Rudolf Haag

Local
Quantum Physics
Fields, Particles, Algebras

Second Revised and Enlarged Edition

With 16 Figures

Springer
Professor Dr. Rudolf Haag
Waldschmidtstrasse 4b, D-83727 Schliersee-Neuhaus, Germany

Editors

Roger Balian Nicolai Reshetikhin

CEA Department of Mathematics
Service de Physique Théorique de Saclay University of California
F-91191 Gif- sur -Yvette, France Berkeley, CA 94720-3840, USA

Wolf Beiglbdck Herbert Spohn

Institut für Angewandte Mathematik Theoretische Physik
Universitüt Heidelberg Ludwig-Maximilians-Universitüt Munchen
Im Neuenheimer Feld 294 Theresienstral3e 37
D-69120 Heidelberg, Germany D-80333 Munchen, Germany

Harald Grosse Walter Thirring

Institut für Theoretische Physik Institut für Theoretische Physik
Universitüt Wien Universitüt Wien
Boltzmanngasse 5 Boltzmanngasse 5
A-1090 Wien, Austria A-1090 Wien, Austria

Elliott H. Lieb
Jadwin Hall
Princeton University, P. 0. Box 708
Princeton, NJ 08544-0708, USA

Library of Congress Cataloging-in-Publication Data. Haag, Rudolf, 1922— Local quantum physics :
fields, particles, algebras / Rudolf Haag, — 2nd rev. and enl. ed. p. cm . — (Texts and monographs in
physics) Includes bibliographical references and index. ISBN 3-540-61049-9 (Berlin : alk. paper)
1. Quantum theory. 2. Quantum field theory. I. Title. II. Series. QC174.12.H32 1996 530.1'2—dc20
96-18937

ISSN 0172-5998
ISBN 3-540-61049-9 2nd Edition (Softcover)
Springer-Verlag Berlin Heidelberg New York
ISBN 3-540-61451-6 2nd Edition (Hardcover)
Springer-Verlag Berlin Heidelberg New York
ISBN 3-540-53610-8 1st Edition (Hardcover) Springer-Verlag Berlin Heidelberg New York
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcast-
ing, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this
publication or parts thereof is permitted only under the provisions of the German Copyright Law of
September 9, 1965, in its current version, and permission for use must always be obtained from
Springer-Verlag. Violations are liable for prosecution under the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1992, 1996. Printed in Germany
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
Typesetting: Data conversion by K. Mattes, Heidelberg
SPIN: 10515879 (soft) / I0542737 (hard) 55/3144-543210 - Printed on acid-free paper
Dedicated to Eugene P. Wigner in deep gratitude
and to the memory of
Valja Bargmann and Res Jost
Preface to the Second Edition

The new edition provided the opportunity of adding a new chapter entitled
"Principles and Lessons of Quantum Physics". It was a tempting challenge to
try to sharpen the points at issue in the long lasting debate on the Copenhagen
Spirit, to assess the significance of various arguments from our present vantage
point, seventy years after the advent of quantum theory, where, after all, some
problems appear in a different light. It includes a section on the assumptions
leading to the specific mathematical formalism of quantum theory and a section
entitled "The evolutionary picture" describing my personal conclusions. Alto-
gether the discussion suggests that the conventional language is too narrow and
that neither the mathematical nor the conceptual structure are built for eter-
nity. Future theories will demand radical changes though not in the direction
of a return to determinism. Essential lessons taught by Bohr will persist. This
chapter is essentially self-contained.
Some new material has been added in the last chapter. It concerns the char-
acterization of specific theories within the general frame and recent progress in
quantum field theory on curved space-time manifolds. A few pages on renor-
malization have been added in Chapter II and some effort has been invested in
the search for mistakes and unclear passages in the first edition.
The central objective of the book, expressed in the title "Local Quantum
Physics", is the synthesis between special relativity and quantum theory to-
gether with a few other principles of general nature. The algebraic approach,
that is the characterization of the theory by a net of algebras of local observ-
ables, provides a concise language for this and is an efficient tool for the study
of the anatomy of the theory and of the relation of various parts to qualitative
physical consequences. It is introduced in Chapter III. The first two chapters
serve to place this into context and make the book reasonably self-contained.
There is a rough temporal order. Thus Chapter I briefly describes the pillars of
the theory existing before 1950. Chapter II deals with progress in understanding
and techniques in quantum field theory, achieved for the most part in the 1950s
and early 1960s. Most of the material in these chapters is probably standard
knowledge for many readers. So I have limited the exposition of those parts
which are extensively treated in easily available books to a minimum: in partic-
ular I did not include a discussion of the path integral and functional integration
techniques. Instead I tried to picture the panorama, supply continuous lines of
argument, and discuss concepts, questions, and conclusions so that, hopefully,
VIII Preface

Chapters I—III by themselves may serve as a useful supplementary text for stan-
dard courses in quantum field theory. Chapters IV—VI address more advanced
topics and describe the main results of the algebraic approach.
The remarks on style and intentions of the book are quoted below from the
preface to the first edition.

April 1996 Rudolf Haag

From the Preface to the First Edition

... Physical theory has aspects of a jigsaw puzzle with pieces whose exact
shape is not known. One rejoices if one sees that a large part of the pieces fits
together naturally and beautifully into one coherent picture. But there remain
pieces outside. For some we can believe that they will fit in, once their proper
shape and place has been recognized. Others definitely cannot fit at all. The
recognition of such gaps and misfits may be a ferment for progress. The spirit
of tentative search pervades the whole book.
The book is not addressed to experts. I hope that at least the essential lines
of argument and main conclusions are understandable to graduate students
interested in the conceptual status of presently existing theory and sufficiently
motivated to invest some time, thinking on their own and filling in gaps by
consulting other literature.
Local Quantum Physics is a personal book, mirroring the perspective of the
author and the questions he has seriously thought about. This, together with
the wish to keep the volume within reasonable bounds implied that many topics
could only be alluded to and that I could not attempt to produce a well balanced
list of references.
Thanks are due to many friends for valuable advice and correction of errors,
especially to Detlev Buchholz, Klaus Fredenhagen, Hans Joos, Daniel Kastler,
Heide Narnhofer, Henning Rehren, John Roberts, Siegfried Schlieder, and Wal-
ter Thirring. I am grateful to Wolf Beiglbbck for his unwavering support of the
project at Springer-Verlag ... , to Michael Stiller, Marcus Speh, and Hermann
Hefiling for their perseverance in preparing files of the manuscript.
Last but not least I must thank Barbara, without whose care and constant
encouragement this book could not have been written.

November 1991 Rudolf Haag

Organization of the Book

The book is divided into chapters and sections. Some sections, but not all,
are subdivided. Thus 111.2.1 denotes Subsection I of Section 2 in Chapter III.
Equations are consecutively numbered in each section. For example (III.2.25) is
equation number 25 in Section 2 of Chapter III. All types of statements (theo-
rems, definitions, assumptions ...) are consecutively numbered in each subsection
and the chapter number is suppressed; thus lemma 3.2.1 is the first statement
in Subsection 3.2 of the running chapter. The index is divided into three parts:
1. Bibliography (listing books).
2. Author Index and References; it combines the listing of authors with the
quotation of articles in journals and the indication of the place in the text (if
any) where the respective article is referred to.
3. Subject Index. Only those page numbers in the text are given where the
item is defined or first mentioned or where some significant further aspect of it
appears.
References from the text to the bibliography are marked in italics with the
full name of the author written out; references to journal articles are indicated
by an abbreviated form of an author's name and the year of publication such
as [Ara 611.
For the benefit of the reader who likes to start in the middle, here are a
few remarks on notation and specific symbols. Vectors and matrix elements in
Hilbert space are usually written in the Dirac notation: IW), (WIA10)• In the
case of vectors the bracket is sometimes omitted. Generically, the symbol A is
used to denote a general *-algebra, whereas the symbol 21 is used in the case of
a C*-algebra generated by observables and R. is used for a von Neumann ring
or W*-algebra.
Contents

I. Background 1

1. Qu antum Mechanics 1
Basic concepts, mathematical structure, physical interpretation.

2. The Principle of Locality in Classical Physics

an d the Relativity Theories 7
Faraday's vision. Fields. 2.1 Special relativity. Poincaré group. Lorentz group.
Spinors. Conformal group. 2.2 Maxwell theory. 2.3 General relativity.

3. Poincaré Invariant Qu antum Theory 25

3.1 Geometric symmetries in quantum physics. Projective representations and
the covering group. 3.2 Wigner's analysis of irreducible, unitary representa-
tions of the Poincaré group. 3.3 Single particle states. Spin. 3.4 Many particle
states: Bose-Fermi alternative, Fock space, creation operators. Separation of
CM-motion.

4. Action Principle 39
Lagrangean. Double rôle of physical quantities. Peierls' direct definition of Pois-
son brackets. Relation between local conservation laws and symmetries.

5. Basic Qu antum Field Theory 42

5.1 Canonical quantization. 5.2 Fields and particles. 5.3 Free fields. 5.4 The
Maxwell-Dirac system. Gauge invariance. 5.5 Processes.

II. General Quantum Field Theory 53

1. Mathematical Considerations and General Postulates 53

1.1 The representation problem. 1.2 Wightman axioms.

2. Hierarchies of Functions 58
2.1 Wightman functions, reconstruction theorem, analyticity in x-space.
2.2 Truncated functions, clustering. Generating functionals and linked cluster
XII Contents

theorem. 2.3 Time ordered functions. 2.4 Covariant perturbation theory, Feyn-
man diagrams. Renormalization. 2.5 Vertex functions and structure analysis.
2.6 Retarded functions and analyticity in p-space. 2.7 Schwinger functions and
Osterwalder-Schrader theorem.

3. Physical Interpretation in Terms of Particles 75

3.1 The particle picture: Asymptotic particle configurations and collision the-
ory. 3.2 Asymptotic fields. S-matrix. 3.3 LSZ-formalism.

4. General Collision Theory 84

4.1 Polynomial algebras of fields. Almost local operators. 4.2 Construction of
asymptotic particle states. 4.3. Coincidence arrangements of detectors.
4.4 Generalized LSZ-formalism.

5. Some Consequences of the Postulates 96

5.1 CPT-operator. Spin-statistics theorem. CPT-theorem. 5.2 Analyticity of the
S-matrix. 5.3 Reeh-Schlieder theorem. 5.4 Additivity of the energy-momentum-
spectrum. 5.5 Borchers classes.

III. Algebras of Local Observables and Fields 105

1. Review of the Perspective 105

Characterization of the theory by a net of local algebras. Bounded operators.
Unobservable fields, superselection rules and the net of abstract algebras of ob-
servables. Transcription of the basic postulates.

2. Von Neum ann Algebras. C*-Algebras. W*-Algebras 112

2.1 Algebras of bounded operators. Concrete C*-algebras and von Neumann
algebras. Isomorphisms. Reduction. Factors. Classification of factors. 2.2 Ab-
stract algebras and their representations. Abstract C*-algebras. Relation be-
tween the C*-norm and the spectrum. Positive linear forms and states. The
GNS-construction. Folia of states. Intertwiners. Primary states and cluster prop-
erty. Purification. W*-algebras.

3. The Net of Algebras of Local Observables 128

3.1 Smoothness and integration. Local definiteness and local normality.
3.2 Symmetries and symmetry breaking. Vacuum states. The spectral ideals.
3.3 Summary of the structure.

4. The Vacuum Sector 143

4.1 The orthocomplemented lattice of causally complete regions. 4.2 The net of
von Neumann algebras in the vacuum representation.
 Contents XIII

IV. Charges, Global Gauge Groups

and Exchange Symmetry 149

1. Charge Superselection Sectors 149

"Strange statistics". Charges. Selection criteria for relevant sectors. The pro-
gram and survey of results.

2. The DHR-Analysis 156

2.1 Localized morphisms. 2.2 Intertwiners and exchange symmetry ("Statis-
tics"). 2.3 Charge conjugation, statistics parameter. 2.4 Covariant sectors and
energy-momentum spectrum. 2.5 Fields and collision theory.

3. The Buchholz-Fredenhagen-Analysis 174

3.1 Localized 1-particle states. 3.2 BF-topological charges. 3.3 Composition of
sectors and exchange symmetry. 3.4 Charge conjugation and the absence of "in-
finite statistics".

4. Global Gauge Group and Charge Carrying Fields 184

Implementation of endomorphisms. Charges with d = 1. Endomorphisms and
non Abelian gauge group. DR categories and the embedding theorem.

5. Low Dimensional Space-Time and Braid Group Statistics 192

Statistics operator and braid group representations. The 2+1-dimensional case
with BF-charges. Statistics parameter and Jones index.

V. Thermal States and Modular Automorphisms 199

1. Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 199

1.1 Introduction. 1.2 Equivalence of KMS-condition to Gibbs ensembles for fi-
nite volume. 1.3 The arguments for Gibbs ensembles. 1.4 The representation
induced by a KMS-state. 1.5 Phases, symmetry breaking and the decomposi-
tion of KMS-states. 1.6 Variational principles and autocorrelation inequalities.

2. Modular Automorphisms and Modular Conjugation 216

2.1 The Tomita-Takesaki theorem. 2.2 Vector representatives of states. Convex
cones in 7i. 2.3 Relative modular operators and Radon-Nikodym derivatives.
2.4 Classification of factors.

3. Direct Characterization of Equilibrium States 227

3.1 Introduction. 3.2 Stability. 3.3 Passivity. 3.4 Chemical potential.

4. Modular Automorphisms of Local Algebras 245

4.1 The Bisognano-Wichmann theorem. 4.2 Conformal invariance and the
theorem of Hislop and Longo.
XIV Contents

5. Phase Space, Nuclearity, Split Property, Local Equilibrium 254

5.1 Introduction. 5.2 Nuclearity and split property. 5.3 Open subsystems.
5.4 Modular nuclearity.

6. The Universal Type of Local Algebras 267

VI. Particles. Completeness of the Particle Picture 271

1. Detectors, Coincidence Arrangements, Cross Sections 271

1.1 Generalities. 1.2 Asymptotic particle configurations. Buchholz's strategy.

2. The Particle Content 279

2.1 Particles and infraparticles. 2.2 Single particle weights and their decompo-
sition. 2.3 Remarks on the particle picture and its completeness.

3. The Physical State Space of Quantum Electrodynamics 289

VIL Principles and Lessons of Quantum Physics.

A Review of Interpretations,
Mathematical Formalism and Perspectives 293

1. The Copenhagen Spirit. Criticisms, Elaborations 294

Niels Bohr's epistemological considerations. Realism. Physical systems and the
division problem. Persistent non-classical correlations. Collective coordinates,
decoherence and the classical approximation. Measurements. Correspondence
and quantization. Time reflection asymmetry of statistical conclusions.

2. The Mathematical Formalism 304

Operational assumptions. "Quantum Logic". Convex cones.

3. The Evolutionary Picture 309

Events, causal links and their attributes. Irreversibility. The EPR-effect. En-
sembles vs. individuals. Decisions. Comparison with standard procedure.

VIII. Retrospective and Outlook 323

1. Algebraic Approach vs. Euclidean Quantum Field Theory 323

2. Supersymmetry 329
 Contents XV

3. The Challenge from General Relativity 332

3.1 Introduction. 3.2 Quantum field theory in curved space-time. 3.3 Haw-
king temperature and Hawking radiation. 3.4 A few remarks on quantum gra-
vity.

Bibliography . 349

Author Index and References 355

Subject Index 387

I. Background

I.1 Quantum Mechanics

Quantum mechanics was born in two different guises, both inspired by the prob-
lem of explaining the existence of stable energy levels in an atom. Heisenberg
had started from the idea that, in some sense, the position coordinates x (i) and
momentum coordinates p(i) of electrons should be replaced by doubly indexed
arrays xa , pni) where the indices n, m label the energy levels and the numbers
4m, AV„ are related to the intensity of spectral lines for a transition between
levels n and m. It was quickly realized by Born and Jordan that, mathemati-
cally, the double indexed schemes should be regarded as infinite matrices and
the essential relations could be simply expressed in terms of matrix multipli-
cation. Schriidinger's starting idea is seen in the title of his first paper on the
subject: "Quantization as an eigenvalue problem". Following up the suggestion
by de Broglie he assumed that classical mechanics results from a more fun-
damental wave theory by the same approximation by which geometric optics
results from wave optics. He replaced the Hamilton function of classical me-
chanics by a differential operator acting on a wave function. Its eigenfunctions
should correspond to the stationary states, its eigenvalues to the allowed energy
levels.
The equivalence of the two approaches and the underlying common abstract
structure emerged soon by the work of Schriidinger, Dirac, Jordan, [Link].
Abstractly one considers a Hilbert space 7-1 and linear operators acting on it.
To each measurable quantity of the classical theory there corresponds a self
adjoint operator acting on 7-1. The spectral values of this operator are the pos-
sible values which may occur in a measurement of the quantity in question. It
is irrelevant for the physical consequences how the Hilbert space is described
and which basis one chooses in it. One only needs a characterization of the "ob-
servables" i.e. the operators corresponding to the measurable quantities. Two
formulations using differently described Hilbert spaces 7 -11 , 7i 2 are equivalent if
there exists a unitary map U from 7-11 onto 71 2 such that the operators A( 1) , A( 2)
corespndigthamobervl tdy

A(2) = UA(1) U^ 1 . (I.1.1)

Some comments are warranted here. First, the characterization of a specific

theory proceeds from the knowledge of a classical theory which tells what the
2 I. Background

relevant observables are and how they are related. Specifically one starts from
the canonical formalism of classical mechanics. There is a configuration space
whose points are described by coordinates qk (k = 1, ... n) and to each posi-
tion coordinate there is a canonically conjugate momentum coordinate pk . The
"quantization" consists in replacing these variables by self adjoint operators
satisfying canonical commutation relations

[gk , ga^ = 0; ^k, pa] = 0; ^k r fill••

= h ska

Next, it can be shown ([Link], Rellich, Stone, [Link]) that under nat-
ural requirements (irreducibility, sufficient regularity) these relations fix the
representation of the operators pk , qk in Hilbert space up to unitary equivalence
provided that n is finite. This statement is mostly called von Neumann's unique-
ness theorem. It does no longer apply if n, the number of degrees of freedom,
becomes infinite, a situation prevailing in quantum field theories and in the
thermodynamic limit of statistical mechanics. For such systems there exists a
host of inequivalent, irreducible representations of the canonical commutation
relations which defies a useful complete classification. This problem will have
to be faced in later chapters.
Returning to quantum mechanics we must keep in mind that even there the
described "quantization" process is far from unique. On the one hand there are
interesting observables besides Pk, qk . In the classical theory they are functions
of the canonical variables. There is an ambiguity in defining the correspond-
ing functions of the noncommuting operators (choice of the order of factors in a
product). Furthermore the classical theory is invariant under canonical transfor-
mations, the quantum description under unitary transformations. The group of
classical canonical transformations is not isomorphic to the unitary group of the
Hilbert space in which the pk, q k are irreducibly represented. This means that
the result of the quantization depends on the choice of the classical canonical
system. Luckily in practice these ambiguities have not caused a serious problem
because one has usually a natural, preferred set of basic variables and simplicity
then turns out to be a good guide.
The next comment leads on to the interpretation, the philosophy and con-
ceptual structure of quantum physics. Dirac introduced the term "observable"
to indicate that these objects do not normally have any numerical value. A
number is produced by the act of measurement which "forces the system to
give a definite answer". A "measuring result" can not be interpreted as reveal-
ing a property of the system which existed (though unknown to us) prior to
the act of measurement. Take the example of a position measurement on an
electron. It would lead to a host of paradoxa if one wanted to assume that the
electron has some position at a given time. "Position" is just not an attribute
of an electron, it is an attribute of the "event" i.e. of the interaction process
between the electron and an appropriately chosen measuring instrument (for in-
stance a screen), not of the electron alone. The uncertainty about the position
of the electron prior to the measurement is not due to our subjective ignorance.
 1.1 Quantum Mechanics 3

It arises from improperly attributing the concept of position to the electron

instead of reserving it for the event.
In the general abstract scheme this interpretation is generalized as follows:
One considers a physical system whose behaviour we want to study and, on the
other hand, an observer with his instruments capable of making measurements
resulting in unambiguous phenomena (events). The need to subdivide the world
into an observed system and an observer has been especially emphasized by
Bohr and Heisenberg ( "Heisenberg cut"). Bohr points out that because "we
must be able to tell our friends what we have done and what we have learnt"
it is necessary to describe everything on the observer side, the instruments and
phenomena, in common language which may, for brevity and without harm,
be supplemented by using concepts and laws from classical physics because of
their unambiguous correspondence to common experience and the deterministic
character of the laws. Heisenberg showed in examples that to some extent the cut
between system and observer may be shifted. Part of the observer's equipment
may be included in the definition of the system. But he stressed that always a cut
must be made somewhere; a quantum theory of the universe without an outside
observer is a contradiction in itself. This is one source of a basic uneasiness
most drastically felt in quantum field theory where one would like to take the
universe as the system under consideration. Does the terminology "observer",
"measurement" mean that the cut is ultimately between the physical world and
the mind to which physical laws are not applied? More concretely, does an event
never become absolutely factual until the mind intercedes? Schrödinger tried to
emphasize this dilemma by his paradox of the cat which is neither dead nor
alive until the veterinarian looks at it. Most physicists seem to feel that "it has
nothing to do with the mind" . 1 Perhaps one has to realize that this problem is
due to an overidealization. Replacing the word "measuring result" by "event"
it is clear that with a high level of confidence a flash on a scintillation screen,
the blackening of a grain on a photographic plate, the death of a cat can (for
the purposes of physics) be considered as facts which happen irrespective of the
presence of an observer. But can one isolate events with absolute precision? Can
one describe the universe as a set of events increasing in number as time goes on?
This would necessitate the introduction of irreversibility on a fundamental level
and a revision of the concepts of space and time. In quantum mechanics this is
circumvented by limiting the class of events considered to the interaction of the
"microscopic system" with a macroscopic measuring device. Other idealizations
could be considered, depending on the regime of physics under consideration. 2
In quantum mechanics in contrast to quantum field theory the "system" is
characterized by its material content i.e. typically it is a certain number of elec-
trons, protons or other particles. One may define the system also by the set of
observables which can be measured on it or by the set of its possible states. The
term "state" suggests intuitively something like mode of existence for an indi-
vidual system. However, since the theory is not deterministic and the optimal
1I am grateful to J.A. Wheeler for this clear statement (private conversation).
2 We shall return to these questions in Chapter VII.
4 I. Background

attainable knowledge of the state allows only the prediction of a probability for
each possible result in a subsequent measurement we have to consider statistical
ensembles of individual systems in order to test the predictions of the theory.
Thus the word state really refers to the "source of the system" i.e. to the ex-
perimental arrangement by which the system is isolated and influenced prior to
the intended measurements. Alternatively speaking, it refers to the statistical
ensemble of individual systems prepared by this source. We may also say that
the state subsumes our knowledge of (the relevant part of) the past history of
the system.
In the mathematical formalism the vectors of unit length in Hilbert space
represent states. In fact they represent "pure states" i.e. optimal attainable
knowledge (finest preparation). General states, corresponding to less than op-
timal knowledge are called "mixtures"; they can be mathematically described
by positive, self adjoint operators with unit trace ("statistical operator", "den-
sity operator"). The special case of a pure state results if this density operator
degenerates to a projection operator on a 1-dimensional subspace. Then it is
equally well characterized by a unit vector which defines this 1-dimensional
subspace. Among the observables, mathematically represented by self adjoint
operators on Hilbert space, the conceptually simplest ones are the propositions
i.e. measurements which have only two alternative outcomes. We may call them
yes and no and assign the measuring value 1 to the "yes"-answer and 0 to the
"no"-answer. Then such a proposition is mathematically represented by an or-
thogonal projector, a self adjoint operator E with E2 = E. The probability for a
yes-answer in this measurement, given a state with density operator g, is given
by
p(E; g) = tr(QE); (I.1.3)
the right hand side denotes the trace of the product of the two operators g and
E. In the special case of a pure state (unit vector W) this degenerates to

p(E)=(W,E WII 2 . (I.1.4)

