Springer New York Berlin Heidelberg Barcelona Budapest Hong Kong London Milan Paris Tokyo Graduate Texts in Mathematics 1 0 4 Editorial Board S.Axler F.W. Gehring P. R. HalmosGraduate Texts in Mathematics 15 16 17 18 19 20 21 22 2B 24 25 26 27 28 29 30 31 32 TAKEUTVZaRING. Introduction to Axiomatic Set Theory. 2nd ed. Oxrosy. Measure and Category. 2nd ed. . SCHAEFER. Topological Vector Spaces. HILTON/STAMMBACH. A Course in Homological Algebra. Mac LaNE. Categories for the Working Hucues/Prrer. Projective Planes. ‘Serre, A Course in Arithmetic. TAKEUTUZARING. Axiomatic Set Theory. HUMPHREYS. Introduction to Lie Algebras and Representation Theory. CoHEN. A Course in Simple Homotopy Theory. Conway. Functions of One Complex Variable I. 2nd ed. Beats. Advanced Mathematical Analysis. ANDERSON/FULLER. Rings and Categories of Modules. 2nd ed. GoLusrTsky/GUILLEMIN. Stable Mappings and Their Singularities. BERBERIAN. Lectures in Functional Analysis and Operator Theory. ‘Winter. The Structure of Fields. RosENBLATT. Random Processes. 2nd ed. HaLMos. Measure Theory. HaLmos. A Hilbert Space Problem Book. 2nd ed. HUSEMOLLER. Fibre Bundles. 3rd ed. HuMPHreys. Linear Algebraic Groups. BarNes/MACk. An Algebraic Introduction to Mathematical Logic. Greve. Linear Algebra, 4th ed. Hotes. Geometric Functional Analysis and Its Applications. Hewrtt/STROMBERG. Real and Abstract Analysis, Manes, Algebraic Theories. Ke.ey. General Topology. ZaRisk/SAMUEL. Commutative Algebra, VoL. ZARISKUSAMUEL. Commutative Algebra. Vol.H. Jacosson. Lectures in Abstract Algebra I.” Basic Concepts. TacOBSON. Lectures in Abstract Algebra Il, Linear Algebra. Jacoason. Lectures in Abstract Algebra IIL. Theory of Fields and Galois Theory. 33 35 36 37 38 39 41 42 43 45 47 49 50 3 52 33 34 5S 56 57 58 59 60 Hirsci. Differential Topology. ‘Sprrzer. Principles of Random Walk. 2nd ed. WerMER. Banach Algebras and Several Complex Variables. 2nd ed. KELLEY/NaMIOKA et al. Linear Topological Spaces. Monk. Mathematical Logic. GRAUERT/FRITZSCHE. Several Complex Variables. Arveson. An Invitation to C*-Algebras. KEMENY/SNELL/KNapP. Denumerable Markov Chains. 2nd ed. AposToL. Modular Functions and Dirichlet Series in Number Theory. 2nd ed. ‘Serre. Linear Representations of Finite Groups. GILLMaN/JeRison. Rings of Continuous Functions. Kenpic. Elementary Algebraic Geometry. Lofve. Probability Theory I. 4th ed. Loéve. Probability Theory II. 4th ed. Moise. Geometric Topology in Dimensions 2 and 3. SAcHs/Wu. General Relativity for Mathematicians. GRUENBERG/WeIR. Linear Geometry. 2nd ed. Epwarps. Fermat's Last Theorem. KLINGENBERG. A Course in Differential Geometry. HARTSHORNE. Algebraic Geometry. Manin. A Course in Mathematical Logic. Graver/WaTkins. Combinatorics with Emphasis on the Theory of Graphs. Brown/PEarcy. Introduction to Operator Theory I: Elements of Functional Analysis. Massey. Algebraic Topology: An Introduction, CROWELL/FOx. Introduction to Knot Theory. Kopitz. p-adic Numbers, p-adic Analysis, and Zeta-Functions. 2nd ed. Lana. Cyclotomic Fields. ARNOLD. Mathematical Methods in ‘Classical Mechanics. 2nd ed. continued after indexB. A. Dubrovin A. T. Fomenko S.P. Novikov Modern Geometry— . Methods and Applications Part II. The Geometry and Topology of Manifolds Translated by Robert G. Burns With 126 Illustrations SpringerB. A. Dubrovin A. T. Fomenko Department of Mathematics and Mechanics Moscow State University Moscow University V-234 Moscow Leninskie Gory Russia Moscow 119899 Russia R. G. Bums (Translator) S. P. Novikov Foca Mathematics Institute of Physical Sciences and Technology York University Maryland University 4700 Keele Street College Park, MD 20742-2431 North York, ON, M3J 1P3 USA Canada Editorial Board S. Axler EF. W. Gehring P.R. Halmos Department of Mathematics Department of Mathematics Department of Mathematics Michigan State University University of Michigan Santa Clara University East Lansing, MI 48824 Ann Arbor, MI 48109 Santa Clara, CA 95053 USA USA USA Mathematics Subject Classification (1991): 53-01, 53B50, 57-01, S8Exx Library of Congress Cataloging in Publication Data (Revised for vol. 2) 7 Dubrovin, B. A. Modern geometry—methods and applications. (Springer series in Soviet mathematics) (Graduate texts in mathematics; 93- ) “Original Russian edition . .. Moskva: Nauka, 1979"—Top. verso. Includes bibliographies and indexes. 1, Geometry. I. Fomenko, A. T. II. Novikov, Sergei Petrovich. III. Title. IV. Series. V. Series: Graduate texts in mathematics; 93, etc. QA445.D82 1984 516 83-16851 This book is part of the Springer Series in Soviet Mathematics. Original Russian edition: Sovremennaja Geometria: Metody i Prilozenia. Moskva: Nauka, 1979. © 1985 by Springer-Verlag New York Inc. 7 All rights reserved. No part of this book may be translated or reproduced in any form without written permission from Springer-Verlag, 175 Fifth Avenue, New York, New York 10010, U.S.A. This reprint has been authorized by Springer-Verlag (Berlin/Heidelberg/New York) for sale in the People's Republic of China only and not for export therefrom. Reprinted in China by Beijing World Publishing Corporation, 1999 98765432 ISBN 0-387-96162-3 Springer-Verlag New York Berlin Heidelberg Tokyo ISBN 3-540-96162-3 Springer-Verlag Berlin Heidelberg New York TokyoPreface Up until recently, Riemannian geometry and basic topology were not included, even by departments or faculties of mathematics, as compulsory subjects in a university-level mathematical education. The standard courses in the classical differential geometry of curves and surfaces which were given instead (and still are given in some places) have come gradually to be viewed as anachronisms. However, there has been hitherto no unanimous agreement as to exactly how such courses should be brought up to date, that is to-say, which parts of modern geometry should be regarded as absolutely essential to a modern mathematical education, and what might be the appropriate level of abstractness of their exposition. The task of designing a modernized course in geometry was begun in 1971 in the mechanics division of the Faculty of Mechanics and Mathematics of Moscow State University. The subject-matter and level of abstractaess of its exposition were dictated by the view that, in addition to the geometry of curves and surfaces, the following topics are certainly useful in the various areas of application of mathematics (especially in elasticity and relativity, to name but two), and are therefore essential: the theory of tensors (including covariant differentiation of them); Riemannian curvature; geodesics and the calculus of variations (including the conservation laws and Hamiltonian formalism); the particular case of skew-symmetric tensors (i.e. “forms”) together with the operations on them; and the various formulae akin to Stokes’ (including the all-embracing and invariant “general Stokes formula” in-n dimensions). Many leading theoretical physicists shared the mathemati- cians’ view that it would also be useful to include some facts about manifolds, transformation groups, and Lie algebras, as well as the basic concepts of visual topology. It was also agreed that the course should be given in as simple and concrete a language as possible, and that wherever practicable thevi Preface terminology should be that used by physicists. Thus it was along these lines that the archetypal course was taught. It was given more permanent form as duplicated lecture notes published under the auspices of Moscow State University as: Differential Geometry, Parts I and II, by S. P. Novikov, Division of Mechanics, Moscow State University, 1972. Subsequently various parts of the course were altered, and new topics added. This supplementary material was published (also in duplicated form) as: Differential Geometry, Part III, by S. P. Novikov and A. T. Fomenko, Division of Mechanics, Moscow State University, 1974. The present book is the outcome of a reworking, re-ordering, and extensive elaboration of the above-mentioned lecture notes. It is the authors’ view that it will serve as a basic text from which the essentials for a course in modern geometry may be easily extracted. To S. P. Novikov are due the original conception and the overall plan of the book. The work of organizing the material contained in the duplicated lecture noies in accordance with this plan was carried out by B. A. Dubrovin. This- accounts for more than half of Part I; the remainder of the book is essentially new. The efforts of the editor, D. B. Fuks, in bringing the book to completion, were invaluable. The content of this book significantly exceeds the material that might be considered as essential to the mathematical education of second- and third- year university students. This was intentional: it was part of our plan that even in Part I there should be included several sections serving to acquaint (through further independent study) both undergraduate and graduate students with the more complex but essentially geometric concepts and methods of the theory of transformation groups and their Lie algebras, field theory, and the calculus of variations, and with, in particular, the basic ingredients of the mathematical formalism of physics. At the same time we strove to minimize the degree of abstraction of the exposition and terminol- ogy, often sacrificing thereby some of.the so-called “generality” of statements and proofs: frequently an important result may be obtained in the context of crucial examples containing the whole essence of the matter, using only elementary classical analysis and geometry and without invoking any modern “hyperinvariant” concepts and notations, while the result’s most general formulation and especially the concomitant proof will necessitate a dramatic increase in the complexity and abstractness of the exposition. Thus in such cases we have first expounded the result in question in the setting of the relevant significant examples, in the simplest possible language appro- priate, and have postponed the proof of the general form of the result, or omitted it altogether. For our treatment of those geometrical questions more closely bound up with modern physics, we analysed the physics literature:Preface vii books on quantum field theory (see e.g. [35], [37]) devote considerable portions of their beginning sections to describing, in physicists’ terms, useful facts about the most important concepts associated with the higher- dimensional calculus of variations and the simplest representations of Lie groups; the books [41], [43] are devoted to field theory in its geometric aspects; thus, for instance, the book [41] contains an extensive treatment of Riemannian geometry from the physical point of view, including much useful concrete material. It is interesting to look at books on the mechanics of continuous media and the theory of rigid bodies ([42], [44], [45]) for further examples of applications of tensors, group theory, etc. In writing this book it was not our aim to produce a “self-contained” text: ina standard mathematical education, geometry is just one component of the curriculum; the questions of concern in analysis, differential equations, algebra, elementary general topology and measure theory, are examined in other courses. We have refrained from detailed discussion of questions drawn from other disciplines, restricting ourselves to their formulation only, since they receive sufficient attention in the standard programme. In the treatment of its subject-matter, namely the geometry and topology of manifolds, Part II goes much further beyond the material appropriate to the aforementioned basic geometry course, than does Part I. Many books have been written on the topology and geometry of manifolds: however, most of them are concerned with narrowly defined portions of that subject, are written in a language (as a rule very abstract) specially contrived for the particular circumscribed area of interest, and include all rigorous founda- tional detail often resulting only in unnecessary complexity. In Part II also we have been faithful, as far as possible, to our guiding principle of minimal abstractness of exposition, giving preference as before to the significant examples over the general theorems, and we have also kept the inter- dependence of the chapters to a minimum, so that they can each be read in isolation insofar as the nature of the subject-matter allows. One must however bear in mind the fact that although several topological concepts (for instance, knots and links, the fundamental group, homotopy groups, fibre spaces) can be defined easily enough, on the other hand any attempt to make nontrivial use of them in even the simplest examples inevitably requires the development ‘of certain tools having no forbears in classical mathematics. Consequently the reader not hitherto acquainted with elementary topology will find (especially if he is past his first youth) that the level of difficulty of Part II is essentially higher than that of Part I; and for this there is no possible remedy. Starting in the 1950s, the development of this apparatus and its incorporation into various branches of mathematics has. proceeded with great rapidity. In recent years there has appeared a rash, as it were, of nontrivial applications of topological methods (sometimes in combination with complex algebraic geometry) to various problems of modern theoretical physics: to the quantum theory of specific fields of a geometrical nature (for example, Yang-Mills and chiral fields), the theory of fluid crystals andVili Preface superfluidity, the general theory of relativity, to certain physically important nonlinear wave equations (for instance, the Korteweg-de Vries and sine— Gordon equations); and there have been attempts to apply the theory of knots and links in the statistical mechanics of certain substances possessing “long molecules”. Unfortunately we were unable to include these applications in the framework of the present book, since in each case an adequate treatment would have required a lengthy preliminary excursion into physics, and so would have taken us too far afield. However, in our choice of material we have taken into account which topological concepts and methods are exploited in these applications, being aware of the need for a topology text which might be read (given strong enough motivation) by a young theoretical physicist of the modern school, perhaps with a particular object in view. The development of topological and geometric ideas over the last 20 years has brought in its train an essential increase in the complexity of the algebraic apparatus used in combination with higher-dimensional geometrical in- tuition, as also in the utilization, at’a profound level, of functional analysis, the theory of partial differential equations, and complex analysis; not all of this has gone into the present book, which pretends to being elementary (and in fact most of it is not yet contained in any single textbook, and has therefore to be gleaned from monographs and the professional journals). Three-dimensional geometry in the large, in particular the theory of convex figures and its applications, is an intuitive and generally useful branch of the classical geometry of surfaces in 3-space; much interest attaches in particular to the global problems of the theory of surfaces of negative curvature. Not being specialists in this field we were unable to extract its essence in sufficiently simple and illustrative form for inclusion in an elementary text. The reader may acquaint himself with this branch of geometry from the books [1], [4] and [16]. Of all the books on the topology and geometry of manifolds, the classical works A Textbook of Topology and The Calculus of Variations in the Large, of Siefert and Threlfall, and also the excellent more modern books [10], [11] and [12], turned out to be closest to our conception in approach and choice of topics. In the process of creating the present text we actively mulled over and exploited the material covered in these books, and their methodology. In fact our overall aim in writing Part II was to produce something like a modern analogue of Seifert and Threlfall’s Textbook of Topology, which would however be much wider-ranging, remodelled as far as possible using modern techniques of the theory of smooth manifolds (though with simplicity of language preserved), and enriched with new material as dictated by the contemporary view of the significance of topological methods, and of the kind of reader who, encountering topology for the first