This action might not be possible to undo. Are you sure you want to continue?
Modern Differential Geometry for Physicists
World Scientific Lecture Notes in Physics
Published
Vol. 47: Some Elementary Gauge Theory Concepts H M Chan and S T Tsou Vol. 48: Electrodynamics of High Temperature Superconductors A M Portis Vol. 49: Field Theory, Disorder and Simulations G Parisi Vol. 50: The Quark Structure of Matter M Jacob Vol. 51: Selected Topics on the General Properties of Quantum Field Theory F Strocchi Vol. 52: Field Theory: A Path Integral Approach A Das Vol. 53: Introduction to Nonlinear Dynamics for Physicists H D I Abarbanel, et a/. Vol. 54: Introduction to the Theory of Spin Glasses and Neural Networks V Dotsenko Vol. 55: Lectures in Particle Physics D Green Vol. 56: Chaos and Gauge Field Theory T S Biro, et a/. Vol. 57: Foundations of Quantum Chromodynamics (2nd edn.) T Muta Vol. 59: Lattice Gauge Theories: An Introduction (2nd edn.) H J Rothe Vol. 60: Massive Neutrinos in Physics and Astrophysics R N Mohapatra and P B Pal
Vol. 61: Modern Differential Geometry for Physicists (2nd edn.) C J lsham
World Scientific Lecture Notes in Physics  Vol. 61
Modern Differential Geometry for Physicists
Second Edition
Chris J lsham
Theoretical Phvsics G r o w Imperial College of Science, fechnology and Medicine UK
World Scientific
Singapore New Jersey. London Hong Kong
Published by
World Scientific Publishing Co. Re. Ltd. P 0 Box 128, Farrer Road, Singapore 912805 USA once: Suite lB, 1060 Main Street, River Edge, NJ 07661 UKoflce: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublicationData A catalogue record for this book is available from the British Library.
First published 1999 Reprinted 2001
World Scientific Lecture Notes in Physics Vol. 61 MODERN DIFFERENTIAL GEOMETRY FOR PHYSICISTS (2nd Edition) Copyright 0 1999 by World Scientific Publishing Co. Re. Ltd.
All rights reserved. This book, or parts thereof; may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without wriften permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN 9810235550 ISBN 9810235623 (pbk)
This book is printed on acidfree paper.
Printed in Singapore by UtcFPrint
Preface
This book is based on lecture notes for the introductory course on modern, coordinatefree differential geometry which is taken by our firstyear theoretical physics PhD students, or by students attending the oneyear MSc course “Fundamental Fields and Forces” at Imperial College. The course is concerned entirely with the mathematics itself, although the emphasis and detailed topics have been chosen with an eye on the way in which differential geometry is applied t o theoretical physics these days. Such applications include not only the traditional area of general relativity, but also the theory of YangMills fields, nonlinear sigma models, superstring theory, and other types of nonlinear field systems that feature in modern elementary particle theory and quantum gravity. The course is in four parts dealing with, respectively, (i) an introduction to general topology; (ii) introductory coordinatefree differential geometry; (iii) geometrical aspects of the theory of Lie groups and Lie group actions on manifolds; and (iv) the basic ideas of fibre bundle theory. The first chapter contains a short introduction to general topology with the aim of providing the necessary prerequisites for the later chapters on differential geometry and fibre bundle theory. The treatment is a little idiosyncratic in so far as I wanted to emphasise certain algebraic aspects of topology that are not normally mentioned in introductory mathematics texts but which are of potential interest and importance in the use of topology in theoretical physics.
V
In preparing this part of the text. The theory of fibre bundles is introduced in chapter five. I have laid considerable stress on the basic ideas of ‘tangent space structure’. but the latter subject is not treated in any detail and readers not familiar with this topic should supplement the text at this point. This is fairly brief since many excellent introductions to the subject aimed at physicists have been published in recent years. thus allowing for a more extensive index. I have not included any material on Riemannian geometry as this aspect of the subject is well covered in many existing texts on differential geometry and/or general relativity. Chapter four is concerned with the theory of Lie groups. with a treatment that emphasises the theory of principle bundles and their associated bundles. some more algebraic.vi PREFACE The second and third chapters contain an introduction to differential geometry proper. and the action of Lie groups on differentiable manifolds. I was particularly conscious of the difficulty which physics graduate students often experience when being exposed for the first time to the rather abstract ideas of differential geometry. The final chapter contains an introduction to the theory of connections and their use in YangMills theory. The second edition of this book differs from the first mainly by the addition of the chapter on general topology. and I have included a few more worked examples. A number of short explanatory remarks have been added in places where readers and students have suggested that it might be helpful: I a m most grateful to all those who drew my attention to such deficiencies in the original text. My experience in teaching this subject for a number of years is that a firm understanding of the various ways of describing tangent spaces is the key to attaining a grasp of differential geometry that goes beyond just a superficial acquiescence in the jargon of the subject. In particular. and there is no great point in replicating that material in detail here. However. In addition. which I develop from several different points of view: some geometric. I have resisted the attention to . I have tried here to emphasise the geometrical foundations of the connection between Lie groups and Lie algebras. it has also been completely reset in LaTeX. I have taken the opportunity to correct misprints in the original text.
Chris Isham Imperial College. June 1998 .PREFACE vii add substantial amounts of new materialother than the chapter on topologysince I wanted to retain the flavour of the original as bona fide lecture notes that could reasonably be read in their entirety by a student who sought an overall introduction to the subject.
.
. .1.4. . . . . . . . . . . . . . .2 Lattices . . . .2 Metric Spaces . . . . . . . 14 18 23 1. . . . . .4. . 10 1. 40 1.1 Partially ordered sets . . . .3.4 Operations on metrics . 42 i x .2. . . . 1. . . . . . . . . . . . . . . . . 1. . . . . .2.1. . . . . . 14 1. . . . . 1. . . . .2. . . 23 1. . .2. . 37 1. . . . . . . . . . .5 A topology as a lattice . . . . . . . . .3 Partially Ordered Sets and Lattices . . . . . 25 1. . . 1. . . . .4 General Topology . . . . . . . . . . . .Contents 1 An Introduction to Topology 1. . . . . . . .3. .3 Examples of metric spaces . . . . .6 The lattice of topologies T ( X )on a set X . . . . . . . .1 Preliminary Remarks . .5 Some topological concepts in metric spaces . 1. . . . . . . . . . . . .2 The idea of a metric space . . . . . .4. . . . . . .4. . . . . . . . .1 An example of nonmetric convergence .4. . . . . . . 32 1. .3 Topological spaces . . . . .1 Remarks on differential geometry . . . 1. .4. 1 1 1 2 3 3 5 8 1. . . . . . . . 1. . . . . . . . . . .2. 1. .2 The idea of a neighbourhood space . 11 1. . . .4 Some examples of topologies on a finite set .1 The simple idea of convergence . . . . . . . .2 Remarks on topology . . . . . . .
. . . .1 Preliminary Remarks . 92 3 Vector Fields and nForms 97 97 97 3. . . . . . .4. . . . . . . 2.6 The tangent space T. .4. .2 68 70 The intuitive idea . . .4. . . . . . . . . .4.7 Some properties of convergence in a general topological space . . . . . . . . . .12 Frames and locales . . .3 3. . . . . . . . . .4. .3 Tangent Spaces .M . . .1. .2 Some examples of differentiable manifolds . . . . . .1 52 54 59 59 2. . . . .3.2. .4 The pushforward of an equivalence class of curves. . . .7 A simple example of the pushforward operation 91 2. . . . .4.2 The Main Definitions . . . . . .1 The main definition . . . . . . .2 The vector field commutator . . . . .1 2. . . . . . 48 1. . . . 102 hrelated vector fields . . . . . . . . . . . .1. . .10 The idea of a homeomorphism . . . . . . . . . . . . .2. . . . . . . . . .3. 51 1. .3. . . . . . . . . 2 Differentiable Manifolds 2. . . . . 64 2. . . . . . . 104 . 2. .2. . . . . 45 1. .8 The idea of a compact space . 60 2. . . . . . . . . . . . . . . . . . . . 70 A tangent vector as an equivalence class of curves 72 2. .3.1. . . . . .3 Differentiable maps . . . . .8 The tangent space of a product manifold . . . . . . 3.3. . . . . .1 Vector Fields . . . 79 2. 60 Coordinate charts . 1. . . . . . . . . . 90 2. . . . . . . . . . . . 77 2. .3.3 The vector space structure on T.3. . . . . . 3. . . . . . . . . 76 2. . . 46 1. . .3. . .X CONTENTS 1. . .V of a vector space V .11 Separation axioms . . . . . . .9 Maps between topological spaces . . 2. . . . . . . .5 Tangent vectors as derivations . . . . . . . . . . . . . .
.. .. . . 140 4 Lie Groups 4. . . . .. . .3 3.4 3. . . . . .. . . .4. . 170 . . . .5 3. . . . .. . . . . . .4. . .2. .. . . . 123 The pullback of a oneform . .4 General Tensors and nForms .1 Leftinvariant vector fields . .3 The exponential map . . . 135 The definition of the exterior derivative .. . .1 Complete vector fields . .3 3. . . . . .. . 138 3.I ) .. . .4. . . . .4 The Lie algebra of GL(n. . . . The algebraic dual of a vector space .. . . .2. . . 132 The idea of an nform 3.2. .. . .... . . . . . .. . . . 111 Local flows . . . .2 149 149 The first definitions . . . .. .2 3.1 4. . .CONTENTS x i 3. . .. ..4 Oneparameter groups of diffeomorphisms . . .2 3. 121 The main definitions . .2 The completeness of a leftinvariant vector field 162 4.. .2 Integral Curves and Flows . . . 169 R 4. . . . . . . . .3. 157 4. . . . .. . . . . . .. . . 107 3. . . 107 3. . . .3. . . . . . .. . . . . . . . . . . . .3. . . .3..1 3. 137 The local nature of the exterior derivative . . . 155 4. . . . . . . . .. . .4 .3 LeftInvariant Forms . . 132 . .2. . . . . . .1. . .. .2.2 3.2. . . . .2. . . . . 126 A simple example of the pullback operation . . . . . . . . ... . 4. 130 The tensor product operation . . . . .. 3. . .1. . . . . . . . ..3. . . . . . . . .. 157 4. .3 3. . . . . . . . . .2 The Lie Algebra of a Lie Group . . . . . . . . . . .1 The Basic Ideas . .3 Cotangent Vectors 3. . . .. . . . . . . .. 129 The Lie derivative . . . . . . . 165 4. . . . . . . . . . . . .2.. 149 The orthogonal group . . 115 117 121 Some concrete examples of integral curves and flows . .. .. . . .. . . . . . . . . . . . .1 3..4. .. .5 DeRham Cohomology . .
. . . . . . .1 Bundles in General . . 183 Some important transitive actions . . . .1. . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . .2 The main result . . .2 5.5. . .1. . . . . .2. .8 Introduction . . . . . .3. 201 The idea of a crosssection . . . . .2 4. . . . 224 5.1. . . . . . . . 214 The pullback operation . . . . . . . .5 5. . . . . . . . . 218 5.1. . . . . . . . . . . . .3. . .5 Infinitesimal Transformations . 185 4. . . . . . . . . . . . . . .3 5. . . . . .4. . . 179 The main theorem for transitive group actions . . . 216 Universal bundles . 213 Maps between bundles . . . . . . . .2. . . . . . . 207 Covering spaces and sheaves . . . .xii 4. . 170 The CartanMaurer form . . . . . . .4 5.1. . 220 Principal bundle maps . . . . . . . . . 199 199 The definition of a bundle . .4 Transformation Groups . .2 Principal Fibre Bundles 5. 232 Associated bundle maps . . .5. . 172 The basic definitions . . . 195 199 5 Fibre Bundles 5. . . . . . . . . .1. . . . 175 Different types of group action . . . .1. . . . .1 The induced vector field . . . . . .1 4. . . . . 232 5. .1 4. . . . . . . .3 Associated Bundles . . . . .4. . . . . 5. . . . .4 CONTENTS The basic definitions .4. . . . . . . . 190 4. . . . . . . . .1 5. . . . 230 5. . . 210 The definition of a subbundle .6 5. . . . . . . .1. .2 The main definition . . . . . . . . . . .3 4. . . . . . . . . . . . . .1 5. . . . .3. . .4. . . . . . . . .2 4. . . . . . . . . . . . .7 5.1 5. . . . . . 190 4.3 Crosssections of a principal bundle . 175 4. . 236 . . . . . . . . . . . . . .2.2 . . . . . . . . . . 220 The main definition . . . . . . . . . . . . . . .
. . .2 Vector bundles as associated bundles . . 271 277 BIBLIOGRAPHY INDEX 281 .1 Parallel transport in a principal bundle . . .1. . . . .2.1 The main definitions . . . . 246 248 5. . . . . . . 240 5. . . . 6. . . . 269 . . . . . 256 6. . . . . 249 6 Connections in a Bundle 253 6. . .4. . . . . . . . . 258 .2. 261 262 . .5 Crosssections as functions on the principle bundle .4 Connections in the frame bundle 6. 6. . . .2. 267 .1 The definition of a connection . . 6.3. . . . . . 6. 5. . . . . . . . . . 248 5. . . . .4 Riemannian metrics as reductions of B(M) . . . . . . . . 243 5. . . . . . . . . . . . . . . .4. . . .4 Vector Bundles . . .2 Parallel transport in an associated bundle . . . . . . .2. . .2 Parallel Transport . . . . . . . . . . . . . . 6. .2 Local representatives of a connection . . . .3. .CONTENTS xiii 5. .1 Connections in a Principal Bundle . .3 Local gauge transformations . 253 6. .1. .3 Covariant differentiation . . .3.1. .1. . 262 . . . .4 The curvature twoform . . . . . . . . . . . . . . . . . . 253 6. . . . .3 Restricting and extending the structure group . . .
.
with little reference to the fact that more than one coordinate system may be needed to cover a manifold. and YangMills theory. in canonical quantum gravity. this is far from being the only use of differential geometry in physics. say. where spacetime is represented mathematically by a fourdimensional differentiable manifold. However.1 1. although there are usually copious discussions of the effects on tensorial objects of changing 1 .1.Chapter 1 An Introduction to Topology 1. and the use of differential geometry in quantum field theory has increased steadily in recent decadesfor example. topological quantum field theory. In particular. However.1 Preliminary Remarks Remarks on differential geometry A physics student is likely to first encounter the subject of differential geometry in a course on general relativity. Such courses usually employ a very coordinatebased approach to the subject. Evidently. the nonlinear amodel. For example. the Hamiltonian and Lagrangian approaches to classical mechanics are best described in this way. an undergraduate course in general relativity can be rather misleading when viewed from the perspective of modern mathematics. superstring theory. no excuse is needed for teaching a course on differential geometry to postgraduate students of theoretical physics. the impression of the subject gained from.
This special collection of subsets is then used to give a purely settheoretic notion of characteristic topological ideas such as ‘nearness’. and is well worth studying in its own right. only rarely is it emphasised that the domains of two coordinate systems may differ. From a physical perspective. on a first encounter with the idea of a topology.2 Remarks on topology The subject of topology can be approached in a variety of ways. The modern approach to differential geometry is very different: although coordinate systems have an important role to play.2 CHAPTER 1. it is not obvious why that particular set of axioms is chosen rather ‘Two classic examples are Bourbaki (1966) and Kelly (1970). ‘convergence of a sequence’. What is neglected in such approaches to differential geometry is the fact that the topology of a manifold may be different from that of a vector space. At the most abstract level. and for this reason it is appropriate to begin any text dealing with modern ‘coordinatefree’ differential geometry with an introduction t o general topology and associated ideas. the key concepts are developed in a way that is manifestly independent of any specific reference to coordinates. However. . In fact. ‘continuity of a function’ etc. one could say that topology is concerned with the relation between points and ‘regions’: in particular.1. a ‘topology’ on a set X consists of a collection of subsets of Xknown as the open sets of the topologythat satisfy certain axioms (they are listed in Theorem 1. and thatfor examplethe familiar expression involving Jacobian transformations is really only valid on the intersection of the domains of the coordinate systems concerned. 1. A N INTRODUCTION T O TOPOLOGY from one coordinate system to another. Concomitantly.3). open sets are what ‘real things’ can exist in. the fact that a manifold is actually a special type of topological space becomes of greater importance. the subject of topology proper is of considerable significance in many areas of modern theoretical physics. and hencein particularit cannot be covered by a single coordinate system. Many excellent books on topology take an abstract approach from the outset’.
in general. But it should be emphasised that. 1. and then to show how this definition is narrowed t o give the standard axioms for general topology. We shall use the idea of the convergence of ’The notation x E X means that x is an element of the set X. However. . we would like t o say that an infinite sequence ( 5 1 .1 Metric Spaces The simple idea of convergence A key ingredient in any topologicaltype structure on a set X is the sense in which a point’ x E X can be said to be ‘near’ to another point y E Xwithout such a concept. what follows cannot be regarded as a comprehensive introduction to topology. the particular introduction to general topology given in Section 1.1. . and it should be supplemented with private study.) of points in X ‘converges’ to a point x E X if the elements of the sequence get arbitrarily near to x in an appropriate way. In addition.2 1. In particular. the points in X are totally disconnected from each other. . Other texts take a somewhat different approach and motivate the axioms for topology by starting first with a metric space: a special type of topological space whose underlying ideas are more intuitively accessible than are those of topology in general. metric spaces play many important roles in theoretical physics in their own right. METRIC SPACES 3 than any other. and for these reasons we shall begin with a short introduction to the theory of such spaces.2. The most I can do in the limited space available is to provide a quick introduction to some of the key ideas.4 is aimed at motivating the axioms for topology by starting with the broadest structure one can conceive with respect to which the notion of a converging sequence makes sense. Q. and the underlying motivation only slowly becomes clear.2. For this reason. I have also included topics that I feel are of potential interest in theoretical physics but which do not appear in the standard texts on topology: a good example is the lattice structure on the set of all topologies on a given set. .
in Section 1. Thus the disks3 Figure 1. 5The notation {z I P ( z ) }means the set of all 2 such that the proposition P ( z ) is true. the structure of a topological space is completely reflected by the convergent collections that it admits.4.z21. this is illustrated in Figure 1.1.4 CHAPTER 1 . the 3An equation of the form cy := p means that the quantity denoted by Q: is defined by the expression.1: A convergent sequence of complex numbers. in the latter case it is necessary to extend the discussion to include the idea of the convergence of collections of subsets of Xwith this proviso. will depend on E ) such that n > n implies o Izn .) ‘converges’ to z means that. As we shall see. B. A familiar example is provided by the complex numbers: the ‘nearness’ of one number z1 to another z2 is measured by the value of the modulus 1 z1. . on the right hand side. A N INTRODUCTION T O TOPOLOGY sequences to develop the theory of metric spaces and.(z) := { z’ E Q= I Iz . . and to say that the sequence (z1. general topological spaces. for all real numbers E > 0. p. That is. . z2. 4The symbol C denotes the complex numbers.z’I < E } ‘trap’ the sequence. .zl < E . in general. there exists an integer no (which.
1) or. ( .y3 Z IR” of and the associated balls Be(?) := { gE IR” I d(Z. 5 2 . aThe symbol Z denotes the vector with components ( 5 1 .y3 .2.2. (1. b‘ dea’. This gives rise to the following definitions. notes the usual ‘dot’ product a’. . Then a sequence of points (denoted Zn + Z) if ‘Lfor all E (1.y3 defined in Eq. If A c B and A # B then A is said to be a proper subset of B.5) 1.g) < E } . METRIC SPACES 5 convergence condition can be rewritten a s “for all E > 0 there exists no such that n > no implies z. .2.1.3). .2 The idea of a metric space The concept of a distance function can be generalised t o an arbitrary set X by extracting the crucial properties visavis convergence of the Euclidean distance d(Z.2. E BE(z)” (1.a&i. .2.3) d ( Z ]y3 := d ( Z .4) Zn E IR” is said t o converge t o Z E R” > 0 there exists no such that n > no implies Zn E BE(Z)”. in terms of the tails T.2) “for all E > 0 there exists no such that Tn0c BE(z)”. (1. 6The notation A c B means that A is a subset of B. ’The symbol R denotes the real numbers.sn).2. This notion of convergence can be generalized at once t o the space7 all ntuples of real numbers with the aid of the distance function (1. := { zk I k > n } of the sequenceI6 (1.2. This does not exclude the possibility that A = B. .2. b‘ := C:=.
y) 2 0. If Eq. Y) = for all z. A N INTRODUCTION T O TOPOLOGY Definition 1. there may be z # y such that d ( z . z = y d(z. It is important to know when two metrics can be regarded as being equivalent.(1.7) (1. d . 2.y) < E } . y.2. “for all E > 0. z E X. For example. there exists no such that Tn0 Bc(z)”. metrics d(’)(z..d) may not converge at all but.y) and d 2 ) ( zy) on a .b) where a E A and b E B.1 1. involved.(z) := { y E X I d(z. with d ( z . A metric on a set X is a mapg lo d : X x X the three conditions + l that satisfies R (1. if it does.z) d(z. 2.2.7) is replaced by the weaker condition “d(z. The set X itself is said to be a metric space. and the function d is a pseudometric. y ) = 0) then X is said to be a pseudometric space. a sequence may converge to more than one point (see later). gThe notation f : A + B means that f is a function (or map) from the set A to the set B.8) +. “If A and B are sets. and = 0 if.2. z ) = 0 for all z E X ” (ie.2. That is. convergence of a sequence in a metric space can be defined in terms of the tails of the sequence being trapped by the balls surrounding a point. This is defined to be the set of all (ordered) pairs (a.Y) d(Y. . We also write a e f(a) to denote that the particular element a in A is mapped to f ( a ) in B.6 CHAPTER 1. As in the example of the complex numbers. (1. Any given sequence of points in a metric space (X. Comments 1. Sometimes this term is applied to the pair ( X .9) c where B.y) 2 0. In more general types of topological space. z + z means . Y) I 4x7 4 + d(z. A x B denotes the Cartesian product of A with B.6) (1. 3. d ) if it is appropriate to make a reference to the specific metric function. and only if.2. it converges to one point only [Exercise!].
2. Of greater interest perhaps is when two metrics lead to the same set of convergent sequences.2. 2.11 for details. A metric d2)is stronger than a metric d(’) (or d(’) is weaker than d 2 ) ) if “for all I E I : X . then a d2)convergent sequence is automatically d(l)convergent [Exercise!]. If d 2 )is stronger than dl). d(l)(z. This motivates the following definition: Definition 1. Comments 1.2 1. and with each sequence converging to the same point in both metrics. It follows that equivalent metrics admit the same set of convergent sequences. 4.2.11) 2. Some of the material needed to prove this will be introduced later.L ( Y ) ) . it can be shown that if two metrics induce the same set of convergent sequences (with the same limits) then they are necessarily equivalent. METRIC SPACES 7 L : set X are said to be isometric if there exists a bijection such that. then the topology associated with d 2 )is stronger than that associated with d(’) in the sense that it has ‘more’ open sets.10) 3. Namely. for all z. see Definition 1.y)= d 2 ) ( L ( Z ) . A result of considerable importance is the converse to this.y E X.1. 3. For any given pair of metrics on a set X it is not necessarily the case that one is either stronger or weaker than the other. If the metric d2)is stronger than the metric d(’) in the sense above. A pair of metrics are equivalent if each one is stronger than the other. E’ >0 (1. for all E > 0. X + X ( 1. 0 . there exists such that B!?(z) c Bil)(z)”.
(1. If a differentiable manifold C is equipped with a Riemannian metric g . Of course. A metric can be defined on any set X by d ( z .2.CE ~ ~ IIi?ll for all i i V I)pi?ll= 1pl 11 61 for all p E C and V'E V 1 + (1.2. 4. I shall refer to the vector space IR". this will also mean using the particular metric Eq.13) 3. A N INTRODUCTION T O TOPOLOGY 1.'$..16) When IR" is viewed as a topological space in this way.2. y ) := 1 ifxfy 0 ifx=y. I shall refer to it as the euclidean space Rn.q) := J(.2.3 Examples of metric spaces 1.8 CHAPTER 1. y E C is defined to be d(x. 2.n yzl P 2 1..17) (1.12) where the infinurn is over all piecewise differentiable curves t ++ y ( t ) in C that pass through the points x and y.2.'.2.2. the set of ntuples IR" also has a natural vector space structure: if it is desirable to emphasise this property.2.14) (1.19) IIv'II> 0 with IIi?ll= 0 only if v'= 0.y3 d(. the distance between a pair of points x . Some equivalent metrics on IR" are d(Z.14). (1. We recall that if V is a complex vector space.2.2.15) := max 1x2 i s 1 .18) (1.(?y3  (1. . (1.y) := inf Y J (saa(r(t))'r"(t)'ra(t))' dt (1. a norm on V is a realvalued map ii H11 ii 11 that satisfies the three conditions (a) (b) (c) ~ ~ i ' + i ? ~ ~ ~ ~ ~Z. usually.
2.. 5. continuous functions on the closed interval [a. Another inequivalent metric is (1.2. A norm gives rise to a metric on V defined by .20) A sequence of vectors that converges with respect t o this metric is said t o be strongly convergent. (1. This is inequivalent t o the metric in Eq. (f(z))’ (1. METRIC SPACES 9 An example is a Hilbert space with 11 GI[:= Note that Eq.2.b].2. On any set X .21).1. let C ( [ ab].22) a sketched in Figure 1.2.23) 6. realvalued.51 . and it is a matter of considerable significance that many of these are metric spaces. d(Z.2.24) Then a distance function can be defined on f2(X)by . let [’(X) denote the set of all realvalued functions f on X such that (i) f(z) 0 for all but a countable set of z E X.b] := { T E R 1 a 5 T 5 b }. Spaces of functions play important roles in theoretical physics.R)denote the space of all .14) is an example of this construction on the vector space R”. G) :=[I a .R)by Another metric is (1. A metric can be defined on C([a. 1 (1.2.2. s (1. = (ii) c~~~ converges./m. For example.
. . g ) := If@)  d t > l< E. Some specific examples of such operations are as follows. i = 1 . 2 .2: A pair of functions satisfying d ( f . y E X . . a new metric. . 2. each of which is greater than or equal t o zero and such that at least one of them is nonzero. defines a metric on X if {al. A N INTRODUCTION TO TOPOLOGY Figure 1. a} is any set of real numbers.10 CHAPTER 1. If dl and d2 are a pair of metrics on X. . called the join of dl and d 2 . If di.4 Operations on metrics There are a number of ways in which metrics on a set X may be combined t o form a new metric. for all s.. 1. .2. 1. . . . can be defined by. n is a finite set of metrics on X then u2.
This can be remedied by defining instead the meet of dl and d2 to be. y ) := min(l. 3. if a metric is present on X this classification can be refined to one in which any z E X belongs to one of three categories.3.y). Thus if we are only interested in metrics up to equivalence. or (ii) it does not. . this fails t o satisfy the triangle inequality Eq. It is interesting to note that the set of all metrics on X forms a lattice under these two operations (see Section 1. .2 for a short introduction to lattices).1. if A is a subset of the metric space X. for all x. y ) ) but.y E X . y} of X. by 'topological' concepts I mean those dealing with the relations of points and subsets12.8).2.define d b ( z . = 2.(fI . [). 13The complement of a subset A of a set X is often written as X .thenin purely settheoretic termsevery point in X is one of just two types in relation t o A c X:either (i) z belongs to A. 12Theword 'topological' comes from the Greek TMTOCwhich. "In general. z . however.y)) for all x . a function f : A + R is said to be bounded if there exists some finite real number K > 0 such that < K for all a E A . means 'place'. nothing is lost by requiring them to be bounded functions on X x X. defined as follows (see Figure 1. y E X. METRIC SPACES 11 One might expect t o be able to use this pair of metrics to define another metric as min(dl(z. in which case it belongs to the complement A" of A defined as13 A" := { z E X 1 x @ A } . If d is any metric on X .A. Then db is a bounded" metric that can be shown to be equivalent to d . For example. . (1.3). roughly speaking. 1.d(z. However.5 Some topological concepts in metric spaces In the present context.d a ( z . where the infinurn is taken over all finite subsets {x = x1.2.5 .2.
Thus a limit point is either an interior point or a boundary point of A. AN INTRODUCTION T O TOPOLOGY Figure 1. The interior.(z) around z E X has the property that B E ( z C A.3: Illustration of an interior. z is an interior point of A if there exists E > 0 such that the ball B. y is an exterior point of A if there exists E > 0 such that the ball &(y) has the property that B. It is easy to see that Int(A) = Ext(A"). Ext(A) and Bd(A). ) 2. 3 .12 CHAPTER 1. exterior. both A and A". a point z E X is said to be a limit point of A if B E ( zn A # 0 for all E > 0. Definition 1. > 0.28) .2. and a boundary point. an exterior. and boundary of A are defined t o be the set of all interior.3 1. z is a boundary point of A if every ball B E ( z ) E .(y) n A = 8. Comments 1. intersects ) Furthermore. exterior and boundary points of A respectively. (1. they are denoted Int(A).
2. It is closed if it contains all its boundary points.32) A n Ext(A) = 0.2. These properties are of fundamental importanceindeed. { 2 E R 1 a 5 z < b } is neither open nor closed. A = Int(A).2. A set A is said to be open if it contains none of its boundary points.2.31) (1. In the example of the real line IR with its usual metric d ( z . Bd(A) = Bd(A"). (1. The analogous properties for closed sets are: 0 the intersection of an arbitraq collection of closed sets is closed. METRIC SPACES 13 Ext(A) = Int(A"). similarly { z E IR I a 5 x 5 b } is an example of a closed set.2. the interval { z E IR.2.1.30) plus Int(A) c A. the intersection of any finite collection of open sets is open. . a general topological space is defined to be a set X with a family of subsets that satisfy these three properties. y) := 3. A" is closed I .29) ( 1. 4. and only if. This will be discussed in detail later. The empty set 0 0 0 and X itself are both open.y ( . Note that [Exercise!] (i) a set A is open if. and only if. (ii) a set A is open if. The collection of all open sets in any metric space is called the topology associated with the space. (1. On the other hand.I a < x < b } (for any a < b) is an open z set. It possesses the following important properties: 0 the union of an arbitrary collection of open sets is open.
A N INTRODUCTION T O TOPOLOGY the union of any finite collection of closed sets is closed.3 Partially Ordered Sets and Lattices 1. .14 0 CHAPTER 1. The relevant concepts are ‘partially ordered set’ and ‘lattice’.3. for example. (ii) a point z E X is a limit point of a subset A if and only if there exists a sequence (x1. . the limit of every convergent sequence (xl. 5. In the latter context. the empty set 0 8 and X itself are closed. a 2sphere. This is the key to proving the result mentioned earlier that two metrics with the same set of convergent sequences are equivalent.’~ 14A useful reference with applications to quantum theory is Beltrametti & Cassinelli (1981). all of which could be considered to be defined on some common abstract 0 set X with the cardinality of the continuum. . .) in A that converges to z.1 Partially ordered sets In developing the general theory of topology. and only if.) of points in A itself lies in A .. It follows that Q. it is therefore significant that: (i) A cX is closed if and only if it contains all its limit points. . 6. the closed sets (and hence the topology) associated with a metric are uniquely determined by its collection of convergent sequences. The topology associated with a metric space is determined equally by either the collection of all open sets or the collection of all closed sets. it is useful to emphasise certain algebraic properties that arise naturally in this context. We will see later the precise sense in which a metric space is a special case of a general topological space. 1. a 2torus and a 1264sphere are coded entirely in their respective distance functions.22. . Thus the topological differences between. both of which play an important role in many branches of rnathematic~. Thus a subset A is closed if.
z E X . A relation R on a set X is a subset of X x X . A preorder on a set X is a relation that is reflexive and transitive.1.y E X either z 5 y or y 5 z. A partially ordered set (or poset) is a set X with a relation on X that is: 5 (Pl) Reffexive: for all 5 EX . 2. 1. PARTIALLY ORDERED SETS AND LATTICES 15 Definition 1 4 . 3. if z 5 y and y 5 z then x 5 z. if z 5 y and y 5 x = y. Note that any particular pair of elements z. and E X is said to be Rrelated to y E X (denoted x R y ) if the pair (5. This is denoted diagramatically by X It is clear that a finite. y) E R c X x X . Thus a poset is a preordered set whose associated relation is also antisymmetric. however. However.z 5 z.y E X may not be related either way. . there are many relations I that are not derived from functions.3.y E X .f(z)) z E X } . if it is true that for any z. An element y in a poset X covers15another element z if z 4 y and there is no z E X such that z 4 z 4 y. Note that a function f : X + X defines a relation {(z. then X is said to be totally ordered. y . then (P3) Tkansitive: for all x . 151t is perfectly possible for an element to cover more than one element. The notation z 4 y will be used if it is necessary to emphasise that z 5 y but z # y. partially ordered set is determined uniquely by its diagram of covering elements. J: (P2) Antisymmetric: for all z.
It should be noted that any equivalence relation R on a set X partitions X into disjoint equivalence classes in which all the elements in any class are equivalent to each other. there always exists a third one that lies between them. y. However. Note that no r E IR possesses a cover since. 2. the integers Z are also totally ordered with respect to 5. If X is any set. given any pair of real numbers. Examples 1. A preorder 5 can be defined on the set Metric(X) of all metric functions on a set X by saying that d(') 5 d2) the open balls for if the two metrics satisfy Eq. An important example in theoretical physics is the set of gauge orbits of the action of a gauge group on the space of vector potentials in a YangMills theory. they admit the same set of convergent sequences). or equal to.11). AN INTRODUCTION T O TOPOLOGY 4. namely n + 1. On the other hand. y E X. For later use we recall also the definition of an equivalence relation on a set X. 3. topologies in general can be partially ordered by the relation of one being stronger than the other. the set of all subsets of X is denoted P(X) and is known as the power set of X . As we shall see later. b E IR. zRy implies yRz. The set of all such equivalence classes is denoted X/R. zRz. . the preorder 5 is not a partial ordering on the set Metric(X) since it is not antisymmetric: d') 5 d 2 )and d2)5 d(') do not imply that d(') = d2) but only that the two metrics are equivalent (that is. This is a relation R that is: ( E l ) Reflexive: for all z E X. and only if. (E3) Transitive: for all z. in which a 5 b means that a E IR is less than. Thus A c X if. zRy and yRz implies zRz. This means that the topology is associated with d2) stronger than that associated with d'). (E2) Symmetric: for all z. z E X. The real numbers IR are totally ordered with respect to the usual ordering 5 . (1. but each n E Z does possess a unique cover.2.16 CHAPTER 1.
c} is b.{ b } } : b} 0 while the diagram for X = {a. In this example. N ( I c ) P(P(X)).3.1) where A . Similarly.1. equivalently. or. {a}. with P(X) = (0. In the general theory of topological convergence (to be developed later) the central idea is to associate with each IC E X a collection N ( x ) of subsets of X (the 'neighbourhoods' of the point z) that determine whether or not a sequence converges to IC. PARTIALLY ORDERED SETS AND LATTICES 17 A E P ( X ) . the C E collection N := { N ( z )I IC E X} can be regarded as a subset of P(P(X)) or as an element of P(P(P(X))). The set P(X) has a natural partial ordering defined by A 5 B means A C B (1. The simplest nontrivial example is the partialordering diagram for the twoelement set X = {a. a cover of a subset A is any subset of X that is obtained by adding a single point to A .3. X. 0 . B C X . Thus N ( I c ) P(X).
in this particular example there is important extra structure given by the propositional functions ‘and’ and ‘or’. let M be a spacetime manifold equipped with a Lorentzian metric. define q 5 p if p lies in the causal future of q: i e .. q can be joined to p by a path whose tangent vector is everywhere timelike or null (Kronheimer & Penrose 1967). Specifically.e. given any two propositions a and b. and are captured in the following definitions: Definition 1. a meet (or greatest lower bound) of a. (b) a A b is the greatest such lower bound: i e .3. given any pair of propositions a and b we can form the propositions ‘ a and b’ and ‘ a or b’. King & McCarthy 1976.18 CHAPTER 1. . Malament 1977). if a is true then b is necessarily true. more precisely. Analogues of these and related operations arise in many different situations. andrather remarkablythe entire metric (up to an overall conformal factor) can be recovered from this ordering (Hawking. 1. usually denoted a A b and a V b respectively.2 Lattices One of the central branches of mathematics in which a poset structure occurs naturally is propositional logic where. However. b E P is an element a A b E P such that: (a) a A b is a lower bound of a and b: thus a A b 5 a and aAb5b. the relation a 5 b is defined to mean that a logically implies b. This is a partial ordering. Then if q . A N INTRODUCTION T O TOPOLOGY 4. This feature has been behind a variety of suggestions that spacetime should s be regarded a a discrete set but still with a causal structure/partial ordering16. In any poset P . . i. Partially ordered sets play an important role in classical general relativity. see Sorkin (1991) . p E M . if there exists c E P such that c 5 a and c 5 b then c 5 a A b. “Rafael Sorkin has been one of the most articulate exponents of this idea. for example.5 1.
The definition of a greater lower bound is analogous.0 5 a. A join (or least upper bound) of a . A null element in a lattice C is an element 0 such that. The lattice is complete if a greatest lower bound and a least upper bound exist for every subset17 S of L (all that is guaranteed by the definition of a lattice is that these bounds will exist for all finite subsets of L ) . .a 5 1.3. . 3. (i) (a‘)‘ = a (ii) a (1. (a) a V b is an upper bound of a and b: thus a Note that. 5.4) (1. for all a E C. we have a A ( b V c) and = ( a A b) V ( a A c) ( 1. E L is any other element such that a 5 2. a join or meet is necessarily unique [Exercise!]. if there exists c E P such that a 5 c and b 5 c then a V b 5 c. these bounds are denoted A S and V S respectively.6) 5 b implies b 5 a’ ’ (iii) a V a’ = 1 and a A a = 0. b E P .3. b E P is an element a v b E P such that: 5 a V b and b5aVb. if it exists. A map a t)a’ of a poset P is an orthocomplementation if. for all a . PARTIALLY ORDERED SETS AND LATTICES 19 2.3. A lattice is a poset L in which every pair of elements possesses a join and a meet.5) (1. a 5 u . A unit element in a lattice C is an element 1 such that.1. ‘?By a least upper bound of a subset S of L is meant an element u E L such that (i) for all a E S. and (ii) if 2. 4. for all a E C. for all a E S then u 5 v.3. (b) a V b is the least such upper bound: i e . If they exist.3. for all a . 6.3) a V ( b A c ) = ( a V b) A ( a V c). A lattice is distributive if. ’ A poset with an orthocomplementation operation is said to be orthocomplemented. b.3.2) (1. c E C.
Thus 1 and 0 are absorptive elements for the Vsemigroup and the Asemigroup respectively. If C is distributive then any complement a' of an element a E 13 is unique.7) 1 V a = l . 2. AN INTRODUCTION T O TOPOLOGY 7. and hence C is also a semigroup with respect to the Voperation with 0 as the unit element. 0 Examples 1. For all a E C . 4.1 A a = a. however. All the lattices we shall be considering have both a unit element and a null element. A Boolean algebra is an orthocomplemented. a 5 b if and only if a A b = a if and only if a V b = b. for all a E L. andOAa=O. . Thus C is a semigroup18 with respect to the &operation with 1 as the unit element. inverse elements need not exist. In addition. However. unlike the case of a group. 3. it is not a group since no element other than 1 has an inverse. distributive lattice.3. (1. Similarly.20 CHAPTER 1. 0 V a = a. Comments 1. A simple example of a Boolean algebra is given by the following diagram in which a A b = 0 and a V b = 1: 0 "A semigroup is a set equipped with an associative combination law and a unit element.
2. The unit and null elements are 1 := 0 := x 0.27). It becomes a lattice under the join and meet operations defined in Eqs. y) by a positive real number that is greater than (resp.3. the meet and join of the family is defined to be the intersection and union respectively of the sets in the family. the set of all linear subspaces of V is a lattice with (1.3.3.3. less than) 1. (1. 5. given any metric d(z. Clearly this lattice is complete. smaller) metric can always be constructed by multiplying d(z. (1.261. 4.3.y). PARTIALLY ORDERED SETS AND LATTICES 2. y E X (this should not be confused with the preorder relation given earlier in Definition 1.12) A W. _< d(')(z. := W. there is no null or unit element since.8) (1.2). If V is a vector space. a larger (resp.10) (1.9) and A 5 B defined t o mean A c B. Two nondistributive lattices with 5 elements: 21 0 '0 3.2. w1 . i E I is any family of subsets of X. Note.y) for all z.3. More generally. The set P(X) of all subsets of a set X is a Boolean algebra with A A B := A ~ A V B := A U B B (1. n W.11) The lattice theory complement A' of A c X is defined to be the settheoretic complement A". The set of all metrics on a set X can be partially ordered by y) defining d ( l ) 5 d@)if d(l)(z. however. if Ai.1.
If V is a Hilbert space. any lattice satisfies the absorptive laws: "If V has an infinite dimension. the lattice of closed linear subs pace^'^ is complemented. 1 20More precisely.22 CHAPTER 1. This is because the basic type of yesno question that can be asked in classical physics is whether the point in state space that represents the state of the system does. (1. for all a . the subspace [Wl. for all a E C. It differs strikingly from the analogous lattice of propositions in classical mechanics in that it is not distributive. Isham 1995). Wz] itself may not be closed. < ' u . (L3) Associativity: ( a v b ) V c = a V ( b V c ) and ( a A b ) A c = a A ( b A c ) . Wz] that contains both W and W. This particular lattice has been used extensively in investigations into the axiomatic foundations of general quantum theory. b E C . which is automatically distributive.13) Correspondingly. lie in any particular subset.20 This distinction between the quantum and classical lattices has given rise to the interesting subject of 'quantum logic' (Beltrametti & Cassinelli 1981. the disjunction WlVW2 of two closed subspaces is defined to be the closure of the smallest subspace [Wl. In addition. the lattice is usually chosen to be the set of all subsets that are measurable with respect to some given measure structure. Thus the propositional lattice of classical physics is essentially the lattice of subsets of state space. >: W' := W l = { 'u E V I for all w E W. (L2) Commutativity: a V b = b V a and a A b = b A a . c E C .3.. AN INTRODUCTION TO TOPOLOGY and with Wl V Wz defined to be the smallest subspace that contains the pair of subspaces W1 and Wz. . the set of all hermitian projection operators on V also forms a complemented lattice. 0 A lattice C satisfies several very important algebraic relations: ( L l ) Idempotency: a V a = a and a A a = a . for all a . where the complement of a subspace W is defined to be its orthogonal complement with respect to the Hilbert space inner product <. or does not. b. w > = O}.
as reflected in the variety of styles to be found in the many textbooks that are available on the subject. Proof [Exercise!] QED 1. ’lA selection of particularly useful references in this context is Bourbaki (1966).3. . A structure of this type will consist of. (1. Dugundji (1996). for each x E X . An excellent problemoriented introduction to set theory and topology is Lipschutz (1965). and Kelly (1970).4 General Topology 1.4. We shall study the theory of general topological spaces in terms of the convergence of sequences and generalizations thereof.(z) in a metric space.1. Conversely.21 In a nonmetric space X. the general theory of topology may be approached in several different ways.1 An example of nonmetric convergence As mentioned earlier. b E L. CsAzdr (1978).1 A n y nonempty set L that i s equipped with binary operations A and V that satisfy the conditions (Ll)(Ld) can be given a partial ordering by defining a 5 b i f a = a A b. GENERAL TOPOLOGY 23 (L4) a A ( a V b) = a and a V ( a A b) = a for all a.14) The resulting structure is a lattice in which the meet and join operations are a A b and a V b respectively.4. it is no longer possible t o define ‘nearness’ using a real number. there is the important theorem: Theorem 1. a collection N ( x ) of subsets of X (the ‘neighbourhoods’ of x) with convergence being defined purely in terms of these subsets. Instead we attempt to trap the tails of a sequence with subsets of X that can serve as some sort of analogue of the balls B.
. see Figure 1. . ..2) Note that this does not rule out the possibility that a sequence may converge to more than one point. .E Z . € 2 . .b]. E . .(t) converges to the real number f ( t ) in the usual way.3) An appropriate family of neighbourhoods of f E F ( [ a ..b] and E is any positive number. in terms of the tails T. such that the value of g at these points is constrained to lie in the open intervals (el. . b ] . there is a set of points t l . R ) the is collection of all finite intersections of sets of the form (1. . Specifically. At this point. := {xk I k > n} of the sequence: o ((forall N E N ( x )there exists n such that Tn0c N.4. it might be useful to give a wellknown example of convergence that is not associated with any metric or pseudometric structure. .) .4.24 CHAPTER 1. . (f If a function g belongs to a finite intersection of n sets of this type. . That isz2 t “for all t E [a.b] c R. E Z ) . A sequence of functions f is said to converge pointwise to a function f if.4) where t is any real number in the closed interval [a. x) if “for all N E N ( x ) there exists n such that n > no implies x. for all E > 0. a sequence (x1. R)denote the set of all realvalued b].(t) . equivalently.b]. E N” o (1. E . in X is said to converge to x with respect to N ( x ) (denoted x. . (en. ) respectively.x2.. 3 no(€. t z . .1) or.4. E ~ ) (.4. and positive real numbers €1. ) such that n impliesIf.f ( t ) I < E. . t.” (1. functions on the closed interval [u. let F([a. the sequence of real numbers f. AN INTRODUCTION TO TOPOLOGY Specifically. for all t E [a. It is interesting to note that this is the type of restriction imposed by Feynman in his analysis of finitetime approximations to a path 22Thesymbol 3 is short for “there exists”.” > no (1.4 for an illustration of a typical neighbourhood Nt. .+ ).
of subsets of X in order to give a notion of convergence that accords with our intuition of what this should mean? 2. ) at times tl. (en. g could represent the position of a particle that is constrained to lie in the intervals (el. GENERAL TOPOLOGY 25 Figure 1.2.4: A neighbourhood in function space. When do two different families of neighbourhoods lead to the same sets of convergent sequences? For example. What properties should be possessed by the collections N ( s ) . n = 1 . . E Z ) . el). integral. in regard to the first question. . a minimal requirement .E z . In this case.4. would seem to be that the constant sequence z := z.2 The idea of a neighbourhood space Two crucial questions that must be addressed are: 1.4. . ( . . 1. t 2 . . E . . . ..t. . . . . z E X . respectively. 2 .
O of subsets of X is finer than the collection cy if.O then it is trivial that . k .2. in particular.. If a c . if n/(')(x) is finer than N ( ' ) ( x ) ( i e . arguably. there exists a subset B E .O. We could try to consider other prima facie requirements on these neighbourhoods. in which case the first important concept is the following: Definition 1. If a. i. .. If a metric d2) a space X is stronger than another metric d') on in the sense of Definition 1. N(')(x)t J ' V ( ~ ) ( ~ ) )any sequence (zl. . A N INTRODUCTION T O TOPOLOGY should always converge to x.. Note that. each N E N(x)must contain the point x: a property.26 CHAPTER 1. If cy I. N ( x )I. 2. (1.e. that is implicit in the use of the word 'neighbourhood' in the first place. In particular. for each A E a .4. This will be denoted by a t . T2.6 1.) that converges to x with rethen that spect to ~ V ( ~ ) ( x ) . The definition Eq.T . . 3. we say that the collection a is coarser than the family .O such that B c A. Equivalently. A necessary and sufficient condition for this is that each neighbourhood of xi.s o N ( ' ) ( x ) t T where T denotes the set of (x).y.x2. the collection . Two families JV(~)(Z) N(')(s) neighbourhoods of a point and of x E X are equivalent (denoted N ( ' ) ( x ) N ( ' ) ( x ) )if each is finer than the other.} of tails of the sequence is finer than the family N ( x ) .O c P(X). tails of the sequencenecessarily converges with respect to N(l) 4.O k y then Q I.. then N(')(z) h / ( 2 ) ( x )for all x E X .O and . Comments 1..thus the k relation is transitive. this condition implies that the empty set 8 is not a member of n/(x) for any x E X .O is finer than a..2) of convergence with respect to N ( x ) is equivalent to the statement that the set T := { T I .e. .. 2.O. but it turns out to be more useful to start with the second question.
As far as convergence of sequences is concerned. 5.2 produce equivalent sets of neighbourhoods. 7. the situation is not totally dissimilar to that in a gauge theory: in particular. then a c p. . if a and . the remark in Comment 6 above implies that neighbourhoods are of interest only up to equivalence. where a neighbourhood of a point z is defined to be any set that contains an 0 open ball that contains z. a is coarser than Therefore. The relation n/(’)(z) N ( l ) ( zdefined above is an equivalence E ) relation on subsets of P(X).. it is easy to see that if a t. a collection of conditions on the elements of n/(z)that select a unique representative from the set of equivalent collections of neighbourhoods. In this context. it is useful to find the analogue of a natural ‘gauge choice’. (ie.a subset U of L is said to be an upper set if a E U implies that b E U for all b E C satisfying a 5 b. Thus each equivalence class contains precisely one upper family.p. Two equivalent sets of neighbourhoods of z E X produce the same collection of sequences that converge to z..4.1. The balls of two metrics that are equivalent in the sense of Definition 1. the first observation is that the convergence properties of the family of neighbourhoods N ( z )is not affected if we append to N ( x )any subset of X that is a superset of an element of this family. However. In any lattice L. Thus our claim is that there is no loss of generality in assuming that each N ( z ) is an upper set in the lattice P ( X ) of subsets of X . p) and p is upper. the set of neighbourhoods of a point z E X in a metric space is defined to be the set of all open sets that contain z. ie. That is: 6.B if and only if a = p. are both upper B it follows that a % . Furthermore. GENERAL TOPOLOGY 27 Here.