The discussion of general observables can be reduced to this situation. We de-

scribe it here only for observables with a discrete spectrum. The result of mea-
surement of such an observable (represented by the operator A) can be any one
of a discrete set of values ak . To each outcome a k we have a projector Ek corre-
sponding to the proposition yes for this particular outcome. These projections
are mutually orthogonal
Ei Ek = SikEi (I.1.5)
and sum up to the unit operator

Ek = ll . (I.1.6)

The observable A itself combines the specification of this orthogonal family of

projectors {Ei } with the assignment of a real number a i (the measuring result)
to each of them. The operator representing the observable is then
 1.1 Quantum Mechanics 5

A= aEEE. (I.1.7)

Given this operator the projectors Ek and the values ak are obtained by the
spectral resolution of A. The probability of result a k in a measurement on a state
A is given by tr(AEk) . The mean value or expectation value of the observable A
in the state A is given by

(A), aEp(EE; A) tr Ag. (I.1.8)

It should be noted that due to (I.1.5), (I.1.6) and (I.1.7)

A2 = a2EE, (1.1.9)

or generally, if F is any real valued function of one real variable,

F(A) = F(aE)EE . (I.1.10)

Thus the change from the operator A to the operator F(A) does not mean
that we change the apparatus; it only labels the measuring results differently,
assigning the value F(a) to the event which was formerly labeled by the value
a. Therefore it would be more appropriate to consider the Abelian algebra
generated by A rather than A itself as the mathematical representor of the
measuring apparatus. The choice of a particular operator in this algebra only
fixes a particular labeling of the measuring values.
Extending this argument Segal [Seg 47] proposed to consider as the primary
object of the mathematical formalism of quantum mechanics the algebra gener-
ated by the hounded observables, equipped with the norm topology. This leads
to the notion of a C*-algebra which will be defined in Chapter III, section 2
and used extensively afterwards. In this view Hilbert space appears only in a
secondary rôle, as the representation space of the algebra. In ordinary quantum
mechanics where one deals with a unique equivalence class of representations,
the C*-algebra formulation and the Hilbert space formulation are equivalent.
One of the most salient aspects of the quantum mechanical description is
the so called superposition principle. Probabilities are calculated as the absolute
squares of amplitudes. In the naive picture of matter waves the superposition of
two waves, corresponding to the addition of wave functions, has a simple intu-
itive meaning. The intensity, being given by a quadratic expression in the wave
function, shows the interference phenomena familiar from wave optics. In quan-
tum mechanics this superposition, the fact that a linear combination of state
vectors with complex coefficients gives again a state vector, is somewhat more
mysterious. On the one hand, if the system consists of more than one particle,
the state vector describes a wave in configuration space, not in ordinary space.
Secondly, while in Maxwell's electromagnetic theory of light the components
of the wave function are in principle measurable quantities, namely electric or
magnetic field strengths, the value of the quantum mechanical wave function
can not possibly be measured. In fact the state vectors W and .tW where a is
a phase factor (complex number of absolute value 1) represent the same phys-
ical state. Pure states correspond to rays rather than vectors in Hilbert space.
6 1. Background

This is physically extremely significant. Otherwise there could be no particles

with half integer spin. Classical electrodynamics is not the quantum theory of
a single photon but a correspondence limit of the quantum theory for infinitely
many photons. We must recognize therefore that the quantum mechanical su-
perposition principle does not have a simple operational meaning. Is it possible
to understand by some deeper lying principle why the formalism is based on
a linear space with complex numbers as coefficients? Is the linearity due to an
approximation and is the field of complex numbers just the simplest possibility?
We shall defer such questions to Chapter VII.
Let us first return to the correspondence between classical and quantum
mechanics. It is a striking fact that notwithstanding the revolutionary change
in our understanding of the laws of nature the formal correspondence between
quantum mechanical relations and those of the canonical formalism in classical
mechanics is very close. This hint was followed in the subsequent development
of quantum electrodynamics and further in quantum field theories of elementary
particles where one is way beyond the regime in which experimental information
about an underlying classical theory exists. The idea that one must first invent
a classical model and then apply to it a recipe called "quantization" has been of
great heuristic value. In the past two decades the method of passing from a clas-
sical Lagrangean to a corresponding quantum theory has shifted more and more
away from the canonical formalism to Feynman's path integral. This provides an
alternative (equivalent?) recipe. There is, however, no fundamental reason why
quantum theory should not stand on its own legs, why the theory could not be
completely formulated without regard to an underlying deterministic picture.
Sometimes Niels Bohr's epistemological arguments are interpreted as denying
this possibility as a matter of principle. Bohr emphasizes the need for a common
language in which phenomena can be objectively described. This is certainly a
prerequisite for doing physics. But, while he used the term "classical" in this
context this does not mean that one needs a full fledged deterministic mirror
theory from which the true one can be obtained by "quantization". In fact,
Bohr did not believe that a classical picture of half integer spin was useful. 3 A
simple but impressive example of a self contained quantum theory is provided
by Wigner's analysis sketched in section 3 of this chapter.
The quantum mechanical formalism is an extension of ordinary probability
theory in the following sense. In standard probability theory one has an event
space E which is a set of subsets of a (total) set Q. The elements of E correspond
to events or propositions. E forms a Boolean algebra with unit element Q; the
union of two subsets in L' (event 1 or event 2) and the intersection of subsets
(event 1 and event 2) are again elements of E. Q itself corresponds to an
event (the trivial, the certain one). On E one has a probability measure. This
measure defines in a natural way an extension of the Boolean algebra to an
Abelian W*-algebra M (this concept is defined in Chapter III). The measure
may be regarded as a normalized, positive linear form on M (a "state"). The
quantum mechanical generalization starts from the fact that one is dealing with
3 Private conversation 1954.
1.2 The Principle of Locality in Classical Physics and the Relativity Theories 7

a large family of event spaces simultanously (one for every maximal observable)
and with a large family of states. Each state defines a probability measure on
all event spaces. In this one assumes that one can measure any observable in
any state. 4
Within the set of states the process of mixing i.e. the throwing together of
several ensembles with arbitrary weights is an operationally well defined pro-
cedure, inherent in the probability concept. With any pair of states, abstractly
denoted here by w 1 , w2 one has the family of states

w),=aw1 +(1 --a)w2; 0<a<1, (I.1.11)

the mixtures of w1 and w 2 with weights a and 1—a respectively. This implies
that the set of states is a convex body IC. Linear combinations of states with
positive coefficients adding up to 1 are again states and the linearity is respected
in the probabilities for all subsequent measuring results. One may embed this
convex body in a real linear space V by considering formal linear combinations
of states with arbitrary real coefficients. The part V+ = {ak : a > 0} is a
convex cone in V. The normalization of states is provided by a distinguished
linear form e on V so that IC = {x E V+ : e(x) = 1}. The dual space of V,
denoted by V*, is the set of all linear forms on V. It also has a positive cone
V*+ = {A E V* : A(w) > 0 for all w E 1C}. The elements of V* correspond to
observables and A(w) to the expectation value of A in the state w.
The general structure expresses essentially only the probabilistic setting and
the assumption that an arbitrary observable can be measured in any state. The
specific mathematical structure of Quantum Mechanics where V+ is the set
of positive operators of trace class on a Hilbert space and V* the set of self
adjoint operators is not so easily derivable from operational principles. Much
effort has been devoted to this and we shall describe some of it in Chapter
VII. Here it may suffice to indicate the most important deviation of quantum
mechanical calculus from its classical counterpart. The convex set 1C is not a
simplex. This means that the same mixed state can arise in many different ways
as a convex combination (mixture) of pure states. It does not uniquely define
the pure components contained in it.

1.2 The Principle of Locality in Classical Physics

and the Relativity Theories

The German term "Nahwirkungsprinzip" is more impressive than the somewhat

colourless word "locality". Certainly the idea behind these words, proposed
4 Note that, to justify this, it is important to distinguish between observables and observa-
tion procedures. Many procedures yield the same observable. An observable is an equivalence
class of procedures; for instance two different measuring devices operated at different times
may correspond to the same observable due to the existence of a dynamical law. This seem-
ingly evident distinction will turn out to be important also in several other contexts. Attention
to this was drawn by Ekstein [Ek 69].
8 1. Background

by Faraday around 1830, initiated the most significant conceptual advance in

physics after Newton's Principia. It guided Maxwell in his formulation of the
laws of electrodynamics, was sharpened by Einstein in the theory of special
relativity and again it was the strict adherence to this idea which led Einstein
ultimately to his theory of gravitation, the general theory of relativity.
In contrast to the picture underlying mechanics, whereby material bodies
influence each other by forces acting at a distance, one assumes that all points
of space participate in the physical processes. Effects propagate from point to
neighbouring point. The vehicle by which this idea is put to work is the concept
of fields. Each point in space is thought to be equipped with dynamical vari-
ables, the field quantities. In electromagnetic theory they are the vectors E and
B of electric and magnetic field strength. The knowledge of these as functions
of the position in space and time is the goal in the description of a particular
physical situation. All laws governing their behaviour are strictly local i.e. they
are a system of partial differential equations.' Initial conditions and boundary
conditions may be freely varied thus allowing the great variety of possible phys-
ical situations. From the specific form of the field equations in electrodynamics,
the Maxwell equations, two further properties can be abstracted which were
raised to the status of general principles, obligatory for all theories, by Einstein
in the development of the concepts of special relativity. The first is the existence
of a limiting velocity for the propagation of effects (hyperbolic character of the
field equation). The second is the invariance under the Lorentz group.

1.2.1 Special Relativity, Poincaré Group, Lorentz Group, Spinors,

Conformal Group
The basic observation leading to special relativity is the following: the com-
parison of times at different places in space is not an objectively well defined
procedure if all laws are local and the speed of signals is limited. It needs a con-
vention for the synchronization of different clocks and there is a certain amount
of arbitrariness in this convention. This prevents us from considering space and
time as two distinct, objectively meaningful, concepts. Rather we must consider
the 4-dimensional manifold of possible pointlike events, the 4-dimensional space-
time manifold. This consequence of Einstein's argument about the relativity of
the notion of equal time was first emphasized by Minkowski. Still the distinc-
tion of past and future of such a pointlike event remains objectively meaningful.
Associated to a point x there is a division of space-time into three parts: 2 the
forward cone V+ containing all events which can be causally influenced from
x, the backward cone V - containing all events from which an influence on x
can come and the complement S of these two cones, the "space-like region" to
the point x. The latter contains all points which can have no causal relation-
ship with x (Fig. I.2.1.). One may call V+ the future, V - the past and S the
'This last conclusion was not drawn by Faraday but is due to Maxwell.
21 apologize for having used the symbol V+ in different context before.
1.2 The Principle of Locality in Classical Physics and the Relativity Theories 9

Fig. 1.2.1.

present of the event x. The boundary of V+ is formed by all possible events

which can be reached by the fastest possible signals sent from x and these are
identified with light signals in physics. The locality principle is thus sharpened
to Einstein's causality principle: No physical effect can propagate faster than
light.
In special relativity one assumes that the causal structure of space-time is
a priori globally given. There is supposed to exist a preferred class of coordinate
systems, the "inertial systems" in which the causal future of a point x = (t, x)
consists of all points s' = (t', x) with
t' — t> fix`— x!. (I.2.1)
Notation. In the context of special relativity we shall always use inertial coordi-
nates. As implied in (1.2.1) we choose units in which the velocity of light equals
unity, i.e. times are measured in cm (1 s = 3 x 10 1° cm). For a space-time point
we denote the time coordinate by x°, the three Cartesian space coordinates by
xi (ti =1, 2, 3). A Euclidean 3-vector with components x i is denoted by a bold-
face letter x. Always Roman indices ti, k, • • • will be employed in this context to
label space components; Greek indices a„ 3, ... µ, v, ... will denote both space
and time components and run from 0 to 3. The use of upper indices for the
coordinates is conventional; it is the customary usage in general relativity. We
use the metric tensor
1 0 0 0 1

gin'
__ 'o —1 0 0
0 0 —1 0
(I.2.2)
0 0 0 --1 /
and the summation convention: if an index appears in an expression twice, once
in an upper position and once in a lower position, a summation over the values
10 I. Background

of this index is automatically implied (contraction of indices). The "Lorentz

distance" square of two space-time points, replacing the Euclidean distance
square of two points in 3-dimensional space, is then given by

dx2 =guv (x'u —

XP) (xiv- xv). (I.2.3)

It is positive for time-like separation, negative for space-like separation and zero
if x and x' can be connected by a light signal. The 4-dimensional continuum
equipped with the (indefinite) metric (I.2.2), (I.2.3) is called Minkowski space
and will be denoted by M.

Poincaré Group, Lorentz Group. The invariance group of Minkowski space

is the Poincaré group, also called the "inhomogeneous Lorentz group". We may
understand invariance in two different ways. Either we consider mappings from
M to .M described in a fixed coordinate system, where the point x is shifted
to x' = gx. This is the active interpretation. g is then a diffeomorphism of M.
Or we may consider g as a coordinate transformation so that x and s' denote
the same point in two different coordinate systems (passive interpretation). In
the present context it does not matter which interpretation we choose. The
invariance group consists of all maps g which do not change the Lorentz dis-
tance between any two points, or alternatively, in the passive interpretation,
which keep the formal expression for the Lorentz distance unchanged by the
transformation. This implies that 3
u = au + Ate, (I.2.4)

where au and At are constant and A satisfies

g4v A A o = g ar). (I.2.5)

If we regard A v and guv as 4 x 4 matrices and A T denotes the transposed

matrix (I.2.5) can be written as

ATgA = g. (I.2.6)

The general element g = (a, A) of the Poincaré group consists of a translation

4-vector au and a Lorentz matrix A. The (full) Lorentz group may be described
as the set of all real 4 x 4 matrices satisfying (I.2.6). This relation implies

(det A) 2 = 1 —> det A = ±1 (I.2.7)

+1
(A00)2 =
1 + E(Ai0)2; AO_ (I.2.8)

Thus the full Lorentz group consists of four disconnected pieces depending on
the sign combinations of det A and A00 . The branch which is connected to the
identity, i.e. the branch with det A = 1, A °° > 1 will be denoted by £ and we
3 See the subsequent discussion of the conformal group.
I.2 The Principle of Locality in Classical Physics and the Relativity Theories 11

shall reserve in the sequel the term Lorentz group for this branch. The other
branches are obtained from e by reflections in time and (or) space. These will
always be separately discussed. The question as to whether and in which way
the reflections correspond to physical symmetries has been an exciting issue in
fundamental physics during the past decades.
The use of both upper and lower indices and of the "metric tensor" (I.2.2)
appears to be unduly cumbersome for computations. It is avoided mostly in the
early literature by the Minkowski trick of using instead of the time coordinate
x° the imaginary coordinate x 4 = ix ° . The (negative) metric tensor will then
have the Euclidean form c5 (a, 3 = 1, • • . 4) and one may identify upper and
lower indices. Conceptually, however, the notation is [Link] and eases
the transition to general relativity. We have to remember that in differential
geometry a vector field V°(x) maps a manifold into the tangent bundle. 4 In
other words, for a given point s E M, V4 (x) is an element of the tangent
space at x. On the other hand a "covariant vector field" W, (x), also called a
1-form, maps into the cotangent bundle. It gives a cotangent vector at x, a
linear function on tangent space, assigning to the tangent vector V the number
W(V) = WW,VI`. The metric provides a mapping between tangent and cotangent
spaces. From the vector V o we may go over to the covector VN, = g, ,V v and,
vice versa, we can define W ' = gm' W, where el' is the inverse matrix to gm,.
In special relativity g is the constant numerical matrix (I.2.2) which is its own
inverse; so go' is also given by (I.2.2).
The fields appearing in classical physics have simple transformation prop-
erties under the Poincaré group q3. Their value at a point refers to the tangent
space at this point in a way described by a certain number of upper and lower in-
dices. For a scalar field 0(x) (no indices) the transformed field under g = (a, A)
is
45' (x) = 0(g-1 x) .
For a vector field:
1/0 (x) = AiVv (9 -1 x),
for a 1-form
Wµ = A ov W„(g -1 x),
where the matrix A is the "contragredient" to A i.e.

 = 0 T ) -1

For tensor fields we have one factor A resp. A for each index e.g.
T^v
(x) = Au Av ,oTâ (g-l x).

4 The
reader unfamiliar with these concepts may find an easily readable introduction in
[Warner 1971] or [Thirring 1981, Vol.1].
12 L Background

Spinors. For many physical consequences, possibly even for the origin of
Lorentz symmetry, it is of great importance that £ is isomorphic in the small
to the group SL(2,C), the group of complex 2 x 2-matrices with determinant 1.
This connection can be most easily seen if one arranges the four components of
a 4-vector V in the form of an Hermitean 2 x 2-matrix V:
0 3 1— 2
V -- (1.2.14)
V + V 2 V° — V3
and notes that
det(V) = gi,,,,VP`VL. (I.2.15)
A transformation
V V = aVa*, (I.2.16)
where a is any complex 2 x 2 matrix with determinant 1 and a* its Hermitean
-

adjoint, leads to an Hermitean V' with the same value of the determinant. Hence
it defines a real linear transformation of the vector components
01' = A ÛV' (I.2.17)
which conserves the Lorentz square of V. Thus A in (I.2.17) is a Lorentz matrix.
To compute it we write (I.2.14) as
V = V'`a^ (1.2.18)
with
1 0 0 1
a° a1
0 1 ' 1 0'
_ 0 —i _ 1 0
a2 a3 (I.2.19)
i 0 ' 0 —1
One has
r t Va/,,,
V^` = 2 tr aa„a*a1,,.
At (a) = (I.2.20)
2
A closer study shows that as a runs through all SL (2,C) the corresponding A (a)
runs twice through L. Obviously a and —a give the same A. As a topological
group L is not simply connected. There are closed paths in L which cannot be
continuously contracted to a point. An example is a rotation by 27r which is
an uncontractible path in 2 from the identity to the identity transformation.
SL(2,C) on the other hand is simply_connected. It is the covering group of 2
and we shall denote it therefore by L. An element of Z may be regarded as
an element of 2 together with the homotopy class of a path connecting this
element with the identity. There are two classes of paths. Correspondingly we
denote by 513 the covering group of the Poincaré group. Its elements are (a, a)
with a E SL(2,C) = E. The consideration of Z, 513 instead of ,L, 513 leads to the
concept of spinors which play a prominent rôle in relativistic quantum theory.
The defining representation of SL(2,C) is given by the matrices a described
above. It acts in a complex 2-dimensional vector space. Elements of this space
are called covariant spinors of rank 1 and the components will be denoted by
Wr (r = 1, 2). Intimately related to this representation there is
I.2 The Principle of Locality in Classical Physics and the Relativity Theories 13

a) the complex conjugate representation a Ti,

b) the "contragredient" representation a ,Q = (aT)-1,

c) the representation a ,Q.

Van der Waerden suggested that spinors transforming under a) should be de-
noted by a dotted index : Wr , spinors transforming under b) by an upper index :
wr , those transforming under c) by an upper dotted index. This reflects the fact
that the representation b) is equivalent to the defining representation, namely
one has
= EaE -1 , a (I.2.21)
or, with fully written indices

f3
, ,

s = E rr ar E s' s e (1.2.22)

with
Ers = —Ers _ 0 1
--1 0 '
(I.2.23)

So E may be regarded as a metric tensor in spinor space. The representations a

and ax are inequivalent but â and 73 are again related by E.
Taking tensor products of these representations one comes to spinors of
higher rank, having several dotted and undotted, upper or lower indices. One
finds that all equivalence classes of finite dimensional irreducible representations
of SL(2,C) are obtained by considering spinors with n dotted, m undotted lower
indices which are totally symmetric with respect to permutations of each of the
two types of indices. If n + m is even such a spinor belongs to a true (single
valued) representation of 2, it corresponds to a Lorentz tensor of some rank.
The simplest case may be read off from (I.2.14), (1.2.16). A 4-vector Vo is
written in spinorial notation as Val. The next simple cases, symmetric spinors
with two undotted or two dotted indices correspond to (complex) self dual skew
symmetric tensors F0,, + icµ„o°F° (for the notation see equation (I.2.64) be-
low). For an excellent brief review of spinor calculus see Laporte and Uhlenbeck,
[Lap 31] and [van der Waerden 1932].