time, desires to learn a reasonable amount in the shortest possible time. It seemed to us sensible to try to benefit (more particularly in Part I, and as far as this is possible in a book on mathematics) from the accumulated methodological experience of the physicists, that is, to strive to make pieces of nontrivial mathematics morePreface ix comprehensible through the use of the most elementary and generally familiar means available for their exposition (preserving, however, the format characteristic of the mathematical literature, wherein the statements of the main conclusions are separated out from the body of the text by designating them “theorems”, “lemmas”, etc.). We hold the opinion that, in general, understanding should precede formalization and rigorization. There are many facts the details of whose proofs have (aside from their validity) absolutely no role to play in their utilization in applications. On occasion, where it seemed justified (more often in the more difficult sections of Part II) we have omitted the proofs of needed facts. In any case, once thoroughly familiar with their applications, the reader may (if he so wishes), with the help of other sources, easily sort out the proofs of such facts for himself. (For this purpose we recommend the book [21].) We have, moreover, attempted to break down many of these omitted proofs into soluble pieces which we have placed among the exercises at the end of the relevant sections. In the final two chapters of Part II we have brought together several items from the recent literature on dynamical systems and foliations, the general theory of relativity, and the theory of Yang-Mills and chiral fields. The ideas expounded there are due to various contemporary researchers; however in a book of a purely textbook character it may be accounted permissible not to give a long list of references. The reader who graduates to a deeper study of these questions using the research journals will find the relevant references there. Homology theory forms the central theme of Part III. In conclusion we should like to express our deep gratitude to our colleagues in the Faculty of Mechanics and Mathematics of M.S.U., whose valuable support made possible the design and operation of the new geometry courses; among the leading mathematicians in the faculty this applies most of all-to the creator of the Soviet school of topology, P. S. Aleksandrov, and to the eminent geometers P. K. RaSevskii and N. V. Efimov. We thank the editor D. B. Fuks for his great efforts in giving the manuscript its final shape, and A. D. Aleksandrov, A. V. Pogorelov, Ju. F. Borisov, V. A. Toponogov and V. I. Kuz’minov, who in the course of reviewing the book contributed many useful comments. We also thank Ja. B. Zel’dovié for several observations leading to improvements in the exposition at several points, in connexion with the preparation of the English and French editions of this book. We give our special thanks also to the scholars who facilitated the task of incorporating the less standard material into the book. For instance the proof of Liouville’s theorem on conformal transformations, which is not to be found in the standard literature, was communicated to us by V. A. Zorié. The editor D. B. Fuks simplified the proofs of several theorems. We are grateful also to O. T. Bogojavlenskii, M. I. Monastyrskii, S. G. Gindikin, D. V. Alekseevskii, I. V. Gribkov, P. G. Grinevit, and E. B. Vinberg.x Preface Translator’s acknowledgments. Thanks are due to Abe Shenitzer for much kind advice and encouragement, to several others of my colleagues for putting their expertise at my disposal, and to Eadie Henry for her excellent typing and great patience.Contents CHAPTER 1 Examples of Manifolds §1. The concept of a manifold 1.1. Definition of a manifold 1.2. Mappings of manifolds; tensors on manifolds 1.3. Embeddings and immersions of manifolds. Manifolds with boundary §2. The simplest examples of manifolds 2.1. Surfaces in Euclidean space. Transformation groups as manifolds 2.2. Projective spaces . 2.3. Exercises §3. Essential facts from the theory of Lie groups 3.1. The structure of a neighbourhood of the identity of a Lie group. The Lie algebra of a Lie group. Semisimplicity 3.2. The concept of a linear representation. An example of a non-matrix Lie‘group §4. Complex manifolds 4,1. Definitions and examples 4.2, Riemann surfaces as manifolds §5. The simplest homogeneous spaces 5.1. Action of a group on a manifold 5.2. Examples of homogeneous spaces 5.3, Exercises §6. Spaces of constant curvature (symmetric spaces) 6.1. The concept of a symmetric space 6.2, The isometry group of a manifold. Properties of its Lie algebra 6.3. Symmetric spaces of the first and second types 6.4. Lie groups as symmetric spaces 6.5. Constructing symmetric spaces. Examples 6.6. Exercises Veen 10 15 19 20 28 31 31 37 41 41 42 49 SL 53 55 58xii §7. Vector bundles on a manifold 7.1. Constructions involving tangent vectors 7.2. The normal vector bundle on a submanifold CHAPTER 2 : Foundational Questions. Essential Facts Concerning Functions on a Manifold. Typical Smooth Mappings §8. Partitions of unity and their applications 8.1. Partitions of unity 8.2. The simplest applications of partitions of unity. Integrals over a manifold and the general Stokes formula 8.3. Invariant metrics §9. The realization of compact manifolds as surfaces in RY §10. Various properties of smooth maps of manifolds 10.1. Approximation of continuous mappings by smooth ones 10.2. Sard’s theorem 10.3. Transversal regularity 10.4. Morse functions §11. Applications of Sard’s theorem 11.1. The existence of embeddings and immersions 11.2. The construction of Morse functions as height functions 11.3. Focal points CHAPTER 3 The Degree of a Mapping. The Intersection Index of Submanifolds. Applications §12. The concept of homotopy 12.1. Definition of homotopy. Approximation of continuous maps and homotopies by smooth ones 12.2. Relative homotopies §13. The degree of a map 13.1, Definition of degree 13.2. Generalizations of the concept of degree 13.3. Classification of homotopy classes of maps from an arbitrary manifold to a sphere 13.4. The simplest examples §14. Applications of the degree of a mapping 14.1. The relationship between degree and integral 14.2. The degree of a vector field on a hypersurface 14.3. The Whitney number. The Gauss~Bonnet formula 14.4. The index of a singular point of a vector field 14.5. Transverse surfaces of a vector field. The Poincaré-Bendixson theorem §15. The intersection index and applications 15.1. Definition of the intersection index 15.2. The total index of a vector field Contents 59 59 62 65 65 69 74 16 1 71 19 83 86 93 95 99 * 99 102 102 102 104 106 108 110 110 112 14 118 122 125 125 127Contents 15.3. The signed number of fixed points of a self-map (the Lefschetz number). The Brouwer fixed-point theorem 15.4. The linking coefficient CHAPTER 4 Orientability of Manifolds. The Fundamental Group. Covering Spaces (Fibre Bundles with Discrete Fibre) §16. Orientability and homotopies of closed paths 16.1. Transporting an orientation along a path 16.2. Examples of non-orientable manifolds §17. The fundamental group 17.1. Definition of the fundamental group 17.2. The dependence on the base point 17.3. Free homotopy Classes of maps of the circle 17.4. Homotopic equivalence 17.5. Examples 17.6. The fundamental group and orientability §18. Covering maps and covering homotopies 18.1. The definition and basic properties of covering spaces 18.2. The simplest examples. The universal covering 18.3, Branched coverings. Riemann surfaces 18.4. Covering maps and discrete groups of transformations §19. Covering maps and the fundamental group. Computation of the fundamental group of certain manifolds . 19.1. Monodromy 19.2. Covering maps as an aid in the calculation of fundamental groups 19.3. The simplest of the homology groups 19.4, Exercises §20. The discrete groups of motions of the Lobachevskian plane CHAPTER 5 Homotopy Groups §21. Definition of the absolute and relative homotopy groups. Examples 21.1. Basic definitions 21.2. Relative homotopy groups. The exact sequence of a pair §22. Covering homotopies. The homotopy groups of covering spaces and loop spaces 22.1. The concept of a fibre space 22.2. The homotopy exact sequence of a fibre space 22.3. The dependence of the homotopy groups on the base point 22.4. The case of Lie groups 22.5. Whitehead multiplication Facts concerning the homotopy groups of spheres. Framed normal bundles. The Hopf invariant 23.1. Framed normal bundles and the homotopy groups of spheres 2 xiii 130 133 135 135 135 137 139 139 141 142 143 144 147 148 148 150 153 156 157 157 160 164 166 166 185 185 189 193 193 195 198 201 204 207 207xiv Contents 23.2. The suspension map 212 23.3. Calculation of the groups 7,, (S") 214 23.4. The groups 44 2(S") 216 CHAPTER 6 Smooth Fibre Bundles 220 §24. The homotopy theory of fibre bundles 220 24.1. The concept of a smooth fibre bundle 220 24.2. Connexions 225 24.3. Computation of homotopy groups by means of fibre bundles 228 24.4, The classification of fibre bundles 235 24.5. Vector bundles and operations on them 241 24.6. Meromorphic functions 243 24.7. The Picard-Lefschetz formula 249 §25. The differential geometry of fibre bundles 251 25.1. G-connexions on principal fibre bundles 251 25.2. G-connexions on associated fibre bundles. Examples 259 25.3. Curvature 263 25.4. Characteristic classes: Constructions 269 25.5. Characteristic classes: Enumeration 278 §26. Knots andiinks. Braids 286 26.1. The group of a knot 286 26.2. The Alexander polynomial of a knot 289 26.3. The fibre bundle associated with a knot 290 26.4. Links 292 26.5. Braids 294 CHAPTER 7 : Some Examples of Dynamical Systems and Foliations on Manifolds 297 §27. The simplest concepts of the qualitative theory of dynamical systems. Two-dimensional manifolds 297 27.1. Basic definitions 297 27.2. Dynamical systems on the horus . 302 §28. Hamiltonian systems on manifolds. Liouville’s theorem. Examples 308 28.1, Hamiltonian systems on cotangent bundles 308 28.2. Hamiltonian systems on symplectic manifolds. Examples 309 28.3. Geodesic flows 312 28.4. Liouville’s theorem 314 28.5. Examples 317 §29. Foliations 322 29.1. Basic definitions 322 29.2, Examples of foliations of codimension 1 327 §30. Variational problems involving higher derivatives 333 30.1. Hamiltonian formalism 333 30.2. Examples 337Contents 30.3. Integration of the commutativity equations. The connexion with the Kovalevskaja problem. Finite-zoned periodic potentials 30.4. The Korteweg—deVries equation. Its interpretation as an infinite-dimensional Hamiltonian system 30.5 Hamiltonian formalism of field systems CHAPTER 8 The Global Structure of Solutions of Higher-Dimensional Variational Problems $31. 32. §33. Some manifolds arising in the general theory of relativity (GTR) 31.1. Statement of the problem 31.2. Spherically symmetric solutions 31.3. Axially symmetric solutions 31.4. Cosmological models 31.5. Friedman’s models 31.6. Anisotropic vacuum models 31.7. More general models Some examples of global solutions of the Yang-Mills equations. Chital fields 32.1. General remarks. Solutions of monopole type 32.2. The duality equation 32.3. Chiral fields. The Dirichlet integral The minimality of complex submanifolds Bibliography Index . xv 347 358 358 358 359 369 374 377 381 385 393 393 399 403 414 419 423CHAPTER 1 Examples of Manifolds §1. The Concept of a Manifold 1.1. Definition of a Manifold The concept of a manifold is in essence a generalization of the idea, first formulated in mathematical terms by Gauss, underlying the usual procedure used in cartography (i. the drawing of maps of the éarth’s surface, or portions of it). The reader is no doubt familiar with the normal cartographical process: The region of the earth’s surface of interest is subdivided into (possibly overlapping) subregions, and the group of people whose task it is to draw the map of the region is subdivided into as many smaller groups in such a way that: (i) each subgroup of cartographers has assigned to it a particular subregion (both labelled i, say); and (ii) if the subregions assigned to two different groups (labelled i and j say) intersect, then these groups must indicate accurately on their maps the rule for translating from one map to the other in the common region (i.e. region of intersection). (In practice this is usually achieved by giving beforehand specific names to sufficiently many particular points (i.e. land-marks) of the original region, so that it is immediately clear which points on different maps represent the same point of the actual region.) Each of these separate maps of subregions is of course drawn on a flat sheet of paper with some sort of co-ordinate system on it (e.g. on “squared” paper). The totality of these flat “maps” forms what is called an “atlas” of the2 1. Examples of Manifolds region of the earth's surface in question. (It is usually further indicated on each map how to calculate the actual length of any path in the subregion represented by that map, i.e. the “scale” of the map is given. However the basic concept of a manifold does not include the idea of length; i.e. as it is usually defined, a manifold does not ab initio come endowed with a metric; we shall return to this question subsequently.) The above-described cartographical procedure serves as motivation for the following (rather lengthy) general definition. 1.1.1. Definition. A differentiable n-dimensional manifold is an arbitrary set M (whose elements we call “points”) together with the following structure on it. The set M is the union ofa finite or countably infinite collection of subsets U, with the following properties. (i) Each subset U, has defined on it co-ordinates x4, a= 1,...,n (called local co-ordinates) by virtue of which U, is identifiable with a region of Euclidean n-space with Euclidean co-ordinates x4. (The U, with their co- ordinate systems are called charts (rather than “maps”) or local co-ordinate neighbourhoods.) (ii) Each non-empty intersection U, 7 U, of a pair of such subsets of M thus has defined on it (at least) two co-ordinate systems, namely the restrictions of (x5) and (x4); it is required that under each of these co- ordinatizations the intersection U,- U, is identifiable with a region of Euclidean n-space, and further that each of these two co-ordinate systems be expressible in terms of the other in a one-to-one differentiable manner. (Thus if the transition or translation functions from the co-ordinates xq to the co- ordinates x7, and back, are given by APE XQreeer Xe HSL (ty XE= Xp XR GEL, then in particular the Jacobian det(ax3/ax4) is non-zero on the region of intersection.) The general smoothness class of the transition functions for all intersecting pairs U,, U,, is called the smoothness class of the manifold M (with its accompanying “atlas” of charts U,). Any Euclidean space or regions thereof provide the simplest examples of manifolds. A region of the complex space C” can be regarded as a region of the Euclidean space of dimension 2n, and from this point of view is therefore also a manifold. Given two manifolds M=\|),U, and N=\),U,, we construct their direct product M x N as follows: The points of the manifold M x N are the ordered pairs (m, n), and the covering by local co-ordinate neighbourhoods is given by MxN=\JU,xV,, Pa§1. The Concept of a Manifold 3 where if x¢ are the co-ordinates on the region U,, and yf the co-ordinates on V,, then the co-ordinates on the region U, x V, are (x¢, yi). These are just a few (ways of obtaining) examples of manifolds; in the sequel we shall meet with many further examples. It should be noted that the scope of the above general definition of a manifold is from a purely logical point of view unnecessarily wide; it needs to be restricted, and we shall indeed impose further conditions (see below). These conditions are most naturally couched in the language of general topology, with which we have not yet formally acquainted the reader. This could have been avoided by defining a manifold at the outset to be instead a smooth non-singular surface (of dimension n) situated in Euclidean space of some (perhaps large) dimension. However this approach reverses the logical order of things; it is better to begin with the abstract definition of manifold, and then show that (under certain conditions) every manifold can be realized as a surface in some Euclidean space. We recall for the reader some of the basic concepts of general topology. (1) A topological space is by definition a set X (of “points”) of which certain subsets, called the open sets of the topological space, are distinguished; these open sets are required to satisfy the following three conditions: first, the intersection of any two (and hence of any finite collection) of them should again be an open set; second, the union of any collection of open sets must again be open; and thirdly, in particular the empty set and the whole set X must be open. The complement of any open set is called a closed set of the topological space. The reader doubtless knows from courses in mathematical analysis that, exceedingly general though it is, the concept of a topological space already suffices for continuous functions to be defined: A map f: X + Y of one topological space to another is continuous if the complete inverse image J ~*(U) of every open set U < Y is open in X. Two topological spaces are topologically equivalent or homeomorphic if there is a one-to-one and onto map between them such that both it and its inverse are continuous. In Euclidean space R", the “Euclidean topology” is the usual one, where the open sets are just the usual open regions (see Part I, §1.2). Given any subset Ac R*, the induced topology on A is that with open sets the intersections A > U, where U ranges over all open sets of R". (This definition extends quite generally to any subset of any topological space.) 