A2. .) converges t o z with respect t o "(z) it also converges with respect to N(z). (1. (1. .) implies that z E A1 n Az n .n. implies Thus n > max(n1. (1.4. .n 2 . "(2) + z.8) z E A. there is no loss of generality in choosing N ( z ) to be algebraically closed under the formation of finite intersections of its members. and so z .. . QED It follows from the above that. for any finite collection A1.4. as far as the convergence of sequences is concerned.9) an upper family. (1. . + E N(z). Lemma Given any family of subsets N ( s ) . such that n > n1 implies n > n2 implies z E . E X. n > n.4.4. to be (i) algebraically closed under finite intersections. . if z N ( x ) z then.10) constitute our 'gauge choice' for the elements in N ( z ) .. from this perspective..28 CHAPTER 1. Thus. ..4.10) (ii) The two conditions Eqs. . . .A.. .zZ.define "(2) to be the collection of all finite intersections of sets belonging to N ( z ) .there exists nl. Conversely. n. . AN INTRODUCTION TO TOPOLOGY The second observation stems from the following lemma. . . . Then a sequence in X converges to z with respect to N ( z )if and only if it converges with respect to N'(z). n z . Al z E A2 . there is no loss in generality z in requiring each family of sets N ( z ) .91. . . Proof Clearly "(z) is finer than N ( z )(since N ( z )C N'(z)) and hence if (zl..This is captured in the following important definition (which also includes the fact that 8 N ( x )for all z E X). n A.
. A neighbourhood structure N on a set X is an assignment to each z E X of a filter N ( z )on X all of whose elements contain the point z. Every filter is a filter base. 2.. It follows from the discussion above that our final form of the idea of convergence can be cast in terms of filters on X : Definition 1. .N ) (or simply X . The pair (X. GENERAL TOPOLOGY 29 Definition 1. A sequence (z1.1.z2.8 1. B E t3 then there exists C E B such that C c A n B. (1. the filter N ( z ) is called the neighbourhood filter of the point z in X. 2.. (b) F is algebraically closed under finite intersections. A filter base B is a family of nonempty subsets of X such that if A . The definition above is about the most general notion of convergence of sequences that one could conceive. it forms the foundation for a variety of special structures in which the filters N ( z ) are restricted in some way.there exists no such that n > no implies z E N”. (c) 3 is an upper family.11) 3.7 A filter 7 on X is a family of subsets of X such that23 (4 0 # F. a ‘topology’ is one such example. 23Moreformally.4. Comments 1. if N is understood) is called a neighbourhood space (Cs6z6r 1978).) converges to z with respect to N if “for all N E N ( z ) . this says that F is a ‘proper dual ideal’ in the lattice P ( X ) . .4. As we shall see shortly.
. F . and if a c P(X) is such that a sx B. . Nothing is lost in doing so since a sequence converges with respect to a filter if and only if it converges with respect to any associated filter base (convergence with respect to a filter base is defined in the obvious way). ? Note that this is true if and only if B is finer than any filter base B ( x ) for n/(x). . The significance of this version of the idea of convergence is that it admits a very important generalisation to an arbitrary filter base B.then a is also a filter base. 5.. If B is a filter base. 4.there exists B E (ii) If A E B such that B c A . Definition 1.9 A filter base L converges to z E X if 23 is finer than N ( z ) .) 22 is a filter base [Exercise!]. This is often useful in practice.) converges t o x with zg respect to N if and only if the filter base T is finer than the filter N(x)”. We note that (i) the collection T of tails of a sequence (zl. This filter is said to be generated by B. and (ii) the condition for convergence of this sequence can be rewritten as “(zl. . Note also that collections of neighbourhoods and n/(2)that are equivalent admit the same set of convergent filters. . AN INTRODUCTION T O TOPOLOGY 3.and the filter base B is said to be a base for this filter. If B is a filter base then ?(a) { B c X I 3A E B such that A := c B } is a filter.30 CHAPTER 1 . In practice. This is contained in the following definition. it is often convenient to deal with filter bases rather than the (frequently much larger) filters that they generate. A nonempty collection B of subsets of X is a base for a specific filter 3 on X if and only if (i) B c 3. . ..
(z). > 0. the definitions of open and closed sets are as before.. as is a finite union of closed sets. And. (1. A is open (i) if and only if A" is closed. Examples 1. We shall return t o this topic in Section 1. 2. boundary point and limit point generalise in a similar way. These form what is known as a filter subbase: that is. and (iii) an arbitrary intersection of closed sets is a closed sets. as is an arbitrary union of open sets..4. 0 . a set A is open if. as in the case of metric spaces.(f) := { g I 1g(t) . and (ii) if and only if A = Int(A). GENERAL TOPOLOGY 31 The extension of the idea of convergence from sequences to filter bases is of considerable importance in the discussion of topological spaces that are not metric spaces. the set of balls B. a collection of neighF b]. (ii) a finite intersection of open sets is an open set. In particular. It is important to note that none of these concepts change if the filter N ( z )is replaced by any filter base that is equivalent to it. [Exercise!] 3. as in the case of a metric space: (i) 8 and X are open and closed. ([a. a bourhoods of a particular function f w s defined in Eq. Comments 1.4. It can be shown that.7. and only if. The definitions of exterior point. It is also the case that A is closed if and only if it contains all its limit points. for all z E A .1. On the space of functions .4) as N. an interior point of a set A C X is any point z such that there exists N E N ( z )with N c A. the set of all finite intersections of sets of this 0 type forms a filter base. In a metric space.4.f(t)l < E } . Similarly. For example. E for the filter N ( z )of all neighbourhoods of z. Many of the topological concepts introduced in the context of metric spaces possess precise analogues in a neighbourhood space. is a filter base 2. there exists N E N ( z )such that N C A. IR).
in a topological space. N )is a topological space if and only if each filter N ( x ) has a filter base consisting of open sets. . Thus. The precise mathematical definition of the new structure is as follows. N E N ( y ) .N ) in which. This is sketched in Figure 1. but it is by no means obvious that it should also hold sway in the quantum realm. The following theorem [Exercise!] is of considerable importance: Theorem 1. The crucial extra requirement that has emerged over the years is that any neighbourhood of a point II: should also be a neighbourhood of all points ‘sufficiently near’ to II: (and. AN INTRODUCTION T O TOPOLOGY 1. settheoretic based. For example. this is not true in a more general neighbourhood space.3 Topological spaces The concept of a neighbourhood space is a considerable generalisation of that of a metric space and. as we have seen. It is intriguing to ponder whether or not this is also relevant for a mathematical model that is t o describe physical spacetime. the collective experience of the mathematical community is that this concept needs to be supplemented with an additional requirement in order to yield a really useful tool. it allows meaningful.2 A neighbourhood space ( X . the neighbourhoods are essentially determined by the open sets alone. for all y E N l .10 A topological space is a neighbourhood space ( X . The problem with a general neighbourhood space is the absence of any generic relation between the filters N ( z )at different points z in X . the notion of ‘sufficiently near’ must itself be defined in terms of the neighbourhoods of z). for all z E X and for all N E N ( z ) . However.5.32 CHAPTER 1 . there exists Nl E N ( z )such that. Definition 1. ideas of ‘nearness’ and ‘convergence’. this is certainly what is assumed in classical general relativity. of course.4.
(1. every metric space is first countable. the set of all balls whose radius is a rational number). Note. 2.(z) := { y E X I d ( s .1. These spaces have many important properties. ( f ) deb]. fined in Eq.4. For example. Every metric or pseudometric space is a topological space since the balls B.5: Neighbourhoods in a topological space. GENERAL TOPOLOGY 33 Figure 1..4) are open [Exercise!] and therefore so are their .e. for all s E X . the sets N t . however.4. Examples 1. i. y) < E } are open [Exercise!] and form a basis for the neighbourhoods of 2. that a metric space with a finite number of points is rather uninteresting: any such space automatically possesses the discrete topology [Exercise!] defined as the topology in which. N ( z ) is the collection of all subsets of X that contain the point s (see below). . there exists a countable basis for the neighbourhoods of each point (for example. IR). In the neighbourhood structure associated with pointwise convergence on the space of functions 3( [ a .
X } : the only open sets are 0 and X. X } and h / ( c ) = { { a .c}. on X = { a .3 Let r be any family of subsets of a set X that satisfies the three conditions: (71) 0 and X belong to r. c}. and (iii) any finite intersections of open sets is open. but only 29 topologies. and hence this function space is a topological space. One of the central properties of a topological space is that the converse is true in the sense of the following theorem s (whose proof is left a an exercise). For example. for a given topology r on X. 24Thenumber of elements in a finite set X is denoted 11 x. Then T ( X )c N ( X ) with an injection i : T ( X )+ N ( X ) where. X } . But these form a filter base for N (f ) . so they cannot form a basis for the neighbourhoods. c} there are 64 neighbourhood structures. (72) an arbitrary union of elements of r belongs to r. Theorem 1. = n there are 2n2n different 3. . Comments 1. (73) any finite intersection of elements of r belongs to r . Let N ( X ) and r ( X ) denote respectively the set of all neighbourhood structures on X and the set of all topologies on X. b. the neighbourhood structure i ( r ) is obtained by defining the neighbourhood filter N ( z ) of any x E X to be the collection of all subsets N of X that contain a Topen set that contains x.34 CHAPTER 1 . c} that is not a topology is N ( a ) = { { a . b. N ( b ) = {{b. Then r is the family of open sets of a topology on X with a neighbourhood base B ( x ) := { 0 E r I x E 0 )f o r all x € X . (ii) an arbitrary union of open sets is open. An example of a neighbourhood structure on X = { a . b } . A N INTRODUCTION T O TOPOLOGY finite intersections. On a finite set X neighbourhood structures [Exercise!] but the number of topologies is less. 0 11 x We have seen that the collection of all open sets in a neighbourhood space satisfies the three conditions: (i) 0 and X are open.
will be denoted by ( X . for a topology.12) we have k o Z ( T ) =T 3.4. 5. it is convenient t o introduce the notion of a ‘base’. Thus a collection B of subsets of X is said to be a base for a topology every Topen set can be written as a union of members of B. . 2.( ~ constitute a complete.( ~ 3 For this reason. (iii) are algebraically closed under finite unions.4. The complements of the elements in this family then form the collection of open sets for a unique topology on X in which the original collection C is the family of closed sets. As in the case of neighbourhood structures. way 3) of defining what is meant by a ‘topology’ on a set X. This defines a map k : N ( X ) + T ( X ) . alternative. 4.and it can be shown that. The axioms ( ~ 1 ) . or ‘subbase’. N ) . T if A collection of subsets is said to be a subbase if the set of all finite intersections of elements of the collection forms a base for the . in the context of the diagram T (X ) +n/(X) *T(X). Thus a topology is also determined by its collection of closed sets.1. a topological space ). Indeed.T ) rather than ( X . N E N ( z ) if and only if there exists 0 E T such that z E 0 c N . An equivalent way of defining a topological structure on a set X is as a collection C of subsets of X that (i) include 8 and X. The open sets of an arbitrary neighbourhood structure N on X obey the conditions in the theorem above and therefore generate a topological space associated with N. (ii) are algebraically closed under arbitrary intersections. . GENERAL TOPOLOGY 35 i e . sets that contain all their limit points. (1. i e . many introductions to general topology start at this point by defining a topology on a space X to be a collection T of subsets of X that satisfies the conditions ( ~ 1 ) . for any topology T .
r2open. In practice. is an open set in the topology on Xj.j. here. A N INTRODUCTION T O TOPOLOGY topology. ~ l ) and (XZ. 6. an interesting topology on any infinite set X is the cofinite topology defined to be the one whose open sets are 8 and the complements of all finite subsets of X . One of the simplest is the collection of all upper sets. topologies are almost always defined in terms of their open sets. Another important example is the ‘product topology’ on the Cartesian product X 1 x X2 of a pair of topological spaces ( X l .j 2 . the product topology on xiGlXi is defined by specifying a subbase to be the collection (1. and 0 is open in the topology on X j .} of indices. which is clearly a topology since it is algebraically closed .4. . there exists a unique ‘smallest’ topology on X for which C is a subbase: namely that topology whose open sets are defined to be all arbitrary unions of the collection of all finite intersections of elements of C. the product xiG=Xi is defined to be the set of functions from I to UicIXi with the property that f ( i ) E X i for all i E I . i E I . The definition of the product topology admits a wideranging generalisation to the product of topological spaces X i . 7.36 CHAPTER 1.13) where j E I . Namely. Partiallyordered sets possess several natural topologies related to their ordering structure. C X j z . and where each subset Oj. It should be noted [Exercise!] that given any family C of subsets of a set X . or of a base or subbase for the open sets. . 8.i = 1 . . n. 1 Thus a basis for this topology is precisely the collection of all sets of the form 0 1 x 0 2 . where I is an arbitrary index set.. . . . 72). 2 . This topology is defined to be the union of all sets of the form 0 x 0 2 where O1 c X1 is Tlopen and 0 2 C X2 is . Thus a basis for the product topology is the collection for all finite collections {jl. For example.
1. it is the intersection of all the closed subsets of X that contain A (that this intersection is indeed a closed set follows from the property that an arbitrary intersection of closed sets is closed).{ a . x. more precisely. and 0 n A # 0 and 0 n A“ # 0. el> {ah (1..4. { c . b. Show that the collection 7 := (0. c . and only if. GENERAL TOPOLOGY 37 under arbitrary unions and intersections.15) defines a topology on the set X := { a . 9. c. 1 4 4 Some examples of topologies on a finite set . follows from the property than an arbitrary union of open sets is open). This is equivalent to saying that x is an interior point of the complement of A. Interior points. The interior of Adefined to be the set Int(A) of all interior points in Ais the ‘largest’ open subset of A. d. This is defined to be the ‘smallest’ closed subset of X that contains A. An associated idea is of the closure A of a subset A of a topological space X . it is the union of all the open subsets of A (that this union is indeed an open set. and boundary points have the following key properties for a subset A of a topological space X : 0 A point x E X is an interior point of A if.c. e } . more precisely. The idea of a topology can be illustrated with the aid of examples on a set X that has only a finite number of elements. d . exterior points. . The collection of all lower sets yields another example.4. and only if. 4. there exists an open set 0 with x E 0 and 0 n A = 0. A point x E X is an exterior point of A if. 4 . there exists an open set 0 such that x E 0 C A. 0 0 A point 2 E X is a boundary point of A if there exists an open set 0 with x E 0.{b. Problem 1.
16) is as T { a } n {c. e } = X E T { c . d. c . d } = { a } E T {a}n{b. (1.d. A N INTRODUCTION T O TOPOLOGY i) List the closed subsets of X . d .b.e} = {b.4.d. c.e} = {a. . d } E T.d}={a. Answer 1. e } satisfies the three axioms (TI)(T~): (TI): (72): 0 and X belong to T by definition. e } = { c . Find the interior and exterior points of A .15) of the set X := { a . c.d.e} = 0 E T {c. d}? iii) Let A be the subset {b. b.d}n{a. c. ii) What are the closures of the sets { a } . b. c . { a .c. d } E 7 {c. d } n { b . d } U {b. c.38 CHAPTER 1. d } n { b . The calculation of the unions of the members of T is as follows: { a } u { C . c .d} E T { a } u { b .C . d } u { a .c.d} U {b.c.e} E T { a . c.d}={c.c. c } and {b.17) to be a topology are satisfied. d } = { a . d } = 0 E T { a } n { a . d .C .e} = X E T. d } E T {"}U{a.4. d.c. d } E T { a . d . Hence show that there are subsets of X that are neither open or closed. and also subsets of X that are both open and closed.c. and hence find the boundary of A in X . ( ~ 3 ) : The calculation of the intersections of the members of follows: (1.c. d } of X . e } = { c .d} E T { c .4. We have to show that the collection of subsets in Eq. d } = { a . c.d. Thus all the conditions for T (1. c. e } = { a .c.d.
 {a} = { a ) { a . e}. d. open subset of A is {c. e } = X {b.1. d. Hence the only exterior point of A is a. b}. and the only open subset of this is { a } . d } . d } . hence the interior points of A are c and d. the only nonempty.20) iii) We have A := {b. by inspection. The open sets are and hence the closed sets are An example of a subset of X that is neither open nor closed is { a . c. ii) The closure A of a subset A c X is defined to be the intersection of all the closed subsets F of X such that A c F c X. c. ie. c } = { a . A point z E X is an interior point of A if there exists an open set 0 such that II: E 0 C A . Which of the following classes of subsets constitutes a topology on X? . (1.. Thus. b. A" = { a . Problem 2. e } . d. and hence bd(A) = { b .d l = {b. it is the 'smallest' closed subset of X that contains A. Hence the boundary points are b and e. A point z E X is a boundary point of A if it is neither an interior point nor an exterior point. b. A point II: E X is an exterior point of A if there exists an open set 0 such that z E 0 E A". However. However. c. e}.4. An example of a subset of X that is both open and closed is { a } .4. el. Let X denote the finite set { a . GENERAL TOPOLOGY 39 i) A closed set is defined to be any subset of X that is the complement of an open set. c.
c. d } } is a topology on X since. The key definitions are as follows. X . b} and { a . i) The set r1 := ( 0 . 1. The join of two ropen sets 01. ii) The set r2 := ( 0 . { a . iii) The set 7 3 := { 0 . { a . 1 2 3.{ a } . b } is not a member of the collection r2.but the intersection { a . b.21) := 0 n 0 2 . c . b. c} is not a member of the collection 7 1 . c} belong to 7 1 .4.4. d } . b. b. X . 0 2 are ropen sets. this collection of subsets of X is algebraically closed under the operations of intersection and union (and it contains 0 and X). for a given topology r on X .22) . b. the set of all 7open sets is a lattice. 2.c. d } = { a . b} U { a . d } } is not a topology on X since { a . 0 0 A 1 0 2 2 is defined as (1. { a } . { a . X . b}. c ) and { a . { a . 0 0 1 is defined as 2 v0 2 := 0 u 0 1 ( 1. b. c } = { a . { a . { a . { a .5 A topology as a lattice It is interesting to note that. b. c} n { a . b. d} belong to r2. c } } is not a topology on X since { a . b}. the partialordering operation is de1 1 fined by 0 5 O2 if and only if 0 C 0 2 . b. but the union { a . d } . The meet of two ropen sets 01. If 0 1 . A N INTRODUCTION T O TOPOLOGY Answer 2. { a .40 CHAPTER 1. by inspection. c } .11 1. Definition 1.4.
Hence {a}” = 8‘ = Int(X .8) = Int(X) = X . Thus {a}’’ is not equal to { a } . 41 ( 1.4. of open sets to be the open set i A~.4. consider the topology (8.27) . Then {a}’ = Int(X .24) 1.0). If desired. X } on the set X := {a. The lattice T inherits the distributive nature of the meet and join operations of P ( X ) .e. The pseudocomplement of the ropen set 0 is defined as 0‘ := Int(X . The unit element 1 and null element 0 are defined as 1 := X and 0 := 8 respectively.1.25) 3.23) 5 .26) For example. we may have the strict inequality O V O ’ < 1. meet and join operations are the same as those on the lattice P ( X ) of all (i. c } ) = 0. Comments (1..E I . this can be remedied by defining the meet of an arbitrary family Oi. The pseudocomplement is generally not an orthocomplement (in the sense that (A’)’ = A ) since one typically has 04 ( ‘ ’ 0 ) with 0 # (0’)’.{ a } ) = Int({b.4. The lattice T is not complete since it is not necessarily closed under the operation of taking arbitrary intersections (all that is guaranteed by the axioms of topology is that the lattice of open sets is closed under the operation of taking finite intersections of its elements). 2. { a } .4. G I (1. GENERAL TOPOLOGY 4. not just open) subsets of X . b. (1.4. c}.4. ( 1. Similarly. The partialordering.~O~ := Int nOi.
b . 4. Thus topological spaces give natural models for this type of logic. Clearly 0’ = { z E C I IzI > l}. 2. In the lattice P ( X ) of a 1 subsets of X we do have both the 2 relations A = A’ and A V A’ = X. with 7 1 5 r2 defined to mean that every rlopen set is r2open (that is.. The set r ( X ) of all topologies on a set X can be partially ordered. In these circumstances. in the topology { 8 . This connection between intuitionistic logic and topology finds its deepest expression in the subject of topos theory. As another example. in what is called intuitionistic logic. one typically has a 4 a” and a V a’ 4 1. and 7 2 is stronger. The first step is to make r ( X ) into a poset: Definition 1. { u } . On the other hand. 1 4 6 The lattice of topologies T ( X ) on a set X .12 1. and hence 0 U 0’ is equal t o C minus the unit circle { z E C I lzl = 1). AN INTRODUCTION T O TOPOLOGY For example. The analogue of the idea of a Boolean algebra in this situation is what is known as a Heyting algebra (MacLane & Moerdijk 1992). or coarser. This is one of the main reasons why lattices of this type can serve as mathematical models of classical logic. One of the very interesting properties of the set r ( X ) of all topologies on a set X is that it can be equipped with a natural lattice structure (this must not be confused with the statement in the previous section that the collection of all open sets for a given topology on X is lattice). than r2. X } on X := {u . or finer. consider 0 V 0’ when 0 is the unit disk 0 := { z E C I Iz( < l} in the complex plane C equipped with its standard metric topology. This notation is compatible with the earlier use of ‘stronger’ and ‘weaker’ in relation to metrics. r2 has ‘more’ open sets than rl).42 CHAPTER 1 . .c}. than rl. the topology r1 is said t o be weaker. we have { u } V {a}’ = { u } U 8 = { u } which is a proper subset of X = 1%4 4.
I.355.28) 2.7 and is equal to 1.4. { u } say. The join operation is 7 V T~ := 1 coarsest topology containing 72). The weakest topology (the indiscrete topology) is just (0.(X) on X can be given a lattice structure with respect to which this ordering is the lattice ordering.29. . Comments 1. The set of all topologies . It is instructive to study a few simple examples where X is a finite set. X}.4. If X lattice: = { u . if 1 is a finite integer n.13 1.(X)( = 2'Ix' = (P(P(X))().1. 2.4. {Ai n A2 I A1 E TI. X } and the strongest topology P(X). The meet operation is 71 A 72 := 7 1 n72 = {A C X I A is open in both T~ and T ~ } . b } there are four topologies arranged in the following . The lattice operations on .( X )5 2"("l).29) 3. The number of topologies that can be placed on a given set X has been calculated for the cases = 1 . it is known that 2" 5 x 1 I T . The strongest topology is P(X) ( i e . for which there is just the single topology (0. The null and unit elements are respectively the weakest topology (0. every subset of X is open) and is called the discrete topology. The simplest case is when X has one element. When X is infinite it can be shown (Frohlich 1964) ~ that the cardinality of T ( X )is two orders of infinity higher than that of X (that is.& E (1.209527and 9535241 respectively. (1.(X) are defined as follows: Definition 1.{ u } } . 11 x In general.6942.4. GENERAL TOPOLOGY 43 3.
The first really interesting example is when X is a set {a. c} of b. cardinality 3. {a. A } is an atom . c } . ab(ab)(ac) means the topology whose open b} c}. For example. 4. A N INTRODUCTION T O TOPOLOGY 3. TA = (0. b.6 using a notation that has been chosen for maximum typographical simplicity. For example. X . The lattice of all topologies on X possesses many interesting properties (Larson & Andima 1975). { b } . More precisely.44 CHAPTER 1. Figure 1. sets other than 0 and X are the subsets {a}. for each A C X .6: The lattice of topologies for X = {a. it is complete and atomic. and {a. The lattice diagram for this case is shown in Figure 1.
5.p) { := AcX I z 6A or A E U } (1.c and (bc).1. The lattice r ( X )is also antiatomic. It can be shown that a necessary and sufficient condition for a topology to be determined by the set of convergent sequences alone is that it be first countable (Bourbaki 1966.31) where U is any ultrafilter (a maximal element with respect to the natural partial ordering of filters) not equal to the principal ultrafilter of all subsets of X containing the point x € X. there exist topologies 7dwith the properties that (i) the maximal topology 1 covers Td.30) In the example above where X = { a . to z. Thus a general topological structure is determined by its collection of convergent filters. c}. ‘rA covers the trivial topology 0).. b. That is. ab(ab)(bc) . unlike the situation for metric spaces.(ab).a.. and every topology r is determined by these atoms in the sense that TA 5 T}. topological space It is important to note that if 2 is a sequence in A C X that converges . (1. then z is a limit point of A. b. . ac(ab)(ac) and ac(ac)(bc). GENERAL TOPOLOGY 45 (that is. bc(ab)(bc) . the antiatoms are the topologies bc(bc)(ac).4. the converse may not be true: a subset A of a general topological space may have a limit point to which no sequence of elements in A converges.(ac). the atoms are the six topologies b. However. 1 4 7 Some properties of convergence in a general . ab(ab)(ac). Kelly 1970). and (ii) every topology is uniquely determined by the antiatoms that lie above it. But what is true is that there will always be a filter on A that converges to the limit point. c}. the antiatoms are topologies of the form r(.4. In general.4. In the example where X = { a .
y) A most important concept in topologyand < 00. It can be shown that there is a onetoone correspondence between filters and nets.4. . tori.32)is a farreaching generalisation of the definition Eq. + X for some directed 3. 25A subspace A of a metric space is bounded if d(z. I have elected t o concentrate on filters because their definition as ‘dual ideals’ in the lattice P(X) is in line with my desire t o emphasise some of the algebraic aspects of general topology.14 1.4. (1. of ‘finite size’. (1. or any other subspaces of Euclidean space R”that are closed and bounded25.4. which means a space that is. A N INTRODUCTION TO TOPOLOGY Generalised convergence can also be discussed using what are called ‘nets’: Definition 1.11) of a convergent sequence. A net on X converges t o a point x E X with respect t o a neighbourhood structure h/ on X if “for all N E h/(x). 1. there exists a E D such that y Q implies f ( y ) E N ” . (1. The classic examples of compact spaces are spheres. and hence Eq.P E D then there exists y E D such that a 5 y and P 5 7. However. Many authors prefer this (for example.32) The set of all positive integers is a special example of a directed set.8 The idea of a compact space one that fundamentally involves generalised convergenceis that of a ‘compact’ space. in some sense.46 CHAPTER 1. 2. and therefore nothing is lost by using the latter. Kelly (1970)) because of the intuitive similarity between nets and sequences. A directed set is a partiallyordered set D with the property that if Q.4. A net on a set X is any function f : D set D.
Another famous theorem is the Tychonoff product theorem which asserts that the product topology on the product of an arbitrary family of compact spaces is itself compact. However. ~ . there exists n’ or. it can be shown that every sequence ( z l . This asserts that every closed and bounded subset of the Euclidean space IR” is compacta result that is extremely useful in practice.for all n. all A E for B. “for all N E N ( z ) . 3. A famous result in topology is the HeineBorel theorem which has been alluded to already. An alternative.15 A topological space X is compact if every filter base B on X has an accumulation point.. in terms of the tails T.4. E N” (1.35) Comments 1. (1. GENERAL TOPOLOGY 47 One characteristic feature of such a set is that any infinite subset of points must necessarily cluster together in some way. it turns out that this is too broad. defined as any point z such that any neighbourhood of z is visited infinitely many times by the sequence: “for all N E N ( z ) .33) N n T. > n such that z ! . in a closed and bounded subset of IR” necessarily has at least one accumulation point. 2.4. That is. wellknown definition involves properties of ‘coverings’ of X by families of open sets.4.34) is true.).34) One might try and reverse this result and define a general compact space to be any topological space in which Eq. (1. More precisely.. there exists z E X such that “for all N E Af(z). (1. N n A # 0”.. and the most useful definition is to strengthen Eq. (1. # 8”. of the sequence. not just the tails of sequences: Definition 1.for all T. .4.4.1.34) by including all filter bases.4. the reader is referred to one of the cited standard texts for further discussion of this idea.
48 CHAPTER 1. Hence it does preserve the lattice operations. in our case this is essentially the lattice of open sets associated with a topology.4. is welldefined even if f : X not onetoone. (1. These are defined as follows.9 Maps between topological spaces A crucial concept in most branches of mathematics is of a structurepreserving map between two sets equipped with the same type of mathematical structure.38) Note that the equality may not hold in Eq. The first relevant question in the present context.' ( B ) with generalisations to arbitrary families. therefore. (1. there are two natural mapsone from P ( X ) to P ( Y ) . This map satisfies (a) (b) + Y is (1. and one from P(Y) to P ( X ) . .38).16 1. Definition 1. The second map is from P ( Y ) to P ( X ) . 2. it should be noted.4. From a purely settheoretic perspective.37) (1.4.and is defined on A C Y as the inverse set map fl(A) := { z E X I f ( ~ E A} ) (1.39) which. is whether a map f : X + Y between a pair of sets X and Y induces any maps between P ( X ) and P ( Y ) that respect the lattice structure.4. AN INTRODUCTION T O TOPOLOGY 1.4.4. (b) f ( A n B ) c f ( A ) n f@).4.41) f'(A U B ) = f'(A) U f .' ( B ) fl(A n B ) = fl(A) n f .36) f(A) := { f ( ~E)Y I z E A } and has the properties (4 f(A u B ) = f ( A )u fW. The induced map from P ( X ) to P ( Y ) is defined on a subset AcXby (1. and hence the induced map from P ( X ) to P ( Y ) does not preserve the lattice structure.40) (1.4.
GENERAL TOPOLOGY 49 The results in Eqs. Thus a continuous function is a structurepreserving map when the subject of topology is viewed from the perspective of the lattice of open sets.41) motivate (and render consistent) the following definition of a ‘continuous’ map.4. f ( z n )converges to f(x). the converse also holds. Definition 1. tNote that when X = Y this reduces to the familiar definition: = R “for all x E R.1. .401. E > 0.4.T ) + (Y. equivalently. for any convergent sequence z + z. That is. it can be shown that a function f : ( X .T’)is continuous if and only if “for all x E X and A E h/(f(x)).17 A map f : ( X .) 3.4. for all O E T ’ . A more intuitive idea of the continuity of a function f is that a ‘small variation’ in x produces only a small variation in the value f(z).42) or. Comments 1. (1.4. for all points z E X and . In the absence of a metric.T ’ ) between topological spaces is continuous if. the concept of ‘small’ must be defined in terms of the neighbourhoods of the points z and f(x): in fact. The ‘small’ variation is often phrased in terms of sequences. The significance of this will be touched on later in the context of the ideas of frames and locales.f(y)l < 6 ” . (1.43) If(.41) that a continuous map between two topological spaces induces a homomorphism from the lattice T’ into the lattice T. there exists N E n/(x) 4 such that f ( N ) c 111’’ (1.4. f’(O) E T .4. . (1.then f is continuous.401. Thus if a function f : X + Y is continuous and if z + z then f(x.4. If X is a metric space. if f : X + Y is such that. It follows from Eqs.yI < b 2 implies . there exists b > 0 such that 1 . T ) + (Y.) + f(x) [Exercise!]. N ( f ( z ) ) f ( n / ( z ) ) . 2.
4. One of the important practical problems in the theory of topology is to find ways of constructing ‘natural’ topologies on a given set. (1. Two of the most useful techniques involve placing a topology on a set X with the aid of maps to or from X and some other topological space Y . In fact. The induced topology on X is then called the subspace topology and consists of all sets of the form X n 0 where 0 is open in the topology T on Y. X A special case is when X is a subset of Y with an injection i : Y . 5. then f(B):= { f ( A ) I A E B} is a filter base on Y . AN INTRODUCTION TO TOPOLOGY For more general spaces this is false. and f is a map from X to Y .401.y T ) := {fl(O) I 0 E T}.41) are + crucial for the success of this construction. if T is a topology on Y .4. ) 4. For example.T ) + (Y. A common example of such a situation is when some equivalence relation R is defined on a set Y . The identification topology on X is defined as p ( ~ := { A c X ) I pl(A) E T}.7’) is continuous if and only if for all 2 E X and for any filter base B on X that converges to II: it is true that f(B)converges to f(x). (1. but what is true is that a function f : ( X .4.50 CHAPTER 1 . an equivalence relation can be defined on . (Exercise: show that if B is a filter base on X and f is any map from X to Y . and X is the space Y / R of equivalence classes with p being the canonical map of an element of Y onto its equivalence class. this example is universal since if p is any surjective map from a space Y onto a set X . Another important example arises when ( Y .44) The key property of the induced topology is that it is the coarsest topology on X such that f is continuous. (1.4. T ) a topological is space and there is a surjective map p : Y + X . It is because of the existence of this special prop6rty that we refer to this topology as ‘natural’.45) The key property of this topology is that it is the finest one on X such that p is continuous. Note that the results Eqs. the induced topology on X is defined to be f .
1. GENERAL TOPOLOGY 51 Y by saying that two points y1 and yz are equivalent if p(yl) = p ( y a ) . (b) f and f’ are continuous.4.Note that the symbol f’ refers here to the map from Y to X that is the actual inverse of the map f from X to Y . this explains the origin of the name ‘identification topology’.7’) is T) (a) f is a bijection. f’(0) Topen. and (ii) for all 0 E T’. When f is invertible the two maps are related by {f’(y)) = f’({y)) for all y E Y . Generally speaking. we are looking for the appropriate meaning of an ‘isomorphism’ in the topological case.4. Note that the points in Y that are mapped into the same point in Y / R are precisely those that are equivalent to each other. Comments 1. We shall write this as ( X . Definition 1.T’). It is then easy to see that a bijection can be established between X and Y / R by mapping the point p(y) E X to the equivalence class of y in Y .39). T ) 21 (Y. In the context of topologyworking from the idea of a continuous map as a structurepreserving mapthis suggests the following definition. f(0)is T’open. an isomorphism between two structures of the same type involves a bijective map between the underlying sets with the property that both it and its inverse are structure preserving.1. It should not be confused with the inverse set map defined in Eq. Hence one says that Y / R is obtained by ‘identifying equivalent points’. A bijection f is a homeomorphism if and only if (i) for all 0 E T .4. is Thus f induces a bijective map between the collections of open sets . + (Y.10 The idea of a homeomorphism A crucial question is when two topological spaces can be regarded as being equivalent.18 a homeomorphism (an isomorphism in the context of general topology) if : A map f (X. (1. More precisely.
4. this remark is related to the analysis by Bohr and R ~ s e n f e l d ~ the need to smear quantum fields with test of ~ functions that are nonvanishing on an open set. From the viewpoint of conventional physics this is related to the idea that if X represents physical space. a segment of a line in the plane is thin whereas a closed disc is fat. By construction. If T is a topology on X and 4 E Perm(X). 260necould say that all open sets are ‘fat’ whereas closed sets come in both thin and fat varieties. they are homeomorphic) and. . Thus the set T(X)of all topologies on X decomposes under the action of Perm(X) as a disjoint union of orbits that are the homeomorphism classes of topology. Thus the induced map is an isomorphism of the lattice structures associated with the two topologies. More precisely. q!(~) is defined to be the topology whose open sets are { 4 ( 0 ) I 0 E T } . conversely.11 Separation axioms An important question in any topological space X is the extent to which points can be distinguished from each other by listing the collection of open sets to which each belongs. It thus seems plausible to argue that it is physically meaningless to distinguish between two points in X if the collections of open sets to which they belong are identical. The relevant mathematical definitions for handling this type of consideration are a s follows. T N @ ( T ) (ie.52 CHAPTER 1 . it cannot exist as a subset of a closed subset unless this has a nontrivial interior. The set Perm(X) of all bijections (‘permutations’) of X onto itself is a group. 0 1. then any real ‘object’ exists inside an open set. A N INTRODUCTION T O TOPOLOGY for the two topologies and with the property that it preserves the algebraic operations of forming unions and intersections.26 In the context of quantum field theory. 2. For example. if 7 1 and 7 2 are a pair of topologies on the same set X that are homeomorphic then there exists 4 E Perm(X) such that 72 = TI). 27The relevant papers are conveniently reprinted and translated in Wheeler & Zurek (1983)..
4. the closure A of any subset A of a topological space X is defined to be the smallest closed set containing A (it can be constructed as the intersection of all closed sets containing . As remarked earlier. (1. J % ) #NY). a partial ordering on X can be defined by ( 1. y E 0 2 and 1 O1 no2= 0. therefore. A topological space X is Toif.45). given any pair of points z. 2. to note that to a n y topological space X there is an associated To space.4. From the remarks made above. and with a . This is equivalent to saying that. if all its points are to have ‘physical meaning’ in the sense of being distinguishable by objects located in open sets.y each one is contained in an open set that excludes the other. given any pair of points z.1. 2. the resulting space is actually T2.46) 3. If X is To. The space is T2.19 1. it could arguably be asserted that any topological space that represents spacetime must be at least Toat least. It is interesting to note that if this procedure is applied to a space with a pseudometric p.if for any pair of points z. Comments 1. y E X there exist open sets O1 and 0 2 such that z E 0 .4. y E X . and can be regarded as the space that is obtained from the original space X once points that cannot be ‘physically separated’ have been identified. GENERAL TOPOLOGY 53 Definition 1. It is easy to see that N(z) = N ( y ) if and only if {z} = {y}. for all z.y E X. or Hausdorff. It is important. This is constructed by defining the equivalence relation R on X zRy if N ( z )=N(y) (1. The space is TI if. 3.A). at least one of them is contained in an open set that excludes the other. The resulting space is To.4.47) and then equipping X / R with the identification topology of Eq.
b } with the topology { O . { a . Note also that (i) every metric space is Hausdorff but (ii) the only Hausdorff topology on a finite set is the discrete topology [Exercise!].12 Frames and locales I wish now to discuss briefly certain algebraic structures that are associated with any topological space and from which the topology can be largely reconstructed. 4. Indeed. for every point x E X . 7. An example of a topology that is TI but not T is the cofinite Z topology on any infinite set X . A space is TI if and only if. filters) in a topological space has a precise answer in terms of the separation properties of the space. := [y]) p(x. o Note that there exist topological spaces that are T but not TI. i = 0 . The question of the uniqueness of the limits of sequences (or. y) where [x] denotes the equivalence class of the point IC. 5. it is easy to see that the only topology on a finite set that is TI is the discrete topology P(X). Considerations of this type lead to the ideas of a T3 topology or a T4 topology. 8. the subset { x } of X is closed. more generally. X . for example. 1 . Specifically. 1.54 CHAPTER 1. it can be shown that a necessary and sufficient condition for a topological space X to be Hausdorff is that every filter on X converges to at most one point in X . One wellknown example is that the . { a } . There exist more refined notions of separation that involve. 2 is also a Ti topology. AN INTRODUCTION T O TOPOLOGY metric topology that is induced by the distance function d([x]. An example is the set X = { a .4. A subspace of a Ti topology. the extension of the Hausdorff axiom to include the ability to distinguish between arbitrary closed sets (not merely single points) with the aid of nonintersecting open sets that contain them. b } } . 6.
{b. otherwise. defined by 2 I+ h.4.X. l}. This inspires an attempt to define a 'generalised point' associated with the lattice to be any homomorphism from the lattice into (0.4.c}} on the set X = {a. The natural map h : X + p t ( r ) . can the topological space be reconstructed (up t o homeomorphisms)? From the discussion in Section 1.c} have isomorphic lattices of open sets but r1 not homeomorphic to is 72. A topology can be constructed on the set p t ( r ) of all such homomorphisms by defining the open sets to be all subsets of the form { h E p t ( r ) 1 h ( 0 ) = 1) where 0 is any ropen subset of X. (1. . Another example is based on the observation in Section 1. A number of important statements can be made concerning this construction:28 "A comprehensive discussion of the subject is Johnstone (1986). Since the only question that can be asked of a point is whether or not it belongs to any particular open set. defined on open sets 0 by h x ( 0 ):= 1. { a } } and 7 2 := {8. The basic step is to try to reconstruct the points of X from the lattice associated with a topology r on X . l}.. 0. if 2 E 0.11 it seems unlikely that the topology can be reproduced completely if X contains points that cannot be separated by specifying the open sets to which they belong: ie. X .b.1.1} of two points. In particular.4. is a homomorphism from the lattice r onto the lattice {0.5 that the collection of subsets of X that constitutes a topology forms a sublattice of P(X).4. For example. see also Vickers (1989). if the topology is not To. x E X . given the lattice..48) It can be seen at once that each h. is clearly continuous with respect to this topology. GENERAL TOPOLOGY 55 topology of a compact Hausdorff space X is uniquely specified by the ring structure of its set C ( X ) of realvalued continuous functions. This raises the interesting question of the extent t o which this lattice structure determines the topology. the topologies T~ := { 8 . : r + (0. it is natural to consider the collection of mappings h.
e. At this point it is important to observe that there is no reason why the constructions above cannot be applied to lattices that are not a przori lattices of open sets in any topology! To see what type of lattice is appropriate for such a treatment note that (i) the lattice of open sets associated with a topology is algebraically closed under arbitrary unions (the join operation). the 29There is a technical difference between these two concepts that comes into play only when structurepreserving maps are being considered. More precisely.49) It is this collection of properties that is axiomatised to construct a purely algebraic definition of a topologylike structure. For example. y E X determine the same homomorphism if. is a c) continuous injection of X / R into pt(7).4. In particular.4. Spaces for which this map is both onetoone and onto (it is then necessarily a homeomorphism) are called soberthey are the spaces whose topology is completely captured by the lattice structure of their open sets. Thus the map h : X + pt(T) is onetoone if and only if the topology T on X is TO. See the literature for more details.47). The topology on pt(7) is To. 3. may not be surjective: i. N ( x ) = N ( y ) . for any x E X. (1. 2. and (iii) the lattice of open sets obeys the infinite distributive law A A V S V{ A A B 1 B E S } = where S is any collection of open sets. If a relation R is defined on X as in Eq. AN INTRODUCTION TO TOPOLOGY 1. (1. there may exist homomorphisms that are not of the form h. (ii) if the meet of an arbitrary family of open sets is defined to be the interior of their intersection. a frame or locale2’ is defined to be any complete lattice that satisfies the infinite distributive law. the lattice of open sets becomes a complete sublattice of P(X). all Hausdorff spaces are sober. Two points x . and only if.. it is clear that the map [x] h. Zoc cit). . the cofinite topology on an infinite set X is not.56 CHAPTER 1. and this has given rise to the interesting subject of ‘pointless’ topology (Johnstone. Many of the ideas in topology generalise to this situation. The map x t) h.
the set of all of them is then given the topology in which the open sets are all subsets of pt(t) of the form { h E pt(l) I h(a) = 1) for some a E l . T ~+ ( Y .1. In this context it is relevant t o note that: 1. 2. If locales/frames are to replace pointset topology there has to be some analogue of the important idea of a continuous map f : ( X . iff is a onetoone map.4. It is clear from these results how one might set about constructing the appropriate generalisations to frames/locales of the ideas of subspace and quotient space. the associated lattice map is injective. T ~ ) one topological space to another. . GENERAL TOPOLOGY 57 ‘points’ associated with any frame l are defined to be the homomorphisms from e into {O’l}. This algebraic generalisation of topology is rather fascinating and it is attractive to speculate that structures of this type might one day form an important ingredient in a proper understanding of the quantum theory of space and time.9 that such a map Thus induces a homomorphism from the lattice 7 2 into the lattice T ~ . homomorphisms between locales replace the idea of continuous maps. The key ) from step here is the result mentioned in Section 1. if f is surjective. the associated lattice map is surjective.4. o the converse holds if T~ is a T topology.
.
and the question arises naturally. a set has no structure other than what it contains. therefore. by itself. at the very least. M should also be a topological space whose open sets represent certain privileged regions in the spacetime. the idea has arisen that.Chapter 2 Differentiable Manifolds 2. Namely. The use of such familiar mathematical models has many important implications. the topological spaces that have actually been used historically t o represent spacetime (or space and time separately) are of a very special type. the number of which is identified as the ‘dimension’ of the space. as to what other mathematical structure should be imposed on M . However. In practice. Thus. threedimensional physical space is represented mathematically by the Euclidean space IR3. the combined notion of ‘spacetime’ is represented by the Euclidean space IR4. they have the property that it is possible t o uniquely label any particular spacetime point by specifying the values of a finite set of real numbers. in Newtonian physics. not the least of which is that dzfferentiation can 59 .1 Preliminary Remarks The minimal mathematical representation of spacetime in theoretical physics would presumably be with some set M whose elements represent ‘spacetime points’. and onedimensional time is represented by IR. But. in special relativity.
However. the metricspace topology induced by the usual metric function. thus opening up the very fruitful idea that the dynamical evolution of a physical system can be modelled by differential equations defined on the spacetime. of which the Euclidean space R4of special relativity is just a special example. Also. This remains one of the major motivations for studying differential geometry. in general relativity a spacetime is modelled by a ‘differentiable manifold’.2 2.2. However. For convenience. and many physicists would be happy with the idea that employing real numbers in this way is an idealisation that might at some future time be replaced with a different picture of the world at very small distances (for example. there is an explicit underlying topology on such spaces: namely.1 The Main Definitions Coordinate charts We must proceed now to give the formal definition of a differentiable manifold.60 CHAPTER 2. Specifically. as emphasised earlier. of course. it remains the case that one of Einstein’s major contributions to physics was his realisation this it is possible to generalise the mathematical model of spacetime whilst keeping the basic ideas of (i) being able to locate a spacetime point via the values of a set of real numbers. differential geometry enters into many other areas of modern theoretical physics. 2. not withstanding such caveats. in relation to the Planck length Lp := ( G ~ / C ~ N ’1035m that arises naturally in quantum theories ) ’~ of gravity). DIFFERENTIABLE MANIFOLDS be defined. and it has become an indispensable tool for many scientists who work in these fields. It is a matter of some philosophical and physical doubt whether such a use of real numbers is totally meaningful: the construction of the real numbers from the rationals is quite an abstract operation. it will be . and unless stated otherwise. and (ii) being able to describe the dynamical evolution of a system using differential equations.
otherwise it is locally defined. know as the components of X .2. 4) where U is an open subset of M (called the domain of the coordinate chart) and 4 : U + IR" is a homeomorphism of U onto an open subset of the Euclidean space R" equipped with its usual metric topology. Hausdorff topological space. 'A topological space is said to be connected if it cannot be written as the union of two disjoint open sets. If U = M then the coordinate chart is said to be globally defined.1: A local coordinate chart. Then the overlap function between the two coordinate charts is the map 42o $1' from the open subset $1(Ul n U 2 )c R" onto the open subset 42(U1n U 2 )c R" (see Figure 2.1 61 1. Any topological space X that is not connected can be decomposed uniquely into a union of disjoint open subsets.42) be a pair of rndimensional coordinate charts with U1 nU2 # 8. . 2.2. Let (U1.$1) and (U2. This is illustrated in Figure 2. Definition 2. An mdimensional (m < m) coordinate chart on a topological space M is a pair (U. THE MAIN DEFINITIONS assumed that M is a connected'.1.2). Figure 2.
i. n An atlas is said to be complete if it is maximalie. 'A function f : R" + R" is said to be C k if all its partial derivatives up to. j E I . The topological space M is then said to be a differentiable manifold. and including.. 3. the family (Ui. it is not For a complete atlas. . A C function is one for which all partial derivatives of all orders exist.62 CHAPTER 2.q 5 i ) i E ~is called a differential structure on M of dimension m. or an mmanifold if it is useful to indicate the dimension explicitly. (b) each overlap function q5j o q5F1. order k exist.2: The overlap function of two coordinate charts. and if the derivatives of order k are all continuous. An atlas of dimension rn on M is a family of rndimensional coordinate charts (Ui. These " definitions extend immediately to a function that is defined only on some open subset of Euclidean space. DIFFERENTIABLE MANIFOLDS Figure 2. is a C" map2 from $i(Ui n U j ) t o @j(Ui U j ) in IR". q5i)iE~ (where I is an index set) such that (a) M is covered by the family in the sense that M = UiEI Ui. contained in any other atlas.
the fact that the coordinates x p are to be thought of a s (local) functions on the manifold plays a key role in the modern development of differential geometry. .. . and the overlap funcR tions are required to be holomorphic. . Comments 1. V is isomorphic if n = dimV. . such that V is a topological vector pace)^. 2. 4"(p)) E I " with respect to the chart ( U . Many of the ideas of differential geometry can be extended t o the situation where the dimension is infinite. p ( ) := zp.1) The coordinate functions are often written as x p . . there is the concept of a complex manifold in which the Euclidean space l " is replaced with C". There are analogous definitions of C k .u. where the coordinate R functions 4 p : U + R. Similarly. . . as (2. ~ " ( p ) ) . 3This is just the standard topology on the vector space IR" to which. p = 1 . . and the coordinates of a particular point p are written as the mtuple of real numbers ( ~ ' ( p )x'(p). 2 . . For a proof of this and related theorems on topological vector spaces see Treves (1967).2. However. .m. are defined in terms of the projection functions up : R" + lR. This latter requirement introduces radical changes in the idea of a manifold and requires a separate study.but considerable care is needed. .p = 1 . 2 . 4'(p). respectively. . 4 ) . has the coordinates (~$'(p). . . are continuous (ie.2. A point p E U c M . . This arises because on any (real) finitedimensional vector space V there is a unique Hausdorff topology with respect to which the vector space operations of vector addition. . THE MAIN DEFINITIONS 63 4. and real analytic. 3. 4.2.m. . and scalar multiplication. of course. < co) and C" mani(k folds in which the overlap functions are required to be ktimes differentiable. As we shall see. each of which is potentially associated with a corresponding type of infinitedimensional manifold in which the vector space is the range space of the coordinatechart functions 4 : U + V . there exist many different types of infinitedimensional topological vector space.
5 . The dimension of M1 x M 2 is the sum of the dimen0 sions of MI and Ma. This involves the important idea of a ‘submanifold’ of a given manifold. The situation that is closest to the finitedimensional cases is when the coordinate spaces are chosen to be Banach spaces.3.2. If M I and M2 are two differentiable manifolds then the Cartesian product M1 x M2 can be given a manifold structure in a natural way [Exercise!]. It would be beyond the scope of the present book to deal properly with infinitedimensional manifolds.4) with $ ( N n U ) = # ( U ) n l Rw h e r e O < k I m . The Euclidean space Rm: The Euclidean space R” equipped with its usual. can be given a differ .2 Some examples of differentiable manifolds 1. or the space of embeddings of a 3manifold in a 4manifold.2 A subset N of a differentiable manifold M is a C”submanifold of M if every point of N lies in some chart (U. usually in the context of spaces of nonlinear functions: for example. k This is sketched in Figure 2. a classic reference on this subject is Lang (1972).64 CHAPTER 2. metric topology. DIFFERENTIABLE MANIFOLDS Infinitedimensional differential geometry arises in several places in modern theoretical physics. A more typical procedure is t o start with a few very special spaces that can trivially be seen to be manifolds (such as the Euclidean space Rnsee below) and then to construct a subspace of one such manifold in a way that is guaranteed to give the subspace a differential structure.2.2) 2. It is most unusual in practice to show that a given space is a manifold by directly examining the differentiability properties of overlap functions associated with various open coverings of the space. (2. noncompact. Definition 2. the nonlinear amodel. but at various points in the text I will indicate briefly whether or not the particular topic under discussion can be extended to the infinitedimensional case.