Conformal Group. 5 The metric (1.2.2) combines two aspects. On the one
hand it defines the causal structure (Fig. I.2.1.); in addition it defines distances
in space-like and time-like directions. One may separate the two aspects and
consider the causal structure as more fundamental. If the rest masses of all
particles were zero only this aspect would remain. Such an idealization may be
of interest as an asymptotic description of the high energy regime and possibly
also if one wants to explain the origin of masses.
The group which conserves only the causal structure, not the metric, is the
conformal group. We want to determine the transformation formulas between
5 This will be used only in a few sections and can be skipped at a first reading.
14 I. Background

two coordinate systems in which the light cones are characterized by the van-
ishing of the right hand side of (I.2.3) with gm„ given by (I.2.2). Let us look
at a small neighbourhood of a point x and an infinitesimal coordinate transfor-
mation.
e — ^^ e + e u^`(x} e —r O.
The line element from x to x + dx

ds 2 = gi „dxPdx„

changes by
{a „
6ds 2 = c (—de dx„ + axA dxPdx^` _ ^ Gm„dxF`dx„ ,
g14„ axe

aum au„
G„ = u = gu „u . (1.2.24)
ax„ +
We have used the fact that g„ is constant. If the form of the light cone equation
shall not change by the coordinate transformation we must have

Gm „(x)dx'`dx „ = 0 whenever gi „dx'`dx„ = 0. (I.2.25)

Thus Gm.„ must be a (possibly position dependent) metric tensor which has the
same light like directions as gm„. This means that

Gov (x) = a(x)9 v . (I.2.26)

We can eliminate a by contracting with g

a = 4 G14„g°„
and get the following condition for the displacement field u of the infinitesimal
transformation:
1 au au 1
il eu e=
ax„ + ax^ — 4 g^„a
O. (I.2.27)
2
The general solution of this equation can, for smooth functions, be easily found.
If one makes a Taylor expansion around the point x = 0

u^(x} = a^li + a^vi x„ + a^v^p x „xP + • • (I.2.28)

one notes that the homogeneous polynomials of different degree decouple and
(I.2.27) gives separate conditions for each a

a ( Û) ... + aÛué a,.. _ l g „a{ )A ...

(I.2.29)

with acv ... totally symmetric in the n-1 last indices. For n = 1 this gives no
restriction for a (l ). For n = 2 one gets
I.2 The Principle of Locality in Classical Physics and the Relativity Theories 15

aû2) = w + agµ v with ww„ _ —w„µ . (I.2.30)

For n = 3 one finds

_Ave = gµvice + g µe cv — guet ,„. (I.2.31)
For n > 3 there is no solution (if the dimensionality of space-time is higher
than 2). This follows from the permutation symmetry of the last indices of a(n).
The admissible form of u is then

uµ (x) = aµ + gµ a wav xv + d xµ + 2xµ c a xA - xa xAcµ, (I.2.32)

where aµ, d, cµ are arbitrary constants and w is antisymmetric. The family of dif-
ferential operators uµ(x)ô µ are the generators of a 15-parametric Lie group, the
conformal group of 4-dimensional space-time with the causal structure (L2.1). 6
aµ, w v generates the Poincaré transformations Thepartwi ms
(I.2.4), the part with d generates dilations, the one with c µ the "proper confor-
mal transformations", also called "conformal translations"

— (x, x)cµ
x'Er,—
___. (I.2.33)
1 — 2(x, c) + (x, x)(c, c) .
These transformations become singular on the submanifold where the denomi-
nator vanishes. Therefore they do not give global symmetries of M but can be
a lim- definphyscalo difemrphsnutablegiofr
ited range of the parameters c. One can, however, compactify M so that the
conformal transformations act as diffeomorphisms defined everywhere in the re-
sulting space. The conformal group plays a rôle in the high energy asymptotics
of quantum field theory.
Computations involving the conformal group are greatly simplified using
a representation of the group by linear transformations. One possibility is to
take a 6-dimensional real space (coordinates ea , a = 0, 1, • • • 5) equipped with a
metric
_ +1 S for a=0,5
gap — —1 yap for a = 1, 2, 3, 4.
The group of pseud o- orthogonal transformations SO(4,2)

e'a = M5 e^,
with matrices M satisfying

gaWW1 = gya; det M = 1, (I.2.36)

yields such a representation. Its relation to conformal transformations in

Minkowski space is established as follows. M E SO(4,2) transforms the cone
6 Note that in 2-dimensional space the argument that a(n) = 0 for n > 3 is no longer true.
In this case the conformal group has infinitely many parameters. It is the group of analytic
functions of a complex variable with nonvanishing derivative in the domain of definition.
16 L Background

C (6) = te : ga,o re = 0} (I.2.37)

into itself. Defining

xt` = (e4 + e5)-leµ, µ = 0 1, 2, 3 (I.2.38)

the transformation (I.2.35) induces a transformation x — ^^ s' in IR4 which turns

out to be a conformal transformation. Introducing

ri = e4 + e5 , Yv = e4 — ^ 5 r (I.2.39)

one gets from (I.2.37), (I.2.38)

(x, x) = gµ„xµx „ = i[ . (I.2.40)

The various subgroups:

(i) The 6-parametric subgroup leaving e4 and E5 fixed

( AA,
M = 1 , A E ,e (I.2.41)
1

yields the homogeneous Lorentz transformations;

(ii) the 4-parametric subgroup

A = e t` + i aµ ; i? = r-1; = s + 2(e, a) + ri(a, a) (I.2.42)

yields the translations

x'A=xt4 +O.
(Note that the transformation law of ic follows from the others by (I.2.40));

(iii) the 4-parametric subgroup

e'l2 = el' — rtcl-'; k' _ ic; iii = r] — 2 (e', c) + ,c(c, c) (I.2.43)

yields the "conformal translations” (I.2.33);

(iv) the dilations

x ' ° = axl`
result from
e 'µ = e ^; 1£' = AK; 7/' = a -1 7 7• (I.2.44)

By a conformal transformation the "Lorentz distance square" between two

points x 1 , x2 E M is changed by a conformal factor
I.2 The Principle of Locality in Classical Physics and the Relativity Theories 17

(x; - x2) 2 = N(X1) -1 N(X2) -1 (X1 — X2)2, (I.2.45)

where, of course N = 1 for Poincaré transformations, N = A -1 for dilations.

For the "conformal translations" (1.2.43) one finds that N is the denominator
in (I.2.33)
N = 77-1 71. (1.2.46)
Instead of realizing the conformal group by pseud o- orthogonal transforma-
tions in a 6-dimensional real space one may also use a realization by pseudo-
unitary transformations in a 4-dimensional complex space. This is an exten-
sion of the spinor calculus for the Lorentz group to the case of the confor-
mal group. The 2-component spinors are replaced by 4-component bispinors
Wa (a = 1, • • • 4), familiar from Dirac wave functions at a point. Following the
suggestion by Penrose we shall call them twistors. 7 The conformal group is iso-
morphic in the small to the group SU(2,2), the group acting on twistor space

( _
which leaves the (indefinite) Hermitean form

OM (0 1 7° 10 (I.2.47)

invariant. Here ( 10) denotes the (naive) positive definite Hermitean

scalarproduct and 'y° is an Hermitean matrix with two eigenvalues +1 and
two eigenvalues —1. The Lie algebra of SU(2,2) is the algebra of the 15 Dirac
matrices
,yµv = Z,yu,yv(µ ,y (I.2.48)
-y"; v); 514 = 2`y5-e; ,y5 = -y°-y 1 -y2 -y3 ,

generated from the 7A defined in section 3.3 below. The correspondence to

SO(4,2) is established by identifying the generators of
- rotations in the e5-eA plane with -y'
- Lorentz transformations with -yAv
- rotations in the C4 -eA plane with -y 5^`
- dilations (rotations in the C4 -e5 plane) with -y 5 .

1.2.2 Maxwell Theory

"War es ein Gott, der diese Zeichen schrieb?" This quotation from Goethe's
Faust was Boltzmann's reaction to Maxwell's "Treatise on Electricity and Mag-
netism". The enormous wealth of physical phenomena predictable from a few
basic equations borders indeed on the miraculous. These equations (in pre-
relativistic vector notation)

as
+ curl E = 0; div B = 0, (I.2.49)

7 See [Pen 67].

There is an essential difference in the transformation law between Dirac
wave functions and twistors. In Dirac theory the translations do not change the index a; in
the case of twistors the translations (and conformal transformations) act on the index a.
18 I. Background

curl H — = 4rr j ; div D = 4irP, (I.2.50)

^
D =EE; B=µH (I.2.51)
connect the electric resp. magnetic field strengh E, B with the electric current
density j and charge density g. In the absence of polarizable media we may
choose units so that D is identified with E and H with B.
Neither Maxwell nor Boltzmann could know that these equations contain
the germs for developments in theoretical physics far transcending electromag-
netism. We have already mentioned Lorentz invariance and special relativity.
It is miraculous how these equations simplify if one uses the 4-dimensional
Minkowski space notation. The electric and magnetic field strengths combine
to a skew symmetric tensor (2-form)

0 -E2
-E1 - E3
El 0 B3 - B2
Fu„ = E2 - B3 0 Bl '
(I.2.52)
E3 B2 - B1 0
the charge and current densities combine to a 4-vector

9 t' = (P,j). (I.2.53)

The homogeneous Maxwell equations (I.2.49) become

attFecy + ;Fop + aUFtte = 0. (I.2.54)

They can be formally solved by introducing a 1-form A A (x), the vector potential,
putting
Fp„=[Link] ---- a„Ap. (I.2.55)
The inhomogeneous Maxwell equations (I.2.50) become
a F1`v = —47r (I.2.56)

Charge conservation which is locally expressed by the continuity equation

ay
-5—
t = —div j

becomes
aµjt`= 0 (I.2.57)
and is read off immediately from (I.2.56) due to the antisymmetry of Fi".
It is tempting at this stage to remark that the equations fit naturally into
differential geometric concepts, specifically the theory of integration over sub-
manifolds (of Minkowski space) and the associated de Rham cohomology. Con-
sidering a piece of a p-dimensional submanifold characterized in parametric form
by x = x(A l , • • • Ap) the oriented "surface element" 8
8 The determinant in (I.2.58) is the Jacobian of the p coordinates in question with respect
to the parameters.
I.2 The Principle of Locality in Classical Physics and the Relativity Theories 19

aek
der-AP = det dA i • • • dA p , (I.2.58)
a,
is invariant under change of parametrization and it transforms like a contravari-
ant tensor of rank p under coordinate transformation. It is totally skew sym-
metric in its indices. Therefore a totally skew symmetric covariant tensor field
XAl ...p, p assigns in an intrinsic (coordinate independent) way and without ref-
erence to the metric a measure on each p-dimensional submanifold of M by
the integral f X i,,,...[Link] . This is the reason why such a tensor is called a p-
form. One finds that the curl-operation, the antisymmetric differentiation, leads
(again in an intrinsic way) from a p-form to a (p + 1)-form, denoted by dX
= Antisymmetr• aX... (I.2.59)
Furthermore this operation, called the coboundary operation in mathematics,
has the property
d2 = 0. (I.2.60)
One recognizes that the first Maxwell equation means that the 2-form F satisfies

dF= 0 (I.2.61)

and that the Ansatz (I.2.55) which reads now

F=dA (I.2.62)

solves (I.2.61) due to (I.2.60). A p-form X satisfying dX = 0 is called a cocycle.

If X = dY then X is called a coboundary which certainly implies that X is also
a cocycle. The question as to whether each cocycle is a coboundary is the basic
question of cohomology theory. For a contractible region, a region which can be
shrunk to a point by continuous deformations, the answer is "yes" (Poincaré's
lemma). Thus (I.2.55) gives a general solution of (1.2.54) as long as we do not
have any holes in Minkowski space or boundary conditions at infinity to worry
about. The potential AA is not uniquely determined by (I.2.62); there remains
the freedom of adding an arbitrary cocycle which locally can be written as do
where 0 is an arbitrary scalar field. This innocent observation contains the germ
of a further very important extrapolation from Maxwell theory, the principle of
local gauge invariance.
Turning now to the second Maxwell equations we may note that it would
be more natural to consider the current density as a 3-form because it is by
integrals over 3-dimensional submanifolds that we obtain from it the physically
interesting quantity: the electric charge. This can be done by defining

Jµve (
S
) ° £µvAoi ° (
X
)

where E is the Lorentz-invariant Levi-Civita numerical tensor

/+1 if pupa. is an even permutation of 0123
EA,,g, = —1 if µ1/ pa. is an odd permutation of 0123
0 if any two indices are equal.
20 I. Background

One can use € to define to each p-form X in 4-dimensional space-time a (4 — p)-

form called its Hodge dual and denoted by *X. For instance J = *j if we
consider j as the 1-form jµ = gµ„j„, and

(*F)µ„ _ ^ EY„^ F0°. (I.2.65)

The second Maxwell equation reads then

d*F= J (1.2.66)
and the continuity equation

dJ = 0. (I.2.67)
On a formal level one might consider the 2-form G = *F as a potential for the
current density J since J is the coboundary of G. This determines G from J up
to a cocycle. Barring cohomological obstructions this arbitrariness can precisely
be lifted by requiring

d*G - dF.- 0, (1.2.68)

which is the first Maxwell equation.
On this level one could say that the Maxwell equations result from Poincaré
invariance and charge conservation purely by a convenient definition of a "charge
potential” G. However, the field F is not merely a computational device. It is
an observable which can be measured by its effect on the motion of the charges.
A point particle of charge q moving with velocity y in an electromagnetic field
experiences the Lorentz force

K = q(E + y x B). (I.2.69)

In relativistic kinematics the trajectory of such a particle is a world line in
Minkowski space which is conveniently described by giving the position xm as a
function of the proper time T (defined by dr 2 = gµ„dx1dx„). The 4-velocity

de(r)
uµ (I.2.70)
dr
—

is a 4-vector having the constant Lorentz square 1 along the trajectory. The
relativistic adaptation of Newton's second law with the Lorentz force acting on
the particle is

dTs=
md2 — gFµ„ (x(^r))u „(^r). (I.2.71)

To obtain a complete theory one must add to the Maxwell equations a model
for the current density and describe the effect of on on the development of j.
If one as sumes the current to be built up from point particles, numbered by an
index i, then
I.2 The Principle of Locality in Classical Physics and the Relativity Theories 21

jA(x) = E qi
f u(r)4(x - x i (Ti))drti (I.2.72)

and for each particle we have the equation of motion (1.2.71). But this does not
give a well defined theory because the insertion of the current density (I.2.72)
into the Maxwell equations gives a field which is singular on the trajectories of
the particles. In (I.2.71) we get an infinite or undefined self force on each particle
resulting from the electromagnetic field generated by it. We cannot just claim
that this self force is zero because an accelerated charge radiates electromag-
netic energy and the energy balance demands that there is a radiation reaction
force on the particle. In a beautiful analysis Dirac has extracted the relevant
finite part of the self force of a point particle [Dir 38]. This may be considered
as a "renormalization" prodedure in the classical theory. It has, however, some
defects. In the form given by Dirac the electromagnetic field is split into an "ex-
ternal field" and a "self field". The equation of motion of the particle is a third
order differential equation in which the external field at the position of the par-
ticle enters. A proper formulation of the initial value problem replaces this dif-
ferential equation by an integro-differential equation which determines the orbit
from the incoming electromagnetic field and the initial (straight line) asymptote
of the world line of the particle. This equation has solutions only if the incoming
field is not too rapidly varying [Haag 55b], [Rohr 61]. For further discussion of
these problems see for instance [Rohrlich 1965], [Thirring 1981, Vol.2]. From
the present day point of view it appears that the point particle model for the
current, while adequate for many problems, is not a good starting point for
a fundamental theory. There are not only the mentioned defects but we have
learned that electromagnetism is closely tied to the principle of local gauge in-
variance which will be discussed later. It demands that the charge should be
carried by a complex field and that j is a quadratic expression in this matter
field, as it is in Schrbdinger's matter waves. But then we are leaving the province
of classical theory.

1.2.3 General Relativity 9

In special relativistic electrodynamics the principle of locality is still not fully

implemented. There remains the rigid metric structure of Minkowski space as-
sumed to be known a priori, fixing the causal order between far apart points
irrespective of the physical situation in between. Also the theory of gravitational
forces existing around 1910 did not fit naturally into the picture of a local field
theory in Minkowski space. The clue to remove both defaults, leading to the
development of the general theory of relativity, was recognized by Einstein to
lie in the equivalence of inertial and gravitational mass. It suggests that the
transition from one coordinate system to another one, arbitrarily accelerated
9 Our purpose here is only to give an outline of basic concepts and ideas underlying gen-
eral relativity. They will only be used in Chapter VIII of this book. For an easily readable
introduction we refer to [Bergmann 1976], for a critical survey of the physical background,
computational techniques and applications to astrophysics to [Weinberg 1972].
22 I. Background

with respect to the first, cannot be physically distinguished from a change of

the gravitational field. Therefore the existence of a preferred class of coordinate
systems, the inertial ones, should not be assumed.
The mathematical structure needed to incorporate these physical ideas is
Riemann's geometry. It refers the concept of metric to the infinitesimal neigh-
borhood of each point, more precisely to the tangent spaces of the points of
the space-time manifold. In a coordinate system, a tangent vector z at a point
x may be characterized by the derivative of the coordinates of a parametrized
curve x(s) passing through the point;

dxµ
z^ = ; x(0) = x.
ds s = 0
The scalar product of two such tangent vectors at x is defined as

(z1 , z2) = 9w , (x)zi z2

where the "metric tensor" gµ„ does not have the a priori given form (I.2.2) but
depends on the point x. One has a metric field g,,,,,(x). Only the signature of the
metric is the same as that of (I.2.2) at every point. One has time-like vectors
(positive length square). The orthogonal complement of a time-like vector is
a 3-dimensional subspace on which the metric is negative definite; it contains
only space-like vectors. Now the notions of tangent bundle, cotangent bundle
become really relevant because we have no natural identification of tangent
spaces at different points. Nor do we have a preferred class of coordinate systems
related by linear transformations. Therefore all relations should be expressed in
a way valid in any coordinate system and we have to handle general coordinate
transformations, x

= x A (x°, . . . x 3) (I.2.75)
where the four functions on the right hand side may be quite arbitrary apart
from being smooth and such that (I.2.75) is invertible i.e. solvable for the xt'
in terms of x'A within the region to which the coordinate systems are applied.
It has been disputed whether the requirement of general covariance of all
laws is a physical principle or merely an esthetic demand which can always
be met irrespective of the physical content of the theory. It is empty indeed if
nothing is said about the nature of the quantities appearing in the theory. But
it is very restrictive if we know that the theory deals with a certain number
of fields which are related in a specific way to the tangent spaces. Thus, a
vector field VA(x) assigns to each point of the manifold a vector in the attached
tangent space; it is a "section in the tangent bundle". Under the coordinate

V(x) r r
change (I.2.75) it transforms as

(I.2.76)
V v(x) "

A covector field X A (x), a section of the cotangent bundle, transforms as

1.2 The Principle of Locality in Classical Physics and the Relativity Theories 23

U
= Xu(x). (1.2.77)
Xµ(x) aX,µ

The matrices A and A in (I.2.10), (I.2.11) are replaced respectively by the

Jacobian matrix of the transformation (1.2.75) or its inverse. For tensor fields
with several upper or lower indices the transformation law is as in (I.2.13)
replacing the Lorentz matrices by the respective Jacobian matrices.
Since the laws shall be formulated as differential equations for fields and
since the field components at a point have to be related to a basis in the tangent
space of the point we must be able to compare vectors in the tangent spaces of
infinitesimally close points even though there is no natural identification of the
tangent spaces of different points in general. The necessary additional structure
is the notion of parallel transport or affine connection. In a coordinate system
it is described by connection coefficients C. The change of a component of a
vector V at x under parallel transport to x + Sx is of the form

5Vu = —C,, eVUSxe, (1.2.78)

(the minus sign is conventional). The splitting

Cie=rÛe+T%, (1.2.79)
where F is symmetric, T antisymmetric in the lower indices has an intrinsic
meaning, it does not depend on the coordinate system. T is called torsion and
transforms as a tensor under coordinate changes. P is not a tensor but obeys
the inhomogeneous transformation law

ax" ax° axT e a^ À 32 xe

_
r,^ = (I.2.80)
uv axe aft' ax^v D oT + axe a^ ^4x'U .
In the standard theory of general relativity the torsion T is assumed to vanish.
F may be interpreted as the gravitational field. The motion of a point particle
which is under the influence of no other than gravitational forces is assumed to
be an isoparallel i.e. the orbit is generated by parallel transport of the velocity
4-vector uI& = dxµ/dT where T is the proper time (dT 2 = gi ,,,dedxv) :

dui` (T)
= --r^U^^^l'. (1.2.81)
dT
Because of the equivalence of inertial and gravitational mass the particle mass
does not appear in this equation of motion and, moreover, one can always choose
a coordinate system so that the inertial forces balance the gravitational forces
at a chosen point which means that the apparent acceleration and hence r
vanish at this point. On the side of the mathematical algorithm this possibility
arises from the fact that P is not a tensor but has an inhomogeneous term in
its transformation law.
It is clear that the connection coefficients must bear a relation to the metric
field because parallel transport should respect the metric structure; the scalar
product of two vectors should not change under parallel transport. It is, however,
24 I. Background

surprising that this requirement determines f completely in terms of the metric

field. One finds

rev = 2 g e (avgeA + a0QU — aeg,I) • (I.2.82)

H. Weyl called this relation "the fundamental fact of differential (Riemannian)

geometry".
The affine connection allows us to define in an intrinsic way the notion of
differentiation of vector and tensor fields. We must, in a coordinate system,
compare the field components at the point x + Sx with those resulting from
parallel transport of the field at x to this neighboring point. The resulting
concept of covariant differentiation gives for a vector field the definition

De V°(x) = 8 0 V` (x) + F%(x)Vv(x). (I.2.83)

This allows a very simple transcription of laws from special to general relativity;
for instance electrodynamics in the presence of a gravitational field results if
we replace in Maxwell's equations the partial derivatives a iL by the symbols
of covariant differentiation DA . One may note here that in the curl-operation
(the coboundary operation) the connection coefficients drop out. Thus, if the
Maxwell system is written using only the d-symbol, then one does not need
to know T for these relations. But not all important differential relations in
physics can be written in terms of the curl-operation. An important example
is energy-momentum conservation which, locally, is expressed as a continuity
equation for the energy-momentum tensor Ti,,,. In general relativity this reads

DATA' = 0. (I.2.84)
The replacement of ordinary differentiation by covariant differentiation relates
to the fact that TA,, includes gravitational contributions.
A further essential concept tied to the existence of an affine connection is
curvature. Transporting a vector V parallel around an infinitesimal loop may
lead to a change of its direction by an infinitesimal Lorentz transformation:

511A(x) = R Û x (x)V'du° ° (I.2.85)

where du is the surface element of the loop. The coefficients R ÛQo form a tensor,
the Riemann curvature tensor. It is, by definition, antisymmetric in the indices
o- since du is antisymmetric and it is an infinitesimal Lorentz matrix in the
first two indices which means that

R,‘,,2,, - DavAR ve, (I.2.86)

is antisymmetric also in Av. There are further general properties of RÀ,,, e° which
follow from its construction viz. the symmetry under the exchange of the first
pair with the second pair of indices

R,N,,, gar = Re°av, (I.2.87)

 1.3 Poincaré Invariant Quantum Theory 25

the cyclicity in the last three indices

RÀvecr + RacrVQ + Raecry = 0, (I.2.88)
and the Bianchi identities
DARAVeer + DQRaJW + DORAvQpt = 0. (I.2.89)
From the Riemann tensor one obtains by contraction the symmetric Ricci tensor

Rµv = gAeRAAgv) (I.2.90)

and, by further contraction, the curvature scalar

R= Ri,,v . (I.2.91)
The Bianchi identities yield then the crucial relation

D (R1L1 — 2 g^`vR ) = 0. (I.2.92)

The remaining hard problem is to replace Newton's law of gravitation

div F = 4irGg (I.2.93)

(g is the mass density, F the gravitational field, G the gravitation constant).
Einstein solved this, replacing the mass density by the energy-momentum ten-
sor Tw (which is a function of the local fields, provided that one has a field
theoretic model of matter) while the left hand side of (I.2.93) is replaced by
a tensor obtained from the curvature. The choice of this is severely limited by
the fact that due to (I.2.84) the covariant divergence of it has to vanish. One
possible choice is suggested by (I.2.92) and this fits with the non relativistic
approximation (I.2.93) when the gravitational force is related to the connec-
tion coefficients by comparing (I.2.81) with Newton's equations of motion. One
obtains then the Einstein equations

1
Ri,,, — igmv R = (1.2.94)
Apart from an additional, so called "cosmological term" ag m,v the left hand side
of (I.2.94) is the only tensor of this type with vanishing covariant divergence
which can be formed from the metric tensor and its first and second derivatives
without using higher powers of the curvature. In this the Einstein equations are
the simplest, most natural possibility allowed in the setting.