1.1.2. Definition. The topology (or Euclidean topology) on a manifold M is given by the following specification of the open sets. In every local co- ordinate neighbourhood U,, the open (Euclidean) regions (determined by the given identification of U, with a region of a Euclidean space) are to be open in the topology on M; the totality of open sets of M is then obtained by admitting as open also arbitrary unions of countable collections of such regions, ie. by closing under countable unions.4 1. Examples of Manifolds With this topology the continuous maps (in particular real-valued func- tions) of a manifold M turn out to be those which are continuous in the usual sense on each local co-ordinate neighbourhood U,. Note also that any open subset V of a manifold M inherits, ie. has induced on it, the structure of a manifold, namely V=\), V,, where the regions V, are given by V=VoU, (2) (2) “Metric spaces” form an-important subclass of the class of all topological spaces. A metric space is a set which comes equipped with a “distance function”, i.e. a real-valued function p(x, y) defined on pairs x, y of its elements (“points”), and having the following properties: @) p(x, y= Ply, x (ii) p(x, x) =0, p(x, y) >Oifx#y; (ili) p(x, y)< p(x, z) + p(z, y) (the “triangle inequality”). For example n-dimensional Euclidean space is a metric space under the usual Euclidean distance between two points x =(x!,...,x"),y=(yl,...."): p(x y) = yy. A metric space is topologized by taking as its open sets the unions of arbitrary collections of “open balls”, where by open ball with centre xo and radius e we mean the set of all points x of the metric space satisfying p(xo, x) 0, and opposite orientations if J <0. Thus Euclidean n-space possesses two possible oriéntations. In the sequel we shall show that more generally any connected orientable manifold has exactly two orientations. 1.2. Mappings of Manifolds; Tensors on Manifolds Let M = (J, U,, with co-ordinates x7, and N =|), V,, with co-ordinates yf, be two manifolds of dimensions n and m respectively. 1.2.1. Definition. A mapping f: M — N is said to be smooth of smoothness class k, if for all p,q for which f determines functions YAlxpy « Xp) =S(x},...,xp)f, these functions are, where defined, smooth of smoothness6 1. Examples of Manifolds class k (i.e. all their partial derivatives up to those of kth order exist and are continuous). (It follows that the smoothness class of f cannot exceed the maximum class of the manifolds.) Note that in particular we may have N = R, the real line, whence m= 1, and f is a real-valued function of the points of M. The situation may arise where a smooth mapping (in particular a real-valued function) is not defined on the whole manifold M, but only on a portion of it. For instance each local co-ordinate x} (for fixed «, p) is such a real-value function of the points of M, since it is defined only on the region U,. 1.2.2. Definition. Two manifolds M and N are said to be smoothly equivalent or diffeomorphic if there is a one-to-one, onto map f such that both f: M—> N and f~':N—M,.are smooth of some class k>1. (It follows that the Jacobian J,, = det(dy4/dxs) is non-zero wherever it is defined, i.e. wherever the functions yf =f(x;,...,x5)4 are defined.) We shall henceforth tacitly assume that the smoothness class of any manifolds, and mappings between them, which we happen to be considering, are sufficiently high for the particular aim we have in view. (The class will always be assumed at least 1; if second derivatives are needed, then assume class > 2, etc.) Suppose we are given a curve segment » x=x(t),a Tyra defined as sending the velocity vector at x of any smooth curve x = x(t) (through x) on M, to the velocity vector at f(x) to the curve f(x(t)) on the manifold N. In terms of local co-ordinates x* in a neighbourhood of x e M, and local co-ordinates y* in a neighbourhood of f(x) € N, the map f may be written as Y=fxt,.jx", Paly...sm. It then follows from the above definition of the induced linear map f, that its matrix is the Jacobian matrix (dy*/dx*), evaluated at x, ic. that it is given by of? ay yf = Se on: be &, (5) For a real-valued function f: M — R, the induced map f, corresponding to each xe€M is a real-valued linear function (ie. linear functional) on the tangent space to M at x; from (5) (with m = 1) we see that it is represented by the gradient of f at x and is thus a covector. Interpreting the differential of a function at a point in the usual way as a linear map of the tangent space, we see that f, at x is just df. 1.2.4. Definition. A Riemannian metric on a manifold M is a point-dependent, . positive-definite quadratic form on the tangent vectors at each point, depending smoothly on the local co-ordinates of the points. Thus at each point x =(x},..., x4) of each region U, with local co-ordinates xf, the metric is given by a symmetric matrix (g{(xp,...,x3)), and determines a (sym- metric) scalar product of pairs of tangent vectors at the point x: Km) = BB nb = Xm o>, le? = <6, e),8 1. Examples of Manifolds where as usual summation is understood over indices recurring as superscript and subscript. Since this scalar product is to be co-ordinate-independent, i.e. BR eons = eS Sane it follows from the transformation rule for vectors that the coefficients g{9 of the quadratic form transform (under a change to co-ordinates xj) according to the rule 8 8x ino 4x77? Gxd av= (6) . The definition of a pseudo-Riemannian metric on a manifold M is obtained from the above by replacing the condition that the quadratic form be at each point positive definite, by the weaker requirement that it be non-degenerate. (It then follows from the smoothness assumption that, provided M is connected, the index of inertia of the quadratic form is constant (cf. §3.2 of Part I).) 1.2.5. Definition. A tensor of type (k, !) on a manifold is given in each local co- ordinate system x; by a family of functions OTH HO) of the points x. In other local co-ordinates xf (embracing the point x) the components T;!---**(x) of the (same) tensor are related to its components in the system xf by the transformation rule Cope onse = SFE, ORG OXF OXF yy alt= Ge Sk eae T ” All of the definitions and results of Chapter 3 of Part I pertaining to tensors defined on regions of Cartesian n-space, now apply without change to tensors on manifolds. A metric g,, on a manifold provides an example of a tensor of type (0, 2) (compare (6) and (7)). On an oriented manifold such a metric gives rise to a volume element Tay an = VII 20.1007 8 = det(Bap)s where €,, 2, is the skew-symmetric tensor of rank n such that é,2.__,= 1 (see §18.2 of Part I). It follows (as in §18.2 of Part I) that the volume element is a tensor with respect to co-ordinate changes with positive Jacobian, and so is indeed a tensor on our manifold-with-orientation. As in Part I, so also in the present context of general manifolds, it is convenient to write the volume element in the notation of differential forms (in arbitrary co-ordinates defining the same orientation): X= /|gl dx! a--+ A dx"§1. The Concept of a Manifold 9 A Riemannian metric dl? on a (connected) manifold M gives rise to a metric space structure on M with distance function p(P, Q) defined by p(P, Q) = min [ dl, yo dy where the infimum is taken over all piecewise smooth arcs joining the points Pand Q. We leave it to the reader to verify that the topology on M defined by this metric-space structure coincides with the Euclidean topology on M. It follows from the results of §29.2 of Part I, that any two points of a manifold (with a Riemannian metric defined on it) sufficiently close to one another can be joined by a geodesic arc. For points far apart this may in general not be possible, though if the manifold is connected such points can be joined by a broken geodesic. 1.3. Embeddings and Immersions of Manifolds. Manifolds with Boundary 1.3.1. Definition. A manifold M of dimension m is said to be immersed in a manifold N of dimension n> m, if there is given a smooth map f: M +N such that the induced map f, is at each point a one-to-one map of the tangent plane (or in other words if in terms of local co-ordinates the Jacobian matrix of the map f at each point has rank m). The map f is called an immersion of the manifold M into the manifold N. (In its image in N, self-intersections of M may occur.) An immersion of M into N is called an embedding if it is one-to-one. Abusing language slightly, we shall then call M a submanifold of N. We shall always assume that any submanifold M we consider is defined in each local co-ordinate neighbourhood U, of the containing manifold N by a system of equations , Pylsboo9)= 0, i" +> where rank (%) =n-—m, 7 : PIM, xt) * P with the property that on each intersection U, U,, the systems (f = 0) and (f7=0) have the same set of zeros. It follows that throughout each neighbourhood U, of N we can introduce new local co-ordinates Ypres Vp satisfying Yat PhO Dee SLIM XB10 1. Examples of Manifolds In terms of these co-ordinates the submanifold M is in each U, given by the equations while Vpreves yp Will serve as local co-ordinates on the submanifold M. 1.3.2. Definition. A closed region A of a manifold M, defined by an inequality of the form f(x) < 0 (or f(x) = 0) where f is a smooth real-valued function on M, is called a manifold-with-boundary. (It is assumed here that the boundary 0A, given by the equation f(x) =0, is a non-singular submanifold of M, i.e. that the gradient of the function f does not vanish on that boundary.) Let A and B be manifolds with boundary, both given, as in the pre- ceding definition, as closed regions of manifolds M and N respectively. A map g: AB is said to be a smooth map of manifolds-with-boundary if it is the restriction to A of a smooth map BUSN, Ol4=0 of an open region U of M, containing A. (If A is defined in M by the inequality f(x) <0, then U is usually taken to be U, = {x| f(x) 0. We conclude this section by mentioning yet another widely used term: a compact manifold without boundary is called closed. §2. The Simplest Examples of Manifolds 2.1. Surfaces in Euclidean Space. Transformation Groups as Manifolds A non-singular surface of dimension k in n-dimensional Euclidean space is given by a set of n—k equations fix',....x)=0, i=1,...,n 7k, () where for all x the matrix (@f,/dx/) has rank n — k. If ata point (x4, ..., x$).on this surface the minor J;,__;,_, made up of those columns of the matrix (0f,/dx/) indexed by j,,. 4» is non-zero, then as local co-ordinates on a neighbourhood of the surface about the point we make take Cy Oe A i, 2) where the hatted symbols are to be omitted (see §7.1 of Part I). Since the surface is presupposed non-singular, it follows that it is covered by the regions of the form U;, __;,_,, where this symbol denotes the set of all points of the surface at which the minor J;, . ;,_, does not vanish.§2. The Simplest Examples of Manifolds it 2.1.1. Theorem. The covering of the surface (1) by the regions Djinn VSS <0* 0, and lie in opposite orientation classes if det A <0. (Thus at each point x of the manifold M, there are exactly two orientation classes of ordered bases of the tangent space at x.) Since a frame t for the tangent space at x can be moved continuously from x to take up the positions of frames for the tangent spaces at nearby points, it makes sense to speak of an orientation class as depending continuously on the points of the manifold. We are now ready for our alternative definition of orientation. 2.1.2. Definition. A manifold is said to be orientable if it is possible to choose at every point of it a single orientation class depending continuously on the points. A particular choice of such an orientation class for each point is called an orientation of the manifold, and a manifold equipped with a particular orientation is said to be oriented. If no orientation exists the manifold is non- orientable. (Imagine a frame moving continuously along a closed path in the manifold, and returning to the starting point with the opposite orientation.) 2.1.3. Proposition. Definition 1.1.3 is equivalent to the above definition of an orientation on a manifold. ; Proor. If the manifold M is oriented in the sense of Definition 1.1.3, then at each point x of M we may choose as our orienting frame the ordered n-tuple (e,j,+-+»€nj) Consisting of the standard basis vectors tangent to the co- ordinate axes of the local co-ordinate system Xfyeeey XP on the local co- ordinate neighbourhood U, in which x lies. If x lies in two local co-ordinate§2. The Simplest Examples of Manifolds 13 neighbourhoods U, and U;, then we shall have two orienting frames chosen at x; however, since M is oriented in the sense of Definition 1.1.3, the Jacobian of the transition function from the local co-ordinates on U; to those on U, is Positive, so that (in view of the transformation rule for vectors) the two frames lie in the same orientation class. Conversely, suppose that M is oriented in the sense of Definition 2.1.2 above, and that there is given at each point x a frame lying in the orientation class of the given orientation of M. Around each point x there is an open neighbourhood (in the Euclidean topology on M, and of size depending on x) sufficiently small for there to exist (new) co-ordinates x!,...,x" on the neighbourhood with the property that at each point of it the standard (ordered) basis (e,,..., e,) of vectors tangent to the axes of x',..., x" in order, lies in the given orientation class; this is so in view of the continuity of the dependence of the given orientation class on the points of M. If we choose one such neighbourhood (with the new co-ordinates introduced on it) for each point of M, then their totality forms a covering of the manifold by local co- ordinate neighbourhoods; furthermore, the transition functions for the regions of overlap all have positive Jacobians, since at each point of such regions the standard frames lie in the same orientation class (namely the one given beforehand on M). This completes the proof. a 2.1.4. Theorem. A smooth non-singular surface M* in n-dimensional space R", defined by a system of equations of the form (1), is orientable. Proor. Let t denote a point-dependent tangent frame to the surface M*. Obviously the (ordered) n-tuple ¢ =(t, grad f,,..., grad f,-,) of vectors is linearly independent at each point (since the (co)vectors grad f; are linearly independent among themselves and orthogonal to the surface). Now choose t at each point of the surface M* in such a way that the frame 7 (for the tangent space of R") lies in the same orientation class as the standard frame (e1,..., @,). Since this orientation class is certainly continuously dependent on the points of R’”, so also will the orientation class of t depend continuously on the points of M*. This completes the proof. a The simplest example of a non-singular surface in R"*' is the n-dimensional sphere S", defined by the equation Meth it is a compact n-manifold. Convenient local co-ordinates on the n-sphere are obtained by means of the stereographic projection (see §9 of Part I). Thus let Uy denote the set of all points of the sphere except for the north pole N =(0,...,0, 1), and similarly let Us be the whole sphere with the south pole S=(0,...,0, —1) removed. Local co-ordinates (u}, ..., uj) on the region Uy are obtained by stereographic projection, from the north pole, of the sphere onto the hyperplane x"*! =0; similarly, projecting stereographically from14 1. Examples of Manifolds My (I+ eg (2) =7 Figure 1. Local co-ordinates on the sphere via stereographic projections. the south pole onto the same hyperplane yields co-ordinates (ui, ..., u3) for the region Us (see Figure 1). It is clear from Figure 1 that the origin and the two points uy(x) and us(x) in the plane x"*! are collinear, and that the product of the distances of uy(x) and us(x) from the origin is unity. From this and a little more it follows easily that the transition function from the co- ordinates (uj, ..., ux) to the co-ordinates (uj, ..., u3) is given by (verify it!) (= (H.