Association for Women in Mathematics Series

Marie José Bertin

Alina Bucur
Brooke Feigon
Leila Schneps Editors

Women in
Numbers Europe
Research Directions in Number Theory
Association for Women in Mathematics Series
Volume 2

More information about this series at http://www.springer.com/series/13764

Association for Women in Mathematics Series

Focusing on the groundbreaking work of women in mathematics past, present, and

future, Springer’s Association for Women in Mathematics Series presents the latest
research and proceedings of conferences worldwide organized by the Association
for Women in Mathematics (AWM). All works are peer-reviewed to meet the highest
standards of scientific literature, while presenting topics at the cutting edge of pure
and applied mathematics. Since its inception in 1971, The Association for Women in
Mathematics has been a non-profit organization designed to help encourage women
and girls to study and pursue active careers in mathematics and the mathematical
sciences and to promote equal opportunity and equal treatment of women and girls
in the mathematical sciences. Currently, the organization represents more than 3000
members and 200 institutions constituting a broad spectrum of the mathematical
community in the United States and around the world.
Marie José Bertin • Alina Bucur • Brooke Feigon
Leila Schneps
Editors

Women in Numbers Europe

Research Directions in Number Theory

123
Editors
Marie José Bertin Alina Bucur
Number Theory Department of Mathematics
Jussieu Institute of Mathematics University of California, San Diego
Paris, France La Jolla, CA, USA

Brooke Feigon Leila Schneps

Department of Mathematics Algebraic Analysis
The City College of New York Jussieu Institute of Mathematics
New York, NY, USA Paris, France

Association for Women in Mathematics Series

ISBN 978-3-319-17986-5 ISBN 978-3-319-17987-2 (eBook)
DOI 10.1007/978-3-319-17987-2

Library of Congress Control Number: 2015946869

Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 2015
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, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media (www.

springer.com)
Preface

The Women In Numbers—Europe workshop (WINE) was held on October 13–18,

2013, at the Centre International de Rencontres Mathématiques, in Luminy, France.
It was the first workshop in the Women in Numbers (WIN) series held outside North
America.
The WIN conferences bring together female researchers in number theory at
multiple career stages to conduct research projects during a weeklong workshop.
The project leaders are generally senior faculty, while the participants are graduate
students, postdocs, and junior and senior faculty. The benefits of these workshops
flow in both directions: senior women meet, mentor, and collaborate with young
researchers, while junior women encounter important new research problems and
develop a network of colleagues, supporters, and mentors. Each of the WIN
conferences has given rise to a proceedings volume, and some of the collaborations
initiated at the conferences went on to generate publications in research journals.
Based on the model of the WIN conferences, the Association for Women in
Mathematics (AWM) now conducts similar events in the Research Collaboration
Conferences for Women (RCCW) series, in areas such as topology (WIT), alge-
braic combinatorics (Combinatorixx), computer vision (WISH), and applied math
(WHAM).
Due to funding restrictions, the WIN workshops were at first limited to North
American researchers; despite strong interest shown by European mathematicians,
most of them could not be accommodated. This problem led to the organization of
the WINE conference, as an attempt to correct this imbalance and to build a bridge
between female number theorists on both sides of the Atlantic.
Prior to the conference, the participants were divided into 10 working groups of
3–7 members each, according to their research interests. Each group contained a mix
of senior and junior mathematicians, including advanced graduate students, with
leaders drawn from the participating senior mathematicians. The leaders suggested
background reading and references before the conference, and arrived there with a
specific research project.
During the conference itself, participants attended background lectures given
by the group leader of each project, with several hours set aside each day for

v
vi Preface

collaborations of the working groups. Project leaders directed the research efforts
of their respective groups, providing specific tasks such as computations or partial
proofs as a way of penetrating the problem. By the end of the week, group
participants had understood their research problems and grasped where they fit into
a larger scheme, and many groups had already taken some concrete first steps in the
direction of a solution. The week closed with short presentations given by junior
members of each research group on their progress, as well as future directions for
the work.
The CIRM mathematics research center in Luminy—after initial hesitation at the
hitherto-unheard of request to organize a conference for women only—turned out
to be an ideal setting for this project. The seamless and hospitable running of the
practical side of things allowed the conference organizers to devote most of their
energies to their respective research groups, while the incredible natural beauty
of the surroundings, with rocky cliffs plunging down to small coves of turquoise
sea, gave inspiration and also much-needed opportunities for some rugged physical
activity. Each day saw some groups working outside in the sunshine, thanks to
moveable blackboards, and evenings after dinner were devoted to group research—
an intense form of socializing.
Nearly all of the groups went on with their collaborations after the conference.
Eventually, after more than a year, the papers started coming in, and according to
standard practice, were sent to anonymous referees for assessment. This volume
contains the final results of the research initiated at the WINE conference. Covering
topics in graph theory, L-functions, p-adic geometry, Galois representations, elliptic
fibrations, genus 3 curves and bad reduction, harmonic analysis, symplectic groups
and mould combinatorics, the articles span a wide swath of number theory.
The WIN conference series is a remarkable initiative that has produced signif-
icant advantages for its participants in terms of research, conference experience,
and networking. More WINs are planned for the future, and we hope the trend will
spread to include anyone, in any country, who would like to participate in a WIN
one day.

Acknowledgements We would like to thank the following organizations for providing financial
support for WINE: Centre International de Rencontres Mathématiques, Clay Mathematics Insti-
tute, Microsoft Research, National Science Foundation, and the Number Theory Foundation.

Paris, France Marie José Bertin

San Diego, CA, USA Alina Bucur
New York, NY, USA Brooke Feigon
Paris, France Leila Schneps
Contents

Explicit Construction of Ramanujan Bigraphs . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

Cristina Ballantine, Brooke Feigon, Radhika Ganapathy,
Janne Kool, Kathrin Maurischat, and Amy Wooding
Classifications of Elliptic Fibrations of a Singular K3 Surface . . . . . . . . . . . . . 17
Marie José Bertin, Alice Garbagnati, Ruthi Hortsch, Odile Lecacheux,
Makiko Mase, Cecília Salgado, and Ursula Whitcher
Shalika Germs for sln and sp2n Are Motivic. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51
Sharon M. Frechette, Julia Gordon, and Lance Robson
The Conjectural Relation Between Generalized Shalika
Models on SO4n .F/ and Symplectic Linear Models on Sp4n .F/:
A Toy Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 87
Agnès David, Marcela Hanzer, and Judith Ludwig
Bad Reduction of Genus Three Curves with Complex Multiplication . . . . . 109
Irene Bouw, Jenny Cooley, Kristin Lauter, Elisa Lorenzo García,
Michelle Manes, Rachel Newton, and Ekin Ozman
Symmetries of Rational Functions Arising in Ecalle’s Study
of Multiple Zeta Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 153
Adriana Salerno, Damaris Schindler, Amanda Tucker
On -Li Coefficients for Rankin–Selberg L-Functions . . .. . . . . . . . . . . . . . . . . . . . 167
Alina Bucur, Anne-Maria Ernvall-Hytönen, Almasa Odžak,
Edva Roditty-Gershon, and Lejla Smajlović
Galois Representations and Galois Groups Over Q . . . . . .. . . . . . . . . . . . . . . . . . . . 191
Sara Arias-de-Reyna, Cécile Armana, Valentijn Karemaker,
Marusia Rebolledo, Lara Thomas, and Núria Vila

ix
Contributors

Sara Arias-de-Reyna Mathematical Research Unit, University of Luxembourg,

Luxembourg, Luxembourg
Cécile Armana Laboratory of Mathematics, University of Franche-Comté,
Besançon, France
Christina Ballantine Department of Mathematics and Computer Science, College
of the Holy Cross, Worcester, MA, USA
Marie José Bertin Jussieu Institute of Mathematics, Pierre and Marie Curie
University, Paris, France
Irene Bouw Institute of Pure Mathematics, University of Ulm, Ulm, Germany
Alina Bucur Department of Mathematics, University of California San Diego,
San Diego, CA, USA
Jenny Cooley Mathematics Institute, University of Warwick, Coventry, UK
Agnés David Mathematical Research Unit, University of Luxembourg, Luxem-
bourg, Luxembourg
Anne-Maria Ernvall-Hytönen Department of Mathematics and Statistics,
University of Helsinki, Helsinki, Finland
Brooke Feigon Department of Mathematics, The City College of New York,
New York, NY, USA
Sharon M. Frechette Department of Mathematics and Computer Science, College
of the Holy Cross, Worcester, MA, USA
Radhika Ganapathy Mathematics Department, The University of British
Columbia, Vancouver, BC, Canada
Alice Garbagnati Department of Mathematics, University of Milan, Milan, Italy

xi
xii Contributors

Julia Gordon Mathematics Department, The University of British Columbia,

Vancouver, BC, Canada
Marcela Hanzer Department of Mathematics, University of Zagreb, Zagreb,
Croatia
Ruthi Hortsch Department of Mathematics, Massachusetts Institute of
Technology, Cambridge, MA, USA
Valentijn Karemaker Mathematical Institute, Utrecht University, Utrecht, The
Netherlands
Janne Kool Max Planck Institute for Mathematics, Bonn, Germany
Kristin Lauter Microsoft Research, Redmond, WA, USA
Odile Lecacheux Jussieu Institute of Mathematics, Pierre and Marie Curie
University, Paris, France
Elisa Lorenzo García Mathematics Institute, University of Leiden, Leiden, The
Netherlands
Judith Ludwig Mathematical Institute, University of Bonn, Bonn, Germany
Michelle Manes Department of Mathematics, University of Hawaii, Honolulu,
HI, USA
Makiko Mase Tokyo Metropolitan University, Tokyo, Japan
Kathrin Maurischat Mathematical Institute, University of Heidelberg,
Heidelberg, Germany
Rachel Newton Max Planck Institute for Mathematics, Bonn, Germany
Almasa Odžak, Department of Mathematics, University of Sarajevo, Sarajevo,
Bosnia & Herzegovina
Ekin Ozman Mathematics Department, University of Texas at Austin, Austin, TX,
USA
Department of Mathematics, Bogazici University, Istanbul, Turkey
Marusia Rebolledo Laboratory of Mathematics, Blaise Pascal University,
Clermont-Ferrand, Aubière, France
Lance Robson Mathematics Department, University of British Columbia,
Vancouver, BC, Canada
Edva Roditty-Gershon Raymond and Beverly Sackler School of Mathematical
Sciences, Tel Aviv University, Tel Aviv, Israel
Adriana Salerno Mathematics Department, Bates College, Lewiston, ME, USA
Cecília Salgado Institute for Mathematics, Federal University of Rio de Janeiro,
Rio de Janeiro, RJ, Brazil
Contributors xiii

Damaris Schindler Hausdorff Center for Mathematics, University of Bonn, Bonn,

Germany
Lejla Smajlović Department of Mathematics, University of Sarajevo, Sarajevo,
Bosnia & Herzegovina
Lara Thomas Pure and Applied Mathematics Unit, ENS Lyon, Lyon, France
Amanda Tucker Department of Mathematics, State University of New York
College at Geneseo, Geneseo, NY, USA
Núria Vila Department of Algebra and Geometry, University of Barcelona,
Barcelona, Spain
Ursula Whitcher Department of Mathematics, University of Wisconsin–Eau
Claire, Eau Claire, WI, USA
Amy Wooding Department of Mathematics and Statistics, McGill University,
Montreal, QC, Canada
Explicit Construction of Ramanujan Bigraphs

Cristina Ballantine, Brooke Feigon, Radhika Ganapathy, Janne Kool,

Kathrin Maurischat, and Amy Wooding

Abstract We construct explicitly an infinite family of Ramanujan graphs which are

bipartite and biregular. Our construction starts with the Bruhat–Tits building of an
inner form of SU3 .Qp /. To make the graphs finite, we take successive quotients by
infinitely many discrete co-compact subgroups of decreasing size.

MSC 2010: 11E39, 11E57, 16W10

C. Ballantine
Department of Mathematics and Computer Science, College of the Holy Cross,
1 College Street, Worcester, MA 01610, USA
e-mail: cballant@holycross.edu
B. Feigon ()
Department of Mathematics, The City College of New York, NAC 8/133,
New York, NY 10031, USA
e-mail: bfeigon@ccny.cuny.edu
R. Ganapathy
Department of Mathematics, The University of British Columbia, 1984 Mathematics Road,
Vancouver, BC, Canada V6T 1Z2
e-mail: rganapat@math.ubc.ca
J. Kool
Max Planck Institute for Mathematics, Vivatsgasse 7, 53111 Bonn, Germany
e-mail: kool79@mpim-bonn.mpg.de
K. Maurischat
Mathematisches Institut, Universität Heidelberg, Im Neuenheimer Feld 288,
69120 Heidelberg, Germany
e-mail: maurischat@mathi.uni-heidelberg.de
A. Wooding
The Department of Mathematics and Statistics, McGill University, Burnside Hall, Room 1023,
805 Sherbrooke W., Montreal, QC, Canada H3A 0B9
e-mail: amy.cheung@mail.mcgill.ca

© Springer International Publishing Switzerland 2015 1

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_1
2 C. Ballantine et al.

1 Introduction

Expander graphs are highly connected yet sparse graphs. By a highly connected
graph we mean a graph in which all small sets of vertices have many neighbors.
They have wide ranging applications, especially in computer science and coding
theory. They also model neural connections in the brain and many other types of
networks. One is usually interested in regular or biregular expanders. The expansion
property is controlled by the size of the spectral gap of the graph. Asymptotically,
Ramanujan graphs are optimal expanders as we will explain below. Infinite families
of regular Ramanujan graphs of fixed degree were first constructed in the late
1980s by Lubotzky et al. (1988), and independently by Margulis (1988). Since
then, the study of problems related to the existence and construction of Ramanujan
graphs has become an active area of research. Until recently, all constructions of
families of regular Ramanujan graphs have been obtained using tools from number
theory, including deep results from the theory of automorphic forms. As a result, the
graphs obtained have degree q C 1, where q is a power of a prime. Using similar
methods, Ballantine and Ciubotaru (2011) gives a roadmap toward the construction
of infinite families of Ramanujan bigraphs, i.e., biregular, bipartite graphs satisfying
the Ramanujan condition, of bidegree .q3 C 1; q C 1/, where q is a power of a prime.
However, they stop short of providing explicit examples. Very recently, Marcus
et al. (2014) used the method of interlacing polynomials to prove the existence
of arbitrary degree Ramanujan bigraphs. By making the two degrees equal, this
implies the existence of arbitrary degree (regular) Ramanujan graphs. Their proof is
non-constructive.
In this article, we follow the roadmap given in Ballantine and Ciubotaru (2011)
to explicitly construct an infinite family of Ramanujan bigraphs. We start with a
quadratic extension, E=Q, and define a division algebra D which is non-split over
E, i.e., D is not isomorphic to the matrix algebra M3 .E/. We then use this to define
a special unitary group G over E from D by means of an involution of the second
kind. We define this involution such that the corresponding local unitary group is
isomorphic to SU3 .Qp / at the place p, i.e., Gp D G.Qp / Š SU3 .Qp /, and compact
at infinity. We also give a concrete description of an infinite family of discrete co-
compact subgroups of Gp which act without fixed points on Gp .
Since the division algebra D is non-split, Corollary 4.6 of Ballantine and
Ciubotaru (2011) guarantees that each quotient of the Bruhat–Tits tree of Gp by
one of the above subgroups satisfies the Ramanujan condition. Therefore, we obtain
an infinite family of Ramanujan bigraphs of bidegree .p3 C 1; p C 1/. We note that
most of this work could be carried out over a general totally real number field but
we often choose to work over Q to simplify the notation.
Explicit Construction of Ramanujan Bigraphs 3

2 Preliminaries and Notation

In this section we introduce the notation used throughout the article and give a brief
review of Ramanujan graphs and bigraphs, unitary groups, and buildings.

2.1 Ramanujan Graphs and Bigraphs

While Ballantine and Ciubotaru (2011) also contains a concise review of this topic,
we find it useful for the reader to have an overview within the current article.
Lubotzky (2012) gives a review of expander graphs with applications within math-
ematics. Hoory et al. (2006) provides a review accessible to the nonspecialist with
many applications, especially to computer science. For an elementary introduction
to regular Ramanujan graphs, we refer the reader to Davidoff et al. (2003).
A graph X D .V; E/ consists of a set of vertices V together with a subset of pairs
of vertices called edges. In this article, all graphs are undirected. Thus, the pair of
vertices forming an edge is unordered. The degree of a vertex is the number of edges
incident to it. A graph is called k-regular if all vertices have degree k. A graph is
called .l; m/-biregular if each vertex has degree l or m. A bipartite graph is a graph
that admits a coloring of the vertices with two colors such that no two adjacent
vertices have the same color. A bigraph is a biregular, bipartite graph.
We denote by Ad.X/ the adjacency matrix of X and by Spec.X/ the spectrum
of X. Thus, Spec.X/ is the collection of eigenvalues of Ad.X/. Since the adjacency
matrix is symmetric, Spec.X/  R. For p a k-regular graph, we have k 2 Spec.X/.
For an .l; m/-biregular graph, we have lm 2 Spec.X/. Moreover, if we denote by
i the eigenvalues of a graph, for a connected k-regular graph we have

k D 0 > 1  2      k:

Thus, k is the largest absolute value of an eigenvalue of X. We denote by .X/

the next largest absolute value of an eigenvalue. If X is bipartite, the spectrum
is symmetric and k is an eigenvalue. Let X be a finite connected bigraph with
bidegree .l; m/, l  m. Suppose X has n1 vertices of degree l and n2 vertices of
degree m. We must have n2  n1 . Then, Spec.X/ is the multiset

f˙0 ; ˙1 ; : : : ; ˙n1 ; 0; : : : ; 0g;

„ ƒ‚ …
n2 n1

p
where 0 D lm > 1      n1  0. Then, with the above notation, .X/ D 1 .
4 C. Ballantine et al.

If W is a subset of V, the boundary of W, denoted by @W, is the set of vertices

outside of W which are connected by an edge to a vertex in W, i.e.,

@.W/ D fv 2 V n W j fv; wg 2 E; for some w 2 Wg:

The expansion coefficient of a graph X D .V; E/ is defined as

 ˇ 
j@Wj ˇ
c D inf ˇ W  V W 0 < jWj < 1 :
minfjWj; jV n Wjg ˇ

Note that, if jVj D n is finite, then

 ˇ 
j@Wj ˇˇ n
c D min W  V W 0 < jWj  :
jWj ˇ 2

A graph X D .V; E/ is called an .n; k; c/-expander if X is a k-regular graph on n

vertices with expansion coefficient c. The expansion coefficient c of a regular graph
is related to .X/, the second largest absolute value of an eigenvalue (Lubotzky et al.
1986, Proposition 1.2) by

.X/
2c  1  :
k
Good expanders have large expansion coefficient. Thus, good expanders have small
.X/ (or large spectral gap, k.X/). Alon and Boppana (Alon 1986; Lubotzky et al.
1988) showed that asymptotically .X/ cannot be arbitrarily small. They proved
that, if Xn;k is a k-regular graph with n vertices, then
p
lim inf .Xn;k /  2 k  1:
n!1

Lubotzky et al. (1986) defined a Ramanujan graph to be a graph that beats the Alon–
Boppana bound.
Definition
p 2.1. A k-regular graph X is called a Ramanujan graph if .X/ 
2 k  1.
Feng and Winnie Li (1996) proved the analog to the Alon–Boppana bound for
biregular bipartite graphs. They showed that, if Xn;l;m is a .l; m/-biregular graph with
n vertices, then
p p
lim inf .Xn;l;m /  l  1 C m  1:
n!1

Then, Solé (1999) defines Ramanujan bigraphs as the graphs that beat the Feng-Li
bound.
Explicit Construction of Ramanujan Bigraphs 5

Definition 2.2. A finite, connected, bigraph X of bidegree .l; m/ is a Ramanujan

bigraph if
p p p p
j l  1  m  1j  .X/  l  1 C m  1:

Solé’s definition is equivalent to the following definition given by Hashimoto

(1989).
Definition 2.3. A finite, connected, bigraph of bidegree .q1 C 1; q2 C 1/ is a
Ramanujan bigraph if
p
j..X//2  q1  q2 j  2 q1 q2 :

Our goal is to construct an infinite family of Ramanujan bigraphs of the same

bidegree and with the number of vertices growing without bound. In general, it
is difficult to check that a large regular or biregular graph is Ramanujan. In this
article, the graphs are quotients of the Bruhat–Tits building attached to an inner
form of the special unitary group in three variables. We then employ a result of
Ballantine and Ciubotaru (2011), which uses the structure of the group, to estimate
the spectrum of the building quotient in order to conclude that the graphs constructed
are Ramanujan.

2.2 Unitary Groups in Three Variables

We denote by F a local or global field of characteristic zero. For a detailed discussion

on unitary groups, we refer the reader to Rogawski (1990). Let E=F be a quadratic
extension and  W E3 E3 ! E be a Hermitian form. Then the special unitary group
with respect to  is an algebraic group over F whose functor of points is given by

SU.; R/ D fg 2 SL3 .E ˝F R/ j .gx; gy/ D .x; y/ 8x; y 2 E3 ˝F Rg

for any F-algebra R. We use SU3 to denote the standard special unitary group
corresponding to the Hermitian form given by the identity matrix; that is,

SU3 .R/ D fg 2 SL3 .E ˝F R/ j t gN g D Id3 g

where gN is conjugation with respect to the extension E=F.

Let D be a central simple algebra of degree three over E and ˛ be an involution
of the second kind, i.e., an anti-automorphism of D that acts on E by conjugation
with respect to E=F. By Wedderburn’s theorem (Knus et al. 1998, Theorem 19.2),
D is a cyclic algebra over E. Let ND denote the reduced norm of D. Then, .D; ˛/
defines a special unitary group G by

G.R/ D fd 2 .D ˝F R/ j ˛.d/d D 1; ND˝F R .d/ D 1g:

Moreover, all special unitary groups are obtained in this way from .D; ˛/ (Rogawski
1990, section 1.9).
6 C. Ballantine et al.

2.3 Buildings

Let F be a non-archimedean local field and let E=F be an unramified separable

quadratic extension. Let G D SU3 be defined as above. Let O D OE be the ring of
integers of E and p be the unique maximal ideal in O. Let k D O=p be the residue
field. We denote by B the Borel subgroup of upper-triangular matrices and by B.k/
the k-points of B. We denote by I the preimage of B.k/ under the reduction mod p
map G.O/ ! G.k/. The group I is an Iwahori subgroup. Then the Weyl group W
of G is the infinite dihedral group. Let s1 and s2 be the reflections generating W.
For i D 1; 2, let Ui D I [ Isi I. These subgroups are the representatives of the G-
conjugacy classes of maximal compact subgroups of G (Hashimoto and Hori 1989).
Moreover, I D U1 \ U2 .
The Bruhat–Tits building associated with G is a one dimensional simplicial
complex defined as follows. The set of 0-dimensional simplices consists of one
vertex for each maximal compact subgroup of G. If K1 and K2 are two maximal
compact subgroups of G, we place an edge between the vertices corresponding to
K1 and K2 if and only if K1 \ K2 is conjugate to I in G. The edges form the set of
one-dimensional simplices of the building. Since they are the faces of the largest
dimension, they are the chambers of the building. The group G acts simplicially on
the building in a natural way. The building associated with SU3 is a .q3 C 1; q C 1/
tree, where q is the cardinality of the residue field k. For more details on buildings,
we refer the reader to Tits (1979) and Garrett (1997).

2.4 Ramanujan Bigraphs from Buildings

Let G be the group SU3 over Qp (or a finite extension of Qp ). Let XQ be the Bruhat–
Tits tree of G. Let E be an imaginary quadratic extension of Q and let D be a central
simple algebra of degree 3 over E and ˛ an involution of the second kind on D. Let
G be the special unitary group over Q determined by .D; ˛/. We have the following
theorem of Ballantine and Ciubotaru (2011) that motivates our work.
Theorem 2.4 (Ballantine and Ciubotaru (2011, Theorem 1.2)). Let  be a
discrete, co-compact subgroup of G which acts on G without fixed points. Assume
that D ¤ M3 .E/, G.Qp / D G and G.R/ is compact. Then the quotient tree
X D X=Q  is a Ramanujan bigraph.

In the rest of this article we give a description of an algebra D together with an

involution ˛ fulfilling the assumptions of the above theorem, as well as an infinite
collection of discrete, co-compact subgroups of G which act on G without fixed
points.
Explicit Construction of Ramanujan Bigraphs 7

3 Choosing the Algebra and the Involution

The goal of this section is to determine explicitly a global division algebra D which
is central simple of degree three over its center E and is equipped with an involution
˛ of the second kind with fixed field F such that the related special unitary group G,

G.R/ D fd 2 .D ˝F R/ j ˛.d/d D 1 and ND˝F R .d/ D 1g;

yields compactness at infinity. Such an algebra exists by the Hasse principle (see
for example Harris and Labesse 2004, p. 657), which actually is much stronger: For
any set of local data, there is a global one localizing to it. We note that in Ballantine
and Ciubotaru (2011) the authors refer to Clozel et al. (2008) for the existence
of the global group (and thus the algebra defining it). The example of central
simple algebra with involution given in Ballantine and Ciubotaru (2011) does not
necessarily lead to Ramanujan bigraphs. It is not a division algebra and the resulting
unitary group has non-tempered representations occurring as local components
of automorphic representations. Therefore, Rogawski’s Theorem (Rogawski 1990,
Theorem 14.6.3) does not apply.

3.1 Cyclic Central Simple Algebras of Degree Three

Let E be a number field. Let L be a cyclic algebra of degree three over E, and let 
be a generator of its automorphism group which is isomorphic to the cyclic group
C3 . Then, define a cyclic central simple algebra D of degree three over E by

D D L ˚ Lz ˚ Lz2 ;

where z is a generic element satisfying z3 D a 2 E subject to the relation

zl D .l/z for any l 2 L:

By a theorem of Wedderburn (Knus et al. 1998, Theorem 19.2), any central simple
algebra of degree three is cyclic. From now on we will assume D is in the form given
above. As D is a vector space over L with basis f1; z; z2 g, we write the multiplication
by d 2 D from the right in terms of matrices to obtain an embedding D ,! M3 .L/,
0 1
l0 l1 l2
7 A.l0 ; l1 ; l2 / WD @ a.l2 / .l0 / .l1 / A ;
d D l0 C l1 z C l2 z2 !
a2 .l1 / a2 .l2 / 2 .l0 /

for l0 ; l1 ; l2 2 L. Let NL=E denote the norm of L=E, and TrL=E denote the trace of
L=E. Then for the reduced norm of D we have
8 C. Ballantine et al.

ND .d/ D det A.l0 ; l1 ; l2 / D NL=E .l0 / C aNL=E .l1 / C a2 NL=E .l2 /

aTrL=E .l0 .l1 /2 .l2 //:

In order for D to be a division algebra, we have to assume that L=E is a field

extension. Since L is a cyclic C3 -algebra over E, it follows that L=E is C3 -Galois.
Additionally, D is a division algebra if and only if neither a nor a2 belongs to the
norm group NL=E of L=E (Pierce 1982, p. 279).

3.2 Involutions of the Second Kind

Let E=F be a quadratic extension of number fields, and let hi Š C2 be its Galois
group. In order to equip a division algebra D over E with an involution ˛ of the
second kind with fixed field F, we need to extend the nontrivial automorphism  of
E to D.
We start by extending  to L,  W L ! D. For this we have two possibilities.
Either, the image L0 WD .L/ equals L or it does not. If L0 does not equal L, then 
gives rise to an isomorphism of L to L0 inside some field extension containing both.
However, L and L0 are not isomorphic as extensions of E, otherwise D would not
be a division algebra. So L=F is not Galois. Notice that in this case L along with L0
generate D. In contrast, if we extend  W L ! L, i.e., .L/ D L, then h; i is an
automorphism group of L=Q of order at least six. That is, the degree six extension
L=F is Galois with Galois group h; i, which is isomorphic to the cyclic group C6
or the symmetric group S3 .

3.3 Compactness at Infinity

We now assume F is totally real. For simplicity, let F D Q.

In order for the unitary group defined by .D; ˛/ to be compact at infinity, we
need E=Q to be imaginary quadratic. To see this, assume E=Q is real quadratic.
Then, E1 D E ˝Q R Š R ˚ R would split. Therefore, L1 would split as well
and we would be able to find an isomorphism D1 Š M3 .R/ ˚ M3 .R/, where the
involution is given by (see Platonov and Rapinchuk 1994, p. 83)

.x; y/ 7! .t y; t x/;

and the reduced norm is given by

ND .x; y/ D det.x/ det.y/:

Explicit Construction of Ramanujan Bigraphs 9

Thus,

G1 WD G.R/ D f.x; y/ 2 D1 j . t yx; t xy/

D .Id3 ; Id3 / and det.x/ det.y/ D 1g Š GL3 .R/

is not compact.
Next we remark that in the case when L=Q is Galois, the Galois group is
necessarily C6 . To see this, assume L=E is a C3 D hi-Galois extension such that
L=Q is S3 -Galois. At infinity, we have

E1 D E ˝Q R Š C

and  acts by complex conjugation. Therefore,

L1 D L ˝Q R Š L ˝E C Š C ˚ C ˚ C;

with the isomorphism given by

l ˝ s 7! .0 .l/s; 1 .l/s; 2 .l/s/ for l 2 L and s 2 E1 :

Notice that ŒL W E D 3, so there is always a real primitive element, and thus

there is an E-basis for L which is -invariant. Here multiplication in L1 is defined
coordinate wise. The S3 -action is given by

.l ˝ s/ D .l/ ˝ s 7! .1 .l/s; 2 .l/s; 0 .l/s/

and

.l ˝ s/ 7! ..l/.s/; 2 .l/.s/; .l/.s//:

Thus, for any .t0 ; t1 ; t2 / 2 L1 we have

.t0 ; t1 ; t2 / D .t1 ; t2 ; t0 /;
.t0 ; t1 ; t2 // D .Nt0 ; Nt2 ; Nt1 /;

with the usual complex conjugation. Without specifying the algebra .D; ˛/ contain-
ing L any further, we read off that D1 is isomorphic to the matrix algebra M3 .C/
with L1 embedded diagonally. This leads to the following result.
Proposition 3.1. Let E,L, and .D; ˛/ be as above, and assume L=Q is S3 -Galois.
Then the split torus

T1 D f.Ntt1 ; t; Nt1 / j t 2 C g  L1

is contained in G1 . In particular, G1 is non-compact.

10 C. Ballantine et al.

Proof of Proposition 3.1. We check the definition of G for elements of T1 . We have

ND ..Ntt1 ; t; Nt1 // D NL1 =E1 ..Ntt1 ; t; Nt1 // D Ntt1  t  Nt1 D 1;

as well as

˛..Ntt1 ; t; Nt1 //  .Ntt1 ; t; Nt1 / D ..Ntt1 ; t; Nt1 //  .Ntt1 ; t; Nt1 /

D .tNt1 ; t1 ; Nt/  .Ntt1 ; t; Nt1 / D 1:

Therefore, T1 defines a non-compact torus of G1 . t

u
In the case when L=Q is Galois, there is an obvious (but not unique) choice of an
involution of the second kind. As  extends to an automorphism of L, it is defined
on any coefficient of A.l0 ; l1 ; l2 /. Thus, the map
0 1
.l0 / .a.l2 // .a2 .l1 //
˛.A.l0 ; l1 ; l2 // WD t .A.l0 ; l1 ; l2 // D @.l1 / .l0 / .a2 .l2 //A
.l2 / .l1 / 2 .l0 /

clearly satisfies the conditions

˛ 2 D id;
˛.A  B/ D ˛.B/  ˛.A/;
˛ jE D :

In order for ˛ to be an involution on D of the second kind, we must have that the
image ˛.D/ is contained in D. Defining

Ql0 D .l0 /; Ql1 D .a/.l2 /; Ql2 D .a/2 .l1 /;

this condition is equivalent to

˛.A.l0 ; l1 ; l2 // D A.Ql0 ; Ql1 ; Ql2 //:

That is,
0 1 0 1
.l0 / .a.l2 // .a2 .l1 // Ql0 Ql1 Ql2
@.l1 / .l0 / .a2 .l2 //A D @ a.Ql2 / .Ql0 / .Ql1 / A :
.l2 / .l1 / 2 .l0 / a2 .Ql1 / a2 .Ql2 / 2 .Ql0 /

This evidently reduces to the following conditions

 D  on L
Explicit Construction of Ramanujan Bigraphs 11

and

a.a/ D 1:

We summarize the above discussion in the following theorem.

Theorem 3.2. Assume the extension L=Q is Galois, and that ˛ is defined by

˛.A.l0 ; l1 ; l2 // D t .A.l0 ; l1 ; l2 //:

Then .D; ˛/ is a division algebra which is central simple over E with involution ˛
of the second kind if and only if the following conditions are satisfied:
(i) a 2 E , and a; a2 … NL=E
(ii) NE=Q .a/ D a.a/ D 1
(iii)  D  on L, i.e. L=Q is C6 -Galois.
Moreover, if these conditions are satisfied, the group G1 is compact.
Proof of Theorem 3.2. The first part of the theorem is proved above. What is left to
show is compactness at infinity. The realization of L inside the diagonal subgroup
of M3 .L/ is chosen such that it is compatible with the isomorphism

L1 D L ˝Q R Š L ˝E C Š C3

induced by the three embeddings of L into C. Indeed, D1 Š M3 .C/ with involution

˛ W M3 .C/ ! M3 .C/ given by ˛.A/ D t A. Therefore,

G1 Š fA 2 M3 .C/ j t A  A D Id3 ; det A D 1g D SU3 .R/;

is induced by the standard hermitian form of signature .3; 0/. Thus, G1 is compact.
t
u
Notice that it is non-trivial to satisfy condition (iii) of Theorem 3.2, as a quadratic
field E=Q does not necessarily allow an extension L of degree three which is C6 -
Galois over Q. However, there are situations which allow for the conditions of
Theorem 3.2 to be satisfied. Below we provide such an example.
p
Example 3.3 (An Example in the Galois-Case). Let E D Q. 3/. Therefore, E
contains a primitive third root of unity, 3 , and Kummer theory applies. That is,pany
cyclic C3 -extension L=E can be obtained by adjoining a third root, p L D E. 3 b/,
where b 2 E n.E /3 . In particular, choose b D 3 . Then, 3 3 D 9 is a
primitive 9th root of unity. Then, L D E.9 / D Q.9 / is a cyclotomic field,
which is tautologically cyclic over Q. Its relative Galois group is Gal.L=E/ D hi,
where .9 / D 3 9 . Extending  (complex conjugation) from E to L means
.9 / D 98 . Thus,

.9 / D .9 /8 D 38 98 D N3 98 D .9 /:

12 C. Ballantine et al.

Now choose an element a 2 E such that a; a2 … NL=E and NE=Q .a/ D 1. One can
take for example
p
2C
3
aD p :
2  3

Then, trivially, NE=Q .a/ D 1, and we verified using Magma that a; a2 … NL=E .
p
Example 3.4 (An Example in the Non-Galois Case). Again, choose E D Q. 3/.
But this time, choose a cyclic degree three extension L D E. /, 3 D b, where
b 2 E n.E /3 is chosen such that L=Q is not Galois. For example, one could choose
b D 23 . The automorphism  of L=E is given by . / D 3 , and the minimal
polynomial is given by X 3  b. Let 0 be a root of X 3  .b/, and let L0 D E. 0 /.
Then (within any field extension containing both) L and L0 are non-equal, but there is
an isomorphism ˛ W L ! L0 extending  given by . / D 0 . For the cyclic algebra
.D; ˛/ with involution, choose L as above and a D .b/, i.e. z may be identified
with 0 . Then the above constraint

˛. / D z

determines an involution on D of the second kind, as ˛.z/ D ˛ 2 . / D . For

convenience, let d D l0 C l1 z C l2 z2 , lj 2 L, be an arbitrary element of D, then

˛.d/ D ˛.l0 / C ˛.l1 / C 2 ˛.l2 /;

and one easily checks ˛ 2 .d/ D d. Using the identification of D with a subring of
M3 .L/ as before, we write down this involution for matrices:
0 1 0 1
010 0 0
z D @0 0 1A !
7 ˛.z/ D @ 0 . / 0 A :
a00 0 0 2 . /

So for an element e0 C e1 z C e2 z2 2 L0 D E.z/  D, i.e. ej 2 E,

˛.A.e0 ; e1 ; e2 // D A..e0 / C .e1 / C .e2 / 2 ; 0; 0/:

As ˛ 2 D id, we read off the image of l D e0 C e1 C e2 2 2 L under ˛ in matrix

form:
0 1
.e0 / .e1 / .e2 /
˛.A.l; 0; 0// D @a.e2 / .e0 / .e1 /A :
a.e1 / a.e2 / .e0 /
Explicit Construction of Ramanujan Bigraphs 13

Thus, finally

˛.A.l0 ; l1 ; l2 // D ˛.A.l0 ; 0; 0// C ˛.A.l1 ; 0; 0// C 2 ˛.A.l2 ; 0; 0//;

that is, for lj D ej0 C ej1 C ej2 2 2 L with ejk 2 E, we find

˛.A.l0 ; l1 ; l2 // D A.Ql0 ; Ql1 ; Ql2 /;

where Qlj D .e0j / C .e1j / C .e2j / 2 .

4 Choosing the Family of Subgroups

Let G be the global special unitary group constructed from the division algebra and
the involution of the second kind given in Example 3.3 of the previous section. Let
p be a place where Gp WD G.Qp / is isomorphic to SU3 .Qp /. In this section, we
will give an explicit infinite family of discrete co-compact subgroups of G which
act without fixed points on the Bruhat–Tits tree of Gp . Before we proceed, we need
to describe the place p explicitly. From Rogawski (1990, 14.2) we have that Gp is
isomorphic to SU3 .Qp / if and only if p is inert in E. In fact, we can see this directly
as shown below. If p does not remain prime (i.e., is not inert), then there are two
cases. Either (i) p ramifies in E (i.e., .p/ D p2 , p D p) or (ii) p splits into two
non-equal prime ideals in E (i.e., .p/ D pp with p 6D p).
p
(i) Theponly prime ramified in E is .p/ D .3/ D p2 , where p D . 3/ D
. 3/ D p. In this case, Ep =Qp is a ramified field extension. The involution
˛ is then trivial on the localization Ep , as ˛.p/ D p D p. The group Gp will
lead to a .p C 1/-regular tree, the Bruhat–Tits building on SL2 .Qp /. This case
has been treated in Lubotzky et al. (1988). p
(ii) There are many primes p which are split in E D Q. 3/. The minimal
polynomial is X 2 C 3. This is reducible modulo p if and only if p is split in E.
Equivalently, the minimal polynomial is reducible if and only if 3 is a square
mod p. The two localizations Ep and Ep here are both equal to Qp . Therefore,
the field extension E=Q localizes as a split algebra Ep D Ep ˚ Ep D Qp ˚ Qp .
The involution ˛ is conjugation on E, so ˛.p/ D p. That is, ˛ exchanges the
two summands of Ep . Then, Dp D Dp ˚ Dp , and for an element .g1 ; g2 / 2 Dp
to be in Gp we need

N.g1 ; g2 / D .1; 1/
and
.1; 1/ D ˛..g1 ; g2 //.g1 ; g2 / D .˛.g2 /; ˛.g1 //.g1 ; g2 / D .˛.˛.g1 /g2 /; ˛.g1 /g2 /:

That is, g2 D ˛.g1 /1 for some g1 in the reduced norm one group, ND1 p , of Dp .
Thus, Gp Š ND1 p Š ND1 p .
14 C. Ballantine et al.

Finally, if p is inert in E, then Ep =Qp is an unramified quadratic field extension.

This is the case if and only if 3 is not a square modulo p. Then Gp Š G.Qp /.
Therefore, only these primes are “good” primes for us, i.e., leading to Ramanujan
bigraphs. By quadratic reciprocity, for a prime p > 3,
  (
3 1; p 1; 7 .mod 12/
D
p 1 p 5; 11 .mod 12/:

Thus the “good” primes are the primes p such that p 5; 11 .mod 12/.
Fix a prime p 5; 11 .mod 12/ and let q be a prime not equal to p. We follow
the notation in Ballantine and Ciubotaru (2011, 4.3). Let ZŒp1  be the subring of Q
consisting of rational numbers with powers of p in the denominator. Notice that G1
and Gp are matrix groups with coefficients in R and Qp , respectively. By abuse of
notation, we denote by G1 .ZŒp1 / and Gp .ZŒp1 / the obvious subgroups in G1
and Gp , respectively. It is clear that G1 .ZŒp1 / and Gp .ZŒp1 / are isomorphic.
Define G.ZŒp1 / WD G1 .ZŒp1 /  Gp .ZŒp1 / to be their product in G1  Gp .
It follows from Borel (1963) that G.ZŒp1 / is a lattice in G1 Gp . For each positive
integer n, we define the kernel modulo qn ,

.qn / WD ker.G.ZŒp1 / ! G.ZŒp1 =qn ZŒp1 /;

and

p .qn / WD .qn / \ Gp :

Then, as shown in Ballantine and Ciubotaru (2011), each p .qn / is a discrete co-
compact subgroup of Gp . It has finite index and no non-trivial elements of finite
order. Thus, each subgroup p .qn / acts on the Bruhat–Tits tree of Gp without fixed
points and the quotient building is a finite biregular graph of bidegree .p3 C1; pC1/.

5 An Infinite Family of Ramanujan Bigraphs

Let G be the inner form of SU3 constructed using the division algebra and involution
of Example 3.3. Let p be a prime congruent to 5 or 11 modulo 12 and q a prime not
equal to p. We denote by XQ the Bruhat–Tits tree associated with Gp . For each positive
integer n, let p .qn / be the subgroup of Gp constructed in the previous section and let
Xn be the quotient of XQ by the action of p .qn /. By Ballantine and Ciubotaru (2011,
Corollary 4.6), Xn is a Ramanujan bigraph. Thus, we have constructed an infinite
family of Ramanujan bigraphs. As p .qnC1 / ¨ p .qn /, the number of vertices of
Xn tends to infinity as n ! 1. Moreover, for each n, the graph Xn is a subgraph
of XnC1 .
Explicit Construction of Ramanujan Bigraphs 15

Acknowledgements The authors would like to thank Dan Ciubotaru, Judith Ludwig, and Rachel
Newton for helpful discussions. This work was initiated at the CIRM workshop, “Femmes en
nombre” in October 2013 and the authors would like to thank CIRM, Microsoft Research, the
National Science Foundation DMS-1303457, the Clay Mathematics Institute, and the Number
Theory Foundation for supporting the workshop.
This work was partially supported by a grant from the Simons Foundation (#245997 to Cristina
Ballantine).
This work was partially supported by grants from the National Science Foundation
(DMS-1201446) and from PSC-CUNY (Brooke Feigon).
This work was partially supported by the European Social Fund (Kathrin Maurischat).

References

Alon, N.: Eigenvalues and expanders. Combinatorica 6(2), 83–96 (1986). Theory of computing,
Singer Island (1984). MR 875835 (88e:05077)
Ballantine, C., Ciubotaru, D.: Ramanujan bigraphs associated with SU.3/ over a p-adic field. Proc.
Am. Math. Soc. 139(6), 1939–1953 (2011). MR 2775370 (2012b:22020)
Borel, A.: Some finiteness properties of adele groups over number fields. Inst. Hautes Études Sci.
Publ. Math. 16, 5–30 (1963). MR 0202718 (34 #2578)
Clozel, L., Harris, M., Taylor, R.: Automorphy for some l-adic lifts of automorphic mod l Galois
representations. Publ. Math. Inst. Hautes Études Sci. 108, 1–181 (2008). With Appendix A,
summarizing unpublished work of Russ Mann, and Appendix B by Marie-France Vignéras.
MR 2470687 (2010j:11082)
Davidoff, G., Sarnak, P., Valette, A.: Elementary Number Theory, Group Theory, and Ramanujan
Graphs. London Mathematical Society Student Texts, vol. 55. Cambridge University Press,
Cambridge (2003). MR 1989434 (2004f:11001)
Feng, K., Winnie Li, W.-C.: Spectra of hypergraphs and applications. J. Number Theory 60(1),
1–22 (1996). MR 1405722 (97f:05128)
Garrett, P.: Buildings and Classical Groups. Chapman & Hall, London (1997). MR 1449872
(98k:20081)
Hashimoto, K.-i.: Zeta functions of finite graphs and representations of p-adic groups. In: Auto-
morphic Forms and Geometry of Arithmetic varieties. Advanced Studies in Pure Mathematics,
vol. 15, pp. 211–280. Academic, Boston (1989). MR 1040609 (91i:11057)
Hashimoto, K.-i., Hori, A.: Selberg-Ihara’s zeta function for p-adic discrete groups. In: Automor-
phic Forms and Geometry of Arithmetic Varieties. Advanced Studies in Pure Mathematics,
vol. 15, pp. 171–210. Academic, Boston (1989). MR 1040608 (91g:11053)
Harris, M., Labesse, J.-P.: Conditional base change for unitary groups. Asian J. Math. 8(4),
653–684 (2004)
Hoory, S., Linial, N., Wigderson, A.: Expander graphs and their applications. Bull. Am. Math. Soc.
(N.S.) 43(4), 439–561 (electronic) (2006). MR 2247919 (2007h:68055)
Knus, M.-A., Merkurjev, A., Rost, M., Tignol, J.-P.: The Book of Involutions. American Mathe-
matical Society Colloquium Publications, vol. 44. American Mathematical Society, Providence
(1998). With a preface in French by J. Tits. MR 1632779 (2000a:16031)
Lubotzky, A.: Expander graphs in pure and applied mathematics. Bull. Am. Math. Soc. (N.S.)
49(1), 113–162 (2012). MR 2869010 (2012m:05003)
Lubotzky, A., Phillips, R.L., Sarnak, P.: Explicit expanders and the ramanujan conjectures. In:
STOC ‘86 Proceedings of the eighteenth annual ACM symposium on Theory of computing,
pp. 240–246 (1986)
Lubotzky, A., Phillips, R.L., Sarnak, P.: Ramanujan graphs. Combinatorica 8(3), 261–277 (1988).
MR 963118 (89m:05099)
16 C. Ballantine et al.

Marcus, A., Spielman, D.A., Srivastava, N.: Interlacing families I: Bipartite Ramanujan graphs of
all degrees. Ann. Math. 182(1), 307–325 (2015)
Margulis, G.A.: Explicit group-theoretic constructions of combinatorial schemes and their applica-
tions in the construction of expanders and concentrators. Problemy Peredachi Informatsii 24(1),
51–60 (1988). MR 939574 (89f:68054)
Pierce, R.S.: Associative Algebras. Graduate Texts in Mathematics, vol. 88. Springer, New York,
1982. Studies in the History of Modern Science, 9. MR 674652 (84c:16001)
Platonov, V., Rapinchuk, A.: Algebraic Groups and Number Theory. Pure and Applied Mathemat-
ics, vol. 139. Academic, Boston (1994). Translated from the 1991 Russian original by Rachel
Rowen. MR 1278263 (95b:11039)
Rogawski, J.D.: Automorphic representations of unitary groups in three variables. Annals of
Mathematics Studies, vol. 123. Princeton University Press, Princeton (1990). MR 1081540
(91k:22037)
Solé, P.: Ramanujan hypergraphs and Ramanujan geometries. In: Emerging applications of number
theory (Minneapolis, MN, 1996), The IMA Volumes in Mathematics and its Applications,
vol. 109, pp. 583–590. Springer, New York (1999). MR 1691550 (2000c:05108)
Tits, J.: Reductive groups over local fields. In: Automorphic Forms, Representations and
L-Functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Part 1,
Proc. Sympos. Pure Math., XXXIII, pp. 29–69. American Mathematical Society, Providence
(1979) MR 546588 (80h:20064)
Classifications of Elliptic Fibrations
of a Singular K3 Surface

Marie José Bertin, Alice Garbagnati, Ruthi Hortsch, Odile Lecacheux,

Makiko Mase, Cecília Salgado, and Ursula Whitcher

Abstract We classify, up to automorphisms, the elliptic fibrations on the singular

K3 surface X whose transcendental lattice is isometric to h6i ˚ h2i.

MSC 2010: Primary 14J28, 14J27; Secondary 11G05, 11G42, 14J33

1 Introduction

We classify elliptic fibrations on the singular K3 surface X associated with the

Laurent polynomial

M.J. Bertin ()

Jussieu Institute of Mathematics, Pierre and Marie Curie University, Paris, France
e-mail: marie-jose.bertin@imj-prg.fr
A. Garbagnati
Department of Mathematics, University of Milan, Milan, Italy
e-mail: alice.garbagnati@unimi.it
R. Hortsch
Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA, USA
e-mail: rhortsch@math.mit.edu
O. Lecacheux
Jussieu Institute of Mathematics, Pierre and Marie Curie University, Paris, France
e-mail: odile.lecacheux@imj-prg.fr
M. Mase
Tokyo Metropolitan University, Tokyo, Japan
e-mail: mtmase@arion.ocn.ne.jp
C. Salgado
Institute for Mathematics, Federal University of Rio de Janeiro, Rio de Janeiro, RJ, Brazil
e-mail: salgado@im.ufrj.br
U. Whitcher
Department of Mathematics, University of Wisconsin-Eau Claire, Eau Claire, WI, USA
e-mail: whitchua@uwec.edu

© Springer International Publishing Switzerland 2015 17

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_2
18 M.J. Bertin et al.

1 1 1 x y y z z x
xC CyC CzC C C C C C C :
x y z y x z y x z

In order to compute the Néron–Severi lattice, the Picard number, and other basic
properties of an algebraic surface, it is useful to identify an elliptic fibration on the
surface. Moreover, in view of different applications, one may be interested in finding
all the elliptic fibrations of a certain type. The fibrations of rank 0 and maximal
torsion lead more easily to the determination of the L-series of the variety (Bertin
2010). Those of positive rank lead to symplectic automorphisms of infinite order
of the variety. Lenstra’s Elliptic Curve Method (ECM) for finding small factors of
large numbers originally used elliptic curves on Q with a torsion-group of order 12
or 16 and rank  1 on Q (Montgomery 1987; Atkin and Morain 1993). One way
to obtain infinite families of such curves is to use fibrations of modular surfaces, as
explained by Elkies (2007).
If the Picard number of a K3 surface is large, there may be an infinite
number of elliptic fibrations, but there is only a finite number of fibrations up to
automorphisms, as proved by Sterk (1985). Oguiso used a geometric method to
classify elliptic fibrations in Oguiso (1989). Some years later, Nishiyama (1996)
proposed a lattice-theoretic technique to produce such classifications, recovering
Oguiso’s results and classifying other Kummer and K3 surfaces. Since then, results
of the same type have been obtained by various authors (Kumar 2014; Elkies and
Schütt 2014; Bertin and Lecacheux 2013).
Recently, the work of Braun et al. (2013) described three possible classifications
of elliptic fibrations on a K3 surface, shining a new light on the meaning of
what is a class of equivalence of elliptic fibrations. In particular, they proposed
a J1 -classification of elliptic fibrations up to automorphisms of the surface and
a J2 -classification of the frame lattices of the fibrations. For our K3 surface, the
two classifications coincide. Thus, it is particularly interesting to exhibit here an
J2 -classification by the Kneser–Nishiyama method, since in general it is not easy to
obtain the J1 -classification. This topic will be explained in detail in Section 2.
Section 3 is devoted to a toric presentation of the surface X, following ideas of
Karp et al. (2013), based on the classification of reflexive polytopes in dimension 3.
More precisely, the Newton polytope of X is in the same class as the reflexive
polytope of index 1529. Since, according to Karp et al. (2013), there is an S4 action
on the vertices of polytope 1529 and its polar dual, there is a symplectic action of
S4 on X: This action will be described on specific fibrations. One of them gives
the transcendental lattice TX D h6i ˚ h2i: We may use these fibrations to relate X
to a modular elliptic surface analyzed by Beauville in Beauville (1982). We also
describe a presentation of X found in Garbagnati and Sarti (2009), which represents
X as a K3 surface with a prescribed abelian symplectic automorphism group.
The main results of the paper are obtained by Nishiyama’s method and are
summarized in Section 4, Theorem 4.2.
Theorem 1.1. The classification up to automorphisms of the elliptic fibrations on
X is given in Table 1. Each elliptic fibration is given with the Dynkin diagrams
Classifications of Elliptic Fibrations of a Singular K3 Surface 19

characterizing its reducible fibers and the rank and torsion of its Mordell–Weil
group. More precisely, we obtained 52 elliptic fibrations on X, including 17
fibrations of rank 2 and one of rank 3.
Due to the high number of different elliptic fibrations, we give only a few cases
of computing the torsion. These cases have been selected to give an idea of the
various methods involved. Notice the case of fibrations #22 and #22b exhibiting
two elliptic fibrations with the same singular fibers and torsion but not isomorphic.
Corresponding to these different fibrations we give some particularly interesting
Weierstrass models; it is possible to make an exhaustive list.

2 Classification of Elliptic Fibrations on K3 Surfaces

Let S be a smooth complex compact projective surface.

Definition 2.1. A surface S is a K3 surface if its canonical bundle and its irregular-
ity are trivial, that is, if KS ' OS and h1;0 .S/ D 0.
Definition 2.2. A flat surjective map E W S ! P1 is called an elliptic fibration if:
1) the generic fiber of E is a smooth curve of genus 1;
2) there exists at least one section s W P1 ! S for E.
In particular, we choose one section of E, which we refer to as the zero section.
We always denote by F the class of the fiber of an elliptic fibration and by O the
curve (and the class of this curve) which is the image of s in S.
The group of the sections of an elliptic fibration E is called the Mordell–Weil
group and is denoted by MW.E/.
A generic K3 surface does not admit elliptic fibrations, but if the Picard number
of the K3 is sufficiently large, it is known that the surface must admit at least
one elliptic fibration (see Proposition 2.3). On the other hand, it is known that a
K3 surface admits a finite number of elliptic fibrations up to automorphisms (see
Proposition 2.5). Thus, a very natural problem is to classify the elliptic fibrations on
a given K3 surface. This problem has been discussed in several papers, starting in
the Eighties. There are essentially two different ways to classify elliptic fibrations on
K3 surfaces described in Oguiso (1989) and Nishiyama (1996). In some particular
cases, a third method can be applied; see Kumar (2014). First, however, we must
introduce a different problem: “What does it mean to ‘classify’ elliptic fibrations?”
A deep and interesting discussion of this problem is given in Braun et al. (2013),
where the authors introduce three different types of classifications of elliptic
fibrations and prove that under certain (strong) conditions these three different
classifications collapse to a unique one. We observe that it was already known by
Oguiso (1989) that in general these three different classifications do not collapse to
a unique one. We now summarize the results by Braun et al. (2013) and the types of
classifications.
20 M.J. Bertin et al.

2.1 Types of Classifications of Elliptic Fibrations

on K3 Surfaces

In this section we recall some of the main results on elliptic fibrations on K3 surfaces
(for example, compare Schütt and Shioda 2010), and we introduce the different
classifications of elliptic fibrations discussed in Braun et al. (2013).

2.1.1 The Sublattice U and the J0 -Classification

Let S be a K3 surface and E W S ! P1 be an elliptic fibration on S. Let F 2

NS.S/ be the class of the fiber of E. Then F is a nef divisor which defines the map
jFj W S ! P.H 0 .S; F/ / which sends every point p 2 S to .s0 .p/ W s1 .p/ W : : : W
sr .p//, where fsi giD1;:::r is a basis of H 0 .S; F/, i.e. a basis of sections of the line
bundle associated to the divisor F. The map jFj is the elliptic fibration E. Hence,
every elliptic fibration on a K3 surface is uniquely associated with an irreducible
nef divisor (with trivial self-intersection). Since E W S ! P1 admits a section, there
exists a rational curve which intersects every fiber in one point. Its class in NS.S/
is denoted by O and has the following intersection properties O2 D 2 (since O
is a rational curve) and FO D 1 (since O is a section). Thus, the elliptic fibration
E W S ! P1 (with a chosen section, as in Definition 2.2) is uniquely associated with
a pair of divisors .F; O/. This pairof divisors spans a lattice which is isometric to
01
U, represented by the matrix , (considering the basis F, F C O). Hence each
10
elliptic fibration is associated to a chosen embedding of U in NS.S/.
On the other hand, the following result holds:
Proposition 2.3 (Kondo (1992, Lemma 2.1) and Nikulin (1980a, Corollary
1.13.15)). Let S be a K3 surface, such that there exists a primitive embedding
' W U ,! NS.S/. Then S admits an elliptic fibration.
Let S be a K3 surface with Picard number .S/  13. Then, there is a primitive
embedding of U in NS.S/ and hence S admits at least one elliptic fibration.
A canonical embedding of U in NS.S/ is defined as follows: Let us denote by
b1 and b2 the unique two primitive vectors of U with trivial self-intersection. An
embedding of U in NS.S/ is called canonical if the image of b1 in NS.S/ is a nef
divisor and the image of b2  b1 in NS.S/ is an effective irreducible divisor.
The first naive classification of the elliptic fibrations that one can consider is the
classification described above, roughly speaking: two fibrations are different if they
correspond to different irreducible nef divisors with trivial self-intersections. This
essentially coincides with the classification of the canonical embeddings of U in
NS.S/.
Following Braun et al. (2013) we call this classification the J0 -classification of
the elliptic fibrations on S.
Classifications of Elliptic Fibrations of a Singular K3 Surface 21

Clearly, it is possible (and indeed likely, if the Picard number is sufficiently

large) that there is an infinite number of irreducible nef divisors with trivial self-
intersection and also infinitely many copies of U canonically embedded in NS.S/.
Thus, it is possible that there is an infinite number of fibrations in curves of genus 1
on S and moreover an infinite number of elliptic fibrations on S.

2.1.2 Automorphisms and the J1 -Classification

The automorphism group of a variety transforms the variety to itself preserving

its structure, but moves points and subvarieties on the variety. Thus, if one is
considering a variety with a nontrivial automorphism group, one usually classifies
objects on the variety up to automorphisms.
Let S be a K3 surface with a sufficiently large Picard number (at least 2). Then
the automorphism group of S is in general nontrivial, and it is often of infinite order.
More precisely, if .S/ D 2, then the automorphism group of S is finite if and only
if there is a vector with self-intersection either 0 or 2 in the Néron–Severi group.
If .S/  3, then the automorphism group of S is finite if and only if the Néron–
Severi group is isometric to a lattice contained in a known finite list of lattices, cf.
Kondo (1989). Let us assume that S admits more than one elliptic fibration (up to
the J0 -classification defined above). This means that there exist at least two elliptic
fibrations E W S ! P1 and E 0 W S ! P1 such that F ¤ F 0 2 NS.S/, where F (resp. F 0 )
is the class of the fiber of the fibration E (resp. E 0 ). By the previous observation, it
seems very natural to consider E and E 0 equivalent if there exists an automorphism
of S which sends E to E 0 . This is the idea behind the J1 -classification of the elliptic
fibrations introduced in Braun et al. (2013).
Definition 2.4. The J1 -classification of the elliptic fibrations on a K3 surface is the
classification of elliptic fibrations up to automorphisms of the surface. To be more
precise: E is J1 -equivalent to E 0 if and only if there exists g 2 Aut.S/ such that
E D E 0 ı g.
We observe that if two elliptic fibrations on a K3 surface are equivalent up
to automorphism, then all their geometric properties (the type and the number
of singular fibers, the properties of the Mordell–Weil group and the intersection
properties of the sections) coincide. This is true essentially by definition, since an
automorphism preserves all the “geometric” properties of subvarieties on S.
The advantages of the J1 -classification with respect to the J0 -classification are
essentially two. The first is more philosophical: in several contexts, to classify an
object on varieties means to classify the object up to automorphisms of the variety.
The second is more practical and is based on an important result by Sterk: the
J1 -classification must have a finite number of classes:
Proposition 2.5 (Sterk (1985)). Up to automorphisms, there exists a finite number
of elliptic fibrations on a K3 surface.
22 M.J. Bertin et al.

2.1.3 The Frame Lattice and the J2 -Classification

The main problem of the J1 -classification is that it is difficult to obtain a

J1 -classification of elliptic fibrations on K3 surfaces, since it is in general difficult
to give a complete description of the automorphism group of a K3 surface and the
orbit of divisors under this group. An intermediate classification can be introduced,
the J2 -classification. The J2 -classification is not as fine as the J1 -classification,
and its geometric meaning is not as clear as the meanings of the classifications
introduced above. However, the J2 -classification can be described in a very natural
way in the context of lattice theory, and there is a standard method to produce it.
Since the J2 -classification is essentially the classification of certain lattices
strictly related to the elliptic fibrations, we recall here some definitions and
properties of lattices related to an elliptic fibration.
We have already observed that every elliptic fibration on S is associated with an
embedding
W U ,! NS.S/.
Definition 2.6. The orthogonal complement of
.U/ in NS.S/,
.U/?NS.S/ , is
denoted by WE and called the frame lattice of E.
The frame lattice of E encodes essentially all the geometric properties of E, as
we explain now. We recall that the irreducible components of the reducible fibers
which do not meet the zero section generate a root lattice, which is the direct sum
of certain Dynkin diagrams. Let us consider the root lattice .WE /root of WE . Then
the lattice .WE /root is exactly the direct sum of the Dynkin diagram corresponding
to the reducible fibers. To be more precise if the lattice E8 (resp. E7 , E6 , Dn , n  4,
Am , m  3) is a summand of the lattice .WE /root , then the fibration E admits a
fiber of type II  (resp. IV  , III  , In4

, ImC1 ). However, the lattices A1 and A2 can be
associated with two different types of reducible fibers, i.e. with I2 and III and with I3
and IV, respectively. We cannot distinguish between these two different cases using
lattice theory. Moreover, the singular non-reducible fibers of an elliptic fibration can
be either of type I1 or of type II.
Given an elliptic fibration E on a K3 surface S, the lattice Tr.E/ WD U ˚ .WE /root
is often called the trivial lattice (see Schütt and Shioda 2010, Lemma 8.3 for a more
detailed discussion).
Let us now consider the Mordell–Weil group of an elliptic fibration E on a
K3 surface S: its properties are also encoded in the frame WE , indeed MW.E/ D
WE =.WE /root . In particular,

rank.MW.E// D rank.WE /  rank..WE /root / and

.MW.E//tors D .WE /root =.WE /root ;

where, for every sublattice L  NS.S/, L denotes the primitive closure of L in NS.S/,
i.e. L WD .L ˝ Q/ \ NS.S/.
Definition 2.7. The J2 -classification of elliptic fibrations on a K3 surface is the
classification of their frame lattices.
Classifications of Elliptic Fibrations of a Singular K3 Surface 23

It appears now clear that if two elliptic fibrations are identified by the
J2 -classification, they have the same trivial lattice and the same Mordell–Weil
group (since these objects are uniquely determined by the frame of the elliptic
fibration).
We observe that if E and E 0 are identified by the J1 -classification, then there
exists an automorphism g 2 Aut.S/, such that E D E 0 ı g. The automorphism
g induces an isometry g on NS.S/ and it is clear that g W WE ! WE 0 is an
isometry. Thus the elliptic fibrations E and E 0 have isometric frame lattices and
so are J2 -equivalent.
The J2 -classification is not as fine as the J1 -classification; indeed, if
h W WE ! WE 0 is an isometry, a priori there is no reason to conclude that there
exists an automorphism g 2 Aut.S/ such that gjWE D h; indeed, comparing
the J1 -classification given in Oguiso (1989) and the J2 -classification given in
Nishiyama (1996) for the Kummer surface of the product of two non-isogenous
elliptic curves, one can check that the first one is more fine than the second one.
The advantage of the J2 -classification sits in its strong relation with the lattice
theory; indeed, there is a method which allows one to obtain the J2 -classification
of elliptic fibration on several K3 surfaces. This method is presented in Nishiyama
(1996) and will be described in this paper in Section 4.1.