I.3 Poincaré Invari ant Quantum Theory

I.3.1 Geometric Symmetries in Quantum Physics.

Projective Representations and the Covering Group

It was stressed in section I.1 that "states" and "observables" in quantum the-
ory have their physical counterparts in instruments which serve as sources of
26 I. Background

an ensemble or detectors of an event. Within the regime of special relativity

we may distinguish two different types of information in the description of an
instrument. On the one hand there is the intrinsic structure of the appara-
tus, given essentially by a workshop drawing. On the other hand, we have to
specify where and how this equipment is to be placed within some established
space-time reference system in the laboratory. We can put its center of mass
in different positions, orient its axes, trigger the apparatus at a time of our
choice, let it rest in the laboratory or move it with a constant velocity. There
are altogether 10 parameters which specify the "placement". They correspond
to the 10 parameters of the Poincaré group. Poincaré invariance of the laws of
nature means that the result of a complete experiment does not depend on the
placement. Specifically, if both the source and the detector are subjected to the
same shift of placement, then the counting rates will remain unaltered.
Suppose we have an ideal source which prepares an ensemble in a pure state.
For a specified placement of this source in our space-time reference system the
prepared state is mathematically described by a ray in a Hilbert space H, say
by W = {A W}, where W is a vector in 7-1 and a an arbitrary complex number.
Similarly, if we have an ideal detector, giving a yes-answer in a pure state
described by the ray 0 and the answer no in the orthogonal complement to 0,
then the probability of detecting an event in this set up of source and detector

_
is given by the ray product

1")12W)
(0,0)(W,
,

where (0, W) denotes the scalar product of Hilbert space vectors. Shifting the
placement' of the source by a Poincaré transformation g = (a, A) we denote
the state prepared by this source by W9. Correspondingly, shifting the detector
by the same element g E 413 leads from P to cP9. Symmetry means that the
probability of an event is unaffected if both source and detector are shifted by
the same group element:
[ 9 IW9l = (I.3.2)
Keeping g fixed and letting ! run through all states (here assumed to correspond
precisely to all rays of 7-0) we get a map Tg of the rays corresponding to the
shift g E 413
T9 -1I19 . (I.3.3)
It must leave the ray product of any pair of rays invariant

[Tg4;$IT9 l = [4;$I l
^ ^
(I.3.4)

and it must satisfy the composition law of the Poincaré group

T9 Tg^ = Tgg ^ (I.3.5)

'We understand the Poincaré transformations here in the "active" sense i.e. not as a change
of the space-time reference system but as a change of placement of hardware in the reference
system.
 I.3 Poincaré Invariant Quantum Theory 27

In particular (1.3.5) implies that Tg has an inverse. So Tg maps the set of all
rays onto itself.
A ray transformation may be replaced in many different ways by a trans-
formation of the vectors of the Hilbert space. All we have to ask for is that the
image vector shall lie on the image ray. However one has the following important
fact, known as the Wigner unitarity-antiunitarity theorem.

Theorem 3.1.1
A ray transformation T which conserves the ray product between every pair of
rays of a Hilbert space can be replaced by a vector transformation T which is
additive and length preserving, i.e. which satisfies

T (W1 + W2 ) = TW1 + TW2, and Jf TW lit =JJ W 11 2 (I.3.6)

This vector transformation is determined uniquely up to an arbitrary phase

factor (complex number of modulus 1) by T and the requirements (I.3.6). It
satisfies either
TaW = aTW (1.3.7)
or
TaW= âTW, (I.3.8)
where a is a complex number, â its complex conjugate. In the first case T is a
linear, unitary operator, in the second case it is called antilinear and antiunitary.

We shall not prove it here. 2 It is closely related to the "fundamental theorem

of projective geometry". In the case of a continuous group the possibility of
antiunitary operators is excluded because each group element is the square
of another one and the product of two antiunitary operators is unitary. We
conclude: To every Poincaré transformation g = (a, A) there corresponds a
unitary operator Ug - U(a, A) acting in the Hilbert space, determined up to a
phase factor eia. The multiplication law (I.3.5) becomes

U91 U92 = 71U gig2 (1.3.9)

Here n is a phase factor which may depend on 91 and 92 . Thus U is a representa-

tion up to a factor, a projective representation of the Poincaré group. Since there
remains the freedom of changing every operator Ug by an arbitrary phase factor
one may use this to simplify the phase function nj. One finds [Wig 39], [Barg 54]:

Theorem 3.1.2
Any ray representation of the Poincaré group can, by a suitable choice of phases,
be made into an ordinary representation of the covering group T.

Thus the factor n can be eliminated in (1.3.9) if we understand by 9k

elements of ,3 rather than f3 and it is reduced to a sign ±1 if we consider the
Poincaré group itself.
2 For a proof see the book by Wigner, appendix to chapter 20 [ Wigner 1959].
28 I. Background

To summarize: Poincaré invariance in quantum theory means that we have

a unitary representation of 513, describing the effect of Poincaré transformations
on the state vectors of Hilbert space. The occurence of the covering group results
from the fact that states correspond to rays rather than vectors of 71.

I.3.2 Wigner's Analysis of Irreducible Unitary Representations

of the Poincaré Group

The building blocks of general unitary representations of 43 are the irreducible

ones i.e. those which act in a Hilbert space not containing any (proper, closed)
subspace which is transformed into itself. These representations should yield
the state spaces of the simplest physical systems in relativistic quantum the-
ory and it is a purely mathematical task to classify them all (up to unitary
equivalence). In a fundamental paper E.P. Wigner recognized these facts and
largely solved the classification problem [Wig 39]. It is curious that this paper
remained practically unknown or unappreciated in the physics community for
more than a decade. We sketch the essential line of argument and the results
without aspiring to mathematical rigour. For more thorough discussions of this
subject the reader is referred to the books [Hamermesh 1964], [Ohnuki 1988].
Let U(a) denote the representor of a translation by the 4-vector a u and U(a)
that of an element a E Z. Hopefully the distinction by the type of argument
of U will be sufficient to avoid confusion. The Roman character stands for a
translation 4-vector, the Greek a for a unimodular 2x 2 matrix which determines
a homogeneous Lorentz transformation A(a).
The translation subgroup is commutative. We can write

U(a) = ezPHa" (I.3.10)

where the infinitesimal generators Pi, are four commuting, self adjoint operators.
The multiplication law in i3 gives the relation
U(a) -1 P" U(a) = 1l v(a)Pv; (I.3.11)
In words: the operators P4` transform like the components of a 4-vector under
the homogeneous Lorentz group. Since they commute they have a simultaneous
spectral resolution. The spectral values are a subset of a 4-dimensional space
(p-space) and we may represent a general vector W E 7-( by the set of spectral
components Wp
W _ {Wp }. (I.3.12)
Wp is a vector in a degeneracy space 7 -(p . If the simultaneous P-spectrum were
discrete 7-129 would be a subspace of 7-1. In the case of continuous spectrum,
which is the case of primary interest, we may consider 7-( p as an infinitesimal
(improper) subspace of H. The scalar product in N becomes

(W'iW) = f(W;Iw)d,) (1.3.13)

where the bracket under the integral denotes the scalar product of the compo-
 1.3 Poincaré Invariant Quantum Theory 29

nents in N and dµ is some positive measure in p-space. In other words, we rep-

resent 71 as the direct integral of the spaces lip with respect to the measure dp.
According to (I.3.11) the operator U(a) maps 7-h onto 7-Ç where p =
A(a) p. Let us call the set of all points in p-space which arise from a single point
p upon application of all homogeneous Lorentz transformations the "orbit of p".
Clearly then all spaces 'Hp with p-vectors lying on one orbit will be mapped on
each other by the action of the operators U(a) whereas spaces with p-vectors
on different orbits will not be connected by any U(a) or U(a). Therefore in an
irreducible representation the P-spectrum has to be concentrated on a single
orbit. This gives a first division of the irreducible representations into the fol-
lowing classes:

class orbit
m+ Hyperboloid in forward cone; p2 = m2 and p° >0.
0+ Surface of forward cone; p2 0 and p° >0.
00 The single point pu = 0.
Space-like hyperboloid; p2 = ^ k 2 (!û real) .
m_ Hyperboloid in backward cone; p2 = m2 and p° < 0.
0_ Surface of backward cone; p2 0 and p° < 0.

Of course p2 = g„p4`pM = (p0 ) 2 — p2 . We shall be concerned only with the

classes m + and 0+ for the following reason. The operator P°, considered in
its rôle as an observable, is interpreted as the total energy of the system and,
correspondingly, P' are the components of the linear momentum. 3 One of the
most important principles of quantum field theory, ensuring the stability, de-
mands that the energy should have a lower bound. This is not the case in the
last three classes. In class 0 0 all states have zero energy-momentum, all states
are translation invariant. If they are also Lorentz invariant we have the triv-
ial representation appropriate to the vacuum state. Other possibilities in 0 0
which carry a non trivial unitary representation of the Lorentz group have only
marginal physical interest, though, mathematically, the analysis of this class,
carried through by Bargmann and by Naimark, is very interesting. See [Barg 47]
and the books [Naimark 1964], [Ohnuki 1988].
Since pp is interpreted as energy-momentum the classification parameter m
(resp. 0) of the first two classes is, of course, the rest mass. The measure dp is
fixed (up to an arbitrary normalization factor) by the requirement that it must
be Lorentz invariant and concentrated on a single hyperboloid. In 4-dimensional
notation it is4
3 Here we go over to natural units, putting ? = 1. This will be done from now on throughout
this book.
4 The step function
0
9 (p°) for { po
= 0 0 <
combined with the 6-function on the hyperboloid is Lorentz invariant; it just singles out the
positive energy mass hyperboloid.
30 I. Background

dµ = 6(p2 — m2)e(p°)d4p (I.3.14)

or, if we use the 3-momentum p to coordinatize the hyperboloid

dapP ;
dµ 2E 6 P= (p 2 + m2 ) 112 . (I.3.15)

To complete the analysis we pick a point p on the orbit and choose for every
p on the orbit an element 0(p) E C so that

AP(p))p = P. (1.3.16)

In the case m+ we may take p = (m, 0, 0, 0). According to the definition we

have
= m71; 13= P° 11+ pfr. (I.3.17)
Thus we can choose
i /2 (eP + po. ) 1/2 (1.3.18)
o(p) = m-
which is well defined because P is a positive matrix. We can use U(,(3(p)) to
identify the degeneracy spaces 7-h with so that the degeneracy spaces can
all be considered as samples of one "little Hilbert space" 4. Then we can write

?-1 = 1j ® 11c (I.3.19)

where 7 -(c = £( 2) (R3 ) with measure (1.3.15). This is most simply described
in Dirac's notation. Choosing a basis in ij labelled by an index k we have an
"improper basis" I p; k) in N and

Ip; k) = U (0(p)) I p; k). (I.3.20)

Now, for any given p on the orbit and any given a E ,C we can make the
decomposition
= 0(P)7(a, PV3-1 (P); p = A (a)p• (1.3.21)
Then 7, defined by (1.3.21) is an element of the stability group of p, i.e. A(7)
leaves p invariant. With our choice of p the stability group is SU(2), the covering
group of SO(3). To obtain an irreducible representation of q3 in 71 we need an
irreducible representation of the "little group" SU(2) in I. These latter are finite
dimensional. The representors are the well known (2s + 1)-dimensional rotation
matrices Ds to angular momentum s (s = 0, 1/2, 1 • • •). In our chosen basis we
can write the complete representation as

ei(p °-P8Î Ip
U(a) I p; k)= ; k) , (I.3.22)

U(a) I p; k) = > Dik ('7(a, p)) I p ; 1); p = fl(a)p.

'
(I.3.23)

One might think that the case of zero mass can be dealt with by taking
the limit m — ^^ 0 in the preceding discussion. This is not exactly true, however.
 1.3 Poincaré Invariant Quantum Theory 31

We can pick now the momentum vector p = (1/2, 0, 0, 1/2) on the orbit. The
corresponding matrix is
_ 1 0
(1.3.24)
—0 0
and the little group (stability group) is generated by the unimodular matrices

e'g0 0 1
and yo = 0 1 ' (1.3.25)
11s41 —

yo E R mod 27r, ?I arbitrary complex. This 3-parameter group is isomorphic to

the Euclidean group of the plane, the generator -y, corresponding to rotations,
and y,, to translations. The irreducible representations of this group can be
found by applying the same technique as used for the Poincaré group, now with
two dimensions less. If y,, is non-trivially represented the representation space
is infinite dimensional. These representations have so far found no application
in physics. So we restrict attention to the representations with U(7,1) = 1. They
lead to a 1-dimensional "little Hilbert space" and

D(y,,o) = einw; n E 7. (1.3.26)

The restriction of n to (positive or negative) integers comes from the fact that
we need a true representation of the matrix group y v , not a representation of
its covering group to get a true representation of q3. Noting that A(y, p ) is a
rotation in the x 1 -x2-plane by an angle Zip we can interpret

(1.3.27)

as the angular momentum along the axis of the linear momentum p. Instead of
the spin (magnitude of the angular momentum in the rest frame) we have in
the zero mass case the helicity, the component of the angular momentum in the
direction of p.

I.3.3 Single Particle States. Spin

The irreducible representations of 513 of class (m+ , s) yield the quantum theory
of a single particle with rest mass m and spin s alone in the world. A general
state vector is here characterized by a (2s + 1)-component wave function 'k(p)
in p-space, the probability amplitude in the basis (I.3.20). The scalar product
is given by

(W'IW) = E Jr 1- k(P) k(P) 2d p

:P
(I.3.28)

We can, instead, use a wave function in x-space, defining

-312 f (p)e_iPxd(p)
bk(x) = (27r)
32 L Background

3
= (270 -3/ 2 J P)Q1^px-Ept) d p. (1.3.29)
^_k(^p^
26p

Due to the fact that p is restricted to the mass hyperboloid Ok(x) satisfies the
Klein-Gordon-equation
2
(D + m 2 )5k(x) = 0 ; D ate - A. (I.3.30)

One may note in passing that this apparently "dynamical" equation results
in the present context from the irreducibility condition for the representation.
The "Casimir operator" POPµ commutes with all representors U(a, a) and must
therefore, in an irreducible representation, be represented by a number. We
must remember still that only the positive energy solutions of (1.3.30) are
admitted. This implies that the initial values tp and ao/at at a fixed time are
not independent; in fact 8'/et is determined by 0. The differentiation a/8t
is described in momentum space by the multiplication operator -iep . This

_
corresponds in position space at fixed time to a convolution with an integral
kernel
e(X - x') _ i(27) -3/2f (p2 + m2)1/2eip(x-x5d3p, (1.3.31)

so that
alp (t, x) = f E(x - - x')b(t, x')d 3x'. (I.3.32)

In the case of spin zero the full description of the representation in terms of
the wave function 0(x) is simple. The transformation law reads

(U(a) ) (x) = (x -
a); (U (A)0) (x) = (11 -1 x) (1.3.33)

and the scalar product can be written as

(W'W) i' ao —a' dix. (1.3.34)

10,4 ax ax
The right hand side is independent of the choice of t (the space-like surface
over which we integrate). One might be tempted to interpret the x-space wave
function (1.3.29) in analogy to wave mechanics as the probability amplitude for
finding the particle at time t = x° at the position x. That this cannot be done
is seen from the form of the normalization integral following from (I.3.34) and
from the non local relation between azb/at and ' at equal times. The question
of how "position of the particle" should be defined in this context has been
discussed by Newton and Wigner [New 49]. 1f one absorbs the factor (2e p ) -1 in
(I.3.28), defining the "Newton-Wigner wave function" in momentum space as

IY NW (P) - ( 2 6 13) -1120(0 (1.3.35)

and, correspondingly, the Newton-Wigner wave function in x-space as its Fourier

transform then the scalar product takes the same form as in the Schrodinger
theory
 1.3 Poincaré Invariant Quantum Theory 33

(W ' (W) = f' NW NWud 3 x(1.3.36)

One may regard tew (x) as the probability amplitude for finding the particle
at time t = x0 at the position x.

Remark: This definition of localization is dependent on the Lorentz frame. A

state W0 , strictly localized at time t = 9 at the point x = 9, given by
0, x) = 6 3 (x),
pô w(
is not strictly localized at x' = O if viewed in a different Lorentz frame. However,
for a massive particle the ambiguity in defining the localization is small, namely
of the order of the Compton wave length. The Newton-Wigner wave function is
related to the covariant wave function (1.3.29) by

pNw (t, x) = f K(x — x')t(t, x')d3x1 , (I.3.37)

and the integral kernel K, being the Fourier transform of (2e p ) 1 /2 , is a fast de-
creasing function of mix — x'1. Thus, for massive particles, the covariant wave
function tells us approximately where the particle is localized if no accuracy
beyond the Compton wave length is needed. For massless particles the concept
of localization is not appropriate.

Spin 1/2 and the Dirac Equation. For particles with spin the orthogo-
nal basis in the little Hilbert space is not well suited for a description of the
state vectors by wave functions in x-space. The transformation law (I.3.23)
contains a matrix depending in a highly non trivial fashion on p. This means
that one has a very complicated transformation law in x-space. One can give,
however, a manifestly Lorentz invariant formulation. In the case of s = 1/2 this
is achieved by going over to

MP) = m112 D EP — P Q )rk /2 ^k(p), (1.3.38)

k

and the corresponding x-space wave function (pr (x) obtained by a Fourier trans-
form as in (1.3.29). These transform like spinorial wave functions namely

(U(a)(,o)r (x) _ E arsçcs (A 1 (a)x). (1.3.39)

Each component cpr satisfies the Klein-Gordon equation (I.3.30). The scalar
product is
((P' 10 = m-1 f 45 (p)( 6p — 13(7)Ci5 (p)dP(p). (I.3.40)

It is convenient to introduce the (conjugate contravariant) spinor x by

(ep — pa)cp -- mX (I.3.41)

with the inversion

34 I. Background

(ep + poU = mcp, (I.3.42)

and use the 4-component wave function

(I.3.43)

This replaces in x-space the Klein-Gordon equation by a first order system, the
Dirac equation
(i7Pam, — m)/P = 0, (I.3.44)
where the Dirac matrices are given by

o 0 Il 0 —Q3
7 (I.3.45)
Il 0 ' cri 0

The transformation law is

(U(a)tp) (x) = S(a)IP (A'(c)x) , (I.3.46)

a 0
S(a) = (1.3.47)

In order to conform with customary notation we denote the complex conjugate

of the Dirac spinor by 0* and reserve the symbol zfi for 0* 7°. The scalar product
in x-space can then be written

(OM = J -
0* (x)^(x)d3x (I.3.48)

independent of the choice of t.

To sum up: the state vectors of a massive spin 1/2 particle may be described
as the positive energy solutions of the Dirac equation. This is a purely group
theoretic result. The motion of an electron in an electromagnetic field and the
prediction of antiparticles, tied to Dirac's original approach, is another matter.
There further physical principles, in particular locality and gauge invariance
play an essential rôle.
One comment should be added about the choice (1.3.45) for the Dirac ma-
trices. One may subject ' to an arbitrary position independent unitary trans-
formation in the component space. Then the form (1.3.44) remains but the
7-matrices are changed by a similarity transformation. The algebraic relations

?Ix + = 2g" (1.3.49)

and the property that 7° is Hermitean, 71' skew Hermitean stays unaltered. The
choice (1.3.45) is group theoretically distinguished because of the block diagonal
form of the transformation matrix (I.3.47); the upper and lower components
of are then separately Lorentz covariant spinors.

Higher Spin. It is tempting to apply here the familiar phrase: "we leave this
problem as an exercise to the reader". But let us add a few remarks. The sim-
 1.3 Poincaré Invariant Quantum Theory 35

plest manifestly covariant generalization to the case of spin s = n/2 is to use a

wave function cp ri ...r. (p), (rk = 1, 2) which is a spinor of rank n, totally sym-
metric in its indices. This has indeed (2s + 1) independent components. The
scalarproduct is then
tt

(g', çC ) = m-tt f (7 H(Ep - Pok)(Pdµ(p) (I.3.50)

1

with ak the Pauli matrix acting on the k-th index. In x-space the product
under the integral sign becomes a differential operator of order n. There are
many equivalent descriptions if one introduces redundant components. Take for
instance the case s = 1. A symmetric spinor of rank 2 corresponds to a self dual
antisymmetric Minkowski tensor. Instead one might use a 4-vector cpA with the
auxiliary condition ppcp„, = Q which again brings the number of independent
components to 3. If the single particle wave equations are used heuristically to
develop a quantum field theory with interactions then different starting points
(equivalent for a single free particle) suggest different theories.