-Y), 6) Sway? |S way while the transition functions in the other direction are obtained by interchanging the letters N and S in this formula. The n-sphere bounds a manifold with boundary, denoted by D"** and called the (closed) (n + 1)-dimensional disc (or ball), defined by the inequality S(x)=xt + -+x2,.- 150. Note finally that the sphere S" separates the whole space R"*! into two non- intersecting regions defined by f(x) <0 and f(x)>0. Finally (before turning to the consideration of the classical transformation groups) we introduce the concept of “two-sidedness”. 2.1.5. Definition. A connected (n — 1)-dimensional submanifold of Euclidean space R" is called two-sided if a (single-valued) continuous field of unit normals can be defined on it. We shall call such a submanifold a two-sided hypersurface, (See the remark below for the justification of this.) 2.1.6. Theorem. A two-sided hypersurface in R" is orientable. Proor. Let v be a continuous field of unit normal vectors to a two-sided hypersurface M. At each point of M choose an ordered basis t for the tangent space in‘such a way that the frame (z, v) and the standard tangent frame (e;,..-,@,) Of R* lie in the same orientation class of R". It follows that the orientation class of t must be continuously dependent on the points of M, yielding the desired conclusion. Qa§2. The Simplest Examples of Manifolds 15 Remark. It will be shown in §7 that any two-sided hypersurface in R" is defined by a single non-singular equation f(x)=0 (and hence is indeed a hypersurface), whence it follows that such a hypersurface always bounds a manifold-with-boundary. Somewhat later, in Chapter 3, it will also be proved that any closed hypersurface in R” is two-sided. A The transformation groups introduced in §14 of Part I constitute import- ant instances of manifolds defined by systems of equations in Euclidean space. Thus in particular: (1) the general linear group GL(n, R), consisting of all n x n real matrices with non-zero determinant, is clearly a region of R”; (2) the special linear group SL{n, R) of matrices with determinant + 1 is the hypersurface in R™ defined by the single equation det A= 1; (3) the orthogonal group O(n, R) is the manifold defined by the system of equations AAT=1. (4) the group U(n) of unitary matrices is defined in the space of dimension 2n? of all complex matrices by the equations AAT=1, where the bar denotes complex conjugation. In §14 of Part I it was shown that these groups (and others) are smooth non-singular surfaces.in R"™ (or R?"*); we can now therefore safely call them smooth manifolds. Note that all of these “group” manifolds G have the following property, linking their manifold and group structures: the maps g: GG, defined by y(g)=g7! (ie. the taking of inverses), and : G x GG defined by w(g, h) = gh (i.e. the group multiplication), are smooth maps. 2.1.7. Definition. A manifold G is called a Lie group if it has given on it a group operation with the property that the maps 9, defined as above in terms of the group structure, are smooth. All of the transformation groups considered in Part I are in fact Lie groups. 2.2. Projective Spaces We define an equivalence relation on the set of all non-zero vectors of R"*! (regarded as a vector space) by taking two non-zero vectors to be equivalent if they are scalar multiples of one another. The equivalence classes under this16 1. Examples of Manifolds relation are then taken to be the points of (real) projective space of dimension n, denoted by RP". (Each projective space comes with a natural manifold structure, which will be-precisely defined below.) We now give an alternative (topologically equivalent) description of RP". Consider the set of all straight lines in R"*! passing through the origin. Since such a straight line is completely determined by any direction vector, and since any non-zero scalar multiple of any particular direction vector serves equally well, we may take these straight lines as the points of RP”. Now each of these straight lines intersects the sphere S" (with equation (y°)* + --- +(y")? = lat exactly two (diametrically opposite) points. Thus the points of RP” are in one-to-one correspondence with the pairs of diametrically opposite points of the n-sphere. We may therefore think of projective space RP" as obtained from S" by “glueing”, as they say, that is by identifying, diametrically opposite points. (We note in passing the consequence that functions on RP" may be considered as even functions on the sphere S": f(y) =f(— y).) Examples. (a) The projective line RP’ has as its points pairs of diametrically opposite points of the circle S'. Since every point of the upper semicircle (where y>0) has its partner in the lower semicircle, we can obtain (a topologically equivalent space to) RP" by taking only the bottom semicircle (together with the points where x= +1) and identifying its end points x = +1. Clearly the result is again a circle; we have thus constructed a one-to- ‘one correspondence (which is in fact a topological equivalence) between RP! and the circle S! (see Figure 2). . The analogous construction can be carried out in the general case, i.e. for RP". One takes the disc D* (obtained as the lower half of the sphere S") and identifies diametrically opposite points of its boundary. (The case n =2 is illustrated in Figure 3.) -1e1 Figure 2 Figure 3§2. The Simplest Examples of Manifolds 17 (b) In §14.3 of Part I a homomorphism from the group SU(2) onto the group SO(3) was constructed, under which each matrix A of SU(2) has the same image as —A (ie. having kernel { +1}), or, in other words, identifying the points A and — A of the manifold SU(2) in the image manifold SO(3). We saw in §14.1 of Part I that there is a homeomorphism between the manifold SU(2) and the 3-sphere S? under which matrices A and —A are sent to diametrically opposite points of S*. Hence we obtain an identification (in fact topological equivalence) of SO(3) with projective 3-space RP?. We now introduce explicitly a (natural) manifold structure on ‘the projective spaces RP". For this purpose we return to our original characterization of RP" as consisting of equivalence classes of non-zero vectors in the space R"*! with: co-ordinates y°,..., y". For each q=0,1,...,n, let U, denote the set of equivalence classes of vectors (y°, ..., y") with y* # 0. On each such region U, of RP" we introduce the local co-ordinates xi, «++, Xq defined by 6) Clearly the regions U,, q =0, 1, ...,n, cover the whole of projective n-space. We next calculate the transition functions. For notational simplicity we do this for the particular pair Uo, U,: the general formulae for the transition functions on U;7U, can be obtained from those for Uy» U, by the appropriate replacement of indices. Now the co-ordinates in Uo are given by 1 2 bade tS, y y y and in U, by 0 2 BoB deh uted. y y y Hence in the region Ugo U,, where both y®, y! #0, the transition function from (x9) to (x,) is obviously rt xB 3 _ 38 2 xdaa xt = 78 = Nxt xb () (Note that x= y'/y® is non-zero on Up NU,.) The Jacobian of this transition function is given by ant #0. isoyn=tey = At | __%o 0... Of Gyr18 1. Examples of Manifolds Since, as noted above, the general transition functions on U,nU, are obtained similarly, it follows that RP" with the U, as local coordinate neighbourhoods is indeed a smooth manifold. The manifold RP? (with n = 2) is called the projective plane; in this case the region U, is called the finite part of the projective plane. Finally we note that, as is easily shown, the one-to-one correspondences S'-+ RP! and SO(3)— RP? described in the above examples, are in fact diffeomorphisms. We define complex projective space CP" similarly; its points are the equivalence classes of non-zero vectors in C"*' under the analogous equivalence relation (i.e. scalar multiples are identified), and the local co- ordinate neighbourhoods, with their co-ordinates, are defined as in the real case, making CP” a 2n-dimensional smooth manifold. By way of an example, we consider in detail the complex projective line CP?. Its points are the equivalence classes of non-zero pairs (z°,z') of complex numbers, where the equivalence is defined by (z°, z') ~ (4z°, 42") for any non-zero complex number 4. Consider the (complex) function Wo(z°, z') =2'/z°; this function is defined (and one-to-one) on all of CP! except (the equivalence class of) (0, 1): we shall formally define wo as taking the value oo at this point. Thus via the function wo the complex projective line CP' becomes identified with the “extended complex plane” (i.e. the ordinary complex plane with an additional “point at infinity”). 2.2.1, Theorem. The complex projective line CP' is diffeomorphic to the 2-dimensional sphere S?. Proor. On the region Ug of the complex projective line consisting of all equivalence classes of non-zero pairs (i.e. non-zero pairs determined only up to scalar multiples) (2°, z') with z° # 0, we introduce local co-ordinates uo, ¥9 defined by up + iv9 = Wo = z!/z2°. (These local co-ordinates may be regarded as defining a one-to-one map from Up onto the real plane R?.) Similarly, uy, 04, defined by u, + iv, =w, =z°/z', will serve as co-ordinates on the region U, consisting of pairs (z°,z') (up to scalar multiples) with z' #0. Clearly the regions Ug and U, cover CP'. The transition function from (up, V9) to (u,, »,) on the region of intersection is given by Ug U9 41, 0)=| 9 — aT > a (ata aia) or, in complex notation, by : 1 ug—ivo u, tiv, =w,=— = . Wo ua +B Since this formula coincides with the formula (5) (in the case n = 2) for the transition functions for the stereographic co-ordinates on the sphere S?, the theorem follows. ia§2. The Simplest Examples of Manifolds 19 It is on account of this result that the extended complex plane is often called the “Riemann sphere”. Note that if w=u+iv provides local co- ordinates u, v for the finite part of the extended complex plane (i.e. for the ordinary complex plane), then 1/w provides local co-ordinates of a (punc- tured) neighbourhood of the “point at infinity” 0. We now return to the consideration of the general complex projective space CP”. From each equivalence class of (n + 1)-vectors we may choose as representative a vector whose tip lies on the unit sphere S?"*", i.e. satisfying JP +o + [27 = 1, by simply multiplying any vector z=(z°,..., 2") in the class by the scalar A=(Y2-0lz4)" "7. The resulting vector (with tip on S"*') is then clearly unique only up to multiplication by scalars of the form e”, ie. by complex numbers of modulus 1. We therefore conclude that: Complex projective space CP" can be obtained from the (unit) sphere S?m*! = {2/2 olz*l? =1}, by identifying all points ez on the sphere (p variable) with z, i.e. by identifying all points differing by a scalar factor of the form e'*, Thus we have a map s*1_, Cp", (8) such that the pre-image of each point of CP" is (topologically equivalent to) the circle S' = {e'*}. In particular, in view of Theorem 2.2.1, we obtain thence amap S58 A2)rwHh (le? +2? =, 2.3. Exercises 1. Prove that the odd-dimensional projective spaces RP?**' are orientable. 2. Prove that the connected component containing the identity element of a Lie group, is a‘normal subgroup. 3. Prove that a connected Lie group is generated by an arbitrarily small neighbour- hood of the identity element. 4. Prove that every Lie group is orientable. 5. Prove that the projective spaces RP" and CP" are compact. 6. Quaternion projective space HP" is defined as the set of equivalence classes of non- zero quaternion vectors in H"**, where two (n + 1)tuples are equivalent if one is a20 1. Examples of Manifolds multiple of the other (by a non-zero quaternion). Define a manifold structure on HP", and verify that H/P' is diffeomorphic to S*. 7. Construct a mapping S***? -+ HP", analogous to the mapping (8), and identify the complete inverse image of a point of HP" under this map. §3. Essential Facts from the Theory of Lie Groups 3.1. The Structure of a Neighbourhood of the Identity ofa Lie Group. The Lie Algebra of a Lie Group. Semisimplicity ‘Every Lie group G (see Definition 2.1.6) has a distinguished point gp = 1¢G (the identity element), and, being by definition a smooth manifold, has a tangent space T= Tj, at that point. For each he G the transformation G + G, defined by g++ hgh™', is called the inner automorphism of G determined by h. Any such transformation of G clearly fixes the identity element go = 1 (since hgh™ ' = go), and therefore the induced linear map of the tangent space T to Gat the identity (see §1.2 above) is a linear transformation of T, denoted by Ad(h): T-> T. From the definitions of the inner automorphism determined by each element h, and the linear map of the tangent space T which it induces, it follows easily that Ad(h~') = [Ad(h)]~' and Ad(h, h,) = Ad(h,) Ad(h,), for alll h, hy, hz in G. Hence the map hr» Ad(h) is a linear representation (i.e. a homomorphism to a group of linear transformations) of the group G: Ad: G + GL{n, R), where n is the dimension of G. (Note that for commutative groups G the representation Ad is trivial, ie. Ad(h) = 1 for all he G.) We shall now express the group operation on a Lie group G in a neighbourhood of the identity, in terms of local co-ordinates on such a neighbourhood. We first re-choose co-ordinates in a neighbourhood of the identity element so that the identity element is the origin: 1 = go =(0,..., 0). We then express in functional notation the co-ordinates of the product g,g2 (if it is still in the neighbourhood) of elements g,=(x',...,x") and 82=(y',...,y") by VW HV IW CHL and the co-ordinates of the inverse g~' of an element g =(x',..., x") by GX) =9(',...x HL ym§3. Essential Facts from the Theory of Lie Groups 21 The functions w(x, y) (=g1g2) and (x) (=g~') obviously satisfy the following conditions (arising from the defining properties of a group): (i) W(x, 0) = HO, x) = x (property of the identity); (ii) W(x, @(x)) = 0 (property of inverses); (iil) W(x, Uy, 2) = W(x, 9), z) (associative property). Given sufficient smoothness of the function w(x, y), it follows from condition (i) (and Taylor’s theorem) that W"(x, y) =x" + y* + bg, xy” + (terms of order > 3) () (where of course bg, = 07y*/dx*dy’ evaluated at the origin). Now let € and » be tangent vectors to the group at the identity, i. elements of the space T, and as usual denote their components, in terms of our co-ordinates x*, by &, * respectively. The. commutator [£,]¢T of tangent vectors ¢, 7 is defined by CE, mJ" = (bj — Big)?” 2) This commutator operation on T has the following three basic properties: (a) (., ] isa bilinear operation on the n-dimensional linear space T (where n is the dimension of G); (>) [2,2] =—[n, 2]; ©) CC6,9),61+CCE6, 21, 2) + CL, 61, ¢] =0 (“Jacobi’s identity”). 