2.1.4 Results on the Different Classification Types

One of the main results of Braun et al. (2013) is about the relations among the
various types of classifications of elliptic fibrations on K3 surfaces. First we observe
that there exists two surjective maps J0 ! J1 and J0 ! J2 , which are in fact
quotient maps (cf. Braun et al. 2013, Formulae (54) and (57)). This induces a map
J1 ! J2 which is not necessarily a quotient map.
The Braun et al. (2013, Proposition C’) gives a bound for the number of
different elliptic fibrations up to the J1 -classification, which are identified by the
J2 -classification. As a Corollary the following is proved:
Corollary 2.8 (Braun et al. (2013, Corollary D)). Let S.a;b;c/ be a K3 surface such
 
2a b
that the transcendental lattice of S is isometric to . If .a; b; c/ is one of the
b 2c
following .1; 0; 1/, .1; 1; 1/, .2; 0; 1/, .2; 1; 1/, .3; 0; 1/, .3; 1; 1/, .4; 0; 1/, .5; 1; 1/,
.6; 1; 1/, .3; 2; 1/, then J1 ' J2 .

2.2 A Classification Method for Elliptic Fibrations

on K3 Surfaces

The first paper about the classification of elliptic fibrations on K3 surfaces is due
to Oguiso (1989). He gives a J1 -classification of the elliptic fibrations on the
Kummer surface of the product of two non-isogenous elliptic curves. The method
24 M.J. Bertin et al.

proposed in Oguiso (1989) is very geometric: it is strictly related to the presence

of a certain automorphism (a non-symplectic involution) on the K3 surface. Since
one has to require that the K3 surface admits this special automorphism, the method
suggested in Oguiso (1989) can be generalized only to certain special K3 surfaces
(see Kloosterman 2006; Comparin and Garbagnati 2014).
Seven years after the paper (Oguiso 1989), a different method was proposed by
Nishiyama in Nishiyama (1996). This method is less geometric and more related
to the lattice structure of the K3 surfaces and of the elliptic fibrations. Nishiyama
applied this method in order to obtain a J2 -classification of the elliptic fibrations,
both on the K3 surface already considered in Oguiso (1989) and on other K3
surfaces (cyclic quotients of the product of two special elliptic curves) to which the
method by Oguiso cannot be applied. Later, in Bertin and Lecacheux (2013),
the method is used to give a J2 -classification of elliptic fibrations on a K3 surface
whose transcendental lattice is h4i ˚ h2i.
The main idea of Nishiyama’s method is the following: we consider a K3 surface
S and its transcendental lattice TS . Then we consider a lattice T such that: T is
negative definite; rank.T/ D rank.TS / C 4; the discriminant group and form of T
are the same as the ones of TS . We consider primitive embeddings of  W T ,! L,
where L is a Niemeier lattice. The orthogonal complement of .T/ in L is in fact the
frame of an elliptic fibration on S.
The classification of the primitive embeddings of T in L for every Niemeier lattice
L coincides with the J2 -classification of the elliptic fibrations on S. We will give
more details on Nishiyama’s method in Section 4.1.
Since this method is related only to the lattice properties of the surface, a priori
one cannot expect to find a J1 -classification by using only this method.
Thanks to Corollary 2.8, (see Braun et al. 2013) the results obtained by
Nishiyama’s method are sometimes stronger than expected. In particular, we will
see that in our case (as in the case described in Bertin and Lecacheux 2013)
the classification that we obtain for the elliptic fibrations on a certain K3 surface
using the Nishiyama’s method is in fact a J1 -classification (and not only a J2 -
classification).

2.3 Torsion Part of the Mordell–Weil Group

of an Elliptic Fibration

In Section 4.2, we will classify elliptic fibrations on a certain K3 surface, determin-

ing both the trivial lattice and the Mordell–Weil group. A priori, steps (8) and (9)
of the algorithm presented in Section 4.1 completely determine the Mordell–Weil
group. In any case, we can deduce some information on the torsion part of the
Mordell–Weil group by considering only the properties of the reducible fibers of
the elliptic fibration. This makes the computation easier, so here we collect some
results on the relations between the reducible fibers of a fibration and the torsion
part of the Mordell–Weil group.
Classifications of Elliptic Fibrations of a Singular K3 Surface 25

First, we recall that a section meets every fiber in exactly one smooth point, so a
section meets every reducible fiber in one point of a component with multiplicity 1
(we recall that the fibers of type In , II  , III  , IV  have reducible components with
multiplicity greater than 1). We will call the component of a reducible fiber which
meets the zero section the zero component or trivial component.
Every section (being a rational point of an elliptic curve defined over k.P1 /)
induces an automorphism of every fiber, in particular of every reducible fiber. Thus,
the presence of an n-torsion section implies that all the reducible fibers of the
fibration admit Z=nZ as subgroup of the automorphism group. In particular, this
implies the following (well-known) result:
Proposition 2.9 (cf. Schütt and Shioda (2010, Section 7.2)). Let E W S ! P1 be
an elliptic fibration and let MW.E/tors the torsion part of the Mordell–Weil group.
If there is a fiber of type II  , then MW.E/tors D 0.
If there is a fiber of type III  , then MW.E/tors  .Z=2Z/.
If there is a fiber of type IV  , then MW.E/tors  .Z=3Z/.
If there is a fiber of type In and n is an even number, then MW.E/tors  .Z=2Z/2 .
If there is a fiber of type In and n is an odd number, then MW.E/tors  .Z=4Z/.

2.3.1 Covers of Universal Modular Elliptic Surfaces

The theory of universal elliptic surfaces parametrizing elliptic curves with pre-
scribed torsion can also be useful when finding the torsion subgroup of a few elliptic
fibrations on the list. It relies on the following definition/proposition.
Proposition 2.10 (See Couveignes and Edixhoven (2011, 2.1.4) or Shioda
(1972)). Let  W X ! B be an elliptic fibration on a surface X. Assume  has
a section of order N, for some N 2 N, with N  4. Then X is a cover of the
universal modular elliptic surface, EN ; of level N.
After studying the possible singular fibers of the universal surfaces above, one gets
the following.
Proposition 2.11. Let EN be the universal modular elliptic surface of level N. The
following hold:
i) If N  5, then EN admits only semi-stable singular fibers. They are all of type
Im with mjN.

ii) The surface E4 is a rational elliptic surface with singular fibers I1 ; I4 ; I1 .

2.3.2 Height Formula for Elliptic Fibrations

The group structure of the Mordell–Weil group is the group structure of the rational
points of the elliptic curve defined over the function field of the basis of the fibration.
It is also possible to equip the Mordell–Weil group of a pairing taking values in
26 M.J. Bertin et al.

Q, which transforms the Mordell–Weil group to a Q-lattice. Here we recall the

definitions and the main properties of this pairing. For a more detailed description,
we refer to Schütt and Shioda (2010) and to the original paper (Shioda 1990).
Definition 2.12. Let E W S ! C be an elliptic fibration and let O be the zero section.
The height pairing is the Q-valued pairing, < ;  >W MW.E/  MW.E/ ! Q
defined on the sections of an elliptic fibration as follows:
X
< P; Q >D .S/ C P  O C Q  O  contrc .P; Q/;
c2C

where .S/ is the holomorphic characteristic of the surface S,  is the intersec-

tion form on NS.S/, C D fc 2 C such that the fiber E 1 .c/ is reducibleg and
contrc .P; Q/ is a contribution which depends on the type of the reducible fiber and
on the intersection of P and Q with such a fiber as described in Schütt and Shioda
(2010, Table 4). P
The value h.P/ WD< P; P >D 2.S/ C 2P  O  c2C contrc .P; P/; is called the
height of the section P.
We observe that the height formula is induced by the projection of the intersec-
tion form on NS.S/ ˝ Q to the orthogonal complement of the trivial lattice Tr.E/
(cf. Schütt and Shioda 2010, Section 11).
Proposition 2.13 (Schütt and Shioda (2010, Section 11.6)). Let P 2 MW.E/ be
a section of the elliptic fibration E W S ! C. The section P is a torsion section if and
only if h.P/ D 0.

3 The K3 Surface X

The goal of this paper is the classification of the elliptic fibrations on the unique K3
surface X such that TX ' h6i ˚ h2i. This surface is interesting for several reasons,
and we will present it from different points of view.

3.1 A Toric Hypersurface and the Symmetric Group S4

Let N be a lattice isomorphic to Zn . The dual lattice M of N is given by Hom.N; Z/;

it is also isomorphic to Zn . We write the pairing of v 2 N and w 2 M as hv; wi.
Given a lattice polytope ˘ in N, we define its polar polytope ˘ı to be ˘ı D
fw 2M j hv; wi  1 8 v 2 Kg. If ˘ı is also a lattice polytope, we say that ˘
is a reflexive polytope and that ˘ and ˘ı are a mirror pair. A reflexive polytope
must contain 0; furthermore, 0 is the only interior lattice point of the polytope.
Classifications of Elliptic Fibrations of a Singular K3 Surface 27

Reflexive polytopes have been classified in one, two, three, and four dimensions. In
three dimensions, there are 4319 reflexive polytopes, up to an overall isomorphism
preserving lattice structure (Kreuzer and Skarke 1998, 2000). The database of
reflexive polytopes is incorporated in the open-source computer algebra software
(Stein et al. 2014).
Now, consider the one-parameter family of K3 surfaces given by

1 1 1 x y y z x z
xC C y C C z C C C C C C C C : (1)
x y z y x z y z x

This family of K3 surfaces was first studied in Verrill (1996), where its Picard–
Fuchs equation was computed. A general member of the family has Picard lattice
given by U ˚ h6i.
The Newton polytope ˘ı determined by the family of polynomials in Equation 1
is a reflexive polytope with 12 vertices and 14 facets. This polytope has the greatest
number of facets of any three-dimensional reflexive polytope; furthermore, there is a
unique three-dimensional reflexive polytope with this property, up to isomorphism.
In the database of reflexive polytopes found in Stein et al. (2014), this polytope has
index 1529.
We illustrate its polar polytope ˘ next to ˘ı in Figures 1 and 2.
Let us recall some standard constructions and notations involving toric varieties.
A cone in N is a subset of the real vector space NR D N ˝ R generated by
nonnegative R-linear combinations of a set of vectors fv1 ; : : : ; vm g  N. We
assume that cones are strongly convex, that is, they contain no line through the
origin. Note that each face of a cone is a cone. fan † consists of a finite collection of
cones such that each face of a cone in the fan is also in the fan, and any pair of cones
in the fan intersects in a common face. We say † is simplicial if the generators of

Fig. 1 ˘ (reflexive polytope

2355)
28 M.J. Bertin et al.

Fig. 2 ˘ı (reflexive polytope

1529)

each cone in † are linearly independent over R. If every element of NR belongs to

some cone in †, we say † is complete. A fan † defines a toric variety V† . If the
fan is complete, we may describe V† using homogeneous coordinates, in a process
analogous to the construction of Pn as a quotient space of .C /n . The homogeneous
coordinates have one coordinate zj for each generator of a one-dimensional cone
of †.
We may obtain a fan R from a mirror pair of reflexive polytopes in two equivalent
ways. We may take cones over the faces of ˘  NR , or we may take the normal fan
to the polytope ˘ı  MR . Let † be a simplicial refinement of R such that the one-
dimensional cones of † are generated by the nonzero lattice points vk , k D 1 : : : q,
of ˘; we call such a refinement a maximal projective subdivision. Then the variety
V† is an orbifold. Then in homogeneous coordinates, we have one coordinate zk for
each nonzero lattice point in ˘. We may describe the anticanonical hypersurfaces in
homogeneous coordinates using polynomials of the form:

X Y
q
pD cx zkhvk ;xiC1 : (2)
x2˘ı \M kD1

Here the cx are arbitrary coefficients. Note that p has one monomial for each lattice
point of ˘ı . If the reflexive polytope ˘ is three-dimensional, V† is smooth and
smooth anticanonical hypersurfaces in V† are K3 surfaces (see Cox and Katz 1999
for details).
The orientation-preserving symmetry group of ˘ and ˘ı is the symmetric
group S4 . This group acts transitively on the vertices of ˘ı . As the authors of Karp
et al. (2013) observe, by setting the coefficients cx corresponding to the vertices of
˘ı to 1 and the coefficient corresponding to the origin to a parameter , we obtain a
naturally one-parameter family of K3 hypersurfaces with generic Picard rank 19:
Classifications of Elliptic Fibrations of a Singular K3 Surface 29

0 1
X Y
q
pD@ zkhvk ;xiC1 A C z1 : : : zq : (3)
x2vertices.˘ı / kD1

Equation 3 is simply Equation 1 in homogeneous coordinates.

If we view S4 as acting on the vertices of ˘ rather than the vertices of ˘ı , we
obtain a permutation of the homogeneous coordinates zk . The authors of Karp et al.
(2013) show that this action of S4 restricts to a symplectic action on each K3 surface
in the pencil given by Equation 3; in particular, we have a symplectic action of S4
on X. In the affine coordinates of Equation 1, the group action is generated by an
element s2 of order 2 which acts by .x; y; z/ 7! .1=x; 1=z; 1=y/ and an element s4 of
order 4 which acts by .x; y; z/ 7! .x=y; x=z; x/.

3.2 The K3 Surface X

Definition 3.1. Let X be the K3 surface defined by F D 0, where F is the

numerator of
1 1 1 x y y z x z
xC CyC CzC C C C C C C :
x y z y x z y z x

The K3 surface X is the special member of the family of K3 surfaces described in (1)
which is obtained by setting  D 0.
We will use three elements of the symplectic group S4 : the three-cycle s3 given
by .x; y; z/ 7! .y; z; x/; the four-cycle s4 and the two-cycle s2 :
We describe explicitly a first elliptic fibration, which gives the main properties
of X:

3.3 A Fibration Invariant by s3

We use the following factorizations

F D .x C y C z C 1/ .xy C yz C zx/ C .x C y C z  3/ xyz (4)

F .x C y C z/ D .x C y C z  1/2 xyz C .x C y C z C 1/.x C y/.y C z/.z C x/:
(5)

If w D x C y C z; we see that w is invariant under the action of s3 : If we substitute

w  x  y for z, we obtain the equation of an elliptic curve, so the morphism

E W X ! P1w (6)
30 M.J. Bertin et al.

.x; y; z/ 7! w D x C y C z

is an elliptic fibration of X.
We use the birational transformation

v C .w C 1/u v.w C 1/  u2
xD ; yD
u.w  3/ v.w  3/

with inverse

u D ..w  3/x C .w C 1//..w  3/y C .w C 1//;

v D ..w  3/x C .w C 1//2 ..w  3/y C .w C 1//

to obtain the Weierstrass equation

  2
v 2 C w2 C 3 uv C w2  1 v D u3 : (7)

Notice the torsion points .u D 0; v D 0/ and u D 0; v D .w2  1/2 of order
 2 2
3 and the 3 points of order 2 with u-coordinate 1 4
w  1 , .w  1/2 , and
 .w C 1/2 :
We use also the Weierstrass form



2 D

 .w  3/ .w C 1/3
 .w C 3/ .w  1/3 (8)

with
1
 2  1
  3 
uD
 w2  1 ; vD  w2 C 3
C w2  1 :
4 8
The singular fibers are of type I6 for w D 1; 1; 1 and I2 for w D 3; 3; 0:
So the trivial lattice of this fibration is Tr.E/ D U ˚ A˚3 ˚3
5 ˚ A1 . Hence the
Picard number of X is 20 and rank.MW.E// D 0. So X is a singular K3 surface.
This elliptic fibration is contained in the Shimada and Zhang (2001, Table 2 line 4)
and thus its transcendental lattice is h6i ˚ h2i.
Moreover, all the fibers have split multiplicative reduction and thus the Néron
Severi group is generated by curves defined on Q.
Remark 3.2. We have already observed that X is a special member of the one-
dimensional family of K3 surfaces defined by Equation 1. Indeed, the transcendental
lattice of X is primitively embedded in U ˚ h6i by the vectors .1; 1; 0/; .0; 0; 1/.
This gives the following proposition:
Proposition 3.3. The Néron–Severi group of the K3 surface X has rank 20 and is
 by divisors which are defined over Q. The transcendental lattice of X is
generated
20
TX ' :
06
Classifications of Elliptic Fibrations of a Singular K3 Surface 31

Remark 3.4. The equation (8) is the universal elliptic curve with torsion group
Z=2Z  Z=6Z and is in fact equivalent to the equation given in Kubert (1976).Thus
this fibration can be called modular: we can view the base curve .P1w / as the modular
 
ab
curve X with  D f 2 Sl2 .Z/ ; a 1 mod 6; c 0 mod 6; b 0 mod 2g.
cd
By (5) we can easily obtain the equation

.w  1/2 xyz C .w C 1/ .x C y/ .y C z/ .z C x/ D 0

and realize X by a base change of the modular rational elliptic surface E6 described
by Beauville in Beauville (1982). We can prove that on the fiber, the automorphism
s3 corresponds to adding a 3-torsion point.

3.4 A Fibration Invariant by s4

If t D zxy ; we see that t is invariant under the action of s4 : Substituting tzx for y
in F, we obtain the equation of an elliptic curve. Using standard transformations
(as in Kumar 2014(39.2); Atkin and Morain 1993 or Cassels 1991) we obtain the
Weierstrass model
 
v 2 D u u2  2t t2 C 1 u C t2 .t C 1/4 :

The point Qt D u D t.t C 1/2 ; v D 2t2 .t C 1/2 is of order 4:
The point Pt D .u D .t C 1/2 ; v D .t2 C 1/ .t C 1/2 / is of infinite
 order.
The singular fibers are 2I1 .t D 0; 1/ C I8 .t D 1/ C 2I1 t2 C t C 1 D 0 :
One can prove that on the fiber, s4 corresponds to the translation by a 4-torsion
point. Moreover, the translation by the point Pt defines an automorphism of infinite
order on X.
Remark 3.5. If we compute the height of Pt we can show, using Shioda formula,
Shioda (1990) that Pt and Qt generate the Mordell–Weil group.

3.5 A Fibration Invariant by s2

If r D yz , we see that r is invariant under the action of s2 : Substituting rz for y we

obtain the equation of an elliptic curve and the following Weierstrass model
 
v 2  r2  1 vu D u .u  2r .r C 1// u  2r2 .r C 1/ :

The point .0; 0/ is a two-torsion point. The point .2r .r C 1/ ; 0/ is of infinite order.
The singular fibers are 
2I6 .r D 0; 1/ C I0 .r D 1/ C I4 .r D 1/ C 2I1 r2  14r C 1 D 0 .
32 M.J. Bertin et al.

Remark 3.6. From Elkies results (Elkies 2010; Schütt 2010) there is a unique K3
surface X=Q with Néron–Severi group of rank 20 and discriminant 12 that consists
entirely of classes of divisors defined over Q. Indeed it is X. Moreover, in Elkies
(2010) a Weierstrass equation for an elliptic fibration on X defined over Q, is given:

4
y2 D x3  75x C .4t  242 C /:
t
Remark 3.7. The surface X is considered also in a slightly different context in
Garbagnati and Sarti (2009) because of its relation with the study of the moduli
space of K3 surfaces with a symplectic action of a finite abelian group. Indeed,
the aim of the paper (Garbagnati and Sarti 2009) is to study elliptic fibrations
EG W SG ! P1 on K3 surfaces SG such that MW.EG / D G is a torsion group. Since
the translation by a section is a symplectic automorphism of SG , if MW.EG / D G,
then G is a group which acts symplectically on SG . In Garbagnati and Sarti (2009)
it is shown how one can describe both a basis for the Néron–Severi group of SG
and the action induced by the symplectic action of G on this basis. In particular,
one can directly compute the lattices NS.SG /G and G WD NS.SG /? . The latter
does not depend on SG but only on G and its computation plays a central role
in the description of the moduli space of the K3 surfaces admitting a symplectic
action of G (see Nikulin 1979; Garbagnati and Sarti 2009). In particular, the case
G D Z=6ZZ=2Z is considered: in this case, the K3 surface SG is X, and the elliptic
fibration EG is (6). Comparing the symplectic action of Z=6Z  Z=2Z on X given
in Garbagnati and Sarti (2009) with the symplectic group action of S4 described
in Section 3.1, we find that the two groups intersect in the subgroup of order 3
generated by the map s3 given by .x; y; z/ 7! .z; y; x/.

4 Main Result

This section is devoted to the proof of our main result:

Theorem 4.2. The classification up to automorphisms of the elliptic fibrations on
X is given in Table 1. Each elliptic fibration is given with the Dynkin diagrams
characterizing its reducible fibers and the rank and torsion of its Mordell–Weil
group. More precisely, we obtained 52 elliptic fibrations on X, including 17
fibrations of rank 2 and one of rank 3.
We denote by r the rank.MW.E//, and we use Bourbaki notations for An ; Dn ; Ek
as in Bertin and Lecacheux (2013).
Outline of the Proof The proof consists of an application of Nishiyama’s method:
the details of this method will be described in Section 4.1. Its application to
our case is given in Section 4.2. The application of Nishiyama’s method gives
us a J2 -classification, which coincides in our cases with a classification up to
automorphisms of the surface by Corollary 2.8.
Classifications of Elliptic Fibrations of a Singular K3 Surface 33

Table 1 The elliptic fibrations of X

Lroot No. Nroot r MW.E /tors
E83 1 A5 ˚ A1  E8 A1 E8 E8 1 .0/
2 A1  E8 A5  E8 A1 A2 E7 E8 0 .0/
E8 D16 3 A5 ˚ A1  E8 A1 D16 1 Z=2Z
4 A5 ˚ A1  D16 A1 D8 E8 1 .0/
5 A5  E8 A1  D16 A1 A1 A2 D14 0 Z=2Z
6 A1  E8 A5  D16 E7 D10 1 .0/
E72 D10 7 A5 ˚ A1  E7 E7 D10 1 Z=2Z
8 A5 ˚ A1  D10 A1 A1 A1 E7 E7 1 Z=2Z
9 A1  E7 A5  E7 D6 A1 D10 1 Z=2Z
10 A1  E7 A5  E7 D6 A2 D10 0 Z=2Z
11 A5  E7 A1  D10 A1 A1 D8 E7 1 Z=2Z
12 A5  E7 A1  D10 A1 A2 D8 E7 0 Z=2Z
13 A1  E7 A5  D10 E7 D6 D4 1 Z=2Z
E7 A17 14 A5 ˚ A1  E7 A17 1 Z=3Z
15 A5 ˚ A1  A17 A9 E7 2 .0/
16 A5  E7 A1  A17 A1 A15 2 .0/
17 A5  E7 A1  A17 A2 A15 1 .0/
18 A1  E7 A5  A17 D6 A11 1 .0/
D24 19 A5 ˚ A1  D24 A1 D16 1 .0/
D212 20 A5 ˚ A1  D12 A1 D4 D12 1 Z=2Z
21 A1  D12 A5  D12 A1 D10 D6 1 Z=2Z
D38 22 A5 ˚ A1  D8 A1 D8 D8 1 Z=2Z
22.b/ A5 ˚ A1  D8 A1 D8 D8 1 Z=2Z
23 A1  D8 A5  D8 A31 D6 D8 1 .Z=2Z/2
D9 A15 24 A5 ˚ A1  D9 A1 A15 2 Z=2Z
25 A5 ˚ A1  A15 D9 A7 2 .0/
26 A5  D9 A1  A15 A3 A13 2 .0/
27 A1  D9 A5  A15 A1 A9 D7 1 .0/
E64 28 A1  E6 A5  E6 A1 A5 E62 0 Z=3Z
A11 E6 D7 29 A5 ˚ A1  A11 A3 D7 E6 2 .0/
30 A5  E6 A1  D7 A21 A11 D5 0 Z=4Z
31 A5  E6 A1  A11 A1 A9 D7 1 .0/
32 A1  E6 A5  D7 A5 A11 2 Z=3Z
33 A5  D7 A1  A11 A9 E6 3 .0/
34 A1  D7 A5  A11 A1 A5 D5 E6 1 .0/
35 A1  E6 A5  A11 A25 D7 1 .0/
D46 36 A1  D6 A5  D6 A1 D4 D26 1 .Z=2Z/2
(continued)
34 M.J. Bertin et al.

Table 1 (continued)
Lroot No. Nroot r MW.E /tors
D6 A29 37 A5 ˚ A1  A9 A1 A9 D6 2 Z=2Z
38 A5  A9 A1  A9 A3 A7 D6 2 .0/
39 A5  A9 A1  D6 A1 A3 A9 D4 1 Z=2Z
40 A1  A9 A5  D6 A7 A9 2 .0/
D25 A27 41 A5 ˚ A1  A7 A7 D25 1 Z=4Z
42 A5  A7 A1  A7 A1 A5 D25 2 .0/
43 A5  A7 A1  D5 A21 A3 A7 D5 1 Z=4Z
A38 44 A5 ˚ A1  A8 A28 2 Z=3Z
45 A1  A8 A5  A8 A2 A6 A8 2 .0/
A24 46 A5 ˚ A1  A24 A16 2 .0/
A212 47 A5 ˚ A1  A12 A4 A12 2 .0/
48 A5  A12 A1  A12 A6 A10 2 .0/
D4 A45 49 A5 D A5 A1  A5 A3 A25 D4 1 Z=2Z
50 A5 D A5 A1  D4 A31 A35 0 Z=2Z  Z=6Z
A46 51 A5  A6 A1  A6 A4 A26 2 .0/

Remark 4.2. The fibration given in Section 3.3 is # 50 in Table 1, the one given in
Section 3.4 is # 41, the one given in Section 3.5 is # 49, the one given in Remark 3.6
is #1. The fibrations of rank 0 may be found also in Shimada and Zhang (2001).
Remark 4.3. We observe that there exists a primitive embedding of TX ' h6i ˚ h2i
in U.2/ ˚ h2i given by the vectors h.1; 1; 1/; .0; 1; 1/i. Thus, X is a special
member of the one-dimensional family of K3 surfaces whose transcendental lattice
is isometric to U.2/ ˚ h2i. The elliptic fibrations on the generic member Y of this
family have already been classified (cf. Comparin and Garbagnati 2014), and indeed
the elliptic fibrations in Table 1 specialize the ones in Comparin and Garbagnati
(2014, Table 4.5 and Section 8.1 case r D 19), either because the rank of the
Mordell–Weil group increases by 1 or because two singular fibers glue together
producing a different type of reducible fiber.

4.1 Nishiyama’s Method in Detail: An Algorithm

This section is devoted to a precise description of Nishiyama’s method. Since the

method is very well described both in the original paper (Nishiyama 1996) and in
some other papers where it is applied, e.g. Bertin and Lecacheux (2013) and Braun
et al. (2013), we summarize it in an algorithm which allows us to compute all the
results given in Table 1. In the next section we will describe in detail some peculiar
cases, in order to show how this algorithm can be applied.
Classifications of Elliptic Fibrations of a Singular K3 Surface 35

Definition 4.4. A Niemeier lattice is an even unimodular negative definite lattice

of rank 24.
There are 24 Niemeier lattices. We will denote by L an arbitrary Niemeier lattice.
Each of them corresponds uniquely to its root lattice Lroot .
In Table 2 we list the Niemeier lattices, giving both the root lattices of each
one and a set of generators for L=Lroot . To do this we recall the following notation,
introduced in Bertin and Lecacheux (2013):

P 
1 Pn 1
˛n D jD1 .n  j C 1/aj ıl D 2
l2
idi C 12 .l  2/dl1 C 12 ldl
nC1
Pl2
PiD1 
1 1 1 1
ıl D iD1 di C 2 .dl1 C dl / ıQl D 2
l2
iD1 idi C 2 ldl1 C 2 .l  2/dl

6 D  2e1 C3e2 C4e3 C6e
3
4 C5e5 C4e6

7 D  .2e1 C3e2 C4e3 C6e24 C5e5 C4e6 C3e7 /

Now let us consider a K3 surface S such that .S/  12. Let us denote by TS its
transcendental lattice. We describe an algorithm which gives a J2 -classification of
the elliptic fibration on S.

(1) The lattice T: We define the lattice T to be a negative definite lattice such that
rank.T/ D rank.TS / C 4 and the discriminant group and form of T are the same
as the ones of TS . The lattice T is not necessarily unique. If it is not, we choose
one lattice with this property (the results obtained do not depend on this choice).
(2) Assumption: We assume that one can choose T to be a root lattice.
(3) The embeddings : Given a Niemeier lattice L we choose a set of primitive
embeddings  W T ,! Lroot not isomorphic by an element of the Weyl group.
(4) The lattices N and Nroot : Given a primitive embedding  we compute the
orthogonal complement N of .T/ in Lroot , i.e. N WD .T/?Lroot and Nroot its
root lattice.
(5) The lattices W and Wroot : We denote by W the orthogonal complement of .T/
in L, i.e. W WD .T/?L and by Wroot its root lattice. We observe that Nroot D
Wroot and N ,! W with finite index.
(6) The elliptic fibration E : The frame of any elliptic fibration on S is a lattice W
obtained as in step 5. Moreover, the trivial lattice of any elliptic fibration on S is
of the form U ˚ Nroot D U ˚ Wroot where Wroot and Nroot are obtained as above.
Hence, we find a J2 -classification of the elliptic fibration on S. In particular
every elliptic fibration on S is uniquely associated with a primitive embedding
 W T ,! L. Let us denote by E the elliptic fibration associated with .
(7) The singular fibers: We already observed (cf. Section 2.1) that almost all the
properties of the singular fibers are encoded in the trivial lattice, so it is clear
that every Nroot .WD ..T/?Lroot /root / determines almost all the properties of the
reducible fibers of E .
(8) The rank of the Mordell–Weil group: Let  be a given embedding. Let r WD
rank.MW.E //. Then r D rank.NS.S//  2  rank.Nroot / D 20  rank.TS / 
rank.Nroot /.
(9) The torsion of the Mordell–Weil group: The torsion part of the Mordell–Weil
group is Wroot =Wroot . W=N) and can be computed in the following way: let
36 M.J. Bertin et al.

Table 2 The Niemeier lattices L: Lroot and L=Lroot

Lroot L=Lroot
E83 h.0/i
E8 D16 Z=2Z D hı16 i
.1/ .1/ .2/
E72 D10 .Z=2Z/2 D h
7 C ı10 ;
7 C
7 C ı10 i
E7 A17 Z=6Z D h
7 C 3˛17 i
D24 Z=2Z D hı24 i
.1/ .2/ .1/ .2/
D212 .Z=2Z/2 D hı12 C ı12 ; ı12 C ı12 i
.1/ .2/ .3/ .1/ .2/ .3/ .1/ .2/ .3/
D38 .Z=2Z/3 D hı8
C Cı8 ı8 ; ı8
C ı8 C ı 8 ; ı 8 C ı 8 C ı8 i
D9 A15 Z=8Z D hı9 C 2˛15 i
.1/ .2/ .3/ .1/ .3/ .4/
E64 .Z=3Z/2 D h
6 C
6 C
6 ; 
6 C
6 C
6 i
A11 E6 D7 Z=12Z D h˛11 C
6 C ı7 i
.2/ .3/ .4/ .1/ .2/ .4/
D46 .Z=2Z/4 D hı6 C ı 6 C ıQ6 ; ı6 C ıQ6 C ı6 ,
.1/ .2/ .4/ .1/ .3/ .4/
ı6 C ı 6 C ıQ6 ; ı6 C ıQ6 C ı 6 i
Z=2Z  Z=10Z D he
.2/ .1/ .2/
D6 A29 ı6 C 5˛9 ; ı6 C ˛9 C 2˛9 i
.1/ .2/ .1/ .1/ .2/ .1/ .2/
D25 A27 Z=4Z  Z=8Z D hı5 C ı5 C 2˛7 ; ı5 C 2ı5 C ˛7 C ˛7 i
.1/ .2/ .1/ .2/ .3/
A38 Z=3Z  Z=9Z D h3˛8 C 3˛8 ; ˛8 C 2˛8 C 2˛8 i
.1/ .2/ .3/
A45 D4 .Z=6Z/2  Z=2Z D h5˛5 C 2˛5 C ˛5 C ı 4 ,
.1/ .2/ .3/ .4/
5˛5 C 3˛5 C 2˛5 C 4˛5 C ı4 ,
.1/ .4/
3˛5 C 3˛5 C ıQ4 i
.1/ .2/ .3/ .4/ .1/ .2/ .3/ .4/
A46 .Z=7Z/2 D h˛6 C 2˛6 C ˛6 C 6˛6 ; ˛6 C 6˛6 C 2˛6 C ˛6 i
.1/ .i/ P 6 e .i/
D64 .Z=2Z/6 D hı4 C ı4 ; i D 1; : : : ; 5; iD1 ı4 i
A24 Z=5Z D h5˛24 i
.1/ .2/
A212 Z=13Z D h2˛13 C 3˛13 i
.1/ .2/ .3/ .4/ .5/
A64 3
.Z=5Z/ D h˛4 C ˛4 C ˛4 C 4˛4 C 4˛4 ,
.1/ .2/ .3/ .5/ .6/
˛4 C ˛4 C 4˛4 C ˛4 C 4˛4 ,
.1/ .3/ .4/ .5/ .6/
˛4 C 4˛4 C ˛4 C 4˛4 C ˛4 i
.1/ P8 .i/
A83 4
.Z=4Z/ D h3˛3 C iD2 ci ˛3 such that .c2 ; : : : ; c8 /
is a cyclic permutation of .2001011/i
.1/ P12 .i/
A12
2 .Z=3Z/6 D h2˛2 C iD2 ci ˛2 such that .c2 ; : : : ; c12 /
is a cyclic permutation of .11211122212/i
.1/ P24 .i/
A24
1 .Z=2Z/12 D h˛1 C iD2 ci ˛2 such that .c2 ; : : : ; c24 /
is a cyclic permutation of .00000101001100110101111/i
 ƒ24 ' L

l C Lroot be a non-trivial element of L=Lroot . If there exist k ¤ 0 and u 2 Lroot

such that k.l C u/ 2 Nroot , then l C u 2 W and the class of l is a torsion element.

Remark 4.5. It is not always true that the lattice T can be chosen to be a root lattice,
and the method can be applied with some modifications without this assumption,
see Braun et al. (2013). Since everything is easier under this assumption and in
our case we can require that T is a root lattice, we described the method with the
assumption (2). In particular, if T is not a root lattice, then one has to consider the
Classifications of Elliptic Fibrations of a Singular K3 Surface 37

primitive embeddings of T in L, but one cannot use the results in Nishiyama (1996,
Sections 4 and 5), so the points (3), (4), and (5) are significantly more complicated.

4.2 Explicit Computations

Here we apply the algorithm described in Section 4.1 to the K3 surface X.

4.2.1 Step 1

From Proposition 3.3, we find that the transcendental lattice of X is

 
60
TX D :
02

According to Nishiyama (1996), Schütt and Shioda (2010) and Bertin and
Lecacheux (2013), TX .1/ admits a primitive embedding in E8 and we can take T
as its orthogonal complement in E8 , that is

T D A5 ˚ A1 :

4.2.2 Step 2

We observe that T is a root lattice.

4.2.3 Step 3

We must find all the primitive embeddings  W T ,! Lroot not Weyl isomorphic.
This has been done by Nishiyama (1996) for the primitive embeddings of Ak in Am ,
Dn , El and for the primitive embeddings of A5 ˚ A1 into E7 and E8 . So we have to
determine the primitive embeddings not isomorphic of A5 ˚ A1 in Am and Dn . This
will be achieved using Corollary 4.7 and Lemma 4.10. First we recall some notions
used in order to prove these results.
Let B be a negative-definite even lattice, let a 2 Broot a root of B. The reflection
Ra is the isometry Ra .x/ D xC.a  x/ a and the Weyl group of B, W .B/, is the group
generated by Ra for a 2 Broot .
Proposition 4.6. Let A be a sublattice of B. Suppose there exists a sequence of roots
x1 ; x2 ; : : : ; xn of A?B with xi  xiC1 D "i and "2i D 1 then the two lattices A ˚ x1 and
A ˚ xn are isometric by an element of the Weyl group of B:
Proof. First we prove the statement for n D 2. Since the two sublattices A ˚ hx1 i
and A ˚ hx1 i are isometric by Rx1 we can suppose that x1  x2 D 1 (i.e. "1 D 1).
38 M.J. Bertin et al.

Then x1 C x2 is also a root and is in A?B : So the reflection Rx1 Cx2 is equal to Id on
A: Let g WD Rx1 ı Rx1 Cx2 then g 2 W.B/ is equal to Id on A: Moreover g .x2 / D
Rx1 .x2 C ..x1 C x2 /  x2 / .x1 C x2 // D x1 ; and so g fits. The case n > 2 follows by
induction. 

 4.7. Suppose n  9; p  6; up to an element of the Weyl group W .Dn /

Corollary
or W Ap there is a unique primitive embedding of A5 ˚ A1 in Dn or Ap :
Proof. From Nishiyama (1996) up to an element of the Weyl group there exists one
primitive embedding of A5 in Dn or Ap : Fix this embedding. If M is the orthogonal
of this embedding then Mroot is Dn6 or Ap6 : So for two primitive embeddings of
A1 in Mroot we can apply the previous proposition. 
We study now the primitive embeddings of A5 ˚ A1 in D8 (which are not
considered in the previous corollary, since the orthogonal complement of the unique
primitive embedding of A5 in D8 is h6i ˚ h2i2 ).
We denote by f"i ; 1  i  ng the canonical basis of Rn :
We can identify Dn .1/ with Dn ; the set of vectors of Zn whose coordinates have
an even sum.
First we recall the two following propositions, see, for example, Martinet (2002).
Proposition 4.8. The group Aut.Zn / is isomorphic to the semi-direct product
f˙1gn Ì Sn , where the group Sn acts on f˙1gn by permuting the n components.
Proposition 4.9. If n ¤ 4, the restriction to Dn of the automorphisms of Zn induces
an isomorphism of Aut .Zn / onto the group Aut .Dn /. The Weyl group W .Dn / of
index two in Aut .Dn / corresponds to those elements which induce an even number
of changes of signs of the "i .
Lemma 4.10. There are two embeddings of A5 ˚ A1 in D8 non-isomorphic up to
W .D8 / :
Proof. Let d8 D "1 C "2 and d8iC1 D "i1 C "i with 2  i  8 a basis of D8 : We
consider the embedding

A5 ,! hd7 ; d6 ; d5 ; d4; d3 i:

By Nishiyama’s results (Nishiyama 1996), Pthis embedding is unique up to an

element of W.D8 / and we have .A5 /?D8 D h 6iD1 "i i˚hx7 i˚hd1 i with x7 D "7 C"8 :
We see that ˙x7 and ˙d1 are the only roots of .A5 /?D8 .
We consider the two embeddings

A5 ˚ A1 ,! hd7 ; d6 ; d5 ; d4; d3 i ˚ hx7 i

A5 ˚ A1 ,! hd7 ; d6 ; d5 ; d4; d3 i ˚ hd1 i:

Suppose there exists an element R0 of W.D8 / such that R0 .x7 / D d1 and R0 .A5 / D
A5 ; we shall show that R0 .d1 / D ˙x7 : If z WD R0 .d1 /, then as R0 is an isometry
z  d1 D R0 .d1 /  R0 .x7 / D d1  x7 D 0: Moreover, z 2 .A5 /?D8 and so z D ˙x7 :
Classifications of Elliptic Fibrations of a Singular K3 Surface 39

Since R0 .A5 / D A5 , we see that R0 jA5 is an element of O .A5 /, the group of

isometries of A5 : We know that O .A5 / =W .A5 /
Z=2Z; generated by the class of
 W d7 $ d3 ; d6 $ d4 ; d5 $ d5 :
Thus, we have R0 jA5 D  2 W .A5 / or R0 jA5 D  with  2 W .A5 / : We can also
consider  as an element of the group generated by reflections Ru of D8 with u 2 A5 :
So, for v in .A5 /?D8 we have Ru .v/ D v if u 2 A5 and then  .v/ D v:
Let R D 1 R0 then R D R0 on .A5 /?D8 : Since R0 .d1 / D ˙x7 and R0 .x7 / D d1
we have R0 jh"7 ;"8 i D ."7 ! "8 ; "8 ! "7 / or ."7 ! "7 ; "8 ! "8 / : Also we have
Rjh"1 ;"2 ;:::;"6 i D Id or "i $ "7i :
In the second case R corresponds to a permutation of "i with only one sign minus;
thus, R is not an element of W .D8 / :

4.2.4 Step 4

For each primitive embedding of A5 ˚ A1 in Lroot , the computations of N and Nroot

are obtained in almost all the cases by Nishiyama (1996, Section 5). In the few cases
not considered by Nishiyama, one can make the computation directly. The results
are collected in Table 3, where we use the following notation. The vectors x3 ; x7 ; z6
in Dn are defined by

x3 WD dn3 C 2dn2 C dn1 C dn ;

x7 WD dn7 C 2.dn6 C dn5 C dn4 C dn3 C dn2 / C dn1 C dn ;
x07 WD 2.dn6 C dn5 C dn4 C dn3 C dn2 / C dn1 C dn ;
z6 WD dn5 C 2dn4 C 3dn3 C 4dn2 C 3dn1 C 2dn ;
ze6 WD dn5 C 2dn4 C 3dn3 C 4dn2 C 2dn1 C 3dn ;

and the vectors x; y in Ep are

x WD e1 C e2 C 2e3 C 2e4 C e5 ; y WD e1 C 2e2 C 2e3 C 3e4 C 2e5 C e6 :

4.2.5 Step 5 (An Example: Fibrations 22 and 22(b))

In order to compute W we recall that W is an overlattice of finite index of N; in fact,

it contains the non-trivial elements of L=Lroot which are orthogonal to .A5 ˚ A1 /.
Moreover, the index of the inclusion N ,! W depends on the discriminant of N.
Indeed
p jd.W/j D jd.NS.X//j D 12, so the index of the inclusion N ,! W is
jd.N/j=12.
As example we compute here the lattices W for the two different embeddings of
A5 ˚ A1 in D8 (i.e., for the fibrations 22 and 22(b)). Thus, we consider the Niemeier
40 M.J. Bertin et al.

Table 3 The orthogonal complement of the primitive embeddings A5 ˚ A1 in Lroot

No. Primitive
D Embedding E Orthogonal Complement
.1/ .1/ .1/ * +
e1 ; e3 ; : : : ; e6 .1/ .1/ .1/ .1/
D E 4e1 C 6e2 C 8e3 C 12e4 C
1 .1/ .1/ .1/ .1/ .1/ ˚
˚ e8 10e5 C 8e6 C 6e7 C 3e8
˝ .1/ ˛ D E D E
.2/ .2/ .3/ .3/
y ˚ e1 ; : : : ; e8 ˚ e1 ; : : : ; e8
D E * +
.1/ .1/ .1/ .1/ .1/ .1/ .1/
e1 ; e3 ; : : : ; e6 .1/ 2e C 3e2 C 4e3 C 6e4 C
2 D E e8 ; 1.1/ .1/ .1/ .1/ ˚
.2/
˚ e1 5e5 C 4e6 C 3e7 C 2e8
˝ .1/ ˛ D .2/ .2/ .2/ .2/
E
y ˚ x ; e2 ; e4 ; : : : ; e8 ˚
D E
.3/ .3/
e1 ; : : : ; e8
* +
4e1 C 6e2 C 8e3 C 12e4 C
3 he1 ; e3 ; : : : ; e6 i ˚ he8 i hyi ˚ ˚
10e5 C 8e6 C 6e7 C 3e8
hd1 ; : : : *; d16 i +
x7 ; d9 ; : : : ; d3 ;
4 hd16 ; d14 ; : : : ; d11 i ˚ hd1 i hz6 i ˚ ˚ he1 ; : : : ; e8 i
* 2d2 C d1 +
2e1 C 3e2 C 4e3 C 6e4 C
5 he1 ; e3 ; : : : ; e6 i ˚ hd16 i hyi ˚ e8 ; ˚
5e5 C 4e6 C 3e7 C 2e8
hd15 i ˚ hx3 ; d13 ; : : : ; d1 i
6 hd16 ; d14 ; : : : ; d11 i ˚ he1 i * 6 i ˚ hx7 ; d9 ; : : : ; d1 i ˚ hx; e2+; e4 ; : : : ; e8 i
hz
D E .1/ .1/
3e1 C 4e2 C 6e3 C
.1/
.1/ .1/ .1/
e2 ; e4 ; : : : ; e7 .1/ .1/ .1/ .1/ ˚
7 8e4 C 6e5 C D 4e6 C 2e7E
.1/
˚he1 i .2/ .2/
hd1 ; : : : ; d10 i ˚ e1 ; : : : ; e7
8 hd10 ; d8 ; : : : ; d5 i ˚ hd1 i Dhz6 i ˚ hx7 i ˚E hd3Di ˚ hd3 C x7E C 2d2 C d1 i ˚
.1/ .1/ .2/ .2/
e1 ; : : : ; e7 ˚ e1 ; : : : ; e7
D E * +
.1/ .1/ .1/ .1/ .1/ .1/
e1 ; e3 ; : : : ; e6 2e1 C 3e2 C 4e3 C
9 D E .1/ .1/ .1/ .1/ ˚
.2/
˚ e1 6e4 C 5e5 C 4e6 C 3e7
˝ .1/ ˛ D E
.2/ .2/ .2/
y ˚ x.2/ ; e2 ; e4 ; : : : ; e7 ˚

*1 ; : : : ; d10 i
hd +
D E .1/ .1/ .1/
.1/ .1/ .1/ .1/ 2e1 C 2e2 C 3e3 C
e2 ; e4 ; : : : ; e7 e1 ; .1/ .1/ .1/ .1/ ˚
10 D E 4e4 C 3e5 C 2e6 C e7
.2/ D E
˚ e1 .2/ .2/ .2/
x.2/ ; e2 ; e4 ; : : : ; e7 ˚ hd1 ; : : : ; d10 i
* +
D E .1/
2e1 C 3e2 C 4e3 C
.1/ .1/
.1/ .1/ .1/
e1 ; e3 ; : : : ; e6 .1/ .1/ .1/ .1/ ˚ hd9 i ˚
11 D 6e4 C 5e5E C 4e6 C 3e7 ˝ ˛
˚ hd10 i .2/ .2/
e ; : : : ; e7 ˚ hx3 ; d7 ; : : : ; d1 i ˚ y.1/
*1 +
D E .1/
.1/ 2e1 C 2e2 C 3e3 C
.1/ .1/
.1/ .1/ .1/
e2 ; e4 ; : : : ; e7 e1 ; .1/ .1/ .1/ .1/ ˚
12 D 4e4 C E 3e5 C 2e6 C e7
˚ hd10 i .2/ .2/
e1 ; : : : ; e7 ˚ hd9 i ˚ hx3 ; d7 ; : : : ; d1 i
* +
D E .1/
x.1/ ; e2 ;
.1/
13 hd10 ; d8 ; : : : ; d5 i ˚ e1 hx7 ; d3 ; d2 ; d1 i ˚ .1/ .1/ ˚
D E e4 ; : : : ; e7
.2/ .2/
e1 ; : : : ; e7 ˚ hz6 i
(continued)
Classifications of Elliptic Fibrations of a Singular K3 Surface 41

Table 3 (continued)
No. Primitive Embedding Orthogonal
* Complement +
2e1 C 3e2 C 4e3 C
14 he2 ; e4 ; : : : ; e6 i ˚ he1 i ˚ ha1 ; : : : ; a17 i
6e4 C 5e5 C 4e6 C 3e7
* a ; : : : ; a17 ; +
P6 9
15 ha1 ; : : : ; a5 i ˚ ha7 i he1 ; : : : ; e7 i ˚ jD1 jaj  6a8 ;

* a7 C 2a8+
2e1 C 3e2 C 4e3 C
16 he1 ; e3 ; : : : ; e6 i ˚ ha1 i hyi ˚ ˚
6e4 C 5e5 C 4e6 C 3e7
* 1 C 2a2 ; a3 ; : : : ; a17 i
ha +
2e1 C 2e2 C 3e3 C
17 he2 ; e4 ; : : : ; e7 i ˚ ha1 i ; e1 ˚
4e4 C 3e5 C 2e6 C e7
ha1 C 2a2 ; a3 ; : : : ; a17*i +
P6
jD1 jaj ;
18 ha1 ; : : : ; a5 i ˚ he1 i hx; e2 ; e4 ; : : : ; e7 i ˚
a7 ; : : : a17
19 hd24 ; d22 ; : : : ; d19 i ˚ hd1 i hz*6 i ˚ hx7 i ˚ hx7 C d1 C 2d2+; d3 ; : : : d17 i
D E .1/ .1/
d1 C 2d2 C d3 C x7 ;
.1/ .1/
.1/ .1/ .1/
d12 ; d10 ; : : : ; d7 .1/ .1/ .1/ ˚
D E
20 .1/ D d3 ; d4 ; dE5 D E D E
˚ d1 .2/ .2/ .1/ .1/
d ;:::;d ˚ z6 ˚ x7
D E D 1 E D 12 E
.1/ .1/ .1/ .1/ .1/ .1/ .1/
d12 ; d10 ; : : : ; d7 z6 ˚ x7 ; d5 ; : : : ; d1 ˚
21 D E D E D E
.2/ .2/ .2/ .2/ .2/
˚ d12 d11 ˚ x3 ; d9 ; : : : ; d1
D E ˝ .1/ ˛ D .1/ E
.1/ .1/ .1/
d7 ; d6 ; : : : ; d3 e
z ˚ x7 ˚
22 D E D6 E D E
.1/ .2/ .2/ .3/ .3/
˚ d1 d1 ; : : : ; d8 ˚ d1 ; : : : ; d8
D E ˝ .1/ ˛ D E
.1/ .1/ .1/ .1/
22 d7 ; d6 ; : : : ; d3 e
z ˚ d1 ˚
D E D6 E D E
.1/ .2/ .2/ .3/ .3/
.b/ ˚ x7 d1 ; : : : ; d8 ˚ d1 ; : : : ; d8
D E D E D E D E D E
.1/ .1/ .1/ .1/ .1/ .1/ .2/
d8 ; d6 ; : : : ; d3 z ˚ x7 ˚ d1 ˚ d7 ˚
23 D E D6 E D E
.2/ .2/ .2/ .2/ .3/ .3/
˚ d8 x3 ; d5 ; : : : ; d1 ˚ d1 ; : : : ; d8
24 hd9 ; d7 ; : : : ; d4 i ˚ hd1 i hz6 i ˚ hx7 ; d1 C 2d2 i ˚ ha1 ; : : : ; a15 i
* P6 jaj  6a8 ; +
jD1
25 ha1 ; : : : ; a5 i ˚ ha7 i hd1 ; : : : ; d9 i ˚ a7 C 2a8 ;
* ; : : : ; a15 +
a 9
a1 C 2a2 ;
26 hd9 ; d7 ; : : : ; d4 i ˚ ha1 i hz6 i ˚ hd1 ; d2 ; x7 i ˚
a3 ; : : : ; a15
* P +
6
jD1 jaj ;
27 ha1 ; : : : ; a5 i ˚ hd9 i hd8 i ˚ hx3 ; d6 ; : : : ; d1 i ˚
* +a7 ; : : : ; a15
D E .2/ .2/
e2 C e3 C 2e4 C e5 ;
.2/ .2/
.1/ .1/ .1/
e1 ; e3 ; : : : ; e6 .2/ .2/ .2/ .2/ ˚
28 D E e1 ;E e3 D; e5 ; e6
.2/ D E ˝ ˛
˚ e2 .3/ .3/ .4/ .4/
e1 ; : : : ; e6 ˚ e1 ; : : : ; e6 ˚ y.1/
(continued)
42 M.J. Bertin et al.

Table 3 (continued)
No. Primitive Embedding Orthogonal
*P Complement +
6
jD1 jaj  6a8 ;
29 ha1 ; : : : ; a5 i ˚ ha7 i ˚
a7 C 2a8 ; a9 ; a10 ; a11
hd1 ; : : : ; d7 i ˚ he1 ; : : : ; e6 i
30 he1 ; e3 ; : : : ; e6 i ˚ hd7 i hyi ˚ hd6 i ˚ hx3 ; d4 ; : : : ; d1 i ˚ ha1 ; : : : ; a11 i
31 he1 ; e3 : : : ; e6 i ˚ ha1 i hyi ˚ hd1 ; : : : ; d7 i ˚ ha1 C 2a2 ; a3 ; : : : ; a11 i
˝ ˛
32 hd7 ; d5 ; : : : ; d2 i ˚ he1 i hz6 i ˚ x07 ˚ hx; e2 ; e4 ; e5 ; e6 i
˚ ha1 ; : : : ; a11 i
˝ 0˛
33 hd7 ; d5 ; : : : ; d2 i ˚ ha1 i * 6 i ˚ x7 ˚+he1 ; : : : ; e6 i ˚
hz
a1 C 2a2 ;
a3 ; : : : ; a11
34 ha1 ; : : : ; a5 i ˚ hd7 i hd6 i ˚ hx3 ; d4 ; : : * : ; d1 i +
P6
jD1 jaj ;
˚ he1 ; : : : ; e6 i ˚
a7 ; : : : ; a11
35 ha1 ; : : : ; a5 i ˚ he1 i hx;
DP e 2 ; e 4 ; e 5 ; e 6 i ˚ E 1 ; : : : ; d7 i ˚
hd
6
ja ; a ; : : : ; a
D E D jD1 E Dj 7 E D 11 E
.1/ .1/ .1/ .1/ .2/ .3/ .3/
d6 ; d4 ; : : : ; d1 z6 ˚ d5 ˚ d1 ; : : : ; d6 ˚
36 D E D E D E
.2/ .2/ .2/ .2/ .2/ .4/ .4/
˚ d6 x3 ; d3 ; d2 ; d1 ˚ d1 ; : : : ; d6
*P +
D E D E 6 .1/ .1/
.1/ .1/ .1/ jD1 jaj  6a8 ;
37 a1 ; : : : ; a5 ˚ a7 .1/ .1/ .1/ ˚
D a7 C 2aE8 ; a9
.2/ .2/
a1 ; : : : ; a9 ˚ hd1 ; : : : ; d6 i
*P +
D E D E 6 .1/
.1/ .1/ .2/ jD1 jaj ;
38 a1 ; : : : ; a5 ˚ a1 hd1 ; : : : ; d6 i ˚ .1/ .1/ .1/ ˚
D a7 ; a8 ;E a9
.2/ .2/ .2/ .2/
a1 C 2a2 ; a3 ; : : : ; a9
D E DP E
.1/ .1/ 6 .1/ .1/ .1/ .1/
39 a1 ; : : : ; a5 ˚ hd6 i jD1 jaj ; a7 ; a8 ; a9 ˚ hd5 i
D E
.2/ .2/
˚ hx3 ; d3 ; d2 ; d1 i ˚ a1 ; : : : ; a9
* +
D E .1/
a1 C 2a2 ;
.1/ D E
.1/ .2/ .2/
40 hd5 ; : : : ; d1 i ˚ a1 hz6 i ˚ .1/ .1/ ˚ a1 ; : : : ; a9
D E D E DP a3 ; : : : ; aE9 D E
.1/ .1/ .1/ 6 .1/ .1/ .2/ .2/
41 a1 ; : : : ; a5 ˚ a7 jD1 jaj C 3a7 ˚ a1 ; : : : ; a7
D E D E
.1/ .1/ .2/ .2/
˚ d1 ; : : : ; d7 ˚ d1 ; : : : ; d7
* +
D E D E DP E .2/
a1 C 2a2 ;
.2/
.1/ .1/ .2/ 6 .1/ .1/
42 a1 ; : : : ; a5 ˚ a1 jD1 jaj ; a7 ˚ .2/ .2/
D E D a3 ; : : : ; a7E
.1/ .1/ .2/ .2/
˚ d1 ; : : : ; d5 ˚ d1 ; : : : ; d5
D E D E DP E D E
.1/ .1/ .1/ 6 .1/ .1/ .1/
43 a1 ; : : : ; a5 ˚ d5 ja ; a7 ˚ d4 ˚
D jD1 j E D E
.1/ .1/ .1/ .2/ .2/
x3 ; d2 ; d1 ˚ a1 ; : : : ; a7
D E D E DP E
.1/ .1/ .1/ 6 .1/ .1/ .1/
44 a1 ; : : : ; a5 ˚ a7 jD1 jaj ; a7 C 2a8 ˚
D E D E
.2/ .2/ .3/ .3/
a1 ; : : : ; a8 ˚ a1 ; : : : ; a8
(continued)
Classifications of Elliptic Fibrations of a Singular K3 Surface 43

Table 3 (continued)
No. Primitive Embedding Orthogonal
*P Complement
+ * +
D E D E 6 .1/ .2/ .2/
.1/ .1/ .2/ jD1 jaj ; a1 C 2a2 ;
45 a1 ; : : : ; a5 ˚ a1 .1/ .1/ ˚ .2/ .2/
aD7 ; a8 E a3 ; : : : ; a8
.3/ .3/
˚ a1 ; : : : ; a8
DP E
6
46 ha1 ; : : : ; a5 i ˚ ha7 i jD1 jaj  6a8 ; a7 C 2a8 ; a9 ; : : : a24
* P6 ja.1/  6a.1/ ; +
D E D E jD1 j 8 D E
.1/ .1/ .1/ .1/ .1/ .2/ .2/
47 a1 ; : : : ; a5 ˚ a7 a7 C 2a8 ; ˚ a1 ; : : : ; a12
.1/ .1/
*P a9 ; : :+: ; a12 * +
D E D E 6 .1/ .2/ .2/
.1/ .1/ .2/ jD1 jaj ; a1 C 2a2 ;
48 a1 ; : : : ; a5 ˚ a1 .1/ .1/ ˚ .2/ .2/
* a7 ; : : : ; a12 + a3 ; : : : ; a12
D E D E .2/
a1 C 2a2 ;
.2/ D E
.1/ .1/ .2/ .3/ .3/
49 a1 ; : : : ; a5 ˚ a1 .2/ .2/ .2/ ˚ a1 ; : : : ; a5
aD3 ; a4 ; a5 E
.4/ .4/
˚ a1 ; : : : ; a5 ˚ hd1 ; : : : ; d4 i
D E D E
.1/ .1/ .2/ .2/
50 a1 ; : : : ; a5 ˚ hd4 i hd3 i ˚ hx3 i ˚ hd1 i ˚ a1 ; : : : ; a5
D E D E
.3/ .3/ .4/ .4/
˚ a1 ; : : : ; a5 ˚ a1 ; : : : ; a5
* +
D E D E DP E .2/
a1 C 2a2 ;
.2/
.1/ .1/ .2/ 6 .1/
51 a1 ; : : : ; a5 ˚ a1 jD1 j ja ˚ .2/ .2/ ˚
D E Da3 ; : : : ; a6 E
.3/ .3/ .4/ .4/
a1 ; : : : ; a6 ˚ a1 ; : : : ; a6

lattice L such that Lroot ' D38 and we denote the generators of L=Lroot as follows:
.1/ .2/ .3/ .1/ .2/ .3/ .1/ .2/ .3/
v1 WD ı8 C ı8 C ı8 , v2 WD ı8 C ı8 C ı8 , v3 WD ı8 C ı8 C ı8 :
Fibration #22: We consider the embedding '1 W A5 ˚ A1 ,! L such that
.1/ .1/ .1/ .1/ .1/ .1/
'1 .A5 ˚ A1 / D hd7 ; d6 ; d5 ; d4 ; d3 i ˚ hd1 i. The generators of the lattice N are
described in Table 3 and one can directly check that N ' h6i ˚pA1 ˚ D8 ˚ D8 .
So, jd.N/j D 6  25 and the index of the inclusion N ,! W is 22 D 6  25 =12. This
implies that there is a copy of .Z=2Z/2  .Z=2Z/3 which is also contained in W
and so in particular is orthogonal to '1 .A5 ˚ A1 /.
We observe that v1 is orthogonal to the embedded copy of A5 ˚ A1 , v2 and v3 are
not. Moreover v2  v3 is orthogonal to the embedded copy of A5 ˚ A1 . Hence v1 and
v2  v3 generates W=N ' .Z=2Z/2 . We just observe that v2  v3 2 W is equivalent
f.2/ f.3/
mod Wroot to the vector w2 WD ı8 C ı8 2 W, so W=N ' .Z=2Z/2 ' hv1 ; w2 i. We
will reconsider this fibration in Section 4.3 comparing it with the fibration #22b.
Fibration #22.b/: We consider the other embedding of A5 ˚ A1 in Lroot , i.e. '2 W
.1/ .1/ .1/ .1/ .1/ .1/
A5 ˚ A1 ,! L such that '2 .A5 ˚ A1 / D hd7 ; d6 ; d5 ; d4 ; d3 i ˚ hx7 i:
The generators of the lattice N is described in Table 3 and one can directly check
that N ' h6i˚A1 ˚D8 ˚D8 . As above this implies that W=N ' .Z=2Z/2 which is
generated by elements in L=Lroot which are orthogonal to '2 .A5 ˚ A1 /. In particular,
v1  v2 and v2  v3 are orthogonal to '2 .A5 ˚ A1 / so v1  v2 2 W and v2  v3 2 W.
f.2/ f.3/ f.2/ f.3/
Moreover, v2  v3 D ı8 C ı8 mod Wroot . So, denoted by w2 WD ı8 C ı8 , we
44 M.J. Bertin et al.

have that W=N ' hv1  v2 ; w2 i. We will reconsider this fibration in Section 4.3
comparing it with the fibration #22.

