 flows with positive topological entropy. Invent Math
Bálint P, Gouëzel S (2006) Limit theorems in the stadium 140:639–650
billiard. Commun Math Phys 263:461–512 Bonano C, Collet P (2006) Complexity for extended
Baraviera AT, Bonatti C (2003) Removing zero Lyapunov dynamical systems. arXiv:math/0609681
exponents, English summary. Ergodic Theory Dynam Bonatti C, Crovisier S (2004) Recurrence et genericite.
Syst 23:1655–1670 Invent Math 158(1):33–104
Barreira L, Schmeling J (2000) Sets of “non-typical” points Bonatti C, Viana M (2000) SRB measures for partially
have full topological entropy and full Hausdorff dimen- hyperbolic systems whose central direction is mostly
sion. Israel J Math 116:29–70 contracting. Israel J Math 115:157–193
Chaos and Ergodic Theory 659

Bonatti C, Viana M (2004) Lyapunov exponents with Carleson L, Gamelin TW (1993) Complex dynamics. In:
multiplicity 1 for deterministic products of matrices. Universitext: tracts in mathematics. Springer,
Ergodic Theory Dynam Syst 24(5):1295–1330 New York
Bonatti C, Diaz L, Viana M (2005) Dynamics beyond Chazottes JR, Gouëzel S (2007) On almost-sure versions
uniform hyperbolicity. A global geometric and proba- of classical limit theorems for dynamical systems. Pro-
bilistic perspective. In: Mathematical physics, III. bab Theory Relat Fields 138:195–234
Encyclopedia of mathematical sciences, vol 102. Chernov N (1999) Statistical properties of piecewise
Springer, Berlin smooth hyperbolic systems in high dimensions. Dis-
Bowen R (1975) Equilibrium states and the ergodic theory crete Contin Dyn Syst 5(2):425–448
of Anosov diffeomorphisms. In: Lecture notes in math- Chernov N, Markarian R, Troubetzkoy S (2000) Invariant
ematics, vol 470. Springer, Berlin/New York measures for Anosov maps with small holes. Ergodic
Bowen R (1979) Hausdorff dimension of quasi-circles. Inst Theory Dynam Syst 20:1007–1044
Hautes Études Sci Publ Math 50:11–25 Christensen JRR (1972) On sets of Haar measure zero in
Bowen R, Ruelle D (1975) Ergodic theory of Axiom abelian Polish groups. Israel J Math 13:255–260
A flows. Invent Math 29:181–202 Collet P (1996) Some ergodic properties of maps of the
Boyle M, Downarowicz T (2004) The entropy theory of interval. In: Dynamical systems, Temuco, 1991/1992.
symbolic extensions. Invent Math 156:119–161 Travaux en Cours, 52. Hermann, Paris, pp 55–91
Boyle M, Fiebig D, Fiebig U (2002) Residual entropy, Collet P, Eckmann JP (2006) Concepts and results in cha-
conditional entropy and subshift covers. Forum Math otic dynamics: a short course. In: Theoretical and math-
14:713–757 ematical physics. Springer, Berlin
Boyle M, Buzzi J, Gomez R (2006) Ricardo almost iso- Collet P, Galves A (1995) Asymptotic distribution of
morphism for countable state Markov shifts. J Reine entrance times for expanding maps of the interval. In:
Angew Math 592:23–47 Dynamical systems and applications. World Sci Ser
Brain M, Berger A (2001) Chaos and chance. In: An Appl Anal, vol 4. World Scientific Publishing, River
introduction to stochastic aspects of dynamics. Walter Edge, pp 139–152
de Gruyter, Berlin Collet P, Courbage M, Mertens S, Neishtadt A, Zaslavsky
Brin M (2001) Appendix A. In: Barreira L, Pesin Y (eds) G (eds) (2005) Chaotic dynamics and transport in clas-
Lectures on Lyapunov exponents and smooth ergodic sical and quantum systems. In: Proceedings of the
theory. Proc Sympos Pure Math, 69, Smooth ergodic International Summer School of the NATO Advanced
theory and its applications, Seattle, WA, 1999. Ameri- Study Institute held in Cargese, August 18–30, 2003.
can Mathematical Society, Providence, pp 3–106 Edited by NATO science series II: mathematics, phys-
Brin M, Stuck G (2002) Introduction to dynamical sys- ics and chemistry, 182. Kluwer, Dordrecht
tems. Cambridge University Press, Cambridge Contreras G, Lopes AO, Thieullen P (2001) Lyapunov
Broise A (1996a) Transformations dilatantes de l’intervalle minimizing measures for expanding maps of the circle,
et theoremes limites. Etudes spectrales d’operateurs de English summary. Ergodic Theory Dynam Syst 21:
transfert et applications. Asterisque 238:1–109 1379–1409
Broise A (1996b) Transformations dilatantes de l’intervalle Cowieson WJ (2002) Absolutely continuous invariant
et théorèmes limites. Astérisque 238:1–109 measures for most piecewise smooth expanding maps.
Bruin H, Keller G, Nowicki T, van Strien S (1996) Wild Ergodic Theory Dynam Syst 22:1061–1078
Cantor attractors exist. Ann Math 143(2):97–130 Cowieson WJ, Young LS (2005) SRB measures as zero-
Buzzi J (1997) Intrinsic ergodicity of smooth interval noise limits. Ergodic Theory Dynam Syst 25:1115–1138
maps. Israel J Math 100:125–161 Crovisier S (2006a) Birth of homoclinic intersections: a
Buzzi J (2000a) Absolutely continuous invariant probabil- model for the central dynamics of partially hyperbolic
ity measures for arbitrary expanding piecewise systems. http://www.arxiv.org/abs/math/0605387
R-analytic mappings of the plane. Ergodic Theory Crovisier S (2006b) Periodic orbits and chain-transitive
Dynam Syst 20:697–708 sets of C1-diffeomorphisms. Publ Math Inst Hautes
Buzzi J (2000b) On entropy-expanding maps. Unpublished Etudes Sci 104:87–141
Buzzi J (2001) No or infinitely many a.c.i.p. for piecewise Crovisier S (2006c) Perturbation of C1-diffeomorphisms
expanding Cr maps in higher dimensions. Commun and generic conservative dynamics on surfaces.
Math Phys 222:495–501 Dynamique des diffeomorphismes conservatifs des sur-
Buzzi J (2005) Subshifts of quasi-finite type. Invent Math faces: un point de vue topologique, 21, Panor synthe-
159:369–406 ses. Soc Math France, Paris, pp 1–33
Buzzi J (2006) Puzzles of quasi-finite type, zeta functions Cvitanović P, Gunaratne GH, Procaccia I (1988) Topolog-
and symbolic dynamics for multi-dimensional maps. ical and metric properties of Henon-type strange
arXiv:math/0610911 attractors. Phys Rev A 38(3):1503–1520
Buzzi J. Hyperbolicity through Entropies. Saint-Flour de Carvalho A, Hall T (2002) How to prune a horseshoe.
Summer probability school lecture notes (to appear) Nonlinearity 15:R19–R68
660 Chaos and Ergodic Theory

de Castro A (2002) Backward inducing and exponential Gatzouras D, Peres Y (1997) Invariant measures of full
decay of correlations for partially hyperbolic attractors. dimension for some expanding maps. Ergodic Theory
Israel J Math 130:29–75 Dynam Syst 17:147–167
de la Llave R (2001) A tutorial on KAM theory. In: Smooth Glasner E, Weiss B (1993) Sensitive dependence on initial
ergodic theory and its applications, Seattle, WA, 1999. conditions. Nonlinearity 6:1067–1075
Proc Sympos Pure Math, vol 69. American Mathemat- Gordin MI (1969) The central limit theorem for stationary
ical Society, Providence, pp 175–292 processes. (Russian). Dokl Akad Nauk SSSR 188:
de Melo W, van Strien S (1993) One-dimensional dynam- 739–741
ics. In: Ergebnisse der Mathematik und ihrer Gouëzel S (2004) Central limit theorem and stable laws for
Grenzgebiete (3), 25. Springer, Berlin intermittent maps. Probab Theory Relat Fields 128:
Denker M, Philipp W (1984) Approximation by Brownian 82–122
motion for Gibbs measures and flows under a function. Gouëzel S (2005) Berry–Esseen theorem and local limit
Ergodic Theory Dynam Syst 4:541–552 theorem for non uniformly expanding maps. Ann Inst
Detmers MF, Liverani C (2007) Stability of statistical H Poincaré 41:997–1024
properties in two-dimensional piecewise hyperbolic Gouëzel S, Liverani C (2006) Banach spaces adapted to
maps. Trans Am Math Soc 360:4777–4814 Anosov systems. Ergodic Theory Dynam Syst 26:
Devaney RL (1989) An introduction to chaotic dynamical 189–217
systems, 2nd edn. Addison–Wesley, Redwood Graczyk J, Świątek G (1997) Generic hyperbolicity in the
Diaz L, Rocha J, Viana M (1996) Strange attractors in logistic family. Ann Math 146(2):1–52
saddle-node cycles: prevalence and globality. Invent Gromov M (1987) Entropy, homology and semialgebraic
Math 125:37–74 geometry. Séminaire Bourbaki, vol 1985/86.
Dolgopyat D (2000) On dynamics of mostly contracting Astérisque No 145–146:225–240
diffeomorphisms. Commun Math Phys 213:181–201 Guckenheimer J (1979) Sensitive dependence on initial
Dolgopyat D (2002) On mixing properties of compact conditions for one-dimensional maps. Commun Math
group extensions of hyperbolic systems. Israel J Math Phys 70:133–160
130:157–205 Guckenheimer J, Holmes P (1990) Nonlinear oscillations,
Dolgopyat D (2004a) Limit theorems for partially hyper- dynamical systems, and bifurcations of vector fields.
bolic systems. Trans Am Math Soc 356:1637–1689 Revised and corrected reprint of the 1983 original. In:
Dolgopyat D (2004b) On differentiability of SRB states for Applied mathematical sciences, vol 42. Springer,
partially hyperbolic systems. Invent Math 155: New York
389–449 Guckenheimer J, Wechselberger M, Young LS (2006) Cha-
Downarowicz T (2005) Entropy structure. J Anal Math 96: otic attractors of relaxation oscillators. Nonlinearity 19:
57–116 701–720
Downarowicz T, Newhouse S (2005) Symbolic extensions Guivarc’h Y, Hardy J (1998) Théorèmes limites pour une
and smooth dynamical systems. Invent Math 160: classe de chaînes de Markov et applications aux
453–499 difféomorphismes d’Anosov. Ann Inst H Poincaré 24:
Downarowicz T, Serafin J (2003) Possible entropy func- 73–98
tions. Israel J Math 135:221–250 Gurevic BM (1996) Stably recurrent nonnegative matrices.
Duhem P (1906) La théorie physique, son objet et sa (Russian) Uspekhi Mat Nauk 51(3)(309):195–196;
structure. Vrin, Paris, 1981 translation in: Russ Math Surv 51(3):551–552
Eckmann JP, Ruelle D (1985) Ergodic theory of chaos and Gurevic BM, Savchenko SV (1998) Thermodynamic for-
strange attractors. Rev Mod Phys 57:617–656 malism for symbolic Markov chains with a countable
Eskin A, McMullen C (1993) Mixing, counting, and number of states. (Russian). Uspekhi Mat Nauk
equidistribution in Lie groups. Duke Math J 71: 53(2)(320):3–106; translation in: Russ Math Surv 53(2):
181–209 245–344
Fiedler B (ed) (2001) Ergodic theory, analysis, and efficient Gutzwiller M (1990) Chaos in classical and quantum
simulation of dynamical systems. Springer, Berlin mechanics. In: Interdisciplinary applied mathematics,
Fiedler B (ed) (2002) Handbook of dynamical systems, vol 1. Springer, New York
vol 2. North-Holland, Amsterdam Hadamard J (1898) Les surfaces à courbures opposees et
Fisher A, Lopes AO (2001) Exact bounds for the polyno- leurs lignes geodesiques. J Math Pures Appl 4:27–73
mial decay of correlation, 1/f noise and the CLT for the Hasselblatt B, Katok A (eds) (2002, 2006) Handbook of
equilibrium state of a non-Hölder potential. Non- dynamical systems, vol 1A and 1B. Elsevier,
linearity 14:1071–1104 Amsterdam
Furstenberg H (1981) Recurrence in ergodic theory and Hasselblatt B, Katok A (2003) A first course in dynamics.
combinatorial number theory. In: MB Porter lectures. In: With a panorama of recent developments. Cam-
Princeton University Press, Princeton bridge University Press, New York
Gallavotti G (1999) Statistical mechanics. In: A short Hennion H (1993) Sur un théorème spectral et son appli-
treatise. Texts and monographs in physics. Springer, cation aux noyaux lipchitziens. Proc Am Math Soc 118:
Berlin 627–634
Chaos and Ergodic Theory 661

Hénon M (1976) A two-dimensional mapping with a Keller G, Liverani C (1999) Stability of the spectrum for
strange attractor. Commun Math Phys 50:69–77 transfer operators. Ann Scuola Norm Sup Pisa Cl Sci
Hesiod (1987) Theogony. Focus/R. Pullins, Newburyport 28(4):141–152
Hochman M (2006) Upcrossing inequalities for stationary Keller G, Nowicki T (1992) Spectral theory, zeta functions
sequences and applications. arXiv:math/0608311 and the distribution of periodic points for Collet-
Hofbauer F (1979) On intrinsic ergodicity of piecewise Eckmann maps. Commun Math Phys 149:31–69
monotonic transformations with positive entropy. Israel Kifer Y (1977) Small random perturbations of hyperbolic
J Math 34(3):213–237 limit sets. (Russian). Uspekhi Mat Nauk 32(1):193–194
Hofbauer F (1985) Periodic points for piecewise mono- Kifer Y (1986) Ergodic theory of random transformations.
tonic transformations. Ergodic Theory Dynam Syst 5: In: Progress in probability and statistics, vol 10.
237–256 Birkhäuser Boston, Inc, Boston
Hofbauer F, Keller G (1982) Ergodic properties of invari- Kifer Y (1988) Random perturbations of dynamical sys-
ant measures for piecewise monotonic transformations. tems. In: Progress in probability and statistics, vol 16.
Math Z 180:119–140 Birkhäuser Boston, Inc, Boston
Hofbauer F, Keller G (1990) Quadratic maps without Kifer Y (1990) Large deviations in dynamical systems and
asymptotic measure. Commun Math Phys 127: stochastic processes. Trans Am Math Soc 321:505–524
319–337 Kifer Y (1997) Computations in dynamical systems via
Hofer H, Zehnder E (1994) Symplectic invariants and random perturbations. (English summary). Discrete
Hamiltonian dynamics. Birkhäuser, Basel Contin Dyn Syst 3:457–476
Host B, Kra B (2005) Nonconventional ergodic averages Kolyada SF (2004) LI-Yorke sensitivity and other concepts
and nilmanifolds. Ann Math 161(2):397–488 of chaos. Ukr Math J 56:1242–1257
Hua Y, Saghin R, Xia Z (2006) Topological entropy and Kozlovski OS (2003) Axiom A maps are dense in the space
partially hyperbolic diffeomorphisms. arXiv:math/ of unimodal maps in the Ck topology. Ann Math
0608720 157(2):1–43
Hunt FY (1998) Unique ergodicity and the approximation Kozlovski O, Shen W, van Strien S (2007) Density of
of attractors and their invariant measures using Ulam’s Axiom A in dimension one. Ann Math 166:145–182
method. (English summary). Nonlinearity 11:307–317 Krengel U (1983) Ergodic theorems. De Gruyter, Berlin
Hunt BR, Sauer T, Yorke JA (1992) Prevalence: a Krzyżewski K, Szlenk W (1969) On invariant measures for
translation-invariant “almost every” on infinite- expanding differentiable mappings. Stud Math 33:
dimensional spaces. Bull Am Math Soc (NS) 27: 83–92
217–238 Lacroix Y (2002) Possible limit laws for entrance times of
Ibragimov I, Linnik Y, Kingman JFC (ed) (trans) an ergodic aperiodic dynamical system. Israel J Math
(1971) Independent and stationary sequences of ran- 132:253–263
dom variables. Wolters-Noordhoff, Groningen Ladler RL, Marcus B (1979) Topological entropy and
Ionescu Tulcea CT, Marinescu G (1950) Théorie ergodique equivalence of dynamical systems. Memoirs Am
pour des classes d’opérations non complètement con- Math Soc 20(219)
tinues. (French). Ann Math 52(2):140–147 Lastoa A, Yorke J (1973) On the existence of invariant
Ishii Y (1997) Towards a kneading theory for Lozi map- measures for piecewise monotonic transformations.
pings. I. A solution of the pruning front conjecture and Trans Am Math Soc 186:481–488
the first tangency problem. Nonlinearity 10:731–747 Ledrappier F (1984) Proprietes ergodiques des mesures de
Jakobson MV (1981) Absolutely continuous invariant Sinaï. Inst Hautes Etudes Sci Publ Math 59:163–188
measures for one-parameter families of one- Ledrappier F, Young LS (1985) The metric entropy of
dimensional maps. Commun Math Phys 81:39–88 diffeomorphisms. I Characterization of measures satis-
Jenkinson O (2007) Optimization and majorization of fying Pesin’s entropy formula. II Relations between
invariant measures. Electron Res Announc Am Math entropy, exponents and dimension. Ann Math 122(2):
Soc 13:1–12 509–539, 540–574
Katok A (1980) Lyapunov exponents, entropy and periodic Leplaideur R (2004) Existence of SRB-measures for some
orbits for diffeomorphisms. Inst Hautes Etudes Sci Publ topologically hyperbolic diffeomorphisms. Ergodic
Math 51:137–173 Theory Dynam Syst 24:1199–1225
Katok A, Hasselblatt B (1995) Introduction to the modern Li M, Vitanyi P (1997) An introduction to Kolmogorov
theory of dynamical systems. In: Encyclopedia of complexity and its applications. In: Graduate texts in
mathematics and its applications, 54. With a supple- computer science, 2nd edn. Springer, New York
mentary chapter by: Katok A, Mendoza L. Cambridge Li TY, Yorke JA (1975) Period three implies chaos. Am
University Press, Cambridge Math Monthly 82:985–992
Keller G (1984) On the rate of convergence to equilibrium Lind D, Marcus B (1995) An introduction to symbolic
in one-dimensional systems. Commun Math Phys 96: dynamics and coding. Cambridge University Press,
181–193 Cambridge
Keller G (1990) Exponents, attractors and Hopf decompo- Liu PD, Qian M, Zhao Y (2003) Large deviations in Axiom
sitions for interval maps. Ergodic Theory Dynam Syst A endomorphisms. Proc R Soc Edinb Sect A 133:
10:717–744 1379–1388
662 Chaos and Ergodic Theory

