Springer Monographs in Mathematics

Springer Science+Business Media, LLC

N.P. Landsman

Mathematical Topics
Between Classical and
Quantum Mechanics

With 15 Illustrations

, Springer
N.P. Landsman
Korteweg-de Vries Institute for Mathematics
University of Amsterdam
Plantage Muidergracht 24
Amsterdam 1018 TV
The Netherlands

Mathematics Subject Classification (1991): 8ISIO, 8IPXX, 58FXX, 8IRXX, 81TXX

Library of Congress Cataloging-in-Publication Data

Landsman, N.P. (Nicolaas P.)
Mathematical topics between classical and quantum mechanics / N.P.
Landsman.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-4612-7242-7 ISBN 978-1-4612-1680-3 (eBook)
DOI 10.1007/978-1-4612-1680-3
1. Quantum theory-Mathematics. 2. Quantum field theory-
Mathematics. 3. Hilbert space. 4. Geometry, Differential.
5. Mathematical physics. 1. TitIe.
QCI74.I7.M35L36 1998
530.12---dc21 98-18391

Printed on acid-free paper.

© 1998 Springer Science+Business Media New York

Origina1ly published by Springer-Verlag New York, Inc.in 1998
Softcover reprint ofthe hardcover 1st edition 1998
AII rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer Science+Business Media, LLC),
except for brief excerpts in connection with reviews or scholarly analysis. Use in connection
with any form of information storage and retrieval, electronic adaptation, computer software, or by
similar or dissimilar methodology now known or hereafter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even if the
former are not especially identified, is not to be taken as a sign that such names, as understood by
the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone.

Camera-ready copy prepared from the author' s JJ.TEiX files.

987 6 5 4 3 2 1

lSBN 978-1-4612-7242-7
 I realize that the disappearance of a culture does not signify the disappearance
of human value, but simply of certain means of expressing this value, yet the fact
remains that I have no sympathy for the current of European civilization and do
not understand its goals, if it has any. So I am really writing for friends who are
scattered throughout the corners of the globe.
Our civilization is characterized by the word "progress". Progress is its form
rather than making progress one of its features. Typically it constructs. It is oc-
cupied with building an ever more complicated structure. And even clarity is only
sought as a means to this end, not as an end in itself For me on the contrary
clarity, perspicuity are valuable in themselves. I am not interested in constructing
a building, so much as in having a perspicuous view of the foundations of typical
buildings.
Ludwig Wittgenstein
Preface

Subject Matter
The original title of this book was Tractatus Classico-Quantummechanicus, but
it was pointed out to the author that this was rather grandiloquent. In any case,
the book discusses certain topics in the interface between classical and quantum
mechanics. Mathematically, one looks for similarities between Poisson algebras
and symplectic geometry on the classical side, and operator algebras and Hilbert
spaces on the quantum side. Physically, one tries to understand how a given quan-
tum system is related to its alleged classical counterpart (the classical limit), and
vice versa (quantization).
This monograph draws on two traditions: The algebraic formulation of quan-
tum mechanics and quantum field theory, and the geometric theory of classical
mechanics. Since the former includes the geometry of state spaces, and even at
the operator-algebraic level more and more submerges itself into noncommutative
geometry, while the latter is formally part of the theory of Poisson algebras, one
should take the words "algebraic" and "geometric" with a grain of salt!
There are three central themes. The first is the relation between constructions
involving observables on one side, and pure states on the other. Thus the reader will
find a unified treatment of certain aspects of the theory of Poisson algebras, oper-
ator algebras, and their state spaces, which is based on this relationship. Roughly
speaking, observables relate to each other by an algebraic structure, whereas pure
states are tied together by transition probabilities (in both cases topology plays
an additional role). The discussion of quantization shows both sides of the coin.
One side involves a mapping of functions on the classical phase space into some
operator algebra; at the other side one has coherent states, which define a map
from the phase space itself into a projective Hilbert space. The duality between
these sides is neatly exhibited in what is sometimes called Berezin quantization.
viii Preface

The second theme is the analogy between the C* -algebra of a Lie groupoid
and the Poisson algebra of the corresponding Lie algebroid. For example, the role
played by groups and fiber bundles in classical and quantum mechanics may be
understood on the basis of this analogy.
Thirdly, we describe the parallel between symplectic reduction in classical me-
chanics (with Marsden-Weinstein reduction as an important special case) and
Rieffel induction (a tool for constructing representations of operator algebras) in
quantum mechanics. This provides an interesting example of the mathematical
similarities alluded to above, and in addition leads to a powerful strategy for the
quantization of constrained systems in physics.
Various examples illustrate the abstract theory: The reader will find particles
moving on a curved space in an external gauge field, magnetic monopoles, low-
dimensional gauge theories, topological quantum effects, massless particles, and
8-vacua. On the other hand, the reader will not find path integrals, geometric
quantization, the WKB-approximation, microlocal analysis, quantum chaos, or
quantum groups. The connection between these topics and those treated in this
book largely remains to be understood.

Prerequisites, Level, and Organization of the Book

This book should be accessible to mathematicians with a good undergraduate
education and some prior knowledge of classical and quantum mechanics, and to
theoretical physicists who have not completely abstained from functional analysis.
It is assumed that the reader has at least seen the description of classical mechanics
in terms of symplectic geometry, and knows the standard Hilbert space description
of a quantum-mechanical particle moving in R3.
The reader should be familiar with the basics of the theory of manifolds, Lie
groups, Banach spaces, and Hilbert spaces, say at the level of a first course. The
necessary concepts in operator algebras, Riemannian and symplectic geometry,
and fiber bundles are developed from scratch, but some previous exposure to these
subjects would do no harm.
It is suggested that the reader start by going through the informal Introductory
Overview as a whole. The main text is of a technical nature. The various chapters
are logically related to each other, but can be read almost independently. To study
a given chapter it is usually sufficient to be familiar with the preceding chapters
merely at the level of the Introductory Overview. Some technical details will, of
course, depend on previous material in a deeper way. One should by all means go
through the list of conventions and notation below.
In the interest of clarity and continuity, no credits or references to the literature
are given in the main text. These may be found in the Notes, which in addition
contain comments and elaborations on the main text. If no reference for a particular
result is given, it is either standard or new (we leave this decision to the reader).
 Conventions and Notation ix

The author would be happy if glaring omissions in the notes or references were
pointed out to him.
In the Index, entries refer only to the location where an entry is defined and/or
occurs for the first time.

Conventions and Notation

Unless explicitly indicated otherwise, or obvious from the context, our conventions
are as follows.

General
• The (Roman) chapter number is used only in cross-referencing between dif-
ferent chapters. In such references, numbers in brackets refer to equations and
those without refer to paragraphs (e.g., 1.2.3) or to sections (such as 1.2).
• The symbol • means "end of proof". The symbol 0 stands for "end of
incomplete proof".
• The equation A := B means that A is by definition equal to B.
• The abbreviation "iff" means "if and only if".
• An index that occurs twice is summed over, i.e., ajaj := Li ajaj.
• Projections between spaces are denoted by T; in case of possible confusion we
write TE->Q for the pertinent projection from E to Q.
• The symbol f means "restricted to".
• The symbol Ix stands for the function on X that is identically one.
• We put 0 E JR+ but 0 f/. N.

Functional Analysis
• Vector spaces are over C, and functions are C-valued. Vector spaces over JR are
denoted by VIR etc.; spaces of real-valued functions are written, for example,
COO(P, JR). The only exception to this rule is formed by Lie algebras 9, which
are always real except when the complexification 9c is explicitly indicated (this
occurs only in 111.1.10, III.l.l1, and IV.3.6).
• The space Co(X), where X is a locally compact Hausdorff space, consists of
all continuous functions on X that vanish at infinity; the space of all compactly
supported continuous functions on X is denoted by Cc(X), and the bounded
continuous functions form Cb(X). These are usually seen as normed spaces
under the sup-norm
11/1100 := sup I/(x)l.
XEX

• When X has the discrete topology (relative to which all functions are continu-
ous), we often write l(X),lc(X),lOO(X),lo(X) for C(X), Cc(X), L OO(X), and
Co(X).
x Preface

• The topological dual of a topological vector space V is denoted by V*; hence

the double dual is V**. The action of () E V* on v E V is denoted by ()(v).
Multilinear forms a are similarly denoted by a(vI' ... , vn ).
• When confusion might arise otherwise, we write X + Y for X + Y in VI EB V2,
where X E VI and Y E V2 (for example, in V EB V the expression X + Y would
be ambiguous, denoting either X + Y +0, where X + Y E V ~ V EBO c V EB V,
or X+Y, orO+X + Y).

Hilbert Spaces
• Inner products (, ) in a Hilbert space 1-l are linear in the second entry and
antilinear in the first.
• If K is a closed subspace of a Hilbert space 1-l, then [K] denotes the orthogonal
projection onto K. If \II E 1-l, we write [\II] for [C\II].
• The symbol S1-l denotes the space of all unit vectors in 1-l. The projective space
of 1-l is called 1P7t; hence IPC N = ClPN - I •
• The symbols ~(1-l), ~o(1-l), ~ I (1-l), ~lh(1-l) stand for the collections of all
bounded, compact, trace-class, Hilbert-Schmidt operators on 1-l. The unit
operator in ~(1-l) is called lL We write VJtN(C) for ~(CN).
• When A and B are operators on 1-l, the symbol [A, B) stands for the commutator
AB - BA. We also use {A, B}1i := i[A, B]/Ii.
• In the context of the previous item, or more generally when A and B are elements
of a Jordan algebra or a C* -algebra, A 0 B denotes ~ (A B + B A). In all other
situations, 0 has its usual meaning of composition; i.e., when f and g are
suitable functions, one has f 0 g(x) := f(g(x».
• We say that two Hilbert spaces are naturally isomorphic if they are related by
a unitary isomorphism whose construction is independent of a choice of basis.
• The Hilbert space L2(JRn ) is defined with respect to Lebesgue measure.

Our convention for the inner product is the one mainly used in the physics
literature. Its motivation, however, is mathematical. Firstly, each \II E 1-l defines a
linear functional on 1-l by \11(<1» := (\II, <1», without the need to change the order.
Secondly, the convention is the same as for "inner products" taking values in a
C*-algebra, which for good reasons are always taken to be linear in the second
entry; see IV.2.

C* -Algebras
• The set of self-adjoint elements in a C* -algebra sa is called salR. Its state space
is S(sa), and its pure state space is P(sa).
• The unitization of a C*-algebra sa is called san.
• States on a C* -algebra are denoted by w; pure states are sometimes also called
p, a, or 1/f. The state space of sa is called S(sa); the pure state space is denoted
by P(sa).
 Conventions and Notation xi

A of A E ~tn{ is the function on P(~) defined by

• The Gelfand transfonn
A(w) := w(A). When~is commutative, this concept is used for general A E ~.
• Representations of a C* -algebra are generically denoted by 7r •
• The GNS-representation corresponding to a state w is called 7rw , with canonical
cyclic vector Qw.
• Equivalence of representations means unitary equivalence.
• The representation of a C* -algebra 21 induced (in the sense of Rieffel) by a
representation 7r x of a C* -algebra ~ on a Hilbert space fix is denoted by
7r x (~), realized on a Hilbert space fix.
• Transfonnation group C* -algebras are called action C* -algebras.

Group Representations and Actions

• Group representations on a Hilbert space are tacitly assumed to be continuous
and unitary.
• The adjoint action of a Lie group G on its Lie algebra 9 is denoted by Ad; the
dual coadjoint action on g* is called Co, i.e., Co(x) := Ad*(x- I ).
• When H eGis a closed subgroup, the representation of G induced (in the
sense of Mackey) from a representation U x (H) on a Hilbert space fi x is denoted
by UX (G), and is realized on a Hilbert space called fix.
• The unitary dual of a group G is denoted by G.
• Equivalence of group representations means unitary equivalence.

Differential Geometry
• All manifolds (Lie groups included) are assumed to be real, smooth, connected,
Hausdorff, finite-dimensional, and paracompact.
• If cp : M ~ N is a smooth map between two manifolds, the pullback is denoted
by cp*, and the pushforward is cp* (often called Tcp or cp' in the literature). In
particular, for g E COO(N) the function cp*g in COO(M) is g 0 cpo
• We denote a point on a manifold Q by q, with coordinates qi (in a given
chart; i = 1, ... , dim(Q». The dependence ofthe coordinates on the chart is
suppressed in the notation. We write ai for a/aqi. The point Pidqi in the fiber
Tq* Q of the cotangent bundle T* Q at q then has canonical coordinates (Pi, qi);
we denote this point by (p, q). Similarly, the point Vi ai in the fiber Tq Q of the
tangent bundle T Q at q has coordinates (Vi, q j), and we sometimes label this
simply as (u, q).
• Theactionof8 E Tq*Qonu E TqQ iswrittenas8q (u). Similarly for multilinear
fonns, e.g., ~(v, w) stands for a Riemannian inner product of v, WE Tq Q.
• The tangent vector (field) to a curve cO is called cO.
• The symbol I\n(Q) stands for the bundle of n-fonns over Q. Also, I\n(Q)
is the dual vector bundle of I\n(Q), i.e., the bundle of totally antisymmetric
contravariant tensors.
xii Preface

• The space of compactly supported smooth sections of a vector bundle E is

denoted by r(E).
Acknowledgements

This book was written between March 1996 and March 1998, based on research
starting in 1989. The project was financed by the E.P.S.R.C., the Alexander von
Humboldt Stiftung, and the Royal Netherlands Academy of Arts and Sciences. It
was carried out at the Department of Applied Mathematics and Theoretical Physics
of the University of Cambridge (October 1989-September 1993, October 1994-
June 1997), the II. Institut fur Theoretische Physik of the University of Hamburg
(October 1993-September 1994), the Korteweg-de Vries Institute for Mathe-
matics of the University of Amsterdam (July 1997-March 1998) and the Erwin
Schr6dinger Institute for Mathematical Physics in Vienna (September-October
1997).
Many mathematical ideas in this book may be traced back to J. von Neumann;
key physical insights originated with P.A.M. Dirac. In addition, the author has been
inspired by the work of EA. Berezin, P. Bona, A. Connes, V. Guillemin, R. Haag,
K. Hepp, C.J. Isham, G.w. Mackey, J.E. Marsden, B. Mielnik, M.A. Rieffel, EW.
Shultz, J.-M. Souriau, S. Sternberg, A. Weinstein, and P. Xu.
The research of the author's Ph.D. students Mark Robson, Urs Wiedemann,
and Ken Wren contributed to this work, as did his collaboration with Noah Lin-
den. Helpful comments, suggestions, and corrections on the manuscript were
received from Hendrik Grundling, Brian Hall, Eli Hawkins, Marc Rieffel, Simon
Ruijsenaars, Erik Thomas, Gijs Tuynman, and Alan Weinstein.
The author is grateful to Jeremy Butterfield, Robbert Dijkgraaf, Gerard Emch,
Klaus Fredenhagen, Chris Isham, Dick Kadison, Daniel Kastler, Jerry Marsden,
and John C. Taylor for moral and other forms of support.
The book was written during a happy time, shared with Imke and our cat Pauli.

Klaas Landsman
University of Amsterdam
Contents

Preface vii
Subject Matter. . . . . . . . . . . . . . . . . . . . Vll
Prerequisites, Level, and Organization of the Book viii
Conventions and Notation. IX
General . . . . . . . IX
Functional Analysis ix
Hilbert Spaces . . . X
C* -Algebras . . . . X
Group Representations and Actions . xi
Differential Geometry xi
Acknowledgements. xiii

Introductory Overview 1
I. Observables and Pure States
Observables . . . . . . I
Pure States . . . . . . . 3
From Pure States to Observables 5
II. Quantization and the Classical Limit 7
Foundations . . . . . . . . . . . 8
Quantization on Flat Space '" 10
Quantization on Riemannian Manifolds . 13
III. Groups, Bundles, and Groupoids . . . . . . 14
Lie Groups and Lie Algebras . . . . . . 14
Internal Symmetries and External Gauge Fields 17
Lie Groupoids and Lie Algebroids 21
IV. Reduction and Induction 25
Reduction . . . . . . . . . . . . . 25
xvi Contents

Induction. 28
Applications in Relativistic Quantum Theory 32

I Observables and Pure States 37

1 The Structure of Algebras of Observables . 37
l.l Jordan-Lie Algebras and C* -Algebras . . 37
1.2 Spectrum and Commutative C*-Algebras 41
1.3 Positivity, Order, and Morphisms . . . . . 45
1.4 States................... 49
1.5 Representations and the GNS-Construction 52
1.6 Examples of C* -Algebras and State Spaces 55
1.7 Von Neumann Algebras . . . . . . . . 58
2 The Structure of Pure State Spaces . . . . . . . . . 60
2.1 Pure States and Compact Convex Sets . . . 60
2.2 Pure States and Irreducible Representations 63
2.3 Poisson Manifolds. . . . . . . . . . . . . . 65
2.4 The Symplectic Decomposition of a Poisson Manifold. 69
2.5 (Projective) Hilbert Spaces as Symplectic Manifolds. 71
2.6 Representations of Poisson Algebras . . . . . . . . 76
2.7 Transition Probability Spaces . . . . . . . . . . . . 80
2.8 Pure State Spaces as Transition Probability Spaces. 82
3 From Pure States to Observables . . . . . . . . . . . 84
3.1 Poisson Spaces with a Transition Probability. 84
3.2 Identification of the Algebra of Observables 86
3.3 Spectral Theorem and Jordan Product 88
3.4 Unitarity and Leibniz Rule . . . . . . . . . 90
3.5 Orthomodular Lattices . . . . . . . . . . . 92
3.6 Lattices Associated with States and Observables . 94
3.7 The Two-Sphere Property in a Pure State Space 98
3.8 The Poisson Structure on the Pure State Space .. 103
3.9 Axioms for the Pure State Space of a C*-Algebra 104

II Quantization and the Classical Limit 108

1 Foundations. . . . . . . . . . . . . . . . 108
1.1 Strict Quantization of Observables 108
1.2 Continuous Fields of C' -Algebras 110
1.3 Coherent States and Berezin Quantization 112
1.4 Complete Positivity . . . . . . . . . . . . 116
1.5 Coherent States and Reproducing Kernels 122
2 Quantization on Flat Space . . . . . . . . . . . . 126
2.1 The Heisenberg Group and its Representations. 126
2.2 The Metaplectic Representation . . . . . . . . 129
2.3 Berezin Quantization on Flat Space. . . . . . . 133
2.4 Properties of Berezin Quantization on Flat Space 137
2.5 Weyl Quantization on Flat Space . . . . . . . . . 140
 Contents xvii

2.6 Strict Quantization and Continuous Fields on Flat

Space . . . . . . . . . . . . . . . . . . . 144
2.7 The Classical Limit of the Dynamics .. 148
3 Quantization on Riemannian Manifolds . 154
3.1 Some Affine Geometry . . . . . . . 154
3.2 Some Riemannian Geometry . . . . 157
3.3 Hamiltonian Riemannian Geometry 159
3.4 Weyl Quantization on Riemannian Manifolds 162
3.5 Proof of Strictness . . . . . . . . . . . . . . . 166
3.6 Commutation Relations on Riemannian Manifolds. 170
3.7 The Quantum Hamiltonian and its Classical Limit. 173

III Groups, Bundles, and Groupoids 178

1 Lie Groups and Lie Algebras . . . . . . . . . . . . . 178
1.1 Lie Algebra Actions and the Momentum Map 178
1.2 Hamiltonian Group Actions. . . . . 183
1.3 Multipliers and Central Extensions . 187
1.4 The (Twisted) Lie-Poisson Structure 192
1.5 Projective Representations . . . . 196
1.6 The Twisted Enveloping Algebra . . 199
1.7 Group C*-Algebras . . . . . . . . . 201
1.8 A Generalized Peter-Weyl Theorem 206
1.9 The Group C* -Algebra as a Strict Quantization . 211
1.10 Representation Theory of Compact Lie Groups . 215
1.11 Berezin Quantization of Coadjoint Orbits 219
2 Internal Symmetries and External Gauge Fields 224
2.1 Bundles. . . . . . . . . . . 224
2.2 Connections............... 227
2.3 Cotangent Bundle Reduction . . . . . . 231
2.4 Bundle Automorphisms and the Gauge Group 235
2.5 Construction of Classical Observables 238
2.6 The Classical Wong Equations . . . . 241
2.7 The H -Connection . . . . . . . . . . 244
2.8 The Quantum Algebra of Observables 249
2.9 Induced Group Representations. . . . 253
2.10 The Quantum Wong Hamiltonian . . . 257
2.11 From the Quantum to the Classical Wong Equations . 260
2.12 The Dirac Monopole ... 264
3 Lie Groupoids and Lie Algebroids . . . 269
3.1 Groupoids... . . . . . . . . . 269
3.2 Half-Densities on Lie Groupoids 273
3.3 The Convolution Algebra of a Lie Groupoid 275
3.4 Action *-Algebras . . . . . . . . . 279
3.5 Representations of Groupoids . . . 282
3.6 The C*-Algebra of a Lie Groupoid 285
xviii Contents

3.7 Examples of Lie Groupoid C* -Algebras 288

3.8 Lie Algebroids . . . . . . . . . . . . . 292
3.9 The Poisson Algebra of a Lie Algebroid 296
3.10 A Generalized Exponential Map . . . . 302
3.11 The Groupoid C*-Algebra as a Strict Quantization. 305
3.12 The Nonnal Groupoid of a Lie Groupoid. . . . . . 308

IV Reduction and Induction 313

1 Reduction. . . . . . . . . . . . . . . . . . 313
1.1 Basics of Constraints and Reduction 313
1.2 Special Symplectic Reduction . . . 316
1.3 Classical Dual Pairs . . . . . . . . . 319
1.4 The Classical Imprimitivity Theorem . 322
1.5 Marsden-Weinstein Reduction . . . . 325
1.6 Kazhdan-Kostant-Stemberg Reduction 328
1.7 Proof of the Classical Transitive Imprimitivity
Theorem . . . . . . . . . . . . . . . . 332
1.8 Reduction in Stages . . . . . . . . . . . 336
1.9 Coadjoint Orbits of Nilpotent Groups. . 341
1.10 Coadjoint Orbits of Semidirect Products 343
1.11 Singular Marsden-Weinstein Reduction 349
2 Induction . . . . . . . . . . . 354
2.1 Hilbert C*-Modules . . . . . . . . . . . 354
2.2 Rieffel Induction . . . . . . . . . . . . 358
2.3 The C*-Algebra of a Hilbert C*-Modu1e 363
2.4 The Quantum Imprimitivity Theorem 366
2.5 Quantum Marsden-Weinstein Reduction. 370
2.6 Induction in Stages . . . . . . . . . . . . 375
2.7 The Imprimitivity Theorem for Gauge Groupoids 378
2.8 Covariant Quantization . . . . . . . . . . 383
2.9 The Quantization of Constrained Systems 386
2.10 Quantization of Singular Reduction . . . 390
3 Applications in Relativistic Quantum Theory . . 393
3.1 Coadjoint Orbits of the Poincare Group 393
3.2 Orbits from Covariant Reduction . . . 396
3.3 Representations of the Poincare Group 399
3.4 The Origin of Gauge Invariancc 403
3.5 Quantum Field Theory of Photons . . 407
3.6 Classical Yang-Mills Theory on a Circle 414
3.7 Quantum Yang-Mills Theory on a Circle. 420
3.8 Induction in Quantum Yang-Mills Theory on a Circle 424
3.9 Vacuum Angles in Constrained Quantization . . . . . 427

Notes 433
Chapter I. 433
 Contents xix

Chapter II . 445
Chapter III . 457
Chapter IV . 469

References 483

Index 521
Introductory Overview

I. Observables and Pure States

The aim of the first chapter is to give two descriptions of classical and quantum
mechanics, each of which enables one to see in a different way what their common
properties as well as their striking differences are. The first description focuses on
the observables of the theory, whereas the second one is based on the pure states.

Observables
Consider a particle moving in the configuration space Q = JR.3. Its phase space
is the cotangent bundle T*JR.3 ~ JR.6, and the collection of classical observables is
taken as Ql~ = c oo (T*JR.3, JR.). This is a real vector space under pointwise addition
and scalar multiplication by real numbers.
Ordinary (pointwise) multiplication of f, g E Ql~, which for the moment we
write as fog, naturally defines a bilinear map on Ql~. This map is commutative and
associative. In addition, in mechanics a key role is played by the Poisson bracket
af ag af ag
{f, g} := api aqi - aqi api'

Hence Ql~ becomes a real Lie algebra under the Poisson bracket. This bracket is
related to 0 by the Leibniz rule, which says that g r-+ {f, g} is a derivation of
o for all f E Ql~, in that {f, go h} = {f, g) 0 h + g 0 {f, hI. Hence one coins
the abstract definition of a Poisson bracket on a commutative (but not necessarily
associative) algebra as a Lie bracket satisfying the Leibniz rule with respect to the
product defining the algebra.
In quantum mechanics the above system is described by an infinite-dimensional
space; to avoid complications we shall instead look at an N -level quantum system
2 Introductory Overview

(N < 00). The set of its observables mlR is the real vector space 9J1 N(C)IR of
Hermitian complex N x N matrices. A symmetric bilinear product on mlR is given
by
A 0 B := ~(AB + BA).
In addition, mlR admits a Poisson bracket defined by
i
{A, Bli, := fi(AB - BA),

where n E iR\{O}; in physics n has a specific numerical value, and is known as

Planck's constant. A difference with the classical case is that 0 now fails to be
associative.
The following algebraic structure of the set of observables of classical or quan-
tum mechanics may be extracted from the above considerations. A Jordan-Lie
algebra mlR is a real vector space equipped with two bilinear maps, 0 and {, }
that are commutative and anticommutative, respectively. For each A EmIR, the
map B ~ {A, B} is a derivation of the Poisson structure (mIR , {, }); this makes
(mIR , {, }) a real Lie algebra. Also, B ~ {A, B} is a derivation of the Jordan
structure (mIR , 0); this is the Leibniz rule. Finally, the associator identity
(A 0 B) 0 C- A 0 (B 0 C) = ~n2{{A, C}, B)

holds, for some constant n E R For n = 0, in which case the commutative product
is associative, one speaks of a Poisson algebra; this associativity is an algebraic
characterization of classical mechanics.
The identity (AoB)oA 2 = Ao(BoA2), where A2 := AoA, which makes (mIR , 0)
a so-called (real) Jordan algebra, is implied by these axioms. A J B-algebra is
defined as a Jordan algebra for which mlR is a Banach space, and the norm and the
Jordan product 0 are related by certain axioms. We refer to a Jordan-Lie algebra
mlR for which (mIR , 0) is a J B-algebraas a J LB-algebra (for Jordan-Lie-Banach).
A C*-algebra is a complex Banach space equipped with an associative
multiplication and an involution *, such that the C* -axioms
IIABII :::; IIAIIIIBII, IIA*AII = IIAI12
are satisfied. It can be shown that any C* -algebra is isomorphic to a norm-closed
subalgebra of SB(H) for some Hilbert space H.
In elementary quantum mechanics one assumes that every (bounded) observable
of a given theory corresponds to a (bounded) self-adjoint operator on a Hilbert space
H, and vice versa. This assumption may be dropped, in which case the system is
said to possess superselection rules. The assumption that the observables form
the self-adjoint part mlR := {A Em I A* = A} of a C*-algebra m then naturally
emerges. A crucial point is now that a J LB-algebra is the self-adjoint part of a
C* -algebra.
The state space Scm) of a C* -algebra m(with unit II) consists of all linear
functionals w on m that are positive (that is, w( A * A) ::: 0 for all A E m) and
normalized (i.e., w(ll) = 1). Such states w are automatically continuous, so that
 I. Observables and Pure States 3

S(Qt) c Qt*. The space S(Qt) is equipped with the w* -topology inherited from Qt*.
If A E (0, 1) and WI, W2 E S(Qt), then AWl + (1 - A)W2 E S(Qt). Moreover, S(Qt)
is a closed subset of the unit ball of Qt*. Hence S(Qt) is a compact convex set.
The state space of Qt = 9J1n (C) consists of the density matrices on C N • At the
opposite extreme, so to speak, one can show that the state space ofQt = C(X) for
a compact Hausdorff space X consists of the probability measures on X.
A representation of a C* -algebra Qt is a linear map rr : Qt --+ '13('H), for some
Hilbert space 'H, such that
rr(AB) = rr(A)rr(B); rr(A*) = rr(A)*.
For the J LB-algebra QtIR this means that rr : QtIR --+ '13(1i)1R satisfies
rr({A, Bll) = (rr(A), rr(B)ll; rr(A 0 B) = rr(A) 0 rr(B);

here (A, Bll := i(AB - BA) and A 0 B := !(AB + BA), etc.

There is a remarkable correspondence between states and representations of
a C* -algebra. It is given by the GNS-construction. Given a state W on a C*-
algebra Qt, this construction produces a representation rrw on some Hilbert space
1iw containing a unit vector Q w that is cyclic for rrw(Qt) (that is, rrw(Qt)Q w is dense
in 'Hw). These objects are related by
(Qw, rrw(A)Qw) = w(A) VA E Qt.

Conversely, let a vector Q E 1i be cyclic for some representation rr(Qt). Then

w(A) = (Q, rr(A)Q) defines a state on Qt whose GNS-representation is equivalent
to rr.

Pure States
A state is called pure if it cannot be written as a convex combination of other
states. The set of pure states of a C*-algebra Qt is denoted by P(Qt); any state w
can be approximated by finite sums Li Pi Pi, where Li Pi = 1 and all Pi are pure.
The pure state space of 9J1n (C) and C(X) may be identified with the projective
space lP'C N and with X, respectively.
It is often convenient to look at A E Qt IR as a function Aon P(Qt); this is accom-
plished by putting A(p) = P (A). The map A 1--+ Ais the Gelfand transform. The
ensuing realization of QtIR as a space of functio~s on its pure state space is faithful.
In this realization II A II equals the sup-norm II A 1100 of A over P(Qt).
A representation rr is called irreducible if the set rr(Qt)\II is dense in 'H for
every \II E 'H. The special significance of pure states in the context of the GNS-
construction is that the corresponding representations are irreducible.
The pure states of a classical system are the points of its phase space P. A
manifold P whose associated space of smooth functions COO(P, 1R) is equipped
with a Poisson bracket (satisfying the Leibniz property with respect to pointwise
multiplication) is called a Poisson manifold. Each function h E COO(P, 1R) then
defines a Hamiltonian vector field /;h by
/;h/ = {h, n·
4 Introductory Overview

Hence on P = T*JR3 we have

ah a ah a
; hapi
=aqi- - - --
aqi api'
Hamilton's equations of motion for a curve a (t) in P are
da(t)
----;jt = ;h(a(t));
solutions are called Hamiltonian flows or curves.
Poisson manifolds form the main source of Poisson algebras. The example P =
T*JR3 is special in that at each point a E P the collection of Hamiltonian vector
fields;h spans the tangent space T" P. Poisson manifolds with this nondegenera-
cy property are called symplectic. Traditionally, classical mechanics used to be
described in terms of symplectic manifolds, but many Poisson manifolds that are
not symplectic have turned out to be relevant in physics. A system whose phase
space is not symplectic may be said to possess classical superselection rules.
The most important result in the theory of Poisson manifolds is that any such
manifold admits a (generally singular) foliation by subspaces on which the ;h span
the tangent space. These subspaces therefore acquire a symplectic structure, and
are accordingly called the symplectic leaves of P. Such leaves are characterized
by the properties that any two of their points can be connected by a piecewise
smooth Hamiltonian curve in P and that any Hamiltonian flow must stay within a
given leaf.
The simplest nontrivial illustration is provided by P = JR3, with Poisson bracket
given by {x, y} = z and its cyclic permutations. The symplectic leaves are the
spheres S; of radius r; there is a jump in dimension of the leaves at r = 0,
rendering the foliation a singular one.
We return to quantum mechanics. Let H E 9J'tn (C)IR, and define if E COO(C N)
by
H(IJI) := (IJI, HIJI).

The Hilbert space 'H = C N (seen as a real manifold) has a natural nondegenerate
Poisson structure, characterized by
- - i--
{A, B}I! = h,(AB - BA).

Since a quantum-mechanical state is normalized to unit length and defined only

up to a phase, the space of pure states is the projective space JID1i, rather than 'H.
Fortunately, the considerations above can be transferred to JID1i almost without
modification. In particular, if may be seen as a function H on JID1i, and the above
Poisson structure projects to one on JID1i. The Hamiltonian flow of H with respect
to that structure is then precisely the projection to JID1i of the unitary time evolution
on 'H that solves the SchrOdinger equation with Hamiltonian H.
Given a Poisson manifold P, we define a representation of the Poisson algebra
Ql~ = COO(P, JR) on a symplectic manifold S (with associated Poisson bracket
{, Js) as a linear map JT : Ql~ -+ COO(S, JR) satisfying three properties, of which
 I. Observables and Pure States 5

the two most important are

n({f, g}) = {n(f), n(g)ls; n(f 0 g) = n(f) 0 n(g).
One recognizes the analogy with the definition of a representation of a J L B-
algebra. A representation n of COO(P, R) on S is always associated to a smooth
Poisson map J : S ~ P through n = J*.
A representation n(Qt~) on S is said to be irreducible if at every point a E S
the collection of Hamiltonian vector fields {!;n(f) (a ), f E Qt~} spans the (real)
tangent space Ta S. Interestingly, the notion of irreducibility for representations of
J L B -algebras (and therefore of C* -algebras), looked upon as spaces of functions
on their pure state spaces, can be shown to be identical to the one for Poisson
algebras.
The pure state space P(Qt) of a C* -algebra Qt is a Poisson manifold in a certain
generalized sense; it is foliated by symplectic leaves of the form lP1i", where each
Hilbert space 1i" corresponds to an equivalence class of irreducible representations
of Qt. The basic theorem on irreducible representations is the same for Poisson
algebras of the type COO(P, R), where P is a finite-dimensional manifold (which
we here consider to be the pure state space of COO(P, R)), and C* -algebras (where
commutative C* -algebras are understood to have the zero Poisson structure), where
S = lP1i for some Hilbert space 1i. It is the following: If a symplectic manifold
S carries an irreducible representation n of a C* -algebra or a Poisson algebra Qt]R,
then S must be isomorphic (as a symplectic manifold) to a symplectic leaf of the
space of pure states of Qt]R, or to a covering space thereof. Up to isomorphism, n (f)
is simply the restriction of f to the leaf in question (composed with the covering
projection if necessary).
Saying that lP1i equipped with a certain Poisson structure is the pure state
space of quantum mechanics clearly does not fully characterize this theory. For by
comparison with classical mechanics we know that the observables of quantum
mechanics do not comprise all functions in C OO (1i, JR.), but only those of the form
iI, where H E wtn (C)]R (or lB(1i)]R).
The essential extra ingredient of quantum mechanics is the existence of transition
probabilities between pure states. A transition probability on a set P is a function
p : P x P ~ [0,1] satisfying p(p, a) = I ~ P = a and p(p, a) =
o ~ p(a, p) = O. All transition probabilities in physics are symmetric in
that p(p, a) = p(a, p). The transition probabilities of classical mechanics are
trivial: p(p, a) = 8pa' In quantum mechanics, on the other hand, where P = lP1i,
the function p assumes the form p(rp, 1/1) = 1(<1>, 1JI)1 2 (where the unit vectors
<1>, IJI E 11 project to rp, 1/1 E lP1i).

From Pure States to Observables

We have seen that classical mechanics is described by Poisson algebras of observ-
abIes of the type Qt]R = C OO ( P, JR.), where P is a Poisson manifold. The algebra of
observables of a quantum-mechanical system (perhaps possessing superselection
rules) is the self-adjoint part Qt]R of a C* -algebra Qt, realized as a certain collection
6 Introductory Overview

of functions on the pure state space P = P(~). This space is a generalized Poisson
manifold, which, like its classical counterpart P, is foliated by symplectic leaves.
Classical and quantum mechanics share the property ofunitarity. This means that
the Hamiltonian flow p r+ p(t) generated by a given observable preserves the
transition probabilities, in that p(p(t), a (t» = p(p, a) for all t for which the flow
is defined.
A Poisson space with a transition probability is, roughly speaking, at the same
time a symmetric transition probability space P and a Poisson manifold, such that
the Poisson structure is unitary.
The quantum mechanics of an N -level system, whose algebra of observables
is OOtn(C}R, has the property that its pure state space P = PeN is irreducible as
a transition probability space. In general, a transition probability space is called
irreducible if it is not the union of two (nonempty) orthogonal subsets. A sector
C of a transition probability space P is a subset of P with the property that
p(p, a) = 0 for all pEe and all a E P\ c. Thus a transition probability space
is the disjoint union of its irreducible sectors. In classical mechanics each point of
P is a sector.
The superposition principle of quantum mechanics (which is normally expressed
in terms of vectors in a Hilbert space) can be described in the present language.
For any subset Q of P we define the orthoplement
Q-L:= {a E Plp(p,a) = OVp E Q}.
The possible superpositions of the pure states p, a are then the elements of
{p, a}H. If p and a lie in different sectors, then clearly {p, a}H = {p, a}.
It turns out that the pure state space of quantum mechanics with (discrete)
superselection rules can be characterized (up to technicalities) by the following
three properties (or axioms):
• QMl: The pure state space P is a Poisson space with a transition probability.
• Q M2: For each pair (p, a) of points that lie in the same sector of P, {p, a} H
is isomorphic to Pe2 as a transition probability space.
• QM3: The sectors of (P, p) as a transition probability space coincide with the
symplectic leaves of P as a Poisson space.
Here 1P'C2 is understood to be equipped with the usual Hilbert space transition
probabilities. The universality of the transition probabilities (and, by implication,
of the Poisson structure) of quantum mechanics is notable, as is the third property
(which is not shared by classical mechanics).
To characterize classical mechanics, one simply postulates
• CM1: The pure state space P is a Poisson space with a transition probability.
• CM2: The transition probabilities are p(p, a) = Dpa.
One can reconstruct the algebra of observables ~R from its pure state space,
equipped with the structure of a Poisson space with a transition probability. Given
a general transition probability space (P, p), we first define the real vector space
~R(P) as a certain subspace of the real Banach space l')(.)(P). For simplicity we
 II. Quantization and the Classical Limit 7

assume that (P, p) has a finite basis (here a basis B of P is a pairwise orthogonal
subset for which LpEB p(p, a) = 1 for all a E P). The space Qt]R(P) in question
then consists of all finite linear combinations Li Ai PPi' where Ai E JR, Pi E p,
and pp(a) := pp". This will be the collection of observables, which are seen to
be essentially linear combinations of the transition probabilities.
Axioms QMl and QM2 imply the existence of a spectral theorem in Qt]R(P),
saying that every A E Qt]R(P) has a spectral resolution A := Lj Aj Pej' where the
e j are pairwise orthogonal and the eigenvalues Aj are real. The spectral theorem
equips QtIR(P) with a squaring map, for given the spectral resolution above one can
define A2 by A2 = Lj A]Pej" Subsequently, one defines a map 0 on QtIR(P) by

A 0 B := ~«A + B)2 - (A - B)2).

Axiom QM2 implies that this map is bilinear, so that 0 indeed defines a Jordan prod-
uct. This product, combined with the sup-norm, turns QtIR(P) into a J B-algebra;
the relevant axioms are satisfied as a consequence of the fact that the Jordan product
comes from a spectral resolution. Had the transition probabilities been trivial, this
Jordan product would have been pointwise multiplication, implying associativity.
Given a Poisson structure on p, any function h on P whose restriction to each
symplectic leaf is smooth defines a Hamiltonian flow a f-+ a(t) on P. This defines
a one-parameter family of maps Cit : QtIR(P) -+ QtIR(P), given by CitU) : a f-+
f(a(t)). It is not difficult to show that unitarity (guaranteed by Axiom QMl)
implies that Cit is a Jordan homomorphism; that is, CitU 0 g) = CitU) 0 Cit(g). The
derivative of the homomorphism property with respect to t yields the Leibniz rule,
since

df(a(t)) = {h, f}(a(t)).

dt
Quite unlike the situation in classical mechanics, in quantum mechanics the
Poisson structure of the pure state space turns out to be determined by the axioms
up to a collection of constants (one for each sector). Suitable rescalings then lead to
a single constant n. It is remarkable that the curious associator "identity" is satisfied
by the ensuing Poisson bracket. Therefore, at the end of the day (QtIR (P), 0, {, }, II .
II) becomes a J L B -algebra. This enables one to endow the complexification QtIR (P)
with the properties of a C*-algebra, of which QtIR(P) is the self-adjoint part. In
analogy with classical mechanics, the algebra of observables QtIR(P) is realized
(even as a Banach space) as a subspace of fOO(P, JR).
In passing from pure states to algebras of observables one has the correspon-
dences listed in Table 1.

II. Quantization and the Classical Limit

The second chapter relates classical and quantum mechanics to each other. Such
a relation is possible on the basis of the structural similarities between the mathe-
8 Introductory Overview

Pure state space Algebra of observables

transition probabilities Jordan product

Poisson structure Poisson bracket
unitarity Leibniz rule

TABLE 1. From pure states to algebras of observables

matical description of these theories laid out in Chapter I, and it can be approached
from the point of view of either observables or pure states.

Foundations
The problem of quantizing a given classical system is as old as quantum mechanics
itself. Initially, the term "quantization" indicated the fact that at a microscopic scale
certain physical quantities assume only discrete values, sometimes called quantum
numbers. This was found to be true particularly for energy levels of bound states,
as well as for, e.g., angular momentum and electrical charge. Such discreteness
is easily understood within the Hilbert space formalism of quantum mechanics,
where self-adjoint operators mayor may not have a discrete spectrum, and is no
longer seen as the defining property of a quantum theory.
In the modem literature "quantization" refers to the passage from a classical to a
"corresponding" quantum theory. This notion goes back to the time that the correct
formalism of quantum mechanics was beginning to be discovered, and from that
time to the present day practically all known quantum-mechanical models have
been constructed on the basis of some quantization procedure. Nonetheless, Barry
Simon wrote:
It seems to me that there has been in the literature entirely too much emphasis
on quantization (i.e. general methods ofobtaining quantum mechanics from
classical methods) as opposed to the converse problem of the classical limit
of quantum mechanics. This is unfortunate since the latter is an important
question for various areas of modern physics while the former is, in my
opinion, a chimera.
In the present book the conception of quantization used in this quotation, which
indeed applies to geometric quantization and related approaches, is replaced by a
different one: We see quantization as the study of the possible correspondence be-
tween a given classical theory, given as a Poisson algebra or a Poisson manifold and
perhaps a Hamiltonian, and a given quantum theory, mathematically expressed as
a certain algebra of observables or a pure state space, and perhaps a time evolution.
For this purpose it is not at all necessary that the quantum theory be formulated in
terms of classical structures. On the basis of this understanding quantization and
the classical limit are two sides of the same coin.
Early thought on both quantization and the classical limit was guided by Bohr's
"correspondence principle", which was a rather vague idea to the effect that quan-
 II. Quantization and the Classical Limit 9

tum mechanics should converge to classical mechanics for Ii ~ 0, and also in

the limit of large quantum numbers. The second aspect, including its relation to
the first, will be studied in Chapter III. The use of the limit Ii ~ 0 is sometimes
criticized on the argument that nis a constant, but what is meant here is simply that
n should be small compared to other relevant quantities of the same dimension;
this includes the case where units have been chosen in which n is dimensionless
and equal to I!
In Chapter I classical and quantum mechanics are formulated in such a way
that they look structurally similar for any value of n. On the observable side one
classically has a Poisson algebra (Q(~, 0, {, D, in which ° is associative, whereas
n
quantum-mechanically one has the self-adjoint part (Q(~, On, {, }r,), -=J- 0, of a
CO-algebra. One now needs a proper way of expressing the idea that (Q(~, ... ) is
the quantization of (Q(~, ... ), and that the latter is the classical limit of the former.
For this to be possible in the first place, the quantum algebra of observables Q(~
must be defined for all values Ii E 10, where 10 is a certain subset ofR that has as
an accumulation point (10 may be discrete, e.g., 10 = {lIn, n EN}, or an interval,
°
such as 10 = (0, 1]; another example would be 10 = R\{O}).
The essence of quantization is now that there should be a family of linear maps
Q/t : Q(~ ~ Q(~, n E 10; the operator Q,,(f) is interpreted as the quantum
observable corresponding to the classical observable f. A mathematically precise
version of Bohr's correspondence principle, at least as far as the algebraic structure
is concerned, is then given by the conditions

and

for all f, g E Q(~; here a possible Ii-dependence of the operations in Q(~ has been
indicated. Together with the continuity of Ii t-+ II QnU)lIn for all f E Q(~, these
conditions define what is meant by a strict quantization.
From the perspective of pure states the classical theory is characterized by a
Poisson manifold (P, {, }). Quantization should relate this to a family of Poisson
spaces with a transition probability (P n, p, {, In), n E 10, satisfying the "QM"
axioms of Chapter I. This relation is given by a pure state quantization, which is
a collection of injections qn : P ~ Ph (Ii E 10 ) that embed the classical pure state
space into its quantum counterpart. These maps should satisfy certain conditions
motivated by the correspondence principle. One such condition is obviously

stating that the quantum-mechanical transition probabilities converge to the classi-

cal ones. It is interesting to relate this condition to the one on the Jordan product of
observables. Assume that P is discrete; then Q(o = eo(p) is generated by functions
of the type p~ : p t-+ opa. Given a pure state quantization q", we can hope to define
a strict quantization Qf of Q(~ by linear extension of Q~ (p~) := Pqh(a). The spec-
\0 Introductory Overview

tral resolution of I E Qt~ is I = La I(a)p~, so that Qff(f) = La I(a)pqh(a)'

For small Ii the right-hand side approximates the spectral representation of Qff(f),
since the qh(a) become almost orthogonal. Therefore, Qff(f)2 is approximately
La l(a)2 Pqh (a), which equals Qff(f2). Hence Qff(f)2 -+ Qff(f2) for Ii -+ 0,
which is equivalent to the condition on the Jordan product.
For nondiscrete P the notion of a pure state quantization has been worked out
only when P = 5 is symplectic and each Ph is irreducible, being equal to F'1ih
for some Hilbert space Jih. In the cases we consider, the sum over points in P is
then replaced by the Liouville measure {LL on 5, locally given by d{LL(P, q) :=
d n pd nq/(2JT)n. In addition, a function c : 10 -+ lR\{O} appears. The conditions
on a pure state quantization qh are stated in terms of the Berezin quantization of
I E Qt~ = Co(5, lR). This is an operator Qff(f) on Hh, defined (for each Ii E 10 )
by its expectation values

(\11, Q~(f)\I1) := c(1i) Is d{LL(a) p(qh(a), 1/I)/(a),

where 1/1 E F'1ih is the projection of the unit vector \11 to F'1ih. This expression
evidently generalizes Qff (f) in the previous paragraph. The function c is fixed by
imposing the first condition Qff(ls) = IT. The second requirement on qh is that in
the limit Ii -+ 0 the above expression with 1/1 = qtlp) converge to I(p) for all
I E Qt~ and all p E 5. Finally, each qh should pull the canonical symplectic form
on F'1ih back to the one on 5.
Let us assume that each qh(a) E F'1i" is the projection of a unit vector \11K E Htt •
The map W : Ji" -+ L2(5, c(Ii){Ld defined by WIJI(a) := (IJIK, IJI) is then a
partial isometry. Defining p to be the projection onto the image of W, and U to be
W, seen as a map from Hh to pL2(5, C(Ii){LL), we obtain
UQff(f)U-1 = pip,
where I is seen as a multiplication operator on L 2(5, c(Ii){Ld. In this way, quantum
observables act on a subspace of L 2 (phase space), rather than on L 2(configuration
space), as is more usual in quantization theory, in an extremely elegant fashion.

Quantization on Flat Space

Our main illustration of strict as well as pure state quantization will come from the
manifold 5 = T*lR n , equipped with its canonical Poisson bracket; this makes 5
symplectic. This manifold is particularly well structured in being both a cotangent
bundle and a Kahler manifold (the latter comprise a class of complex manifolds of
which more examples will be encountered in the next chapter). It turns out that a
Kahler manifold often admits a strict Berezin quantization, which is derived from a
pure state quantization as explained above. The observables on cotangent bundles,
on the other hand, are best quantized using a prescription going back to Weyl,
which is not directly related to a pure state quantization. The phase space T*lR n ,
then, may be quantized either way; Berezin quantization enjoys the advantage of
positivity, whereas Weyl quantization has better symmetry properties.
 II. Quantization and the Classical Limit 11

In both methods the Heisenberg group Hn plays a central role; this is the
connected and simply connected Lie group whose Lie algebra ~n = JR2n + 1 is
described by
[Pi, Qj] = -8/ Z; [Pi, Z] = [Qj, Z] = 0,
in terms of a suitable basis {Pi, Qj, Z}i,j=l, ... ,n' The Heisenberg group is nilpotent,
and the exponential map Exp : ~n -+ Hn is a diffeomorphism. For each n i= 0
there exists an irreducible representation Vi on 7t = L 2(JRn), given by
h

Vi (Exp(-uQ
h
+ vP + tZ»\lI(x):= e-i(t+luv-ux)/h
2 \lI(x - v),

where u Q := Ui Qi, etc. Of special significance are the Weyl operators

Vi (p, q) := Vi (p, q, 0) = e*(PQ~-qpi),
h h

where Q~,i = Xi and Pl. i = -i na / axi are the position operator and momentum
operator of elementary quantum mechanics.
Both Berezin quantization Qff and Weyl quantization Q';i are defined for n E
10 = JR\{O}, and map ~~ = Co(T*JRn,JR) into ~Il = 1B0(L2(Rn»]R (the self-
adjoint part of the C*-algebra of compact operators on L 2(JRn». Both are given by
an expression of the form

For Weyl quantization one puts A = 2n P, where P is the (nonpositive) parity

operator P on L 2(Rn), defined by P\lI(x) := \lI( -x). To obtain Berezin quanti-
zation one chooses the positive operator A = [\lI~], which is the projection onto
the (unit) vector
\lI~(x) := (nn)-n I 4 e -x 2/ (2h),

The pure state quantization q:

associated with Berezin quantization is given by
7th = L2(JRn) for all n i= 0, and q:(p, q) = 1/I~p,q), where the right-hand side is
given by projecting the unit vector

\lI~p,q):= Vi(P,q)\lI~
h

to JPlL 2 (Rn), In terms of z := (q + ip)/./2, the transition probabilities between

quantized pure states are
p(q:(z), q:(w» = e-lz-wI2/t"
which evidently converges to the classical transition probability 8zw as n -+ O.
The Hilbert space L 2(T*JRn, I1-d is naturally isomorphic to L 2(Cn, I1-G), where
I1-G is a suitable Gaussian measure on C n. The projection p in the preceding
section then projects on the subspace of functions on Cn that are entire in Z.
Accordingly, Berezin quantization on flat space assumes the pulchritudinous form
12 Introductory Overview

of sandwiching a multiplication operator on L 2(Cn , /Lc) between two identical

projections, whose image is a space of entire functions.
A comparison between classical dynamics and its quantum counterpart is of
central importance to the theory of quantization and the classical limit. If the
classical Hamiltonian h lies in 2(~, this comparison is straightforward. In that
case, the quantum Hamiltonian Hh := Qh(h) lies in 2(~, and Dirac's property
implies that for fixed f and for all ! E 2(~ one has

lim II Qh(a?U» - a~(Qh(f»11 =

/i,-+O
o.
Here a?U) : u 1-+ !(u(t», and a;(A) := eitHn/h Ae-itH,,/h.
It so happens that most Hamiltonians on T*JR.n used in physics are unbounded,
so that the above norm-convergence is somewhat unrealistic. A silver lining on this
generic unboundedness, however, is the fact that for Hamiltonians that are at most
quadratic in the canonical variables (p, q) the excellent equivariance properties
ofWeyl quantization imply that Qr (a?(f» - a?(QhU» = 0 for any It. For Qf
instead of Qr this equation holds for Hamiltonians that in addition are O(2n)-
invariant.
Convergence from quantum to classical dynamics for more general unbounded
Hamiltonians may be achieved by looking at the time evolution of particular
pure states. Most literature on this subject is concerned with the time-dependent
WKB method, where one assumes that the initial wave function is of the form
\II h(x) = Ph (x) exp(i S (x) / It), where S is real and independent of It, and Ph is a real
formal power series in It (of which only the zeroth -order term is relevant in the clas-
sicallimit). An approximate solution \IIh(X, t) to the time-dependent Schr6dinger
equation is then constructed in terms of a classical trajectory between x(O) and
x(t) = x, where x(O) is determined by the requirement that the trajectory with
initial data (d S(xo), xo) indeed arrives at x after time t. Such initial pure states are
quite peculiar, since in the classical limit they typically converge to mixed states on
2(0: The support of the mixed state on Co(T*JR.n) in question, which is a probability
measure on T*JR.n, is the so-called Lagrangian submanifold of T*JR.n defined by S
and Po (this is the collection of points (d S(q), q), where q E supp (Po». Moreover,
the WKB method works without further ado only if the projected flow defines a
diffeomorphism of the configuration space JR.n for all t' E [0, f].
We concentrate on a different method, which works well if the initial state
\II h converges to a pure state in the classical limit. We will specifically look
at the (coherent) state \II~p,q) defined earlier, whose classical limit is the point
(p, q) E T*JR.n. The method is based on Taylor-expanding the quantum Hamilto-
nian H = H(Pt" Q~), which, up to suitable ordering, is obtained by substituting
(Pt" Q~) for (p, q) in the classical Hamiltonian h(p, q) around the classical tra-
jectory (p(t), q(t». This method works well for classical Hamiltonians of the
type

h(
p,q
) = (p - eA(q»2
2m
+ V( q,)
 II. Quantization and the Classical Limit 13

which describe a particle moving in an external potential V and magnetic field

V x A. One can prove that for all f E Ql~, and Qh = QJi'
or Qh = Qf" one has

lim ('I1Ap.q), [Qh(a~(f» - a~(Qh(f»] 'I1}:.Q») = o.

h->O

Quantization on Riemannian Manifolds

According to the general theory of relativity, gravitational fields are described by
a pseudo-Riemannian metric g on spacetime. To describe the motion of a test
particle in a static external gravitational field we therefore assume that space is
a Riemannian manifold (Q, g); the corresponding phase space is the cotangent
bundle S = T* Q, whose canonical symplectic structure is independent of g. The
metric provides an isomorphism between T* Q and the tangent bundle T Q, and
it turns out to be easier to discuss mechanics on T Q. The natural Hamiltonian on
TQ is
h(v,q) = ~gij(q)Vivj.

The Hamiltonian flow (v(t), q(t» on T*Q is known as geodesic motion, since q(t)
is a geodesic on Q; the tangent vector to this geodesic is v(t), which is parallel
transported along the geodesic.
Using the geometric structure, it is possible to generalize the Weyl quantiza-
tion method on the flat space ]Rn to any Riemannian manifold. The key to this
generalization lies in rewriting Weyl's prescription as

Q:;(f)'I1(x) = r
Ji*.n
dny KfLfJ(x, y)'I1(y),

with kernel Kf LfJ(x, y) = n-n J«x - y)/n, ~(x + y». Here J(v, q) is the partial
Fourier transform of f (p, q) in the fiber direction of T* Q; this is a function on T Q.
We now recognize ~ (x + y) as the midpoint of the geodesic connecting x and y, and
(x - y) as its tangent vector at this midpoint; the map (x, y) 1-+ «x - y), ~(x + y»
provides a diffeomorphism between lRn x lRn and TlRn.
When (Q, g) is complete (in that the motion generated by h is defined for all
times), and in addition has the property that any two points are connected by a
unique geodesic, one has Q :::::: lRn as a manifold. Moreover, the obvious gener-
alization of the geodesic construction above provides a diffeomorphism between
Q x Q and T Q. In general, one has to proceed locally, using the geodesic midpoint
construction to obtain a diffeomorphism between a neighborhood of the diagonal
embedding o(Q) in Q x Q and the zero section Q in TQ. On a suitable choice
of functions in Ql o = Co(T* Q), this still enables one to generalize the Weyl pre-
scription to obtain a strict quantization map QJ:' .For suitable (real) f, the operator
QJ:' (f) is a compact (self-adjoint) operator on L2(Q) (defined with respect to the
canonical Riemannian measure on Q).
The single most important property of QJ:' is that it is equivariant under isome-
tries. To explain this, we first note that the group Diff(Q) of diffeomorphisms of
Q acts on T* Q by pullback; call this action pO. Accordingly, each cP E Diff( Q)
14 Introductory Overview

defines an automorphism ag(f) = f 0 ({J* of ~o. Furthermore, there is a natu-

ral representation ph of Diff( Q) on L 2( Q), which defines automorphisms a; on
~ = l}3o(L 2( Q», as explained before in the context oflRn. If, then, ({J is an isometry
of (Q, g), for all suitable f one has
a~ (QJi(f») = QJi(ag(f».
It is possible to extend QJi to certain unbounded classical observables, in par-
ticular to functions that are polynomial in the canonical momenta. The Weyl
quantization of the classical Hamiltonian is
QJi (h) = -ih2(~ - ~R),
containing not only the Laplace-Beltrami operator ~, but picking up an additional
term proportional to the Ricci scalar R. If (Q, g) is complete and R is bounded, this
quantum Hamiltonian is essentially self-adjoint on the domain C~(Q) C L2(Q).
However, even when these conditions are not met one can prove results on the
convergence of quantum to classical dynamics similar to those in the flat case.

III. Groups, Bundles, and Groupoids

In Chapter III we construct Poisson algebras and C* -algebras from well-known
geometric objects, namely Lie groups and their Lie algebras, and principal fiber
bundles and their associated "infinitesimal" objects. These Poisson and C*-
algebras tum out to be related by a strict quantization. The theory of Lie groupoids
and algebroids then provides a perspective unifying these seemingly diverse classes
of examples, as well as providing new ones.

Lie Groups and Lie Algebras

Let 9 be a Lie algebra. The (minus) Lie-Poisson structure on the dual g* is given
by the Poisson bracket

{X, Yl- = -[X, V],

where each X E 9 defines a linear function X(e) := e(X) on g*. Physically, the
associated Poisson algebra COO(g~, 1R) is the classical algebra of observables of
an immobile particle whose only degrees of freedom are "internal". For example,
when 9 = .50(3) it describes a spinning particle, the magnitude of whose spin is
not fixed.
In the spirit of Chapter lone may then look for representations of COO(g~, 1R)
on a symplectic manifold S. Such a representation corresponds to a Poisson map
J : S -+ g~. The representation theory of COO(g~, 1R) is closely related to the
existence of g-actions on S, i.e., homomorphisms X f-+ ~x from 9 into the space
of vector fields on S. For given such a representation, one finds a g-action by
h := ~lx' with Jx := J* X. Conversely, ag-action X f-+ ~J'x generated by some
 III. Groups, Bundles, and Groupoids 15

smooth map J : S ~ g~ in this way is called Hamiltonian, and J is known

as a momentum map for the action. It does not follow that the momentum map
of a Hamiltonian g-action is a Poisson map with respect to the Poisson bracket
displayed above: In general, one has
{lx, Jy}s = -J[x.Y] - f'(X, y),
where r is constant on S, defining a so-called 2-cocycle on g.
A smooth action of a Lie group G on a manifold S leads to a g-action through
~xf(a) = df(Exp(tX)a)Jdtlt = 0; the G-action is said to be Hamiltonian when
the associated g-action is. When J : S ~ g~ is a Poisson map (i.e., r = 0), the
G-action and the g-action are called strongly Hamiltonian, and J is said to be
equivariant.
The coadjoint action Co of G on g* is the dual of the adjoint action. The
main theorem on the Lie-Poisson structure is that the symplectic leaves of g~
are precisely the coadjoint orbits. This endows the coadjoint orbits with the Lie
symplectic structure. For example, the coadjoint orbits in 50(3)* are two-spheres;
picking an orbit fixes the magnitude of the classical spin. More generally, a coad-
joint orbit plays the role of a classical charge. When G is abelian, as in the theory
of electromagnetism (where H = U (1 )), the charge is just a number. The signifi-
cance of the coadjoint orbits in representation theory is that (up to covering spaces)
every irreducible representation 1fo of the Poisson algebra COO(g~, 1R) is realized
on such an orbit O.
In quantum mechanics the focus is on G-actions on a projective Hilbert space
JID?t. These actions should not merely respect the Poisson structure on JID?t, but
must in addition preserve the quantum-mechanical transition probabilities. Such
G-actions on JID?t turn out to be given by linear unitary G-"actions" U on 11. itself,
which satisfy U(x)U(y) = c(x, y)U(xy). Here c : G x G ~ U(l) is a so-called
multiplier, which measures to what extent U differs from a true representation of
G. A multiplier on G is the "global" analogue of a 2-cocycle r on the Lie algebra
g. Indeed, let J be the momentum map of the associated action on JID?t; it is given
by
lx(1/I) = in(\II, dU(X)\II),
where the unit vector \II is a lift of 1/1 E JID?t to 11., and
d
dU(X) := dt U(Exp(tX))lt=o.

The presence of a multiplier in the G-action on 11. is then reflected by

{ix, Jy}1l = -J[X,y] - nr(X, y).

What is the quantum-mechanical counterpart of the Lie-Poisson algebra

COO(g~, 1R)? This turns out to be the group C*-aIgebra C*(G). Here the shift
from the "infinitesimal" object g* to the "global" object G in passing from classi-
cal to quantum mechanics is typical. To define C*(G) (for a group whose left and
right Haar measures dx coincide, for simplicity) one starts from the convolution
16 Introductory Overview

operation on, say, Cgo(G), that is,

f * g(x):= i dy f(xy-I)g(y).

One adds an involution f*(x) := f(x- 1) so as to tum Cgo(G) into a * -algebra. In

the associativity of convolution and the involutive nature of f 1--+ f* this algebra
reflects the corresponding properties (xy)x = x(yz) and (xy)-l = y-l X-I of the
group G itself. One then equips Cgo(G) with an appropriate nonn, and closes it so
as to obtain the C*-algebra C*(G).
The quantum analogue of the correspondence between g-actions on symplectic
manifolds and representations of COO(g~, ~) is then a basic theorem about C*(G),
stating that there is a bijective correspondence between representations U of G
on Hilbert spaces and nondegenerate representations rr of C*(G) as a C*-algebra,
given by rr(f) = Ie dx f(x)U(x).
When G is compact there is a neat quantum analogue of the decomposition of
g~ as the union of its symplectic leaves (which, as we saw, are just the coadjoint
orbits). From the Peter-Weyl theorem, one has the decomposition
C*(G) ~ EBm1 dy (C),
YEG

where G is the space of all (equivalence classes of) irreducible representations of

G, and d y is the dimension of a given such representation.
The analogy between the Poisson algebra COO(g~, JR) and the C*-algebra C*(G)
is further illustrated by the construction of a strict quantization relating the two.
One here chooses Qlo = Co(g*) and QlJj = C*(G), and, roughly speaking, defines
the quantization map QJj : Ql~ ---+ Ql~ by

Qt,(f)(Exp(X» := 1 0*
dn()
--
(2rr fon
i
e,,8(X) f«().

For compact or nilpotent Lie groups one can show that this indeed defines a strict
quantization. The nature of this prescription may be illustrated by the fact that in
any representation rr of C*(G) one obtains (transgressing the realm of bounded
operators)
rr(Qh(X» = iMU(X).
Hence from the Lie-Poisson bracket above and the property [dU(X), dU(y)]
= dU([X, Y]) one immediately verifies that

~ [rr(Qh(X»,rr(Qh(Y»] = rr(QJj({X, Y}_».

Rather than COO(g~, R) one may try to quantize the Poisson algebra COO(O, R),
where 0 is a coadjoint orbit in g*. For compact G this is indeed possible, with the
interesting feature that Planck's constant is "quantized"; this reflects the compact-
ness of the classical phase space O. One starts from an irreducible representation
U y (G) on a (finite-dimensional) Hilbert space 1i y labeled by a highest weight y,
 III. Groups, Bundles, and Groupoids 17

with highest weight vector wy • Let J : lP1{y ~ g~ (as displayed above) be the
momentum map of the G-action on lP1{ associated to the representation U on 'H.
One then has to assume that 0 contains J (1/1 y). When this is the case, for k E Z
one defines 'Hh for Ii = II k as 'H~/t, (i.e., the carrier space of the representation
with highest weight ky), upon which the map qh : 0 ~ lP1{h, defined by

qh(CO(X)y) := rH h ->lP1ih(Uy/h(X)W y/h),

is a pure state quantization of 0 on 10 := liN. The associated Berezinquantization

Qg then turns out to be strict. It is also G-equivariant: With a~(f) := f 0 Co(x- 1)
for arbitrary x E G and f E COO(O), one has

Qf/k(a~(f» = Uky(x)Qf/k(f)Uky(X)*.

Internal Symmetries and External Gauge Fields

The description of purely spatial degrees of freedom of a single particle having
been given in Chapter II, and the treatment of purely internal variables having just
been sketched, the goal is now to combine these.
The appropriate mathematical tool is the theory of principal fiber bundles.
When Q is a manifold and H a Lie group, a principal H -bundle P is defined by a
free H -action on P and by a projection r : P ~ Q. These must be such that locally
P :::::: Q x H, relative to which the H -action becomes the canonical right action
on the second variable, and r is projection onto the first. In particular, PI H :::::: Q.
This setup is the starting point for the classical as well as the quantum theory of a
particle that moves on Q and has internal degrees of freedom related to H.
More generally, a bundle over a manifold Q with typical fiber F is a space B
with a projection r : B ~ Q such that locally B :::::: Q x F, and r is projection
onto the first variable. Apart from principal bundles, where F is a Lie group, an
important class is formed by vector bundles, where F is a vector space. A section
of B is a map s : Q ~ B for which r 0 s = id.
A most important concept, used in classical as well as in quantum mechanics, is
that of an associated bundle: Given a smooth H -action L on some manifold M,
the associated bundle M = P X H M is (P x M) I H, where the H -action on P x M
defining the quotient is given by h : (x, m) ~ (xh- 1 , Lh(m». This is a bundle over
Q with typical fiber M. The projection rM->Q is given by rM->Q([x, m]H) = rex).
The classical theory is based on the Poisson manifold (T* P)I H. Here the H-
action on T*P is the pullback of the given action on P; the canonical Poisson
bracket on the cotangent bundle T * P quotients to one on (T* P) I H , defining its
Poisson structure. The symplectic leaves of (T*P)I H are of the form J-1(0)1 H,
where J : T*P ~ ~~ is the momentum map of the associated ~-action on T*P,
and 0 is a coadjoint orbit in ~*. The choice of an orbit 0 specifies a classical
charge; the orbit contains internal degrees of freedom, which in physics couple to
an external gauge field. There is a correspondence
18 Introductory Overview

between the coadjoint orbits 0 in £:1*, here in the guise of the irreducible representa-
tions no of COO(£:1~, lR), and the irreducible representations of the Poisson algebra
COO«T*P)/ H, lR): For each such orbit one obtains an irreducible representation
nO on the symplectic manifold (T*P)o := J-1(0)/ H.
A symplectic leaf (T* P)o is locally of the form T* Q x O. However, to separate
the spatial and internal degrees of freedom in an intrinsic fashion, one needs to
choose a connection on P. This is a decomposition of each tangent space Tx P
into an (intrinsically defined) vertical subspace (which projects to zero under i),
and a complement, called the horizontal tangent space at x. Choosing such a
decomposition turns out to be equivalent to the specification of an £:1-valued 1-
form A on P, with certain properties. The part of A that lives on Q (relative to a
local factorization P ~ Q x H) is the physicist's gauge field or Yang-Mills field.
Let us introduce the manifold
P *Q T*Q := {(x, a) E P x T*QI ip~Q(X) = ipQ--->Q(a)}.
This is a principal H -bundle over T* Q if one defines its projection to be the
one onto the second variable, and its H -action to be essentially the H -action
on P. Choosing a connection then leads to the realization of (T* P)o as a bundle
associated to P*Q T* Q by the coadjointrepresentation of H on O. In this realization
the Poisson bracket on (T*P)o depends on A.
The basic tool in the construction of (unbounded) physical observables on the
phase space (T* P)o is the group Aut(P) of automorphisms of the bundle P; this
group consists of those diffeomorphisms on P that commute with the H -action.
Any diffeomorphism on P pulls back to one of T* P; a bundle automorphism in
addition maps J- 1(0) into itself, and quotients to a Poisson map on (T*P)o. The
momentum map for this reduced action p~ of Aut(P) on (T* P)o then gives the
classical observables that are linear in the (conventional) momentum. Functions
of the configuration variable q are more easily obtained, namely from the natural
projection i(pp)O--->Q.
From the perspective of ("classical") representation theory the symplectic space
(T* PP therefore plays a double role: It firstly carries the irreducible representation
nO of the Poisson algebra COO«T*P)/ H, lR), and secondly it supports the Poisson
action p~ of the group Aut(P).
To specify a "natural" Hamiltonian hO on all leaves (T*P)o in one go, one
needs a Riemannian metric gQ on Q and a connection A on P, as above. In the
A-dependent realization of (T* P)o as the associated bundle (P *Q T* Q) x H 0
one then simply puts
hO(p, q, (J) = 4g1/(q)PIlPv.

For simplicity this has been expressed in local coordinates, but hO is an intrinsically
defined function. In the original definition of (T*P)o as a subspace of (T*P)/ H
this reads
h~(p, q, (J) = 4g~v(q)(pll - (JiA~(q»(pv - (JjA~(q».
The associated equations of motion are the so-called Wong equations.
 III. Groups, Bundles, and Groupoids 19

We now turn to the associated quantum theory. For compact H an appropriate

(complexified) quantum algebra of observables is ~o(L 2(p»H, the C* -algebra of
compact operators on L 2(P) (defined with respect to some H-invariant measure
equivalent to the Lebesgue measure) that commute with the representation U R of
H, given by

On the basis of the quantization theory on Riemannian manifolds in Chapter II

one shows that ~O(L2(p»H is a strict quantization of Co«T*P)/ H). Noticing
that Co«T*P)/ H, JR) ~ Co(T*P, JR)H, the associated quantization map is simply
given by restriction of the Weyl quantization map Q~ on T*P (where P has been
equipped with an H -invariant Riemannian metric).
From a representation-theoretic viewpoint, the quantum counterpart of (T*P)o
is a Hilbert space 'H. x , constructed as follows. One starts with a representation
U x (H) on a Hilbert space 'H. x , and then considers the vector bundle HX := P x H 'H. x
associated to P by the representation U x; locally HX ~ Q x 'H. X. In contrast with
the classical situation, the relevant object is not this associated bundle itself, but
rather its space of smooth sections r(HX) (with compact support). This space may
be realized as a space of maps ",x : P -+ 'H. x satisfying the equivariance condition
",X(xh-') = Ux(h)"'X(x)
for all x E P and h E H. Exploiting the fact that the fiber 'H. x is a Hilbert space,
one can equip r(HX) with an inner product; its closure is 'H.x.
In analogy to the classical situation, Jix plays a double role in representation
theory. Firstly, it carries a representation n x of the C* -algebra ~o(L 2(p»H , given
by

nX(K)"'X(x) = i dJi,(Y) K(x, y)"'X(y).

This representation is irreducible iff U x (H) is irreducible, so that we obtain a

correspondence
UX(H) +--+ nX(~o(L2(p»H)

analogous to the classical correspondence no B- nO. Secondly,1i X carries the

induced representation U x (Aut(P», defined by

dv(cpQ'(r(x))) ",X(cp-'(x»,
dv(r(x»

where v is a measure on Q that is naturally defined by Ji" and CPQ is the

diffeomorphism of Q associated to cP in the obvious way.
The well-known construction of induced representations of a (Lie) group G is
a special case: one takes P = G, defined as a bundle over Q = G / H through
the canonical right action of H C G on G. The left action of G on itself then
realizes G as a subgroup of Aut(G), so that the equation above applies. This is
20 Introductory Overview

called Mackey induction. When v is G-invariant we obtain

UX(y)\IIX(x) = \IIX(y-IX).
We return to the case of a general bundle P. The correspondence between the
classical and the quantum theory is further illustrated by the remarkable equation
iIu1UX(~) = JrX(Qf(1~».
Here ~ is an element of the Lie algebra of Aut(P), that is, an H -invariant vector
field on P, and J is the momentum map for the Aut(P)-action on T*P. Since
the latter commutes with the H -action, its momentum map is H -invariant, so
that each J~ lies in Coo(T* P, JR)H. Extending the definition of Qf to suitable
unbounded functions, the right-hand side is therefore well-defined. When U x (H)
actually corresponds to a coadjoint orbit 0, one may regard JrX(Qf(1~» as the
quantization of Jp,but in the absence of such a correspondence the right-hand
side still makes sense as the quantum "~-momentum" in the sector X . In particular,
one has
i
/i[Jrx(Qf (1~», Jrx(Qf (17]))] = Jrx (Qf ({J~, J7]}))'

The quantum Hamiltonian on fix defined by Weyl quantization is

H; = -1fl,2 (~~ - iRQ + 12F2 - CX) .
Here ~ ~ is a gauge-covariant Laplacian, and the other terms are geometric objects
acting as multiplication operators, all constructed from gQ and A.
As in Chapter II, the possible convergence of the classical equations of motion
generated by h~ to their quantum counterparts generated by H; can be analyzed.
One has to find a suitable analogue of the coherent states \IIt q) used for this
purpose when only spatial degrees of freedom are present. Our discussion on the
quantization of the Poisson algebra Coo(O, JR.), where 0 is a coadjoint orbit in
fl*, suggests how to proceed. We assume that 0 is associated with an irreducible
representation U x (H), labeled by a highest weight X, in that 0 contains J (1/J x)·
For "quantized" Ii = 1/ k, kEN, we then replace \II~p,q) by the unit vectors
W(p,q,h) '- \II(p,q) .0, U (h)\II
Ilk .- Ilk I(Y kx kx

in L 2(l~.n) ® 7-lkx, where WkX is a normalized highest weight vector in fikx' The
desired convergence may then be shown for k --+ 00. The proof makes essential
use of the G-equivariance of the Berezin quantization of Coo(O, JR).
Everything said so far may be explicitly calculated in the simplest nontrivial
example, where the bundle P(Q, H) is SO(3)(S2, SO(2». This bundle supports
a certain canonical connection, which in physics terms describes the field of a
magnetic monopole sitting at the origin. The symplectic leaves in .50(2) = JR. are
just numbers e, identified with the electric charge of the particle moving on S2.
The symplectic leaves (T" S 0(3)Y are diffeomorphic to T* S2, but one still sees
the effect of a nonzero charge e in all relevant quantities, such as the momentum
map for the reduced SO(3) action on (T* SO(3)Y.
 III. Groups, Bundles, and Groupoids 21

j j
C*(G)

FIGURE 1. Groupoids and algebroids in quantization

The quantum theory is formulated in terms of the line bundles Hn , nEZ,

defined by the irreducible representations Un(a) = exp( -ina) of SO(2). These
representations are the quantum analogues of the coadjoint orbits e, illustrating
the quantization of electrical charge. It should be mentioned that the Hilbert space
1t n is insensitive to the topology of the line bundle Hn; the relevance of this bundle
in quantum mechanics lies in the fact that the space of sections r(Hn ), which does
"see" the topology, provides a domain of essential self-adjointness for the basic
quantum observables.

Lie Groupoids and Lie Algebroids

Our aim is to explain Figure 1. Let us first look at a case where we already know
what all the entries and arrows mean, namely when G = G is a Lie group and
18 = 9 its Lie algebra. In that case, we have seen that we can canonically associate
a Poisson algebra COO(g~, JR) with 9 and a C*-algebra C*(G) with G, in such a
way that under favorable circumstances (e.g., when G is compact), C*(G) is a
strict quantization of COO(g~). The central ingredient in the construction of the
quantization map Q~ = Q/i was the usual exponential map Expw = Exp : 9 -+
G.
The quantization of a system with configuration space Q fits into this diagram
as well. We already know three of the four comers: For 18 we would like to read the
tangent bundle T Q, with associated Poisson algebra Coo (T* Q, JR), and we wish
C* (G) to be the algebra of compact operators 23o(L \ Q».
The object G should then
be chosen as Q x Q, equipped with structures such that one may firstly construct
T Q as an associated infinitesimal object (in analogy to the construction of a Lie
algebra from a Lie group), and secondly can define the C*-algebra C*(Q x Q)
through the construction of a convolution and an involution on C~(Q x Q).
The appropriate starting point is the concept of a groupoid. This is a gener-
alization of a group, in which multiplication is only partially defined. When it is
defined, it is associative, and each element has an inverse. For example, one may
say that two elements (q\, q;) and (q2, q~) of Q x Q can be multiplied iff q; = q2,
in which case (q\, qD(q;, q~) := (q\, q~). This reflects the way arrows are com-
posed; one therefore interprets a point (q, q') E Q x Q as an arrow from q' to q.
22 Introductory Overview

Thus the inverse is (q, q,)-I := (q', q). The ensuing object is the pair groupoid
on Q.
Every groupoid G may be thought of as a collection of arrows connecting points
on some space Q, called the base of G. The collection of all elements of the form
y y -I , Y E G, is naturally isomorphic to the base Q, and this isomorphism leads
to an inclusion t : Q ~ G. Arrows in t(Q) evidently start and end at the same
point. In a pair groupoid one has t(q) = (q, q).
In a group all arrows start and end at the unit e, so that any two elements may be
composed. An intermediate possibility is an action groupoid. Given a group G and
a G-action on a set Q, we look at G x Q as a collection of arrows between points in
Q, in such a way that (x, q) starts at x-1q and ends at q. Accordingly, the product
(x, q)(y, q') is defined when q' = x-1q, in which case (x, q)(y, x-1q) := (xy, q).
The inverse is (x, q)-I := (X-I, x-1q). Hence t(q) = (e, q).
When all relevant objects are manifolds and all operations are smooth, one
speaks of a Lie groupoid. Given a Lie groupoid G, one can tum C~(G) into
a convolution *-algebra that reflects the basic properties of the groupoid opera-
tions. For example, for a Lie group this reproduces the *-algebra we have already
encountered. For a pair groupoid one obtains

1* g(ql, q2) = fa dv(q) I(ql, q )g(q, q2)

and f*(ql, q2) = l(q2, ql). On an action groupoid one has

I*g(x,q)= fa dyl(xy,q)g(y-I,y-Ix-I q )
and f*(x, q) = I(x- I , x-1q).
One sees that in these two cases the involution is defined by
f*(y) = I(y-I);

this is, in fact, always true. One can put a norm on C~(G), and complete it so
as to obtain a C*-algebra C*(G). For the pair groupoid Q x Q one then finds
C*(Q x Q) = 'l30(L2(Q».
Given a principal H -bundle P over Q, one may form the "quotient" of the pair
groupoid P x P by H, obtaining the gauge groupoid P x H P of the bundle. This
is a groupoid with base Q; an arrow [x, y]H starts at r(y) and ends at rex) (where
r is the bundle projection on P). The C*-algebra C*(P XH P) of this groupoid
turns out to be isomorphic to 'l30(L2(Q» ® C*(H). For compact H this is nothing
but the C* -algebra 'l3 o(L 2(p»H we have already encountered, and C*(P x H P)
is in every respect the correct generalization of Bo(L 2(P»H to the case where H
is noncompact. In particular, given a representation Ux(H) one may construct an
induced representationrr x ofC*(P XH P), which is irreducible iff U x is. This leads
to a bijective correspondence
Ux(H) +------+ rrX(C*(P XH P»
between the representations of H and the representations of C*(P XH P).
 III. Groups, Bundles, and Groupoids 23

The C*-algebra of an action groupoid G x Q is usually written as C*(G, Q),

and is called an action C*-algebra. The C*-algebra C*(G, Q) has the remarkable
property that each of its representations corresponds to a system ofimprimitivity
(U, if), where U is a representation of G, and if is a representation of Ca( Q),
satisfying the covariance condition
U(x)if(j)U(x)* = if (axCi».
Here ax (i) : q 1-+ i(x- Jq). This condition is an integrated fonn of the "canonical"
commutation relation

tii [Qt!(X),
- -
Q/l(f)] =
-
Q,,(~x j),

for i E C~(Q) and X E g. Here Q/l(X') := iMU(X) and Q,,(j) := if(j); recall
the definition of the linear function X' E COO(g*, JR) and of the vector field h on
Q.
Turning to the top right comer in Figure I, we now describe the "infinitesimal"
object Q5 associated to a Lie groupoid G, generalizing the concept of a Lie algebra.
The Lie algebroid of a Lie groupoid with base Q is a vector bundle over Q,
which apart from the bundle projection 'f : Q5 ~ Q enjoys another linear map
'fa : Q5 ~ T Q, called the anchor. In addition, there is a Lie bracket [, ] on the
space of sections of Q5, which is related to the usual commutator on vector fields
on Q through the anchor. These objects are all constructed from G; the bundle Q5
itself is built from the geometry of the map t : Q ~ G (it is the nonnal bundle
of this inclusion), the Lie bracket is derived from the commutator of left-invariant
vector fields on G (much as in the case of a Lie algebra), and the anchor is the
derivative of the map from G to Q that assigns to an arrow its starting point.
For example, the Lie algebroid of the pair groupoid Q x Q is the tangent bundle
T Q with the obvious Lie bracket; the anchor is, of course, the identity map. The
Lie algebroid of the action groupoid G x Q is the action algebroid 9 x Q, regarded
as a trivial bundle over Q. Identifying sections of 9 x Q with g-valued functions
X (.) on Q, the Lie bracket on constant sections is simply the bracket [, ]g in g.
More generally, one has
[X, Y]gxQ(q) = [X(q), Y(q)]g + ~yX(q) - ~x Y(q).

The anchor comes out as 'fa (X, q) = -h(q). Finally, the Lie algebroid of the
gauge groupoid P XH Pis (TP)/ H as a vector bundle over Q, with commutator
inherited from the usual one on vector fields on P.
This brings us to the downward arrow on the right in Figure I, namely the
construction of a Poisson algebra from the Lie algebroid Q5. The Poisson mani-
fold in question is the dual bundle Q5*, the Poisson bracket on C OO (Q5*, JR) being
detennined by the special cases
{f, g}- = 0;

----
{S, fl- = -'fa 0 sf;
{sJ, S2}- = -[sJ, S2]18.
24 Introductory Overview

s
Here f and g are functions on the base Q, and is a linear function on I!)* defined
by a section s of I!) in the obvious way.
For I!) = T Q this is simply the usual Poisson structure on the cotangent bundle
T* Q; we see that this structure ultimately derives from the groupoid operations on
Q x Q. The relevant special cases of the bracket on the Poisson algebra Coo(g"'- x
Q, JR.) determined by the action groupoid G x Q are as follows. Firstly, for functions
f, g depending only on Q one has the obvious {f, g} _ = O. Secondly, on constant
sections (identified with linear functions on gO) one has {X, Y}- = -[X,Y].
Finally, the "mixed" bracket is {X, j} _ = h j. One sees from these Poisson
brackets that a representation of the Poisson algebra Coo (g* x Q, JR) on a symplectic
manifold is essentially a classical system of imprimitivity, being the classical
analogue of the system of imprimitivity determined by a representation of the
corresponding groupoid C*-algebra C*(G, Q).
Now to the top horizontal arrow. It turns out that the exponential map Exp :
9 --+ G on a Lie algebra can be generalized to a map Exp W : I!) -~ G from the Lie
algebroid I!) into a corresponding Lie groupoid G. This generalized exponential
map, however, depends on the choice of a connection (or covariant derivative) on
the vector bundle I!) over Q. Since for a Lie algebra the base space of this bundle
consists of only one point, there is no need for a connection in this case. In an
action Lie groupoid 9 x Q one does not need a connection either in order to define
Expw. In terms ofExp: 9 --+ G the map Expw : 9 x Q --+ G x Q is given by

ExpW (X, q) = (Exp(X), Exp(~X)q).

On a pair Lie algebroid T Q one does need a connection; this is, of course,
nothing but an affine connection. The latter leads to an exponential map exp in the
sense of affine geometry, in terms of which Exp W : T Q --+ Q x Q is

where T := TTQ-->Q' For example, the affine connection may be the Levi-Civita
connection provided by a Riemannian metric on Q.
At last, we are now in a position to define the generalized Weyl quantization
map QJ:' : COO (I!) *, JR) --+ C*(G)]R (restricted to suitable bounded functions); this
is the bottom line of Figure 1. In analogy with the prescription for groups, in a
rough sketch it is given by

where the integration is over the fiber of I!)* above T®-->Q(X),

Using the above formulae for Expw, one verifies that this prescription indeed
reduces to the Weyl quantization of Coo(T* Q, JR) explained in Chapter II, as well
as to the quantization of Coo(g"'-, JR) discussed above.
 IV. Reduction and Induction 25

IV. Reduction and Induction

In symplectic geometry one has the concept of symplectic reduction, whose aim is
the construction of new symplectic manifolds from old ones; it may be interpreted
as a tool in the representation theory of Poisson algebras. There exists an analogous
technique in the representation theory of C* -algebras, called induction. The final
chapter develops the analogy between reduction on the classical side and induction
on the quantum side. In physics these techniques playa central role in the classical
and quantum theory of constrained systems.

Reduction
The general concept of symplectic reduction is as follows. Let (S, w) be a symplec-
tic manifold, and let C be a submanifold of S. The restriction Wc of the symplectic
form w to C is closed, but not necessarily nondegenerate. Suppose Wc is degen-
erate. The tangent bundle TC to C then contains a subbundle Nc := TC n TC~,
where T C~ consists of all vectors in T C on which Wc identically vanishes. Under
favorable circumstances, the collection of all curves in C that are tangent to Nc
defines a foliation <l>c of C, whose quotient SC := C / <l>c is a manifold.
The essential point is now that the reduced space SC is equipped with a sym-
plectic form w C , whose pullback to C under the projection from C to SC is Wc.
This is possible because the "directions of degeneracy" Nc of Wc have disappeared
in the construction of the reduced space.
In physics the submanifold C C S is defined by constraints on the allowed initial
states of a given dynamical system; Gauss's law in electrodynamics is a typical
example. Flows along Nc are often generated by gauge transformations, which
do not modify the physical state of the system, and correspond to a redundancy in
the description of the system in terms of the degrees of freedom in S. The passage
from S to C then implements the constraints, whereas the subsequent step from
C to SC eliminates the gauge redundancy. In any case, one should firmly keep in
mind that symplectic reduction is generically a two-step procedure (except when
Wc is nondegenerate, so that C itself is symplectic, and SC = C).
Suppose that TC~ is contained in TC, in which case C is called coisotropic.
The collection of all smooth functions on S that are constant on the leaves of <l>c
is then a Poisson algebra, which in physics is the algebra of weak observables Ql~
ofthe system. Each f E Ql~ evidently "reduces" to a well-defined function JTc (f)
on the reduced space, and the map JT c is a representation of Ql~ in C"'\Sc , lR).
The following specialization of the above reduction scheme plays a central
role in this chapter. Suppose one has a pair of symplectic manifolds (S, ws) and
(Sp, wp), a Poisson manifold P, and a pair of Poisson morphisms 1 : S ~ P-
and lp : Sp ~ P (here P- is P with minus its Poisson bracket). One then takes
S = S x Sp, equipped with the symplectic form w := Ws + wp. We write <l>p for
the null foliation <l>c. The submanifold

C = S *p Sp := {(a, a) E S x SpIJ(a) = lp(a)}

26 Introductory Overview

is then coisotropic, and leads to a reduced space

sj := (S *p Sp)/4>p.

Most physically relevant examples of symplectic reduction are a special case of

this construction, which we call special symplectic reduction. From the point of
view of representation theory the main feature of the construction emerges when
one has a second Poisson manifold P2, and a Poisson map J2 : S -+ P2, such that
the pullback J;C oo (P2, JR) in Coo(S, JR) Poisson-commutes with J*Coo(P, JR).
The map JP : sj -+ P2 , given by
JP([a, a]<I>p) = h(a)

is then well-defined, and is a Poisson map. Pulling back, one obtains a represen-
tation (JP)* of the Poisson algebra C oo (P2, JR) on sj, which is said to be reduced
by the representation J; of Coo(P, JR).
Writing 1t for J etc., we denote this situation by
h h
P2 +-- S -+ PI.

Denote the set of all f E Coo(S, JR) for which {f, J;g} = ofor all g E C oo (P2, JR)
by J* Coo (P2, JR)'. The existence of the manifold S and the maps J 1, h implies that
PI and P2 stand in a certain relationship to each other, which is particularly close
if J;C oo (P2, JR)' = JtCoo(PI, JR) as well as JtCoo(P1, JR)' = J;C oo (P2 , JR), and
1t and h are surjective, with connected and simply connected level sets in S.
If, given PI and P2, one can find S, J I , and h such that these, and some
additional technical conditions are met, one says that PI and P2 are Morita
equivalent. The classical imprimitivity theorem then states that Coo(P I , JR)
and C oo (P2 , JR) have equivalent representation theories. Specifically, every repre-
sentation of C oo (P2, JR) is reduced from some representation of Coo(PI , JR), and
vice versa, and this bijection preserves irreducibility.
The idea of the proof of this theorem is as simple as it is elegant, and is most
easily formulated if we use Poisson maps J rather than representations n = J*
(one may always pass from one to the other). Given a Poisson map J p : Sp -+ Ph
one constructs the reduced space Sf' by special symplectic reduction. As explained
above, this leads to a Poisson map Ji : sf' -+ P2 • One now turns the diagram
P2 ~ J 1 PI around
J2 S -+ ,0btammg
" PI ~ h P2. App I'
J 1 S - -+ ymg specla. I sympI '
ectlc
reduction once again, this time from J = J i ' one obtains a reduced space S2 and
(J :

a Poisson map J2 : S2 -+ PI' Using all the assumptions involved in the Morita
equivalence of PI and P2, one then shows that S2 is symplectomorphic to Sp, such
that J 2 is equivalent to Jp- This works in the opposite direction as well.
Specializing special symplectic reduction, we now assume that P = 1)* (where
1)* is the dual of the Lie algebra I) of a connected Lie group H), and J : S -+
I):' is an equivariant momentum map coming from a strongly Hamiltonian H-
action on S. Moreover, we take Sp to be a coadjoint orbit 0 in 1)* (equipped
with the Lie symplectic structure), so that Jp is simply the inclusion map. The
ensuing doubly specialized reduction procedure is called Marsden-Weinstein
 IV. Reduction and Induction 27

reduction. The reduced space Sf obtained by special symplectic reduction from

these data is easily seen to be diffeomorphic to 1-'(0)/ H, which thereby acquires
a symplectic structure. The symplectic space 1-'(0)/ H is called a Marsden-
Weinstein quotient.
Since H acts on S by Poisson maps, the Poisson bracket on S descends to
a Poisson bracket on S / H. However, the latter is not symplectic, unless H is
discrete. As a Poisson manifold, S / H is foliated by its symplectic leaves. It turns
out that these leaves are precisely the Marsden-Weinstein quotients 1-'(0)/ H.
This allows us to see the phase spaces (T*P)o of Chapter III in a new light. In
particular, when P and H are connected and simply connected, the correspondence
no ~ nO between the irreducible representations of COO(~~, R) and those of
COO«T*P)/ H, R) found in Chapter III comes out as a consequence of the classical
imprimitivity theorem.
In a generalization of this construction, which we call Kazhdan-Kostant-
Sternberg reduction, the inclusion of 0 into ~* is replaced by a general Poi~son
map 1p : Sp ~ ~~, where Sp is symplectic. We assume that 1p is minus the
momentum map of a strongly Hamiltonian H -action on Sp. Special symplectic
reduction then leads to to a reduced space (T* P)P.
In the special case that P = G is a Lie group, seen as a principal bundle over
Q = G / H, we thus obtain a reduced space (T*G)P for each strongly Hamiltonian
H -space Sp' This reduced space carries a classical system of imprimitivity. Firstly,
the left G-action on T*G (pulled back from the left action on G) reduces to a G-
action on (T*G)p. This yields a Poisson map 10) : (T*G)P ~ g~. Secondly, since
(T*G)P is a bundle over G/H, one has a map 1&) : (T*G)P ~ G/H; this is a
Poisson map with respect to the zero Poisson structure on G / H. These combine
to form a representation of the Poisson algebra Coo (g~ x G / H, R) of the action
algebroid defined by the canonical G-action on G / H.
Without any connectedness assumptions, the classical transitive imprimitivity
theorem now states that any classical system of imprimitivity for G and G / H ,
in other words, any representation of the Poisson algebra COO(g~ x G / H, R), is
equivalent to one of the above form.
So far, we have (tacitly) assumed that S/ H and each 1- 1(0)/ H are manifolds.
When H is compact this is the case when the H -action on S is free. A fascinating
situation arises when one drops this assumption. Without loss of generality, we
may restrict ourselves to the case 0 = {OJ. It turns out that the reduced space
S~ is the (disjoint) union of certain symplectic manifolds S~Kl' each of which
corresponds to the conjugacy class [K] of the stabilizer K C H of some point
in S. The reduced space has a Poisson structure, which restricted to each S~Kl is
equivalent to the symplectic structure of that subspace. Any Hamiltonian flow in
1- 1(0)/ H necessarily stays inside a given subspace S~Kl' In view of the last point,
the decomposition of S~ is somewhat reminiscent of the foliation of a Poisson
manifold by its symplectic leaves.
28 Introductory Overview

Induction
Almost every aspect of special symplectic reduction has a counterpart in the context
of Hilbert spaces and C' -algebras, albeit with subtle changes.
In special symplectic reduction one starts from a Poisson map J : S -+ p-.
Equivalently, one has a representation j* : COO(P-) -+ COO(S, JR), which may
alternatively be regarded as an antirepresentation of COO(P, JR). The quantum ana-
logue of the (complexified) Poisson algebra COO(P) is taken to be a C*-algebra
1J3; the antirepresentation J* should then correspond to a right action JT R (IJ3) on a
linear space of some sort. These are easy analogies. In the absence of an underlying
space P for 1J3, it is clear that the equivalent classical objects J and j* should be
disentangled in quantum theory.
The quantum counterpart of J in operator theory is a Hilbert C* -module over
the C* -algebra 1J3. This consists of a complex linear space £, a linear right action
JT R oPE on £, and a "1J3-valued inner product" (, h : £ x £ -+ 1J3.
The sesquilinear form (, }'B must firstly satisfy (\II, <I>); = (<I>, \II}'B, generaliz-
ing the behavior of an ordinary (>valued inner product under complex conjugation.
Secondly, the lJ3-valued inner product should intertwine JT R with the canonical right
action of IJ3 on itself (given by multiplication on the right); in other words, one
requires that (\II, JTR(B)<I»'B = (\II, <1»'13 B. Furthermore, one imposes positive def-
initeness, in that (\II, \11)'13 ::: 0, with equality iff \II = O. It is finally required that
£ be complete in the norm I \II II := I (\II, \11)'13 111/2.
For example, IJ3 is a Hilbert C* -module over itself, with JTR(B)A := A Band
(A, B)'B := A*B. Also, a Hilbert space 1t is a Hilbert C*-module over C in its
inner product.
So far, we have stated the first half of the input for "quantum induction". In
special symplectic reduction one furthermore has a second Poisson map J p :
Sp -+ P, where Sp is a symplectic manifold. In quantum theory Sp is replaced
by a Hilbert space 'Hx. There is no quantum counterpart of J p , but the associated
representation J; : COO(P, JR) -+ COO(Sp, JR) corresponds to a representation JT x
of IJ3 on 'H x .
The construction of the classical reduced space sj is replaced by a procedure
called Rieffel induction. Table 2 presents a summary of the analogy between
special symplectic reduction and Rieffel induction, which proceeds as follows. One
first equips £ ® 1t x with a sesquilinear form (, )~, defined by linear extension of

(\II ® v, <I> ® w)~ := (v, JTx«(\II, <I»'B)wh,

where \II, <I> E £ and v, w E 1t x' This form is positive semidefinite, because (, h
and (, )'13 are. Then form the quotient of £ ® 'Hx by the null space N x of (, )~;
this is evidently a pre-Hilbert space. The induced space

1t x := (£ ® 'Hx/Nx)-
is the completion of £ ® 1t x / N x in the inner product inherited from (, )~.
For the quantum counterpart of the reduced representation (J P)*(C OO ( P2 » on sj
in special symplectic reduction, we define the notion of an adjointable operator
 IV. Reduction and Induction 29

Special symplectic reduction Rieffel induction

Poisson algebra COO(P) C* -algebra IJ3

symplectic manifold S linear space £
J* : CCC(P) --+ COO(S) JfR (IJ3) on £
Poisson map J : S --+ P- lJ3-valued inner product ( , )Ql
symplectic manifold Sp Hilbert space 'H. x
representation J; : COO(P) --+ COO(Sp) representation Jf x : IJ3 --+ 1J3('H.x )
Cartesian product S x Sp tensor product £ 18> 'H. x
constraint manifolds S *p Sp £ 18> 'H. x
null foliation <t> p null space N x
S'j = (S *p Sp)/<t>p 'H.x = (£ 18> 'H.x/Nx )-
J*COO(p)' c COO(S) C*(£, 1J3)
Poisson algebra C OO (P2 ) CO-algebra 2l
reduced representation (JP)*(C OO (P2» induced representation Jf x (21)

TABLE 2. Special symplectic reduction and Rieffel induction

on E. This is an operator A that has an adjoint with respect to the ~-valued inner
product; in other words, one has

(\II, A<I»'B = (A*\II, <l>h

for some operator A* on E. The space of all adjointable operators on E is a C*-
algebra, denoted by C*(E, 23). An adjointable operator A has the property that
A ® llx maps the null space N x into itself (here llx is the unit operator on Ji x )'
Hence A ® lIx induces an operator on E ® Ji x /Nx in a natural way; under suitable
boundedness assumptions the latter operator extends to an operator nX (A) on Jix.
To complete the picture, suppose one has a morphism of a C* -algebra Ql into
C*(E, 23). Composing with this morphism, one may (with slight abuse of notation)
look at n x as a representation of Ql on the induced space Jix. This representation
is said to be induced (in the sense of Rieffel) by the representation n x(23) with
respect to the Hilbert C*-module E over~.
With regard to the analogies listed in Table 2, it is remarkable that the constraint
manifold of classical mechanics has no quantum counterpart. In other words, in
quantum mechanics it is not necessary to impose the constraints (at least in the
case that all constraints are first class in the sense of Dirac, which is the case in
special symplectic reduction). As opposed to classical reduction, which is a two-
step procedure, the construction of the induced space Jix in Rieffel induction has
only one step, corresponding to the second step of symplectic reduction.
The physical interpretation of Jix is that it is the physical state space of the
system, in which all gauge (and perhaps other unphysical) degrees of freedom
have been removed. (Traditional approaches to constrained quantization instead
try to mimic the first step of symplectic reduction, imposing the constraints on
the Hilbert space of states of the unconstrained system. This has turned out not to
work, except in the very simplest examples.)
30 Introductory Overview

We now explain the quantum analogue of Morita equivalence of Poisson al-

gebras. In preparation, we need a refinement of the C*-algebra C*(£, IB), which
is analogous to the restriction of the C*-algebra IB(H) of all bounded operators
on a Hilbert space H to the C*-algebra IBa(H) of all compact operators on H.
Namely, we define CO(£, IB) as the C*-subalgebra of C*(£, IB) that is generated
by all operators of the type T!f <4" where 1lJ, <l> E £, and

T!f,¢>Z := 1lJ(<l>, Z)'B.

Two C* -algebras Q( and IB are now said to be Morita equivalent when there exists
o
a full Hilbert C*-module £ over IB under which Q( :::::: C (£' IB). We write Q( ~
£ ~ lB. Here a Hilbert C* -module is called full when the collection {(IlJ, <I»'B}.
where 1lJ, <l> run over £, is dense in lB.
For example, IBa(H) is Morita equivalent to C, with £ = H.
As in the classical case, Morita equivalence implies that Q( and IB have equivalent
representation theories, the bijection preserving irreducibility. More precisely, the
quantum imprimitivity theorem states that every representation of Q( is equiva-
lent to one that is Rieffel-induced from some representation Ji x (IB), and vice versa.
The proof uses exactly the same idea as its classical counterpart. The crucial step
of turning P2 13- s ~ PI around to PI ~ s- ~ P2 in the classical proof now
works as follows. The conjugate space E is equal to £ as a real vector space, but has
the conjugate action of complex scalars. The replacement of S by S- corresponds
to the replacement of £ by E. Moreover, the expression
(1lJ, <l>)c~(£.'B) := T!f,¢>,
o
in combination with the right action JiR(A)1lJ := A *1lJ, where A E C (£' IB),
defines E as a full Hilbert C* -module over CO(£, IB). Similarly, the right action
of IB on £ is turned into a left action on E by acting with the adjoint. Hence
Q( ~ £ ~ IB turns around to IB ~ E ~ Q(.
The theorem is then proved by starting with a representation Ji x (lB), Rieffel-
inducing with respect to Q( ~ £ ~ IB to obtain a representation Ji x (Q(), using the
latter to construct an induced representation of IB with respect to IB ~ E ~ Q(, and
finally showing that this representation of IB is equivalent to Ji x. This procedure
works in both directions.
In view of the Morita equivalence IBa(H) ~ H ~ C, an immediate corollary
of the quantum imprimitivity theorem is that the C* -algebra of compact operators
has only one irreducible representation.
There is a quantum analogue of Marsden-Weinstein reduction. Instead of a
strongly Hamiltonian H -action on a symplectic manifold S, we start from a repre-
sentation U (H) on a Hilbert space H, and the role of the Poisson algebra COO(f):')
is now played by the group C* -algebra IB = C* (H). Suppose, for simplicity, that
H is compact. We then take £ = H (actually, £ is a certain completion of H, but
we will not bother with this detail), on which C*(H) acts from the right by

JiRU) = l dh f(h)U(h)-I.
 IV. Reduction and Induction 31

The C*(H)-valued inner product on 'H is defined by letting ('It, <1>}c*(H) be the
function h ~ ('It, U (h)<1».
We may now proceed with Rieffel induction from some representation rr x of
C*(H); as we have seen in Chapter III, we may equivalently assume that we have
a representation U x of H. The form (, )~ on Ji ® 1t x reads

Although for compact H this integral may be explicitly computed, we leave the
expression as it stands, and remark that it is valid for noncompact groups as well.
The only difference with the compact case is that in general £ is a suitably chosen
dense subspace of Ji (for in the noncompact case the convergence of the H-
integration needs attention).
A most interesting instance of "quantum Marsden-Weinstein reduction" arises
in the context of a principal H-bundle P. We take Ji = L 2 (P) (defined with
respect to some H -invariant measure), which carries the unitary representation
U(H) := U R(H) naturally constructed from the given right action of H on P.
Hence we obtain a right action of C*(H) on L 2 (P), eventually leading to a Hilbert
C* -module £ over C*(H). The Hilbert space Jix constructed by Rieffel induction
from Ux(H) is then naturally isomorphic to the space Jix defined earlier in the
context of Mackey induction.
We may compute the C*-algebra q(£, C*(H». Remarkbly, this turns out to
be the C*-algebra C*(P XH P) of the gauge groupoid of the bundle. It follows
that C*(P XH P) and C*(H) are Morita equivalent. The bijective correspondence
Ux(H) ++ rrX(C*(P XH P» found in Chapter III then follows from the quantum
imprimitivity theorem. Let UL be the canonical representation of Aut(P) on L 2 (P).
For each cp E Aut(P) the operator U(cp) commutes with rrR(C*(H», which implies
that it is adjointable. The Rieffel-induced representative rrX(U(cp» coincides with
the induced representative U x (cp) defined in Chapter III.
Specializing to the case where P = G is a Lie group, and realizing that the
action C* -algebra C* (G, G / H) is isomorphic to the gauge groupoid C* -algebra
C*(G XH G), we conclude that C*(G, G/ H) and C*(H) are Morita equivalent.
Applied to this situation, the general quantum imprimitivity theorem then implies
the quantum transitive imprimitivity theorem. To explain what this theorem
means, first observe that Jix carries a transitive system of imprimitivity, in which
U(G) = UX (G), and ff(Co(G / H» is defined by

The theorem now states that for any system of imprimitivity for G with Q = G / H
there exists a representation Ux(H) such that the system is equivalent to the one on
Jix just defined. This is the exact quantum counterpart of the classical transitive
imprimitivity theorem discussed earlier.
32 Introductory Overview

Applications in Relativistic Quantum Theory

The most interesting applications of Rieffel's and other induction techniques in
physics appear to be to the mathematically rigorous study of gauge field theories.
We wish to clarify three aspects of such theories.
Firstly, there is a close (but not universal) relationship between gauge invariance
and the masslessness of the field quanta of a quantum gauge theory. A shadow of
this relationship may already be seen in the classical theory of massless relativistic
particles and fields; infinite-dimensional Marsden-Weinstein reduction will be
seen to playa central role.
Secondly, the passage from an unconstrained Yang-Mills theory to its physical
sector involves a tremendous reduction in degrees of freedom; when the underlying
space is a circle this reduction even leads to a finite-dimensional theory. This will
be proved with induction techniques.
Thirdly, it was discovered in the sixties and seventies that the quantization of
certain physical systems, notably gauge theories, may involve parameters (be-
yond n) without a classical analogue. These so-called vacuum angles or O-angles
emerge in a transparent way when one quantizes constrained systems using induced
representations.
We start with a description of the coadjoint orbits of the Poincare group.
This group is the semidirect product P = L ~ p M of the Lorentz group
L = SO(3, 1) and the additive group M = ]R4 (equipped with the Minkowski
metric diag(l, -1, -1, -1), of which L is the connected isometry group). The
action p of L on M with respect to which the semidirect product is formed is
simply the defining action of SO(3, 1) on ]R4. A central role in the description of
the coadjoint orbits of such semidirect products is played by the dual action p*(L)
on M* c:::: ]R4.
A nontrivial analysis shows that each coadjoint orbit oP in p* is isomorphic
(as a symplectic manifold) to a Marsden-Weinstein quotient of the type (T* p)'?,
which we have encountered before in a different context. Here we have to take
P = L, whereas 0 is a coadjoint orbit of the stabilizer L p of some point p in M*
under the action p*(L). Hence the orbits are fibered over T*(L / L p), with typical
fiber O.
The phase space of a massless relativistic particle with positive energy is ob-
tained by choosing p = (1, 0, 0, -1). Its stabilizer is isomorphic to the Euclidean
group E(2) := SO(2) ~p ]R2 in dimension 2. Hence classical massless particles
are further classified by the coadjoint orbits 0 of E(2). The dual of the Lie algebra
of E(2) is]R3, whose coadjoint orbits are either cylinders C, = S; x ]R (where the
circleS; of radius r > 0 lies in the (x, y)-plane, and ]R is the z-axis), or points
(0,0, h).
Only the latter are believed to be of physical relevance; the parameter h is called
the helicity of the particle. For example, a classical photon has helicity 1 or -1,
and a classical graviton has helicity ±2. The phase space 0b.+.h of such particles
is diffeomorphic to T*(L/ E(2», but the Poisson bracket contains an additional
term (beyond the canonical cotangent bundle bracket) proportional to h.
 IV. Reduction and Induction 33

There is an almost complete parallel between the coadjoint orbits and the ir-
reducible representations of P. The only difference lies in the fact that the latter
are classified by the irreducible representations of the stabilizers L jj, rather than
by their coadjoint orbits. Relativistic massless quantum particles are therefore
classified by the irreducible representations of £(2). The physically relevant rep-
resentations UO,+,h are again labeled by the helicity h, which in quantum theory
assumes only (half-) integral values. The Hilbert space rt°,+,h carrying UO,+.h is
L2(L/ £(2».
In a remarkable twist of nature and mathematics, the physically relevant irre-
ducible representations of P describe both quantized particles and classical fields.
However, the massless relativistic fields occurring in the Lagrangians and Hamilto-
nians of classical field theory do not transform under UO,+.h , but under a so-called
covariant representation of P. This is a (generally nonunitary) representation
n A that is (Mackey) induced from a (nonunitary) representation of L. The lack of
unitarity does not matter for classical physics, since the "covariant" action of P
on the space of fields should be seen in a symplectic context; it is, indeed, strongly
Hamiltonian.
Gauge fields A transform under the covariant vector representation v (P), n
defined by

In order to reach UO.+,±I, as a first step one imposes the infinite number of con-
straints DAIl = 0 on the space SV of all gauge fields, and performs symplectic
reduction. This leads to a symplectic space So,+, v, whose configuration space
part consists of all solutions of the above wave equation whose Cauchy data are
square-integrable in a suitable sense.
The second step of the passage from SV to rt°,+,±I, then, involves the gauge
group 9. This is the real Hilbert space of real solutions). of the wave equation
0). = 0 on M whose (weak) derivative a). (seen as a four-vector with components
all).) lies in SO,JR, v, The connection between gauge invariance and masslessness in
classical free field theories is now as follows. The gauge group acts on SO,JR, v by

this action is strongly Hamiltonian, with momentum map J, and the Marsden-
Weinstein quotient J- ' (0)/9 is rt°,+,1 EB rt°,+,-I. Moreover, the reduction of the
covariant action nV (P) on SV to SO,JR, v further reduces to an action on J- ' (0)/9,
which coincides with the representation UO,+,I EB UO,+,-'.
We now quantize this reduction procedure with the aid of a generalization of the
quantum Marsden-Weinstein induction technique, which is suitable for dealing
with infinite-dimensional groups. We start as if the gauge group were locally com-
pact, and consider a Hilbert space rt carrying a representation U (9). To construct
rt we exploit the fact that SV, previously looked upon as a symplectic space, may
34 Introductory Overview

be turned into a Hilbert space 'Hv. For 'H we then take the bosonic Fock space
00

exp('H v ):= EB®~1iv,

'=0

where ®~ 'Hv is the symmetrized tensor product of I copies of 1iv. We define a

map JEXp : 1i v -+ exp(1iv) by

JExp(A):= L-;
®'A
00

1=0 --/IT
it follows that the inner product of two exponential vectors is

(JEXP(A), JEXP(B») = e(A,B)1t v .

Since the exponential vectors are dense in 1i, the natural representation U(Q)
we use is characterized by its matrix elements

( JExp(A), U F(A)JExP(B») = e(A,B)1t v e -~ IIAII~v e(A,JA)1t V -(aA,B)1t v .

Mimicking the finite-dimensional situation, we would like to integrate the above

function of A over Q with respect to a Q-invariant "Lebesgue" measure. This is
impossible, but fortunately one may combine the nonexistent Lebesgue measure
on Q with the factor exp( - ~ I AII~v). This combination yields a Gaussian measure
f-1, on a certain completion Qc of Q. We then put

( JExp(A), JEXP(B»): := e(A,B)1t v £c df-1,(A)e(A,JA)1t V -(oA,B)1t v .

One may proceed with the construction of the induced space 'HId as usual, obtaining
the correct quantum field theory of photons. In particular, the gauge group Q is
trivially represented in 1i id , and Gauss's law is satisfied.
Following this treatment of the connection between masslessness and gauge
invariance in classical and quantum electromagnetism, we tum to a different class
of models for a discussion of the remaining two points of interest in gauge theories.
Classical Yang-Mills theory on a circle with structure group H is defined by
the configuration space AIR = L2(SI, ~), with phase space
S = T*AIR:::: A = L2(S', ~C),
Here the inner product in L2 is defined with respect to an Ad(H)-invariant
inner product on ~. The gauge group Q of the model is the Sobolev loop
group 'H, (SI, H), consisting of those g E C(S', H) whose (weak) derivative
g := g-ldg/da lies in AIR. These definitions guarantee that the action

g :A 1-+ A8 := Ad(g)A + gdg-' = g(A _ g)g-'

of Q on AIR is smooth. This action lifts to a strongly Hamiltonian action on the phase
space A, given by the same formula (with A replaced by a complex connection
Z).
 IV. Reduction and Induction 35

The point is now that the Marsden-Weinstein quotient SO = J-1(0)IQ (where

J the momentum map of the above Q-action) is symplectomorphic to the finite-
dimensional cotangent bundle T*(H 1Ad(H». The identification of this reduced
space with the physical phase space rests on the fact that J-1(0) is the subspace
of S on which Gauss's law holds. The key element of the proof of this symplecto-
morphism is the construction of the Wilson loop W : AIR -+ H. We first define
W: L2(SI, ~) -+ C([O, 1], H) as the solution of the differential equation
(a~ +A)WA(a)=o,
with initial condition WA(O) = e (here WA := W(A». With W(A) := WA(l),
one shows that W(Ag) = W(A) for all based gauge transfonnations g E Qe (i.e.,
g(O) = e). Hence W quotients to a map from AIRIQe to H; the peculiar feature
of the model is that this map is a diffeomorphism. Moreover, W complexifies to a
map We from the phase space A to the complexification He of H. Since He is dif-
feomorphic to T* H, the map We restricts and quotients to a symplectomorphism
between Sa and T* H.
The quantization of this reduction procedure follows the lines of our earlier
treatment of photons. The unconstrained phase space S is quantized as the bosonic
Fock space H := exp(A), on which the gauge group is represented by

UF(g)JExp(Z):= e-~lIgIl2+(g,Z\/Exp(Zg).
Hence the matrix element of U F (g) between two exponential vectors again contains
a Gaussian factor exp( - ~ IIi 11 2 ), which we wish to combine with the nonexistent
Haar measure on Q. This leads to a version of the well-known Wiener measure
fL w, conditioned to the space of continuous loops on H. We may therefore put

( JExp(W), JEXP(Z»): := iH dfL W (g) e(W,Z8)+(g,Z).

For technical reasons the integral is over LH = C(SI, H) rather than over the
gauge group, which is a sup-dense subspace of LH that happens to have fL W_
measure zero.
The induced space H'd defined by induction with respect to the above fonn is
naturally isomorphic to the subspace of L 2(H) that is invariant under the represen-
tation defined by the adjoint action. In fact, replacing LH in the above integral by
the space LHe of based loops, the induced space H: is L2(H) itself. A function
f E COO(H) defines W f E COO(A) by Wf(A) := f(W(A». The quantization of
the observable Wf on L2(H) then comes out to be the multiplication operator f.
When f is a class function, this operator has a well-defined restriction to H;d.
The identification ofH: with L 2(H) makes essential use of the Hall coherent
states q/h in L 2(H); this is a recently discovered family of coherent states that is
labeled by the points z in He. The complexified Wilson loop We : A -+ He of
the classical theory has a quantum counterpart, which (up to normalization) maps
-W(Z)
Y"EXP(Z) to \11 1/ 2 •
36 Introductory Overview

We finally tum to vacuum angles. In view of the equalities

noW) := 9/90 = no(LH) = nl(H),

where 9 0 is the identity component of 9, the gauge group is disconnected whenever

the first homotopy group of the structure group H is nontrivial. For example, in the
abelian case H = U (l) one has no W) = n I (U (l» = Z. This motivates us to make
some general comments on the quantization of Marsden-Weinstein quotients by
a disconnected group 9.
The space SO = J-'(0)/9 may be constructed in two steps: one firstly forms
J- I (0)/9°, which is a symplectic space. Secondly, one quotients the latter by the
discrete group no(Q), again obtaining a symplectic space, which is isomorphic to
So. We isolate the second step. Although on the classical side noW) possesses
only the trivial coadjoint orbit {O}, on the quantum side it will have nontrivial
irreducible representations, which have no classical counterpart. A vacuum angle
is an element of the unitary dual ;;;(g); for Yang-Mills theory on a circle this is
the same as ;;(ii). For H = U (1) one therefore finds 9 = z = u (1), explaining
the alternative name 8-angles.
Pick a () E ;;;(g), with corresponding representation Ue(noW». Instead of
forming the physical state space by induction from the trivial representation of
the gauge group, as we have done so far, we have the freedom of inducing from
the representation UeW), derived from Ue via the canonical projection from 9 to
9/9°. The quantum observables of the gauge-invariant system, such as the physical
Hamiltonian, then explicitly depend on 8, since these operators are constructed by
an induction procedure that depends on 8. Hence one obtains a different physical
theory for each 8 E ~. In other words, the gauge-invariant theory admits
inequivalent quantizations, classified by 8.
The 8-dependence may be shown quite explicitly in the U(l) gauge theory on
a circle. As we have seen, the reduced classical theory of this model describes
a particle moving on U(l); the corresponding 8-dependent quantum theory turns
out to be a description of the Aharonov-Bohm effect.
CHAPTER

Observables and Pure States

1 The Structure of Algebras of Observables

1.1 Jordan-Lie Algebras and C*-Algebras
In this section we specify the key algebraic and functional-analytic structures
relevant to classical and quantum mechanics. Our main aim is to look at a C*-
algebra from the point of view of its self-adjoint part. From this perspective the
relationship between the respective algebraic structures of classical and quantum
mechanics is particularly transparent.
Recall that an algebra is a vector space with a (not necessarily commutative or
associative) bilinear and distributive mUltiplication o. We write A2 := A 0 A.
Definition 1.1.1. A (real) Jordan algebra is a (real) algebra where
A 0 B = BoA; (1.1)
A 0 (B 0 A2) = (A 0 B) 0 A2. (1.2)
The simplest motivation for (1.2) is that it is automatically satisfied when the
Jordan product 0 comes from an associative product via A 0 B = 4(AB + BA).
However, not all Jordan algebras arise in this way.
Theories of dynamical significance have a second algebraic operation.
Definition 1.1.2. A Jordan-Lie algebra is a real vector space ~lR equipped with
two bilinear maps 0 and {, } (referred to as the Jordan product and the Poisson
bracket, respectively), such that the following conditions are satisfied. Firstly, one
has
A 0 B = Bo A;
{A, B} = -{B, A} (1.3)
38 I. Observables and Pure States

for all A, B E !2ljR. Secondly,for each A, the map B ~ {A, B} is a derivation of

(!2ljR, 0) as well as of(!2ljR, {, D. This means that the Leibniz rule

fA, B 0 C} = fA, B} 0 C + B 0 fA, C} (1.4)

as well as the Jacobi identity

{A, {B, Cn = {{A, B}, C} + {B, {A, Cn (1.5)

must hold for all A, B, C E !2ljR. Finally,for all A, B, C E !2ljR and some 1t2 E R
one requires the associator identity
(A 0 B) 0 C - A 0 (B 0 C) = ~lt2{{A, C}, B}. (1.6)

A Jordan-Lie algebra in which 0 is associative is called a Poisson algebra.

It follows from these axioms that (!2lIR, 0) is a real Jordan algebra, whereas
(!2lIR, (,D is a real Lie algebra. In connection with Jordan-Lie algebras, the
terminology (non) associative always refers to the Jordan product.
The following definitions are recorded for later use.
Definition 1.1.3. A Jordan morphism between Jordan-Lie algebras !2l1R and
$IR is a linear map P : !2l1R ~ $IR satisfying P(A 0 B) = P(A) 0 P(B) for all
A, B E !2l1R. Similarly, a Poisson morphism between such algebras is a linear map
satisfying P({A, B}) = (P(A), P(B)}. A map between Jordan-Lie algebras that
is simultaneously a Jordan morphism and a Poisson morphism is called simply a
morphism. An invertible (Jordan, Poisson) morphism a : !2l1R ~ !2ljR is called a
(Jordan, Poisson) automorphism, and an invertible (Jordan, Poisson) morphism
a: !2l1R ~ $IR is a (Jordan, Poisson) isomorphism.
We now equip the algebras introduced above with a norm.
Definition 1.1.4. A J B-algebra is simultaneously a real Jordan algebra and a
Banach space in which for all A, B E !2l1R one has

IIA 0 BII ~ IIAIIIIBII; (1.7)

IIAII2 ~ IIA2 + B 2 11. (1.8)

The motivation for the axioms of a J B -algebra will emerge in due course.
Putting A = B in (1.7) and B = 0 in (1.8), one sees that given (1.7), axiom (1.8)
is equivalent to the pair
IIA211 = IIAII2; (1.9)
IIA211 ~ IIA2 + B 2 11. (1.10)

Definition 1.1.5. A J LB-algebra is a J B-algebra!2l1R equipped with a Poisson

bracket that makes it a Jordan-Lie algebra for some 1t2 ~ O.
A J LB-algebra with It = 0 may alternatively be regarded as a Poisson alge-
bra with zero Poisson structure. A Poisson algebra with nonzero Poisson bracket
cannot, in general, be normed in such a way that the bracket is defined on the
 1 The Structure of Algebras of Observables 39

norm-completion of the algebra. For this reason Poisson algebras are usually not
studied in the setting of Banach spaces.
A J L B -algebra 21JR turns out to be the real part of a complex associative algebra
21 of a much-studied type.
An involution on a complex algebra is a real-linear map A ~ A* such that for
all A, BE 21 and J.. E C one has

A** = A; (1.11)
(AB)* = B* A*; (1.12)
(J..A)* = IA*. (1.13)

A *-algebra is a complex associative algebra with an involution.

Definition 1.1.6. A C*-a1gebra is a complex Banach space 21 that is at the same

time a *-algebra, such that for all A, B E 21 one has

IIABII :s IIAII IIBII; (1.14)

IIA* All = IIAII2. (1.15)

Combining the two axioms for a C* -algebra leads to II A II :s II A* II; replacing A

by A* and using (1.11) yields

IIA*II = IIAII· (1.16)

It can actually be shown that (1.15) implies (1.14), but this highly nontrivial fact
distracts from the guiding idea that a C* -algebra is a specialization of a Banach
algebra. This is a complex Banach space and an associative algebra, in which all
A, B satisfy (1.14). This property guarantees that left and right multiplication are
bounded, hence continuous; in fact, multiplication is ajointly continuous operation.
For example, the space I.B(B) of all linear maps on a Banach space B is a Banach
algebra under the norm

IIAII := sup{IIAWIlI WEB, IIWII = I}. (1.17)

The C* -axioms are motivated by the following example. Consistent with the
above terminology, a *-a1gebra of bounded operators on some Hilbert space 1{
is a collection of bounded operators on 1{ that is closed under addition, scalar
multiplication, operator multiplication, and taking adjoints. Thus the role of the
involution is played by the adjoint. Recall the definition of the norm of a bounded
operator:

IIAII2 := sup{(AW, AW)I \II E 1{, (W, \II) = I}. (1.18)

Since in a Hilbert space the norm is defined by 11\11112 = (\II, W), eq. (1.18) is
evidently a special case of(1.17). Hence IIA \1111 :s IIAIIII\IIII, which implies (1.14).
Moreover, we estimate
40 I. Observables and Pure States

so that IIAII2 ::::: IIA* All, which is ::::: IIA*IIIIAIl by (1.14). This leads to (1.16) by
the argument preceding that equation; the ensuing inequality II A *A II ::::: II A 112 then
implies (1.15).
Definition 1.1.7. A morphism between C*-algebras~, SB is a linear map rp :
~ --+SB such that
rp(AB) = rp(A)rp(B); (1.19)
cp(A *) = cp(A)* (1.20)
for all A, B E ~. An isomorphism is a bijective morphism. 1Wo C* -algebras are
isomorphic when there exists an isomorphism between them.
It is clear that a C* -algebra morphism between ~ and SB restricts to a morphism
(in the sense of 1.1.3) between the associated J LB-algebras ~IR and SBIR (cf.
1.1.9), and vice versa. Morphisms between C* -algebras have excellent properties;
see 1.3.10. For example, an isomorphism is automatically isometric.
Theorem 1.1.S. A norm-closed *-algebra ~ in SB(?t) is a C* -algebra (with oper-
ator multiplication as the product, etc.). Conversely, any C* -algebra is isomorphic
to a norm-closed *-algebra in SB(?t), for some Hilbert space ?t.
The computation following (1.18) establishes the first half. The proof of the
converse will be given at the end of 1.5. D
An element A of a *-algebra ~ for which A * = A is called self-adjoint. The
self-adjoint part ~lR is the collection of all self-adjoint elements in ~, seen as a
real vector space. Since one may write
A = A' + jA":= ~(A + A*) + i-=?(A - A*), (1.21)
every element of ~ is a linear combination of two self-adjoint elements.
A commutative C* -algebra is a C* -algebra in which the associative multiplica-
tion is commutative. The connection between J B-algebras, Jordan-Lie algebras,
and C* -algebras is as follows.
Theorem 1.1.9. Let ~ be a C*-algebra, and choose hE 1R\{0}. Equipped with
the norm inherited from ~, and the operations
A 0B:= ~(AB + BA);
i
{A, B}1l := /irA, B], (1.22)

the self-adjoint part ~lR of~ is a J LB-algebra. When ~ is noncommutative, the

parameter h in (1.6) equals Ii in (1.22); in particular, one has h 2 > O. When ~ is
commutative, ~IR is a Poisson algebra with zero Poisson bracket.
Conversely, given a J LB-algebra ~IRfor which h2 ::: 0, its complexification ~
is a C* -algebra under the operations
AB := A 0 B - ~ih{A, B}; (1.23)
(A + i B)* := A - i B; (1.24)
 I The Structure of Algebras of Observables 41

1
IIAII:= IIA*AII2. (1.25)
In (1.24) we assume that A, B E QtjR, and concerning (1.25) we remark that
A* A E QtjR for any A E Qt.
To prove the first half of the theorem, first note that by (1.16) the involution in Qt
is continuous, so that QtjR is a closed subspace of Qt. The axioms for a J LB-algebra
are trivially verified, except (1.10). We defer the proof of this property to the end
of 1.4.
In the opposite direction, it is trivially verified that the product (1.23) is asso-
ciative as a consequence of the properties of a Jordan-Lie algebra. When (1.25)
defines a norm, the property (1.15) holds by construction.
When QtjR is associative, so that Qt is commutative, the norm (1.25) on Qt simply
1
becomes IIA + iBIl = IIA2 + B2112, where A, B E QtjR. All axioms for a norm are
then easily derived from (1.9) and (LlO).
The proof that (1.25) is a norm also in the noncommutative case, as well as the
proof of (1.14), will be given at the end of 1.4, too. 0
One could replace the minus sign on the right-hand side of (1.23) by a plus sign;
that choice leads to a C* -algebra as well, which is anti-isomorphic to the one based
on the minus sign.

1.2 Spectrum and Commutative C* -Algebras

We are going to examine to what extent the closely related notions of spectrum
and functional calculus of a (bounded) self-adjoint operator A on a Hilbert space
1{ generalize to the context of C* -algebras and J L B -algebras. On the way we
discuss the structure of commutative C* -algebras. In this section we do not use
Theorem 1.1.9, except in the commutative case, for the outstanding part of the
proof of this theorem will depend on some of the results below.
Recall that the spectrum a(A) of A E ~(1{) is the set of those z E C for which
A - z[ has no (bounded two-sided) inverse; when A is self-adjoint, the spectral
radius rCA) appears in the fundamental equality
IIAII = rCA) := sup{lzll z E a(A)}. ( 1.26)
Since the presence of a unit is crucial in these definitions, we have to go through
the following considerations. A unit [in a J B-algebra QtjR is an element such that
A 0 I = A for all A E QtjR; a J B-algebra with a unit is called unital. When QtjR is
a J L B -algebra, its unit becomes a unit of the C* -algebra Qt, in that [A = A[ = A
for all A E Qt. This follows by putting B = C = [in (1.4), implying {A, [} = 0 for
all A, and subsequently applying (1.23). In any case, taking the adjoint of[*[ = [*
yields [*[ = [; hence [* = [. Also, (1.15) then implies 11[11 = l.
When a C* -algebra or a J L B -algebra has no unit, one may add one.
Proposition 1.2.1. For every C* -algebra without unit there exists a unique unital
C* -algebra Qt[, called the unitization of Qt, and an isometric (hence injective)
morphism Qt ---+ Qt[, such that Qt[/Qt c::: C.
42 I. Observables and Pure States

Let 2)(~) be the Banach algebra of all bounded linear maps on ~. Whether or
not ~ is unital, the map p : ~ ~ 2)(~), given by
p(A)B:= AB, ( 1.27)
is isometric. To see this, note that IIp(A)1I ::: IIAII by (1.14), whereas the opposite
inequality follows from (1.15).
Now let ~ be a nonunital C*-algebra, and form ~[ := ~ EB C. Extend p to ~[
by p(A+z)B := AB + zB, so that p(O+l) = H (the unit in 2)(~». Equipped
with the norm (1.17) and the algebraic structure of 2)(~), and with the involution
p(A+z)* := p(A*) + ZH, the vector space p(~[) is easily shown to be a unital
C* -algebra. Since p is a vector space isomorphism between ~[ and p(~[), one
may transfer the CO-algebraic structure on the latter to ~[. Restricted to ~ C ~[,
one recovers ~ as a C* -algebra. Uniqueness follows from 1.2.4.4 below. 0
Definition 1.2.2. Let ~ be either a unital C* -algebra, or the complexification of
a unital J LB-algebra ~JR. The spectrum a(A) of A E Ql is defined as the set of
those z E C for which A - zH has no (two-sided) inverse in Ql.
When ~ is nonunital, one puts a(A) := a[(A+O), where a[ stands for the
spectrum in the unitization of~.
In the nonunital case 0 always lies ina(A), as it is obvious from 1.2.1 that A E Ql
never has an inverse in QlIT. The theory of Banach algebras shows that a(A) is a
compact subset of C. For later use we note that
a(zA) = za(A); (1.28)
a(AB) U {OJ = a(BA) U {OJ (1.29)
for all A, B E ~; the first property is obvious, and the second follows, because for
z I- 0 the invertibility of AB - z implies the invertibility of BA - z. Namely, one
computes that (BA - z)-' = B(AB - z)-'A - z-'H.
For any locally compact Hausdorff space X, we regard the space Co(X) of all
continuous functions on X that vanish at infinity as a Banach space in the sup-
norm. A basic fact of topology and analysis is that Co(X) is complete in this
norm. Convergence in the sup-norm is the same as uniform convergence. What's
more, it is easily verified that C o(X) is a commutative C* -algebra under pointwise
addition, multiplication, and complex conjugation (defining the involution). When
X is compact, the function 1x, which is 1 for every x, is the unit H. One checks
that the spectrum of f E C(X) is simply the set of values of f. On Co(X), with X
noncompact, one has to supplement this set with zero.
Theorem 1.2.3. Let Ql be a commutative C* -algebra. There exists a locally com-
pact Hausdorffspace X for which Ql is isomorphic to Co(X). When ~ is unital, X
is compact, so that ~ ~ C(X). The space X is unique up to homeomorphism.
Similarly, an associative J LB-algebra ~JR is isomorphic to some Co(X, lR),
where X is locally compact; when QlJR has a unit, X is compact.
For simplicity we assume that ~ is unital; if it isn't, one would start by adjoining
a unit. The proof is based on a technique that applies to general commutative unital
 l The Structure of Algebras of Observables 43

Banach algebras; we state the main facts without proof. Consider the space ~(m)
of all nonzero multiplicative linear functionals w on m (that is, w : m ---* C satisfies
w(AB) = w(A)w(B) for all A, B). Each w E ~(m) is continuous, and satisfies
IIwll = w(lI) = 1. Thence it is easily seen that ~(m) is a closed subspace ofm' with
the w*-topology. By the Banach-Alaoglu theorem, ~(m) is therefore a compact
Hausdorff space in the relative w* -topology.
The Gelfand transform of A E m is the function A on ~(m) defined by
,.1.(w) := w(A). (1.30)
Since the relative w*-topology on ~(m) coincides with the weakest topology that
makes all functions A continuous, it is clear that the Gelfand transform maps m
into C(~(m». It is immediate that the image ofm in C(~(m» separates points. Re-
garding C(~(m» as a commutative Banach algebra in the sup-norm, as explained
above, the multiplicativity of each w E ~(m) implies that the Gelfand transform
is a homomorphism. The spectrum of A E m coincides with the set of values of A
on ~(m); in other words,
a(A) = a(,.1.) = {,.1.(w)1 w E ~(m)}. (1.31)
This implies that
11,.1.1100 = rCA), (1.32)
where the spectral radius rCA) is defined in (1.26). In any Banach algebra,
commutative or not, one has
rCA) = n-->oo
lim IIA n ll l / n . (1.33)

Now assume that m is a commutative C* -algebra; accordingly, regard C(~(m»

as a commutative C* -algebra. We first show that w E ~(m) is real on mlR. Pick
A EmIR, and suppose that w(A) = a + if3, where a, f3 E R Since w(lI) = 1, one
has weB) = if3, where B := A - all is self-adjoint. Hence for t E lR one computes
Iw(B + itlI)12 = f32 + 2tf3 + t 2. On the other hand, using IIwll = 1 and (1.15) we
estimate Iw(B + itlI)e :s IIBII2 + t 2. Hence f32 + tf3 :s IIBII2 for all t E R For
f3 > 0 this is impossible. For f3 < 0 we repeat the argument with B replaced by
-B. Hence f3 = 0, so that w(A) is real when A = A*. Consequently, by (1.30)
the function A is real-valued. Writing ((I(A) := A, condition (1.20) follows.
Secondly, for A EmIR one combines (1.15) with (1.33) to obtain IIAII = rCA),
which with (1.33) implies I AII 00 = II A II. For general A this equality then follows
via (1.15) in both m and C(~(m», and the fact that A* A E mjR. Thus the Gelfand
transform is isometric, and therefore injective. Finally, surjectivity follows from
the Stone-Weierstrass theorem.
The uniqueness of X follows from the fact that when X and Yare compact Haus-
dorff spaces, the commutative Banach algebras C(X) and C(Y) are isomorphic iff
X and Y are homeomorphic; this is equivalent to the statement that ~(C(X» is
homeomorphic to X. The homeomorphism is given by letting x E X correspond
with Wx E ~(C(X», defined by wAf) := f(x). The assumption that X is compact
and Hausdorff, hence normal, is used to prove that the evaluation map x 1---* Wx
44 I. Observables and Pure States

is injective, whereas the compactness of X implies that the evaluation map is

surjective.
The case of a commutative J L B -algebra Qt]R may be reduced to that of a
commutative C*-algebra by complexification; see Theorem 1.1.9. 0
Theorem 1.2.4. Let Qt be either a unital C* -algebra or the complexification of a
unital J L B -algebra QtJR.
1. The spectrum a(A) of A E Qt]R in Qt is equal to its spectrum in the C*-algebra
C*(A) generated by A and n. In particular, a(A) is a subset of the real line.
2. The compact Hausdorff space Ll(C*(A» of Theorem 1.2.3 is homeomorphic
to a(A), and the C*-algebra c*(A) of the preceding item is isomorphic to
C(a(A». Under this isomorphism the function A E C(X) is mapped into
ida (A) : t 1--+ t.
3. The continuous functional calculus for self-adjoint operators A on a Hilbert
space applies verbatim toQtJR: In particular, for each A E Qt]R and f E' C(a(A»
there exists an operator f(A) E Qt that is the obvious expression when f is
polynomial (and in the general case is given by uniformly approximating f
by polynomials on the basis of the Stone-Weierstrass theorem), and has the
properties
a(f(A» = f(a(A»; (1.34)
IIf(A)11 = IIflloo· (1.35)
4. For A E QtJR the norm is given by (1.26); for general A one has II A II =
-Jr(A * A). Hence the norm in a C* -algebra is unique, in that a * -algebra that is
a C* -algebra in some norm admits no other norm in which it is a C* -algebra.
If A = A*, then A z := A - z is normal for any z E C (i.e., A z commutes
with its adjoint Af). Suppose that z 1. a(A), so that A z is invertible. Consider the
commutative C*-algebra C*(Az' A;I) generated by A z , its inverse, and the unit.
By Theorem 1.2.3 one has C*(Az' A;I) ~ C(X), where X = Ll(C*(Az' A;I».
Since A z is invertible in C*(Az' A;I), it must be that Az(x) i= 0 for all x E X. It
is then elementary that A;I is a uniform limit of polynomials in A z , Af , and Ix.
Transferring this back to C*(Az' A;I) by the inverse of the Gelfand transform, it
follows that C*(Az' A;I) = C*(A z ) = C*(A). Hence when A - z is invertible in
2t its inverse lies in C*(A), which implies the first claim in 1.2.4.1.
Consider 1.2.3 and its proof with Qt = C*(A). We see from (1.31) and the fact
that A is real-valued for A E 2tJR that the spectrum of A in C*(A) is real. When
now Qt has the meaning of the present Theorem 1.2.4, the second claim in 1.2.4.1
follows from the first one just proved.
For 1.2.4.2, consider the map A : X -+ lR. It is clear from (1.31) that the image
of A is a(A). To prove injectivity, assume A(Wl) = A(W2)' Then wl(A) = w2(A)
by (1.30), whence wl(A)n = w2(A)n by multiplicativity of Wi E Ll(c*(A».
Since the linear span of all polynomials is dense in C*(A), and the Wi are con-
tinuous, this yields WI = W2. The map A is continuous, because A E C(X)
by 1.2.3. To prove continuity of the inverse, one checks that for z E a(A) the
 I The Structure of Algebras of Observables 45

functional A-l(Z) E L'l.(C*(A» maps A to Z (and hence An to zn, etc.). Fi-

nally, given the homeomorphism L'l.(C*(A» :::: a(A), the second isomorphism
in C*(A) :::: C(L'l.(C*(A») :::: C(a(A» follows from the topological fact that a
compact Hausdorff space X is detennined by C(X), and vice versa; cf. the proof
of 1.2.3.
The existence of the continuous functional calculus should now be obvious.
Since f(a(A)) is the set of values of f on a(A), (1.34) follows from (1.31),
with A replaced by f(A). The fact that for C*-algebras the Gelfand transfonn is
isometric yields (1.35).
The corresponding statements for a J L B -algebra follow by complexification,
using the commutative part of 1.1.9.
The first claim in 1.2.4.4 follows from (1.35) with f = id. The second claim
follows from the first, (1.15), and the property A* A E s,u1R. Hence the nonn is
detennined by the algebraic structure. •

For later use we record that for A E s,uIR, Theorem 1.2.4 implies

a(A) =0 {} A = O. (1.36)

1.3 Positivity, Order, and Morphisms

Recall that a (bounded) operator A E ~(H) on a Hilbert space is called positive
when (\}I, A \II) ~ 0 for all \II E H; this property is equivalent to A * = A and
a(A) ~ JR+. This notion of positivity induces a partial ordering :s in ~(H), in
which A :s B when B - A ~ O. Our aim is to generalize these concepts to
C*-algebras and J LB-algebras.

Definition 1.3.1. A partially ordered vector space (s,uIR, :s) consists of a real
vector space s,u1R and either one of the following equivalent data:

• A positive cone s,u+ in s,u1R; this is a subset for which (i) A E s,u+ and t E JR+
implies t A E s,u+, (ii) A, B E s,u+ implies A + B E s,u+, and (iii) s,u+ n -s,u+ = O.
• A linear partial ordering, i.e., a partial ordering :s in which A :s B implies
A + C :s B + C for all C E s,u1R and t A :s t B for all t E JR+.

The equivalence between these two structures is as follows: Given s,u~ one defines
A :s B if B - A E s,u~, and given:s one puts s,u~ = {A E s,u1R 10 :s A}.

Definition 1.3.2. Let s,u1R be a J LB-algebra or the self-adjoint part of a C*-

algebra. An element A E s,u1R is called positive when its spectrum is positive; i.e.,
a(A) C JR+. We write A ~ 0 or A E s,u+, where

(1.37)

It is immediate from (1.31) that A E s,u1R is positive iff its Gelfand transfonn A is
pointwise positive in C(a(A».
46 I. Observables and Pure States

Theorem 1.3.3. The set (1.37) of all positive elements of a C* -algebra Qt or a

J L B -algebra QtJR is a positive cone. This cone may alternatively be expressed as

Qt~ = (Azi A E QtJR} (1.38)

= (B* BI B E Qt}. (1.39)
Property (i) in 1.3.1 follows from (1.28). Since a(A) ~ [0, rCA)], we have
Ic - tl :s c for all t E a(A) and all c ::: rCA). Hence SUPtEa(A) lc1 a (A) - AI :s c
by (1.31) and 1.2.4.2, so that lIc1 a (A) - ,11100 :s c. Gelfand transforming back
to C*(A), this implies licIT - A II :s c for all c ::: II A II by 1.2.4.3. Inverting this
argument, one sees that if II cIT - A II :s c for some c ::: II A II, then a (A) c lR+ .
Using this with A replaced by A + B and c = IIAII + IIBllleads to property (ii).
Finally, when A E Qt+ and A E -Qt+, it must be that a(A) = 0, hence A = 0 by
(1.36). This proves property (iii).
If a(A) C lR+ and A = A *, then ,JA E QtJR is defined by the continuous
functional calculus for f = ,.f and satisfies ,JA2 = A. Hence Qt+ ~ {A21 A E
QtJR}. The opposite inclusion follows from (1.34) and 1.2.4.2. This proves (1.38).
The inclusion Qt+ ~ (B* BI B E Qt} is trivial from (1.38).
Lemma 1.3.4. Every A E QtJR has a decomposition A = A+ - A_, where
A+, A_ E Qt+ and A+A_ = O. Moreover, IIA± II :s IIA II.
Apply the continuous functional calculus with f = ida(A) = f+ - f-, where
id a (4)(t) = t, f+(t) = max{t, O}, and f-(t) = max(-t, O}. The bound follows
from (1.35) with A replaced by A±. •
Apply this decomposition to A = B* B (noting that A = A *); it follows that
(A_)3 = -(BA_)* BA_. Since a(A_) C lR+ as A_ is positive, we see from (1.34)
with f(t) = (3 that (A_)3 ::: O. Hence -(BA_)* BA_ ::: O.
Lemma 1.3.5. If -C*C E Qt+ for some C E Qt, then C = O.
Write C = D + iE, where D, E E QtJR (cf. (1.21», so that C*C = 2D2 +
2E2 - CC*. Applying (1.29) with A replaced by C and B by C*, we see that the
assumption a(C*C) C lR- implies a(CC*) C lR-; since C*C is the sum of three
positive terms, and Qt+ is a positive cone, it follows that C*C E Qt+. 1;-Ience the
starting assumption a(C*C) C lR- implies a(C*C) C lR+, so that a(C*C) = O.
Hence C*C = 0 by (1.36).
In a C*-algebra this implies C = 0 by (1.15). In a complexified J LB-algebra
we replace C by C* in the above argument, so that CC* = 0 as well as C*C = 0;
hence D2 + £2 = 0, whence D = £ = 0 by (1.8). •
The last claim before the lemma therefore implies BA_ = O. As (A_)3 =
-(BA_)* B A_ = 0, we see that (A_)3 = 0, and finally A_ = 0 by the continuous
functional calculus with f(t) = t 1/3. Hence B* B = A+, which lies in Qt+. •

When A = A * one checks the validity of

- II A II IT :s A :s II A II IT (l.40)
 I The Structure of Algebras of Observables 47

by taking the Gelfand transform of C*(A). The implication

- B ~ A~ B => IIAII ~ IIBII (1.41)

then follows, because - B ~ A ~ B and (1.40) with A replaced by B yield

-IIBII][ ~ A ~ liB II ][, so that a(A) S; [-IIBII, IIBII]; hence IIAII ~ IIBII by
(1.26).
An important consequence of (1.39) is the fact that inequalities of the type A I ~
A2 for A I, A2 E mlR are stable under conjugation by arbitrary elements B E m, so
that Al ~ A2 implies B* AlB ~ B* A2B. This is because AI ~ A2 is the same as
A2 - Al 2: 0; by (1.39) there is an A3 Em such that A2 - Al = A;A3. But then
(A3 B)* A3 B 2: 0, and this is nothing but B* A I B ~ B* A2 B. For example, replace
A in (1.40) by A* A, and use (1.15) in a C*-algebra, or (1.25) in a complexified
J LB-algebra. This yields A* A ~ IIA 11 2][. Applying the above principle to any
A, BE m gives

(1.42)

Definition 1.3.6. A positive map Q : m ~ lB between two C* -algebras is a

linear map with the property that A 2: 0 in mimplies Q(A) 2: 0 in lB.

Proposition 1.3.7. A positive map between C*-algebras is *-preserving and

bounded.

One infers from 1.3.4 that Q(mll~J S; lBlR; the C-linearity of Q then proves the
first claim.
For the second claim, let us first show that boundedness on m+ implies bound-
edness on m. Using (1.21) and 1.3.4, we can write A = A~ - A~ + iA~ - iA':.,
where A~ etc. are positive. Since IIA'II ~ IIAII and IIA"II ~ IIAII by (1.21), we
have II BII ~ IIAII for B = A~, A~, A~, or A':. by 1.3.4. Hence ifllQ(B)1I ~ cliBIl
for all B E m+ and some c > 0, then IIQ(A)II ~ 4c1lAIi.
Now assume that Q is not bounded; by the previous argument it is not bounded
on m+, so that for each n E N there is an An E mi such that II Q(An) II 2: n 3 (here
mi consists of all A E m+ with II A II ~ 1). The series L:o n- 2An obviously
converges to some A E m+. Since Q is positive, we have Q(A) 2: n- 2 Q(An) 2: 0
for each n. Hence by (1.41), IIQ(A)II 2: n- 2 I1Q(A n )1I 2: n for all n EN, which
is impossible. Thus Q is bounded on m+, and therefore on m by the previous
paragraph. •

Since a morphism cp : m ~ 'B satisfies cp(B* B) = cp(B)*cp(B), it is clear from

(1.39) that a morphism is a positive map. For later reference we collect this, and
other good properties of morphisms. In preparation, we define a left ideal in a
C*-algebra m as a closed linear subspace 'J S; m such that A E 'J implies BA E 'J
for all B E m. Similarly, a right ideal is a closed linear subspace 'J S; m such that
A E 'J implies ABE 'J for all B E m. An ideal is both a left and a right ideal.
A proper ideal cannot contain a unit ][; in order to prove properties of ideals one
needs a suitable replacement of a unit.
48 I. Observables and Pure States

Definition 1.3.8. An approximate unit in a nonunital C* -algebra Qt consists of

a directed set A (i.e., a set with a partial order ~ in which for all AI, .1..2 there is
a A :::: Ai, i = 1,2) and a family (l[AhEA of elements ofQtfor which H~ = HA and
a(HA) C [0, 1] (whence IIHAII ~ 1), so that for all A E Qt one has
lim IIHAA - All
A-+oo
= A->OO
lim IIAHA - All = O. (1.43)

For example, an approximate unit in Co(R) may be constructed with A = N,

taking Hn to be a continuous function that is 1 on [-n, n] and vanishes for Ix I >
n + 1. More generally, it can be shown that every nonunital C* -algebra Qt has an
approximate unit; when Qt is separable, A may be chosen countable. The technique
of approximate units allows us to prove the main properties of ideals in C* -algebras.
Theorem 1.3.9. Let J be an ideal in a C* -algebra Qt.
1. Ifl E JthenJ* E J;inotherwords,everyidealinaC*-algebraisself-adjoint.
2. The quotientQt/J is a C*-algebra in the norm IIr(A)1I := infJEJ IIA + lli. the
multiplication r(A)r(B) := r(AB), and the involution r(A)* := r(A*).
Note that the involution in Qt/J is well-defined because of 1.3.9.1.
Put J* := {A*I A E J}, and note that J n J* is a C*-subalgebra of Qt. Hence it
has an approximate unit {HA}. Pick 1 E J, and use (1.15) and 1.3.8 to estimate
IIJ* - J*HAII 2 ~ 11(1*1 - J* JHA)II + 11(1 J* - 1 J*H}JII.
Now J* 1 and 1 J* both lie in J n J*, so that both terms on the right-hand side
vanish for A --* 00. Hence J* is a norm-limit of elements in J; since J is closed,
it follows that J* E J.
We omit the well-known proof that Qt/J is a Banach algebra in the given norm
and multiplication. To prove the C*-property (1.15), we first note that
IIr(A)II = lim IIA - AHA II, (1.44)
A->OO

for any A E Qt and approximate unit {HA} in J. To prove this, we first add a unit to
Qt if necessary. For any 1 E J we have A - AHA = (A + l)(H - HA) + 1 (HA - H),
so that IIA - AHA II ~ IIA + 111 IIH - HAil + IIJHA - 111. Since
IIH - HAil ~ 1 (1.45)
from 1.3.8 and the proof of 1.3.3, we obtain limhoo II A - AHA II ~ II A + 1 II. For
each E > 0 we can choose 1 E J such that IIr(A)II + E :::: IIA + 111. Using this
1 in the previous inequality, letting E --* 0, and noting the obvious II A - AHA II ::::
IIr(A)II, we obtain (1.44).
Successively using (1.44), (1.15) in Qt, (1.45), (1.44) once again, and the def-
inition of the C*-operations in Qt/J, we obtain IIr(A)1I 2 ~ IIr(A)r(A)*II. By the
argument preceding 1.1.7, this implies (1.15). 0
Theorem 1.3.10. Let qJ : Qt --* ~ be a morphism between C* -algebras.
1. The kernel of qJ is an ideal in Qt. Conversely, every ideal in a C* -algebra is the
kernel of some morphism.
 I The Structure of Algebras of Observables 49

2. One has IIrpll = 1, and therefore IIrp(A)1I ::: IIAII for all A E Ql.
3. The map rp is isometric when it is injective.
4. The image rp(Ql) is a C* -subalgebra of~.
5. The map rp is positive.
It is clear that a (rp(A* A» ~ a(A* A), so that the inequality in 1.3.10.2 follows
from 1.2.4.4. It follows that rp is continuous, so that ker(rp) is closed. It is then
obvious from (1.19) that ker(rp) is an ideal. On the other hand, a given ideal J is
the kernel of the canonical projection t' : Ql ~ QljJ. Now QljJ is a C*-algebra
and rp := t' is a morphism with J = ker(rp).
Assume that there is aBE Ql for which IIrp(B)1I =I=- IIBII. By (1.15), (1.19), and
(1.20) this implies IIrp(B* B)II =I=- IIB* BII. Put A := B* B, noting that A* = A. By
(1.26) we must have a(A) =I=- a(rp(A». Since a(rp(A» ~ a(A) in any case, this
implies a(rp(A» C a(A). By Urysohn's lemma there is a nonzero f E C(a(A»
that vanishes on a(rp(A», so that f(rp(A» = O. By the continuous functional
calculus we have rp(f(A» = 0, proving 1.3.10.3 by reductio ad absurdum.
Define 1/1 : Qlj ker(rp) ~ ~ by 1/1 0 t' = rp, with t' : Ql ~ Qlj ker(rp) the
canonical projection. Then 1/1 is an injective morphism, so that it is isometric by
1.3.10.3. Hence IIrpll = 1, since IIt'il = 1. Since 1/I(Qlj ker(rp» = rp(Ql), it follows
that rp has closed range in ~. Since rp is a morphism, this implies that rp(Ql) is a
C*-algebra in the norm of lB. •

1.4 States
We now change our perspective, and pass from observables to states.
Definition 1.4.1. A state on a C* -algebra Ql is a linear map w : Ql ~ C for
which w(A) ~ 0 for all A E Ql~ (positivity) and IIwll = 1 (normalization). The
state space S(Ql) ofQl is the set of all states on Ql.
For example, on Ql = ~(1i) every unit vector 0 E 1i defines a state w by
w(A) = (0, 1l'(A)O). (1.46)
This is, indeed, the original notion of a state as used in quantum mechanics.
Combining 1.3.4 with positivity, we see that a state is real-valued on QlIR; in
view of (1.21) we then infer that a state is a Hermitian functional on Ql, in that
w(A*) = w(A) (1.47)
for all A E Ql. In particular, a state is determined by its values on Ql+. Combining
the positivity ofw with (1.39) one sees that (A, B){J) := w(A* B) defines a pre-inner
product on Ql. Hence from the Cauchy-Schwarz inequality we obtain the useful
bound
Iw(A* B)1 2 ::: w(A* A)w(B* B). (l.48)
Proposition 1.4.2. A linear map w : Ql ~ C on a unital C* -algebra is positive
iJfw is bounded and IIwll = well). Hence a state w on a unital C*-algebra may
50 I. Observables and Pure States

equivalently be characterized as a positive linear functional for which w(lI) = 1

or as a bounded linear functional for which IIwll = w([) = 1.
A state w on a C* -algebra without unit has a unique extension to a state Wn on
the unitization Qtn, given by

wn(A + A[) := w(A) + A. ( 1.49)

When w is positive and A = A * we have, using (1.40), the bound Iw(A)1 ::::
w([) II A II. For general A the same inequality follows from (l.48) with A = [,
(1.15), and the bound just derived. The upper bound is reached by A = [.
To prove the converse claim, we first note that the argument after (1.33) may
be copied, showing that w is real on QtlR. Next, we show that A ::: 0 implies
w(A) ::: O. Choose s > 0 small enough so that 11K - sAil :::: 1. For w i- 0 one
has 11K - sAil ::: Iw([ - s A)I/w([), so that Iw([) - sw(A)1 :::: w([). This is only
possible when w(A) ::: o.
As to the positivity of Wn, we observe that Iw(A - A[)JI ~ 0 for any approximate
unit in Qt. Using (1.48) with B = [,\., this leads to Iw(A)1 2 :::: w(A *A). Combining
this inequality with 0.47), the definition (1.49) leads to w[«A + A[)*(A + A[» :::
Iw(A) + 3:1 2 ::: O. Hence w is positive by (1.39). •

An important feature of a state space S(Qt) is that it is a convex set. (A convex

set C in a vector space V is a subset of V such that the convex sum AV + (1 - A)W
belongs to C whenever v, W E C and A E [0, 1]. Geometrically, this means that
the line segment between any two points in C lies in C. It follows that a finite sum
Li Pi Vi belongs to C when all Pi ::: 0 and Li Pi = 1, and all Vi E C.) In the
unital case it is clear that S(Qt) is convex, since both positivity and normalization
are clearly preserved under convex sums. In the nonunital case one arrives at this
conclusion most simply via (1.49).
Let S(Qt) be the state space of a unital C* -algebra Qt. Each element w of S(Qt)
is continuous, so that S(Qt) c Qt*. Since w* -limits obviously preserve positivity
and normalization, we see that S(Qt) is closed in Qt* if the latter is equipped with
the w* -topology. Moreover, S(Qt) is a closed subset of the unit ball of Qt*, so that
S(Qt) is compact in the relative w*-topology by the Banach-Alaoglu theorem. It
follows that the state space of a unital C* -algebra is a compact convex set.
The very simplest example is Qt = <C, in which case S(Qt) is a point. The next
case is Qt = <C EB <C = <c2 . The dual is <c 2 as well, so that each element of (<C 2 ) * is
of the form W(A+fL) = CtAt + C2A2. Positive elements of <C EB <C are of the form
A+fL with A ::: 0 and fL ::: 0, so that a positive functional must have Ct ::: 0 and
+
C2 ::: O. Since [ = 1 1, normalization yields Ct + C2 = 1. Identifying 0 with the
functional mapping A+fL to A, and 1 with the one mapping it to fL, we conclude
that S(<C EEl <C) may be identified with the interval [0, 1].
Now consider Qt = Wh (<C). We identify Wh (<C) with its dual through the pairing
w(A) = TrwA. It follows that S(Qt) consists of all positive 2 x 2 matrices p with
Tr p = 1; these are the density matrices of quantum mechanics_ To identify S(Qt)
 1 The Structure of Algebras of Observables 51

with a familiar compact convex set, we parametrize

_
p- ~ ( 1+x Y + iz ) , ( 1.50)
2 y - iz 1- x

where x, y, Z E lR. The positivity of this matrix then corresponds to the constraint
x 2 + y2 + Z2 :s 1. Hence S(VJ12(1C» is the unit ball in lR 3 .
There are lots of states:
Proposition 1.4.3. For every A E Qt and a E a(A) there is a state Wa on Qtfor
which w(A) = a. When A = A* there exists a state W such that IW(A)I = IIAII.
If necessary we add a unit to Qt; in the present context this is justified by (1.49).
Define a linear map wa : ICA $IC[ --+ IC by wa(AA + J1.[) := Aa + J1.. Since
a E a(A), one has Aa + J1. E a(AA + J1.[); this easily follows from the definition of
a(A). In any Banach algebra one has r(A) :s II A II; applying this with A replaced
by AA + J1.[ implies IWa(AA + J1.[)1 :s /lAA + J1.[II. Since wa([) = I, it follows that
IIwII = 1. By the Hahn-Banach theorem there exists an extension Wa of W to Qt of
norm 1. Since IIwa II = wa([) = I, this extension is a state, which clearly satisfies
wa(A) = wa(A) = a.
Since a(A) is closed, there is an a E a(A) for which r(A) = lal. For this a, and
A = A*, one has IW(A)I = lal = r(A) = IIAII by (1.26); cf. 1.2.4.4. •
Corollary 1.4.4. For all A E Qt]R one has
IIAII = sup{lw(A)1 I W E S(Qt)}. (1.51)
Hence ijw(A) = Ofor all states wE S(Qt), then A = O.
Our goal is to give a geometric realization of a unital C* -algebra Qt as a certain
function space, somewhat in the spirit of Theorem 1.2.3. A function f on a convex
set K is called affine if it preserves convexity, that is, if

The space A (K, lR) of all real-valued affine continuous functions on a compact
convex set K has a positive cone A(K, lR)+, consisting of all positive functions
(cf. 1.3.1). Equivalently, A(K, lR)+ has a linear partial ordering, in which f :s g
when f(w) :::: g(w) for all W E K. Also, A(K, lR)+ is a Banach space in the
sup-norm in the case that K is Hausdorff.
Theorem 1.4.5. The selfadjoint part Qt]R of a unital C* -algebra Qt is isomorphic
as a partially ordered Banach space to the space A(S(Qt), lR) of all real-valued
affine continuous functions on the state space S(Qt) ofQt (equipped with the relative
w* -topology).

The isomorphism in question is given by (1.30), now seen as a map from Qt to

the space of functions on the state space S(Qt). It is immediate that this transform
is injective. It is a well-known fact in functional analysis that a Banach space Qt]R
may be identified under (1.30) with the subspace of its double dual Qt~* consisting
52 I. Observables and Pure States

of all w*-continuous linear functionals on ~IR' Since linear functionals on ~* are

automatically affine on S(~), and since we know that a state is real-valued on
~lR, the transform (1.30) maps ~lR into A(S(~),IR). Reinterpreting (1.51) as the
equality IIAII = 11,11100, it follows that the map A ~ A is isometric.
Without proof we state that any Hermitian functional q; E ~* on a C* -algebra ~
has a unique decomposition q; = tiwi - t2w2, where Wj E S(~) and tj E R+. This
implies that an element f of A(S(~), R) has a unique extension to a Hermitian
linear functionalj on ~* , which is evidently w* -continuous, and is therefore given
by an element A E ~ (cf. the preceding paragraph). The function A in (1.30) is
evidently f, so that the image of~ under A ~ A is all of A(S(~), R).
Finally, it is trivial from the pertinent definitions that the transform (1.30)
preserves positivity. 0
As promised, we now complete the proof of Theorem 1.1.9. We discuss the
unital case; the nonunital case may be reduced to this, using 1.2.1.
For the first half it remained to be shown that the norm on the self-adjoint part of
a C*-algebra satisfies (1.10). This follows from the order inequality A2 ::::: A2 + B2
(which is derived from (1.38) and the linearity of the partial ordering) and (1.41).
In the second half we need to prove that (1.25) defines a norm on ~. Firstly, the
property IIAII = 0 => A = 0 follows from Lemma 1.3.5. Secondly, the triangle
inequality follows by successively using (1.25) with A replaced by A + B, (1.51),
(1.48), and again (1.25) and (1.51), this time from right to left. Finally, we use
(1.42); taking the norm in ~lR and using (1.25) yields (1.14). •
Thus from now on a J L B -algebra and the self-adjoint part of a C* -algebra will
be one and the same object.

1.5 Representations and the GNS-Construction

In the theory of C* -algebras, Hilbert spaces are most naturally regarded as modules,
and the material of this section explains how the usual Hilbert space framework
of quantum mechanics emerges from the algebraic setting.
Definition 1.5.1. A representation of a C* -algebra ~ on a Hilbert space 1f is a
morphism n : ~ ---+ !.l3(1f).
From the Jordan-Lie point of view, this means that n : ~lR ---+ !.l3(1f)lR is a
morphism of Jordan-Lie algebras (cf. 1.1.3); here the Jordan-Lie structure on both
spaces is given by (1.22). In view of 1.3.10.2 a representation n is automatically
continuous; hence lin (A) II ::::: II A II for all A E ~. When n is faithful, this sharpens
to IIn(A)1I = IIAII by 1.3.10.3.
There is a natural equivalence relation in the set of all representations of ~:
Two representations nl, n2 on Hilbert spaces 1f1o 1f2, respectively, are called
equivalent if there exists a unitary isomorphism U : 1f1 ---+ 1f2 such that
Unl(A)U* = n2(A) for all A E ~.
The map n(A) = 0 for all A E ~ is a representation; more generally, such trivial
n may occur as a summand. To exclude this possibility, one says that a represen-
 I The Structure of Algebras of Observables 53

tation is nondegenerate if 0 is the only vector annihilated by all representatives

of~.
A representation Jr is called cyclic if its carrier space H contains a cyclic vector
Q for Jr; this means that the closure of Jr(~)Q (which in any case is a closed
subspace of H) coincides with H.
Proposition 1.5.2. Any nondegenerate representation Jr is a direct sum of cyclic
representations.
The proof uses a lemma that appears in many other proofs as well.
Lemma 1.5.3. Let 9)1 be a *-algebra in !J3(H), IJ1 a nonzero vector H, and p
the projection onto the closure of9)1lJ1. Then p E 9)1' (that is, [p, A] = ofor all
A E 9Jt).
If A E 9)1, then ApH ~ pH by definition of p. Hence pJ.. Ap = 0 with
pJ.. = IT - p. When A = A* this yields [A, p] = 0; by (1.21) this is true for all
A E 9)1. •

Apply this lemma with 9)1 = Jr(~); the assumption of nondegeneracy guarantees
that p is nonzero, and the conclusion implies that A f-* pJr(A) defines a cyclic
subrepresentation of ~ on pH. This process may be repeated on pJ..H, etc. •
If Jr is a nondegenerate representation of a C* -algebra ~ on H, then any unit
vector Q E H defines a state w E S(~), referred to as a vector state relative to Jr ,
by means of (1.46). Conversely, from any state w E S(~) on ~ one can construct
a cyclic representation Jr w on a Hilbert space Hw with cyclic vector Q w in the
following way. We restrict ourselves to the unital case; the general case follows by
adding a unit to ~ and extending w to ~ll by (1.49).
Construction 1.5.4.
1. Given w E S(~), define the sesquilinear form (, Xl on ~ by
(A, B)~:= w(A*B). ( 1.53)
Since w is a state, hence a positive functional, this form is positive semidefinite
(this means that (A, A)O' 2: 0 for all A). Its null space
N w = {A E ~ I w(A* A) = O} (1.54)
is a left ideal in ~.
2. The form (, )0' projects to an inner product (, )w on the quotient ~/Nw' If
Vw : ~ --+ ~/ N w is the canonical projection, then by definition
(VwA, V",B)", := (A, B)~. (1.55)
The Hilbert space H", is the closure of~/Nw in this inner product.
3. The representation Jr",(~) is firstly defined on ~/N", c H", by
Jr",(A)V",B := V",AB; (1.56)
it follows that Jr", is continuous. Hence Jr",(A) may be defined on all ofH", by
continuous extension of (1.56).
54 I. Observables and Pure States

4. The cyclic vector is defined by Ow = Vw][, so that

(Ow, JTw(A)Ow) = w(A) VA E ~. (1.57)
We now prove the various claims made here. First note that the null space N w
of (, )~ can be defined in two equivalent ways, since
N w := {A E ~ I (A, A)~ = O} = {A E ~ I (A, B)~ = 0"1 B E ~}. (1.58)
The equivalence follows from (1.48). With the continuity of w, the equality (1.58)
implies that N w is a left ideal. Hence JT w in (1.56) is well-defined on the dense
subspace ~/Nw C 'H. w, where it clearly satisfies (1.19), with qJ ~ JTw. Also, (1.20)
may be verified from (1.55) and (1.53).
To prove thatJTw is bounded on ~/Nw, we compute IIJTw(A)1II112 for 111 = VwB,
where A, B E ~. From (1.55) and (1.53) one has IIJTw(A)1II112 = w(B* A* AB).
By (1.42) and the positivity of w one has w(B*A*AB) S IIAI1 2w(B*B). But
w(B*B) = 1111111 2, so that IIJTw(A)1I s IIAII. Hence JT w may be extended to all of
'H. w, where (1.55) and (1.53) hold by continuity.
Proposition 1.5.5. Ifa representation JT (~) on 'H. is cyclic, then the GNS-represen-
tation JTw(~) on 'H. w defined by any vector state 0 (corresponding to a cyclic unit
vector 0 E 'H.) is equivalent to JT(~).
The operator U : 'H. w ~ 'H. implementing the equivalence is initially defined
on the dense subspace JTw(~)Ow by UJTw(A)Ow = JT(A)O; this operator is well-
defined, for JTw(A)Ow = 0 implies JT(A)O = 0 by the GNS-construction.1t fonows
from (1.57) that U is unitary as a map from 'H. w to U'H. w, but since 0 is cyclic for
JT, the image of U is 'H.. Hence U is unitary. One verifies that U intertwines JT w
and JT. •
Corollary 1.5.6. If the Hilbert spaces 'H. I, 'H.2 oftwo cyclic representations JTI, JT2
each contain a cyclic vector 01 E 'H.I, 02 E 'H.2 such that
wl(A):= (0" JTI(A)OI) = (02, JT2(A)02) =: w2(A)
for all A E ~, then JTI (~) and JT2(~) are equivalent.
By 1.5.5 the representation JTI is equivalent to the GNS-representation JT wl , and
JT2 is equivalent to JTW2 • On the other hand, JTW1 and JTW2 are induced by the same
state WI = W2, so they must coincide. •
The state w is called faithful when its GNS-representation JTw is faithful. This
is guaranteed when N w = 0, but note that even in that case 'Hw does not coincide
with ~, as the topology on ~ in the operator-norm is finer than the topology of the
norm IIAII~ := (A, A)~.
The GNS-construction leads to a simple proof of Theorem 1.1.8, which uses the
following notion.
Definition 1.5.7. The universal representation JTu ofa C* -algebra ~ is the direct
sum ofall its GNS-representations JT w , w E S(~); hence it is defined on the Hilbert
space 'H. u = EBwES(~)'H.W'
 1 The Structure of Algebras of Observables 55

Theorem 1.1.8 then follows by taking H = Hu; the desired isomorphism is

nu. If nu(A) = 0 for some A E Qt, then nw(A)Q w = 0 for all states w, whence
Ilnw(A)Qw 112 = w(A * A) = 0 by the GNS-construction, so that A *A = 0 by
1.4.3, and finally A = 0 by (1.15). Hence nu is faithful, and therefore isometric by
1.3.10.3. •

1.6 Examples of C* -Algebras and State Spaces

In this section we give some elementary examples of C* -algebras.

Example 1.6.1. Commutative C* -algebras

Let X be a discrete space. Take Qt := t:o(X), which is the closure (in the sup-
norm) of t:c(X). The space t:o(X) is a C*-algebra under pointwise multiplication
and complex conjugation; see 1.2. By elementary Banach space theory, the dual
of Qt is Qt* = t:1(X) under the pairing p(f) = Tr pf := LXEx p(x)f(x). The
positive cone in Qt or Qt* consists of the positive functions f or p. The state space
S(Qt) is the set of those positive functions p for which Tr p = Lx p(x) = l.
For example, for a given y EX, the function p = 8y , defined by 8y (x) : = 8xy , is a
state; one clearly has 8y (f) = f(y). Hence by Corollary 1.5.6 the one-dimensional
representation ny, defined on Hy = <C by ny(f) := f(y), is equivalent to the
GNS-representation n8 y (the pertinent cyclic vector in <C is simply Q = I).
A positive normalized function on X defines a faithful state when it is strictly
positive on X. The GNS-representation n p(t:o(X» of a faithful state p is equivalent
to the representation n on H = £2(X) (with counting measure) by multiplication
operators, i.e., n(f)'I1(x) := f(x )'11 (x). To see this, we first write the inner product
in t: 2 (X) as ('11, <1» = Tr'l1*<I> := Lx 'I1(x)<I>(x). Then note that since Tr p = I,
one has pl/2 E £2(X). It is clear from the property p(x) > 0 for all x that pl/2
is a cyclic vector for n(£o(X», with the property (pI/2, n(f)pI/2) = p(f) for all
f E £o(X). The equivalence between n p and n then follows from Corollary 1.5.6.
Adding the fact that the double dual of Qt is t:o(X)** = £OO(X), we summarize
the situation by

(1.59)

When X is finite all inclusions are replaced by equalities; when X is infinite all
inclusions are strict.
Now take X to be a locally compact Hausdorff space, and put Qt := Co(X)
with the sup-norm; this is the closure of Cc(X). Recall that a Radon measure is
a Borel measure that is inner regular with respect to compact sets. By the Riesz
representation theorem, Qt* is the space of all complex Radon measures JL on X
with finite total mass JL(X). With Qt+ consisting of the positive functions in Qt, the
dual cone Qt*+ is the subspace of Qt* of nonnegative finite Radon measures. The
state space S(Qt) = Mt(X) then consists of the probability measures on X. The
GNS-representation nil of a state JL E S(Qt) is realized on Hil = L2(X, JL), on
which nll(f) is f as a multiplication operator.
56 I. Observables and Pure States

Example 1.6.2. Noncommutative C* -algebras

When a Hilbert space H = (f is finite-dimensional, the "maximally noncom-

mutative" C* -algebra of operators on H is the algebra VJ'tN(C) of N x N matrices.
The appropriate generalization of VJ'tN(C) to the infinite-dimensional case is as
follows. A projection in a *-algebra is an element satisfying p2 = p* = p.

Definition 1.6.3. Let H be a Hilbert space. The *-algebra lJ3 j(H) of finite-rank
operators on H is the (finite) linear span ofallfinite-dimensional projections on 1i.
In other words, an operator A E lJ3(H) lies in lJ3 j(H) when A H := {A \II I \II E H}
is finite-dimensional.
The C*-algebra lJ3 o(H) of compact operators on H is the norm-closure of
lJ3 f (H) in lJ3 (H) (with all C* -algebraic operations borrowed from lJ3 (H»).
It is clear that lJ3 j (H) is a *-algebra, since p* = p for any projection p. It is
obvious that lJ3 j(H) is closed under right multiplication by elements of lJ3(H);
since it is a * -algebra, it is therefore also closed under left multiplication. By
continuity of multiplication in lJ3(H), it follows that lJ3o(H) is an ideal in lJ3(H). It
is easily verified that the unit operator [ lies in lJ3 o(H) iff H is finite-dimensional.
We know from the theory of single operators on a Hilbert space that the image of
the unit ball in H under an element A E lJ3 o(H) is compact (in the strong topology
on H); this explains the name of lJ3 o(H). A self-adjoint operator A E lJ3(H) is
compact iff A = Li aj [\IIi] (norm-convergent sum), where each eigenvalue aj has
finite multiplicity, and limi-+oo lai I = 0 (where the eigenvalues have been ordered
so that aj :::; a j when i > j). In other words, the set of eigenvalues is discrete, and
can have only 0 as a possible accumulation point.
We now wish to determine the state space of lJ3 o(H). This involves the study of
a number of other subspaces of lJ3(H), whose definition we recall.

Definition 1.6.4. The Hilbert-Schmidt norm II A 112 of A E lJ3(H) is defined by

IIAII~ := L IIAe;ll2 = Tr(IAI 2), (1.60)

where {edj is an arbitrary basis ofH; the right-hand side is independent of the
choice of basis. Also, IAI := .j A* A is defined by the continuous junctional cal-
culus. The Hilbert-Schmidt class lJ3 2(H) consists of all A E lJ3(H) for which
IIAII2 < 00.
The trace norm II A III of A E lJ3(H) is defined by
IIAII, := IIIAII/211~ = TriAl. (1.61)
The trace class lB, (H) consists of all A E lJ3(H) for which II A II, < 00.

The noncommutative analogue of (1.59) is as follows.

Theorem 1.6.5. One has the inclusions

lB j (H) ~ lJ3, (H) = lJ3 o(H)* ~ lJ3 2 (H) ~ lJ3 o(H) ~ lJ3(H) = lJ3 o(H)**,
(1.62)
 I The Structure of Algebras of Observables 57

where the (isometric) identification of fJ3 ICH) with fJ3 oCH)* is made through the
pairing
p(A) = TrpA, (1.63)
and fJ3(H) is (isometrically) identified with fJ3 1(H)* = fJ3o(H)** through the
pairing A(p) = Tr pA.
The definition of the nonns in (1.6.4) easily leads to IIA II s II A IIi for i = 1,2.
Since fJ3o(H), fJ31 (H), and fJ3 2(H) are the completions of fJ3 j (H) in the nonns II . II,
II . Ill. and II . 112, respectively, these inequalities imply that fJ3 i (H) ~ fJ3 o(H) for
i = 1,2. Using the characterization of self-adjoint compact operators mentioned
above, one then infers from 1.6.4 that II A 112 S II A Ill, so that fJ31 (H) ~ fJ3 2(H).
The inclusions fJ3 1(H) ~ fJ3 o(H)* and fJ3(H) ~ fJ3 1(H)* both follow from the
(nontrivial) estimate
ITrpAI s IIAlillplll. (1.64)
To show that fJ3o(H)* ~ fJ3 1(H) one restricts a given element p E fJ3 o(H)* to
fJ3 2 CH), on which it is continuous. Now, the operator space fJ3 2(H) is a Hilbert
space in the inner product (A, B) := Tr A* B, so that by Riesz-Fischer there must
be an operator p E fJ3 2 (1t) such that p(A) = Tr pA for all A E fJ32(H). One then
shows that ITr pip II slip II for any finite-dimensional projection p, which implies
that Ilplll s Ilpll, so that p E fJ3 1(H). With the opposite inequality from (1.64),
this proves that fJ3 1(1t) = fJ3 o(H)* isometrically.
To establish the inclusion fJ3 1(H)* ~ fJ3(H), pick A E fJ3 1(H)* , and define a
quadraticform QA on H by QA (\}I, <1» := ,,1(1<1» (\}II). Here the operator 1<1» (\}II is
defined by 1<1»(\}IIQ := (\}I, Q)<1>. This fonn is easily seen to be bounded by 11,,1 II,
so that it is implemented by a bounded operator A, in that Q A (\}I, <1» = (\}I, A <1».
By linear extension to fJ3 j(H) and subsequently continuous extension to fJ3 1(H),
this implies thatA(p) = Tr pA, with IIA II s 11,,1 II. Since (1.64) implies the opposite
inequality, this proves the last claim. 0
Corollary 1.6.6. The state space S(fJ3 o(H» of the C*-algebra fJ3 o(H) of all
compact operators on some Hilbert space H consists ofall density matrices, where
a density matrix is an element p E fJ3 1(H) that is positive (p 2: 0) and has unit
trace (Tr p = 1), and the corresponding state is defined in (1.63).
Since p E fJ3 1(H) is compact, one may diagonalize it by p = Li Pi [\}Ii]. Using
A = [\}Iil, which is positive, the condition p(A) 2: 0 yields Pi 2: O. Conversely,
when all Pi 2: 0, the operator p is positive. The nonnalization condition II pili =
L Pi = 1 completes the characterization of S(fJ3 o(H». •
Proposition 1.6.7.
1. For each unit vector \}I E H the GNS-representation rr 1ft (fJ3 o(H» corresponding
to the density matrix p = [\}I] is equivalent to the defining representation.
2. The GNS-representation rr p corresponding to a faithful state p on fJ3o(1t) is
equivalent to the representation R-p(fJ3 o(H» on the Hilbert space fJ3 2(H) of
Hilbert-Schmidt operators given by left multiplication, i.e., R-p(A)B := AB.
58 I. Observables and Pure States

The first claim is immediate from the property Tr [\{J] A = (\{J, A \{J) and 1.5.5.
For the second, it is obvious from 1.6.4 that for A E l13(H) and B E 113 2 (H)
one has I A B 112 ::: II A II II B 112, so that the representation Tr p is well-defined. When
p E 113 1(1i) and p ::: 0, then pl/2 E 113 2(H), and it is easily seen that pl/2 is
cyclic for Tr p(l13 oCH» when p is faithful. Using the fact that for A, B E 113 2 (H)
one has Tr AB = Tr BA, we compute (pI/2, Trp(A) p I /2) = peA). The equivalence
between Trp and Trp then follows from 1.5.6. •

1.7 Von Neumann Algebras

In this section we state some basic facts about von Neumann algebras (which
will be used only as ancillary tools). The com mutant 9:n' of some collection 9:n
of bounded operators on a Hilbert space is the set of all bounded operators that
commute with all elements of 9:n; the bicommutant 9:n" is the commutant of 9:n'.
One verifies that 9:n'" = 9:n'. The main result is the so-called double commutant
theorem, which we will first state in the finite-dimensional case.
Proposition 1.7.1. Let H = en be afinite-dimensional Hilbert space, and let 9:n
be a *-algebra (and hence a C*-algebra) in l13(H) = 9:n n (C) containing IT. Then
9:n" = 9:n.

Choose some \{J E H, form the linear subspace 9:n\{J of H, and consider the
projection p = [9:n\{J] onto this subspace. By Lemma 1.5.3 one has p E 9:n'. Hence
A E 9:n" commutes with p. Since ][ E 9:n, we therefore have \{J = IT\{J E 9:n\{J,
so \{J = p\{J, and A\{J = Ap\{J = pA\{J E 9:n\{J. Hence A\{J = Ao\{J for some
Ao E 9:n.
Choose \{JI, ... , \{In E H, and regard Q := \{JI+'" +\{In as an element of
H n := tBnH :::::: H ® en (the direct sum of n copies of H), where \{Ji lies in the ith
copy. Identify l13(Hn) with the algebra 9:n n (l13(H» of n x n matrices with entries
in l13(H), and embed 9:n in 9:nn (l13(H» by A f--7 8(A) := AIT~, where IT~ is the
unit in 9:nn (l13(H»; this is the diagonal matrix in 9:n n (l13(H» in which all diagonal
entries are A.
Now use the first part of the proof, with H, 9:n, A, and \{J replaced by Hn,
8(9:n), A := 8(A), and Q, respectively. Hence given \{JI, ... , \{In and 8(A) E 8(9:n)
there exists Ao E 8(9:n)" such that 8(A)Q = AoQ. For arbitrary B E 9:nn(l13(H»,
compute ([B, 8(A)])ij = [B ij , A]. Hence 8(9:n)' = 9:nn (9:n'). It is easy to see that
9:nn (9:n')' = 9:n n (9:n"), so that 8(9:n)" = 8(9:n"). Therefore, Ao = 8(A)o for some
Ao E 9:n. Hence A\{Ji = AO\{Ji for all i = 1, ... , n. Since the \{Ji were arbitrary,
this proves that A = Ao E 9:n". •

As it stands, Proposition 1.7.1 is not valid when 9:n n (C) is replaced by l13(H),
where dim(H) = 00. To describe the appropriate refinement, we define two locally
convex topologies on l13(H) that are weaker than the norm topology we have been
using so far.

°
The seminorms plJ!(A) := IIA \{J II define the strong topology on l13(H), so that
A). -+ A strongly when II(A). - A)\{J II -+ for all \{J E H. In the proof of 1.7.2
 I The Structure of Algebras of Observables 59

we will use the fact that a neighborhood basis of A is given by all sets of the form
{B E ~(1t) I II(A - B)\Ildl < E for all i = 1, ... , n}, where E > 0, n EN, and
\III, ... , \Il n E 1t.
The weak topology on ~(1t) is defined by the seminorms p"'.<I>(A) :=
I(\II, A<l»I, so that A" ~ A weakly when I(\II, (A" - A)\II)I ~ 0 for all \II E 1t.
The norm topology is stronger than the strong topology, which in tum is stronger
than the weak topology.
Theorem 1.7.2. Let rot be a *-algebra in SB(1t) containing n. Thefollowing are
equivalent:
1. rot" = rot.
2. rot is closed in the weak operator topology.
3. rot is closed in the strong operator topology.
It is easily verified from the definition of weak convergence that the commutant
!J1' of a *-algebra !J1 is always weakly closed. If rot" = rot, then rot = S)1' for
!J1 = rot', so that rot is weakly closed. Hence 1 => 2. Since the weak topology is
weaker than the strong topology, 2 => 3 is trivial.
To prove 3 => 1, we adapt the proof of 1.7.1 to the infinite-dimensional situation.
Instead of rot\II , which may not be closed, we consider its closure rot\II , so that p =
[rot\Il]. Hence A E rot" implies A E rot\Il; in other words, for every E > 0 there is
an A€ E rot such that II (A - A€)\Il1l < E. For1t n this means that II«S(A - Af)QII2 <
E2. The left-hand side of this inequality equals the sum E7=1 II(A - Af)\IldI 2 , so
that II(A - Af)\Ili II < E for all i = 1, ... , n. It follows that Af ~ A strongly for
E ~ O. Since all A€ E rot and rot is strongly closed, this implies that A E rot, so
that rot" £ rot. Since trivially rot £ rot", this proves 3 => 1. •
This theorem is remarkable, for it relates a topological condition (rot being
closed in certain topologies) to an algebraic one (rot being its own bicommutant).
A similar but simpler example of such a theorem states that a linear subspace lC of
a Hilbert space is closed iff lC = lC..l.J. (where lC1. is the orthogonal complement
of lC).
Definition 1.7.3. A *-algebra rot (containing the unit operator) ofbounded oper-
ators on some Hilbert space is called a von Neumann algebra ifit satisfies one
(hence all) of the conditions in 1.7.2.
We know from 1.6.5 that SB(1t) = ~I (1t)*; the pertinent w*-topology on SB(1t)
is often called the a-weak topology. This topology is generated by the seminorms
pp(A) := ITr pAl, and is clearly stronger than the weak topology (but weaker than
the norm topology). Hence a von Neumann algebra rot ~ ~(1t) is closed in the
(relative) a-weak topology.
Moreover, a von Neumann algebra rot is closed in the norm topology (defined
by the norm (1.18» as well, so that it is a C* -algebra. A state on rot £ ~(1t) of the
form (1.63) for a density matrix p (cf. 1.6.6) is called normal. The linear span of
all normal states in rot* is called the predual rot* of rot. For example, the predual
of SB(1t) is ~ I(1t), and more generally one has rot = rot: as a Banach space. The
60 I. Observables and Pure States

set N(9J1) := S(9J1) n 9J1* of all normal states on 9J1 is called the normal state
space of 9J1.
All von Neumann algebras in this book are of the form 9J1 = JT('2t)", where JT
is a representation of some C* -algebra '2t. In particular, one may take JT = JTu; cf.
1.5.7.
Proposition 1.7.4. The bidual '2t** ofa C* -algebra '2t is isomorphic (as a Banach
space) to JTu('2t)". Through this isomorphism, '2t** acquires the structure of a von
Neumann algebra (and therefore of a C* -algebra).
The proof is a highly nontrivial generalization of the proof of 1.6.5. The equality
!Boo-i)* = !B I (H) is now replaced by the fact that '2t* is the linear span of all func-
tionals of the form A ~ (\II, JTu('2t)<l», where \II, <l> E Hu. This characterization is
then used to show that '2t* is the predual of JTu ('2t)", so that '2t** = JT u('2t)". 0
In the context of Theorem 1.4.5, we note that when K is a Hausdorff compact
convex set, the bidual of A(K, JR) (with sup-norm) is the space Ab(K, JR) of all
bounded real-valued affine functions on K. Hence for a C* -algebra '2t one has
'2t~* ::: JTu('2t)~ ::: A b (S('2t), JR). The predual of '2t** is obviously '2t;* = '2t*, and
the normal state space is N('2t**) = S('2t). More generally, for any von Neumann
algebra 9J1 one has 9J11R ::: Ab(N(9J1), JR) as partially ordered Banach spaces.
This isomorphism maps the a -weak topology on 9J11R to the topology of pointwise
convergence on A b (N(9J1), JR).
The center of a von Neumann algebra 9J1 is 9J1n9J1'; this is the set of all elements
of 9J1 that commute with every element in the algebra. The following proposition
allows one to regard JT('2t)" as a von Neumann subalgebra of'2t**.
Proposition 1.7.5. If JT is a cyclic representation of a C* -algebra '2t, there exists
a projection p in the center ofJT u ('2t)" such that JT('2t)" is isomorphic (as a von
Neumann algebra) to PJT u ('2t)".
The idea of the proof is that the morphism JT 0 JT u- 1 from JT u('2t) to JT('2t) is
a-weakly continuous, so that it can be extended to a morphism from JT u ('2t)" to
JT ('2t)". The kernel of this extension is a a -weakly closed ideal in JTu('2t)". It can be
shown that a a-weakly closed ideal in a von Neumann algebra 9J1 is of the form
q9J1, where q is a projection in the center of 9J1. Applying this to the case at hand
yields 1.7.5, with p = [- q. 0

2 The Structure of Pure State Spaces

2.1 Pure States and Compact Convex Sets
In this section we look at a subspace of the state space on a C* -algebra, which may
be interpreted as a quantum analogue of the phase space of a classical system.
Let us return to 1.4. One observes that the compact convex sets one naturally
has in mind have a boundary; this particularly applies to the state spaces of the
 2 The Structure of Pure State Spaces 61

C* -algebras C, C ED C, and 9J1 2 (C). The intrinsic definition of this boundary is as

follows.
Definition 2.1.1. An extreme point in a convex set K is a member W of K that
can be decomposed as W = AWl + (1 - A)w2, where A E (0, 1), iff WI = W2 = w.
The collection 8e K of extreme points in K is called the extreme boundary of K .
An extreme point in the state space K = s(m) of a C* -algebra mis called a
pure state. A state that is not pure is called mixed. We write p(m), or simply p,
for 8e S(m), referred to as the pure state space ofm.
Thus the single state on C is pure, the pure states on C ED C are the points 0 and
1 in [0, 1], and the pure states on 9J12 (C) are the matrices p in (1.50) for which
x 2 + y2 + Z2 = 1. These are the projections onto one-dimensional subspaces of
C2, and we see that P(9J12(C» may be identified with the unit sphere in JR3. More
generally, one has
Proposition 2.1.2. The pure state space of S}3 0 (H) consists ofall one-dimensional
projections, so that any pure state on !.Bo(H) is a vector state ( J.46) in H.
This is immediate from 1.6.6, the spectral theorem applied to a density matrix,
and 2.1.1. •
A useful reformulation of the notion of a pure state is as follows.
Proposition 2.1.3. A state is pure iff 0 S P S W for a positive functional p
implies p = twfor some t E JR+.
We assume that mis unital; if not, use 1.2.1 and (1.49). For p = 0 or p = W
the claim is obvious. When w is pure and 0 S p S w, with 0 i- p i- w, then
o < p(lI) < I, since w - p is positive; hence IIw - p II = w(lI) - p(lI) = 1 - p(II).
Hence p(lI) would imply w = p, whereas p(lI) = 0 implies p = 0, contrary to
assumption. Hence WI := (w - p)/(1 - p(II» and W2 := pi p(lI) are states, and

.
w = AWl + (1 - A)Wz with A = 1 - p(II). Since w is pure, by 2.1.1 we have
p = p(lI)w.
Conversely, if w is decomposed as in 2.1.1, then 0 S AWl S W, so that AWl =
tw by assumption; normalization gives t = A, hence WI = W = W2, and w is
~.