THE MAIN DEFINITIONS 65 Figure 2.. ntuple ( v 1 . . . .x") ( i e .y) R2 I x2 y2 = 1) of the Euclidean space IR2. More generally.urn) E R".2. where m = dim V . any finitedimensional vector space V can be regarded as a differentiable manifold by choosing any basis set of vectors for V and using this t o map V isomorphically onto IR" in the usual way'. v 2 . the elements of the ntuple). an ntuple of real numbers) Ic' are dex2 fined to be the components (d. the vector v = . . . . of W may be very nontrivial topologically even though its 'carrier' space I " is trivial in this respect (more precisely. . IR" is a contractible R topological space).. xzlviei is mapped to the . 2.2. .em} is a basis for V. However. . The circle S1: The circle S1 can be thought of as the subset {(z. ential structure with a globallydefined coordinate chart in which the coordinates of a vector (ie. . If the latter E + 41f {el..3: A submanifold n/ of a manifold M . . e2. Note that any open subset W of IR" inherits an mdimensional differential structure from IR" in which the coordinate charts have domains that are the intersection of appropriate open subsets of W with the domain (R") the globallydefined chart on IR".
66 CHAPTER 2. (2..4. + To see that the overlap functions are differentiable consider. then S1is clearly a closed and bounded subset. DIFFERENTIABLE MANIFOLDS is equipped with its usual metric topology. In fact. and henceby the HeineBore1 theoremits subspace topology is compact (see Section 1.3) 'Note that referring to the subset ((2.e.e. one example where this i s possible is the Euclidean space R".2. y ) means the point on the circle with this (constrained) value of II: and yi. for example. Of course. involves a slight abuse of language in the sense that any other subset of R2 that is homeomorphic to this particular subset would be just as good a choice from a topological perspective. there is no way of parameterising the circle globally with a single coordinate function. any mdimensional manifold M for which there is a single globallydefined coordinate chart is necessarily homeomorphic t o IR". In U1 n U3 we have y = (1 . Another explicit set of possible coordinate charts is shown in Figure 2. y) E R2 I x2 y2 = 1) as 'the' circle. not two.4. this is not defined globally on the circle in the sense that it is not a continuous function at the point where 6 = 0 and 6 = 277. it is not possible to locate a point anywhere on a typical mdimensional manifold with just a single coordinate chart. In the case of the circle S1.z2)i with 0 < y < 1 and 0 < x < 1.8). + .one possibility would be to use a pair of overlapping angular coordinates. the overlap of Ul and U3.5 In trying to give S1a differential structure it might seem natural to invoke the familiar angular coordinate 6. more generally. However. This situation is generally true: i.. the circle is a space of dimension one. where Note that although the coordinate functions have been written as functions of both 3: and y it is understood that these variables are subject to the constraint 1c2 y2 = 1 and that (z.
THE MAIN DEFINITIONS 67 Y Figure 2. The nsphere: The nsphere can be regarded as the subset S" := (2E Rn+' I 2.2. As the product of two compact topological spaces. 3.2= l}. (1 . Thus @3(z)1 (z.8) and can be equipped with the product of the differential structures on each individual S1.z ) 2i 41 0 $. Note that if R2 is regarded as a manifold. The torus T 2 : The torus is defined to be the Cartesian product S' x S1 of two circles. then the subset S1 is a compact.2. as such it R" . twodimensional manifold can be parameterised locally by specifying the values of two angles.).z').4) which is indeed infinitely differentiable for this range of values for z and y. T 2 is itself compact (see Section 1.4.2. This compact.'(z) (2. = and so = (1 .4: An atlas for the circle S'. of the Euclidean space I " . onedimensional submanifold. 4.
3 Differentiable maps A very important general idea in mathematics is that of a structurepreserving map between two sets that are equipped with the same .x: .5: Stereographic projection from an nsphere.x. of the overlap functions shows that this gives S" the structure of a differentiable manifold and that it is a submanifold of IRnfl. It can be given an explicit differential structure by means of stereographic projection from the north and south poles (the maps & and $2 respectively) as illustrated in Figure 2. 2. .xi  . Figure 2.2. DIFFERENTIABLE MANIFOLDS acquires a topology whichby the HeineBore1 theoremis compact.68 CHAPTER 2.). .5. . Detailed consideration where xn+l := +2/(1 .
3 1. In differential geometry. THE MAIN DEFINITIONS 69 type of mathematical structure. in group theory this would be a homomorphism.4) and (V.5) Figure 2. (2. the local representa' tives are C functions as defined in the standard real analysis of functions between the topological vector spaces R" and IR".$)on M and N respectively. The local representative of a function f (from a manifold M to a manifold N) with respect to the coordinate charts (U.6: A local representative of a function f : M +N . which is defined as follows: Definition 2. a structurepreserving map is a continuous map between two topological spaces. A map f : M + N is a C'function if. For example.6. 2.2. the role of a structurepreserving map is played by a 'C'function' between two manifolds. for all coverings of M and N by coordinate neighbourhoods.2. in topology. . is the map 7/lofo(61:(b(u)cR"+IRn that is illustrated in Figure 2.2.
This is based in part on the intuitive geometric idea of a tangent plane to a surface. A function that is C" is also said to be smooth. The C' nature of the local representative with respect to any other coverings by coordinate charts then follows from the differentiability requirement on the overlap functions.3 2. Two manifolds that are diffeomorphic ( i e . DIFFERENTIABLE MANIFOLDS In particular. a differentiable function is defined to be a C1 function. 2.1) . 3. In trying t o show that any given map f : M + N is C'. The set of all diffeomorphisms of a manifold M onto itself forms a group Diff(M).1 Tangent Spaces The intuitive idea One of the most important ideas in differential geometry is that of a 'tangent space' to a manifold. (2. A function f : M + N is a CTdiffeomorphism iff is a bijection with the property that both f and its inverse are C functions.70 C H A P T E R 2.'=O}. ' Comments 1. 3. Thus the tangent space at the point 2 E S" looks as if it should be defined by TZSn:={i7~IR"+' IZ*. as sketched in Figure 2.3. .3. 2. Groups of this type play an important role 0 in several branches of modern theoretical physics. it suffices (fortunately!) t o show that the local representative of f is C ' for a single covering of M and N by coordinate charts. This is in the same sense that two groups that are isomorphic can be regarded as being the 'same' group.7. there exists a diffeomorphism between them) are 'equivalent' in the sense that they can be regarded as two different concrete copies of a single abstract Platonic manifold situated somewhere in the realms above.
7: A tangent space for the sphere S" embedded in I" R" However. however. One of the main aims of this section is to present these different approaches to the definition of a tangent space and to show that they are indeed equivalent. In fact. quite different and many of the finitedimensional definitions are no longer appropriate. As we shall see. However. for finitedimensional manifolds these definitions are ultimately equivalent. some of which emphasise the geometric picture and some the algebraic. rather a s in the picture above for the case of a sphere.2. The situation with regard to infinitedimensional manifolds is. there are severalsuperficially quite differentdefinitions of a tangent space. This particular approach is slanted heavily towards the geometric view of a tangent space as being 'tangent' to the manifold. Thus the discussion is firmly grounded in the world of finitedimensional differential geometry. and it is largely a matter of taste that determines the approach t o be adopted in any particular case. the presentation will begin with the particular definition that does often generalise t o an infinitedimensional manifold. and this gives a more algebraic slant to the idea. Nevertheless. TANGENT SPACES 71 Figure 2.3. it transpires that tangent space structure is also deeply connected with the local differentiable properties of functions on the manifold. the problem .
The critical question. In Figure 2.72 CHAPTER 2. The crucial point here is that the curve itself lies in the manifold S". For example.and therefore this idea of a tangent vector as a 'tangent to a curve' can be generalised to an arbitrary manifold without it first needing to be embedded in a higherdimensional vector space. The answer is that a tangent vector is to be understood as something that is tangent to a curve in the manifold. In particular.2 A tangent vector as an equivalence class of curves .3. This leads to the idea of a tangent vector as an appropriate equivalence class of curves. at the point 2 E S". in particular. in the case above of S"which is embedded in a convenient s p a c e i t is clear that there are many curves on the manifold that are 'tangent' to the given vector v' (see Figure 2. DIFFERENTIABLE MANIFOLDS with this picture is that it is just thata picture. and neither should the definition of a tangent space. the definition of a differentiable manifold itself made no reference to such a space. Thus the general idea is to define a tangent vector in terms of a curve to whichpictoriallyit might be deemed to be tangent. and then regarding a tangent space as a specific linear subspace of this vector space. not in the surrounding Rnfl. But one of the main aims of modern differential geometry is to present ideas in a way that is intrinsic to the manifold itself and. is to understand what should replace the intuitive idea of a tangent vector as something that is tangent to a surface in the usual sense and which. its tangent vector (in the usual sense of a curve in a Euclidean space) is equal to the vector v'.7 it is clear that a curve can be drawn on the sphere with the property that. 'sticks out' into the containing space. is not dependent on a n embedding in some (largely arbitrary) higherdimensional vector space.8). and this is the definition that we shall now present. in particular. things are not quite as simple as this because. therefore. for example. it involves embedding the sphere in a particular vector space R"". 2. However.
not the set of image points in M .9. x:") around the point. It is important to remember this distinction between a function and its set of image points.. . Two curves g 1 and cr2 are tangent at a point p in M if (a) 4) az(0) = p . ie. it is independent of coordinate system. Note that if g1 and uz are tangent in one coordinate system. 2 .3. Cw) map c from some interval ( . . = (b) in some local coordinate system ( x l .. this is illustrated in Figure 2. 2. m.8: Many curves are tangent to the same vector v'...x2.. A curve on a manifold M is a smooth ( i e . TANGENT SPACES 73 Figure 2. then they are tangent in any other coordinate system that covers the point p E M . Definition 2. .2. . the two curves are 'tangent' in the usual sense as curves in IRm: f o r i = 1 .4 1.6 . . Note that the 'curve' is defined to be the map itself. Thus the definition is an intrinsic one. E ) of the real line into M ..
3.74 CHAPTER 2.10). the tangent space TpM (see Figure 2. This definition of a tangent vector is consistent with the intuitive geometrical picture whenas in the case of a sphere embedded . The equivalence class of a particular curve o will be denoted [o]. A tangent vector at p E M is an equivalence class of curves in M where the equivalence relation between two curves is that they are tangent at the point p . There is a natural projection map 7r : T M + M that associates with each tangent vector the point p in M at which it is tangent. The inverse image (the fibre over p ) of any point p under the map 7r is thus the set of all vectors that are tangent to the manifold at that pointie. 4. The tangent space TpM to M at a point p E M is the set of all tangent vectors at the point p . DIFFERENTIABLE MANIFOLDS Figure 2. This is a special case of the general idea of a fibre bundlesomething to which we shall return later.9: A curve is a function. The tangent bundle T M is defined as T M := UpEM T p M . Comments 1..
in a Euclidean spacethere is a natural way of representing tangent vectors as elements of the containing vector space.3. w = [I. ie.) The idea that a tangent vector can be regarded as a type of differential operator is a crucial ingredient in the othermore algebraicdefinitions of 0 tangent space structure.3. (2.i7)>~IRn+' xIRn+' I Z. This remark also extends to the tangent bundle T M which. A tangent vector v in T.?=land Zi7=0}.3) 2. (Exercise: Show that this definition does not depend on the particular choice of o in the equivalence class w..3. in the case of a sphere. .'.M can be used as a 'directional derivative' on functions f on M by defining: (2.2. can be represented as: T S n = ((.4) where c is any curve in the equivalence class represented by v.10: The tangent bundle T M . TANGENT SPACES 75 Figure 2. We shall return to this point later.
Proof Figure 2. and (ii) a real multiple of a tangent vector is itself a tangent vector.11: The addition of two equivalence classes of curves. S" embedded in JRn+'. This suggests strongly that it should be possible to make TpM into a vector space within the framework of the general definition of a tangent vector given above.3. We use a coordinate chart (U. Theorem 2.and then consider the image curves 4 o o1 and q5 o o2 which map an open interval into R". pictorial representation of tangent vectors to. it is clear that (i) any two tangent vectors at a point p can be added t o give a third.76 CHAPTER 2.3 The vector space structure on T. DIFFERENTIABLE MANIFOLDS 2. for example.1 The tangent space TpM carries a structure of a real vector space. o1and o2are representative curves for the two tangent vectors v1 and v2 in T p M . .4) around p E M with the property that $ ( p ) = 6 E R".11.M In the heuristic. In Figure 2.
a tangent space can be regarded as a local 'linearization' of the manifold.3.2.M is a real vector space.3. like both a1 and a2.3. 2. R passes through the null vector 6 when t = 0. the curves a and 02 cannot be 'added' directly since 1 the set M in which they take their values is not a vector space. and hence it is legitimate to consider the sum t c) 4 o a1( t ) 4 o a ~ ( twhich is a ) curve in l " and which. the local coordinate space R" i s a vector space. More details of this will be given later. $ ) and representatives 01 and ~2 of the tangent vectors v1 and v2. It follows that the map + is a curve in M that passes through the point p when t = 0. Then define (i) v1+ v2 := [+' o (4 o a1 (ii) rv := [$' o It can be shown [Exercise!] that (a) these definitions are independent of the choice of chart ( U . that maps the tangent spaces of M linearly into the tangent spaces of n/ according to the following definition. This is the concept of the 'pushforward' map h. Note that this means that T M is a vector bundle.3.7) ( ~ o 4 for all T a)] R. . and it is of considerable importance that a map h between two manifolds M and N can be 'linearized' with the aid of the tangent spaces and the vector space structure that they carry.6) (2. However. (b) under these definitions] T.4 The pushforward of an equivalence class of curves To some extent. TANGENT SPACES 77 Of course. + 4 o a2)] E (2.
Show that the map h.3.M and T E IR. T N !% T P .12) is h. Show that this definition is independent of the particular choice of the curve a in the equivalence class w E TpM.78 Definition 2.(u).11) TM M h.w2 E T.+ N and v E TpM then the ‘pushforward’ h.10) and k : N 3. TP (2.. DIFFERENTIABLE MANIFOLDS If h : M .12: The pushforward operation on tangent vectors. = k. for all v1. If M . is linear. A N TM and 1 1 A P I MIcoh’P I (rcoh. and if h : M P show that ( k o h).h. which relates to the diagrams + N + (2. N and P are manifolds.12) 1 .3. L(vi + uz) h.3..5 CHAPTER 2. (2.9) (2.3. (2.8) Figure 2. 2.3. That is.(rw) = = h*(vi)+ h*(vz) rh.(v) in Th(p)N defined by (see Figure 2. Exercises 1.(v) := [ho a] where w = [a].
. (2. and which is based on the idea of a ‘directional derivative’. the definition of a tangent vector given above can be related to the familiar ‘object with a superscript index’ ( p much favoured in such approaches. Recall that the directional derivative of a function f along a tangent vector v is defined as (Eq. A good introduction to a categorybased view of differential geometry is given in Lang (1972).3.11) says that if g : M + P is a map from a manifold M to a manifold P. I will not introduce such ideas formally in this book.5 Tangent vectors as derivations So far.3. The ensuing interplay between the geometric and the algebraic ways of understanding tangent vectors is one of the central features of modern differential geometry. a reader who has already encountered the coordinatebased approach t o differential geometry (for example. then the pushforward g*(v) E Tg(p)P v is independent of factorising of g through some intermediate manifold N as M %N lc\ P with g = k o h. and if v is a tangent vector at p E M . Eq. rather abstract. rather than geometric. our description of a tangent vector has been (hopefully!) geometrically appealing but. or how.3. In particular. A powerful way of uncovering this relation is t o discuss first an alternative rigorous approach t o the idea of a tangent vector that is essentially algebraic in form.13) which enables the equivalence class of curves v E TpM t o act as a type of differential operator on the space C m ( M )of realvalued 6The natural language in which to describe this type of property is ‘category theory’. although they do lie behind many of the constructions made in differential geometry. TANGENT SPACES 79 In effect.2. This ‘functorial’6 property is necessary for the idea of the pushforward operation t o be of much use: otherwise it would be as if sending a letter from London t o New York by air produces a different outcome if the plane stops t o refuel somewhere along the way! 2.3.3. (2.4)) v(f) := dt where [o] v = (2. admittedly. and a functor is a structurepreserving operation between categories. by taking an introductory course on general relativity) may be wondering if.
3.M as compared to the construction in Theorem 2.3.3.3. Comments 1.3. The first 'algebraic' way of defining tangent vectors is to turn Eq.15) merely assert that w is a linear map from the vector space C"(M) to R.142. general rule: geometric ideas appear most naturally in the geometric approach to tangent vectors. (2.80 CHAPTER 2. DIFFERENTIABLE MANIFOLDS differentiable functions on M . More precisely: Definition 2.The term 'derivation' is used because of the crucial 'Leibniz rule' in Eq.1 of the vectorspace structure on T p M .g E C"(M) v ( r f ) = ~ w ( f ) all f E C " ( M ) and T E R.13) around and define T.16) (d 2. The space of derivations DpM can readily be given the structure of a real vector space by defining (211 + vz)(f) ( T W ) := := + wz(f) 4f) Vl(f) T (2. (2.14) (2. This illustrates an obvious. f = f(p)v(g)+ g(p)v(f) for all f' 9 E C"(M).3.w in D.w2. f E C " ( M ) and E R.6 1. and algebraic ideas appear most naturally in the algebraic approach. A derivation at a point p E M is a map w : C"(M) that + IR such (2. for + + (4 . Note how much more natural (and easier!) is this definition of linear structure on D.3. 2.18) for all w1.(M). but important.M as a space of directional derivatives.3.3. (2. Equations (2. . and then regarding a tangent vector as a derivation map from this ring into R. This idea is developed by thinking of C"(M) as a ring over R.16) which is clearly analogous to the rule in elementary calculus for taking the derivative at a fixed point of the product of two functions. The set of all derivations at p E M is denoted D.15) (i) v(f g) = v(f) w(g) for all f.M. 3.17) (2.
13). (2.13) defines a map from TpM t o DpM that is clearly linear.3. a set of derivations at p is defined as7 .3.2 that TpM and DpM are.3.3.. Unfortunately. p = 1 . . and we shall show in Theorem 2. Of course.19) as meaning the 'partialderivative' of f evaluated at the point p E M . . since the algebraic scheme is problematic in infinite dimensions. Specifically. partialderivatives can be defined in the usual way for the local representatives f o 4l : 4 ( U ) c R" + R o f f . 2 . . . in fact. d i m M . notwithstanding the alluring nature of the notation. 4) containing the point p of interest (see Figure 2. which generally does not possess the vectorspace structure needed for a partialderivative to be defined.3. It follows that we can adopt whichever approach we wishgeometric or algebraicaccording t o the type of problem under discussion.2. 0 A very important example of a derivation is provided with the aid of any coordinate chart (U. TANGENT SPACES 81 Equation (2. (2.whichlike all definitionsmeans no more and no less than it says. The function f is defined on the manifold M . (2. and this is what is exploited in the definition Eq. There are a number of explicit examples of this procedure throughout the book. this equality of approach to tangentspace structure only holds in finitedimensional differential geometry. isomorphic.13: A local coordinate chart used t o define a derivation.19) Figure 2.19). 70ne should firmly resist any temptation to interpret the left hand side of Eq.
(2.82 CHAPTER 2. where. 0 E IR".20) (2. .x m ) and such that x p ( p ) = 0 for all p = 1 .4) be a coordinate chart around p E M with associated x coordinate functions ( d . . .21) for all q in some open neighbourhood of p ..22) function + R denotes the constant Let F := f o 4l defined on some open ball B around for a' E B.M . Proof V=l c m zV f u . (2. . '1 have taken this lemma from the elegant little book by Hicks (1965).3.21) means that the function f can be written locally as f = kf@) + k. ' Then. : M whose value is everywhere c. for any c E R. .3.3. . 2 . Lemma Let (U. m = d i m M . Then for any f E C w ( M )there exists fp E Cm(M) such that (2. 2 .3. . DIFFERENTIABLE MANIFOLDS We shall now show that these quantities form a basis set for the vector space D. . Note that Eq.
M.~ ~ .22) certainly exists. that Icl = kl x k l ) . To this end we apply the derivation at p to both sides of Eq. aUp ~ 2 .27) However. . . . . 7 zrn(q)) = F(Q = 2 f b ) + cz P ( d p=l + p=l m z p ( dF p ( x l ( q ) . . The next step is to find an explicit form for f p ( p ) . . (2. which is defined initially on some open ' ( f 0 4l)(4(4)) = F(z1(q)7 * .3. .26) and hence the desired representation Eq. TANGENT SPACES = F(O.3.0) 83 + where Fp(ti) := 1 1 dF dt . ..0 . Then f(4) = c R".3. . . we get'. then.)) = 2CV(kl) = 2'u(kc) (2.2. . (2.. (2. using the Leibniz property of the derivation u . .O. if 'u E D. more precisely.3.22) to give (&) P (2.3. for any c E R.28) gThis is one of the nicest uses I know of the truebut otherwise unremarkablefact that 1 = 1 x 1 (or. . 0 ) 1 .( ~ 1 . := F o 4.t a p .xrn(q)) fp(4) (2.3. .3. v(lc.25) is a C"function on B Now let f p neighbourhood of p E M but which can then be extended arbitrarily to the rest of M . . .) = w(ck1) = = = cv(k1) = CV(k1 x k1) C(kl(P)'u(kl) + kl(P)'u(kl)) c ( l x V(k1) + 1 x v(k.
. approaches. where v” := u(zp). . [Exercise!].urn}.3.30) is a key result in relating modern. .rn then define new coordinate functions y” := x” . 2 .84 CHAPTER 2. DIFFERENTIABLE MANIFOLDS and hence v(kc) = 0. = 1 . QED Comments 1. v”) first arises in the modern approach t o differential geometry: indeed. Eq. . This is where ‘an object with a superscript index’ (ie.3. .kzr(p) and apply the Lemma using this coordinate system. We see here an example of the importance of the idea that the ‘coordinates’ d’. (2.. since (&)p z’ = 6. (2. .3.27) (and re= membering that d ’ ( ~ ) 0) gives which. The set of real numbers {v1.3.v2. Applying this result t o Eq. are called the components of the derivation v E DpM with respect t o the given coordinate system. QED Corollary If v E D p ( M )then v = C v(xp) p=l m (2. m are p . 2 . .30) Proof If d ’ ( p ) # 0 for p = 1 . Then gives the equation between functions f( 1 = b(P)( )+ ”=l cY”() m fp( 1 Applying the derivation t o both sides of this equationand remembering that v vanishes on the constant functions q I+ f ( ~ and ) q H xp(p)gives the result in Eq. . proves the lemma. . coordinatebased. (2.. ‘coordinatefree’ differential geometry t o the more traditional. .30). .
If z1 = cr!l (ax. we have 0 = w(x”) = up. for some 211. . . ular. .2 The linear map (2.Then consider Eq. In partic2. .. . .m.32) for f ranging over the set of local coordinate functions ( d .2. so we must show that (i) it is injective. From the equality of ~ ( q ) and ~ ( z 1 ~ ) follows that it L (i) (2.2.m = dim M are linearly independent and span the vector space DpM.we have ~ ( q ) ~ ( ~ 1 2with w1 = [a11 and = ) w2 = [a2].32) is meaningful even if the function f is defined only on some open neighbourhood of the point o ( 0 ) E M . is an isomorphism.M = dim M . Now suppose that. dim D. . . Proof We know already that the map L is linear. 1.. Note that the components of a given derivation will be different in different coordinate systems. w2 E T p M .3. 2 . and (ii) it is surjective. . is injective: To show that L is injective we note first that Eq.= 0 for some set of real numbers { d . . TANGENT SPACES 85 actually (local) functions on the manifold M : the components up of the derivation ‘u are obtained by letting w act on these functions. .3. up 2. .u 2 .13)) L : TpM + DpM defined by (see Eq. . We shall return to this later.3. . 2 . for all u = 1 .3.P}. (2.3. .33) . p = 1 .P }then. (2.. The result above can be used t o prove the critical fact that DpM and TpM are isomorphic as vector spaces: Theorem 2. This means that the set of vectors (&).
The isomorphism of TpM and DpM means that either can be used according to the situation under consideration. it follows in particular that dim TpM = dim M .](f) v(f). any tangent vector v E T. 2. Hence tive. . L : TpM + DpM is surjecQED Comments 1. . and hence wl = [GI] = [o2]= v2. have we = which means that [o. if f E Cm(M). +M such that: (i) %(O) = P. Thus the curves 01 and 0 2 are tangent at the point p = al(0)= a2(0). thus L is injective. .M and DpM are isomorphic. Since T. Then. (ii) (E. Any formula proved for tangent vectors defined as equivalence classes of curves in TpM must have an analogue within the definition as a derivation in DpM.3.m. .86 CHAPTER 2.3. 2 . DIFFERENTIABLE MANIFOLDS for all p = 1 .34) (2.35) This is a trivial task using a local coordinate system on M . L E) is surjective: Let v E DpM and construct a curve a.M . In particular. : (2. and vice versa.
38) p=l with d .. we can write u= up ( & ) p where up = v(xp). and then. (2.. since h.39) t=O 3. ..v(x P ) .38). the pushforward map h. p=l (L) . if h : M +n/.3. : TpM N_ DpM + Th(p$ N_ induced by a smooth map h : M + N . Specifically. local coordinate systems around the points p E M and h(p) E N respectively. within the algebraic language of derivations.xm} and {y1..2. p = 1 . the map h. we have D m h. in this form. .3.3. (2. is manifestly lineara good example of the general maxim that algebraic properties of tangent spaces are best seen in the algebraic approach. dxp P (2.42) hb) .30) to give m (2. Similarly.8)acquires a new form. . = (2. According to Eq.3. 2 . if v E T p M .yn} be . m. A useful example of an application of these results is to find a concrete expression for the local representative of the pushforward map h.40) h) for all f E C"(N). suppose the manifolds M and n/ have dimension m and n respectively.3. .)n/ is defined on any v E DpM as (2. .v)(f) := v(f 0 4.v = x v p h .x2. and let {x1.3. (&) (A)pg = (h* ( ) (&) &j (2.3.3.y2. the local coordinate system around h(p) E Jz/ can E Thcp)n/ as be used to expand the vector h. (2. : DpM +E q p ) N is linear.defined for equivalence classes of curves by Eq.41) On the other hand. .. TANGENT SPACES 87 can be expanded in components by thinking of it as a derivation and then using Eq. Note that. Specifically.P x O O ( t ) l  dt where [o] v. (h. .then h.3. : DpM + Dh(.
in this particular context. a curve in M . .41). rather heuristically.v E Dh(p)n/: m n (2. . according to the expansion result Eq.3. Note that the expression in Eq.3. the precise expression (&)p h” is often written.40).19) to the standard partial derivative of the local representatives of these functions with respect to the coordinates xp. (2. DIFFERENTIABLE MANIFOLDS However.n are realvalued functions (the components of the map h) defined on the coordinate neighbourhood of the point p E M .3.we have (h* (&)J = (h* (&).88 CHAPTER 2.3. (2. as ah”(p)/azp.3.3.3. ie.43) follows from Eq.3. inserting Eq. Then. (2.3.45) h” = 1 . for all f E C”(M).44) where Eq.43) (2. .3.46) u = l u=l Note that.3.44) into Eq.47) 5. E ) + M . Finally..40) is to a map c : ( . (2. Another example of the use of Eq. Eq. 2 .E .3.44)which is called the Jacobian matrix of the function h with respect to the given local coordinates on M and Nis related via Eq.46) becomes (2. we get the following expression for the local representation of h.) (y”) = (y”o axp (L) h ) (2.3. (2. .3. (2.30) applied to the tangent space Dh(p)N. (2. terms of which in Eq. (2. (2. and where h” := 3” o h (2.
52) t=O .. Note that in writing Eq. Specifically.. (2.3. Put more abstractly. and (ii) the m real numbers { ~ ( d ) . TANGENT SPACES 89 (2. It can now be seen that the tangent bundle T M has a natural structure of a 2mdimensional differentiable manifold.40) implies so that (2. . .3. the Euclidean space IR" has been identified with any of its tangent spaces TzRm by associating with w' E R" the derivation ~ ( w ' in DzR" defined by ) dt ie.50) for all f E Cm(R). 6.3.Y(z")} that are the components of the derivation Y Y(z'). ~ " ( p ) } . 4) is a local coordinate chart around p E M then define the bijection where TMlu denotes the 'restriction' of T M to the subset U c M . . the 2m coordinates of a vector v E TpM are defined to be (i) the m real numbers { ~ ' ( p )~. ' ( p ) . .2.51). w' E I " generates the curve t R 5 + tw' in IR". denotes the derivation on Cm(R)defined by (2. with respect to this coordinate system. that fix the point p in a local coordinate system on M .49) where (&). .  (2. if (U.3. . This identifica0 tion is discussed in more detail in the following subsection.3.3.
let us think of T.3.o = Tf(W d dt + tw)lt. We must now show that and onto.(e( ?J) t q w . dt + .o (2.90 CHAPTER 2. Then. suppose that w1.6 The tangent space T. DIFFERENTIABLE MANIFOLDS 2. and hence V are such . dt x : V + D. = X(TW)f d f(v dt + trw)lt..54) x(w1)f+ x(w2)f.f ( v dt + tw)lt=o (2.w2)=0. = = f(v d + twl)lt=o + f(v + twz)lt=o d dt (2.3.V by d x ( w ) f := .3. .wq)) It=O = !(w.53) for all f E C"(V).V at ?J E V of the ring of smooth functions C"(V).  (2. 2 E w + In particular. dt Also.V is linear. we note that x is linear. onetoone. we have d (2.3.57) . x ( w l . To show that x is onetoone. w2). for all f E C"(V). d 0 = . that x(w1) = ~ ( w q )ie.55) = TX(W)f. for all T E R.3.3. To show linearity. is isomorphic to V itself in some natural way.V of a vector space V It is intuitively obvious that the tangent space at any point in a vector space V considered as a manifold. this is true for the function on V given by any linear : and then map ! V + R. Then we can define a map x : V +D.w2))lt=o = 0.V as the set of derivations D.56) f(v t ( w . To show this explicitly.
(2. Then we invoke the isomorphism of the space of equivalence classes of curves with the space of derivations at the point ‘u E V .3. to show that x is onto. &+ Answer If we use the natural global coordinates {z1. Problem The map h : IR2 + R2is defined by (2. Thus Eq. 2 ) .the components of the vector field v := 4x1& 32z a22 are v1 = 4s’ and v 2 = 3x2. b) := (az . it is advisable to use a different letter of the alphabet to denote the global coordinates on the target manifold R2 from that used (x) on the domain manifold R2.To this end. + 2. Finally.3.2b.3. 4a3b2).3. then w = 0.(4x1 3x2&)(l. Compute the pushforward of the derivation h. the linear maps on a finitevector space ‘separate’ the points in the space in the sense that if t ( w ) = 0 for all such linear maps l .7 A simple example of the pushforward operation It may be helpful to illustrate the ideas of the pushforward operation with the aid of a simple example of a map between two vector spaces considered as differentiable manifolds.x2}on the domain manifold R2of the map h : IR2 + IR2. are the natural coordinates x2} on R ~ . we denote the natural global coordinates on the target manifold R2by . To avoid confusion. and hence x is onetoone. TANGENT SPACES 91 However.57) implies that w 1 = WZ. it suffices to note that any equivalence class of curves through ‘u clearly contains one of the form t t+ v tw for some w E V.where {d.2.58) h(a. which have the a + values 4 and 6 respectively at the specific point ( 1 .2) at the point (1’2) E R2.
components of h.47). 2>+~~((ah’/a~:”)(l.= 2 b) (ah’/lad)(a. q ) in a product M x N can be ‘decomposed’ into a component that lies ‘along’ M .3. b) (ah’/az’)(a.16). this expression is what is really meant by the notation used in the definition Eq. ~ ) ’ ~‘(dh’/&’)(l. In particular. according t o ( h . 0 ’ J (2.3.59) Indeed. and then the components of the map h are hl(a. it is intuitively clear that a tangent vector at a point ( p . If one takes the torus S’x S1 as an example.63) 2.3.2)+v2(ah2/az2)(1. Thus.3. )~ 2) = (2.2b ’ h’(a. ~ = v1(ah2/az’)(l. The Jacobian matrix of the transformation is (ahl/az’)(a.b) = U . b) = 12a2b2.8 The tangent space of a product manifold We come now to some technical results concerning the tangent space structure of the product M x N of two manifolds M and N . = 4 ~ 4 8 + 6 ~ 1 6288. DIFFERENTIABLE MANIFOLDS {y’. These will be of importance in our discussions later of Lie groups and fibre bundles.3. y’}.v are the (2.92 CHAPTER 2. (2. b) := ? h(a.b) := 9’ o h(a. . the point with zcoordinates (1.b) = 4a3b2. 4 ~ 2 + 6 ~ ( . b ) in the domain manifold Eq.2)is mapped by h t o the point with ycoordinates (3. = 2a. (2. = 8a3b b) at any point ( a .3.62) so that (2.2 ) 4 = 2) = = (2.3.61) and ( h . and another that lies along N .3.60) R2. The formal expression of this intuition is contained in the following theorem.58)of the function h. (ah2/laz’>(a.
68) and so [a11 = prl*[a].3. operations are defined as (i) (v1.w) with v E V and w E W . and is hence the desired isomorphism.)Mx N + TpM CB TqN by This is clearly onetoone and surjective.)M xN..3.65) (2.w1 w ~ ) and (ii) T(V.3. z ( t ) ) namely. b1 = a1* 1 ).wg) := (v1 v2. QED "The direct sum V @ W of two real vector spaces V and W is defined to be the set of all pairs (v.2.3 O n the product manifold M x isomorphism1o N there i s a natural (2. TW)for T E B.w) := ( T V .3..as claimed (where Eq. Define a linear map x : T@.3. and where the vector space . al(t) := prloa(t) and a2(t) := pr2 O 4 t ) .( = Prl*[al 0 (2. Then there exist unique curves 01 and a in M and N respectively 2 a : such that a ( t )= (al(t).66) are the usual projection maps. w1) (vg. Proof Let a : ( . TANGENT SPACES 93 Theorem 2..49) has been used).E .( = (Prl 0 O a)*. + + + . E ) + M x N be a curve such that [a]= 21 E T(. Furthermore. ( o) = Prl*g* ). (2.3.64) where (2.
o i.73) iq*Q b+ +jp*P 2.Y) fojp:N Y + P.. o i. : M + P z I + (2. o j p = P o .N (Q. = q( ) Pr.69) (2.3. Let P be another manifold and let f : M x N + P. 0 (2. idM) is the identity map of M (resp.:= (f P) Q*Q + (f 0 j p ) * P f o i .)M x N. : M + M x N .3.b!.)P . N) into itself.3.67) is the map + qP.. Thus the inverse of the isomorphism TpM @ T.77) f(wd f(P. pr..3.3. Y which satisfy the relations pr. (2. DIFFERENTIABLE MANIFOLDS 1. It is convenient to define the injections (for each p E M and q EN): i .76) (2. = idM .70) I + (w) (P.)M x N + Tf@. J: (2.3.f* : qP..)M x N where and 4 TpM @ TqN i Tf(.72) where idM (resp. o j p = idhl (2. I + .Y> j p : N + M x N .3..94 Comments CHAPTER 2. p E M ) .3. Then .. p ( )) is the constant map on N (resp.3.P ) x in Eq. and q( ) (resp. (2.71) (2.)P factorises as T(..75) .74) (2. pr. M ) that takes every point into q E N (resp.3.
The diagonal map associated with a manifold M is the map + M x M defined by A(p) := (pip).2.a ) = (2.p)M M x we have TpM 6 TpM 3 (2. (2. a f (f 0 A)* = (f 0 i p > * + (f 0 j p > * .3. TANGENT SPACES 95 A :M 3. Thus. on T p M . In the chain of maps TpM %T@.3.3. 0 A*(a) ( a .80) .(a) f*(&(a)) = = 0 x 0 &(a) = f ( a l ) for all Q E TpM.79) In particular] if f : M x M + P then f o A : M + P with f o A(p) = f ( p l p ) and (f 0 n).78) x for all a E T p M .3.
.
1 1. to each point in some open subset of M ) . for all f E Cm(M). A critical requirement here is that the tangent vectors should vary ‘smoothly’ as we move around the manifolda concept that is formalised in the following definition. more generally. 97 . E T. Definition 3.1 3. functhe tion Xf : M + R defined by is infinitely differentiable.M at each point p E M .1.Chapter 3 Vector Fields and nForms 3. A vector field X on a C”manifold M is a smooth assignment of a tangent vector X. where ‘smooth’ is defined t o mean that. We come now t o an important generalisation of this idea that involves a ‘field’ of tangent vectors in the sense that a tangent vector is assigned t o each point of M (or.1 Vector Fields The main definition So far we have dealt only with tangent vectors defined at a single point in the manifold.
the brackets are usually left i out for linear maps unless it is necessary to add them for clarity. 3.1. It is straightforward to show that this definition of a smooth assignment agrees with the one above. or sometimes Lxf.2) TX noX=id~. Comments 1.1. The map f I+ Xf has the following properties with respect to the ring structure on C"(M): (i) X(f + g) = Xf + X g for all f . (3.1. . and is then usually denoted L x f .5) (ii) X ( r f ) = rXf for all f E C w ( M ) . (3.1) gives rise to a map X : C w ( M )+ C " ( M ) in which' the image Xf off E C " ( M ) is defined by Eq. g E Cm(M) + gXf for all f. linear operators)it is conventional to write the image X ( f ) o f f E Cw(M) simply as X f .1). e . A vector field X on M gives rise in an obvious way t o a vector field. The function Xf is often known as the Lie derivative of the function f along the vector field X. M In this case. 4.1.3.1.4) (3. (2.T E l R (iii)X(fg) = f X g 'The map X : Cw(M) + Cw(M) is linear andas with all linear maps from a vector space to itself ( i e . . denoted Xu. 'smooth' means that X is a C"map from the manifold M to the 2mdimensional manifold T M (using the differential structure defined in relation to Eq. (3.1.g E C"(M) (3.51)).98 CHAPTER 3. A vector field on M can also be defined as a smooth 'crosssection' of the tangent bundle T M : TM 1 (3. A vector field defined via Eq.1. .1) is required to be true for all f E C"(V) and points p in the subset U C M . on any open subset U of M . A vector field on an open subset U of M is defined in the same way except that the condition in Eq. 2. The vector field Xu is said to be the restriction of X to U .3) (3. VECTOR FIELDS AND NFORMS 2.
This procedure can be reversed in the sense that it is possible to define a vector field X as a derivation of the ring of functions Cm(M). (3.1.3.7) + + + for all f E C"(M).3. (3. to each point p in M such a derivation assigns a linear map X.16) for a derivation in DpM.h c Cm(M) and if X and Y are vector fields on M.142. and we shall employ it where appropriate.5).6) for each f in Cm(M).1. satisfies the three conditions Eqs. and (ii) Eq. but it is conventional).1.1.5) shows that X is a derivation of the ring Cw(M)in the usual sense of the word (in this respect. (2. Note that (i) Eqs. Y are vector fields on M and a .1.33.3.142. rather than the converseis useful in many situations. VECTOR FIELDS 99 which should be contrasted with the equations Eqs. The set of all vector fields on a manifold M is denoted VFld(M) and carries a natural structure of a real vector space.1. Indeed. to that satisfies the three equations Eqs. if g . a map from Cm(M) Cm(M) i. IR defined by (3. It is important to note that this definition can be extended to give VFld(M) the structure of a module over the ring Cm(M).1. : C"(M) j . f + for all p E M and f E C m ( M ) .e. (3. 5. This alternative approachin which we go from vector field t o derivation at a point p E M.1.8) .33.1.16) for a derivation at a point. But then the map p t) X.1). (2. (2. (3. and it is clear that this assignment is smooth in the sense of Eq. (3.4) show that X is a linear map from the vector space Cm(M) into itself. Y if X.3.1..f := 9(P)Xpf W Y . More precisely. a new vector field gX hY can be defined by + (9X+ W. the use of the term 'derivation' in relation to Eq.3. and it is easy to see that X.16) is potentially confusing.1. b E IR then aX b is defined as (ax bY)f := a X f bYf (3.3.Specifically. assigns a field of tangent vectors in the sense of Definition 3. 6.
The functions {Xxp. . .11) p=l where there is an implicit understanding that this equation is mean. 2 .1. Specifically.1. (2.100 CHAPTER 3. which we write as xu = p=1 c xuxp a m 8 .4). (3. ingful only on the open set on which the coordinates {” The expression Eq.3. It should be noted that the subscript U is often dropped. VECTOR FIELDS AND NFORMS 7. (3.10) in which the module structure on VFld(U) is used to expand the vector field Xu in terms of the local coordinate vector fields p = 1 . .4’) be another coordinate chart with the property that U U’ # 0 so that the components XIC’” Xxp can be meaningfully compared on the open and n . 2 . They are usually denoted Xp. (3.m } on U are known as the /I components of the vector field X with respect to the coordinate system associated with the chart (U. . the actual form of these component functions depends on the coordinate chart in use. . Of course.1. let (U’. m where the coefficient functions X x p are defined on the open subset U .1.1.4) be a coordinate chart on the manifold M .9)) shows the precise sense in which a vector field can be thought of as a firstorder differential operator on the functions on a manifold. . a (3.} are defined. 8. Let (U. . Then at each point p in the open set U we can use the expansion Eq. Thus.10) becomes simply m &. so that Eq.30) for the derivation X p associated with a vector field X defined on M .10) (interpreted as Eq. = 1 . p  (3. for all p E U .
and all attempts to construct a nowherevanishing vector field necessarily fail (this is known colloquially as the ‘hairy billiardbail problem’).11) can be applied againthis time to the vector field d/axp expanded with respect to the coordinates x’”to get (3.1. topological properties of the manifold.12) and equating terms gives (3.14) which is a familiar expression in elementary.a ax‘” (3. X. In this context it is important to appreciate that the precise meaning of the expression ax’”/axp in Eq. on the open subset U U’ C M . For this reason. A famous example is the theorem showing that a necessary condition for the existence of a nowherevanishing vector field X (ie. (3.1.13) The components of X in the two coordinate systems are Xp := Xxp and X”’ := Xx’” respectively. coordinatebased approaches to differential geometry.3.1.11) plays a key role in understanding the precise connections between older approaches t o differential geometry and the more modern.1.13) into Eq. VECTOR FIELDS 101 set U U’.1.1.1. Then using Eq.11) with respect t o both systems of coordinates gives. coordinatefree formalism that is being developed here.12) and Eq.1.1. (3.1. (3. (3. . m p=l n a dxp n m X2’”. Eq. see Bott & Tu (1982)). (3. (3. substituting Eq. # 0 for all p E M ) is that the Euler number of the manifold M is zero (for example.14) is the function on the open set U U’ defined n 9. One of the important properties of vector fields is their ability to encode certain global. This is nicely illustrated by the 2sphere S2:this has Euler number 2.. Then.
1.17) from Eq. the composite map X o Y : C"(M) + C"(M) can be defined as X o Y(f) := X ( Y f ) and this is certainly linear. it satisfies Eqs. When Y O X is subtracted from X o Y these secondorder terms cancel and only . Indeed.2 The vector field commutator We come now to the rather important question of whether or not two vector fields X and Y can be 'multiplied' together to get a third field. the 2torus S' x S' has Euler number 0.1.102 CHAPTER 3.1.1. we also have Y 0 X(fg) = YgXf + gY(Xf) + Y fXg + fY(Xg) (3.1. However. it is n o t a vector field since it fails to satisfy the crucial derivation property in Eq. 0 3. VECTOR FIELDS AND NFORMS On the other hand. ie. (3.16) gives (XoY Y oX)(fg) = g(X0Y Y o X ) f + f ( X o Y Y 0X)g (3.11) of a vector field as a firstorder partialdifferential operator it is clear that the simple composition X o Y cannot be a vector field since it includes secondorder differential operators as well as those of first order.33. and it is easy to see that nowherevanishing vector fields exist on this space. Comments 1.17) and subtracting Eq. (3.Y o X is a vector field even though the individual composites X o Y and Y o X are not.. However.Y o X is called the commutator of the vector fields X and Y and is written as [X. (3. (3.1.5). In terms of the local representation Eq.1. which does not equal gX(Yf) + fX(Yg) as it would have to if it was to be a derivation.4). Since X and Y can be viewed as linear maps of C " ( M ) into itself.Y]. This new vector field X o Y .1.18) which illustrates the vital point that X o Y . (3.1.
Y ] are readily computed [Exercise!] to be (3. B .20) 3.1. . which satisfies the following conditions: Antisymmetry: For all A. (3.Y ]= [Y. (3. The components of the commutator [X.B ) G.1.1. The vector field commutator is related in an important way to the idea of a Lie algebra.denoted ( A .23) is not an identity in the strict sense. it is just one of the conditions that the multiplication operation has to satisfy in order to be a Lie algebra.B ) I+ [AB] thought of a type of ‘multiplication’.Y ]is clearly antisymmetric in the sense that [ X .1.3.23) shows that it is not associative.1. the expression Y p . a real Lie algebra is defined to be a real vector space L with a bilinear map L x L + L. B E L. If X . 2.[AB].1. Y and 2 are any three vector fields on M then their commutators satisfy the socalled Jacobi identity: 4. (3. v denotes the derivative of the function Y” with respect to the coordinate xv in the sense of Eq. VECTOR FIELDS 103 the firstorder differential operators are left. as is conventional. is the condition Eq.19) where.1. Note that the word ‘identity’ is somewhat misplaced here: Eq. XI. (3. [AB]= [AB] Jacobi identity: For all A. The algebraic pairing of vector fields X .15).23) If the operation ( A .1.22) [A[BC]] [B[CA]] [C[AB]] 0. = + + (3. (3. Y to give a new field [ X . C E C. In general.
Given a vector field X on M and a smooth map h from M to N.1. if h is not onetoone) then the definition in Eq.5 we defined the pushforward h. We shall return later to this idea and also to the important question of when the infinitedimensional Lie algebra VFld(M) possesses a nontrivial finitedimensional subalgebra.e.3 hrelated vector fields In Section 2.24) but. In this case. and the question arises of whether it is similarly possible to define an induced map from the vector fields on M to those on N .1. the equation [ A .23) is a genuine ‘identity’. B .[C. (3. (3.(Xp2).1. C E M ( n .104 CHAPTER 3.[ B . andas in the matrix casethe equation Eq.. with [AB]:= A B . a natural choice for an induced vector field h.1). : TpM + ThcP. as can be seen with the aid of the following diagram (Figure 3.B ] ]= 0 is automatically satisfied for all A.BA. Eq.A]] C [C. 0 3.203. This result is related to thesomewhat heuristicassertion that a vector field is a ‘generator of infinitesimal transformations’ on the manifold.21) is a genuine identity.21) have the important implication that VFld(M) has the structure of a real Lie algebra. ] ] [ B . (3. this may fail to be welldefined for two reasons: (i) If there are points pl and pz such that h ( p l ) = h ( p z ) (i.1.) # h. + + The two results in Eqs. ie.IR) by virtue of the rule for matrix multiplication. V E C T O R FIELDS AND NFORMS An important example of a Lie algebra is the set M(n.R) of all n x n real matrices.24) will be ambiguous if h.3.1.[A.X on n/ might appear to be: ( h * X ) h ( p ) := h*(X*) (3.1. ..(X. This algebra is infinitedimensional andwith due carecan be regarded as the Lie algebra of the infinitedimensional Lie group Diff(M) of diffeomorphisms of M .N induced by a map h : M + N.1. This was a linear map between the vector spaces TpM and Th(plN. (3..
1. and if f E Cm(n/). However.2 Let h : M + N and let X and Y be vector fields on M and N respectively.1.1. If X and Y are hrelated.1: The problem of pushingforward a vector field. Note that if h is a dzgeomorphism from M t o N then neither of these objections apply and an induced vector field h. VECTOR FIELDS 105 Figure 3.X. Then X and Y are said t o be hrelated if. (3.1.then . and this motivates the following definition: Definition 3.25) Comments 1. it is possible that in certain cases the idea will work. (3. even if h is not a diffeomorphism. (3.24).24) does not specify the induced field outside the range of h. h*(XP)= Y h ( P ) and we then write Y = h. at all points p in M .3. (ii) If h is not surjective then Eq.X can be defined via Eq.
] is hrelated to [Yl.] = [h*Xl.X. Eq.h*Xz].106 CHAPTER 3. this relation plays an important role in the development of the theory of the Lie algebra associated with a Lie group. Thus [ X I . with Yz] h*[ X l .1.X.27) can be used to write which means that for all p E M . r L h Figure 3.X1 = Yl and h. 2.1.31) As we shall see later. . If h. (3.2).Xz = Y2then. (3. VECTOR FIELDS AND NFORMS which is a useful way in which to express the property of hrelatedness (see Figure 3.2: The functions involved in hrelatedness. for all f E C”(M).