Liverani C (1995a) Decay of correlations. Ann Math 142: Nagaev SV (1957) Some limit theorems for stationary
239–301 Markov chains. Theor Probab Appl 2:378–406
Liverani C (1995b) Decay of correlations in Piecewise Newhouse SE (1974) Diffeomorphisms with infinitely
expanding maps. J Stat Phys 78:1111–1129 many sinks. Topology 13:9–18
Liverani C (1996) Central limit theorem for deterministic Newhouse SE (1989) Continuity properties of entropy.
systems. In: Ledrappier F, Lewowicz J, Newhouse Ann Math 129(2):215–235; Erratum: Ann of Math
S (eds) International conference on dynamical systems, 131(2):409–410
Montevideo, 1995. Pitman research notes. Math 362: Nussbaum RD (1970) The radius of the essential spectrum.
56–75 Duke Math J 37:473–478
Liverani C (2001) Rigorous numerical investigation of the Ornstein D (1970) Bernoulli shifts with the same entropy
statistical properties of piecewise expanding maps – a are isomorphic. Adv Math 4:337–352
feasibility study. Nonlinearity 14:463–490 Ornstein D, Weiss B (1988) On the Bernoulli nature of
Liverani C, Tsujii M (2006) Zeta functions and dynamical systems with some hyperbolic structure. Ergodic The-
systems. (English summary). Nonlinearity 19: ory Dynam Syst 18:441–456
2467–2473 Ovid (2005) Metamorphosis. W.W. Norton, New York
Lyubich M (1997) Dynamics of quadratic polynomials. I, Palis J (2000) A global view of dynamics and a conjecture
II. Acta Math 178:185–247, 247–297 on the denseness of tinitude of attractors. Asterisque
Mañé R (1982) An ergodic closing lemma. Ann Math 261:335–347
116(2):503–540 Palis J, Yoccoz JC (2001) Fers a cheval non-uniformement
Mañé R (1985) Hyperbolicity, sinks and measures in one- hyperboliques engendres par une bifurcation homocline
dimensional dynamics. Commun Math Phys 100: et densite nulle des attracteurs. CRAS 333:867–871
495–524 Pesin YB (1976) Families of invariant manifolds
Mañé R (1988) A proof of the C1 stability conjecture. Publ corresponding to non-zero characteristic exponents.
Math IHES 66:161–210 Math USSR Izv 10:1261–1302
Margulis G (2004) On some aspects of the theory of Pesin YB (1977) Characteristic exponents and smooth
Anosov systems. In: With a survey by Richard Sharp: ergodic theory. Russ Math Surv 324:55–114
periodic orbits of hyperbolic flows. Springer mono- Pesin Ya (1997) Dimension theory in dynamical systems.
graphs in mathematics. Springer, Berlin In: Contemporary views and applications. Chicago lec-
Melbourne I, Nicol M (2005) Almost sure invariance prin- tures in mathematics. University of Chicago Press,
ciple for nonuniformly hyperbolic systems. Commun Chicago
Math Phys 260(1):131–146 Pesin Ya, Sinaĭ Ya (1982) Gibbs measures for partially
Milnor J, Thurston W (1988) On iterated maps of the hyperbolic attractors. Ergodic Theory Dynam Syst 2:
interval. In: Dynamical systems, College Park, MD, 417–438
1986–87. Lecture Notes in Math, vol 1342. Springer, Piorek J (1985) On the generic chaos in dynamical sys-
Berlin, pp 465–563 tems. Univ Iagel Acta Math 25:293–298
Misiurewicz M (1976a) A short proof of the variational Plykin RV (2002) On the problem of the topological clas-
principle for a Zn+ action on a compact space. Bull sification of strange attractors of dynamical systems.
Acad Polon Sci Sér Sci Math Astronom Phys 24(12): Uspekhi Mat Nauk 57:123–166. Translation in: Russ
1069–1075 Math Surv 57:1163–1205
Misiurewicz M (1976b) Topological conditional entropy. Poincare H (1892) Les methodes nouvelles de la
Stud Math 55:175–200 mecanique céleste. Gauthier–Villars, Paris
Misiurewicz M (1979) Horseshoes for mappings of the Pollicott M, Sharp R (2002) Invariance principles for inter-
interval. Bull Acad Polon Sci Sér Sci Math 27:167–169 val maps with an indifferent fixed point. Commun Math
Misiurewicz M (1995) Continuity of entropy revisited. In: Phys 229:337–346
Dynamical systems and applications. World Sci Ser Przytycki F (1977) On U-stability and structural stability of
Appl Anal, vol 4. World Scientific Publishing, River endomorphisms satisfying. Axiom A Studia Math 60:
Edge, pp 495–503 61–77
Misiurewicz M, Smítal J (1988) Smooth chaotic maps with Pugh C, Shub M (1999) Ergodic attractors. Trans Am Math
zero topological entropy. Ergodic Theory Dynam Syst Soc 312:1–54
8:421–424 Pujals E, Rodriguez-Hertz F (2007) Critical points for
Mora L, Viana M (1993) Abundance of strange attractors. surface diffeomorphisms. J Mod Dyn 1:615–648
Acta Math 171:1–71 Puu T (2000) Attractors, bifurcations, and chaos. In: Non-
Moreira CG, Palis J, Viana M (2001) Homoclinic tangen- linear phenomena in economics. Springer, Berlin
cies and fractal invariants in arbitrary dimension. Rees M (1981) A minimal positive entropy homeomor-
CRAS 333:475–480 phism of the 2-torus. J Lond Math Soc 23(2):537–550
Murray JD (2002, 2003) Mathematical biology. I and II An Robinson RC (2004) An introduction to dynamical sys-
introduction, 3rd edn. In: Interdisciplinary applied tems: continuous and discrete. Pearson Prentice Hall,
mathematics, 17 and 18. Springer, New York Upper Saddle River
Chaos and Ergodic Theory 663

Rousseau-Egele J (1983) Un théorème de la limite locale Szász D (ed) (2000) Hard ball systems and the Lorentz gas.
pour une classe de transformations dilatantes et mono- In: Encyclopedia of mathematical sciences, 101. Math-
tones par morceaux. Ann Probab 11:772–788 ematical physics, II. Springer, Berlin
Ruelle D (1968) Statistical mechanics of a one-dimensional Tsujii M (1992) A measure on the space of smooth map-
lattice gas. Commun Math Phys 9:267–278 pings and dynamical system theory. J Math Soc Jpn 44:
Ruelle D (1976) A measure associated with axiom-A 415–425
attractors. Am J Math 98:619–654 Tsujii M (2000a) Absolutely continuous invariant mea-
Ruelle D (1978) An inequality for the entropy of differen- sures for piecewise real-analytic expanding maps on
tiable maps. Bol Soc Brasil Mat 9:83–87 the plane. Commun Math Phys 208:605–622
Ruelle D (1982) Repellers for real analytic maps. Ergodic Tsujii M (2000b) Piecewise expanding maps on the plane
Theory Dynam Syst 2:99–107 with singular ergodic properties. Ergodic Theory
Ruelle D (1989) The thermodynamic formalism for Dynam Syst 20:1851–1857
expanding maps. Commun Math Phys 125:239–262 Tsujii M (2001) Absolutely continuous invariant measures
Ruelle D (2004) Thermodynamic formalism. In: The math- for expanding piecewise linear maps. Invent Math 143:
ematical structures of equilibrium statistical mechanics, 349–373
2nd edn. Cambridge Mathematical Library, Cambridge Tsujii M (2005) Physical measures for partially hyperbolic
University Press, Cambridge surface endomorphisms. Acta Math 194:37–132
Ruelle D (2005) Differentiating the absolutely continuous Tucker W (1999) The Lorenz attractor exists. C R Acad Sci
invariant measure of an interval map f with respect to f. Paris Ser I Math 328:1197–1202
Commun Math Phys 258:445–453 Vallée B (2006) Euclidean dynamics. Discrete Contin Dyn
Ruette S (2003a) On the Vere-Jones classification and Syst 15:281–352
existence of maximal measures for countable topolog- van Strien S, Vargas E (2004) Real bounds, ergodicity and
ical Markov chains. Pacific J Math 209:366–380 negative Schwarzian for multimodal maps. J Am Math
Ruette S (2003b) Chaos on the interval. http://www.math. Soc 17:749–782
u-psud.fr/%7Eruette Vasquez CH (2007) Statistical stability for
Saari DG (2005) Collisions, rings, and other Newtonian diffeomorphisms with dominated splitting. Ergodic
N-body problems. In: CBMS regional conference Theory Dynam Syst 27:253–283
series in mathematics, 104. Published for the Confer- Viana M (1993) Strange attractors in higher dimensions.
ence Board of the Mathematical Sciences, Washington, Bol Soc Bras Mat (NS) 24:13–62
DC. American Mathematical Society, Providence Viana M (1997a) Multidimensional nonhyperbolic attractors.
Saperstone SH, Yorke JA (1971) Controllability of linear Inst Hautes Etudes Sci Publ Math No 85:63–96
oscillatory systems using positive controls. SIAM Viana M (1997b) Stochastic dynamics of deterministic
J Control 9:253–262 systems. In: Lecture notes 21st Braz Math Colloq
Sarig O (1999) Thermodynamic formalism for countable IMPA. Rio de Janeiro
Markov shifts. Ergodic Theory Dynam Syst 19: Viana M (1998) Dynamics: a probabilistic and geometric
1565–1593 perspective. In: Proceedings of the International Con-
Sarig O (2001) Phase transitions for countable topological gress of Mathematicians, vol I, Berlin, 1998. Doc Math
Markov shifts. Commun Math Phys 217:555–577 Extra I:557–578
Sataev EA (1992) Invariant measures for hyperbolic map- Wang Q, Young LS (2003) Strange attractors in
pings with singularities. Uspekhi Mat Nauk 47: periodically-kicked limit cycles and Hopf bifurcations.
147–202. Translation in: Russ Math Surv 47:191–251 Commun Math Phys 240:509–529
Saussol B (2000) Absolutely continuous invariant mea- Weiss B (2002) Single orbit dynamics. American Mathe-
sures for multidimensional expanding maps. Israel matical Society, Providence
J Math 116:223–248 Ya S (1972) Gibbs measures in ergodic theory. Uspekhi
Saussol B (2006) Recurrence rate in rapidly mixing Mat Nauk 27(166):21–64
dynamical systems. Discrete Contin Dyn Syst 15(1): Yomdin Y (1987) Volume growth and entropy. Israel
259–267 J Math 57:285–300
Shub M (1987) Global stability of dynamical systems. Yoshihara KI (2004) Weakly dependent stochastic
Springer, New York sequences and their applications. In: Recent topics on
Simon R (1972) A 3-dimensional Abraham-Smale exam- weak and strong limit theorems, vol XIV. Sanseido Co
ple. Proc Am Math Soc 34:629–630 Ltd, Chiyoda
Starkov AN (2000) Dynamical systems on homogeneous Young LS (1982) Dimension, entropy and Lyapunov expo-
spaces. In: Translations of mathematical mono- nents. Ergodic Theory Dynam Syst 6:311–319
graphs, vol 190. American Mathematical Society, Young LS (1985) Bowen–Ruelle measures for certain
Providence piecewise hyperbolic maps. Trans Am Math Soc 287:
Stewart P (1964) Jacobellis v Ohio. US Rep 378:184 41–48
664 Chaos and Ergodic Theory

Young LS (1986) Stochastic stability of hyperbolic Books and Reviews

attractors. Ergodic Theory Dynam Syst 6:311–319 Baladi V (2000b) The magnet and the butterfly: thermo-
Young LS (1990) Large deviations in dynamical systems. dynamic formalism and the ergodic theory of chaotic
Trans Am Math Soc 318:525–543 dynamics. (English summary). In: Development
Young LS (1992) Decay of correlations for certain qua- of mathematics 1950–2000. Birkhäuser, Basel,
dratic maps. Commun Math Phys 146:123–138 pp 97–133
Young LS (1995) Ergodic theory of differentiable dynam- Bonatti C (2003) Dynamiques generiques: hyperbolicite et
ical systems. Real and complex dynamical systems. transitivite. In: Seminaire Bourbaki vol 2001/2002.
Hillerad, 1993. In: NATO Adv Sci Inst Ser C Math Asterisque No 290. pp 225–242
Phys Sci, vol 464. Kluwer, Dordrecht, pp 293–336 Kuksin SB (2006) Randomly forced nonlinear PDEs and
Young LS (1998) Statistical properties of dynamical sys- statistical hydrodynamics in 2 space dimensions. In:
tems with some hyperbolicity. Ann Math 585–650 Zurich lectures in advanced mathematics. European
Young LS (1999a) Recurrence times and rates of mixing. Mathematical Society (EMS), Zürich
Israel J Math 110:153–188 Liu PD, Qian M (1995) Smooth ergodic theory of random
Young LS, Wang D (2001) Strange attractors with one dynamical systems. In: Lecture notes in mathematics,
direction of instability. Commun Math Phys 218:1–97 vol 1606. Springer, Berlin
Young LS, Wang D (2006) Nonuniformly expanding 1D Ornstein D, Weiss B (1991) Statistical properties of chaotic
maps. Commun Math Phys 264:255–282 systems. With an appendix by David Fried. Bull Am
Young LS, Wang D (2008) Toward a theory of rank one Math Soc (NS) 24:11–116
attractors. Ann Math 167:349–480 Young LS (1999b) Ergodic theory of chaotic dynamical
Zhang Y (1997) Dynamical upper bounds for Hausdorff systems. In: XIIth International Congress of Mathemat-
dimension of invariant sets. Ergodic Theory Dynam ical Physics, ICMP ’97, Brisbane. Int Press, Cam-
Syst 17:739–756 bridge, MA, pp 131–143
 is to study those objects. One of the main tools
Ergodic Theory: Fractal Geometry is the fractal dimension theory that helps to
extract important properties of geometrically
Jörg Schmeling “irregular” sets.
Center for Mathematical Sciences, Lund
University, Lund, Sweden
Definition of the Subject