Here is another example of a pure state space.

Proposition 2.1.4. The pure state space of the commutative C* -algebra Co(X)
(equipped with the relative w* -topology) is homeomorphic to X.
The case that X is not compact may be reduced to the compact case by passing
from m= Co(X) to mx = C(X) (where X is the one-point compactification of
X); cf. (2.2) below. In view of the proof of Theorem 1.2.3, we then merely need to
prove that any pure state on C(X) is multiplicative, and vice versa; P(C(X» and
~(C(X» are both equipped with the relative w* -topology.
Let Wx E ~(C(X» (cf. the proof of 1.2.3), and suppose a functional p satisfies
o S p S W x ' Then ker(wx ) S; ker(p), and ker(wx ) is a maximal ideal, so that
62 I. Observables and Pure States

ker(wx ) = ker(p). Since two functionals on any vector space are proportional
when they have the same kernel, it follows from 2.1.3 that Wx is pure.
Conversely, let W be a pure state, and pick agE C(X) with 0 :::: g :::: Ix. Define
a functional Wg on C(X) by wg(f) := w(fg). Since w(f)- wg(f) = w(f(I - g»,
and 0:::: I - g :::: lx, one has 0:::: Wg :::: w. Hence Wg = tw for some t E JR.+ by
2.1.3. Putting f = Ix yields t = w(g). Since any function is a linear combination
of functions g for which 0 :::: g :::: I x, it follows that W is multiplicative. •
It could be that a given convex set contains no extreme points at all; think of
an open convex cone. When K is compact, this possibility is excluded by a basic
theorem in functional analysis, which we state without proof. The convex hull
co( V) of a subset V of a vector space is defined by
co(V) := {AV + (1 - A)W I v, W E V, A E [0, I]). (2.1)
Theorem 2.1.5. A compact convex set K embedded in a locally convex vector
space is the closure of the convex hull of its extreme points. In other words, K =
co(aeK).

Although the state space of a C*-algebra ~ without unit (such as 23 o{1i) or

Co(X» is not compact, Theorem 2.1.5 may nonetheless be used. For the pure
state space of ~ may be described in terms of the pure state space P(~ll) of its
unitization ~ll; cf. 1.2.1. Define a functional Woo by woo(A + Ali) = A for all A;
this is easily seen to be a pure state on ~ll. Taking (1.49) into account, one obtains
a homeomorphism
(2.2)
The extreme boundary ae K of a compact convex set is not necessarily closed, so
that the pure state space P(~) of a unital C* -algebra ~, while always a Hausdorff
space, is not generally compact. Nonetheless, it is interesting to realize ~IR as a
subspace of C(~IR(P), JR.), somewhat in the spirit of 1.4.5. To do so, we replace
~(~) in the definition (1.30) of the Gelfand transform by the pure state space of
an arbitrary C* -algebra; cf. 2.1.4.
Definition 2.1.6. Let 2l.IR be the self-adjoint part of a C* -algebra ~. The Gelfand
transform of A E 2l.IR is the function A : P(~) -+ JR. defined by (1.30). The
subspace {A I A E ~Ild of£OO(P(~), JR.) is denoted by IR . m
The extension of the Gelfand transform from ~IR to ~ is useful only for com-
mutative C* -algebras; in the noncommutative case the first claim below would not
hold if ~IR were replaced by ~.
Theorem 2.1.7. The Gelfand transform is an isomorphism between ~IR and mlR ~
C(P(~), JR.), seen as partially ordered Banach spaces (here the order in ~IR is
defined by 1.3.3 and 1.3.1, whereas the order in C(P(~), JR.) is defined by the cone
ofpointwise positive functions).
The equality mlR = C(P(~), JR.) occurs iff ~ is commutative and unital, in which
case P(~) is closed.
 2 The Structure of Pure State Spaces 63

The first claim follows from 1.4.5 and 2.1.5: Any A E A(K, R) is determined by
a
its values on e K (for it is affine and continuous). The inclusion!it1R 5; C(P(~), R)
is immediate from the definition of the relative w* -topology. The claim about the
order is a trivial consequence of the pertinent definition, too. The last claim follows
from 1.2.3 and 2.1.4. •
An alternative proof may be obtained from the following sharpening of
Proposition 1.4.3.
Proposition 2.1.8. For every A E ~IR and a E a(A) there is a pure state Wa on
~for which wa(A) = a. There exists a pure state w such that Iw(A)1 = IIAII.

We extend the state in the proof of 1.4.3 to C*(A) by multiplicativity and conti-
nuity, that is, we put eVa (An) = an, etc. It follows from 2.1.4 that this extension is
pure. One easily checks that the set of all extensions of eVa to!.21 (which extensions
we know to be states; see the proof of 1.4.3) is a closed convex subset Ka of S(~);
hence it is a compact convex set. By Theorem 2.1.5, Ka has at least one extreme
point Wa. If Wa were not an extreme point in S(~), it would be decomposable as
in 2.1.1. But in that case WI andw2 would both coincide on C*(A) with eVa, so that
Wa cannot be an extreme point of Ka. •
In any case, when !.21 is noncommutative one would like to characterize !it1R in
C(P(!.21), R). This will be done in Theorem 3.2.1.

2.2 Pure States and Irreducible Representations

In this section we start our analysis of irreducible representations of C* -algebras
and their connection to pure states.
Definition 2.2.1. A representation Jr of a C* -algebra !.21 on a Hilbert space 1t is
called irreducible if a closed subspace of1t that is stable under Jr(!.21) is either 1t
orO.
This definition should be familiar from the theory of group representations. The
defining representations of!mn (C), ll3o(1t), and s:B(1t) are evidently irreducible.
Proposition 2.2.2. Each of the following conditions is equivalent to the
irreducibility of Jr (!.21):
1. Jr(!.21)' = C][, or, equivalently, Jr(!.21)" = s:B(1t) (Schur's lemma).
2. Every nonzero vector Q in 1t is cyclic for Jr(!.21) (i.e., Jr(!.21)Q is dense in 1tfor
all Q =f=. 0).
The commutant Jr(!.21)' is a *-algebra in s:B(1t), so when it is nontrivial it must
contain a self-adjoint element A that is not a multiple of][. It follows from Theorem
1.7.2 and the spectral theorem that the projections in the spectral resolution of A lie
in Jr (!.21)' if A does. Hence when Jr (!.21)' is nontrivial it contains a nontrivial projection
p. But then p1t is stable under Jr(!.21), contradicting irreducibility. Hence 2.2.1 ~
2.2.2.1.
64 I. Observables and Pure States

Conversely, when ;rr(Qt)' = ClI and ;rr is reducible, one finds a contradiction
because the projection onto the alleged nontrivial stable subspace ofH commutes
with ;rr(Qt).
When there exists a vector \}! E H for which ;rr(Qt)\}! is not dense in H, we can
form the projection onto the closure of Jr(Qt)\}!. By Lemma 1.5.3, with 9J1 = ;rr(Qt),
this projection lies in Jr(Qt)', so that by Schur's lemma Jr cannot be irreducible.
Hence 2.2.1 =::} 2.2.2.2. The converse is trivial. •
The connection between representations and states (see 1.5) can be refined when
a state is pure.

Theorem 2.2.3. The GNS-representation Jr",(Qt) ofa state W E S(Qt) is irreducible

iff W is pure.
When W is pure yet Jr",(Qt) reducible, there is a nontrivial projection p E Jr",(Qt)'
by Schur's lemma. Let Q", be the cyclic vector for Jr",. If pQ", = 0, then ApQ", =
pAQ", = 0 for all A E Qt, so that p = 0, since Jr", is cyclic. Similarly, p-1Q", = 0
is impossible. We may then decompose W = A1/1 + (1 - A)1/I-1, where 1/1 and 1/1-1
are states defined as in (1.46), with \}! := pQ",1 II pQ",1I , \}!-1 := p-1Q",1 II p-1Q", II,
and A= IIp-1Q,,, 112. Hence W cannot be pure, so that Jr", is irreducible by reductio
ad absurdum.
In the opposite direction, suppose Jr", is irreducible, yet W decomposable as in
2.1.1. Then AWl - W = (1 - A)w2, which is positive; hence AWI(A* A) :::: w(A* A)
for all A E Qt. By (l.48) this yields IAwl (A * B)1 2 :::: w(A *A)w(B* B) for all A, B.
This makes the quadratic form Q on Jr",(Qt)Q", by Q(Jr",(A)Q"" Jr",(B)Q",) :=
AW I (A * B) well-defined. Furthermore, Qis bounded with norm I, so that Qcan
be extended to 'H", by continuity. Since w is a Hermitian functional, one has
Q(<I>, \}!) = Q(\}!, <1».
Thus there exists a self-adjoint operator Q on H", such that Q(\}!, <1» = (\}!, Q<1»
for all \}!, <I> E H. In other words, one has (Jr",(A)Q"" QJr",(B)Q",) = AWl (A* B).
Since Jr", is a representation, one computes that [Q, Jr",(C)] = 0 for all C E Qt, so
that Q E Jr",(Qt)'. Since Jr", is irreducible, one must have Q = tlI for some t E lR..
Hence WI is proportional to w, and therefore equal to w by normalization, so that
w is pure. •
Here are some easy consequences of this result, culminating in 2.2.6.
Proposition 2.2.4. If (Jr(Qt), H) is irreducible, then the GNS-representation
(Jr",(Qt), H",) defined by any vector state w (corresponding to a unit vector Q E 'H)
is equivalent to (Jr(Qt), H). In particular, any vector state in an irreducible
representation is pure.
Immediate from 2.2.2.2, 1.5.5, and 2.2.3.
Corollary 2.2.5. Every irreducible representation of a C* -algebra comes from a
•
pure state via the GNS-construction.
Combine 2.2.4 and 2.2.3.
•
 2 The Structure of Pure State Spaces 65

This leads straight to a basic result in the theory of C* -algebras:

Theorem 2.2.6. The C* -algebra l.l3 o(1t) ofall compact operators on some Hilbert
space possesses only one irreducible representation, up to equivalence, namely the
defining one.

This is immediate from 2.1.2,2.2.5, and 2.2.4.

•
Two pure states p and a on a C* -algebra 2t are said to be equivalent if the
associated GNS-representations Ji p and Ji(J are equivalent; we write p '" a. It
is easily verified from the definition of this notion of equivalence that '" is an
equivalence relation in P(2t). It follows from 2.2.4 that all vector states in rip are
equivalent to p. Conversely, any state a '" p is given by a vector state in rip,
for if U : H(J --+ Hp intertwines Ji(J and Ji p, then the vector state defined by
UQ(J E Hp coincides with a. Since the intertwiner is unique by Schur's lemma,
one thus obtains a bijection between the equivalence class [p] of a given pure state
p and the set of vector states in H p •
The topological aspects of this bijection will be clarified in 2.5. For now, we
are led to a manageable refinement of the the universal representation (cf. 1.5.7),
which is still faithful.

Definition 2.2.7. The reduced atomic representation Jim ofa C* -algebra 2t is the
direct sum over irreducible representations Jira = EBpE(P(QI)]Jip (on the Hilbert space
Hra = EBpE[p('<l)]H p ), where one includes one representative of each equivalence
class in P(2t).

The specific choice of pure states in each equivalence class affects the reduced
atomic representation only within (unitary) equivalence. Replacing the use of 1.4.3
in the proof of Theorem 1.1.8 by 2.1.8, one infers that Jira is indeed faithful. If p
and a are inequivalent pure states, Schur's lemma implies that

Jira(2t)" = EB"'E[P('l)]I.l3(H",). (2.3)

If 2t is commutative, so that 2tR ~ C(P(2t), JR) (see 2.1.7), one easily infers that
e
Ji ra (2t)" = OO (p(2t». On the noncommutative side, we infer

Proposition 2.2.8. Every finite-dimensional C* -algebra is a direct sum ofmatrix

algebras.

Since 2t is finite-dimensional and 2t ~ Jira(2t), the right-hand side must be finite-

dimensional. Hence by 1.7.1 and (2.3) one has 2t ~ EB"'E(p(QI)]l.B(H",), where each
H", is finite-dimensional and the sum is finite. •

2.3 Poisson Manifolds

We return to Poisson algebras (cf. 1.1.2). The main source of such algebras is the
following.
66 I. Observables and Pure States

Definition 2.3.1. A Poisson manifold is a manifold P equipped with a bilinear

operation {, } : C'~\P, JR) x Ceo(P, JR) ~ eeo(p, JR) with the property that
°
(Ceo(P, JR), 0, {, }), where is pointwise multiplication, is a Poisson algebra.

By definition, the map g ~ {f, g} (for fixed f E Ceo(P, JR» is a derivation on

COO(P, JR), and this implies that the Poisson bracket (f, g}(a) depends only on the
differentials df and dg at a E P. Therefore, there exists a smooth anti symmetric
tensor field B E r(J\2(P» such that

{f, g} = B(df, dg). (2.4)

The Jacobi identity implies that the Poisson tensor B must satisfy

tBdtBa =0 Va E J\3(p). (2.5)

(Recall that the insertion t of A E J\n(P) into fJ E J\n+m(p) produces an element

tAfJ E J\rn(p) defined by (tAfJ)(C) = fJ(A J\ C) for all C E !\m(P).) If P is
finite-dimensional, this can be conveniently stated in terms of local coordinates
faa}: If Bab(a) = Bry(da a J\ dab), so that Bba = _Bab, then

aB k aB~ aB ro
B ea _ _ + B ec _ _ + B eb _ _ = O. (2.6)
aa e aa e aa e
Conversely, an element B E r(J\2(P» satisfying (2.5) (or 2.6» defines a Poisson
bracket by (2.4).
The Poisson tensor B defines a linear map B~ : T* P ~ T P by

(B~(a»(fJ) := B(a, fJ), (2.7)

where a and fJ lie in the same fiber in T* P. If h E Ceo(P, JR), the image B~(dh)
is usually written as ~h, and called the Hamiltonian vector field of h. Hence

(2.8)

By virtue of the Jacobi identity, one has

(2.9)

and
(2.10)

for all j, g E Ceo(P, JR), where L is the Lie derivative. Hence j ~ ~f is a

homomorphism from eeo(p, JR) into the subspace of reT P) (regarded as a Lie
algebra under the commutator) of vector fields preserving the Poisson structure.
If c : I ~ P (where I ~ JR is some interval containing 0) is a curve in P
for which e(O) = a, we write a(t) for e(l). Given h E Ceo(P, JR), Hamilton's
equations of motion for such a curve are
da(t)
- - = ~h(a(t». (2.11)
dt
 2 The Structure of Pure State Spaces 67

A curve satisfying this equation for some h is called a Hamiltonian curve. The
corresponding flow, given by
Ft(a) = a(t), (2.12)
is called the Hamiltonian flow of h. A trivial consequence of (2.11) and (2.8) is
Proposition 2.3.2. A function h E COO(P, JR) is constant along the flow
trajectories it generates.
The theory of ordinary differential equations (Picard iterations) guarantees ex-
istence and uniqueness of a local solution for each initial value c(O) E P and t
in some compact interval around O. When the motion exists, one has the property
Fs 0 Ft = Fs+t . Given h and c(O), it may happen that the motion is not defined for
all t E JR, in which case the vector field ~h is called incomplete. If ~h has compact
support, it is always complete.
Given h E COO(P, JR) with Hamiltonian flow a(t), one constructs a
one-parameter family of linear maps a~ : COO(P, JR) -4 COO(P, JR) by
a~(f)(a) := f(a(t)). (2.13)
This family is evidently defined only for those t for which the solution of (2.11) is
defined for any initial value. One infers from (2.8) and (2.11) that the infinitesimal
version of (2.13) is

da;~f) = {h, a~(f)}; (2.14)

here the derivative is understood pointwise. The following result is a local version
of the "infinitesimal" fact (2.10).
Proposition 2.3.3. Ifa~(f) satisfies (2.14), then a? is a morphism (cf 1.1.3) of
COO(P, JR)for each t for which it is defined.
The Leibniz rule and (2.14) imply d[at(fg)]/dt = d [at (f)at (g)]/dt; the
proposition follows by integrating this relation. •
If the motion pertinent to h is defined for all t E JR, one obtains a one-parameter
group of automorphisms in this way. Equation (2.14) evidently makes sense in any
Poisson algebra.
Definition 2.3.4. An element h of a Poisson algebra is called complete if the
one-parameter family ofautomorphisms defined by (2.14) is defined for all t E JR.
For Poisson algebras of the type COO(P, JR) this amounts to saying that the flow
of ~h is complete.
We will frequently need the notion of a Poisson map J : (PI, B I) -4 (P2, B2);
this is a smooth map such that, in obvious notation,
1*{f, gh = {1* f, 1*gh (2.15)
for all f, g E C (P2, JR). Equivalently,
OO

B;(J*a, 1*/3) = BJ(O')(a, /3) (2.16)

68 I. Observables and Pure States

for all a E PI and all a, fJ E TJ(tY)P2 ,

It follows from the Jacobi identity that the flow Ft : P -+ P of each Hamiltonian
vector field ~f is a Poisson map (for all t for which the flow is defined). Moreover,
a chasing of the definitions shows the validity of
Proposition 2.3.5. For any Poisson map J : PI -+ P2 one has

(2.17)
for all f E C"'(P2 , JR). Moreover, the image of the flow of~J* f under J is the flow
of~f'
In the present setting, symplectic spaces are regarded as special instances of
Poisson manifolds.
Definition 2.3.6. A Poisson manifold for which the map BO is an isomorphism is
called symplectic. If B-;. : T P -+ T* P is the inverse of BO, the symplectic form
WE r(/\2(p» is defined by

W(X, Y) := (B~(X»(Y). (2.18)

As a consequence of the Jacobi identity (or (2.5», W is closed (dw = 0). In
terms of the symplectic form, the Poisson bracket reads
{f, g} = -w(~f' ~g), (2.19)
where ~J and ~g are defined as in (2.8), that is,
~f = BU-I(df), (2.20)
and this is equivalent to the connection

i~fw = df· (2.21)

The following characterization of symplectic manifolds follows directly from
the definition and the local existence of Hamiltonian flows.
Proposition 2.3.7. A Poisson manifold is symplectic iff one of the following
equivalent properties is satisfied:
• The collection of Hamiltonian vector fields {~f' f E C"'(P, JR)}, or,
equivalently, the image of B~, spans TtY P at each a E P .
• Any two points of P can be connected by a piecewise smooth Hamiltonian
curve.
When P is finite-dimensional, the first condition simply states that at every point
the rank of BO (that is, the dimension of the image of B~ at a given point) equals
the dimension of P.
The cotangent bundle T* Q of any manifold Q is symplectic.
Definition 2.3.8. The canonical symplectic form w on a cotangent bundle T* Q
is given by w = -de, where e is a one-form on T* Q defined by
(2.22)
 2 The Structure of Pure State Spaces 69

where t := tT'Q-,>Q.

In canonical coordinates (p, q) on T* Q this reads

UJ = dqi /\dpi' (2.23)

and the associated Poisson bracket is given by

{j,g}:= af ~ _ af}.~. (2.24)

api aq' aq' api
A diffeomorphism J : Sl -+ S2 between two symplectic manifolds that is a
Poisson map is called a symplectomorphism; Sl and S2 are symplectomorphic
when such a map exists.

2.4 The Symplectic Decomposition of a Poisson Manifold

In this section we argue that an arbitrary finite-dimensional Poisson manifold is
foliated by symplectic subspaces; this is somewhat analogous to the decomposition
of a finite-dimensional C* -algebra as a direct sum of matrix algebras; cf. 2.2.8. In
preparation, we recall some differential geometry.

Definition 2.4.1. A distribution D on a manifold P is a subset of the tangent

bundle T P such that Da := D n Ta P is a vector spacefor each a E P. The rank
of D at a is the dimension of Da.
A distribution is called smooth iffor every a E P and vEDa there is a smooth
vector field ~, defined on a neighborhood N of a, such that ~ (p) E D p for all
pEN, and ~(a) = v. Such a ~ is called a local section of D.
A distribution is called involutive iffor any pair ~ I , ~2 of local sections one has
[~l' ~2](P) E D p in their common domain of definition.
A distribution D on P is completely integrable when each point a E P lies in
an immersed submanifold Sa ~ P whose tangent space at a is Dao

(One sometimes speaks of a generalized distribution when the rank of D is not

constant on P; we will, instead, speak of a regular distribution when the rank is
constant.)
Hence a completely integrable distribution defines a foliation of P, whose leaves
are the Sa. The leaves of a completely integrable foliation may have varying di-
mension. (Such a foliation is sometimes called singular; again, we will rather use
the adjective regular when the leaf dimension is constant.)
For smooth regular distributions the question of complete integrability is settled
by the well-known Frobenius theorem, which states that D is completely inte-
grable iff it is involutive. In general, one needs a stronger condition (the "singular
Frobenius theorem") to arrive at completely integrability, which we state without
proof.

Lemma 2.4.2. A smooth distribution D is integrable iff at each a E P one can

choose local sections ~ I, ... , ~rank(D") that span Da with the property that for an
70 I. Observables and Pure States

arbitrary local section ~ of D (defined around a) one has

[~, ~j](a(t)) = A{ (t)~j(a(t)) (2.25)

for small enough t. Here a ~ a(t) is the (local)flow generated by ~, and the A{
are certain functions of t.

We return to Poisson manifolds. In general, the map B~ may fail to be surjective.

The image of B~ defines a distribution D on T P , which is easily seen to be smooth;
for it is generated by the Hamiltonian vector fields, each of which is a smooth
section of the tangent bundle. The rank of B~ is not necessarily constant, so that
D may not be regular.

Definition 2.4.3. A symplectic leaf in a Poisson manifold (P, B) is a maximal

set ofpoints that are equivalent under the following equivalence relation: p "-' a
iff p and a can be connected by a piecewise smooth Hamiltonian curve.
The terminology will be justified shortly. This equivalence relation leads to a
decomposition P = Ua Sa, where each Sa is a symplectic leaf.

Lemma 2.4.4. The rank of B~ is constant on each symplectic leaf.

This is simply because the flow of each Hamiltonian vector field ~f is a Poisson
map, and such maps leave B (and therefore BU) invariant, cf. (2.10) and (2.16).
In particular, the pushforward of a Hamiltonian flow Ft maps the image of B~ at
some a into its image at Ft(a). •

Using 2.4.4, 2.4.2 (with ~j = ~fj for suitable fi), and (2.9), one infers that D
is completely integrable, and it will become clear shortly that the leaves of the
foliation defined by D are just the symplectic leaves of S.
In general, a given symplectic leaf Sa C P may not be a submanifold of P.
Nonetheless, one may tum Sa into a manifold by a standard procedure of (singular)
foliation theory. In the present context, this is accomplished by defining a chart
around a given a E Sa in the following way. Let the rank of BU at a be n, and
choose functions iI, ... , fn such that {~Ji}j=I ..... n spans the image of BU at a.
There is an E > 0 and an E-ball OE C Rn around 0 such that F : OE --+ P, defined
by

F(t), ... , tn) = F1: 0 ... 0 F,:(a),

where F/ denotes the flow of ~fj' is a bijection.

Lemma 2.4.5. Applying the above procedure for a sufficient number of points
a E Sa leads to an atlas on Sa that is well-defined and independent, up to smooth
equivalence, of the choice of the fi at each point. The dimension of Sa with this
manifold structure is the rank of BU.

The pushforward of each F/ , and therefore of F,: 0 ••• 0 F,:, maps the image of
Btt at a into its image at F/ (a). •
 2 The Structure of Pure State Spaces 71

Equipped with this manifold structure and topology, each Sa is an injectively

immersed submanifold of P; that is, the inclusion La : Sa '-+ P is continuous and
of constant rank, equal to the dimension of Sa, at each point.
The singular Frobenius theorem 2.4.2 applies; indeed, the leaf of the pertinent
foliation is locally given by F( Of)'
Lemma2.4.6. Iff E COO(P, JR) vanishes on Sa, then {f, g}(a) = Ofor all a E Sa
and all g E COO(P, JR). Therefore, one can define a Poisson bracket {, }a on Sa by

{t~f, t~g}a := t~{f, g}. (2.26)

Each Sa is a symplectic manifold, and each inclusion La is a Poisson map.

If f = 0 on Sa, then {f, g} = -l;g/ = 0, since l;g is tangent to Sa. •
Thus we arrive at
Theorem 2.4.7. For each finite-dimensional Poisson manifold P there exists a
family {Sa} of symplectic manifolds, and injective Poisson immersions ta : Sa '-+
P, such that P = UaLa(Sa) (disjoint union). Each subset ta(Sa) is a symplectic
leafof P as defined in 2.4.3. The value ofthe Poisson bracket {f, g} at some a E P
depends only on the restrictions of f and g to the symplectic leaf through a.
In the text preceding the theorem we have made no notational distinction be-
tween Sa and ta (Sa)' Indeed, if each Sa is a submanifold, one can simply say that
P = Ua Sa as manifolds.

2.5 (Projective) Hilbert Spaces as Symplectic Manifolds

In this section we look at the geometric structure of P(Q3 oCJt».

Definition 2.5.1. The projective space lP1t of a Hilbert space 11. is the space of
one-dimensional complex linear subspaces of1t. Equivalently, lP1t is the quotient
§1t / U (1) of the unit sphere

§1t:= {\II E 11. I (\II, \II) = I} (2.27)

by the action ofU(l) ~ T, given by z : \II t--+ z\ll, where Izl = 1.

The identification of vector states in 'N, one-dimensional projections on 11., and

points of lP1t is immediately clear from this realization. Hence we conclude from
2.1.2 that lP1t ~ P(1J3 0 (1t» (as collections of linear functionals on 113 0 (11.) for
the moment), and lP1t S; P(Q3(1t»; when 11. is infinite-dimensionallP1t does not
nearly exhaust P(Q3(1t».
The space lP1t can be topologized by restricting the usual (norm) Hilbert space
topology on 11. to §1t, and quotienting it to lP1t ~ §1t/ U(1). We will denote the
image of \II E §1t in lP1t under the canonical projection T : §1t -+ lP1t by 1/1;
conversely, given 1/1 E lP1t, such a \II E §'N will stand for an arbitrary preimage
of 1/1 (and similarly for rp, <1>, etc.).
72 I. Observables and Pure States

We now give IP1i the structure of a real manifold. For 1/1 E IP1i and \lI E §1-{,
define a neighborhood N", := {cp E IP1i I (\lI, <1» -=1= OJ; this is indeed an open set
in the quotient topology. Then N", is mapped into \lI..L C 1-{ by
<I> [\lI..L ] <I>
F",(cp) = (\lI, <1» - \lI = (\lI, <1» (2.28)

(which depends only on the lift <1», where [1lJ..L] is the projection onto \lI..L C 1-{.
Clearly, 1/1 is mapped into the null vector, and the image of this map is open in \lI..L.
It is easily checked that this map is a homeomorphism between N", and its image.
We now let 1/1 (more precisely, 1lJ) vary over a basis in 1-{, and for each such 1/1 we
construct an (arbitrary) reference isomorphism between \lI..L and a fixed reference
Hilbert space 1-{' with two (real) dimensions less than 1-{. This leads to a collection
of charts, making IP1i a Hilbert manifold, modeled on 1-{' (equipped with the strong
topology). We will refer to the topology on IP1i considered so far as its manifold
topology.
Proposition 2.5.2. Thefollowing topologies on 1P'1-{ coincide:
1. The manifold topology.
2. The w* -topology relative to IP1i C lJ3 o(1-{)*.
3. The w* -topology relative to IP1i C 1J3(1-{)*.
It is quite trivial to verify that the topology on IP1i that is inherited from the
strong topology on 1-{ is stronger than the topology in 2.5.2.3, which in turn is
stronger than the one of2.5.2.2. Using the fact that lJ3 o(1-{) is generated by the one-
dimensional projections on 1-{, one verifies that the topology in 2.5.2.2 coincides
with the one induced by the weak topology on 1-{. Since the strong and the weak
Hilbert space topologies coincide on §1-{, the equivalence between 2.5.2.2 and
2.5.2.3 follows.
It follows from (2.28) that for arbitrary cP EN"" one has
cp(A)
cp([IlJ]) = (F",(cp), AF",(cp»+(IlJ, AF",(cp» + (F",(cp), A\lI) + (1lJ, AIlJ). (2.29)

It is clear from this equation that F",(CPn) ~ F",(cp) strongly implies CPn(A) ~
cp(A), so that CPn ~ cp in the topology of 2.5.2.3. Hence the manifold topology
on IP1i is stronger than the topology of 2.5.2.3. Conversely, if CPn(A) ~ cp(A),
then each term on the right-hand side of (2.29) must converge, so that F",(CPn) ~
F",(cp) weakly and (F",(CPn), AF",(CPn» ~ (F",(CPn), AF",(CPn»' Taking A = II,
these conditions imply F",(CPn) ~ F",(cp) strongly, so that the topology of 2.5.2.3
is stronger than the manifold topology. Hence the topologies in 2.5.2.1 and 2.5.2.3
coincide. •
Corollary 2.5.3. The pure state space ofthe C* -algebra lJ3 o(1-{) (with relative w*-
topology) is homeomorphic to the projective space IP1i (with manifold topology).
Theorem 2.5.4. The pure state space p(m) ofa C* -algebra mis a disjoint union
p(m) = UalP1ia, where 1-{a is isomorphic to the irreducible GNS-representation
 2 The Structure of Pure State Spaces 73

space of an arbitrary state in IntO'. All states in a given subspace IntO' are equiv-
alent, and any two states lying in different such subspaces are inequivalent. The
inclusion map of any IntO' (equipped with the manifold topology) into P(~) (with
the w* -topology) is continuous.
The set-theoretic part of this claim follows from the comments after the proof
of 2.2.6. The topological part is a consequence of (1.57) and the equalities of the
topologies in 2.5.2.1 and 2.5.2.3. •
Of course, the disjoint union in 2.5.4 is meant in a set-theoretic rather than a
topological sense (the IntO' are not necessarily components of P(~».
We now embark on a description of Int as a symplectic manifold, starting with
the corresponding analysis of 'H. Regarding 'H as a real vector space, we identify
the tangent bundle T'H with 'H x 'H in the usual way: For any \II E 'H, an element
<I> E 'H defines a tangent vector V (<I» E TIjt 'H by
df
V(<I»ljtf = d"t(\II + t<l»II=O. (2.30)

If V(<I» is tangent to §'H, the derivative r* will project it to an element v(<I» of

TInt. This applies to tangent vectors of the form V(iA\II), where A* = A in
~('H), and \II is arbitrary. We observe that for any \II E 'H the collection of vectors
{iA\II1 A E ~('H)Ild, while not being equal to 'H because of the restriction to
~('H)IR, contains \11.1. It then follows from the above discussion of the manifold
structure of Int that for all 1/1 E Int one has
T",Int = {v(iA\II) I A E ~('H)Ild. (2.31)
We now show that 'H and Int are both examples of (real) symplectic manifolds
(the real structure depends on the choice of a basis). Further to the identification
T'H c::::: 'H x 'H (see (2.30» we identify T*'H with 'H x 'H: For <I> E 'H the one-form
0(<1» is defined by
(O(<I»)(V(Q»:= Re (<I>, Q). (2.32)
Note that 0(<1» = df(!J, where f(!J(Q) := Re (<I>, Q).
A Poisson tensor on 'H may be defined for any It E lR\{O} by
1
B(O(<I», O(Q» := - 2ltIm (<I>, Q). (2.33)

It follows that
1
BU(O(<I») = - 21t V (i<l». (2.34)

Since this map is evidently invertible, one infers

Proposition 2.5.5. The Poisson manifold ('H, B) is symplectic. The symplectic
form w is given by
w(V(<I», V(Q» = 2ltIm (<I>, Q). (2.35)
74 I. Observables and Pure States

Let H be a self-adjoint element of lB(1-{). Define iI E C OO (1-{, JR) by

H(\II) := (\II, H\II). (2.36)

The corresponding Hamiltonian vector field is

(2.37)

The (real) linear span of V(\II) and all ~ii(\II) is Tlllft. The Poisson bracket of
functions of the type (2.36) is (cf. (1.22»
.......... i----
{A, B} = /i([A, BD = {A, BJn. (2.38)

If U is a unitary operator on 1-{, the pullback U* Aequals U-I AU. It then follows
from (2.38) and (2.15) that each such U defines a Poisson map.
The Schrodinger equation "H\II(t) = ifid\ll(t)/dt" of quantum mechanics is
nothing but (2.11) with (2.37). The solution of this equation is the Hamiltonian
flow generated by H, given by
(2.39)
We now pass to 1P1t. Recall the action of U(1) on 1-{ (cf. 2.5.1); it is easily
checked that this is a Poisson map for each z E U(l). Consider 1-{* := ft\{O};
since each point of1-{* has the same stabilizer (namely {e}), it follows that 1-{* / U (1)
is a manifold. Moreover, ft* / U (1) is a Poisson manifold: If T : 1-{* ~ 1-{* / U (1)
is the canonical projection, then T*B(\II) = T*B(z\ll) for all z and \II, so that we
can consistently define a Poisson tensor B R on 1-{* / U (l) at some point 1/1 = T (\II)
by BR('I/!) = T*B(\II). Equivalently, the Poisson bracket {, } on 1-{* / U(l) is taken
to be
T*{f, g}R = {T* f, T*g}, (2.40)
which is well-defined by the same argument. The Jacobi identity and the Leibniz
rule follow from the fact that they are satisfied on P.
Although ft* / U (1) may be infinite-dimensional, the statement of Theorem 2.4.7
actually applies.
Proposition 2.5.6. The symplectic leaves ofthe Poisson manifold 1-{* / U (1) are the
spaces Sr = 1-{r / U (1), where 1-{r = {\II E 1-{ I (\II, \II) = r2}, so that 1-{* / U(1) =
Ur>oftr/ U(l). The projective space IP1t may be identified with SI. Hence IP1t is
symplectic; the symplectic form w is explicitly given by

w",(v(iA\II), v(iBW» = -in[A,B](1/I), (2.41)

and the corresponding Poisson bracket is

"" i--- ---

{A, B} = /i[A, B] = {A, B}It, (2.42)

cf (1.22) and (2.38).

 2 The Structure of Pure State Spaces 75

See the text below (2.30) for the definition of v. We will show that Sl is a
symplectic leaf of 1t* / U(I); the argument for the other S, is similar. For each
H E !.B(1t)IR we here have introduced the function iI on Int by
if(1/I) = if(r(\II» := H(\II), (2.43)
where H is given by (2.36), and \II is now assumed to be a unit vector. Note that
(2.44)
where the norm on the left-hand side is the operator norm in !.B(1t). Indeed, our
notation if is motivated by the fact that (2.43) is a special case of the Gelfand
transform (1.30). It follows directly from the definition of the manifold structure
of Int that if is smooth for each H E !.B(1t)IR' Equation (2.37) implies

~H(r(\II» = -v (~H\II ) . (2.45)

The fact that each S, is symplectic now follows from Propositions 2.3.7 and 2.2.2,
and (2.37) or (2.45). The Poisson bracket (2.42) is derived from (2.38); it is, of
course, consistent with (2.19), (2.41), and (2.45).
Finally, the continuity of the inclusion of S I into 1t* / U (1) is immediate from
Proposition 2.5.2. •
It follows from the comment after (2.37) that the Poisson structure is completely
determined by the special case (2.42).
If 1t = eN is finite-dimensional, the symplectic form defined by (2.41) is fi
times the well-known Fubini-Study form on !PeN.
As on 1t, each unitary operator U (projected to a map on Int) is a Poisson map
with respect to (2.42). The Schr6dinger equation, projected to Int, is a special case
of (2.11): If, in somewhat sloppy notation, 1/I(t) is the flow obtained by projecting
\II(t) (cf. (2.39» from §1t to Int, one has from (2.45)

d1/l(t) = ~. (1/1 (t). (2.46)

dt H

In particular, the flow is complete for any H. As a matter of notation, we write the
solution as
(2.47)
The right-hand side is by definition the projection of (2.39) to JP1t.
Eigenvalues and eigenvectors have a neat description in the present language,
too.
Proposition 2.5.7. A vector \II E 1t is an eigenvector of an operator H E 1"l3(1t)IR
iff 1/1 = r(\II)~ is a critical point of if (i.e., d if(1/I) = 0); the corresponding
eigenvalue is H (1/1).
This is perhaps obvious from the minimax description of eigenvalues, but here
is a direct proof. The property d if (1/1) = 0 is the same as X if (1/1) = 0 for all
X E T",JP1t. By (2.31) and (2.39), this is equivalent to (\II, (H A - AH)\II) = 0,
76 I. Observables and Pure States

or (AIJI, HIJI) = (AIJI, HIJI), for all A E ~(1-0IR. Hence (IJI, Hct» E IR for all
ct> E IJI-L, which is possible only if (IJI, Hct» vanishes for all ct> E IJI-L. This implies
that IJI must be an eigenvector of H. •

From the symplectic point of view, the two steps in the construction of PH
appear in reverse order. Firstly, one pulls the symplectic form w on H back to §H;
here it is degenerate. Secondly, this degeneracy is removed upon quotienting §H
by U(l), arriving at JlD1i once more. See IV. 1.5.

2.6 Representations of Poisson Algebras

We look at symplectic manifolds as the classical analogues of modules for Poisson
algebras (cf. the opening remark in 1.5).

Definition 2.6.1. A representation of a Poi.sson algebra (1.2(1R, c, (, }) is a linear

map Jr : I.2(IR --+ COO(S, IR), where S is a symplectic manifold, satisfying

Jr(f 0 g) = Jr(f)Jr(g);
{Jr(f), Jr(g)}s = Jr({f, g}) (2.48)

(where {, }s is the Poisson bracket on S), as well as preserving completeness.

The condition (2.48) says simply that Jr : I.2(IR --+ COO(S, IR) is a morphism,
assuming that the Jordan product in COO(S, IR) is represented by pointwise multi-
plication (cf. 1.1.3). The completeness requirement means that the flow of ~lC(h) is
defined for all times if h is complete in QlIR, cf. 2.3.4. It is imposed to eliminate
constructions of the type QlIR = COO(P, IR), pi f P open in P, and Jr being simply
restriction to P'.
There is a natural notion of equivalence. Namely, two representations Jrl :
I.2(jR --+ COO(SI, IR) and Jr2 : QljR --+ C OO (S2, IR) are caned equivalent if there exists
a symplectomorphism J : SI --+ S2 such that J*Jr2(f) = Jrl(f) for all f E QlIR·
We can analyze the structure of representations of Poisson algebras of a slightly
more general type than COO(P, IR), where P is a Poisson manifold.

Definition 2.6.2. A Poisson space P is a Hausdorff topological space together

with a linear subspace QlIR C C(P, IR) and a collection Sa of symplectic manifolds
(called the symplectic leaves of P ), as well as continuous injections ta : Sa "---+ P,
such that:

1. P = Uata(Sa) (disjoint union).

2. QlIR separates points.
3. QlIR ~ C,[,(P, IR), where C'['(P, JR) consists of all f E C(P, IR) for which
t~ f E COO(Sa, JR) for each Cf.
4. QljR is closed under the Poisson bracket

{f, g}(ta(a» = {t~f, t~g}a(a). (2.49)

 2 The Structure of Pure State Spaces 77

If the ambient space P carries additional structure, such as a uniformity or a

smooth structure, one can refine the above definition in the obvious way; such
refinements will play an important role in later sections.
Definition 2.6.3. A uniform Poisson space is a Poisson space P in which the
topology is defined by a uniformity on P and that satisfies Definition 2.6.2 with
C(P, JR) replaced by the space Cu(P, JR) of uniformly continuous functions on P.
Similarly, a smooth Poisson space is a Poisson space for which P is a manifold
and C(P, JR) is replaced by C'X)(P, JR). By Theorem 2.4.7, a smooth Poisson space
with Qt]R = COO(P, JR) is nothing but a Poisson manifold. The more general concept
of a Poisson space is useful when the symplectic leaves do not fit together to
form a manifold. This happens in the context of singular symplectic reduction, cf.
IV.I.II. Moreover, we will show in 3.2.2 that the pure state space of a C* -algebra
is a uniform Poisson space. In any case, the object C,[,(P, JR) defined in 2.6.2.3 is
the function space intrinsically related to a (general, uniform, or smooth) Poisson
space P.
Definition 2.6.2 does not entail that Qt]R is a Poisson algebra under pointwise
multiplication as the Jordan product, but an interesting result arises when one
makes that assumption. In preparation for this, we remark that the notion of a
Poisson map makes sense in the context of Poisson spaces: It is still defined by
(2.15).

Proposition 2.6.4. Let (P, Qt]R) be a locally compact Poisson space for which Qt]R
is a Poisson algebra under pointwise multiplication. If rr : Qt]R ~ COO(S, JR) is
a representation ofQt]R on a finite-dimensional symplectic manifold S, then there
exists a continuous map J : S ~ P such that rr = J*.
For simplicity we show this for compact P, and assume that Qt]R contains the
unit function I p . The Stone-Weierstrass theorem then implies that Qt]R is dense in
C( P , JR) in the su p-norm. Take a point a E S, and define a linear functional la
on Qt]R by la(!) = (rr(f»(a). By the first member of (2.48), this functional is
multiplicative. If it were defined on all of C (P, JR), we could immediately conclude
from this that fa is continuous; a positivity argument shows that this follows in
the present case as well. Hence we extend fa to all of C(P, JR). It follows that fa
defines a pure state, and pure states on C(P, JR) correspond to points of P (see
2.1.4). Hence la corresponds to a point J (a) in P, and this defines the desired map
J : S ~ P. The continuity of J follows from a technical argument in the theory
of commutative C* -algebras. The second member of (2.48) obviously implies that
J is a Poisson map. 0
Corollary 2.6.5. IfQt]R = COO(P, JR)for a Poisson manifold P and rr : Qt]R ~
COO(S, JR) is a representation, then there exists a smooth Poisson map J : S ~ P
such that rr = J*.
The smoothness of J follows from the property rr = J*.
•
There is a natural notion of irreducibility for representations of Poisson algebras.
78 I. Observables and Pure States

Definition 2.6.6. A representation 7r ofa Poisson algebra Q(IR is called irreducible

if
(2.50)
If S is infinite-dimensional, it is understood that one takes the closure of the
left-hand side in the definition. The finite-dimensional irreducible representations
of a Poisson algebra associated with a locally compact Poisson space (Definition
2.6.2) can be described concretely.
Theorem 2.6.7. Under the assumptions ofProposition 2.6.4, let 7r be irreducible.
Then S is symplectomorphic to a symplectic leaf Sa of P, or to a covering space
thereof
In fact, it will follow from the proof below that an irreducible representation
space S of a locally compact Poisson space has to be finite-dimensional. The proof
of this theorem is based on 2.6.4; we have 7T -=- j* for 1 : S -+ P. For each
a E S, let Sa(a) be the symplectic leaf for which la(a)(Sa(a» contains l(a) (cf.
2.6.2); we will henceforth identify la(a)(Sa(a» and Sa(a)' By irreducibility, any
X ETaS can be written as X = ~rr(j) for some f E Q(IR. We define a linear map
1* : TaS -+ TJ(a)Sa(a) by
(2.51)
The notation is consistent: If P is a manifold, 1* is indeed the pushforward of 1,
cf. (2.17). The fact that 1* is well-defined follows from its injectivity, which we
will now demonstrate. If l*~rr(j)(a) = 0, then (j, g}(l(a» = 0 for all g E Q(IR,
since 1 is a Poisson map. But then (U;(~rr(f), ~rr(g» = 0 for all g, where (US is
the symplectic form on S. Since (US is nondegenerate, this implies ~rr(f) = 0,
which proves injectivity. Now, 1* is evidently surjective as well, because Sara) is
symplectic. Hence 1* is an isomorphism.
Combining this result with Propositions 2.3.5 and 2.3.7, we conclude that
1 (S) S; Sa(a) , where Sand Sa(a) are locally symplectomorphic (since 7r is a repre-
sentation). The completeness of 7r (see Definition 2.6.1) implies that 1 (S) = Sa(a)'
For if the inclusion l(S) S; Sa(a) were proper, we could take a neighborhood N
of a boundary point of l(S) in Sa(a), and take al E N n l(S) and az E N but
az ¢. 1 (S), such that al and a2 are connected by a Hamiltonian curve tangent to
~f (cf. 2.3.7). We then consider the Hamiltonian curve in S tangent to ~rr(j) and
passing through s" where l(SI) = al (Sl may not be unique). By 2.3.5 this curve
is mapped onto the Hamiltonian curve connecting al and a2, but this is impossible
because of our assumptions on a2. Hence the curve in S in question must suddenly
stop somewhere, contradicting the completeness of 7r .
A similar argument shows that 1 is a covering projection. For 1 not to be a
covering projection, there must exist a point a3 E Sa (a ), a neighborhood Na3 of a3,
and a connected component 1;-1 (NaJ of 1- 1(NaJ such that 1(1;-/ (NaJ) C Na3
is a proper inclusion. Butin that case we could choose points al E 1(1;-1 (Na » and
a2 E Na3 buta2 ¢. 1(1;-1 (Na3 », which can be connected by a smooth Hamiltonian
curve, tangent to some vector field ~g. Letal = 1 (SI) for some Sl, and consider the
 2 The Structure of Pure State Spaces 79

flow of ~rr(g) through s I in S. Then this flow must either suddenly stop, contradicting
the completeness of:rr , or continue outside Ji- 1(Na ) to a point S2 for which J (S2) =
a2, contradicting the assumptions on az. Hence J must be a covering projection,
and Theorem 2.6.7 is proved. •
We now return to C* -algebras and their pure state spaces. Take a C* -algebra 2l
with pure state space P = P(21) (equipped with the w*-topology), and identify
its self-adjoint part 2lJR with a subspace of C(P, R) by the Gelfand transfonn
A E C(P, R); see (1.30). We will occasionally drop the hat on A.
Proposition 2.6.8. The pure state space P = P(21) ofa C* -algebra 2l (where 2lJR
is identified with a subspace ofC(P, R) through the Gelfand transform (1.30)),
equipped with the irreducible representation spaces Sa = Inta and the inclusion
maps La, is a Poisson space.

This is a trivial consequence of 2.5.4; note that the Poisson bracket in the sense
of 2.6.2 coincides with the one (1.22) originally defined on 2lJR. Recall that the
choice of each Ha is arbitrary within unitary equivalence; the Poisson structure
on P(21) is independent of the particular choices made by the comment following
(2.42). The spaces Inta are now seen to be the symplectic leaves of P. •

Proposition 2.6.8 recognizes the fact that (the self-adjoint parts of) C* -algebras
fall under the theory of Poisson spaces. This point of view receives further support
from a reconsideration of the notion of a representation :rr of a C* -algebra on a
Hilbert space 1t (see 1.5.1). As explained in 2.5, we may identify ~(H)JR with a
subspace of the Poisson algebra COO(H, JR), so that:rr maps A E 2lJR to;(A) E
COO(Int, JR) (cf. (2.43». It follows from 2.5.6 that:rr : 2lJR ~ COO(H, JR), thus
interpreted, is a Poisson morphism.

Proposition 2.6.9. A representation :rr of a C* -algebra 2l on a Hilbert space H is

irreducible ifffor every 1{1 E Int the set {v(i:rr(A)III) I A E 2lJR} of tangent vectors
is dense in T",Int (Poisson irreducibility).

This follows from (2.31) and 2.2.2.2. Note that T",H equals {V(AIII) 1111 E
~(H)}, but does not equal {V(AIII) 1111 E ~(H)IId; nonetheless, T",Int is given
by (2.31). This is because the orthogonal complement of {V (A 111) 1111 E ~(H)JR}
in T", H projects to zero in T",Int. •

Combining 2.6.9 and (2.45), we see that the notions of irreducibility of a repre-
sentation of a C* -algebra (Definition 2.2.1) and of a Poisson algebra (Definition
2.6.6) coincide (cf. 2.2.1). Therefore, on the Poisson side there is a close fonnal
similarity between C* -algebras and Poisson algebras as far as their respective rep-
resentation theories are concerned. Indeed, combining Theorems 2.5.4 and 2.6.7
and Proposition 2.6.8, we obtain (under the above identifications)

Corollary 2.6.10. Let 2lJR be either a Poisson algebra defined by a locally compact
Poisson space P, or the selfadjoint part of a C* -algebra with pure state space
P. Then, up to equivalence, every irreducible representation of21JR is given by the
80 I. Observables and Pure States

restriction of~R to a symplectic leafof P (or p, respectively), or by this restriction

preceded by the covering projection on a covering space of such a leaf
Note that lP1t has no nontrivial covering spaces. If the Poisson space P is
compact and the Poisson algebra ~R contains the unit function, then !ZlR is dense
in C(P, JR) (cf. the proof of 2.6.4); in that case, P is actually the pure state space
of ~ as a C* -algebra (cf. 2.1). If P is merely locally compact, the same conclusion
holds if!ZlR is contained in Co(P).
The difference between representations of C* -algebras and Poisson algebras
lies on the Jordan side; from the point of view of pure states, the Jordan structure
on !ZlR eventually originates from a novel structure on P(~).

2.7 Transition Probability Spaces

Here is the structure alluded to at the end of the preceding section.
Definition 2.7.1. A transition probability on a set Pis a function
p :P xP ~ [0, 1] (2.52)
that satisfies
p(p, a) = 1 {:=} P= a (2.53)
and
p(p, a) = 0 {:=} p(a, p) = O. (2.54)
A set with such a transition probability is called a transition probability space.
The following set of definitions is natural and self-evident.
Definition 2.7.2. A family of subsets of a transition probability space P is called
orthogonal if p(p, a) = 0 whenever p and a do not lie in the same subset. The
space P is called reducible ifit is the union oftwo (nonempty ) orthogonal subsets;
if not, it is said to be irreducible. A component C ofP is a subset C C P such
that C and P\ C are orthogonal. An irreducible component of P is called a sector.
Thus any transition probability space is the disjoint union of its sectors.
Certain subsets of P are of special significance. The orthoplement of Q C P
is defined by
Q1- = {a E P I p(p, a) = 0 VP E Q}. (2.55)
It is immediately obvious that if R ~ S, then S1- ~ R1-, and that T ~ T1-1-.
Putting R = Q and S = Q1-1- shows that Q1-1-1- ~ Q1-; putting T = Q1- yields
Q1- ~ Q1-1-1-. Hence Q1- = Q1-1-1-. Accordingly, the orthoclosure of a subset
Q ~ P is defined as Q1-1-, and Q is called orthoclosed if Q = Q1-1-. It follows
that Q1- (and therefore Q1-1-) is always orthoclosed. Also, one easily sees that any
component C ~ P is orthoclosed; this applies in particular to P itself, and to any
sectorofP.
 2 The Structure of Pure State Spaces 81

Definition 2.7.3. A basis of a transition probability space P is an orthogonal

family B of points ofP with the property that
L pep, a) = I Va E P (2.56)
pEB

(if B is infinite the sum is defined as the least upper bound of all finite partial
sums).
The transition probability space is called symmetric if
pcp, a) = pea, p) Vp, a E P. (2.57)
The simplest example of a symmetric transition probability space is obtained by
taking any set P, and putting
pcp, a) := Dpo. (2.58)
Proposition 2.7.4. In a 5ymmetric transition probability space all bases have the
same cardinality.
Let BI and B2 be two bases. If both are finite, (2.56) shows that the cardinality
card(BI) of BI is given by LpEB , LOEB2 pcp, a). But then the symmetry of p
implies that this must equal card(B2). The same calculation shows that it is im-
possible that B\ is finite and B2 infinite (and vice versa). Let both be infinite. For
fixed a E B 2 , define R(a) = (p E BI I pcp, a) > O}. By (2.56), R(a) can be at
most countable. Hence the set UoE B2R(a) has the same cardinality as B 2 . On the
other hand, this set is contained in B\, so that card(B2) S card(Bd. The symmetry
of p leads to the opposite inequality, so that card(BI) = card(82). •
Consequently, one can define the dimension of a symmetric transition prob-
ability space as the cardinality of any of its bases. If B is a basis, then 81-1- =
P.
Clearly, any subset of P is a transition probability space if one simply restricts
p to it. Not every orthoclosed subset is necessarily the orthoclosure of a maximal
orthogonal subset contained in it, however: There exist examples of orthoclosed
subsets that do not have any basis. To exclude pathological cases, we impose the
following
Definition 2.7.5. A transition probability space is well-behaved if:
• It is symmetric.
• Every orthoclosed subset Q ofP has the property that any maximal orthogonal
subset of Q is a basis of Q.
In a well-behaved transition probability space any set of the type Q1- is
orthoclosed. Moreover, any orthogonal subset S has the property

S1-1- = {p E PI L
OES
pcp, a) = I} , (2.59)

since one can complete S with a basis of S1- to form a basis of P.

82 I. Observables and Pure States

For each point p in an arbitrary transition probability space p, the function p p

on P is defined by
pp(a) := pep, a). (2.60)
Proposition 2.7.6. Let P be a well-behaved transition probability space. For each
orthoclosed subset Q s:; P the function
dim(Q)

PQ:= L
i=!
Pe, (2.61)

is independent of the choice of basis {ei} of Q. Moreover,

Q={pEP/PQ(p)=I}. (2.62)
Choose a basis B = {ei} U {u j} of P that contains the given basis of Q; clearly,
u j E Q1.. for all j. By (2.56), PQ = Li Pe, = 1- Lj PUj' in which the right-hand
side is clearly independent of the choice of basis of Q. We now prove (2.62). If
p E Q, then Lj pep, Uj) = 0, so that PQ(p) = 1 by (2.56). If PQ(p) = I, then
P (p, U j) = 0 for all j, so that p E (U j U j)1.. = QH = Q. •

2.8 Pure State Spaces as Transition Probability Spaces

This section is devoted to the result that the pure state space of a C* -algebra is a
well-behaved transition probability space. To see this in perspective, we start in
the more general context of compact convex sets; cf. 1.4.5 and preceding text. We
will routinely omit the hat on the Gelfand transform.
Let K be a compact convex set (in a Hausdorff vector space). An extreme
point p E aeK is called norm-exposed if there exists some A E Ab(K, 1R), with
IIAII = I, such that {w E K / A(w) = I} = p. Equivalently, A satisfies A(p) = I
and A(w) < 1 for all WE K\{p}.
Proposition 2.8.1. Let K be a Hausdorff compact convex set with the property
that every extreme point is norm-exposed. Then the formula
pep, a):= inf {A(p) / A E Ab(K, 1R), 0 ~ A ~ IK, A(a) = I} (2.63)
defines a transition probability on the extreme boundary ae K of K. The expression
(2.63) is not changed if the infimum is taken over A(K, 1R) instead of Ab(K, 1R).
If p = a, then pep, a) = 1 by definition. The converse follows immediately
from the extra requirement on K. Condition (2.54) is easily verified if one rewrites
(2.63) as
pea, p) = 1 - sup {A(a) / A E Ab(K, 1R), 0 ~ A ~ IK, A(p) = OJ. (2.64)
The claim that we may minimize over A in A(K, 1R) follows from the density of
A(K, JR) in Ab(K, 1R) in the topology of pointwise convergence. •
If Q{ is a commutative unital C* -algebra, its self-adjoint part is of the form
Q{1R = C(P(Q{), 1R) by Theorem 2.1.7, where the pure state space is a compact
 2 The Structure of Pure State Spaces 83

Hausdorff space. Since ~R.* contains eOO(p(~», one immediately sees that (2.63)
leads to (2.58). Ifone minimizes (2.63) only over A(K, R) ~ C(p(~), R),thesame
result follows from Urysohn's lemma; since p(~) is compact and Hausdorff, it is
normal.
Theorem 2.8.2. The pure state space p(~) of a C* -algebra ~ is a well-behaved
transition probability space under (2.63). The transition probabilities are explicitly
given by p(p, a) = 0 if P and a are inequivalent, and
p(p, a) = I(Qp, QO')1 2 (2.65)
if p and a are equivalent. Here Qp, QO' E §'Jia are (arbitrary) preimages of
p, a E P'Jia (cf. 2.5 and 2.2).
Note that this implies that the transition probabilities are given by (2.58) if ~ is
commutative.
We may assume that ~IR has a unit. If it hasn't, we use 1.2.1 and (2.2); the special
point Woo satisfies p(woo , p) = ofor all p =I- Woo. To see what is happening, we first
prove the theorem for finite-dimensional C* -algebras. By Proposition 2.2.8 these
are direct sums of matrix algebras, i.e., ~ = $arotNa(C). We write A = $aAa
for A E ~. The pure state space of ~ is P = UalPC Na . We now take a fixed ex; if
a E IPC Na C p, then a(A) = a(Aa) = (QO', AaQO'), where QO' E CNa is defined
as in the statement of the theorem. The projection [QO'] onto QO' may be regarded
as an element of ~ by adding zero operators. Then
(2.66)
if p E IPC Na (i.e., it is equivalent to a), and [QO' ](p) = 0 otherwise. In particular,
[QO' ](a) = 1, and [QO' ](p) < 1 if p =I- a. This shows firstly that every pure state
is norm-exposed, and secondly that p(p, a) vanishes if p and a are inequivalent
(note that 0 < [QO'] < nand II [QO'] II = 1, since [QO'] is a projection).
We now assume that p and a are equivalent, and without loss of generality,
put ~ = rotN(C). We claim that the infimum in (2.63) is reached for A = [QO'].
For suppose there exists an A E ~IR = rotN(C)IR for which 0 < A < [QO'] and
A(a) = (QO', AQO') = 1. Choose a basis {e" ... , eN} in C N that projects onto
{a, ... , eN} in PCN. Since 0 < A < [QO'] and [QO'](ej) = 0 for i = 2, ... , it
must be that A(ej) = 0 for i = 2, ... , N. Also, clearly, A(e,) = [QO' ](e,) (since
e, = a). Then B = [QO'] - A satisfies B > 0, and B(ej) = (ej, Bei) = 0 for all
i. The latter is impossible for a positive definite matrix. Hence we can compute p
by p(p, a) = [QO' ](p), which, with (2.66), proves (2.65).
The proof of Theorem 2.8.2 for general C* -algebras follows the same idea; the
direct sum of matrix algebras is now replaced by the reduced atomic representation
1Tra of ~ (see 2.2.7 and the subsequent theory). The projection [QO'] E fJ3('Ji0') is
regarded as an element of 1Tra(ll)" by adding zero operators; hence it lies in ~R.*. As
in the finite-dimensional case, this shows that every pure state is norm-exposed,
while additionally reducing the proof to the situation where p and a are equivalent.
We then observe that A(S(~), R) ~ 1Tra(~IR)" ~ Ab(S(~), R) (with equalities
only for finite-dimensional algebras), so that we may take the infimum in (2.63)
84 I. Observables and Pure States

over all A in n m (Q(lld'. The remainder of the proof is then the same as in the finite-
dimensional case, since the property of positive definite matrices we used holds
for arbitrary positive definite operators on a Hilbert space.
A basis of P is obtained by using the decomposition P = Ua lP1-(, (see 2.5.4);
one chooses an orthonormal basis (in the usual Hilbert space sense) in each Ha,
and projects it to IP1i a. This yields a basis (in the sense of 2.7.3) of IP1ia as a
transition probability space. Combining these bases by taking the union over all a
then produces a basis ofP. The fact that P is a well-behaved transition probability
space then follows from elementary Hilbert space theory. •
The transition probability between pure states on a C* -algebra Q( may be related
to the norm on Q(*, in that

pep, a) = 1 - jllp - a1l 2 • (2.67)

If p and a are equivalent, so that they are vector states in the same Hilbert space
(cf. the comments following 2.2.6), then (2.67) is equivalent to
(2.68)
either equality follows from a simple calculation with 2 x 2 matrices. If p and a
are inequivalent, one can show that II p - a II = 2.

3 From Pure States to Observables

3.1 Poisson Spaces with a Transition Probability

We have encountered two kinds of structure on the pure state space P(Q() of a
C*-algebra Q(. Firstly, it is a Poisson space (cf. 2.6.2 and 2.6.8), and secondly, as
established in 2.8.2, it is a transition probability space. We will now examine how
these structures are interrelated. Recall (2.60).
Definition 3.1.1. The real normed vector space Q(:(P), regarded as a subspace
oflOO (P, JR) (with sup-norm), consists of all finite linear combinations of the type
L~l Ci Pp" where Ci E JR and Pi E P. The closure of Q(:(P) is called Q(~(P).
The double dual of Q(~(P) will playa central role in what follows, so that we
use a special symbol:
(3.1)

Since Q(~(P) ~ lo(P, JR), one has Q(1R(P) ~ lo(P, JR)** = loo(P, JR). The space
Q(1R(P) is the function space intrinsically related to a transition probability space
P. It is a partially ordered Banach space in the obvious way. We will now identify
this space in the case that P is the pure state space of a C* -algebra.
According to 1.7.5 there exists a central projection p in Q(** such that nm(Q()" c:::
pQ(** (cf. 2.2.7). Hence nra(Q()" is contained in Q(** in a natural way. By w*-
continuity, elements ofQ(1R c::: A(S(Q(), JR) are determined by their values on P(Q().
 3 From Pure States to Observables 85

This is, in general, not the case for arbitrary elements of Ql~* ~ Ab(S(Qt),JR.).
However, A E pQlJr:' c Ql~* is determined by Ii E £oo(P(Ql), JR.); this follows either
from the explicit expression (2.3) or from a more abstract argument. Therefore, the
Gelfand transform (1.30) maps pQl~*, and hence JTra(Ql)~, isometrically into some
closed subspace of £oo(P(Ql), JR.).

Proposition 3.1.2. IfP is the pure state space of a C* -algebra Qt, equipped with
the transition probabilities (2.65), then the Gelfand transform (1.30) isomorphi-
cally maps JTra(Ql)i ~ pQl~* (as a partially ordered Banach space) to Ql]R(P).ln
particular, Ql]R(P) = £oo(P, JR.) ifQl is commutative.

As a visual aid in proving this proposition, we define a (locally nontrivial) fiber

bundle B(P), whose base space B is the space of sectors, equipped with the discrete
topology, and whose fiber above a given base point ex is ~(TiO'h~; here TiO' is such
that the sector ex is JlD1t0'. Moreover, by (2.5.4) the pure state space P itself may be
seen as a fiber bundle over the same base space; now, the fiber above ex is JlD1t0'. We
will denote the projection of the latter bundle by r. A cross section s of B(P) then
defines a function son P by s(p) = [s(r(p))](p); in this description, we identify
a bounded self-adjoint operator H on TiO' with the corresponding function iI on
JlD1tO', cf. (2.43). By (2.44), this identification is isometric if we define the norm
of a cross section of B(P) by lis II = SUPO'EB Ils(ex)1I (where the right-hand side, of
s
course, contains the operator norm in ~(TiO'» and the norm of as the sup-norm
in £oo(P, JR.).
It follows directly from its definition that the space Ql~o(P) consists of sections
s of B(P) with finite support for which s(ex) has finite rank for each ex. Its closure
Ql~(P) contains all sections for which ex 1-+ IIs(ex)1I vanishes at infinity, and s(ex) is
a compact operator. It follows from 1.6.5 that the dual Qt~ (P)* may be realized as
the space of sections for which s(ex) is of trace class and ex 1-+ IIs(ex) II 1 (cf. (1.61»
is in £I(B, JR.). The bidual Ql]R(P) then consists of all sections of B(P) for which
ex 1-+ IIs(ex)1I is in £oo(B, JR.). It follows from (2.3) that this is precisely the image
of the Gelfand transform (1.30) of JTnJQt)i. •

IfP is simultaneously a (general, uniform, or smooth) Poisson space (cf. 2.6.2,

2.6.3) and a transition probability space, two function spaces are intrinsically as-
sociated with it: C,[,(P, JR.), defined in 2.6.2.3, and Ql]R(P), respectively. The space
naturally tied with both structures in concert is therefore

QldP, JR.) := Ql]R(P) n C'['(P, JR.). (3.2)

For example, if P is a smooth Poisson space (i.e., a Poisson manifold) equipped

with the transition probabilities (2.58), then Ql]R(P) = £oo(P, JR.), so that
QldP,JR.) = Cf,'b(P, JR.). The corresponding equation for C*-algebras is (3.6)
below.
In general, since elements of Ql L (P, JR.) are smooth on each symplectic leaf of
p, they generate a well-defined Hamiltonian flow (2.11), which, of course, stays
inside a given leaf.
86 I. Observables and Pure States

Definition 3.1.3. A (general, uniform, or smooth) Poisson space that is simulta-

neouslya transition probability space is called unitary if the Hamiltonian flow on
P defined by each element ofQtL (P, JR) preserves the transition probabilities. That
is, if p(t) and a(t) are Hamiltonian curves (with respect to a given H E QtdP, JR»)
through p(O) = p and a(O) = a, respectively, then
p(p(t), a(t» = p(p, a) (3.3)
for each t for which both flows are defined.
We infer from (2.46), (2.47), 3.1.2, 2.6.8, and 2.8.2 that the pure state space of
a c· -algebra is unitary. Also, a Poisson manifold with (2.58) is evidently unitary.
Definition 3.1.4. A (general, uniform, or smooth) Poisson space with a transition
probability is a set P that is a well-behaved transition probability space (Defini-
tion 2.7.5) and a unitary (general, uniform, or smooth) Poisson space (Definitions
2.6.2,2.6.3, and 3.1.3),Jor which QtjR = QtdP, JR) (defined in (3.2)).
This definition imposes two closely related compatibility conditions between the
Poisson structure and the transition probabilities: Firstly, it makes a definite choice
for the space QtjR appearing in the definition of a Poisson space, and secondly, it
imposes the unitarity requirement.
We collect the previous findings in
Theorem 3.1.5.
• The pure state space of a C* -algebra equipped with the w* -topology, the tran-
sition probabilities (2.63), and the Poisson structure 2.6.8, is a Poisson space
with a transition probability.
• A Poisson manifold equipped with the transition probabilities (2.58) is a smooth
Poisson space with a transition probability.

3.2 Identification of the Algebra ofObservables

This section is devoted to the following result, which shows how a unital C*-
algebra Qt can be recovered from its pure state space. We recall that Qt~(P) was
defined in 3.1.1, and that QtjR(P) ~ ioo(p, JR) (cf. (3.1). Also, we regard QtjR as a
closed subspace of Cb(P(Qt), JR) c iOO(P(Qt), JR) through the Gelfand transform
(1.30).
Theorem 3.2.1. Let P(Qt) be the pure state space of a unital C* -algebra Qt,
equipped with the transition probabilities (2.63) and the w* -uniformity inherited
from Qt*. Then
(3.4)
where CAP(Qt), JR) is the space of real-valued uniformly continuous functions on
P(Qt).
Before starting with the proof, we clarify the content of the theorem.
 3 From Pure States to Observables 87

Firstly, the w* -uniformity on p(~) may be defined by its subbase consisting of

all subsets of P x P of the type {(p, a) E P x P IIp(A) - a(A)1 < E}, where
A E ~JR and E > O. It is noteworthy, however, that the subspace of functions in
lOO(P(~), R) that are uniformly continuous with respect to any uniformity on P(~)
is closed. This generalizes the well-known fact that the subspace of continuous
functions relative to any topology on P is sup-norm closed; the proof of our
observation proceeds by the same e /3 argument.
Secondly, we know from 1.4.5 that ~JR :::::: A(S(~), R) = ~** n C(S(~), R).
There may, however, exist spurious elements of ~IR* that happen to be w*-
continuous on P(~) but not on S(~). Therefore, an arbitrary C* -algebra ~ does not
satisfy ~JR = 2t1R* n C(P(2t), R) (although a large class of such algebras does, see
below). The theorem shows that these spurious elements fail to be uniformly con-
tinuous on P(2t), and that uniform continuity on P(2t) can be used to characterize
2tJR.
We now pass to the proof of Theorem 3.2.1. According to 3.1.2, we may iden-
tify 2tJR(P) with the image of the Gelfand transform of Jr.. (2t)i (or p2tIR*) in
lOO(P(2t), R); we denote this image by rotJR' Hence we can write the right-hand
side of (3.4) as rot]R n C u ('P(2t), R). Since we identify 2t]R with its Gelfand trans-
form, and because 2tJR ~ Jrra (2t}R' we can say that 2tJR ~ rot]R. The inclusion
2t]R ~ Cu(P, R) is immediate from the definition of the w*-uniformity, so that
2tJR ~ rotJR n Cu(P, R).
We note that P(2t) = aeN(rot), where N(rot) is the normal state space of the
complexification rot ofrot]R, cf. 1.7 (recall that rot is a von Neumann algebra). For
any von Neumann algebra rot of the form rot = 123** (where 23 is a C* -algebra)
one has

(3.5)

which sharpens Theorem 2.1.5, since N(rot) ~ S(rot) may be a proper inclusion.
We apply this with Sl3]R = 2t~(P) (cf. the proof of 3.1.2), for which 123** indeed
equals our rot. As a corollary of Theorem 2.1.5, note that if L C K is a closed
subset of K for which co(L) = K, then aeK ~ L. It then follows from (3.5)
that P(rot)- ~ (aeN(rot»- = P(2t)-, where the closures are taken in the w*-
topology on rot*. Therefore, one can approximate any p and a in P(rot)- in the
w* -topology on rot* by elements of P(2t), so that Pa ~ P and ap ~ a for nets
{Pa} and lap} in P(2t). If we choose these such that P = a on 2t]R, then clearly
lima,fJ(Pa(A) - ap(A» = 0 for all A E 2t]R.
Now choose B E rot n C u ('P(2t), R). By the definition of the w* -uniformity on
2t1R , the uniform continuity of B implies that lima.fJ(Pa(B) - afJ(8» = O. Hence
p(B) = 0'(8).
Without proof, we now invoke a deep corollary of the Stone-Weierstrass theorem
for C* -algebras: If 2t and 23 are unital C* -algebras with 2t ~ 23, and B E 23 is
such that p(8) = 0'(8) for any pair p, a coinciding on 2t, then B E 2t.
Returning to the previous paragraph, this corollary implies that 8 E 2tJR . 0
Combining Theorem 3.1.5 and (3.4) we infer
88 I. Observables and Pure States

Corollary 3.2.2. The pure state space of a C*-algebra equipped with the w*-
uniformity, the transition probabilities (2.63), and the Poisson structure 2.6.8 is a
uniform Poisson space with a transition probability.
We briefly return to (3.2). According to Propositions 3.1.2 and 2.6.8, and the
comment below (2.44), for pure state spaces P = P(2l.) of C* -algebras one has
2l.]R(P) n Cu(P) c Cf(P, JR).1t then follows from Theorem 3.2.1 that
(3.6)
A unital C* -algebra 2l. is called perfect if
(3.7)
In that case, C u in (3.6) may be replaced by C.
If P(2l.) is closed (hence compact), then 2l. is obviously perfect. Hence com-
mutative C*-algebras are perfect (cf. 2.1.7, which actually implies 3.2.1 in the
commutative case), and so are finite-dimensional C*-algebras. On the basis of
Proposition 2.5.2 one might expect that the unitization !Bo(1i)n of !B o(1i) cannot
be perfect, but the opposite is true. While 2.5.2 does show that any element of fJ3('Jt)
is continuous on all points of P(fJ3 o(1i)n) except Woo (cf. (2.2», only members of
!Bo(1i)n are continuous at Woo, too. Finally, deeper analysis shows that fJ3(,}-{) is
perfect for any Hilbert space '}-{.

3.3 Spectral Theorem and Jordan Product

Given a C*-algebra 2l., one can use Proposition 3.1.2 to endow 2l.]R(P), and hence
2l.]R (cf. (3.4», with the structure of a J LB-algebra; cf. 1.1.9. It is enlightening,
however, to derive this structure from the pure state space P = P(2l.). By Theo-
rem 3.1.5 this is a Poisson space with a transition probability; our first goal is to
reconstruct the Jordan product on 2l.]R(P) from the transition probabilities.
Definition 3.3.1. Let P be a well-behaved transition probability space (cf 2.7.5).
A spectral resolution ofan element A E foo(P, JR) is an expansion (in the topology
ofpointwise convergence)
A = I:>-.jpQj, (3.8)
j

where A. j E JR, and {Q j} is an orthogonal family of orthoclosed subsets ofP (cf

(2.61 »for which Lj PQj equals the unit function on P.
Proposition 3.3.2. IfP = Ua lF1i a with transition probabilities (2.65), then any
A E 2l.:(P) (cf 3.1.1) has a unique spectral resolution.
By 2.5.4 and 2.8.2 this applies, in particular, to the pure state space of a C*-
algebra.
Firstly, the case of reducible P may be reduced to the irreducible one by grouping
the Pi in A = L~l CiPPI into mutually orthogonal groups, with the property
that (Up)..L..L is irreducible if the union is over all Pi in a given group. Thus we
 3 From Pure States to Observables 89

henceforth assume that P is irreducible, hence of the form P = IfP1-{ with the
transition probabilities (2.65).
If P is finite-dimensional, the proposition is simply a restatement of the spectral
theorem for Hermitian matrices. In the general case, let A be as above, and Q :=
{PI, ... , PN }H. If a E Q, then A(a) = Lj AjpQj(a) for some Aj and mutually
orthogonal Q j c Q, since the situation is finite-dimensional. If a E Q-L, this
equation trivially holds, as both sides vanish.
Let us assume, therefore, that a lies neither in Q nor in Q-L. Define C{JQ (a) by the
following procedure: Lift a to a unit vector E in 1{, project E onto the subspace
defined by Q, normalize the resulting vector to unity, and project back to IfP1-{. In
the Hilbert space case relevant to us, the transition probabilities satisfy
pea, p) = pea, C{JQ(a))p(C{JQ(a), p) (3.9)

for P E Q and a ~ Q-L. We now compute A (a) by using this equation, followed by
the use of the spectral theorem in Q, and subsequently recycle the same equation
in the opposite direction. This calculation establishes the proposition for a ~
Q-L. •

Proposition 3.3.3. lfP is the pure state space of a C' -algebra, A = Lj Aj PQj
is the spectral resolution of A E '2l~(P), and A2 is defined by A2 = Lj A7PQj'
then the product 0 defined by
(3.10)

turns '2l~o(P) into a Jordan algebra. Moreover, this Jordan product 0 can be
extended to '2l~(P) (cf 3.1.1) by (norm-) continuity, which thereby becomes a
J B-algebra. Finally, the bidual '2l1R (P) (with sup-norm inherited from eOO(p, JR))
is turned into a J B -algebra by extending 0 by w' -continuity.

The bilinearity of (3.10) is not obvious, and would not necessarily hold for
arbitrary well-behaved transition probability spaces in which a spectral theorem
(in the sense of 3.3.2) is valid. In the present case, it follows from the explicit form
of the transition probabilities in IfP1-{. The quickest way to establish bilinearity, of
course, is to look at a function PQ (where Q lies in a sector IfP1-{ of P) as the
Gelfand transform of a projection operator on 1{ (cf. (2.43)).
Given bilinearity and the spectral theorem 3.3.2, the proof of (1.2) reduces to
showing that (p p 0 Pr) 0 p" = Pp 0 (Pr 0 p" ) for p, a orthogonal and r arbitrary.
Through the (inverse) Gelfand transform this reduces to a calculation with 3 x 3
matrices. The first Jordan algebra axiom is trivially satisfied by (3.10).
We now show that the axioms (1.7), (1.8) hold in '2l~(P); the norm-closure
'2l~(P) will then be a J B-algebra. If A is given by (3.8), and A := SUPj IAjl, then
on the one hand IIAII ::: A, since each Aj is a possible value of A (assumed at any
point in Qj)' On the other hand, IA(a)1 :::; A Lj pQj(a) = A by (2.56), so that
II A II :::; A. Hence II A II = A. With our definition of A2, this immediately establishes
(1.9) and (LlO) (which are equivalent to (1.8)). Axiom (1.7) follows if we assume
that II A II :::; 1 and II B II :::; 1, use (3.10), the observation that if f, g E eoo are both
90 I. Observables and Pure States

positive then IIf - gil is majorized by max{1I fll, IIgll}, and the triangle inequality;
these steps yield II A 0 B II :::: 1.
The statement about the bidual is a direct consequence of Lemma 3.3.4
below. •

Without proof we state a generalization of Proposition 1.7.4.

Lemma 3.3.4. Let (2tJR, 0) be a J B -algebra. Then the Jordan product on 2tJR has
a bilinear extension (called 0 as well) to 2tlit such that the maps A ~ A 0 Band
A ~ BoA are w* -continuous and II A 0 B II :::: II A II II B II for all A, B E 2tlit. An
extension with these properties is unique.
Of course, this discussion includes the situation where 2t is commutative. In that
case, the trivial transition probabilities (2.58) and the above construction imply that
the Jordan product on 2tJR('P) = lOO(P, JR) is pointwise multiplication, as it should
be.

3.4 Unitarity and Leibniz Rule

The following result shows that the Leibniz rule (1.4) in a J LB-algebra 2tJR is
a consequence of the unitarity condition relating the Poisson structure and the
transition probabilities on P(2t).

Proposition 3.4.1. Let P be a Poisson space with a transition probability (see

3.1.4) in which every A E 2t:(P) has a unique spectral resolution (in the sense
of 3.3.1). Assume that for each h E 2t L (P, JR) (cf (3.2)) the map A ~ {h, A}
is bounded on 2tL(P, JR) (with sup-norm). If a Jordan product 0 is defined on
2t dP, JR) through the transition probabilities, in the manner of Proposition 3.3.3,
then 0 and the Poisson bracket satisfy the Leibniz rule.
The boundedness assumption holds when P is the pure state space of a C*-
algebra; it is made mainly to simplify the proof. The proposition evidently holds
when 2t L (P, JR) is a Poisson algebra, for which the assumption is violated.
Writing 8h(A) for {h, A}, the boundedness of 8h implies that the series at(A) =
L:otn8Z(A)/n! converges uniformly and defines a uniformly continuous one-
parameter group of maps on 2tL (P, JR). On the other hand, if a (t) is the Hamiltonian
flow of h on P (cf. 2.3), then at as defined by (2.13) must coincide with the
definition above, for they each satisfy the differential equation (2.14) with the same
initial condition. In particular, the flow in question must be complete. Moreover,
it follows that the Leibniz rule (yet to be established) is equivalent to the property
that at is a Jordan morphism for each t; this, in tum, can be rephrased by saying
that at(A 2) = at(A)2 for all A E 2tL(P, JR).
Let A E 2t~o(P) n 2tdP, JR). By (3.8) and (2.61), A = Lk AkPek' where all ek
are orthogonal. Unitarity implies firstly that at(A) = Lk AkPek(-t), and secondly
that the ek( -1) are orthogonal. Hence at(A) is given in its spectral resolution, so
that (at(A»2 = Lk A~Pe.(-t). Repeating the first use of unitarity, we find that
this equals at(A 2 ). Hence the property holds on 2t:(P). Now 2t:(p) is dense
 3 From Pure States to Observables 91

in ~lu~.(P) in the topology of pointwise convergence in eOO(p, R). But AI. ---+ A
pointwise clearly implies at(AA) ---+ at(A) pointwise. This, plus the w*-continuity
of the Jordan product (cf. 3.3.4), proves the desired result. •
We return to the pure state space P of a unital C* -algebra. Through the results
3.3.3 and 3.4.1, the fact that Qlll~'<P) c QldP, R), and the observation that the
associator identity (1.6) is a consequence of the special form of the transition
probabilities, we have reconstructed QlJR(P) as a J LB-algebra. The final ingredient
on P that allows one to reconstruct the C* -algebra Ql whose pure state space it is, is
its uniform structure (namely, the w* -uniformity defined by QlJR). The J L B -algebra
QlJR is given by (3.4), and the C*-algebra Ql is then constructed as in 1.1.9.
Corollary 3.4.2. Let Ql be a unital C* -algebra with pure state space P(Ql), the
latter seen as a uniform Poisson space with a transition probability. Then a :
QlJR ---+ QlJR is an automorphism (cf. 1.1.3) iff the map a* : P ---+ P, defined by
a* peA) := p(a(A»,
1. is a bijection ofP;
2. is uniformly continuous, along with its inverse;
3. is a Poisson map;
4. leaves the transition probabilities invariant.
This is now obvious, as we have seen that the data preserved by a determine
P(Q1), whereas the data preserved by a* determine QlJR. •
Corollary 3.4.3. A bijection of Int that preserves transition probabilities is
induced by a unitary or an antiunitary operator on 'H..
We start with Ql = ~o('H.), for which QlJR(P) = ~('H.)JR (cf. 1.6.5 and 3.1.2).
By Proposition 3.3.3, the Jordan structure on ~('H.)JR is therefore determined by
the transition probabilities on P(Ql) = Int. Hence the given bijection of P must
correspond to a Jordan automorphism of ~('H.)JR. The corollary then follows from
the following lemma. 0
Lemma 3.4.4. Any Jordan automorphism a of~('H.)JR is (anti) unitarily imple-
mented. That is, a(A) = U AU* for some unitary or antiunitary operator U on
'H..
To start, extend a to ~('H.) by (complex) linearity. The definition of a Jordan
morphism then implies, after some manipulations, that
(a(AB) - a(A)a(B»(a(AB) - a(B)a(A» = 0
for all A, B E ~('H.)JR. Since ~('H.) acts irreducibly on 'H., it follows that a must
either be a morphism (i.e., a(AB) = a(A)a(B» or an antimorphism (a(AB) =
a(B)a(A». If a is a morphism, one defines the unitary operator U as follows. Take
an arbitrary unit vector Q E 'H.; since Q is cyclic for ~('H.), one may start defining U
on vectors of the type AQ, where A E ~('H.). Let the range of the projectiona([Q))
be CQ(h where Q a is a unit vector. Then define U AQ := a(A)Q a • The property
lIa(B)1I = liB II for all B E ~('H.)(withB = A[Q])showsthatllUAQIl = IIAQII,
92 I. Observables and Pure States

so that U is well-defined and unitary. The property a(A) = U AU* easily follows
from the fact that a is a morphism.
The case where a is an antimorphism can be reduced to the previous paragraph.
Define a bya(A) = a(A *); this is an antilinear morphism of ~(1i). The operator
U is then constructed as in the previous paragraph, and evidently turns out to be
an@~~ •

3.5 Orthomodular Lattices

In this section we collect some material from the theory of lattices that will be
used in what follows.
.c
Definition 3.5.1. A lattice is a partially ordered set (poset) in which any two
elements x. y have a supremum (or least upper bound) x V y (that is, x :s: x v y
and y :s: x v y, and ifx :s: z and y :s: zfor some z, then x v y :s: z) and an infimum
(or greatest lower bound) x /\ y (i.e., x ~ x /\ y and y ~ x /\ y. and if x ~ z and
y ~ zfor some z, then x /\ y ~ z).
An equivalent definition of a lattice is that it is a set .c
equipped with two
idempotent, commutative, and associative operations V. /\ : x.c .c -+ .c
that
satisfy x v (y /\ x) = x and x /\ (y V x) = x. The partial ordering is then defined
by x :s: y if x /\ y = x. The largest element in the lattice, if it exists, is denoted by
I, and the smallest one (if it exists) by o. Hence 0 :s: x :s: I for all x E .c.
.c .c
A lattice is called complete when every subset of has a supremum as well
.c
as an infimum. An atom of a lattice with 0 is an element a for which 0 :s: x :s: a
implies x = 0 or x = a. A lattice with 0 is called atomic if for every x f. 0 in L
there is an atom a f. 0 such that a :s: x. All lattices occurring in this section are
complete and atomic.
The "classical" example of a lattice is obtained by taking a set S and defining .c
as the power set 2 s of S (i.e., the set of all subsets of S). The lattice structure of .c
consists of v := U and /\ := n.1t follows that I = S, whereas 0 = 0 is the empty
set. Such a lattice is distributive, in that
x V (y /\ z) = (x v y) /\ (x v z); (3.11)
this is equivalent to the same property with v and /\ swapped.
One can weaken the distributivity property by requiring only (3.11) if x :s: z;
thus a lattice is said to be modular if
x :s: z ===> x V (y /\ z) = (x v y) /\ Z Vy. (3.12)
The canonical example of a nondistributive modular lattice is the collection L(V)
of all linear subspaces of a (left) vector space V (over an arbitrary division ring [J);
the reader may keep [J) = 1R or C in mind). The lattice operations are x /\ y : = x ny,
while x v y := x + y is the linear span of x and y. Equivalently, the partial order
is given by inclusion. Evidently, I = V and 0 = O.
Definition 3.5.2. An orthocomplementation on a lattice .c with 0 and I is a map
x ~ x.L, satisfying (for all x. y E .c)
 3 From Pure States to Observables 93

• x1-1-=x.
• x:::: y {=:} y1- ::::x1-.
• x 1\ x1- = o.
• x V x1- = f.
It follows that 11- = 0 and 01- = f, and that
(x v y)1- = x1- 1\ y1-; (x 1\ y)1- = x1- V y1- (3.13)
(de Morgan's laws). A lattice with an orthocomplementation is called an
orthocomplemented lattice. For example, in the lattice £ = 2s an orthocom-
plementation is given by the set-theoretic complement.
A lattice homomorphism between two orthocomplemented lattices is a map
preserving:::: and ..L (and hence V and 1\). A lattice isomorphism is a bijection that
with its inverse is a homomorphism; we write £( c::: £2 if £( and £2 are isomorphic.
Similarly, a lattice automorphism is an isomorphism between a lattice and itself.
The following weakening of the modular law (3.12) will soon tum out to be of
prime relevance.
Definition 3.5.3. An orthocomplemented lattice £ is called orthomodular if
(3.12) holds for y = x1-, that is,
x :::: z ==} x V (x1- 1\ z) = z. (3.14)
The following reformulation of orthomodularity will be used later on.
Lemma 3.5.4. An orthocomplemented lattice £ is orthomodular if.! x :::: z and
x1- 1\ Z = 0 imply x = z.
If (3.14) holds and x1- 1\ Z = 0, then z = x V 0 = x. Conversely, if x :::: z, then
z v (x1- 1\ z) = z, so that x v (x1- 1\ z) :::: z. Assuming that x1- 1\ Z = 0, one infers
(x V (x1- 1\ z))1- 1\ Z = o. Now apply the condition stated in the lemma with x
replaced by x V (x1- 1\ z). •
Let (, ) : V x V be a Hermitian form (that is, a nondegenerate sesquilinear
form) on V, defined relative to an involution A f-+ I of l!) (think of complex
conjugation for l!) = C, and of the identity map on R). The orthoplement x1- of
x E L(V) is defined in the obvious way by x1- := {\II E V I (\II, Cl» = OVCl> EX};
this is an element of L(V) as well. One easily verifies that x U 1- = x1- (cf. (2.55)
and subsequent text), but in general x :::: X 1-1- , rather than the equality required in
Definition 3.5.2.
Therefore, one considers the lattice LeV) of orthoclosed subspaces of V, that
is, x E L(V) lies in LeV) iff XU = x. The lattice operation 1\ is the same as
in L(V), but v in LeV) is defined by x V y = (x + y)u (this is the smallest
orthoclosed subspace containing x and y). This lattice is evidently complete. One
can show that LeV) is modular iff V is finite-dimensional. In fact, in general,
any finite-dimensional linear subspace of V is orthoclosed, so that L(V) = LeV)
if V is finite-dimensional. Even in the finite-dimensional case, ..L need not be an
orthocomplementation on £(V).1t is almost trivial, however, to check the following
necessary and sufficient extra condition.
94 I. Observables and Pure States

Proposition 3.5.5. The map x ~ x~ is an orthocomplementation on C(V) iff

(x + x~)~ = ofor all x E C(V), which is equivalent to the property ('11, '11) =
o {} '11 = 0 (that is, (, ) is anisotropic). If, in addition, x + x~ is orthoclosed
(implying x + x ~ = V) for all x E C( V), then C( V) is orthomodular.
To show that the additional assumption implies orthomodularity, note that on
this assumption, for any x one has z = z 1\ V = Z 1\ (x + x~). If x ::: z, this
equals x + z 1\ x~ by the modular law (3.12) in L(V) (with y = x~; recall + is
v in L(V». Taking the double orthoplement of the equation Z = x + z 1\ x~ thus
found yields z~~ = z for the left-hand side (since z E C(V) by assumption) and
(x + z 1\ X~)H = X v (z 1\ x~) by the definition of v in C(V). This proves the
orthomodular law (3.14). •
Corollary 3.5.6. The lattice CUt) of all closed subspaces of a Hilbert space is
complete, atomic, and orthomodular.
This follows from Proposition 3.5.5, since a linear subspace of a Hilbert space
is closed iff it is orthoclosed. •
The lattices C(V) (and in particular C(Ji» enjoy the property of irreducibility.
Here a lattice is said to be reducible if it is (isomorphic to) a nontrivial Cartesian
product C = C 1 x C2 (with lattice operations defined componentwise). If not, it is
called irreducible. The key tool in analyzing reducibility of orthocomplemented
lattices is the center C(C) of C. This consists of the elements c E C for which
x = (x 1\ c) V (x 1\ c~) for all x E C. Clearly, 0, I E C(C).

Proposition 3.5.7. An orthocomplemented lattice C is irreducible iff the cen-

ter is trivial, in that C(C) = {o, l}. In general, any c E C(C) corresponds
to a factorization C ~ [0, c] x [0, c~], where the isomorphism is given by
x *+ (x 1\ C, X 1\ c~). The orthocomplementation in [0, c] x [0, c~] is defined
by (x, y)~ := (x~ 1\ C, y~ 1\ c~).

Here [0, c] = {x E C 10::: x ::: c}, etc. Note that I *+ (c, c~). The proof of this
proposition is a straightforward definition-chasing. 0

3.6 Lattices Associated with States and Observables

The connection between states and observables is further elucidated by consider-
ing various lattices naturally defined in terms of these. Also, one such lattice in
particular will playa central role in the axiomatization of pure state spaces.
Proposition 3.6.1. The collection of projections in a von Neumann algebra !JJ1
forms a complete orthomodular lattice C(!JJ1), in which the partial ordering is
given by the usual order structure inherited from !JJ1 (seen as a partially ordered
space, cf 1.3), and the orthocomplementation is x~ = I - x, where I =]I.
The lattice C(~(Ji» is isomorphic to C(Ji).
We first demonstrate the last claim. The isomorphism is obtained by identifying
a projection [K] E C(~(H» with the closed subspace K E C(H) onto which it
 3 From Pure States to Observables 95

projects (given a projection p, the subspace K ~ 1t consists of those \}I E 1t for

which p \}I = \}I; the fact that K is closed follows from elementary estimates). The
definition of the order :s and the orthocomplementation 1- in the two lattices then
rapidly leads to the conclusion that this identification leads to a lattice isomorphism;
note that for projections, x :s y iff x = xy. One then applies Proposition 3.5.5.
The proposition itself then follows from Definition 1.7.3 and Theorem 1.7.2, which
allow us to regard C(!JJt) as a sub lattice of C(1t) for some Hilbert space 1t. The
completeness of C(!JJt) is equivalent to the property that!JJt is strongly closed. Note
that!JJt is determined by C(!JJt) in the sense that !JJt = C(!JJt)". •
There are two von Neumann algebras naturally associated with a C* -algebra ~.
Firstly, one can take the bidual !JJt = ~** = Jru(~)" (cf. 1.7.4); through Propo-
sition 3.6.1 this defines the complete orthomodular lattice C(~**). The atoms of
C(~**) are the minimal projections; this lattice is atomic only for a limited class
of C* -algebras. Atomicity holds, for example, if ~ = lBo(1t), in which case
~** = 1l3(1t), so that C(~**) ::::: C(1t) by 3.6.1.
Secondly, one may choose!JJt = Jrra(~)". In view of the isomorphism Jrra(~)~ :::::
~~(P) (cf. (3.1) and 3.1.2), we write

C(~R(P» := C(Jrra(~)"). (3.15)

This is a complete atomic orthomodular lattice.
The lattice C(~**) turns out to be isomorphic to a certain lattice defined in terms
of the state space S(~). This requires the following concept.
Definition 3.6.2. A face F of a convex set K is a convex subspace that is closed
under "purification". That is, F is a face of K iff given a decomposition W =
AWl + (l - A)w2for some A E [0,1], the condition WI, W2 E F implies WE F,
and conversely, W E F implies WI, W2 E F.
Clearly, a face consisting of a single point is an extreme point of K. The set K
is a face, and we regard the empty set 0 as a face, too. For example, the faces of
an equilateral triangle (interior plus boundary) in 1l~2 are the empty set, the three
comers, the three (closed) sides of the triangle, and the triangle itself.
The set F(K) of all faces of K is partially ordered by inclusion, and has a
minimal element 0 = 0 and a maximal element I = K. The intersection of an
arbitrary family offaces is a face as well. Hence F(K) is a complete lattice with 0
and I, for which x /\ y = x ny, and x v y is the intersection of all faces containing
x U y. The atoms of F(K) are the extreme points.
Even if K = S(~), the lattice F(K) is not particularly well behaved. As the
following result shows, it turns out to be preferable to look at a smaller set of faces.
Proposition 3.6.3. The collection F(S(~» of all norm-closed faces in the state
space ofa unital C* -algebra is a complete orthomodular lattice under the following
operations: :s is ~, and F -'- is the supremum of the set of all norm-closed faces
that are orthogonal to F (here we say that F2 is orthogonal to FI , or F2 1- F" if
there is an element A E [0, [] ~ ~llt ::::: Ab(S(~), R.) such that A(w) = 1 for all
°
W E FI and A(w) = for all W E F2)' Finally, F(S(~» is isomorphic to C(~**).
96 I. Observables and Pure States

Given a projection P E QI.]R*, the set Fp := {w E S(QI.) I w(p) = I} is easily

shown to be a norm-closed face in S(QI.). Equally easily, given a norm-closed
face F, the set IF := {A E QI.]R* I w(A* A) = OVw E F} is seen to be the self-
adjoint part of a a -weakly closed left ideal in QI.**. A technical argument in the
theory of von Neumann algebras shows that any such ideal must be of the form
IF = QI.]R* p.l for some projection p.l .It is then easily checked that F p = F, and that
the correspondence p # Fp thus established yields an isomorphism of £(QI.**)
and F(S(QI.» as lattices. Note that under this isomorphism minimal projections
correspond to pure states.
We now tum to the orthocomplementation. If A = 0 on Fp and A E [0, II], then
vA E IFp. Hence vA = Bp.l for some B, so that A = Ap.l. Since A* = A,
this implies A = p.l A, so that A = p.l Ap.l. Now, p.l Ap.l ~ p.l IIp.l = p.l,
since A ~ II, so that A ~ p.l. A similar argument shows that A = I on F p and
A E [0, II] imply p ~ A. Therefore, if A = 0 on F p2 , A = Ion F p1 , and A E [0, II],
then PI ~ A ~ pi, from which PI ~ pi; we say that PI .1 P2. The converse is
obvious, so that we have shown that FPl .1 FP2 is equivalent to PI .1 P2. We now
notice that p.l equals the supremum of alI q for which p .1 q, and conclude that
the bijection p # F p preserves orthocomplementation. D
We are going to show that the lattice £(Ql.IR(P» is isomorphic to a certain lattice
defined by the transition probabilities on P(QI.). For the moment, however, we
return to the general setting of transition probability spaces (cf. 2.7).
Proposition 3.6.4. The collection of orthoclosed subsets of a well-behaved tran-
sition probability space P forms a complete atomic orthomodular lattice £(P)
under the operations x /\ y = x ny, x v y = (x U y).l.l (equivalently, ~ is ~),
and.l is given by (2.55).
The orthomodularity follows from Lemma 3.5.4: Assume x ~ z, and choose a
basis B(z) ofz containing a basis B(x) of x . It follows from (2.56) and the definition
/\ = n that B(z)\B(x) is a basis of x.l /\ z. If this equals 0, then B(z) = B(x), and
hence z = x.
Equations (2.59) and (2.53) imply that pH = P for all PEP; hence each point
ofP lies in £(P), and the definition of ~ implies that these points are precisely the
atoms of £(P). The completeness of £(P) is obvious, since arbitrary intersections
of orthoclosed subsets are orthoclosed, and the lattice is orthocomplemented. •
Proposition 3.6.5. Let P = U a P a be the decomposition of a well-behaved
transition probability space into its sectors (cf 2.7.2). Then
(3.16)

Here each factor £(Pa ) is irreducible; in particular, P is irreducible iff £(P) is

irreducible.
If C is a component of P (so that P = C U C.l), and Q, R E £(P) satisfy
Q ~ C and R ~ P\C = C.l, then Q U R = Q v R; this follows from repeated
application of (3.13), and from Q.l.l = Q (etc.). Since sectors are components,
 3 From Pure States to ObselVables 97

this shows that c := P a (regarded as an element of C(P» is in the center C(C(P».

Equation (3.16) then follows from Proposition 3.5.7.
Now suppose that P is irreducible, yet C(C(P» nontrivial. This means that
there exists an orthoclosed subset c <; P (where c i=- 0 and c i=- P) such that
Q = (Q /\ c) V (Q /\ c-L) for all orthoclosed Q <; P. Since P is irreducible, one
cannot have P = c U c-L , so that there is an atom p lying in neither c nor c-L . Taking
Q = p we thus find Q /\ C = Q /\ c-L = 0 (recall that /\ = n in C(P». This shows
that such c cannot exist, and therefore C(P) must be irreducible. •

Theorem 3.6.6. If2! is a C*-algebra with pure state space P(2!), one has the
lattice isomorphism (cf (3.15) and 3.6.4)

C(2!IR(P» ~ C(P(2!». (3.17)

By Theorems 2.5.4, 2.8.2, and 3.6.5, one has C(P(2!» ~ Op C(Inip)' On the
other hand, C(2!IR(P» (which by definition is C(11ra(2!)f~), cf. Proposition 3.1.2)
equals C(E9p~(1{p» by (2.3). The center of this lattice is generated by the minimal
central projections [1{p], and by 3.5.7 and 3.6.1. One therefore obtains C(2!IR (P» ~
Op C(1{p)' Finally, if K is a closed subspace of some Hilbert space 1{, and K] :=
K n §1{, then K ++ r(K j ) (where r : §1{ ~ Ini is the canonical projection, cf.
2.5) establishes an isomorphism between the lattices C(1{) and C(Ini). •

The lattice C(P(2!» occurs in an interesting reformulation of the spectral the-

orem. In preparation, recall from basic measure theory that the a-algebra 13(JR)
of Borel subsets of JR is an orthocomplemented lattice in which S is <; (hence
B] V B2 = B] U B 2, B j / \ B2 = B] n B 2, 0 = 0, and I = JR). This lattice is
not complete, but merely a-complete (i.e., V and /\ exist for arbitrary countable
families).

Theorem 3.6.7. For each self-adjoint element A of a C* -algebra 2! there exists

a lattice homomorphism CPA : 13(JR) ~ C(P(2!» with the property that

(3.18)

if the B j are mutually disjoint. For each p E P(2!) the Gelfand transform A then
has the spectral resolution

A(p)= r A.dp~ (A.),

JIR A
(3.19)

where the Borel measure P:A on JR is defined by (cf (2.61))

P~A (B) := P'PA(B)(P)· (3.20)

This follows from the usual spectral theorem for self-adjoint operators on a
Hilbert space, applied to 11m(A) (cf. 2.2.7). It is easily checked that the precise
choice of 11m (which, we recall, depends on choosing a pure state in each sector
of P(2!» does not affect any of the statements in the theorem, as different choices
lead to equivalent realizations. •
98 I. Observables and Pure States

3.7 The Two-Sphere Property in a Pure State Space

The lattice £(P(I.((» plays a central role in the proof of the following char-
acterization of the pure state space of a C* -algebra as a transition probability
space.
Definition 3.7.1. A well-behaved transition probability space P (with associated
lattice £(P») is said to have the two-sphere property iffor any two points p, a
(with pi-a) lying in the same sector ofP, the space p Va is isomorphic as
a transition probability space to the two-sphere S2, with transition probabilities
given by p(z, W) = ~(l + cos8(z, w», where 8(z, w) is the angular distance
between z and w, measured along a great circle.
To understand the nature of this property, recall that a two-sphere S2 with radius
I may be regarded as the extreme boundary of the unit ball B3 C JR.3. The latter
is affinely isomorphic to the state space S(VJ1 2 (C» of the C*-algebra of 2 x 2
matrices, so that S2 is the pure state space of this algebra. Concretely, we identify
a state on VJ12 (C) with a density matrix p on C 2 , which may be parametrized as in
(1.50). Restricted to the extreme boundary, this parametrization leads to a bijection
between IP'C 2 and S2. Under this bijection the transition probabilities (2.65) on 1P'C2
are mapped into the ones stated in 3.7.1. In other words, the two-sphere property
states that there exists a fixed two-sphere S;, : : :
1P'C 2 , equipped with the standard
Hilbert space transition probabilities p = PC2 given by (2.65), and a collection
of bijections Tpva : p V a --+ S;"defined for each orthoclosed subspace of the
type p V a (where p and a #- p lie III the same sector of P), such that for all
p', a' E p Va,
pC2(Tpva (p'), Tpva(a'» = p{p',a'). (3.21)
Theorem 3.7.2. Let a well-behaved transition probability space P (with associ-
ated lattice £(P») have the two-sphere property. IfP has no sector ofdimension 3,
then P :::::: U a lP'7t a as a transition probability space (for somefamily Pia} ofHilbert
spaces), where each sector lP'7ta is equipped with the transition probabilities (2.65 ).
This statement is not necessarily false when P does have sectors of dimension
3 (in fact, we believe it to be true in that case as well); unfortunately, the proof
below does not work in that special dimension.
If p and a lie in different sectors of P, then p V a = {p, a}; this follows from
repeated application of (3.13) and pl...L = P (etc.). In any case, it is sufficient
to prove the theorem for each sector separately, so we may assume that P is
irreducible. The first step in the proof is then to construct the lattice C{P) (cf.
3.6.4). The strategy of the proof is to characterize £(P), and then use the so-called
coordinatization theorem in lattice theory to show that C{P) = £(1t) for some
Hilbert space 1t.
We already know that £(P) is orthomodular, atomic, and complete (as estab-
lished in Proposition 3.6.4); by 3.6.5 it is irreducible (in the sense of 3.5.7) as well.
A lattice £ is called atomistic if every element is the supremum of the collection
of its atoms. If £ is orthomodular, atomic, and complete, then it is atomistic. For let
 3 From Pure States to Observables 99

x be the supremum of the atoms in some Z E C, and assume x < z. By (3.14) one
then has x1- /\ Z =1= 0, so that x1- /\ Z must contain an atom, which is a contradiction.
To apply the coordinatization theorem, we need to establish a further property
of C(P). An atomistic lattice C with 0 is said to have the covering property if
for an atom a E C and an arbitrary element x E C with a /\ x = 0, the inclusions
x S y S x Va for some y E C imply y = x or y = x Va. For example, the
lattices C(V) in 3.5 have the covering property.

Lemma 3.7.3. C(P) has the covering property.

Consistent with previous notation, we denote atoms of C(P) (hence points of

P) by p, a, and arbitrary elements by Q, Qi, R, S.
Let n = dim(Q) (cf. 2.7); for the moment we assume n < 00. We will first use
induction to prove that if p fit Q, then the element (p v Q) /\ Q1- is an atom.
To start, note that if Q I S Q2 for orthoclosed Q I, Q2 sets of the same finite
dimension, then Q I = Q2. For an orthoclosed set in P is determined by a basis of it
(cf. (2.62)), which in tum determines its dimension. This implies that dim(p v Q) >
dim(Q) if p fit Q (take QI = Q and Q2 = p v Q). Accordingly, it must be that
(p v Q) /\ Q1- > 0, for equality would imply that dim(p v Q) = dim(Q).
For n = 1, Q is an atom. By assumption, p v Q is S2; hence (p v Q) /\ Q1-
is the antipodal point to Q in p v Q, which is an atom, as desired. Now assume
n > 1. Choose a basis {edi=l, ... ,dim(Q) of Q; then Q = V7=lei' Put R = v7::iei;
then R < Q, whence Q1- < R1-, so that (p v Q) /\ Q1- S (p v Q) /\ R1-. The
assumption (p v Q) /\ Q1- = (p v Q) /\ R1- is equivalent, on use of Q = R v en,
(3.13), and the associativity of /\, to«p v Q)/\ R1-)/\e; = (p v Q)/\ R1-, which
implies that (p v Q) /\ R1- S e;. This is not possible, since the left-hand side
contains en. Hence
0< (p v Q) /\ Q1- < (p v Q) /\ R1-. (3.22)

It follows from the orthomodularity of C(P) that if R S Sand R S Q, then

(3.23)

Since R < Q and R S p v R, one has p v Q = (p v R) v Q. Now use (3.23) to

find

The right-hand side equals a v en, where a := (p v R) /\ R1- is an atom by the

induction hypothesis. The equality a = en would imply that p E Q, hence a =1= en.
But then (3.22) and the two-sphere property imply 0 < dim«p v Q) /\ Q1-) < 2,
so that (p v Q) /\ Q1- must indeed be an atom.
It follows that dim(p v Q) = dime Q)+ 1. Hence any S C P satisfying Q S S S
P v Q must have dimeS) equal to dim(Q) or to dim(Q) + 1. In the former case, it
must be that S = Q by the dimension argument earlier. Similarly, in the latter case
the only possibility is S = p v Q. All in all, we have proved the covering property
for finite-dimensional sublattices. A complicated technical argument involving the
 100 I. Observables and Pure States

dimension theory of lattices then shows that the covering property holds for all
x E C(P). 0
Atomistic lattices with the covering property are known as AC-Iattices. At this
stage we can sum up by saying that C(P) is a complete irreducible orthomodular
AC-lattice. This allows us to use the following classical coordinatization theorem;
cf. 3.5 for the definition of C(V). This involves the notion of a chain, which is
a totally ordered subset of the lattice C. The length (also called rank, height,
or dimension) of an AC-lattice C is the cardinality of a maximal chain (which
contains 0 and l) minus I, which is well-defined because of the covering property.
This number coincides with the minimal number of atoms Pi for which [ = V Pi;
for C = C(P) it is equal to the dimension of P (as defined in 2.7).
The coordinatization theorem for AC-Iattices is the following.
Theorem 3.7.4. Let C be a complete irreducible orthomodular AC-lattice of
length:::: 4. There exists a vector space V over a division ring JD) (both unique
up to isomorphism), equipped with an anisotropic Hermitian form (defined rel-
ative to an involution ofll)), and unique up to scaling), such that C ~ C(V) as
orthocomplemented lattices.
We omit the lengthy and complicated proof of this theorem. In the context of
our lattice C(P), the essential point is that the division ring Il)) is constructed by
choosing two atoms p and a i- p, whereupon Il)) ~ (p va) \ a. The vector space
V is constructed in terms of a basis {ei}, which corresponds to a basis {ei} of P
(or, more generally, of the set of atoms in C); hence the length of C is equal to
the dimension of V. We will need neither the explicit form of the addition and
multiplication in Il)), nor the scalar multiplication in V (which are given in terms
of a certain geometric procedure). To proceed, the following information suffices.
Lemma 3.7.5. Let V be 3-dimensional, and let C(V) carry a topology for which
the lattice operations V and /\ are jointly continuous. Then Il)) (regarded as a
subset of the collection of atoms in C(V»), equipped with the topology inherited
from C(V), is a topological division ring (i.e., addition and multiplication are
jointly continuous).
This is clear from the explicit construction of addition and multiplication in
Il)). •

Let F E C(P) be finite-dimensional. We can define a topology on [0, F] (i.e.,

the set of all Q E C(P) for which Q :::: F) through a specification of convergence.
Given anet {Qd in F, we say that QA ~ Q when eventually dim(QA) = dim(Q),
and if there exists a family of bases {et} for {Qd, and a basis {e j } of Q, such that
Li,j p(et, ej) ~ dim(Q). This notion is actually independent of the choice of all
bases involved, since L j p(p, e j) is independent of the choice of the basis in Q
for any PEP, and similarly for the bases of QA (to see this, extend {ej }~~I(Q) to
a basis {e j } ~~I(P) , and use (2.56». An equivalent definition of this convergence is
that QA ~ Q if P(PA' a) ~ 0 for all a E F /\ Q-.l and all {pd such that PA E QA'
 3 From Pure States to Observables 10 1

Lemma 3.7.6. The above construction defines a topology on F, which is

Hausdorff.

The first claim is easily verified. For the second, let QA -+ Q and QA -+
R. Then P(PA' a) -+ 0 for all a E Q~ V R~ = (Q /\ R)~, and all {PAl for
which PA E Q A' Choose a basis {e j 1of Q that extends a basis of Q /\ R. Then
"dim(QAR)
L..j=! P( PA' ej ) _
-
I , but aIso "dim(Q)
L..j=! p ( PA' .) -
eJ -
l,SInce
' Q A -+ Q . Hence
P(PA' a) -+ 0 for all a E Q /\ (Q /\ R)~. This leads to a contradiction unless
Q=R. •

Lemma 3.7.7. The restriction of this topology to any two-sphere P Va :::= S2 in

F induces the usual topology on S2 . Moreover, v and /\ are jointly continuous on
any [0, Fl, where F is a 3-dimensional subspace of C(P).

If we restrict the topology to the atoms in F, then PA --+ P if P(PA' p) -+ 1.

On F = P v a :::= S2, one can easily show from the explicit form of the transition
probabilities P that the convergence p( 1/IA' 1/1) -+ 1 is equivalent to p( 1/IA' q;) -+
p(1/I, q;) for all q; E P Va. Namely, if P(1/IA' 1/1) -+ 1 in lP1i (for any Hilbert space
1i; the case of relevance is 1i = C 2 ), then I(\IIA, \11)1 -+ 1 for arbitrary lifts \IIA' \II
in §1i. Choose an orthonormal basis {ei 1in 1i containing \II; the equation (\II, <1» =
Li(\II, ei )(ei, <1» then rapidly leads to the conclusion 1(\IIA, <1»1 -+ 1(\11, <1»1 for all
<I> E §1i. The corresponding topology is the projection of the usual topology on
C2 to S2 :::= IPC 2 , which demonstrates the first claim.
We turn to the proof of joint continuity of v and /\. Assume that F E C(P) is 3-
dimensional. We firstly show that PA -+ P and a A -+ a, where P and a are atoms,
impliesPA VaA -+ pva.LetrA = (PA vaA)~/\F,andr = (pva)~/\F;theseare
atoms. Let p~ be the antipodal point to PA in PA vaA(i.e., p~ = Pi: /\
(PA vaA »,
and let a{ be antipodal to a A in PA V a A. Then {PA' p~, rAl is a basis of F, and
so is {aA, a{, rAJ. The definition of a basis and of PA -+ p, a A -+ a implies that
p(p, r A) -+ 0 and p(a, r A) -+ O. Hence p(r, r A) -+ 1. Now take an arbitrary atom
ex A E rf /\ F, and complete to a basis {ex A, ex~, rd, where ex~ E PA V a A. Again,
the definition of a basis implies that p(ex A , r) -+ O. By our second definition of
convergence, one therefore has PA V a A -+ P Va.
Secondly, we show that QA -+ Q and RA -+ R, where Q and R are two-
dimensional subspaces of F, implies QA /\ RA -+ Q /\ R (we assume Q =I- R,
so eventually QA =I- RA)' Let ex = Q~ /\ F, fJ = R~ /\ F, Y = Q /\ R, and
YA = QA /\ R A ; as a simple dimension count shows, these are all atoms. By
assumption, P(YA' ex) -+ 0 and P(YA' fJ) -+ O. Since (ex U fJl = (ex V fJ)~ by
definition of v, and (ex v fJ) is two-dimensional, Y is the only point in F that is
orthogonal to ex and fJ. Hence P(YA, y) -+ 1; if not, the assumption would be
contradicted. But this is precisely the definition of QA /\ RA -+ Q /\ R, and the
proof is finished. •

Corollary 3.7.8. The division ring ]I)) equals C, and the involution relative to
which the Hermitian form of Theorem 3.7.4 is defined is complex conjugation.
102 I. Observables and Pure States

It follows from the previous lemma that Jl)) is locally compact and connected.
According to the classification of locally compact division rings, there exist only
three connected ones: Jl)) = R., C, and !HI (the quatemions). Of these, only C is
homeomorphic to (p v a)\a c:::: ]R2. Note that the algebraic structure is therefore
entirely determined by the topology.
Moreover, Lemma 3.7.7 implies that the orthocomplementation 1. is continu-
ous on 3-dimensional subspaces. If one inspects the way the involution A 1-+ Iof
Jl)) is constructed in the proof of Theorem 3.7.4, one immediately infers that this
involution (of C in our case) must be continuous as well. It can be shown that C
possesses only two continuous involutions: complex conjugation and the identity
map. The latter cannot define a nondegenerate sesquilinear form (so that in par-
ticular, the lattice .c(V) cannot be orthomodular). Hence one is left with complex
conjugation. •
Note that we have used the two-sphere property twice, for different purposes:
firstly for deriving the covering property of .c(P), and secondly for identifying
Jl))=C.
With this corollary in hand, the definition of a Hermitian form implies that
(\}I, \}I) must be real for all \}I E V, and the anisotropy means that (\}I, \}I) must be
nonzero and have the same sign for all \}I. If necessary, one may change the sign
of the form so as to make it positive definite. Accordingly, V is equipped with an
inner product in the usual sense, that is, it is a pre-Hilbert space. The fact that V
is actually a Hilbert space follows from the orthomodularity of .c(P) c:::: .c(V) by
a rather technical result, whose proof we omit.
Proposition 3.7.9. A pre-Hilbert space V over C is complete iff the associated
orthocomplemented lattice .c( V) is orthomodular.
We conclude that .c(P) is isomorphic to the projection lattice .cO-O of some
complex Hilbert space 'H. Therefore, their respective collections of atoms P and
JP>1i must be isomorphic. Accordingly, we may identify P and JP>1i as sets. Denote
the standard transition probabilities (2.65) on JP>1i by PH' With p the transition
probabilities in P, we will show that p = P1t.
Refer to the text following 3.7.1. We may embed S;r
isometrically in JP>1i; one
then simply has p = P1t on S;f' Equation (3.21) then reads
(3.24)
in particular, PH(Tpvu (p'), Tpvu(a'» = 0 iff p(p', a') = O. On the other hand, we
know that P and PH generate isomorphic lattices, which implies that PI-(p', a') = 0
iff p(p', a') = O. Putting this together, we see that
PH(Tpvu(p'), Tpvu(a'» =0
iff PH(P', a') = O. A fairly deep generalization of Corollary 3.4.3 states that a
bijection T : JP>1i1 ---+ JP>1i2 (where the 'Hi are separable) that merely preserves
orthogonality (i.e., P H2(T(p'), T(a'» = 0 iff PHI (p', a') = 0) is induced by a
unitary or an antiunitary operator U : 'HI ---+ 'H2. We use this with 'HI = P va,
 3 From Pure States to Observables 103

1t2 = S;f'
and T = Tpv(J' Since Tpv(J is induced by a unitary or an antiunitary
map, which preserves PH, we conclude from (3.24) that PH(P', a') = p(p', a').
Since P and a (and p', a' E pYa) were arbitrary, the proof of Theorem 3.7.2 is
finished. •

3.8 The Poisson Structure on the Pure State Space

We now further investigate how the transition probabilities and the Poisson
structure on the pure state of a C* -algebra are related.

Theorem 3.8.1. IfP is the pure state space of a C* -algebra, then the symplectic
leaves of P as a Poisson space coincide with the sectors of P as a transition
probability space.

This is immediate by combining 2.5.4 and 2.8.2 with 2.6.8.

•
If the C* -algebra in question is commutative (so that its transition probabilities
are (2.58», it is understood to be equipped with the zero Poisson structure. cf.
the comment following 1.1.5. If, however, 2lIR is a Poisson algebra associated to a
Poisson space P, then the natural transition probabilities on P are given by (2.58).
and the above result fails: The sectors of P are its points, whereas its symplectic
leaves are nontrivial if the Poisson bracket does not identically vanish.
We now show that the symplectic structure on lP1t is determined by the transition
probabilities (2.65) and unitarity; recall Definition 3.1.3.

Theorem 3.8.2. Let lP1t, equipped with the transition probabilities (2.65) and its
usual manifold structure, be a unitary Poisson space for which the Poisson structure
is symplectic. Then the Poisson structure is determined up to a multiplicative
constant, and is given by (2.42)for some Ii =1= o.

It follows from (3.2) and Proposition 3.1.2 that 2ldP, R) equals the Gelfand
transform of ~(1t)IR. According to the definition of unitarity, the Hamiltonian flow
generated by any function A on lP1t (where A is a bounded self-adjoint operator
on 1t, cf. (2.43» must preserve the transition probabilities (2.65). Corollary 3.4.3
and Stone's theorem imply that such a flow must be of the form 1/I(t) = e- itC (A)1/I;
cf. (2.47), where C(A) is some self-adjoint operator depending on A in an as yet
unknown way. Antiunitary flows are excluded, for they cann<!t satisfy 1/1(0) = O.
We now compute the Poisson bracket of the functions A and B (see 2.43).
Using (2.8), (2.11), and the preceding paragraph, we obtain {A, B}(1/I) =
1, B (eXP(itC(A»1/I ) It = O. The right-hand side equals i[CW,B](1/I). The anti-
symmetry of the left-hand side implies that C(A) = Ii-I A for some Ii-I E R. The
value Ii-I = 0 is excluded unless 1t is one-dimensional, for otherwise the Poisson
structure would be degenerate. In other words, the Poisson bracket is given by
(2.42). Since the collection of all differentials dA spans the cotangent bundle at
each point 1/1 of lP1t. the Poisson structure is completely determined. •
104 I. Observables and Pure States

Theorems 3.8.1 and 3.8.2 show that the Poisson structure on the pure state
space P(Qt) = Ua JPl1ta of a C* -algebra (cf. 2.5.4) is to a large extent determined
by unitarity. The only freedom resides in a possible sector-dependence of n; here
n- n
I i- 0 except in one-dimensional sectors (in which the value of is irrelevant, as

the Poisson bracket identically vanishes at such points). The choice (1.22) for the
Poisson bracket on Qt]R corresponds to taking nto be a sector-independent constant.
n
We may regard as a function on P(Qt), which is constant on each sector. If A
denotes an element of Qt]R, the restriction of A to a sector JID1t a corresponds to an
operator Aa (cf. (2.43». The sector in which p E P(Qt) lies is called a(p). With
this notation, the Poisson bracket is then given by
" " i---
{A, B}(p) = n(p) [Aa(p), Ba(p)](P)· (3.25)

The following result shows that under a natural topological requirement the sector-
n
dependence of cannot be arbitrary.
Lemma 3.8.3. Equip P(Qt) = Ua JlD1ta with the Poisson structure (3.25). Assume
thatP(Qt) is equipped with a uniformity for which Qt]R (as defined in (3.4)) is closed
n
under Poisson brackets. Then the function is uniformly continuous on P(Qt).
This applies in particular to the w* -uniformity on P(Qt). Suppose n is not
uniformly continuous. We then take A, B E Qt]R in such a way that Aa and
Ba are independent of a in a neighborhood of a point a of discontinuity of
n, with [Aa(a), Ba(a)] i- o. Then the real-valued function on P(Qt) defined by
p ~ n(p){A, B}(p) is certainly uniformly continuous near a, since its value at
p is equal to i[Aa~(p)](p). But, by assumption, {A, B} is uniformly contin-
uous as well. Because of the factor n, the product n{A, B} cannot be uniformly
continuous. This leads to a contradiction. •
We can always rescale the Poisson bracket by multiplying it with n(·); the
resulting Poisson structure is then the same in all sectors. In view of Lemma 3.8.3,
in the given situation Qt]R will be closed under the rescaled Poisson bracket as well.

3.9 Axioms for the Pure State Space of a C* -Algebra

We can sum up part of the preceding discussion as follows.
Theorem 3.9.1. The pure state space P of a unital C* -algebra Qt, equipped with
the w* -uniformity, the transition probabilities (2.65), and the Poisson structure
( 1.22), has the following properties.

C* 1: P is a uniform Poisson space with a transition probability (Definition 3.1.4).

C*2: P has the two-sphere property (Definition 3.7).
C* 3: The sectors ofP as a transition probability space coincide with the symplectic
leaves ofP as a Poisson space.
C*4: The space Qt]R (defined through C* 1 by (3.2)) is closed under the Jordan
product constructed from the transition probabilities in 3.3.
 3 From Pure States to Observables 105

C* 5: The pure state space P(Qt) ofQt (as defined in 2.1.1) coincides with P.
Recall that Qt]R is the self-adjoint part of Qt, and that the norm on Qt]R is equal
to the sup-norm that Qt]R inherits from its inclusion in i(~\P, JR.). Property C* 1
was established in Corollary 3.2.2; C*2 is immediate from (2.65); C*3 holds by
Theorem 3.8.1; C*4follows from (3.6) and Theorem 3.2.1; and C*5 holds because
the uniformity on P(Qt) used to establish 3.2.1 is precisely the w* -uniformity. •
We now turn things around, and claim that the properties C* l-C* 5 actually
characterize pure state spaces of unital C* -algebras. Property C* 5 is then regarded
as an axiom restricting the possible uniformities on P. As an axiom, the precise
meaning of C*5 is as follows. From Axioms C*I, C*2, and C*4 the space Qt]R
emerges as a J B-algebra, which is contained in C(P, JR.) as a partially ordered
Banach space. Hence each element of P defines a pure state on Qt]R by evaluation;
Axiom C* 5 requires that all pure states of Qt]R be of this form (note that by C* 1,
the function space Qt]R ~ Q(1R already separates points).
Theorem 3.9.2. If a set P satisfies C* l-C* 5 (with P as a transition probability
space containing no sector of dimension 3), then there exists a unital C* -algebra
Qt, whose self-adjoint part is Qt]R (defined through C* 1). This Qt is unique up to
isomorphism, and can be explicitly reconstructed from p, such that
1. P = P(Qt) (i.e., P is the pure state space ojQt).
2. The transition probabilities (2.63) coincide with those initially given on P.
3. The Poisson structure on each symplectic leafofP is proportional to the Poisson
structure imposed on the given leaf by (1.22).
4. The w* -uniformity on P(Qt) defined by Qt is contained in the initial uniformity
onP.
5. The CO-norm on Qt]R C Qt is equal to the sup-norm inherited from the inclusion
QtIR C iOO(P, JR.).
As stated after Theorem 3.7.2 (which is an important step in the proof of 3.9.2),
we believe that the restriction to dimension =I- 3 can be dropped.
The proof of this theorem essentially consists in the descri ption of the construc-
tion of Q(; practically all the work has already been done. Axioms C* I and C*2
entirely determine P as a transition probability space by Theorem 3.7.2. Hence
QtIR(P) is determined by Proposition 3.1.2. We now use Axiom C*3, which implies
that each symplectic leaf of P is a projective Hilbert space Ini a. For the moment
let us assume that each leaf Inia has a manifold structure (e.g., the usual one)
relative to which all functions Ii (cf. (2.43», where H E !E(1t a )]R, are smooth.
Then QtIR(P) n Cu(P, JR.) C C'l':'(P, JR.) by the explicit description of Qt]R(P) in
3.1.2. It then follows from Axiom C* 1, in particular from (3.2), that
(3.26)
This space is norm-closed by one of the remarks following 3.2.1. The condition
in Proposition 3.3.2 holds, so that we can construct a Jordan product in Qt]R by the
procedure in 3.3. By Proposition 3.3.3 and Axiom C*4, this turns Qt]R into a J B-
algebra. At this stage we can already construct the pure state space P(Qt) through
106 1. Observables and Pure States

2.1.1 and 1.4.1; the property 3.9.2.1 then holds by Axiom C*5, whereas 3.9.2.2
follows from Theorem 2.8.2.
We may regard the restriction of !2l]R to a given sector IP1t", as the Gelfand
transform of a Jordan sub algebra of 23(1i",)]R. This subalgebra must be weakly
dense in 23(1i",)]R, for otherwise Axiom C*5 cannot hold.
By Axiom C*3 and a straightforward modification of Theorem 3.8.2 (taking into
account that the restriction of !2ldP, R) to IP1t", is weakly dense in 23 (1i)]R, rather
than coinciding with it), the Poisson structure in each sector ofP is determined up to
a constant, which implies 3.9.2.3. By Lemma 3.8.3 and Axiom C* 1 (which, through
the definition of a uniform Poisson space, requires that!2l]R be closed under Poisson
brackets) we can rescale the Poisson bracket so as to make Planck's "constant"
a constant on P. By Proposition 3.4.1 the Leibniz rule (1.4) is then satisfied as
a consequence of the unitarity imposed by Axiom C* I. By the remark after the
proof of 3.4.1, the associator identity (1.6) holds for the rescaled Poisson bracket.
Hence !2l]R becomes a J L B-algebra by Definition 1.1.5, and the complexification
!2l is a C* -algebra by Proposition 1.1.9.
From 3.9.2.1 and 3.9.2.2 we infer that !2l]R(P) = !2l]R(P(!2l». The w*-uniformity
appearing in (3.4) is the weakest uniformity relative to which all elements of!2l]R
are uniformly continuous. Property 3.9.2.4 then follows from Theorem 3.2.1 and
(3.26). Property 3.9.2.5 is evident from Proposition 3.3.3.
Finally, let us assume that some IP1t", have an exotic manifold structure such that
!2l]R(P) n Cu(P, R) is not contained in C'{'(P, R), so that!2l]R C !2l]R(p)nCII (p, R)
is a proper inclusion (rather than the equality (3.26». It follows from Axiom C*5
that the weak density mentioned two paragraphs ago must still hold. This weak
density suffices for the subsequent arguments to be valid, and we can construct
a C*-algebra!2l with pure state space P. The proper inclusion above would then
contradict (3.4). Hence such exotic manifold structures are excluded by the axioms
(if they exist at all). •

Certain simplifications of this characterization suggest themselves. For example,

if one amends C* l-C* 5 in Theorem 3.9.1 by deleting the word "uniform" from C* 1
and replacing (3.4) in C*3 by (3.7), then Theorem 3.9.2 is correct if one replaces
"( w* -) uniformity" by "( w* -) topology" and "C* -algebra" by "perfect C* -algebra"
(cf. 3.2). Greater simplification is achieved by imposing finite-dimensionality on
P (as a transition probability space, cf. 2.7):

Corollary 3.9.3. The pure state space P of a finite-dimensional C* -algebra is

characterized by the following properties:

QM 1: P is a finite-dimensional Poisson space with a transition probability.

QM2: P has the two-sphere property (Definition 3.7).
QM3: the sectors ofP as a transition probability space coincide with the symplectic
leaves ofP as a Poisson space.

Compare this with the characterization of classical mechanics:

 3 From Pure States to Observables 107

Corollary 3.9.4. The pure state space P of a classical mechanical system is

characterized by the following properties:
CM1: P is a smooth Poisson space with a transition probability.
CM2: The transition probabilities are p(p, a) = 8p(J'
CHAPTER II

Quantization and
the Classical Limit

1 Foundations
1.1 Strict Quantization of Observables
The aim of quantization theory as presented in this book is to relate Poisson algebras
or Poisson manifolds to C* -algebras or their pure state spaces. A slightly awkward
feature of the first relationship is that usually Poisson algebras are not Banach
spaces; a nonzero Poisson bracket on some Poisson subalgebra 2t~ of Cf'(P, JR)
cannot be extended to the closure ~~ of 2t~ in the sup-norm.
Apart from this complication, the following definition is largely motivated by
Theorem 1.1.1.9; in particular, recall 1.(1.22).

Definition 1.1.1. A strict quantization of a Poisson algebra 2t~ (which is dense ly

contained in the selfadjoint part ~~ ofa commutative C* -algebra ~o ) consists of
a collection ofpoints 10 s;: JR that has 0 ¢ 10 as an accumulation point (we write
I := 10 U {OJ), a collection ofC*-algebras {~lilnE[, and a collection of linear
maps {QIi : 2t~ --+ ~~}fjEJ (where Qo is the identity map), such that the following
conditions hold:

1. Rieffel's condition: For all f E 2t~, the function Ii f--+ II Qfj(f) II is continuous
on I. In particular, one has

lim
h~O
IIQh(f)1I = 11111· (1.1)

2. von Neumann's condition: For all f, g E 2t~ one has

(1.2)
 1 Foundations \09

3. Dirac's condition: For all I, g E 2i~ one has

lim II{Qh(f), Q/i(g)}/i - Q/i({f, g})11 = O. (1.3)
/i->O
4. Completeness: The collection {Q/i(f) liE 2i~} is dense in m~/or each n E I.
A strict quantization of a Poisson manifold P is a strict quantization 01 some
Poisson subalgebra 2i~ oIC'b(P, JR.), equipped with the sup-norm 11111 = 1111100,
whose closure contains Co(P, JR.).
For a given classical observable 1 E 2io, one construes the operator Qr,(f) as
the quantum observable (at the given value of n) whose physical interpretation
corresponds to that of its classical counterpart I.
We have suppressed the possible n-dependence of the C* -algebraic operations
in mh in our notation. The completeness condition is not crucial: If it fails to
be satisfied for a given m/i, one may simply replace mil in the definition by the
C*-algebra generated by all Q/i(f).
We may extend Q/i to a map (denoted by the same symbol) from 2i0 to mil by
complex linearity. Conditions 1.1.1.2 and 1.1.1.3 then imply that
lim IIQ,,(f)Q/i(g) - Qh(fg)1I = O. (1.4)
/i->O
Definition 1.1.2. A strict quantization (mil, {QIl}) is called a strict deforma-
tion quantization when Q/i(2io) is closed under multiplication (in mh) and Q/i is
nondegeneratelor each nin that QIl(f) = 0 iff 1 = O.
Note that Rieffel's condition implies that a strict quantization is always non-
degenerate for small enough n, so the last requirement is a modest one. The
terminology is justified by the fact that a strict deformation quantization of 2i~
allows one to define an associative "deformed" product· /i in 2i0 with the property
Q/i(f)Q/i(g) = Q,,(f '/i g) (and, of course, 1 '0 g := Ig). The conditions on
a strict quantization may then be rephrased in terms of this product in the obvi-
ous way. There are many examples of strict quantization that are not deformation
quantizations, in particular those related to pure states (see 1.3).
The maps Q/i are highly nonunique, depending on what physicists call
an operator-ordering prescription. Hence two strict quantizations (m~, {Q~}),
(m~, {Q~}), where m~ = m~ for all n, are called equivalent if for each 1 E 2i~ the
function
n~ IIQ~(f) - Q~(f)1I
is continuous on I. It follows that lim/i->o II QA(f) - Q~(f)1I = O. In the next
section we will construct an object from a given strict deformation quantization
that is invariant under changes to equivalent quantizations.
A strict (deformation) quantization is called positive if each Q" is positive
(that is, 1 :::: 0 in m~ implies Q/i(f) :::: 0 in m~). In many physically relevant
applications, including the premier example ofWeyl quantization, the quantization
fails to be positive. However, a nonpositive quantization is sometimes equivalent
110 II. Quantization and the Classical Limit

to a positive one (cf. 2.6.3). If a positive quantization can be extended from Q(~ to
Qt~ (where the property (1.3) is evidently lost) such that it remains positive, the
maps Q" : Qto ~ Qth are automatically continuous; see 1.1.3.7.

1.2 Continuous Fields of C* -Algebras

The notion of strict quantization is closely related to an object intrinsic to the
theory of C* -algebras.
Definition 1.2.1. A continuous field of C*-algebras (\t, (Qtx, CPx }XEX) over a
locally compact Hausdorff space X consists of a C* -algebra It, a collection of
C*-algebras {QtxhEX' and a set (cpx : \t ~ QtX}XEX of surjective morphisms, such
that:

1. The function x f--+ IIcpAA)1I is in CoCX) for all A E \to

2. The norm of any A E \t is IIAII = SUPXEX IIcpxCA) II·
3. For any f E Co(X) and A E \t there is an element fA E \tfor which cpAf A) =
fC x)cpxC A) for all x E X.

A section of the field is an element {Ax }XEX of OXEX Qtx for which there is an
A E \t such that Ax = cpxCA)for all x E X.
It is clear that \t may be identified with the space of sections of the field, seen
as a C* -algebra under pointwise scalar multiplication, addition, adjointing, and
operator multiplication, by means of (CPx(A)}xEX ++ A. In particular, A = B iff
CPx(A) = cpAB) for all x.
The simplest example is obtained by taking Qtx = Q( for all x, and letting
\t = Co(X, Qt) with CPx(A):= Ax. Such a field is called trivial.
Lemma 1.2.2. The C* -algebra \t of (sections of) a continuous field is locally
uniformly closed. That is, if A E Ox Qtx is such that for every y E X and every
E > 0 there exists a BYE \t and a neighborhood NY of y in which II A x - BJ II < E
for all x E NY, and also limx-+oo IIAxll = 0, then A E \to
Alternatively, if the junction x f--+ II Ax - Cx II lies in Co(X) for each C E It,
then A E \to
Inthesituationofthefirstpart,thereisacompactsetK S; X for which II Ax II < f
outside K, as well as a finite cover {NX 1 , ••• , NXn} of K. Taking a partition of
unity {ud on K subordinate to this cover, the operator B := Li ujW' lies in \t
because of 1.2.1.3, and satisfies SUPXEX IIAx - Bx II < f. Hence A E \t by 1.2.1.2
and the completeness of \to
Given any A E Ox Qtx and y EX, because CPy is surjective there is a BYE \t such
that Ay = B~. The assumption in the second part then implies that the conditions
in the first part are satisfied, such that A E \to •
Proposition 1.2.3. Suppose one has a family {QtX}XEX ofC*-algebras indexed by
a locally compact Hausdorff space X, as well as a subset it S; Ox Q(x that satisfies
the following conditions:
 1 Foundations III

1. The set {Ax I A E ~} is dense in mx for each x E X.

2. Thefunction x 1-+ II Ax II is in Co(X) for each A E ~.
3. The set ~ is a * -algebra (under pointwise operations).

There exists a unique continuous field of C* -algebras (It, {mX, CPx }xeX) whose
collection of sections contains ~. Namely, It consists of all A E Ox mx for which
the function x 1-+ IIAx - Cxlllies in CO(X) for each C E ~,regarded as a C*-
algebra under pointwise operations, and the norm ofl.2.1.2. Finally, CPx(A) := Ax
is the evaluation map.

We first show that It as defined above is locally unifonnly closed. With the
objects A, y, E, BY, and N as specified in Lemma 1.2.2, take C E ~ arbitrary, and
define the functions fAC : x 1-+ IlAx - Cx II and fBC : x 1-+ II BI - CxII. Using the
general Banach space inequality

I(IIXII -IIYI!)I ~ IIX - n, (1.5)

one obtains IfAC(x) - fBc(X) I < E for all x EN. By assumption, /Bc is con-
tinuous, so that IfBc(X) - f Bc(Y) I < E for all x in some neighborhood N' of y.
Combining the two inequalities yields IfAc(X)- fAc(Y) I < 3E for all X E NnN'.
Hence fAc is continuous at y, which was arbitrary, so that A E It by definition of
It.
Using this property, it is easily shown that It is a C* -algebra, and that condition 3
in Definition 1.2.1 is satisfied. It is clear from 1.2.1.1 and the definition of It in
1.2.3 that It is maximal. On the other hand, according to the second part of Lemma
1.2.2, It is minimal, so that it is unique. •

We are now in a position to connect Definitions 1.1.1 and 1.2.1.

Theorem 1.2.4. Suppose one has a strict quantization of a Poisson algebra m~,
except perhaps for (1.3). When I is not compact, the function Ii 1-+ IIQfi(f)1I is
assumed to be in Co(I) for all f E mO. Furthermore, assume that either I is
discrete, or that all mfi are identical for Ii =1= 0 and the function Ii 1-+ Qfi(f) is
continuous for all f E mO.
There exists a unique continuous field of C*-algebras (It, {m fi , CPfi}fief) whose
collection of sections {cpfi(A)} fief, A E It, contains all {Qfi(f)}fief, f E mO.
Moreover, any strict quantization equivalent to the given one leads to the same
continuous field.

One defines ~ C Ofi mfi as the complex linear span of all expressions of the
fonn Ii 1-+ Qfi(fd··· Qfi(fn), where Ii E mO. We first show that each function
of the type Ii 1-+ II Qfi(!1) ... Qfi(fn)1I is continuous. It follows from (1.4) that
limfi-->o II Qfi(fdQ/i(f2 ... fn) - Q/i(fl ... fn)1I = 0, so that by induction one has
(1.6)

Equation (1.5) then yields limfi-->O II Q/i(!1) . .. Qfi(fn)1I - II Q/i(!1 ... fn)1I = 0,
so that, finally, limfi-->O II Q/i(!1) . .. Q/i(fn)1I = IIfl ... fn II by (1.1). This proves
112 II. Quantization and the Classical Limit

continuity at Ii = O. Using (1.6) and the continuity of each function Ii I--? Qh(f),
the same result follows for polynomials in the QhC!;).
When I is discrete, continuity away from 0 is trivial. In the alternative case, for
monomials this follows from an inductive argument based on the inequalities
I II Qh(ft) ... Qh(fn) II - II Qh,(ft> '" Qh,(fn)1I1
~ IIQh(ft)··· Qh(fn) - Qh/(ft)··· Qh/(fn)1I
~ II Qh(ft) - Qh/(ft) II II Qh(h) ... Qh(fn)1I
+ II Qh/(ft) II II Qh(h) ... Qh(fn) - Qn:(h)'" Qh'(fn)lI. (1.7)
The extension of this argument to polynomials is a trivial application of the
triangle inequality. Since condition 1.2.3.2 is evidently satisfied, one is therefore
in the situation of Proposition 1.2.3, and the first claim follows. The second is clear
from the proof of 1.2.3 and the definition of equivalent quantizations. •
If one wishes to take Definition 1.2.1 as a canonical starting point of the theory
of quantization, one might contemplate the following definition of quantization
(specialized to the case of Poisson manifolds, for simplicity).
Definition 1.2.5. Let I £; 1R contain 0 as an accumulation point. A continuous
quantization of a Poisson manifold P consists of
1. A continuous field ofC*-algebras (It, {Qth, <Ph}hE/).
2. A Poisson subalgebra §to ofC~(P) whose closure Qto contains Co(P).
3. A linear map Q : §to --? It that with Qh(f) := <Ph(Q(f»for all f E §to and
Ii E I satisfies Qo(f) = f and Qh(f*) = Qh(f)*, and for all f, g E §to
satisfies Dirac's condition (1.3).
Provided that 1.1.1.4 is satisfied, a continuous quantization is strict. Conversely,
Proposition 1.2.4 gives conditions, which will be satisfied in all examples in this
book, under which a strict quantization is continuous.

1.3 Coherent States and Berezin Quantization

Having introduced quantization theory from the point of view of observables, we
now look at quantization from the dual perspective of pure states. Recall 1.2.1.
Definition 1.3.1. Relative to a continuous field of C* -algebras (It, {Qtx, <Px }xEK ),
a continuous field of (pure) states is afamily {w~}~~~, where each w~ is a (pure)
state on Qtx, and A is an index set, such that
1. For each.le E A and A E It the function x I--? w~(Ax) lies in Co(X).
2. For each x E X the collection {W~}AEA isfaithful, in that n AEA ker(lTc.f.) = 0,
where IT(OI.(QtX ) is the GNS-representation defined by w~ (in other words, the
represent~tion EEl AEA IT x (QtX) is faithful).
(01.

In the context of quantization theory, the following result allows one to construct
continuous fields of pure states by checking a simple condition.
 1 Foundations 113

Lemma 1.3.2. Under the assumptions of Theorem 1.2.4, suppose for each Ii E I
one has a state Wh on ~h such that the function Ii f-+ Wh(Qh(f» is continuous on
I for each f E §lo. Then Ii f-+ wh(Ah) is continuous on I for all A E \to

We first assume that A = Qh(ft)··· Qh(fn) (the extension to polynomials is

trivial). Continuity at Ii = 0 is an immediate consequence of (1.6) and the as-
sumption in 1.3.2. Away from 0, one uses the completeness assumption 1.1.1.4 to
approximate Q/i,(ft)· .. Qh(fn) by Qh(f), and notes that the proof of 1.2.4 estab-
lishes the continuity of Ii f-+ Qh(ft)··· Qh(fn); simply omit the first inequality in
(1.7). Combined with the continuity of Ii f-+ Wh(Qh(f», this does the job.
Finally, if A is as specified in the last paragraph of 1.2.3, one uses the last
sentence in the proof of 1.2.2, from which the result trivially follows. •
Given a continuous quantization of a Poisson manifold P (cf. 1.2.5), with ~o =
Co(P), itis natural to take A = P and wg (f) = f(a) for alIa E P. Writingqh(a)
for wI:., one may then look at qh as a map that "quantizes" classical pure states.
Such maps may be studied in their own right, even in the absence of a continuous
quantization of P. We will do so in the special case that P = S is a symplectic
manifold of dimension 2n < 00. One may then anticipate that ~h = 230(7th).
In what follows, the transition probability p is the standard one defined on a
projective Hilbert space, given by 1.(2.65). The canonical symplectic form on IP1t
is denoted by W1f. (cf.1.2.5). A measure on a manifold is said to be locally Lebesgue
if it is equivalent to Lebesgue measure in each local chart.

Definition 1.3.3. Let 10 £ R be as in 1.1.1. In a pure state quantization of a

symplectic manifold (S, ws) one specifies, for each Ii E 10 , a separable Hilbert
space 7th, a smooth injection qh : S ~ IP1th (cf 1.2.5.1), and a Radon measure
J.th on S that is locally Lebesgue, such that

1. for all Ii E 10 and alil/J E 1P1t/i one has

Is dJ.th(a) p(qh(a),l/J) = 1; (1.8)

2. for allfixed f E Cc(S) and PES, the function

Ii f-+ Is dJ.th(a) p(qh(P), qh(a»f(a)

is continuous on 10 and satisfies

lim
h-+O Jsf dJ.t/i(a) P(qh(P), qh(a»f(a) = f(p); (1.9)

3. the map qh is an approximate symplectomorphism, in that (pointwise)

lim qhW1i = Ws. (1.10)

h-+O

Since f E Cc(S), the requirement 1.3.3.2 is equivalent to the continuity of the

function Ii f-+ p(qh(p), qh(a» for fixed P and a. Moreover, we will shortly see
114 II. Quantization and the Classical Limit

that (1.9) and (1.8) imply the conceptually pleasing result

lim p(qli(P), qli(a»
Ii--'>O
= 8pu • (1.11)

The (over) completeness condition (1.8) should be compared with 1.(2.56), but
note that elements of a basis of a transition probability space are by definition
orthogonal, whereas the family {qli (p) I pES} becomes approximately orthogonal
only in the limit limli--'>o, as guaranteed by (1.11).
Combining 1.(2.56), applied to the transition probability space JPYHIi, with (l.8),
the volume volli (S) of S with respect to I-Lli is found to be
volli (S) = dim (71.1i). (1.12)
In all examples in this book, (1.10) holds without the limit for alln E 10 • In
addition, the measure I-Lli will always be of the form
(1.13)
where c : 10 --+ 1R\{0} is some positive continuous function, and the Liouville
measure I-L L on S is defined by
1 f n
(1.14)
I-LLCf) := (21T)nn! Js fw s ·

The Liouville measure stands out by its invariance under any Hamiltonian flow, as
Proposition 1.2.3.3 implies that I-LLCf) = I-LLCap(f» for all t; cf. 1.(2.13).
It is clear from (1.13) and (1.12) that 71.1i is finite-dimensional iff S is compact,
n
and that only certain discrete values of are allowed in that case. As 0 :5 p(., .) :5 1,
eq. (1.8) then implies limli--'>o c(1i) = 00, so that limli--'>o dim (71.1i) = 00.
A pure state quantization naturally leads to the quantization of observables.
Definition 1.3.4. Let {71.1i, qli, I-LlilliEio be a pure state quantization ofa symplectic
manifold S. The Berezin quantization of a function f E L OO(S) is the family of
operators {Qff(f)}IiEio ' where Qff(f) E r.13(71.1i) is defined by polarizing

it(Q~(f»:= 1 dl-Lli(a) p(qli(a), it)f(a). (1.15)

Here it E JPYHIi; the integral converges because of(1.8}.

Here LOO(S) is defined with respect to any locally Lebesgue measure, such as
I-LL. If Qff
takes values in r.13 0 (1i1i), the left-hand side coincides with the Gelfand
transform of Qff(f) evaluated at it, namely Qf(j)(it). Iff E L '(S, I-Lh)nLOO(S),
the operator Qff
(f) may be written as a Bochner integral

Qff(f) = 1 dl-L,,(a)f(a)[q,,(a)], (1.16)

where [q,,(a)] is the projection onto the one-dimensional subspace in 11." whose
image in JPYHIi is qli(a). A number of properties of Qff
are immediately evident.
Most trivially, (1.9) may be rewritten as
lim qli(p)(Q~(f) = f(p). (1.17)
Ii.... 0
 1 Foundations 115

This leads to (l.11), as follows. According to Urysohn's lemma, there is a function

f E Co(S, JR) such that IIflloo = 1, and f(p) = - f(a) = 1. From 1.(2.67) and
(1.19) below we infer
IqJj(p)(Q~(f» - qJj(a)(Q~(f» I ~ 2J1 - p(qJj(p), q,,(a» ~ 2.

Letting h ~ 0, eq. (1.11) then follows from (1.17) and 1.(2.53).

Theorem 1.3.5. Assume that f E Loo(S, JR). Then:

• Q~ is positive (that is, f 2: 0 almost everywhere on S implies Qf(f) 2: 0 in

1J3(1tJj»).
• Q~ (f) is selfadjoint.
• Iff E LI(S, /lll). then Q~(f) E 1"J3 1(1th) (i.e .• Q~(f) is of trace-class). with

Tr Q~(f) = /lJj(f) = Is d/lh(a)f(a). (1.18)

• The operator Q~ (f) is bounded by

IIQf(f)1I :s IIflloo. (1.19)

• If f E Co(S). then Qf(f) E lJ3o(1tJj) (i.e .• Qf(f) is compact). and Q~ :

Co(S) ~ 1J3 0 (1th) is continuous.

Positivity and self-adjointness are obvious from (1.15). To show that Qf (f)
is trace-class for f E LI(S, /lit) n Loo(S), we first assume f 2: O. Then
Qf (f) 2: 0, so that the trace norm is II Qf (f) III = Tr Qf (f). Choose a basis
{en} in 1t1l. Then I::=I(en , [qh(a)]en) = I::=I p(qll(a), en) :s 1 for N < 00.
Since f E LI(S, /lh), the monotone convergence theorem says that Tr Q~(f) ex-
istsandequalsjsd/lh(a)f(a). Thus IIQ~(f)111 = /lr!(f) for f 2: O.Forarbitrary
f we write f = fl - 12, with fl' h 2: 0 a.e. Hence II Qf(f) II 1 :s 00; linearity
of the trace then yields (1.18).
The conclusion from (1.15) that for f E L 00 (S) the operator Q~ (f) is bounded,
with bound (1.19), uses the following (slightly more general) argument. Let A
be a symmetric operator such that 1(1lJ, AIlJ)1 :s cllllJ1I2 for some c > 0 and for
all IlJ in its domain. One then replaces IlJ by IlJ ± A IlJ / c, and subtracts the two
inequalities thus obtained. This implies the inequality II A IlJII :s c II 1lJ1I, showing
that A is bounded with norm:s c. This argument with (1.8) implies (1.19).
Finally, the last claim follows from the second and the third: Start with f E
Cc(S), and use (1.19). •
There is a clear intuitive connection between the respective conditions 1.1.1.1,
1.1.1.3 on the observable side, and 1.3.3.2, 1.3.3.3 on the pure state side. More-
over, 1.1.1.2 is closely related to (1.11). For the latter equation implies that the
projections [qll(a)] in (1.16) become approximately orthogonal as h ~ 0, so that
the integral should approximate the spectral resolution of Q~ (f). This implies that
Q~(f)2 should approach Q~(f2) for small h (cf. 1.3.3), which is the essence of
von Neumann's condition.
116 II. Quantization and the Classical Limit

On the other hand, the completeness conditions 1.1.1.4 and 1.3.3.1 are not
related. Even if a Berezin quantization satisfies 1.1.1.1-4, it may not define a strict
deformation quantization. On the positive side, we have
Proposition 1.3.6. Let f E Co(S), and assume that 1ih is independent of 1i
whenever 1i varies through a connected subset of10 • Then Rieffel's condition holds;
in particular,
(1.20)

We initially assume that f E Cc(S), and at the end extend the result to f E
Co(S, R) using the continuity of Qg. The function 1i 1-+ Qg(f) from any given
connected subset of 10 to 23(1ih) is continuous with respect to the trace norm,
hence certainly relative to the operator norm on 23(1ih). Therefore, 1i 1-+ II Qg (f) II
is continuous on 10 by (1.5).
To prove (1.1), note that (1.19) implies
lim sup II Qg (f) II ~ IIflloo. (1.21)
h-...O
On the other hand, for f E Co(S) we can find PES for which IIflloo = If(p)l.
By (1.9) and the obvious inequality II Qg(f) II :::: Iqh(p)(Qg(f))l, we have
lim inf IIQg(f)1I :::: IIflloo. (1.22)
h-...O
Hence (1.20) follows.
In the examples in this book, the Berezin quantizations constructed from certain
•
pure state quantizations do satisfy all of 1.1.1.1-4. Unfortunately, the proofs of
1.1.1.2-4 seem to involve special features of these examples.
Corollary 1.3.7. In the situation of Definitions 1.3.3 and 1.3.4, suppose that 1ih
is independent of 1i whenever 1i varies through a connected subset of 10 , and that
the Berezin quantization map Qg, defined on ~o = Co(S), satisfies 1.1.1.2.
The collection {wh }h::l, where wh := qh(a), is a continuous field ofpure states
(cj. 1.3.1) relative to the continuous field of C* -algebras of Theorem 1.2.4.
It is clear from Proposition 1.3.6 and its proof that the assumptions of Theorem
1.2.4 hold. Condition 1.3.3.2 implies that the assumption in 1.3.2 is met. Finally,
(1.8) implies that the faithfulness assumption in 1.3.1, where ~x = ~h = 'BO(1ih),
is satisfied for 1i =f:. O. Hence the claim follows from Lemma 1.3.2. •
This corollary applies to all pure state and Berezin quantizations considered in
this book.

1.4 Complete Positivity

Theorem 1.3.5 shows that Qg is a positive linear map from Co(S) into 'BO(1ih)'
It has, in fact, a stronger positivity property. The study of this property is further
motivated by the idea that a positive map Q (cf. Definition 1.3.6) generalizes the
 1 Foundations 117

notion of a state, in that the C in w : ~ ~ C is replaced by a general C* -algebra

~ in Q : ~ ~ ~. One would like to see whether one can generalize the GNS-
construction, and it turns out that for this purpose one needs to impose the stronger
positivity property in question.
For a given C*-algebra~, and n E N, we first introduce the C*-algebra VRn(~)'
The elements of VRn(~) are n x n matrices MI with entries in ~; multiplication
is done in the usual way, i.e, (MIN)ij := MIikNkj. with the difference that one
now multiplies elements of ~ rather than complex numbers. In particular, the
order has to be taken into account. The involution in VR n (~) is, of course, given by
(MI*)ij = MIji' in which the involution in ~ replaces the usual complex conjugation
in C. One may identify VRn(~) with ~ ® VRn(C) in the obvious way.
When 1f is a faithful representation of ~ (which exists by Theorem 1.1.1.8), one
obtains a faithful realization 1fn of VRn(~) on 11. ® cn , defined by linear extension of
1fn(MI)Vi := 1f(MIij)Vj; we here look at elements of11.®cn asn-tuples(v" ... , vn),
where each Vi E 11.. The norm IlMIIi of MI E VRn(~) is then simply defined to be
the norm of 1fn (MI). Since 1fn(VRn(~» is a closed *-algebrain ~(11.®cn) (because
n < (0), it is obvious that VR n (~) is a C* -algebra in this norm. The norm is unique
by I. 1.2.4.4, so that its definition does not depend on the choice of 1f .
Definition 1.4.1. Given a linear map Q : ~ ~ ~ between C* -algebras ~ and
~,and n EN, define the map Qn : VRn(~) ~ VRn(ll~) by (Qn(lW»ij := Q(MIij).
In other words, seen as a map from ~ ® VRn(C) to 'B ® VRn(C), one defines Qn
by linear extension of Q ® id on elementary tensors.
A linear map Q : ~ ~ 'B between C* -algebras is called completely positive
if Qn is positive for all n E N.
The point is now that completely positive maps that in addition are normalized
(like a state) have a generic structure, which is of central importance for quantiza-
tion theory. Recall that a partial isometry is a linear map W : 11., ~ 11.2 between
two Hilbert spaces, with the property that 1t, contains a closed subspace K, such
that (W'l1, W<I>h = ('l1, <1», for all \11, <I> E K 1, and W = 0 on Kt. Hence W is
unitary from K, to WK1.1t follows that W*W = [Ktl and WW* = [K 2 ], where
K2 is the image of W, are projections.
Theorem 1.4.2. Let Q : ~ ~ ~ be a completely positive map between C*-
algebras with unit, such that Q(ll) = ll. By Theorem 1.1.1.8, we may assume that
~ is faithfully represented as a subalgebra 1fx(~) ~ ~(1tx),for some Hilbert
space 1t x .
There exists a Hilbert space 1t x , a representation 1fX of ~ on 1t x , and a partial
isometry W : 1t x ---+ 1t x (with W*W = ll) such that
1f x (Q(A» = W*1fX(A)W (1.23)

for all A E ~. With P := WW* (the targetprojectionofW on 1t X), itx := p1t x =

W1tx c 1t x , and U : 1t x ---+ itx defined as W, seen as a unitary mapfrom 1t x
to it x , one has the equivalent relation
(1.24)
118 II. Quantization and the Classical Limit

The proof consists in a modification of the GNS-construction (cf. 1.1.5). We

denote elements of1t x by v, w, with inner product (v, who
Construction 1.4.3.
1. Define the sesquilinear form (, )~ on Q( ® 1t x (algebraic tensor product) by
(sesqui- )linear extension of
(A ® v, B ® w)~ := (v, Jrx(Q(A* B»wh. (1.25)
Since Q is completely positive, this form is positive semidefinite; denote its null
space byNx '
2. The form (, )~ projects to an inner product (, )X on Q( ® 1t x / N x · If Vx :
Q( ® 1t x ~ Q( ® 1t x /Nx is the canonical projection, then by definition
(1.26)
The Hilbert space 1t x is the closure ofQ( ® 1t x / N x in this inner product.
3. The representation Jr x (Q() is initially defined on Q(® 1t x / N x by linear extension
of
JrX(A)Vx(B ® w):= Vx(AB ® w); (1.27)
this is well-defined, because Q( ® IIxNx ~ N x . One has the bound
IIJrX(A)11 ~ IIAII, (1.28)
so that Jr x (A) may be defined on all of 1t x by continuous extension of ( 1.27).
This extension is a representation of Q( on 1tx.
4. The map W : 1t x ~ 1t x , defined by
(1.29)
is a partial isometry. Its adjoint W* : 1t x ~ 1t x is given by (continuous
extension of)
W*VxA ® v = Jrx(Q(A»v, (1.30)
from which the properties W' W = II and (1.23) follow.
We now prove the various claims made in this construction. Firstly, to show that
the form defined by (1.25) is positive, we write
L(Ai ® Vi, Aj ® Vj)& = L(vi, Jr x (Q(A7 Aj»vj)x' (1.31)
i.j i.j
Now consider the element A of!mn(Q() with matrix elements Aij = A7 A j . Taking
a faithful representation Jr (Q(), from which one constructs Jrn (!mn (Q(» as explained
above 1.4.1, one sees that
(z, Jrn(A)z) = L(Zi, Jr(A7 Aj)zj) = L(Jr(Ai)Zi, Jr(Aj)Zj) = IIAzII2 ::: 0,
i.j i.j
where Az := Li Jr(Ai)Zi. Hence A::: 0. Since Q is completely positive, it must
be that $, defined by its matrix elements $ij := Q(A7 Aj), is positive in !mn(lB).
 1 Foundations 119

Repeating the above argument with A and 7f replaced by la and 7f x' respectively,
one concludes that the right-hand side of (1.31) is positive.
It follows from (1.25) that (C®v, AB®w)~ = (A*C®v, B®w)~,sothatA®][x
leaves N x stable; compare the corresponding argument for the GNS-construction
based on 1.(1.58).
To prove (1.28) one uses 1.(1.42) in VJtn(~)' Namely, for an arbitrary collection
A, B I, ... , Bn E ~ we conjugate 0 ~ A * A][n ~ II A 112][n with the matrix la, whose
first row is (BI •...• Bn). and which has zeros everywhere else; the adjoint la* is
then the matrix whose first column is (Bi • ...• B:)T. and all other entries zero.
This leads to 0 ~ la* A* Ala ~ IIAII 21a*1a. Since Q is completely positive. one
has Qn (la* A *Ala) ~ II A 112 Qn (la*la). Hence in any representation 7f x(~) and any
vector (v[, ... vn) E Xx ® en one has

(1.32)
i,j i,j

With \II = Li Vx Bi ® Vi, from (1.25), (1.27), and (1.32) one then has

I17fX(A)\II1I2 ~ IIAII2~)Vi.7fX(Q(BtBj»vj>X = IIAII211\11112.

i,j
To show that W is a partial isometry one merely uses (1.29). (1.25). and Q(][) = ][.
Equation (1.30) is then trivially verified from the defining property (w. W*\II)x =
(Ww. \II)X for all WE 1t x and \II E 1tx.
To verify (1.23), one uses (1.29) and (1.30). Since W is a partial isometry, one
has p = W W* for the projection p onto the image of W, and in this case, W* W = ][
for the projection onto the subspace of 1t x on which W is isometric; this subspace
is 1tx itself. Hence (1.24) follows from (1.23), since
U7f x (Q(A»U- 1 = W7f x (Q(A»W* = WW*7fX(A)WW* = p7fX(A)p. •

When Q fails to preserve the unit. the above construction still applies. but W is
no longer a partial isometry; one rather has II WII 2 = II Q(][)II. Thus it is no longer
possible to regard 1t x as a subspace of 1tx.
If ~ and perhaps ~ are nonunital, the theorem holds if Q can be extended (as a
positive map) to the unitization of ~ (cf. 1.1.2.1). such that the extension preserves
the unit ][ (perhaps relative to the unitization of ~). When the extension exists but
does not preserve the unit, one is in the situation of the previous paragraph.
The relevance of all this to Berezin quantization is as follows.
Proposition 1.4.4. A positive map between a commutative unital C* -algebra and
a C* -algebra is completely positive.
We write Q : ~ ~ ~ for the map in question. By Theorem 1.1.2.3 one has
~ = C(X) for some compact Hausdorff space X. We may then identify VJtn(C(X»
with C(X. VJtn(C». Take G E C(X, VJtn(C» and pick E > O. Since X is compact,
there is a finite collection of points Xl • •••• Xn and a finite cover {O!" ...• O!/}
with the property that IIG(Xi) - G(x)1I < E for all X E O!i' Using a partition
of unity {ud subordinate to this cover. one constructs F, E C(X, VJtn(C» by
120 II. Quantization and the Classical Limit

Ft(x) := L:=l Ui(X)G(Xj). One then has 11Ft - Gil < E. Hence elements of the
form F, where F(x) = Li fi(x)Mi for fi E C(X) and M j E 91tn(C), and the sum
is finite, are dense in C(X, 91tn(C».
It is easily seen that such F is positive iff all fi and Mi are positive, so that positive
elements G ofC(X, 91tn(C» can be approximated by positive F's. On such F, one
has Qn(F) = Li Q(fi) OS) Mi. Now, each operator Bi OS) M is positive in 91tn(~)
when Bi and M are positive (as can be checked in a faithful representation). Since Q
is positive, it follows that Qn maps each positive element of the form F = Li fi Mi
into a positive member of 91tn(~).
We know from 1.1.3.7 that Q is continuous; the continuity of Qn follows because
n < 00. A norm-limit A = limn An of positive elements in aC*-algebra is positive,
because by 1.( 1.39) we have An = B; B n , and lim Bn = B exists because of
1.(1.15). Finally, A = B* B by continuity of multiplication, i.e., by 1.(1.14). Hence
if Fk -+ G :::: 0 in C(X, 91tn(C», then Qn(G) = limk Qn(Fd is a norm-limit of
positive elements, which is positive. •
The application to Berezin quantization is obvious from 1.1.3.5 and 1.4.4: We
take 2l = Co(S), 23 = 230 ('Jih), 'Ji x = 'Jir., and Q = Qg. Theorem 1.4.2 then
applies, for we can extend Qg
to the unitization Co(S)n of Co(S) (which consists
of all functions of the form f + Al s, f E Co(S) and A E C) by linear extension of
(1.15). Since Co(S) + CIs c LOO(S), this extension is still positive by Theorem
U.S, and satisfies Qg(1s) = If because of (1.8).
Corollary 1.4.5. The image Qg(Co(S» is closed in ~o('Jir.). In particular, if
Qg(C~(S» is dense in 23o('Jir.), then Qg(Co(S» = 23o('Jir.).
Taking rrx as in the proof of 1.4.2, the image rrX(Co(S» is closed by Theorem
1.1.3.10.4, so that prrX(Co(S»p = Qg(Co(S» is closed as well. •
In the opposite direction, one may ask whether a given positive map Q can be
written in a form similar to (1.16).
Proposition 1.4.6. Let Q : Co(S) -+ 23('Ji) be positive (where S is a locally
compact Hausdorffspace), and such that Q(f) E ~l('Ji)forali f E Cc(S), Then
there exists a regular Borel measure J1, on S and a (weakly) measurable family
a ~ p(a) of density matrices, such that (weakly)

Q(f) = Is dJ1,(a)f(a)p(a). (1.33)

Given the assumptions, the map f ~ Tr Q(f) defines a positive linear func-
tional, which by the Riesz representation theorem corresponds to a positive regular
Borel measure J1, on S. Also, for each unit vector \II E 'Ji we obtain a posi-
tive linear functional f ~ (\II, Q(f)\II), hence a positive regular Borel measure
J1,q, on S. Since (\II, Q(f)\II) :::: Tr Q(f), we see that J1,q, is absolutely con-
tinuous with respect to J1,. Hence we obtain the Radon-Nikodym derivatives
pq,(a) := dJ1,q,/dJ1,(a), and subsequently the operators p(a) by polarization.
Equation (1.33) follows by construction; the claimed properties of the p(a) are
then obvious. •
 1 Foundations 121

Obviously, the given condition is always met if 'H is finite-dimensional. It

remains to be investigated whether each p(a) is a one-dimensional projection,
and if so, whether the ensuing map S ~ Jnl is smooth. There is, however,
a generalization of (1.33) that applies to any positive map on a commutative
C* -algebra.
Definition 1.4.7. Let X be a set with a a-algebra :E of subsets of X. A positive-
operator-valued measure, or POVM, on X in a Hilbert space 'H is a map Il ~
A(ll)from :E to ~('H)+ (the set ofpositive operators on 'H), satisfying A(0) = 0,
A(X) = n, and A(U;Il;) = L; A(Il;) for any countable collection of disjoint
Il; :E (where the infinite sum is taken in the weak operator topology).
E
A projection-valued measure, or PVM, is a POVM that in addition satisfies
A(Il, n 1l 2 ) = A(1l 1)A(1l 2 )for all Il" 112 E :E.
Note that the above conditions force 0 ::5 A(1l) ::5 n. A PVM is usually written
as Il ~ E(Il); it follows that each E(Il) is a projection (take Il, = 112 in the
definition). This notion is familiar from the spectral theorem.
Proposition 1.4.8. Let X be a locally compact Hausdorff space, with Borel struc-
ture :E. There is a bijective correspondence between positive maps Q : Co(X) ~
~ ('H) that can be extended to CO(X)I by a unit-preserving positive map and POVMs
Il ~ A(1l) on X in 'H, given by

Q(f) = Ix dA(x) f(x). (1.34)

The map Q is a representation ofCo(X) iff Il ~ A(1l) is a PVM.

The precise meaning of (1.34) will emerge shortly. Given the assumptions, in
view of 1.1.2.3 we may as well assume that X is compact.
Given Q, for arbitrary '" E 'H one constructs a functional jlljl,1/J on C(X) by
jlljl,I/J(f) := ("', Q(f)"'). Since Q is linear and positive, this functional has the
same properties. Hence the Riesz representation theorem yields a probability mea-
sure J.,L1jI,1jI on X, For Il E :E one then puts ("', A(Il)"') := J.,L1/J,1/J(1l), defining an
operator A(1l) by polarization. The ensuing map Il ~ A(1l) is easily checked to
have the properties required of a POVM.
Conversely, for each pair"', <1> E 1{ a POVM Il ~ A(Il) in'H defines a signed
measure J.,L1jI,<fJ on X by means of J.,L1/J,<fJ(Il) := ("', A(Il)<1». This yields a positive
map Q : C(X) ~ ~('H) by ("', Q(f)<1» := Ix dJ.,LI/J,<fJ(x) f(x); the meaning of
(1.34) is expressed by this equation.
Approximating f, g E C(X) by step functions, one verifies that the property
E(Il? = E(Il) is equivalent to Q(fg) = Q(f)Q(g); then use 1.1.3.7. •
Corollary 1.4.9. Let Il ~ A(Il) be a POVM on a locally compact Hausdorff
space X in a Hilbert space 'Hx. There exist a Hilbert space 'H x , a projection p on
'H x , a unitary map U : 'Hx ~ p'H x , and a PVM Il ~ E(1l) on 'Hx such that
U A(Il)U-' = pE(ll)p for all Il E :E.
Combine Theorem 1.4.2 with Proposition 1.4.8.
•
122 II. Quantization and the Classical Limit

Suppose X is the phase space S of a physical system, and one is in the situation
discussed prior to 1.4.5. One then obtains a POVM Ll ~ A(Ll) on S in 1{/i asso-
ciated to the Berezin quantization map Q = Qg : Co(S) -+ ~o(1{/i). According
to 1.1.6.6, one may identify a state on 'B o(1{/i) with a density matrix p on 1{/i. The
physical interpretation of the map Ll ~ A(Ll) is then contained in the statement
that the number
(1.35)
is the probability that in a state p the system in question is localized in Ll C S
(localization in phase space). Transferring the situation to it/i by means of the
unitary U in 1.4.2, and writing p := U pU- I , one simply has pp(Ll) = Tr pE(Ll),
where Ll ~ E(Ll) is the PVM on 1{x given by 1.4.9.
When X is a configuration space Q, on the other hand, the Poisson bracket
between any two functions on X normally vanishes, so that the conditions (1.2)
and (1.3) can be satisfied by taking Q to be a representation rr of Co(Q) on 1{.
By Proposition 1.4.8, the situation is therefore described by a PVM Ll ~ E(d)
on Q in 1{; the probability that in a state p the system is localized in Ll C Q
(localization in configuration space) is
pp(Ll) := Tr pEed). (1.36)

1.5 Coherent States and Reproducing Kernels

One can find an explicit realization of 1{/i := 1{x and of the projection p in 1.4.2
if a further assumption is made, which is satisfied in many cases of interest.
Definition 1.5.1. A pure state quantization {1{/i, q/i, f-L/i}hElo of S is said to be
coherent if each qh(a) E lnlh can be lifted to a unit vector WI:' E 1{h, and the
ensuing map a ~ WI:' from Sto 1{h is continuous. The unit vectors WK coming
from a coherent pure state quantization are called coherent states.

In terms of coherent states, the polarized form of (1.8) is

Is df-Lh(a)(W\. Wh>(WK. W2) = (W 1 ,W2) (1.37)

for all WI , W2 E 1ih' We write

Kh(p,a):= (Wf,Wh>; (1.38)
as a consequence of the continuity assumption above, K/i is jointly continuous.
Also, one notices that
(1.39)
Proposition 1.5.2. Let {1{h. WK. f-L/i}aES,hElo be a coherent pure state quantiza-
tion, with associated Berezin quantization Qg. One may put
1{x = 1{h := L 2(S. dJlh); (l.40)
rrx (f)4>(a) := f(a)4>(a) (1.41)
 1 Foundations 123

in 1.4.2. Furthermore:
• For each Ii E 10, the map W : 1t n -+ 1t1l defined by
W\II(O") := (\11K, \II) (1.42)
is a partial isometry (with WW* = p a projection, and W* W = n). We denote
its image W1t1l in Jill by ii. n.
• The projection p : 1t1l -+ ii. 1l is given by

(1.43)

• The elements ofii. 1l C 1t1l may be chosen to be continuous functions.

• For each PES, the function 0" 1-+ K,1l(0", p) lies in ii. n.
• The evaluation map <1> 1-+ <1>(0") is continuous for all <1> E ii. 1l and all 0" E S.
• For each f E VX)(S), the operator Qg(f) := WQg(f)W* on ii. 1l , which
provides an equivalent realization of the Berezin quantization of S, is given by
Qg(f)<l>(O") = pf(O")<l>(O"). (1.44)
The first two claims follow from (1.37). The Cauchy-Schwarz inequality applied
to (1.42), and the continuity of qll prove the third claim. The next claim is immediate
from (1.42), since K,h(" p) = W\IIt:. To show the continuity of the evaluation map,
we write

(1.45)

which, as a consequence of (1.43), holds for all <l> E ii. h and all pES. The right-
hand side is (K,h(" p), <l» (inner product in 1t/j), which, combined with the previous
item, proves the claim. Finally, (1.44) is immediate from the definitions. •
Comparing, e.g., (1.44) with (1.24), we see how the above construction provides
an explicit realization of the objects defined in 1.4.2. As a case in point, we may
rewrite (1.35) in an appealing way. Note that because of (1.41), the PVM ~ 1-+
E(~) in 1.4.9 is given by E(M = Xli (the characteristic function of ~). Assuming
that p is a pure state p = [\II], where \II E 81t/j, the discussion after (1.35) and
(1.42) then implies that the probability that the system is localized in ~ is

p[ljIj(M = i d/l-/j(O") 1(\11K ,\11)12. (1.46)

An interesting feature to be abstracted from 1.5.2 is the following.

Definition 1.5.3. Let S be some set, and let 1t be a Hilbert space offunctions (of
some class) on S. A reproducing kernel of1t is afunction K, : S x S -+ C such
that:
• For each PES, the function 0" 1-+ K,( 0", p) lies in 1t.
• The reproducing property
\II(p) = (K(., p), \II) (1.47)
124 II. Quantization and the Classical Limit

holds for all'll E ?t, pES.

Taking IV = K(·, p), we obtain

1I,q·, p)1I = JK(p, p); (1.48)

in particular, K(p, p) ::: 0 for all p. Putting'll = K(·, a), one observes that
K(a, p) = K(p, a).

Proposition 1.5.4. A Hilbert space offunctions on S has a (necessarily unique)

reproducing kernel ifJeach evaluation map Err : 'll ~ 'll(a) is continuous.
The uniqueness of K follows by assuming that two reproducing kernels K 1, K2
exist, and showing that IIK 1 (-, p) - Kd·, p)1I has to vanish because of the
reproducing property. The rest is obvious. •
Lemma 1.5.5. If 'Ii has a reproducing kernel K, then strong convergence 'lln -)-
'll in ?t implies uniform convergence as functions on all subsets of S where a ~
K(a, a) is bounded.
One has

Then use (1.48).

This situation becomes particularly interesting when S is a topological space
•
and K is jointly continuous. In that case, (1.47) and (1.48) imply that ?t consists
of continuous functions. Moreover, if we equip C(S) with the topology of uni-
form convergence on compact sets, then Lemma 1.5.5 implies that the canonical
injection ?t ~ C(S) is continuous. This motivates the following abstract con-
siderations, which provide an interesting perspective on the reproducing kernel of
?t.
Definition 1.5.6. A Hilbert subspace of a topological vector space V is a
Hilbert space ?t with continuous linear injection ?t ~ V. In other words, ?t
is a continuously embedded subspace of V.
The Riesz-Fischer theorem then leads to an antilinear map (j ~ 0 from V* to
?t (and hence to V), defined by the property (j(w) = (0, w) for all w E ?t. When
V* separates points in V, the range V* of this map is dense in ?t. To guarantee
this, we assume that V is locally convex and Hausdorff. In any case, one obtains a
positive sesquilinear form Q on V* by
Q«(j, 11) := (ii, 0). (1.49)

In the situation of the paragraph preceding 1.5.6, the dual of V = C(S) is the
space of complex Radon measures f..L on S with compact support. Hence

jl(p) = Is df..L(a)K(p, a) (1.50)

 1 Foundations 125

from (1.47), so that the quadratic form Q is given by

Q(/L, v) = ( dv(p)d/L(a)lC(p, a). (1.51)

Jsxs
In particular, the reproducing kernel itself is recovered by lC(p, a) = Q(80' , 8p ),
where 80' is the Dirac measure at a (i.e., 80' (f) := f (a», etc. Hence IC is completely
determined by the embedding 1t "-+ C(S). If, in addition, we suppose that 1t C
L 2(S, d/L) (defined with respect to some Radon measure /L), the projection p :
L 2(S, d/L) ~ 1t is given by generalizing (1.43) to

p'P(p) = Is d/L(a)lC(p, a)'P(a). (1.52)

We know that lC(a, a) :::: 0 for all a E S; let us further assume that lC(a, a) > 0
for all a (equivalently, there are no points in S at which all elements of 1t vanish).
Then one obtains a family of unit vectors \110' in 1t, defined by

'PO' (p):= lC(p, a) . (1.53)

.jJC(a, a)

These satisfy the overcompleteness property

Is d/L(a)K.(a, a)('PJ, 'PO')('PO' , 'P2) = ('PI. 'P2) (1.54)

for all 'PI, 'P2 E 1t; cf. (1.37), and notice thatthe inner product (, ) is the one in 1t,
inherited from L 2 (S, dJ-L). Hence these unit vectors satisfy the key property (1.8)
of coherent states; via the reproducing kernel they are eventually defined through
the evaluation map.
The Hilbert space 1tn is defined as the image of 1t under the unitary transforma-
tion U : L 2(S, d/L) ~ L 2(S, d/Ln) (where d/Ln(a) := d/L(a)lC(a, a» defined by
U'P(a) = 'P(a)/.jJC(a, a). This space 1tn has a reproducing kernellCn , namely

K. ( a)'- ('P P 'PO') _ lC(p, a) (1.55)

n p, .- , - .jJC(p, p)1C(a, a)

This kernel is normalized, in that ICn(a, a) = 1 for all a; equivalently, one has
IIlCn(·, p)1I = 1 in 1tn. Its reproducing nature in 1tn may be derived from the
corresponding property of IC in 1t.
A Berezin operator QB(f), depending on f E UXJ(S), may then be defined on
1t (or 1tn ) as in (1.44), with p given by (1.52) (with IC replaced by ICn ). On 1t this
operator then assumes the form (cf. (1.16»

(1.56)

whereas on 1tn one has the same equation with 'P'" replaced by U'P'" = ICnh a).
It remains to be seen, of course, whether one can introduce Ii in a suitable way, so
as to arrive at a pure state quantization or a strict quantization.
126 II. Quantization and the Classical Limit

2 Quantization on Flat Space

2.1 The Heisenberg Group and its Representations
The manifold P = T*]Rn is equipped with its canonical cotangent bundle Pois-
son bracket 1.(2.24). Regarding V f := (aflap, aflaq) as a vector in ]R2n, and
introducing the 2n x 2n matrix

J = (~n ~), (2.1)

we can write
{f, g} = (V f, J V g) (2.2)
in terms of the natural inner product in ]R2n. This Poisson bracket is symplectic;
as in 1.(2.23), the symplectic form is
(2.3)
A central role in the study of T*]Rn is played by the so-called Heisenberg
group Hn. A concrete form of its Lie algebra ~n = ]R2n+l is obtained by taking
the coordinate functions Pi, qj as well as the unit function on T*]Rn as basis
elements, and equating the Lie bracket with minus the Poisson bracket. This basis
is traditionally denoted by {Pi, Qj, Zl;,j=l .... ,n. The Lie brackets are
[Pi, Pj] = [Qi, Qj] = 0;
[Pi, Qj] = -d! Z;
[Pi, Z] = [Qj, Z] = o. (2.4)
Definition 2.1.1. The Heisenberg group fIn is the unique connected and simply
connected Lie group with Lie algebra ~n'
Clearly, fIn = ]R2n+l is nilpotent, and the exponential map Exp : ~n -+ fIn
is a diffeomorphism. Following the physics literature, we parametrize fIn by
coordinates u, v E ]Rn and s E ]R so that

(u, v, s):= Exp(-uQ + vP + sZ), (2.5)

where u Q := Uj Qi , etc. The composition rule in fIn then follows from (2.4) and
the CBH-formula Exp(A)Exp(B) = Exp(A + B + HA,
B]); the higher-order
commutators vanish in this case. This yields
(u, v, s)· (u', v', s') := (u + u', v + v', s + s' - ~(uv' - vu'», (2.6)
where v u' = Vi u;, etc. Regarding w := (u, v) as a vector in the linear symplectic
space ]R2n, equipped with the (symplectic) form w = dv i /\duj (cf. (2.3», we may
write (2.6) as
(w, s)· (w', s') = (w + w', s + s' + ~w(w, w'». (2.7)
 2 Quantization on Flat Space 127

One often works with a version of the Heisenberg group in which the s-
coordinate is compactified; the group Hn is the quotient of fIn by the discrete
normal subgroup (0,0, 2JrZ). Hence the projection T : fIn --+ Hn is given by
T(U, V, s) = (u, v, exp(-is». The composition law in Hn then follows from (2.6)
as
(u, v, z)· (u ' , Vi, z'):= (u + u' , V + Vi, zz'e~i(UVI-VUI»). (2.8)

A Lie algebra anti-isomorphism p~1 ~ f)n between the Poisson algebra p~1 of
polynomials on T*]Rn of degree ::s 1 and the Heisenberg Lie algebra is given by
p(u.v.s)(p, q) = vp - uq +s ~ vP - uQ + sZ. (2.9)

One may regard X E f)n as a function X on the dual f)~ by putting X(O) :=
O(X) for 0 E f)~; this yields an inclusion f)n C C)O(f)~). We use coordinates
(p, q, c) on f)~ = ]R2n+1 (where p, q E ]Rn and c E ]R), which represent the point
p P + q Q+ c Z. Here {Pi, Qi, ZJi.i= I ..... n is the basis of f)~ dual to the given one
in f)n' The functions Pi, Qi then coincide with the coordinate functions Pi, qi.
The differentials of all functions X span the cotangent bundle T* f)~, so that a
possible Poisson structure on f)~ is determined by the Poisson brackets of the X.
Thus one may put

{X, YJ- := -fX,YJ. (2.10)

The reason for the minus sign will become clear in III.I.l. This leads to the Poisson
c»
::i::J,
bracket (we omit the argument (p, q,

{f,gJ-=c(:;i::i - (2.11)

cf. 1.(2.24). The symplectic leaves of f)~ come in two types. Firstly, one has the
manifolds T*IR~ := ]R2n X {c} for c =f- 0, with symplectic form We = C dqi /\ dpi.
The "usual" T*]Rn with Poisson bracket 1.(2.24) is the leaf corresponding to c = 1.
Secondly, each point (p, q) in T*lR n x {OJ is a leaf.
There is a different way of looking at these leaves. The so-called coadjoint
action Co of fIn on f)~ is defined by
(Co(u, v, s)e)(Y) := O(Ad«u, v, S)-I)y), (2.12)
where Ad is the adjoint action of fIn on f)n' The CBH-formula yields
Proposition 2.1.2. The coadjoint action of the Heisenberg group is given by
Co(u, v, s)(p, q, c) = (p + cu, q + cv, c). (2.13)
Accordingly, the orbits in f)~ under the coadjoint action coincide with the symplectic
leaves of the Poisson structure (2.10).
The result may be recast in the language of Chapter I.
Proposition 2.1.3. Unless it is defined on a zero-dimensional space, any ir-
reducible representation Jrel of the Poisson algebra COO(f)~) associated to the
128 II. Quantization and the Classical Limit

Heisenberg group is equivalent to rr~1 ,for some real c i- 0, defined on the symplectic
manifold (T*JRn , c dqi A dpi) by
rr~I(f) = fre;
frc(P, q) := f(p, q, c). (2.14)
If rr d is zero-dimensional, there is a point (p, q) E T*Rn such that
rr(~.q)(f) = f(p, q, 0). (2.15)
This is immediate from Theorem 1.2.6.7.
The corresponding representation of ~n C COO(~~) on T*JR~ is simply
•
Pi~Pi;
Qi ~ qi;
Z ~ cIr*JR". (2.16)
In particular, CCXl(T*JRn) with the canonical Poisson structure 1.(2.24) may be seen
as the representative rrll(COQ(~~».
Proposition 2.1.3 has an exact parallel in quantum mechanics. Consider the
following family of representations of fIn. For each real A i- 0, construct the
operator Uf(u, v, t) on the Hilbert space L2(Rn) by

(2.17)
It is easily checked that the Uf are unitary, and indeed furnish a representation
of fIn, called the Schrodinger representation. The irreducibility of Uf will be
proved in 2.5.5. We see that Uf(O, 0, s) = exp( -i)..s)[; hence for).. E Z the
representation uf is defined on Hn as well, satisfying
U~(O, 0, z) = zn[. (2.18)
A useful equivalent version of uf is given by

UJ..(u, v, s)\II(x) := e-iJ..{s+!uv-UX)\II(x - v); (2.19)

one has vufv* = UJ.. for the unitary V: L2(Rn) ~ L2(JRn) defined by V\ll(x) =
)..n/2\11(AX). The corresponding representations of the Lie algebra ~n are given by
(cf. III.(1.69»

(2.20)
and
dUJ..(Qi) = _i)..xi;

dUJ..(P·) -a ..
= -ax}'
}
dUJ..(Z) = -iH, (2.21)
 2 Quantization on Flat Space 129

respectively; here xi is meant as a multiplication operator, i.e., (xi \II)(x) = Xi \II (x).
These operators are defined and essentially self-adjoint on S(JRn) c L2(JRn), on
• (S) (S) (S) ~ I..
which [dU.. (X), dU.. (Y)] = dU.. ([X, Y]) lor all X, Y E '}n'
The representation U.. is of particular use for).. = lin. For later convenience,
we introduce the Weyl operator
i ( S S)
Ul(p, q):= U!(p, q, 0) = eX pQh-qPh , (2.22)
h h

where
(2.23)

and

(2.24)

are the physicists' position operator and momentum operator, respectively; cf.
(2.20) and (2.37). These operators are both defined and essentially self-adjoint on
S(JRn), on which domain one has the canonical commutation relations
Cph,i'
s Qs.j] _
Ii -
. r. .j][.
-I,w i ' (2.25)
cf. (2.4). One might add here that
ihdUdZ)
h
= n. (2.26)
Theorem 2.1.4. Unless it is one-dimensional, any irreducible representation U
of fin is equivalent to uf for some).. t= O. When U is one-dimensional, there is a
point (p, q) E T*JRn such that U equals
U(p,q) ( u, v, s ) = ei(uq-vp) . (2.27)
When U (0, 0, s) = nfor all s E JR, the representation must be one-dimensional,
so that (2.27) is a restatement of the representation theory of the abelian group JR2n .
A proof of the remainder of this celebrated theorem will be given at the end of
III.3.7. Another appropriate proof is obtained by combining either Corollary 2.6.7
or Proposition 111.1.8.4 with Corollary 1.2.2.6; the statement in 2.1.4 concerning
).. t= 0 is equivalent to the uniqueness of the irreducible representation of the
C* -algebra of compact operators. 0

2.2 The Metaplectic Representation

As we have seen in the previous section, the Heisenberg group is closely re-
lated to the Poisson algebra p~l of polynomials on T*JRn of degree ~ 1. At
the next level, the Poisson algebra p2 of quadratic polynomials on T*JRn turns
out to be anti-isomorphic to the Lie algebra of the symplectic group Sp(n, JR).
This group consists of the linear Poisson isomorphisms of T*JRn ::::: JR2n; a ma-
trix M E GL(2n, JR) lies in Sp(n, JR) iff MT J M = J (cf. (2.1». For the Lie
algebra this means that a 2n x 2n matrix X lies in sp(n, JR) iff J X + XT J = 0
130 II. Quantization and the Classical Limit

(equivalently, Xl is symmetric). The maximal compact subgroup of Sp(n, R) is

Sp(n, R) n O(2n); ifR2n is identified with through en
. (qj + ipj)
Zl = -'---:~- (2.28)
-Ii
(where pj := Pj), then Sp(n, R) n O(2n) = U(n). It follows from the theory of
noncompact semisimple Lie groups that the homotopic properties of Sp(n, R) are
determined by its maximal compact subgroup; hence Sp(n, R) is connected, but
not simply connected, since Jrl (Sp(n, R» = JrI(U(n» = Z. Note that in terms of
complex coordinates the symplectic form (2.3) reads

(2.29)

Hence in terms of the usual inner product on en one has

w(z, z') = 21m (z, z'). (2.30)

This expression renders it self-evident that U (n) C Sp(n, R). With a := a/ az and
a := a/az, the Poisson bracket 1.(2.24) now reads
{j, g} = i(ajag - ajag). (2.31)

Further to the notation w = (u, v), we put a := (p, q); also recall (2.1). For
X E .sp(n, R) we define the quadratic polynomial

PX(a) := t(J Xa, a), (2.32)

where the inner product is the usual one in R2n. Using (2.2), for X, X' E .sp(n, R)
one easily verifies that

{PX , P X'} = -PIX,X']' (2.33)

which proves that (2.32), which is clearly bijective, defines a Lie algebra anti-
isomorphism between p2 and .sp(n, R).
The group Sp(n, R) acts on Hn: the matrix ME Sp(n, R) maps (w, s) E Hn to
(Mw, s). Writing h for (w, s), we say simply that M maps h into Mh. We may
therefore build the semidirect product Sp(n, R) ~ Hn , whose elements are pairs
(M, h), with M E Sp(n, R) and h E Hn. The group multiplication is given by
(M, h) . (M', h') := (M M', h . M h'), where the product· in Hn is given by (2.6).
Note, in particular, that

(M, 0)· (e, h)· (M- I , 0) = (e, Mh), (2.34)

where e and 0 are the identity elements in Sp(n, R) and Hn , respectively. The
"mixed" Lie bracket in the Lie algebra .sp(n, R) ~ I)n is

[M, (w, s)] = (Mw, s). (2.35)

Let PI, P2 be polynomials of degree ::'S 2 in (Pi, qj). The space p:::2 of such
polynomials is easily seen to be closed under the Poisson bracket 1.(2.24).
 2 Quantization on Flat Space 131

Proposition 2.2.1.
• Under the correspondence (2.9) the Poisson algebra p:::1 of polynomials of
degree::: 1 is anti-isomorphic to the Lie algebra ~n of the Heisenberg group
fIn.
• The Poisson algebra p2 of quadratic polynomials is anti-isomorphic to the Lie
algebra sp(n, R) of Sp(n, R) under the correspondence PM(a) ~ M.
• By linear extension of the preceding two items, the Poisson algebra p:::2 of
polynomials of degree::: 2 is anti-isomorphic to the Lie algebra sp(n, R) ~ ~n
of the semidirect product Sp(n, R) ~ fIn.
The first item was shown in the previous section. The second is proved by (2.33).
The third claim follows from (2.35). •
One can easily solve the equations of motion for Hamiltonians in p:::2. The
Hamiltonian flow generated by p(w,s) (cf. (2.9» is a(t) = a + tw (cf. (2.13)
with c = 1), and the flow generated by P x is a(t) = Exp(tX)a. These flows are
compatible with the natural action pO of Sp(n, R) ~ fIn on T*Rn, under which
(M, (w, s»mapsa tOP?M.(W,S»(a) = Ma+w.IfPx ~ X under the isomorphism
of2.2.1 (X E sp(n, R) ~ ~n), one verifies that Exp(tX) maps a to Exp(tX)a =
a(t), where a ~ a(t) is the Hamiltonian flow generated by P x on T*Rn. Hence
Exp(X)a = a(1). (2.36)
We will now construct an important integrable Hilbert space representation of
sp(n, R). Let P(Pi, qj, 1) be a polynomial on T*Rn. We define
Q'j( (P(Pi, qj, 1» := A[P(Pl. i , Q:.,j), 1I], (2.37)

cf. (2.24) and (2.23). This expression means that one substitutes pi, Q~, for p, q
in p, and symmetrizes; thus A[ ... ] denotes complete symmetrization. For example,
A[A h ... , An] = L1rES. A 1r (I)' .. A 1r (n)/n!, where the sum is over all n! elements
7C of the permutation group Sn.
Given its construction from UK (Exp(u Q - v P», it follows from standard repre-
sentation theory that Q'j( (P) is well-defined as an unbounded operator on L 2(Rn)
with domain S(Rn). If P is real, then Q'j( (P) is symmetric on this domain.
Proposition 2.2.2.
• Restricted to at most quadratic polynomials, Q'j( is a Lie algebra homomor-
phism, in that for all PI, P2 E p:::2 one has

hi[W
Q" (PI), Q/iW]
(P2) = Q"W({PI, P2}). (2.38)

• Hence dph, defined by

(2.39)

(where X E sp(n, R) ~ ~n corresponds to P x under the anti-isomorphism

between sp(n, R) ~ ~n and p:::2, cf. 2.2.1 ),jurnishes a representation ofthe Lie
132 II. Quantization and the Classical Limit

algebra sp(n. R) P< ~n as unbounded operators on the common invariant dense

domain S(Rn) C L2(Rn).
• For each X E sp(n. R) P< ~n the operator idP"'(X) is essentially selfadjoint on
this domain.
• There exists a double covering Mp(n. R) of Sp(n. R) (known as the meta-
plectic group) and a representation ph of Mp(n. R) P< fIn (where the action
of Mp(n. R) on fIn factors through Sp(n. R) in the obvious way) on L 2(Rn),
whose derived representation ofsp(n. R) P< ~n is dph.
• Restricted to fIn' the representation pli coincides with UI/IiUln).
The restriction of pli to Sp(n. R) is called the metaplectic representation.
A simple calculation shows that the commutation relations (2.38) are satisfied on
S(Rn); hence the first claim follows from 2.2.1. The equationdph(X) = dUI/Ii(X)
for X E ~n is immediate from (2.23), (2.24), and (2.26).
A technical result in functional analysis, involving the existence of a dense set
of analytic vectors (here given by the linear span of the Hermite polynomials),
shows that dpli(sp(n, R» exponentiates to a representation ph(Sp(n, R», where
Sp(n, R) is the unique connected and simply connected covering group of Sp(n. R)
(one has Sp(n. R)/ Sp(n. R) ~ IE). This argument also leads to the essential self-
adjointness property mentioned. It can be shown that the metaplectic representation
ph is double-valued on Sp(n. R) (that is, pli(M)ph(M') = ±p\MM'), where the
sign depends on M and M'), so that there is a double covering group Mp(n. R) of
Sp(n. R) on which ph is single-valued (i.e., is a representation). 0

From 2.2.2, (2.22), and (2.34) we have the equivariance property

p\M)Ul(a)ph(M)* = Ul(Ma). (2.40)

h h

where M E Sp(n, R). We may reformulate this result in terms of dynamics. We

regard a real polynomial h on T*Rn as a classical Hamiltonian, denoting its flow
by a ~ a(t). Its quantization, the quantum Hamiltonian Hh, is taken to be the
unbounded operator

Hh := Q:i (h(P;. Q~», (2.41)

cf. (2.37), (2.24), and (2.23). Let h = Px E p2. From (2.39) we see that Hh =
i hdph(X), so according to (2.36) we can rewrite (2.40) as
(2.42)

We tum to the reducibility of ph(Sp(n, R». The following result will be of

central importance in the construction of the Weyl quantization map in 2.5.