4.2.6 Step 6

We recalled in Section 2.1 that each elliptic fibration is associated with a certain
decomposition of the Néron–Severi group as a direct sum of U and a lattice,
called W. In step 5 we computed all the admissible lattices W, so we classify
the elliptic fibrations on X. We denote all the elliptic fibrations according to their
associated embeddings; this gives the first five columns of the Table 1.

4.2.7 Step 7

Moreover, again in Section 2.1, we recalled that each reducible fiber of an elliptic
fibration is uniquely associated with a Dynkin diagram and that a Dynkin diagram is
associated with at most two reducible fibers of the fibration. This completes step 7.

4.2.8 Step 8

In order to compute the rank of the Mordell–Weil group it suffices to perform the
suggested computation, so r D 18  rank.Nroot /. This gives the sixth column of
Table 1.
For example, in cases 22 and 22(b), the lattice Nroot coincides and has rank 17,
thus r D 1 in both the cases.

4.2.9 Step 9

In order to compute the torsion part of the Mordell–Weil group one has to identify
the vectors v 2 W=N such that kv 2 Nroot for a certain nontrivial integer number
k 2 Z; this gives the last column of Table 1. We will demonstrate this procedure
in some examples below (on fibrations #22 and #22.b/), but first we remark that in
several cases it is possible to use an alternative method either in order to completely
determine MW.E/tors or at least to bound it. We already presented the theoretical
aspect of these techniques in Section 2.3.
Probably the easiest case is the one where r D 0. In this case MW.E/ D
MW.E/tors . Since r D 0, this implies that rank.Nroot / D 18 D rankN, so N D Nroot .
Hence W=N D W=Nroot , thus every element w 2 W=N is such that a multiple
is contained in Nroot , i.e. every element of W=N contributes to the torsion. Thus,
MW.E/ D W=N D W=Nroot . This immediately allows to compute the torsion for
the 7 extremal fibrations #2; 5; 10; 12; 28; 30; 50:
Classifications of Elliptic Fibrations of a Singular K3 Surface 45

Fibration #50 Nroot ' A˚3 ˚3 ˚3 ˚3

1 ˚ A5 (r D 0) The lattice N D Nroot is A1 ˚ A5 , then
3 3 2
jd.N/j D 2 6 and jW=Nj D 2  6. Moreover W=N  L=Lroot ' .Z=6Z/  Z=2Z.
This immediately implies that W=N D Z=6Z  Z=2Z.
Fibration #1 Nroot ' A1 ˚ E8˚2 (fibers of special type, Proposition 2.9) The
presence of the lattice E8 as summand of Nroot implies that the fibration has a fiber
of type II  (two in this specific case). Hence MW.E/tors is trivial.
Fibration #29 Nroot ' A3 ˚ D7 ˚ E6 (fibers of special type, Proposition 2.9) By
Proposition 2.9 if a fibration has a fiber of type IV  , then the Mordell–Weil group is
a subgroup of Z=3Z. On the other hand, a fiber of type D7 , i.e., I3 can only occur in
fibrations with 4 or 2-torsion or trivial torsion group. Therefore MW.E/tors is trivial.
Fibration #25 Nroot ' A7 ˚D9 (the height formula, Section 2.3.2) Suppose there is
a non-trivial torsion section P. Then, taking into account the possible contributions
of the reducible fibers to the height pairing, there is 0  i  7 such that one of the
following holds:

i.8  i/ i.8  i/
4D C 1 or 4 D C 1 C 5=4:
8 8
After a simple calculation, one sees that neither of the above can happen and
therefore the torsion group MW.E/tors is trivial.
Fibration #22 Nroot ' A1 ˚ D8 ˚ D8 We already computed the generators
of W=N in Section 4.2.5, W=N ' .Z=2Z/2 ' hv1 ; w2 i. A basis of Nroot is
.1/ .j/
hx7 i ˚ hdi iiD1;:::8;jD2;3 . So

.1/ .1/ .1/ .1/ .1/ .1/ .1/ .1/

2v1 D d1 C 2d2 C 3d3 C 4d4 C 5d5 C 6d6 C 2d7 C 3d8
0 0 11
X3 7
X
C @d7.i/ C d8.i/ C 2 @ .i/
d j AA
iD2 jD1

.1/ .1/ .1/ .1/ .1/ .1/ .1/ .1/

and 2v1 62 Nroot since d1 C 2d2 C 3d3 C 4d4 C 5d5 C 6d6 C 2d7 C 3d8 is
.1/
not a multiple of x7 . Vice-versa

.2/ .3/
2w2  D8 ˚ D8 2 Nroot :

Thus MW.E/ D Z  Z=2Z.

Fibration #22.b/ Nroot ' A1 ˚ D8 ˚ D8 Similarly, we consider the generators of
W=N ' .Z=2Z/2 ' hv1  v2 ; w2 i computed in Section 4.2.5. A basis of Nroot is
.1/ .j/
hd1 i ˚ hdi iiD1;:::8;jD2;3 . So

2w1 62 Nroot and 2w2 2 Nroot :

Thus also in this case MW.E/ D Z  Z=2Z.

46 M.J. Bertin et al.

4.3 Again on Fibrations #22 and #22.b/

As we can check in Table 1 and we proved in the previous sections, the fibrations
#22 and #22.b/ are associated with the same lattice N and to the same Mordell–Weil
group. However, we proved in Lemma 4.10 that they are associated to different
(up to Weyl group) embeddings in the Niemeier lattices, so they correspond to
fibrations which are not identified by the J2 -fibration and in particular they cannot
have the same frame. The following question is now natural: what is the difference
between these two fibrations? The answer is that the section of infinite order, which
generates the free part of the Mordell–Weil group of these two fibrations, has
different intersection properties, as we show now in two different ways and contexts.
Fibration #22: We use the notation of Section 4.2.5. Moreover we fix the following
.1/ .j/ .j/
notation: ‚11 WD x7 and ‚i WD di , i D 1; : : : 8, j D 2; 3 are, respectively, the non
trivial components of the fibers of type I2 , I4 , and I4 , respectively.
The class P WD 2F C O  v1 is the class of a section of infinite order of the
fibration, generating the free part of MW.E/ and the class Q WD 2F C O  w2 is the
class of the 2-torsion section of the fibration. The section P meets the components
‚11 , ‚21 , ‚31 and Q meets the components ‚10 , ‚27 , ‚37 . We observe that h.P/ D 3=2
and h.Q/ D 0 which agree with Schütt and Shioda (2010, Formula 22) and the fact
that Q is a torsion section, respectively. We also give an explicit equation of this
fibration and of its sections, see (10).
Fibration #22.b/: We use the notation of Section 4.2.5. Moreover we fix the
.1/ .j/ .j/
following notation: ‚11 WD d1 and ‚i WD di , i D 1; : : : 8, j D 2; 3
are respectively the non-trivial components of the fibers of type I2 , I4 , and I4 ,
respectively. The class Q WD 2F C O  w2 is the class of the 2-torsion section
of the fibration. Observe that Q meets the components ‚10 , ‚27 , ‚37 . The class

P D 2F C O C v1  v2  ‚21  ‚22  ‚23  ‚24  ‚25  ‚26  ‚27

is the class of a section of infinite order, which intersects the following components
of the reducible fibers: ‚11 , ‚27 , ‚30 . This agrees with the height formula. We also
give an explicit equation of this fibration and of its sections, see (11).
Remark 4.11. The generators of the free part of the Mordell–Weil group is clearly
defined up to the sum by a torsion section. The section P˚Q intersects the reducible
fibers in the following components ‚11 , ‚20 , ‚37 (this follows by the group law on the
fibers of type I2 (or III) and I4 ).
Remark 4.12. Comparing the sections of infinite order of the fibration 22 and the
one of the fibration 22(b), one immediately checks that their intersection properties
are not the same, so the frames of the elliptic fibration 22 and of elliptic fibration
22(b) are not the same and hence these two elliptic fibrations are in fact different
under the J2 -classification.
We observe that both the fibrations #22 and #22.b/ specialize the same fibration,
which is given in Comparin and Garbagnati (2014, Section 8.1, Table r D 19,
Classifications of Elliptic Fibrations of a Singular K3 Surface 47

case 11)). Indeed the torsion part of the Mordell–Weil group, which is already
present in the more general fibrations analyzed in Comparin and Garbagnati (2014),
are the same and the difference between the fibration 22 and the fibration 22(b)
is in the free part of the Mordell Weil group, so the difference between these two
fibrations involves exactly the classes that correspond to our specialization.
Here we also give an equation for each of the two different fibrations #22 and
#22(b). Both these equations are obtained from the equation of the elliptic fibration
v
#8 (9). So first we deduce an equation for #8: Let c WD .w1/ 2 . Substituting v by

c .w  1/2 in (7), we obtain the equation of an elliptic curve depending on c, which

corresponds to the fibration #8 and with the following Weierstrass equation

Ec W ˇ 2 D ˛ ˛ 2 C 6c2 ˛  c3 .c  4/ .4c  1/ : (9)

˛ yc2 .4c1/
Fibration #22 Putting n0 D c2 .4c1/
; ˇD 4n03
; cD x
4n03
; in (9) we obtain
 
En0 W y2 D x x2  n0 n02  6n. 0/ C 1 x C 16n04 (10)

with
 0 singular fibers of type 2I
4 .n D 0; 1/

C I2 C 2I1 : We notice the point P D
.n  1/ n ; 2n .n  1/ of height 32 , therefore P and Q D .0; 0/ generate
2 0 02 02

the Mordell–Weil group of En0 . To study the singular fiber at n0 D 1 we do the

transformation N 0 D 1n ; y D Nˇ061 ; x D N˛041 and P D .˛1 ; ˇ1 / with ˛1 D .N 0  1/2 N 0

and ˇ1 D 2N 02 N 02  1 : We deduce that the section P intersects the component
of singular fibers at 0 and 1 with the same subscript, so this fibration corresponds
to fibration #22.
2˛
Fibration #22(b) Putting n D c.4c1/ ; ˇ D yc.4c1/
4n ; c D x
2n in (9) we obtain

 
En W y2 D x x2 C 2n n2 C 3n C 4 C n4 (11)

with
 singular fibers of type 2I4 .n D 0; 1/ C I2 C 2I1 : We notice the point P D
4; 2.n C 2/2 of height 32 , therefore P and Q D .0; 0/ generate the Mordell–Weil
group of En . Since P does not meet the node of the Weierstrass model at n D 0,
the section P intersects the component ‚0 of the singular fiber for n D 0; so this
fibration corresponds to #22(b).
Remark 4.13. Let us denote by E9 and E21 the elliptic fibrations #9 and #21,
respectively. They satisfy Tr.E9 / ' Tr.E21 / and MW.E9 / ' MW.E21 /, but E9 is
not J2 -equivalent to E21 since, as above, the infinite order sections of these two
fibrations have different intersection properties with the singular fibers. Indeed these
two fibrations correspond to different fibrations (Comparin and Garbagnati 2014,
Case 10a) and case 10b), Section 8.1, Table r D 19) on the more general family of
K3 surfaces considered in Comparin and Garbagnati (2014).

Acknowledgements We thank the organizers and all those who supported our project for their
efficiency, their tenacity and expertise. The authors of the paper have enjoyed the hospitality of
CIRM at Luminy, which helped to initiate a very fruitful collaboration, gathering from all over the
48 M.J. Bertin et al.

world junior and senior women, bringing their skill, experience, and knowledge from geometry and
number theory. Our gratitude goes also to the referee for pertinent remarks and helpful comments.
A.G is supported by FIRB 2012 “Moduli Spaces and Their Applications” and by PRIN 2010–
2011 “Geometria delle varietà algebriche.” C.S is supported by FAPERJ (grant E26/112.422/2012).
U.W. thanks the NSF-AWM Travel Grant Program for supporting her visit to CIRM.

References

Atkin, A.O., Morain, F.: Finding suitable curves for the elliptic curve method of factorization.
Math. Comput. 60, 399–405 (1993)
Beauville, A.: Les familles stables de courbes elliptiques sur P1 admettant quatre fibres singulières.
C. R. Acad. Sci. Paris Sér. I Math. 294, 657–660 (1982)
Bertin, M.J.: Mesure de Mahler et série L d’une surface K3 singulière. Actes de la Conférence:
Fonctions L et Arithmétique. Publ. Math. Besan. Actes de la conférence Algèbre Théorie Nbr.,
Lab. Math. Besançon, pp. 5–28 (2010)
Bertin, M.J., Lecacheux, O.: Elliptic fibrations on the modular surface associated to 1 .8/.
In: Arithmetic and Geometry of K3 Surfaces and Calabi-Yau Threefolds. Fields Institute
Communications, vol. 67, pp. 153–199. Springer, New York (2013)
Braun, A.P., Kimura, Y., Watari, T.: On the classification of elliptic fibrations modulo isomor-
phism on K3 surfaces with large Picard number, Math. AG; High Energy Physics (2013)
[arXiv:1312.4421]
Cassels, J.W.S.: Lectures on Elliptic Curves. London Mathematical Society Student Texts, vol. 24.
Cambridge University Press, Cambridge (1991)
Comparin, P., Garbagnati, A.: Van Geemen-Sarti involutions and elliptic fibrations on K3 surfaces
double cover of P2 . J. Math. Soc. Jpn. 66, 479–522 (2014)
Couveignes, J.-M., Edixhoven, S.: Computational Aspects of Modular Forms and Galois Repre-
sentations, Annals of Math Studies 176. Princeton University Press, Princeton
Cox, D., Katz, S.: Mirror Symmetry and Algebraic Geometry. American Mathematical Society,
Providence (1999)
Elkies, N.D.: http://www.math.harvard.edu/~elkies/K3_20SI.html#-7[1 (2010)
Elkies, N.D.: Three Lectures on Elliptic Surfaces and Curves of High Rank. Lecture notes,
Oberwolfach (2007)
Elkies, N., Schütt, M.: Genus 1 fibrations on the supersingular K3 surface in characteristic 2 with
Artin invariant 1. Asian J. Math. (2014) [arXiv:1207.1239]
Garbagnati, A., Sarti, A.: Elliptic fibrations and symplectic automorphisms on K3 surfaces.
Commun. Algebra 37, 3601–3631 (2009)
Karp, D., Lewis, J., Moore, D., Skjorshammer, D., Whitcher, U.: On a family of K3 surfaces with
S4 symmetry. In: Arithmetic and Geometry of K3 Surfaces and Calabi-Yau Threefolds. Fields
Institute Communications. Springer, New York (2013)
Kloosterman, R.: Classification of all Jacobian elliptic fibrations on certain K3 surfaces. J. Math.
Soc. Jpn. 58, 665–680 (2006)
Kondo, S.: Algebraic K3 surfaces with finite automorphism group. Nagoya Math. J. 116, 1–15
(1989)
Kondo, S.: Automorphisms of algebraic K3 surfaces which act trivially on Picard groups. J. Math.
Soc. Jpn. 44, 75–98 (1992)
Kubert, D.S.: Universal bounds on the torsion of elliptic curves. Proc. Lond. Math. Soc. 33(3),
193–237 (1976)
Kumar, A.: Elliptic fibrations on a generic Jacobian Kummer surface. J. Algebraic Geom.
[arXiv:1105.1715] 23, 599–667 (2014)
Kreuzer, M., Skarke, H.: Classification of reflexive polyhedra in three dimensions. Adv. Theor.
Math. Phys. 2, 853–871 (1998)
Classifications of Elliptic Fibrations of a Singular K3 Surface 49

Kreuzer, M., Skarke, H.: Complete classification of reflexive polyhedra in four dimensions. Adv.
Theor. Math. Phys. 4, 1209–1230 (2000)
Martinet, J.: Perfect Lattices in Euclidean Spaces, vol. 327. Springer, Berlin/Heidelberg (2002)
Montgomery, P.L.: Speeding the Pollard and Elliptic Curve Methods of Factorization. Math.
Comput. 48. (1987) 243–264.
Nikulin, V.V.: Finite groups of automorphisms of Kählerian K3 surfaces. (Russian) Trudy Moskov.
Mat. Obshch. 38, 75–137 (1979). English translation: Trans. Moscow Math. Soc. 38, 71–135
(1980)
Nikulin, V.V.: Integral symmetric bilinear forms and some of their applications. Math. USSR Izv.
14, 103–167 (1980)
Nishiyama, K.: The Jacobian fibrations on some K3 surfaces and their Mordell–Weil groups. Jpn.
J. Math. (N.S.) 22, 293–347 (1996)
Oguiso, K.: On Jacobian fibrations on the Kummer surfaces of the product of nonisogenous elliptic
curves. J. Math. Soc. Jpn. 41, 651–680 (1989)
Schütt, M.: K3 surface with Picard rank 20 over Q. Algebra Number Theory 4, 335–356 (2010)
Schütt, M., Shioda, T.: Elliptic surfaces. In: Algebraic Geometry in East Asia–Seoul 2008.
Advanced Studies in Pure Mathematics, vol. 60, pp. 51–160. The Mathematical Society of
Japan, Tokyo (2010)
Shimada, I., Zhang, D.-Q.: Classification of extremal elliptic K3 surfaces and fundamental groups
of open K3 surfaces. Nagoya Math. J. 161, 23–54 (2001)
Shioda, T.: On the Mordell–Weil lattices. Commun. Math. Univ. St. Pauli 39, 211–240 (1990)
Shioda, T.: On elliptic modular surfaces J. Math. Soc. Jpn. 24(1), 20–59 (1972)
Stein, W.A., et al.: Sage Mathematics Software (Version 6.1). The Sage Development Team. http://
www.sagemath.org (2014)
Sterk, H.: Finiteness results for algebraic K3 surfaces. Math. Z. 180, 507–513 (1985)
Verrill, H.: Root lattices and pencils of varieties. J. Math. Kyoto Univ. 36, 423–446 (1996)
Shalika Germs for sln and sp2n Are Motivic

Sharon M. Frechette, Julia Gordon, and Lance Robson

Abstract We prove that Shalika germs on the Lie algebras sln and sp2n belong to
the class of so-called motivic functions defined by means of a first-order language
of logic. It is a well-known theorem of Harish-Chandra that for a Lie algebra g.F/
over a local field F of characteristic zero, the Shalika germs, normalized by the
square root of the absolute value of the discriminant, are bounded on the set of
regular semisimple elements grss , however, it is not easy to see how this bound
depends on the field F. As a consequence of the fact that Shalika germs are motivic
functions for sln and sp2n , we prove that for these Lie algebras, this bound must
be of the form qa , where q is the cardinality of the residue field of F, and a is
a constant. Our proof that Shalika germs are motivic in these cases relies on the
interplay of DeBacker’s parametrization of nilpotent orbits with the parametrization
using partitions, and the explicit matching between these parametrizations due
to Nevins (Algebra Representation Theory 14, 161–190, 2011). We include two
detailed examples of the matching of these parametrizations.

1 Introduction

In this paper we prove that Shalika germs for the Lie algebras of type sln and
sp2n belong to the class of so-called motivic functions, and explore some of the
consequences of this fact.
Shalika germs first appeared in the papers of Shalika (1972) and Harish-Chandra
(1973). The survey of their role in harmonic analysis on p-adic groups is beyond the
scope of this paper; we refer the reader to the beautiful article by Kottwitz (2005),
and to Harish-Chandra 1999 for the detailed definitions and main results regarding
them. We note that Shalika germs, by definition, are functions on the set of regular
semisimple elements in a Lie algebra, yet, except for those defined on a few Lie

S.M. Frechette
Dept. of Mathematics & Computer Science, College of the Holy Cross, 1 College Street,
Worcester, MA 01610
J. Gordon () • L. Robson
Mathematics Department, The University of British Columbia, Vancouver, BC, Canada
e-mail: gor@math.ubc.ca

© Springer International Publishing Switzerland 2015 51

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_3
52 S.M. Frechette et al.

algebras of small rank, their exact values elude computation. Here we use a general
theorem about uniform bounds for motivic functions proved in Shin and Templier
2015, Appendix B to estimate the absolute values of the Shalika germs in a uniform
way over all local fields of a fixed (sufficiently large) residue characteristic.
First, let us recall the definitions. Let F be a local field, G be a connected
reductive algebraic group over F, and g its Lie algebra. In our results, G D SLn
or Sp2n , although several of the background results hold in greater generality. Let
X 2 g.F/, with adjoint orbit OX D fAd.g/X j g 2 G.F/g and stabilizer CG .X/.
(Since here we are dealing with the classical Lie algebras, the Adjoint action is
just conjugation: Ad.g/X D gXg1 .) The space OX with the p-adic topology is
homeomorphic to G.F/=CG .X/, which carries a G-invariant quotient measure. For
the fields F of characteristic zero, it was proved by Ranga Rao (1972) that when
transported to the orbit of X, this measure is a Radon measure on g.F/, i.e., it
is finite on compact subsets of g.F/. (Strictly speaking, it is the group version of
this statement that is proved in Ranga Rao 1972, but in characteristic zero this is
equivalent to the Lie algebra version.) Denote this quotient measure on G.F/=CG .X/
by d g. The orbital integral at X is the distribution X on Cc1 .g.F// defined by
Z
X .f / D f .Ad.g/X/d g: (1)
G.F/=CG .X/

For the fields of sufficiently large positive characteristic (with an explicit bound on
the characteristic), convergence of orbital integrals was proved by McNinch (2004).
There are finitely many nilpotent orbits in g.F/, provided the field F has
characteristic zero or sufficiently large positive characteristic (depending on the root
system of g). The Shalika germ expansion expresses the regular semisimple orbital
integrals as linear combinations of nilpotent ones, in a neighbourhood of the origin.
More precisely, let Nil.F/ denote the finite set of nilpotent orbits in g.F/, let g.F/rss
denote the set of regular semisimple elements in g.F/, and for each O 2 Nil.F/
let O be the orbital integral over O (it is a linear functional on Cc1 .g.F//). For
every f 2 Cc1 .g.F// there exists a neighbourhood Uf of zero in g.F/, and functions
O .X/ defined on g.F/rss \ Uf , such that for all X 2 g.F/rss \ Uf , we have the
expansion
X
X .f / D O .X/O .f /: (2)
O2Nil.F/

The functions O are called provisional Shalika germs, using the terminology of
Kottwitz (2005, §6, §17). These provisional Shalika germs are well defined as germs
of functions at the origin; moreover, they possess a natural homogeneity property,
and using this they can be extended canonically to the entire set g.F/rss (see Kottwitz
2005, § 17 for details).
The goal of this paper is to prove that provisional Shalika germs belong to
the class of the so-called motivic functions. This was proved for g D sp2n by
L. Robson in his M.Sc. essay (Robson 2012); we include this case here since it
Shalika Germs for sln and sp2n Are Motivic 53

was not published elsewhere. We also study the case g D sln which is in some ways
simpler, but has a technical issue that does not arise in the sp2n case (namely, the
dependence of the set of nilpotent orbits on the field F); this was the content of our
WIN project. We present both cases in detail here in preparation for a general proof
for all Lie algebras, which will appear elsewhere.
The class of motivic functions was defined by R. Cluckers and F. Loeser
(Cluckers and Loeser 2008). Concretely, motivic functions are complex-valued
functions on p-adic manifolds defined uniformly in p by means of a first-order
language of logic, called Denef-Pas language, which we will define below. We
include a brief, simplified version of the definition of motivic functions. For the
details, as well as a survey of the applications of this class of functions in harmonic
analysis on p-adic groups, we refer the reader to the survey (Cluckers et al. 2011a)
and the original papers (Cluckers and Loeser 2008; Cluckers et al. 2011b). The aim
of this paper is to add Shalika germs to the list of functions arising in harmonic
analysis that can be studied via motivic integration techniques, in the case of g D sln
or sp2n .
Cluckers, Hales, and Loeser (2011b) prove that regular semisimple orbital
integrals are motivic, and in Cluckers et al. (2014a) the same statement is proved
for all, and in particular, nilpotent orbital integrals. Thus, once we have shown that
the functions O are motivic, we see that both sides of (2) are motivic functions.
As an immediate consequence of the Transfer Principle proved by Cluckers and
Loeser (2008), this shows the Shalika germ expansion holds for fields of sufficiently
large positive characteristic. (This was previously proved by DeBacker (2002a); our
results give an alternative proof.) More importantly, the uniform boundedness result
from Shin and Templier (2015, Appendix B) then implies the uniform bound on
Shalika germs normalized by the square root of the discriminant (see Theorem 17
below).
Our main results are stated and proved in Section 6. The rest of the paper provides
a review of all the prerequisites, thus experts may want to turn immediately to the
last section. The proof that provisional Shalika germs are motivic functions has
two main ingredients: first we must establish a way to describe nilpotent orbits in
the motivic context, and second, we find definable test functions which allow us to
isolate individual Shalika germs in the linear combination. The first step requires a
parametrization of nilpotent orbits that is as field-independent as possible, and this
is where partitions are advantageous. For the second step, it is convenient to use
DeBacker’s parametrization of orbits. The proof of the main theorem essentially
works in much greater generality than stated here, except we do not quite have the
structure in Denef-Pas language that would capture the set of nilpotent orbits in
general. Here, for the special cases of sln and sp2n , we use the matching between
the two parametrizations of nilpotent orbits that was established by Nevins (2011).
Since all three of the authors found this material challenging to absorb, in Section 5
we include detailed examples for sl3 and sp4 ; we hope they will be useful for future
students.
54 S.M. Frechette et al.

2 Motivic Functions

This section is included in order for the paper to be self-contained. However, this
overview of the definitions has appeared in various forms in several papers on the
topic; the present version is quoted nearly verbatim from Cluckers et al. (2011a),
except for Section 2.3, which is new and specifically adapted for the purposes of
this paper.
Informally, motivic functions are built from definable functions in the Denef-
Pas language. Thus they are given independently of the field and can be interpreted
in any non-Archimedean local field. We first recall the definition of the Denef-Pas
language.

2.1 Denef-Pas Language

Denef-Pas language is a first order language of logic designed for working with
valued fields. Formulas in this language will allow us to uniformly handle sets and
functions for all local fields. We start by defining two sublanguages of the language
of Denef-Pas: the language of rings and Presburger language.

2.1.1 The Language of Rings

Apart from the symbols for variables x1 ; : : : ; xn ; : : : and the usual logical symbols
equality ‘D’, parentheses ‘.’, ‘/’, the quantifiers ‘9’, ‘8’, and the logical operations
conjunction ‘^’, negation ‘:’, disjunction ‘_’, the language of rings consists of the
following symbols:
• constants ‘0’, ‘1’;
• binary functions ‘’, ‘C’.
A (first-order) formula in the language of rings is any syntactically correct
formula built out of these symbols. (One usually omits the words ‘first order’.) If
a formula in the language of rings has n free variables, then it defines a subset of
Rn for any ring R. For example, the formula “9x2 .x2  x1 D 1/” defines the set of
units R in any ring R. Note that by convention, quantifiers always run over the ring
in question. Note also that quantifier-free formulas in the language of rings define
constructible sets, as they appear in classical algebraic geometry.

2.1.2 Presburger Language

A formula in Presburger language is built out of variables running over Z, the logical
symbols (as above) and symbols ‘C’, ‘’, ‘0’, ‘1’, and for each d D 2; 3; 4; : : :, a
symbol ‘ d ’ to denote the binary relation x y .mod d/. Note the absence of the
symbol for multiplication.
Shalika Germs for sln and sp2n Are Motivic 55

2.1.3 Denef-Pas Language

The Denef-Pas language is a three-sorted language in the sense that its formulas
utilize three different “sorts” of elements: those of the valued field, of the residue
field, and of the value group (which will always be Z in our setting). Each variable
in such a formula runs over only the elements of one of the sorts, so there are
three disjoint sets of symbols for the variables of the different sorts. To create a
syntactically correct formula, one must pay attention to the sorts when composing
functions and inserting them into relations.
In addition to the variables and the logical symbols, the formulas use the
following symbols:
• In the valued field sort: the language of rings.
• In the residue field sort: the language of rings.
• In the Z-sort: the Presburger language.
• the symbol ord./ for the valuation map from the nonzero elements of the valued
field sort to the Z-sort, and the symbol ac./ for the so-called angular component,
which is a multiplicative function from the valued field sort to the residue field
sort (more about this function below).
A formula in this language can be interpreted in any discretely valued field F
which comes with a uniformizing element $, by letting the variables range over F,
over its residue field kF , and over Z, respectively, depending on the sort to which
they belong; ord is the valuation map (defined on F  and such that ord.$/ D 1),
and ac is defined as follows: if x is a unit (that is, ord.x/ D 0), then ac.x/ is the
residue of x modulo $ (thus, an element of the residue field); for all other nonzero
x, one puts ac.x/ WD $ ord.x/ x mod .$/. Thus, for x ¤ 0, ac.x/ is the residue class
of the first non-zero coefficient of the $-adic expansion of x. Finally, we define
ac.0/ D 0.
Thus, a formula ' in this language with n free valued-field variables, m free
residue-field variables, and r free Z-variables gives naturally, for each discretely
valued field F, a subset '.F/ of F n  kFm  Zr : namely, '.F/ is the set of all the
tuples for which the interpretation of ' in F is “true”.
We will denote this language by LDP .

2.2 Definable Sets and Motivic Functions

The LDP -formulas introduced in the previous section allow us to obtain a field-
independent notion of subsets of F n  kFm  Zr for all local fields F of sufficiently
large residue characteristic. The reason behind the restriction on characteristic is
explained below in Remark 3.
56 S.M. Frechette et al.

Definition 1. A collection X D .XF /F of subsets XF  F n  kFm  Zr is called a

definable set if there is an LDP -formula ' and an integer M such that XF D '.F/
for each F with residue characteristic at least M (cf. Remark 3), where '.F/ is as
described at the end of Section 2.1.3.
By Definition 1, a definable set is actually a collection of sets indexed by non-
Archimedean local fields F; such practice is not uncommon in model theory and has
its analogues in classical algebraic geometry. A particularly simple definable set is
.F n  kFm  Zr /F , for which we introduce the simplified notation VFn  RFm  Zr ,
where VF stands for valued field and RF for residue field. We apply the typical set-
theoretical notation to definable sets X; Y, e.g., X  Y (if XF  YF for each F),
X  Y, and so on.
Definition 2. For definable sets X and Y, a collection f D .fF /F of functions fF W
XF ! YF is called a definable function and denoted by f W X ! Y if the collection
of graphs of the fF is a definable set.
Remark 3. There is a subtle issue here, due to the fact that the same definable set
can be defined by different formulas. Technically, it would be more elegant to think
of a definable set as an equivalence class of what we have called definable sets in
Definition 1, where we call two such definable sets equivalent if they are the same
for all F with sufficiently large residue characteristic. To ease notation, we will not
emphasize this point, but because of this all results presented in this paper will only
be valid for fields with sufficiently large residue characteristic. In particular, we
assume hereafter that char.F/ ¤ 2.
We now come to motivic functions, for which definable functions are the
building blocks. We note that while definable functions, by definition, must be
VFn  RFm  Zr -valued for some m; n; r, the motivic functions are built from
definable sets and functions, and can be thought of as complex-valued functions
(although here they will naturally be Q-valued). This does not require thinking of
rational or complex numbers in the context of logic; these are just usual complex-
valued functions that happen to be built from definable ingredients as prescribed by
the following definition.
Definition 4. Let X D .XF /F be a definable set. A collection f D .fF /F of functions
fF W XF ! C is called a motivic function on X if and only if there exist integers N,
N 0 , and N 00 , such that, for all non-Archimedean local fields F,

X
N Y
N0  Y
N 00 
˛ .x/ 1
fF .x/ D qFiF .#.YiF /x / aijF .x/ ; for x 2 XF ; (3)
1  qaFi`
iD1 jD1 `D1

for some
• nonzero integers ai` ,
• definable functions ˛i W X ! Z and ˇij W X ! Z,
• definable sets Yi  X  RFri ,
Shalika Germs for sln and sp2n Are Motivic 57

where, for x 2 XF , .YiF /x is the finite set fy 2 kFri j .x; y/ 2 YiF g, and qF is the
cardinality of the residue field kF .
We call a motivic function on a one-point set a motivic constant.
In Theorem 17, we will need to allow the square root of the cardinality of
the residue field as a possible value of a motivic function. Hence, we will use
the slightly generalized notion of a motivic function introduced in Cluckers et al.
(2014a, §B.3.1). Namely, given an integer r > 0 and a definable Z-valued function
f =r
f , expressions of the form qF h, where h is a motivic function as above, will also be
called motivic functions.

2.3 Adding Constants to the Language

We will need to extend Denef-Pas language by adding finitely many constant

symbols in the valued field sort, whose role will be to encode units whose angular
components form a set of representatives of kF =.kF /m , where m is a fixed integer.
Such extensions were first used by T.C. Hales, and an extension very similar to the
one we define here appears first in J. Diwadkar’s thesis (Diwadkar 2006, § 2.2.3).
Specifically, let m be a fixed integer. We add m constant symbols d1 ; : : : ; dm to
the valued field sort of Denef-Pas language, to obtain the language LDPm .
Now we need to define their interpretation, given a local field F with a
uniformizer $ and residue field k.
If the set k =.k /m has m elements, then we want d1 ; : : : ; dm to be interpreted
as units such that their angular components form a set of representatives of distinct
.k /m -cosets. Specifically, we can write a formula

9y1 ; : : : ; ym 2 F  ; ord.yi / D 0; Àz W yi D yj zm if i ¤ j:

This formula is true for F under our assumption. Then we can set the values of
d1 ; : : : ; dm in F to be any collection fy1 ; : : : ; ym g satisfying this formula.
If the cardinality of kF =.kF /m is equal to ` < m, we write a similar formula
stating that fy1 ; : : : ; y` g are distinct representatives of .kF /m -cosets, with the
convention that the trivial coset is always represented by the constant symbol 1.
More precisely, for every divisor ` of m, let `;m be the following formula, with the
quantifiers ranging over the residue field sort:

`;m .y1 ; : : : ; y` / WD ‘Àz W yi D yj zm for i ¤ j ^ 8x9z; x D yi zm for some 1  i  `:’

(4)
(This formula is written slightly informally; in reality, it contains a conjunction
of `.`  1/=2 formulas, and a disjunction of ` formulas.) This formula states that
y1 ; : : : ; y` are distinct representatives of kF =.kF /m in kF .
58 S.M. Frechette et al.

For a given finite field k and fixed m, exactly one of the statements

`;m WD ‘9y1 ; : : : ; y` ; `;m .y1 ; : : : ; y` /’

holds, as ` runs over all divisors of m. If `;m holds in kF , we interpret the constant
symbols d1 ; : : : ; d` as units of the valued field such that `;m .ac.d1 /; : : : ; ac.d` //
holds. Set the rest of the di equal to 1.
None of the constructions and theorems of motivic integration change if we add
finitely many constant symbols. Hereafter, we fix an integer n (coming from a fixed
Lie algebra sln or sp2n ), and say that a set or function is definable if it is definable in
the language LDPm for some m  nP.n/, where P.n/ is the number of partitions of
n. We shall see later that we may need to consider the union of languages LDPm as
m varies over a set of integers associated with partitions of n; however, it does not
matter how many constants we add, as long as it is a finite number that is fixed in
advance.
In the same way that a non-Archimedean local field F with a choice of the
uniformizer is a structure for the language LDP , we note that a structure for LDPm is
a non-Archimedean local field F with a choice of the uniformizer of the valuation,
and a choice of a collection of units whose angular components form a set of
representatives of kF =.kF /m .
The theory of motivic integration works as usual in this setting; this is a very
special case of the set-up of Cluckers and Loeser (2015), where the new constants
can be thought of as part of the theory T (in the setting of that paper).
With this terminology, we can now state this paper’s goal precisely: to show that
Shalika germs are motivic functions, up to dividing by a motivic constant, in the
sense that they are motivic functions where we use the language LDPm with some
finite m. We note that not all motivic constants are invertible in the ring of motivic
functions, which is why we require the “up to motivic constant” provision.

3 Classification of Nilpotent Orbits of sln and sp2n

Via Partitions

As discussed in the Introduction, we study two parameterizations of nilpotent orbits

in sln and in sp2n , with a view toward defining these orbits by formulas in Denef-
Pas language. In this section we recall a well-known parametrization involving
partitions, and in Section 4 we recall a parametrization due to DeBacker (2002b),
involving the Bruhat-Tits building for g. In fact, a proof of definability of the
nilpotent orbits for sln using DeBacker’s parametrization is carried out explicitly in
Diwadkar’s thesis (Diwadkar 2006). Here we recast it in a slightly simpler form,
taking advantage of the explicit matching between the two parametrizations, as
proved by Nevins (2011), and also of recent developments in the theory of motivic
integration that allow us to slightly simplify Diwadkar’s terminology.
Shalika Germs for sln and sp2n Are Motivic 59

3.1 Notation

Hereafter, F will stand for a non-Archimedean local field with char.F/ ¤ 2, and
F for a separable closure of F. The ring of integers of F will be denoted by O (or
OF if there is a possibility of confusion), the maximal ideal by P, and the residue
field by kF . We will always assume that F comes with a choice of the uniformizer
of the valuation $, and when talking about the language LDPm , with a choice of
representatives for O =.O /m as discussed above in Section 2.3.

3.2 Parametrization of Nilpotent Orbits in sl.n/

Using Partitions

For a positive integer n, a partition  D .1 ; 2 ; : : : ; t / of n is a weakly decreasing

sequence of positive integers ( i.e. 1  2      t ) whose sum is n. We say
the i are the parts of the partition , and the length of  is t. For each 1  j  n,
the multiplicity mj ./ is the number of parts of  satisfying i D j. We denote the
greatest common divisor of the parts i by gcd./.
It is well known that when the characteristic of F is greater than n, the set of
nilpotent orbits of sln .F/ is in one-to-one correspondence with the set of partitions
of n (see Collingwood and McGovern 1993 or Waldspurger 2001, for instance).
A nilpotent orbit corresponds to the partition whose parts are determined by the
blocks in its Jordan normal form. Specifically, let  D .1 ; 2 ; : : : ; t / be a partition
of n, and let Ji denote the i  i -matrix whose .j; j C 1/ entries are equal to 1 for
1  j  i , with all remaining entries equal to 0. Let J denote the n  n-matrix in
Jordan normal form whose Jordan blocks are the Ji , and let O denote the nilpotent
orbit in sln .F/ with representative J .
The explicit correspondence between partitions and F-rational nilpotent orbits is
described in the following proposition. The number of F-rational nilpotent orbits
depends both on the partition  and on the characteristic of F, in a controlled way.
Proposition 5 (Nevins (2011), Prop. 4). Let  be a partition of n, and
m D gcd./. For any d 2 F  define the n  n-matrix D.d/ D diag.1; 1; : : : ; 1; d/.
(1) For each d 2 F  , the matrix Xd D J D.d/ represents an F-rational orbit in
O .F/, and conversely every orbit has a representative of this form.
(2) The SLn .F/-orbits represented by J D.d/ and J0 D.d0 / coincide if and only if
 D 0 and d d0 in F  =.F  /m .
Example 6. In the case of sl3 , we have three partitions:  D .3/, .2; 1/, and
.1; 1; 1/. The corresponding nilpotent orbits O in sl3 .F/ have representatives
0 1 0 1 0 1
010 010 000
X.3/ D @0 0 1A, X.2;1/ D @0 0 0A, and X.1;1;1/ D @0 0 0A respectively.
000 000 000
60 S.M. Frechette et al.

The nilpotent orbits O.2;1/ and O.1;1;1/ do not split further into distinct F-rational
orbits, since m D gcd./ D 1 for these partitions. Since m D 3 for the first partition,
the nilpotent orbit O.3/ splits into jF  =.F  /3 j distinct F-rational orbits, represented
by the matrices
0 1
010
Xd D J.3/ D.d/ D @0 0 dA ;
000

one for each distinct equivalence class of d in F  =.F  /3 . By our assumptions, F has
residue characteristic ¤ 2 and its residue field kF has q D pk elements, where p is
q1
prime. By standard results in group theory, the number of cubes in kF is gcd.3;q1/ ,
 3
and so the cardinality of kF =.kF / is gcd.3; q  1/. Thus the number of distinct


F-rational orbits in this case is

(
ˇ  ˇ ˇ ˇ
ˇF =.F  /3 ˇ D 3  ˇk =.k /3 ˇ D 9 if 3 j .q  1/,
F F
3 otherwise:

3.3 Parametrization of Nilpotent Orbits in sp2n

Using Partitions

In the case of sp2n , classes of quadratic forms over F take the place of the cosets
F  =.F  /m that we have seen in the parametrization of nilpotent orbits in the sln
case. Thus, we begin by recalling the classification of quadratic forms.

3.3.1 Quadratic Forms

Let V be a finite-dimensional vector space over F, and Q a non-degenerate quadratic

form defined on V. Recall that the quadratic space .V; Q/ over F is anisotropic if
there is no nonzero x 2 V such that Q.x/ D 0, and is isotropic otherwise.
2
 form q0 W F ! F that is represented in the standard basis
Consider the quadratic
01
by the matrix q0 D . The quadratic space .F 2 ; q0 / is the hyperbolic plane, a
10
key example in the theory of quadratic forms. Since char.F/ > 2, if .V; Q/ is a
non-degenerate quadratic space over F, then by the Witt decomposition (see Lam
2005 for instance), the quadratic form Q can be decomposed into an orthogonal
direct sum

0 ˚ Qaniso ;
Q D qm (5)
Shalika Germs for sln and sp2n Are Motivic 61

for some m  12 dim.Q/, where .Vaniso ; Qaniso / is anisotropic and uniquely deter-
mined up to isometry. The integer m is called the Witt index of .V; Q/ and the
quadratic form Qaniso is called the anisotropic part of Q. Moreover, quadratic forms
of a given dimension may be classified by their discriminant and Hasse invariant.
Since char.F/ ¤ 2, we have F  =.F  /2 ' Z=2Z  Z=2Z; thus, there are at most
8 nondegenerate quadratic forms over F of a given dimension. On the other hand,
the maximum possible dimension of an anisotropic form over F is four. Thus, by
the Witt decomposition, to list all equivalence classes of quadratic forms, it suffices
to list the classes of anisotropic forms. Representatives for these classes are given in
the following lemma.
Lemma 7 (Nevins (2011, Lemma 3)). Let F be as above. If 1 2 .F  /2 , let ˛ D "
be a fixed nonsquare unit in F. If 1 … .F  /2 , let ˛ D 1 and " D 1. Given a
quadratic form Q, its anisotropic part Qaniso is either the zero subspace, or isometric
to one of the 15 anisotropic forms in Table 1.
Given a class of quadratic forms, a matrix representative of the form Q D qm 0 ˚
Qaniso (or Q D qm 0 ), where Q aniso is one of the diagonal matrices given in the above
table, will be called a minimal matrix representative of the class.

3.3.2 Partition Parametrization of the Nilpotent Orbits in sp2n

 0 I
Embed Sp2n into GL2n as Sp2n D fg 2 GL2n W gt Jg D Jg, where J D I 0 . Note
that this is a different embedding than the one used by Waldspurger (2001), however,
the parametrization below follows his methods. Let V denote the vector space of the
natural representation of sp2n , with symplectic form defined by hx; yi D xt Jy.

Table 1 Explicit representatives of the 15 nonzero equivalence classes of

anisotropic quadratic forms over a local F with residue characteristic not 2
Dimension disc.Q/ Hasse.Q/ Representative
1 1 1 1
1 " 1 "
1 $ 1 $
1 "$ 1 "$
2 ˛ 1 diag.1; ˛/
2 ˛ 1 diag.$; ˛$/
2 tt0 $ .t; t0 $/F diag.t; t0 $/ t; t0 2 f1; "g
3 t 1 diag.˛t; $; ˛$/ t 2 f1; "g
3 ˛t$ .˛; $/F diag.1; ˛; t$/ t 2 f1; "g
4 1 1 diag.1; "; $; "$/
62 S.M. Frechette et al.

The nilpotent Adjoint orbits in sp2n .F/ are parametrized by partitions  of 2n

in which the odd parts have even multiplicity (Collingwood and McGovern 1993,
Corollary 4.1.8). For such a partition , let O denote the geometric nilpotent orbit
corresponding to .
The F-points, O .F/, of this orbit may fail to be a single Sp2n .F/-orbit, so
the set of partitions  is no longer sufficient to parametrize the nilpotent orbits
over F. Instead, there is a set defined in terms of classes of quadratic forms
corresponding to the partition  that parametrizes the Sp2n .F/-orbits in O .F/. Let
Q D .Q2 ; : : : ; Q2n / be an n-tuple of isometry classes of quadratic forms over F.
We say that Q corresponds to the partition  of 2n (whose odd parts have even
multiplicities) if dim.Qi / D mi ./ for each i D 2; : : : ; 2n.
Theorem 8 (Nevins (2011, Proposition 5), Due to Waldspurger (2001)). Let 
be a partition of 2n, and suppose the odd parts of  have even multiplicity. Then
O .F/ is a union of Sp2n .F/-orbits parametrized by the n-tuples

Q D .Q2 ; : : : ; Q2n /

corresponding to  (as defined above), where Qi is an isometry class of a

nondegenerate quadratic form over F.
Following Nevins (2011), for each pair .; Q/ we give an explicitly defined X 2
sp2n .F/ in the corresponding nilpotent orbit. We first give a decomposition of the
vector space V corresponding to the partition , and then define X by its action on
each component.
Let fp1 ; : : : ; pn ; q1 ; : : : ; qn g denote a symplectic basis for V; that is, a basis
such that hpi ; qj i D ıij , hqi ; pj i D ıij , and hpi ; pj i D hqi ; qj i D 0. For each
X1
i 2 f1; : : : ; 2ng, let si D j mj . Then the elements si are integers such that
j<i
2
0 D s1  s2      s2n  n. For each j with mj ¤ 0, let V.j/ be the subspace
given by

V.j/ D spanfpsjC1 ; : : : ; psj C 1 jmj ; qsj C1 ; : : : ; qsj C 1 jmj g: (6)

2 2

M
Then V D V.j/, so we may define X by its action on each subspace V.j/.
jWmj ¤0
If j is odd, let  D .j; : : : ; j/, a partition of 12 j mj , and define the restriction of X
to V.j/ with respect to the basis given in (6) by
!
J 0
XjV.j/ D : (7)
0 Jt
Shalika Germs for sln and sp2n Are Motivic 63

If j D 2N is even, define XjV.j/ with respect to the basis given in (6) by

m
!
JNmj j Z ˚ .1/N Qj
XjV.j/ D m ; (8)
0 .JNmj j /t

where Z is the mj .N  1/  mj .N  1/ zero matrix and Qj is the minimal matrix

representative of Qj . Then we have the following correspondence:
Theorem 9 (Nevins (2011), Adapted from Proposition 6). Let  be as above. The
matrix X 2 sp2n .F/ defined by (7) and (8) is a representative of the Sp2n .F/-orbit
in O .F/ corresponding to the n-tuple Q.

4 Parametrization of Nilpotent Orbits Via the Building

4.1 Preliminaries Regarding the Building

Following the notation and terminology of Nevins (2011), we briefly recall the
necessary facts about the standard affine apartment of the Bruhat-Tits building
B.G/ D B.G; F/ for G a connected reductive algebraic group over F. However,
since this is the only case we need in this paper, we assume that G is split over F,
which simplifies these definitions substantially.
Let T be a split maximal torus of G. Let X  .T/ be the group of F-rational
characters of T and let X .T/ be the group of F-rational cocharacters. Let h ; i W
X  .T/  X .T/ ! Z denote the natural pairing. Let ˆ D ˆ.G; T/ denote the set of
roots of T in G; it is a finite subset of X  .T/, and g has the root space decomposition
M
gDt˚ g˛ ;
˛2ˆ

where t is the Lie algebra of T and the root subspace g˛ is defined by

g˛ D fX 2 g j Ad.t/X D ˛.t/X for all t 2 Tg:

The standard affine apartment A in the building B.G/ is the affine space underlying
the vector space X .T/ ˝Z R, together with a hyperplane structure.
Let W.ˆ/ D W.G; T/ denote the Weyl group of T in G. The Weyl group is
generated by reflections through hyperplanes corresponding to each root ˛ 2 ˆ; its
action on X  .T/ preserves ˆ.
For each ˛ 2 ˆ and n 2 Z, we consider the affine functional, or affine root,
˛ C n W A ! R defined for each x D  ˝ s 2 A by

.˛ C n/.x/ D h˛ C n;  ˝ xi D sh˛; i C n:
64 S.M. Frechette et al.

Put ‰ D f˛ C n j ˛ 2 ˆ; n 2 Zg; and for each D ˛ C n 2 ‰ consider the

hyperplane
H D fx 2 A j .x/ D 0g:

The set of all such hyperplanes forms a hyperplane structure on A.

4.1.1 The Standard Apartment for sl.n/

When G D SLn , let T be the diagonal torus and consider the apartment A
corresponding to T. We identify X .T/ with Zn , and think of cocharacters explicitly
as functions t 7! diag.tx1 ; : : : ; txn /, for t 2 F  and .x1 ; : : : ; xn / 2 Zn . Identify A
with X .T/ ˝ R, and for each i with 1  i  n, define the mapping ei W A ! R by
ei .f ˝ s/ D sxi , for f D .t 7! diag.tx1 ; tx2 ; : : : ; txn // 2 X .T/ and s 2 R. Then the
set of roots is given by

ˆ D fei  ej j 1  i ¤ j  ng: (9)

Each of the root spaces gei ej is one-dimensional. We may view gei ej as being
spanned by the matrix Eij whose entries are all zero except for the .i; j/-entry, which
equals 1.

4.1.2 The Standard Apartment for sp2n

When G D Sp2n , again let T be the diagonal torus, whose elements are of the form
 D diag.t1 ; t2 ; : : : ; tn ; t11 ; t21 ; : : : ; tn1 / by our choice of the embedding.
The rank of sp2n is n, so we have X  .T/ ' X .T/ ' Zn , as abelian groups.
Similar to the sln case, for 1  i  n, define the mapping ei W A ! R by ei .f ˝ s/ D
sxi , for f D .t 7! diag.t1x1 ; t2x2 ; : : : ; tnxn ; t1x1 ; t2x2 ; : : : ; tnxn // and x 2 R. Then ˆ is
the set

ˆ D fei  ej ; ˙.ei C ej /; ˙2ei j 1  i ¤ j  ng: (10)

Finally, let A be the standard apartment of sp2n relative to this root datum.
Below in Sections 5.3 and 5.4, we discuss in detail the examples of sl.3/ and
sp.4/.

4.2 DeBacker’s Parametrization Using the Building

Generalizing the work of Barbasch and Moy (1997), DeBacker (2002b) developed
a parametrization of nilpotent orbits that relies on facets in the Bruhat-Tits building
Shalika Germs for sln and sp2n Are Motivic 65

of g and is valid for any reductive Lie algebra over F, provided the residue
characteristic is sufficiently large. This is actually a family of parametrizations that
depends on a real parameter r, although for our purposes it suffices to consider the
case corresponding to r D 0, in the notation of DeBacker (2002b). To ease notation,
we omit the r-dependence in DeBacker’s notation and state suitably modified
versions of the relevant theorems with r D 0.
Let A denote the standard affine apartment corresponding to the Lie algebra g.
The set A has the structure of a simplicial complex (generally, polysimplicial, but
the groups we are considering in this paper are simple). Let us define the facets (i.e.
the simplices) in the apartment A. For x 2 A and n 2 Z, define the sets

ˆx D f˛ 2 ˆ j ˛.x/ 2 Zg and Hn D fx 2 A j jˆx j D ng: (11)

For an integer n, a facet of A is defined to be any connected component F of Hn .

We denote by A.F ; A/ the smallest affine subspace of A containing F . With this,
we define the dimension of a facet to be dim.F / D dim A.F ; A/, hence facets of
the apartment A have bounded dimension.
Given a subspace H of A, a facet F  H is said to be maximal if the dimension
of F is maximal among the dimensions of facets contained in H. An alcove is the
closure of any facet of maximal dimension in A.
For example, H0 consists of all x 2 A for which ˛.x/ … Z for all ˛ 2 ˆ. This is
the set of points x 2 A that do not lie on any of the hyperplanes H˛m for any ˛ 2 ˆ
and any m 2 Z. Thus any connected component of H0 is the interior of some alcove
in A. For instance, in the case of sl3 or sp4 , these facets will be two-dimensional.
Also for sl3 or sp4 , any connected component of H1 is the edge of an alcove, and
any connected component of H2 is a vertex of an alcove.

4.2.1 Moy-Prasad Filtration Lattices, and Generalized Facets

For each pair .x; r/ with x 2 A and r 2 R, Moy and Prasad (1994) define certain
O-lattices gx;r giving a filtration of g. The parameter r is referred to as the depth of
the lattice. Since we consider only the case r D 0, we suppress r from the notation
throughout.
Associated with each root ˛ 2 ˆ we have the root subgroup U˛ , a T.F/-invariant,
closed one-parameter subgroup of G, and the root subspace g˛ , which coincides
with the tangent space of U˛ . (For the rest of this section, we reserve boldface letters
for algebraic groups, and their non-boldface counterparts for the groups of rational
points.) Note that our groups G, T and U˛ are in fact defined over Z, and thus we
can talk about the well-defined subgroups G.O/, U˛ .O/, etc.
With this notation (see Rabinoff 2005, § 3 for more detail of the notation) for
each x 2 A, define the parahoric subgroup Gx as

Gx D hT.O/; U˛ .Pb˛.x/c / j ˛ 2 ˆiI

66 S.M. Frechette et al.

its pro-unipotent radical is

1d˛.x/e
GC
x D hT.P C 1/; U˛ .P / j ˛ 2 ˆi:

(This turns out to be equivalent to the more complicated standard definition, cf.
Rabinoff 2005, (3.1).)
Similarly, for each x 2 A, we have the corresponding lattices gx  gC
x in the Lie
algebra:

gx D hh.O/; Pb˛.x/c X˛ j ˛ 2 ˆi (12)

and
1d˛.x/e
x D hh.P/; P
gC X˛ j ˛ 2 ˆi; (13)

where the root space g˛ is spanned by the element X˛ , and h D Lie.T/ is the
Cartan subalgebra of g corresponding to T. (More precisely, by choosing a splitting
.B; T; fx˛ g/ of G, defined over Z, we would then have the corresponding generators
X˛ of g˛ . For our classical Lie algebras, X˛ are the standard generators of the
corresponding root spaces; see examples in Section 5 below.)
If x; y 2 A are contained in the same facet F , then we have Gx D Gy and
GCx D Gy , as well as gx D gy and gx D gy . For a given facet F , we will simply
C C C
C
write gF and gF for the lattices associated with any x 2 F . We also have need of
the quotient of these lattices, denoted VF D gF =gCF , which is a Lie algebra over kF .
In order to state DeBacker’s parametrization theorem, we need to define an
equivalence relation on facets, and in order to do that, we require the notion of a
generalized facet. For each x 2 B.G/, the set

F D fy 2 B.G/ j gx D gy and gC C
x D gy g (14)

is called the generalized facet containing x. We say two generalized facets F1 and F2
are strongly associate if A.F1 \ A; A/ D A.F2 \ A; A/ ¤ ;, for some apartment A.
If there exists an element g 2 G such that F1 and gF2 are strongly associate, then
we say F1 and F2 are associate. For two facets F1 and F2 contained in a given
apartment A, we say that F1 and F2 are associate if the generalized facets they
determine are associate.
Remark 10. In this paper, thanks to the explicit parametrization of orbits and
Nevins’ matching theorem, we need not interpret this notion of associate using
Denef-Pas language; the fact that this notion involves the whole building and not
just a single apartment is one of the main obstructions we currently perceive to
obtaining our main result for general Lie algebras.
Shalika Germs for sln and sp2n Are Motivic 67

4.2.2 DeBacker’s Parametrization

We say an element v 2 VF is degenerate if the coset it parametrizes contains a

nilpotent element (i.e. if there exists a nilpotent X 2 gF such that v D X C gC
F ). Let
I.F/ be the set given by

I.F/ D f.F ; v/ j F  A is a facet, and v 2 VF is a degenerate elementg: (15)

DeBacker defines an equivalence relation

on I.F/: say that .F1 ; v1 /
.F2 ; v2 /
if and only if there exists g 2 G such that A.F1 ; A/ D A.gF2 ; A/, and such that
under the resulting natural identification of VF1 with Ad.g/VF2 , the elements v1 and
Ad.g/v2 lie in the same orbit under Gx for any x 2 F1 .
Let Nil.F/ denote the set of rational nilpotent orbits in g. Using the theory of sl2 -
triples, DeBacker proves the following results regarding the relationship between
the sets I.F/ and Nil.F/.
Lemma 11 (DeBacker (2002b)). Suppose the residue characteristic of F is suffi-
ciently large, and .F ; v/ 2 I.F/. Then
1. (Lemma 5.3.3, r D 0 case) There exists a unique nilpotent orbit of minimal
dimension which intersects the coset v nontrivially. We denote this nilpotent orbit
by O.F ; v/.
2. (Lemma 5.4.1, r D 0 case) The map  W I.F/=
! Nil.F/ defined by
.F ; v/ 7! O.F ; v/ is a well-defined, surjective map.
However, this map is not injective. (A detailed explanation of this phenomenon
is given in Nevins 2011.) To obtain a one-to-one correspondence, we must restrict
to the subset of distinguished pairs. We say a pair .F ; v/ 2 I.F/ is distinguished if
v is not an element of any proper Levi subalgebra of the VF . Let

I d .F/ D f.F ; v/ 2 I.F/ j .F ; v/ is distinguishedg: (16)

Theorem 12 (DeBacker (2002b), r D 0 Case of Theorem 5.6.1). Suppose

the residue characteristic of F is sufficiently large. Then there is a bijective
correspondence between I d .F/=
and the set of nilpotent orbits in g.F/ given
by the map which sends .F ; v/ to O.F ; v/.
Proof. Theorem 5.6.1 from DeBacker (2002b) contains this statement for a slightly
different parameter space, allowing the facets F of I d .F/ to run over the enlarged
Bruhat-Tits building of g. By virtue of Theorem 5.6 from Nevins (2011), one may
substitute the parameter space I d .F/ defined above.
68 S.M. Frechette et al.

5 Explicit Matching Between the Two Parametrizations

Following Nevins (2011), we give a correspondence between the parametrization

involving partitions and DeBacker’s parametrization defined in terms of the build-
ing, for the cases g D sln and sp2n . (In order for both parametrizations to be valid, we
must again assume that the residue characteristic of F is not 2, and the characteristic
of F itself is zero, or sufficiently large.) We also work out the examples of sl3 and
sp4 in complete detail.
In Section 6 we will associate certain definable functions with each nilpotent
orbit. For that purpose it would be very convenient to use DeBacker’s parametriza-
tion, however, the set Nil.F/ itself is more easily understood through Waldspurger’s
parametrization via partitions. Thus, it is necessary to understand the explicit
matching between these two parametrizations.

5.1 The Matching for sln

For each partition  D .1 ; 2 ; : : : ; t / of n, and each diagonal matrix D D

diag.d1 ; d2 ; : : : ; dn / 2 T, define the set
X
I D f1; 2; : : : ; ng n f1 ; 1 C 2 ; : : : ; i D ng:
i

I represents the set of locations of the nonzero entries of the matrices J D.d/ 2
O .F/ described in Proposition 5. For each i 2 I , the value diC1 is the .i; i C 1/-
entry of X, and all remaining entries are zero.
Recall the hyperplanes H'Cn D fx 2 A j '.x/ D ng, defined for roots ' 2 ˆ.
Also recall the standard notation ˛i D ei  eiC1 for the simple roots of SLn . With
the notation as above, define
\
H;D D H˛i Cval.diC1 /  A:
i2I

Note that when  D .1; 1; : : : ; 1/, we have X D 0 and I D ;. Nevins (2011) states
that the zero orbit corresponds to the associate class of the interior of any alcove
in the apartment. The following theorem of Nevins establishes the remainder of the
correspondence between the two parametrizations of the nilpotent orbits for sln .
Theorem 13 (Nevins (2011), Theorem 2 with r D 0). Let ; D, and H;D  A
be as above, and let F be any facet of maximal dimension in H;D . For any x 2 F ,
we have X D J D 2 gF ; set v to be its image in VF . Then .F ; v/ 2 I d .F/ and
O.F ; v/ D Ad.sln .F// X.
Shalika Germs for sln and sp2n Are Motivic 69

5.2 The Matching for sp2n

Let X 2 O .F/ be the nilpotent element corresponding to the n-tuple Q, as in

Theorem 9. For each odd j, let

Ij D f1; : : : ; 12 jmj g n fj; 2j; : : : ; 12 jmj g

and let Sj D Sj1 denote the set of simple roots

Sj1 D fesj Ck  esj CkC1 j k 2 Ij g:

For each even j, suppose Qj D qm 0 ˚ Qaniso is the minimal matrix representative for
Qj , where m is the Witt index of Qj (0  2m  mj ), and set Mj D . 12 j  1/mj . Then
we take Sj D Sj1 [ Sj2 , where

Sj1 D fesj Ck  esj CkCmj j 1  k  Mj g [ fesj CMj C2i1 C esj CMj C2i j 1  i  mg

and

Sj2 D f2esjCMj Ci j 2m < i  mj g:

If Qaniso D diag.a2mC1 ; : : : ; amj /, define for each root ˛i D 2esj CMj Ci the integer
v˛i D val.ai / for 2m C 1  i  mj . Let H;Q be the common intersection (over
all j) of the hyperplanes H˛ for ˛ 2 Sj1 and H˛Cv˛ for ˛ 2 S2j 2
. Finally, the following
theorem of Nevins gives the correspondence between the two parametrizations of
nilpotent orbits for sp2n :
Theorem 14 (Nevins (2011), Theorem 4 with r D 0). The affine subspace
H;Q  A is a nonempty union of facets. Let F be any maximal facet in H;Q ,
and let v denote the projection of X in VF . Then .F ; v/ 2 I d .F/ and O.F ; v/ D
Ad.sp2n .F// X.