Article Outline The connection between fractal geometry and

dynamical system theory is very diverse. There
Glossary is no unified approach and many of the ideas arose
Definition of the Subject from significant examples. Also the dynamical
Introduction system theory has been shown to have a strong
Preliminaries impact on classical fractal geometry. In this article
Brief Tour Through Some Examples there are first presented some examples showing
Dimension Theory of Low-Dimensional nontrivial results coming from the application of
Dynamical Systems – Young’s Dimension dimension theory. Some of these examples require
Formula a deeper knowledge of the theory of smooth
Dimension Theory of Higher-Dimensional dynamical systems then can be provided here.
Dynamical Systems Nevertheless, the flavor of these examples can
Hyperbolic Measures be understood. Then there is a brief overview of
General Theory some of the most developed parts of the applica-
Endomorphisms tion of fractal geometry to dynamical system the-
Multifractal Analysis ory. Of course a rigorous and complete treatment
Future Directions of the theory cannot be given. The cautious reader
Bibliography may wish to check the original papers. Finally,
there is an outlook over the most recent develop-
Glossary ments. This article is by no means meant to be
complete. It is intended to give some of the ideas
Dynamical system A (discrete time) dynamical and results from this field.
system describes the time evolution of a point
in phase space. More precisely a space X is
given and the time evolution is given by a Introduction
map T : X ! X. The main interest is to describe
the asymptotic behavior of the trajectories In this section some of the aspects of fractal
(orbits) T n(x), i.e. the evolution of an initial geometry in dynamical systems are pointed out.
point x  X under the iterates of the map T. Some notions that are used will be defined later on
More generally one is interested in obtaining and I intend only to give a flavor of the applica-
information on the geometrically complicated tions. The nonfamiliar reader will find the defini-
invariant sets or measures which describe the tions in the corresponding sections and can return
asymptotic behavior. to this section later. The geometry of many invari-
Fractal geometry Many objects of interest ant sets or invariant measures of dynamical sys-
(invariant sets, invariant measures etc.) exhibit tems (including attractors, measures defining the
a complicated structure that is far from being statistical properties) look very complicated at all
smooth or regular. The aim of fractal geometry scales, and their geometry is impossible to
© Springer-Verlag 2009 665
C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6_179
Originally published in
R. A. Meyers (ed.), Encyclopedia of Complexity and Systems Science, © Springer-Verlag 2009
https://doi.org/10.1007/978-3-642-27737-5_179
666 Ergodic Theory: Fractal Geometry

describe using standard geometric tools. For some However, there are numerous examples of
important classes of dynamical systems, these dynamical systems exhibiting pathological behav-
complicated structures are intensively studied ior with respect to fractal geometrical characteris-
using notions of dimension. In many cases it tics. In particular higher-dimensional systems
becomes possible to relate these notions of dimen- seem to be as complicated as general objects con-
sion to other fundamental dynamical characteris- sidered in geometric measure theory. Therefore, a
tics, such as Lyapunov exponents, entropies, clean and unified theory is still not available.
pressure, etc. The study of characteristic notions like entropy,
On the other hand tools from dynamical sys- exponents or dimensions is an essential issue in the
tems, especially from ergodic theory and thermo- theory of dynamical systems. In many cases it
dynamic formalism, are extremely useful to helps to classify or to understand the dynamics.
explore the fractal properties of the objects in Most of these characteristics were introduced for
question. This includes dimensions of limit sets different questions and concepts. For example,
of geometric constructions (the standard Cantor entropy was introduced to distinguish non-
set being the most famous example), which a isomorphic systems and appeared to be a complete
priori, are not related to dynamical systems invariant for Bernoulli systems (Ornstein). Later,
(Furstenberg 1967; Pesin and Weiss 1996). the thermodynamic formalism (see (Ruelle 1978))
Many dimension formulas for asymptotic sets of introduced new quantities like the pressure. Bowen
dynamical systems are obtained by means of (1979) and also Ruelle discovered a remarkable
Bowen-type formulas, i.e. as roots of some func- connection between the thermodynamic formalism
tionals arising from thermodynamic formalism. and the dimension theory for invariant sets. Since
The dimension of a set is a subtle characteristic then many efforts were taken to find the relations
which measures the geometric complexity of the between all these different quantities. It occurred
set at arbitrarily fine scales. There are many that the dimension of invariant sets or measures
notions of dimension, and most definitions carries lots of information about the system, com-
involve a measurement of geometric complexity bining its combinatorial complexity with its geo-
at scale ε (which ignores the irregularities of the metric complexity. Unfortunately it is extremely
set at size less than ε) and then considers the difficult to compute the dimension in general. The
limiting measurement as ε ! 0. A priori (and in general flavor is that local divergence of orbits and
general) these different notions can be different. global recurrence cause complicated global behav-
An important result is the affirmative solution of ior (chaos). It is impossible to study the exact
the Eckmann–Ruelle conjecture by Barreira, (infinite) trajectory of all orbits. One way out is to
Pesin and Schmeling (1999), which says that for study the statistical properties of “typical” orbits by
smooth nonuniformly hyperbolic systems, the means of an invariant measure. Although the
pointwise dimension is almost everywhere con- underlying system might be smooth the invariant
stant with respect to a hyperbolic measure. This measures may often be singular.
result implies that many dimension characteristics
for the measure coincide.
The deep connection between dynamical sys- Preliminaries
tems and dimension theory seems to have been
first discovered by Billingsley (1978) through Throughout the article the following situation is
several problems in number theory. considered. Let M be a compact Riemannian man-
Another link between dynamical systems and ifold without boundary. On M is acting a dynam-
dimension theory is through Pesin’s theory of ical system generated by a C1þα diffeomorphism
dimension-like characteristics. This general the- T : M ! M. The presence of a dynamical system
ory is a unification of many notions of dimension provides several important additional tools and
along with many fundamental quantities in methods for the theory of fractal dimensions.
dynamical system such as entropies, pressure, etc. Also the theory of fractal dimensions allows one
Ergodic Theory: Fractal Geometry 667

to draw deep conclusions about the dynamical Note that this limit exists. mH(s, Z ) is called the
system. The importance and relevance of the s-dimensional outer Hausdorff measure of Z. It
study of fractal dimension will be explained in is immediate that there exists a unique value s*,
later sections. called the Hausdorff dimension of Z, at which
In the next sections some of the most important mH(s, Z ) jumps from 1 to 0.
tools in the fractal theory of dynamical systems In general it is very hard to find optimal cover-
are considered. The definitions given here are not ings and hence it is often impossible to compute the
necessarily the original definitions but rather the Hausdorff dimension of a set. Therefore a simpler
ones which are closer to contemporary use. More notion – the lower and upper Box dimension – was
details can be found in (Pesin 1997b). introduced. The difference to the Hausdorff dimen-
sion is that the covering balls are assumed to have
Some Ergodic Theory the same radius ε. Since then the limit as ε ! 0 does
Ergodic theory is a powerful method to analyze not have to exist one arrives at the notion of the
statistical properties of dynamical systems. All the upper and lower dimension.
following facts can be found in standard books on
ergodic theory like (Petersen 1983; Walters 1982). Dimension of a Measure
The main idea in ergodic theory is to relate Definition 1 Let Z  ℝN and let m be a probabil-
global quantities to observations along single ity measure supported on Z. Define the Hausdorff
orbits. Let us consider an invariant measure: m- dimension of the measure m by
( f 1A) ¼ m(A) for all measurable sets A. Such a
measure “selects” typical trajectories. It is impor- dimH ðmÞ  inf dimH ðK Þ:
KZ:mðK Þ¼1
tant to note that the properties vary with the invari-
ant measures. Any such invariant measure can be
decomposed into elementary parts (ergodic
components).
Pointwise Dimension
An invariant measure is called ergodic if for
Most invariant sets or measures are not strongly
any invariant set A ¼ T 1A one has m(A)m(M ∖ A) ¼
self-similar, i.e. the local geometry at arbitrarily
0 (with the agreement 0  1 ¼ 0), i.e. from the
fine scales might look different from point to
measure-theoretic point of view there are no non-
point. Therefore, the notion of pointwise dimen-
trivial invariant subsets.
sion with respect to a Borel probability measure is
The importance of ergodic probability mea-
defined.
sures (i.e. m(M) ¼ 1) lies in the following theorem
Let m be a Borel probability measure. By
of Birkhoff.
B(x, ε) the ball with center x and radius ε is
denoted. The pointwise dimension of the mea-
Theorem 1 (Birkhoff) Let m be an ergodic prob-
sure m at the point x is defined as
ability measure and ’  L1(m). Then
n1
lim 1
’ Tkx ¼ M’ dm m  a:e: log mðBðx, eÞÞ
n!1 n
k¼0 dm ðxÞ≔ lim
e!0 log e

Hausdorff Dimension if the above limit exists. If dm(x) ¼ d, then for

With Z  ℝN and s  0 one defines small ε the measure of small balls scales as
m(B(x, ε))  εd.
mH ðs, Z Þ ¼ lim inf diamðBi Þs : Proposition 1 Suppose m is a probability mea-
d!0 fBi g
i
sure supported on Z  ℝN. Then dm(x) ¼ d for m –
sup diamðBi Þ < e and [Bi  Zg: almost all x  Z implies dimH(m) ¼ d.
i i
668 Ergodic Theory: Fractal Geometry

One should not take the existence of a local relation of “round” balls to “oval” dynamical
dimension (even for good measures) for granted. Bowen balls. If one understands how metric
Later on it will be seen that the spectrum of the balls can be efficiently used to cover dynamical
pointwise dimension (dimension spectrum) is a balls one can use the dynamical and relatively
main object of study in classical multifractal easy relation to compute notion of entropy to
analysis. determine the dimension. However, in higher
The dimension of a measure or a set of mea- dimensions this relation is by far nontrivial.
sures is its geometric complexity. However, under A heuristic argument for comparing “round”
the presence of a dynamical system one also wants balls with dynamical balls is given in section
to measure the dynamical (combinatorial) com- “The Kaplan–Yorke Conjecture”.
plexity of the system. This leads to the notion of
entropy. The Pressure Functional
A useful tool in the dimension analysis of dynam-
Dimension-Like Characteristics and ical systems is the pressure functional. It was
Topological Entropy originally defined by means of statistical physics
Pesin’s theory of dimension-like characteristics (thermodynamic formalism) as the free energy
provides a unified treatment of dimensions and (or pressure) of a potential j (see for example
important dynamical quantities like entropies (Ruelle 1978)). However, in this article a
and pressure. The topological entropy of a contin- dimension-like definition (see (Pesin 1997b)) is
uous map f with respect to a subset Z in a metric more suitable. Again an outer measure using
space (X, r) (in particular X ¼ M – a Riemannian Bowen balls will be used. Let ’ : M ! ℝ be a
manifold) can be defined as a dimension-like continuous function and
characteristic. For each n  ℕ and ε > 0, define the
Bowen ball Bn(x, ε) ¼ {y  X : r(Ti(x), Ti( y)) ε
for 0 i n}. Then let mP ðZ, a, e, n, ’Þ ¼ lim inf exp  ani
n!1
i

mh ðZ, a, e, nÞ≔ n
þ sup ’ Tkx :
x  Bni ðx, eÞ k¼0
lim inf
n!1
e ani
: ni > n, [Bni ðx, eÞ  Z :
i
i
This defines an outer measure that jumps from
This gives rise to an outer measure that jumps 1 to 0 as α increases. The threshold value α is
from 1 to 0 at some value α . This threshold called the topological pressure of the potential j
value α is called the topological entropy of Z denoted by P(’). In many situations it does not
(at scale e). However, in many situations this depend on ε.
value does not depend on e. The topological There is also a third way of defining the pres-
entropy is denoted by htop(T| Z ). sure in terms of a variational principle (see
If Z is f – invariant and compact, this definition (Walters 1982)):
of topological entropy coincides with the usual
definition of topological entropy (Walters 1982). Pð’Þ ¼ sup hm þ ’ dm
The entropy hm of a measure m is defined as minvariant M
hm ¼ inf {htop(T| Z ) : m(Z ) ¼ 1}. For ergodic
measures this definition coincides with the Kol-
mogorov–Sinai entropy (see (Pesin 1997b)). Brief Tour Through Some Examples
One has to note that in the definition of entropy
metric (“round”) balls are substituted by Bowen Before describing the fractal theory of dynamical
balls and the metric diameter by the “depth” of the systems in more detail some ideas about its role
Bowen ball. Therefore, the relation between are presented. The application of dimension the-
entropy and dimension is determined by the ory has many different aspects. At this point some
Ergodic Theory: Fractal Geometry 669

(but by far not all) important examples are con- following example explains some of those
sidered that should give the reader some feeling difficulties.
about the importance and wide use of dimension Let 1/2 < l < 1 and consider the maps
theory in dynamical system theory. Fi : [0, 1]! [0, 1] given by F1(x) ¼ lx and
F2(x) ¼ lx þ (1  l). Then the images of F1,
Dimension of Conformal Repellers: Ruelle’s F2 have an essential overlap and J ¼ [0, 1]. If one
Pressure Formula randomizes this construction in the way that one
Computing or estimating a dimension via a pres- applies both maps each with probability 1/2 a
sure formula is a fundamental technique. Explicit probability measure is induced on J. This mea-
properties of pressure help to analyze subtle char- sure might be absolute continuous with respect to
acteristics. For example, the smooth dependence Lebesgue measure or not. Already Erdös realized
of the Hausdorff dimension of basic sets for that for some special values of l (for example for
Axiom- A surface diffeomorphisms on the deriv- the inverse of the golden mean) the induced
ative of the map follows from smoothness of measure is singular. In a breakthrough paper
pressure. B. Solomyak (1995) proved that for a.e. l the
Ruelle proved the following pressure formula induced measure is absolutely continuous.
for the Hausdorff dimension of a conformal repel- A main ingredient in the proof is a transversality
ler. A conformal repeller J is an invariant set condition in the parameter space: the images of
T(J ) ¼ J ¼ {x  M : f nx  V 8 n  ℕ and arbitrary two random samples of the (infinite)
some neighborhood Vof J} such that for any x  J applications of the maps Fi have to cross with
the differential DxT ¼ a(x)Isox where a(x) is a nonzero speed when the parameter l changes.
scalar with ja(x) j > 1 and Isox an isometry of This is a general mechanism which allows one
the tangent space TxM. to handle more general situations.

Theorem 2 (Ruelle 1982) Let T : J ! J be a Homoclinic Bifurcations for Dissipative

conformal repeller, and consider the function Surface Diffeomorphisms
t ! P(t log| DxT|), where P denotes pressure. Homoclinic tangencies and their bifurcations play
Then a fundamental role in the theory of dynamical
systems (Palis and Takens 1987, 1993, 1994).
Pðs log jDx TjÞ ¼ 0 , s ¼ dimH ðJ Þ Systems with homoclinic tangencies have a com-
plicated and subtle quasi-local behavior.
Newhouse showed that homoclinic tangencies
can persist under small perturbations, and that
Iterated Function Systems horseshoes may co-exist with infinitely many
In fractal geometry one of the most-studied sinks in a neighborhood of the homoclinic orbit
objects are iterated function systems, where there and hence the system is not hyperbolic
are given n contractions F1,   , Fn of ℝd. The (Newhouse phenomenon).
unique compact set J fulfilling J ¼ [ i Fi ðJ Þ is Let Tm : M2 ! M2 be a smooth parameter
called the attractor of the iterated function system family of surface diffeomorphisms that exhibits
(see (Falconer 1990)). One question is to evaluate for m ¼ 0 an invariant hyperbolic set Λ0
its Hausdorff dimension. Often (for example (horseshoe) and undergoes a homoclinic bifurca-
under the open set condition) the attractor J can tion. The Hausdorff dimension of the hyperbolic
be represented as the repeller of a piecewise set Λ0 for T0 determines whether hyperbolicity is
expanding map T : ℝd ! ℝd where the Fi are the the typical dynamical phenomenon near T0 or not.
inverse branches of the map T. In general it is by If dimH Λ0 < 1, then hyperbolicity is the prevalent
far not trivial to determine the dimension of J or dynamical phenomenon near f0. This is not the
even the dimension of a measure sitting on J. The case if dimH Λ0 > 1.
670 Ergodic Theory: Fractal Geometry

More precisely, let NW m denote the set of Theorem 5 (Eggleston 1952) The Hausdorff
nonwandering points of Tm in an open neighbor- dimension of Xp is given by
hood of Λ0 after the homoclinic bifurcation. Let ‘
denote Lebesgue measure. dimH Xp ¼ ð1= log 2Þp
½ log p  ð1 pÞ logð1 pÞ :

Theorem 3 (Palis and Takens 1987, 1993) If

dimHΛ0 < 1, then The underlying dynamical system is E2(x) ¼
2x mod 1 and the dimension is the dimension
‘ m  ½0, m0 : NW m is hyperbolic of the Bernoulli p measure. Other cases included
lim ¼ 1:
m0 !0 m0 that by Rényi who proposed generalization of the
base d expansion from integer base d to noninteger
base β. In this case the underlying dynamical sys-
The hyperbolicity co-exists with the tem is the (in general non-Markovian) beta shift
Newhouse phenomena for a residual set of param- β(x) ¼ βx mod 1 studied in (Schmeling 1997).
eter values. There are many investigations concerning the
approximation of real numbers by rationals by
Theorem 4 (Palis–Yoccoz) If dimHΛ0 > 1, then using dynamical methods. The underlying
dynamical system in this case is the Gauss map
‘ m  ½0, m0 : NW m is hyperbolic T(x) ¼ (1/x) mod 1. This map is uniformly
lim < 1: expanding, but has infinitely many branches.
m0 !0 m0