3) (The first two of these properties are almost immediate from the definition of the commutator operation. Here is a sketch of the proof of (c) as a consequence of the associative law (iii) above: From (1) we obtain that UWE, y), 2) = W(x, y) + 2 + BG, W(X, y)2” + (terms of degree > 3 in y', z/). Substitution in this from (1) yields an expansion of *(W(x, y), z) in terms of x“, y’, 2” in which the coefficient of x" y’z” is 5,4. Repeating this procedure for "(x, (y, z)) and comparing the coefficient of x“y’z’ with that obtained in the case of w“(W(x, y), z), we find that Bp be = bapbey 4 On the other hand from the definition of the commutator we obtain CLE, n), C1 = (bf, — b5p)LE, IPC” = (bf, — bog )(bhy — BE, )E"N"C?. It follows that Jacobi’s identity is equivalent to (Bj) — D5p)(bhy — BE) + (fy — Dig) (Dha — Bf) + (Bjy — biig)(bE, — bf.) = 0, which is easily seen to be a consequence of (4), as required.) Thus the tangent space to G at the identity is with respect to the commutator operation a Lie algebra; since it arises from G it is called the Lie algebra of the Lie group G. (Cf. §24.1 of Part I.)22 ‘ 1. Examples of Manifolds Ife,,..., , are the standard basis vectors of T (in terms of the co-ordinates x',...,x") then since [e,, e,] is again a vector in T, we may write Leg, €,] = chyeas whence by bilinearity [é, a)" = 5,0", (5) for all vectors ¢,4 in T. The constants c§,, which clearly determine the commutator operation on the Lie algebra, and which are skew-symmetric in the indices B, y, are called the structural constants of the Lie algebra. A one-parameter subgroup of a Lie group G is defined to be a parametrized curve F(t) on the manifold G such that F(0)=1, F(t, + t2)=F(t,)F(t2), F(—t)= F(t). (Thus the one-parameter subgroup is determined by a homo- morphism t++ F(t) from the additive reals to G.) (Before proceeding we note parenthetically that left multiplication by a fixed element h of an abstract Lie group G defines a diffeomorphism G — G(g++hg); the induced map of tangent spaces is defined as before to send the tangent vector g(t) to a curve g(t), to the tangent vector (d/dt)(hg(t)) to the curve hg(t).) Now if F(t) is a one-parameter subgroup of G, then aF dFt+e)| _d — py Fol: Fe (eg ae OF Ono = FOG] je=0 where the last equality follows from the preceding parenthetical definition of the tangent-space map induced by left multiplication on the group by the element F(t). Hence F(t) = F(t)F(0), or F(t)~' F(t) = F(0), ie. the induced action of left multiplication by F(t)~! sends F(t) to F(0) = const. Conversely, for each particular tangent vector A of T, the equation F-\F=A 6 is satisfied by a unique one-parameter subgroup F(t) of G; to see this note first that (6) is (when formulated in terms of the function (x, y) defining the multiplication of points x, y of G) a system of ordinary differential equations, and therefore by the appropriate existence and uniqueness theorem for the solutions of such systems, has, for some sufficiently small e>0 a unique solution F(t) for |t| M(N, R). We leave it to the reader to verify that p, is actually a representation of the Lie algebra g, i.e. that it is a Lie algebra homomorphism: as well as being linear, it preserves commutators: pall, 0] = [046,047].§3. Essential Facts from the Theory of Lie Groups 29 (We note that it can be shown that if p is continuous, then it will automatically be smooth.) A representation p: G + GL(N, R) (or G+ GL(N, C)) is called faithful if it is one-to-one, i.e. if its kernel is trivial: p(g) #1 unless g=1. A matrix Lie group trivially has a faithful Lie representation (i.e. a representation which is also a topological equivalence). However, as we shall now show (by means of an example) not every Lie group can be realized (i.e, has a faithful Lie representation) as a matrix Lie group. As our example we take the group G=SL/(2, R) consisting of all transformations of the real line of the form nix xrex4 2nat 4 In (16) i 1—ze* where x eR, ae R, zeC, |z| <1, and In denotes the main branch of the natural logarithmic function, i.e. the continuous branch determined by In 1 =0. (Note that in (16) the argument of the function In is a fraction whose numerator and denominator are complex conjugates; hence the fraction has modulus 1, so that its natural logarithm is either zero or purely imaginary, and therefore the image of x in (16) is indeed real (in fact between — 7 and z).) It is not difficult to see that the group SL(2,R) is a connected 3-dimensional Lie group (with the obvious co-ordinates a and the real and imaginary parts of z). The subgroup isomorphic to Z consisting of those transformations (16) with ae Z and z=0, is easily seen to be central in the whole group SL(2, R) (ie. each of its elements commutes with all elements). (We shall see below that in fact it coincides with the centre of SL, R).) Note also for later use that the transformations (16) with ae R and z=0 forma one-parameter subgroup of SLE, R). Each transformation (16) has the property that if x-+y under the transformation, then x+2nk++y+2nk for all keZ. Hence each such transformation yields a transformation w = e*++ e” of the unit circle |w| = 1. It is easily verified that the latter transformation has the explicit form wre 2 gaia (17) 1-zw If one conjugates this by the linear fractional transformation z = i[(1 — w)/ (1 + w)], which maps the unit circle (with one point removed) to the real line (and the interior of the unit circle to the open upper half-plane) then one finds that the group of such transformations is isomorphic to SL(2, R)/{+1}. (Alternatively one may use the results of §13.2 of Part I to get that the group of transformations (17) is isomorphic to SU(1, 1)/{+1} ~ SL{2, R)/{+1}.) We thus have a homomorphism from our group SL(2, R) onto SL(2, R) with kernel the above central subgroup isomorphic to Z. Since SL(2, R)/{+ 1} has trivial centre, it follows that the centre of SL(2, R) is precisely that infinite cyclic subgroup.30 1. Examples of Manifolds 3.2.3. Theorem. The group SLO, R) has no faithful Lie representation. Proor. While the above-mentioned one-parameter subgroup (consisting of the transformations (16) with a R arbitrary and z= 0) clearly has infinite intersection with the centre of SL, R) (which consists, as just shown, of the transformations (16) with a € Z and z = 0), it is obviously not contained in the centre. We shall show that this is incompatible with the existence of a faithful Lie representation of SLO, R). Thus suppose that there is a subgroup G of GL(n, C) which is identical with SLO, R) as far as its Lie group structure is concerned. Denote by H the one- parameter subgroup of G corresponding to the one-parameter subgroup of SLO, R) just mentioned. It follows from §14.2 of Part I and §3.1 above that H has the form {exp tA|teR}, where A is some fixed nxn matrix. By conjugating, if need be, by a suitable matrix from GL(n, C) we may bring A into its Jordan canonical form; hence we may assume that A is in block diagonal form, with blocks each of the type A ay 0 , (18) where a; = 0 or 1. (We are supposing here that different blocks correspond to different A, i.e. that the degree of each block is equal to the multiplicity of the eigenvalue 4.) The matrix exp(tA) (see §14.2 of Part I) will then also be in block diagonal form with blocks of the same size as those of A, and with the block corresponding to (18) having the form e*'B,(t) where 1 at 4a,a,t? 3a,a,a5¢? a att 01 at 4a,a3t? 2 th! B= Since for infinitely many t the matrix with blocks e“B,(t) lies in the centre of G, it follows that every element of G also has the same block diagonal form as A, ie. has blocks of the same degree in the same order. The set P of alln xn matrices (including the singular ones) which commute with every element of G, clearly forms a linear subspace of the vector space C” of all n x n matrices; the intersection P 4 G is the centre of G. For reasons similar to before, every element of P again has the same block diagonal form as the matrices in G.§4. Complex Manifolds 31 The condition that any given n x n matrix of that block diagonal form lie in P is equivalent to the condition that its image in the quotient space C"/P be zero, and is therefore expressible as a homogeneous system of linear equations in the entries in the diagonal blocks of the given matrix, which can be so arranged that any single equation involves only those entries in a single block. Applying this to the matrix exp tA with blocks e“'B,(t), we obtain, on multiplying each equation by e~“ for the appropriate 4, a system of polynomial equations in ¢. Since such a system is satisfied by either all or else only finitely many values of t, this contradicts the fact that the group H = {exp tA} is not contained in the centre of G but has infinite intersection with it. Oo EXERCISES. 1. Calculate the Lie algebra of the Lie group SL(2, R). 2. Verify that the above-described group homomorphism SL(2, R) -» SL{2, R)/{ +1} is a local isomorphism in some neighbourhood of the identity. §4. Complex Manifolds s 4.1. Definitions and Examples We now introduce the concept of a complex manifold. 4.1.1. Definition. A complex analytic manifold of complex dimension n is a manifold M of dimension 2n, for which the charts U,(M = (J, U,) with their local co-ordinate systems z¢=x¢+iyg, «=1,...,n, ate identifiable with regions of n-dimensional complex space C”. It is further requited that on each region of intersection U, U,, the transition functions from the co-ordinates 7% to the co-ordinates 2§ and in the reverse direction, be complex analytic (see §12.1 of Part I): 624 azh ae” 6x4 We define a holomorphic map between complex manifolds to be one which is complex analytic (in terms of the given complex local co-ordinates on the manifolds). Holomorphic maps from a complex manifold to the complex line C will be called analytic or holomorphic functions on the manifold. A bijection between complex manifolds will be said to be biholomorphic if both it and its inverse are holomorphic. If two complex manifolds are such that there exists a biholomorphic map between them, we shall say that they are biholomorphi- cally equivalent or complex diffeomorphic. (1)32 1. Examples of Manifolds One important property of a complex manifold is that it always comes with an orientation. 4.1.2. Theorem. A complex analytic manifold is oriented. Proor. Let M be a complex manifold and let zg = xq + iyi, 2 = x5 + iy, be local co-ordinates on charts U,, U,, respectively. By Lemma 12.2.2 of Part I, on the region of overlap of each such pair of neighbourhoods, the (real) Jacobian of the transition function from the co-ordinates x4, yy to the co- ordinates x}, yp, satisfies a, (3) Since such Jacobians are therefore all positive, the theorem follows. Q FPa|seP= ? The complex projective spaces CP", introduced in §2.2 above, provide examples of complex analytic manifolds. Complex local co-ordinates on CP” are defined as in the real case; the transition functions, exemplified by formula (7) of §2.2 (with the x/ replaced by the complex co-ordinates z/) are clearly complex analytic. The manifolds C P" are compact (see Exercise 5 of §2.3). It follows from the discussion in §2.2 that the complex manifold CP! is biholomorphic to the extended complex plane with complex local co- ordinates w= 1/z in the neighbourhood of the point at infinity (and with w=Oat o). The simplest examples of complex manifolds are furnished by regions of C”. Further important examples are provided by the non-singular complex surfaces in C". Such a manifold is defined by a system of equations 2) ‘\ 7 Saale! -..,2)=0 where the functions f,,..., f,-, are all complex analytic, and at every point the rank of the matrix (af,/dz/) is largest possible (namely n—k). The verification that a non-singular complex surface in C” is indeed a complex analytic manifold is carried out in a manner analogous to that of the real case (see the proof of Theorem 2.1.1), with the aid of results from §12 of Part I. In contrast with the real case (see §9), compact complex analytic manifolds are not realizable as non-singular complex surfaces in some C". This is a consequence of the following theorem. 4.1.3. Theorem. A holomorphic function on a compact, connected complex manifold is necessarily constant. Proor. If f: MC is a holomorphic function on a compact, connected complex manifold M, then it follows by means of a well-known argument§4. Complex-Manifolds 33 (using the continuity of |f| and the compactness of M) that |f| attains a largest value; ie. there is a point Py of M such that |f(P)| <|f(Pp)| for all points P of M. The constancy of f on M follows from the connectedness assumption, and the following basic result of complex function theory (to state and prove which, we now interrupt the present proof). 4.1.4. Lemma (“Maximum-Modulus Principle”). Let f be a function holomorphic on some region U of n-dimensional complex space C". If the function | f| has a local maximum at some point Po of U, ie. if |f(P)| <\f(Po)| for all points P of U sufficiently close to Po, then f is constant in some neighbourhood of Po. Proor. Since the function | {| will clearly have a local maximum at Pp, on any complex line through Pp, it suffices to prove the lemma for the case n = 1. We may also assume without loss of generality that Pp =0, and that f(0) 40 (since if (0) =0, the assertion of the lemma is trivial). By multiplying the function f, if necessary, by an appropriate complex number we may further suppose that f(0) is a positive real number. In §26.3 of Part I we gave a proof of the “Residue Theorem” of complex function theory; the well-known “Cauchy integral formula” for holomorphic functions of a complex variable is an almost immediate corollary of that result: _ i f(z) dz IO=55 pe, where » is any circle enclosing the origin. Putting z=re'? where r is any constant small enough to ensure that both y and its interior are contained in U, this becomes ae 10)= 5 | sea, @) which formula obviously must also hold if in it f is replaced by Re f (or Re(f(0) —f(z)) is non-riegative on some neighbour- hood of the origin, since, by hypothesis, for all z sufficiently close to 0 we have -£(0)—|f(2)| = 0, and since also |Re f(z)| <|f(z)|. On the other hand since formula (3) continues to hold with f replaced by g, it follows that ae j g(re'*) dp =0. ° for all sufficiently small r. Hence throughout some neighbourhood of the origin g(z) must be identically zero, ie. Re f(z) =f(0). Since | f(z)| G, olg)=s"', are everywhere complex analytic (i.e. holomorphic). Thus the above groups are examples of matrix “complex Lie groups”. 4.1.6. Definition. A Lie group G which is a complex analytic manifold, is called a complex Lie group if the above maps w and are complex analytic. 4.1.7. Theorem. Every compact, connected, complex Lie group G is commutative.§4. Complex Manifolds 35 Proor. As usual we denote by g the Lie algebra of the group G. It is not difficult to see that the adjoint representation Ad: G + GL(n, C) is a complex analytic map (between complex manifolds). Since G is compact and connec- ted, Theorem 4.1.3 implies that Ad is a constant map; since Ad is a homomorphism, we must in fact have that Ad(g) = 1 for all geG. If g(t) is any (smooth) curve in G passing through the identity element, with &(0) =X say, and if Y is any element of g, then, as we showed in the proof of part (i) of Theorem 3.1.4, Ad(g(e))(¥) = ¥ + tLX, ¥] + O(?). Since Ad(g(t))(¥) = Y, we conclude that [X, Y] =0 for all X, Y eg, whence by Corollary 3.1.2, G is commutative, as required. Qo It can be shown that in fact the only compact, connected, complex Lie groups are the “complex tori”, which we shall now consider. As usual let 1,--+,€2, denote the standard basis vectors in R?