5.3 Example: The Correspondence, in the Case of sl3

We examine these parametrizations and their correspondence in the case of the Lie
algebra g D sl3 . Representatives X for the nilpotent orbits O are given above in
Example 6. Following the construction given in Section 5.1, we compute the sets
H;D.d/ as in Theorem 13, where D D D.d/ D diag.1; 1; : : : ; 1; d/, for d 2 F  . We
then determine the maximal facet F  H;D corresponding to each orbit O.1;1;1/ and
O.2;1/ , and to each of the F-rational nilpotent orbits contained in O.3/ .
For the partition  D .1; 1; 1/; we have I.3/ D ; and X D 0. In this trivial case,
the corresponding maximal facet is the interior of any alcove in the apartment A.
70 S.M. Frechette et al.

Hα1
Hα1

−1
H(α 1+ α 2)− 1

Hα 1+ α 2
1
2−
α
H

1
2+
2
α
H

α
H

Fig. 1 Standard affine apartment of sl3 .F/. Solid edges outline an alcove, and dotted lines indicate
affine hyperplanes. Dots indicate the sets H.3/;D.d/

Fig. 2 Facets Fi in the given H(α 1+ α 2)− 1

alcove of the standard affine F5 F4
apartment of sl3 .F/. The
facet F2 is associate with the
other edges F1
Hα 1 Hα 2
F2

0 = F3

The standard apartment for sl3 is shown above in Figure 1. We may choose the
alcove given by the outlined region, which is bounded by the hyperplanes H˛1 , H˛2 ,
and H.˛1 C˛2 /1 . This is denoted by F1 above in Figure 2.
When  D .2; 1/; we have I.2;1/ D f1g and H.2;1/;D.d/ D H˛1 for any d 2 F  .
Any facet of maximal dimension in H.2;1/;D , is therefore an edge in an alcove of A.
Since the three edges of any alcove are associates, it suffices to consider a single
edge, denoted F2 .
For  D .3/;
we have  I.3/ D f1; 2g and H.3/;D.d/ D H˛1 \ H˛2 Cval.d/ : Recall that
0 10
we have X.3/ D 0 0 1 and this orbit splits into 3  gcd.3; q 1/ orbits O .F/ whose
0 00
representatives are given by
0 1
010
Xd D @0 0 dA ; one for each distinct equivalence class of d in F  =.F  /3 .
000
(17)
Shalika Germs for sln and sp2n Are Motivic 71

When 3 − .q1/ we have d 2 f1; $; $ 2 g. When d j .q1/, we fix a non-cubic unit

" 2 F  , and have d 2 f1; "; "2; $; "$; "2 $; $ 2 ; "$ 2 ; "2 $ 2 g. In any case, H.3/;D.d/
will be a single point, thus any facet of maximal dimension in H.3/;D.d/ will consist
of a single element.
Specifically, when val.d/ D 0, then H.3/;D.d/ D f0g, with corresponding facet
denoted by F3 in Figure 2. Taking val.d/ D 1 gives H.3/;D.d/ D H˛1 \ H˛2 C1 which
is not in the chosen alcove. For our purposes in handling definability, it is convenient
to fix a single alcove. With this in mind, we note that G.F/ acts on A via the affine
Weyl group; and so, reflecting this point across the hyperplane H˛1 C˛2 to the upper-
right vertex of the alcove, we see that this facet and the facet denoted by F4 are
associates. Similarly, val.d/ D 2 gives H.3/;D.d/ D H˛1 \ H˛2 C2 , which maps to
facet F5 under the affine Weyl group action (e.g. by reflecting across H˛2 and then
H˛1 C1 ).
In order to calculate the lattices associated with each facet Fi , we first determine
the root spaces g˛ for ˛ 2 ˆ. By (9) (with ˛i D ei  eiC1 ), we have

ˆ D f ˛1 ; ˛2 ; ˛1 C ˛2 ; ˛1 ; ˛2 ; .˛1 C ˛2 / g:

Each root space is one-dimensional, and the generators for the six root spaces are
given by the matrices
0 1 0 1 0 1
010 000 00 1
X˛1 D @0 0 0A X˛2 D @0 0 1A X˛1 C˛2 D @0 0 0A
000 000 00 0
0 1 0 1 0 1
000 00 0 000
X˛1 D @1 0 0A X˛2 D @0 0 0A X.˛1 C˛2 / D @0 0 0A
000 01 0 100

Sample calculations for the lattices associated with the facets F1 and F4 are
given in detail below, followed by a table with full results for each of the five facets
mentioned above. Although the lattices associated with H.3/;D.d/ when val.d/ D
1 will be different from the lattices gF4 and gC F4 , the quotients will be identical.
Therefore we may consider F4 when determining the image v of Xd in the quotient.
The case val.d/ D 2 is handled similarly. In the same spirit, we compute gF for the
edge that is labelled F2 in Figure 2; though the lattices gF and gC F are different for
the other two associate edges, the quotients VF for them are isomorphic.

Facet F1 : x 2 F1 if and only if 0 < ˛1 .x/; ˛2 .x/; .˛1 C ˛2 /.x/ < 1: For any
x 2 F1 this gives b˛1 .x/c D 0 and d˛1 .x/e D 1, while b˛1 .x/c D 1
and d˛1 .x/e D 0. We have Pb˛1 .x/c D P0 D O and Pb˛1 .x/c D
P1 D P, hence
72 S.M. Frechette et al.

0 1 0 1
0O 0 0 0 0
Pb˛1 .x/c X˛1 @
D 0 0 A
0 and P b˛1 .x/c @
X˛1 D P 0 0A :
0 0 0 0 0 0

We may similarly calculate Pb'.x/c X for each of the roots ' D

0 1
O 0 0
˙˛2 ; ˙.˛1 C ˛2 /. Finally, h.O/ D @ 0 O 0 A, and so we obtain
0 0 O
0 1
OOO
gF1 D @P O OA :
PPO
1d˛1 .x/e
Next, P D O and P1d˛1 .x/e D P, and similarly for ˙˛2
and ˙.˛1 C ˛2 /. Computing the corresponding lattice representative
0 1
P 0 0
for each root, and using the fact that h.P/ D @ 0 P 0 A ; we find
0 0 P
0 1
POO
that gC D @P P OA : Since kF D O=P, taking the quotient gives
F1
PPP
0 1
kF 0 0
VF1 D gF1 =gC @ 0 kF 0 A :
F1 D
0 0 kF
Facet F4 : This vertex lies on the hyperplanes H˛1 1 , H˛2 , and H.˛1 C˛2 /1 , so
x 2 F4 if and only if ˛1 .x/ D .˛1 C ˛2 /.x/ D 1 and ˛2 .x/ D 0.
Thus Pb'.x/c D P1 and P1d'.x/e D O for ' D ˛1 and ˛1 C ˛2 ,
and similarly Pb'.x/c D P and P1d'.x/e D P2 for ' D ˛1 and
.˛1 C ˛2 /. For the remaining roots, we have Pb˙˛2 .x/c D O and
P1d˙˛2 .x/e D P. Adding the corresponding matrix representatives, we
obtain
0 1 0 1
O P1 P1 P OO
gF 4 D @P O O A and gC @ 2 A
F4 D P P P :
2
P O O P PP

Identifying Pa with $ a O, we get isomorphisms of Pa =PaC1 with

0 1
kF kF kF
@
kF D O=P. Thus taking the quotient gives VF4 D kF kF kF A :
kF kF kF
Shalika Germs for sln and sp2n Are Motivic 73

Parahoric gF Pro-unipotent gC VF
0 1 0 F1 0 1
OOO POO kF 0 0
B C B C B C
gF1 D @P O OA gC
F1 D @P P OA VF1 D@0 kF 0 A
PPO PPP 0 0 kF
0 1 0 1 0 1
OOO POO kF 0 0
B C B C B C
gF2 D @P O OA gC
F2 D @P P PA VF2 D@0 kF kF A
POO PPP 0 kF kF
0 1 0 1 0 1
OOO PPP kF kF kF
B C B C B C
gF3 D @O O OA gC
F3 D @P P PA VF3 D @kF kF kF A
OOO PPP k kF kF
0 1 0 1 0 F 1
O P1 P1 P OO kF kF kF
B C B 2 C B C
gF4 D @P O O A gC
F4 D @P P PA VF4 D @kF kF kF A
2
P O O P PP kF kF kF
0 1 0 1 0 1
O O P1 P P O kF kF kF
B C B C B C
gF5 D @O O P1 A gC
F5 D @ P P O A VF5 D @kF kF kF A
2 2
PP O P P P kF kF kF

Finally, we determine the image v of X in VF for each representative X and

Xd as above. The nilpotent orbit O.1;1;1/ has representative X.1;1;1/ D 0, with
corresponding facet F1 . Its image v in VF1 is simply the zero matrix. Similarly,
0 1
010
O.2;1/ has representative X.2;1/ D @0 0 0A, whose corresponding facet is an
000
0 1
010
associate of F2 . Its image in VF2 is v.2;1/ D @0 0 0A : For  D .3/, with d D "a $ b
000
(0  a; b  2) and Xd as in Equation (17), the image of Xd in the quotient VFbC2 is
0 1
01 0
vd D @0 0 ac.d/A :
00 0

5.4 Example: sp4

We now examine the two parametrizations and their correspondence in the case of
the Lie algebra g D sp4 . As in Theorem 8, consider only the partitions of 4 whose
odd parts have even multiplicity, i.e.  D .4/; .2; 2/; .2; 1; 1/; and .1; 1; 1; 1/: Each
orbit O splits into a certain number of F-rational orbits, depending on . (For
details, see Nevins (2011), Table 1.) Below we give the details for the rational
nilpotent orbits contained in the algebraic orbit O.4/ .
74 S.M. Frechette et al.

The partition  D .4/ corresponds to the nilpotent matrix

0 1 0 1
01 0 0 111 0
B0 0 0 1C B 1C
XDB C D YJ.4/ Y 1 ; where Y D B0 1 1 C:
@0 0 0 0A @0 0 0 1A
00 1 0 001 1

We have m4 ./ D 1 and mi ./ D 0 for i D 1; 2; 3, and therefore the vector

space V ' F 4 satisfies V D V.4/. By Theorem 9, we may use Equation (8) with
X D X jV.4/ , j D 4, N D 2, and mj D 1 to determine representatives of the
F-rational nilpotent orbits in O .F/. By the results of Section 3.3, there are four
minimal matrix representatives of quadratic forms of dimension m4 D 1; given by
0 1
0 1 0 0
B0 0 0 aC
Xa D B
@0
C;
0 0 0A
0 0 1 0

such that a runs over the set f1; "; $; "$g, where " is a fixed non-square unit in F.
We now turn to the correspondence in Section 5.2. Fixing X D Xa , we have
Sj D ; for j ¤ 4. For j D 4, by the given construction we see that the Witt index m
of Q4 is equal to 0, and M4 D 1. Thus

S41 D fe1  e2 g; S42 D f2e2 g:

We have Qaniso D diag.a/, so for ˛1 D 2eM4 C1 D 2e2 , we have v˛1 D val.a/. Thus

H;Q D He1 e2 \ H2e2 Cval.a/ :

Now val.a/ is either 0 or 1, since " is a unit. Figure 3 shows the standard apartment
of sp4 , along with the hyperplanes He1 e2 , H2e2 , and H2e2 C1 and the associated
intersections H;Q of these hyperplanes. From the diagram, it is clear that there
is a unique maximal facet Fa (vertex) in each set H;Q , and Fa consists of a single
element.
In order to calculate the associated lattices, we first determine the root spaces g˛
for ˛ 2 ˆ. By (10), we have

ˆ D f˙.e1  e2 /; ˙.e1 C e2 /; ˙2e1 ; ˙2e2 g:

Each root space is one-dimensional, and the generators for the root spaces are given
by the matrices
Shalika Germs for sln and sp2n Are Motivic 75

0 1 0 1
0 1 0 0 0 00 0
B0 0 0 0C B1 00 0C
Xe1 e2 DB
@0
C Xe2 e1 DB C
0 0 0A @0 00 1A
0 0 1 0 0 00 0
0 1 0 1
00 01 000 0
B0 0 1 0C B0 0 0 0C
Xe1 Ce2 DB
@0 0
C X.e1 Ce2 / DB C
0 0A @0 1 0 0A
00 00 100 0
0 1 0 1
0 01 0 00 00
B0 00 0C B0 0 0 0C
X2e1 DB
@0
C X2e1 DB C
00 0A @1 0 0 0A
0 00 0 00 00
0 1 0 1
0 00 0 00 00
B0 00 1C B0 0 0 0C
X2e2 DB
@0
C X2e2 DB C
00 0A @0 0 0 0A
0 00 0 01 00

(2 e1)∨ (e1 + e2 )∨

(2e2)∨

(e2 − e1 )∨

e2
H2e2 +1

−
H2e2

e1
H

Fig. 3 The standard affine apartment of sp4 .F/. Arrows indicate positive co-roots, and dotted lines
indicate affine hyperplanes. Dots indicate the sets H;Q
76 S.M. Frechette et al.

Identify Fa with the single element that it contains. By referencing Figure 3, we

can calculate ˛.Fa / for ˛ 2 ˆ. Clearly ˛.Fa / D 0 for all ˛ when Fa D 0 (i.e.
when a D 1; "). Suppose a D $ or "$, so that Fa D 1 2 .e1 C e2 / . This gives
_

the values ˙2e1 .Fa / D 1, ˙2e2 .Fa / D 1, ˙.e1 C e2 /.Fa / D 1, and

˙.e1  e2 /.Fa / D 0:
Using the definition of the lattices gF and gC
F given in (12) and (13), respectively,
we compute the following:
If a D 1 or ", we have
0 1 0 1
O OO O P P PP
BO OO OC B P PC
gF a DB C and gC D BP P C:
@O OO OA Fa @P P P PA
O OO O P P PP

If a D $ or "$, we have
0 1 0 1
O O PP PP P2 P2
B O O C
P PC B P2 P2 C
gF a DB and gC BP P C:
@P1 Fa D @
P1 O OA OO P PA
P1 P1 OO OO P P

It is clear that we have kF in each entry of the quotient in both cases, hence
VF ' sp4 .kF /. Finally, we determine the image va of Xa in VF . Then we have
0 1
0 1 0 0
B0 0 0 ac.a/C
va D B
@0
C for each a 2 f1; "; $; "$g.
0 0 0 A
0 0 1 0

6 Shalika Germs

6.1 The Main Results

Here we prove that the so-called provisional Shalika germs are motivic (in the
terminology of Kottwitz 2005, § 6). Harish-Chandra defined Shalika germs on the
full Lie algebra, using their homogeneity (see Kottwitz 2005, § 17 for a detailed
discussion). Here we will show, roughly, that for every nilpotent orbit there exists
a motivic function that coincides (up to a motivic constant) with the Shalika germ
corresponding to that orbit on a definable neighbourhood of the origin. However,
rescaling any given element of the Lie algebra so that it would fall into this
neighbourhood presents a slight problem from the definable point of view, and
so we shall address the full question of homogeneity elsewhere. It turns out that
Shalika Germs for sln and sp2n Are Motivic 77

the existence of motivic functions that represent the Shalika germs in some small
neighbourhood of the origin is sufficient for the application we have in mind,
namely, the uniform-in-p bound on the normalized Shalika germs, which appears
in Theorem 17 below.
Theorem 15. Let g D sln or sp2n . Let N be the set of nilpotent elements in g. Then
(1) There exists a definable set E, such that EF is finite for all fields F of sufficiently
large residue characteristic, and a definable function h W N ! E, such that for
every d 2 E, h1 .d/ is an adjoint orbit, and each orbit appears as the fibre of h.
(2) There exist motivic functions  on E  grss and C on E, and a constant M > 0,
such that for all local fields F of residue characteristic greater than M, for
every d 2 EF , the function C1 F .d; / is a representative of the Shalika
germ on grss corresponding to d, i.e., coincides with the Shalika germ on some
neighbourhood of the origin.
Proof. (1). First, note that the set of nilpotent elements N is, indeed, definable: it is
defined by the formula X n D 0 in sln and by the formula X 2n D 0 in sp2n .
For g D sp2n , recall the parametrization of the nilpotent orbits from Theorem 9,
and let E be the set of pairs .; Q/ as in Theorem 8. Note that for each pair
.; Q/ 2 EF , there is an explicit representative X.;Q/ 2 NF . The definition of X.;Q/
involves only constant symbols in the extended Denef-Pas language; hence, the orbit
of X.;Q/ is a definable set, and the map h can be seen explicitly in Theorem 9.
For g D sln , the proof is essentially carried out in Diwadkar (2006, Section 6);
here, we reinterpret it using the most recent version of motivic integration, and
state it more generally. We are assuming that we are working with G D SLn ,
and n is fixed. There is a certain awkwardness to the proof caused by the fact that
quotients, even by very nice definable equivalence relations, are not easy (sometimes
impossible) to code in a first-order language.
Here we need to make a construction that allows us to handle the quotient
F  =.F  /m , where F is the valued field, and so we use the union of the languages
LDPm defined above in Section 2.3, as m runs over the divisors of n.
More precisely, for every partition  of n, add the symbols for constants of the
valued field sort d;1 ; : : : ; d;m , where m D gcd./.
Now, define the set E as the disjoint union over all partitions  of n, of sets
E , defined as follows. Given a partition , let m D gcd./ as above. We have m
constant symbols corresponding to this partition, d;1 ; : : : ; d;m , in the language.
Recall the formulas `;m from (4) in Section 2.3. With this value of m, exactly
one of the formulas `;m WD ‘9y1 ; : : : ; y` `;m .y1 ; : : : ; y` /’ holds. If `;m holds,
we interpret the constant symbols d;1 ; : : : ; d;` as units of the valued field such
that `;m .ac.d;1 /; : : : ; ac.d;` // holds. Set the rest of the di equal to 1. (Note that
this construction of the language is consistent with Section 2.3, but incorporates the
union over .) Then let

E WD [kD0
m1
f$ k d;1 ; : : : ; $ k d;m g:

This is a definable set since it consists of just the constant symbols in the language.
78 S.M. Frechette et al.

Now, for every partition , and every d 2 E≥ F , we have the elements Xd as

defined in Proposition 5. The disjoint union of the orbits of these elements is
precisely NF .
We will also need two observations about the set E for the proof of the second
part of the theorem:
(1) The set E a parametrizing orbits of dimension at least a is definable for a D
0; : : : ; n.
(2) The cardinality of the set E a is bounded independently of the field.
The first observation holds since the dimension of the orbit depends only on the
partition ; hence, the set E a is the disjoint union of E over a prescribed set of
partitions  that depends only on a. The second observation is immediate from the
definition. Indeed, with the above notation,
P in the case of sln the upper bound on the
size of E a is given by N.n; a/ WD  gcd./2 , where the sum is over the partitions
 of n that give rise to orbits of dimension at least a. In the case of sp2n , the statement
is trivial since the number of orbits of a given dimension is field-independent from
the start.
Now we turn our attention to Part (2).
(2). First, let us discuss the restriction on the residue characteristic of F. We will be
using Cluckers et al. (2014a, Corollary 4.4) which states (in our case, without
exponentials):
Given a family of definable test functions ffa ga2S  Cc1 .g/ with some definable set
S, there exists a constant M and a motivic function h on g  S such that for all non-
Archimedean local fields F of residue characteristic greater than M,

X .fa / D hF .X; a/:

Here we use this corollary with S D E. Recall from Part (1) that there exists a
constant M0 such that for all F with residue characteristic greater than M0 , for
every d 2 EF , we have an element Xd 2 g.F/ (an explicit matrix whose entries are
constant symbols in the language LDPm for some m), and the set fXd gd2EF is a set of
representatives of nilpotent orbits in g.F/. By the matching theorem (Theorem 13
for the case of sln and Theorem 14 for sp2n ), there exists a unique pair .F ; v/ that
corresponds to the orbit of Xd ; in particular, Xd 2 gF and v D Xd mod gC F 2 VF .
Note that gC C
F is an open compact subset of g.F/, and so is its translate Xd C gF . Let
C
fd be the characteristic function of the coset Xd C gF . It is definable by Cluckers
et al. (2014a, Lemma 3.2). Thus we have a family of definable test functions ffd gd2E
indexed by the definable set E. Let M be the maximum of M0 and the constant
from the statement of Cluckers et al. (2014a, Corollary 4.4) quoted above, for this
specific family. This will be the constant that appears as the restriction on the residue
characteristic in our theorem.
Now we are ready to prove the statement of Part (2), for fields F with residue
characteristic greater than M. The argument proceeds by downward induction on
the dimension of the nilpotent orbit. The base case is an orbit of the top dimension,
and the idea is to construct a definable test function whose support intersects only
Shalika Germs for sln and sp2n Are Motivic 79

this orbit, which allows us to isolate the Shalika germ attached to the chosen orbit.
For orbits of smaller dimension it is of course not possible to isolate a single Shalika
germ, but it is possible to construct a definable test function whose support intersects
only the given orbit and orbits of strictly higher dimension. This is where a theorem
of Barbasch and Moy, refined by DeBacker and quoted above as Lemma 11, is
needed, and this is how downward induction on the dimension proceeds.
Thus, for the base case, let  D .n/ be the partition that gives rise to the orbits
of the maximal dimension, which we denote by amax . Let F be a local field with
residue characteristic greater than M, and let d 2 E≥ F . Let Xd be the explicit
representative of the corresponding orbit, as above. Let fd be the corresponding test
function constructed above, i.e., the characteristic function of the coset Xd C gF C ,
with .F ; v/ the pair corresponding to Xd .
By Lemma 11, the orbit of Xd is the unique nilpotent orbit of minimal dimension
intersecting Xd C gC F ; since there are no orbits of dimension greater than amax , the
orbit of Xd is the unique nilpotent orbit intersecting the support of the test function fd .
Thus, for the test function fd the Shalika germ expansion has only one term, namely

X .fd / D Xd .X/Xd .fd /;

where the expansion holds for X 2 Ufd \ g.F/rss , with Ufd some neighbourhood
of the origin (which depends on the test function fd ). By Cluckers et al. (2014a,
Corollary 4.4), Xd .fd / is a motivic function of d (that is, for a fixed d, a motivic
constant, which we denote by C.d/), and X .fd / is a motivic function of X and d.
More precisely, there exists a motivic function .X; d/ such that F .X; d/ D X .fd /.
(Note that here we are using our definition of the constant M, and the assumption
that the residue characteristic of F is greater than M.) If necessary, we can shrink
Ud to make it definable. (Since Ud is open, there exists a lattice of the form gx;r ,
i.e., defined entirely by inequalities on the valuations of the entries of X, which is
contained in Ud .) This establishes the base case.
Now, let us assume the statement of the theorem holds for the orbits of dimension
at least a (where a is an even integer). Let F be as above, and let d 2 EF be a point
such that the orbit of Xd has dimension a  2. As above, there exists a unique pair
.F ; v/ that corresponds to the orbit of Xd , (i.e. the orbit of Xd is the unique orbit
of minimal dimension intersecting Xd C gC F ). Let fd be the characteristic function
of the coset Xd C gC F , as above. Then the intersection of its support with NF is the
union of its intersection with the orbit of Xd , and subsets of orbits of strictly higher
dimension, i.e., of dimension at least a. Then there exists a neighbourhood of 0,
which we will denote by Ufd , such that for X 2 Ufd ,
X
X .fd / D Xd .X/Xd .fd / C Xd0 .X/Xd0 .fd /;
d 0 2EFa
80 S.M. Frechette et al.

where the sum runs over the set of representatives of nilpotent orbits of dimension
at least a. As in the base case, we can shrink Ufd to make it definable. Then, for
X 2 Ufd , we have
X
Xd .X/d .fd / D X .fd /  Xd0 .X/Xd0 .fd /: (18)
a
d 0 2EF

The right-hand side of (18) is almost a motivic function. (Almost, because the
Shalika germs corresponding to the orbits of greater dimension labelled by the
points d0 are ratios of motivic functions and motivic constants.) More precisely,
by the inductive assumption, for the Shalika germs occurring on the right-hand side
of (18), we have Xd0 D C.d0 /1 F .X; d0 / in some definable neighbourhood Ud0 of
the origin. Let U be the intersection of Ufd and all the Ud0 where d0 runs over E a .
Clearing denominators on bothQsides, we see that it remains only to prove that the
product of the motivic constants E a C.d0 / is itself a motivic constant. In the case
of sp2n this is clear, since the indexing set in the product is field-independent, so we
just have a fixed finite product of motivic constants. In the case of sln , recall the sets
E a defined in Part (1) above. Let Pa be the set of partitions  that give rise to the
orbits of dimension at least a, so that E a D [2Pa E . Recall from Part (1) that for
each partition  we have the constant symbols d;1 ; : : : ; d;m , where m D gcd./,
and some of these constants specialize to 1 in a given field F, depending on the
number of roots of unity in F. By definition, we have

G mG
 1

E a D f$ j d;1 ; : : : ; $ j d;m g:
2Pa jD0

Let us define, for each  2 Pa and each `; j with 0  j  m  1 and 1  `  m, a

constant function
(
1 if d;` D 1
'`;j WD
C.d / if d 0 D $ j d;` with d;` ¤ 1:
0

Then we can write

0 1
Y Y Y
m mY
 1 mY
 1

C.d0 / D @ '`;j C.$ j /A :

E a 2Pa `D1 jD0 jD0

(Note that here, because of our convention on the interpretation of the symbols d;`
in the given field, the trivial coset of k =.k /m plays a special role. The functions
'`;j are defined in order to remove all occurrences of the trivial coset from the
product, and the last factor in the product formula above re-introduces the trivial
Shalika Germs for sln and sp2n Are Motivic 81

Q
coset, counted with correct multiplicity.) Thus we have represented E a C.d0 / as a
product of a fixed (i.e. field-independent) number of motivic constants, which proves
it is itself a motivic constant, and completes the proof of the induction step. t
u
We note for future reference that the motivic constants C.d0 / are positive, since
they are obtained as products of volumes of definable sets.

6.2 Corollaries

The first consequence of Theorem 15 is an alternate proof that Shalika germ

expansion holds in large positive characteristic. (This is already known, thanks to the
work of DeBacker.) Indeed, if an equality of motivic functions holds in characteristic
zero, it holds in large positive characteristic by the Transfer Principle of Cluckers
and Loeser (2008).
However, the results of Appendix B to Shin and Templier 2015, allow us to also
prove a different type of corollary. First, we must recall some notation and a theorem
of Harish-Chandra, which we quote here from Kottwitz (2005, Q Theorem 17.9).
For a regular semisimple element X of g, let D.X/ D ˛2ˆ j˛.X/j be the Weyl
discriminant of X (cf. Kottwitz 2005, § 7 for alternative definitions). For d 2 EF , let
 d .X/ WD jD.X/j1=2 d .X/ be the normalized Shalika germ. Note that here we mean
the canonical Shalika germ, not just the provisional Shalika germ considered above
in Theorem 15; thus, it is a function defined on the set of all regular semisimple
elements in g.F/. Let T be a maximal torus of G, and t its Lie algebra. Harish-
Chandra proved the following result.
Theorem 16 (Harish-Chandra (1999), Kottwitz (2005, Theorem 17.9)). Every
normalized Shalika germ  d is a locally bounded function on t. (Here the local field
F is assumed to have characteristic zero.)
Now, suppose we fix a definable compact subset in t (or, more generally, a family
of such definable compact subsets), so that we can vary the local field and still talk
about the bound for the normalized Shalika germs, restricted to the specific set. We
can ask, how does the bound on  d depend on the field F? (Or, on the compact
subset in question?) The next theorem answers both questions. Note that it is more
convenient for us to talk about subsets of g rather than subsets of t. Since there
are finitely many conjugacy classes of tori (with an upper bound on their number
independent of the field), and since Shalika germs are conjugation-invariant, the
local boundedness on grss follows.
We would like to state a general result on the dependence of the bound for Shalika
germs restricted to a compact subset of g.F/ on the field F and on the compact
subset in question. Typically, the compact subsets one is interested in are Moy-
Prasad filtration lattices or other similar subsets. Since here we are working with
82 S.M. Frechette et al.

explicitly defined Lie algebras, we will say that the compact sets Kn form a family
of congruence lattices if every set Kn is defined by the formulas

ord.Xij /  ˛ij .n/;

where ˛ij are Z-valued Presburger-definable functions of the parameter n. We

believe most natural situations where questions about uniform bounds arise should
satisfy this property; for example, Moy-Prasad filtration lattices in classical Lie
algebras satisfy this condition.
Theorem 17. Let g D sln or sp2n . Let Kn be a family of congruence lattices in g,
indexed by a parameter n 2 Z. Then there exists a constant M > 0 (that depends
only on the formulas defining the family Kn ), and definable Z-valued functions a
and b on E, such that for all local fields F with residue characteristic greater than
M, for every d 2 EF ,

j d .X/j  qa.d/Cb.d/n for X 2 KnF \ grss .F/; (19)

where q is the cardinality of the residue field of F.

Proof. Note that since EF is a finite set with an upper bound on its cardinality
independent of F, an equivalent formulation would be to demand the existence of
constants a and b such that (19) holds, independently of d.
Let Ud be the neighbourhood given in the proof of Theorem 15. Recall that this
is a definable neighbourhood on which the Shalika germ expansion holds for the
specific test function fd constructed in that proof. Let U be the intersection of the
definable sets Ud , for d 2 EF . (It is non-empty and definable since the cardinality of
EF is bounded independently of F.) Then on the set U, by Theorem 15, we have that

 d .X/ D jD.X/j1=2 C.d/1 d F .X/;

where C is a motivic function of d, and d is a motivic function of d and X. The

discriminant D.X/ is a definable function since it is a polynomial in the entries of
X (cf. Kottwitz 2005, § 7.5), hence, jD.X/j1=2 is a motivic function in our sense
(cf. Cluckers et al. 2014a, §B.3.1). Thus the right-hand side is the ratio of a motivic
function of X and d, and a motivic function C.d/. Since for each d, C.d/ is a positive
motivic constant, i.e., an element of ZŒq1 ; .1  qi /1 ; i > 0, and since #EF is
bounded independently of F, there exist constants a1 ; a2  0 such that qa2  C.d/ 
qa1 for all d 2 EF . By Harish-Chandra’s Theorem, quoted above as Theorem 16,
for every local field F of characteristic zero, there exists a constant Ad F (that depends
on F) such that

j d .X/j  AdF for X 2 UF :

Shalika Germs for sln and sp2n Are Motivic 83

Therefore, for the fields F of characteristic zero, we have an estimate for the motivic
function jD.X/j1=2 dF .X/, given by

jD.X/j1=2 jd F .X/j D C.d/j d .X/j  qa2 AdF for X 2 UF :

Then by the uniform boundedness principle for motivic functions, Shin and
Templier (2015, Theorem 14.6), there exist constants M and ad 2 Z such that for all
local fields F with residue characteristic greater than M, we have

jD.X/j1=2 jd F .X/j  qad for X 2 UF :

Finally, we obtain, for X 2 UF , that j d .X/j  qad Ca1 . Let a.d/ D ad Ca1 . Note that
since all of the functions involved were motivic functions of d, this constant a.d/
depends definably on d (though as noted at the beginning of the proof, this seems to
be unimportant). Thus, we have proved the theorem for one specific definable open
compact set – namely, U.
Now we can extend it to an arbitrary family of congruence lattices fKn gn>0 using
the homogeneity of Shalika germs. Namely, for every n there exists an integer j.n/
such that $ j.n/ Kn  U, and j.n/ is a Presburger-definable function of n. Indeed,
U has to contain some congruence lattice defined by ord.Xij /  ˇij , with some
constants ˇij 2 Z. Then we can take j.n/ WD maxi;j .˛ij .n/ C ˇij /, where ˛ij are the
functions in the definition of the family fKn gn>0 . Then, by definition of the canonical
Shalika germs d , we have

d .X/ D j$ j.n/ jm d .$ 2j.n/ X/;

where m is the dimension of the nilpotent orbit with parameter d. Note that by
definition, for t 2 F  , we have jD.tX/j D jtjdim.g/r jD.X/j, where r is the rank of g
(cf. Kottwitz 2005, (17.11.2)). Putting this together, we obtain, for X 2 KnF ,

 d .X/ D jD.X/j1=2 d .X/ D jD.$ 2j.n/ X/j1=2 j$ j.n/ jdim.g/r j$ j.n/ jm d .$ 2j.n/ X/
D qj.n/.dim.g/rm/  d .$ 2j.n/ X/:

Therefore, we have

j d .X/j  qa.d/Cj.n/.dim.g/rm/ :

It remains only to observe that since j.n/ is a Presburger function on Z, it is

piecewise-linear, and the statement follows. u
t

Acknowledgements This paper clearly owes a debt to the ideas of T.C. Hales and to the thesis
of Jyotsna Diwadkar. The second author is grateful to Raf Cluckers and Immanuel Halupczok for
multiple helpful communications. We thank the organizers of the WIN workshop in Luminy who
made this collaboration possible. The second and third authors were supported by NSERC.
84 S.M. Frechette et al.

References

Barbasch, D., Moy, A.: Local character expansions. Ann. Sci. École Norm. Sup. (4) 30(5), 553–567
(1997) (English, with English and French summaries). MR1474804 (99j:22021)
Cluckers, R., Loeser, F.: Constructible motivic functions and motivic integration. Invent. Math.
173(1), 23–121 (2008)
Cluckers, R., Loeser, F.: Motivic integration in all residue field characteristics for Henselian
discretely valued fields of characteristic zero. J. Reine Angew. Math. 701, pp. 1–31, (2015)
Cluckers, R., Gordon, J., Halupczok, I.: Local integrability results in harmonic analysis on
reductive groups in large positive characterisitic. Ann. Sci. Ec. Norm. Sup. http://arxiv.org/
abs/1111.7057 47(6) pp. 1163–1195 (2014a)
Cluckers, R., Gordon, J., Halupczok, I.: Motivic functions, integrability, and applications to
harmonic analysis on p-adic groups. Electron. Res. Announc. (2014b). http://arxiv.org/abs/
1111.7057 21, pp. 137–152, (2014)
Cluckers, R., Hales, T., Loeser, F.: Transfer principle for the fundamental lemma. In: Clozel,
L., Harris, M., Labesse, J.-P., Ngô, B.-C. (eds.) On the Stabilization of the Trace Formula.
International Press, Boston (2011b)
Collingwood, D.H., McGovern, W.M.: Nilpotent Orbits in Semisimple Lie Algebras. Van Nostrand
Reinhold Mathematics Series. Van Nostrand Reinhold Co., New York (1993). MR1251060
(94j:17001)
DeBacker, S.: Homogeneity results for invariant distributions of a reductive p-adic group. Ann. Sci.
École Norm. Sup. (4) 35(3), 391–422 (2002a) (English, with English and French summaries).
MR1914003 (2003i:22019)
DeBacker, S.: Parametrizing nilpotent orbits via Bruhat-Tits theory. Ann. Math. (2) 156(1),
295–332 (2002b). MR1935848 (2003i:20086)
Diwadkar, J.M.: Nilpotent conjugacy classes of reductive p-adic Lie algebras and definability in
Pas’s language. Ph.D. Thesis, University of Pittsburgh (2006)
Harish-Chandra: Harmonic analysis on reductive p-adic groups. In: Moore, C.C. (ed.) Harmonic
Analysis on Homogeneous Spaces. Proceedings of Symposia in Pure Mathematics, vol. 26, pp.
167–192. American Mathematical Society, Providence (1973). MR0340486 (49 #5238)
Harish-Chandra: Admissible invariant Distributions on Reductive p-Adic Groups. University
Lecture Series, vol. 16. American Mathematical Society, Providence (1999). With a preface
and notes by Stephen DeBacker and Paul J. Sally, Jr. MR1702257 (2001b:22015)
Kottwitz, R.E.: Harmonic analysis on reductive p-adic groups and Lie algebras. In: Harmonic
Analysis, the Trace Formula, and Shimura Varieties, pp. 393–522 (2005). MR2192014,
American mathematical Society, Providence, RI for Clay Mathematics Institute, Cambridge,
MA, (2006m:22016)
Lam, T.Y.: Introduction to Quadratic Forms Over Fields. Graduate Studies in Mathematics, vol.
67. American Mathematical Society, Providence (2005)
McNinch, G.J.: Nilpotent orbits over ground fields of good characteristic. Math. Ann. 329(1),
49–85 (2004). MR2052869 (2005j:17018)
Moy, A., Prasad, G.: Unrefined minimal K-types for p-adic groups. Invent. Math. 116(1–3),
393–408 (1994). MR1253198 (95f:22023)
Nevins, M.: On nilpotent orbits of SLn and Sp2n over a local non-Archimedean field. Algebra
Representation Theory 14, 161–190 (2011)
Rabinoff, J.: The Bruhat-Tits building of a p-adic Chevalley group and an application to
representation theory. Harvard Senior Thesis (2005)
Ranga Rao, R.: Orbital integrals in reductive groups. Ann. Math. (2) 96, 505–510 (1972).
MR0320232 (47 #8771)
Robson, L.: Shalika Germs are Motivic. M.Sc. Essay, University of British Columbia, Vancouver
(2012)
Shalika, J.A.: A theorem on semi-simple P -adic groups. Ann. Math. (2) 95, 226–242 (1972).
MR0323957 (48 #2310)
Shalika Germs for sln and sp2n Are Motivic 85

Shin, S.W., Templier, N.: Sato-Tate Theorem for Families and Low-Lying Zeroes of Automorphic
L-functions. Preprint. http://arxiv.org/abs/1208.1945. With appendices by R. Kottwitz and
J. Gordon, R. Cluckers and I. Halupczok, Invent. Math. (2015)
Waldspurger, J.-L.: Intégrales orbitales nilpotentes et endoscopie pour les groupes classiques non
ramifiés. Astérisque 269, vi+449 (2001)
The Conjectural Relation Between Generalized
Shalika Models on SO4n .F/ and Symplectic
Linear Models on Sp4n .F/: A Toy Example

Agnès David, Marcela Hanzer, and Judith Ludwig

Abstract We show that if an irreducible admissible representation of SO4 .F/ has a

generalized Shalika model, its theta lift to Sp4 .F/ is non-zero and has a symplectic
linear model.

1 Introduction

The recent progress towards proving the Local Langlands Conjectures for classical
groups (cf. Arthur 2013; Gan and Takeda 2010a, 2011; Cogdell et al. 2011; Jiang
and Soudry 2012; Ginzburg et al. 2001 and many more) increased the interest in
understanding characterizations of images of Langlands functorial transfers and the
finer structures of L- and A-packets. One way of distinguishing representations is
by the models they have. As an example of how models can be used to characterize
images of transfers consider the following situation: Let F=Qp be a finite extension
and let  be an irreducible unitary supercuspidal representation of GL2n .F/. Then
(cf. Jiang et al. 2010b, Theorem 1.1)  is a local Langlands functorial transfer
from SO2nC1 .F/ if and only if  has a Shalika model. Furthermore it turns out
that the existence of certain models of representations of different groups is very
much related through Langlands type correspondences. In this article we investigate
how generalized Shalika models on the split group SO4 .F/ are related to symplectic
linear models on Sp4 .F/ via the local theta correspondence.

A. David
Mathematical Research Unit, University of Luxembourg, Luxembourg, Luxembourg
M. Hanzer ()
Department of Mathematics, University of Zagreb, Zagreb, Croatia
e-mail: hanmar@math.hr
J. Ludwig
Mathematical Institute, University of Bonn, Bonn, Germany

© Springer International Publishing Switzerland 2015 87

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_4
88 A. David et al.

More precisely in Jiang et al. (2010a) the authors conjecture the following:
Conjecture 1.1 (Jiang et al. (2010a), p. 542). Let  be an irreducible admissible
representation of SO4n .F/ which has a generalized Shalika model. Then the
representation ./ of Sp4n .F/ associated with  via the local theta correspondence
is non-zero and has a symplectic linear model.
Here, and in the remainder of the paper, F=Qp is a finite extension. Furthermore
.; m/ denotes the “small” theta lift of a representation  and ‚.; m/ denotes the
“big” theta lift of  to the symplectic group Sp2m .F/ (cf. Kudla 1996, p. 33). If m is
understood, we denote ‚.; m/ by ‚./ and .; m/ by ./. The dual pair used
in this theta correspondence consists of a symplectic and a full orthogonal group
and the restriction to the special orthogonal group is explained below.
The goal of this article is to prove
Theorem (Theorem 5.2). Conjecture 1.1 is true for n D 1.
The result that led to the conjecture in the first place and provides evidence
for it is
Theorem 1.1 (Jiang et al. (2010b), Theorems 1.1 and 1.2). Let  be an irre-
ducible unitary supercuspidal representation of GL2n .F/ which has a Shalika model.
Then the Langlands quotient  of the induced representation  1=2 ÌSO4n .F/ 1 has a
generalized Shalika model. The Langlands quotient  of the induced representation
 1=2 ÌSp4n .F/ 1 has a symplectic linear model. Furthermore ./ D .
Here  denotes the character of GL2n .F/ obtained by composing the determinant
with the norm on the non-archimedean field F and we may regard any character of
F  as a character of GL2n .F/ analogously. Throughout the text we use Zelevinsky’s
notation for the parabolic induction for the general linear groups and for classical
groups as introduced, e.g., in Tadić (1998) Sections 1 and 2.
We will eventually prove Theorem 5.2 by reducing it to a calculation of Jacquet-
modules. The following two results make this reduction possible.
Theorem 1.2 (Jiang et al. (2013), Theorem 1.2). Let  be an irreducible admissi-
ble representation of SO4n .F/ and assume it has a generalized Shalika model. Then
there exists an irreducible admissible representation  of GL2n .F/ such that  is a
quotient of the induced representation

 1=2 ÌSO4n .F/ 1  :

The following theorem is the specialization of Theorem 1.3 in Jiang et al. (2013)
to our situation .n D 1/.1

1
The N1 model is the Shalika model for GL2 , the other one the symplectic model, see the
paragraph before Theorem 1.3 in Jiang et al. (2013).
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 89

Theorem 1.3 (Jiang et al. (2013), Theorem 1.3). Let  be an irreducible admis-
sible representation of GL2 .F/. If the induced representation  1=2 ÌSO4 .F/ 1 has a
generalized Shalika model, then  either has a symplectic model or a Shalika model.

The strategy to prove Theorem 5.2 is as follows: The theorems quoted above
allow to first study the representations . We determine the set of representations
 that admit a symplectic model or a Shalika model. We then analyze them case
by case and study the representations obtained when we parabolically induce  to
representations of SO4 .F/ and Sp4 .F/, respectively. We check when these induced
representations have the respective models. In particular we need to make sure in any
of the cases that the set of ’s for which the induction to SO4 .F/ has a generalized
Shalika model agrees with the set of ’s for which the induction to Sp4 .F/ has
a symplectic linear model. Finally we verify that the models factor through the
relevant quotients of the inductions and that these quotients are related via the theta
correspondence.
The key to get our hands on these representations is the following: We can
show the existence of some of the models by proving that a certain (twisted)
Jacquet-module has a trivial quotient. We prove the existence of these quotients by
calculating the Jacquet-modules explicitly using the Geometric Lemma of Bernstein
and Zelevinsky (see Theorem 5.1 in Bernstein and Zelevinsky 1977).
As the groups we study have such small rank we can explicitly describe the
possible representations  and can then do explicit Jacquet-module calculations to
find all the information on the models we need. In higher rank we cannot pin down
the representations  as explicitly. From the general version of Theorem 1.3 we
know that they all have a Nr -model, but there is no explicit description of such
representations. Therefore our method of analyzing everything explicitly case by
case will not be successful and proving Conjecture 1.1 in general will require a
different approach.
The plan of the paper is as follows: In Section 2 we recall the various models and
determine the set of representations  of GL2 .F/ that admit a symplectic model or
a Shalika model. We also give some background on the theta correspondence. We
then prove Theorem 5.2 case by case in Sections 3–5 depending on the properties
of . Section 3 deals with the square-integrable case. In Section 4 we study the
case where  is a character. We finish the proof by treating the case where  is an
irreducible principal series representation in Section 5.

2 Notation and Preliminaries

We recall the various models occurring in these notes specialized to the case at hand.
For the general definitions we refer to Jiang et al. (2010a) Section 2. Let F=Qp be a
finite extension and fix a non-trivial additive character W F ! C . Let
90 A. David et al.

0 1
1
B 1 C
Jn WD B
@ ...
C 2 GLn .F/
A
1

and set J WD J4 . In the special orthogonal group SO4 , whose F-points are given by

SO4 .F/ D fA 2 GL4 .F/j T AJA D J; detA D 1g;

we fix the maximal diagonal torus T and the Borel subgroup B of upper triangular
matrices. We let P D MN be the standard maximal parabolic subgroup, whose Levi
subgroup M is isomorphic to GL2 .
It is embedded via
!
g 0
 W GL2 .F/ ,! SO4 .F/; g 7! 1
0 J2 T g J2

and the F-points of its unipotent radical N are given by all matrices
 
I2 X
y.X/ D ;
0 I2

such that T X D J2 XJ2 . We refer to P as the Siegel subgroup. The subgroup
H  P.F/ generated by all .g/ for g 2 Sp2 .F/ and all y 2 N.F/ is called the
generalized Shalikasubgroup 4 .F/. We extend
 ofSO to a character H W H ! C
1 0
by H .y.X// D tr X and by demanding it is trivial on .Sp2 .F//.
0 1
Definition 2.1. An irreducible admissible representation  of SO4 .F/ is said to
have a generalized Shalika model if

HomH .; H/ ¤ 0:

Definition 2.2. Let  be an irreducible admissible representation of GL2 .F/.

• The representation  has a Shalika model if

HomS .; S/ ¤ 0;
 ˇ 
a x ˇˇ
where S D a 2 F 
; x 2 F  GL2 .F/ is the Shalika subgroup and
0a ˇ
 
ax
we have extended to a character S W S ! C ; S 
D .x=a/.
0a
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 91

• The representation  has a symplectic model if

HomSp2 .F/ .; 1Sp2 .F/ / ¤ 0:

•  has a linear model if

HomGL1 .F/GL1 .F/ .; 1GL1 .F/GL1 .F/ / ¤ 0:

Remark. In the previous definitions, we followed the conventions from Jiang

et al. (2010a). As it is clear from the definitions, and will be mentioned below
in Lemma 2.2, an irreducible representation of GL2 .F/ having a Shalika model
necessarily has trivial central character. There is also a more general definition of
the Shalika model, accounting for the representations which do not necessarily have
trivial central character (cf. Gan and Takeda 2010b).
Theorem 2.1 (Jacquet and Rallis (1996), Section 6). If an irreducible admissible
representation  of GL2n .F/ has a Shalika model, then  has a linear model.
In the symplectic group Sp4 , whose F-points are given by
 ˇ    
ˇ 0 J2 0 J2
Sp4 .F/ D A 2 GL4 .F/ˇˇ T A AD ;
J2 0 J2 0

we fix the maximal diagonal torus T and the Borel subgroup B of upper triangular
matrices. We have a standard maximal parabolic subgroup P D MN with Levi M Š
GL2 embedded via
 
g 0
GL2 .F/ ,! Sp4 .F/; g 7! :
0 J2 T g1 J2

The group Sp2 .F/  Sp2 .F/ injects into Sp4 .F/ via
0 1
a b
   
ab wx B wx C
; 7! B C
@ y z A: (1)
cd y z
c d

Definition 2.3. An irreducible admissible representation  on Sp4 .F/ has a sym-

plectic linear model if

HomSp2 .F/Sp2 .F/ .; 1Sp2 .F/Sp2 .F/ / ¤ 0:

Remark. Note that in the definitions of models all representations are assumed to
be irreducible. If the corresponding Hom-space for an admissible not necessarily
92 A. David et al.

irreducible representation is non-zero, we speak of functionals instead of models,

so, e.g., if  is a possibly reducible admissible representation of SO4 .F/ such that
HomH .; H / ¤ 0, we say that  has a non-zero generalized Shalika functional.
Lemma 2.2. Let  be an irreducible admissible representation of GL2 .F/.
1. If  has a symplectic model, then  is a character.
2. The representation  has a Shalika model if and only if  is generic with trivial
central character.

Proof. For the proof of the first part note that Sp2 D SL2 . So  has a symplectic
model if and only if there exists a non-zero functional  2 HomSL2 .F/ .; 1SL2 /. Then
V =ker./ Š 1SL2 .F/ as a representation of SL2 .F/. The restriction of any smooth
irreducible representation of GL2 .F/ to SL2 .F/ decomposes into a finite direct sum
of irreducible representations, each occurring with multiplicity one (cf. Labesse and
Langlands 1979, Lemmas 2.4 & 2.6). Furthermore the representations occurring in
the restriction are permuted by any set S of representatives of GL2 .F/= SL2 .F/F  .
More precisely for any two irreducible smooth representations 1 and 2 occurring
in jSL2 .F/ , there exists g 2 S such that 1 Š g 2 , where g 2 is the representation
given by g 2 .h/ D 2 .g1 hg/ for h 2 SL2 .F/. Therefore if jSL2 .F/ contains the
trivial representation as a subrepresentation,  D  is a character.
For the second part we unravel the definitions to see that any non-zero  2
HomS .; S / is in fact a Whittaker functional on V . Furthermore we have
   
t 0
.! .t/v/ D   v D .v/
0t

for all v 2 V and t 2 F  if and only if the central character ! is trivial. t

u
So here are the options for :
1. The representation  is supercuspidal. Note that having trivial central character
implies that  is unitary and so we are in special case of Theorem 1.1 above,
where the implication of Conjecture 1.1 is known to hold.
2.  is a generic subquotient of a reducible principal series. Then  is an essentially
square-integrable representation with trivial central character and in particular
 is unitary. It follows that

 ,!  1=2   1=2 Š . 1=2   1=2 /

and therefore  Š  StGL2 .F/ ; where StGL2 .F/ denotes the Steinberg representa-
tion of GL2 .F/: The condition that the central character is trivial furthermore
implies that 2 D 1.
3. The representation  is a character. Note we can write  D 0  s , where 0 is
unitary and  s D jdetjs , where s 2 R.
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 93

4.  is an irreducible principal series representation,  Š 1  s1  2  s2 , where

1 ; 2 are unitary characters and s1 ; s2 2 R. The condition that the central
character is trivial gives

1 2 D 1 and s1 C s2 D 0:

Therefore  Š  s  1  s for a unitary character .

We briefly explain how to restrict the theta correspondence between symplectic
and full orthogonal groups to the correspondence between representations of
symplectic and special orthogonal groups. Let Ø2n .F/ be the element
0 1
In1
B 1 C
DB
@
C:
A
1
In1

For an irreducible admissible representation  of SO2n .F/; we denote by   the

representation of SO2n .F/ on the same space, defined by   .g/ D .g 1 /: Recall
that we can pass between irreducible admissible representations of O2n .F/ and
SO2n .F/ as follows:
Lemma 2.3 (cf. Mœglin et al. (1987) 3.II.5, Lemme).
1. Let  be an irreducible admissible representation of O2n .F/. Then jSO2n .F/ is
irreducible if and only if  ©  ˝ det.
2. Let  be an irreducible admissible representation of SO2n .F/. Then either (A)
O .F/
 ©   , in which case IndSO2n2n .F/ ./ DW  is irreducible and satisfies  D
O .F/
 ˝ det, or (B)  Š   in which case IndSO2n2n .F/ ./ is reducible and the direct
sum of two non-equivalent irreducible representations  and  ˝ det.
We fix a non-trivial additive character  of F. All Weil representations occurring
in this article will be with respect to this character. Furthermore for n D 1; 2 we
fix the splittings O2n .F/  Sp2n .F/ ! Mp4n2 .F/ and for later purposes O4 .F/ 
Sp2 .F/ ! Mp8 .F/ as described explicitly in Kudla (1994).
Remark. Note that we do not demand that  D ; where entered the definition
of the generalized Shalika model. The theta correspondence in general does depend
on the character  of F; but in our case, the theta lifts of representations of SO4 .F/
we consider are the same for every choice of : This follows from our explicit
computations of theta lifts below and can be explained, e.g., by the description of
these representations as Langlands quotients, cf. Appendix C of Gan and Ichino
(2014).
Using the above lemma we can restrict the theta correspondence, i.e., we
can relate the largest Sp2n .F/-invariant quotient which is an isotype of  in the
appropriate Weil representation with the similar quotient corresponding to  as
94 A. David et al.

follows: Let  be an irreducible admissible representation of SO2n .F/. Then

.; n/ WD .; n/ if (A) (2)

.; n/ WD .; n/ ˚ . ˝ det; n/ if (B)

Remark. We often use the following fact: assume that  is an irreducible represen-
tation of O4 .F/ such that  Š  ˝ det: Then, the first occurrence index of  in
theta correspondence, denoted by n./; is exactly 2; i.e.,  occurs for the first time
in theta correspondence with Sp4 .F/. This follows from the general fact (Sun and
Zhu 2012):
Theorem 2.4. Assume that  is an irreducible admissible representation of the split
O2n .F/: Then the following holds:

n./ C n. ˝ det/ D 2n:

3 The Case of Square-Integrable 

From now on, if  is a standard representation of a classical group, we denote by

L./ its Langlands quotient.
Lemma 3.1. Let  be a quadratic character of F  . Then the representation
 StGL2 .F/  1=2 ÌSp4 .F/ 1 is of length three if  ¤ 1 and of length two if  D 1.
In the case  D 1 there is an irreducible, tempered subrepresentation (necessarily
generic), and in the case  ¤ 1 two non-equivalent, irreducible, square-integrable
subrepresentations. The Langlands quotient L. StGL2 .F/  1=2 ÌSp4 .F/ 1/ has a
symplectic linear model.
Proof. The reducibility issues are dealt with in Sally and Tadić (1993), Proposi-
tion 5.4 and Theorem 5.2. On the other hand, we know that  StGL2 .F/ has a Shalika
model and so by Theorem 2.1 it also has a linear model. Then we reason as in
Ginzburg et al. (1999), p. 878 to conclude that  StGL2 .F/  1=2 ÌSp4 .F/ 1 has a non-
zero symplectic linear functional. Since irreducible generic representations cannot
have a symplectic linear model (cf. Ginzburg et al. 1999, Theorem 1), in the case of
 D 1 we see that this functional factors through the Langlands quotient and gives
a model. In the case  ¤ 1; for a fixed non-degenerate character ; one of the two
square-integrable representations is -generic, and the other is not. It is not difficult
to see that the other one is generic with respect to some other generic character, and
the conclusion follows. t
u
Lemma 3.2. Let  be a quadratic character of F  . Then the representa-
tion  StGL2 .F/  1=2 ÌSO4 .F/ 1 is of length two and its Langlands quotient
L. StGL2 .F/  1=2 ÌSO4 .F/ 1/ has a generalized Shalika model.
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 95

Proof. In this proof, we use the equality sign to denote the equality of the
representations up to the semisimplification. Note that  ÌO2 .F/ 1 is reducible. Thus,
the same Jacquet module calculation as in (cf. Sally and Tadić 1993, Proposition 5.4)
gives us that the representation  1=2 StGL2 .F/ ÌO4 .F/ 1 is of length three and we have
(thus, up to the semisimplification)

 1=2 StGL2 .F/ ÌO4 .F/ 1 D L. 1=2 StGL2 .F/ Ì1/ C 1 C 2 ;

where the i ; i D 1; 2 are (mutually non-isomorphic) square integrable represen-

tations. Note that also 1 ,!  1 Ì ; and 2 ,!  1 Ì  ˝ det; because
 Ì 1 D 1O2 .F/ ˚ detO2 .F/ : From this it follows that 1 ˝ det ,!  1 Ì  ˝ det;
and analogously, 2 ˝ det ,!  1 Ì ; so that 1 Š 2 ˝ det. We conclude
that L. 1=2 StGL2 .F/ ÌO4 .F/ 1/ ˝ det Š L. 1=2 StGL2 .F/ ÌO4 .F/ 1/. Therefore the
restriction of  1=2 StGL2 .F/ ÌO4 .F/ 1 to SO4 .F/ decomposes as

 1=2 StGL2 .F/ ÌO4 .F/ 1jSO4 .F/

D . 1=2 StGL2 .F/ ÌSO4 .F/ 1/ ˚ . 1=2 StGL2 .F/ ÌSO4 .F/ 1/
D L. 1=2 StGL2 .F/ ÌO4 .F/ 1/jSO4 .F/ C 1 jSO4 .F/ C 2 jSO4 .F/
D L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/ C .L. 1=2 StGL2 .F/ ÌSO4 .F/ 1// C 1 jSO4 .F/
C 2 jSO4 .F/

where 1 jSO4 .F/ D 2 jSO4 .F/ and these two representations are irreducible
(as 1 © 2 ). We conclude

 1=2 StGL2 .F/ ÌSO4 .F/ 1 D L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/ C 1 jSO4 .F/ :

The existence of the generalized Shalika model on L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/ now
follows from the general result (Jiang and Qin 2007, Theorem 3.1). Alternatively
we will see that we can argue directly: the subquotient L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/
appears again in Proposition 4.2 and in the proof there we can see the existence of
the model directly. t
u
Next we determine the theta lift of L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/. As far as we know,
unlike in the case of an odd orthogonal-metaplectic pair (cf. Gan and Savin 2012),
there still is no explicit description of the theta correspondence for a symplectic-even
orthogonal dual pair at the same level. We, therefore, directly calculate the lift by
calculating enough bits of one of its Jacquet modules. The main ingredient in these
calculations is Kudla’s filtration of the Jacquet modules of the Weil representations
involved (cf. Kudla 1996, III.8).
For a parabolic subgroup P of a group G and an admissible representation  of
G.F/ we denote by rP ./ the Jacquet module of  with respect to P. We define
P1 D M1 N1 to be the standard parabolic subgroup of Sp4 with Levi M1 isomorphic
96 A. David et al.

to GL1  SL2 and we define Q1 D M10 N10 to be the standard parabolic of O4 with
Levi M10 isomorphic to GL1  O2 : The Jacquet-module functor rP1 induces a functor
from Rep.O4 .F/  Sp4 .F// to Rep.O4 .F/  M1 /, which we will again denote by rP1 .
For n D 1; 2 we denote by !n;n the Weil representation of the double cover
Mp4n2 .F/ of Sp4n2 .F/ viewed as a representation of a dual pair .Sp2n .F/; O2n .F//
(inside Sp4n2 .F/). The Weil representation of Mp8 .F/ viewed as a representation of
.SL2 .F/; O4 .F// will be denoted by !1;2 .
The next proposition could be derived from the much more general results of
Gan and Ichino (2014, Proposition C.4(ii) in Appendix C), but we would have to
adapt their argument a little bit since, in the notation of Gan and Ichino (2014), in
our case l D 1 but Proposition C.4(ii) in loc. cit. describes the case of l D 1. So
we decided to give a direct proof.
Proposition 3.3. Assume 2 D 1. Then the theta lift of the representation
L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/ to Sp4 .F/ is

.L. 1=2 StGL2 .F/ ÌSO4 .F/ 1// D L. 1=2 StGL2 .F/ ÌSp4 .F/ 1/:

Proof. Theorem 2.4 guarantees that

n.L. 1=2 StGL2 .F/ ÌSO4 .F/ 1// D 2

with .L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/; 2/ D .L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2/ by the
calculation in the proof of Lemma 3.2 and (2). To determine the theta lift
.L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2/ we compute a part of its rP1 -Jacquet module. We
have

!2;2  L. 1=2 StGL2 .F/ ÌO4 .F/ 1/ ˝ .L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2/;

so that there is a non-zero intertwining operator, say T; of O4 .F/GL1 .F/SL2 .F/-

modules such that
1
T W rP1 .!2;2 /  L. 1=2 StGL2 .F/ ÌO4 .F/ 1/ ˝ rP1 . .L. 2 StGL2 .F/ ÌO4 .F/ 1/; 2//:

Now, rP1 .!2;2 / has a filtration

.1/ .0/
f0g  J1  J1 D rP1 .!2;2 /;

.1/ GL .F/SL .F/O .F/

such that J1 Š IndGL11 .F/SL22 .F/GL4 1 .F/O2 .F/ .1 ˝ !1;1 /; where 1 is the represen-
tation of GL1 .F/  GL1 .F/ by left and right translations on the space of smooth,
compactly supported functions on GL1 .F/ (denoted by S.GL1 .F//). Furthermore
.0/ .1/
we know that J1 =J1 Š  0 ˝ !1;2 : We have TjJ .1/ ¤ 0; as otherwise we
1
would have
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 97

 0 ˝ !1;2  L. 1=2 StGL2 .F/ ÌO4 .F/ 1/ ˝ rP1 . .L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2//

which would mean n.L. 1=2 StGL2 .F/ ÌO4 .F/ 1//  1; and that is impossible. The
second Frobenius reciprocity gives us that the space of GL1 .F/SL2 .F/GL1 .F/

G
O2 .F/-intertwining operators from 1 ˝ !1;1 to

rQ1 .L. 1=2 StGL2 .F/ ÌO4 .F/ 1// ˝ rP1 . .L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2//

is non-zero. Now, we use the fact that

1
rQ1 .L. 2 StGL2 .F/ ÌO4 .F/ 1// D  0 ˝ . 1 Ì 1/

(we shall prove this in Lemma 3.4). This means that there is an GL1 .F/-
epimorphism from 1 to  0 : But the maximal isotypic component of  0 in
S.GL1 .F// (when we view it as a GL1 .F/  GL1 .F/-module) is again  0 : This
means that rP1 . .L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2// has an irreducible subquotient of
the form  0 ˝ ; where is some representation of SL2 .F/:
Now we settle the case of  ¤ 1: Then, we can read off the cuspidal support of

.L. 1=2 StGL2 .F/ ÌSO4 .F/ 1/; 2/

(e.g. Kudla 1996); it is a subquotient of  1   0 ÌSp4 .F/ 1: The representation

 1   0 ÌSp4 .F/ 1 is of length six (cf. Proposition 5.4 of Sally and Tadić 1993),
and has analogous subquotients as a representation  1   0 ÌO4 .F/ 1: This means
that

rP1 .L. 1=2 StGL2 .F/ ÌSp4 .F/ 1// D  0 ˝  1 ÌSL2 .F/ 1;

and the representation L. 1=2 StGL2 .F/ ÌSp4 .F/ 1/ comes with the multiplicity two in
 1   0 ÌSp4 .F/ 1: Now, from the expression for rP1 . 1   0 ÌSp4 .F/ 1/ in the
1
proof of Lemma 3.4, we see that L. 2 StGL2 .F/ ÌSp4 .F/ 1/ is the only subquotient of
 1   0 ÌSp4 .F/ 1 having  0 ˝ in its rP1 -Jacquet module and the conclusion
follows.
Now let  D 1: We continue with the analysis of the Jacquet module rP1 .!2;2 /:
Since ‚.1SL2 ; 2/ D  1 ÌO2 .F/ 1, we have a non-zero intertwining from  0 ˝  0 ˝
!1;1 !  0 ˝  0 ˝ 1SL2 .F/ ˝  1 ÌO2 .F/ 1. Therefore,

rP1 . .L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2//   0 ˝ 1SL2 .F/

in the appropriate Grothendieck group. We conclude that

.L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2/ 2 fL. 1=2 StGL2 .F/ ÌSp4 .F/ 1/; L. 1   0 ÌSL2 .F/ 1/g:
98 A. David et al.