Example Continued fraction approximation of

numbers
Some Applications to Number Theory Consider the continued fraction expansion of
Sometimes dimension problems in number theory x  [0, 1], i.e.,
can be transferred to dynamical systems and
1
attacked using tools from dynamics. x¼ 1
¼ ½a1 , a2 , a3 , a4 ,    ,
a1 þ a þ 1
2 1
a3 þ
a4 þ⋱
Example Diadic expansion of numbers: Con-
sider a real number expanded in base 2,
and the approximants pn(x)/qn(x) ¼ [a1, a2, . . .,
i.e. x ¼ 1
n¼1 xn =2 : Let
n
an] (i.e. the approximation given by the finite
continued fraction after step n).
n1
1 The set of numbers which admit a faster
Xp ¼ x : lim xk ¼ p :
n!1 n approximation by rational numbers is defined as
k¼0
follows. For t  2, let F t
Borel showed that ‘(X1/2) ¼ 1, where ‘ denotes
Lebesgue measure. This result is an easy conse- p 1
Ft ¼ x  ½01 : x for infinitely many p=q :
quence of the Birkhoff ergodic theorem applied to q qt
the characteristic function of the digit 1 (which in
this simple case is the Strong Law of Large Num- It is well known that this set has zero measure
bers for i.i.d. processes). for each t  2. Jarnik (1931) computed the
One can ask how large is the set Xp in general. Hausdorff dimension of F t and showed that
Eggleston (1952) discovered the following won- dimH(F t) ¼ 2/t. However, nowadays there are
derful dimension formula, which Billingsley methods from dynamical systems which not only
(1978) interpreted in terms of dynamics and allow unified proofs but also can handle more
reproved using tools from ergodic theory. subtle subsets of the reals defined by some
Ergodic Theory: Fractal Geometry 671

properties of their continued fraction expansion known to be fractal. However, its complete
(see for example (Aihua et al. 2005; Pollicott and description is still not available.
Weiss 1999)).
Embedology and Computational Aspects of
Infinite Iterated Function Systems and Dimension
Parabolic Systems Tools from dynamical systems are becoming
In the previous section a system with infinitely increasingly important to study the time evolution
many branches appeared. This can be regarded as of deterministic systems in engineering and the
an iterated function system with infinitely many physical and biological sciences. One of the main
maps Fi. This situation is quite general. If one ideas is to model a “real world” system by a
considers a (one-dimensional) system with a par- smooth dynamical system which possesses a
abolic (indifferent) fixed point, i.e. there is a fixed strange attractor with a natural ergodic invariant
point where the derivative has absolute value measure. When studying a complicated real world
equal to 1, one often uses an induced system. system, one can measure only a very small num-
For this one chooses nearby the parabolic point a ber of variables. The challenge is to reconstruct
higher iterate of the map in order to achieve uni- the underlying attractor from the time measure-
form expansion away from the parabolic point. ment of a scaler quantity. An idealized measure-
This leads to infinitely many branches since the ment is considered as a function h : Mn ! ℝ.
number of iterates has to be increased the closer The main tool researchers currently use to
the parabolic point is. The main difference to the reconstruct the model system is called attractor
finite iterated function system is that the setting is reconstruction (see papers and references in (Ott
no longer compact and many properties of the et al. 1994)). This method is based on embedding
pressure functional are lost. with time delays (see the influential paper
Mauldin and Urbański and others (see for (Cruchfield et al. 1980), where the authors attri-
example (Aihua et al. 2005; Mauldin and bute the idea of delay coordinates to Ruelle),
Urbański 1996, 2000, 2002)) developed a thermo- where one attempts to reconstruct the attractor
dynamic formalism adapted to the pressure func- for the model using a single long trajectory.
tional for infinite iterated function systems. Then one considers the points in R pþ1 defined
Besides noncompactness one of the main prob- by (xk, xkþt, xkþ2t, . . ., xkþpt).
lems is the loss of analyticity (phase transitions) Takens (1981) showed that for a smooth
and convexity of the pressure functional for infi- T : Mn ! Mn and for typical smooth h, the map-
nite iterated function systems. ping ’ : Mn ! ℝ2nþ1 defined by x ! (h(x),
h( ft(x)),   , h( f 2nt(x)) is an embedding. Since
Complex Dynamics the box dimension of the attractor Λ may be
Let T : ℂ ! ℂ be a polynomial and J its Julia set much less than the dimension of the ambient
(repeller of this system). If this set is hyperbolic, manifold n, an interesting mathematical question
i.e. the derivative at each point has absolute value is whether there exists p < 2n þ 1 such that the
larger than 1 the study of the dimension can be mapping on the attractor ’ : Λ ! ℝp defined by
related to the study of a finite iterated function x ! (h(x), h( f(x)), . . ., h( f p(x)) is 1  1? It is
system. However, in the presence of a parabolic known that for a typical smooth h the mapping j
fixed point this leads to an infinite iterated func- is 1  1 for p > 2dimB(Λ) (Casdagli et al. 1991).
tion system.
If one considers the coefficients of the polyno- Denjoy Systems
mial as parameters one often sees a qualitative This section will give some ideas indicating the
change in the asymptotic behavior. For example, principle difficulties that arise in systems with low
the classical Mandelbrodt set for polynomials complexity. Contrary to hyperbolic systems (each
z2 þ c is the locus of values c  C for which the vector in the tangent space is either contracted or
orbit of the origin stays bounded. This set is well expanded) finer mechanisms determine the local
672 Ergodic Theory: Fractal Geometry

behavior of the scaling of balls. While in hyper- Theorem 6 Assume that 0 < δ < 1 and that
bolic systems the dynamical scaling of small balls α  (0, 1) is of Diophantine class n  (0, 1).
is exponential in a low-complexity system this Then an orientation preserving C1þδ
scaling is subexponential and hence the lineariza- diffeomorphism of the circle with rotation number
tion error is of the same magnitude. Up to now α and minimal set Oda satisfies
there is no general dimension theory for low com-
plexity systems. d
A specific example presented here is consid- dimH Oda  :
n
ered in (Kra and Schmeling 2002). Poincaré
showed that to each orientation preserving
homeomorphism of the circle S1 ¼ ℝ/ℤ is associ- Furthermore, these results are sharp, i.e. the
ated a unique real parameter α  [0, 1), called the standard Denjoy examples attain the minimum.
rotation number, so that the ordered orbit structure
of T is the same as that of the rigid rotation Rα,
Return Times and Dimension
where Rα(t) ¼ (t þ α) mod 1, provided that α is
Recently an interesting connection between the
irrational. Half a century later, Denjoy (1932)
pointwise dimensions, multifractal analysis, and
constructed examples of C1 diffeomorphisms
recurrence behavior of trajectories was discov-
that are not conjugate (via a homeomorphism) to
ered (Afraimovich et al. 2000; Barreira and
rotations. This was improved later on by Herman
Saussol 2001a; Boshernitzan 1993). Roughly
(1979). In these examples, the minimal set of T is
speaking, given an ergodic probability measure
necessarily a Cantor set Ω.
m the return time asymptotics (as the neighbor-
The arithmetic properties of the rotation number
hood of the point shrinks) of m-a.e point is deter-
have a strong effect on the properties of T. One area
mined by the pointwise dimension of m at this
that has been well understood is the relation
point. The deeper understanding of this relation
between the differentiability of T, the differentiabil-
would help to get a unified approach to dimen-
ity of the conjugation and the arithmetic properties
sions, exponents, entropies, recurrence times and
of the rotation number. (See, for example, Herman
correlation decay.
(1979)) Without stating any precise theorem, the
results differ sharply for Diophantine and for
Liouville rotation numbers (definition follows).
Roughly speaking the conjugating map is always Dimension Theory of Low-Dimensional
regular for Diophantine rotation numbers while it Dynamical Systems – Young’s
might be not smooth at all for Liuoville rotation Dimension Formula
numbers.
In this section a remarkable extension of Ruelle’s
Definition 2 An irrational number α is of dimension formula by Young (1982) for the
Diophantine class n ¼ n(α)  ℝ+ if dimension of a measure is discussed.

1
k qa k< Theorem 7 Let T : M2 ! M2 be a C2 surface
qm
diffeomorphism and let m be an ergodic measure.
has infinitely many solutions in integers q for Then
m < n and at most finitely many for m > n where
k  k denotes the distance to the nearest integer. 1 1
dimH ðmÞ ¼ hm ðf Þ  ,
l1 l2
In (Kra and Schmeling 2002) the effect of the
rotation number on the dimension of Ω is studied. where l1 l2 are the two Lyapunov exponents
There the main result is for m.
Ergodic Theory: Fractal Geometry 673

In (McCluskey and Manning 1983), Manning Dimension Theory of Higher-

and McCluskey prove the following dimension Dimensional Dynamical Systems
formula for a basic set (horseshoe) of an
Axiom-A surface diffeomorphism which is a set- Here an example of a hyperbolic attractor in
version of Young’s formula. dimension 3 is considered to highlight some of
the difficulties. Let △ denote the unit disc in ℝ2.
Theorem 8 Let Λ be a basic set for a C2 Axiom-A Let f : S1 Δ ! S1 Δ be of the form
surface diffeomorphism T : M2 ! M2. Then
dimH(Λ) ¼ s1 þ s2, where s1 and s2 satisfy T ðt, x, yÞ ¼ ’ðtÞ, c1 ðt, xÞ, c2 ðt, yÞ ,

with
P s log DT x jEux ¼0

P slog DT x jEsx ¼0: @ 1 @ 2

0 < max S1 △ c ðt, xÞ < min S1 △ c ðt, yÞ
@x @y

where Es and Eu are the stable and unstable direc- < l < 1:
tions, respectively.
The limit set
Some Remarks on Dimension Theory for Low-
L≔ \ T n S1 D
Dimensional Versus High-Dimensional nℕ
Dynamical Systems
Unlike lower dimensions (one, two, or conformal is called the attractor or the solenoid. It is an
repellers), for higher-dimensional dynamical sys- example of a structurally stable basic set and is
tems there are no general dimension formulas one of the fundamental examples of a uniformly
(besides the Ledrappier–Young formula), and in hyperbolic attractor.
general dimension theory is much more difficult. The following result can be proved.
This is due to several problems:
Theorem 9 (Bothe 1995; Hasselblatt and
1. The geometry of the Bowen balls differs in a Schmeling 2004) For all t, the thermodynamic
substantial way from round balls. pressure
2. Number theoretic properties of some scaling
rates (Pollicott and Weiss 1994; Przytycki and @ 2
P dimH Lst log c ðt, yÞ ¼ 0:
Urbański 1989) enter into dimension calcula- @y
tions in ways they do not in low dimensions
(see section “Iterated Function Systems”). In particular, the stable dimension is indepen-
3. The dimension theory of sets is often reduced dent of the stable section.
to the theory of invariant measures. However,
there is no invariant measure of full dimension In this particular case the invariant axes for
in general and measure-theoretic consider- strong and weak contraction split the system
ations do not apply (McCluskey and Manning smoothly and the difficulty is to show that the
1983). strong contraction is dominated by the weaker.
4. The stable and unstable foliations for higher In particular one has to ensure that effects as
dimensional systems are typically not C1 described in section “Iterated Function Systems”
(Hasselblatt 1994; Schmeling 1994). Hence, do not appear. In the general situation this is not
to split the system into an expanding and a the case and one lacks a similar theorem in the
contracting part is far more subtle. general situation. In particular, the unstable
674 Ergodic Theory: Fractal Geometry

foliation is not better than Hölder and does not Consider a small ball B in the phase space. The
provide a “nice” coordinate. image TnB is almost an ellipsoid with axes of
length

Hyperbolic Measures ew1 n ,   , ewp n :

Given x  M and a vector v  TxM the For 1 i s, cover T nB by balls of radius ewi n .
Lyapunov exponent is defined as Then approximately

1 exp wiþ1 n exp wp n

wðx, vÞ ¼ lim log Dx T n v : ... :
n!1 n
exp½wi n exp½wi n
provided this limit exists. For fixed x, the numbers balls are needed for covering. The dimension can
w(x, ) attain only finitely many values. By ergo- be estimated from above by
dicity they are m-a.e. constant, and the
corresponding values are denoted by
j>i wj
dimB L þ ðp iÞ≔dimiL L: ð1Þ
j wi j
w1  wp ,
This is the Kaplan–Yorke formula.
where p ¼ dim M. Denote by s (s stands for
stable) the largest index such that ws < 0.
General Theory
Definition 3 The invariant ergodic measure m is In this section the dimension theory of higher-
said to be hyperbolic if wi 6¼ 0 for every i ¼ 1, dimensional dynamical systems is investigated.
. . ., p. Most developed is this theory for invariant mea-
sures. There is an important connection between
The Kaplan–Yorke Conjecture Lyapunov exponents and the measure theoretic
In this section a heuristic argument for the dimen- entropy and dimensions of measures that will be
sion of an invariant ergodic measure will be given. presented here.
This argument uses a specific cover of a dynami- Let m be an ergodic invariant measure. The
cal ball by “round” balls. These ideas are essen- Oseledec and Pesin Theory guarantee that local
tially developed by Kaplan and Yorke (1979) for stable manifolds exist at m-a.e. point. As for the
the estimation of the dimension of an invariant set. Kaplan–Yorke formula the idea is to consider the
Their estimates always provide an upper bound contributions to the entropy and to the dimension
for the dimension. Kaplan and Yorke conjectured in the directions of wi.
that in typical situations these estimates also pro- Historically the first connections between
vide a lower bound for the dimension of an attrac- exponents and entropy were discovered by
tor. Ledrappier and Young (1985a, b) showed that Margulis and Ruelle. They proved that
in case this holds for an invariant measure, then
this measure has to be very special: an SBR- s

measure (after Sinai, Ruelle, and Bowen) hm ð f Þ  wi

i¼1
(an SRB-measure is a measure that describes the
asymptotic behavior for a set of initial points of for a C1 diffeomorphism T. Pesin (1977) showed
positive Lebesgue measure and has absolutely that this inequality is actually an equality if the
continuous conditional measures on unstable measure m is essentially the Riemannian volume
manifolds). on unstable manifolds. Ledrappier and Young
Ergodic Theory: Fractal Geometry 675

(1985a) showed that this is indeed a necessary The hypotheses of this theorem are sharp.
condition. They also provided an exact formula: Ledrappier and Misiurewicz (1985) constructed
an example of a nonhyperbolic measure for
Theorem 10 (Ledrappier–Young (1985a, b)) which the pointwise dimension is not constant
With d0 ¼ 0 for a C2 diffeomorphism holds a.e. In (Pesin and Weiss 1997a), Pesin and Weiss
present an example of a Hölder homeomorphism
s with Hölder constant arbitrarily close to one,
h m ðf Þ ¼  wi di d i1 where the pointwise dimension for the unique
i¼1
measure of maximal entropy does not exist
where di are the dimensions of the (conditional) a.e. There is also a one-dimensional example by
measure on the ith unstable leaves. Cutler (1990).

The proof of this theorem is difficult and uses

the theory of nonuniform hyperbolic systems Endomorphisms
(Pesin theory).
In dimension 1 and 2 the above theorem resem- The previous section indicates that the dimen-
bles Ruelle’s and Young’s theorems. sional properties of hyperbolic measures under
The reader should note that the above theorem invariant conditions for a diffeomorphism are
includes also the existence of the pointwise understood. However, partial differential equa-
dimension along the stable and unstable direction. tions often generate only semi-flows and the
Here the question arises whether this implies the corresponding dynamical system is noninvertible.
existence of the pointwise dimension itself. Also, Poincaré sections sometimes introduce sin-
gularities. For such dynamical systems the theory
The Existence of the Pointwise Dimension for of diffeomorphisms does not apply. However, the
Hyperbolic Measure – The Eckmann–Ruelle next theorem allows under some conditions appli-
Conjecture cation of this theory. It essentially rules out similar
In (Barreira et al. 1999), Barreira, Pesin, and situations as considered in section “Iterated Func-
Schmeling prove that every hyperbolic measure tion Systems”.
has an almost local product structure, i.e., the mea-
sure of a small ball can be approximated by the
product of the stable conditional measure of the Definition 4 A system (possibly with singulari-
ties) is almost surely invertible if it is invertible on
stable component and the unstable conditional
a set of full measure. This implies that a full
measure of the unstable component, up to small
exponential error. This was used to prove the exis- measure set of points has unique forward and
backward trajectories.
tence of the pointwise dimension of every hyper-
bolic measure almost everywhere. Moreover, the
pointwise dimension is the sum of the contributions Theorem 11 (Schmeling–Troubetzkoy (1998))
from its stable and unstable part. This implies that A two-dimensional system with singularities is
most dimension-type characteristics of the measure almost surely invertible (w.r.t. an SRB–measure)
(including the Hausdorff dimension, box dimen- if and only if Young’s formula holds.
sion, and information dimension) coincide, and
provides a rigorous mathematical justification of
the concept of fractal dimension for hyperbolic Multifractal Analysis
measures. The existence of the pointwise dimen-
sion for hyperbolic measures had been conjectured A group of physicists (Halsey et al. 1986)
long before by Eckmann and Ruelle. suggested the idea of a multifractal analysis.
676 Ergodic Theory: Fractal Geometry

The Dynamical Characteristic View of gE ðxÞ≔hm ðxÞ  Y ¼ ℝ

Multifractal Analysis
The aim of multifractal analysis is an attempt to and
understand the fine structure of the level sets of the
fundamental asymptotic quantities in ergodic the- GE ðZÞ ¼ htop ðTjZÞ
ory (e.g., Birkhoff averages, local entropy,
Lyapunov exponents). For ergodic measures the associated multifractal spectrum f Em ðaÞ is
these quantities are a.e. constant, however may called the entropy spectrum.
depend on the underlying ergodic measure.
Important elements of multifractal analysis entail The Dimension Spectrum
determining the range of values these characteris- This is the classical multifractal spectrum.
tics attain, an analysis of the dimension of the Let m be an invariant ergodic measure on a
level sets, and an understanding of the sets complete separable metric space X.
where the limits do not exist. A general concept Set
of multifractal analysis was proposed by Barreira,
Schmeling and Pesin (1997b). An important field gD ðxÞ≔d m ðxÞ  Y ¼ ℝ
of applications of multifractal analysis is to
describe sets of real numbers that have constraints and
on their digits or continued fraction expansion.
GD ðZÞ ¼ dimH Z:
General Multifractal Formalism
m ðaÞ is
The associated multifractal spectrum f D
In this section the abstract theory of multifractal
analysis is described. Let X, Y be two measurable called the dimension spectrum.
spaces and g : X ∖ ℬ ! Y be any measurable
function where ℬ is a measurable (possibly The Lyapunov Spectrum
empty) subset of X (in the standard applications Let
Y ¼ ℝ or Y ¼ ℂ). The associated multifractal
decomposition of X is defined as 1
n1
gL ðxÞ≔wðxÞ ¼ n!1
lim c k x  Y ¼ ℝ
n
X ¼ ℬ [ [ K ga k¼0
aY
where c(x) ¼ log j DxTj and
where
GD ðZ Þ ¼ dimH Z or GE ðZÞ ¼ htop ðsjZÞ:
K ga ≔fx  X : gð x Þ ¼ ag
L ðaÞ and
The associated multifractal spectra f D
For a given set function G : 2 ! ℝ the multi-
X
f EL ðaÞ
are called Lyapunov spectra.
fractal spectrum is defined by It was observed by H. Weiss (1999) and
Barreira, Pesin and Schmeling (1997b) that for
F ðaÞ≔G K ga : conformal repellers