*=C". (In fact for our purposes any basis for R” will serve.) The complex torus T?" is defined to have as its points the equivalence classes of vectors, where two vectors are equivalent if they differ by an integral linear combination of the given basis vectors: 2n z~zt+ yy ney neZ. a= (Such integral linear combinations form a subgroup I of C" called the integral lattice determined by the given basis e,,...,€2,.) Thus T2* is the quotient group of C" by I: . T# = CP. Obviously two integral lattices [ and I’, determined by bases e,,..., 2» and f;,..., fon Tespectively, coincide precisely if the vectors f, lie in T and the ein Tl’: Sante, — es=mif. Since the matrices (n}) and (mj) have integer entries and are mutual inverses, it follows that their determinants are both +1. Thus far we have defined the group structure of the torus T™". We now endow it with its manifold structure by taking as local (complex) co-ordinate neighbourhoods the images of appropriately chosen open subsets of C” under the natural map Ct T*=C"/F, where these open subsets are chosen on the one hand sufficiently small for the restriction to each of the natural map to be one-to-one, and on the other hand so that their images cover T?". We leave to the reader the details of the verification that with this manifold structure and the above abelian group structure, T" is indeed a complex Lie group.36 1. Examples of Manifolds Functions on T?" may obviously be regarded as 2n-fold periodic functions on C" f (: + ¥ nes) =f(2). It follows from Theorem 4.1.3 that: A holomorphic 2n-fold periodic function on C" is constant. By way of an interesting example we consider the special case n= 1. A complex torus T? is determined by a basis for R? = C, ie. by a pair of non- zero complex numbers z,,z, such that z, ¢Rz,. Multiplying by zj* we obtain a pair of the form (1, t) where t(= 2,/z, € C) is non-real (since 1 and t are linearly independent over R). Since, as it is easy to see, the multiplications of C by z, and z; ' induce holomorphic maps between the tori determined by the pairs (z;, 22) and (1, 2), it follows that those tori are biholomorphically equivalent. Hence each one-dimensional torus is determined, at least up to biholomorphic equivalence, by a complex number t with non-zero imaginary part.. 4.1.8. Lemma. If t and 7’ are two non-real complex numbers related by a linear-fractional transformation of the form _mt+n mt+q’ where the matrix (" ") is integral and has determinant +1, then the tori determined by t and v' are biholomorphically equivalent. Proor. In view of the conditions on the coefficients m, n, p,q, the integral lattices determined by the pairs of vectors (1, ) and (pt+q‘1, mt+n-1) coincide. The lemma now follows from the fact that the second pair defines a torus which, by the remark preceding the lemma, is biholomorphically equivalent to that determined by (1, 7’). a Remark. It can be shown (using the theory of elliptic functions) that tori determined by complex numbers t, 1’ sufficiently close to one another, are not biholomorphically equivalent. Regarded as merely a (2-dimensional) real manifold, the torus T? is diffeomorphic to the familiar 2-dimensional real torus S' x S!, where one of the two circles is obtained by identifying points on the straight line determined by 0 and z, which differ by an integral multiple of z,, and the other circle is obtained by carrying out a similar identification of points on the line through 0 and z2. Similarly, the torus T" is diffeomorphic to the 2n-dimensional real torus S' x --- x S! (2n factors). Returning to the complex case, suppose we have a torus T?" determined by a basis e,,...,@2, (not necessarily standard) for R2"=C". Among these 2n$4. Complex Manifolds 37 vectors there will be n which are linearly independent over C; by re-indexing if necessary, we may suppose that e;,..., e, are linearly independent over C. If we express the remaining vectors e,41,---, 2, in terms of the first n, say . free = bye, = k=1-. we obtain a complex matrix B =(b,,), which (as in the particular case n= 1 examined above) determines the torus up to a biholomorphic equivalence. It is easy to see.that the imaginary part of the matrix B must be non-singular, since otherwise the vectors €,,...,@2, would be linearly dependent over R. 4.1.9. Definition. A complex torus T?" is said to be abelian if for some basis €1,-++,@2, Of its integral lattice, the above-defined matrix B=(b,,) is symmetric and its imaginary part H = (h,,) = (Im b,,) is positive definite; i.e. if ba=by and hy,g*e/>0, for all non-zero real vectors (£1, ..., £"). For example, the one-dimensional complex torus determined (up to a biholomorphic equivalence—see above) by a complex number t with Imt >, is abelian; since the tori determined by t and —t clearly coincide, it follows that in fact all one-dimensional complex tori are abelian. However even for n=2 non-abelian tori exist. EXERCISE Show that almost all 2-dimensional complex tori T* are non-abelian. On an abelian torus the Jacobi-Riemann @-function 6(2,,...,2,) Of n complex variables, is defined by Bless +++ 20) = E exif 5 bum + Ema, 4 where the summation is over all n-tuples (m,,...,m,) of integers. The condition that the imaginary part of the matrix B = (b,,) be positive definite, guarantees convergence of the series. 4.2. Riemann Surfaces as Manifolds A Riemann surface is defined (cf. §12.3 of Part I) as a non-singular surface in C? given by an equation of the form f(z, w) =0, (5) where f(z, w) is an analytic function of z and w (for instance a polynomial in z and w). The condition for non-singularity, which makes the surface a one~38 1, Examples of Manifolds dimensional complex manifold (i.e. complex curve), is as follows (see §12.3 of Part I, especially Theorem 12.3.1): grade f= (£2) #0. If we solve equation (5) for w it may happen that we obtain a multi-valued function; for instance: (i) if f(z, w) = w? — P,(z) where P,(z) is a polynomial without multiple roots (see 12.3.2 of Part I), we obtain the two-valued function w= /P,(z) (a “hyperelliptic” Riemann surface); (ii) if f(z, w) =e” —z, we obtain w=In z =In|z| +i arg z + 2zin. (The geometric meaning of multi-valuedness of w(z) is that (some of) the surfaces z = const. meet the surface f(z, w) =0 in more than one point.) Consider the case where f(z, w) is a polynomial of degree n in the variables z,w. On making the substitution z = y'/y®, w = y?/y®, we obtain 1 z, w) = 5 0,9 y's Ys SG HAY) where Q, is a homogeneous polynomial in three variables. This furnishes a device for re-realizing our surface f(z, w)=0 in C?, as the surface in the projective space C P? given by the equation 2,9 y', y?) =0, 6) except that the points of the latter surface for which y° =0 correspond to “points at infinity” on the original Riemann surface (5). The adjunction of these points at infinity has compactified our surface: 4.2.1. Lemma. The Riemann surface in complex projective space CP? defined by equation (6), is compact. Proor. The set of zeros of Q, is clearly a closed set in CP?. Since CP? is compact, and any closed subset of a compact space is compact, the lemma follows. Thus the original Riemann surface f(z, w)=0 gives rise to a compact 2-dimensional real manifold. What do these manifolds actually look like in the case where f(z, w) = w? — P,(z), ie. when f(z, w) is as in (a) above? We first examine cases of low degree, and from these infer a general result. (It turns out that the points at infinity on such a surface are singular points, so that they may not all appear on the manifolds which we are about to construct (as realizations of such surfaces), while those which do appear should strictly speaking be removed from the manifolds.)§4. Complex Manifolds 39 Examples. (a) Let f(z, w) = w? — z; then Q,(y°, y', y?) =(y?)? — y!y®. We join the points z=0, z=oo in the domain of the (extended) multiple-valued function w= Jz by a line segment « (or to put it more vividly, we “cut” the Riemann sphere S?, diffeomorphic to CP', to obtain the sphere with the segment a removed, depicted in Figure 4). It is not difficult to see intuitively that the restriction of the (extended) surface f(z, w) = 0 to those z of S? off the slit «, consists of two disjoint connected components, each of which can be seen, by projecting onto the extended z-plane, to be diffeomorphic to the sphere S? with the line segment a removed (see Figure 5). (These connected pieces are called the “branches” of the multi-valued function.) At the points 0 and oo the (extended) function w = Va is single-valued. The desired surface is obtained by identifying the boundary segment a, of the component denoted I in Figure 5, with the boundary segment , of the region II, and the boundary segment f, of region I with the boundary segment «, of region II of the surface (as indicated in Figure 5). It is intuitively plausible that as a result of this cutting and pasting, we obtain a manifold diffeomorphic to S?. (b) We next consider the case f(z, w)=w?—P,(z), where P,(z) is a polynomial of degree 2 with simple roots z=2z,, z= 22, 2; #22. Join the points z,, z, by a straight line segment on the z-plane. For z outside that line segment the surface f(z, w) =0 falls into two disjoint connected parts. If we adjoin a point at infinity to each of these connected parts, they will be as shown in Example (a), with the difference that here z, # 00 (see Figure 6). As in that example, on identifying the appropriate boundary segments (a, ~ 62 and B,~«,), we see that the Riemannian manifold is in this case also diffeomorphic to S? (with two points removed). oO Figure 640 1. Examples of Manifolds q, Z, 29 °° Figure 7 Figure 8 Figure 9 (c) Consider f(z, w) = w? — P3(z) where P,(z) is a polynomial of degree 3 with distinct roots Zo, z;, 22. Make cuts on S? as indicated in Figure 7; for z off these slits the extended surface f(z, w)=0 again falls into two disjoint connected pieces, as shown in Figure 8. On identifying the appropriate edges of the slits on these two pieces (a, with 82, a, with B,,, with 62, y. with 6,,as indicated in Figure 8), we obtain the 2-dimensional torus (or “sphere-with- one-handle”—see Figure 9) with one point removed. (d) As a final example, consider f(z, w) = w? — P,(z), where P,(z) is a polynomial, of degree 4 with distinct roots z9,2,, 22,23. By cutting and pasting as in Example (c) (with z, playing the role of co), we again obtain the 2-dimensional torus. 4.2.2. Proposition. The Riemann surface of a function of the form w = ./ P,(z), where P,(z) is a polynomial of degree n without multiple roots, is diffeomorphic to a sphere with g handles where n = 2g + 1 or n= 2g + 2 (strictly speaking with certain points removed, namely those corresponding ‘to the points at infinity of the original surface.) Proor. Suppose first that n is even, and write n = 2g + 2. Pair off the roots of P,(z) arbitrarily, and join the members of each pair by an arc in such a way that no two arcs intersect (see Figure 10). If we cut the z-plane along each of these g + 1 arcs, i.e. if we remove the points on these arcs, then the surface falls into two disjoint connected parts U, and U,. (If we move around any pair of§5. The Simplest Homogeneous Spaces 41 ze %y Zn By @, Gye7 4 qo. le 2% 23 7 2n-1 Figure 10 roots on the original surface, we stay on the same branch.) The edges a, 8; of the ith cut lie on (or rather are boundary segments of) different connected pieces U,, U,. We now glue these edges back together as follows: (U1,4)~(U2,B), (U1, Bi) ~(U2, 4). (This is justified by the fact that if on the original surface we move along the piece U, approaching the edge a,, then on crossing it we pass smoothly over onto the branch U, (with corresponding edge ;), and similarly if we approach on U, the edge B;, we cross over onto U, (with corresponding edge a,).) ia For odd n the construction is similar, with z,,, = 00 taken as the (n + 1)st branch point. §5. The Simplest Homogeneous Spaces 5.1. Action of a Group on a Manifold We begin with the definition of such an action. 5.1.1. Definition. We say that a Lie group G (e.g. one of the matrix Lie groups considered in §14 of Part I) is represented as a (Lie) group of transformations of a manifold M (or has a left (Lie)-action on M) if there is associated with each of its elements g a diffeomorphism from M to itself xHT(x), xeM, such that T,, = 7,7, for all g, h e G (whence T, = 1), and if furthermore T,(x) depends smoothly on the arguments g, x (i.e. the map (g, x)++ T,(x) should be a smooth map from G x M to M). The Lie group G is said to have a right action on M if the above definition is valid with the property T,T, = T,, replaced by T,T, = Thy. If G is any of the Lie groups GL{n, R), O(n, R), O(p, 9), or GL{n, C), U(n), U(p, q) (where p + q =n), then G acts in the obvious way on the manifold R" or R7*=C" moreover, in these cases the elements of G act as linear transformations. (Note that if, more generally, a Lie group has a Lie action on”42 1. Examples of Manifolds the manifold R’, which is linear, then that action yields a Lie representation of the Lie group.) The action of a group G on a manifold M is said to be transitive if for every two points x, y of: M there exists an element g of G such that T,(x) = y. 5.1.2. Definition. A manifold on which a Lie group acts transitively is called a homogeneous space: of the Lie group. In particular, any Lie group G is a homogeneous space for itself under the action of left multiplication: T,(h) = gh; in this context G is called the principal (left) homogeneous space (of itself). (The action T,(h) = hg~' makes G into its own principal right homogeneous space.) Let x be any point of a homogeneous space of a Lie group G. The isotropy group (or stationary group) H, of the point x is the stabilizer of x under the action of G: H, = {g\T,(x) =x}. 5.1.3. Lemma. All isotropy groups H, of points x of a homogeneous space, are isomorphic. Proor. Let x, y be any two points of the homogeneous space and g be an element of the Lie group such that T,(x) = y. It is then easy to check that the map H,— H, defined by ht+ghg™' is an isomorphism (assuming a left action). oa 5.1.4. Theorem. There is a one-to-one correspondence between the points of a homogeneous space M of a group G, and the left cosets gH of H in G, where H is the isotropy group (and G is assumed to act on the left). Proor. Let xo be any point of the manifold M. Then with each left coset 8H,, we let correspond the point T,(xo) of M. It is straightforward to verify that this correspondence is well defined (i.e. independent of the choice of representative of the coset), one-to-one, and onto. Qo For right actions the analogous result holds with right cosets instead of left cosets. Remark. It can be shown under certain general conditions that the isotropy group H is a closed subgroup of G, and that the set G/H of left cosets of H with the natural quotient topology can be given a unique (real) analytic manifold structure such that G is a Lie transformation group of G/H. 5.2. Examples of Homogeneous Spaces (a) The group O(n + 1) clearly acts (in the natural way) of the sphere S” (defined as the surface in Euclidean space R"*' given by the equation “(xl +--+ +(x"*})? = 1), It is easy to see that this action is transitive, so that S" is§5. The Simplest Homogeneous Spaces 43 a homogeneous space for the Lie group O(n + 1) of orthogonal transform- ations of R"*!. The isotropy group of the point x=(1,0,...,0)eS" is comprised of all matrices of the form 10 ( “p Ae O(n). Hence by Theorem 5.1.4 above S" can be identified with O(n + 1)/O(n) (where the quotient denotes merely the set of left cosets of the isotropy group, which is not normal in O(n+ 1)). In fact S"= O(n + 1)/O(n), where = denotes diffeomorphism (cf. the above remark). The group SO(n + 1) is also transitive on S", and, analogously to the above, the isotropy group is isomorphic to SO(n), so that we may identify S" with SO(n + 1)/SO(n), and again S" = SO(n + 1)/SO(n), as quotient space of SO(n + 1). (b) From the definition of real projective space RP" as consisting of the straight lines through the origin in R"*', we obtain a transitive action of O(n + 1) on the manifold RP". The subgroup of orthogonal transformations fixing the straight line through O with direction vector (1,0,...,0) is comprised of all matrices of the form +1 0 ( a Ae O(n). Hence the isotropy group is isomorphic to the direct product O(1) x O(n), and, again essentially by Theorem 5.1.4, we have RP" = O(n + 1)/O(1) x O(n). (©) The additive group of reals R acts (transitively) on the circle S' = {e?**} in the following way: Tex errle+n, teR. From the equality e?*' = 1 it follows that the isotropy group is exactly the group of integers. More generally the group of all translations of R" (which group it is natural to denote also by R") acts transitively on the n-dimensional torus T” =(S'J,, in the following way: if y=(t,,...,t,)€R", and z=(e?",..., e2ln) is a point of the n-dimensional torus, define T(z) = (e2Hlort00, amend), Clearly the isotropy group consists of all vectors y with integer components, ie. the isotropy group of this homogeneous space is the integral lattice T of R". Hence (cf. §4.1) T'=RT.44 7 - 1, Examples of Manifolds (d) Stiefel manifolds. For each n, k (k M by setting Fry = SiS where x= (—T/2) (see Figure 11). This map has the following three important properties: (i) fr, , moves each point of y through a time-interval T along the geodesic (as indicated in Figure 11): A(T) ¥(t + T); (ii) fr, , parallel transports vectors along the geodesic; (iii) for any fixed geodesic y, the transformations f;, , with T variable, form a one-parameter subgroup of the isometry group G: Srittiy Sr ftar S-1.y=(fr.)