This follows from the fact that

 0 Ì 1SL2 .F/ D L. 1=2 StGL2 .F/ ÌSp4 .F/ 1/ ˚ L. 1   0 Ì 1/;

and these two representations are the only irreducible subquotients of  1  0 ÌSp4 .F/ 1
with  0 ˝ 1SL2 .F/ in their Jacquet module.
Assume now that .L. 1=2 StGL2 .F/ ÌO4 .F/ 1/; 2/ D L. 1   0 Ì 1/: Then, we have
an epimorphism

rP1 .!2;2 /  L. 1=2 StGL2 .F/ ÌO4 .F/ 1/ ˝ rP1 .L. 1   0 Ì 1//;

and since we have an epimorphism rP1 .L. 1   0 Ì 1// !  1 ˝  0 ÌSL2 .F/ 1; there
is a non-zero epimorphism, say T

rP1 .!2;2 /  L. 1=2 StGL2 .F/ ÌO4 .F/ 1/ ˝  1 ˝  0 ÌSL2 .F/ 1:

Now we analyze the restrictions of T to the terms of the filtration of rP1 .!2;2 /: We
get that there is a non-zero GL1 .F/  GL1 .F/  SL2 .F/  O2 .F/-intertwining

1 ˝ !1;1 !  0 ˝  1 ÌO2 .F/ 1 ˝  1 ˝  0 ÌSL2 .F/ 1;

which is impossible. This proves the proposition. t

u
2
Lemma 3.4. Assume  D 1. Then

rQ1 .L. 1=2 StGL2 .F/ ÌO4 .F/ 1// D  0 ˝  1 Ì 1:

Proof. We use the structure formula (*) on page 2 of Ban (1999) to compute the
Jacquet module of the induced representation  WD  1   0 ÌO4 .F/ 1 with respect
to Q1 . We get that

rQ1 ./ D  1 ˝ det C  1 ˝ 1O2 .F/ C  1 ˝ det

C  1 ˝ 1O2 .F/ C 2 0 ˝  1 Ì 1:

Since the multiplicity of L. 1=2 StGL2 .F/ ÌO4 .F/ 1/ in  is two the lemma follows.
t
u

4 The Case of the Representation  s Ì 1

In this section, we consider the case when the representation  is a character of

GL2 .F/. We write  as  s , with  a unitary character and s in R.
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 99

4.1 SO4 .F/-Side

First of all we prove that for any unitary character  and any s 2 R, the
representation  WD  s Ì 1 of SO4 .F/ has a generalized Shalika model by showing
that a certain twisted Jacquet-module of  has a trivial quotient.
The calculation is done by applying the Geometric Lemma and we adapt the
notation as follows: As before let H be the Shalika subgroup of SO4 .F/. We let
P WD P.F/ D MV denote the F-points of the Siegel subgroup, and we write H D
NV, where N Š SL2 .F/. We form the twisted Jacquet module rV; ./ of  with
respect to the group V and the character WD H jV . Recall that it is defined as the
SL2 .F/-module given by the quotient of  by the space

V; WD spanC f.X/f  .X/f W X 2 V; f 2 g:

Lemma 4.1. For any unitary character  of GL2 .F/ and any s in R, the represen-
tation  s ÌSO4 .F/ 1 has a non-zero generalized Shalika functional.
Proof. It follows from the definitions that if rV; . s ÌSO4 .F/ 1/ has a trivial quotient,
 s ÌSO4 .F/ 1 has a non-zero generalized Shalika functional. The geometric lemma
Bernstein and Zelevinsky (1977) gives a description of the composition of functors
F WD rV; ı iP;SO4 .F/ from GL2 .F/-representations to SL2 .F/-representations. Here
iP;SO4 .F/ denotes the functor of normalized parabolic induction. In order to apply
it note the following: Firstly, under the action of H by right translation, the space
P n SO4 .F/ decomposes into two orbits
  n SO4 .F/ D P [ Pw1 H (we easily get
P
0 I2
that Pw1 P D Pw1 H), where w1 D : Furthermore all the requirements such
I2 0
as good decomposition, etc. of Section 5.1 of Bernstein and Zelevinsky (1977) for
the triples .P; M; V/ and .H,N; V/ are satisfied. Abbreviate Y WD Pw1 H. There is an
SL2 .F/-invariant subspace Y of  s Ì 1 which consists of all functions in  s Ì 1
which vanish outside of Pw1 H. We apply the Jacquet functor rV; to the filtration of
SL2 .F/-representations

f0g  Y   s Ì 1:

Then Theorem 5.2 of Bernstein and Zelevinsky (1977) implies that rV; . s Ì1=Y /
is the SL2 .F/-module given by the restriction of  s from GL2 .F/ to SL2 .F/. We get
the same result for rV; .Y /. We conclude that the SL2 .F/–representation rV; . s Ì
1/ has length two and each subquotient is isomorphic to the trivial representation.
We conclude that rV; . s Ì 1/ has a trivial quotient. t
u
We remind the reader that we need to determine the irreducible quotients of  s Ì
1 and decide when they have a generalized Shalika model.
Proposition 4.2. Whenever the representation  s Ì 1 of SO4 .F/ is irreducible it
has a generalized Shalika model.
100 A. David et al.

1. Assume that 2 ¤ 1: Then the representation  s Ì 1 of SO4 .F/ is irreducible.

2. Assume that 2 D 1. The representation  s ÌSO4 .F/ 1 is reducible if
and only if s D ˙ 12 : Then, in the appropriate Grothendieck group,
1 1
 2 ÌSO4 .F/ 1 D L. StGL2 .F/  2 ÌSO4 .F/ 1/ C L. 1 Ì /. The representation
1
L. StGL2 .F/  2 ÌSO4 .F/ 1/ has a generalized Shalika model and L. 1 Ì / does
not admit one.
Proof. Note that if the representation  s Ì 1 of O4 .F/ is irreducible, then the
representation  s ÌSO4 .F/ 1 of SO4 .F/ is irreducible, since we saw that  s Ì
1jSO4 .F/ D  s ÌSO4 .F/ 1C. s ÌSO4 .F/ 1/ : So, the reducibility points for  s ÌSO4 .F/ 1
are among the reducibility points for  s Ì 1. A necessary condition for reducibility
here is 2 D 1: In that case, using the spinor norm, we have  s Ì 1 Š . s Ì 1/.
We can extend, as we already saw, the Jacquet module calculations from the case
of Sp4 .F/ to the O4 .F/ case if the rank-one reducibilities are the same (i.e., in
the case when  Ì 1 reduces in SL2 .F/). We conclude (cf. Sally and Tadić 1993,
Proposition 5.4.) that the only cases of reducibility are s D ˙ 12 ; and the length
of the representation  s Ì 1 is three. Now, analogously as in the case of the
1
representation  2 StGL2 .F/ Ì1, when restricting to SO4 .F/; we obtain that the
1
length of  2 ÌSO4 .F/ 1 is two. Namely, in O4 .F/
1 1
 2 Ì 1 D L. 2 StGL2 .F/ Ì1/ C L. 1 Ì / C L. 1 Ì  ˝ det/:
1 1
We get that L. 2 StGL2 .F/ Ì1/ Š L. 2 StGL2 .F/ Ì1/˝det and L. 1 Ì/jSO4 .F/ D
L. 1 Ì  ˝ det/jSO4 .F/ ; so that
1 1
 2 ÌSO4 .F/ 1 D L. 2 StGL2 .F/ ÌSO4 .F/ 1/ C L. 1 Ì /jSO4 .F/ :
1
We already know that L. 2 StGL2 .F/ ÌSO4 .F/ 1/ has a generalized Shalika model.
Take  D 1 for a moment. Then, it is easy to see that L. 1 Ì 1/jSO4 .F/ is actually the
trivial representation of SO4 .F/ and it does not admit a generalized Shalika model,
since is a non-trivial character of H. Similarly, if  ¤ 1; then  composed with
the spinor norm is equal to one, since the spinor norm on the Shalika subgroup is
trivial (cf. Lemma 2.2 on p. 79 of Kudla 1996 and Zassenhaus 1962 for the unipotent
1
elements in H/: We can thus also directly see that L. 2 StGL2 .F/ ÌSO4 .F/ 1/ has a
1
generalized Shalika model, since   2 ÌSO4 .F/ 1 has one and L. 1 Ì /jSO4 .F/ does
1 1
not, and L. 2 StGL2 .F/ ÌSO4 .F/ 1/ is a quotient of   2 ÌSO4 .F/ 1: t
u
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 101

4.2 Sp4 .F/-Side

We now analyze the representation  s ÌSp4 .F/ 1, where as above  is unitary and
s 2 R. In the following R.G/ denotes the category of smooth representations of the
topological group G.
Let S.SL2 .F// denote the space of smooth, compactly supported functions on
SL2 .F/. It comes equipped with an action R of the group SL2 .F/  SL2 .F/ given by
R.g1 ; g2 /.x/ D .g1
1 xg2 /, for  2 S.SL2 .F//, .g1 ; g2 / 2 SL2 .F/  SL2 .F/. In the
study of symplectic linear functionals on  s ÌSp4 .F/ 1 the following lemma will be
very useful.
Lemma 4.3. We have an exact sequence of SL2 .F/  SL2 .F/-representations

0 ! S.SL2 .F// !  s Ì 1jSL2 .F/SL2 .F/ !  sC1=2 Ì 1 ˝  sC1=2 Ì 1 ! 0: (3)

Proof. Again, we use Theorem 5.2 of Bernstein and Zelevinsky (1977) and adapt
our notation. So, P D MU denotes the Siegel parabolic subgroup of Sp4 .F/ (with
our previous choice of the Borel subgroup). We let Q D NV with Q D N D
SL2 .F/  SL2 .F/ and V D feg: We decompose rV;1 ı iP;Sp4 .F/ . s /. Here rV;1 turns
out to be just the restriction to SL2 .F/  SL2 .F/. To meet the requirements of the
geometric lemma (decomposability with respect to Q), we have to take M D P and
U D feg. We describe PnSp4 .F/=Q using (Kudla 1996, Chapter 4, Proposition 2.1).
We have just two orbits of Q-action on P n Sp4 .F/; there is an open orbit Pw1 Q,
 
I2 0
where w D , and (closed) orbit PQ: We have the following filtration
I2 I2
( D  s Ì 1)

0  1  ;

where 1 is a subset of functions in  vanishing outside of PwQ. In the notation of

Bernstein and Zelevinsky (1977) we get that the two subgroups which we use to
decompose rV;1 ı iP;Sp4 .F/ are
82 3 9
ˆ
ˆ ab 0 b >
>
<6   =
cd c 0 7
7 W a b 2 SL2 .F/
M0 D 6
ˆ40 0 a b5 c d >
>
:̂ ;
00 c d

and
82 3 9
ˆ
ˆ a 0 0 b >
>
<6   =
0 d c 07 a b
N 0 D wM 0 w1 D 6 7W 2 SL .F/ :
ˆ4 0 05 cd
2
b a >
>
:̂ ;
c 0 0 d
102 A. David et al.

Then, restricted to 1 ; rV;1 acts as iN 0 ;Q ı w ı rV 0 ;1 . s /. Note that the restriction

rV 0 ;1 W R.P/ ! R.M 0 / applied to  s gives the trivial character of M 0 , and
with conjugation by w, it again gives the trivial character of N 0 , so we have
iN 0 ;Q .1/ (compact induction). Note that N 0 is isomorphic to SL2 .F/, so that this
isomorphism gives an embedding SL2 .F/ ! SL2 .F/SL2 .F/ with g 7! .g; J2 gJ2 /:
By Proposition 2.3 in Chapter 4 of Kudla (1996) iN 0 ;Q .1/ Š S.SL2 .F//. The
intertwining operator from iN 0 ;Q .1/ to S.SL2 .F// is given by T.f /.x/ D f .1; J2 xJ2 /.
On the other hand, rV;1 on =1 is composed of iN 0 ;Q ı e ı rV 0 ;1 W R.P/ ! R.Q/;
where now N 0 D M 0 D P \ Q, so that rV 0 ;1 denotes the restriction of the
representation of P to representation of P \ Q and iN 0 ;Q is compact induction from
representations of P \ Q to representations of Q. It is easy to see that P \ Q consists
of matrices of the form
2 3
a1 0 0 b1
60 a2 b2 0 7
6 7:
40 0 a1 0 5
2
0 0 0 a1
1

Note that, in the case of the normalized induction,  s Ì 1 as a representation

1=2
of Sp4 .F/ is actually induced from the representation  s ıP , with our choice
 When we restrict to P \ Q, we get .
sC3=2
of M D   U D feg).
P (so that ˝
a 1 b1 a b2
 sC3=2 / ; 2 1 . When we then induce to SL2 .F/SL2 .F/, we get
0 a1
1 0 a2
 sC3=2 Ì0 1˝ sC3=2 Ì0 1 as a representations of SL2 .F/SL2 .F/ (the prime denotes
the unnormalized induction, so in our usual notation of the normalized induction,
we get  sC1=2 Ì 1 ˝  sC1=2 Ì 1). t
u
We now analyze (3) to see if  s Ì1jSL2 .F/SL2 .F/ has a trivial quotient. We use the
following fact: for any irreducible smooth representation  of SL2 .F/; the largest -
isotypic component of S.SL2 .F//jRSL2 .F/1 is  ˝ Q ; as an SL2 .F/  SL2 .F/-module.
Proposition 4.4. Let  be a unitary character and s 2 R. Assume .; s/ ¤
.1;  23 /. Then the representation  s ÌSp4 .F/ 1 has a symplectic linear model. The
representation  3=2 ÌSp4 .F/ 1 has the trivial representation as a subquotient, and
the trivial representation obviously has a symplectic linear model.
Proof. Assume  is a ramified character. Then, using the Bernstein center decompo-
sition, we get that in that case the epimorphism S.SL2 .F// ! 1 ˝ 1 can be extended
to  s Ì1jSL2 .F/SL2 .F/ , since it is non-zero anyway only on the Bernstein component
in which 1 ˝ 1; as a representation of SL2 .F/  SL2 .F/; lies. This component is
different from the component in which  sC1=2 Ì 1 ˝  sC1=2 Ì 1 lies. This means
that  s Ì 1 has a non-zero linear symplectic model.
Assume that  D 1 and s D 12 ; so that  sC1=2 Ì 1 ˝  sC1=2 Ì 1 has 1 ˝ 1 as a
trivial quotient. Then obviously,  s Ì 1jSL2 .F/SL2 .F/ has a trivial quotient.
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 103

Assume that  is unramified, but  sC1=2 ¤  ˙1 : For a smooth representation 

of SL2 .F/  SL2 .F/ we denote by rV1 ;V2 ./ the Jacquet module of  with respect
to the upper-triangular unipotent subgroup of the first and, then, of the second copy
of SL2 .F/ (we can view it as rV1 .rV2 .///: We get then a smooth GL1 .F/  GL1 .F/
module. We apply this Jacquet functor, which is exact, on the exact sequence (3).
Since S.SL2 .F//  1SL2 .F/  1SL2 .F/ , we have rV1 ;V2 .S.SL2 .F//   1 ˝  1 ;
so that  1 ˝  1 is a subquotient of rV1 ;V2 . s Ì 1jSL2 .F/SL2 .F/ /: Projectivity of
cuspidal representations (Lemma 26. of Bernstein 1992) gives us an epimorphism

rV1 ;V2 . s Ì 1jSL2 .F/SL2 .F/ /   1 ˝  1 :

Actually, Lemma 26 of Bernstein (1992) requires finite-length representations, but

we can just apply it to the representation  s Ì 1=W; where W is a subspace
of S.SL2 .F// such that S.SL2 .F//=W Š 1SL2 .F/  1SL2 .F/ ; i.e., we apply it on
rV1 ;V2 . s Ì 1=W/ which is then clearly of finite length (as a GL1 .F/  GL1 .F/-
module). Frobenius reciprocity then gives rise to an SL2 .F/  GL1 .F/-intertwining
operator, say T;

T W rV2 . s Ì 1/ !  1 Ì 1 ˝  1 :

If the image of this operator is  1 Ì1˝ 1 ; then, there would exist an epimorphism

T1 W rV2 . s Ì 1/  StSL2 .F/ ˝ 1 :

If T1 jrV2 .S.SL2 .F// D 0; this would give us an epimorphism

rV2 . sC1=2 Ì 1 ˝  sC1=2 Ì 1/ D  sC1=2 Ì 1 ˝ rV2 . sC1=2 Ì 1/  StSL2 .F/ ˝ 1 ;

which is impossible, by the requirement  sC1=2 ¤  1 : So, we conclude that

T1 jrV2 .S.SL2 .F// ¤ 0; so that we have an epimorphism

rV2 .S.SL2 .F//  StSL2 .F/ ˝ 1 :

By the Frobenius reciprocity, this would give us a non-zero intertwining map

S.SL2 .F// ! StSL2 .F/ ˝ 1 Ì 1:

The image of this intertwiner is StSL2 .F/ ˝ 1 Ì 1 or StSL2 .F/ ˝1SL2 .F/ : Note that the
maximal isotypic quotient of StSL2 .F/ in S.SL2 .F// is ‚.StSL2 .F/ / D StSL2 .F/ ; so we
would have an epimorphism StSL2 .F/   1 Ì 1 or StSL2 .F/  1SL2 .F/ ; which is
impossible in both of the cases.
We conclude that we have an epimorphism rV2 . s Ì 1/ ! 1SL2 .F/ ˝  1 :
Now we continue analogously: the Frobenius isomorphism gives us a non-zero
SL2 .F/  SL2 .F/-intertwining operator, say T2 ,  s Ì 1 ! 1SL2 .F/ ˝  1 Ì 1:
104 A. David et al.

If the image of this intertwiner were to be 1SL2 .F/ ˝  1 Ì 1, we would have an

epimorphism, say T3 ; from  s Ì1 to 1SL2 .F/ ˝StSL2 .F/ : If T3 restricted to S.SL2 .F//
is zero, this would give an epimorphism

 sC1=2 Ì 1 ˝  sC1=2 Ì 1  1SL2 .F/ ˝ StSL2 .F/ ;

which is impossible by our choice of  s : So, T3 restricted to S.SL2 .F// is non-

zero, but this is impossible with the form of the isotypic component, recalled above.
Thus, the image of T is 1SL2 .F/ ˝ 1SL2 .F/ ; which is what we wanted. t
u
Using Propositions 4.2 and 4.4, we conclude
Proposition 4.5. For any irreducible subquotient  of  s ÌSO4 .F/ 1 having a
generalized Shalika model, its (“small”) theta-lift to Sp4 .F/ is non-zero and has
a linear symplectic model.
Proof. Assume 2 ¤ 1. Then, the representations  s ÌSO4 .F/ 1 and  s ÌSp4 .F/ 1
are irreducible. From Theorem 2.4 we get that n. s ÌSO4 .F/ 1/ D 2: Moreover, in
the same way as in Proposition 3.3, we get that . s ÌSO4 .F/ 1; 2/ D  s ÌSp4 .F/ 1
and then apply Propositions 4.2 and 4.4. Assume that 2 D 1: Then, if  ¤ 1
then for s ¤ ˙ 12 the representations  s ÌSO4 .F/ 1 and  s ÌSp4 .F/ 1 are both
irreducible and the conclusion follows as previously. For s D 12 ; we know that
1
L. 2 StGL2 .F/ ÌSO4 .F/ 1/ is a subquotient of  s ÌSO4 .F/ 1 and has a non-zero
generalized Shalika model, and in Proposition 3.3 we have already proved that
1 1
.L. 2 StGL2 .F/ ÌSO4 .F/ 1// D L. 2 StGL2 .F/ ÌSp4 .F/ 1/. The other subquotients of
 s ÌSO4 .F/ 1 do not have the generalized Shalika models (Proposition 4.2). We have
1 1
also proved that .L. 2 StGL2 .F/ ÌSO4 .F/ 1// D L. 2 StGL2 .F/ ÌSp4 .F/ 1/ (the case of
 D 1) in (Proposition 3.3), so the conclusion is the same. Note that in the case
s D 32 the small theta lift of  3=2 ÌSO4 .F/ 1 is the trivial representation of Sp4 .F/
(cf. Theorem 5.1 (ii) of Kudla 1996), and all the cases are covered. t
u

5 Final Case and Proof of the Main Theorem

5.1 The Case of Irreducible Principal Series

In this section, we consider the case where  is an irreducible principal series

of GL2 .F/ with trivial central character. The representation  is of the form
 s  1  s , with  a unitary character and s in R. The irreducibility condition for
 is: .2 ; jsj/ ¤ .1; 12 /.
Proposition 5.1. Let  be a unitary character of GL2 .F/ and s in R, with
.2 ; jsj/ ¤ .1; 12 /, and let  be the representation  s  1  s .
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 105

1  
Then, the representation  2 ÌSO4 .F/ 1 has  s ÌSO4 .F/ 1 as unique irreducible
quotient. Its theta lift to Sp4 .F/ is:
(i)  s ÌSp4 .F/ 1 if .; s/ ¤ .1; ˙ 32 /;
(ii) 1Sp4 .F/ if .; s/ D .1; ˙ 32 /.
This theta lift has a symplectic model.
Proof. We first note that the map sending f to s 7! f .s/ gives an isomorphism:

1 1
 
. 2 ÌSO4 .F/ 1/ D  sC 2  1  sC 2 ÌSO4 .F/ 1

1 1
 1 1
Š  sC 2   s 2 ÌSO4 .F/ 1 D  s . 2    2 / ÌSO4 .F/ 1:

The last representation has a unique irreducible quotient, namely  s ÌSO4 .F/ 1.
1
 the representation  ÌSO4 .F/ 1 has a unique irreducible quotient,
Thiss implies that
2

 ÌSO4 .F/ 1 .
 
Since  s ÌSO4 .F/ 1 and  s ÌSO4 .F/ 1 are non-isomorphic, these two rep-
resentations have the same theta lift to Sp4 .F/ (Lemma 2.3 and Relation (2)).
Proposition 4.5 and its proof then give the desired result. t
u

5.2 Main Theorem

Theorem 5.2. Let  be an irreducible smooth admissible representation of SO4 .F/

with a generalized Shalika model. Then ./ is non-zero and has a symplectic linear
model.
Proof. By Theorem 1.2 the representation  is a quotient of  1=2 ÌSO4 .F/ 1 for an
irreducible admissible representation  of GL2 .F/. If  is supercuspidal, everything
is known by Theorem 1.1. So we may assume that  is not supercuspidal.
Lemma 2.2 combined with Lemma 3.2, Propositions 4.2 and 5.1 implies that 
must then be one of the representations in the first column of the following table.

 ./
L.StGL2 .F/  1=2 ÌSO4 .F/ 1/ L.StGL2 .F/  1=2 ÌSp4 .F/ 1/
 s ÌSO4 .F/ 1, s.t. .2 ; s/ ¤ .1; ˙ 12 /
 s ÌSp4 .F/ 1
and .; s/ ¤ .1; ˙ 32 /
 ˙3=2 ÌSO4 .F/ 1 1Sp4 .F/
 s 
 ÌSO4 .F/ 1 , s.t. .2 ; jsj/ ¤ .1; 1=2/  s ÌSp4 .F/ 1

Note that in the first column, the first three entries indeed all have a generalized
Shalika model. The last entry of the first column might have a generalized Shalika
model. Note furthermore that all entries in the second column are non-zero. We
106 A. David et al.

have shown in Lemma 3.1 and Proposition 4.4 that all the representations in the
second column have a symplectic linear model. Finally by Propositions 3.3 and 4.5
a representation in the second column is indeed the theta lift of the representation in
the same line in the first column, which completes the proof. t
u

Acknowledgements This project started at the WIN-Europe conference in October 2013. We

would like to thank the organizers of the conference and the CIRM in Luminy for providing such
excellent working conditions. We are grateful to the referee for several helpful comments. MH has
been supported in part by the Croatian Science Foundation under the project 9364. JL would like
to thank Imperial College London for providing financial support in form of a Doris Chen Mobility
Award.

References

Arthur, J.: The Endoscopic Classification of Representations. American Mathematical Society Col-
loquium Publications, vol. 61. American Mathematical Society, Providence (2013). Orthogonal
and symplectic groups
Ban, D.: Parabolic induction and Jacquet modules of representations of O.2n; F/. Glas. Mat. Ser.
III 34(54), 147–185 (1999)
Bernstein, I.N., Zelevinsky, A.V.: Induced representations of reductive p-adic groups I. Ann. Sci.
École Norm. Sup. 10(4), 441–472 (1977)
J. Bernstein, Draft of: Representations of p-adic groups, preprint, 1992, available at http://www.
math.tau.ac.il/~bernstei
Cogdell, J.W., Piatetski-Shapiro, I.I., Shahidi, F.: Functoriality for the quasisplit classical groups.
In: On Certain L-Functions. Clay Mathematics Institute, vol. 13, pp. 117–140. American
Mathematical Society, Providence (2011)
Gan, W.T., Ichino, A.: Formal degrees and local theta correspondence. Invent. Math. 195, 509–672
(2014)
Gan, W.T., Savin, G.: Representations of metaplectic groups I: epsilon dichotomy and local
Langlands correspondence. Compos. Math. 148, 1655–1694 (2012)
Gan, W.T., Takeda, S.: The local Langlands conjecture for Sp(4). Int. Math. Res. Not. IMRN 15,
2987–3038 (2010a)
Gan, W.T., Takeda, S.: On Shalika periods and a theorem of Jacquet-Martin. Am. J. Math. 132,
475–528 (2010b)
Gan, W.T., Takeda, S.: The local Langlands conjecture for GSp.4/. Ann. Math. 173(2), 1841–1882
(2011)
Ginzburg, D., Rallis, S., Soudry, D.: On a correspondence between cuspidal representations of
GL2n and Sep2n . J. Am. Math. Soc. 12, 849–907 (1999)
Ginzburg, D., Rallis, S., Soudry, D.: Generic automorphic forms on SO.2n C 1/: functorial lift to
GL.2n/, endoscopy, and base change. Int. Math. Res. Not. 14, 729–764 (2001)
Jacquet, H., Rallis, S.: Uniqueness of linear periods. Compos. Math. 102, 65–123 (1996)
Jiang, D., Qin, Y.: Residues of Eisenstein series and generalized Shalika models for SO4n .
J. Ramanujan Math. Soc. 22, 101–133 (2007)
Jiang, D., Soudry, D.: Appendix: On the local descent from GL.n/ to classical groups [appendix
to mr2931222]. Am. J. Math. 134, 767–772 (2012)
Jiang, D., Nien, C., Qin, Y.: Symplectic supercuspidal representations and related problems. Sci.
China Math. 53, 533–546 (2010a)
Jiang, D., Nien, C., Qin, Y.: Symplectic supercuspidal representations of GL.2n/ over p-adic fields.
Pac. J. Math. 245, 273–313 (2010b)
The Conjectural Relation Between Generalized Shalika Models on SO4n .F/. . . 107

Jiang, D., Nien, C., Qin, Y.: Generalized Shalika models of p-adic SO4n and functoriality. Israel
J. Math. 195, 135–169 (2013)
Kudla, S.S.: Notes on the local theta correspondence (lectures at the European School in Group
Theory), preprint, 1996, available at http://www.math.utoronto.ca/~skudla/castle.pdf
Kudla, S.: Notes on the local theta correspondence (lectures at the European School in Group
Theory) (1996)
Labesse, J.-P., Langlands, R.P.: L-indistinguishability for SL.2/. Can. J. Math. 31, 726–785 (1979)
Mœglin, C., Vignéras, M.-F., Waldspurger, J.-L.: Correspondances de Howe sur un corps p-adique.
Lecture Notes in Mathematics, vol. 1291. Springer, Berlin (1987)
Sally, Jr., P.J., Tadić, M.: Induced representations and classifications for GSp.2; F/ and Sp.2; F/.
Mém. Soc. Math. France (N.S.) 52, 75–133 (1993)
Sun, B., Zhu, C.-B.: Conservation relations for local theta correspondence. http://arxiv.org/pdf/
1204.2969v2.pdf (2012)
Tadić, M.: On reducibility of parabolic induction. Israel J. Math. 107, 29–91 (1998)
Zassenhaus, H.: On the spinor norm. Arch. Math. 13, 434–451 (1962)
Bad Reduction of Genus Three Curves
with Complex Multiplication

Irene Bouw, Jenny Cooley, Kristin Lauter, Elisa Lorenzo García,

Michelle Manes, Rachel Newton, and Ekin Ozman

Abstract Let C be a smooth, absolutely irreducible genus 3 curve over a number

field M. Suppose that the Jacobian of C has complex multiplication by a sextic CM-
field K. Suppose further that K contains no imaginary quadratic subfield. We give a
bound on the primes p of M such that the stable reduction of C at p contains three
irreducible components of genus 1.

MSC 2010: 11G15, 14K22, 15B33

1 Introduction

In Goren and Lauter (2007), Goren and Lauter study genus 2 curves whose
Jacobians are absolutely simple and have complex multiplication (CM) by the ring
of integers OK of a quartic CM-field K, and they show that if such a curve has
bad reduction to characteristic p then there is a solution to the embedding problem,
formulated as follows Goren and Lauter (2007):

I. Bouw
Institute of Pure Mathematics, University of Ulm, Ulm, Germany
J. Cooley
Mathematics Institute, University of Warwick, Coventry, UK
K. Lauter ()
Microsoft Research, Redmond, WA, USA
e-mail: klauter@microsoft.com
E. Lorenzo García
Mathematisch Instituut, Universiteit Leiden, Leiden, The Netherlands
M. Manes
Department of Mathematics, University of Hawaii, Honolulu, HI, USA
R. Newton
Max Planck Institute for Mathematics, Bonn, Germany
E. Ozman
Mathematics Department, University of Texas at Austin, Austin, TX, USA
Department of Mathematics, Bogazici University, Istanbul, Turkey

© Springer International Publishing Switzerland 2015 109

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_5
110 I. Bouw et al.

Let K be a quartic CM-field which does not contain a proper CM-subfield, and let
p be a prime. The embedding problem concerns finding a ring embedding  W OK ,!
End.E1  E2 /, such that the Rosati involution coming from the product polarization
induces complex conjugation on OK , and E1 ; E2 are supersingular elliptic curves
over Fp .
In this paper, we consider genus 3 curves whose Jacobians have CM by a sextic
CM-field that does not contain a proper CM-subfield. By analogy with Goren and
Lauter (2007), we formulate an embedding problem for the genus 3 case as follows.
Problem 6.3 (The Embedding Problem). Let O be an order in a sextic CM-field
K, and let p be a prime number. The embedding problem for O and p is the problem
of finding elliptic curves E1 ; E2 ; E3 defined over Fp , and a ring embedding

i W O ,! End.E1  E2  E3 /

such that the Rosati involution on End.E1  E2  E3 / induces complex conjugation

on O. We call such a ring embedding a solution to the embedding problem for O
and p.
In this paper, we prove the following result on solutions to the embedding
problem. We refer to Section 6.3 for the precise statement.
Theorem 6.9. Let K be a sextic CM-field such that K does not contain a proper
CM-subfield. Let O be an order in K. There exists an explicit bound on the rational
primes p for which the embedding problem has a solution, and this bound depends
only on the order O.
As in the genus 2 case, Theorem 6.9 yields a bound on certain primes of bad
reduction for the curve C. However, the result is not as strong as in the genus 2 case,
since there are more possibilities for the reduction of C. We discuss the statement
of the result.
Let C be a smooth, absolutely irreducible genus 3 curve over a number field M
whose Jacobian has CM by an order O in a sextic CM-field K. We say that C has
bad reduction at a rational prime p if there exists a prime p of M above p at which
C has bad reduction. In Corollary 4.3, we observe that if C has bad reduction at
a prime p, there are two possibilities for the stable reduction Cp of C at p. Either
Cp contains three irreducible components of genus 1 or Cp contains one irreducible
component of genus 1 and one of genus 2.
In this paper, we restrict our attention to the first of these two possibilities. In
Proposition 6.5, we show that if C has bad reduction at a prime p above p and the
stable reduction contains three genus 1 curves, then the embedding problem for O
and p has a solution. Theorem 6.9 therefore yields the following result on the primes
of bad reduction of C.
Bad Reduction of Genus Three Curves with Complex Multiplication 111

Theorem 6.8. Let C be a genus 3 curve whose Jacobian has CM by an order O

in a sextic CM-field K that does not contain a proper CM-subfield. There exists an
explicit bound on the primes p where the stable reduction contains three irreducible
components of genus 1.
We do not consider all primes of bad reduction of C in Theorem 6.8 for
the following reason. If the stable reduction of C at p contains three irreducible
components of genus 1, then the reduction J p of the Jacobian J of C is isomorphic
to the product E1  E2  E3 of elliptic curves as polarized abelian varieties
(Proposition 4.2). This yields a ring embedding

 W O D End.J/ ,! End.J p / D End.E1  E2  E3 /;

which has the property that the Rosati involution on End.E1  E2  E3 / restricts to
complex conjugation on the image of O (Section 4.3). This is precisely the statement
that  is a solution to the embedding problem for O and p.
Consider a prime p where the curve C has bad reduction, but the stable reduction
Cp contains an irreducible component E of genus 1 and an irreducible component
D of genus 2 (Corollary 4.3). In this case — an example of which is described
in Section 5.2 — the reduction J p of the Jacobian of C is the product of E with
the Jacobian of D as polarized abelian varieties. The abelian variety J p is still
isogenous to a product of elliptic curves (Theorem 4.5), but J p is not isomorphic to a
product of elliptic curves as polarized abelian varieties. This suggests that a different
formulation of the embedding problem would be needed to draw conclusions for
such primes p. We do not discuss the correct formulation of the embedding problem
for this case in the present paper, but leave it as a direction for future work.
The assumption that the CM-field K does not contain a proper CM-field is also
present in the genus 2 case in Goren and Lauter (2007). However, in the genus
2 case, this assumption is equivalent to the assumption that the CM-type of the
Jacobian J is primitive. We refer to Section 3.4 for more details. In characteristic
zero, the condition that the CM-type corresponding to J is primitive is equivalent to
the assumption that J is absolutely simple (Theorem 3.2).
In the genus 3 case, the assumption that the CM-field K does not contain a proper
CM-subfield still implies that the CM-type of the Jacobian J is primitive. However,
the converse does not hold. Even in the case that the sextic CM-field K contains
a proper CM-subfield there exist primitive CM-types (Section 3). In Section 6.4,
we discuss why the embedding problem needs to be formulated differently for such
CM-fields. We show that, in the case where K contains a proper CM-subfield, the
embedding problem as we have formulated it has solutions for any prime p and some
order O of K.
Finally, we have not included the condition that the elliptic curves Ei are
supersingular in the formulation of the embedding problem, in contrast to the
formulation in genus 2, because for a set of Dirichlet density 1=2, the elliptic curves
Ei are ordinary.
112 I. Bouw et al.

1.1 Relation to a Result of Gross and Zagier

One of the motivations of Goren and Lauter for studying solutions of the embedding
problem in genus 2 was generalizing a result of Gross and Zagier on singular moduli
of elliptic curves in Gross and Zagier (1985). Recall that singular moduli are values
j./ of the modular function j at imaginary quadratic numbers . Gross and Zagier
define the product
0 14=w1 w2
Y
J.d1 ; d2 / D @ .j.1 /  j.2 //A ;
Œ1 ;Œ2 

where the product runs over equivalence classes of imaginary quadratic numbers
i with discriminants di , where the di are assumed to be relatively prime. Here
wi denotes the number of units in Q.i /. The function J is closely related to the
value of the Hilbert class polynomial of an imaginary quadratic field at a point 
corresponding to a different imaginary quadratic field.
Under some assumptions, Gross and Zagier show that J.d1 ; d2 / is an integer, and
their main result gives a formula for the factorization of this integer. The result of
Gross and Zagier may be reinterpreted as a formula for the number of isomorphisms
between the reductions of the elliptic curves Ei corresponding to the i at all rational
primes p. This problem is equivalent to counting embeddings of End.E2 / into the
endomorphism ring of the reduction of E1 at p.
Goren and Lauter (2007, Corollary 5.1.3) prove a generalization of the result
of Gross and Zagier. They consider curves of genus 2 with CM by a quartic CM-
field. In their result, the function j is replaced by suitable Siegel modular functions
f =‚k . Here f is a Siegel modular form of weight 10k with values in a number field
and ‚ is a concrete Siegel modular form of weight 10. The modular function f =‚k
has the property that for any  in the Siegel upper half plane the genus 2 curve
corresponding to  has bad reduction at the primes dividing the denominator of
.f =‚k /./. (See Goren and Lauter 2007, Corollary 5.1.2 for the precise statement.)
The Igusa class polynomials are an analog of the Hilbert class polynomials for
quartic CM-fields, where the j-invariant is replaced by the absolute Igusa invariants.
Goren and Lauter and collaborators (see, for example, Goren and Lauter 2007, 2013;
Lauter and Viray 2012) deduce results on the denominators of the coefficients of the
Igusa class polynomials from results on the embedding problem for quartic CM-
fields. A different approach to generalize the results of Gross and Zagier using
arithmetic intersection theory can be found in the work of Bruinier, Kudla, Yang,
and collaborators (see, for example, Bruinier and Yang 2006).
The embedding problem for curves of genus 3 studied in this paper does not
immediately yield a statement analogous to that of Gross and Zagier. One of the
ingredients that is missing is finding good coordinates for the moduli space of curves
of genus 3, analogous to the absolute Igusa invariants in genus 2.
Bad Reduction of Genus Three Curves with Complex Multiplication 113

In this paper, we discuss several differences between the reduction of CM-

curves in genus 2 and in genus 3. The embedding problem in the formulation of
Problem 6.3 does not cover all types of bad reduction. Also, in the case that the
sextic CM-field K contains a proper CM-subfield the embedding problem should be
adapted. It would be interesting to study the implication of these differences for a
possible analog of the Igusa class polynomials for sextic CM-fields.

1.2 Outline

The structure of this paper is as follows. Section 2 gives the possibilities for the
Galois group of the Galois closure of a sextic CM-field, following work of Dodson
(Dodson 1984). Section 3 describes the possible CM-types for a sextic CM-field.
We note which of the CM-types are primitive, meaning that they can arise as the
CM-type of a simple abelian variety. In Section 4, we describe the possibilities for
the reduction of a genus 3 curve and its Jacobian to characteristic p > 0. We also
give some properties of the Rosati involution attached to a polarized abelian variety,
which will be used in Section 6. In Section 5, we give various examples of genus 3
curves with CM; we calculate their CM-types and the reductions of the curves and
their Jacobians to characteristic p > 0. In Section 6, we consider a genus 3 curve C
over a number field M such that its Jacobian has CM by a sextic CM-field K with no
proper CM-subfield. We prove a bound on primes such that there exists a solution to
the embedding problem, and we use that to give a bound on the primes p such that
the stable reduction of C at p contains three elliptic curves. We show that if we drop
the assumption that K has no proper CM-subfield, then the embedding problem as
stated cannot be used to give a bound on the primes p as above.
We include as an appendix a collection of conditions that a solution to the
embedding problem must satisfy, written as equations in the entries of certain
matrices in the image of the embedding. These equations may be useful for future
work. A refinement of the embedding problem (for example, a version which
includes conditions pertaining to the CM-type) would result in extra equations in
addition to those in the appendix. It is to be hoped that studying this larger set
of equations would yield an explicit bound on the primes for which they have a
solution. This would give a bound on the primes p such that the stable reduction
of C at p contains three curves of genus 1, even in the case where the CM-field K
contains a proper CM-subfield.

1.3 Notation and Conventions

We set the following notation, to be used throughout.

• Fp is the finite field with p elements.
• N is a primitive Nth root of unity.
114 I. Bouw et al.

• For a field k, k is an algebraic closure.

• K is a sextic CM-field, i.e., K is a totally imaginary quadratic extension of K C ,
where K C is a totally real cubic extension of Q.
• O is an order of K.
• F and L are Galois closures of K=Q and K C =Q, respectively, with G D Gal.F=Q/
and GC D Gal.L=Q/.
• is a complex embedding K ,! C, and  is complex conjugation. Hence f ;  ı
g is a conjugate pair of embeddings.
• .K; '/ is a CM-type, i.e., a choice of one embedding from each pair of complex
conjugate embeddings.
• A is an abelian variety, End.A/ is the endomorphism ring of A, and End0 .A/ is
End.A/ ˝ Q.
• For f 2 End.A/, f _ 2 End.A_ / is the dual isogeny. The Rosati involution
associated with a fixed polarization is denoted by f 7! f  ; End0 .A/ ! End0 .A/.
• E is an elliptic curve, j.E/ is the j-invariant of E.
• We denote an isomorphism between two abelian varieties over an algebraic
closure of the field of definition by '.
• We denote an isogeny between two abelian varieties over an algebraic closure of
the field of definition by
.
• M is a number field,  (or p) is a finite place of M, O is the valuation ring of ,
and k is the residue field.
• C is a curve over a number field with Jacobian J and genus g D g.C/. A curve
C is always assumed to be smooth, projective, and absolutely irreducible, unless
explicitly mentioned otherwise.
• Bp;1 is the quaternion algebra ramified at p and 1, and R is a maximal order of
Bp;1 .
• For a matrix T, Tr.T/ denotes the sum of its diagonal entries, the trace.
• TrK=K1 denotes the trace of a field extension K=K1 .
• For an element of a central simple algebra, Nrd denotes the reduced norm.
• NmK=K1 denotes the norm of a field extension K=K1 ; we use Nm when the
extension is clear.

2 The Galois Group of the Galois Closure

of a Sextic CM-Field

Let K be a sextic CM-field, i.e., K is a totally imaginary quadratic extension of a

totally real field K C with ŒK C W Q D 3. We denote the Galois closure of K C =Q by L
and the Galois closure of K=Q by F. We write G D Gal.F=Q/ and GC D Gal.L=Q/.
The following proposition lists the possibilities for G.
Bad Reduction of Genus Three Curves with Complex Multiplication 115

Proposition 2.1. Let K be a sextic CM-field, and let G be the Galois group of the
Galois closure of K=Q. Then G is one of the following groups:
(1) C2  C3 ' C6 ,
(2) C2  S3 ' D12 ,
(3) .C2 /3 Ì GC with GC 2 fC3 ; S3 g acting by permutations on the three copies
of C2 .
In particular, if K=Q is Galois, then the Galois group G D Gal.K=Q/ ' C6 is
cyclic.
Proof. This is proved in Section 5.1.1 of Dodson (1984), for example. t
u
In the rest of this section, we sketch the proof of Proposition 2.1, following
Dodson. Since we restrict to the case of sextic CM-fields, the presentation can be
simplified. In the course of the proof, we also give more details on the structure
of the extensions F=Q and K C =Q in the different cases. In particular, we show
that Case 3 is precisely the case where K does not contain an imaginary quadratic
subfield.
Galois theory implies that we have the following exact sequence of groups:

1 ! Gal.F=L/ ! G ! GC ! 1:

Lemma 2.2. We have

Gal.F=L/ ' .C2 /v ; 1v3

and

GC 2 fC3 ; S3 g:

Proof. This lemma is a special case of the proposition in Section 1.1 of Dodson
(1984). We give the proof here for convenience.
p
We first remark that K D K C . ı/ for some totally positive square-free
ı 2 K C . We write ı1 WD ı; ı2 ; : : : ; ır for the GC -conjugates of ı. It follows that
p p
F D L. ı1 ; : : : ; ır /:
p p
Every element h 2 Gal.F=L/ sends ıi to ˙ ıi . Moreover, h is determined
by its action on these elements. It follows that Gal.F=L/ ' .C2 /v is an elementary
abelian 2-group.
Since ı 2 K C it follows that ŒQ.ı/ W Q divides 3. We conclude that the number
of GC -conjugates of ı is at most 3.
The statement on GC immediately follows from the fact that ŒK C W Q D 3. This
proves the lemma. t
u
116 I. Bouw et al.

F = LK1

K
L
2 2

K+

K1
3
2

Fig. 1 Field extensions in Case 2

Proof of Proposition 2.1. We start the classification. Note that Gal.K=K C / is gener-
ated by complex conjugation. It follows that complex conjugation is also an element
of G. This element, which we denote by , is an element of the center of G.
Case I: K=Q Galois.
Since K=Q is Galois, G D Gal.K=Q/ is a group of order 6, hence either cyclic
or S3 . Since the Galois closure L of K C =Q is a totally real subfield of K, it
follows that K C D L. This implies that Gal.K=K C / is a normal subgroup of G
which has order 2. It follows that G ' C6 is cyclic. Note that K contains the
imaginary quadratic subfield K1 WD K C3 and K D K1 K C . This corresponds to
Case 1 of Proposition 2.1.
Case II: K=Q is not Galois and K contains an imaginary quadratic field K1 .
Since K contains an imaginary quadratic field K1 , we have F D LK1 and G '
C2  GC . If GC ' C3 , then L D K C and K=Q is Galois, which contradicts
our assumption. It follows that GC ' S3 and G ' C2  S3 . This is Case 2 of
Proposition 2.1. We obtain the field diagram in Figure 1.
Case III: K=Q is not Galois and K does not contain an imaginary quadratic
subfield.
This case corresponds to Case 3 of Proposition 2.1. In this case the integer v
from Lemma 2.2 is not equal to 1, i.e., we have v D 2 or 3. The following claim
completes the proof of Proposition 2.1.
Claim. The case v D 2 does not occur. This claim is a special case of the
second proposition in Section 5.1.1 of Dodson (1984). We give the proof here for
completeness.
Recall that  2 Gal.F=L/ denotes complex conjugation and is contained in the
center of G. Let  2 GC be an element of order 3. Then  acts on Gal.F=L/ D .C2 /v
by conjugation. This action has two orbits of length 1, corresponding to the identity
element and . All other orbits have length 3. It follows that 3 j .2v  2/. The claim
follows. t
u
Bad Reduction of Genus Three Curves with Complex Multiplication 117

23
K

L
2

K+

Fig. 2 Field extensions in Case 3

Of primary interest to us in the rest of this paper is Case 3 of Proposition 2.1,

in which K does not contain an imaginary quadratic subfield. We have seen that
G ' .C2 /3 Ì GC with GC 2 fC3 ; S3 g. Figure 2 describes the field extensions in
Case 3.

3 Primitive CM-Types

Let K be a sextic CM-field. As in Section 2, we write K C for the totally real

cubic subfield of K. The complex embeddings K ,! C come in pairs f ;  ı g,
where  denotes complex conjugation. Recall that a CM-type .K; '/ is a choice of
one embedding from each of these pairs. The goal of this section is to determine
the primitive CM-types. We start by recalling the definition from Milne (2006),
Section 1.1. For examples we refer to Section 5.
Definition 3.1. Let .K; '/ and .K1 ; '1 / be CM-types. We say that .K; '/ is induced
from .K1 ; '1 / if K1 is a subfield of K and the restriction of ' to K1 coincides with '1 .
A CM-type is called primitive if it is not induced from a CM-type on any proper
CM-subfield of K.
Let A be an abelian variety and let K be a CM-field with ŒK W Q D 2 dim.A/.
We say that A has complex multiplication (CM) by K if the endomorphism algebra
End0 .A/ D End.A/ ˝ Q contains K. We say that a curve C has CM by K if its
Jacobian has CM by K. We say that A (or C) has CM if there exists a CM-field K
such that A (or C) has CM by K. If End.A/ is an order O in a CM-field K with
ŒK W Q D 2 dim.A/, we say that A has CM by O.
The following theorem gives a geometric interpretation of what it means for
the CM-type of a CM-abelian variety to be primitive in characteristic zero. For
convenience, we say that an abelian variety A defined over a field M is simple if it
is absolutely simple, meaning that A ˝M M is not isogenous to a product of abelian
118 I. Bouw et al.

varieties of lower dimension. Similarly, we say that two abelian varieties A1 ; A2

defined over M are isogenous if there exists an isogeny ' W A1 ! A2 defined over
the algebraic closure of M.
Theorem 3.2. Let A be an abelian variety defined over a field of characteristic zero.
Suppose that A has CM with CM-type .K; '/. Then the CM-type .K; '/ is primitive
if and only if the abelian variety A is simple.
Proof. This is proved in Theorem 3.5 of Chapter 1 of Lang (1983). See also
Remark 1.5.4.2 of Chai et al. (2014). t
u
We refer to Section 1.5.5 of Chai et al. (2014) for an explanation of why we need
to assume that A is defined over a field of characteristic zero in Theorem 3.2.
The following result gives a useful criterion for determining whether a given
CM-type is primitive. For a proof, we refer to Theorem 3.6 of Chapter 1 of Lang
(1983). For a CM-type .K; '/ and h 2 Aut.K/, we write

'h D f'i ı h j 'i 2 'g:

Proposition 3.3. Let .K; '/ be a CM-type. We write .F; ˆ/ for the induced CM-
type of the Galois closure of K=Q. Let

Hˆ D fh 2 G D Gal.F=Q/ j ˆh D ˆg:

Then .K; '/ is primitive if and only if

K D F Hˆ :

We now determine the primitive sextic CM-types in each of the cases of

Proposition 2.1. We first consider Case 3. Recall that in the proof of Proposition 2.1
we showed that Case 3 is precisely the case where K does not contain an imaginary
quadratic subfield.
Corollary 3.4. Suppose that we are in Case 3 of Proposition 2.1, i.e., K does not
contain an imaginary quadratic field. Then every CM-type .K; '/ is primitive.
Proof. Suppose that .K; '/ is not primitive. Then K contains a proper CM-subfield
K1 . Since K is sextic, K1 is an imaginary quadratic field. This yields a contradiction.
t
u

3.1 Primitive Types in Case 1

We now consider Case 1 from Proposition 2.1. This is the case in which K=Q is
Galois, with Galois group G ' C6 . We choose a generator  of G. Note that
Bad Reduction of Genus Three Curves with Complex Multiplication 119

complex conjugation corresponds to  3 . Up to replacing ' by its complex conjugate,

every CM-type .K; '/ may be written as

'a;b D f1;  a ;  b g; 0 < a; b < 6; a 1 .mod 3/; b 2 .mod 3/:

We find 4 cases:

fa; bg 2 ff1; 2g; f1; 5g; f4; 2g; f4; 5gg:

Note that changing the generator  of G to  1 changes f4; 5g to f1; 2g, therefore
we do not have to consider the choice f4; 5g.
We write Ha;b for the subgroup fixing the CM-type as in Proposition 3.3. Then
H1;2 D H1;5 D f1g and H4;2 D h 2 i ' C3 . Note that K1 WD K H4;2 is the imaginary
quadratic subfield of K, which is a CM-field. We conclude that '4;2 is induced from
K1 , and hence imprimitive. The other CM-types are primitive.

3.2 Primitive Types in Case 2

We now consider Case 2 from Proposition 2.1. We refer to Section 2 for a description
of the fields involved. Recall that K D K1 K C . Therefore, an embedding W K ,! C
corresponds to an ordered pair . 1 ; C /, where 1 W K1 ,! C is an embedding of
K1 and C W K C ,! C is an embedding of K C . Since K C is totally real, the image
of C is contained in R. We denote the three possible complex embeddings of K C
by i for i D 1; 2; 3. We fix a complex embedding of K1 and denote it by 1. We
denote the other complex embedding of K1 by 1.
A CM-type .K; '/ consists of a triple of these ordered pairs in which no two
of the pairs are complex conjugates. Since Gal.K1 =Q/ is generated by complex
conjugation, we simply choose one of the two complex embeddings of K1 for each
embedding i of K C . This means that we may write

' D f.i ; i / j i D 1; 2; 3g; i 2 f˙1g:

Identifying ' with its complex conjugate yields four different CM-types.
We determine the imprimitive types. The only CM-field properly contained in
K is the imaginary quadratic field K1 . The restriction of the embedding .i ; i / to
K1 is just i . Therefore, the CM-type ' D f.i ; i /g is imprimitive if and only if i
is independent of i. We conclude that there is a unique imprimitive CM-type. The
other three are primitive.
120 I. Bouw et al.

3.3 Examples of CM-Types

We give examples of CM-types illustrating each of the three cases of Pro-

position 2.1.
Example 3.5 (K=Q is Galois with Galois Group G ' C6 ). Let K be Q.7 / where
7 is a primitive seventh root of unity. The maximal totally real subfield of K is
K C D Q.7 C 71 /, which has degree three over Q (the minimal polynomial of
7 C 71 over Q is x3 C x2  2x  1). The field K is a totally imaginary quadratic
extension of K C .
5
The automorphism  which p maps 7 to 7 generates Gal.K=Q/. The fixed field of
2 4 2
h i is Q.7 C7 C7 / D Q. 7/. This is the unique imaginary quadratic extension
of Q contained in Q.7 /. Therefore, the only imprimitive CM-type admitted by K is
'2;4 D f1;  2 ;  4 g; the CM-types 'a;b D f1;  a ;  b g for fa; bg ¤ f4; 2g with a 1
.mod 3/; b 2 .mod 3/ are all primitive.
The following examples have been taken from the database of Klüners and Malle
(2011).
Example 3.6 (The Galois Closure of K=Q is D12 ). Let K be the sextic field obtained
by adjoining a root of the irreducible polynomial f .x/ D x6 3x5 Cx4 C10x2 9xC3.
Then K is a totally imaginary quadratic extension of the totally real cubic field
K C D Q.˛/ where the minimal polynomial of ˛ is g.x/ D x3  7x2 C 12x  3.
The Galois closure F of K=Q is the compositum of the Galois closure of K C with
the unique imaginary quadratic subfield K1 of K, given by the minimal polynomial
x2 C 3x C 3. The Galois group of F is isomorphic to S3  C2 ' D12 . Denote the
roots of g.x/ by ˛1 WD ˛; ˛2 ; ˛3 .
Let i W ˛1 7! ˛i denote the three real embeddings of K C and ˙1
denote the two complex embeddings of K1 . Then the CM-type ' D f.1; 1 /;
.1; 2 /; .1; 3 /g of K is imprimitive, since its restriction to the quadratic
imaginary subfield K1 is also a CM-type. The remaining three CM-types
of K are primitive. For clarity, the primitive CM-types are as follows:
f.1; 1 /; .1; 2 /; .1; 3 /g; f.1; 1 /; .1; 2 /; .1; 3 /g; f.1; 1 /; .1; 2 /; .1; 3 /g.
Example 3.7 (The Galois Closure of K=Q is .C2 /3 Ì C3 ). Let K D Q.ˇ/ be the
degree 6 extension of Q where the minimal polynomial of ˇ is f .x/ D x6  2x5 C
5x4  7x3 C 10x2  8x C 8. Let F be the Galois closure of K. Then Gal.F=Q/
is .C2 /3 Ì C3 . Moreover, K is a CM-field since K is a totally imaginary quadratic
extension of K C D Q.˛/ where the minimal polynomial of ˛ over Q is g.x/ D
x3  7x2 C 14x  7. Note that K contains no quadratic subfield, hence every CM-
type is primitive.
Bad Reduction of Genus Three Curves with Complex Multiplication 121

3.4 Comparison with the Genus 2 Case

The following proposition characterizes primitive CM-types for quartic CM-fields.

Proposition 3.8. Let K be a quartic CM-field. The following are equivalent.
(1) The CM-type is primitive.
(2) The CM-field K does not contain an imaginary quadratic subfield.
Proof. We recall the argument from Example 8.4.(2) of Shimura (1998) in which
we find a classification of the possible Galois groups of quartic CM-fields K
together with the possible CM-types. It follows from this classification that if K
contains a proper CM-subfield K1 ¤ Q then K=Q is Galois with Galois group
G ' C2 C2 . Moreover, in this case all CM-types are imprimitive. Namely, denoting
again complex conjugation by , we may write G D f1; ; ; g. Then the possible
CM-types are f1; g and f1; g, which are fixed by hi and hi, respectively.
Therefore, the statement follows from Proposition 3.3. t
u
Proposition 3.8 explains why Goren and Lauter (2006) and Goren and Lauter
(2007) restrict to the case where the quartic CM-field does not contain an imaginary
quadratic subfield. For quartic CM-fields, this is equivalent to requiring that the CM-
type is primitive. However, as we have seen in our discussion of the primitive types
in Cases 1 and 2 of Proposition 2.1, these two properties are not equivalent for sextic
CM-fields.
We give two concrete examples of genus 2 curves with CM to illustrate
Proposition 3.8. These are similar to the genus 3 examples given in Section 5.1.
We consider two smooth projective curves defined by the following affine equations

D1 W y5 D x.x  1/;
D2 W y8 D x.x  1/4 :

One easily verifies that both curves have genus 2.

The curve D1 has CM by K1 WD Q.5 / with CM-type .1; 2/ in the notation
of Section 5.1. The Galois group of K1 =Q is cyclic of order 4, hence its unique
subgroup of order 2 is generated by complex conjugation, which cannot fix the CM-
type. Indeed, the Jacobian of D1 is simple. In the genus 2 case, all CM-types of
a cyclic CM-field are primitive. We have already seen that this does not hold in
general for genus g  3.
The curve D2 has CM by K2 WD Q.8 /. The corresponding Galois group is
isomorphic to C2  C2 , hence the CM-type is imprimitive. Indeed, the CM-type
is .1; 3/ which is fixed by h3i  .Z=8Z/ : The CM-type .1; 3/ is induced from the
CM-type of the elliptic curve E WD D2 =hi, where .x; y/ D .1=x; y3 =x.x  1// is
an automorphism of order 4.
122 I. Bouw et al.

4 Reduction of CM-Curves and Their Jacobians

Our main result (Theorem 6.8) deals with curves C of genus 3 defined over some
number field whose Jacobians have CM by a sextic CM-field K. In this section, we
describe the possibilities for the reduction of these curves and their Jacobians to
characteristic p > 0.

4.1 The Theorem of Serre–Tate

Let C be a curve of genus g  2 defined over a number field M, and let J WD Jac.C/
be its Jacobian. In the course of our arguments, we allow ourselves to replace M
by a finite extension, which we still denote by M. Let  be a finite place of M. We
write O for the valuation ring of  and k for its residue field. We write k for an
algebraic closure of k .
Recall that the abelian variety J has good reduction at  if there exists an abelian
scheme J over O with J ˝O M ' J. This implies that the reduction J WD
J ˝O k is an abelian variety. We say that J has potentially good reduction at  if
there exists a finite extension M 0 =M and an extension  0 of  such that J ˝M M 0 has
good reduction at  0 .
The following theorem is Theorem 6 of Serre and Tate (1968).
Theorem 4.1 (Serre–Tate). Let J be an abelian variety with CM defined over a
number field M. Let  be a finite place of M. Then J has potentially good reduction
at .
Since there are at most finitely places where J does not have good reduction,
there exists a finite extension of M over which J has good reduction everywhere.