1.3.4 Many Particle States

Bose-Fermi Alternative. In quantum mechanics the Hilbert space of state

vectors of an N-particle system is the N-fold tensor product of the single par-
ticle Hilbert spaces provided that the particles are not of the same species. For
identical particles one has to take into account a further principle, the permuta-
tion symmetry: two state vectors which result from each other by a permutation
of the indices distinguishing the particles must describe the same state. It may
be reasoned that this principle is a consequence of the fact that in quantum the-
ory we cannot keep track of individual particles without disturbing the system
continuously by observations. We cannot know who is who. 1f a permutation
of the particles is irrelevant for the state and if a state is represented by a ray
of state vectors then such a permutation can change a state vector only by a
numerical factor. These factors must give a 1-dimensional representation of the
permutation group and there are only two such representations: the trivial one
where each permutation is represented by the number 1 and the antisymmetric
one where odd permutations are represented by --1 and even ones by +1. Which
of the two possibilities is realized must depend on the particle species. One says
that a particle "obeys Bose statistics", in short that the particle is a boson, if
the trivial representation of the permutation group applies and that "it obeys
Fermi statistics" (the Pauli principle), if the antisymmetric representation has
to be used.
There has been an extended discussion as to whether the Bose-Fermi alter-
native really exhausts all possibilities. More general schemes called para-
statistics, have been proposed following a suggestion by [Link] [Green 53].
The argument for the Bose-Fermi alternative sketched above looks convincing
but one has to recognize that the starting assumption, namely that states of
36 I. Background

several indistinguishable particles may be described uniquely up to a factor by

wave functions in configuration space, is not operationally justified. The po-
sition of an individually named particle is not an observable, only symmetric
functions of the single particle observables are meaningful. Chapter IV of this
book will be devoted to a derivation of the intrinsic significance of statistics as
a consequence of the locality principle and to an analysis of the possibilities.
Le us denote by (7-l®N) s the symmetrized, by (7-0'N ) A the antisymmetrized
N-fold tensor product of the Hilbert space It These are the representation
spaces for a system of N indistinguishable bosons (resp. fermions) and, if the
particles do not interact, the representation of T3 in this space is determined
from the irreducible representations in the single particle space:

U = U®9 ) s for bosons, (I.3.51)

U®^ ) for fermions.
A
Fock Space. Creation Operators. In high energy particle physics processes
in which particles are created or annihilated are essential. Then the particle
number N becomes a dynamical variable and we need for a kinematical de-
scription of all relevant states the "Fock space " 7 -lF which is the direct sum
(denoted by E®) of all the N-particle spaces
^
7-1 E ® (®N)
=S,A
(I.3.52)
N =ü

A state vector in 7-iF is an infinite hierarchy of symmetric (resp. antisymmetric)

wave functions
c
^_ 0i (^) (1.3.53)
'02(6 s 6 )
where we have written the argument e to denote both position (or momentum)
and spin component; tPN is the probability amplitude for finding just N particles
and those in the specified configuration; c E C is the probability amplitude to
find the vacuum.
A more convenient way to describe vectors and operators in 71F is the
following. We choose some complete and orthonormal basis of single particle
state vectors, labeled by an index k (k = 1, 2..). Let

(n) = n1, n2, • • • (I.3.54)

denote an occupation number distribution; nk is the number of particles in the

k-th state. Note that (n) is an infinite sequence of occupation numbers since
there are infinitely many orthogonal quantum states for one particle. But only
such sequences are admitted for which the total particle number is finite, no
matter how large:
E nk < oo. (1.3. 55)
 1.3 Poincaré Invariant Quantum Theory 37

In the case of fermions each nk can, of course, only assume the values 0 or 1.
Let W( n) denote the normalized state vector corresponding to the occupation
number distribution (n). These vectors form a complete orthonormal basis in
One may introduce then a system of annihilation and creation operators
in this basis.
In the bosonic case the annihilation operator ak for a particle in a state k
is defined by
akW(n,) = nki2 W(n-6 k )• (1.3.56)
Here 6k = (0, 0 • • • 1, 0 • • .) with 1 in the k -th position. The creation operator
(the adjoint of ak) is
4w0.0 = (nk +
1 ) 112 (n+6k) (1.3.57)

The factor nk/ 2 in (1.3.56) is not a matter of convention but ensures the co-
variant transformation property of the annihilation operators under a change of
the basis in single particle space. If one changes from the orthonormal system
cpk to ço; by a unitary matrix

^t = UwPk (I.3.58)

then the creation operators in the primed basis are related to the original ones
by
al*
= E
Uikak• (1.3.59)
This may be checked by direct calculation from the representation (I.3.53) of
general state vectors in 7 -lF but more simply by observing that (I.3.56), (I.3.57)
imply the commutation relations

[ak, at] = 0, [ak, ai ] = 6kt, (1.3.60)

which generalize, for arbitrary single particle state vectors to

[a(01), a(ç2)] = 0, [a(921), a * (o2)] _ (cp 1 , (,02). (1.3.61)

The creation operators depend linearly, the annihilation operators conjugate

linearly on the wave function.
For fermions the definition (1.3.56) is replaced by

0 if nk =
akW(n) = (I.3.62)
( -1 ) /V(n_6k) if nk = 1
where p = >t<k nti is the number of occupied levels preceding k. Instead of the
commutation relations (I.3.60) one then has

[ak, at]+ = 0, [ak, ai ]+ = 6k1, (I.3.63)

where the symbol [A, B] + = AB + BA denotes the anticommutator.

Empirically, particles with half integer spin are fermions, particles with in-
teger spin are bosons. It has been one of the major achievements of general
38 1. Background

quantum field theory to show that this observed connection between spin and
statistics follows from its general principles. This will be discussed in section 5
of Chapter II.

Separation of Center of Mass Motion. Let us consider a general (reducible)

unitary representation of 513 which contains no zero mass subrepresentation.
Then we may split off the center of mass motion writing
71 = h ® 1Ic, (I.3.64)
where 1Ic = £(2) (RR3) with measure dp(p) = d3p. Vectors in may be viewed as
r)
corresponding to states in the center of mass system (total 3-momentum zero).
Vectors in fc describe the motion of the center of mass. A general state vector
W E 1-1 is then given by a function
P E 13 — CI)) E 4, (1.3.65)
with
(ow) = f('(P)kt?(P))d3P (I.3.66)
where the integrand on the right hand side is the scalar product in ii.
In this description the generators of the Poincaré group are expressed in
terms of operators Pk, Xk (k =1, 2, 3) acting in 11c, a mass operator M and
angular momentum operators lk (k =1, 2, 3) acting in r); see Bakamjian and
Thomas [Bakam 53]. The nonvanishing commutation relations are
[Pk , Xi] = —i 6kt , (I.3.67)
[4, 12] = il3 and cyclic permutations. (1.3.68)
Of course the Pk , X k commute with the lk; M commutes with all the others.
Pk are the generators of spatial translations. Time translations are generated
by
P° _ (P2 + M2)1/2. (I.3.69)
The generators of spatial rotations are (in 3-vector notation)
L=XxP+1. (1.3.70)
The generators of boosts are 5
K = ( P°X + XP°) + (P° + M) -1 P x 1. (I.3.71)
2
X may be (qualitatively) interpreted as the center of mass position at time zero.
By (1.3.71) it is closely related to the generators of boosts. These formulas have
served as a starting point for a Lorentz invariant formulation of phenomenolog-
ical interparticle forces by writing M as a sum of single particle contributions
plus an interaction potential (see in particular Coester [Coest 65]). It is a highly
nontrivial task to ensure in this procedure even macroscopic locality if more than
two particles are involved.
5 These
are the special Lorentz transformations changing the velocity; Kk generates pseudo-
rotations in the x 0 -x k plane.
 I.4 Action Principle 39

1.4 Action Principle

The belief that the actual world is the best of all possible worlds, or that God
gave laws of nature optimally designed to achieve an end, has provided through
centuries an inspiration to fundamental physics. It brought a teleological ele-
ment which has been extremely fruitful but which, in spirit, appears to be quite
opposed to the principle of locality. The simplest example illustrating the sur-
prising reconciliation between the local and the teleological point of view is the
fact that of all curves connecting two points the straight line is the one with
the shortest length. Correspondingly, in Riemann's geometry an isoparallel is
also a geodesic. The "straightness" (isoparallelism) is locally determined, the
minimality of length (geodesic property) is an optimization of the path between
a given beginning and end.
The field equations of classical physics mentioned in section 2 can all be
regarded as the Euler equations of a variational principle, demanding that a
certain quantity called action shall be stationary if the actual field distribution
in space-time is infinitesimally varied. The action S is an integral of a local
densityl L called the Lagrangean which at a point x is a function of the fields
and their derivatives
S =f
L d 4x. (1.4.1)
Some merits of the action principle are evident. It is much more economical
to write down a Lagrangean than the full system of field equations resulting
from it; it gives, at least if combined with requirements of simplicity, heuristic
guidance of great value in the search for new theories. Closely tied to the action
principle is the canonical formalism which historically played a central rôle in
the evolution of quantum theory.
The single most salient feature which a classical theory formulated by an
action principle has in common with quantum theory is the double rôle of each
physical quantity. The quantity can be measured, yielding a number in any
given physical situation. But it also defines an infinitesimal transformation of all
other quantities, by the Poisson bracket in classical theory, by the commutator
in quantum theory. The definition of Poisson brackets is usually given with the
help of the machinery of the canonical formalism but, as Peierls has shown
[Pei 52], they can be defined directly from the Lagrangean. This avoids the
asymmetric treatment of the time coordinate and uses only relativistic causality
instead. It assumes, however, that the quantity in question is an integral of a
local function of the fields and their derivatives over a space-time region which
is bounded at least in time-like extension. Given such a quantity X we consider
the modified Lagrangean

GE = L + E X, E -> 0 later. (I.4.2)

'Actually the integrand in (I.4.1) should be a 4-form. In special relativity we may regard
G as a scalar (the Hodge dual of a 4-form). In general relativity this procedure replaces the
integrand by LIg11/2 where 191 is the absolute value of the determinant of the metric tensor
and L is a scalar.
40 I. Background

The Euler equations of the modified Lagrangean agree outside of the support
of X with those of L. Let us generically denote all fields occurring in the theory
by the single symbol O. Each solution 0(x) of the Euler equations of the orig-
inal Lagrangean determines a solution 0ÉRx of the Euler equations of ,CE which
coincides with 0 in the remote past and another one, 45Ax , which agrees with
45 in the remote future (the upper indices standing for retarded and advanced,
respectively). Define

S P = ^^(^x — 4,)
E=o
sX^ = de k' x — ^) E= a
(1.4.3)

and
Sx$ = bp — S0 . (1.4.4)
One proves that 0 + e6x0 satisfies, to first order in E, the Euler equations of
the original Lagrangean. Thus 6x gives an infinitesimal transformation in the
solution space of the field equations.
We illustrate this fact in a simple example. It shows the essential elements for
a proof in the general case. Suppose we have a single scalar field cp in Minkowski
space and a Lagrangean density

G 2
= apcPaF`cP — 4 gcp4 . (I.4.5)

The wave equation is

❑cp + gcp3 = 0. (I.4.6)
If cp is a solution of (I.4.6) the condition that ço + 6cp is again a solution reads,
for infinitesimal 6cp,
❑Scp + 3g(p2 Scp = 0. (I.4.7)
This is a linear equation for 6cp since we regard cp (the solution around which
we vary) as given. On the other hand, taking the difference between the Euler
equations for the modified and the original Lagrangean, one sees that 6XCp and
60 are solutions of an inhomogeneous equation

no + 3gcp20 = gx(c,o) (I.4.8)

with the respective boundary conditions & = 0 for t —* —oo in the case 1p = 6 ço
and 0 = 0 for t +oo if 0 = O. The inhomogeneous term qx drops out
when we take the difference between the retarded and advanced solutions. Thus
6,0p as defined in (I.4.4) satisfies (1.4.7). ❑
Now let Y be another quantity, expressable like X as an integral of a local
function of the fields and their derivatives over a space-time region of bounded
temporal extension. Then the action of ox on Y can be identified with the
Poisson bracket {X, Y} as originally defined within the canonical formalism.
One has
{X, Y} = 6xY = —OyX. (1.4.9)
 I.4 Action Principle 41

It suffices to verify this for the basic quantities (linear functionals of the fields)
which is straightforward.
The double rôle of physical quantities is at the root of the very important
relation between conservation laws and symmetries. A symmetry may be defined
as a transformation 0 -- 0', x -- x' with the two properties:

a) If 0 satisfies the field equations so does V.

b) The transformation respects locality i.e. 0'(x') depends only on «(z) (and
possibly derivatives of 0 at this point).

Here, as above, 0 and V are understood as denoting all fields which occur in
the theory. We have seen that each physical quantity X defines an infinitesimal
transformation 6x satisfying a). If X is an integral of a local function of the
fields over a 3-dimensional space-like surface Eo then b) will be satisfied for
those points which lie on this surface but not, in general, for later or earlier
points. If, however, the integrand is a closed 3-form

Jaatid^aS 'r, (I.4.10)

X=^ ^o
dJ = 0 (I.4.11)
then the integral (I.4.10) is independent of the choice of the surface; we may
choose it to pass through any point. So b) will be satisfied and X defines an
infinitesimal symmetry. In physics it is customary to write (1.4.10) as a surface
integral of a conserved current; (1.4.11) is then the continuity equation:

X = j^`d^ 0 , (1.4.12)
f
apjP = 0. (I.4.13)
This results from (1.4.10), (I.4.11) by taking the Hodge duals (see section 2.3).
In special relativity the above reasoning is unaffected if j has additional
tensor indices. For instance energy-momentum is the integral of the energy-
momentum tensor TPv over a spacelike surface

P M' = f Ply day, (I.4.14)

E
and the conservation law is locally expressed by the continuity equation
apTuv = 0. (1.4.15)

As a consequence PP, considered as defining an infinitesimal transformation by

means of the Poisson bracket, generates a symmetry. It gives the infinitesimal
translations in space-time. The same argument can be used for the angular mo-
mentum tensor elivo in space-time (for its definition see e.g. [ Wentzel 19491). Its
surface integral leads to the generators MP' of Lorentz transformations. Note
incidentally that M°k is not a constant of motion, corresponding to the fact that
42 I. Background

Lorentz transformations do not commute with time translations. Therefore we

have used the property b) to define the notion of symmetry which is more nat-
ural and more general than the frequently made statement that a symmetry
should commute with time translations (compare the discussion in section 3.2
of chapter III). Correspondingly we regard the continuity equation as the ex-
pression of a conservation law rather than restricting attention to constants of
motion.
The converse argument, leading from a continuous symmetry group to a
conservation law, is known as Noether's theorem. Formulations and proofs of it
at various levels of generality are found in many books.
In the past decades the direct use of a Lagrangean in the formulation of
quantum theory has come more and more to the foreground. The basic idea,
suggested by Feynman [Feyn 48], see also [Feynman and Hibbs 1965] is that in
quantum mechanics every path leading from a point qi in configuration space at
time t1 to another point q 2 , t2 should be considered, not only the path following
the classical equation of motion. The latter is the path giving an extremal
value to the action. According to Feynman's proposal each path T contributes a
probability amplitude exp i/hS(F) and the addition of all these amplitudes, the
functional integration over all paths, gives the quantum mechanical transition
amplitude from q 1 , t1 to q2, t2 . The passage to the classical theory for h — 0
is visible then from the method of stationary phase in the evaluation of the
functional integral. The path for which S is stationary with respect to small
variations, i.e. the classical path, gives the main contribution.
The transfer of this idea to quantum field theory leads to the problem of
integration over all sections of a field bundle. For its practical application in the
computation of vacuum expectation values see e.g. [Glimm and Jaffe 19871 and
the references given there.

1.5 Basic Quantum Field Theory

1.5.1 Canonical Quantization

The original objective of quantum field theory was to develop a quantum version
of Maxwell's electrodynamics. This was to some extent achieved between 1927
and 1931 by applying to the Maxwell theory the same formal rules of "quantiza-
tion" which had proved so successful in the transition from classical to quantum
mechanics. For the free Maxwell equations (absence of charged matter) this ap-
proach was a full success. It led to a correct description of the properties of
light quanta. The two free Maxwell equations taken together imply that each
component of the field strength tensor Fp , obeys the wave equation

OF,, = 0; o = g' 'c9, ,. (1.5.1)

In addition, the first order equations give constraints which couple the differ-
ent components and yield the transversality of electromagnetic waves. Before
 I.5 Basic Quantum Field Theory 43

discussing the complications due to the constraints let us treat as a prelimi-

nary exercise the case of a scalar field 0 obeying the Klein-Gordon equation,
considered as a classical field equation

❑ o+m2 'P= 0. (1.5.2)

This example will already provide the essential lesson to be learnt from the
canonical quantization of a linear field theory. The Lagrange density is

L = 1 rap0av0 (1.5.3)
— m202.
To copy the quantization rules of section 1 one has to treat space and time in
an asymmetric fashion, considering at some fixed time the spatial argument x
of 0 as a generalization of the index i so that the field is regarded as a mechan-
ical system with continuously many degrees of freedom. The partial derivatives
a/aq, are replaced by the variational derivatives 6/60(x) and the Kronecker
symbol 6ik is replaced by the Dirac 6-function 6 3 (x — x') . The Lagrange func-
tion results from the Lagrange density (1.5.3) by integration over 3-dimensional
space at a fixed time, say t = O. The canonically conjugate momentum to 'P (at
this time), denoted by ir(x), turns out to be equal to the time derivative of 0
and one imposes the canonical commutation relations for the initial values at
t= 0

[r(x), 0(3 0] = —i 63 (x — 30; [0(x), 0(x')] = [r(x), r(x )] = 0. (I.5.4)

The Hamiltonian is

H f (^2 + (grad c2 + m2^2 ) d3x. (I.5.5)

=2
Since this procedure is described in all early presentations of quantum field
theory, see e.g. [Wentzel 1949] there is no need to elaborate on it.
A more elegant method, avoiding the preference of a particular Lorentz
system and fixed time is provided by Peierls' method (see section 4). One finds
directly the commutation relations between fields at arbitrary (not necessarily
equal) times 1
[0(x), ''(x')] = id(x — s') (I.5.6)
where d is the difference between the retarded and the advanced Green's func-
tion of the Klein-Gordon equation (I.5.2). It is a singular integral kernel, solu-
tion of the Klein-Gordon equation with initial conditions

aa = —63 (z).
z°=o = 0; az° zo =o
(1.5.7)

Two important remarks follow from (1.5.6) and (I.5.7):

1 The commutation relation (I.5.6) was first proposed by Jordan and Pauli (1928) in the
case rn=O.
44 I. Background

a) The commutator of fields at different space-time points is a "c-number",

(a multiple of the identity operator); it commutes with all observables.

b) The commutator vanishes if x and x' lie space-like to each other.

While property a) is special to free field theories (linear field equations) prop-
erty b) has general significance and a simple physical reason: the measurement
of the field in a space-time region 01 cannot perturb a measurement in a re-
gion 02 which is space-like to 0 1 because there can be no causal connection
between events in the two regions. Therefore such observables are compatible
(simultaneously measurable).

1.5.2 Fields and Particles

The most important consequence of the quantization is seen by Fourier trans-

forming the field. Due to the field equations the support of the Fourier transform
in p-space is concentrated on the two hyperboloids

p° = ±(p2 + m 2 ) 1/2 (1.5.8)

and we can write

0(t, x) = (27)- 3/2 (a(p)ei_6Pt) + a *(P)e-ox-epc)1 dap (1.5.9)

j i 2ep

with
+ m2)1/2.
^p = (p2 (1.5.10)
If the classical field is real valued then the quantum field should be Hermitean
which implies that a*(p) is the Hermitean adjoint of a(p). The canonical com-
mutation relations (I.5.4) give

[a(p), a * (P )] = 2e1,63 (p —
p'); [a(p), a(P )l = [a * (P), a * (P')} = 0. (1.5.11)

Comparing with the discussion in Section 3, equation (1.3.61), we see that the
canonical quantization of the free scalar field 0 leads to the quantum theory of a
system of arbitrarily many identical, noninteracting, spinless particles obeying
Bose statistics.
To be precise we have to supplement the relations (I.5.11) by the require-
ment that the representation space contains a vector qio, interpreted as the
vacuum, which is annihilated by all the a(p):

a(p)W0 = 0 for all p. (1.5.12)

Then the a(p) can be interpreted as annihilation operators, the a*(p) as creation
operators in the Fock space generated by repeated application of the creation
operators on the vacuum. The requirement (1.5.12) has to be added to the al-
gebraic relations (I.5.11) because the latter allow many inequivalent irreducible
 I.5 Basic Quantum Field Theory 45

Hilbert space representations of which (1.5.12) selects the one in Fock space.
This is due to the fact (mentioned in Section 1) that in the case of infinitely
many degrees of freedom one does not have von Neumann's uniqueness theorem
for the representation of the canonical commutation relations.
There is another problem which has to be faced. The quantum field 0 at a
point cannot be an honest observable. Physically this appears evident because
a measurement at a point would necessitate infinite energy. The mathematical
counterpart is that 0(x) is not really an operator in 7lF. It is an "operator valued
distribution" (see Chapter II, section 1.2) or, alternatively, it may be defined
as a sesquilinear form on some dense domain in 7-lF. This means that matrix
elements (W2 I0(x)IWi) are finite if both vectors W1 and W2 belong to a dense
domain D c 7iF characterized by the property that the probability amplitudes
for particle configurations decrease fast with increasing momenta and increasing
particle number. The problem is then that we cannot directly multiply "field
operators" at the same point. This problem has plagued quantum field theory
throughout its history. It is a fierce obstacle when one wants to define nonlinear
local field equations. In the case of free fields to which our discussion so far was
confined, there is, however, a simple solution. In this case (I.5.9) gives a clear
separation of the annihilation part and the creation part of 0(x). We may write

0(x) = a(x) + a*(x). (1.5.13)

One observes [Wick 501 that any product of the form a* (x)na(x)m is a well
defined sesquilinear form on a dense domain or, alternatively speaking, it is an
operator valued distribution. If A and B are two functionals of the field 0 one
defines the normal order product, also called the Wick product of A and B by
the prescription that each field operator occurring in A or B is expanded in
the form (1.5.13) and in the resulting formal expression for the product AB
the order of factors is changed so that all creation operators stand on the left,
all annihilation operators on the right in each term. Denoting the normal order
product by : AB : we have for instance

: «(s)2 : = a* (x)a*(x) + 2a* (x)a(x) + a(x)a(x). (I.5.14)

The Hamiltonian is obtained from the classical expression (1.5.5) by inserting

for 1$ the quantum field and taking the products in the normal order. One
obtains
H = P° = JEpa*(p)a(p)dL(p). (I.5.15)

Similarly one gets the other generators of the Poincaré group from their classical
counterparts.
The apparent miracle that canonical quantization of a free, scalar field leads
to Fock space and an interpretation of states in terms of particle configurations
has been seen as a manifestation of the wave-particle duality lying at the roots
of quantum mechanics. The generalization of this observation to a field-particle
duality has dominated thinking in quantum theory for decades and has been
heuristically useful in the development of elementary particle theory. A prime
46 I. Background

example is Fermi's theory of the 0-decay which associates a field with each
of the relevant particle types: the proton, the neutron, the electron and the
neutrino. From these fields taken at the same point in x-space one builds a
Lagrangean density which has matrix elements for the transition n—> p+ e- +Ûe.
Yet the belief in field-particle duality as a general principle, the idea that to
each particle there is a corresponding field and to each field a corresponding
particle has also been misleading and served to veil essential aspects. The rôle
of fields is to implement the principle of locality. The number and the nature of
different basic fields needed in the theory is related to the charge structure, not
to the empirical spectrum of particles. In the presently favoured gauge theories
the basic fields are the carriers of charges called colour and flavour but are not
directly associated to observed particles like protons.