2 3.2. given a vector field X on M is it possible t o ‘fill in’ M with a family of curves in such a way that the tangent vector to the curve that passes . Namely.2. INTEGRAL CURVES A N D FLOWS 107 3.1) appear frequently in the physicsoriented literature but the deceptive simplicity of this equation covers up a fair amount of subtlety. It seems intuitively clear that if we have a smooth.3.3.2. as shown in Figure 3. We start by recalling that tangent vectors can be regarded either as equivalence classes of curves or as derivations defined at a point in the manifold.1 Integral Curves and Flows Complete vector fields In this section we shall discuss the precise sense in which a vector field can be regarded as a generator of an ‘infinitesimal’ diffeomorphism of a manifold. nonintersecting family of curves on a manifold then the tangent vectors at each point can be taken together to form a vector field. Figure 3. More interesting however is the converse question. Expressions such as c5xp = E X P ( Z ) (3.3: A field of tangent vectors.
and the family of curves that ‘fit’X.2) (3. play an important role in the theory.3) for all t in some open interval ( . For example. (3. Comments 1. where H : S + IR is the energy function.2.4) . Definition 3. From a physical perspective. E) of IR. the curves associated with the vector field X H . this idea is extremely important. Returning to the case of a general manifold M . the derivation X ? . .2. X can be written as (see Eq.2. The manifold S is equipped with a structure that associates to each such function f : S + IR a vector field XIon S. in the general canonical theory of classical mechanics. Then an integral curve of X passing through the point p is a curve t e a ( t ) in M such that 4 0 ) =P and (3.11)) (3. In local coordinates. and (ii) is such that the tangent vector at every point along this curve agrees with the vector field evaluated at that point.108 CHAPTER 3. and physical quantities are represented by realvalued differentiable functions on S. This idea is contained in the following definition.3 Let X be a vector field on a manifold M and let p be a point in M .1. the state space of a system is represented by a certain (evendimensional) manifold S. we consider first the simpler case of finding a single curve that (i) passes through a specified point p E M . are just the dynamical trajectories of the system. In particular. i e . VECTOR FIELDS AND NFORMS through any particular point p E M is just the vector field evaluated at that point.E .
Solutions in overlapping open sets can be patched together to some extent but this does not guarantee the existence for all values o f t E R.E . the classical Picard iterative method or one of the more 0 modern approaches based on fixedpoint theorems. This motivates the important definition: Definition 3. . it is a different matter when it comes to the existence of solutions thah are defined for all values of the parameter t. One approach is to employ the equivalent integral equation z’(a(t)) = xp(p) + / X’(a(s)) ds t 0 (3. (3.4 A vector field X on a manifold M is complete if.m (3. . INTEGRAL CURVES AND FLOWS and. (3. .6) subject t o the boundary condition z(() ’a0) p = 1 .5) and hence the components X p of X determine the integral curve t I+ a(t) according to the ordinary differential equation X’((T(t)) = z”(a(t)) p dt = z”(p) d = 1 . . 2 . the integral curve that passes through p can be extended to an integral curve for X that is defined for all t E JR. E ) of R.(t. 2 . However.6) with initial data Eq.2.2.3. at every point p in M . (3. . .8) which may be solved uniquely for some range of values for t using. .3) implies that 109 (xx’)(G(t)) = x. (3. By some such means we can demonstrate the existence of an integral curve that passes through the specified point p and is defined for t lying in some open interval ( .7) can be tackled by any of the familiar analytical methods. .2. for example.2.2.7) 2.2.2.m.(x’) = a* (g) t x = xp ’ d dt o a(t) (3. . Eq.2. in the present case. if M is noncompact a locallydefined integral curve may run off the ‘edge’ of M after only a finite range o f t values. For example. The discussion of the existence and uniqueness of solutions to the system of firstorder differential equations in Eq.
On the other hand. this corresponds to the wellknown difficulty in defining a selfadjoint momentum operator p^ in the situation where there is an infinite potential barrier at z = 0 that confines the particle to the positive real axis. In wave mechanics.p^l = i (3. 2. the vector field xd/dx is complete on R+.&the differential operator i X p is symmetric. In the problem of quantising a system whose classical configuration space is R+. + + Examples 1.this means that it is better to use the observables z and xp rather than the usual pair IC. This is closely related to the fact that for any tolerably wellbehaved pair of selfadjoint operators 53. Since it is complete vector fields that can most readily be identified with the Lie algebra of the diffeomorphism group it follows that the study of Diff(M) is considerably easier when M is compact than when it is not. The vector field d/dx is complete on IR but not on the open submanifold IR+ of positive real numbers. X is complete (Abraham & Marsden 1980). p that are appropriate for quantum . 2. it can be shown that i X p is a genuine selfadjoint operator if and only if. a situation that often arises is to have a vector field X on a manifold M that is equipped with a volume element/measure d p and a realvalued as a linear operator on the function p on M such thatregarded complex Hilbert space L 2 ( M .2.110 CHAPTER 3. When using differential geometry in theoretical physics. It can be shown that if the manifold M is compact then any vector field is complete (Abraham & Marsden 1980). or in a gauge theory.9) it can be shown that the eigenvalue spectra of both operators must be the whole real line. j3 that satisfy the canonical commutation relation [i?. as a vector field. This is highly relevant in discussions of the quantisation of a classical mechanical system whose configuration space is a nontrivial differential manifold. asfor e x a m p l e i n 0 a system with constraints. VECTOR FIELDS AND NFORMS Comments 1. However.
INTEGRAL CURVES AND FLOWS 111 mechanics on R.2. > 0. If 7r := xp then the analogue of the CCR Eq.2. oneparameter group o f local diffeomorphismsat a point p in M consists of (see Figure 3. The appropriate starting point is the following definition: Definition 3. not the unit operator) appears 0 on the right hand side.2.9). (iii) a family {$t I It1 < E } of diffeomorphisms of U onto the open set & ( U ) c M .3.2. (3.2 Oneparameter groups of diffeomorphisms We turn now to consider the sense in which a vector field.9) is the socalled ‘affine commutation relation’ [E.2.5 A local.2. genuine selfadjoint operators can be found with the spectrum of 2 being just R+ as required. (3. (ii) a real constant E with the following properties: 2For an extensive discussion see Isham (1984). = i2 Z] (3. whenever there is some generalization of Eq.’ 3. complete or otherwise. can be regarded as the generator of ‘infinitesimal’ transformations on the manifold.9) it is almost invariably a more complicated version of Eq.2. However.4): (i) an open neighbourhood U of p . (3..10)in the sense that a canonical operator (ie.9). (3. (3. This particular example gives just the vaguest hint of how the usual quantum mechanical rules might need t o be amended when dealing with a situation in which the classical configuration space is nontrivial. In the general case there is no natural analogue at all of the conventional CCR of Eq. which is closely locked to systems whose configuration and phase spaces are simple vector spaces.10) for which. .2. unlike Eq.
(ii) For all t . s E R such that Itl. $o(Q) = Q . The second occurrence of the word refers to the fact that the diffeomorphisms are .E) x U (t. $s(q) and q5t+s(q) also belong to U .2.4: A local.112 CHAPTER 3. oneparameter group of local diffeomorphisms. The first occurrence of the word ‘local’ in the definition refers to the fact that the diffeomorphisms $t are only defined for t lying in some small open interval ( .2.4) + M 4th) I+ (3.13) Figure 3. (3.E . and for all points q E U such that 4 t ( q ) . V E C T O R FIELDS AND NFORMS (i) The map (E. we have + SI (iii) For all q E U . Comments 1. E ) of the real line. Is1 and It are bounded above by E .11) is a smooth function of both t and q.
and hence we can have some confidence in the procedure adopted here. E ) x U + M t o be differentiable would be equivalent to requiring the differentiability of ( . whose domain and target spaces are both finitedimensional manifolds. in Eq.E) + DiffU(M). One may feel instinctively that one should require the map (E.2. in which case $2 o 41 also belongs t o DiffU(M). The resulting vector field X4 .E) x U + M . this would involve subtle questions of infinitedimensional differential geometry and. (3.e. t t+ 4t has been replaced by the. the answer is “yes”.5). but to give a proper meaning t o this requires making DiffU(M) into a differentiable manifold in its own right. i. Since there is such a curve passing through any point in U we can obtain a vector field on U by taking the tangents t o this family of curves at each point (see Figure 3..11). E ) + DiffU(M) if the latter could be made meaningful. t o avoid this. By this means. It turns out that in analogous problems that are welldefined. For each open set U . t ++ q5t.E . this begs the question of whether requiring the single map CP : ( . the parameterised family of maps (E.2. The term ‘oneparameter group’ in the definition refers t o Eq. t o be ‘smooth’ in some sense. no problem in requiring CP to be differentiable. therefore.3.12) which shows that the map t t) 4t can be thought of as a representation of part of the additive group of the real line by local diffeomorphisms of M . we avoid the need to consider issues in infinitedimensional differential geometry.E . Through each point q E U there passes the local curve t t) & ( q ) . a standardbut highly usefultrick has been employed. But. 4. in lieu of the more problematic requirement on (6. There is. they are defined only on the (perhaps very ‘small’) open subset U c M . (3. INTEGRAL CURVES AND FLOWS 113 defined only locally on M . 2.2. E) + DiffU(M).E) + DiffU(M). we denote the set of all such local diffeomorphisms by DiffU(M). and note that a pair of such diffeomorphisms 42. cpl can be combined in the form 4 2 o only if 4 l ( U ) c U . of course. Thus DiffU(M) is only what is known as a partial group. Namely. equivalent. single map CP : (E. However. 3.
According t o the definition of an integral curve in Eq. 0 The following proposition seems intuitively obvious but is nevertheless worth proving properly: Theorem 3.14) for all points q in U cM.5: The vector field induced by a local oneparameter group. is said to be induced by the family of local diffeomorphisms and is defined as (3. E ) + M by & ( t ) := q5t(q). (3.2. V E C T O R FIELDS AND NFORMS Figure 3.2.3) we want to show that (3.E .114 CHAPTER 3.1 For all points q E U the curve t curve o f X 4 for It/ < E .15) . Proof * 4t(q) is a n integral Define the curve q5q : ( .2.
Local flows always exist. Then the right hand side of Eq. (3.6 Let X be a vector field defined on an open subset U of a manifold M and let p be a point in U c M .~ 2. by virtue of the usual existence and uniqueness theorems of ordinary differential equation the01y.3. going from a vector field to a group of diffeomorphisms.2.2.2. Then a local flow of X at p is a local oneparameter group of local diffeomorphisms defined on some open subset V of U such that p E V c U and such that the vector field induced by this family is equal to the given field X . But the problem of interpreting an equation like Eq. Comments 1. ie. e) can be extended 3 F ~ a proof of this and related analytical theorems on vector fields see Abrar ham 8z Marsden (1980). .16) Let t = s + u. and are unique.2. If X is a complete vector field then V can always be chosen to be the entire manifold M .16) is (3..2.3 Local flows We have seen how to go from a local group of local diffeomorphisms to a vector field.17) 3. INTEGRAL CURVES A N D FLOWS But 115 (3.1) involves the converse procedure.2. (3. The central concept here is of a ‘local flow’: Definition 3. and the interval (e.
“(q) around the value t = 0 gives which can be regarded as a reasonably rigorous version of the heuristic Eq. 3. Eq. If X and Y are two vector fields on M it can be shown that [Exercise!] (3. (3. It is a measure of the extent to which the vector field Y changes as it is ‘dragged’ along the integral curves of X. (3.2.2. the local flow differential equation (cf. then the map t C) 4.20) where (3. VECTOR FIELDS AND NFORMS to the entire real line.116 CHAPTER 3.6)) 4.2.2. is an isomorphism from the entire additive group of the real line into the group Diff(M) of diffeomorphisms of M. A Taylor expansion of the coordinates d’(qhT(q)) of the transformed point 4.21) The left hand side of Eq. satisfies the (3.2. In a local coordinate system. 4. denotes the family of diffeomorphisms that is the flow for X.1) and the claim that a vector field generates ‘infinitesimal transformations’. We can say that this defines a oneparameter subgroup of Diff(M) and hence an element of its Lie algebra (see later for a full discussion of the relation between Lie groups and Lie algebras). (3. This is the sense in which a vector field is said to belong to the Lie algebra of Diff(M).20) is often called the Lie derivative of Y with respect to X and is denoted LxY. if 4. In this case. .18) around any point q in the open set U c M .2.
depending on the values of the real numbers a and b: .aY(t))= bo"(t).2. (3. (3. (3. Problem 1. + Answer (i) In the usual global Cartesian coordinates (2.23) = where X" and X Y are the components of X in this coordinate system.22) with respect to t and substituting Eq.2.3. compute the flow 4f of the vector field X = y&. y) on IR2. Differentiating Eq.2. and show by explicit calculation that. INTEGRAL CURVES AND FLOWS 117 3. if Y := x & the commutator [ Y .2. and where a p ( t ) := z p o a ( t ) .YY) . X ] satisfies the relation [ Y . (i) Find the integral curves of the vector fields ay$ bxg on the manifold IR2.24) and hence there are three types of integral curve. X Y ( o " ( t ) .2.22) (3.4 Some concrete examples of integral curves and flows The ideas of integral curves and flows can be usefully illustrated with the aid of some examples of a vector field on a simple vector space manifold.2. the integral curves t I+ ( a z ( t )aY(t)) of the vector field X := ay& . where a and b are an arbitrary pair of real numbers.2. b x g satisfy the differential equations + (t) dt (t) day dt da" = X " ( a " ( t ) aY(t))= aaY(t) . Are these fields complete? (ii) Using these results. (3.23) in the result gives (t) d2a" dt2 = a(t) doY dt = aba"(t). (0iy. X ]= limt+.
23) is now az(t) = Acoswtf Bsinwt  ay(t) = b(Asinwt Bcoswt) (3.2. d ) = (c + dt. d) (3.d ) E IR~.2. LCO = a z ( 0 ) and yo = aY(0)) is aY(t) = yo. A = a5(0)and B = aY(O)/b. + yo + bzot.2. (3) ab < 0: The solution to Eqs. To compute the desired limit. (3.Bcoshwt) (3. and subtract their values at a general point ( c .2. if a=O.2.Y and Y . d ) E R2.27) (3. Thus the flow q5$ of X is the oneparameter family of diffeomorphisms of R2given by zo + q$(C.2.223. We note that the vector fields are complete in all three cases since the solutions can all be defined for an arbitrarily large value of the parameter t.25) (3.Once again.26) where w := and A and B are integration constants.118 CHAPTER 3. aY(t) = (2) ab > 0: The solution to the differential equations Eqs.2. it is necessary to work out the components of the vector fields q5$. (3.23) is now 0% (t) = A cosh w t + B sinh w t ay(t) = b(Asinhwt . (ii) The integral curves for the vector field X = y& are # ( t ) = yot. Similarly. . a z ( t )= zo ayot. Thus we must subtract Y&) from 4F*(Y+XL~(C. a..d)).223. Clearly.29) for all points ( c . the components Zp of a vector field 2 are As usual.28) where w := Jab. aY(t) = yo. A = a z ( 0 ) and B = aY(O)/b. the integral curves are a"(t) = zo.yo) (ie.2. VECTOR FIELDS AND NFORMS (1) b = 0: The integral curve passing through the point (zo.2.
"*YY) t = 2 a dx .36) and therefore lim tto ((b. for each t E equation R.31) (C.x . INTEGRAL CURVES A N D FLOWS 119 d ) = (cdt.". d ) .33) Hence where we have also used the result ( & Y p ) ( f ) and thus.d) = ( c . d dy (3.30) (3.2. Hence.2.(f o 4).38) (3.X] =  xyx ) dY : x  a ( a ax  y. ay a .37) However [Y.2. d ) E R2.dt)t = (c dt) =Y.the vector fields 4.3. and we get ((b~.+y+X. for an arbitrary point (c.39) which proves the result.Y and Y satisfy the (3. using the fact that 4".d t .2.2. y (x).y.2. (3. equal to Z ( x @ ) .d))"=5 = (' .2.'(c. QED . hence.1(c.dt) (g) (x+ yt) (3.2. (3.
the sum X Y is not complete since the integral curve Eq.2. l/r). and hence cannot be defined for all t E R. VECTOR FIELDS AND NFORMS Let X := y2& and Y := zz& be a pair of vector fields on Show that they are complete but that X Y is incomplete.44) is the integral curve of X Y that passes through the point (l/r. Thus Eq. + (iii) The integral curves of X ferential equations + Y = y2& + z2gsatisfy the dif = (X" Y")(a"(t). a z ( t ) = zo ty. Y too is complete.43) ) = ~ dt which are not easy to solve explicitly. unlike X and Y individually. it is clear that the equations are consistent with setting a"@)= aY(t). the aY(t))of the vector field X := y2& satisfy integral curves t e (a"(t). + R2.2. do" (t) dt day (t) = X"(a"(t).2. + + . Thus the integral curve of Y := x2& is a"(t)= 50.Y ( t ) = ( ~ " ( t ) ) (3.44) has a singularity at t = r .2.2. aY(t))= 0 (3.120 Problem 2. (3..yo) E R2 is aY(t) = yo.44) da" (t) + + rt for any T E R. However. (3.2. shows that aY(t) = yo tII::.y ( t ) ) ( ~ " ( t ) ) (3. by inspection. Clearly this curve is defined for any value of t E IR.42) a ) ~ dt day (t) = ( X y Y y ) ( a " ( t a. However. Answer (i) In the usual global Cartesian coordinates (s.41) dt from which it follows at once that the integral curve of X that passes through the point (zo.y) on R2.and hence the vector field X is complete.2. aY(t))= (aY(t))2 (3. but with II: and y interchanged. the differential equations = XY(a"(t).40) + (ii) The same analysis. and then the solutions of the single ensuing differential equation da"(t)/dt = ( a z ( t ) ) are 2 1 a"@) a y t ) = = (3. CHAPTER 3.
If V is a real vector space.L E V " and w E V.1 Cotangent Vectors The algebraic dual of a vector space For any real vector space V there is an associated 'dual' space of all realvalued linear maps on V. ?I) := r ( L . COTANGENT VECTORS 121 3.?I) for all T E IR.en} is a basis for V . First we briefly recall some of the principle ideas in the theory of dual spaces.2 . If L : V + W is a linear map from a vector space V to a vector space W .3. if it is necessary to emphasise the vector space involved. Lz E V ' .2) 3. fn} in V* that are determined uniquely by the requirement (fi.w ) . The value of L E V * on w is conventionally written as ( L . .3. If V has a finite dimension n. .3.ej) = b! 3 (3. 'u E V. The set V* can be given the structure of a real vector space by defining (3.?I)V := ( k . 4. the dual L* : W * + V * of L is defined on k E W * by (L*k. .3.M.3) for all i.3. .4) . 2. qw. or as ( L .7 1. and if {el. w ) + ( L 2 . v ) . . n. (3. .3 3.1) (L1 + L2. p E M . w ) := (L1. (3. j = 1 . When applied to the vector spaces T. . . the (algebraic) dual of V is defined to be the set V* of all linear maps L : V + R.3. v ) for all L1. the dual basis for the vector space V * is the set of vecf tors {f'. Definition 3. e 2 .3. and ( T L . . this wellknown construction turns out to be of the greatest value.
and hence V and V* are isomorphic as vector spaces. there is a unique Hausdorff topology on V for which it becomes a topological vector space. in the infinitedimensional case there are typically many linear maps from V to R that are not continuous. (3.3.3. which is invariably chosen so that V is a topological vector space. This isomorphism is noncanonical (in the sense that there is no basisindependent isomorphism) but. any linear map from V to R is necessarily continuous. This requirement is relatively unimportant if V has a finite dimension since.122 CHAPTER 3. 2. As we shall see shortly. If V has a finite dimension] then the map ! x in Eq.3) means that dim V = dim V*. it is important t o take explicit note of the topology on V. one may wonder what the point is of constructing V* if it is isomorphic to the starting space V.4). For any vector space V there is a canonical embedding (V")"defined by x :V + (x('u>. However.5) for all 'u E V and f E V*. The question of the dual space is quite different if V has an infinite dimension.3.3. Comments 1.5) is an isomorphism. We note that the map L* : W* + V* is linear with respect to the vector space structures on the dual spaces W * and V*.* := (4 'u)V (3. and hence it is necessary to distinguish between the algebraic dual V*defined to be the set of all linear maps from V t o Rand the topological dual V'defined to be the set of all continuous linear maps from V t o IR. as mentioned earlier. 4. i e . 3. nonetheless. The existence of the dual basis defined by Eq. in the context of differential geometry with V = TpM the answer lies in the existence of the pullback map defined in Eq. Furthermore. (3.The existence of this important distinction adds much complexity . (3. the addition of two vectors and the multiplication of a vector by a scalar are required t o be continuous operations. . VECTOR FrEms AND NFORMS for all 'u E V. In this case.
M of all such linear maps. it can be equipped with a natural structure of a 2mdimensional differentiable manifold. 4.6) It is a vector bundle and. and is one of the reasons why the development of infinitedimensional differential geometry requires the use of considerably more sophisticated tools drawn from functional analysis than does the finitedimensional case. COTANGENT VECTORS 123 t o the discussion. 0 3.3.. and (ii) dim Tp*M = dim TpM [= dim MI.3. In this context. PEM (3. 3. as in the case of the tangent bundle T M .3. The value of k on the tangent vector w E TpM will be written as (k’w). A oneform w on M is a smooth assignment of a cotangent vector wp to each point p E M . The cotangent space at p E M is the set T.5). The cotangent bundle T*M is the set of all cotangent vectors at all points in M : T*M := u T. Definition 3. Section 2.3. ~if) it is ~ necessary to emphasise the point in the manifold with which we are concerned. ‘smooth’ means . ie. 2.3. it is the dual of the vector space TpM.8 1.2 The main definitions We come now to the topic of cotangent vectors and differential formsideas that are of the greatest importance in the mathematical development of differential geometry and in its applications to theoretical physics. see below (cf.M.or ( k . It follows from the discussion above that (i) this dual space is itself a vector space. A cotangent vector at a point p in a manifold M is a real linear map k from TpM into IR.
X. Equivalently.).3. 2. not a real number.M D. (3.3. For our purposes. It might perhaps be construed as asserting that a oneform is a dual of the space of vector fields in the same sense that a cotangent vector is an element of the dual space of a space T.7). 3. since (w.7) M (w.7)is a function X) on M . this is certainly not the case if the set of vector fields is regarded as a vector space over IR. As was shown in Section 2. and as discussed above.3.5. there is an associated basis for the .3.8) of derivations for the tangent space T. but even then care is needed as these are infinitedimensional spaces and topological subtleties arise in discussions of the precise structure of the dual of such a space.M of tangent vectors.3. A very important application of these ideas is to classical Hamiltonian dynamics. it is safest just to accept (3. VECTOR FIELDS AND NFORMS that for any vector field X on M the realvalued function on (3. a oneform can be defined as a smooth crosssection of the cotangent bundle (see below). ( P >:= ( U P . However. as defined in Eq. (3. As usual in linear algebra. If such a classical mechanical system has a configuration space Q then the space of classical states is defined to be the cotangent bundle T*Q.M. The analogy becomes closer if the vector fields are regarded instead as a module over the ring C m ( M ) . Care needs to be taken when interpreting Eq.7) as it stands without trying to reinterpret it in the full language of duals of topological vector spaces. any local coordinate system around a point p E M generates a basis set (3.124 CHAPTER 3. W is smooth. Comments 1.3.
(3. and we write m w=Xwpdxp p=l (3. These remarks extend to a local coordinate representation for oneforms.3. v = 1 . 4. COTANGENT VECTORS dual space Tp*M.3. Eq.3. satisfying (3.14) which is short hand for .*M can be expanded as k = C kp(dXp).9) and is defined by (cf.3.30)) up = u ( z p ) .3.This is denoted 125 (3. p= 1 m (3. .11) where the components of k with respect t o the given coordinate system are the set of real numbers k.3.. can be u) written in terms of these local coordinate components as (k. . kpVP (3. Thus any k E T. It also follows that the pairing (k.13) where (see Eq.3. = u) p=l c m .10) for all p. (2.12) where p = 1 . .3)) (3.m.m. 2 .. . . .3. . .3.3. 2 .
VECTOR FIELDS AND NFORMS for all p in the domain of the coordinate chart. 0 3.. More precisely.4) is a local coordinate chart on M then the associated 2m coordinates of a cotangent vector k at any point p in U are defined to be (i) the m real numbers { d ( p ) .2 in the context of Lie group theory.3.km} where k p := ( k . We shall see an important example of this idea in Section 4.16) for all q E U cM. . .P ( p ) } .z 2 ( p ) . 6. There are some situations in which it is useful t o generalise the notion of a cotangent vector t o allow for linear maps from TpM t o a general real vector space V. Clearly this is analogous to the way in which a vector field can be defined as a smooth crosssection of the tangent bundle T M . . . if (U. as a smooth crosssection of the cotangent bundle T * M . be used t o ‘pull back’ a oneform on N. Here the components wp of the oneform w are the functions defined on the domain U of the coordinate chart by (3. The cotangent bundle T * M can be given the structure of a 2mdimensional differentiable manifold in a way that closely parallels the manner in which this was done for the tangent bundle T M . and (ii) the m real numbers { k l . . This construction provides an alternative way of defining a oneform: namely. . Namely.3 The pullback of a oneform Now we come to a most important property of oneforms. even though it is not true in general that a map h : M + N between two manifolds M and N can be used to ‘push forward’ a vector field on M . .3. it is deeply connected with the way in which global topological structure of a manifold are reflected .3. ( & ) p ) p .126 CHAPTER 3. k 2 .This feature of forms is of the utmost significance. 5. not just the real numbers. A oneform of this type is called a Vvalued oneform. it CUR.
17) L h Figure 3.6: The definition of the pullback h*k. .3.4). . if w is a oneform on N then the pullback of w is the oneform h*w on M defined by (h*w. := (k. The first step is to define the ‘pullback’ map h* : TLjp)N Tp’M + on the individual cotangent spaces using the construction defined in Eq.)N and Y E T p M .18) E for all points p E M and all tangent vectors TpM.3. (3.(cf.. If h : M + N we have the linear map h. : T. Specifically. (3.3. (h*k. := ( w .)p h*Y)h(p) as illustrated in Figure 3. 2. It is also closely linked with the fact that fibre bundles also ‘pull back’.9 1. Definition 3. for all k E T&. h*Y)h(p) Y (3.M + Th(p)N.3. and then the map h* : Ti(p)N Tp*M is defined as the dual of h. COTANGENT VECTORS 127 in the DeRham cohomology groups of M defined using differential forms.4)). It is matter of the greatest importance that this definition can be extended to oneforms.3.3. (3. + Thus.6. Eq. not ‘push forward’ (see later).
46) of the pushforward of a tangent vector expressed in local coordinates on the domain and target manifolds.P}and {y’.3.gives the result (3.(a/dzp).22) where we have used the definition Eq.128 Comments CHAPTER 3. we saw earlier in Eq. There is an analogue for the pullback of a oneform of the expression Eq.)” (3. can be expressed in terms of the Jacobian at matrix of the map h. . (3.3.19) T.3.3. if h : M + N .22).3.18) of the pullback operation and Eq.(p) = (h*w.:= ( w . p lies in the domain of the zcoordinate system on M .23) . Then if w is a oneform on N . in addition. It is therefore not necessarily correct to think of h* as inducing a map from the complete cotangent bundle T*N into T’M: this will only be the case if h is a surjective map from M to N . . However.)h(p) (3. dinates on M and N respectively.)” the pushforward of the derivation of (a/az”)p the point p .3.11)) (k o h)* = h*k*.. (3.3.then (cf. h.21) = u=l x n wu(h(p))(h*(a/azP). . If. . The pullback map h* is only defined on those cotangent spaces q E n/ is such that there exists p E M with q = h ( p ) . 3. Eq. . let {d.3.3. (3.3. . we have the local coordinate representation Wh(p) = u=l x n w u ( W )( W ) h ( p ) (3.13). (2. (2.(a/azp)).y“} be local coorx2. N and P are three differentiable manifolds with maps h : M + N and k : N + P. VECTOR FIELDS AND NFORMS 1.20) for all points p E M such that h ( p ) lies in the domain of the ycoordinate system on N . Inserting this into Eq.*Nfor which the point 2. with respect to this coordinate system of the pullback h*w are (h*w). . Specifically. then the components (h’w). y2.44) that the components (h. (3.3.(a/az@). If M . (2.
b) = U l ( h ( U .23).b) := (a2.4 A simple example of the pullback operat ion Let us illustrate the pullback operation with the aid of the same map h : IR2 + IR2 that was used in Section 2.b) (3.b) + wz(h(a. y 2 }on the target manifold R2 by W(C.d) (3.25) for all points (c. = w1(a2 .3. we get ( h * w ) l ( ab) . and the expressions in Eq.3. (3. (3.2b)212~2b2 . b ) in the domain manifold R2.2b.3. 4a3b2).2b)4a3b2 x 2~ + (a2 .24) Compute the pullback by h of the oneform w defined in global coordinates { y 1 .3.7 to illustrate the pushforward of a tangent vector.27) at any point ( a .b))&% b) (3.3. the components of the pullback h*w with respect to the global coordinate system { d .3. on the domain x2} manifold R2 are (h*w)&. (2.b ) d) hs2( a .3.2b. b) b) + wz(h(a.b ) ) s ( a . Answer According to Eq. (2.60) for the Jacobian matrix of the map h. 4 ~ ~ b ~ ) 1 2 a ~ b ~ = + (a2.46).3. 4a3b2)2a w ~ ( u ’. d ) E R ~ .26) dhl (h*w)z(a.d) + C2(dY2)(c.3.d) := C+Y1)(C. COTANGENT VECTORS 129 which is the analogue for pullbacks of the local representation of a pushforward in Eq. = w1(h(a.2b. Using the explicit form of h.3.b dh’ ah2 ) ) s ( a .3. 0 3. Problem The map h : R2 + IR2 is defined by h ( a .
3.5a’b + 6b’).2b)4a3b2 x ( . VECTOR FIELDS A N D NFORMS = ( h * w ) 2 ( b).28) w~(u’.b) 4a2b2(5a4 16a’b 12b2)(dx1)(a.2 b ) 2 8 ~ 3 b 8a3b(a4.3.10 The Lie derivative Lxw of a oneform w with respect to a vector field X is defined to be the rate of change of w along the flow lines of the oneparameter group q5f of local diffeomorphisms associated with X : dt (3.130 CHAPTER 3.2b.b) in the domain manifold IR2. (3.b) (3. oneparameter group of local diffeomorphisms t + q5:. and this oneparameter family of forms can be used to describe the way in which w changes along the flow lines of X .5a’b 6b2)(dx2)(.3.31) t=O where the derivative on the right hand side is defined in analogy to Eq. = ~ = = + 126’) (3. 4a3b2)(2) + w2(a2 .2b. More precisely.2. Comments 1. (3. E ) .30) + + at any point ( a . we have the following definition.3. 3.21). the final result is (h*w)(a.2 ) + (a2. Definition 3. 4a3b2)8a3b (a2 . we have the pullback oneform q5:*(w) (also on M ) .. For each t E (e.3.5 The Lie derivative An important special case of the pullback operation on a oneform w on a manifold M occurs in the context of a vector field X with an associated local.29) 4a2b2(5a4 16a’b  Hence.b) = +8a3b(a4 . It can be shown that [Exercise!]: .
37) C”(N). Recall that (i) on a function f .3. = C ( w p wX”+ w u x ’ > p 1.33) L x ( w .36) Exercises Show that exterior differentiation commutes with the pullback operation in the sense that.L X Y ) for all vector fields X and Y . it is convenient for this reason to think of a function as a ‘0form’) and which is of the greatest significance: Definition 3.3.Y ] (see Eq. and (iii) In a local coordinate system.35) (3. COTANGENT VECTORS 131 (3.20)). (3.34) 2. . := Xf = L x f X) for all vector fields X on M .Y ) = ( L x w . In local coordinates: (3.3.3. then h*(df)= d(f 0 h ) for all f E (3.3. (ii) L x Y := [X.32) (ii) (3.3.Y )+ ( w .3.11 The exterior derivative of a function f E Cm(M) is the oneform df defined by (df. u=l m (3. The Lie derivative L x f := Xf of a function f along X arises also in the definition of the ‘exterior derivative’ of fa concept that generalises to a general nform (see later. the components of the oneform L x w are related to those of w and X by @ x u ) .2. L x f := Xf.3. if h : M + N .
1) V@W It is easy to show that the pair (V @ W. there is an isomorphism x : V + (V*)* given by x : v + v** x(v)(k) := ( k . R ) is the vector space of bilinear maps from the Cartesian product V* x W * of the duals of V and W into the real numbers R. 4 (3. for finitedimensional vector spaces. VECTOR FIELDS AND NFORMS 3. if V is finitedimensional. This isomorphism is defined by j ( . .4.p ) is unique up to isomorphisms. General Tensors and nForms The tensor product operation In order to proceed from tangent and cotangent vectors t o general tensors and nforms it is necessary to introduce the idea of the tensor product V @ W of two vector spaces4 V and W .4. there is a natural isomorphism j : V 63 W + B(V* x W * . 5 ) ) that.4. w). (3.132 CHAPTER 3. written V 63 W .I) := (Ic. This definition is very general.)(I.. and for our purposes it is useful to exploit the fact that.3) 4Most of the material that follows is only welldefined for vector spaces whose dimension is finite.2) This can be regarded as an extension of the wellknown fact mentioned earlier (see Eq. The most general and powerful approach is via the ‘universal factorisation property’ in which a tensor product of V and W is defined to be a vector space. . ( 3 . 3 . and a bilinear map p : V x W + V @ W with the property that given any other vector space 2 and a bilinear map b : V x W + 2 there exists a unique linear map 6 : V @ W + 2 such that the following diagram commutes (3. The tensor product of infinitedimensional topological vector spaces requires careful consideration of the topologies concerned. R ) where B(V* x W * . but it is also rather abstract. €3 w)(Ic.4 341 .
4. and TpM stimes. (3. . It is convenient to allow this to include the vaiue T = 0.4) Comments 1.5) from the Cartesian product of Tp*M rtimes.12 A tensor of type (r.4. Special cases of type (T.s ) at the point p E M is as a multilinear map & TiM x >s< TpM+IR (3.4. (3. Using Eq.4.2) and Eq.4.s ) at a point p E M is an element of the tensor product space (3.s) tensors include P (i) TpJM = T*M (ii) Tp'l'M = (Tp*M)* TpM (iii) Tp'p'M is called the space of rcontravariant tensors (iv) T:+M is called the space of scovariant tensors. with the product in this case being defined to be JR. The notation & V means the vector space formed by taking the tensor product of V with itself rtimes.3) for the finitedimensional spaces TpM and Tp*M it follows that an alternative definition of a tensor of type ( r . 3. GENERAL TENSORS AND NFORMS 133 for all vectors v E V and k E V * Applying these ideas in the context of differential geometry leads to the following definition. 8 2. Definition 3. into the real numbers.3. In using this convention it should be noted that there is a canonical isomorphism W 1R 2 W for any real vector space W .
..’M with respect to in which a E Tp’l’M is viewed as a multilinear map from the product 0 of Tp*M rtimes and T. s ) t o each point p E M .. .S+s‘M defined ‘ is to be the multilinear map 6. The definition extends at once to include the idea of a ( r . This is clear from the fact that if {el. s ) tensor field on M .5) is used instead?) 5.4. I) E V”x W * .2). and (ii) when Eq.M stimes..4. (Exercise: What would ‘smooth’ mean in the case when (i) the tensor product is defined by Eq. . In the abstract definition (3. This definition extends to multiple tensor products. In particular.134 CHAPTER 3. .v)(Z.f2.4. of this basis are cy E T. In particular. w) E V @ W . e2.. n and j = 1 . . If dim V < 00 and dim W < 00 then dim(V8W) = dim V dim W . .4.1) of the tensor product V @ Wof two vector spaces V and W . In the more concrete definition (3.. (3.. This is simply a ‘smooth’ assignment of a tensor of type ( T . if cy E Tp’9”M and cy’ E Tp”1”M then cy @ a E Tp‘+T’. . VECTOR FIELDS AND NFORMS 4.en} is a basis set for V and if{fl.m}. the tensor product of 21 and w is defined to be the unique bilinear map from V *x W* into I R that takes on the value ( k . the tensor product of 21 E V and w E W (denoted @ w) is defined to be the element p(u. (3. . w) on ( k .. . .fm} isabasisset f o r W ( s o n = d i m V a n d m = d i m W ) then a basis set for V @ Wis { e i @ f j I i = 1. . a basis set for Tp’+Mwith respect to a local coordinate system is the set of all vectors of the form and the expansion coefficients cy&l.4).
2 The idea of an nform A very important example of a tensor field is an 'nform' where 0 5 n 5 d i m M . However.n . .n) that is totally skewsymmetric in the sense that. . 2 .X. the wedge product. .3.9) where X . w(xl. are arbitrary vector fields on M and deg(P) IX is the degree of the permutation P . .e.4. Xnl+l. . This is remedied with the aid of the following definition: + + + + Definition 3. .14 If w1 E A"'(M) and w z E A n 2 ( M ) . 2 . We shall be interested in applying the tensorproduct operation to nforms. and a lform was defined in Section 3.X. if w1 E A"' (M) and w2 E An2(M) then w 1 @w2 is an (nl nz)covariant tensor field but it will not be an (nl nz)form since it does not satisfy the alternating property with respect to all its 'indices'.. GENERAL TENSORS AND NFORMS 135 3. or exterior product..13 An nform is a tensor field w of type (0. . . w1 @ w z ( X 1 . . for any permutation P of the indices 1 .1 if P is odd. ..4. it is +1 if P is even and . . . A 0form is defined t o be a function in G"(M). .3. The formal definition of a general nform is as follows... i.4. .e.. . P ( z ) . xz. of w1 and w2 is the (n1 nz)form WI A w2 defined by + ~1 A w z :=  1 n1 !nz!permsp C (l)deg(P)(~l @~ 2 ) (3. will antisymmetric under permutations of the indices that take an index k in the range 1 5 lc 5 n1 into the range n1 1 5 k 5 n l nz.> x P ( n ) ) (3.10) ~ . . Definition 3. .4.Xnlfn2) not be . x . . 7 xTl) = (l)deg(P)W(Xp(l). We shall write the set of all nforms on M as A"(M). i.
if h : M + N and if a and . the pullback of a 1form defined in Eq. if w is any tensor field of type (O. (3. VECTOR FIELDS AND NFORMS where. on the manifold M .11) for all vector fields X I .X Z . More precisely. * .4. . elements then the coefficients in the expansion w ~ ~ ~ ~skew. This wedge product has many important properties with respect to the other structures that can be imparted to differential forms. O then h*(oA p) = (h*o)A (h*P). 2. X n ) :=W(XP(l).XP(n)) (3. The wedge product turns the vector space A ( M ) := @$ZoMAn(M) a graded algebra with into ~1 A ~2 = ( .14) Furthermore. . XP(Z).. (3.13) if w1 E A n l ( M )and w2 E An2( M ) . (3. It can be shown [Exercise!] that a basis set for the nforms evaluated at a point p E M is the set of all vectors of the form (dd")).. . Eq.4.l ) n 1 n z ~ 2 A ~1 (3. .136 CHAPTER 3. . A (dz'"). the pullback off E C"(N) by h : M + N is defined to be the function f o h in Cm(M).. .18) can be generalised in an obvious way to an arbitrary nform.4. A ( d ~ " ) ) . 3. . (In the case of a function/Oform. the 'permuted' tensor field w p of type (0. (3.4.3.A . xz. X ..cf.3.. if a differential form w expanded locally in terms of such are . .37) in this context . ~ ~ 0 symmetric under the exchange of any pair of indices.n).n)is defined to be WP(X1. are differential forms on N. For example.12) Another major example of structure preserved by pullback will appear below in the discussion of the exterior derivative of a general nform. .) This generalised pullback has the very significant property of 'commuting' with the wedge product. f Comments 1. .
.3. * .15 If w is an nform on M with 1 5 n < d i m M then the exterior derivative of w is the (n + 1)form dw defined by + i=l x(l)i+’w([Xi. Y is dw (X. ~ ~ dzp2 A . Xz. A (3. ~ . y y>> X>> (w. . A dxpn. X I .Y ( ( W ..dx”~ . . . (3. . . 2. . A d z p n (with all indices summed over) then dw = w ~ . .4.35) we defined the exterior derivative df of a 0form f E Cm(M)as a 1form and showed that it commuted with pullbacks in the sense of Eq.3. F i . . . The symbol K means . . .37).15) then simplifies considerably. .>= W w .4.. that the particular field X in the series is omitted. . Y I ) . T j .3. .Xn+l) i<j for all vector fields X1. P . . If an nform w is expanded locally as w = ~ ~ ~ ~ ~ . .” . Comments 1.  (3. We wish now to extend this definition to an arbitrary nform and to construct an ‘exterior derivative’ that will convert it into an (n 1)form in a way that commutes with pullbacks.4. The complicated formula in Eq. . The case of a 1form is of particular importance in the theory of fibre bundles and Lie groups. + Definition 3. Y) 3. ~ . in C w ( M ) . . . There are no nforms for n > d i m M and therefore it is conventional to define dw := 0 if w is a dimMform.4.15) Note that the notation X((w. Y ) )means the effect of acting with the vector field X on the function (w.Xn+l. .4.16) . . and the 2form dw acting on any pair of vector fields X .. . Xj]. (3. (3.dxpl A dxp2 A . . GENERAL TENSORS AND NFORMS 137 3. A.3 The definition of the exterior derivative In Eq. .
Y ) for all vector fields X and Y on M . property of the exterior derivative is that the value at a point p E M of the function dw(X1.138 CHAPTER 3. .4. for any f E C m ( M ) . .4. the exterior derivative dw is C"(M)linear in the sense that. (3..l ) d e g wA ~ l l dw2. but not obvious. VECTOR FIELDS AND NFORMS 4.17) 5. . ..18) 0 3. X . 2 .4 The local nature of the exterior derivative An important.+l) = fdw(Xl.19) for all i = 1 ..X. (3. The definition (3.20) Proof .2 If w is a oneform on a manifold M . for all f E C " ( M ) .X2. . (3..Y )= f d w w .15) of exterior differentiation gives commutativity with the act of pulling back: if h : M + N and w is an nform on N then d(h*w)= h*(dw). For the sake of simplicity we shall only prove this result for the case of the exterior derivative of a oneform..4..X 2 .4. ..4.n 1... + Theorem 3.15)..X.depends only on the values of the vector fields X 1 ..) defined in Eq.X 2 .fXZ... ....X2. This is a corollary to the result that. The exterior derivative behaves rather nicely with respect to the wedge product and the associated graded algebra structure on A ( M ) ... the exterior derivative is 'flinear' in the sense that dw(Xl.4. X.specifically d(w1 A Q ) = dwl A w 2 + ( . .4. . at the point p .Xi. (3. d w ( fx.+l) (3.
Y ) = dw(fY.. )= dw(Y. f(p)Xp.4. as functions on M . Then on the open set U . QED Comments 1. for all g E Cm(M).X) cancel.4.4. (3. using Eq. (3. (fX).23).4 ) in a neighbourhood of p E M . for ie. = Cr==l Xp(q)(ap).4. Secondly. fX) f ( w . To show that d w ( X .4.]= f[X.23) ( w . GENERAL TENSORS AND NFORMS 139 and the first observation is that (fX)(h) = fX(h) all h E C m ( M ) . . Y ) ( p ) depends only on the values of X and Y at the point p E M . using the derivation property once more. ) = fdw(X. so. the vector field at q E U can be written as X.24) and Eq.Y)(p) and = . (3. X ) = and so using the derivation property of the vector field We also have. Y 2.4..4.Thus Y(f)X. (fX). the Y(f)(w.= (Up. choose a coordinate chart (U.22) where the last step follows because wp is IRlinear.4. so &(X. fX). (3. lation dw(fX. X ) fdw(Y. and we end up with the reterms Y Y). means that Y X) = z dw(X.) ( u p . f(P)XP) f(P)(% = = XP) (3. )is flinear in Y as well as in X.27) f X Y). so that dw(X. The equality dw(X.] Y Y  However.)= fdw(X. we note that = (w. (3.21). Y. (3.3. when Eq. Therefore.26) are inserted in Eq. we get and [fX.
2) in which the product of any two consecutive maps is zero by virtue of Eq. 5A special case of Eq. oAc(M)AA~(M)AA~(M)A . But X”(p) = X p ( z p ) .4.a v ) ( P ) = x p ( P ) y v ( P ) ~ ( av>(P) y” a. I have effectively introduced a space A’(M) := 0 with d : A .. This result is usually written succinctly as + d2 = 0 (3. and so these components (and hence hP) depend only on the values of the vector fields at p .28) and so & ( X ..l ( M ) +A o ( M )= P ‘ ( M )being defined to be the trivial map. AA~(M)+o . (3.5.1). Note that.5.5 DeRham Cohomology One of the most important mathematical applications of differential forms stems from the observation that if the exterior operator d is applied twice to any nform w then the resulting (n 2)form d (d (w)) is identically zero. A central ingredient in DeRham theory is the sequence of real vector spaces (3. VECTOR FIELDS AND NFORMS &(x’”a. (3. Y ” ( p )of the components of the vector fields X . QED 0 3.~ can be proved directly from the general formula (3. in local coordinates it is a simple consequence of the fact that mixed partial derivatives commute.1) It and is of the greatest ~ignificance.15) for the exterior derivative. for convenience.1) is the wellknown identity curl(grad(q5)) = 0 in standard vector calculus.5.4.5. Y at p E M . Y ) ( p )depends only on the values X P ( p ) . (3.140 CHAPTER 3.
5.% \ C .6) is a proper subset of . Thus boa=cob=doc=eod= or.6) In general.5.7. It is a special example of the more general concept of a differential complex which is defined t o be any sequence of maps and vector spaces (unconnected with differential geometry in general) of the form . DERHAM COHOMOLOGY 141 This sequence of maps and spaces is called the DeRham complex of the differentiable manifold M .5.5..4) c Ker(c)..2) there are the subspace inclusions Im(d: A"'(M) + A " ( M ) )c Ker(d : A " ( M ) + A"+l(M)). (3.5.. = O (3. Im(c) C Ker(d). the property d2 = 0 implies that in the sequence (3..3) with the property that the composition of any consecutive pair of the linear maps is zero..5) has a direct parallel in the case of differential forms. E ~ .. (3. Im(b) as shown in Figure 3. . whenever relations of this type arise there is always an associated cohomology theory in which the cohomology groups measure the extent to which the image set in Eq. . f o e = . Im(a) c Ker(b). (3.3.7: A differential complex.5. (3.5. Specifically. . equivalently.% D ~ .+ A . (3. F f".5. The sequence of inclusions in Eq.5) Figure 3.% B .
In the present context of differential forms this idea is captured in the following central definitions of the DeRham theory.O = w . IR" is contractible).thus B " ( M ) = Im ( d : A"'(M) + A " ( M ) ) = { w E A " ( M ) I w = d.16 1. it can be shown (Poincark's lemma) that on a Euclidean space I " every closed nform is exact for n > 0. Indeed. and this is contained in the following definition.O for some p E A"l(M)}.5. An nform w is closed if dw = 0. The set of exact nforms is denoted B " ( M ) .7) {w E A" ( M ) 1 dw = O}. A closed nform is also said to be an ncocycle for the DeRham cohomology.6 Definition 3.O such that d. (3. But when are they equal? This question of when a closed form is also exact has far reaching implications since the problem of proving that a given closed nform is exact involves global topological properties of the manifold. and hence the vector space B " ( M ) is a subspace of Z n ( M ) .5.142 CHAPTER 3.1)form . An nform w is exact if there exists some (n. For this reason it is desirable to find some precise mathematical construction that faithfully reflects the departure from exactness. It is clear from Eq.8) An exact nform is also said to be an ncoboundary for the DeRham cohomology. Hence R the failure of Z n ( M )to equal B " ( M ) does indeed reflect the degree to which the manifold M is not topologically trivial in the sense that the space IR" is (specifically. 'For a comprehensive discussion see Bott & Tu (1982) .5. Z"(M) := Ker ( d : A " ( M )+ = A"+l(M)) (3. VECTOR FIELDS AND NFORMS the kernel set. (3.6) that every exact nform is closed. The set of all closed nforms thus is denoted Z n ( M ) . 2.
5.11) for all connected manifolds M . Thus B o ( M ) (0) and Z o ( M ) R and hence (3. ie. a real number. 0 I n 5 d i m ( M ) . (3.(M) IR (3. On M = IR.. But then dw = g(x)dX dX a A d~ + g(z)d(dX) = 0 (3. If M is not connected but has k connected components. Since A'(M) := (0) it follows that Bo(M)= (0).5.3.5. then it is straightforward to see that Eq. if we define f ( 4:= &dY) dY (3. This implies that HhR(IR) E (O). since d i m R = 1). E IRk (3. and indeed. = g(x)(dx)z with respect t o the global coordinate system on IR.9) 1.5. DERHAM CONOMOLOGY Definition 3.the most general lform can be written as w. Comments (3.5.14) then df = g(x)dz and hence the lform w is necessarily exact.13) and so every lform is closed (as expected.5. : of a manifold M are the quotient vector spaces H g R ( M ):= Z n ( M ) / B n ( M ) .5.5.17 143 The DeRham cohomology groups HE. Since M is connected (by assumption) it follows that a closed Oform is simply some constant function. However. On the other hand.(M).10) HL.(M) R. since the Poincark lemma .12) which can be thought of as the direct sum of k copies of the real line 2.11) generalises t o = Hg. a function f E A ( M ) is closed if and only if its partial derivatives vanish in any coordinate system.
it follows that R" we have Zn(Rm) B"(Rm)for all E E HgR(IRm) (0) R forn > 0 for n = 0. (ii) if w is an exact form with w = dp then h*(d.144 CHAPTER 3.as before. . This equation has a very significant implication for the cohomological use of differential forms since it implies that (i) if w is closed then so is h*w.5.4 it was shown that exterior differentiation commutes with the act of pullingback in the sense that (Eq. However. unlike the case M = R. it is also true that every 1form is necessarily closed.17) where h : M + N and w is an nform on N. so that A1(S1) 2 Z1(S'). 4.5. Since this is connected we have H g R ( S 1 )% IR.5.18)) d ( h * w ) = h*(dw) (3.16) which reflects the most important topological difference between the circle and the real line: namely. are sin n0 and cos no. the former is not simply connected but has 7r1(S1) Z. it is not true that every 1form is exact. (3. in angular coordinates. In Section 3. the oneform d8 is not exact (not withstanding its appearance) since the angular variable 8 is not continuous or differentiable when considered as a function on the entire space S1.O)= d ( h * P ) and hence h*w is also exact.15) 3. since S1 is onedimensional. For example. and there is no differentiable function on the circle such that d8 = df . (3.4.7 It is not difficult to deduce from this that HhR(S1) R. VECTOR FIELDS A N D NFORMS implies that for any manifold n > 0. And. 7Examples of functions on the circle that are globallydefined and differentiable. "= (3. Now consider the onedimensional manifold S1.
a differentiable function h : M + N induces a linear map h* : H Z R ( N ) + HZR(M). (3. Thus. Two continuous functions f and g between topological spaces X and Y are said to be homotopic if they can be 'deformed' into each other continuously. But then the two results above imply that the pullbacks h*(wl)and h*(w2) lie in the same equivalence class in Z " ( M ) / B " ( M ) .18) A procedure of this type is an intrinsic feature of any useful cohomology theory: specifically. The important result for us is that this property can be shown to hold also for the DeRham cohomology of a differentiable manifold. This is no coincidence. suppose that w1 and w2 are nforms that belong t o the same equivalence class in the quotient space Z"(Af)/Bn(Af). cohomology structures.5.In summary.z) = 9(4.18) if they are homotopic to each other. the cohomology groups/vector spaces probe the topological properties of the manifold or topological space with which they are associated.z) = f ( x ) and F(1. there must be some continuous function F : [0.5. if f and g are two differentiable functions from a manifold M to a manifold ~lf induce identical homomorthen they phisms (3. for example) is that under these circumstances the associated homomorphisms f* : H " ( Y ) + H " ( X ) and g* : H n ( Y ) + Hn(X) (3.5.5.O such that w1 .3. 6. More precisely. A characteristic property of traditional cohomology theories (using simplicial decompositions. DERHAM COHOMOLOGY 145 Now. It follows from the above that the DeRham cohomology theory for differentiable manifolds has many general properties that are shared by the usual simplicial. or singular. Indeed. there Then must be some ( n . one of the most remarkable and powerful results in the theory of the global structure of differential manifolds is DeRham's theorem which asserts that the DeRham cohomology .1)form . and the properties of maps between such spaces are reflected in the induced homomorphisms between the groups attached to the spaces.1] x X + Y such that F ( 0 .w 2 = do. 5.19) are in fact equal.