*. From the last of these properties and §3.1 above, it follows that for each geodesic y the one-parameter subgroup fr, , of G has the form Sr,y = exp(TB,), ©) SS Figure 1150 1. Examples of Manifolds where B, is a certain vector in the Lie algebra g of G (namely the tangent vector to the curve fy, , at T = 0). We denote by L' the linear subspace of the algebra g spanned by the vectors B, € g, where y ranges over all geodesics through xp, and by L° the Lie algebra of the isotropy group H,, of the point Xo. It follows (essentially from the fact that corresponding to each direction on the tangent plane to M at Xo there is a geodesic through x9 with that direction (see §29.2 of Part I)) that g=L°+L' (direct sum of subspaces). (2) If y,, 72 are two geodesics through the point xq then it can be shown that for small ¢ the product funbant-ont-on sends x9 to a point whose distance from x, is of order e°. (This follows without difficulty from the properties of the Riemann curvature tensor. Verify!) We deduce from this (e.g. using formula (7) of §3.1) that we must have [B,,, By] € Lo, whence [L', L'] c L®. Suppose now that gy = exp(TA) is a one-parameter subgroup of G leaving Xo fixed (so that A € Lo). Let y be any geodesic through xo, and denote by > the image of y under the map g,. It then follows, again from the isometric property of the maps involved, that for small enough ¢ the map g, fr, ,g-. translates the points of the geodesic } along that geodesic (which of course also passes through xo). Hence the tangent vector to the one-parameter subgroup g, fr, ,g-. (with parameter 7), is in L,. We now look for this tangent vector. From the basic facts about Lie algebras described in §3.1 it follows that for any two elements X, Y of the Lie algebra of a Lie group we have 2 exp(tX) exp(tY) = exp (wx +Y¥)+ 5X, » Y) + higher-order terms). (This is a weak form of the “Campbell-Baker-Hausdorff formula”. It is an easy consequence of this that exp(tX) exp(tY) exp(—tX) = exp(tY + t?[X, Y] + higher-order terms). Putting t= 1, X =eA, Y= TB,, we deduce that the desired tangent vector is B,+e[A,B,]. Since B,eL', it follows that [A,B,]eL', whence [DP L]cL. We include these facts in the following 6.2.1. Lemma. With G and g = L° + L' as above, we have [L,b9}eL, [LY L°%eL', [LY L}c 1. Q) A Lie algebra which decomposes as the direct sum of two subspaces satisfying (3) is called a Z,-graded Lie algebra, since (3) can be rewritten as [LY LI] c Lit mea 2, (4)§6. Spaces of Constant Curvature (Symmetric Spaces) 51 6.2.2. Corollary. With G as above, and g = L° + L! (direct) its Lie algebra, the linear operator agg whose restriction to L° is the identity map 1, and whose restriction to L' is the reflection —1, is a Lie algebra automorphism (i.e. also preserves commutators). (The map a is an “involution”, i.e. o? = 1.) (The converse is also true. To each involuntary automorphism o of a Lie algebra g, there corresponds a Z,-grading g = L° + L' (direct sum of spaces) of the Lie algebra, where L° is the set of elements fixed by o and L! the set of elements which o negates.) In view of the homogeneity of the manifold M, the local geometry around any point x is determined by a scalar product on the tangent space Rj, at the point. (Here n = dim M.) We now elicit a certain property (familiar from Part I) which this scalar product must have. Note first that the tangent space Ri, can be identified naturally with the space L! c g, Let A be any element of L°, and consider the one-parameter subgroup g; = exp(TA). For each T we have the map Ad(gr):Et+er, 9 CEL. As was shown in the course of proving part (i) of Theorem 3.1.4, we have or = Ad(g7)(6) = & + TLA, 6] + O(T*), whence diy] FT pag 7 A G1 = Od ME. 6) In view of the fact that g7 is an isometry of M the inner product on L' should be invariant under-Ad(g7), i.e. Cor mr) =<); whence on differentiating with respect to T at T = 0, and using (5), we obtain <[A, £1, 2) + <6, [4,7]> =0. 6) This is the condition on the metric (ie. scalar product) on L! = Ry, that we were seeking. (Cf. the definition of a Killing metric in §24.4 of Part I.) 6.3. Symmetric Spaces of the First and Second Types In the preceding subsection we obtained what might be called the algebraic model of a symmetric space. In principle all symmetric spaces can be classified (in the framework of the classification of compact Lie groups). In the present subsection we consider the most important examples of such spaces.52 1. Examples of Manifolds The simplest examples of (simply-connected) symmetric spaces are provided of course by the Euclidean spaces R” and pseudo-Euclidean spaces Rj, , (where the curvature is zero). Here the group G is the full isometry group of R" (or Rj, ,). (For the structure of some of these groups see §4 of Part I.) The isotropy group H is O(n) (or O(p, q)), and the space L' = R® consists of the translations. We have, as in the general case, g= L° + L', with the usual commutator relations between L° and L! (as in Lemma 6.2.1), and with the additional relation [L, L']=0. Non-simply-connected symmetric spaces (earlier called “locally symmetric”) can be obtained from these simply- connected ones by identification under various discrete groups I consisting of translations (and possibly also involving certain reflections, as for instance in the case of the Klein bottle—see §18 below). In the remainder of our examples the group G will be semisimple. Recall from §3.1 that semisimplicity is equivalent to non-degeneracy of the Killing form on the Lie algebra g. Recall also that the Killing form ¢ , > on g is defined by: = —tr(AB). As a basis for the Lie algebra we may take 01 0 0 -1 0 a=(o oh (Po = (5 0): We then have§6. Spaces of Constant Curvature (Symmetric Spaces) 53 and «41,429 =—1, (Ai, Ax) = (Az, As) =0, (As, A3)=-2, (Ay, A) = Cg, A2) = 0. It follows essentially from the fact that the matrices in H = SO(2) have the form cosy sing -sing cosg/’ that the subalgebra L°cg is comprised of the matrices of the form 4(A, — Az). The subspace L' of g is spanned by the vectors A, + 42, As. It is easy to check that the inclusions (3) hold. The restriction of the Killing form to the subspace L! is positive definite; this reflects the positive definiteness of the metric on the Lobachevskian plane. EXERCISE Investigate the general cases S" and L". 6.4. Lie Groups as Symmetric Spaces A Lie group Q endowed with a Riemannian metric invariant under left and tight multiplications by group elements, can itself be regarded as a symmetric space. The isometry group of Q has a subgroup isomorphic to Q x Q, whose action on Q is defined by (81,82): g++ 81982". The isotropy group of this action is clearly the diagonal subgroup H = {(q, q)lq € Q}, which is isomorphic to Q: clearly H(1) = 1. For each qin Q the corresponding symmetry is defined by Sq xt gx gq. (Verify that this does indeed define a symmetry.) In particular, s,(x) =x~'. We shall examine in detail the case where Q is a compact connected subgroup of SO(m), with the Euclidean metric ¢A, B) = tt(AB"), (72) where BT denotes the transpose of the matrix B. (Recall that it can be shown that a Lie group is compact if and only if it is a closed subgroup of some O(m).) EXERCISE Show that the scalar product (7) is the Killing metric on SO(m) determined by the Killing form on its Lie algebra (cf. §24.4 of Part I).54 : 1. Examples of Manifolds (The formula for the curvature of the Killing metric was derived in §30.3 of Part I. It follows from that formula that the Ricci tensor R,, is positive definite. In the same subsection it was shown that the geodesics through the identity are precisely the one-parameter subgroups of Q.) As noted in §24.4 of Part I, the group SO(m) lies on the sphere of radius Jm (in the Euclidean space R”’ of all m x n matrices with the metric (7)), since for A € SO(m), we have AAT = 1, whence = m; thus SO(m) c s™*-} 6.4.1. Lemma. The (Euclidean) scalar product (7) is invariant under right and left translations (i.e. multiplications) by elements of SO(m). Proor. Let g € SO(m), and let A, B be any m x m matrices. Then . The bi- invariance implies that for all elements A, B, C of the Lie algebra L of Q, and all gy = exp(AT), we have ¢ = ¢B, C, (8) whence it follows, just as in the derivation of (6) above, that <(A, B], C> + =0. @) Now let 2,5, Za» be two metrics on Q satisfying (8), (9). Then any linear combination g,, — Ag,» will also be Ad-invariant (or equivalently skew ad- invariant). Let 4, be any eigenvalue of the pair of quadratic forms g,,, Zap, i.€. det(g., — 4, ¥25) = 0. (The symmetry of g,,, 4 implies that A, is real.) The subspace R,, of all eigenvectors corresponding to the eigenvalue , is easily seen (from (9)) to be a (non-zero) ideal of L, whence by the assumed simplicity of L, we must have R,,=L. Hence g,,= 4,8,» Since the Killing metric satisfies (9), the desired result follows. Qa§6. Spaces of Constant Curvature (Symmetric Spaces) 55 6.4.4. Corollary. Every simple subgroup Q of the group SO(m) endowed with the Killing metric, can be isometrically embedded in the sphere S™’-* endowed with a metric proportional to the usual metric on the sphere. 6.4.5. Corollary. Since the Ricci tensor Ry (determined by the Killing metric 8ap On a group) also satisfies (8), (9), it follows that for simple groups, Ray = Agap where 4 = const. Note finally that, as remarked above, it follows essentially from §30.3 of Part I that R,, is positive definite for compact connected Lie groups. This is true also for semisimple groups, since each such group is (locally) a direct product G, x --- x G, of simples, and the sign of A; is easily determined for each of the simple factors G;. 6.5. Constructing Symmetric Spaces. Examples We now return to general symmetric spaces. In the notation of §6.2 above write M=G/H, g=L°+L' (direct sum of subspaces), where M is a given symmetric space, G is its isometry group, L° is the Lie algebra of the isotropy group H, and L’ is identifiable with the tangent space Rx, to M at the point xo (fixed by H, ic. H = Hx). Recall also that by virtue of the homogeneity of M, its metric is determined locally by a metric on L* satisfying (6); in what follows we shall assume the metric on M to be obtained from the Killing form on g (see below). 6.5.1, Lemma. The subspaces L° and L' of the Lie algebra g are orthogonal with respect to the Killing form. Proor. From Lemma 6.2.1, it is immediate that for all A € L°, Be L' we have ad A(L9)c 19, ad A(L!) cL, ad BIL’) c L°, ad B(L®) cL. It follows readily (using a basis of g which is the union of bases for L° and L') that tr(ad A ad B) =, as required. oO We deduce at once from this that in terms of a basis for g of the kind just mentioned, the Killing form on g has the form (0) (g.)= (3 wy (10) where a, B range over the indices of the basis for L°, and y, 6 over the indices56 1. Examples of Manifolds of the basis for L'. (The form g‘} is often called the Killing form of the symmetric space M.) Since the Killing form (10) on g satisfies (9), so also does the form g‘9 on L°. Hence by the proof of Lemma 6.4.3, if L® is simple then the form g°9) will be a constant multiple of the Killing form on the algebra L°. However, in the important examples the algebra L° is not simple, but rather semisimple of the form L° = L? ® L9 where L° and L9 are simple. From Lemma 6.5.1 (with now L in the role of g), we see that in this situation the restrictions of the form g°9 to the factors L9, L9 will be constant multiples (by 4,, 4, say) of the Killing forms on those factors. It can be seen that if H is compact and the metric g‘¥) is positive definite, then with respect to suitable bases for L° and L' the matrices ad A, A € L°, are skew-symmetric. Hence (A, A) = —tr(ad A)? is positive, and therefore in view of —tr(ad A)? = —[tr(ad A)}o + tr(ad A)}.J, it follows that , > (A, ADzo. (i) Hence for compact H (and positive-definite metric on the symmetric space M) the restriction to L° (namely r7) of the Killing form on the Lie algebra g, is positive definite. (Cf. the fact that the Killing form on the Lie algebra of a compact Lie group (e.g. on the Lie algebra L° of H) is non-negative.) We see that in order to construct a symmetric space it essentially suffices to choose a suitable subalgebra L° of g on which the restriction of the Killing form of the enveloping algebra g is non-degenerate; then L’ is defined as the orthogonal complement of L° in g. However the inequality (11) greatly restricts the choice of L°: If the Killing form on g is indefinite (type II) then for symmetric spaces with Riemannian metric the subalgebra L° < g must be such that the restriction of the Killing form to its orthogonal complement is either positive or negative definite, and at the same time L° must be the Lie algebra of a compact group, and therefore of a subgroup of SO(m). Remark. A given symmetric space can be realized as a submanifold of the group G in such a way that the geodesics of M are geodesics also in the manifold G. This embedding can be obtained in any one of the following three (equivalent) ways: (i) by considering all one-parameter subgroups of G emanating from the identity in the direction of vectors Be L' (show that these geodesics sweep out a submanifold of G diffeomorphic to M); (ii) via the map @: M — G, defined by g(x) =s;,'s, (where s,,,s, are the appropriate symmetries); (iii) by means of an “involution” 6:G— G (by which we mean an anti- automorphism of the group (6(g,g2) = 3(g2)6(g,)) such that the map§6. Spaces of Constant Curvature (Symmetric Spaces) 57 induced on g is the identity map on L° and the negative of the identity map on L'); M c Gis then the image under the map g++ ga(g~'). EXERCISE Show the equivalence of these embeddings. What follows is a list of the basic examples:of connected symmetric spaces of type I. (As an exercise, find in each case the corresponding direct decomposition g = L° + L') (1) SO(2n)/U(n). (2) SU(n)/SO(n). (3) SU(2n)/Sp(n). (4) Sp(n)/U(n). (5) SO(p + q)(SO(p) x SO(q)). Grassmannian manifolds (including (6) SU(p + q)/(SU(p) x U(q)). the projective spaces and spheres). (1) Sp(p + a)/(Sp(p) x Sp(q)). The following are examples of symmetric spaces of type II (with positive- definite metric). (The simply-connected ones among such spaces turn out to have the topology of Euclidean space R".) (1) SO(p, 4)(SO(p) x SO(q)). (For q=1 this is the Lobachevsky space L?.) (2) SU(p, q)/(U(p) x SU(q)). (For q=1 this is the unit ball in C’, as a complex manifold; if also p = 1, this manifold is identifiable with L? = SU(1, 1)/U(1)) (3) Sp(p, /(Sp(p) x Sp(q)).. (4) SL{n, R)/SO(n). (5) SL{n, C)/SU(n). (6) SO(n, C)/SO(n, R). We conclude with a list of symmetric spaces of dimension 4 with metric of signature (+ —— —). (These spaces are of potential importance for the general theory of relativity since (by Corollary 6.4.5) the metric g,, satisfies the equation R,, — Ag, = 0 (see §37.4 of Part I). I. Spaces of constant curvature with isotropy group H = SO(1, 3): (1). Minkowski space Rf, 5. (2) The de Sitter space S,=SO(1,4)/SO(1,3); note that S, is homeomorphic to R x S?. Here the curvature tensor R is the identity operator on the space of bivectors A?(R‘): R = 1. (3) The de Sitter space S_ = SO(2, 3)/SO(1, 3); this space is homeomorphic to S! x R?, and its “universal covering space” §_ = §0(2, 3)/SO(1, 3) (see §18) is homeomorphic to R*. Here the curvature tensor R = —1.58 1. Examples of Manifolds IL. Reducible spaces (products of spaces of constant curvature): (1) H=S0(3); M=R, x M2__, where M3__ is a space’ of constant curvature, with signature (— — —). (2) H=SO(1, 2); M=R_ x M3__, where M3 __ is a space of constant curvature with signature (+ — —). (3) H =SO(2) x SO(I, 1); M=M?2_ x M2_, the product of two 2-dimen- sional spaces of constant curvatures. Ill. The symmetric spaces M, of plane waves. (For these the isotropy group is abelian, and the isometry group is soluble.) In terms of a certain system of global co-ordinates the metric has the form dP? = 2 dx, dx, + [(cos t)x} + (sin t)x3] dx2 + dx} + dx}, > cost>sint. In terms of the tetrad (see §30.1 of Part I) given by the 1-forms p=dx,, q=dx,+Kdx,, x=dx;, y=dx4, the curvature tensor is constant, of the form R= —4[cos t(p A x) @(p x) +sin t(p y)@(pa y)}. Remarks. 1. A simply connected symmetric space is uniquely determined by its curvature tensor at a point. To see this let R: A7(V)— A?(V) be the curvature tensor, and denote by 6 the Lie algebra of skew-symmetric linear operators on the space V generated by those operators of the form R(x, y), x, ye V. (Then b is the Lie algebra of the isotropy group (previously denoted by L°).) Let g denote the Lie algebra V + b, where the commutator operation on this direct sum of spaces is defined by Cu, a), (v, b)] = (av — bu, [a, b] + R(u, »)). Then in terms of the pair g, ) the structure of the original symmetric space is naturally reproduced on the symmetric space M=G/H. ° 2. The problem of classifying all curvature tensors of symmetric spaces with a given isotropy group H reduces to that of finding the H-invariant tensors R of the type of the curvature tensor, for which R(x, y) belongs to the Lie algebra of H for all x, y in V. 6.6. Exercises 1. Show that for symmetric spaces of type II with positive-definite metric, the dimension of the subalgebra L° of g is equal to the number of positive squares in the Killing form on g.