At this point some classical and concrete exam- log 2

L ð aÞ ¼ f mh
fD ð2Þ
D
ples of this general framework are considered. a

The Entropy Spectrum where the dimension spectrum on the right-hand

Let m be an ergodic invariant measure for side is with respect to the measure of maximal
T : X ! X. If one sets entropy, and
Ergodic Theory: Fractal Geometry 677

f EL ðaÞ ¼ f EmD ðdimH L  aÞ ð3Þ of the pressure functional with respect to a suit-
able chosen family of potentials. In the remaining
where the entropy spectrum on the right-hand side items this is no longer the case. Analyticity and
is with respect to the measure of maximal convexity properties of the pressure functional are
dimension. lost. However, the authors succeeded to provide a
The following list summarizes the state of the satisfactory theory in these cases.
art for the dynamical characteristic multifractal
analysis of dynamical systems. The precise state- Multifractal Analysis and Large Deviation
ments can be found in the original papers. Theory
There are deep connections between large devia-
tion theory and multifractal analysis. The varia-
• (Pesin and Weiss 1997b; Weiss 1999) For con- tional formula for pressure is an important tool in
formal repellers and Axiom-A surface the analysis, and can be viewed (and proven) as a
diffeomorphisms, a complete multifractal anal- large deviation result (Ellis 1985). Some authors
ysis exists for the Lyapunov exponent. use large deviation theory as a tool to effect multi-
• (Barreira et al. 1997b; Pesin and Weiss 1997b; fractal analysis.
Weiss 1999) For mixing a subshift of finite
type, a complete multifractal analysis exists
for the Birkhoff average for a Hölder continu- Future Directions
ous potential and for the local entropy for a
Gibbs measure with Hölder continuous The dimension theory is fast developing and of
potential. great importance in the theory of dynamical sys-
• (Barreira and Saussol 2001b; Pesin and tems. In the most ideal situations (low dimensions
Sadovskaya 2001) There is a complete multi- and hyperbolicity) a generally far reaching and
fractal analysis for hyperbolic flows. powerful theory has been developed. It uses
• (Takens and Verbitzki 1999) There is a gener- ideas from statistical physics, fractal geometry,
alization of the multifractal analysis on sub- probability theory and other fields.
shifts with specification and continuous Unfortunately, the richness of this theory does
potentials. not carry over to higher-dimensional systems.
• (Barreira and Saussol 2001c; Barreira et al. However, recent developments have shown that
2002b) There is an analysis of “mixed” spectra it is possible to obtain a general theory for the
like the dimension spectrum of local entropies dimension of measures. Part of this theory is the
and also an analysis of joint level sets deter- development of the analytic tools of nonuniformly
mined by more than one (measurable) hyperbolic systems.
function. Therefore, the dimension theory of dynamical
• (Pollicott and Weiss 1999) For the Gauss map systems is far from complete. In particular, it is
(and a class of nonuniformly hyperbolic maps) usually difficult to apply the general theory to
a complete multifractal analysis exists for the concrete examples, for instance if one really
Lyapunov exponent. wants to compute the dimension. The general
• (Iommi 2005) A general approach to multi- theory does not provide a way to compute the
fractal analysis for repellers with countably dimension but gives rather connections to other
many branches is developed. It shows in con- characteristics. Moreover, in the presence of neu-
trary to finitely many branches features of non- tral directions (zero Lyapunov exponents) one
analytic behavior. encounters all the difficulties arising in low-
complexity systems.
In the first three statements the multifractal Another important open problem is to under-
spectra are analytic concave functions that can stand the dimension theory of invariant sets in
be computed by means of the Legendre transform higher-dimensional spaces. One way would be to
678 Ergodic Theory: Fractal Geometry

relate the dimension of sets to the dimension of Barreira L (1996a) A non-additive thermodynamic formal-
measures. Such a connection is not clear. The ism and applications to dimension theory of hyperbolic
dynamical systems. Ergodic Theory Dynam Syst 16:
reason is that most systems do not exhibit a mea- 871–927
sure whose dimension coincides with the dimen- Barreira L (1996b) A non-additive thermodynamic formal-
sion of its support (invariant set). But there are ism and dimension theory of hyperbolic dynamical
some reasons to conjecture that any compact systems. Math Res Lett 3:499–509
Barreira L. Variational properties of multifractal spectra.
invariant set of an expanding map in any dimen- IST preprint
sion carries a measure of maximal dimension (see Barreira L, Saussol B (2001a) Hausdorff dimension of
(Gatzouras and Peres 1996; Kenyon and Peres measure via Poincaré recurrence. Commun Math Phys
1996)). If this conjecture is true one obtains an 219(2):443–463
Barreira L, Saussol B (2001b) Multifractal analysis of
invariant measure whose unstable dimension hyperbolic flows. Commun Math Phys 219(2):443–463
coincides with the unstable dimension of the Barreira L, Saussol B (2001c) Variational principles and
invariant set. There is also a measure of maximal mixed multifractal spectra. Trans Am Math Soc
stable dimension. Combining these two measures 353(10):3919–3944. (electronic)
Barreira L, Schmeling J (2000) Sets of “non-typical” points
one could establish an analogous theory for have full topological entropy and full Hausdorff dimen-
invariant sets as for invariant measures. sion. Israel J Math 116:29–70
Last but not least one has to mention the impact Barreira L, Pesin Y, Schmeling J (1997a) Multifractal
of the dimension theory of dynamical systems on spectra and multifractal rigidity for horseshoes. J Dyn
Contr Syst 3:33–49
other fields. This new point of view makes in Barreira L, Pesin Y, Schmeling J (1997b) On a general
many cases the posed problems more tractable. concept of multifractal rigidity: multifractal spectra
This is illustrated in examples from number the- for dimensions, entropies, and Lyapunov exponents.
ory, geometric limit constructions and others. The Multifractal rigidity. Chaos 7:27–38
Barreira L, Pesin Y, Schmeling J (1999) Dimension and
applications of the dimension theory of dynamical product structure of hyperbolic measures. Ann Math
systems to other questions seem to be unlimited. 149:755–783
Barreira L, Saussol B, Schmeling J (2002a) Distribution of
frequencies of digits via multifractal analysis.
J Number Theory 97(2):410–438
Bibliography Barreira L, Saussol B, Schmeling J (2002b) Higher–
dimensional multifractal analysis. J Math Pures Appl
Primary Literature (9) 81(1):67–91
Afraimovich VS, Chernov NI, Sataev EA (1995) Statistical Belykh VP (1982) Models of discrete systems of phase
properties of 2D generalized hyperbolic attractors. synchronization. In: Shakhildyan VV, Belynshina LN
Chaos 5:238–252 (eds) Systems of phase synchronization. Radio i Svyaz,
Afraimovich VS, Schmeling J, Ugalde J, Urias J (2000) Moscow, pp 61–176
Spectra of dimension for Poincaré recurrences. Discret Billingsley P (1978) Ergodic theory and information.
Contin Dyn Syst 6(4):901–914 Krieger
Aihua F, Yunping J, Jun W (2005) Asymptotic Hausdorff Blinchevskaya M, Ilyashenko Y (1999) Estimate for the
dimensions of Cantor sets associated with an asymp- entropy dimension of the maximal attractor for
totically non-hyperbolic family. Ergodic Theory k-constracting systems in an infinite-dimensional
Dynam Syst 25(6):1799–1808 space. Russ J Math Phys 6(1):20–26
Alexander J, Yorke J (1984) Fat Baker’s transformations. Boshernitzan M (1993) Quantitative recurrence results.
Ergodic Theory Dynam Syst 4:1–23 Invent Math 113:617–631
Ambroladze A, Schmeling J (2004) Lyapunov exponents Bothe H-G (1995) The Hausdorff dimension of certain
are not stable with respect to arithmetic subsequences. solenoids. Ergodic Theory Dynam Syst 15:449–474
In: Fractal geometry and stochastics III. Progr Probab, Bousch T (2000) Le poisson n’a pas d’arêtes. Ann IH
vol 57. Birkhäuser, Basel, pp 109–116 Poincaré (Prob-Stat) 36(4):489–508
Artin E (1965) Ein mechanisches System mit quasi- Bowen R (1973) Topological entropy for noncompact sets.
ergodischen Bahnen, collected papers. Addison Wes- Trans Am Math Soc 184:125–136
ley, pp 499–501 Bowen R (1979) Hausdorff dimension of quasi-circles.
Barreira L (1995) Cantor sets with complicated geometry Publ Math IHES 50:11–25
and modeled by general symbolic dynamics. Random Bylov D, Vinograd R, Grobman D, Nemyckii V (1966)
Comput Dyn 3:213–239 Theory of Lyapunov exponents and its application to
Ergodic Theory: Fractal Geometry 679

problems of stability. Izdat “Nauka”, Moscow Grassberger P, Procaccia I, Hentschel H (1983) On the
(in Russian) characterization of chaotic motions, Lect notes. Physics
Casdagli M, Sauer T, Yorke J (1991) Embedology. J Stat 179:212–221
Phys 65:589–616 Halsey T, Jensen M, Kadanoff L, Procaccia I, Shraiman
Ciliberto S, Eckmann JP, Kamphorst S, Ruelle D (1971) B (1986) Fractal measures and their singularities: the
Liapunov exponents from times. Phys Rev A 34 characterization of strange sets. Phys Rev
Collet P, Lebowitz JL, Porzio A (1987) The dimension A 33(N2):1141–1151
spectrum of some dynamical systems. J Stat Phys 47: Hasselblatt B (1994) Regularity of the Anosov splitting
609–644 and of horospheric foliations. Ergodic Theory Dynam
Constantin P, Foias C (1988) Navier-Stokes equations. Syst 14(4):645–666
Chicago University Press Hasselblatt B, Schmeling J (2004) Dimension product
Cruchfield J, Farmer D, Packard N, Shaw R (1980) Geom- structure of hyperbolic sets. In: Modern dynamical
etry from a time series. Phys Rev Lett 45:712–724 systems and applications. Cambridge University
Cutler C (1990) Connecting ergodicity and dimension in Press, Cambridge, pp 331–345
dynamical systems. Ergodic Theory Dynam Syst 10: Henley D (1992) Continued fraction cantor sets, Hausdorff
451–462 dimension, and functional analysis. J Number Theory
Denjoy A (1932) Sur les courbes défines par les équations 40:336–358
différentielles á la surface du tore. J Math Pures Appl 2: Hentschel HGE, Procaccia I (1983) The infinite number of
333–375 generalized dimensions of fractals and strange
Ding M, Grebogi C, Ott E, Yorke J (1993) Estimating attractors. Physica 8D:435–444
correlation dimension from a chaotic times series: Herman MR (1979) Sur la conjugaison différentiable des
when does the plateau onset occur? Phys D 69:404–424 difféomorphismes du cercle á des rotations. Publ l’Inst
Dodson M, Rynne B, Vickers J (1990) Diophantine Math Hautes Études Sci 49:5–234
approximation and a lower bound for Hausdorff dimen- Hunt B (1996) Maximal local Lyapunov dimension bounds
sion. Mathematika 37:59–73 the box dimension of chaotic attractors. Nonlinearity 9:
Douady A, Oesterle J (1980) Dimension de Hausdorff Des 845–852
Attracteurs. CRAS 290:1135–1138 Iommi G (2005) Multifractal analysis for countable Mar-
Eggleston HG (1952) Sets of fractional dimension which kov shifts. Ergodic Theory Dynam Syst 25(6):
occur in some problems of number theory. Proc Lond 1881–1907
Math Soc 54:42–93 Jarnik V (1931) Über die simultanen diophantischen
Ellis R (1984) Large deviations for a general class of Approximationen. Math Zeitschr 33:505–543
random vectors. Ann Prob 12:1–12 Jenkinson O (2001) Rotation, entropy, and equilibrium
Ellis R (1985) Entropy, large deviations, and statistical states. Trans Am Math Soc 353:3713–3739
mechanics. Springer Kalinin B, Sadovskaya V (2002) On pointwise dimension
Falconer K (1990) Fractal geometry, mathematical foun- of non-hyperbolic measures. Ergodic Theory Dynam
dations and applications. Cambridge University Press, Syst 22(6):1783–1801
Cambridge Kaplan JL, Yorke JA (1979) Functional differential equa-
Fan AH, Feng DJ, Wu J (2001) Recurrence, dimension and tions and approximation of fixed points. Lecture notes.
entropy. J Lond Math Soc 64(2):229–244 In: Mathematics, vol 730. Springer, Berlin, pp 204–227
Frederickson P, Kaplan J, Yorke E, Yorke J (1983) The Katznelson Y, Weiss B (1982) A simple proof of some
Liapunov dimension of strange attractors. J Differ Equ ergodic theorems. Isr J Math 42:291–296
49:185–207 Kenyon R, Peres Y (1996) Measure of full dimension on
Frostman O (1935) Potential d’équilibre Et Capacité des affine invariant sets. Ergodic Theory Dynam Syst 16:
Ensembles Avec Quelques Applications à la Théorie 307–323
des Fonctions. Meddel Lunds Univ Math Semin 3: Kesseböhmer M (1999) Multifrakale und Asymptotiken
1–118 grosser Deviationen. Thesis U Göttingen, Göttingen
Furstenberg H (1967) Disjointness in ergodic theory, min- Kingman JFC (1968) The ergodic theory of subadditive
imal sets, and a problem in Diophantine approximation. stochastic processes. J R Stat Soc B30:499–510
Math Syst Theory 1:1–49 Kleinbock D, Margulis G (1998) Flows on homogeneous
Furstenberg H (1970) Intersections of Cantor sets and spaces and Diophantine approximation on manifold.
transversality of semigroups I. In: Problems in analysis. Ann Math 148:339–360
Sympos Salomon Bochner. Princeton University Press, Kra B, Schmeling J (2002) Diophantine classes, dimension
pp 41–59 and Denjoy maps. Acta Arith 105(4):323–340
Gatzouras D, Peres Y (1996) The variational principle for Ledrappier F (1981) Some relations between dimension
Hausdorff dimension: a survey. In: Pollicott M et al and Lyapounov exponents. Commun Math Phys 81:
(eds) Ergodic theory of Zd actions. Proc of the Warwick 229–238
symposium. Lond Math Soc Lect Note Ser vol 228. Ledrappier F (1986) Dimension of invariant measures.
Cambridge University Press, pp 113–125 Proceedings of the conference on ergodic theory and
680 Ergodic Theory: Fractal Geometry