Lemma 2.2.3. The parity operator P on L 2(Rn). defined by

PIlI(x):= 1lI(-x), (2.43)

commutes with all ph(M), M E Sp(n, R). The eigenspaces L2(Rn)± C L2(Rn).
characterized by the property PL 2(Rn)± = ±lIL2(Rn)±. are irreducible under
ph(Sp(n, R». Hence the commutant of ph(Sp(n. R» is spanned by P and lI.
 2 Quantization on Flat Space 133

Simple computations show that [P, dp"(X)] = 0 for all X E 51'(2, R).
Since Sp(n, R) is connected and the exponential map is onto, it follows that
[P, p\M)] = 0 for all M E Sp(n, R). Hence p"(Sp(n, R» is reducible,
and $±L2(Rn)± obviously decomposes L 2(JRn). The fact that the L2(JRn)± are
irreducible follows from an uninteresting technical argument. D
This lemma implies that for all M E Sp(n, R) one has
p"(M)Pp\M)* = P. (2.44)

2.3 Berezin Quantization on Flat Space

Mter this preparatory material we tum to the quantization of T*JRn. A suit-
able choice of the Poisson algebra we wish to quantize turns out to be !it~ :=
C~(T*JRn, JR); this is a dense subspace of~~ = Co(T*JRn, JR) under the sup-norm.
We write (J' = (p, q); the Poisson bracket is given by 1.(2.24).
We now construct a Berezin quantization of ~~ from a pure state quantization,
as outlined in 1.3. The strategy is generic.
Proposition 2.3.1. Put I = Rand?-t" := L2(Rn) for all Ii =f. O. For each
(p, q) E T*R n, define a unit vector l.JI~p,q} E ?-th by
I.JIhP,q}:= Uk(p,q)l.JIg; (2.45)
I.JIg(x) := (rrh)-n/4 e -x 2/(2h), (2.46)
cf (2.22). Explicitly, one has
I.JIhP,q}(x) = (rr h)-n/4e-!iPq/"eiPX/"e-(X-q)2/(2h) , (2.47)

Denote the projection ofl.JlhP,q) E §?-th to JP1{" by l/IhP,q}. Then the choices
q:(p, q) := l/Ihp,q), (2.48)
dnpdnq
d/L"(p, q):= (2rrli)n (2.49)

yield a coherent pure state quantization ofT*JRn .

This is established by simple computations. In fact, (1.10), without the limit,
and (1.8) are valid for any unit vector I.JIg; the explicit choice (2.46) is used only
to prove (1.9). Here the decisive intermediate result may be expressed in terms of
complex variables (see (2.28» as

("'(w) ",(z» _ It'" r:; ) _ (-!ww-!zHWZ)/h. (2.50)

Vh 'Vh - I'vh\Z, W - e ,
cf. (1.38). Hence (1.11) is immediate from the corollary

The Berezin quantization

p (q:(w), q:(z») = e-lz-wI2/h,
Qg defined by (2.48) (cf. 1.3.4) is given by
•
B f dnpdnq (pq)
Q" (f) = IT*lR" (2rrli)n f(p, q)[I.JI" , ], (2.51)
134 II. Quantization and the Classical Limit

where f E Loo(T*Rn). For f E Co(T*Rn, R) the Gelfand transfonn (1.15) is

--;-- _f d npdnq (P.q)
(2.52)
Q/i (f)(1{I) - iT-an (2rrn)n P(1{Ih ,1{I)f(p, q).

In tenns of the complex variables (2.28), the measure (2.49) reads

dnzd"z
dJL/i(z, Z) = (2rr ni)n . (2.53)

Proposition 2.3.2. In the context of Proposition 1.5.2 (in which, using complex
coordinates, 11.1'1 = L 2(Cn, JL,,»), the Hilbert space it" consists of all junctions of
the type W(z, z) = exp(-zz/(2n»III(Z), where III is an entire function for which
fen d nzd nzexp(-zz)IIII(z)1 2 < 00.
We call the space of functions of the stated type iJ 2(C n ); elementary analysis
shows that it has IC" (cf. (2.50) as a reproducing kernel. By the argument given in
the proof of 1.5.2, nonn-convergence in iJ2 (C n ) implies uniform convergence. This
shows that iJ 2 (C n ) is complete. Moreover, the fact that entire functions are given
by Taylor series (unifonnly convergent on compact sets) shows that the functions
{<i>"}~I=O' where a := (al, ... , an) is a multi-index, with lal := al + ... + an,
and

(2.54)

where za := ~I ••• ~n, fonn an orthononnal basis in it". The orthononnality

follows from an elementary computation in polar coordinates.
Using (2.47) and (2.28), we write (1.42) as

WIII(z, z) = (rrfi}-n/4 e-(Z"l+z2)/(2") f d"x 111 (x)e( _tx2+~Z)/". (2.55)

The integral converges unifonnly in Z on compact sets, so Will is exp( -zz/(2fi})

times an entire function in Z. The square-integrability of Will follows from
the fact that W is a partial isometry. Hence it" £ iJ 2(Cn ). For example,
IC/i(', w) E it/i for each W E Cn , as it should be. One computes (<1>." IC,,(-, w» =
2-"/2 exp( -tww)w.,. It follows that (<I>, IC/i(-, w» = 0 for all w implies <I> = 0,
so that the span of the collection of functions IC/i(-, w) E it/i, W E Cn , is dense in
it/i. Since these are the images of the coherent states in 11.1'1 under W, the proof is
complete. •
With hindsight, we can now fonnulate a unitarily equivalent fonnulation of
Berezin quantization on T*R": We start with the Hilbert space H~(cn) of
conjugate-entire functions on C n , whose inner product with respect to the Gaussian
measure on C n is finite, namely
d"zdnz -
(III, <1» := n-n [ - - .-e- zz /"III(z)<I>(Z) < 00. (2.56)
Cn (2rr t)n
 2 Quantization on Flat Space 135

The functions <1>", occurring in the proof of 2.3.2 form an orthonormal basis of
H~(cn). The latter plays the role of1t in Definition 1.5.3. The Berezin quantization
in this realization, which we denote by Q~ to avoid confusion with the equivalent
version (2.51), is then given by
Qf(f) = pip· (2.57)
Here I is regarded as a multiplication operator on it"
:= L 2(Cn , n-n jt~) (where
jt~ is the Gaussian measure occurring in (2.56», and p is the projection onto the
subspace H~(cn) of entire functions ofz in it". Compare with (1.44).
The Hilbert space H~(cn) has an (unnormalized) reproducing kernel, the so-
called Bergman kernel, given by
(2.58)
Hence by Proposition 1.5.4 each evaluation map E z : '11 ~ 'I1(Z) is continuous.
The coherent states 'I11t' are defined as in (1.53). As in the passage from (1.44) to
(1.56), we may then rewrite (2.57) as
(dnwdnw _
len
v

Qf,(f)'I1(z) = n-n (27ri)n e-ww/"~;}h(Z' w)/(w, W)'I1(W). (2.59)

As explained in 1.3, this can be transferred to the Hilbert space 1t n , which possesses
the normalized reproducing kernel (1.55). In the present setting, 1tn coincides with
H" (cf. 2.3.2), since the rescaled measure jtn is just the Liouville measure times
n-n. Hence we indeed have

(2.60)

There is yet another, closely related, way of looking at Berezin quantization, or

rather the coherent states behind it. For any Hilbert space 1(, with inner product
(, k, we introduce the exponential Hilbert space, or bosonic Fock space, exp(l()
as follows. Let the Hilbert space ®~I( be the symmetrized tensor product of I
copies of 1(; this is the invariant subspace of ®I I( under the natural action of the
permutation group SI. The closure of the direct sum of all ®I I( is

E9 ®~I(.
00

exp(K) := (2.61)
1=0

This space is separable iff I( is. The element 1 E C = ®0l( is denoted by Q;

elements of I( are called w or Z. We define a map JEXp: I( -+ exp(l() by
m= ~®IW w®w
v Exp( w) := L... '" = Q +w + Mj + ... ; (2.62)
1=0 vi! V 2!
this is called an exponential vector. This map is clearly injective, since the compo-
nent of JEXp( w) in I( c exp(K) is w itself. The inner product of two exponential
vectors is
(2.63)
136 II. Quantization and the Classical Limit

For one thing, this equation easily entails that JEXP is continuous. It is not difficult
to show that the collection of exponential vectors is linearly independent and total;
i.e., the linear span IE of all JExp(w), w E K, is dense in exp(K).
For K = C n it is clear from the fact that (2.54) provides an orthonormal basis
-2
that the map Vh : exp(Cn ) -+ H It(Cn ), defined by extension of
I
(V/tw! ®s'" ®s w/)(z):= ,IiI"b;"(z, wlkn .. ·(z, w[kn (2.64)
yl!nn
is unitary. Hence the subspace ®~cn of exp(Cn ) corresponds to the subspace of
lth order monomials in H~(cn). Note that

(2.65)

cf. (2.58). Hence, with the convention (2.28), the coherent states (2.45) in L 2(]Rn)
correspond to the vectors JEXP (w/Jli),
up to normalization. Using (2.63), we
may therefore rephrase Proposition 2.3.1 as
Corollary 2.3.3. For finite-dimensional K the unit vectors (cf 1.5.1)

\II):' := e-k(w.wlx:/ltJExp (w/JIi) (2.66)

define a coherent pure state quantization of K into Hit := exp(K) for all ni- o.
Conceptually, one should stress that K, although a Hilbert space, is to be seen
as a classical phase space. In particular, q/t(w) depends on the phase of w, so
that qh does not quotient to a function on the projective space IP'K. In the infinite-
dimensional case the conditions (1.8) and (1.9) are not defined in the absence of
a Liouville measure on K, but (2.66) makes sense, and comes from a map qh that
satisfies the crucial condition (1.11).
Corollary 2.3.3 and (2.45) suggest that one look for a realization of the repre-
sentation U! (fin) on exp(K); what follows holds whatever the dimension of K.
n
For each z E K the annihilation operator a(z) is an unbounded operator on the
dense domain <E C exp(K) satisfying
a(z)JExp(w) = (z, wkJExp(w). (2.67)
The map z ~ a(z) is evidently antilinear. It can be shown that a(z) is closable;
the domain of its adjoint a(z)* contains <E. The map z ~ a(z)* is linear; a(z)* is
called a creation operator. The domain <E is evidently invariant under a(z); it can
be shown that a(w)*<E is contained in the domain of the closure of each a(z). The
commutator [a, a*] is therefore well-defined on <E; it is given by
[a(z), a(w)*] = (z, wk. (2.68)
The unbounded operators exp(a(z» and exp(a(z)*) are defined on <E as well, where
their action is given by a strongly convergent power series expansion. From (2.67)
one obtains
(2.69)
 2 Quantization on Flat Space 137

ea(z)* JExp(w) = JExp(z + w). (2.70)

In terms of these, the analogue on exp(K) of the Weyl operator (2.22) is

U 1 (z) := e f,;[a(z)*-a(zlJ; (2.71)
h

the unitarity of U is obvious from this expression, and z f-+ U 1 (z) (with (2.28»
h
yields a representation of the Heisenberg group that is equivalent to the one defined
in 2.1 under the same name. We then see from the eBH-formula and (2.68) that
we may rewrite (2.66) as
(2.72)
The position and momentum operators (2.23), (2.24) may then be expressed in
terms of the a and a* as Q~.i = .ffli(ai + an and Pl. i = .ffli(ai - ani i, where
a(z) = aj i , etc.

2.4 Properties of Berezin Quantization on Flat Space

Berezin quantization on flat space has the following pleasant property.
Theorem 2.4.1. Putting §to = C~(T*JRn) and 2(t, = ~o(L 2(JRn» for Ii i= 0,
the Berezin quantization map Q~ : §t~ -+ 2(~ defined by (2.51) is a nondegener-
ate strict quantization of the Poisson manifold T*JRn (with its canonical Poisson
bracket 1.(2.24)) on 1 = JR. Moreover,
Q~(Co(T*JRn» = ~o(L\JRn». (2.73)
Hence Qf, is a strict deformation quantization, except for (1.3), of 2(0 =
Co(T*JRn).
Before starting with the proof, we note that Qr, determines a continuous field
of C* -algebras by Proposition 1.2.4; this will be further developed in 2.6.
The nondegeneracy of Q~ is an easy corollary of 2.3.2. For qJ I, qJ2 E Ji" and
W defined by (1.42) we have (qJI, Q~(f)qJ2) = (WqJI, fWqJ2), where the inner
product is in Jih = L 2(S, d JL,,). Since one can construct a basis of the latter Hilbert
space consisting of functions of the type WqJI WqJ2, the property Qr,(f) = 0
implies f = 0 almost everywhere, which means that f = 0 for f E C~(T*JRn).
The converse is trivial.
The fact that Qr, maps C~(T*JRn, JR) into ~o(L 2 (JRn»]R follows from 1.3.5. To
show that Qr,(C~(T*JRn» is dense in ~0(L2(JRn», one observes that Q~(f) is
Hilbert-Schmidt for f E C~(T*JRn). If one assumes that f(p, q) = fl(p)!z(q),
the kernel K (x, y) of Qr, (f) factorizes as a function of the variables x ± y. Each
factor is then easily seen to be dense in L2(JRn ) as fi runs through C~(JRn). Equation
(2.73) then follows from 1.4.5.
Rieffel's condition and (1.1) hold by Proposition 1.3.6.
We now tum to the proof of (1.2), using (2.28). For mEN we will use
IIgllm.oo:= L lIaaa fJ flloo, (2.74)
lal+lfJl :'Om
138 II. Quantization and the Classical Limit

where a and fJ are multi-indices, and aa := a;' ... a~", etc. For \Ii E L 2(JRn) we
write (\liK, \Ii) = exp( -zz/(2n)\Ii(Z); cf. 2.3.2. From (2.51) and (2.50) we obtain,
after a shift of one of the integration variables,

(\Ii, Q~(f)Q~(g)\Ii) = f dJ.ih(Z, Z)dJ.ih(~, ~)e-(zzHf+zf)/"

x \Ii(z)\Ii(z + ~)f(z, z)g(z +~, Z + n (2.75)

One now expands g(z + ~, z + ~) in a Taylor series around (z, z). The zeroth-order
term leads to (\Ii, Q~(fg)\Ii). The remainder is :s ClIglIl,oolH for a constant C
of order 1 (further contributions to this constant will be absorbed without change
of notation). We take II f II 00 out of the integral, and of the factor exp( -~ ~ In) we
put exp( -~~1(2n» into the measure. We then apply Cauchy-Schwarz to the ~­
integral, factorizing the ~ -dependent integrand into I~ I times the rest. The first of
the ensuing two ~ -integrals is a Gaussian integral of I~ 12 , which is proportional to
n(which appears under a square root, so it will lead to a factor nI/2 ). There remains
an integral over z and ~. Here we apply Cauchy-Schwarz to the z-integration. The
resulting triple integral factorizes after a shift in one of the variables, and can be
performed; two of the factors are equal to II \Ii II. Hence

(2.76)

By 1.2.1.8, 1.2.5.3, and 1.(1.57), for each A E ~(1{)n~ there is a unit vector
\Ii E 7-{ such that

IIAII = 1(\Ii,A\Ii)I· (2.77)

Hence (1.2) follows from (2.76), which implies

lim IIQ~(j)
h--->O
0 Q~(g) - Q~(fg)1I = o. (2.78)

The proof of (1.3) is similar. We consider (2.75) with (\Ii, Q~(f)Q~(g)\Ii) re-
placed by (\Ii, wg(f), Q~(g)]\IJ). On the right-hand side one then has the terms
fez, Z)g(z + ~, z + ~) - fez + ~, z + ~)g(z, z), instead of fez, z)g(z +~, z + ~).
One now expands g(z + ~, z + ~) as well as fez + ~, z + ~) in a Taylor series
around (z, z). The zeroth-order term obviously vanishes. The linear term can
be evaluated by also expanding exp(-z~/n)\Ii(z + ~) in powers of f The ~­
integration can then be performed: The only nonzero contribution comes from
factors ~ f A partial integration in z then shows that the linear term equals
n(\Ii, Q~(-i{f, g} + gaaf - faag)\Ii), where the Poisson bracket is given by
(2.31).
The quadratic term contains ~~(faag - gaaj). In the ~-integral only the
zeroth-order term in ~ from exp( -z~1n)\Ii(z + ~) contributes, and the result may
be expressed as n(\Ii, Q~«(faag - gaaj)\Ii). This cancels the additional term
from the linear contribution. Hence the linear term and the expression with ~~ in
the quadratic term together produce n(\Ii, Q~( -i {f, g })\Ii). The remainder of the
quadratic term has a part proportional to ~ ~ , which vanishes upon integration, and
 2 Quantization on Flat Space 139

a part proportional to ~ g. After g-integration, and partial integration in z, the latter

part is easily seen to be bounded by h 2 11f114,00 IIg1l4,00 11111112.
The contribution of the higher-order terms is estimated as follows. Taylor's
formula with remainder of third order yields an object bounded by 211fll3,00
IIg 113,001~ 13 . We now proceed as in the proof of (1.2): Practically the only dif-
ference is that the Gaussian integral of I~ 12 in that proof is now replaced by one of
I~ 16 , which leads to an overall factor of order h3/ 2 • All this leads to the estimate

~[Qf(f), Qf(g)] - Qf({f, g})lII)l ~

1(111,

(Cdll3.oo IIgll3,oohl/2 + C211f1l4,00 IIgIl4,ooh) 11111112. (2.79)

Equation (1.3) now follows in the same fashion as (1.2) above. •
We turn to the equivariance properties of Qf. In preparation:
Definition 2.4.2. An automorphic action a of a group G on a C* -algebra Ql is
a homomorphism x t-+ ax, such that each ax is an automorphism ofQl, In other
words, apart from the linearity and bijectivity of each ax : Ql -+ Ql one has the
properties ax 0 a y = axy , ax(AB) = ax (A)a x (B), and aAA*) = ax(A)*.
Consider the natural action pO of Sp(n, JR.) D< JR.2n on T*JR.n ~ JR.2n, according to
which (M, w) maps a E T*JR.n to prM,W)(a) := Ma + w (cf. 2.2). This leads to
an automorphic action a O of Sp(n, JR.) D< JR.2n on Qlo = Co(T*JR.n), given by
(2.80)
Also, one has an automorphic action a h of G on Qlh = ~o(L2(JR.n», given by the
representation constructed in 2.2.2. That is,
afM,W)(A):= ph(M, w)Aph(M, w)*. (2.81)
Theorem 2.4.3. Foreach(M, w) E U(n)D<JR.2n, whereU(n) = Sp(n, JR.)nO(2n)
(cf. the text surrounding (2.28)), and all f E Co(T*JR.n), one has
Qf(arM,w)(f) = afM,W)(Qf(f)· (2.82)
To prove this, we rewrite (2.51) as a weak integral

(2.83)

The equivariance under JR.2n is obvious from this formula, the last claim in 2.2.2,
(2.22), (2.6), and (2.19).
Lemma 2.4.4. ffU E U(n) and ph is the metaplectic representation of Sp(n, JR.)
on 7th = L2(JR.n), then
(2.84)

This is most easily proved in the realization ~n ith , described in 2.3.2. From
(2.55) and (2.50) we have WIII~(z, z) = exp( -zz/(2h». If U E Sp(n, JR.)n O(2n),
140 II. Quantization and the Classical Limit

then Wph(U)W* can be shown to be given by

Wp'~(U)W*\II(z, z) = ~\II(U-1Z' U-1z), (2.85)

det(U)
from which (2.84) is immediate. o
The equivariance under U (n) follows from this lemma and (2.40). Since each
element of a semidirect product factorizes, Theorem 2.4.3 follows. •
This theorem can be reformulated in terms of dynamics on T*JRn.
Corollary 2.4.5. Define a class of classical Hamiltonians on T*JRn by
h(p, q) = 4(p, Ap) + 4(q, Aq) + (p, Bq) + (e, p) + (d, q), (2.86)
where A and B are real n x n matrices such that AT = A and B T = - B, the
inner products are in JRn, and e. d E JRn. Denote the time evolution generated by h
on the classical observables by a?(cf 1.(2.13)). Define the quantum Hamiltonian
Hhby
(2.87)
which is an unbounded operator with domain S(JRn). Then Hh is essentially self-
adjoint on S(JRn). The one-parameter automorphism group a~ on ll3 o(L 2 (JRn» is
defined by

(2.88)
Then one has
(2.89)
A matrix X E 9Jt2n (C) lies in U(n) when it satisfies J X + XT J = 0 and
XT + X = O. The polynomial (2.32) is then precisely of the form of the quadratic
term in (2.86). For h of the form (2.86), one computes

Q~(h) = 4(Pi, APi) + 4(Q~, AQ~) + (Pi. BQ~) + (c, pi) + (d, Q~) (2.90)
in terms of (2.24) and (2.23). This follows by calculating the matrix elements
between coherent states (which indeed lie in the domain of Hh). The expression
(2.90) coincides with (2.41), and therefore the essential self-adjointness of Hh, is
a consequence of Proposition 2.2.2. Corollary 2.4.5 now follows from Theorem
2.4.3, exactly as in the derivation of (2.42). •

2.5 Weyl Quantization on Flat Space

Theorem 2.4.3 suggests that one look for a quantization that is equivariant under
the full affine symplectic group Sp(n, JR) ~ JR2n . It is obvious from Lemma 2.4.4,
in particular from (2.84) and (2.44), how this may be accomplished: One simply
replaces the projection [\II~] in (2.83) by (a constant times) the parity operator P.
 2 Quantization on Flat Space 141

This leads to the definition of the Weyl quantization of a suitable function 1 on

T*Rn as the operator on L2(Rn) given by

(2.91)

The normalization has been chosen so that Q]i(1PlRn) = n. Note that at least
in a heuristic sense, Q]i (8) = (Jr h.)-n P (where 8 is the Dirac delta function on
T*Rn ~ R2n), which places the parity operator in a remarkable light.
Since the Fourier transform will play an important role in what follows, we
choose the Schwartz space
(2.92)

as the Poisson algebra to be quantized. Clearly, the closure of §to in the sup-norm
is ~o = Co(T*Rn). We define (2.91) for 1 E S(T*Rn); it is immediate that Q]i
maps S(T*Rn, R) into ID(L 2(Rn»lR.
We will shortly see that Q]i (f) E lDo(L2(Rn». Given our motivation for
constructing Q]i, the following comes as no surprise.
Theorem 2.5.1. Let a O and a h be as in (2.80) and (2.81), respectively. For each
(M, w) E Sp(n, R) ~ R 2n and all 1 E S(T*Rn), one has

Q]i (a?M.w)(f» = afM.w)(Q]i (f). (2.93)

The proof is similar to that of 2.4.3, with (2.44) replacing (2.84). o
Corollary 2.5.2. Let the classical Hamiltonian h be an arbitrary real polynomial
on T*Rn 01 degree::: 2 in (p, q). The quantum Hamiltonian Hh := Q]i (h) (see
(2.41)) is well-defined as an unbounded operator on the domain S(Rn), on which
it is essentially self-adjoint. With the one-parameter automorphisms a~ and a~
defined as in 1.(2.13) and (2.88), respectively, one has
Q]i (a~(f» = a~(Q]i (f). (2.94)
Equation (2.90) is valid (and proved by the same method) also if Qg is replaced
by Q]i, which settles the domain and self-adjointness issues. The corollary then
follows from Proposition 2.2.1. •
The notation Q]i used here and in (2.41) will be justified shortly.
Weyl quantization may be rewritten in various ways. Firstly, one has
w f dnpdnq w
(2.95)
Qh (f) = JPlRn (2Jrnf I(p, q)Oh (p, q),

where O]i (p, q) E ID(L2(Rn» is defined by

o]i (p, q)\II(x) := 2ne2ip(x-q)/h\ll(2q - x). (2.96)
The function 1 may be recovered from Q]i (f) by the formula
I(p, q) = Tr Q]i (f)o]i (p, q). (2.97)
142 II. Quantization and the Classical Limit

This equation may be proved by noting that Q"ii (f) is trace-class for! E S(T*Rn)
(see below), so that Q"ii (f)n"ii (p, q) is trace-class as well, because n"ii (p, q) is
bounded. If K(·,·) is the kernel of Q"ii(f)n"ii(p,q), its trace is dnx K(x,x), J
which easily leads to (2.97).
More generally, the Weyl symbol a/iw LAJ of an operator A E ~(L2(lRn» is a
distribution in $'(T*Rn) defined by
atLAJ(f):= (21f1i) nTrAQ:r(f). (2.98)
If at LAJ is a locally integrable function, we see from (2.95) that one may write
at LAJ(p, q) = Tr An:r (p, q). (2.99)
Comparing this with (2.97), for! E S(T*Rn) one infers that
Q:r (a/iw LAJ) = A. (2.100)
(Using distribution theory it is possible to make sense of this equation even when
! E S'(T*lRn).) Hence at is the inverse of Qr

--- i
Analogously to (2.52), we can write the Gelfand transform of Q"ii (f) as
dnpdnq
Q"ii(f) (1/1) = (2 )n W/iL1/IJ(p,q)!(p,q), (2.101)
PIll" 1f
where the (real-valued) Wigner function is given by
W/iL1/I J(p, q) = li-n(\{J, n"ii (p, q)\{J). (2.102)
Since n"ii (p, q) is 2n times a unitary operator, the Cauchy-Schwarz inequality
implies that IIW/iL1/IJlloo :::: (2j1i)n (if \{J had not been normalized, the bound
would contain an additional factor II \{J 11 2 ). It is then easy to show that W/i L1/1 J E
L2(T*lRn) n Co(T*Rn). The expression (2.102) is often written as

W/iL1/I J(p, q) = ( dnv eipv\{J(q + kliv)\{J(q - kliv). (2.103)

JIll"
It may be inferred from (2.102) that Q"ii is not positive, since there exist vectors
\{J for which W/i L1/1 J is not positive definite. For such \{J, the Wigner function
may not even be in L 1(T*lRn). Here Berezin quantization is much better behaved.
Comparing (2.101) with (2.52), one sees that the Wigner function W/i L1/1 J in Weyl
quantization replaces the positive definite expression (p, q) f-+ Ii-n p( 1/1 q , 1/1) t
(whose L1-norm is 1 by (1.8» in Berezin quantization.
It follows from (2.98) and (2.102) that for a unit vector \{J E L2(Rn) one has
a{ L[\{J]J = lin W/i L1/1 J, (2.104)
or, by (2.100),
(2.105)
Consequently, the transition probabilities 1.(2.65) in p(~o(L2(lRn») may be
expressed in terms of the overlap of the pertinent Wigner functions as

p(p,a) = lin i
TO Ill"
dnpdnq
(2 )n W/iLpJ(p, q)W/iLaJ(p, q);
1f
(2.106)
 2 Quantization on Flat Space 143

note that the integral on the right-hand side is well-defined, since we have just seen
that Wn E L 2(T*JRn).
The image of Q~ in !B(L2(JRn» is best studied by rewriting (2.91) as

Q~ (f)\II(x) = 1
PIR"
dn d n
P Y eip(x-y)/n f (p, 1(X
(211 li)n
+ y») \II(y). (2.107)

In other words, Q~ (f) is an integral operator

Q~ (f) \II (x) f dny KfLfJ(x, y)\II(y);

= (2.108)
JIR"
Kf LfJ(x, y) := Ii- n l«x - y)/Ii, 1(x + y». (2.109)

1 E S(TJRn) of f
, 1
Here the partial (fiberwise) Fourier transform E S(T*JRn) is
dnp.
f(v, q):= - - e 1PV f(p, q). (2.110)
T* IR"
q
(211)n

Proposition 2.5.3. The map Q~ is an isomorphism between S(T*JRn ) and the

space 113 2(L 2(1Rn» ofHilbert-Schmidt operators on L 2(JRn) with kernel in S(JR 2n ).
This is immediate from the above expressions.
Corollary 2.5.4•
•
• The image Q~(sito) is a norm-dense subalgebra offl! = lJ3o(L2(JRn», and
therefore acts irreducibly on L2(JRn) .
• The quantization Q~ is non degenerate (cf 1.1.2).
Finally, we rewrite (2.91) as

Q~(f) = f dnudnv leu, v)U;' (Exp(-uQ + vP». (2.111)

JIR"
For f E S(T*JRn) we have defined the symplectic Fourier transform I E S(JR2n)
by inverting

(2.112)

Hence we see (with Weyl) that Q~ corresponds to a particular operator ordering,

in which the function (p, q) ~ exp(iuq - ivp) on T*JRn (smeared with a test
function) is quantized by the operator U;,(Exp(-uQ + vP» on L 2(JRn). Ignoring
the test functions, one may repeatedly differentiate with respect to u and v; the
linearity of Q~ then indicates that polynomials P on T*JRn are Weyl-quantized
by (2.37). An interesting corollary to (2.111) is
Proposition 2.5.5. The Schrodinger representation (2.17) is irreducible.
If U;, were reducible, by Schur's lemma there would exist E E IJ3(L2(JRn» such
that [E, U;, (u, v, z)] = 0 for all (u, v, z) E Hn. Equation (2.111) and the definition
of a weak integral then imply that [E, Q~ (f)] = 0 for all f E S(T*Rn). But we
saw in 2.5.4 that Qr (S(T*Rn» acts irreducibly on L2(Rn); cf. I.2.2.2. •
144 II. Quantization and the Classical Limit

2.6 Strict Quantization and Continuous Fields on Flat Space

In this section we show that Weyl quantization is strict, and even continuous, like
its Berezin counterpart. The continuous field of C*-algebras generated by Q:i or
Qg will be described in terms of the Heisenberg group fin.
Theorem 2.6.1. The Weyl maps Q:i define a strict deformation quantization of
!i(~ = S(T*JRn, JR) (with Poisson bracket 1.(2.24)) over 1 = JR, with 2(1i = 2( :=
~o(L2(JRn»for Ii f. O.

Given that it is strict, the fact that Q:i is a deformation quantization follows
from 2.5.4. A key ingredient of the proof of strictness is an estimate we borrow
from the theory of pseudo-differential operators.
Lemma 2.6.2. There exists a constant C > 0 such that for all f E S(T*JRn)
II Qr (f) II ::::: CllfII2n+l,00, (2.113)
where,for mEN (cf (2.74)),
IIfllm,oo:= E II a; a: flloo. (2.114)
Ictl+llll:::m
Here actP aPIctl ... actn.
'=
• Pn'
similarly for afJ
q.
This lemma is useful also for Ii f. 1, because Q:i (f) = Qf
(fli), with
fli(p, q) := f(lip, q). Indeed, it now rapidly follows that Ii ~ Q:i (f) is con-
tinuous as a function from JR\{O} to ~o(L2(JRn»; this implies the continuity
of Ii ~ IIQ:i(f)1I for Ii f. O. Also, (1.2) and (1.3) follow straightforwardly
from (2.113) by computing f 'Ii g. As in 1.1, this is defined by the property
Q:i (f)Q:i (g) = Q:i (f 'Ii g), and can be computed from (2.109).
To prove continuity at Ii = 0, we use the following facts. Firstly, a simple
computation shows that the Wigner function (2.102) of the coherent state (2.46)
is a Gaussian:

(2.115)

Secondly, the connection between Weyl quantization and Berezin quantization is

given by
(2.116)
where * is convolution in T*JRn :::::: JR2n. This may alternatively be written as
Qg(f) = Q:i (e~lil\2n f), (2.117)
where ~2n is the Laplacian on T*JRn :::::: JR2n .
Proposition 2.6.3. For each f E S(T*JRn ) the function
Ii ~ II Qg(f) - Q:i (f)11
is continuous on lR. That is, the Weyl and Berezin quantizations of S(T*JRn) are
equivalent.
 2 Quantization on Flat Space 145

This follows from (2.116) and an application of Lemma 2.6.2. Note that
n
continuity at = 0 simply follows from
lim II QJi (f) - Qf (f) II = 0 (2.118)
n-->O

(from the same lemma), since Qii (f) = Qg (f) = f·

The continuity of n ~ II QJi (f) II at n= 0 (and, indeed, at any fi) is now obvious
•
from 2.6.3 and 2.4.1 (or 1.3 .6), finishing the proof of 2.6.1. •
A different and much more general proof of Rieffel's condition for Weyl quan-
tization will be given in Theorem 111.3.11.4. For now, we return to the Heisenberg
group. One may extend (the inverse of) (2.112) to obtain an isomorphism be-
tween S(Hn) and S(lJ~). Thus the (symplectic) Fourier transform j E S(Hn) of
f E S(~~) is defined by
~
fe u v s) '=
" •
1 f)~
d n pdnqdc e,(uq-pv-cs)f(p
(2rr)2n+l
. q c).
' ,
(2.119)

What follows is a special case of a general construction explained in 111.1.7. One

can define an associative product· on S(Hn) by convolution, i.e.,

j. g(u, v, s):= f_ dnu'dnv'ds' j«u, v, s)· (u', v', S,)-I)g(U', v', s'), (2.120)
lil.
as well as an involution * by
/*(u, v, s) = feu, v, S)-l. (2.121)
A representation U of Hn on a Hilbert space H defines a linear map rr : S (Hn) ~
IJ3(H) by

rr(j):= f_ dnudnvdt j(u, v, s)U(u, v, s). (2.122)

lu o

Using (2.120), one easily checks that any representation U(Hn) thus defines a
representation rr of S(Hn) as a * -algebra.
We firstly use this construction with H = it:= L2(Hn,dnudnvds),andU(Hn)
defined by
U(u, v, s)W(u', v', s') = W«u, v, S)-I . (u', v', Sf». (2.123)
It is clear that the ensuing representation ir(S(Hn» defined by (2.122) is faithful.
One now puts a norm on S(Hn) by
11111 := lIir(j)II; (2.124)
this is evidently a C* -norm. The completion of S(Hn) in this norm is denoted by
C*(Hn). All representations of the convolution algebra S(Hn) extend to C*(Hn)
by continuity. Now recall Definition 1.2.1 and (2.14).
Proposition 2.6.4. Define rr5 : C*(Hn) ~ Co(T*Rn) by rr5(j) := fro, extended
from S(Hn) to C*(Hn) by continuity; this yields a representation of C*(Hn) on
146 II. Quantization and the Classical Limit

L2(T*JRn) by multiplication operators. For n -# 0, define a representation n% of

C(Hn) on L2(JRn) by putting U = U% in (2.122); see (2.17).
The triples (C*(Hn), {m h , niJilEIR) and (C*(Hn), {mil, nKlnEoUI/Z), where
mo := Co(T*JRn) and mil := lBoCL2(JRn» for n -# 0, are continuous fields of
C* -algebras.

Analogous to (2.111), one derives the remarkable relation

(2.125)
One may then imitate the method of proof of Rieffel 's condition in Theorem 2.6.1,
concluding that the function n r-+ IIn;;(i) II lies in C(JR) for i E S(Hn). Moreover,
one infers from Lemma 2.6.2 and the fact that fr!! decreases rapidly in that thisn
function even lies in Co(JR). Since n,7 is continuous, this property holds for any
i E C*(Hn). Hence condition l.2.1.1 is satisfied.
Consider the Hilbert space H := L 2 (JR;(2n)- 2n dnlnl n) ® 'B2(L2(JR n»; el-
ements of H are functions on JR taking values in lB 2(L 2(JRn», with inner
product

JIR ~
(\II, <I» = [
(2n) n
Inln Tr \II (n)* <I> (n). (2.126)

°
For q, E S(Hn) C if and n -# one then defines the operator Wq,(n) on L2(JRn)
by Wq,Cfi) := n%Cq,). We know from Proposition 2.5.3 that Wq,(n) is a Hilbert-
Schmidt operator. An explicit calculation, using (2.109), shows that W : S(Hn) --+
H is unitary, so that W can be extended to a unitary isomorphism from if to H.
Writing n := WiT W* of C*(Hn), the point is now that
(2.127)

for all j E C*(Hn). Using (2.122), (2.123), and (2.17), this is initially proved
for j E S(Hn), and extended to C*(Hn) by continuity. The product of n%(j) E
lB(L2(JRn» and \II(n) E 'B2(L2(JRn» lies in 'B 2(L 2 (JRn», because lB 2 (H) is a
(two-sided, nonclosed) ideal in lB(H).
Condition 1.2.l.2 now follows, since from (2.127), (2.124), the unitarity of W,
and 1.2.1.1 just proved, one has

lIill = sup IIn%(i)ll. (2.128)

IlEIR
It follows from 2.5.4 and (2.125) that ni,(S(Hn» is dense in 'B o(L 2(JRn». For
n -# °
one therefore has
(2.129)
since the left-hand side is norm-closed by 1.1.3.10.4. This is consistent with (and
could alternatively have been derived from) 2.1.4, 111.1.7.5, and 1.2.2.2.1. Similarly.
since n5(S(Hn» is dense in Co(T*JRn), by 1.1.3.10.4 one has

»
n5(C*(Hn = Co(T*JRn). (2.130)
 2 Quantization on Flat Space 147

The results just proved imply that S(Hn ), regarded as a subspace of nllEiR Qtll,
satisfies the three conditions in Proposition 1.2.3, Moreover, it is obvious from
property 1.2.1, the definition of the norm in C*(Hn ), and the continuity of each
n
representation rrK that the function 1-+ IIrrK(j) - rr%(g) II lies in Co(lR) for each
j E C*(Hn) and g E S(Hn). In view of the uniqueness part in the statement of
Proposition 1.2.3, the continuous field determined by S(Hn ) C nt!EiR Qtll through
1.2.3 therefore coincides with the field ( C*(Hn), {Qtll, rr/~lnEiR)'
The statement about C* (Hn) is obvious from the comment preceding (2.18). •

Theorem 2.6.5. The quantization maps Q:i of Weyl and Q~ of Berezin both
satisfy the assumptions of Theorem 1.2.4, and therefore lead to a continuous quan-
tization ofT*Rn (cf 1.2.5). The continuous field ofC* -algebras determined by Weyl
quantization according to Theorem 1.2.4 coincides with the one determined by
Berezin quantization, and is equal to the continuous field ( C*(Hn)' {Qtll, rr%JhEIR)
of the C*-algebra of the Heisenberg group.

First observe that limll-+±oo I Q:i (f) II = 0 for all f E S(T*Rn). This is most
easily proved by (2.108), (2.109), and the inequality IIA II ::: IIA 112; see the comment
after 1.1.6.5. Combining this with Theorem 2.6.1 implies that the first claim in 2.6.5
holds.
J
Similarly, it follows from (2.51) that II Qg (f) II ::: n-n f, so that for f E
C~(T*Rn) one has limll-+±oo II Qf(f) II = O. With Theorem 2.4.1, this leads to
the second claim in 2.6.5. The continuous fields determined by Weyl and Berezin
quantization then coincide by Propositions 2.6.3 and 1.2.3 (used in the context of
the proof of 1.2.4).

Lemma 2.6.6. The continuous field determined by S(Hn ) through Proposi-

tion 1.2.3 coincides with the continuous field determined by Weyl quantization
according to Theorem 1.2.4.

It is clear from Proposition 2.6.4 that S(Hn) satisfies the three conditions in
1.2.3. Similarly, we know from the part of the proof of 2.6.5 that has already been
given that the assumptions in 1.2.4 are met. Now note that for any compact set
K cRone may choose j E S(Hn) such that frll does not depend on n for n E K.
This shows that the second field defined in 2.6.6 is contained in the first.
Conversely, let A E nllEK Qtt. lie in the first field. It then satisfies the first
("if") condition in Lemma 1.2.2, where each B Il' is of the form n
1-+ rrK(jt()

for some jll' E S(Hn). Hence for each n'

E K there exists a function p:
E
S(Hn) and a neighborhood Nt( such that IIAh - rr%(jIl')1I < E for all n E Nil'.
Employing the partition of unity in the proof of 1.2.2, define C E nllEK Qth by
Ch := Lj u j(n)Q:i (!r~j)' Since f~} lies in S(T*Rn), the section C lies in the
second field because of 1.2.1.3. As fh j E S(!J~), one can choose the neighborhoods
Nil} small enough so that Ilf hj (., n) - fllj(., nj)1I2n+l,oo < E/C for all n E Nil};
cf. (2.113). Using (2.125) and (2.113), one finds that II All - CIlIl < 2E uniformly
148 II. Quantization and the Classical Limit

on K. Since both fields vanish at infinity, this shows that A lies in the second field
in 2.6.6. The claim follows. •
Theorem 2.6.5 follows from this lemma, since by the proof of Proposition 2.6.4
the first field in 2.6.6 is (C*(Hn), {mh, n!.lnEIR). •
Corollary 2.6.7. The restriction ofthe continuous field of2.6.5 to R\ {OJ is trivial:
if A : R\ {OJ ---+ ~o(L 2(Rn» is in Co(R\ {O}, ~o(L2(Rn))), then A is the restriction
of some element ofC*(Hn) (seen as an element ofTIhEIR mh) to R\{O}.
As in the paragraph following 2.6.2, for Ii =1= 0 the map Ii ~ QJ; Ur") is
continuous as a function from R\{O} to ~o(L 2(Rn». The claim then follows from
2.6.5 and the proof of 1.2.3. •
A fascinating perspective on Theorem 2.6.5 will be given in III.3.12.

2.7 The Classical Limit of the Dynamics

We turn our attention to the connection between classical and quantum dynamics
on flat space. Equation (2.94) does not hold if h ¢ p:::2; for general Hamiltonians h
one merely has asymptotic results. For the moment we proceed in a more general
context, and consider a general strict quantization Qh, defined with respect to some
m~ C; Co(T*Rn) and mh C; ~(L 2(Rn)). The sharpest convergence occurs when
h itself lies in m~; then H" := Qh(h) is in m". We use the notation of (2.88) and
preceding text.
Proposition 2.7.1. The flow of h E m~ is complete. For f E m~, assume that
apU) E m~ for all t. For any strict quantization Qh (such as Q" = QJ; or
Qh = Qf,), for all fixed t one then has
lim IIQ,,(apU» - a~(Q,,(f))ll = O. (2.131)
h->O

For Qh = Qf, and Q" = QJ; we had m~ = C~(T*Rn) and m~ = S(T*R n ),

respectively; since the Hamiltonian flow is smooth the assumption that at (f) E m~
is therefore satisfied in those cases.
The completeness of the flow of h follows, by a standard argument, from the
fact that its Hamiltonian vector field ~h is bounded on T*Rn (the components of
~h in canonical coordinates are themselves in m~). To prove (2.131) we write

Qh(a?(f» - a:'(Q,,(f» = lot ds :s a;'-s(Q,,(a.?(f)))

= lot ds a;'-s ( Q,,({h, a?um - ~[Qh(h), Qh(a?(f))]) . (2.132)

Using the fact that automorphisms are norm-preserving, we therefore see that the
norm II Qh(a?U)) - a~(Qh(f)1I is bounded by

1t ds II Qh({h, a?um - ~[Qh(h), Qh(a?u))]ll·

 2 Quantization on Flat Space 149

By (1.3), the integrand vanishes as Ii ~ O. •

Since most realistic Hamiltonians in physics are not bounded, this instructive re-
sult is of limited practical use. Many physically relevant one-particle Hamiltonians
are of the form
h(
p,q
) = (p - eA(q»2
2m
+ V( q ) , (2.133)

where m > 0, e E JR, the function A : JRn ~ JRn is the magnetic field potential,
and V : JRn ~ R is a scalar potential. It is not necessary to assume that V and
A lie in coo(JRn); for the existence of local solutions (p(t), q(t» to the classical
equations of motion with initial value (p(O), q(O» it suffices that V V and V A
be Lipschitz around q(O). A formal application of the Weyl prescription (2.37)
indicates that h is quantized by the Schrodinger operator (cf. (2.24), (2.23»
( ps eA(QS»2
HI, = h(P;, Q~) = I, - 2m I! + V(Q~). (2.134)

Theorem 2.7.2. Given (p, q) E T*JRn, assume that

• the classical motion (p(t), q(t» with initial conditions (p(O), q(O» = (p, q)
exists for ti < t < t f;
• V and A are c 3 (JRn) in a neighborhood of each point (p(t), q(t»;
• V and A2 are O(exp(x2/2»for x ~ 06.
If Ii < 1, the expression HI, in (2.133) is symmetric on the domain Do consisting
of the span of all coherent states (2.47). If A = 0, the operator HI! has at least
one self-adjoint extension; for arbitrary A, assume this to be the case. By abuse of
notation, let the symbol HI! stand for an arbitrary self-adjoint extension of(2.133 ),
generating the unitary one-parameter group exp(it HI!/Ii) on L 2(JRn). Then, with
the notation (2.88), 1.(2.13), and (2.47),for all t E (ti, tf),for QI! = (and Qf:'
f E S(T*JRn») or QI! = Q~ (and f E C~(T*JRn)), one has

lim (\II~P.q), [Q1!(a~(f» - a;(QI!(f))]\II~p,q)) = O. (2.135)

r,--+o
Since Do is contained in S(JR n ), and the growth conditions postulated on V and
A imply that the multiplication operators V(Q~), Ai(Q~), and A(QD2 map Do
into itself, it easily follows that HI! is indeed symmetric on Do. If A = 0, then HI!
commutes with the conjugation \II 1-+ \II on L 2(JRn); hence it has equal deficiency
indices.
We now write a = (p, q) and R~ = (P;, QD. Given a particular a, we expand
HI! around the solution a(t) of the classical equations of motion aa /dt = {h, a},
with initial value a (0) = a. That is,
(2.136)
with
H(2)(t) := Ho + HI(t) + H2(t), (2.137)
Ho := h(a(t»[, (2.138)
ISO II. Quantization and the Classical Limit

ah s .
Hl(t):= - . (a(t»(R/i - a(t»', (2.139)
aa'
o2h s . s .
H 2(t) := ! . .
aa'oa l
(a (t»(R/i - a(t)Y (R/i - a(t»l , (2.140)

while H 3 (t) is defined as the remainder.

The operator H(2)(t) has a semiclassical interpretation. Firstly, h(a (t» in (2.138)
is just the classical Hamiltonian evaluated at the classical path. This is independent
of t. Secondly, writing S := T*Rn, consider the function h(l) on T S defined by

h(I)(V, a) := dhu(v) = aah. vi. (2.141)

a'
One sees that Hl(t) is obtained from h(l) by a "partial" quantization along the
trajectory a(t):
Hl(t) = h(I)(R~ - a(t), a(t». (2.142)
Secondly, for each fixed t and a define a function h(2)(t) on Tu(,)S by
h(2)(V, t) := !(h")ij(t)viv j , (2.143)
where

(h " )ij(t) = a.2h . (a(t». (2.144)

oa'aa l
Clearly,
(2.145)
Both h(l) and h(2) generate linearized equations of motion, but they do so in a
different sense. The Hamitonian flow a ~ a(t) on S generated by h pushes
forward to a flow (v, a) ~ (v(t), a(t» on the tangent bundle T S. By definition
of the pushforward, one may think. of the latter as follows: if a(t, a) is a one-
parameter family of solutions of the equations of motion on S (where a E (-E, E)
for some E > 0) neighboring a given trajectory a(t) = a(t, 0), and (v, a) E TuS
equals aa(O, a)/aala = 0, then (v(t), a(t» = aa(t, a)/aala = O.
Now, T S is a symplectic manifold in a natural way: The map Bn : T S -+-
T* S (cf. 1.2.3.6) defines a symplectic form w* := - B;w on T S (where w is the
canonical symplectic form on T* S, cf. Definition 1.2.3.8). If (Pi, qi) are canonical
coordinates on S, we denote the coordinates induced on T S by (Pi, iji , Pi, qi);
these stand for the point Pia/api + ijia/oqi E 1(~.q)TS. In terms of these, the
form w* is given by
w* = diji A dPi + dqi A dpi. (2.146)
The associated Poisson bracket is
* af ag af ag
{f,g} = api aiji + 0Pi aqi - f ++ g. (2.147)

Accordingly, the pushforward flow (v, a) ~ (v(t), a(t» on TS is Hamiltonian

with respect to the Poisson bracket (2.147) and the Hamiltonian (2.141).
 2 Quantization on Flat Space 151

Alternatively, if the trajectory a (t) is already known, one can describe the tangent
part of the flow (v, a) ~ (v(t), aCt»~ as a Hamiltonian system in the v-variable.
This is done as follows. Since in the present case the tangent bundle T S is globally
trivial, there is a natural identification of all fibers of T S; in particular, T(1(t) is iden-
tified with T(1 S for all t. The vector space T(1 S is a linear symplectic space, whose
symplectic form W(1 is simply the canonical symplectic form W on S, evaluated at
T(1S; writing v = (p, ij), one has W(1 = diji /\ dpi. The time evolution v ~ vet)
(where v E T(1 S) then coincides with the Hamiltonian flow on T(1 S generated by
the time-dependent Hamiltonian h(2)(t) (regarded as a function on T(1S through
the above identification). The corresponding Hamiltonian equations of motion are
given by

dv = (h(2)(t), v}. (2.148)

dt
Since this system is linear in v, it is solved by
vet) = M(t)v, (2.149)
where the 2n x 2n matrix M(t) is the solution of

dM(t) = lh"(t)M(t) (2.1 SO)

dt
with initial condition M(O) = [2n; here 1 is given by (2.1).
We return to the quantum theory. To understand the nature of HI (t) we define
",(p,q)(t) '= e iS(t)/Il",(p(t).q(t» (2.1S1)
Il cl' Il'

with the classical action

S(t):= 1t ds [~(p(s)q(s) - jJ(s)q(s» - h(p(s), q(s»]. (2.1S2)

We can evidently write this as

"'hP,q)(t)cl = U?,q)(t)"'hP,q) , (2.1S3)
where

(2.1S4)

is the classical propagator. The point is now that the classical equations of motion
and the relation
(2.1SS)

imply that U?,q)(t) is the solution of

d
in dt u?·q)(t) = (Ra + HI(t»U?,q)(t) (2.1S6)

with initial condition u~p,q)(O) = l

152 II. Quantization and the Classical Limit

We now incorporate H2(t). In terms of the metaplectic representation pli

constructed in 2.2.2, we define
uiP,q)(t):= u~p,q)(t)Ul(p, q)pli(M(t»Ul(p, q)*. (2.157)
h h

It follows from 2.2.2 and (2.148)-(2.150) that pli(M(t» is the propagator for
the Hamiltonian H(2)(t) := !(h")ij(t)R~·i R~·j. Indeed, a short calculation, using
dp(M(t»/dt = dpli(M(t)M(t)-1 )pli(M(t», and subsequently (2.150), (2.39), and
(2.32), shows that

ili!!:.-.pli(M(t» = H(2)(t)p"(M(t». (2.158)

dt
Consequently, from (2.156), (2.158), and (2.155) one derives

i Ii!!:.-.
dt U(p,q)
2
(t) -- H(2) (t )U(P.q)
2
(t) . (2.159)

Hence the object

w(p,q)(t) '= U(P·q)(t)W(p,q) (2.160)
h sc· 2 h

satisfies the semiclassical Schrodinger equation

d
i Ii- w~p,q)(t)sc = H(2)(t)WhP,q)(t)sc' (2.161)
dt
We refer to uiP,q)(t) as the semiclassical propagator. This terminology is
motivated by the following result.
Proposition 2.7.3. With HIi, W~p,q>, and W~P·q>Ct)", given by (2.134), (2.45), and
(2.160), respectively, one has
lim lIe-iIHh/IiW~P.q) - wt q)(t)", 11 = o. (2.162)
h-+O
To prove this, we follow the strategy of the proof of Proposition 2.7.1, and write
(with U(t):= exp(-itHh/li»

(U(t) - uip,q)(t)) W~P.q) = -U(t) 10t d u(s)*uiP·q)(s)W~p,q)

ds ds

= -~U(t) l' ds U(S)(Hh - H(2)(s»uiP·q\s)w~{'q), (2.163)

where (2.159) has been used. The existence of the strong derivative d/ds follows
from the growth conditions imposed on V and A. We now insert the expansion
(2.136), and use the explicit form (2.157) to obtain the estimate

II(U(t) - uiP,q)(t»wtq)1I s hIt

10 ds IIH3(S)/'(M(S»W~0,0)1I· (2.164)

Here H3(S) := U1/h(P(S), q(s»* H3(s)Ul/h(P(S), q(s»; this is just H3(S) with
Rl- aCt) replaced by R~. Using (2.39), one finds p"(M(s»Wko.O)(x) to be pro-
portional to li- n / 4 exp( -(Nx, x)/(2h), where N is a nonsingular complex matrix
 2 Quantization on Flat Space 153

(composed from the entries of M) whose real part is positive definite. We then use
the explicit action (2.24), (2.23) of the operators in H 3(s), upon which Taylor's
formula with remainder and the growth conditions on V and A lead to the conclu-
sion that the integrand in (2.164) is O(h3 / 2 ). Hence the left-hand side is O(h 1/ 2 ),
and (2.162) is proved. •
Using (2.157), as well as (2.93) (with (2.80) and (2.81», we obtain

(utq)(t)\II~,q), Q:i (f)U?,q)(t)\II}tq») =

( °
/i (a(M(t),a(t»-l (I)
\II/i(0,0) ' QW ) \II/i(0,0») . (2.165)

A short calculation shows that

( \IIkp,q), Q:i (f)\IIkP,q») = e/it:.'bI/ 4 I(p, q); (2.166)

cf. (2.117). This equation, or a combination of (1.17) and (2.117), implies

lim (\II}(,q), Q:i (f)\II}(,q») = I(p, q). (2.167)

/i---+O
By (2.162) and (2.167) we then obtain

~~ (\II~p,q), a~(Q:i (f»\IIkp,q») = a?M(t),a(t»-l (f)(O). (2.168)

By (2.80), the right-hand side equals I(a(t» = I(p(t), q(t», as (M(t), a(t»
acting on 0 yields just a(t). Theorem 2.7.2 then follows for Q/i = Q:i, since by
(2.167) one has

lim (\IIkP,q) , a~(Q:i (f)\II}(,q») = I(p(t), q(t». (2.169)

/i---+O
For Q/i = Qg we can use 2.4.3 to write
Uk(p, q)*Qg(f)Ui(p, q) = Qg(a?_p,_q)(f», (2.170)

where we have identified (- p, -q) with (H2n , (p, q»-l. We apply this with (p, q)
replaced by (p(t), q(t». An explicit computation establishes that

lim (P\M(t»\IIko,O), Qg(f)P/i(M(t»\IIko,O») = 1(0,0); (2.171)

/i---+O
cf. (1.17). The desired result then follows as for Q:i.
In fact, the above proof for Qg works for Q:i as well; in either case the essential
•
ingredients of the proof are the equivariance of Qg and Q:i under translations in
T*JRn and the fact that the quadratic term M(t) does not contribute to the limit in
(2.135).
It is remarkable that while the classical motion generated by h may be in-
complete, the quantum evolution generated by H/i is defined for all times. Hence
classical incompleteness is generically traded for quantum-mechanical nonunique-
ness, for the self-adjoint extension H/i may not be uniquely determined by the
formal expression (2.133).
154 II. Quantization and the Classical Limit

3 Quantization on Riemannian Manifolds

3.1 Some Affine Geometry
We now replace the configuration space IRn by a general n-dimensional connected
manifold Q. In general, whenever it is convenient to employ (local) coordinates
qi on Q, we will use them; recall that ai := a/aqi.
We start with a geometric structure on the tangent bundle.
Definition 3.1.1. An affine connection on the tangent bundle T Q is a collection
of linear maps V'~ : reT Q) ~ reT Q), defined for each vector field ~ E reT Q),
such that V' f~ = fV'~ and V'd17 = ~(f)17 + fV'(fJ for all f E COO(Q) and all
~,17Er(TQ).

It follows from this definition that in local coordinates the covariant derivative
can be expressed by
(3.1)
where ~ = ~i ai and 17 = 17 i ai are vector fields, and the connection coefficients
r(k are certain functions on Q.
A curve (v(t), q(t» in T Q (covering, as the notation indicates, a curve q(t) in Q)
is called horizontal if V'q(f)V(t) = 0; although the covariant derivative is defined
as acting on vector fields, this condition makes sense because V'q(t) involves only
the behavior of the section it acts on along the curve q(t).
In that case one says that vet) E Tq(t)Q is the parallel transport of v(o) E
Tq(o)Q, and that (v(t), q(t» is a horizontal lift of q(t). Each curve q(t) has a
unique lift i(v,q)(q(t» through a given point (v, q) E TQ. The collection of all
vectors in T(q,v) T Q that are tangent to some horizontal curve through (v, q) forms
the horizontal subspace Tt:'q) T Q of T(v,q) T Q. One may equally well speak of the
horizontal lift i(v,q)(X) of a vector X E Tq Q; this is the unique vector in T(~:q)(T Q)
that projects to X under LT(TQ)-4TQ'
U(v i , qi) are canonical coordinates on T Q (standing for the point Vi ai E Tq Q),
we denote the coordinates induced on T(T Q) by (Vi, iP, Vi, qi); these stand for
the point via/av i + iFa/aqi E Trv,q)TQ. In terms of these, it follows from (3.1)
that horizontal vectors in Trv,q) are of the generic form
_ i j k iii (3,2)
i(v,q)(W, q) - (-rjk(q)w v ,W ,v ,q ),
One has a natural isomorphism T(~:q)TQ c:::: TqQ, under which X E 1(l~:q)TQ
corresponds to L*X E Tq Q; in coordinates, (-r~k(q)wj v k , Wi, Vi, qi) c:::: (Wi, qi).
In contrast, the vertical subspace T(~:q) T Q C T(v,q) T Q consists of all tangent
vectors to vertical curves (v(t), q), which lie in Tq Q. In other words, T(~:q)T Q :=
kerL* n T(v,q)TQ, where L := LTQ-4Q, hence L* = LT(TQ)-4TQ. Such vertical
vectors are of the form (Wi, 0, Vi, qi). Also here one has a natural isomorphism
1(~:q)TQ c:::: TqQ, because T(~:q)TQ = T(TqQ) c:::: TqQ. In coordinates one has
(Wi, 0, vi, qi) c:::: (Wi, qi). Hence the decomposition
T(v,q)TQ = T(~:q)TQ $ T(~:q)TQ c:::: TqQ $ TqQ. (3.3)
 3 Quantization on Riemannian Manifolds 155

An affine connection on T Q defines a vector field ~c on T Q by ~f := fx(X).

The integral curves of ~c are the geodesic flow on TQ. However, the name
("affinely parametrized") geodesic is reserved for a curve in Q that is the pro-
jection of such a flow in TQ under rTQ->Q' It is customary to denote geodesics
by y(.).

Proposition 3.1.2. With Y := d y / d t, a geodesic satisfies the equation

v yY = O. (3.4)

This is obvious from the definition of a horizontal lift and of ~c. •

Putting w = v in (3.2), we see that the coordinate fonn of the geodesic equation
(3.4) is

~yi(t) + r,ik(y(t))yj(t)yk(t) = O. (3.5)

dt
We write y (q, v; .) for the parametrized geodesic starting at y (q, v; 0) = q with
tangent vector y(q, v; 0) = v. Existence and uniqueness of such a geodesic for
small enough t routinely foIIow from the theory of ordinary differential equations.
However, there is no guarantee that a geodesic exists for all t.
An important role in affine geometry is played by the exponential mapping
exp, which is defined through geodesics. It maps a certain set 0 C T Q into Q,
and is defined by

exp(X) := y(rTQ->Q(X), X; I). (3.6)

The set 0 is simply the set of those X for which the geodesic in question is defined
at t = I; this is an open subset of T Q, evidently containing the zero section Q.
The restriction of exp to Oq := Tq Q n 0 is denoted by eXPq. For good global
properties of geodesics a special assumption has to be made.

Definition 3.1.3. A manifold with affine connection is called geodesic ally

complete when all geodesics exist for arbitrary values of the parameter t.

Clearly, Q is geodesicaIIy complete iff 0 = T Q; in other words, for all q E Q

the map eXPq is defined on all of Tq Q. A weaker notion would be completeness at
a point q, meaning that eXPq is defined on Tq Q. The issue of completeness will be
taken up further in the next section, where a special fonn of the affine connection
leads to interesting results in this context.
From the tangent bundle we pass to the cotangent bundle. The cotangent bundle
S = T* Q is equipped with the canonical symplectic fonn 1.(2.23) and the associ-
ated Poisson bracket 1.(2.24). Recall the notation (Pi, qj) := Pidqi for canonical
coordinates on T* Q. The following functions on T* Q will be of basic importance
in what follows. Firstly, a function g E COO(Q, JR.) induces the smooth function

J g := r*g (3.7)
156 II. Quantization and the Classical Limit

on T*Q (with -r := -rT*Q~Q)' Secondly, a smooth vectorfield~ on Q hasasymhol

J~ E COO(T* Q), defined by

(3.8)

in coordinates, if ~(q) = ~; (q)a;, this reads J~(p, q) = p;~; (q). The basic Poisson
brackets between these functions, which comprise the essence of the canonical
Poisson structure on T* Q, are

{Jg, Jii} = 0; (3.9)

{J~, J g} = JH; (3.10)
{J~J' J~2} = J[~1.~21· (3.11 )
These functions and Poisson brackets have a group-theoretical interpretation.
Firstly, regard Cgo(Q, JR) as an abelian group (under addition). The Lie algebra
of this group is the same space, equipped with the trivial Lie bracket. Then (3.9)
shows that the map g H- J g is a Lie algebra antihomomorphism of Cgo(Q, JR) into
COO(T* Q, JR).
Secondly, consider the group Diff(Q) of (smooth) diffeomorphisms of Q with
compact support (that is, a diffeomorphism cp E Diff (Q) is the identity map outside
some compact set). It is possible to equip Diff(Q) with the structure of an infinite-
dimensional Lie group (though not one modeled on a Banach manifold). Since
one-parameter subgroups ofDiff(Q) by definition generate flows on Q, one infers
that, at least formally, the Lie algebra Diff(Q) of Diff(Q) is the set r c(T Q) of
(smooth) vector fields ~ on Q with compact support. In the opposite direction, the
exponential map on Diff(Q) is given by (exp~)(q) = CPt(q), where CPt is the flow
defined by the vector field ~. Unfortunately, with this identification the Lie bracket
in this Lie algebra equals minus the commutator of vector fields; in what foHows
the notation [~l, ~2] stands for the latter (as usual). Evidently, (3.11) shows that the
map ~ H- J~ is a Lie algebra antihomomorphism of Diff(Q) into COO(T* Q, JR).
FinalIy, we can define the semidirect product

QQ := Diff(Q) ~ C;:O(Q, JR) (3.12)

through the natural action of Diff(Q) on Cgo(Q, JR): cP E Diff(Q) maps g E

Cgo(Q) to (cp-l)*g. The Lie algebra of QQ is denoted by QQ. The corresponding
"mixed" Lie bracket is [~, g] = -H; the minus sign reflects the one in cp-l above.
Hence (3.10) shows
Proposition 3.1.4. The map J : ~ g + H- J~ + J g is a Lie algebra
antihomomorphism ofQQ into COO(T*Q, JR).

We describe the Hamiltonian flow on T*Q generated by J g and J~.

Proposition 3.1.5. Define an action Po ofQQ on T* Q by

Po(g) : a H- a - dg(-r(a»;
Po(cp) : a H- (cp-l)*a, (3.13)
 3 Quantization on Riemannian Manifolds 157

and Po(q;, g) := Po(g) 0 Po(q;).lfX E 9Q. then a ~ Exp(tX)a is the Hamiltonian

flow on T* Q generated by Ix.

This is shown by a straightforward computation in coordinates. D

This and similar results will be placed in their proper context in III.2A.

3.2 Some Riemannian Geometry

We now assume that Q is equipped with a Riemannian structure, i.e., with a
metric. The following remarks are mainly intended to establish some notation and
conventions. The metric g provides each tangent space Tq Q with an inner product
~, that is, a bilinear symmetric positive definite map ~ : Tq Q ® Tq Q ~ Q. The
positive-definiteness means that ~(X, X) ::: 0 for all X E TqQ, with ~(X, X) =
o ¢} X = O. This, of course, implies that ~ is nondegenerate. Throughout this
chapter g is assumed to be smooth (C<~).
The length of a parametrized C I curve (c(t) I t E [ti, t f]} is
If
[ (3.14)
i(c):= Ii dt J&:(t)(c(t), c(t»;

this is evidently independent of the parametrization. The length of a piecewise C I

curve is the sum of the lengths of its C 1 pieces. The distance d between two points
in Q is the infimum over the lengths of all piecewise C 1 curves connecting the
points. (If Q were not connected, this definition would apply if the points lie in
the same component; if they don't, the distance is 00.) It is easily shown that this
distance defines a metric on Q in the sense of point-set topology, making (Q, d) a
metric space in that sense.
Let us use the metric to define a bundle homomorphism gn : T* Q ~ T Q. This
maps T; Q into Tq Q, and is defined by the property

~(gU(a), X) = a(X) (3.15)

(where a E Tq* Q and X E Tq Q). The nondegeneracy of g implies firstly that gU

is well-defined by (3.15), and secondly that it is a bijection; its inverse is denoted
by gu : TQ ~ T*Q. The smoothness of g then leads to the conclusion that gU
and gu are diffeomorphisms. One application is the definition of the gradient of a
function:

(3.16)

One writes &j(q) := ~(ai' aj ); the inverse of the matrix (gij(q)} is denoted
by (gij(q)}, so thatgik(q)gkj(q) = at. Ifa = aidqi E TqQ, thengU(a) = aia j ,
with a j = gij (q)a j; hence aj = gij (q)a j. A similar notation is used for general
tensors.
All concepts of the preceding sections apply; recall Definition 3.1.1.
158 II. Quantization and the Classical Limit

Definition 3.2.1. The Levi-Civita connection, or covariant derivative, is the

unique affine connection V' on T Q that is torsion-free in that
(3.17)
and metric in that

~g(1]I, 1]2) = g(V';-1]1, 1]2) + g(1]1 , V';-1]2) (3.18)

for all vector fields ~, 1]1, 1]2.

For the Levi-Civita connection the object r appearing in (3.1) takes the form
r ijk I ilea jgkl
. - zg
.- + akglj - a ).
19jk (3.19)
In this context r is known as the Christoffel symbol.
The Riemann curvature tensor R is defined by
(3.20)
where Rq(~, 1]) : TqR -+ Tq Q. Remarkably, this is a local expression that indeed
defines a tensor. If we write R(X, Y)Zi = R~klZj Xkyl, then from (3.20) and (3.1)
one has

R~kl = akr~l - alr~k + r~krj, - r~lrJ1. (3.21)

Lowering the first index, one has the symmetries
Rijkl = -Rjikl = -Rijik = R klij . (3.22)
The Ricci scalar is defined by
·- gijRkikj·
R .- (3.23)
The Levi-Civita connection leads to geodesics satisfying (3.5) with (3.19). A
set U C Q is called geodesically convex if any two points in U can be joined by
a unique geodesic of minimum length, that lies in U. A neighborhood Uq of q is
called normal if eXPq is a diffeomorphism between some neighborhood of 0 in
Tq Q and Uq. Clearly, a geodesically convex neighborhood is normal.
In local Riemannian geometry one can prove the following
Proposition 3.2.2. Consider the ball B~ := {X E Tq Q I ~(X, X) < E2}. For
each q there exists an E > 0 such that B~ C (\' and U: := eXPq B~ is geodesically
convex.
We will usually drop the E on U~. On a normal neighborhood Uq of q one
can often use normal coordinates qtn) to simplify computations. These depend
on the choice of a fixed orthonormal basis {ei} of Tq Q. By definition, the normal
coordinates of a point y(q, viei; 1) (assumed to lie in Uq ) are qtn) = vi. Obviously,
the normal coordinates of q itself are qin) = 0, and geodesics simply have the form
qin/t) = tv i • One can show that in these coordinates,

gij(q(n) = oij - 1Rikjiq~)qin) + O(q{n). (3.24)

 3 Quantization on Riemannian Manifolds 159

Hence gij(q) = 8ij and akgij(q) = 0, so that the Christoffel symbols r~k vanish
at q. Furthermore,
(3.25)
A fundamental theorem of global Riemannian geometry gives equivalent forms
of completeness. Recall our standing assumption that Q is connected.
Theorem 3.2.3. The following conditions are equivalent:
• (Q, g) is geodesically complete at one point.
• (Q, g) is geodesically complete.
• (Q, d) is complete (as a metric space).
If any (hence all) of these conditions is satisfied, then any two points may be
joined by a minimal geodesic; this is a geodesic whose length equals the distance
between the points.
In view of this theorem, we simply call (Q, g) complete iff it is geodesically
(hence metrically) complete.
The geodesic of the last claim in this theorem is not necessarily unique.
Definition 3.2.4. The cut locus C(q) ofa givenpointq in a complete Riemannian
manifold Q is the collection ofpoints q' in Q for which there exists more than one
minimal geodesic between q and q'.
Global Riemannian geometry yields the following decomposition of Q.
Theorem 3.2.5. In a complete Riemannian manifold Q, let o;ax consist of all
X E Tq Q for which y(q, X; t) is minimal for all t E [0, I]. The cut locus is
(3.26)
where ao;;, is the boundary of 0;;' in TqQ. The set U:;ax eXPq(O;ax) is a
normal neighborhood ofq, which coincides with the set ofpoints in Q that can be
connected to q by a unique minimal geodesic. Hence for each q, Q is the disjoint
union
(3.27)

Heuristically, U:;ax is the largest neighborhood on which normal coordinates can

be defined.
Corollary 3.2.6. Let (Q, g) be complete. For each given q E Q, the set of all
points q' E Q for which there is a unique minimal geodesic between q and q' is
open and dense in Q.

3.3 Hamiltonian Riemannian Geometry

We move on to perturbations of geodesics. Some of this material is interesting in
its own right; other parts will be used in the study of Weyl quantization.
160 II. Quantization and the Classical Limit

Definition 3.3.1. A Jacobi field J along a geodesic y is a vector field satisfying

the equation of geodesic deviation (or Jacobi equation)

(3.28)

This is a second-order differential equation, whose solution J(t) is determined

by the initial data J(O) and VyJ(O).
To derive and interpret (3.28), one looks at a family {y(., a)}aE[O.E] of geodesics,
smoothly depending on a parameter a, such that y (t, 0) = y (t). The value of J
at yet) is defined by J(y(t)) := dy(t, a)/dala = O. The geodesic equation (3.4)
satisfied by Y(', a) for each a leads to VJVyY = O. Since [at, aa] = 0, one
has [y, J] = 0, which, in view of the fact that the Levi-Civita connection V is
torsion-free, implies VJY = VyJ. Combined with (3.20) this results in (3.28).
A Jacobi field along y(q, v;·) is generically denoted by J(q, v; .). In normal
coordinates based at q the equation of geodesic deviation (3.28) at q reads

:t22 Ji(q) + ~R~kl(q)yj(q)y/(q)Jk(q) = O. (3.29)

Our aim is to show that the evolution equations (3.4) and (3.28) may be brought
into Hamiltonian form. Consider the classical Hamiltonian h* on T* Q, defined by
(3.30)

In coordinates this reads

(3.31)

For simplicity we have put a possible mass parameter m equal to 1; cf. (2.133).
Also, we have omitted a possible potential energy from (3.31); the metric tensor
already represents a (static) gravitational field in which the particle moves. The
Hamiltonian flow a 1-+ a (t) on T* Q generated by h* is known as cogeodesic
flow. This terminology is explained by the following
Proposition 3.3.2. Suppose thata(t) satisfies the Hamiltonian equation ofmotion
da(t)/dt = ~h*(a(t)). Then Ya(t) := TT'Q-+Q(a(t)) is a geodesic on Q with
tangent vector field Ya(t) = gP(a(t)). Accordingly, aCt) is equal to the parallel
transport of a along yO'.

This is most easily proved by a coordinate calculation; in a local chart, one

needs to establish that the motion (p(t), q(t)) is such that q(t) is a geodesic with
q(t) = gn(p(t)). This follows from 1.(2.24), (3.5), and (3.19). •

Proposition 3.3.2 suggests that it is more natural to transfer the situation from
T* Q to T Q by the isomorphism gP. Thus the Hamiltonian on T Q, which we
denote by h, has the two equivalent expressions
(3.32)
(3.33)
 3 Quantization on Riemannian Manifolds 161

Proposition 3.3.3. The Hamiltonianflow (v, q) f-+ (v(t), q(t» on T Q generated

by h coincides with the geodesic flow, and is given by parallel transporting the
tangent vector v = v(O) along the geodesic q(t) = y(q, v; t)on Q (withq = q(O»).

In the coordinates (vi, qi), the Poisson bracket on T Q reads

_ ij (af ag af ag mn I af ag ) (3.34)
{j, g} - g avi aqj - aqi av j +g v (ajg ml - amgjl) av n avi .

The claim is then easily derived from (3.5).

Alternatively, one may regard (Pi, qi) as canonical coordinates on T Q, which
•
are related to the noncanonical ones (Vi, qi) by
(3.35)
In either case, the Hamiltonian equations of motion derived from the canonical
Poisson bracket 1.(2.24) come out as
q= v;
Vvv = O. (3.36)
One sees that Q is complete iff the Hamiltonian h is complete in the sense of
Definition 3.1.3.
To find the Hamiltonian form of (3.28) we recall the discussion surrounding
(2.147), which equally well applies to the present case S = T Q. Hence T(T Q) is
a symplectic manifold, and the pushforward of the geodesic flow on T Q to T (T Q)
is generated by h(l), which is constructed from h in (3.33) by (2.141).
Theorem 3.3.4. The Hamiltonian equations of motion on T(T Q) generated by
h(l) take the form (3.36), supplemented by

(3.37)
where we have decomposed X E T(v,q) as X = Xhor + xvcr, and have identified Xhor
and xver with elements ofTq Q in accordance with (3.3) and preceding text. Hence
the Jacobi equation (3.28) along a given geodesic q(.) is Hamiltonian on T(T Q),
ifwe use (3.3 ) in the opposite direction to identify J(t) and V vJ (t) in Tq(t) Q with
a horizontal and a vertical vector in T(q(t).q(t), respectively.
We give a computational proof. The coordinates (ii, iJi, Vi, qi) on T(T Q) (cf.
the paragraph after (3.1» are not canonical with respect to the symplectic structure
on T(T Q); they are related to canonical coordinates (Pi, iF, Pi, qi) by
(3.38)
cf. (3.35). The Hamiltonian (2.141) on TS derived from (3.33), expressed in
canonical coordinates, then reads
(3.39)
162 II. Quantization and the Classical Limit

From (2.147), (3.39), and (3.38) one finds that di/ /dt = Ii. Using the coordi-
nate expressions for horizontal and vertical vectors, and making the identification
with vectors in Tq Q mentioned in the theorem, the time-derivative d / d t may be
converted into a directional derivative along the curve q(t). Using q(t) = v(t),
this leads to the first member of (3.37). Note here that the coordinate expression
of X + VvX, where X E TqQ and VvX E TqQ are embedded in T(q,v)TQ as
horizontal and vertical vectors, respectively, is simply (Vi = v(X i ), iF = Xi).
The proof of the second member of (3.37) is a straightforward but lengthy
computation, which may be simplified by working in normal coordinates on Q
based at q. Using the same simplification, the Hamiltonian equation of motion for
Vi at q is calculated to be dv i /dt + ~Rjklvjvlqk. Converting the time-derivative
into a directional derivative as in the previous paragraph and comparing with (3.29)
then leads to the second member of (3.37). 0

3.4 Weyl Quantization on Riemannian Manifolds

Our goal is the quantization of a suitable subspace of Q(o := Co(T* Q). The most
natural way of doing this is based on a Riemannian generalization of the kernel
(2.109) characterizing Weyl quantization. Hence we start by generalizing the par-
tial Fourier transform (2.110) to the Riemannian setting. The invariant measure on
Q is called JL, the one on the fiber Tq Q is JLq, and the measure on Tq* Q is denoted
by JL~. In coordinates one has

dJL(q) = dnqJdetg(q);
dJLq(v) = dnvy'detg(q);
dnp
d JL * (p) = -:-:-:-::-i:::;:=:=;:::::;: (3.40)
q (211)n v'det g(q)

Here det g(q) denotes the determinant of the matrix gij (q) in given coordinates.
The natural measure on T* Q constructed from (3.40) coincides with the Liouville
measure ILL, since the factors Jdet g(q) cancel. That is, for f E Ll(T* Q) one has

(
JT*Q
dJLL(P, q) f(p, q) = f Q
dJL(q) (
JT;Q
dJL;(p)f(p, q). (3.41)

The fiberwise Fourier transform of a suitable function f on T* Q is the function

j on T Q defined by

(3.42)

where X E Tq X. In coordinates, this simply amounts to (cf. (2.110»

(3.43)
 3 Quantization on Riemannian Manifolds 163

The fiberwise convolution l * g is the Fourier transform of the pointwise product

f g, which gives

l * g(v, q) = f dJ-Lq(v' ) l(~v - v', q)gOv + v', q). (3.44)

Similarly, the Fourier transform of the Poisson bracket 1.(2.24) is

{j, g}(v, q) = i ( dJ-Lq(v' ) lqv - v', q)

JTqQ

a 1
<-
IiI I i a '1 I
~ ]
x [ -.(zv+v) +(-zv+v)-. g(zv+v,q). (3.45)
aq' aq'
Definition 3.4.1. Let M J and M2 be manifolds, and t : MJ "--+ M2 an embedding
(i.e., an injective immersion). The pullback t*T M2 is the manifold

(3.46)

containing M J as a distinguished submanifold (the so-called zero section) through

the identification (0, m) == m; compare 1l1.(2.2).
The normal bundle of the embedding t is the manifold

(3.47)

containing MJ as the zero section by the identification inherited from t*T M 2.

These definitions may be rephrased as follows. Firstly, the pullback t*(T M2) is
justtherestriction T M2 I t(M t ) ofT M2 to t(Mt) C M2; this is the union UmEM l Vm
of the vector spaces Vm := riit2~M,<t(m», with topology inherited from T M 2.
Secondly, the push forward t*(TMt) C TM2 is a subspace of l*(TM2); it is the
union UmEM l V~, where the vector space V~ C Vm consists of all vectors that are
tangent to l(Md. Finally, the quotientt*T M 2/l* T M J is UmEM l Vm/ V~, equipped
with the quotient topology.
The normal bundle is isomorphic to a subbundle of T M2 r M t, but not naturally
so. The following is a fundamental theorem of differential geometry.

Theorem 3.4.2. Let l(M t ) be a closed submanifold of a manifold M 2•

• There exist a tubular neighborhood N'(Md of M t C N'MJ (where MJ is

identified with the zero section), a neighborhood N;(M t ) C M2 of l(M J), and
a diffeomorphism cp : N'(M t ) --+ N;(M J) satisfying rp(m) = l(m) for all
mEM t •

Let, in addition, M2 have a Riemannian metric, and define (T M2 r MJ)-L as the

union UmEM l V,;, where Vm is the orthogonal complement T,(m)l(Md.L ofT,(m)t(M t )
in T,(m)M2 (with topology inherited from T M2) .

• There is a diffeomorphism 7], : N'M t --+ (TM2 r MJ).L such that 7], is linear
on each vector space Vm/ V~ and 7],(Vm/ V~) = V'; for all m.
164 II. Quantization and the Classical Limit

• The tubular neighborhood N'(M I ) may be chosen in such a way that the dif-
feomorphism <p of the first part of this theorem is given by the restriction of
expo7], : N'MI ~ N toN'(M,).

A special case is that of M, C M2 a submanifold of M2, and I the inclusion

map. We further specialize to the case M, = Q, M2 = Q x Q, and I = 0, the
diagonal embedding defined by o(q) := (q, q).

Lemma 3.4.3.

• The normal bundle N 8 Q of the diagonal embedding is isomorphic to T Q.

• Equip Q x Q with the Riemannian metric g Ell g; then V8 = exp 07]1J : T Q ~
Q x Q is given on Xq E Tq Q by
(3.48)

One identifies T(q.q)(Q x Q) with Tq QEElTq Q. The fiber of o*T(Q x Q) ata point
(q , q) is Tq Q Ell Tq Q. The fiber of the push forward bundle 0* T Q at (q , q), on the
+
other hand, consists of all vectors of the type X X, X E Tq Q. Hence NiJ Q ~ T Q
by the definition (3.47). With the metric g Ell g, the orthogonal decomposition of
X +Y has the component i(X + Y)+i(X + Y) ino*TQ and i(X - Y)+i(Y -X)in
1-
(T(Q x Q) r 0(Q» . Hence 7]8 maps Xq E TqQ to IXq+ - IXq E T(q,q)Q x Q,
1 • I

and (3.48) follows. •

To appreciate the following quantization prescription it is helpful to understand

the geometric meaning of the diffeomorphism (3.48): Namely, vi'(q, q') in T Q
is the tangent vector to the geodesic from q' to q at its midpoint.
We are now in a position to define the (generalized) Weyl quantization map Q}[.
We take the dense sub algebra of Co(T* Q) of quantizable functions to be
Qio := C~(T* Q). (3.49)

These are by definition the functions f on T* Q whose Fourier transform j is

in Crgo(T Q); the motivation for this choice will become clear shortly. The space
C~(T* Q, R) is a Poisson subalgebra of COO(T* Q, R); this follows from an in-
spection of (3.45), using the fact that Crgo is closed under convolution and pointwise
multiplication.
The map QJitakes values in Qlll := lBo(L2(Q», where the Hilbert space L2(Q)
is defined with respect to the Riemannian measure Jj, on Q; cf. (3.40).

Definition 3.4.4. The Weyl quantization of f E C~(T* Q) is given,for Ii i= 0,

by the integral operator

Q}[(f)\}J(x):= fo dJj,(y)K,:vLfJ(x,y)\}J(y). (3.50)

For (x, y) tt M(Q) we define K,:v LfJ(x, y) := 0, whereasfor (x, y) E M(Q) we

put
(3.51)
 3 Quantization on Riemannian Manifolds 165

with V8 given by (3.48). Here K is a smooth function on T Q with the following

properties:
• K = 1 in a neighborhood j;{8(Q) C N8(Q) of Q (regarded as the zero section
in TQ).
• K has support in N8(Q).
• K(-V, q) = K(V, q).
For later convenience, we shall in fact assume that for each q the support of
K(', q) is contained in a geodesic ally convex (hence normal) neighborhood of
o E Tq Q. As in 2.5, we write the argument of 'II as x rather than q to avoid
confusion with the argument (p, q) of f.

Proposition 3.4.5. For f E C~(T* Q, JR) one has Q:' (f) E !Bo(L 2(Q»IR'
By definition of C~(T* Q), the kernel of Q:' (f) is in C~(Q x Q), so that
Q:' (f) is a Hilbert-Schmidt operator, hence compact. Also, (3.42), (3.48), and
the symmetry of K in v guarantee that K:' is Hermitian, so that for real f the
operator Q:' (f) is self-adjoint. •

The presence of the cutoff function implies that the kernel K:' is smooth;
K
unfortunately, Q:' (f) seems to depend on the choice of this function (as well as
of j;{8(Q». However, since j has compact support by our choice of lito, there is
a value h.o > 0 (that depends on f) such that Q:'
(f) does not depend on these
choices for Ii E (0, h.o). Namely, h.o is the smallest value of Ii for which h.osupp (j)
lies in j;{8(Q). If the tubular neighborhoods may be chosen as N8(Q) = T Q and
Ni,(Q) = Q x Q, then one may obviously putK = 1. This is possible for Q = JRn
(with flat metric), in which case (3.50) and (3.51) reduce to (2.108), and (2.109),
respectively.
The same conclusion of K-independence formally holds true if j is a distribution
with compact support; if f is polynomial in the momenta pi, the support of j is
localized at the zero section of T Q, and Q:'
(f), now defined as an unbounded
operator on the domain C~(Q), is independent of K for any Ii. This will be further
explored in 3.6.
As in the flat case (cf. (2.101» there is a Wigner function.

Proposition 3.4.6. The Gelfand transform of Q:' (f) is given by

Qf(j)(l{f) = ( dILL(P, q)WhLl{f J(p, q)f(p, q), (3.52)

lpQ
with Wigner function (cf. (2.103))

WIiLl{f J(p, q) = 1 TqQ

dlLq(V) K(liv, q)J(q, v; ~Ii)
--"--.,,......,----;-..,...,....,..
X eipv'II(y(q, v; ~1i»'II(y(q, v; -~Ii», (3.53)

where J is a Jacobian defined in (3.55) below.

 166 II. Quantization and the Classical Limit

To prove this, we initially assume that 'l'T*Q->-Q( supp (f» is contained in a

suitably small geodesically convex set U C P, on which we use coordinates qi .
The linearity of Q~, the fact that 'l'T*Q->-Q( supp (f» is compact, and the existence
of (smooth) partitions of unity on Q then imply the result for general f E 2,l~. We
change integration variables in

Qf(iKl/f) = ( d/-L(q\)d/-L(q2) Ki':' LfJ(q\, q2)I.{J(q\)I.{J(q2): (3.54)

lQxQ

If an arbitrary function F E C( Q x Q) has support inside U xU, we put

fUxU
d/-L(q\)d/-L(q2) F(q\, q2) =

( d/-L(q)d/-Lq(v) J(q, v; k)F(y(q, v; k), y(q, v; -k», (3.55)

lTV
which, with the property J(q,tv;V = J(q,v;kt), defines J in (3.53). The
proposition then follows from (3.54), (3.55), (3.50), (3.51), and (3.43). •

The Jacobian J will be studied in detail in the next section; we will find that
J (q, v, k11,) = 1 + 0(11,2). Also, J will be seen to have the symmetry property
J( -v, q; t) = J(v, q; t), which, with (3.53), confirms that Wfj L1/1 J is realfor real-
valued f. Given that Q~ (f) is bounded for f E 2t~, this property is equivalent to
the self-adjointness of Q~ (f).

3.5 Proof of Strictness

_Recalling Definitions 1.1.1 and 1.2.5, the aim of this section is to prove

Theorem 3.5.1. The map Q~ defined in 3.4.4 is a nondegenerate strict and

continuous quantization of 2t~ = C~(T* Q, lR), so that 2,l0 = Co(T* Q), and
2,lfj = !l3 o(L2(Q» for 11, i=- 0 (with the possible exception of the completeness
condition 1.1.1.4).

It is clear that 1.1.1.4 is not satisfied if the cutoff function K in (3.51) is not
equal to unity. If Q x Q is diffeomorphic to TQ by the map V8 (cf. (3.48», the
quantization Q~ does satisfy 1.1.1.4, since the collection {Q~ (f)} is dense in the
set of Hilbert-Schmidt operators on L2(Q).
The nondegeneracy is obvious: Q~ (f) = 0 implies j = 0 by (3.51), which
implies f = O. Continuity follows from strictness by Theorem 1.2.4.
In the following discussion we will assume that 11, > 0; the arguments for 11, < 0
are a trivial modification. The necessary computations are greatly simplified by
the possibility of localization.

Lemma 3.5.2. Let f, g E 2t~.lfthe projection 'l'T*Q->-Q( supp (f» of the support
of f is disjoint from that of g, then there is nf,g > 0 such that Q~ (f)Q~ (g) = 0
for 11, E (0, nt,g)'
 3 Quantization on Riemannian Manifolds 167

It follows from (3.51) and the fact that j has compact support (particularly
in the fiber direction) that K ~ Lf J (x , z) is nonzero only if the (Riemannian) dis-
tance from both x and z to 'l"PQ-+Q( supp(f) is 0(11.); similarly for K~ LgJ(z, y).
Hence for fixed x, y the kernel J d/L(z) K~ LfJ(x, z)K:' 19J(z, y) of the product
Qf(f)Qf(g) vanishes for sufficiently small 11. < h(x, y),foracertainh(x, y) > O.
Since j and g have compact support also on in the base direction, this vanishing
can be achieved uniformly in (x, y). •

For the reasons stated in the proof of 3.4.6, Lemma 3.5.2 allows us to assume
that 'l"PQ-+Q( supp(f» and 'l"T*Q-+Q( supp(g» are contained in an arbitrarily small
open set U C Q. For U we choose a geodesically convex set Uq (cf. the paragraph
following 3.4.6).
Since (3.28) is a second-order differential equation, for given X, Y E Ty(o)Q
there exists a unique Jacobi field:r for which :r(0) = X and Vy:r(O) = Y. If we
write :r = :ri ai in given coordinates, one may equally well pose the initial condi-
tions :ri(O) = Xi, ji(O) = yi, with unique solution :ri(t). We write Jcj)(q, v;·)
for the Jacobi field with initial conditions

.:Tr.~)(q, v;O) = cS~;

!rr.~)(q, v;O) = 0, (3.56)

and j(j)(q, v;·) for the Jacobi field with initial conditions

~~)(q, v; 0) = 0;
d -. )
( dt.:Tr.'j) .
(q, v;O) = cSj. (3.57)

The n x n matrices M(q, v;t) and M(q, v;t) are then defined by their matrix
elementsM(q, v;t)~ := .:Tr.~)(q, v;t)andM(q, v;t)~:= ~~)(q, v;t),respectively.
These are combined in the 2n x 2n matrix

M2:= ( M(q, ~;t) ~(q, ~;t) ). (3.58)

M(q, v, -t) M(q, v, -t)

Lemma 3.5.3•

• The Jacobian in (3.55) is given (for arbitrary t, as long as the geodesics in

question exist) by

J(q, v; t) = It- n I[detg(y(q, v; t» detg(y(q, v; -t»] 1/2

x detg(q)-ll detM2(q, v;t)l. (3.59)

• If G E C(Q) has support in U, and q E U is such that U is contained in the

image of the exponential map on Tq Q, then

1 u
d/L(q')G(q') = (
JTqQ
d/Lq(v) i(q, v; I)G(y(q, v; 1», (3.60)
168 II. Quantization and the Classical Limit

with Jacobian
i(q, v; t) = It-nl[detg(y(q, v; t»1 detg(q)]!/21 det M(q, v; t)l. (3.61)

To derive (3.55), one passes from the coordinates (qt, q~) to qi, Vi via the expres-
sion q! = y(q, v; t), q2 = y(q, v; -t) (where t = ! in the special case above).
The definition of a Jacobi field implies that

aq~(q, v; t)/av j = J(~)(q, v; -t),

aq~(q, v; t)laqj = .:lr.~)(q, v; -t);
(3.62)
which leads to (3.55). The derivation of (3.60) is analogous. •
We assume the support conditions on f and g stated after 3.5.2, and take
an arbitrary W E L2(Q). From (3.50), (3.51), Lemma 3.5.3, and the property
y(q, v, lit) = y(q, nv, t), we obtain

(w, [QIi (f)QIi (g) - QIi (fg)]w) = 1

U
dJL(q) (
lTqQ
dJLq(v) (
lTqQ
dJLq(v /)

W(y(q, v; nI2»W(y(q, v; -nI2»[(n, q, v, v'), (3.63)

with
[(n, q, v, v') = J(q, v; n12) [n2n i(q, v'; h)

x K:' Lf J(y(q, v; n12), y(q, v'; n»K:' LgJ(y(q, v'; h), y(q, v; -nI2»
- j(q, !v - V')g(q, ~v + VI)] . (3.64)

For a fixed value (q, v, v'), we now make a Taylor expansion of [(n, q, v, v')
n.
in Here [ is a function on T U ® T U, so we may proceed in any coordinate
system.
By evaluating (3.28) in normal coordinates at t = 0 (cf. (3.29» it follows
immediately that jj(q, v, n) = M~(1 + 0(n2» and .Jj(q, v, n) = 8~(1 + 0(n2».
Combined with the explicit form (3.24) of the metric in normal coordinates, we
thus infer from (3.59) and (3.61) that (in any coordinates)
J(q, v; n12) = 1 + 0(n2);
-
J(q, vI ; n) = 1 + O(n2 ). (3.65)
To deal with the terms involving K:'
in (3.64) we use (3.25) and perform a Taylor
expansion of K:' LfJ around the point (y(q, 4v - v'; n12), y(q, 4v - v'; -nI2»),
and of QIi (g) around (y(q, 4v + v'; n12), y(q, 4v + v'; -nI2»). Using (3.51),
the result is then rewritten in terms of the j and g. The 0(1) term vanishes. In
computing the O(h) term, one encounters expressions of the type

a +-.a) Kfi LfJ(q!=y(q'i: v -

(-. W I
vI ;nI2),q2=y(q'i:I v - v I ;-nI2»,
aq: aq~
(3.66)
 3 Quantization on Riemannian Manifolds 169

to be expressed in normal coordinates as indicated above. This is done by in-

verting (3.62) and expanding in powers of Ii. The result is that lin times (3.66)
equals (aiJaqi)(q, 4v - v') + O(Ii). Using (3.45), one then finds that the in-
tegrand of the analogue of (3.63), with Q:i (f)Q:i (g) - Q:i (fg) replaced by
i [Q:i (f), Q:i (g»)/Ii - Q:i ({f, g}), is O(Ii).
In either case one is left with an expression of the type (3.63), with 1 replaced
by a remainder 1(1) of O(Ii). One then replaces the integration variables (q, v, v')
in (3.63) by (q', v, v'), with q' = y(q, v; 1i/2); this introduces a Jacobian, which
is 1 + O(1i2 ), as in the argument leading to (3.65). This Jacobian may be absorbed
into 1(1), which then remains 0 (Ii). Then apply the Cauchy-Schwarz inequality to
the q'-integration, splitting the integrand into "'(q') and the rest. This takes out a
term <Iv dJ,L(q') 1",(q')1 2 )1/2 ::: II '" II. The second factor produced by the Cauchy-
Schwarz inequality is majorized by taking out another factor II '" II, and bounding
the rest of the q' -integrand by taking its supremum over q'. This leads to

1("', [Q:i(f)Q:i(g) - Q:i(fg»)"') Clill'" 112 II ill I,oollgll 1,00

1 :::

x sup [J,Lq(supp(j) n TqU)J,Lq(supp(g) n TqU»), (3.67)

qeV

and a similar inequality for 1("', (i [Q:i (f), Q:i (g»)/Ii - Q:i ({f, g })"')I, in which
the norms II . 111,00 in (3.67) are replaced by II . 112,00' Hence (1.2) and (1.3) follow
as in the proof of 2.4.1.
It remains to prove Rieffel 's condition. Since a very general proof of this property
will be given in Theorem III.3.11.4, we will merely sketch how the proof in flat
space may be generalized.
Firstly, continuity at Ii =1= 0 can be proved in several ways, e.g., by proving
continuity with respect to the Hilbert-Schmidt norm of Q:i (f). To prove continuity
for Ii ---+ 0, we shall construct a positive map Qh', which is equivalent to Q:i in
the sense that the function Ii ~ II Q:i (f) - Qh'(f)II is continuous on R\ {OJ and
limli-+o II Q:i (f) - Qh'(f)II = O. This map may be shown to satisfy (1.1), which
then implies the same for Q:i. In the proof of 2.6.1 we had Qh' = Q:; in the
present case the construction of Qh' is motivated by Q: on flat space, but unlike
Q:i it holds no intrinsic significance on curved spaces.
We define Qh'(f) (where f E Qi~) through its Gelfand transform Qf(j),
defined as a function on IPL 2 (Q), by

Qf{j)(tfJ) = ( dJ,Ldp, q) W:'LtfJ J(p, q)f(p, q), (3.68)

JT*Q
where

WPOS L·I'J ( ) '= Ii- n

Ii 'I' p, q .
I( dJ,Lq(v)
JTqQ (rrli)n/4

K(V, q)J J(q, v; 1)e-iPv/lie-v2/(21i)"'(y(q, v; 1)1 2

, (3.69)
170 II. Quantization and the Classical Limit

where v 2 := ~(v, v), and j is given by (3.61). It is easy to see that (3.69) defines
a bounded operator Q'h'(f) on L2(Q) (see 3.5.4 below). In flat space one has
Q'h' = Qg; in general, (3.69) depends on the cutoff function K, cf. (3.53).

Lemma 3.5.4. The map Q'h' : sii~ ---+ ~(L 2(Q» is positive and takes values in
~o(L 2( Q»jR. It satisfies, for all f E sii~,

lim II Q'h'(f) - Q~ (f) II = 0 (3.70)

n~O

and

lim IIQh'(f)1I
n~O
= IIflloo. (3.71)

The positivity of Q'h' is obvious from (3.69). Since f and K are compactly
supported in q and v, respectively, Q'h'(f) is an integral operator with smooth
compactly supported kernel; hence it is Hilbert-Schmidt and therefore compact.
Self-adjointness is immediate from the reality of Wr.
The proof of (3.70) and (3.71) is very tedious, and will be omitted.
Given this lemma, the corresponding argument in the proof of 2.6.1 leads to

lim IIQ~(f)1I
n~O
= 1111100, (3.72)

and the proof of Theorem 3.5.1 is finished. o

The continuous field of C* -algebras defined by Q:i through Theorems 3.5.1
and 1.2.4 will be identified in 111.3.12.

3.6 Commutation Relations on Riemannian Manifolds

We would like to quantize certain unbounded smooth functions f on the phase
space T* Q. This can be done by the prescription (3.50), (3.51) if f is polynomial
in p. The domain on which the ensuing unbounded operator Q:i (f) is defined is
initially taken to be C;;o( Q), since on this domain the formal manipUlations used in
computing Q~ (f) are well-defined. In this section we examine certain intrinsically
defined functions on T* Q of order zero and one in the canonical momenta (the
Hamiltonian, which is of order two, will be dealt with in the next section).

Proposition 3.6.1. The WeyJ quantizations (in the sense of 3.4.4) of f = Jg (cf
(3.7)) and of f = J~ (cf (3.8)), defined on C;;o(Q) c L2(Q), are given by

Q~ (Jg ) = g; (3.73)
Q:i(J~) = -in(~ +!V. ~). (3.74)
Here g and ! V . ~ (:= ! Vi~i) are multiplication operators.
The computation of Q:i (.. ·)'It(x) is best done in normal coordinates q! based
at x (cf. 3.2). In these coordinates the point q and the vector Xq E Tq Q for which
(exPq<!Xq), eXPq(-!Xq » = (0, v)(see (3.48» are given simply by q! = ~Vi and
 3 Quantization on Riemannian Manifolds 171

x~ = -Vi. Hence from (3.50), (3.51), and (3.43) one obtains

det g(v) 1 •
detg(~v)f(lip, i V )'I1(v),
(3.75)
for the functions f under study this expression will tum out to be independent of
the cutoff function K. For f polynomial in p one has the (oscillatory) integral

f dn-pe -ipv p il ... p"=

-
(2Jl')n
i a) (.
(.1 - ,
8v11
... 18
- .)0
8v 1"
~(n)( v,) (3.76)

where 8(n) is the n-dimensional Dirac delta distribution. This leads to (3.73) and
(3.74). •

Straightforward computation leads to the following "canonical" commutation

relations (valid on the domain C~(Q»:

(3.77)

(3.78)

(3.79)

These reflect the classical Poisson brackets (3.9)-(3.11), in that Dirac's relation
i [ QJi' (f), QJi' (g)] / Ii = QJi' ({f, g}) is satisfied for the functions in question.
The "canonical" commutation relations may be interpreted in terms of a certain
representation Ph of the group (}Q (see (3.12» on L2(Q).

Proposition 3.6.2. The linear action of(}Q on L2(Q) defined by

Ph(g)'I1(q):= e- ig (q)/h'l1(q), (3.80)

dJL(rp-l(q» '11( -l( »

dJL(q) rp q, (3.81)

and Ph(rp, g) := Ph(g) 0 Ph(rp), is unitary, hence a representation. The derived

representation of the Lie algebra gQ is given in terms of the map J (cf. 3.1.4) by

(3.82)

The Radon-Nikodym derivative under the square root exists because rp-l is
a diffeomorphism, under which the locally Lebesgue measure class is invariant.
Given the square root, unitarity is immediate from the definitions. For the remaining
calculation one combines the identity

8i log Jdetg = rb (3.83)

with (3.40) and (3.1). o

172 II. Quantization and the Classical Limit

The representation Ph induces a *-automorphic action a h of 9Q on 2(h

23o(L2(Q» by

at""g)(A) := Ph(CP, g)Aph(rp, g)*, (3.84)

Writing a; := at""O) and a; := a[:d,g)' the kernel of A QJ;(f) (cf. (3.51»

transforms as
d/L(cp-I(X» d/L(cp-l(y»
(a~(K:: LfJ))(x, y) =
d /L(x) d /L(y)
X K,~ LfJ(cp-l(x), cp-l(y»; (3.85)
(aI(K:: Lf J») (x, y) = ei[g(y)-g(x)]lh K:: LfJ(x, y). (3.86)

The classical analogue of this automorphism is evidently given by

a~,g)(f) '= f 0 Po«cp, g)-I), (3.87)
where Po is defined in (3.13).
An isometry of (Q, g) is a diffeomorphism cp for which cp*g = g; an infinites-
imal isometry is a vector field I; on Q for which L~g = 0 (where L is the Lie
derivative ).
Theorem 3.6.3. If cp is an isometry of (Q, g), then
a;(QJ; (f» = QJ; (a~(f) (3.88)
for all f E C~(T*Q) (cf (3.49)), and n small enough so that QJ; (f) is indepen-
dent of the cutoffK (cf the comments following 3.4.4), IfK is invariant under the
(pushforward) action of cp to T Q, eq. (3.88) holds for alln i= 0,
If I; is an infinitesimal isometry whose flow is complete, then, under the same
conditions on f, on the domain C~(Q) one has

(3.89)

If cp is an isometry, the Radon-Nikodym derivatives in (3.85) equal unity.

Equation (3.88) then follows from (3.85), (3.51), (3.42), (3.48), and the property
(3.90)
which, because cp is an isometry, holds by the definition of the exponential map.
Equation (3.89) follows from (3.88), (3.82), Proposition 3.1.5, and the following
interesting result. •
Proposition 3.6.4. If a vector field I; on Q is complete, then QJ; (J~) is essentially
self-adjoint on the domain C~(Q).
By (3.82) we might as well consider dph(I;). Let CPt be the flow generated
by 1;; since I; is complete, the flow exists for all t. Then (3.81) defines a one-
parameter unitary group t f-+ Ph(CP/) on L2(Q). A routine calculation shows that
dp,,(cp/)\II/dtlt = 0 exists for all \II E C~(Q), and equals 1;\11. Hence C~(Q) is
 3 Quantization on Riemannian Manifolds 173

contained in the domain of the generator of the unitary group, and this generator
equals -i~ on C;:O(Q). Furthennore, by (3.81), P/i(cpt) leaves the dense domain
C;:O( Q) invariant for all t, since CPt is a diffeomorphism. A lemma in functional
analysis states that if t r-+ exp(i t H) is a unitary group in a Hilbert space that leaves
a dense linear subspace 1)0 c D(H) invariant, then H is essentially self-adjoint
on 1)0' This implies the proposition. •

3.7 The Quantum Hamiltonian and its Classical Limit

We extend Proposition 3.6.1 to the most important function on T* Q that is
quadratic in the momenta.
Definition 3.7.1. Given a Riemannian metric g on Q, the Laplace-Beltrami
operator ~ is an elliptic second-order differential operator on Q, defined by

(\II, ~<I» = - £ d/L(q)f,q(V\ll(q), V<l>(q», (3.91)

where \II, <I> E C;:O(Q); the gradient V is defined in (3.16).

In coordinates, one has

~ = gij\1·a· = _1-a.(Jdetggija.)· (3.92)

'J Jdetg' J '

here ai acts on everything to its right, including the (omitted) \II. In flat space ~
clearly reduces to the Laplacian.
Proposition 3.7.2. The Weyl quantization of the Hamiltonian h* (cf (3.31)) is
given (on the domain C;:O(Q) C L2(Q») by
Hti := QJi (h) = -~n,z(~ - ~R). (3.93)
Here the Ricci scalar R (cf (3.23)) is seen as a multiplication operator.
The proof of (3.93) follows the same steps as in 3.6.1; here one additionally
uses (3.25). D
The functional analysis of the first tenn of (3.93) is given by a result of the same
type as 3.6.4, but somewhat deeper.
Theorem 3.7.3. When (Q, g) is complete (cf 3.1.3), the Laplace-Beltrami
operator is essentially self-adjoint on C;:O(Q).
The symbol ~ stands for the differential operator (3.92) defined on the domain
C;:O(Q); its closure is denoted by ~. We can look at (3.91) as the definition of ~
as a quadratic fonn with initial domain C;:O(Q).
It is easily verified that ~ is symmetric. It is evident from the definition (3.91)
that~, and therefore ~, is negative. It follows that ~ has equal deficiency indices,
so that self-adjoint extensions exist. (This conclusion also follows from the fact
that ~ commutes with the conjugation \II r-+ \II on L 2 (Q).) The domain D(~) of
174 II. Quantization and the Classical Limit

II is the set of vectors \II E L 2(Q) for which there exists a sequence \II j E C;:O(Q)
such that \II j --+ \II, and II \II j converges to an element in L 2(Q); the latter is then
by definition II \II. The domain D(ll *) of the adjoint is the collection of vectors
in L2(Q) C V(Q)' (the distributional dual of C;:O(Q) = V(Q) with the Schwartz
topology) for which II * \lilies in L 2(Q); here II * is given by the expression (3.92),
understood in the sense of weak (distributional) derivatives. The theorem states
that D(ll) = D(ll *) if Q is complete.
The following fact will be used: If A is a positive closed operator, then the
dimension of ker (A - A) is constant for A E C\[O,oo). Suppose II is not self-
adjoint. Then the deficiency indices are nonzero (and equal), so that the equation
II *\11 = i \II has a nonzero solution in D(ll *). By the above fact, there is a nonzero
solution \II = \III of ll*\II = \II. The theory of elliptic PDE's shows that \III E
C oo ( Q) ("elliptic regularity"), so that the weak derivatives in II * are actually strong
ones. Abbreviating the right-hand side of(3.91) as (V\II, V¢), the idea of the proof
is to write

(V\III. V\III) = -ell *\1110 \II]) = -(\III. \III) ::: O.

forcing V\II] = 0 (in L2(Q» and therefore ll*\II1 = \III = O. However, the partial
integration leading to the first equality is not a priori justified unless \III has compact
support (hence if Q were compact the proof would be finished here). Hence one
uses the following device. Pick a fixed qo E Q, and define a family of functions
jk : Q --+ [0, I] by jk(q) = j(d(q. qo)/k) (k EN), where d is the distance
function on Q x Q (see 3.2), and j : [0, 00) --+ [0, I] is a smooth cutoff function
that is 1 in [0, I] and 0 on [2, 00). At this point the completeness of (Q, g) is used:
It follows from Theorem 3.2.3, guaranteeing metric completeness, that each jk has
compact support (since a closed and bounded set in a finite-dimensional complete
metric space is compact). One clearly has jk --+ 1 pointwise.
The distance d(q, qo) is a differentiable function of q except at qo and at the
cut locus C(qo) (cf. 3.2.4). By Corollary 3.2.6 the set C(qo) is of JL-measure zero,
so that each component of Vd(·, qo) is well-defined as an element of L~C<Q). The
triangle inequality leads to Id(ql, qo) - d(q2, qo)1 ::: d(ql, q2)' This Lipschitz
condition implies that in normal coordinates centered at qo the metric is absolutely
continuous with respect to each variable, with Id;d(q, qo)1 < 1. Hence, by the
chain rule, Id;jk(q)1 ::: 11/1100/ k, so that
(3.94)

Trivially, Uk\lll, jk\ll]) :::: O. Moving the second jk to the left, replacing the
second \III by II * \III, and performing a partial integration (now allowed, since
j;\II1 has compact support), one rewrites this inequality as

IIjk V \IIIII~::: 21(\11 1 V jko jk V \111)1,

where II . 112 is the norm derived from (, ). The Cauchy-Schwarz inequality then
yields IUk V \II tll~ ::: 211 \III V jk 112 IUk V \II db which in tum leads to the bound
IUk V \111112 ::: 211\11] V A112. Accordingly, by (3.94) one has limk IUk V \11]112 = O.
 3 Quantization on Riemannian Manifolds 175

Fatou's lemma and limk jk = I (pointwise) then imply that II V \11 1112 = 0; as we
have seen before, this implies \III = O. •

Self-adjointness of Q';i (h) in (3.93) then follows from a mild assumption.

Proposition 3.7.4. Let (Q, g) be complete with Ricci scalar R bounded. Then
Q';i (h) is essentially self-adjoint on C~(Q) and self-adjoint on D(I::!.).
This is immediate from the Kato-Rellich theorem on perturbations of self-
adjoint operators. 0
In order to generalize the proof of Theorem 2.7.2 to Riemannian spaces (Q, g),
we have to make a simplifying assumption, namely that Q :::::: an as a manifold.
If (Q, g) is complete, by Theorem 3.2.5 this would follow from the assumption
that the cut locus C(qo) is empty for some point qo E Q. The globally defined
coordinates Xi on an may then taken to be normal coordinates based at qo. However,
given the assumption that Q :::::: an, we need not assume that the cut loci defined by
the metric are empty, and neither is it necessary that the classical motion defined
by h be complete. (The case that (Q :::::: an, g) is complete and R is bounded is, of
course, covered; cf. 3.7.4.)
In the present case the notation L2(Q) stands for L 2 (a n ,dn xy'detg(x». We
can define normalized coherent states in L2(Q) by

\IIhP,q)(x) := (n n)-n/\det g(X»-1/4 eip(X-~q)/lie-(x-q)2/(21i); (3.95)

this slightly generalizes (2.47).

We shall merely assume that g is a metric on Q = an for which gij and its
derivatives are o (exp(x2 /2» for x ---+ 00. Ifn < I, the operator Hii in (3.93) is then
symmetric on the domain Do consisting of the span of all coherent states (3.95),
and has one or more self-adjoint extensions (since it commutes with complex
conjugation). As in 2.7.2, with slight abuse of notation the unitary one-parameter
group exp(itHIi/1i) on L2(Q) is understood to be generated by an arbitrary self-
adjoint extension of (3.93). We use the notation (2.88) and 1.(2.13).

Theorem 3.7.5. Let (Q, g) be as detailed in the preceding paragraph. Fix (p, q) E
T* Q, assuming that the cogeodesic motion (p(t), q(t» with initial conditions
(p(O), q(O» = (p, q) exists for ti < t < t f. Then with Q';i defined by 3.4.4 and
\IIhP,q) given by (3.95 ),/or all t E (ti, t f) one has

lim (\IIJ(',q), [Q';i (ct?(f) - ct~(Q';i (f»] \IIhP,q»)

Ii-->O
= O. (3.96)

The coordinates (Pi, qi) are globally defined on T* Q, and from (3.74), (3.1),
and (3.83) we obtain

Q~ := Q';i (qi) = Xi;

Pli,i := Q';i (Pi)

a
= -indetg(X)-1/4 axi detg(x)I/4, (3.97)
176 II. Quantization and the Classical Limit

defined as operators on Vo or on C~(Q) (cf. the comment after (3.92». We may

then write

(3.98)

(rather than the complete symmetrization ~A [Q:; (gij), Qj; (Pi), Q:;Cp j)], as
might have been expected on the basis of the flat-space case (2.37». The trans-
formation V : L2(Q) ---+ L2(JRn) defined by VIl1(x) = detg(x)1/41l1(x) is clearly
unitary, and satisfies

V Q~ V* = Q~.i;
V Pft,i V* = pli' (3.99)

cf. (2.23), (2.24). In particular, one infers that Q~ and Pft,i are essentially self-
adjoint on V o or on C~(Q). Moreover, the canonical commutation relations are
the same as in the flat-space case; see (2,25). From (3.98) and (3.99) we obtain

V Hft V* = ~ [gij (Q~)Pli pl j - iMjgij (Q~)Pli - ~Ii?aiajgij (Q~)]. (3.100)

The final virtue of V is that up to a phase, it maps the coherent states (3.95) into
their flat-space analogues (2.47). Hence one can transfer the entire situation to
L 2 (JRn).

Lemma 3.7.6. With 1l1};,q) the coherent state (3.95), one hasforall (p, q) E T*Q,
all (rp, g) E YQ' and all f E C~(T*Q)

lim (Il1(P,q)
ft->o ft '
[aft - (Qw(f» - QW(ao - (f»]Il1(P,Q)) = 0
(IP,g) ft Ii (IP,g) Ii '
(3.101)

Denote the Fourier transform (cf. (3.42» of a?IP,g/f) by a?IP,g)(j). From (3.87),
for X E TQ Q we obtain the expression

a~(j)(X) = dJ1-d:~~~q» j(rp:;l X); (3.102)

a~(j)(X) = e-i(Xg)(q) j(X), (3.103)

As in the proof of Theorem 3.5.1, Lemma 3.5.2 allows the assumption that
the set iT'Q->Q(supp(f) is contained in an arbitrarily small open set U C Q,
which we choose to be some geodesically convex set U. It is clear from (3.85),
(3.102), and (3.95) that both terms in (3.101) vanish in the limit Ii, ---+ 0 if rp(q) i
iT' Q-> Q ( supp (f); hence we assume the converse.
We treat rp = (rp, 0) and g = (id, g) separately; the result for the two combined
follows in an obvious way. We start with alP' We write both terms in (3.101) in
the form (3.54), change variables firstly by qj t-+ rp(qj), and secondly by (3.55),
with subsequent rescaling v t-+ li,v. We now write q' for the object q in (3.55) to
avoid confusion with the label q on ll1~p,q). In the first term we then use (3.85) and
(3.51). In the second term we have the expression
li,n Ki:' Lao(f)J(rp(y(q', v; ~Ii,), rp(y(q', v; -~Ii,)).
 3 Quantization on Riemannian Manifolds 177

By definition, rp(q') = rp(y(q', v;O» and rp*(v) = tdy(q', v; th)/dnln = O.

Hence
rp(y(q', v; ±th) = y(rp(q'), rp*(v); ±tn) + 0(n2 ),
where the order symbol is meant in the sense of smooth functions evaluated on
both sides. From (3.51) and (3.102) we then see that the expression above equals
[dJ.t(q')/dJ.t(rp(q'»]j(v, q') + O(li). In the first term in (3.101) we expand the
terms involving F (see (3.85» and j around li = O. To 0(1) the first and the
second term are then seen to cancel out. The remainder of O(n) is easily shown to
n
vanish for ~ 0, since j has compact support.
The argument for a g is analogous. This time the cancellation of the 0(1) term
is effected by (3.86), (3.103), and the fact that
[g(y(q', v; -~li» - g(y(q', v; -1n)))/li = -vg(q') + O(n),

which follows because by definition v = y(q, v; 0).

One then proceeds in precisely the same fashion as in the proof of 2.7.2, and
•
obtains (2.162). The final stage is analogous to the procedure to prove 2.7.2 for
Q/i = Q~, except that the use of 2.4.3 is replaced by Lemma 3.7.6. This is possible
because (up to a phase) the operators Ul(P(t), q(t» used in that step are of the
h
form Vp,,(rp, g)V*; cf. 3.6.2. •
CHAPTER III

Groups, Bundles,
and Groupoids

1 Lie Groups and Lie Algebras

1.1 Lie Algebra Actions and the Momentum Map
This section describes the main class of examples of Poisson manifolds that are
not symplectic. Here G is a Lie group, g its Lie algebra, and g* is the dual of g.

Definition 1.1.1. The (±) Lie-Poisson structure on g* is given by the Poisson

bracket

{f, g}±(I1) := ±11([dfe, dge]); (1.1)

here the differential dfe of f E COO(g*, JR) at 11 E g*, which is a linear map from
Tog* c:::: g* to lR, is identified with an element ofg c:::: g**, so that the right-hand side
of( 1.1) is the Lie bracket in g. The space g* equipped with the Poisson bracket (1.1)
is denoted by gl; hence COO(gl, JR) stands for the associated Poisson algebra.

For X E g let X be the linear function on g* defined by X(I1) := I1(X); clearly

X E COO(g*, JR). From (1.1) one then obtains
{X, y}± = ±[X, Y); (1.2)

cf. 11.(2.10) and surrounding text. In fact, the Poisson structure is determined by
the special case (1.2). For let {Ta} be a basis of g, with [Ta, n) = C~bTc, and
dual basis {w a } of g*, defined by w a(Tb) = 0b' We then have global coordinates l1a
on g* (so that 11 = l1awa), and ta is simply the coordinate function l1a. We know
that {f, g} depends linearly on df and dg; cf. 1.(2.4). Since df = (aflal1a )dta ,
the claim follows. Evidently, {ta , t b }± = ±C~btc, or {ea, I1b}± = ±czbec' Thus,
 1 Lie Groups and Lie Algebras 179

omitting the argument a, (1.1) may be written as

c af ag
{f, gl± = ±cabac aaa aab' (1.3)

We now look at the representation theory of Coo (g~, JR) (in the sense of 1.2.6.1).
By Corollary 1.2.6.5 a representation of Coo(g~. JR) corresponds to a symplectic
manifold S and a smooth map J : S ~ g*, such that {J* f, J* g Js = J* {f, g }_
for all f, g E Coo(g*, JR). That is, J : S ~ g~ is a Poisson map. Let

Jx:= J*X, (1.4)

which is in Coo(S, JR); in other words, lx(a) := (J(a»(X).

Proposition 1.1.2. A smooth map J : S ~ g~ is Poisson iff {lx, J y } s = -J[X,y]
E g.
for all X, Y

If B S is the Poisson tensor on S, then {J* f, J*g}s = B;(J*df. J*dg}. As in

the previous paragraph, this implies that

{J* f, J*gJs(a) = af (J(a»~(J(a»{J*Ta, J*Tb}(a). (1.5)

aaa aab
By assumption {J*Ta, J*TbJs(a) = -C~bTc(J(a», so that the right-hand side of
(1.5) is (f, g}_(J(a». •

Define ~x := ~Jx' which is the Hamiltonian vector field of lx. Assuming that
J is indeed a Poisson morphism, the Jacobi identity on the Poisson bracket of S
(or (1.2.9» implies that
(1.6)
here the left-hand side contains the commutator of vector fields, whereas on the
right-hand side [. ] stands for the Lie bracket in g.
Let us refer to a linear map X t-+ ~x of 9 into the space of vector fields l(T S)
on S satisfying (1.6) as a g-action on S. Here r(T S) may be regarded as the
Lie algebra of the diffeomorphism group of S, whose Lie bracket is minus the
commutator (cf. 1.3.3), so (1.6) corresponds to a Lie algebra homomorphism as
appropriate. When various g-actions playa role we sometimes write ~1 for h. If
{Ta} is a basis of g, we abbreviate ~a := ~Ta' One speaks of a Poisson g-action
when S is a Poisson manifold and Lh B = 0 for all X E g, where B is the Poisson
tensor. When (S, w) is symplectic, which is the only case we shall consider in the
context of g-actions, this condition is equivalent to L~xw = 0 for all X.
We infer from 1.(2.10) that a representation of COO (g "'- , JR) on S leads to a Poisson
g-action on S. Conversely, one may ask whether a given g-action on a symplectic
manifold S is related to a representation of Coo(g~, JR) on S.
Definition 1.1.3. A momentum map for a g-action X t-+ h on S is a map
J : S ~ g* for which

~Jx =h (1.7)
180 III. Groups, Bundles, and Groupoids

for all X E g; here 1x is defined by (J.4).

This definition applies to general Poisson manifolds, but we will use it only
when S is symplectic. We will occasionally write Ja for h a • A Hamiltonian g-
action on a symplectic manifold is a g-action given by a momentum map as in
0.7). It is clear from (1.7) and 1.(2.10) that a Hamiltonian g-action is Poisson.
When a momentum map J exists, (1.7) is equivalent to

(1.8)

for all X E g; to prove this, contract both sides with ~f (which is the most general
type of local vector field, since S is symplectic) and use 1.(2.19).
Conversely, when a g-action is Poisson, the properties dw = 0 and Lhw = 0
and the identity L~ = i~d + di~ imply dihw = 0, so that by Poincare's lemma a
function J x satisfying (1.8) must exist at least locally.

Proposition 1.1.4. Sufficient conditions for the existence ofa momentum map for
a Poisson g-action on a symplectic manifold (S, w) are Hd~(S, JR.) = Oor 9 = [g, g)
(equivalently, Hl(g, JR.) = 0).

Here Hd~(S, JR.) := ZlR(S, JR.)/ B1R(S, JR.) is the first de Rham cohomology group
of S; recall that ZlR(S, JR.) and Bd~(S, JR.) are the spaces of all closed and all exact 1-
forms on S, respectively. The sufficiency of the condition H~ (S, JR.) = 0 is evident
from the paragraph preceding this proposition.
The vector space Hl(g, JR) is the first cohomology group of g; since Bl(g, JR.)
is identically zero, HI (g, JR) is defined as the subspace Z 1 (g, JR) c g* of all () for
which ()([X, Y)) = 0 for all X, Y E g. The equivalence between the conditions
9 = [g, g) and Hl(g, JR) = 0 is obvious.
If 9 = [g, g), then an arbitrary X E 9 can be written as X = L j X j
with X j = [Yj , Zd for appropriate Yj , Zj E g. If X = [Y, Z], we choose
1x = w(~y, ~z), which, by an elementary calculation, using (1.6) and dw = 0,
satisfies (1.8) and hence (1.7). For arbitrary X we define 1x by linear extension of
this expression. •

The existence of a momentum map J in itself does not imply that J preserves
the Poisson bracket. To detect the extent to which it does we define a function r
ong x 9 x S by

{lx, Jy 15 = -J[x.Yl - r(X, Y). (1.9)

It is clear that r is bilinear and anti symmetric in X, Y. Taking the Poisson bracket
of both sides of (1.9) with an arbitrary f E COO(S), and using (1.7), (1.6), and the
Jacobi identity, we infer that {r(X, Y), f} = 0 for all X, Y. Since S is symplectic,
this shows that r does not depend on its argument in S. A bilinear function r :
9 ® 9 --+ JR satisfying
r(x, Y) = -r(Y, X), (1.1 0)
r(X, [Y, Z)) + r(Z, [X, Y)) + r(Y, [Z, X)) = 0 (1.11)
 1 Lie Groups and Lie Algebras 181

is called a 2-cocycle on 9 (with values in JR). The space of all 2-cocycles on 9 is

denoted by Z2(g, JR). It follows from the Jacobi identity on both {, }s and [, ] that
r as defined in (1.9) is indeed an element of Z2(g, JR).
We are now motivated to define a modified Lie-Poisson bracket on g* by
{f, g}~ := {f, g}± ± r(df, dg); (1.12)
this is indeed a Poisson bracket on account of (1.11). Generalizing (1.3), in
coordinates one has

(1.13)

( 1.14)
As for r = 0, one shows that this modified Poisson bracket is determined by the
special case

{X, f}~ = ± ([X,Yj + reX, y»). (1.15)

Definition 1.1.5. The space g* equipped with the Poisson bracket (1.13) is denoted
by g('r)±; we sometimes write Cr(g~) for the associated Poisson algebra.
Generalizing Proposition 1.1.2, one easily proves
Proposition 1.1.6. A smooth map J : S ~ g('r)- is Poisson iff (1.9) holds.
The essence of the preceding discussion may now be summarized as follows.
Theorem 1.1.7. There is a bijective correspondence between representations Jr
ofCr(g~) (in the sense ofl.2.6.1) and Hamiltonian g-actions with given complete
momentum map and associated 2-cocycle r. Given Jr : Cr(g~) ~ COO(S) one
constructs a Poisson map J : S ~ g(r)- by 1.2.6.5, and subsequently defines the
g-action X ~ h by (1.7). Conversely, given a g-action with associated complete
momentum map J (yielding a 2-cocycle r), one puts Jr = 1*.
The fact that X ~ h is indeed a g-action follows from the argument leading
to (1.6); even when r f= 0 the additional term in (1.12) is a constant function, so
that the Jacobi identity on the Poisson bracket still implies (1.6). It is Poisson by
1.(2.10), and Hamiltonian with 2-cocycle r by construction. In the converse the
fact that 1* is a representation is immediate from 1.1.6. •
A strongly Hamiltonian g-action is a Hamiltonian g-action possessing a mo-
mentum map J : S ~ g~ that is Poisson; in other words, there exists a J for
which r = 0 in (1.9). A Hamiltonian g-action with 2-cocycle r may alternatively
be described as a strongly Hamiltonian action of a certain r -dependent Lie algebra
containing g.
Definition 1.1.8. The central extension gr of a Lie algebra 9 by JR relative to
some r E Z2(g, JR) is gr := gtBJR as a vector space, equipped with the Lie bracket
[X, Y]r = [X, Y] + reX, Y)To; (1.16)
[X, To]r = 0 (1.17)
182 III. Groups, Bundles, and Groupoids

for X, Y E g, and To a basis vector of the extension R

The Jacobi identity for [, ]1' is a consequence of (1.11). We have an embedding

I : 9 "-+ gr by leX) = X +0 (which is not a Lie algebra homomorphism unless
r = 0), as well as a quotient 9 ~ gr /JR as Lie algebras.
Proposition 1.1.9. There is a bijective correspondence between Poisson maps
1 : S -+ g(r)- (or, equivalently, Hamiltonian g-actions with 2-cocycle r) and
Poisson maps 11' : S -+ gr- (or strongly Hamiltonian gr-actions) in which
(Jr(a»(To) = 1 for all a E S (equivalently, n(To) = Is, so that ~To = 0). This
correspondence preserves irreducibility.

Let w O be the basis element in gr dual to To. Then 11 : g(r)- -+ gr- given by
1 1(f) := f)+w o is a Poisson map (where 9* is embedded in gr as the annihilator
of the extension JR); this follows from (1.15), (1.16), (1.2), and ltTo = 19"
Given 1 : S -+ g(r)-' one constructs lr : S -+ gr-- by lr := 11 0 1.
Conversely, when a given lr is as stated, the equality lfcTo) = Is and Proposition
1.1.6 imply that lx(a) := (Jr),(X) (a) with (1.4) is a Poisson map 1 : S -+ 9(1')-'
Finally, Definition 1.2.6.6 and the fact that ~To = 0 lead to the last part of the
proposition. •

As a by-product of the proof we have

Proposition 1.1.10. The canonical identification of COO(gr_)/ ker(Jt) with

Cf(g"'J is a Poisson isomorphism.

This is immediate from the definitions and (1.13).

The theory so far has been concerned with a Hamiltonian g-action with given
•
momentum map 1. However, when some 1 exists, then any map l' = 1 + f)o,
where f)o E g*, is obviously a momentum map for the given g-action as well.
Having found a particular 1 for which r -I 0, when can we redefine 1 r-+ l' so
as to make l' a Poisson morphism?
A 2-cocycle r E Z2(g, JR) is said to be trivial when

reX, Y) = f)o([X, Y]) (1.18)

for some f)o E g*. The subspace of trivial 2-cocycles is called 8 2(g, JR). The
quotient H\g,JR) := Z2(g, JR)/8 2(g, JR) is the second cohomology group of g.

Proposition 1.1.11. If a momentum map of a Hamiltonian g-action defines a 2-

cocycle satisfying (1.18), then the g-action is strongly Hamiltonian. In particular,
when H\g, JR) = 0 any Hamiltonian g-action is strongly Hamiltonian.
Given (1.18), the redefined momentum map l' = 1 + f)o : S -+ g:' is a Poisson
map by (1.9). The condition H2(g, JR) = 0 implies that any 2-cocyc1e r is given
by (1.18). •

Combining 1.1.11 and 1.1.4 we obtain

 I Lie Groups and Lie Algebras 183

Corollary 1.1.12. Let Hi(g, R) = H2(g, R) = O. Any Poisson g-action is

strongly Hamiltonian, and is associated with a unique Poisson momentum map
J: S ~ g~.
If one has a Poisson momentum map J, one may still shift J t-+ J + eo,
preserving the Poisson property, iff eo annihilates [g, g]; when 9 = [g, g] this
forces eo = o. •
The conditions Hi (g, R) = 0, i = 1, 2, are satisfied, for example, when G is
semisimple. The consistency between 1.1.9 and 1.1.11 is guaranteed by
Proposition 1.1.13. If r(X, Y) = eo ([X , YD for some eo E g*, then g. is iso-
morphic to the trivial extension go = 9 EB R as a direct sum of Lie algebras. In
particular, when H2(g, R) = 0, any central extension of 9 by R is trivial.
If r(X, Y) = eo([X, YD, then X t-+ X + eo(X)To for X E 9 and To t-+ To is
the desired isomorphism between g. and 90. •

1.2 Hamiltonian Group Actions

We will now relate g-actions to G-actions, where G is a Lie group with Lie algebra
g. Recall that a (left) action L of a group G on a manifold S is a map L : G x S ~ S,
satisfying L(e, a) = a and L(x, L(y, a» = L(xy, a) for all a E S and x, y E G.
If G is a Lie group, we assume that L is smooth, unless the contrary is explicitly
stated. We write Lx(a) = xa := L(x, a).
Given a Lie group action, one defines a linear map X t-+ I;x by
d
I;xf(a):= dtf(Exp(tX)a)lt=o, (1.19)

where Exp : 9 ~ G is the usual exponential map. One sees that (1.6) holds, so
that X t-+ I;x is a Lie algebra homomorphism from 9 into the Lie algebra r(T S) of
Diff(S). We say that the ~x generate the G-action, and call ~x a generator defined
by X.
Conversely, one may ask whether a representation of COO(g~, R) on S is de-
rived from a G-action, in which case the representation is called integrable. This
question is partly answered by the following statement.
Theorem 1.2.1. Let X t-+ ~x be a homomorphism as above, and suppose the
flow of each I;x, X E g, is complete (this is the case iff there is a basis {Ta} of 9
such that the flow of each ~a is complete). Then the I;x generate an action of the
a
simply connected Lie group whose Lie algebra is g.
The construction of the a-action is, of course, done with the flow of the genera-
tors; that is, if a t-+ a(t)istheflowgeneratedbY~x,oneputsExp(tX): a t-+ a(t).
Such one-parameter groups generate a (which is connected), but it remains to
check that one indeed obtains a smooth group action. 0
Note that the statement about the basis is nontrivial, since in principle the sum of
two complete vector fields may be incomplete. Clearly, the hypothesis is automat-
184 III. Groups, Bundles, and Groupoids

ically satisfied when S is compact, or more generally when all h have compact
support. In any case, if the theorem leads to a a-action and if no h is identically
zero, there is a discrete normal subgroup Ds C a such that Ds is the maximal
subgroup of a that acts trivially on S. If G = a I D for some discrete central
subgroup DCa (recall that any Lie group with Lie algebra 9 is of this form),
then the h generate a G-action if D ~ Ds.
When the g-action associated to a G-action on a symplectic manifold S is
Hamiltonian, one speaks of a Hamiltonian group action. Similarly, a strongly
Hamiltonian group action is an action for which a momentum map J : S -+ g*
exists that is a Poisson map; cf. (1.19) and 1.1.3. It is immediate from the comment
preceding 1.2.3.5 that a Hamiltonian G-action automatically consists of Poisson
maps. Further to this, the conditions for a G-action on a symplectic manifold to be
(strongly) Hamiltonian are entirely determined by the properties of the associated
g-action, and are therefore given by Propositions 1.1.4 and 1.1.11 and Corollary
1.1.12.
The Hamiltonian version of Noether's theorem is as follows.

Proposition 1.2.2. Given a Hamiltonian G-action on a symplectic manifold S,

when h E COO(S, JR) is G-invariant (i.e., h(xa) = h(a)forall x E G), each lx is
constant along the Hamiltonian flow lines of h.

Putting x = Exp(tX) in h(xa) = h(a), evaluating dldt at t = 0, and using

(1.7) leads to {lx, h} = 0, which by 1.(2.8) and 1.(2.11) implies the claim. •

In view of this, in physics the components Jx of the momentum map usually

play the role of conserved charges.
A Hamiltonian G-action enjoys a certain equivariance property. Recall the ad-
joint action Ad of G on g, defined by Ad(x)Y := xYx- 1 (more precisely, if
Y = dy(t)ldtlt = 0, then Ad(x)Y = dxy(t)x-1It = 0). The derived rep-
resentation of 9 is then given by ad(X)Y = [X, Y] (where we simply write
ad for the awkward dAd). The coadjoint action Co of G on g* is defined by
(Co(x)8)(Y) := O(Ad(x- 1)y), with derived action of 9 ong* written as co := dCo.
One has ad(Ta)Tb = C~bTc, whence

(1.20)

As a first application of these definitions we note:

Proposition 1.2.3. The Lie-Poisson structure is invariant under the coadjoint

action (in other words, the map Co(x) is a Poisson map for each x E G).

By the comment following (1.2) it suffices to show that {X 0 Co(x), Y 0

Co(x)}± = .-!&Y}± 0 Co(x) for all X, Y E g and x E G. Since X 0
Co(x) = Ad(x-1)X etc., this is evident from the fact that the adjoint action is
an automorphism of the Lie algebra. •
 I Lie Groups and Lie Algebras 185

Given a choice J of a momentum map associated to a Hamiltonian G-action on

S via the derived g-action, we define y : G x S ~ g* by

y(x, a) = I(xa) - Co(x)J(a). (1.21)

Lemma 1.2.4. The function y is independent of a, and satisfies

y(xy) = y(x) + Co(x)y(y). (1.22)

A smooth map y : G ~ g* with property (1.22) is called a l-cocycle on G with

values in g*; the space of such 1-cocycles is denoted by Z 1 (G, Co, g*). The proof
that (1.21) is independent of a is similar to the argument after (1.9). For arbitrary
f E COO(S) we compute {IyoL x , f}sforfixedx E GandY E g,anduse(1.7)and
the invariance of the Poisson bracket under Lx. This shows that {Iy 0 Lx, f}s =
{hd(r1)Y, f} s· Since S is symplectic, yy (in obvious notation) is therefore constant
in a for all Y. The property (1.22) is then immediate from the definition and the
a-independence of y and the equality Co(xy) = Co(x)Co(y). •
Corollary 1.2.5. A momentum map I for a Hamiltonian G -action is equivariant
with respect to the modified coadjoint G-action on g* defined by
CoY(x)e := Co(xW + y(x), (1.23)

where y is given by (1.21). That is, I 0 Lx = CoY (x) 0 I for all x E G (recall that
y(x, a) = y(x»).
In particular, for a strongly Hamiltonian G-action the momentum map I is Co-
equivariant, or simplyequivariant, in that I oLx = Co(x)o I forallx. Moreover,
infinitesimal Co-equivariance (in the sense that I : S ~ g~ is a Poisson map) is
equivalent to global Co-equivariance.
Note that (1.22) guarantees that CoY is an (affine) action.
Only the final claim is not immediately obvious. It is clear from (1.9) and (1.21)
that the 2-cocycle r defined by the momentum map of the g-action corresponding
to the group action is given in terms of y by

(1.24)

If we put y = Exp(t Y) in (1.22) and differentiate with respect to t, the right-hand

side vanishes when the g-action is strongly Hamiltonian (as r = 0 in that case).
Hence the vanishing of the left-hand side says that y(x) is constant in x; since
y(e) = 0, y identically vanishes. The equivariance is then stated by (1.21). •

It should be remarked here that the anti symmetry of r as defined by (1.24) is

not automatic; it is guaranteed, however, when y is of the form (1.21).

Theorem 1.2.6. Assume that G is simply connected, and let r E Z2 (g, lR) and y E
Z 1 (G, Co, g*) be related by (1.24). There is a bijective correspondence between
representations of Cr(g~) and Hamiltonian G-actions with 1-cocycle y whose
momentum map I is complete:
186 III. Groups, Bundles, and Groupoids

• Given a representation 7r = 1* : Cr(g~J -+ COO(S), define the vector fields

h := ~Jx (with (1 A)) on S; then the corresponding G-action exists andsatis.fies
(1.21).
• Given a Hamiltonian G-action on S with momentum map J and 1-cocycle y
(defined by (1.21)), the corresponding representation 7r is given by 7r = 1*.
This follows from Theorems 1.1.7 and 1.2.1, and the fact that y is uniquely
determined by [' and (1.24). To see this, define r : 9 -+ g* by (r(X»(Y) :=
[,(Y, X). This leads to an affine map cor : g* -+ g* ,given by cor (X)8 = co(X)O+
r(X). It follows from (1.11) that cor (X) is a Lie algebra homomorphism; note
that the Lie bracket of two affine maps Ai = L j + Vi, where L is linear, is defined
by [AI, A2]O := (AIA2 -A2AI)8+A1V2 -A2VI. As in the linear case, when Gis
simply connected there is a unique affine action CoY (G) on g* whose derivative is
cor (g); it is given by (1.23), where y satisfies (1.22), which is equivalent to (1.11).
This y then has to coincide with the same symbol defined by (1.21), stripped of
its vacuous a -dependence. •
When G = GJD (in the notation used after 1.2.1) is not simply connected, one
has to assume integrability of the G-action. In tum, this guarantees integrability of
[' to y; given its existence, y is uniquely determined by the property 9 = yo L{;-->G.
Recall the definition of the space of l-cocycles ZI(G, Co, g*) below (1.22);
define B I(G, Co, g*) C Z I(G, Co, g*) as the subspace of maps of the form
y(x) = Co(x)Oo - 00 (1.25)
for some 00 E g*. The 2-cocycle [' derived from y E B 1(G, Co, g*) by (1.24)
lies in B2(g, lR); it is remarkable that such a [' is automatically antisymmetric.
In general, elements of Z2(g,lR) derived from y E ZI(G, Co, g*) by (1.24) may
fail to be antisymmetric. Elements of ZI(G, Co, g*) that do give rise to an an-
tisymmetric [' are called symplectic cocycles, forming the space Z1(G, Co, g*).
The first cohomology group of G relative to the coadjoint representation is
Hsl(G, Co, g*) := Z1(G, Co, g*)J Bl(G, Co, g*).
Proposition 1.2.7. When G is simpLy connected, H 2(g, lR) and H}(G, Co, g*)
are isomorphic.
The proof of Theorem 1.2.6 shows that any [' E Z2(g, lR) corresponds to a
unique y E ZI(G, Co, g*). The claim then easily follows from the paragraph
preceding the proposition. •
More generally, further to 1.1.11 we have
Proposition 1.2.8. When a Hamiltonian G-action with momentum map J satisfies
(J.21) with (1.25), the action is strongly Hamiltonian. When H/(G, Co, g*) = 0,
any Hamiltonian action ofG is strongly Hamiltonian.
When y is of the form (1.25), the redefined momentum map J' = J + 00 is
equivariant, as is clear from (1.21). If Hsl(G, Co, g*) = 0, then any y is of this
bm •
 1 Lie Groups and Lie Algebras 187

Note that the affine map A : () 1-* () + ()o satisfies ACoY (x)A -I = Co(x) for all
x, as well as {A* f, A*g}~ = A*{f, g}±.

Corollary 1.2.9. A Hamiltonian action of a compact Lie group is always strongly

Hamiltonian.
The cohomology group HI(G, Co, g*) := ZI(G, Co, g*)/ BI(G, Co, g*) is zero
when G is compact. Hence HsI(G, Co, g*) c HI(G, Co, g*) must be trivial as
~. D
The group analogue of 1.1.12 is
Corollary 1.2.10. Let H1(g, R) = H 2(g, R) = 0, and let G act on a symplectic
manifold by Poisson maps. There exists a unique equivariant momentum map J
associated with this action.
Example 1.2.11. Let S = T*Rn with its canonical symplectic structure.
1. TheactionofG = Rn (whoseLiealgebrag = Rn has a basis (1ih=I .... ,n) on S,
given by a : (p, q) 1-* (p, q + a), is Hamiltonian, with equivariant momentum
map Ji(P, q) = Pi.
2. The action ofG = SO(n) (whose Lie algebra 9 = (\2(Rn) has a natural basis
{T;j}i<j=I, .... n) on S, given by R : (p, q) 1-* (Rp, Rq), is Hamiltonian, with
equivariant momentum map Jij(p, q) = Piqj - Pjqj, qi := qi.
3. Let the abelian Lie algebra 9 = R2n have a basis {Pi, Qj}i.j=I ..... n. The corre-
sponding Lie group G = R2n is parametrized by (u, v) := Exp(-uQ + vP);
cf II. (2.5). It acts on S by
(u, v) : (p, q) 1-* (p + u, q + v);
see 1l.(2.J3), in which we have put c = 1. This action is Hamiltonian, with
momentum map Jp;(p, q) = Pi and JQ;(p, q) = qi. The 2-cocycle r is
r(Pi, Pj) = r(Qi, Qj) = 0;
(1.26)
The central extension gr is the Heisenberg Lie algebra l)n; see II.(204).
4. Finally, the map J of Proposition II.3 .104 is an equivariant momentum map for
the group action II.(3.J3).
The first two examples are a special case of Lemma 2.3.1 below. As to the third,
it should be remarked that since G is abelian, it must be that B 2(R2n, R) = 0; cf.
(1.18). Also, (1.11) is identically satisfied, so that a 2-cocycle on R 2n is simply an
antisymmetric bilinear map on R2n. The dimension of the space of antisymmetric
m x m matrices is ~m(m - 1); hence H 2 (R 2n, R) = Rn(2n-1).

1.3 Multipliers and Central Extensions

Definition 1.1.8 and the ensuing discussion have an analogue at the group level.
The best way to approach this matter is via the following concept.
188 III. Groups, Bundles, and Groupoids

Definition 1.3.1. A central extension of a Lie group G by U (1) is a short exact

sequence
rp T
e ~ U(1) ~ G ~ G ~ e, (1.27)
in which G is a Lie group, ({J and r are smooth homomorphisms, and ({J(U(1)) is
contained in the center ofG (by definition of an exact sequence, the image of each
map is the kernel of the next).
It is most natural to analyze this structure in terms of principal fiber bundles;
the reader unfamiliar with this notion may either skip the following geometric
discussion and resume at Theorem 1.3.3, or jump ahead and read 2.1 before
proceeding.
It is quite easy to see that the group G in (1.27) is a principal U (1 )-bundle
G(G, U(1), r) over the base G; in particular, the action of U(1) on G is given by
Rz(X) := X({J(z). (Identifying U(1) with 1', we write its elements as z.) We now
choose a section s : G ~ G of this bundle (that is, r(s(x)) = x for all x E G).
Since r is a homomorphism, it must be that r(s(x)s(y)) = r(s(xy)). Hence there
exists a function c : G x G ~ U (1) such that
s(x)s(y) = c(x, y)s(xy) (1.28)
for all x, y E G. Since s«xy)z) = s(x(yx)) by associativity of the multiplication
in G, one must have the identity
c(x, y)c(xy, z) = c(x, yz)c(y, z) (1.29)
for all x, y, Z E G. We may restrict ourselves to sections satisfying s(e) = e
(where on the left-hand side e E G and on the right-hand side e E G). Then
c(e, x) = c(x, e) = 1 (1.30)
for all x. Moreover, while s may not be globally smooth, it may always be chosen
so as to be (Borel) measurable, and smooth in a neighborhood of e E G. In that
case c is smooth near (e, e). This motivates the following
Definition 1.3.2. A multiplier on a Lie group G is a measurable function c :
G x G ~ U(l) that is smooth near (e, e) and satisfies (1.29) and (1.30).
The set ofall multipliers on G is called Z2(G, U(l)); this isa vector space when
the group operation in the abelian group U(1) is written additively.
As explained in 2.1, a section s leads to a trivialization 1/1s : G ~ G x U (1) of
G, which is generally discontinuous with respect to the product manifold structure.
According to (2.3) one has 1/Is-l(X, z) = s(x)z. Transferring the group operations
from G to G x U(l), using (1.28) one obtains
(x, z)· (y, w) = (xy, zwc(x, y));
(x, Z)-l = (X-I, zc(x, X-l )). (1.31)
The c-extension G c of G is G x U (1) with the above group operations, and with
the manifold structure inherited from G via 1/Is. Thus G e, which is a Lie group, is
 1 Lie Groups and Lie Algebras 189

simply a trivialized version of G, depending on c via the choice of the section s.

Conversely, one may start from a multiplier.
Theorem 1.3.3. Let c be a multiplier on G. Equip the set G c := G x U(1) with
the group law (1.31). There exists a manifold structure on G c that turns it into a
Lie group with the given multiplication. If c is smooth on G x G, the manifold
structure is that of the direct product.
Associativity of the group multiplication in G c is a consequence of (1.29). The
remainder of the proof is a technical exercise in the definition of a Lie group;
the idea is that the product manifold structure may be used in a neighborhood of
(e, I); the group operations are then smooth because c is smooth near (e, e). This
local manifold structure is subsequently transferred to all of G c using the group
~ D
The embedding t : G ~ G c by t(x) = (x, I) is not a homomorphism unless
c = 1. When necessary for unambiguity we will denote the subgroup U(I) C G c
defining the central extension by U c (1).
If one starts from the diagram (1.27) and then passes to G c via a section s,
one may examine the effect of a change in son c and hence on G c • Given some
measurable function b : G -+ U (1) that is smooth near e, one may pass from s to
s', defined by s'(x) := s(x)Rb(x) = s(x)cp(b(x». This leads to the replacement of
c by
, b(x)b(y)
c (x, y) = c(x, y). (1.32)
b(xy)
Two multipliers c, c' related by (1.32) for some b are called equivalent. Thus an
appropriate cohomology theory is defined through the subspace B2(G, U(I» of
Z2(G, U(I», consisting of multipliers of the form
b(xy)
c(x, y) = b(x)b(y) (1.33)

for some measurable function b : G -+ U(l) that is smooth near e. Hence one
forms the cohomology group H2(G, U(I» := Z2(G, U(1»/ B2(G, U(l». Equiv-
alent multipliers then define the same element of H2(G, U(I». In particular, the
multiplier c in (1.33) is equivalent to 1.
The connection between equivalent multipliers and isomorphic group extensions
is now as follows.
Proposition 1.3.4.
1. Two multipliers c and c' are related by (1.32) iff the extensions G c and G c' are
isomorphic as Lie groups.
2. In particular, when c is of the form (1.33), the corresponding c-extension is
isomorphic to the direct product of G and U (1).
3. Thus when H2(G, U(l» = 0, any c-extension ofG is trivial.
4. When G is simply connected, every multiplier is equivalent to one that is smooth
on G x G, so that as a manifold G c is a trivial U (I )-bundle over G (cf 2.1 ).
190 III. Groups, Bundles, and Groupoids

The map (x, z) 1-+ (x, Zb(x)-I) provides the desired isomorphism from G e
to G e,. Conversely, when G e and G e, are isomorphic to G, they must be related
by an isomorphism cp : G e ~ G e' of the form cp(x, z) = (x, ¢(x )z), where
¢ : G ~ Ue (1). Since cp is in particular a group homomorphism, the choice
b(x) := ip(x )-1 satisfies (1.32). The second point follows from the first by choosing
c' = 1.
When H2(G, U(1» = 0, any c is given by (1.33), which implies 1.3.4.3.
The last statement is a consequence of Theorem 1.3.3 and the fact that a
U(1) bundle over a contractible space is necessarily (isomorphic to) a trivial
bundle. •
Under certain conditions there is a correspondence between extensions of Lie
algebras and of Lie groups. We identify JR. in 1.1.8 with the Lie algebra ue (1) of
Ue (1), and write Exp : u(l) ~ U(1) for the exponential map, conventionally
realized as Exp(X) = exp(-iX). In a neighborhood Ne x Ne of (e, e) we can
write c = Exp(X), where X : Ne x Ne ~ uc(1) (for simply connected G this can
be done on all of G x G). Then define r : 9 ® 9 ~ JR. by
d d
r(X, Y):= - -
ds dt
[x (Exp(tX), Exp(sY» - X (Exp(sY), Exp(tX»]1 -1-0'
s--
(1.34)
For example, in the setting of Example 1.2.11.3 the multiplier c : JR.2n X JR.2n ~
U (1) is given by
c«u, v), (u ' , Vi» = e i (uv'-vu')/2. ( 1.35)
Through (1.34) (with c = exp(-ix» and 11.(2.5) this indeed reproduces the 2-
cocycle r in (1.26). The Heisenberg group Hn is nothing but the central extension
JR.~n defined by r; cf. 11.(2.8) and 1.3.6 below. The multiplier c' «u, v), (u', v'» :=
exp(iuv') leads to the same r; it is related to c by (1.32), with b(u, v) =
exp(-~iuv).

Lemma 1.3.5. The map r defined by (1.34) is an element of Z2(g, JR.). If c E

B2(G, U(1», then r E B2(g, JR.).

Wewriter(X, Y) = X*(X, Y)-X*(Y, X) for x* : gxg ~ JR.. With this notation,

(1.29) combined with its cyclic permutations in x, y, z implies X*(X, [V, Z]) +
cycl. = X*([Y, Z), XHcycl., which leads to(1.11). The second claim is immediate
from (1.33) and (1.18). •
We now discuss the inverse process of passing from r to c.
Proposition 1.3.6.
1. When G is simply connected there exists a multiplier c E Z2(G, U(J» that is
related to a given 2-cocycle r E Z2(g, JR.) by (1.34).
2. When G = G/ D (where G is simply connected and D = 1T1 (G) is a central
subgroup of G), such aCE Z2(G, U(1» exists iff D ~ Z(G c )/Uc (1) (here
Ge is the central extension given by the previous item and 1.3.3, and we have
identified G with Gc/Uc (1»).
 1 Lie Groups and Lie Algebras 191

3. When c exists, its equivalence class in H2(G, U(1» or H2(G, U(l» is uniquely
determined by r.

By Lie's third theorem there is a simply connected Lie group Gr with Lie
algebra gr (unique up to isomorphism); as a manifold Gr = G x R. Define
c: G x G -+ JR. by (x, 0)· (y, 0) = (xy, c(x, y», where· is the multiplication in
Gr that comes with its construction. The associativity of . implies that c satisfies
(1.29) (if the group law in JR. is written multiplicatively). If r : JR. -+ U (1) is the
covering projection, we put c := roc and verify that c satisfies (1.29), since r is
a homomorphism. This is the desired multiplier c, and Ge , defined as in 1.3.3, is
a quotient of Gr by the central subgroup IE c JR.. In particular, the Lie algebra of
Ge is gr. This proves the first claim.
As to the not simply connected case, the necessity of the stated condition is
obvious, for D must lie in the center Z(G) of G. To prove sufficiency, consider
Dr C G r ; as a set Dr := D x JR., which is a subgroup of Gr. The assumption
implies that Dr is abelian, so that there must be an isomorphism </J : Dr -+ D x JR.,
where this time the symbol x stands for the direct product of groups. Hence
Dr := </J-l(D x IE) is a discrete central subgroup of Gr. Then one easily infers
that the Lie group G e := Gr / Dr is a central extension of G. Its multiplier cis
defined by the property (x, 0)· (y, 0) = (xy, c(x, y», proving its existence.
Finally, uniqueness in cohomology follows from Lie's third theorem In
combination with 1.3.4.1. •

Given G and r, this proposition gives conditions for the existence of a central
extension G e with Lie algebra gr.

Corollary 1.3.7. When G is simply connected one has

( 1.36)

The first isomorphism is clear from 1.3.5 and 1.3.6; the second one follows from
Proposition 1.2.7. •

We return to symplectic geometry. The group analogue of 1.1.9 is

Corollary 1.3.8. Let a Hamiltonian G -action on a connected symplectic manifold

with CoY -equivariant momentum map be given, with r defined by (1.24). Assume
that G and r are such that a central extension G e (defined through 1.3.6 and 1.3.3)
with Lie algebra gr exists. Then the G e-action obtainedfrom the G-action through
projection on G = Gel Ue(1) is strongly Hamiltonian.

This is immediate from (1.12), 1.1.8, and 1.1.6.

•
According to Proposition 1.2.8, the special case y E B 1(G, Co, g*) implies that
the G-action has an equivariant momentum map; in other words, it is strongly
Hamiltonian. This is consistent with 1.3.8, for Y E Bl(G, Co, g*) implies r E
B2(g, JR.) by the proof of 1.2.7; this, in tum, leads to c E B2(G, U(l» by 1.3.6.3,
which means that G c :::::: G x U(l) as a Lie group by 1.3.4.
192 III. Groups, Bundles, and Groupoids

1.4 The (Twisted) Lie-Poisson Structure

We now turn to an analysis of the Poisson algebra Cf(g~), starting with an in-
teresting realization of it. This involves the geometry of T*G, which we briefly
review first. We denote the right- and left-invariant vector fields on G by ~: and
. I .
"I:L
x ' respective y; I.e.,
L d
h f(y):= dt f(yExp(tX»lt=O. (1.37)

R d
~x f(y):= dt f(Exp(tX)y)lt=o. (1.38)

For the commutator one has

[ I:L,R I:L,R] _ ±I:L,R . (1.39)
5X '''Y - "[X,y]'

here and in what follows the upper sign enters for "L", and the lower one for "R".
We write ~L,R := ~!:.R. One sees that
a '.

(1.40)

The left or right Maurer-Cartan form 0t!~ is an element of A l(G) 0 9 (i.e.,

a g-valued I-form on G), defined by .
OMC(I:L,R) ' -
L,R "x ,- X. (1.41)
The connection between the two follows from (1.40) as
Ad(xWt!c(x) = rli/c(x). (1.42)

In terms of a basis {Ta} of 9 we expand Ot!~ (x) = Of R(X )Ta, defining a collection
of left- or right-invariant I-forms Of R(X): Define th~ G-actions Land R on G by
Lx(Y) := xy; (1.43)
Rx(Y) := yx- I • (1.44)
One may then equivalently define Bf(x) := L:_1wa and B~(x) := R;_IWa, where
the w a form a basis of g* = Te* G dual to the basis {Ta }. The Maurer-Cartan
equations

dBf,R(x) = =FtcgcBf.R(X)!\ °f.R(x) (1.45)

are an immediate consequence of (1.39).

One defines two (globally valid) trivializations
TG ~ 9 x G; (1.46)
in the left trivialization one maps Y E Tx G to (L r I )* Y E 9 x G, whereas in the
right trivialization one maps the same Y to (Rx)S. Conversely, (Y, X)L stands
for (Lx)*Y E TxG, and (Y,X)R corresponds to (Rrl)*Y E TxG. For example,
the left trivialization of ~f(x) is (Y, X)L, and the left trivialization of ~:(x) is
(Ad([ I ) Y, x) L; cf. (1.40).
 1 Lie Groups and Lie Algebras 193

Similar to (1.46), one has

T*G ~ g* x G; (1.47)
in the left trivialization a E Tx*G is mapped to (L~a, x) in g* x G, and in the right
trivialization a is mapped to (R;_,a, x). We write (0, X)L for L;_,O E Tx*G, and
(0, X)R for R;O E Tx*G. The connection between these trivializations is
(0, X)L = (Co(x)(J, X)R. (1.48)
Given a cocycle r E Z2(g, lR) (regarded as an element of A;(G», one can define
a 2-form r L on G by rdx):= L;_,r; similarly, rR(x):= R;r. In other words,
rL,R(~~,R, ~~,R) = reX, Y). Hence rL.R(X) = rab0f,R(X) A 0f,R(X); cf. (1.14).
Thus we obtain a 2-form r~.R := r*r L.Ron T*G, where r := rpG4G (this
notation will be used throughout this section).
We define G-actions Ax := L;_, and Px := R;_, on T*G; in the trivializations
defined above their expressions are
Px(O, yh := (CoY (x)(J, yx-I)L; (1.49)
Ax(O, yh := (0, XY)L; (1.50)
Px(O, y)R := (0, YX-1)R; (1.51)
Ax(O, y)R := (CoY (x)(J, XY)R; (1.52)
here we assume that r is related to y by (1.24). To derive these expressions, one
uses relations of the type (cf. (1.42»
R;O:!C = Ad(x )O:!c. (1.53)
Recall the coordinates Oa on g* introduced after (1.2).
Proposition 1.4.1. Letw be the canonical symplecticform on T*G (cll.2.3.8),
and equip T*G with the 2-form W~·R := w + rl.R"

• The form W~·R is symplectic.

• In the above trivializations the corresponding Poisson bracket on T* G is given
by

(1.54)

• The actions p and A commute and are Hamiltonian, with CoY -equivariant
momentum mappings j Rand j L, respectively, given by
jR(O, X)L = -0; (1.55)
jL(O, X)L = CoY(x)(J; (1.56)
jR(O, X)R = -COY(x-I)(J; (1.57)
jL(O, X)R = O. ( 1.58)
The 2-form rtR is closed as a consequence of (1.11). For an arbitrary manifold
Q, it is easily verified that the form w + r*a on T* Q is symplectic for any closed
194 III. Groups, Bundles, and Groupoids

2-form a on Q. To derive (1.54), one first observes that 8 in 1.(2.22) is given by

8(0, x) = Oa 1\ '*0f,R(X), and then uses the Maurer-Cartan equation to show that
(1.59)
Equation (1.54) then follows from 1.(2.19). Given this Poisson bracket, one verifies
that the momentum maps (1.58) generate the actions (1.52); this, in turn, guarantees
that the group actions in question are Hamiltonian. The equivariance of J Rand
J L is trivially verified. •
Let Cf(T*G)R stand for the set of p-invariant functions in COO(T*G), with
Poisson bracket (1.54). This is a Poisson sub algebra, since each Px is a Poisson
map.
Corollary 1.4.2. The map (JL)* : Cf(g~) --+ Cf(T*G)R is an isomorphism of
Poisson algebras.
We see from (1.52) that in the right trivialization, Cf(T*G)R consists of those
f (p, x) R that are independent of x. One then immediately infers from (1.54), where
the lower sign applies, that such functions satisfy the Poisson bracket (1.12); cf.
(1.3). Then use (1.58). •
The obvious generalization of Proposition 1.2.3 is
Proposition 1.4.3. The Poisson structure (1.12) is invariant under the G-action
(1.23); in other words, the map CoY (x) is a Poisson map for each x E G.
Proceeding as in the proof of 1.2.3, the claim follows if
y(x)([X, Y]) = r(Ad(x-1)X, Ad(x-1)y) - r(X, Y). (1.60)
To prove this, we write the left-hand side as d[y(x)(Ad(Exp(t X»y)]/dt at t = O.
The expression in square brackets equals [Co(Exp( -tX»y(x)](Y). Using (1.22),
this equals [y(Exp(-tX)x) - y«Exp(-tX»](Y). Writing xx-1Exp(-IX)X for
Exp( -I X)x and using (1.22) once again, as well as (1.24), we eventually obtain
(1.60). •
A coadjoint orbit 0 in g* is an orbit under the coadjoint action. Similarly, a
CoY -orbit in g* is defined with respect to the action (1.23). The CoY -orbit 0;
through 0 E g* is of the form 0; = G / G~, where G~ is the stability group of e
under the CoY -action; we see from (1.23) and (1.24) that its Lie algebra is
g~ = {X E gIO([X, Y]) - r(X, Y) =OVY E g}. (1.61)
Theorem 1.4.4. Let y be a symplectic cocycle, and define r by (1.24). The sym-
plectic leaves ofg* with respect to the Poisson structure (1.12) are the CoY -orbits
of G. In particular, the symplectic leaves of the Lie-Poisson structure coincide
with the coadjoint orbits.
We know from 1.2.4.7 and 1.2.3.7 that the tangent space at some point e of a
given leaf L(J is spanned by the Hamiltonian vector fields ~f(e). These depend
linearly on df = (aflafJa}dfJa = (aj/aea)dTa. We now use an ancillary result.
 1 Lie Groups and Lie Algebras 195

Lemma 1.4.5. The Hamiltonianjlow on g± (computed with respect to the Poisson

structure (1.12)) generated by X is O(t) = CoY (Exp(=t=tX»)O. In other words,
~: = =t=~x, where h is defined by (1.19) and the CoY -action (1.23) on g*, and~:
is the Hamiltonian vector field defined by the Poisson structure (1.12).

For clarity we start with the proof for y = f' = 0, choosing the minus sign for
concreteness. Since the linear functions separate points in g* , it suffices to compute
d - - -
dt Y(Co(Exp(tX»)8) = -O(Ad(Exp(-tX»[X, Y)) = {X, y}-(O(t»,

from which the claim follows. For general y one in addition uses the identity
d
dt y(Exp(tX»(y) = -f'(X, Y) - y(Exp(tX»([X, Y)), (1.62)

which follows by putting x = Exp(tX) and y = Exp(sX) in (1.22), dividing by s

and letting s -+ 0, and subsequently applying (1.60).
The calculation for the plus sign is analogous. •

By this lemma, the Hamiltonian vector fields ~fa span the tangent space at 0
to the CoY -orbit through 0; this implies the claim locally. Globally, since G is
connected, it is generated by the image of Exp(g) in G. This ends the proof of
Theorem 1.4.4. •

It follows that any CoY -orbit (and in particular any coadjoint orbit) OY is an
even-dimensional symplectic manifold; for y = 0 the immersion of 0 := 0° in
g± defines the (±) Lie symplectic form w~ on O. Equipped with this form, we
denote 0 by O±. (As in the general case of symplectic leaves, 0 is not necessarily
a submanifold of g*; cf. the text following 1.2.4.4.) For general y we see from
1.(2.19) and (1.2) that w~Y is given by
oy
(w±
± ±
Mh, ~y ) = =t=(O([X, Y) + f'(X, Y»; (1.63)

since this action is trivially transitive, (1.63) suffices to define w~Y. It is clear from
(1.63) or 1.2.3 that w~Y is invariant under the CoY -action.
Lemma 1.4.5 has the following

Corollary 1.4.6. The momentum map for the coadjoint action ofG on a coadjoint
orbit O± is given by h(O) = =t=0.
Recall Definition 1.1.5. Theorem 1.4.4 leads to

Corollary 1.4.7. Let 7f : Cf(g~) -+ COO(S) be an irreducible representation

(in the sense ofl.2.6.6). Then S must be (symplectomorphic to) a CoY -orbit in g*
(equipped with the symplectic structure (1.63)), or a covering space thereof

This is immediate from Theorems 1.2.6.7 and 1.4.4. The symplectomorphism

in question is given by the momentum map. •

Note that Proposition 11.2.1.2 is a special case of 1.4.7.

196 III. Groups, Bundles, and Groupoids

Corollary 1.4.8. Let S be a connected symplectic space with a transitive Hamil-

tonian action of a Lie group G. There is ayE ZI(G, Co, g*) such that S is
(symplectomorphic to) a CoY -orbit in g* (equipped with the symplectic structure
( 1.63)), or a covering space thereof
If, in addition, HI (g, 1R) = H 2 (g, 1R) = 0, then S is (symplectomorphic to) a
coadjoint orbit in g*, or a covering space thereof

If the G-action is Hamiltonian, there is a momentum map 1. The transitivity of

the G-action on S implies that 1* is irreducible in the sense of Definition 1.2.6.6.
Then apply 1.4.7. The second claim is then immediate from Corollary 1.2.10. •

Finally, the central extension G c introduced in 1.3.3 may be used to shed light
on the Poisson structure (1.54) on T*G, which in this context we write as (T*G)c'

Proposition 1.4.9. Let Uc(l) act on T*G c (equipped with the canonical cotan-
gent bundle Poisson structure) by lifting the action h : x ~ xh- I on G e . An
equivariant momentum map 1c : T*G c ~ ueO) = IRfor this action is given in
the right trivialization by 1c(80, 8 1, ... , 8n ) = -80 . Then (T* G)e ::::: 1c- 1(l)/ Uc(l)
as Poisson manifolds.

Realizing that the additional structure constants of G c (compared with G) are

given by C~b = rab and Cb j = 0 for all i, j = 0, ... , n, this follows from (1.54),
first applied to G c and then to G. •

This statement will be properly understood in the setting of 2.3 and IV.I.5.

1.5 Projective Representations

We specialize the discussion to the setting relevant to quantum mechanics: The
symplectic manifold S is a projective Hilbert space Ini (see I.2.5), and the G-
action Lx : l{! ~ xl{! on Ini should preserve the transition probabilities I.(2.65)
for all x E G. In all cases of interest it turns out that requiring smoothness of a Lie
group action on Ini would force 7t to be finite-dimensional, so we here assume
the action to be merely continuous.
This continuity may be restated as follows. We equip the group U(7t) of all
unitary operators on 7t with the strong operator topology (or the weak one, which
coincides with the strong topology on the unitaries). Denote the central subgroup
of all mUltiples of II by 1l'lI, and form the quotient U(7t)/'ll'lI, endowed with the
quotient topology. Continuity of the G-action is then equivalent to continuity of
L, seen as a homomorphism from G into U(7t)/1l'lI.
By I.3.4.3, for each x E G there exists a unitary or an antiunitary operator U (x)
on 7t such that TO U(x) = Lx 0 T (where T := T§1t~lP1l)' For G connected, all
U(x) must be unitary, since in a connected Lie group each x is a square, and a
square of either a unitary or an antiunitary operator is unitary. Two different U's
projecting to the same map on 1P'7t must differ by a phase. Now consider

G := {(x, U) E G x U(7t) I Lx = T(U)}, ( 1.64)

 I Lie Groups and Lie Algebras 197

where -r(V) is the map ofJnl defined by V through the canonical projection of§1t
to Jnl. Inheriting the group operations and topology from G x U(1t), one verifies
that G is a Lie group, which by the previous paragraph is a central extension of G
by V(1); cf. (1.27). Following the discussion after Definition 1.3.2, we choose a
measurable section s : G ~ G, smooth near e, associated with a multiplier c on
G. The trivialization G ~ G x V(I) (as a set) defined by s leads to the choice of
a representative V(x) for each map Lx. This choice obviously satisfies
V(x)V(y) = c(x, y)V(xy). (1.65)
Hence V is a projective representation of G on 1t with multiplier c; we sometimes
say that V is a c-representation.
For example, the abelian group JR2n has a projective representation on L2(JRn)
given by

vf(u, v)\}I(x) := eiu (x-1 v)\}I(x - v); (1.66)

cf. 1I.(2.17). The multiplier is given by (1.35), which should not be surprising in
view of the definition of the Heisenberg groups fIn and Hn = JR~n .
A redefinition V'(x) = b(x)V(x), where b : G ~ V(I), leads to the modifi-
cation (1.32); we say that V and V' are equivalent. Clearly, V is equivalent to a
representation iff cis ofthe form (1.33).
Proposition 1.5.1. There is a bijective correspondence between c-representations
V ofG and representations Vc ofG c in which V c(1) is represented by the defin-
ing representation (times the identity operator). This correspondence preserves
irreducibility.
Given a c-representation V(G), define Vc(G c) by Vc(x, z) := zV(x). Con-
versely, if a representation Vc(G c) satisfies Vc(e, z) = zH, then V(x) := Vc(x, 1)
satisfies (1.65). The last claim is obvious from Schur's lemma and the fact that
V c (1) is a central subgroup of G c • •

f
Indeed, the projective representation V (JR2n ) defined in (1.66) is the restriction
of Vf(Hn) (see 1I.(2.17» to JR2n, identified with (JR2n, 1) C Hn.
The classical analogue of 1.5.1 is Proposition 1.1.9.
Proposition 1.5.2. If H2( G, V (1» = 0, then any projective representation of G
is equivalent to a representation.
As already pointed out, we see from (1.32) that V is equivalent to a representation
iff c is of the form (1.33). •
We now look at the corresponding concepts for Lie algebras. A projective
representation of a Lie algebra g on a complex vector space V is a linear map
R : g ~ L(V) (the space of linear maps on V) such that
[R(X), R(Y)] = R([X, YD - if(X, Y)H (1.67)
for some 2-cocycle f E Z2(g, JR); the Jacobi identity on the commutator on the
left-hand side enforces (1.11). One may speak of a f-representation of g. If R
198 III. Groups, Bundles, and Groupoids

is modified to R'(X) := R(X) + iOo(X)1I for some (fixed) 00 E g, then (1.67) is

satisfied for R', with

r'(x, y) = r(X, Y) + Oo([X, Y]); (1.68)

cf. (1.18). Such representations R and R' are called equivalent.
Proposition 1.5.3. There is a bijective correspondence between r -represen-
tations of 9 (on complex vector spaces) and representations of g(' in which the
generator To is represented by -ill; cf Definition 1.1.8.
A r-representation R(g) defines a representation Rr(gr) by Rr(X) := R(X)
and Rr(To) := -ill, and vice versa. •

Proposition 1.5.4. If r(X, Y) = Oo([X, Y]) for some 00 E g*, then a r-

representation is equivalent to a representation. Hence when H 2 (g, JR.) = 0 any
r -representation is equivalent to a representation.
Compare with 1.1.13. This is obvious from (1.68): A projective representation
whose r is of the above form is equivalent to a representation by the shift R' (X) : =
R(X) - iOo(X)II. •
In relating (projective) representations of a Lie group G to (projective) represen-
tations of its Lie algebra g, we need to discuss a technical point. This discussion is
necessary, because when 1i is infinite-dimensional, the G-action on lP'1i provided
by a representation is not necessarily smooth or even C I. In other words, for a given
\II E 1i the map from G to 1i defined by x ~ U(x)\II may not be differentiable,
so that the curve U (Exp(t X»\II is not necessarily C I; this would make it difficult
to define the generating vector field ~x at \II.
To simplify the discussion somewhat we assume that U, when it is projective,
defines a multiplier c that is smooth on G x G (the case where c = 1 is therefore
included). As we have seen in 1.3.4, when G is simply connected this can always
be achieved.
Definition 1.5.5. A smooth vector for a (projective) representation U is an
element \II E 1i for which the map x ~ U (x) \II is smooth.
It can be shown that the set 1i'tj of smooth vectors for U is a dense linear
subspace of 1i.

Proposition 1.5.6. Under the above conditions on U and 1i, the subspace 1i'tj
is stable under U(G). For each X E 9 the operator dU(X), defined by
d
dU(XW := dt U(ExP(tX»\IIlt=O (1.69)

is essentially self-adjoint on 'H'tj. Finally, 1i'tj is stable under d U (g).

The stability of 1i'tj follows from (1.65) and the smoothness of c and of group
multiplication. The second claim is then shown as in the proof of II.3 .6.4. The last
point is evident. •
 I Lie Groups and Lie Algebras 199

It follows from (1.65) and (1.34) that on 1irJ one has

[dU(X), dU(Y)] = dU([X, YD - ir(X, Y)][, (1.70)
where r is defined by (1.34) with c = exp(-ix); cf. (1.67).
Conversely, given a (projective) representation R of 9 by symmetric operators
on some common domain D, one may ask whether there is a (projective) repre-
sentation U (G) such that R = d U. As in the classical case, such a representation
of 9 is then called integrable.
When 1i is finite-dimensional and G is simply connected, every representation of
9 by skew-Hermitian matrices is integrable; this is already a difference with the case
of group actions on general finite-dimensional manifolds, caused by the fact that
the flow of a skew-Hermitian matrix is always complete. In the infinite-dimensional
case further conditions are required.

1.6 The Twisted Enveloping Algebra

Recall the definition I.(2.43) of the function A E C oo (lP1i), where A E ~(1i)IR'
Since dU(X) tends to be unbounded, the functions idiJ{X) are defined only on
IP1trJ; cf. the preceding section. One can topologize 1irJ so that IP1trJ is a Frt5chet
submanifold of 1P1t, and the G-action restricted to IP1trJ is smooth, with smooth
momentum mapping. Since this is technically involved, we will merely state a key
result.

--
Theorem 1.6.1. Let 1i and U be as stated above 1.5.5. Then
1x := iIidU(X), (1.71)
defined on IP1trJ. is a momentum map for the G-action on lP1t derived from the
representation U on 1i. It satisfies
{Jx, Jrh = -J[x.YJ - lir(X, y)IJ1>'H. (1.72)
where r is defined by (1.34) or (1.70).
With 1/1 = 'l"S1-l-->JI>'H(\II) one obtains h(1/I) = v(X\II) (where v is defined after
I.(2.30», so that (1.71) follows from I.(2.45). Equation (1.72) is then derived from
(1.69), I.(2.42), and (1.70). •

Hamiltonian G-actions on lP1t will, in general, fail to preserve the transition

probabilities 1.(2.65), and will therefore not be given by a (projective) represen-
tation U(G) on 1i. For this reason the Poisson algebra Cf(g~) is not useful in
quantum mechanics.
There are three algebras that do playa role in quantum mechanics analogous to
the job performed by Cf(g~) in classical mechanics. One will be constructed in
this section, the other two in the next.
Definition 1.6.2. The enveloping algebra U(gc) of 9 is the quotient of the
complexified tensor algebra 7(gc) = EB~o ®n gc (where ®OgC := C) by the
200 III. Groups, Bundles, and Groupoids

two-sided ideal I generated by all elements of the form X ® Y - Y ® X - [X, V],

where X, Y E ge.
In other words, T (gc) is the complex vector space consisting of linear combina-
tions of complex elements of the form X I ® ... ® Xko Xi E ge, with algebra product
given by concatenation. The productinU(gc) is the one inherited from T(gc). The
quotienting procedure imposes the relation X ® Y - Y ® X - [X, Y] = 0 for all
X, Y E ge inU(gc). We denote the image of XI ® ... ® Xk E T(ge) inU(gc)
simply by X I ' .. Xko so that XY - Y X = [X, Y] in U(gc).
Definition 1.6.3. The twisted enveloping algebra Ur(gc) of g relative to r E
Z2(g, Ii) is the quotient of T(gc) by the two-sided ideal Ir generated by all
elements of the form X ® Y - Y ® X - [X, Y] + ir(X, Y).
Here r(X, Y) is seen as an element of ®Oge. This time one has the relation
X ® Y - Y ® X - [X, Y] + ir(X, Y) = 0 inUr(gc).
Proposition 1.6.4. The twisted enveloping algebra Ur (gc) is isomorphic to the
quotient ofU (gr ) by the two-sided ideal generated by To + i.
This is obvious.
The algebra Ur(gc) (and hence its special case U(ge» has a natural involution
•
(cf. 1.1.1), given by linear extension of (XI ... Xk)* = (_I)k Xk'" XI (as well as
A* = I for A E C); this is well-defined, and descends from a similarly defined
evolution on T(gc) because the relation mentioned in 1.6.3 is stable under it.
Definition 1.6.5. As a real vector space, the Jordan-Lie algebra ~~(g), defined
for each Ii 1= 0, is the subspace of elements OfUr/h(gc) that are invariant under
the involution. Its Jordan product and Poisson bracket are given by the projection
oftheoperationsAoB:= !(A®B+B®A)and{A, B}h = i(A®B-B®A)/Ii,
defined on T (gc), to Ur /h(gc), respectively.
Note that the projection of these operations to ~~(g) is well-defined, since
A Ir/h c Ir/h and {B, Ir}h C Ir/h for all A, B E T(gc). One verifies 1.(1.6).
0
The analogue of Proposition 1.4.3 is
Proposition 1.6.6. One obtains an automorphic group action fJ y / h of G on
Ur/h(gc) and thence on ~~(g) by defining
fJI(X) := Ad(x)X - iy(x-I)(X) (1.73)

on g C T(gc), extending this action to T(gc) by fJI (1) = 1 and fJI (X I ® ... ®
Xk):= fJI X I ®·· '®fJI Xk,projecting the action toUr/h(gc),andfinally restricting
it to ~~(g).
The fact that one indeed has a group action follows from (1.22). On T(ge) the
action is automorphic by construction. The fact that it quotients well to Ur(gc)
follows from the property
fJI (X ® Y - Y ® X - [X, y] + ir(X, Y» = Ad(x)X ® Ad(x)Y-
 1 Lie Groups and Lie Algebras 201

Ad(x)Y ® Ad(x)X - [Ad(x)X, Ad(x)Y] + ir(Ad(x)X, Ad(x)y), (1.74)

which is a consequence of (1.60) and the fact that Ad(x) is an automorphism of
g. Finally, the restriction to 1.2l~(g) is well-defined because pi (X*) = (Pi (X»*
etc. •
The Jordan-Lie algebra 1.2l~(g) is a quantum analogue of the Poisson algebra
Pr(g~) of (real) polynomials on g*, equipped with the Poisson bracket (1.12)
inherited from Cf(g~). A quantization map is constructed as follows.
Theorem 1.6.7. The map Q~ : Pr(g~) ~ 1.2l~(g) defined by Qn(lg.) := 1 and
linear extension of
(1.75)

satisfies
{Q~(A), Q~(B)}n = Q~({A, B}~) + O(h); (1.76)
Q~(A) 0 Q~(B) = Q~(AB) + O(h); (1.77)
Q~(Pr(g~J) = 1.2l~(g). (1.78)
Here the symmetrization operation A is defined after 11.(2.37). The first two
equations are a matter of checking the definitions. The third one follows from the
fact that symmetrization establishes a vector space isomorphism between Ur(gc)
and the symmetric tensor algebra 8(g); this follows from the Poincare-Birkhoff-
Witt theorem, which is well known for U(gc), hence valid for U(grc), and holds
for U r (gc) in view of Proposition 1.6.4. •
Proposition 1.6.8. With f3Y and CoY given by (1.73) and (1.23), respectively,for
all x E G and A E P(g*) one has the equivariance property
(1.79)
The proof is a simple calculation. o
Due to the integrability problem, not all representations of 1.2l~(g) are related
to unitary G-actions. In addition, 1.2l~(g) has the drawback of not being a J LB-
algebra, so that much of the functional-analytic apparatus developed in Chapter I
is not available.