I.5.3 Free Fields

Equation (I.5.9) may be read either as giving the decomposition of a (scalar)

quantum field into the set of creation-annihilation operators in the Fock space of
a species of (spinless) particles or, conversely, as the construction of a covariant,
local field from the representation U, n of the Poincaré group (for s = 0) via
,s

(1.3.51), (1.3.52), (1.3.61). The construction of a local, covariant field can also
be carried through for the other relevant types of irreducible representations
of3. The step from the single particle theory to the theory in Fock space is
often called second quantization because, with a due amount of sloppiness, one
regards the single particle theory to result by "quantization" from the classical
mechanical model of a point particle and, interpreting the single particle theory
as a classical field theory, the transition to Fock space is again a quantization,
achieved by replacing the single particle wave functions by creation-annihilation
operators with the "canonical commutation relations" (1.3.61).
For s = 1/2 the construction leads to the Majorana field. If ()MA denotes
a single particle wave function (r = 1, 2) as in equation (I.3.38) we may write
the corresponding creation operator as 2

a* (ço) = Erg f as(p)cpr(p)d,a(p) (I.5.16)

which defines the (improper) creation operators â(p) in p-space.

To obtain a local quantum field Or (x) we must combine the creation and
annih il ation operators as in (I.5.9)

Or(x) = ar(x) + âr(x). (I.5.17)

We assume Fermi statistics. The resulting commutation relations are then

08(y)1+ = En-6 (x - y) • (1.5.18)

2Er8
is the "metric tensor in spinor space" (see section 2).
 1.5 Basic Quantum Field Theory 47

Furthermore, since the Majorana field describes a single species of particles (no
antiparticles) the Hermitean adjoint 0T of cP, is related to 0:

is'
41(x) = E;.sm-1 â — aV 4.9, (x). (I.5.19)

Thus 4(x) anticommutes with 0(y) and with 0*(y) if x and y lie space-like to
each other. If we say that local observables are formed by taking the normal
ordered product of an even number of factors 0, 0* at the same point then we
have commutativity of local observables at space-like separation as demanded
by the locality principle. The transformation property of the field is

U(a)4',.(x)U -1 (a) = (a -1 ),,T47., (Ax); A = A(a). (1.5.20)

The analogous construction can be carried through for particles of higher

spin if we describe the single particle wave functions without redundant compo-
nents as symmetric spinors of higher rank (see section 3) . One then arrives at
local fields transforming like symmetrized spinor fields of higher rank. The case
of rank 2 with mass zero gives the free Maxwell field. The symmetric spinors
Ors and their complex conjugates 41:9 correspond to the self-dual tensors

,Fµv = (Ft,v f i*Fp,p). (1.5.21)

A plane wave solution of the Maxwell equations fixes them uniquely up to

a normalization factor. They describe respectively right hand and left hand
circularly polarized waves or, if one wishes, wave functions of a particle with
sharp momentum, zero rest mass and helicity +1 and —1 respectively. We may
introduce annihilation operators a ± (p) for these modes and creation operators
a±*(p). From these we can build up a local, Hermitean quantum field F,„(x)
acting in the Fock space of photons. One cannot, however, find a covariant
quantum field corresponding to the vector potential acting in this Fock space.
The standard method to incorporate the vector potential is due to Gupta and
Bleuler, [Gupt 50], [Bleu 50]. 3 It uses a "Hilbert space with indefinite metric", a
space equipped with a sesquilinear scalar product for which (WIC) is not positive
definite. In this space one characterizes a certain subspace called the "physical
state space", in which the metric is semidefinite. It still contains vectors of zero
length. Then one forms equivalence classes; two vectors in the ("physical") space
are called equivalent if their difference has zero length. Each such equivalence
class correponds to a vector in the Fock space generated by the electromagnetic
fields.
Note that the Fock space construction must always be modified if one uses
single particle wave functions with redundant components, e.g. if one describes
spin 1 particles by 4-vector wave functions. In that example we have a constraint
3 For this and the remaining material in this section see any standard text on quantum
field theory e.g. [Jauch and Rohrlich 1976] We restrict the exposition to a few remarks.
48 I. Background

i"pt,(p) = 0 and the creation operators a*(cp) do not define creation operators
a^ (p) in p-space. If one wants to define such operators and, correspondingly,
vector fields , (x) one needs a larger space than flF and subsidiary conditions
to single out the physical states.
Reviewing our construction of free fields from the irreducible representations
of q3 we notice that the two most important fields, the Dirac field and the
electromagnetic potential are not directly obtained. The Dirac field results if one
starts from two types of particles (particle and antiparticle) whose description
is interwoven so that it is not equivalent to two Majorana fields. We define the
first two components of the field Or(x) by combining the annihilation operators
of the particle (as its negative energy part) with the creation operators of the
antiparticle (as its positive energy part). Then P and .1.* are independent; the
Hermiticity condition (I.5.19) disappears. As in (I.3.41), (1.3.42) one goes over
)
to the 4-component field 1p _ . The physically important point is seen if
X
we introduce the charge as the difference between the number of particles and
the number of antiparticles. All components of b lower the charge by one unit,
the components of 0* raise it. This may serve as a reminder that the deeper
significance of the fields is to effect a local change of charge, not of particle
number. The observable fields are formed by multiplying components of •0* and
V) at the same point. The most important one is the current density which, in
customary notation, is written as

?(x) = q : Ip(x)-elp(x) : (I.5.22)

with
'0 _
,y
0* ° . (1.5.23)
Due to their construction the observables do not only commute with the charge
but they remain unchanged if 1p is subjected to a local gauge transformation

Ip(x) e'A(x) (x) (1.5.24)

with an arbitrary function A(x).

I.5.4 The Maxwell-Dirac System

The Dirac equation, originally conceived as the wave equation for a particle
with charge q in an electromagnetic field described by the vector potential AA,
is
{'yµ (OA + qA i4 (x)) — m} z/i(x) = 0. (1.5.25)
It results from the free Dirac equation by the substitution

4,-, Di, = âµ - iqA i,,, (1.5.26)

which corresponds to the old observation in the classical electron theory that the
Lorentz force results in the Hamiltonian formalism by the simple replacement
 I.5 Basic Quantum Field Theory 49

pp, --t p, + qA tt . (I.5.27)

But the root of this rule is not visible in classical mechanics. It is local gauge
invariance: if we subject 0 and A simultaneously to the gauge transformations

0(x) —^^ Ili (x) = e igA(x) 0(x); Am (x) —4 A;,(x) = A, (x) + âu a(x) (I.5.28)

then (I.5.25) remains valid for the transformed fields and A m defines the same
electromagnetic field strengths as AA . In differential geometric language the
Dirac wave function is a section in a fiber bundle whose base space is Minkowski
space and whose typical fibre is a complex 4-dimensional space with the struc-
ture group U(1). The latter, because a point in the fiber (the value of the Dirac
spinor 0) is physically relevant only up to a phase factor which may be con-
sidered as an element of the group U(1). The observable fields are of the form
TpT with F = -yu, -ru-r • • - . They remain all unchanged if 0 is multiplied by a
phase factor i.e. if the section is subjected to a gauge transformation (1.5.28).
The vector potential provides a U(1) connection in the bundle, analogous
to the rôle of the affine connection, the parallel transport, in the case of the
tangent bundle (see subsection 2.3). It allows the comparison of phases of
in infinitesimally neighbouring points. The electromagnetic field strength is the
curvature of this connection. If it does not vanish then we have no natural
comparison of phases at a distance. The operator D,. of (I.5.26) is the covariant
derivative in this bundle.
To complete the system of field equations we have to add to (I.5.25) the
Maxwell equation (I.2.56) with the current density j" given by (1.5.22) and
the expression (1.2.55) for the electromagnetic field in terms of the potential.
These relations are obviously all invariant under the gauge transformations
(1.5.28). The full system of equations follows from a variational principle with
the Lagrange density
1
,C = 0(11 1`Du — m )11i — 4 F t UFu„ (I.5.29)

where it is understood that FA,, is written in terms of A m,, and the basic variables
which are varied independently in the variational principle are 1/i, Ti) and A. We
can write
.C = ,CD +.Cm+ LI (I.5.30)
where GD is the Lagrange density of the free Dirac theory

GD = 0(i748).4 — m)0, (I.5.31)

GM the Lagrange density of the fr ee Maxwell theory

1
Lm = — 4 F^`V F11„ (I.5.32)

4 The second part of (I.5.28) is often called a gauge transformation of the second kind,
while the first part, the transformation of the charged field, is called a gauge transformation
of the first kind.
50 I. Background

and Li the Schwarzschild interaction Lagrangean coupling the Dirac and the
Maxwell fields:
.CI = rill,. (1.5.33)
Remarks: 1) As long as we consider « (x) and A 4 (x) as numerical valued func-
tions we may omit the Wick ordering symbol in (1.5.22). The system of field
equations (1.5.25), (I.2.56) with the definitions (1.2.55), (I.5.22) constitutes a
mathematically well defined classical field theory which, moreover, is built from
principles remarkably similar to those used in the construction of the general
theory of relativity. But for physics we cannot interpret //) as a classical field
nor can we interpret it in general as a quantum mechanical wave function of
a single particle because, if the vector potential is time dependent, one cannot
separate the positive and negative energy parts of . We must regard /p and A
as quantum fields. Then the definition of the products 'cb(x)A(x) occurring in
the Dirac equation and Tplp occurring in (1.5.22) poses a problem which is not
as easily dealt with as in the fr ee field case where we can define a Wick prod-
uct. In fact, this problem was considered as unsurmountable without a radical
change of the basic concepts until the successful development of renormalized
perturbation theory suggested that it could be overcome.
2) In classical Maxwell theory the vector potential is introduced as a con-
venient auxiliary tool. It serves to solve the homogeneous Maxwell equation
(I.2.54) and it also allows to derive the Lorentz force on a charged particle from
a variational principle by adding the Schwarzschild term (I.5.33) to the La-
grangean of the free motion. The interpretation of A A as the connection in a
fiber bundle appears only when charged matter is described by a complex matter
field, a feature which (at least in its physical interpretation) is tied to quan-
tum physics. Otherwise there is nothing which A t,, connects. We may conclude
therefore that in quantum physics the vector potential acquires a fundamen-
tal significance which it does not have in the classical theory. We encounter
in the Maxwell-Dirac theory besides the pointlike observables like ju, Fu„ also
"stringlike" observables associated to any path T from a point x to a point y,
namely
A(F) = (x) exp (_iq f A u (x`)dx'!`) (y). (I.5.34)

This is brought into sharp focus by the observation of Aharonov and Bohm
[Ahar 59]: If the accessible space available to an electron is not simply connected
e.g. by excluding a toroidal tube, then we may produce a situation in which the
electromagnetic field strengths vanish everywhere in the accessible region while
the vector potential cannot vanish there if there is a magnetic flux inside the
tube. The line integral of the vector potential around the tube must equal this
flux. The time development of the electron wave function is influenced by the
vector potential (i.e. by the total flux in the tube) even if the electron is confined
entirely to the force-free region.
 1.5 Basic Quantum Field Theory 51

1.5.5 Processes

The quantum field theory based on the Maxwell-Dirac system is called Quan-
tum Electrodynamics (QED). To study its physical consequences one started
from the free field theory, resulting if L i is neglected, as a zero order approxi-
mation. There we have the Fock space of noninteracting photons, electrons and
positrons. Li is considered as a perturbation producing transitions between par-
ticle configurations. The elementary process is the emission or annihilation of a
photon combined with the corresponding change of momentum of an electron or
positron or the creation or annihilation of an electron-positron pair. Originally
Dirac's time dependent perturbation theory, giving a transition probability per
unit time, ("the golden rule of quantum mechanics" as it was called by Fermi)
was used to derive the probability for spontaneous decay of an excited state
of an atom, the cross sections for the photo effect, Compton effect, electron-
electron scattering, pair creation, Bremsstrahlung. An excellent presentation of
this early stage of the theory is given in Heitler's book [Heider 1936] . The re-
sults of the lowest order of perturbation theory were found to be in very good
agreement with experiment. But there remained a dark cloud. One could not
continue the approximation. The higher order terms lead to divergent integrals.
The hope that QED contained any sensible physical information beyond the
lowest order calculations dwindled during the thirties. Faith was restored in a
spectacular way by joint progress in experimental precision and theoretical un-
derstanding between 1946 and 1949. Experiments established small deviation
of the hydrogen spectrum from the theoretical predictions of wave mechanics
("Lamb shift") and a deviation of the gyromagnetic ratio of the free electron
from the value 2 which it should have according to the 1-particle Dirac theory
("anomalous magnetic moment") . On the theoretical side it was the develop-
ment of an elegant, manifestly covariant form of the perturbation expansion
by Tomonaga, Schwinger, Feynman together with the idea of renormalization,
suggested by H.A. Kramers and first applied to the Lamb shift problem by
Bethe which allowed the computation of higher order corrections, and it turned
out quickly that these corrections could account for the experimental findings.
In subsequent years, as the experiments were refined and the theoretical com-
putations were pushed to the 4-th order and ultimately to the 6-th order, the
agreement between experiment and theory became truly spectacular.
We end this chapter with a quotation from the basic paper by Dyson [Dy 49]:
"Starting from the methods of Tomonaga, Schwinger, Feynman and using no
new ideas or techniques, one arrives at an S-matrix from which the well known
divergences seem to have conspired to eliminate themselves. This automatic
disappearance of divergences is an empirical fact, which must be given due
weight in considering the future prospects of electrodynamics."
II. General Quantum Field Theory

II.1 Mathematical Considerations

and General Postulates

The developments mentioned at the end of the last chapter had restored faith in
quantum field theory. On the other hand it could not be overlooked that in spite
of the great success of QED there remained ample reasons for dissatisfaction.
It was not understood how the theory could be formulated without recourse to
the perturbation expansion. The detailed renormalization prescriptions, needed
to eliminate all infinities, had become quite complicated and not easily commu-
nicable to one who had not acquired familiarity with the procedure the hard
way, namely by doing the computations and learning to avoid pitfalls. Apart
from QED there existed models for a theory of weak interactions which could be
compared in lowest order with experiment but which was not renormalizable,
and there were the meson theories of strong interaction where perturbation ex-
pansions did not prove to be very useful. This mixture of positive and negative
aspects of quantum field theory provided the motivation in the fifties to search
for a deeper understanding of the underlying principles and for a more concise
mathematical formulation. K.O. Friedrichs described his feelings about the lit-
erature on quantum field theory as akin to the challenge felt by an archeologist
stumbling on records of a high civilization written in strange symbols. Clearly
there were intelligent messages, but what did they want to say? (private con-
versation 1957). His answer to the challenge was his book on the mathematical
aspects of quantum field theory [Friedrichs 1953] where, among other things, he
pointed out the existence of inequivalent representations of the canonical com-
mutation relations and discussed examples of these under the heading "myriotic
fields".

II.1.1 The Representation Problem

The breakdown of von Neumann's uniqueness theorem (see 1.1) in the case
of infinitely many degrees of freedom had already been noted in 1938 by von
Neumann himself but it took a long time until this fact became widely known
and its significance for quantum field theory was recognized. The phenomenon
is easily understood if we start from the relations in the form (1.3.60) which
54 II. General Quantum Field Theory

results from the Heisenberg relations (I.1.2) by substituting

ak = 2-1"2 (pk — iqk); ak = 2-1I2 (pk + iqk) . (11.1.1)

Using the notation of (I.3.54) we note that an occupation number nk is an

eigenvalue of the positive operator Nk = akak and the commutation relations
imply that nk must be a non negative integer. An occupation number distribu-
tion (n) is an infinite sequence of such integers. We can divide the set of such
sequences into classes, saying that (n (1) ) and (n (2) ) are in the same class if the
sequences differ only in a finite number of places. Application of a creation or
annihilation operator changes (n) only in one place. Therefore the basis vectors
W(n) introduced in subsection 1.3.4, with (n) restricted to one class, already span
a representation space of the ak, ak and it is evident that representations belong-
ing to different classes cannot be unitarily equivalent. The set of representations
obtained in this way does, however, not exhaust by far all possibilities. There
are representations in which the W(n) do not appear as normalizable vectors. Let
us illustrate this in the case of the anticommutation relations (I.3.63) to which,
of course, the same reasoning as above may be applied. There the numbers
nk are restricted to the value 0 or 1 and an occupation number distribution,
being an infinite sequence of such numbers, is in one to one correspondence
with a real number in the interval [0, 1] because it may be regarded as defining
an infinite binary fraction. One may then construct representations acting in a
Hilbert space of functions over this interval, square integrable with respect to
some measure.
A systematic study of the classification problem of irreducible representa-
tions of canonical commutation and anticommutation relations was undertaken
by Giirding and Wightman [Gard 54a, 54b]. An exhaustive, practically usable
list of representations appears unattainable. This follows from the study of the
nature of the algebras, see [Mackey 57] and [Glimm 61].
The first examples in quantum field theory where "strange representations"
appeared were given by Friedrichs loc. cit. and van Hove [Hov 52]. It soon be-
came evident that the occurrence of such representations was not an exception
but a generic feature [Haag 54, 55a] . Specifically: the standard quantization pro-
cedure of a classical field theory leads to field quantities which satisfy canonical
commutation relations at a fixed time. Let us take the simplest example, a
scalar field cp with Lagrangean (1.4.5). For g = 0 we have a free field. For g 0
the field equations are non linear and one expects a theory with interacting
particles. The equal time commutation relations between ça and its canonically
conjugate momentum r apiax° are independent of g; they are given by
(1.5.4) (replacing by cp). We need a representation of the relations (1.5.4) in
a Hilbert space it the space of physical states. In fact, we want an irreducible
representation because the field at arbitrary time is determined by the canonical
quantities due to the field equations and thus all observables can be expressed
in terms of the cp(x), 7r(x).
We want to argue that the representations must be inequivalent for different
values of g. If this were not so we could use one Hilbert space in which cp(x) and
 1I.1 Mathematical Considerations and General Postulates 55

7r(x) are represented for different values of g by the same (improper) operators.
For each g we expect to have one physically distinguished state, the physical
vacuum. Its state vector 12g must depend on g because it is the lowest eigenstate
of the Hamiltonian and the latter depends on g. On the other hand the operators
U(a), representing translations in 3-space, would not depend on g because, for
any g, they give the same transformation of the canonical quantities

U(a)cp(x)U -1 (a) = cp(x + a), U(a)7r(x)U -1 (a) = 7r(x + a), (II.1.2)

and, in an irreducible representation, they are determined (up to a numerical

factor of modulus 1) by (II.1.2). The vacuum is invariant under space transla-
tions,
U(a) f2g = (29. (II.1.3)
For g = 0 we have a fr ee field and the Fock representation as described earlier.
There the vacuum state vector is the only invariant vector under the U(a). If,
for some other value of g, we also had the Fock representation then (II.1.3)
would imply 09 = 00 . More generally, we shall see in 111.3 that in an irreducible
representation space of the field operators there can be at most one invariant
vector under space translations. Thus if f29 99 then the representations must
be inequivalent. We conclude therefore that the determination of the represen-
tation class of (1.5.4) is a dynamical problem. This conclusion is usually referred
to as Haag's theorem, (compare also [Coest 60}).
Actually the above discussion touches only the tip of an iceberg. It appears
that an infinite renormalization of the field is needed. In the renormalized per-
turbation expansion one relates formally the "true" field to the canonical field
cp which satisfies (I.5.4) by
4, = Z-1/2 v (1I.1.4)
where Z is a constant (in fact zero). This means that the fields in an interacting
theory are more singular objects than in the free theory and we do not have the
canonical commutation relations (1.5.4). 1 While this decreases somewhat the
interest in canonical commutation relations the representation problem remains
significant whenever we have to deal with infinitely many degrees of freedom.
We note that the algebraic structure, whatever it is, does not, in general, fix
the Hilbert space representation. This phenomenon will concern us in chapter
IV. For canonical commutation relations it plays an important rôle in the non
relativistic many body problem, the statistical mechanics in the infinite volume
limit. "Strange representations" appear naturally e.g. by Bogolubov transfor-
mations [Bog 58} and in the description of thermal states given by Araki and
Woods [Ara 634 We shall come back to this in chapter V.
1 In constructive field theory it was proved that field renormalization is finite in models
based on 2-dimensional space-time. See [Glimm and Jaffe 1981. So canonical commutation
relations for interacting fields can be upheld in such cases. In Minkowski space theories an
infinite renormalization seems to be needed and the relations (I.5.4) get lost.
56 II. General Quantum Field Theory

11.1.2 Wightman Axioms

The need to base the discussion of quantum field theory on clearly stated pos-
tulates was felt acutely by several authors in the early fifties. Wightman and
Girding began in 1952 to isolate those features of quantum field theory which
could be stated in mathematically precise terms and to extract general postu-
lates which looked trustworthy in the light of the lessons learned from renor-
malization. This led to the "Wightman axioms" . 2
Let us state here the essential postulates. The problem of physical interpre-
tation and description of particles will be postponed to sections 3 and 4.