20) and it is the existence of this entity that makes cohomology potentially more powerful than homology. simplicial group H”(M. the wedge product plays the role of the cup product. (3.5. In the conventional singular/simplicial theory this is a pairing homomorphism u : HP(X) (8HyX) + HP+4(X) (3. on the other hand.13) which states that differential forms have a similar graded structure under the wedge product. V E C T O R FIELDS AND NFORMS groups are actually isomorphic to the real cohomology groups computed using singular or simplicial theory..4.4. This possibility is strongly supported by Eq.R)is known. (3.5. Is there an analogue of this cup product for the DeRham theory? A clue is given by the skew commutativity of the product in the simplicial/singular case: u U b = (l)pqb U u (3. This identification of the DeRham and ‘topological’ cohomology groups can be used either way. From all perspectives.This is strikingly reminiscent of Eq. there may be cases where the simplicial groups can be computed using the differential forms on the manifold. differential forms are basically defined in a local way that rests heavily on the differential structure: the ability to probe global properties stems entirely from the special properties of the operation of exterior differentiation. Alternatively. Another satisfying link between the various cohomology theories concerns the ‘cup product’.17) which means that the conditions of closedness and exactness are preserved under . in DeRham cohomology. This really is very striking since the singular/simplicial groups depend only on the underlying topological structure of a manifold (ie. What is suggested therefore is that.146 CHAPTER 3.21) if a E W ( X ) and b E H‘J(X). DeRham’s theorem is of major importance. The question of the existence of closed but nonexact nforms can be answered at once if the (say) real. they know nothing of its differential structure of coordinate charts) while. 7.
5. The full version of DeRham’s theorem confirms this conjecture and shows that the wedge product is indeed an exact analogue for differential form cohomology of the cup product in the other theories..5. .5.(M) + H. DERHAM COHOMOLOGY the wedge product in such a way as t o induce a linear map 147 A : H.3.22) that satisfies Eq.21). (3.(M) 8 HL.2(M) (3.
.
A good example is the set { z E C I (z(’ = l} of all complex numbers of modulus one.1 The Basic Ideas The first definitions The key idea of a Lie group is that it is a group in the usual sense. Lie groups are of the greatest importance in modern theoretical physics. but with the additional property that it is also a differentiable manifold. It should really be supplemented by private reading on the general theory of Lie algebras. The formal definition of a Lie group is as follows. and the subject is one that can be approached from a variety of perspectives. and (ii) those aspects of Lie group theory that are especially appropriate for the section of the course that deals with fibre bundle theory.1. and their use in modern theoretical physics (for example.Chapter 4 Lie Groups 4. and in such a way that the group operations are ‘smooth’ with respect to this structure. in the classification of elementary particles). but it is also a manifold: namely. This is clearly a group under the action of multiplication.1 4. the circle S1. 149 . The discussion that follows is relatively elementary and is concerned mainly with (i) the geometrical relation between a Lie group and its Lie algebra.
CHAPTER 4 .:G+G 9/ 9g1  (4. the maps p:GxG h92) + I+ G g1g2 (4. The right and left translations of G are diffeomorphisms' of G labelled by the elements g E G and defined by rg:G Y ' Comments + I+ G.3) 1. 2.1. and (iii) a topological group with respect to the subspace topology. (b) a differentiable manifold with the properties that taking the product of two group elements. LIE GROUPS 1. A real Lie group G is a set that is (a) a group in the usual algebraic sense.150 Definition 4 1 .1. The maps p : G x G + G and L : G + G can be used to express the basic axioms of a group purely in maptheoretic terms. In the algebraic case they are only required to be bijections. Specifically. 'Note that left and right translations exist for any group G. A Lie subgroup of G is a subset H of G that is (i) a subgroup of the group G in the algebraic sense. are smooth operations. not just for Lie groups.2) are both C". 99 ' l. . (ii) a submanifold of the differentiable manifold G .1) and L : G + G 9 ++ 9 '  (4.1. and taking the inverse of a group element. 3. they are required to be homeomorphisms. if G is a topological group.
A somewhat weaker notion is that of a topological group. " It is a very deep and nontrivial theorem that every real C Lie group admits a real analytic structure in the sense that a set of coordinate charts covering G can always be found with the property that the associated overlap functions are C". the associativity condition gI(g2g3) = (g1g2)g3for all G. 4. This is defined to be a group G that is at the same time a topological space.g3E For example. THE BASIC IDEAS 151 g1.] Writing the axioms in this way is sometimes technically convenient in standard group theory. The left and right translations satisfy the relations '91 '92 = '9192 (4.4. where idc : G + G denotes the identity map on G. There is an analogous theory of real analytic Lie groups and complex analytic Lie groups in which the underlying manifold is required to be a real or complex analytic manifold respectively.1.1. 3. [Exercise: translate the remaining axioms for a group into this language. it constitutes the solution t o Hilbert's fifth problem. Gleason and Zippin (Montgomery & Zippin 1955). 2.4) p x idG between the maps G x G x G + G x G G and G x G x G + G x G %+ G.6) . It also allows a generalisation t o a 'grouplike object' in a category which is other than the category of sets.5) and (4.1. and in such a way that the maps p : G x G + G and L : G + G are continuous.g2. but the converse is not true. is equivalent t o the relation p o (idc x p ) = p o ( p x idc) idc x p (4. This famous result is due to Montgomery.1. Every Lie group is a topological group.
.1. This group is usually denoted U (1) and belongs to a series U ( n ) n = 1 . space. A map q5 : G1 + G2 is an antiisomorphism if it is a bijection and if both q5 and q5l are antihomomorphisms. a structurepreserving map) is defined to be a group homomorphism that is also a smooth map between the two underlying differentiable manifolds.7) This 6dimensional real Lie group plays an important role in the geometrical approach to complex analysis and in the theory of relativistic strings. This also applies to the maps G t) Diff(G) afforded by the left and right translations in so far as the infinitedimensional group Diff(G) can be equipped with the differential structure of a genuine Lie group. (4. 4. the maps g ++ I. d E C! with ad  bc = 1. 2. When equipped with its natural differential structure. As a topological . 2 . b. . the vector space IR" is an ndimensional (abelian) Lie group. The real.. Furthermore.1. 0 Examples 1.8) 2A map q : G1 + G2 between groups G1 and G2 is said to be an anti5 homomorphism if q5(ab) = q5(b)q5(a) for all a .152 CHAPTER 4 . 3. b E GI. it is compact. IR) := { A E M ( n . LIE GROUPS respectively. . In general. and g t T. are both ) injective and hence define an isomorphism and an antiisomorphism2 respectively from G into the group Diff(G) of diffeomorphisms of G. c. and defining the group combination law as multiplication of complex numbers.R) I det A # 0) (4. The group of Mobius transformations of the complex plane C! is defined to be the subset of diffeomorphisms with the special form zt) +b cz + d az where a . general linear group in ndimensions is defined as GL(n. The circle S' := { z E C! I IzI = 1) can be given the structure of a real onedimensional Lie group by employing the usual differential structure. of Lie groups (see below). a homomorphism between two Lie groups GI and G2 (ie. .
This property is important in many ways: for example. it has the structure of a.R) is an open subset of M ( n . or less than. and is hence open. this poses severe difficulties for the quantisation of any system that possesses such a group: a good example is the problem of quantum gravity. 6 .A I+ detA. in which case one has the famous Haar measure. R)II Rn2. as an open subset of the Euclidean space Rn2. 5. . diffeomorphisms.4. R ) . The subset GL(n. T H E BASIC IDEAS 153 and is thus the subset of the set M ( n . 0. C) I det A # 0). noncompact.R) has a noncompact topology.R) is the complement of this set. R ) is in bijective correspondence with R ' and ". although it is not compact.1.. One of the major problems with infinitedimensional Lie groups is that they are not locally compact. In particular. each point has a neighbourhood whose closure is compact.R) is locally compact.C).R) is the same as that of M ( n . R ) of all real n x n matrices made up of those matrices that have an inverse.the group GL(n.C) := {A E M ( n . locallycompact .9) As. and the role of the infinitedimensional group of spacetime. As such. dimenthe sion of G L ( n . Note that. 2n2dimensional Lie group. (4. or spatial. R ) + IR.R)namely n2and it decomposes into two disjoint components.an open subset of the 2n2 (real) dimensional manifold M ( n . The complex. Note that the map det : M ( n . ie. according t o whether det A is greater than. and it is with respect t o this structure that it is a Lie group. Since GL(n. and hence the Haar measure does not exist. general linear group in ndimensions is defined as GL(n. The space M ( n . Note that.R) (since (0) is a closed subset of IR). is continuous and hence detl{O} is a closed subset of M(n.1. the existence of a measure on a Lie group that is quasiinvariant under left or right translations is only guaranteed if the group is locally compact. the group GL(n. therefore it has a natural topological and differential structure. it acquires both a topological and a differentiable structure from M ( n .
154 CHAPTER 4. R ) I det A > 0) (4. R ) (ie. = (iii) det A' (det A)' and hence A E GL+(n. Put another way. locallycompact . 7. The special linear groups are defined as the following noncompact. LIE GROUPS Note thatunlike GL(n. 8.11) SL(n.C) I det A = 1).R) implies Al E GL+(n. R ) .IR):= {A E G L ( n .10) and is itself a noncompact. Lie group of dimension n2 . R ) I det A = l} and (4.12) .1.R)GL(n. ) which are precisely the three conditions for a subset of a group G to be a subgroup. the component containing the unit element 1)is denoted GL+(n.. The connected component of G L ( n .R . is a subgroup of GL(n.R)note R) that: (i) det(1) = 1 so that the unit element 1belongs to G L + ( n . subgroups of the general linear groups: S L ( n . C) is a connected space since any matrix A with detA > 0 can be connected continuously to a matrix B for which det B < 0 by a path in G L ( n .1. (ii) det AB = det A det B and hence A .C) := {A E G L ( n .R). locally compact. C ) on which the determinant is a complex number. R ) := { A E G L ( n . B E GL+(n. removing the single point 0 from the real line R disconnects it into two pieces whereas the same operation performed on Q: leaves a connected space.1. R) implies AB E GL+(n. To show directly that GL+(n. (4.
R ) := { A E G L ( n .1.R) = n2. Definition 4.1.1.14) plus the fact that dim G L ( n . R ) is a submanifold of G L ( n .13) Comments 1. the defining constraint AAt = I is equivalent to the equations I which impose (n2 n)/2 polynomial constraints on the matrices regarded as elements of M ( n .R)is algebraically closed under multiplication. 3. R ) of dimension + + . R ) . and hence the unit element belongs to O ( n . To see that this subset of G L ( n . } (4.R) defined b Y O ( n . 2. R )I AAt = I . In terms of the matrix elements of the members of O ( n .1.R). (ii) if AAt = B and BBt = II then ( A B ) t ( A B ) BtAtAB = II and = hence O ( n .R) in terms of the (n2 n)/2 polynomial constraints Eq.R). R ) .2 The orthogonal group A very important role in theoretical physics is played by the orthogonal groups and the unitary groups. (iii) if AAt = B then A' A'A = B and hence = At which means Al(A')t = AtA = A E O ( n . T H E BASIC IDEAS 155 4.2 The (real) orthogonal group is the subgroup of GL(n. The definition of O ( n .R)implies that A' E O( n . R ) is indeed a subgroup note that: (i) I l l t = B. (4. suggests that O ( n .4.
and p is the multiplication function p : G L ( n . Eq.8)it follows that the Lie group O ( n . (4. . For example. in general.IR). consider the chain of maps + where T : GL(n. Furthermore.IR) into itself with the property that O ( n . This has major implications in many different directions.^ However. this means that O ( n . 3This is not trivial since.14) implies that (4. R ) x GL(n.IR) + GL(n. p ( A .1.1)/2.156 CHAPTER 4.IR).IR)+ G L ( n .1.(n2 n)/2 = n(n . Now the composite function p := p o (id x T ) o A is a continuous map from GL(n.In turn.1. (4. (4.B ) := A B . LIE GROUPS n2 .JR) is a closed submanifold of G L ( n . the representation theory of a compact Lie group is far more tractable than is that of a noncompact group.14). and hence O ( n .16) But the inverse image of a closed set (such as { I ) under a continuous } map is itself a closed set. even without embarking on such a detailed analysis. R ) is defined by T ( A ) := A t . R ) is a closed and bounded subset of the vector space M ( n . the solution surface of a collection of polynomial constraints has singular point. for example.17) so that O(n. I R ) E Rn2.JR) is actually a closed subset of the sphere Sn2lc M ( n .IR) is a compact manifold.R) = / {} +I. there are important pieces of information that can be extracted from the general form of Eq.4. see Postnikov (1986).1. and a more careful investigation shows that this is indeed SO. R ) E IRn2 and henceby the HeineBore1 theorem (see Section 1. For example.
19) and is a compact Lie group of real dimension n2 ..q . The condition AAt = 1 implies that (det A)’ = 1. R)and SO(p. . In an analogous way the unitary group U ( n ) is defined by U ( n ) := {A E GL(n. .l ) .R). The definition of the orthogonal group has an important generalisation.2. q . with p ‘+l’entries and q ‘1’ entriesand such that p g = nthen the subset of GL(n. A case of particular importance in theoretical physics is O(3. C) with real dimension n’. the subgroup of matrices satisfying det A = 1 is denoted SO(p. + Apart from O ( n . R) are noncompact.1 . q = 0 (or q = n.18) and is a compact Lie subgroup of GL(n. R). . The closed (and hence4 compact) subset of O ( n . . if the vector space R”is equipped with the diagonal metric ( + l .q . 4. The special unitary group S U ( n ) is defined as SU(n) := {A E U ( n ) I det A = 1) (4. C) I AAt = 3) (4.4. and is denoted SO(n. and is denoted O(p.1 The Lie Algebra of a Lie Group Leftinvariant vector fields One of the most important features of a Lie Group G is the existence of an associated Lie algebra that encodes many of the properties of 41n general. A similar remark applies to the special orthogonal groups. R ) with p = n.1.q. THE LIE ALGEBRA OF A LIE GROUP 157 4. a closed subset of any compact space is itself compact.l)the Lorentz group that plays a major role in the physics of special relativity.IR). 5. . + l .2. R). .q. R) that leaves this metric fixed is a Lie group. all the groups O(p. The orthogonal group O ( n . locallycompact Lie groups. .2 4. Namely.R)and SO(n.I )introduced above corresponds to the R special case of O ( p . . 6. p = 0).1. R ) of matrices A satisfying det A = 1 is a subgroup [Exercise!] known as the special orthogonal group. so that [ detA = f l .1. It has dimension n(n .1)/2.
. rg. i. (from Eq.2. (3. rg. In particular. this even includes certain global topological properties of G such asfor examplewhether or not it is a compact space.related to itself for all g E G (see Eq.2...g’ Comments E G. LIE GROUPS the group. (4..2.X2] is hrelated to the commutator [Yl. equivalently.1) E.[X1.25)). A vector field X on a Lie group G is leftinvariant if it is 1. Rather remarkably.25)) that if vector fields X I and X z on a manifold M are hrelated to vector fields Yl and Y2respectively on a manifold n/. The key definition is the following. i.(Xgt) = Xgtg for all g.) = Xggt for all g.Y2]. Similarly. (X. this means that if XI and X2 are leftinvariant vector fields on G then lg.where h : M + N. 2.2) 2.1.2.3) or. tangentspace related structure around the entire group manifold. (3.4) 1.3 1. (4.X2] = = [l..158 CHAPTER 4. Definition 4. The critical observation now is based on the general fact (Eq.5) .. The crucial property of a Lie group that enables this to occur is the existence of the left and right translations of G onto itself that can be employed to map local.XZI.e.X1. It is clearly a real vector space.. then the commutator [Xl. equivalently. =X for all g E G (4.2. The set of all leftinvariant vector fields on a Lie group G is denoted L(G). (3.XZ] [Xl. X is rightinvariant if it is r.25)) (4.e.related to itself for all g E G.X or.X = X for all g E G (4. g’ E G. Z.1. ~.1.
6) (4.1 There i s a n isomorphism of L ( G ) . THE LIE ALGEBRA OF A LIE GROUP 159 Thus X l .~. /. what is the dimension of the real Lie algebra L(G)? The answer to these vital questions is given 0 comprehensively by the theorem that follows.7) where we have used I. (4. This result is very elegant. It is called the Lie algebra of G. X . how many of them are there? Or to be more precise..I*/.G t o G at the unit element e E G. and similarly L: = B.injective. and if such fields do exist.7) that LA is indeed a leftinvariant vector field. and so it remains to prove that it is onetoone and onto.A = A. Theorem 4.2. = l. regarded as a real vector space.. Proof An explicit isomorphism i : TeG + L(G) will be constructed by exploiting the lefttranslation operation. with the tangent space T. and hence = l g l g * . It follows from Eq.+ It is clear that A + L A is a linear map. Hence the map i : TeG + L ( G ) . (i) If LA = LB then. L t = L:.! o 1. 3.A Note that. and so the set L ( G ) of all leftinvariant vector fields on G is a sub Lie algebra of the infinitedimensional Lie algebra of all vector fields on the manifold G .g’ E G.4.2. for all g E G.2.2. Specifically. (4. .. if A E TeG let i ( A ) := LA where L A is the vector field on G defined by L t := 1. in particular. but it leaves open the question of whether there are any nontrivial leftinvariant vector fields on any given Lie group G. for all g. But L! = 1. E L(G) implies that [ X l . X z ] E L ( G ) . A L ~is .
2.2. 2. for all g E GI. Comments 1.B. Thus ~ ( A L ) L.l*L.8) Clearly. since L is leftinvariant. and define A t E T. Specifically. (4. and hence i : TeG + L(G) is surjective.+AL = Zg.(AL)= L. nontrivial subalgebra of the Lie algebra of all vector fields on G. the idea of a commutator is defined only for vector fieldsnot for vectors at a fixed point in Gand hence the Lie bracket on TeG has to be constructed from the commutator on the leftinvariant vector fields L(G) with the aid of the isomorphism i : TeG + L(G) discussed in the theorem. A significant implication of the theorem is that it should be possible to regard the tangent space T. LIE GROUPS (ii) Now let L be any leftinvariant vector field on G. and hence (4. G is a finitedimensional manifold it follows that L(G) is a finitedimensional.. It is clear that L F+ AL is the inverse of A C) LA.10) [AB]:= [LA.2.9) /. equivalently.(Le) = L. = QED N. (4.LB] or. [= Z.G as the Lie algebra of the Lie group G. B E TeG then the Lie bracket [AB] E TeG is defined to be the unique element in TeG such that LWI = [LA.G by AL := L. Since. Corollary The dimension of the vector space L(G) is equal t o dimTeG = dim G. .LB].160 CHAPTER 4. and Z. if A . by assumption. Z.11) The power and elegance of this way of viewing the Lie algebra of a Lie group is reflected nicely in the following result.x*(Lt) = A .. (4.2. However.
1.2. (3.6) is the definition R.2. from Eq.13) which says that the vector fields LA and Lf*(A) frelated..” := rg. Proof Since f is a group homomorphism it follows that. = f*(LA’)]j(e) = [L f * (ALf*(A’)]. and hence for all g E G f 0 1.LA’]e) [f*(LA). (4. THE LIE ALGEBRA OF A LIE GROUP 161 Theorem 4.6).2.30).A (4.) = = lf(g)*f*(A) Lf(.En}. Thus n (4. Then f*(L.2.G with the associated leftinvariant vector field LA E L( G ) (so that A = L t ) as defined in Eq.) = f*W 0 f)*(A) (4. f*[AA’] = f*([LA. = (4. ( A ’ ) ] ). [ f * ( A ) f .G. = lf(g) 0 f. 2. . A similar construction to the above can be carried out with a set of rightinvariant vector fields associated with the elements of T. g’ E G . . But are then. is a basis set for L ( G ) TeG. f ( g g ’ ) = f ( g ) f ( g ’ ) .A) = = f*l. = dimG.2. .16) . QED Comments 1. Then the induced map f* : TeG + TeH is a homomorphism between the Lie algebras of the groups.12) Let A be in T.E 2 . for all g.2 Let f : G + H be a smooth homomorphism between the Lie groups G and H .2.14) and hence f* : T.2.G + TeH is a Lie algebra homomorphism. If { E l .4. (4. The analogue of Eq.*(A) (f 0 1g>*(A) (If(.2. (4.15) f=1 for some set of real numbers Cagy that are known as the structure constants of the Lie group/algebra. n then the commutator of any pair of these vector fields must be a linear combination of them.
Recall that an integral curve t + a X ( t ) a vector field X is said to be ‘complete’ (see Section 3.. (4.2)G G + Tglg2G x which will be analysed with the aid of the discussion in Section 2.i)* L t R”4A’I (4.RA’] = f p ’ A 1 [RA.2. (4. a preliminary result is needed concerning the ‘infinitesimal’ form of the product group law on G.g.” = (Ad.G%Tglg2G x From Eq.G @ Tg2G through which the induced map p* factorises a (see Eq.2.17) (4.2.)G x G onto the space T. To prove this.21) fib.22) .2.18) (4. (2.3.3. Thus let p : G x G + G be the group product. It can be shown that (4 (b) (c) [RA.. However.74)) s x T~... even if G is noncompact..2 The completeness of a leftinvariant vector field The next result is of central importance in the theory of Lie groups and their associated Lie algebras. = p o x . the key result in Lie group theory is that every leftinvariant vector field on a Lie group G is complete.5.75).19) where the adjoint map Ad. so that p(gl.2.3. (2. : G + G is defined for each g E G by Adg(g’) := 9g’gl. every vector field on a compact manifold is complete.P ) = (P 0 ig2)*a + (P 0 jg1 >*P (4.2.. LA’]= 0 R.162 CHAPTER 4. In particular. LIE GROUPS with the ‘inverse’ A R := Re (= rgl*(R.2. 9 2 ) := glg2..)for any g E G). As remarked in Section 3.20) 4.2.. We wish to consider the induced tangent space map p* : T(gl.2) if it of is defined (or can be extended to a curve that is defined) for all values of the parameter t.~GG&TglG@Tg. where p . but there are plenty of examples of incomplete vector fields on a noncompact space. there is a map from T(.
THE LIE ALGEBRA OF A LIE GROUP 163 p o i = rgz and p o jgl I. + . Let t + a X ( t )be an integral curve of X through e E G that is defined for all I1 < E for some E > 0. (4.p) E T. This implies that $z(q5t(g)) = #+tZ ( 9 ) and so oX(tl t 2 ) = o X ( t 1 ) a X ( t 2 ) (4. then X is complete. .2.23) Hence Eq.2. This for all tl. This is the result that will be needed in the proof of the following main theorem.4.3 If X is a leftinvariant vector field o n a Lie group G.27) It follows that if 4 f ( g ) + t 2 I < E .then t + 4f is a local flow (see Section 3.2. . = (4. + 1..2) for the vector field X ..G @ T..25) and hence which shows that t point g E G. It21 < E and Itl equation can be read as saying that the map t + o X ( t )is a local homomorphism from the additive group IR of the real line into the Lie group G.2.22) becomes for all ( a .2. Theorem 4. o o X ( t ) an integral curve of X through the is := g o X ( t ) . Thus t (4. Proof 1.G.t2 such that Itl/ < E .
= I.2..28) by t ++ ax@). is necessary to show that for I1 < 2 ~ . (4. and hence to the whole real line. factorise the expression (4. (4.31) can be written as (using also the fact that the vector field X is leftinvariant): (7.164 CHAPTER 4. for all g E G) and hence the first term on the right hand side of Eq. Thus the crucial task is to show that the curve defined in (4. 3.32) .x(s/2)*(x. LIE GROUPS 2. e.3.x(s/2)) 2 1 +&7x(s/2)*(xgx(s/2))} (4. This suggests a means whereby the integral curves on G can be extended consistently using the group structure.24)has been used.X(s/Z))*Xe.2. (4. (4.2.28) which is consistent with the original definition if It1 < E since then. if I1 < 2 ~ . Temporarily denoting the curve defined by the right hand side of Eq. an integral curve of x . o r . the left and right translations of a group commute (2. (4.{~.2. and A : G + G x G is the usual diagonal map A(g) = ( g . X ( ~ )To do this.2. (4.2.. ax = p o A o ux o H where H : IR + IR is defined by H ( t ) := $. (2. ol.x(f)) (4. However. = .79).2.2.29) Clearly this procedure can be iterated to I1 < n for any positive t E integer n. g ) . it s 8 : = X .31) where Eq.2.2. r.21) has been used. from Eq.28) is. Thus ( ) i where the factorisation from Eq.28) as .X(S/2) 0 k+(s/2))*Xe = (krX(42) 0 T. t define 2 a X ( t ) := (. in fact. Namely. Then.
(4. / ~ ) ) Z . X can be extended to all values o f t E R. The exponential map exp : TeG + G is defined by exp A := exp t A It=l. A = atA(&)o. (4.31) reads (4.2. ~ ( . of the leftinvariant vector field LA (with aLA(0) e) that is defined = for all t E R by virtue of the above theorem is written as t + exptA where A E TeG.27).2. QED 4. and hence Thus the right hand side of Eq. and hence Eq.2. and hence the leftinvariant vector field X is complete.2.2. and (4.28) is an t 1 extension of the original integral curve ax from the interval I < E to I1 < 2 ~ Iterating the procedure then shows that the integral curve of t .4.2.31). (4. But the second term on the right hand side of this equation is clearly the same.2. this is the first term on the right hand side of Eq.2. 2. Definition 4.34) which is precisely the statement that ex defined by Eq.4 1. (4. X ( . (4.2.2. / 2 ) ~ X ~ x = / 2 ) (~ X ( . The unique integral curve t + aLA(t).3 The exponential map The theorem just proved leads naturally to the important idea of the 'exponential' map.35) (4. THE LIE ALGEBRA OF A LIE GROUP But.32) is l .36) . T ~ x ( o ax = l u x ( t ) o ~ ) 165 ax. from Eq.
On the other hand. (4. If G is not compact. for all tl. and (ii) both it and its inverse (defined on the image of V in G under exp) are smooth. LIE GROUPS 3. t 2 E IR. The following proposition shows that the converse is also true: namely. Of necessity it is then not injective as else it would establish a diffeomorphism between the noncompact vector space TeG and the compact group G. ‘For example. whose Lie algebra is simply IR. then the exponential map may not be surjective (there are some famous simple examples6 of this phenomenon). every oneparameter subgroup of G is of the form t t)exp t A for some A E TeG E L(G). it is now possible for the map to be globally injective. It can be shown5 that the exponential map is a local difleomorphism from the tangent space TeG at the unit element e E G into G. so that values of T E R that differ by an integral multiple of 27r are mapped to the same point in U(1). see also Helgason (1962). (4. 3.e. It follows from Eq.37) Comments 1.. It can be shown that exp T = ezr. The statement in Comment 1 still applies however and hence the map is onetoone in some neighbourhood of e E G (it is assumed here that G is connected).2. 2.2. Thus there is some open neighbourhood V of 0 E TeG such that (i) the exponential map restricted to V is bijective. see Helgason (1962).27) (and its equivalent for the fully extended integral curve) that t e.t A is a oneparameter subgroup exp of G. If G is a compact Lie group then the exponential map is a surjective function from TeG onto G. A oneparameter subgroup of a Lie group G is a smooth homomorphism x from the additive group of the real line IR into G. i. Thus there is a onetoone association between oneparameter subgroups of the Lie group G and its Lie 5A classic text for this type of result is Chevalley (1946). .166 CHAPTER 4 . A good example is the case G = U(1).
QED Corollary Let f : G + H be a smooth homomorphism between two Lie groups G and H whose exponential maps are denoted expG : TeG + G and exp. and hence t t+ X(t) is an integral curve for LA 6 L(G). THE LIE ALGEBRA OF A LIE GROUP 167 algebra. (4. T e H .) o x for all s E IR. Theorem 4. for all t E R.37) implies that Hence x o I.4 If x : R + G is a oneparameter subgroup of G then.2.2. The neighbourhood of e E G onto which exp : TeG + G maps diffeomorphically is ‘filled’ with the images of these subgroup maps.4. Then we have the commutative diagram TeG f .2. It follows from the theorem that ~ ( t = expH t B where B := x* ) 0 (4. : TeH + H respectively.41) . ~ ( t = exp tA where A := x* ) 0 (i) = Proof The relation (4. lX(.39) so that x is a oneparameter subgroup of H . The result follows from the uniqueness of such curves.2.
2.* maps TeG to TeG. for the Lie group SU(3). the adjoint representation is eight dimensional.(exp B ) = g exp B 9l. (2. Note that its kernel is the centre of G. (4. B = !*(A). proves the result in Eq. (4.B = g expB9l for all B E TeG..41) reads which.2. (4. Then. and the kernel of the action is Z3. LIE GROUPS Let k E C”(H).A)(k).40)..2.168 CHAPTER 4.2. Therefore.45) QED The map g t) Ad. (4. hence Eq. from the corollary above. the G is a homomorphism of the Lie group onto mapping Ad. gives a representation of G on TeG E L ( G ) known as the adjoint representation. But L t = A and hence B ( k ) = A(k o f ) = (f. .. QED Corollary If Ad. However. < exp Ad. : G itself. (kox) = d k(X(t))/ dt t=O = Lt(k0 f) = kof(expGtA) dt d (4. and hence Ad. for t = 1.39).(e) = e.. B = Ad. (4.42) where the last step follows because t H expGtA is an integral curve for the leftinvariant vector field LA. for each g E G . from Eq. This is the representation that is used in the famous ‘eightfold way’ classification of the elementary particles.44) is well defined.3.2. Thus Eq. For example.2. ie.44) Proof We note that Ad.(g’) := 9g’gl for each g E G then expAd.
.2. 2 .50) .R .n. (4.2.4. R ) is a matrix and hence we can consider the curve t ++ etA in G L + ( n . R ) . In particular. (4.47) and (LAxzj) = . furthermore.49) M ( n .2. . R ) of the general linear group G L ( n . R ) of all real n x n matrices: hence the tangent space at any point in G (and especially at e E G) can be identified in a natural way with the vector space M ( n .46) ) where g E GL+(n. j = 1 . But the tangent vector to this curve at t = 0 is obviously the matrix A and.IR) and i . . THE LIE ALGEBRA OF A LIE GROUP 169 4. . Let A E TeG M ( n .2.2.R).(x”(g exptA)) t=O 9 dt d . Eq.47) n = C g i k A k j= (gA)i’ k=l (4. . R ) where etA refers to the usual exponential of a matrix. Hence etA = exptA for all t E R and A E TeG can be rewritten using (4. This group appears as an open subset of the linear space M ( n .2.2. A natural system of coordinates on GL+(n.R)valid in some neighbourhood of the unit matrix 1are the ‘matrix elements’ defined by . R). x”(g) := gaj (4.48) However. it is clear that it defines a oneparameter subgroup of GL+(n.4 The Lie algebra of GL(n.R). Then the coordinate representation of the leftinvariant vector field associated with A is (4.IR) An important special example of leftinvariant vector fields is when the Lie group concerned is the connected component G L + ( n . R ) . A E M ( n . which can therefore in turn be associated with the Lie algebra of GL+(n.2.
Two of the most important applications for leftinvariant forms . The representation (4.52) 2.2.2. 2 M ( n .2. 3.54) (Eij)kl := &kJjl and the associated leftinvariant vector fields are (4. can be R) written in coordinate form as (4.A] is the usual matrix commutator: hence the Lie algebra structure induced on T. Similarly.170 CHAPTER 4.51) Comments 1.LA] = LP’A (4. the rightinvariant vector fields on GL+(n.3.R) is just the commutaR) tor of the matrices.2.1 The basic definitions The concept of leftinvariance for vector fields on a Lie group G has a natural analogue with respect to the dual structure of differential forms.51) gives [LA’. R ) is the set of matrices as Eij defined (4. GL+(n.2.53) where [A’.55) 4.3 LeftInvariant Forms 4.2. LIE GROUPS so that the leftinvariant vector field L A has the localcoordinate rep (4. A natural basis for M ( n .
18) follows Z.3.4. A rightinvariant differential form is defined in an analogous way.1 that there is an isomorphism i : + L ( G ) in which i ( A )= LA.G 9 (4. T.(dw) = d(Ziw). From Eq. LEFTINVARIANT FORMS 171 are (i) the development of a groupinvariant DeRham cohomology for Lie groups. Similarly. However.2) for all g' E G.3. and (ii) the theory of connections in fibre bundles. and hence i f w is leftinvariant so is its exterior derivative dw."G E L*(G)is called the dual Lie algebra of G. The latter topic has been much discussed by theoretical physicists in recent years in relation t o the general mathematical framework for YangMills theories. this nomenclature is potentially confusing in so far as L*(G) . z. (4.(wg') = Wglg' (4.G 2. It was shown in Theorem 4. In this context. Comments 1. T.3. particularly in investigations of anomalies in the quantisation of such systems. there is an isomorphism between Te*Gand the set L*(G)of leftinvariant oneforms on G that associates to each d E T.5 An nform w E A"(G) on a Lie group is leftinvariant if. for all 9 E G. Definition 4. This is one of the starting points for the development of a Ginvariant version of the DeRham differential form cohomology.4.'G the leftinvariant oneform Ad on G defined b Y Ad := Zgi*(d) E T.3) for all g E G. (3.3. equivalently. The discussion that follows of invariant differential forms is not very comprehensive and is partly concerned with developing the minimal set of tools necessary t o set up the theory of connections in bundles.1) 1gw = w or.
. . .4.G and d as an 0 element of the algebraic dual vector space TlG. En}. of leftinvariant vector fields: (4.15) we introduced the idea of structure constants with n respect to a particular basis { E l . .3) stand in a dual relationship to the leftinvariant vector fields LA according to the equation (Ad.5) for a basis of leftinvariant oneforms. (4. (4. This intuition can be substantiated by examining the twoforms dW* via their effect on a pair of the basis elements for L(G) using Eq.E2. whereas the commutator of two vector fields is itself a vector field. (4. = dim G . L A ) . w 2 . 4. However.3.3.3.3. .172 CHAPTER 4. .6) butin seeking t o derive an analogue of Eq. (4.5)it should be noted that.4) where ( d . (3. 3.5) A dual basis { w ' . A ) refers to the pairing between A E T. A ) for all g E G (4. a twoform can also be reached from a oneform by exterior differentiation. the natural combination of a pair of oneforms (with a wedgeproduct) leads not to a oneform but to a twoform.3. LIE GROUPS is not itself a Lie algebrait is only the linear dual of the underlying vector space of the Lie algebra L(G). which suggests that both this and the operation of taking a wedgeproduct may be involved in the analogue of Eq.w n } for L*(G) can be defined by (w". .2 The CartanMaurer form In Eq. = (d.3.Ep) := 6 . (4.3.15): . The leftinvariant oneforms defined in Eq.2. .
if ( Z.3.4.) = w6 8 w'(Ep. (4.G. (4.3. . E. LEFTINVARIANT FORMS 173 But (wa. E.3.7) gives dwa(Ep. (4. E.G it is also possible (and perhaps potentially less confusing) to regard the CartanMaurer form a s taking its values . (4. 2.10) gives W ' A w'(Ep.3.W' 8 w'(Er. (4.the expression Eq.) . On the leftinvariant vector fields L A .A)(g')= L. Definition 4. the basic definition of the wedge product in Eq.3. (4.3.". The CartanMaurer form is leftinvariant.12) 3.10) for the exterior derivative of a leftinvariant oneform. Comments 1.3.) = b.) = Cora. L. which is a constant and is hence annihilated by the vector field Ep.6 The CartanMaurer form E is the L(G)valued oneform on G that associates with any v E T. Ep) = $8. v ) denotes this leftinvariant vector field then ( Z .11) for all v E T.3.11) becomes (E.3.3. v ) ( g / ) := Z9'*(Z9'*v) (4.q. (4.5) in the last term of Eq.4. Specifically. (3. Thus using Eq. Since L(G) T.E.8) to give the famous CartanMaurer equation (4.8) However.G the leftinvariant vector field on G whose value at g E G is precisely the given tangent vector U.9) which can be combined with Eq.& 3 ).
174 CHAPTER 4 .3. (4.13) and the explicit expression Eq. R ) . Then R*E is a L(G)valued oneform on M .Lf) = A (4. a matrixvalued) differential form on the group. R could be thought of as a ‘gaugefunction’ associated with a YangMills theory using the Lie group G. rather than in L ( G ) .16) which is often written rather symbolically as R*E= RldR.3.2. when G = GL(n. Let R : M + G where M is some differentiable mdimensional manifold: for example.13) which. When G is a group of matrices.14) can be used t o write the components of R*E with respect to some local coordinate system on G as for all p E M in the coordinate chart..R)valued (ie.R).it follows that 9 . (4.3.=ij = C( 9 k=l n 1 ) i k (d z k j >g .3.3. (4.17) . E can be viewed as an M ( n . serves to define E precisely.51) for LA in the case of G L ( n . (4. With this interpretation Eq. if M is spacetime or physical space.3. Eq.14) 4. (4. In particular. (4.3. in fact. From Eq. LIE GROUPS in TeG.12) becomes (E. R) with TeG M ( n . Hence (4.
(rg+(E). the CartanMaurer form transforms as rg.3. or some set of references frames.4. if there is a homomorphism g ++ T~ from G into the group Perm(M) of bijections of M . Definition 4. this could be spacetime. or space. A linear representation of a group G is a special example of a Gaction in which the set on which the group acts is a vector space. for any v E TgtG. and with the action respecting the linear structure.7 A group G acts on a set M on the left. . v)). TRANSFORMATION GROUPS 175 Exercises Show that under the action g H rg of right translation of G onto itself.4. Thus the idea of transformation groups is of fundamental importance in theoretical physics. 4.( E) = Adgi*(E) (4. Thus Comments 1. or something somewhat more exotic.4 4.4.18) in the sense that. or has a left action on M.v) the element of TeG is given by Adgl * (( E.1 Transformation Groups The basic definitions In almost all applications of group theory in theoretical physics the groups arise as groups of transformations of some space of physical significance. for example. This motivates the following sequence of definitions.
but it is complicated since these groups are infinitedimensional and hence necessarily involve the nontrivial subtleties of infinitedimensional topology and differential geometry. If M is a topological space (resp.92 E G. g2 E G. Such a step is not impossible. (4.4.4.6) Writing the image point S g ( p ) in this case as p g . o 6. (4.1) and (4.4. differentiable) that the maps g t)yg or g with respect to this new structure.4) (4 3.gz E G. . and the group of gauge transformations in YangMills theory.4. The image point y. which uses the same 71mportant exceptions are the group of diffeomorphisms of space. or spacetime. diffeomorphisms) of M. This is an antihomomorphism g t) S from G into . The resolution of this problem is contained in the definition that follows.4. g2(glP) = (g2gl)P for all 91.5) S . especially since most of the Lie groups G with which theoretical physicists are concerned are themselves7 finitedimensional.4.4. However.5) and (4.3) (4. be continuous (resp.6) become (i) p e = p for all p E M.(p) =p for all p E M. This seems an unnecessary complication.176 CHAPTER 4.4. in which case equations (4. differential) structure on the group of homeomorphisms (resp.2) become (i) e p = p for all p E M. (4. Lie) group then it is natural to try to put a topological (resp. LIE GROUPS 2.. There is an analogous definition of a right action of a group G on a set M. if G is a topological (resp. Thus (i) (ii) S. Perm(M).4. (4.8) (4 4.4. = bylg2 for all 91. equations (4.7) (4.(p) is usually written as simply g p . diffeomorphisms) of M and then to require 6. differentiable manifold) then usually it is appropriate to require that the bijections involved are restricted to the subgroup of Perm(M) of homeomorphisms (resp. (Pg1)gz = P(glg2) for all g1.
2. an equivalent definition for a right action of a Lie group on a differentiable manifold. In the present case. As remarked earlier. this suggests we should study the map (4. This is required to be smooth in the sense above.p)G M + TSpM. a special case of a left action is a linear representation of a Lie group on a finitedimensional vector space. 2.4. in which case the basic homomorphism conditions in equations (4.1) and (4. it is always interesting to look at the associated pushforward operation on the tangent spaces.11) and the 0 theory of flows of vector fields. 4.4. M into Comments 1.8 A left action of a Lie group G on a differentiable manifold M is a homomorphism g e yg from G into the group of diffeomorphisms Diff(M) of M with the property that the map : G x M + M defined by is a smooth map from the differentiable manifold G x the manifold M . x and we shall essefitially do so in Section 4.4.2) become 3.5. : T(g. TRANSFORMATION GRO UPS 177 type of trick discussed earlier in the context of Eq. Note that. Whenever one encounters a map between two manifolds. (3.4. in practice. Definition 4. . of course. There is.12) r. The definition of a left action is often given directly in terms of this map : G x M + M.4.
16) 6. a map f : M + M‘ is said to be equivariant with respect to the group actions g ts 7. A more sophisticated version of this idea allows for two Lie groups G and G that act on manifolds M and M’ respectively. . =f M’ lr’ (4.14) so that 7L. for all g E G. and vice versa. G ’ x M ’ (4. and ’ with a homomorphism p : G + GI. and g t)7. Namely. (4. on M and M’ respectively if the following diagram commutes M f.4. There is a welldefined concept of a structurepreserving map between a pair of group actions of a Lie group G on manifolds M and MI. given a left action g e 7ga right action can be defined by 8. pequivariant).4.13) 5.17) M commutes.15) is equivalent to the statement 9f ( P ) = f(SP) for all g E G and p E M. M’ (4.4. := rg1 (4. (4.4. a map f : M + M‘ is said to be equivariant (or. Note that Eq. if the diagram G x M pxf.8 (4.f = f 07. In this case. Specifically.178 CHAPTER 4 .4.15) for all g E G. so that lr f. LIE GROUPS there is no need to consider leftactions and rightactions separately since they are in onetoone correspondence.4.18) rlOpf x or ‘Note that the idea of equivariance makes sense for a general group acting on a set: it is not necessary to use only Lie groups acting on manifolds.4. perhaps better.
. 3. TRANSFORMATION GROUPS or. 2. 179 p ( g ) f ( p )= f k P ) for all g E G and p E M .4.19) An important physical application of this idea arises in the context of the theory of spinor fields on a general manifold M . or (ii) there is a unique g E G such that p = gq. (4. if M is physical space. (4. then the groups G and G’ are SL(2. either (i) there is no g E G such that p = g q .4. 0 4. The kernel of a Gaction is the subgroup of G defined by K := { g E G I g p = p for all p The group action is effective if K = E M}. q E M can be ‘connected’ by an element of the group.4. i e .9 1. { g E G I g p = p } = { e } .4.4. l ) respectively. The Gaction is free if. Definition 4. This is complementary to the idea that in an effective action the whole of G can be ‘probed’ by its action on M . for all p . If M is spacetime. . this means that given any pair of points p . The Gaction is transitive if any pair of points p . q E M . equivalently. Thus in a free action every point in M is moved away from itself by every element of G with the exception of the unit element e only.2 Different types of group action We come now to a series of definitions concerned with various aspects of the action of a group G on a set M .20) {e}. In a transitive action the whole of M can be ‘probed’ by the Gaction. for all p E M .C ) and S O ( 3 . q E M there exists g E G such that p = gq. In particular. the groups are S U ( 2 ) and SO(3) respectively.
(4. The kernel of this action is the centre C(G) of G. For example. the Z3 centre is represented trivially and it is the quotient group S U ( 3 ) / Z 3 . If the Gaction on M is transitive then the stability group.180 CHAPTER 4 . LIE GROUPS 4. (4. in the adjoint action of S U ( 3 ) on itself. this is a closed subgroup of G for each p E M . It is a normal subgroup of G and the action of G on M passes naturally to an action on M of the quotient s group G/K that a effective. An example is afforded by the adjoint action of G on itself in which Ad. it is useful to define the orbit 0. of the Gaction through p E M to be the set of all points in M that can be reached from p (see Figure 4. or little group. Given that a generic Gaction is not transitive. The kernel measures the part of the group that is not involved at all in the Gaction on M. 5. Comments 1. Gp of the action at a point pEMis G.21) Figure 4.22) In the Lie case.1: The orbits of a group action on M . or isotropy group.4. := { q E M I 3 g E G with q = gp}.(g’) := gg’gl.1): 0.4. := {g E G I gp = p}.
2. Thus the eightdimensional representation of S U ( 3 ) (the 'octet' representation) is a faithful representationg of S U ( 3 ) / Z 3 . (4. i e . For example. irrespective of whether they are Lie groups. it suffices to show that there exists some point po E M with the property that. The same remarks apply to the induced linear representation Ad. TRANSFORMATION GROUPS 181 that acts effectively. (ii) The adjoint action Eq. in the case of Lie groups (with H a closed subgroup) it can be shown that there is a unique 'A linear representation g + L ( g ) E Aut(V) of a group G on a vector space V is said to be faithful if ker(L) = e. 3. and then from Po to 4 . .4.4.representation of SU(3) is a faithful representation of S U ( 3 ) itself. q E M one can go from p to q by going first from p to po (using the inverse of the group element that transforms po to p ) . then given two points p . any faithful linear representation of G is effective but it is never free since if 0 is the null vector in the representation ' vector space then. A free action is effective but the converse is not necessarily true. g 6 = 6 for all g E G.+ on the vector space T. 2. Ad. This will play a central role in our discussion later of principal fibre bundles. (4.23) M This action is transitive for any pair of groups G and H .4. On the other hand. 4.G. A very important example of a transitive Gaction is when = G / H where H is a closed subgroup of G. To show that an action of a group G on a set M is transitive. for any other p E M I there exists (at least one) g E G such that p = gpo.(e) = e for all g E G. if this is the case.20) of a group G on itself is not free since e E G is left fixed by every g E G. However. Note that: (i) A wellknown example of a free action of any group G is left (or right) translation on itself. The left action of g E G is defined by rg(dH) := ( 9 d ) H . . by linearity. the threedimensional ('quark'). Indeed.
This leads to an equivalence relation on M in which two points are defined to be equivalent if and only if they lie in the same orbit. If M is a topological space.45)). Figure 4. 5. the orbit space does have a decent topological/differential structure. non Note that the orbit space of the right action of a subgroup H of a group G on G is just the space of left cosets G / H .24) "A classic reference is Helgason (1962). The kernel K of a group action is related to the stability groups b Y K= n G. namely the unique analytic structure mentioned above.the orbits) by this equivalence relation. space carries a natural identification topology (see Eq. . 6. but Hausdorff).1).4.. Note that: (i) The action of G on G / H is effective if and only if H contains no normal subgroup of G. The set of action on M then the orbit (1. LIE GROUPS analytic manifold" structure on the coset space G / H such that G is a Lie transformation group on this space.4. with H a closed subgroup of G . It is never free since h(e H) = eH for all h E H .182 CHAPTER 4. (ii) A linear representation is never transitive since the null vector 6 cannot be taken to any other element in the vector space by the action of any g E G. An important physical example is the action of the gauge group on the space of YangMills fields. In the case of a Lie group. equivalence classes is called the orbit space of the Gand is denoted M / G . M is partitioned into disjoint subsets (ie. it is frequently rather unpleasant (for example. As usual. It is clear that the orbits through any pair of points in M are either equal to each other or are disjoint (cf. PEM (4..
25) so that the stability groups along an orbit are always conjugate (see Figure 4.4.4.4. If p and q lie on the same orbit of a Gaction on M then there is some g E G such that p = gq.3 The main theorem for transitive group act ions Next is a theorem of considerable significance. . The word 'effectively' covers up a fair number of sins and it is best first to state the result from a purely group/settheoretic perspective with no reference to topology or differential structure. TRANSFORMATION GROUPS 183 7. The theorem of interest asserts the converse: if M is any space on which G acts transitively then it is effectively of the form G / H for some H c G. As noted above.2: Stability groups along an orbit are conjugate. It follows that Gp = gG.g' (4. especially for the use of Lie groups in theoretical physics.2). 4.4. the action of G on the coset space G / H is transitive. 0 Figure 4.