1.7 Group C* -Algebras

We will now construct an object free from these drawbacks. To simplify the nota-
tion, we assume that the multiplier c is globally smooth; when it isn't, one should
replace C~(G) in the discussion below by the space B;;o(G) of bounded mea-
surable functions with compact support that are smooth near e. We also assume
that G is unimodular; that is, each left Haar measure is also right-invariant. This
assumption is not necessary, but simplifies most of the formulae. We denote Haar
measure by dx; it is unique up to normalization. When G is compact we choose
202 III. Groups, Bundles, and Groupoids

the normalization so that fa dx = 1. The Banach space LI(G) and the Hilbert
space L2(G) are defined with respect to Haar measure.
Given a multiplier c, we define the (twisted) convolution

f * g(x):= fa dy C(xy-I, y)f(xy-I)g(y). (1.80)

This certainly makes sense for f, g E C;:O(G). The associativity of * is a conse-

quence of the associativity of group multiplication and the invariance properties
of the Haar measure (and, when c f. 1, of (1.29». Moreover, we can define an
involution on C;:O(G) by

/*(x):= c(x, x-I)f(x- I ). (1.81)

The property (f * g)* = g* * /* reflects the law (xy)-I = y-Ix-l in a group
(for c f. lone in addition needs (1.29». Hence C;:O(G) has been turned into a
*-algebra.
A representation 1r of C;:O(G) on a Hilbert space 'It is defined as a morphism 1r :
C;:O(G) -+ 1B('It). An example of a representation is 1rL : C;:O(G) -+ IB(L2(G»,
defined by
1rdf)'II := f * 'II. (1.82)
In Lemma 1.7.2 we will see that this operator is bounded. Moreover, one easily
*
verifies that JrL(f g) = 1rdf)1rL(g) and 1rL(f*) = 1rdf)*. Introducing the
left-regular representation U L of G on L2(G) by
Udy)'II(x) := c(y, y-IX)'II(y-1x), (1.83)

it follows that

(1.84)

Definition 1.7.1. The (twisted) reduced group C* -algebra q(G, c) is the small-
est C*-algebra in IB(L2(G» containing 1rdC;:O(G». In other words, q(G, c)
is the closure of the latter in the norm Ilflir := l11rdf)lI. We write q(G) for
q(G,I).
Perhaps the simplest example of a reduced group algebra is obtained by taking
G = ]Rn. Since the Fourier transform f ~ 1
turns convolution into pointwise
multiplication, the algebra q(]Rn) is commutative. Indeed, the left-regular rep-
resentation 7rL on L2(]Rn) is Fourier-transformed into the action on L2(]Rn) by
multiplication operators. Hence
IIfllr = 1111100, (1.85)
so that by the Riemann-Lebesgue lemma and the Stone-Weierstrass theorem,
(1.86)
This generalizes to arbitrary abelian Lie groups G (and, more generally, to locally
compact abelian groups). Let GC be the set of all irreducible c-representations of
 1 Lie Groups and Lie Algebras 203

G; for c = 1 this is the set of characters, and we write G := G1• It is well known
that G is itself a locally compact abelian group, in terms of which the Fourier
f
transform (which is a function on G) of f E LI(G) may be defined as

f(y):= fa dx f(x)Uy(x). (1.87)

By the same arguments as for IRn , one obtains

(1.88)
We return to the general, possibly twisted case.
Lemma 1.7.2. Let U be an arbitrary continuous c-representation of G on a
Hilbert space 1t. Then rr(f), defined by

rr(f):= fa dx f(x)U(x), (1.89)

is bounded, with
( 1.90)
Since U is unitary, we have 1(\11, rr(f)\II)1 :::: (F, F)u(G) for all \II E ri,
where F(x) := 1I\IIIIJlf(x)l. The Cauchy-Schwarz inequality then leads to
1(\11, rr(f)\II)1 :::: II 111111\11112. The argument in the proof of 11.1.3.5 then leads to
(1.90). (A more sophisticated proof uses properties of Bochner integrals to argue
that IIrr(f)1I :::: JG dx If(x)1 IIU(x)1I = IIfliI·) •
The following result generalizes the correspondence between UL in (1.83) and
rr L in (1.84) to arbitrary representations.

Theorem 1.7.3. There is a bijective correspondence between nondegenerate

representations rr of the *-algebra C~( G) that satisfy (1.90) and continuous c-
representations U of G. This correspondence is given in one direction by (1.89),
and in the other by
(1.91 )
where fX(y) := c(x, x-1y)f(x-1y). This bijection preserves direct sums, and
therefore irreducibility.
It is technically convenient to extend the *-algebra C~(G) to a Banach algebra
Ll(G, c); this is Ll(G) as a Banach space. The operations (1.80) and (1.81) are
easily seen to be continuous on the LI-norm, so that they may be extended from
C~(G) to Ll(G). Recall from 1.1.5.2 that any nondegenerate representation of a
C* -algebra is a direct sum of cyclic representations; the same can be shown to be
true of L 1(G , c). Thus Q in (1.91) stands for a cyclic vector of a certain cyclic
summand of1t, and (1.89) defines U on a dense subspace of this summand; it will
be shown that U is unitary, so that it can be extended to all of 1t by continuity.
Given U, it follows from easy calculations, using (1.65), that rr(f) in (1.89)
indeed defines a representation. It is bounded by Lemma 1.7.2. The proof of non-
degeneracy makes use of the existence of an approximate unit in LI(G, c), which
204 III. Groups, Bundles, and Groupoids

heuristically converges to the Dirac delta function De. This is constructed as fol-
lows. Consider a basis of neighborhoods N).. of e, partially ordered by inclusion.
Choose II).. = N)..XN)., which is the characteristic function of N).. times a normal-
ization factor ensuring that II II).. II 1 = 1. One can then show that lim).. II).. * f = f (in
*
I . Iii) for all f ELI (G, c)( and similarly for f II)..); for c =1= 1 it is at this point that
the continuity of c near e is used. Since Jr is continuous, one has lim).. Jr(II)..) = II
strongly, proving that Jr must be nondegenerate.
To go in the opposite direction we use the approximate unit once more; it follows
from (1.91) (from which the continuity of U is obvious) that U(x)Jr(f)Q =
lim).. Jr(IIDJr(f)Q. Hence U(x) = lim).. Jr(II~) strongly on a dense domain. The
property (1.65) then follows from (1.91) and (1.80). Since IIJr(lIDIl :::: 1I1Ifili = 1,
we infer that II U (x) II < 1 for all x. Hence also II U (x -I) II < 1. From (1.65) we
derive

(1.92)

so that II U (x )-111 :::: 1. We see that U (x) and U (x) -I are both contractions; this is
possible only when U(x) is unitary.
Finally, if U is reducible, there is a projection E such that [E, U (x)] = 0 for all
x E G (see 1.2.2.2). It follows from (1.89) that [Jr(f), E] = 0 for all f; hence Jr
is reducible. Conversely, ifJr is reducible, then [E, Jr(IID] = 0 for all x E G; by
the previous paragraph this implies [E, U(x)] = 0 for all x. D

This theorem suggests looking at a slightly different object from C:(G, c).
Inspired by 1.1.5.7 one makes the following

Definition 1.7.4. The (twisted) group C*.aIgebra C*(G, c) is the closure of the
convolution algebra Cgo(G) in the norm

11111 := IIJru(f)1I = sup IIJr(f)II, (1.93)

7f

where Jr u is the direct sum of all nondegenerate representations Jr of crgo (G) that
are bounded as in (1.90). We write C*(G)for C*(G, 1).
By Theorem 1.7.3 the representations Jr occurring in the sum are those associated
with representations U(G) via (1.89).

Corollary 1.7.5. There is a bijective correspondence between nondegenerate

representations Jr of the C*-algebra C*(G, c) and continuous c-representations
U ofG, given by (continuous extension of) (1.89) and (1.91). This correspondence
preserves irreducibility.
Hence one may alternatively define C*(G, c) as the closure ofCrgo(G) in the
norm (1.93), where now the sum is over all representations Jr of Crgo(G) that
correspond to an irreducible representation U (G) via (1.89).
The second part follows from the last statement of the first part and the faithful-
ness of the reduced atomic representation; cf. 1.2.2.7 etc. Hence one obtains the
same norm in (1.93) by restricting the Jr 's to be irreducible. •
 1 Lie Groups and Lie Algebras 205

Since a Lie group is separable as a topological space (separability being part of

the definition of a manifold used in this book), the algebras Cg'"(G) (inheriting the
norm of Ll(G» and hence C*(G) are (norm) separable. Therefore, all irreducible
representations are on separable Hilbert spaces, and one would obtain the same
C* -algebra by restricting the rr's in Definition 1.7.4 to be on separable Hilbert
spaces.
In conjunction with (1.85), the second definition of C*( G) stated in 1.7.5 implies
that for abelian groups, C*(G) always coincides with C;(G). The reason is that for
y E G one has rry{f) = j(y) E C, so that the norms (1.93) and (1.85) coincide.
For future reference we single out (cf. (1.86»
(1.94)

In any case, looking at 1.7.1, we see that

C;(G, c) = rrL(C*(G, c» ::::: C*(G, c)/ ker(rrd. (l.95)

A Lie group is said to be amenable when the equality C;(G) = C*(G) holds;
in other words, rrdC*(G» is faithful iff G is amenable. This turns out to imply
that also C;(G, c) = C*(G, c) for arbitrary multipliers c; we shall not prove this
remarkable result. We have just seen that all locally compact abelian groups are
amenable, so that the previous comment implies that the Heisenberg group Hn
is amenable. Hence the object C*(Hn) constructed in 11.2.6 is indeed the group
C*-algebra of Hn. It follows from the Peter-Weyl theorems in the next section
that all compact groups are amenable. It may be shown that also all solvable Lie
groups are amenable, as are direct products of the amenable groups mentioned.
To provide an alternative characterization of amenability we first describe the
connection between the representation theories of C*(G, c) and of C;(G, c).

Definition 1.7.6. The c-unitary dual GC of a group G is the collection of

equivalence classes ofirreducible c-representations ofG .In other words (cf Corol-
lary 1.7.5), GC is the set of equivalence classes of irreducible representations of
C*(G, c).
The reduced unitary dual G~ is the set of equivalence classes of irreducible
representations ofC;(G, c). For c = 1 we write G(r) := Gtr)' and speak of the
(reduced) unitary dual.

The earlier definition of the unitary dual of an abelian group is evidently a

special case of 1.7.6. The following notion provides the key to describing G~. We
say that a representation V 1(G) is weakly contained in a representation V 2 (G)
when ker rr2(C*(G, c» ~ ker rrl (C*(G, c»; here V j is related to rrj by (1.89).
For example, every subrepresentation properly contained in a representation is
weakly contained in it. However, the notion of weak containment is more general
than proper containment. Consider the regular representation V L of R.n on L 2(R.n);
since the associated representation rrL(C*(JR» is faithful (see (1.94) and the pre-
ceding discussion), its kernel is {O}. Hence every irreducible representation of R.n
is weakly contained in VL, although none is properly contained in it.
206 III. Groups, Bundles, and Groupoids

Proposition 1.7.7. The reduced c-unitary dual G~ consists of those irreducible

representations that are weakly contained in the left-regular representation U L (G).
Consequently, G is amenable iff all its irreducible representations are weakly
contained in U L.
This directly follows from the above definitions.
Remarkably, one may show that the above condition for amenability is equivalent
•
to the weak containment of merely the trivial representation in U L • Either way,
when G is a noncompact semisimple Lie group it can be shown that the trivial
representation is neither properly nor weakly contained in the (left- or right-)
regular representation. Hence such groups are not amenable.
A comparison between Theorem 1.2.6 and Corollary 1.7.5 indicates that C*(G)
is a quantum analogue of CXl(g"'J. More generally, Corollaries 1.4.7 and 1.7.5
suggest that C*(G, c) is a quantum analogue of C~(g~), and that CoY -orbits are a
classical version of projective irreducible representations. In particular, coadjoint
orbits are analogous to irreducible representations.
In addition, we can formulate an "integrated" version of Proposition 1.6.6 (and
thereby a quantum version of 1.4.3):
Proposition 1.7.8. One obtains an automorphic group action a(c) of G on
C*(G, c) by putting
a~C)(f): y 1-+ c(x, Ad(x-l)y)C(Y, x)f(Ad(x-1)y) (1.96)

for f E C;o(G), and extending to C*(G, c) by continuity.

In the universal representation one has

(1.97)

which firstly shows that a~c) is an automorphism, secondly that it can be extended
to C*(G, c), and thirdly that (1.96) defines a group action. •

1.8 A Generalized Peter-Weyl Theorem

Further to the left-regular representation U d G) in (1.83), which is a c-
representation, consider the right-regular representation UR(G) on the Hilbert
space L2(G), defined by
(1.98)

this is a c-representation. Note that

1T/i(f)\II := L dx f(X)UR(X)-I\ll = \11* f, (1.99)

where convolution (1.80) is defined with respect to c. It immediately follows from

(1.99) and (1.82) that ULand U R commute; this may also be verified directly, using
(1.29).
 1 Lie Groups and Lie Algebras 207

Recall Definition 1.7.6. Each Y E {;c has a conjugate Y E {;e; a representative

Uy of the class y is obtained by defining a representative U y of y on the conjugate
Hilbert space 11. y rather than on 11.y (hence 11.y = 11.y ).
After these definitions we recall (a version of) the Peter-Weyl theorem. This
theorem states that for a compact Lie group G one has
L 2(G) ~ E911. y ® 11.y , (1.100)
YEG

under which decomposition

UdG) ® UR(G) ~ E9 Uy(G) ® Uy(G). (1.101)
YEG

The direct sum is, of course, meant in the Hilbert space sense. This is usually
stated and proved for c = 1, but is, in fact, valid for any multiplier; see below.
One may identify 11.y ® 11.y with !mdy (C) (where dy is the dimension of the
representations in the class y) as Hilbert spaces by letting v ® W E 11.y ® 11.y
correspond to the operator mapping U E 11.y to (w, u)v, and extending by linearity.
The inner product on !mdy (C) is then given by (M, N) = Tr M* N. We accordingly
rewrite (1.100) as
L\G) ~ i}(G):= E9 !mdy(C). (1.102)
YEG

Writing (h,R(X) for the operator on i.2(G) that is equivalent to UL,R(X) on L2(G)
under the isomorphism (1.102), we may rephrase (1.101) as
Udx)\I1(y) = Uy(x)\I1(y); (1.103)
U R(x)\I1(y) = \I1(y)Uy (x)*. (1.104)
The essential step in the proof of the Peter-Weyl theorem consists in showing
that the Plancherel transform V : L2(G) ~ f2(G), defined by

"'(Y) := V\I1(y) = .[d; fa dx \I1(x)Uy(x), (1.105)

is unitary. The inverse transform can then be computed from unitarity as

V-1",(X) = E#rTr["'(y)Uy(X)*]. (1.106)
yEa

The Peter-Weyl theorem (with multiplier) has an interesting reformulation, also

valid for c =I=- I, which in a certain sense is a quantization of Theorem 1.4.4.
Theorem 1.8.1. For a compact Lie group G one has
C:(G, c) ~ C*(G, c) ~ E9 !mdy(C). (1.107)
YEO c

Here the direct sum oJmatrixalgebras includes those sums $yMy of matrices for
which the function y 1-+ liMy II is in lo(G C ).
208 III. Groups, Bundles, and Groupoids

Note that the definition of the direct sum is different from the one in (1.102).
The proof below uses some elementary aspects of the theory of induced group
representations. This subject will be studied in great generality in 2.9; for the mo-
ment we need just a very special and simple case. Let H be a compact subgroup
of a unimodular locally compact group G c , and let Vx be a I-dimensional repre-
sentation of H. (In the application below, G c will indeed be a central extension,
but for the moment it is arbitrary as stated above. We will, accordingly, denote its
elements simply by x.) The Hilbert space 11. x C L2(G c ) is defined as the set of
functions \II E L2(G c ) that satisfy the equivariance condition
(1.108)
for (almost) all x E G c, h E H; the inner product on 11. x is the one inherited from
L2(G c ). In other words, 11. x is the subspace of L2(G c ) that transforms trivially
under V R ® Vx(H), where V R(h)\II(x) := \II(xh). The left-regular representation
(1.83) (with c = 1) restricts to a representation VX(G c ) on 1t x , which is said to
be induced by V X. In other words, for \II E 11. x one has
VX(y)\II(x) := \II(y-I X). (1.109)

The projection VX : L2(G c ) ---+ 1t x defined by

Vx\ll(x):= £ dh \II (xh)Ux (h) (1.110)

obviously intertwines V L and V x, i.e., Vx 0 V L = V X 0 Vx . Moreover, when H is

abelian, Fourier analysis on H shows that EBXEH Vx = II, so that for such H

L2(G c ) ~ EBXEH1tX;
VdG c ) ~ EBXEHVX(G C ). (1.111)

An equivalent realization of VX may be defined on L\GcIH) (defined with

respect to a suitably normalized G-invariant measure, which exists because G c
and H are unimodular), as follows. Choose a cross section s : G cI H ---+ G c (i.e.,
r 0 s = id, with T := rGc~GclH)' and define VX : L2(G c ) ---+ L2(G cl H) by
VX\lI(q) := \II(s(q», (1.112)
with adjoint
(1.1l3)
It follows that V x is a partial isometry, which is unitary on the image of (V X)*,
which is 1t x . Putting V/ := V x V x (V X)* , one obtains
VsX(x)ct>(q) = Vx(S(q)-lxs(X-Iq»ct>(X-lq). (1.114)

We will apply this to the case where G c is as defined in 1.3.3 and H = Vc(l),
so that Gel H = G. In the following result G is not necessarily compact.
Lemma 1.8.2. The representation Vc(G c) associated (by 1.5.1) with the c-
representation VdG) on 1t = L2(G) defined by (1.83) is equivalent to the
representation VI(G c ) induced by the defining representation VI ofVc(l).
 1 Lie Groups and Lie Algebras 209

This is verified using the cross section s : G -+ G c given by s (x) = (x, 1). The
property
UI(e, z) = zIT (1.115)
follows from (1.108), (1.109), and the fact that Uc (1) is central.
Let us now assume that G (and hence G c ) is compact. The well-known Frobe-
•
nius reciprocity theorem states that the number of times a given irreducible
representation U(G c) occurs in}(X is equal to the number of times Ux(H) occurs
in U(G c f H) (i.e., the restriction of U to H).
Hence a given irreducible representation U y (G c) occurs in the decomposition of
Uc(G c) with multiplicity equal to the number of times the defining representation
of Uc(l) occurs in U y . By 1.5.1 this multiplicity must equal dy, since Uc(l) is
always in the defining representation times the unit matrix. Hence by 1.5.1 all
c-irreducible representations y of G occur in L 2( G) with multiplicity d y , as in the
case c = 1.
Clearly, the Hilbert space VJtdy(C) carries a c-representation U~2)(G) given by
U?)(x)M := Uy(x)M, which is the irreducible c-representation Uy(G) with mul-
tiplicity d y • Here U y is some representative of y; everything that follows depends
on the choice of this representative, but other choices lead to equivalent statements.
We recall the orthogonality relations for a compact group K: Given an
irreducible representation UI«K) of dimension dl( one has

dl( l dx (\III, UI«x)\II 2)(Udx)\II3 , \11 4 ) = (\III, \11 4 )(\11 3 , \11 2), (1.116)

It follows from these relations for G c that Py : L2(G) -+ VJtdy(C), defined

by Py\ll := q,(y) (see (1.105», is a partial isometry (note that L2 S;;; Lion
compact spaces, so that P y is we11-defined). Trivially, PyUL = U~2) P y . From the
preceding two paragraphs we conclude that the map V : L2(G) -+ i.2(G) given
by V := ffiYEGCPy is unitary, and satisfies VU L = (h V. Of course, V is the
Plancherel transform (1.105).
This shows that, as in the case c = 1, the left-regular representation UL on
L 2(G) contains all irreducible c-representations. It then follows from 1.7.1 and the
comment after 1.2.2.7 that C;(G, c) = C*(G, c).
For f E C~(G) we have from (1.84) and (1.105) that
irL(f) := VJrdf)V- 1 = ffiYEGCJry(f) ® ~y; (1.117)

cf. (1.89). The map f f-+ VJrL(f)V- 1 is a *-isomorphism from C~(G) into
ffiyEGc VJtdy (C), seen as direct sum of matrix algebras, since Jr L is a faithful
representation and V is unitary. It can therefore be extended by continuity. The
irreducibility statement in 1.7.5 implies that Jr y (C; (G, c» = VJtdy (C).
Finally, to prove that the direct sum in (1.107) should be defined as stated, first
note that JrL(f) E s:B I (L 2(G» for f E C~(G). Hence JrL(f) E s:B O(L2(G» by
1.(1.62), so that JrdC*(G» c s:B O(L2(G» by the continuity of JrL. Since V is
unitary, it follows that VJrL(C*(G»V-I E s:BO(L2(G». It is then easy to adapt
210 III. Groups, Bundles, and Groupoids

the standard proof that the eigenvalues of a compact self-adjoint operator (ordered
from large to small) go to zero to conclude that limy-->oo lI]l'y(f)1I = O.
Theorem 1.8.1 follows. •
We write ]l' k for the representation of C* ( G c) corresponding to the representation
Uk(G c ) induced by Uk(Uc(l» (see (1.89», where k E Z and Uk(Z) := Zk for
=
Z E 'll' U (1). For G possibly noncompact, the first stage of the above proof leads
to
Corollary 1.8.3. For each k E Z there are isomorphisms
C;(G, e k )::::::: ]l'k(C*(G c»::::::: C*(Gc)/ker(]l'k). (1.118)
Explicitly, under the first isomorphism the function rrk(f) E C;(G, ek ) (where
f E C~(Gc) C C(G c is »
rrk(f) : x ~ l dz l f(x, z). (1.119)

Here dz is the normalized Haar measure on 'll'. This corollary is proved by a

straightforward generalization of Lemma 1.8.2: Given a ek-representation U of G,
one defines an associated representation Uck of G c by Uck(X, z) := zkU(x), and
verifies that Uck ::::::: Uk. •

For k = lone should compare 1.8.3 with 1.1.10. As we have seen in 1.3, the
multiplier c is a derived object, the intrinsic object being the central extension
(1.27). Hence C;(G, c) is not quite intrinsic either, but Corollary 1.8.3 shows
how to define the intrinsic analogue of C;(G, c): It is C*(G)/ ker(rr 1). This C*-
algebra is, of course, isomorphic to C;(G, c), and also to any C;(G, e'), where c'
is equivalent to c. The case of general k will be used in the next section.
Corollary 1.8.3 is closely related to the decomposition
C;(G c ) ::::::: EDkeZ rrk(C*(G c », (1.120)

»
which follows from C;(G c) = rrL(C*(G c and (1.111). Equation (1.118) shows
that C;(G, c) is isomorphic to a (closed 2-sided) ideal in C;(Gd, namely the one
»
that is isomorphic to rrl(C*(G c by (1.120).
As an application of (1.118) we prove
Proposition 1.8.4. Let e be the mUltiplier on ]R2n given in (1.35). Then for all
k E Z\{O} there are isomorphisms
C*(]R2n, ek ) ::::::: C;(]R2n, ck ) ::::::: ~o(L 2(]Rn». (1.121)

We will not prove the first isomorphism here; the proof is identical to that
of Theorem 3.7.1 below. As to the second, we saw (after (1.35» that ]R~n =
Hn. We use the notation of Lemma 1.8.2. Using (1.115), 11.(2.18), and Theorem
11.2.1.4, or direct calculation, one shows that Uk (Hn) is a multiple of the irreducible
SchrOdinger representation Uf(Hn) defined in 11.(2.17). The second isomorphism
in (1.121) then follows from 1.8.3 and 11.(2.129) (with fIn replaced by Hn).
 I Lie Groups and Lie Algebras 211

Here is a direct proof as well. Define a map Kk : S(JR 2n ) --* S(JR2n) by

(1.122)

with inverse

(1.123)

The well-known properties of the Fourier transform show that Kk is an isomor-

phism (of vector spaces to begin with). We identify the image S(JR.2n) of Kk with
S(JRn x JR.n), regarded as a space of kernels on JR.n, and as such as a subspace of
~o(L2(JR.n»; cf. 11.2.5.3. One calculates from (1.80) and (1.81) with (1.35) that
Kk is a *-isomorphism between dense subspaces of C;(JR.2n, ck ) and ~o(L 2(JRn».
Then extend Kk by continuity. •
Note that the multiplier has a crucial effect: Without it, one has the isomorphism
ct)(JR2n) ~ Co(JR.2n) by Fourier transformation; cf. (1.86).
In the light of Example 1.2.11.3, the classical analogue of Proposition 1.8.4 (for
k = 1, say) states that the Poisson algebra Cf(JR.2n, JR.) (where JR2n is seen as the
dual Lie algebra g* for G = JR2n) is coo(T*JRn , JR.) equipped with the canonical
Poisson bracket 1.(2.24); cf. 1.9.6.
To close this section, we "quantize" the realization of Cf(g~) stated in 1.4.2.
Denote the operators in a C* -algebra Q( C ~(L 2(G» that commute with U R(G)
(cf. 0.98» by Q(R. The following result is nontrivial even for c = 1.
Proposition 1.8.5. When G is compact, JrdC*(G, c» = ~O(L2(G»R.
The Plancherel transform (1.105) maps U R into {; R := V U RV-I, given by
{;R(X)q,(y) = q,(y)Uy(x)-I. The result then follows from (1.l17). •

1.9 The Group C* -Algebra as a Strict Quantization

When G is compact the C*-algebra C*(G) turns out to be related to the
(complexified) Poisson algebra COO(g~) by a strict quantization.
Analogously to 11.(3.49), we define C~(g*) as the class of functions on g*
whose Fourier transform j is in C?,,(g). Here the Fourier transform of j E L I(g*)
is defined by (cf. 11.(3.42»

j(X) := 1 g*
n
d (}
(2Jr)n
eiB(X) j«(}), (1.124)

where d n() is Lebesgue measure on g* ~ JR.n, whose normalization is fixed by that

of the Haar measure dx on G, as follows. When j has support near e, we can write

1 dx j(x) = 1 d n X J(X)j(Exp(X», (1.125)

where d n X is a Lebesgue measure on g, and J is some Jacobian. The normalization

is now fixed by the condition J (0) = 1. In turn, the normalization of the Lebesgue
212 III. Groups, Bundles, and Groupoids

measure dnO on fl* is fixed by requiring the inversion formula

f(O) = ~ d n X e-ili(X) j(X). (1.126)

As in the argument after 11.(3.49), one infers that C~(fll, JR) is a Poisson subalgebra
of COO(fll, JR).
We choose a smooth cutoff function K on fl that equals I in a neighborhood jj
of 0, is invariant under inversion X H - X, and has support in the neighborhood
N of 0 on which Exp is a diffeomorphism; cf. 11.3.4.4. When G is compact one
may assume that K is Ad-invariant; i.e., satisfies K(Ad(y)X) = K(X) for all y E G.
This may always be achieved by averaging.
Definition 1.9.1. For an n-dimensional Lie group G, the RietTel quantiza-
tion Qf : C~(fl~) -+ C*(G) is defined as follows: For x rJ. Exp(N) we put
Q~ (f)(x) = 0, whereas for x E Exp(N) we put

Qf,(f)(x) := lI,-n K (Exp-' (x»j(Exp-' (x)/Il,). (1.127)

Analogously to QJ[ in 11.3.4, the restriction f E C~(fl*) implies that for small
enough II, the operator Qf, (f) is independent of K. When G is exponential (in that
Exp : fl -+ G is a diffeomorphism) the cutoff K can be omitted altogether. For
general f E COO(fl*) the object Qf(f) is a distribution on G. In particular, when
f is a polynomial, one obtains a distribution with support at e. One may identify
the set of such distributions with the enveloping algebra U(gc), but even on this
space Qf, does not coincide with Qf, in (1.78) unless the Jacobian 1 appearing in
(1.125) equals unity. However, for unimodular groups, and therefore in particular
for compact groups, one has leX) = I + O(X 2 ), and this property suffices to
guarantee that, at least formally, Qf,(X) = Q~(X) for all X E g. Since Qf,(X) is
not defined as an element of C*(G), one may pass to a representation 1l'(C*(G»,
related to U(G) by (1.89). Formally, one then has
1l'(Qf,(X» = iMU(X). (1.128)
Theorem 1.9.2. Let Qto = CO(fl*) and Qtll = C*(G) for II, rJ. O. When G is
compact, the map Qf, in 1.9.1 yields a strict and continuous quantization ofSil~ :=
C~(fl~, JR) on 1= JR, up to condition II.I.l.lA.

Strictness, which implies continuity (cf. 11.1.2.5) by Theorem 11.1.2.4, will fol-
low from the fact that Qfis a special case of the generalized Weyl quantization
prescription on Riemannian manifolds (cf. 11.3.4).
Lemma 1.9.3. A compact Lie group G admits a right-invariant Riemannian
metric g such that the exponential map eXPe obtained from g coincides with the
map Exp defined by the Lie group structure.
Choose an inner product (, ) on TeG = fl that is invariant under the adjoint
action of G. and define g by the property g;;(~:. ~:) := (X, Y); cf. (1.38). this is
evidently right-invariant, but due to the Ad-invariance of (. ) it is left-invariant as
well. Such metrics are called bi-invariant.
 1 Lie Groups and Lie Algebras 213

For any right-invariant metric g and any point of G one has the identity

g (v~{~:, ~:) = k{g ([~:,~:J,~:) - g ([~:, ~n, ~n + g ([~:, ~n, ~n},

(1.129)
obtained from 11.(3.17) and 11.(3.18) with various permutations of the entries, using
the x-independence of gx(~:, ~:). When g is Ad-invariant, the last two terms
cancel, upon which the nondegeneracy of g implies that V~:~: = k[H, ~:J due
to the nondegeneracy ofg. In particular, V~:~: = 0, so that by 11.(3.4) we infer that
the curves x(t) = Exp(t Y)x are geodesic for all Y E g. The claim follows. •
We identify C*(G) with nLCC*(G» (see (1.95) and 1.8.1), which in turn is
expressed as in 1.8.5. Also, we identify C~(g*) C COO(g*) with C~(T*G)R c
COO(T*G)R, as in 1.4.2 (with r = 0). Choosing a metric on G as in the lemma, it
follows from 11.(3.51), 11.(3.48), (1.82), (1.80) (with c = 1), and (1.127) that under
the above identifications one has Qf = QJion C~(T*G)R (confirming 11.3.6.3).
The theorem then follows from Theorem 11.3.5.1. •
Proposition 1.9.4. Suppose that G admits a metric g as specified in Lemma 1.9.3,
and that the cutoff K is Ad-invariant (these assumptions are satisfied when G is
compact). With a := a(l) given by (1.96),for all x E G and f E §to one has
ax(QfU» = Qf(Co(x- I )* f). (1.130)
The metric g of Lemma 1.9.3 is bi-invariant, and g.. is invariant under the adjoint
representation Ad(G). Identifying g ~ JRn this implies that Ad(G) ~ SO(n); in
particular, Ad(G) and Co(G) leave the Lebesgue measures invariant on g and g*,
respectively. The claim then follows from (1.127), (1.124), and the Ad-invariance
of the cutoff K .
Alternatively, the claim follows from Proposition 1.2.3, equation (1.52), and
Theorem 11.3.6.3. •
This proposition is a "bounded" version of 1.6.8.
Theorem 1.9.2 can be generalized to the twisted case, at the cost of It being
defined only at a discrete set. Let r E Z2(g, JR) be related to c E Z2(G, U(1» by
(1.34). This leads to a Poisson algebra C~(g~, JR) (see 1.1.5), a central extension
G c with Lie algebra gr (see 1.3.3 and 1.3.6), a group C*-algebra C*(G c ), and a
twisted group C*-algebra C*(G, c) (cf. 1.7).
Theorem 1.9.5. Let Q(o := Co(g*) and Q(1i := C*(G, c l / Ii ) for It = 1/ k, where
k E IE, and let sit~ := C~(g(r)-' JR) be equipped with the Poisson bracket (1.12),
(1.13), taking the minus sign. When G is compact, the map Qf, defined in 1.9.1
with C*(G) replaced by C*(G, clift), is a strict quantization ofQi~ on 10 = I/Z
(except possibly for the completeness condition lI.1.1.1.4).
The signs may be checked from (1.128), (1.67), and (1.15). The proof is based
on the analogy between 1.1.10 and 1.8.3. Extend f E C~(g*, JR.) to a function
i E C~(flf, JR.), such that fee) = i(1, e) and
i(eo i= 1, e) < i(eo = 1, e) = fee); (1.131)
214 III. Groups, Bundles, and Groupoids

in particular, one has

Ilflloo = 1111100' (1.132)

In view of (1.13) this automatically means that

(j, g}-(l, e) = {j, g}~(e), (1.133)

since the left-hand side does not involve derivatives with respect to eo.
We denote the map of Definition 1.9.1 as defined on C~(g*, JR.), taking values
Qf,
in C*(G, c 1/ 1i ), by whereas the map defined in the same way, but now on
C~(g~, JR.), taking values in C* (G c), is written as Qf. A short computation using
(1.119) and an elementary oscillatory integral shows that

(1.134)
for Ii E I/Z small enough so that the right-hand side is independent of K. In
particular, the left-hand side depends only on the value of I at eo = 1; this is
a special case of the fact that, for Ii small enough, nk(Qf(I» depends only on
I(eo = kli). This follows by a calculation similar to the one leading to (1.134).
Theorem 1.9.2 applied to G c implies that limli--->o II Qf(I)1I = 1111100' On the
other hand, according to (1.120) one has IIAII = SUPkEZ IInk(A)1I for all A E
C* (G c). Combining the two of these equations with the last remark of the preceding
paragraph and the property (1.131), we conclude that
(1.135)

Together with (1.132) and (1.134) this proves II.(1.1) for Qf.
Conditions IL(1.2) and 11.(1.3) in Theorem 1.9.5 now follow from (1.134),
Corollary 1.8.3, (1.133), Theorem 1.9.2 (once again applied to G c ), and the
inequality IInk(A)1I :s IIAII in C*(G c); cf. U.S. •

The obvious generalization of 1.9.4 (in which Co is replaced by CoY, where y

and r are related by (1.24» is not valid except in special cases (see below).
While proved for compact G, Theorem 1.9.5 may hold in other situations.
Proposition 1.9.6. Let G :::::: g* = JR.2n, with r given by (1.26) and c defined in
(1.35). Then the statement of Theorem 1.9.5 holds (without the final qualification).

Using Proposition 1.8.4 one obtains that Qf = Qj; (cf. 1I.2.5), so the
proposition follows from Theorem IL2.6.1. •
In this case one does have the "twisted equivariance property"
(1.136)
»,
for all x E JR.2n and f E C~(T*JR.n) (or S(T*JR.n where y is related to r by
(1.24). This follows by direct computation; in (1.23) only the term y(x) con-
tributes, yielding CoY (u, v) : (p, q) f-+ (p + u, q + v). Alternatively, one uses the
corresponding property II.(2.93) of Qj;.
 1 Lie Groups and Lie Algebras 215

1.10 Representation Theory o/Compact Lie Groups

Following the discussion of the Weyl quantization of g*, we tum our attention to
the possible quantization of coadjoint orbits C) in g*. In view of later applications
to gauge theories we restrict ourselves to the case that G is compact. We start with
a brief review of the relevant representation theory, assuming familiarity with the
standard Cartan-Weyl approach. Throughout this section G is a compact connected
Lie group unless stated otherwise, and all representations are finite-dimensional.
Firstly, let G be abelian; it then has to be a torus G = T = '['r = U (l)' . Each
irreducible representation of T is one-dimensional, and is a character UA : ']I" ---*
C (a character of an arbitrary group is a one-dimensional representation). The label
'A of the character is an element of t* ::::: IR r (the dual of the Lie algebra t ::::: IR r of
T), related to UA by

dUA(X) = -i'A(X). (1.137)

It follows that 'A E A := 'ZI c t*. Conversely, each 'A E A defines an irre-
ducible representation of T by exponentiation, so that we have found a bijective
correspondence between the unitary dual f and the lattice A C t*.
For a general Lie group, we note that (1.61) (with r = 0) implies that 0 E g*
satisfies O([X, Y]) = 0 for all X, Y E ge (where gil is the Lie algebra of the
stabilizer Gil of 0 under the co adjoint action). In other words, 0 : gil ---* IR is a Lie
algebra homomorphism.
Definition 1.10.1. A coadjoint orbit C) E g* is called integral iffor some (hence
all) 0 E C) the functional 0 f gil exponentiates to a character ofG e.
Inotherwords,O isintegraliffthere is a character UII of Go such that 0 = idUII on
gil. If this holds for one 0 E C),itholdsforall, since one has UCo(x)1I = UooAd(x- 1 ).
Obviously, if G is a torus T, its coadjoint action is trivial, so that its coadjoint
orbits are the points of t*; the integral orbits are precisely the elements of the
lattice A. Consequently, one has a bijective correspondence between f and the set
of integral coadjoint orbits of T. The following theorem generalizes this idea.
Theorem 1.10.2. There exists a bijective correspondence between the unitary
dual G and the set of integral coadjoint orbits in g*.

We will merely sketch the proof in explaining how this parametrization of Gis
related to the Cartan-Weyl theory. This theory starts by choosing a maximal torus
T (i.e., a maximal connected abelian subgroup) of G, with associated Weyl group
W := N(T)/T (where N(T) is the normalizer of T). The integer r := dim(T)
is called the rank of G; it does not depend on the choice of T, since all maximal
tori are conjugate. The Weyl group acts on T by conjugation, and hence it acts
on t and t*. The latter action is the projection of the coadjoint action of N(T).
It maps A C t* (called the weight lattice in the present context; elements of A
are traditionally called weights) into itself; the W -action on A coincides with the
natural W -action on f under the identification of f with A explained above. The
Cartan-Weyl theory states
216 III. Groups, Bundles, and Groupoids

Theorem 1.10.3. Let T be some maximal torus in G. There exists a bijective

correspondence between the unitary dual {; and the set of W -orbits in f . that is,
{;~f/W=A/W.
The relationship between this parametrization of (; and the one in 1.10.2 follows
from an important lemma in the structure theory of compact Lie groups, which we
state without proof.
Lemma 1.10.4. In the notation of 1.10.3 there is a bijection
G/Ad(G) ~ T/W. (1.138)
Denoting the set of coadjoint orbits in g* by g* / G, one therefore has
g*/G ~ t*/W. (1.139)
The concrete association of a coadjoint orbit in g* with a W -orbit in t* is as
follows. Restrict the adjoint representation of G (extended to the complexification
gc) to T, and decompose gl(: under Ad(T) as gl(: = tc EB ti:, where ti: is the
sum of all eigenspaces with nonzero eigenvalues; this leads to a decomposition
g = tEB tl., where tl. := ti: n g. This coincides with the orthogonal decomposition
of 9 under an arbitrary Ad( G)- invariant inner product (, ). One may, for example,
take a faithful representation U of G, and define the invariant inner product by
(X, Y):= -TrdU(X)dU(Y). (1.140)
An arbitrary compact Lie group is of the form G = S X ']['k , where S is semisimple
(i.e., a product of simple factors with finite discrete center) and ']['k is a torus. If G
is semisimple, the adjoint representation is faithful, and may be used in (1.140);
this defines the Killing form on g. More generally, all invariant metrics have
the property that the direct summands in g are mutually orthogonal, and that the
metric restricted to a given simple summand is proportional to the Killing form. For
concreteness' sake, in what follows we assume that (, ) restricted to the semisimple
part of g coincides with the Killing form.
The extension e().. ) E g* of A E t* obtained by putting e(A) = 0 on t..L and
e(A) = A on t is therefore independent of the choice of the metric on g. Thus the
coadjoint orbit OA := OO(A) associated to A is the coadjoint orbit through e(A); it
is obvious from the definition of the W -action on t* that all points of the W -orbit
of A are mapped into OA'
To go in the opposite direction one needs (1.138) to show that the stabilizer
of any point in 0 is connected, and that it contains a maximal torus. As any two
maximal tori are conjugate to each other, and Gco(x)e = x Gllx -\ , there accordingly
exists a e E 0 for which T ~ Gil' Hence we can define A(e) = e r t. Note that
(1.61) and T ~ Gil imply that e r tl. = O.
If Gil = T the coadjoint orbit 011 through e is said to be regular; it is of maximal
dimension among all coadjoint orbits. Otherwise, it is called singular. For regular
orbits one immediately sees that Gco(x)O == T implies that x E N(T), so that
different choices of e for which Gil = T map into the same W -orbit of A(e). It
follows from (1.138) that the same is true for the singUlar orbits.
 1 Lie Groups and Lie Algebras 217

Theorem 1.10.2 now follows from Theorem 1.10.3 and Lemma 1.10.4 by
restricting the isomorphism to weights and integral orbits. •
A functional A E t* is called regular when WA = A for W E W implies
W = e (and singular otherwise); this defines the sets t:
and A, := t:
n A of
regular elements and regular weights in t* , respectively. In the context of 1.10.2,
elements of t; evidently correspond to regular coadjoint orbits, and similarly for
the singular case.
Each connected component C of t; is called a Weyl chamber; this is an open
convex cone in t*. Singular weights clearly lie on the boundary of some Weyl
chamber. One singles out an arbitrary Weyl chamber Cd' and declares a weight
dominant if it lies in the closure Cd' The point is now that each W -orbit intersects
a given closed Weyl chamber C in exactly one point. Hence Theorem 1.10.3 may
now be restated:
Corollary 1.10.5. In the notation of 1.10.3 there is a bijection between G and
the set Ad := A n Cd of dominant weights.
Any Hilbert space 1i carrying a representation U(G) decomposes under U(T)
as 1i ~ tB).Etl o(U)1i)., where each 1i). carries the representation U).(T) (perhaps
with multiplicity). The set 1'1o(U) C A contains the weights of U. This applies, in
particular, to the adjoint representation Ad. The nonzero weights of Ad are called
roots; one writes 1'1 for 1'1o(Ad)\{O}, with elements generically denoted by a.
The decomposition of 9c under Ad takes the form 9c = tBaE tl9a tB ie, where
each 9a is one-dimensional. Writing 9a = CEa for some nonzero vector E a , we
have
[X, Eal = -ia(X)Ea (1.141)
for X E t. It follows that if a E 1'1, then -a E 1'1, since ga = 9-a (where the
complex conjugation is the usual one on gc = 9 tB ig).
Given a choice of Cd' a root is called positive if (a, A) > 0 for all A E Cd (here
the inner product on 9 has been transferred to 9* in the usual way). The collection of
positive roots is called 1'1 +. A root lies either in 1'1+ or in - 1'1 + . Singular dominant
weights A have the property that (a, A) = 0 for some a E 1'1+; a weight is regular
iff (a, A) I- 0 for all roots a.
lt is not difficult to show from (1.141) and the Jacobi identity that 9c has a basis
{Hj, E a , E-a}j=I ..... r;aEtl+' normalized such that (Ea, E-a) = 1, satisfying

[Hj, Hd = 0;
[Hj , E±a] = ~iajE±a;
[Ea , E-al = -iajHj;
[Ea, Epl = Na.pEa+p (fJ I- -a), (1.142)
where a E 1'1+, fJ E 1'1, a j := a(Hj ), and the Na.p are constants that vanish iff
a + fJ is not a root (in which case E a +p is, of course, not defined).
The bijection in 1.10.5 is now implemented by the following fact:
218 III. Groups, Bundles, and Groupoids

Corollary 1.10.6. In the notation of 1.10.3, a Hilbert space 1{~w carrying an

irreducible representation Uy corresponding to a dominant weight y has a unit
vector Illy. unique up to a phase, on which

dUy(X)llI y = -iy(X)llIy (1.143)

for all X E t, whereas for all ex E ~+ one has

(1.144)

The unit vector Illy is called a highest weight vector; it is unique up to a phase.
It is easily inferred that

(1.145)

for all ex E ~ +. since the Lie brackets (1.142) imply that dU(E_a)llIy must either
be zero or a vector with weight y - ex :f. y.
One may now see the correspondence in 1.10.2 in a clearer light. Let J : lP1{y -*
g* be the momentum map for the G-action on lP1{y defined by Uy• given by (1.71);
this may be rewritten as

h(1/I) = i(llI, dUy(X)IlI), (1.146)

where the unit vector III is a lift of 1/1 E lP1{y to §1{~w.

Proposition 1.10.7. The coadjoint orbit Oy corresponding to an irreducible

representation Uy with highest weight vector Illy contains J1/Iy'
In fact, J : JIDUy(G)llI y -* Oy is a symplectomorphism when JIDUy(G)llI y
inherits the usual symplectic structure of lP1{~w (with n= 1), and Oy is endowed
with the (minus) Lie symplectic form (1.63).

Equations (1.146), (1.143), (1.144), and (1.145) imply that (J(1/Iy»(X) equals
y(X) for X E t and equals 0 for X E t.l. Hence J(1/Iy) is precisely the element
8(y) E g* discussed after the proof of 1.10.4. proving the first claim.
By (1.146), the stability group GJ(o/y) of J(1/Iy) consists of those x E G for
which (Uy(x)llI y, dUy(Y)Uy(x)llI y) = (Illy, dUy(Y)llIy) for all Y E g. Since Uy
is irreducible, this implies that Illy and Uy (x)llI y define the same element of lP1{y,
proving that G J(o/y) S; Go/ y • The opposite inclusion is trivial from the equivariance
of J, which can either be checked directly from (1.146), or may more abstractly
be inferred from 1.2.5, it having been realized from (1.72) (with r = 0) and 1.1.2
that J is a Poisson map on lP1{y. •

For general 11, one would have a factor 11, on the right-hand side of (1.146).
It is actually quite easy to give an explicit description of the Lie algebra gy
of G y := GI!(y)' From (1.61) (or the above proof), 0.142), and the previously
discussed fact that y(Ea) = 0 for all ex E ~ we infer that

(1.147)
 1 Lie Groups and Lie Algebras 219

where g~ := 9 n (ga $ g-a)' It follows that the dimension dim(Oy) of the orbit
through a dominant weight y is given by
dim(Oy) = dim(g) - dim(t) - 2 Card {a E I::!.. +I(y, a) = O}. (1.148)
It follows from Proposition 1.10.7 that PUy(G)Wy is a symplectic submanifold
of P1i~w; this is not necessarily true if CWy is replaced by an arbitrary one-
dimensional subspace of 1i';. However, the same statement as in 1.10.7 evidently
applies to any vector of the type Uy(w)W y , where W is a lift ofw E W = N(T)IT
to N (T) C G. For J maps all vectors of this type into the same coadjoint orbit.
In fact, Uy(w)W y has weight wy, showing that all weights wy, WE W, occur in
'l../hw
I ~y •

1.11 Berezin Quantization of Coadjoint Orbits

Coadjoint orbits of compact Lie groups are interesting partly because they lead to
coherent pure state quantizations indexed by a discrete set 1 3 Ii; cf. 11.1.5.1 and
II.1.3.3. We will use the label y to denote a dominant weight in Ad C t*, as well
as the corresponding element O(y) of the coadjoint orbit Oy C g*.
Theorem 1.11.1. Let G be a compact connected Lie group, and Oy an inte-
gral coadjoint orbit (cf 1.10.1), corresponding to a highest weight y E Ad' For
Ii = II k, kEN, define 1i1i := 1i~ili' i.e., the carrier space of the irreducible rep-
resentation Uy/Ii(G) with highest weight y Iii = kyo The map qli : Oy ~ P1i1i,
given by
qli(Co(x)y) := T'lln ..... P1th(Uy/Ii(X)W y/Ii), (1.149)
is well-defined and injective. Together with

Ji-Ii = dy/IiJi-L, (1.150)

where d}.. := dim(1i~W), this provides a pure state quantization ofOy (equipped
with minus the Lie symplectic structure) on 10 := liN.
One should note here that ky E Cd when y E Cd' since Weyl chambers are
convex cones. In what follows, Oy stands for (Oy)_; see the notation introduced
before (1.63).
The map qli is well-defined and injective by the equation Gky = G y plus the
argument on stability groups used in the proof of 1.10.7. In fact, if we define
JIi: P1i1i ~ g* by (1.71), with Ii = 11k (equivalently, by (1.146) with the right-
hand side divided by k), it follows from 1.10.7 that JIi takes values in Oy' and is a
left inverse of qli.
We start from the fact that the Haar measure on G (with total mass 1) pushes
forward to the Liouville measure derived from the Lie symplectic structure under
the canonical projection G ~ Oy ~ GIG y . Using the invariance of the Haar
measure and the unitarity of Uky, we then have

1
Oy
dJi-da) p(PIi, ali)f(a) = ( dx I(Wky, Uky (X)Wky)1 2 fy(yx)
10
(1.151)
220 III. Groups, Bundles, and Groupoids

for all f E C(Oy), where fy = fC->-G/G y f E C(G) is a right-Gy-invariant func-

tion,andyissuchthatfG->-G/G/Y) = P E Oy. Choosing f = 1, the orthogonality
relations (1.116) for compact groups and (1.150) then imply 11.(1.8).
Equation 11.(1.10), even without the limit, follows from 1.10.7. To prove 11.(1.9)
we need a lemma.

Lemma 1.11.2. Let Yi be dominant weights with highest weight representations

and vectors UYi and\llyi' respectively (i = 1,2). Thenfor each x E G one has

(WYl' Uyl(x)W yl )' (W y2 , UYz(x)W Yz ) = (W yl +y2 , UYI+Yz(X)WYI+Y2)' (1.152)

This is immediate from 1.10.6 and the connectedness of G.
This lemma implies that
•
(Wky, Uky(X)Wky) = (W y , Uy(x)Wyi. (1.153)
Using (1.151), we can write the left-hand side ofIl.(1.9) as

lim ( dI-LL(a) p(Ph, ah)f(a) = lim ( dl-Lk(X)fy(Yx),

h->-O JOy k->-oo JG
where I-Lk is a probability measure on G defined by
dl-Lk(X) := dky dxl(W y , Uy(x)W y )1 2k . (1.154)
It is obvious that each I-Lk is right-Gy-invariant. It follows from (1.144), (1.145),
and the fact that the exponential map is surjective for compact Lie groups, that
1(\11 y, U y (x) Wy ) I, which is evidently ::::: 1, equals 1 iff x E G y. Hence for large k
the support of I-Lk is increasingly concentrated on G y. This suggests that

lim I-Lk(f) = ( dh f(h) (1.155)

k->-oo JG y
for all f E C(G), where dh is the normalized Haar measure on G y • This is
confirmed by more detailed analysis (cf. the proof of 1.11.4 below). For the right-
Gy-invariant function fy E C(G)G y one therefore obtains

lim I-Lk(fy) = fy(e).

k->-oo
This proves 11.( 1.9), which finishes the proof of 1.11.1. o
The Berezin quantization Qg associated with the pure state quantization in
1.11.1 (cf. 11.1.3.4) is defined on §to := COO(Oy)' By 11.(1.16), one has

Qf/k(f) = dky L dx fy (X)[Uky (X)Wky]; (1.156)

this is an element of Qt 1/ k = Vltdky (C). The most important property of Qg is its

G-equivariance. For x E G we write
(1.157)
 1 Lie Groups and Lie Algebras 221

where A E milk, and for f E mO we put

a2(f) := Co(x- 1)* f. (1.158)
Proposition 1.11.3. For all kEN, x E G, and f E LOO(Oy) one has
Qflk(a~(f» = a~/k(Qf/k(f»· (1.159)
This is immediate from (1.156), (1.158), the fact that (Co(y)* f)y = L;/y, the
right-invariance of the Haar measure, and (1.157). •
Theorem 1.11.4. The Berezin quantization (1.156), defined on the space
COO(Oy, 1R), is strict.
Recall that Oy := (Oy)_. Rieffel's condition 11.1.1.1.1 follows from Theorem
1.11.1 and Proposition 11.1.3.6. The completeness condition II. 1.1. 1.4 is an easy
consequence of Schur's lemma and the irreducibility of Uky.
We will now prove von Neumann's condition II. 1. 1. 1.2 and Dirac's condition
II.l.l.i.3. We pick a unit vector <l>k in each 1t~~ and use the invariance of the Haar
measure and (1.116) to write (using the notation of the proof of 1.11.1)
(<I>k' (Qf/k(f)Qf/k(g) - Qf1k(fg»<I>k)

= dky fa dx fy(X)(<I>k, Uky(x)Wky)Ik(x), (1.160)

where

Ik(x) := dky L dy (Wky, Uky(y)Wky)F;(y), (1.161)

F;(y):= (Uky(Xy)Wky, <l>k)[gy(XY) - gy(x)]. (1.162)

In the notation used after (1.151), the function F;
on G corresponds to a function
F onO y •
X

Using (1.153), we can write (Wky, Uky(y)Wky) = exp(-nSy(y», where

Sy (y) := - log(wy, U y (y) Wy) (in view of the exponentiation, the choice of the
branch cut of the logarithm is irrelevant). The function Sy is right-Gy-invariant;
we denote the corresponding function on G / G y by S. We identify G / G y with
0Y' so that the coset [G y] E G / G y is identified with y E Oy.
Putting S:(y) := -log I(Wy, Uy (y)Wy )1, the absolute value of exp( -nS) is
exp( -nS+). As in the argument preceding (1.155), we see that S+ takes values
in [0,00] and assumes its unique absolute minimum 0 at y. Since F:
in (1.161)
is bounded, a standard argument implies that to O(exp(-n» we may replace the
integration over G / G y by one over any neighborhood of y.
We identify Ty Oy with 9/ 9y, and use complex coordinates {Za, Za }ae6t' where
l:!..~ consists of those positive roots for which (y, a) =1= O. By the definition of
a highest weight, this implies that (y, a) > 0 for all a E l:!..~. The coordinates
(Za, Za) correspond to the point in Oy given by

Co [ Exp (i a~t (ZaEa - ZaE-a») ] y.

222 III. Groups, Bundles, and Groupoids

A simple computation, using (1.142) and 1.10.6, leads to

S(Za, Za) = E (y, a)ZaZa + O(ld). (1.163)
aet.t
Hence to O(exp(-k» we may approximate ft(x) by

d ky 1 (n
g/gy aet.t
dzadza ) J(
2
1T
- ) -kLaE,,+(y,a)ZaZa FX(
Za, Za e Y
-)
Za, Za ,

where J is a Jacobian, and F:

has been extended to g/ gy by, say, the exponential
map. If we omit the factor [. , .J in (1.162), the integral (1.161) can be evaluated,
using the orthogonality relations (1.116). On the other hand, we can compute the
above integral to lowest order in the steepest descent approximation; this avoids
the need to compute J (0). Comparing the results computes the prefactor in the
steepest descent approximation as unity. As a by-product we obtain the asymptotic
expression for k ~ 00

TIaet.+(y, a) I d' (0 )
d rv y k'i 1m y (1.164)
ky J(O) ,
where dim(Oy) is given by (1.148). (Comparison with the Weyl dimension formula
then yields J(O) = TIaet.+(a, 8), where 8 := Laet.+ a.) !
Thus the steepest descent approximation to the above integral, and therefore to
(1.161), reads
N 1
ft(x) = ~ TfD1(J r)(O) + O(k- N - 1), (1.165)

where, abbreviating aa := a/aZa and 8 a := 8/8z a , we have put

(1.166)

Substituting this expansion in (1.160) we see that

(<I>ko (Qf/k(f) 0 Qf/k(g) - Qf/k(fg»<I>k) = 0(1/ k). (1.167)
To analyze the remainder of 0(1/ k) we note that the Ith term in the expansion
leads to an x-integrand in (1.160) of the form

(Uky(X- 1)<I>ko Wky )(Uky(X- 1 )<I>ko W(l,»!a I2 [j' gy(x),

where Ij ::: I and the '11(/,) are given by the action of products of dUky(Ea) and
dUky(E_a) on Wky ' The important point is now that the orthogonality relations
(1.116) (applied to the x-integration) then imply that the O(k- N -I) term is bounded
by ClI<I>k 112/ k N + 1 for some constant C. Hence 1I.(1.2) follows by 1I.(2.77).
To prove 11.(1.3) we need the I = 1 term in (1.165). We substitute (1.166), and
perform some partial integrations in the remaining x -integral (using the invariance
of the Haarmeasure). We abbreviate A := (<I>ko Uky(X)Wky); then (1.144) implies
 I Lie Groups and Lie Algebras 223

that aaA and BaA vanish at Za = Za = O. Terms of the form aaBaA (or A) drop out
in the commutator, as do contributions from] (whose first derivatives at 0 already
vanish identically). What remains is
(<I>b ik[Qf/k(f), Qf/k(g)] - Qf/k({J, g}-)<I>k) = O(1lk), (1.168)
where, in the realization of f, gas Gy-invariant functions f y , gy on G,
. ~ 1 L L
{Jy, gy}± := ±I ~ -(-);a fy Lagy· (1.169)
aE8 y y, ex

Here the left-invariant vector fields ;~a on G are defined as in (1.37), the element
E±a of ge having been expressed in terms of elements of g. Also, ~y is ~~ u ~~,
i.e., the set of all roots ex for which (y, ex) =I- O.
To finish the proof, we remark that (1.169) is precisely the Lie-Poisson bracket
on Oy; this may be verified at the point y E Oy (or e E G) by direct computation
from (1.3), from which the general statement follows by the G-invariance of the
Poisson structure.
It is manifest that the right-hand side of (1.169) is left-G-invariant if fy and
gy are; its right-Gy-invariance is not so obvious. The latter may be verified at the
infinitesimal level from (1.147), (1.142), and the fact that for fJ =I- ±ex one has
N_ a -/3,/3 = -Na ./3. (1.170)
This follows from the Ad(g)- invariance of the inner product on ge, combined with
the normalization of the Ea. Invariance of (1.169) under gy implies invariance
under G y, which is connected.
The higher-order terms in (1.168) are dealt with as in the above proof oflI.(1.2).
This proves IL(1.3), finishing the proof of Theorem 1.11.4. •
It is possible to regard ah := qh(a), defined in (1.149) for Ii = 11k, as a state
an on the group algebra C*(G) by
(1.171)
The following result is analogous to IL(2.167).
Proposition 1.11.5. With Q~ defined in 1.9.1,for all a E Oy and f E C~(g*)
one has
lim an(Q~(f» = f(a) (1.172)
n->-O

along the sequence Ii = 11k, kEN.

This follows from a straightforward calculation. One starts by using (1.124),
(1.125), and rescaling X f-+ XI k. The k-dependence is firstly in ](Xj k)K(XI k),
which goes to 1 for k ~ 00. Secondly, one uses (1.153) and subsequently
lim (Illy, Uy(Exp(Xjk»llIy)k = il/ly·dUy(X)l/Iy ).
k->-oo

This can be computed by (1.143) and (1.144). The result then follows from the
well-known representation of the delta function as an oscillatory integral. •
224 III. Groups, Bundles, and Groupoids

Finally, we remark that the results in this section have an obvious yet somewhat
cumbersome generalization: If the orbit Oy is not integral, but such that Oy/c is
integral for some c E lR. \ {O}, we can construct a strict quantization for the values
n= c/k,k E N.

2 Internal Symmetries and External Gauge Fields

2.1 Bundles
Many constructions where some form of symmetry plays a role, and in partic-
ular the mathematical description of gauge field theories, involve the notion of
a (smooth locally trivial fiber) bundle. We have already encountered the tangent
bundle and the cotangent bundle of a manifold; here is a general definition.

Definition 2.1.1. A bundle B(Q, F, r) consists of manifolds B (the total space),

Q (the base), F (the typical fiber), and a smooth surjection r : P ~ Q with
the following property: Each q E Q has a neighborhood Na such that there is
a diffeomorphism 1{!a : r-'(Na ) ~ Na x F C Q x F for which r = rQ 0 1{!a
(where rQ : Q x F ~ Q is the projection onto the first factor).

The maps 1{!a are called local trivializations. To avoid cumbersome expressions
we shall often say "B ~ Q x F (locally)", omitting reference to N. Similarly, we
then loosely write "1{! : B ~ Q x F (locally)". We factorize 1{!a = (r, 1{!:) so
that 1{!: restricted to r-'(q) provides a diffeomorphism between the latter and the
typical fiber F. Each subset r -, (q) is called a fiber of B. One may think of B as
Q with a copy of F attached at each point.
Throughout this chapter Q will be physically interpreted as the space on which
a particle moves, or perhaps as some more general configuration space.
Two bundles Bj(Qj, Fj , rj) (i = 1,2) are said to be isomorphic if there is a
diffeomorphism 1{! : B, ~ B2 that preserves fibers. Such a bundle isomorphism
defines a diffeomorphism of the base spaces and typical fibers in question. The
bundle is said to be trivial if there is a bundle isomorphism 1{! : P ~ Q x F. Any
bundle over a contractible base is trivial.
By definition, a section of B is a map s : Q ~ B satisfying r 0 s = id. It can
be shown that (Borel) measurable sections always exist, whereas the existence of
smooth sections is not guaranteed (they certainly exist if B is trivial). However,
one can always choose smooth local sections Sa : Na ~ B. In the spirit of the
paragraph before the last, we may say "s : Q ~ B (locally)" when s is actually
defined on some N c Q.

Definition 2.1.2. Given two bundles B, and B2 over the same base Q, with
projections r" r2 and typical fibers F, and F2 , respectively, the fiber product of
B, and B2 is
(2.1)
 2 Internal Symmetries and External Gauge Fields 225

with manifold structure inherited from the Cartesian product. This may be regarded
as a bundle over Q with projection rex, y) = rl(x) = r2(Y) and typical fiber
FI x F2.
Let B( Q, F, r) be a bundle over Q and let f : M ~ Q be a smooth map from
some manifold M to Q. Then the pullback bundle

f*B := B *Q M = {(x, y) E B x M I rex) = fey)} (2.2)

is a bundle over M under projection r(2) onto the second variable and typical fiber
F.
Hence BI *Q B2 can be equipped with a bundle structure in three different ways:
It is a bundle over Q as explained above, it is a bundle rtB2 over BI with typical
fiber F2 under the projection r(l) onto the first variable, and finally it is a bundle
r;BI over B2 with typical fiber FI under the projection r(2) onto the second variable.
One can specialize the bundle structure. For example, in a vector bundle each
fiber is a (topological) vector space (where the linear operations are smooth with
respect to the ambient manifold structure), and the local trivializations respect the
linear structure in the obvious sense. Clearly, T* Q and T Q are vector bundles.
Even when it is nontrivial, a vector bundle always admits a smooth global section,
namely the zero section so(q) := O. We will generically denote vector bundles by
the letter V, unless the typical fiber is a Hilbert space, in which case we write H,
and speak of a Hilbert bundle.
When the Bi in 2.1.2 are both vector bundles Vi (with finite-dimensional or
Hilbert fibers), one may form two different vector bundles over Q from VI *Q V2
by declaring the typical fiber to be either the tensor product VI ® V2 or the direct
sum VI EB V2. One accordingly writes VI ® V2 or VI EB V2. One can also form the
dual bundle V* of a vector bundle V, whose typical fiber is the dual V* of V, and
whose local trivializations are dual to those of V.
Here is the "mother" of all bundles in which group actions playa role.

Definition 2.1.3. A principal bundle P(Q, H, r) is a bundle for which the typical
fiber is a Lie group H (the structure group) with smooth (left) action R on P such
that Q = PI H, and 1/Ia 0 Rh = Rf 0 1/Ia, where the action Rf : Q x H ~ Q x H
on the right-hand side stands for Rf(q, k) = (q, kh- I ).

To stress the role of Hand Q, one may speak of a principal H -bundle over Q
for clarity. It follows that the H -action Rh must be free, and that r(Rh(x» = rex).
In agreement with the above, we will write Rh as x 1-+ xh- I for x E P. In
contrast with a vector bundle, it can be shown that a principal bundle admits
smooth global sections iff it is trivial. In a trivial bundle P = Q x H one obviously
has Rh(q, k) = (q, kh- I ).
In a principal bundle a given local trivialization 1/Is is equivalent to a smooth
local sections: Givens one can put 1/Is(s(q» = (q, e) and subsequently extend 1/Is
by H -equivariance; that is,

1/Is(s(q)h) = (q, h). (2.3)

226 III. Groups, Bundles, and Groupoids

Conversely, given 1/1 = 1/Is one defines s(q) := 1/I-I(q, e). If various local sections
Sa are involved, we will write 1/Ia for 1/Isa.
If Q is covered by open sets of the type Na and q E Na n N fJ , it must be that
(2.4)

(no sum over ex), where the smooth maps gafJ : Na nNfJ ~ H are called transition
functions. More generally, two different systems oflocal trivializations are related
in this way.
In an interesting special case one takes P to be the universal covering space Q of
Q, so that H = 7r1 (Q) is the first homotopy group of Q (regarding discrete groups
as zero-dimensional Lie groups). For another example the reader could now skip
ahead to 2.7.
Definition 2.1.4. Given a principal H -bundle P over Q and a smooth H -action
L on some manifold M, the associated bundle M = P XH M is (P x M)/H,
where the H-action on P x M is given by h : (x, m) 1-+ (xh- I , Lh(m». This is a
bundle over Q with typical fiber M and projection rM-> M ([x , m]fI) = rex), which
is well-defined in being independent of the representative (x, m) E P x M in the
equivalence class [x, m]fI E (P x M)/ H.
The following result will be used on many occasions.
Proposition 2.1.5. A section W(L) : Q ~ M of a bundle M associated to a
principal bundle P( Q, H, r) may alternatively be represented:
• As a map WL : P ~ M that is H -equivariant in that
(2.5)
This is related to W(L) by W(L)(r(x» = [x, WL(X)]H, which is independent of
the choice of x E r - l o r (x) because of (2.5 ).
• Given a section s : Q ~ P, as a map WsL : Q ~ M, in terms of which
W(L)(q) = [seq), W:(q)]lI;
W:(q) = WL(s(q»;
WL(X) = L(hs(x)WsL(r(x», (2.6)
where hs(x) is determined by xhs(x) = s(r(x».
This follows directly from the definitions involved. Note that hs(xk) =
k- I h s (x), ensuring the consistency of the relation between Wf and WL • •

The space of smooth compactly supported sections of a vector bundle V (where

compact support is defined with reference to the zero vector in each fiber) is denoted
by reV); when M is a vector space V, the symbol r(P Xli V) will specifically
refer to the first realization discussed above. The second realization will be called
rs(p XH V).
We will see that interesting classical phase spaces arise by taking M to be a
coadjoint orbit of G in this construction. Alternatively, in classical as well as
 2 Internal Symmetries and External Gauge Fields 227

quantum mechanics one encounters the case where M is a linear space V, carrying
a linear H -action L; in that case one speaks of an associated vector bundle. The
case relevant to quantum mechanics is that in which V is a Hilbert space 1i x
carrying a representation Ux(H); cf. 2.9.
A (local) trivialization 1/Ia = ('r, 1/1;;) of P leads to a (local) trivialization 1/1/: :
M -+ Q x M of an associated bundle M by putting
(2.7)

This is well-defined, since 1/I;;(xh) = 1/I;;(x)h. Conversely, (1/I/:)-I(q, m) =

[sa(q), m]H, in terms of the (local) section associated with 1/Ia; cf. 2.3.
If two vector bundles V I, V2 are both associated to a principal H -bundle P over
Q by H-actions L j , then VI ® V2 and VI E!) V2 are associated to P by the actions
L I ® L2 and LIE!) L2, respectively. The dual bundle V* of an associated vector
bundle V defined by an H -action L * is defined by the dual H -action q := (Lh-I)*
on V*.

2.2 Connections
In preparation for the definition of a connection on a principal bundle, note that
the tangent bundle TP of a principal bundle TP has a natural subbundle

VP:= {v E TPI 't'*v = OJ. (2.8)

Elements of V P are called vertical vectors, and the linear space Vx P := Tx P n V P

is called the vertical tangent space at x. It is easily seen that VP is stable under
the lifted H -action R*. Indeed, V P is spanned by vector fields of the type ~ x given
by the H-action on P; cf. (1.19). It is customary to define
/:1 . /:p
(2.9)
'iX·= -'iX'

referred to as a fundamental vector field. The vector ~1:(x) is evidently tangent

to the curve x(t) = xExp(tX). One easily shows that

(Rh)*~1: = ~{d(h)X· (2.10)

A lift of X E Ty(x)Qtox E Pisanelementlx(X) E TxX for which t'*lx(X) = X;

a lift is evidently unique up to the addition of vertical vectors. Each lift Hx of TT(X) Q
to x E P satisfies Hx P E!) Vx P = Tx P, but there is no canonical choice of such a
complement to VxP.
Definition 2.2.1. A connection on a principal H -bundle is a smooth assignment
x ~ HxP C TxP such that HxP E!) VxP = TxP and
(2.11)

Elements of Hx are called horizontal vectors, and each Hx is called the hor-
izontal subspace of Tx P. The collection of all Hx is the horizontal subbundle
HP of TP. The horizontal lift ix(X) of X E TT(X)Q to TxP is the unique vector
228 III. Groups, Bundles, and Groupoids

in HxP satisfying ,*(ex(X» = X. From (2.11) we infer

eRh(x)(X) = (Rh)*lx(X). (2.12)

Similarly, a horizontallifU(q(·» of a curve q(.) in Q is a curve x(·) in P for which

,(x(t» = q(t) and x(t) E Hx(r)P for all t. Such a lift is unique if one specifies
x = x(O) (at which ,(x) = q(O». The parallel transport of x E P to Cl(q(t»
along a C l curve q(.) in Q (with ,(x) = q(O» is by definition x(t), where x(·) is
the horizontal lift of q(.) through x.
These notions may be transferred to any bundle M associated to P. The horizontal
lift of a C l curve q(.) in Q through [x, m]H E M (with ,(x) = q(O» is the curve
[x(·), m]H, where xO is the horizontal lift of q(.) through x. Similarly, the parallel
transport of [x, m]H to 'M~Q(q(t» is the point

Pq-.+q(r)[x, m]H := [x(t), m]H, (2.13)

where x(t) is as defined above.

Proposition 2.2.2. There is a bijective correspondence between connections on P

and smooth sections A of A I(p) ® fJ (i.e., fields offJ-valued i-forms on P), called
connection I-forms, satisfying

A(~{) = X, (2.14)
RZA = Ad(h)A (2.15)

for all X E fJ and h E H.

Given the Hx, one defines A by (2.14) and Ax(X) = 0 for all X E HxP; equation
(2.15) follows from (2.14), (2.10), and (2.11). Given A, one defines HxP as the
subspace of Tx P annihilated by Ax; (2.11) follows from (2.15). •

In a local trivialization 1/1 : P ~ Q x H (locally) associated to a section

s : Q ~ P (locally; cf. (2.3» we can write

(2.16)

where s* A E A I(Q) ® fJ and etC is defined in (1.41). This expression is enforced

by 2.1.3, (2.14), (2.15), and (1.53). Connections on a trivial principal bundle Q x H
are, then, necessarily of the above form.
A connection I-form A determines a projection 'v :
TP ~ VP (mapping TxP
onto VxP) by

'v(X) = ~{(X). (2.17)

The complementary projection 'h : TP ~ HP is given by

'h(X) = X - ~{(X). (2.18)

The curvature of A is an f)-valued 2-form F on P, defined by

(2.19)
 2 Internal Symmetries and External Gauge Fields 229

Some rearrangements lead to the expression

F = dA + [A, A). (2.20)
Writing s* A = Ii A~dqJL and s*F = Ii F~vdqJLdqV (in terms of coordinates qJL on
Q and a basis {Ii) of ~, omitting explicit reference to the section s) one therefore
has the physicists' formula
j _ j
FJLV - aJLA v - avAJL
j
+ CjkAJLA
j j k
v' (2.21)
A most important aspect of a connection on P is that it defines a certain first -order
differential operator on all vector bundles associated to P. We write V := P x H V
(defined relative to some H -action L on V). Recall that the space of all compactly
supported smooth sections of V is denoted by r(V).
As in 11.3.1.1, a covariant derivative on a vector bundle V( Q, V, r) is a linear
association ~ ~ Y'~, where ~ E r(TQ) and Y'~ : reV) ~ rev) satisfies Y'n =
fY'~ and Y'~ (f\ll) = Hf)\II + fY'~ \II for all f E C OO ( Q) and \II E r(V). One
may look at a covariant derivative as a map Y' : reV) ~ r(A I (Q) ® V), so that
Y'~ IJI = Y'\II(~).
We can (locally) choose a moving frame, that is, a collection of sections Sj :
Q ~ V, such that {Si(q)}j=I, .... dim(V) is a basis of C1(q) for all q (in some open
subset of Q). A function 1/1 : Q ~ lRdim(V) then defines a section IJI : q ~ 1/I i Sj(q)
of V. With obvious abuse of notation we may then write
(2.22)

where A is a matrix-valued I-form on Q, defined by the property Y'Sj = Afsj;

it evidently depends on the moving frame. This dependence is controlled by the
transformation property
(2.23)
where A is the I-form determined by a moving frame Sj(q) := sj(q)M-1(q)f,
where M : Q ~ GL(lRdim(V» (locally). This property, which is immediate from
the definition of A, guarantees that Y' ~ = MY' 1/1, where ~ = M 1/1 defines the
same section \II as 1/1 does, but in terms of the frame S.
By slight abuse of terminology (cf. the case of a principal bundle) one often refers
to A as a connection, and instead of saying that there is a covariant derivative on
V one says that V has a connection.
Proposition 2.2.3. A connection (with associated I-form A) on P(Q, H, r) de-
fines a covariant derivative Y' A on any vector bundle V(Q, H, r) associated to P
by
(2.24)

when the limit exists; here q(.) is a curve through q with tangent vector ~q. and
the limit is independent of the specific choice of the curve.
In terms of the realization \ilL (cf 2.1.5) one then has (with abuse of notation)
Y'tIJlL(x) := ~\(~T(x»\IIL(X). (2.25)
230 III. Groups, Bundles, and Groupoids

In terms of the realization Wf the covariant derivative is

(2.26)

where A := s* A.

Here dL(X) := dL(Exp(tX»/dtlt = 0 for X E I); hence dL(s*A) E

A I(Q) ® dL(I) (locally). Note that the right-hand side of (2.24) is well-defined
as the difference between two elements of a vector space, in that both terms in the
numeratorlie in the same fiber r,,-: Q(q) because of the parallel transport operation
(2.13) involved. The properties of a covariant derivative then easily follow from
the ordinary Leibniz rule.
We take the horizontalliftx(·) of q(.) in V through x, and note that (2. 13) implies
that

(2.27)

from which the equivalence of (2.24) and (2.25) is immediate. Note that (2.25) is
well-defined in that VtWL(ph-l) = L(h)VtWL(p) because of (2.12).
To derive (2.26), which is obviously a special case of (2.22), we notice that
vtwf(q) = is(q)(X)WL(S(q», use (2.18) to write is(q)(X) = s*(X) - ~{(s.(X))'
and then use the definition of ~ f and the equivariance of WL • •

It follows most easily from (2.26) and a well-known identity for the exterior
derivative d that in a given trivialization the curvature s*F of A is related to the
covariant derivative by (cf. 11.(3.20»

(2.28)

In complete analogy to the special case of an affine connection (see 11.3.1),

given a covariant derivative V on some vector bundle V(Q, V, r) one may define
the horizontal lift i of a vector or of a curve. In terms of the matrix-valued 1-
form A appearing in (2.22) and the identification r-I(q) ::::::: lRdim(V) given by a
moving frame (that is, the components Vi are defined by v =: ViSi(q», one has
iv(X) = (-v j A~(X), X).
The following construction will not be used until 3.1 0, but logically fits in at this
point. We follow the notation of Definition 2.1.2, except that the general bundle B
is now a vector bundle V.

Proposition 2.2.4. Let f*V be the vector bundle defined by (2.2). Then a covariant
derivative on V pulls back to a covariant derivative on f*V.

Sections of f*V have the form W(x) = (WI (x), x), where WI(X) E r-I(f(x».
Choose a (local) moving frame {Si} on V, with associated connection A, and define
l/II : M -+ lRdim(V) (locally) by WI(x) = 1/Ii(x)s;{f(x». The desired covariant
derivative is then given by VW(x) = (VW1(x), x), where

(2.29)
 2 Internal Symmetries and External Gauge Fields 231

This is well-defined (cf. (2.22) and subsequent text): A change of moving frame
and corresponding change in A and 1/11 does not affect the total expression, because
of properties like f* M = M 0 / . •

The corresponding horizontal lift .e in f*V is given by

(2.30)
seen as an element of Tv V $ TxM :J 1(v,x)(V *Q M); one immediately sees that
(2.30) is indeed tangent to V *Q M.

2.3 Cotangent Bundle Reduction

In preparation for the main theorem of this section we will first look at connections
more specifically from the point of view of the cotangent bundle T*P and the
momentum map. We start with a more general statement.
Lemma 2.3.1. The pullback 0/ an H -action on a manifold M to the cotangent
bundle T* M is strongly Hamiltonian, with momentum map J : T* M -+ ~~ given
by
Jx(er) = er(h). (2.31)
Note that in order to obtain a (left) action on T* M one has to put h ~ RZ- 1 , if
the H-action on M is denoted by R. The claim most easily follows from 1.(2.24),
(1.19), and (1.7) in canonical coordinates, in which
Jx(p, q) = Pi~~(q). (2.32)
Here ~~ is defined by h(q) := ~~(q)a/aqi.
The annihilator of VP (cf. (2.8» in T*P is
•
VOp := (er E T*P I er(X) = 0 \IX E VP}. (2.33)
Elements of VOp are called horizontal I-forms, and V~P is the horizontal cotan-
gent space at x. Let R* be the pullback of the H -action on P to T* P (if we speak
of "the H -action R* on T* P" we mean the left action in question, i.e., we silently
incorporate the shift h ~ h- I , as in the comment preceding (2.32». By the above
lemma this action is strongly Hamiltonian, with momentum map J : T*P -+ ~~.
From (2.9) and (2.31) we conclude that
VOp = J-1(0). (2.34)

The projection dual to Th : TP -+ HP is 'h* : T*P -+ VOp (that is, ,:(w) =

w From (2.18), (2.31), (2.14), and (2.9) we see that
0 Th)'

'h*(W) = W + J(w) 0 A; (2.35)

here J(w) E ~* hits the ~-part of A E A I(p) ® ~.
In a (local) trivialization we have T* P ~ T* Q x T* H (locally). We then put
T* H ~ ~* x H in either the left or the right trivialization (see 1.4), and choose
232 III. Groups, Bundles, and Groupoids

coordinates e j (i = 1, ... , dim (H» on f) * such that e = e jui in terms of a basis {ui }
off)* dual to a basis {1j} off). Choosing also canonical coordinates (PIJ-, qIJ-) on T* Q
(locally) we have the quadruple (p, q, h, ekR representing pIJ-dqIJ- + ejeLR(h).
From 2.1.3, (1.49), (1.51), (1.55), and (1.57) we obtain
RZ- 1 (p, q, e, h)L = (p, q, Co(k)O, hk- 1 )L, (2.36)
J(p, q, e, h)L = -e, (2.37)
RZ-1(p, q, e, h)R = (p, q, e, hk-I)R, (2.38)
J(p, q, e, h)R = -Co(h-1W. (2.39)
Hence from (2.35), (2.16), (2.36) etc., and (1.42) one derives the coordinate
expression of rh*:
rh*(pIJ-, qIJ-, e j , h)L = (PIJ- - Ad(h- I )'A~(qWk. qIJ-, 0, hh; (2.40)
rh*(pIJ-, qIJ-, e j, h)R = (PIJ- - A~(qWj. qIJ-, 0, hk (2.41 )
With (2.37) and (2.39) this confirms (2.34). In this trivialization the canonical
Poisson bracket on T*P (cf. I.(2.24» reads simply
{f, g} = {f, g} T' Q + {f, g} T' H , (2.42)
where the first term is given by I.(2.24) and the second by (1.54) (with r = 0 and
G replaced by H).
The aim of the following considerations is to factorize f)* from T* P in an intrinsic
way, so as to facilitate the study of the momentum map. Recall Definition 2.1.2.
Definition 2.3.2. Let P be a principal H -bundle over Q. Then H acts on PH: =
P *Q T Q and P~ := P *Q T* Q through its action on P (combined with the trivial
action on T(*)Q); e.g., on P~ one has h : (x, a) 1-+ (xh- I , a). With this action,
and projection onto the second factor, PH = rTQ-'>QP and P~ = rT'Q-'>QP are
principal H -bundles over T Q and T* Q, respectively.

Lemma 2.3.3. Regarding all spaces in question as bundles over P with the obvious
projections, there are natural H -equivariant isomorphisms between

• P x f) and VP, where H acts on the former by pA := R x Ad and on the latter

by R*;
• P~ and V OP, where H acts on the former as specified in 2.3.2 and on the latter
by (R*)-I.

The first isomorphism is given by (x, X) B- ~{(x). The desired equivariance

follows from (2.10). Given (x, a) E P~, the map r; : T,*(x)Q -+ Tx*P produces an
element r;a E Vxop. Given WE VxOP one defines a w E T;(x)Q by
aw(X) := w(lAX»; (2.43)
the nonuniqueness of the lift does not matter, since w annihilates vertical vec-
tors. One easily checks that these maps are diffeomorphisms, and are each other's
inverse. The H -equivariance follows from the property r 0 Rh = r. •
 2 Internal Symmetries and External Gauge Fields 233

Further to 2.3.3, one has

Lemma 2.3.4. For any given connection, HP is H -equivariantly isomorphic to
PH as a bundle over P.
The isomorphism is obtained by letting (x, v) E HP correspond to lAv). The
equivariance follows from (Rh)*(,,(V) = lRh(x)(v) (see (2.11». •
Let H act on P x ~* by pC := R x Co; that is,
PK(x, a, 0) = (xh- I , a, Co(h)O). (2.44)
Proposition 2.3.5. A connection on P leads to H -equivariant isomorphisms
TP:::::: PH ED (P x ~); (2.45)
T*P :::::: P~ ED (P x ~*). (2.46)
Here all objects are seen as vector bundles over P, on which the associated
diffeomorphism is trivial.
All claims are immediate from 2.3.2, 2.3.3, and 2.3.4.
Corollary 2.3.6. Regarding the objects involved as manifolds (rather than
•
bundles), (2.46) is an H -equivariant diffeomorphism
T*P:::::: P~ x ~*, (2.47)
and similarly for (2.45). The momentum map of the H -action (2.44) on P~ x ~* is
J(x, a, 0) = -0. (2.48)
To prove (2.48) one adopts a local trivialization, compares (2.44) with (2.36),
and uses (2.37). •
We shall need the explicit fOnD of the isomorphism 1{1A : T* P -+ P~ x ~*,
which follows from (2.35); one has
1{IA(W) = ('PP-4p(w), aw+J(w)oA, -J(w», (2.49)
with a ... given by (2.43). The inverse is
1{IA 1(X, a, 0) = ';.p-4TQ(a) + 00 Ax. (2.50)
The canonical Poisson structure 1.(2.24) is invariant under the H -action on
T*P (any diffeomorphism of a manifold M pulled back to T* M is a Poisson map).
Consequently, one obtains a Poisson structure on (T* P)1H. The associated Poisson
algebra is the classical algebra of observables of a particle moving on Q = PI H
with all possible "classical" charges 0 C ~*. There is a natural isomorphism
between COO«T*P)I H) and COO(T*p)H, the space of smooth functions on T*P
that are invariant under the H -action R*.
Theorem 2.3.7. Each symplectic leaf of the Poisson manifold (T*P)I H
(inheriting the canonical Poisson structure ofT*P) is of the form
(rp)" := J-1(0)1 H, (2.51)
234 III. Groups, Bundles, and Groupoids

where 0 C ~* is a coadjoint orbit of H.

Any connection on P leads to a diffeomorphism

(T*P)/H ~ P~ XH ~*, (2.52)

seen as the vector bundle over T* Q associated to the principal H -bundle P~ (cf
2.3.2) by the coadjoint representation on ~*. Consequently,for each 0 and a given
connection A one obtains a diffeomorphism

1//1: (PP)o ~ P~ Xli 0, (2.53)

where (PP)o is a bundle over T* Q (with typical fiber 0) associated to P~.
The first claim is clear from (2.38), (2.39), 1.4.4, and (2.42). The second and
third follow from the first and 2.3.6. The map 1/1;( is simply the reduction of 1/1A in
(2.49); that is, the latter quotients to a map [1/1 A]lI from (T*P)/ H to P~ Xli, which
provides the diffeomorphism in (2.52), and 1/1;( is then the restriction of [1/1A]lI to
the symplectic leaf T* pO in (T* P) / H . •

The tilde on (PPP signifies that the quotient in (2.51) is taken with the entire
group H, rather than with its identity component HO; cf. IY.1.6.
From Corollary 1.2.6.10 we now have
Corollary 2.3.8. Up to equivalence, each irreducible representation of the Pois-
son algebra COO«T*P),LH, JR) (in the sense ofl.2.6.6) is realized on a symplectic
manifold of the type (T*PP, or, equivalently, of the type P~ XH 0 (ora covering
space thereof).

The space P~ x H ~* becomes a Poisson manifold by declaring the diffeomor-

phism (2.52) to be a Poisson map. Similarly:

Definition 2.3.9. The symplectic space P~ XH-2, equipped with the2oisson

structure that it inherits as a symplectic leaf of(T*P)o, is denoted by (T*P)~.
Equivalently, the Poisson structure on (PP)~ could be defined by stipulating
that the diffeomorphism 1/1;( in (2.53) be a Poisson map.
The corresponding Poisson bracket is easily computed in a local trivialization
of the associated bundles in question corresponding to a local trivialization of P
(see (2.7». Both (T*P)/ H and P~ x H ~* trivialize to T* Q x ~* (locally), and
from (2.49) with either (2.37) and (2.40), or (2.39) and (2.41) (the difference
between these expressions fades when passing to the quotient by H and taking
local trivializations), we infer that locally the diffeomorphism IfA : (T*P)/ H ~
P~ X 1I ~* is given by

lfA(p, q, (}) = (PIL - A~(q)(}j, qIL, (}i). (2.54)

It follows from (2.42) and 1.4.2 that the Poisson bracket on P~ Xli 1)* is
°
{j, g}* =
af ag
-a -aIL -
af ag
-aIL -a- -
i af
(}iFILo(q)-a -a
ag
PIL q q PIL PIL Po
 2 Internal Symmetries and External Gauge Fields 235

ck 0 (_ af ~ Ai ( ) af ~ Aj af ag ). (2.55)
+ ij k aOi aOj + J1, q apJ1, aO j + J1, aOi apJ1, ,
for brevity we have omitted the argument (p, q, (J) on both sides. One notices that
this is the sum of the canonical Poisson brackets on T* Q and ~~ and terms in
which the connection A and its curvature F (see (2.21)) enter.
As an immediate corollary one infers that
(2.56)

as symplectic manifolds, where T* Q is equipped with the canonical symplectic

structure 1.2.23.

2.4 Bundle Automorphisms and the Gauge Group

We have seen in n.3.3 that when H is trivial the group gQ defined by 11.(3.12)
plays an important role in the construction of observables on T* Q. In particular,
the map J in 11.3.1.4 is the momentum map for the action of g Q on T* Q defined
in 11.3.1.5 (this observation is a special case of the results below).
If H is nontrivial and P is a principal H-bundle over Q, neither Diff(Q) ~
C~(Q, 1R) nor Diff(P) ~ C~(P, 1R) is sensitive to the bundle structure. We now
prepare ourselves for the specification of the correct group.
Definition 2.4.1. The group Aut(P) is the group ofsmooth bundle isomorphisms of
P with compact support on Q .In other words, Aut(P) is the restriction of Diff(P)
to those Coo diffeomorphisms cp that satisfy
(2.57)
for all h E H, and for which the projection (to Q under 'f) of the set where cp
differs from the identity is compact.
The gauge group Gau(P) C Aut(P) consists of those cp E Aut(P) satisfying
'foCP='f·

Proposition 2.4.2. The space Aut(P) is isomorphic to f(P x H P), that is, to the
collection of sections of the bundle P x H P associated to P by the given H -action
on P. Here the group operation in r(p x H P) (realized as H -equivariant maps
cpM : P -+ P) is cp:dcp~d = cp~d 0 cp~d.
The gauge group Gau(P) is isomorphic to r(p XH H), where the associated
adjoint bundle P x H H is defined with respect to the adjoint action of H on itself.
The group operation in f(P XH H) (realized as H-equivariant maps g := gM :
P -+ H) is pointwise multiplication.
Hencefor trivial P = Q x H the gauge group is isomorphic to C~(Q, H) (with
pointwise multiplication as the group operation).
The first claim is immediate from Proposition 2.1.5. Define a map g : P -+ H
by the property
Cpg(X) = Rg(X)-l(X) = xg(x). (2.58)
236 III. Groups, Bundles, and Groupoids

Since by definition one has

g(xh-') = Ad(h)g(x) = hg(x)h-', (2.59)

the second claim follows.

The relation between Aut(P) and Diff(Q) is described by the exact sequence of
•
groups

1 -+ Gau(P) -+ Aut(P) -+ Diff(Q) -+ 1; (2.60)

that is, the image of each homomorphism is the kernel of the next one. The second
arrow is given by inclusion, and the third is the map ({J ~ ({JQ, where ({JQ is the
element of Diff(Q) defined by ({J through the bundle projection T; that is,

T 0 ({J = ({JQ 0 T. (2.61)

As in 11.3.3 we say, in a somewhat loose sense, that the Lie algebra ()iff(P) of
Diff(P) is the collection r(TP) of smooth compactly supported vector fields on P.
Since Aut(P) and Gau(P) are, equally loosely, Lie subgroups of Diff(P), we can
discuss their respective Lie algebras. In preparation, note that the space (TP)/ H
is a bundle over Q, with projection inherited from TTP->P->Q. The space r(TP)H
consists of all H -invariant vector fields ~ on P for which T( supp (~» is compact.
It is a Lie algebra under (minus) the commutator borrowed from r(TP).

Proposition 2.4.3. One may identify r«TP)/ H) and r(TP)H, upon which the
Lie algebra llut(P) of Aut(P) is isomorphic to r«TP)/ H).
The Lie algebra gllu(P) ofGau(P) of the gauge group is isomorphic to r(P XH
~), where P XH ~ is the vector bundle (over Q) associated to P by the adjoint
representation of H on ~, and the Lie bracket on r(p XH ~) is the pointwise
bracket in ~.
Hence for trivial P = Q x H the Lie algebra of the gauge group is isomorphic
to C~(Q,~) (with pointwise Lie bracket).

Since an element ~P E llut(P) by definition satisfies

(2.62)

it is clear that llut(P) = r(TP)H. With [X]H denoting the equivalence class of
X E TP in (TP)/ H, the map r(TP)H 3 ~ ~ ~ E r«TP)/ H) defined by
~(q) = [~(S(q»]H is therefore independent of the sections : Q -+ P.Conversely,
one puts ~(x) = rit->p(x) n Tit->(TP)/ H(~(rp->Q(x»); this intersection consists of
one point, since the H -action is free. These two maps provide a bijection between
r«TP)/ H) and llut(P).
The Lie algebra gllu(P) of Gau(P) comprises all vertical H -invariant vector
fields on P; the second claim therefore follows from Lemma 2.3.3. Alternatively,
it is obvious from Proposition 2.4.2. •

The relationship between P XH P and (TP)/ H will be elucidated in 3.8.8.

 2 Internal Symmetries and External Gauge Fields 237

For later use, we record that elements).. := ).. Ad E rep x H ~) of gau(P) satisfy
(cf. (2.59»
)..(xh- 1) = Ad(h»)"(x). (2.63)
The diagram (2.60) infinitesirnalizes to an exact sequence of Lie algebras
o -+ gau(P) -+ aut(P) -+ iliff(Q) -+ 0, (2.64)
where we recall that iliff(Q) = r(TQ) (with minus the commutator as its Lie
bracket). The corresponding exact sequence of vector bundles (allover Q) is
o -+ P XH ~ -+ (TP)/H -+ TQ -+ 0; (2.65)
taking sections, we recover (2.64). A connection A on P is then equivalent to a
splitting of the sequence (2.64) (in the sense of a map from iliff(Q) to aut(P) that
is a left inverse to the arrow in the opposite direction), since ~ E iliff(Q) has a
horizontal lift e(~), which lies in aut(P) because of (2.11).
As shown in (2.3), a (local) section s : Q -+ P is equivalent to a (local) trivial-
ization 1/1S : P -+ Q x H. In a fixed such trivialization 1/1s, a gauge transformation
q;g : P -+ P then induces a (local) diffeomorphism gS : Q x H -+ Q x H by
requiring that gS o1/ls = 1/Is 0 g ("active picture"). This yields
gS(q, h) = (q, gs(q)h), (2.66)
where gs is related to g as in 2.1.5; that is,
gs(q) = g(s(q». (2.67)
In the "passive picture" the gauge transformation q;g defines a new section Sg by
the property 1/Isg = 1/Is 0 q;-;l. Using (2.59) this gives (cf. (2.4»
Sg(q) = s(q)gs(q). (2.68)
It easily follows from (2.14), (2.15), and (2.57) that q; E Aut(P) has the prop-
erty that the pullback (q;-I)*A is a connection I-form if A is. This particularly
applies to q;g E Gau(P). It is interesting to compute the action of Gau(P) in a local
trivialization given by a section s.
Proposition 2.4.4. For q;g E Gau(P) (cf. 2.4.2) one has
s* M(q) = Ad(gs(q»s* A(q) + gs(q)dg;l(q), (2.69)
where Ag := (q;-;I)*A. The right-hand side of(2.69) describes both the value of
(q;-; I )* A in the fixed trivialization 1/1s defined by s (active picture). and the value
of A in the trivialization defined by the transformed section Sg-I (passive picture).
Alternatively, iJtwo sections Sa and sp are related by (2.4), and Aa := s~A etc.,
then, writing gpa = g;;J. one has
Ap(q) = Ad(gpa(q»Aa(q) + gpa(q)dgaP(q). (2.70)
The last claim follows from s* 0 (q;-;l)* = s* -I' as is immediate from (2.68).
To derive (2.69), which should be comparelwith (2.23), one evaluates the left-
hand side on X = dq(t)/dtlt = 0; this yields A(d/dt s(q(t»g;l(s(q(t»lt = 0),
238 III. Groups, Bundles, and Groupoids

where we used (2.67). The differentiation dldt firstly hits s(q(t»; one uses (2.15)
to obtain the first tenn on the right-hand side of (2.69). It secondly hits g;l( . .. );
this time one inserts gs(q)-lgs(q) after s(q), and uses (2.14) to find the second
tenn. •

The second tenn in (2.69) equals (g; I )*8fc (q), and is, of course, also equal to
-dgs(q)g;l(q). Suppressing the dependence of g(q) on the section s, physicists
write (2.69) as

(2.71)

In any case, the second tenn drops out in the transfonnation of the curvature:
From (2.20) and (2.69), or (2.21) and (2.71), one infers that

S*F8(q):= s* ° (rp,;-I)*F(q) = Ad(gs(q»s*F(q). (2.72)

2.5 Construction of Classical Observables

In this section we construct a complete set of classical observables on a symplectic
leaf (,PP)o of(T*P)1 H in tenns of the momentum map of a certain group action.
In preparation, consider the following general construction (which includes (2.78)
below as a special case).

Definition 2.5.1. Let Ll be an H -action on M, denoting the bundle associated to

P through this action by Ml (here I is some label). Then Aut(P) acts on MI by

(2.73)

This action is well-defined on account of (2.57).

Returning to the main theme, our basic group is the semidirect product

g~ := Aut(P) ~ C~(Q, 1R), (2.74)

which is defined through the action rp : g ~ (rpQI)*g, where rp E Aut(P) and

g E C~(Q, 1R), and rpQ is the diffeomorphism of Q defined by rp; see (2.61).
Theorem 2.5.2. Consider the g~-action on T*P defined by

Po(g) : w ~ w - d-,;*g(w); (2.75)

Po(rp) : w ~ (rp-I)*w, (2.76)

and Po (rp , g):= Po(g) 0 Po(rp), where g E Cgo(Q, lR)andrp E Aut(P). This action
is strongly Hamiltonian and commutes with the H -action R*, so that there is a
reduced strongly Hamiltonian action p~ on C[;;-"P)o.
Recall (2.53). The actions p~(g) := 1/Ir ° Po(g) and p~(rp) := 1/IA~ ° Po(rp),
where A'" := (rp-I)*A, on P~ XH 0 are then given by

p~(g) : [x, a, 8]H ~ [x, a - dg, 8]H; (2.77)

p~(rp) : [x, a, 8]H ~ [rp(x), (rpQI)*a, 8]H. (2.78)
 2 Internal Symmetries and External Gauge Fields 239

In particular, for a gauge transformation f/lg one simply has

p~(f/lg) : [x, a, 8]H 1-+ [x, a, Co(g(X»e]H, (2.79)

where g E f'(P XH H) is ~ated to f/lJ!y (2.58).

Finally, each p~(f/l) : (T*P)~ 1-+ (T*P)~~ is a Poisson map; cf 2.3.9.

Here [x, a, 8]H E P~ XH 0 is the H-equivaIence class of (x, a, 8) E P~ X o.

Note that the reduced actions p~(f/l) and p~(g) are well-defined because of (2.57)
and (2.59), respectively. Equation (2.77) is obvious. To derive (2.78) one uses the
property f/l*~1 = ~1, which follows from (2.57). Specializing (2.78) to a gauge
transformation, the associated diffeomorphism f/lQ of Q is the identity, so that f/lg
maps (x, a, 8) into (xg(x), a, 8); cf. (2.58). Since [xh, a, O]H = [x, a, CO(h)e]H
by definition of the H -equivalence class in question, (2.79) follows.
The fact that the group action on T*P is strongly Hamiltonian follows from
2.3.1, and from the existence of an equivariant momentum map: For Aut(P) one
has (2.31), whereas for the C~(Q, 1R.)-action on T*P one has

(2.80)

Compare with 11.(3.7).

Since the action Po on T* P commutes with the H -action on this space, the
momentum map J for Po reduces to a well-defined map JO on the reduced space
Cf;'p)o. Because of the definition of the Poisson structure and the reduced group
cr
action p~ on 7 pp, the map JO is an equivariant momentum map for p~. This
shows that p~ is strongly Hamiltonian.
The last claim is obvious from 2.3.9 and the rest of the theorem. •

Let 1/1 : P -+ Q x H be a (local) trivialization of P; see the text preceding (2.36).

A local expression for the action p~ (f/lg) : (,f";'p)o -+ Cf;'p)o of the gauge group
m~ be derived from (2.66). Pulling this action back to T*P and subsequently to
(T* P)o ~ T* Q x 0 (locally), one obtains

p~(f/lg) : (aq , O)s 1-+ (aq + O(dg;'(q)gs(q», Co(gs(q»e)s. (2.81)

Here a q E Tq*Q, and dg;'(q)gs(q) E I); cf. the proof of 2.4.4. Also, we have
explicitly indicated the dependence on the section s. In other words, the point
(PJL, qJL, OJ)s is mapped to (PJL + (oJLg(q )-' gs(q»iO j , q, Co(gs(q »{ OJ)s.
Relative to a fixed section s, a p-form fA on T* Q x 0 (perhaps defined only
locally), also depending on the connection A = s* A, is said to be gauge-covariant
when
(2.82)

This ~operty, sacred in physics, states the fact that f is an expression of a function
on (T* P)o, depending on the connection A, in a local trivialization. A case in point
is the covariant momentum, which is a I-form defined by

(2.83)
240 III. Groups, Bundles, and Groupoids

In coordinates this reads p~(p, q, 0) = Pll - OJ A~(q).1t then follows from (2.81)
and (2.69) (or (2.71» that this function is indeed gauge-covariant. More generally,
an expression of the type f 0 pA is gauge-covariant.
Similarly, we could compare (local) expressions for a function on cF"pp in
two (local) trivializations whose (local) sections are related by (2.4). With slight
abuse of notation we define

(2.84)

With the relation (2.70), the condition for gauge-covariance is

fAp 0 p[?(g/Ja) = fAa. (2.85)

It follows from (2.54) and (2.82), or from direct calculation, that the local
expression for p[?(f(Jg) : (PP)~ ---+ (PP)~go cf. (2.79) and (2.69), is simply

(p[?(f(Jg»(aq , O)~ = (a, Co(gs(q)wt g , (2.86)

where the explicit dependence on the connection has been displayed. In other
words, the momentum on P~ XH 0 is gauge-covariant. The fact that p[?(f(Jg) is a
Poisson map may then be verified from 1.(2.15), (2.86), and (2.55).
A (local) trivialization o/s, corr~onding to a (local) section s of P (cf. (2.3»,
induces a (local) trivialization (T* pp ~ T* Q x 0, whose inherited Poisson
structure is simply the sum of the canonical bracket 1.(2.24) on T* Q and minus
the Lie-Poisson bracket (1.3) on 0; that is,
° af ag af ag k af ag
(2.87)
{t, g} = apll aqll - aqll apll - CjjOk aOj ao j '

Theorem 2.5.3. Relative to an arbitrary connection A on P and a (local) section

s, the equivariant momentum map J O of the reduced g~-action p[? on (PP)o (cf
2.5.2)isasfollows. With slight abuse ofnotation we simply write J O for (o/s-I)* J O,
putting the s-dependence into the argument of JO.
For the C~(Q, lR)-action (2.75) on (PP)o one has

Jf(aq, O)s = g(q). (2.88)

For general ~ E aut(P), identified with a vector field ~p on P, with (2.83), ~Q :=

'l"*~P, and A := s*A, one has

J~q,(aq, O)s = P~q.IJ)(~Q) + () 0 As(q)(~p). (2.89)

For pure gauge transformations this specializes to

Jf\aq , ()s = O(As(q», (2.90)

where A E rep x H I) and As := A 0 s; cf 2.4.3 and 2.1.5.

Finally,for the horizontalliJt of some vector field ~ Q on Q one has

Jl~~Q)(aq, O)s = P~q.IJ)(~Q)· (2.91)

 2 Internal Symmetries and External Gauge Fields 241

The equivariance of J O is immediate from 2.5.2, and may be verified by explicit

computation using the above formulae. The expression (2.80) for the momen-
tum map of the C~(Q, R.)-action on T*P entails (2.88). To derive the remaining
expressions we decompose
(2.92)

where we recall that ~Q = r*~P, and t'(~Q) = s*~Q -~I(~Q)' This follows from the
identity ~P = s*~Q +(~p -s*~Q), in which the second term is vertical, and (2.17).
It is obvious from (1.19) and (2.9) that (1/Is)*~{ = ~*; with (2.31), this leads to'
(2.90). Also, the property (1/Is 0 s )*~ Q = ~ Q, combined with the previous equation
and (2.31), leads to (2.91). Equation (2.89) then follows from (2.92), (2.91), and
(2.90). •
One verifies that J O is gauge-covariant in the sense of (2.82); in case the section
s explicitly occurs, according to Proposition 2.4.4 one should interpret the symbol
M in (2.82) by substituting Sg-I for s; cf. (2.68). For example, the invariance
of the last term in (2.89) may be checked using (2.15) and (2.62). Similarly, the
gauge-invariance of (2.90) follows from (2.68) and (2.63).
Since the map p~(cp) in (2.78) maps ('PP')~ to (PP)~., which has a different
Poisson structure, there is no concept of a momentum map unless A'" = A (i.e.,
the connection is invariant under (cp-l)*). In that case the momentum map ]0 is
given by

(2.93)
which is weB-defined, and easily follows from (2.89) and (2.50).
The momentum map for p~(g) always exists, and, analogously to (2.88), is
given by
(2.94)

2.6 The Classical Wong Equations

We tum to the Hamiltonian. In the spirit of 11.3.3 we assume that there is a Rie-
mannian metric gQ on Q. On each symplectic space (PP)~ = P~ X H 0 one then
has a natural Hamiltonian

hO([x,a,8]H):= tgQl(a,a); (2.95)

cf. 11.(3.30). In coordinates this is simply
hO(p,q,8) = ~g~V(q)P/LPV' (2.96)

where g~V = (ggv)-l. Since (2.95) explicitly depends neither on 8 E 0 nor

on the connection A, the Hamiltonian hO is evidently invariant under the gauge
transformation (2.79) or (2.86). Since, by Theorem 2.5.2, the diffeomorphism
p~(cpg) is a Poisson map from (PP)~ to (PP)~g, it follows from 1.2.3.5 and the
242 III. Groups, Bundles, and Groupoids

gauge-invariance of hO that the Ham~nian flow on (PP)~g is the image under

i/~(q;g) of the Hamiltonian flow on (T*P)~.
From the Poisson bracket (2.55) one then obtains the equation of motion
qJl = g~V(q)pv; (2.97)
.
PJl a pu()
-- -"2I JlgQ q PpPu + FiJlV (q )PV().
j, (2.98)
• k .
()i = -Cij()kpJl A~(q). (2.99)
These are known as the Wong equations, describing the motion of a "colored"
particle in the external gauge field A. The word "color" refers to the classical
charge, represented by the coadjoint orbit O. The equations have a simple geo-
metric interpretation. Firstly, the projection of the motion to T* Q (cf. the remark
following (2.53» is cogeodesic motion distorted by the Lorentz force, that is, by
the last term in (2.98). Secondly, the motion ()(t) in the fiber 0 is given by parallel
transport in the bundle P x H 0 associated to P by the coadjoint action of H on
O. Indeed, looked at as an equation in the vector bundle P~ x H 1)* , with covariant
derivative VA defined by the connection A (see 2.2.3), equation (2.99) combined
with (2.97) reads vt() = o.
We will show that the Hamiltonian flow on each leaf is the reduction of a single
Hamiltonian flow on T*P.
Proposition 2.6.1. There is a bijective correspondence between H -invariant
Riemannian metrics g on P satisfying
~(~L(h)X' ~ld(h)Y) = ~(~1, ~t) (2.100)

for all x E P and X, Y E I). and triples (gQ, (~I}qEQ' A). where gQ is a Rie-
mannian metric on Q, each g;
is a bi-invariant Riemannian metric on H (the
g;
dependence on q being smooth in that (X, Y) E COO(Q)for all X, Y E TeH =
I)), and A is a connection on P(Q, H, T).
Given g, for each q E Q one defines a bilinear form on Te H = I) by
g:(X, Y):= g.(q)(~l,~t); (2.101)
by (2.10) and the right invariance of g this is independent of the section s. Since
g: is Ad-invariant by (2.100), one subsequently obtains a bi-invariant metric (with
the same name) on H by left or right translation. The smooth dependence on q is
immediate from the smoothness of g.
A connection on P is constructed by defining Hx P C Tx P as the orthogonal com-
plement of Vx P; condition (2.11) is satisfied because g is H -invariant. Equivalently,
the connection I-form A may be directly constructed as
Ax = (g:)u 0 J 0 gu, (2.102)
where gu : TP -+ T*P is defined below II.(3.15), J is the momentum map for
the H-action on T*P, and (g:)U : 1)* -+ I) is obtained from g~X) in the usual way.
Equation (2.14) is then satisfied because of (2.31), whereas (2.15) is a consequence
of the equivariance of J.
 2 Internal Symmetries and External Gauge Fields 243

Finally, the metric gQ is constructed from g by gQ(X, Y) := g(e(X), e(Y»,

where the point to which one lifts X, Y horizontally is immaterial in view of
the H -invariance of g. In other words, gQ is the unique metric on Q for which
horizontal lifting is an isometry.
In the opposite direction one defines g on VP by reading (2.101) from right to
left, declares HP to be orthogonal to VP, and manufactures the metric on HP as
in the previous paragraph. •
In a trivialization P ~ Q x H (locally) the correspondence is
gn = gij;
ij
A ip. = gHgP.j;
Q _ g
g p.v _ gij g .g . (2.103)
- p.v H p.j VI'

where g% is the inverse of gn (which may differ from gij). All expressions depend
on q E Q but not on h E H. These comments equally well apply to the expressions
below. In the opposite direction we obtain
gij = gn;
H .
gip. = gp.i = gijA~;
_ Q H i j
gp.v - gp.v +gijAp.Av· (2.104)
In preparation for the following theorem, we mention the obvious fact that
any H-invariant function f on T*P is is well-defined on (T*P)/ H, and therefore
defines a reduced function fO on each leaf (PP)o by restriction.
Theorem 2.6.2. Let the equivalent data in 2.6.1 satisfy the condition that
independent ofq. Then the Hamiltonian h* on T*P, defined by 1I.(3.30) through
be g:
the metric g on P, reduces to a function h~ on each leaf P~ x H 0 ~ (PP)o,
which differs from h O in (2.95) by a constant. In other words, the equations of
motion ofh~ are the Wong equations (2.97)-(2.99).
This is most quickly established in local coordinates; inverting g using (2.104),
one obtains
h*(p, q, e, h)R = ~g1/(q)(pp. - eiA~(q»(pv - ejA~(q» + g%(q)()iej. (2.105)
Using the inverse of (2.54), the reduced Hamiltonian on P~ x H 0 is

h~(p, q, e) = ~g(/(q)PP.PV + g%(q)()iej. (2.106)

When g%(q) is independent of q, one computes from (2.55) that the last term in
(2.105) Poisson-commutes with every function on P~ XH 0. This computation
exploits the fact that C}kgf{ is totally antisymmetric, which is a restatement of the
Ad-invariance of gH. Since P~ x H 0 is a symplectic space, this means that the
term in question must be a constant. •
Rather than on (PP)~ one can work on (PP)O, equipped with the Poisson
bracket (2.87); cf. (2.53). This is more natural when the bundle P is trivial, i.e.,
244 III. Groups, Bundles, and Groupoids

P:::: Q x H, for in that case one has (PP)o :::: T*Q x O. Also, formulating the
dynamics on (PPp is better suited for studying the relation between the classical
and the quantum theory; see 2.11 below. The following considerations equally well
apply to a local trivialization of a possibly nontrivial bundle P.
Instead of (2.106), which generates the Hamiltonian flow on (PP)o as trans-
formed by the diffeomorphism (2.54), and relative to the Poisson bracket (2.55),
°
one now uses the Hamiltonian h~ : = h 0 1{Ii]. By (2.96), the coordinate expression
of h~ is
(2.107)
This may equally well be obtained from (2.105), omitting the last term (which
does not contribute to the equations of motion).
The comment following (2.83) evidently applies; one has

h~g 0 p~(cpg) = h~, (2.108)

so that the gauge transformation p~(cpg) maps the Hamiltonian flow on (PP)o
generated by h~ into the Hamiltonian flow generated by h~g.
From (2.107) and (2.87) one deduces the Hamiltonian equations
(2.109)
(2.110)
(2.111)

where P~ := PI-' - 8; A~(q), as before. A different form of (2.111) is obtained by

transferring the motion on 0 to a certain flow h (.) on H; this correspondence wiIl
be used in the proof of Theorem 2.11.1. We assume that 0 = O(J is the coadjoint
orbit through 8, and note that hh -I and the connection form A are elements of the
Lie algebra ~.
Proposition 2.6.3. Let the flow h (.) in H be the solution of
(2.112)

with initial condition h(O) = h.lf8(t) and h(t) are related by 8(t) = Co(h(t»8,
so that
8j (t) = 8(Ad(h(t)-IT;), (2.113)

then 8; (t) solves (2.Jll) if h(t) solves (2.112).

One evaluates both sides of (2.112) in the coadjoint representation, acts on 8,
and uses (1.20). •

2.7 The H -Connection

Let us illustrate the preceding concepts in the case that the principal bundle P is
G (G / H, H, r), where G is a Lie group having H as a closed subgroup (hence H
 2 Internal Symmetries and External Gauge Fields 245

is automatically a Lie group), r is the canonical projection from G to G I H, and

the H -action on G is given by
(2.114)
Many nontrivial bundles are obtained in this way. The simplest and most famous
one will be studied in 2.12.
The restriction to H of the adjoint representation Ad(G) on 9 quotients to a
representation Is(H) on g/f), called the isotropic representation. This leads to the
vector bundle G XH (g/f) overGIH associated with G(GIH, H, r). Let f)0 C g*
be the annihilator of f) c g; clearly, f)0 ~ (g/f)*. The coisotropic representation
Ci(H) on f)0 is the restriction of the representation Co(G) on g* to H. This defines
the associated vector bundle G x H f)0.
Lemma 2.7.1. There are diffeomorphisms (which are bundle isomorphisms whose
associated diffeomorphism of G I H is the identity)
T(GI H) ~ G XH (g/f); (2.115)
r(GI H) ~ G XH f)0. (2.116)
Furthermore, there is an isomorphism
G~ ~ G x f)0 (2.117)
as principal H -bundles over T*(G I H). Here the H -action on G x f)0 is given by
Rh(x, e) = (xh- 1, Ci(h)(), which defines G x f)0 as a principal H -bundle over
T*(GIH) through the isomorphism (2.116).
Recall that G~ is defined by 2.3.2 (with P replaced by G). The canonical left
action L of G on G I H pushes forward to a G-action L* on T(G I H). One identifies
T[elH (G I H) with g/f) in the obvious way. Under the isomorphism (2.115) the point
[x, Y]H E G XH (g/f) then corresponds to (Lx)*Y E T(GI H). One verifies the
independence of the representative (x, Y).
Similarly, T[;lH(GIH) ~ f)0, and [X,e]H E G XH f)0 corresponds to L;_,e E
T*(GIH).
Finally, (2.117) is obtained by letting (x, [y, e]H) E G~ correspond to the point
(x, Is(x-1y)() E G x f)0; note that x-1y E H by the definition of G~, and that
this is evidently independent of the choice of (y, e) "" (yh- 1 , Is(h)(). •
Recall (2.51). Using the right trivialization of T*G (see 1.4), and noticing thatthe
momentum map for the H -action on T*G (from the right) is simply the restriction
of the momentum map of the G-action to f) c g, we see from (1.51) and (1.57)
that
(2.118)
Here [X]H := xH; since 0 is stable under Co(H), the right-hand side is
independent of the choice of x.
On the other hand, in the left trivialization of T*G (1.55) and (1.49) yield
(T-;C)o = {[e, X]H elf) E -OJ, (2.119)
246 III. Groups, Bundles, and Groupoids

where the H -equivalence classes are the orbits of the H -action given by
h : (e, x) ~ (Co(hW, Xh-l). (2.120)
We note that G acts on itself by left multiplication; this action evidently com-
mutes with the H -action (2.114), allowing us to regard G as a subgroup of the group
Aut(G) of all bundle automorphisms of G(G / H, H, T). According to (1.52), in
°
the right trivialization the reduced G-action A on (T-;:'G)o is

A~(e, [Y]H) = (Co(xW, [XY]H), (2.121)

whereas in the left trivialization, (1.50) yields

A~([e, Y]H) = [e, XY]H. (2.122)

By (1.58) and (1.56), the momentum map J O : (i;'C)o -+ g~ for this action in
the right and the left trivialization, respectively, is
JO(e, [X]H) = e; (2.123)
JO([e,X]H) = Co(x)e. (2.124)
Choosing an orthonormal basis {Ta} of 9 (with ensuing coordinates ea on g*, cf.
1.1) allows one to write (2.123) as

Jao(e, [X]H) = eaR. (2.125)

Here Ja°(-) := JO(.)(Ta) and e: is the coordinate function ea relative to the right
trivialization, regarded as a function on (T-;:'G)o (it is, of course, equally well a
function on T*G and on (T*G)/ H).
We now further specialize the discussion to the following situation.
Definition 2.7.2. A closed subgroup H eGis called reductive if there exists a
linear space meg such that Ad(H)m = m and
9 = I) E9m. (2.126)

One calls (2.126) a reductive decomposition of g. Any compact subgroup is

reductive, for one can equip 9 with an H -invariant inner product, and define m to
be the orthogonal complement of I).
Choose a basis {Ti} of I). Given a reductive decomposition (2.126) of g, let lei}
be elements of g* with the properties ei(Tj) = 8~ for all i, j = 1, ... , dim(H)
and ei(X) = 0 for all X E m.
Proposition 2.7.3. Let H be a reductive subgroup ofG with associated reductive
decomposition (2.126). The H-connection
dim(H)
AH(x):= L ei(x)® T; (2.127)
i=l

is independent of the choice of basis (within the class of bases considered), and
defines a connection i-form on G( G / H, H, T).
 2 Internal Symmetries and External Gauge Fields 247

Thus AH is the restriction of the left Maurer-Cartan form efc to ~ (see 1.4)
in an adapted basis. The defining properties of a connection are easily verified:
(2.15) follows from (1.53), whereas (2.14) is equivalent to [~, m] ~ m (which is
immediate from the reductivity of H). The basis independence of A H will be clear
from the proof of the following corollary. •
Let us note that A H is a G-invariant connection, because 9fc is G-invariant
under the canonical left action of G on T* G.

Corollary 2.7.4. There is a bijective correspondence between reductive decom-

positions (2.126) and G-invariant connections on G(Gj H, H, 't').
We use the left trivializations TG ~ 9 x G and T*G ~ g* x G; cf. (1.46)
and (1.47). The vertical tangent vectors in TxG are those of the form (X, X)L,
where X E ~, whereas the horizontal ones are defined by A H to be the vectors
(y, X)L, where Y E m. Hence the choice of m as a complement to ~ in 9 defines
the connection, and vice versa. In view of (1.50), the G-invariance of a connection
A forces it to be of the form A H • •

Using (2.117), the diffeomorphism (2.47) implemented by AH assumes the

simple form

(2.128)
Comparing with (1.47) one sees that this amounts to a factorization g* ~ ~o x ~*
as a manifold; as a vector space this actually sharpens to g* ~ ~o E9 ~* .It is evident
from (2.7.4) that an H -connection provides such a decomposition, for apart from
the canonical embedding ~o ~ g* it defines an embedding ~* ~ g* through the
identification of ~* with mO C g*.
Using (2.128), we denote elements of G~ x ~* by (x, J.1" 9), where x E G,
J.1, E ~o, and 9 E ~* = mO. The H-action on G~ x ~* is

Rh(X, J.1" e) = (xh- 1, Ci(h)J.1" Co(h)e), (2.129)

which is consistent with (1.49). The momentum map for this action, given in
general form by (2.48), then simply reads J(x, J.1" 9) = -e, which complies with
(1.55). Obviously, the references to (1.49) and (1.55) are on the understanding that
the pertinent G-action on T*G is restricted to H. Hence by (2.53) one has the
symplectomorphism

(2.130)
for the symplectic leaf (T7CJ.2. in (T*G)j H. For 0 = to} this reproduces (2.116).
The bundle projection of (PGp onto T*(Gj H) is given simply by

(2.131)

In the present formulation, the reduced G-action (2.122) on (T7CJ)o is

(2.132)
248 III. Groups, Bundles, and Groupoids

This is a special case of (2.78); the I-form a E T* Q appearing in the latter is

not to be confused with /-L E 1)0 in (2.l32). In the present context and notation the
momentum map (2.124) of this action reads
JO ([x, /-L, (}]H) = Co(x)(/-L + (}). (2.l33)
This depends on the connection, which enabled us to identify () E 1)* with an
element of g* .
It is often more informative to~ve the momentum map relative to a (local)
trivialization of P, and hence of (T*Gp ~ T*(G / H) x 0 (locally). Specializing
(2.89) to the case at hand we obtain, using (2.127), (1.42), and (2.83),
(2.134)

where the index a on JO and ~Q stands for Ta. Specializing 2.6.1 to the situation
at hand, we have
Proposition 2.7.5. An Ad(H)-invariant inner product (, ) on 9 defines
• a left-G-invariant and right-H -invariant Riemannian metric g on G;
• a G-invariant Riemannian metric gG / II on G / H;
• a bi-invariant Riemannian metric gH on H;
• an H-connection AH on G(G/ H, H, r).

Proposition 2.6.1, where g:

The metric g satisfies (2.100), and the four objects listed are related as in
is independent of q.
The metric g is obtained from (, ) by left translation and the identification
9 = Te G . The restriction of (, ) to m ~ g/I) yields an inner product on T[HJ G / H,
and subsequently a G-invariant Riemannian metric gG/ II by left translation. The bi-
invariant metric on H is defined by left or right- translating the restriction of (, ) to
I). The H -connection is obtained by declaring m to be the orthogonal complement
of I); cf. 2.7.4. Equation (2.100) is a restatement of the Ad(H)-invariance of (, ).
It is easily verified that gG/ JI, gH, and A defined in 2.6.1 coincide with the ones
above. •
We are therefore in the situation of Theorem 2.6.2. With respect to the basis
{Ta} of 9 introduced earlier, the free Hamiltonian 11.(3.30) on T*G in the left
trivialization is simply
dim(G)
h*«(}, X)L =i L ();. (2.l35)
a=!

If (, ) happens to be Ad(G)-invariant, the expression for h*«(}, X)R in the right

trivialization is the same. In that case, equations (1.58) and (2.l35) imply
dim(G)
h* = i L (1/;)2, (2.136)
a=!

where JL is the momentum map for the left action of G on T*G. Being right-H-
invariant, this momentum map reduces to the momentum map JO of the reduced
 2 Internal Symmetries and External Gauge Fields 249

G-action on (T""";"'GP; cf. (2.125). Hence the reduced Hamiltonian h~ on (T7i;)o ,

defined as in (2.106), is given by essentially the same expression:
dim(G)
h~ = ~ L (1;;)2. (2.137)
a=l

2.8 The Quantum Algebra ofObservables

In 2.3 we defined the Poisson algebra COO«T* P)/ H), and noticed the isomorphism
COO«T*P)/ H) ~ coo(T*p)H. We now tum to the quantization of this algebra. For
simplicity we assume that we have an H -invariant measure JL on P that is locally
Lebesgue. This may be the measure obtained from an H -invariant Riemannian
metric g on P, cf. 2.6.1 and 11.(3.40). In any case, one may define the Hilbert space
L 2 (P) := L 2 (P, JL).
On L 2(p) we have a representation U R of H, given by

UR(h)W{x) = W(xh). (2.138)

Hence we can define the C* -algebras 23 o(L 2(p»H and 23(L 2{p»H of compact and
bounded operators on L 2 (P) that commute with each U R(h), h E H. The latter is
not particularly useful in the present context, whereas the former is empty unless
H is compact. We will therefore proceed on the assumption that H is compact.
This assumption leads to a particularly clean analogy with the classical case. The
noncompact situation will be treated, with new techniques, in 3.7.1.
According to II.3.4 one may think of 23 o(L \P»IR as the quantization of the
Poisson algebra C~{T*P, JR); a quantization map is provided by the generalized
Weyl quantization QJi in 11.3.4.4. In analogy with Theorem 1.9.2 (corresponding
to the special case P = H = G) we are led to
Theorem 2.S.1. Let QJi : C~{T*P) ~ 23 o{L 2(p» be as defined in 1l.3.4.4, and
assume that H is compact and K is H -invariant.
With~(o = C~«T*P)/H),sothat~O = Co«T*P)/H),and~h = 23 o(L 2 (P»1l
forli E JR\ (O), the map QJi :
~o ~ ~h definesanondegenerate strict quantization
of(T*P)/ H, with the possible exception of the completeness condition 1l.1.1.1.4.

The H -invariance of K may be achieved by averaging over the compact group

n
H, but this is not essential: Since for small enough QJi
(f) is independent of K,
it would suffice to use any quantization that maps C~(T*P, JR)H into 23 o{L 2(P»:
and coincides with QJi for small n.
The theorem is immediate from Theorems I1.3.S.1 and 11.3.6.3. •

It is not difficult at all to determine the structure of the (complexified) quantum

algebra of observables 23 o{L 2(p»H. In preparation, we note that the H -invariant
measure JL on P defines a unique measure v on Q, satisfying

i dJL{x) f(x) = k 1 dv(q) dh f{s(q)h) (2.139)

250 III. Groups, Bundles, and Groupoids

for any f ELI (P) and any measurable section s : Q -+ P. This measure is locally
Lebesgue; if J1, comes from a Riemannian metric g, then v is just the Riemannian
measure determined by gQ (cf. 2.6.1). This measure is used in the construction of
L2(Q) := L2(Q, v).
Proposition 2.8.2. Each measurable section s : Q -+ P determines an
isomorphism Q3 o(L 2(p»H ~ Q3 o(L 2( Q» ® C*(H).
Here the tensor product is defined as the norm-closure of the algebraic tensor
product in the natural representation on L2(Q) ® L 2(H).
The section s determines a trivialization of P by (2.3), which leads to a uni-
tary transformation Us : L 2(P) -+ L2(Q X H) ~ L2(Q) ® L2(H) defined by
Us qJ (q , h) := qJ(s(q)h). Considerthe space 1.B 2(L 2 (P»11 of H-invariantHilbert-
Schmidt operators on L \P), whose elements K are characterized by a kernel
K E L 2 (P X p)H satisfying
K(xh, yh) = K(x, y) (2.140)
almost everywhere. We construct a map p : L 2(P x P)1I -+ L2(Q X Q x H)
by (p(K»(q, q', h) = K(s(q)h, seq'»~. We then identify L2(Q x Q x H) with
1.B 2(L 2(Q» ® L 2(H), where 1.B 2(L 2(Q» and L2(H) are seen as (dense) sub-
spaces of I.B O(L2(Q» and C*(H) (in tum identified with C;(H), see l.7). It is
then verified that UsK U; = p(K). Since the norm-closures of 1.B 2(L 2(P»11 and
1.B 2(L 2(Q» ® L 2(H) are l.Bo(L 2(p»H and Q3 o(L 2(Q» ® C*(H), respectively, the
claim follows. •
Corollary 2.8.3. Up to equivalence there is a bijective correspondence be-
tween the irreducible representations rrX of I.B O(L 2(P»H and the irreducible
representations Ux of H, X E fi.
This follows from 2.8.2, l.7.5, and 1.2.2.6.
An analogous statement holds for arbitrary representations, but the stated form
•
is helpful in understanding the analogy with the classical result 2.3.8.
The representations rr X(l.B o(L 2 (P»H) may be explicitly realized in various
forms. The first one is rr{ on
1-l; = L 2(Q) ® l-l x ' (2.141)

The proof of Proposition 2.8.2 leads to

rrsX(K)qJ;(q) = l dv(q') l dh K(s(q)h, s(q'»Ux(h)IJ1;(q'). (2.142)

One may realize 1-l x in a fashion that is directly analogous to the realization of the
symplectic manifold (PP/) as the associated bundle P~ XII O. The following
construction is valid as it stands whether or not H is compact, as long as H is
unimodular.
Definition 2.8.4. Consider the Hilbert bundle
HX = P XII l-lx (2.143)
 2 Internal Symmetries and External Gauge Fields 251

associated to P by the representation Ux of H on 1t x (here Ux may be reducible,

but for convenience we use notation pertinent to the irreducible case). There is a
natural inner product on the space of sections r(HX), given by

(qJ(x), <I>(x):= fa dv(q)(qJ(x)(q), <I>(x)(q»x. (2.144)

where (, )x is the inner product in the fiber rH"/~Q(q) :::: fix (different identifi-
cations of the fiber with rtx lead to the same inner product). The Hilbert space
L2(HX) is the completion ofr(HX) under this inner product.
As explained in 2.1.5, we may realize r(HX) as the set of smooth functions
qJx : P -+ rtx satisfying the equivariance condition (2.5); that is,
qJX(Xh-l) = Ux(h)qJX(x). (2.145)

Moreover, the projection of the support of qJX to Q must be compact. in this

realization the inner product is given by

(qJX, <l>X):= fa dv(r(x» (qJX(x), <l>X(x»x; (2.146)

the integrand indeed depends only on x through rex) because of (2.145). The
Hilbert space rt x is the completion ofr(HX) in this inner product.
We return to the case that H is compact. Then rt X is a subspace of L \P) ® rtx:
the latter carries a representation U R ® Ux of H (cf. (2.138», and it follows by
definition that rt X is the subspace of L 2(p) ® rtx transforming trivially under
UR ® UX' We already encountered a special case of this in 1.8.
Proposition 2.8.5. The representation rrx (~O(L2(p»H) on 1t x , defined by

rrX(K)qJX(x) = i dJ.L(Y) K(x, y)qJX(y) (2.147)

(initially defined on ~2(L2(p»H and extended to ~O(L2(p»H by continuity),

corresponds to Ux(H) as in 2.8.3.
Note that the left-hand side satisfies (2.l45), and that the integrand on the right-
hand side is a function of r (y) because of (2.140). The expression may therefore
be rewritten as

rrX(K)qJX(x) = fa dv(r(y» 1dh K(xh, y)Ux(h)qJX(y). (2.148)

To relate rr x to the realization (2.142) we realize r (HX) in the second manner

mentioned in 2.1.5. In this realization the inner product on rsCHX) is simply the
one in (2.141), which therefore is the closure of r s (HX). The relation (2.6) between
qJ; and qJX defines a unitary map T/ : 1t x -+ rtf, given by
T/qJX(q) = qJX(s(q»; (2.149)
(T/)-lqJf(x) = Ux(hs(x»qJf(r(x». (2.150)
For all A E I.B O(L 2(P»H one then has T/ rrx (A)(T/)-l = rrf (A).
•
252 III. Groups, Bundles, and Groupoids

In the spirit of the idea of bundles, it is more elegant to take not a single (possibly
discontinuous) section s, but rather a collection of smooth local sections Sa : Na -+
P, relative to a cover {Na } of Q; cf. 2.1. On regions of overlap Na n N p, the
appropriate sections are related by (2.4). This leads to yet another realization of
the carrier space 1t x , which we denote by 1tfs). This is defined as the closure (under
the inner product (2.152) below) of the space of all objects wtl' defined as follows.
A vector Wt) consists of a collection {W;} of smooth functions W; : Na -+ 1t x '
which are related on overlap regions by
(2.151)

(no sum over fJ). Furthermore, one requires that (wt), wt) < 00 in terms of the
inner product on Hr.), defined by

(wt), <l>fs) := L J( dv(q)lPa(q)(W;(q), <I>~(q»x'

a Q
(2.152)

Here {lPa} is a partition of unity subordinate to the cover {Na }; the inner product
is independent of its precise choice.
Proposition 2.8.6. The realization Jrt) is given by (q E Up)

Jrt)(K)W;(q) =L ( dv(q')lPa(q') JH( dh K(sp(q)h, sa (q'»Ux (h)W;(q').

a JQ
(2.153)
If ga : Na -+ H is such that s(q) = sa(q)ga(q), the realizations Jr/~) and
Jrl are intertwined by the unitary Vs,/s) : 1t; -+ 1tfsl' given by Vs,/s)W:(q) :=
Ux (ga(q»Wf (q). With (2.142) this leads to (2.153). •
The classical inclusion ('f~p)o C (T* P)/ H has a quantum analogue; the Hilbert
space 1t x carrying an irreducible representation Jr x (Q3 o( L 2(p»H) may be naturally
realized as a subspace of L 2(p). This is done through the following result (of which
(1.111) is a special case).
Proposition 2.8.7. For compact H one has the decomposition
L 2(p) ~ E91tX ® 1tx:, (2.154)
XEH

under which
Q3o(L 2(p»H ® JrR(C*(H» ~ E9 Jrx (Q3 o(L 2(p»H) ® Jrx:(C*(H». (2.155)
xE11
Here one could replace JrR(C*(H» and Jrx:(C*(H» by UR(H) and Ux:(H),
respectively. This proposition is proved by mapping 1t x ® 1tx: into a subspace of
L 2(P) so as to intertwine Jrx ® 1rx: with the defining representation of~o(L2(P»H
tensored with UR(C*(H». Define VX : 1t x ® 1tx: -+ L 2 (P) by linear extension of
(2.156)
 2 Internal Symmetries and External Gauge Fields 253

where v E 1ix and d x = dim(1i x)' Note that V x is indeed linear, since 1ix = 1i X'
Equation (2.139) and the orthogonality relations (1.116) then imply that VX is a
partial isometry. Using (2.140) and (2.145) one verifies that
VX 0 JrX(A) ® Ux(h) = A ® UR(h) 0 VX (2.157)
for all A E lBo(L 2(p»H and h E H.
The simplest way to prove that EBXEH VX = ][ is to use the isomorphism 2.8.2;
the operator Us featured in the proof of 2.8.2 accomplishes
UslBo(L 2(p»H ® JrR(C*(H»Us* = lBo(L 2(Q» ® JrdC*(H» ® JrR(C*(H»,
(2.158)
where JrR and JrL are defined via (1.89) by the right- and the left-regular represen-
tations (1.98) and (1.83) of H on L2(H) (with c = 1), respectively. The desired
result then follows from the Peter-Weyl decomposition (1.100) of L 2 (H). •

One may select a copy of1i x by picking a fixed unit vector v E 1i x ' The operator
Pf on L 2 (P) defined by

Pf'l1(x) = dx L dh (v, U x (h)v)x 'l1(xh) (2.159)

lies in the commutant of lBo(L 2(p»H, and is a projection for which Pvx L 2(p)
~ 1i x and Pf A ~ Jrx (A) for all A E lBo(L2(P»H.

2.9 Induced Group Representations

We tum to an important application of principal bundles and their associated vec-
tor bundles. In what follows we do not assume that H is compact, unless stated
otherwise.
Recall (2.73), which in the present case defines a H -action UX on the Hilbert
bundle HX associated to P by a representation U x(H) on a Hilbert space 1i x .
Definition 2.9.1. Let a principal bundle P( Q, H, i) and a representation Ux(H)
on 1ix be given. The induced representation U(x) of the group Aut(P) of
automorphisms ofP on the Hilbert space L2(HX) is given by

dv(rp(/(q» UX( )'l1(X)( -I( ». (2.160)

dv(q) rp rpQ q

It is essential here that the measure v on Q is locally Lebesgue, for this guarantees
that v 0 rpQI and v are equivalent. This means that these measures have the same
null sets; one says that v is quasi-invariant under Diff(Q). In view of the square
root it is easily checked that U(x)(rp) is unitary, hence defines a representation.
In the realization 1i x of H -equivariant functions \IIX : P ~ 1ix satisfying
(2.145) this reads

(2.161)
254 III. Groups, Bundles, and Groupoids

In the realization 1ifs) one has

(2.162)

where we assume that q E Na and qJQl(q» E N fJ , and hfJ(x) E H is defined by

the property xhfJ(x) = sfJ(r(x»; cf.2.1.5.
In the realization 1i; one of course has the same expression, with the indices a
and fJ omitted.
These representations of Aut(P) may be extended to representations of g~ (see
(2.74». For any Ii =1= 0 the (additive) group C~(Q, R) is represented on L2(HX)
by the appropriate analogue of 11.(3.81), namely
Pkx\,g)W(x)(q) := e-;g(q)/hw(x)(q). (2.163)

Essentially the same expression is valid on 1i; or on 1i~); on 1i x one has

(2.164)

We then obtain a representation P~ (9~) by letting P~ (Aut(P» coincide with

UX(Aut(P», and putting p~(qJ, g) = p~(g)p~(qJ). These induc~epresentations
are the quantum analogues of the reduced actions pg>(9~) on (T*P)o in 2.5.2; in
the quantum case the relevant action is on the space of sections of the appropriate
associated bundle, rather than on the associated bundle itself (as in the classical
case).
When H is compact this construction of induced representations can be refor-
mulated as follows. Firstly, g~ is contained in gp (see 11.(3.12» by the inclusions
Aut(P) C Diff(P) and C~(Q, R) ~ C~(P, R)H C C~(P, R). Now define a
representation Ph of g~ c gp = Diff(P) ~ C~(P, R) on L 2 (P) by restriction of
the representation Ph(9P) given by 11.3.6.2 (with Q replaced by P). Subsequently,
extend Ph(9~) to a representation Ph ® Ix on L 2 (P) ® 1i x , where [x is the unit
operator on 1i x ' The restriction of Ph(9~) ® [x to the subspace 1i x C L 2(p) ® 1ix
is then given by p~. Note that this restriction is well-defined, because Ph(9~) lies
in the commutant of UR(H).
We give yet another description of P~ .

Lemma 2.9.2. Let H be compact. Under the decomposition (2.154) the restriction
of Ph(9~) to 1i X ® 1ix is p~ ® [x'
This follows from (2.159) and the orthogonality relations (1.116). •

As explained at the beginning of 11.3.6, one can extend the Weyl quantization
prescription Q]i (initially defined on C~w(T*P, R» to certain unbounded func-
tions. This equally well applies to the restriction of Q]i to C~w(T*P, R)H, with H
compact. A representation JrX(~O(L2(p»H) may be extended to the H-invariant
unbounded operators on L 2(p) thus encountered in an obvious way. The following
result is the "quantization" of Theorem 2.5.3, whose notation we use.
 2 Internal Symmetries and External Gauge Fields 255

Theorem 2.9.3. Let J be the momentum map for the gp-action (2.75), (2.76) on
T*P. Relative to an arbitrary connection A on P one has
:n;X (Q~(Jg») = g; (2.165)

:n;x (Q~(J~p») = -in [vtQ +!v .~Q -dUx(A(~P»]; (2.166)

:n;x (Q~(h») = indUx(A); (2.167)

:n;x (Q~ (Jf(~Q») = -in [vtQ + !v . ~Q], (2.168)

defined as unbounded operators on the domain r(HX) C L 2 (HX).

It should be clear what these expressions stand for in the various realizations
1tx, 1t;, and 1t&). For example, g and V . ~Q are functions on Q, which on
1t; and 1t&) are realized as multiplication operators; on 1tx, though, one should
pull these functions back to P with r~->Q' In all cases g should more properly
be written as g ® lx' On 1t x the object dUX (F), where F is A(~P) or A, acts
like dUX (F(x»'-li x (x), whereas on 1t; and 1trs) one should replace F by F 0 s
(recall that A 0 s = As) and tensor with the unit operator on L2(Q). The covariant
derivative V~Q is defined in 2.2.3; on 1t; one has, from (2.26),
V~'-lI:(q) = (ajt + A~(q)dUxCf;))'-lI:(q), (2.169)

where A = s* A. The corresponding expression on 'H&) is obtained by replacing

'-lI; by '-lit and A by s~A.
The proof of 2.9.3 starts with an equation of independent interest, namely
indUX(X) = :n;X(Q~ (lx», (2.170)
for all X in the Lie algebra gp of gpo This equation is defined on r(HX), and follows
directly from 11.(3.82) and Lemma 2.9.2. Now note that on 1t x one simply has
11.(3.73) and 11.(3.74). This easily leads to (2.165)---(2.168); for example, (2.166)
follows from (2.25) and (2.92). •
Equations (2.165), (2.166), (2.167), and (2.168) are to be compared with their
classical counterparts (2.88), (2.89), (2.90), and (2.91), respectively.
Proposition 2.9.4. The operators in (2. J65 )-(2. J68) are essentially self-adjoint
on r(HX) C L 2(HX).
This follows as in the proof of Proposition II.3.6.4. o
Since J is equivariant (cf. 2.5.3) and UX is a representation, on account of
(2.170) the following equation holds on r(HX) for all X, Y E gp:

~ [:n;x (Q~ (J x ») , :n;X(Q~ (ly»] =:n;x (Q~ ({lx, Jy}); (2.171)

this is a version of Dirac's condition 11.(1.3).

This is the right place to mention the quantum-mechanical counterpart of the
discussion on gauge covariance in 2.5. Firstly, the analogue of the classical con-
dition (2.82) in the "active picture" is as follows. Letsg-l(q) = S(q)g;l(q), as in
256 III. Groups, Bundles, and Groupoids

2.4.4,anddefinetheunitaryoperatorUg : L2(Q)®'lt x --+: L2(Q)®'lt x (regarded

as a map from 'It; to 1t;g -I) by Ug\llsx(q) := U x (gs(q»\II; (q). An operator OA
that involves dUx (A), where A = s* A, is said to be gauge-covariant if
U g OAU*g = OAg , (2.172)

where Ag is related to A by (2.69). The covariant derivative (2.169) is a case in

point. The meaning of this condition is similar to that of (2.82); it is satisfied when
OA is of the form T/ OA(T/)-I; cf. (2.149) and (2.150).
Secondly, the quantum counterpart of (2.85) in the "passive picture" is that an
operator OA on 1t&J satisfy

(2.173)

t,
where U gafi \II~ := \II as defined as in (2.151).
We now specialize the construction of induced representations to the case that
the principal bundle P is G( G / 1/, 1/, 'f); see 2.7. Given a representation Ux of 1/
on a Hilbert space 'lt x , we construct the associated vector bundle HX := G x Il'lt x
as in the general case. A central ingredient in the definition of the induced Hilbert
space 1t x is the measure v, which is constructed from an 1/ -invariant measure on
G. We use a right-invariant Haar measure dx := dfL(x) on G for this purpose. In
the present context, the space 1t x is the closure (in the inner product (2.146» of
the space of smooth functions \II x : G --+ 'It x whose projected support on G / 11
is compact and that satisfy (2.145).
Recall from 2.7 that G C Aut(G). The induced action of Aut(P) on HX given
by (2.73) specializes to G :3 Y by
(2.174)
Equation (2.161) then specializes to

dV('f(y-I X » x -I
dv('f(x» \II (y x). (2.175)

We continue to denote points in the base space Q = G / 1/ by q, and denote the

canonical left action of G on G/1/ by y : q f-+ yq, where y E G. On 1t&J we
then have, in the notation of (2.162),

dV(y-Iq) -I -I x-I
dv(q) Ux(s,Aq) YStl(y q»\II tl (y q). (2.176)

As in the general case, the realization of u,x (y) on 1t; = L 2 (G / 1/) ® 1t x (defined
with respect to a single measurable section s : G / 1/ --+ G) is obtained from
(2.176) by simply omitting the indices a and f3.
This special case of induced group representations is called Mackey induction;
compare with the corresponding classical theory described in 2.7. The formulae for
Mackey induction simplify in the case that G and 1/ are unimodular, which implies
that v is not merely quasi-invariant but actually invariant under the canonical left
 2 Internal Symmetries and External Gauge Fields 257

action of G on G j H. Hence in that case the square roots in (2.175) and (2.176)
are identically 1 and can be omitted.
Combining (2.125) and (2.170), we infer that

rr;X(Q';(8:» = ifidUX(Ta ), (2.177)

where eaR is regarded as a function on (T*G)j H; cf. the comment after (2.125).
To put this in perspective, let us return to Weyl quantization on T*G (where G
is seen as a Riemannian manifold). Using 11.(3.74) (in which the divergence term
vanishes in view of the invariance of the measure) and (1.58), we obtain
(2.178)

on C;o (G) c L 2 ( G), in terms of the left-regular representation U L defined in (1.83)

(with c = 1). In view of 2.8.7, equation (2.177) therefore follows from (2.178), at
least when H is compact. For later reference (cf. (2.192» we give the corresponding
formula in the left trivialization of T*G. Defining e/; as 8a relative to the left
trivialization, we infer from (2.178) and the relation eaR(ax ) = Ad(x-I)~e;(ax),
where ax E Tx*G (cf. (1.42», that

Q';( (e;) = -ilidUR(Ta ). (2.179)

Combining (2.177) and (2.166), and assuming that G and H are unimodular,
we obtain the geometric expression
(2.180)

on 'H.x; on'H.;
(and analogously on 'H.~}) one replaces \II X (y) by \II; (q) andAy(~:>
by As(q)(~:). The right-hand side is, of course, independent of A.
We now assume a reductive decomposition (2.126); reca112.7.3. Specializing to
the associated H-connection AH on the bundle G(Gj H, H, r), equation (2.166)
becomes (cf. the corresponding classical expression (2.134»

rr;X (Q';( (Ja») = -in [v~; - CO(s(q»~dUx(Ii)]. (2.181)

See IV.2.8 for applications of Mackey induction in physics.

2.10 The Quantum Wong Hamiltonian

We now look at the quantization of the Hamiltonian, assuming that H is compact.

Definition 2.10.1. Given a Riemannian metric gQ on Q and a connection A on

P(Q, H, r), the Laplace-Bochner operator D.~ is a second-order differential
operator on the space a/sections l(HX) a/the Hilbert bundle HX associated to P.
For \II X ,<l>x E r(HX) it is defined by the property

(\II X, D.~<l>X) := - ~ dv(r(x»gQI(VA\IIX(x), VA<l>X(x»x. (2.182)

258 III. Groups, Bundles, and Groupoids

Cf. II.3.7.1; recall that elements of r(HX) by definition have compact support.
An analogous definition can be given on fs(HX) or f{s}(HX), where 'l'(x) and x
are replaced by q. In any case, the expression ( .. .)x in (2.182) is an element of
Tq* Q ® Tq* Q, so that the integrand is a scalar.
In coordinates, on fs(HX) or f{s}(HX) one has (cf. (2.169»

L\~ = g~v(V/l + dUX (A/l»(8 v + dUx(A v », (2.183)

where V is the Levi-Civita connection defined by gQ. Here dUx(A/l) may be

rewritten as A~dUx(1;), where A~ is a multiplication operator on L2(Q). Note
that L\ ~ is gauge-covariant; cf. (2.172).
Proposition 2.10.2. In the notation and circumstances ofTheorem 2.6.2 the quan-
tum Hamiltonian H; := Jrx (Qf (h*» in the sector X (for the moment defined as
an unbounded operator on the domain f(HX») is
H ItX = _l/i?
2
(AAX - lR
3 Q
+ ..lF2
12
- eX) '
(2.184)
where RQ is the Ricci scalar on Q, the Yang-Mills Hamiltonian is given in an
arbitrary section and coordinate system by
F2 '=
•
g!~g/lPgvO'
IJ Q Q
Fi/l v FjpO" (2.185)
and the constant e X is
eX := e~ (H) + fig~je~ke!t. (2.186)

Here the Casimir element e~ (H) in the representation U x (H) is defined

through Li(dUx (T;»2 = -e~(H)[x; it is a (positive) constant, as U x is irre-
ducible. It is possible to give an intrinsic definition of F2, but note that (2.185) is
independent of the section in which F = s*F is computed because of (2.72) and
the Ad-invariance of gH. The terms RQ and F2 in (2.184) depend on q, and are to
be seen as multiplication operators on 1tx.
Equation (2.184) follows from 11.(3.93) and two identities. Firstly, if L\ is the
Laplace-Beltrami operator on P defined by g, one has
A = L\~ - e~(H) (2.187)

on 1t x (seen as a subspace of L 2(p) ® 1t x )' This easily follows from 2.6.1 if one
decomposes \ltV and \leI> in 11.(3.91) in a horizontal and a vertical part; cf. the
text surrounding (2.35). Secondly, the Ricci scalars on P and Q are related by the
famous identity
R = RQ - 41 F2 + 4gH
I kl e i e j
jk iI' (2.188)
This may be verified from 11.(3.23), 11.(3.21), (2.21), and (2.104). One obtains
some additional terms, whose sum vanishes on account of the Ad-invariance and
q-independence of gH. •
One tool in the analysis of the possible self-adjointness of H; is the following.
If tV E (fh.. 1t).., then tV).. denotes the component of tV in 1t)...
 2 Internal Symmetries and External Gauge Fields 259

Lemma 2.10.3. Let 1i = Efh 1ii.., and let a closed operator Ai.. be given for each A.
The operator A := EDi..Ai.. is defined on the domain V(A) consisting of all \{I E 1i
for which \{Ii.. E 'D(Ai..) and Li.. Ai.. \{Ii.. E 1i (this operator is easily seen to be
closed). Then A is self-adjoint iff each Ai.. is self-adjoint.

The adjointA* of A is easily seen to be A* = EDi..A~. If the equation A*\{I = ±i\{l

has no solution in V(A*), then none of the equations A~\{I = ±i\{li.. can have a
solution, and vice versa. •

Corollary 2.10.4. If(P, g) is complete with Ricci scalar R bounded, then Hi is

essentially selfadjoint on r(HX)for all X E fI.
In particular, Hi is essentially self-adjoint on r(HX) when Q is compact.

This follows from 11.3.7.4, 2.8.7, 2.10.3, and the inclusion C~(P) C D(ED x H~),
where H~ is the closure of Hi as defined in 2.10.2. •

When Q is compact the hypothesis is obviously satisfied.

This result, and similarly Proposition 2.9.4, gives a hint as to why vector bundles
are relevant in quantum mechanics. For one might question this relevance on the
grounds that 1i x is the same for all vector bundles over Q with fiber 1i x ; it is the
dense subspace r(HX) that is sensitive to the topology of the bundle. We therefore
conclude that this topology is relevant for the specification of the domain of the
key observables of the quantum theory, such as the Hamiltonian.
We may look at the quantum Hamiltonian Hi from a different point of view. If h*
is the usual classical Hamiltonian on T*P (defined with respect to an H-invariant
metric g on P), under the appropriate hypotheses the operator HI! = Qr (h*) de-
fines a self-adjoint operator on L 2(p), interpreted as the quantum Hamiltonian of
a particle moving on P. This leads to an action of IR as a one-parameter automor-
phism group a~ on ~o(L 2(p», defined as in 11.(2.88). By the H -invariance of HI!
this restricts to ~o(L 2(p»H. One may then ask whether at is implemented in a
representation Jrx (~o(L 2(p»H), in that there exists a unitary group t 1-+ U;n(t)
on 1t x for which

(2.189)

for all A E ~O(L2(p»H and all t E lR. This is evidently the case in the present
circumstances, and the Hamiltonian remerges as the generator of U;n' that is,
U;n(t) = exp(-itHi In).
Specializing the theory to the case where the principal bundle P is chosen to be
G(G I H, H, r), as in 2.7, allows one to give a purely group-theoretic formulation
of the various geometric objects encountered. We first look at the situation on G
itself.

Lemma 2.10.5. Let a unimodular Lie group G be equipped with a left-invariant

metric g, with corresponding orthonormal basis {Tal ofg = TeG. Then the La-
place-Beltrami operator ~ on C~(G) C L2(G) is given in terms of the Casimir
260 III. Groups, Bundles, and Groupoids

element
(2.190)

and the right-regular representation (1.98) (with c = 1) by

(2.191)
We start from the definition 11.(3.91), and write df = 8f~aL f (cf. 1.4). Since
sab, the Riemannian volume element defined by the metric must be
g-I(8 a , 8 b ) =
proportional to a left-invariant Haar measure (unique up to scale), which is also
right-invariant by unimodularity. Hence we can partially integrate in the right-hand
side of 11.(3.91), and obtain (2.191) on account of dUR(X) = ~f, X E g. •
Comparing (2.191) with II.(3.93) and (2.135), we infer that

Q;;(h*) = i ~ Q;;«8aL )2) = -ill? (~dUR(Ta2) - ~RG); (2.192)

note that the Ricci scalar RG

is a constant. To put this in perspective, one should
recall (2.179), inferring that QJi «8;)2) =I- (QJi (8;»2.
We now assume that H is a reductive subgroup of G, so that we are in the situation
of Proposition 2.7.5. Comparing (2.191) and (2.187), and using the H -equivariance
of elements of 7{x , we conclude that

(2.193)

where the Ta occurring in the sum form a basis of m. This expression may be
substituted into the Hamiltonian (2.184); in the present case the last three terms
on the right-hand side of (2.184) are constants. If, in addition, the inner product
(, ) on 9 is Ad(G)-invariant, one has

L
dim(lJ)
~~H = dUX(Ta2 ) + Ci(H), (2.194)
a=l

which should be compared both with its classical counterpart (2.137) and with the
more general quantum formula (2.187).

2.11 From the Quantum to the Classical Wong Equations

Let us now investigate the possible classical limit of the dynamics generated by the
quantum Hamiltonian Hi on 7{x. As in Theorem 11.3.7.5 we assume that Q = IRn,
so that the bundle P = IRn x H is necessarily trivial. Accordingly, given a coadjoint
orbit 0 C ~*, one has (PP)o = T*Q x O. Also, 'HI = L2(Q)®7{x with respect
to the natural section s : Q --+ P given by seq) = (q, e); cf. (2.141).
Recall that H is a compact Lie group. As in 1.10, we assume that the object X
labeling the irreducible representation Ux(H) is a highest weight, corresponding
 2 Internal Symmetries and External Gauge Fields 261

to an integral coadjoint orbit Ox C f)*. We extend X E t* to an element (J(X) of f)*,

as explained in 1.10, and denote the stabilizer of (J(X) under the coadjoint action
by Hx. Then Ox :::::: HIHx'
Inspired by 1.11 we put Ii = 11k, kEN, and study the Hamiltonians H;A
on 'Hkx = L2(Q) ® 'H kx as k --+ 00. The corresponding time evolution is then
expected to converge to the flow on cF"p)o generated by the classical Hamiltonian
h ~x given by (2.107).
The precise formulation of this convergence will be in terms of pure state quan-
tizations, as in Theorem 11.2.7.2. Combining the coherent states 1I.(3.95) in L 20Rn)
with their counterparts Ukx(h)'Pkx E 1ikx , where h E H (see (1.149», we define
unit vectors in L 2(JR.n) ® 'Hkx by
'P(p,q,h) '- 'P(p,q) (0, U (h)'P (2.195)
11k .- 11k '01 kx kx'
Recall that 'Pkx is a normalized highest weight vector in 'Hkx'
Analogously to Iy in 1.11, for I E C(T*JR.n x Ox) we define Ix E C(T*JR.n x H)
as the pullback of I under the canonical projection from T*JR.n x H to T*JR.n x Ox;
the function Ix is right-Hx-invariant. The Berezin quantization corresponding to
the pure state quantization (2.195) is then given by

Q~(f):= dx/n l'JRn ~~:~~: l dh Ix(p, q, h) ['PhP,q'h)] , (2.196)

defined for I E C~(T*JR.n x Ox' JR.), taking values in mn = lBo(L2(JR.n» ®

9J1dX / h (C)JR. In (2.196) and all subsequent expressions in this section, Ii = 1I k.
Theorem 2.11.1 below generalizes Theorems 1I.2.7.2 and 11.3.7.5. As before,
we write a?(f) for the function (p, q, (J) f-+ I(p(t), q(t), (J(t», where the time
evolution is defined for t E (ti' tf) as the Hamiltonian flow on T*JR.n x Ox generated
by h~x; in other words, (p(t), q(t), (J(t» is the solution of the system (2.109)-
(2.111) with initial conditions (p, q, (J). Since the terms tli2qRQ -izF2 + CX)
in (2.184) do not contribute for Ii --+ 0 (as will be clear from the proof below), in
what follows one may replace Hi ln by

(2.197)

Here Ph~~ := -ili"VfL' P,~v := -iMfL (as in (11.2.24», and

TP := iIidUx/n(Tj ); (2.198)
this is, of course, consistent with (2.107), 1I.(2.24), and (2.167) with (2.90). We
assume that g and A are C 3 near the classical trajectory (p(t), q(t), (J(t», and such
that each multiplication operator occurring in Hi lh is O(exp(tx2» for x --+ 00.
Then for Ii < 1 the operator Hen is symmetric on the domain Dc consisting of
the linear span of all states (2.195). Subsequently, we assume that each operator
He h thus defined has at least one self-adjoint extension, which we denote by the

same symbol. This abuse of notation is justified by the fact that as in 1I.2.7.2, for
times that the classical flow exists different self-adjoint extensions will have the
262 III. Groups, Bundles, and Groupoids

same classical limit. We then put

• i HX/~ i Hx/h
a;'(A) := e',t ~ Ae-;;t ~ . (2.199)

Theorem 2.11.1. Let g and A be as specified above. For given (p, q, h) E T*JR.n x
H (corresponding to(p, q, Co(h)O(X» E T*JR.n x 0x»),/orall t E (ti> tf)andall
/ E C~(T*JR.n x Ox, JR.), one has

~~ (W~P.q.h>, [Qg(a~(f» - a~(Qg(f))]W~P.q.h)) = 0 (2.200)

along the sequence (h = 1/ khEN.

This may, of course, be restated as

(2.201)

Equation (2.200) is proved along the lines of the proof of 11.(2.135), of which it
is obviously a generalization. For simplicity we restrict the argument to the case
where the metric g on Q is the flat Euclidean one; nontrivial metrics are easily
incorporated by the method of proof of Theorem 11.3.7.5.
As in 11(2.136), 11.(2.137) we expand H;/Ii = H(2)(t) + H3(t). Here
o
Ho := h / (p(t), q(t), O(t»H, (2.202)

where the classical Hamiltonian h ~x (p, q, 0) is given by (2.107), and subsequently

the argument (p, q, 0) is replaced by the solution (p(t), q(t), O(t» of the Wong
equations (2.109)-(2.111) with initial conditions (p, q, 0). Also,
A . .
H\(t) := p/l(t)(8P/l - avA~(t)Oj(t)8Qv - A~(t)8Tj), (2.203)

in which p~(t) := P/l(t) - Oi(t)A~(t), where A~(t) := A~(q(t»; furthermore,

8P/l := PK./l - P/l(t), along with 8Qv := Q~'v - qV(t) (cf. 112.23» and 8Tj :=
Tjli - OJ (t); cf. (2.198). Finally,

H2(t) := &(8P/l - aVA~(t)Oi(t)8Qv - A~(t)8T;)2

- &p~(t)apatTA~(t)Oj(t)Oi(t)A~(t)8QP8QtT. (2.204)

Generalizing 11.(2.154), we introduce the classical propagator

U~p·q·h)(t) = eHS(,)- f A)U! (p(t), q(t»Ux/li(h(t»Ux/li(h)*U !(p, q)*, (2.205)

~ ~

in which J A is shorthand for J~ ds q/l(s)A ~ (s )OJ (s); the classical action S(t) is
given by 11.(2.152), with h replaced by h~x; andh(t) is the solution of(2.112) with
initial condition h (the parameter appearing in (2.200». Using the Wong equations
(2.109), (2.110), and (2.112), as well as 11.(2.155), the relation
d .
-Ux/Ii(h(t»
dt
= dUx/li(h(t)h(t)- 1
)Ux/Ii(h(t)),
 2 Internal Symmetries and External Gauge Fields 263

and (2.113), with () replaced by ()(X), one verifies that U}p,q,h)(t) satisfies II.(2.156)
with initial condition U}p,q,h)(O) = ]L
The next step in the proof of 11.2.7.2 is not as easily generalized, since there
no longer exists a simple expression for the propagator of a Hamiltonian that is
quadratic in pi,..,Q~'v, and T/,. In any case, we introduce

H(2)(t):= ! (Pi,.. - avA~(t)()j(t)Q~'v - A~(t)t/'(t)f

- ! p:(t)apaI1A~ (t)(}j(t)A~ (t)(}j(t)Q~'P Q~,I1, (2.206)
in which
(2.207)
In view of (2.113), we could write «(}(X»(Ad(h(t)-I)T;) instead of (}j(t) on the
right-hand side. It is not difficult to prove that H(2)(t) is essentially adjoint on V e ,
since each vector (2.195) is analytic for H(2)(t), and the linear span of such vectors
is dense.
The unitary operator ug)(t), which plays the role of pli(M(t» in the present
context (cf. 11.(2.158», is defined as the solution of

ih~U(2)(t)
dt D
= H(2)(t)U(2\t)
D
(2.208)

with initial condition ug)(O) = I. An explicit form of this operator is

ug\t) = f:
k=O
(_i)k
h
r dS I 1or dS2'"
1o
tH dSk H(2)(SI)'"
1o
H(2)(Sk),

(2.209)
the sum converging strongly on 'Dc. As in 11.(2.157) we define
Utq,h)(t) := U~P,q,h)(t)U! (p, q)Ux/li(h)ug)(t}Ux/h(h)*U! (p, q)*. (2.210)
• •
Using 11.(2.155) and the fact that U x /Ii is a representation, one obtains the relation
H2(t) = U!(p(t), q(t»Ux/li(h(t»H(2)(t)Ux/li(h(t»*U!(p(t), q(t»*. (2.211)
• •
This suffices to prove 11.(2.159) and 11.(2.161), with the label (p, q) replaced by
(p, q, h).
The analogue of Proposition 11.2.7.3 may now be proved by the same method,
substituting the series (2.209) for pli(M(s» in 11.(2.164) and replacing "'kO,O) by
"'kO,O) ® "'x/Ii. The essential point is that piw Q~'v, and 'it\t) each contribute a
"fluctuation" factor v'ii to the norm on the right-hand side of 11.(2.164). For 'it"'<t)
this is a consequence of (1.142)-( 1.145) and the fact that
-Ii
("'x/Ii,1I (t)"'x/Ii) = o. (2.212)
The appropriate generalization of 11.(2.170) is
(2.213)
264 III. Groups, Bundles, and Groupoids

where fo(p, q, h) : (p', q', 0) ~ f(p'+p, q'+q, Co(h)(). This is a consequence

of 11.2.4.3, (1.159), and the definition (2.196). As explained after the proof of
11.2.7.2, this equivariance property lies at the heart of (2.200). To finish the proof
of 2.11.1 we therefore need the analogue of 11.(2.171), namely

lim (ug)(t)\}Iho.o. e ), Q:(f)ug)(t)\}I~o.o.e») = fx(O, 0, e). (2.214)

h~O

This can indeed be proved from (2.209) and a lengthy combinatorial argument.
Theorem 2.11.1 then follows from 11.2.7.3, (2.210), (2.213), and (2.214). 0

2.12 The Dirac Monopole

It is now time to illustrate some of the abstract concepts introduced in this chapter
in an example that is simple yet instructive. We will work in the setting of 2.7,
specializing to G = SO(3) and H = SO(2), seen as the subgroup of rotations
around the z-axis. The coset G j H is a two-sphere S2, and the principal bundle
SO(3)(S2, SO(2), r) is a quotient by the discrete group 112 of the famous Hopf
fibration S\S2, SO(2), r).
We use the standard generators Ta , a = 1,2,3, of the Lie algebra 9 = 50(3) ~
JR.3. The commutation relations in 50(3) are [Ta, Tb] = Eabe Te, where Eabe is the
fully anti symmetric symbol with EI23 = 1. We label elements x of SO(3) by the
Euler angles (¢, 0, 1/1), so that
(2.215)

where 0 :::: a < 21r, 0 :::: {J < 1r, 0 :::: y < 21r. Since H = (Exp( 1/1)} is
the stability group of the point e z = (0, 0, 1) in JR.3 with respect to the defining
action of S 0(3) on JR.3, we may realize G j H as Gez = S2. The bundle projection
r : SO(3) ---+ SO(3)jSO(2) is then given by r(x) = xe z. This yields
r(R(¢, 0,1/1» = (sin¢ sine, - cos¢ sinO, cosO). (2.216)

We denote this point in S2 by (¢, e). (These coordinates are related to the usual
spherical coordinates (¢s, Os) by ¢s = ¢ - k1r, Os = 0.)
The standard bi-invariant metric g on G is defined by declaring that {Ta} be
an orthonormal basis. By Proposition 2.7.5 this defines an SO (3)-invariant metric
gS2 on S2, as well as an SO(3)-invariant connection on SO(3)(S2, SO(2), i).
The explicit form of the connection will be determined shortly. The metric gS2 is
diagonal in (¢, e), and is easily seen to be given by
gS2 = d0 2 + (sinO)2d¢2; (2.217)

this coincides with the pullback of the Euclidean metric on JR.3 to the unit sphere.
(One could introduce an arbitrary radius r of the two-sphere by multiplying g with
r2, which leads to an overall factor r2 in (2.217) as well.)
Choosing m to be the linear span of TI and T2, the decomposition (2.126) is
reductive, as is easily verified from the commutation relations. This is consistent
with 2.7.5, since the decomposition in question is indeed orthogonal with respect
 2 Internal Symmetries and External Gauge Fields 265

to g. Hence by 2.7.3 and 2.7.4, or alternatively by 2.7.5, there is an associated H-

connection AH on the bundle SO(3)(S2, SO(2), r). To describe AH explicitly we
introduce two sections s± : N± -+ S 0(3) of the bundle. Here N+ and N- consist
of S2 minus the northpole «() = 0) and minus the southpole «() = Jl), respectively.
The sections in question are defined by
s±(CP, 0) = R(CP, (), ±CP); (2.218)
evidently, L is discontinuous at the south pole, whereas s+ is so at the northpole.
The transition function g_+ in (2.4) relating s+ and L is
(2.219)
Proposition 2.12.1. In the gauges s± the H -connection A± := slAII is given by
cos() ± 1
A±(CP, ()) = . ()
sm
w'" ® T3 , (2.220)

where w'" := sin()dcp.

The result is stated in the given form because w'"
rather than dCP has unit norm
with respect to g. We see that A_ would be singular at the southpole «() = Jl),
whereas A+ would be singular at the northpole «() = 0); happily, these points do
not lie in the relevant domain of definition. Combining (2.127) with (2.14), one
sees that
(2.221)
If x(n) are the coordinates of Exp(x(n)Ta) in an arbitrary Lie group G (whenever
these coordinates, which are normal in the sense of 11.3.1, are defined), one has
the relation
(2.222)
where
M(x) := (1 - e-Ad(X)) Ad(x)-I, (2.223)
regarded as a matrix acting on 9 relative to the basis {Ta }. For G = SO(3) the
matrix M can be calculated explicitly, yielding

M(x)a =8asinllxll
b b IIxll
+ xtn)x(n)
IIxll2
(1- sin IIxlI) + Eabcx<n)(I_Cosllxll) (2.224)
IIxll IIxll2 '
where IIxll2 := x(n)x(n)' We now use the identity
(2.225)
Hence the normal coordinatesofL(cp, ()) are (-() sin cP, () cos cP, 0); note that there-
fore IIxll2 = 0 2 • The object A_ can now be computed from (2.221), (2.222), and
(2.224). To find A+ we apply the gauge transformation (2.70), using (2.219); cf.
the comment following the proof of 2.4.4. This yields A+ = A_ + 2dCP ® T3, and
one obtains (2.220). •
266 III. Groups, Bundles, and Groupoids

Being gauge-invariant (for an abelian structure group), the curvature F =

slFH = dA± is the same for A+ and A_, and is given by
F(q" e) = - sine de /\ dq, ® T3; (2.226)
this is -T3 tensored with the volume 2-fonn on S2 with respect to the SO(3)-
invariant metric gS2 defined by g (cf. 2.7.5). The field A± describes a magnetic
monopole (of unit strength) located at the origin oflR3 •
By the theory of the H-connection, in particular Corollary 2.7.4, combined
with the uniqueness of the reductive decomposition (2.126) of 50(3), the mag-
netic monopole field is the unique SO(3)-invariant connection on the bundle
SO(3)(S2, SO(2), r). (Note that the local fonns A± are only SO(3)-invariant
up to a gauge transfonnation.)
Since H = SO(2), whose coadjoint action on ~* = lR is trivial, the coadjoint
orbits of H are simply points in lR. Physically being the electric charge, these orbits
are traditionally denoted by the symbol e. From (2.53) or (2.130), combined with
(2.116), we conclude that the reduced space (T*GY is simply T* S2 as a mani-
fold. Identifying (T* G)e with T* S2 automatically implements the diffeomorphism
(2.53), so that the Poisson bracket on (T*GY is given by (2.55). Since H is abelian,
the C~ in that expression vanish, so that from (2.226) we obtain (with q I = q, and
q2 = e)
e at ag at ag . (at ag at ag ) (2.227)
{f,g}*=-----+esme - - - - - .
apit aqlt aqlt apit apo ap¢ ap¢ apo
We now compute the momentum map rfor the reduced SO(3)-action on
(T*G)e with respect to the sections s±. This is done by specializing (2.134) to the
case at hand. Firstly, for e = 0 one has
+ cosq, Po )
- sinq, cot () p¢
jO(p¢, Po, q" e)± = ( cosq, cote ~: + sinq, P o . (2.228)

This follows from the well-known expression for the vector fields ~a := ~f
generating the SO (3)-action on S2; these are nothing but jO, with P... replaced by
ala .... Because e = 0 in (2.81), the "free" momentum map (2.228) is independent
of the section.
Furthennore, since the coadjoint representation of S 0(3) is the same as its
defining representation (as is its adjoint one), the tenn Co(s±(q" e»~e3 occurring
in (2.134) is simply R(q" e, ±q,)a3' Seen as a vector in lR3 this is given by (2.216),
and evidently coincides with the unit vector pointing at (q" e). Using (2.220) one

1
therefore obtains (omitting the argument (p¢. Po, q" e»

-.-(1
sinq, ± cose)
sme
j~ = jO+e ( -~osq,(1 ±cose) . (2.229)
sme
±1
 2 Internal Symmetries and External Gauge Fields 267

Since by (2.81) the map P[,(f{Jg+_) sends (Pt/>' Po, ¢, () to (Pt/> + 2e, Po, ¢, (), one
has the relation P[,(f{Jg+_)* J~ = J".... Hence Je is gauge-covariant, as it should be;
cf. the comment following the proof of Theorem 2.5.3.
The Hamiltonian he on G e = T* S2 is equivalently given by (2.107) or by
(2.137); using (2.217) and (2.220) one obtains

(2.230)

Using (2.229) it may be verified that (2.137) equals h~ + (egi; cf. the comment
following (2.107). The gauge-covariance of h~ is verified as in the case of J±.
By the third Wong equation (2.99), whose right-hand side evidently vanishes, the
charge e is conserved in time; this is obvious anyway, because the motion cannot
leave the coadjoint orbit O. The extra term in (2.227) then leads to a perturbation
of the cogeodesic motion on T* S2.
The fact that H is abelian allows the introduction of a free parameter g E
lR\{O} in the definition of the principal bundle we started from. The group H g ,
isomorphic to SO(2), but parametrized by fJ E [0, 2Jrg), acts on P = SO(3) by
RfJR(¢, (),1/1) := R(¢, (), 1/1- fJIg). This leads to a modified bundle P(S2, H g , 'l'),
and has the effect that the right-hand side of (2.220) (and hence of (2.226» should
be multiplied by g. Consequently, the parameter e in (2.227), (2.229), and (2.230)
becomes ego
In the general context of mechanics on the bundle SO(3)(S2, SO(2), 'l'),
by Corollary 2.3.8 the significance of the classical parameter e E 50(2)* =
lR is that it classifies the irreducible representations of the Poisson algebra
COO«T* SO(3»1 SO(2),lR). As we have seen (cf. Theorem 2.8.1), one should think
of the J LB-algebra~IR = iBo(L2(SO(3»>i°(2) as the quantization of this Poisson
algebra. By Corollary 2.8.3, the irreducible representations of ~ are classified by
the unitary dual SO(2) = Z of SO(2). Hence each integer n E Z corresponds
to an irreducible representation Jrn of~. For g = 1 this integer is the quantum
counterpart of the classical charge e E lR; the rescaling T3 1-+ T3 I g means that in
this consideration e should be replaced by eg.
Labeling elements of H = SO(2) by fJ E [0, 2Jr), the representation Un is

(2.231)

The Hilbert bundle Hn that the representation Un associates to the principal bundle
SO(3)(S2, SO(2), 'l') (cf. 2.8) is a line bundle over S2 (that is, the typical fiber
is C). The Hilbert space L 2(Hn) of square-integrable sections of this line bundle
carries both the irreducible representation Jrn of the algebra of observables ~IR and
the induced representation un of SO(3); cf. 2.8 and 2.9. It is therefore a central
object in the quantum mechanics of a charged particle moving on S2.
The first realization of L2(Hn) is the space 1{.n of L 2-functions on SO(3) sat-
isfying the equivariance condition (2.145); in Euler angles this condition reads
q,n(¢, (), 1/1 + fJ) = exp(in{3)q,n(¢, (), 1/1). Hence 'lin is exp(in1/l) times a function
of (¢, (). Therefore, this realization has no particular advantage over the other two,
which directly work with functions on S2.
268 III. Groups, Bundles, and Groupoids

The Hilbert space 1il := 1i:±, where one has to choose a sign, is

1i~ = L 2(S2) = L 2 ([0, 2n] x [0, n], sin ededl/J) ; (2.232)

cf. (2.141). The most important operators on 1il are the angular momenta

i1iLiU n(Ta) = Jrn(Q]i(Ja»; (2.233)

cf. (2.170). For n = 0 one obtains the well-known angular momentum operators
(here expressed in shifted spherical coordinates; cf. (2.216»
. a a
- sm l/J cot e al/J + cos l/J ae
a . a
Jr1(Q]i(J» = -in cosl/Jcote al/J + sml/J ae (2.234)
a
a</>

As with its classical counterpart (2.228), this is independent of the section.

For general n, one may compute the left-hand side of (2.233) directly from the
definition (1.69), or one evaluates the right-hand side using (2.181), (2.169), and

1
(2.220). Either way, one obtains

-.-(1
sinl/J ± cose)
sme
n~(Q]i (J» = Jr\Q]i (J» + nn ( - ~os l/J (1 ± cos e) . (2.235)
sme
±1
This expression, then, is the quantization of (2.229). It follows from Proposition
2.9.4 that these operators are essentially self-adjoint on the domain r(Hn) of smooth
sections of the line bundle Hn. For the specification of this domain in the realization
of L 2 (Hn) as 1i~ one needs to take into account that the section s _ is discontinuous
at the southpole. This complication may be resolved by observing that for e =1= 0, Jr
the section \II~ E COO(N_) is related to \11+ E COO(N+) by
\II~(l/J, e) = e-2in</>\II~(l/J, e), (2.236)
compare (2.151) combined with (2.219) and (2.231). Accordingly, \II~ is in r(Hn)
iff it is in COO«O, 2Jr) x (0, Jr» and in addition satisfies the boundary conditions

lim ~\II~(l/J, e) = 0; (2.237)

8 ..... 0 al/J

lim
8 ..... 11'
(~
al/J
+ 2in) \II~(l/J, e) = 0; (2.238)

ak ak
lim - k \II~(l/J, e) = lim - k \II~(l/J, e) (2.239)
</> ..... 0 al/J </> ..... 211' al/J
for all k E {O UN}. It is interesting to verify that condition (2.238) guarantees
that the operators Jr~(Q]i (J» are well-defined on r(H n ), in that the differential
 3 Lie Groupoids and Lie Algebroids 269

operators conspire with the boundary condition so as to effectively replace (1 -

cos e) / sin e in the second tenn by (1 + cos e) / sin e near e = 1T •
On H~ the conditions for 'l1~ to be in r(Hn) are analogous: One requires (2.237)
for e -+ 1T rather than e -+ 0, and in (2.238) one replaces e -+ 1T bye -+ 0 and
+2in by -2in. Condition (2.239) is the same.
The story on H{,s} is entirely analogous: The pair ('l1~, 'l1~) comprising 'l1~} is
related by (2.236), and to be in r(Hn) one imposes the above conditions for H±.
The quantum Hamiltonian HI: is given by (2.184), which may be evaluated
using either (2.194) or (2.183) with (2.217) and (2.220), as well as (2.226) and
RS2 = 1, C~ (S 0 (2» = n 2 . There is not much point in writing down the resulting
expression. The important facts are that HI: is essentially self-adjoint on r(Hn)
(cf. Corollary 2.10.4), and that the parameter e, or, more generally, eg, in the
classical Hamiltonian he is replaced by nn in the quantum Hamiltonian. The same
substitution applies to the angular momentum (2.235). This phenomenon is the
Dirac quantization condition eg = nn, expressing the fact that in quantum
theory electric charge is quantized in the presence of a magnetic monopole field.

3 Lie Groupoids and Lie Algebroids

3.1 Groupoids
A groupoid is a certain generalization of a group, in which multiplication is only
partially defined. When it is defined, it is associative, and each element has an
inverse in a suitable sense. Lie groupoids, which are groupoids with an appropriate
smooth structure, and their infinitesimal objects, Lie algebroids, enable one to give
a unified description of a large class of examples in quantization theory. Although
the definition of a groupoid appears complicated, it will become clear through the
examples that one is studying a most natural object.

Definition 3.1.1. A groupoid G(Q, is, it, t, " I), sometimes written as G ~ Q,
consists of a set G (the total space), a set Q (the base), a map is : G -+ Q (the
source projection), a map it : G -+ Q (the target projection), a map t : Q '-+ G
(the inclusion), and a multiplication· : G2 -+ G, where

and a map I : G -+ G (the inversion). subject to the following conditions. We

write Yt Y2 for Yt . Y2·

1. If(Yt, Y2) E G2 (so that Yt Y2 is defined), then is(Yt Y2) = is (Y2) and it(Yt Y2) =
it(Yt).
2. If i,(Yt) = i t(Y2) and i s(Y2) = it(Y3) (so that (Yt Y2)Y3 and Yt(Y2Y3) are
defined), then (YtY2)Y3 = Yt(Y2Y3).
3. One hasrs(t(q» = it(t(q» = q far all q E Q, and yt(is(Y» = t(it(Y»Y = Y
for all Y E G.
270 III. Groups, Bundles, and Groupoids

4. Writing y- I ;= I(y), the inversionsatisjiesrs(y- I ) = rl(Y)' rl(y- I ) = rs(Y),

y-Iy = t(rs(Y», and yy- I = t(rl(Y».
Note that the second equation in 3.1.1.3 is well-defined on account of its prede-
cessor; similarly, the third and fourth equations in 3.1.1.4 are meaningful because
of the first and the second ones, respectively. It follows from 3.1.1.3 that rs and rt
are onto (i.e., surjective).
It can be shown from these axioms that the inverse is unique, so that one could
omit I from the data specifying a groupoid, and require its existence subject
to the last condition above. It also follows that I is involutive with respect to
multiplication; that is,
(y-I)-l = y; (3.2)
(YIYZ)-1 = Y2- I YI-l. (3.3)

In addition, 3.1.1.3 and 3.1.1.4 imply

yy-Iy = y. (3.4)

In fact, when Yo E t(Q) and(y, Yo) E G2, so that yYo is defined, then yYo = y. This
follows since Yo = t(rs(YI» for some Yl; then Yo = YI-I Yl by 3.1.1.4, but since
rs(Y) = rt(Yo) = rt(YI-lyj) = r,(YI), we see that Yo = t(r,(YI» = y-I y , so that
finally YYo = yy-Iy = Y by 3.1.1.4, as claimed. Similarly, when (Yo, y) E G2,
then YoY = y.
Hence elements of t(Q) act like units for the partially defined multiplication,
and one sometimes calls an element of the form t(q) a unit, referring to Go ;= t(Q)
as the unit space in G. For Yo E G to be a unit it suffices to find one y for which
yYo = y, for in that case the above argument implies that Yo = t(rs(Y».
One thinks of elements of Q as "objects", and of elements of G as "arrows".
The arrow y then points from rs(Y) to rt(y), and has an inverse y-I pointing in
the opposite direction. Arrows are composed from right to left; two arrows YI, Y2
can be composed to YI Yz iff the endpoint of Yz matches the starting point of YI.
The arrow t(q) connects q with itself, but it may not be the only arrow to do so.
The collection of all arrows connecting q with itself is the isotropy group

(3.5)

of q; this is clearly a group under the multiplication inherited from G. We conclude

from the preceding paragraph that t(q) is the unit element of the group Gq in the
usual sense.
An equivalent definition of a groupoid is obtained by saying that a groupoid
consists of a set G, a subset Gz C G x G, a map y f-+ Y -I from G to G, and a map
. ; G2 --+ G, such that: (i) Gz contains {(y, y-I) lYE G}; (ii) if (YI, yz) E Gz and
(yz, Y3) E G2, then (YI)'2, Y3) and (Yl, Y2Y3) are in Gz, and (YI Y2)Y3 = YI (YZY3);
(iii)(y-l)-l = y;(iv)if(YI, yz) E G z, then (YIYZ)Yz- 1 = YI andYI-I(YIYz) = yz.
The connection with Definition 3.1.1 is then established by identifying Q with
Go throught, which in tum is identified with the set{yy- I lYE G}. The projections
 3 Lie Groupoids and Lie Algebroids 271

it and is are given by it (y) = y Y -I and is (y) = Y -I y. The equivalence of the

two definitions is then easily checked.
We see that a group G = G is a special case of a groupoid, in which Q consists
of one point e, identified with its image l( e) E G, which is the unit in the group.
Since i,(y) = is(y) = e for all y E G, all elements can be multiplied, and the
group axioms follow from 3.1.1. At the opposite extreme we have
Definition 3.1.2. The pair groupoid Q x Q ~ Q is defined by the operations
is(ql, q2) := q2, i,(ql, q2) := q" t(q) := (q, q), (q/, q2)' (q2, q3) := (q/, q3), and
(q/, q2)-1 := (q2, ql).
This time Gq = (q, q) consists of one point only.
When H eGis closed under multiplication and inverses, we can form the
subgroupoid H(is(H), is, iI, t,', I); note that i,(H) = is(H). A subgroupoid of
the pair groupoid Q x Q that contains all identities is evidently the same as an
equivalence relation on Q; this follows from straightforward definition-chasing.
More generally, a groupoid G ~ Q gives rise to an equivalence relation on Q:
the map it x is : G -+ Q x Q is a morphism (in the obvious sense) of G into
the pair groupoid Q x Q ~ Q, whose image is a sub groupoid of the latter. By
3.1.1.3 one has it X is(t(q» = (q, q), so that the image contains all identities.
From the previous paragraph one therefore obtains an equivalence relation'" on
Q. This relation is simply that q '" q' iff there exists ayE G for which is (y) = q
and it(Y) = q'. Hence the equivalence class of q is the same as the orbit Gq of q
under G, if we define the latter as the set of all q' E Q for which there is ayE G
satisfying is(Y) = q and i,(y) = q'. If Gq = Q, then G is said to be transitive.
Proposition 3.1.3.
• A groupoid is the disjoint union of transitive subgroupoids.
• A transitive subgroupoid G ~ Q is isomorphic to a groupoid of the form
Q x H x Q ~ Q, where the group H is isomorphic to the isotropy group GqO
of an arbitrary point qo E Q.
Here the groupoid operations in Q x H x Q ~ Q are is(ql' h, q2) := q2,
i,(ql, h, q2) := ql, t(q) := (q, e, q), (q/, h, q2) . (q2, k, q3) := (ql, hk, q3), and
(ql, h, q2)-1 := (q2, h- I , q,).

When G is not transitive, each orbit Gq in Q gives rise to a transitive subgroupoid

Gq := is-I(Gq) = it-I(Gq). When G ~ Q is transitive, one chooses an arrow
yo(q): qo -+ q for each q, in terms of which (q/, h, q2) f-+ yo(qdhYo(q2)-1 is an
isomorphism between Q x GqO x Q and G. •
Definition 3.1.4. Let a group G act on a set Q. The action groupoid G x Q ~ Q
is defined by the operations is (x, q) := x -I q and i, (x , q) := q, so that the product
(x, q). (y, q') is defined when q' = x-1q. Then (x, q). (y, x-'q) := (xy, q). The
inclusion is t(q) := (e, q), andfor the inverse one has (x, q)-I := (X-I, x-1q).
Hence (x, q) is an arrow from x -I q to q. The isotropy group Gq coincides with
the usual isotropy group G q of the G -action.
272 III. Groups, Bundles, and Groupoids

Definition 3.1.5. A Lie groupoid is a groupoid G(Q, 1's , rt , t, " I), where G and
Q are manifolds, the maps 1's and 1't are surjective submersions, and multiplication
and inclusion are smooth.
Proposition 3.1.6. In a Lie groupoid:
1. The inclusion t is an immersion.
2. The inverse I is a diffeomorphism.
3. G2 is a closed submanifold ofG x G.
4. For each q E Q the fibers rs-I(q) and rt-l(q) are submanifolds ofG.
S. The isotropy group of any point q is a Lie group.
We omit the proof, which is a nontrivial exercise in differential geometry.
In the second definition of a groupoid given above one obtains a Lie groupoid
by requiring that G be a manifold and that inversion and multiplication be smooth.
A Lie group is clearly a Lie groupoid. When Q is a manifold, the pair groupoid
Q x Q ~ Q is a Lie groupoid. Thirdly, when the G ·action on Q is smooth, the
action groupoid G x Q ~ Q is a Lie groupoid.
A new example of a Lie groupoid may be constructed if one is given a principal
H -bundle. Recall that the equivalence class [x, y]H is defined by the equivalence
relation (x, y) ~ (xh, yh) for all hE H.
Definition 3.1.7. The gauge groupoid P XH P ~ Q of a principal bundle
P(Q, H, 1') is defined by the projections rs([x, y]H) := 1'(Y) and rt ([x , y]H) :=
rex), and the inclusion t(r(x» := [x, X]H. Accordingly, the multiplication
[x, y]H . [x', y']lI is defined when y and x' lie in the same fiber of P, in which case
[x', y']H = [y, Z]fl for some Z = y'h, h E H. Then [x, y]H . [y, Z]H := [x, Z]H.
Finally, the inverse is [x, y]]/ := [y, X]fl.
The isotropy group Gq consists of all [x, y] H for which r (x) = r (y) = q. Each
y E Gq is of the form [s(q)h, S(q)]H' where s is an arbitrary section of P. Hence
[s(q)hl, S(q)]H . [s(q)h2, S(q)]H = [s(q)hlh2, S(q)]II, so that for all q E Q one
concludes that Gq c:::: H as a group.
It is plain that a gauge groupoid is transitive. If the bundle P is trivial and brought
into the form P = Q x H, there is a smooth isomorphism P x H P ~ Q c::::
Q x H x Q ~ Q (see 3.1.3). This is given by [(ql, h), (q2, e)]H t--+ (ql, h, q2).
More generally, when P is nontrivial, a section s : Q --+ P leads to a nonsmooth
isomorphism of the above type through Proposition 3.1.3. For one may choose
yo(q) := [seq), S(qO)]H to obtain the isomorphism [s(ql)h, S(q2)]H t--+ (ql, h, q2).
We learn frorr Proposition 2.4.2 that the space of sections of the total space
P x II P of a gauge groupoid may be identified with the group of automorphisms
of the bundle P(Q, H, r).
Proposition 3.1.8. Let G (G / H, H, r) be the principal H -bundle defined in 2.7.
Then the gauge groupoid G x H G ~ G / H and the action groupoid G x (G / H) ~
G / H are isomorphic (in the obvious sense).
The correspondence [x, y]H ~ (xy-I, rex»~, combined with the identity map
on the base Q, is easily seen to provide the desired isomorphism. •
 3 Lie Groupoids and Lie Algebroids 273

3.2 Half-Densities on Lie Groupoids

As we have seen in 1.7. one can associate a C*-algebra C*(G) with a Lie group G.
and establish a correspondence between representations of G and representations
of C*(G). One can generalize this to Lie groupoids. The material in this section is
preparatory for this purpose.
For a vector bundle V over a manifold M with n-dimensional typical fiber V. the
bundle A(V) is defined as A nv minus the zero section. This is a principal C* -bundle
over M. whose fiber at x is the n-fold antisymmetric tensor product of Vx (the fiber
of Vat x). with 0 omitted (here C* is C\ {O}, seen as a multiplicative group). We
write AM := A(T M).