§7. Vector Bundles on a Manifold 59 2. Show that in the complex case (e.g. where G = SL(n, C) or SO(n, C)) the numbers of negative and positive square terms in the complex Lie algebra g are equal, and dim L°=4dimg. Find the subalgebra L° of the Lie algebra g of the group G=SL(n, ©). 3. Show that for symmetric spaces of type II with positive-definite metric one always has M=G/H, where H is a maximal compact subgroup er G. Investigate the particular cases SL(n, R)/SO(n), SL{n, C)/SU(n). 4. Show that a simply-connected, symmetric space of type II always has the topology of Euclidean R". For the next few exercises, note that, as for Lie groups (see §30.3 of Part I), so also for symmetric-spaces do we have KREME treo =3CLE 1 (De, EGER, =L 5. Show that for spaces of type I, the Ricci tensor Rey is positive definite, and the “sectional curvature” (R(é, n)é, > (where €, n span a parallelogram of unit area) is non-negative. 6. Show that for spaces of type II the sectional curvature is non-positive. Deduce from this that a simply-connected, symmetric space of type II is topologically the same as R" (assuming the metric Riemannian). 7. Decide which of the 7 symmetric spaces of type I and 6 spaces of type II listed above have non-vanishing sectional curvature. Investigate the spaces S", CP”, H P* of type I, and the spaces L", SU(n, 1)/U(n), SL{n, R/SO(n), SL(n, C)/SU(n) of type IL. 8. Prove that in dimensions n= 2, 3 the only simply-connected, symmetric spaces with positive-definite metric are L', S*, R*. (Hint. Show that the isotropy group HG must be SO(n) (n = 2, 3), and thence deduce (for n = 3) the constancy of all sectional curvatures.) 9. Prove thata simply-connected symmetric space M with semisimple G = G, x --- x Gy (where the G; are simple) has the form M =(G,/H,)x --- x(Gi/Hy), with the metric decomposing as a direct product of metrics on the factors M, = G,/H,,each of which is proportional to the Killing metric on the subspace L} of the corresponding Lie algebra g; = L? + L}. §7. Vector Bundles on a Manifold 7.1. Constructions Involving Tangent Vectors From any n-dimensional manifold M we can construct a 2n-dimensional manifold, called the tangent bundle L(M) of M as follows. The points of the manifold L(M) are defined to be the pairs (x, ¢) where x ranges over the points60 1. Examples of Manifolds of M and é ranges over the tangent space to M at x. Local co-ordinates are introduced on L(M) in the following way. Let U, be a chart of M with local co-ordinates xj. Then in terms of the usual standard basis e, = 4/Ax§ (of operators on real-valued functions on M), any vector ¢ in the tangent a to M at a point x of U, can be written in terms of components as ¢ = Cfe,. A: a typical chart Uf of UM) we take the set of all pairs (x, €) where x Sages over U,, with local co-ordinates (Yas VE) = Og Mp Shee SO The transition functions on the region of intersection of two charts Ut and US (with co-ordinates x4) are then of the form exh Ob A= 0 B= (seh wt 44). The Jacobian matrix of such a transition function is then clearly Oy )= (Gj ap (#) i axh ) A=(22), H=(22 es), (% axe axa ax 7 whence the Jacobian is (det A)* > 0. This gives immediately the 7.1.1. Proposition. The tangent bundle L(M) is a smooth, oriented 2n-di- mensional manifold. Note by way of an example that the tangent bundle on a region U of Euclidean space R" is diffeomorphic to the direct product U x R". If the manifold M comes with a Riemannian metric, then we can delineate in L{M) a submanifold, the unit tangent bundle L,(M), consisting of those points (x, €) with |€| = 1. The dimension of L, is 2n — 1. (It is defined in L(M) by the single non-singular equation f(x, €) = Bap SE =1) Example. Each tangent vector & at a point of the n-sphere S" (defined in Euclidean space R"*! by the equation )'2~o (x)? = 1) is perpendicular to the radius vector x from the origin to the point x. Hence in the case M = S", the unit tangent bundle L,(M) of pairs (x,é) with |é|=1, is (intuitively) identifiable with the Stiefel manifold V,, 1, 2 (see §5.2). In particular for n = 2, the unit tangent bundle L,(S7) is identifiable with V4, , = SO(3) (which is in turn diffeomorphic to RP?—see §2.2). A smooth map f: M > N from a manifold M to a manifold N, determines a smooth map of the corresponding tangent bundles: LM) > LN), (x, 2) (f(s fas where f, is the induced map of the tangent spaces (see §1.2). We note briefly a few other constructions similar to that of the tangent bundle.§7. Vector Bundles on a Manifold 61 (i) One often meets with the manifold L,(M) whose points are the pairs (x, t) where t ranges over the straight lines through the origin, in the tangent space R" to M at the point xe M. (ii) Given any n-dimensional manifold M, we may construct from it the tangent n-frame bundle E = E(M) having as points the pairs (x, t) with xeM and t=(&,,..., ¢,) any ordered basis (ie. frame) for the tangent space to M at x. (iii) If M is oriented then £ = E(M) is defined as in (ii) except that the frames tare required to be in the orientation class determining the orientation of M. (iv) If M is a Riemannian manifold, then Ey = Eo(M) is defined as in (ii) with the frames t restricted to being orthogonal. Further examples of such constructions will be considered in Chapter 6. We now define the cotangent bundle L*(M) on a manifold M. The points of L*(M) are taken to be the pairs (x, p) where p is a covector (i.e. 1-form on M) at the point x. Local co-ordinates xj on a chart U, of M determine the local co-ordinates (x, pp.) on the corresponding chart of L*(M), where the Ppa are defined by P= Ppa dy (ie. they are the components of p with respect to the standard dual basis of 1-forms on U,). The transition functions from co-ordinates (x5, Pp.) to co-ordinates (x4, pgp) on U, WU, are as follows: (x8, Pap) = (ster The Jacobian matrix is then A? 0 x5 axe (“a 4) 4=(38)-9=(sad) whence the Jacobian is 1, and the manifold L*(M) is oriented. The existence of a metric g,, on the manifold M gives rise to a map L(M)— L*(M), defined by (x8, EH (x4, BapDE"), ite. by means of the tensor operation of lowering of indices (see §19.1 of Part I). Since the expression w = p,dx* (a differential form on M) is invariant under transformations of the form (1), it can be regarded as a differential form on L*(M). Its differential Q = dw = Yea dp, 0 dx* (see §25.2 of Part I) is a non-degenerate (skew-symmetric) 2-form on L*(M), which is, obviously, closed, ie. d2=0. We conclude that: The manifold L*(M) is symplectic. (Recall that in Part I we defined a symplectic space to be one equipped with a closed (skew-symmetric) 2-form.) axt 550m): ()62 1. Examples of Manifolds 7.2. The Normal Vector Bundle on a Submanifold Let M be an n-dimensional Riemannian manifold with metric g,,, and let N be a smooth k-dimensional submanifold of M. The normal (vector) bundle vy(N) on the submanifold N in M, is defined to have as its points the pairs (x, v) where x ranges over the points of N, and vis a vector tangent to M at the point x, and orthogonal to N at x (i.e. orthogonal to the tangent space to N at x, which 4s a subspace of the tangent space to M at x). Assuming (as always— see §1.3) that the submanifold N is defined by a non-singular system of (n — k) equations, then (as noted in §1.3) in terms of suitable local co-ordinates y',...,y" on M these equations take the simple form y‘*! =0,..., y"=0, and y',..., y* serve as local co-ordinates on N. In terms of such local co- ordinates y',..., yon M, the normal bundle v,(N) is then determined as a submanifold of L(M) by the system of equations ¥*2=0,...,¥=0, gag(=0, a=... k Since this system is non-singular (verify it!), it follows that v,(N) is an n-dimensional submanifold of L(M). Examples. (a) Let M be Euclidean n-space R", and suppose N is defined by the non-singular system of (n — k) equations AM =0,.. f= y=) where y',...,y" are Euclidean co-ordinates on R". Then the vectors grad f,,..., grad f,_, are at each point of N perpendicular to the surface N and linearly independent (see §7.2 of Part I). Hence we see that vp-(N) has the structure of a direct product: Vp(N) & Nx RYTK, More generally if N is defined as a submanifold of a manifold M by a non- singular system of equations Si) =0,..-5 fr-n(%) = 0, then at each point x of N the vector fields eis) = grad f(0) = ay, i=l,...,n—k, are orthogonal to N and linearly independent, whence any vector normal to Nat xe N has the form v = v'e(x). The correspondence (x, v) +(x, v',..., ¥) is then a diffeomorphism: vy(N) = N x RY, An important special case arises from the consideration of a manifold 4 with boundary, defined by an inequality f(x)<0 in M. Here N is the boundary 0A of A defined by the single equation f(x) =0, and of dimension§7. Vector Bundles on a Manifold 63 n—1.'The normal bundle to the boundary then decomposes as a direct product: vy(GA) = 0A x R. (b) Suppose M=N x N where N is a Riemannian manifold. A typical tangent vector to M at a point is then a pair (€, ) of tangent vectors to N. Define a Riemannian metric on M by setting (Ea, ta)s (S25 M2)> = Sas ba) + Cm Ma? Consider the diagonal A = {(x, x)|x € N} of M; this is 4 submanifold of M manifestly identifiable with N. The tangent vectors to A at any point will have the form (f, ¢); hence a tangent vector v=(¢, 7) will be perpendicular to A precisely if 0=C0, (8 > =< o +0 for all tangent vectors ¢ to N. since this is possible if and only if =—n, it follows that the vectors normal to the diagonal AXN have the form v=(&, —). Hence we conclude that: Vy x n(A) = L(N). (c) Let vy(N) be the normal bundle on the submanifold N of the Riemannian manifold M. We define a map h, the geodesic map from vy(N) to Mas follows. Let (x, v) be any point of v(N) and let y(t) be the geodesic of M emanating from x with initial velocity vector yv; thus }(0) = v. Then define h by A(x, v) = y(1). 7.2.1. Lemma. The Jacobian of the map h is non-zero at every point of vy(N) of the form (x, 0). Proor. We give the proof only for the case when M is the space R" with the usual Euclidean metric, and N is a hypersurface in R" given (locally) by parametric equations x'=x!(u!,...,u"~1), i=1,...,n. Then as local co- ordinates for the points (x,v)e€vg-(N) we may take the n-tuples (u',...,u"~1, 0), where x = x(u), v = tn(u); here n(u) is the unit normal to the surface N at the point x(u). In terms of these co-ordinates the geodesic map h is clearly given by h(u',...,u"~4, t) = x(u) + tn(u). Hence its partial derivatives are as follows: Gh Ox | On oh ad aa ae On putting t= 0 we obtain the non-singular Jacobian matrix (h/du, dh/dt) = (0x/du, n), whence the lemma. im64 1, Examples of Manifolds 7.2.2. Corollary. Suppose that the submanifold N is compact. Then the geodesic map h maps the region Vie = {(, v)I Iv] R by: f= te 7 x¢€U,(M), o(tp(x,M)) if xe U(M), where U,(M) is the region of M appearing in the corollary, and where the plus sign is taken if x lies in a particular one of the two disjoint connected regions comprising R" — M, and the minus sign if x is in the other. (It is here that we are using the two-sidedness of M in R".) Then M is defined in R" by the equation f(x) =0. Oo Figure 12CHAPTER 2 Foundational Questions. Essential Facts Concerning Functions on a Manifold. Typical Smooth Mappings. The present chapter is devoted to foundational questions in the theory of smooth manifolds. The proofs of the theorems will play no role whatever in the development of the basic topology and geometry of manifolds contained in succeeding chapters. Consequently in this chapter the reader may, if he wishes, acquaint himself with the definitions and statements of results only, without thereby sacrificing anything in the way of comprehension of the later material. The subject matter of the chapter falls into two parts. In the first part “partitions of unity”, so-called, are constructed, and then used in proving various “existence theorems” (which are in many concrete instances self- evident): the existence of Riemannian metrics and connexions on manifolds, the rigorous verification of the general Stokes formula, the existence of a smooth embedding of any compact manifold into a suitable Euclidean space, the approximability of continuous functions and mappings by smooth ones, and the definition of the operation of “group averaging” of a form or metric on a manifold with respect to a compact transformation group.- The second part, beginning with “Sard’s theorem”, is concerned with making precise ideas of the “typical” singularities of a function or mapping. This part will be found very useful in subsequent concrete topological constructions, so that the definitions and statements of results contained in it merit closer study. §8. Partitions of Unity and Their Applications We first introduce some notation. The space of all (real-valued) functions on a manifold M, with continuous partial derivatives of all orders, will be denoted by C%(M) (these will be our “smooth” functions); the supremum (i.¢. least66 2. Foundational Questions. Functions on a Manifold. Typical Smooth Mappings upper bound) of the values f(x) taken by a function f will be denoted by sup f(x); and supp f will denote the support off, i.e. the closure of the set of all points x at which f(x) #0. 8.1. Partitions of Unity We begin with a lemma concerning Euclidean space R". 8.1.1. Lemma. Let A, B be. two non-intersecting, closed subsets of Euclidean space R", with A bounded. Then there exists a C®-function y on R" such that (x)= 1 on A and g(x) =0 on B (see Figure 13). Moreover such a @ can be found satisfying 0<¢(x)<1. Proor. Let a, b be two real numbers with 0b, Wx)4 =1 for r?R" is the identification of U, with a region ¢,(U,) of Euclidean R". Let S, be any compact subset of U,. Since ¢.(U,) is an open subset of R”, there exists by Lemma 8.1.1 a smooth function J, on R" such that f(x)=1 on g,(S,) and supp f,¢ 9,(U,), ie. f,(x)=0 outside ~,(U,). Consider the function F,(P) on M defined by _ Shl@AP)) for PeU,, FAP= {é for P¢ U,. Clearly F,¢ C(M), F,(P)=1 on S,, and F,(P)=0 outside U,. : We are now ready to turn our attention to the compact subset C of M contained in the open subset V (as in the lemma). In view of the compactness of C we can find a finite collection of (possibly new) local co-ordinate neighbourhoods U,,..., Uy and compact subsets S,,..., Sy, such that N U U,cV. 1 Ny S.cU, Ccl)S, at a By what we have just shown, for each a=1,...,.N there exists a function F,¢C*(M) such that F,=1 on S, and F,=0 outside u,. The function F=1-[]¥.1 (1—F,) then belongs to C*(M), is identically 1 on C, and vanishes outside (J¥_, U,, so that certainly F(P) =0 outside V. a 8.1.3. Theorem (Existence of “Partitions of Unity”). Let M be a compact, smooth manifold and let {U,} (10 for all xe M. If we take 9, =w,/y, then these g, satisfy the requirements of the theorem. This completes the proof. a The family of functions g, is called a partition of unity subordinate to the covering {U,}. Remark. The assumption that the manifold M be compact is not essential. It is readily seen that the proof of the existence of partitions of unity carries over to manifolds having suitable “locally finite” coverings (such a covering being one for which there is a neighbourhood of each point intersecting only finitely§8. Partitions of Unity and Their Applications 69 many regions of the covering). Recall that a Hausdorff topological space is called paracompact if every open covering has a locally finite refinement which covers the space. Thus the above proof of the existence of partitions of unity works more generally for any manifold which is paracompact. 8.2. The Simplest Applications of Partitions of Unity. Integrals Over a Manifold and the General Stokes Formula The theorem on the existence of partitions of unity has useful consequences; we shall now consider some of these. For the sake of simplicity we shall assume throughout that the manifolds we deal with are compact. 8.2.1. Corollary. On any compact manifold a Riemannian metric can be defined. Proor. Let {U,}, 1R"; h,(B") = U,, be fixed co-ordinate maps, where B” is the unit open ball in R" (with fixed co-ordinates x',..., x"). Thus h, assigns co- ordinates to the chart U,. By choosing the U, sufficiently small and arranging the co-ordinatization appropriately, we may assume (by virtue of the Implicit Function Theorem) that every intersection @D m U, which is non-empty is