. In the present case. there is a bajection j p : GIG.184 CHAPTER 4.26) is a diffeomorphism. Since the Gaction is transitive.defined M i bY j p : G / G p + M gGp ++ QP where Gp is the stability group at the point p . This is precisely the condition that glGp = g2Gp ( i e . (4. Suppose jp(glGp)= jp(g2Gp). g1 and g2 belong to same right Gporbit) and so j p is onetoone.26) is indeed independent of the representative chosen from the coset. Hence j p is surjective. (4. Thus the definition in Eq.26) 2. Then g l p = g2p and hence gc1g2p = p .5 Let G be a group that acts transitively on a set M . LIE GROUPS Theorem 4. When G is a Lie group that acts smoothly and transitively on a differentiable manifold M . Helgason 1962) (i) The stability group Gp is a closed subgroup of G . . QED Comments 1. every point in M can be written as g p for some g E G. and hence GIGp can be equipped with the unique analytic manifold structure referred to earlier. a key question is whether or not the bijection j p in Eq.4. which means that gT1g2 belongs to the stability group G.26) is welldefined.. if g1 and g2 are in the same coset then g1 = g2h for some h E Gp.4. Proof 1.4.e. it is necessary t o show that the definition is independent of the particular representative that is selected. The two major results in this direction are: (Chevalley 1946.4. (4. Then. 3. . Such a step is mandatory whenas is the case h e r e a function is defined on a space of equivalence classes in terms of a particular representative in each class: i. for each p E M . (4. . The first step is to ensure that the map in Eq.and hence g1p = g2hp = g2p.
6 T h e nsphere S" is diffeomorphic to the coset space O(n 1.O. and if G is a locally compact Lie group. then the bijection j p is a diffeomorphism from G/G. 1 . . onto M . TRANSFORMATION GROUPS 185 (ii) If M is locally compact and connected. The critical question in the Lie group case is to decide when the spaces G /G pcan be regarded as embedded submanifolds of M . . O . O ) . . If the Gaction on M is not transitive then M decomposes into a disjoint union of orbits. (O. . + Proof 1. . . It follows that each orbit 0. This is a complicated question and for further information on these topological matters should be sought in the technical literature. 0 4. let (v'. O ) . c M . Then the linear + . + To this end. . 2 . .d)= Cy2: viwi denote the usual inner product on the real vector space Rn+' where {vi I i = 1 . According to the main theorem 4. which first requires finding out when the isotropy groups of the different orbits are closed.4. it is necessary to find a transitive action of the group O ( n 1. . R)subgroup.R ) / O ( n . is in bijective correspondence with the coset G / Gpwhere Gpis the stability group for p E 0. on each of which G does act transitively and with a specific stability group. .5.4. . . IR). . .4. 2. . . ( 0 . . .4 Some important transitive actions We shall now give several important examples of familiar manifolds that admit a transitive group of transformations by a Lie group G and are hence diffeomorphic to a coset space G / H for some subgroup H of G. R)on the nsphere and then to show that the stability group at any conveniently chosen 'fiducial' point is a O(n. Theorem 4.1)}. . .n 1) are the components of the vector v'with respect to the usual orthonormal basis set { ( l .
has + + .IR) on the nsphere. in particular. .186 CHAPTER 4. + Then v' satisfies the matrix equation v' = Avo. The first task is to show that this action is transitive. Since v' is any unit vector in the nsphere. of the form gHg' where g E O ( n 1. .IR).} is such that the matrix A satisfies the equation AAt = A t A = I and hence A E I O ( n 1.IR) on this vector space leaves invariant the inner product and hence. .27) 2. e. Let 3 := (1. whose component form is However. and let v' be any other unitlength vector. (4.30) Hence the isotropy group is this particular" 1. the set of unitlength vectors + S" := { a € R"+l I (G. . .. (4. the choice of the vectors {ei.R . . e.4.IR) is transitive. e..0. Choose a set of orthonormal vectors {ei.. as there are many subgroups of O ( n 1.} such that the complete set {G.e. 0) E S" c IR"+'. . e:.R)) this property.4. . .R)that are isomorphic to the subgroup of matrices satisfying Eq. 0 X (4. . indeed any subgroup that is conjugate to this subgroup H (ie.4. . .lR) subgroup of QED "I say 'particular' subgroup. e:. . ) + O(n.29) 0 where X is any n x n matrix that satisfies XXt O(n =XtX = (4.30). This is the desired action of O ( n + 1.')= l}.4. this shows that the action of O ( n 1.. . + + 3. The isotropy group at the particular vector wo is clearly all matrices of the form + 1 0 0 0 0 . . LIE GROUPS action of O ( n 1. ei.} is an orthonormal basis set for IR"+l. ..
4.4. . R).35) to belong t o SU(2). It can also be obtained directly [exercise!] from a study of the implications of the defining relations for a matrix ( :) ( 8: ) ( y ) (z :) = and det ( :) = 1 (4. (4.. then the matrix A in Eq. e. The stability group is a U ( n ) subgroup.34) This resultthat the group space of SU(2) is diffeomorphic t o a 3sphereis important in a number of different ways. If the vectors {e. (4. . .4.4. . . IR) subgroup of O ( n 1.31) Cp='. (4.} C ei..4.28)satisfies det A = 1 and hence belongs t o the S O ( n + l . .4.32) into itself andas in the case abovethis action can be shown t o be transitive. which gives the result s2n+1( n + l ) / U ( n )= SU(n + l)/SU(n). TRANSFORMATION GRO UPS 187 Comments 1. .4.IR). is the same as that arising from the usual basis.' ufwi A similar discussion applies to the invariance of (C. since SU(1) = { e } . . 2.R)/SO(n.v)! z Iz2 = 1} u1 (4. It then follows that + S" z S O ( n + 1. e.} are chosen such that the orientation on IRn+' induced by the orthonormal basis set { . e:. e. zU A particular example is n that = (4. it follows % s3 U(2)/U(l) SU(2).w') = under the linear action of U ( n + 1) on C"". .4. This action will map the (271 + 1)sphere S2n+1 {u E a="+' = I n+l (u.33) 1 when.
4.R)on M ( n .188 CHAPTER 4 . R) where RP" denotes the real projective space of all lines in (4. R). This clearly maps complex lines into complex lines and hence passes to an action on the complex projective space CPn (which is defined to be the space of all such lines in C"+l).38) Rn+'. (4. R)/O(n. and the little group is a U ( n ) subgroup. The first step is to show that the subspace S. since CP' S2 (the complex Riemann sphere) proves the famous result A particular example is CP' E E s22 S U ( 2 ) / U ( 1 ) . Theorem 4. LIE GROUPS 3.40) where g E G L + ( n . R)by T J A ) := g A g t (4. is . denote the set of real.37) RP" E S O ( n + 1. Proof G L + ( n .R ) / S O ( n .4.4. Unlike the original linear action.39) 1. R )and A E M ( n .R) mapped into itself by this action. (4. n x n symmetric matrices. Then S. Another important example of a transitive group action can be obtained by considering the linear action of SU(n 1 ) on C"". 0 The next result is significant in the context of the grouptheoretical structure of Riemannian geometry. this new SU(n 1 ) action is transitive. R ) . Define a left action of GL+(n.4.7 Let S. positivedefinite. c M ( n . Hence: + + S U ( 2 ) / U ( l ) which.
where A is regarded as a linear whose inner product (6. which is thus the 'fiducial' matrix for this action: in particular. so is gAgt for all g E GL+(n. 3.Av') > 0 for all v' # 6.41) is of the form Eq. Then (v'. c M ( n .40). . But this is precisely the R (g g subgroup SO(n.4.40) and hence since OD: E GL+(n. is indeed mapped into itself under R) the GL+(n. (4. 2. Furthermore. . This shows that every S E S. (gAgt)G) (gtv'. and hence (g'v'. ) such that ~ ~ = 1l ) t = 1. since S is positive definite it follows that di > 0 for all i = 1 . . (4.4.R) and a diagonal matrix D = diag(d1.lR)action on the subspace S. n.4.) such that S = ODOt. .4. This result plays a central role in the fibre bundle approach to the theory of Riemannian metrics on a manibld where. . . since a symmetric matrix S is diagonalisable.'> operator on the real vector space R" is defined as usual. a Riemannian metric on a manifold M assigns an element of S. The isotropy group at the fiducial matrix lt is the set of all g E GL+(n. Agtv') since the transpose of a matrix = is defined such that (v'. there exists some 0 E SO(n.'.R)is invertible. . (ii) To say that A is positive definite means that (. This completes the proof of the theorem. action defined in Eq. (4. .40). to each point in M . dz. so that the image of a symmetric matrix is symmetric. .4. But g E GL+(n. These two statements show that S. TRANSFORMATION GROUPS 189 (i) If A = At then (gAgt)t= (gt)tAtgt= gAgt. To see that the action is transitive note that.41) But Eq. (4.R)it follows that Eq. . 2 .R). w').AT?) > 0 for all v' E R"such that v' # 6.4.41) is indeed a transformation of the type in Eq.d. BG) = (Btv'.4. (4.4. the GL+(n. I R ) is transitive. Agt$ > 0 for all v' # 6 if and only if (5. and hence the square root of D can be taken to give s = (OD:) 1( 0 D i ) t . can be obtained from the unit matrix 1 E S. Hence if A is positive definite. QED Comments 1.R). (4. roughly speaking.
By these means a map is obtained from the Figure 4. 4. The basic idea is as follows (see Figure 4. 1 ) is relevant in the mathematics of 0 Lorentzian geometries in a fourdimensional spacetime. the properties of ‘infinitesimal transformations’ and their relation to the full.1 Infinitesimal Transformations The induced vector field The next topic is of considerable importance in any programme that uses Lie transformation groups: namely.5 4. the matrices in Sn. and hence a vector field on M that is tangent to this family everywhere. In particular. ..1 GL+(4. l ) where 2. If a Lie group G acts as a group of transformations on a differentiable manifold M then each oneparameter subgroup of G will produce a manifoldfilling family of curves. ). the relation S3.1 .~ GL+(n..l have signature (+.5. LIE GROUPS IR)/SO(n . . +. R ) / S O ( 3 . Lie algebra element associated with the oneparameter subgroup into . globallydefined group action.3).3: The orbits of a oneparameter subgroup of G.+.190 CHAPTER 4. There is a similar result Sn.
dt = 6exp tA .” A E TeG} =” 1 (4. through a point p E M of the Gaction is a smoothly embedded submanifold. and 6. INFINITESIMAL TRANSFORMATIONS 191 the infinitedimensional Lie algebra of all vector fields on M . this is a generalisation of the way in which the Lie algebra of G reflects the properties of G itself. Comments 1.4. A (4. is defined as (4.11) for a right action is a smooth map I? : M x G + M with (i) r ( p . tiable manifold M .g2 E G. (4.5) . but it is not a significant restriction and the results obtained all have exact analogues for a left action. Definition 4.10) and (4.4.glgz) for all p E M and gl.5. If the orbit 0. (4.5.2) 2. Then the vector field X A on M induced by the action of the oneparameter subgroup t + exp tA. then Tp(Op) {X. A E TeG..5.10 Let G be a Lie group that has a right action g + 6 on a differen.4.5.4) (ii) r(r(p. e) = p for all p E M .gi). The analogue of equations (4. In the discussion that follows a rightaction of G on M (rather than leftaction) is used.(p) has been abbreviated to pg. The reason for this choice will emerge later. The interesting question is the extent to which this linear map reflects the details of the action of the Lie group G on M : as will be shown below.5.5. the flow is ie. The induced vector field X A has a flow q5$(p) = pexptA = SeXpt~(p).gz) = r(p.1) where f E Cm(M).3) 3.
5.5.5.*L. M. .5. (4.12) which is a useful alternative definition of the induced field X A . (4. (4.5.5.5.7) for all g E G and p E M where r..15)) of M .12) is used in the next result that deals with the infinitesimal version of the equivariance of a group action of the Lie group G on a pair of manifolds M and M’ ( c j . for all p E M and g E G.11) that.(exptA) = f(pgexptA)/ d . 0 The expression Eq. X. to each p E M associate the map M .. LIE GROUPS +M by S. and hence Eq. 9 ) pg in terms of which Eq.(g/) := g’g.) (f)= A(f 4.A)(f) L.related for each p E M . . = x .. it follows from Eq.5.192 Then each g E G defines a map M CHAPTER 4. (4. (4.9) M . (4. O (M.) = O d f O M. = M.*L.8) (4. : G r.8)). for all p E M .10) t=O dt Thus. Eq.*A)(f = But Mp) = A(f O M .).L. for all f E C”(M).5.5.5) asserts the equivariance (see Eq.A(f 0 Mp) = P. : G + M defined by r M p ( g ):= F ( P . there is another way of using : M x G + M : namely. 0 1. for all g E G.11) which shows that the vector fields LA on G and X A on M are M.(p) := r(p. o r.5. since L t = A E TeG.5. = S 0 M . (4. 0 1.g ) =pg.”= Mp*(A) (4.6) However. as M . Note that. + G is the right translation A E Now.4. (4.8) implies (M. if LA is the leftinvariant vector field on G associated with TeG then. (4. (4. (4.
Thus. In other words..13) for all p E M and g E G .17) Similarly. the equivariance condition Eq. (4. Now let us consider a very special case of a right action of G on a manifold M : namely. i. QED. the leftinvariant vector field LA is induced by the right translation of G on itself. Proof From Eq.(A) = Lf = (4.e. INFINITESIMAL TRANSFORMATIONS 193 Theorem 4.13) implies that.18) . and let f : M + M’ be equivariant under this action. Then the vector fields X A and XIA are f related for all A E TeG.8 Let G be a Lie group that acts on the right on manifolds M and M’ with induced vector fields X A and XIA respectively. f(P9) =f( P h (4.(A)= Zg. the rightinvariant vector field RA.5. for all f E C”(G).5.Also.12) we have X t = M p * ( A ) . and hence which is precisely the condition that the vector fields X A and XIA be frelated. (4.5. is induced from the lefttranslation operation as (4.4.(g‘) = gg’ = lg(g’). for all g E G and p E M . (4. when M is G itself and dg = rg for all g E G. A E TeG. Then X.5.5.” Gg.5.16) since G.5.
5..5.Texp tA. (4. QED Comments 1. We know from Eq. Thus. (4.19) Proof In general.lexp tA.17). according to Eq. using .5. (4.20). Theorem 4.B ( t (4. (4. and if Y is any other vector field then (see Eq. =  lim t+O B B . of the adjoint action of G on itself. LIE GROUPS This way of looking at the vector fields LA and RA is useful in a variety of situations. Using this result and Eq.5.19) it can be deduced that exp t A exp B exp tA = exp (t[AB] O ( t 2 ) ) .(B) = g(exp B)g' for all g E G.2. + (4. tA lim tiO . on the Lie group G is 4 = TexptA. B E TeG with Lie bracket [AB].5.194 CHAPTER 4 . A E TeG. B ).44) that exp Ad. [AB]= [LA.5.2 1) which proves the result.20)) Now. (B) t  lim t+O Adexp tA. A good example is the following theorem which deals with the relation between the commutator of a pair of Lie algebra elements and the pushforward Ad. (4.9 Let A. ( B ) t Adex.LB]. if 4 is a flow for a vector field X on a manifold M .. (3.2.22) 0 . a flow for the leftinvariant vector field LA.Then. Eq.
the Lie algebra of the Lie group G is ‘represented’ by the vector fields on the manifold M on which G acts.. Theorem 4. rather technical looking.i*(X.G.2 The main result We can now prove what is undoubtedly one of the central results in the theory of ‘infinitesimal’ transformations and which explains why the subject is important in so many applications. theorem will be used in the proof of the main result in the next section. T h e n the map A t) X A .26) 4. o Ad.A..Ad. Then Eq.1 = oM.(A) P = = &. The theorem concerned asserts that the map that associates the vector field X A with A E T. (4. (4.G the induced vector field X A o n M .oAd.25) so that M.1MPg.5.5. f o r all g E G and A E T.24) = p(gg’g’) = MpgOrgi(g’) 6.5. (4. is a homomorphism of . = 6.24) implies that XAdg*(A)d.5..5.5.i*M.10 If the Lie group G acts o n a manifold M then.)* ( A ) .(g’) (4. which associates to each A E T.(A) But M...G is actually a homomorphism of Lie algebras: i e . Theorem 4. .). Proof We have Xfg*(”)= M.4.11 Let M be a manifold o n which a Lie group G has a right action.(g’) = ( M . INFINITESIMAL TRANSFORMATIONS 195 The next. o Ad.
(4. action of G on M then the result in Eq.e. (4.19). (4. using Eq. X dx tA. the linear map A t+ XAis continuous and hence the limtto can be moved inside the square brackets to give [xA. Eq. Proof It is clear from Eq.28) Hence.5.23). LIE GROUPS L ( G ) 2 TeG into the injinitedimensional Lie algebra of all vector fields on M . xB] = XitoAepA()B/ lm'(dxt.196 CHAPTER 4.5.27) for d l A.3)) a flow for XA is &.(B)B A ep = lim t+O t (4.27) would have become . so the key task is to prove Eq. Since (see Eq.5.5. rather than a right. (4.5.5.5. (4.X"] = X"W (4. i.29) where the last step follows from the linearity properties XA.27). (4.XA'= X(AA') and XA = XA. Comments QED 1. (4.20) can be used to write (4. B E TeG 2 L ( G ) . Since TeG is a finitedimensional vector space.12) that the map A c) XA is linear.5.. [X".(tA).B)t  XWI by virtue of Eq.5.5. If the same procedure had been followed throughout but with a left.
Then the oneparameter subgroup t e exptA acts trivially on M.” As a final remark on this whole issue of infinitesimal transformations one might wonder when the process discussed above can be reversed. That is. But if a manifold M and Lie group G are such as to admit a homomorphism from L(G) into the vector fields on M.5. SU(2) and SO(3) have the same Lie algebra and . Several properties of the Gaction on M are faithfully reflected in the homomorphism x : TeG + VFld(M). rather than a homomorphism. This means that p E M is left fixed by the oneparameter subgroup t H exptA. these being different global coverings of each other (for example. This is the reason why a right action has been used throughout this section.4. which is consistent with the effective nature of the action only if A = 0. INFINITESIMAL TRANSFORMATIONS 197 so that the linear map A e X A would then be an antihomomorphism from TeG E L(G) into the Lie algebra of vector fields on M . 2. when is it true that there is a global group action of G on M such that the given homomorphism is precisely the one that derives from this action? One obvious problem is that usually more than one Lie group has a given Lie algebra.A # 0. ” (ii) Suppose the action is free and that A E TeG is such that Xp” vanishes at some point p E M . which contradicts the free nature of the action unless A = 0. so that x ( A ) = 0. For example: (i) Suppose the action is effective and let A E Ker(X). Hence we conclude: “If G acts effectively on M then the map A t)X A is an isomorphism from TeG L(G) into the Lie algebra of vector fields on M . we have started with a group action of G on M and derived a homomorphism from the Lie algebra of G into the vector fields on M . Hence: “If G acts freely on M then the induced vector field X A is nowhere zero for all A E TeG. x(A) := X A .
” . LIE GROUPS there is a twotoone covering homomorphism 7r : S U ( 2 ) + S O ( 3 ) . It is thus sensible to start with the assumption that the group concerned is the unique simplyconnected Lie group with the given Lie algebra. It is also necessary to restrict the discussion to those cases where the vector fields in the range of the homomorphism x are complete since the fields induced by a group action always have this property. A famous result encompassing some of these features is Palais’ Theorem.1. see Eq. then there exists a unique Gaction on M such that x ( A ) = XAthe induced vector field for that action. “If x is a homomorphism from L(G) into the vector fields on M .8)).198 CHAPTER 4. (5. and if G is simply connected and M is compact.
Chapter 5 Fibre Bundles
5.1
5.1.1
Bundles in General
Introduction
In many situations in theoretical physics we encounter fields carrying indices. Some of these refer to tensorial properties with respect to the space on which the field is defined (often spacetime'); others are associated with internal structure on the space in which the field takes its values.
For example, consider the question of the precise mathematical structure that underlies the existence of an 'ntuple' of real scalar fields {$'(x), +'(x), . . . , @(x)} defined on a spacetime M . One possibility is that it stems from a map (b : M + V from M to a real, ndimensional vector space V. More precisely, if { le'2, . . . ,Zn} is Z, some basis set for V , the value +(x) E V of the field at 2 E M can
'The most common domain space for fields in modern physics is spacetime. For this reason I shall usually refer to the space on which a field is defined as 'spacetime', but it should be appreciated that this is not the only possibility. For example, in the canonical approach to a field theory the fields are defined on physical 3space. And in special relativity, it is natural to consider fields whose domain is the 'massshell hyperbola' p .p = m2 in momentum space.
199
200 be expanded a s
CHAPTER 5. FIBRE BUNDLES
q5(x) :=
c
n i=l
@(Z)Z,
(5.1.1)
where, for all z E M , the quantities @(z),i = 1 , 2 , .. . , n, are real numbers. The coefficient functions z ++ qY(z), i = 1 , 2 , .. . ,n, are then the 'fields with indices'. In many cases of physical importance, the vector space V arises as a representation space of some internal symmetry group. Vectorvalued fields are not the only possibility. Another impor: M + N is a map from spacetime (or tant example is when space) M to a manifold N that is not a vector space. For example, in the nonlinear amodel, the basic field takes its values in a homogeneous spaces G I H where H is a closed Lie subgroup of the Lie group G ; a s we saw in Chapter 2, such a space G I H is typically a nontrivial manifold. In cases such as this, we are to think of having a local coordinate chart ( U , $ ) on N so that $ : U c N + IR" with +(y) := (+'(y), +'(y), . . . ,$P(y)) E IR". The coefficients of our spacetime field q5 are now defined by
+
(5.1.2) which makes sense provided the image point 4(z) E n/ lies in the open set U c N . Note that vectorspace valued fields are actually a special case of this situation in which we use the (global) coordinate chart on V that is naturally associated with the basis set {Zl, Zz, . . . ,Zn} (ie., the coordinates of a vector v' E V are defined to be its components w z with respect to the basis set: v' = Cy=lwig). At first sight, the idea of a manifoldvalued field q5 : M + N might appear to exhaust the possibilities of what is meant by a scalar 'field'. However, it has become clear that, in fact, this is not the case, and a genuine generalisation must be considered. This is the situation in which the space in which a field takes its values varies from point to point in the space M on which it is defined. Thus rather than writing 4 : M + N we must think now of a family of target spaces N, labelled by the spacetime points z E M , and with +(z) E N, for each z E M ; We are thus lead naturally to consider a bundle of spaces, parameterised by points in spacetime, in which our field appears as a (crosssection'.
5.1. BUNDLES I GENERAL N
201
Note that we have already encountered the idea of a bundle in Section 2.3 where the tangent spaces TxM of a manifold M (parameterised by the points x E M ) were ‘glued’ together t o form the tangent bundle T M . In particular, recall that a vector field on M can be defined as a map X that associates to each point x E M a point X , in the tangent space T x M ;ie., it is a crosssection of the tangent bundle, and hence an example of a ‘generalised field’ of the type under consideration.
As we shall see later, tensor fields of all types appear naturally as crosssections of various bundles associated with a spacetime manifold. Thus fibre bundle theory plays a central role in the underlying mathematics of differential geometry, and h e n c e i n particularf general relativity. However, this is not the only use of fibre bundle theory in physics. Another, very important, example arises in YangMills theory in which the natural mathematical setting is a bundle of copies of the internal symmetry group G (for example, G might be S U ( 2 ) or S U ( 3 ) ) ,one at each point 5 in spacetime. The way in which these fibres twist around the spacetimetypically Euclideanised and compactifiedreflect topological properties of the gauge theory such as instanton number. A related application is the use of bundles in the context of KaluzaKlein theories and the associated concept of ‘dimensional reduction’.
Fibre bundle theory is an important part of pure mathematics too, especially algebraic topology. This is because the different ways in which the fibres of a bundle over a space M can,twist around are determined by global topological properties of M that are specified by certain cohomology groups (see later). Indeed, in one powerful approach to algebraic topology (Ktheory) the cohomology groups of a space M are essentially defined by the different bundles that can be built on it.
5.1.2
The definition of a bundle
Faced with a family of spaces F, parameterised by the elements z E M , it seems natural t o require that the dependence of Fx on the spacetime point z be continuous, or even differentiable, in some
202
CHAPTER 5. FIBRE BUNDLES
way. For example, for the concrete case of a 2sphere embedded in R3 it seems intuitively clear that the tangent spaces do indeed vary smoothly as we vary the point to which they are tangent, although it may not be obvious how this intuition should be made concrete in mathematical terms, especially if we do not wish to invoke the ‘ambient’ space R3 in which the 2sphere appears as a hypersurface. One possibility might be to think of x + F, as a map from M to the collection 7 of all topological spaces. One could then require this map to be continuous with respect to some topology placed on 7, providedof c o u r s e t h a t a ‘topologising’ of 7 is possible. However, if 7 could be given a topology it would become a topological space, and hence a member of itself, thereby encountering Bertrand Russell’s famous problem of barbers who shave themselves! But every set can be given at least one topology (for example, the discrete or indiscrete topology), and hence 7 is not a set but some more general class of objects.
UxEMF,
A more productive approach is to consider the collection E := of all the ‘fibres’ F,, x E M , and require this to be a topolog
ical space in its own right. This idea leads to the basic mathematical definition of a bundle’ :
Definition 5.1
1. A bundle is a triple ( E ,7 r , M ) where E and M are topological spaces and 7r : E + M is a continuous map. 2. The space E is called the bundle space, or total space, of the bundle; M is the base space of the bundle; the map T is the projection; and the inverse image 7r’({x}) is the fibre over
XEM.
The general idea of a bundle is sketched in Figure 5.1.
’Some of the best references on fibre bundles are also some of the oldest.
A classic workwhich makes heavy use of the idea of local bundle chartsis Steenrod (1951). A text that takes a more modern approachand which has
strongly influenced my own presentationis Husemoller (1966)
5.1. BUNDLES IN GENERAL
203
Figure 5.1: A diagrammatic representation of a general bundle.
204
Comments
CHAPTER 5. FIBRE BUNDLES
1. There is no loss in generality in supposing that n : E + M is surjectiveif it is not, simply consider the bundle over the image
n ( E )c M .
2. A C" bundle is as above but where E and M are C"manifolds, and with n a C"map. In what follows it is to be understood that any continuous map is also required to be C" if the spaces concerned are differentiable manifolds.
3. By common consent, the total space E is often called the 'bundle' even though, strictly speaking, this refers to the triple ( E ,n, M ) .
4. We often denote a bundle (ie., the triple) with a Greek letter like [ or 7 . In this case, the total and base spaces of a bundle [ are denoted E ( [ ) and M(E) respectively.
5 . The definition above of a bundle is very general. However, in all existing applications in physics (and for most uses in pure mathematics too) the bundles that arise have the special property that the fibres ~ ' ( { z } ) z E M , are all homeomorphic (diffeomorphic, in the , manifold case) to a common space F . In this situation, F is known as the fibre of the bundle, and the bundle is said to be a fibre bundle. This condition will always apply in the cases of interest to us, and from now on we shall mainly assume that it is satisfied.
If ( E , r , M )is a bundle with fibre F it is often convenient to indicate this by writing F + E 4 M or, sometimes,
F+
E
(5.1.3)
M
1.
0
depending on what is typographically convenient.
Examples
1. One of the simplest examples of a fibre bundle is the product bundle over M with fibre F . This is defined simply as the triple ( M x F, prl, M ) . It is sketched in Figure 5.2.
for example. the fibre could be taken to be the closed interval [1. it can be represented with the aid of the diagram in Figure 5. A famous example of a fibre bundle is the Mobius band which is a twisted strip whose base space is the circle S1. 3. as shown in Figure 5. 'c') on the two sides of the strip are to be identified.1] (and neither is it homeomorphic to it). the two points 'b' (resp. This is rather difficult to draw since it cannot be embedded in Euclidean 3space.1].2: A product bundle. in this example. 2. However.4 which shows how it is obtained from a cylinder by reflecting in the diameter de and then identifying points.1. the circle is represented by the line with the two endpoints 'u' identified.3in particular. It can be represented by taking a rectangle and identifying the short edges. note that the total space E is not the product space S1x [1. Another famous example is the Klein bottle. BUNDLES IN G E N E R A L 205 1 F " I Figure 5. Note that. . However. Thus.5.
4: The Klein bottle. b C e d C b a S1 Figure 5. a . FIBRE BUNDLES b C Figure 5.206 CHAPTER 5.3: The Mobius strip.
In YangMills theory. is different: a crosssection of a bundle of this type can at best be . we saw that vector fields and oneforms can be regarded as crosssections of the tangent bundle and cotangent bundle respectively. of the internal symmetry group G. (5. conversely. 7r o s = idM. z E M . Definition 5.1. but this is not necessarily the case for a general fibre bundle.1.5. a crosssection s defines a unique function S : M + F given by for all z E M .1.’ ( { Z } ) over z.3 The idea of a crosssection As mentioned already. BUNDLES IN GENERAL 207 5. an important example of this phenomenon will be given below in the context of principal bundles. each such function S gives a unique crosssection s defined as Eq.2 A crosssection of a bundle ( E .1.T . prl.4) Comments 1. (5. 2. and we shall see later how this idea can be extended to incorporate general tensor fields. a crosssection is equivalent to a normal function from the base space M to the fibre F .5. all fields (with the exception of the YangMills field itself) are crosssections of various vector bundles that are associated with the fundamental YangMills ‘principal’ fibre bundle of parameterised copies G. from the viewpoint of applications to theoretical physics. in a product bundle. In the earlier chapters on differential geometry.. M ) is a map s : M + E such that the image of each point z E M lies in the fibre T . M ) . Thus. The situation for a nonproduct fibre bundle F + E 2 M .More precisely.5). one of the most important ideas in fibre bundle theory is that of a ‘crosssection’. Crosssections of vector bundles always exist. In the case of a product fibre bundle ( M x F. This is sketched in Figure 5.
5: A crosssection of a product bundle. FIBRE BUNDLES Figure 5. .208 CHAPTER 5.
5.6) . S n ) with a bundle space defined as E(TS") := {(Z..6. For example. The tangent bundle TS" of the nsphere S" can be represented as a subbundle of the product bundle (S" x IRn+'.1. 2. r .y'= 0) (5. which is usually not true. 1 b C a Figure 5. Then ( G . a Examples 1. prl. 1 that is antzperiodic around the circle. a necessary condition for the existence of such sections is that G is diffeomorphic to G / H x H .1. This shows that a crosssection of the Mobius bundle is equivalent to a function from S1 to [1. G / H ) is a bundle with fibre H . In general there will be no (smooth) crosssections for fibre bundles of this type since. Let H be a closed Lie subgroup of the Lie group G.6: A crosssection of a Mobius bundle. BUNDLES IN GENERAL 209 regarded as a type of 'twisted' function from M to E.Define the map 7r : G + G / H by r ( g ) := g H . as will become clear below. a crosssection can be represented using the identification diagram in Figure 5.fj') E S" x Rnfl1 2 . in the case of a Mobius band.
which plays a crucial role in some fundamental areas of pure mathematics. (5.1. Another example of a covering of the circle is the map 7r : IR + S' defined by ~ ( r:= eir in which the covering space is the real line.7. This is sketched in Figure 5. with this definition. A crosssection of v(Sn) called is a normal field of S" as a submanifold of IRnf'. Similarly.210 CHAPTER 5.3 A topological space M covers another space M if there is a continuous map n. 5. where S' is represented by the set of all complex numbers of modulus one. especially topos theory.7) Note that. ) .1.: M + M with the property that each z E M has an open neighbourhood U c M such that n'(U) is a disjoint union of open sets Ui. each of which is mapped homeomorphically onto U by the map n. Examples 1. = 2 1) in IR"+'.$ E S" x Rn+'3r E IR such that I c= r2}. the normalbundle v(S") of S" embedded in Rnf' is defined to be the set of all vectors in IR"" that are normal to points on the sphere: E(v(S")):= {(?. Definition 5. A simple example of a covering space is the double covering of a circle by itself under the map 7r : S' + S1defined by 7r(eie):= e2ie. FIBRE BUNDLES where the nsphere S" is identified as the surface (2E Rnf' I 2. a crosssection of TS" is clearly equivalent to a vector field on S". The idea of a covering is also a useful introduction to the theory of sheaves.4 Covering spaces and sheaves A covering space is a special type of bundle that arises naturally in a number of places in theoretical physics. 2.
9) Viewed topologically. Sheaves play a fundamental role in many branches of pure mathematics (for example: complex function theory. it defines a twotoone covering of R.7: A double covering of a circle by a circle. Viewed group theoretically. this map is a group homomorphism with kernel (5. algebraic geometry. 3. 2 .5. A more sophisticated example is the map T : SU(2) + SO(3) which takes a 2 x 2 complex matrix A E SU(2) t o the orthogonal matrix 1 T ( A ):= ~ (AriA'rj) ~ tr (5.P3 (the group manifold of SO(3)) by S3 (the group manifold of S U ( 2 ) ) .1.1. are the Pauli spinmatrices.1. the theory of topoi and the logical foundations of mathematics) andalthough they have not yet appeared much in theoretical physicsit is worth giving a few of the .S) 2 where ri. a = 1 . BUNDLES IN GENERAL 211 S1 Figure 5. Running parallel to the theory of bundles is the theory of 'sheaves'. 3 . generalised cohomology theory.
. A sheaf of abelian groups is an &ale bundle ( E .rr.1. There are analogous definitions for a sheaf of rings and other such algebraic structures. not every &talemap is a covering.rr(U) is an open subset of M . M ) in which each fibre . and an assignment.rr(U). FIBRE BUNDLES Definition 5. 2.212 basic definition^. of a homomorphism puv : A(V) + A(U) (called the restriction map) such that (i) puu = 1. The product bundle ( M x F.^ CHAPTER 5.10) A wellknown example is the assignment to each open subset U of the ring C"(U) of Coofunctions on U. to each pair U C V of open sets in M . . 2. Comments 1.. There is an intimate connection between the idea of a sheaf and that of a 'presheaf' on a space M . the 'restriction 3A comprehensive and sophisticated reference to sheaves in the context of topos theory is MacLane & Moerdijk (1992). in some sense. z E M . A bundle (E. and (ii) if U C V c W then Puv Pvw = Puw. is an abelian group.4 1. Thus. M ) can never be &ale if F is " any connected C manifold. Thus covering maps are local homeomorphisms in a rather strong sense. 3. A classic example of an Qtalebundle is a covering space. and (ii) 7rp is a homeomorphism of U with . However. prl.M)is said to be &ale if the projection map IT : E + M is a local homeomorphismie. to each p E E there exists some open set U C E with p E U and such that (i) . (5. &ale bundles are complementary to our intuitive idea of a twisted product of manifolds.T . In this case. The latter is an assignment of an abelian group (or ring etc) A ( U ) to each open subset U c M .rr'({z}).
M ) is a subbundle of the 4A local crosssection of a bundle is a crosssection of the bundle restricted to some open subset of the base space.M ) gives rise naturally to an associated presheaf namely the presheaf of local crosssections4 of A. M ) has fibre F then so does its restriction to N c M . if ( E .7r.5 The definition of a subbundle Earlier. A bundle (E’. (b) M’ c M .1.5.T .M ) if c E. 2. The definition of a ‘restriction’ of a bundle is given in Definition 5. Definition 5.d. A similar situation arises for fibre bundles and leads to the idea of a ‘subbundle’ of a bundle. more tractable. Then the restriction of ( E . More generally.5. BUNDLES IN GENERAL 213 map’ is precisely what the name implies: Le. prl. ) to M N is defined to be the bundle (r’(N).. M ) are in onetoone correspondence with functions S : M + F according to Eq.5). M’) is a subbundle of a bundle ( E . Conversely. Every sheaf (A. We recall that crosssections of a product bundle ( M x F. when discussing the general ideas of differentiable manifolds. if ( E .1. from each presheaf one can construct an associated sheaf of ‘germs’. (5. it was pointed out that many important examples arise naturally as submanifolds of other. if U c V . T .1.T . (a) E’ (c) r ’ = TIE!. l ~ N ) . .T . Let N be a subspace of M . singular cochains can take their values in locally specified abelian groups.5 1. Presheaves and sheaves play an important role in algebraic topology where. 5. manifolds. The formal definition is as follows. function the puv : C”(V) + C”(U) is defined to take a C“function on V to its restriction to the subset U . for example.~ Note that ~ + ~ ~ .
This involves continuous maps from E to E’ and M to M‘ which map fibres over M into fibres over M’ (as usual. E’. More precisely: Definition 5.12) If e.. M) and (E’.11) where S : M + F is any function such that.7r’.1. this is a morphism in the category of bundlesi. maps fibres f) into fibres. M) and (E’. and M’ are differentiable manifolds)..13) Comments 1. S ( 4 ) (5. 5More precisely. the pair of maps (ZL.1.7 r . these maps are required to be smooth if E . the map f is uniquely determined by the map u. 5. I 2 \ E’ 1 (5.1. S(z)) E E. 2. . M’) is a pair of maps (u.f) where u : E + E’ and f : M + M’ such that the following diagram is commutative E M f.1. M ) then crosssections of the former have the form 44 = (z. The equation (5.6 A bundle map between a pair of bundles ( E . (z. Since 7r is surjective. structurepreserving map between the relevant objects in the category. M’).1. M’ 2.13) implies precisely that. for all 1c E M.T ’ . 7r‘ou= f 07r (5. for all z E M. M. ie.. FIBRE BUNDLES product bundle (M x F .e.6 Maps between bundles Now we come to the important idea of a ‘bundle map’5 between a pair of bundles ( E .7 r . p r .214 CHAPTER 5. u(7r’({z})) c 7r’’({f(z)}). a .
in some appropriate sense. Definition 5.d. Given maps f : M + M‘ and f’ : M’ +MI’. M”) is the bundle map between the bundles E and E“ defined as (u’ou. d .d.idM) is called a Mbundle map. Note that in this case each fiber in the bundle is mapped into itself.M)to ( E ’ ..14) (5. M’).T”. M’ & M“ 5. M ) such that U‘ 0 U = idE. there exists an open neighbourhood U of x such that (lu and qlu are Uisomorphic.15) f’o f =idM. 4. (5.1. . f’) M’) + (E”. This restriction is not contained in the definition of a bundle given above (although..1.5. M ) + (E’.rr. for each x E M .f ) : ( E . This is defined to be a bundle map (u. M ) and (E’.M) (E”.1.f’) from (E’.16) M f. a fibre bundle looks like a product ‘in the small’.1. If M = M’. a bundle map (u.rr. the composition f’ o f : M + M” is defined by f’ o f (z) := f’(f (x)) for all x E M . Two bundles ( and 7 with the same base space M are said to be locally isomorphic if.T”.T .T’. BUNDLES IN GENERAL 215 3. M’) and the bundle map (u’.T’. : (E’. Similarly. U 0 U’ = idEt.f ) from (E.M”).f‘o f ) : (E. 0 Note that fibre bundles are usually thought of as being some sort of ‘twisted product’ of the base space with the fibres rather pictorial concept which implies that. A special example of a bundle map is an isomorphism between a pair of bundles ( E . M’) t o ( E .T . we can then require a general fibre bundle to look locally like a product bundle. + The following commutative diagram helps to illustrate this: E Iff 4 E’ ITt E“ 1 P 7 (5. f 0 f’=idMt. in fact. all the examples given are of this type) and the first task is to define what it means for two bundles to look like each other locally. M ’ ) for which there is another bundle map (u’. r . the composition of the bundle map (u.7 1.
216
CHAPTER 5. FIBRE BUNDLES
2. A fibre bundle ( E 7 7 r , M is trivial if it is Misomorphic to the ) product bundle ( M x F , p r , , M ) for some space F . It is locally trivial if it is locally isomorphic to the product bundle. Note that
(a) The relation of being locally isomorphic is an equivalence relation on the set of all bundles over the topological space (or differentiable manifold) M .
(b) When a bundle is trivial there is not in general any distinguished, or uniquely determined, trivialisation.
Examples
The bundles in the previous examples are all locally trivial. Note that a locally trivial bundle is necessarily one with a fibre. In fact, an alternative definition of a fibre bundle with fibre F is a triple ( E , 7 r , M )such that, for each z E M , there exists an open neighbourhood U c M of z and a homeomorphism h : U x F + n'(U) such that 7r(h(z,y)) = z for all z E U, y E F. (5.1.17) From now on, all bundles under discussion will be assumed to be locally trivial.
5.1.7
The pullback operation
We come now t o an idea of the greatest importance. Recall that one of the most significant features of differential forms (when compared with vector fields) is that, given a pair of manifolds M and N and a map f : M + N , differential form w on N can be 'pulled back' any to give a differential form f*(w) M . It is highly significant that on a fibre bundle can also be pulled back by a map into its base space. In particular, this shows the deep consistency between bundle theory in general and the specific use of differential forms on the base space M of a bundle to describe the twists of the fibres around the base
5.1. BUNDLES IN GENERAL
217
space. As might be expected, a central role is played here by the cohomology groups of M and their relation to differential forms via DeRham’s theorem.
Definition 5.8
Let [ = (E,.rr,M) a bundle with base space M, and let be f : M’ + M be any map from M’ to M. Then the induced bundle (or pullback) of [ is the bundle f*(Q over M’ with a total space
E’ := {(d, E M’ x E f(d)= r ( e ) } e)
and with the projection map 7r’ : E’
2’.
I
(5.1.18)
+ M’ defined by ~’(2, e) :=
Comments
1. There is a natural bundle map from the induced bundle (E’,d, M’) to the original bundle ( E ,T , M ) ,defined as:
E ( f * ( I ) ) % E ( 0
(5.1.19)
M’
where f t ( d , e) := e.
k
f,
M
Iff
Note that T ’  ’ ( { Z ’ } ) = { ( x ’ , e ) E { x ’ } x E I f(d) r ( e ) } = { d } x = n’({f(d)}), and hence ft maps each fibre T ’  ~ ( { Z ’ } ) C E ( f * ( [ ) ) homeomorphically onto the fibre r  ’ ( { f ( x ’ ) } )C E(<).In particular, this shows that each fibre of the pullback f*([) is homeomorphic to the fibre F of the original bundle
c.
2. If j : N + M is the inclusion map for a subspace N of M then the restriction [ I N is Nisomorphic to the induced bundle j * ( ( ) .
3. If g : M“ + M’ and f : M’ + M , then if [ = (E,.rr,M) is a bundle over M, the pullbacks obey the same type of functorial property as do the pullbacks of differential forms. More precisely, the bundles g*(f*([)) and (f o g ) * ( ( ) are MI’isomorphic.
218
CHAPTER 5. FIBRE BUNDLES
: M’
is a smooth map between manifolds M‘ and M then the functions on M can be pulled back by f to give functions on M’. Specifically, if 4 E C m ( M ) ,define f*(q5) E C”(M’) by f*(+)(z):= +(f(s))for all z E M’. It is noteworthy that crosssections of bundles pull back too. Thus if s : M + E(<)is a crosssection of <, an associated crosssection S : M’ + E ( f * ( < ) can be ) defined by Z ( 2 ) := (2, ( f ( z ) ) . s (5.1.20)
4. Iff
+M
In studying this definition it is helpful to look at the following diagram
which commutes in the sense that
fc o S = s o f .
5.1.8
Universal bundles
There is a deep algebraictopological connection between the classification of bundles over some fixed base space and the pullback operation. This starts with the observation that if = F + E + N is a fibre bundle over n/,and if f is a map from M into N, then it seems intuitively clear that the pullback bundle f *(<) will not change dramatically if f is varied in a ‘continuous’ way. More precisely, if f,g are a pair of maps from M into N that are homotopic then it is fairly easy to show that f*(<) and g*(<) are isomorphic bundles over
<
M.
Thus, if B F ( M )denotes the set of isomorphism classes of Fbundles over M I any Fbundle over N generates a map
<
f
a t : [MINI + B F ( M )
! + 
f*(O
(5.1.22)
where [MINIdenotes the set of homotopy classes of maps6 from M into N.
‘Strictly speaking, these maps have to be basepoint preserving, ie., we choose points zo E M , yo E N and require that all maps f from M to N under consideration satisfy f ( ~ = yo. )
5.1.
BUNDLES IN GENERAL
219
In general, the map in Eq. (5.1.22) is neither onetoone nor onto. However, for a wide class of spaces M (for example CWcomplexes, which includes all differentiable manifolds) there exist universal bundles U F = F + EF + BF with the property that the map
is both onetoone and onto. Thus:
1. if f , g : M + BF then f * ( U F ) N g*(UF) if, and only if, f and g are homotopic maps; 2. if 77 is any Fbundle over M , there exists some f : M such that 77 is Misomorphic to f*(UF).
+ BF
It can be shown that a n y Fbundle whose total space EF is contractible can serve as a model for a universal F  b ~ n d l e . We shall ~ see some concrete examples of this in the next section on principal bundles. Thus the set of isomorphism classes of Fbundles over M is in bijective correspondence with the set of homotopy classes of maps from M into B F . This procedure is rendered meaningful by the observation that different universal Fbundles have base spaces that are necessarily homotopically equivalent. Hence the problem of classifying bundles over M ‘reduces’ to the problem of computing the homotopy set [ M ,BF]a task that is tailormade for study with the aid of algebraic topology. The general idea of ‘characteristic classes’ can also be seen clearly from this construction. Thus let w be an element of some cohomology group of B F and let be a bundle over a space M induced by a map $ : M + BF, i.e., 77 I f ( U F ) . Then the characteristic class of 77 I corresponding to the ‘universal’ class w is defined as
w9 := g ( w ) .
(5.1.24)
Note that this object is unambiguously defined since any pair of maps $1 and $2 from M into BF that induce isomorphic bundles over M
’For example, see Husemoller (1966), or Milnor & Stasheff (1974).
220
CHAPTER 5. FIBRE BUNDLES
are homotopic, and hence the induced cohomology classes 4;(w) and &(w) are equal. Note also that if $ : M + N and q5 : n/ + B F with 7 = +*(IAF), then (4 o $)*(IAF)21 $*(c$*(UF)N $*(q), and hence
Thus the association of a characteristic class wg with a bundle 7 is ‘covariant’ with respect to the pullback operation of bundles and cohomology classes. In particular, if a model can be found for B F that is a differentiable manifold, then one can define real characteristic classes of an Fbundle on a manifold M as the pullbacks of DeRham cohomology classes on B F .
5.2
5.2.1
Principal Fibre Bundles
The main definition
We start now to consider the wideranging question of how fibre bundles are to be classified, and how the different types can be constructed. A crucial idea is that of a ‘principal fibre bundle’, which is a bundle whose fibre is a Lie group in a certain special way. These have the important property that all nonprincipal bundles are associated with an underlying principal bundle. Furthermore, the twists in a bundle associated with a particular principal bundle are uniquely determined by the twists in the latter, and hence the topological implications of fibrebundle theory are essentially coded into the theory of principal fibre bundles. The crucial definitions are as follows.
Definition 5.9
1. A bundle ( E ,7r, M ) is a Gbundle if E is a right Gspace and if ( E ,7r, M ) is isomorphic to the bundle ( E ,p, E / G ) where E/G is the orbit space of the Gaction on E and p is the usual projection
5.2. PRINCIPAL FIBRE BUNDLES
map:
221
E
M
I IT\
A
E
lp
(5.2.1)
E/G
Note that the fibres of the bundle are the orbits of the Gaction on E , and h e n c e i n generalwill not be homeomorphic to each other. 2. If G acts freely on E then ( E ,T , M ) is said to be a principal Gbundle, and G is then called the structure group of the bundle. The freedom of the Gaction implies that each orbit is homeomorphic to G, and hence we have a fibre bundle with fibre
G.
Examples
1. The product bundle ( M x G, prl, M ) is a principal Gbundle under the obvious right Gaction (z,go)g := (2,909). One might expect a 'trivial' Gbundle to be defined as one that is isomorphic to the product bundle. However, this requires a definition of a 'morphism' between two principal bundles, and this is not the same thing as a bundle map since the Gaction must also be preserved; we shall return to this later.
' 2. Let G be the cyclic group 2 2 = { e , a } with a = e, and let the nonidentity element of this group act on the nsphere S" by exchanging the antipodal points: Ze := 2,Za := here S" is viewed as the set of unitlength vectors 2 in lR"+l. This Zzaction is free and gives rise to a principal Zzbundle whose base space is diffeomorphic to the real projective space RP" N Sn/Zz. A special case is the bundle Zz + S3 + R P 3 discussed earlier in the context of the covering map Eq. (5.1.8).
c
3. One of the most important examples of a principal bundle occurs when H is a closed Lie subgroup of a Lie group G. Then H acts on the right on G by group multiplication with an action that is clearly free and whose orbit space is simply the space of cosets G/H. Thus we get a principal Hbundle (G,T,G/H) whose fibre is
222
CHAPTER 5. FIBRE BUNDLES
the group H . When this bundle is locally trivial (see below) the Lie group G can be thought of globally as being the 'twisted' product of H with G I H .
For example, the U1 action on SU(2) gives rise t o a bundle U1 + SU(2) + SU(2)/Ul, which is S1 + S3 + S2. This famous fibering of the threesphere by the circle is known as the Hopf bundle.
Other examples of principal bundles obtained in this way are
U ( n ) + SU(n + 1) + GP"
and and
(5.2.2) (5.2.3) (5.2.4)
SO(n) + S O ( n
SU(n) + S U ( n
+ 1) + S"
+ 1) + S2"+l.
4. In YangMills theory with the internal symmetry group SU(2), the bundle corresponding t o instanton number one is a principal SU(2)bundle whose base space is a foursphere regarded as the onepoint compactification of 'Euclideanised' Minkowski spacetime. We recall that the group space of SU(2) is a threesphere, and in fact this instanton bundle is of the form S3 + S7 + S4. This is clearly not a product bundle since S7 is not homeomorphic t o S3 x S4. 5. Another example of a principal bundle can be obtained by considering the action of the multiplicative group G* of nonzero complex numbers on the space G""  (0) given by
(ZI,Z2,.
. . ,&+l)X
:= ( Z J ,
%A,. . . , Z n + J )
(5.2.5)
for all X E
6,.
This is a free action and the orbit space is the space GP" of complex lines in the vector space Thus we get the principal bundle G* + G"+l  (0) + CP" (5.2.6)
en+'.
where the projection map associates with each n
(z1, z 2 , . .