Definition 3.2.1. For a > O. the bundle of a-densities IA III (V) is the line bundle
over M associated to A(V)(M. C*. r) by the representation Z ~ Izl-Il o/C* on
C. An a-density on V is a section of the bundle ofa-densities. We put IA III M :=
IAIIl(T M).
A I-density is called simply a density; of interest to us are the cases a = I and
a = ~. According to Proposition 2.1.5. we describe a section qJll of IAIIl M as
a (smooth and compactly supported) map qJll : AM ---+ C satisfying qJll (AX) =
IAlllqJll(X) for all A E R and all X E AM. Such a section may be represented
by an equivalence class qJll = [f. v]ll. where f E C;;o(M) and v is a positive
measure on M that is locally Lebesgue. The equivalence relation defining the
class [f. v]1l is (f. v) '" (g. J1,) when g = f(dv/dJ1,)1l (the Radon-Nikodym
derivative d v/ d J1, is well-defined. since J1, and v are equivalent). It follows from
the multiplicative property of the Radon-Nikodym derivative that this is indeed
an equivalence relation. The section defined by the class [f. v]1l takes the form

[f. v]Il(iJ I A ... A an) := f(x) ( dv (X»)1l • (3.6)

dJ1,L
where al A ... A an E Ax M. and d J1,L := dx 1 ... dx nis the Lebesgue measure
in the particular chart used. The point of introducing densities is that they can be
integrated over M (even if it is not orientable) without specifying a measure, i.e.,
when qJ 1 is a density, the object

L L qJ1 := dnx qJl (A7=1 a~i) (3.7)

is independent of the choice of (local) coordinates; cf. the usual definition of the
integral of an n- form over M. In the realization of qJll as an equivalence class [f, v]1l
this simply reads fM[f. V]I = fM dVf. Similarly. since qJlll qJIl2 is an element of
r(IAl ll l+1l2M), the product of two half-densities may be integrated on M; here

L[f'V]~[g,J1,]~ = LdVJ~~fg= LdJ1,J::fg. (3.8)

Accordingly. one can form the Hilbert space of half-densities L 2(M).

274 III. Groups, Bundles, and Groupoids

Proposition 3.2.2. Each bundle IA la M is trivial, and any measure p, on M that is

locally Lebesgue defines a trivialization 1/1Il of IAla M. Accordingly, p, establishes
a natural bijection between r(IAl a M) and C~(M).

Since the notation of 2.1.4 would contain no information about a, we write

elements of IAl a M as equivalence classes [X, Ala (rather than [X, A]c.). In a
given local chart we define a local trivialization of IAl a M by

1/I1l([a l 1\ ... 1\ an, Ala) := (d:;"L )a A. (3.9)

By the C*-equivariance property [pX, Ala = [X,lpl-aAla this is actually

independent of the chart, and defines a global trivialization.
A section [f, vl a then trivializes to f: :
M --+ C, given by

f;(x) = f(x) ( : : (X») a (3.10)

Conversely, a function f : M --+ C corresponds to a section [f, P,la. •

When applicable, integration of such trivialized sections is then done with
respect to p" on which the numerical value of the integral does not depend.
Let now G ~ Q be a Lie groupoid. Recall from 3.1 that q '"" q' on Q when
there exists Y E G for which Ts(Y) = q and Tt(Y) = q'.

Lemma 3.2.3. The fibers G~ := Ts-I(Ts(Y» and G~ := Tt-l(rt(Y» are

diffeomorphic; their common dimension is denoted by d Y •
The fibers G~ and G~, (and similarly G~ and G~, ) are diffeomorphic when rs (y)
and rs(Y') are equivalent.

If q ~ q' by such a y, then R y , defind by

Ry(Y') := y'y (3.11)

when (y', y) E G 2, maps Ts-1(Tt(Y» into rs-l(rs(Y»; since y has an inverse, this
map is a diffeomorphism. Similarly, L y , defined by

Ly(Y') := yy' (3.12)

whenever (y, y') E G2, maps rt-I(rs(y» diffeomorphic ally to rl-I(r,(y». The
inversion I : y t-+ y-I is a diffeomorphism between Ts-l(rs(Y» and rl-I(rs(y»,
which is equal to T,-I(rl(y-I). Thus we have

Ry-l : G~ --+ G~_l;

Ly-l : G~ --+ G~_l;
I ·G
.
sy --+Gsy-l· (3.13)

Hence Ry 0 I : G~ --+ G~ is a diffeomorphism; its inverse is Ly 0 I : G~ --+ G~.

Since rl(Y-') = rs(Y) and Tt(Y) are equivalent points in Q (namely by y), the
last claim follows as well. •
 3 Lie Groupoids and Lie Algebroids 275

Lemma 3.2.4. For YI E Tt- 1{Ts{Y» there are canonical isomorphisms

Ty_,G S -I ~ Ty,G t ;
I 1'1
yI
Tyy, G~YI ~ TI'l G~I ;
TYYlG~y, ~ TyG~;
Ty_,G t -I ~ TyG ty . (3.14)
~ 1'1

These isomorphisms are given by the pushforwards of the inversion map, of

Ly-" of Ry,-" and of Lyy, , respectively. •
The space NG (or AtG) is defined as the line bundle over G whose fiber at Y
is the complexified d Y -fold antisymmetric tensor product of TyG~ (or TyG~); this
fiber is evidently one-dimensional. Note that TyG~ is the kernel of (Ts)*, etc. We
can then form the tensor product AS G ® AI G.
Definition 3.2.5. The symbol N®tG stands for the line bundle ASG ® AtG with
the zero section omitted; this is a principal C* -bundle over G.
An s ® t -density on G is a smooth compactly supported section ofthe line bundle
I
.JjXfs®tG associated to As®tG by the representation z 1-+ Izl-2' ofC*.
One may equivalently define .JjXfs®t G as .J1ATs ® .Jfi\f,
where .JjXfs.t is
the line bundle associated to As.tG, minus the zero section, seen as a principal
I
C*-bundle over G, by the representation z 1-+ Izl-2'.
Pulling back, Lemma 3.2.4 leads to an isomorphism

Al'l : M~y, G ® M~y, G ® M~,-, G ® M~,-, G

-+- JiAT~G®M:G® IAI~,Gy. (3.15)
Similarly, the pullback of the inversion map leads to an isomorphism

A; : M~-,G ® My-,G -+- M~G ® MyG. (3.16)

3.3 The Convolution Algebra of a Lie Groupoid

After this preparation we come to the definition of convolution.
Proposition 3.3.1. Let 'IIs®t, <l>s®t be s ® t-densities on G. Convolution on the
groupoid is defined by

Ws®t * <l>s®t(y) = 1 1',-1(1',(1'»

AI ('IIS®I<l>S®I), (3.17)

where AI ('IIS®I <l>S®I) is defined for YI E Tt-I{Ts(Y» by Al ('IIS®I <l>s®t) : YI 1-+

Al'l (ws®t{YYI) ® <l>s®t(YI-I».lnvolution is defined by

('IIs®t)*(y):= A;('IIs®t{y- I ». (3.18)

With these operations. the vector space r(../lXTs®tG) is a *-algebra.

276 III. Groups, Bundles, and Groupoids

The fact that r<Jii\TS®I G) is closed under convolution follows from 3.1.6.3,
the smoothness of groupoid multiplication (cf. 3.1.5), and the smoothness of the
isomorphism AI. Similarly, c10sedness under involution is a consequence of the
smoothness of inversion and of the isomorphism A- •
To verify associativity of the convolution product, one uses the associativity of
multiplication in the groupoid and a self-evident property of AI.
The equalities (\IIS®I)** = ",s®t and (\IIs®t ® <l>s®t)* = (CI>s®t)*(\IIs®t)* follow
from (3.2) and (3.3), respectively. •

Up to the maps Al and A-, there is therefore a direct correspondence between

the key properties of the groupoid operations (i.e., the associativity of multipli-
cation and the involutive nature of the inversion) and those in the corresponding
convolution *-algebra (where the role of the inversion is played by the adjoint).
Although it shows that one can always associate a convolution *-algebra
with a Lie groupoid, the definition above is not easy to use in practice. Fortu-
nately, one may trivialize the bundle .JfATs®t G. What follows adapts the general
considerations in 3.2 on measures and a-densities to groupoids.

Definition 3.3.2. A t-system on a Lie groupoid G ~ Q is afamily {J-t~}qEQ of

positive measures such that:

1. The measure J-t~ is defined on 7:t- 1(q) C G (or, equivalently, on G with support
in 7:t- l (q) C G).
2. Each J-t~ is locally Lebesgue (recall from 3.1.6 that each fiber 7:t- l (q) is a
manifold).
3. For each f E C'g'"(G) the map q r-+ fr,-I(q) dJ-t~(Y)f(Y)from Q to C is smooth.

Similarly, an s-system is defined as above, with t replaced by s.

If, in addition, the family of measures defining a t-system is invariant under
all maps L y , the t-system is called a left Haar system. Similarly, a right Haar
system is an s-system invariant under all maps R y •

It is clear from (3.13) that the inversion I maps a given t-system {J-t~} into an
associated s-system {J-t~} = {I*IL~}, by which a left Haar system is mapped into a
right Haar system (and vice versa). This is possible because of the diffeomorphism
G~ ~ G~ discussed in Lemma 3.2.3.

Proposition 3.3.3. Every Lie groupoid G ~ Q possesses a left Haar system.

Consequently, the bundle .JfATs®t G is trivial.

The proof of the first claim will be given at the end of 3.8. For the second, it
is enough to have a t-system. The construction of the global trivialization gen-
eralizes (3.9). For Y E G, choose local coordinates {Xi} and {yi} on G~ and
G~, respectively, with associated Lebesgue measures dlL~ := dxl ... dx n and
dJ-tL := dyl ... dyn. (These coordinates may be different even if G~ = G~.) The
 3 Lie Groupoids and Lie Algebroids 277

map

,I, ([!lX
'I'jt VI /\ .•. /\ vd
!lX!lY
® vI /\ ..• /\ vd
!lY
,/I.
'])
:= ,
/I.
dJ.Ll
- d- dJ.L~
- (3.19)
y y J.L' dJ.L'
is well-defined (in not depending on the representative in the equivalence class),
smooth, independent of the chart, and defines a global trivialization. •
One may represent a section \II.®I of .JjAf.®1 G by an equiValence class that
we denote by [j, {V~}l!®l' where f E Cgo(G) and {v~} is a t-system on G. With
2 2
{v~} the s-system associated to the given t-system, the equivalence relation is
(i, {v~}) ,...., (g, {J.L~}) when for all Y E G one has

g(y) = f(y) (dV~(y) dV:,(y) (y»)!

dJ.Lr.ry) dlLr,(y)
a:
For af /\ ... y E A~®I G, the section in question is then duly given as a C*-
equivariant map from A·®IG to C by (cf. (3.6»

[f {v'}] 1 1 (aX
, q i®i I
/\ ... /\ axd y
®
dV S dv'
aIY /\ ••• /\ adY ) :=
f(y) ( ~~(y)
dJ.Ll dJ.L~
y
)!
(3.20)
As in (3.10), a fixed t -system {v~} on G leads to a trivialization of s ® t -densities
as complex-valued functions on G, tied to the trivialization of .JjAf.®1 G defined
by {v~}. Generalizing (3.10), a section [j, {v~}]!®! as above is trivialized by
2 2
f~ : G ~ C given by

(3.21)

Proposition 3.3.4. A left Haar system {J.L~} on a Lie groupoid G ~ Q defines an

isomorphism between the convolution *-algebra r<.JIAT.®1 G) and ego (G), turned
into a *-algebra asfollows. Convolution and involution are defined as

f * g(y) := 1
r,-l(r,(y)
dJ.L~,(y)(YI) f(YYI)g(YI-I); (3.22)

f*(y):= f(y- I ). (3.23)

The isomorphism is given by letting f E Cgo(G) correspond to the equivalence
class [j, {JL~}l!®! in r<.JIAT.®I G). The fact that (3.22) is thus mapped into (3.17)
2 2
follows from the definition of AYl in (3.15) by Lemma 3.2.4, and the comment
following 3.3.2. Similarly, the correspondence between (3.23) and (3.18) follows
from the definition of A-in (3.16) and the fact that the right Haar system occurring
in, e.g., (3.21) is defined from the given left Haar system by the inversion map. •
278 III. Groups, Bundles, and Groupoids

Let us look at some examples of the preceding abstract constructions.

Proposition 3.3.5. ffG = G is a Lie group, then each locally Lebesgue measure
defines a t-system, which in this case is the same as an s-system. A left-invariant
Haar measure on G provides a left Haar system. The ensuing convolution algebra
is the group algebra, restricted to C~(G).

This is obvious, as G = <t-I(e) = <s-I(e). A measure on G defines a left Haar

system iff it is left-invariant in the usual sense. •

Note that the right Haar system defined by a left Haar system coincides with it
when G is unimodular.
Combined with 3.3.4, this proposition shows that one may define the group
algebra of a Lie group without specifying a Haar measure, but since one still needs
the isomorphism (3.15) in (3.17), there is not much advantage in this.

Proposition 3.3.6. On a pair Lie groupoid Q x Q ~ Q any measure v on Q

that is locally Lebesgue defines a left Haar system.
The corresponding pair groupoid *-algebra is C~(Q x Q), with operations

f * g(ql, q2) = Ldv(q) f(q" q)g(q, q2); (3.24)

j*(q" q2) = f(q2, qd. (3.25)

To construct the left Haar system one identifies <,-1 (q) = {q} x Q with Q for
each q. The above formulae then follow from (3.22), (3.23), and 3.1.2. •

In particular, when Q is a finite set with cardinality n, the convolution algebra

is simply Vltn (C).

Proposition 3.3.7. On a gauge groupoid P x H P ~ Q an H -invariant measure

/L on P which is locally Lebesgue produces a left Haar system.
The corresponding gauge groupoid *-algebra C~(P x H P) is given by

f * g([x, y]H) = 1 d/L(z) f([x, Z]H )g([z, y]H); (3.26)

f*([x, y]H) = f([y, X]lI)' (3.27)

When H is compact one may identify C~(P Xli P) with C~(P x p)H, seen
as a subalgebra of the convolution algebra C~(P x P) of the pair Lie groupoid
P x P ~ P; cf. 3.3.6.

We identify <t-I(q) with P through the choice of a measurable section s : Q ~

P; that is, we let [seq), X]H E <t-I(q) correspond to x E P. Since [x, S(q)]H .
[seq), y]H = [s«(x», yhs(x)]H, where hs is defined below (2.6), the first claim
follows. The second follows from the above consideration and the H -invariance
of the measure. •
 3 Lie Groupoids and Lie Algebroids 279

3.4 Action *-Algebras

We now look at the convolution algebra of an action Lie groupoid.
Proposition 3.4.1. In an action Lie groupoid G x Q ~ Q each left-invariant
Haar measure dx on G leads to a left Haar system. The operations in the action
groupoid *-algebra C;:O(G x Q) are

f*g(x,q)= fa dy f(xy,q)g(y-', y-Ix-I q ); (3.28)

f*(x,q) = f(x-',x-'q). (3.29)

To construct the left Haar system one identifies <t- I (q) with G, letting (x, q) E
<t-I(q) correspond tox E G. Equations (3.28) and (3.29) then follow from (3.22),
(3.23), and Definition 3.1.4. •
We will now construct a seemingly different *-algebraic structure on the function
space C~(G x Q), which turns out to be isomorphic to the one above. The data
for the definition still constitute a smooth action of a Lie group on a manifold Q;
the associated action of G on C;:O(Q) is denoted by
- - -I
ax(f) :q H- f(x q). (3.30)
We look at C;:O( Q) as a commutative *-algebra in the obvious way, i.e., as a dense
subalgebra of the C* -algebra Co(Q).
To exhibit the natural structure of the construction, we first consider the
following generalization. Recall Definition 11.2.4.2.
Definition 3.4.2. A smooth C* -dynamical system consists of a Lie group G, a
dense subalgebra !it of some C* -algebra ~, and an automorphic action a of G on
2(, such that for each fixed A E !it the function x H- ax (A) from G to 2( is smooth.

It follows that x H- ax(A) is continuous for all A E 2(. (More generally, when G
is merely locally compact and the latter continuity property is satisfied for !it = 2(,
one speaks of a C* -dynamical system.) The term "dynamical system" comes from
the example G = Rand 2( = Co(S), where R acts on S andat(f) : (1 H- (1 (t); cf.
1.(2.13). Anotherexarnple is, of course, provided by 2( = Co(Q) with !it = C;:O( Q),
where the G-action on 2( is defined as in (3.30). The smoothness of the G-action
on Q then implies that one indeed has a smooth C* -dynamical system. In any case,
given a smooth C*-dynamical system, one considers the space C~(G, !it), made
into a *-algebra by the operations

F * G(x):= fa dy F(y)ay(G(y-l x »; (3.31)

F*(x):= ax(F(x- I )*). (3.32)

In the nonunimodular case one here needs to assume that the Haar measure is left-
invariant. The *-algebra C~(G,!it) thus defined is called the (smooth) crossed
product *-algebra, or simply the crossed product, of G and !it, and is denoted
by C*(G, !it). If!it = C~(Q), we call C*(G, C~(Q» an action *-algebra.
280 III. Groups, Bundles, and Groupoids

Lemma 3.4.3. The action groupoid *-algebra C;;o (G x Q) is isomorphic to the

action *-algebra C*(G, C;;o(Q».
A function FE C*(G, C;;o(Q» defines f E C;;o(G x Q)by f(x, q) = F(x)(q).
The definition of the function spaces in question easily implies that this correspon-
dence is bijective. Under this correspondence, a shift in the integration variable in
(3.28) and the left invariance of the Haar measure reproduces (3.31). The equality
of (3.29) and (3.32) is immediate. •
Generalizing 1.7.3, we have
Theorem 3.4.4. There is a bijective correspondence between
• nondegenerate representations rr of the crossed product C*(G, it) that are
bounded as in

IIrr(F)1I :::: IIFIlI := £

dx IIF(x)lI; (3.33)

• pairs (U, ir), where U is a representation of G, and ir is a nondegenerate

representation of ~ that for all x E G and A E ~ satisfies the covariance
condition
U(x)ir(A)U(x)* = ir(ax(A». (3.34)
This correspondence is given in one direction by

rr(F) = £ dx ir(F(x»U(x); (3.35)

in the other direction one defines AF : x t-+ AF(x) and ax(f) y t-+
aAf(x- 1y», and puts
U(x)rr(F)Q = rr(aAF»Q; (3.36)
ir(A)rr(F)Q = rr(AF)Q, (3.37)
where Q is a cyclic vector for a cyclic summand ofrr(C*(G, it».
The proof of this theorem is analogous to that of 1.7.3. The analogue of the
Banach algebra L I(G, c) used in that proof is LI(G, ~), the closure of C;;o(G, it)
in the norm (3.33). The rest of the proof may be read off from 1.7.3. 0
We return to G-actions on a manifold Q.
Definition 3.4.5. Given a G-action on Q, a smooth system ofimprimitivity of
G on Q is a pair (U(G), ir(C;;o(Q»), where U is a continuous representation of
G, and ir is a nondegenerate representation ofC;;o(Q) (seen as a commutative
*-algebra in the obvious way), satisfying the covariance condition
U(x)ir(j)U(x)* = ir(aAi». (3.38)
The meaning of the conditions on the pair (U, ir) may be clarified by expressing
them in infinitesimal form. For X E g, i E C;;o(Q) we put
Q~(X):= ihdU(X); (3.39)
 3 Lie Groupoids and Lie Algebroids 281

n
Q (j) := it(j), (3.40)

and obtain (on the domain H(J; see III.l.S.S)

~[Qn(j), QnCg)] = 0; (3.41)

i - - -----
h[Qn(X), Q~(Y)] = Qh(-[X, Y]); (3.42)

i - - -
h[Q~(X), Q~(f)] = Qn(hf). (3.43)

As in 1.1 we have X E COO(g*, JR); also recall the definition (1.19) of gx. Equation
(3.41) is evident from the fact that it (Cgo(Q» is a representation (but is a weaker
property); equation (3.42) is equivalent to (1.70) (with r = 0); and finally, (3.43) is
an infinitesimal restatement of the covariance condition (3.38). These commutation
relations may be seen as a version of Dirac's condition 11.(1.3); cf. 3.11.

Corollary 3.4.6. There is a bijective correspondence between

• nondegenerate representations 1f of the action *-algebra C*(G, Cgo(Q» (or,

equivalently, of the action groupoid *-algebra Cgo(G x Q» that are bounded
as in

1I1f(F)1I ::: IIFIII = sup { dx If(x, q)l; (3.44)

qEQJO

• smooth systems of imprimitivity of G on Q.

Further to 3.1.8 we naturally have

Proposition 3.4.7. The action groupoid *-algebra Cgo( G x (G j H» and the gauge
groupoid *-algebra Cgo(G XH G) are isomorphic.

For simplicity we prove this only for unimodular G. We identify COO( G x H G)

with COO(G x G)H in the obvious way, so that Cgo(G XH G) is identified with a
certain subspace of COO(G x G)H. One then establishes the desired bijection f ++
i between Cgo(G x (Gj H» and Cgo(G XH G) by i([x, y]H) = f(xy-I, rex»~,
with inverse I(x, q) = i([s(q), x-1S(q)]H)' Here s : G j H -+ G is an arbitrary
section, on which the right-hand side clearly does not depend. Using the invariance
of the Haar measure under inversion and right translation, one verifies that this
bijection duly intertwines the *-algebraic operations stated in 3.4.1 and 3.3.7. •

Corollary 3.4.8. There is a bijective correspondence between nondegenerate rep-

resentations 1f ofthe gauge groupoid *-algebra Cgo( G x H G) that (for unimodular
G) are bounded as in

1I1f(f)1I ::: sup ( dx If([y, X]H )1, (3.45)

YEOJO

and smooth systems ofimprimitivity (U(G), it (Cgo(G j H»).

282 III. Groups, Bundles, and Groupoids

U sing the invariance of the Haar measure under inversion, it is easy to see that
the bound (3.45) is equivalent to (3.44). •
Using (3.27) and thefact that rr by definition satisfies rr(f*) = rr(f)*, one infers
that the bound (3.45) is equivalent to the same expression in which f([y, X]H) is
replaced by f([x, Y]H). The significance of this comment will become clear in the
light of (3.57). See also Corollary 3.7.6.

3.5 Representations of Groupoids

In 1.7 the (reduced or full) C* -algebra of a Lie group G was defined in direct
relationship with the representations of G. This motivates the following definition
of a representation of a general groupoid.
Definition 3.5.1. A representation U of a groupoid G =*
Q consists ofa family
{1t}qEQ of Hilbert spaces indexed by the base Q and a collection (U(Y)}YEG of
maps such that
1. U (y) : 1t r,(y) --* 1t r,(y) is unitary.
2. U(YIY2) = U(Yl)U(Y2) whenever (YI, Y2) E G2 •
3. U(y-l) = U(y)* for all Y E G.
Hence, with reference to the third paragraph after (3.4), one thinks of a repre-
sentation of G ~ Q as a functor converting the points of Q into Hilbert spaces
and the arrows in G into unitary maps connecting these Hilbert spaces.
The simplest example of a representation is obtained by choosing 1tq = C for
all q E Q and U(y) := 1 for all Y E G. This representation is evidently not
faithful if there are nontrivial isotropy groups Gq •
For all our applications, and also for the purpose of defining the groupoid C*-
algebra, it will be sufficient to assume that all 1tq are separable and of the same
dimension. Let us give an example. Since each isotropy group Gq is a Lie group,
we can form 1tq := L2(Gq, JL%), where JL% is a left or right Haar measure on Gq.
For W E 1t r,(y) we then define Ulr(Y)W E 1t r,(y) by

(3.46)

The Radon-Nikodym derivative occurring here makes sense, since Gr,(y) and Gr,(y)
are diffeomorphic by h 1-+ Y -I h Y , so that the measure class of the Haar measure
is preserved (recall that a Haar measure on a Lie group is locally Lebesgue). It is
easily verified that this indeed furnishes a representation of G in the sense of 3.5.1.
In a variation on this example, in the base of each transitive sub groupoid one
picks a point qo of G (cf. the proof of 3.1.3), and a function yo: Gqo :--* 1's-l(qO)
satisfying 1'. (Yo(q» = qo and 1'1 (Yo(q» = q. One then replaces (3.46) by

dJL:'(y)(y-1 hyo(1'1 (y »Yo( 1's(Y »-1 )

dJL~(y)(h)
 3 Lie Groupoids and Lie Algebroids 283

Since it depends on the function Yo, this representation is not really intrinsic (unlike
its predecessor), but different choices of Yo lead to equivalent versions. In any case,
it will be reconsidered shortly.
In the regular representation one takes 1tq := L 2 (T,-'(q), JL~). Defined for a
general Lie groupoid, this representation is given by
(3.48)
This makes sense, since y-'y' E T,-t(Ts(Y» when T,(y) = Tr(y'), and is unitary
because of the left invariance of the left Haar measure.
To relate representations of G ~ Q in the above sense to representations of
the convolution algebra C~(G) by a *-algebra of bounded operator.s on a Hilbert
space in the usual sense, one has to choose a measure v on Q. As we shall see, it
is not sufficient to limit one's attention to measures that are locally Lebesgue.
Definition 3.5.2. Let a Borel measure von Q and a left Haar system {JL~}qEQ on
G, with associated right Haar system {JL~}, be given. One obtains measures v x JL'
and v x JLs on G, defined by

v X JL'(f):= ( dv(q) j dJL~(Y) fey); (3.49)

lQ <,-l(q)

VXJLS(f):= (dV(q)j dJL~(Y)f(Y). (3.50)

lQ <,-l(q)

The measure v is said to be quasi-invariant when v x 11' and v x f.Ls are equivalent,
and invariant when they are equal.
If v is quasi-invariant, the Radon-Nikodym derivative
dv x JLs
p.- ---'-- (3.51)
.- dv x f.L'
is well-defined on G, equaling unity in the invariant case. An example of a quasi-
invariant measure on the base of an arbitrary Lie groupoid is a measure that is
supported and locally Lebesgue on an arbitrary orbit Gq • A measure that is locally
Lebesgue on Q (so that it is supported on all of Q) is quasi-invariant as well, since
v x JL' and v x JLs are both locally Lebesgue on G. More examples are given in the
following proposition, whose main goal it is to examine when a quasi-invariant
measure is invariant. Here the measure v on Q is as specified above.
Proposition 3.5.3.
• A left Haar measure on a Lie group G is invariant iffG is unimodular.
• For a pair Lie groupoid Q x Q ~ Q, a measure is invariant iff it is a multiple
of the fixed measure on Q defining the left Haar system (cf 3.3.6).
• For a gauge groupoid P x H P ~ Q, a measure v is invariant when H is
unimodular, and when v is related to JL (the measure defining the left H aar
system, as in 3.3.7) by (2.139).
284 III. Groups, Bundles, and Groupoids

• For an action Lie groupoid G x Q =* Q, a measure v is invariant iff it is

G-invariant.
This follows directly from the definitions.
If in a given representation U of G =*
Q one has 1tq = K for all q, one can
•
form L 2( Q, v) ® K, which will be the Hilbert space carrying the representation :rr
of the convolution algebra C;?,"(G) associated to U.
In general, assuming that each 1tq is separable, the subset Qn ~ Q for which
dim('Hq) = n for all q E Qn should be measurable for each n E N U 00. One then
considers the space r of all functions '11 : Q -+ {'H q } satisfying '11 q E 1tq ; such a
function is called a section of the field {1t q }. The idea is to tum a suitable subset
of r into a Hilbert space with inner product

('11, <1»:= 10 dv(q) ('I1q, <l>q)q, (3.52)

where (, )q is the inner product in 1tq •

Such a suitable subset is obtained by specifying a sequence of sections 'I1n
satisfying the two conditions that firstly the function q 1-+ ('I1 n(q), 'I1m(q»q be
measurable for all n, m, and secondly that for each fixed q the 'I1n span 1tq • There
then exists a unique maximal linear subspace roof r that contains {'11n }, and for
which all functions q 1-+ ('I1 q , <l>q)q are measurable. The direct integral

1t = 10$ dv(q)1tq (3.53)

is then by definition the subset of ro of functions '11 for which ('11, '11) < 00, as
defined by (3.52). It depends on the choice of the sequence {'I1 n}, but in all practical
applications it is clear that all reasonable such choices lead to the same result, so
that this dependence will be suppressed.
The simplest example of a direct integral is 'H = fIR dx'H x , where 1tx = C for
all x. Choosing the sequence {'I1n} to consist of a single strictly positive measurable
function then leads to 1t = L2(lR). When 1tq = K for all q, one takes {'I1n} to
be a strictly positive measurable function on Q tensored with a basis in K. The
corresponding direct integral is nothing but 1t = L2(Q, v) ® K. Since Q is a
finite-dimensional manifold, 'H is separable.
As an example relevant to Lie groupoids, we construct the direct integral

'H(qo) = ($ dv(q) L 2(Gq , JL~/) (3.54)

lGqo

over the Hilbert spaces 1tq = L2(Gq , JL%) considered in the context of (3.46),
with v supported on a given orbit Gqo. One picks the function Yo mentioned
before (3.47) so that it is measurable, and identifies each stability group Gq with
G qO through Gq 3 h 1-+ yo(q)-lhYo(q) E G qo • Inserting the appropriate Radon-
Nikodym derivative, this leads to a unitary map Vq : L2(G qO ' JL~) -+ 1tq . One
then picks a basis {en} in L2(G qo ) and a strictly positive measurable function f
on Q, and defines the sequence {'I1n} as 'I1 n(q, h) := f(q)VqeJL(h). The resulting
 3 Lie Groupoids and Lie Algebroids 285

direct integral clearly consists of those sections '" of the given field for which
q ~ V;I"'q lies in L2(GqO x Gqo ).
In a third example, we consider the field 1tq = L 2 (.,-I(q), IL~) featured in
(3.48), and a measure v on Q that is locally Lebesgue. One then tacitly chooses
the sequence {'lin} in such a way that (cf. (3.49»

10$ dv(q) L 2(.,-I(q), IL~) = L 2(G, v x IL'). (3.55)

3.6 The C* -Algebra of a Lie Groupoid

In preparation for the definition of the C* -algebra of a Lie groupoid we show
how a representation of a Lie groupoid determines a representation of its groupoid
*-algebra.
Proposition 3.6.1. Assume that one has a Lie groupoid G ~ Q with
• a left Haar system {IL~};
• a representation U of G on a collection {1tq} of separable Hilbert spaces 1t
for which the function
y ~ ('IIl,(y), U(y)'IIr,(y)hiT,(y)
is measurable for all'll, <l> E 1t;
• a quasi-invariant measure v on Q;
• an associated direct integral1t = f~ d v(q) 1tq.
For each f E C~(G) the operator 1t(f) on 1t, defined by

1t(f)'IIq := ! l,-L(q)
dlL~(Y) #(y)f(y)U(y)'IIr,(y), (3.56)

where p is given by (3.51), is bounded, with

111t(f)1I ~ IIflll := max{lIflls, IIfII,}; (3.57)

IIflls., := sup!
qEQ l,~,L(q)
dlL~·'(Y)lf(Y)I. (3.58)

Then 1t is a nondegenerate representation of the groupoid *-algebra C~(G) on

1t.
To derive the bound, we generalize the proof of Lemma 1.7.2. Writing F(y) :=
J p(y)lf(y)III'IIr,(y)1I and G(y) := Jlf(y)III"'r,(y)lI, we use 3.5.1.1 to majorize
1('11, 1t(f)'II)1 by fGdv x IL' FG. Applying the Cauchy-Schwarz inequality and
using the argument in the proof ofIl.1.3.5 then leads to (3.57).
*
To verify that 1t preserves multiplication, one writes out 1t(f g) using (3.56)
and (3.22), changes variables YI ~ Y2 = YI-I and subsequently y ~ y' = yY2- 1•
The range of(y',)I2) is then y' E .,-l(q) and Y2 E .,-1 ('s (y'». Finally, one needs
3.5.1.2 and the properties
p(yy') = p(y)p(y');
286 III. Groups, Bundles, and Groupoids

(3.59)
The first is a nontrivial consequence of the left invariance of {JL~} and the right
invariance of {JL~}, and the second is immediate from the definition of {JL~}.
The proof that rr preserves the involution follows from 3.5.1.3 and the fact that
the measure v x JL',JP on G is invariant under inversion.
The nondegeneracy of rr is an easy consequence of the surjectivity of is,1 and
the unitarity of U. •
An important example of such a representation is the regular one.
Theorem 3.6.2.
1. The regular representation rri of C~(G) on H~ = L2(G, v x JL ' ), given
by (3.48), (3.56), and a locally Lebesgue measure v on Q, is equivalent to
rrl<C~(G» on Hi := L2(G, v x JL S), given by
(3.60)
Here the convolution is given by (3.22).
2. The regular representation is faithful on C~(G). It may be decomposed as a
direct integral over the reduced regular representations rr~, q E Q, defined
*
on L2(is-l(q), JL~) by rr~(f)\II = f \II, that is, by the restriction of(3.60).
3. For given qo E Q, the representation rr;o(C~(G» is equivalent to the
representation rrqO on H(qo) (see (3.54)) given by (3.56) with (3.47).
4. Finally, rr;o may be realized on the Hilbert space L 2( i s- I(qo» of half-densities
on i s- I(qo), so that it can be defined without the choice of a (left or right) Haar
system.
The representation rri is given on H~ = L2(G, v x JLI) (cf. (3.55» by

rri(f)\II(y) = ( dJL~,(y)(Yl),JP(Ydf(YI)'lI(YI-ly). (3.61)

JG~
Now perform the unitary transfonnation V : H~ ~ Hi := L2(G, v x JL S),
1
defined by V'lI(y) := p(y )-2 'lI(y). Using (3.59), changing integration variables,
and using the left invariance of JLI one obtains Vrri V* = rr[; cf. (3.60).
(Incidentally, the price for the simplicity of (3.60) is that compared with (3.48)
the corresponding representation U[ (G) on L2(is-l(q), JL~) now contains an addi-
tional Radon-Nikodym derivative.) In analogy to (3.55), one decomposes Hi as
a direct integral over Q by

L 2(G, v x JLS) = 1a(JJ dv(q) L 2(is- l (q), JL~); (3.62)

rr[ = 1a(JJ dv(q)rr~. (3.63)

The latter means that rr[(f)'lIq = rr~(f)'lIq for (almost) all q E Q, where rr~ is
the operator on Hq = L2(is-l(q), JL~) specified in the theorem.
 3 Lie Groupoids and Lie Algebroids 287

To prove the equivalence between 7r~o and 7rqO ' one uses 3.1.3 to realize both
representations in the model Gqo x GqO x Gqo, where they coincide.
Finally, analogously to (3.15), we can use (3.14) to obtain an isomorphism

(3.64)

In a slight modification of (3.17), if III is a half-density on 's-I (qo) and <l>s®t is an

s ® t-density on G, we can write convolution as

<l>s®t * lI1(y) = 1.,-l(.,(y»

XI(<I>s®t<l», (3.65)

where the definition of XI in terms of XYl is analogous to that of Al in terms of Ai'"

The left hand side is then by definition equal to 7r;(<I>s®t)lI1. •

One should mention that 7r: mayor may not be reducible, and that these
representations mayor may not be equivalent when q varies.
For example, for a group there only is one q, and 7r% is the left-regular
representation, which is reducible.
For a pair groupoid Q x Q ~ Q the reduced regular representation 7r:omay
be realized on L2(Q) (defined with respect to a locally Lebesgue measure v). For
f E C~(Q x Q) it takes the form

7rs(f)II1(q) = fo dv(q')f(q, q')II1(q'). (3.66)

We have written tr s for 7r:o ' since this representation does not depend on qQ. We
see from Proposition 3.3.6 that 7r S is faithful and irreducible; the representations
7r: are trivially equivalent for all q E Q.
For an action groupoid G x Q ~ Q all possibilities may occur, depending
on the group action. For unimodular G the regular representation is realized on
L 2(G x Q). If, for simplicity, we assume that Q has a G-invariant locally Lebesgue
measure, and define L2(Q) accordingly, we have

7ri(f)II1(x, q) = 1 dy f(xy, q)lI1(y-l, y-Ix-I q ). (3.67)

By Corollary 3.4.6 there are representations Ui<G) and ii'i<C~(Q» associated to

7ri (C~(G x Q». We infer from (3.36) that

~=~®~, ~~
where UL is defined on L2(G) by (1.83) (with c = 1), and uF is (analogously)
defined on L2(Q) by UF(x)lI1(q) := lI1(x- 1q). From (3.37) we see that

ii'i(j)II1(x, q) = J(q)lI1(x, q). (3.69)

To reduce this we perform a unitary transformation U on L2(G x Q), defined
by VII1(x, q) := 111 (x , xq); this step is necessary because x and q are mixed up in
288 III. Groups, Bundles, and Groupoids

'S-I(q), and hence in (3.68). Then Jrl := VJrf V-I is given by

JrI(f)\II(x, q) = 10 dy f(xy-I, xq)\II(y, q), (3.70)

with associated representations

ul = U L ® ll; (3.71)
iil(j)\II(x, q) = j(xq)\II(x, q). (3.72)

This evidently reduces as a direct integral over q; the representation Jr~o is simply
obtained by fixing q = qo in (3.70). One sees that Jr~o and Jr~l are inequivalent
when qo and ql lie in orbits of different type.
We are now in a position to paraphrase Definitions 1.7.1 and 1.7.4.
Definition 3.6.3. The reduced groupoid C*-algebra q(G) of a Lie groupoid
G ~ Q is the completion ofC~(G) in the norm
IIfll, := IIJrL(f)1I = sup IIJrq(f)II. (3.73)
qeQ

Here Jr L stands for any of the realizations of the regular representation discussed
in 3.6.2; likewise for Jrq.
The groupoid C*-algebra C*(G) of a groupoid G is the closure of C~(G) ill
the norm

IIfII:= IIJru(f)1I = sup II Jr (f) II , (3.74)

where Jr u is the direct sum of all nondegenerate bounded representations Jr of

C~(G) satisfying (3.57).

In the definition of C*(G) the bound (3.57), which depends on the choice of a left
Haar system, may be replaced by an appropriate intrinsic continuity condition. In
the presence of a given Haar system one obviously has the inequalities (cf. (3.57»

1Ifll, ~ IIfII ~ IIfll/. (3.75)

3.7 Examples olLie Groupoid C*-Algebras

We will now determine the structure of the C* -algebras of some of the Lie
groupoids we have been looking at so far.
Theorem 3.7.1.
• When G = G is a unimodular Lie group, the (reduced) groupoid C*-algebra
ct)(G) coincides with the (reduced) group C*-algebra defined in 1.7.1 alld
1.7.4.
• For a pair Lie groupoid Q x Q ~ Q one has
C*(Q x Q) ~ C;(Q x Q) ~ !B o(L2(Q». (3.76)
 3 Lie Groupoids and Lie Algebroids 289

• For a gauge groupoid P x H P ~ Q each measurable section s : Q ~ P

determines isomorphisms
C;(P x H P) ~ 23o(L 2(Q» ® C;(H); (3.77)
C*(P x H P) ~ 23 o(L 2(Q» ® C*(H). (3.78)
In particular, lor compact H one has C*(P XH P) ~ ~0(L2(p»H.
Note that there is no ambiguity in the definition of the tensor product, as one of
the factors is 2300t).
For a Lie group II I III = II I III, so the first claim is obvious from 1.7.4.
The reduced regular representation JrS of C~(Q x Q) is faithful; cf. (3.66). The
norm-closure ofJrS(C~(Q x Q» is 230(L2(Q), so that one infers (3.76) for C;.
To prove the same result for the full groupoid C* -algebra, one needs to show that
every nondegenerate bounded r-=presentation Jr of C~(Q x Q) on some Hilbert
space 'It is a multiple of JrS on L 2(Q). This can be done by a method whose
significance will emerge at the end of IV.2.4.
Given (Jr, 'It), we equip the algebraic tensor product Cgo(Q) ® 'It with a
sesquilinear form (, )0, defined by linear extension of
(g ® \II, I ® c:J»0 := (\II, Jr(f x g)c:J>ht, (3.79)
where I x g(q, q') := I(q)g(q'). This form is easily seen to be positive semidef-
inite; if No is its null space, the completion of (Cgo(Q) ® 'It)/No in the inherited
inner product is a Hilbert space, denoted by 'ito. Subsequently, define a linear map
V : C~(Q) ® 'Ito ~ 'It by linear extension of
VI ® [g ® \II] := Jr(f x g)\II. (3.80)
Here [g ® \II] E 'ito is the image of g ® \II E Cgo(Q) ® 'It under the canonical
projection. The map V is well-defined: Firstly, if [g®\II] = 0 thenJr(f xg)\II = 0,
as can be checked using (3.79). Secondly, Jr is bounded, so that the right-hand side
exists for all \II E 'It, and accordingly for all [g ® \II] E 'ito. Moreover, using (3.79)
and the property Jr(f x g)* = Jr(g x 7), as well as (3.24) and (3.25), one verifies
that V satisfies
(3.81)
for all 0 1 , 02 E C~(Q) ® 'ito. Since in addition, the image of V is dense in 'It as
a consequence of the nondegeneracy of Jr , it follows that V can be extended to a
unitary map from L 2(Q) ® 'Ito to 'It, which we call V as well. The point is that V
intertwines Jrs ® nand Jr in that V Jrs (f) ® n= Jr (f)V, as is trivially verified from
(3.80). One concludes that Jr is equivalent to the direct sum of dim('lt°) copies of
Jrs. This implies (3.76).
We now come to the last claim of the theorem, which obviously generalizes
Proposition 2.8.2. The section s establishes an isomorphism 's-l(qO) ~ P by
identifying [x, S(qO)]H E 's-l(qO) with x E P. Hence, using Proposition 3.5.3,
we have L 2('s-l(qO» = L 2(P, JL) for some fixed H-invariant locally Lebesgue
measure JL on P. The reduced regular representation Jrr := Jr~o onL 2 (p, JL) depends
290 III. Groups, Bundles, and Groupoids

neither on qo nor on s. It is faithful, and reads

n'r(f)\I1(x) = i dJ,L(Y) f([x, Y1H)\I1(y). (3.82)

As in the proof of 2.8.2, we in addition uses to map L 2(P) into L2(Q)®L2(H)

(in case H is not unimodular, one should note that the measure dh occurring in
(2.139) is a right Haar measure, whereas the one used to define L2(H) is a left
Haar measure J,Ld. One then obtains

Us7T,U)Us*\I1(q, h) = fa dv(q') L dJ,LLCk)fs(q, k, q')\I1(q', k-1h), (3.83)

where fs(q, k, q') := f([s(q)k, S(q')1H). Hence U s7T,U; factorizes into the prod-
uct of the defining representation (3.66) in the argument (q, q') E Q x Q and the
left-regular representation TTL in the argument h E H; cf. (1.84). With Definition
1.7.1 this immediately implies the isomorphism (3.77).
Let C~(Q x H x Q) be the image of C~(P XH P) under the map f ~ fs, and
define C~(Q) as the space of functions on Q of the type f(q) = fP(s(q», where
fP E C~(P); the space C~(Q x Q) is defined similarly.
Rather than with C:(Q x H x Q), we may work with its subspace C:(Q x
Q) ® C~(H). This is justified by the fact that the two spaces in question have the
same closure in the norm II . III (cf. (3.57». In particular, C:(Q x H x Q) and
C~(Q x Q) ® C~(H) have the same closure in the C* -norm (3.74).
Transferring the *-algebraic operations from P x H P, one sees that as a *-algebra
C:(Q x Q) ® C~(H) is the direct product of C~(Q x Q) (with operations (3.24)
and (3.25» and C~(H) (with operations (1.80) and (1.81), in which c = 1). This
reflects the fact that s leads to an isomorphism P x H P ~ Q x H x Q as groupoids
over Q, as explained after Definition 3.1.7.
The argument used to prove (3.76) works equally well when C~(Q) is replaced
by C~(Q). Combining this with Theorem 1.7.3 and the above factorization, one is
led to (3.78). The final claim then follows from Proposition 2.8.2 or 3.3.7. •
Recall 2.8.4. Generalizing 2.8.3 and 2.8.5, we have
Corollary 3.7.2. Up to equivalence there is a bijective correspondence between
the nondegenerate representations 7T x ofC*(P x H P) and the representations U x
of H. Here 7T x is realized on 1i x by

(3.84)

The representation 7TX(C*(P XH P» is irreducible ijJUx(H) is irreducible.

Choosing a section s : Q --+ P and following the proof of (3.78) above, one
sees, by first restricting to C~(Q x Q) ® C~(H), that
7TX(C*(P x H P» ~ ~o(L 2(Q» ® 7T x (C*(H».

Then use Corollaries 1.7.5 and 1.2.2.6.

•
 3 Lie Groupoids and Lie Aigebroids 291

The C*-algebra of an action groupoid G x Q ~ Q is called an action C*-

algebra, denoted by C*(G, Q) (which is short for C*(G, Co(Q»). As a slight
variation on Definition 3.4.5 we put
Definition 3.7.3. A system of imprimitivity of G on Q in a Hilbert space 1t
is a pair (U, jf) where U is a continuous representation ofG on 1t, and jf is a
nondegenerate representation ofCo ( Q), satisfying the covariance condition (3.38).
There is, in fact, no real difference between 3.4.5 and 3.7.3.
Corollary 3.7.4. There is a bijective correspondence between nondegenerate
representations 7f of C*(G, Q) and systems of imprimitivity of G on Q. This
correspondence is given by continuous extension of (3.35 H 3.37).
Combine Corollary 3.4.6 and Definition 3.6.3.
The structure of action C* -algebras is in general fairly complicated, except for
•
the following corollary to Proposition 3.4.7.
Corollary 3.7.5. One has the isomorphisms
C*(G, G/ H) ~ C*(G XH G) ~ fJ30(L2(G/ H» ® C*(H), (3.85)
and similarly for the reduced C* -algebras (i.e., C* is replaced by C:).
Since the *-algebras C;;"(G x (G/H» and C;;"(G XH G) are isomorphic by
3.4.7, they in particular have the same representation theory. Hence all claims
follow from Definition 3.6.3 and Theorem 3.7.1. •
In particular, we may look at the C* -algebra fJ3 o(L 2(Rn» in two different ways:
It is the C* -algebra of the pair groupoid Rn x R n ~ Rn, as well as the action
C* -algebra defined by the canonical action of Rn on itself.
What follows is the quantum transitive imprimitivity theorem.
Corollary 3.7.6. There is a bijective correspondence between systems ofimprim-
itivity (UX(G), jf(Co(G / H») and representations Ux of H. The system (UX, jfX)
is irreducible (in the sense that the only bounded operators commuting with all
UX(x) and jfX(j) are AlI, A E C) lffUx(H) is irreducible.
Up to equivalence, the pair(UX, jfX) is realized on 1t x (cf. 2.8.4), where UX(G)
is the induced representation defined in (2.175), and jfx is given by
(3.86)

In the realization of the carrier space as 1tfs) (cf. 2.8) one has (U{~), jff,), given
by (2.176) and
jf{~)(j)W:(q) = i(q)w:(q), (3.87)

and analogously on 1t; .. see (2.141).

Combine Corollaries 3.7.4, 3.7.5, 3.4.8, and 3.7.2. The irreducibility of(UX, jfX)
is equivalent to the irreducibility of7fX(C*(G XH G». •
292 III. Groups, Bundles, and Groupoids

A deeper understanding of this result will be achieved in IV.2.7.

We close this section with a remarkable consequence of Corollary 3.7.6, namely
Theorem 11.2.1.4. We are going to apply 3.7.6 with H = {e} and G = Q = Rn,
in which case there is only one irreducible system of imprimitivity, as iI consists
of a single element.
A representation UOo)(H") on a Hilbert space 1t for which
(3.88)

where A. E R\{O}, defines a system of imprimitivity (U(Rn), ii'(Co(R"))) on 1t by

U(v) := Uo.)(Exp(vP));

ii'(j):= Ln d"u Up.) (EXP(-iQ)) j(u), (3.89)

where j is the Fourier transform of f. (Here j E L'(R"), so that ii' thus defined
is to be extended from the image of the Fourier transform of L'(Rn) in Co(Rn) to
Co(R") by continuity.) This may be verified using 11.(2.6).
Conversely, given a system of imprimitivity (U(Rn), ii'(Co(R"») on 1t and a
real number A. =1= 0, one obtains a representation Up.)(Hn) on 1t for which (3.88)
holds by

UO.)(Exp( -u Q» := ii'(hu);
Up.)(Exp(vP» := U(v), (3.90)
where ep(x) := exp(ipx). Here the representation ii'(Co(Rn» has been extended
to Cb(Rn ) by the functional calculus obtained from the spectral theorem.
These two constructions are each other's inverse; the uniqueness of the ir-
reducible representation satisfying (3.88) follows from the uniqueness of the
irreducible system of imprimitivity (U(R"), ii'(Co(R"»). In summary, the clas-
sification of the irreducible representations of the Heisenberg group with nonzero
central element follows from the uniqueness of the irreducible representation of
the group with one element. •

3.8 Lie Algebroids

The construction of the Lie algebra of a Lie group as an "infinitesimal" object can
be generalized to the setting of Lie groupoids. Like a Lie algebra, the object in
question may be defined in its own right.

Definition 3.8.1. A Lie a1gebroid V ~ Tg on a manifold Q is a vector bundle

V(Q, V, T)over Q, which apartfrom the bundle projection T : V ~ Q is equipped
with a vector bundle map Ta : V ~ T Q (called the anchor), as well as with a Lie
bracket [, ]v on the space f'(V) of (smooth compactly supported) sections of V,
satisfying
(3.91)
 3 Lie Groupoids and Lie Algebroids 293

where the right-hand side is the usual commutator of vector fields on reT Q), and
[sJ, fS2]V = f[sJ, S2]V + (Ta 0 SJ/)S2 (3.92)
for all SJ, S2 E reV) and f E COO(Q).
It is part of the definition of a bundle map that the anchor is fiber-preserving and
linear on each fiber.
When Q is a point, one has reV) = V := g, and the only nontrivial requirement
is that [, ]9 be a Lie bracket. Hence in that case the Lie algebroid 9 is simply a
real Lie algebra. The next simplest example is

Definition 3.8.2. The pair algebroid T Q ~ TQQ consists ofthe tangent bundle with
its usual projection and commutator, and anchor Ta = id.
The property (3.92) then reads [~J, f~2] = f[~J, ~2]+(~J/)~2, which is, indeed,
identically satisfied.
Definition 3.8.3. Let P(Q, H, T) be a principal H-bundle over Q. The gauge
algebroid (TP)/ H ~ TQQ is defined by the obvious projections (both inherited from
T), and the Lie bracket on r«TP)/ H) obtained by identifying this space with
r (T p)H (as in the proof of Proposition 2.4.3), and borrowing the commutator
from r(TP).
Thus the bundle V is the central term in the exact sequence (2.65).
Definition 3.8.4. Suppose one has a g-action on a manifold Q,- see 1.1. The action
algebroid 9 x Q ~TQQ has V = 9 x Q (as a trivial bundle over Q), with anchor
Ta(X, q) := -h(q).ldentifying sections ofg x Q with g-valuedfunctions X(·)
on Q, the Lie bracket on reg x Q) is
[X, Y]gxQ(q):= [X(q), Y(q)]g + ~yX(q) - hY(q). (3.93)
Similar to Proposition 3.1.8, we have
Proposition 3.8.5. Let G (G / H, H, T) be the principal H -bundle defined in 2.7.
The gauge algebroid (TG)/ H ~ TYl///) and the action algebroid 9 x (G / H)~ T~G///)
are isomorphic (in the obvious sense).
The quotient of the right trivialization T G c:::: 9 x G by H provides a
diffeomorphism (TG)/H c:::: 9 x (G/H). •
We now explain how one may associate a Lie algebroid Q; ~ Tg with a given
Lie groupoid G ~ Q.
A left-invariant vector field ~ L on G is a vector field satisfying (Tt )*~ L = 0
and (Ly)*~L(y') = ~L(yy') for all (y, y') E Gz. Note that the second condition
is well-defined because of the first one. The space of all left-invariant vector fields
on G is denoted by r(TG)L.
This definition may be restated in terms of the corresponding flow: If ~(y) = y
for some flow yeA), then ~ is a left-invariant vector field iff Tt(Y(A» is independent
of A and y'(y(A» = (y'y)(A) whenever y'y is defined; once again, the second
294 III. Groups, Bundles, and Groupoids

condition is well-defined because of the first one. We call such a y(.) a left-
invariant flow on G.
Lemma 3.8.6.
1. The vector space r(TG)L is a Lie algebra under the usual commutator
borrowed from r(TG).
2. A left-invariant vector field is determined by its values on the unit space Go =
t(Q).
3. The tangent bundle ofG at the unit space has a decomposition
T,(q)G = T,(q)Go EB T.~q)G, (3.94)
T1G := ker(Tt )* C TG. (3.95)
To prove 3.8.6.1 it suffices to remark that since for any smooth map rp the
commutator satisfies rp*[~t. ~2] = [rp*~" rp*~2], the space r(TG)L is closed under
the commutator.
Since y = y(y-'y), left invariance implies that ~L(y) = (Ly)*~L(y-'Y),
which proves 3.8.6.2.
Because nonzero elements of T,(q)(G o) are tangent to curves t(q(s», for which
Tt(t(q(s») = q(s) =J:. q for small enough s, it follows that T,(q)G o n T.~q)G = o.
Now note that the image of T,(q)G under ker( Tt )* on the one hand equals Tq Q, and
on the other hand is isomorphic to the quotient T,(q)G/ T.Cq)G as a vector space. A
dimension count then establishes (3.94). •
Definition 3.8.7. The Lie algebroid (.5 ~ TQQ of a Lie groupoid G ~ Q is given
by the following (cf Definition 1l.3.4.1).
• The vector bundle V = (.5 over Q is the normal bundle Nt Q defined by the
embedding t : Q ~ G; accordingly, the projection T : Nt Q -+ Q is given by
Ts or Tt (these projections coincide on Go).
• Identifying N:(q)Q with T.Cq)G by (3.94), the anchor is given by Ta := (Ts )* :
TG -+ TQ (restricted to ker(Tt)*).
• Identifying a section of N:(q) Q with a left-invariant vector field on G through
the previous item and 3.8.6.2, so that ['«(.5) = ['(Nt Q) ~ [,(TG)L c r(TG),
the Lie bracket [, ]0 is given by the commutator on r(TG) (this is consistent
because of3.8.6.1).
The required equality (3.91) is automatically satisfied (as it holds for all vector
fields on G). To verify (3.92), note that from 3.1.1.4, for f E COO(Q) and ~ E
r(TG) one has (LyMf~(Y-'Y» = f(Ts(y»(Ly)*~(Y-'Y). Hence the action of
COO(Q) on r(TG)L is given by (f~L)(y) = f(Ts(y»~L(y). Equation (3.92) then
follows as in the case of the pair algebroid.
Rather than defining (.5 in terms of the normal bundle Nt, one may put
(.5' := t*TtG = TtG *G Q = {(X, q) E TtG x Q I TTG-?G(X) = L(q)}; (3.96)

cf. 2.1.2. This is simply the restriction of Tt G to t( Q), seen as a bundle over Q
through projection onto the second variable. The anchor is defined as Ta := (Ts )*,
 3 Lie Groupoids and Lie Algebroids 295

as above, and the Lie bracket is obtained by extending sections of ~' to left-
invariant vector fields on G. The isomorphism between ~ and ~' is then obvious
from Definition 3.8.7 and (3.94).
Proposition 3.8.8.
• A Lie algebra 9 is the Lie algebroid 0/ a Lie group G.
• The pair algebroid T Q ~ TQQ is the Lie algebroid o/the pair groupoid Q x Q ~
Q.
• The gauge algebroid (TP)I H ~ TQQ is the Lie algebroid o/the gauge groupoid
P XH P~ Q.
• The action algebroid 9 x Q ~ TQQ is the Lie algebroid 0/ the action groupoid
G x Q~ Q.
For a Lie group G the base Q consists of a point, which t maps to e E G; the
normal bundle is TeG. The construction of the Lie algebroid then amounts to the
usual identification of 9 = TeG with the space of left-invariant vector fields on G.
The Lie algebroid of a pair groupoid is identified by Lemma 11.3.4.3, since t is
the diagonal embedding. The isomorphism Nt'(q) Q :::::: ~(q) G identifies X E Tq Q
with O+X E T(q.q)(Q x Q) :::::: TqQ ED TqQ. Hence the anchor is the identity, the
left-invariant vector fields are of the form ~ L (q, q') = O+~ (q'), and the Lie bracket
is simply the usual one on r(T Q).
In discussing the gauge groupoid, one first notes that when P(Q, H, r) is a
principal fiber bundle, the tangent bundle T Q has the following description. One
defines (TP)II as the bundle over P whose fiber at x is TxP/VxP; see (2.8). In
view of (2.10), the H -action on (TP) 1I (pushed forward from the H -action on P)
is well-defined, and one has T Q :::::: «TP)I 1)1 H. (The dual of this isomorphism
is (2.56).)
We apply the same procedure to the principal H -bundle P x P over P x H P.
Identifying T(x.y)P x P with TxPED TyP, the role of VxP in the definition of I is now
played by the space of all vectors of the form ~i(xH~i(Y) E T(x.y)p x P, X E ~.
The vector bundle Tr(p XH P) over Q is then a double quotient «VP x P)I /)1 H.
The restriction of T r(P X H P) to Go equals the H -quotient of the restriction
of (VP x P)I I to the diagonal. The fiber T(x,x)(VP x P)I I is isomorphic to TxP
+
through the identification of the equivalence class [X Y] E (VX P ED Tx P)I I with
Y - X E TxP. Taking the H-quotient, and using the isomorphism Nt'(q)Q ::::::
~(q)(P XH P) given by (3.94), we arrive at N'Q :::::: (TP)I H.
Following the steps in the above derivation, one immediately infers from its
definition that the anchor is the canonical projection from (TP1H) to T Q, and
that the Lie bracket is as stated.
In the action groupoid the identification Nt Q :::::: TeG = 9 is immedi-
ate from Definition 3.1.4; the normal bundle is automatically identified with
TeG C T(e.q)G x Q. The anchor then follows from (1.19). To compute the
Lie bracket on r(g x Q) one notes that since (y, q) = L(y,q)(e, y-lq), a sec-
tion ~ : q f-+ X(q) E 9 defines a left-invariant vector field on G x Q by
~L(y, q) := L(y,q)*~(e, y-lq) = ~;(y-lq)(Y)' Here ~L is defined in (1.37), and
296 III. Groups, Bundles, and Groupoids

~.L(y) E TyG, regarded as a subspace of T(y,q)G x Q. The expression (3.93) then

easily follows. •
To close this section, we provide the missing proof of the first claim in Proposi-
tion 3.3.3. Indeed, a given strictly positive smooth density p on the vector bundle
(8 associated to G ~ Q by 3.8.7 can be (uniquely) extended to a left-invariant
density p on the vector bundle T'G, which in turn yields a left Haar system by
J.L~(f) := J,,-I(q) pf· •

3.9 The Poisson Algebra of a Lie Algebroid

We saw in 3.6 that one can associate a C* -algebra to a Lie groupoid. The clas-
sical analogue is the construction of a Poisson algebra of a Lie algebroid. This
generalizes the Lie-Poisson structure on COO(g*, R) introduced in 1.1.
Proposition 3.9.1. Given a Lie algebroid V ~ TQQ with anchor r a , the dual vector
bundle V* is a Poisson manifold V~ under the Poisson bracket on COO(V*, R)
defined by the following special cases:
{f,g}± = 0; (3.97)

--
{s, f}± = ±ra 0 sf;
{Sl' S2}± = ±[Sl, S2]V.
(3.98)
(3.99)
Here f := rvo-"QiE COO(V*, R) is defined by j E COO(Q, R), and similarly
for g. Also, S E COO(V*, R) is defined by a section s E J(V) through s(8) :=
8 (s(rvo-., Q(8»).
Note that the function S is linear (in the sense of being linear on each fiber of
V*), and that any such (smooth) function is of this form. Hence the collection
of differentials df, ds spans the cotangent space at every point of V*, so that the
Poisson bracket is indeed completely defined by (3.97)-(3.99).
In a local trivialization of V one has s(q) = sa(q)ea (where lea} is a basis of
the typical fiber V of V), hence s(q, 8) = 8asa(q) in terms of the coordinates 8a
on V* defined by the dual basis. We write [ea, eb](q) = C~b(q)ec and ra(ea , q) =
A~(q)a/aqlL, in terms of which the Poisson tensor is given by B(9,q)(dq lL, dqV) =
0, B(9,q)(d8a, dqlL) = ±A~(q), and B(9,q)(d8a, d8b) = ±C~b(q)8c. The conditions
(3.91) and (3.92) then lead to identities on C and A that are used to prove the
Jacobi identity 1.(2,6). •
Proposition 3.9.1 has the following converse.
Proposition 3.9.2. If a Poisson manifold V* is a vector bundle over Q, such that
the Poisson bracket of two linear functions is linear, then V* is the dual of a Lie
algebroid V, and the Poisson bracket on COO(V*, R) is the one in 3.9.1.
The Lie bracket on f'(V) is defined by reading (3.99) from right to left. To define
the anchor, we note that the Leibniz rule yields
(3.100)
 3 Lie Groupoids and Lie Algebroids 297

We take f as described in 3.9.1. Since the left-hand side and the first term on the
right are linear, as is S2, it follows that lSI, f} is constant on the fibers, defining a
1
function on Q. Applying the Leibniz rule to {s, fg}, the map H- {s, f} is seen
1
to be a derivation on COO(Q, 1R), so it must be that lSI, f} = ~s for some vector
field ~s on Q.
Hence s H- ~s is a map from reV) to reT Q). To prove that it is given by a
bundle map ra : V -+ T Q, we must show that ~ js = 1 ~s for all 1 E C OO ( Q).
This follows from the Leibniz rule {fs, g} = f {s, g} + {f, g}s, which may be
rewritten as ~ jsg -l~sg = {f, g }s. The left-hand side is a function on Q, whereas
the right-hand side is linear; this is possible only when (3.97) holds. Therefore,
~s = ra 0 s for some bundle map ra.
The Jacobi identity on the Poisson bracket and the definition of ra imply (3.91).
Finally, (3.100) is equivalent to (3.92). •
Combining Propositions 3.9.1 and 3.9.2, we conclude that there is a com-
plete equivalence between Lie algebroids and linear Poisson structures on vector
bundles. We now apply this to our usual list of examples; cf. 3.8.8.
Proposition 3.9.3.
• A Lie algebra g yields the ± Lie-Poisson structure ( J. J) on g*.
• The pair algebroid T Q ~ TQQ leads to ± the canonical Poisson bracket 1.(2.24)
on T*Q.
• The gauge algebroid (TP)/ H ~ TQQ is associated with ± the Poisson structure
on (T*P)/ H specified prior to 2.3.7.
• The ± Poisson bracket on g* x Q associated to the action algebroid g x Q ~ TQQ
is given by

(3.101)

Compare 1.1 for the notation used in (3.101).

The first claim is obvious. The second is most easily proved in a local trivial-
ization, using canonical coordinates. The section seq) = (8/8q iL, q) then leads
to s(p,q) = Pw The brackets (3.97)-(3.99) thus spec~to {qiL,qV}+ = 0,
{PiL' qV}+ = (8/8qiL)qV = 8~, and {PiL' Pv}+ = [8/8qiL, 8/8qV] = 0, respec-
tively. This proves the claim. Note that the linear function ~ E COO(T*Q, 1R)
corresponding to ~ E reT Q) is simply the usual symbol of the vector field.
The third point follows from the second, quotienting by H.
To prove (3.101) one simply verifies that it includes (3.97)-(3.99) with (3.93)
as special cases; do not forget the minus sign in the anchor. •
It is worth giving the main special cases of (3.101). For X E g we regard i as
1
a function on g* x Q by i(O, q) = O(X), and as before, E C~(Q, JR) defines a
function on g* x Q by f(o, q) = f(q). We then have
{f, g}- = 0; (3.102)
{i, Yl- = -[X, V]; (3.103)
298 III. Groups, Bundles, and Groupoids

{X, !}- = hi. (3.104)

As will become clear in 3.11 below, these brackets are the classical counterpart of
the commutation relations (3.41), (3.42), and (3.43). We denote the action Poisson
algebra on g* x Q with bracket (3.101) by COO(g± x Q, Ill); the corresponding
Poisson manifold is, of course, written as g± x Q.
Proposition 3.9.4. The symplectic leaves of g~ x Q are classified by pairs
(oG, OH), where OG is a G-orbit in Q, the group H ~ G is the stabilizer of
an arbitrary point in OG, and OH is a coadjoint orbit in 1)*. The leaf L(oo ,OH)
corresponding to (OG, OH) is given by
L(OO,oH) = {(O, q) E g* x Q I q E OG, (-Co(S(q)-I)O r 1)*) E OH}, (3.105)
where s : OG ~ G / H ~ G is an arbitrary section of the bundle G( G / H, H, r).
One infers from (3.101) that any Hamiltonian flow starting in g~ x OG stays
in this subspace. Using the right trivialization of T*G one sees that g* x OG is
diffeomorphic to (T* G) / H. Equipping the latter, and therefore g* x OG, with
the Poisson structure inherited from T*G, and comparing (1.54) with (3.101),
one infers that the injection of g* x OG into g~ x OG is a Poisson map. The
proposition then follows from Theorem 2.3.7, applied to the bundle P(Q, H, r) =
G(G/ H, H, r), and (2.118). •
We saw in 3.4 that an action *-algebra is a special case of a crossed product
*-algebra. Similarly, the Poisson algebra of an action Lie algebroid is a special
case of the following classical analogue of a crossed product *-algebra.
Definition 3.9.5. Let a Lie group G act on a Poisson manifold P by Poisson maps
L. The semidirect product of P and T*G (equipped with the canonical Poisson
structure 1.(2.24)) is the quotient T*G XG P under the product action p x L of
G (cf (1.51)), equipped with the unique Poisson structure making the canonical
projection r : T*G x P ~ T*G XG P a Poisson map.
The diffeomorphism [(0, X)R, a]G f-+ (0, xa) between T*G XG P and g* x P
equips the latter with a Poisson structure; the associated Poisson algebra COO(g~ x
P, Ill) is called a crossed product Poisson algebra.
In self-evident notation, the Poisson bracket is explicitly given by

(3.106)

Hence putting P = Q with the zero Poisson structure shows that (3.101) is indeed
a special case of (3.106); this is the classical version of Lemma 3.4.3.
In the following classical analogue of Theorem 3.4.4 the boundedness condition
(3.33) is replaced by an integrability condition.
Theorem 3.9.6. There is a bijective correspondence between
• Poisson maps J : S ~ g~ x P for which the associated g-action is integrable
(here S is a symplectic manifold);
 3 Lie Groupoids and Lie Algebroids 299

• Pairs consisting of a strongly Hamiltonian G-action on S and a Poisson map

p : S ~ P such that the classical covariance condition
~ip*{j) = p*{~: j) (3.107)

holds for all X E 9 and i E COO{P, JR.); cf (1.19).

It is not necessary that S be connected; the proof works for each component
separately. Given J, one defines a g-action on S by X ~ ~J'x' where the definition
of X is similar to the one given prior to (3.1 02). When this action is integrable, the
corresponding G-action is strongly Hamiltonian by definition. Also, the restriction
of J* to COO{P) evidently defines a representation on S. Condition (3.107) is then
satisfied because of (3.106) and the fact that J is a Poisson map.
Conversely, a strongly Hamiltonian G-action is associated with a Poisson map
J(1) : S ~ g~ (see 1.2). Writing J(2) := p, one obtains a map J = (J(I), J(2» :
S ~ g~ x P. Using (3.106) and an argument similar to the proof of 1.1.2 one
shows that J is a Poisson map. 0
One would like to sharpen this result by saying that there is a bijective correspon-
dence between representations J* of the Poisson algebra COO(g~ x P, JR.) (which
by definition implies that the Poisson map J is complete) and pairs as stated in
3.9.6, for which in addition the map p is complete (so that p* is a representation
of COO(P ,JR.) on S). This works in one direction when G is simply connected, for
in that case the completeness of f(l) 0 J : S ~ g* (where f(l) : g* x P ~ g* is
the projection onto the first variable) implies that the g-action on S is integrable
by Theorem 1.2.1. However, in the opposite direction it is in general not clear
that the completeness of J(l) and J(2) = P implies that of J. Paraphrasing the first
ingredient of 3.9.6, we obtain
Definition 3.9.7. Given a G-action on Q, a classical system ofimprimitivity of
G on Q consists of a symplectic manifold S, along with a strongly Hamiltonian
G-action on S and a nonzero representation if : COO{Q,JR.) ~ COO(S,JR.) (where
Q has the zero Poisson structure), such that the integrated classical covariance
condition
a~(if(j» = if(ap(i» (3.108)

holds for all x E G and i E COO(Q,JR.). Here a~(f)(u) .- f{x-1u) and

ap(j)(q) = i(x-1q).
By I.2.6.5 there exists a complete Poisson map J(2) : S ~ Q for which if = J(i).
Condition (3.108) is evidently equivalent to the G-equivariance of J(2)o that is, one
has J(2)(XU) = xJ(2)(U) for all x E G and u E S. Moreover, equation (3.108)
implies that ~iif(1) = if(~~ 1>,cf. (3.107), and is equivalent to this condition
when G is connected.
This time we have spoken of a representation of COO(Q), rather than merely a
Poisson map p : S ~ Q. This is justified by the classical version of Corollary
3.4.6, which specializes and sharpens Theorem 3.9.6:
300 III. Groups, Bundles, and Groupoids

Corollary 3.9.8. When G is connected there is a bijective correspondence be-

tween classical systems of imprimitivity of G on Q and representations 1C of the
Poisson algebra CCXl(g~ x Q, R) whose associated g-action is integrable.
As remarked above, the integrability condition is automatically satisfied when G
is simply connected. The new issue relative to the proof of3.9.6 and the subsequent
comment is that in this special case the completeness of J(l) and J(2) = p does
imply the completeness of J. The proof of this requires advanced techniques in
symplectic geometry that we have not developed. 0

In analogy to Proposition 3.7.5 we have a result suggested by 3.8.5.

Proposition 3.9.9. The Poisson manifolds (T*G)/ H and g~ x (G / H), and the
Poisson algebras COO«T*G)/ H, R) and COO(g~ x (G / H), R), are isomorphic.
The isomorphism mentioned in the proof of 3.8.5 can be "dualized", defining a
Poisson map. •
Paraphrasing the comment after 3.7.5, it therefore follows that we may look at
the Poisson algebra coo(T*Rn , R) in two different ways: It is the Poisson algebra
of the pair algebroid TR", as well as the Poisson algebra of the action algebroid
defined by the canonical action of Rn on itself.
The classical analogue of 3.7.6 is
Corollary 3.9.10. When G is connected, each coadjoint orbit q!!f H leads to an
irreducible classical system ofimprimitivity of G on G / H in (T*G)o.
Recall the definition (2.51), with P = G, of (T7G)o, as well as (2.118) and
(2.130). The claim follows from Corollaries 2.3.8 and 3.9.8. •
In view of the possibility of covering spaces in 2.3.8, this analogy is not
quite perfect. The connectedness assumption and the restriction to irreducible
representations will, however, be removed in IV.l.6.4.
In order to generalize Theorem 1.1.7, we need an appropriate concept of the
action of a Lie algebroid on a manifold.
Definition 3.9.11. A (left) groupoid action ofG ~ Q on a space S consists of
maps Jp : S --+ Q and L : G *Q S --+ S, where
G *Q S := {(y, a) E G x S I Ts(Y) = Jp(a)}. (3.109)
Writing ya := L(y, a), these maps must satisfy
Jp(ya) = r,(y); (3.110)
£(Jp(a»a = a; (3.111)
y(y'a) = (yy')a, (3.112)
whenever (y', a) E G *Q Sand (y, y') E G2.
When G is a Lie groupoid, one speaks of a smooth groupoid action if S is a
manifold and J p and L are smooth.
 3 Lie Groupoids and Lie Algebroids 301

As to the last definition, note that the surjectivity of rs implies that G *Q S is a

submanifold of G x Q.
For example, taking S = G and J p = rt leads to G *Q S = G 2 , and L is
simply multiplication in the groupoid. Alternatively, choosing S = Q and J p = id
reproduces the action of G on its base.
In the smooth case one obtains a linear map s 1--+ ~s from r(<!5) to r(T S)
(where <!5 is the Lie algebroid of G), defined as follows. Identifying the section
s E r(<!5) with a left-invariant vector field ~L on G, which in tum corresponds to
a left-invariant flow yO on G (i.e., ~L(y) = y), we put

d -I
~s(a) := d)" (t(Jp(a»)()..) aIA=O· (3.113)

It follows from 3.1.1.4, the left invariance ofy(·), and (3.111) that (t(Jp(a »)()..)-I a
lies in G *Q S, so that ~s is well-defined. The definition of the Lie bracket on r (<!5)
implies that s 1--+ ~s is a Lie algebra homomorphism. Finally, (3.110) and the
definition of the anchor ra in 3.8.7 entail

(3.114)

for all s E r(<!5). Thus we are led to

Definition 3.9.12. An action of a Lie algebroid V -=+ TQQ on a manifold S consists

of a smooth map J p : S --* Q and a Lie algebra homomorphism s 1--+ ~s from
r(V) into r(TS) such that (3.114) holdsfor all s E r(V).

This definition is further motivated by the thought that the pair algebroid T S is the
most natural Lie algebroid; to express the idea that an action should "preserve"
the anchor ra one in addition needs the map J p • One could, equivalently, use
an antihomomorphism s 1--+ ~s, in which case the condition on the anchor reads
(Jp)*gs = -ra 0 s. Thus one has generalized the definition of an action of a Lie
algebra 9 on a manifold; cf. 1.1.

Proposition 3.9.13. Let V -=+ Tg be a Lie algebroid with associated Poisson al-
gebra COO(V*, JR.). A representation Jr : COO(V*, JR.) --* COO(S, JR.) (in the sense of
/.2.6.1), where S is a symplectic manifold, leads to a V-action on S.

Given Jr, one obtains a Poisson map J : S --* V* by 1.2.6.5, and subsequently
defines the V-action by J p := r 0 J and gs := ~vs (i.e., the Hamiltonian vector
field of J*s). Equations 1.(2.15) and (3.99) then imply that [gSI' ~S2] = [Sl, S2]V,
whereas 1.(2.8), 1(2.15), and (3.98) imply the condition on the anchor. •

One may then define a strongly Hamiltonian V-action on a symplectic manifold

essentially as an action given by a "momentum map" J as above. This, then, leads
to the obvious generalization of Theorem 1.1.7 from Lie algebras to Lie algebroids,
which is a classical counterpart of the correspondence between representations of
a Lie groupoid and representations of the associated C* -algebra; cf. 3.6.
302 III. Groups, Bundles, and Groupoids

3.10 A Generalized Exponential Map

The theory of Lie groupoids and algebroids suggests a unifying principle behind
the various strict quantizations we have discussed so far. It turns out that these may
be formulated in terms of a generalized exponential map Exp W : ~ ~ G from a
Lie algebroid into a corresponding Lie groupoid.
Lemma 3.10.1. The vector bundles rtG and "Cs*~ (over G) are isomorphic.
Recall (3.95) and 2.1.2. The pullback bundle
"Cs*~ := {(X, y) E ~ x G I "C(X) = "Cs(Y)} (3.115)
is a vector bundle over G with projection onto the second variable. The isomorphism
of 3.10.1 is proved via the vector bundle isomorphism ~ :::: ~'; see (3.96) and
subsequent text. Thus replacing ~ in (3.115) by ~', one checks that (Ly-l)* :
T~G ~ T;_lyG is the desired bundle isomorphism (note that the inverse is (Ly)*,
and cf. Lemma 3.2.3). •
For a Lie group G we have T' G = T G, and Lemma 3.10.1 simply reproduces
the left trivialization T G :::: {I x G. For a pair Lie groupoid Q x Q ~ Q we
identify T' (Q x Q) with Q x T Q, where the first Q is seen as the zero section
in T Q; the projection is "CT'(Qx Q)--> QxQ(q, Y) = (q, "CTQ-->Q(Y»' On the other
hand, the lemma says that Q x TQ should be isomorphic to {(Y, q', q) E TQ x
Q x Q I "CTQ-->Q(Y) = q}, with projection "Cr:TQ-->QxQ(Y, q', q) = (q', q). This
isomorphism is immediately obvious. Similarly for a gauge groupoid.
For an action Lie groupoid G x Q ~ Q we identify T' (G x Q) with T G x Q
(where Q is the zero section of TQ), with projection "CT'(GxQ)-->GxQ(Y, q) =
("CTG-->G(Y), q). The lemma identifies TG x Q with {(Y, x-1q, X, q) E {I x Q x
G x Q} with (Y, x, q) through the left trivialization of TG.
Let us now assume that ~ has a covariant derivative (or, equivalently, a con-
nection), with associated horizontal lift lf8; cf. the paragraph following (2.28). By
Proposition 2.2.4 and Lemma 3.10.1 one then obtains a connection on T'G (seen
as a vector bundle over G, whose projection is borrowed from TG). Going through
the definitions, one obtains that the associated horizontal lift l of a tangent vector
X = Y := dy(t)/dtlt = 0 in TyG to Y E T~G is

ly(y) = :t [LY(')*"~y_ll.y("Cs(y(t)))l=o' (3.116)

which is an element of Ty(TIG) (here lf8( ... ) lifts a curve).

Example 3.10.2.
• For a Lie group G = G the base space is a point, so that no connection needs to
be chosen, and a horizontallijt is always zero. In the left trivialization (where
(X, x) := d/dt(xExp(tX)/dt)lt = 0; cf. (1.37», the expression (3.116) then
reads
d
l(y.x)(X, x) = dt (Y, xExp(tX»,=o = (0, Y, X, x). (3.117)
 3 Lie Groupoids and Lie Algebroids 303

Here T(TG) ~ T(g x G) ~ g x g x TG ~ g x g x g x G.

• A connection in a pair Lie algebroid T Q ~ TQQ is the same as an affine con-
nection, with horizontal lift lTQ : T Q ---+ T(T Q); cf. ][.3.2.1. Using the
identification Tt(Q x Q) ~ Q x TQ as above, the left-hand side of (3.116)
is of the form l(q,.y)(q', q), where Y E TqQ. The right-hand side then assumes
the form d/dt(q'(t), l~Q(q(t»lt = 0, so that

l(q'.rM', q) = (q, l~Q(q». (3.118)

This is a vector in T(T t (Q x Q» ~ T Q x T(T Q), as it should be. An analogous

computation may be done for gauge groupoids .
• For an action Lie groupoid G x Q ~ Q we just saw that Tt(G x Q) ~
TG x Q ~ g x G x Q. Using the notation of(3.117), the right-hand side of
(3.116) is a vector in 'r(r,x)(TG) x TqQ, namely
fO",x,q)(X, x, q) = (0, Y, X, x, q). (3.119)
Since the bundle Tt G ---+ G has a connection, one can define the geodesic flow
X 1-* X(t) on TtG in precisely the same way as on a tangent bundle with affine
connection; see 11.3.1. To recapitulate, the flow X(t) is the solution of
X(t) = lX(t)(X(t», (3.120)
with initial condition X (0) = X.
Definition 3.10.3. Let the Lie algebroid ~ ~ TQQ of a Lie groupoid G ~ Q
be equipped with a connection. Relative to the latter, the left exponential map
ExpL : ~ ---+ G is defined by
EXpL(X) := Yx,(I) = rT'G~G(X'(I», (3.121)

whenever the geodesicjiow X'(t) on TtG (defined by the connection on TtGpulled

back from the one on ~) is defined at t = 1. Here X' E ~' = TtG I Go is the
image of X under the isomorphism ~' ~ ~; cf. (3.96) etc.
Our goal, however, is to define a "symmetrized" version of EXpL .
Lemma 3.10.4. For all X E ~ for which EXpL(X) is defined one has
rt(ExpL(X» = reX). (3.122)
We write X for X' in (3.121). One has rt(yx(O» = reX) and
d
dt <t(yx(t» = (rt 0 rTIG~G)*lx(t)(X(t» = (rt)*X(t) = 0,
since lx(Y) covers Y, and X(t) E TtG = ker«t)* n TG.
We combine this with the obvious «tX) = r( -tX) to infer that
•
rt(ExpL(tX» = rt(ExpL(-tX» = rs(ExpL(-tX)-I).
Thus the (groupoid) multiplication in (3.123) below is well-defined.
304 III. Groups, Bundles, and Groupoids

Definition 3.10.5. The Weyl exponential map Expw : ® -+ G is defined by

Expw (X):= EXpL(_~X)-lExpL(~X). (3.123)

The following result is closely related to the tubular neighborhood theorem

11.3.4.2, and includes Lemma 11.3.4.3 as a special case.
Theorem 3.10.6. The maps EXpL and Exp ware diffeomorphisms from a neigh-
borhood N' of Q c ® (as the zero section) to a neighborhood N; of t(Q) in G,
such that EXpL(q) = Expw (q) = t(q)for all q E Q.
The property EXpL(q) = t(q) is immediate from Definition 3.10.3. The push-
forward of ExpL at q is Exp~ : Tq® -+ 7;(q)G. Now recall the decomposition
(3.94). For X tangent to Q C ® one immediately sees that Exp~(X) = t*X.
For X tangent to the fiber r-1(q), which we identify with T.(q)G (cf. (3.96)
etc.), one has Exp~(X) = XI, as follows by the standard argument used to
prove that eXPq in the theory of affine geodesics is a local diffeomorphism: FOI
a curve Xes) = sX in T.(q)G one has EXpL(X(S» = yx'(s)(l) = yx,(s), so that
d/ds[ExpL(X(s»]ls = 0 = XI. Since Exp~ is a bijection atq, the inverse function
theorem implies that ExpL is a local diffeomorphism. Since it maps Q pointwise to
t( Q), the local diffeomorphisms can be patched together to yield a diffeomorphism
of the neighborhoods stated in 3.10.6; we omit the details of this last step, since it
is identical to the proof of the tubular neighborhood theorem.
As for Expw, we have Exp: (X) = t*X for X E Tq Q c Tq®. Also,

:s [EXpL(_~SX)-lExpL(~SX)]s=o = -F*X I + ~XI,

where 1* is the pushforward of the inversion in G. The right-hand side lies in
ker«rs )* + (r/)*) C TG, and every element in this kernel is of the stated form.
Similarly to (3.94), one may prove the decomposition
(3.124)
It follows that Exp: is a bijection at q, and the second part of the theorem is
derived as for ExpL . •

Our standard list of examples illustrates Definition 3.10.5.

Proposition 3.10.7.
• For a Lie group G no connection is needed, and one has
EXpL(X) = Expw (X) = Exp(X), (3.125)

where X E 9 and Exp : 9 -+ G is the usual exponential map.

• For a pair Lie groupoid Q x Q ~ Q one chooses an affine connection 'V on
T Q, with associated exponential map exp : T Q -+ Q. Then
EXpL(X) = (r(X), eXPr(X)(X»; (3.126)
Expw (X) = (exPr(x/ -~X), eXPr(X)(~X», (3.127)
 3 Lie Groupoids and Lie Algebroids 305

where X E T Q and T := TTQ-+Q.

• For a gauge groupoid P x H P ~ Q one chooses an H -invariant affine connec-
tion on TP, with exponential map exp : TP -+ P. This induces a connection
on (TP)/ H, in terms of which

EXpL([X]H) = [T(X), eXPr(X)(X)]I/; (3.128)

Expw ([X]H) = [exPT(X)( -~X), eXPT(X)(~X)]H' (3.129)

where T := TTP-+P, and [X]H E (TP)/ H is the equivalence class of X E TP

under the H -action on T P.
• For an action groupoid G x Q ~ Q the trivial connection on 9 x Q -+ Q
yields

EXpL(X, q) = (Exp(X), q); (3.130)

Expw (X, q) = (Exp(X), Exp(~X)q). (3.131)

We infer from (3.117) and (3.120) that the geodesic flow in TtG = TG ~ gxG
is determined by the differential equation (Y, i) = (0, Y); this suggestive notation
should actually read (0, Y, Y, x). Recalling that we work in the left trivialization,
this equation is solved by (Y(t), x(t» = (Y, xExp(tY». Now X Egis identified
with X' E TeG, so that (3.125) follows.
The geodesic flow on Tt(Q x Q) is (X(t), Y(t» = (X(O), Yet»~, where Y(t) is
the flow on T Q determined by the affine connection. This immediately leads to
(3.126), and hence to (3.127), since (x, y_)-I(X, y+) = (y_, y+).
An H-invariant connection on TP by definition satisfies (Rh)* V~YJ = V~TJ for
all h E H and all~, TJ E nTP)H. This implies H-invariance of the geodesic flow
on TP in that (RhMX(t» = «Rh)*X)(t) for all h and X. The exponential map is
then H-invariant in the sense of 11.(3.90), which, with (3.126), leads to (3.128).
Equation (3.130) follows from (3.119), in analogy with the derivation of (3.126).
Subsequently, (3.131) is derived from (3.123) and the definitions of multiplication
and inversion in an action groupoid (see 3.1.4). •

3.11 The Groupoid C* -Algebra as a Strict Quantization

Theorem 11.2.6.1 shows that C~(T*Rn), regarded as a subalgebraofthe complex-
ified Poisson algebra coo(T*Rn), is quantized by Q3o(L2(Rn». This is generalized
to arbitrary Riemannian manifolds Q in Theorem 113.5.1, in which C~(T*Q)
is quantized by Q30(L2(Q». For compact G, in Theorem 1.9.2 the complexified
Lie-Poisson algebra C~(g~) is quantized into the group C* -algebra C*( G). When
P(Q, H, -r) is a principal fiber bundle with compact structure group H, we saw in
Theorem 2.8.1 that the complexified Poisson algebra C~«T*P)H) is quantized
by Q3 0(L2(p»H.
If we look at this list, as well as at Propositions 3.8.8 and 3.9.3 and Theorem
3.7.1, we discern that in all cases an appropriate subspace of the Poisson algebra
COO(~*) canonically associated to a given Lie algebroid ~ ::;T~ (see 3.9.1) is
306 III. Groups, Bundles, and Groupoids

quantized by the C*-algebra C*(G) (defined in 3.6.3) of a Lie groupoid G whose

Lie algebroid is ~. In the cases at hand one has C*(G) = C:(G).
In all cases the quantization map Q/i is a special case of the following construc-
tion. We start by defining a fiberwise Fourier transform j E C<'O(~) of suitable
f E COO(~*). This transform depends on the choice of a family {JL~}qEQ of
Lebesgue measures, where JL~ is defined on the fiber .-l(q). We will discuss the
normalization of each JL~ in due course; for the moment we merely assume that
the q-dependence is smooth in the obvious (weak) sense.
For a function j on ~ that is Lion each fiber we put

f(O) = 1
r-I(q)
dJL;(X)e- i9 (X) j(X), (3.132)

where. := '~->Q and X E .-l(q). Each JL~ determines a Lebesgue measure JL~*
on the fiber .;a/-+Q(q) of ~*, whose normalization is fixed by requiring that the
inverse to (3.132) be given by

j(X) = I-I
T0'~Q(q)
dJL;*(8)e i9 (X) f(O). (3.133)

The fiberwise Fourier transforms 11.(3.42) and (1.124) are clearly special cases
of (3.133). On the action Lie algebroid 9 x Q, equipped with the trivial connection,
we simply have

J<X, q) = { (~:~n ei9 (X) f(8, q). (3.134)

As in (1.124), the normalization of d n 8 is determined by the normalization of the

Haar measure on G.
Having constructed a Fourier transform, we define the class C~(~*) as con-
sisting of those smooth functions on ~* whose Fourier transform is in C;;o(~);
cf. 11.(3.49). Generalizing the procedure in Definition 11.3.4.4, we pick a function
K E COO(~, R.) with support in Nt (cf. 3.10.6), equaling unity in some smaller
tubular neighborhood of Q, as well as satisfying K( -X) = K(X) for all X E ~.

Definition 3.11.1. Let G be a Lie groupoid with Lie algebroid ~. For Ii =1= 0 the ±
Weyl quantization of f E C~(~*) is the element Q~ (f)± E C;;o(G) (regarded
as a dense subalgebra of C*(G) or C:(G), defined by Q~ (f)±(y) := 0 when
Y tt N.., and by
(3.135)
Here the Weyl exponential Expw : ~ -+ G is defined in (3.123), and the cutoff
function K is as specified above.
This definition is possible by virtue of Theorem 3.10.6. By our choice of
C~(~*), the operator Q~ (f)± is independent of K for small enough Ii.

Proposition 3.11.2. For real f the operator Q~ (f)± is selfadjoint in Ct)(G).

 3 Lie Groupoids and Lie Algebroids 307

This is immediate from (3.23) and (3.123). •

It is evident from Proposition 3.10.7 that the previously constructed Weyl quanti-
zation maps 11.(2.108) with 11.(2.109) on T*]Rn ,11.(3.50) withll.(3.51) on T* Q, and
the map defined in Theorem 2.8.1 on (T*P)/ H correspond to Q'!r (f)-, whereas
the quantization defined in (1.127) on g* corresponds to Q'!r(f)+. In all cases,
Q'!r O± defines a strict quantization of the appropriate subalgebra of COO(I5~, ]R)
into C*(G). A new case is the action Lie groupoid G x Q ~ Q. Here the cutoff
K is independent of q, and coincides with the function appearing in (1.127).

Theorem 3.11.3. For small enough Ii, afunction I E C~(g* x Q) is quantized

according to (3.135) by

Q'!r (f)±(Exp(X), q) = 1 g*
dnfJ i
--~- e iil1 (X) 1(±fJ, Exp(-tX)q).
(2n n)n
(3.136)

When G = ]Rn and Q has a G-invariant measure, Q'!r O± defines a strict

quantization 01 the Poisson algebra C~(g~ x Q, 1R) olthe action Lie algebroid
IRn x Q into the action C* -algebra C*(lRn, Q) = q(lR n, Q) on I = R
Equation (3.136) follows from (3.135), (3.131), and (3.134). Conditions
1I.1.l.1.1 and 2 hold by Theorem 3.11.4 below. We prove 1I.1.l.l.3. For an action
C*-algebra C*(G, Q) the bound (3.75) reads

11111 :::: IIfIIl = sup

qEQ
1
G
dx I/(x, q)l, (3.137)

where I E C;;o(G x Q); cf. (3.44) and (3.57). We put Q'!r (f) := Q'!r (f)+, and
substitute (3.136) and (3.137) in II Q'!r (f)Q'!r (g) - Q'!r(fg)lIl (do not confuse
I E C*(G, Q) in (3.137) with I E C~(g"'- x Q) used in this step). One rescales
some integration variables so that Ii occurs only in expressions of the generic form
I (fJ, Exp(/iX)q), where X E 9 = ]Rn. One then Taylor-expands I and g in Ii, e.g.,
l(fJ, Exp(liX)q) = l(fJ, q) + liXa~p l(fJ, q) + O(li2); (3.138)
cf. (1.19). Expressions of the form xa exp(ifJbXb) in the O(Ii) term in (3.138) are
rewritten as -i8/8fJa exp(ifJbXb), upon which one partially integrates in fJ. Two
integrations in the O(Ii) term can then be done explicitly, and using (3.101) (in
which the structure constants C of course vanish) and j, g E C~(lRn x Q) one
proves 11.(1.3) via (3.137). •
When Q = G/ H, one may use Proposition 3.9.9 and Corollary 3.7.5 to show
that the prescription (3. 135) applied to C~(g"'- x Q) (mappingitintoC*(G, G / H»,
that is, (3.136), coincides with its application to C~«T*G)/ H) (thereby mapped
into C*(G XH G». Specifically, one should use Q'!rO+ on g"'- x (G/H) and
Q'!r 0- on (T*G)/ H; this is because the two relevant Poisson brackets stated in
3.9.3 differ by a sign. Taking G = ]Rn and H trivial, we see that Theorem 3.11.3
is essentially Theorem 11.2.6.1.
To further understand the prescription (3.136), we pass to some representation n
of C*(G, Q), for example, to the regular representation nf, cf. (3.67). This has the
308 III. Groups, Bundles, and Groupoids

advantage that Q::' may be extended to certain unbounded functions. With Xand
j as defined prior to (3.102), and U(G) and n(Co(Q» associated to Jr(C*(G, Q»
as in 3.4.6 or 3.7.4, easy formal manipulations (which in the case of Jri are valid,
on, say, C~(G x Q) c L2(G x Q» yield
w - .
Jr(Q/i (X)+) = lIUlU(X); (3.139)
Jr(Q::' (j)+) = n(j). (3.140)

Comparing this with (3.39) and (3.40), we see that Q~(g) = Jr(Q::'(g» when
g is eitherX or j. In particular, if we omit the arbitrary representation Jr, eqs.
(3.41)-(3.43) and (3.102)-(3.104) lead to a strong version of Dirac's condition
II.( 1.3),

(3.141)

To reiterate, this is valid when f and g are of the form j or X, and strictly speaking
holds in any representation Jr of C*(G, Q) on a suitable domain (e.g., 1irJ) of the
carrier space of Jr .
Motivated by these examples, one would like (3.135) to provide a strict quanti-
zation for any Lie groupoid G. However, Dirac's condition 11.(1.3) has been proved
only in cases featuring a good correspondence between the symplectic leaves of <!S*
and the irreducible representations of C;(G); cf. the proofs of Theorems 11.2.6.1,
11.3.5.1, 1.9.2, and 2.8.1. The other conditions, though, always hold.
Theorem 3.11.4. The map (Q::')± : C:(<!S~, R) ~ C;(G)1R defined by (3.135)
satisfies conditions lI.l.l.l.l and 2 of Rieffel and von Neumann.

3.12 The Normal Groupoid o/a Lie Groupoid

The essence of the proof of Theorem 3.11.4 is to regard <!S as a Lie groupoid, and
glue it to G so as to obtain a new Lie groupoid containing both G and <!S.
Definition 3.12.1. Let G =* Q be a Lie groupoid with associated Lie algebroid
<!S ~ TQQ as defined in 3.B.7. The normal groupoid GN is a Lie groupoid with base
R x Q, defined by the following structures.
• As a set, G N = <!S U {R\{O} x G}. We write elements OfGN as pairs (n, u),
where u E <!S for n = 0 and U E Gfor n "1= O. Thus <!S is identified with {OJ x <!S .
• As a groupoid, G N = {O x <!S} U {R\{O} x G}. Here <!S is regarded as a Lie
groupoid over Q, with 'Cs = 'Ct = 'C and addition in the fibers as the groupoid
multiplication. The groupoid operations in R\{O} x G are those in G.ln other
words,
'Cs(O, X) := 'Ct(O, X) = (0, 'C(X»; (3.142)
'C,(n, y) := (n, 'Cs(y»; (3.143)
'Ct(n, y) := (n, 'Ct(Y»; (3.144)
 3 Lie Groupoids and Lie Algebroids 309

t(O, q) := (0, q); (3.145)

t(n, q) := (n, t(q»; (3.146)
(0, X) . (0, Y) := (0, X + Y); (3.147)
(n, y.) . (n, Y2) := (n, YI Y2); (3.148)
(0, X)-I := (0, -X); (3.149)
(n, y)-I := (n, y-I). (3.150)
n
Here =1= 0; in (3.145) it is understood that Q c (!S as the zero section .
• The smooth structure on GN is as follows. To start, the open subset 0 1 :=
R\{O} x G c GN inherits the product manifold structure. Let Q C Nt C (!S
and t(Q) c.N; c G, as in Theorem 3.10.6.
Let 0 := {(n, X) I nX E Nt}; this is an open subset ofR x (!S, containing
{OJ x (!S. Define l/J : 0 ~ GN by
l/J(O, X) := (0, X);
l/J(n, X) := (n, Expw (nX». (3.151)
Since Expw : Nt ~ .N; is a diffeomorphism (cf 3.10.6), we see that 0 is a
bijection from 0 to O 2 := {O x (!S} u {R\{O} x .N;}. This defines the smooth
structure on O 2 in terms of the smooth structure on O. Since 0 1 and O 2 cover
GN , this specifies the smooth structure on GN •
The fact that GN is a Lie groupoid follows from the corresponding property of
G. The given chart is defined in terms of the Weyl exponential, which depends
on the choice of a connection in (!S. However, one may verify that any (smooth)
connection, or, indeed, any (Q-preserving) diffeomorphisms between Nt and.N;,
leads to an equivalent smooth structure on GN • For example, we could have used
ExpL instead of Expw . Also, the smoothness of Expw makes the above manifold
structure on GN well-defined, in that open subsets of 0 1 n 02 are assigned the
same smooth structure.
The normal groupoid of a pair Lie groupoid Q x Q ~ Q is known as the
tangent groupoid of Q, and is sometimes described by saying that one "blows up"

°
the diagonaI8(Q) in Q x Q. Convergence in (Q x Q)N in the manifold topology is
as follows: If nn ~ then (/in, qn, q~) ~ (0, X) iff qn ~ r(X), q~ ~ r(X), and
Yn (t) / /in ~ X, where Yn is an affinely parametrized geodesic with Yn (0) = qn and
Yn (1) = q~. This convergence is independent of the affine connection defining the
geodesic in question. In local coordinates, where X = (vI-', ql-'), the convergence
condition is simply that q~ ~ ql-', (q~)1-' ~ ql-', and «q~)1-' - q~)//in ~ vI-'.
We now pick a left Haar system {J.L~ }qeQ on G =* Q; cf. 3.3.3. The vector bundle
(!S, regarded as a Lie groupoid as in 3.12.1, has a left Haar system consisting of the
family {J.L~ }qeQ of Lebesgue measures on each fiber, already used in the construc-
tion of the Fourier transform. Since we have a Lie groupoid, the Radon-Nikodym
derivative Jq(X) := dJ.L~(Expw (X»/dJ.L~(X) is well-defined and strictly positive
onNt (since both measures are locally Lebesgue on spaces with the same dimen-
sion). We now fix the normalization of the J.L~ by requiring that limx-> 0 Jq(X) = 1
310 III. Groups, Bundles, and Groupoids

for all q. This leads to a left Haar system for GN , given by

ilL.
lit
""'(O,q) .=
•
""'q'
t I;:-n t
f-t(h.q) := n f-tq, (3.152)
where n is the dimension of the typical fiber of (!S. The factor fi-n is necessary in
=
order to satisfy condition 3.3.2.3 at fi 0, as is easily verified using the manifold
structure on G N defined in 3.12.1.
To avoid confusion between functions on G N and on (!S* , we denote the former
by f, g. Thus the *-algebraic structure on C~(GN) defined by (3.22) and (3.23)
with 3.12.1 and (3.152) becomes

f * g(O, X) = ( df-t~(x)(Y)f(O, X - Y)g(O, Y); (3.153)

11
IT- 1oT(X)

f * g(fi, y) = fi- n
T,- (r,(y»
df-t~,(y)(YI)f(fi, YYI)g(fi, YI- 1); (3.154)
-:=--:=
f*(O, X) = f(O, -X); (3.155)
f*(fi, y) = f(fi, y- l ). (3.156)
The reduced normal groupoid C* -algebra C;(G N ) is the closure of the *-algebra
C~(GN) in the norm (3.73); cf. Definition 3.6.3.
Let 'Jh be the ideal in C;(G N) generated by those functions in C~(GN) that

°
vanish at fi. The canonical map f t-+ [flh from q(G N) to q(G N)/'Jh is given by
=
[fhO f(fi, .). However, in view of the factor fi-n in (3.154), for fi #- this map
is only a *-homomorphism from q(G N ) to q(G) if we add a factor fi-n to the
definition (3.22) of convolution on G. Since we would like to identify q(G N )/'Jh
with q(G), in which convolution is defined in the usual, fi-independent, way, we

For fi °
should therefore renormalize the canonical projection.
= one has C*(G N )/'Jo ~ q«(!S), which in tum is isomorphic to Co«(!s*)
by the fiberwise Fourier transform (3.132). This motivates the definition of 'Po :
q(G N) ---+ Co«(!s*) for fi = 0, and 'Ph : C;(G N) ---+ C;(G) when fi #- 0, by
continuous extension from f E C~(GN) of

'Po(f) : 0 t-+ f(O, 0),

'Ph(f) : Y t-+ fi-nf(fi, y). (3.157)
Here f(O, 0) and f(O, X) are related as f(O) and I(X) are in (3.132).
Theorem 3.12.2. Let G =*
Q be a Lie groupoid, with associated Lie algebroid
(!S -=+ TQQ. The triple (q(GN), (Qth, 'PhhelR), where Qto = Co«(!S*), andQth = q(G)
for fi #- 0, is a continuous field of C* -algebras.
To prove this, we need some standard concepts in the theory of C* -algebras.
A primitive ideal in a C* -algebra Qt is an ideal that occurs as the kernel of an
irreducible representation of Qt.
Definition 3.12.3. The primitive spectrum Prim(Qt) of a C* -algebra Qt is a
topological space whose elements are the primitive ideals in Qt, and whose topology
 3 Lie Groupoids and Lie AIgebroids 311

is defined by the following closure operation: The closure of S C Prim(~) is the

set of all primitive ideals of ~ containing the intersection of the elements of S
(which is an ideal in ~).
The topology defined here is known as the Jacobson topology; it can be defined
in a much wider context. We let 1f) be the irreducible representation annihilating J;
i.e.,J = ker(1f). One seesthatJn ~ JintheJacobsontopologywhen1f).(A) = 0
for all sufficiently large n implies 1f)(A) = O. For example, when X is a locally
compact Hausdorff space, one may identify Prim(Co(X» with X by identifying
J x := (f I f(x) = O} with x. The Jacobson topology then coincides with the
original topology on X.
Lemma 3.12.4. Let fl be a C* -algebra, and let 1/1 : Prim(fl) ~ X be a continuous
and open map from the primitive spectrum Prim(fl) (equipped with the Jacobson
topology) to a locally compact Hausdorff space X. Define J x := n1/l-1(x); i.e.,
A E J x ijJ1f)(A) = Ofor all J E 1/1-1 (x). Note that J x is an ideal in fl.
Taking ~x = fl/Jx and Ox : fl ~ ~x to be the canonical projection, the triple
(f!, {~x, Ox }XEX) is a continuous field ofC*-algebras.
We omit the long and difficult proof of this lemma, and instead apply it, with
f! = C;(G N ) and X = I = R. In order to verify the assumption in 3.12.4, we first
note that Jo ~ Co(lR\{O}) ® C;(G), as follows from a glance at the topology of
GN • Hence Prim(Jo) = R\{O} x Prim(C;(G», with the product topology. Since
C*(GN )/Jo ~ Co(~*), one has Prim(C*(G N )/Jo) ~ ~*.
We need a second lemma, proved by straightforward definition-chasing.
Lemma 3.12.5. Let J be an ideal in a C* -algebra ~, and decompose
(3.158)
where Prim)(~) consists of those primitive ideals containing J, and Prim)(~)
is its complement. Then one has the homeomorphisms Prim)(~) ~ Prim(~/J)
and Prim)(~) ~ Prim(J). Moreover, Prim)(~) is closed and Prim\~) is open in
Prim(~).

We apply this lemma with ~ = C;(G N ) and J = J o. Then C;(G N )/Jo ~

Co(~*), and a glance at the topology of GN shows that Jo ~ Co(R\ to}) ® C:(G).
Thus the decomposition (3.158) reads
(3.159)

For example, when G = Q x Q ~ Q is a pair Lie groupoid, the right-hand

side of the decomposition (3.159) is T*Q U R\{O}. In that case the topology is
easily computed: The closure of a set (0, 141) is T* Q U (0, 141]. This illustrates the
fact that in general, the primitive spectrum is not Hausdorff.
Equation (3.159) with 3.12.5 does not provide the full topology on Prim( C;(GN »,
but it is sufficient to know that ~* is not open. If it were, R\ (OJ x Prim(G) would be
closed, and this possibility can be excluded using the convergence criterion men-
tioned after 3.12.3. Using (3.159), we can define a map 1/1 : Prim(C;(G N» ~ R
312 III. Groups, Bundles, and Groupoids

by 1/10) = 0 for all J E 18* and 1/I(n, J) = n for n =I- 0 and J E Prim(c;(G».1t is
clear from the preceding considerations that 1/1 is continuous and open. Using this
in Lemma 3.12.4,one sees thatJh is indeed the ideal in C:(GN) generated by those
/ E C~(GN) that vanish at n. Hence ~o :::::: C o(I8*) as above, and ~h :::::: C:(G)
n
for =I- O. Theorem 3.12.2 then follows from Lemma 3.12.4 and the argument
leading to (3.157). •
We now prove Theorem 3.11.4. Recalling the cutoff K in 3.11.3, for each / E
C~(I8*)we define a function Q(f) on GN by

Q(f)(O, X) := i(X);
Q(f)(n, Expw (X» := K(X)i(X/n);
Q(f)(n, y) := 0 Vy f. N;. (3.160)
It is clear from Definition 3.12.1 that Q(f) is smooth on GN • Although Q(f)
n,
does not have compact support in using (3.75) and (3.152) one may argue that
it lies in c;(GN)' Comparing (3.160), (3.157), and (3.135), we have Q':(f)+ =
rph(Q(f». Hence {Q': (f)+helR is a section of the continuous field of Theorem
3.12.2. Definition 11.1.2.1 then implies Rieffe1's condition 11.1.1.1.1, as well as
II.(1.4), which in turn implies von Neumann's condition 11.1.1.1.2.
To also cover functions of the type Q~ (f)-, one equips GN with a differ-
ent smooth structure, obtained by replacing Expw(X) in 3.12.1 by Expw(-X).
The original "+" smooth structure is equivalent to the modified "-" one by the
diffeomorphism (0, X) ~ (0, - X) and (n, y) ~ (n, y). •

Corollary 3.12.6. When 11.(1.3) holds, the field (c;(G N), {~h, rphhelR) o/Theo-
rem 3.12.2, the space §to = C~(I8*), and the map Q in (3.160) define a continuous
quantization 0/18* (cf 11.1.2.5).
We now see that the continuous field of Theorem II.2.6.5 is a special case of the
one in Theorem 3.12.2, with G given by the pair groupoid IRn x IRn, so that GN
is the tangent groupoid of IRn. More generally, the continuous field generated by
Wey1 quantization on a Riemannian manifold Q (cf. Theorem II.3.5.1) is given by
putting G = Q x Q in 3.12.2.
CHAPTER IV

Reduction and Induction

1 Reduction
1.1 Basics of Constraints and Reduction
We start with a geometric description of symplectic reduction in a rather general
form, and subsequently relate this to the notion of a constraint.
Recall Definition 1.2.4.1 and subsequent paragraph.

Definition 1.1.1. Let (S, w) be a symplectic manifold, and let C be a closed

submanifold of S. The null distribution Ne on C is the kernel of the restriction
We = t*w ofw to C (here t : C,-* S is the canonical embedding).
Note that although w is by definition closed and nondegenerate, its restriction to
C, while closed, may be degenerate. Namely, given a vector X E TuC, no vector
Y E Tu S for which w(X, Y) '# 0 may be tangent to C.
We denote the annihilator in T*S of a subbundle VeTS by Vo. For example,
~ consists of all I-forms a on S that satisfy a(X) = 0 for all X ENe. The
symplectic orthogonal complement in TS of V, on the other hand, is called V.L;
it consists of all Y E TS such that w(X, Y) = 0 for all X E V (assuming, of
course, that X and Y lie in the same fiber of TS). In this notation we obviously
have

Ne = TC n TC.L. (1.1)

Theorem 1.1.2. When the rank ofWe is constant on C. the null distribution Ne
is smooth and completely integrable; hence Ne defines the null foliation cl>e of
C. When the space

(1.2)
314 IV. Reduction and Induction

of leaves of this foliation is a manifold in its natural topology, there is a unique

symplectic form (IF on SC satisfying
(1.3)
where r := rc->sc maps a to the leaf of the null foliation in which it lies.
We omit the technical proof that.Nc is smooth (when the rank. of Wc varies on
C the null distribution is not necessarily smooth). Use of the well-known iden-
tity dwc(Xo, X), X2) = Xo wc(X) , X2) - ... - w([X), X2], Xo) for any 2-form
Wc, combined with dwc = 0, shows that.Nc is involutive. Consequently, it is
completely integrable by Frobenius's theorem.
Using the identity Lx = dLx + Lxd, one shows that w is invariant under any
flow along the leaves of <l>c. Thus we can define w C by
c - -
w (X, Y) := wc(X, y), (1.4)
where Xis a preimage of X in TC under r*. By construction, w C does not depend
on the particular choice of the preimage, since any two possible choices differ
by an element of .Nc , which is annihilated by Wc. The 2-form w C satisfies (1.3)
by construction, and is then seen to be closed because Wc is closed and r* is
surjective (alternatively, one uses the identity above). Finally, w C is nondegenerate
by construction; or note that its rank is equal to the dimension of SC . The uniqueness
of w C is obvious, for any w C satisfying (1.3) must satisfy (1.4). 0
We now pass to the elementary theory of constraints. In an expression of the
type "g ETC" one refers, of course, to the restriction of a vector field g to C.
Lemma 1.1.3. For f E COO(S, JR} the property df = 0 on C is equivalent to
gj E TCl..
Here gf is the Hamiltonian vector field of f; recall 1.(2.8}. The claim is
immediate from 1.(2.21}. •
Lemma 1.1.4. The property df E (TCl.}o is equivalent to gf ETC.
The condition df E (TCl.}o implies that Xf = 0 for all X E TCl.. Now take
X = gg (at some point), and use Lemma 1.1.3 and the equality gf g = -gg/. This
proves that~f ETC, because the map g f-+ ~g is surjective, since S is symplectic.
This argument works in the opposite direction as well. •
Definition 1.1.5. A function fP E COO(S, JR} satisfying fP = 0 on C is called a
constraint. A first-class constraint is a constraint satisfying dfP E (TCl.}o. A
constraint that is not first class is called second class.~
This definition is of great significance, as there is a fundamental difference
between situations with first- and second-class constraints.
Proposition 1'.1.6. Each ofthe following conditions is necessary and sufficientfor
afunction fP E COO(S, JR) to be afirst-class constraint (up to a possible constant):
1. Its Hamiltonian vector field gl{! lies in TC n TCl..
 1 Reduction 315

2. The Poisson bracket {({J, ({J'} with any other constraint ({J' is itself a constraint.
Moreover, at each point ofC the space TC n TCl- is spanned by the Hamiltonian
vector fields offirst-class constraints.
The first characterization is immediate from 1.1.5, 1.1.3, and 1.1.4. The second
follows from 1.(2.19), 1.(2.8), and the first one (using 1.1.3 once again). The final
claim holds because S is symplectic, so that f 1-+ ~f is surjective onto TC; then
use 1.1.3 and 1.1.4 in the opposite direction. •
Hence the Hamiltonian flow generated by the first-class constraints sweeps out
the leaves of the null foliation of C. In physics this flow is regarded as unphysical,
corresponding to the fact that the Hamiltonian equations of motion are underde-
termined on C. Passing from C to SC is then a means of eliminating redundancy
and indeterminism. Note that using the comment after 1.(2.16), one may reconfirm
the invariance of w under flows tangent to the null distribution.
There are four special cases of interest (which are neither exhaustive nor
mutually exclusive).
Definition 1.1.7. A submanifold C ofa symplectic manifold (S, w) is called
• isotropic when T C ~ T C l-;
• coisotropic when TCl- ~ TC;
• Lagrangian when it is at the same time isotropic and coisotropic, in other
words, when T Cl- = T C, so that Wc = 0;
• symplectic when TC n TCl- = 0, so that Wc on C is symplectic.
Some authors ascribe theomorphic status to Lagrangian submanifolds, but in
this book they hardly play a role. Locally a symplectic submanifold C may be
described as the set on which a collection of second-class constraints vanishes,
whereas a coisotropic submanifold is locally described as the null set of a set of
first-class constraints.
Definition 1.1.8. A weak observable on S is afunction f E COO(S, lR)for which
df r C lies in.N2:.
The restriction of a weak observable to C is evidently constant on the leaves of
the null foliation of C.
Proposition 1.1.9. A smooth function f is a weak observable
1. iff~f lies in TC + TCl-;
2. iffits Poisson bracket with any first-class constraint vanishes on C.
The proof of the first characterization is like that of 1.1.4, adding the fact that
(TCnTCl-l equals TC + TCl-. The second follows from the equation {({J, f} =
~rpf in combination with (1.1) and 1.1.6.1. •
Proposition 1.1.10. When C is coisotropicaUy embedded in S, the collection Ql~
of weak observables is a Poisson algebra.
316 IV. Reduction and Induction

Since Qt~ is characterized by a differential condition, it is obvious that it is closed

under the pointwise product as well as under linear operations. Closure under the
Poisson bracket is proved as follows. In the coisotropic case, Proposition 1.1.9
states that f is a weak observable iff ~f ETC. To check whether ~{f.g} E TC
when ~f and ~g are, we compute ~(f.g}qJ for an arbitrary constraint that is first class
by assumption. Using 1.(2.8), the Jacobi identity, and then applying 1.1.9.2 twice,
one shows that ~{f.g}qJ = 0 on C. •
Now assume that the reduced space SC constructed in 1.1.2 is a manifold.
Definition 1.1.11. The reduced representation ll c (f) of a weak observable f
is the unique element of COO(Sc, JR) that satisfies
r~-,>scllc(f) = f r c. (1.5)
The Poisson algebra of observables of the system whose constraint hypersurface
is S is then defined as
(1.6)
Using the tubular neighborhood theorem, one easily sees that any smooth function
on a submanifold C may be extended to a smooth function defined on a neighbor-
hood of C. When C is closed, a further smooth extension to S is always possible.
Hence Qtc ~ COO(S, JR) as Poisson algebras.
It follows from 1.1.9 that when C is coisotropic, so that TC + TC.L = TC,
the Hamiltonian flow of f does not leave C when it starts there, and that this flow
projects onto the flow of ll c (f) in SC. For general C, it can be shown that one
can always decompose f = fl + h (at least in a neighborhood of C) such that
fl satisfies ~ft E TC (so that its Hamiltonian flow stays in C) and h vanishes on
C. In physics one is given a Hamiltonian h on C that should be a weak observable
by construction (of C), and one subsequently tries to extend h to S such that ~h is
tangent to C.

1.2 Special Symplectic Reduction

We are now going to describe a special case of the construction in the previous
section that includes many physically relevant examples.
Definition 1.2.1. In special symplectic reduction one starts from:
• A pair of symplectic manifolds (S, ws) and (Sp, wp).
• A Poisson manifold P (we denote the same manifold equipped with minus the
Poisson bracket by P-).
• A pair of Poisson morphisms J : S --+ P- and Jp : Sp --+ P.
The total space is then S = S x Sp, equipped with the symplectic form w :=
Ws + w p , and the constraint manifold C in S is
C = S *p Sp := {(a, a) E S x SpIJ(a) = Jp(a)}. (1.7)
 1 Reduction 317

This situation is denoted by

J Jp
S~ P+--Sp' (1.8)
The reduction of S by C is described by the following
Theorem 1.2.2. Let either J* or (J p)* (or both) be surjective at all points relevant
to S *p Sp. Then S *p Sp is a coisotropic submanifold of S x Sp, and Wc is (locally)
of constant rank.
The null distribution Nc of S *p Sp is spanned by the collection of vector fields
~f' where f E C'O(P, R.) and
(1.9)
Here the Poisson bracket is the one corresponding to the symplectic form w.
We first show thatC is asubmanifold. Define f : Sx Sp ~ P x Pby f(a, a) :=
(J(a), Jp(a», and let D be the diagonal in P x P. Then C = f-I(D) and
f*-I1(p,p)D = 1(u,a)C whenever J(a) = Jp(a) = p. The surjectivity condition
implies that f intersects D transversally, which in tum guarantees that C is a
submanifold of S.
We next prove that C is coisotropic. Let X E Tu S and Y E Ta Sp; then X + Y E
T(u,a)C iff J*X = (Jp)*Y. The dimension of T(a,a)C at any point (a, a) E C equals
dim S + dim Sp - (rank J*)(a), so that
dim(T(u,a)C.L) = (rank J*)(a). (1.10)

Let Mu,a) denote the linear span of the collection of vector fields ~f taken at (a, a),
where f runs through COO(P, R.). Then
dim(Mu,a» = (rank J*)(a). (1.11)

We now argue that Mu,a) ~ T(u,a)C.L, so that(1.11) and (1.10) imply that
(1.12)
Namely, let X + Y E T(u,a)C, as above; then, since J*X = (Jp)*Y, one has
w(u,a)(X + Y, ~f) = d(J* f - J; f)(u,a)(X + y) = O.
Moreover, Mu.a) C 1(u.a)C by a similar calculation, which uses Proposition
1.2.3.5. Therefore, according to (1.12) the submanifold C is coisotropically im-
mersed in S x SP' and one has N = N c . It then follows from (1.11) and the
condition stated in the theorem that Wc has constant rank on each connected
component of S x Spo The above argument is symmetric in J and J p ' •

We are therefore in a position to apply Theorem 1.1.2, obtaining a reduced

symplectic space (Sc, wc), which we assume to be a manifold. To indicate the
dependence on the given data, we will denote the reduced space by (sj, wj),
where
(1.13)
318 IV. Reduction and Induction

Here <I> = <l>c is the null foliation generated by Nc . When S and J are fixed, we
sometimes simply write (SP, wP) for (sj, wj). The collection of weak observables
is then called ~~.
It follows from 1.1.5 or 1.1.6 that the constraints defined by C (cf. (1.7» are
precisely the functions of the form J* f - J; f, where f E COO(P, R).
Inspired by the theory of von Neumann algebras, we define the Poisson
commutant of some subspace ~s c COO(S, R) by
~~ := {g E COO(S, R) I {j, g} = 0 Vf E ~s)}. (1.14)

It follows from the Jacobi identity and the Leibniz rule that ~~ is a Poisson algebra
(even when ~s isn't). The operation ~s f-+ ~; plays the role of the "weak closure"
of ~s; the previous remark implies that ~~ is always a Poisson algebra. In general,
~s may be strictly contained in ~; even when the former is a Poisson algebra.
Similary, one defines the Poisson center of ~s as ~s n ~~. This may not be a
Poisson algebra, but its "weak closure" ~; n ~~ is.
An important subspace ofCOO(S, R) is J*COO(P, R)'. This maybe regarded as a
Poisson subalgebra of COO(S x SP' R) under the obvious embedding of COO(S, R)
in the latter. Combining 1.1.8 and 1.1.5 (or 1.1.9 and 1.1.6), one infers that
rcOO(p, R)' ~ ~~. (1.15)
Hence Definition 1.1.11 applies. We write
rrj : rCOO(p, R)' ~ COO(SP, R),
or simply rr P , for rrc. Denoting a point in SP by an equivalence class [a, a]ct> under
the null foliation, one has simply
(1.16)
Because f E J*COO(P, R)', this is independent of the choice of a in the given
equivalence class. The same construction applies, of course, to Poisson subalgebras
~s of J*COO(P, R)'.

Corollary 1.2.3. In the context of Definition 1.2.1, suppose one has a second
Poisson manifold P2 and a Poisson map h : S ~ P2 such that JiCOO(P2, R) ~

h JI Jp
P2 •
S PI • Sp

j
~ sPI
reduction

FIGURE 1. Special symplectic reduction; Sf := S~l etc.

 1 Reduction 319

J*CXJ(P, R)'. The map JP : SP ---+ P2 defined by

JP([a, a]<I» = J2 (a) (1.17)

is well-defined, and is a Poisson map.

Relabeling (J, P) as (it, Pd for clarity, and writing Si
and Ji
for SP = Sj,
and J P, respectively, we can summarize this situation pictorially as in Figure 1.
Equivalently, there is a reduced representation rri
:= (Ji>* of C OO (P2, R) on SP
rr
that is given in terms of P by

rr2P = rr P 0 J*
2' (1.18)

1.3 Classical Dual Pairs

In this section we define various duality relationships of increasing strength
between Poisson manifolds. The strongest of these will relate the respective
representation theories of these manifolds to each other.
A foliation <Il of S is always understood to be smooth in the sense that its
associated distribution T <Il (consisting of the vectors in T S which are tangent to
the pertinent leaf of <Il) is smooth as defined in 1.2.4.1.
Definition 1.3.1. A symplectically complete foliation <Il ofa symplectic manifold
S is afoliation with the property that the distribution T <Il.L is completely integrable.
Here is an alternative characterization.
Proposition 1.3.2. A foliation <Il is symplectically complete iff the space
COO(S, R)<I> of all smooth functions on S that are constant on each leaf of <Il is
a Poisson algebra.
A function f is constant along <Il iff df E T<Il°. As in 1.1.4, this is equivalent
to ~f E T<Il.L. As in 1.2.4, when COO(S, R)<I> is a Poisson algebra one uses 1.(2.9)
and Lemma 1.2.4.2 to show that the distribution T<Il.L is completely integrable.
Conversely, complete integrability implies that T<Il.L is involutive, so that the
first step of the proof establishes the claim in the opposite direction. D
The foliation generated by T <Il.L is called <Il.L. Recall (1.14).
Proposition 1.3.3. Let <Il be a symplectically complete foliation, and assume that
the functions in COO(S, R)<I> separate the leaves of<ll.
Then (COO(S, R)<I>)' = COO(S, R)<I>.L.
As before, one shows that f E COO(S, R)<I>.L implies ~f E T<Il. Hence, without
the additional assumption, one has COO(S, R)<I>.L ~ (COO(S, R)<I>)' by 1.(2.8).
Let g E (COO(S, R)<I>)'; by 1.(2.8) this is equivalent to ~gf = 0 for all f E
COO(S, R)<I>. The assumption now implies that ~g E T<Il. Since <Il is smooth, so is
<Il.L. Hence any two points in a connected leaf of <Il.L can be joined by a (piecewise)
smooth curve c(·). Using 1.(2.21) one computes dg(c(t))/dt = w(~g, c(t». This
vanishes when~g E T<Il, so thatg is constant along c(·), sog E COO(S, R)<I>.L. •
320 IV. Reduction and Induction

The assumption in 1.3.3 is satisfied when the leaf space S / <I> is a manifold.
We now look at the case where P is a Poisson manifold, and J : S ---+ P- is a
Poisson map for which the level sets of J define a foliation <I> of S; that is, the
leaf of <I> through a E Sis J-l(i(a». This is, for example, the case when J is a
submersion, which guarantees that each subspace J-I(a) is a submanifold of S.
In addition, we assume that S / <I> is a manifold. When all this holds, one has the
equality
(1.19)

Since J*COO(P, lR) is evidently a Poisson algebra, the foliation <I> is symplectically
complete by Proposition 1.3.2.
The situation is particularly neat when the associated foliation <I>.L is itself given
by the level sets of a Poisson map h : S ---+ P2 • In view of the symmetry between
(J, P) and h P2 we relabel the former as (iI, PI) in what follows.
Definition 1.3.4. A classical dual pair (S, PI. P2, h h) consists of a con-
nected symplectic manifold S and a pair of Poisson manifolds PI, P2, together
with Poisson maps J I : S ---+ P I- and h : S ---+ P2, such that:
1. The level sets of J I and h define foliations <1>1 and <1>2 of S, respectively, with
the property that <1>2 = <l>t (and hence <I> I = <l>t).
2. The leaf spaces S / <I> I and S / <1>2 are manifolds.
3. The maps J I and h are surjective submersions.
4. The level sets Jil(al) and J2- I(a2) are connected for all ai E Pi (i = 1,2).
5. The level sets J1-I(al) and J2- I(a2) are simply connectedfor all ai.
6. The maps J I and h are complete.
We denote classical dual pairs by

(1.20)
using this notation also when not all of the above conditions are satisfied.
It follows from 1.3.3 and (1.19) that when 1.3.4.1-3 are obeyed one has
J;COO(PI' lR)' = J;C OO (P2, lR),
J;C OO (P2, lR)' = J;COO(PI' lR). 0.21)
As in the comment after (1.19), we infer that <1>1 and <1>2 are symplectically
complete.
Lemma 1.3.5. When conditions 1, 3, and 4 in 1.3.4 are satisfied, the foliation <1>2
(or <1>1) coincides with the foliation defined by all Hamiltonian vector fields of the
form ~J: j, (or ~J; j), where f E COO(PI , lR) (or f E C OO (P2, lR»).
This is immediate from the proof of 1.3.2.
Corollary 1.3.6. In a classical dual pair with connected leaves there is a bijective
•
correspondence LI ++ L2 := h(J-I(L» between the symplectic leaves LI and
L2 in PI and P2, respectively.
 I Reduction 321

Givena E Ph Lemma 1.3.5 for <1>1 shows that the leaf JI-I(a) of <1>1 is generated
by the Hamiltonian flow of the vector fields glr f' where f E C OO ( P2, JR.). By 1.2.3.5
and 1.2.4.3 the set h(JI-I(a» is a (connected) symplectic leaf in P2. When a' lies
on the same leaf as a, there is a (piecewise) smooth Hamiltonian curve c in PI that
connects a' and a. Using 1.2.4.3 once again, and subsequently 1.3.5 for <1>2, we
infer that JI-I(C) lies in a single leaf of <1>2. Hence h(JI-I(a» = h(J,I(a'», and
».
J I- I (L I)/<I>2 ~ J 2(J I- I (L I Thus we obtain a bijection between the symplectic
leaves reached in this way. Because J I and h are surjections, all symplectic leaves
are included. •

We saw in 1.2.6 that a symplectic leaf in P may be regarded as an irreducible

representation of the Poisson algebra COO(P, JR.), so that (up to possible cov-
ering spaces of the symplectic leaves in question) Corollary 1.3.6 expresses a
bijective correspondence between the irreducible representations of two Poisson
algebras connected by a classical dual pair. This result can be generalized to all
representations. The following concept is central to this generalization.

Definition 1.3.7. Two Poisson manifolds PI, P2 are called Morita equivalent
when theyformpartofa classical dual pair (S, PI. P2, JI, h).
Despite the terminology, this definition fails to define an equivalence relation
in the class of all (finite-dimensional) Poisson manifolds, because not all Poisson
manifolds are Morita equivalent to themselves.

Proposition 1.3.8. Morita equivalence defines an equivalence relation in the

subclass of all Poisson manifolds that are Morita equivalent to themselves.

Reflexivity (i.e., P ~ P) being satisfied by definition, symmetry (that is, PI ~

P2 implies P2 ~ PI) holds because from the diagram P2 1;:.. S ~ PI one obtains
PI 1..!- S- ~ P2. Finally, transitivity is true by the following argument. When
M M
PI '" P2 by SI and P2 '" P3 by S2, so that one has
III S
P I +--- ll2 n h2 S l23 n
I ----+ '2 + - 2 ----+ '3,

one obtains a symplectic manifold Sri

by special symplectic reduction from the
middle three spaces in the diagram. Using Corollary 1.2.3 with respect to Jll and
h3, this leads to the classical dual pair PI +- Sri
-4 P3. •

Here are some simple examples of Morita equivalence.

Proposition 1.3.9.
• If S is a connected and simply connected symplectic manifoLd, and P is a
connected manifold with the zero Poisson structure, then S x P is Morita
equivaLent to P.In particular, S is Morita equivalent to a point.
• Two connected symplectic manifolds SI and S2 are Morita equivalent iff their
fundamental groups are isomorphic.
322 IV. Reduction and Induction

In the first case the pertinent classical dual pair is

Sx P /.!- S x T* P ~ P,

with h = «(1), <pp->p 0 «2» and h = <pp->p 0 «2) (here <(i) is the projection
onto the ith variable).
We pass to the second example. Let (SI, wd and (S2, W2) have isomorphic
fundamental groups Jl'1(SI) ~ Jl'2(S), with isomorphism ¢ : Jl'1(Sd -+ Jl'2(S),
and denote the universal covering spaces by 5i • Then Jl'1 (S) acts on 51 X 52 by
x : (UI, U2) ~ (XUI, ¢(X)U2). Since Jl'i(Si) is discrete, the form Wi := rii-,>Si Wi on
5i is symplectic (i = 1,2). Equip 51 x 52 with the symplectic formw12 := W2 -WI'
The quotient S12 := (51 x 5 2)/Jl'I(SI) has a unique symplectic form Wl2 whose
pullback to 51 x 52 under the canonical projection is W12.
The classical dual pair is now given by

with the obvious projections J i : SI2 -+ Si. One easily verifies all pertinent prop-
erties. For example, the completeness of the Ji follows from the path lifting lemma
of homotopy theory. Also, for each UI E SI the leaf JI-I(UI) is homeomorphic to
52, which is indeed connected and simply connected; analogously, J2- I(U2) ~ 51.
Conversely, assume that one has a classical dual pair S2 ::- S ~ SI. Some
algebraic topology then shows that Jl'1 (Si) ~ Jl'1 (S) for i = 1, 2. 0

1.4 The Classical Imprimitivity Theorem

We now state and prove the classical imprimitivity theorem.
Recall the definition of completeness after 1.2.6.1. In the situation of 1.3.7
the pullbacks Jt and J{ are representations of COO(P I , JR) and C OO (P2 , JR) on
S, respectively (cf. 1.2.6.1).

Theorem 1.4.1. Let PI and P2 be Morita-equivalent Poisson manifolds. There

is a bijective correspondence between the representations of C OO ( PI, JR) and
C OO (P2, JR) preserving irreducibility. Equivalently (by Corollary 1.2.6.5), there is
a bijection between complete Poisson maps Jp : Sp -+ PI and Ja : Sa -+ P2 pre-
serving symplectic leaves (or their covering spaces). This correspondence arises
as follows.
Let P2 ::- S ~ PI be a classicaL duaL pair impLementing the Morita equivalence
between PI and P2. When Ja : Sa -+ P2 is a compLete Poisson map (where Sa
is sympLectic), there exist a symplectic manifoLd Sp and a complete Poisson map
Jp : SI -+ PI such that Sa is symplectomorphic to the reduced space Si obtained
by special symplectic reduction (and J a ~ Jt'), where Si and Jf are defined in
Figure 1.
In the opposite direction, given a complete Poisson map Jp : Sp -+ PI (where
Sp is symplectic) there exist a symplectic manifold Sa and a complete Poisson map
 I Reduction 323

Jp
S PI • Sp

S- P2 •
~ I
1 · - SU
Sp·-

PI
~•
2
SU
2

FIGURE 2. Classical imprimitivity theorem: Sf ~ Sp and J2 ~ Jp

Ju : S2 --+ P2 such that Sp is symplectomorphic to the reduced space (S-)~ (and

Jf ), dejined as in Figure 1, but now with respect to PI :::.. S- ~ P2.
Jp ::::::
Taking Su = Sf
and Ju = Jf asjustdejined, one has (S-)~ :::::: Sp and Jf : : : Jp.
Conversely, taking Sp = (S-)~ and Jp = Jf, one has sf::::::
Su and Jf :::::: Ju .

See Figure 2. We write S~ for (S-)~.

The existence of the Poisson maps Jf : sf --+ P2 and Jf : S2 --+ PI follows
from Corollary 1.2.3. The reinterpretation of the data (1.20) as PI :::.. S- ~ P2
will be used again in the proof of Theorem 1.8.1.
Starting from Jp : Sp --+ PI, constructing Jf : Sf --+ P2 as indicated, which we
relabel as Ju : Su --+ P2, and subsequently defining Jf : S2 --+ PI. the essence
of the proof of the theorem consists in the construction of a symplectomorphism
cp : S2 --+ Sp.
Consider the space S*P, SP*P2 S, defined as the subset of Sx Sp xS- consisting of
triples (ai, a, a2) satisfying J I (al) = Jp(a) and h(al) = h(a2). By construction,
the space SU is obtained from this by a double foliation (cf. 1.2.2). The first
one, <1>/ on S x Sp, is generated by the Hamiltonian vector fields defined by
the functions Jj f - J; f, where f E Coo(PI, JR); we denote points of the leaf
space SP by [ai, al<!>/. The second foliation, <1>11 on S x S-, is generated by
the Hamiltonian vector fields defined by the functions «i/i g - <t2)Ji g, with
g E C oo (P2, JR) and <(I), «2) the projections onto the first and second variables
in S x S-, respectively. Its leaf space has elements [al,a2]<I>1/" By (1.17) the
p

equivalence classes [(ai, al<!>l' a2l<!>11 correspond to elements of S1.

Take a triple (al,a,a2) E S *P, Sp *P2 S, projecting to [(al,a]<I>l'a2l<!>11" By
Lemma 1.3.5 the Hamiltonian vector fields defined by the functions Jj f, with f E
Coo (PI, JR), generate a foliation that coincides with the foliation by the connected
level sets of h. Recall that h(ar) = h(a2); for simplicity we initially assume
324 IV. Reduction and Induction

that there is a smooth Hamiltonian curve CI1I~1120 in S connecting 0"1 and 0"2.
whose tangent vector is ~J: f. for some f E cOO(PI. R). Using a cutoff function if
necessary, we may pick an f with compact support, so that ~f is complete in PI.
Let a2 = CI1I~112(tO) (and, of course, al = CI1I~112(0». We can move a in Sp along
the Hamiltonian curve Ca (-) that is generated by - J;
f and starts at Ca (0) = a.
Let i'x := ca(to); this makes sense, since by assumption the map J p is complete.
The definition of [.. .]<1>1 then implies that
(1.22)
Let now a2 move around a closed Hamiltonian curve CI12 ->112(-)' generated by
some J;g, where g E COO (PI , R), and, say, CI12~112(0) = CI12 ->112(1) = a2. We put
Cl1r?112 := {C I12 -> 112 (t)1 t E [0, I]}. According to 1.2.3.5 with J = JIo the curve
CIO := JI(CI12~112('» in PI is Hamiltonian, being generated by g. The curve
CI := {c1(t)1 t E [0, I]} is closed, since CI(O) = CI(1) = JI(y). Our assumption
that the level sets of lz are simply connected implies that C112 -> 112 , and hence Clo is
contractible. Using 1.2.3.5 with J = Jp , one infers that the Hamiltonian curve cp
in Sp that is generated by J;g and starts at cp(O) = ii, covers CI. The latter being
contractible, the monodromy lemma of homotopy theory (or a direct argument)
implies that cp (1) must be closed; i.e., c p (1) = ii. This, in tum, guarantees that
ii is independent of the Hamiltonian path from a I to y. which in addition implies
that ii is independent of the choice of (aI, a) in the class [ai, a]<I>/. Finally, it is
clear from the construction that ii does not change when a different representative
of the given class [.. .]<1>/1 is chosen.
When the Hamiltonian curve CI1I -?I12(') is merely piecewise smooth, one simply
uses the above argument for each smooth piece, with the same conclusion. Thus
we may define
(1.23)
where i'x is determined by (1.22). We have just seen that this map is well-defined.
We now use Lemma 1.3.5 once more, this time saying that the foliation determined
by the Hamiltonian vector fields ~Ji f' where f E c OO (P2 , R), coincides with the
foliation by the level sets of h Since J I (a2) = Jp(ii), we infer that the equivalence
class [.. .]<1>/1 is uniquely determined by ii, and conclude that ({J is a bijection (this
could alternatively be established by a dimension count; cf. the proof of 1.2.2).
Moreover, ({J is a Poisson map, as is obvious from Theorem 1.2.2. Hence ({J is a
symplectomorphism. By (1.17) we have
Jf ([[a2. ii]<I>1' 0"2]<1>/1) = iJ(a2)'
But JI (a2) = Jp(ii), so that under the above symplectomorphism the map J2 is
transformed into Jp •
The construction can, of course, be carried out in the opposite direction as well.
All relevant constructions preserve completeness, and the proof of bijectivity is
finished.
When Sp = L is a symplectic leaf of PI and Jp = l is the inclusion map, the
fiber product S *P1 L coincides with JI-I(L), and the foliation <1>/ is nothing but
 I Reduction 325

the foliation <1>2 by the level sets of h By (1.17) we have J{([a]<I>2) = }z(a)
for a E J1-1(L); since }z(Jil(L» ::::: J 1- 1(L)/<1>2 = sf, we conclude that J{
injectsSf into P2. The proof of Corollary 1.3.6 then shows that }z(JI-I(L» is a
symplectic leaf in P2, so that sf is symplectomorphic to such a leaf by J{.
When Sp = L is a covering space of L, one analogously obtains that covers Sf
}z(JI-I(L». Theorem 1.2.6.7 then shows that the bijection preserves irreducibility,
as claimed. •

1.5 Marsden-Weinstein Reduction

We will now look at a further specialization of the reduction procedure in 1.2;
under suitable assumptions this will produce examples of classical dual pairs as
well. Throughout this section H is a Lie group, and until the last paragraph we
consider a strongly Hamiltonian H -action on a symplectic manifold (S, w), with
Co-equivariant momentum map J : S ---+ ~~.
Lemma 1.5.1. The map J* : T(fS ---+ TJ«f)~* ::::: ~* is surjective iff the stabilizer
H(f is discrete.
By III.(1.7) an element X E ~ is annihilated by the image ofJ*(a) iffh(a) = O.
Hence the dimension of H(f equals the dimension of the annihilator of this image.
This common dimension is 0 when J*(a) is surjective. •
We now take a coadjoint orbit 0 in ~*, and specialize Definition 1.2.1 to the
case P = ~~ and Sp = 0+, with J p = to the inclusion map (we will omit the "+"
whenever it is convenient). In other words, the situation is
J
S ---+ h* to /,...,
'J+ ~ v+. (1.24)
Recall that a map between two manifolds is proper when the inverse image of
every compact set is compact.
Definition 1.5.2. A Lie group action L : H x S ---+ S on a manifold is called
proper when the map (x, a) ~ (xa, a)from H x S to S x S is proper. In other
words,for sequences {an} in Sand {x n } in H the convergence of{an} and of {Xnan}
must imply that {xn} has a convergent subsequence.
Roughly speaking, this means that nearby points in S can be mapped into each
other by x E G only if x is near e. We collect some relevant properties of proper
group actions.
Proposition 1.5.3.
• The action of a compact group is always proper.
• Under a proper action the stabilizer H(f of every point a E S is compact.
• The quotient of a manifold by a proper and free action is a manifold.
The first claim is immediate from the definition. The second follows because
(a, a) E S x S is compact, so that its inverse image {(H(fa, a)} must be compact,
326 IV. Reduction and Induction

too. The third statement is a nontrivial theorem of differential geometry, whose

proof we omit. 0
Theorem 1.5.4. Given a strongly Hamiltonian action of a Lie group H on a
symplectic manifold (S, w), with Co-equivariant momentum map J : S -+ ~~,
assume that there is a () E 0 such that H acts freely and properly on J-I«().
Then the hypothesis in Theorem 1.2.2 is met, and the reduced space S? := S~o is
a symplectic manifold. There are symplectomorphisms

(1.25)
where HO is the identity component of H, and Ho is the stabilizer of () under the
coadjoint action.
Since the H-action is free, Lemma 1.5.1 and the equivariance of J (cf. III.1.2.5)
imply that the hypothesis of Theorem 1.2.2 is satisfied.
We first take 0 = to}, in which case (1.9) and preceding text implies the first
isomorphism in (1.25); recall that the functions J* X generate the HO-action. For
general orbits we perform a shifting trick that reduces the situation to the zero orbit.
Namely, we consider S = S x 0, on which H acts by the product of the original
action and the coadjoint action. By III.1.4.6 the momentum map j : S -+ ~~ is
given by j(a, () = J(a) - (), so that S *~. 0 = j-I(O). This trick establishes the
first isomorphism in (1.25) in the general case. The second isomorphism follows
from the transitivity of the H -action on 0 and the equivariance of J. •

It should be mentioned that although under the stated assumptions J-1(0) can
be shown to be coisotropically embedded in S, the quotient J-1(0)1 HO does not
coincide with the quotient J- 1(0)/4> by the null foliation 4>; in fact, when H
and Ho are connected, it can be shown that r l (O)/4> ~ (1-1(0)1 HO) x O.
Also, unless Ho is connected, the space J-1«()/(Ho n HO) is not the same as the
quotient of J-I«() (which is equally well coisotropically embedded in S) by its
null foliation, for the latter quotient is J-1«()/(Ho n HO)o.
The reduced space J-1(0)1 H is called the Marsden-Weinstein quotient of S
with respect to 0 (and the given group action). As expressed by (1.25), when H is
connected, this quotient is essentially the same as the reduced space S?
To show
how they are related when H fails to be connected, we form the discrete group
1fo(H):= HIHo, and see that

S? := J-'(O)I H ~ S? 11fo(H). (1.26)

Hence S? is symplectic, as is S? (this follows only because 1fo(H) is discrete).

One may look at Marsden-Weinstein quotients from a different angle.
Theorem 1.5.5. Let the H -action on S be free, proper, and strongly Hamiltonian.
Then the space S1 H is a manifold with a unique Poisson structure for which the
canonical projection T : S -+ S 1 H is a Poisson map.
The symplectic leaves of S 1 H are the connected components of the Marsden-
Weinstein quotients J-1(0)1 H, where 0 C ~* is a coadjoint orbit.
 1 Reduction 327

The Poisson structure on SI H is obtained by identifying COO(SI H, R) with the

space COO(S, R)H of H -invariant smooth functions on S; this space is a Poisson
algebra, since the symplectic and hence the Poisson structure on S is H -invariant
(see III. 1.2). The proof of the second claim uses the following
Lemma 1.5.6. Let <I> I and <1>2 be the foliations of S by the level sets of J and by
the H -orbits, respectively. Then <1>+ = <1>2 (hence <l>f = <1>1)'
By the definition III.(1.7) of the momentum map, the distribution T<I>2 is gen-
erated by the vector fields ~Jx' where X runs through f). The claim is then obvious
from 111.(1.8). •
Corollary 1.5.7. Under the assumptions of 1.5.4, specializing (J .20) to the case
SI H :- S ~ f)~ satisfies conditions 1.3.4.1-3.
Continuing the proof of 1.5.5, Lemmas 1.3.5 and 1.5.6 show that T<I>I is spanned
by Hamiltonian vector fields of the form ~" t, where f E COO(SI H, R}. Using the
equivariance of J, one sees that 1'(J- I «(}» = J-I(O)I H; cf. (1.25). Since l' is a
Poisson map, Proposition 1.2.3.5 applies, showing that all Hamiltonian vector fields
in S I H are tangent to the subs paces J -I (0) I H. It follows that each connected
component of the latter must be a symplectic leaf, and it is plain that all leaves are
then of this form. •
For later use we collect more precise information about this situation.
Proposition 1.5.8. In the situation of 1.5.4 and 1.5.7 the maps l' and J are
complete.
Pick a function g E COO(SI H, R) with complete Hamiltonian flow [a ]HO, and
take any smooth curve cO covering [a]HO. For the Hamiltonian flow a(·) of
1'*g E COO(S, R)H through c(O) we make the ansatz a(t) = x(t)c(t), where x(·)
is some curve in H starting at the identity. Using the equation of motion 1.(2.11)
(with h replaced by 1'*g) and the H-invariance of ~r'h, one obtains the equation

~!-lX = i: - ~r'h ( 1.27)

along cO; here ~1(a) := -d Exp(AX)a IdAIA = 0 for X E f) (cf. III.(2.9», and
we have written X-I X for the cumbersome (Lx-l )*x; cf. 111.(1.43).
Applying 1.2.3.5, one sees that 1'*(i: - ~r'h) = 0, so that there exists a curve
XO in f) for which i: - ~r'h = ~1 along c(·). Comparing with (1.27), we see that
x (t)-I x(t) = X (t). This can be solved for all t, so that a (t) exists for all t as well.
Hence l' is complete.
We tum to the question of the completeness of the momentum map J. Choose a
function f E COO(f)*, R) with complete Hamiltonian flow in f)~. Denote the flow
of J* fin S by a(·). According to Lemma 1.3.5, the J* Xgenerate the group action
on S, so we can make the ansatz a(t) = x(t)a for some flow xO in H. Applying
J to both sides and using its equivariance as well as Proposition 1.2.3.5, we obtain
x(t)(} = (}(t), where (}O is the Hamiltonian flow of f in f)~. Using III.(1.54)
(with r = 0) and III.(1.58) we infer that x(t) = 1'pll.-+H(a(t», where aO is the
328 IV. Reduction and Induction

Hamiltonian flow on T* H (with its canonical symplectic structure) generated by

(JL)* f.
The momentum map JL : T* H -+ ~"- for the left action of H on T* H coin-
cides with the projection r P : T* H -+ (T* H)I H, when we identify the quotient
(T* H)I H under the right action with ~* (using the right trivialization of T* H).
Hence J L is complete by the first part of the proposition.
It follows that xO and therefore a(·) is defined for all t. •

We now recognize that the projective Hilbert space JP"H, looked at in 1.2.5 as a
symplectic leaf in the Poisson manifold 1-{* 1U (1), may alternatively be described
as a Marsden-Weinstein quotient. The group U(l) acts on 1-{ by z : \II H- z\ll
(where z E C with Izl = 1). Using 1.(2.35), 1.(2.37), and m.(1.8), as well as the
standing convention T = -i for the single generator T of u( 1), one easily derives
that the momentum map J = h for this action is

J(\II) = (\II, \II). (1.28)

Thus JP"H :::::: J- 1(1)1 U(l); cf. the closing comment 0fI.2.5.
To close this section, we remark that the Marsden-Weinstein reduction pro-
cedure is easily generalized to the case where the momentum map J is
CoY -equivariant; cf. m.1.2.5. Instead of a coadjoint orbit in ~* we simply take
a CoY -orbit (cf. IlL 1.4.4), and proceed as in the Co-equivariant case.

1.6 Kazhdan-Kostant-Sternberg Reduction

We can now see the theory in m.2.3 in the light of Marsden-Weinstein reduction.
Indeed, given a principal bundle P(Q, H, r), the hypotheses in 1.5.4 evidently
apply to the case S = T* P, and we infer that the spaces ('f~p)o defined in IlI.(2.51)
are nothing but Marsden-Weinstein quotients.
We may generalize the construction of the reduced space (rp)o, in replacing
the inclusion LO : 0 -+ ~~ by a general Poisson map Jp : Sp -+ ~~, where Sp
is symplectic. Such maps usually come from a strongly Hamiltonian H -action on
Sp whose associated momentum map is minus J p ; we assume that we are indeed
in this slightly more special situation. Hence we fill out Figure I as in Figure 3,
where J is the momentum map for the H -action on T* P pulled back from the
given action on P, and rR := rT*P-->(T*P)/H is the canonical projection. We refer
to this situation as (generalized) Kazhdan-Kostant-Sternberg reduction.
We may look at this instance of special symplectic reduction as a special case
of Marsden-Weinstein reduction: As in the proof of 1.5.4 we take S := T * P x Sp ,
equipped with the product H -action. The associated momentum map] : S -+ ~"'­
is simply] = J - Jp , so that the submanifold T*P *~. Sp of T*P x Sp (defined
as in (1.7» coincides with ]-1(0). Hence for connected H the reduced space is

(1.29)
 1 Reduction 329

7: R J
..- - -
I
(T*P)/H -+-- T*P - -..... ~~

reduction

(T*P)P

FIGURE 3. Kazhdan-Kostant-Stemberg reduction

We will use the right-hand side as the definition of the reduced space also when
H is not connected; as in (1.26) we write the reduced space as
(1.30)
Theorem m.2.5.2 then generalizes: The group Qp defined in III.(2.74) acts on
CPP)P in strongly Hamiltonian fashion by the obvious generalization ofIII.(2.75}-
I1I.(2.76). Theorem I1I.2.3.7 generalizes to
Theorem 1.6.1. E~ connection A on a princ!I!E1 bundle P(Q, H, 7:) defines
a projection 7:A : (T*P)P ~ T*Q that makes (T*P)P a bundle over T*Q with
typical fiber Sp.
Using the (A-dependent) factorization III.(2.47), we denote points of T*P by
triples (x, a, e); recall that x E P, a E T*Q, and e E ~*, with the constraint
7:T*Q-->Q(a) = 7:P-->Q(x). From m.(2.48) and Definition III.2.3.2 we have
Cp"'Pt ~ (P *Q T* Q x Sp)/ H, (1.31)
where the H-action is h : (x, a, a) ~ (xh- 1 , a, hOI). Hence 7:A(X, a, a) := a is
the desired projection. •
Theorem 1.4.1 leads to the following classical imprimitivity-type theorem.
Theorem 1.6.2. Let P and H be connected and simply connected. Given a com-
plete Poisson map J p : Sp ~ ~~, one obtains a strongly Hamiltonian H -action on
Sp and a complete Poisson map JP from the corresponding reduced space (T*P)P
to (T*P)/ H.
Conversely, for any complete Poisson map J : S ~ (T*P)/ H (where S is
symplectic) there exists a symplectic manifold Sp and a strongly Hamiltonian H-
action on Sp, with momentum map -Jp, such that S is symplectomorphic to the
reduced space (T* P)P.
This correspondence is bijective.

By Corollary 1.5.7, the diagram (T*P)/H ~ T*P ~ ~~satisfiesconditions

1,2,3, and 6 in 1.3.4; the maps 7: R and J are complete by Proposition 1.5.8. When
P and H are connected and simply connected, we see from 111.(2.48) and from the
330 IV. Reduction and Induction

fact that the level sets of r: R are the H -orbits (as the H -action is free) that the level
sets of J and r: R are connected and simply connected. Hence we have a classical
dual pair, and Theorem 1.4.1 applies.
This almost leads to the theorem: Rather than H -actions one obtains f)-actions.
However, the f)-action on (T*P)P is integrable, because it is derived from an inte-
grable f)-action on T*P (namely the H -action pulled back from the given H -action
on P). At the other side, the f)-action on Sp is integrable by Theorem 111.1.2.1, for
H is simply connected and J p is complete. •

We now take P = G = H in the bundle P(Q, H, r:), where G is a Lie group,

so that the base is a point. Recall the momentum maps J L and J R for the left and
the right action of G on T* G, respectively; cf. II1.I.4, and in particular III. ( 1.55}-
III.(1.58). The top line

(T*P)/ H :.::- T*P ~ f): (1.32)

of Figure 3 then specializes to

(1.33)

since the momentum map JL : T*G -+ g~ for the left action of G on T*G
coincides with the projection r: R : T*G -+ (T*G)/G, when we identify the
quotient (T*G)/ G under the right action with g*; cf. the proof of 1.5.8.
The structure of (1.33) is illuminated by an easy calculation.

Proposition 1.6.3. Pick a coadjoint orbit 0+ in g~, and specialize Figure 3 to

P = 1/ = G and Sp = ~with J p = LO the inclusion map.
The reduced space (T*G)o+ is symplectomorphic to 0+, and J O is equivalent
to -LO : 0+ -+ g~.

Using the left trivialization, we identify T*G with g~ x G, with Poisson bracket
IlI.(3.101); cf. Proposition III.3.9.9. From 111.(1.55) we then obtain

T*G *0+ 0 = {(-O, y.O)1 y E G. 0 EO}. (1.34)

The reduced space (T7(;)o is the orbit space of the G-action x : (-0, y. e) f--*
(-Co(x)O, yx-', Co(x)01..,cf. III.(1.49). One easily sees that the desired sym-
plectomorphism from (T*G)o+ to 0+ is given by [-0, y. O]G f--* yO, so that
[-y-'e, y, y-'O]o f--* O.
From (1.17) one has JO([-y-10. y. y-'O]o) = JL(-y-'O, y)L; on account of
III.(1.56), this equals -0. •
We now assume that the bundle P( Q. H. r:) is of the form G (G / H. H. r:). where
G is a Lie group with closed subgroup H acting on G from the right; see 111.2.7.
This time we use the right trivialization T*G ~ g~ x G, so that (1.32) becomes

(1.35)
 I Reduction 331

compare with Prop,?sition III.3.9.9. Here

,R(O,x) = (0, (X]H); (1.36)
Jt~(O,x) = -(Co(x-1)0) r~, (1.37)

where (X]H is the cosetxH E GJ H,and JR is the momentum map for the H-action
on T*G from the right; see III.(1.57). As always, the restriction Jt~ of J R to ~ is
just the momentum map for the right action of H on T*G. Writing ,R = (,(1), '(~»'
we see from III.(1.58) that,tt) :
T*G -+ g~ equals the momentum map of the left
action of G on T* G, that is,
R = JL .
'(I) (1.38)
Given a strongly Ham1!!0nian H-space Sp, with momentum map -Jp : Sp -+
~~, the reduced space (T*G)P (cf. (1.26) or (1.30» consists of equivalence classes
[0, x, a]H, where
Jp(a) = -(Co(x-1)0) r ~, (1.39)
and the H-action is given by (cf. III.(1.51)}
h : (0, x, a) H- (0, xh- 1, ha). (l.40)
Since G C Aut(G) C go (cf. the paragraphs following III.~} 18) and (1.30»,
one obtains a "reduced" action)...P of G on the reduced space (T*Gy. By III. ( 1.52)
the explicit fonn of this action is
)...i([O, y, a]H} = [Co(x)O, xy, a]H; (1.41)
this generalizes III.(2.121}. By III.(1.58) the momentum map for)...P is
J0)([0, y,a]H) = 0. (1.42)
Given (l.40), this is clearly well-defined, generalizing 111.(2.123). See Figure 4,
in which J; = -J0)' The minus sign in front of JL is explained by Proposition
1.6.3: It changes -to to to.
Specializing to the case where Sp is a coadjoint orbit 0+ in ~~ and J p = to is
the inclusion map, we recover the reduced space (f;:G)o+ already encountered in

J~
g~ ...- - - T*G ---.... ~~ ...- - -

-
(T*G)P

FIGURE 4. The reduced space (T7G)P

332 IV. Reduction and Induction

111.(2.118). From our present perspective one has

(T7GP+ := (Jr~rl(O)j H :::: (T*G)(h j7ro(H). (1.43)
Recall Definition III.3.9.7. A classical system of imprimitivity for a given Lie
group G is called transitive when the G-action on Q is transitive; hence Q = G j H
for some (closed) subgroup H c G. Theorem 1.6.2 now has a refinement, in which
the special assumptions on P = G and H are dropped. In particular, we do not
assume that G or H is connected. The following classical transitive imprimitivity
theorem is the classical analogue of Corollary III.3.7.6.
Theorem 1.6.4. Let Sp carry a strongly l!pmiltonian H -action. Then the
Kazhdan-Kostant-Sternberg reduced space (T*G)P carries a (transitive) classical
system of imprimitivity for G on G j H.
Conversely, each symplectic manifold carrying a transitive classical system of
imprimitivity for G on G j H is a reduced space of the type CPG)P (!!E, to a G-
equivariant symplectomorphism), and the correspondence Sp ~ (T*G)P thus
achieved is bijective.

1.7 Proof of the Classical Transitive Imprimitivity Theorem

This section contains the proof of Theorem 1.6.4.
Since the proof of 1.4.1 as it stands breaks down without the connectedness
assumptions, we seek a modification of the construction of the respective reduced
spaces. We have already replaced the reduced space (T*G)P as originally defined
by the Marsden-Weinstein quotient (T7G)P :::: (T*G)P j7ro(H).
When H is connected, Corollary 1.2.3 yields a complete Poisson map J P :
(T*G)p ~ (T*G)j H :::: g~ x Gj H. Using the explicit description of (T*G)P in
the previous section, as well as (1.17) and (1.36), this map is explicitly given by
JP([e, x, a]H) = .R(e, x, a) = (e, [X]H). (1.44)
One sees from (1.40) that this map is indeed well-defined, and remains so when
H is no longer connected. Hence we obtain (using the same notation) a complete
Poisson map J P : (T7G)P ~ g~ x G j H.
Splitting JP into 1~) : (T7G)P ~ g~ and J&) : (T7G)P ~ G j H, we see that
J~) is the momentum map for the G-action).,P on (T7G)P specified in (1.41). In
particular, it follows that the strongly Hamiltonian g-action on CFC)p defined by
J~) is integrable. Thus one obtains a classical system of imprimitivity on (T7G)P
defined by the G-action ).,P and the representation if = (1&)* of COO(G j H. R);
the covariance property is easily verified from (1.44) and (1.41).
Reducing in the opposite direction, we will exploit the following insight.
Lemma 1.7.1. Consider the situation in Figure 5, where
J = (J(1). 1(2) :S ~ g:' x GjH
is a complete Poisson map corresponding to a given classical system of
imprimitivity of G on G j H. The space g~ x G is T* G - in the right trivialization.
 1 Reduction 333

J
I)~ ...- - - g~ x G - - - " g~ x GjH ...- - - S

j reduction

J
T*G ,R

When G is connected, the reduced space T*G;R is a manifold that is

symplectomorphic to

(1.45)
where <f>G/H is the foliation of S generated by the distribution NG/H spanned by
all Hamiltonian vector fields of the form ~J*(2) f-' where j E CXJ(G j H, JR).
Moreover, minus the reduced Poisson map -J;R : T*G;R -+ I)~ is equivalent
to the well-defined quotient
J(12) := [J(I)r~ r J(2/([e]H)]4>GIH (1.46)
of the momentum map J(I) r I) of the restriction of the G-action on S to H.
The reduced space T*G;R by definition consists of equivalence classes
[e, x, a]4>,
where e is determined by J(l)(a) = e, and x is constrained by
J(2)(a) = [X]H'
Recall that <f> is the foliation defined by all vector fields ~(,R)' f - ~J* f' where
f E COO(g* x G j H, JR). Since G is connected, the distribution spanned by these
vector fields is simply the sum of NG/H and the vectors tangent to the G-orbits
of the action x: (e, y, a) t--+ (Co(x)e, xy, xa). This action is derived from (1.38)
and 111(1.52). By m.( 1.54) in the right trivialization, the flow of ~(,R)' f leaves x
in (O,x) E T*G inert.
The constraints on (e, x, a) and the G-covariance of J(2) (which holds by def-
inition of a classical system of imprimitivity) imply that J(2)(x- 1a) = [e]H. It is
then easily verified that the map [0, x, a]4> t--+ [x-1a ]4>GIH defines a symplecto-
morphism from T*G;R to J(2/([e]H )j<f>G/H (given that these spaces are manifolds,
see below). Note that J(2/ ([e]H) is indeed stable under the Hamiltonian flow of any
J(~J, j E COO(G j H, JR). Denoting this flow by a(·), we use Proposition 1.2.3.5
to compute
d d
dt J(2)a(t) = dt q(t) = 0,
where q(t) is the Hamiltonian flow of j in GjH. Since COO(GjH,JR) is
commutative as a Poisson algebra, one has q(t) = q(O) for any initial condition.
334 IV. Reduction and Induction

By (1.17), III.(1.57), and the equivariance of J(l) we have

J;R([(), x, a]CI» = JR«(), x) r I) = -J(l)(x-Ia) r I). (1.47)
Hence under the above symplectomorphism, under which J;R is transformed to,
rR , we find that JrR([a]H) = -J(l)(a) r I), where a E J(2) ([e]H).
I
say, J~ ~

It is instructive to verify that J(J) r I), restricted to J(2)1 ([e]II), is constant on the
leaves of the foliation <POI II. Picking X E I) and j E COO(G / H, JR) we compute,
using the notation in the paragraph before the last,
d - s -
dt J(l)x(a(t» = (J(~)f, J(I)x}(a(t» = -hJ(~)f(a(t»,

where we have used the antisymmetry of the Poisson bracket and III.(1.7). By
1.2.3.5 (as above) and the equivariance of J(2) under the action of G, and hence
of H, this equals _~~/H f(q(O», where, for general X E g, the vector field ~~/H
is defined by the left action of G on Gj H. This vanishes, since q(O) = [e]H and
X E I).
Finally, J(2)1([e]H)/<P oIH is a manifold. It is immediate from its equivariance
that J(2) is a surjective submersion, so that J(2)I([X]H) is a submanifold of S for
any x E G (this is, of course, consistent with the fact that T*G *(T*O)/H S is a
submanifold of T*G x S, since rR is a surjective submersion; cf. the proof of
Theorem 1.2.2).
It follows from 1.(2.9), the fact that J(2) is a Poisson map, and the commutativity
of the Poisson algebra COO(G / H, JR) that ~r(2) ,- 1--+ dj([e]H) is an isomorphism
between the abelian Lie algebra spanned by the ~J(iJ' restricted to J(2/([e]H),
and the vector space T[;lH (G / H). Under this isomorphism the foliation <POI H of
J(2)1 ([e]H ) is given by the orbits of a Lie group that acts freely and properly; hence
the quotient space is a manifold. •
In view of this lemma, we declare that even when G is not connected, the reduced
space defined by the triple

T*G- ~ (T*G)/H ~S (1.48)

is P'G:R rather than T*G: R.As in the proof of Lemma 1.7.1 ,one obtains a Poisson
map J(J) r I) : P'G:R ~ I)~, which is complete when J is.
Following the proof of Theorem 1.4.1 (with J I and h interchanged) we now
construct the reduced space depicted in Figure 6. Here g~ x G / H is (T*G)/ H in
the right trivi~,?~ whereas g~ x Gis T*G in the left trivialization. The tilde
in the name g~ x G JR of the reduced space has the same significance as in (1.30).
r~
Elements of this reduced space have the form [(), x, [a]CI>G/H]H, where () E g*,
--J
X E G, and [a]CI>G/H E T*G rR (where a E S) are constrained by

J(l)(a) r I) = () r I), (1.49)

J(2)(a) = [e]H. (1.50)
 1 Reduction 335

,R

I
g~ x G I H - g~ x G fJ~ ...- - -

reduction
_____ (12)

g+. X GJR
I~

------ (12) _
FIGURE 6. Classical imprimitivity: 9~ x G JR :::::: Sand
I~
Jjt;) : : : J
I~

The H -action defining the orbit space is given by

h : (e, x, [O']"'G/H) ~ (Co(h)e, xh- I , [hO' J"'G/H); (1.51)

see III. ( 1.49). Equation (1.49) has the implication that there exists a unique lift O'(J E
S of [O']"'G/H satisfying J(1)(O'(J) = e. This follows by regarding j E COO(GIH, R)
as a function on g~ x G I H, so that J* j = J(2/; according to Proposition I.2.3.5,
the flow 0' (t) of J* jon S projects to the flow of j on g~ x G I H under J. Using
III.(1.54) with "L" and III.(1.37), one sees that the variable in GIH as well as
e e
r fJ remain fixed, whereas any with given restriction to fJ can be reached with
such a flow. This implies the existence of O'(J; uniqueness follows, for example,
from the argument at the end of the proof of Lemma 1.7.1.
- - - - - (12)
Consider the map ({J : g+. x G JR ~ S, defined by
r~

(1.52)

This is clearly independent of the particular point in the H -orbit, and a dimension
count shows that ({J is locally a diffeomorphism. To prove that ({J is injective, assume
thatxO'(J = X'O'O,. Equation (1.50) implies thatx' = xh- I for some h E H;applying
J(l) to both sides and using the G-equivariance of this map leads to e' = he, and
applying J(2) finally shows that the primed variables are related to the unprimed
ones by the H-action (1.51). Hence ({J is injective. The G-covariance of J(2) implies
that ({J is surjective; hence ({J is a diffeomorphism. The definition of the Poisson
- - - - - (12)
bracket on g+. x G JR and the fact that G acts on S by Poisson maps easily imply
r~
that ({J is a Poisson map. In conclusion, ({J is a symplectomorphism.
By (1.17) and III.(1.48) the map j~~2) in Figure 6 is given by
r~

(1.53)

Using (1.49), (1.50), and the G-equivariance of both J(1) and J(I), one sees that the
right-hand side of (1.53) equals J(xO'(J). Hence by (1.52) the map ({J intertwines
-(i2)
JJR andJ.
r~
336 IV. Reduction and Induction

This proves Theorem 1.6.4, except for the bijectivity claim. To prove this claim
we have to show that when we start from a Poisson map Jp : Sp ~ ~~, construct
J p : (T7i:J)P ~ g~ x G / H, and subsequently put S = ('Pay and J = J p in
(1.48), the ensuing reduced space is symplectomorphic to SP' with J(I) f ~ being
equivalent to - J p'
We use the description of S given around (1.40). One has
(1.54)

from which we see that the constraint J&) = [e]H forces x E H. Hence we can
label points in (J&»-I([e]H) by (e, e, a); by (1.39) one has
Jp(a) = -e r ~. (1.55)
Once again using the argument at the end of the proof of Lemma 1.7.1, we conclude
that the map [e, e, a]C\lG/H 1-+ a from (J&)-I([e]H)/<I>G/H to Sp is a symplecto-
morphism. Using the final claim in Lemma 1.7.1 (and the comment ending the
paragraph following the lemma) and (1.42), we infer from (1.55) that this symplec-
tomorphism intertwines J(l) f ~ and -Jp , so that it intertwines the corresponding
integrated H -actions as well.
This finishes the proof of 1.6.4. •

1.8 Reduction in Stages

For the time being we return to the setting of special symplectic reduction in order
to prove a theorem on (special) symplectic reduction in stages. The statement
appears to be rather complicated, but the thrust of the result will become obvious
when the application to Marsden-Weinstein reduction is considered.
Let S ~ P ~ Sp be as specified in 1.2.1; since there will be quite a few spaces
•
and maps m what J:lollows, we reIabel these 0 b'~ects as S I III
~ PI
lp
+- Sp' Now assume
that Sp is itself a reduced space; this means that there are data S2 ~ P3 ~ S3 as
in 1.2.1, so that Sp = sij
:= s;;;.
The right-hand side is, of course, defined as
in (1.13), and in what follows one could replace equality by symplectomorphism.
In addition, we require that there be a map J21 : S2 ~ PI for which the reduced
Poisson map Jij : si3
~ PI (defined as in (1.17» coincides with Jp : Sp ~ PI.
II
The reduced space SP is consequently called Sfl'
Since in particular we are given Poisson maps J 11 : SI ~ P I- and hi : S2 ~
PI, we are in the setting of Definition 1.2.1 for the third time, now with data
S2 ~ PI- ~ S I. Theorem 1.2.2 then leads to a reduced symplectic space that we
denote by SJf .
Theorem 1.8.1. With the above notation:
1. There is a Poisson map J,j} : sJI ~ P3-. We are in the setting of 1.2.1 and
1.2.2 for thefourth time, with data Sn ~ P3 ~ S3.
 1 Reduction 337

2. The resulting reduced space S~3 is symplectomorphic to

21
Sn .
33

3. Suppose one has a Poisson manifold P2 and a Poisson map 112 : Sl ~ P2,
such that li2COO(P2, JR) ~ J*COO(P I , JR)'. By 1.2.3 this leads to a Poisson map
In :sg
33 33
~ P2.
33
There exists a Poisson map 11~3 : S~3 ~ P2 that is equivalent to lf~ (under the
21 21
symplectomorphism mentioned above).

See Figure 7.
The Poisson map lil is constructed as in Figure 1, with (clockwise) P2, 12, S,
h, PI, lp, SP' Sf, and lP replaced by P3, h3, S:;, hi, PI, 1 11 , SI, and lN, sJL
respectively.
33
The symplectomorphism S~3 ~ Sf~ is a consequence of the fact that both spaces
21
are symplectomorphic to (SI *PI S2*P3 S3)/cfJ. Here SI *PI S2*P3 S3 consists of those
triples (x, y, z) in SI x S2 X S3 for which lII(X) = hl(y) and h3(y) = h3(Z),
and cfJ is the foliation generated by the distribution spanned by all vector fields of
the form liJ - l;J + 1;3g - 133g, where f E C"'(PJ, JR) and g E C OO (P3, Ii).
This follows from the definition of the reduced space, as well as from the equations
(in self-explanatory notation) lil([y, zb) = hl(Y) and IN([x, yh) = h3(Y); cf.
(1.17).
To define 11~3 we notice that there is a Poisson map l?l : SJf ~ P2 that
21

. 0 b·
IS tamed b · . the data S2 ~
y remterpretmg ht P-I <E- J II PI <E-
JII S I as S I ~ hi S2, and
subsequently looking at SJf as Sf: . One is then back at Figure 1, with the clockwise
data listed a paragraph ago replaced by P2, 112. SI, 111, PI, hi, S2, Sf:, and l?l-
The Poisson map liP is then constructed once again as in Figure 1, this time with
21
entries P2, If}, SJ:, lN, P3, h3, S3, S~3, and liP. •
21 21

Thus the theorem is in essence a consequence of the associativity of the fiber

product, namely,

(1.56)

We apply Theorem 1.8.1 to the Marsden-Weinstein case. Given a strongly

Hamiltonian H-space SP' with equivariant momentum map -lp : Sp ~ I:J~,
the reduced space (,PG)P has been defined in 1.6; cf. Figure 4.
Given a strongly Hamiltonian G-space S, with equivariant momentum map
1: S ~ g~, we specialize (1.8) to S ~ g~ ~ (T7G)P, denoting the reduced
space defined by these data by
_P -' P
S~ := (S *g. (T*G)P)/ G ~ S~ /1I:0(G). (1.57)

Here the G-action on S x (T7G)P is the product of the given action and the action
p
')..P given in (1.41), and S~ is defined as in (1.13). where the foliation cfJ coincides
with the foliation by the orbits of the GO -action.
338 IV. Reduction and Induction

33
J23 JI~3
S23
33
11 P2 ~. ________ ~2~1_ _ _ _ _ _ _ __
SII33
11
21

33 33
FIGURE 7. Reduction in stages: s~?
21
:::: sg and ll~3 :::: ltt
21

Instead of T*G one may consider an arbitrary strongly Hamiltonian H-space

S with equivariant momentum map J H , and specialize the diagram in (1.8) to
JH Jp •
S -+ f:J~ +-- Sp. TIus leads to a reduced space

(1.58)
whose elements are equivalence classes [a, a]H, constrained by JH(a) = Jp(a),
where the equivalence relation on S x Sp is given by the orbits of the H -action
h : (a, a) H- (ha, ha). Compare with (1.13).
Theorem 1.8.2. Suppose one has a strongly Hamiltonian G-action on a sym-
plectic manifold S, with equivariant momentum map J, as well as a strongly
Hamiltonian H -space Sp, with equivariant momentum map -J p.
-P -
The reduced space S~ defined in (1.57) is symplectomorphic to the space sjrlJ
defined in (1.58), where J H = Jrf) is the momentum map for the H -action on S
obtained by restricting the given G-action.
Consider the specialization of Figure 1 shown in Figure 8.

Lemma 1.8.3. In Figure 8 one has the symplectomorphism f"";;G"'~JL ~ S that

leads to the equivalence J!JL ~ JrlJ'
 1 Reduction 339

I
~~ ..-.--- T*G ---.... g~ " - 4 - - - S

reduction

T7(;j_jL

~J

FIGURE 8. PO _JI. ::::::: S and J~JL ::::::: Jrb

Using (1.7) and III.(1.58), one sees that in the right trivialization of T*G one
has
T*G *9. S = {(-J(a), y, a)1 a E S, Y E G}.
By 1.2.2, specifically (1.9), and 111.(1.52), for connected G the null foliation 4> of
this space coincides with the foliation by the orbits of the G-action x : (e, y, a) ~
(Co(x)e, xy, xa). This also defines the equivalence classes for general G. The
equivariance of J and the fact that the G-action preserves the symplectic structure
of S then imply that the map [(-J(a), y, a)]<\> ~ y-1a is a symplectomorphism.
Choose H = G. Using (1.17), and subsequently III.( 1.57) and the equivariance
of J, one obtains
J!jL([(-J(a), y, a)]<\» = JR(-J(a), y) = Co(y-l)J(a) = J(y-1a).
For H = G the last claim of the lemma is then immediate from the previous
paragraph. For general H one simply restricts J to ~. •
Using this lemma, we fill out Figure 7 as in Figure 9; this immediately leads
to Theorem 1.8.4 for connected H and G (where Marsden-Weinstein reduction
coincides with special symplectic reduction).
To proceed for general H and G we take a closer look at the proof of Theorem
1.8.1. In the case at hand the space Sl *P, S2 *P, S3 consists of those triples
(a, w, a) E S x T*G x Sp for which J(a) = -JL(w) and Jr~(w) = Jp(a). In the
right trivialization of T*G, where w = (e, y), these conditions read J(a) = -e
and -(Co(y-')e) r ~ = Jp(a); see III.(1.57) and 111.(1.58). The foliation 4>
coincides with the foliation by the orbits of the GO x HO-action
(x, h) : (a, e, y, a) ~ (xa, Co(x)e, xyh-', ha);
p

cf. III.(1.51) and III.(1.52). The symplectomorphism S; ~ Sjr~ is then given by

1.8.1 as
[a, -J(a), y, al<\> ~ [y-'a, a]Ho.

From this we see that replacing (T*G)o+ by rFGP+ amounts to replacing HO

by H in the last-mentioned equivalence class, too. This is precisely what is needed
340 IV. Reduction and Induction

p
p JRJ JP
Jr~
SR P2 • sP
J Jr~

1,1
S
~ S

J
j
g~
/
• (T*G)p-
~ I)~
JPR

-J'1
T*G
~ J, 1
Jp
o
FIGURE 9. Marsden-Weinstein reduction in stages: s~ ~ s~~

to prove the theorem, which now follows, because Lemma 1.8.3 holds whether or
not G is connected. •
Applying Theorem 1.8.2 to the special case (1.43) we obtain
Corollary 1.8.4. Suppose one has a strongly Hamiltonian G-action on a symplec-
tic manifold S with equivariant momentum map J, leading to the reduced space
S1 defined in (1.57).
~O

This reduced space is symplectomorphic to the Marsden-Weinstein quotient

~o
SJr~ = Jr~-I (0)/ H.
A different application of Theorem 1.8.1 leads to
Proposition 1.8.5. Let G I and G 2 be Lie groups, and let S carry a strongly
Hamiltonian action ofG t x G 2, with equivariant momentum map J = J 1 EB h :
S -+ g~ EBg~. (When Gland G2 are compact this is equivalent to having commuting
actions ofGt and G2, both of which are strongly Hamiltonian.)
Then the Marsden-Weinstein quotient S?
:= J11(O)/G 1 carries a strongly
Hamiltonian G 2-action, with equivariant momentum map Jf. and similarly for
1 B- 2. One has the symplectomorphisms
(J?)-I(O)/G 1 ~ (Jf)-I(O)/G2 ~ J-1(O)/G 1 X G 2. (1.59)
 1 Reduction 341

s o
o 0
FIGURE to. Reduction in stages: S,~ ~ S,~

The G I X G2-equivariance of J implies that JI is G I-equivariant and G2-

invariant, and similarly for 1 ~ 2. (When two commuting and strongly
Hamiltonian actions of compact Lie groups G I and G2 are given, one may con-
versely achieve that J = J 1 $ his G 1 x G2-equivariant by averaging JI over
G2 and hover G 1.) With III.(1.19), 111.(1.7), and 1.(2.8), the invariance property
guarantees that JrCOO(g1, R) and J;COO(g;, R) Poisson-commute. The first sym-
plectomorphism in (1.59) now follows by specializing Figure 7 to Figure 10; the
second follows from the proof of Theorem 1.8.1. •

1.9 Coadjoint Orbits ofNilpotent Groups

We will now look at a situation where (T-::(:;)o+ (see (1.43» is (symplectomorphic
to) a coadjoint orbit of G, and where every coadjoint orbit is of this form. In
such a situation Corollary 1.8.4 is particularly useful, stating that any Marsden-
Weinstein quotient with respect to G may more simply be constructed by reducing
with respect to a suitable subgroup H C G.
As we shall see in this section, such a scenario applies when G is nilpotent;
recall that this means that [X J, [X2 • ... [Xk-I. X k] ... J] vanishes for all Xi E g
342 IV. Reduction and Induction

and all integers k :::: ko, for some ko :::: 1. For example, the Heisenberg group fIn
is nilpotent.

Theorem 1.9.1. Let a Lie group G be nilpotent, connected, and simply connected,
and pick a coadjoint orbit O~ through 0 E g*.
There exists a subgroup Po ~ G of dimension dim(G) - ~ dim(O~), and a
pointa E P~ (which is stable under Co{Po)' and therefore is a coadjoint orbit),for
which the reduced space (T7G)o+, defined as in (1.43) with H = Pol!!!9 0 = a,
is symplectomorphic to (O~)+. Moreover, the reduced G-action on {T*G)a (with
associated momentum map -J~, where J~ := J;R ) is equivalent to the coadjoint
IPo
action on O~ (with momentum map -tof)'

Note that Po is a proper subgroup unless O~ is zero-dimensional.

The idea of the proof is that the map J~ in Figure 4 (with Sp ---+ a, ~ ---+ Po' and
J; ---+ J~) is a symplectomorphism. To show this, one relies on a fundamental
fact in the theory of nilpotent Lie groups, which we state without proof.

Lemma 1.9.2. For every 0 E g* there exists a subalgebra Po of g of dimension

Hdim(g) + dim(go» (where go is the Lie algebra of the stabilizer Go ofe under
the coadjointaction) that contains go' and has the property that e([X, YD = Ofor
all X, Y E Po.

The Lie algebra Po is called a polarizing subalgebra of g. Its defining property

is equivalent to the stabilityofa := 0 r Po underCo(Po), where Po is the connected
and simply connected Lie group with Lie algebra Po' Indeed, in the theorem we
take Po and a as indicated above.
According to III.(2.1l9), in the left trivialization of T*G the reduced space
(T7G)a consists of H -equivalence classes [-0, X )Pti' with 0 r Po = a and x E G.
The Po-action defining the equivalence classes is given by III.{2.120).
The crucial technical point is the equality

{O E g*1 (0 r Po) = 0 r Po} = Co(Po)O. (1.60)

It is obvious that the right-hand side is contained in the left-hand side, for

(CoG(h)O) r Po = CoPO(h)(e) r Po) (1.61)

for all h E Po; for clarity we have denoted the coadjoint action on g* and on p&
by CoG and Co Po , respectively. The opposite inclusion is proved by observing
that the left-hand side of (1.60) is a copy of jRdim(9)-dim(po). It is a nontrivial fact
about connected, simply connected, and nilpotent groups that the orbit Co(Po)O is
homeomorphic to jRdim(Pii)-dim(9ii). By Lemma 1.9.2 the dimensions match, so that
(1.61) has been proved.
By (1.17) and I1I.(2.124) one has

J~([-O, x]po) = Co(x}8. (1.62)

 1 Reduction 343

Combining the information collected so far, one infers from Corollary 1.2.3 that
J; is a symplectomorphism between rFiJ)a and O~ .
The reduced G-action on (T-;;;:;)a is given by 111.(2.122); the final claim in
Theorem 1.9.1 is then immediate from (1.62). D

Consider, for example, the Heisenberg group fIn introduced in 11.2.1; its coad-
e
joint orbits are listed after 11.(2.11). For the zero-dimensional orbit = (p, q, 0)
one has go = Po = ~n, so that the claims of the theorem are self-evident. Accord-
e
ing to 11.(2.13), a 2n-dimensional orbit through = (0,0, c #- 0) has stabilizer
go = RZ, and one may choose Po = RQ E9 RZ, which is abelian. (Alternatively,
one could pick, e.g., RP E9 RZ, which illustrates the fact that polarizing subalge-
bras are not necessarily unique.) The left-hand side of (1.60) is {(O, R, c)}, which
by 11.(2.13) indeed coincides with the right-hand side. Thus one verifies without
any difficulty that the coadjoint orbits of fIn are indeed as described by Theorem
1.9.1.
We return to the general case. The quantum counterpart of Theorem 1.9.1 is the
following. Recall Mackey induction from III.2.9.

Theorem 1.9.3. Let U be an irreducible representation ofa connected and simply

e
connected nilpotent Lie group G. There exists a point E g* and a polarizing
subalgebra Po such that U is equivalent to the representation U Oinducedfrom the
one-dimensional representation

Uo(Exp(X» := e-iO(X) (1.63)

of the connected and simply connected group Po with Lie algebra Po'
Two induced representations U o; (i = 1,2) of this type are equivalent iffe
l and
e2 lie in the same coadjoint orbit. In particular, different choices of the polarizing
subalgebra Po lead to equivalent representations.

Note that Uo is indeed a representation of Po as a consequence of the property

mentioned at the end of Lemma 1.9.2.
We will not prove this theorem, merely illustrating it for the Heisenberg group.
e
Choosing = (0,0, c i= 0) and Po = RQ E9 RZ as above, one computes from
11.(2.6) and III.(2.176), in which one takes the section s(q) = (0, q, 0), that U(o,o·c)
coincides with Uc as defined in 11.(2.19), with A = c. The one-dimensional rep-
e
resentation 11.(2.27) corresponds, of course, to = (p, q, 0). Theorem 11.2.1.4 is
now seen to be a corollary of Theorem 1.9.3.

1.10 Coadjoint Orbits of Semidirect Products

Another illustrative and physically relevant situation where Corollary 1.8.4 applies
with force is given by semidirect product Lie groups G = L ~ p V, where V is
a vector space carrying a linear L-action p. Following the habit of physicists, we
denote elements x of G by pairs x = (A, v) E L x V; elements of the dual V* are
344 IV. Reduction and Induction

generically called p. The multiplication rule in G is

(AI, V)(A2, V2) := (A)A 2 , v) + p(A I )V2). (1.64)
The Lie algebra g = [EEl V then has the bracket
[(X, v), (Y, w)] = ([X, y], dp(X)w - dp(Y)v). (1.65)
The coadjoint action of G on g* = [* EEl V* is
CoCA), v)(e, p) = (Co(A)e + (p*(A)p) 1\ v, p*(A)p), (1.66)
where p' is the dual action of G on V*; that is, (p*(A)p)(v) := p(p(A -)v). For
p E V* and v E V, the element p 1\ V E [* is defined by

P 1\ veX) := p(dp(X)v). (1.67)

In terms of the dual p; of the map Pp : g --+ V*, defined for fixed p E V* by
pp(X) := -dp*(X)p, one has the equivalent definition p 1\ V = p;v.
The stabilizer of p E V* under p* is denoted by Lp (as usual), with Lie alge-
bra [po Now recall Definition 111.2.1.4 and the first paragraph of m.2.7. Given a
coadjoint orbit OLp in [;, we can form the bundle L XLp OLp associated to the
principal bundle L(L/L p, L p , r) by the coadjoint action of Lp on OLp. This as-
sociated bundle has base L/ Lp, which we identify with the L-orbit O~ = p*(L)p
in V*, and typical fiber OLp.
We now fix a pair (8, p) E g*, for simplicity denoting the coadjoint orbit O~(rp
in [~ by O~p. The first characterization of the coadjoint orbits of L ~ p V is as
follows. Note the similarity with Proposition III.3.9.4.
Proposition 1.10.1. The coadjoint orbit O~.p) ofG = L ~ p V through (8, p) E g*
is a fiber bundle over L XLp O~p with typical fiber [~(the annihilator of[p in [*).
This provides a bijection between the set of coadjoint orbits in g* and the pairs
(OL, O~p) consisting of an L-orbit OL in V* and a coadjoint orbit O~p in [~
(where L p is the stabilizer of an arbitrary point p in OL ).
The bundle projection roc : O~.p) --+ L XLp O~p is given by

roc (Co(A, v)(8, p»:= [A, 8 r [p]Lp- (1.68)

This projection is well-defined because of a property analogous to (1.61). The coad-
joint orbit O~.p) in g* determines a pair (Ot, O~p); conversely, a pair ( OL, O~p)
as stated corresponds to an orbit CO~.p) with the property that 8 r [p lies in O~p,
and p has L p as its stabilizer. The fact that this correspondence is bijective and
independent of all choices follows from elementary verifications, using the equality
{p 1\ vi v E V} = ~, (1.69)
which is easily checked.
•
 I Reduction 345

We now tum to an alternative description of the coadjoint orbits of L ~ p V that

places the bijective correspondence stated in Proposition 1.10.1 in a new light.
To formulate this result, we first infer from (1.66) that for each p E V* and each
+
coadjoint orbit CJL p in I; the set CJL p pel; EI1 V* is a coadjoint orbit of L p ~ p V*.

Theorem 1.10.2. The coadjoint orbit CJGe- _ (with the H+" Lie symplectic struc-
(.p) ~
ture) is symplectomorphic to the reduced space (T*G)o+, defined as in (1.43) with
G = L ~p V,
H:= Lp ~p V, (1.70)

and CJ = CJ~P +p. Moreover, the reduced G-action on d-;(])o+ (with associated
momentum map -Jr; := -Jr:;) is equivalent to the coadjoint action on CJ(~ _)
~ ,p
(with momentum map -Loc; ).
9

What follows is a more complicated version of the proof of Theorem 1.9.1.