related topics II (Georgenthal, 1986). Teubner–texte Palis J, Viana M (1988) On the continuity of Hausdorff
Math 94:137–173 dimension and limit capacity for horseshoes. Lecture
Ledrappier F, Misiurewicz M (1985) Dimension of invari- notes in math, vol 1331. Springer
ant measures for maps with exponent zero. Ergodic Pesin Y (1977) Characteristic exponents and smooth ergo-
Theory Dynam Syst 5:595–610 dic theory. Russ Math Surv 32(4):55–114
Ledrappier F, Strelcyn JM (1982) A proof of the estimate Pesin Y, Pitskel B (1984) Topological pressure and the
from below in Pesin’s entropy formula. Ergodic Theory variational principle for noncompact sets. Funct Anal
Dynam Syst 2:203–219 Appl 18:307–318
Ledrappier F, Young LS (1985a) The metric entropy of Pesin Y (1992) Dynamical systems with generalized
diffeomorphisms. I Characterization of measures satis- hyperbolic attractors: hyperbolic, ergodic and topolog-
fying Pesin’s entropy formula. Ann Math 122:509–539 ical properties. Ergodic Theory Dynam Syst 12:
Ledrappier F, Young LS (1985b) The metric entropy of 123–151
diffeomorphisms, II. Relations between entropy, expo- Pesin Y (1993) On rigorous mathematical definition of
nents and dimension. Ann Math 122:540–574 correlation dimension and generalized spectrum for
Lopes A (1989) The dimension spectrum of the maximal dimensions. J Stat Phys 71(3–4):529–547
measure. SIAM J Math Anal 20:1243–1254 Pesin Y (1997a) Dimension theory in dynamical systems:
Mãné R (1981) On the dimension of compact invariant sets rigorous results and applications. Cambridge Univer-
for certain nonlinear maps. Lecture notes in mathemat- sity Press, Cambridge
ics, vol 898. Springer Pesin Y (1997b) Dimension theory in dynamical systems:
Mãné R (1990) The Hausdorff dimension of horseshoes of contemporary views and applications. In: Chicago lec-
surfaces. Bol Soc Bras Math 20:1–24 tures in mathematics. Chicago University Press,
Mauldin RD, Urbański M (1996) Dimensions and mea- Chicago
sures in infinite iterated function systems. Proc Lond Pesin Y, Sadovskaya V (2001) Multifractal analysis of
Math Soc 73(1):105–154 conformal axiom A flows. Commun Math Phys
Mauldin RD, Urbański M (2000) Parabolic iterated func- 216(2):277–312
tion systems. Ergodic Theory Dynam Syst 20(5): Pesin Y, Tempelman A (1995) Correlation dimension of
1423–1447 measures invariant under group actions. Random
Mauldin RD, Urbański M (2002) Fractal measures for Comput Dyn 3(3):137–156
parabolic IFS. Adv Math 168(2):225–253 Pesin Y, Weiss H (1994) On the dimension of determin-
Mauldin DR, Urbański M (2003) Graph directed Markov istic and random cantor-like sets. Math Res Lett 1:
systems. In: Geometry and dynamics of limit sets. 519–529
Cambridge tracts in mathematics, vol 148. Cambridge Pesin Y, Weiss H (1996) On the dimension of deterministic
University Press, Cambridge and random cantor-like sets, symbolic dynamics, and
McCluskey H, Manning A (1983) Hausdorff dimension the Eckmann-Ruelle conjecture. Commun Math Phys
For horseshoes. Ergodic Theory Dynam Syst 3: 182:105–153
251–260 Pesin Y, Weiss H (1997a) A multifractal analysis of equi-
Moran P (1946) Additive functions of intervals and librium measures for conformal expanding maps and
Hausdorff dimension. Proc Cambr Philos Soc 42:15–23 Moran-like geometric constructions. J Stat Phys 86:
Moreira C, Yoccoz J (2001) Stable intersections of cantor 233–275
sets with large Hausdorff dimension. Ann Math 154(1): Pesin Y, Weiss H (1997b) The multifractal analysis of
45–96 Gibbs measures: motivation. Mathematical foundation
Neunhäuserer J (1999) An analysis of dimensional theo- and examples. Chaos 7:89–106
retical properties of some affine dynamical systems. Petersen K (1983) Ergodic theory. Cambridge studies in
Thesis, Free University Berlin, Berlin advanced mathematics 2. Cambridge University Press,
Oseledets V (1968) A multiplicative ergodic theorem. Cambridge
Liapunov characteristic numbers for dynamical sys- Pollicott M, Weiss H (1994) The dimensions of some self
tems. Trans Moscow Math Soc 19:197–221 affine limit sets in the plane and hyperbolic sets. J Stat
Ott E, Sauer T, Yorke J (1994) Part I Background. In: Phys 77:841–866
Coping with chaos. Wiley Ser Nonlinear Sci. Wiley, Pollicott M, Weiss H (1999) Multifractal analysis for the
New York, pp 1–62 continued fraction and Manneville-Pomeau transfor-
Palis J, Takens F (1987) Hyperbolicity and the creation of mations and applications to Diophantine approxima-
homoclinic orbits. Ann Math 125:337–374 tion. Commun Math Phys 207(1):145–171
Palis J, Takens F (1993) Hyperbolicity and sensitive cha- Przytycki F, Urbański M (1989) On Hausdorff dimension
otic dynamics at homoclinic bifurcations. Cambridge of some fractal sets. Stud Math 93:155–167
University Press, Cambridge Ruelle D (1978) Thermodynamic formalism. Addison-
Palis J, Takens F (1994) Homoclinic tangencies for hyper- Wesley
bolic sets of large Hausdorff dimension. Acta Math Ruelle D (1982) Repellers for real analytic maps. Ergodic
172:91–136 Theory Dynam Syst 2:99–107
Ergodic Theory: Fractal Geometry 681

Schmeling J (1994) Hölder continuity of the Holonomy Weiss H (1992) Some variational formulas for Hausdorff
maps for hyperbolic basic sets II. Math Nachr 170: dimension, topological entropy, and SRB entropy for
211–225 hyperbolic dynamical system. J Stat Phys 69:879–886
Schmeling J (1997) Symbolic dynamics for Beta-shifts and Weiss H (1999) The Lyapunov spectrum of equilibrium
self-normal numbers. Ergodic Theory Dynam Syst 17: measures for conformal expanding maps and axiom-A
675–694 surface diffeomorphisms. J Stat Phys 95(3–4):615–632
Schmeling J (1998) A dimension formula for endomor- Young LS (1981) Capacity of attractors. Ergodic Theory
phisms – the Belykh family. Ergodic Theory Dynam Dynam Syst 1:381–388
Syst 18:1283–1309 Young LS (1982) Dimension, entropy, and Lyapunov
Schmeling J (1999) On the completeness of multifractal exponents. Ergodic Theory Dynam Syst 2:109–124
spectra. Ergodic Theory Dynam Syst 19:1–22
Schmeling J (2001) Entropy Preservation under Markov Books and Reviews
codings. J Stat Phys 104(3–4):799–815 Bowen R (1975) Equilibrium states and the ergodic theory
Schmeling J, Troubetzkoy S (1998) Dimension and of Anosov diffeomorphisms. Lecture notes in mathe-
invertibility of hyperbolic endomorphisms with sin- matics, vol 470. Springer
gularities. Ergodic Theory Dynam Syst 18: Eckmann JP, Ruelle D (1985) Ergodic theory of chaos and
1257–1282 strange attractors. Rev Mod Phys 57:617–656
Schmeling J, Weiss H (2001) Dimension theory and Federer H (1969) Geometric measure theory. Springer
dynamics. AMS Proc Symposia Pure Math Ser 69: Hasselblatt B, Katok A (2002) Handbook of dynamical
429–488 systems, vol 1, Survey 1. Principal structures. Elsevier
Series C (1985) The modular surface and continued frac- Katok A (1980) Lyapunov exponents, entropy and periodic
tions. J Lond Math Soc 31:69–80 orbits for diffeomorphisms. Inst Hautes Études Sci Publ
Simon K (1997) The Hausdorff dimension of the general Math 51:137–173
Smale–Williams solenoidwith different contraction Katok A, Hasselblatt B (1995) Introduction to the modern
coefficients. Proc Am Math Soc 125:1221–1228 theory of dynamical systems. Cambridge University
Simpelaere D (1994) Dimension spectrum of axiom-A Press, Cambridge
diffeomorphisms, II. Gibbs measures. J Stat Phys 76: Keller G (1998) Equilibrium states in ergodic theory. In:
1359–1375 London Mathematical Society student texts 42. Cam-
Solomyak B (1995) On the random series S ln (an Erdös bridge University Press, Cambridge
problem). Ann Math 142(3):611–625 Mañé R (1987) Ergodic theory and differentiable dynam-
Solomyak B (2004) Notes on Bernoulli convolutions. In: ics. In: Ergebnisse der Mathematik und ihrer
Fractal geometry and applications: a jubilee of Benoît Grenzgebiete 3, vol 8. Springer
Mandelbrot. Proc Sympos Pure Math, vol 72, Part Mario R, Urbański M (2005) Regularity properties of
1. American Mathematical Society, Providence, Hausdorff dimension in infinite conformal iterated
pp 207–230 function systems. Ergodic Theory Dynam Syst 25(6):
Stratmann B (1995) Fractal dimensions for Jarnik limit sets 1961–1983
of geometrically finite Kleinian groups; the semi- Mattila P (1995) Geometry of sets and measures in Euclid-
classical approach. Ark Mat 33:385–403 ean spaces. In: Fractals and rectifiability. Cambridge
Takens F (1981) Detecting strange attractors in turbulence. University Press, Cambridge
Lecture notes in mathematics, vol 898. Springer Pugh C, Shub M (1989) Ergodic attractors. Trans Am Math
Takens F, Verbitzki E (1999) Multifractal analysis of local Soc 312(1):1–54
entropies for expansive homeomorphisms with specifi- Takens F (1988) Limit capacity and Hausdorff dimension
cation. Commun Math Phys 203:593–612 of dynamically defined Cantor sets. Lecture notes in
Walters P (1982) Introduction to ergodic theory. Springer math, vol 1331. Springer
 Index

A Automorphism, 81, 294, 427, 440

Abelian variety, 589 group, 440
Absolute Cesàro boundedness, 480 Average-distropy-number, 176
Absolute continuity, 322 Axiom A diffeomorphisms, 384
Absorption theorem, 501 Axiom A flow, 586
Abstract Cauchy problem, 481 Axiom A system, 17–18
Adic transformations, 26
Adjacency matrix, 434
AF-equivalence relation, 498, 499, 501–504 B
Affable equivalence relation, 494, 502–504 Baire space, 545
Ageev, Danilenko, 121 Baker’s theorem, 577, 590, 592
Akcoglu–Krengel theorem, 91 Banach lattice, 464
Akcoglu’s theorem, 476 Banach principle, 84
Alexeyev’s theorem, 115–116 Banach spaces, 642
σ-algebra, 238 Baumslag–Solitar group, 493
Algebraic dynamical systems, 590 Benchmarking, 529, 534
Algebraic K-theory, 451, 622 Benedicks and Carleson analysis, 656
Algorithmic complexity, 650 Bergelson’s question, 278
Almost conjugacy, 427, 441, 447 Bernoulli factors, 262
Almost finite actions, 517 Bernoulli flow, 211
Alpern’s lemma, 237 Bernoulli measure, 9
Ambrose-Kakutani theorem, 239 Bernoulli processes, 177
Amenable group, 214 Bernoulli property, 51, 207
Amenable radical, 521 Bernoulli shift, 7, 9, 37, 39, 40, 45, 51, 56, 177, 202,
Analytic number theory, 578 538–539, 643
Analytic set, 529, 570 explicit isomorphisms, 205
Anosov diffeomorphism, 16, 18, 319, 384, 541, 542, 617 isomorphic transformation to, 208
Anosov flow, 16, 18, 586, 618 isomprphism of, 205–207
Anosov systems, 16–17 Kolmogorov–Sinai entropy, 205
Aperiodicity, 293, 303, 304 one sided, 203
Area-preserving smooth flow, 333 Ornstein’s isomorphism theorem, 206–207
Arithmetic function, 293 Berry–Esseen inequality, 649
Arithmetic progression, 582–584, 588, 593 Besicovitch sequences, 478
Arnold Cat Map, 17 Bi-Lipschitz equivalence, 558
Arnold flows, 335, 337 Billiard flow, 12
rescalings of, 361 Billiard map, 12, 13
special representation of, 342 Billiard systems, 12–13
Asymmetric singularities, 343 Binomial coefficients, 172–173
Asymptotically orthogonal powers (AOP), 165, 304 Birkhoff ergodic theorem, 64, 534
Asymptotic equipartition property, 375 Birkhoff pointwise ergodic theorem, 44, 638
Attractor, 6 Birkhoff recurrence theorem, 392–393
reconstruction, 671 Birkhoff’s theorem, 82
Auslander-Ellis theorem, 398 Birkhoff theorem, 639

© Springer Science+Business Media, LLC, part of Springer Nature 2023 683

C. E. Silva, A. I. Danilenko (eds.), Ergodic Theory,
https://doi.org/10.1007/978-1-0716-2388-6
684 Index

Block multiplication, 8 Column, 38

Blokhin flows, 337 Combinatorial freedom, 637, 652
Blum-Hanson (BH) property, 470 Combinatorial number theory, 41–42, 391
Bochner-Herglotz theorem, 111 Compact group rotation, 266
Bohr recurrence, 73, 74 μ-compatible metric, 63, 64
Boltzmann’s ergodic hypothesis, 43 Completely deterministic point, 293
Boltzmann’s hypothesis, 327 Completely-positive entropy, 51, 181
Borel classes, 547–548 Complete order isomorphism, 521, 522
Borel flow, 333 Complex dynamics, 22
Borel hierarchy, 569–571 Complexity, 375
Borel measure, 19 Computability theory, 543
Borel probability, 638 W-condition, 352
measure, 3, 8, 20, 374 Conditional distropy, 175
Borel reducibility, 543 Conditional information function, 175
Borel set, 11 Conditional probability, 175
Borel’s theorem, 9, 581 Cone method, 327
G-boundary, 520, 522 Conjugacy, 5, 451, 507, 508
Boundary map, 520 problem, 437–440
Bounded gaps, 64, 69, 71 Connes-Feldman-Weiss theorem, 492
Bounded prime volume (BPV), 305 Conservative system, 63
Bowen ball, 668 Conservativity, 234–236
Bowen–Dinaburg formulae, 650 Consistent temporal modeling, 42
Bratteli diagram, 494, 499–502 Constrained systems, 435
Bratteli-Elliott-Krieger theorem, 501 Context-free systems, 9
Bratteli-Vershik map, 240 Continued fraction digits, 580
Bratteli-Vershik system, 500 Continued fraction map, 10
Bratteli-Vershik transformations, 498, 500 Continuity properties of the topological dynamics, 654
Brownian motion, 220 Continuous action, 495
Bufetov functionals, 351 Continuous orbit equivalence, 493, 494, 506–510
Bunimovich billiards, 13 Continuous-time flows, 431
Butterfly effect, 334 Contracting subspace, 14
Conze–Lesigne factor, 99
Countable amenable group, 512, 519
Countable equivalence relations, 551
C Countable generator, 238
Cameron-Martin theorem, 243 Countable Lebesgue spectrum, 50, 358
Canonical anticommutation relations (CAR), 284 Counting measure, 217
Cantor minimal systems, 494, 495, 497, 503, 504, 506, Critical points, 644, 645
508, 509, 517 C0-semigroup, 480
Cantor set, 18, 21 Cuntz comparison for open sets, 514, 515
Centralizer, 217 Curtis-Hedlund-Lyndon theorem, 433
Central Limit Theorem, 647–650 Curvature, 577, 585–587
Cesàro averaging, 474 Cylinder set, 6, 37
Cesàro bounded operator, 461–463, 480
Chacón–Ornstein theorem, 90
Chacon’s transformation, 158
Chacon system, 8 D
Chacon transformations, 38 Decay of correlations, 53
Chain recurrent set, 641, 653 Deep neural networks (DNNs), 42, 43
Chaos, 634 Deformation rigidity, 623
Chouqet simplex, 558 Denjoy systems, 504, 671–672
Chowla conjecture, 165, 296–298 Derived from Anosov (DA) attractor, 320
Classical multifractal spectrum, 676 Descriptive complexity, 542
Co-analytic set, 529, 546, 570–572 continuous reductions, 544
Coarse topology, 178 ill-founded tree, reduction of, 545–546
Coded systems, 9 pre-orderings, 544
Cohomological equation, 338 reduction of sets, 543
Coin-flipping measure, 552 reductions of equivalence relations, 543–544
Collet–Eckmann unimodal maps, 649 Descriptive set theory, 569
Index 685

Devaney chaos, 636 flow (R-action), 192–193

C1 diffeomorphism, 653 function, 375
Differentiable dynamics, 382 as function of measure, 375
Axiom A diffeomorphisms, Anosov diffeomorphisms, generators, 176
Sinai-Ruelle-Bowen measures, 383–385 information function, 174–175
Hausdorff dimension of conformal repellers, 385 of invariant measures, 374–375
interval maps, indifferent fixed point, 383 non-amenable groups, 196–198
Markov maps of the interval, 382–383 Ornstein theory, 182
Differentiable rigidity, 620–623 Pinsker-field and K-automorphisms, 181–182
Dimension group, 427, 440 production, 371, 386
Dimension representation, 440 relative entropy, 380–381
Diophantine analysis, 578, 579 Shannon’s noisy channel theorem, 173–174
mixing and additive relations, 590–592 skew product and random walk in random scenery,
orbit growth and convergence, 589–590 182–184
Diophantine approximation, 577, 580, 597, 599, 601, 603, spectrum, 676
609 time reversibility, 176–177
Diophantine exponent, 600, 604 topological, 184–192
Diophantine-like conditions, 353 transformation, 177–180
Diophantine problem, 577, 591 Zd-actions, 193–195
Directional recurrence, 279 Entropy-expanding maps, 653, 657
Dirichlet’s theorem, 599 Entropy theory
Disjointness, 150, 153 definition, 402
from mixing systems, 359 entropy n-tuples, 404–406
of rescalings, 360–361, 363 Kolmogorov system, 403–404
theory, 295 maximal pattern entropy, 408–409
Dispersing billiards, 13 null systems, 408
Dissipative partially hyperbolic diffeomorphisms, 325 sequence entropy, 406–408
Distortion estimates, 320 topological vs measurable, 402–403
Distribution entropy, 171–173 tuples, 409
Doeblin–Fortet, 644 weak horseshoe, 406
Doubling map, 37, 63, 64, 69, 71 Equidistribution, 577, 578, 581, 582
Dyadic (von Neumann-Kakutani) adding machine, 6 Equilibrium distribution, 372
Dynamical systems, 8, 334, 532 Equilibrium measures, 7
complexity in structure theory, 546–550 Equilibrium states, 376–378
complexity of classification theory, 550–556 nonuniqueness of, 378
descriptive complexity (see Descriptive complexity) Equivalence relation
Hamilton dynamics, 534–535 complexity of classification theory, 550–556
measure structure, 537–540 countable, 551
presentations, 540 and Friedman-Stanley jump operator, 553
rotations of unit circle, 535–536 on homeomorphisms, 535
smooth dynamics, 540–542 measure conjugacy, 533
symbolic shifts, 536–537 natural, 535
Dynamical zeta function, 577, 587, 589 Polish Group action, 555
reductions of, 543–544
S1-actions, 553–555
E topological conjugacy, 533
Eckmann–Ruelle conjecture, 675 turbulence, 555
Effros–Handelman–Shen Theorem, 500, 502 with uncountable classes, 551–553
Elliott invariant, 505, 515, 516, 518 Ergodic average, 579, 581, 582
Ellis-Namakura theorem, 398 Ergodic decomposition, 45, 46, 48, 69, 237–238, 394
Embedding, 427, 433 theorem, 537
theorem, 452 Ergodic flow, 333
Endomorphism, 81, 675 Ergodic hypothesis, 36, 578, 579
Energy surface, 36 Ergodicity, 36, 43–46, 394
Ensemble averages, 638 of Gaussian cocycles, 280
Entropy, 51, 92, 170–171, 176, 269–272, 293, 435 of powers, 279
Bernoulli processes, 177 Ergodic measure preserving systems, 549, 550
determinism and zero-entropy, 180–181 Ergodic properties, 222
distribution, 171–173 Ergodic Ramsey theory, 71, 98
686 Index