4.2 Reduction of Genus 3 Curves with CM

We now describe the restrictions imposed by Theorem 4.1 on the reduction of the
curve C.
Recall that C is a curve of genus g.C/  2 defined over a number field M. We
say that C has good reduction at a finite place  of M if there exists a model C over
O with C ˝O M ' C such that the reduction C WD C ˝O k is smooth. Similarly,
C has potentially good reduction at  if it has good reduction over a finite extension
of M.
We say that C has semistable reduction at  if there exists a model C over O with
C ˝O M ' C such that the reduction C is semistable. This means that C is reduced
and has at most ordinary double points as singularities. The corresponding model
C D C is called a semistable model of C at . The Stable Reduction Theorem
Bad Reduction of Genus Three Curves with Complex Multiplication 123

(Deligne and Mumford 1969, Corollary 2.7) states that every curve C admits a
semistable model at  after replacing M by a finite extension. Since we assume
that g.C/  2, there exists a unique minimal semistable model, which is called the
stable model at . Its special fiber C is called the stable reduction of C at . The
minimality of the stable model implies that C has potentially good reduction if and
only if the stable reduction C is smooth. If the finite place  is fixed, we usually
omit it.
We now turn to our situation of interest, namely that of a genus 3 curve whose
Jacobian has CM by a sextic CM-field. The following proposition is a consequence
of Theorem 4.1.
We say that C has bad reduction at  if it does not have potentially good reduction
at . This is equivalent to the stable reduction C having singularities. We say that the
reduction CN of C is tree-like if the intersection graph of the irreducible components
of C is a tree. Note that we always consider the reduction J (resp. C) as an abelian
variety (resp. curve) defined over the algebraically closed field k for convenience.
Proposition 4.2. Let C be a curve of genus 3 defined over a number field M such
that its Jacobian J D Jac.C/ has CM. Let  be place of M where C has bad
reduction. Then
(a) the stable reduction C of C is tree-like, and
(b) the reduction J of J is the product of the Jacobians of the irreducible
components of C (as polarized abelian varieties).
Proof. Let  be a finite place of M. After replacing M by a finite extension and
choosing an extension of , we may assume that C has stable reduction at . Let C
be the stable model of C. Set S D Spec.O /, and define Pic0 .C=S/ to be the identity
component of the Picard variety. Since the stable reduction C of C is reduced,
Theorem 1 in Section 9.5 of Bosch et al. (1990) states that Pic0 .C=S/ is a Néron
model of J.
Theorem 4.1 implies that J has potentially good reduction, i.e., there exists an
abelian variety J over S with generic fiber J. Proposition 8 of Section 1.2 in Bosch
et al. (1990) shows that J =S is a Néron model. Since two different Néron models
are canonically isomorphic, it follows that Pic0 .C=S/ 'S J . In particular, it follows
that the special fiber Pic0 .C=S/ ˝O k ' Pic0 .C/ is an abelian variety.
Example 8 of Section 9.2 in Bosch et al. (1990) shows that Pic0 .C/ is given by
an exact sequence
Y
1 ! T ! Pic0 .C/ ! B WD Jac.CQ i / ! 1; (4.1)
i

where B is an abelian variety and T is a torus. The product on the right-hand side
is taken over the irreducible components of C. We denote the normalization of an
irreducible component Ci of C by CQ i . The torus T satisfies

T ' Gtm;k

124 I. Bouw et al.

for some t  0. The torus Gm is not compact, and hence not an abelian variety.
Since Pic0 .C/ is an abelian variety, the exact sequence (4.1) implies that t D 0, i.e.,
Pic0 .C/ contains no torus. By Corollary 12.b of Bosch et al. (1990), this means that
the intersection graph of the irreducible components of C is a tree. Both statements
of the proposition follow from this. t
u
The corollary below follows immediately from Proposition 4.2. In Section 5, we
give examples of each of the cases.
Corollary 4.3. Let C be a genus 3 curve with CM defined over a number field M,
and let  be a finite place of M. One of the following three possibilities holds for the
irreducible components of C of positive genus:
(i) (good reduction) C is a smooth curve of genus 3,
(ii) C has three irreducible components of genus 1,
(iii) C has an irreducible component of genus 1 and one of genus 2.
Note that the stable reduction C may contain irreducible components of genus 0.
This happens for the stable reduction C1 to characteristic 3 of the curve C1 from
Lemma 5.3, for example. One may show that C1 has four irreducible components:
one of genus 0 and three of genus 1. The three elliptic curves each intersect the
genus 0 curve in one point but do not intersect each other. Since the irreducible
components of genus 0 do not contribute to the Jacobian, we have not listed them in
Corollary 4.3.
Remark 4.4. Let C be a curve of genus 3 with CM, defined over a number field
M. Suppose that C has bad reduction at a finite place  of M. In Case (ii) of
Corollary 4.3, the reduction C of C contains three irreducible components Ei of
genus 1. Proposition 4.2 implies that

J ' E1  E2  E3

as polarized abelian varieties, i.e., the polarization on J is the product polarization.

In Case (iii) of Corollary 4.3, C contains an irreducible component E of genus 1
and an irreducible component D of genus 2. In this case, we have

J ' E  Jac.D/

and the polarization on J is the product polarization induced by the principal

polarizations on the components. We show below that in this case J is still isogenous
to a product of elliptic curves (Theorem 4.5). However, it is not true that the
polarization of J is induced by polarization on the three elliptic curves as we had in
Case (ii).
Even in the case where C has good reduction (Case (i) of Corollary 4.3), the
reduction J of the Jacobian need not be simple even if J is. In this case, the
polarization of J is induced by the embedding of C in its Jacobian and hence is
not a product polarization.
Bad Reduction of Genus Three Curves with Complex Multiplication 125

The following theorem is a generalization of Theorem 3.2 to positive

characteristic.
Theorem 4.5. Let J be an abelian variety of dimension 3 with CM, defined over a
number field M. Suppose that the reduction J of J at a finite place of M is not simple.
Then J is isogenous to the product of three copies of the same elliptic curve E.
Proof. The result is essentially a special case of Theorem 1.3.1.1 of Chai et al.
(2014). For the convenience of the reader we sketch a direct proof in our situation.
By assumption, J has CM by the sextic CM-field K. This implies that we have
an embedding

K ,! End0 .J/:
Q
Decompose J into isotypic components: J
i Ani i where the Ai are simple and
Ai 6
Aj for i ¤ j. Since J is not simple by assumption, for dimension reasons
0
Q exists 0j such that Aj D E is an elliptic curve. We have K ,! End .J/ D
there
i Mni .End .Ai //. Projecting this ring homomorphism on the jth factor gives an
injection K ,! Mnj .End0 .E//. A dimension argument shows that nj D 3 and
therefore J
E3 . t
u
Proposition 4.6. Let C be a genus 3 curve with CM, defined over a number field M.
Suppose that C has bad reduction at a finite place  of M. Then the reduction J of
the Jacobian J of C is supersingular or K contains an imaginary quadratic field K1 .
Proof. Let C and J be as in the statement of the proposition. Since C has bad
reduction at , Corollary 4.3 shows that C has an irreducible component E1 of
genus 1. It follows that we may regard E1 as abelian subvariety of J. (This is slightly
weaker than the statement in Remark 4.4.) In particular, J is not simple. Theorem 4.5
implies therefore that J is isogenous to the product of three copies of an elliptic
curve E. Note that J is supersingular if and only if E is.
We assume that E is ordinary. Since E may be defined over a finite field, it has
CM and K1 WD End0 .E/ is an imaginary quadratic field contained in the center of
End0 .E3 / D M3 .K1 /. Since J is isogenous to E3 , we obtain an embedding

K D End0 .J/ ,! End0 .J/ ' End0 .E3 / D M3 .K1 /:

Theorem 1.3.1.1 of Chai et al. (2014) states that K is its own centralizer in M3 .K1 /.
Since the center of M3 .K1 / is K1 , we conclude that K1 is contained in K and the
result follows. t
u
The following corollary summarizes the results so far in the case that the CM-
field K does not contain an imaginary quadratic subfield K1 .
126 I. Bouw et al.

Corollary 4.7. Let C be a genus 3 curve with CM by K, defined over a number

field M. Suppose that K does not contain an imaginary quadratic subfield. Then the
following holds:
(a) the CM-type .K; '/ of J is primitive, and J is absolutely simple,
(b) if C has bad reduction at a finite place , then the reduction of J at  is
supersingular.
Proof. Part (a) follows from Corollary 3.4 and Theorem 3.2. Part (b) follows from
Proposition 4.6. t
u

4.3 Polarizations and the Rosati Involution

In the rest of this section, we recall some results on the Rosati involution following
Sections 20 and 21 of Mumford (1970) and Section 17 of Milne (2008). For precise
definitions and more details, we refer to these sources. Let A be an abelian variety
and  W A ! A_ be a polarization associated with an ample line bundle L on A.
The polarization  is an isogeny and therefore has an inverse deg1  _ D 1 2
Hom.A_ ; A/ ˝Z Q.
The Rosati involution on End0 .A/ D End.A/ ˝ Q is defined by

f 7! f  D 1 ı f _ ı :

It satisfies

.f C g/ D f  C g ; .fg/ D g f  ; a D a

for f ; g 2 End0 .A/ and a 2 Q. In the case where  is a principal polarization, i.e.,
deg./ D 1, the Rosati involution acts as an involution on End.A/. This is because
1 is in Hom.A_ ; A/ and not just in Hom.A_ ; A/ ˝Z Q. The natural polarization
on a Jacobian is a principal polarization.
The Rosati involution is a positive involution Theorem 1 of Section 21 in
Mumford (1970). This means that

.f ; g/ 7! Tr.f  g /; End0 .A/ ! Q

defines a positive definite quadratic form on End0 .A/. We refer to Section 21 of

Mumford (1970) for the precise definition of the trace. In the case that A D E is an
elliptic curve, we choose the polarization  defined as

 W E ! Pic0 .E/; P 7! ŒP  ŒO:

The corresponding Rosati involution sends an isogeny f to its dual isogeny f _ and
Tr.f  f _ / is deg.f /, the degree of the endomorphism f .
Bad Reduction of Genus Three Curves with Complex Multiplication 127

Proposition 4.8. Let A be a simple abelian variety defined over a field of

characteristic zero with principal polarization . Assume that A has CM by a
field K. Then the Rosati involution associated with  induces complex conjugation
on the CM-field K.
Proof. Since A is simple, the endomorphism algebra End0 .A/ equals K and the
proposition is proved, for example, in Lemma 1.3.5.4 of Chai et al. (2014). t
u
Remark 4.9. Let A be a simple abelian variety with End0 .A/ D K as in the
statement of Proposition 4.8. Let M be a number field over which A can be defined,
and let p be a prime of M at which A has good reduction. Write A for the reduction.
We obtain an embedding

K ,! End0 .A/:

The Rosati involution on End0 .A/ is an extension of the Rosati involution on

End0 .A/ D K, which is complex conjugation by Proposition 4.8.
The following proposition was used in the proofs of Goren and Lauter (2007) but
not stated there explicitly.
Proposition 4.10. Suppose that A D En is a product of elliptic curves as polarized
abelian varieties. Then the Rosati involution acts as

Mn .End.E// ! Mn .End.E//; .fi;j / 7! .fj;i_ /:

Proof. The result is well known to the experts. We sketch the argument. The proof
we present here is a variant of the proof of Proposition 11.28 (ii) of van der Geer
and Moonen (2011).
Let A D En be as in the statement of the lemma, and write pi W A ! E for
the projection on the ith coordinate. Then any line bundle L on A satisfies L D
p1 L1 ˝    ˝ pn Ln for suitable line bundles Li on E.
Consider the natural map A_ D Pic0 .A/ ! .Pic0 .E//n D .E_ /n which sends
a line bundle L 2 Pic0 .A/ to the n-tuple .LjEi /i 2 .Pic0 .E//n of the restrictions
of L to the ith copy Ei WD .   ; 0; E; 0;    / of E. One shows that this map is an
isomorphism. The product polarization A W A ! A_ D .E_ /n is induced by the
natural polarization  W E ! Pic0 .E/ on E. In particular, it is also a principal
polarization.
Using this identification, it suffices to prove the proposition in the case that f 2
End.A/ corresponds to an n n matrix with an endomorphism ˛ 2 End.E/ as .j; i/th
component and zeros everywhere else. The endomorphism f _ W A_ ! A_ induced
by f sends a line bundle L on A to pi .˛  Li /. We conclude that f  W A ! A D En
corresponds to the matrix with the dual isogeny ˛ _ in the .i; j/th coordinate and
zeros elsewhere. This proves the proposition. t
u
128 I. Bouw et al.

5 Examples

In this section we discuss some examples of genus 3 curves with CM.

5.1 Cyclic Covers

The first type of examples we consider are N-cyclic covers of the projective line
branched at exactly three points, see also Sections 1.6 and 1.7 of Chapter 1 of Lang
(1983). More precisely, let C be a smooth projective curve defined over a field of
characteristic zero which admits a Galois cover  W C ! P1 whose Galois group is
cyclic of order N such that  is branched exactly at three points. We may assume
the three branch points to be 0; 1; 1 2 P1 .
Kummer theory implies the existence of integers 0 < a1 ; a2 < N with
gcd.N; a1 ; a2 / D 1 such that the extension of function fields corresponding to  is

Q.x/  Q.x/Œy=.yN  xa1 .x  1/a2 /:

The Galois group of  is generated by ˛.x; y/ D .x; N y/, where N is a primitive

Nth root of unity.
Define 0 < a3 < N by a1 C a2 C a3 0 .mod N/. Then a chart at 1 may be
given by

wN D za3 .z  1/a2 ;

where z D 1=x. The condition that  is branched at 1 is therefore equivalent to

a3 .a1 C a2 / 6 0 .mod N/. The Riemann–Hurwitz formula shows that

3
X
2g.C/  2 D 2N C .N  gcd.N; ai //:
iD1

In Lemma 5.1 below, we show that the endomorphism ring of Jac.C/ contains
Q.N /. Therefore Jac.C/ has CM by Q.N / if and only if 2g.C/ D '.N/, where '
denotes Euler’s totient function. For example, this condition is satisfied if N is an
odd prime. This case is discussed by Lang (Section 1.7 of Chapter 1 of Lang 1983).
The condition 2g.C/ D '.N/ is satisfied for exactly three curves Ci , up to
isomorphism over C. Kummer theory implies that two tuples .N; a1 ; a2 ; a3 / and
.M; b1 ; b2 ; b3 / define isomorphic curves if and only if N D M and there exists an
integer c with gcd.c; N/ D 1 and a permutation  2 S3 such that bi ca .i/
.mod N/ for all i. This is similar to the argument in Section 1.7 of Chapter 1 of
Lang (1983).
Bad Reduction of Genus Three Curves with Complex Multiplication 129

The three curves satisfying this property are:

C1 W y9 D x.x  1/3 ;
C2 W y7 D x.x  1/2 ;
C3 W y7 D x.x  1/:

An alternative equation for C1 is

y3 D z4  z; where z3 D x: (5.1)

We put KNi D Q.Ni / and GNi D .Z=Ni Z/ . In the three cases we consider in
Lemma 5.1, we have GNi ' C6 . For j 2 .Z=Ni Z/ , we denote the corresponding
element of Gal.KNi =Q/ by

j
j W Ni 7! Ni ;

or also by j when no confusion can arise.

The following lemma summarizes the properties of the curves Ci .
Lemma 5.1. (a) The curve C1 has CM by Q.9 /. The CM-type is .1; 2; 4/. This
type is primitive.
(b) The curve C2 has CM by Q.7 / and CM-type .1; 2; 4/. This type is imprimitive.
(c) The curve C3 has CM by Q.7 / and CM-type .1; 2; 3/. This type is primitive.
Proof. It is easy to check that the automorphism ˛ of Ci has a fixed point. Using this
point to embed the curve Ci in its Jacobian, we see that ˛ induces an endomorphism
˛ 2 End.Jac.Ci // of multiplicative order Ni .
We may regard ˛ 2 End.Jac.Ci // as a primitive Ni th root of unity. In all three
cases, we have 2g.Ci / D 6 D '.Ni / D ŒQ.Ni / W Q. It follows that Ci has CM
by KNi .
To calculate the CM-type of Ci we follow the strategy of Section 1.7 of Chapter 1
of Lang (1983), and identify the cohomology group H 0 .Ci ; / of holomorphic
differentials with the tangent space of Jac.C/. It suffices to find a basis of H 0 .Ci ; /
consisting of eigenvectors of ˛  , the map induced by ˛ on H 0 .Ci ; /. Such a basis
is computed in Section 1.7 of Chapter 1 of Lang (1983). The statement on the CM-
type easily follows from this. (The fact that the action of h˛i on H 0 .Ci ; / does not
factor through the action of a quotient group provides a second proof that ˛ defines
an endomorphism of order Ni of Jac.Ci /.)
We explain what happens for C1 . We use a slightly different notation from
Theorem 1.7.1 of Chapter 1 of Lang (1983). A basis of H 0 .C1 ; / is given by

y dx y2 dx y4 dx
!1 D ; !2 D ; !4 D :
x.x  1/ x.x  1/ x.x  1/2
130 I. Bouw et al.

Note that ˛  !i D 9i !i . The statement on the CM-type of Jac.C1 / follows.

Primitivity is shown in Section 3.1.
In Example 3.5 we have determined all primitive CM-types for Q.7 /. The
statements on the (im)primitivity of the CM-types of C2 and C3 follow from this.
t
u
Remark 5.2. Lemma 5.1.(b) implies that Jac.C2 / is not simple. We may also check
this directly. The curve C2 admits an automorphism
 
1 y2
ˇ.x; y/ D ; :
1x 1x

The curve E WD p C2 =hˇi has genus 1. This curve has CM by the field K1 D
Q.7 /h2 i D Q. 7/.
One checks that h˛; ˇi ' Z=7ZÌZ=3Z is a non-abelian group. Using the method
of Kani and Rosen (1989) or Paulhus (2008), one may also deduce from this that

Jac.C2 /
E3 :

Our next goal is to describe the reduction behavior of the curves C1 and C3 .
Lemma 5.3. (a) The curve C1 has bad reduction at p D 3 and good reduction at
all other primes.
(b) The reduction J 1;p of the Jacobian J1 of C1 to characteristic p is ordinary if
and only if p 1 .mod 9/ and supersingular if and only if p D 3 or p 2
.mod 3/.
(c) If p 4; 7 .mod 9/, then the abelian variety J 1;p is simple.
Proof. It is easy to see that C1 has good reduction to characteristic p ¤ 3.
Indeed, (5.1) still defines a smooth projective curve in characteristic p ¤ 3. We
consider the reduction at p D 3. In this case, the extension of function fields

F3 .z/  F3 .z/Œy=.y3  z.z3  1//

defines a purely inseparable field extension. This implies that F3 .z/Œy=.y3 z.z3 1//
is the function field of a curve of genus 0. This does not imply that C1 has bad
reduction to characteristic 3, since there could be a different model.
We claim that there does not exist a curve of genus 3 in characteristic 3 with
an automorphism of order 9. This claim implies that C has bad reduction to
characteristic 3. Indeed, if C has potentially good reduction, then the automorphism
group Aut.C/ of the reduction C of C contains Aut.C/. Hence, in particular, Aut.C/
contains an automorphism of order 9.
To obtain a contradiction, we assume that X is a curve of genus 3 in characteristic
3 with an automorphism  of order 9. We consider the Galois cover

X ! X=h i:
Bad Reduction of Genus Three Curves with Complex Multiplication 131

This cover is wildly ramified of order 9 above at least one point. We apply the
Riemann–Hurwitz formula to this cover. It follows from Theorem 1.1 of Obus and
Pries (2010) that the contribution of a wild ramification point with ramification
index 9 to 2g.X/  2 in the Riemann–Hurwitz formula is at least 2  .9  1/ C 5 
.3  1/ D 26, which contradicts the assumption that X has genus 3. This proves (a).
We have shown that C1 has bad reduction to characteristic 3. Let C1;3 be
the stable reduction of C1 to characteristic 3. Then C1;3 contains at least 2
irreducible components of positive genus (Corollary 4.3). Furthermore, there is an
automorphism of order 9 acting on C1;3 . The only way this is possible is if C1;3
contains three irreducible components of positive genus, which are then elliptic
curves, each with an automorphism of order 3. The automorphism of order 9
permutes these components. There is a unique elliptic curve with an automorphism
of order 3, namely the elliptic curve with j D 0. In characteristic 3 this curve may
be given by

w3  w D v 2 : (5.2)

This curve is supersingular by the Deuring–Shafarevich formula (Crew 1984).

We conclude that the reduction J 1;3 of the Jacobian of C1 to characteristic 3
is supersingular. Proposition 4.2.(b) implies that J 1;3 is in fact superspecial: the
Jacobian J 1;3 is isomorphic to three copies of the supersingular elliptic curve (5.2)
as a polarized abelian variety.
The rest of (b) may be deduced from Yui (1980). For p 4; 7 .mod 9/, Yui’s
results (Yui 1980) imply that J is neither ordinary nor supersingular. In fact, her
results imply that J has p-rank zero, but is not supersingular. Theorem 4.5 therefore
implies that J is simple. t
u
The situation for C3 is similar but somewhat easier.
Lemma 5.4. (a) The curve C3 has good reduction at p ¤ 7 and potentially good
reduction at p D 7.
(b) The reduction J 3;p of the Jacobian J3 of C3 to characteristic p is ordinary if and
only if p 1 .mod 7/ and supersingular if and only if p D 7 or p 1; 3; 5
.mod 7/.
Proof. The fact that C3 has good reduction to characteristic p ¤ 7 follows as in the
proof of Lemma 5.3. The curve C3 has potentially good reduction to characteristic
7 as well, see Example 3.8 of Bouw and Wewers (2012). The curve C3 does not
have good reduction over Q7 but acquires good reduction over the extension Q7 .7 /
of Q7 .
Statement (b) for p ¤ 7 follows from Yui (1980). We consider the reduction C3;7
of C to characteristic 7. In characteristic 7, the reduction C3;7 is given by

w7  w D v 2 ;
132 I. Bouw et al.

by Example 3.8 of Bouw and Wewers (2012). By the Deuring–Shafarevich formula,

it follows that the Jacobian J 3;7 of C3;7 has p-rank 0. To show that it is supersingular,
it suffices to find an elliptic quotient of the curve C3;7 .
The curve C3;7 admits an extra automorphism of order 3 given by

ˇ.v; w/ D .3 v; 32 w/;

where 3 2 F 7 is an element of order three. The automorphism ˇ has exactly two

fixed points, namely the points with w D 0; 1. It follows that E3;7 WD C3;7 =hˇi is
an elliptic curve. This shows that J 3;7 is supersingular. t
u

5.2 A Picard Curve Example

We end this section by considering Example 3 from Section 5 of Koike and Weng
(2005), wherein Koike and Weng study Picard curves with CM. We show that the
curve in the aforementioned example has bad reduction to characteristic p D 5, and
that the stable reduction consists of an elliptic curve and a curve of genus 2. We will
show that the Jacobian has superspecial reduction in this case. This is an example
where the reduction J of the Jacobian is isomorphic to E3 , but the polarization is
neither that of a smooth curve nor the product polarization E  f0g  f0g C f0g 
E  f0g C f0g  f0g  E.
A Picard curve is a curve of genus 3 given by an equation

y3 D f .x/;

where f .x/ 2 CŒx is a polynomial of degree 4 with simple roots. Every Picard curve
admits an automorphism ˛.x; y/ D .x; 3 y/. Therefore, the endomorphism ring of
the Jacobian contains Q.3 /.
Let C4 be the smooth projective curve defined by

y3 D f .x/ WD x4  13  2  72  x2 C 23  13  5  47  x  52  31  132 :

Koike and Weng show that the Jacobian of C4 has CM by the field K D K C K1 with
K1 D Q.3 / and K C D QŒt=.t3  t2  4t  1/. The CM-field K is Galois over Q,
hence we are in Case 1 of Proposition 2.1. One may show that the corresponding
CM-type is primitive. For example, one may check using Bouw (2001) that the
reduction J 4;7 of the Jacobian J4 of C4 to characteristic 7 has p-rank 1, and hence
is neither ordinary nor supersingular. It follows from this that the Jacobian J4 is
simple. The primitivity of the CM-type follows from this, by Theorem 3.2.
We now consider the reduction of C4 . The discriminant of f is 212  56  134 which
shows that C4 has good reduction for p ¤ 2; 3; 5; 13. One may check that C4 also
has good reduction at p D 2; 13. We do not consider what happens for p D 3.
Bad Reduction of Genus Three Curves with Complex Multiplication 133

We determine the reduction at p D 5. Note that

f .x/ x2 .x C 2/.x  2/ D x4 C x2 .mod 5/: (5.3)

Therefore, the stable reduction of C4 contains an irreducible component D of genus

2 given by the equation

yN 3 D xN 2 .Nx2 C 1/: (5.4)

The reason that this curve has genus 2 rather than 3 is that the 3-cyclic cover .Nx; yN / 7!
xN has only 4 branch points in characteristic 5, and not 5 branch points as it had in
characteristic zero. It follows that the curve C4 has bad reduction to characteristic
5, and the reduction of C4 consists of the curve D of genus 2 intersecting with
an elliptic curve. (We do not actually have to compute the elliptic component to
conclude this.) The reduction J 4 of the Jacobian of C4 is therefore isogenous to
the product of an elliptic curve and the abelian surface Jac.D/. To determine the
reduction type of J 4 , we first consider the Jacobian Jac.D/ of the curve D given by
Equation (5.4).
One may show by computing the Hasse–Witt matrix of D that the Jacobian J.D/
is supersingular. This is a similar calculation to the one we did in Section 5.1.
However, since D has genus 2, it suffices to compute the p-rank. In fact, the
Hasse–Witt matrix is identically zero, which shows that J.D/ is superspecial, i.e.,
isomorphic to the product of two supersingular elliptic curves.
Alternatively, we may note that D has additional automorphisms given by
   
1 yN 1 yN
.Nx; yN / D .Nx; yN /; .Nx; yN / D  ; 2 ;  ı .Nx; yN / D ; :
xN xN xN xN 2

Note that  fixes the two points with xN D 0; 1 and  fixes the two points with
xN 2 D 1. The quotients C4 =hi and C4 =hi are elliptic curves, each with an
automorphism of order 3. In particular, these elliptic curves have j D 0. Since
p D 5 2 .mod 3/, they are supersingular. Theorem 4.5 implies that J is isogenous
to E03 , where E0 denotes the supersingular elliptic curve over F5 with j D 0.
Remark 5.5. The examples we discussed in this section all have the property that
the CM-field K contains a CM-subfield K1 with Q ¨ K1 ¨ K. In Section 6.4,
we will show that this implies that the embedding problem, which we formulate in
Section 6, has degenerate solutions for every prime. This explains why we exclude
this case in Theorem 6.8.
Remark 5.6. Let K be a sextic CM-field. It is known how to construct genus 3
curves C in characteristic 0 with CM by K (Shimura 1998, Sections 6.2 and 14.3).
We sketch the construction.
134 I. Bouw et al.

We fix a CM-type .K; '/. Let ıK=Q be the different. For any ideal a of OK we
consider the lattice '.a/ D .'1 .a/; '2 .a/; '3 .a//. Then

A WD C3 ='.a/

is an abelian variety with CM by .K; '/. Shimura (Theorem 3 of Section 6.2 in

Shimura 1998) shows that all CM abelian varieties occur in this way.
In Section 14.3 of Shimura (1998), Shimura also describes all Riemann forms
defining principal polarizations on A. Such a Riemann form exists if the following
two conditions are satisfied.
• The ideal ıK=Q aa D .a/ is principal.
• There exists a unit u 2 OK such that ua is totally imaginary and the imaginary
part of 'i .ua/ is negative for all i.
Every principally polarized abelian variety of dimension 3 is isomorphic to the
Jacobian of a (possibly singular) genus 3 curve C by Theorem 4 of Oort and Ueno
(1973). More precisely, Oort and Ueno show that the curve C is of compact type,
meaning that A is isomorphic to the product of the Jacobians of the irreducible
components of positive genus of C. (This notion is essentially the same as the notion
“tree-like” that we used in Section 4.2.) In our situation, the abelian variety A is
simple, and it follows that the curve C is smooth.

6 Embedding Problem

6.1 Formulation of the Embedding Problem

Let C be a genus 3 curve defined over some number field M. We assume that the
Jacobian J D Jac.C/ has CM by a sextic CM-field K. After replacing M by a finite
extension if necessary, we may assume that J has good reduction (Theorem 4.1) and
that C has stable reduction at all finite places of M.
In this section, we make the following important assumption.
Assumption 6.1. We assume that K does not contain an imaginary quadratic
subfield.
Recall that Assumption 6.1 implies that the CM-type of C is primitive (Corol-
lary 4.7). The reason for making this assumption is discussed in Section 6.4.
Let p be a finite prime of M where the curve C has bad reduction. We write
k for the algebraic closure of the residue field at p and let p denote the residue
characteristic. We want to bound these primes p. (See Theorem 6.8 for the precise
statement of our result.) Recall from Corollary 4.3 that there are two possibilities for
the reduction C of C. In this section, we only deal with the case where C has three
irreducible components of genus 1 and postpone the other case for future work. To
summarize, we make the following assumption on the prime p.
Bad Reduction of Genus Three Curves with Complex Multiplication 135

Assumption 6.2. Let p be a finite prime of M, such that the stable reduction C D
Cp of C at p contains three elliptic curves as irreducible components (Case (ii) of
Corollary 4.3).
Let p be as in Assumption 6.2. We write E1 ; E2 ; E3 for the three elliptic curves
that are the irreducible components of C. We write J for the reduction of J at p.
Recall from Remark 4.4 that we have an isomorphism

J ' E1  E2  E3

as polarized abelian varieties, i.e., the polarization on J is the product polarization.

Corollary 4.7 implies that the Ei are supersingular. In particular, they are isogenous.
(This also follows from Theorem 4.5.)
Let End.J/ D O  OK . Reduction at the prime p gives an injective ring
homomorphism

O ,! End.J/ ' End.E1  E2  E3 /:

Problem 6.3 (The Embedding Problem). Let O be an order in a sextic CM-field

K, and let p be a prime number. The embedding problem for O and p is the problem
of finding elliptic curves E1 ; E2 ; E3 defined over a field of characteristic p, and a ring
embedding

i W O ,! End.E1  E2  E3 /

such that the Rosati involution on End.E1  E2  E3 / induces complex conjugation

on O. We call such a ring embedding a solution to the embedding problem for O
and p.
The following result states that if we have a solution to the embedding problem
then the elliptic curves Ei are automatically isogenous. The proof we give here works
directly with the abelian variety E1  E2  E3 without considering it as the reduction
of an abelian variety in characteristic zero. However the proof is essentially the same
as the proofs of Theorem 4.5 and Proposition 4.6.
Lemma 6.4. Let K be a sextic CM-field. Suppose that there exist elliptic curves
E1 ; E2 ; E3 defined over a field of characteristic p > 0 and an injective Q-algebra
homomorphism

i W K ,! End0 .E1  E2  E3 /:

Then the elliptic curves E1 , E2 , and E3 are all isogenous. Furthermore, if K contains
no imaginary quadratic subfield, then the Ei are supersingular.
136 I. Bouw et al.

Proof. First suppose that no two of the elliptic curves E1 , E2 , E3 are isogenous.
Then
0 1
End0 E1 0 0
i W K ,! End0 .E1  E2  E3 / D @ 0 End0 E2 0 A
0 0 End0 E3
D End0 E1  End0 E2  End0 E3 :

Projecting on the factor End0 Ei gives a ring homomorphism K ,! End0 Ei . Since

K is a field, this ring homomorphism must be injective. But End0 Ei is either an
imaginary quadratic field or a quaternion algebra, neither of which can contain a
sextic field.
Now suppose that exactly two of the elliptic curves are isogenous. Without loss
of generality, we may assume that E1
E2 and E1 6
E3 . Then
0 1
End0 E1 End0 E1 0
i W K ,! End0 .E1  E2  E3 / D @End0 E1 End0 E1 0 A
0 0 End0 E3
D M2 .End0 E1 /  End0 E3 :

Again, projecting on the factor End0 E3 , we see that K ,! End0 E3 . This is

impossible for dimension reasons. Thus, we have proved that all three elliptic curves
are isogenous.
Now suppose that K contains no imaginary quadratic subfield and that the elliptic
curves Ei are ordinary. Then End0 E1 D K1 for some imaginary quadratic field K1
and

i W K ,! End0 .E1  E2  E3 / D M3 .K1 /:

Let ˇ be a generator for K over Q and let f be its minimal polynomial, which
has degree 6. The matrix i.ˇ/ 2 M3 .K1 / has a minimal polynomial of degree at
most 3 over K1 . Since i is an injective Q-algebra homomorphism, this means that f
splits over K1 . Since K1 is quadratic, this implies that K1 ,! K, contradicting the
assumption that K contains no imaginary quadratic subfield. t
u
Proposition 6.5. Let C be a genus 3 curve such that O WD End.Jac.C// is an order
in a sextic CM-field K satisfying Assumption 6.1. Let M be a number field over which
C is defined, and let p be a prime of bad reduction of C such that Assumption 6.2
is satisfied. Write p for the residue characteristic of p. Then there exists a solution
to the embedding problem for O and p. Moreover, in this situation the three elliptic
curves are supersingular.
Proof. Let C be as in the statement of the proposition. Then the CM-type of its
Jacobian J is primitive (Corollary 4.7.(a)). Therefore the Rosati involution acts
Bad Reduction of Genus Three Curves with Complex Multiplication 137

as complex conjugation on End0 .J/ D K by Proposition 4.8. The canonical

polarization on the Jacobian J is a principal polarization, therefore the Rosati
involution also acts on End.J/ D O.
Assumption 6.2 implies that the reduction J of the Jacobian at p is isomorphic
to a product of three elliptic curves Ei as polarized abelian varieties. These elliptic
curves are supersingular (Corollary 4.7.(b)). Remark 4.9 shows that we obtain a
solution to the embedding problem. t
u

6.2 Endomorphisms of J as 3  3 Matrices

In this section we describe the ring End.E1  E2  E3 / from the embedding

problem (Problem 6.3). Recall that we may assume that the Ei are isogenous
(Lemma 6.4). We recall from Proposition 4.10 the description of the Rosati
involution corresponding to the product polarization on E1  E2  E3 .
We can view an element f 2 End.E1  E2  E3 / as a matrix
0 1
f1;1 f1;2 f1;3
f D @f2;1 f2;2 f2;3 A ;
f3;1 f3;2 f3;3

where fi;j 2 Hom.Ej ; Ei /. Given two endomorphisms f ; g the composition f ı g

corresponds to multiplication of matrices. Since the polarization on J D E1 E2 E3
is the product polarization, the Rosati involution f 7! f  sends f to
0 1
_ _ _
f1;1 f2;1 f3;1
B _ _ _C
@f1;2 f2;2 f3;2 A
_ _ _
f1;3 f2;3 f3;3

where fi;j_ denotes the dual isogeny of fi;j .

For i D 2; 3, let i W E1 ! Ei be an isogeny of degree ıi . The composition

.1; 2; 3/ .1;ı21 _ 1
2 ;ı3
_
3 /
E1  E1  E1 / E1  E2  E3 / E1  E1  E1

induces an injective Q-algebra homomorphism

End0 .E1  E2  E3 / ,! End0 .E1  E1  E1 / D M3 .End0 E1 /: (6.1)

Let ˆ denote the composite map

ˆ W K ,! End0 .E1  E2  E3 / ,! M3 .End0 E1 /:

138 I. Bouw et al.

It is easily seen that

0 1
1 0 0
@0 ı2 0 A ˆ.O/  M3 .End E1 /:
0 0 ı3

Under the assumptions made in Section 6.1, we may assume that the elliptic
curves Ei in the formulation of the embedding problem are supersingular (Proposi-
tion 6.5). We therefore recall some well-known facts on the endomorphism ring of
a supersingular elliptic curve.
Let p 2 Z>0 be the rational prime lying below p.
Proposition 6.6. Let E be a supersingular elliptic curve defined over a field of
characteristic p. Then End0 E is a quaternion algebra over Q ramified at precisely
the places fp; 1g. This quaternion algebra is non-canonically isomorphic to the
";p
algebra Bp;1 , where Bp;1 D . 1;1Q / if p D 2 and if p is odd, Bp;1 D . Q /
where
8
ˆ
<1 if p 3 .mod 4/;
ˆ
" D 2 if p 5 .mod 8/;
ˆ
:̂` if p 1 .mod 8/:

In the case that p 1 .mod 8/, ` 2 Z>0 is a prime such that ` 3 .mod 4/ and `
is not a square modulo p. Any isomorphism sends End E to an order of Bp;1 and the
involution given by taking the dual isogeny corresponds to the canonical involution
on Bp;1 .
Proof. The fact that the endomorphism algebra End0 .E/ of a supersingular elliptic
curve is a quaternion algebra over Q ramified precisely at fp; 1g is proved, for
example, in Section 21 of Mumford (1970). The statement on the Rosati involution
is also proved in loc. cit. The uniqueness of the quaternion algebra is proved, for
example, in Theorem III.3.1 of Vignéras (1980).
For p D 2, let Q D . 1;1
Q
/. For every odd prime p, let " be as in the statement
of the proposition and let Q D . ";p
Q / be the corresponding quaternion algebra.
The statement that Q is exactly ramified at the places fp; 1g follows easily from the
properties of the Hilbert symbol (page 37 of Vignéras 1980). t
u
For b 2 Bp;1 , we write Nrd.b/ D bb (where b represents the involution on
the quaternion algebra) for the reduced norm of b. The reduced norm corresponds
to the degree of an endomorphism under the identification in Proposition 6.6.
Lemma 6.7 (Elements of Small Norm Commute, Goren and Lauter (2007,
Corollary 2.1.2)). Let R be a maximal order of Bp;1 . If k1 ; k2 2 R and
p
Nrd.k1 /; Nrd.k2 / < p=2, then k1 k2 D k2 k1 .
Bad Reduction of Genus Three Curves with Complex Multiplication 139

6.3 Bounding the Primes of Bad Reduction for C

Recall that J D Jac.C/ is the Jacobian of a genus 3 curve C which has complex
multiplication by an order O in a sextic CM-field K which does not contain an
imaginary quadratic field (Assumption 6.1). Let K C denote the totally real cubic
subfield of K. The main result of this section is Theorem 6.8 which gives an upper
bound on the primes of bad reduction for C satisfying Assumption 6.2.
Theorem 6.8. Suppose that K does not contain an imaginary quadratic subfield.
Letpp j p be a prime of bad reduction for C satisfying Assumption
p 6.2. Write K D
Q. ˛/ for some totally negative element ˛ 2 K C n Z with ˛ 2 O D End.J/.
Then p  4 TrK C =Q .˛/6 =36 :
The existence of such ˛ is guaranteed because the sextic CM-field K contains
no imaginary quadratic subfield. By Proposition 6.5, the following result implies
Theorem 6.8.
Theorem 6.9. Suppose that K does not contain an imaginary quadratic subfield.
Let p be a prime such that there exists a solutionp to the embedding problem
p O of K. Write K D Q. ˛/ for some totally negative
(Problem 6.3) for some order
element ˛ 2 K C n Z with ˛ 2 O. Then p  4 TrK C =Q .˛/6 =36 :
We break down the proof of Theorem 6.9 into several lemmas.
Let
0 1
r s t
Q D @ u v wA
x y z
p
be the image of ˛ in End.E1  E2  E3 /. By Proposition 4.8, the Rosati involution
corresponds to complex conjugation on K, so we have
0 1 0 1
r _ u _ x_ r s t
@ s_ v _ y_ A D @ u v wA : (6.2)
t_ w_ z_ x y z

Lemma 6.10. We may assume that the homomorphisms s W E2 ! E1 and t W E3 !

E1 are both nonzero.
Proof. Suppose for contradiction that both s and t are zero. Then the image of ˛ in
End.E1  E2  E3 / is
0 1
rr_ 0 0
Q2 D @ 0 vv _  ww_ vw C wz A :
0 w_ v  zw_ w_ w  zz_
140 I. Bouw et al.

For i D 2; 3, let i W E1 ! Ei be an isogeny of degree ıi . As seen in (6.1), the i

induce an injective Q-algebra homomorphism End0 .E1  E2  E3 / ! End0 .E1 
E1  E1 / D M3 .End0 E1 / sending Q2 to
0 1
rr_ 0 0
SD@ 0 vv _  ww_ ı21 2_ .vw C wz/ 3
A:
0 ı31 3_ .w_ v  zw_ / 2 w_ w  zz_

Since .vw C wz/_ D w_ v  zw_ , the entries of S commute and therefore
form a subfield L of End0 E1 . Since S is the image of ˛ under an injective Q-
algebra homomorphism, the minimal polynomial of S over L divides the minimal
polynomial of ˛ over Q. Recall that rr_ 2 Z is the degree of r. Now rr_ is an
eigenvalue of S and therefore a root of its minimal polynomial. But this means that
the minimal polynomial of ˛ over Q has a root in Z, contradicting its irreducibility.
Therefore, at least one of s; t is nonzero. Using E2 in place of E1 , we see that at
least one of s; w is nonzero. Using E3 in place of E1 , we see that at least one of t; w
is nonzero. Putting all these conditions together and reordering the elliptic curves
E1 ; E2 ; E3 if necessary, we may assume that s and t are both nonzero. t
u
Henceforth, we assume that s and t are nonzero. Therefore, we can use s_ and t_
to give an injective Q-homomorphism
p End0 .E1  E2  E3 / ,! End0 .E1  E1  E1 /
as in (6.1). The image of ˛ in M3 .End0 E1 / is
0 1
r ı2 ı3
T D @1 svs_ =ı2 swt_ =ı2 A ; (6.3)
1 tw_ s_ =ı3 tzt_ =ı3

where ı2 D deg.s/ and ı3 D deg.t/.

Since K contains no imaginary quadratic subfield, Lemma 6.4 shows that the
elliptic curves E1 ; E2 , and E3 are supersingular. By Proposition 6.6, we may choose
an isomorphism End0 E1 ! Bp;1 . The isomorphism sends End E1 to a order of
Bp;1 and the Rosati involution on End E1 corresponds to the usual involution p on
Bp;1 . We abuse notation slightly by continuing to write T for the image of ˛ in
M3 .Bp;1 /.
Lemma 6.11. Supposep that K contains no imaginary quadratic subfield. Let T
denote the image of ˛ in M3 .Bp;1 /. Then the entries of the matrix T do not all
commute with each other.
Proof. Suppose for contradiction that the entries of T commute. Let K1 denote the
subfield of Bp;1 generated by the entries of T. A subfield of Bp;1 is either Q or
a quadratic subfield which splits Bp;1 . But Bp;1 is ramified at the infinite place,
so it is not split by any real field. Thus, K1 is either Q or an imaginary quadratic
field. By assumption,
p K contains no imaginary quadratic subfield. Thus, the minimal
polynomial of ˛ over Q remains irreducible over K1 .
Bad Reduction of Genus Three Curves with Complex Multiplication 141

p polynomial of T over K1 . The degree of g is at most 3.

Let g denote the minimal
Since T is the image of ˛ under
p an injective Q-algebra homomorphism, g divides
the minimal polynomial
p of ˛ over Q, which has degree 6. Thus, the minimal
polynomial of ˛ over Q factorizes over K1 , giving the required contradiction. u
t
We restrict to the case where p is odd; the case p D 2 is very similar. By
Proposition 6.6, Bp;1 has a Q-basis 1; i; j; k where i2 D ", j2 D p , ij D k,
ji D ij and " is as in Proposition 6.6. We embed Bp;1 into M4 .Q/ via
0 1 0 1 0 1 0 1
100 0 0 " 0 0 0 0 p 0 0 0 0 "p
B0 1 0 0C B1 0 0 0C B0 0 0 pC B0 0 p 0 C
1 7! B
@0 0 1
C ; i 7! B C ; j 7! B C ; k 7! B C:
0A @0 0 0 "A @1 0 0 0A @0 " 0 0 A
000 1 0 0 1 0 0 1 0 0 1 0 0 0

This induces an embedding M3 .Bp;1 / ,! M12 .Q/. Let U denote the image of
˛ in M12 .Q/. Write Tr.T 2 / for the sum of the elements on the diagonal of T 2 .
Define Tr.Q2 / in the same way. It is easily checked that Tr.T 2 / D Tr.Q2 /. By
the construction of the embedding Bp;1 ,! M4 .Q/, we have

Tr.U/ D 4 Tr.T 2 /: (6.4)

p
Lemma 6.12. Let T denote the image of ˛ in M3 .Bp;1 /. Then Tr.T 2 / D
TrK C =Q .˛/.
Proof. Let ˛ D ˛1 ; ˛2 ; ˛3 denote the conjugates of ˛. The characteristic polynomial
of U is .X  ˛1 /m1 .X  ˛2 /m2 .X  ˛3 /m3 for some m1 ; m2 ; m3 2 Z>0 with m1 C
m2 C m3 D 12. The trace of U is m1 ˛1 C m2 ˛2 C m3 ˛3 2 Q. If we can show that
m1 D m2 D m3 D 4, then Equation (6.4) gives

4 Tr.T 2 / D Tr.U/ D m1 ˛1 C m2 ˛2 C m3 ˛3 D 4.˛1 C ˛2 C ˛3 / D 4 TrK C =Q .˛/:

(6.5)

Therefore, it is enough to show that m1 D m2 D m3 . Since ˛ 2 OK C , we have

˛1 C ˛2 C ˛3 2 Z and therefore .m2  m1 /˛2 C .m3  m1 /˛3 2 Q. Suppose
for contradiction that we are not in the case m1 D m2 D m3 . Then, without loss of
generality, .m2 m1 / ¤ 0 and since ˛2 … Q it follows that .m3 m1 / ¤ 0. Therefore,
˛3 D ˛2 for some  2 Q. But ˛3 is a Galois conjugate of ˛2 and the Galois group
of the Galois closure of K C =Q is either C3 or S3 . Therefore, the automorphism
sending ˛2 to ˛3 has order dividing 6 and hence  is a sixth root of unity in Q.
Therefore,  D 1 and ˛3 D ˛2 . But this gives TrK C =Q .˛/ D ˛1 C ˛2 C ˛3 D ˛1 .
So ˛ D ˛1 D TrK C =Q .˛/ 2 Q, which is a contradiction. t
u
Proof of Theorem 6.9. Suppose for contradiction that p > 4 TrK C =Q .˛/6 =36 . We
will show that the entries of the matrix T commute, contradicting Lemma 6.11. The
key ingredients will be Lemma 6.7 (which states that elements of a maximal order
p
whose reduced norms are smaller than p=2 commute) and Equation (6.7) below.
142 I. Bouw et al.

Recall that
0 1
r ı2 ı3
T D @1 svs_ =ı2 swt_ =ı2 A (6.6)
1 tw_ s_ =ı3 tzt_ =ı3

where ı2 D deg.s/ and ı3 D deg.t/. We have

0 1
1 0 0
@0 ı2 0 A T 2 M3 .End E1 /:
0 0 ı3

We have chosen an isomorphism End0 E1 ! Bp;1 , sending End E1 to a order of

Bp;1 . The dual on End E1 corresponds to the usual involution on Bp;1 . We identify
End0 E1 with Bp;1 and write Nrd.f / D deg.f / D ff _ for f 2 End E1 .
By Lemma 6.12, we have Tr.T 2 / D TrK C =Q .˛/. Writing out the entries on the
diagonal of T 2 gives

0 < deg.r/ C 2 deg.s/ C 2 deg.t/ C deg.v/ C 2 deg.w/ C deg.z/

p
D  TrK C =Q .˛/ < 3 6 p=4: (6.7)

Note that the sum of degrees is a sum of non-negative integers. We want to use (6.7)
0 1
1 0 0
to bound the reduced norms of the non-scalar entries of @0 ı2 0 A T: Recall that,
0 0 ı3
in light of Lemma 6.10, we are assuming that s and t are nonzero. Therefore,
deg.s/; deg.t/  1 and (6.7) gives
p p
i) Nrd.r/ D deg.r/ < 3 6 p=4  4 < p=2,
p
ii) 2 deg.s/ C deg.v/ < 3 6 p=4, p
iii) 2.deg.s/ C deg.t/ C deg.w//
p < 3 6
p=4,
iv) 2 deg.t/ C deg.z/ < 3 p=4.
6

Observe that Nrd.swt_ / D deg.s/ deg.w/ deg.t/ D Nrd.tw_ s_ /. So it remains to

bound the reduced norms of svs_ , swt_ and tzt_ . Let a 2 R>0 . The maximum of
the function f .x/ D x2 .a  2x/ for x  0 is achieved
p at x D a=3 and we have
f .a=3/ D .a=3/3 . Applying this to ii) with a D 3 6 p=4, we see that
p p
Nrd.svs_ / D deg.s/2 deg.v/ < . 6 p=4/3 D p=2:

Similarly, using iv) we get

p p
Nrd.tzt_ / D deg.t/2 deg.z/ < . 6 p=4/3 D p=2:
Bad Reduction of Genus Three Curves with Complex Multiplication 143

Using iii), we get

p
Nrd.swt_ / D deg.s/ deg.w/ deg.t/  .deg.s/ C deg.w//2 2 deg.t/ < . 6 p=4/3
p
D p=2:
0 1
1 0 0
Therefore, by Lemma 6.7, the entries of @0 ı2 0 A T commute. Since the entries
0 0 ı3
0 1
1 0 0
of @0 ı2 0 A T are just scalar multiples of the entries of T, this means that the
0 0 ı3
entries of T commute. But this contradicts Lemma 6.11. Therefore, the assumption
p > 4 TrK C =Q .˛/6 =36 does not hold. t
u

6.4 Solutions to the Embedding Problem in the Case that K

Contains an Imaginary Quadratic Subfield

In this section, we consider the case where the sextic CM-field K contains an
imaginary quadratic subfield K1 . We show that the embedding problem 6.3 has
solutions for every prime p (Corollary 6.15). The solutions are constructed via the
reduction at p of a CM-abelian variety A D E3 in characteristic zero, where E is
an elliptic curve. In particular, the CM-type of A is imprimitive (Theorem 3.2).
The solutions we construct may therefore be called degenerate solutions to the
embedding problem.
The point is that if K is a CM-field which contains an imaginary quadratic
subfield then there always exist imprimitive CM-types for K. This is what allows
for the existence of degenerate solutions to the embedding problem. Recall from
Corollary 3.4 that there do not exist imprimitive CM-types .K; '/ for CM-fields
that do not contain a proper CM-subfield.
The proof of Theorem 6.8 relied on showing non-existence of solutions of the
embedding problem for sufficiently large primes (Theorem 6.9) in the case where
the sextic CM-field contains no proper CM-subfield. In contrast, if C is a curve
whose Jacobian has CM by a sextic CM-field K which contains an imaginary
quadratic field, then this strategy breaks down because there the embedding problem
has degenerate solutions for all primes p (Corollary 6.15). The embedding problem,
as formulated in Problem 6.3, does not take the CM-type into consideration. It may
be possible to prove an analogous result to Theorem 6.8, in the case that K contains
a proper CM-subfield, using a more refined formulation of the embedding problem
that includes the CM-type as part of the data.
144 I. Bouw et al.

Proposition 6.13. Let K be a sextic CM-field containing a proper CM subfield K1 .

Let E be an elliptic curve over an arbitrary field and suppose that there exists an
embedding K1 ,! End0 .E/. Then there exists an order O of K and a ring embedding

O ,! End.E3 / D M3 .End.E//

such that the Rosati involution on End.E3 / corresponding to the product polariza-
tion on A D E3 induces complex conjugation on O.
Proof. It suffices to give an injective Q-algebra homomorphism

K ,! End0 .E3 / D M3 .End0 .E//: (6.8)

This can be achieved as follows. Write K D K C K1 where K C =Q is a totally real

field with ŒK C W Q D 3. Choose a primitive element ˛ of K C =Q, so K C D Q.˛/.
Embed K1 diagonally via the fixed embedding of K1 into End0 .E/. Map ˛ to a
symmetric matrix Q 2 M3 .Q/ which has the same minimal polynomial as ˛. Since
all the conjugates of ˛ are real, the existence of the matrix Q is proved in Theorem 4
of Bender (1968). Extend to a Q-algebra homomorphism. t
u
In Remark 5.6, we reviewed the construction in characteristic 0 of genus 3 curves
with CM by a sextic CM-field K. Similarly, when K1 is an imaginary quadratic
field, elliptic curves with CM by K1 exist in characteristic zero. For example, we
may take E D C=OK1 , where we consider the maximal order OK1 of K1 as lattice
in C (Silverman 1994, Remark II.4.1.1). Then End.E/ D OK1 . Moreover, j.E/ is
an algebraic integer (Silverman 1994, Theorem II.6.1). (This can be deduced from
Theorem 4.1 which states that E has potentially good reduction.) In particular, E can
be defined over the number field M WD Q.j.E//.
We now show the existence of elliptic curves with CM by K1 in positive
characteristic. As above, E=M is an elliptic curve defined over the number field
M with End.E/ D OK1 . We choose a rational prime p, and let p be a prime of M
above p. After extending M if necessary, we may assume that E has good reduction
at p. Write Ep for the reduction of E at p. We obtain an embedding

OK1 D End.E/ ,! End.Ep /:

This proves the following lemma.

Lemma 6.14. Let p be a prime. Then there exists an elliptic curve Ep in character-
istic p with OK1 ,! End.Ep /.
The following result follows immediately from Lemma 6.14 and Proposi-
tion 6.13.
Corollary 6.15. Let K be a sextic CM-field containing an imaginary quadratic
field K1 . Then there exists an order O of K for which there exists a solution to
the embedding problem for O and p for every prime number p.
Bad Reduction of Genus Three Curves with Complex Multiplication 145

Corollary 6.15 does not specify whether the elliptic curve Ep from Lemma 6.14
is ordinary or supersingular. The following proposition answers this question. Note
that it follows that the set of primes where the elliptic curve Ep is supersingular has
Dirichlet density 1=2.
Proposition 6.16 (Deuring’s Theorem). Let E=M be an elliptic curve with CM by
OK1 . Let p be a rational prime and p be a prime of M above p such that E has good
reduction at p. Then the reduction Ep of E at p is supersingular if and only if p is
inert or ramified in K1 .
Proposition 6.16 is well known, but hard to find explicitly in the literature. The
statement can be proved using Theorem 10 of Section 10.4 of Lang (1987). We give
the idea of the proof of the proposition. Let E=Fq be an elliptic curve. Write  for
its q-Frobenius endomorphism. Then E is supersingular if and only if there exists
integers n; m such that  n D Œpm , where Œp denotes multiplication by p. (See, for
example, the proof of the Theorem of Deuring in Section 22 of Mumford 1970.)
The theorem from Lang (1987) shows that this happens if and only if p is inert or
ramified in K1 .

Appendix: Equations

In this section, we list the equations obtained from a possible solution to the
embedding problem. We start by setting some notation.
Let K C be the maximal real subfield of the sextic CM-field K D K C .
/. Take
an integral basis of OK C , so OK C D ˛1 Z ˚ ˛2 Z ˚ ˛3 Z. We may assume that
K C D Q.˛1 /. We fix the following notation:
• TrK=K C .
/ D a1 ˛1 C a2 ˛2 C a3 ˛3
• NmK=K C .
/ D b1 ˛1 C b2 ˛2 C b3 ˛3
• fi .x/ D x3 C mi x2 C ni x C si is the characteristic polynomial of ˛i over Q for
i D 1; 2; 3:
A solution to the embedding problem (Problem 6.3) gives us three elliptic curves
E1 ; E2 ; E3 and an embedding of  W OK ,! End.E1  E2  E3 / such that Rosati
involution on E1  E2  E3 restricts to complex conjugation in the image of OK .
This gives the following conditions on .˛i / and .
/:

1. Commutativity:
(a) .˛i /.
/ D .
/.˛i / for all i D 1; 2; 3.
(b) .˛i /.˛j / D .˛j /.˛i / for all i ¤ j 2 f1; 2; 3g.
2. Characteristic polynomial: fi ..˛i // D 0 for all i D 1; 2; 3.
3. Norm: .
/.
/ D b1 .˛1 / C b2 .˛2 / C b3 .˛3 /, where  denotes the conjugate
transpose.
4. Trace: .
/ C .
/ D a1 .˛1 / C a2 .˛2 / C a3 .˛3 /.
146 I. Bouw et al.

5. Duality/Complex conjugation: .˛i / D .˛i / for all i D 1; 2; 3. Since we are

interested in the case that Rosati involution induces complex multiplication and
since
can be chosen so that
2 2 K C is totally negative, we have .
/ D .
/.

In the rest of this appendix, we will only write the conditions for i D 1 which
is enough if we have a power basis. In any case, the other relations for i D 2; 3 are
similar. We now write the conditions above in terms of matrix coefficients. We are
using the conventions and maps introduced in Section 6.2.
0 1 0 1
abc pq r
Let M D .˛1 / be the matrix @ d e f A and N D .
/ be the matrix @ s t uA :
g h` v wy

Equations for Duality/Complex Conjugation Condition

0 1
ab c
The relation .
/ D .
/ translates into M D M _ i.e., @ d e fA D
g h `
0 1
a_ d _ g_
@ b _ e_ h _ A :
c_ f _ `_
This gives us the following relations.
Remark 6.17. Note that we name the relations with respect to the variables we
intend to use later on. Our aim is to simplify the equations and write everything
in terms of the upper triangular entries of our matrices which are a; b; c; e; f ; ` in the
case of M and p; q; r; t; u; y in the case of N.
(b-d) d D b_
(c-g) g D c_
(f-h) h D f_
(int) a; e; ` are integral and in Q, hence they are integers.
0 1 0 _ 1
pq r p s_ v _
The relation .
/_ D .
/ translates into: @ s t uA D @ q_ t_ w_ A :
v wy r_ u_ y_
This gives us the following relations:
(q-s) s D q_
(r-v) v D r_
(u-w) w D u_
(trace) p D p_ , t D t_ , and y D y_
i.e., p, t, and y have trace zero in End.E1 /, End.E2 /, and End.E3 /
respectively.
Bad Reduction of Genus Three Curves with Complex Multiplication 147

Equations for Commutativity Condition

Using M and N as above, the condition means MN D NM which translates into the
following equations:
(i-i) ap C bs C cv D pa C qd C rg. (By equation (int) in section “Equations for
Duality/Complex Conjugation Condition”, a is an integer. Hence ap D pa
and bs C cv D qd C rg.)
(i-ii) aq C bt C cw D pb C qe C rh
(i-iii) ar C bu C cy D pc C qf C r`
(ii-i) dp C es C f v D sa C td C ug
(ii-ii) dq C et C fw D sb C te C uh (By equation (int) in section “Equations for
Duality/Complex Conjugation Condition”, e is an integer. Hence et D te
and dq C fw D sb C uh.)
(ii-iii) dr C eu C fy D sc C tf C u`
(iii-i) gp C hs C `v D va C wd C yg
(iii-ii) gq C ht C `w D vb C we C yh
(iii-iii) gr C hu C `y D vc C wf C y` (By equation (int) in section “Equations for
Duality/Complex Conjugation Condition”, ` is an integer. Hence `y D y`
and gr C hu D vc C wf .)

Combining Duality and Commutativity Conditions

Now we will plug in the equations we obtained in section “Equations for Dual-
ity/Complex Conjugation Condition” into the equations we obtained in section
“Equations for Commutativity Condition”. Note that our aim is to simplify the
equations and write everything in terms of the upper triangular entries of our
matrices which are a; b; c; e; f ; ` in the case of M and p; q; r; t; u; y in the case of N.

Relation Obtained using:

bq_ C cr _ C rc_ C qb_ D 0 (i-i), (c-g), (q-s), (v-r)
pb C qe C rf _  aq  bt C cu_ D 0 (i-ii), (u-w), (f-h)
ar C bu C cy  pc  qf  r` D 0 (i-iii)
b_ p  eq_  fr _ C q_ a  tb_  uc_ D 0 (ii-i), (b-d), (q-s), (r-v), (q-s)
b_ q  fu_ C q_ b  uf _ D 0 (ii-ii), (b-d), (u-w), (q-s), (f-h)
dr C eu C fy C q_ c  tf  u` D 0 (ii-iii), (q-s)
c_ p  f _ q_ C .a  `/r _ C u_ b  yc_ D 0 (iii-i), (c-g), (f-h), (s-q), (r-v), (u-w), (b-d), (int)
c_ q C f _ t C .e  `/u_ C r _ b  yf _ D 0 (iii-i), (c-g), (f-h), (u-w), (r-v), (int)
c_ r C f _ u C r _ c C u_ f D 0 (iii-i), (f-h), (u-w), (r-v)
148 I. Bouw et al.

Equations for Characteristic Polynomial Condition

The characteristic polynomial condition for i D 1 translates into 0 D M 3 C m1 M 2 C

n1 M C s1 . Combining this equality with Equation (int) of section “Equations for
Commutativity Condition” gives the following equations. For instance, for the top
left corner of the matrix sum we get

0 D a3 CabdCacgCbdaCbedCbfgCcgaCchdCc`gCm1.a2 CbdCcg/Cn1 aCs1 :

If we apply Condition (int) this turns into

.2a C e C m1 /bd C .2a C ` C m1 /cg C bfg C chd C a3 C m1 a2 C n1 a C s1 D 0:

The following is the list of equations coming from all nine entries.
(i) .2a C e C m1/bd C .2a C ` C m1/cg C bfg C chd C a3 C m1 a2 C n1 a C s1 D 0
(ii) .a2 C ae C e2 C m1 a C m1 e C n1 /b C .e C ` C m1 C a/ch C bdb C bfh C cgb D 0
(iii) .a2 C a` C `2 C m1 a C m1 ` C n1 /c C .a C e C ` C m1 /bf C bdc C cgc C chf D 0
(iv) .a2 C ea C e2 C m1 a C m1 e C n1 /d C .e C a C ` C m1 /fg C dbd C dcg C fhd D 0
(v) .a C 2e C m1 /db C .2e C ` C m1 /fh C dch C fgb C e3 C m1 e2 C n1 C s1 D 0
(vi) .a C ` C e C m1/dc C .e2 C e` C `2 C m1 e C m1 ` C n1 /f C dbf C fgc C fhf D 0
(vii) .a2 C`a C`2 Cm1 a Cm1 `Cn1 /g C.a CeC`Cm1 /hd Cgbd Cgcg Chfg D 0
(viii) .e2 C `e C `2 C m1 e C m1 ` C n1 /h C .a C e C ` C m1 /gb C gch C hdb C hfh D 0
(ix) .a C 2` C m1 /gc C .e C 2` C m1/hf C gbf C hdc C `3 C m1 `2 C n1 ` C s1 D 0

Combining Duality and Characteristic Polynomial Conditions

Now we will plug in the equations we obtained in section “Equations for Dual-
ity/Complex Conjugation Condition” into the equations we obtained in section
“Equations for Characteristic Polynomial Condition”. Note that our aim is to
simplify the equations and write everything in terms of the upper triangular entries
of our matrices which are a; b; c; e; f ; ` in the case of M and p; q; r; t; u; y in the case
of N. Note that Nrd.x/ D xx_ ; Tr.x/ D x C x_ denote the reduced norm and trace
of an element. Since the norm and trace are scalars, they commute with everything
else.
We start with the relations coming from M:

(I) .2a C e C m1 / Nrd.b/ C .2a C ` C m1 / Nrd.c/ C Tr.bfc_ / C a3 C m1 a2 C

n1 a C s1 D 0
(II) .a2 C ae C e2 C m1 a C m1 e C n1 C Nrd.b/ C Nrd.c/ C Nrd.f //b C .a C e C
` C m1 /cf _ D 0
(III) .a2 C a` C `2 C m1 a C m1 ` C n1 C Nrd.b/ C Nrd.c/ C Nrd.f //c C .a C e C
` C m1 /bf D 0
Bad Reduction of Genus Three Curves with Complex Multiplication 149

(IV) .a2 C ae C e2 C m1 a C m1 e C n1 C Nrd.b/ C Nrd.c/ C Nrd.f //b_ C .a C

e C ` C m1 /fc_ D 0
(V) .a C 2e C m1 / Nrd.b/ C .2e C ` C m1 / Nrd.f / C Tr.b_ cf _ / C e3 C m1 e2 C
n1 e C s1 D 0
(VI) .e2 C e` C `2 C m1 e C m1 ` C n1 C Nrd.b/ C Nrd.c/ C Nrd.f //f C .a C ` C
e C m1 /b_ c D 0
(VII) .a2 C a` C `2 C m1 a C m1 ` C n1 C Nrd.b/ C Nrd.c/ C Nrd.f //c_ C .a C
e C ` C m1 /f _ b_ D 0
(VIII) .e2 C e` C `2 C m1 e C m1 ` C n1 C Nrd.b/ C Nrd.c/ C Nrd.f //f _ C .a C
e C ` C m1 /c_ b D 0
(IX) .a C 2` C m1 / Nrd.c/ C .e C 2` C m1 / Nrd.f / C Tr.c_ bf / C `3 C m1 `2 C
n1 ` C s1 D 0
Write Tr.X/ for the sum of the entries on the main diagonal of a matrix X. Notice
p
that if we take
D ˛1 like in Section 6.3, then

m1 D Tr.˛1 / D Tr.N 2 / D Tr.M/ D a C e C `;

where the first equality follows by definition, the second equality is Lemma 6.12,
p
the third equality holds because we took
D ˛1 , and the final equality is the
definition of Tr.M/. This implies that Equation (II) = Equation (IV), Equation (III)=
Equation (VII) and Equation (VI)=Equation (VIII).
Combining m1 D a C e C ` with relations (I)–(IX), we deduce the following
relations on the coefficients m1 ; n1 ; s1 of the characteristic polynomial of ˛1 :
(1) m1 D .a C e C `/
(2) n1 D ae C e` C a`  Nrd.b/  Nrd.c/  Nrd.f / (using Equation (1) together
with Equations (II), (III) and (VI)).
(3) s1 D a Nrd.f / C e Nrd.c/ C l Nrd.b/  ae`  Tr.bfc_ / (using Equation (1)
together with Equations (I), (V) and (IX)).