We see from III. (2. I 19) that in the left trivialization of T*G the reduced space
(T7G)o+ consists of H -equivalence classes [-e, - p, A, V]H, with e I Ip E CJ~P,
and according to III.(2.120), (1.64), and (1.66) the H -action is given by
(A[, w): (-e, -p, A 2 , v) t-+ (-Co(A,)I1-pAw, -p, A2AI', V-p(A2AI')W).
(1.71)
According to (1.17) and III. (2. 124) one has
Jr;([-e, -p, A, V]H) = CoCA, v)(e, p), (1.72)
which is given by (1.66). Using (1.69), it is now easily verified th~r; is injective.
Since (A, v) varies freely in G, it is obvious from (1.72) that Jr;«T*G)o+) contains
CJGe- _ • Combined with Proposition 1.10.1, the restriction 8 I Ip E CJe~P implies
( ,p)
that the image of Jr; is precisely CJ~.p)'
Hence Jr; : (T--;:-'G)o+ --+ CJGe- _) is a diffeomorphism that is even a
( ,p
symplectomorphism by Corol~ 1.2.3.
The reduced G-action on (T*GP+ being given by III.(2.122), the final claim is
immediate from (1.66) and (1.72). •
To state the quantum versions of Proposition 1.10.1 and Theorem 1.10.2, we
take p E V* and a E Lp, so that Uu(Lp) is an irreducible representation of Lp.
Define H as in (1.70), and note that U u (L p) extends to an irreducible representation
Up,u(H) by

(1.73)
Theorem 1.10.3. Suppose that the semidirect product G is regular in that each
L-orbit in V* is (relatively) open in its closure. Then the representation UP,U(G)
induced by an irreducible representation U p,u (H) of the above type is irreducible
for any choice of p and a, and for every irreducible representation U (G) there
exists a pair (p, a) such that U is equivalent to U p,u .
346 IV. Reduction and Induction

Two representations uPI.al(G) and u h a2(G) are equivalent iff the Pi lie in
the same orbit OL (so that P2 = p*(A)PI for some A E L), and Ua2 0 AdA is
equivalent to Ual' In other words, the unitary dual G is parametrized by pairs
(OL, a), where OL is an L-orbit in V* and a is a member of the unitary dual of
the stabilizer of an arbitrary point in OL.
Let U be a representation of G on a Hilbert space 1f. One easily sees from (1.64)
that
U(A)U(v)U(A)* = U(p(A)v), (1.74)
where A := (A,O) and v := (e, v) etc. By III.(1.89) the restriction of U to V
defines a representation 1r of the group C*-algebra C*(V). Using III.(1.88) with G
replaced by V, this yields a representation if(Co(V*». We see from (1.74) that the
pair (U(L), if (Co(V*») is a system ofimprimitivity of Lon V* in 1f; cf. Definition
III.3.7.3. Conversely, such a system determines a representation U(G) on 1f, and
the correspondence thus obtained is bijective. Theorem 1.10.3 then follows from
Theorem 2.7.3 below and Corollary III.3.7.4. •
The general description of the explicit form of induced representations in I1I.2.9
simplifies somewhat. because G / H is n~w equal to L / L P ~ O~. To obtain a
(measurable global, or smooth local) sectIOn s : G / H ~ G we merely need to
choose a ( ... ) section b : L/ Lp ~ L (where the name "b", standing for "boost".
comes from physics), in terms of which s is given by s(p) := (b(p), 0). The carrier
space of the realization uta (cf. the text below III.(2.176)) is then

(1.75)
where 1fa is the carrier space of Ua(Lp), and we have replaced the suffix s by b.
Assuming that O~ possesses an L-invariant measure (which will be the case in
our applications), one is able to simplify m.(2.176) to
uta (A, v)wt· a (p) = eiPVUa (b(p)-I Ab(p*(A -I)p») w{a (p*(A -I)p). (1.76)
Note that the argument of Ua indeed lies in L p.
We return to the classical setting. From 1.10.2 we are led to a third description
of the coadjoint orbits in question. Recall that we identify the L-orbit O~ in
V* with L/ Lp; this leads to an embedding (i.e., an injective homomorphism)
X: V ~ COO(L/Lp,R) (as additive groups),givenby X(v): p H> -p(v),
where p E Oft. Using the natural embedding L C Aut(L), where L is seen as the
total space of the principal bundle L(L/ L p, L p, -r) (cf. 111.2.7), we observe that X
extends to an embedding X : L ~p V ~ Aut(L) ~ COO(L/ L p, R). This enables us
to regard G as a subgroup of Aut(L) ~ COO(L/ L P' R). In particular, G acts on the
reduced space (PL)o;P by restriction of the action Po of Aut(L) ~ COO(L/ L p, R)
defined in Theorem m.2.5.2.
Theorem 1.10.4. The coadjoint orbit ( O~'P») + is symplectomorphic to the re-
- OLp L-
duced space (T*L) 9 ,defined as in (1.43) with G = L, H = L p, and 0 = 0/.
 1 Reduction 347
_ Lp
The G-action on (T* L)o& explained above is equivalent to the coadjoint action
on O~.ji)"
Given pE V*, we regard T* L as a submanifold of T*G by the embedding
(e, I) ~ (e, -p, I, 0); this is a Poisson map. The momentum map J r1, is simply
the restriction of J r~ (where ~ is the Lie algebra of the group H specified in 1.10.2)
to [ji, regarded as a subalgebra of ~ by the embedding e ~ (e, 0). One then easily
infers from (1.71) that (Jr1.)-I(O~')/ Lji is diffeomorphic to (Jr~)-I(O~' +p)/ H
under the bijection P

(1.77)

where e r [ji E O~p. Since the embedding T* L "--+ T*G above is a Poisson
map, this diffeomorphism is a Poisson map, and therefore a symplectomorphism,
by definition of the quotient Poisson structure. The first claim then follows from
Theorem 1.1 0.2.
Comparing (1.72) with (1.77), and using (1.66) with v = 0, one sees that the
_ l.p
pertinent symplectomorphism ip : O~.ji) --+ (T* L)o& is given by

(1.78)
We now prove that ({J intertwines the G-actions in question. Firstly, it is obvious
from (1.77), (1.64) and III.(2.122) that ({J intertwines the L-actions. Secondly,
we note that in the left trivialization the one-form df(x) E Tx*G (where f E
COO(G, R» is represented by (ef' X)L, where ef(X) := ~f f. Regarding xCv) as a
right- L p-invariant function on L, it then follows from III.(l.37) and (1.69) that in
the left trivialization, dX (v) at A is represented by (p /\ peA -I )v, A)L. Hence the
action III.(2.77) reads v: (-e, A)L ~ (-e - p /\ peA -I)V, Ah., This quotients
to the action
v: [-e, AlL p ~ [-e - p /\ peA -I)V, AlL p
_ Lp
on (T* L)oe . The second claim in 1.10.4 then follows from (1.77), (1.71),
111.(2.122), (1.66), and (1.72); cf. the end of the proof of 1.10.2.
As a check on this computation we note that by definition of X , the momentum
map 111.(2.80) reads Jx(a)([-e, A]Lp) = -(p*(A)p)(a). By (1.66) and Corollary
III. 1.4.6 this coincides with J(O.a)(Co(A, O)(e, p», where J is the momentum map
for the coadjoint action of G on its coadjoint orbit ( O~.ji») +' We conclude from
(1.72) and (1.77) that JCj 0 ({J intertwines Jx(a) and J(O.a)' •
_ Lfi

We infer from this proof that the G-action on (T* L)o& is given by
(AI, v): [-e, AZ]L p ~ [-e - p /\ p(AIAz)-lv, A I AzlL;;- (1.79)
This formula may alternatively be derived from (1.66) and (1.78).
Applying Theorem 111.2.3.7 (or its generalization 1.6.1), we conclude that O~. ji)
is a bundle over T*O~ with typical fiber O~p; the bundle projection depends on
348 IV. Reduction and Induction

the choice of a connection on the principal bundle L(L/L p , L p, r). When Lp is

reductive one may choose the H -connection; see 111.2.7.
Alternatively, since (PL)o;p is by definition equal to (irfp)-I(Ot p)/ L p' it may
be regarded as a bundle L XLp rr- 1( -ot p), where rr : [* ~ [ft is the restriction
map rr(t9) := t9 r [po As always, we identify L/ Lp with the L-orbit O~ through
p in V*. This is the bundle over L / L p that is associated to the principal bundle
L(L/ L p, L p, r) by the coadjoint action of Lp on rr- 1( -otp) c [*; note that this
subset is indeed stable under the restriction of Co(L) to L p. This bundle structure
is evidently independent of the choice of a connection, and may also be derived
from Proposition 1.10.1. For example, when E [~,eo p
so that O~p
110
= {O}, one infers
from 111.(2.116) that

(1.80)

By 111.(2.56) this is even a symplectomorphism.

As a simple example, consider L = SO(3) and V = JR.3, with p the defining
representation of SO(3). The latter coincides with the coadjoint action of SO(3)
on [* = JR.3, as well as with the action p* on V* :::' JR.3 (where JR.3 and its dual have
been identified through the Euclidean inner product). Hence both the coadjoint
orbits in [* and the p*(SO(3»-orbits in V* are either spheres S; with radius r > 0
or the origin (r = 0). The expression p 1\ v in (1.67) is easily seen to coincide with
the usual exterior product of p and v.
We take G = £(3) := SO(3) ~p JR.3, which is the Euclidean group. When
p = 0 one has Lo = SO(3), so that the G-coadjoint orbit O~( , 0) is simply the
S o (3)-orbit through iJ, whose Lie symplectic structure is r times the volume form
on S2. For p =1= 0 we may choose p = r3 := (0,0, r), so that Lp = SO(2). By
Theorem 1.10.4 the orbit O~,r3) is a bundle over T* S2 with typical fiberiJ3 =: e E
JR., which is just a point. Thus oC!.
(lI,r3)
is T* S2 as a manifold, which as a symplectic
space is the space T* G e discussed in III.2.12. In particular, the Poisson bracket on
O~ is III.(2.227).
(0,r3)
For a more complicated illustration of the formalism see 3.1.
We close this section with a result on reduction in stages for semidirect products.
Theorem 1.10.2 and Corollary 1.8.4 have the

Corollary 1.10.5. Let G = L ~ p V act on a symplectic manifold S in strongly

Hamiltonianfashion, with equivariant momentum map J : S ~ g~, and choose
·· orb'It v//"1 := v(8,p)
a coad'JOint ,"'IG . *
In g+.

The Marsden-Weinstein quotient S?+ = J-1(O)/G is symplectomorphic to

---- OLp
(irl} (p)/V) JrLp' that is, to the symplectic space obtained by first reducing S by V
with respect to the coadjoint orbit p E V*, and then reducing by Lp with respect
to the coadjoint orbit OLp C [ft.
 1 Reduction 349

Applying 1.10.2 and 1.8.1 in succession shows that we need to prove only that
--- c/ p
(Jr"\/(p)1 V)JrL p coincides with the reduction of S by H = Lp ~p V with respect
to the orbit 0 = O~P +p in I)~.
The G-equivariance of J implies that Jry(la) = p*(I)Jrv(a). It follows that Lp
maps Jr"\}(p) into itself. Moreover, the rule (1.64) easily implies that the action of
L p quotients to an action on the reduced space J -I (p) 1V. It is quite straightforward
to verify that the reduced L p-action is strongly Hamiltonian, with equivariant
momentum map given by the quotient of Jrlp-
Hence the second reduction is well-defined. By the same argument, the map
[[a]V]L p 1--+ [alLp><pv,wherea E Jrvl(p),iswell-defined,andisalmosttrivially
seen to be a diffeomorphism. Using Theorem 1.5.4 one infers that it is even a
symplectomorphism. •
Note that Theorem 1.1 0.4 follows from 1.1 0.5 by taking S = T* G and observing
rv
that J 1(p)1 V is symplectomorphic to T* L; in our derivation of 1.1O.4from 1.10.2
this was used in the opposite direction.

1.11 Singular Marsden-Weinstein Reduction

In this section we look at what happens to Theorem 1.5.4 when H does not act
freely. We do assume that the strongly Hamiltonian H -action on S is proper,
as reasonable results are available only in that case. We look only at reduction
from 0 E 1)*; by the shifting trick in the proof of 1.5.4 this entails no loss of
generality. The reduced space is therefore SO = J-I(O)I H; for simplicity we omit
the subscript J in the notation S~ (cf. (1.26».
When the H -action fails to be free, but each stabilizer Hu is discrete for a E
J- 1(0), the space J-I(O) is still a submanifold of S by Lemma 1.5.1, and the
possible singularities in the reduced space J- I(0)1 H come from taking the quotient
by H. This case is, of course, included in what follows.
When the dimension of some stabilizer is greater than zero, there is no guarantee
that J-I(O) is a submanifold of S. For a given subgroup K ~ H, define
SK := {a E SI Hu = K}; ( 1.81)
S[KJ := HSK = {a E SI Hu is conjugate to K}. (1.82)
These spaces are empty when K is not compact; cf. Proposition 1.5.3. Using
the compactness of each stabilizer H u , it can be shown that SK and S[KJ are
submanifolds of S. We can say more:
Lemma 1.11.1. The space SK is a symplectic manifold, with symplectic form
inherited from S.
The subspace Tu SK of Tu S, where a E S K , consists of the K -invariant vectors
(since a is K -invariant, the pushforward of the K -action maps Tu S into itself).
Since K is compact, we can choose a K -invariant inner product (, ) on Tu S, and
by linear symplectic geometry there exists an invertible linear map J on Tu S such
350 IV. Reduction and Induction

that (X, JY) = wu(X, Y) for all X, Y. By the K-invariance of the symplectic
form w and of the inner product, the map J commutes with the K -action, so
that the restriction of J to Tu SKis well-defined. Then the assumption that for
some X E TuSK the number wu(X, Y) vanishes for all Y E TuSK implies that
(X, JY) = 0 for all Y E TuSK; hence (X, X) = 0, so that X = O. The claim
follows. •
This lemma will be used in the proof of Proposition 1.11.3.
Proposition 1.11.2. Let a(·) be the Hamiltonian flow of f E COO(S, R)H. For
any t for which the flow exists, the point a(t) lies in J-1(0) n S[I/a]' where a =
a(O). Moreover, any two points in a connected component of J-1(0) n SK may
be connected by a piecewise smooth Hamiltonian curve, where the pieces are
generated by H-invariant Hamiltonians.
By Proposition III.1.2.2 the set J-1(0) is invariant under the flow. The H-
invariance of f implies that h(a(t» = (ha)(t) for all h E H, so that Hu(t) ~ Hu.
Inverting the flow leads to the opposite inclusion, so that Hu(t) = Hu. It follows that
the flow preserves J-1(0) n SI/a. Using the stability of J-1(0) under H resulting
from the equivariance of the momentum map and the second equality in (1.82),
one obtains
(1.83)
Since we have just seen that the right-hand side is preserved by Hamiltonian flows,
the first claim follows. The second claim results from the equality
(1.84)
Here the inclusion of the right-hand side in TuJ-1(0) follows from Noether's
Theorem (Proposition 111.1.2.2). Its inclusion in TuSHa follows from the first line
in the proof of 1.11.1 combined with 1.(2.8) and the H -invariance of the Poisson
bracket. The proof of Corollary 1.5.7 then yields the equality in (1.84). •
The reduced space SO is trivially a disjoint union
-0 -0
S = U[K]S[K]' (1.85)
where [K] varies over all conjugacy classes in H, and

(0) n S[K])/ H.
-0 -I
S[K] := (J (1.86)
Proposition 1.11.2 suggests that from a Hamiltonian point of view this is an inter-
esting decomposition, since the flow of an H -invariant Hamiltonian on S projects
to a flow that stays inside a given subspace SPK]"
Proposition 1.11.3. Let S~ be the union of those components of S K whose
intersection with J-1(0) is not empty.
The natural action of the group NH(K)/ K on S~ is free and strongly
Hamiltonian; denote its equivariant momentum map by h. The (regular)
Marsden-Weinstein quotient Jil(O)/(NI/(K)/ K) is homeomorphic to SPK] (with
 1 Reduction 351

the quotient topology). Since the former is a symplectic manifold, the space SfKl
thereby becomes a symplectic manifold as well.
Here N H(K) is the nonnalizer of Kin H (that is, the collection of elements of H
that commu te with all members of K). It is clear from the definitions that N H (K) / K
acts on SK by restricting and quotienting the H -action, and that this action is free.
From the second equality in (1.82) we then infer that S[KI/ H ~ SK/(NH(K)/ K).
Taking intersections with J-1(0) leads to the desired homeomorphism.
To interpret (J-I(O) n SK )/(NH(K)/ K) as a Marsden-Weinstein quotient, we
first note that Lemma 1.11.1 implies that S~ is symplectic. Secondly, since J is
equivariant and S~ consists of K -stable points, the momentum map for the H-
action restricted to S~ takes values in the space (I)*)K of Co(K)-invariant points
in 1)*. Since Jx for X E t generates the K -action on S~, which is trivial, each Jx
must be constant on each component. The constants are all zero, as S~ intersects
J-1(0). Hence J r S~ takes values in (I)*)K n to, where to is the annihilator of t
in I).
Now observe that (I)*)K n to is naturally isomorphic to the dual of the Lie algebra
of N H (K) / K: This is immediate from the definition of the nonnalizer and of the
(co) adjoint action, combined with the isomorphism T[;lx (H / K) ~ to (cf. the
proof of Lemma 111.2.7.1).
We conclude that J r S~ may be interpreted as the momentum map h for the
NH(K)/ K -action on S~. Since this action is free (and evidently proper), Theorem
1.5.4 applies. •
Since the reduced space SO is not (necessarily) a manifold, there is no self-
evident definition of the space of "smooth" functions COO (So , JR). In the present
context the following approach is appropriate.
Definition 1.11.4. A continuous function f on SO is said to be smooth when there
exists an H -invariant smooth function on S whose restriction to J -I (0) quotients
to f.In other words,
(1.87)
where JoH := .10 n cOO(S, JR)H, and .10 is the ideal of smooth functions on S that
vanish on J -I (0).
When SO is a manifold, one recovers the usual definition of cOO(So, JR). Oth-
erwise, the main advantage of Definition 1.11.4 is that one obtains a Poisson
algebra.
Proposition 1.11.5. The space COO (So , JR) is a Poisson algebra under the Poisson
bracket inheritedfrom cOO(S, JR). This bracket coincides with the one correspond-
ing to the symplectic structure on each subspace SfKl.In other words, the inclusion
of each symplectic manifold SfKl in SO is a Poisson map.
We already know that cOO(S, JR)H is a Poisson algebra under the bracket inher-
ited from cOO(S, JR); cf. the proof of Theorem 1.5.5. We need to prove that JoH
is a Poisson ideal in cOO(S, JR)H. It is trivially an ideal with respect to pointwise
352 IV. Reduction and Induction

multiplication. To show that in addition it is a Lie algebra ideal under the Poisson
bracket, we pick f E coo(S, ~)H and g E J OH , and study {j, g}. This function is
in coo(S, ~)H because of the H -invariance of the Poisson bracket, and vanishes
on J-I(O) by 1.(2.8), 1.5.7, and 1.3.5.
This proves the first claim. The remainder is obvious from Proposition
1.11.3. •
In the case that SO is not a manifold, one cannot define the Hamiltonian flow
of h E coo(So,~) as the solution of 1.(2.11), since the notion of a tangent vector
is problematic at singular points. However, one can simply say that a(·) is the
Hamiltonian flow of h iff 1.(2.14) is satisfied for all f E coo(So, ~). Existence
of the flow may be proved by lifting the flow to S, and uniqueness is eventually
a consequence of the fact that due to the properness of the H -action on 5, the
function space coo(So,~) separates points in So.
We sum up.
Theorem 1.11.6. Assume that one has a proper and strongly Hamiltonian H-
action on a symplectic manifold 5, with equivariant momentum map J.
• The Marsden-Weinstein quotient SO = J-I(O)/ H can be decomposed as a
(disjoint) union of symplectic manifolds SrK]' defined by (1.86) with (1.82),
referred to as the symplectic pieces of SO.
• The function space coo(So, ~), defined by (1.87), is a Poisson algebra.
• The inclusion of each SrK] in SO is a Poisson map.
• Any Hamiltonian flow on Sa preserves the decomposition in question; in fact,
any two points in a connected component of a given subspace SrK] can be
connected by a piecewise smooth Hamiltonian curve.
Combine (1.85) with Propositions 1.11.3 and 1.11.5. The second claim in the
final item follows from Proposition 1.11.2. •
It follows that the pair (So, '2tIR = COO (So , ~» is a Poisson space in the sense of
Definition 1.2.6.2.
The abstract theory may be illustrated by what is probably the simplest nontrivial
example. Consider the standard action of H = 50(2) on Q = ~2, given by
e : (ql, q2) f-+ (ql cose - q2 sine, ql sine + q2 cose).
According to Lemma III.2.3.1 this lifts to a strongly Hamiltonian action on 5 =
T*~2, in which (PI, P2) transforms in exactly the same way as (q I, q2). This lifted
action is not free at the point (0, 0, 0, 0), whose stabilizer is 50(2); the stabilizer
of all other points is trivial. The momentum map is
(1.88)
Hence the level set J -I (0) consists of those (p, q) for which p = t q or q = 0 for
some t E ~; this set is {(lR.2\(O, 0» x ~}U~2, where (ql, q2, t) in (~2\(0, 0» x ~
stands for (ql, q2, tPI, tP2), and the second ~2 represents the points (PI, P2, 0, 0).
Hence the quotient J- I (0)/50(2) may be identified with ~+ x~, which may be
 I Reduction 353

thought of as the cotangent bundle T*JR.+. The singularity in the reduced space
takes the form of a boundary.
The topology on the reduced space may be computed by noting that the copy
of JR.2 in T*JR.2 defined by the equations P2 = q2 = 0 is contained in J-I(O),
and has the property that every SO(2)-orbit in J-I(O) cuts it in two points. These
are related by the action of 'Z} that maps (PI, ql) to (±pI, ±ql). Since the JR.2
in question evidently has the canonical symplectic structure, and Z2 acts on it by
Poisson maps, one infers that
(1.89)
as symplectic spaces. Here 1R2 /Z2 is seen as a topological space with the quotient
topology, at the same time being the union of the symplectic manifolds (0, 0) and
(1R2\(0, 0»/Z2. Thus the isomorphism (1.89) means that one has a homeomor-
phism in the usual sense, under which the appropriate symplectic subspaces are
mapped into each other symplectomorphically.
In summary, the decomposition (1.85) consists of (0,0) and (1R2\(0,0»/Z2,
each of which is a symplectic manifold in its own right. Using invariant theory, it
may be shown that smooth functions on SO (in the sense of 1.11.4) must correspond
to smooth functions h on S that depend only on the SO(2)-invariants (p, p) :=
pi + p~, (q, q), and (q, p). Since dh = 0 at (0, 0, 0, 0), one verifies that the two
symplectic pieces of SO are indeed stable under Hamiltonian flows; cf. Proposition
1.11.2.
This example illustrates a deeper property of singular Marsden-Weinstein
reduction, which we will not prove.
Proposition 1.11.7. Under the assumptions of 1.11.6, in each connected
component of 05° one of the symplectic pieces SPKO] is open and dense.

Since the Poisson bracket on COO(So, 1R) is evidently determined by the sym-
plectic form on S~KO]' Proposition 1.11.5 implies that the symplectic structure on
all other symplectic subspaces is determined by S~KO].
Finally, we give an example in which the group action is not proper, but the claims
of Theorem 1.11.6 nonetheless hold. We continue with the symplectic manifold
S = T*JR.2, but now consider an action of H = JR., namely
t : (PI, P2, ql, q2) 1-+ (PI, P2, ql + Pit, q2 - P2t). (1.90)
An equivariant momentum map for this action is
J(pJ. P2, ql, q2) = t(pi - p~). (1.91)
It follows that J* fails to be surjective at all "singular" points of the form
(0,0, q I , q2), at which it is identically zero; the lR-action is not proper precisely at
these singular points. The singular points have stabilizer JR., whereas the stability
group of all other points is trivial. This opens the possibility that J -I (0) might not
be a submanifold of S, and this is indeed the case.
The Marsden-Weinstein quotient SO = J-I(O)/IR is connected as a topological
space, but it does not have a constant dimension as a "manifold": if we look at SO
354 IV. Reduction and Induction

as fibered over the subspace PI = ±P2 of 1R2 , then the fiber above (0,0) is two-
dimensional, whereas at all other points it is one-dimensional. By Definition 1.11.4,
the space COO(So, 1R) consists of smooth functions f E COO(S, 1R), restricted to
J-I(O), that satisfy {J, f} = 0 on J- 1(0). It follows that such an f arbitrarily
depends on the Pi, but depends on the qi through the combination q I P2 + q2 PI.
A study of the Hamiltonian flow on S defined by such functions, and therefore
of the corresponding flow on SO obtained by projection, shows that SO may be
decomposed into five symplectic pieces. These are given by the equations PI =
P2 > 0, PI = P2 < 0, PI = - P2 > 0, PI = - P2 < 0, and PI = P2 = O. Any
point in a given piece cannot leave the piece under a Hamiltonian flow. Hence we
have the same situation as for proper group actions.

2 Induction
2.1 Hilbert C* -Modules
What follows is the most important mathematical concept in the quantization
theory of classical systems obtained by special symplectic reduction (see 1.2).

Definition 2.1.1. A Hilbert C* -module over a C* -algebra 23 consists of

• A complex linear space E.

• A right action Jr R of23 on E (i.e., JrR maps 23 linearly into the space of all linear
operators on E, and satisfies JrR(AB) = JrR(B)JrR(A»,for which we shall write
q,B:= JrR(B)q" where q, E E and BE 23.
• A sesquilinear map (, )'13 : E x E -+ ~, linear in the second and antilinear in
the first entry, satisfying

(q" <1»; = (<I>, q,)'13; (2.1)

(q" <l>B)'13 = (q" <I>)S]3B; (2.2)
(q" q,)'13 2: 0; (2.3)
(q" q,)S]3 =0 ¢> q, = 0, (2.4)

for all q" <I> E E and B E 23.

The space E is complete in the norm

(2.5)

We say that E is a Hilbert 23-module, and write E r= 23.

One checks that (2.5) is indeed a norm: 1Iq, 112 equals sup{w( (q" q,)'13)}. where the
supremum is taken over all states w on 23. Since each map q, ~ ..jw«(q" q,)'13) is
a seminorm (i.e., a norm except for positive definiteness) by (2.3), the supremum is
a seminorm, which is actually positive definite because of Corollary 1.1.4.4 (which
applies because (q" q,)'13 is self-adjoint by (2.1) or (2.3» and (2.4).
 2 Induction 355

The ~-action on £ is automatically nondegenerate: The property \II B = 0 for

all B E ~ implies that (\II, \II).BB = 0 for all B, hence (\II, \II)!B = 0 (when ~
is unital this follows by taking B = ][; otherwise, one uses an approximate unit in
~), so that \II = 0 by (2.4).
When all conditions in 2.1.1 are met except (2.4), so that II • II defined by (2.5)
is only a seminonn, one simply takes the quotient of £ by its subspace of all null
vectors and completes, obtaining a Hilbert C* -module in that way.
It is useful to note that (2.1) and (2.2) imply that

(\IlB, <I»!B = B*(\II, <l>h. (2.6)

Example 2.1.2.

1. Any C* -algebra 2l is an 2l-module 2l ~ 2l over itself, with (A, B)<.1 := A * B.

Note that the norm (2.5) coincides with the C* -norm by 1.(1.15).
2. Any Hilbert space 'It is a Hilbert C-module 'It ~ C in its inner product.
3. Let H be a Hilbert bundle over a compact Hausdorff space Q. The space of
continuous sections £ = ro(H) ofH is a Hilbert C*-module ro(H) ~ C(Q)
over ~ = C(Q);for \II, <I> E ro(H) the function (\II, <I»C(Q) is defined by

(\II, <I>)c(Q) : q 1-+ (\II(q), <I>(q», (2.7)

where the inner product is the one in the fiber .-l(q). The right action of
C(Q) on ro(H) is defined by stipulating that \II f is the section that maps q to
f(q)\II(q)·

In the third example the nonn in ro(H) is II \II II = SUPq(\II(q), \II (q»2" ,so that it is
I

easily seen that £ is complete.

Many Hilbert C*-modules of interest will be constructed in the following way.
A pre-C* -algebra is a *-algebra satisfying all properties of a C* -algebra except
perhaps completeness. Given a pre-C* -algebra ~,define a pre-Hilbert ~-module
e ~ ~ as in Definition 2.1.1, except that the final completeness condition is
omitted.

Proposition 2.1.3. In a pre-Hilbert ~-module (and hence in a Hilbert ~-module)

one has the inequalities

II'IIBII ~II'IIIIIIBII; (2.8)

(\II, <I»!B(<I>, \II)!B ~ 11<1>112 (\II, \II)!B; (2.9)
II(\II,<I»!BII ~ 11'111111<1>11· (2.10)
To prove (2.8) one uses (2.6), 1.(1.42), 1.(1.41), and 1.(1.15). For (2.9) we sub-
stitute <1>(<1>, \Ilh - \II for \II in the inequality (\II, \II)!B ~ O. Expanding, the
first tenn equals (\II, <I»!B(<I>, <l>h(<I>, \II)!B. Then use 1.(1.42), and replace <I> by
<1>/11<1>11. Equations 1.(1.15), (2.1), and (2.9) then imply (2.10). •
Corollary 2.1.4. A pre-Hilbert ~-module e~ ~ can be completed to a Hilbert
~-module.
356 IV. Reduction and Induction

t c.
One first completes in the norm (2.5), obtaining Using (2.8), the ~-action
t
on extends to a ~-action on c.
The completeness of ~ and (2.10) then allow
t
one to extend the !.B-valued sesquilinear form on to a ~-valued one on It is c.
easily checked that the required properties hold by continuity. •
In Example 2.1.2, it is almost trivial to see that m, 'H, and r o(H) are the closures
t,
of§{ (defined over Q{), of a dense subspace and of r(H) (defined over C~(X»,
respectively.
A Hilbert C*-module £ :;=: ~ defines a certain C*-algebra C*(c,~) that plays
an important role in the induction theory in 2.2. A map A : ~ for which there c c
c
exists a map A * : ~ £ such that
(111, Aet>hB = (A*III, et»!B (2.11)
for all 111, et> E C is called adjointable.
Theorem 2.1.5.
1. An adjointable map is automatically C-linear, ~-linear (that is, (AIII)B =
A (111 B) for all 111 E £ and B E ~), and bounded.
2. The adjoint of an adjointable map is unique, and the map A ~ A * defines an
involution on the space C*(c, ~) of all adjointable maps on c.
3. Equipped with this involution, and with the norm 1.(1.17), defined with respect
to the norm (2.5) on C, the space C*(c,~) is a C*-algebra.
4. Each element A E C*(c, ~) satisfies the bound
(Alii, AIII)!B :s: II A 112(111, III)!B (2.12)
for all 111 E c.
5. The (defining) action ofC*(£,~) on £ is nondegenerate.
We write C*(c, ~) ~ c :;=: ~.

The property of C-linearity is immediate. To establish ~-linearity one uses

(2.6); this also shows that A* E C*(£,~) when A E C*(c, ~).
To prove boundedness, fix 111 E C and define T\jJ : c ~ ~ by T\jJet> :=
(A * A 111, et»!B' It is clear from (2.10) that II T\jJ II :s: IIA* A 11111, so that T\jJ is
bounded. On the other hand, since A is adjointable, one has T\jJet> = (111, A* Aet»!B,
so that, using (2.1 0) once again, one has II T\jJ et> II :s: II A * A et> II 1111111. Hence
sup{ II T\jJ III 1111111 = I} < 00 by the principle of uniform boundedness (here it
c
is essential that is complete). It then follows from (2.5) that IIA II < 00.
Uniqueness and involutivity of the adjoint are proved as for Hilbert spaces; the
former follows from (2.4), the latter in addition requires (2.1).
The space C*(c,~) is norm-closed, since one easily verifies from (2.11) and
(2.5) that if An ~ A then A~ converges to some element, which is precisely A *.
As a norm-closed space of linear maps on a Banach space, C*(c, ~) is a Banach
algebra, so that its satisfies 1.( 1.14). To check the remaining axiom, one infers from
(2.5) and the definition (2.11) of the adjoint that IIA 112 :s: IIA* A II; using 1.(1.14) and
the argument leading to 1.(1.16), one first obtains IIA*II = IIAII, and subsequently
1.(1.15).
 2 Induction 357

Finally, it follows from (2.3), 1.(1.39), and (2.11) that for fixed lJ1 E C the map
A ~ ('II, AlJ1)!B from C*(c,~) to ~ is positive. Replacing A by A* A in 1.(1.40)
and using 1.(1.15) and (2.11) then leads to (2.12).
To prove the final claim, we note that for fixed'll, <I> E C, the map Z ~
'11(<1>, Z)!B is in C*(c, ~). When the right-hand side vanishes for all'll, <1>, it
follows from (2.2) that A (<I>, Z)!B = 0 for all A in the C*-algebra in ~ generated
by (c, c)!B. In any C*-algebra, the property AB = 0 for all A implies B = 0;
use an approximate unit if necessary. Hence (<I>, Z)!B = 0 for all <I> E c. Taking
<I> = Z, we conclude that Z = 0 by (2.4). •

Under a further assumption (which is by no means always met in our examples)

one can completely characterize C*(c, ~). A Hilbert C*-module over ~ is called
self-dual when every bounded ~-Iinear map cP : c ---+ ~ is of the form cp('II) =
(<I>, 'II)!B for some <I> E c.

Proposition 2.1.6. In a self-dual Hilbert C* -module C ~ ~ the C* -algebra

C*(c, ~) coincides with the space £(c)!B of all bounded C.-linear and ~-linear
maps on C.

In view of Theorem 2.1.5 we need to show only that a given map A E £(c)!B
is adjointable. Indeed, for fixed'll E C define CPA,1jI : C ---+ ~ by CPA,IjI(Z) :=
('II, AZ)!B' By self-duality this must equal (<I>, Z)!B for some <1>, which by definition
is A*'II. •

In the context of Example 2.1.2.1, one may wonder what C*(2l, 2{) is. The map
p : 2{ ---+ ~(2{) given by 1.(1.27) is easily seen to map 2{ into C*(2{, 2{). This map
is isometric (hence injective). Using (2.11), one infers that Ap(B) = p(AB) for all
A, B E 2{. Hence p(2{) is an ideal in C*(2{, 2l). When 2{ has a unit, one therefore
has C*(2{, 2{) = p(2{) :::::: 2l; cf. the proof of 1.1.2.1.
When 2{ has no unit, C*(2{, 2{) is the so-called multiplier algebra of 2{. One
may compute this object by taking a faithful nondegenerate representation 1f :
2{ ---+ ~('Jt); it can be shown that C* (2{, 2{) is isomorphic to the idealizer of 1f (2{)
in ~(1t) (this is the set of all B E ~(1t) for which B1f(A) E 1f(2{) for all A E 2{).
One thus obtains

C*(Co(X), Co(X» = Cb(X); (2.13)

C*(~o(1t), ~o(1t» = ~(1t). (2.14)

Equation (2.13) follows by taking 1f(Co(X» to be the representation on L2(X) by

multiplication operators (where L 2 is defined by a measure with support X), and
(2.14) is obtained by taking 1f(~o(1t» to be the defining representation; see the
paragraph following 1.1.6.3.
In Example 2.1.2.2 the C*-algebra C*(1t, C) coincides with ~(1t), because
every bounded operator has an adjoint. Its subalgebra ~o(1t) of compact operators
has an analogue in the general setting of Hilbert C* -modules as well; see 2.4.
358 IV. Reduction and Induction

2.2 Rieffel Induction

Given a Hilbert ~ -module E, the goal of the Rieffel induction procedure described
in this section is to construct a representation 1C x of C* (E, ~) from a representation
1Cx of ~. In order to explicate that the induction procedure is a generalization of
the GNS-construction I.1.5.4, we first induce from a state Wx on ~, rather than
from a representation 1Cx .
Construction 2.2.1. Suppose one has a Hilbert C* -module E ~ ~.
1. Given a state Wx on ~, define the sesquilinear form f.)~ on E by
----x
(\II, cf»o := wx({\II, cf»!B). (2.15)
Since Wx and {, )!B are positive (cf. (2.3)), this form is positive semidefinite. Its
null space is
- ----x
Nx = {\II EEl (\II, \11)0 = OJ. (2.16)
-x -x -
2. The form (, )0 projects to an inner product (,) on the quotient E / N x . If
Vx : E ~ E /Jilx is the canonical projection, then by definition
_ _ x ----x
(Vx \II, Vxcf» := (\II, cf»o. (2.17)
The Hilbert space i£x is the closure ofE/Jilx in this inner product.
3. The representation ii"X(C*(E, ~» is firstly defined on E/Jilx c i£x by
1C X(A)Vx \ll := VxA\II; (2.18)

itfollows that ii'x is well-defined and continuous. Since E /Jilx is dense in i£x,
the operator ii'X(A) may be defined on all ofi£x by continuous extension of
(2.18), where it satisfies 1.(1.19) and 1.(1.20).
The GNS-construction I. 1.5.4 is a special case of 2.2.1, obtained by choosing
E= ~ = m, as explained in Example 2.1.2.1.
The analogue of I.(1.58) and the property (2.11) imply that AJilx £;; Jilx , so
that (2.18) is well-defined. The continuity of :if x follows from (2.18) and (2.17),
- ---- x
which imply that llii'x (A)Vx \11112 = (A \II, A \11)0. Using (2.15), (2.12), and (2.10)
in succession, one obtains
(2.19)
On the other hand, 1.(1.51) applied to ~, used with the definition of IIAII for
A E C*(E, ~), implies that
IIAII = sup{IIii'X(A)II, Wx E S(~)}. (2.20)
As a corollary, one infers a useful property that will be used, e.g., in the proof
of Theorem 2.3.3.
Lemma 2.2.2. Let A E C*(E, ~) satisfy (\II, A\II)!B ~ 0 for all \II E E. Then
A ~O.
 2 Induction 359

It follows from (2.20) that E9 wx ES(')3)ii x is faithful; the condition on A implies

that ii x (A) :::: 0 for all wX• •

An illustration of the construction of the induced space is obtained by special-

izing Example 2.1.2.3 to the case H = HX, which is a vector bundle over Q with
typical fiber 1i x ; cf. III.(2.143). Hence we take £ = ro(HX), the space of contin-
uous sections fo(HX) of HX, and 1)3 = C(Q), made into a Hilbert C*-module by
(2.7) and the canonical right action of C(Q) on ro(HX). For any pure state Wx = q
on C(Q), the induced space is just 1i x •
When one starts from a representation 1Tx (1)3) rather than from a state, the general
construction proceeds as follows.
Construction 2.2.3. Start/rom a Hilbert C*-module £ ;:=': 1)3.

1. Given a representation 1T x (1)3) on a Hilbert space 1i x' with inner product (, >X'
the sesquilinear form (, )~ is defined on £ ® 1ix (algebraic tensor product) by
sesquilinear extension of
(2.21)

where v, W E 1i x . This form is positive semidefinite, because (, >x and ( , )')3

are. The null space is

N x = {Ijt E £ ® 'Hx I (Ijt, Ijt)~ = OJ. (2.22)

As in 1.(1.58), we may equally well write

(2.23)
2. Theform (, )~ projects to an inner product (, )X on the quotient £ ® 1i x /Nx ,
defined by
(2.24)

where Vx : £ ® 1ix -+ £ ® 1ix/Nx is the canonical projection. The Hilbert

space 1i x is the closure off ® 1ix/Nx in this inner product.
3. The representation 1T x (C*(£, 1)3» is then defined on 1i x by continuous extension
of

(2.25)
where lIx is the unit operator on 1i x ; this is well-defined, and the extension in
question is possible, since
(2.26)
To prove that the form defined in (2.21) is positive semidefinite, we assume
that 1Tx (1)3) is cyclic (if not, the argument below is repeated for each cyclic
summand; see 1.1.5.2). With q, = Li \IIi Vi and Vi = 1T X (Bi)Q (where Q is
a cyclic vector for 1T x (I)3», one then uses (2.21), (2.6), and (2.2) to obtain
--x . --x
(\11,\11)0 = (V,1T X«<P,<P)')3)v)x Wlth<P:= Li\lljBj. Hence (\11,\11)0 ~ Oby
360 IV. Reduction and Induction

(2.3) and the positivity of 1l' X' By (2.11) and (2.23), the operator A ® lix maps N x
to itself, so that (2.25) is well-defined.
To prove continuity, one computes II1l'X(A)Vx q, 112 = (v, 1l'x«A<I>, A<I>}!8)v>X
from (2.24) and (2.25); according to (2.12) and the property 1I1l'x(A)Il ::: IIA II
(cf. the text after 1.1.5.1), this is bounded by IIAII2(v, 1l'x«(<I>, <I>}!8)v)x' Since the
second factor equals II Vx q, 11 2 , this proves (2.26). •

Similarly, 1l'x is faithful and nondegenerate when 1l'x is.

To interrelate the above two formulations, one assumes that 1l'x is cyclic, with
cyclic vector Ox' Then define a linear map {j : e ~ e ® 'ltx by

(jw:=w®Ox' (2.27)

According to (2.15), (2.21), and 1.(1.57), this map has the property

(2.28)

By (2.17) and (2.24) the map (j therefore quotients to a unitary isomorphism

U : i£x ~ 'lt x , which by (2.18) and (2.25) duly intertwines if x and 1l'x.
Of course, any subspace of C*(e, IB) may be subjected to the induced represen-
tation 1l' x. This particularly applies when one has a given (pre-) C* -algebra 21 and
a morphism 1l' : 21 ~ C*(e, IB), leading to an induced representation 1l'X(21) on
1tx. Further to an earlier comment, one verifies that 1l'x is nondegenerate when 1l'
and 1l'x are. With slight abuse of notation we will write 1l'X(A) for 1l'X(1l'(A». The
situation is depicted in Figure 11.
We tum to a practical method of obtaining alternative and more explicit
realizations of 1l'X(C*(e, IB».

Proposition 2.2.4. Suppose one has a Hilbert space 1t; (with inner product
denoted by (, )!) and a linear map {j : e ® 1t x ~ 1t; satisfying

(2.29)

forallq" <i> E e®1t x ' Then {j quotientstoanisometricmapbetweene®1t x /Nx

and the image of {j in 'It;. When the image is dense this map extends to a unitary

1l' 1l'x

I
21 e ~

IB 1tX

~ 1t X
induction

FIGURE 11. Rieffel induction

 2 Induction 361

isomorphism U : 'H.x -+ 'H.;. Otherwise, U is unitary between 'H.x and the closure
of the image of U.
In any case, the representation Jr x (C* (£, ~» is equivalent to the representation
Jrf(C*(£, ~», defined by continuous extension of
(2.30)

It is obvious that NX = ker(U), so that, comparing with (2.25), one indeed has
U 0 Jrx = Jr: 0 U. •
As an abstract illustration of this technique, consider the space eel, 'H. x )'13 of
all antilinear maps f : £ -+ 'H. x satisfying
f('IIB) = Jrx(B*)f('II) (2.31)

for all 'II E £ and B E 23. Define a map U : £ ® 'H. x -+ e(l, 'H. x )'13 by linear
extension of
(2.32)

Taking the inner product of (U ('II ® v»( ct» with an arbitrary vector W E 'H. x' one
sees from (2.21) that U(q/) is the zero map iff q/ E Nx • The image of U may be
equipped with an inner product designed to satisfy (2.29), i.e., we put
(U(\II ® v), U(Cf> ® w»! := (v, Jrx«('II, Cf»'13)w>X. (2.33)

Comparing with (2.21), one sees that (2.29) is indeed satisfied, so that the comple-
tion of U(£ ® 'H. x ) in this inner product may be identified with the Hilbert space
'H.; of the preceding paragraphs. As we shall see, in practical applications one can
sometimes obtain a direct characterization of the space 'H.; thus defined.
So far, we have presented the simplest version of Rieffel induction, in which £
is a Hilbert C*-module. One may consider the following generalizations.
Firstly, it is not necessary that £ ;::::! 23 be complete. When t ;::::! 23 isn't,
an operator A satisfying (2.11) need neither be bounded on t,
nor automatically
satisfy (2.12). Let an adjointable operator A on satisfyt
(A'll, A'II)'13:::: C~('II, '11)'13 (2.34)

for some positive number CA. Using the reasoning leading to the bound (2.19), one
sees that this bound is still satisfied, with II A II replaced by CA. Moreover, defining
II A II as the smallest number C A for which (2.34) holds, one can still derive the
equality (2.20) in the same way. This equality, then, implies that II . II thus defined
is a norm on the space C*(t, 23) of all maps on £ satisfying (2.11) and (2.34).
The proof that C*(£, 23) is a C* -algebra in the complete case may then by copied,
showing that in the above norm C*(t, 23) is a pre-C*-algebra.
Using (2.24), (2.15), and (2.5), one shows that II V'll II :::: 11'1111, where the norm
on the left-hand side is in i£x, and the norm on the right-hand side is the one defined
in (2.5). It follows that the induced space 'H.x (or i£X) obtained by Rieffe1-inducing
from a pre-Hilbert C* -module is the same as the induced space constructed from
its completion.
362 IV. Reduction and Induction

Secondly, whether or not t

~ ~ is complete, one may drop the positivity
condition (2.3), as long as Jr x( (lit, 1It)'8) is a positive operator on 1{x for all lit E t.
For the latter condition is sufficient to guarantee that the form (2.21) is positive
t
semidefinite. In that case ~ ~ is called Jr x-positive.
This suggests that one may generalize the Rieffel induction procedure by alto-
gether omitting the C* -algebra ~ and its representation Jr x. The price one pays for
e
the absence of a ~-action on is the stringent positivity condition (2.36) below
(which in Rieffel induction is automatically satisfied).
Construction 2.2.5. Suppose one has a vector space t and a Hilbert space 1{ x'
where t is equipped with a sesquilinear form (, )'8(1t x ) that takes values in ~ (1{ x),
and for all lit, <I> E t satisfies
(lit, <I»;(1t x ) = (<1>, 1It)'8(1t x ); (2.35)
n
L(v;, (lit;, IItj}Vj)x ::: 0 (2.36)
;,j=1

for each n EN and all VI, ••. , Vn E 1{x and lit I , .•• , IIt n E t.In other words, the
matrix M E VJtn(~(1{x» with entries M;j = (lit;, IItj }'8(1t x ) is positive (cf II. I A).
Then
1. The form (, )~ on £ ® 1{x is defined by
(lit ® v, <I> ® w)~ := (v, (lit, <I»'8(1t x )wh. (2.37)

2. The induced Hilbert space 1{x is the closure of £ ® 1{x /Nx (where the null
space N x is defined as in (2.22» in the inner product (, )X inherited from (, )~,
defined as in (2.24).
3. The induced action Jrx (A) on Rx ofan adjointable operator A on £ satisfying
(2./1) and (2.12) with ~ """"* ~(1{x) is defined as in (2.25).
4. The induced action JrX(B) on Rx of an operator B E ~(1{x) satisfying
B(IIt, <I>}'8(1t x ) = (lit, <I>}'8(1t x )B (2.38)

for all lit, <I> E £ is defined by continuous extension of

JrX(B)Vx~:= VxlI® B~, (2.39)

where II is the unit operator on t.

The form (, )~ is positive because of the assumption (2.36). Equation (2.38)
implies that B maps N x into itself, so that Jrx (B) is well-defined. Using 1.(1.40),
one easily proves the inequality
(lI® B~, lI® B~)~ ~ IIBII2(~, ~)~; (2.40)
as in the proof of (2.26) this leads to the bound
IIJr X (B)1I ~ IIBII· (2.41)
This equally well holds for adjointable operators A on t. •
 2 Induction 363

This generalized induction procedure is known as Fell induction. Construction

11.1.4.3 is a special case of 2.2.5: Given a completely positive map Q : Il ~ ~
t
and a representation rr x (~) on 1{x' we take = Il, and define
(A, B)'13(1f.x) := rrx(Q(A* B)). (2.42)
The Hilbert space 1{x and the representation rrX constructed in 11.1.4.3 are then
exactly the same as the objects defined in 2.2.5.
Interestingly, the map W in 11.(1.29) is a special case of a map intrinsically
defined in Fell induction in general. Namely, in the setting of Construction 2.2.5,
pick a <I> E t,and define Wei> : 1{x ~ 1{x by
WeI>V:= Vx<l> ® v. (2.43)
Using (2.35) and (2.37), one computes its adjoint W; : 1ix ~ 1ix as
W; Vx \11 ® v = (<1>, \I1)'13(1f.x)v. (2.44)
Comparing (2.43) and (2.44) with 11.(1.29) and 11.(1.30), respectively, one sees
that in the special case 11.1.4.3 one has to put <I> = lL
One may pass from Rieffel induction to Fell induction by defining
(2.45)
Condition (2.36) then holds by the argument in the proof of 2.2.3, and may alterna-
tively be derived from Lemma 2.2.2. This is especially useful when one has found
a candidate for (, )'13 that fails to be positive, but that is rr x -positive. Alternatively,
one may have a family of C*-algebras ~(n) and sesquilinear forms (, )'13(0) for
which one would like to take a limit n ~ 00, which makes no sense at the level
of C* -algebras. It may then nonetheless be the case that the expressions (2.37),
defined via (2.45), do converge. We will see Fell induction in action in 2.10, 3.3,
3.5, and 3.8. In these applications one has 1{x = C, for which condition (2.36)
reads simply
(\11, \11k :::: o. (2.46)
A detailed comparison between classical reduction and quantum induction will
be given in 2.9.

2.3 The C* -Algebra of a Hilbert C* -Module

In preparation for the quantum imprimitivity theorem in the next section, and also
as a matter of independent interest, we introduce the analogue for Hilbert C*-
modules of the C* -algebra ~o(1{) of compact operators on a Hilbert space. This
is the C*-algebra most canonically associated to a Hilbert C*-module.
Definition 2.3.1. The collection Co(c, ~) of "compact" operators on a Hilbert
c
C* -module ~ ~ is the C* -algebra generated by the adjointable maps of the
type T:'eI>' where \11, <I> E C, and

T::eI>Z:= \11(<1>, Z)'13. (2.47)

364 IV. Reduction and Induction

We write Co(£' s.B) ~ £ ~ s.B, and call this a quantum dual pair.
The word "compact" appears between quotation marks because in general, ele-
ments of Co (£, s.B) need not be compact operators. The significance of the notation
introduced at the end of the definition will emerge from Theorem 2.3.3 below.
Using the (trivially proved) properties
(T~4»* = T!\,; (2.48)
AT~4> = Tl",,4>; (2.49)
T~4>A = T~A'4>' (2.50)
where A E C*(£, s.B), one verifies without difficulty that Co(£' s.B) is a (closed
2-sided) ideal in C*(£, s.B), so that it is a C*-algebra by Theorem 2.1.5. From (2.8)
and (2.10) one obtains the bound
II T: 4> II ~ II \1111 II <I> II. (2.51)
One sees from the final part of the proof of Theorem 2.1.5 that CO (£, s.B) acts
nondegenerately onE. When CO(£, s.B) has a unit, it must coincide with C*(£, s.B).
Proposition 2.3.2.
1. When £ = s.B = ~ (see Example 2.1.2.1) one has
Co(~,~) ~~. (2.52)
This leads to the quantum dual pair ~ ~ ~ ~ ~.
2. For £ = 11. and s.B = C (see Example 2.1.2.2) one obtains
o
C (11., C) = s.Bo(11.), (2.53)
whence the quantum dual pair s.B o(11.) ~ 11. ~ C.
One has T;ff 4> = p(\II<I>*); see 1.(1.27). Since p : ~ -+ s.B(~) is an isometric
morphism, th~ map f{J from the linear span of all T:'
4> to ~, defined by linear
extension of f{J(T;ff 4» = \11<1>*, is an isometric morphism as well. It is, in particular,
injective. When i has a unit it is obvious that f{J is surjective; in the nonunital case
the existence of an approximate unit implies that the linear span of all \II <1>* is
dense in~. Extending f{J to Co(~'~) by continuity, one sees from 1.1.3.10.4 that
f{J(CQ(~, ~)) = ~.
Equation (2.53) follows from Definition 1.1.6.3 and the fact that the linear span
of all T~.4> is s.B /(11.). •
In Example 2.1.2.3 one derives that CQ(ro(H), C(Q» is the C*-algebra of the
continuous field of C*-algebras over Q determined by H (in which ~q = 9'Jtn (C)
for all q E Q).
A Hilbert C*-module £ over s.B is called full when the collection {(\II, <I>)~},
where \II, <I> run over £, is dense in s.B. A similar definition applies to pre-Hilbert
C* -modules.
Given a complex linear space £, the conjugate space is equal to £ as a real e
vector space, but has the conjugate action of complex scalars.
 2 Induction 365

Theorem 2.3.3. Let e be a full Hilbert f13-module. The expression

(\II, cl>)co(t:.~) := T:'<I> (2.54)

in combination with the right action 1l'R(A)\II := A*\II, where A E C~(e, (13),
defines £ as a full Hilbert C* -module over CO(e, (13).ln other words, from e ~ f13
one obtains £ ~ C~(e, 1J3).Theleftaction1l'L(B)\II:= \IIB*ojf13 on £ implements
the isomorphism

(2.55)

We define l.2l to be CO(e, 1J3); in the references to (2.1) etc. below one should
substitute l.2l for f13 when appropriate. The properties (2.1), (2.2), and (2.3) follow
from (2.48), (2.50), and Lemma 2.2.2, respectively.
To prove (2.4), we use (2.54) with cl> = \II, (2.47) with Z = \II, (2.2), (2.6),
and (2.5) to show that (\II, \II)!! = 0 implies 11('11, 'II)~II = O. Since ('II, 'II)~ is
positive by (2.3), this implies (\II, 'II)~ = 0; hence'll = 0 by (2.4).
It follows from (2.6) and (2.50) that each 1l'L(B) is adjointable with respect to
(, }!!. Moreover, applying (2.5), (2.54), (2.51), and (2.8) one verifies that 1l'L(B) is
a bounded operator on £ with respect to II . II!!, whose norm is majorized by the
e
norm of B in 1J3. The map 1l'L is injective because is nondegenerate as a right
f13-module.
Let £e be the completion of £ in II . II!!; we will shortly prove that £e = £.
It follows from the previous paragraph that 1l'L(B) extends to an operator on £e
(denoted by the same symbol), and that 1l'L maps f13 into C*(£e, l.2l). It is trivial
from its definition that 1l'L is a morphism. Now observe that

1l'L«'II, cl>}~) = T::'<I>' (2.56)

for the definitions in question imply that

T::'<I>Z = 'II(cl>, Z}!! = Tl<l>'II = Z{cl>, 'II}~. (2.57)

e
The fullness of ~ IJ3 and the definition of CO(£e, l.2l) imply that 1l'L : I.B ~
C~(£co l.2l) is an isomorphism. In particular, it is norm-preserving by 1.1.3.10.5.
e
The space is equipped with two norms by applying (2.5) with IJ3 or with l.2l;
we write II . II~ and II . 11'<1' From (2.54) and (2.51) one derives

(2.58)

For'll E e we now use (2.5), the isometric nature of and (2.56) to obtain
1l'L'

11'11 II ~ = II T::' ",11 2 • From (2.51) with f13 replaced by l.2l one then derives the converse
1

inequality to (2.58), so that lI'IIh = II'IIII~. Hence £e = £, as e is complete in

II'II~ by assumption. In other words, the completeness of e as a Hilbert f13-module
is equivalent to the completeness of £ as a Hilbert l.2l-module.
We have now proved (2.55). Finally, noticing that as a Hilbert C*-module over
l.2l the space £ is full by definition of CO(e, 1J3), the proof of Theorem 2.3.3 is
complete. •
366 IV. Reduction and Induction

For later reference we record the remarkable identity

(Z, <I>)c;(t:,~)W = Z(<I>, W)~, (2.59)
which is a restatement of (2.57).

2.4 The Quantum Imprimitivity Theorem

Our aim in this section is to prove an operator-algebraic version of Theorem
1.4.1, in which special symplectic reduction is replaced by Rieffel induction. This
theorem will be based on the following concept.
Definition 2.4.1. Tho C* -algebras !2l and ~ are Morita equivalent when there
exists a full Hilbert C*-module £ over ~ under which!2l ~ CQ(£, ~). We write
!2l ~ ~ and!2l ~ £ ~ ~.
This definition is better behaved than its classical counterpart 1.3.7, for we have
Proposition 2.4.2. Morita equivalence is an equivalence relation in the class of
all C* -algebras.

The reflexivity property ~ ~ ~ follows from (2.52), which establishes the

quantum dual pair ~ ~ ~ ~ ~. Symmetry is implied by (2.55), proving that
!2l ~ £ ~ ~ implies ~ ~ "£ ~ !2l.
The proof of transitivity is more involved. When !2l ~ ~ and ~ ~ Q: we have
the chain of quantum dual pairs
!2l ~ £1 ~ ~ ~ £2 ~ Q:.

We then form the linear space £1 ®~ £2 (which is the quotient of £1 ® £2 by the

ideal I~ generated by all vectors of the form WIB ® '112 - WI ® BW2), which
carries a right action 7r:>(Q:) given by
7r~(C)(WI ®~ '11 2) := WI ®~ (W2C), (2.60)
Moreover, we can define a sesquilinear map (, )~ on £1 ®~ £2 by
(WI ®~ '112, <1>1 ®~ <l>2)~ := ('112, ('lit. <l>lh<l>2}e:. (2.61)
With (2.60) this satisfies (2.1) and (2.2); as explained prior to (2.6), one may
therefore construct a Hilbert C* -module, denoted by £® ~ Q:. (Remarkably, if
one looks at (2.61) as defined on £1 ® £2, the null space of (2.5) is easily seen to
contain I~, but in fact coincides with it, so that in constructing £® one only needs
to complete £1 ®~ £2.)
f
Apart from the right action 7r:> (Q:), the space £® carries a left action 7r (!2l): The
operator
7r~(A)(Wl ®~ '112) := (A 'Ill) ®~ '112 (2.62)
is bounded on £1 ®~ £2 and extends to £®. We now claim that
q(£®, Q:) = 7rf(!2l)· (2.63)
 2 Induction 367

Using (2.47), the definition of ®'ll, and (2.2), it is easily shown that

n~(T:'(1/I2.<t>2)'lI.<t»OI ®'ll Q2 = \III ®'ll (\112, <1>2(<1>1, OI)'ll)'l102. (2.64)

Now use the assumption CQ(£2,1t) = ~; as in (2.54), with ~ and £ replaced by
It and £2, this yields (\II, <I»'ll = T$.<t>. Substituting this in the right-hand side of
(2.64), and using (2.47) with ~ replaced by It, the right-hand side of (2.64) becomes
\III ®'ll \112(<1>2(<1>1, Ol)'ll, 02k Using \II B* = ndB)\II (see 2.3.3), (2.11) with It
instead of~, (2.61), and (2.47) with ~ replaced by It, we eventually obtain
(2.65)

o
This leads to the inclusion C (£®' It) £ nT(Q1.). To prove the opposite inclusion,
one picks a double sequence {\II~, <I>~} such that L~ T:, <t>i is an approximate unit
2' 2
in ~ = CQ(£2, It). One has limN L~ \II~(<I>~, Z)( = Z from (2.47), and a short
computation using (2.47) with (2.61) then yields
N

lim " T.~ "" .hi ... "" ... i =

N L...J "'1 ""'lI'" 2."'1 ""'lI"'2
n~(T:" <t> 1 ).
i

Hence nT(~l) £ CQ(£®, It), and combining both inclusions one obtains (2.65).
Therefore, one has the quantum dual pair m~ £® ~ It, implying that m~ It.
This proves transitivity. •
The following simple example of this concept will have nontrivial consequences.
Proposition 2.4.3. For any Hilbert space 11., the C* -algebra ~o(11.) of compact
operators on 11. is Morita equivalent to C, with quantum dual pair ~o(11.) ~ 11. ~
C. In particular, the matrix algebra VRn (C) is Morita equivalent to C.
This is immediate from (2.53). In the finite-dimensional case one has VRn(C) ~
cn n
;=! C, where VR (C) and C act on cn
in the usual way. The double Hilbert
C* -module structure is completed by specifying
(z, w)c = ziw i ;
«(z, w) m.(c»ij = Zi wi , (2.66)
from which one easily verifies (2.59).
In practice, the following way to construct quantum dual pairs, and therefore
•
Morita equivalences, is useful.
Proposition 2.4.4. Suppose one has
• two pre-C* -algebras !it and ~;
• a full pre-Hilbert ~-module e;
e, e
• a left action offit on such that can be made into afull pre-Hilbert fit-module
with respect to the right action nR(A)\II := A*\II;
• the identity
(2.67)
368 IV. Reduction and Induction

(for all \II, <1>, Z E t) relating the two Hilbert C* -module structures;
• the bounds (for all A E §l and B E ~)

(\liB, \IIB}Q ~ liB 112(\11, \II}Q; (2.68)

(A\II, A\II)~ ~ IIAII2(\II, \II)~. (2.69)

Then!2l !!'!- ~, with quantum dual pair!2l :r=: c :r=: ~, where c is the completion
of t as a Hilbert ~-module.

Using Corollary 2.1.4 we first complete t to a Hilbert ~-module C. By (2.69),

t,
which implies II A \1111 ~ II A II II \1111 for all A E §l and \II ~ the action of §l on t
t
extends to an action of!2l on c. Similarly, we complete to a Hilbert !2l-module
e e
c ; by (2.68) the left action nL(B)\II := \II B* extends to an action of ~ on c • As
in the proof of Theorem 2.3.3, one derives (2.58) and its converse for E ! t,
so
that the ~-completion c of £ coincides with the !2l-completion ec of £; that is,
ec=e.
e c
Since is a full pre-Hilbert §l-module, the !2l-action on is injective, hence
faithful. It follows from (2.67), Theorem 2.3.3, and (once again) the fullness ofe
that!2l ~ q(c, ~). In particular, each A E !2l automatically satisfies (2.11). •

Clearly, (2.67) is inspired by (2.59), into which it is turned after use of 2.4.4.
For example, one may take m= ~O(L2(Q» (where Q is a manifold), whose
dense subalgebra §l consists of the Hilbert-Schmidt operators with kernel in
C~(Q x Q). This subalgebra acts on t = C~(Q) in the obvious way. Further
t,
taking ~ = ~ = C, with self-evident action on one generalizes (2.66) to

(\II, <I>}c = (\II, <1»;

(\II, <I>}cgo(QxQ)(q, q') = \II(q)<I>(q'). (2.70)

The bounds (2.68) and (2.69) are trivially satisfied, so that in this case Proposition
2.4.4 reconfirms 2.4.3.
After this preparation, we pass to the quantum imprimitivity theorem; cf. its
classical analogue Theorem 1.4.1.

Theorem 2.4.5. There is a bijective correspondence between the nondegener-

ate representations of Morita-equivalent C* -algebras !2l and~, preserving direct
sums and irreducibility. This correspondence is as follows.
c
Let the pertinent quantum dual pair be m :r=: :r=: ~. When no (m) is a repre-
sentation on a Hilbert space 11.0 , there exists a representation nx(~) on a Hilbert
space 11.x such that no is equivalent to the Rieffel-induced representation nX
defined by (2.25) and the above quantum dual pair.
In the opposite direction, a given representation nx(~) is equivalent to the
Rieffel-induced representation nO, defined with respect to some representation
e
no(m) and the quantum dual pair ~ :r=: :r=: !2l.
Taking no(m) = nX(m) as just defined, one has nO (~) ~ nx(~)' Conversely,
taking nx(~) = nO(~), one has nX(!2l) ~ no (!2l).
 2 Induction 369

1fx
IB
~
~ ~
1{x

£ ~
~
~
~ 1{x := 1{u

IB
~ 1{u

FIGURE 12. Quantum imprimitivity theorem: 1t" :::: 1tx and 1f" :::: 1fx

See Figure 12. The idea of the proof is the same as in 1.4.1; its execution is,
in fact, simpler. Starting with 1fx (IB), we construct 1f x (~) with Rieffel induction
e
from the quantum dual pair~ ~ ~ IB, relabel this representation as 1fu(~), and
move on to construct 1fu (IB) from Rieffel induction with respect to the quantum
dual pair IB ~ £ ~ ~. We then construct a unitary map U : 1{u ~ 1{x that
intertwines 1fU and 1fX'
e
We first define U : £ ® ® 1{x ~ 'fix by linear extension of

(2.71)

Note that U is indeed (:-linear. Using (2.71), the properties 1.(1.20) and 1.(1.19)
with qJ replaced by 1fx' (2.21), and (2.47), one obtains
- - 'B X
(UWI ® q>l ® VI, UW2 ® q>2 ® V2)X = (q>l ® VI, T1iI1 •1iI2 q>2 ® V2)O' (2.72)

Now use the assumption ~ = CQ(e, IB) to use (2.59), and subsequently (2.24)
and (2.25), all read from right to left. The right-hand side of (2.72) is then seen
to be equal to (Vxq>, ® V"1fX«w,, W2).<t)VX q>2 ® V2)X. Now put 1fx = 1fu and
1{x = 'fi u , and use (2.21) and (2.24) from right to left, with X replaced by u. This
shows that

(UWI®q>I®V" UW2®q>2®V2>X = (Vu(W,®Vxq>,®v,), Vu (W2®Vx q>2®V2»u.

(2.73)
e
In particular, U annihilates W ® ci>, where ci> E ® 1{x, whenever ci> E N x or
W ® Vx ci> E N u . Hence we see from the construction firstly of 1{x = 'fiu from
e ® 'fix, and secondly of 'fiu from £ ® 1{u (cf. 2.2.3), that U descends to an
isometry U : 'fiu ~ 1{x, defined by linear extension of
370 IV. Reduction and Induction

Using the assumptions that the Hilbert C* -module £ ~ ~ is full and that the
representation Jr x (~) is nondegenerate, we see that the range of [; and hence of
U is dense in ?ix, so that U is unitary.
To verify that U intertwines Jr q and Jr x' we use (2.74) and (2.25), with X replaced
by a, to compute
(2.75)
where the left action of B E ~ on \II E "£ is as defined in 2.3.3. Thus writing
JrL(B)\II = \IIB*, using (2.6), 1.(1.19) with rp replaced by Jr x ' and (2.74) from
right to left, the right-hand side of (2.75) is seen to be Jr x(B)U Va (\II ® Vx <I> ® V).
Hence UJrq(B) = Jrx(B)U for all B E~.
Using the proof that the Morita equivalence relation is symmetric (see 2.4.2),
one immediately sees that the construction works in the opposite direction as well.
It is easy to verify that Jrx = Jr x ' $ Jr X2 leads to JrX = Jrx' $ JrX2. This also
proves that the bijective correspondence Jr x (~) ~ Jr x (!X) preserves irreducibility:
When Jr x is irreducible and Jrx isn't, one puts JrX = Jrq as above, decomposes
Jrq = Jrq' $ Jrq2, then decomposes the induced representation Jrq(~) as Jra =
Jra' $ Jr q2 , and thus arrives at a contradiction, since Jr q ~ Jr X. •

Combined with Proposition 2.4.3, this theorem leads to a new proof of Corollary
1.2.2.6. Furthermore, in the light of the example given after the proof of Proposition
2.4.4, the first part of the proof of Theorem III.3. 7.1 is now seen to be an application
of Theorem 2.4.5.

2.5 Quantum Marsden-Weinstein Reduction

We come to a class of examples of Hilbert C* -modules and Rieffel induction which
is of central importance to applications in physics. What follows may be seen as the
quantum counterpart of the Marsden-Weinstein symplectic reduction procedure
in l.5. For simplicity, proofs are given only for the unimodular case (recall that
every compact group is unimodular).
For clarity of presentation, we do not start with the most general assumptions;
the following result will be generalized in due course.
Theorem 2.5.1. Let U be a representation ofa compact Lie group H on a Hilbert
space?i, with corresponding representation Jr of the group C*-algebra C*(H);
cf III. (1.89). The formula

JrR(f) = L dh f(h)U(h)-1 (2.76)

defines a right action JrR of C*(H) by continuous extension from f E Crgo(H). In

conjunction with the map (, }c'(H) : ?i x ?i ~ C*(H), defined by
(\II, <I>}c'(H) : h ~ (\II, U(h)<I», (2.77)
one obtains a pre-Hilbert C* -module?i ~ C*(H). Completion therefore produces
a Hilbert C*(H)-module.
 2 Induction 371

Using III.(1.80) with c = 1, it is trivial to check that (2.76) defines a right

action. The verification of (2.1) and (2.2) is equally straightforward. Because U is
continuous, the function defined by (2.77) lies in C(H), and therefore in C*(H).
For \11 i= 0 this function is nonzero, as it is nonzero at least at the identity e. Hence
(2.4) follows.
To prove (2.3), consider the isomorphism C*(H) ~ 7rL (C*(H»; this follows
from III.(1.95) and the fact that compact groups are amenable, so that C*(H) ~
C;(H). Picking Q E L 2(H), a shift of variables shows that

(2.78)

on the right-hand side Q is regarded as an element of C*(H), which is justified,

since for compact H one has the inclusion L2(H) C C*(H). This proves the
desired positivity.
A different proof of positivity, which does not use the isomorphism between
C* (H) and C;( H), is as follows. Consider the function 1H. Assuming that the Haar
measure dh is normalized to unit volume, one sees from 111.(1.80) that 1H 1H = *
I H, whereas III.(1.8l) shows that IH is self-adjoint in C*(H). Hence IH lies in the
positive cone ofC*(H) by 1.(1.39). Accordingly, for any representation7rp (C*(H»
on a Hilbert space 1f.P ' with inner product (, )p, one has

Applying this with 7r p = 7r ®7r x' where 7r x is arbitrary, and choosing \11 p = \11 ® \11 x
for some \11 x E 1f.x' one obtains

(\11 x' 7rX «\11, \I1)c'(H»\I1 x)x 2: O.

Hence (\11, \I1)c'(H) 2: 0 by the proof of Theorem 1.1.1.8. We have now verified
all conditions for a pre-Hilbert C*-module; completion is possible by Corollary
21A •

To see what the completion may look like, consider the example H = U(1)
in the regular representation 7r = 7rL on 1f. = L2(U(1». Fourier-transforming
L2(U(l» to 12, one infers that C*(U(l) ~ lo; cf. 1.1.6.1 and III.(1.86). From
(2.77) one derives that on 12 one has

(W, <1ȣ0: n ~ Wn-<l>n. (2.79)

Since the norm in lo is the sup-norm, it follows that the Hilbert C* -norm (2.5) on
12 is the sup-norm as well. We conclude that the completion of 12 ~ lo is lo ~ lo
(sine the commutative C*-algebra lo is already complete).
In Theorem 2.5.6 we will generalize this example to arbitrary Lie groups with
multiplier. Partly in preparation for this generalization, and partly as a matter of
interest for physics, we first consider a generalization of 2.5.1 to representations
with multiplier (see III. 1.3 and III. 1.5).
372 IV. Reduction and Induction

Proposition 2.5.2. Let c be a multiplier on a compact Lie group H, and let U (H)
be a c-representation on a Hilbert space ?to Replacing C*( H) by the twisted group
algebra C*(H, c), all statements andformulae of Theorem 2.5.1 hold.

The argument is essentially the same as for 2.5.1. In the first proof of positivity
one should, of course, include the factor c in 111.(1.83). Using 111.(1.65), 111.(1.92),
and III.(1.29), which leads to the cancellation of all factors c, (2.78) still follows.
In the second proof of positivity one should take :rrx to be a representation of
C*(H, c), so that:rr p is a representation of C*(H) (since U and Ux have mUltiplier
c and c, respectively, and cc = 1). •

Here it is crucial that (2.76) contains U(h)-I rather than U(h- I ). In view of
Theorems III.1.4.4 and 111.1.9.5, and the last paragraph in 1.5, we may regard the
above "twisted" version of the construction in 2.5.1 as a quantum analogue of
Marsden-Weinstein reduction for momentum maps that are not Co-equivariant.
We now tum to Rieffel induction in the case at hand. While an almost trivial
matter, this is nonetheless fairly instructive, especially for the purpose of com-
parison with the noncompact case treated below. For simplicity we look only at
ordinary representations (c = 1).

Proposition 2.5.3. In the situation of Theorem 2.5.1, apply Construction 2.2.3

with ~ = C* (H) and a non degenerate representation :rrx(C* (H», or, equivalently,
a representation Ux(H).
The induced space?t x is isomorphic to P;d(?t ®?t x ), where P;d is the projection
onto the subspace of?t ®?t x transforming trivially under the representation
U ® Ux(H).
Any bounded operator A on 1t commuting with U(H) satisfies (2.11) and
(2.12), so that its induced representative :rrX(A) on ?t x may be defined. Under
the isomorphism of the previous paragraph, :rrX(A) is the restriction of A ® llx to
P;i?t ® ?t x )·

By the discussion surrounding (2.34), we may start from the pre-Hilbert C*-
module?t ~ C*(H). Using III.(1.89) with :rr replaced by :rrx and (2.77), one
computes the inner product (2.21) as

(\II ® v, ct> ® w)& = i dh (\II, U(h)ct»(v, Ux(h)wh· (2.80)

For arbitrary q" <I> E ?t ®?t x one therefore has

(q" <1»& = i dh ("'. U ® U x (h)<I»1i®1i x • (2.81)

Since the integrand is bounded and JH d h = 1, one may bring the integral over H
inside the inner product. The well-known expression

P;d = i dh U(h) (2.82)

 2 Induction 373

for the projection on the trivial representation of a compact group gives

- - x - -
(\11, <1»0 = (\11, P;d <I> hUsH, . (2.83)

Hence the null space N"x is the orthogonal complement (P;d('It ® 'ltx))1., and since
for a closed subspace K; c 'It one has 'It/K;1. ~ K;, the first claim follows.
To prove (2.12) for A E U(H)' we take a vector state Wx on C*(H) such that
wx(B) = (Q x ' Jtx(B)Qx)x' and use (2.83) to obtain

Wx «(A \11, A \I1)c'(H» = (A ® lIx \11 ® \11 x' p;d(A ® lIx \11 ® \11 x )ht®H, . (2.84)
The assumption A E U(H)' implies that A ® lIx commutes with P;d' so that the
right-hand side is bounded by II A 112(\11 ® \11 X' P;d(\I1 ® \11 x ))H®H x • In this expression
we rewrite the second factor by using (2.84) with A = 1I from right to left. This
yields

w x «(A\I1, A\I1}C*(H»:::: IIAII 2 w x «(\I1, \I1)c·(H».

Since this is true for all vector states, (2.12) follows.
Finally, the identification of Jt x (A) with A I P;d('It ® 'lt x ) is obvious. •

Let us now examine a possible generalization of the construction in Theorem

2.5.1 to the case where the Lie group H is merely locally compact. When H is
noncompact, the function defined by (2.77) will not, in general, lie in C*(H) for
all \11, <I> E 'It. For example, for 'It = L2(JR) and U = UR, equation III.(1.94)
and its derivation implies that (\11, \I1)c'(lR) as defined by (2.77) lies in C*(JR) iff
\11 is such that its Fourier transform lies in Co(JR). There certainly exist functions
\11 ¢ L 1(JR), \11 E L 2(JR), for which this is not the case.
The way out is simply to try to find a dense subspace of 'It that does have the
required property. As this subspace will not be stable under the action of C*(H),
one in addition needs to identify a suitable dense subalgebra of C*(H).1t simplifies
the discussion to do this once and for all, taking C'g"(H).

Theorem 2.5.4. Let U be a representation of a Lie group H on a Hilbert space

'It containing a dense subspace Ethat for each f E C'g"(H) is stable under JtR(f)
(as defined by (2.76)) andfor which the function h ~~ (\11, U(h)<I» lies in C'g"(H)
for all \11, <I> E E.
For all \11 E Ethe function h f-+ (\11, U(h)\I1) is a positive element ofthe reduced
group algebra q(H).
Hence when H is amenable, the operator (\11, \I1)c'(H), defined as in (2.77),
is positive for all \11 E E, so that the right action (2.76) makes E ;=: C'g"(H)
a pre-Hilbert C'g"(H)-module, which may be completed to a Hilbert C*-modu/e
e ;=: C*(H). Under the assumption of amenability, a bounded operator A on 'It
that, along with its adjoint A *, leaves Estable and commutes with U (H) satisfies
(2.11) and (2.12).
More generally, the diagram E ;=: C'g"(H) defined by (2.76) and (2.77)
is Jtx-positive when Ux(H) is weakly contained in the (left or right) regular
representation.
374 IV. Reduction and Induction

The first proof of positivity of Theorem 2.5.1 generalizes to the noncompact

case; the only change is that one should take <I> E C'g"(H) c L 2(H). (For a
bounded operator, positivity on a dense subspace of a Hilbert space implies posi-
tivity.) The second proof of positivity has to be modified by the introduction of an
approximation method, as follows.
It can be shown that an amenable group H has a family of subsets {Uj}jEJ,
where J is a directed index set, with the following properties:
• Each Uj is measurable, with finite Haar measure /L(Uj) .
• The Uj eventually fill up H in the precise sense that the family of functions gj E
Ll(H) C C*(H) defined by gj = (/L(U j »-1/2 XUj (with XE the characteristic
*
function of a Borel set E) satisfies lim j g j gj = 1H pointwise on H.
*
Clearly, each gj gj is a positive element of C*(H).
In the second proof of positivity one now replaces IH by gj * gj. Using the
bound gj * gj ::::: 1H and the Lebesgue dominated convergence theorem, one can
interchange lim j and the H -integration, and the result follows.
The first part of the second claim follows from our definition "C*(H) = C;(H)"
of amenability. The fact that A as specified satisfies (2.11) is trivial. The proof
of (2.12) uses the above approximation technique as well. The operator p~ :=
IH *
dh U ® Ux(h)gj gj(h) is well-defined, as gj has compact support. One then
proceeds as in the proof of Proposition 2.5.3, replacing Pid by p~. Taking the limit
in j yields the claim.
The final point is obvious from Proposition 111.1.7.7, according to which one
may work with C;(H) even when H is not amenable. •
As in the noncompact case, the above construction may be generalized.
Proposition 2.5.5. Let c be a multiplier on a Lie group H, and let U(H) be a
c-representation on a Hilbert space 1i. Replacing C'g"(H) by the space B'g"(H) of
bounded measurable functions with compact support that are smooth near e (cf.
1/1.1.7). regarded as a dense subalgebra of the twisted group algebra C*(H, c),
all statements andformulae of Theorem 2.5.4 hold.
The verification of this claim is similar to that of 2.5.5. D
The amenability of H is sufficient, but by no means necessary, for the positivity
of (\II, \II)c'(H).
Theorem 2.5.6. Let c be a multiplier on a Lie group H, and apply the construction
in 2.5.4 to the case 1i = L2(H) and the c-representation U = U R, defined by
1Il.(1.98), and £ = B'g"(G) (or, when c = 1, by C'g"(H».
The ensuing Hilbert C*-module is C*(H, c) ~ C*(H, c); cf. Example 2.1.2.1.
In particular, the completion of £ in the norm (2.5) is C*(H, c).
Let Wx be a vector state on C*(H, c), as in the proof of 2.5.3. Analogously to
(2.78), one derives
(2.85)
 2 Induction 375

This firstly proves positivity, and secondly shows that the norm (2.5) of \II coincides
with its norm as an element of C*(H, c). The rest is obvious from III.(1.99) and
(\II, <I»C'(H,c) = \11* * <1>, (2.86)
which follows from 111.(1.80), III.(1.81), and lll.(1.29). •
The C* -algebra of "compact" operators is therefore given by (2.52).
Rieffel induction with noncompact groups differs essentially from the compact
case. Proposition 2.5.3 breaks down, because the H -integration may no longer be
brought inside the inner product, and the would-be projection Pid does not exist.
However, (2.80) and (2.81) are still valid, and are often computable. A simple
example is given around (2.123).

2.6 Induction in Stages

This section parallels its classical counterpart 1.8. We return to general induction
in order to prove a theorem on Rieffel induction in stages. The comments on its
classical version Theorem 1.8.1 apply here as well.
Theorem 2.6.1. Suppose that 23 and It are C* -algebras, and that one has a Hilbert
C*-module e2 ;:::: It, a homomorphism 1f2 : 23 -+ C*(e2' It), and a representation
1fy(lt) on a Hilbert space 1iy. On Rieffel induction, these data lead to an induced
space 1i Y and an induced representation 1f y (23) on 1i Y ; cf. Figure 11.
Now assume that one in addition has a Hilbert C* -module el ;:::: 23 and a
morphism 1f1 : 2l -+ C*(eJ, 23) (where ~ is a C*-algebra), and take 1ix and
1fX in Figure 11 (with e and 1f replaced by el and 1f1) to be equal to 1iY and
1f Y , respectively. We denote the corresponding induced Hilbert space 1i x and
representation 1fX by 1iHY and 1f HY , respectively.
Define a It-valued sesquilinear form (, )~ on el ® e2 by sesquilinear extension
of(cf. (2.61)
(2.87)
and let It act on el ® e2 from the right by [ ® 1fR (where 1fR is the given right
action oflt on e2). Then the conditions (2.1) and (2.2) (with 23 replaced by It)
are met,formula (2.5) defines a seminorm on el ® e2, and one obtains a Hilbert
C*-module el®e2 over It by removing the null space and completing; cf. the text
preceding (2.6). Moreover, the given action 1f1 of~ on el leads to a morphism
1f~ : ~ -+ C*(el®e2, <1:), obtained by quotienting and extending 1f1 ® [.
The induced space 1i~ defined by Rieffel induction from the Hilbert <1:-module
e I ®e2 and the representation 1fY (<1:) is equivalent to 1iH Y , and the corresponding
induced representation 1f~(~) is equivalent to 1fHY (~).
All statements except those in the final paragraph are easily verified from the
pertinent definitions. The situation is summarized in Figure 13, which should be
compared with its classical counterpart Figure 7.
Let \IIj, <l>j E ej (i = 1, 2), and v, W E 1iy. We denote the element Vy \112 ® v E
1iY by [\112 ® v]ll, and, similarly, for Q E 1iY we denote the projection of \III ® Q
376 IV. Reduction and Induction

1iY ••- - -

FIGURE 13. Rieffel induction in stages: 1C'Y :::::: 1t~ and 7r"Y :::::: 7r~

to 1i7CY by [w ® Q]/. Also, the projection ofwl ® W2 to el®e2 is called Wl®W2,

and the projection of a ® v (where a E el®e2) to 1i~ is written as [a ® v]l/I.
The essential point is the equality

([Wl®[W2®V]/J]/, [<I>l®[<I>2®W]/J]/)7C Y = (v. rry({w2. rr2«(Wl. <l>lha)<I>2)e:)W)y.

(2.88)
where the left-hand side is an inner product in 1i7CY , the right-hand side being in
1i y • This follows by applying (2.21) twice. Using (2.21), (2.87), and (2.1) (for
(. )e:), one computes that the right-hand side coincides with the inner product of
[Wl®W2 ® V]III and [<I>l®<I>2 ® w]I/1 in 1i~.
From this, one easily infers that the map from the pertinent dense subspace of
1i7CY to 1i~ that sends [W 1 ® [W2 ® V]I 1]/ to [W 1® W2 ® V]I 11 is well-defined, and
extends to a unitary isomorphism U between 1i7CY and 1i~. One then trivially sees
from their definitions that rr 7CY and rr~ are intertwined by U. •

With trivial modifications, the theorem might equally well have been formulated
and proved in terms of pre-Hilbert C*-modules and pre-C*-algebras.
In complete analogy with the classical case (cf. 1.8), we may specialize Theorem
2.6.1 to the case of quantum Marsden-Weinstein reduction in stages. We repeat
the setting of Theorem 2.5.4, with H replaced by G.
 2 Induction 377

C*(G) -----. 1t X "0--- C*(H)

FIGURE 14. Quantum Marsden-Weinstein reduction in stages: 1t[l) ::::: 1tb) and 1r(~) ::::: 1r(~)

Theorem 2.6.2. Suppose G is a Lie group, with closed subgroup H. Let U be

a representation of G on a Hilbert space 1t containing a dense subspace t that
for each f E C~(G) is stable under JrR(f) (as defined by (2.76) with H replaced
by G), andfor which the function (\11, ct>)c*(G) : x t-+ (\11, U(x)ct» lies in C~(G)
t,
for all \11, ct> E such that (\11, \11) C*(G) is positive for all \11. This defines a pre-
t
Hilbert C* -module ;= C~(G), which may be completed to a Hilbert C* -module
£;= C*(G).
Now choose a representation Ux (H), and apply Rieffelinduction to £ ;= C* (G)
from the representation JrX(C*(G» on 1t x (cf. DefinitionIIl.2.8.4 with P replaced
by G), corresponding to the representation UX(G) induced from Ux(H); see
&)
//I. (2.175). Call the doubly induced Hilbert space 1tb), with Jr the corresponding
doubly induced representation of some (pre) C*-algebra ~ ~ C*(£, C*(G».
Restrict U(G) to H, and Rieffel-induce on £ ;= C*(H)from Ux(H), obtaining
a Hilbert space 1t2l) and an induced representation Jr(1)(~).
There exists a unitary U : 1tb) ~ 1t2l) intertwining Jrft)(~) and Jra)(~).

We specialize Figure 13 to Figure 14. In this specialization we have replaced

£\®£2 by 1t, andJr? by Jr\. This is justified by the following quantum counterpart
of Lemma 1.8.3.
378 IV. Reduction and Induction

Lemma 2.6.3. Define the pre-Hilbert C*-module £®C~(G) ;::= C~(H) as in

the proofof2.6.1, with £1 replaced bye as specified in 2.6.2, and £2 = C~(G) C
L2(G) carrying the left-regular representation 11:L ofCrgo(G) C C*(G).
This module is equivalent (in the obvious sense) to £ ;::= Crgo(H), defined by
restricting t;::= Crgo(G)from G to H.
Using (2.87) with ~ = C*(G) and 11:2 = 11:L, one easily shows that
(\}II ® \}I2. <1>1 ® <l>2k,?"(JI) = (11:R(\}I2)\}II.11:R(<I>2)<I>lk,?"(H). (2.89)

where 11:R is defined as in (2.76) with H replaced by G. Hence we define V:

£ ® Crgo(G) --+ £ by linear extension of
(2.90)
By (2.89) this map quotients well to V : £®Crgo(G) -+ £, and carries the pre-
Hilbert C*-module £®C;:"(G) ;::= C;:"(H) into £ ;::= C~(H). •
Theorem 2.6.2 now follows from 2.6.1 and 2.6.3.
This particular case of Theorem 2.6.1 has a further specialization that historically
•
was the first example of a theorem on induction in stages.
Corollary 2.6.4. Let K be a Lie group, and let H C G C K be closed subgroups.
The representation UY(K) induced from a representation Uy(G) that is itself in-
ducedfrom a representation Ux(H) (so that Uy(G) ~ UX (G» is equivalent to the
representation UX(K) directly induced from Ux(H).

Apply Theorem 2.6.2 with m= C*(K), 1-£ = L 2(K), £ = C;:"(K), and 11:\ =
11: L. We will show in the next section that the positivity condition is met; see the
argument following (2.92), with H replaced by G and P by K. •

2.7 The Imprimitivity Theorem/or Gauge Groupoids

Recall the gauge groupoid P x H P ~ Q defined by a principal bundle P( Q, H. r)
(cf. 111.3.1.7) and its C* -algebra C*(P x H P) (see III.3.6.3). Using Proposition
2.4.3 and Theorem 111.3.7.1, particularly III.(3.78), one may infer that C*(P x H P)
and C*(H) are Morita equivalent. The aim of this section is to restate this result
with the machinery developed in this chapter, by directly constructing a quantum
dual pair. This provides considerable insight into the situation, using the material
of 2.4 in an instructive way.
Theorem 2.7.1. For any principal bundle P(Q, H. r) with associated gauge
groupoid P XH P ~ Q. the groupoidC*-algebra C*(PXH P) is Morita equivalent
to C*(H).
We apply the construction of Theorem 2.5.4 with 1-£ = L 2 (P), U = U R, and
t = C~(P); see III.2.8 and III.(2.138). The presence of an H-invariant measure
IL on P is not essential, but simplifies some of the formulae. For the same reason,
 2 Induction 379

we retain our standing assumption that H is unimodular. Hence the right action
1fR(C~(H» is given by specializing (2.76), yielding

1fR (f)'II(x) = L dh 'II(xh-1)f(h), (2.91)

and the map ('II, <I>}c*(H) is defined by (2.77), which specializes to

('II, <I>}c-(H) : h 1--+ l dt-t(x) 'II(x)<I>(xh). (2.92)

It is easily shown that ('II, 'II}c-(H) is positive in C*(H) (whether or not H is

amenable). Indeed, we may proceed as in the proof of positivity in Theorem 2.5.6.
Generalizing (2.85), we derive

wx«('II, 'II}c-(H) = L L dv(q) II dh 'II(s(q)h)Ux(h)Qx 11 2, (2.93)

where s : Q --* P is an arbitrary measurable section, and we have used III. (2. 139).
Thus one obtains a pre-Hilbert C* -module C~(P) ~ C~(H), which by Corol-
lary 2.1.4 may be completed to a Hilbert C* -module that in analogy with 2.5.6 is
denoted by C(P) ~ C*(H). (When q is a point, the above construction reduces to
2.5.6, so that C(P) = C*(H). However, when P(Q, H, r) = G(Gj H, H, r), the
space C(G) is different from C*(G).)
Applying (2.47) to the case at hand, one derives that for'll, <1>, Z E C~(P) the
b
operator T."',<I> IS gIVen y
C-(H). •

CO(H)
T",,<I> Z(x) = Jp dt-t(y) K",,<I>
( CO(H)
(x, y)Z(y), (2.94)

1 --
with
C*(H)
K",,<I> (x, y):= H dh 'II(xh)<I>(yh). (2.95)

Since the H -action on P is proper, the integrand in (2.95) has compact support in
h. Noting that Kf,~~H) is invariant under the H-action h : (x, y) 1--+ (xh- 1, yh- 1),
one infers that Kf,:~H) lies in C~(P x H P).
In view of the bound (2.51) (from whose derivation one sees that it holds in a
C*(H)
pre-Hilbert C*-module as well), the operator T", <I> may be extended to C(P) by
continuity, and may be defined for all'll, <I> E C(P). In particular, one obtains the
bound (2.12).
Since Kf~H) E C~(P XH P), we may regard T;,~H) as an element of C*(P XH
P). However, in order to identify Co(C(P), C*(H» with C*(P XH P), we need
to show that the norm of T;,~H) in Co(C(P), C*(H» coincides with its norm
III.(3.74) in C*(P XH P). This indeed follows from the isomorphism 111.(3.78).
C*(H) C*(H)
Choosing a section s : Q --* P, one passes from K",,<I> (x, y) to K",,<I>,s (q, h, q')
as indicated in 111.(3.83), and computes its operator norm 1.(1.17) using (2.5) with
~ = C*(H). This computation yields a norm that coincides with the norm in
~O(L2(Q» ® C*(H), proving the claim.
380 IV. Reduction and Induction

Passing to the C* -algebra generated by all T;,~H), Definitions 2.3.1 and III.3.6.3
show that CO(C(P), C*(H» is isomorphic to C*(P XH P). Hence the theorem
follows from Definition 2.4.1. •
It follows from this proof that one has a quantum dual pair
C*(P XH P) ;= C(P) ;= C*(H)
in which 7r R (C*(H» and (, )c.(H) are given by (2.91) and (2.92), respectively,
and the left action 7rL(C*(P XH P» coincides with the representation 7r, given in
III.(3.82). Finally, according to (2.95) and (2.59) one has

('11, <1>)c·(px H P) : (x, y) ~ In dh qs(xh)<1>(yh). (2.96)

These formulae are all understood to be defined on the pre-quantum dual pair
used above, namely C~(P XH P) ;= C~(G) ;= C~(H), and extended to the
completion displayed above by continuity.
In particular, Corollary III.3.7.2 follows from Theorem 2.4.5. Moreover, the
theory of induced group representations of III.2.9 may be reinterpreted in the
light of Theorems 2.7.1 and 2.4.5. To do this, we explain how the construction in
Definitions III.2.8.4 and III.2.9.1 is a special case of Rieffel induction.
We start from Construction 2.2.3, in which we take the Hilbert C* -module & ;=
~ to be C(P) ;= C*(H), defined by (2.76) and (2.77). Furthermore, we take 11. x
to be the carrier space of a representation Ux(H) (or, equivalently, of 7rx (C*(H».
As explained in 2.2, one obtains the same induced space 11. x if one starts from
suitable dense subspaces C(P) and C~(H) of C(P) and C*(H), respectively.
When H is compact, we are in the situation of Proposition 2.5.3. Hence we start
from the pre-Hilbert C*-module L 2 (P) ;= C*(H), and obtain the induced space
11. x as the subspace of all *E L 2 (P) ® 11. x satisfying Pid* *.
= Since U = U R,
this condition is nothing but the equivariance condition III.(2.145). For 'P E Aut(P)
and'll E L 2 (P) we define UL('P)qs by the right-hand side ofIII.(2.161), with qsx
replaced by'll. This defines an operator Ud'P) on L 2(p), which is easily seen to
be adjoin table as a consequence of the fact that elements of Aut(P) by definition
commute with the H-action on P. We identify L 2 (P) ® 11. x with L 2 (P, 11. x ), and
use the description of tr x (A) in 2.5.3. It is then obvious that 7rX(UL('P» coincides
with UX('P) as defined in III.(2.161).
The noncompact case is slightly more involved, since the induced space is no
longer a subspace of L 2(p) ® 11. x ' We are now in the situation of Theorem 2.5.4,
with 11. = L 2 (P),t = C~(P), and U = U R • To identify the reduced space, we use
the method of Proposition 2.2.4. We take 11.~ to be what is, prophetically, called
11. x in Definition III.2.8.4. Consider the map V : C~(P) ® 11. x --1- 11.~ defined by
linear extension of

V'll ® v(x):= In dh qs(xh)Ux(h)v. (2.97)

Note that the equivariance condition I1I.(2.145) is indeed satisfied by the left-hand
side, as follows from the invariance of the Haar measure. Using (2.80) or (2.81),
 2 Induction 381

one verifies the fundamental equation (2.29). It is clear that (;(C~(P) ® 1t x ) is

dense in 1tx , so by Proposition 2.2,4 one obtains an isomorphism U : 1tx ~ 1t!.
Morover, using (2.30), one may repeat the last sentence of the discussion of the
compact case.
Had we not known Definition m.2.8,4 beforehand, it would have been possible
to obtain the form of 1t x on the basis of the method explained after (2.31). For
the image of {; as defined in (2.32) coincides with the image of (; as defined in
(2.97), under the identification of 1t! with a subspace of C(£, 1t x )'13 via the inner
product in L 2(p).
As in the second half of m.2.9, we now specialize to Mackey induction, where
the principal bundle P(Q, H,.) is G(G/H, H, .); cf. m.2.7. As always, the
explicit expressions below apply to the unimodular case.
Corollary 2.7.2. The action C*-algebra C*(G, G/ H) is Morita equivalent to
the group C*-algebra C*(H), with quantum dual pair C*(G, G / H) .= C(G) .=
C*(H). ThispairisobtainedbycompletingC~(GxG/ H).= C~(G).= C~(H),
defined by the maps (2.91), (2.92), and

JrL(f)IJI(x) = L
dy f(xy-l, [X]H )1JI(y); (2.98)

(1JI,4>)c'(G,G/H) : (x, [y]H) 1-+ Ldh lJI(yh)4>(x- 1 yh). (2.99)

Every nondegenerate representation rr ofC*(G, G/ H) is equivalent to an in-

duced representation rr x, realized on 1t x (see Definition III.2.B,4 with P replaced
by G), where Ux is a representation of H. Explicitly, one has

(2.100)

cf. (2.98). The representation rrX(C*(G, G / H» is irreducible iff Ux(H) is

irreducible.
This follows upon combining Theorems 2.7.1 and 2,4.5, including the explicit
construction of induced representations above, with Corollary 111.3.7.5. Equations
(2.98) and (2.99) are derived from the formulae stated after Theorem 2.7.1, spe-
cialized to the case at hand using Proposition 111.3,4.7. Similarly, the stated form
of Jrx follows from 111.(3.84). •
Corollary III.3.7.6 is then seen to be a consequence of Corollaries 2.7.2 and
III.3.7,4, combined with Definition III.3.7.3.
Under a regularity assumption on the group action (which is always satisfied
when G is compact), it is possible to classify the irreducible representations of an ar-
bitrary action C* -algebra. This classification is the quantum version of Proposition
III.3.9,4.
Theorem 2.7.3. Let a smooth action of a Lie group G on a manifold Q be
regular, in that each orbit is (relatively) open in its closure. Then the irreducible
representations ofC*(G, Q) are classified by pairs (0, Ux(H», where 0 is a
382 IV. Reduction and Induction

G -orbit in Q, the group H is the stabilizer Gqo of an arbitrary point qo in 0, and

U x is an irreducible representation of H.
The representation rrfj corresponding to such a pair may be realized on the
Hilbert space 'Hx defined inlll.2.8.4, and is given by (2.100), with [X]H replaced by
xqo. Equivalently, the system ofimprimitivity (U(G), if X (Co(Q))) corresponding
to the representation rrX(C*(G, Q» by Corollary 1ll.3.7.4 is given by U = ux
(cf. Ill. (2. 175)) and

(2.101)

We start from a given irreducible system of imprimitivity (U(G), if(Co(Q») on

some Hilbert space 'H. Since G is a Lie group and Q is a manifold, it can be shown
that C*(G, Q), and therefore 'H are separable. Using the spectral theorem applied
to if (Co(Q», one can bring 'H into the form

(2.102)

where the Vi are mutually singular Borel measures, and the 'Hi are multiplicity
spaces. The representation if is then given in terms of multiplication operators in
the obvious way. The fact that if can be extended to a system of imprimitivity
easily implies that each measure Vi is quasi-invariant under G (that is, its measure
class is G-invariant).
A measure V on a G-space Q is said to be ergodic when a G-invariant Borel
function is constant almost everywhere. Equivalently, for a G-invariantBorel set B
either its complement Q\B or B itself must have measure zero. It is quite obvious
that (U, if) can be irreducible only when there is a single term in (2.102) in which
the measure V is ergodic; for otherwise one could decompose the carrier space as
L2(Q, v) = L2(Q\B, v) ED L2(B, v).
Let S be the smallest closed set in Q for which v(Q\S) = 0; this set is auto-
matically G-invariant, and we can write L 2(S, v) instead of L2(Q, v). Suppose S
contains two orbits Gql i- Gq2. First assume that [qllG (i.e., the closure of Gql
regarded as a point [qdG in Q/G) does not contain [q2]G. Then rQ~Q/G([qdG)'
which is a closed G-invariant subset of S, does not contain Gq2. This contradicts
either ergodicity or the definition of S. Hence [q2]G E [qIlG' We combine this
inclusion with the regularity assumption. Accordingly, there is an open set N in S
that contains Gq, (the closure of Gql in Q, hence in S) but is disjoint from Gq2'
The G-translate GN of N has the same properties, so that its complement S\ GN
is a closed G-invariant subset of S, which does not contain Gq I. This again leads
to a contradiction with either ergodicity or the definition of S. Hence v must be
concentrated on a single orbit in Q.
Furthermore, the regularity assumption turns out to be equivalent to the statement
that each orbit (equipped with the topology inherited from Q) is homeomorphic to
G /G qo ' Hence if(Co(Q» ~ Co(Gqo) ~ Co(G / G qo )' so that the situation reduces
to the representation theory of the action C*-algebra C*(G, G / H), with H = G qo •
The theorem now follows from the second half of Corollary 2.7.2. D
 2 Induction 383

2.8 Covariant Quantization

Let us return to quantization theory, and ask in some generality what happens to
Berezin quantization in the presence of a classical symmetry group. The following
notion, which generalizes Definition III.3.7.3, is natural in this context.

Definition 2.8.1. Given a smooth G-action on a manifold X, a generalized sys-

tem of imprimitivity of G on X in a Hilbert space 11. is a pair (U, Q) in which
U is a representation of G on 11., and Q : Co(X) -+ ~(11.) is a positive map,
satisfying the covariance condition

U(x)Q(j)U(x)* = Q(aAi», (2.103)

where ax(j) : y H- i(x-1y). The system is called transitive when X = G/ H

with the natural G-action.

When X is compact, or when Q may be extended to CO(X)I, this condition may

be equivalently stated in terms of the POVM t!. H- A(t!.) associated to Q (cf.
11.1.4.8) as

U(X)A(t!.)U(X)-1 = A(xt!.). (2.104)

Every (ordinary) system of imprimitivity is evidently a generalized one as well,

since a representation is a particular example of a positive map. A class of exam-
ples of truly generalized transitive systems of imprimitivity arises as follows. Let
(U(G), it (Co(G/ H») be a system ofimprimitivity on a Hilbert space JC, and sup-
pose that U(G) is reducible. Pick a projection p in the commutant of U(G); then
(pU(G), pit p) is a generalized system of imprimitivity on 11. = pJC. Of course,
(U, it) is described by Corollary III.3.7.6, and must be of the form (UX, itX). Under
an innocent technical assumption, this class turns out to exhaust all possibilities.
What follows generalizes Corollary III.3. 7.6 (or the second part of Corollary 2.7.2)
to the case where the representation it is replaced by a positive map Q.

Theorem 2.8.2. Let (U(G), Q(Co(G / H))) be a transitive generalized system of

imprimitivity on 11., where Q may be extended to the unitization ofCo(G / H) such
that the extension preserves the unit.
There exists a representation Ux(H), with corresponding induced representa-
tion UX (G) on 11. x and system ofimprimitivity (UX, itX) as described in Corollary
111.3.7.6, and a projection p on 11. x in the commutant of UX(G), such that
(pUX(G), pit x p) and (U(G), Q(Co(G/ H») are equivalent.

We apply Theorem 11.1.4.2. To avoid confusion, we denote the Hilbert space

11. x and the representation]f x in Construction 11.1.4.3 by i{x and it x, respectively;
the space defined in III.2.8.4 and the induced representation of III.(2.175) will still
be called 11. x and ]fx, as in the formulation of the theorem above. Indeed, our
goal is to show that (it X, i{X) may be identified with (]fx, 11.X). We identify ~ in
11.1.4.2 and 11.1.4.3 with ~(11.), where 1t is specified in 2.8.2; we therefore omit
the representation ]fx occurring in 11.1.4.2 etc., putting 1t x = 1t. For x E G we
384 IV. Reduction and Induction

define a linear map U(x) on Co( G / H) ® 'li by linear extension of

U(x)f ® \II := (Xx (f) ® U(x)\II. (2.105)
Since (Xx 0 (Xy = (Xxy, and U is a representation, U is clearly a G-action. Using the
covariance condition (2.103) and the unitarity of U(x), one verifies that
(U(x)f ® \II, U(x)g ® <I»~ = (f ® \II, g ® <I»~, (2.106)

where (, )~ is defined in 11.( 1.25). Hence U(G) quotients to a representation Ux (G)

on ilx. One checks that (UX, fiX) is a system ofimprimitivity on il x (compute on
Co( G / H) ® 'li and then pass to the quotient). By Corollary 111.3.7.6, this system
must be of the form (UX, fiX) (up to equivalence).
Finally, the projection p defined in 11.1.4.2 commutes with all UX(x). This is
verified from 11.(1.29), 11.(1.30), and (2.103). The claim follows. •
The power of this result is clear in the following application. Let a phase space
S be a coadjoint orbit in g*, so that S ~ G / H, equipped with the Lie symplectic
structure (see3.1 for examples). A pure state quantization {q/i, 'li/i, IL/i} (cf. 11.1.3.3)
of S leads to an associated Berezin quantization Qg. This quantization is called
covariant when each 'li/i carries an irreducible representation U/i(G), such that
(2.107)
for all f E Co(G/H) and x E G, where (XO is defined by 111.(1.158). We have
already seen examples of this in 111.1.11. Theorem 2.8.2 then implies
Corollary 2.8.3. The representation U/i(G) occurring in a covariant Berezin
quantization of Co(G / H) must be an irreducible sub representation of an induced
representation U x (G). When p is the corresponding projection on 'li/i C 'lix, the
Berezin quantization Qg must have the form Qg(f) = pfix(f)p, where
(2.108)
Cf. III.(3.86). In particular, if 'lix = L 2( G / H) (i.e., one has induced from
the trivial representation of H), then fiX (f) is simply f, seen as a multiplication
operator, and its Berezin quantization consists in squeezing this operator into a
subspace.
In view of (2.104) and the discussion following 11.1.4.9, Theorem 2.8.2 and
its Corollary 2.8.3 describe covariant localization in phase space. There is a
corresponding theory of covariant localization in configuration space, which
we now briefly discuss. To set the stage, we start on Euclidean space Q = 1R3 •
In elementary quantum mechanics, a particle moving on 1R3 with spin j E N is
described by the Hilbert space
'li j = L \1R3 ) ® 'li j , (2.109)
where'lij = C 2 j+1 carries the irreducible representation Uj(SO(3». The basic
physical observables are represented by the unbounded operators Q% (position),
pl (momentum), and Jl
(angular momentum), where k = 1, 2, 3. These operators
 2 Induction 385

satisfy the commutation relations (say, on the domain S(R.3) ® Hi)

[Q%, Q!] = 0; (2.110)
[Pt, Qt] = -iMkl; (2.111)
[Jt, Qt] = iMklmQ!; (2.112)
[Pt, p/] = 0; (2.113)
[Jt, J/] = iMklmJ~; (2.114)
[Jt, p/] = iMklmP~, (2.115)
justifying their physical interpretation.
The momentum and angular momentum operators are best defined in terms of
a unitary representation U i of the Euclidean group E(3) = SO(3) ~ R3 on Hi,
given by
(2.116)
In terms of the standard generators Pk and Tk of R3 and SO(3), respectively,
one then has pt = ihdUi(Pk) and Jt = ihdUi(Tk); the commutation relations
(2.113)-(2.115). follow from III.(1.70) and (1.65).
Moreover, we define a representation fri of CoOR,3) on Hi by
(2.117)
where j is seen as a multiplication operator on L 2(R3). The associated PVM
/:l. ~ E(ll.) on R3 in Hi (see 11.1.4.8) is E(/:l.) = XI!. ® Hio in terms of which the
position operators are given by Q% = JR.] dE(x)xk; cf. the spectral theorem for
unbounded operators. Equation (2.110) then reflects the commutativity of Co(R3),
as well as the fact that fr i is a representation.
Identifying Q =]R.3 with Gj H = E(3)jSO(3) in the obvious way, one checks
that the canonical left action of E(3) on E(3)jSO(3) is identified with its defining
action onR3. It is not hard to then verify from (2.116) that (U i (E(3», fri(Co(R3»)
is a system of imprimitivity as defined in III.3.7.3. The commutation relations
(2.111), (2.112) are a consequence of the covariance relation III.(3.38).
Rather than using the unbounded operators Q%, pt, and JkS and their commuta-
tion relations, we therefore state the situation in terms of (U i (E(3», fri (CO(R 3»).
Such a pair, or, equivalently, a nondegenerate representation rri of the action C*-
algebra C*(E(3), R 3) (cf. III.3.7.4), then by definition describes a quantum system
that is localizable in R3 and covariant under the defining action of E (3). It is natural
to require that rri be irreducible, in which case the quantum system itself is said
to be irreducible.
Proposition 2.8.4. An irreducible quantum system that is localizable in R3 and
covariant under E(3) is completely characterized by its spin j E N. The corre-
sponding system oJimprimitivity (U i (E(3», fri(Co(R3))) is equivalent to the one
described by (2.109), (2.116), and (2.117).
This follows from Corollary III.3.7.6. The induced representation (E(3» ul
defined by the section s : R3 ~ E(3) given by seq) := (e, q) (see III.2.9)
386 IV. Reduction and Induction

is precisely (2.116); cf. Lemma 3.3.2 below. The corresponding representation

111.(3.86) of CO(R.3) is (2.117). •

This is a neat explanation of spin in quantum mechanics, though it would be a

mistake to think that spin has no classical counterpart.
Generalizing this approach to an arbitrary homogeneous configuration space
Q = G I H , a nondegenerate representation 1T of C* (G, G I H) on a Hilbert space
1t describes a quantum system that is localizable in G I H and covariant under
the canonical action of G on GI H. By Corollary 111.3.7.4 this is equivalent to a
system of imprimitivity (U(G), ir(Co(GI H») on 1t, and by Proposition 11.1.4.8
one may instead assume that one has a PVM I:!.. Ho E (I:!..) on G I H in 1t and a
unitary representation U(G) that satisfy
U(X)E(I:!..)U(X)-l = E(xl:!..) (2.118)

for all x E G and I:!.. E ~; cf. (2.104). The physical interpretation of the PVM
is given by 11.(1.36); the operators defined in 111.(3.39) play the role of quantized
momentum observables. Generalizing Proposition 2.8.4, we have

Theorem 2.8.5. An irreducible quantum system that is localizable in Q = G I H

and covariant under the canonical action of G is completely characterized by an
element X E iI ofthe unitary dual ofH. The corresponding system ofimprimitivity
(UX(G), irj(Co(GI H») is equivalent to the one described by III.2.B.4, III. (2. 175 ),
and 1II.(3.B6).
This is immediate from Corollary 111.3.7.6.

With an analogous notion of classical localization, the corresponding situation

•
in classical mechanics is described by Theorem 1.6.4. Thus the different possibil-
ities allowed by Theorem 2.8.5 do not correspond to inequivalent quantizations
of the classical phase space T*(G I H) (for which notion cf. 3.9), but rather to
quantizations of various symplectic leaves in (T*G)I H. For by Theorem 1.5.5 the
latter coincide with the spaces defined in (1.43), which in tum carry all irreducible
transitive classical systems of imprimitivity for G on G I H (up to possible covering
spaces).
It is suggested by 111.1.11 and 1II.2.10 that for n E N the Hilbert space 1tnx
carrying a system ofimprimitivity (UX(G), irX(Co(GI H») should be seen as the
quantization of the symplectic1eaf(T7'GiJx of(T*G)1 H for the value Ii = lin of
Planck's constant; here X is an integral weight. In particular, for fixed Ii = 1, differ-
ent values of X correspond to the quantization of different symplectic manifolds.
The phase space T*( G I H) is quantized by L 2( G I H); none of the nontrivially
induced representations should be seen as the quantization of T*(G I H).

2.9 The Quantization of Constrained Systems

We will now look at Rieffel induction as a quantum analogue of the classical special
symplectic reduction procedure of 1.2. For the purpose of comparing the two, the
 2 Induction 387

second approach in 2.2 to Rieffel induction is more appropriate; see Construction

2.2.3.
Firstly, ~lR should be regarded as the quantization of the Poisson algebra
e
COO(P, R), whereas the Hilbert ~-module is the simultaneous quantum analogue
of the Poisson map J : S ~ P- and its pullback J* : COO(P, R) ~ COO(S, R).
The representation 1fx (~) on 1t x is the quantum counterpart of the Poisson map
Jp : Sp ~ P; under favorable conditions, the Hilbert space 1t x may even be
thought of as the quantization of the symplectic manifold Sp. Quotienting e ® 1tx
by the null space N x ' which is the decisive step in the construction of the induced
Hilbert space 1t x , is clearly the analogue of quotienting by the null foliation <I> in
(1.l3).
Recall Definition 1.1.8. A suitable quantum analogue is given by
Definition 2.9.1. A weak quantum observable is a linear map A on e ® 1t x
e
that for all q, , <i> E ® 1t x satisfies
- -x - -x
(A \II, <1»0 = (\II, A <1»0 ; (2.119)
- -x 2--x
(A\II, A\II)o :5 IIAII (\II, \11)0' (2.120)

The collection of all weak quantum observables is called m~.

Using (2.23), we see that (2.119) implies that ANx £ N x . Hence we may
define 1f X (A)Vq, := VA q" generalizing (2.25). Because of (2.120), the analogue
of (2.19) still holds.
e
If we weaken 2.9.1 to allow for operators defined on a dense domain ® 1t x ,
thereby giving up (2.120), the operator 1fX(.4) may be unbounded, with (initial)
domain V(e ® 1t x ). In view of (2.24), condition (2.119) guarantees that 1fX(A)
is then symmetric on this domain. Moreover, the vector space m~ (defined with
or without (2.120» is easily seen to be a Jordan-Lie algebra under the operations
1.(1.22).
The quantum version of (1.15) is the inclusion
(2.121)
where C*(e, 23) is identified with C*(e, 23) ® Hx' This follows, because (2.11)
and (2.21) guarantee that a self-adjoint element A of C*(e,~) satisfies (2.119).
Finally, (2.25) is evidently the quantum analogue of (1.16).
It is instructive to reconsider the example e = ro(HX) (given after 2.2.2) in the
light of the analogy with special symplectic reduction. In Definition 1.2.1 we take
S = CFp)x, which is defined as in (1.30), in which we replace Sp by Sx to avoid
notational ambiguity, and P = Q (with zero Poisson structure). Regarding S as
a bundle over T*Q (cf. Theorem 1.6.1), and therefore over Q, the Poisson map
J : S ~ Q is the bundle projection. Finally, we reduce from Sp = q, seen as a
symplectic leaf in Q, and p the inclusion map. To compute the reduced space we
look locally: By Theorem 1.6.1 we locally have S ~ T* Q x SX' The constraint
hypersurface S *Q Sp is the fiber J-I(q) of S above q. The null distribution N is
tangent to the fibers of T* Q, so that finally the reduced space is sq ~ Sx.
388 IV. Reduction and Induction

However, mesmerized by the analogy between special symplectic reduction

and Rieffel induction, one should not overlook a cardinal difference between the
classical and the quantum reduction procedure. In the classical case one had to
restrict the full space S x Sp to the constraint hypersurface S *p Sp before taking
the quotient by the null foliation; the latter would, indeed, not even be defined
without this restriction. In the quantum case, on the other hand, one passes straight
from £ ® 1t x to the quotient £ ® 1t xlN x . There is no analogue of the constraint
hypersurface; in other words, in quantum mechanics there is no need to impose
the constraints. (This comment applies only to the case where all constraints are
first class, which is appropriate here in view of Theorem 1.2.2 and the comment
preceding 1.1.8.)
In an effort to better mimic the classical procedure, one may introduce an ad-
ditional step in Rieffel induction, in which one at the very beginning passes from
£ ® 1t x to £ ®Q3 1t x . Here the tensor product over IB means that £ ®Q31t x consists of
equivalence classes [\II ® v h under the equivalence relation \II B ® v ~ \II ®n x(B)v
for all B E lB. In other words, this new first step would consist in dividing £ ® 1t x
by the vector space generated by all expressions ofthe form \II B®v - \II®n x(B)v.
Although there is a notational similarity between S *p Sp and £ ®Q31t x , this step
still amounts to taking a quotient rather than imposing the constraints in some way
or another. More importantly, this first step is entirely unnecessary, because all
vectors \liB ® v - \II ® n x (B)v lie in the null space N x , as a simple computation
shows. Hence they automatically disappear in the construction of 1tx.
In physics one would interpret the above discussion in the light of the quan-
tization of constrained systems. Here "quantization" may refer to any procedure
deserving that name (such as the methods of strict or pure state quantization dis-
cussed in this book). The problem is to quantize the symplectic manifold SC
defined by (1.2), given that it has been obtained by symplectic reduction from S.
One could, of course, try to quantize SC without this knowledge, but in case it is
simpler and less problematic to quantize S, it makes sense to try to quantize S first,
and then use some method that mimics symplectic reduction in order to construct
what should be the quantization of SC. See Figure 15.
For the purpose of this discussion we assume that C is coisotropically embedded
in S; in other words, that all constraints are first class (we also put Ii = 1). The
traditional approach is to quantize S as if there were no constraints. This yields

reduction
S SC

quantization
j j quanH",Hon

induction
Q(S) Q(Sc)

FIGURE 15. Constrained quantization: Q is some quantization procedure

 2 Induction 389

a Hilbert space Q(S) = H and some quantization Q(COO(S, JR.» of the classical
observables as operators on H. In particular, the constraints C{Ji are quantized by
operators Q(C{Ji), which are usually unbounded and symmetric on some common
dense domain. For example, when C is defined by Marsden-Weinstein reduction at
zero, that is, when C = 1 -I (0) for some strongly Hamiltonian H -action on S with
equivariant momentum map 1 : S -* f)~, the constraints C{Ji are the components
1i , i = 1, ... , dim(H). In that case one hopes that Q(Ji) = idU(T;), where U is
a representation of H on H; cf. 111.(1.128) and III.(2.167).
The Dirac method of quantizing systems with first-class constraints now
consists in defining the "physical state space" Q(Sc) = HD as
(2.122)
When f E COO(S, JR) is a weak observable, it is hoped that [Q(f), Q(C{Ji)] is
proportional to a linear combination of the Q(C{Jj); cf. 1.1.9.2. If so, Q(f) leaves HD
stable, so that one may define the "physical observable" QC (f) as the restriction
of Q(f) to 'H D ·
The Dirac method attempts to quantize the first step of classical reduction,
namely imposing the constraints. The second step of quotienting by the null fo-
liation, i.e., passing from C to C / ct>c, has no counterpart. Compare this with the
opposite state of affairs in the method based on Rieffel induction.
In the case C = 1- 1(0) with H compact, the Dirac method is successful. The
condition dU(T;)\II = 0 for all i (where the T; form a basis of f) is equivalent to
U(h)\II = \II for all h E HO (the component of e in H); in other words, the physical
state space HD is the subspace Ho of'li that transforms trivially under HO. When
Q(f) commutes with all U (h), the physical observable QC (f) is the restriction
of Q(f) to H o· Indeed, we see from Proposition 2.5.3 that Ho coincides with
the Hilbert space HO obtained by Rieffel induction from the trivial representation
Uid of H on 'H.d = C, and that the physical observable QC (f) is nothing but the
induced representative Jrid(Q(f».
However, when the quantum constraints Q( C{J j) fail to have zero in their discrete
spectrum or fail to have a joint eigenvector for this value, the space HD is empty.
For Marsden-Weinstein reduction at zero this habitually happens when H is not
compact. Examples are provided by Theorem 2.5.4. However, probably the sim-
plest example is S = T*JRn, with its standard symplectic structure, equipped with
the constraint PI = O. Analogously to Example III.I.2.11, this constraint is the
momentum map for the JR-action given by

a..( PI, P2, ... , q I ,q 2 , ... ) f-+ ( PI, P2, ... , q I + a, q 2 )

, ....
It is easy to show that Marsden-Weinstein reduction at zero leads to the re-
duced phase space SC :::::: T*JRn-l. All functions of (P2, ... , q2, ... ) are weak
observables, their reduced action on SC being the obvious one.
Quantizing S by the Hilbert space H = L2(JRn), with Q(PI) = -i8/8x l , it is
clear that the spectrum of the quantized constraint is absolutely continuous, and
equal to JR. Hence HD = {O}, which is not the desired quantization L 2(JRn-l) of
SC.
390 IV. Reduction and Induction

The method based on quantum induction handles this problem quite effortlessly.
In the context of Theorem 2.5.4 we take 1i = L 2 (Rn ), as in the Dirac method. The
representation U(R) is taken to be
U(a)\II(x', x 2 , ••• ) := \II(x' - a, x 2 , ••• ); (2.123)
this representation satisfies Q( h) = i d U (T) = Q(p,), where T is the standard
generator of R. Choosing £ = C~(Rn), one verifies the properties required in
2.5.4. We now induce from the trivial representation Uid ofR, so that 1ix = 1iid =
C. Hence

(\II,<I»~= {da ( d n x\ll(x'-a,x 2 , ... )<I>(x',x 2 , ... )

JR JR"
A A 2
= \11(0, x 2 , ••• )<1>(0, x , ... ), (2.124)
where the Fourier transform (indicated by the hat) is taken only in the first variable.
To identify the induced space 1iid we use Proposition 2.2.4. Our guess is 1i~ =
L 2(Rn-'), consisting of the square-integrable functions of (x 2 , ••• ). This is proved
by constructing (; : £ ~ L 2(Rn-' ) as
- 2 2
U\II(x , ... ):= \11(0, x , ...). (2.125)
A

It is evident from (2.124) that (2.29) is satisfied.

We identify L2(Rn) with L2(R) ® L 2(R n-') in the well-known way. Bounded
operators A on L2(Rn) that do not act on the first variable in \II have the form
A = 1I, ® A2, where 1I, is the unit operator on L 2(R), and A2 E ~(L2(Rn-'».
Such operators lie in C*(£, ~) (cf. the text following (2.34», and may be Rieffel-
induced to bounded operators Jrid(A) on 1-{'d. One easily sees from (2.30) thatJrid(A)
is simply A2.
Thus we have a satisfactory quantization of the classical situation, which, de-
spite its simplicity, is not so easily amenable to treatment by other constrained
quantization techniques.

2.10 Quantization of Singular Reduction

Basing constrained quantization on Rieffel induction allows one to quantize sin-
gular Marsden-Weinstein quotients. It is interesting to see how the singularities
in the reduced space, and particularly its decomposition into symplectic pieces, is
reflected in the quantum theory. In the absence of a general theory of this reflection,
we approach this matter through specific examples.
Consider the case S = T*R2 and H = SO(2) discussed in 1.11. A suitable
quantization of S is given by the Hilbert space 1i = L2(R2), carrying the rep-
resentation of SO(2) given by U(h)\II(q) := \II(h-'q), in terms of the defining
action of SO(2)onR2. Mapping L2(R2) to L2(R+, r dr)®L 2(SO(2» in the usual
way, the representation U is the tensor product of the unit 1I on L2(R+, r dr) and
the left-regular representation UL on the second factor. We then induce from the
trivial representation Uid of SO(2), reflecting the zero in SO = J-'(O)/ H. Ac-
cording to Proposition 2.5.3, the induced space 1iid is L 2(R+, r dr). Adjointable
 2 Induction 391

operators on 1£ are necessarily SO(2)-invariant, so that they are linear combina-

tions of operators of the form A = A I ® A2, where A2 commutes with U L. The
induced representation is then given by linear extension of

(2.126)

The identification of l£ id with L 2(JR+ , r dr) is a pleasant result, since we have seen
that the reduced space SO was T*JR.+. On the other hand, the origin 0 is a null set
with respect to the measure r dr, so that at first sight the quantum theory contains
no analogue of the lower-dimensional symplectic piece of So.
However, the singular structure of the classical reduced phase space is reflected
in the domains of unbounded observables. It is remarkable that unbounded oper-
ators, notably differential operators, of the form ;rid(A) on l£ id that would not be
essentially self-adjoint on the natural domain C~(JR+\{O}) are often essentially
self-adjoint on the domain Vx C~(JR2) inherited from 1£ (where Vx is defined be-
low (2.24». The domain of self-adjointness will then generically be such that wave
functions in this domain vanish near the origin, providing a quantum analogue of
the fact that motion in the highest-dimensional (and, indeed, in any) symplectic
piece of SO cannot cross the barrier to the lower-dimension piece(s).
For example, the Hamiltonian H = -~ + V(r) on L2(JR2) is SO(2)-invariant
and essentially self-adjoint on C~(JR2). The unitary map U : L2(JR+, rdr) --+
L 2(JR+, dr), defined by UqJ(r) := .jTqJ(r), transforms ;rid(H) into

* d2 1
d
U;r' (H)U = - dr2 - 4r2 + V(r). (2.127)

While the analysis of this expression is quite straightforward for any reasonable
potential V, the free case V = 0 already suffices to illustrate the main point.
Defined on C~(JR+\{O}), the operator (2.127) then has deficiency indices (1, 1),
so that it is not essentially self-adjoint. However, defined on Vx C~(JR2), which
consists offunctions of the type qJ (r) = .jTf (r2) with f E C~(JRt), the operator
in question is indeed essentially self-adjoint. The closure of the latter operator is
an extension of the closure of the former, to whose domain one adds functions
of the indicated type in order to achieve essential self-adjointness. The boundary
condition qJ(O) = 0 corresponds to a hard wall potential at the origin.
We now turn to the quantization of the example at the end of 1.11. As above,

Fourier transform qJ 1-+ *,

the quantization of S = T*JR 2 is taken to be 1£ = L 2(JR2), of which we take the
so that we work in p-space. The JR-action (1.90) on S
is quantized by the representation (; (JR), given by

(2.128)

This is motivated by the fact that the generator idU(T) (where T is the standard
generator of JR) on 1i is

' - i dU (T) -- -21

Q(<p) .-
(a 2
-- + -
-a(xl)2 a-
a(x2)2
2
)
, (2.129)
392 IV. Reduction and Induction

which is the quantization of the single classical constraint rp = J in the SchrOdinger

representation; cf. (1.91).
It is possible now to follow the procedure of Theorem 2.5.4 with H = R, with
a suitable replacement of C~(H). However, it is easier to approach this situation
using Fell induction rather than Rieffel induction. Hence we apply Construction
.c
2.2.5, in whichE c 2(R) is taken to consist of those 4t E C~(R2) for which there
exists an E > 0 (depending on \II) such that 4t = O(lpn for Ipl := Jpi + p~ ~
O. The motivation for this choice of E will emerge shortly. Since the classical
reduced phase space is ]-I(O)/R, we put 'Hx = 'Hid = C in 2.2.5, so that also
~('Hx) = C. With v = w = 1 in (2.37), the form (\II, <1»& then coincides with
(\II, <I>)c. Proceeding as if we were performing Rieffel induction from the trivial
representation ofllt, we define (\II, <I»~ by (2.80) with Uid(h) = 1. Using (2.128),
this yields

(2.l30)

defined as an oscillatory integral. This can be computed as such, resulting in

(\II, <I»~ = 1
IR
-dk- ~A
\II(k, k)<I>(k, k)
211' Ikl
+ \II(k, -k)<I>(k, -k)
A A ]
. (2.l31)

This expression is well-defined by virtue of our choice of E.

Using the method of Proposition 2.2.4, the induced space 'Hid may be identified
with
'H~:= L2(R,dk/211'IkD®C2. (2.l32)
For we may define fj : E ~ 'H~ by
fj\ll(k) := (f/, \II)+(fk-' \II), (2.l33)
where the functions f k± E c oo (R2) are defined for k E R by
f k± (x 1, x 2 ) := eik (xl'FX2) , (2.l34)
and the pairing on the right-hand side of (2.133) is defined as if \II and f k± both
were in L2(R2). The right-hand side of (2.133) equals 4t(k, k)+4t(k, -k), so that
one easily verifies (2.29) from (2.l31). (Equation (2.125) may be rewritten in a
similar way, since the right-hand side equals (fx2 .... , \II), where fx2 .. ..<;I,;2, ... )
is 8(x 2 - ;2, ... ), but the gain in doing so is small.)
This way of writing the map fj is of fundamental importance, as the f k± form the
complete set of linearly independent solutions of the quantum constraint equation
Q(rp) f = 0; cf. (2.129). These solutions do not lie in L 2 , so that the Dirac method
fails in this example, but we see that the solutions of the quantized constraint
equations still playa formal role in the quantization procedure.
An operator A that commutes with all U(t) is adjointable. Now suppose that
A : E ~ E satisfies the stronger condition of commuting with Q(rp). When the
action of A may be extended to an action on each fk± , it follows that Aft is again a
 3 Applications in Relativistic Quantum Theory 393

solution of the quantized constraint equations, so that it may be expanded in terms

of the solutions ik±' The induced representation Jrid(A) may then be expressed in
terms of the expansion coefficients. To make this argument precise, suppose that
A as above is such that (ik, A\II) = (A* ik, \II), where A* ik := (A(k, .), f.)*, for
some kernel A. Here ik is regarded as a 2-component vector, so that each A(k, k')
is a 2 x 2 matrix, the inner product (, )* being in L 2(R, dkj2Jr Ikl) ® (:2. It then
follows from (2.30) and (2.133) that
Jr!d(A)\IIid(k) = (A(k, .), \IIid)*. (2.135)
We now examine the question to what extent the quantum theory reflects the
decomposition of the classical reduced phase space SO into its symplectic pieces;
see the final paragraph of 1.11. Firstly, as in the previous example, the lowest-
dimensional piece PI = P2 = 0 does not occur in the quantum theory. Secondly,
we have seen that smooth functions on SO have to descend from smooth functions
on S that depend only on PI, P2, and q I P2 +q 2 p, . In the SchrOdinger representation
on L2(R2) (position space) these are quantized by
Q(Pl) = -i8j8xl;
Q(P2) = -iajax 2;
Q(ql P2 + q2p,) = -i(x I 8j8x 2 + x 28j8x'), (2.136)
respectively. The induced action on 1t~ is then given by
Jr!d(Q(pd) = diag (k, k);
Jr!d(Q(P2» = diag (k, -k);

Jr!d(Q(ql P2 + q2 PI» = -ik diag (:k' - :k) . (2.137)

Here k is seen as a multiplication operator on L 2(R, dk j2Jr Ik I), and the diagonal is
meant to be of a 2 x 2 matrix. The operators in (2.137) are defined and essentially
self-adjoint on the space of C~-functions in 1t~.It is obvious from these expres-
sions that the four subspaces L2(R+, dkj2Jrlkl) ® ej, L 2(R-, dkj2JrlkD ® ej,
where i = 1,2, of L2(R, dkj2Jrlkl) ® (:2 do not mix under the action of these
operators. More precisely, each of these spaces is stable under the group generated
by the Lie algebra spanned by the operators in (2.137). Each of these four sectors
is plainly the quantum counterpart of the appropriate symplectic piece in SO.

3 Applications in Relativistic Quantum Theory

3.1 Coadjoint Orbits of the Poincare Group
We will now apply the formalism of 1.10 to the case that G is the Poincare group.
Assuming that the reader is familiar with the theory of special relativity, we recall
some notation. Minkowski space M is R4 equipped with the metric tensor ~ : =
diag(l, -1, -1, -1); this tensor is used to raise and lower indices as explained
394 IV. Reduction and Induction

in 11.3.2. Greek indices run from 0 to 3, whereas Latin ones go from 1 to 3. The
pairing between M and its dual M* is pv := p(v) := P/Lv/L = pov o + PiVi,
but p2 := gM(P, p) = P5 - p2, where p2 := PiPi, etc. In general, p stands for
(PI, P2, P3).
Furthermore, 0(1, 3) is the subgroup of G L( 4, JR.) consisting of elements that
leave g,M invariant. For convenience of notation, the Lorentz group L is defined
as the connected component of 0(3, 1) containing the identity (this subgroup of
0(3, 1) is often called Lt). The Lie algebra ( of the Lorentz group consists of
those real 4 x 4 matrices M~ for which the components M/Lv are antisymmetric.
The dual (* is identified with ( under the pairing M(N) := M/Lv N /LV' This is
useful, because P /\ V defined in (1.67) is now the matrix with entries

(3.1)

Moreover, the coadjoint action of L is given by Co(M) : N f-+ M N MT.

A convenient basis of [ is {Ji, Bd, where Ji := €ijkEjk and Bi := EOi -
Ejo; here E/Lv is the matrix with entry 1 at position jLV, and 0 elsewhere. The
commutation relations are [Jj, h] = €jjkJko [Bi, Bj] = -€jjkJk, and [Ji , Bj ] =
€ijk Bk· Physically, the J j generate rotations, whereas the Bi generate boosts. Hence
( has the reductive decomposition

(= .50(3) EEl b, (3.2)

in which .50(3) and b are the linear spans of the Ji and of the B i , respectively.
The Poincare group is P = L~pM,inwhichpisthedefiningactionp(A)q/L =
Aq/L := All;,qV of L c 0(3, 1) on M. The dual action is p*(A)pl-' = API-' =
A; PV' We will often omit the symbols p and p*.
In principle, each coadjoint orbit of P is the "covariant" phase space of some
relativistic particle. As we have seen in Proposition 1.10.1, the first step in the clas-
sification of these orbits is the study of the p*(L)-orbits in M*. The classification
of the latter is well known: The orbit types are
P'lL ._ P'lL -
\../0 . - \../(0,0.0,0) -
(OJ',

O~2.± := ot±m,O,o.O) = {p E M* I p2 = m 2, ±po > O} ~ L/SO(3);

o~± := Ot±I.O.O,-1) = {p E M* I p2 = 0, ±po > O} ~ L/ E(2);
O~m2 := O~.O.O.m) = {p E M* I p2 = _m 2} ~ L/SO(I, 2), (3.3)

where m > O. Here SO(3) is a subgroup of L in the obvious way (which corre-
sponds to the decomposition (3.2». The embedding in L of the Euclidean group
E(2) := SO(2) ~p JR.2 in dimension 2 is specified by looking at its generators: A
basis {T;} of E (2) is obtained by putting

T 1 := BI - J2 ;
T2 := B2 + JI;
T3:= h. (3.4)
 3 Applications in Relativistic Quantum Theory 395

This time [ has the decomposition

[= e(2) (£) m, (3.5)
where m is the linear span of T4 := B, + h, T5 := B2 - J" and T6 := B3. This
decomposition fails to be reductive, but on the other hand m is a Lie subalgebra of
[, unlike the subspace b in the massive case (3.2).
To further describe the data called for in 1.10.1, one should classify the coadjoint
orbits of SO(3), £(2), and SO(1, 2). The P-orbits related to the L-orbit O~m2
describe tachyonic particles, which are believed not to exist in nature. In what
follows, we will therefore look only at the coadjoint orbits that are related to the
L-orbits O~2.± and O~.±; the vacuum O~ is relevant to (quantum) field theory, but
not to particle dynamics.
Proposition 3.1.1. The physically relevant coadjoint orbits ofthe Poincare group
are as follows .
• One family, O~.±.s' related to the p*(L)-orbits O~2.± with m > 0, is further
labeled by a parameter s ::: O. Each value of the spin s labels a coadjoint orbit
s1of SO(3). One has dijfeomorphisms

v m.±.s "-'
",.,p - T*1Ill3
~ X
S2
s' (3.6)

• A secondfamily Ot,±.s' related to the p*(L)-orbits O~.±, is further labeled by a

parameter hER Each value of the helicity h labels a coadjoint orbit (0,0, h)
of E(2). There are difJeomorphisms
Ot,±.s ~ T*(L/ E(2». (3.7)
We start with the massive case; the parameter m is the mass of the particle
living on any of the P-orbits to be described. The stabilizer of p = (±m, 0, 0, 0)
is Lp = SO(3). As we have seen in 1.10, the coadjoint orbits of SO(3) are the
two-spheres S;with radius r; the case r = 0 is included here as the origin. By
Proposition 1.10.1 the orbit type of the P -orbit 0MP.. _depends on the restriction of
.p
Nt to -50(3) ~ JR.3. This restriction passes through a unique SO(3) co adjoint orbit
S;, and we correspondingly label the P-orbit as O~.±.s'
The simplest case is s = 0; the orbit O~.±.o is the one through (0, p) E p*.
We infer from (1.66) and (3.1) that the stability group of the above point in this
orbit is SO(3) x JR., where JR. c M is the zeroth copy JR.eo. By (1.80) we then
simply have O~.±.o ~ T*(L/ SO(3». The space L/ SO(3), and therefore the orbit
O~2.±' is diffeomorphic to JR.3, because SO(3) is the maximal compact subgroup
of L. Hence O~.±.o ~ T*JR.3. This is the usual phase space of a spinless particle,
with the difference that in the present description the base JR.3 is momentum space,
whereas the fiber of the cotangent bundle is position space.
For s f. 0 the stability group is SO(2) x JR., where SO(2) c SO(3) is the
stabilizer of some point in s1, and JR. is as above. We are now in a position to use
Theorem 1.10.4 (and especially the subsequent comments) in its full glory. Since
(3.2) is a reductive decomposition, there exists an L-invariant connection on the
396 IV. Reduction and Induction

principal bundle L(L/SO(3), SO(3), r), namely the H-connection; see III.2.7.
Hence Theorem III.2.3.7 asserts that there is an L-equivariant identification of
O~,±,s with a bundle over T*JR3 with typical fiber S;,
providing a Lorentz-invariant
splitting of spin and orbital degrees of freedom. Since the base of this bundle is
contractible, we therefore obtain (3.6). Note that the symplectic structure on the
right-hand side of (3.6) does not factorize; this is what physicists call spin-orbit
coupling.
We pass to the massless case. As we have seen, the stabilizer of p =
(± 1,0,0, -1) is E(2). Its coadjoint orbits may be described either in analogy
to those of E (3) (cf. 1.1 0), or by using Proposition 1.1 0.1, or by direct calculation
from (1.66), in which P 1\ v equals the number PI v 2 - p2V I . Either way, identifying
t(2)* with JR3 using the basis {Ii} (cf. (3.4», one derives that each point (0, 0, h)
is an orbit, and that the remaining orbits are cylinders C, = S;
circleS; ° x JR, where the
of radius r > lies in the p-plane, and JR is the z-axis. The stabilizer of
(0,0, h) is E(2) itself, whereas the stabilizer of a point in C r is JR (embedded in
the JR2 of S 0(2) ~ p JR2 in a way depending on the choice of the point).
The only orbits that are believed to be of physical relevance are the points
(0,0, II). The P-orbit corresponding to such a point is denoted by Or; ± h' The
stabilizer of a point in Ot,±,h is conjugate to E(2) ~ p JR, where JR lies 'i~ M as
JR(l, 0, 0, -1). This group is equal to SO(2) ~ p JR3, where the SO (2)-action p on
JR3 is given by rotations in the (x, y)-plane. By Theorems 1.10.4 and 111.2.3.7 we
infer that Ot,±,h is diffeomorphic to T*(L/ E(2». •

However, since E(2) is not a reductive subgroup of L, one cannot choose an

L-invariant connection A on the bundle L(L/ E(2), E(2), r). Hence the pertinent
diffeomorphism cannot be chosen in a Lorentz-invariant way. Moreover, the Pois-
son structure on T*(L/ E(2» is not the canonical one; it depends on h. Indeed,
using coordinates p, q on the cotangent bundle that are canonical with respect to
the standard symplectic form, but interchanging p and q in the light of the comment
concluding the treatment of O~,±,o above, the Poisson bracket on T*(L/ E(2» is

h at ag at ag at ag
{j, g} = -a -a"
qi P
- -a,'-a -
P qi
hFij(p)-a -a '
qi qj
(3.8)

where F is the curvature of the connection A; cf. III.(2.55). In particular, the

canonical position coordinates do not Poisson-commute.

3.2 Orbits from Covariant Reduction

According to Theorem 1.10.2, the coadjoint orbits oP of P may be obtained
by Kazhdan-Kostant-Sternberg reduction from a particular coadjoint orbit of a
subgroup H c P of the form (1.70). It is interesting to see what happens when
one instead reduces from a coadjoint orbit of the Lorentz group. In particular, we
shall investigate how one may recover the orbits OP in that way. Apart from being
an illustration of symplectic reduction and the theory of constraints, this turns out
 3 Applications in Relativistic Quantum Theory 397

to be the classical analogue of a standard procedure in relativistic quantum theory,

to be discussed in the next section.
Lemma 3.2.1. Let OL be a coadjoint orbit of the Lorentz group L. The reduced
space (T* p)()L , defined by (1.43) with G replaced by P and H by L, is symplec-
tomorphic to T* M* x O~, in such a way that the reduced P-action 1ll.(2.122) on
(T* p)()L becomes the natural action on T* M* x O~.
Here the "natural P-action" is the product of the pullback to (T* M)- ::::: T* M*
of the defining action on M and the coadjoint L-action on OL (on which M acts
trivially). The symplectic structure on T* M* x O~ is not the direct product one,
although its restriction to each factor is the one indicated by the notation; see below.
The proof is a simple calculation; as in the proof of 1.1 0.2 one uses the left
trivializationofT* P. The space I f"[1 (OL) consists of points (M, p, A, v) in _OL x
M* x L x M; elements of the reduced space are equivalence classes of such points
under the L-action derived from III. (1. 49) and (1.64). The map [M, p, A, V)]L t--+
(-Ap, v, -OL(A)M) is a diffeomorphism from I f"[I(OL)/ L to M* x M X OL.
(This map is a combination of minus the momentum map J L and the identity on
M; cf. III.(1.56), (1.66), and Figure 4.)
The symplectic structure on this space may be computed from III.(1.54) and
(1.65). The claim about the reduced P-action may be verified from III.(1.50) and
(1.64). •
Our aim is now to construct the physically relevant coadjoint orbits of P by
symplectic reduction of T* M* x O~. In the simplest case O~ = to} this is done
by imposing the constraints
(3.9)
and ±po > O. The constraint ({J is a momentum map for the JR.-action on T* M*
given by
t : (PIL' qIL) t--+ (PIL' qlL - pILt).
The symplectic reduction (T* M*)'I' of T* M* with respect to this constraint is
symplectomorphic to T*JR.3 (with p-space as the base); for one may map (PIL' qIL) E
C (the subspace of T* M* where ({J = 0) to (Pi, qi - qO pi / Po). Similarly, one
has T*0!;2,± ::::: T*JR. 3 , so that finally (T* M*)'I' ::::: O~,±,o by (3.6). One may put
m = ohere.
This construction may be generalized to arbitrary values of the spin. We use
the matrices M as parameters on 0, in addition employing Ki := MOi and Ri :=
~€ijkMjk> as well as K2 .- KjKj and R2 := RjRj. It may be shown that the
conditions
KjRj=O;
R2 _ K2 = S2 (3.10)
select a coadjoint orbit O~ in [*; this is the only type of orbit we shall need in order
to reach the physical orbits in p*. These orbits are plainly four-dimensional.
398 IV. Reduction and Induction

The massive case is qualitatively different from the massless situation, but we
present the results in united fashion.
Theorem 3.2.2. Let m ~ O. and consider the constraints rp in (3.9) and rpv :=
p/L M/Lv, In addition. impose ±po > O.
• The symplectic reduction ofT* M* x O~ by rp is T*0;'2.± X O~. Thefunctions
rpv are well-defined on this reduced space (and in what follows are regarded as
junctions on T*0;'2.± x Of J.
• For m > 0 the constraints rpv = 0 are second class. and the subspace of
T*0;'2.± x O~ on which they hold is symplectomorphic to the coadjoint orbit
(O~.±.s)+ of P.
• For m = 0 the constraints rpv = 0 are first class. and the symplectic reduction
ofT*O~.± x Of by these constraints is symplectomorphic to the union of the
coadjoint orbits (0c'±.s)+ and (oc,±.-s)+'
• The reduction ofthe natural P-action on T* M* x Of (cf. 3.2.1) to the reduced
spaces above is equivalent to the coadjoint action.
The first claim of the proposition is immediate from the calculation preceding
3.2.2. In what follows p is either (±m, 0, 0, 0) or (± I, 0, 0, -1), and L p denotes
either S 0(3) (for m > 0) or £(2) (when m = 0). In the former case [~ may be
identified with b (cf. (3.2».
We identify T*OLm.2 ± with L XL-P [op.; cf. (3.2) and 111.(2.116). The constraints
p/L M /LV = 0 are then the components of <I> : (L x Lp ~) X 0; --+ M*, defined by
<I>([A, N]L p ' M):= MAp.
Contracting <I> with p, or rpv with pV, one immediately sees that at most three
components of rpv are independent. Since M itself satisfies the condition of lying
in 0;, only two components of <I> are actually independent. The solution set
C;' of these constraints consists of those ([A, N]L p ' M) for which N E (~ and
M E A([p nO;). This is well-defined, for Lp maps (p n Of into itself, so that
changing A by AA makes no difference as long as A E L p.
For m > 0 the set (p n 0; contains all matrices M in ( for which K j = 0 and
R2 = S2; with (p = 50(3), identified with its dual, it is clear that this set is the
coadjoint orbit S; of SO(3). We now define 1/1 : C;' --+ L xSO(3) rrl(S;) (where
rr : (* --+ (T is the restriction map) by
1/1 ([A , N]SO(3), M) := [A, N - A-I M]so(3)' (3.11)
It is obvious from the above description of C;' that 1/1 is a diffeomorphism.
Now recall from 1.10 that L XLp rrl(-O~p) is symplectomorphic to [email protected]);
see the paragraph preceding (1.80). Applied to the case at hand, this means
that L XSO(3) rrl(S;) is symplectomorphic to O~.±.s' Chasing the definitions
of the relevant symplectic structures, one may (tediously) verify that C;' is in fact
symplectomorphic to O~.±.s' This proves, in particular, that C;' is symplectic.
For m = 0 the set Ip nO; consists of all matrices M in ( for which K I = - R2,
K2 = RI, R3 = ±l, and K3 = O. Hence C? is the union of two components
 3 Applications in Relativistic Quantum Theory 399

c~'±. This time the analogue of (3.11) fails to be a diffeomorphism. The selection
of two independent constraints <I> I, <1>2 is made via the choice of a (local) section
b: L/ £(2) ~ 0k± ~ L. The <Pj in question are then simply the first and second
components of <l>b, defined by <l>b([A, N]E(2). M) := b(Ap)-I<I>([A, N]E(2), M).
The following claims may be verified by a straightforward but tedious local
analysis. The two independent constraints generate a free and proper action of
]R2 on T* 0k± x 0;; the momentum map J of this action is, of course, given by
J j = <1>7. We are therefore in the situation of Marsden-Weinstein reduction at zero.
concluding from Theorems 1.5.4 and 1.2.2 that J-1(0) = c~ is coisotropically
embedded in T*Ok± x 0;. While the explicit form of the ]R2-action depends
on b, the Marsden-Weinstein quotient J- 1(0)/]R2 does not. The analogue of the
map (3.11) quotients well to J- 1(0)/]R2, and is a diffeomorphism. The remainder
of the argument is then as in the massive case. It is elementary to verify that the
pertinent P-actions are intertwined by the symplectomorphism between C:Z and
O~ ± s constructed above. D

3.3 Representations of the Poincare Group

In this section we relate the irreducible representations of the Poincare group P
to its covariant representations; these are by definition the ones that are in-
duced from some finite-dimensional representation of the Lorentz group L. We
will first quickly go through the classification and realization of the irreducible
representations of p. assuming that the reader has seen this material before.
In order to apply Theorem 1.10.3 we need to verify that P is a regular semidirect
product. All L-orbits in M* are closed. except 0k±. The closure of the latter is
obtained by adding the origin. so that it is obvious that the regularity condition is
met. Hence the irreducible representations of P are classified by 1.10.3. As in the
classical case, we will concentrate on the representations related to the L-orbits
0;'2,±. for m 2 ::: O.
The orbits 0;'2.± possess an L-invariant measure v, which, upon identification
of 0;'2,± with JR.3 by simply omitting the variable

(3.12)
is dv(p) = dp1dp2dp3/(161'(3Wp)' Hence we can form the representation space
(1.75) and the representation (1.76). For m > 0 and Lp = SO(3) the label C1 is the
spin s. taking values in N U 0, so that 1-£s = C 2s +1 is the space carrying the well-
known irreducible representation Us (SO(3». For m = 0 we are interested only
in irreducible representations of Lp = £(2) that correspond to the SO(2)-orbit
(0,0) in ]R2. Such representations Uh are labeled by the helicity h E SO(2) = Z,
and are realized on 1-£h = C.
Comparing these representations with the coadjoint orbits of the classical theory
in 3.1, one sees that the only difference in the parametrization lies in the fact that in
quantum theory spin and helicity assume integral values. (It can be shown that the
projective representations of P are given by representations of its covering group
400 IV. Reduction and Induction

P = SL(2, C) ~p M, where the action jj projects to an action of L via the well-

known covering projection from SL(2, C) onto L. For the above classification the
replacement of L by SL(2, C) has the consequence that s and h may now be half-
integers as well.) The following summary should be compared with Proposition
3.1.1.
Proposition 3.3.1.
• The Poincare group P is a regular semidirect product.
• For m > 0, P has a family of irreducible representations V m.±,s, where s E
N U 0, which is realized on 'It;,±,s = L 2(O~2,±) ® C 2s +1 by (1.76).
• For m = 0, P has, among others, the series of irreducible representations
VO.±.h, where h E Z, realized on 'It~.±.h = L 2( O{;,±).

Here b : O~2.± ~ Lj Lp -+ L is a measurable section, as explained above (1.75).

The explicitform of these representations is given by (1.76).
Our goal is now to relate these representations to the covariant representations
of P. These are defined by taking G = P and H = L in III.2.9 and taking the
representation V)..(L) from which one induces to be finite-dimensional. This is
motivated by relativistic field theory, in which the basic fields of a theory generi-
cally transform according to such a covariant representation. However, since L is
semisimple and noncompact, it can be shown that its finite-dimensional representa-
tions are all nonunitary, except the trivial one. Thus we discard the symbols 'It).. and
V).., suggesting unitarity, in favor of S).. and R).., respectively. The corresponding
induced representations R)..(P) are nonunitary as well, except the trivially induced
representation Rid.
This is not a problem, because the induced space 'It).. will be regarded primarily
as a symplectic manifold, henceforth called S)... The point, then, is that R).. defines
a strongly Hamiltonian action of P. The relationship between the covariant and
irreducible representations will, accordingly, be achieved by symplectic reduction.
The notion of irreducibility, however, is the one from unitary representation theory
(rather than the transitivity of the P-action).
In any case, one may construct the induced space S).. = 'It).. as a Hilbert space in
the same way as for unitary induction; see, in particular, the text below III.(2.176).
The next lemma should be compared with 3.2.1. The analogue of (3.14) for E(3)
instead of P has already been encountered in (2.116).
Lemma 3.3.2. The Hilbert space S).. and the induced representation R)..(P) may
be realized as
S).. = L 2 (M) ® S)..; (3.l3)
R)..(A, v)'I1)..(q) = V).. (A)'I1).. (A -I(q - v», (3.14)
or, after a Fourier transform on '11).., as
S).. = L 2 (M*)® S)..; (3.15)
n)..(A, v)q,\p) = eipv V).. (A)q,)..(A -I p). (3.16)
 3 Applications in Relativistic Quantum Theory 40 1

Here we have simply written S').. for S;, etc.; the inner product on S').. (and similarly
on S').. ) is given by

(3.17)

One notes that P / L ~ M, so that L2(p / L) = L 2(M) := L 2(JR4), and chooses

the section s : P / L --* P to be s(q) = (1, q). Since the Lebesgue measure on M
is Lorentz-invariant, 111.(2.176) then simplifies as stated. •
The definition of the symplectic form with respect to which the action R}(P)
is strongly Hamiltonian, and the corresponding identification of the canonical
variables, depend on the form of 'R')... We shall discuss only two cases, of relevance
to physics. Firstly, we take 'R').. = 'Rid, where 'Rid (A) = 1 for all A E L. The
inner product on Sid = (; is the usual one, so that 1.(2.35) with (3.17) defines a
symplectic form relative to which the action 'Rid(P) is strongly Hamiltonian; cf.
Theorem III.l.6.1. As we have already remarked, this case is rather atypical, in
that 'Rid(P) is unitary.
Secondly, we pick the vector representation 'R').. = 'R v , defined on Sv := (;4
by
(3.18)
Elements of Sy are traditionally denoted by A, since in physics gauge fields are
examples of fields transforming according to 'R y. The complex structure and inner
product on Sy are not the usual ones on (;4, however; they are defined by
i(Ao, AI, A2, A 3):= (-iAo. iA I , iA 2• iA3); (3.19)
(A, B)v := AoBo + A,B, + A2B2 + A3B3. (3.20)

The point of these definitions is that 1.(2.35) now defines a Lorentz-invariant

symplectic form on Sv, for (with slight abuse of notation) one has
(3.21)
The same definition, but now applied to the inner product (3.17), then defines a
symplectic form on SV that is invariant under the P-action 'R v. Consequently, this
action is strongly Hamiltonian.
In both cases the identification of S').. as a phase space is equivalent to regard-
ing L 2(M) ® (;N, where N = 1 or 4, as the cotangent bundle of L2(M. JRN);
cf. 111.(2.28)-(2.31). This comment, however, would not apply to arbitrary
representations 'R')...
In all cases, the first step of the reduction of S').. is performed by imposing the
infinite number of constraints
(3.22)
on S').., where 0 := 8/.L8/.L is the d' Alembertian. On S').. these constraints read
(3.23)
402 IV. Reduction and Induction

cf. (3.9). Unfortunately, this equation has no solutions in the given phase space.
Although conceptually this situation is purely classical, one is reminded of the
discussion about Dirac's constrained quantization method in 2.9. In the present
context the problem arises because of the infinite number of degrees of freedom.
Regarding ",A as an infinite-component coordinate qi, i E M*, the form (3.23)
°
shows that the constraints in question amount to putting qi = for a certain subset
i E M~ c M*. If the index set M* were finite, the reduced phase space would
simply be T*JRM·\M c. Formally regarding M~ as the set where (3.12) holds, the
space M*\M~ may be identified with JR3 through the elimination of Po as an
independent variable. To make this argument precise, one first expands solutions
of (3.23) with given sign ± of the energy Po by

\lJA(q) = ( d 3 p e-ipqq,A(p), (3.24)

± JR.J 16rr 3wp ±

where Po = ±wp • One subsequently declares the reduced space to be

sm,±,A := {\IJ~ E S'(M) ® SA I
(D+m 2)\lJ A =0, q,~ E L2(JR3,d3p/(16rr3wp»®SA); (3.25)

the inner product (\IJ±, <l>±) in sm,±,A is by definition equal to the one (q,±, <I>±) in
L 2(JR3, d 3 p/(16rr 3wp»®S),. Here the fact that all solutions of (3.23) inS'(M)®S),
admit an expansion of the type (3.24) is a consequence of (3.23) and the fact that the
Fourier transform maps S'(M) into S'(M*), The extension of L 2(M) to S'(M)
is possible, since in classical physics one is not tied to the choice of L 2(M*) as
the unconstrained phase space, and is free to enlarge it.
It follows that one may identify sm,±,A with L2(O~2,±) ® SA' The reduced
P-action Rm,±,A is given as in (3.16), with the understanding that Po is given
by (3.12), We look at sm,±,A as a symplectic manifold in a manner analogous
to the interpretation of SA explained above, and we regard Rm,±,A as a strongly
Hamiltonian action, rather than as a representation.
Notice, however, that sm,±,id = 1tm,±,o and Rm,±,id = Um,±,o, which is, excep-
tionally, a unitary representation. This brings us in a position to justify the choice
of (3.25) as the reduced phase space by a completely different argument. Here SA
and UA "go along for the ride", so we omit them.

Proposition 3.3.3. Regarding L 2(M) as a Hilbert space, Fell induction on the

basis of the quantum constraint (3.22), supplemented by ±po > 0, yields the
induced Hilbert space L2(O~2,±).
The representation (3.14) on L 2(M) is thereby induced to a representation on
L2(O~2.±)' which coincides with the irreducible representation Um,±,o(P).

The following construction may be seen as the quantum counterpart of the re-
duction of T* M* by the constraint (3.9). The construction is possible, and provides
an unexpected quantum twist to an otherwise classical situation, because the action
Rid(P) is not merely strongly Hamiltonian, but unitary.
 3 Applications in Relativistic Quantum Theory 403

The quantum reduction procedure is essentially the same as the one discussed
in 2.10 for the constraint ~(pi - p~) on T*R2; see (2.128) etc. Thus we consider
the representation V(R) on L2(M*) defined by

V(f)W(p) := e~it(p2-m2)W(p). (3.26)

We then perform Fell induction, applying Construction 2.2.5. This time we may
t
simply take c L2(M*) to be Cgo(M*). Adding the condition ±po > 0, and
rescaling the inner product by a factor of 4 (which could have been avoided by
rescaling the constraint if desired), the induced space then emerges as S1 =
L2(O~2.±)' on which the representation U~(p) Fell-induced from Uid(P) is equal
to the irreducible representation Um.±.o(P). •

We have added the suffix F here in order to distinguish between induction from
representations of L and Fell induction. In the current analogue of the functions
(2.134), the label k is replaced by p, and one now has Jp±(ql1) = e ipq , where
Po = ±wp • These functions do not lie in 'Ji, yet they nonetheless form a complete
set of linearly independent solutions of the constraints (3.22).

3.4 The Origin a/Gauge Invariance

For scalar fields we have achieved our goal of relating the covariant action Rid(P)
to the irreducible representation Rm.±.id(P). For nontrivially induced actions R)..
it is necessary to impose further constraints in order to achieve irreducibility. This
will eventually bring us to the main theme of this section, namely the relationship
between masslessness and gauge invariance.
The first step towards irreducibility consists in defining new variables by means
of the (bounded and invertible) map Ub, defined on sm.±.).. by
-).. 1 -)..
UbW±(p) := R)..(b(p)- )W±(p), (3.27)

where p = (±wp , p), and b : O~2.± --+ L is a section, as before. This map is
obviously not unitary, but it is a symplectomorphism when the symplectic form
on sm.±.).. has been defined appropriately. It is customary in physics to denote the
left hand side of (3.27) by a~(p), suppressing the b-dependence. The point of the
transformation (3.27) is that the P-action R;·±·).. := UbRm·±·)..U; is given by
R;·±·)..(A, v)a~(p) = eipvR).. (b(p)-I Ab(A -I p») a~(A -I p). (3.28)

Compare with (1.76), recalling that the argument of U(j in that formula, and there-
fore the argument of R).. in (3.28), lies in L jj (which, we recall, is S 0(3) for m > 0
and £(2) for m = 0). The only difference is that U(j in (1.76) is a unitary irre-
ducible representation of L jj , whereas the restriction R)..(L r Ljj) ofR)..(L) to Ljj
is possibly nonunitary and reducible.
Let us briefly discuss the massive case, which is well understood.
Proposition 3.4.1. Suppose that R)..(L rSO(3» contains Us(SO(3», and let
P)..-->s be the projecforon S).. whose image is the subspace carrying Us(L r SO(3».
404 IV. Reduction and Induction

The constraints

(3.29)

are Poincare covariant, in that W~ satisfies (3.29) ijfRm·±·J..(A, v)W~ satisfies

(3.29). In particular, the left-hand side of(3.29) does not explicitly depend on the
section b.
The subspace of sm.±.J.. satisfying the constraints (3.29) is precisely the Hilbert
space 1tm.±.s carrying the irreducible representation Um.±.s(P). Tn particular, this
subspace is symplectic, so that the constraints (3.29) are second class.

The first point is obvious from (3.28) and the fact that pf--.s commutes with all
RJ..(-)' The remainder follows from the discussion after (3.27). •

For example, the representation Rv(L r SO(3» reduces as Uo EB UI, under

[eo], so that with p = (m, 0, 0, 0), we may write pil- Ail- = °

which the carrier space C4 decomposes as Ceo EB C3 • The projection PJ--..+ I equals
for P-t...... 1A = O.
Since (A -I): = Ail-v' and by definition of b one has b(p)lL"pv = Pil-' (3.29) may
be written as the covariant equation Pil-AjJ.(p) = O.
Returning to our original starting point, one may impose (3.29) and (3.23) or
(3.22) on SJ.., and, choosing a sign of the energy Po, conclude that the reduced
phase space is 1t m.±.s. All values of sEN can be reached in this way, and any
covariant wave equation for massive fields is of this form. (To reach s E N/2 one
needs to start with a representation of SL(2, C) rather than of L.) Wave equations
are usually posed in M rather than M*; for example, in the situation discussed
above, the complete set of constraints consists of (0 + m 2 )AjJ. = 0 and ajJ.AjJ. = O.
For massless fields a similar procedure may be used: For any given helicity
h E Z there exist representations of L whose restriction to E(2) contains Uh. For
example, to reach U±I one may use the representation on antisymmetric tensor
fields FjJ.v' However, in physics a key role in the description of massless fields
with helicity ±1 is played by vector fields Aw Let us therefore investigate the
restriction of Rv(L) to E(2).

Proposition 3.4.2. Equip Sv = C 4 with the symplectic form (3.21); the repre-
sentation Rv(L) defined in (3.18) defines a strongly Hamiltonian L-action on Sv.

se
Seen as a representation, the restriction ofRv to E(2) is indecomposable.
The symplectic reduction of Sv by the constraint pjJ. AjJ. = 0, where p =
(1, 0, 0, -1), is C 2 with its usual symplecticform 1.(2.35). The reduced E (2)-action
on Se is the representation UI EB U-I.

Choosing a basis {UI := e}, U2 := e2, U± := !(eo ± e3)}, and taking the
generators (3.4) in the defining representation, one calculates that P := Cu_ = C P
is invariant under E(2), as is the span :r of {u}, U2, u-l. The representation is
indecomposable, because Ta U a = u_ for a = 1, 2.
Since pjJ. = (1, 0,0, 1), the solution of the constraints is :r.One computes that
:r :r
the null space of the symplectic form on is p, so that S~ = IP ~ C2 • Explicit
computation of the action of the generators yields the final claim. 0
 3 Applications in Relativistic Quantum Theory 405

This reduction of S~ is of the Marsden-Weinstein type: With H = C acting on

C4 by
>.. : AIL t-+ AIL - i>"PIL' (3.30)
one computes from 111.(1.8) and 1.(2.32) and the subsequent text line that
J(A) := _2pIL AIL (3.31)
is an equivariant momentum map for this action. Hence 3 = J- 1(0) and S~ =
J- 1(0)/ H. The space P consists precisely of those vectors in Sy that are "pure
gauge", that is, of the form AIL = >"PIL for some>.. E H.
The symplectic orthogonal complement of a subspace V (which is not necessar-
ily linear) of a linear symplectic space x:; (such as a Hilbert space with symplectic
form 1.(2.35» is (cf. 1.1)
V i := {z E x:; I w(z, w) = OVw E V}. (3.32)
Applying this to x:; = Sy and V = P, one sees that pi = 3.
Before proceeding on the basis of these insights, we should first pay attention
to a subtle point concerning real fields. The representation ny(L) is real, in that
lR.4 C (:4 is invariant under L. In physics one defines a real vector field with mass
m 2: 0 as an element of
(3.33)

The Fourier coefficients A± are now dependent variables, related by A_(p) =

A+(p). This leads to an identification of the real Hilbert space sm.V.JR. with the
complex Hilbert space sm.+. v, endowing sm.lR. v with the structure of a complex
n
Hilbert space, as well as with the action m.+. v (P) (rather than with the restriction
ofnm .+. v$n m .-. y to the real subspace in question, as might have been expected).
This juggling is justified by the physical requirement of positive energy. In any
case, the symplectic form on sm.JR.. Y may be expressed as

(3.34)

where E C M is an arbitrary Cauchy surface for (3.22), and we have written A

for V(A), etc. The norm in sm.JR.. V (seen as a Hilbert manifold) is

2
IIAII:=
1 JR.3
d3 p
16 3
~ - 2
L.J IAIL(p)1 .
7r Wp IL=O
(3.35)

This norm is not Poincare invariant, but the topology it induces is. A similar
procedure applies to scalar fields.
After these preparations we come to our main point of relating masslessness to
gauge symmetry.
Definition 3.4.3. Let Qc c S'(R4)/lR. consist 0/ all real solutions>.. o/the wave
equation 0>.. = 0 on M, modulo the constants, and let the gauge group Q consist
o/those>.. E Qc whose (weak) derivative a>.. (seen as a/our-vector with components
406 IV. Reduction and Induction

all A) lies in SO.RV. The Lie algebra 9 is identified with Q, which becomes a Hilbert
space in the norm

2 [ d3P - 2
IIAII = (A, A)g := (aA, aA)so.Rv = JJR3 (2n)3 wpIA(p)1 . (3.36)

Here wp = n, and). equals ).+, defined as in (3.24).

The following result transfers the setting of Proposition 3.4.2 from Sv to S :=
SO.R v, at the same time being an infinite-dimensional analogue of the part of
Theorem 3.2.2 concerning m = O.
Theorem 3.4.4. The action of the gauge group Q on SO,JR, v , given by
(3.37)
is strongly Hamiltonian. The Marsden-Weinstein quotient SO := J-1(0)/Q with
respect to this action is symplectomorphic to the Hilbert space '}-{O,+,I EI7 '}-{O.+.-I
(cf 3.3.1), with symplecticform 1.(2.35). The action nO,+,v of Pan S reduces to
an action on SO that is equivalent to the representation UO,+,I $ UO,+,-I,
The last-mentioned representation of P is reducible, but becomes irreducible
when spatial reflections are included. Particles transforming under UO,+,1 $UO,+,-I
are called photons.
Our choice of the (Hilbert) manifold structure on S and Q implies that the action
(3.37) is smooth, since addition in a Hilbert space is smooth. Following the steps
leading to (3.31), one then verifies that a momentum map for (3.37) is given by

1W: p f--)- -pIlA Il (p)/p2, (3.38)

where PO = n. When A lies in SO.R v , the function JW is indeed an element

of g* :::::: Q. Since the coadjoint action is trivial, one verifies that J is equivariant.
In the setting of (real) Hilbert manifolds, the theory of Marsden-Weinstein
reduction is essentially the same as for finite-dimensional manifolds. Definition
3.4.3 and (3.37) easily imply that the Q-action is free. However, since Lemma 1.5.1
was proved using a dimension-counting argument, which does not generalize to
the infinite-dimensional case, we now prove directly that 0 is a regular value of
J. The derivative J*(B) at A is independent of A, and equal to J(B). Hence J* is
surjective for all A, so that 0 is regular by definition.
U sing the second formulation of properness in Definition 1.5.2, one immediately
shows thatthe (I-action on S is proper, for An ---+ A in (I precisely when allAn ---+ allA
in S.
The Marsden-Weinstein quotient SO therefore exists as a symplectic manifold
by Theorem 1.5.4. The set J- 1(0) consists of all A E S satisfying all All = 0,
or, equivalently, pll A(p) = O. (In the present context this equation should not
be thought of as a gauge-fixing condition, but as Gauss's law.) This is a closed
subspace of S, which easily implies that SO is a Hilbert space. The identification
of the reduced space and the reduced P -action proceeds in complete analogy with
the proof of Proposition 3.4.2. •
 3 Applications in Relativistic Quantum Theory 407

The comments following (3.31) may be repeated verbatim: defining P C SO,JR, v

as the fields that are pure gauge, i.e., of the form AJ.L = 8p.A for some A E g, and
putting:r := J-1(0), one has :r= pi. and SO ~ jP.:r

3.5 Quantum Field Theory of Photons

On the basis of the general idea that symplectic reduction is to be quantized by
induction (in the sense of Rieffel or Fell), we shall formulate an analogue of
Theorem 3.4.4 in quantum field theory. In preparation, we define an important
class of C* -algebras.
Definition 3.5.1. The CCR algebra 2!1(K) over a Hilbert space K is the twisted
group C* -algebra C*(Kd , c), where Kd is K as an additive group, equipped with
the discrete topology, and the multiplier is

(3.39)
Here the symplectic form w on K is defined by 1.(2.35).
It is customary to write W(z) forthefunction f satisfying f(z) = 1 and f(w) =
o for all w =1= z. We assume that the Haar measure on Kd is normalized so that
each point has measure 1. From III.( 1.80) and 111.(1.81) we then have the relations

W(z)W(w) = e-!iW(Z,W)W(z + w); (3.40)

W(z)* = W( -z). (3.41)
Equation (3.40) describes the canonical commutation relations in Weyl form,
which explains the names CCR and 2!1. Note that substitution of (3.41) into~)
shows that W(z) is unitary. We will denote the linear span of all W(z) by 2!1(K);
this is plainly a dense subalgebra of 2!1(K), playing an important role in what
follows.
It can be shown that the CCR-algebra is simple, so that all nondegenerate
representations are faithful. A most important representation of 2!1(K) is the
Fock representation 7rF on the bosonic Fock space exp(K) defined in 11.(2.61).
Recalling 11.(2.67) etc., this representation is defined by continuous extension of
7rF(W(Z» := ea(z)'-a(z) (3.42)

from the span IE of the exponential vectors; cf. 11.(2.62). On use of the CBH-
formula, 11.(2.69), and 11.(2.70), this is equivalent to

+ z).
I
7rF(W(z»JExp(w) = e-i(z.z}-(z.w) JExp(w (3.43)

For example, whenK = en,the operator7rF(W(z» coincides with UI(Z) as defined

in 11.(2.71).
While the preceding paragraph in conjunction with 11.2.3 relates the CCR-
algebra to Berezin quantization (at least when K is finite-dimensional), there
are equally close links between the CCR-algebra and Weyl quantization; cf. the
comments following 11.(2.112).
408 IV. Reduction and Induction

Proposition 3.5.2. Let U be an operator on K that preserves the symplectic

form 1.(2.35); this is the case,for example, when U is unitary. Then linear and
continuous extension of au (W(z» := W(U z) defines an automorphism ofW(K).
Linear extension of u F,JEXP(z) := ,JEXP(Uz) defines an operator U F on
IE c exp(K). When U is unitary, the map U F i~ unitary, so that it may be extended
to all ofexp(K). It then implements au, in that UFlfF(A)(U F)* = lfF(au(A»
for all A E W(K), In particular, a group representation Ux(G) on K leads to a
representation U: (G) on exp(K).

It is clear that (3.40) is preserved by au, and so is (3.41) by the linearity of U.

The unitarity of U F for unitary U is immediate from 1l.(2.63).
It is not obvious that U F as given is well-defined, since the exponential vectors
form an overcomplete set. However, putting UF(WI ®s .. , ®s w n) := UWl ®s
... ®s U W n, and using 11.(2.62), defines the same operator in an unambiguous way.
The implementing property follows from a simple calculation, using (3.43). •

The following result will be used in relationship to the weak algebra of observ-
abIes of the quantum field theory of photons. Restricting w to V, one obtains a
C*-algebra W(V) by Definition 3.5.1. It is clear that the linear span of all W(z),
z E V, may be regarded as a subspace of W(K) by extending the functions in
question to K with the value 0 outside V. Looking in the Fock representation, one
sees also that the completion W(V) is a subalgebra of W(K). Similarly, W(V.l)
may be defined as in 3.5.1; cf. (3.32).

Proposition 3.5.3. The commutant W(V)' ofW(V) in W(K) is W(V.l),

The inclusion W(V.l) ~ W(V)' is immediate from (3.40); the hard part of the
proof is the opposite inclusion.
For f E lOO(Kd ) one has the inequalities

lI!IIoo ::: 1I!II2 ::: II !II , (3.44)

where the first norm is the sup-norm, the second norm is in l2(Kd) (with respect
to the Haar measure on K d ), and the third is in W(K). The first inequality is
obvious (given the discreteness of the underlying measure space), and the second
follows from the existence of the state wo, defined by continuous extension of
wo(f) = f(O); indeed, II f II~ = wo(f* f). It follows that W(K) as a Banach space
(with its C*-norm) is continuously embedded in lo(Kd) (with sup-norm), for any
element of the former is the limit of a Cauchy sequence in le(Kd ); by (3.44) this
sequence must also converge in the sup-norm, so that its limit must lie in lo(K d ).
Now take an arbitrary f E W(K), and a Cauchy sequence fn in le(Kd) con-
verging to f in W(K). It then follows from (3.40) that the commutator [fn, W(z)]
is the function f~z) : W H- 2ifn(w - z) sine -4w(w, z». Now, limn f~z) exists in
W(K), hence in lo(Kd)' The function W H- sine - iw(w, z» lies in lb(Kd ), which
is the multiplier algebra of lo(K d ); cf. (2.13). Hence f~z) ---* f<z) (defined like
f~z), with fn replaced by f) is in lo(Kd)' By uniqueness of the limit, we infer
f~z) ---* f<z) in W(K). We conclude that [f, W(z)] = f<z).
 3 Applications in Relativistic Quantum Theory 409

Now, f is in2U(V)' iff[f, W(z)] vanishesforallz E V. The preceding paragraph

then yields II f<z) II = 0, whereupon (3.44) implies that f<z) identically vanishes for
= =
such z. Therefore (evaluating f<z) at w w' + z, and using w(z, z) 0), f must
vanish whenever its argument does not lie in Vl. ,and the proposition follows. •
Applied to the cases /C = Sv and /C = so. JR. v of the preceding section, we see
that in either case Proposition 3.5.3 implies the equality
2U(P)' = 2U(3). (3.45)
Note that 2U(P) is abelian, because w vanishes on P. For P c Sv this is because
piJ. piJ. = 0, and for P C SO.JR. v the reason is that aiJ.aiJ.).. = O.
To interpret this result, we look in the Fock representation; in what follows /C
is either Sv or SO.JR. v, and H is C or g, respectively. The Hilbert space exp(/C)
carries a representation U F(H), defined by
(3.46)
where).. E P; we have identified H with the corresponding subspace P c /C. It is
obvious from (3.40) and the fact that w vanishes on P that UF defines a linear action
of H, and the unitarity of each W()") implies that this action is unitary. (Note that
U F is not an example of a representation of the type U: mentioned in 3.5.2.) The
explicit form (3.43) shows that U F()..) performs the gauge transformation (3.37) on
the argument of .JEXP; the remaining term may be thought of as a factor included
to make U F()..) unitary. Identifying 2U(/C) with its faithful Fock representative,
(3.45) then means that the gauge-invariant subalgebra of 2U(/C) is 2U(3).
We are now going to "quantize" Proposition 3.4.2 and Theorem 3.4.4. We would
like to use the specialization of Rieffel induction to quantum Marsden-Weinstein
reduction explained in 2.5. In particular, we take our cue from Theorem 2.5.4, in
which we put 1i = exp(/C), U = U F, and t = ~; recall from 11.2.3 that ~ is
the linear span of all exponential vectors. However, for /C = Sv and H = C it is
not clear how to construct a pre-Hilbert C~(H)-module that is stable under the
action of a suitable subalgebra of 2U(Sv). For /C = SO.JR. v and H = 9 the group
C* -algebra does not even exist; infinite-dimensional topological vector spaces do
not support nontrivial translation-invariant Borel measures, so that 9 does not have
a Haar measure. We therefore use Fell induction, in which we fill out the data as
if we were performing Rieffel induction from the trivial representation Uid(H); cf.
(2.81).
We first specialize to /C = Sv and H = C; the following theorem is the quantum
counterpart of Proposition 3.4.2.
Proposition 3.5.4. In Construction 2.2.5, put t = ~, 1ix = Sl3(1i x ) = 1iid = C,
and
id._
(\II, <1»0 .-
[d)"dI
- - . (\II, UF()..)<I». (3.47)
c 2m
Here we have written (\II, <1»0 for (\II, <I>}c, which is the same by (2.37), in order
to stress the analogy with (2.80) or (2.81).
410 IV. Reduction and Induction

1. This form is finite for'll, <I> E ~, and satisfies (2.35) and (2.46).
2. The induced space 1{~ := 1{x may therefore be constructed as in 2.2.5.2. This
space is naturally isomorphic to exp(S~); recall that :1 jP ~ S~ is the classical
reduced space.
3. Each gauge transformation U F()..) is adjointabLe, and acts trivially on the
induced space, in thatfor aLL)" E C one has
(3.48)

-----
4. The pre-C*-aLgebra TCF(!ID(:1» Leaves IE s~ and consists of adjointabLe

----
operators. The induced representative TC~(211(:1» defined by 2.2.5.3, where
TC~(A) := TCid(TCF(A», is isomorphic to 211(S~).
5. Defining R~(E(2» on ~ by taking U = Rv(A) in 3.5.2, where A E E(2),
each R~(A) leaves ~ stabLe and is adjointable. The induced action U~ on 1{~,
given by U~(A) := TCid(R~(A», is linear and unitary, and equivaLent to the
representation (UI $ U_I)F (cf 3.5.2).
6. Since the E (2)-action Rv on Sv is sympLectic, it defines an automorphic action
a of E(2) on 211(Sv), as expLained in 3.5.2. Because:1 C Sv is stabLe under
Rv(E(2», this action restricts to 211(:1). In the representation TC~ the Latter
automorphism group is impLemented by U~.
We will show that (3.47) is finite by explicit calculation. The property (2.35)
holds because H = C is unimodular. Equation (2.46) follows as in the proof of
Theorem 2.5.4 (as C is amenable); a different proof is given below.
Claim 3 follows from the property
(3.49)
for all ).. E 9 and'll, <I> E ~. This is a simple consequence of (3.47) and the
translation invariance of the Haar measure on C.
To express (3.47) in a convenient form, we decompose A = A L + AT, where
AT = (0, AI, A 2 , 0) lies in the orthogonal complement (in the Hilbert space sense)
in:1 of p, and AL = (Ao, 0, 0, A3) is the orthogonal complement of AT in Sv.
Recall that n is defined below 11.(2.61). A Gaussian integration results in

(JEXP(A), JEXP(B»): = (JEXP(A L ), n): (n, JEXP(B »): L T

e(AT,B ).

(3.50)
The identification of the reduced space uses the method of Proposition 2.2.4.
Our guess is 1i~ = exp(S~), and this is proved by defining [; : IE -+ exp(S~) by
linear extension of

(3.51)

where [AT]p is the equivalence class of AT E :1 in :1 jP ~ C2 ; cf. 3.4.2. Since

(AT, BT) equals ([AT]p, [BT]p)l(;2, it is obvious from (3.50) that (2.29) holds.
Moreover, this construction establishes (2.46), for 1{~ is a Hilbert space. It is clear
 3 Applications in Relativistic Quantum Theory 411

that (j ~ is dense in exp(S~), so that the guess of the latter as the induced space
has been vindicated, proving 3.5.4.2
It is easily seen that an operator on ~ is adjointable iff it commutes with all

------
U F(A.). By (3.45) the intersection of lI'F(2U(Sv» with the space of all adjointable
operators on ~ is 11'F(2U(..1»; this implies the first part of 3.5.4.4.
All remaining claims in 3.5.4 easily follow from (2.30); the use of exponential
vectors has reduced these verifications to Proposition 3.4.2. •
After this warm-up we tum to the quantization of Theorem 3.4.4. Thus we
specialize the discussion preceding 3.5.4 to the case K, = SO,R,V and H = g. In
order to define an integral of the type (3.47), we first use (3.46), (3.43), and (3.36)
to compute the integrand in the attempted generalization of (3.47) as

(v'EXP(A), U F(A)v'EXP(B») = e(A,B)e-!1I'·1I 2 e<A,o>.)-(o>'.B). (3.52)

Knowing that topological vector spaces support certain Gaussian measures, this
suggests combining the Gaussian factor in (3.52) with the nonexistent flat measure
on g. Unfortunately, the ensuing combination defines a set function on 9 that is
merely finitely additive, and therefore fails to be a measure (which by definition
is countably additive). To have a measure, it is necessary to enlarge g. Partly for
later use, we present the general setting.
Recall Definition 11.1.5.6 and the subsequent theory.
Theorem 3.5.5. Let 1i be a real Hilbert subspace of a quasi-complete locally
convex Hausdorff vector space V, and take y > O. Suppose that V carries a
Radon measure JLy whose Fourier transform is given by

Iv dJLy(v)e i9 (V) = e-!YQ(9,9). (3.53)

Here() E V*, and the quadraticform Q on V* is defined by 11.(1.49). A measure with

this property is called Gaussian, with covariance Q, and is uniquely determined
by its Fourier transform (3.53).
1. The map 01--+ y-I/2e from V* to L2(V, Ji.y), defined by O(v) := O(v), is well-
defined and isometric, so that it extends to an isometry w 1--+ y-I/2 Wfrom 1i
w
to L2(V, Ji.y). For each W E 1i this defines as an element of L2(V, JLy); we
write (w, v) := w(v), which makes sense for almost all v E V with respect to
JLy.
2. In (3.53) one may replace 0 by W, yielding

(3.54)

3. The translate of JLy by W E V is disjoint from JLy when W ¢ 1i, and equivalent
to JLy when W E 1i, with Radon-Nikodym derivative given by the general
Cameron-Martin formula

(3.55)
412 IV. Reduction and Induction

An elementary computation yields (e, e)LZ(V,/Ly) = yQ(fJ, 19), so that the first
claim follows by 11.(1.49). In particular, the property fJ(v) = 0 for all v E 1-£
e
implies (}(v) = 0 for JLy-almost all v E V, so that the map 01-+ is well-defined.
To derive (3.54), take a sequence {On} in V* for which en -+ w in 1-£, so that
en -+ win L 2(V, JLy) by 3.5.5.1. Hence exp(ien) -+ exp(iw). Since In -+ I in
L 2 implies (In, 1) -+ (j, 1) when 1 E L2, one obtains (3.54) from (3.53).
The last claim is the general Cameron-Martin theorem. The proof of the
disjointness property is beyond the scope of this book. In view of its importance
for what follows, we do outline the proof of (3.55). For simplicity we put y = 1
and JL : = JL I· First take w = ij, where rJ E V*, and take the Fourier transform of
each side of (3.55). Using (3.53), the left hand side is immediately found to be

Iv dJL(v + ij)eiO(v) = e-!Q(O,O)-iQ(1/,O). (3.56)

The right-hand side of (3.55) may be heuristically computed by formally applying

(3.53) with 19 replaced by 19 + irJ. The result indeed equals the left-hand side.
To proceed rigorously, we remark that a continuous linear (hence measurable)
map rp : VI -+ V2 between two quasi-complete locally convex Hausdorff vector
spaces pushes a Radon measure JL on VI forward to a Radon measure rp*JL on V2;
the definition rp*JL(B) := JL(rp-1 (B» (where B is a Borel set in V2) implies that
for each IE L I (V2, rp"JL) one has

( dJL(v) I(rp(v» = ( dqJ*JL(u) I(u). (3.57)

lv, lvz
In particular, the pushforward of a Gaussian measure on VI is again Gaussian,
with covariance Q", given by Q",(a, fJ) = Q(qJ*a, qJ* fJ). Here rp* : Vi -+ V~ is
the transpose of qJ : VI -+ V2.
We apply this with VI = V and V2 = lR?, with rp : V -+ ]R2 given by rp(v) :=
(O(v), rJ(v». Since 19 and rJ lie in V*, this map is continuous. Applying (3.57) and
11.( 1.49), the Fourier transform of the right-hand side of (3.55) becomes

e-!(ij,ij) ( dJL(v)e iO (v)-1/(v) = e-!Q(1/,1/) { dqJ*JL(x)eiX,-xz. (3.58)

h kz
Inspecting all possibilities, one verifies that

( dv(x)eia(x)-fJ(x) = e-!Qz(a+ifJ,a+ifJ) (3.59)

llRz
for any Gaussian measure v on ]R2 with covariance Q2; in other words, for V = ]R2
equation (3.53) is correct also for complex 0 E Vc' Note that Q2 is seen as a bilinear
(rather than a sesquilinear) form, as we have assumed that 1-£ is real. To compute
(3.58), we now take a = (1,0) and fJ = (0, 1), so that rp*a = 0 and rp* fJ = rJ.
Combining (3.58) and, (3.59), one recovers the right-hand side of (3.56).
Hence (3.55) follows for w E V*. For general W E 1-£ the result then follows by
a continuity argument similar to the one used in proving (3.54). 0
 3 Applications in Relativistic Quantum Theory 413

Key differences with the finite-dimensional situation are that generically one
has l1y(H) = 0, and that the measures l1y tend to be disjoint for different y. In
applying induction methods to the quantization of gauge theories, one is given the
gauge group as a Hilbert Lie group H, but the space V in 3.5.5 has to be guessed,
and is generally not unique. However, any successful choice leads to (3.55) for
W E H, and this turns out to guarantee that the gauge group acts trivially in the
induced space.

Corollary 3.5.6. The inclusion Q ~ Qc (cf 3.4.3) is continuous, and V = Qc

carries a Radon measure satisfying (3.53). Hence for A E Qc and 17 E Q one has

19,
dI11(A)ei(~.A) = e-~1I~1I2; (3.60)

dl1l(A + 17) = e-~1I~1I2-(~.A)dl1l(A). (3.61)

The dual of Qc is Q: = So(lR,4)/Io, where So(lR,4) consists of those f E S(lR,4)

whose Fourier transform / vanishes at 0, and Io is the annihilator in S(lR,4) of
the space of solutions of DA = 0 in S'(lR,4). The first claim then follows from the
estimate I(A, f)1 S IIAII 11]11, where ](p) := /(w p , p)/w~, and 11]11 is defined by
the right-hand side of (3.36). Note that 11]11 < 00 because /(0) = 0, so that the
constant term in the Hermite expansion of / vanishes.
The existence of 111 eventually follows from the fact that Q:
is a nuclear space
(a property it inherits from S(lR,4», so that its dual Qc = Q:* is conuclear. 0

By 3.5.5.1, the right-hand side of (3.52) is meaningful for A E Qc. For

example, (A, aA) stands for -ia(A), where a E Q is defined by (cf. 3.4.3)
a(p) := (Ao(p)/w p ) + PiAi(P)/W~. We then postulate that the counterpart of
(3.47) is

(3.62)

The following theorem closely parallels Proposition 3.5.4, and quantizes Theo-
rem 3.4.4. This time Q; stands for the pertinent subspace of exp(SO.R, v); the spaces
J and P are defined at the end of 3.4.
Theorem 3.5.7. Apply Construction 2.2.5 with £ = Q;, Hx = Il3(H x) = Hid = C,
and define (\II, <I»~ by sesquilinear extension of(3.62).