Ergodic Ramsey theory and recurrence, 582–583 Ergodic transformation, 2, 540

polynomialization and IP-sets, 584 Étale equivalence, 493, 495–499, 501, 502, 504
sets of primes, 584 Euclidean coordinates, 12
topology and coloring theorems, 583–584 Euclidean space, 14
Ergodic theorems, 532, 579, 582 Even shift, 431
definition, 79 Expanding maps, 319
local, 86 Expanding subspace, 14
mean, 579
for measure-preserving maps, 81–85
F
for non-amenable groups, 101
Factor map, 183, 197, 427, 433, 529, 537
pointwise, 88–90, 579
Farey map, 10
subadditive and multiplicative, 90–92
Fatou set, 22
Ergodic theory, 1, 2, 299, 446–447, 597, 598, 667
Filtering problems, 160
adding machines, 6
Finitary isomorphisms, 210
Adic transformations, 26
Finite equivalence, 427, 441, 442
Anosov systems, 16–17
Finite lattice, 373
Axiom A system, 17–18
Finite partition, 16
Borel probability measure, 401
Finite set, 6
chaos and complexity theory, 400–401
Finite-state coding theorem, 445
characteristic factors, 414–416
Finite statistical description, 639–641, 644, 646, 647
complex dynamics, 22
First digits, 581
continued fraction map, 10
First-return map, 24
cutting and stacking, 25–26
Fock space, 221
disjointness, 399
Foïaş-Strătilă property, 267
entropy (see Entropy)
Foiaş-Stratila theorem, 126–128, 229
factors, 23
Follower sets, 9
Farey map, 10
Følner sequence, 93, 214
f-expansions, 10–11
Fractal geometry, 665, 669
full shifts and shifts of finite type, 6–8
Free action, 493, 496, 517–519
Gaussian systems, 11
Friedman-Stanley jump operator, 553
geodesic flow on manifold of negative curvature, 15
Full shift, 427, 431, 447
Hamiltonian systems, 11–13
Full tent map, 635, 636, 638, 639, 642, 650, 652
history, 390
Furstenberg boundary, 493, 520, 521
horocycle flow, 15–16
Furstenberg correspondence principle, 584
induced transformations, 24–25
Furstenberg disjointness, 308
infinite ergodic theory, 22–23
Furstenberg ergodic averages, 71
interval exchange maps, 6
Furstenberg systems, 294, 303, 307
inverse limits, 27
Furstenberg theorem, 280
joinings, 28
Furstenberg-Zimmer’s structure theorem, 412
Markov partitions and coding, 16
multiple ergodic averages, 412–414
natural extension, 27–28 G
non-uniformly hyperbolic systems, 19 GAGs, 228, 229
partially hyperbolic dynamical systems, 18–19 Gaussian and Poisson systems, 226
physically relevant measures and strange attractors, Gaussian centralizer, 225
19–22 Gaussian-compact factor, 226
problems, 391 Gaussian distribution, 219
products, 23 Gaussian dynamical system, 221, 592
recurrence (see Recurrence) Gaussian factor, 224
rigid rotation of compact group, 5–6 Gaussian-Kronecker automorphism, 124
Rokhlin’s lemma, 26–27 Gaussian-Kronecker system, 228
β-shifts, 11 Gaussian process, 1, 217
skew products, 24 Gaussian random variable, 11
smooth expanding interval maps, 9–10 Gaussian space, 217
smooth uniformly hyperbolic diffeomorphisms and Gaussian stationary processes, 219
flows, 14–18 Gaussian systems, 11
structure theorems, 411–412 Gaussian transformations, 243, 265
subshifts, 8–9 Gauss–Kuz0 min law, 580
suspension flows, 25 Gauss map, 581
Index 687

Gauss measure, 10, 581 Hidden Markov models (HMMs), 42

Gelfand-Naimark theorem, 522 Higher dimensional shift space, 427, 447–450
General averaging methods, 474–476 Hilbert space, 36, 39, 52, 68, 492, 511, 521
Generalized Gaussian systems, 220 Hille-Yosida theorem, 479
Generalized Krengel class, 260 Hindman’s theorem, 582–584
Generalized Morse systems, 8 Hoeffding’s inequality, 43
General nonsingular Bernoulli shifts, 260 Hölder functions, 9
Generic point, 294 Holonomy invariance, 640
Geodesic(s), 585–587 Homeomorphism, 494–498, 501–504, 506–508, 511, 512,
flow, 578, 580, 585–587 514–517
Geometric interpretation, 220 Homoclinic bifurcations, 669–670
Gibbs distribution, 372 Homogeneous space, 597, 600, 601, 606
Gibbs–Markov map, 650 Homorphisms, 39
Gibbs property, 372 Hopf Argument, 321–322
class of regular potentials, 379 Hopf’s ergodic theorem, 95
Perron-Frobenius theorem, 379–380 Horocycle flow, 15–16, 618
relative entropy, 380–381 Horseshoe maps, 17
subshifts of finite type, 378–379 Host-Kra-Maass theorem, 417
Gibbs states, 381 Host’s theorem, 162
Glimm-Effros dichotomy, 551–553 Hurewicz pointwise Ergodic theorem, 236
Glimm-Effros theorem, 238–239 Hyperbolic (Anosov) dynamical system, 616
Global-global mixing, 257 Hyperbolic billiards and hard sphere gases, 327–328
Global rigidity, 625–627 Hyperbolic flows, 586
Global simplicity, 652, 653 Hyperbolicity, 4, 14, 53, 633, 641–643, 646, 649, 650,
Golden mean shift, 431 653, 654, 656, 657
G-operator system, 521, 522 Hyperbolic map, 589
Graph-joining, 273 Hyperbolic matrix, 618
Group endomorphisms, 587–588 Hyperbolic measure-preserving transformation, 4
Group extensions of dynamical systems, 24 Hyperbolic set, 14
Group measure space construction, 492, 507 Hyperbolic systems, 318, 641
Hyperbolic toral automorphism, 586
Hyperplane absolute winning (HAW) property, 608
H
Haar measure, 5
Hahn–Banach theorem, 82 I
Hajós–Minkowski theorem, 602 Idempotent ultrafilter, 583
Halmos topology, 537 Induced transformation, 24–25, 235
Halmos-von Neumann theorem, 395, 539 Infinite direct product of finite types (IDPFT), 243
Hamana boundary, 493, 520–522 Infinite divisibility, 217
Hamiltonian function, 12 Infinite ergodic index, 253
Hamiltonian systems, 11–12 Infinite ergodic theory, 22–23
billiard systems, 12–13 Infinite iterated function system, 671
KAM-systems and stably non-ergodic behavior, 13 Infinitely divisible, 217
Hamilton’s equations, 12, 534 Infinite Markov shifts, 242
Hamming metric, 173 Infinite measure preserving system, 65, 75
Harrington-Kechris-Louveau dichotomy, 552 Infinite partition, 10
Hausdorff a-measure, 64 Infinite sequence, 375
Hausdorff dimension, 348, 597, 600, 603, 606, 608, 638, Infinite set, 582
642, 645, 647, 667, 669 Information theory, 445
Hausdorff space, 497, 498, 506–508, 511, 513–515, 517, Intermediate convergents, 10
520, 522 Intermittent maps, 21–22
Height, 38 Internal disjointness, 303, 304
Hellinger distance, 242 Intersective polynomials, 70
Hénon diffeomorphisms, 22 Intersymbol interference, 444
Hénon family of maps, 329 Interval exchange maps, 6
Hénon-like maps, 646 Interval exchange transformations (IETs), 38, 131–136,
Hénon map, 656 340–341, 351, 362
Herglotz’s theorem, 67 joining properties, 360
Heterodimensional homoclinic intersections, 654 spectral measures of, 357–358
688 Index

Interval maps, 383, 643–645, 649, 652, 654–657 Kneading invariants, 652, 653, 657
Invariant integrals, 462 Kochergin flows, 354
Invariant measures, 294, 374 Kolmogorov–Arnold–Moser (KAM) theory, 314
entropy of, 374–375 Kolmogorov automorphism, 182
Invariant probability measure, 635, 639, 644, 650–652 Kolmogorov-Chaitin complexity, 371
Invariants of étale equivalence relations, 497 Kolmogorov extension theorem, 37
Inverse limits, 27 Kolmogorov process (K-process), 181–182
Invertible measure preserving system, 66, 70, 71 Kolmogorov–Sinai entropy, 205, 223
Ising chain, 382 Kolmogorov-Sinai theorem, 177
Ising model, 370 Kontsevich-Zorich conjecture, 350
Isomorphism, 39, 493–498, 501, 502, 505, 507, 508, 510, Koopman mixing, 257
511, 516, 521, 522 Koopman operator, 36, 39, 221, 539
Iterated function systems, 669 Koopman representation
Iterated Kreiss condition, 479 Alexeyev’s theorem, 115–116
ITPFI transformation, 251–252 Alpern-Tikhonov topology, 123
Banach problem, 112
disjoint, 114
J entropy, 114
Jacobs-deLeeuw-Glicksberg decomposition, 477 ergodic dynamical system, 112
Jakobson one-dimensional theorem, 647 Gaussian-Kronecker automorphism, 124
Jakobson’s method, 329 Lamperti theorem, 113
Jewett–Krieger theorem, 402, 506 Markov operators, 113
Jewett-Krieger type realization, 494, 506 maximal spectral types, 114
Jiang-Su algebra, 516 multiplicity (see Multiplicity function)
Jiang-Su stability, 516 pairwise independence property, 114
Joinings, 294, 298 spectral isomorphism, 112
Joinings in ergodic theory Thouvenot. Funny rank one, 123
and conjectures in number theory, 164 Koopman unitary operator
definition, 149, 150 associated with nonsingular Gaussian transformation,
disjointness, 153 269
and factors, 155 associated with nonsingular Poisson transformation,
filtering problems, 160 268–269
future research, 166 for nonsingular system, 267–268
and isomorphism, 154 Kreiss bounded operators, 479
Markov intertwinings, 156 Krengel class, 259–260
and multiple ergodic averages, 163 Krengel entropy, 269
Ornstein’s and Krieger’s theorems, 161 Krengel-Pinsker factor, 270
and Rohlin’s multifold mixing, 161, 163 Krengel-Sucheston concept of mixing, 257
self-joining (see Self-joinings) Krickeberg mixing, 255
set of, 152 Krieger embedding theorem, 442–443
Julia set, 22, 213 Krieger generator theorem, 177
Krieger’s theorem, 161, 246, 494, 505
Kronecker factor, 48, 395
K Kronecker subset, 217
Kac’s formula, 235 Krylov–Bogoliubov theorem, 346, 537, 638
Kakutani equivalence, 211–212, 562–563, 573
Kakutani-Rokhlin tower decomposition, 511, 517, 518
Kakutani transformation, 241 L
Kaplan–Yorke formula, 674 Lacunary sequence, 66
Katok formulae, 650 Lagrange multiplier, 372
Katok fundamental class, 345 Laplace–Beltrami operator, 585, 586
K-automorphism, 258 Large deviations principles, 371
Keane conjecture, 347 Lasota–Yorke inequality, 644
Keane’s condition, 339 Lebesgue density theorem, 639
Khintchine–Groshev Theorem, 600 Lebesgue measure, 5–7, 9, 10, 18, 19, 22, 37, 45, 86, 316,
Khintchine’s recurrence theorem, 75 599, 605
Khintchine’s transference principle, 600 Lebesgue number, 185
Kingman’s theorem, 91 Lebesgue probability space, 27, 39, 42
King’s weak closure theorem, 278 Lebesgue space, 23, 46
Index 689

Lebesgue spectrum, 116–117, 358 Maximal analytic equivalence relation, 558

Ledrappier–Young formula, 646 Maximal inequality, 82
Light mixing, 49 Maximum entropy, 639, 640
Lindenstrauss’s theorem, 102 measure, 640, 641, 643, 646, 652, 657
Linear flow, 362 Maxwell-Boltzmann-Gibbs formula, 370
special representation of, 342 Mobius orthogonal, 1
spectral measures of, 357–358 Mean and pointwise ergodic theorems, 236–237
on torus, 335 Mean dimension for a single homeomorphism, 511
on translation surfaces, 335–336 Mean dimension for topological dynamical systems, 511
Linearization, 346 Mean ergodic theorem, 460–464, 579
Liouvillean rotation number, 572 Measurable distality, 411
Liouville measure, 15 Measure conjugacy, 550
Liouville’s fomula, 12 Measure distal, 550
Liouville’s theorem, 534 Measure isomorphism, 559
Little shift equivalence problem, 440 Measure of maximal entropy, 381
Livshits’s criteria, 620 Measure-preserving dynamical system, 150
Local complexity, 651 Measure-preserving homomorphism, 39
Local entropy theory, 391 Measure-preserving systems, 4
Local ergodicity, 642 Measure-preserving transformation, 36–40, 42–56
Local ergodic theorem, 86 Measure rigidity, 627–628
Local limit theorem, 649 Measures of maximal entropy, 427
Locally Hamiltonian flow, 333, 334, 336–337, 362–363 Measure space, 184
genericity notions for, 345 Measure-theoretic dynamical system, 294
special representation of, 342–343 Measure-theoretic entropy, 36, 578, 640, 651
spectra of, 358–359 Measure-theoretic isomorphism, 39, 40
Local product structure, 16 Mertens’ theorem, 586, 588
Local rigidity, 623–625 Meshalkin’s map, 178–179
Local stable manifolds, 14 Metric Diophantine approximation, 597, 598
Local unstable manifolds, 14 Metric entropy, 8, 640
Local variational principle, 652 Metric number theory, 597, 599
Logarithmic Furstenberg systems, 297, 306, 307 Mildly mixing flow, 333
Logarithmic singularities, 342 Mild mixing, 48, 49, 279, 353
Lower entropy factor theorem, 443–444, 452 Minimal action, 493–496, 498, 501, 503–505, 507, 511,
Lusin’s separation theorem, 571 514, 515, 518–520
Lyapunov exponent, 19, 633, 636, 637, 639, 642, 643, 646, Minimal components, 339
650, 655, 674 Minimal homeomorphism, 529, 533, 550, 566
Lyapunov spectra, 676 Minimality, 338
Minimal radon uniquely ergodic model, 239
Minimal self-joinings, 8, 149, 158, 353
M Minkowski measure, 10
Mackey Borel structure, 138 Mixing, 578, 586, 589–593
Maharam extension, 247–248 flow, 333
Mahler’s compactness criterion, 601 light, 49
Mahler’s Criterion, 602 mild-mixing, 48, 49
Mandelbrot set, 22 partially, 49
Mañé Stability Theorem, 641 strong, 36, 39, 41, 45–51, 55
Markov chain, 7, 40, 45, 47, 51, 446–447, 644 weak, 46–49, 52, 53, 55, 56
Markov-Feller operators, 463, 468 Möbius function, 420
Markov intertwinings, 149, 156 aperiodicity, 304
Markov maps, 382–383 logarithmic Furstenberg systems, 306, 307
Markov matrix, 179 multiplicativity, 303–304
Markov odometers, 240 short intervals, 304–306
Markov partition, 16, 430, 586, 587, 641 Möbius orthogonality, 294, 295, 298–302
Markov process, 7, 179 Möbius randomness principle, 164
Markov random fields, 214 Modulated ergodic theorem, 461–464, 476
Markov shift, 7, 37, 75, 179–180, 213, 633, 643, 653 Moment of order, 217
Masur’s criterium, 348 Morse function, 345
Masur-Veech measure, 345 Morse shift, 432, 436
Matrix subshift, 448 Morse-Smale diffeomorphism, 541, 563
690 Index