A. Hilbert Space and Poincaré Group.

1. We deal with a Hilbert space 71 which carries a unitary representation of

T.
2. There is precisely one state (ray in 7t), the physical vacuum, which is
invariant under all U(g), g E q3.

3. The spectrum of the energy-momentum operators PP is confined to the

(closed) forward cone
p2 > 0; p° > 0. (I1.1.5)

B. Fields.

1. Fields are "operator valued distributions" over Minkowski space. 3

This needs an explanation. As already mentioned in subsection I.5.2 a
quantum field 45(x) at a point cannot be a proper observable. It may be
regarded as a sesquilinear form on a dense domain D C 7-l; this means
that the matrix element (W2145(x)I 1 ) is a finite number when both Wi
and W2 are in D and that it depends linearly on WI , conjugate linearly on
W2. To obtain an operator defined on the vectors in D one has to average
("smear out") 'P with a smooth function f on Minkowski space i.e. take

^
(f) = f'P(x)f(x)d 4x. (I1.1.6)

If f belongs to the "test function space" then 0(f) is an (unbounded) oper-

ator acting on 7-1, defined on some dense domain D C N. The test function
space is usually taken as the space S of Laurent Schwartz, the space of
infinitely often differentiable functions decreasing as well as their deriva-
tives faster than any power as x moves to infinity in any direction. 0(f)
2 The first published account is [Wight 57]. An extended version is given in [Wight 64]. Sim-
ilarly motivated work [Haag 54, 55a], Lehmann, Symanzik and Zimmermann [Leh 55] placed
less emphasis on mathematical precision and was more concerned with physical interpretation
and particle aspects of the theory.
3 Apart from the quoted work of Girding and Wightman this has been recognized by
Schmidt and Baumann [Schmidt 56].
 II.1 Mathematical Considerations and General Postulates 57

depends linearly on f and its matrix elements are continuous functions

of f with respect to the Laurent Schwartz topology of S. If, as we shall
always assume, the space of allowed test functions is S, the distribution
is called tempered. Its Fourier transform is again a tempered distribution.
The mathematical theory of distributions is presented in [Schwartz 19571,
[Gelfand and Shilov 1964].
2. The domain D should contain the vacuum state vector and be invariant
under the application of the operators U(a, a) and 0(f).
Of course one will usually have to deal with several fields, each of which
may have several tensor or spinor components. Correspondingly we must
take test functions for each type (index i) and each component (index a)
and understand 0(f) generically as

0( f) = E f(x)f(x)d4x. (II.1.7)

Remarks: Renormalization theory suggests that it is essential to smear out 0

both in space and time, in contrast to the case of free fields, where an averag-
ing over 3-dimensional space at a fixed time is sufficient. Due to the stronger
singularities (see above) one cannot assume well defined commutation relations
of fields at equal time.

C. Hermiticity.
The set of fields contains with each 0 also the Hermitean conjugate field
0*, defined as a sesquilinear form on D by

(W2I 0* (x)IWi) = (Wii 0(x)IW2)- (II.1.8)

D. Transformation Properties.
The fields transform under as
U(a, a)02),(x)U_1(a, a) = 1141}e(a-1)Ve(A(a)x + a),

here M(a) is a finite dimensional representation matrix of a E 2. Of

course this should be properly written in terms of the smeared out fields
but this rewriting is evident using (II.1.7).
E. Causality.

The fields shall satisfy causal commutation relations of either the bosonic
or fermionic type. If the supports of the test functions f and h are space-
like to each other then either

[0'(f ), (h)] = 0
or
[0i (f ), v(h)1+ = 0.
58 II. General Quantum Field Theory

F. Completeness.

By taking linear combinations of products of the operators O(f) one

should be able to approximate any operator acting on it This may be
expressed by saying that D contains no subspace which is invariant under
all 0(f) and whose closure is a proper subspace of ?l. Alternatively one
may say that there exists no bounded operator which commutes with all
0(f) apart from the multiples of the identity operator (Schur's lemma).

G. "Time-slice Axiom". "Primitive Causality".

There should be a dynamical law which allows one to compute fields at

an arbitrary time in terms of the fields in a small time slice

Ot,,. ={x: Ix° — ti < 6l. (II.1.12)

Therefore the completeness, demanded under F should already apply when we

restrict the support of the test functions f to O. This postulate was Added
somewhat later and in most of the work analyzing the consequences of the ax-
ioms it was not used. In chapter III we shall introduce a sharpened version of
this postulate, taking the hyperbolic character of the dynamical laws in rela-
tivistic theories into account.

Remarks: The endeavour to study the consequences of the assumptions A to

G (possibly supplemented by further assumptions concerning the particle spec-
trum and the physical interpretation) has, following Wightman, been commonly
called axiomatic quantum field theory. This terminology is appropriate in the
sense that the starting point, the total input, is clearly stated. On the other hand
the word "axiom" suggests something fixed, unchangeable. This is certainly not
intended here. Indeed, some of the assumptions are rather technical and should
be replaced by more natural ones as deeper insight is gained. We are concerned
with a developing area of physics which is far from closed and should keep an
open mind for modifications of the assumptions, additional structural principles
as well as information singling out a specific theory within the general frame.
We shall endeavour to do this in later chapters and, following R. Jost, we prefer
to use the term "general quantum field theory" rather than "axiomatic quan-
tum field theory" because it allows more flexibility. We regard the postulates
A—G as well as subsequent modifications and further structural assumptions
as working hypotheses rather than as rigid axioms.

II.2 Hierarchies of Functions

The impact of covariant perturbation theory and the analysis of the conse-
quences of the axioms lead in the fifties to a description of quantum field theory
 II.2 [Link] of Functions 59

in terms of infinite hierarchies of functions.' Different choices of hierarchies were

considered; the properties of the functions and their interrelation were studied,
using as an input the general postulates (axioms supplemented by more specific
information coming from experience with renormalized perturbation theory).
The present section is intended as a brief description of the resulting picture.

II.2.1 Wightman Functions, Reconstruction Theorem,

Analyticity in z-Space

We shall suppress the indices i, a distinguishing different fields and components

whenever they are not absolutely essential. Let us denote the state vector of the
vacuum by O. The vacuum expectation values of products of fields

wn (xl , • • . xn) = (,f0 ( x 1 ) . • . 0 ( xn )1,0 ) (II.2.1)

are tempered distributions over IR 4n. This follows directly by the "nuclear the-
orem" of L. Schwartz from axiom B. The tun are called Wightman distribùtions
because, in a fundamental paper, Wightman pointed out the following facts
[Wight 56]:
First, if the hierarchy of distributions wn (n = 0, 1, 2, • • .) is known then the
Hilbert space and the action of 0(f) in it can be reconstructed. This reconstruc-
tion theorem is closely related to the Gelfand-Naimark-Segal construction which
will be explained in the next chapter. The main point can be simply described.
Due to the completeness axiom F the linear span of vectors

cP(fl) ... i(fk)Q, k = 0,1, • • • co (II.2.2)

is dense in 1-1. If 0(x) is contained in the set of fields so is *(x) (axiom C);
hence the scalar product of two state vectors of the form (II.2.2) is known once
we know all the tun and the same is true for the matrix element of any OW
between such state vectors.
Secondly, the wn are boundary values of analytic functions of 4n complex
variables. We sketch the argument using freely somewhat heuristic manipula-
tions. A precise discussion is given in the books [Streater and Wightman 1964],
[Jost 19651.
The vacuum is translation invariant

U(a)S2 = O. (II.2.3)

Due to the translation covariance of the fields (axiom D)

0(x) = U(x)0(0)U -1 (x). (II.2.4)

Expressing U(x) in terms of the generators as in (I.3.10) we may write

'The term "function" is understood here in a broad sense. It includes generalized functions,
distributions.
60 II. General Quantum Field Theory

_ (fl^^(p)e-zP^^^-^^}^(p)e-^ p(x2-s3) . . • ^(0) Q)

= Wn (e 1 , • • • en -1)
with
ek = xk — xk+1 •

By a Fourier transformation

gi
Wn(el, • • • Sn -1) = J Wn(g1, • • . qn-1)e'1E4k kd4 .. . d4q^ 1-

(II.2.5) becomes

Wn(gla (in i)_ (2 7) - 4(n - 1 ) f en

Wnl g"kd4e1 ...
d4en -1

= (Q1cP(0)64(P — g 1 )cP(0)64 (P — q2) ... 6 4 (P — qn-1) 0 ( 0) fly. (II.2.8)

Therefore Wn vanishes when any one of the qk lies outside of the spectrum of P
and thus, by axiom A3, at least when any one of the qk lies outside the forward
cone. If we now take complex 4-vectors

zk = ek — ink (II.2.9)

instead of the real vectors ek then the integrand on the right hand side of (II.2.7)
becomes
Wn e - i EgkG e -
> gk'lk
If all rik are taken as positive time-like vectors then we get in (II.2.7) an expo-
nential damping factor and the integral, as well as its derivatives with respect to
any zk , converges because Wn is a tempered distribution. Thus Wn
is an analytic function of all its arguments as long as the negative imaginary
parts of all the zk lie in the forward cone.
Actually the domain of analyticity of these functions is much larger than
the primitive domain

Tn —Im zk E V+ ,k = 1, • • • n — 1 (II.2.10)

found by the above argument. As a first step one shows that the domain of holo-
morphy includes all "points" z = (z1 ... zn_ 1 ) which arise from a point in the
,

primitive domain Tn by a complex Lorentz transformation which is connected

to the identity, i.e. by application of any complex matrix A with

AT gA = g; det A = 1 (If.2.11)

to all the zk . By axiom A2

U(a)fl = D. (II.2.12)
Therefore, according to D (Equ. (II.1.9)) a Wightman function with Lorentz
transformed arguments is a linear combination of other Wightman functions
with different components of the fields. The coefficients arise from products of
 II.2 Hierarchies of Functions 61

representation matrices of Z. One notes that the number of spinorial fields in wn

must be even for non-vanishing wn and therefore the product of representation
matrices belongs to a true (tensorial) representation of Z. The coefficients are
then single valued, analytic functions and the relation (II.1.9) may be extended
to complex Lorentz transformations. The resulting domain reached by the ap-
plication of complex Lorentz transformations to the primitive domain is called
the extended tube Tn'. It contains real points, the so called Jost points. These
are the points where in (IL2.9) all 4 as well as their convex combinations are
space-like and all r1k = 0 :

z:7]k=0; k< 0 ; (>.xkk) 2 <o for ak>0, E ak>0}.

(I1.2.13)
A further extension of the domain of holomorphy results from local corn-
mutativity (axiom E). A permutation of the arguments of wn again gives a
Wightman function. Since the definition of the extended tube depends on the
order of arguments in wn, wn will be an analytic function of the original ar-
guments in a different domain. Due to E this function will coincide with the
original one for space-like separation of the x k and a careful study shows that
there is a real open neighbourhood in the Jost regions of both extended tubes
where the two functions agree. This is enough to conclude that the two func-
tions are analytic continuations of each other. The domain of holomorphy of wn
is thus extended to the union of the "permuted extended tubes".
In the theory of functions of several complex variables a basic observation
is that the domain of analyticity cannot have an arbitrary shape. If we know
that a function is analytic in a certain region then it must be automatically
analytic in a (usually larger) region, the so called holomorphy envelope of the
given region. This leads to a further enlargement of the holomorphy domain
of wn. By these methods considerable information about the functions w' for
small n-values has been obtained.
Apart from the restrictions for the individual functions wn of the hierarchy
(analyticity, transformation properties) there is an infinite system of inequalities
between the different wn which must hold in order to make the metric in the
Hilbert space, constructed from the wn, positive definite. They are summarized
in the demand
(fl tA*Alf2) > 0 for all A= E ff n(x l , • • • x n ).p(x l ) ••• 0(xn) 11 d4xk.
(II.2.14)
The analysis of these restrictions, directly in terms of the wn, was sometimes
called the "nonlinear programme".

II.2.2 Truncated Functions. Clustering

Consider a configuration (x 1i • • • xn ) consisting of several clusters. A cluster is

a subset of points such that all the points in one cluster have a large space-like
separation from all the points in any other cluster. Then we expect that
62 II. General Quantum Field Theory

w n (x1 , ... x n ) ... H wnr yr,1 , ... yr,nr)

( ,
(II.2.15)

the approximate equality becoming exact in the limit of infinite separation

between the clusters. In (II.2.15) the index r labels the clusters, n,. is the
number of points in the r-th cluster and yr,k , k = 1... nr, a subset of the
x k , denote the points belonging to this cluster. The reason for this asymptotic
factorization is that the correlation between two compatible observables A, B
in a state is measured by the difference
(AB) — (A) (B) (I1.2.16)
where the bracket denotes the expectation value in the state. We expect that
in the vacuum state the correlations of quantities relating to different regions
decrease to zero as the separation of the regions increases to space-like infinity.
This physically plausible guess is, in fact, not an independent assumption but
follows from the axioms as will be discussed in detail in section 4.
It is therefore convenient to introduce another hierarchy of functions {wT}
in terms of which the hierarchy {wn} is obtained as
wn (x1, • • .xn) = EP Hr WT (yr,k)• (II.2.17)

Here P denotes a partition of the set of points {x i } into subsets (potential

clusters), labeled by the index r. The sum extends over all possible partitions.
We may note that wT = 0, w° = 1. We call Or' the truncated functions or
correlation functions. Given the hierarchy {wn} the truncated functions can be
computed recursively:
u4 (x) = wl (x)
u4( x i , x2) = w2 ( x 1 , x2 ) — w 1 1 x 1) w 1 (x2) (II.2.18)

The term wT always occurs on the right hand side in (II.2.17) coming from
the trivial partition where all the points are taken together. The other terms
involve truncated functions with a smaller number of arguments. So (II.2.17)
may be inverted, beginning with (II.2.18). The asymptotic property (II.2.15)
of the wn corresponds now to the simpler property
wT(x 1 , • .. xn ) —> 0 (II.2.19)
when any difference vector x i — x i goes to space-like infinity. Actually the cor-
relation functions in the vacuum tend also to zero with increasing time-like
separation of the points. The rate of decrease in space-like and time-like direc-
tions depends on the details of the energy-momentum spectrum of the theory.
We shall return to these questions in section 4 of this chapter.
For free fields the situation is particularly simple. All truncated functions
with n 2 vanish:
wT =O for n 542. (II.2.20)
Thus, for a system of fr ee fields it suffices to know the 2-point functions. The
others are then given by (I1.2.17) with the additional simplification that only
such partitions contribute which are pairings of points.
 II.2 Hierarchies of Functions 63

Generating Functionals. Linked Cluster Theorem. Let us consider a hier-

archy of functions pn(x i , . . . xn ), totally symmetric under permutations of their
arguments. 2 This set of functions may be assembled into one generating func-
tional P{ f } of a function f (x) :

P{ f } = f pn(xi, • xn) f(xi) • • • f (xn)dxl ... dx n . (II.2.21)

n=o
From it we can recover the pn as the Taylor coefficients
n
pn (xi,... xn) = P{f} (II.2.22)
f(xl) ... b .f(xn) f.„0

The relation between the hierarchy {pn} and the hierarchy of truncated func-
tions {pT} is very simply expressed in terms of the respective generating func-
tionals P and PT , namely

P{f} = exp PT{f }; PT{f} = 1nP {f }. (IL2.23)

Note that we assume here p ° = 1, pz, = O. This relation, sometimes called the
linked cluster theorem, combines in an elegant way the combinatorics involved
in the definition (II.2.17) and its inversion (11.2.18).
We give an example of the application of (II.2.23) taken from the classical
statistical mechanics of a 1—component real gas. It is the example from which
the idea of introducing truncated functions originated (see Ursell [Urs 27], Mayer
[May 37], Kahn and Uhlenbeck [Kahn 38]).
A grand canonical ensemble with inverse temperature ,Q and chemical po-
tential µ defines a hierarchy of functions pn, the probability densities for find-
ing n, molecules with the configuration x i , • • • xn in 3-dimensional space. If the
molecules interact by two-body forces, given by a potential V, the Gibbs pre-
scription leads to
pn (xz • • • xTL) = v— ne —p/2 ^ i^i V^^
, (II.2.24)
Here = V (xt — xi ) . The constant y has the dimension of a volume; it is
proportional to f33/2 e -431' and is approximately equal to the specific volume per
particle in the gas. The grand partition function
1
G =E fpn(x i , . • . xn)d 3 x2 • • . d3 xn = 'P{1} (II.2.25)

is precisely the generating functional taken for the constant function f = 1.

Thus, according to (II.2.23) , the "grand potential" Q, defined as the logarithm
of G, is given by
2 The Wightman functions of a single field are totally symmetric in their analyticity domain
due to E. This will be used in subsection 2.7. They are of course, not symmetric for real
points with time-like separation because a permutation changes the direction from which
the boundary value should be approached. Examples of symmetric functions of real points
in Minkowski space are the time ordered functions discussed below. The generalization to
several fields is trivial. We just have to replace f by the multicomponent function PA as in
(II.1.7).
64 II. General Quantum Field Theory

Q = In G = PT{1} = E n! f pT{x l .. . x,,,)d3x1 ... d3x.n . (11.2.26)

The essential point is now that the truncated functions are practically zero
unless the points in their argument form a linked cluster i.e. unless they can be
connected by a bridge which at no step is wider than the range of the potential.
Therefore the sum in (II.2.26) is essentially an expansion in powers of a 3 /v
where a is the range of the potential. In the case of low density this converges
fast and only a few terms are needed in the sum over n to obtain the grand
potential, from which all thermodynamic information can then be obtained by
differentiating with respect to (3 and µ. The first truncated functions are easily
computed from (II.2.18), beginning

pT = u-1; p 2 = u -2 ( e /2V1_c2) _ 1) ; ... (11.2.27)

II.2.3 Time Ordered Functions

This is the most widely used hierarchy of functions in quantum field theory.
It is computed in the Feynman-Dyson form of perturbation theory and it is
closely related to the scattering matrix by the LSZ-formalism. One introduces a
time-ordering operation T which, applied to a product of field operators 45(x),
shall mean that the actual order of the factors is taken to be the chronological
one, beginning with the earliest time on the right, irrespective of the order in
which the factors are written down. In the case of Fermi fields a sign factor —1
has to be added if the permutation needed to come to from the original order to
the chronological order of the Fermi fields is odd. For example, for Bose fields
one defines 3
0(x 1 )cP(x 2 ) if t 1 > t2
(x l )^{x 2 ) _ P(x2 )P(x l ) if t 1 < t2.
5(x1)0(x2) T.1 .2. 28 )
(II

If both fields are Fermi fields then the second line will be —P(x 2)0(x i ). The
vacuum expectation values of time ordered products will be called ir-functions.
They are often also refered to as Green's functions or Feynman amplitudes:
rn(x l , • • • x,) = (Q1T0(x i ) • • • 0 (x.)1 0). (1I.2.29)
Of course the cluster decomposition leading to the hierarchy of truncated func-
tions proceeds in exactly the same way as for the Wightman functions. More-
over, since the ir-functions are symmetric under permutation of their arguments,
they can be subsumed by a generating functional and the linked cluster theorem
can be applied.
t1 = t2 the upper and the lower line on the right hand side of (II.2.28) agree according
3 For
to axiom E as long as the points do not coincide. For coincident points there is an ambiguity.
The T-product is not defined as a distribution by (II.2.28). This ambiguity has been used
by Epstein and Glaser in their treatment of renormalized perturbation theory [Ep 71]. Apart
from the ambiguity for coincident points the definition of the T-product is Lorentz invariant
because the chronological order depends on the Lorentz frame only when points lie space-like
to each other and then the reordering is irrelevant.
 II.2 Hierarchies of Functions 65

11.2.4 Covariant Perturbation Theory. Feynman Diagrams.

Renormalization

This is the backbone of the study of experimental consequences of QED and has
remained one of the essential sources of guidance in the subsequent development
of quantum field theory. Although it is described in many books we give here
a brief sketch based on the "magic formula" of Gell-Mann and Low [Gell 51]
using the terminology described above. We assume that the theory is obtained
from a classical field theory with a Lagrange density

L = L0 + LI. (II.2.30)

G° is the Lagrange density of a free field theory (compare (I.5.30)). For simplic-
ity we assume that no derivatives occur in .C I . We shall denote the interacting
fields generically by 0(x) and the corresponding free fields, which result if we
drop the interaction Lagrangean .CI , by cP°(x). For the generating functional of
the 'r-hierarchy of the interacting fields we shall write T{ f }. Gell-Mann and Low
loc. cit. have written down an explicit expression for this functional in terms
of quantities defined in the free field theory. This "magic formula of covariant

N{o}
perturbation theory" (still without renormalization) is

T{ f } = (11.2.31)
{}

N{ f} = (Q0IT ei 0 (f) +i f E7(x)`4x I Qo) (II.2.32)

Here Q0 is the vacuum state vector of the free theory, .C7 means that the free
fields P° (x) instead of the interacting fields 0(x) are inserted into the expression
for the interaction Lagrangean density. 4 The formula (II.2.32) is symbolic in
the sense that the exponential should be expanded into a power series and then
in each term the chronological ordering should be applied. The term with n
factors 0° and p factors L7 gives the contribution to rn in the p-th order of
perturbation theory. It is
'rn(xl , ... xn )(n)

=
^ p!N {0}
f (QoT 0 (xi) . • • 0°(xn) ,c7(yi) • • ,c7(yp)I f207d 4 yr • . • d 4
yp•

(1I.2.33)
The integrand on the right hand side is a r-function of free fields and can be
evaluated easily using the cluster expansion, the analogue of (II.2.17), remem-
bering that for free fields all truncated functions except the 2-point functions
vanish. The integrand of (II.2.33) reduces therefore to a sum of products of
2-point r-functions (Feynman propagators). Each term in this sum corresponds
4 G9(x) is a polynomial of fields at the same point. Since the fields entering here are
free fields a prescription defining the product exists, namely the Wick ordering. We shall
understand GÎ as the Wick ordered product of its factors.
66 II. General Quantum Field Theory

f X3
tl

f X3

X1 Y2 X2 X1 Y1 y2 Y3 X2

(a) (b)

f X3

f X3

^• < ^ x
X1 X2 X, Y1 X2
Y1

(c) (d)
Fig.11.2.1.