1. Thisform is finite for \II, <I> E Q;, and satisfies (2.35) and (2.46).
2. The induced space H~ is naturally isomorphic to exp(So), where the classical
reduced space SO ::::::: HO.+. 1 EElHo.+.- 1 is defined in 3.4.4.
3. Each gauge transformation UF(A) is adjointable, and, satisfying (3.48), acts
trivially on the induced s~
4. The pre-C* -algebra Jl'F(W(J» leaves Q; ~, and consists of a:!l!!!!:!able
operators. The induced representative Jl'~(W(J» is isomorphic to W(SO).
414 IV. Reduction and Induction

5. Each (no.+ Y (A, V»F, where (A, v) E P, leaves <E stable and is adjointable.
The induced P-action (no.+. v)} is unitary, and equivalent to the representation
(Uo.+. 1 $ Uo.+.-Il.
6. The automorphic P-action a on W(S) obtained/rom the symplectic P-action
nO.+. v on S restricts to 2l1(.J). In the representation Jr~ this restriction is
implemented by U~.
Compared with the proof of Proposition 3.5.4, the components AL,T of A are
now given in momentum space by
-T 2
A (p) = (0, Ai(p) - PiPjAj(p)/p );
-L 2
A (p) = (Ao, PiPjAj(p)/p ). (3.63)

The integral in (3.62) is computed by transforming the Gaussian integration over

9c to one over lR2 with the aid of (3.57) and the subsequent expression for Q",; cf.
the proof of 3.5.5. Moreover, (3.61) guarantees that (3.49) still holds. Otherwise,
similar steps as in the proof of 3.5.4 lead to a tedious verification of all claims. 0

In view of this theorem we conclude that the induced Hilbert space and the
various representations it carries describes a quantum field theory of photons.

3.6 Classical Yang-Mills Theory on a Circle

In order to explain certain topological features of quantum field theory in the
simplest possible model, we shift our attention from Minkowski space in dimension
3+ 1 to the cylinder lR x SI. Here lR stands for the time axis, whereas the circle
SI represents space. The field theory to be studied comes from a relativistically
invariant model, but this time we are not interested in aspects of special relativity
and the Poincare group. The model will therefore be presented in a partial gauge
fixing (" Ao = 0") that breaks relativistic invariance. We look at the circle S 1 as
the interval [0, 1] with boundary points identified; it is parametrized by a E [0, 1).
Any principal H -bundle over the circle is isomorphic to the trivial bundle P =
SiX H. Recall from the comment after III.(2.l6) that a connection on a trivial
bundle P = Q x His an element of A I(Q)®~. When Q = SI, the space A I(SI)
of smooth I-forms is simply C OO (SI, lR), so that a (smooth) connection on SI x H
is an element of COO(SI, ~). Similarly, according to Proposition III.2.4.2, the space
of smooth gauge transformations may be identified with COO(S', H).
Although the classical theory can be defined on the basis of smooth connections
and gauge transformations, the corresponding quantum theory requires a more
general class. It clarifies matters to include certain nonsmooth connections and
gauge transformations already at the classical level. For the following definition
we equip ~ with an Ad(H)-invariant inner product; by the compactness of H,
this is always possible. The real Hilbert space L2(SI,~) = L2([0, 1]) ® ~ is then
defined with respect to the Lebesgue measure and the above inner product on ~.

Definition 3.6.1. Let H be a compact connected Lie group.

 3 Applications in Relativistic Quantum Theory 415

• The configuration space AIR of classical Yang-Mills theory on a circle with

structure group H is L 2(SI, fJ). The phase space is S = T* AIR .
• The gauge group 9 of this theory is the Hilbert manifold HI (Sl, H), consisting
ofall g E C(SI, H) whose (weak) derivative g := g-Idg Ida lies in L2(SI, fJ).
Elements g of 9 are loops in H, so that 9 is a (Sobolev) loop group of H.
It can be shown that g E 9 is absolutely continuous, and that g exists almost
everywhere. Physically, one could say that 9 consists of all continuous loops with
finite kinetic energy. To obtain an alternative characterization of 9, take a faithful
representation U(H) on some cn; then Wln(C) is a normed space in the usual
way, so that one can define the Hilbert space HI (Sl, Wln(C» as the completion
of COO(SI, Wln(C» in the p = 1 Sobolev norm. The gauge group HI (Sl, G)
is the subset of HI (Sl, Wln(C» consisting of those functions that take values
in U(H). This endows Hl(SI, H) with the structure of a Hilbert manifold. The
continuous inclusion 'Hl(SI, H) C C(SI, H) is then a consequence of the Sobolev
embedding theorem, from which it also follows that HI (Sl, H) is not contained
in any CP(SI, H) for p > O.
The Lie algebra 9 of 9 is H1(St, fJ), with dual g* = HI(St, fJ*). We will often
use the notation
(3.64)
We write elements of S as pairs (E, A), where E E L2(SI, fJ*) and A E AIR.
Using the inner product on fJ, we identify fJ* with fJ, and subsequently identify S
with A = L 2(St, fJc). Rather than 11.(2.28), we use the convention
Z:=A+!iE. (3.65)
This has the advantage that the complex connection Z behaves in the same way as
A under gauge transformations; cf. 3.6.2 below. (Whereas in the previous sections
no confusion was possible, we will now denote complex connections by Z or W,
and real ones by A.)
Lemma 3.6.2. The action Ill. (2. 69) of9 on AIR pulls back to a 9-action on S.
Writing Ad(g)A for the junction a t--+ Ad(g(a»A(a), and similarly for Co(g)E,
this action is given by
g : (E, A) t--+ (Co(g)E, Ag), (3.66)
Ag := Ad(g)A + gdg- t = g(A _ g)g-l. (3.67)
On complex connections (3.65) this reads Z t--+ zg, where zg is defined as in
(3.67). This action is smooth, proper, and strongly Hamiltonian.
The smoothness of this action is a technical exercise in Hilbert manifold theory
that we omit. The main point is that AIR and 9 have been defined precisely so that
g lies in AIR when g E 9; cf. 3.4.4. Properness follows as in 3.4.4. The last point
follows from Lemma III.2.3.1, as usual. 0
The 9-action is not free, unless H is abelian. This may be handled by realizing
that 9 is isomorphic to the semidirect product H ~ p ge, where ge is the subgroup
416 IV. Reduction and Induction

of based loops or based gauge transformations, i.e., loops starting and ending
at e E H. The structure group H acts on ge by
r(h)g : ct ~ hg(ct)h- I . (3.68)
A homomorphism from 9 to H ~ p ge is given by g ~ (g(O), gg(O)-I), with
inverse (h, g) ~ g h. Now the action of ge on AR is free, and Marsden-Weinstein
reduction (at 0) of S by the above 9-action, which is our goal, may be carried out
in two steps; see Corollary 1.10.5. One firstly reduces by the ge -action, yielding
a Marsden-Weinstein quotient that is duly a manifold, and secondly performs
singular reduction by 9 1ge ~ H. As we shall see, the first step yields a finite-
dimensional reduced space, and the second step is easy.
Another interesting feature of Marsden-Weinstein reduction in the present
situation is that 9 may well be disconnected.

Proposition 3.6.3. The group JroW) := 919 0 (where 9 0 is the identity component
of9) is isomorphic to the first homotopy group JrI(H) of H.
To put this in perspective, consider the loop group LH = C(st, H), equipped
with the topology of uniform convergence (with respect to the metric topology of
H inherited from the Riemannian structure, or from H ~ U(H) as above). This
topology coincides with the compact-open topology, so that one has Jro(LH) =
Jrl (H) by definition of Jrt.
The group Jrl (H) is isomorphic to a discrete subgroup D of the center of the
universal covering group H of H (i.e., H = HID). Under this isomorphism an
element [8] E Jrl (H) is the equivalence class of loops in H that are homotopic
to the projection (from H to H) of a path from e to 8 in H. We thus label the
components L H [8] of L H by 8 ED. Since the inclusion 9 c L H is continuous
with respect to the manifold topology on 9, the proposition will follow if each
intersection 9[8] := 9 n LH[8] is connected in the topology of 9. By the reasoning
in the previous paragraph, this follows from the obvious fact that any two paths
with finite kinetic energy in H between e and 8 are homotopy-equivalent in the
topology of Ht (S', H). •
For later use, we infer from this proof that
LH!e] := LHe n LH[e] ~ LHe, (3.69)

where the based loop group LHe is defined similarly to ge.

Elements of 9[e] are called small gauge transformations. whereas members
of the other 9[8] are large gauge transformations. For example, if H = U (1), it
follows that

JroW) = Jro(LU(l» = Z. (3.70)

The members of a given component 9[n], nEZ, are labeled by the winding number
of the loop. An example of an element of 9[n] is, of course,

(3.71)
 3 Applications in Relativistic Quantum Theory 417

The fact that 9 is disconnected when 11", (H) is nontrivial is important mainly in
the quantum theory of this model, but even in the classical theory it is enlightening
to calculate both s~ and S~; cf. (1.26). We put 9Je] := ge n 9[e].
Lemma 3.6.4. Under the 9-action 1/1.(2.69) on AIR, there are the diffeomorphisms
AIR/9!e] ~ iT, AIR/ge ~ H, and AIR 19 ~ HI Ad(H).
We shall first sketch the proof of the second diffeomorphism; the first is then
obvious from (3.69). We define a map W: AIR --+ C([O, 1], H) as the solution of
the differential equation (valid for almost every a)

(a~ +A)WA(a)=o, (3.72)

with initial condition WACO) = e; we have written WA for W(A). According to

the theory of product integration, the solution is absolutely continuous in a, and
may be written as a uniform limit

WA(a) = J~moo TI Exp [PA (0 - n ~ l)a) - PA ((1- ~)a) J. (3.73)

Here PA(a) := foa df3 A(f3) is the primitive of A.

Simple manipulations show that the function gWA : a t--+ g(a)WA(a) on S'
satisfies d(gWA)lda = A8 gWA, where A8 is given in III.(2.69) or 3.6.2. When
g E ge one has the initial condition gWA(a) = e, so that gWA = W AK . Since
g(l) = g(O) = e, this implies WA(l) = WAK(l). Accordingly, the Wilson loop

(3.74)
is invariant under based gauge transformations. In other words, for all g E ge one
has W(Ag) = W(A).
Suppose that W(A) = WeB). Then g := WAW;' lies in ge. Using (3.72), one
sees that dglda = -Ag + gB; in other words, A = Bg. Noting thatfor compact
connected groups Exp is surjective, we conclude that A t--+ W(A) induces a
bijection from AIldge to H. It is a nontrivial technical task to prove that this
bijection is a diffeomorphism; we omit this part of the proof.
The third diffeomorphism follows from the isomorphism H ~ 91ge (by the
discussion following the proof of 3.6.2) and the intertwining property
W(Ad(h)A) = hW(A)h-'. (3.75)
Here h E H, seen as a subgroup of 9. This is the case a = 1 of the property
WAd(h)A(a) = hWA(a)h-', which is immediate from (3.72). 0
Hence a gauge-invariant function of A E AIR is a function of W(A).
Theorem 3.6.5. An equivariant momentum map for the 9-action (3.66) is given
by
(3.76)
418 IV. Reduction and Induction

where A. E 9 and DAE := dE + [A, E].

There are symplectomorphisms
rl(O)/g!e] :::::: T* if; (3.77)
rl(O);ge :::::: T* H; (3.78)
rl(O)/g:::::: T*(H/Ad(H». (3.79)

The d in dE above is a weak derivative; the integral (d E, A. )AR is well-defined,

since dA. lies in L 2(SI, I). Similarly, ([A, E], A.)AR is well-defined because A and
E are in L2(SI, I) and A. E C(SI, I).
In somewhat symbolic notation, the Poisson bracket on COO(T* AIR, JR) is

8f 8g ) (8 f 8g ) (3.80)
{f,g} = ( 8E' 8A AR - 8A' 8E AIR'

One derives from (3.66) that the generator of A. is given by

1;>./ = (88Ef , CO(A.)E) AR

- (88Af , DAA.) AIR
. (3.81)

Combining (3.80), (3.81), 111.(1.7), and 1.(2.8), one verifies (3.76).

The symplectomorphisms all follow from I1I.(2.S6), which is valid also for
Hilbert manifolds, and Lemma 3.6.4; see below for comments on the potentially
singular third case. •

An explicit expression for the symplectomorphism (3.78) may be derived as

follows. By (3.76), the condition (E, A) E J-I(O) forces DAE = 0; in physics
this is seen as the Gauss law constraint of Yang-Mills theory; for abelian Hone
simplyhasdE = O. Gauss's law implies thatE(a) = Co(WA(a»E, where E E 1)*
is independent of a.
This suggests that it is convenient to use the variables (E, A), where

(3.82)

Define a map fPs : T*AJR -+ T*H by fPs(E, A) := (EI' W(A», where EI :=

101 da E(a), and T* H is identified with 1)* x H through the left trivialization.
As in the proof of (3.6.4), one verifies that on J-I(O) this map quotients to a
diffeomorphism from J-'(O)/ge to T* H, and checks from (3.80), and III.(l.S4)
that this diffeomorphism is a Poisson map.
We may implement (3.78) in a more elegant way (which will be essential in
quantum theory) on the basis of the following notion.

Definition 3.6.6. Let H be a compact connected Lie group. The complexification

He of H is the unique Lie group that contains K as a closed subgroup, and has
the property that for any complex-analytic Lie group K and real Lie group ho-
momorphism 1/1 : H -+ K there exists a unique complex-analytic homomorphism
1/Ie : He -+ K that on H (regarded as a subgroup of He) coincides with 1/1.
 3 Applications in Relativistic Quantum Theory 419

The Lie algebra of He is the complexification I)e of I) (avoiding a potential

ambiguity of notation); when H is simply connected, He is the connected and
simply connected Lie group whose Lie algebra is I)e. For H = U(l) one has
He = C*. It can be shown that H is always a maximal compact subgroup of He,
and that every finite-dimensional representation of H extends to a holomorphic
representation of He. In the present context the following property of He is crucial.
Proposition 3.6.7. The cotangent bundle T* H is diffeomorphic to He.
Identifying T* H with 1)* x H (in the left trivialization), and subsequently
equating 1)* with I) through the Ad-invariant inner product on I), it turns out that
<PH : (X, h) f-+ hExp(-iiX) is a diffeomorphism from T* H to He. 0
This diffeomorphism equips He with a symplectic structure.
The differential equation (3.72) still makes sense when A E AIR is replaced by
a complex connection Z E A. Thus the Wilson loop map W : AIR ---+ H may be
analytically continued to a map We : A ---+ He.
Proposition 3.6.8. The map We, restricted to J- l (O) (seen as a subspace of A
through the identification of A with T* AIR explained after (3.64)), quotients to a
symplectomorphismJrom J- l (O)/Ye to He.
We regard E(a) as an element of I) rather than of 1)* (cf. 3.6.7); replacing Co
in (3.82) by Ad, this applies to E as well. Given (E, A) E T* AIR and Z E A,
related to (E, A) by (3.65), consider the functions f, g : [0, 1] ---+ He defined by
f(a) := WA(a)vh;i(a) and g(a) := Wz(a). Using (3.82), with Co replaced by
2
Ad, and (3.72), one verifies that f and g satisfy the same first-order differential
equation. Since f(O) = g(O) = e, it follows that f(1) = g(l). On J-1(0) the
function E is constant, so that W!;£(1) = Exp(-iiEd. Hence
2

f(l) = W(A)Exp(-iiEd = g(l) = WdZ).

The claim then follows from the proof of 3.6.7 and the paragraph below
(3.82). •

Note that (3.75) with A replaced by Z reconfirms (3.79). When H is nonabelian,

the space H / Ad(H) is not a manifold; the cotangent bundle is then defined as
follows. Let Ad* be the pullback to T* H of the adjoint action of H on itself; in a
trivialization T* H ~ 1)* x H this is given by Ad*(h) : (e, k) f-+ (Co(h)(), hkh- l ).
The momentum map JAd for Ad* is given by Jtd(a) = a(~: - H). One then
puts T*(H / Ad(H» := (J Ad )-I(O)/ Ad*(H) as a topological space; away from the
singularities of H / Ad(H), this is the usual cotangent bundle.
The structure of T*(H / Ad(H» may be described with the theory in 1.11. We
will not do so here, other than saying that by Lemma III. 1. lOA, the space H / Ad(H)
is homeomorphic to the StietTel chamber T / W. For example, for H = S U (2)
the latter is the closed interval [0, 1], so that the singularities are merely boundary
points. This is true in other cases as well. In the abelian case this phenomenon
does not arise; for example, for H = U(l) we simply have S~ ~ T* SI.
420 IV. Reduction and Induction

In any case, Theorem 3.6.5 is spectacular, showing that the reduced phase space
of a particular field theory is finite-dimensional. In the present model this feature
is peculiar to the one-dimensionality of space.

3.7 Quantum Yang-Mills Theory on a Circle

We now tum to the quantum theory of the model in the previous section, expecting
to see that quantum induction naturally leads to L 2(H) or L 2(H / Ad(H».
The CCR-algebra 2!J(A) is defined as in 3.5.1. The gauge group 9 acts on 2!J(A)
by automorphisms, defined by (extension of)
(3.83)
The analogue of the gauge-invariant algebra 2!J(.J) in the quantum field theory of
photons is now the sub algebra 2!J(A)9 of 2!J(A) consisting of those B E 2!J(A)
for which Clg(B) = B for all g E g.
Proposition 3.7.1. There exists a representation UF(Q) of the gauge group on
the Fock space exp(A) that on exponential vectors takes the form

(3.84)

and implements the automorphism (3.83) in the Fock representation Jr F(2!J(A».

Here and in what follows, inner products of the type (g, Z) are in A. After the
proof of 3.7.3 below we show that (3.84) is indeed the restriction of an operator
on exp(A). Granted this, unitarity follows from 11.(2.63), the representation prop-
erty comes from (3.67), and finally (3.83) with (3.43) leads to the implementing
property JrF(Clg(W(Z))) = UF(g)JrF(W(Z»UF(g)*. 0
As in (3.46) (combined with (3.43», we see that U F performs a gauge transfor-
mation on the argument of .y'EXp, with additional factors guaranteeing unitarity.
It is not of the form (3.46), because the automorphisms (3.83) are not inner.
Apart from the fact that it has led us to (3.84), which plays a central role in what
follows, the CCR-algebra is hardly of any use in quantum Yang-Mills theory. This
is because it fails to contain such crucial observables as the quantized functions of
the Wilson loop. In order to construct the latter, as well as to verify that (3.84) is
well-defined, we return to the setting of Definition 11.1.5.6 and Theorem 3.5.5.
Proposition 3.7.2.
1. The obvious inclusion of 1t = AIR = L \0, 1) ® IJ into the Schwartz space
V = V'(O, 1) ® IJ is continuous.
2. The ensuing map from (V'(O, 1) ® IJ)* = V(O, 1) ® IJ to AIR, defined after
Il.l.5.6, is the natural inclusion.
3. The assumption of Theorem 3.5.5 is satisfied.
4. The support of J..iy is contained in the set of distributions that are the (weak)
derivative of a continuous function.
 3 Applications in Relativistic Quantum Theory 421

The first claim is immediate from the Cauchy-Schwarz inequality, and the second
is obvious. The third claim follows because V'(O, 1) ® ~ is conuclear. The fourth
statement is a well-known property of white noise, whose proof we omit. 0

We may, accordingly, construct the Hilbert space L2(V'(0, 1) ® ~,JLy). The

map A 1-+ Ad(g)A may be extended from AR to V'(O, 1) ® ~ by dualizing its
restriction to 1)(0, 1) ® ~ C AIR, so that the gauge transformation (3.67) applies
to L 2(1)' (0, 1) ® ~, JLy) as well. For each g E g we then define an operator U y (g)
on L 2 (V'(0, 1) ®~, JLy) by

Uy (g)'l1(A) := e[ -~lIgIl2_!(A.dgg-I)]/y 'l1(AK- 1 ). (3.85)

It follows from the uniqueness of a measure satisfying (3.53), and the invariance
of the inner product in AIR, that JLy is invariant under A 1-+ Ad(g)A. Using this
in conjunction with (3.55), which applies in view of Definition 3.6.1, one shows
that Uy(g) is unitary. It is then easily checked that (3.85) defines a representation
of the gauge group g on L 2 (V'(0, 1) ®~, JLy).

Lemma 3.7.3. There exists a unique unitary operator Vy : exp(A) -.

L2(V'(0, 1) ®~, JLy) that maps JEXp(Z) to

(3.86)

Here (A, Z) is defined as in 3.5.5.1. The uniqueness of Vy is clear, as the expo-

nential vectors are total in exp(A). The fact that Vy is well-defined follows from
an alternative expression. Note that the linear span of all vectors of the form ®n Z,
where Z E A with II Z II = 1, and n EN, is dense in exp(A). We put

Vy ®n Z : A 1-+ ~Hn (y-I/2(A, Z») , (3.87)

vn!
where Hn is a Hermite polynomial. This operator is clearly well-defined, and
easily shown to be bounded, hence extendible to exp(A). A somewhat lengthy but
elementary computation then shows that Vy = Vy. 0

Since one computes from (3.84)-(3.86) that VI/4UF(g)Vij! = UI/4(g), it is

clear that (3.84) is well-defined.
We now tum to the construction of the quantized Wilson loop. Using Propo-
sition 3.7.2.4, it can be shown that the limit in (3.73) exists for almost every
A E V'(O, 1) ® ~ with respect to JLy. In particular, the Wilson loop W(A) is
well-defined for such A.

Definition 3.7.4. Let f E COO(H, R), and define WI E COO(AIR) by

W/(A):= f(W(A». (3.88)

The observable QI/4(WI) on L 2(V'(0, 1) ® ~, JLI/4) is the operator defined by

(3.89)
422 IV. Reduction and Induction

On exp(A) one has the operator

QF(Wf) := ViI! Q1/4(Wf) V1/4. (3.90)

To understand the choice y = 1/4, consider the abelian case. When H = U (1),
identified with l' C C, the (classical) Wilson loop is a numerical function on AIR
(or on S), which coincides with the function A t-+ exp(-A1). Here A1 := 1 A; J0
more generally, for I E AIR one has the linear functions

Af(E, A) := (A, f); (3.91)

Ef(E, A) := (E, f). (3.92)

Note that {Ef' Ag} = (f, g) by (3.80). These linear functions are quantized on
Fock space exp(A) by

QF(A f) := ! (a(f) + a(f)*) ; (3.93)

QF(Ef):= -i (a(f) - a(f)*). (3.94)
It follows from 11.(2.68) and the above Poisson bracket that Dirac's original con-
dition i[QF(Ef)' QF(A f )] = Q({Ef , Ag}) is satisfied. For H = U(1) we define
Q1/4(W) as in (3.89), with the function I simply omitted (or taken to be the iden-
tity map), and subsequently define QF(W) by (3.90). A partial justification of
Definition 3.7.4 then lies in the equality
QF(W) = e-QF(A 1). (3.95)

This may be derived on a vector y'EXp(Z) by use of (3.93), the CBH-fonnula,

11.(2.68),11.(2.69),11.(2.70),3.7.4, and (3.86).
Our aim is to quantize Theorem 3.6.5. Since the motivation for the steps to
follow is similar to that for the steps leading from Theorem 3.4.4 to 3.5.7, we will
not repeat the arguments given there. The counterpart of (3.52) for Yang-Mills on
the circle is found, from (3.84), 11.(2.63), and (3.67), to be

(VEXP(W), UF(g)VEXP(Z») = e-!lIgI1 2 e(W,z8)+(g,Z). (3.96)

In the situation of Definition 11.1.5.6, we take 1t = AIR and V = C([O, 1], ~)o,
seen as a Banach space in the sup-nonn; the suffix 0 indicates that V consists of the
continuous paths in ~ satisfying X(O) = O. Here AIR is seen as a Hilbert subspace
of C([O, 1], ~)o by identifying it with the set of all X E C([O, 1], ~)o whose nonn
with respect to the inner product

(X, Y)1 := 11 da X'(a)Y'(a) (3.97)

is finite. Equivalently, AIR is injected into C([O, 1], ~)o by the primitive mapping
A t-+ 'P A defined below (3.73). The pertinent analogue of Corollary 3.5.6 is then
as follows.
Proposition 3.7.5. The inclusion AIR ~ C([O, 1], ~)o by'P is continuous, and
lor each y > 0 there is a Radon measure JL~o on C([O, 1], ~)o satisfying (3.53).
 3 Applications in Relativistic Quantum Theory 423

Since PA(a) = (A, X[O.al), one has IIPAlioo ::: IIAII by the Cauchy-Schwarz
inequality. The second claim, whose proof we omit, defines the Wiener measure
on C([O, 1], £)0 with variance y. 0
Our interest, of course, lies in a suitable measure on a completion of the gauge
group 9 in some topology, rather than of its Lie algebra AIR. For this completion
we take the loop group LH = C(SI, H). Assuming that H C mtn(C) for some
n, this loop group is seen as a submanifold of the Banach space C([O, 1], VJ'tn(C».
The same comment applies to its subspace C([O, 1], H)e of continuous paths that
start at e. We define Ito's map i : C([O, 1], £)0 ~ C([O, 1], H)e by

'Ix(a) :=
A
!] +
J~oo N-I Exp [X((1 - ---,;-)a
• n 1) (
- X (1 - N)a
n)] . (3.98)

It can be shown that the limit exists for almost every X with respect to IL~o. The
image of IL~o under Ito's map is the Wiener measure IL~' on C([O, 1], H)e with
variance y. Ito's map is a bijection up to null sets of IL~o and IL~" Comparing
(3.98) with (3.73), it is clear that i 0 P = W; cf. the comment preceding 3.7.4.
Hence the Wiener measure IL~' may equivalently be defined as the image of the
measure ILy on'D'(O, 1) ® £) under W; see Proposition 3.7.2.
Let C([O, 1], H)~ be the space of continuous paths g in H for which g(O) = e
and g(l) = h. Each such space carries a Radon measure IL~:' characterized by
the disintegration property

1 C([O.II.H).
dIL~' f(y) = J 1
H
dh
e([O.II.Ht.
Wh
dILy' f(y) (3.99)

for all f E LI(C([O, 1], H)e, IL~e). In particular, this assigns a measure to the
based loop group LHe = C([O, 1], H):. As to the loop group LH itself, as in the
case of 9 (cf. 3.6), we can write LH = H ~ p LHe as groups.
Definition 3.7.6. Writing LH ~ H x LHe as Borel spaces, the Wiener measure
on LH with variance y is the direct product IL~ = ILH x IL~: ofthe Haar measure
Wh
ILH on H and the measure ILy' on LHe.

Note that neither IL~: nor IL~ is a probability measure, unlike IL~o and IL~"
The behavior of all these Wiener measures under translations follows from the
general Cameron-Martin formula (3.55) for IL~o (which applies because of 3.7.5)
and the (almost sure) bijectivity of Ito's map. For example, for X E HI (SI, H)
one obtains
dIL~:X(I) (g X) = e-[!IIXIl 2 +(g.Ad(X)X)]IY dIL~: (g), (3.100)

where the second term in the exponential is defined by Theorem 3.5.5.1 and Propo-
sition 3.7.2. The Radon-Nikodym derivative d IL~ (g X) / d IL~ (g), where once again
X E HI (SI, H), will be equally important; it is given by the same expression. The
translate of IL~ (etc.) by X ¢ HI (SI, H) is singular with respect to IL~.
424 IV. Reduction and Induction

After this intermezzo we return to Yang-Mills theory. Motivated by (3.96)

and the surrounding discussion, we define (, )~ on (!! C exp(A) by sesquilinear
extension of

(3.101)

The expressions of the type (g, Z), which in (3.96) were well-defined for g E
HI(SI, H) as inner products, make sense for general g E LH by 3.5.5.1 with
3.7.2; cf. (3.100) and subsequent comment. Indeed, the justification of (3.101) lies
in the property (3.49), with A replaced by g; this follows from the counterpart of
(3.100) for JLf. As in 3.5.7.3, it will follow that the gauge group 9 is trivially
represented in the induced space (to be defined in Theorem 3.8.1 below). Since the
Cameron-Martin formula, and therefore (3.49), is not valid for all h E LH, the
choice of the gauge group in 3.6.1 (instead of LH) has hereby to some extent been
justified. In this connection it is worthwhile to remark further that the representation
U F in (3.84) cannot be extended from 'HI (Sl, H) to C(SI, H).

3.8 Induction in Quantum Yang-Mills Theory on a Circle

We now come to our main result on quantum Yang-Mills theory on a cylinder.
The context of the following theorem is the same as in 3.5.7, and the notation is
similar. We write H'} for the induced space and n id for the corresponding induced
representation; this convention is used in order to be able to write n:(A) :=
nid(QF(A» for the quantization of some observable given as an operator QF(A)
on exp(A).
Theorem 3.8.1. Apply Construction 2.2.5 with t = (!! c exp(A), 'Hx
~(Hx) = Hid = C, and define ('l1, <l»~ by sesquilinear extension of(3.101).

1. Thisform isfinite on (!!, and satisfies (2.35) and (2.46).

2. The induced space Hit is naturally isomorphic to L 2( H)Ad(H); this is the
subspace of L 2(H) that is invariant under UAd(H), defined by
UAd(k)'l1(h) := 'l1(k- l hk). (3.102)
3. For each g E 9 the gauge transformation U F(g) is adjointable, and acts trivially
on the induced space.
4. Let f E COO(H) be a class junction, and recall (3.90). The operator n:(Wf )
is the restriction of the multiplication operator f on L2(H) to L 2(Ht d(H).
5. Replacing LH by LHe in (3.101) yields an induced space 'Hit,e that is natu-
rally isomorphic to L 2(H). For all f E COO(H) one has n~e(Wf) = f as
multiplication operators on L 2(H). .
6. Replacing LH by LHrlleads to the induced space L2(lf).
The first item is proved as in 3.5.7, granted that a map [; may be constructed; this
will be done below. We compute the integral in (3.101) in two steps. According to
Definition 3.7.6, the integration over LH factorizes into an integral over of LHe,
and an additional integration over H. We start with the former.
 3 Applications in Relativistic Quantum Theory 425

The computation will be based on the connection between the Wiener measure
and the heat equation on H. For fixed y > 0, this is the PDEdfldt-!y L\H f = 0,
where L\H := Li T? is the Laplacian on H. Here the orthonormal basis {T;} of
~, seen as a set of right-invariant vector fields on H, is defined with respect to the
Ad(H)-invariant inner product on ~ already used in the definition of L2([0, 1], ~);
the Laplacian depends on this inner product, but not on the basis. It can be shown
that the fundamental solution Py is in COO(H) for t > 0.
Givenk Borel sets BI, ... , Bk in H, and k elements al, ... , ak in [0,1], define
a class of subsets of C([O, 1], H)e by
C~I:.:'.:~k := {g E C([O, 1], H)e I g(al) E BJ, ... , g(ak) E Bd. (3.103)
Using the theory of stochastic processes, it can be shown that

/L~' (C~I:.:::~k) = n1
k

i=1 Bi
dh i
k
npy(hjhj~l'aj
j=1
-aj-I), (3.104)

(3.105)

where, of course, ak := 1 and hk := h. Therefore, one has

Wh (
/Ly' C([O, 1], H)eh) = py(h, 1) = PI(h, y). (3.106)
Lemma 3.8.2. The integral (3.101) over LHe yields

1 Life
d/L~: (g)e(W,ZS)+(g,Z) = e![(W,W)+(Z,Z)]PI(WdZ)-IWdW), 1). (3.107)

To derive this result, one starts with Z, W E AIR. Then substitute the identity
Z = -(dWz/da)WZ- 1 (which is immediate from (3.72», and use (3.100) with
X = Wz.Intermsofthenewvariableg = gX theintegraiisoverC([O, 1], H),{,,(Z),
since X(O) = e and X(I) = Wz (l) = W(Z). Perform the transformation g t-+
g-I, under which the Wiener measure is invariant. The integral is now over the
space C([O, 1], H),{,,(Z)-I. Repeating the above trickfor W, and using (3.106), leads
to (3.107). Of course, at this stage one has W = W and Z = Z.
We now use the nontrivial fact that the fundamental solution Py (', t) of the
heat equation on H has a unique analytic continuation to He (containing H as a
subgroup; see 3.6.6). Since we know from 11.(2.62) and (3.101) that the left hand
side of (3.107) is analytic in Z and antianalytic in W, and that the complexified
Wilson loop We : A --+ He is analytic, the result follows. (Alternatively, one may
extend the Cameron-Martin formula by analytic continuation in X). 0
We refer to the induced space defined by (3.101), with LH replaced by LHe ,
as 7-l~,e' with corresponding induced representation 7r~,e of the set of adjointable
operators on IE. We now wish to apply Proposition 2.2.4. In order to put the result
in a neat form, we define an appropriate analogue of the coherent states 11.(2.47)
426 IV. Reduction and Induction

for compact connected Lie groups. We formulate this in terms of He rather than
T* H; cf. Proposition 3.6.7.

Definition 3.8.3. Let H be a compact connected Lie group. For z E He, the Hall
coherent state ~~ in L2(H) is the function h 1--+ p/i(h-1z, 1), where Py is the
analytic continuation of the fundamental solution of the heat equation on H.

The linear span of all ~~ (as Z varies in He) is dense in L 2(H). The Hall coherent
states are not normalized, and mayor may not provide a pure state quantization
of T* H in the sense of 11.(1.3.3); in the present context this is entirely irrelevant.
Note that p/i(h-'z, 1) = Pl(h-1z, Ii).
Our guess for 1t},e is L 2(H); this plays the role of 1t~ in 2.2.4. We define
Ue : ~ ~ L2(H) by linear extension of
UeJExp(Z):= e~(Z,Z)~m(Z)· (3.108)

Equation (2.29) follows from (3.107) and the identity

py(h, 1) = L dk py(hk- 1, Vpy(k, ~). (3.109)

This identity follows from the definition of the heat kernel, as well as from the
properties py(hk, t) = py(kh, t) (from the Ad-invariance of the Laplacian 6.H)
and py(h- 1, t) = py(h, t) (from the invariance of 6.H under 1'; 1--+ -Tj). One
should compare (3.108) with its classical analogue 3.6.8.
The definition of the Wiener measure JL~: on L He and the invariance of the inner
product on AIR under Ad(H) imply that JL~: is invariant under the (outer) automor-
phisms r(h) defined in (3.68). It follows that for h E H, identified with a constant
function on Sl, the operator U F(h) on exp(A) is adjointable with respect to the
inner product (3.101), with LH replaced by LHe. By (3.84) and (3.64) one sim-
ply has UF(h).JEXi)(Z) = .JEXi)(Ad(h)Z), as h = O. Using (2.30), (3.108), and
the Ad(H)-invariance of the complexified heat kernel, the representation U F(H),
induced to L 2(H), is equal to UAd; see (3.102).
To complete the induction with respect to (3.101), the integration over H men-
tioned in the second paragraph of the proof must still be performed. This is done
at the present stage, and is a special instance of quantum Marsden-Weinstein re-
duction (itself a special case of Rieffel induction; see 2.5). We are in the setting of
Theorem2.5.1,with1t = L2(H)andU = UAd. Inducing from the trivialrepresen-
tation of H, Proposition 2.5.3 leads to the induced space L 2 (H)Ad(H) announced
in 3.8.1. This proves 3.8.1.2.
The claim 3.8.1.3 follows from the Cameron-Martin formula (3.100) for JLf,
as we explained at the end of 3.7.
To prove 3.8.1.5, which immediately implies 3.8.1.4, we first note that 3.8.3,
(3.106), and (3.100) imply that Ue in (3.1 08) may be rewritten as

(3.110)
 3 Applications in Relativistic Quantum Theory 427

Now recall Lemma 3.7.3. By (3.86), with y = 1/4, and (3.110), the action of the
operator Ue.l/4 := Vl/4Ue Vi/~ on '11 E Vl/4~ is given by

Ue.l/4'11(h) =( I dJ.L~r (g)'I1(g). (3.111)

JC([O.II.Ht.-

Hence by (2.30), (3.90), and (3.88) one has

JT!d(QF(Wt»Ue,I/4'11(h) = 1C([O.I].Ht.-
I dJ.L~r (g) f(W(g»'I1(g). (3.112)

However, one rapidly derives from (3.74) and (3.72) that W(g) = g-I(1). Since
g(l) = h- I by definition of the space over which one integrates in (3.112), we
conclude that f(W(g» = f(h).1t follows that
JT;d(QF(Wt »Ue.I / 4'11(h) = f(h)Ue. I / 4'11(h).

Since the vectors Ue. I /4'11 are dense in L2(H) and f is bounded as a function,
hence as a multiplication operator, this proves 3.8.1.5.
If in (3.101) one integrates over LHJe1 rather than LHe, the isomorphism (3.69)
shows that the above argument still goes through, with H replaced by its universal
covering group H. It follows from the structure theory of compact Lie groups that
H is the direct product of Rm and a compact group; in the former, one uses the
coherent states 11.(2.47), and in the latter, one employs those of Hall.
This proves 3.8.1.5, concluding the proof of Theorem 3.8.1. •

3.9 Vacuum Angles in Constrained Quantization

In this final section we will explain why the discussion so far of the quantization of
Yang-Mills theory on a cylinder has been incomplete. More generally, whenever
the classical reduced space is given as a Marsden-Weinstein quotient with respect
to a disconnected group H, there turns out to be a certain freedom in the induction
process quantizing the reduced space.
For simplicity, we initially assume that H is a finite-dimensional unimodular
Lie group, acting on a symplectic manifold S in strongly Hamiltonian fashion,
with equivariant momentum map J : S -+ 9*. Recall that HO is the component
of H containing the identity e, and that JTo(H) := HI H O• One then considers the
reduced spaces S~ = J-1(0)1 H and S~ = J-1(0)1 H O• By (1.26) the former may
be obtained by a two-step reduction process: one firstly reduces S by H O, which
results in S~, and secondly reduces S~ by JTo(H), which yields S~ ~ S~/JTo(H).
Note that the incompletely reduced space S~ is already symplectic.
We now abstract the second step, and consider the reduction of a general sym-
plectic manifold So by a discrete group D whose given action on So consists of
Poisson maps. We regard D as a zero-dimensional Lie group, with Lie algebra {O}.
The momentum map J of the D-action may then be thought of as being identically
zero, so that J-1(0) = So. The reduced space is then simply Sol D. In particular,
there are no constraints.
428 IV. Reduction and Induction

Following our general strategy, we wish to quantize SolD by Rieffel (or Fell)
induction. Hence we assume that we have a unitary representation Uo(D) on some
Hilbert space Ho that "quantizes" the D-action on So. Although on the classical
side D possesses only the trivial coadjoint orbit {O}, on the quantum side it will have
nontrivial irreducible representations. At first sight, one should Rieffel-induce from
the trivial representation of D, but in the absence of classical constraints there is
actually no good reason not to induce from an arbitrary (irreducible) representation
Ue(D) defined on a Hilbert space He.
The induced space Hg is constructed as in Proposition 2.5.3, amended if nec-
essary when D is not compact (cf. the end of 2.5). One chooses a suitable dense
subspace £ c Ho (when D is compact one may take £ = Ho), and considers the
sesquilinear form on £ ® He defined by
(W, <i»g = })w, Uo ® Ue(8)<i»'Ho®'H8; (3.113)
&ED

cf. (2.81). Quotienting by the null space of this form and completing then leads
to the induced space Hg in the standard way. This space carries an induced repre-
sentation 77:g of the algebra of weak observables mw of the model in question; see
2.9.1. The algebra of observables
(3.114)
in general explicitly depends on () E D, even when Ue is one-dimensional.
For example, take a not simply connected Lie group G with universal covering
group G, so that G ~ G 177:1 (G). Here D = 77:1 (G) is a discrete subgroup of the
centerofG. The cotangent bundle T*G is then symplectomorphicto (T*G)I77:I(G),
where 77:1(G) acts on T*G by pullback of its action on G. Physically, this describes
a particle moving on Q = G, with phase space S = T*G.

---
To quantize, we take Ho = L 2(G), on which Uo (77: 1(G» acts as the right-regular
representation; that is, Uo(8)'II(x) := 'II(x8). We now choose a () E 77:1(G), and
realize that we are in the situation discussed in the paragraph containing (2.97),
with P = G, £ = C~(G), H = 77:1(G), and Ux = Uo. It follows from that
discussion that the induced space Hg obtained from the induction process on
L2(G) is isomorphic to the Hilbert space 1i() carrying the representation U() of G
that is Mackey-induced by U()(77:I(G».
In the realization 1i~ given by a section s : G 4 G (cf. the text after III.(2.162»,
the induced space is simply L 2(G)®Ho. The corresponding induced representation
77:!(G) is then given by 111.(2.176). Moreover, according to Corollary 2.7.2 the space
1{() carries an irreducible representation 77:() of the action C*-algebra C*(G, G).
Conversely, every irreducible representation of C*(G, G) is equivalent to one of
this form.

---
Noting that C*( G, G)]R may be thought of as the quantum algebra of observables
of a particle moving on G, we see that a quantum particle moving on a not simply
connected Lie group has a family of superselection sectors labeled by 77: 1(G).
For a simple illustration of this scheme, we take G = Rand D = 277:Z, so
that G = U (1). This is the setting for a particle moving on a circle. Note that the
 3 Applications in Relativistic Quantum Theory 429

group U (1) plays a double role: It is the configuration space Q of the particle,
as well as the unitary dual Z. A configuration space variable will be called {3,
whereas an element of Z is denoted by O. Both variables take values in [0, 21r).
The representation Ue(Z) corresponding to 0 E Zis
Ue(n) := eine , (3.115)
defined on ?te = C. Hence we obtain a family of induced representations U eOR)
and JTe(C*(IR, U(1))), the latter irreducible. As recalled above, these representa-
tions may be realized on the Hilbert space rtf,
which is the same for all 0 and
equal to L2(U(I». We choose the section s : U(I) -+ IR to be s({3) = {3. From
III. (2. 176) the explicit form of U:
(1R) is
U:(2JTn + {3/)"':({3) = eine "':({3 - {3/) (3.116)

when {3 - {3' E [0, 2JT); in the case that {3 - fJ' E (-2JT, 0) one has
(3.117)
The corresponding representation JT!(C*(IR, U(1») is most easily described
through Corollary III.3.7.4 and Theorem III.3.4.4. With U!(IR) given above, it
remains to state the representation ir!(Co(U(1))). By III.(3.87) and subsequent
comment, this is given by
(3.118)
It follows from (3.116), (3.117), and III.(1.69) that the associated representation
d U: of the generator T of IR is

(3.119)

The O-dependence of this operator lies in its domain; by Proposition III. 1.5.6 one
should initially define Pe on the space of smooth vectors for U: (1R), on which it
is essentially self-adjoint. It is a simple technical matter to show that the domain
of the self-adjoint closure Pe of Pe thus defined is
V(jj9) = Ve := {'" E AC([O, 2JT]) I "'(2JT) = e-ie",(O)}, (3.120)
where AC stands for the space of absolutely continuous functions.
Physicists like to see the O-dependence of Pe in the explicit form of the operator.
This is achieved by the unitary transformation Ve : L2(U(1» -+ L 2(U(I»,defined
by Ve"'(fJ):= exp(ifJO/(2JT»"'(fJ). One obtains

-v:-1 = -.1
V.ePe d - -0- (3 . 121)
(J dfJ 2JT'
which is self-adjoint on the domain Vo; cf. (3.120).
Comparing (3.121) with the classical covariant momentum III.(2.83), one is
tempted to interpret the term 0/ (2JT) as an external electromagnetic potential A.
This interpretation is correct, and provides a physical realization of the superselec-
tion sector o. The fact that the particle undergoes scattering despite the fact that the
430 IV. Reduction and Induction

field strength F of the potential in question vanishes is called the Aharonov-Bohm

effect. At a time when it was thought that only the field strength (rather than more
general gauge-invariant functions of the potential A) was an element of "physical
reality", this effect was considered very surprising. (The Wilson loop defined by
this A is nontrivial.) The truly surprising feature of the Aharonov-Bohm effect is
its periodicity in e with period 2Ir. In the above description this is an immediate
consequence of the periodicity of (3.115).
After this intermezzo we return to the main theme. The discussion at the be-
ginning of this section suggests that the space S~ admits a family of inequivalent
quantizations 'Hg, where () E ;;(ii). These are defined as follows. As always,
one starts with a representation U (H) on a Hilbert space 'H that quantizes the
given H -action on S. Then take a representation Uo of Iro(H) on 'Ho, leading to a
representation Uo(H) by
(3.122)

Rather than inducing from the trivial representation of H, one now induces from
Uo(H). As we have seen, classical Marsden-Weinstein reduction from a discon-
nected group may be split into two steps. The quantum induction procedure may be
split up in a similar way. In the first step one induces from the trivial representation
of HO. This is done by putting the form

(\II, <1»8:= { dh (\II, U(h)<1» (3.123)

JHO

on some domain t ~ 'H and constructing the induced space, now called 'Ho, and
the induced representation Iro as usual. The operator U(h) is adjointable for all
h E H (and not merely for all h E HO). Since HO is trivially represented on 'H o,
it follows that
(3.124)

defines a representation of Iro(H) on 'Ho. The second step of the induction proce-
dure then consists in induction on 'Ho with respect to the representations Uo and
Uo of D = Iro(H), in the way explained prior to (3.113).
It is instructive to illustrate this two-step procedure in the example of Yang-Mills
theory on a cylinder, with structure group H = U(1). Recall (3.70) and (3.115); it
follows that the inequivalent quantizations of S = T* AIR are labeled by e E U (1).
In the first step of the induction procedure we integrate over L U (1)° = L U (1 )[0].
We denote the left-hand side of (3.101), with LU(1) replaced by LU(1)o, by (, )8.
With A = L2(Sl), this leads to the expression

( v'Exp(W), v'EXP(Z»)O = e(w,Z) ( dl-tf (g) e(g·z-W). (3.125)

° JW(l)O

We see from (3.69) that for the present purpose we may put H = JR.
In that case the fundamental solution of the heat equation is Py(x, t) =
(2Iryt)-l/2 exp(-x 2 /(2yt». Using this in Lemma 3.8.2, or calculating directly,
 3 Applications in Relativistic Quantum Theory 431

one obtains

(JEXP(W), JEXP(Z»): = (21l')-1/2e4[(WL,W/')+(ZL,ZL)]+WIZI. (3.126)

We put 1l'~(A) := 1l'o(QF(A», where QF(A) is some adjointable operator on

exp(A). Recall (3.93) and (3.94).

Proposition 3.9.1. The induced space 1to is naturally isomorphic to L 2(JR.). On

this space one has
(3.127)

(3.128)

(3.129)

Here we have identified the generator of U(1) with -i, as usual. The first two
e
operators are defined and essentially self-adjoint on the linear span H of all
Hermite polynomials.
Compare the first two expressions with II.(2.23) and 11.(2.24).
As always, we employ Proposition 2.2.4, this time omitting the suffix The *.
guess 1to = L 2(JR.) is substantiated by defining Uo : <e -+ L 2(JR.) by linear
extension of

(3.130)

One checks (2.29) from (3.126) and a standard Gaussian integration. Since Uo<e
e
coincides with H as defined above, and the latter is dense in L2(JR.), we conclude
that the guess of L 2(JR.) for the induced space was a good one.
In the classical abelian theory both Al and EI are gauge-invariant under small
gauge transformations. In the quantum theory this is reflected by the fact that on
the domain <e the operators QF(A I ) and QF(EI) (see (3.93) and (3.94» commute
with U F (g), where gEL U (1)°, and are therefore adjointable. The induced action
of a(1) is found from 11.(2.67), (2.30), and (3.130) to be
d
1l'o(a(I» = x + ~ dx; (3.131)

by construction, this operator is defined on the domain Uo<e. Since the induction
procedure preserves the adjoint of adjointable operators, it follows from (3.131)
that 1l'o(a(1)*) = x - ~d/dx. From (3.93) and (3.94) we then obtain (3.127) and
(3.128), respectively. Recalling the definition of QF(W) in the abelian case, given
prior to (3.95), Equation (3.129) is then obvious. •
We now compute the representation Uo(Z) defined in (3.124). Any g E LU(l)[n]
isoftheformg = gOgn, where gO E LU(I)o,andgn is defined in (3.71). By (3.84),
(2.30), and (3.130) one obtains

Uo(n)'I1(x) = 'I1(x + 21l'n). (3.132)

432 IV. Reduction and Induction

Theorem 3.9.2. The induced space 1-lg defined by induction from the representa-
tion Uo(LU (1», given by (3.122) and (3.115), is naturally isomorphic to L2(U(I».
Writing 1l'~(A) := 1l'g(QF(A», one has
(3.133)
seen as a multiplication operator. Moreover, the quantized "electric.field" is

(3.134)

cf (3.119). This operator is defined and essentially self-adjoint on the domain

provided by Rieffel induction, and its closure is p().
Since (3.132) is the right-regular representation of 1l',(U (1» = 21l'Z on G = JR,
we are exactly in the situation leading to (3.115). In the present case, (3.113), with
1-l() = C, reads

(\II, <I»g = 2rr I: e r dx \II (x ) <I> (x + 2rr n),

inO (3.135)
nEZ Ja
where we have normalized the Haar measure on Z as 2rr times the counting mea-
sure. Bothrr~(E ,) andrr~(W) are adjointable withrespectto (3.135). (The operator
rr~(A ,) isn't, because among the multiplication operators on L 2(JR) only functions
with period 2rr are adjointable.) The map (2.97), transferred from 1-l x to 1-l; , then
reads
U:\{I(f3) = 2rr I: e inO \{I (f3 + 2rrn), (3.136)
nEZ

where \II E £H C L2(JR). Equation (3.133) then follows from (3.129), (2.30), and
(3.136). Similarly, from (3.128), (2.30), and (3.136) one obtains (3.134). It is crucial
that this unbounded operator is defined on the natural domain £H provided byU:
Rieffel induction. It follows from (3.136) that functions in this domain are smooth,
and satisfy the boundary condition \{I(2rr) = exp(-iO)\{I(O). The final claim then
follows from standard functional analysis. •
On the basis of these considerations one expects that any quantum gauge theory
on a compact space whose gauge group 9 is disconnected possesses inequivalent
quantizations labeled by the unitary dual of rroW). In physics one refers to elements
of ,r;;(Q) as vacuum angles or O-angles. Such angles do not label superselection
sectors (defined as inequivalent representations of the algebra of observables).
Rather, each vacuum angle defines its own algebra of observables (3.114). What-
ever their physical interpretation, in the description suggested here, vacuum angles
emerge if one constructs the algebra of observables by induction from a nontrivial
representation of the gauge group.
Notes

Chapter I
1.1.1 Jordan algebms were introduced by Jordan [1932] in connection with quantum me-
chanics (of which Jordan had been one of the founders), and were further studied by Jordan et
a1. [1934]. There is a substantiallitemture on such algebms; an interesting modern textbook
is Faraut and Koninyi [1994].
Jordan-Lie algebms appeared in Orgin and Petersen [1974], who claimed that in certain
cases the associator identity (1.6) follows from the other axioms. It was added as an extm
postulate by Emch [1984], who also noted that for /i2 i= 0 the Jacobi identity (1.5) follows
from the other axioms. Also cf. Ayupov et a1. [1997]. For Poisson algebras see the notes to
2.3.
The study of infinite-dimensional Jordan algebras was initiated by von Neumann [1936].
J B-algebms were introduced by Alfsen et al. [1978]. See Emch [1972] and especially Emch
[1984] for a nice overview with historical perspective. Axiom (1.7) can actually be derived
from (1.9) and (l.lO); see the comment on p. III of Alfsen and Shultz [1976], Alvermann
[1985], and Rodriguez Palacios [1988]. Hanche-Olsen and St0rmer [1984] is a textbook
on Jordan algebra and J B-algebms, and Upmeier [1987] presents an overview with many
applications. These works also describe a structure theory (mostly already present in Alfsen
et a1. [1978]), whose main conclusion is that any J B-algebm 2l contains an ideal 'J such
2lj'J is isomorphic (as a J B-algebra) to a norm-closed Jordan subalgebra of ~(1-l)R for
some Hilbert space 1-l.
The history of C' -algebms is told by Kadison [1982, 1994]; also see the Introduction in
Bmtteli and Robinson [1987]. Standard references are Dixmier [1977], Pedersen [1979],
Takesaki [1979], and Kadison and Ringrose [1983, 1986]. The most extensive analysis of
the Oelfand-Neumark Theorem l.l.8 is in Doran and Belfi [1986]. See Connes [1994],
Domn [1994], Fillmore [1996], and Davidson [1996] for modern surveys of C*-algebms.
Introductions that relate C*-algebms to quantum mechanicsand that are more oriented to-
wards physicists are Bmtteli and Robinson [1987, 1981], Thirring [1981, 1983], Emch
[1972, 1984], Haag [1996], and Landi [1997].
434 Notes

As shown by Araki and Elliott [1973], axiom (1.14) can actually be derived from (1.15);
see Doran and Belfi [1986] for an exhaustive study of the axioms and their consequences.
The relevance of C*-algebras to quantum mechanics was recognized by Segal [1947], and
received considerable impetus from the work of Haag and his collaborators on algebraic
quantum field theory; see Haag and Kastler [1964] and Haag [1996].
Theorem 1.1.9 is taken from Landsman [1997], who provides an alternative proof of the
second half. Namely, as shown by Wright [1977], the complexification 2( of a J B-algebra
2(1R can be normed and made into a so-called Jordan C*-algebra (alternatively called J B*-
algebra). One then adds the Lie structure and uses the result of Rodriguez Palacios [1988]
that if the Jordan product in a J B* -algebra 2( is the anticommutator of an associative product,
then 2(, equipped with this associative product and the original norm, is a C* -algebra.
If liZ < 0 one can tum 2( itself into an associative algebra through (1.23) with the i
omitted. This leads to a so-called real C*·algebra, or R*·algebra, which is isomorphic to
an algebra of bounded operators on some real or quaternionic Hilbert space; see Goodearl
[1982]. Attempts to model quantum mechanics on such spaces have been unsuccessful; cf.
Beltrametti and Cassinelli [1984) and references therein.

1.1.2 This material may be found in all textbooks on C* -algebras. Takesaki [1979],
Kadison and Ringrose [1983), and Davidson [1996] are particularly efficient. Palmer [1994]
is an encyclopedic treatise on Banach algebras.
Most of the theory holds for general J B -algebras; see Alfsen et al. [1978] and Hanche-
Olsen and St0rmer [1984].

1.1.3 Recall that a partial ordering ::: on a set is a binary relation von satisfying: (i)
=
x ::: x for all x; (ii) if x ::: y and y ::: x then x y; (iii) if x ::: y and y ::: z then x ::: z.
One writes x < y if x ::: y and x of. y; also, y ? x (or y > x) is the same as x ::: y (or
x < y). The general theory of partially ordered topological vector spaces is given by Wong
and Ng [1973] and by Asimow and Ellis [1980].
All C* -algebraic results in this section may be found in the standard textbooks.

1.1.4 Definition 1.4.1 is due to von Neumann [1932] (for 2( = ~(H» and Segal [1947)
(for general C*-algebras); both were motivated by quantum mechanics.
All books on C* -algebras discuss the basic properties of states and state spaces. The
general theory of compact convex sets may be found in in Alfsen [1971] and Asimow
and Ellis [1980]. Kadison and Ringrose [1983] discuss unital C'-a1gebras and their state
spaces in the light of this general theory; Alfsen et al. [1978], Asimow and Ellis [1980],
and Hanche-Olsen and St0rmer [1984] do so for general unital J B-algebras.
The use of more general compact convex sets and partially ordered Banach spaces than
those provided by C* -algebras is central to the so-called operational approach to quantum
mechanics, for which we refer to Haag and Kastler [1964], Schwinger [1970], Davies and
Lewis [1971], Hartkiimper and Neumann [1974], Mielnik [1974], Davies [1976], Gudder
[1979], Beltrametti and Cassinelli [1981], Holevo [1982], Ludwig [1985], Lahti and Buga-
jski [1980, 1985], and Busch et al. [1995]. The starting point is the duality between the state
space K, assumed to be a compact convex set, and the partially ordered Banach space of
observables A(K, 1R) or Ab(K, 1R). Theorem 1.4.5 is a special case ofthis theory, and should
be seen in its light. The connection between this approach and the theory of J B-algebras
has been studied by Kummer [1991].
The decomposition of rp used in the proof of 1.4.5 is Thm. 4.3.6 in Kadison and Ringrose
[1983]. The final step of the proof of Theorem 1.1.9 was inspired by the proof of Lemma
8.5 in Alfsen et al. [1978].
 Observables and pure states 435

l.l.S Representations and the GNS- (Gelfand-Neumark-Segal) construction are dis-

cussed in all books on C' -algebras. For more on the universal representation and the closely
related proof of Theorem 1.1.8, see Dixmier [1977], Pedersen [1979], or Kadison and
Ringrose [1983, 1985).
1.1.6 For the Riesz representation theorem see, for example, Pedersen [1989]. A detailed
proof of Theorem 1.6.5 may be found in Reed and Simon [1972], Takesaki [1979], or
Pedersen [1989].
1.1.7 For the history of von Neumann algebras cf. Kadison [1958,1982]. Von Neumann
developed his theory of "rings of operators" partly in order to generalize his own Hilbert
space formalism of quantum mechanics (von Neumann [1932]). He eventually came to
believe that quantum mechanics should be described by so-called type 111 factors; see von
Neumann [1981). Bub [1981] provides historical comments.
A clear discussion of the various topologies on m(1t) is in Takesaki [1979] or Pedersen
[1989]. The theory of von Neumann algebras, which started with Theorem 1.7.2 due to von
Neumann himself, is covered by Pedersen [1979], Thkesaki [1979], Kadison and Ringrose
[1983, 1985], and Connes [1994]. In addition to the first three of these books, cf. Dixmier
[1977] for the description of the bidual of a C' -algebra 21 as a von Neumann algebra, Le.,
Proposition 1.7.4. Proposition 1.7.5 is Theorem 10.1.12 in Kadison and Ringrose [1985];
equivalent statements are in Pedersen [1979] and Takesaki [1979]. It states the "universal"
property of the universal representation. Some authors use the term W' -algebra for an
abstract C* -algebra that is the dual of a Banach space, a von Neumann algebra then meaning
a W' -algebra realized on a Hilbert space.
1.2.1 The insight that an extreme point of a convex set (as defined by Minkowski) is
precisely a pure state in the sense of quantum mechanics is due to von Neumann. Extreme
points of general compact convex sets are studied in Alfsen [ 1971], Asimow and Ellis [1980],
and Pedersen [1989]; in connection with C' -algebras see Pedersen [1979], Takesaki [1979],
and Bratteli and Robinson [1987]. Kadison and Ringrose [1983] is particularly efficient,
and contains a proof of the Krein-Milman Theorem 2.1.5. For (2.2) see Dixmier [1977],
§3.2.4.
1.2.2 Most of this section is standard; for the reduced atomic representation see Pedersen
[1979], Kadison and Ringrose [1983], and Akemann and Shultz [1985]. Proposition 2.2.8
is given, for example, in Kadison and Ringrose [1986], Prop. 6.6.6, and in Davidson [1996],
Thm. III. 1. 1.
1.2.3 Poisson algebras and Poisson manifolds in the setting of function spaces go back
to Lie [1890]; the modem era started with Kirillov [1976] and Lichnerowicz [1977]. See
Marsden and Ratiu [1994] (p. 293) for historical comments, and Huebschmann [1990] for
extensive references. The abstract concept of a Poisson algebra appeared implicitly in Falk
[1951]; Dirac [1950,1964] deserves major credit as well. The theory of Poisson manifolds
can be found in Weinstein [1983], Libermann and Marie [1987], Marsden and Ratiu [1994],
and Vaisman [1994]; also see the survey by Weinstein [1998].
Equation (2.5) is equivalent to the statement that the so-called Schouten (or Nijenhuis)
bracket of B with itself vanishes; see the books quoted in the previous paragraph.
For technical properties of flows in connection with classical mechanics, see Abraham
and Marsden [1985]. For the theory in infinite dimension, cf. Marsden [1974] and Chernoff
and Marsden [1974]. A symplectic structure of the type we consider is called a strong
symplectic structure in the literature (see the two books just quoted). A closed 2-form lJ) is
called strongly symplectic if the map Bu, defined by reading (2.18) from right to left, is an
isomorphism. In contrast, for a weak symplectic structure this map is merely injective.
436 Notes

There is a huge literature on symplectic manifolds; Abraham and Marsden [1985), Arnold
[1989], Arnold and Givental [1990]. and Slawianowski [1991] are particularly useful in the
context of classical mechanics.
1.2.4 For Definition 2.4.1 see, for example. Libermann and Marie [1987), App. 3. Here
one also finds the singular Frobenius theorem 2.4.2 (due to Sussmann [1973)) as Thm. 3.10.
The ordinary Frobenius theorem is included as Thm. 4.2; for the latter also cf. Choquet-
Bruhat et aI. [1982), A readable review ofthe theory of (singular) distributions and foliations
is Dazord [1985].
Theorem 2.4.7 is due to Kirillov [1976]; our discussion follows Marsden and Ratiu
[ 1994], § 10.6, where further details concerning the connection with singular foliation theory
may be found (note that these authors use the terminology "symplectic stratification theo-
rem", although the singular foliation obtained is not a stratification in the sense of Goreski
and McPherson [1988] or Sjamaar and Lerman [1991)). Other detailed treatments are in
Libermann and Marie [1987], §III.12, and Vaisman [1994], Ch. 2, who includes examples.
1.2.5 The insight that the (normal) pure states of an irreducible quantum system corre-
spond to points in lP1-i goes back (at least) to Weyl [1931], p. 75. For lP''H as a symplectic
manifold see, e.g., Cirelli et al. [1983], Abbati et al. [1984], Cirelli et al. [1994], or Mars-
den and Ratiu [1994]. For infinite-dimensional manifolds in general see Marsden [1974],
Abraham and Marsden [1985], Choquet-Bruhat et al. [1982], or Lang [1995].
Proposition 2.5.2 is due to Cirelli et al. [1983]. It follows from (2.68) and Prop. 2.6.15
of Bratteli and Robinson [1987] that the norm-topology relative to both lP''H c ~o('H)*
and lP1-i c ~('H)' coincides with the manifold topology on lP''H as well. Theorem 2.5.4 is
similar to Prop. 4.2 in Roberts and Roepstorff [1969]; we have added the appearance of lP''H
with its manifold topology.
Our construction of lP''H as a symplectic leaf in 'H*I U (1) is not standard, but forms an
instance of the general procedure of reduction; see IV.1.5, in particular Theorem IV. 1.5.5.
If a compact Lie group G acts smoothly on a manifold M, then MH (the collection of points
in M with stability group H) and MHIG are manifolds (MIG may not be). See Bredon
[1972]. We could have worked with 'HI U(I); this is not a Poisson manifold, but a Poisson
space in the sense of Definition 2.6.2. Also see IV.I.II.
The projected Schrodinger equation (2.46) in the given (symplectic) context goes back
at least to Hermann [1973]. It is physically not very interesting for bounded H. If the
Hamiltonian H is unbounded, the function il and its Hamiltonian vector field ~ (j are
defined only on a dense submanifold of lP''H, namely the projection of the domain of H.
This situation can be handled by the theory of densely defined vector fields on infinite-
dimensional manifolds, see Marsden [1974] and Chernoff and Marsden [1974]. Even in
that case the flow 1/I(t) is defined on all of lP''H.
Using the standard complex structure J on'H (defined by JV(<<I» = V(i«l») or on lP1-i
(where similarly, J = i in each local chart (2.28», one can define a Kahler metric g by

g(X, f) = ~w(X, Jf).

See, e.g., Griffiths and Harris [1978] (N < 00) and Marsden [1974] (N = 00) for the
mathematics, and Strocchi [1966], Hermann [1973], Marsden [1974], Cirelli et aI. [1983],
Cirelli and Lanzavecchia [1984], Abbati et al. [1984], Marsden and Ratiu [1994], Cirelli et
al. [1994], Hughston [1995], and Ashtekar and Schilling [1997] for applications in quantum
mechanics. On 'H the Kahler metric is given by

g(V(<<I», V(Q» = liRe (<<I>, Q).

 Observables and pure states 437

On P1l one has the remarkable equation, found by Cirelli et al. [1990],

fig(~A' ~iJ)(l/I) = AoB(l/I) - A(l/I)B(l/I).

This relates the Fubini-Study metric on IP'1l to the Jordan product 0 on !B(1l), much as the
Fubini-Study symplectic form on P1l is related to the Poisson bracket on !B(1l), cf. (1.22).
Since this relation is not easily generalized to other C· -algebras than !B(1l) or 230(1l), it
plays no role in our approach.
1.2.6 The point of view in this section, as well as Proposition 2.6.4, originate in Lands-
man [1996a]. The definition of a Poisson space was partly inspired by that of a stratified
symplectic space in Sjamaar and Lerman [1991].
Here is the technical argument alluded to at the end of the proof of Proposition 2.6.4.
If necessary, one adds the unit function I p to !2l, and extends n by linearity and the rule
n( 1p) = Is; this still defines a representation, and j remains multiplicative on the extended
algebra 2t (cf. Lemma 2.3.26 in Bratteli and Robinson [1987]). Evidently, X,(lp) = I. The
multiplicativity of j immediately implies that it is positive on 2tnC(p), and by the previous
equation j must therefore be continuous with norm 1. Hence it can be extended to all of
C(P), where it remains multiplicative. By continuity, l(n(j))(u)1 ::: IItII"" for all u E S,
so that IIn(j)II"" ::: 11/1100. Hence n : !2l ~ Cgo(S) c Cb(S) is continuous as a map
between Banach spaces. We extend n to all of C(P) by putting (n(j))(u) = J(u)(j); this
is precisely its extension by continuity. Since multiplication is continuous in the sup-norm,
n = J* is a Jordan morphism of C(P) into Cb(S) (where the Jordan product 0 is pointwise
multiplication). The continuity of J now follows (cf. Thm. 3.4.3 in Kadison and Ringrose
[1983]): Since P, being compact and Hausdorff, is completely regular, a subbase for the
topology of P is given by (f-I(O)}, where I and 0 range over C(P) and the open sets
in JR, respectively (cf. Kelley [1955], p. 117). Now, J- 1(j-l(O)) = n(j)-l(O), which is
open since n(j) is continuous. Hence J is continuous.
If P is not compact, the proof undergoes only minor changes. The Stone-Weierstrass
theorem (see, e.g., Pedersen [1989]) now says that!2l n Cc(P) is dense in Co(P). It follows
that Xr is normalized, and the remainder of the argument is the same. Note that general
positive functionals defined on a dense subalgebra of Ca( P) may not be extendible to positive
functionals on the unitization of this subalgebra; the argument needs the preservation of
multiplicativity on this extension to conclude positivity (and hence boundedness).
Definition 2.6.6 is taken from Landsman [1996a] (written in 1992); it was rediscovered
in Gotay et ai. [1996]. Theorem 2.6.7, in the special case that P is a Poisson manifold, is
in Landsman [1996a].
1.2.7 Transition probabilities were introduced by Born [1926] in the context of quantum-
mechanical collision theory. Curiously, he initially thought that (in modern notation) the
transition probability between two unit vectors \II and <I> was given by the inner product
(\II, <1», and stated the correct expression 1(\11, <1»1 2 only as a note added in proof. The abstract
notion of a transition probability space is due to von Neumann [1981], who thereby went
beyond the general situation in quantum mechanics laid out in von Neumann [1932]. The
condition of symmetry (which has nothing to do with the invariance of the laws of physics
under time-inversion, cf. Haag [1996]) was not included in his definition of a transition
probability space. The concept was revived by Mielnik [1968], who introduced the notion
ofa basis and proved Proposition 2.7.4. Further work is in Zabey [1975], Belinfante [1976],
and Pulmannova [1986]; see Beltrametti and Cassinelli [1984] for a concise review.
Here is an example of a transition probability space that is not well-behaved (mentioned,
with an error, in Zabey [1975], who attributes it to Mielnik). The elements of Pare equiv-
438 Notes

alence classes of subsets of R of the type P ~ [0, n] (where the integer n 2: 3) of Lebesgue
measure /-L(p) = 1; two subsets are equivalent when they differ by a null set. The transition
probabilities are defined by pep, u) = /-L(p n u). Note that dim(P) = n. For Q ~ P, let
Q ~ [0, n] denote the union of all members of Q. Then Ql. consists of all elements of P
that are disjoint (up to null sets) from Q. If /-L(Q) > n - 1 there are no such elements, so that
Ql.l. = P in that case. If /-L(Q) :s n - 1, on the other hand, Ql.l. consists of all elements
of P that are contained in Q. Therefore, if Q c [0, n] is a given subset and Q is defined as
the collection of all elements of P that lie in Q, then Q is orthoclosed iff /-L(Q) :s n - I.
However, Q has a basis only if /-L( Q) is integral. For another example see BeJtrametti and
Cassinelli [1984].
Any orthogonal subset of P is a sample space in the sense of classical probability theory.
The second requirement of 2.7.5 may then be rephrased by saying that any maximal "clas-
sical" subspace of P is complete, in the sense that an arbitrary point in P must make a
transition to some member of the given subspace. Moreover, this kind of completeness
continues to hold if one restricts to orthoclosed subspaces.
The existence of a multitude of maximal classical subspaces in a nontrivial transition
probability space is very hard to interpret. In the case of quantum mechanics, the so-called
Copenhagen interpretation (see Jammer [1974]), which is based on the somewhat obscure
philosophy of complementarity of Niels Bohr, says that the choice of some such subspace
is determined by the experimental arrangement set up by a physicist. The author's opinion
is laid out in Landsman [1991, 1995b].

1.2.8 The expression (2.63) was proposed by Mielnik [1969] for arbitrary convex sets K,
but this formula does not actually define a transition probability without the extra condition
we have added. The expression is motivated by operational considerations about filters and
preparation procedures in quantum mechanics (see the references in the notes to 1.4). We
will not give this motivation here, since in our approach transition probabilities are funda-
mental and irreducible properties of pure state spaces, from which a possible operational
interpretation of the theory is to be derived, rather than the converse.
Theorem 2.8.2 may be generalized, stating that the pure state space of a J B-algebra QlJR
is a symmetric transition probability space under (2.63). Firstly, it follows from equation
(4.3) and Cor. 7.3 in Alfsen and Shultz [1978] that every pure state in a J B-algebra is norm-
exposed, so that by 2.8.1, equation (2.63) indeed defines a transition probability. Secondly,
we need to show that the transition probabilities thus defined are symmetric. As explained
on p. 159 of Alfsen and Shultz [1978], and also in Prop. 1.13 of Alfsen and Shultz [1979], in
J B -algebras there is a bijective correspondence between pure states p of QlR and minimal
idempotents (projections) p in QlR *; here an idempotent p in a J B-algebra is an element
°
satisfying p2 = P (hence :s p :s 1I), and the minimality of p means that there is no
nonzero projection q such that q :s p (our notation is different from the reference cited).
This correspondence is given by the equation pep) = 1, which uniquely determines one
entry given the other. It then follows from Thm. 2.17 in Alfsen and Shultz [1976] that the
transition probability pep, u) as defined by (2.63) is given by &(p). The symmetry of the
transition probabilities then follows from equation (4.5) and Cor. 7.3 in Alfsen and Shultz
[1978].
Equations (4.3) and (4.5) in Alfsen and Shultz [1978], which are central to the above
proof, are two of the three "pure state properties", which they show to be satisfied by the
pure state space of a J B -algebra. In a slightly more general context, these or closely related
properties were first postulated by Gunson [1967] and Pool [1968].
 Observables and pure states 439

There is an alternative way of looking at the symmetry of the transition probabilities for
J B-algebras. If 2(R is a finite-dimensional J B-algebra, there exists a (real) inner product
(, ) on 2(R ::::: Rn such that the positive cone 2(~ is self-dual. Using the inner product to
identify 2(a with 2(it, this means simply that 2(~ = (2(it)+. The symmetry of the transition
probabilities then eventually follows from the symmetry of the inner product. The close
relationship between the self-duality of the positive cone and the symmetry of the transition
probabilities in a finite-dimensional (formally real) Jordan algebra has been stressed by
Haag [1996] in connection with the foundations of quantum mechanics. It turns out that
finite-dimensional J B-algebras are characterized by the self-duality of the positive cone,
plus the fact that the subgroup of G L(2(IR) that maps 2(~ into itself acts transitively on
the interior of 2(~. See, for example, Faraut and Koninyi [1994]. An infinite-dimensional
analogue of this result is discussed in Iochum and Shultz [1983] and Iochum [1984]; the
central property of facial homogeneity (originally due to Connes) occurring in their work is
further analyzed in Ajupov et al. [1990], in which it is admitted that the physical relevance
of this property is obscure.
Equation (2.67) appears in Roberts and Roepstorff [1969], who also prove the related
result that the spaces f"Ha in the decomposition P(2() = Uaf"H a (cf. 2.5.4) are precisely the
components ofP(2() in the norm-topology. The result that lip-a Ii = 2 for inequivalent pure
states is due to Glimm and Kadison [1960]; the statement actually holds for arbitrary disjoint
states (these are states whose GNS-representations have no equivalent subrepresentations),
see Cor. 10.3.6 in Kadison and Ringrose [1986].
The transition probabilities (2.65) can be expressed in terms of the Fubini-Study metric
g on f"H. This metric can be normalized in such a way that
pep, a) = ~(l + cosd(z, w»,
where d is the distance defined by g. For example, for 11. = (:2 the Fubini-5tudy distance
d(z, w) is just the angular distance measured along the (shortest) great circle connecting z
and w (cf. 3.7.1). This expression has led to interesting connections with information theory,
entropy, uncertainty, and statistical inference; cf., e.g., Hilgevoord and Uffink [1991], Petz
[1994], and Brody and Hughston [1998]. In a different direction, the ensuing connection be-
tween quantum mechanics and Riemannian geometry has been exploited by Anandan [1991]
and Ashtekar and Schilling [1998]. Another way to look at (2.65) relates this transition
probability to the projective cross-ratio of algebraic geometry; see Hughston [1995].
Continuity properties of the transition probabilities (2.65) are studied in Archbold and
Shultz [1989].
Although it is somewhat contrary to the spirit of the present work, one can define transition
probabilities between mixed states (the physical relevance of such transition probabilities
has been questioned by Roberts and Roepstorff [1969]). For general '-algebras this was
done in Uhlmann [1976]; it was shown by Alberti [1983] that for unital C'-algebras 2(
Uhlmann's general expression reduces to
P(Wl, W2) = inf {wl(A)W2(A- l )IA > 0, A E 2(, A-I E 2(}.

For density matrices on a Hilbert space 11. (that is, states on 93 0 (11.» this is equivalent to a
formula due to Araki [1972], namely

Further information and references may be found in Uhlmann [1993]. Cantoni [1975] defines
transition probabilities between arbitrary states in the context of lattice theory (cf. 3.6); see
440 Notes

Gudder [1979] for a review. The equivalence between Cantoni's and Uhlmann's transition
probabilities on state spaces of unital C* -algebras is shown in Raggio [1982] and Araki and
Raggio [1982]. In more general situations the various approaches do not even coincide on
the pure state spaces; see Pulmannova [1989].
1.3.1 This material is mainly taken from Landsman [1997]. Following a seminar the
author gave in Gottingen, 1995, A. Uhlmann informed him that in his lectures on quantum
mechanics ~(P) had long been employed as the space of observables; also see Uhlmann
[1996].
1.3.2 Theorem 3.2.1 originates in the following result of Shultz [1982]: If A E p~**
is such that A, A* A, and AA* are in C.(P(~», then A E p~. It was then shown by
Brown [1992] that the hypothesis on A* A and AA* can be dropped. Our proof is based on
that of Brown [1992], which also contains the corollary of the Stone-Weierstrass theorem
that is used in the proof. The original Stone-Weierstrass theorem for CO-algebras, due to
Glimm (see Dixmier [1977] for a very detailed presentation), states that if ~ is a unital
C* -subalgebra of a unital C* -algebra IB which separates P(IB)- , then ~ = lB. The usual
Stone-Weierstrass theorem then follows by taking IB to be commutative.
Equation (3.5) follows from the property that the predual determines the order ofIDt (that
is, w(A) 2: 0 for all pure normal states w implies A 2: 0), and Lemma 3.4.1 in Dixmier
[1977]. There is no such result for arbitrary von Neumann algebras, which may even have
no pure normal states at all. In the given setting the pure normal states are abundant, because
N(IB**) = S(IB).
If~ has no unit, Theorem 3.2.1 can be adapted in two essentially equivalent ways. Firstly,
one has (~[)a = ~R(P(~» n Cu(P(~), R); this follows from the proof as given, plus
=
Corollary 8 in Brown [1992]. Secondly, ~R ~R(P(~» n Cu(P(~) U 0, R) (cf. Theorem
6 of Brown [1992]). Indeed, the Stone-Weierstrass theorem for nonunital C· -algebras is as
in the unital case, but with P(IB)- replaced by P(IB)- U O.
Perfect C*-algebras were introduced in Shultz [1982], and studied in detail in Ake-
mann and Shultz [1985]. For nonunital algebras the definition is that ~ is perfect if
~R = ~1R(P(~» n C(P(~) U 0, R); this means that ~ is perfect iff its unitization is.
The main motivation was that if~ is perfect, the Stone-Weierstrass theorem can be sharp-
ened so as to state that ~ = IB if ~ separates P(IB) U 0 (where "U 0" may be omitted in
the unital case). The perfectness of lBo(1t) is a special case of the following result (Shultz
[1982]): If P(~)- consists of multiples of normal states on 1l"rd(~)"' then ~ is perfect. Also,
arbitrary direct sums of perfect C* -algebras are perfect.
The physical meaning of uniform structures on state spaces in quantum mechanics is
discussed by Werner [1983].
1.3.3 This material is from Landsman [1997]; the second half of the proof of Proposition
3.3.3 is based on the proof of Thm. 12.12 in Alfsen and Shultz [1976]. For the proof of
Lemma 3.3.4 see Alfsen et ai. [1978], Shultz [1979], or Hanche-Olsen and St0rmer [1984].
1.3.4 Proposition 3.4.1 is taken from Landsman [1997]. Exhaustive information on deriva-
tions and one-parameter automorphism groups on Banach spaces may be found in BratteH
and Robinson [1987].
Corollary 3.4.2 was inspired by Thm. 18 in Shultz [1982], but is phrased in different
language and has an entirely different proof. In Shultz's result the condition that a* be a
Poisson map is replaced by the requirement that a* preserve the orientation of P(~).
There are similar results relating properties of an (auto)morphism a of a unital C* -algebra
~ to properties of its dual a* , seen as a map on the entire state space S(~). Kadison [1965]
showed that a is a Jordan automorphism of~ iff a* is an affine (w*-) homeomorphism of
 Observables and pure states 441

S(Qt) (also cf. Bratteli and Robinson [1987]). Shultz [1981] extended this by proving that
a is a morphism of Qt iff a* in addition preserves orientation.
Corollary 3.4.3 is a famous theorem due to Wigner [1931]. An antiunitary operator U
is an antilinearbijection on 'H (i.e., UclJl = cUlJI for all cEq that satisfies (UlJI, U<1» =
(<1>, lJI)foralllJl, <1> E 'H. See Bargmann [1964],RobertsandRoepstorff[1969], Varadarajan
[1985], Beltrametti and Cassinelli [1984], Shultz [1982], Cirelli et a1. [1983], or Thynman
and Wiegerinck [1987] for various alternative approaches to this theorem.
Lemma 3.4.4 goes back to Kadison, but we here refer to Bratteli and Robinson [1987]
(Example 3.2.14) for a detailed presentation. The following generalization is due to Alfsen
et a1. [1980] (Prop. 2.4): if 7f; : Qt ~ ~('H;)1R (i = 1,2) are irreducible Jordan-equivalent
representations of a J B-algebra Qt, then there exists a unitary or an antiunitary map from
'HI to 'H 2 that implements the equivalence (here Jordan equivalence means that there is a
Jordan automorphism f3 : ~('HI)R ~ ~('H2)1R such that 7f2 = f3 o7fI).

1.3.5 A classic on lattice theory is Birkhoff [1967]. For orthomodular lattices see Maeda
and Maeda [1970] or Kalmbach [1983] (which is highly readable and contains many attrac-
tive historical quotations and excursions). The connection between Hermitian forms and
orthocomplementations is thoroughly discussed in Baer [1952], and Varadarajan [1985], as
well as in Maeda and Maeda [1970]; recent reviews are Piziak [1991] and Holland [1995].
The fact that C(V) is not modular in infinite dimension follows, after an elementary but
somewhat lengthy argument, from the existence of closed subspaces whose sum is not
closed; cf. Kalmbach [1983] or Beltrametti and Cassinelli [1984].
A division ring Jl)) (sometimes called a skew-field) is a ring in which the equations
xa = band ay = b can be solved for x and y whe~ever a i- O. An involution of Jl))
is a linear bijection A ~ I satisfying A/l = iiI and I = A. For general division rings,
the definition of a sesquilinear form is the same as for Jl)) = C, namely a bilinear map
(, ) : V x V ~ Jl)) satisfying (AlJI, /l<1» = I(lJI, <1»/l and (<1>, lJI) = (lJI, <1». Such a form is
said to be nondegenerate if (lJI, <1» = 0 for all <1> implies lJI = O.
A detailed proof of Proposition 3.5.7 is in Kalmbach [1983], Thm. 3.1; also cf. Birkhoff
[1967], §8.

1.3.6 For a direct proof of Proposition 3.6.1 cf. Kalmbach [1983]. Detailed discussions of
projections in von Neumann algebras are in Takesaki [1979] and in Kadison and Ringrose
[1983]. It can be shown that C(9R) is irreducible iff 9R is a factor, that is, 9R n 9R' = ICH.
This is because the center C(C(9R» consists of the projections in 9R n 9R'. One should
therefore be aware that 9R may not act irreducibly on a Hilbert space (in the sense of 2.2.1),
while C(9R) is nonetheless irreducible.
A modem reference on the lattices C(Qt**) and F(S(Qt» is Akemann and Pedersen [1992],
where Proposition 3.6.3 may be found; it goes back to Prosser [1963] and Effros [1963].
The "technical argument" on left ideals used in the proof is due to Effros [1963], and also
appears in Pedersen [1979], Thm. 3.6.11. As shown by Topping [1967], these lattices have
the property of semimodularity (also called M-symmetry), cf. Birkhoff [1967] or Maeda
and Maeda [1970]. Call (y, z) a modular pair if (3.12) holds for all x ::: z. Semimodularity
then means that (z, y) is a modular pair whenever (y, z) is. In an orthocomplemented atomic
lattice, semimodularity is equivalent to the covering property defined in 3.7, cf. Thm. 30.2
in Maeda and Maeda [1970].
Proposition 3.6.3 can be generalized to a certain class of partially ordered Banach spaces
(which includes unital J B-algebras and C*-algebras); see Alfsen and Shultz [1976] and
Edwards and Riittimann [1985].
442 Notes

Proposition 3.6.4 is due to Zabey [1975] and Belinfante [1976]; also cf. Beltrametti and
Cassinelli [1984] and Pulmannova [1986].
A third lattice associated with a unital C' -algebra 2l, which is not isomorphic to either
£(2l") or £(2lR (P», is the lattice F(2l) of w' -closed faces of S(2l) (where ~ is ~). Since
a w' -closed face is norm-closed, one can associate a projection p to each such face by
the construction in the proof of 3.6.3. Any projection thus associated to a w' -closed face is
called closed; ifll - p is closed, then p is said to be open. There is a bijective correspondence
between open projections p and norm-closed left ideals 2l" p n 2l in 2l. The lattice F(2l)
is isomorphic to the lattice .c(2l) of all closed projections, but note that the latter is not a
sublattice of £(2l"), since x v y is not necessarily closed if x and yare. Hence x v y has to be
redefined as the smallest closed projection containing x and y. An intrinsic characterization
of an open projection is that it is the ultraweak limit of an increasing net in 2l. It can be shown
that all open and closed projections lie in 2lR (P). The terminology comes from the special
case where 2l is abelian: In that case the open projections are precisely the characteristic
functions of open sets in P(2l) (with the w' -topology). It is clear from this example that the
lattices .c(2l) and F(2l) do not admit a (natural) orthocomplementation.
More information on this subject, sometimes called noncolDDlutative topology, may be
found in Akemann [1969], Giles [1970], Giles and Kummer [1971], Borceux and van den
Bossche [1989], and Akemann and Pedersen [1992].
Yet another lattice (the fourth) associated with 2l consists of all projections in a(2l), which
is the so-called sequential completion of 2l. This is defined as the smallest a -complete C'-
algebra on Hra containing Jrm(2l); here a C'-algebra concretely acting on a Hilbert space
H is called a-complete if it contains the limits of all weakly convergent sequences in
it. See Plymen [1968]. This lattice is generally neither atomic nor complete, though it is
orthocomplemented. Also cf. Roberts and Roepstorff [1969] for the use of a-complete
CO-algebras in algebraic quantum mechanics.
Proposition 3.6.5 and Theorem 3.6.6 are from Landsman [1997]. From the point of view
of quantum logic, the first claim of Theorem 3.6.7 is that each A E 2l is an observable
on the lattice £(P(2l» if the Borel sets Bj are mutually disjoint; an observable on a a-
complete orthocomplemented lattice £ is defined as a lattice homomorphism A : B(lR) --+ £
satisfying A(v~1 B i ) = V~I A(Bj ), cf. Varadarajan [1985], §1II.2. The study of observables
on lattices is closely related to measure theory on lattices; see Varadarajan [1985], Rilttimann
[1985], and Schindler [1990]. Yet another way of looking at this situation is that a given
observable A defines a map p t-+ tL~ from (pure) states into probability measures on lR
(supported on the spectrum of A).
Theorem 3.6.7 places our approach in the context of Mackey 's [1963] axioms for quantum
mechanics. See Plymen [1968], Roberts and Roepstorff [1969], Gudder [1979], and Holland
[1995] for further development of Mackey's methodology.

1.3.7 The two-sphere property was inspired by Alfsen et al. [1980] and Shultz [1982].
Theorem 3.7.2 is from Landsman [1997]. The exclusion of dim(Pa ) = 3 is a consequence
of the use of Theorem 3.7.4, which leads to the desired result for dim(P,,) 2: 4 only (the
case dim(P,,) = 2 is covered directly by the axiom).
The covering property in atomistic lattices is equivalent to the exchange property, stating
that if a and b are atoms and x is such that x 1\ a = 0, then a ~ x v b implies b ~ x Va.
This, in tum, is equivalent to BirkhotT's exchange axiom; cf. Thm. 7.10 in Maeda and
Maeda [1970] or Prop. 10.1 in Kalmbach [1983].
Lemma 3.7.3 is taken from Landsman [1997]. The final step in the proof is as follows.
According to Ramsay [1965], a complete orthocomplemented lattice £ with the covering
 Observables and pure states 443

property is a so-called dimension lattice (cf. Kalmbach [1983, 1986] for a detailed discus-
sion). This implies that there is a function d : [, ~ JR+ with certain properties; in our
case, [, = ['(P), it is easily verified that d is proportional to the dimension defined in 2.7.
More precisely, our proof of the finite-dimensional covering property (which was inspired
by the proof of Prop. 6.15 in Alfsen et al. [1980]) implies that each interval [0, Q], where
dim(Q) < 00, is a dimension lattice.
By Thm. 13.2 in Kalmbach [1983] (or, equivalently, Prop. 8.2 in Kalmbach [1986]), a
complete orthomodular lattice is modular iff it is a dimension lattice in which d(l) < 00.
Thus each [0, Q] is modular as long as dim(Q) < oo,as is 1:= {Q E [,(P) Idim(Q) < oo}.
The sub lattice I is an ideal of [,(P) (in the sense that y ::: x and x E I imply y E I), which
is supremum-dense (this means that an arbitrary x E [,(P) may be written as x = VjXj for
some Xj E I). The existence of a supremum-dense modular ideal means, by definition, that
[,(P) is locally modular. Thm. 8.17 in Kalmbach [ 1986] states that a complete orthomodular
and locally modular lattice is a dimension lattice. Thm. 8.20 in Kalmbach [1986] says that
a dimension lattice has the exchange property (we use only the implication (i)~(ii) of
this theorem, since the converse, while true, has an incomplete proof). As remarked in
the previous paragraph, in our context the exchange property is equivalent to the covering
property. This completes the proof of Lemma 3.7.3.
Theorem 3.7.4 originated in projective geometry; the main contributions were by von
Staudt, Hilbert, von Neumann, and Birkhoff (junior). Complete modem proofs may be found
in Baer [1952], Freyer and Halperin [1956], and Varadarajan [1985], who also explain
the connection between lattice theory and projective geometry (the connection between
quantum mechanics and projective geometry clearly fascinated von Neumann; cf. Piron
[1976] and Varadarajan [1985] for a full explanation of this connection). Summaries are in
Maeda and Maeda [1970], Birkhoff [1967], BeItrametti and Cassinelli [1984], Kalmbach
[1986], and Holland [1995]. The fact that length 3 is excluded is caused by the existence of
so-called non-Desarguesian projective geometries in dimension 3; see Freyer and Halperin
[1958] for a certain analogue of the coordinatization procedure in that case. Various other
generalizations exist; for example, when [, is not necessarily atomic, but modular, one can
coordinatize [, in terms of a so-called (von Neumann) regular ring (instead of a division
ring). If [, has no atoms at all, this leads to the subject of continuous geometry (cf. Maeda
[1958] and von Neumann [1981]), created by von Neumann in connection with quantum
mechanics and his work on rings of operators.
Lemma 3.7.5 is due to Kolmogorov [1932], and was used in exactly the same way in
Zierler [1961] and in Cirelli and Cotta-Ramusino [1973].
The criteria setting out when a definition of convergence defines a topology are given in
Kelley [1955]. They are almost trivially verified in Lemma 3.7.6, since our convergence is
defined through convergence in R
Lemma 3.7.7 is taken from Landsman [1997]; the first argument in the proof is Lemma
3.3 in Cirelli and Cotta-Ramusino [1973].
The classification of topological division rings used in the proof of Lemma 3.7.8 is due
to Pontrjagin [1946] (also cf. Weiss and Zierler [1958]). The classification of (continuous)
involutions of JR, C, and lIll is discussed in Varadarajan [1985] (§1I.2 and Lemma IV.4.5);
also cf. Wilbur [1977] for conditions guaranteeing the continuity of the involution.
Proposition 3.7.9 is due to Amemiya and Araki [1966] (it had previously been stated, with
an incorrect proof, by Piron); also cf. Maeda and Maeda [1970] (Thm. 34.9), Varadarajan
[1985] (Lemma 4.42), or Kalmbach [1986] (Thm. 11.9).
The generalization of Wigner's theorem used at the end of 3.7 is Theorem 4.29 in
Varadarajan [1985]. For dim(1t) ~ 4, it is equivalent to the fact that the group of lattice
444 Notes

automorphisms of P(1t) (for separable 'It) is precisely the group of unitary and antiu-
nitary operators on 'It. More precisely, for any automorphism rp of C('It) there exists a
unitary or antiunitary operator U on 'It such that rp(K) = UK, where K is a closed sub-
space of 'It (if we look at C('It) as the lattice C(~('It» of projections p on 'It, one would
have rp(p) = UpU*). This is also derived in Varadarajan [1985], §IV.3 (the results are
stated for infinite-dimensional 'It, but this restriction is not essential). The proof is not easy,
relying on several steps in the proof of the coordinatization theorem 3.7.4. Wigner's gen-
eralized theorem (like its weaker counterpart) holds in any dimension, for one can embed
a low-dimensional Hilbert space isometrically in a higher-dimensional one, and choose the
bijection so that the embedded space is mapped into itself.
Section 3.7 should be seen in the light of a large body of work in which axioms on an
orthocomplemented lattice C are given so as to make it isomorphic to Celt). This program
goes back to Birkhoff and von Neumann [1936], and received considerable impetus from
Mackey [1963]. See Zierler [1961], Wilbur [1977], Piron [1976], Gudder[1979], Beltrametti
and Casinelli [1984], Kalmbach [1986], Piziak [1991], and Holland [1995]. The orthomod-
ularity of C is somewhat justified from Mackey's layout of the logical structure of quantum
mechanics. The covering property can to some extent be physically motivated in an opera-
tional framework (cf. Gunson [1967], Pool [1968], and Beltrarnetti and Casinelli [1984]),
whereas irreducibility amounts to the absence of superselection rules. Completeness and
atomicity seem more a matter of mathematical convenience.
Having arrived at a lattice of the type C( V), the main difficulty in the traditional approach
lies in the determination of the division ring lDl. An important mathematical breakthrough
is the work of Soler [1995] (reviewed in Holland [1995]), who gave surprisingly minimal
conditions on C implying that V must be a Hilbert space over R C, or JEll. Her main condition
on C is equivalent to the existence of an infinite orthogonal sequence in V, and therefore
her theorem applies only to infinite-dimensional separable Hilbert spaces (moreover, her
conditions are very hard to interpret physically).
Since the fields Rand 1HI are as irrelevant to quantum mechanics as other more exotic
division rings, our approach has been to put in the choice of C as early as possible. Since it
enters through an axiom on the transition probabilities, this has been done in a physically
meaningful way (this was inspired by Schwinger [1970], who introduces C through the
properties of filters). As an added bonus, the covering property did not have to be postulated
separately, but could be derived.

1.3.8 This material, like that of the next section, is taken from Landsman [1997]. A
simplified version appeared in Landsman [1998b].

1.3.9 Theorems 3.9.1 and 3.9.2 should be compared with the work of Alfsen et al. [1980],
who characterized unital CO-algebras in terms of their state spaces (cf. Alfsen [1977] and
Asimow and Ellis [1980] for reviews); see Landsman [1997] for such a comparison. Araki
[1980] provides a certain simplification of this characterization in the finite-dimensional
case. The general program has been continued by Alfsen and Shultz [1998].
An interesting argument leading from J B-algebras to C'-algebras is that only C*-
algebras admit a satisfactory notion of a tensor product, allowing one to combine physical
systems; see Araki [1980] and Hanche-Olsen [1985].
The normal state space of a J BW -algebra or a von Neumann algebra has been character-
ized by Iochum and Shultz [1983]. The situation is qualitatively different from J B -algebras
or C' -algebras, since a normal state space may have no extreme points.
 Quantization and the classical limit 445

Chapter II

11.1.1 The quotation in the Introductory Overview is from Simon [1980]; it is sometimes
displayed in papers and seminars, and R. Haag has repeatedly expressed similar sentiments.
For the early history of quantization and the classical limit, based on Bohr's
correspondence principle, cf. Jammer [1974] and Mehra and Rechenberg [1982].
In condition F of sections III. I and III.5 of von Neumann [1932], it is stated that if a self-
adjoint operator R "corresponds" to a function !R on classical phase space, and F : R -+ R
is an arbitrary function, then F(R) (defined by the functional calculus) should correspond
to F 0 !R. Construing the correspondence !R 1-+ R as a quantization map R = Q/i(!R), this
stringent condition cannot hold in general, not even for F(t) = /2. For this choice, in view
of 1.(3.10) the condition is equivalent to (1.2) without the limit, explaining our terminology.
Dirac's condition (1.3), also without the limit, is proposed in §21 of his [1930] book; Dirac
did recognize that his condition could not always be satisfied, and added the qualifying
remark that the condition should be satisfied only by "the simpler" commutators.
The idea of deformation quantization, which appreciates the fact that the conditions
of von Neumann and Dirac can hold only asymptotically, goes back to Berezin [1974,
1975a,b] (also cf. Vey [1975] and Bayen et al. [1978]). The mathematical framework was
developed in a series of papers starting with Gerstenhaber [1964J. In the original setting
one constructs a "deformed" associative product ./i on a given Poisson algebra, in such a
way that f .r. g -+ f· g for Ii -+ 0 and j(f .r. g - g ·Ii f)/Ii converges to {f, g} in the same
limit. Here f .Ii g is defined by a formal power series expansion, and the Ii -+ 0 limit is
handled accordingly.
This subject of "formal" deformation quantization reached a high point in the work of
Fedosov [1994, 1996], who showed that every regular Poisson manifold P (or rather its
associated Poisson algebra COO(P, R» is quantizable in the given sense (regularity here
means that the rank of B~ is constant; cf. 1.2.3). Also cf. Weinstein [1994]. The culmination
of the subject is Kontsevich [1998], who proved that every finite-dimensional Poisson
manifold can be quantized in the sense of formal deformation quantization.
In its current development, formal deformation quantization is remote from quantum
mechanics and even from Hilbert space theory, using essentially different techniques from
the ones described in this book. Moreover, no version of von Neumann's condition is
imposed.
Strict deformation quantization was introduced by Rieffel [1989a, 1994], who in partic-
ular proposed what we call Rieffel 's condition. In his approach the norm and the product in
the C· -algebras 2(1i depend on Ii; in particular, the product in 2(/i is analogous to the product
.r. in formal deformation quantization. Further work in this setting, especially on Rieffel's
condition, may be found in Nagy [1992,1997, 1998a] and Blanchard [1996]. These papers
contain applications to the theory of quantum groups (in the C' -algebraic setting introduced
by Woronowicz [1987, 1995]; also cf. Lance [1995]), as do Rieffel [1993b, 1995], Nagy
[1993, 1998b], and Sheu [1996,1997]. Related work may be found in Landstad [1994] and
Landstad and Raeburn [1997].
Definition 1.1.1 is taken from Landsman [1993b]. The reformulation of strict (deforma-
tion) quantization in terms of the maps Q/i simply adopts the perspective of a physicist, who
looks at quantization in precisely this way. Mathematically, this reformulation is closely
related to the concept of E-theory and its associated asymptotic morphisms (see Connes
[1994]). The connection between E -theory, quantization, and operator K -theory is further
446 Notes

developed by Nagy [1996, 1997] and Rosenberg [1996]; for the last two topics also see
Rieffel [1993c].
We shall not discuss geometric quantization or prequantization in this book, referring
the interested reader to Souriau [1969, 1997], Kostant [1970], Sniatycki [1980], Kirillov
[1990], Woodhouse [1992], and Chernoff [1995]. The construction of the prequantization
line bundle provided by this technique is often useful in the context of Berezin-Toeplitz
quantization on Kahler manifolds (cf. the notes to 1.5). On the other hand, the prequantiza-
tion of functions on phase space does not easily fit into a C· -algebraic framework, because
the prequantization of a bounded function is always an unbounded operator, a property that
may persist even after the second step of quantization. Moreover, one works at a fixed value
of Ii.
11.1.2 See Fell [1962] or Dixmier [1977] for the traditional theory of continuous fields
of C· -algebras. Definition 1.2.1 is taken from Kirchberg and Wassermann [1995]; Lemma
1.2.2 and Proposition 1.2.3 show that their definition is equivalent to Dixmier's.
Blanchard [1996] defines a continuous field of C'-algebras over a locally compact
Hausdorff space X as a C*-algebra 11: equipped with a nondegenerate morphism from
Co(X) to the center of the multiplier algebra of 9 (cf. IV.2.1), such that 1.2.1.1 holds
with 21"" := I1:/Co(XYI1: (where Co(XY is the ideal in CoCX) of functions vanishing at x,
and ({J" the canonical projection. Condition 1.2.1.2 is then automatically satisfied, so that
one obtains a continuous field in the sense of Definition 1.2.1. Conversely, given 1.2.1, one
has ker«({Jx) = Co(XYI1:, so that the canonical isomorphism ((J,,(I1:) ~ 11:/ ker«({Jx) leads to
the equivalence between the two definitions in question.
The connection between strict (deformation) quantization and continuous fields of
C· -algebras was recognized by Rieffel [1989a]; it was initially thought that any such quan-
tization would define a continuous field, but it was quickly realized (Rieffel [1993a)) that
further assumptions were needed. Definition 1.2.5, which is a slight variation on a definition
proposed in Rieffel [1998], seems a good compromise between Rieffel's earlier definitions
and those in Landsman [1993b]. Results analogous to Theorem 1.2.4 are given in Nagy
[1992, 1998a].
Somewhat against the spirit of the founding fathers, one could omit Dirac's condition
from Definition 1.2.5. In some cases the ensuing continuous fields of C· -algebras (in which
210 is commutative) may nonetheless be seen as quantizations. For examples see Matsumoto
[199Ia,b], Matsumoto and Tomiyama [1992], Borthwick et al. [1993], and Exel [1994]. A
unified approach to these examples is developed in Abadie and Exel [1997], where Dirac's
condition reappears through the back door.
Partitions of unity are discussed in Pedersen [1989] and Jiinich [1994]. Since these will
often be used, we recall their definition. Let Q be a Hausdorff space, and let {Nal aEl be
a locally finite open cover of Q (i.e., each point of Q has a neighborhood that intersects
only a finite number of the sets Na ). A partition of unity subordinate to the given cover is
a collection of positive functions (UalaEl such that u" E Cc(N" , R) and LaEl U" = 1. A
partition of unity always exists when Q is paracompact; Hormander [1983] proves that the
u" may be taken to be smooth when Q is a manifold.
11.1.3 Definition 1.3.1, anticipated by Emch [1984], Rieffel [1989b], and Landsman
[1993a], is due to Nagy [1992, 1998a] and Blanchard [1996]. The last two authors look at 11:
as a Co(X) module, so that each set (w~ IXEx defines a Co(X)-linearfunction cp~ : 11: ~ Co(X)
by ((J~(A) : x ~ w~(Ax)'
Suppose one has a triple (11:, {21X , ({Jxl"Ex) as in Definition 1.2.1, that satisfies conditions
2 and 3. One may then still use Definition 1.3.1. Under the assumption that each 21"" is
 Quantization and the classical limit 447

separable and nonzero, Blanchard [1996] proves that such a triple satisfies 1.2.1.1 (so that it
is a continuous field of C*-algebms) iff it admits a continuous field of states. The proof uses
Kasparov's [1981] generalization of Stinespring's Theorem 1.4.2 (also see Lance [1995]).
Analogous results are given in Nagy [1992, 1998a]. Previous applications of continuous
fields of states by Rieffel [1989b] and Landsman [1993a] in proving Rieffel's condition
(1.1) in certain models may be seen as embryonic versions of these results; Proposition
1.3.6 is a case in point.
Definition 1.3.3 is an abstmction of the notion of a coherent state, rewritten in the language
of tmnsition probabilities. For the standard theory of coherent states and their various
generalizations, see Klauder and Skagerstam [1985], Perelomov [1986], Zhang et a1. [1990],
and Ali et al. [1995]. The usual definition stipulates as a minimal requirement that coherent
states form a family {nO'la E S} for which the map a H- nO' is (strongly) continuous,
and fsdp,(a)[nO'] = [weakly, for some measure p, on S; various requirements may be
added. For example, pammetric dependence on Ii and good behavior for Ii ~ 0 were
already studied in a special example by SchrOdinger [1926]. Further work on the role of
coherent states in the classical limit of quantum mechanics is cited in the notes to 2.7; also
cf. Simon [1980], Yaffe [1982], as well as the first two books cited above. From our point
of view, equation (1.11) is of centml conceptual importance, for it shows that the tmnsition
probabilities on qr.(S) that are inherited from JP'Hr. become classical when Ii ~ O.
The main ideas of what is here called Berezin quantization go back to Davies and Lewis
[1970], Holevo [1973] (cf. Davies [1976] and Holevo [1982] for a textbook presentation
of the approach in these papers), and Berezin [1972, 1974, 1975a,b] (also cf. Perelomov
[1986] for a summary of these four papers). Whereas the other authors concentrated on
operational ideas and measurement theory, it was the specific contribution of Berezin to
study operators of the type Qg (f) in connection with quantization theory and the classical
limit, in particular analyzing their Ii-dependence. In doing so he discovered, for example,
the "quantization" of Planck's constant when one quantizes a compact phase space. In view
of this, and of Berezin's premature death in a drowning accident (cf. Bogolyubov et a1.
[1981] and Dobrushin et al. [1996]), it seems reasonable to name the quantization method
involving Qg after him.
Equation (1.16) appears in Ali and Doebner [1990] under the name prime quantization
(with weaker conditions on the coherent states). Berezin [1972] calls f the contravariant
symbol of Qg (f); for an arbitmry bounded opemtor on 'It, the covariant symbol of an
opemtor A is the function on S defined by a H- (q,,(a»(A) (the terminology lower and
upper symbol, respectively, is also found in the litemture, e.g., Simon [1980]). Berezin
actually looks for operators whose covariant symbol is well-behaved for Ii ~ 0, and
regards such an operator as the quantization of the Ii ~ 0 limit of its covariant symbol. The
Berezin transform

Bf(p) = Is dp,,,(a) p(q,,(p), qr.(a»f(a),

which is well-defined as map from L""(S) to itself, maps the contmvariant symbol into the
covariant one. Our condition (1.9) evidently states that the Berezin transform becomes the
identity for Ii ~ O. For a study of the Berezin transform on so-called bounded symmetric
domains (these are certain bounded subspaces of eN; cf. Helgason [1978]) see Berezin
[1975a], Peetre [1990], Unterberger and Upmeier [1994], and Englis [1996].
A different approach to quantization theory based on coherent states is due to Klauder
[1988, 1995]. For the connection between geometric quantization and coherent states see
Rawnsley [1978], Thynman [1987b], Odzijewicz [1988,1992], and Rawnsley et al. [1990].
448 Notes

The coherent states constructed in these papers should satisfy Definition 1.3.3; on those
Kahler manifolds that are not coadjoint orbits the limit in (1.10) is strictly necessary,
and (1.13) does not hold. In fact, starting from Berezin-Toeplitz quantization on Kahler
manifolds (see the notes to 1.5), the easiest way to find J1,h is to use (1.18).
Bochner integrals, such as (1.16), will frequently occur in this chapter. The theory of
such integrals may be found in Yosida [1980]. A function! : S ~ B taking values in a
Banach space B is Bochner-integrable with respect to a measure J1, on S iff (i) ! is weakly
measurable (that is, for each functional WE B* the function a ~ w(f(a» is measurable),
(ii) there is a null set So C S such that {f(a)la E S\So} is separable, and (iii) the function
defined by a t-+ 1I!(a)1I is integrable. It will always be directly clear from this whether
a given operator- or vector-valued integral may be read as a Bochner integral; if not, it is
understood as a weak integral, in a sense always obvious from the context. The Bochner
Is
integral dJ1,(a)!(a) may be manipulated as if it were an ordinary (Lebesgue) integral.
For example, one has

11.1.4 Theorem 1.4.2 is due to Stinespring [1955]; also cf. Paulsen [1986] (where the
nonunital version may be found) and Kadison [1994]. Stinespring also proved 1.4.4.
Proposition 1.4.6 is equivalent to Theorem V.I. I in Berezanskii [1968]. Positive-operator-
valued measures are discussed abstractly by Riesz and Sz.-Nagy [1990], Appendix, and, in
the context of quantum mechanics, by Davies [1976], Holevo [1982], Busch et al. [1995],
and Schroeck [1996]. The measure theory in the proof of Proposition 1.4.8 is discussed in
Pedersen [1989], §4.5. Corollary 1.4.9, due to Neumark, is actually valid for any space X
with a a-algebra; see Schroeck [1996], §II.lI.F.
11.1.5 For coherent states see the notes to 1.3. The theory of Hilbert spaces with a
reproducing kernel may be found in Aronszajn [1950] or Meschkowski [1962]; also see
Ali [1985] for a summary. Schroeck [1996] contains a generalization to matrix-valued
reproducing kernels. Definition 1.5.6 and the ensuing theory are due to Schwartz [1964],
who develops a far-reaching generalization of the theory of reproducing kernels. Overviews
of the connection between coherent states, reproducing kernels, and POV-measures are given
by Davies [1976] and Ali and Doebner [1990].
If S is a complex manifold, (usually taken to be homogeneous and Kahler in this type
of application), and itt! consists of (anti-) holomorphic functions, operators of the type
(1.44) are known as (generalized) Toeplitz operators. (Strictly speaking, the term "Toeplitz
operator" refers to the case where S = SI and p projects onto the Hardy subspace of
functions with positive Fourier coefficients only.) The reproducing kernel in it h is then
known as a Bergman kernel, and p is sometimes called the Szego projection; we refer to
Meschkowski [1962] and Helgason [1978] for the first steps in the theory of this kernel.
See Boutet de Monvel and Guillemin [1981], Guillemin [1984], and Upmeier [1996] for
the theory of generalized Toeplitz operators (the latter book is particularly relevant, since
it describes the CO-algebras generated by these operators in great detail). More generally,
the projection p in (1.44) may project onto the space of (anti)holomorphic sections of
a holomorphic line bundle over S. In either case, the quantization procedure defined by
(1.44) is known as Berezin-Toeplitz quantization. In the context of homogeneous Kahler
manifolds this quantization was introduced by Berezin [1974, 1975a]; also cf. Guillemin
[1984], Berger and Coburn [1986], and Tuynman [1987a,b] for early work. More recent
work on Berezin-Toeplitz quantization is cited in the notes to 2.4.
 Quantization and the classical limit 449

11.2.1 For exhaustive information on the Heisenberg group, and nilpotent Lie groups
in general, see Corwin and Greenleaf [1989] or Leptin and Ludwig [1994]. Much useful
information on fin and its irreducible representations is also contained in Folland [1989].
The representation theory of Lie algebras by unbounded operators on infinite-
dimensional Hilbert spaces, which is relevant here as well as in the remainder of this
chapter, may be found in Warner [1972] or Barut and Ra~ka [1977]; some relevant notions
are also reviewed in 1II.!.5. For G = fin and U = Uf (or Uc) on 1t = L 2 (Rn), it is not
difficult to see that the space of smooth vectors is S(Rn); see Howe [1980] or Corwin and
Greenleaf [1989].
11.2.2 Proposition 2.2.1, as well as the construction of ph in 2.2.2, go back to van Hove
[1943]. In this context the Groenewold-van Hove theorem should be mentioned: This
states, roughly speaking, that there exists no decent map dph from Coo (T*Rn , R) to some Lie
algebra of unbounded operators on a Hilbert space for which (2.38) can be extended beyond
p:S2 and the restriction to p:S1 gives an irreducible representation of ~n. See Groenewold
[1946], van Hove [1943], Guillemin and Sternberg [1984b], Abraham and Marsden [1985],
and Gotay et al. [1996].
Equation (2.32) is a special case of the momentum map; see Ill!.!.
For 9 = .sp(n, R) and R = dpn on 1t = L2(Rn) a dense set of analytic vectors is given
by the linear span of the Hermite polynomials. By the integrability conditions proved in
the references cited in the notes to 2.1, there exists a unitary representation ph of Sp(n, R)
whose derivative in the sense explained above is indeed dpl!. The explicit form of pI! and the
metaplectic group Mp(n, R) are discussed in, e.g., Segal [1959], Bargmann [1961], Shale
[1962], Voros [1977], Guillemin and Sternberg [1984b], Littlejohn [1986], Folland [1989],
and Kirillov [1990]. A different approach to the construction of pn(Mp(n, R» is based on
the fact that Sp(n, R) is contained in the automorphism group of fin. Therefore, for each
M E Sp(n, R) the map h ~ UI/h(Mh) defines an irreducible representation of fin. which
in view of 2.!.4 is equivalent to Ul/Ii. The unitary implementer is ph(M); cf. (2.40). In any
case, it turns out that ph and pr.: are equivalent iff Ii and Ii' have the same sign.
11.2.3 The coherent states of 2.3.1 were discovered by SchrOdinger [1926]; also see the
notes to 1.3. In this case the Berezin transform becomes simply

cf. (2.117), where the prefactor of 1i!:l.2n is different.

Proposition 2.3.2 is due to Bargmann [1961]. He absorbs exp( -iii Ii) into the measure on
Cn , so that the elements of fi/i are entire functions. Like many other authors, he actually uses
holomorphic rather than antiholomorphic functions, but the latter choice is more natural in
the context of coherent states (cf. Klauder and Skagerstam [1985]). The Hilbert space of
entire analytic functions in L2(Cn, exp( -zz)dzazl(2Jl'i» is often called the Bargmann-
Fock space, since the use of this space in the context of the canonical commutation relations
goes back to Fock [1928] (who defined the inner product directly in terms of the Taylor
coefficients of a function, rather than through a Gaussian measure on cn). See Folland
[1989] for a thorough discussion.
Exponential Hilbert spaces, called bosonic Fock spaces by physicists, go back to Segal
[1956]; also see Guichardet [1972], who proves all claims we make on exponential vectors.
The realization of the "canonical" coherent states in such spaces is due to Klauder [1970].
Also cf. Klauder and Skagerstam [1985]. A very detailed analysis of creation and annihila-
tion operators is in Bratteli and Robinson [1981]. These operators derive their name from
450 Notes

their action on the subspaces ®~ JC of exp(JC), which we do not need; it turns out that a(z)
maps ®~JC to ®~-lJC (for n = 0 one has a(z)O = 0), whereas a(z)* maps it to ®~+lJC.
The Riemannian geometry of qr.(JC) as a submanifold of JP'exp(JC) is studied in Field
[1996]; the most interesting result is that qr.(JC) has zero intrinsic curvature with respect to
the induced Fubini-Study metric (the extrinsic curvature is nonzero).

11.2.4 Theorem 2.4.1 is due to Coburn [1992] and Borthwick et al. [1993]; both ac-
knowledge Klimek and Lesniewski [1992a] for the organization of the proof. Using less
crude estimates they show that (2.76) is even O(Ii). Our proof of 0.1) is different from these
references; the proof of nondegeneracy is taken from Berger and Coburn [1986]. By the non-
degeneracy of Qg and the open mapping theorem, (2.73) implies that IIflloo :::: en Qg(f)1I
for some C > 0, and f E Co(S). See Berger and Coburn [1994] for a study of the constant
C, and for deeper inequalities.
One can study Qg for function spaces larger than Co(cn), so that one leaves the compact
operators. This is not particularly useful for physics (it introduces spurious superselection
sectors), but leads to fascinating mathematical structures; see Guillemin [1984], Berger and
Coburn [1986], Coburn and Xia [1995], and Upmeier [1996].
Theorem 2.4.1 has an analogue for the Berezin-Toeplitz quantization of bounded sym-
metric (Cartan) domains: See Borthwick et aI. [1993], and Borthwick et al. [1995]. Riemann
surfaces have been treated in Klimek and Lesniewski [1992a,b, 1994, 1996]. The analysis
aspects of the proofs are similar to the one for e" , but one has to add detailed information
about the structure of such domains (which are related to Jordan algebras; cf. Upmeier [1987,
1996]). Also see Cahen et al. [1994, 1995] for an approach through formal (rather than strict)
deformation quantization. Bordemann et al. [1994] and Sheu [1996] apply Berezin-Toeplitz
quantization to certain compact Kiihlermanifolds. Here Ii can assume only quantized values
(cf. (1.12», as we will confirm in certain special cases in III.I.ll. See Cahen et al. [1993]
for the same problem in formal deformation quantization. A strict Berezin quantization of
the upper half-plane is given by Radulescu [1998]. In all cases discussed so far, one does
not obtain a strict deformation quantization from Berezin-Toeplitz quantization.
Theorem 2.4.3 is due to Berezin [1974, 1975a].

11.2.5 Weyl quantization is due to Weyl [1931], whose definition was (2.111). There is a
huge mathematical literature on this subject in the context of the theory of pseudodifferential
operators; principal sources are Grossmann et al. [1968], Voros [1977, 1978], Hormander
[1979, 1985a], Howe [1980], Robert [1987], Folland [1989], and Rieffel [1993a].
IdentifyingT*JRn withJR2n , one initially defines an isomorphism Q}r : S(JR2n) -+ S(JR 2n)
by (2.107). By duality, one then immediately has Q}r : S(JR2n) -+ S'(JR2n). The Schwartz
kernel theorem (cf. Reed and Simon [1975] or Hormander [1983]) identifies S'(JR2n) with
the space of continuous maps from S(lRn) to S'(JRn), so one eventually has a continuous
map Q}r (f) : S(JRn) -+ S'(JRn) for each f E S'(JR 2n).
To get back into the realm of (possibly unbounded) operators on L 2(JRn), as well as other
instances of good behavior (relevant to the theory of partial differential equations), one has
to impose certain restrictions on f. One firstly assumes that f E COO(JR2n), and secondly
imposes conditions on the behavior of f and its derivatives at infinity. If these are satisfied,
the expression (2.110), already meaningful as the Fourier transform of a distribution, makes
direct sense as a so-called oscillatory integral (Hormander [1983]), and defines the kernel
(2.109) as an element of S'(JR2n ). The corresponding operator Q}r(f) then maps S(JRn)
into itself (rather than into S'(JRn), as for general f). For example, Q}r (f) thus defined
lies in !Bo(L2(JRn» if f E CO'(T*JRn) (see Voros [1977]); as mentioned in the main text,
 Quantization and the classical limit 451

QJi : CQ'(T*]Rn) ~ !Bo(L2(Rn)) is not continuous. Oscillatory integrals have routinely

and correctly been computed by physicists since Dirac [1930].
In Rieffel [1993a] a similar procedure is used to define QJi (f) for functions f, all of
whose derivatives (including the zeroth) are bounded. Such functions form a dense subspace
of Cb(T*Rn), and the image of this space under QJi is contained in the C'-algebra of
pseudodifferential operators of order 0 (see, e.g., Cordes [1987, 1995]). Rieffel [1993a,
1994] shows that this defines a strict deformation quantization.
The more traditional calculus of pseudodifferential operators, which goes back to Kohn
and Nirenberg [1965], uses the "quantization"

Q~N(f)II1(x):= ( d n pdny eip(x-y)/fl f(p, x)II1(y),

JT*Rn (2n Ii)n
rather than (2.107). This is not useful for quantum mechanics, because Q~N does not preserve
self-adjointness. It is, however, local in x, so that it may effortlessly be extended to manifolds
(unlike QJr); see Taylor [1984].
Early mathematical studies of Weyl quantization are Segal [1963] and Pool [1966J.
Among other things, these authors showed that the map f ~ QJi (f) is unitary from
L2(T*Rn) to !B2(L 2(Rn»; this follows from a straightforward calculation. More recent
work is, for example, Daubechies [1980, 1983] and Gracia-Bondfa and Varilly [1988].
The deformed product defined by the Weyl quantization (which first occurred in von
Neumann [1931]) may formally be written as (Groenewold [1946])

(f 'flg) = (a
f exp [ --:- 8 - --
Ii - -
21 op oq oq op
a8)] g.

This is (historically inaccurately) sometimes called the Moyal product, after Moyal [1949].
Though an attractive formal expansion, the Moyal product plays no role in our setting, but
it can be given a precise meaning in various ways; see Voros [1977, 1978], Bayen et al.
[1978], Folland [1989], Gracia-Bondfa and Vanlly [1988], and Estrada et al. [1989].
The expression (2.95), and the ensuing connection between Weyl quantization, the parity
operator, and the Dirac delta function, are due to Grossmann [1976]. Equation (2.102) was
first written down by Royer [1977]. The idea of inverting expressions of the type (2.95) by
(2.97) goes back to Stratonovic [1957]. See Gadella [1995] for a review ofWeyl quantization
and Wigner functions from this perspective. Theorem 2.5.1 goes back to van Hove [1951];
also cf. Voros [1977], Hormander [1979, 1985a], Folland [1989], Graffi and Parmeggiani
[1990], and Borsari and Graffi [1994]. Our proof stresses the role of the parity operator.
Wignerfunctions were introduced in Wigner [1932]. Moyal [1949] was the first torecog-
nize the connection between Weyl's quantization and Wigner's function. One may evidently
define the Wignerfunction ofa mixed state through (2.101) as well; in fact, in physics these
functions are often used in quantum statistical mechanics. See Hillery et al. [1984] for a sur-
vey ofWignerfunctions in nonrelativistic physics (cf. de Groot et al. [1980] and Carinena et
al. [1990] for the relativistic case), and Folland [1989] for mathematical aspects. The physi-
cal interpretation of the Wigner function is that it is to some extent a probability distribution
on phase space. For the right-hand side of (2.10 I) is of the form IT*Rn Pf, which looks like
the expectation value of f in a mixed state p in classical mechanics. However, in classical
mechanics p is a probability measure, which the Wigner function fails to define because
of its potential non~ositivity. This failure is a consequence of the uncertainty relations of
quantum mechanics, which forbid sharp localization in phase space; see, e.g., Schroeck
[1996].
452 Notes

Positivity of the Wigner function is actually quite rare: Hudson [1974] shows that for
pure states WhL", J is positive iff \11 is a complex Gaussian; also cf. Folland [1989], and see
Brocker and Werner [1995] for the case of mixed states.

11.2.6 Theorem 2.6.1 is due to Rieffel [1993a, 1994], who actually proved it for the larger
function space ila = Cg"(T*IR", 1R) of smooth functions that together with all derivatives
are bounded (cf. the notes to the previous section). Our proof is different from Rieffel's,
and derives from Landsman [1993b]. A third approach to results of this type may be found
in Elliott et al. [1996]. Equation (2.117) was found by Berezin [1974].
A sharper version of Lemma 2.6.2 was originally proved by Calderon and Vaillancourt
[1971] in the setting of the Kohn-Nirenberg calculus; a simple proof may be found in
Hwang [1987]. See the references at the beginning of this section for the Weyl version.
Both versions playa central role in the theory of pseudodifferential operators.
The smearing (2.116) goes back to Husimi [1940]; his motivation was that it leads to
a positive phase space distribution function (which in our setting is the analogue of the
Wigner function for Berezin quantization) W~ L", J : (p, q) ~ Ii-II p( ",~P.q), "'), which is
sometimes called the Husimi function; cf. Lee [1995] for a recent survey of its applications
in physics. It is easily shown (cf. Prop. 1.99 in Folland [1989]) that replacing",g in (2.116) by
an arbitrary pure state"', where \11 E L2(IR"), defines a collection of positive maps as well.
These more general maps may not always correspond to a strict deformation quantization,
though.
Proposition 2.6.3 is due to Helffer et al. [1987].
Convolution algebras of may be defined for any locally compact group; see Ill. 1.7.
The unitary transformation W is the Plancherel transform for fIn, and the measure
dhllil n/(27f)2n is the Plancherel measure on the unitary dual of fIll (up to a set of Planche rei
measure zero, namely the collection of one-dimensional representations of fIll)' See Dixmier
[1977] for this transform for locally compact unimodular groups in general, and Corwin
and Greenleaf [1989] for the details for nilpotent Lie groups.
Proposition 2.6.4 has its roots in Dixmier [1960]; for a modem approach see, e.g., Packer
and Raeburn [1992], Thm. 1.2 and Example 1.4.(2). Lemma 2.6.6 and Corollary 2.6.7 are
stated without proof in Elliott et al. [1993].

11.2.7 Proposition 2.7.1 is the easiest version of a number of results in the literature that
may be seen as adaptions of Egorov's theorem in the theory of pseudodifferential operators
(cf. Taylor [1984] or Hormander [1985a]) to the setting of quantization theory. Our approach
follows Rieffel [1996], who considered the special case QJj = QJr, and proved the stronger
version in which hand f are allowed to be in Cg"(T*IR", 1R). The strongest result is Thm.
IV-9 in Robert [1987] (also cf. Prop. 1.5 in Helffer et al. [1987]), who proves (2.131) for
Qn = QJr and f E Cg"(T*IR", 1R) under the following assumption on Hh: There is a
classical Hamiltonian h E C""(T*IRn , 1R) satisfying

for each multi-index (a, (3) (the notation is explained below (2.114», and II H n - QJr (h)1I =
D(Ii). This includes all Hamiltonians considered by Rieffel [1996], as well as certain
unbounded ones; cf. Theorem 2.5.1.
In all these cases, the convergence in (2.131) may actually be shown to be uniform in t;
the same comment applies to (2.135) and (2.162). The completeness of the classical flow
of h follows from Prop. 2.1.21 in Abraham and Marsden [1985].
 Quantization and the classical limit 453

Under similar assumptions, one may show that Qh(ot~(f» '" ot~(Qh(f» + O(Ii) as an
asymptotic expansion in the sense of pseudodifferential operators; see Wang [1986], Robert
[1987, 1998], and Paul and Uribe [1995].
The theory of SchrOdinger operators may be found in Reed and Simon [1975, 1978],
Cycon et aI. [1987], and Hislop and Sigal [1996]. For a positive potential, self-adjointness
in an external magnetic field is guaranteed by a theorem of Leinfelder and Simader [1981]
(also cf. Cycon et aI. [1987]), which states that (2.134) is essentially self-adjoint on C,:"'(lRn)
when V ~ 0, V E L~(lRn),Ai E L!c(Rn)foreachi,and8iAi E L~(Rn).Alsocf.Combes
et al. [1978] and Hogreve [1983].
Theorem 2.7.2 generalizes a result of Hepp [1974], who proved the special case A = 0,
Qh = =
Q:-, and !(p, q) expi(uq - pv). For the inclusion of abelian as well as nonabelian
(Yang-Mills) gauge fields see 111.2.11 and its notes.
Hepp assumes only V E c2+8(lRn) for some 8 > 0; the proof then uses the Holder
continuity of V(2) rather than the second-order Thylor series of V. The meaning of the
various steps in Hepp's proof needed some clarification. Much useful information on small
fluctuations, coherent states, and their connection with the metaplectic representation may
be found in Littlejohn [1986]. The theory of linearizing Hamiltonian equations of motion
is in Marsden et al. [1991]; also cf. Marsden and Ratiu [1994] for a quick review. For
arbitrary symplectic spaces S, the linearization in terms of h 1 proceeds exactly as in the case
S = T*Rn. However,linearization along a classical trajectory with h2 is more complicated
in the general case, as one needs a so-called symplectic connection to identify the tangent
spaces TO'(I)S with TO'S, as well as to define the second derivative h".
Proposition 2.7.3 is of interest in its own right, as it shows that the quantum fluctuations
around the classical path are controlled by the quadratic term in the expansion of H/i around
this path; see Hepp [1974] and Littlejohn [1986].
Further work in the direction of Theorem 2.7.2 is Yajima [1979], Hagedorn [1980,1981,
1985], Robert [1987, 1998], Robinson [1988a, 1988b, 1993], Wang [1991], Combescure
[1992], and Arai [1995]. Heuristic work includes Kurchan et al. [1989], Barnes et al. [1994],
and Nauenberg et aI. [1994], who review experiments on the semiclassical evolution of
coherent states).
Hepp's approach differs fundamentally from the time-dependent WKB method (cf.
the Introductory Overview), for which we refer to Truman [1976, 1977], DeWitt-Morette
et al. [1979], Schulman [1981] (these two references explain what happens to the WKB
approximation beyond caustics; also cf. DeWitt-Morette et al. [1983]), Maslov and Fedoriuk
[1981], Saksenaet al. [1991], Maslov [1994], and Robert [1998]. As in the time-independent
case, rigorous work on the WKB approximation has been submerged into the theory of
Fourier integral operators (see Hormander [1985b]); the connection between the two is
discussed in Guillemin and Sternberg [1977], Voros [1977, 1978], Robert [1987, 1992,
1998], and Paul and Uribe [1995]. Rather then follow any of these authors, a C* -algebraist
would rather introduce Ii into microlocal analysis through the theory of pseudodifferential
operators on a tangent groupoid; cf. Nistoret al. [1997] and Monthubert and Pierrot [1997].
A sample of other approaches to the semiclassical behavior of nonstationary states is
Marsden [1974], Albeverio and H0egh-Krohn [1977], Voros [1977, 1978], Berry and Balazs
[1979], lona-Lasinio et al. [1981], Blanchard and Sirugue [1985], Slawianowski [1991],
Saksena et al. [1991], Omnes [1994, 1997a], Osborn and Molzahn [1995], Paul and Uribe
[1995], Werner [1995], and Rezende [1996]. References to literature on the connection
between the classical limit of quantum mechanics and the small-time limit of diffusion
processes may be found in the notes to 3.7.
454 Notes

The semiclassical approximation of eigenvalues and eigenstates of Schrooinger operators

is a topic with a different flavor, and a huge literature. The traditional theory of the time-
independent WKB method is contained in Maslov and Fedoriuk [1981], and is summarized
in Arnold [1989]. Geometric aspects of Maslov's version of the time-independent WKB
method are discussed in Guillemin and Sternberg [1977], Arnold and Givental [1990],
Woodhouse [1992], Littlejohn [1992], and Bates and Weinstein [1995]. An interesting
heuristic overview may be found in Gutzwiller [1990]. Analytical aspects, closely related
to microlocal analysis (cf. Hormander [1985a,b]), are reviewed in Robert [1987, 1992,
1998] and Helffer [1988]. Different analytical techniques, closer to the usual theory of
Schrooinger operators, are described in Hislop and Sigal [1996].
A given (normalized) energy eigenfunction II1Eh (with eigenvalue Eli) defines a pure
state Wfh on mh = 1B0(L2(jRn» by 1.(1.46). In the context of Definition 1.3.1, in which
the continuous field of CO-algebras is given by Theorem 2.6.5, one may try to choose the
sequence (EIi}h in such a way that Wfh converges to a state wg on mo = Co(T*jR"), that is,
to a probability measure J-Lg on T*jRn. A necessary condition for this to happen is that the
sequence {En} converges to some E; there is a formidable literature on the behavior of such
sequences of eigenvalues, reviewed in Robert [1992, 19981. It follows from Lemma 1.3.2,
Theorem 2.6.5, and (2.52) that wg, if it exists, is the weak (vague) limit of the sequence
(J-LEh In of measures on T*Rn, defined by dJ-L Eh(p, q) := (21T n)-" p(1{!AP.q) ,1{! E h)d" pdnq.
For example, if Hn = Q:i(h) and the flow of the classical Hamiltonianh on h-1(E) C
T*Rn is ergodic, then under suitable assumptions almost all sequences of the above type
converge to the Liouville measure on h-1(E); see Helffer et at. [1987]. WKB states II1n "-
exp(i Sin), on the other hand, converge to measures supported on the so-called Lagrangian
submanifold defined by S; cf. Colin de Verdiere and Parisse [1994] and Werner [1995].
Further work in this direction may be found in Knauf [1989], Duclos and Hogreve [1993],
and Paul and Uribe [1996]. It is quite remarkable that pure quantum states may well converge
to mixed classical states.

11.3.1 The necessary background in affine and Riemannian geometry may be found in,
for example, Helgason [1978], Klingenberg [1982], Gallot et a1. [1990], Choquet-Bruhat et
a1. [1982], or Lang [1995].
For the group gQ and its action on T* Q cf. Guillemin and Sternberg [1984b] or Isham
[1983]; also cf. the notes to 3.6. For the topology of Diff( Q), and some of its modifications,
see Ebin and Marsden [1970], Marsden [1974], Ismagilov [1996], and Omori [1997].

11.3.2 In the Riemannian case, geodesics satisfying (3.4) are automatically affinely
parametrized; this means that t is an affine function of the length of the geodesic. The
equation satisfied by arbitrarily parametrized geodesics contains additional terms.
A proof of Proposition 3.2.2 may be found in most books on the subject; e.g., Helgason
[1978] (Thm. 1.6.2), Gallot et a1. [1990] (Thm. 2.92), or Klingenberg [1982] (Thm. 1.9.7).
Theorem 3.2.3 is named after Hopf and Rinow; a proof may be found in Klingenberg
[1982], Thm. 2.1.3, or Gallot et a1. [1990], Thm. 2.103 and Cor. 2.105. Theorem 3.2.5 is
in Klingenberg [1982] (Thm. 2.1.14) and Gallot et at. [1990] (prop. 2.113 and Scholium
3.78).

11.3.3 An extensive treatment of Riemannian geodesic and cogeodesic motion is in Klin-

genberg [1982]; also cf. Abraham and Marsden [1985]. Theorem 3.3.4 is closely related
to Lemma 3.1.17 in Klingenberg [1982], and is suggested by comments in Marsden et a1.
[1991]. Linearization in the second sense explained in 2.7 (Le., using the quadratic Hamilto-
nian (2.143» is possible but awkward in the general Riemannian case, since the symplectic
 Quantization and the classical limit 455

connection needed to identify the tangent spaces along a geodesic will usually not coincide
with the Levi-Civita connection. This leads to cumbersome expressions.
One may equip T Q with a metric 9 such that the Hamiltonian flow on T Q is geodesic
with respect to g. Using the decomposition (3.3), the metric 9 is simply defined by declaring
the two copies of Tq Q to be orthogonal, and putting 9 = g on each copy. It follows from the
construction of 9 that the projection i : T Q ~ Q is an isometric submersion in the sense
that i * is an isometry between the orthogonal complement of its kernel and its image. This
implies that the horizontal lift of a curve is a geodesic in T Q iff the curve is a geodesic.
It is then immediate from this property, combined with the definition of horizontal curves
in T Q and the description of the flow in T Q in 3.3.3, that the Hamiltonian flow on T Q is
geodesic with respect to g.

11.3.4 This section is mostly adapted from Landsman [1993b].

For the normal bundle see, e.g., Klingenberg [1982], 1.3.11-1.3.14 (which includes a
detailed discussion of the diagonal map 8), and many other books on differential geom-
etry. Theorem 3.4.2 is the tubular neighborhood theorem; see Lang [1995]. The proof
uses a partition of unity argument; recall that our definition of a manifold includes para-
compactness. Our choice (3.48) corresponds to the choice of {X +- X} as a complement
to 1(q,q)8(Q) in 1(q,q)(Q x Q). The more commonly used map v';(Xq) = (exPq(X q ), q)
corresponds to the complement T. Q ffi 0 = {X +O}.
The geometric meaning of vi~ explained after Lemma 3.4.3 may be used to give an
elegant direct proof of the tubular neighborhood theorem.
The abbreviation C~ in (3.49) stands for Paley-Wiener, and 210 consists of the COO
functions I on T* Q satisfying the following conditions: (i) the support of I, projected to
Q, is compact; (ii) for each fixed q E Q the function 1(', q) on T; Q has an extension to
Tq* Qc ~ 1C" as an entire analytic function Iq such that for every N there are constants
C N and H (independent of N) such that Ilq(z)1 :::: CN(l + Izl)-N exp(H Imz). These are
simply the conditions of the Paley-Wiener theorem, which characterizes functions whose
Fourier transform is in C~(JR.n); see Reed and Simon [1975] or Hormander [1985a].
When Q is compact, one can show that the Wigner function (3.53) is continuous. The
proof is analogous to the flat-space case (cf. the comment following (2.102»; see Landsman
[1993b]. In the noncompact case one shows that W" E Co(T*Q, JR.) if the following condi-
tion is satisfied by the metric g on Q: The constant SUPQloQ2 [dp,(p(q,; q2»/dp,(q2)] should
be finite. Here the supremum is taken over all pairs of points that can be connected by a
unique geodesic, and p(q,; q2) is the geodesic reflection of q, inq2 (that is, p(q,; q2) = y(I),
where y is the affinely parametrized geodesic for which y(O) = q, and y(l/2) = q2).
There are alternative attempts to generalize Weyl quantization to Riemannian manifolds.
The proposal by Underhill [1978] (who only assumes the existence of an affine connection
on T Q) corresponds to a Wigner function where, compared with (3.53), the factor J is
absent. Liu and Qian [1992], on the other hand, have our factor J as well as an additional
factor j(y(q, v, t), v; -/0'/2. As mentioned in their paper, this quantization is not always
self-adjoint, which can be traced back to this extra factor. Different from all these is the
proposal of Emmrich [1993a], who did not compute the Wigner function (but cf. the notes
to 3.7). For yet another Wigner function in curved space see Habib and Kandrup [1989].
A different approach is due to Upmeier [1991], who generalized the definition (2.91)
of flat-space Weyl quantization (rather than (2.107»; also cf. Unterberger and Unterberger
[1988]. This requires that the Riemannian structure on Q admits the analogue of the parity
operator P, that is, Q must be a symmetric space (see Helgason [1978]). One may also
attempt to generalize (2.99), (2.100); this turns out to be possible if the phase space is
456 Notes

a homogeneous symplectic manifold, see Vanlly et aJ. [1990], Figueroa et aI. [1990],
and Gracia-Bondia [1992]. Another generalization ofWeyl quantization requiring a group
structure may be found in Manchon [1993].
If one is interested merely in quantizing functions on T* Q that are independent of,
or linear in p, one does not even need an affine connection; the manifold structure of Q
suffices. See Abraham and Marsden [1985]. The analogue of the (non-self-adjoint) Kohn-
Nirenberg "quantization" (cf. the notes to 2.5) on a manifold with connection is developed
by Bokobza-Haggiag [1969], Widom [1980], and Pflaum [1995].
11.3.5 The proof is from Landsman [1993b], with certain improvements added. Much
useful computational information on Jacobi fields and their associated determinants is in
Azencott et a1. [1981] and in Molzahn et aJ. [1990]. The Jacobians (3.59) and (3.61) may
alternatively be expressed through the derivative of the exponential map, which itself may
be written in terms of Jacobi fields.
A different way to construct a positive quantization on a Riemannian manifold is given
in Colin de Verdiere [1985], who uses a so-called Friedrichs "quantization" (see Thylor
[1984]). A general construction of coherent states on Riemannian manifolds may be found
in Paul and Uribe [1995,1996].
11.3.6 Proposition 3.6.2 goes back (at least) to Goldin [1971] and Goldin et aJ. [1980],
who see quantization theory on Q as the problem of finding general unitary representations
of 9Q. A closely related approach is developed in Doebner and Tolar [1975] and Angermann
et aJ. [1983]. See Isham [1983] and Ali and Goldin [1991] for reviews; also cf. Albert in
[1991], Chernoff [1995], and Ismagilov [1996].
Theorem 3.6.3 is due to Landsman [1993b]. When I{J is not an isometry, the classical and
the quantum action of 9Q are related only in the limit Ii ~ 0; see the above reference, and
Lemma 3.7.6 for a special case.
Proposition 3.6.4 is stated without proof in Abraham and Marsden [1985], which contains
the lemma used in the proof as Lemma 2.6.13; also cf. Theorem VIII. 10 in Reed and Simon
[1972].
11.3.7 Quantum theory on Riemannian manifolds was discussed almost immediately after
the birth of modern quantum mechanics; cf. Dowker [1974] for early references.
The coefficient ~ in (3.93) is sensitive to the precise quantization scheme that is used;
the scheme in Liu and Qian [1992] produces i, Underhill [1978] finds fi, whereas the
i
1
value has been found from geometric quantization (cf. Sniatycki [1980], Woodhouse
[1992], and Wu [1998]). Emmrich [1993a] obtains the value zero. Our seems somewhat
preferred by physicists: It is equivalent to having an extra term -iii R in the classical
Lagrangian, which is "naturaIIy" induced by the measure in the path integral on curved
space (see Dowker [1974]). The need to have the Ricci scalar in the quantum Hamiltonian
was apparently first recognized by Pauli in 1950 (see Pauli [1973], pp. 161-174).
The proof of Theorem 3.7.3 is mainly based on Strichartz [1983]; also cf. Davies [1989].
Alternative proofs are in Chernoff [1973], Cheeger et al. [1982], and Cordes [1987]. These
references, as well as Rosenberg [1997], should also be consulted for additional results in
the analysis of the Laplace-Beltrami operator. Aspects of the theory of unbounded operators
used in our proof may be found in Reed and Simon [1975]. The existence of \111 in case
that"K is not self-adjoint follows from the Corollary to their Thm. X.I on p. 137. For
elliptic regularity see §IX.6 of this reference. Completeness is sufficient but not necessary
for essential self-adjointness on C~(Q): For examples where (Q, g) is incomplete but f).
is nonetheless essentiaIIy self-adjoint on C~(Q), see Horowitz and Marolf [1995]. On the
other hand, it is easy to give examples where (Q, g) is incomplete and f). fails to be essentially
 Groups, bundles, and groupoids 457

self-adjoint on C~(Q). The simplest one is Q = (0, I) with flat metric (cf. Reed and Simon
[1975], p. 178). More generally, boundary value problems always involve Laplacians whose
different self-adjoint extensions describe the possible boundary conditions one imposes on
the solutions. See, e.g., Berezanskii [1968] or Cordes [1987] for a modem treatment.
For the Kato-ReUich theorem see Reed and Simon [1975]. It states that A + B (where A
and B are densely defined linear operators on a Hilbert space) is self-adjoint on D(A) (and
essentially self-adjoint on any core of A) if D(A) 5; D(B) and IIBIIIII :::: a II A 111 II + blllllil
for all III E D(A) and some a < I (called the relative bound) and arbitrary b. In our
application A = d, and B = R is bounded, so its domain as a multiplication operator is all
°
of L2(Q). Hence a = and b = IIRlloo = SUPq IR(q)l.
Theorem 3.7.5 is analogous to a result of Hogreve [1983], which does not involve Wey]
quantization and is a direct generalization of Hepp's version of Theorem 2.7.2 (cf. the notes
to 2.7). His proof is somewhat different from ours; a third proof, using the time-dependent
WKB approximation to the kernel of the propagator, is in Landsman [1993b]. A major
advantage of the present proof along the lines of Hepp [1974] is that there are no difficulties
with caustics (see below). As in the entire chapter, the metric is assumed to be smooth,
though for the proof to go through this could be relaxed to be C S • It would be interesting
to generalize 3.7.5 to arbitrary, and particularly to incomplete Riemannian manifolds. See
Paul and Uribe [1995] for a micro local approach.
Elworthy and Truman [1981] give a certain analogue of Proposition 2.7.3 that for suf-
ficiently small t is valid for arbitrary Riemannian manifolds, but holds for initial wave
functions of the WKB type III = p exp(i S / h) rather than for coherent states; also cf.
DeWitt-Morette et al. [1979], Schulman [1981], and Elworthy et a1. [1985]. In general, the
time-dependent WKB approximation in curved space that is used in these papers suffers
from similar problems with caustics as in the flat case. In contrast to the flat case, the
geodesic WKB-like approximation to the kernel of the propagator now has problems with
caustics as well because of the cut locus. See Molzahn et a!. [1990, 1992] for the approxi-
mation up to the cut locus, and DeWitt-Morette et al. [1979] and Azencott et a1. [1981] for
the situation beyond it.
The limit h -+ 0 in quantum mechanics is similar to the limit t -+ 0 of a diffusion
process; cf. Nagasawa [1993] for the general connection between the Schrodinger equation
and diffusion theory. The relation between small-time diffusion and the classical limit of
quantum mechanics on Riemannian manifolds is analyzed in DeWitt-Morette et al. [1979],
Elworthy and Truman [1981], and Azencott and Doss [1985].
We have not studied the connection between eigenfunctions of the Laplace-Beltrami
operator and geodesic flow; the enormous literature on this topic is reviewed in Robert
[1992, 1998]. In fact, the ergodicity result of Helffer et a1. [1987] quoted in 11.2.7 was di-
rectly inspired by an analogous theorem concerning ergodic geodesic flow on a Riemannian
manifold, whose final proof is due to Colin de Verdiere [1985]. Further references may be
found in the notes to III.2.11.

Chapter III
111.1.1 The Lie-Poisson structure, coadjoint orbits, and the momentum map (in a special
case) are all in Lie [1890], but were rediscovered in the sixties in the work of KirilIov,
Souriau, Kostant, Smale, and others. The momentum map, whose original definition was
(1.8), plays a central role in modem symplectic geometry. For textbook accounts, which
458 Notes

contain most results in this section, as well as in 1.2 and 1.4, see Guillemin and Sternberg
[1984b], Abraham and Marsden [1985], Libermann and Marie [1987], Marsden and Ratiu
[1994] (which contains historical notes, as well as a generalization of the momentum map
to Poisson manifolds), and Souriau [1969, 1997]. These accounts all start from actions of G
rather than g, but much of the theory depends only on the g-action, and is valid even when
it is not integrable to a G-action.
General references for the cohomology of Lie groups and Lie algebras are Guichardet
[1980] and de Azcarraga and Izquierdo [1995] (which contains many interesting applica-
tions to physics, particularly to the theory of anomalies). In the present context also cf.
Guillemin and Sternberg [1984b] and Libermann and Made [1987], as well as Varadarajan
[1985] for central extensions of Lie algebras. The vanishing of H1(g, JR.) and H 2(g, JR.) for
semisimple Lie algebras is known as the Whitehead lemma(s).
111.1.2 Symplectic group actions are "canonical" in the terminology of classical me-
chanics, and are therefore a "classical" subject. See the notes to the preceding section for
references. Theorem 1.2.1 is due to Palais [1957], which contains the full proof. A text-
book account is in Hector and Hirsch [1986], §3.1.3. This result is particularly interesting
in comparison with the well-known theorems on (essential) self-adjointness of unbounded
symmetric operators on Hilbert spaces (for which see Reed and Simon [1972, 1975]).
The construction starting with (1.21) was introduced by Souriau [1969, 1997]. A good
account is also in de Azcarraga and Izquierdo [1995].
Corollary 1.2.9 is due to Gotay and Thynman [1991], whose proof is entirely different.
Our proof relies on a theorem of van Est [1953] (also see Guichardet [1980]) to the effect
that for a compact Lie group the cohomology with coefficients in any (linear) representation
space is trivial.
The examples in 1.2.11 are "classical", but the modern formulation is due largely to
Souriau [1969, 1997].
111.1.3 The analytic theory of multipliers for Lie groups and Lie algebras may be found
in the book by Varadarajan [1985]. Our Theorem 1.3.3 is Thm. 7.21 of that book, and
our Corollary 1.3.7 is a special case of Thm. 7.37. The last part of our Proposition 1.3.4 is
Varadarajan's Cor. 7.30. Also cf. Guichardet [1980] and de Azcarraga and Izquierdo [1995].
The geometric approach to multipliers and central extensions may be found in the en-
lightening article of Thynman and Wiegerinck [1987], who introduced the cohomology
group H2(G, U(l» as defined in our main text (the analogous and better-known cohomol-
ogy group defined on the basis of smooth cocycles is only relevant to topologically trivial
central extensions). In particular, H2(G, U(l» is not necessarily isomorphic to H 2(g, JR.),
since not all r E Z2(g, JR.) can be integrated to an element C E Z2(G, U(l».
Proposition 1.3.6 is from Neeb [1996b], who reformulated a condition due to Thynman
and Wiegerinck [1987]. There are two alternative formulations of the condition stated in the
second item. Firstly, regard r as a 2-form on G by left translation, and identify X with the
corresponding right-invariant vector field ~: on G. A necessary and sufficient condition for
r to define a U(l)-extension of G with Lie algebra gr is that for each X E 9 the I-form ix r
a
be exact. Secondly, consider the group c appearing in the proof of 1.3.6. The adjoint action
of JR. Cae on the Lie algebra gc = gr is trivial (since JR. is central in a), so that the adjoint
a
action of c quotients to a well-defined action of = a aclJR. a
on gc. Since G = /1fl(G),
we obtain an action of 1fJ(G) on gc by restriction of the a-action. The second equivalent
criterion is that this 1f} (G)-action be trivial.
From the proof of 1.2.9 we therefore infer that H2(G, U(l» = 0 for compact simply
connected G, and H2(g, JR.) = 0 for a Lie algebra 9 whose associated simply connected Lie
 Groups, bundles, and groupoids 459

group is compact. Since this restriction implies that g is semisimple (see, e.g., Thm. 3.8.2
in Barut and Ra\=ka [1977]), the latter statement alternatively follows from the more general
property stated after the proof of 1.1.12.
111.1.4 For the initial part of this section cf. Abraham and Marsden [1985], Libermann
and Marie [1987], and Marsden and Ratiu [1994].
Theorem 1.4.4 is due to Kirillov [1976]. Corollary 1.4.8 has an alternative version,
in which transitive symplectic G-spaces correspond to orbits in the space of 2-cocycles
Z2(g, JR.); this has the advantage that only a single G-action on Z2(g, JR.) needs to be con-
sidered, and the disadvantage that Z2(g, R) is less intuitive than g*. See Guillemin and
Sternberg [1984b].
Martinez Alonso [1979] proved that given G, there exists a single central extension 0
(which was first constructed by Carinena and Santander [1975] in the context of projective
representations) such that any transitive symplectic G-space is a coadjoint orbit of 0 (or a
covering space thereof).
111.1.5 Projective unitary group representations have a long history; the traditional theory
may be found in Varadarajan [1985]. Our Proposition 1.5.1 is essentially his Thm. 7.5.
A modem presentation is given by Rieffel [1979], whose work, among other things,
removes a number of separability assumptions in Varadarajan [1985]. In particular, the
approach based on (1.64) is due to Rieffel.
For smooth vectors for U see Warner [1972] and Barnt and Ra\=ka [1977]. A seemingly
alternative treatment of dU(g) is in terms of the Garding subspace HG C H, which
consists of all vectors of the type fG dx/(x)U(x)\If, where / and \If run through C~(G)
and H, respectively. The density of H't/ in H is then most easily proved by showing that
HG ~ H't/, and subsequently that HG is dense in H. However, it is shown in Dixmier and
Malliavin [1978] that for connected G one actually has the equality H't/ = H G.
Sufficient conditions for the integrability ofrepresentations of Lie algebras by unbounded
operators are reviewed in Barut and Ra\=ka [1977].
111.1.6 The seminorms Pal"'"n \If := IIdU(Tal)· .. dU(Tan)\If II define a topology on H't/,
relative to which each dU(X) is a continuous map on H't/, and U (seen as a map from
G x H't/ to H't/) is separately continuous; see Corwin and Greenleaf [1989], Appendix.
The analysis of the momentum map on H't/ is done in Michor [1990], who uses a
particular notion of smoothness in infinite-dimensional manifolds due to Frohlicher and
Kriegl [1988]. He shows that the G-action on H't/ (and hence on Ini.'t/) is smooth, and that
the momentum map (1.71) is smooth as well. The manifold lP'H't/ is weakly symplectic.
A great deal is known about the image of the momentum map for infinite-dimensional
group representations; see Wildberger [1992], Arnal and Ludwig [1992], and Neeb [1995,
1996a]. Also cf. the notes to 1.10.
For the usual enveloping algebra see, e.g., Warner [1972] or Barnt and Ra\=ka [1977].
The author is indebted to A. Kent for the proof of Proposition 1.6.4.
Definition 1.6.5 is taken from Landsman [1993c]. Theorem 1.6.7 goes back to Berezin
[1967]; alsocf. Gutt [1983].
111.1.7 The theory in this section is usually discussed in the general setting of locally
compact groups. For a discussion of unimodularity see, e.g., Gaal [1973]. A sufficient
condition for a locally compact group G to be unimodular is that the identity e have a compact
neighborhood that is invariant under inner automorphisms. It follows that all compact and
all locally compact abelian groups are unimodular. Also, G is unimodular if it coincides
with its commutator subgroup; this applies to all semisimple Lie groups.
460 Notes

A detailed discussion of group C' -algebras may be found in Dixmier [1977] and in
Pedersen [1979]. Twisted group CO-algebras were introduced in Auslander and Moore
[1966], and further studied in, e.g., Kleppner and Lipsman [1973], Green [1978], and Packer
and Raeburn [1992]; also consult the review by Rosenberg [1994]. Theorem 1.7.3 is a special
case ofThm. 3.3 in Busby and Smith [1970]. Twisted group C' -algebras belong to the class
of twisted covariance algebras; references on the structure of such algebras are Green
[1978] and Packer and Raeburn [1989, 1990]. For (untwisted) covariance algebras, also
called crossed products, see the notes to 3.4.
Amenability is discussed, e.g., in Pedersen [1979] and Paterson [1988]. The original def-
inition (due to von Neumann) is different from the one above, and amounts to the existence
of an invariant mean on the C'-algebra Cb(C). The fact that the amenability of C implies
C;(C, c) = C'(C, c) follows from Thm. 3.11 in Packer and Raeburn [1989]; see Packer
[1994] for more general results in this direction.
111.1.8 For the Peter-Weyl theorem and the Plancherel transform in the unimodular case
see Dixmier [1977], The nonunimodular case is treated in Kleppner and Lipsman [1972,
1973], which also contains various results equivalent to Theorem 1.8.1 and Corollary! .8.3.
For the theory of induced representations see the notes to 2.9.
Lemma 1.8.2 is a rather trivial case of the "Mackey machine", in which one constructs
representations of a group extension by inducing from representations of the normal sub-
group defining the extension; see Mackey [1958], Green [1978], Rieffel [1979J, and the
review by Rosenberg [1994].
Proposition 1.8.4 is well known, and contained (usually in practically unrecognizable
generalizations) in all references for twisted group C'-algebras given above. The equality
C*(lE2 n, c) = C;(lR 2n, c) alternatively follows from the fact that JR2n is amenable; see the
notes to the preceding section.
Equation (1.121) has the following generalization to the case where c is degenerate.
Define r by (1.34), and decompose JR2n = Vo EB Vlo such that r vanishes on Vo and is
nondegenerate on V j (hence Vo and V j are even-dimensional). Then

Here we have written Vo in order to indicate that one has taken the Fourier transform in the
variables in Vo. This degenerate case is investigated in Kastler and Mebkhout [1990].
111.1.9 The map Q~ in 1.9.1 is due to Rieffel [1990a], who was the first to recognize
that the group C'-algebra C'(C) should be thought of as the quantization of the Lie-
Poisson algebra COO(g~). He also proved a version of Theorem 1.9.2 for exponential groups.
Theorems 1.9.2 and 1.9.5 are due to Landsman [1998d]. For Lemma 1.9.3 see Helgason
[1978], Ch. II.3, and Milnor [1976], §5. The general theory of (Riemannian and other)
connections on Lie groups may be found in Kobayashi and Nomizu [1963, 1969].
111.1.10 The Cartan-Weyl theory is discussed in a large number of textbooks; good
modem presentations are, e.g., Wallach [1973], Brocker and tom Dieck [1985], and Knapp
[1986]. The reformulation ofthe Cartan-Weyl theory in terms of coadjoint orbits is due to
Kostant [1970]. An explicit construction of U y from Oy is done through the Borel-Wei!
theory (see Wallach [1973], Knapp [1986], or Vogan [1987]), which in this application
coincides with the approach through geometric quantization; see the notes to II.I.I for
references, and in the present context in particular cf. Hurt [1983].
Equation (1.147) is closely related to Prop. 4.12 in Knapp [1986].
 Groups, bundles, and groupoids 461

In the context of 1.10.7 one may ask which properties single out the coadjoint orbit Oy
through J (1/Iy) among all coadjoint orbits contained in J (IP'?-{y)' This question was answered
by Kostant [1973] and Atiyah [1982] in terms of convexity properties. A brief summary
of the situation is as follows. One has a natural projection 'l' := 'l'0*-+1* given by restricting
a (J E g* to t; the object 'l' 0 J is then evidently the momentum map for the T -action on
lP'1ty given by restricting U y to T. A general theorem, due to Atiyah [1982] and Guillemin
and Sternberg [1982, 1984a], states that the image ofthe momentum map ofa torus action
on a compact connected symplectic manifold is a convex polytope (see Kirwan [1984] for
a generalization and Audin [1991] or Guillemin [1994] for reviews). Applying this to the
situation at hand, it turns out that Oy is singled out by the property that it contains the
extreme points of 'l' 0 J (lP'1t y ); these extreme points are precisely the imgages of the highest
weight state 1/Iy and its transforms under the Weyl group.
An overview of the role of coadjoint orbits in the representation theory of noncompact
Lie groups is given in Guichardet [1985], Vogan [1987, 1992], and Kirillov [1990]; see the
notes to 11.2.1 for analogous references relevant to the nilpotent case.
111.1.11 Coherent states of the type studied in this section were introduced by Klauder
[1963], and were rediscovered by Perelomov [1972], who added the perspective of the
Cartan-Weyl theory. See Perelomov [1986] and Klauder and Skagerstam [1985] for more
references, reprints of the original papers, and a general overview. A wealth of rigorous
information is contained in Simon [1980]. The connection with the Borel-Weil theory is
explained in Onofri [1975].
The fundamental idea of rescaling the label of an irreducible representation by multiply-
ing with lin (which accordingly has to be quantized in the compact case) is due to Berezin
[1975a,b]. He in addition investigated certain noncompact Lie groups; also cf. Perelomov
[1986]. In a more intuitive setting, this rescaling was explicit in the early years of quantum
mechanics, and seems to comprise one of the faces of Bohr's correspondence principle; see
Mehra and Rechenberg [1982].
One would expect that Qg(X) equals iIidUy/n(X), but after an arduous calculation (due
to Simon [1980]) one actually obtains

B - i
QI/k(X) = k + c(y) dUky(X),
where c(y) := 2(y, 8)/(y, y); recall that {) was defined after (1.164). While this result is
expected to be true in general, Simon [1980] acknowledges that his proof is limited to the
case that y is a multiple of a fundamental weight.
Lemma 1.11.2 is from Gilmore [1979] (whose proof, as remarked in Simon [1980],
is unnecessarily complicated). A detailed proof of (1.155) is in Duffield [1990], Prop. 4.
Related results are in Berezin [1972], Lieb [1973], Simon [1980], and Hogreveet al. [1983].
For the Weyl dimension formula see, e.g., Wallach [1973], Brocker and tom Dieck [1985],
or Knapp [1986].
Theorem 1.11.4 is due to Landsman [1998c]. An entirely different proof of Dirac's
condition, valid for arbitrary compact Klihlermanifolds, is given by Bordemann et a!. [1994].
Another relevant paper, which stresses the Kahler geometry behind Berezin quantization
on coadjoint orbits of compact Lie groups, is Barmoshe and Marinov [1994].
The steepest descent method used in the proof of 1.11.4 is in Hormander [1983].
111.2.1 A standard introductory reference for bundles and connections is Kobayashi and
Nomizu [1963, 1969]; for a full meal see Greub et al. [1972, 1973]. Choquet-Bruhat et a!.
[1982] and de Azcarraga and Izquierdo [1995] are presentations directed at physicists.
462 Notes

Much of the theory of bundles (with the evident exception of the theory of connections)
applies to general topological spaces and groups.
In the construction of principal bundles one may start with Q, H, a cover of Q by suitable
open sets N", and a collection of transition functions satisfying the relation hap ( q )h py (q) =
h"y(q) whenever q EN" n N p n Ny. The bundle P(Q, H, r) can then be reconstructed
from these data.
111.2.2 The insight that the physicists' gauge fields are the mathematicians' connections
on a bundle goes back to Hermann [1975], Wu and Yang [1975], and Konopleva and Popov
[1983] (relevant parts of which apparently date back to the sixties). The entire theory of
elementary particle interactions is currently based on gauge fields; see Weinberg [1995,
1996].
111.2.3 The history of cotangent bundle reduction is described in Marsden [1993];
applications of this construction are surveyed in Marsden .1l992].
A more common way of obtaining the reduced space (T*P)" is via Marsden-Weinstein
reduction; see IV. 1.5.
Lemma 2.3.1 is "classical"; see, e.g., Cor. 4.2.11 in Abraham and Marsden [1985]. Con-
nections are discussed from the cotangent bundle point of view in Guillemin and Sternberg
[1984b].
Theorem 2.3.7, straightforward as it is in its final formulation, is the culmination of a de-
velopment involving the work of Smale [1970] (who did the abelian case), Sternberg [1977],
and Weinstein [1978]. The Poisson bracket (2.55) was first computed by Montgomery et al.
[1984]. The symplectic form on P~ XH 0 is discussed in Guillemin and Sternberg [1984bj.
Writing 0 = H / Hj.L' it is possible to embed Cr.. .
P)" as a symplectic submanifold of
T*(P / Hj.L)' equipped with a modified symplectic structure; see Abraham and Marsden
[1985], Thm. 4.3.3, Kummer [1981], and Marsden [1992].
111.2.4 The exact sequences (2.60) and (2.64), along with the pertinent interpretation
of connections, are due to Atiyah [1957]; cf. Mackenzie [1987a], App. A, for a detailed
discussion. Propositions 2.4.2 and 2.4.3 are from Atiyah and Bott [1983]. In identifying
aut(P) as the Lie algebra of Aut(P) it is worth mentioning that an H -invariant vector field
on P is complete iff its projection to Q is complete; see Kumpera and Spencer [1972],
§33. Since the latter condition is satisfied as a consequence of the compact support on Q,
elements of aut(P) are automatically complete on P.
111.2.5 The construction of observables on P~ x H 0 through the momentum map of
the 9~-action on T*P is taken from Landsman [1993b]; also see Meinrenken [1994] and
Robson [1994, 1996]. This generalizes the approach ofIsham [1983] and Guillemin and
Sternberg [1984b] to nontrivial structure groups; cf. the notes to 11.3.1.
The second term on the right-hand side of (2.89) can be understood from a Lagrangian
point of view as the contribution to the Noether conserved charge due to the gauge field. In
general, if the Lie derivative L~pA vanishes for some vector field ~ P on P, it does not follow
that L~Qs' A is zero, too. This leads to the above-mentioned contribution, see Jackiw and
Manton [1980].
Cotangent reduction for more general symplectic structures on T*P than the canonical
one is discussed in Alekseevsy et al. [1994].
111.2.6 The Wong equations were first proposed by Wong [1970] on the basis of a heuristic
study of the classical limit of the equation of a scalar quantum field coupled to a Yang-
Mills field. The symplectic formulation is due to Sternberg [1977] and Weinstein [1978];
Montgomery [1984] related this to the construction of these equations due to Kerner [1968],
and thereby proved Theorem 2.6.2.
 Groups, bundles, and groupoids 463

The Wong equations are generalized to spinning particles in Kiinzle [ 1972] (which covers
the abelian case), Arodz [1988], and Hamad and Pare [1991]; also cf. Linden et al. [1996].
Other aspects of the Wong equations are treated in, for example, Duval and Horvathy [1982],
Balachandran et al. [1983, 1984], Chiang et al. [1985], Feher [1986], and Chrusci6ski and
Kijowski [1996].
Proposition 2.6.1 is known in physics as the Kaluza-Klein construction. It is, in fact
valid in the more general situation that gil is merely right-invariant (in which case (2.100)
is, of course, not valid). An amazing aspect of this construction is that the Einstein equations
for g are equivalent to the coupled Einstein-Yang-Mills equations for gQ and A; this follows
from (2.188), interpreting Q as space-time rather than space. This has led to the physical idea
that all Yang-Mills fields as well the gravitational field on four-dimensional spacetime are
shadows of the gravitational field in a higher-dimensional world, some of whose dimensions
are compact, and so small as to be invisible. (The original version of Kaluza and Klein
described the electromagnetic field in this way, assuming that the universe is 5-dimensional.)
This idea is also fundamental to string theory, but there is no evidence that it is correct
other than as a mathematical artifact. With certain restrictions, the construction is valid in
the pseudo-Riemannian case as well; see Choquet-Bruhat and DeWitt-Morette [1989] for
a precise statement. For an overview of "Kaluza-Klein physics" see the reprint volume
Appelquist et al. [1985]; mathematical aspects are discussed in Coquereaux and Jadczyk
[1988].
The correspondence between (2.112) and (2.111) is a special case of the passage between
Lie-Poisson equations on 0* and second-order equations on TG; see Marsden and Ratiu
[1994].
111.2.7 For the H -connection see Kobayashi and Nomizu [1963J, §II.11. The more gen-
eral theory of connections invariant under some group action is in Kobayashi and Nomizu
[1969J, applications to physics being discussed in Forgacs and Manton [1980], Jackiw and
Manton [1980], Hamad et al. [1980J, and Cant [1981].
The main reason for the popularity of the H -connection among physicists is that it solves
the Yang-Mills equations; see, e.g., Laquer [1984]. In particular, many famous "topological"
Yang-Mills configurations, such as instantons and monopoles, are special cases of the
H -connection. See Bais and Batenburg [1985] and the notes to 2.12.
A different type of application of the H -connection is to the theory of the Berry phase;
see Vinet [1988] and Giavarini and Onofri [1990]. For quantization theory on homogeneous
spaces see the notes to IV.2.8.
11I.2.S This material is mainly adapted from Landsman [1993b] (which generalizes the
treatment of the homogeneous case in Landsman [1990a, 1992]), but 2.8.2 is an elementary
special case of Thm. 3.1 in Muhly et al. [1987], with a different proof. The special case
P = G is Thm. 2.1 in Green [1980]. Corollary 2.8.3 is discussed in IV.2.7.
For the existence of the measure v in (2.139) cf. Bourbaki [1963J, Prop. VII.2.3 and 4.
Equation (2.154) is Lemma 5.3 in Guillemin and Uribe [1986].
When 1t x is separable, an equivalent construction of 1t x starts from the vector space
ilx of all Jl-measurable functions ~x : P ~ 1tx that satisfy the properties that (2.145)
holds, the function x ~ (~x (x), ~ x (x)x) is locally integrable, and (~x, ~ X) < 00. The
Hilbert space 1t x is then the quotient of ilx by the vectors of zero norm. This is proved in
Moscovici [1969]; to apply his proof, note that the action of a structure group on a principal
bundle is always proper.
111.2.9 An even more general construction of induced representations, which applies
when P and H are merely locally compact, and the H -action on P is not necessarily free,
464 Notes

is presented in Moscovici [1969] (though no bundle-theoretic interpretation is given). One

then tries to define an induced representation of some group G that acts on P; the G- and
H -actions must commute. For the construction (which is entirely analogous to the one in
our main text) to apply, the H -action must be proper, and Q = P/ H must possess a positive
measure v that is quasi-invariant under the natural G-action.
A detailed description of Mackey induction for locally compact groups may be found
in Warner [1972], Gaal [1973], Barut and Rar;:ka [1977], Varadarajan [1985], or Knapp
[1986]. The formulation in terms of vector bundles appeared in Hermann [1966], following
special cases in Bott [1957]. A discussion of harmonic analysis on vector bundles of the type
G XH 1ix may be found in Wallach [1973]. Attractive general overviews with applications
and history, by the founder of the modem theory of induced group representations, are
Mackey [1968, 1978, 1992]. Also cf. the notes to IV.2.8.
In the special case A = A H , equation (2.180) is due to Doebner and Tolar [1990]. A
different perspective on the quantum analogue of the classical momentum map for the
9~-action on T*p o is offered by Meinrenken [1994].
111.2.10 For the Laplace-Bochner operator cf. Wallach [1973] (who simply calls it the
Laplacian) and Kuwabara [1982]. Equation (2.187) is due to Berline and Vergne [1985] and
Guillemin and Uribe [1986]. For (2.188) see the notes to 2.6. An intrinsic definition of F2
is presented in Atiyah and Bott [1983].
Lemma 2.10.3 is a special case ofThm. 3 in Nussbaum [1964].
An explicit expression for the Ricci scalar RG on a Lie group G may be found, e.g., in
Coquereaux and Jadczyk [1988], §2.5.
The identity (2.193) appears, e.g., in Strathdee [1983], Siebarski [1987], and in the present
context of quantization theory also in Landsman [1992].
111.2.11 Theorem 2.11.1 is a reformulation of Theorem 6.1 of Hogreve et al. [1983].
Their derivation of their (6.33) provides a detailed proof of the claim preceding our (2.212),
and their proof of their (6.35), in particular their Lemma 4.7, also proves our (2.214). The
theorem and the given proof still apply if one includes a scalar potential V(q) and a term
Ao(q i Zi in the Hamiltonian (as is done in the above reference). Our proof relates to that of
Hogreve et al. [1983] in much the same way that our proof of Theorem 11.2.7.2 stands to the
proof of the corresponding result in Hepp [1974]; cf. the notes to 11.2.7. Moschella [1989]
claims to simplify the proof of the convergence of the quantum equations of motion to the
classical Wong equations, but in fact his argument proves convergence in a certain class of
mixed states, and with respect to different observables from the ones we are interested in.
Accordingly, he does not obtain a quantization condition on n.
The right-hand side of (2.209) may be written as T exp ( -i J~ H(2)(s)/n), where T
stands for "time-ordering"; see Dollard and Friedman [1979], and cf. Iy'(3.73).
There is an impressive body of mathematical literature on the semiclassical asymptotics
of the eigenfunctions and eigenvalues of the Laplace-Bochner operator, and their connec-
tion with Hamiltonian trajectories given by solutions of the classical Wong equations. For
example, Schrader and Taylor [1984, 1989] and Zelditch [1992] extend the work on clas-
sical limits of energy eigenstates cited in the notes n.2.7 to the case at hand. Eigenvalue
asymptotics are investigated in Guillemin and Uribe [1985, 1986, 1989, 1990], Taylor and
Uribe [1992], Brummelhuis and Uribe [1992], and Brummelhuis et al. [1995]. Other aspects
of the motion of a quantum particle in a Yang-Mills field are discussed in Arodz [1983],
Belov and Maslov [1990], and Oh [1996].
111.2.12 The quantum theory of magnetic monopoles started with the paper by Dirac
1
[1931], who arrived at the quantization condition eg = 1nn. The factor corresponds to
 Groups, bundles, and groupo ids 465

the use of the covering groups SU(2) and U(l) of SO(3) and SO(2). This leads to the Hopf
fibration SU(2)(S2, U(1), T); see, e.g., any of the books cited in the notes to 2.1.
There exists a gigantic body of literature on monopoles. The modem understanding of
the quantum case in terms of the line bundles Hn is due to Greub and Petry [1975] and
Wu and Yang [1975, 1976]. The description in terms of induced representations is due to
Langlands [1987]. Much information about both standard angular momentum and the role
of SO(3) in the theory of magnetic monopoles may be found in Biedenharn and Louck
[1981a,b]. The fact that rotational symmetry forces the field to be a monopole configuration
may be found in, e.g., Cant [1981] and Horvathy [1981].
One should actually start with the theory on R3, on which the monopole field potential
is given by A(r,,p, () = A(,p, ()/r. The magnetic field is B = -gT3 ® er /r 3, which
is evidently singular at the origin O. Hence one declares that the configuration space of
a charged particle moving in a monopole field is R 3\{0} :::: S2 x R+. The theory on S2
contains all essential features of the situation on R 3 \{0}.
Our treatment mainly follows Landsman [1990b], which includes a detailed discussion of
the passage from S2 to R 3 \ {O}. The proof of Proposition 2.12.1 is taken from Landsman and
Linden [1991]; the ancillary result (2.224) may be found in Choquet-Bruhat et al. [1982],
Problem III.5(8), in which we corrected a sign error.
The eigenfunctions of the operator d vn( C 2(SO(3))), which is usually taken as the quan-
tum Hamiltonian, are so-called monopole harmonics; see Wu and Yang [1976], Biedenham
and Louck [1981b], Kuwabara [1982, 1984], and Landsman [1990b]. In the realization of
the theory on the space 1{n of S o (2)-equivariant functions on S 0(3), a monopole harmonic
is simply a matrix element Uj (x)':,., where Vj (S 0(3» is the usual irreducible representation
of spin j, and m runs from - j to j. It may be checked that the corresponding functions in
1{± are indeed in r(Hn).
As in the general case, the justification for using a Hilbert space of sections of a line
bundle in quantum theory is that the smooth sections provide a domain of essential self-
adjointness of the relevant operators (angular momentum and Hamiltonian). The topology
of the line bundle enters the quantum-mechanical description in this way; the total Hilbert
space L2(H") is not sensitive to this topology.

ID.3.1 For an overview of the theory of groupoids cf. Brown [1987] and Weinstein
[1996a], as well as Renault [1980] and Mackenzie [1987a]. The shortest definition is that a
groupoid is a small category with inverses. In the context of group representation theory a
groupoid is sometimes called a virtual group; see Ramsay [1971] for an interesting account.
Most of the theory that is not purely algebraic is done in the context of topological or
measurable groupo ids.
Lie groupoids were introduced by Ehresmann [1958]. The main modem sources are
Mackenzie [1987a], Coste et al. [1987], and Albert and Dazord [1988]. Mackenzie refers
to differentiable groupoids, reserving the name Lie groupo ids for transitive differentiable
groupo ids. Weinstein [1996b] remarks that one can omit the smoothness of the inclusion
from Definition 3.1.5. The theory of gauge groupoids is developed in Mackenzie [1987a,
1989].

DI.3.2 For half-densities etc. see Guillemin and Sternberg [1977]; our treatment is some-
what different. For the Hilbert space of half-densities, due to Mackey, see Abraham and
Marsden [1985]. The bundle v'fATs®'G is mentioned in Weinstein [1991] and Connes
[1994], following a special case in Connes [1980] (involving the holonomy groupoid of
a foliation).
466 Notes

111.3.3 Convolution of sections of v'JAjs®t G was introduced by Connes [1980, 1994].

He symbolically writes convolution on a Lie groupoid as

With an isomorphism similar to A in (3.17) added, the equivalence between Connes's expres-
sion and (3.17) is easily established, using Prop. 1.1.2 in Mackenzie [1987a]. An equivalent
construction of the convolution algebra of a Lie groupoid appears in Bigonnet [1988],
whereas a third approach is in unpublished lecture notes by Renault (see Ramazan [1998]).
For left Haar systems and the corresponding convolution see Hahn [1978b] (who ac-
knowledges Westman [1968]) and Renault [1980], who work in the context of topological
groupoids. As in lhe group case, one can introduce a twist (cocycIe) into lhe convolution;
see Renault [1980]. The first part of Proposition 3.3.3 is due to Ramazan [1998].
111.3.4 Crossed products were introduced by Doplicher et al. [1966]. A crossed product
is usually defined as the C*-completion of C*(G, 21), denoted by C*(G. 21); see Pedersen
[1979] for lhe basic theory, and Green [1978] and Packer [1996] for advanced results. It is
not necessary to assume that G is unimodular; the standard definition of a crossed product
C* -algebra contains a factor ~(x )-1 on the right-hand side of (3.32), but our definition leads
to an isomorphic algebra.
A complete proof of Theorem 3.4.4 may be found in Busby and Smith [1970], Thm. 3.3;
a slightly different approach is in Pedersen [1979], Prop. 7.6.4. All these authors work in
the L 1 rather than C':' context, allowing general locally compact groups G.
Action C* -algebras are usually called transformation group algebras; see the notes to
3.7.
Systems of imprimitivity (for general locally compact groups) were introduced by
Mackey in his study of induced group representations; see Barut and Rat;ka [1977] and
Mackey [I 968](forthe case Q = G/ H), Varadarajan [1985], Mackey [1978], and Mackey
[1992]. The definition in lhese books is stated in terms of projection-valued measures on Q;
see IV.(2.118) and surrounding text. The equivalent approach we use goes back to Glimm
[1962] (who, in the locally compact setting, of course worked with Co(Q) rather than
C':'(Q».
111.3.5 Representations of locally compact groupoids in the sense of Definition 3.5.1 are
studied in Hahn [1978b] and Renault [1980]. For lhe more general case of an action of a
groupoid on some space see 3.9.11 or Mackenzie [1987a]. Quasi-invariant measures are
studied in Hahn [1978a]; also cf. Renault [1980]. The regular representation is developed
in Hahn [1978b].
Our definition of a direct integral Hilbert space (originally due to von Neumann) follows
Bratteli and Robinson [1987], §4.4.1. Also cf. Takesaki [1979] or Kadison and Ringrose
[1986], among others.
111.3.6 The C* -algebra of a locally compact groupoid first appeared in Connes [1980]
and Renault [1980]. The structure of such algebras is beginning to be analyzed, cf. Muhly
et al. [1987] and the series of papers by Muhly and Williams [1990,1992,1995].
Renault [1980], 11.1.22, shows that every representation of C':' (G) on a separable Hilbert
space that is continuous with respect to the inductive limit topology on C':'(G) and the weak
topology on s.B(H) is automatically bounded in lhe sense of (3.57); his proof is for locally
compact groupo ids, involving an additional condition that is automatically satisfied for Lie
groupoids. A detailed proof of (3.59) is in Hahn [1978a], Thm. 3.8.
 Groups, bundles, and groupoids 467

The groupoid analogue of the Banach algebra L 1(G) is L I (G), defined as the space of
measurable functions on G for which q t-+ 1,'rS.1-l(q )dJ1!q·t(Y)lf(Y)1 is essentially bounded
with respect toa locally Lebesgue measure on Q. It is a Banach algebra under the norm 11·111
defined in (3.57) and the continuous extension of multiplication and involution in C~(G).
It follows from Proposition 3.6.1 that every nondegenerate representation of C~(G) on a
separable Hilbert space that is bounded (in the sense of (3.57» extends to L I (G). See Hahn
[1978b] for the proof of these claims. The appropriate generalization of Theorem 1.7.3
holds here, in that each such representation corresponds to a representation U of G as in
3.6. I. This result is due to Renault [1987] (following a special case in Renault [1980]).
The first two points of Theorem 3.6.2 are in Hahn [1978b] and Renault [1980]. The
fourth point was inspired by Connes [1994], which contains a version of (3.65).
Parallel to the group case there is a concept of amenability of (locally compact) groupoids,
which is expressed by the equality C;(G) = C*(G), either as a theorem or as a definition;
see Renault [1980].

111.3.7 Theorem 3.7.1 is a special case of Thm. 3.1 in Muhly et al. [1987], with a
different proof. Equation (3.76) is proved in Renault [1980] via the correspondence between
representations of C*(G) and those of G, as mentioned above. Our proof is based on Rieffel
[1972]. For P = G, equation (3.78) is a special case ofThm. 2.13 in Green [1980] (who
states the result for the transformation group C*-algebra C*(G, G / H».
Starting with Effros and Hahn [1967], transformation group C'-algebras (which we
call action C* -algebras) are much studied by C' -algebraists; see the review by Packer [1994]
for history and references. Corollary 3.7.4 is a special case ofThm. 7.6.6 in Pedersen [1979].
Corollary 3.7.6 is Mackey's (transitive) imprimitivity theorem; see the books listed in
the notes to 3.4. Like Corollary 3.7.2, it is a special case ofRieffel's imprimitivity theorem,
and will be rederived as such in IV.2.7.
The special case C*(G, G/ H) of the CO-algebra C'(G, Q) was introduced by Glimm
[1962]. See the notes to IV.2.8 for applications to physics.
The proof of the Stone-von Neumann uniqueness theorem at the end of the section is
due to Mackey [1963, 1968, 1978]. The technical details of Mackey's proof are slightly
different, because he uses a different (equivalent) notion of systems of imprimitivity. In any
case, the argument can be generalized to arbitrary locally compact abelian groups.

111.3.8 Lie algebroids were introduced by Pradines [1966], who also showed how they
could be constructed from Lie groupoids. References and further development of the theory
until 1987 may be found in Mackenzie [1987]; a more recent reference is Vaisman [1994].
Pradines's "grand scheme to generalise the standard construction of a simply connected
Lie group from a Lie algebra to a corresponding construction of a Lie groupoid from a Lie
algebroid" was completed by Mackenzie [1987] in the locally trivial case, and by Brown
and Mucuk [1995, 1996] in general. Unlike the case of Lie algebras, there is a potential
cohomological obstruction, so that not every Lie algebroid corresponds to a Lie groupoid.
Apart from principal fiber bundles, the main context is the theory of foliations.
Equation (3.93) appears in Mackenzie [1987b], who attributes it to A. Weinstein (our
derivation in the proof of Proposition 3.8.8 is different).

111.3.9 Propositions 3.9.1 and 3.9.2 are due to Courant [1990]. A different approach to
the correspondence between Lie algebroids and linear Poisson structures, including a more
intrinsic definition of the Poisson structure on ~', may be found in Coste et al. [1987].
Definition 3.9.5, which includes the Poisson bracket (3.101) on g x Q as a special case,
is due to Krishnaprasad and Marsden [1987], and was further studied by Weinstein [1987].
468 Notes

Theorem 3.9.6 is due to Xu [1992], as is Corollary 3.9.8 (which is a special case of his
Cor. 4.1). We now sketch the remaining part of the proof of the latter; this depends on the
concept of a symplectic groupoid discussed in the notes to IV. 1.2. We have to show that
the completeness of the map p : S ~ Q associated to it : C""(Q) ~ C""(S) by p* = it
implies that J = (1(1), p) : S ~ g~ x Q is complete.
Thinking of Q as a Poisson manifold with zero Poisson structure, it is integrable, with
symplectic groupoid T* Q. The existence of an equivariant momentum map for the pullback
G-action on T*Q (see Lemma 2.3.1) implies that one may equip G := T*G x T*Q with
the structure of a symplectic groupoid with base g~ x Q; when the G-action on Q is
trivial this would be the direct product with respect to the groupoid structures mentioned in
the notes to IV. 1.2. Applying Thm. 3.1 in Xu [1991 b) in the direction "complete Poisson
map ~ symplectic groupoid action" to p, there exists a symplectic T* Q-action on S. The
covariance condition (3.107) implies that p intertwines the G-actions on Sand Q. By Thm.
4.1 in Xu [1992] this in tum entails that there exists a symplectic G-action on S associated
to J. Applying Thm. 3.1 above in the opposite direction then leads to the desired conclusion
that J is complete.
Corollary 3.9.10 is a special case of the classical transitive imprimitivity theorem IV. 1.6.4.
Definition 3.9.11, originally due to Pradines, may be found in Mackenzie [1987]. How-
ever, Mackenzie's notion of a Lie algebroid action (which he calls a representation) is
different from ours. A representation in his sense is a morphism of r(V) into the Lie al-
gebra of derivations on sections of some vector bundle. This concept is not appropriate in
relationship with the representation theory of the Poisson algebra C""(V*, R).
A different line of research relating Lie groupoids to Poisson structures is discussed in
the notes to IV. 1.2. Here the generalized momentum map associated to an action of a Lie
groupoid G on a symplectic manifold takes values in the base Q rather than in the dual of
the Lie algebroid~, as in our approach.
111.3.10 Lemma 3.10.1 is Prop. I1I.3.3 in Mackenzie [1987]. The map ExpL was
introduced by Pradines [1968].
111.3.11 The role of Lie algebroids and groupoids in strict quantization as explained in
this section originates with Landsman [1993b, 1996a).
In the context offormal deformation quantization, Dirac's condition has been proved for
arbitrary Lie groupoids by Nistor et al. [1997], and by Ramazan [1998].
Theorem 3.11.3 was first mentioned by Rieffel [1989a], though not in the context of Lie
algebroids and groupoids or Weyl quantization. This paper gives many interesting examples
of strict deformation quantizations that do not fit into our scheme, such as the quantization of
the Poisson algebra of functions on the 2-torus, equipped with a suitable Poisson structure,
by the so-called noncommutative torus. As these examples show, a given C* -algebra may
be a strict deformation of more than one Poisson algebra.
An "unbounded" version of this deformation, generalizing the procedure in 1.6 from Lie
algebras to Lie algebroids, may be found in Nistor et al. [1998].
111.3.12 The normal groupoid is constructed in Hilsum and Skandalis [1987], §3.1, and
is further discussed in Weinstein [1989]. These authors use I = [0, 1] instead of I = R, and
construct the manifold structure in a slightly different way. The special case of the tangent
groupoid (again, for I = [0, I)) is due to Connes [1994] (circulating in the eighties).
Lemma 3.12.4 is due to Lee [1976]. As explained in Elliott et al. [1993], it can be used
to simplify the proof of Rieffel's condition in certain examples in Rieffel [1989b]. For the
primitive spectrum and the Jacobson topology cf. Dixmier [1977], who gives our Lemma
3.12.5 as §§3.2.1,2.
 Reduction and induction 469

The realization of C;(G N ) as the C' -algebra of a continuous field has the following
generalization (G. Skandalis, private communication, June 1997). Let G be a Lie groupoid
with base Q, and let p be a continuous and open map from Q to some Hausdorff space X
which is G-invariant in the sense that p 0 rs = port. Define Gx := (p 0 rs)-l(x) (this is
a subgroupoid of G because of the G-invariance of p), and 2(x := C*(G x ). Then the triple
(C;(G), (C;(G x ), IPx}xEX), where IPAf):= f f Gx , is a continuous field of CO-algebras at
those points x where C*(Gx ) = C;(Gx )'
We apply this to our situation by taking G = GN and X = JR, whence Q = JR x Q,
and p is just projection onto the first variable. Continuity away from Ii = 0 follows from
the triviality of the field for Ii =1= 0 (whether or not C;(G) = C*(G». The result above
may be used to prove continuity at Ii = 0 by noticing that C;(<5) = C*(<5); this follows
because both sides are isomorphic to Co(<5*). In other words, from this point of view it is
the amenability of <5, regarded as a Lie groupoid as explained in 3.12.1, that lies behind
Theorem 3.11.4.
An interesting application of Corollary 3.12.6 would lie in the development of gener-
alizations of the Atiyah-Singer index theorem, noting that recent proofs of this theorem
through deformation quantization (see Connes [1994], Fedosov [1996], and Elliott et al.
[1996)) may be interpreted in the light of the special case of 3.12.6 in which GN is the
tangent groupoid of a manifold. The first step towards such generalizations, namely a good
definition of the analytical index, has already been taken in Monthubert and Pierrot (1997].
An entirely different application of the normal groupoid to the classical limit of quantum
mechanics is given Bellissard and Vittot [1990].

Chapter IV

IV.t.t The theory of constraints and reduction has a venerable tradition; the modem era
started with the work of Dirac [1950, 1964]. Efforts to put his approach on a geometric and
rigorous footing were initiated by the Warszaw school of ThJczyjew and collaborators in the
sixties and seventies; see, e.g., the books by Kijowski and ThJczyjew [1979] and Binz et al.
[1988], and references therein. Further contributions were made by Lichnerowicz [1977]
and Gotay et al. [1978] (also cf. Gotay and Nester [1980)). The approach of the latter starts
from a presymplectic manifold S, and is therefore more general than the one presented in
the main text. Also, the infinite-dimensional case is included.
In physics the constraints on the phase space S are usually derived from a Lagrangian; the
so-called constraint algorithm then leads to the final constraint hypersurface C, on which the
equations of motion defined by a Hamiltonian h are well-defined. Hence our C is supposed
to be the endproduct of this algorithm. Apart from the references above, see Sundermeyer
[1982] or Henneaux and Teitelboim [1992]; these books contain a wealth of information
and examples from physics.
Theorem 1.1.2 goes back to Cartan [1958]; a heuristic version is implicit in Dirac [1950,
1964]. The given formulation may be found, e.g., in Libermann and MarIe [1987], §14,
which contains a proof that the null distribution is smooth in App. 4, Prop. 3.7.
The concept of a weak observable is due to Dirac and Bergmann (see Sundermeyer
[1982] for an extensive list of references to the original literature ).
470 Notes

The decomposition mentioned in the last paragraph of the section is well known to
physicists; as a theorem it is proved, e.g., by Lichnerowicz [1977] (also cf. Thm.III.14.11
in Libermann and Marie [1987]).