+ 1tuple
., &+I)
the point in CP" whose homogeneous coordinates are (z1; z 2 ; . . . ; ,zn+l)
The total space B ( M ) of the bundle of frames is defined to be the set of all frames at all points in M .2. M . PRINCIPAL FIBRE BUNDLES 223 A closely connected bundle is obtained by restricting to the U1 subgroup of G.7) 6. . + . i = 1 . . which acts freely on the real (2n+l)sphere embedded in a="+' .zm).d ) as the coordinates for a differential z2. .9) for some nonsingular matrix map h : U x GL(m. B ( M ) becomes a manifold of dimension rn m2. structure on B ( M ) . 2 . b2. .zm. and the projection map 7~ : B ( M ) + M is defined to be the function that takes a frame into the point in M to which it is attached. z 2 .'(U) E GL(m. .. . R). m (5.bm) of basis vectors for the tangent space T. Thus we can define a and then use (zl. Then any base b = ( b l . b2.( 0 ) . . . A natural. A linear frame (or base) b at a point z E M is an ordered set (bl.M.. The bundle space B ( M ) can be given the structure of a differentiable manifold as follows. IR) on B ( M ) is defined by m m m for all g E GL(m. A principal fibre bundle of major importance in differential geometry is the bundle of frames attached to an mdimensional differentiable manifold M . .. (5.. . .R)... . . free right action of GL(m.2. z E U .bm) for the vector space T .2. Let U C M be a coordinate neighbourhood with coordinate functions ( 2 1 . By these means.5. can be expanded uniquely as bi=xg(&) j=1 5 m .R) + . . This gives rise to the principal U1bundle u1+ S2"+l + GP".
. it is equal to the unit element of G for all spacetime points that lie outside some compact subset of M . more generally it is the group of automorphisms of the principal bundle (see below). ie. between a pair of principal Gbundles f) ( P. . Another important bundle in YangMills theory is obtained by considering the set A of all YangMills potentials in the theory.2 Principal bundle maps After this fairly extensive set of examples we must return to the general theory and.2. A bundle map (u. In either case.10 1.224 CHAPTER 5. 5. Definition 5. to the generalisation of the idea of a bundle map to include the preservation of the group action in a principal bundle. With some care in the precise selection of G and/or A it can be shown that this action is free and hence that A is a principal Gbundle with base space dG (Singer 1978). the physical configurations of the theory are identified with potentials that are related by gauge transformations.. M') is said to be a principal bundle map 'More precisely. / A related example arises in general relativity with the space RiemM of Lorentzian (or Riemannian. FIBRE BUNDLES 7. ie.T . if that is appropriate) metrics on M on which the diffeomorphism group of M acts as a type of analogue of the gauge group G of YangMills theory with g E RiemM being taken to q g under the action of #I E DiffM. then the corresponding gauge group G is the set of alls smooth maps from M into G. they lie in the same orbit of the action of the gauge group G on A.. It is important to make this distinction in order to separate genuine gauge transformations from 'rigid' (2. a gauge transformation is required to have compact support. M) and (P'. in particular. spacetime independent) transformations that may appear as symmetries of the Hamiltonian or action. Both these examples differ from the previous ones in that the spaces concerned are infinitedimensional. e. If the underlying Gbundle over spacetime M is trivial. and hence some subtlety arises in the detailed constructions.n'.
12) An example of the generalised situation arises in the theory of spinor fields on a Riemannian manifold M (see also the remarks following Eq. 1. T . R). PRINCIPAL FIBRE BUNDLES 225 (or just a principal map) if u : P + P’ is Gequivariant in the sense that u(Pg) = u(P)g (5. (5. the corresponding spinor group Spin(3. In this case. In the special case of SO(3.11) for all p E P and g E G. l . We come now to a theorem that is of some significance. although it is easy enough to prove. C). ) : M P ’ 1 P’ 1\ (5. This can be generalised to the case in which (P. (4.d. 2. M’) is a G’bundle.13) MldnZM Then u is necessarily an isomorphism.idm) be a principal Mmap between a pair of principal Gbundles ( P .T’.2.4. Theorem 5.T .M ) and (P’.2.5. see Husemoller (1966). ( P > W for all p E P .2. and called the spin group of Mthat acts on the bundle of spinor frame^. R).2.1 Let (u.(pl) = d ( u ( p l > )= ’For a full discussion of the spinor group Spin(m. and G is its double covering groupdenoted Spin(rn. Then . and g E G. We then require u(Pg)= . Thus let p l E P and p2 E P be such that u(p1) = u ( p 2 ) . and A : G + G’ is a group homomorphism.17)). G’ is the orthogonal group SO(rn. (P’.R) (where rn = dim M ) that acts on the bundle of orthonormal frames. G’ is the Lorentz group SO(m 1 . .^ In the case of a Lorentzian manifold. R)happens to be isomorphic to SL(2. 1. M ) is a Gbundle. R ) . l).l.’ Proof (i) First we shall show that u is onetoone.
. in the case of differentiable manifolds). But G acts freely on P . FIBRE BUNDLES d ( u ( p 2 ) ) = 7r(p2). it is a famous theorem'' that H + G + G/H is locally trivial when G is a finitedimensional Lie group. Thus pl =p2.. including showing that the maps involved are continuous. r . and therefore p l and p2 belong to the same fibre. This is called the automorphism group of the bundle and is denoted Aut(<). Armed with the idea of a principal bundle map we can now give a proper definition of what it means to say that a principal Gbundle ( P . the automorphism group of the bundle plays the same role as the familiar group of 'gauge transformations' in the instantonzero sector (when "See Husemoller (1966) for full proofs of results of this type. prl. which proves that u is onetoone. M ) .T . that there is a principal bundle map from ( P . Namely. This is left as an exercise. M ) t o the product bundle ( M x G. although this is not always a minor matter t o prove: for example. Hence there exists g E G such that p l = p29.226 CHAPTER 5 . It follows from the above that the set of all principal maps from a principal Gbundle = ( P . and ) . 'lFor example. . In YangMills theory in a nontrivial instanton sector. ) choose p E P such that ~ ( p= ~ ' ( p ' ) Then d(u(p)) = ~ ( p= ~ ' ( p ' ) and hence p' and u ( p ) are in the same fibre. M ) t o itself forms a group. Thus let p' E P'. and thus u(p1) = u(p2g) = u(p2)g = u(p1)g. M ) is locally trivial as a bundle." QED Comments 1. and for each open set U c M on which the restriction of the bundle is trivial there is a trivialising map h : U x G + . 2. (iii) Finally. All the specific principal bundles considered above are locally trivial. The idea of local triviality is then defined in the obvious way as the requirement that ( P . (ii) Next let us show that u is surjective. and so u is surjective. and hence g = e. it must be shown that the inverse map is continuous (or smooth. see Chevalley (1946).T .'(U) that is equivariant with respect to the G action on U x G and r'(U). M )is trivial. Therefore there exists g E G such that p' = u(p)g = u ( p g ) .
rr(pg) = n ( p ) = f (XI). Theorem 5.2 Let ( be the product bundle ( M x G. f ( z .s) = (5.M) be a principal Gbundle and let f : M’ + M .) . This induces Clearly this action is an action on P’ since . = x It has been remarked already that one of the significant features of fibre bundles is the way in whichlike differential formsthey can be pulled back via a map between a pair of manifolds. ( 4 g g ’ ) = u x ( z .2.r. T h e n the automorphisms u : M x G + M x G of this principal bundle are in onetoone correspondence with m a p s x : M + G such that u ( z . Aut(() is isomorphic to the group C m ( M . G )of smooth gauge transformations. The next result is important since it shows that the property of being a principal bundle is preserved in this process.3 Let ( = (P. Proof Let f *(() = (P‘. x ( z ) g ) . Proof Let u be an automorphism of (. x u ( z ) g ) where I have used the obvious right Gaction on the product space M x G. M ) .2. PRINCIPAL FIBRE BUNDLES 227 the bundle is trivial). g Conversely. M’) and consider the diagram P’ P‘/G 3.14) Define a Gaction on M’ x P by ( 2 . . x ( z ) g ) . g ) = (z. x u ( z ) ) g = (z. Theorem 5. given x define u x ( z . Then we have QED ( u x ( w > > g ’ (x.g ) = (u(z. p ) g := ( z ‘ . 99’). ) := (z. e Then u(z. as shown by the following theorem. h M‘ . Indeed.5. Then u ( z . prl. g ) ) for some f : M x G + G. T h e n the induced bundle f’(0i s also a principal Gbundle.e ) ) g = (2.d. Define a map xu : M + G by xu(z):= f(z. I! T M IT (5. P f. p g ) . when ( is trivial.
produces a bundle Z + IR + S’that is a universal Zbundle. (iii) h ( ( z ‘ .pg)G = T’((z’. Hence [ M . p )as . FIBRE BUNDLES free.S1]. ~ ) G (z’.Q)G = h(z’.Now r1(S1) Z and N all other homotopy groups of S1 vanish. NOW define h : P’/G + M’ by h ( ( z ’ . l 21 H 1 ( M . i e .r w eir. Since the twists of any fibre bundle are inherited from the parent principal bundle. (ii) h is injective. 2.228 CHAPTER 5. n ) is defined to be any topological space with the property that x n ( K ( x . The theorem shows in particular that the restriction Eru of E to a subspace U of M is itself a principal bundle. Thus. P 2 ) G . ~ then z = z .Pl)G = ( X h . for example. := ’ (iv) h and hl are continuous [Exercise!]. which means precisely that (Z:. p ) 6 P’ ~ with f(z’) = ~ ( p )Then .M’) is a principal GQED Comments 1.pz)~) and hence ~ ( p l = f(zi)= f(zi) = 7r(pz) which implies that ) there exists g E G such that pl = p z g . . p )denotes the Gorbit that passes through ( z ’ . f*(E) = (P’. i .n’. and if n is any positive integer. but then h(z’. these are characterised by the property that E G is a contractible space. ~ ) G )= 2’ := where ( z ’ . giving rise to S] Z). Thus the pullback bundle bundle. n ) )E x and with all other homotopy groups vanishing. Thus the principal Zbundles over any manifold M are classified by the homotopy set [ M . p ) ~ ) z clearly defines a map that is surjective.pl)~)= h((zL. S1 is an EilenbergMaclane12 space K ( Z . if R is any group. it follows that particular interest is attached to the universal Gbundles G + E G + BG. an EilenbergMaclane space K ( x . p ) ~ ) T ‘ ( ( Z ’ .q ) then there = ~ exists g E G such that q = pg. since if h((zi. It can . 1). “In general. (i) The map h is well defined since if ( z ’ .required.pg)G) = 2’ = h ( ~ ’ . the infinite covering of S1by the real line under the map 7r : IR + S1. As for all universal bundles.
2... .. PRINCIPAL FIBRE BUNDLES the classification result 229 &(M) = H1(M. . (5. . 2 . . is equal to the singular n)] cohomology group H " ( M . . .2. shows that 7r2(CPoo)N Z and that all other homotopy groups vanish. . . . (5. A more sophisticated example is afforded by the sequence of principal U1bundles (see Eq.5. (5.e. and (ii) the homotopy set [ M . z.2.17) is a universal U1bundle.17) Now.16) Taking the limit n + 00 in an appropriate sense. e (5. n ) are homotopically equivalent. . This in turn implies the famous classification result B.). (5. i. generalises to the feature that all the homotopy groups of S" vanish. Thus Eq. zn)ezO:= (eiOzl.2.2.18) 4. n = 1 . we get the bundle u1+S " + CP". . whichsince it is a CW compleximplies that S" is a contractible space.2). be shown that (i) any two EilenbergMaclane spaces of type ( T .K ( T ..7)) U1 + SZn+l + CP". (5.) in SZn+' (identified as the set of all vectors z' E C"" such that IziI2= 1) as (zl. and where eiO E U1 acts on an element (z1.2.CP"]. A similar argument to that employed for the infinitesphere.2. .yl(M)N H2(M. .ei0 z. and the set of Riemannian metrics RiemM are both contractible spaces. i0 z2. Z). . and hence U1bundles over an arbitrary manifold M are classified by the homotopy set [M. the fact that ri(Sn)N 0 for i < n.15) 3. The set of YangMills potentials A.z2. T ) of M with coefficients in the group A. Thus the fundamental 'gaugebundles' G + A + d/G and DiffM + RiemM + RiemM/DiffM are universal bundles for the gauge group 4 and the diffeomorphism group DiffM respectively. in which the U1 group is regarded as the group of all complex numbers of modulus one.Z). z 2 . CP" is an EilenbergMaclane space K ( Z .
as appropriate).e. Then.x. and (ii) uu(Pg) = (71. The resulting function xu : P + G is unique because the Gaction on the bundle space P is free.Xa(P))g = uu(P)g. 4 M x G is h.3 Crosssections of a principal bundle We come now to a theorem that is of considerable significance in a variety of applications of fibre bundle theory in differential geometry and theoretical physics. : P defined by h. prl. Hence u. e). Then.(pg) = x.x. an isomorphism.(p) E G such that p = u(~(p))x. M ) ) if and only if at possesses a continuous crosssection. this is left as an exercise.(p)g. if h : M x G + P is a principal map then we can define a crosssection Oh : M + P by C J h ( 5 ) := h(5. Clearly X. is a principal map and therefore. : P + M x G by u. by the theorem above. :M x G + P . (i) u. QED Comments 1.(q g) := o ( x ) g . Now define u. especially in those areas where global topological properties are important. The inverse of the map u. there must exist x.4 A principal Gbundle (P. M isomorphic to ( M x G.. The converse statement is trivial.230 CHAPTER 5. This theorem states thatontrary to what is sometimes felt to be intuitively ‘obvious’a principal fibre bundle does not have any smooth crosssections unless it is untwisted! More precisely: Theorem 5. M ) is trivial (i. Namely. is a bundle map. FIBRE BUNDLES 5.(Pg)) = ( 4 P ) l X U h M = (7r(P).7 r . for any point p E P .(p)).2.(P). The proof of the theorem is completed by showing that all the various maps are continuous (or smooth.(p) := (7r(p). Proof Let cr : M + P be a crosssection of the bundle.(p).
for example. 3. It is clearly of some importance to know when a principal bundle is trivial. 13A Hausdorff space X is paracompact if for every open covering of X there exists another covering that refines the first. . The GribovSinger effect means that there is an intrinsic obstruction t o choosing a gauge that works for all physical configurations. This is primarily a global problem and involves a delicate interplay between the topological structures of the fibre and base space.5. A similar situation arises in the corresponding situation in general relativity with the principal bundle DiffM + RiemM + RiemMIDiffM. then to the 2simplices. a partition of unity is a collection of positive. one for each open set of the covering and which vanishes outside the open set. Partitions of unity are used to patch together locally defined geometrical objects to give a global object. One wellknown result is that every principal bundle defined over a contractible. and with the property that the sum of the functions is 1 everywhere. 4. An example of a nontrivial principal bundle arises in YangMills theory and is responsible for the socalled ‘Gribov effect’. PRINCIPAL FIBRE BUNDLES 231 2.2. and then trying to construct a crosssection by first defining it on the 0simplices (this is always possible since the 0simplices are just points) then extending it t o the 1simplices. The precise extent to which a global crosssection of a principal bundle does not exist can be used t o give a topological classification of the twists in the bundle. paracompact13 base space is necessarily trivial (see later). realvalued smooth functions. (Helgason 1962). Roughly speaking. as Singer showed. and so on.. Choosing a crosssection of this bundle corresponds physically to choosing a gauge. This is the bundle 4 + A + d / B referred to earlier in which d is the set of all YangMills potentials and B is the gauge group of the theory. For details see. whether or not this has any real significance in the theory is still a matter for debate. and hence to relate to the theory of characteristic classes. One approach (which is applicable for any fibre F ) involves triangulating the base space sufficiently finely that the bundle is trivial over any simplex. At some stage. However. this particular principal bundle is definitely not trivial and hence no smooth crosssections exist (Singer 1978). The importance in differential geometry of paracompact spaces lies in the existence of a partition of unity for every locallyfinite covering of X .
232 CHAPTER 5. Let X and Y be any pair of right Gspaces. In particular.nrcl(F)). .11 1. this analysis shows that potential obstructions to constructing a crosssection lie in the cohomology groups H'"(M.3.Ic = 1 . Then the Gproduct of X and Y is the space of orbits of the Gaction on the Cartesian product X x Y .(F))..n. we can form a fibre bundle with fibre F for each space F on which G acts as a group of transformations. 2 . if there exists g E G such that y) y') x = xg and y' = yg.1 Associated Bundles The main definition In this section we shall discuss a method of constructing a great variety of fibre bundles that are associated in a precise way with some principal bundle. and hence generates an element of the cohomology group H"+l(M. Thus there will be some nsimplex A and a map s from A into the fibre F which cannot be extended to the interior of A.r.M ) with structure group G. It can be shown that this association to A of an element of 7rn(F)depends only on the cohomology class of the (n+l)cell interior of A. The basic idea is that given a particular principal bundle (P. FIBRE BUNDLES if the bundle is nontrivial. First we must introduce the concept of a 'Gproduct' of a pair of spaces on which G acts. a dimension n will be reached for which the crosssection cannot be extended to all the simplices of the next higher dimension. But A is homeomorphic to an nsphere. ' . = (d. This gives the primary obstruction to constructing a continuous section of the bundle. Definition 5. Thus we define an equivalence relation on X x Y in which (x. For discussion of this and related topics see Steenrod (1951). . . and the inability t o extend s is equivalent to the statement that the element in r n ( F ) corresponding to the homotopy class of s is nontrivial. 5. In general.3 5. this analysis can be applied to a universal principal bundle G + E G + BG and yields the basic universal characteristic classes for all principal Gbundles.
given any right Gspace Y .y1]) ~ ( [ g ~ . G is clear. Define PF := P X G F where (p. YZ) = (Sl. it should be shown that the map L and its inverse are continuous.y]) := ygl is a Proof (i) This map is welldefined since [gl. Y2) = (91gT1g2. the group G can itself be used as a right Gspace X to give the Gproduct G X G Y . V ) < . E X x Y is written as ( 5 .3. ASSOCIATED BUNDLES 233 2.y2] implies that there exists g E G such that g2 = glg and y2 = y1g.v)g := (pg. The Gproduct is denoted X X G Y .Yl).v2] then there exists g E G with (PZ.l v ) and then ~ ( P Z= .It is relevant for our considerations later that this space is homeomorphic to Y : Lemma The map homeomorphism. and hence ( 9 2 . QED Now we come to the crucial definition of an 'associated fibre bundle'.5. As usual. y 2then = ]) YlS? = Y 2 9 i 1 and so ( 9 2 . this is left as an exercise (or see Husemoller (1966)).y ) ~ or just [z. M ) is a fibre bundle over M with fibre F that TF.rr(plg) = ~ ( p l ) . p . and the equivalence class of (x.g . L : G XG Y +Y defined by L([g. and define a map T F : PF + M by TF([P.12 Let = ( P . Finally. vl] = [P2. (iii) The map L is injective since if ~([gl. ([F]:= (PF.yl] = [g2. Note that TF is welldefined since if Ipl. as is the corresponding proof of differentiability in the manifold case with G being a Lie group.y1gT1g2).glv). Z ) = ( p l g . is said to be associated with the principal bundle ( via the action of the group G on F . Definition 5. Note that.v]) := ~ ( p ) Then . Therefore Yzg? = Y199'9? = Ylg? (ii) The map L is surjective because L([e.y]) = y. y] if the group y) .M ) be a principal Gbundle and let F be a left Gspace.
Theorem 5. if [p. we must show that t[F]really is a fibre bundle. FIBRE BUNDLES Note also that in order for the definition to make sense. QED Comments 1. and therefore j p o is a homeomorphism (see Husemoller (1966) for a discussion of the continuity properties of the translation function). pis the so] v ) called translation function of the principal bundle t. Proof E M. t[F] is Misomorphic to the product bundle (M x F. In particular. The crucial step is contained in the next theorem. M) and also by the way in which the group G acts on the fibre F . within the category of associated bundles. p ) vBut. . Now.pg)g'v = ~ ( p 0 . gv = Y for all g E G and all Y E F ) then. Choose some particular point po E 7r'({x}) so that i p o ( F ) 7rF1({x}). t[F]is only deemed to be trivial if the underlying bundle t is trivial as a principal Gbundle.vI t+ ( T ( P ) . Now.234 CHAPTER 5.v]. M) via the map b. . w ] = ~ ( p and hence i p ( F ) c v ) 7rF1({7r(p)}). see below for further remarks on this potentially confusing 0 nomenclature. considered only as a bundle.5 For each x to F . ~ into ~ ( p 0 .v] E 7rF1({x)) define jpo : 7rF1({x}) + c F as the map that takes [ p . . The map j p o is welldefined since T(p0. prl. p ) where ~ ( p 0 . 7 r ~ [ p . if G acts trivially on F ( i e .7 r . It is clear that the 'twists' in an associated bundle ([F]are determined by the twists in the underlying principal bundle t = (P . 4 However. whose value is defined as the unique element in G such that p = p o ~ ( p 0 . Hence the maps ip0 and j p o are inverses of each other. the space 7rF1({x}) is homeomorphic To each point p E P(t)there is associated the map iP : F + PF defined by ~ ~ ( :=) lp. p ) .
(b) Let Z2 act nontrivially on [1. ie. Let be the principal Zzbundle formed from the action of the group Z2 = {e.1) .the trivial action).3. ASSOCIATED BUNDLES 235 Examples 1. Then the associated bundle B ( R m ) = B ( M ) XGL(". the principal bundle of interest is the bundle of frames B ( M ) on the mdimensional manifold M .and ( is simply the double covering of the circle by a circle that was discussed earlier in the context of Figure 5.R).1]. Then the associated bundle <[F] just is a cylinder.1] by a E Zz taking T E [1.3. in which a E Zz sends 2 E S1 into 2. A simple case is when F = IR" and the action of GL(m.IR) is just the usual linear group of transformations. The structure group is GL(m. In this case the ensuing associated bundle is the Klein bottle.1] by a E Z2 taking T E [1. a } on the circle S1. Thus we see that some of the most familiar examples of fibre bundles are obtained from what is arguably the simplest nontrivial principal bundle.T . Then the associated bundle ( [ F ]is the Mobius strip. In this case. 2. The next example is of major importance in differential geometry as it provides a 'bundletheoretic' way of viewing general tensorial structures.7..1] into . The base space S1/Zz is also diffeomorphic to S1. especially vector spaces. < A number of interesting bundles can formed from this particular principal bundle by the process of association. which can act on a variety of spaces. For example: (a) Let Z2 act on F = [1.5.. considered solely as a bundle it is isomorphic to the product space S1 x [1.R) IR" can be identified with the tangent bundle T M via the map (5.1] into itself (ie. (c) Let Z2 act on F = S1 by a E Zz reflecting the elements of the circle in a diameter.
IR). .R) a group of linear transformations of a real vector as space V. FIBRE BUNDLES where b is a base for the tangent space TxM (ie. and crosssections of this vector bundle are tensor fields of type p.. . 3.. The bundle space is diffeomorphic to the product M x R and the crosssections are in onetoone correspondence with the usual realvalued functions on M . Then the corresponding associated bundle is the bundle of absolute scalars. (c) More generally let V = R m ~ R m ~ . ~ ~ m ~ ( R m ) * ~ .R) + AutV be any representation of GL(rn. we mean by this merely that c [ F ]is diffeomorphic .T m ) .R) act on v' E V by m (det kl . 5.2 Associated bundle maps At various places above. T2. and let a E GL(rn.ki=l hi .hk=l c c m The associated bundle is known as the bundle of tensor densities of weight w . @ ( R ) where the first tensor product is taken k times and I"' the second 1 times. For example: (a) Let V = R and let p be the identity transformation p(a) = 1 for all a E GL(m. b E Bx(M))and r'= ( T I . .. . let p : GL(rn. The ensuing associated bundle is called the bundle of scalar densities of weight p..236 CHAPTER 5. Generally speaking. . More generally.. we have implied that the total space c [ F ]of an associated bundle is 'like' some other space 2 that is well known. This example is quite useful in so far as the definition of densities in a 'nonbundle' way tends to be a little awkward. contravariant of rank k and covariant of rank 1. (b) Now let V = R again but this time set p(a) = (det for some w E R. Then B[V] is called the bundle of tensors of type p (it is actually a vector bundlesee below). . .3. .
The critical definitions are as follows: Definition 5. Two associated bundles <[F]and [[F]are said to be locally isomorphic if the underlying principal bundles are locally isomorphic. but this would not take into account the ‘associated’ status of the bundles and the corresponding relations to their underlying principal bundles. MI). (5. 9lvl) = [u(pg). Note that this implies that [[F] is trivial as a bundle. This is welldefined since p u]. An associated fibre bundle E[F]is trivial if its underlying principal bundle t is trivial. ie. 4. if for each z E M .3) and it is clearly a bundle map [Exercise!]. t[F]and J’[F] defined to be a bundle map of the type above.13 1. Not every bundle map is necessarily of this type. A map between the associated bundles P X G F and P’ X G F can be defined by u ~ ( [ w]).T I . this could be defined simply as a bundle map.) be a principal bundle map between a pair of principal f Gbundles t = (P. But this raises the question of or is meant by a genuine ‘morphism’ between two associated bundles. The most useful definition is to start with a principal bundle map between these underlying bundles. is An associated bundle map between a pair of associated bundles 2..3. glul = [ u ( p ) g . ASSOCIATED BUNDLES 237 to 2. is isomorphic to 2 as a bundle. 3. For example (as remarked above). M ) and t’ = (PI.3.5. and then use this to construct a bundle map between the two associated bundles.7r.glwl = [ U ( P > . w([pg. := [u(p). but the converse is false. if t is any nontrivial principal Gbundle and F is any space on which G acts trivially. and it follows therefore that this definition is more restrictive than the first one. thenas a b u n d l e ( [ F ]is trivial but it is not trivial as an associated bundle. Let (u. Of course. there exists an open neighbourhood U c M of z such that t1~ [lv are isomorphic as principal and bundles. . A failure to observe this subtle distinction can cause some confusion when reading the literature.
there exists an open neighbourhood U C M of z and a bundle isomorphism h : U x G + P ( [ ) l u .w) = [o(z). We know that it is a bundle. Then there is a bundle MIisomorphism between (i) the pullback f*([[F])[ [ F ] . we can define a map u’ : P F + U~x F by ~ u’([h(z. (5.238 CHAPTER 5. prl. w]. FIBRE BUNDLES We consider now the important question of the status of the pullback of an associated fibre bundle. and hence.B. IfFisaGspace. := (z. w] e which is clearly a bundle Umap. M ) with a map L : (M x G)Gx F + M x F defined by L([(X.w) := [h(z. Comments 1. ) . If h comes from a local section o : U + P ( [ ) then h’(z.wecandefineh’: U x F + Pplu b Y (5. P>.5) . g>.M) be a principal Gbundle with base space M and let G act o n a space F to produce the associated bundle [ [ F ] . p r .I.w].3. := (z.4) h’(z.3. QED A simple corollary to the above is the result that the restriction ~ [ F ] ofNthe associated bundle [ [ F ]to a subspace N c M is NI isomorphic (as a bundle) to the bundle ~ I N [ which is associated t o F] to the restriction [IN N of the principal bundle [. w]):= [(x. [p.Let M’ be some other space and let f : M‘ + M. Let [ = (M x G. 4) 2. > s). M) and let F be any Gspace. of and (ia) the bundle obtained by first pulling back [ to give the principal bundle f*([) and then forming the associated bundle f * ( [ ) [ F ] . Then the associated bundle [ [ F ]is Misomorphic to the product bundle (M x F .6 Let [ = ( P .T . As a bundle. 4. Then it is a straightforward p exercise [!I to show that j is an isomorphism. Proof Construct a map j : f * ( [ [ F ] c M’ x ( P X G F ) + f * ( [ ) [ F c ) ] (M’x P ) X G F by j ( z . (N. but is it associated to a principal bundle? The answer is contained in the following theorem: Theorem 5. To see this note that a principal bundle E is locally trivial.)Similarly. ) . [ [ F ]is locally trivial. . for each z E M.
These relations are characteristic in the sense that any family of Gvalued functions satisfying these relations are necessarily the transition functions of some fibre bundle.gv)we can write U F as the map from M x F into itself defined by (z. 4. every automorphism of ( is of the form u ( z . or transition functions of the bundle. This is the foundation of the approach t o fibre bundle theory based on local bundle coordinate chartssee Steenrod (1951) for further details. Uz. ) := [u(p). ( [ F ]is locally trivial. On the overlap U1 n Uz there is therefore a principal bundle morphism h. hence. bundleisomorphism of PF with M x F defined aboveb([(z. the natural definition of an automorphism of an associated bundle (IF] is a bundle map U F defined by u ~ ( I p . 3. the structure functions. are known as (z. ( z ) v ) . g ) . guzul(z)g). Thus there exists a map gulu2 : Ul n Uz + G such that h. v) tj (2. g). Thus h' defines a bundle Uisomorphism from U x F t o P F ~ .] where u E Aut [ is an automorphism of w] w the principal bundle (. g ) = (5. U3 with U1 n U2 n U3 # 8. According to the definition above. They satisfy the relations for all U1. : Uz x G + qu2.: o h ~ .This is relevant to understanding how a gauge x transformation in YangMills theory acts on the matter fields in the system.5. g ) = ( z . x ( z ) g ) some map x : M + G. and hb2 : Uz x F + P ~ l u ~ together with the pair of principal bundle local trivialisations hu1 : U1 x G + Pp1 and hu. .3. ASSOCIATED BUNDLES 239 It is easy to see that h' o u' = idpFIU and u' o h' = iduxF. w ] )[(z. In particular. : (U1 n U Z )x G + (UI n U z ) x G.: 0 hu.These functions guiu. If U1 and UZ are a pair of neighbourhoods of z E M over which [ is trivial.x(z)g).w] using the = or. prl. w ] ) = (z. there will exist a pair of corresponding maps h& : U1 x F + P ~ l u . on the associated for bundle ( [ F ] we have u ~ ( [ ( z . M ) is the product bundle. as a bundle. if ( = ( M x G. Then.
T.sl) := Pg. 6. the crosssections of such a bundle are the fields for a ‘twisted’ nonlinear amodel. Suppose there exists a bundle map u from P(q) to P(E) that is equivariant with respect to the subgroup H . Then G is itself a left Gspace under left multiplication.. FIBRE BUNDLES 5. Let E be any principal Gbundle. it is in fact a principal Gbundle. Then. u(ph) = u(p)h for all h E H c G.3. we can form the associated bundle q[G] = P XH G which is a bundle with fibre G. More generally] if H is a subgroup of G and if q is a principal Hbundle. [ g I H ) G ) := b g ] H (5. . and hence constructing P X G G / H . where H is a closed subgroup of G.240 CHAPTER 5.e. in physical terms. Note that.3.3 Restricting and extending the structure group The example abovewhere a principal Hbundle q was converted into a bundle q[G]with fibre Gis of considerable importance and merits further discussion. Definition 5.M) any principal Gbundle.14 Let H be a closed subgroup of G and let q and be a principal Hbundle and Gbundle respectively.10) k ( b ] H ):= ( p .9) whose inverse k : P/H + P XG H is given by (5. a very important is class of associated bundles is obtained by considering the action of G on the coset space G / H . [ e ] H ) G . If = (P.3. defined over the same base space. defined by j ( ( P . 5. This is closely related t o the bundle ( / H whose total space is defined to be the orbit space P/H of the Haction on P(E) and whose projection map onto M is defined as 7 r ’ ( [ p ] ~ := ~ ( p )There is an Misomorphism j : P X G G/ H + P/H ) . The ensuing associated bundle J[G] is Misomorphic to E by the map defined by i : P X G G + P with i(b. i. As we shall see below.
T . in YangMills theory it is related to the question of the spontaneous breakdown of the internal symmetry group from G to H . More generally. ~ ( p := [ q . v)G. Now we must return to the nontrivial matter of deciding when a principal Gbundle admits a restriction to any specific subgroup H c G. u ( p ) := b.7 A principal Gbundle J = (P. In differential geometry. Thus the Gextension of an Hrestriction of a bundle is isomorphic to the bundle. and if F is a left Gspace (and hence. Then there always exists a Gextension of 7 for any group G that contains H as a closed subgroup: simply form the bundle P X H G mentioned above and note that this can be given a Gbundle structure by defining [p. there is a bundle Mmap of q [ F ]with ( [ F ]in which ( 4 .lR)are in onetoone correspondence with Riemannian metrics on M (see below). Let J be any principal Gbundle with an Hrestriction q. Comments 1.3.7r. a fortiori. reductions of the structure group GL(rn. v ) is~ mapped to (4. Then a principal map is defined by u : P + P X H G . For example. M) has a restriction to a Hbundle i f and only if the bundle (/H over M has a crosssection.go]g := [p. g ] where ( g . (b) 241 t is called a Gextension of q. . a left Hspace).4 2. 3. Then there is a principal isomorphism of J with q[G] defined by L : P ( J )+ P ( q ) X H G . if t is a principal Gbundle with an Hrestriction q . gag].5. IR) of the frame bundle B(M) to the subgroup O(rn. Let q = (P. M) be a principal Hbundle. g ) E P ( q ) x G is any pair of ) elements such that p = u ( q ) g (the map L does not depend on the choice) and where u : P ( q ) + P ( J )is the embedding map of q in J . This question is of considerable physical interest. ASSOCIATED BUNDLES (a) 7 is called a Hrestriction of J . A theorem of major importance is the following: Theorem 5.
and hence o*([) a principal Hbundle over M . Then define a crosswhere p is any section of J/H by o : M + P ( J ) / H .. let o : M section of < / H : + P ( J ) / H N P(J) X H G / H be a cross(5. . If F is a cell then all its homotopy groups vanish. P ( J ) / H ) . . FIBRE BUNDLES Let u : P ( q ) + P(J)be a restriction map.2. This follows from the theory discussed briefly in Section 5. 2 . = k 1 .11) P(J) M * l p P(J)/H If p(p) := [ p ] we have a principal Hbundle [ := (P(J). Comments 1.}). then G / H is always a cell.o(z) := [ u ( p ) ] ~ element in T. . Thus. we define P(7) := { ( x . 2 . p ) := p . We shall see an important example of this shortly. Finally.. Assuming that M is paracompact it can be shown that any fibre bundle whose fibre is a cell (ie. as the bundle is space of the Hbundle q being sought. In particular. . it is homeomorphic to IR" for some n ) always admits a crosssection (for example. 2. define the bundle embedding map QED u : P ( q ) + P(J)by u ( x . Conversely.'({.. then the structure group of any principal G bundle over M can be reduced from G to H ..3 where it was argued that the obstructions to constructing a crosssection of a fibre bundle lie in the cohomology groups H'"(M. p ) E M x P(J)1 o(z) = [ p ] H } . Specifically.rkI(F)). If these homotopy groups are nonzero then the potential obstructions . and if the homotopy groups of G / H satisfy n i ( G / H ) N 0 for all i = 1 .rn1. and hence so do the potential obstructions. This example can be generalised in a very significant way.242 Proof CHAPTER 5. ~ p.3. it follows from obstruction theory that if M is an rndimensional differentiable manifold. Hence a bundle whose structure group G is noncompact can always be reduced to its maximal compact subgroup. if H is the maximal compact subgroup of any noncompact Lie group G. see Husemoller (1966)). .
.. i = 1.Z)N Z.3. there are no obstructions to reducing SU(2) to { e } . Since the cohomology groups H i ( M ) vanish when i > 4. 7r3(S3)) = H4(M.5.rri(G/H). . and hence there is no obstruc1 tion to reducing the group from SU(3) to SU(2). . . on a threedimensional manifold C. This comes from the nonvanishing cohomology group H 4 ( M . notice thatsince S U ( 2 ) N S3 and 7r3(S3) N Zthere is a potential obstruction to reducing the structure group from SU(2) to the trivial group { e } (ie.33) that SU(3)/SU(2) N S5. in general. (4. .7ril(G/H)).3. m . This is of relevance when considering the canonical quantisa0 tion of YangMills theory on the physical threespace C.2. in fourdimensional spacetimes. there are no topological properties of SU(3) YangMills theories over and above the ones that already arise for the group S U ( 2 ) . M ) over an arbitrary fourdimensional. there is an obstruction to constructing a crosssection of the principal SU(2)bundle).3. 4. the only nonvanishing obstructions to reduction must come from . r*(Sn)1 0 for i < n. the instanton number is precisely the value of this integer. (Recall that if M has N dimension m then H*(M) 0 for all i > m. But. i = 1 . An example where the reduction can take place arises in considering an ‘instanton’ YangMills principal bundle ( P . This shows how Riemannian metrics arise within the context of the bundle of frames. For example. In this context. compact (and therefore ‘Euclideanised’) spacetime manifold M . Using the same line of argument one can see that. ASSOCIATED BUNDLES 243 to constructing a crosssection of <[G/H]and hence of reducing the group from G to Hare represented by elements of the cohomology groups Hi(M. in fact.4. Then we know from Eq.) 3. This is basically the reason why. 5. suppose that G = SU(3) and we are interested in the reduction to the usual SU(2) subgroup.4 Riemannian metrics as reductions of B(M) Now we come to an example of great significance in differential geometry.7 r . 2 .
r'j. and define O ( M )to be the subset of B ( M ) made up of all sets of orthonormal basis vectors at all the points of M .14) (iii) g ( X . Proof A Riemannian metric g is defined to be a symmetric..R). 2)= r g ( X .15) thus defined is clearly bilinear .e.8 There is a onetoone correspondence between the Riemannian metrics o n a n mdimensional differentiable manifold M and the reductions of the structure group of the bundle of frames B ( M ) f r o m GL(m.3. Suppose first that g is a Riemannian metric on an mdimensional manifold M .. (5. (5. R ) and let O ( M ) denote the reduced bundle. There is also an isomorphism of this associated bundle with the tangent bundle T M via the map in Eq. FIBRE BUNDLES Theorem 5.R)t o O ( m .R) to O ( m . and define (5. i.ri] ++ C p l r r i b i . Combining these two maps we can evidently regard Lb as a map from R" t o T. This is clearly an O ( m .e.1). Thus.3.R) R" defined by ~ b ( q:= [b. If u . where x E M is the point at which b E B ( M ) is a base. L~(F')= CEl ribi.3. 2covariant tensor field. Now regard O ( M ) as a subbundle of B ( M ) . s E R(5.3. [b. suppose that the group of the principal bundle B ( M ) is reducible from GL(m. The object Eq. X ) (5. positive definite. v E TxM choose a base b E 7r& ({z}). 2)for all T. i. if X .244 CHAPTER 5.15) where ( . 2 are any vector fields on M we have (i) g(rX + sY. IR) preserves the orthonormality) and the required bundle map O ( M ) + B ( M ) is simply the subspace embedding. X ) 2 0 and = 0 only if X = 0. For each frame b E B ( M ) there is an injection Lb : R" + B ( M ) XGL(". Y )= g ( Y .2)+ sg(Y. ) denotes the usual Cartesian inner product on the real vector space R".12) (4 g ( X .R)bundle (since this subgroup of GL(m. Conversely. Y .3.M.13) (5.3.
IR) to the noncompact group O ( m.IR)/O(m1.IR) transformations of the inner product on IR” means that Eq. the coset space GL(m.F2 E Rm.1.f3. j = 1 .15) is independent of the particular choice of the base b for T. 2 . bz. If b is extended to a local section of B(M) then the set of local vector fields { b l .. and the invariance under O ( m . m.).I )is the maximal compact subgroup of GL(rn. .M [Exercise!].3.. 3. .which corresponds to finding a Lorentzian signature metric on M . (5. Thus we have acquired a welldefined Riemannian metric on M ..3. g.5.} is called an rnbein and we have. QED Comments 1. If b E B(M) then (TI.7 5 ) = g. 1475)= Cz=.IR)/O(rn.IR). . l). We can also consider the reduction of GL(rn.l) is not a cell. and hence to find Riemannian ) metrics on M . bj) If b is a base at z E M then bi = C~lbr(z)(d.(bi.16) where the components g p v ( z ) of g with respect to the local coordinate system are Hence the components of the metric thus defined are related t o the components of the base fields by w h e r e i . Since O ( m .. . and hence there may be topological obstructions . j = l cc m m gpv(z)b:(z)bj”(z)rf. 1 z (FlY2) = .(L&.3. (5.F E IR”.R) R it follows from the earlier discussion of obstruction theory that it is always possible to construct crosssections of the associated bundle B(M) x q m . However. 2. ASSOCIATED BUNDLES 245 and positive definite.. for all Fl. b.p=l i . for all 7 . ~G L ( m .
F .~ ~ ( P ) ] b. FIBRE BUNDLES to the construction of such pseudoRiemannian metrics. T . 5.246 CHAPTER 5.4(P)l 4 i s defined (5.(b(pg)I = b g .5 Crosssections as functions on the principle bundle Finally we come to a result concerning crosssections of associated bundles that has been widely used in induced representation theoryin which vectors in the Hilbert space are crosssections of certain vector bundlesand in studies of anomalies in quantised gauge theories.~]. ) Conversely. Proof Given 4.20). Ms a n associated fibre bundle then its i) crosssections are in bijective correspondence with maps 4 : P ( [ ) + F that satisfy 4(Pd = gl4(p) S+ for all P E P ( 0 (2nd 9 G.' ( { x } ) and L~ : F + nF1({x}) is the injection of F into PF (and onto nF1({x})) defined previously by ~ ~ ( 2 1 := lp. let s : P ( J ) + F by M + PF be a crosssection and define 4s : 4s(P> := q 1 ( s ( 4 ) (5. Furthermore.3. (5.3.3.3.20) where p is any point in n'({x}).3. This is independent of the = choice of P E n'({x}) since bg. (5. n ~ ( s + ( x =)n ( p ) = x. define s+ by Eq. In fact. s . Theorem 5.19) by The crosssection corresponding to such a m a p s + ( 4 := [P.21) where p E n . this coset space is homotopic to the real projective space RPm'which has many nonvanishing homotopy groups. ) . This topological property was exploited in the theory of 'metric kinks'one of the earlier examples 0 of the use of topological methods in theoretical physics.9 If ( P F . 4(p)l. and hence s+ is a crosssection.
26) 2.25) and hence.27) and so the local representatives s g and sv2 are related by the gauge transformat ion SUl(.3..w] o~ L.)>l. note that the maps each other since 4 e s+ and s e q5s are inverses of = [P. 4s(P)l 44 (5.(S(Z)) =.g’w] = Lpg(gP1w). w Comments 1. (5. ( )) Finally. 4s(fl(zL. &)I) = 4.22) implies that. ’ we see that s z and su(z) are related by () (5. as required.w) := [a(z).3. Eq. Now. (5.3. for all z E U .and ~ ( $ ~ =pgl+s(p).for all z E Ul n U2. . ASSOCIATED BUNDLES Now.3.3.g) := a(z)g. L ~ ( W )= 247 hence 4.(P> = Lp1(s+(4) = Lp’([P.22) (5. h’(z. h(z. the local representative su : U + F of a section s of PF is defined by s u ( z ) := 4s(a(z)).g) = a1(z)g and h2(z. If a : U c M + P ( J )is a local trivialising crosssection of J. Note that the associated trivialisations satisfy hl (z.> =R(+s(4. using the local trivialisations h : U x G + r’(U).23) 4s. (5. g ) = h2(01(4g) = &l(a2(z)fl(4lg) = (z. Hence hZ1 0 h ( z .5.28) for all z E Ul n U2.3. Lpl(+))l = sdz) = [P.g) = az(z)g. = Then s u 2 ( 4 = 4s(a2(4) = ds(al(z)R(z)) = ~ ( + $ s ( ~ l ( 4 ) (5.(pg) = [P. = [Pg.R(z)lg) (5.3.29) and so n(z)is just the transition function gulu2(z)discussed earlier. (5.w].1 . and h : U x F + r F 1 ( U ) . then there exists some local ‘gauge function’ R : Ul n U2+ G such that a2(z) al(z)R(z).3. If a1 : U1 + P and a2 : U2 + P are two local sections of P with Ul n U2 # 8.3.3.24) 45) = [+>.
for each z E M there must exist some neighbourhood U c M of z and a local trivialisation h : U x IR" + r . sl. A vector bundle map between a pair of vector bundles ( E .T . the map h : {y} x IR" +n'({y}) is linear. it can play a role analogous to that of the more familiar linear space of functions on a manifold. a vector bundle is simply an associated bundle in which the fibre happens to be a vector space. it is common to present them in a framework that makes no direct reference to an underlying principal bundle. Furthermore. because vector bundles arise so often in practice. From one perspective. . We shall start this section with this approach.1 Vector Bundles The main definitions Vector bundles are of considerable importance in theoretical physics because the space of crosssections of such a bundle carries a natural structure of a vector space. as such. the associatedprincipal definition will be given at the end.M ' ) is a bundle map (u. for all y E U . in which the restriction f) of u : E + E' to each fibre is a linear map. An ndimensional real (resp.M ) and ( E ' . r ( E ) has the structure of a real vector space in which the real numbers are represented by constant functions on M . 2.s2 E r ( E ) (b) (fs)(z):= f(z)s(z)for all x E M . realvalued functions on M .15 1. Definition 5.' ( U ) such that. M ) is equipped with a natural module structure over the ring C ( M ) of continuous.4 5. complex) vector space. 3.248 CHAPTER 5.s E r ( E ) . The space r ( E )of all crosssections of a vector bundle ( E . complex) vector bundle ( E . defined by: (a) (sl + s2)(z):= sl(z)+ s2(x) for all z E M .4.7 r . therefore.7 r . f E C ( M ) .T'. FIBRE BUNDLES 5. M ) is a fibre bundle in which each fibre possesses the structure of an ndimensional real (resp. However. In particular.
rndimensional submanifold of R" for some n. The projection map is 7r : E(y. 3.. . and if U' is the corresponding set in IRP".2 ( 2 .4. e.. Let M be a smoothly embedded. a onedimensional vector bundle) over the projective space IRP" with total space E(y. then a local homeomorphism h : U' x R + T'(U') can be defined by h ( [ z ] .2) in which denotes the line that passes through z E IR"". for all G E T. The same is true of the cotangent bundle T M .2 Vector bundles as associated bundles Let us now show that vector bundles can also be discussed within the associatedprincipal bundle framework introduced in the last section. The tangent bundle T M of a differentiable manifold is a real vector bundle whose dimension is equal to the dimension of M . VECTOR BUNDLES 249 In the case where all the spaces are manifolds.1)sphere in IR". Examples 1. ~E IRP" x Wf1 1 v ) = Xz for some X E IR} (5.4. := ([z]. There is a 'canonical' real line bundle 3.) := and projection map E 7r MxlR" I v'w'= 0 .Ax). the space of crosssections is a module over the ring Cm(M) smooth functions of on the base space M . 2.5. Local triviality is demonstrated by noting that if U c S" is any open set that is sufficiently small that it contains no pair of antipodal points. .4. For any ndimensional vector space V. thus 7r'({[~]}) is the line in R"" that passes through z (and x). An example is the normal bundle associated with the embedding of the ( n . A) [XI 5. u ):= [z]. the product space M x V is an ndimensional vector bundle over M . Then the normal bundle of M in IR" is an (nm)dimensional vector bundle v ( M )over M with total space E ( v ( M ) ) {(z.) := { ( [ z ] .4.1) : E(v(M)) + M defined by ~ ( z5):= z. 4.) + IRP" defined by T ( [ z ] .M} (5.
IR. To give T& ({x}) a vector space structure. It is clear that the restriction h' : {x} x IR" +7r&i({x})of the local trivialising map h' is linear for each x E U C M . .10 Let ( = (P.'({4>. QED Comments 1.w] where h : U x G L ( n . is a homeomorphism from R"onto ?r& ({x}). v] where p E 7 r . R ) act on R" in the usual way. and the is associated bundle ( [ V ] a vector bundle.R)+ e ?r'(U) is a trivialising map of P over U c M . More generally. finitedimensional vector space we can consider the group GL(V. ' LP'(Ui) + LP'(zIh) = Lp'(Ui  + glv2) Lpg(Sl(v1 + 4)= L p ( U 1 + v2) = LP(V1) + L P ( U 2 ) .' ( { x } ) .R) of automorphisms of V . To see that this bundle is locally trivial we define a map h' : U x IR" + ?r&(U).250 CHAPTER 5.4. all X E := for If p' E ?r'({x}) is any other choice such that ~ ~ ' ( v '= ) some v E R" then. for some g E G L ( n . L ~ ( Z I ) for (ii) h P ( v ) L ~ ( X Z I )for all v E R"..R ) . Proof The map L~ : R"+ ~ $ ( { x } ) L. + =I ) ? Lpg(glvl .?r. if V is any real. In particular. h'(z. FIBRE BUNDLES Theorem 5.v) := [h(x. the bundle associated with a linear representation of the group G L ( m . Thus all tensor bundles are vector bundles.M)be a principal GL(n. ) . if 5 is the bundle of frames on an mdimensional manifold.3) and hence the vector space structure is independent of the choice of P E . choose any p E ?r'({x}) and define (i) bP(vl) + ~ ~ ( 2 1 := h P ( w l 2) + v2) for all ' ~ 1 . ~ ( W ):= lp. 2 1 2E R".R)bundle and let G L ( n . Then the associated can bundle ([R"] be given the structure of an ndimensional real uector bundle. (5.R) on any vector space V gives rise t o a vector bundle.
If 5 is a principal G L( n .4. 3. : f) 4. : 5 + 5' is a principal GL(n. every vector bundle is bundle isomorphic to an associated bundle of this type. VECTOR BUNDLES 251 2. . The converse statement to the one in the theorem is also true: namely.IR)bundle over M and f is a function from M' to M . (uF.5. If (u. hence vector bundles pullback to vector bundles.R)map then f) 5[IR"] + ['[IR"] is a vector bundle map. then the pullback bundle f*(5[IR"]) admits a natural structure of an ndimensional vector bundle over M'.
.