Acknowledgements The authors would like to thank the Centre International de Rencontres
Mathématiques in Luminy for sponsoring the Women in Numbers - Europe (Femmes en Nombre)
workshop and for providing a productive and enjoyable environment for our initial work on this
project. We would especially like to thank the organizers of WINE, Marie José Bertin, Alina Bucur,
Brooke Feigon, and Leila Schneps for making the conference and this collaboration possible. We
also thank the referee for the detailed and helpful report.

The work of MM was partially supported by NSF-DMS 1102858.

References

Bender, E.: Characteristic polynomials of symmetric matrices. Pac. J. Math. 25, 433–441 (1968).
MR0229619 (37 #5193)
Bosch, S., Lütkebohmert, W., Raynaud, M.: Néron models. Ergebnisse der Mathematik und ihrer
Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], vol. 21. Springer, Berlin
(1990). MR1045822 (91i:14034)
150 I. Bouw et al.

Bouw, I.: The p-rank of ramified covers of curves. Compos. Math. 126(3), 295–322 (2001).
MR1834740 (2002e:14045)
Bouw, I., Wewers, S.: Group actions on curves and the lifting problem. http://math.arizona.edu/
swc/aws/2012/2012BouwWewersNotesPreliminary.pdf (2012)
Bruinier, J., Yang, T.: CM-values of Hilbert modular functions. Invent. Math. 163(2), 229–288
(2006). MR2207018 (2008b:11053)
Chai, C.-L., Conrad, B., Oort, F.: Complex Multiplication and Lifting Problems. Mathematical
Surveys and Monographs, vol. 195. American Mathematical Society, Providence (2014).
MR3137398
Crew, R.: Etale p-covers in characteristic p. Compos. Math. 52(1), 31–45 (1984). MR742696
(85f:14011)
Deligne, P., Mumford, D.: The irreducibility of the space of curves of given genus. Inst. Hautes
Études Sci. Publ. Math. 36, 75–109 (1969). MR0262240 (41 #6850)
Dodson, B.: The structure of Galois groups of CM-fields. Trans. Am. Math. Soc. 283(1), 1–32
(1984). MR735406 (86i:11063)
Goren, E., Lauter, K.: Evil primes and superspecial moduli. Int. Math. Res. Not. 19, 53864 (2006).
MR2250004 (2007f:11061)
Goren, E., Lauter, K.: Class invariants for quartic CM fields. Ann. Inst. Fourier (Grenoble) 57(2),
457–480 (2007). MR2310947 (2008i:11075)
Goren, E., Lauter, K.: A Gross-Zagier formula for quaternion algebras over totally real fields.
Algebra Number Theory 7(6), 1405–1450 (2013). MR3107568
Gross, B., Zagier, D.: On singular moduli. J. Reine Angew. Math. 355, 191–220 (1985). MR772491
(86j:11041)
Kani, E., Rosen, M.: Idempotent relations and factors of Jacobians. Math. Ann. 284(2), 307–327
(1989). MR1000113 (90h:14057)
Klüners, J., Malle, G.: A database for number fields. http://galoisdb.math.upb.de/ (2011)
Koike, K., Weng, A.: Construction of CM Picard curves. Math. Comput. 74(249), 499–518
(electronic) (2005). MR2085904 (2005g:11103)
Lang, S.: Complex Multiplication. Grundlehren der Mathematischen Wissenschaften [Fundamen-
tal Principles of Mathematical Sciences], vol. 255. Springer, New York (1983). MR713612
(85f:11042)
Lang, S.: Elliptic Functions. Graduate Texts in Mathematics, 2nd edn., vol. 112. Springer,
New York (1987). With an appendix by J. Tate. MR890960 (88c:11028)
Lauter, K., Viray, B.: An arithmetic intersection theory formula for denominators of Igusa class
polynomials. http://arxiv.org/abs/1210.7841 (2012)
Milne, J.S.: Complex multiplication. http://www.jmilne.org/math/CourseNotes/cm.htm (2006)
Milne, J.S.: Abelian varieties. http://www.jmilne.org/math/CourseNotes/av.html (2008)
Mumford, D.: Abelian Varieties. Tata Institute of Fundamental Research Studies in Mathematics,
No. 5. Published for the Tata Institute of Fundamental Research/Oxford University Press,
Bombay/London (1970). MR0282985 (44 #219)
Obus, A., Pries, R.: Wild tame-by-cyclic extensions. J. Pure Appl. Algebra 214(5), 565–573 (2010).
MR2577662 (2011h:12011)
Oort, F., Ueno, K.: Principally polarized abelian varieties of dimension two or three are Jacobian
varieties. J. Fac. Sci. Univ. Tokyo Sect. IA Math. 20, 377–381 (1973). MR0364265 (51 #520)
Paulhus, J.: Decomposing Jacobians of curves with extra automorphisms. Acta Arith. 132(3),
231–244 (2008). MR2403651 (2009c:14049)
Serre, J.-P., Tate, J.: Good reduction of abelian varieties. Ann. Math. (2) 88, 492–517 (1968).
MR0236190 (38 #4488)
Shimura, G.: Abelian Varieties with Complex Multiplication and Modular Functions. Princeton
Mathematical Series, vol. 46. Princeton University Press, Princeton (1998). MR1492449
(99e:11076)
Silverman, J.: Advanced Topics in the Arithmetic of Elliptic Curves. Graduate Texts in Mathemat-
ics, vol. 151. Springer, New York (1994). MR1312368 (96b:11074)
Bad Reduction of Genus Three Curves with Complex Multiplication 151

van der Geer, G., Moonen, B.: Abelian varieties. Preliminary version, http://www.math.ru.nl/
bmoonen/research.html (2011)
Vignéras, M.-F.: Arithmétique des algèbres de quaternions. Lecture Notes in Mathematics, vol.
800. Springer, Berlin (1980). MR580949 (82i:12016)
Yui, N.: On the Jacobian variety of the Fermat curve. J. Algebra 65(1), 1–35 (1980). MR578793
(82m:14016)
Symmetries of Rational Functions Arising
in Ecalle’s Study of Multiple Zeta Values

Adriana Salerno, Damaris Schindler, Amanda Tucker

Abstract In Ecalle’s theory of multiple zeta values he makes frequent use of certain
properties that express symmetries of rational functions in several variables. We
focus on the properties of push-invariance, circ-neutrality, and alternality. Ecalle
states and uses several implications about the relations between these symmetries.
In this paper we investigate two of these implications and prove two results: first,
that push-invariance and circ-neutrality imply the first alternality relation, but not the
more general alternality relations, and second, that alternality does, indeed, imply
circ-neutrality.

1 Introduction

The multiple zeta values are the numbers

X 1
.n1 ; : : : ; nr / D ;
0<k1 <<kr
k1n1    krnr

for ni 2 N with nr > 1. Here, r is the depth and n D n1 C    C nr is the weight

of .n1 ; : : : ; nr /, where 1  r < n. When the depth r D 1 these values are known
as single zeta values and are nothing other than the special values of the Riemann
X 1
zeta function .n/ D . Very little is known about the algebraic nature of the
0<k
kn

A. Salerno ()
Mathematics Department, Bates College, Lewiston, ME, USA
e-mail: asalerno@bates.edu
D. Schindler
Hausdorff Center for Mathematics, University of Bonn, Bonn, Germany
A. Tucker
Department of Mathematics, State University of New York College at Geneseo,
Geneseo, NY, USA

© Springer International Publishing Switzerland 2015 153

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_6
154 A. Salerno et al.

special values of the Riemann zeta function (let alone multiple zeta values). Euler
knew already in 1735 Euler (1775) that

2 4
.2/ D ; .4/ D
6 90
and that, more generally,

B2n .2/n
.2n/ D .1/nC1 2  2n Q:
2.2n/Š

A widely believed folklore conjecture, for example, states that the numbers , .3/,
.5/, .7/; : : : ; .2nC1/ are algebraically independent over Q for any integer n  1.
In studying algebraic independence of zeta values one is led naturally to the
study of similar questions for multiple zeta values. In this context it turns out to be
useful to define what is called the Q-algebra of formal multiple zeta values, which
is generated by symbols of the form Z.k1 ; : : : ; kr / modulo the standard regularized
shuffle and stuffle algebraic relations. The structure of this algebra is a tantalizing
and much-studied question. Several authors, particularly Hoffman (1997) and
Zagier (1993) have made seminal contributions to its study. J. Ecalle has designed
and implemented a vast program to study it, using his own personal language and
theory, which yields beautiful and natural generalizations, restatements and proofs
of some of the important facts and conjectures concerning multiple zeta values.
The key to his theory is to place the whole situation within a bigger universe,
known as the theory of moulds; in this paper, we consider only “moulds” which are
in fact rational functions of several variables and which are the most relevant moulds
in the study of multiple zeta values. An essential feature of Ecalle’s theory is the
study of symmetry properties of moulds, such as the properties of circ-neutrality,1
push-invariance, and alternality (defined in Section 2) that we concentrate on in this
paper. Ecalle’s seminal article Ecalle (2011) contains a grand survey of some of his
main ideas. However, because of the length and depth of the theory, there are few
proofs. The following assertion (for general moulds, but we restrict it to rational
functions here) appears in section 11.9 of Ecalle (2011).
Assertion 1.1. If A is a rational function that is push-invariant and circ-neutral,
then A is alternal.
This statement turns out not to be true in full generality,2 but we prove that push-
invariance and circ-neutrality do imply the first alternality relations in Section 4.
Moreover, we found via Maple calculations that there is no counterexample to the
more general case in two variables and low degree, but we found one in degree three
with five variables, which we give in Section 4.

1
Ecalle uses the terminology pus-neutrality for this property.
2
However, the statement forms part of the proof of a result for which Ecalle gave a very different
but complete proof in a subsequent paper.
Symmetries of Rational Functions Arising in Ecalle’s Study of Multiple Zeta Values 155

In Section 2.4 of Ecalle (2011) we find the following assertion (again for general
moulds):
Assertion 1.2. Every alternal rational function is circ-neutral.
The main goal of Section 5 is to detail the proof of this assertion. Our strategy
to prove Assertion 1.2 is to first reduce it to the polynomial case, which is done
in Section 3. We can then treat the polynomial case using properties of certain Lie
algebras and Ecalle’s multiplication operator mu, whose definition we introduce in
Section 5.

2 Definitions

There are three properties of rational functions that are of interest: push-invariance,
circ-neutrality, and alternality.
If A.u1 ; : : : ; ur / is a rational function in r variables, we define

circ A.u1 ; : : : ; ur / D A.ur ; u1 ; : : : ; ur1 /

and, for u0 D u1  u2      ur ,

push A.u1 ; : : : ; ur / D A.u0 ; u1 ; : : : ; ur1 /:

Definition 2.1. A is push-invariant if

push A D A:

Definition 2.2. A is circ-neutral if

A C circ A C circ2 A C    C circr1 A D 0:

Definition 2.3. A is alternal if

X
A.w/ D 0
w2sh.u1 u2 :::uk /.ukC1 :::ur /

for all 1  k  d 2r e, where sh.w; w0 / is the set of all possible shuffles of the words
w and w0 , that is, permutations of the letters of w and w0 that preserve the ordering
in w and the ordering in w0 .
A more technical definition of the shuffle of two words is the following. Let A
be the set of all words in the alphabet fa1 ; : : : ; arCs g and suppose w D a1 a2    ar
and w0 D arC1 arC2    arCs . Let the symmetric group on r C s letters SrCs act on A
by permuting the indices. Let
156 A. Salerno et al.

T D f 2 SrCs j  1 .1/ <  1 .2/ <    <  1 .r/ and

 1 .r C 1/ <  1 .r C 2/ <    <  1 .r C s/g

and note that if x  r and y > r there is no relation whatsoever imposed on  1 .x/
and  1 .y/. Then

sh.w; w0 / D fw00 2 A j w00 D .ww0 / for some  2 Tg:

Example 2.4. An example of a polynomial that is push-invariant, circ-neutral, and

alternal is A.u1 ; u2 / D 2u31  3u21 u2 C 3u1 u22 C 2u32 .
Remark 2.5. We note here that circ-neutrality and alternality are additive properties
that respect the multi-degree of a polynomial. Hence, once we have reduced to the
polynomial case in Section 3 we can further reduce to the case of monomials of
fixed multi-degree.

3 Reduction to the Polynomial Case

In this section, we reduce the study of Assertion 1.2 and similar questions to the
case of polynomials. It turns out that to show that a family of rational functions
satisfy circ-neutrality, push-invariance, or alternality, it suffices to show that a
corresponding family of polynomials satisfies that property.
Lemma 3.1. Every rational function A.u1 ; : : : ; ur / can be written in the form
A D P=Q where P and Q are polynomials in u1 ; : : : ; ur such that Q.u1 ; : : : ; ur /
is invariant under push and any permutation of u1 ; : : : ; ur . If A D P=Q is such an
expression for A, then A is alternal (resp. push-invariant, resp. circ-neutral) if and
only if P is alternal (resp. push-invariant, resp. circ-neutral).
Proof. Consider a rational function A 2 Q.u1 ; : : : ; ur /. We note that

circr D id and pushrC1 D id:

Write A D p=q with p; q 2 QŒu1 ; : : : ; ur . Set u0 D u1      ur , and let the

symmetric group SrC1 act on the r C 1 indices 0; 1; : : : ; r by permutation. We define
Y
Q.u1 ; : : : ; ur / D q.u .1/; : : : ; u .r/ /:
 2SrC1

Then the polynomial Q in r variables is push-invariant and invariant under any

permutation of the r indices 1; : : : ; r. Then we simply set
Symmetries of Rational Functions Arising in Ecalle’s Study of Multiple Zeta Values 157

Y
P.u1 ; : : : ; ur / D p.u1 ; : : : ; ur / q.u .1/ ; : : : ; u .r/ /;
 2SrC1 ;  ¤id

so that A D P=Q, which is of the desired form.

Finally, because Q is invariant under push and any permutation of u1 ; : : : ; ur , we
have that A is circ-neutral (resp. push-invariant, resp. alternal) if and only if P is
circ-neutral (resp. push-invariant, resp. alternal), which proves the lemma. t
u

4 push-Invariance and a Counterexample to 1.1

Checking Assertion 1.1 in Maple (2013), one finds no counterexample for poly-
nomials in fewer than three variables in degree five (or seven variables in degree
four), but in 3 variables and degree 5, a nice counterexample appears; we give it
at the end of this section. First, let us show that, even though push-invariance and
circ-neutrality do not imply alternality in general, they do imply the first alternality
relation (shuffling one variable into the rest).
Theorem 4.1. If A is
X circ-neutral and push-invariant, then the first alternal relation
holds. That is, A.w/ D 0.
w2sh.u1 /.u2 ;:::;ur /

Proof. The circ-neutral relation guarantees that

A C circ A C circ2 A C    C circr1 A D 0:

That is,

A.u1 ; : : : ; ur / C A.ur ; u1 ; : : : ; ur1 / C    C A.u2 ; u3 ; : : : ur ; u1 / D 0:

Now, assuming A is push-invariant, we have that A under any change of variables is

push-invariant. It follows that circk A is push-invariant for all k. Thus, we can push
each term an appropriate number of times, preserving equality, to get

push A C pushr1 circ A C pushr2 circ2 A C    C push2 circr1 A D 0:

So we see that

A.u0 ; u1 ; : : : ; ur1 / C A.u1 ; u0 ; : : : ; ur1 / C A.u1 ; u2 ; u0 ; : : : ; ur1 / C    C

A.u1 ; u2 ; : : : ; u0 ; ur1 / C A.u1 ; : : : ; ur1 ; u0 / D 0;

which is precisely the first alternal relation, written in terms of the variables
u0 ; u1 ; : : : ; ur2 ; ur1 . Just as push-invariance implies push-invariance under any
158 A. Salerno et al.

change of variable, so does knowing the first alternal relations with one set of
variables imply the first alternal relations after any change of variables. This
concludes the proof. t
u
Next we observe that Assertion 1.1 is trivially true for the case of linear forms in
any given number of variables. Indeed, in the following lemma we show that there
are no nonzero linear push-invariant forms at all.
Lemma 4.2. There are no non-trivial push-invariant linear forms.
Proof. Assume that the linear form A.u1 ; : : : ; ur / D a1 u1 C    C ar ur is push-
invariant. Then the equation A D push A implies that

a1 u1 C    C ar ur D a1 .u1      ur / C a2 u1 C    C ar ur1 :

We compare coefficients on both sides and obtain the system of linear equations

a1 D a1 C a2
a2 D a1 C a3
::
:
ar1 D a1 C ar
ar D a1 :

If ar D 0, then all the other ai have to be zero. If ar is nonzero, then we may assume
after normalization that ar D 1. The last equation then gives that a1 D 1; the first
equation gives a2 D 2; the second a3 D 3, until we obtain from the penultimate
equation that ar D r, which is a contradiction to ar D 1. t
u
Remark 4.3. Determining the dimension of the subspace of push-invariant polyno-
mials is non-trivial. We include here Maple calculations of the dimension of the
space of push-invariant, circ-neutral polynomials for small values of r (number of
variables) and n (degree) (Maple 2013):
0 1
rj
n
12 3 4 5 6 7 8 9
B2 00 1 0 1 1 1 1 2C
B C
B3 01 0 2 1 3 2 5 3C
B C
B4 00 1 1 3 3 6 7 11C
B C
B C
B5 01 2 3 5 11 14 24 34C
B C
B6 00 1 3 8 14 28   C
B C
B7 01 2 8 13 31 55   C
B C
@8 00 3 5 19 43    A
9 01 2 11 29    
Symmetries of Rational Functions Arising in Ecalle’s Study of Multiple Zeta Values 159

Entering the rows individually into the On-line Encyclopedia of Integer Sequences
(OEIS), we note that the dimensions for r D 2 appear to correspond to the
dimensions Œ.2n C 2/=4  Œ.2n C 4/=6 of cusp forms of weight 2n C 6 on SL2 .Z/.
Apart from this the significance of these dimensions is not clear.
Despite push-invariance being a very strong property, it turns out, as observed at
the beginning of this section, that push-invariance and circ-neutrality together are
not enough to ensure that a given rational function is alternal in higher degree.
The following polynomial P is the smallest counterexample that we found to
Assertion 1.1. The Maple code used can be found in Appendix.
P D u21 u3 C u21 u4 C 2u1 u2 u4  2u1 u2 u5  u1 u23  2u1 u3 u4 C u1 u24
C2u1 u4 u5 C u22 u3  u22 u5 C u2 u23 C 2u2 u3 u5  2u2 u4 u5  u2 u25  u23 u4
Cu23 u5  u3 u24 C u3 u25

5 circ-Neutrality and a Proof of Assertion 1.2

The goal of this section is to prove that any alternal rational function is, in fact,
circ-neutral.
Lemma 5.1. To show any alternal rational function is circ-neutral, it suffices to
show that any alternal polynomial is circ-neutral.
Proof. Let A.u1 ; : : : ; ur / be an alternal rational function. Then by Section 3, we
can write A D P=Q with P alternal and Q invariant under any permutation of the
variables u1 ; : : : ; ur . Thus, to show A is circ-neutral, it suffices to show that P is. u
t
The remainder of this section is devoted to proving that if a polynomial P is
alternal then it must also be circ-neutral (see Theorem 5).
Our approach is to first prove a proposition that gives a useful characterization
of circ-neutral polynomials. We then show that an alternal polynomial must fit
this characterization. In order to prove this proposition, we first introduce Ecalle’s
multiplication operator mu on two polynomials.
Definition 5.2. If A and B are polynomials in rA (resp. rB ) variables, then mu.A; B/
is a polynomial in r D rA C rB variables defined as

mu.A; B/ D A.u1 ; : : : ; urA /B.urA C1 ; : : : ; urA CrB /:

Furthermore, we set ŒŒA; B D mu.A; B/  mu.B; A/.

Note that mu.B; A/ D B.u1 ; : : : ; urB /A.urB C1 ; : : : ; urA CrB / is also a polynomial in
r variables.
Proposition 5.3. Let A and B be monomials of degree dA (resp. dB ) in rA (resp. rB )
variables. Then M WD ŒŒA; B is circ-neutral.
160 A. Salerno et al.

Proof. M is a homogeneous polynomial of degree d D dA C dB in r D rA C rB

variables given by

M D A.u1 ; : : : ; urA /B.urA C1 ; : : : ; urA CrB /  B.u1 ; : : : ; urB /A.urB C1 ; : : : ; urA CrB /:

ar br
Writing A.u1 ; : : : ; urA / D ua11 : : : urAA and B.u1 ; : : : ; urB / D ub11 : : : urBB , we have

a b b a
M.u1 ; : : : ; ur / D ua11 : : : urArA ubrA1C1 : : : ur rB  ub11 : : : urBrB uarB1C1 : : : ur rA :

If we now consider the sum

X
r
M C circ M C circ2 M C    C circr1 M D M.ui ; : : : ; ur ; u1 ; : : : ; ui1 /;
iD1

we see that the positive monomial from the i-th term cancels with negative
monomial in the i C rA -th term (with indices taken modulo r in the set f1; : : : ; rg).
Thus, in fact,

X
r
M.ui ; : : : ; ur ; urC1 ; : : : ; ui1 / D 0;
iD1

so M is indeed circ-neutral. t
u
Remark 5.4. It follows directly from Proposition 5.3 and additivity of circ-
neutrality that if A and B are any polynomials, not necessarily monomials, then
the polynomial ŒŒA; B is circ-neutral.
Note that both alternality and circ-neutrality are empty conditions for a polyno-
mial in one variable, so we adopt the convention of saying alternality implies circ-
neutrality in this case. If A is a polynomial in r > 1 variables, because alternality
and circ-neutrality both respect degree, we can assume that A is homogenous of
multi-degree d. Additivity and Proposition 5.3 together imply that, in order to prove
Assertion 1.2, it suffices to show that any alternal A is a linear combination of terms
of the form ŒŒB; D; this is the method we use to prove the following theorem.
Theorem 5.5. If A is a homogeneous alternal polynomial of degree d in r > 1
variables, then A is circ-neutral.

Proof. We capitalize on properties of a certain Lie algebra to show that if A is

alternal then it is of the desired form ŒŒB; D.
Let QŒu1 ; : : : ; ur  be the ring of polynomials in r commuting variables and let
R D Qhx; yi be the ring of polynomials in two non-commuting variables. Let Rr
denote the Q vector subspace of R spanned by monomials containing exactly r y’s.
Note that R0 is spanned by the monomials that are powers in x, including x0 D 1.
As a Q vector space, R is the direct sum of the Rr ’s for all r.
Symmetries of Rational Functions Arising in Ecalle’s Study of Multiple Zeta Values 161

For r  1, define the map of Q vector spaces

r W QŒu1 ; : : : ; ur  ! Rr (1)

by extending linearly from the map ua11    uar r 7! Ca1 C1 Ca2 C1 : : : Car1 C1 Car C1 ,
where

Ci D ad.x/i1 y D Œx;    x; Œx; Œx; y    ;

and Œx; y D xy  yx is the standard Lie bracket. So, for example, we have

r .1/ D r .u01    u0r / D C1r D yr ;

r .u1 / D r .u11 u02    u0r / D C2 C1r1 D Œx; yyr1 D .xy  yx/yr1 ;
r .u21 / D r .u21 u02    u0r / D C3 C1r1 D Œx; Œx; yyr1 ;
r .u2 / D r .u01 u12 u03    u0r / D C1 C2 C1r1 D yŒx; yyr2 ;
r .7u21  5u2 / D r .7u21 u02    u0r  5u01 u12 u03    u0r / D 7C3 C1r1  5C1 C2 C1r2 ; and
r .u22 ur / D r .u1 u22 ur / D C1 C3 C1r3 C2 :

A key observation is that if B and D are two polynomials in s and t variables,

respectively, then we have sCt .mu.B; D// D s .B/t .D/, so that

sCt .mu.B; D/  mu.D; B// D s .B/t .D/  t .D/s .B/

D Œs .B/; t .D/; (2)

where mu is the operator from Definition 5.2 above.

Let LieŒx; y be the free Lie algebra generated by x and y under the Lie bracket,
Œf ; g D fg  gf , and let Lier Œx; y be the Q vector subspace of LieŒx; y spanned by
brackets of exactly r y’s with any number of x’s, i.e., the space of Lie polynomials
with homogeneous degree r in y.
Lemma 5.6. If A is an alternal polynomial in r variables, then r .A/ lies in
Lier Œx; y.
Proof. We note that if A is an alternal polynomial in the variables u1 ; : : : ; ur , then the
polynomial r .A/ considered as a polynomial in the variables Ci satisfies the shuffle
relations. By Theorem 1.4 (Section 1.5) in Rautenauer (1993) any polynomial that
satisfies the shuffle relations in the variables Ci is a Lie polynomial in the variables
Ci . Now we recall that Ci is defined as

Ci D ad.x/i1 .y/;

and, hence, is a Lie polynomial in the variables x and y. Thus, any Lie polynomial
in the variables Ci is also a Lie polynomial in the variables x and y. t
u
162 A. Salerno et al.

Let Alr denote the subspace of alternal polynomials in QŒu1 ; : : : ; ur .

Lemma 5.7. The map r W Alr ! Lier Œx; y is a bijection for every r  1.
Proof. The main point is that the Ci are algebraically independent. To see this
we first observe that by Lazard elimination we have a direct sum LieŒx; y D
LieŒx ˚ LieŒC1 ; C2 ; : : : (see Proposition 10 a) in Bourbaki 2006). Part b) of the
same proposition proves that if C10 ; C20 ; : : : are non-commutative indeterminates,
then the map induced by Ci0 7! Ci is an isomorphism from the free Lie algebra
LieŒC10 ; C20 ; : : : to LieŒC1 ; C2 ; : : :. Hence the latter is a free Lie algebra. Finally,
Theorem 1 (b) in § 3, no. 1, of Bourbaki (2006) shows that the universal enveloping
algebra of LieŒC10 ; C20 ; : : : is just the free non-commutative polynomial algebra
QhC10 ; C20 ; : : :i, and the previous isomorphism extends uniquely to an isomorphism
of universal enveloping algebras. This shows that the universal enveloping algebra
of LieŒC1 ; C2 ; : : :, namely the polynomial algebra QhC1 ; C2 ; : : :i is also free on
these variables, showing that they are indeed algebraically independent. Thus, any
element of LieŒC1 ; C2 ; : : : has a unique expression as a polynomial in the Ci , and
thus a unique preimage under the map (1). So the map r is both injective and
surjective, which concludes the proof. t
u
Remark 5.8. The above lemma shows that the map r W Lier Œx; y ! Alr ,
which is the restriction to the Lie algebra of the map on QhC1 ; C2 ; : : :i defined by
a1 1
r .Ca1 : : : Car / D u1 : : : urar 1 , is an explicit inverse to r .
Lemma 5.9. Let r > 1. Then any element of Lier Œx; y can be written as a sum
P
i ai Œfi ; gi  with fi ; gi 2 LieŒC1 ; C2 ; : : : for all i.

Proof. Note that the assertion of the Lemma is equivalent to saying that one can
decompose any element of Lier Œx; y into a sum of brackets in which none of the
fi ; gi is equal to x.
We observe that every element in Lier Œx; y for r > 1 can be written as a linear
combination of Lie brackets of r y’s and any number of x’s. By additivity it is hence
enough to prove the lemma for a single Lie bracket of r y’s and s x’s. For this we
note that any Lie bracket is of the form Œf ; g with f ; g 2 LieŒx; y. If both f and g are
themselves Lie brackets, or if f or g is equal to y, then we are already in the desired
form. Hence it remains to consider the case where f or g is equal to x. Without
loss of generality we may assume that f D x. Thus we have reduced the P proof to
showing that a Lie bracket of the form Œx; g can be rewritten in the form i ai Œfi ; gi 
with none of the fi ; gi equal to x.
We prove this claim by induction on the degree of the bracket, which equals r C s
in the notation above. Recall that we have assumed r > 1 and hence r C s  4, since
we are considering brackets of the form Œx; g where g is a Lie bracket containing at
least 2 y’s, so of degree at least 3. The base
P case is thus the example g D Œy; Œx; y.
To write this as a linear combination i ai Œfi ; gi  with none of the fi ; gi equal to
x, we use the Jacobi relation

Œa; Œb; c C Œc; Œa; b C Œb; Œc; a D 0;

Symmetries of Rational Functions Arising in Ecalle’s Study of Multiple Zeta Values 163

with a D x, b D y and c D Œx; y. This yields

Œx; Œy; Œx; y C ŒŒx; y; Œx; y C Œy; ŒŒx; y; x D 0:

The middle term is zero, so we can rewrite Œx; g D Œx; Œy; Œx; y D Œy; ŒŒx; y; x,
which is of the desired form. This completes the base case of the induction.
Next we assume that all Lie brackets up to total degree r C s  1 in x and y can be
written in the form stated in the lemma. Consider a Lie bracket of the form Œx; g in
degree r C s. Then g is a Lie bracket
P of degree r C s  1, and hence by our induction
hypothesis we can write g D i ai Œfi ; gi  with none of the fi ; gi equal to x. Now we
use again the Jacobi relation to rewrite the problematic Lie bracket as
X X X
Œx; g D ai Œx; Œfi ; gi  D  ai Œgi ; Œx; fi   ai Œfi ; Œgi ; x:
i i i

Since none of the fi and the gi are equal to x, this completes the proof of the lemma.
t
u
We can now complete the proof of Theorem 5.
Let A be an alternal polynomial in r > 1 variables. Then r .A/ is in Lier Œx; y by
Lemma 5.6. Thus, by Lemma 5.9, we can write
X
r .A/ D ai Œfi ; gi 
i

where none of the fi or gi is equal to x. For each i in the sum, we thus have fi 2
Liesi Œx; y for some si  1, and gi 2 Lieti Œx; y for some ti  1. Now, by Lemma 5.7,
si and ti are surjective, so there exist polynomials Bi and Di satisfying si .Bi / D fi
and ti .Di / D gi .
Summing this up and using (2) and the linearity of r , we have
X
r .A/ D ai Œfi ; gi 
i
X
D ai Œsi .Bi /; ti .Di /
i
X
D ai si Cti .mu.Bi ; Di /  mu.Di ; Bi //
i
!
X
D r ai ŒŒBi ; Di  :
i
164 A. Salerno et al.

Finally, by the injectivity of r proved in Lemma 5.7, we have that

X
AD ai ŒŒBi ; Di :
i

By Proposition 5.3 and additivity this is enough to guarantee that A is circ-neutral,

which concludes the proof of Theorem 5. t
u

Appendix

This is the Maple code for creating a generic push-invariant, circ-neutral polynomial
of degree n in r variables and checking whether it fails on the second alternality
relation. One inputs the degree n of the polynomial and the number r of variables at
the beginning of the program as can be seen below. We used this code to obtain the
counterexample to Assertion 1.1 in Section 4.
#
#(1) Create a r b i t r a r y polynomial in r v a r i a b l e s of
degree n
# ( 2 ) S o l v e t h e s y s t e m s f o r push  i n v a r i a n t and
circ n e u t r a l
# ( 3 ) Check w h e t h e r p o l y n o m i a l s a t i s f i e s s e c o n d
alternality
# relation
n :=3:
r :=5:
# p r o c e d u r e t o co m p u te p u sh ( P )
p u sh P : = p r o c ( P , r )
l o c a l Q:
Q: = ex p a n d ( s u b s ( { u [1]=  add ( u [ k ] , k = 1 . . r ) , s e q ( u [ k ] = u [ k 1 ] ,
k =2.. r )} ,P ) ) :
r e t u r n Q:
end p r o c :
# p r o c e d u r e t o co m p u te c i r c ( P )
c i r c P := proc (P , r )
l o c a l Q:
Q: = ex p a n d ( s u b s ( { u [ 1 ] = u [ r ] , s e q ( u [ k ] = u [ k 1 ] , k = 2 . . r ) } , P ) ) :
r e t u r n Q:
end p r o c :
U: = [ s e q ( u [ i ] , i = 1 . . r + 1 ) ] :
with ( combinat ) :
# C r e a t i o n o f g e n e r i c p o l y n o m i a l o f deg n i n r v a r s w i t h
#indeterminate coefficients
Symmetries of Rational Functions Arising in Ecalle’s Study of Multiple Zeta Values 165

C1 : = p a r t i t i o n ( n+ r ) :
C2 : = [ ] :
f o r i f r o m 1 t o n o p s ( C1 ) do
i f ( n o p s ( C1 [ i ] ) = r ) t h e n C2 : = [ op ( C2 ) , op ( p e r m u t e ( C1 [ i ] ) ) ]
fi :
od :
C3 : = [ ] :
f o r i f r o m 1 t o n o p s ( C2 ) do
C3 : = [ op ( C3 ) , [ s e q ( C2 [ i ] [ j ] 1 , j = 1 . . r ) ] ]
od :
P:=0:
f o r i f r o m 1 t o n o p s ( C3 ) do
P : = P+a [ i ] p r o d u c t ( u [ j ] ^ C3 [ i ] [ j ] , j = 1 . . r ) :
od :
# Th at ’ s i t , P i s t h e g e n e r i c p o l y n o m i a l
#Now make l i n e a r s y s t e m f o r t h e c i r c  n e u t r a l i t y r e l a t i o n
#P+ c i r c ( P ) + . . . + c i r c ^ ( r 1 ) ( P ) = 0
PP [ 0 ] : = P :
f o r i f r o m 1 t o r 1 do PP [ i ] : = c i r c P ( PP [ i 1 ] , r ) : od :
Q: = add ( PP [ i ] , i = 0 . . r  1 ) :
COEFFScir cn eu t : = { c o e f f s (Q, U ) } :
#Now make l i n e a r s y s t e m f o r push  i n v a r i a n c e
Q: = ex p a n d ( Pp u sh P ( P , r ) ) :
COEFFSpushinv : = { c o e f f s (Q, U ) } :
S o l s : = s o l v e ( COEFFScir cn eu t u n i o n COEFFSpushinv ) :
P : = ex p a n d ( s u b s ( S o l s , P ) ) :
p r i n t ( " g e n e r i c push  i n v a r i a n t , c i r c  n e u t r a l p o l y n o m i a l
of
degree " ,n , " in " , r ," v ar i ab l es " ) :
print (P ):
# Test whether i t s a t i s f i e s the second a l t e r n a l i t y
relation
# c o r r e s p o n d i n g t o sh ( ( 1 , 2 ) , ( 3 , . . , r ) )
C4 : = c h o o s e ( r , 2 ) :
# build shuffle permutations
f o r i f r o m 1 t o b i n o m i a l ( r , 2 ) do
s h u [ i ] : = [ s e q (m, k = 1 . . r ) ] :
s h u [ i ] [ C4 [ i ] [ 1 ] ] : = 1 :
s h u [ i ] [ C4 [ i ] [ 2 ] ] : = 2 :
cc : = 3 :
f o r k f r o m 1 t o r do
i f ( s h u [ i ] [ k ] =m) t h e n s h u [ i ] [ k ] : = c c : c c : = c c + 1 : f i :
od :
od :
166 A. Salerno et al.

# build shuffle relation

Q: = 0 :
f o r i f r o m 1 t o b i n o m i a l ( r , 2 ) do
Q: =Q + s u b s ( { s e q ( u [ k ] = u [ s h u [ i ] [ k ] ] , k = 1 . . r ) } , P )
od :
Q: = ex p a n d (Q ) :
i f (Q=0 ) t h e n p r i n t ( " s a t i s f i e s s e c o n d s h u f f l e r e l a t i o n " )
fi :
i f (Q< >0) t h e n p r i n t ( " f a i l s s e c o n d s h u f f l e r e l a t i o n " ) f i :

Acknowledgements We would like to express our sincerest thanks to our group leader, Leila
Schneps, for all of her help and support and for suggesting such an intriguing problem. We would
also like to thank the conference organizers of “Women in Numbers—Europe 2013” for all their
hard work planning, funding, running, and following up on the wonderful conference that provided
us with such a stimulating work environment.

References

Bourbaki, N.: Éléments de Mathématique. Groupes et Algèbres de Lie: Chapitres 2 et 3, 2nd ed.,
Springer (2006)
Ecalle, J.: The flexion structure and dimorphy: flexion units, singulators, generators, and the
enumeration of multizeta irreducibles. In: Asymptotics in Dynamics, Geometry and PDEs;
Generalized Borel Summation, vol. II, CRM Series, Edizioni della Normale, Scuola Normale
Superiore Pisa, 27–211 (2011)
Euler, L.: Meditationes circa singulare serierum genus. Novi Commun. Acad. Sci. 15(20), 140–186
(1775)
Hoffman, M.E.: The algebra of multiple harmonic series. J. Algebra 194, 477–495 (1997)
Maple 17: Maplesoft, a division of Waterloo Maple Inc. (2013)
Rautenauer, C.: Free Lie Algebras. London Mathematical Society Monographs New Series
(Book 7). Oxford University Press, Oxford (1993)
Zagier, D.: Periods of modular forms, traces of Hecke operators, and multiple zeta values.
Sūrikaisekikenkyūsho Kōkyūroku 843, 162–170 (1993)
On -Li Coefficients for Rankin–Selberg
L-Functions

Alina Bucur, Anne-Maria Ernvall-Hytönen, Almasa Odžak,

Edva Roditty-Gershon, and Lejla Smajlović

Abstract The generalized -Li criterion for a certain zeta or L-function states
that non-negativity of -Li coefficients associated to this function is equivalent
to non-vanishing of this function in the region Re s > =2. For  2 Œ1; 2/ and
positive integers n, we define -Li coefficients n .   0 ; / associated to Rankin–
Selberg L-functions attached to convolutions of two cuspidal, unitary automorphic
representations  and  0 . We investigate their properties, including the archimedean
and non-archimedean terms, and the asymptotic behavior of these terms.

1 Introduction

A simple positivity criterion for the Riemann hypothesis, proved by Li (1997)

states that the Riemann hypothesis is equivalent to non-negativity of a sequence
of numbers
X   
1 n
.n/ D 1 1 ;

2Z./

A. Bucur ()
Department of Mathematics, 9500 Gilman Drive #0112, La Jolla, CA 92093, USA
e-mail: alina@math.ucsd.edu
A.-M. Ernvall-Hytönen
Department of Mathematics and Statistics, University of Helsinki, 00014 Helsinki, Finland
e-mail: ernvall@mappi.helsinki.fi
A. Odžak • L. Smajlović
Department of Mathematics, University of Sarajevo, Zmaja od Bosne 35,
71000 Sarajevo, Bosnia & Herzegovina
e-mail: almasa@pmf.unsa.ba; lejlas@pmf.unsa.ba
E. Roditty-Gershon
Raymond and Beverly Sackler School of Mathematical Sciences,
Tel Aviv University, Tel Aviv, Israel
e-mail: rodittye@post.tau.ac.il

© Springer International Publishing Switzerland 2015 167

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_7
168 A. Bucur et al.

where the sum runs over the set Z./ of non-trivial zeros of the Riemann zeta
functionPand in the sum indicates that it is taken in the sense of the limit
lim .
T!1 jImjT

The Li criterion is generalized to many classes of functions. For Dirichlet and

Hecke L-functions it is proved in Li (2004), for automorphic L-functions the Li
criterion is deduced in Lagarias (2007). In Smajlović (2010), a class S ][ that
contains both the Selberg class S and the class of automorphic L-functions is
introduced and the Li criterion for this class is obtained. For the Rankin–Selberg
L-functions the Li criterion is proved in Odžak and Smajlović (2010).
The Li coefficients for various classes of zeta and L-functions may be generalized
in different ways. Droll (2012), following Freitas (2006), defined for  2 Œ1; 2/
generalized -Li coefficients F .n; /, for F 2 S ][ as

X   n 
 
F .n; / D 1 (1)

2Z.F/

where the sum is taken over the set Z.F/ of all non-trivial zeros of F and proved that
non-negativity of -Li coefficients F .n; / for all positive integers n is equivalent
to non-vanishing of F in the region Res > =2, or, equivalently, in the region
Res < 1  =2.
We will refer to this criterion as -Li criterion. Let us note that the coefficients
F .n; / are a generalization of the Li coefficients F .n/, introduced in Smajlović
(2010), for F 2 S ][ , in the sense that F .n; 1/ D F .n/.
Another existing generalization of Li coefficients is given by Sekatskii (2013),
in the case of the Riemann zeta function. Namely, for an arbitrary real number a ¤
1=2, generalized Li coefficients are defined as

X   n 
 a
.n; a/ D 1 :
Ca1
2Z./

The generalized Li criterion for the Riemann zeta function, proved in Sekatskii
(2013) states that the Riemann hypothesis is equivalent to non-negativity of sums
.n; a/ for any a 2 R n f1=2g.
Rankin–Selberg L-function attached to the Rankin–Selberg convolution of two
cuspidal, unitary automorphic representations  and  0 of GLm .AF / and GLm0 .AF /
does not belong to the class S ][  S, since it might have poles at points on the line
Res D 1, different from s D 1 and its functional equation is such that, in general
case (if the Ramanujan hypothesis is not assumed) it might have trivial zeros inside
the critical strip. Therefore, the results of Droll (2012) do not apply in this setting.
In this paper, we define the analogue of generalized -Li coefficients ( 2 Œ1; 2/)
for the Rankin–Selberg L-function and prove -Li criterion for those functions.
Then, we deduce arithmetic formulas for generalized -Li coefficients. The gener-
alized -Li coefficient for the Rankin–Selberg L-function can be written as a sum of
On  -Li Coefficients for Rankin–Selberg L-Functions 169

two terms: archimedean term, coming from the archimedean part of the completed
L-function and non-archimedean term, coming from the finite part L-function. We
derive full asymptotic expansion of the archimedean part of the -Li coefficient
and investigate the asymptotic behavior of non-archimedean part, as n ! 1, for a
fixed .

2 Preliminaries

Let F be a number field of degree d D ŒF W Q, and let AF denote the ring of
adeles over F. For a unitary cuspidal automorphic representation  of GLm .AF /,
the completed L-function associated to  can be written as an absolutely convergent
Euler product over all places v of F
Y 
ƒ.s; / D L.s; v / D L s; f L .s; 1 /
v

inthe half-plane
Res > 1, see Jacquet and Shalika (1981a, p. 555, Th. 5.3). Here,
L s; f denotes the product over all finite places of F, whereas L .s; 1 / is the
product over infinite places. At the prime ideal p where p is unramified there is a
set of m non-zero Satake parameters f˛ .p; j/g such that

Y
m
L.s; p / D .1  ˛ .p; j/Nps /1 , (2)
jD1

where Np denotes the absolute norm of the ideal p.

At the prime ideal p where p is ramified, the local L-function is defined in terms
of the Langlands parameters of p . It can be expressed as P .Nps /1 , where P is
a polynomial
 of a degree at most m, equal to 1 at zero. Therefore, all local factors
of L s; f can be written in the form (2), with the convention that some of ˛ .p; j/
may be zero.
The local factor at infinite places is

Y
m
L .s; v / D v .s C  .v; j//,
jD1

where f .v; j/gm

jD1 are the Langlands parameters associated to v and v .s/ D
 s=2
.s=2/ D R .s/; if v is real and v .s/ D 2 .2/s .s/, if v is complex.
The Rankin–Selberg L-function attached to the product   Q 0 of two unitary
cuspidal automorphic representations of GLm .AF / and GLm0 .AF / is given, for Res >
1, by an absolutely convergent Euler product of local factors
170 A. Bucur et al.

 Y
L s; f  Q f0 D L.s; p  Q p0 /, (3)
p<1

as proved in Jacquet and Shalika (1981a, Th. 5.3). Here, Q denotes the contragra-
dient representation of . For any place v of F, Q v is equivalent to the complex



conjugate  v (Gelfand and Kazhdan 1974), hence L s; f  Q f0 D L s; Q f  f0 .
For prime ideals p at which p and p0 are unramified, the local factors are
given by

Y m

m Y 0
1
L.s; p  Q p0 / D 1  ˛ .p; j/˛ 0 .p; k/Nps :
jD1 kD1

At finite places p ramified for  or  0 the local L-function is defined in terms of the
Langlands parameters. It can be written as Q .Nps /1 , where Q is a polynomial of
a degree at most mm0 , equal to 1 at zero. Therefore, we can write all local factors at
finite places as
0
Y
m Y
m
L.s; p  Q p0 / D .1  ˛Q 0 .p; j; k/Nps /1 ;
jD1 kD1

with the convention that some ˛Q 0 .p; j; k/ might be equal to zero, where
˛Q 0 .p; j; k/ D ˛ .p; j/˛ 0 .p; k/, for the prime ideal p unramified for both 
and  0 .

The logarithmic derivative of L s; f  Q f0 , for Re s > 1; can be written as an
absolutely convergent series over all integral ideals n of the ring of integers OF of F

L0 X ƒ.n/aQ 0 .n/ X c;Q 0 .n/

 .s; f  Q f0 / D D ; (4)
L n
Nns n
Nns

where ƒ.n/ D log Np if n D pk , for some integer k  1, and ƒ.n/ D 0, otherwise.

Similarly, at infinite places v, the archimedean local factor L.s; v  Q v0 / can be
written as a product
0
Y
m Y
m
L.s; v  Q v0 / D v .s C Q 0 .v; j; k//,
jD1 kD1

where Q 0 .v; j; k/ D  .v; j/ C  0 .v; k/, at infinite places v unramified

for both  and  0 . Complex numbers Q 0 .v; j; k/ satisfy the trivial bound
ReQ 0 .v; j; k/ > 1 (see the calculations in Rudnick and Sarnak 1996,
Appendix).
On  -Li Coefficients for Rankin–Selberg L-Functions 171

Let us put
 Y
L s; 1  Q 1
0
D L.s; v  Q v0 /,
v2S1

where S1 denotes the set of all infinite places of F consisting of r1 real and 2r2
complex places. Then, as proved in Shahidi (1981), Shahidi (1984), Shahidi (1985),
Shahidi (1990), Jacquet and Shalika (1981a), Jacquet and Shalika (1981b), Moeglin
and Waldspurger (1989) and Gelbart and Shahidi (2001) (see also Cogdell 2007,
Ths. 9.1 and 9.2), the complete Rankin–Selberg L-function

ƒ.s;   Q 0 / D L.s; f  Q f0 /L.s; 1  Q 1

0
/

extends to a meromorphic function of order 1 on the whole complex plane, bounded

(away from its possible poles) in vertical strips. It has simple poles at s D 1Cit0 and
s D it0 , arising from L.s; f  Q f0 / if and only if m D m0 and  0 Š  ˝ jdetjit0 , for
some t0 2 R. Otherwise, it is a holomorphic function. Finally, ƒ.s;   Q 0 / satisfies
the functional equation
 1=2s
ƒ.s;   Q 0 / D    Q 0 QQ 0 ƒ.1  s; Q   0 /; (5)

where QQ 0 > 0 is the arithmetic conductor and  .  Q 0 / is a complex number

of modulus 1.
We call the zeros of ƒ.s;   Q 0 / the non-trivial zeros of L.s; f  Q f0 /. The set of
non-trivial zeros of L.s; f  Q f0 / is denoted by Z.L.s; f  Q f0 //, or shortly by Z.L/,
in the case when  and  0 are fixed.
By the functional equation and the Euler product representation, all those
zeros lie in the critical strip 0  Re s  1 (actually, Shahidi has proved that
L.1 C it; f  Q f0 / ¤ 0, for all t 2 R). Other trivial zeros arise from the
poles of the function L.s; 1  Q 1 0
/. In the critical strip, the trivial zeros of
L.s; f  Q f / are at the points s D Q 0 .v; j; k/, for those v 2 S1 , j and k
0

such that ReQ 0 .v; j; k/  0. Furthermore, the functional equation implies that
Z.L.s; f  Q f0 // D 1  Z.L.s; f  Q f0 //.
Let N; 0 .T/ denote the number of non-trivial zeros  of L.s; f  Q f0 / such that
jImj  T. Then,

dmm0
N; 0 .T/ D T log T C c; 0 T C O; 0 .log T/; as T ! 1; (6)

where

1 dmm0
c; 0 D log QQ 0  .1 C log 2/:
 
172 A. Bucur et al.

The complete proof is given in Iwaniec and Kowalski (2004, Th. 5.8) for the
counting function N; 0 .T/ of both trivial and (finitely many) non-trivial zeros in
the critical strip.
To be fully precise, in the case we consider, a slight modification of the argument
from Iwaniec and Kowalski (2004) should be made to include the case when the
L-function has poles at it0 and 1 C it0 , for some t0 2 Rn f0g. It is easily done by
considering the entire function

ƒc .s;   Q 0 / WD .s  it0 /ı; 0 .t0 / .s  1  it0 /ı; 0 .t0 / ƒ.s;   Q 0 / (7)

instead of .s .1  s//r ƒ.s;   Q 0 / (cf. Iwaniec and Kowalski 2004, (5.23)). Here,
we put

1; if m D m0 and  0 Š  ˝ jdetjit0 , for some t0 2 R;
ı; 0 .t0 / D
0; otherwise.
C
0 .T/ and N; 0 .T/ denote counting functions for the non-trivial zeros 

Let N;
of L.s; f  Q f / with 0 < Im  T and T  Im < 0, respectively. Repeating the
0

arguments given in Lagarias (2007, Th. 2.1) it can be easily shown that
 
˙ 1 dmm0
N; 0 .T/ D T log T C c ; 0 T C O; 0 .log T/; (8)
2 
as T ! 1.
Function ƒc .s;   Q 0 /, defined by (7) is an entire function of order one, non-
vanishing at s D 0, and hence possesses a representation as a Hadamard product

Y  
s s=
ƒ .s;   Q / D e
c 0 asCb
1 e ; (9)

2Z.L.s;Q 0 //

where eb D ƒc .0;   Q 0 /. Furthermore, in Odžak and Smajlović (2010), Propo-

sition 4.1, using an
Pexplicit formula for the Rankin–Selberg L-function it is proved
1
that the sum 
is -convergent and that
2Z.L.s;Q 0 //

X  1
aD : (10)

2Z.L.s;Q 0 //

3 -Li Coefficients for the Rankin–Selberg L-Function

Let  and  0 be arbitrary, but fixed unitary cuspidal automorphic representations

of GLm .AF / and GLm0 .AF /. In this section, we show that -Li coefficients for
the Rankin–Selberg L-function L.s; f  Q f0 / are well defined and we deduce two
formulas for evaluation of -Li coefficients.
On  -Li Coefficients for Rankin–Selberg L-Functions 173

In the sequel, we assume that representations  and  0 are fixed, therefore, to

shorten the notation, we denote

L.s/ D L.s; f  Q f0 /; G.s/ D L.s; 1  Q 1

0
/; ƒc .s/ D ƒc .s;   Q 0 /:

Furthermore, we put Z.L.s;   Q 0 // D Z.L/.

Repeating arguments given in Odžak and Smajlović (2010, Section 3) it is easy
to see that we may write G.s/ as

Y
dmm 0  
dmm0 s=2 s=2 s C .`/
G.s/ D  QQ 0  ; (11)
2
`D1

where .`/ D Q 0 .`/ D Q 0 .v; j; k/ for r1 C r2 places v 2 S1 and .`/ D

Q 0 .`/ D Q 0 .v; j; k/ C 1 for the rest of r2 places v 2 S1 (j D 1; : : : ; m,
k D 1; : : : ; m0 ).
Definition 3.1. Let  2 Œ1; 2/. For an arbitrary positive integer n, the nth -
Li coefficient associated to the Rankin–Selberg L-function L.s/ attached to the
product   Q 0 of two unitary cuspidal automorphic representations of GLm .AF /
and GLm0 .AF / is defined as

X  

n 
0
n .  Q ; / D 1 : (12)

2Z.L/

The following theorem shows that the coefficients n .  Q 0 ; / are well defined.
Theorem 3.2. The coefficients n .  Q 0 ; / introduced in Definition 3.1 have the
following properties
(1) The series defining n .  Q 0 ; / is *-convergent for every positive integer n.
(2)

X   n 
0 
Ren .  Q ; / D Re 1  ;

2Z.L/

where the sum on the right is absolutely convergent.

(3) The series
ˇ  ˇ
X 1 C ˇRe  ˇ

 ˇ  ˇ 2
1C  ˇ ˇ
2Z.L/

converges.
174 A. Bucur et al.

Proof. Claim .1/ for n  2 follows from the fact that ƒc .s/ is an entire function
ofPorder one. In the case when n D 1, the claim follows from the fact that the sum
1

is -convergent.
2Z.L/
The claim .2/ follows from the claim .3/ by using Bombieri and Lagarias (1999,
Lemma 1).
In order to prove claim (3), we need to prove that
ˇ  ˇ
X 1 C ˇRe  ˇ

 ˇ  ˇ 2
1 C ˇ ˇ
2Z.L/ 

ˇ ˇ  ˇˇ
converges. Estimate first ˇ1 C ˇRe  ˇˇ  2. Since ƒc .s/ is an entire function of
order one, the sum
X 1
 ˇ ˇ 2
2Z.L/ 1 C ˇˇ


converges, and the proof is complete. t

u
Theorem 3.3. The coefficients n .  Q 0 ; / can be expressed in the following form
 n 
1 d  n1
n .  Q 0 ; / D  s log ƒc
.s/ : (13)
.n  1/Š dsn sD

Proof. The proof closely follows the corresponding proof in Droll (2012,
Lemma 2.1.2), so we omit details. The right-hand side of (13) may be written as
 n 
1 d  n1
 s log ƒc
.s/
.n  1/Š dsn sD
!  
X
n1
n  nk dnk
D log ƒ .s/
c
:
kD0
k .n  1  k/Š dsnk sD

Let us next study the part

 
d nk
log ƒc .s/ :
dsnk sD

The Hadamard product representation (9), together with formula (10) implies that
0
ƒ c .s/ X  1
D ;
ƒc .s/ s
2Z.L/

and, analogously
On  -Li Coefficients for Rankin–Selberg L-Functions 175

  X
dnk  .n  k  1/Š
log ƒc .s/ D :
dsnk sD .  /nk
2Z.L/

We have now derived

 n 
1 d  n1
 s log ƒc
.s/
.n  1/Š dsn sD
!
X n
n1
 nk X  .n  k  1/Š
D
k .n  1  k/Š .  /nk
kD0 2Z.L/
!   n 
X X
n1
n  nk X  
D D 1 ;
k .  / nk 
2Z.L/
kD0 2Z.L/

which completes the proof. t

u
Next, we will give an alternate description of the -Li coefficients associated to
the Rankin–Selberg L-function.
Theorem 3.4. The coefficients n .  Q 0 ; / can be expressed in the following form
 
1 1
n .  Q 0 ; / D d n1 1  ; L ; (14)
n 

where dn .z0 ; L/ are

 1the
power series coefficients in the expansion of the logarithmic
1
derivative of ƒc 1s around z0 ¤ 1 which is not a zero of ƒc . 1s /; i.e., in a small
neighborhood of z0 we have
  X
1
d 1
log ƒc D dn .z0 ; L/.s  z0 /n :
ds 1s nD0

Proof. This proof also follows the corresponding proof in Droll’s thesis (2012,
1
Lemma 2.1.2). We start by noting that if 1s … Z.L/ and s ¤ 1, we can use (9)
to get
0 !1
  X
d 1 1 1 1
log ƒc D @a C C A
1s .1  s/2 1 
1s  
ds
2Z.L/

a X  1 1 1

D C C
.1  s/2 1  s 1   C s .1  s/2
2Z.L/

(15)

Let z0 D 1  1=. Using the following two identities (see Droll 2012, Lemma 2.1.2)
176 A. Bucur et al.

1  1 1 
D and D
1   C s    1   .sz0 / 1s 1  .s  z0 /


we may write (15) in the following form

!
a X 2 1 1 1
2
  .sz /
 :
.1  s/  1 0 1  .s  z0 / .1  s/2
2Z.L/ 

1 1
For s close enough to z0 we may write  .sz0 / and 1 .sz0 / as geometric series,
1 
hence we have (see Droll 2012, p. 59)
1   !
a X 2 X  n X1
1
 .s  z0 /n  m .s  z0 /m 
.1  s/2    nD0    .1  s/2
2Z.L/ mD0

X
1 X  nC1 ! !
 11
D  nC1
1 .s  z0 /  nC1
n

nD0 2Z.L/
 2 .1  s/2
a
C :
.1  s/2

We can now split the sum over Z.L/ into two -convergent sums, using
equation (10), to get

a a X 1 1

2
 2
.1  s/ 2.1  s/ nD0 2n
0 1
X 1 X   nC1 !

C  nC1 @ 1 A .s  z0 /n
  
nD0 2Z.L/

X
1 X
1
D  nC1 nC1 .  Q 0 ; /.s  z0 /n D dn .z0 ; L/.s  z0 /n ;
nD0 nD0

which by uniqueness of series expansions concludes the proof. t

u

4 Li-Type Criterion for the Zero-Free Regions

We are now ready to prove the Li-type criterion for zero-free regions in the case of
the Rankin–Selberg L-function.
On  -Li Coefficients for Rankin–Selberg L-Functions 177

Theorem 4.1. The function ƒc .s/ will have all its zeros in the strip 1 2  Res  
2
if and only if Ren .  Q 0 ; /  0 for every positive integer n.
Proof. The proof will also closely follow the proof of Droll (2012, Th. 2.1.3).
We have
X   n 

Ren .  Q 0 ; / D Re 1  :

2Z.L/

By Bombieri and Lagarias (1999, Th. 1), we know that if R is a multiset of complex
numbers with 1 62 R and
X 1 C jRe./j
< 1;
2R
.1 C jj/2

1
then the conditions Re  2 for every  2 R and

X    
1 n
Re 1  1  0
2R


are equivalent. Let us now put

n o
RD W  2 Z.L/ :


Then 1 62 R. Furthermore, by Theorem 3.2, we have

ˇ ˇ
X 1 C ˇRe  ˇ

 ˇ  ˇ 2 < 1;
ˇ ˇ
2Z.L/ 1 C 

so we may use Bombieri and Lagarias (1999, Th. 1). Hence, the conditions
Re   12 , i.e. Re  2 and

X    
1 n
Re 1  1  0
2R


are equivalent. Recalling the definition of our set R, we see that the condition
Re  2 is equivalent to

X     X   n 
 n 
Re 1  1  D Re 1   0:
 
2Z.L/ 2Z.L/
178 A. Bucur et al.

Recalling that Z.L/ D 1  Z.L/ completes the proof, since now bounds Re./  2
and Re./  1  2 are equivalent for all zeros  of ƒc .s/ (i.e., all non-trivial zeros
of L.s/). t
u

5 Arithmetic Formulas for -Li Coefficients

From the expression (13) for the nth  Li coefficient we easily get
!  k 
Xn
n  k1 d
n .  Q ; / D 
0
log ƒ .s/
c
: (16)
kD1
k .k  1/Š dsk sD

Now we wish to analyze this expression, and to separate the archimedean and non-
archimedean parts. We will treat the case  2 .1; 2/, since the case  D 1 was
thoroughly investigated in Odžak and Smajlović (2010).
Lemma 5.1. The following expressions hold true
8
ˆ P0
dmm

ˆ
ˆ dmm0 1 1 sC.`/
dk <  2 log  C 2 log QQ 0 C 2 2
; k D 1I
`D1
log G.s/ D dmm0

dsk ˆ 1 P
ˆ .k1/ sC.`/
; k > 1:
:̂ 2k 2
`D1

0
where .s/ is classical digamma function, i.e. .s/ D  .s/.