IV.l.2 Theorem 1.2.2 is a reformulation of Prop. 2.1 in Xu [199Ia], which in tum gen-
eralizes Thm. 3.12 in Mikami and Weinstein [1988]. In these papers special symplectic
reduction is approached through the following theory, due to Karasev [1987], Weinstein
[1987b], and Zakrzewski [l990a,b]. We give only a brief summary (following Weinstein
[1991]); for more information we refer to Coste et a1. [1987], Albert and Dazord [1990],
Karasev and Maslov [1993], Vaisman [1994], and Weinstein [1998].
A symplectic groupoid (G, w) is a Lie groupoid G ~ Q with a symplectic form w, with
the property that the graph {(y, y', yy')} (where y, y' E G) of groupoid multiplication is
a Lagrangian submanifold of G x G x G-. This entails that the inversion J : G ~ G-
is a Poisson map. The most important consequence of the definition is that there exists a
Poisson structure on Q for which !(Q) is a Lagrangian submanifold of G, and Ts : G ~ Q
and Tt : G ~ Q- are Poisson maps. Moreover, the Poisson subalgebras T;cOO(Q) and
T;cOO(Q) of COO (G) commute; when the fibers of Ts (and hence of T/) are connected, these
subalgebras are even the Poisson commutants of one another.
The simplest example is G = T* Q (with its canonical symplectic form), where Ts =
T/ = TpQ--+Q' and groupoid "multiplication" is addition in a fiber of r* Q. The associated
Poisson structure on Q is the zero bracket. When Q is a Lie group G, still using the canonical
symplectic form, one can put a different groupoid structure on T* G by identifying T*G with
G x g* in the right trivialization (cf. 111.1.4), and regarding G x g* as an action groupoid
with respect to the coadjoint action (cf. I1I.3.1.4). Hence Ts = _J R and Tt = JL; this
assigns the (+) Lie-Poisson structure to g* .
One may ask whether a given Poisson manifold P is integrable, in that there exists a
symplectic groupoid whose base is P. As we just saw, any manifold P with the zero bracket
and any dual Lie algebra g* with the Lie-Poisson structure are integrable; not every Poisson
manifold is.
Recall Definition I1I.3.9.11 of a smooth action of a Lie groupoid G ~ Q on a manifold
S. For later use, define the orbit Ga of a E S under a G-action on S in the obvious way, i.e.,
Ga := {ya I (y, a) E G *Q S}. Mikami and Weinstein [1988] define such a Lie groupoid
action to be symplectic when the graph fey, a, ya)} of the groupoid action is a Lagrangian
submanifold of G x S x S- . It easily follows that J p : S ~ Q is a Poisson map. (Compare
this with Proposition I1I.3.9.13, where as an alternative proposal we associate a generalized
momentum map J : S ~ 18* to the G-action on S.) A deeper result is that Jp is necessarily
complete, and that conversely, when the Ts -fibers of G are connected, the completeness of
a given Poisson map J p : S ~ Q implies that there exists a symplectic G-action on S
associated with Jp ; see Thm. 3.1 in Xu [199Ib]. The main example of such a symplectic
groupoid action is derived from an ordinary group action with equivariant momentum map
J : S ~ g~ by putting J p = -J. The action of (0, X)R E T*G on a E S is then defined
when (1 = J(xa), and (J(xa), x)Ra = xa.
Let now P be an integrable Poisson manifold, so that there exists a symplectic groupoid
G ~ P. Xu's [199Ia] formulation of special symplectic reduction starts from a pair of
symplectic G-actions on Sand Sp. Using the above definition of an orbit, one can form
the quotient space (S *p Sp)/G under the diagonal G-action on S *p Sp. When the G-
orbits are connected, Prop. 2.1 in Xu [\99Ia] then shows that this quotient coincides with
the reduced space sj as defined in (1.13). When there are disconnected orbits one has
a situation similar to the one discussed around (1.26). As Mikami and Weinstein [1988]
 Reduction and induction 471

remark, this formulation of special symplectic reduction "turns out to be purely groupoid-
theoretic, involving no symplectic geometry at all". (They work in the special situation that
Sp is a symplectic leaf of P, with Jp the inclusion map.)
In the form given here, Definition 1.2.1 and Theorem 1.2.2 appear in Landsman [1995a].
Our formulation without symplectic groupo ids is motivated by the clean analogy with
the quantum situation; this analogy is obscure in the version cited above. For our use of
transverse intersections see, e.g., exercise 1.6E in Abraham and Marsden [1985], or §27 in
Guillemin and Sternberg [1984]. The dimension counting in the proof goes back to Kazhdan,
Kostant, and Sternberg [1978].
Theorem 1.2.2 can actually be generalized to the case where Sand Sp are Poisson
manifolds. As in the symplectic case, the reduced space SP is defined as the quotient of
S *p Sp by the foliation defined by the vector fields ~f' j E COO(P, R); the alternative
description in terms of the null foliation of We is, of course, not available here. Rather
than being symplectic, the manifold SP is a Poisson space, which carries a reduced Poisson
structure in the sense of Marsden and Ratiu [1986]. Exactly as in the proof of 1.2.2, one first
shows thatN eTC. Secondly, one has BU(.Nlj) C TxC. To show this local property, take
a = dg l +dg2 , withg; = r;'h;, where the 7:; are the natural projections rl : SxSp ~ Sand
7:2 : S x Sp ~ Sp, and hI E COO(S), h2 E COO(Sp)' Then the property that a E BU(Nlj) is
equivalent to the equality {J* j, hd = {pO j, h 2} for all j E COO(P). Hence J.~gl = P.~g2'
which proves the claim. The Poisson generalization of Theorem 1.2.2 now immediately
follows from the "Poisson Reduction Theorem" in Section 2 of Marsden and Ratiu [1986].

IV.l.3 Symplectically complete foliations were first defined by Libermann [1983], which
contains Proposition 1.3.2; also see Prop. III.9. 7 in Libermann and MarIe [1987]. See Dazord
and Delzant [1987] for further developments.
Proposition 1.3.3 is given by Weinstein [1983] for the case where S/cf) is smooth. Our
proof of the more general statement follows the proof of Thm. I in Karshon and Lerman
[1997].
When conditions I and 2 in 1.3.4 are satisfied, one speaks of a Weinstein dual pair; the
attribute "Weinstein" is sometimes omitted. When in addition 1.3.4.3 is met, one has a full
dual pair. The theory of such dual pairs is due (independently) to Karasev [1989] (whose
unpublished Russian original is from 1981) and Weinstein [1983] (the latter contains 1.3.2
as well). For a review also see Vaisman [1994] and Choquet-Bruhat and DeWitt-Morette
[1989]. Weinstein [1990] introduces the concept of a symplectic affinoid space, which is
a generalization of a full dual pair.
A symplectic groupoid G ~ Q (cf. the notes to 1.2) provides an interesting example of
a dual pair: The diagram in (1.20) then becomes Q ~ G ~ Q-.
Weinstein [1983] remarks that his dual pair is the classical analogue of a Howe dual
pair; this is a pair of reductive subgroups of a symplectic group Sp(2n, R), which are each
other's centralizer. Such pairs were introduced by Howe [1989] (which had been around
for a decade prior to publication). Howe dual pairs are studied from the perspective of
constrained quantization in Landsman [1994] and Bowes and Hannabuss [1997]. In the
author's opinion, the true quantum analogue of a classical dual pair is a quantum dual pair;
cf. Definition 2.3.1. Corollary 1.3.6 is due to Weinstein [1983], who acknowledges Kazhdan
et a!. [1978].
Definition 1.3.7 and Proposition 1.3.9 are due to Xu [199Ib] (which contains a great
deal of additional information on Morita equivalence in the present context, as does Xu
[1992]). The second example in 1.3.9 is attributed to Weinstein. The algebraic topology
needed to complete the proof is the exact sequence (see Bott and Tu [1982], equation
472 Notes

(17.4» 7l'1(Jj-I(O";» -+ 7l'1(S) -+ 7l'1(Sj) -+ 7l'O(Jj-I(O"j». By assumption, the first and the
last entry are the trivial group, so that the claim follows.
IV.I.4 Theorem 1.4.1 is due to Xu [199Ib]. Our proof, however, is taken from Landsman
[1995a]; the relevant homotopy theory may be found in Janich [1994]. The proof in Xu
[199Ib] is based on the theory of symplectic groupoids. Xu assumes that the Poisson
manifolds in question are integrable; combining Thms. 4.18 and 5.2 in Weinstein [1990],
one infers that this is always the case in the given situation. Xu's proof follows the lines of
first showing that integrable Morita-equivalent Poisson manifolds have Morita-equivalent
symplectic groupoids, which in tum have equivalent categories of complete symplectic
realizations.
IV.1.S Lemma 1.5.1 is due to Smale [1970]. For proper group actions and a proof of
Proposition 1.5.3 see Abraham and Marsden [1985], particularly Prop. 4.1.23, or Cushman
and Bates [1997], App. B.
Marsden-Weinstein reduction is due to Meyer [1973] and Marsden and Weinstein [1974];
it has a long pedigree in classical mechanics. Our formulation as a special case of special
symplectic reduction coincides with the construction of Marsden-Weinstein quotients in
Kazhdan et al. [1978]. There is a great deal more to say about this subject. For example,
the construction may be carried out under less stringent conditions than the surjectivity of
J. (Le., the regularity of 8). Extensive treatments may be found in Guillemin and Stern-
berg [1984], Abraham and Marsden [1985], and Libermann and Marie [1987]. Marsden
[1992] gives an overview of applications of Marsden-Weinstein reduction in mechanics.
A generalization to general Poisson manifolds is presented in Marsden and Ratiu [1986,
1994].
Theorem 1.5.5 is due to Marle; see Libermann and Marie [1987], prop. IV.6.8.1t is also
mentioned by Weinstein [1983].
Proposition 1.5.8 is implicit in, e.g., Weinstein [1983] and Xu [1992]. The completeness
of J is explicitly proved in the latter paper using a different method based on Thm. 3.1 in
Xu [199Ib]. The procedure to (re)construct dynamics on S given the dynamics on SI H
used in the proof of the completeness of r is due to Marsden and Weinstein [1974]; also
see Marsden et al. [1990] and Marsden [1992]. The existence of a complete solution of the
equation X(t)-I.i(t) = X(t) is proved, e.g., in Dollard and Friedman [1979], Ch. I.
When H does not act freely on S one may still ask whether 1.3.3 holds in the context
of Marsden-Weinstein reduction. For compact H this is analyzed by Karshon and Lerman
[1997].
IV.I.6 In the special case P = G the reduced space T' pP of Figure 3 appeared in
Kazhdan et al. [1978]; also see Guillemin and Sternberg [1984] and Zakrzewski [1986]. In
this context one usually speaks of symplectic induction.
Theorem 1.6.1 is due to Duval et al. [1992]. Many interesting generalizations of Theorem
1.6.2 may be found in Xu [1992]. Theorem 1.6.4 is due to Ziegler [ 1996].
IV.I.7 Our proof of 1.6.4 is based on Theorem 1.4.1 (which was not used by Ziegler).
The last part of the proof of Lemma 1.7.1, however, is based on Ziegler's [1996] proof of
Theorem 1.6.4.
IV.I.S Theorem 1.8.1 is due to Landsman [1995a]. It is motivated by Theorem 2.6.1 on
Rieffel induction in stages. Theorem 1.8.2 is a straightforward generalization of Corollary
1.8.4, which appeared in Landsman [1995a]. Many special cases were known; see, for
example, Marsden et al. [1984] and Guillemin et al. [1996]. Lemma 1.8.3 is Prop. A.4 in
Weinstein [1987a].
 Reduction and induction 473

Proposition 1.8.5 is taken from Sjamaar and Lerman [1991], who prove it for singular
reduced spaces as well. Extensive information on strongly Hamiltonian product actions may
be found in Libermann and MarIe [1987].
IV.I.9 All unproved statements in this section may be found in Leptin and Ludwig [1994]
or in Corwin and Greenleaf [1989]. The latter contains Lemma 1.9.2 as Thm. 1.3.3 or Prop.
3.1.18, whereas the "nontrivial fact" (due to Chevalley and Rosenlicht) used in the proof
of 1.9.1 is Thm. 3.1.4. Finally, Theorem 1.9.3, which combines the work of Dixmier and
Kirillov, is Thms. 2.2.2-4.
Theorem 1.9.1 is taken from Landsman [1995a]. Our use of the structure of the coadjoint
orbits of nilpotent Lie groups should be distinguished from the "orbit philosophy" of Kirillov
[1962, 1990], Souriau [1969], Kostant [1970], and others; oursole aim is the correspondence
between symplectic reduction and the theory of induced representations, which is seen to
be quite perfect in the case of connected, simply connected nilpotent Lie groups.
IV.I.IO Proposition 1.10.1 is due to Rawnsley [1975]. Theorem 1.10.2 appeared in Lands-
man [1995a] and in Guillemin etal. [1996]. Theorem 1.10.3 was first given byWigner [1939]
for the case that G is the Poincare group. Barut and Ra~ka [1977] and Varadarajan [1985]
are good sources for the theory of induced representations of semidirect products. Our proof
is a straightforward C' -algebraic reformulation of the proof given in these references.
Theorem 1.10.4 appeared in Marsden et al. [1984], which includes extensive references
to related results, as well as applications; also cf. Guillemin and Sternberg [1984]. Both
groups of authors use the equality in the opposite direction.
The quantization theory of Isham [1983] may be reconsidered in the light of 1.10.4.
Given a homogeneous configuration space Q = L / H, he first looks for a vector space
V with an L-action p, such that Q is diffeomorphic to some p*(L)-orbit in V*. He then
accepts any irreducible representation of G = L D< p V as a possible quantization of the
cotangent bundle T* Q. However, having found aft E V* for which H = Lp, the space T* Q
is symplectomorphic to f:L 0 = (Jr1p )-' (0)/ Lp, which by 1.10.4 is symplectomorphic to
the coadjoint orbit og,P)' Hence Isham's proposal amounts to accepting any irreducible
representation of G as a possible quantization of this particular coadjoint orbit. Cf. Robson
[1994, 1996] for a related discussion.
Corollary 1.10.5 is due to Leonard and Marsden [1997], who provide a very detailed
proof, as well as giving applications to the motion of underwater vehicles.
Baguis [1998] gives a detailed study of the symplectic geometry of the coadjoint orbits
of semidirect products.
IV.I.ll The mathematical theory of singular Marsden-Weinstein reduction started with
Arms et al. [1981], who proved that the singularities in J-'(O) are conical. Arms etal. [1990]
look at the singular case of general symplectic reduction, comparing various approaches,
and include a good bibliography. A very detailed treatment is given in Cushman and Bates
[1977], App. B.
Lemma 1.I1.1 is Lemma 27.1 in Guillemin and Sternberg [1984]. Proposition 1.11.2 is
taken from Cushman and Bates [1977], App. B.5.17 (our proof is a trifle different). Propo-
sition 1.11.3, which in our presentation is of fundamental importance, is due to Sjamaar and
Lerman [1991]. Definition 1.11.4 was first proposed by Arms et al. [1991], which contains
the first half of Proposition 1.11.5; the second half is due to Sjamaar and Lerman [1991].
Theorem 1.11.6 summarizes results of Arms et al. [1981], Otto [1987], Arms et al. [1991],
and Sjamaar and Lerman [1991]. Using more sophisticated techniques, the latter prove that
the decomposition (1.85) is locally finite and satisfies the condition of the frontier; that is,
the closure of each piece is the union of other pieces in the decomposition. Indeed, they show
474 Notes

that the decomposition in question is a stratification in the sense of Goreski and MacPherson
[1988]. Note that none of these properties is in general satisfied by the decomposition of a
Poisson manifold into its symplectic leaves, which otherwise is somewhat comparable with
the decomposition of a singular Marsden-Weinstein quotient into its symplectic pieces.
The S o (2)-example was given in Gotay and Bos [1986], who also generalize itto S O(n).
Lennan et al. [1993] further analyze this example (among many others), and refer to Schwarz
[1975] for the proof of the claim on smooth functions on So. Proposition 1.11.7 is due to
Sjamaar and Lennan [1991].
The second example is taken from Landsman [1998a], who also explains its relevance
to cosmology. Similar examples appear in the literature; see, e.g., Sniatycki and Weinstein
[1983] and Anns et al. [1990].
Singular Marsden-Weinstein reduction is of great importance to general relativity and
Yang-Mills theories, where field configurations with symmetry project to singular points of
the physical phase space (obtained by fonning a Marsden-Weinstein quotient with respect to
the gauge group; see IV.3). These applications are studied in Fischer et al. [1980], Isenberg
and Marsden [1981], Anns [1981, 1986], Anns et al. [1981], and Emmrich and Romer
[1990]. The two-dimensional case enables one to perfonn explicit calculations; see IV.3.6
and notes thereto.
Moduli spaces of flat connections on a compact Riemann surface provide closely related
examples; from the large body of literature on this topic, starting with Atiyah and Bott
[1983], we select Hitchin [1990], Weinstein [1995], Huebschmann [1996], and Jeffrey and
Weitsman [1997] (and references therein to earlier work of these authors). For the moduli
space of all Yang-Mills connections on a compact Riemann surface see the review by
Sengupta [1997], and references therein.
IV.2.1 The theory of Hilbert C" -modules over commutative C' -algebras was initiated by
Kaplansky [1953]. The generalization to the noncommutative case was studied by Paschke
[1973], containing all results in this section except Corollary 2.1.4. Simultaneously, Rieffel
[1974a] introduced pre-Hilbert CO-modules. A recent textbook is Lance [1995], which
contains 2.1.4.
For multiplier algebras in the present context see Wegge-Olsen [1993] or Lance [1995].
The advanced theory of Hilbert C" -modules, which we do not cover, is mainly due to Kas-
parov [1980,1981]. In his work, Hilbert C"-modules are a basic tool in the K-theory of
C" -algebras, which is a noncommutative generalization of the theory of vector bundles. See
Wegge-Olsen [1993] for a "friendly introduction" to this topic, and Connes [1994] for a
high-level treatise. Blackadar [1986] and Skandalis [1991] review Kasparov's [1981] gener-
alization of operator K -theory, known as K K -theory. Frank [1998] contains an exhaustive
bibliography on all aspects of Hilbert C* -modules. A detailed study of self-duality is in
Frank [1990].
IV.2.2 Rieffel induction is due to Rieffel [1974a], which contains historical comments. He
works entirely in the setting of pre-Hilbert C'-modules on which (2.4) does not necessarily
hold (which he refers to as pre-~-Hilbert spaces).
The construction revolving around (2.32) is taken from Hannabuss [1984]. Fell induction
is due to Fell [1978]; also cf. Fell and Doran [1988] (which is an encyclopedic treatment
of induction techniques in representation theory, including a vast bibliography). A related
induction procedure is given by Bennett [1978].
IV.2.3 Operators of the type (2.47) appear in Rieffel [1974a], and the C"-algebra
q(E, ~) is defined in Paschke [1973]. Theorem 2.3.3 is a "completion" of Prop. 6.18
in Rieffel [1974a]; an essential step in the proof, namely the equality 11"'1121 = II "'1I'lI, is
 Reduction and induction 475

equivalent to Prop. 3.1 in Rieffel [1979]. In the present fonn, Theorem 2.3.3 is a special
case of Prop. 7.1 in Lance [1995].
IV.2.4 For the history of the concept of Morita equivalence see Morita's obituary by
Arhangel'skii et al. [1997]; see, e.g., Bass [1968] for a textbook treatment. Morita's theorem
in pure algebra states that two rings 2l and !B have isomorphic categories of (left) modules
iff 2l is isomorphic to the endomorphism ring of a !B-module £, where £ and !B are each a
direct summand of some (possibly different) power of each other. In that case 2l and !B are
said to be Morita equivalent. An appropriate version of this concept is applied to C* -algebras
and W*-algebras in Rieffel [1974b], and to general Banach algebras in Gr~nbrek [1995].
With an appropriate definition of modules and category equivalence, Rieffel [197 4b] shows
that two von Neumann algebras 9Jt and IJ1 have isomorphic categories of (left) modules iff
9Jt = C'(£, 1J1) for some Hilbert C'-module £ ~ 1J1.
What we (following, e.g., Skandalis [1991] and [Lance [1995]) for simplicity call Morita
equivalence in the main text should more properly be called strong Morita equivalence,
which is indeed the tenninology used in most of the literature. The original definition of
this equivalence relation by Rieffel [197 4a] consisted in the conditions of Proposition 2.4.4,
on the basis of which he fonnulated and proved Theorem 2.4.5. Various generalizations of
Rieffel's imprimitivity theorem are studied in Fell [1978] and Fell and Doran [1988].
The correspondence between the representations of 2l and !B established in Theorem
2.4.5 can be shown to preserve weak containment, but not cyclicity.
An interesting result, due to Brown et al. [1977], is that two C* -algebras 2l, !B with
countable approximate identity (this is automatic when the algebras are separable) are
strongly Morita equivalent iff they are stably isomorphic; that is, when 2l ~ !B ® !BoO-£)
for separable 1t. Also cf. Lance [1995]. The notion of stable isomorphism appearing here
is a noncommutative generalization of the same concept for vector bundles.
IV.2.S The idea of looking at Hilbert CO-modules coming from a group representation
first appeared, in a different context, in Rieffel [1988]. It was rediscovered in Landsman
[1995a], which contains most results in this section.
It would be interesting to have a criterion on U or H guaranteeing that the dense subspace
t C 1t assumed in Theorem 2.5.4 exists. More generally, one could ask for conditions
guaranteeing the existence of an t such that the function defined by (2.77) lies in L I(H) for
all \II, ¢ Et. This question is well known when Ll is replaced by L2; see, e.g., Dixmier
[1977].
The fact about amenable groups used in the proof of Theorem 2.5.4 may be found in
§3.6 of Greenleaf [1969], or in §II.3 of Renault [1980] (where the existence of the Uj is
even given as the definition of amenability).
One may generalize the construction to the case where U is a representation of a Lie
groupoid as defined in III.3.5.1, with associated direct integral Hilbert space III.(3.53). The
generalization of (2.76) is

Equation (2.77) becomes

(\II, ¢}cgo(G) : Y ~ (\II't(Y)' U(Y)¢"(Y»'t(Y)'

For <I> = \II this is positive when r is amenable as defined by Renault [1980]; the relevant
part of the proof of Theorem 2.5.4 may simply be copied.
476 Notes

IV.2.6 Theorem 2.6.1 is due to Rieffel [1974a]; also cf. Fell and Doran [1988]. The
special case Corollary 2.6.4 is due to Mackey, and holds for locally compact groups. For the
original proof see the books cited in the notes to III.2.9. A somewhat different derivation of
Theorem 2.6.1 may be found in Rieffel [197 4a] and Fell and Doran [1988]. The intermediate
case Theorem 2.6.2 incorporates the generalization of 2.6.4 given by Moscovici [1969].

IV.2.7 A different derivation of Mackey's transitive imprimitivity theorem from Rieffel's

imprimitivity theorem may be found in Rieffel [1974a] and Fell and Doran [1988]. These
authors do not use the groupoid C* -algebra C*(P x H P) and then specialize to P = G,
but directly construct an equivalence bimodule between C*(G, G I H) and C*(H). A very
efficient proof of the transitive imprimitivity theorem is given by 0rsted [1979].
Theorem 2.7.3 is contained in Glimm [1961, 1962]. Our proof combines the easy part
of the proof of Thm. I in Glimm [1961] (namely the implications I ---+ 2 and 3 ---+ 4) with
§V1.5 of Varadarajan [1985] and some elements of the proof of Prop. 8.1 in Rieffel [1979].
A different approach to the proof is contained in the proof of Thm. 2.2 in Glimm [1962],
which could be somewhat simplified by using Lemma 1.1 in Effros and Hahn [1967].
The fact that Gqo is homeomorphic to GIG qO under the regularity assumption is con-
tained in Thm. I of Glimm [1961]. A simpler proof may be given by first noting that the
orbit is Hausdorff (which is a trivial consequence of the regularity assumption), and then
proceeding as in the proof of Prop. 7.1 in Rieffel [1979].
An interesting application of Theorem 2.7.3 to the theory of quantum groups is contained
in Koomwinder and Muller [1997].
When the regularity assumption on (G, Q) does not hold, the classification of the ir-
reducible representations of C*(G, Q) appears to be impossible, and the right object to
study is the primitive ideal space of C*( G, Q). This study was initiated by Effros and Hahn
[1967], whose main conjecture on the structure of this space was proved by Gootman and
Rosenberg [ 1979].
IV.2.8 Definition 2.8.1 is usually given in terms of covariant POV-measures; see all
books on POV-measures cited in the notes to 11.1.4 also for the covariant case. Theorem
2.8.2 is due to Poulsen [1970], and was rediscovered in physics by Neumann [1972], with
further contributions by Scutaru [1977], Cattaneo [1979], and Castrigiano and Henrichs
[1980]. Applying this theorem to the covariant Berezin quantization of the coadjoint orbits
of compact Lie groups studied in III.I.II, one is naturally led to the Borel-Wei! theory (cf.
the notes to 1II.1.1 0), in a way independent of geometric quantization.
The analysis of covariant localization on ]R3 was initiated by Newton and Wigner [1949],
and was reformulated in terms of systems of imprimitivity by Wightman [1962]. He used
the original definition of these systems, namely (2.118). The generalization to arbitrary
homogeneous spaces is due to Mackey [1968, 1978]; see Varadarajan [1985] for a detailed
technical account, and consult Mackey [1992, 1998] for historical comments.
The Mackey-Wightman approach was further developed in Doebner and Tolar [1975],
who suggested that the natural quantum Hamiltonian on 1t x is the middle term in III.(2.192);
cf. the classical expression 111.(2.135). Moreover, they remarked that one could state the
theory in terms of the action C* -algebra C*(G, G I H). This approach was further developed
by Majid [1988, 1990] and Landsman [1990a,b, 1992]. Other aspects of quantization theory
on homogeneous spaces are discussed in Emch [1982, 1983], Camporesi [1990], Landsman
and Linden [1991], Marinov [1995], Robson [1994,1996], and Wu [1998]. The last two
references confirm the picture sketched at the end of the section via geometric quantization.
Unlike a massive particle, a photon cannot be localized in the configuration space JR.3
in an E(3)-covariant way. It may, however, be covariantly localized in the generalized
 Reduction and induction 477

sense nonnally applied to phase space localization; see, for example, Ali and Emch [1974],
Kraus [1977], and Brooke and Schroeck [1996]. In the notation of (2.109) etc., the wave
function \II of a photon belongs to the subspace of elements \II E 'HI that satisfy the
transversality condition V . \II(x) = O. The projection p onto this subspace (given by
p\ll = \II - .6.L: I VV. \II, where.6. L is the Laplacian) commutes with U I (E(3», so that one
is in the setting of 2.8.2, with G = E(3), H = SO(3), u x = U l , and ir x = ir l as defined
in (2.117).
By Proposition 11.1.4.8, this leads to an E(3)-covariant POVM .6. t4 A(.6.) on ]R3 in
p'Hl. In line with Corollary 11.1.4.9, this POVM is given by A(.6.) = pE(.6.)p, where
E(.6.) = Xt;. ® ][1 (cf. the main text). The position operators Qk = JIll dA(x)Xk (cf.
11.(1.34) and the main text) do not commute with each other. The classical counterpart
of this phenomenon is mentioned after (3.8). Indeed, Duval and Elhadad [1992] show that
the geometric quantization of the canonical classical position variables precisely yields the
quantum position operators Qk just defined.
A different approach to photon localization, based on microlocal analysis, has been
initiated by Omnes [1997b].

IV.2.9 The problem of quantizing constrained systems has been faced since the earliest
days of quantum mechanics. A good historical overview of the treatment of gauge invariance
in quantum electrodynamics is given in Weinberg [1995], which with Weinberg [1996] also
contains an up-to-date treatment of heuristic techniques used by physicists to deal with
gauge invariance and constraints.
Books more specifically dealing with constrained quantization include Dirac [1964]
(which initiated the modern era), Sundenneyer [1982], Govaerts [1991], and Henneaux and
Teitelboim [1992]. The technique of BRST quantization developed in the last two books
(as well as in Weinberg [1996]) seems to perfonn well in quantum field theory and string
theory, especially in their path-integral version. Applied to finite-dimensional systems, the
operatorial BRST technique faces similar functional-analytic problems as the Dirac method;
cf. Landsman and Linden [1992] for simple examples. Nonetheless, the BRST method
remains the most highly developed and widely used method of constrained quantization to
date. See Duval et a1. [1991] for a "bosonic" refonnulation of BRST.
There is an extensive literature on the geometric quantization of constrained systems;
see, for example, Gotay [1986], Ashtekar and Stillennan [1986], Blau [1988], Thynman
[1990], Woodhouse [1992], and Robson [1994, 1996].
A CO-algebraic approach to constrained quantization that is closer in spirit to the Dirac
method than the technique described in the main text, has been developed by Grundling and
Hurst [1985, 1987, 1988a,b]. Applications to quantum field theory are given in Grundling
[1988]; similar techniques are used by Thirring and Narnhofer [1992], and Acerbi et a1.
[1993a,b]. This approach has the advantage of being able to handle second-class constraints
(which in our method have to be brought into first class fonn by refonnulating the classical
situation), but lacks the connection with symplectic reduction and Hilbert C* -modules.
Other mathematically sound attempts to rescue the Dirac method include Ashtekar and
Tate [1994] ,Ashtekar et a1. [1995], and Klauder [1997]. In Klauder's approach the projec-
tion Pid is replaced by approximate projections in the spirit of the p~ used in the proof of
Theorem 2.5.4.
The use of Rieffel induction in constrained quantization started with Landsman [1995a];
Definition 2.9.1 appeared in Landsman [1998a].
The idea of constructing an inner product by group averaging as in (2.81) (with 'Hx = C)
goes back at least to Nachtmann [1968]. In the context of constrained quantization see
478 Notes

Teitelboim [1982, 1984], Higuchi [1991], Halliwell and Hartle [1991], and Ashtekar et al.
[1995].
IV.2.lOThe analysis of the operator (2.127) is done with Weyl's method; see, for example,
Thm. X.7 in Reed and Simon [1975]. Our operator is in the limit circle case at 0 and in the
limit point case at 00. The essential self-adjointness on Vx C;"'(]R2) follows from Thm. 3 in
Nussbaum [1964], or from a direct argument. Some abstract theory behind this example is
developed in Wren [1997], who in addition discusses a profound generalization.
There is a great deal of literature on generalized eigenfunctions of the type fl and
the corresponding expansions; see Berezanskii [1968] for an old but still adequate, and
Poerschke et al. [1989] and Poerschke and Stolz [1993] for a more recent treatment. The
group generated by the Lie algebra defined by (2.137) is the two-dimensional Poincare
group.
It is instructive to replace the classical constraint rp = ~ (p? - pi) by rp± = rp± ~ exp( 4q I).
Interestingly, the Hamiltonian flow of rp_ on T*]R2 is incomplete, so that the constraint fails
to generate an action of R
The constraints are quantized on L2(lR2) by Q(rp±) = Q(rp) ± &exp(4q I), where the
last term is a multiplication operator. The incompleteness of rp_ is reflected in the quantum
theory, because Q(rp_) is not essentially self-adjoint on C;"'(lR2). What follows applies to
any self-adjoint extension.
The most fundamental difference between rp and rp± is that the spectrum of (the closure
of) Q(rp+) and of Q(rp_) is lR with mUltiplicity one, whereas the spectrum of Q(rp) is lR with
multiplicity two. Consequently, for fixed k only one of the two generalized eigenfunctions
of Q(rp±) plays a role in the construction of the induced space (as opposed to the pair fl
in the main text), which is naturally isomorphic to L2(lR, dk/2rclki). These eigenfunctions
may be deduced from the elementary theory of Bessel functions, but one needs more specific
Hilbert space techniques to decide which one occurs in the spectral decomposition of 'H.
These techniques may be found, for example, in Picard [1989].
The details of the quantum treatment of rp and rp± may be found in Landsman [1998a].
These constraints are motivated by quantum cosmology and the question what the "wave
function of the universe" should be; see Landsman [1995c] for this context. See Marolf
[1997] and references therein for an analogous treatment of the constraints of quantum
cosmology.
The above consideration on multiplicity is relevant to the constrained quantization
proposal of Hiijicek [1994], who suggests that all generalized solutions of the quantum
constraints should be used in the construction of the physical Hilbert space of pure quantum
states.
A different approach to the quantization of singular Marsden-Weinstein quotients is
presented by Emmrich and Romer [1990], who face the problem of having to decide
which self-adjoint extension of the reduced Hamiltonian to choose. See Sniatycki and
Weinstein [1983] for yet another approach (further discussed in Wren [1997]). Meinrenken
and Sjamaar [1998] look at the problem in the context of geometric quantization.
Despite the existence of a large number of papers on the quantization of the moduli space
of flat connections on a compact Riemann surface (cf. the notes to 1.11), the effect of the
singularities in this space on the quantum theory is not well understood (see Jeffrey and
Weitsman [1992] for a careful treatment of the singular points in geometric quantization).
IV.3.t Forthe Poincare group and its use in physics see Barut and Racka [1977], Varadara-
jan [1985], Woodhouse [1992], or Weinberg [1996]. The coadjoint orbits of the Poincare
group were first described by Souriau [1969, 1997] and Arens [ 1971 a,b]. Further discussions
 Reduction and induction 479

of these orbits may be found in Guillemin and Sternberg [1984], Carinena, Gracia-Bondia,
and Vanlly [1990], Woodhouse [1992], Duval and Elhadad [1992], and Schroeck [1996].
The noncommutativity of position coordinates of particles with nonzero spin or helicity is
discussed by Bacry [1988].
IV.3.2 For m > 0, Theorem 3.2.2 is suggested in §I.20 of Guillemin and Sternberg [1984].
The general case is taken from Landsman and Wiedemann [1994], where the details of the
proof may be found.
IV.3.3 For Proposition 3.3.1, originally due to Wigner [1939], see the first four references
in the notes to IV.3.1 above. It is usually thought that the irreducible representations of P
provide an adequate description of elementary particles in asymptotic states (and this was
Wigner's motivation as well), but this description fails even for electrons (because of the
photon cloud always surrounding them), and also, for different reasons, for quarks and
gluons. See Buchholz [1996] for a promising new approach.
The trick involved in (3.20) is due to Carey et al. [1977, 1978]. Proposition 3.3.3 comes
from Landsman [l995a]. For covariant representations see the notes to the next section.
One may wonder whether covariant Berezin quantizations of the coadjoint orbits of the
Poincare group exist; see the discussion following the proof of Theorem 2.8.2. It unfor-
tunately turns out that for the physical orbits the condition in Corollary 2.8.3 cannot be
satisfied; see Schroeck [1996] (who reaches this conclusion in a different way), and ref-
erences therein on this issue. For attempts to construct nonetheless a covariant relativistic
quantum mechanics of single particles, see Ali [1985], Ali et al. [1995], and Schroeck
[1996]. As explained in Ali [1998], this leads to a hyperplane-dependent notion of local-
ization in phase space. (For the analogous proposal of hyperplane-dependent localization
in Minkowski space, see Butterfield and Fleming [1998].) At least for the massive orbits,
covariant Weyl quantization turns out to be possible; see Carinena et al. [1990]. This is
reminiscent of the nonrelativistic theory, in which Weyl quantization has better covariance
properties than Berezin quantization, too; cf. Theorems 11.2.4.3 and 11.2.5.1.
IV.3.4 Covariant representations of P and Proposition 3.4.1 have a long tradition (going
back to Pauli and Wigner), culminating in the work of Weinberg [1995] (in which references
to his original work in the early sixties may be found). For a more mathematical treatment
of wave equations for massive fields see Barut and Ra~ka [1977] and Asorey et al. [1985].
The finite-dimensional representations of the Lorentz group are labeled by two positive
integers jl, jz. The decomposition of Rt,h ,h) under the restriction to SO(3) is given by the
well-known Clebsch-Gordan series, so that all integral spins between Ijl + jz I and Ijl - jz I
occur. In the restriction to E(2) the only helicity that occurs as a proper subrepresentation
is h - h; see Weinberg [1995]. This result misses representations on reduced spaces, and
therefore fails to explain the connection between masslessness and gauge invariance.1t does
explain why helicity 0 occurs in the proof of Proposition 3.4.2, since Rv = R(l/2,1/2)'
The second half of 3.4.2 is taken from Landsman and Wiedemann [1994], as is Theorem
3.4.4. This paper also treats helicity ±2, relating the masslessness of the graviton to the
infinitesimal diffeomorphism invariance of linearized gravity.
The idea of gauge invariance lies at the basis of modem high-energy physics. The con-
nection between masslessness and gauge invariance holds for most, but not all, interacting
theories. The most famous exception is massless quantum electrodynamics in d = 2, as
recognized by Schwinger [1962]; see Lowenstein and Swieca [1971] for the definitive
treatment of the Schwinger model.
1V.3.5 A standard reference for the CCR-algebra is Bratteli and Robinson [1981], which,
however, does not give Definition 3.5.1. The equivalence between our definition (which
480 Notes

we learned from H. Grundling) and the standard one may be proved using a uniqueness
theorem due to Slawny [1972]. As is clear from the fact that to define the CCR-algebra one
quite unnaturally has to equip lC with the discrete topology, it is an object best avoided.
Our definition of the Fock representation is also different from, but equivalent to, that
in Bratteli and Robinson [1981]. In physics this representation is known as the second
quantization of lC; the representation Ui defined in Proposition 3.S.2 is usually called
r(Ux)' See, for example, Reed and Simon [197S].
Proposition 3.S.3 is due to Grundling and Hurst [1987]; the present proof, due to H.
Grundling, fills a gap in the proof in that reference. Araki [1963], Thm. I(S), gave an
arduous proof of the corresponding von Neumann algebra result Jr(2U(V»" = Jr(2U(VJ. »'
for any regular representation Jr. This result follows immediately from 3.S.3, so that the
regularity assumption may evidently be dropped.
The Fock representation of 2U(So.lR. v), with the conventions (3.20), was introduced by
Carey et al. [1977, 1978], who regard it as a rigorous version of the Fermi representation
of quantum electromagnetism.
The treatment of Radon measures based on his own theory of Hilbert subspaces (cf. the
notes to 1I.1.S) may be found in Schwartz [1973]. Related approaches to measure theory on
infinite-dimensional topological vector spaces are presented in Kuo [197S] and Guichardet
[1972]. Malliavin [1997] is entirely concerned with Gaussian measures. Physicists will
enjoy the discussion in DeWitt-Morette et al. [1979] and Choquet-Bruhat et al. [1982].
The Gaussian measure /-ty on V defined by (3.S3) is the image of a so-called cylinder
measure /-t~ on ft, but this way of looking at things is complicated by the fact that in infinite
dimension /-ty(ft) = 0, although /-t~(ft) = I.
Theorem 3.S.S, generalizing the original result of Cameron and Martin [1944] for ft =
L2([0, I], ]Rn) and V = C([O, 1], ]Rn)o, is due to Thomas [1983]. Related results are in Kuo
[I97S] and Malliavin [1997]. A locally convex space is called quasi-complete when all
closed and bounded sets are complete.
Theorem 3.S.7 is due to Landsman and Wiedemann [1994] (also see Wiedemann [1994]),
who used cylinder measures on 9. The conventional treatment of the quantized free
electromagnetic field may be found in Weinberg [199S].
There actually exists a construction of a group algebra for certain infinite-dimensional
groups; see Grundling [1997]. It would be interesting to try to formulate Theorem 3.5.7
using Rieffel induction on this group algebra.

IV.3.6 The literature on two-dimensional Yang-Mills theories is formidable; much of

it was triggered by the work of Rajeev [1988] and Witten [1991, 1992]. A review of the
Euclidean theory is given by Sengupta [1997].
Definition 3.6.1 is due to Rajeev and Rossi [199S]. For Sobolev loop groups see Frenkel
[1984] and Freed [1988]. (A basic reference on smooth loop groups is Pressley and Se-
gal [1986].) Sobolev spaces of paths on Riemannian manifolds are studied in Klingenberg
[1982]. In general dimensions, Sobolev gauge groups and spaces of connections are con-
sidered, e.g., in Mitter and Viallet [1981], Freed and Uhlenbeck [1984], and Kondracki and
Sadowski [1986].
A proof of the smoothness claimed in Lemma 3.6.2 may be found in Freed and Uhlenbeck
[1984], App. A, or in Rajeev and Rossi [I99S]. The latter also contains the description of
9 as a semidirect product, as well as Lemma 3.6.4 (whose heuristic version first appeared
in Rajeev [1988]).
 Reduction and induction 481

Wilson loops come from physics, where one writes (3.73) as

W(A) = P Exp ( - [ da A(a») .

Here P stands for path-ordering; see Dollard and Friedman [1979] for a rigorous discussion
of such "path-ordered" or "product" integrals. Wilson loops are used in Yang-Mills theory
as well as in gravity. An interesting monograph on this topic is Gambini and Pullin [19%],
where further references may be found.
Proposition 3.6.3 and Theorem 3.6.5 are taken from Landsman and Wren [1997]. Gross
[1993], Thm. 2.5, proves the statement preceding Theorem 3.6.5 for almost every A with
respect to the measure tL~' defined in 3.7.
For Definition 3.6.6 see Brocker and tom Dieck [1985]; Proposition 3.6.7 is due to Hall
[1997b].
Langmann and Semenoff [1993] relate the appearance of the Weyl group W in the Stieffel
chamber T / W to the so-called Gribov problem, which occurs when one tries to fix the gauge
in this model. The geometry of Stieffel chambers is discussed in Brocker and tom Dieck
[1985]. The effect of the singularities in these chambers on constrained quantization is
analyzed in Wren [1997, 1998b].
IV.3.7 Proposition 3.7.1 is due to Dimock [1996]. The representation (3.85) has been
considered, in various realizations, by many authors, such as Albeverio and Hoegh-Krohn
[1978], Frenkel [1984], and Ismagilov [1996]. It is a special case of a general class of
"energy" representations of gauge groups in various dimensions introduced by Gelfand
et al. [1977]. Further to these authors, Albeverio et al. [1981] and Wallach [1987] show
that such representations are irreducible when the dimension of space is 2: 3, and provide
criteria for irreducibility in dimension 2. In our case of dimension I, the representation U y
is evidently reducible, but Driver and Hall [1998] prove that as in higher dimensions, it has
no trivial subrepresentation.
A complete proof of Lemma 3.7.3 may be found in Guichardet [1972], Thm. 7.1. Equation
(3.89) in Definition 3.7.4 is motivated by a construction of Dimock [1996], who uses
stochastic calculus. The approach through (3.90) is due to Wren [1998a,b].
For Ito's map and all Wiener measures in this section see Frenkel [1984] and Malliavin and
Malliavin [1990]. The Cameron-Martin formula (3.1 (0) appears as Prop. (5.2.7) in Frenkel
[1984], Thm. 1.3 in Malliavin and Malliavin [1990], and Thm. XI. 1.4.3 in Malliavin [1997].
A further generalization to paths on Riemannian manifolds is given by Hsu [1995]; also cf.
Malliavin [1997].
Different approaches to the quantization of Yang-Mills theory on a cylinder are presented
by Hetrick [1994], Dimock [1996], and Hall and Driver [1998]. For the Euclidean theory
see Witten [1991, 1992].
IV.3.8 For H = U(1), Theorem 3.8.1 and its proof are due to Landsman and Wren
[1997]; the general case was proved by Wren [1998a,b]. The connection between the Wiener
measure and the heat kernel on ]Rn is classical; for loop groups see Frenkel [1984] and
Malliavin and Malliavin [1990]. For smoothness and other properties of general heat kernels
cf. Davies [1989]. The analyticity arguments used in the proof of 3.8.1.5 are developed in
Hall [1994] (who establishes the unique analytic continuation of the heat kernel on H) and
Wren [1998b].
Definition 3.8.3 is due to Hall [1994], who actually defined two inequivalent families
of coherent states for compact Lie groups (a third family was added in Hall [1998]). The
associated Segal-Bargmann transform is studied in Hall [1994, 1997a]. For a review cf.
482 Notes

Hall [1997c]. Hall's coherent states appear, in a different way, also in the quantization of
Yang-Mills theory on a circle by Hall and Driver [1998].
We have not addressed the rather difficult issue of the Hamiltonian of the theory; see
Dimock [1996] and Wren [1998a,b].
1V.3.9 The role of ]1') (Q) and the associated "B-angles" in the quantization of a particle
on a multiply connected configuration space was independently discovered by Schulman
[1968, 1981] and Souriau [1969, 1997]. Another important early paper is Laidlaw and
DeWitt [1970]. More recent treatments, all different from ours, are Sniatycki [1980], Isham
[1983], Horvathy et al. [1989], Balachandran et al. [1991], and Giulini [1995].
In the works of all these authors the fundamental group ]I')(Q) plays a central role.
To relate this to discrete reduction, we generalize the discussion of multiply connected
Lie groups in the main text. Recall that a multiply connected space Q may be written as
Q = Q/]I')(Q), where Q is the universal covering space of Q. As for Q = G, we have
T' Q ::::: (T' Q)/]I') (Q). Hence the inequivalent quantizations of T' Q are labeled by the
unitary dual of ]I')(Q).
The emergence of B-angles in quantum field theory was discovered by Lowenstein and
Swieca [ 1971]. It later turned out that such angles are relevant to quantum chromodynamics;
the physics literature is reviewed by Jackiw [1985] and Weinberg [1996]. For U(l) gauge
theory on the circle also see Manton [1985]. There is a fundamental difference between
8-angles in quantum field theories on a compact space, which are of a purely topological
nature, as discussed in the main text, and also in Asorey [1981] and Jackiw [1985], and
B-angles in theories on a noncompact space. The latter are of a dynamical origin, and
are closely related to the infrared behavior of the theory. See Acerbi et al. [1993b], and
Loffelholz et al. [1996] for a rigorous discussion.
The treatment of quantum mechanics on the circle is taken from Landsman [1990b];
an alternative mathematical discussion may be found in Isham [1983] and in Asorey et al.
[1983]. The Aharonov-Bohm effect was discovered by Aharonov and Bohm [1959]; for a
rigorous discussion see Asorey [1982] and Ruijsenaars [1983]. The easiest way to prove
(3.120) is to use the theorem of Dixmier and Malliavin [1978] quoted in the notes to III. 1.5.
This yields the boundary condition on IJI; the precise domain then follows from Example 1
in Section X.I of Reed and Simon [1975].
Our approach to 8-angles in constrained quantization, taken from Landsman and Wren
[1997], is intended to explain the origin of B-angles in quantum Marsden-Weinstein reduc-
tion by a disconnected gauge group. (The theory and applications of discrete reduction in
classical mechanics may be found in Marsden [1992].) We hereby complement treatments
based on Dirac's quantization method (such as the one of Jackiw [1985]).
References

Abadie, B. and R. Exel [1997] Defonnation quantization via Fell bundles. e-print Junct-
anI970600.
Abbati, M.C., R. Cirelli, P. Lanzavecchia, and A. Mania [1984] Pure states of general
quantum-mechanical systems as Kahler bundles. Nuovo Cim. B83, 43-59.
Abraham, R. and J .E. Marsden [1985] Foundations ofMechanics, 2nd ed. Addison Wesley,
Redwood City.
Acerbi, F., G. Morchio, and F. Strocchi [1993a] Infrared singular fields and nonregular
representations of canonical commutation relation algebras. J. Math. Phys. 34 (1993),
899-914.
Acerbi, E, G. Morchio, and F. Strocchi [1993b] Theta vacua, charge confinement and
charged sectors from nonregular representations of CCR algebras. Lett. Math. Phys. 27,
1-11.
Aharonov, Y. and D. Bohm [1959] Significance of electromagnetic potentials in quantum
theory. Phys. Rev. 115, 485-491.
Ajupov, S.A., B. Iochum, and N.D. Yadgorov [1990] Symmetry versus facial homogeneity
for self-dual cones. Lin. Alg. Appl. 142, 83-89.
Akemann, C.A. [1969] The general Stone-Weierstrass problem. J. Funct. Anal. 4, 227-294.
Akemann, C.A. and G.K. Pedersen [1992] Facial structure in operator algebra theory. Proc.
London Math. Soc. 64, 418-448.
Akemann, C.A. and F.W. Shultz [1985] Perfect CO-algebras. Mem. Amer. Math. Soc. 326,
1-117.
Albert, C. and P. Dazord [1988] Theorie des groupoides symplectiques. I. Theorie generale
des groupoides de Lie. Publ. Dept. Math. Univ. C. Bernard-Lyon J (nouv. ser.) B88-4,
51-105.
Albert, C. and P. Dazord [1990] Theorie des groupoides symplectiques. II. Groupoides
symplectiques. Publ. Dept. Math. Univ. C. Bernard-Lyon J (nouv. ser.), 27-99.
484 References

Alberti, P.M. [1983] A note on transition probabilityover C* -algebras. Lett. Math. Phys. 7,
25-32.
Albertin, U.K. [1991] The diffeomorphism group and flat principal bundles. 1. Math. Phys.
32,1975-1980.
Albeverio, S. and R. H0egh-Krohn [1977] Oscillatory integrals and the method of stationary
phase in infinitely many dimensions, with applications to the classical limit of quantum
mechanics. Inv. Math. 40, 59-106.
Albeverio, S. and R. H0egh-Krohn [1978]. The energy representation of Sobolev-Lie
groups. Compositio Math. 36, 37-52.
Albeverio, S., R. H0egh-Krohn, and D. Testard [1981] Irreducibility and reducibility for the
energy representation of the group of mappings of a Riemannian manifold into a compact
semisimple Lie group. 1. Funct. Anal. 41, 378-396.
Alekseevsy, D., J. Grabowski, G. Marmo, and P.W. Michor [1994] Poisson structures on
the cotangent bundle of a Lie group or a principle bundle and their reductions. 1. Math.
Phys.3s,4909-4927.
Alfsen, E.M. [19701 Compact Convex Sets and Boundary Integrals. Springer, Berlin.
Alfsen, E.M. [1977J On the state spaces of Jordan and C*-algebras. In: Connes, A. (ed.)
Algebres d' operateurs et leurs applications en physique mathematique. Editions CNRS,
Paris.
Alfsen, E.M., H. Hanche-Olsen, and F.W. Shultz [1980] State spaces of C* -algebras. Acta
Math. 144, 267-305.
Alfsen, E.M. and F.W. Shultz [1976] Non-commutative spectral theory for affine function
spaces on convex sets. Mem. Amer. Math. Soc. 172, 1-120.
Alfsen, E.M. and F.w, Shultz [1978] State spaces of Jordan algebras. Acta Math. 140,
155-190.
Alfsen, E.M. and F.W. Shultz [1979] On non-commutative spectral theory and Jordan
algebras. Proc. London Math. Soc. 38, 497-516.
Alfsen, E.M. and F.w, Shultz [1998] On orientation and dynamics in operator algebras. Part
1. Commun. Math. Phys. 194,87-108.
Alfsen, E.M., F.W. Shultz, and E. St0rmer [1978] A Gelfand-Neumark theorem for Jordan
algebras. Adv. Math. 28, 11-56.
Ali, S.T. [1985] Stochastic localisation, quantum mechanics on phase space and quantum
space-time. Riv. Nuovo Cim. 8 (11),1-128.
Ali, S.T. [1998] Systems of covariance in relativistic quantum mechanics. Int. 1. Theor.
Phys.37,365-373.
Ali, S.T., J.-P. Antoine, J.-P. Gazeau, and U.A. Mueller [1995] Coherent states and their
generalizations: a mathematical overview. Rev. Math. Phys. 7, 1013-1104.
Ali, S.T. and H.-D. Doebner [1990] Ordering problem in quantum mechanics: prime
quantization and a physical interpretation. Phys. Rev. A41, 1199-1210.
Ali, S.T. and G.G. Emch [1974] Fuzzy observables in quantum mechanics. 1. Math. Phys.
15,176--182.
Ali, S.T. and G.A. Goldin [1991] Quantization, coherent states and diffeomorphism groups.
In: Henning, J.D., W. LUcke, and J. Tolar (eds.) Differential Geometry, Group Rep-
resentations, and Quantization, pp. 147-178. Lecture Notes in Physics 379. Springer,
Berlin.
 References 485

Alvennann, K. [1985] The multiplicative triangle inequality in noncommutative J B and

J B'-algebras. Abh. Math. Sem. UnL Hamburg 55, 91-96.
Amemiya, I. and H. Araki [1966] A remark on Piron's paper. Publ. RIMS (Kyoto) A2,
423-427.
Anandan, J. [1991] A geometric approach to quantum mechanics. Found. Phys. 21, 1265-
1284.
Angennann, B., H.-D. Doebner, and J. Tolar [1983] Quantum kinematics on smooth man-
ifolds. In: Andersson, S.l. and H.-D. Doebner (eds.) Nonlinear Partial Differential
Operators and Quantization Procedures, pp. 171-208. Lecture Notes in Mathematics
1037. Springer, Berlin.
Appelquist, T., A. Chodos, and P.G.O. Freund [1985] (eds.) Modern Kaluza-Klein Theories.
Benjamin-Cummings, Menlo Park.
Arai, T. [1995] Some extensions of the semiclassical limit n-+ 0 for Wigner functions on
phase space. J. Math. Phys. 36, 622-630.
Araki, H. [1963] A lattice of von Neumann algebras associated with the quantum theory of
a free Bose field. J. Math. Phys. 4,1343-1362.
Araki, H. [1972] Bures distance function and a generalization of Sakai's non-commutative
Radon-Nikodym theorem. Publ. RIMS Kyoto 8, 335-362.
Araki, H. [1980] On a characterization of the state space of quantum mechanics. Commun.
Math. Phys. 75, 1-24.
Araki, H. and G.A. Elliott [1973] On the definition of C'-algebras. Publ. RIMS 9, 93-112.
Araki, H. and G.A. Raggio [1982] A remark on transition probability. Lett. Math. Phys. 6,
237-240.
Archbold, R.J. and EW. Shultz [1989] Characterization of C"-algebras with continuous
trace by properties of their pure states. Pac. J. Math. 136, 1-13.
Arens, R. [1971a] Classical Lorentz invariant particles. J. Math. Phys. 12, 2415-2422.
Arens, R. [1971a] Classical relativistic particles. Commun. Math. Phys. 21, 139-149.
Arhangel'skii, A.V., K.R. Goodearl, and B. Huisgen-Zimmennann [1997] Kiiti Morita
1915-1995. Notices Amer. Math. Soc. 44, 680-684.
Anns, J.M. [1981] The structure of the solution set for the Yang-Mills equations. Math.
Proc. Camb. Phil. Soc. 90, 361-372.
Anns, J.M. [1986] Symmetry and solution set singularities in Hamiltonian field theories.
Acta Phys. Polon. B17, 499-523.
Anns, J. [1996] Reduction of Poisson algebras at nonzero momentum values. J. Geom.
Phys. 21, 81-95.
Arms, J.M .. R.H. Cushman, and M.J. Gotay [1991] A universal reduction procedure for
Hamiltonian group actions. In: Ratiu, T. (ed.) The Geometry ofHamiltonian Systems, pp.
33-52. Springer, New York.
Anns, J.M., M.J. Gotay, and G. Jennings [1990] Geometric and algebraic reduction for
singular momentum maps. Adv. Math. 79, 43-103.
Anns, J .M., J.E. Marsden, and V. Moncrief [1981] Symmetry and bifurcation of momentum
mappings. Commun. Math. Phys. 78, 455-478.
486 References

Anns, J .M., J .E. Marsden, and V. Moncrief [1982] The structure of the space of solutions
of Einstein's equations II: Several Killing fields and the Einstein-Yang-Mills equations,
Ann. Phys. (N.Y.) 144, 81-106.
Arnal, D. andJ. Ludwig [1992] Convexity of the moment mapping ofa Lie group.J. Funct.
Anal. 105, 256-300.
Arnold, V.l. [1989] Mathematical Methods of Classical Mechanics, 2nd ed. Springer, New
York.
Arnold, V.I. and A.B. Givental [1990] Symplectic Geometry. In: Arnold, V.I. and S.P.
Novikov (eds.) Dynamical Systems N, pp. 1-136. Springer, Berlin.
Arodz, H. [1983] Motion of a wave packet in an external Yang-Mills field. Acta Phys.
Polon. B14, 757-773.
Arodz, H. [1988] Lagrangian and Hamiltonian fonnulations of dynamics of classical
particles with spin and color. Acta Phys. Polon. B19, 697-708.
Aronszajn, N. [1950] Theory of reproducing kernels. Trans. Amer. Math. Soc. 68, 337-404.
Ashtekar, A., J. Lewandowski, D. Marolf, J. Mourao, and T. Thiemann [1995] Quantization
of diffeomorphism invariant theories of connections with local degrees of freedom. J.
Math. Phys. 36, 6456-6493.
Ashtekar, A. and T.A. Schilling [1998] Geometrical fonnulation of quantum mechanics.
e-print gr-qc/9706069.
Ashtekar, A. and M. Stillennan [1986] Geometric quantization and constrained systems. J.
Math. Phys. 27, 1319-1330.
Ashtekar, A. and R.S. Tate [1994] An algebraic extension of Dirac quantization: examples.
J. Math. Phys. 35,6434-6470.
Asimow, L. and A.J. Ellis [1980] Convexity Theory and its Applications in Functional
Analysis. Academic Press, London.
Asorey, M. [1981] Some remarks on the classical vacuum structure of gauge field theories.
J. Math. Phys. 22,179-184.
Asorey, M. [1982] Regularity of gauge equivalence in quantum mechanics and the
Aharonov-Bohm effect. Lett. Math. Phys. 6, 429-435.
Asorey, M., L.J. Boya, and J.F. Carinena [1985] Covariant representations in a fibre bundle
framework. Rep. Math. Phys. 21, 391-404.
Asorey, M., J.G. Esteve, and A.F. Pacheco [1983] Planar rotor - The 9-vacuum structure,
and some approximate methods in quantum mechanics. Phys. Rev. D27, 1852-1868.
Atiyah, M.F. [1957] Complex and analytic connections in fibre bundles. Trans. Amer. Math.
Soc. 85, 181-207.
Atiyah, M.F. [1982] Convexity and commuting Hamiltonians. Bull. London Math. Soc. 23,
1-15.
Atiyah, M.F. and R. Bott [1983] The Yang-Mills equations over Riemann surfaces. Phil.
Trans. R. Soc. London A308, 524-615.
Audin, M. [1991] The Topology of Torus Actions on Symplectic Manifolds. Birkhauser,
Basel.
Auslander, L. and C.C. Moore [1966] Unitary representations of solvable Lie groups. Mem.
Amer. Math. Soc. 62.
 References 487

Ayupov, S.A., A. Rakhimov, and S. Usmanov [1997] Jordan, Real and Lie Structures in
Operator Algebra. Kluwer, Dordrecht.
Azcarraga, A. de and J.M. Izquierdo [1995] Lie Groups. Lie Algebras. Cohomology, and
some Applications in Physics. Cambridge University Press, Cambridge.
Azencott, R. et al. [1981] Geodesiques et diffusions en temps petit. Seminaire de probabilites
Universite de Paris VII. Asterisque 84-85.
Azencott, R. and H. Doss [1985] L'equation de Schrooinger quand Ii ~ 0: une approche
probabiliste. Lecture Notes in Mathematics 1109,1-17.
Bacry, H. [1988] Localizability and Space in Quantum Physics. Lecture Notes in Physics
308. Springer, Heidelberg.
Baer, R. [1952] Linear Algebra and Projective Geometry. Academic Press, New York.
Baguis, P. [1998] Semidirect products and the Pukanszky condition. J. Geom. Phys. 25,
245-270.
Bais, EA. and P. Batenburg [1985] A new class of higher-dimensional Kaluza-Klein
monopole and instanton solutions. Nucl. Phys. B253, 162-172.
Balachandran, A.P., G. Marmo, N. Mukunda, J.S. Nilsson, E.C.G. Sudarshan, and E Za-
ccaria [1984] Non-abelian monopoles break color. 1. Classical Mechanics. Phys. Rev.
D29,2919-2935.
Balachandran, A.P., G. Marmo, B.S. Skagerstam, and A. Stem [1983] Gauge Symmetries
and Fiber Bundles - Applications to Particle Dynamics. Lecture Notes in Physics 188.
Berlin, Springer.
Balachandran, A.P., G. Marmo, B.S. Skagerstam, A. Stem [1991] Classical Topology and
Quantum States. World Scientific, Singapore.
Bargmann, V. [1961] On a Hilbert space of analytic functions and an associated integral
transform. Comm. Pure Appl. Math. 14, 187-214.
Bargmann. V. [1964] Note on Wigner's theorem on symmetry operations. J. Math. Phys. 5,
862-868.
Barmoshe, D. and M.S. Marinov [1994] Realization of compact Lie algebras in Kahler
manifolds. J. Phys. A27, 6287-6298.
Barnes, I.M.S., M. Nauenberg, M. Nockleby, and S. Tomsovic [1994] Classical orbits and
semiclassical wavepacket propagation in the Coulomb potential. J. Phys. A27, 3299-
3321.
Barut, A.O. and R. Ra~ka [1977] Theory ofGroup Representations and Applications. PWN,
Warszawa.
Bass, H. [1968] Algebraic K-Theory. Benjamin, New York.
Bates, S. and A. Weinstein [1995] Lectures on the Geometry of Quantization. Berkeley
Mathematics Lecture Notes 8. University of California, Berkeley.
Bayen, E, M. Flato, C. Fronsdal, A. Lichnerowicz, and D. Stemheimer [1978] Deformation
theory and quantization I, II. Ann. Phys. (N.Y.) 110, 61-110, 111-151.
Belinfante, J.G.E [1976] Transition probability spaces. J. Math. Phys. 17,285-290.
Bellissard, J. and M. Vittot [1990] Heisenberg's picture and non commutative geometry of
the semi classical limit in quantum mechanics. Ann.lnst. H. Poincare A52, 175-235.
Belov, V.V. and V.P. Maslov [1990] Quasi-classical trajectory coherent states in quantum
mechanics with gauge fields. Dokl. Akad. Nauk. 311, 849-854.
488 References

Beltrametti, E.G. and G. Cassinelli [1984] The Logic of Quantum Mechanics. Cambridge
University Press, Cambridge.
Bennett, J.G. [1978] Induced representations of C* -algebras and complete positivity. Trans.
Amer. Math. Soc. 243,1-36.
Berezanskii, Ju.M. [1968] Expansions in Eigenfunctions of Self-Adjoint Operators.
American Mathematical Society, Providence.
Berezin, F.A. [1967] Some remarks about the associative envelope of a Lie algebra. Funct.
Anal. Appl. 1,91-102.
Berezin, F.A. [1972] Covariant and contravariant symbols of operators. Math. USSR Izv. 6,
1117-1151.
Berezin, F.A. [1974] Quantization. Math. USSR Izv. 8, 1109-1163.
Berezin, F.A. [1975a] Quantization in complex symmetric spaces. Math. USSR Izv. 9, 341-
379.
Berezin, F.A. [1975b] General concept of quantization. Commun. Math. Phys. 40, 153-174.
Berger, C.A. and L.A. Coburn [1986] Toeplitz operators and quantum mechanics. J. Funct.
Anal. 68, 273-299.
Berger, C.A. and L.A. Coburn [1994] Heat flow and Berezin-Toeplitz estimates. Amer. J.
Math. 116, 563-590.
Berline, N. and M. Vergne [1985] A computation of the equivariant index of the Dirac
operator. Bull. Soc. Math. France 113, 305-345.
Berry, M.V. and N.L. Balazs [1979] Evolution of semiclassical quantum states in phase
space. J. Phys. A12, 625-642.
Biedenham, L.C. and J.D. Louck [198Ia] Angular Momentum in Quantum Physics.
Encyclopedia of mathematics and its applications, Vol. 8. Addison Wesley, Reading.
Biedenham, L.C. and J.D. Louck [1981 b] The Racah-Wigner Algebra in Quantum Physics.
Encyclopedia of mathematics and its applications, Vol. 9. Addison Wesley, Reading.
Bigonnet, B. [1988] Construction d'une C*-algebre associee a certains feuilletages de
Stefan. C. R. Acad. Sci. Paris Ser.1 Math. 307, 307-310.
Binz, E., J. Sniatycki, and H. Fischer [1988] The Geometry of Classical Fields. North-
Holland, Amsterdam.
Birkhoff, G. [1967] Lattice Theory (3d ed.). Amer. Math. Soc. Coli. Publ. 25. American
Mathematical Society, Providence.
Birkhoff, G. and J. von Neumann [1936] The logic of quantum mechanics. Ann. Math. 37,
823-843.
Blackadar, B. [1986] K-Theoryfor Operator Algebras. Springer, New York.
Blanchard, E. [1996] Deformations de C'-algebras de Hopf. Bull. Soc. math. France 124,
141-215.
Blanchard, Ph. and M. Sirugue [1985] Large deviations from classical paths. Hamiltonian
flows as classical limits of quantum flows. Commun. Math. Phys. 101, 173-185.
Blau, M. [1988] On the geometric quantisation of constrained systems. Class. Quantum
Grav. 5, 1033-1044.
Bogolyubov, N.N. et al. [1981] Berezin, Feliks Aleksandrovich - Obituary. Russ. Math.
Surv. 36, 209-216.
 References 489

Bokobza-Haggiag, J. [1969] Operateurs pseudo--differentiels surune variete differentiable.

Ann.lnst. Fourier Grenoble 19,125-177.
Borceux, E and G. van den Bossche [1989] An essay on noncommutative topology. Topology
and Its Appl. 31, 203-223.
Bordemann, M., E. Meinrenken, and M. Schlichenmaier [1994] Toeplitz quantization of
Kiihlermanifolds and gl(N), N ~ 00 limits. Commun. Math. Phys. 165,281-296.
Born, M. [1926] Zur Quantenmechanik der Stossvorgiinge. Z. Phys. 37, 863-867.
Borsari, I. and S. Graffi [1994] Symbols of operators and quantum evolution. J. Math. Phys.
35, 4439-4450.
Borthwick, D., A. Lesniewski, and M. Rinaldi [1995] Notes on the structure of quantized
Hermitian symmetric spaces. Rev. Math. Phys. 7, 871-891.
Borthwick, D., A. Lesniewski, and H. Upmeier [1993] Non-perturbative deformation
quantization of Cartan domains. J. Funct. Anal. 113, 153-176.
Bott, R. [1957] Homogeneous vector bundles. Ann. Math. 66, 203-248.
Bott, R. and L. Th [1982] Differential Forms in Algebraic Topology. Springer, New York.
Bourbaki, N. [1963] Integration. Hermann, Paris.
Boutet de Monvel, L. and V. Guillemin [1981] The Spectral Theory of Toeplitz Operators.
Annals of Mathematics Studies 99. Princeton University Press, Princeton.
Bowes, D. and K.C. Hannabuss [1997J Weyl quantization and star products. J. Geom. Phys.
22,319-348.
Bratteli, O. and D.W. Robinson [1987] Operator Algebras and Quantum Statistical Me-
chanics, Vol. I: C' - and W' -Algebras, Symmetry Groups, Decomposition of States, 2nd
ed. Springer, Berlin.
Bratteli, O. and D.W. Robinson [1981] Operator Algebras and Quantum Statistical
Mechanics, Vol. II: Equilibrium States, Models in Statistical Mechanics. Springer, Berlin.
Bredon, G. [1972] Introduction to Compact Transformation Groups. Academic Press, New
York.
BrOCker, T. and T. tom Dieck [1985] Representations of compact Lie groups. Springer,
Berlin.
BrOCker, T. and R.E Werner [1995] Mixed states with positive Wigner function. J. Math.
Phys. 36, 62-75.
Brody, D.C. and L.P. Hughston [1998J Geometrisation of statistical mechanics. e-print
gr-qcl9708032.
Brooke, J.A. and EE. Schroeck [1996] Localization of the photon on phase space. J. Math.
Phys.37,5958-5986.
Brown, L.G. [1992] Complements to various Stone-Weierstrass theorems for C'-algebras
and a theorem of Shultz. Commun. Math. Phys. 143, 405-413.
Brown, L.G., P. Green, and M.A. Rieffel [1977] Stable isomorphism and strong Morita
equivalence of C' -algebras. Pac. J. Math. 71, 349-363.
Brown, R. [1987] From groups to groupoids: a brief survey. Bull. London Math. Soc. 19,
113-134.
Brown, R. and O. Mucuk [1995] The monodromy groupoid of a Lie groupoid. Cah. Top.
Geom. Diff. Categ. 36, 345-369.
490 References

Brown, R. and O. Mucuk [1996] Foliations,locally Lie groupoids and holonomy. Cah. Top.
Geom. Diff. Categ. 37,61-71.
Brummelhuis, R. and A. Uribe [1991] A semi-classical trace formula for Schrooinger
operators. Commun. Math. Phys. 136, 567-584.
Brummelhuis, R., T. Paul, and A. Uribe [1995] Spectral estimates around a critical level.
Duke Math. J. 78, 477-530.
Bub, J. [1981] Hidden variables and quantum mechanics - a sceptical review. Erkenntnis
16, 275-293.
Buchholz, D. [1996] Quarks, gluons, color: facts or fiction? Nucl. Phys. B469 (1996),
333-353.
Busby, R.C. and H.A. Smith [1970] Representations of twisted group algebras. Trans. Amer.
Math. Soc. 149,503-537.
Busch, P., M. Grabowski and P.J. Lahti [1995] Operational Quantum Physics. Springer,
Berlin.
Butterfield, J.N. and G. Fleming [1998] Strange positions. In: Butterfield, J.N. and C. Pag-
onis (eds.) From Physics to Philosophy. A Festschrift for Michael Redhead. Cambridge
University Press, Cambridge.
Cahen, M., S. Gutt, and J. Rawnsley [1990] Quantization of Kahler manifolds I. J. Geom.
Phys. 7, 45-62.
Cahen, M., S. Gutt, and J. Rawnsley [1993] Quantization of Kahler manifolds II. Trans.
Amer. Math. Soc. 337,73-98.
Cahen, M., S. Gutt, and J. Rawnsley [1994] Quantization of Kahler manifolds III. Lett.
Math. Phys. 30,291-305.
Cahen, M., S. Gutt, and J. Rawnsley [1995] Quantization of Kahler manifolds IV. Lett.
Math. Phys. 34, 159-168.
Calderon, A. and R. Vaillancourt [1971] On the boundedness of pseudo-differential
operators. J. Math. Soc. Japan 23, 374-378.
Cameron, R.H. and W.T. Martin [1944] Transformation of Wiener integrals under
translations. Ann. Math. 45, 386-396.
Camporesi, R. [1990] Harmonic analysis and propagators on homogeneous spaces. Phys.
Rep. 196,1-134.
Cant, A. [1981] Invariant connections and magnetic monopoles. J. Math. Phys. 22, 2283-
2288.
Cantoni, V. [1975] Generalized "transition probability". Commun. Math. Phys. 44, 125-128.
Carey, A.L., J.M. Gaffney, and C.A. Hurst [1977] A C"-algebraic formulation of the
quantization of the electromagnetic field. J. Math. Phys. 18, 629-640.
Carey, A.L., J.M. Gaffney, and C.A. Hurst [1978] A C"-algebraic formulation of gauge
transformations of the second kind for the electromagnetic field. Rep. Math. Phys. 13,
419-436.
Carinena, J .F., J .M. Gracia-Bondia, and J .C. Varllly [1990] Relativistic quantum kinematics
in the Moyal representation. J. Phys. A23, 901-933.
Carinena, J.F. and M. Santander [1975] On the projective unitary representations of
connected Lie groups. J. Math. Phys. 16, 1416-1420.
Cartan, E. [1958] Lerons sur les Invariants Integraux. 2nd edt Hermann, Paris.
 References 491

Castrigiano, D.P.L. and R. W. Henrichs [1980] Systems of covariance and subrepresentations

of induced representations. Lett. Math. Phys. 4, 169-175.
Cattaneo, U. [1979] On Mackey's imprimitivity theorem. Comment. Math. Helvetici 54,
629-641.
Cheeger, J., M. Gromow, and M.E. Taylor [1982] Finite propagation speed, kernel esti-
mates for functions of the Laplace operator, and the geometry of complete Riemannian
manifolds. J. Diff. Geom. 17, 15-53.
Chernoff, P.R. [1973] Essential self-adjointness of powers of generators of hyperbolic
equations. J. Funct. Anal. 12,401-414.
Chernoff, P.R. [1995] Irreducible representations of infinite dimensional transformation
groups and Lie algebras I. J. Funct. Anal. 130, 255-282.
Chernoff, P.R. and J.E. Marsden [1974] Properties of Infinite Dimensional Hamiltonian
Systems (LNM 425). Springer, Berlin.
Chiang, C.c., S.C. Lee, and G. Marmo [1985] Lagrangian dynamics on higher-dimensional
spaces with applications to Kaluza-Klein theories. J. Math. Phys. 26, 1083-1092.
Choquet-Bruhat, Y., C. DeWitt-Morette, and M. Dillard-Bleick [1982] Analysis, Manifolds,
and Physics, 2nd ed. North-Holland, Amsterdam.
Choquet-Bruhat, Y. and C. DeWitt-Morette [1989] Analysis, Manifolds, and Physics. Part
II: 92 Applications. North-Holland, Amsterdam.
Chruscinski, D. and J. Kijowski [1996] Equations of motion from field equations and a
gauge-invariant variational principle for the motion of charged particles. J. Geom. Phys.
20, 393-403.
Cirelli, R. and P. Cotta-Ramusino [1973] On the isomorphism of a 'quantum logic' with
the logic of projections in a Hilbert space. Int. J. Theor. Phys. 8, 11-29.
Cirelli, R., P. Lanzavecchia [1984] Hamiltonian vector fields in quantum mechanics. Nuovo
Cimento B79, 271-283.
Cirelli, R., P. Lanzavecchia, and A. Mania [1983] Normal pure states of the von Neumann
algebra of bounded operators as a Kahler manifold. J. Phys. A16, 3829-3835.
Cirelli, R., A. Mania, and L. Pizzochero [1990] Quantum mechanics as an infinite-
dimensional Hamiltonian system with uncertainty structure. J. Math. Phys. 31,
2891-2897.
Cirelli, R., A. Mania, and L. Pizzochero [1994] A functional representation for
non-commutative C* -algebras. Rev. Math. Phys. 6, 675-697.
Coburn, L.A. [1992] Deformation' estimates for the Berezin-Toeplitz quantization.
Commun. Math. Phys. 149,415-424.
Coburn, L.A. and J. Xia [1995] Toeplitz algebras and Rieffel deformations. Commun. Math.
Phys. 168, 23-38.
Colin de Verdiere, Y. [1985] Ergodicite et fonctions propres. Commun. Math. Phys. 102,
497-502.
Colin de Verdiere, Y. and B. Parisse [1994] Equilibre instable en regime semi-c1assique. I.
Concentration microlocale. Comm. Partial Differential Equations 19, 1535-1563.
Combes, J.M., R. Schrader, and R. Seiler [1978] Classical bounds and limits for energy
distributions of Hamilton operators in electromagnetic fields. Ann. Phys. (N.Y.) Ul,
1-18.
492 References

Combescure, M. [1992] The squeezed state approach of the semiclassical limit of the time-
dependent Schrodinger equation. J. Math. Phys. 33, 3870-3880.
Connes, A. [1980] A survey offoliations and operator algebras. In: Kadison, R.Y. (ed.) Op-
erator Algebras and Applications, Proc. Symp. Pure Math. 38(1), pp. 521-628. American
Mathematical Society, Providence.
Connes, A. [1994] Noncommutative Geometry. Academic Press, San Diego.
Coquereaux, R. and A. Jadczyk [1988] Riemannian Geometry, Fibre Bundles, Kaluza-
Klein Theories and all that. World Scientific, Singapore.
Cordes, H.O. [1987] Spectral Theory oj Linear Differential Operators and Comparison
Algebras. (LMS Lecture Notes 76). Cambridge University Press, Cambridge.
Cordes, H.O. [1995] The Technique oJPseudodifferential Operators. (LMS Lecture Notes
172). Cambridge University Press, Cambridge.
Corwin, L. and EP. Greenleaf [1989] Representations oJNilpotent Lie Groups and Their
Applications, Part I. Cambridge University Press, Cambridge.
Coste, A., P. Dazord, and A. Weinstein [1987] Groupoides symplectiques. Publ. Dept. Math.
Univ. C. Bernard-Lyon I2A, 1-62.
Courant, TJ. [1990] Dirac Manifolds. Trans. Amer. Math. Soc. 319, 631-661.
Cushman, R.H. and L.M. Bates [1997] Global Aspects oJIntegrable Systems. Birkhauser,
Basel.
Cycon, H.L., R.G. Froese, W. Kirsch, and B. Simon [1987] SchrOdinger Operators, with
Applications to Quantum Mechanics and Global Geometry. Springer, Berlin.
Daubechies, I. [1980] On the distributions corresponding to bounded operators in the Weyl
quantization. Commun. Math. Phys. 75, 229-238.
Daubechies, I. [1983] Continuity statements and counterintuitive examples in connection
with Weyl quantization. J. Math. Phys. 24, 1453-1461.
Davidson, K.R. [1996] C'-Algebras by Example. Fields Institute Monographs 6. American
Mathematical Society, Providence (RI).
Davies, E.B. [1976] Quantum Theory oJOpen Systems. Academic Press, London.
Davies, E.B. [1989] Heat Kernels and Spectral Theory. Cambridge University Press,
Cambridge.
Davies, E.B. and J.T. Lewis [1970] An operational approach to quantum probability.
Commun. Math. Phys. 17,239-260.
Dazord, P. [1985] Feuilletages a singularites. Indag. Math. 47, 21-39.
Dazord, P. and T. Delzant [1987] Le probleme general des variables actions-angles. J. Diff.
Geom. 26, 223-251.
DeWitt-Morette, C., A. Maheshwari, and B. Nelson [1979] Path integration in
non-relativistic quantum mechanics. Phys. Rep. 5, 255-372.
DeWitt-Morette, C., B. Nelson, and T.-R. Zhang [1983] Caustic problems in quantum
mechanics with applications in scattering theory. Phys. Rev. D28, 2526--2546.
Dimock, J. [1996] Canonical quantization of Yang-Mills on a circle. Rev. Math. Phys. 8,
85-102.
Dirac, P.A.M. [1930] The Principles oJQuantum Mechanics. Clarendon Press, Oxford.
Dirac, P.A.M. [1931] Quantized singularities in the electromagnetic field. Proc. R. Soc.
London A133, 60-72.
 References 493

Dirac, P.A.M. [1950] Generalized Hamiltonian Systems. Can. J. Math. 12, 129-148.
Dirac, P.A.M. [1964] Lectures on Quantum Mechanics. Belfer School of Science, Yeshiva
University, New York.
Dixmier, J. [1960] Sur les representations unitaires des groupes de Lie nilpotents, IV. Can.
J. Math. 12, 324-352.
Dixmier, J. [1977] C'-Algebras. North-Holland, Amsterdam.
Dixmier, J. and P. Malliavin [1978] Factorisations de fonctions et de vecteurs indefinements
differentiables. Bull. Soc. Math. France 102, 305-330.
Dobrushin, R.L., R.A. Minlos, M.A. Shubin, and A.M. Vershik (eds.) [1996] Contempo-
rary Mathematical Physics. F.A. Berezin Memorial Volume. AMS Translations 2-175.
American Mathematical Society, Providence.
Doebner, H.D. and J. Tolar [1975] Quantum mechanics on homogeneous spaces. J. Math.
Phys. 16, 975-984.
Doebner, H.D. and J. Tolar [1990] Mackey's quantization and invariant connections.
In: Niederle, J. and J. Fischer, (eds.) Selected Topics in Quantum Field Theory and
Mathematical Physics, pp. 234-238. World Scientific, Singapore.
Dollard, J.D. and C.N. Friedman [1979] Product Integration. Addison-Wesley, London.
Doplicher, S., D. Kastler, and D. W. Robinson [1966] Covariance algebras in field theory
and statistical mechanics. Commun. Math. Phys. 3, 1-28.
Doran, R.S. [1994] (ed.) C'-algebras: 1943-1993. Con temp. Math. 167. American
Mathematical Society, Providence.
Doran, R.S. and V.A. Belfi [1986] Characterizations o/C' -algebras. M. Dekker, New York.
Dowker, J.S. [1974] Covariant Schrooinger equations. In: Arthurs, A.M. (ed.) Functional
Integration and its Applications, pp. 34-52. Clarendon Press, Oxford.
Driver, B.K. and B.C. Hall [1998] A note on the one-dimensional energy representation.
University of California at San Diego preprint.
Duclos, P. and H. Hogreve [1993] On the semiclassical localization of the quantum
probability.J. Math. Phys. 34,1681-1691.
Duffield, N.G. [1990] Classical and thermodynamic limits for generalized quantum spin
systems. Commun. Math. Phys. 127, 27-39.
Duval, C., and J. Elhadad [1992] Geometric quantization and localization of relativistic
spin systems. Contemp. Math. 132,317-330.
Duval, C., J. Elhadad, M.J. Gotay, J. Sniatycki, and G.M. Thynman [1991] Quantization
and bosonic BRST theory. Ann. Phys. (NY) 206, 1-26.
Duval, C., J. Elhadad, and G.M. Thynman [1992] Pukanszky's condition and symplectic
induction. J. Dijf. Geom. 36, 331-348.
Duval, C. and P. Horvathy [1982] Particles with internal structure: the geometry of classical
motions and conservation laws. Ann. Phys. (N.Y.) 142, 10-33.
Ebin, D.G. and J.E. Marsden [1970] Groups of diffeomorphisms and the motion of an
incompressible fluid. Ann. Math. 92, 102-163.
Edwards, C.M. and G.T. Riittimann [1985] On the facial structure of the unit balls in a
G L-space and its dual. Math. Proc. Camb. Phil. Soc. 98, 305-322.
Effros, E.G. [1963] Order ideals in a C'-algebra and its dual. Duke Math. J. 30, 391-412.
494 References

Effros, E. and F. Hahn [1967] Locally compact transformation groups and C* -algebras.
Mem. Amer. Math. Soc. 75.
Ehresmann, C. [1958] Categories topologiques et categories differentiables. In: Col-
loque de Geometrie Differentielle Globale, pp. 137-150. Centre Beige des Recherches
Mathematiques, Brussels.
Elliott, G.A., T. Natsume, and R. Nest [1993] The Heisenberg group and K-theory. K -Theory
7,409-428.
Elliott, G.A., T. Natsume, and R. Nest [1996] The Atiyah-Singer index theorem as passage
to the classical limit in quantum mechanics. Commun. Math. Phys. 182,505-533.
Elworthy, D. and A. Truman [1981] Classical mechanics, the diffusion (heat) equation and
the Schrooinger equation on a Riemannian manifold. J. Math. Phys. 22, 2144--2166.
Elworthy, D., A. Truman, and K. Watling [1985] The semi-classical expansion for a charged
particle on a curved space background. J. Math. Phys. 26, 984-990.
Emch, G.G. [1972] Algebraic Methods in Statistical Mechanics and Quantum Field Theory.
Wiley, New York.
Emch, G.G. [1982] Quantum and classical mechanics on homogeneous Riemannian
manifolds. J. Math. Phys. 23, 1785-1791.
Emch, G.G. [1983] Geometric dequantization and the correspondence problem. Int. J. Theor.
Phys.22,397-420.
Emch, G.G. [1984] Mathematical and Conceptual Foundations of 20th Century Physics.
North Holland, Amsterdam.
Emmrich, C. [1993a] Equivalence of extrinsic and intrinsic quantization for observables
not preserving the vertical polarization. Commun. Math. Phys. 151,515-530.
Emmrich, C. [1993b] Equivalence of Dirac and intrinsic quantization for non-free group
actions. Commun. Math. Phys. 151,531-542.
Emmrich, C. and H. Romer [1990] Orbifolds as configuration spaces of systems with gauge
symmetries. Commun. Math. Phys. 129, 69-94.
EngliS, M. [1996] Berezin quantization and reproducing kernels on complex domains. Trans.
Amer. Math. Soc. 348, 411-479.
Est, W.T. van [1953] Group cohomology and Lie algebra cohomology in Lie groups. I, II.
Proc. Kon. Ned. Akad. Wet. A56, 484-504.
Estrada, R., J .M. Gracia-Bondfa, and J .C. Varilly [1989] On asymptotic expansions of
twisted products. J. Math. Phys. 30,2789-2796.
Exel, R. [1994] The soft torus: a variational analysis of commutator norms. J. Funct. Anal.
126, (1994), 259-273.
Falk, G. [1951] Uberringe mit Poisson-Klammem. Math. Ann. 123,379-391.
Faraut, J. and A. Koranyi [1994] Analysis on Symmetric Cones. Clarendon Press, Oxford.
Fedosov, B.V. [1994] A simple geometrical construction of deformation quantization. J.
Diff. Geom. 40, 213-238.
Fedosov, B.Y. [1996] Deformation Quantization and Index Theory. Akademie-Verlag,
Berlin.
Feher, L.Gy. [1986] Classical motion of colored test particles along geodesics of a Kaluza-
Klein spacetime. Acta Phys. Hung. 59, 437-444.
 References 495

Fell, J.M.G. [1962] The structure of algebras of operator fields. Acta Math. 106, 237-268.
Fell, J .M.G. [1978] Induced Representations and Banach' -algebra Bundles. Lecture Notes
in Mathematics 582. Springer, Berlin.
Fell, J.M.G. and R.S. Doran [1988] Representations 0/* -Algebras, Locally Compact Groups
and Banach '-Algebraic Bundles, Vol. 2. Academic Press, Boston.
Field, T.R. [1996] The Quantum Complex Structure. D. Phil. thesis, Oxford University.
Figueroa, H., J .M. Gracia-Bondfa, and J .C. Varilly [1990] Moyal quantization with compact
symmetry groups and noncommutative harmonic analysis. J. Math. Phys. 31, 2664-2671.
Fillmore, P.A. [1996] A User's Guide to Operator Algebras. Wiley-Interscience, New York.
Fischer, A.E., J .E. Marsden, and V. Moncrief [1980] The structure of the space of solutions
of Einstein's equations I: One Killing field. Ann. Inst. H. Poincare (phys. Theor.) 33,
147-194.
Fock, V. [1928] Verallgemeinerung und Losung der Diracschen statistischen Gleichung. Z.
Physik 49,339-357.
Folland, G.B. [1989] Harmonic Analysis on Phase Space. Princeton University Press,
Princeton.
Forgacs, P. and N.S. Manton [1980] Space-time symmetries in gauge theories. Commun.
Math. Phys. 72, 15-35.
Frank, M. [1990] Self-duality and C' -reflexivity of Hilbert C' -modules. Zeitschr. Anal.
Anw.9,165-176.
Frank, M. [1998] Hilbert CO-modules and related subjects - a guided reference overview.
http://www.mathematik.uni-leipzig.de/Ml/franklhilmod.html.
Freed, D.S. [1988] The geometry of loop groups. J. Diff. Geom. 28, 223-276.
Frenkel, I. B. [1984] Orbital theory for affine Lie algebras.lnv. Math. 77, 301-352.
Freed, D.S. and K.K. Uhlenbeck [1984]lnstantons and Four-Manifolds. Springer, New
York.
Freyer, K.D. and I. Halperin [1956] The von Neumann coordinatization theorem for
complemented modular lattices. Acta Scient. Math. 17,203-249.
Freyer, K.D. and I. Halperin [1958] On the construction of coordinates for non-Desarguesian
complemented modular lattices. Proc. Kon. Ned. Akad. Wet. A61, 142-161.
Frohlicher, A. and A. Kriegl [1988] Linear Spaces and Differentiation Theory. Wiley,
Chichester.
Gaal, S.A. [1973] Linear Analysis and Representation Theory. Springer, Berlin.
Gadella, M. [1995] Moyal formulation of quantum mechanics. Fortschr. Phys. 43, 229-264.
Gallot, S., D. Hulin and J. Lafontaine [1990] Riemannian Geometry. Springer, Berlin.
Gambini, R. and J. Pullin [1996] Loops, Knots, Gauge Theories and Quantum Gravity.
Cambridge University Press, Cambridge.
Gelfand, I.M., M.I. Graev, and A.M. Vershik [1977] Representations of the group of smooth
mappings of a manifold into a compact Lie group. Compositio Math. 35, 299-334.
Gerstenhaber, M. [1964] On the deformation of rings and algebras. Ann. Math. 79, 59-103.
Giavarini, G. and E. Onofri [1990] Vector coherent states and non-abelian gauge structures
in quantum mechanics. Int. J. Mod. Phys. AS, 4311-4331.
Giles, R. [1970] Foundations for quantum mechanics. J. Math. Phys. 11, 277-322.
496 References

Giles, R. and H. Kummer [1971] A non-commutative generalization of topology. Indiana

Univ. Math. J. 21, 91-102.
Gilmore, R. [1979] The classical limit of quantum nonspin systems. J. Math. Phys. 20,
891-893.
Giulini, D. [1995] Quantum mechanics on spaces with finite fundamental group. Helv. Phys.
Acta 68, 438-469.
Glimm, J. [1962] Families of induced representations. Pac. J. Math. 12, 885-911.
Glimm, J. and R.V. Kadison [1960J Unitary operators in C*-algebras. Pac. J. Math. 10,
547-556.
Goldin, G.A. [1971] Non-relativistic current algebra as unitary representation theory of
groups. J. Math. Phys. 12,462-487.
Goldin, G.A., R. Menikoff and D.H. Sharp [1980] Particle statistics from induced
representations of local current groups. J. Math. Phys. 21, 650-664.
Goodearl, K. [1982] Notes on Real and Complex C' -algebras. Shiva Pub!., Nantwich.
Gootman, E.C. and J. Rosenberg [1979] The structure of crossed product C* -algebras: a
proof of the generalized Effros-Hahn conjecture. Inv. Math. 52, 283-298.
Goreski, M. and R. MacPherson [1988] Stratified Morse Theory. Springer, New York.
Gotay, M. [1986] Constraints, reduction, and quantization. 1. Math. Phys. 27,2051-2066.
Gotay, M.J. and L. Bos [1986] Singular angular momentum mappings. J. Diff. Geom. 24,
181-203.
Gotay, MJ., H.B.G.S. Grundling, and G.M. Tuynman [1996J Obstruction results in
quantization theory. J. Nonlinear Sci. 6, 469-498.
Gotay, MJ. and J .M. Nester [1980] Generalized constraint algorithm and special presym-
plectic manifolds. In: Kaiser, G. and J.E. Marsden (eds.) Geometric Methods in
Mathematical Physics. Lecture Notes in Mathematics 775. Springer, Berlin.
Gotay, M.J., J.M. Nester, and G. Hinds [1978] Presymplectic manifolds and the Dirac-
Bergmann theory of constraints. J. Math. Phys. 19,2388-2399.
Gotay, M.J. and G.M. Tuynman [1991] A symplectic analogue of the Mostow-Palais the-
orem. In: Dazord, P. and A. Weinstein (eds.) Symplectic Geometry, Groupoids, and
Integrable Systems, pp. 173-181. Springer, New York.
Govaerts, J. [1991] Hamiltonian Quantization and Constrained Dynamics. Leuven
University Press, Leuven.
Gracia-Bondla, J.M. [1992] Generalized Moyal quantization on homogeneous symplectic
spaces. Contemp. Math. 134,93-114.
Gracia-Bondia, J.M. and J.C. Varilly [1988] Algebras of distributions suitable for phase-
space quantum mechanics I, II. J. Math. Phys. 29, 869-879, 880-887.
Graffi, S. and A. Parmeggiani [1990] Quantum evolution and classical flow in complex
phase space. Commun. Math. Phys. 128,393-409.
Green, P. [1978] The local structure of twisted covariance algebras. Acta Math. 140, 191-
250.
Green, P. [1978] The structure of imprimitivity algebras. J. Funct. Anal. 36, 88-104.
Greenleaf, F. [1969] Invariant Means on Topological groups. Van Nostrand Reinhold, New
York.
 References 497

Greub, W., S. Halperin, and R. Vanstone [1972] Connections, Curvature, and Cohomology.
Vol. 1: de Rham Cohomology of Manifolds and Vector Bundles. Academic Press, New
York.
Greub, W., S. Halperin, and R. Vanstone [1973] Connections, Curvature, and Cohomology.
Vol. ll: Lie Groups, Principal Bundles and Characteristic Classes. Academic Press, New
York.
Greub, W. and H.-R. Petry [1975] Minimal coupling and complex line bundles. J. Math.
Phys. 16,1347-1351.
Grgin, E. and A. Petersen [1974] Duality of observables and generators in classical and
quantum mechanics. J. Math. Phys. 15,764-769.
Griffiths, P. and J. Harris [1978] Principles ofAlgebraic Geometry. Wiley, New York.
Groenewold, H.J. [1946] On the principles of elementary quantum mechanics. Physica 12,
405-460.
Gn.:mbrek, N. [1995] Morita equivalence for Banach algebras. J. Pure Appl. Alg. 99, 183-219.
Groot, S.R. de, W.A. van Leeuwen, and Ch.G. van Weert [1980] Relativistic Kinetic Theory.
North-Holland, Amsterdam.
Gross, L.P. [1993] Uniqueness of ground states for Schrodinger operators over loop groups.
J. Funct. Anal. 121, 373-441.
Grossmann, A. [1976] Parity operator and quantization of delta-functions. Commun. Math.
Phys.48, 191-194.
Grossmann, A., G. Loupias, and E.M. Stein [1968] An algebra of pseudodifferential
operators and quantum mechanics in phase space. Ann. Inst. Fourier (Grenoble) 18,
343-368.
Grundling, H.B.G.S. [1988] Systems with outer constraints. Gupta-Bleuler electromag-
netism as an algebraic field theory. Commun. Math. Phys. 114,69-91.
Grundling, H.B.G.S. [1997] A group algebra for inductive limit groups. Continuity problems
of the canonical commutation relations. Acta Appl. Math. 46, 107-145.
Grundling, H.B.G.S. and C.A. Hurst [1985] Algebraic quantization of systems with a gauge
degeneracy. Commun. Math. Phys. 98, 369-390.
Grundling, H.B.G.S. and C.A. Hurst [1987] Algebraic structures of degenerate systems and
the indefinite metric. J. Math. Phys. 28, 559-572.
Grundling, H.B.G.S. and C.A. Hurst [1988a] A note on regular states and supplementary
conditions. Lett. Math. Phys. 15,205-212; Err. ibid. 17, 173-174.
Grundling, H.B.G.S. and C.A. Hurst [1988b] The quantum theory of second class
constraints: kinematics. Commun. Math. Phys. 119, 75-93; Err. ibid. 122, 527-529.
Gudder, S.P. [1979] A survey of axiomatic quantum mechanics. In: Hooker, C.A. (ed.)
Logico-Algebraic Approach to Quantum Mechanics, pp. 323-363. D. Reidel, Dordrecht.
Gudder, S.P. [1979] Stochastic Methods in Quantum Mechanics North-Holland, Amster-
dam.
Guichardet, A. [1972] Symmetric Hilbert Spaces and Related Topics. Lecture Notes in
Mathematics 261. Springer, Berlin.
Guichardet, A. [1980] Cohomologie des Groupes Topologiques et des Algebres de Lie.
Nathan, Paris.
498 References

Guichardet, A. [1985] Theorie de Mackey et methode des orbites selon M. Duflo. Expo.
Math. 3, 303-346.
Guillemin, V. [1984] Toeplitz operators in n dimensions. Int. Eq. Op. Th. 7, 145-205.
Guillemin, V. [1994] Moment Maps and Combinatorial Invariants of Hamiltonian P-
Spaces. Birkhiiuser, Boston.
Guillemin, V., E. Lerman, and S. Sternberg [1996] Symplectic jibrations and multiplicity
diagrams. Cambridge University Press, Cambridge. 1996.
Guillemin. V. and S. Sternberg [1977] Geometric Asymptotics. Math. Surveys 14. American
Mathematical Society, Providence.
Guillemin. V. and S. Sternberg [1982] Convexity properties of the moment mapping. I./nv.
Math. 67. 491-513.
Guillemin, V. and S. Sternberg [1984a] Convexity properties of the moment mapping. II.
Inv. Math. 77. 533-546.
Guillemin, V. and S. Sternberg [1984b] Symplectic Techniques in Physics. Cambridge
University Press, Cambridge.
Guillemin, V. and A. Uribe [1985] Band asymptotics on line bundles over S2. J. Diff. Geom.
21. 129-133.
Guillemin, V. and A. Uribe [1986] Clustering theorems with twisted spectra. Math. Ann.
272. 479-506.
Guillemin, V. and A. Uribe [1989] Circular symmetry and the trace formula.lnv. Math. 96,
385-423.
Guillemin, V. and A. Uribe [1990] Reduction and the trace formula. J. Diff. Geom. 32,
315-347.
Gunson, J. [1967] On the algebraic structure of quantum mechanics. Commun. Math. Phys.
6,262-285.
Gutt, s. [1983] An explicit '-product on the cotangent bundle of a Lie group. Lett. Math.
Phys. 7, 249-258.
Gutzwiller, M.C. [1990] Chaos in Classical and Quantum Mechanics. Springer, New York.
Haag, R. [1996] Local Quantum Physics, 2nd ed. Springer, Berlin.
Haag, R., and D. Kastler [1964] An algebraic approach to quantum field theory. J. Math.
Phys. 5, 848-861.
Habib, S. and H.E. Kandrup [1989] Wigner functions and density-matrices in curved spaces
as computational tools. Ann. Phys. (N.Y.) 191, 335-362.
Hagedorn, G.A. [1980] Semiclassical quantum mechanics I: The Ii --+ 0 limit for coherent
states. Commun. Math. Phys. 71, 77-93.
Hagedorn, G .A. [1981] Semiclassical quantum mechanics III: The large order asymptotics
and more general states. Ann. Phys. (N.Y.) 135, 58-70.
Hagedorn, G.A. [1985] Semiclassical quantum mechanics IV: Large order asymptotics and
more general states in more than one dimension. Ann. Inst. H. Poincare A42, 363-374.
Hahn, P. [1978a] Haar measure for measure groupoids. Trans. Amer. Math. Soc. 242, 1-33.
Hahn. P. [1978b] The regular representation of measure groupoids. Trans. Amer. Math. Soc.
242,35-72.
 References 499

Hiijicek, P. [1994] Quantization of systems with constraints. In: Ehlers, J. and H. Friedrich
(eds.) Canonical Gravity: From Classical to Quantum. Lecture Notes in Physics 434, pp.
113-149. Springer, Berlin.
Hall, B.C. [1994] The Segal-Bargmann "coherent state" transform for compact Lie groups.
J. Funct. Anal. 122, 103-151.
Hall, B.C. [1997a] The inverse Segal-Bargmann transform for compact Lie groups. J. Funct.
Anal. 143,98-116.
Hall, B.C. [1997b] Phase space bounds for quantum mechanics on a compact Lie group.
Commun. Math. Phys. 184, 233-250.
Hall, B.C. [1997c] Quantum mechanics in phase space. Contemp. Math. 214,47-62.
Hall, B.C. and B.K. Driver [1998] Yang-Mills theory and the Segal-Bargmann transform.
University of California at San Diego preprint.
Halliwell, U. and J.B. Hartle [1991] Wave functions constructed from an invariant sum
over histories satisfy constraints. Phys. Rev. D43, 1170-1194.
Hanche-Olsen, H. [1985] JB-algebras with tensor product are CO-algebras. Lecture Notes
in Mathematics 1132, 223-229.
Hanche-Olsen, H. and E. St0rmer [1984] Jordan Operator Algebras, Pitman, Boston.
Hannabuss, K.C. [1984] Holomorphic and abstract inducing. Math. Proc. Camb. Phil. Soc.
96, 453-468.
Hamad, J. and J.P. Pare [1991] Kaluza-Klein approach to the motion of non-abelian charged
particles with spin. Class. Quant. Grav. 8, 1427-1444.
Hamad, J., S. Shnider, and L. Vinet [1980] Group actions on principal bundles and invariance
conditions for gauge fields. J. Math. Phys. 21, 2719-2724.
Hartkiimper, A. and H. Neumann [1974] (eds.) Foundations o/Quantum Mechanics and
Ordered Linear Spaces. Lecture Notes in Physics 29. Springer, Berlin.
Hector, G. and U. Hirsch [1986] Introduction to the Geometry o/Foliations. Part A, 2nd ed.
Vieweg, Braunschweig.
Helffer, B. [1988] Semi-classical Analysis for the Schrodinger Operator and Applications.
Lecture Notes in Mathematics 1336. Springer, Berlin.
Helffer, B., A. Martinez, and D. Robert [1987] Ergodicite et limite semi-classique. Commun.
Math. Phys. 109,313-326.
Helgason, S. [1978] Differential Geometry, Lie Groups, and Symmetric Spaces. Academic
Press, New York.
Henneaux, M. and C. Teitelboim [1992] Quantization 0/ Gauge Systems. Princeton
University Press, Princeton.
Hepp, K. [1974] The classical limit of quantum mechanical correlation functions. Commun.
Math. Phys. 35, 265-277.
Hermann, R. [1966] Lie Groups/or Physicists. Benjamin, New York.
Hermann, R. [1973] Topics in the Mathematics 0/ Quantum Theory. MathSci Press,
Brookline.
Hermann, R. [1975] Gauge Fields and Cartan-Ehresmann Connections. MathSci Press,
Brookline.
Hetrick, J. E. [1994] Canonical quantization of two-dimensional gauge fields. Int. J. Mod.
Phys. A9, 3153-3178.
500 References

Higuchi, A. [1991] Quantum linearization instabilities of de Sitter spacetime. I, II. Class.

Quantum Grav. 8,1961-1981,1983-2004.
Hilgevoord, J. and J. Uffink [1991] Uncertainty in prediction and in inference. Found. Phys.
21,323-34l.
Hillery, M., R.E O'Connel, M.O. Scully, and E.P. Wigner [1984] Distribution functions in
physics - Fundamentals. Phys. Rep. 106, 121-167.
Hilsum, M. and G. Skandalis [1987] Morphismes K-orientes d'espaces de feuilles et
fonctorialite en theorie de Kasparov. Ann. scient. Ec. Norm. Sup. (4' s.) 20, 325-390.
Hislop, P.D. and I.M. Sigal [1996] Introduction to Spectral Theory with Applications to
Schrodinger Operators. Springer, New York.
Hitchin, N.J. [1990] Flat connections and geometric quantization. Commun. Math. Phys.
131, 347-380.
Hogreve, H. [1983] The Classical Limit ofQuantum Theories: Particles in External Metrics
and with Spin. Ph.D. thesis, Freie Universitat Berlin.
Hogreve, H., J. Potthoff, and R. Schrader [1983] Classical limits for quantum particles in
external Yang-Mills potentials. Commun. Math. Phys. 91, 573-598.
Holevo, A.S. [1973] Statistical decision theory for quantum systems. J. Multivariate Anal.
3,337-394.
Holevo, A.S. [1982] Probabilistic and Statistical Aspects of Quantum Theory. North-
Holland, Amsterdam.
Holland, S.S. Jr. [1995] Orthornodularity in infinite dimensions; a theorem of M. Soler.
Bull. Amer. Math. Soc. 32, 205-234.
Honnander, L. [1979] The Weyl calculus of pseudo-differential operators. Comm. Pure
App!. Math. 32, 359-443.
Honnander, L. [1983] The Analysis ofLinear Partial Differential Operators, Vo!.I. Springer,
Berlin.
Honnander, L. [1985a] The Analysis of Linear Partial Differential Operators, Vol. JIl.
Springer, Berlin.
Honnander, L. [1985b] The Analysis of Linear Partial Differential Operators, Vol. IV.
Springer, Berlin.
Horowitz, G.T. and Marolf, D. [1995] Quantum probes of spacetime singularities. Phys.
Rev. D52, 5670-5675.
Horvathy, P.A. [1981] Rotational symmetry and Dirac's monopole. Int. J. Theor. Phys. 20,
697-707.
Horvathy, P.A., G. Morandi, and E.C.G. Sudarshan [1989] Inequivalent quantizations in
multiply connected spaces. Nuovo Cim. DU, 201-228.
Hove, L. van [1951] Sur certaines representations unitaires d'un groupe infini de
transfonnations. Mem. Acad. Roy. de Belgique, Classe des Sci. 26,61-102.
Howe, R. [1980] Quantum mechanics and partial differential equations. J. Funct. Anal. 38,
188-254.
Howe, R. [1989] Remarks on classical invariant theory. Trans. Amer. Math. Soc. 313, 539-
570.
Hsu, E.P. [1995] Quasi-invariance of the Wiener measure on the path space over a compact
Riemannian manifold. J. Funct. Ana!' 134,417-450.
 References 501

Hudson, R.L. [1974] When is the Wigner quasi-probability density non-negative? Rep.
Math. Phys. 6, 249-252.
Huebschmann, J. [1990] Poisson cohomology and quantization. J. reine angew. Math. 408,
57-113.
Huebschmann, J. [1996] Poisson geometry of flat connections for SU(2)-bundles on
surfaces. Math. Z. 221, 243-259.
Hughston, L.P. [1995] Geometric aspects of quantum mechanics. Lect. Notes Pure Appl.
Math. 169,59-79.
Hurt, N.E. [1983] Geometric Quantization inAction. Reidel, Dordrecht.
Husimi, K. [1940] Some formal properties of the density matrix. Prog. Phys. Math. Soc.
Japan 22,264-314.
Hwang, I.L. [1987] The L 2 -boundedness of pseudodifferential operators. Trans. Amer.
Math. Soc. 302, 55-76.
Iochum, B. [1984] Cones Autopolaires et Algebres de Jordan. Lecture Notes in Mathematics
1049. Springer, Heidelberg.
Iochum, B. and EW. Shultz [1983] Normal state spaces of Jordan and von Neumann
algebras. J. Funct. Anal. 50, 317-328.
Isenberg, J. and J.E. Marsden [1982] A slice theorem for the space of solutions of Einstein's
equations. Phys. Rep. 89, 179-222.
Isham, c.J. [1983] Topological and global aspects of quantum theory. In: DeWitt, B.S.
and R. Stora (eds.) Relativity, Groups and Topology 2, pp. 1059-1290. North-Holland,
Amsterdam.
Ismagilov, R.S. [1996] Representations of Infinite-Dimensional Groups. American
Mathematical Society, Providence.
Jackiw, R. [1985] Topological investigations of quantized gauge theories. In: Treiman, S.B.
et al. (eds.) Current Algebra and Anomalies, pp. 211-359. World Scientific, Singapore.
Jackiw, R. and N.S. Manton [1980] Symmetries and conservation laws in gauge theories.
Ann. Phys. (N.Y.) 127, 257-273.
Jammer, M. [1974] The Philosophy of Quantum Mechanics. Wiley, New York.
Jiinich, K. [1994] Topologie. 4. Auflage. Springer, Berlin.
Jeffrey, L.C. and J. Weitsman [1992] Bohr-Sommerfeld orbits in the moduli space of flat
connections and the Verlinde dimension formula. Commun. Math. Phys. 150, 593-630.
Jeffrey, L.C. and J. Weitsman [1997] Toric structures on the moduli space of flat connections
on a Riemann surface. II. Inductive decomposition of the moduli space. Math. Ann. 307,
93-108.
Jona-Lasinio, G., E Martinelli, and E. Scoppola [1981] The semi-classical limit of quantum
mechanics: a qualitative theory via stochastic mechanics. Phys. Rep. 77, 313-327.
Jordan, P. [1932] Uber eine Klasse nichtassoziativer hyperkomlexen Algebren. Nachr. Ges.
Wiss. Gottingen, 569-575.
Jordan, P., J. von Neumann, and E.P. Wigner [1934] On an algebraic generalization of the
quantum mechanical formalism. Ann. Math. 36, 29-64.
Kadison, R.V. [1958] Theory of operators, part II. Operator algebras. Bull. Amer. Math.
Soc. 64, 61-85.
502 References

Kadison, R.V. [1965] Transfonnations of states in operator theory and dynamics. Topology
3,177-198.
Kadison, R.V. [1982] Operator algebras - the first forty years. In: Kadison, R.V. (ed.)
Operator Algebras and Applications, Proc. Symp. Pure Math. 38(1), pp. 1-18. American
Mathematical Society, Providence.
Kadison, R.V. [1994] Notes on the Gelfand-Neumark theorem. In: Doran, R.S. (ed.) C*-
algebras: 1943-1993. Contemp. Math. 167, pp. 21-53. American Mathematical Society,
Providence.
Kadison, R.V. and J.R. Ringrose [1983] Fundamentals o/the Theory o/Operator Algebras
1. Elementary Theory. Academic Press, New York.
Kadison, R.V. and J.R. Ringrose [1986] Fundamentals o/the Theory o/Operator Algebras
11. Advanced Theory. Academic Press, New York.
Kalmbach, G. [1983] Orthomodular Lattices. Academic Press, London.
Kalmbach, G. [1986] Measures and Hilbert Lattices. World Scientific, Singapore.
Kap1ansky, I. [1953] Modules over operator algebras. Trans. Amer. Math. Soc. 75, 839-858.
Karasev, M. V. [1989] The Maslov quantization conditions in higher cohomology and
analogs of notions developed in Lie theory for canonical fibre bundles of symplectic
manifolds. I, II. Selecta Math. Soviet. 8, (1989),213-234,235-258.
Karasev, M.V. and V.P. Maslov [1993] Nonlinear Poisson Brackets: Geometry and
Quantization. American Mathematical Society, Providence.
Karshon, Y. and E. Lennan [1997] The centralizer of invariant functions and division
properties of the moment map. Ill. J. Math. 41, 462-487.
Kasparov, G.G. [1980] Hilbert CO-modules: theorems of Stinespring and Voiculescu. J.
Operator Theory 4, 133-150.
Kasparov, G.G. [1981] The operator K -functor and extensions of C* -algebras. Math. USSR
fzv. 16, 513-572.
Kastler, D. and M. Mebkhout [1990] Revisiting the Mackey-Stone-von Neumann theorem.
The C' -algebra of a presymplectic space. Nucl. Phys. Proc. Suppl. B18, 200-211.
Kazhdan, D., B. Kostant, and S. Sternberg [1978] Hamiltonian group actions and dynamical
systems of Calogero type. Commun. Pure Appl. Math. 31, 481-507.
Kelley, J.L. [1955] General Topology. Van Nostrand, London.
Kerner, R. [1968] Generalization of the Kaluza-Klein theory for an arbitrary non abelian
gauge group. Ann.lnst. H. Poincare 9, 143-152.
Kijowski, J. and W. Tulczyjew [1979] A Symplectic Framework/or Field Theories. Lecture
Notes in Physics 107. Springer, Berlin.
Kirchberg, E. and S. Wassennann [1995] Operations on continuous bundles of C' -algebras.
Math. Ann. 303, 677-697.
Kirillov, A.A. [1962] Unitary representations of nilpotent Lie groups. Russ. Math. Surv. 17,
53-104.
Kirillov, A.A. [1976] Local Lie algebras. Russ. Math. Surv. 31,55-75.
Kirillov, A.A. [1990] Geometric Quantization. In: V.I. Arnold and S.P. Novikov (eds.)
Dynamical Systems N, pp. 137-172. Springer, Berlin.
Kirwan, F. [1984] Convexity properties of the moment mapping. I1I.lnv. Math. 77,547-552.
 References 503

Klauder, J.R. [1963] Continuous representation theory. II. Generalized relation between
quantum and classical dynamics. J. Math. Phys. 4, 1058-1073.
Klauder, J.R. [1970] Exponential Hilbert space: Fock space revisited. J. Math. Phys. 11,
609-630.
Klauder, J.R. [1988] Quantization is geometry, after all. Ann. Phys. (N.Y). 188, 120-141.
Klauder, J.R. [1995] Quantization without quantization. Ann. Phys. (N.Y). 237,147-160.
Klauder, J .R. [1997] Coherent state quantization of constrained systems. Ann. Phys. (N.Y.)
254,419-453.
Klauder, J.R. and B.-S. Skagerstam [1985] (eds.) Coherent States. World Scientific,
Singapore.
Kleppner, A. and R.L. Lipsman [1972] The Plancherel formula for group extensions. I. Ann.
Scient. Ec. Norm. Sup. 5, 459-516.
Kleppner, A. and R.L. Lipsman [1973] The Plancherel formula for group extensions. II.
Ann. Scient. Ec. Norm. Sup. 6, 103-132.
Klimek, S. and A. Lesniewski [1992a] Quantum Riemann surfaces, I. The unit disc.
Commun. Math. Phys. 146, 103-122.
Klimek, S. and A. Lesniewski [1992b] Quantum Riemann surfaces, II. The discrete series.
Lett. Math. Phys. 24, 125-139.
Klimek, S. and A. Lesniewski [1994] Quantum Riemann surfaces, III. The exceptional case.
Lett. Math. Phys. 32,45-61.
Klimek, S. and A. Lesniewski [1996] Quantum Riemann surfaces for arbitrary Planck's
constant. J. Math. Phys. 37,2157-2165.
Klingenberg, W. [1982] Riemannian Geometry. de Gruyter, Berlin.
Knapp, A. W. [1986] Representation Theory of Semisimple groups. An Overview Based on
Examples. Princeton University Press, Princeton.
Knauf, A. [1989] Coulombic periodic potentials: the quantum case. Ann. Phys. (N.Y.) 191,
205-240.
Kobayashi, S. and K. Nomizu [1963] Foundations of Differential Geometry. Vol. I. Wiley,
New York.
Kobayashi, S. and K. Nomizu [1969] Foundations ofDifferential Geometry. Vol. II. Wiley,
New York.
Kohn, J. and L. Nirenberg [1965] An algebra of pseudo-differential operators. Comm. Pure
Appl. Math. 18,269-305.
Kolmogorov, A. [1932] Zur Begriindung der projektiven Geometrie. Ann. Math. 33, 175-
176.
Kondracki, W. and P. Sadowski [1986] Geometric structure on the orbit space of gauge
connections. J. Geom. Phys. 3, 421-434.
Konop1eva, N.P. and V.N. Popov [1981] Gauge Fields. Chur, Switzerland.
Kontsevich, M. [1998] Deformation quantization of Poisson manifolds, I. e-print
q-alg/9709040.
Koomwinder, T.H. and N.M. Muller [1997] The quantum double of a (locally) compact
group. J. Lie Theory 7,101-120.
Kostant, B. [1970] Quantization and unitary representations. Lecture Notes in Mathematics
170, 87-208.
504 References

Kostant, B. [1973] On convexity, the Weyl group and the Iwasawa decomposition. Ann.
scient. Ec. Norm. Sup. 6, 413-455.
Kraus, K. [1977] Position observables of the photon. In: Price, W.C. and S.S. Chissick
(eds.) The Uncertainty Principle and Foundations o/Quantum Mechanics, pp. 293-320.
Wiley, New York.
Krishnaprasad, P.S. and J.E. Marsden [1987] Hamiltonian structure and stability for rigid
bodies with flexible attachments. Arch. Rat. Mech. An. 98, 137-158.
Kurchan, J., P. Leboeuf, and M. Saraceno [1989] Semiclassical approximation in the
coherent-state representation. Phys. Rev. A40, 6800-6814.
Kummer, H. [1991] The foundation of quantum theory and noncommutative spectral theory.
I, II. Found. Phys. 21,1021-1069,1183-1236.
Kummer, M. [1981] On the construction of the reduced phase space of a Hamiltonian system
with symmetry. Indiana Univ. Math. J. 30,281-291.
Kumpera, A. and D.C. Spencer [1972] Lie Equations. Vol. 1. Princeton University Press,
Princeton.
Kiinzle, H.P. [1972] Canonical dynamics of spinning particles in gravitational and
electromagnetic fields. J. Math. Phys. 13,739-744.
Kuo, H.H. [1975] Gaussian Measures in Banach Spaces. Lecture Notes in Mathematics
463. Springer, Berlin.
Kuwabara, R. [1982J On spectra of the Laplacian on vector bundles. J. Math. Tokushima
Univ. 16, 1-23.
Kuwabara, R. [19841 Spectrum and holonomy of the line bundle over the sphere. Math. Z.
187,481-490.
Laidlaw, M.G.G. and C.M. DeWitt [1970] Feynman functional integrals for systems of
indistinguishable particles. Phys. Rev. D3, 1375-1378.
Lahti, PJ. and S. Bugajski [1980] Fundamental principles of quantum theory. Int. J. Theor.
Phys. 19,499-514.
Lahti, P.J. and S. Bugajski [1985] Fundamental principles of quantum theory 2. From a
convexity scheme to the DHB theory. Int. J. Theor. Phys. 24, 1051-1080.
Lance, E.C. [1995] Hilbert CO-Modules. A Toolkit/or Operator Algebraists. LMS Lecture
Notes 210. Cambridge University Press, Cambridge.
Landau, L.J. [1996] Macroscopic observation of a quantum particle in a slowly varying
potential- on the classical limit of quantum-theory. Ann. Phys. (N.Y.) 246, 190--227.
Landi, G. [1997] An Introduction to Noncommutative Spaces and their Geometries.
Springer, Berlin.
Landsman, N.P. [1990a] Quantization and superselection sectors I. Transformation group
CO-algebras. Rev. Math. Phys. 2,45-72.
Landsman, N.P. [1990b] Quantization and superselection sectors II. Dirac Monopole and
Aharonov-Bohm effect. Rev. Math. Phys. 2,73-104.
Landsman, N.P. [1991] Algebraic theory of superselection sectors and the measurement
problem in quantum mechanics. Int. J. Mod. Phys. A6, 5349-5372.
Landsman, N.P. [1992] Induced representations, gauge fields, and quantization on
homogeneous spaces. Rev. Math. Phys. 4, 503-528.
 References 505

Landsman, N.P. [1993a] Defonnations of algebras of observables and the classical limit of
quantum mechanics. Rev. Math. Phys. 5, 775-806.
Landsman, N.P. [1993b] Strict defonnation quantization of a particle in external
gravitational and Yang-Mills fields. J. Geom. Phys. 12,93-132.
Landsman, N.P. [1993c] Quantization and classicization: from Jordan-Lie algebras of
observables to gauge fields. Class. Quantum Grav. 10, S 10 I-S 108.
Landsman, N .P. [1994] The infinite unitary group, Howe dual pairs, and the quantization
of constrained systems. e-print hep-thl9411 171.
Landsman, N.P. [1995a] Rieffel induction as generalized quantum Marsden-Weinstein
reduction.J. Geom. Phys. 15,285-319; Err. ibid. 17 (1995) 298.
Landsman, N .P. [1995b] Observation and superselection in quantum mechanics. Stud. Hist.
Phil. Mod. Phys. 26, 45-73.
Landsman, N.P. [1995c] Against the Wheeler-DeWitt equation. Class. Quantum Grav. 12,
LlI9-Ll23.
Landsman, N .P. [1996a] Classical and quantum representation theory. In: de Kerf, E. A. and
H.G.J. Pijls (eds.) Proceedings Seminar Mathematical Structures in,Field Theory.CWI-
syllabus 39, pp. 135-163. Mathematisch Centrum, CWI, Amsterdam.
Landsman, N .P. [1996b] Classical behaviour in quantum mechanics: a transition probability
approach. Int. J. Mod. Phys. B, 1545-1554.
Landsman, N.P. [1997] Poisson spaces with a transition probability. Rev. Math. Phys. 9,
29-57.
Landsman, N.P. [1998a] The quantization of constrained systems: from symplectic re-
duction to Rieffel induction. In: Strasburger, A., S.T. Ali, J.-P. Antoine, J.-P. Gazeau,
and A. Odzijewicz (eds.) Quantization, Coherent States and Poisson Structures. Proc.
X1Vth Workshop on Geometric Methods in Physics, Bia/owieia, 1995, pp. 79-95. Polish
Scientific Publishers, Warsaw.
Landsman, N.P. [1998b] Simple new axioms for quantum mechanics. Int. J. Theor. Phys.
37, 343-348.
Landsman, N.P. [1998c] Strict quantization of coadjoint orbits. J. Math. Phys., to appear.
Landsman, N.P. [1998d] 1Wisted Lie group C' -algebras as strict quantizations. Lett. Math.
Phys., to appear.
Landsman, N.P. and N. Linden [1991] The geometry of inequivalent quantizations. Nucl.
Phys. B365, 121-160.
Landsman, N.P. and N. Linden [1992] Superselection rules from Dirac and BRST
quantization of constrained systems. Nucl. Phys. B371, 415-433.
Landsman, N.P. and U.A. Wiedemann [1995] Massless particles electromagnetism, and
Rieffel induction. Rev. Math. Phys. 7, 923-958.
Landsman, N.P. and K.K. Wren [1997] Constrained quantization and O-angles. Nucl. Phys,
B502 [PM], 537-560.
Landstad, M.B. [1994] Quantizations arising from abelian subgroups. Int. J. Math. S, 897-
936.
Landstad, M. B. and I. Raeburn [1997] Equivariant defonnations of homogeneous spaces.
J. Funct. Anal. 148,480-507.
Lang, S. [1995] Differential and Riemannian Manifolds, 3d ed. Springer, New York.
506 References

Langlands, R.P. [1987] The Dirac monopole and induced representations. Pac. J. Math.
126,145-151.
Langmann, E. and G.W. Semenoff [1993] Gribov ambiguity and non-trivial vacuum
structure of gauge theories on a cylinder. Phys. Lett. B303, 303-307.
Laquer, H.T. [1984] Stability properties of the Yang-Mills functional near the canonical
connection. Michigan Math. J. 31, 139-159.
Lee, H.-W. [1995] Theory and applications of the quantum phase space distribution
functions. Phys. Rep. 259, 147-211.
Lee, R.-Y. [1976] On the C' -algebras of operator fields. Indiana Univ. Math. J. 25, 303-314.
Lee, T. and P. Oh [1994] Non-abelian Chern-Simons quantum mechanics and non-abelian
Aharonov-Bohm effect. Ann. Phys. (N.Y.) 235, 413-434.
Leinfelder, H. and C. Simader [1981] SchrMinger operators with singular magnetic vector
potentials. Math. Z. 176, 1-19.
Leptin, H. and J. Ludwig [1994] Unitary Representation Theory ofExponential Lie Groups.
De Gruyter, Berlin.
Leonrad, N .E. and I.E. Marsden [1997] Stability and drift of underwater vehicle dynamics:
mechanical systems with rigid motion symmetry. Physica DI05, 130--162.
Lerman, E., R. Montgomery, and R. Sjamaar [1993] Examples of singular reduction. In:
Salomon, D. (ed.) Symplectic Geometry, LMS Lecture Notes Series 192, pp. 127-155.
Cambridge University Press, Cambridge.
Libermann, P. [1983] Problemes d 'equivalence et geometrie symplectique. Asterisque 107-
108,43-68.
Libermann, P. and C.-M. MarIe [1987] Symplectic Geometry and Analytical Mechanics.
Reidel, Dordrecht.
Lichnerowicz, A. [1977] Les varietes de Poisson et leurs algebres de Lie associees. J. Diff.
Geom. 12, 253-300.
Lie, S. [1890] Theorie der Transformationsgruppen. S. Teubner, Leipzig.
Lieb, E.H. [1973] The classical limit of quantum spin systems. Commun. Math. Phys. 62,
327-340.
Linden, N., A.J. MacFarlane, and I.W. van Holten [1996] Particle motion in a Yang-Mills
field - Wong's equations and spin 112 analogs. Czech. J. Phys. 46, 209-215.
Littlejohn, R.G. [1986] The semiclassical evolution of wave packets. Phys. Rep. 138, 193-
291.
Littlejohn, R.G. [1992] The Van Vleck formula, Maslov theory, and phase space geometry.
J. Stat. Phys. 68, 7-50.
Liu, Z.-I. and M. Qian [1992] Gauge invariant quantization on Riemannian manifolds.
Trans. Amer. Math. Soc. 331,321-333.
Loffelholz, J., G. Morchio, and F. Strocchi [1996] A quantum mechanical gauge model and
a possible dynamical solution of the strong C P-problem.Ann. Phys. (N.Y.) 250, 367-388.
Lowenstein, I.H. and I.A. Swieca [1971] Quantum electrodynamics in two dimensions.
Ann. Phys. (N.Y.) 68, 172-195.
Ludwig, G. [1985] An Axiomatic Basis for Quantum Mechanics. Volume 1: Derivation of
Hilbert Space Structure. Springer, Berlin.
 References 507

Mackenzie, K. [1987a] Lie Groupoids and Lie Algebroids in Differential Geometry.

Cambridge University Press, Cambridge.
Mackenzie, K. [1987b] A note on Lie algebroids which arise from groupoid actions. Cah.
Top. Geom. Diff. Cat. 28, 283-302.
Mackenzie, K. [1989] Classification of principal bundles and Lie groupoids with prescribed
gauge group bundle. J. Pure Appl. Alg. 58,181-208.
Mackey, G.W. [1958] Unitary representations of group extensions. I. Acta Math. 99, 265-
311.
Mackey, G.w. [1963] The Mathematical Foundations of Quantum Mechanics. New York,
Benjamin.
Mackey, G.W. [1968] Induced Representations. Benjamin, New York.
Mackey, G.W. [1978] Unitary Group Representations in Physics Probability and Number
Theory. Benjamin, New York.
Mackey, G.W. [1992] The Scope and History of Commutative and Noncommutative
Harmonic Analysis. American Mathematical Society, Providence.
Mackey, G.W. [1998] The relationship between classical mechanics and quantum
mechanics. Contemp. Math. 214,91-110.
Maeda, F. [1958] Kontinuerliche Geometrien. Springer, Berlin.
Maeda, F. and S. Maeda [1970] Theory of Symmetric Lattices. Springer, Berlin.
Majid, S. [1988] Hopf algebras for physics at the Planck scale. Class. Quantum Grav. 5,
1587-1606.
Majid, S. [1990] Physics for algebraists: non-commutative and non-cocommutative Hopf
algebras by a bicrossproduct construction. J. Algebra 130, 17-64.
Malliavin, P. [1997] Stochastic Analysis. Springer, Berlin.
Malliavin, M.-P. and P. Malliavin [1990]. Integration on loop groups. I. Quasi invariant
measures. J. Funct. Anal. 93, 207-237.
Manchon, D. [1993] Weyl symbolic calculus on any Lie group. Acta Appl. Math. 30, 159-
186.
Manton, N. S. [1985] The Schwinger model and its axial anomaly. Ann. Phys. (N.Y.) 159,
220-251.
Marinov, M.S. [1995] Path integrals on homogeneous manifolds.J. Math. Phys. 36, 2458-
2469.
Marolf, D. [1997] Refined algebraic quantization: systems with a single constraint. Banach
Center Publ. 39, 331-344.
Marsden, J.E. [1974] Applications of Global Analysis in Mathematical Physics. Publish or
Perish, Boston.
Marsden, J.E. [1981] Lectures on Geometric Methods in Mathematical Physics. SIAM,
Philadelphia.
Marsden, J.E. [1992] Lectures on Mechanics. Cambridge University Press, Cambridge.
Marsden, J.E. [1993] Steve Smale and geometric mechanics. In: Hirsch, M.W., J.E. Marsden,
and M. Shub (eds.) From Topology to Computation: Proceedings 0/ the Smale/est, pp.
499-516. Springer, New York.
Marsden, J.E., R. Montgomery, and T. Ratiu [1990] Reduction, symmetry, and phases in
mechanics. Mem. Amer. Math. Soc. 436, 1-110.
508 References

Marsden, J .E. and T.S. Ratiu [1986] Reduction of Poisson manifolds. Lett. Math. Phys. 11,
161-170.
Marsden, J.E. and T.S. Ratiu [1994] Introduction to Mechanics and Symmetry. Springer,
New York.
Marsden, J.E., T. Ratiu, and G. Raugel [1991] Symplectic connections and the linearization
of Hamiltonian systems. Proc. Royal Soc. Edinburgh 177A, 329-380.
Marsden, J.E., T. Ratiu, and A. Weinstein [1984a] Semidirect products and reduction in
mechanics. Trans. Amer. Math. Soc. 281, 147-177.
Marsden, J.E., T. Ratiu, and A. Weinstein [1984b] Reduction and Hamiltonian structures
on duals of semidirect product Lie algebras. Con temp. Math. 28, 55-100.
Martinez Alonso, L. Group-theoretical foundations of classical and quantum mechanics. II.
Elementary systems. J. Math. Phys. 20,219-230.
Maslov, V.P. [1994] The Complex WKB Methodfor Nonlinear Equations. Birkhauser, Basel.
Maslov, V.P. and M.V. Fedoriuk [1981] Semi-Classical Approximation in Quantum
Mechanics. Reidel, Dordrecht.
Matsumoto, K. [199Ia] Noncommutative three-dimensional spheres. Japan. J. Math. (N.S.)
17, 333-356.
Matsumoto, K. [199Ib] Noncommutative three-dimensional spheres. II. Noncommutative
Hopf flbering. Yokohama Math. J. 38, 103-111.
Matsumoto, K. and J. Tomiyama [1992] Noncommutative lens spaces. J. Math. Soc. Japan
44,13-41.
Mehra, J. and H. Rechenberg [1982] The Historical Development of Quantum Theory. Vol.
1. The quantum theory ofPlanck, Einstein, Bohr, and Sommerfeld: its foundation and the
rise of its difficulties, 1900-1925. Springer, New York.
Meinrenken, E. [1994] Coherent states and classical limits. J. Phys. A27, 3257-3265.
Meinrenken, E. and R. Sjamaar [1998] Singular reduction and quantization. e-print dg-
ga19707023.
Meschkowski, H. [1962] Hilbertsche Riiume mit Kernfunktion. Springer, Berlin.
Meyer, K. [1973] Symmetries and integrals in mechanics. In: Peixoto, M.M. (ed.) Dynamical
systems, pp. 259-272. Academic Press, New York.
Michor, P.W. [1990] The moment mapping for unitary representations. Ann. Global Anal.
Geom. 8, 299-313.
Mielnik, B. [1968] Geometry of quantum states. Commun. Math. Phys. 9, 55-80.
Mielnik, B. [1969] Theory of filters. Commun. Math. Phys. 15, 1-46.
Mielnik, B. [1974] Generalized quantum mechanics. Commun. Math. Phys. 37, 221-256.
Mikami, K. and A. Weinstein [1988] Moments and reduction for symplectic groupoids.
Publ. RIMS Kyoto Univ. 24,121-140.
Milnor, J. [1976] Curvatures ofleft invariant metrics on Lie groups. Adv. Math. 21, 293-329.
Mitter, P.K. and C.M. Viallet [1981] On the bundle of connections and the gauge orbit
manifold in Yang-Mills theory. Commun. Math. Phys. 79,457-472.
Molzahn, F.H., T.A. Osborn, and S .A. Fulling [1990] Gauge invariant asymptotic expansions
of SchrOdinger propagators on manifolds. Ann. Phys. (N.Y.) 204, 64-112.
 References 509

Molzahn, F.H., T.A. Osborn, and S.A. Fulling [1992] Multi-scale semiclassical ap-
proximations for Schrodinger propagators on manifolds. Ann. Phys. (N.Y.) 214,
102-141.
Montgomery, R. [1984] Canonical formulation of a classical particle in a Yang-Mills field
and Wongs's equations. Lett. Math. Phys. 8,59--67.
Montgomery, R., J .E. Marsden, and T. Ratiu [1984] Gauged Lie-Poisson structures.
Con temp. Math. 28,101-114. American Mathematical Society, Providence.
Monthubert, B. and F. Pierrot [1997] Indice analytique et groupo"ides de Lie. C. R. Acad.
Sci. Paris Ser. I Math. 325, 193-198.
Moschella, U. [1989] Classical limit of a quantum particle in an external Yang-Mills field.
Ann. Inst. H. Poincare A51, 351-370.
Moscovici, H. [1969] Generalized induced representations. Rev. Roum. Math. Pures et Appl.
14,1539-1551.
Moyal, J.E. [1949] Quantum mechanics as a statistical theory. Proc. Camb. Phil. Soc. 45,
99-124.
Muhly, P.S., J.N. Renault, and D.P. Williams [1987] Equivalence and isomorphism for
groupoid C'-algebras. J. Operator Th. 17,3-22.
Muh1y, P.S., and D.P. Williams [1990] Continuous trace groupoid C' -algebras. Math. Scand.
66,231-241.
Muhly, P.S., and D.P. Williams [1992] Continuous trace groupoid C'-algebras. 2. Math.
Scand. 70, 127-145.
Muhly, P.S., and D.P. Williams [1995] Groupoid cohomology and the Dixmier-Douady
class. Proc. London Math. Soc. 71, 109-134.
Nachtrnann, O. [1968] Dynamische Stabilitat im de-Sitter-Raum. Sitz. Ber. Ost. Akad. Wiss.
[[176,363-379.
Nagasawa, M. [1993] Schrodinger Equations and Diffusion Theory. Birkhiiuser, Basel.
Nagy, G. [1992] A Framework for Deformation Quantization. Ph.D. thesis, University of
California at Berkeley.
Nagy, G. [1993] On the Haarmeasure of the quantum SU(N) group. Commun. Math. Phys.
153,217-228.
Nagy. G [1996] E-theory with *-homomorphisms. J. Funct. Anal. 140,275-299.
Nagy, G. [1997] Deformation quantization and K -theory. Contemp. Math. 214, 111-134.
Nagy, G. [1998a] A deformation quantization procedure for C*-algebras. J. Operator Th.,
to appear.
Nagy, G. [1998b] A rigidity property for quantum SU(3) groups. In: Brylinski, J.-L., R.
Brylinski, N. Handzy, and B. Tsygan (eds.) Advances in Geometry and Mathematical
Phyisics, Vol. I. Birkhauser, Basel.
Nauenberg, M., C. Stroud, and J. Yeazell [1994] The classical limit of an atom. Sci. Amer.
270(6), 44-49.
Neeb, K.-H. [1995] On the convexity of the moment mapping for unitary highest weight
representations. J. Funct. Anal. 127,301-325.
Neeb, K.-H. [I 996a] Coherent states, holomorphic extensions, and highest weight
representations. Pac. J. Math. 174, 497-542.
510 References

Neeb, K.-H. [1996b] A note on central extensions of Lie groups. J. Lie Theory 6,207-213.
Neumann, H. [1972] Transfonnation properties of observables. Helv. Phys. Acta 25, 811-
819.
Neumann, J. von [1931] Die Eindeutigkeit der Schrooingerschen Operatoren. Math. Ann.
104, 570-578.
Neumann, I. von [1932] Mathematische Grundlagen der Quantenmechanik. Springer,
Heidelberg.
Neumann, I. von [1936] On an algebraic generalization of the quantum mechanical
fonnalism (part I). Math. Sb. 1,415-484.
Neumann, I. von [1981] Continuous geometries with a transition probability. Mem. Amer.
Math. Soc. 252, 1-210 (edited by I.S. Halperin; MS from 1937).
Newton, T.D. and E.P. Wigner [1949] Localized states for elementary systems. Rev. Mod.
Phys. 21, 400-406.
Nistor, v., A. Weinstein, and P. Xu [1997] Pseudodifferential operators on differential
groupoids. e-print junct-anI970200.
Nussbaum, A.E. [1964] Reduction theory for unbounded closed operators in Hilbert space.
Duke Math. J. 31, 33-44.
Odzijewicz, A. [1988] On reproducing kernels and quantization of states. Commun. Math.
Phys. 114, 577-579.
Odzijewicz, A. [1992] Coherent states and geometric quantization. Commun. Math. Phys.
150,385-413.
Oh, P. [1996] Classical and quantum mechanics of non-abelian Chern-Simons particles.
Nucl. Phys. 462, 551-570.
Omnes, R. [1994] The Interpretation o/Quantum Mechanics. Princeton University Press,
Princeton.
Omnes, R. [1997a] Quantum-classical correspondence using projection operators. J. Math.
Phys.38,697-707.
Omnes, R. [1997b] Localization of relativistic particles. J. Math. Phys. 38, 708-715.
Omori, H. [1997] Infinite-Dimensional Lie Groups. American Mathematical Society,
Providence.
Onofri, E. [1975] A note on coherent state representations of Lie groups. J. Math. Phys. 16,
1087-1089.
0rsted, B. [1979] Induced representations and a new proof of the imprimitivity theorem. J.
Funct. Anal. 31, 355-359.
Osborn, T.A. and EH. Molzahn [1995]. Moyal quantum mechanics: the semiclassical
Heisenberg dynamics. Ann. Phys. (N.Y.) 241, 79-127.
Otto, M. [1987] A reduction scheme for phase spaces with almost Kahler symmetry.
Regularity results for momentum level sets. J. Geom. Phys. 4, 101-118.
Packer, I.A. [1994] Transfonnation group C' -algebras: A selective survey. In: Doran [1994],
pp. 183-217.
Packer, I.A. [1996] Crossed product C'-algebras and algebraic topology. Rev. Math. Phys.
8,623-637.
Packer, I.A. and I. Raeburn [1989] Twisted crossed products of C* -algebras. Math. Proc.
Cam. Phil. Soc. 106, 293-311.
 References 511

Packer, J .A. and I. Raeburn [1990] 1Wisted crossed products of C' -algebras. 2. Math. Ann.
287,595-612.
Packer, J.A. and I. Raeburn [1992] On the structure of twisted group C'-algebras. Trans.
Amer. Math. Soc. 334, 685-718.
Palais, R.S. [1957] A global fonnulation of the Lie theory of transfonnation groups. Mem.
Amer. Math. Soc. 22, 1-123.
Palmer, T.W. [1994] BanachAlgebras and the General Theory oj -Algebras. Vol.f: Algebras
and Banach Algebras. Cambridge University Press, Cambridge.
Paschke, W.L. [1973] Inner product modules over B'-algebras. Trans. Amer. Math. Soc.
182,443-468.
Paterson, A.L [1988] Amenability. American Mathematical Society, Providence.
Paul, T. and A. Uribe [1995] The semi-classical trace fonnula and propagation of wave
packets. J. Funct. Anal. 132 (1995), 192-249.
Paul, T. and A. Uribe [1996] On the pointwise behavior of semi-classical measures.
Commun. Math. Phys. 175,229-258.
Pauli, W. [1973] Ausgewiihlte Kapitel aus der Feldquantisierung. E.T.H. ZUrich Lecture
Notes 1950-51. In: Enz, c.P. (ed.) Pauli Lectures in Physics, Vol. 6: Selected Topics in
Field Quantization. MIT Press, Cambridge (MA).
Paulsen, V.I. [1986] Completely Bounded Maps and Dilations. Longman, Harlow.
Pedersen, G.K. [1979] C'-Algebras and their Automorphism Groups. Academic Press,
London.
Pedersen, G.K. [1989] Analysis Now. Springer, New York.
Petz, D. [1994] Geometry of canonical correlation on the state space of a quantum system.
J. Math. Phys. 35,780-795.
Peetre, J. [1990] The Berezin transfonn and Haplitz operators. J. Operator Th. 24, 165-186.
Perelomov, A.M. [1972] Coherent states for arbitrary Lie groups. Commun. Math. Phys.
26, 222-236.
Perelomov, A. [1986] Generalized Coherent States and their Applications. Springer, Berlin.
Pflaum, MJ. [1995] Local Analysis of Deformation Quantization. Ph.D. thesis, Ludwig-
Maximilians-Universitat MUnchen.
Picard, R. [1989] Hilbert Space Approach to some Classical Transforms. Longman, Harlow.
Piron, C. [1976] Foundations oJQuantum Physics. Benjamin, Reading (Mass.).
Piziak, R. [1991] Orthomodular lattices and quadratic spaces: a survey. Rocky Mount. J.
Math. 21,951-992.
Plymen, R. [1968]. CO-algebras and Mackey's axioms. Commun. Math. Phys. 8, 132-146.
Poerschke, T., G. Stolz, and J. Weidmann [1989]. Expansions in generalized eigenfunctions
of self-adjoint operators. Math. Z. 202,397-408.
Poerschke, T. and G. Stolz [1993] On eigenfunction expansions and scattering theory. Math.
Z. 212, 397-357.
Pontrjagin, L. [1946] Topological Groups. Princeton University Press, Princeton.
Pool, J.C.T. [1966] Mathematical aspects of the Weyl correspondence. J. Math. Phys. 7,
66-76.
512 References

Pool, J.C.T. [1968] Semimodularity and the logic of quantum mechanics. Commun. Math.
Phys. 9, 218-228.
Poulsen, N .S. [1970] Regularity Aspects of the Theory of Infinite-Dimensional Representa-
tions of Lie Groups. Ph.D. thesis, MIT.
Pradines, J. [1966] Theorie de Lie pour les groupoldes differentiables. Relations entre
proprietes locales et globales. C. R. Acad. Sc. Paris A263, 907-910.
Pradines, J. [1968] Geometrie differentielle au-dessus d 'un groupolde. C. R. Acad. Sc. Paris
A266, 1194-1196.
Pressley, A. and G. Segal [1986] Loop Groups. Oxford University Press, Oxford.
Prosser, R.T. [1963] On the ideal structure of operator algebras. Mem. Amer. Math. Soc. 45.
Pulmannova, S. [1986] Transition probability spaces.J. Math. Phys. 27,1791-1795.
Pulmannova, S. [1989] Mielnik and Cantoni transition probabilities. Int. J. Theor. Phys. 28,
711-718.
Radulescu, F. [1998] The r -equivariant form of the Berezin quantization of the upper half
plane. Mem. Amer. Math. Soc. 630.
Raeburn, I. [1988] Induced C' -algebras and a symmetric imprimitivity theorem. Math. Ann.
280,369-387.
Raggio, G.A. [1982] Comparison of Uhlmann's transition probabilitywith one induced by
the natural cone of von Neumann algebras in standard form. Lett. Math. Phys. 6, 233-236.
Ramazan, B. [1998] Deformation Quantization of Lie-Poisson Manifolds. Ph.D. thesis,
Universite d'Orleans.
Ramsay, A. [1965] Dimension theory in complete orthocomplemented weakly modular
lattices. Trans. Amer. Math. Soc. 116,9-31.
Ramsay, A. [1971] Virtual groups and group actions. Adv. Math. 6, 253-322.
Rawnsley, J.H. [1975] Representations of a semi-direct product by quantization. Math.
Proc. Camb. Phil. Soc. 78, 345-350.
Rawnsley, J.H. [1977] Coherent states and Kahler manifolds. Quart. J. Math. Oxford (2)
28,403-415.
Rajeev, S.G. [1988] Yang-Mills theory on a cylinder. Phys. Lett. B212, 203-205.
Rajeev, S.G. and L. Rossi [1995] Some rigorous results for Yang-Mills theory on a cylinder.
J. Math. Phys. 36, 3308-3319.
Reed, M. and B. Simon [1972] Methods of Modern Mathematical Physics. I: Functional
Analysis. Academic Press, New York.
Reed, M. and B. Simon [1975] Methods of Modern Mathematical Physics. II: Fourier
Analysis. Self-adjointness. Academic Press, New York.
Reed, M. and B. Simon [1978] Methods of Modern Mathematical Physics. N: Analysis of
Operators. Academic Press, New York.
Renault, J. [1980] A Groupoid Approach to CO-algebras. Lecture Notes in Mathematics
793. Springer, Berlin.
Renault, J. [1987] Representation des produits croises d' algebres de groupoi"des. J. Operator
Th. 18,67-97.
Rezende, J. [1996] Stationary phase, quantum mechanics and semi-classical limit. Rev.
Math. Phys. 8, 1161-1185.
 References 513

Rieffel, M.A. [1972] On the uniqueness of the Heisenberg commutation relations. Duke
Math. J. 39, 745-753.
Rieffel, M.A. [1974a] Induced representations of C* -algebras. Adv. Math. 13, 176-257.
Rieffel, M.A. [1974b] Morita equivalence for C*-algebras and W*-algebras. J. Pure Appl.
Alg. 5, 51-96.
Rieffel, M.A. [1979] Unitary representations of group extensions: an algebraic approach to
the theory of Mackey and Blattner. Adv. Math. Suppl. Stud. 4, 43-82.
Rieffel, M.A. [1988] Projective modules over higher-dimensional noncommutative tori.
Canad. J. Math. 40, 257-338.
Rieffel, M.A. [1989a] Deformation quantization of Heisenberg manifolds. Commun. Math.
Phys. 122, 531-562.
Rieffel, M.A. [1989b] Continuous fields of C* -algebras coming from group cocycles and
actions. Math. Ann. 283,631-643.
Rieffel, M.A. [1990a] Lie group convolution algebras as deformation quantizations of linear
Poisson structures. Am. J. Math. 112,657-686.
Rieffel, M.A. [1990b] Proper actions of groups on C*-algebras. In: Araki, H. and R.Y.
Kadison (eds.) Mappings oj Operator Algebras, pp. 141-182. Birkhauser, Boston.
Rieffel, M.A. [1993a] Deformation quantization for actions of JR.d. Mem. Amer. Math. Soc.
506.
Rieffel, M.A. [1993b] Compact quantum groups associated with toral subgroups. Contemp.
Math. 145, 465-49t.
Rieffel, M.A. [1993c] K -groups of C* -algebras deformed by actions of JR.d. J. Funct. Anal.
116, 199-214.
Rieffel, M.A. [1994] Quantization and C*-algebras. In: Doran, R.S. (ed.) CO-algebras:
1943-1993. Contemp. Math. 167, pp. 67-97. American Mathematical Society,
Providence.
Rieffel, M.A. [1995] Non-compact quantum groups associated with abelian subgroups.
Comm. Math. Phys. 171, 181-20t.
Rieffel, M.A. [1996] The classical limit of dynamics for spaces quantized by an action of
JR.". Can. J. Math. 49,160-174.
Rieffel, M.A. [1998] Quantization and operator algebras. In: Bracken, A.1., D. De Wit,
M. Gould, and P. Pearce, (eds.) Meeting with the Platypus. Proc. Xllth Int. Congress oj
Mathematical Physics, Brisbane 1997. International Press, Singapore.
Riesz, F. and B. Sz.-Nagy [1990] Functional Analysis. Dover, New York.
Robert, D. [1987] Autour de I'Approximation Semi-Classique. Birkhauser, Basel.
Robert, D. [1992] (ed.) Methodes Semi-Classiques. Asterisque 207, 1-212, 210, 1-384.
Robert, D. [1998] Semi-classical approximation in quantum mechanics. A survey of old
and recent Mathematical results. He/v. Phys. Acta 71, 44-116.
Roberts, J.E. and G. Roepstorff [1969] Some basic concepts of algebraic quantum theory.
Commun. Math. Phys. 11,321-338.
Robinson, S.L. [1988a] The semiclassical limit of quantum mechanics. I. Time evolution.
J. Math. Phys. 29, 412-419.
Robinson, S.L. [1988b] The semiclassical limit of quantum mechanics. II. Scattering theory.
Ann. Inst.ll. Poincare A48, 281-296.
514 References

Robinson, S.L. [1993] Semiclassical mechanics for time-dependent Wigner functions. 1.

Math. Phys. 34, 2185-2205.
Robson, M.A. [1994] The Geometric Quantization of Constrained Systems. Ph.D. thesis,
University of Cambridge.
Robson, M.A. [1996] Geometric quantization of reduced cotangent bundles. J. Geom. Phys.
19 (1996) 207-245.
Rodriguez Palacios, A. [1988] Jordan axioms for C* -algebras. Manuscripta Math. 61, 297-
314.
Rosenberg, J. [1994] C* -algebras and Mackey's theory of group representations. In: Doran
[1994], pp. 151-181.
Rosenberg, J. [1996] Behavior of K -theory under quantization. In: Doplicher, S., R. Longo,
J.E. Roberts, and L. Szido (eds.) Operator Algebras and Quantum Field Theory, pp.
404-415. International Press, Cambridge (MA).
Rosenberg, S. [1997] The Laplacian on Riemannian manifolds. LMS Student Texts 31.
Cambridge University Press, Cambridge.
Royer, A. [1977] Wigner function as the expectation of the parity operator. Phys. Rev. AIS,
449-450.
Ruijsenaars, S.N.M. [1983] The Aharonov-Bohm effect and scattering theory. Ann. Phys.
(N.Y.) 146, 1-34.
Riittimann, G.T. [1985] Facial sets of probability measures. Prob. Math. Stat. 6, 187-215.
Saksena, A., T.A. Osborn, and EH. Molzahn [1991] An asymptotic analysis of quantum
evolution with electromagnetic fields. J. Math. Phys. 32, 938-955.
Schindler, C. [1990] The unique Jordan-Hahn decomposition property. Found. Phys. 20,
561-573.
Schrader, R. and M.E. Taylor [1984] Small Ii asymptotics for quantum partition functions
associated to particles in external Yang-Mills potentials. Commun. Math. Phys. 92, 555-
594.
Schrader, R. and M.E. Taylor [1989) Semi-classical asymptotics, gauge fields and quantum
chaos.J. Funct. Anal. 83, 258-316.
Schrodinger. E. [1926] Der stetige Ubergang von der Mikro- zur Makromechanik.
Naturwiss. 14, 664-666.
Schroeck, EE. [1996] Quantum Mechanics on Phase Space. Kluwer, Dordrecht.
Schulman, L.S. [1968] A path integral for spin. Phys. Rev. 176, 1558-1569.
Schulman, L.S. [1981] Techniques and Applications of Path Integration. Wiley, New York.
Schwartz, L. [1964] Sous-espaces hilbertiens d'espaces vectoriels topologiques et noyaux
associes. (Noyaux reproduisants). J. Anal. Math. 13, 115-256.
Schwartz, L. [1973] Radon Measures on Arbitrary Topological Spaces and Cylindrical
Measures. Bombay, Oxford University Press.
Schwarz, G.W. [1975] Smooth functions invariant under the action of a compact Lie group.
Topology 14, 63-68.
Schwinger, J. [1962] Gauge invariance and mass. II. Phys. Rev. 128,2425-2429.
Schwinger, J. [1970] Quantum Kinematics and Dynamics. W.A. Benjamin, New York.
Scutaru, H. [1977] Coherent states and induced representations. Lett. Math. Phys. 2, 101-
107.
 References 515

Segal, I.E. [1947] Postulates for general quantum mechanics. Ann. Math. 48, 930--948.
Segal, I.E. [1956] Tensor algebras over Hilbert spaces. I. Trans. Amer. Math. Soc. 81,
106-134.
Segal, I.E. [1959] Foundations of the theory of dynamical systems of infinitely many degrees
offreedom I. Mat.-Fys. Medd. Dansk. Vid. Selsk. 31, 1-38.
Segal, I.E. [1963] Transforms for operators and symplectic automorphisms over a locally
compact abelian group. Math. Scand. 13,31-43.
Sengupta, A. [1997a] The moduli space of Yang-Mills connections over a compact surface.
Rev. Math. Phys. 9, 77-121.
Sengupta, A. [1997b] Yang-Mills on surfaces with boundary: Quantum theory and
symplectic limit. Commun. Math. Phys. 183,661-705.
Sengupta, A. [1997c] Gauge theory on compact surfaces. Mem. Amer. Math. Soc. 600.
Shale, D. [1962] Linear symmetries of free Boson fields. Trans. Amer. Math. Soc. 103,
149-167.
Sheu, A.J.-L. [1996] The Weyl quantization of Poisson SU(2). Pac. J. Math. 173,223-240.
Sheu, A.J .-L. [1997] Compact quantum groups and groupoid C' -algebras. J. Funct. Anal.
144 (1997) 371-393.
Shultz, EW. [1979] On normed Jordan algebras which are Banach dual spaces. J. Funct.
Anal. 31, 360--376.
Shultz, EW. [1982] Pure states as dual objects for CO-algebras. Commun. Math. Phys. 82,
497-509.
Shultz, EW. [1981] Dual maps of Jordan homomorphisms and • -homomorphisms between
CO-algebras. Pac. J. Math. 93, 435-441.
Simon, B. [1980] The classical limit of quantum partition functions. Commun. Math. Phys.
71, 247-276.
Sjamaar R. and E. Lerman [1991] Stratified symplectic spaces and reduction. Ann. Math.
134, 375-422.
Skandalis, G. [1991] Kasparov's bivariant K -theory and applications. Exp. Math. 9, 193-
250.
Slawianowski, J. [1991] Geometry ofPhase Spaces. PWN, Warszawa and Wiley, Chichester.
Slawny, J. [1972] On factorrepresentations and the C· -algebra of the canonical commutation
relations. Commun. Math. Phys. 24, 151-170.
Slebarski, S. [1987] The Dirac operator on homogeneous spaces and representations of
reductive Lie groups I. Amer. J. Math. 109, 283-301.
Smale, S. [1970] Topology and Mechanics. I.lnv. Math. 10,305-331.
Sniatycki, J. [1980] Geometric Quantization and Quantum Mechanics. Springer, Berlin.
Sniatycki, J. and A. Weinstein [1983] Reduction and quantization for singular momentum
mappings. Lett. Math. Phys. 7, 155-161.
Soler, M.P. [1995] Characterization of Hilbert spaces with orthomodular spaces. Comm.
Algebra 23, 219-243.
Souriau, J .-M. [1969] Structure des Systemes Dynamiques. Dunod, Paris.
Souriau, J.-M. [1997] Structure of Dynamical Systems: a Symplectic View of Physics.
Birkhiiuser, Basel.
516 References

Sternberg, S. [1977] On minimal coupling and the symplectic mechanics of a classical

particle in the presence of a Yang-Mills field. Proc. Nat. Acad. Sci. 74, 5253-5254.
Stinespring, W. [1955] Positive functions on C'-algebras. Proc. Amer. Math. Soc. 6, 211-
216.
Strathdee, J. [1983] Symmetry aspects of Kaluza-Klein theories. In: Ferrara, S. et al. (eds.)
Supergravity 1982, pp. 170-182. World Scientific, Singapore.
Stratonovic, R.L. [1957] On distributions in representation space. Sov. Phys. JETP 4,891-
898.
Strichartz, R.S. [1983] Analysis of the Laplacian on a complete Riemannian manifold. J.
Funct. Anal. 52, 48-79.
Strocchi, F. [1966] Complex coordinates in quantum mechanics. Rev. Mod. Phys. 38, 36-40.
Sundermeyer, K. [1982] Constrained Dynamics. Lecture Notes in Physics 169. Springer,
Berlin.
Sussmann, II.J. [1973] Orbits of families of vector fields and integrability of distributions.
Trans. Amer. Math. Soc. ISO, 171-188.
Takesaki, M. [1979] Theory of Operator Algebras I. Springer, Heidelberg.
Taylor, M.E. [1984J Pseudo Differential Operators. Princeton University Press, Princeton.
Taylor, M.E. and A. Uribe [1992] Semi-classical spectra of gauge fields. J. Funct. Anal.
nO,I-46.
Teitelboim, C. [1982] Quantum mechanics ofthe gravitational field. Phys. Rev. D25, 3159-
3179.
Teitelboim, C. [1984] Explicit evaluation of group-invariant measure as by-product of path
integration over Yang-Mills fields. J. Math. Phys. 25, 1093-1101.
Thirring, W. [1981] Quantum Mechanics ofAtoms and Molecules. Springer, Berlin.
Thirring, W. [1983] Quantum Mechanics of Large Systems. Springer, Berlin.
Thirring, W. and H. Narnhofer [1992] Covariant QED without indefinite metric. Rev. Math.
Phys. Special issue (dedicated to R. Haag) 193-211.
Thomas, E.G.F. [1983] A simple proof of the Cameron-Martin theorem making use of
Schwartz reproducing kernels. Bal. Soc. Mat. Mexicana 28, 67-76.
Topping, D. [1967] Asymptoticity and semimodularity in projection lattices. Pac. J. Math.
20,317-325.
Truman, A. [1976] Feynman path integrals and quantum mechanics as Ii ~ O. J. Math.
Phys. 17, 1852-1862.
Truman, A. [1977] The classical action in non-relativistic quantum mechanics. J. Math.
Phys. 18, 1499-1509.
Thynman, G.M. [1987a] Generalized Bergman kernels and geometric quantization. J. Math.
Phys.28,573-583.
Thynman, G.M. [1987b] Quantization: towards a comparison between methods. J. Math.
Phys.28,2829-2840.
Thynman, G.M. [1990] Reduction, quantization, and nonunimodular groups. J. Math. Phys.
31,83-90.
Thynman, G.M. and W.A.J.J. Wiegerinck [1987] Central extensions in physics. J. Geom.
Phys. 4, 207-258.
 References 517

Uhlmann, A. [1976] The "Transition Probability" in the state space of a "-algebra. Rep.
Math. Phys. 9, 273-279.
Uhlmann, A. [1993] Density operators as an arena for differential geometry. Rep. Math.
Phys.33,253-263.
Uhlmann, A. [1996] Spheres and hemispheres as quantum state spaces. J. Geom. Phys. 18,
7fr92.
Underhill, J. [1978] Quantization on a manifold with connection. J. Math. Phys. 19, 1932-
1935.
Unterberger, A. and J. Unterberger [1988] Quantification et analyse pseudodifferentielle.
Ann. Scient. Ec. Norm. Sup. 21, 133-158.
Unterberger, A. and H. Upmeier [1994] The Berezin transform and invariant differential
operators. Commun. Math. Phys. 164, 563-579.
Upmeier, H. [1987] Jordan algebras in analysis, operator theory, and quantum mechanics.
CBMS Reg. Conf. Ser. Math. 67, 1-85.
Upmeier, H. [1991] Wey 1quantization of symmetric spaces. I. Hyperbolic matrix domains.
J. Funct. Anal. 96, 297-330.
Upmeier, H. [1996] Toeplitz Operators and Index Theory in Several Complex Variables.
Birkhiiuser, Basel.
Vaisman, I. [1994] Lectures on the Geometry o/Poisson Manifolds. Birkhliuser, Basel.
Vaisman, I. [1996] Reduction of the Poisson-Nijenhuis manifolds. J. Geom. Phys. 19,
90-98.
Varadarajan, V.S. [1985] Geometry o/Quantum Theory (2nd ed.). Springer, New York.
Varilly, J.e., J.M. Gracia-Bondia, and W. Schempp [1990] The Moyal representation of
quantum mechanics and special function theory. Acta Appl. Math. 18, 225-250.
Vey, J. [1975J Deformation du crochet de Poisson sur une variete symplectique. Commun.
Math. Helv. 50, 421-454.
Vinet, L. [1988] Invariant Berry connections. Phys. Rev. D37, 2369-2372.
Vogan, D.A. [1987] Unitary Representations o/Reductive Lie Groups. Princeton University
Press, Princeton.
Vogan, D.A. [1992J Unitary representations of reductive Lie groups and the orbit method. In:
Tirao, J. and N. Wallach (eds.) New Developments in Lie Theory and their Applications,
pp. 87-114. Birkhiiuser, Basel.
Voros, A. [1977J Asymptotic Ii-expansions of stationary quantum states. Ann. Inst. H.
Poincare 26, 343-403.
Voros, A. [1978] An algebra of pseudodifferential operators and the asymptotics of quantum
mechanics. J. Funct. Anal. 29, 104-132.
Voros, A. [1989] Wentzel-Kramers-Brillouin method in the Bargmann representation.
Phys. Rev. A40, 6814-6825.
Wallach, N. [1973] Harmonic Analysis on Homogeneous Spaces. Dekker, New York.
Wallach, N. [1987] On the irreducibility and inequivalence of unitary representations of
gauge groups. Compositio Math. 64, 3-29.
Wang, X.-P. [1986] Approximation semi-classique de l'equation de Heisenberg. Commun.
Math. Phys. 104, 77-86.
518 References

Wang, X.-P. [1991] Semiclassical resolvent estimates for N -body SchrOdinger operators.
J. Funct. Anal. 97, 466-483.
Warner, G. [1972] Harmonic Analysis on Semi-simple Lie Groups, Vol. I. Springer, Berlin.
Wegge-Olsen, N.E. [1993] K -theory and C'-algebras. Oxford University Press, Oxford.
Weinberg. S. [1995] The Quantum Theory of Fields. Vol. I. Cambridge University Press,
Cambridge.
Weinberg. S. [1996] The Quantum Theory of Fields. Vol. II. Cambridge University Press,
Cambridge.
Weinstein, A. [1978] A universal phase space for particles in a Yang-Mills field. Lett. Math.
Phys. 2,417-420.
Weinstein, A. [1983] The local structure of Poisson manifolds. J. Diff. Geom. 18,523-557.
Err. ibid. 22 (1985) 255.
Weinstein, A. [1987] Poisson geometry of the principal series and nonlinearizable structures.
J. Diff. Geom. 25, 55-73.
Weinstein, A. [1989] Blowing up realizations of Heisenberg-Poisson manifolds. Bull. Sc.
math. (2) 113, 381-406.
Weinstein, A. [1990] Affine Poisson structures. Int. J. Math. 1,343-360.
Weinstein, A. [1991] Noncommutative geometry and geometric quantization. In: Symplectic
Geometry and Mathematical Physics. Progr. Math. 99, pp. 446-461. Birkhauser, Basel.
Weinstein, A. [1995a] Deformation quantization. Sem. Bourbaki 789. Asterisque 227, 389-
409.
Weinstein, A. [1995b] The symplectic structure on moduli space. In: The Floer Memorial
volume. Progr. Math. 133, pp. 627-635. Birkhauser, Basel.
Weinstein, A. [1996a] Groupoids: unifying internal and external symmetry. Notices Amer.
Math. Soc. 43, 744-752.
Weinstein, A. [1996b] Lagrangian mechanics and groupoids. Fields Inst. Commun. 7, 207-
231.
Weinstein, A. [1997] The modular automorphism group of a Poisson manifold. J. Geom.
Phys. 23, 379-394.
Weinstein, A. [1998] Poisson geometry. Diff. Geom. Appl. 9, 213-238.
Weinstein, A. and P. Xu [1991] Extensions of symplectic groupoids and quantization. J.
reine angew. Math. 417, 159-189.
Weiss, E. and N. Zierler [1958] Locally compact division rings. Pac. J. Math. 8, 369-371.
Werner, R.E [1983] Physical uniformities in the state space of nonrelativistic quantum
mechanics. Found. Phys. 13, 859-881.
Werner, R.E [1995] The classical limit of quantum theory. e-print quant-ph/95040I6.
Westman, J. [1968] Harmonic analysis on groupoids. Pac. J. Math. 27,621-632.
Weyl, H. [1931] The Theory of Groups and Quantum Mechanics. Dover, New York.
Widom, H. [1980] A complete symbolic calculus for pseudodifferential operators. Bull. Sc.
math., 2' serie 104, 19-63.
Wiedemann, U.A. [1994] Constraints and Spontaneous Symmetry Breaking in Quantum
Field Theory, Ph.D. thesis, University of Cambridge.
 References 519

Wiedemann, U.A. and N .P. Landsman [1996] The Stueckelberg-Kibble model as an example
of quantized symplectic reduction. J. Math. Phys. 37, 2731-2747.
Wightman, A.S. [1962] On the localizability of quantum mechanical systems. Rev. Mod.
Phys. 34, 845-872.
Wigner, E.P. [1931] Gruppentheorie und ihre Anwendung auf die Quantenmechanik der
Atomspektren. Vieweg, Braunschweig.
Wigner, E.P. [1932] On the quantum correction for thermodynamic eqUilibrium. Phys. Rev.
40, 749-759.
Wigner, E.P. [1939] Unitary representations of the inhomogeneous Lorentz group. Ann.
Math.40,149-204.
Wilbur, WJ. [1977] On characterizing the standard quantum logics. Trans. Amer. Math.
Soc. 233, 265-282.
Wildberger, N.J. [1992] The moment map of a Lie group representation. Trans. Amer. Math.
Soc. 330,257-268.
Witten, E. [1991] On quantum gauge theories in two dimensions. Commun. Math. Phys.
141, 153-209.
Witten, E. [1992] Thto dimensional gauge theories revisited. J. Geom. Phys. 9, 303-368.
Wong, S.K. [1970] Field and particle equations for the classical Yang-Mills field and
particles with isotopic spin. Nuovo Cim. A65, 689-694.
Wong, Y.-C. and K.-F. Ng [1973] Partially Ordered Topological Vector Spaces. Clarendon
Press, Oxford.
Woodhouse, N.MJ. [1992] Geometric Quantization. 2nd ed. Clarendon Press, Oxford.
Woronowicz, S. L. [1987] Compact matrix pseudogroups. Commun. Math. Phys. 111,613-
665.
Woronowicz, S. L. [1995] C' -algebras generated by unbounded elements. Rev. Math. Phys.
7,481-521.
Wren, K.K. [1997] Quantization of constrained systems with singularities using Rieffel
induction.J. Geom. Phys. 24,173-202.
Wren, K.K. [1998a] Constrained quantization and 9-angles. II. Nucl. Phys. 8521 [PM],
471-502.
Wren, K.K. [1998b] Constrained Quantization of Yang-Mills Theory via Rieffel Induction.
Ph.D. thesis, University of Cambridge.
Wright, J.D.M. [1977] Jordan C'-algebras. Michigan Math. J. 24, 291-302.
Wu, T.T. and C.N. Yang [1975] Concept of non-integrable phase factors and global
formulation of gauge fields. Phys. Rev. D12, 3845-3857.
Wu, T.T. and C.N. Yang [1976] Dirac monopoles without strings: monopole harmonics.
Nucl. Phys. 8107, 365-380.
Wu, Y. [1998] Quantization of a particle in a background Yang-Mills field. J. Math. Phys.
39. 867-875.
Xu, P. [1991a] Morita equivalent symplectic groupoids. In: Dazord, P. and A. Weinstein
(eds.) Symplectic Geometry, Groupoids, and Integrable Systems, pp. 291-311. Springer,
New York.
Xu, P. [1991b] Morita equivalence of Poisson manifolds. Commun. Math. Phys. 142,493-
509.
520 References

Xu, P. [1992] Morita equivalence and symplectic realizations of Poisson manifolds. Ann.
Sc. Ec. Norm. Sup. 25, 307-333.
Yaffe, L.G. [1982] Large N limits as classical mechanics. Rev. Mod. Phys. 54, 407-428.
Yajima, K. [1979] The quasi-classical limit of quantum scattering theory. Commun. Math.
Phys. 69, 101-129.
Yosida, K. [1980] Functional Analysis, 6th ed. Springer, Berlin.
Zabey, P.C. [1975J Reconstruction theorems in quantum mechanics. Found. Phys. 5, 323-
342.
Zakrzewski, S. [1986] Induced representations and induced Hamiltonian actions. J. Geom.
Phys. 3, 211-219.
Zakrzewski, S. [1990a] Quantum and classical pseudogroups. I. Union pseudogroups and
their quantization. Commun. Math. Phys. 134, 347-370.
Zakrzewski, S. [ 1990b] Quantum and classical pseudogroups. II. Differential and symplectic
pseudogroups. Commun. Math. Phys. 134,371-395.
Zelditch, S. [1992] On a "quantum chaos" theorem ofR. Schroder and M. Taylor. J. Funct.
Anal. 109, 1-21.
Zhang, W.M., D.H. Feng, and R. Gilmore [1990J Coherent states: theory and some
applications. Rev. Mod. Phys. 62, 867-927.
Ziegler, F. [1996J Methode des Orbites et Representations Quantiques. Ph.D. thesis,
Universite de Provence.
Zierler, N. [1961] Axioms for non-relativistic quantum mechanics. Pac. J. Math. 11, 1151-
1169.
Index

AC-lattices, 100 associator identity, 2, 38

action atom, 92
CO-algebra, 23, 291 atomic lattice, 92
• -algebra, 279 atomistic lattice, 98
algebroid, 23, 293 automorphic action, 139
groupoid, 22, 271 axioms for the pure state space of a
Poisson algebra, 298 C· -algebra, 104
adjoint action, 184
adjoint bundle, 235 Banach algebra, 39
adjointable operator, 28, 356 Bargmann-Fock space, 449
affine base
connection, 154 of a bundle, 224
function, 51 of a groupoid, 22, 269
geometry, 154 based
affinely parametrized geodesic, 454 gauge transformations, 416
Aharonov-Bohm effect, 36, 430 loops, 416
algebra, 37 basis of a transition probability space, 7,
• -algebra, 39 81
of bounded operators, 39 Berezin quantization, 10, 114
of finite-rank operators, 56 on coadjoint orbits, 219
a-density, 273 on flat space, 133
amenable, 205 Berezin transform, 447
anchor, 23, 292 Berezin-Toeplitz quantization, 448
anisotropic form, 94 Bergman kernel, 135, 448
annihilation operator, 136 bi-invariant metric, 212
anti unitary operator, 441 bicommutant, 58
approximate unit, 48 Birkhoff's exchange axiom, 442
associated bundle, 17, 226 Bochner integral, 448
associated vector bundle, 227 bosonic Fock space, 34, 135
522 Index

bounded symmetric domain, 447 2-cocycle on g', 181

bundle, 224 cogeodesic flow, 160
automorphism, 235 coherent pure state quantization, 122
of a-densities, 273 coherent states, 122
coisotropic representation, 245
C* -algebra, 2, 39 coisotropic submanifold, 25, 315
of a Lie group, 204 commutant, 58
of a Lie groupoid, 285 commutation relations on a Riemannian
of compact operators, 56 manifold, 170
Cameron-Martin formula commutative C' -algebra, 4D
general, 41 1 compact Lie groups, 215
Cameron-Martin theorem complete
general, 412 element of a Poisson algebra, 67
canonical commutation relations, 129 lattice, 92
in Weyl form, 4D7 Riemannian manifold, 159
canonical symplectic form, 68 completely integrable, 69
Casimir element, 258 completely positive map, 117
CCR algebra, 4D7 completeness of a quantization, 109
center complex connection, 415
of a lattice, 94 complexification, 418
of a Poisson algebra, 318 component of a transition probability
of a von Neumann algebra, 60 space, 80
central extension, 181, 188 condition of the frontier, 473
c-extension, 188 conjugate representation, 207
chain, 100 conjugate space, 364
character, 215 connection, 18, 227
Christoffel symbol, 158 G-invariant,247
classical I-form, 228
action, 151 coefficients, 154
constrained systems, 313 constrained quantization, 386
covariance condition, 299 and vacuum angles, 427
dual pair, 320 author's method, 387
imprimitivity theorem, 26, 322 BRST method, 477
limit, 8 Dirac's method, 389
on a Riemannian manifold, 173 constrained systems
on flat space, 148 classical, 313
propagator, 151 quantum, 386
system of imprimitivity, 24, 299 constraint, 314
transitive imprimitivity theorem, 27, continuous field
332 of C' -algebras, 110
closed projection, 442 of states, 112
Co-equivariant momentum map, 185 continuous functional calculus, 44
coadjoint action, 15, 127, 184 continuous quantization, 112
coadjoint orbit, 194 contravariant symbol, 447
from covariant reduction, 396 convex hull, 62
of a nilpotent group, 341 convolution
of a semidirect product, 343 on a Lie group, 202
of the Poincare group, 393 on a Lie groupoid, 275
l-cocycle on G, 185 cotangent bundle reduction, 231
 Index 523

covariance, 411 pure states, 65

covariance condition, 280 quantizations, 109
covariant representations
representations, 399 of a C* -algebra, 52
Berezin quantization, 384 of a Poisson algebra, 76
derivative, 158,229 equivariant momentum map, 15, 185
localization in configuration space, ergodic measure, 382
384 Euclidean group, 348
localization in phase space, 384 exchange property, 442
momentum, 239 exponential
representation, 33 Hilbert space, 135
symbol,447 mapping, 155
covering property, 99 vector, 135
creation operator, 136 extreme boundary, 61
c-representation, 197 extreme point, 61
crossed product, 279
"-algebra, 279 Face, 95
Poisson algebra, 298 factor, 441
C"-dynamical system, 279 faithful state, 54
c-unitary dual, 205 Fell induction, 363
curvature, 228 Fermi representation, 480
cut locus, 159 fiber, 224
cyclic representation, 53 fiber product, 224
cyclic vector, 53 first cohomology group of g, 180
first cohomology group of G, 186
Dimension of a transition probability first-class constraint, 314
space, 81 Fock representation, 407
Dirac Frobenius reciprocity theorem, 209
condition, 109 Frobenius theorem, 69
method,389 Fubini-Study symplectic form, 75
monopole, 264 full Hilbert C* -module, 30, 364
quantization condition, 269 full dual pair, 471
direct integral, 284 fundamental vector field, 227
distance in a Riemannian manifold, 157
distribution on a manifold, 69 Gllrding subspace, 459
distributive lattice, 92 g-action, 179
division ring, 441 r -representation, 197
dominant weight, 217 gauge
double commutant theorem, 58 algebroid, 293
dual bundle, 225 group, 33, 34,235,405
groupoid, 22, 272
Egorov's theorem, 452 invariance, 403
enveloping algebra, 199 gauge-covariant, 239, 256
equation of geodesic deviation, 160 Gauss law, 418
equivalent Gaussian measure, 411
multipliers, 189 Gelfand transform
projective representations in a C* -algebra, 3, 62
of a Lie algebra, 198 in a commutative Banach algebra, 43
of a Lie group, 197 generalized
524 Index

generalized (continued) subbundle, 227

distribution, 69 subspace, 154,227
system of imprimitivity, 383 vectors, 227
generator, 183 Howe dual pair, 471
geodesic, 155 Husimi function, 452
flow, 155
motion, 13 Ideal,47
geodesically complete, 155 left,47
geodesically convex, 158 right, 47
geometric quantization, 446 implemented automorphism, 259
G-invariant connection, 247 imprimitivity theorem
GNS-construction, 3, 52 classical, 26, 322
Groenewold-van Hove theorem, 449 for gauge groupoids, 378
group CO-algebra, 15,204 quantum, 30, 368
group action, 183 inclusion, 269
groupoid, 21, 269 incomplete vector field, 67
C* -algebra, 288 induced representation, 19, 253
action, 300 induction, 25
Fell,363
Half-densities on Lie groupoids, 273 in stages, 375
Hall coherent states, 35, 426 Mackey, 378
Hamiltonian, 15 quantum Marsden-Weinstein, 377
g-action, 180 Rieffel,375
curve, 67 Mackey, 208,256
flow, 4, 67 Rieffel, 28, 29, 358
group action, 184 inequivalent quantizations, 36, 430
on a Riemannian manifold, 173 infinitesimal isometry, 172
Riemannian geometry, 159 integrable
vector field, 3, 66 Poisson manifold, 470
H -connection, 246 representation of Coo (g~ , IR), 183
heat equation, 425 representation of g, 199
Heisenberg group, II, 126 integral coadjoint orbit, 215
helicity, 32, 395 invariant measure, 283
Hermitian form, 93 inversion, 269
Hermitian functional, 49 involution, 39, 441
highest weight vector, 218 involutive, 69
Hilbert irreducible
C* -module, 28, 354 lattice, 94
lB-module, 354 representation
bundle, 225 of a C* -algebra, 3, 63
space of half-densities, 273 of a Poisson algebra, 5, 78
subspace, 124 transition probability space, 6, 80
Hilbert-Schmidt norm, 56 isometry, 172
Hopf fibration, 264 isomorphic bundles, 224
horizontal isomorphism
I-forms, 231 of C' -algebras, 40
curve, 154 of Jordan algebras, 38
cotangent space, 231 of Poisson algebras, 38
lift, 154, 227 isotropic, 315
 Index 525

isotropic representation, 245 algebra action, 178

isotropy group, 270 algebroid, 23, 292
Ito's map, 423 algebroid action, 301
groupoid, 22, 272
Jacobi Poisson structure, 14, 178
equation, 160 symplectic form, 195
field, 160 symplectic structure, 15
identity, 38 lift,227
Jacobson topology, 311 linear partial ordering, 45
J B-algebra, 2, 38 Liouville measure, 10, 114
J LB-algebra, 2,38 local section, 69, 224
Jordan local trivialization, 224
algebra, 2, 37 localization
automorphism, 38 in configuration space, 122
isomorphism, 38 in phase space, 122
morphism, 38 in 1R3 , 385
product, 2, 37 of a photon, 477
Jordan-Lie algebra, 2, 37 locally Lebesgue, 113
locally uniformly closed, 110
Kaluza-Klein construction, 463 loop group, 415
Kato-Rellich theorem, 457 Lorentz force, 242
Kazhdan-Kostant-Sternberg reduction, Lorentz group, 394
27,328 lower symbol, 447
Killing form, 216
Krein-Milman theorem, 435 Mackey induction, 20, 256
in stages, 378
Lagrangian, 315 magnetic field potential, 149
Laplace-Beltrami operator, 173 magnetic monopole, 266
Laplace-Bochner operator, 257 manifold topology, 72
Laplacian, 425 Marsden-Weinstein quotient, 27, 326
large gauge transformations, 416 Marsden-Weinstein reduction, 27, 325
lattice, 92 in stages, 338
automorphism, 93 quantum, 370
homomorphism, 93 singular, 349
isomorphism, 93 Maurer-Cartan equations, 192
left Maurer-Cartan form, 192
exponential map, 303 maximal torus, 215
Haar system, 276 metaplectic group, 132
ideal,47 metaplectic representation, 132
invariant flow, 294 metric, 157, 158
invariant vector field, 293 minimal geodesic, 159
regular representation, 202 Minkowski space, 393
trivialization, 192 modular lattice, 92
Leibniz rule, 38 modular pair, 441
length momentum map, 15, 179
of a curve, 157 equivariant, 15, 185
of a lattice, 100 momentum operator, 11, 129
Levi-Civita connection, 158 Morita equivalence
Lie classical, 26, 321
526 Index

Morita (continued) perfect C'-algebra, 88

quantum, 30, 366 Peter-Weyl theorem, 207
morphism photons, 406
between C' -algebras, 40 7f x -positive, 362
between Jordan-Lie algebras, 38 Plancherel transform, 207
moving frame, 229 Planck's constant, 2
Moyal product, 45 1 Poincare group, 394
multiplier, 15, 188 coadjoint orbits, 393
multiplier algebra, 357 irreducible representations, 399
Poisson
Naturally isomorphic, x {I-action, 179
nilpotent, 341 algebra, 2, 38
Noether's theorem, 184 of a Lie algebra, 178
noncommutative topology, 442 of a Lie algebroid, 296
noncommutative torus, 468 automorphism, 38
nondegenerate bracket, 1, 37
quantization, 109 commutant,318
representation, 53 irreducibility, 79
norm-exposed, 82 isomorphism, 38
normal manifold, 3, 66
bundle, 163 map, 67
coordinates, 158 morphism, 38
groupoid, 308 space, 76
neighborhood, 158 space with a transition probability, 6,
state, 59 86
state space, 60 tensor, 66
normalization of a state, 49 polarizing subalgebra, 342
normalized functional, 2 position operator, 11, 129
null distribution, 313 positive
null foliation, 313 cone, 45
element of a C· -algebra, 45
Open projection, 442 functional, 2
orbit, 271, 470 map, 47
orthoclosed, 80 operator, 45
orthoclosure, 80 quantization, 109
orthocomplementation, 92 root, 217
orthocomplemented lattice, 93 positive-operator-valued measure, 121
orthogonal, 80 positivity, 49
orthomodular, 93 POVM,121
orthoplement, 80 pre-C' -algebra, 355
oscillatory integral, 450 pre-Hilbert ~-module, 355
predual,59
Pair algebroid, 293 prime quantization, 447
pair groupoid, 22, 271 primitive ideal, 310
parallel transport, 154, 228 primitive spectrum, 310
parity operator, 11, 132 principal
partial ordering, 434 H -bundle over Q, 225
partially ordered vector space, 45 bundle, 225
partition of unity, 446 fi ber bundle, 17
 Index 527

projection, 56 symplectic, 25
projection-valued measure, 121 reductive
projective decomposition, 246
Hilbert space, 71 subgroup, 246
representation regular
of a Lie algebra, 197 coadjoint orbit, 216
of a Lie group, 197 distribution, 69
space, 71 foliation, 69
proper group action, 325 Lie group action, 381
pullback, 163 representation, 283
pullback bundle, 225 semidirect product, 345
pure state, 3, 61 weight, 217
quantization, 9, 113 relative bound, 457
space, 61 representation
PVM,121 of a C' -algebra, 3, 52
of a groupoid, 282
Quantum of a Poisson algebra, 4, 76
constrained systems, 386 reproducing kernel, 123
dual pair, 364 Ricci scalar, 158
field theory of photons, 407 Rieffel
imprimitivity theorem, 30, 368 condition, 108
Marsden-Weinstein reduction, 370 induction, 28, 358
in stages, 377 in stages, 375
transitive imprimitivity theorem, 31, quantization, 212
291 Riemann curvature tensor, 158
quasi-invariant, 253, 283 Riemannian geometry, 157
right
Radon measure, 55 Haar system, 276
rank ideal, 47
of a compact Lie group, 215 regular representation, 206
of a distribution, 69 trivialization, 192
real C* -algebra, 434 root, 217
real vector field, 405 R*-algebra, 434
reduced
atomic representation, 65 Scalar potential, 149
group C* -algebra, 202 SchrOdinger
groupoid C* -algebra, 288 equation, 74
regular representation, 286 operator, 149
representation, 26, 316 representation, 128
space, 25 Schur's lemma, 63
unitary dual, 205 second
reducible class constraint, 314
lattice, 94 cohomology group of g, 182
transition probability space, 80 quantization, 480
reduction section, 17
in stages, 336 of a bundle, 224
Kazhdan-Kostant-Sternberg, 27, 328 of a continuous field of C' -algebras,
Marsden-Weinstein, 27, 325 110
special symplectic, 26, 316 of a field of Hilbert spaces, 284
528 Index

sector, 6, 80 group action, 184

self-adjoint structure group, 225
element of a • -algebra, 40 subgroupoid, 271
part of a C· -algebra, 2, 40 sup-norm, ix
self-dual Hilbert C' -module, 357 superselection rules, 2
semiclassical propagator, 152 symbol, 156, 297
semiclassical Schrooinger equation, 152 symmetric transition probability space,
semidirect product, 298 5,81
semimodularity,441 symplectic
shifting trick, 326 affinoid space, 471
a-weak topology, 59 cocycles, 186
singular decomposition of a Poisson manifold,
coadjoint orbit, 216 69
foliation, 69 form, 68
Marsden-Weinstein reduction Fourier transform, 143
classical, 349 group, 129
quantum, 390 groupoid, 470
weight, 217 induction, 472
small gauge transformations, 416 leaf, 70
smooth leaves, 4
C' -dynamical system, 279 Lie groupoid action, 470
distribution, 69 orthogonal complement, 313
groupoid action, 300 piece, 352
Poisson space, 77 Poisson manifold, 4, 68
system of imprimitivity, 280 reduction, 25, 313
vector, 198 reduction in stages, 336
source projection, 269 submanifold, 315
special symplectic reduction, 26, 316 symplectically complete foliation, 319
spectral symplectomorphic, 69
radius, 41 symplectomorphism, 69
resolution, 88 system of imprimitivity
theorem, 88 classical, 24, 299
spectrum, 42 quantum, 23, 291
spin, 395 Szego projection, 448
s-system, 276
stable isomorphism, 475 Tangent groupoid, 309
state, 49 target projection, 269
state space, 2, 49 f)-angle, 32, 36, 432
Stieffel chamber, 419 time-dependent WKB method, 12, 453
strict deformation quantization, 109 time-independent WKB method, 454
strict quantization Toeplitz operators, 448
of a Poisson algebra, 9, 108 torsion-free connection, 158
of a Poisson manifold, 109 torus, 215
strong total space
Morita equivalence, 475 of a bundle, 224
symplectic form, 435 of a groupoid, 269
operator topology, 58 transformation group C' -algebras, 467
strongly Hamiltonian transition
g-action, 15,181 functions, 226
 Index 529

probability, 5,80 curve, 154

probability space, 80 subspace, 154
transitive tangent space, 227
genemlized system of imprimitivity, vectors, 227
383 von Neumann algebm, 59
groupoid, 271 von Neumann's condition, 108
imprimitivity theorem
classical, 27, 332 Weak
quantum, 31, 291 classical observable, 25, 315
system of imprimitivity, 332 opemtor topology, 59
trivial quantum observable, 387
bundle, 224 symplectic form, 435
cocycle, 182 weakly contained, 205
continuous field of C' -algebras, 110 weight, 215
(-system, 276 weight lattice, 215
tubular neighborhood, 163 weights of a representation. 217
tubular neighborhood theorem, 455 Weinstein dual pair, 471
twisted well-behaved, 81
convolution, 202 Weyl
covariance algebra, 460 chamber, 217
enveloping algebra, 200 exponential map, 304
group C' -algebm, 204 group, 215
Lie-Poisson structure, 192 operator, 129
reduced group C' -algebm, 202 opemtors, 11
two-sphere property, 98 quantization
typical fiber, 224 on a Riemannian manifold, 164
on flat space, 11, 141
Uniform on the dual of a Lie algebra, 212
Poisson space, 77 on the dual of a Lie algebroid, 306
unimodular, 201 symbol,142
unit, 41, 270 Wiener measure, 35
unit space, 270 on loops in H, 423
unital J B-algebra, 41 on paths in ~, 423
unitarity, 6 on paths in H, 423
unitary,86 Wigner function, 142
unitary dual, 205 Wilson loop, 35,417
unitization, 41 Wong equations
universal representation, 54 classical, 18, 242
upper symbol, 447 quantum, 257

Vacuum angle, 32, 36, 432 Yang-Mills Hamiltonian, 258

vector Yang-Mills theory on a circle
bundle, 17,225 classical, 34, 414
representation, 33, 401 quantum, 420
state, 53
vertical Zero section, 163,225