Multifold mixing, 162 Markov odometers, 240

Multifractal analysis, 676, 677 Markov shift, 242
Multifractal decomposition, 676 mean and pointwise ergodic theorems, 236–237
Multifractal spectrum, 676 minimal radon uniquely ergodic model, 239
Multiple averages, 420 multiple and polynomial recurrence, 258–259
Multiple recurrence, 69–71, 583 natural extensions of nonsingular endomorphism, 243
Multiple weak mixing, 55 nonsingular product odometers, 240
Multiplicative orthogonality criterion (MOC), 303 nonsingular restricted orbit equivalence, 272
Multiplicativity nonsingular transformation, 234
AOP property, 304 orbit theory (see Orbit theory)
internal disjointness, 303, 304 Parry’s generalization, 271
Multiplicity function Poisson suspension, 242–243, 263–264
cocycle approach, 119–120 rational ergodicity and rational weak mixing, 254–255
Gaussian and Poissonian automorphism, 120 representations of ergodic flows, 239
maximal spectral type, 119 smooth nonsingular transformation, 276–277
Rokhlin multiplicity problem, 121 spectral theory, 265–269
Multi-valued Hamiltonian flow, 336 topological groups, 244–246
tower transformation, 240
weak mixing, 252–254
N Nonsingular flows, 239
Natural extension, 27–28 Nonsingular joining, 273
Newhouse phenomenon, 669 Nonsingular Markov shifts, 242, 262–263
Newhouse theorem, 657 Nonsingular Poisson suspensions, 242–243
Newtonian mechanics, 4 Nonsingular product odometers, 240
Newton’s laws of motion, 4 Nonsingular subadditive Ergodic theorem, 237
Nilmanifold, 592 Non-spectral approach, 481
Nilrotation, 295 Non-uniform expansion/contraction, 645, 646
Noisy channel theorem, 173–174 Non-uniformly hyperbolic systems, 19
Non-compact surfaces, 363 Nonzero Lebesgue measure, 635
Non-degenerate elliptic fixed point, 13 Nonzero topological entropy, 635
Nonemptiness problem, 448 Normal number, 580, 581
Nonequilibrium steady states, 386 Novikov problem, 334, 335
Non-hyperbolic systems, 5 Nuclear dimension, 515
Non-invertible transformations, 212 Number theory, 598, 670
Non-lacunary sequence, 67, 74
Non-reversibility, 360
Nonsingular Bernoulli shifts, 241–242 O
Nonsingular Chacón map, 241 2-odometer, 6
Nonsingular dynamical system, 233, 234 G-odometers, 508
Nonsingular endomorphism, 243 Off-diagonal self-joinings, 150, 157
Nonsingular ergodic theory, 233 Operator ergodic theory, 460
applications, 279–284 general averaging methods, 474–476
Bernoulli shifts, 241–242 mean ergodic theorem, 461–464
Chacón map, 241 modulated ergodic theorem, 461–464
conservativity and ergodicity, 234–236 rates of convergence, 464–465
directional recurrence, 279 stability, 471–473
dynamical properties of Bernoulli and Markov shifts, strong Cesàro convergence, 468–469
259–263 uniform ergodic theorem, 465–468
dynamical properties of IDPFT systems, 259 weak stability and mixing, 469–471
entropy, 269–272 Orbit complexity, 650
ergodic decomposition, 237–238 on the set of measures, 651
future research, 284–285 Orbit-counting
Gaussian transformations, 243, 265 exotic orbit growth, 588–589
generator, 238 and geodesics, 585–587
Glimm-Effros theorem, 238–239 group endomorphisms, 587–588
IDPFT transformations, 243 Pólya–Carlson dichotomy, 589
joinings and factors, 272–276 Orbit Dirichlet series, 578, 588
K-automorphism, 258 Orbit equivalence of minimal actions of a finitely
Krickerberg mixing, 256 generated abelian group, 503
Index 691

Orbit equivalence of minimal ℤ-actions on the Cantor Poincaré recurrence, 64–65, 579, 583
set, 503 theorem, 390, 394
Orbit theory, 246 Poincaré sequences, 394
cocycles of dynamical systems, 250–251 Point process, 217
continuous orbit equivalence, 248–249 Pointwise dimension, 667
full groups, 246–247 Pointwise ergodic theorems, 88–90, 579
ITPFI transformation, 251–252 Poisson entropy, 270–271
Maharam extension, 247–248 Poisson factor, 226
normalizer of the full group, 249–250 Poisson flow, 262
Ordered Bratteli diagram, 500 Poisson point process, 1, 218–220
Orientation preserving, 6 Poisson processes, 219
isometries, 15 Poisson suspension, 220–223, 263–264, 281
Ornstein’s isomorphism theorem, 206–207 Polish group action, 530, 544, 551, 555, 557–558
Ornstein’s theorem, 161, 206 Polish space, 530–532, 545, 546, 549, 552, 553, 569, 570,
Ornstein’s theory, 56, 182 572
Ornstein-Weiss tiling theorem, 511, 517 Pólya–Carlson dichotomy, 589
Orthogonality of sequences, 295 Polynomial deviations of ergodic averages, 350
Orthonormal basis, 50 Polynomial rate (PR), 305
Oseledec’s theorem, 325 Pontryagin duality, 540
Oseledec and Pesin theory, 674 Positive definite sequence, 67
Oseledets theorem, 642 Positive upper density, 582
Power-bounded operator
on Banach space, 471, 473
P ergodic, 464, 477
Pairwise independence property (PID), 114 on Hilbert space, 473
Pairwise independently determined system on reflexive Banach space, 462
(PID), 162 Power series, 465
Palis conjecture, 640 Predecessor sets, 9
PAPs, 229 Pressure functional, 668
Parabolic systems, 671 Prime number theorem, 585, 586
Parameter exclusion, 329 Prime shift, 431
Parry entropy, 270 Prime volume, 305
Partial hyperbolicity, 18, 324 Probabilistic limit theorems, 648
Partially hyperbolic dynamical systems, 18 Probability distribution, 374
Anosov flows, time-one maps of, 18 Probability measure, 8
uniformly hyperbolic systems, compact group Probability space, 4, 39, 43, 44, 170–172
extensions of, 18 Probability vector, 7, 171, 175, 177
Partially mixing, 49 Product measure, 7
Pascal adic transformation, 56 Product transformation, 40
Periodic components, 339 Prouhet-Thue-Morse (PTM) sequence, 8, 432
Periodic points, 381 Proximal action, 520
Perron-Frobenius theorem, 379–380
Perron-Frobenius theory, 446
Perron numbers, 452 Q
Pesin’s theory of dimension-like characteristics, 668 Qualitative theory, 533, 540
Pesin theory, 19, 325, 642, 656 Quantitative nondivergence method, 608
Phase space, 12, 35 Quantitative Poincaré recurrence, 64–65
Phase space average, 638, 648 Quantitative weak mixing, 358
Phase-structure grammars, 9 Quasi-compact operator, 466
Physical measure, 639, 656 Quasi-ergodic hypothesis, 36, 577, 579
Piecewise C2 expanding maps, 10 Quasi-genericity vs. logarithmic quasi-genericity, 295
Piecewise monotone maps, 652 Quasi-hyperbolic automorphisms, 590
Pigeon hole principle, 236
Pinsker algebra, 214
Pinsker σ-ALGEBRA, 51 R
Pinsker-algebra theorem, 182 Radius of comparison, 512–514
Pinsker factor, 181 Radon-Nikodym derivative, 22, 244, 258
Pinsker field, 181 Radon–Nikodym theorem, 85
Poincaré map, 12, 22 Ramsey theory, 420, 582–584
692 Index

Random dynamical systems, 24 Rudolph’s counterexample machine, 208

Random walks Rudolph structure theorem, 207
boundaries of, 282 Ruelle’s pressure formula, 669
recurrence for, 281 Ruelle transfer operator, 644
Rank function, 513 Run-length limited shift, 428
Rank one attractors, 647 Ryzhikov’s theorem, 163
Rank-one transformations, 38, 39, 45, 52, 55
Rates of convergence, 464–465 S
Rational ergodicity, 254–255 Saddle connection, 339
Rational maps, 213 Sarnak’s conjecture, 165
Rational spectrum, 48 Chowla conjecture, 299
Ratner property, 355 Möbius orthogonality, 301
Ratner’s technique, 360 (Slog) vs. (Clog), 301
Realization problem, 561 strategies, 302
Recurrence, 578, 579, 584, 590 strong MOMO property, 300
combinatorics and number theory, 71–73 (S) vs. (Slog), 302
multiple, 69–71 zero entropy system, 298
quantitative Poincaré recurrence, 64–65 Second quantization operators, 221, 222
set of, 583 Self-joinings, 150, 218
subsequence, 66–69 and commuting transformations, 157
topological multiple, 582 minimal, 158
Recursive reduction, 536 relative properties with respect to factor, 160
Reduced group C⁎-algebra, 522 simple system, 159
Reduction property, co-analytic set, 572 Semisimplicity, 160
Regular expressions, 9 Sensitivity on initial conditions, 633, 637, 643
Regular languages, 9 Shannon-McMillan-Breiman theorem (SMB-theorem),
Regular summability matrix, 474 93, 180, 375
Relative entropy, 380–381 Shannon’s noisy channel theorem, 173–174
Renyi inequality, 254 Shape, 515, 517
Representations, realizations and genericity, 54–55 Shearing mechanism, 355, 360
Restricted orbit equivalence, 212 Shift equivalence, 428, 439, 440
Reversibility, 360 strong, 428, 438, 439
Riemann hypothesis, 585 Shift map, 37
Riemannian manifold, 14, 15, 18 Shift of finite type, 428, 433–435
Riemannian metric, 15 β-Shifts, 11
Riemann sphere, 22 Shift spaces, 428
Riemann zeta function, 585 higher dimensional, 447–450
Rigid flow, 333 invariant measures and entropy, 373–375
Rigidity, 615 Ising chain, 382
differentiable, 620–623 Markov chains, finite alphabets, 381–382
future research, 628 measure of maximal entropy and periodic points, 381
global, 625–627 run-length-limited, 444
local, 623–625 shift of finite type and sofic shifts, 433–435
measure, 627–628 and sliding block codes, 431–433
sequence, 49, 279 Short intervals, 303–306
Rigid rotation, 5 S1-actions, 553–555
Rigid sequence, 128–130 Simplicity, 150
Ritt condition, 478 Sinai billiards, 13
Road-colorings, 441 Sinai–Ruelle–Bowen measure, 323, 371, 384, 634, 635,
Road problem, 427, 442 639, 641–643, 646, 647, 655
Rokhlin cocycle, 117, 121–122 Singularities, 336
Rokhlin extensions of rotations, 306 Skew products, 24
Rokhlin lemma, 514, 517 group extensions of dynamical systems, 24
Rokhlin property, 244 random dynamical systems, 24
Rokhlin’s lemma, 26 Slit torus, 349
Rosendal theorem, 558 Smooth ergodic theory
Rotation number, 535, 536 absolute continuity, 322
Roth’s theorem, 75 Anosov diffeomorphism, 319
Roth-type condition, 350 critical points and singularities, 327
Index 693

DA attractors, 320 Stationary stochastic process, 4, 81

definition, 313 Statistical learning theory, 42
dissipative partially hyperbolic diffeomorphisms, 325 Statistical properties, 648, 650
distortion estimates, 320 Statistical stability, 634, 654, 655
examples of, 317–318 Stirling’s formula, 172
expanding maps, 319 Stochastic matrix, 7
Hopf Argument, 321–322 Stochastic perturbations, 655
hyperbolic billiards and hard sphere gases, 327–328 Stochastic stability, 634, 655
hyperbolic blocks, 326 Stolz region, 478
hyperbolic systems, 318 Stone-Weierstrass theorem, 296
interval maps and parameter exclusion, 328 Stopping point, 338
Lebesgue measure and local properties of volume, 316 Strict comparison, 513, 515–517, 519
near critical diffeomorphisms, 329 Strict inequality, 650
nonuniformly hyperbolic attractors, 326 Strong Cesàro convergence, 468–469
partial hyperbolicity, 324 Strong mixing, 36, 39–41, 45–51, 55
partially hyperbolic conservative diffeomorphism, 325 Strong MOMO property, 307
Sinai-Ruelle-Bowen measures, 323 Strong orbit equivalence, 494, 505, 506
Smooth flow, 333 Strong pinsker conjecture, 56
cohomological equation for, 349–350 Structural stability, 14, 623, 634, 653, 654
Smoothly isomorphic flows, 333 Structure theory, 533
Smooth manifolds, 532, 540 Sturmian systems, 8
Smooth measure-preserving transformations, 561–562 Subsequence recurrence, 66–69
Smooth nonsingular transformation, 276–277 Subshift, 8–9, 295
Smooth reparametrization, 338 finite type, 634, 641, 652
Smooth transformations, 563 Subspace theorem, 590
Sofic shift, 428, 433–435 Super-rigidity, 557
Sofic systems, 9 Suspension flows, 25
Solenoids, 17–18 Switchable Ratner property, 355
Space average, 579, 581 Symbolic dynamics, 373–374, 428, 451–453, 637,
Spacers, 38 641–643, 652, 657
Spectral decomposition theorem, 17 origins of, 429–431
Spectral disjointness, 359 Symbolic extension, 652
Spectral measure, 218 Symbolic modeling approach, 451
Spectral properties, 36, 40 Symbolic shifts, 536–537
Spectral theorem, 112 Symmetric logarithmic, 356–357
Spectral theory, 82, 265–269 Synchronized automata, 304
Banach problem, 118–119 Systems with discontinuities, 643, 644
Borel probability space, 116 Szemerédi’s theorem, 69, 71, 72, 583, 584, 593
3-dimensional Heisenberg group, 136–138
Foiaş-Stratila theorem, 126–128
Gaussian dynamical systems, 125 T
Heisenberg odometer, 138 Tail entropy, 651
interval exchange transformations, 131–136 Tail field, 181
Lebesgue spectrum, 116–117 Tent map, 635, 636, 638, 639, 642, 650, 652
parabolic dynamical system, 130 Thermodynamic formalism, 453, 671
perturbations of affine cocycles, 119–120 Thermodynamic formalism, for finite systems, 371–372
properties, 119 equilibrium distributions and Gibbs property, 372–373
time-changes of nilpotent system, 131 finite lattice, 373
time-changes of unipotent system, 130–131 Time-change flow, 338
weighted operators analysis, 116 special representations, 344
zero entropy automorphism, 128 Time-reparametrizations, 338
Stability, 653 Time-zero partition, 176
Stable maps, 645 Toeplitz systems, 9
Standard Gaussian systems, 219 Toms-Winter conjecture, 516, 517, 519
State splitting, 428, 438, 442 Topological complexity, 639, 640, 657
algorithm, 452 Topological conjugacy, 428, 433, 563–567
Stationary action, 282 Topological distal, 548
Stationary Gaussian process, 219 Topological dynamical system, 295, 470, 493–495,
Stationary process, 81, 218 506–508, 511, 514
694 Index

Topological dynamics Uniform Rokhlin property, 514, 515

chaos and complexity theory, 400–401 Unimodal maps, 20–21
characteristic factors, 416–418 Unimodular lattice, 601
disjointness, 399–400 Unique ergodicity, 45
Ellis semigroup, 397–399 Unique invariant measure, 651, 655
history, 390 Uniquely ergodic model, 295
Kronecker systems, 403–404 Unique trace property, 520, 522, 523
measurable mild mixing, 397 Unstable manifold, 14
measurable mixing, 396 Upper density, 67, 71, 72, 74
measurable weak mixing, 396
minimality and transitivity, 392–395
multiple ergodic averages, 418–420 V
problems, 391 Van der Corput’s method, 68
recurrence, 391 Variational principle, 187, 189, 377, 650
structure theorems, 410–411 Vertex shifts, 434
topological weak mixing, 396 von Neumann algebras, 283, 492, 493
weak disjointness, 400 Von Neumann ergodic theorem, 68
Topological entropy, 4, 184–192, 428, 435, 446, 578, 586, von Neumann flows, 344, 353
589, 637, 639, 640, 643, 650–652, 654, 657, 668 rescaling of, 361
Topological genericity, 634, 641 von Neumann mean Ergodic theorem, 44, 236
Topologically transitive homeomorphism, 533 von Neumann’s theorem, 82
Topologically weak mixing, 352
Topological Markov chain, 371
Topological Markov shifts (TMS), 187 W
Topological mixing, 396 Wandering set, 22
Topological multiple recurrence, 582 Weak Bernoulli property, 40
Topological orbit equivalence for actions, 492 Weak mixing, 39, 41, 46–49, 52, 53, 55, 56, 252–254
Topological Pinsker factor, 404 flow, 333
Topological pressure, 668 full measure of, 352
Topological recurrence, 66, 73 in linear flows and IETs, 351
Topology, 55 logarithmic singularities, 357
Toral automorphism quantitative, 358
hyperbolic, 586, 587 Weak Pinsker conjecture, 56
quasi-hyperbolic, 587, 588 Weak rational ergodicity, 254
Totally non-symplectic (TNS) condition, 622 Weak stability and mixing, 469–471
Tower, 514, 517, 518 Weyl chamber flow, 619
map, 25 Weyl rationally almost periodic (WRAP), 304
transformation, 240 Weyl’s criterion, 67
Trachtman’s approach, 442 Wiener lemma, 111
Transfer function, 338 Wiener–Wintner theorem, 96, 476, 477
Transition matrix, 378
Transition probability, 461
Translation surface, 335–336 Y
Trigonometric polynomial sequence, 476 Yomdin theory, 646
Turbulence, 555 Young’s dimension formula, 673
Twist map, 13

Z
U Zeeman’s tolerance stability conjecture, 654
Uniform approximation, 599 Zero-entropy, 180–181
Uniform ergodic theorem, 465–468 continuous interval maps, 304
Uniform hyperbolicity, 641, 642, 650, 653, 654, 656 detection, 307
Uniform Kreiss resolvent condition, 479 Gaussian system, 224
Uniformly expanding map, 641, 649, 652, 653, 655 Zeta function, 428, 436, 437, 439, 577, 585–587, 589
Uniformly recurrent subgroup, 523 ℤd-odometers, 509