Proof. Definition (11) of G.s/ implies that

X1
dmm 0  
d dmm0 1 s C .`/
G.s/ D  log  C log QQ 0 C :
ds 2 2 2 2
`D1

Now it is clear that

X 1
dmm 0  
dk .k1/ s C .`/
log G.s/ D ;
dsk 2k 2
`D1

when k > 1. t
u
We are now ready to formulate and prove the arithmetic formula for the -Li
coefficients.
Theorem 5.2. For any positive integer n and any  2 .1; 2/ one has

n .  Q 0 ; / D S1 .n; / C SNA .n; /; (17)

On  -Li Coefficients for Rankin–Selberg L-Functions 179

where

S1 .n; / D n log QQ 0  dmm0 log 
2
! 0  
Xn
k X 1
dmm
n .k1/  C .`/
C (18)
k .k  1/Š 2k 2
kD1 `D1

and
  n  n 
it0 1 C it0
SNA .n; / D ı; 0 .t0 / 2  
it0   1 C it0  
!
X n
n k X c;N 0 .n/
 . log Nn/k1 ; (19)
kD1
k .k  1/Š n
Nn

where the coefficients c;N 0 .n/ are defined as the coefficients of the Dirichlet
series (4).
h k i
d
Proof. Let us start with formula (16). We need to find ds k log ƒc
.s/ : We have
sD

dk dk
log ƒc
.s/ D ı; 0 .t0 / log ..s  it0 /.s  1  it0 //
dsk dsk
dk dk
C k
log L.s/ C k log G.s/:
ds ds
The derivative involving the function G has already been evaluated in the previous
lemma, and it yields the total contribution
! 0  
n  Xn
n k X 1
dmm
 C .`/
0 .k1/
dmm log  C log QQ C
0
2 k .k  1/Š 2k 2
kD1 `D1

This is exactly the archimedean contribution S1 .n; /.

It thus suffices to concentrate on the other terms that yield the contribution of the
non-archimedean term SNA .n; /. Let us start with evaluating
 
dk
ı; .t0 / .log.s  it0 / C log.s  1  it0 //
0
dsk sD

D ı; 0 .t0 /.k  1/Š.1/ k1
.  it0 / C .  1  it0 /k :
k
180 A. Bucur et al.

Let us now calculate the total contribution of these terms. We have

!  
Xn
n k 1 1
ı; 0 .t0 /.k  1/Š.1/k1 C
kD1
k .k  1/Š .  it0 /k .  1  it0 /k
!  k  k !
Xn
n  
D ı; 0 .t0 /  
kD1
k   it0   1  it0
  n  n 
it0 1 C it0
D ı; .t0 / 2 
0  : (20)
it0   1 C it0  

dk
The only term that remains is dsk
log L.s/. Differentiating (4) we get

  " #
dk dk1 X c;N 0 .n/
log L.s/ D
dsk sD dsk1 n Nns
sD
" #
X c;N 0 .n/
D  . log Nn/ k1

n
Nns
sD
X c;N 0 .n/
D  . log Nn/k1
n
Nn

Substituting this to the sum defining n .  Q 0 ; /, together with (20) yields the
expression for SNA .n; /. The proof is complete. u
t

6 Asymptotic Behavior of -Li Coefficients

In this section we investigate the asymptotic behavior of the nth -Li coefficient
attached to a Rankin–Selberg L-function, as n ! 1, while  2 Œ1; 2/ is fixed.
We will separately investigate the asymptotic behavior of archimedean and non-
archimedean contribution in the arithmetic formula (17).

6.1 Evaluation of the Archimedean Part of -Li Coefficient

For the evaluation of the archimedean part (18) it is useful to write it in terms of the
Hurwitz zeta function. First, we use a recurrence relation for the digamma function
(Abramowitz and Stegun 1964, 6.4.6)
On  -Li Coefficients for Rankin–Selberg L-Functions 181

.n/ .n/
.z C 1/ D .z/ C .1/n nŠzn1 ;

for n  0, to make a variable shift and then the fact that (Abramowitz and Stegun
1964, 6.4.10)
.n/
.z/ D .1/nC1 nŠ.n C 1; z/;

for z ¤ 0; 1; 2; : : : to get

! 0  
Xn
n k X 1
dmm
 C .`/
.k1/
k .k  1/Š 2k 2
kD1 `D1
! 0  
Xn
n k X 1
dmm
 C .`/ C 2
.k1/
D
k .k  1/Š 2k 2
kD1 `D1

X
dmm0
.`/
n
C  dmm0
 C .`/
`D1
  X ! dmm0  
n
 k X
dmm0
 X  0  C .`/ C 2
n
 C .`/ C 2
Dn C   k;
2  2 k 2 2
`D1 kD2 `D1

X
dmm 0  n
.`/
C  dmm0 ;
 C .`/
`D1

thus the archimedean contribution is

X  0   C .`/ C 2 
0
n  n dmm
S1 .n; / D log QQ 0  dmm log  C
0
2 2  2
`D1
!  
n
 k X
Xn dmm0
 C .`/ C 2
C  k;
k 2 2
kD2 `D1

X
dmm0
.`/
n
C  dmm0 : (21)
 C .`/
`D1

The following theorem gives us the full asymptotic expansion of the archimedean
contribution to the nth -Li coefficient. The proof follows the lines of the analogous
theorem proved in Odžak and Smajlović (2010).
Theorem 6.1. Let  and  0 be two automorphic unitary cuspidal representations of
GLm .AF / and GLm0 .AF /, respectively. Then, for  2 .1; 2/ and an arbitrary K 2 N
182 A. Bucur et al.


S1 .n; / D dmm0 n log n
2
n

 
C log QQ 0 C dmm0 log C 1
2 2
X
0  n dmm0
1X
dmm
 2 .`/
C 0
dmm C C .`/
4  C .`/ 2
`D1 `D1

 X
K
B2k 2kC1
dmm0 n C OK .n2K /;
2 kD1
2k

as n ! 1, where B2k are the Bernoulli numbers.

Proof. Let us first investigate the sum
!   dmm
n
 k
0 0
X
dmm Xn
 C .`/ C 2 X
 k; D S1 .n; `/;
k 2 2
`D1 kD2 `D1

appearing in (21).
Calculus of residues implies that
Z
.1/n
S1 .n; `/ D nŠ f` .s/ds; (22)
2i R

where R is positively oriented rectangle with vertices at points 32 ˙ i and n C 1

2 ˙i
and
 
.s  n/
 s  C .`/ C 2
f` .s/ D  s; :
.s C 1/ 2 2

Namely, poles of the function f` .s/ inside R are simple poles at s D 2; 3; : : : ; n
which come from the gamma function .s  n/, since functions 2 and
s


 k;  C.`/C2
2
are holomorphic in R. Residues are easily found using the fact
.1/n
that RessDn .s/ D nŠ
, thus
 
1
 k  C .`/ C 2 .1/nk
RessDk f` .s/ D  k; ;
kŠ 2 2 .n  k/Š

for k D 2; 3; : : : ; n, and hence (22) holds true.

The function f` .s/ is uniformly bounded on the real segment joining n C 12 and
e , hence, the rectangle R can be deformed to the line .en  i1; en C i1/. Further
n

singularities of the function f` .s/ are a simple pole at s D 0 and a pole s D 1 of

order 2. Therefore, we get
On  -Li Coefficients for Rankin–Selberg L-Functions 183

S1 .n; `/ D .1/n1 nŠ .RessD0 f` .s/ C RessD1 f` .s//

 Z 1ˇ ˇ  ˇ ˇ 
ˇ .en C it  n/ ˇ
 en ˇ ˇ
n ˇ 2 C  C .`/ ˇ
CO nŠ ˇ ˇ  e ;
ˇ .en C it C 1/ ˇ 2 ˇ 2 ˇ dt
1

(23)

Residue at s D 0 is simple and given by

 
 C .`/ C 2  C .`/ C 1
.1/n1 nŠRessD0 f` .s/ D  0; D ;
2 2

since .0; x/ D 12  x.
Residue at s D 1 can be found using Laurent series representations of the factors
appearing in f` .a/. Namely,
   
2 C  C .`/ 1  0 2 C  C .`/
 s; D  C
2 s1  2

 s 
 
D C log .s  1/ C   
2 2 2 2
1
D 1 C .  1/.s  1/ C   
.s C 1/
.1/n1 1 .1/n1  0
.s  n/ D C .n/ C    :
.n  1/Š s  1 .n  1/Š 

and thus
   
n 0 0 2 C  C .`/ 
.1/n1 nŠRessD1 f` .s/ D .n/  C log C 1 :
2   2 2

Asymptotic expansion of the digamma function appearing in the above residue is

given by the Stirling formula for the digamma function (Abramowitz and Stegun
1964, 6.3.18)

0 1 X B2k K
.n/ D log n   n2k C OK .n12K /;
 2n kD1 2k

as n ! C1, thus
  
 n  0 2 C  C .`/
.1/n1 nŠRessD1 f` .s/ D n log n C 
2 2  2

    X B2k 2kC1
K
C log C 1   n C OK .n2K /;
2 4 2 kD1 2k

as n ! 1.
184 A. Bucur et al.

For the estimation of the integral appearing in (23), let us notice that
ˇ
 n  ˇ ˇˇ
ˇ  e ˇ ˇ ˇ
n ˇ 2 C  C .`/ ˇ ˇ
ˇ  ;
ˇ 2 e ˇ 2 ˇ ˇ D o.1/;

as n ! 1, since  2 Œ1; 2/ and Re.`/ > 1 for all ` D 1; : : : ; dmm0 .

The part containing the gamma functions can be modified using the reflection
formula and a functional equation for the gamma function
ˇ ˇ ˇ ˇ
ˇ  .en C it  n/ ˇ ˇ.1/n1 ˇ
ˇ ˇ
ˇ  .en C it C 1/ ˇ D Qn jen C it  n C jj
jD0

and it decays rapidly enough since

Z ˇ ˇ Z 1
ˇ .en C it  n/ ˇ
1
dt
ˇ ˇ
nŠ ˇ ˇ dt D 2nŠ Q
(24)
1 .e C it C 1/
n n
0
..en C j  n/2 C t2 /1=2
jD0
Z 1
dt
 2nŠ
0 ..en  n/2 C t2 /n=2
Z 1
2nŠ dt
D n
.e  n/ n1
0 .1 C t2 /n=2
.n C 1/nC1=2 e.nC1/
  en  n2K ;
.en  n/n1

as n ! 1, for all K 2 N, by the Stirling approximation formula for the gamma

function (Abramowitz and Stegun 1964, 6.1.37). Now,

n

0
X
dmm
 
S1 .n; `/ D dmm0 n log n C dmm0 log C   1 
2 2 2
`D1

dmm0  
n X  0 2 C  C .`/  C2
 C dmm0 C
2  2 4
`D1
0
1X  X B2k 2kC1
dmm K
C .`/  dmm0 n C OK .n2K /;
2 2 kD1 2k
`D1

as n ! 1. Combining obtained result with (21) completes the proof. t

u
On  -Li Coefficients for Rankin–Selberg L-Functions 185

6.2 Evaluation of the Non-archimedean Part

of the Rankin–Selberg -Li Coefficient

For the evaluation of the non-archimedean part of a -Li coefficient attached

to a Rankin–Selberg L-function let us notice that its main part comes from the
logarithmic derivative of the Rankin–Selberg L-function evaluated at . It can be
written in the form of a Dirichlet series as in (19), or in terms of coefficients
appearing in the Taylor (Laurent) series representation of L0 =L-functions around
. Namely, for  2 .1; 2/ (case  D 1 is treated in Odžak and Smajlović 2010),  is
not a pole of the L-function, thus

L0 X 1
.s/ D  .l/.s  /l ;
L lD0

or, equivalently

L0 X 1
.s C / D  .l/sl : (25)
L lD0

Differentiation, for k  1 implies

 .k1/ X
1
L0
.s C / D  .l/l.l  1/ : : : .l  k C 2/slkC1 ;
L lD0

and passing to the limit as s ! 0C , we obtain

 .k1/
L0
./ D .k  1/Š .k  1/:
L

Now, the non-archimedean contribution can be written in the form

!
Xn
n k
SNA .n; / D   .k  1/
kD1
k
  n  n 
it0 1 C it0
Cı; .t0 / 2 
0  ; (26)
it0   1 C it0  

which will be used for the further asymptotic expansion. Namely, we will prove that
this contribution to the nth -Li coefficient
p can be written in
pterms of the incomplete
-Li coefficient up to the height n and the error term O. n log n/.
186 A. Bucur et al.

Incomplete -Li coefficient up to the height T is denoted by n .  Q 0 ; ; T/ and

defined by
X  

n 
n .  Q ; ; T/ D
0
1 :

2Z.L/
jImj<T

We will skip  and  0 in the notation of the incomplete -Li coefficient up to the
height T and denote it simply by n .; T/.
Theorem 6.2. Let  and  0 be two automorphic unitary cuspidal representations
of GLm .AF / and GLm0 .AF /, respectively, and  2 .1; 2/. Then,
p p
SNA .n; / D n .; N/ C O. n log n/:

Proof. For the proof we will use contour integration along a suitably chosen
rectangle and approximation of the integrals along its sides. The proof follows the
lines of the proofs of corresponding theorems in Lagarias (2007) and Odžak and
Smajlović (2010), with a slightly modified integrand. Let
!

 n Xn
n
 j
kn .s/ D 1 C 1D ;
s jD1
j s

0
and let us integrate the function g.s/ D kn .s/ LL .s C / over the p rectangle R.n/
formed by the p lines Res D 0 for 3 < 0 < 2, Res D 2 n, Ims D ˙T
where T D n C "n ; 0 < "n < 1 is such that the horizontal lines Ims D ˙T
do not approach closer than O. log1 n / to any zero of L.s/. This is possible due to
equation (6).
Poles of g.s/ inside R.n/ are at s D 0, at points that correspond to trivial
zeros (denoted by
) and non-trivial zeros (denoted by ) of the Rankin–Selberg
L-function, and at points corresponding to possible poles of L.s/ at 1 C it0 and it0 .
When ˇ D
or ˇ D ; residues of g.s/ at simple poles ˇ   are given by
 n  n
 ˇ
RessDˇ g.s/ D 1 C 1 D  1:
ˇ ˇ

When ı; 0 .t0 / ¤ 0 residues at 1 C it0   and it0   are

 n  n
1 C it0 it0
RessD1Cit0  g.s/ D 1  ; RessDit0  g.s/ D 1  :
1 C it0   it0  

Residue at s D 0 is easily found using (25); it is equal to

!
Xn
n k
RessD0 g.s/ D   .k  1/:
kD1
k
On  -Li Coefficients for Rankin–Selberg L-Functions 187

Cauchy residue theorem implies

1
Z X   n  X 

n 
g.s/ds D 1 C 1
2i R.n/ 


2Z.L/
jImj<T
  n  n 
1 C it0 it0
Cı; 0 .t0 / 2  
1 C it0   it0  
!
Xn
n k
C   .k  1/ (27)
kD1
k

Using arguments similar to ones used in Lagarias (2007) it is easy to conclude

that the contribution from the trivial zeros is O.1/, since  2 Œ1; 2/. Since the sum
over  is equal to an incomplete -Li coefficient up to a height T, multiplied by
.1/, we get
Z
1
g.s/ds D SNA .n; /  n .; T/ C O.1/: (28)
2i R.n/

It is left to estimate the integral on the left side of the above equation. It is done
analogously as in Odžak and Smajlović (2011), using the approximation

L.s/0 X 1
D C O; 0 .log jsj/ ;
L.s/ 
s

which implies that

L.s/0 X 1
D C O; 0 .log T/ ;
L.s/ s
jTImj1

for s D  CiT, uniformly in 1    2, as noticed in Odžak and Smajlović (2010).

Last observation, together with formula (6) for the number of non-trivial zeros of
the Rankin–Selberg L-function up to a height T and the bound on the length of the
sides of rectangle R.n/ yields the bound
Z
1 p
g.s/ds D O. n log n/:
2i R.n/

Inserting the above bound into (28) completes the proof. t

u
188 A. Bucur et al.

7 Further Research

Results presented in this paper may be generalized to different classes of functions.

Besides that, various numerical computations might be conducted in order to pose
conjectures related to the asymptotic behavior of generalized -Li coefficients,
as n ! 1, for different values of the parameter  and for different classes of
L-functions.
In the paper Ernvall-Hytönen et al. (2015) we define, for 0 ; 1 2 R a very broad
class S ][ .0 ; 1 / of functions F satisfying following four conditions:
(i) (Dirichlet series) The function F possesses a Dirichlet series representation that
converges absolutely for Res > 0 :
(ii) (Analytic continuation) There exist finitely many non-negative integers
Q
n
m1 ; : : : ; mn and complex numbers s1 ; : : : ; sn such that function .s  si /mi F.s/
jD1
is entire function of finite order.
(iii) (Functional equation) The function F satisfies the functional equation

F .s/ D w F .1  sN/;

where
2MCı.1 /
Y
r Y
F .s/ D F.s/QsF  .j s C j / .s  si /mi
jD1 iD1

Y
N
.s  si /mi .1  s  si /mi ;
iD2MC1Cı.1 /

with jwj D 1, QF > 0, r  0, j > 0, j 2 C, j D 1; : : : ; r. We assume that

poles of F are arranged so that the first 0  2M C ı.1 /  N poles are such
that s2j1 C s2j D 1 , for j D 1; : : : ; M, and ı.1 / D 1 if 1 =2 is a pole of F in
which case s2MCı.1 / D 1 =2; otherwise ı.1 / D 0.
(iv) (Euler sum) The logarithmic derivative of the function F possesses a Dirichlet
series representation

F0 X1
cF .n/
.s/ D  ;
F nD2
ns

converging absolutely for Res > 0 .

The class S ][ .0 ; 1 / contains the class S ][ as well as Rankin–Selberg
L-functions. Besides classical L-functions, the class S ][ .0 ; 1 / also contains all
finite products of complex shifts of functions from the class S ][ .
On  -Li Coefficients for Rankin–Selberg L-Functions 189

We show that for   1 the generalized -Li coefficients are well defined for all
F 2 S ][ .0 ; 1 / and derive an analogue of the -Li criterion for zero-free regions.
Furthermore, we derive arithmetic formulas for generalized -Li coefficients and
deduce their asymptotic behavior.
We also conduct numerical computations in order to evaluate -Li coefficients for
finite products of shifts of Riemann zeta functions and deduce certain conjectures
on the asymptotic behavior of -Li coefficients, as n ! 1, for different values of
  1, depending on the real parts of shifts.
In the paper Bucur et al. (2015), we define generalized -Li coefficients for
]
functions F in the class SR that consists of all functions F belonging to the extended
]
Selberg class S , introduced in Kaczorowski and Perelli (1999) with the property
]
that  is a zero of F if and only if 1   is. Functions from the class SR do not
necessarily satisfy the generalized Riemann hypothesis. For example, Davenport-
]
Heilbronn L-function (introduced in Davenport and Heilbronn 1936) belongs to SR
and possesses infinitely many zeros in the half-plane Res > 1.
]
We prove that generalized -Li coefficients are well defined for functions in SR
and derive an arithmetic formula for their computation. The asymptotic behavior of
generalized -Li coefficients as n ! 1, and as  ! 1 (since in this case we might
have no zero-free regions) will be a subject of our future investigation. We intend
to conduct a numerical investigation using Davenport-Heilbronn type L-functions
(see Bombieri and Ghosh (2011)) in order to conjecture asymptotic behavior of
generalized -Li coefficients as  ! 1, in the case when the function violates
Riemann hypothesis, but still possesses a positive proportion of zeros on the critical
line.

Acknowledgements The authors of the paper would like to thank the organizers of the WINE
conference. The conference was funded by CIRM, Microsoft research, Number theory foundation,
NSF and Clay Mathematics Institute. Their support is gratefully acknowledged. The research of
A.-M. E.-H. was supported by the Academy of Finland, grant no. 138337. The research of A.B.
was supported by Simons Foundation Award #244988.

References

Abramowitz, M., Stegun, I.: Handbook of Mathematical Functions, With Formulas, Graphs, and
Mathematical Tables. NBS Applied Mathematics Series 55, National Bureau of Standards,
Washington, DC (1964)
Bombieri, E., Ghosh, A.: Around Davenport-Heilbronn function. Uspekhi Math. Nauk 66, 15-66
(2011) [translated in Russ. Math. Surv. 66, 221–270 (2011)]
Bombieri, E., Lagarias, J.C.: Complements to Li’s criterion for the Riemann hypothesis. J. Number
Theory 77, 274–287 (1999)
Bucur, A., Ernvall-Hytönen, A.-M., Odžak, A., Smajlovi ´ c, L.: On a Li-type criteria for-zero free
regions of certain Dirichlet series with real coefficients (2015, in preparation)
Cogdell, J.W.: L-functions and converse theorems for GLn , Automorphic forms and applications,
IAS/Park City Math. Ser. 12, Amer. Math. Soc, Providence, RI, 2007, 97–177
190 A. Bucur et al.

Davenport, D., Heilbronn, H.: On the zeros of certain Dirichlet series II. J. Lond. Math. Soc. 11,
307–312 (1936)
Droll, A.D.: Variations of Li’s criterion for an extension of the Selberg class. Ph.D. thesis, Queen’s
University, Kingston (2012)
Ernvall-Hytönen, A.-M., Odžak, A., Smajlović, L., Sušić, M.: On the modified Li criterion for a
certain class of L-functions. J. Number Theory 156, 340–367 (2015)
Freitas, P.: A Li-type criterion for zero-free half-planes of Riemann’s zeta function. J. Lond. Math.
Soc. 73, 399–414 (2006)
Gelbart, S., Shahidi, F.: Boundedness of automorphic L-functions in vertical strips. J. Amer. Math.
Soc. 14, 79–107 (2001)
Gelfand, I.M., Kazhdan, D.: Representation of the group GL.n; K/, where K is a local field. In:
I.M. Gelfand (ed.) Lie Groups and Their Representations, pp. 95–118. Wiley, New York (1974)
Iwaniec, H., Kowalski, E.: Analytic Number Theory. American Mathematical Society Colloquium
Publications, vol. 53. American Mathematical Society, Providence (2004)
Jacquet, H., Shalika, J.A.: On Euler products and the classification of automorphic representations
I. Am. J. Math. 103, 499–558 (1981a)
Jacquet, H., Shalika, J.A.: On Euler products and the classification of automorphic representations
II. Am. J. Math. 103, 777–815 (1981b)
Kaczorowski, J., Perelli, A.: On the structure of the Selberg class, I: 0  d  1. Acta Math. 182,
207–241 (1999)
Lagarias, J.C.: Li’s coefficients for automorphic L-functions. Ann. Inst. Fourier 57, 1689–1740
(2007)
Moeglin, C., Waldspurger, J.-L.: Le spectre résiduel de GL.n/. Ann. Sci. École Norm. Sup. 22,
605–674 (1989)
Li, X.-J.: The positivity of a sequence of numbers and the Riemann hypothesis. J. Number Theory
65, 325–333 (1997)
Li, X.-J.: Explicit formulas for Dirichlet and Hecke L-functions. Ill. J. Math. 48, 491–503 (2004)
Odžak, A., Smajlović, L.: On Li’s coefficients for the Rankin-Selberg L-functions. Ramanujan J.
21, 303–334 (2010)
Odžak, A., Smajlović, L.: On asymptotic behavior of generalized Li coefficients in the Selberg
class. J. Number Theory 131, 519–535 (2011)
Rudnick, Z., Sarnak, P.: Zeros of principal L-functions and random matrix theory. Duke Math. J.
81, 269–322 (1996)
Sekatskii, S.K.: Generalized Bombieri-Lagarias’ theorem and generalized Li’s criterion (2013)
[arXiv:1304.7895]
Shahidi, F.: On certain L-functions. Am. J. Math. 103, 297–355 (1981)
Shahidi, F.: Fourier transforms of intertwining operators and Plancherel measures for GL.n/. Am.
J. Math. 106, 67–111 (1984)
Shahidi, F.: Local coefficients as Artin factors for real groups. Duke Math. J. 52, 973–1007 (1985)
Shahidi, F.: A proof of Langlands’ conjecture on Plancherel measures. Complementary series for
p-adic groups. Ann. Math. 132, 273–330 (1990)
Smajlović, L.: On Li’s criterion for the Riemann hypothesis for the Selberg class. J. Number Theory
130, 828–851 (2010)
Galois Representations and Galois Groups
Over Q

Sara Arias-de-Reyna, Cécile Armana, Valentijn Karemaker,

Marusia Rebolledo, Lara Thomas, and Núria Vila

Abstract In this paper we generalize results of P. Le Duff to genus n hyperelliptic

curves. More precisely, let C=Q be a hyperelliptic genus n curve, let J.C/ be
the associated Jacobian variety, and let N` W GQ ! GSp.J.C/Œ`/ be the Galois
representation attached to the `-torsion of J.C/. Assume that there exists a prime
p such that J.C/ has semistable reduction with toric dimension 1 at p. We provide
an algorithm to compute a list of primes ` (if they exist) such that N` is surjective.
In particular we realize GSp6 .F` / as a Galois group over Q for all primes ` 2
Œ11; 500;000.

1 Introduction

In this paper we present the work carried out at the conference Women in numbers—
Europe (October 2013), by the working group Galois representations and Galois
groups over Q. Our aim was to study the image of Galois representations attached
to the Jacobian varieties of genus n curves, motivated by the applications to the
inverse Galois problem over Q. In the case of genus 2, there are several results in
this direction (e.g., Le Duff 1998; Dieulefait 2002a), and we wanted to explore the
scope of these results.

S. Arias-de-Reyna
Mathematical Research Unit, University of Luxembourg, Luxembourg, Luxembourg
C. Armana
Laboratory of Mathematics, University of Franche-Comteé, Besançon, France
V. Karemaker
Mathematical Institute, Utrecht University, Utrecht, The Netherlands
M. Rebolledo
Laboratory of Mathematics, Blaise Pascal University, Clermont-Ferrand, Aubière, France
L. Thomas
Pure and Applied Mathematics Unit, ENS Lyon, Lyon, France
N. Vila ()
Department of Algebra and Geometry, University of Barcelona, Barcelona, Spain
e-mail: nuriavila@ub.edu

© Springer International Publishing Switzerland 2015 191

M.J. Bertin et al. (eds.), Women in Numbers Europe, Association for Women
in Mathematics Series 2, DOI 10.1007/978-3-319-17987-2_8
192 S. Arias-de-Reyna et al.

Our result is a generalization of P. Le Duff’s work to the genus n setting, which

allows us to produce realizations of groups GSp6 .F` / as Galois groups over Q, for
infinite families of primes ` (with positive Dirichlet density). These realizations are
obtained through the Galois representations N` attached to the `-torsion points of
the Jacobian of a genus 3 curve.
The first section of this paper contains a historical introduction to the inverse
Galois problem and some results obtained in this direction by means of Galois
representations associated with geometric objects. Section 3 presents some theoretic
tools, which we collect to prove a result, valid for a class of abelian varieties A of
dimension n, that yields primes ` for which we can ensure surjectivity of the Galois
representation attached to the `-torsion of A (see Theorem 3.10). In Section 4, we
focus on hyperelliptic curves and explain the computations that allow us to realize
GSp6 .F` / as a Galois group over Q for all primes ` 2 Œ11; 500;000.

2 Images of Galois Representations and the Inverse Galois

Problem

One of the main objectives in algebraic number theory is to understand the absolute
Galois group of the rational field, GQ D Gal.Q=Q/: We believe that we would get
all arithmetic information if we knew the structure of GQ . This is a huge group, but
it is compact with respect to the profinite topology. Two problems arise in a natural
way: on the one hand, the identification of the finite quotients of GQ , and on the
other hand, the study of GQ via its Galois representations.
The inverse Galois problem asks whether, for a given finite group G, there exists a
Galois extension L=Q with Galois group isomorphic to G. In other words, whether
a finite group G occurs as a quotient of GQ . As is well known, this is an open
problem. The origin of this question can be traced back to Hilbert. In 1892, he
proved that the symmetric group Sn and the alternating group An are Galois groups
over Q, for all n. We also have an affirmative answer to the inverse Galois problem
for some other families of finite groups. For instance, all finite solvable groups and
all sporadic simple groups, except the Mathieu group M23 , are known to be Galois
groups over Q.
A Galois representation is a continuous homomorphism

 W GQ ! GLn .R/;

where R is a topological ring. Examples for R are C, Z=nZ or Fq with the discrete
topology, and Q` with the `-adic topology. Conjectures by Artin, Serre, Fontaine-
Mazur, and Langlands, which have experienced significant progress in recent years,
are connected with these Galois representations.
Galois Representations and Galois Groups Over Q 193

Since GQ is compact, the image of  is finite when the topology of R is discrete.

As a consequence, images of Galois representations yield Galois realizations over
Q of finite linear groups
ker 
Gal.Q =Q/ ' .GQ /  GLn .R/:

This gives us an interesting connection between these two questions and provides
us with a strategy to address the inverse Galois problem.
Let us assume that  is an `-adic Galois representation associated with some
arithmetic-geometric object. In this case, we have additional information on the
ramification behavior, like the characteristic polynomial of the image of the Frobe-
nius elements at unramified primes or the description of the image of the inertia
group at the prime `. This gives us some control on the image of mod ` Galois
representations in some cases and we can obtain, along the way, families of linear
groups over finite fields as Galois groups over Q.
More precisely, let X=Q be a smooth projective variety and let

` W GQ ! GL.Hétk .XQ ; Q` //;

be the `-adic Galois representation on the k-th étale cohomology. We know that:
• ` is unramified away from ` and the primes of bad reduction for X,
• if p is a prime of good reduction and p ¤ `, the characteristic polynomial of
` .Frobp / has coefficients in Z, is independent of ` and its roots have absolute
value pk=2 .
Let us consider an attached residual Galois representation

  W GQ ! GLn .F`r /;

where  is a prime in a suitable number field, dividing ` and r  1 an integer. To

determine the image of   , we usually need to know the classification of maximal
subgroups of GLn .F`r /, as well as a description of the image of the inertia group at `
and the computation of the characteristic polynomial of   .Frobp /, for some prime
of good reduction p ¤ `.
Let us summarize the known cases of realizations of finite linear groups as Galois
groups over Q, obtained via Galois representations.
In the case of two-dimensional Galois representations attached to an elliptic
curve E defined over Q without complex multiplication, we know, by a celebrated
result of (Serre 1972), that the associated residual Galois representation is surjective,
for all but finitely many primes. Moreover, it can be shown that if we take, for
example, the elliptic curve E defined by the Weierstrass equation Y 2 C Y D X 3  X,
then the attached residual Galois representation is surjective, for all primes `.
Thus we obtain that the group GL2 .F` / occurs as a Galois group over Q, for all
194 S. Arias-de-Reyna et al.

primes `. Actually we have additional information in this case: the Galois extension
Q.EŒ`/=Q is a Galois realization of GL2 .F` /, and it is unramified away from 37
and `, since E has conductor 37.
The image of two-dimensional Galois representations, attached to classical
modular forms without complex multiplication, has been studied by Ribet (1975).
The image of the residual Galois representations attached to a normalized cuspidal
Hecke eigenform without complex multiplication is as large as possible, for all
but finitely many primes . This gives us that the groups PSL2 .F`r / or PGL2 .F`r /
can occur as Galois groups over Q. Moreover, we have effective control of primes
with large image for the mod ` Galois representation attached to specific modular
forms. This gives us Galois realizations over Q of the groups PSL2 .F`r /, r even, and
PGL2 .F`r /, r odd; 1  r  10, for explicit infinite families of primes `, given by
congruence conditions on ` (cf. Reverter and Vila 1995; Dieulefait and Vila 2000).
Recently, it has been proven that the groups PSL2 .F` / are Galois groups over
Q for all ` > 3, by considering the Galois representations attached to an explicit
elliptic surface (see Zywina 2014).
Results on generically large image of compatible systems of three-dimensional
Galois representations associated with some smooth projective surfaces and with
some cohomological modular forms are obtained in Dieulefait and Vila (2004).
The effective control of primes with large image for the residual three-dimensional
Galois representations attached to some explicit examples gives us that the groups
PSL3 .F` /, PSU3 .F` /, SL3 .F` /, SU3 .F` / are Galois groups over Q, for explicit
infinite families of primes ` (cf. Dieulefait and Vila 2004).
In the case of four-dimensional Galois representations, we have results on large
image for compatible systems of Galois representations attached to abelian surfaces
A defined over Q such that EndQ .A/ D Z, to Siegel modular forms of genus two and
to some pure motives (cf. Le Duff 1998; Dettweiler et al. 2001; Dieulefait 2002b;
Dieulefait and Vila 2011). The effective control of primes with large image in some
explicit cases gives us that the groups PGSp4 .F` /, for all ` > 3; and the groups
PGSp4 .F`3 /, PSp4 .F`2 /, PSL4 .F` / and PSU4 .F` /, for explicit infinite families of
primes `, are Galois groups over Q (cf. Arias-de-Reyna and Vila 2011; Dettweiler
et al. 2001; Dieulefait 2002b; Dieulefait and Vila 2008).
In the next section we consider the image of residual Galois representations
attached to principally polarized abelian varieties of dimension n, which provides
Galois realizations over Q of the general symplectic group GSp2n .F` /, for almost
all `.
Finally, we remark that, using these methods, we can expect to obtain realizations
of the groups PSL2 .F`r /, PGL2 .F`r /, PGSp2n .F`r /, and PSp2n .F`r / as Galois groups
over Q. In fact, by considering compatible systems of Galois representations
attached to certain automorphic forms, we know (cf. Wiese 2008; Dieulefait and
Wiese 2011; Khare et al. 2008; Arias-de-Reyna et al. 2013) that these groups are
Galois groups over Q, for infinitely many integers r and infinitely many primes `.
More precisely, we have:
Galois Representations and Galois Groups Over Q 195

• “Vertical direction”: For every fixed prime `, there are infinitely many positive
integers r, such that PSL2 .F`r / can be realized as a Galois group over Q.
Moreover, for each n  2, there are infinitely many positive integers r, such
that either PGSp2n .F`r / or PSp2n .F`r / are Galois groups over Q (cf. Wiese 2008;
Khare et al. 2008).
• “Horizontal direction”: For every fixed r, there is a positive density set of primes `,
such that PSL2 .F`r / can be realized as a Galois group over Q. Moreover, for each
n  2, there is a set of primes ` of positive density for which either PGSp2n .F`r /
or PSp2n .F`r / are Galois groups over Q (cf. Dieulefait and Wiese 2011; Arias-de-
Reyna et al. 2013).

3 Galois Representations Attached to Abelian Varieties

3.1 The Image of the `-Torsion Galois Representation

Let A be an abelian variety of dimension n defined over Q. The set of Q-points of

A admits a group structure. Let ` be a prime number. Then the subgroup of the Q-
points of A consisting of all `-torsion points, which is denoted by AŒ`, is isomorphic
to .Z=`Z/2n and it is endowed with a natural action of GQ . Therefore, it gives rise
to a (continuous) Galois representation

A;` W GQ ! GL.AŒ`/ ' GL2n .F` /:

As explained in Section 2, we obtain a realization of the image of A;` as a Galois

group over Q.
In this section, we will consider principally polarized abelian varieties, i.e. we
will consider pairs .A; /, where A is an abelian variety (defined over Q) and  W
A ! A_ is an isogeny of degree 1 (that is, an isomorphism between A and the dual
abelian variety A_ ), induced from an ample divisor on A. Not every abelian variety
A admits a principal polarization  and, when it does, it causes certain restrictions
on the image of A;` .
Let V be a vector space of dimension 2n, which is defined over F` and endowed
with a symplectic (i.e., skew-symmetric, nondegenerate) pairing h; i W V V ! F` .
We consider the symplectic group

Sp.V; h; i/ WD fM 2 GL.V/ W 8v1 ; v2 2 V; hMv1 ; Mv2 i D hv1 ; v2 ig

and the general symplectic group

` such that 8v1 ; v2 2 V;

GSp.V; h; i/ W D fM 2 GL.V/ W 9m 2 F
hMv1 ; Mv2 i D mhv1 ; v2 ig:
196 S. Arias-de-Reyna et al.

When A is a principally polarized abelian variety, the image of A;` lies inside the
general symplectic group of AŒ` with respect to a certain symplectic pairing. More
precisely, denote by ` .Q/ the group of `-th roots of unity inside a fixed algebraic
closure Q of Q. Recall that the Weil pairing e` is a perfect pairing

e` W AŒ`  A_ Œ` ! ` .Q/:

If .A; / is a principally polarized abelian variety, we can consider the pairing

e`; W AŒ`  AŒ` ! ` .Q/

.P; Q/ 7! e` .P; .Q//

which is a non-degenerate skew-symmetric pairing (i.e., a symplectic pairing),

compatible with the action of GQ . This last condition means that, for any  2 GQ ,

.e`; .P; Q// D e`; .P ; Q /:

Note that GQ acts on ` .Q/ via the mod ` cyclotomic character ` , so that
.e`; .P; Q// D .e`; .P; Q//` . / . If we fix a primitive `-th root of unity ` , we
may write the pairing e`; .; / additively, i.e. we define

h; i W AŒ`  AŒ` ! F`

as hP; Qi WD a such that  a D e`; .P; Q/.

In other words, we have a symplectic pairing on the F` -vector space AŒ` such
that, for all  2 GQ , the linear map ./ W AŒ` ! AŒ` satisfies that there exists a
scalar, namely ` ./, such that

h./.P/; ./.Q/i D ` ./hP; Qi: (1)

That is to say, the image of the representation A;` is contained in the general
symplectic group GSp.AŒ`; h; i/ ' GSp2n .F` /. Therefore, below we will consider
A;` as a map into GSp.AŒ`; h; i/ ' GSp2n .F` / and we will say that it is surjective
if ImA;` D GSp.AŒ`/ ' GSp2n .F` /.
The determination of the images of the Galois representations A;` attached to
the `-torsion of abelian varieties is a topic that has received a lot of attention. A
remarkable result by Serre quoted in Serre (2000, n. 136, Theorem 3) is:
Theorem 3.1 (Serre). Let A be a principally polarized abelian variety of dimen-
sion n, defined over a number field K. Assume that n D 2; 6 or n is odd and
furthermore assume that EndK .A/ D Z. Then there exists a bound BA;K such that,
for all ` > BA;K ,

ImA;` D GSp.AŒ`/ ' GSp2n .F` /:

Galois Representations and Galois Groups Over Q 197

For arbitrary dimension, the result is not true (see, e.g., Mumford 1969 for an
example in dimension 4). However, one eventually obtains symplectic image by
making some extra assumptions. For example, there is the following result of C. Hall
(cf. Hall 2011).
Theorem 3.2 (Hall). Let A be a principally polarized abelian variety of dimension
n defined over a number field K, such that EndK .A/ D Z, and satisfying the
following property:
(T) There is a finite extension L=K so that the Néron model of A=L over the ring of integers
of L has a semistable fiber with toric dimension 1.

Then there is an (explicit) finite constant BA;K such that, for all `  BA;K ,

ImA;` D GSp.AŒ`/ ' GSp2n .F` /:

Remark 3.3. In the case when A D J.C/ is the Jacobian of a hyperelliptic curve C
of genus n, say defined by an equation Y 2 D f .X/ with f .X/ 2 KŒX a polynomial
of degree 2n C 1, Hall gives a sufficient condition for Condition (T) to be satisfied
at a prime p of the ring of integers of K; namely, the coefficients of f .X/ should
have p-adic valuation greater than or equal to zero and the reduction of f .X/ mod p
(which is well-defined) should have one double zero in a fixed algebraic closure of
the residue field, while all the other zeroes are simple.
Applying the result of Hall with K D Q yields the following partial answer to
the inverse Galois problem:
Corollary 3.4. Let n 2 N be any natural number. Then for all sufficiently large
primes `, the group GSp2n .F` / can be realized as a Galois group over Q.
Remark 3.5. Several people, including the anonymous referee, pointed us to the
following fact: if we consider a family of genus n hyperelliptic curves Ct defined
over Q.t/, with big monodromy at `, then Hilbert’s Irreducibility Theorem provides
us with infinitely many specializations t D t0 2 Q such that the Jacobian Jt0 of the
corresponding curve Ct0 satisfies that ImJt ;` ' GSp2n .F` /. Such families of curves
0
exist for any odd ` (see, e.g., Hall 2008 or Zarhin 2014). In particular, for any n 2 N
and any odd `, the Inverse Galois problem has an affirmative answer for the group
GSp2n .F` /. Although ensuring the existence of the desired curve, this fact does not
tell us how to find such a curve explicitly.
In the case of curves of genus 2, Le Duff has studied the image of the
Galois representations attached to the `-torsion of J.C/, when Condition (T) in
Theorem 3.2 is satisfied. The main result in Le Duff (1998) is the following:
Theorem 3.6 (Le Duff). Let C be a genus 2 curve defined over Q, with bad
reduction of type (II) or (IV) according to the notation in Liu (1993) at a prime
p. Let ˆp be the group of connected components of the special fiber of the Néron
model of J.C/ at p. For each prime ` and each prime q of good reduction of
C, let Pq;` .X/ D X 4 C aX 3 C bX 2 C qaX C q2 2 F` ŒX be the characteristic
198 S. Arias-de-Reyna et al.

polynomial of the image under J.C/;` of the Frobenius element at q and let
Qq;` .X/ D X 2 C aX C b  2q 2 F` ŒX, with discriminants P and Q , respectively.
Then for all primes ` not dividing 2pqjˆp j and such that P and Q are not
squares in F` , the image of J.C/;` coincides with GSp4 .F` /.
Using this result, he obtains a realization of GSp4 .F` / as Galois group over Q for
all odd primes ` smaller than 500,000.

3.2 Explicit Surjectivity Result

A key point in Hall’s result is the fact that the image under A;` of the inertia
subgroup at the place p of L which provides the semistable fiber with toric dimension
1 is generated by a nontrivial transvection (whenever ` does not divide p nor the
cardinality of the group ˆp of connected components of the special fiber of the
Néron model at p). A detailed proof of this fact can be found in Proposition 1.3 of
Le Duff (1998).
We expand on this point. Given a finite-dimensional vector space V over F` ,
endowed with a symplectic pairing h; i W V  V ! F` , a transvection is an element
T 2 GSp.V; h; i/ such that there exists a hyperplane H  V satisfying that the
restriction TjH is the identity on H. We say that it is a nontrivial transvection if T
is not the identity1 . It turns out that the subgroups of GSp.V; h; i/ that contain a
nontrivial transvection can be classified into three categories as follows (for a proof,
see, e.g., Arias-de-Reyna et al. 2014, Theorem 1.1):
Theorem 3.7. Let `  5 be a prime, let V be a finite-dimensional vector space over
F` , endowed with a symplectic pairing h; i W VV ! F` , and let G  GSp.V; h; i/
be a subgroup that contains a nontrivial transvection. Then one of the following
holds:
1. G is reducible. L
2. There exists a proper decomposition V D i2I Vi of V into equidimensional
non-singular symplectic subspaces Vi such that, for each g 2 G and each i 2 I,
there exists some j 2 I with g.Vi /  Vj and such that the resulting action of G
on I is transitive.
3. G contains Sp.V; h; i/.
Remark 3.8. Assume that V is the `-torsion group of a principally polarized abelian
variety A defined over Q and h; i is the symplectic pairing coming from the
Weil pairing. If G D ImA;` satisfies the third condition in Theorem 3.7, then
G D GSp.V; h; i/. Indeed, we have the following exact sequence

1
We adopt the convention that identity is a transvection so that the set of transvections for a given
hyperplane H is a group.
Galois Representations and Galois Groups Over Q 199

1 ! Sp.V; h; i/ ! GSp.V; h; i/ ! F

` ! 1;

where the map m W GSp.AŒ`; h; i/ ! F ` associates with M the scalar a satisfying
that, for all u; v 2 V, hMu; Mvi D ahu; vi. By Equation (1), the restriction of m to
Im.A;` / coincides with the mod ` cyclotomic character ` . We can easily conclude
the result using that ` is surjective onto F`.

Even in the favorable case when we know that Im.A;` / contains a nontrivial
transvection, we still need to distinguish between the three cases in Theorem 3.7.
In this paper, we will make use of the following consequence of Theorem 3.7 (cf.
Corollary 2.2 of Arias-de-Reyna and Kappen 2013).
Corollary 3.9. Let `  5 be a prime, let V be a finite-dimensional vector space
over F` , endowed with a symplectic pairing h; i W V  V ! F` and let G 
GSp.V; h; i/ be a subgroup containing a nontrivial transvection and an element
whose characteristic polynomial is irreducible and which has nonzero trace. Then
G contains Sp.V; h; i/.
In order to apply this corollary in our situation, we need some more information
on the image of A;` . We will obtain this by looking at the images of the Frobenius
elements Frobq for primes q of good reduction of A.
More generally, let A be an abelian variety defined over a field K and assume
that ` is a prime different from the characteristic of K. Any endomorphism ˛ of A
induces an endomorphism of AŒ`, in such a way that the characteristic polynomial
of ˛ (which is a monic polynomial in ZŒX, see, e.g., 3, Chapter 3 of Lang
1959 for its definition) coincides, after reduction mod `, with the characteristic
polynomial of the corresponding endomorphism of AŒ`. In the case when K is
a finite field (say of cardinality q), we can consider the Frobenius endomorphism
q 2 EndK .A/, induced by the action of the Frobenius element Frobq 2 Gal.K=K/.
Then the reduction mod ` of the characteristic polynomial of q coincides with the
characteristic polynomial of A;` .Frobq /. This will turn out to be particularly useful
in the case when A D J.C/ is the Jacobian of a curve C of genus n defined over K,
since one can determine the characteristic polynomial of J.C/;` .Frobq / by counting
the Fqr -valued points of C, for r D 1; : : : ; n.
As a consequence, we can state the following result, which will be used in the
next section.
Theorem 3.10. Let A be a principally polarized n-dimensional abelian variety
defined over Q. Assume that there exists a prime p such that the following condition
holds:
(Tp ) The special fiber of the Néron model of A over Qp is semistable with toric dimension 1.

Denote by ˆp the group of connected components of the special fiber of the Néron
model at p. Let q be a prime of good reduction of A, let Aq be the special fiber of the
Néron model of A over Qq , and let Pq .X/ D X 2n C aX 2n1 C    C qn 2 ZŒX be the
characteristic polynomial of the Frobenius endomorphism acting on Aq .
200 S. Arias-de-Reyna et al.

Then for all primes ` which do not divide 6pqjˆp ja and such that the reduction
of Pq .X/ mod ` is irreducible in F` , the image of A;` coincides with GSp2n .F` /.
Remark 3.11. The condition that ` does not divide a corresponds to the Frobenius
element having non-zero trace modulo `. Note that the theorem is vacuous
when a D 0.

4 Galois Realization of GSp2n .F` / from a Hyperelliptic

Curve of Genus n

Let C be a hyperelliptic curve of genus n over Q, defined by an equation Y 2 D f .X/

where f .X/ 2 QŒX is a polynomial of degree 2n C 1. Let A D J.C/ be its Jacobian
variety. We assume that A satisfies condition .Tp / for some prime p. In this section
we present an algorithm, based on Theorem 3.10, which computes a finite set of
prime numbers ` for which the Galois representation A;` has image GSp2n .F` /. We
apply this procedure to an example of a genus 3 curve using a computer algebra
system.

4.1 Strategy

First, to apply Theorem 3.10, we restrict ourselves to hyperelliptic curves of genus

n whose Jacobian varieties will satisfy Condition (Tp ) for some p. Namely, we fix
a prime number p and then choose f .X/ 2 ZŒX monic of degree 2n C 1 such that
both of the following conditions hold:
1. The polynomial f .X/ only has simple roots over Q, so that Y 2 D f .X/ is the
equation of an hyperelliptic curve C over Q.
2. All coefficients of f .X/ have p-adic valuation greater than or equal to zero, and
the reduction f .X/ mod p has one double zero in Fp , and its other zeroes are
simple. This ensures that A D J.C/ satisfies Condition (Tp ) (see Remark 3.3).
Any prime of good reduction for C is also a prime of good reduction for its
Jacobian A. Primes of good reduction for the hyperelliptic curve can be computed
using the discriminant of Weierstrass equations for C (see Lockhart 1994). In our
case, it turns out that any prime not dividing the discriminant of f .X/ is of good
reduction for C, hence for A.
We take such a prime number q of good reduction for A. Recall that Pq .X/ 2
ZŒX is the characteristic polynomial of the Frobenius endomorphism acting on the
fiber Aq .
Let Sq denote the set of prime numbers ` satisfying the following conditions:
(i) ` divides neither 6pqjˆp j nor the coefficient of X 2n1 in Pq .X/,
(ii) the reduction of Pq .X/ modulo ` is irreducible in F` .
Galois Representations and Galois Groups Over Q 201

Note that if the coefficient of X 2n1 in Pq .X/ is nonzero, condition (i) rules out
only finitely many prime numbers `, whereas if it vanishes, condition (i) rules out
all prime numbers `. By Theorem 3.10, for each ` 2 Sq the representation A;` is
surjective with image GSp2n .F` /. Also, primes in Sq can be computed effectively
up to a given fixed bound.
Since we want the polynomial Pq .X/ (of degree 2n) to be irreducible modulo
`, its Galois group G over Q must be a transitive subgroup of S2n with a 2n-cycle.
Therefore, by an application of a weaker version of the Chebotarev density theorem
due to Frobenius (Stevenhagen and Lenstra 1996, “Theorem of Frobenius”, p. 32),
the density of Sq is

#f 2 G  S2n W  is a 2n-cycleg

:
#G
This estimate is far from what Theorem 3.2 provides us, namely that the density of
`’s with Im.A;` / D GSp2n .F` / is 1.
This leads us to discuss the role of the prime q. First of all, we can see that
[
Sq D f` primeW ` − 6pjˆp j and A;` surjectiveg;
q

where the union is taken over all primes q of good reduction for A. Note that the
inclusion  follows directly from Theorem 3.10. To show the other inclusion ,
suppose now that ` − 6pjˆp j and that the representation at ` is surjective. Its
image GSp2n .F` / contains an element with irreducible characteristic polynomial
and nonzero trace (see, for instance, Proposition A.2 of Arias-de-Reyna and Kappen
2013). This element defines a conjugacy class C  GSp2n .F` / and the Chebotarev
density theorem ensures that there exists q such that A;` .Frobq / 2 C, hence ` 2 Sq .
Moreover, if, for some fixed `, the events “` belongs to Sq ” are independent as
q varies, the density of primes ` for which A;` is surjective will increase when we
take several different primes q. A sufficient condition for this density to tend to 1 is
that there exists an infinite family of primes q for which the splitting fields of Pq .X/
are pairwise linearly disjoint over Q.
Therefore, it seems reasonable to expect that computing the sets Sq for several
values of q increases the density of primes ` for which we know the surjectivity of
A;` . This is what we observe numerically in the next example.

4.2 A Numerical Example in Genus 3

We consider the hyperelliptic curve C of genus n D 3 over Q defined by Y 2 D f .X/,

where

f .X/ D X 2 .X  1/.X C 1/.X  2/.X C 2/.X  3/ C 7.X  28/ 2 ZŒX:

202 S. Arias-de-Reyna et al.

This is a Weierstrass equation, which is minimal at all primes ` different from

2 (see Lockhart 1994, Lemma 2.3), with discriminant 212  7  73  1;069;421 
11;735;871;491. Thus, C has good reduction away from the primes appearing in this
factorization. Clearly, p D 7 is a prime for which the reduction of f .X/ modulo 7 has
one double zero in F7 and otherwise only simple zeroes. Therefore, its Jacobian J.C/
satisfies Condition (T7 ). As we computed with MAGMA, the order of the component
group ˆ7 is 2. Recall that Pq .X/ coincides with the characteristic polynomial of the
Frobenius endomorphism of the reduced curve C modulo q over Fq .
Our method provides no significant result for q 2 f3; 5g because for q D 3 the
characteristic polynomial Pq .X/ is not irreducible in ZŒX and for q D 5 it has zero
trace in Z. So in this example, we first take q D 11. The curve has 11; 135, and 1247
points over F11 , F112 , and F113 , respectively. The characteristic polynomial P11 .X/ is

P11 .X/ D X 6  X 5 C 7X 4  35X 3 C 77X 2  121X C 1331

and it is irreducible over Q. Its Galois group G has order 48 and is isomorphic to
the wreath product S2 o S3 . This group is the direct product of 3 copies of S2 , on
which S3 acts by permutation (see James and Kerber 1981, Chapter 4): An element
of S2 o S3 can be written as ..a1 ; a2 ; a3 /; /, where .a1 ; a2 ; a3 / denotes an element of
the direct product S2  S2  S2 and  an element of S3 . The group law is defined as
follows:

..a1 ; a2 ; a3 /; /..a01 ; a02 ; a03 /;  0 / D ..a1 ; a2 ; a3 /.a01 ; a02 ; a03 / ;  0 /;

where .a01 ; a02 ; a03 / D .a0 .1/ ; a0 .2/ ; a0 .3/ /. One can also view the wreath product
S2 o S3 as the centralizer of .12/.34/.56/ in S6 , through an embedding W S2 o S3 !
S6 whose image is isomorphic to the so-called Weyl group of type B3 (James and
Kerber 1981, 4.1.18 and 4.1.33). More precisely, under , the image of an element
..a1 ; a2 ; a3 /; / 2 S2 oS3 is the permutation of S6 that acts on f1; 2; : : : ; 6g as follows:
it first permutes the elements of the sets E1 D f1; 2g, E2 D f3; 4g and E3 D f5; 6g
separately, according to a1 , a2 , and a3 respectively (identifying E2 ; E3 with f1; 2g
in an obvious way) and then permutes the pairs E1 ; E2 ; E3 according to the action
of  on the indices. For example, denoting S2 D fid; g, the image under of
..; id; id/; .123// is the 6-cycle .135246/.
Let us now determine the elements of S2 o S3 which map to 6-cycles in S6 through
the embedding . For an element in S2 o S3 to be of order 6, it has to be of the form
..a1 ; a2 ; a3 /;  / with  a 3-cycle in S3 . Now, sends an element ..a1 ; a2 ; a3 /;  /
where either one or three ai ’s are id, to a product of two disjoint 3-cycles in S6 .
So the elements of S2 o S3 which are 6-cycles in S6 are among the eight elements
..id; id; /;  /, ..id; ; id/;  /, ..; id; id/;  / and ..; ; /;  / with  D .123/ or
 D .132/. Moreover, James and Kerber (1981, Theorem 4.2.8) (see also Gramain
2008, Lemma 3.1 or Taylor 2012) ensures that these 8 elements are conjugate. Since
..; id; id/; .123// D .135;246/ is a 6-cycle, we deduce that the 8 elements listed
above are exactly the elements of S2 o S3 which are 6-cycles in S6 .
Galois Representations and Galois Groups Over Q 203

To conclude, the Galois group G, viewed as a subgroup of S6 , contains exactly 8

elements that are 6-cycles. Therefore, the density of S11 is 8=48 D 1=6.
We can compute Pq .X/ using efficient algorithms available in MAGMA Bosma
et al. (1997) or SAGE Stein (2014), which are based on p-adic methods. We found
that there are 6891 prime numbers 11  `  500;000 that belong to S11 . For these
`, we know that the image of A;` is GSp6 .F` /, so the groups GSp6 .F` / are realized
as Galois groups arising from the `-torsion of the Jacobian of the hyperelliptic curve
C. For instance, the first ten elements of S11 are

47; 71; 79; 83; 101; 113; 137; 251; 269; 271:

Also, the proportion of prime numbers 11  `  500;000 in S11 is about 0.1659,

which is quite in accordance with the density obtained from the Chebotarev density
theorem.
By looking at polynomials Pq .X/ for several primes q of good reduction, we are
able to significantly improve the known proportion of primes `, up to a given bound,
for which the Galois representation is surjective. Namely, we computed that
[
f` prime; 11  `  500;000g  Sq :
11q571

As a consequence, for any prime 11  `  500;000, the group GSp6 .F` / is realized
as a Galois group arising from the `-torsion of the Jacobian of the hyperelliptic curve
C. This is reminiscent of Le Duff’s numerical data for GSp4 .Fl / (see Theorem 3.6).
Combining all of the above suggests that the single hyperelliptic curve C might
provide a positive answer to the inverse Galois problem for GSp6 .F` / for any
prime `  11.

Acknowledgements The authors would like to thank Marie-José Bertin, Alina Bucur, Brooke
Feigon, and Leila Schneps for organizing the WIN-Europe conference which initiated this
collaboration. Moreover, we are grateful to the Centre International de Rencontres Mathématiques,
the Institut de Mathématiques de Jussieu, and the Institut Henri Poincaré for their hospitality
during several short visits. The authors are indebted to Irene Bouw, Jean-Baptiste Gramain, Kristin
Lauter, Elisa Lorenzo, Melanie Matchett Wood, Frans Oort, and Christophe Ritzenthaler for several
insightful discussions. We also want to thank the anonymous referee for her/his suggestions that
helped us to improve this paper.
S. Arias-de-Reyna and N. Vila are partially supported by the project MTM2012-33830 of
the Ministerio de Economía y Competitividad of Spain, C. Armana by a BQR 2013 Grant from
Université de Franche-Comté and M. Rebolledo by the ANR Project Régulateurs ANR-12-BS01-
0002. L. Thomas thanks the Laboratoire de Mathématiques de Besançon for its support.
204 S. Arias-de-Reyna et al.

References

Arias-de-Reyna, S., Dieulefait, L., Shin, S.-W., Wiese, G.: Compatible systems of symplectic
Galois representations and the inverse Galois problem III. Automorphic construction of
compatible systems with suitable local properties. Math. Ann. 361(3), 909–925 (2015)
Arias-de-Reyna, S., Dieulefait, L, Wiese, G.: Classification of subgroups of symplectic groups over
finite fields containing a transvection. Demonstratio Math. (2014, preprint)
Arias-de-Reyna, S., Kappen, C.: Abelian varieties over number fields, tame ramification and big
Galois image. Math. Res. Lett. 20(1), 1–17 (2013)
Arias-de-Reyna, S., Vila, N.: Tame Galois realizations of GSp4 .F` / over Q. Int. Math. Res. Not.
IMRN 9, 2028–2046 (2011)
Bosma, W., Cannon, J., Playoust, C.: The Magma algebra system. I. The user language. J. Symb.
Comput. 24(3–4), 235–265 (1997). Computational algebra and number theory (London, 1993)
Dieulefait, L.V.: Explicit determination of the images of the Galois representations attached to
abelian surfaces with End.A/ D Z. Exper. Math. 11(4), 503–512 (2002a)
Dieulefait, L.V.: On the images of the Galois representations attached to genus 2 Siegel modular
forms. J. Reine Angew. Math. 553, 183–200 (2002b)
Dettweiler, M., Kühn, U., Reiter, S.: On Galois representations via Siegel modular forms of genus
two. Math. Res. Lett. 8(4), 577–588 (2001)
Dieulefait, L., Vila, N.: Projective linear groups as Galois groups over Q via modular representa-
tions. J. Symb. Comput. 30(6), 799–810 (2000). Algorithmic methods in Galois theory
Dieulefait, L., Vila, N.: On the images of modular and geometric three-dimensional Galois
representations. Am. J. Math. 126(2), 335–361 (2004)
Dieulefait, L., Vila, N.: Geometric families of 4-dimensional Galois representations with generi-
cally large images. Math. Z. 259(4), 879–893 (2008)
Dieulefait, L., Vila, N.: On the classification of geometric families of four-dimensional Galois
representations. Math. Res. Lett. 18(4), 805–814 (2011)
Dieulefait, L., Wiese, G.: On modular forms and the inverse Galois problem. Trans. Am. Math.
Soc. 363(9), 4569–4584 (2011)
Gramain, J.-B.: On defect groups for generalized blocks of the symmetric group. J. Lond. Math.
Soc. (2) 78(1), 155–171 (2008)
Hall, C.: Big symplectic or orthogonal monodromy modulo l. Duke Math. J. 141(1), 179–203
(2008)
Hall, C.: An open-image theorem for a general class of abelian varieties. Bull. Lond. Math. Soc.
43(4), 703–711 (2011). With an appendix by Emmanuel Kowalski
James, G., Kerber, A.: The Representation Theory of the Symmetric Group. Encyclopedia of
Mathematics and its Applications, vol. 16. Addison-Wesley, Reading (1981). With a foreword
by P. M. Cohn, With an introduction by Gilbert de B. Robinson
Khare, C., Larsen, M., Savin, G.: Functoriality and the inverse Galois problem. Compos. Math.
144(3), 541–564 (2008)
Lang, S.: Abelian Varieties. Interscience Tracts in Pure and Applied Mathematics. No. 7. Inter-
science, New York/London (1959)
Le Duff, P.: Représentations galoisiennes associées aux points d’ordre ` des jacobiennes de
certaines courbes de genre 2. Bull. Soc. Math. France 126(4), 507–524 (1998)
Liu, Q.: Courbes stables de genre 2 et leur schéma de modules. Math. Ann. 295(2), 201–222 (1993)
Lockhart, P.: On the discriminant of a hyperelliptic curve. Trans. Am. Math. Soc. 342(2), 729–752
(1994)
Mumford, D.: A note of Shimura’s paper “Discontinuous groups and abelian varieties”. Math. Ann.
181, 345–351 (1969)
Ribet, K.A.: On `-adic representations attached to modular forms. Invent. Math. 28, 245–275
(1975)
Galois Representations and Galois Groups Over Q 205

Reverter, A., Vila, N.: Some projective linear groups over finite fields as Galois groups over Q.
In: Recent Developments in the Inverse Galois Problem (Seattle, WA, 1993). Contemporary
Mathematics, vol. 186, pp. 51–63. American Mathematical Society, Providence (1995)
Serre, J.-P.: Œuvres. Collected papers IV. Springer, Berlin (2000). 1985–1998
Serre, J.-P.: Propriétés galoisiennes des points d’ordre fini des courbes elliptiques. Invent. Math.
15(4), 259–331 (1972)
Stein, W.A., et al.: Sage Mathematics Software (Version 6.0). The Sage Development Team. http://
www.sagemath.org (2014)
Stevenhagen, P., Lenstra, H. W.: Chebotarëv and his density theorem. Math. Intell. 18(2), 26–37
(1996)
Taylor, J.: Families of irreducible representations of S2 o S3 . https://documents.epfl.ch/users/j/jt/
jtaylor/www/PDF/representations_of_S2wrS3.pdf (2012)
Wiese, G.: On projective linear groups over finite fields as Galois groups over the rational
numbers. In: Modular Forms on Schiermonnikoog, pp. 343–350. Cambridge University Press,
Cambridge (2008)
Zarhin, Y.G.: Two-dimensional families of hyperelliptic jacobians with big monodromy (preprint,
2014) [arXiv:1310.6532]
Zywina, D.: The inverse Galois problem for PSL2 .Fp / (preprint, 2013) [arXiv:1303.3646]