The obvious natural vector fields on P are those that arise from the right action of G on P. In the case of Riemannian geometry on a manifold M . E L(G). (4.5. The basic idea of parallel transport/covariant differentiation is t o compare the points in ‘neighbouring’ fibres in a way that is not dependent on any particular local bundle trivialisation. (6.1)) which represents the Lie algebra homomorphically in the sense that (Eq. the wellknown Christoffel symbol is essentially a connection on the bundle of frames B(M).5. Eq. B X”] ‘A classic reference is Kobayshi & Nomizu (1963). According t o the discussion in Section 4.27)) [X”. the YangMills field can be identified with a connection on the bundle space P. For a YangMills theory with a principal bundle G + P + M . to each element A € L ( G ) there corresponds a vector field X A on P(cj.1.1.1 6.Chapter 6 Connections in a Bundle 6. This suggests looking for vector fields on the bundle space P that ‘point’ from one fibre to another. = X[””] for all A .5.1) 253 . together with the associated concepts of ‘parallel transport’ and ‘covariant differentiation’. (4.1 Connections in a Principal Bundle The definition of a connection An idea of great significance’ in fibre bundle theory is that of a ‘connection’.
the vectors Xp” are tangent t o the fibre a t p E P . p E P (6. = dim L( G ) = dim G. This motivates the following definition of a ‘connection’.e.1 A connection in a principal bundle G + P + M is a smooth assignment t o each point p E P of a subspace HpP of TpP such that (a) TpP N V. Comments 1.P for all p E P . these particular fields are not suitable for our purposes since they do not point from one fibre t o another.e.1.254 CHAPTER 6.P 1 T*T = o} (6..4) (b) hg*(HpP) HpgPfor all g E G..5) = where h g ( p ):= pg denotes the right action of G on P (the meaning of the word ‘smooth’ in this context is explained below). Definition 6. What is needed is some way of constructing vectors that point away from the fibre. and hence they point along the fibre. i. rather than away from it. i.1. Xp”is said to be a vertical vector.P @ H. the map L : L(G) + VFlds(P). However.3) where T : P + M is the projection in the bundle. The equation (6. in particular. Technically.1. On the contrary. P .1. elements of TpP that complement the vertical vectors in VpP. dim V. It is easy to see that the map A ++ Xp” is an isomorphism of L ( G) onto V .4) implies that any tangent vector T E TpP can be decomposed uniquely into a sum of vertical and horizontal . A tj X A (6. (6.2) is an isomorphism of L ( G ) into the Lie algebra of all vector fields on P. CONNECTIONS IN A BUNDLE Furthermore.1. since the action of G on P is free. it belongs t o the vertical subspace VpP of TpP defined by v p P := { T E T.
e. .these components will be denoted by ver(r) and hor(r) respectively.) := Ll(ver(r)) (6.6.1. ver(X).5) implies that the verticalhorizontal decomposition of the tangent spaces of P is compatible with the right action of G on P .10) 5. 3 A. i. (6. 2 .8) (b) bg*w = Ad.P.6) Then Eq.+1 is said to be an exact sequence if.9) = (c) r E HpP if and only if up(.1. (6.1.1 we have Ker(ai+l) = Im(ai). then so are ver(X) and hor (X) . for all i = 1 .1. .define up(.P +T~ P T* M +O . An exact sequence of the form 0 % B C % D +0 (where 0 denotes the trivial vector space with just one element) is said to be a short exact sequence. A E L ( G ) = (6. If r E T p P . (6.2). For all p E P .1. This introduction of a connection oneform is very useful: in fact. . 2. . .lw. Thus: (a) wP(XA) A V p E P .1. A connection can be associated with a certain L(G)valued oneform w on P in the following way.) = 0. (6. a vector field X on P gives rise to a pair of vector fields ver(X) and hor(X) on P with the property that.1. E H.4) shows that : HpP + T.P induced by Eq.7) where L is the isomorphism of L(G) with V. an alternative way of defining a connection is to say that it 2A sequence of maps and vector spaces A1 3 A2 3 . the projection map T : P +M induces a map rr* : TpP + TT(. E VpP and hor(X).. Similarly.VT E TpP(6.1. for all p E P . (Ss*w)p(~) Adgl(wp(r)). 3.)M whose kernel is VpP. . The equation (6. 4.rt .)M is an isomorphism.(.1. CONNECTIONS riv A PRINCIPAL BUNDLE 255 components lying in VpPand HpP. The word 'smooth' in the phrase 'smooth assignment' in the definition means that if X is a smooth vector field. Thus we have the short exact sequence2 o + V.
.f o r)wZ for any smooth function f E C m ( M ) . corresponding to the fact that A transforms according to the adjoint representation of G under rigid (i. we see that the YangMills field can be regardedat least locallyas a Liealgebra valued oneform on M . then so is the affine sum (f o r ) w l (1 . . Let h : U x G + .1. (6.1. if we write (6..e.2 Local representatives of a Connection Now we come to the crucial question of what a connection oneform w ‘looks’ like. which leads once more to the directsum decomposition of T. The precise relation between this field and the oneform connection on the bundle space P . . . Thus.9). . (6.1.11) p=l a=l where { E l . + 6. Define the local arepresentative o f w to be the L(G)valued oneform wu o n the open set U c M given by wu := a*w. In particular this is true for constant functionsie. .dim G. E2. A horizontal vector is then defined to be one that satisfies Eq. 2 .. spacetime independent) gauge transformations. If w1 and w2 are a pair of connection oneforms on P . ‘ U C P be the local trivialisation of P induced by d according to () h ( z ..1.1 Let a : U c M + P be a local section of a principal bundle G + P + M which is equipped with a connection oneform w .g ) := a ( s ) g .1. CONNECTIONS IN A BUNDLE is a L(G)valued oneform on P that satisfies Eqs. the set of all connection oneforms is a cone n the real vector space of all L(G)valued oneforms on P.256 CHAPTER 6. (6. . 6. A YangMills field is usually written in the form A.10).1. is given by the following theorem. Theorem 6. where p is a spacetime index and the index ‘a7ranges from 1 . and how this is related to YangMills theory.4).86. .P in Eq. . E d i m ~ } is a basis set for L(G).
1.. Putting together these results with Eq.3.byEq.6 = L f for some A E L(G)in fact. (4.g)(a. (h*w)(z.15) S o j p ( g ) = b(p. we have P u ( z ) * ( L t ) X&.g)(U G ) II TzU @ TgG (cf. ie. i f ( c Y .)(% P ) = (6*4(u(z).114.1.1. Proof Factor the map h : U x G + P as UxG (x7 9 ) Then. Eq.g> 6 . Now.e. (6.P) E TzU @ T.13). (h*W)(z.5. ) x the local representative h*w of w on U x G can be written in terms of the local 'YangMills'field wu as = (h*w)(z.. (.6.)g)(o*a) P + Wu(z)g(Pu(z)*P).1.3. (6. 9 t) ( P J ) .(2. (2. b o i g =ag: P+ P (6. (6. and of course w ( X A )= A E L(G)..g)(O*% P> (6 0 j u ( z ) ) * P ) (6.1.3.73)). ~ E T(z.17) we finally obtain: (h*w)(z. (4.1.g)(% dxlQ t)  PxG (+>.1.14) .13) P) = ((fl x id)*6*4(. We also know that . I )  P 4 4 9 (6. Furthermore.) = Adgl(wu(z)(O*a)) Eg(P) P + = Adgl(W. (6.g)(a.17) = .1.ig:P+PxG.W u ( z ) g ( ( S0 i g ) * m * a + where.g). bg*wu(z)g Adgl ( w u ( z ) ) according to the characteristic property Eq.69).G.18) QED .3. A = Eg(P) according to the definition of the CartanMaurer form given in Eq.U(4)+ %(P) (6..12).pe(p.)g.1.16) Therefore.u(a)> + =:.. 6ojp=Pp:G+P. i. CONNECTIONS I N A PRINCIPAL B U N D L E 257 Then.g)(% ) = (bg*wu(.11).1.9) of a connection oneform. ac= cording to Eq.g)=pg.andj:G+PxG. g ) = p g .(6.(PI for all ( a .P) Adg1(w.3. so that b o i g ( p ) = b ( p .12) where Z is the CartanMaarer L(G)valued oneform o n G defined a n Eqs.
There are two answers. If f is close enough to the identity map that U fl f ( U ) # 8. In general.19) between the components Xp and X”’ of a vector field with respect to local coordinate charts whose domains are U and U’ respectively.1. ‘passive’ view of the gauge group. This can be related to the ‘active’ view by noting that if (U7q5)is a coordinate chart on M . whose subtle difference depends on whether the gauge group is thought of in an active way or a passive way. In the YangMills case. and if f : M + M is a diffeomorphism then a new coordinate chart can be defined as the pair ( f ( U ) . and if w is a connection on P . it seems relevant to ask how a local representative wu is affected by such a gauge transformation. This raises the interesting question of how the familiar gauge transformations arise in the fibre bundle formalism.1.19) can be applied for all z in this intersection.1. (3. CONNECTIONS IN A BUNDLE 6. then Eq. We shall answer this question in the course of considering the alternative.(X). The analogue of this in ordinary differentiable geometry is the relation defined on z E U f l U’ (see Eq. it is easy to check that @ ( w ) is an L(G)valued oneform that satisfies all the axioms for a connection. a gauge transformation in the principal bundle G + P + M is defined to be any principal automorphism of the bundle. If q5 : P + P is such a map. with a similar definition for any other tensor field on M .14)) (6.q5 o f’). I shall leave as an exercise [!] the task of investigating the exact relation between . This active view is analogous to the way in which a vector field X on a manifold M can be defined to transform under the action of a diffeomorphism h : M + M as the pushforward h. (6.3 Local gauge transformations The discussion above shows us that a connection oneform w can be decomposed locally as the sum of a YangMills field on M plus a fixed L(G)valued oneform on G.1.258 CHAPTER 6. Henceat least locallyspecifying a connection is equivalent to giving a YangMills field. The pullback @ ( w ) is said to be the gauge transform of w by the gauge transformation 4 E Aut(P).
1. Let A.20) the local representatives are related on U1 n U2 by Proof If 8. (6. can be rewritten as .1. CONNECTIONS IN A PRINCIPAL BUNDLE 259 the active transformation X t) f * ( X ) and the corresponding local coordinate transformation Eq. denotes the local vector field then A ~ ) ( x ) (u~w).1. := and using the relation Eq.22) I+ (ul(4. by the same type of argument used in the proof of Eq. nu. ++ P cJl(z)qX) (6. Theorem 6. (6.2 Let w be a connection on the principal bundle G + P + M and let u1 : U1 + P and u2 : U2 + P be two local trivialisations on open sets Ul.1. l ( Z ) W . + p a s & ulnu2 X n + u1 x P ~ G 6 .20) to factorise the local crosssection g2 :u. (6.) and A) denote the local representatives of w with respect to u1 and .19).(8. (6.1. c2 respectively. U2 c M such that Ul n U2 # 8. The precise statement for the YangMills case is contained in the following theorem.qX)) we get which.12).6.).2(4 =.1. Then if IR : Ul n U2 + G is the unique local gauge function defined by .
25) 1.1.1. 4.3. if the bundle is trivial. CONNECTIONS IN A BUNDLE If G is a matrix group. (6. (4. then an active gauge transformation 4 : P + P induces a transformation A t) u*(c#Pw) = (4 o u ) * w .U j will be related on U i n U j by Eq. 3.1.260 Corollary CHAPTER 6.16) as Af)(z) Comments = fl(x)lAf)(z)fl(x) + fl(~)~'8~fl(z).2. it is not possible to describe the connection w in terms of a single YangMills field on M . (6. The relation with the transformation for a single field is as follows. it is.1. (6.a(x) = $ 0 u(z)fl(z).1.(s)Rl(z) + fl(z)8. Note that these functions Rij : Ui n Uj + G are precisely the bundle transition functions discussed in Section 5.26) Note that. if {Ui}iEI is some indexed set of trivialising open sets that cover M with corresponding local sections ui : Ui + P and .25) with the corresponding local gauge function f l i j (x)satisfying the relation ui(z)= uj(z)Rij(x).(z) t ) R(z)A. in a sense.R(z)1. Conversely. (6. and then the argument used in the proof of the theorem shows that the transformation of the local representative A can be written in the familiar form: A. this result can be rewritten using Eq. one must cover M with local trivialising charts. a crosssection u can be defined on all of M. 2. (6. Eq.26) refers to a globallydefined L(G)valued oneform on M . and then the local YangMills fields associated with any pair of overlapping charts Ui.25) looks just like the familiar YangMills gauge transformation which. If u : U + P is a local section of G + P + M with A := a * ( w ) . But note that it relates a pair of local YangMills fields whose regions of definition in M are typically different (albeit overlapping)it does not refer to a single YangMills field. Instead. But there exists some f2 : U + G such that. and then Eq.3. for all x E U . If the principal bundle is nontrivial.
28) Then where we have used the general result Eq.. for all z E U r l U’ (6.)..1. 2 . a ‘patching’ argument shows that a unique connection oneform on P is defined by any set of local L(G)valued oneforms A(i)on Ui c M that are related pairwise by Eq.1. .m} is some basis for the Lie algebra M ( m .4 Connections in the frame bundle An interesting example of the constructions above is a connection in the principal G L ( m .$’) such that unu’#0.R)(the set of all m x m real matrices) of GL(m. ..1. 6. $) on M provides a local section u : U + B(M) by associating with z E U c M the local . . .1.R). . m (a. If w .25). frame ((a/azl)z. and consider the relation between r and the local oneform I” associated with another coordinate chart (U’.1. .27) (6.R))valued oneform on U .6.25) for a connection in a bundle whose structure group is a matrix group. (6. = u=l JL(x)(du). (6.R) is just the Jacobian of the coordinate transformations: (a/adm). Any local coordinate chart (U.~). we can .X = 1 . CONNECTIONS IN A PRINCIPAL BUNDLE 261 i transition functions Rij : U n Uj + G... ( a / a ~ ~ ) ~ ) is a connection oneform on B(M) let r := u*w denote the associated L(GL(m. and the transition function J : U n U’ + G L ( m ..1. If {Gi 1 x. .R)bundle B(M) of frames on an mdimensional manifold M . The local frame associated with the local section u’ of B(M) is z ++ ((a/~z”).
Definition 6.1. as (. := 6.r) X.1.=. .262 CHAPTER 6.2 Parallel Transport 6.) = (b) ver(XJ) = 0. (6.P + T. there is a technique for generating horizontal vector fields.). the transformation in Eq. Eq. (4 T*<X.. t o each vector field X on M there exists a unique vector field. CONNECTIONS IN A BUNDLE write the matrixvalued oneform r. as in the transformation Eq.25) in the analogous case of YangMills theory. (6. then Eq. for all p E P . e. 6. (6.29) becomes the wellknown transformation law for the components rphC an affine connection on M : of Note that. denoted Xr. (6. if we pick the natural basis set (G. l?.2..30) is valid globally only if there is a globallydefined coordinate chart.XX(Gt).T . on P such that. which for a generic manifold is not the case. In particular. First.1. This vector field X r is known a s the horizontal lift of X.. fields whose flow lines move from one fibre into another.1 Parallel transport in a principal bundle We can now describe what is meant by 'parallel transport' in a principal bundle ( = (P.1. i.@)M is 'an isomorphism. In particular. = Cy.30) is only defined for points x E M that lie in the intersection of the two coordinate charts involved. I) ? .2 Since 7r* : H.M ) equipped with a connection w .6.
Horizontal lifting has several pleasant algebraic properties: (a) (Xr +'Y') = (X Y)' + (b) If f E C " ( M ) . b ] ) . PARALLEL TRANSPORT 263 Comments 1. (6.(hor[Xt.'>= qg 2. The act of horizontal lifting is Gequivariant in the sense that &*(X.31) for 'hrelatedness' we see that T.rr(a'(t)) a ( t ) for all t E [a.2.Y'] = [ X .b] + P which is horizontal (ie. A horizontal lift of a is a curve at : [a. Then Y = X t .b] c I R into M (ie. 3. ver[a'] = 0) and such that . and g E G since we can then define unambiguously X.rr*([X'.Y] .(Yp) any p E for T .2..Y]' Definition 6.Y']) = [. (6.b]. a is the restriction to [a.rr*Xf.6. Yr]) = .3 = hor([X'.1) from which it follows that [X. := .rr.rrXr (c) Using the general rule Eq.rr. (3. then (fX)T = f o .2..2) Let a be a smooth curve that maps a closed interval [a. A sufficient condition for a vector field Y on P t o be the horizontal lift of a field on M is that (a) ver(Y) = 0. and (b) hg*(Yp)= Ypg for all p E P .b] of a smooth curve defined on some open interval containing [ a . = The key to understanding parallel translation is the following theorem concerning the existence of horizontal lifts of curves: . Y?]).' ( { x } ) .1.
a horizontal vector field).4) where the last equality follows since ac*w is a connection oneform on P ( a * ( ( ) )and [PI is horizontal (because P is the integral curve of . But ) of this is clearly unique and equal to 0.264 CHAPTER 6. (6. lift of a. + P is any other horizontal lift of a . Clearly ~ ( a T ( t = a ( t ) (so that )) a f ( t )E P(a*(<))) and w a t ( t .b].E .tj (a?' o ~ ) ( t:= ( t . Suppose that P : [a. QED Comments 1. ( [ ~ f l ) waqt. it can be pulled b] back by at to give the curve in P ( a * ( ( ) ) .' { a ( a > }there exists a unique . In practice. but nonhorizontal. horizontal lift of a such that a f ( a ) = p . Proof Extend a to the open interval I := (u . and a t * ( w ) is a connection is oneform on P ( a * ( ( ) ) [Exercise!].3) I 1 * M 3 . Thus af is horizontal. ~ ( t to) P(a*(<)) the curve t t) t in I is horizontal.y ( t ) ) t ) with a t ( t .3 For each point p E ~ .2. P a&p) :=p.' o 4) 0. Let : I + P ( a * ( ( ) ) the integral curve that passes through be ( a . + G b] . p ) E P(a*(<)) 1 x P of the unique horizontal lift of the vector c field d / d t on I . and so the lift = t t+ ( t . CONNECTIONS IN A BUNDLE Theorem 6. this usually means making a comparison between the horizontal lift atand some other 'natural'. + P(a*(<))aE. Then there exists some unique function g : [a.2.(bt = O P )= (JJar(t. If y : [a. y ( t ) )= y(t). Hence (ac*w)([a.(at*[PI) ( " t * W ) P ( t ) ( [ P I ) = 0 I = (6. b E ) for some E > 0. In order to make full use of the ideas above (for example. . Then a*(<) a principal Gbundle over I . Therefore cxf is unique. when defining covariant derivatives) it is necessary to have as explicit an expression as possible for the horizontal lift at of a curve (Y in terms of the connection oneform w . Define af := a€o P. ~ ( p ( t ) ) a ( t ) = for all t E [a.b] + P is such a curve. i e .
10) can be reexpressed as the matrixvalued integral equation * .9) becomes where x p are the local coordinates in this chart.2.2. If u : U + P is a local section of 6 associated with a coordinate chart U C M . If the boundary conditions on t g ( t ) are g ( a ) = go E G.[a]. (6. which we have But u*w identified earlier with a YangMills field A.b]. Eq. Eq. * P P(t>dt> (6. ( ~ ) ( [ a ] ) . In terms of this field. and hence ~ p ( ~ l ( [ P ] ) = ( a * ~ ) . a'(t) = P(t>s(t>.2.b] + U is p(t) := a ( a ( t ) ) Then [PI = a.5) [a.g(t)> 6 .6.2. a natural choice for a lift of a curve a : [a. PARALLEL TRANSPORT 265 such that. 2.2. This relation can be factored as (6. .b] t and then * PxG (P(t>.9) This is the differential equation that determines the horizontallift function g ( t ) in terms of the connection w. for all t E [a.6) (6.2. is the local representative w u . (6.
14) ) . If A is subjected to a gauge transformation of the type in Eq.1. If G is a group of matrices. 0 .(a(s))dip(s))go.(a(t))(P exp S.266 CHAPTER 6. .12) Thus the final local expression for the horizontal lift a of the curve t a : [a.(a(s))&p(s) + J. The analogous object in general relativity is a central ingredient in the 'loopvariable' approach to the canonical quantisation of gravity. This integral equation for g ( t ) can be solved as the pathordered integral g ( t ) = (P exp  S. (6.t d s A ~ ( ~ ( S i)y p ( s ) c) ) O(n(t))'(Pexp  t ds A p ( a ( s )b p ( s ) ) f l ( a ( a ) )(6.then Eq.)go.2.2.2((Y(S2))~~1(S1)~p2(S2) + .26) then it is fairly easy to see [Exercise!] that the pathordered integral transforms homogeneously as P exp  J. and if a is a closed loop (so that a(.2.. t (6.14) shows that Wa[A] t r (P exp := ds A. It is known as the Wilson loop function and plays an important role in many approaches to the quantisation of gauge fields. ds t ~.(a(s)) iYp(s)) (6.2.) = a ( b ) ) . and where summation of the dummy index p is understood. .b]+ U C M is a'(t) = c.t dsl lS1 ds2 A.))A. d s A. ds t A ~ ( ~ ( S cip(s))go ) ) := (1  S.13) 3.15) is a gaugeinvariant function of the gauge field A . (6. CONNECTIONS IN A BUNDLE where we have written the components of the tangent vector [a]as & p ( t ) := d z p ( a ( t ) ) / d t .2. 4.(~(s. (6.
.16) Thus we have a natural map from the loop space of M into G. and in general we get a nontrivial map from T'({(Y(cL)}) onto itself given by (6. it plays an important role in understanding the relation between the connection and certain topological properties of M . As remarked already. if pl = ~ ( pthen plg = T ( p ) g = T ( p g ) . for all g E G. PARALLEL TRANSPORT 267 We can now give a precise meaning to the concept of parallel translation: Definition 6. There is no reason why the horizontal lift should also be closed.2.' ( { ( Y ( ~ ) }obtained by ) ) associating with each point p E T . T O 6. Comments 1. 6. 2. The subgroup of all elements in G that can be obtained in this way is called the holonomy group of the bundle at the point a(0) E M .4 Let (Y : [a. a loop) in M . . In particular. This is performed by means of the following definition of the 'vertical' and 'horizontal' subspaces of a tangent space to the associated bundle.~ ( { c Y ( c L the point at@)E )}) ~'({a(b)}) where at is the unique horizontal lift of a that passes though p at t = a.2.2. = dg o 7 . Since a horizontal curve is mapped into a horizontal curve by the right action 6 of G on P we have.b] + M be a curve in M . The parallel translation along Q is the map T : ~ . which implies that ) T is a bijection of fibres.2 Parallel transport in an associated bundle The next important step is to extend the ideas of connections and parallel transport to associated fibre bundles.l ( { a ( a ) )+ T .6. an interesting case is when a is a closed curve ( i e .
This leads to the concept of parallel translation (or transportation) in the associated bundle as the map TF : 7rF1({a(a)}) 7rF1({a(b)}) + obtained by taking each point y E 7rF1({a(a)})into the point a.b]+ G obeys the differential equation (6. M ) .10).is defined as (cf.7 r .w] be any point in 7rF1({a(a)}).(af(t)) = [ a f ( t ) . F 2. Let a : [a.b]+ M and let lp. a left action of G on F . CONNECTIONS IN A BUNDLE Definition 6.19) is the horizontal lift of a to Pp that passes through lp. w] E P .B(t)g(t) where g : [a. Then the curve a.18) Comments 1. Let k. h ( z . y E PF. be defined by k. (6. g ) := a(z)g.M ) be the bundle associated to ( via . Let af be the unique horizontal lift of a to P ( ( )such that a'(a) = p .5 1. o ag = k.with inverse u : 7r'(U) + U x G defined as u ( p ) := ( 7 r ( p ) . w E F .2.2. w) in the equivalence class of 9 = lp. As usual. where t ++ a$(t)is the horizontal lift of a to PF that passes through Y.(PF). the horizontal subspace of the tangent space T I ~ .@) := k.(p) := lp. I ( P F )independent is of the choice of elements ( p .B(t) := o ( a ( t ) ) . this local section generates the trivialisation map h : U x G + 7r'(U). . the natural choice for a lift of a to P(J)is .. and let ( [ F ]= ( P F T F . Eq. Since kgl.3)) Then 2. Let w be a connection in the principal Gbundle ( = (P.@).~].2. 3. and then a f ( t )= . : P ( ( ) + PF. . If CT : U + P ( ( ) is a local section of the principal bundle. The vertical subspace of the tangent space T. x g ( p ) ) where xg : P + G is the . the definition of H ~ . ~ ( P F ) is defined as Hg.v] P F ) := (H p P ) ( (6.268 CHAPTER 6. w] at t = a.1. v ] (6.
when the bundle is equipped with a connection this can be used to 'pullback' the fibre over the second point to the fibre over the first. the ' + image of (rL(t) is the curve t t)(a(t). (6. be a curve in M such that ( ~ ( 0 = 20 E M .T .w) at t = 0 is given by Eq. Then the covariant derivative of $J in the direction (Y at 20. is < where rb is the (linear) paralleltransport map from the vector space xV'((a(t)}) the vector space T~'({Q}). (6. Then the image of (Y? in U x G is 21 O 4 t ) = (a(t>.g(t)). xu(P(t)))s(t) = (a(t>.2.' ( U ) . these values lie in two different fibres and although they could be compared by using a local bundle trivialisation. E > 0. E ] + M .g(t)w) E u x F. and let ) 1c. xu(P(t>g(t))) = ( 4 1 .2.M ) be a principal Gbundle and let V be a vector space that carries a linear representation of G. PARALLEL TRANSPORT 269 unique map satisfying p = a(~(p))x.2. the result would depend on the trivialisation chosen.2. However. : M + Pv be a crosssection of the associated vector bundle. The precise definition is as follows. and then a subtraction can be performed unambiguously.6 Let = (P. Definition 6. to Comments 1. In a local bundle chart.(p) for all p E T . More precisely. using the local trivialisation u : xF1(U) U x F . Let (Y : [0.21) 6. (6.3 Covariant differentiation The basic problem of defining a derivative of a crosssection $J : M + Pv of a vector bundle is the lack of any natural way of comparing the values of $J at any pair of neighbouring points in M . the horizontal lift of a ( t )t o U x V that passes though (20.20) Similarly.21) as the curve .2.6.
2.. (6.2.23) that if the curves a and .2.26)ie.23)Eq.2.2. (6.’ + u ( a ( t ) ) . if the local representative of is +u : U + V satisfying Eq.$. w h e r e w i t h the aid of Eq.L? are tangent at zo E M . Thus the local representative of Va+ is the element of V given by: t + (6. where a is any curve in M that belongs to the equivalence class of v. g ( t ) v ) where g ( t ) satisfies the differential equation Eqs.2.270 CHAPTER 6. := Comments (6.2.25) is V.26) . (6. := Va. (6.rb+(a(t)) is = g ( t ) .2. the covariant derivative along X is the linear operator V X : r(PV) 4 r ( P v ) on the set r(PV) of crosssections of the vector bundle PV defined by + ( V x N z ) VX.3. A particular case of Eq.’ ) / d t = 0. the covariant derivative of the section of PV along v is defined to be V. It follows from Eq. CONNECTIONS IN A BUNDLE tj ( a ( t ) .10) with the boundary condition g ( 0 ) = e. 2. (6.2.96. + ~ ( z ) ) +(z)the element of V that represents .25) becomes the familiar expression (V.7 1.(z)$(z). (6. 2.2. Thus the following definition is meaningful: Definition 6.2.25) 1.10) that relates g ( t ) to the local YangMills field. If X is a vector field on M . h ’ ( z .+ := Va$. (6.23) w h e r e i n deriving the last expressionwe have used (i) the result Eq.24) for all crosssections $ of the vector bundle. If v E T. and (ii) the identity d ( g ( t ) g ( t ) . then va+ = V@$ (6. (5.!Nz) = WW + A. Then.M.
it is linear with respect to the set of vector fields on M regarded as a module over Cm(M)) the sense that in 4.3.2. Thus the function from P to V that corresponds to the crosssection V X $ is simply xr( O d .v].6.2. the covariant derivative of the crosssection $ can be expressed in terms of the equivalent function & : P + V that satisfies & ( p g ) = g'&(p) as Vx.30) are characteristic in the sense that one way of defining a covariant derivative on a vector bundle (Pv.X E VFlds(M).Note that V X is also linear in the subscript X ( i e . that satisfies these condi+ tions. (6. We know from the heuristic discussion in YangMills theory . .19). (5. One can then track backwards and eventually arrive at the idea of a connection oneform on the underlying principal bundle I. 3.2. Using the notation in Eq. L ~ ( u := [p.T V .it also possesses a derivation property in the form for all f E C m ( M ) .27) where L is the usual map V + Pv. The conditions expressed by Eqs.$ = LP(XL4lL) (6. Not only is V X a linear operator on r(PV).M ) is as any family of linear maps V x : r(Pv) r ( P V ) . cl 6.4 The curvature twoform All hhat remains to be done to relate the heuristic theory of YangMills fields to the theory of connections is to explain how the wellknown YangMills field strength Fpvarises in this geometrical approach.2.286. PARALLEL TRANSPORT 271 2.2. and X is the ) r horizontal lift of the vector field X to P [Exercise!].
Xk+l} of vector fields on P(J). X 2 . .horX2. CONNECTIONS IN A BUNDLE (which necessarily neglects all global aspects sincein effectit deals with only a local trivialising chart on M ) that FPvwill turn out to be some sort of ‘covariant curl’ of A. The new idea that must be invoked is a special extension of the exterior derivative to incorporate the verticalhorizontal splitting of the tangent spaces.2.Xk+1) = dw(horX1. the curvature twoform The relation of Dw to the familiar YangMills field strength is contained in the famous Cartan structural equation: Theorem 6. . where [wp(X)wp(Y)]denotes the Lie bracket in L(G) between the Lie algebra elements wp(X) and wp(Y). Definition 6. f o r all p E P ( 0 .26). X2. If w is a connection oneform on P€) (‘. .8 1.31) + i.2. the exterior covariant derivative of w is the horizontal (k 1)form Dw defined by Dw := dw o hor (6. (6.32) for any set {XI.272 CHAPTER 6. Dw(X1.4 If G = Dw is the curvature 2fom of the connection w. . .. If w is any kform on a principal bundle space P ( c ) . arranged to transform covariantly under the gauge transformation in Eq.horXk+1) (6.. but the discussion above of covariant differentiation is not applicable here since the connection/YangMills field is not a crosssection of a vector bundle but rather a oneform defined on the principal bundle space. .1. .. of w is defined as G := Dw.. . .. then o n a n arbitrary pair of vector fields X and Y o n P ( ( ) we have. e. 2.
6.2. PARALLEL TRANSPORT
273
Proof
Since both sides of Eq. (6.2.33) are linear functions of X and Y it suffices to prove the relation for the three choices: (i) X and Y both horizontal; (ii) X and Y both vertical; (iii) one of the pair X , Y is horizontal, the other is vertical. (i) If X and Y are both horizontal then w ( X ) = 0 = w ( Y ) , and D w ( X ,Y ) = d w ( X ,Y ) . Hence Eq. (6.2.33) is satisfied. (ii) If X and Y are both vertical then there exist A , B E L ( G ) such that X p = X t and Yp= X: . Now, according to the definition following Theorem 3.2, a tensor expression like dwp(X,Y ) is independent of which local vector fields are chosen to extend X , and Ypaway from p E M . In particular, we can evaluate this expression using X A and X B . Then the right hand side of Eq. (6.2.33) becomes
&p(XA, X B ) + I ~ p ( X A ) ~ p ( X B ) I 
(6.2.34)
Now, using the expression Eq. (3.4.15) for the exterior derivative of a twoform we have d w ( X A , X B ) = X A ( w ( X B ) ) X B ( w ( X A ) )w ( [ X " , X " ] ) . But w ( X " ) is the constant Lie algebra element B , and hence X " ( w ( X " ) ) = 0. Similarly, X " ( w ( X " ) ) = 0. By Eq. (4.5.27), [ X " , X B ] = XiAB],and hence w ( [ X A , X B ] = w ( X [ " ~ ] ) [AB]. ) = Thus, since w p ( X A )= A and w p ( X B )= B , we see that Eq. (6.2.34) vanishes. However, the left hand side of Eq. (6.2.33) vanishes identically since X and Y are vertical. Thus Eq. (6.2.33) is satisfied. (iii) If X is horizontal and Y is vertical, then G ( X ,Y ) = 0 (because Y is vertical), and [ w p ( X ) w p ( Y = 0 since w ( X ) = 0. Thus it remains )] only to show that dw(X, Y ) = 0. Evaluating this tensorial object at p E P by invoking the same argument as above, we can replace Y with X A for some A E L ( G ) . Now, X ( w ( X A ) )= 0 since w ( X " ) is the constant Lie algebra element A; and w ( X ) = 0 since X is horizontal. Thus dwp(X, ) = wp([X,X " ] ) . However, if X generates the flow Y t I+ +f of diffeomorphisms of M , we know from Eq. (4.5.20) that [x, ]= limt,,,(Y  + f * ( Y ) ) / tIn particular, Y .
[x, = ',%(&xp x"]
tA,(X)  X ) / t *
(6.2.35)
274
CHAPTER 6. CONNECTIONS IN A BUNDLE
Thus, if X is horizontal, so is [X,XA], and hence w ( [ X , X A ]= 0. )
QED
Comments
1. If { E l ,E 2 , .. . , earn^} is a basis for the Lie algebra L(G), we can write w = waEa,and then Eq. (6.2.33) becomes
dim G
G" = dw"
C CbcaWbA
Wc
(6.2.36)
b,c=l
where Cbc" are the structure functions of L(G) (and where summing is understood over the repeated indices). This expression for G" should be contrasted with the CartanMaurer equation Eq. (4.3.10) dw" 1 dimG CbcaWb  0 for the leftinvariant oneforms W" on a Lie 5 cb,c=1 group G.
+
2. If 0 : U + P is a local section of the principle bundle, the local representative A := a*w of w is supplemented with the local representative F := a*G of the curvature 2form. It then follows from Eq. (6.2.36) that F a = dA"+ ~$~~ CbcaAbAAc. inserting Or, coordinate indices, we get the familiar result
FP a
=
 A&
+ C Cbc"ALAE).
b,c=l
dim G
(6.2.37)
3. It is easy to prove the Bianchi identity DG = 0.
+ P and 0 2 : U + P are a pair of local sections with 2 Ul n U2 # 0, there exists some local gauge function R : Uln U2 + G
4. If
01
: U1
such that 02(x) = a l ( x ) R ( x ) .Correspondingly, there are two local representatives for the curvature 2form Gnamely F(') := crl*G and F(2):= az*G. Using an analysis very similar to that employed in the derivation of the gauge relation Eqs. (6.1.216.1.25), it can be shown [Exercise!] that these curvature representatives are related by
F$) (x)= IR (x) 'Fit' (x)R(x)
for all
II:
(6.2.38)
0
E U, n U,.
6.2. PARALLEL TRANSPORT
275
This completes the derivation of the basic relation between the mathematical theory of connections in principal and associated bundles, and the physicists’ familiar theory of the YangMills field and its gauge transformations. Note that, although we have talked above about the YangMills field, the same analysis applies also to the Riemannian connection in the GL(m,R)bundle of frames B(M). In this case, the parallel transport and covariant derivatives coincide with the familiar operations from elementary Riemannian geometry, and the curvature 2form that takes its values in the Lie algebra of GL(rn,IR) is nothing but the usual curvature tensor in a nonholonomic basis.
Bibliography
Abraham, R. & Marsden, J. (1980), Foundations of Mechanics, Benjamin, London. Beltrametti, E. & Cassinelli, G. (1981), The Logic of Quantum Mechanics, AddisonWesley, London. Bott, R. & Tu,L. (1982), Differential Forms in Algebraic Topology, SpringerVerlag, New York. Bourbaki (1966), Elements of Mathematics: AddisonWesley, London.
General Topology,
Chevalley, C. (1946), Theory of Lie Groups, Princeton University Press, Princeton. C s M r , A. (1978), General Topology, Adam Hilger, London. Dugundji, J. (1996), Topology, Allyn and Bacon, Boston. Frohlich, 0. (1964), Math. Ann. 156,79. Hawking, S., King, A. & McCarthy, P. (1976), J. Math. Phys. 17, 171. Helgason, S. (1962), Differential Geometry and Symmetric Spaces, Academic Press, New York. Hicks, N. (1965), Notes o n Differential Geometry, Van Nostrand, New York. Husemoller, D. (1966), Fibre Bundles, McGrawHill, New York.
277
278
BIBLIOGRAPHY
Isham, C. (1984), Topological and global aspects of quantum theory, i n B. DeWitt & R. Stora, eds, ‘Relativity, Groups and Topology II’, NorthHolland, Amsterdam, pp. 10621290. Isham, C. (1995), Lectures o n Quantum Theory: Mathematical and Structural Foundations, Imperial College Press, London. Johnstone, P. (1986), Stone Spaces, Cambridge University Press, Cambridge. Kelly, J . (1970), General Topology, Van Nostrand, London. Kobayshi, S. & Nomizu, K. (1963), Foundations of Differential Geometry: Volume I, Wiley, New York.
63,481. Kronheimer, E. & Penrose, R. (1967), Proc. Camb. Phil. SOC.
Lang, S. (1972), Differential manifolds, Addison Wesley, London. Larson, R. & Andima, S. (1975), Jour. Math. 5, 177. Lipschutz, S. (1965), General Topology, Schaum Publishing, New York. MacLane, S. & Moerdijk, I. (1992), Sheaves in Geometry and Logic: A First Introduction to Topos Theory, SpringerVerlag, London. Malament, D. (1977), J. Math. Phys. 18, 1399. Milnor, J . & Stasheff, J. (1974), Characteristic Classes, Princeton University Press. Montgomery, D. & Zippin, L. (1955), Topological Transformation Groups, Interscience, New York. Postnikov, M. (1986), Lectures in Geometry: Lie Groups and Lie Algebras, Mir Publishers, Moscow. Singer, I. (1978), ‘Some remarks on the Gribov ambiguity’, Comm. Math. Phys. 60, 712.
L. Academic Press. Spacetime and causal sets. S. Topology via Logic. (1951). Cambridge. Urrutia & F. (1989). Topological Vector Spaces. Treves. pp. Vickers. Rosenbaum. Steenrod. London. NahmadAchar. E. (1967). Zertuche. The Topology of Fibre Bundles. N. Princeton. Princeton University Press. Ryan. in J. World Scientific. R. . F. M. Cambridge University Press. eds.BIBLIOGRAPHY 279 Sorkin. 150173. D’Olivo. (1991). M. ‘Relativity and Gravitation: Classical and Quantum’. Singapore.
.
215 locally trivial. 237 Bianchi identity. 237 map between pair of. 212 associated. 214 composition of pair of. 215 normal. 233 automorphism group of. 204 fibre over 5. 236 . 215 local isomorphism between a pair of. 47 adjoint map. 202 crosssection of.202 isomorphism between a pair of. 20. of an embedding. 152 associated bundle. 42 boundary of a set. 210 normal. 237 trivial. 215 Ctale. 249 of tensors. see associated bundle base space. 220 28 1 Mmap. 274 billiardball hairy. 12. 202 Coo. 12 point. 101 Boolean algebra. 162 antihomomorphism. 216 map.Index accumulation point. 204 G. 239 local isomorphism between pair of. see crosssection fibre. 31 bundle.
35 closure. 272 curve definition of. 254 continuous map. 247 principal bundle of.282 principal. 257 Cartesian product. 224. see map. 36. 123 vector. 26 finer than. 53 collection of subsets coarser than. 111 canonical. 230 product bundle of. continuous coordinate chart. 98 curvature twoform. 47 connect ion affine sum of pair of. 84 of the ring C w ( M ) . see principal bundle product. 31. 270 covering space. 173 form 173. 218 tangent bundle of. 256 local representative of. 219. 63 cotangent bundle. 263 tangent pair of. 73 DeRham complex. 6.99 diffeomorphism ddfinition of. 61 INDEX atlas of. 207. 202 pullback. 216 universal. 200. 213 sub. 229 oneparameter group of. 217 derivation at a point. 272 Cart an. 232 vector. 80 components of. 63 coordinates.Maur er equation. 241 associated bundle of. see vector bundle Cartan structural equation. 171 DeRham’s theorem. 64 characteristic class. 213 total space. 62 domain of. 141. 62 . 123 space. 61 coordinate functions. 26 commutation relations affine. 70 group of. 13. 70. 228. 110 compact space. 204 projection map of. 231 closed set. 219. 123 covariant derivative. 202 trivial. 256 principal bundle in. 111 differentiable manifold. 210 crosssection. 73 horizontal lift of. 246 local representative of. 217 restriction to subset of base space. 269. 207 pullback of.
first countable c*. 232. see nform. complex. 121 dual map between pair of. 63 infinitedimensional. 232 gauge group. 69 smooth. 229. 47. . 70 distance. 145 inverse set. 48 local representative of. 29 neighbourhood. 95 differentiable. 226. 179 orbit of. 175 free. 49 diagonal. see neighbourhood. 224. 218 Gproduct. exterior derivative of covariant. 242 general linear. 46 directional derivative. 231 group action on a set. 30 definition of. see relation. see topology. homotopic pair of. 131 filter base. 231. 45 first countable space. 239. 70 function. 228 equivalence relation. function 283 121 EilenbergMaclane space. 272 function of. 31 exterior derivative nform of. filter subbase. 255 exponential map. 69 continuous. 219. 50 base. 16. 180 orbit space of. 12. 178 homotopic pair of. . 201. convergence with respect to. 62 directed set. 12 point. 80 transitive. 217.INDEX complex. 31 ultra. 229. 224. 258 Gribov effect. 179 adjoint representation ( 168 cohomology. 258 orbit. 29. 121 dual basis of. 5 equivariant . equivalence exact sequence. 179 kernel of. 63 differential structure. 16 transformation. 182 stability group of.165 exterior of a set. 75 dual vector space.
268 instanton. 151 unitary. 130 vector field of. 19. see function. 157 orthogonal. 226.169 dual. 45 atomic. 42 isotropy group. 19. 161 Lie derivative function of. 157 partial. 31 . 98 oneform of. 41 definition of. see Lie group Lorentz. 171 Lie group of. 242. 12 point. 19 distributive. 222. 157 spin. 246 left translation of. 43 unit element in. 157 group action. real. 229. 22 null element in. 52 Hausdorff space. action on a set. 52 orbit of. 150 special linear. stability group of Jacobi identity. 27 Lie algebra. 155 special. 19 upper set in. 113 right translation of. 205 lattice a topology a an example. 51 homotopic functions. 88 KaluzaKlein theory.lR). 41 example of closed linear subspaces of a Hilbert space. 154 special unitary. 19 structure on space of topologies. see vector field. see topology. 103 Jacobian matrix. 267 homotopy. 103 of GL(n. 201. 12. 243 integral curve. homotopic pair of Hopf bundle. 228. 116 153 general linear. see group. 159 structure constants of. 152 holonomy. 201 Klein bottle. Hausdorff Heyting algebra. 44 complete. integral curve of interior of a set. 225 topological. 42 homeomorphism. 150 Lie. 222 horizontal subspace. s 40 antiatomic.284 INDEX intuitionistic logic.
135 closed. 26. 7 metrics equivalent. 7. 171 rightinvariant. orbit space of overlap function. 8 one form Lie derivative of. 35 linear frame. 14. 6 bounded. 240 norm. 224. see function metric. 7. 16 isometric. 34 space. 177 definition of. 11 partial ordering of. 16. 7 join of. 27 of a point. 27. 56 locally compact space.INDEX Lie group action on a manifold. 166 subgroup of. 152 oneparameter subgroup of. see space. 18. see Lie derivative. 11 meet of. 10 lattice structure on. 241. 123 pullback of. 17. action on a set. 126 definition of. 46 nonlinear amodel. 13. 137 exterior product of. 245 Riemannian. 223 little group. 37 Mobius band. 150 homomorphism between pair of. 205 Mobius transformations. 35 pseudocomplement of. 152 map continuous. metric stronger than. 171 neighbourhood filter. 127 open set. 126 components of. 229. stability group of locale. see group. 153 lower set. 224. 23. one form of oneform Vvalued. 244 space. 243. 46 convergence of. 142 exterior derivative of. see group. 142 exact. 50 map. 21 285 nform. 31. 41 orbit space. 29. 135 leftinvariant. action on a set. 29. 61 . 8. 34 neighbourhoods equivalent families of. 11 Lorentzian. 32 net. 26 weaker than. 12. 150 limit point. 29 structure. 31.
see topology. 18 join operation in. 241 automorphism group of. 133 contravariant. 6. 31. 53 relation definition of. 234 trivial. 68 submanifold.286 INDEX Riemann sphere. 54 neighbourhood. 15 principal bundle. 188 scalar density. 31 tails of. see topology spontaneous symmetry breakdown. product pseudometric. 241 Hrestriction of. 3. 6. 226 connection in. 221 Gextension of. pointwise convergence of. 221 translation function of. 14. 29 of functions. 3. 241 stability group. 26. stability group of stereographic projection. 47 as a filter base. 268 partially ordered set. see poset Poincar6 lemma. space topological. 226 map. 74 differential structure on. 24. 18 orthocomplemented. 15 definition of. 142 poset covering element in. see connection. 64 tangent bundle. 5. 74 of a product manifold. 212 space metric. 226 product topology. 225 structure functions of. see neighbourhood. 24. 19 least upper bound in. 19 meet operation in. 74 pushforward of. 6 . 50 . 267. action on a set. 15 equivalence. 133 Palais’ Theorem. 30 sheaf. 5. 239 structure group of. 24. see group. 89 space. 92 vector. 36 preorder. 78 tensor. 19 topologies on. 198 parallel translation. 16. 15 greatest lower bound in. 26. principal bundle in locally trivial. 236 sequence convergent.
242 product. see topology. 262 induced. 248 vector field Cm(M)modul~? structure on set of. see topology. 54 finer. Hausdorff base of. 40 of a metric space. 105 as derivation of ring C ( M ) . 115 restriction to opan subset. 133 field . 52 stronger. 236 topological space. see topology topology To. 231. 43 induced. 53 T2. veciior field of local flow of. 32. 191 integral curve of'. 53 indiscrete. 35 discrete. 134. 135 tensor density. 114 induced by oneparameter subgroup. 158 Lie derivative of. 98 rightinvariant.INDEX covariant. 50 lattice structure on. 1 58 vector space structure on set of. see Lie derivative. 123. 35 subspace. 13 paracompact . 61 connected. 109 components of. 13. stronger first countable. 36 287 vector horizontal. 27. 256 vertical. 102 complete. 50. 77 map between pair of. 42 torus. 42 subbase of. 54 component of. 67 upper set. 108 leftinvariant. 43. 248 TM as an exarnple of. 61 definition of. 50 weaker. 97 horizontal lift of. 42 topos theory. 53 identification. 99 hrelated pair of. 100 definition of.99 commutator of pair of. see topology. 52 lattice structure on open sets. 33 Hausdorff. 36. 53 Tl. weaker cofinite. 36 separation axioms. 254 vector bundle. 35 coarser. 99 vector space .
162 A L . 70 d w .142 B d M ) .5 a U b.80 Diff(M). w. 24 V[. 63.bI. 269 ( L . 268 vector. 137 d x m .121 II4L 8 5 . 95 d f . 188 C([a. 222 YangMills theory.11 Ad. 256.153 GL(n. 231. 12 BE(% 5 B n ( M ) .9 D w . 9 3. 180 h*(lc). 18 a V b. 233 Y F + E 3 M. 160 A"(M)..p. 48 f:A+B. 24 f . 79 CPoo.62 a=. w. 15 34 A. 266 INDEX AxB.IR). 125 d(X. 165 bl. . vertical Wilson loop. 271 CO"(M). 4 &6.V ) . 224 rp). C ) . 207. 222. 272 A. 226 a f . 146 a A b. v ) . Ext(A). 16. 53 Aut(J). 131 DpM. 19 A C . 265.]. 122 vertical subspace. 202 (x.127 h*w. see vector.M).288 t o p o l o g d .223 Coo. 201. 160 [a. 4 c.248 G L ( n .6 51.IR). 263 AcB. 6 W ) 204 ..229 C P . 204 [AB]. 154 Gp.135 11 x. 229.W.rr.Y). 224.152 GL+(n. ) G . 26 Q := p..6 Bd(A). 12 expA. 239. 127 (E.bI.5 a !. 218 B ( M ) .
255 h*(v). 98 k x f .133 T.26 N ( X >34 . 155 O(P. 162 w. 133 u p .171 I. 74 T * M . & V. 87 ver(r). l R ) . N(4. 105 L ~ 250 . 219 121 G P . 255 4. 74 S L ( n . 157 O. 157 9(x). 34 U ( n ) . 75 VFld(M). 126 . 65 SOo.INDEX 289 r g . 98 e2(x>. 143 hor(T). 99 m. 159 Ad. 192 d ) ( z ) IRn. 132 3 wP. P ) . 52 5 IR+.. 93 wl). 254 n/(')(z). 5 T ( X ) . 157 S1. 225 bl. vglw. 4.n).IR).X. C ) . 17 o(n. 150 q5.(M).228 LA. 154 SL(n.M. 16 M . 94 K(.. 188 SO(n. 123 Tp*M. 157 S U ( n ) . 16 Perm(X).R).157 UF.. Int(A). 94 j p . 204 Metric(X).IR). 9 L x Y . 219 V C W . 150 L X W .f ). 180 PF.130 L x f . 116 M ( n .110 RA. 12 i. 74 Tn. v*. 104 Wt).R). 5 I R P .158 L*(G).rr. 68 W P .229 Spin(rn. 218 T ( P o . 78 h.123 Tp""M.IR).154 Sn. 234 TM. 233 P ( X ) . 171 L(G).4.
A.3 x N(X) x. 11 XA.5 x. X/R. 262 3.. [ X . 63 X . 24 . 142 . 16 Z " ( M ) . X?. 98 XEX.290 A ~ 2 135 . 97 ( XJfh 99 { Iw 4 1 1 4 . 233 x p .5 E[FI. 100 ~1 INDEX fl81 Zn 3 2. + 101 X XG Y . 191 Xfl97 XP. I 102 Z. 233 g.Y . 16 x u ..