to a complete pairing of the occurring fields and can be characterized by a Feyn-

man diagram: the arguments x1, yj are represented by points, a contraction (a
pairing of two fields) by a line connecting the respective points.
Let us illustrate this in QED where
— o ry1- ii;°Aµ
C°=e:^ :. (II.2.34)
Then each of the points y 1 (vertices) is the source of three lines, one corre-
sponding to , one to Tb and one to A. The only nonvanishing propagators
are the contractions of V) with V and of A with A. This is incorporated in the
graphical representation by using for a solid line with an arrow pointing to-
wards its source, for i a solid line with an arrow pointing away from its source
and for A a dashed line. Thus, for instance, for the 3-rd-order contribution to
(,fl 1Tt (x l ) 3 (x2) A„ (x3 )1,fl) i.e. for the case n = 3, p = 3, one has the following
four possible diagrams 5 (apart from a permutation of the y 3 ).
If we go over to the truncated r-functions the situation simplifies still fur-
ther. The contribution of a disconnected diagram (like the one in Fig. II.2.1.c)
is just the product of the contributions of its connected parts. This is eliminated
in the truncation process. So we may write
Usually one does not mark the points x 4 in a Feynman diagram. The lines emanating
5
from them are called external lines.
 II.2 Hierarchies of Functions 67

TT { f } = E conn. Feynman diagrams, (11.2.35)

which is a short hand notation for saying that a function TT results from sum-
mation over all the connected Feynman diagrams with n end points (external
lines) in the expansion of N{ f }. In particular, the normalization factor N{0}
is then eliminated. It corresponds to the sum over all Feynman graphs without
external lines.
The connected graphs still lead to divergent integrals if they contain loops of
certain types. Renormalization theory gives rules for removing these infinities.
To convey some flavour of the computations we illustrate them in the example
of diagram II.2.1.a. Its contribution is 6

f {SF(x1 — y1)1/ ° SF(yi — y3)`eSF(y3 — y2)eY6SF(y2 — x2)}as

x g orDF(y3 — x3)9 A vDF(yi — y2)d4yid4y2d4 y3 • (11.2.37)

By Fourier transformation this becomes

f exp ( —
z (pi (x1 — x 3 ) + p2(x3 — x2))) DF(P1 — P2)

x SF(p1) f„(pi,p2)SF(p2)d4pld4p2 (1I.2.38)

with

p2) = Ye f SF(q) Y„SF(q + p2 p1)DF(pi — q)d 4q 7g. (11.2.39)

Inserting the expressions (1I.2.36) for the propagators one sees that there are
two powers of the components of q in the numerator, six in the denominator
of the 4-dimensional integration. The integral diverges logarithmically at large
values of q. However, the difference of T„ at two different values of the arguments
pi gives a convergent integral and one may write formally

ri = C'y v + ruen (ial , p2) (11.2.40)

where c is an infinite constant and rren is a finite function. The divergent

constant part of r„ corresponds to the replacement of the triangle in the diagram
by a single point with the local interaction c/(y)ry „0(y)A„(y) which is of the
same form as the basic interaction Lagrangean and may be regarded as an
(infinite) change of the parameter e ("charge renormalization") or cancelled by
an (infinite) local counter term in the Lagrangean.
To show that a finite number of local counter terms in the Lagrangean
suffices to subtract the infinities from all diagrams in a consistent manner is
6 SF is the contraction between and V, g,,„DF the contraction between A4 and A. The
Fourier transforms of these Feynman propagators are
-k
SF(q) ~ i(quryu — m)(q2 — m2 + is)-1; DF(q) = (q2 + iE) . (II.2.36)
68 II. General Quantum Field Theory

a rather involved combinatorial problem. It is achieved by a structure analysis

of Feynman graphs, started by Dyson [Dy 49] and carried further by many
authors.
We have presented the magic formula (I1.2.31), (I1.2.32) without giving any
reasons. Let us try to justify it. For this purpose the simplest starting point is
the so called "interaction representation" developed mainly by Tomonaga and
Schwinger. To avoid confusion with other uses of the term "representation" we
shall call it "interaction picture" or "Dirac picture". In the previous sections
we used the "Heisenberg picture" for the description of the dynamics and we
shall adhere to this with few exceptions throughout the book. In this picture
a state is described by a vector (or density matrix) which does not change in
the time interval between the completion of the preparation of the state and
the subsequent measurement of an observable. The operator A representing an
observable depends on the time of measurement. It changes with time according
to the Heisenberg equation of motion
ôA _
(1I.2.41)
ât — z [H, Ai.
In non relativistic Quantum Mechanics the most commonly used description
is the "Schrodinger picture" where the state vector changes according to the
Schrödinger equation
aw
at
= — iHW,
whereas the operator A describing an observable depends only on the placement
in 3-space, not in the time of measurement. Dirac's time dependent perturba-
tion theory suggests a third picture which provided the intuitive basis for the
development of the covariant perturbation expansion in Quantum Field Theory.
If the Hamiltonian is split in the form

H = Ho + Hi ,

where Ho is the Hamiltonian of a free field theory serving as the zero order
approximation and H1 is regarded as the interaction, then one may divide the
time dependence between observables and states, letting the observables change
as in the free field theory and expressing the rest of the dynamical law as a
change of the state vectors. We distinguish the quantities in the three pictures by
an upper index H, S, D, respectively. In relating the mathematical expressions
for the observables in the three pictures it must be understood that we talk
about observables which are expressed in terms of a basic set of quantities
referring to a sharp time, for instance the basic fields and their canonically
conjugate momenta at the respective time. Note that in the Heisenberg picture
Ho and H1 are not fixed operators in 7 -L and that in the interaction picture the
same holds for H and H1 . So to avoid confusion we should regard these symbols
as denoting fixed functionals of the basic fields and momenta at a specified time
and we shall express this by writing for instance Hj {cP' ) (t) } for the interaction
Hamiltonian at time t in the Dirac picture. Since we assumed for simplicity that
 II.2 Hierarchies of Functions 69

the interaction Lagrangean does not contain any derivatives of fields the form
of HI is
11I {0(t)} _ — f GI(sP(x)d3x with x° = t. (11.2.42)
A (D ) (t) and A(H)(t) are related by a unitary operatur U(t),

A ( D) (t) = U(t)A (1-1) (t)U(t) -1 .

The time dependence of U is obtained by comparing (II.2.41) with the corre-

sponding relation where A (H) is replaced by A(D ) and H by H° . We have

aA(D) (t) [H0 {D( t) }, A(D) (t)]

at

= aU A(H) (t) U- 1 + iU [H{((t)}, A (H)(t)^ U1 - UA ^H ) (t)U -1 U--1

at
= at U-1A'D) (t) + i [H{cf(''(t)} , A( D) (t)1 — A(D) (t) U-1.
at

at time t we have

Defining
aU(t)
at
_
Since H — H° = HI and A(t) is an arbitrary functional of the basic quantities

_iHI {o(D)(t)}U(t). (II.2.43)

U(t2 r t 1 ) = U(t2)U (t1) -1 ,

(1I.2.44)
we get an operator which establishes the relation between A (D) and A ( H) at time
t2 if t1 is chosen as the reference time at which A(D ) = A(H) . The information
from the differential equation (II.2.43) and the initial condition U(t,, t 1 ) = It
are combined in the integral equation

U(t2 , t 1 ) = Il — i f t2 HI{0 (D) (t')U(t', t1 )dt'.

t,
(II.2.45)

It can be solved by iteration, yielding the perturbation expansion (for t2 > t1 )

00

U= Il + U(n),
.
n-1 (II.2.46)
U(n) (t2, t1) = ( — Z) n HI(t'n) . . . HI(ti)dt1. . . den .
t 2 >t;n >...ti>tl

Using the time ordering symbol T and inserting (11.2.42) we get

n
U(n) (t2 , t1) = Tri(xi) . . . GI(x n)d4x1 . . . d4 xn, (1I.2.47a)
t2 >x?>tl

or, symbolically
U(t2, t 1 ) = T exp i f G(x)d4x. (II.2.47b)
t2 >2o>tl
70 II. General Quantum Field Theory

We have omitted the index D since all fields are now understood in the Dirac
picture i.e. they are free fields. From the definition (II.2.44) follows the compo-
sition law
U(t3, t2 )U(t2 , t1) = U(t a , ti). (II.2.48)
For Heisenberg fields coinciding with 0(D) at a time t o we get (for x2 > x?)

0(11)(x2)0(11)(x1 ) = U(x2, to) -1 45(D)(x2)U(x2, x°)0(13)(x 1 )U(x°, to).

Choosing some arbitrary final time tf larger than x2 we may write the first
factor as U(t f, to) -1 U(t f, x2) and write symbolically in the case of n factors

TO(H) (x l ) ... 0(H) (x n )

(II.2.49a)
= U(t f, t o ) - 'T0 (D) (x l ) ... 0 (D) (xn ) expi GI(x)d 4x
t f >x°>to

which is generalized for the generating functional of T-products of Heisenberg

fields to

T exp i J '(x)h(x)d 4x = U(t f to ) -1 T exp i f

,

t f >x°>to
(11(x) + 0(x)h(x)) d 4x.
(II.2.49b)
If all the steps in the above argument were well defined then this remarkable
formula would give us operators 0(11) (x) in Fock space, obeying the Heisenberg
equations of motion in the time interval between to and tf , explicitly expressed in
terms of free fields. Furthermore these fields would transform covariantly under
the representation of the Poincaré group in the Fock space of free particles as
long as the shifted arguments of the fields 0 stay in the time interval between
to and tf. This latter circumstance would imply that in the limit to — ^^ —oo
tf — . +oo the physical vacuum should be represented by the Fock vacuum since
this is the only Poincaré invariant vector in Fock space. Thus we would obtain
the formula by Gell-Mann and Low.
The problems with the argument come from three sources. The first con-
cerns the ultraviolet divergencies. GI (x) is a Wick product of free fields. We
shall call any Wick polynomial of the free fields and their derivatives at a point
a "generalized field" or "composite field" (compare the end of section II.5.5).
Any such field is an operator valued distribution on a dense domain of 7-1 (see
section 1I.1.2). Ordinary products G1(xi) ... Gk(xk) of such composite fields are
also well defined distributions for test functions of class 84k i.e. smooth func-
tions f (x i , ... xk) of fast decrease. The time ordered product, however, differs
from this by commutators of the Gj multiplied by step functions in the time
differences. Their discontinuity introduces an ambiguity. Since the commutators
of fields vanish for space-like separation of points the discontinuity of the step
functions can only introduce a possible ambiguity when two or more points in
the product of the Gi coincide. The ambiguities will therefore be of the form of
some other composite field multiplied by an undefined (formally infinite) con-
stant. The amazing fact is now that there exist finite sets of composite fields L i
such that the ambiguous terms arising in the T-products between (j=1.N)
 II.2 Hierarchies of Functions 71

any of them lead only to composite fields within the same set. If the original
Lagrangean belongs to such a set then the theory can be "renormalized" by
redefining the T-products, replacing inductively in each subsequent order of
the perturbation expansion the formally infinite coefficients of the troublesome
terms by unknown finite constants. Then one obtains in each order a well de-
fined expression involving a certain number (maximally N) unknown constants
which cannot be determined from the theory but have to be fixed empirically.
The existence of renormalizable theories results from the fact that one can
order the composite fields by their canonical dimension. A naive power counting
argument indicates that in the T-product of two fields with dimensions d1, d 2
1 + d2 — k thecompsifldwabguoscefinthavdmso
with k > 4. So the theory has a chance of being renormalizable if the dimension
of GI is less or equal to 4, which means that the coupling constant should be
dimensionless (or have the dimension of a positive power of energy). A precise
proof of renormalizability is a highly nontrivial endeavor.
The second source of difficulties concerns the integration over infinite space-
time. On the level of the operator formulas (II.2.49) this may be remedied
by replacing the integral over GI be f G j (x)g(x)d 4x where g is a smooth test
function which is constant in a large region and vanishes at infinity. Then the
formulas give operators which satisfy the Heisenberg equations of motion in a
large region. The Poincaré covariance is modified and there is the problem of
defining the physical vacuum. We shall not discuss this further. In some models
of field theories this problem may present a serious obstacle as indicated by the
slogan of "infrared slavery" in Quantum Chromodynamics.
The third question is the convergence of the perturbation series. For this
we have to refer to work in Constructive Quantum Field Theory. For realistic
theories the answer is presumably negative.
There has been a great effort to divorce the formalism from the canonical
quantization of a classical field theory, and to derive from general principles the
structural relations between amplitudes with different number of arguments,
implied in the diagrammatic description. If one writes the T-functions as a power
series with respect to some parameter, starting with the T-functions of a system
of uncoupled free fields as the zero order terms, then the general postulates
A-E, when supplemented by conditions limiting uniformly in all orders the
growth (of the Fourier transforms) for large momenta, lead to the renormalized
perturbation series with only a few possible choices of the interaction terms.?

II.2.5 Vertex Functions and Structure Analysis

Feynman diagrams suggest that T-functions may be constructed from more

primitive building blocks. A first step was the truncation process, reducing the
problem to the consideration of connected diagrams only.
7 lmportant contributions to this program are due to Bogolubov and Parasiuk, Zimmer-
mann, Symanzik, Hepp, Epstein and Glaser, Steinmann. For detailed accounts see [Bog 571,
[Bogolubov and Shirkov 1959], [Hem) 1969], [Zi 70], [Sym 70, 71], [Ep 71], [Steinmann 1971].
72 II. General Quantum Field Theory

Fig. II.2.2.

tadpole irreducible s.e. part reducible s.e. part

Fig. 11.2.3.

To carry this one step further call two points of a diagram strongly connected
if, by cutting a single line, the graph cannot be disconnected so that the two
points lie in different pieces. One sees that the strong connection property is
transitive. If x is strongly connected to y and y to z then x is strongly connected
to z. So one can divide the points of a diagram uniquely into subsets of strongly
connected points. If we picture each such subset by a circle and omit all lines
connecting points within such a subset then the original diagram reduces to a
tree graph i.e. a diagram without closed loops (Fig. II.2.2.). Let us call such a
circle an f-vertex if f lines emanate from it. The cases f = 1 and f = 2 are
special. A 1-vertex ("tadpole") occurs only as an end point of a diagram A
2-vertex ("irreducible self energy part") followed by other 2-vertices still gives a
subgraph with only two free lines. If we add the contributions from all possible
diagrams to a truncated n-point function then the result can be expressed as
a sum of contributions of tree graphs in which internally only f-vertices with
f > 3 occur and where the connecting lines correspond to the full 2-point
T-function instead of the free Feynman propagator. An f-vertex corresponds
analytically to a vertex function Ef(x i , . • • xf). It is the sum over all 1-particle
irreducible diagrams8 which allow one external line to be attached to each of
the points x k (k = 1, • • • f). These attached lines are all considered as cut
(amputated). The conclusion of this analysis is that the T-functions can be
constructed from the hierarchy of vertex functions E f with f > 3 and the full
2-point function. The relation between the T-hierarchy and the E f-hierarchy
can also be expressed in terms of their generating functionals [Sym 67]. See also
[Itzykson and Zuber 1980].
8 In
our context we should better call it 1-line irreducible. It means that we cannot dis-
connect the subdiagram by cutting a single line. In the intuitive picture a line is thought to
correspond to a virtual particle. Symanzik carried the structure analysis further by considering
2-particle irreducibility next.
 II.2 Hierarchies of Functions 73

Apart from simplifying the book-keeping and the discussion of renormaliza-

tion in perturbation theory the vertex functions have the advantage of faster
decrease for large separation of their points (also in time-like directions) as
compared to the r-functions. The lines in a diagram may be considered as ties
between the points they connect. The more ties there are between two points
the tighter they are pulled together.

II.2.6 Retarded Functions and Analyticity in p-Space

Another hierarchy, used especially in the study of dispersion relations are the
r-functions, the vacuum expectation values of retarded, iterated commutators
(resp. anticommutators); see Lehmann, Symanzik, Zimmermann [Leh 57], Nishi-
jima [Nish 57], Glaser, Lehmann, Zimmermann [Glas 57]. They are defined as

r (n) (x? xi, . . . Xn -1)

=( - i)n - i E O (^ - ^1 11 . • . ® (xn -2 - Xn -1)

Perm 1

X PI [• • • [[ 45 (x), 45 (x1)]? 0(X2)1 . • . 0 (xn-1)] I '°) (1I.2.50)

Note that the first argument is distinguished. The symmetrization (sum over
permutations) refers to the remaining arguments x1 , • • • xn _ i . O is the step func-
tion in time:
e( x ) — 1 if x° > 0 (I1.2.51)
0 if x° < 0.
Using space-like commutativity (axiom E) and the Jacobi identity for iterated
commutators one shows that r(n) vanishes unless all the difference vectors

ek = X — Xk (II.2.52)

lie in the forward cone and that r(n ) is a Lorentz covariant function of the 6,
(apart from the ambiguity for coincident points which r-functions share with
r-functions). The support of the r-functions in x-space is analogous to that
of the w-functions in momentum space. It is confined to the forward cone in
all the k. Thus the Fourier transforms of the r(n) with respect to the ek, the
r-functions in momentum space, are analytic in an "extended tube".

II.2.7 Schwinger Functions and Osterwalder-Schrader Theorem

While the r-functions have gradually drifted into oblivion the "Euclidean
Green's functions" or "Schwinger functions" have moved to the center of the
stage. The Minkowski trick of replacing the time coordinate x ° by

^4 = 2^ ° (II.2.53)

changes the Lorentzian metric to the Euclidean metric

74 II. General Quantum Field Theory

(x, x) = — E(?) 2 , A = 1,2,3,4. (II.2.54)

Configurations of points x k with real coordinates xk (A = 1, • • • 4) lie in the

"permuted extended tube", the analyticity domain of the Wightman functions
described in subsection 2.1. We denote the Wightman functions for configura-
tions of points x k in real Euclidean space (real coordinates xk, A = 1, • • • 4)
by
Sn(x 1 • •Xn)=w^( — ix14 ,X1,••• — iXn4 7 Xn ).
,
• (II.2.55)
These functions have some remarkable properties.
(i) If we exclude the coincidence of points i.e. remain outside of the sub-
manifold
4={x 1 ,••• x.^ : xk = xi for some k j} (II.2.56)
then the Sn are analytic functions. We may regard Sn as a distribution over
1R471 -- A. The appropriate space of test functions is S (1R4 — d), the smooth
functions decreasing fast for large x and vanishing with all their derivatives on
A.
(ii) S' is Euclidean invariant.

Sn(gx 1 , . . . gxn ) = Sn( x1, ... x,a), (II.2.57)

where gx = Rx + a; R E SO(4).
(transcription of axiom D).
(iii) Sn is symmetric under permutations.

1 ... n
Sn (x P 1 , • • • xp n ) = Sn(x1 , • • • x,a) for any permutation
P1 Pn '
(II.2.58)
(transcription of axiom E).

Thus Sn for a general configuration of points is immediately given if it is

known for (Euclidean) time ordered configurations in the positive half space

Û<x^<x^ -1 <...x 1 • (II.2.59)

We denote the test functions with support in (II.2.59) by S ± (1114n). So S'b may
be regarded as a version of Tri with complex arguments and this was the original
point of view of Schwinger [Schwing 59] in considering this hierarchy. See also
Ruelle [Rue 61], Symanzik [Sym 66].
Due to the permutation symmetry (iii) one can combine the hierarchy {Sn}
into one generating functional 6{f} and one has the linked cluster theorem.
There is an analogue of the magic formula of Gell-Mann and Low, expressing
S{ f } formally explicitly in terms of the interaction Lagrangean. This is the
Feynman-Schwinger functional integral in the Euclidean domain

61 fl = Je 4 (f)_f)d4xdp,() (I1.2.60)
 II.3 Physical Interpretation in Terms of Particles 75

where dµ(0) is the Gaussian measure over classical field distributions corre-
sponding to free field theory. For a precise discussion and applications see
[ Glimm and Jaffe 19871.
There remains the question of whether and how one can return from the
Schwinger functions to the physically interesting functions in Minkowski space.
The conditions on the hierarchy {Sn} which guarantee that their analytic con-
tinuation leads to distributions in Minkowski space satisfying the Wightman
axioms have been found by Osterwalder and Schrader [Ost 73, 751. This impor-
tant work shows that besides the conditions (i) to (iii) the essential requirement
is reflection positivity. Let fn E S+ (1R4n ) and define an involution

(®fn)(xi, ... xn) = fn(ixn, ... 9x 1 ), (II.2.61)

where i x = (—x 4 , x) gives the Euclidean time reflection. Let the symbol
denote the tensor product in the tensor algebra of test functions

T= ® s(lR4n ), (II.2.62)
n

(fn ® gm)(xi , ...

x n , y i , ... Yin ) = fn( 2 i, • • xn)gm (yl, ... y m ). (II.2.63)
Then the reflection positivity demands

E sn+m(®fn ® fm)
n,m
? 0, (II.2.64)

for all elements f E T with fn E s+ (1R 4n ) .

This is the counterpart of the positivity condition (II.2.14) in the Wight-

man hierarchy. To check it directly would again be a formidable task but the
Feynman-Schwinger functional integral reduces this problem to a simpler one
which can be checked in concrete models for GI . Thus one may regard (II.2.60) as
a representation theorem guaranteeing that the hierarchy {Sn} has the needed
properties if Li is suitably chosen.
The Osterwalder-Schrader theorem states that properties (i), (ii), (iii) to-
gether with (II.2.64) and some limitation of the growth of S'n with n are equiv-
alent to the Wightman axioms. Therefore these conditions have been called
the axioms of Euclidean quantum field theory. Moreover, given the Schwinger
functions, one has a reconstruction theorem which gives the physical Hilbert
space fl with the action of the physical space-time translations on it. To ob-
tain the action of the local field operators in 71 one needs, however, an analytic
continuation.

II.3 Physical Interpretation in Terms of Particles

II.3.1 The Particle Picture

Suppose we have a theory satisfying the general postulates of section 1 and

suppose even that we have "solved it", for instance by having computed the
76 IL General Quantum Field Theory

w-functions or T-functions. How does it relate to physical phenomena? Exper-

iments in high energy physics are described in terms of collision processes of
particles. Therefore we must know which state vectors in 7-( correspond to spe-
cific configurations of particles before and after collision. In the present section
we give a heuristic discussion of this question; it will be followed by a more
thorough and precise analysis in section 4. In both sections the complications
caused by the presence of zero mass particles and long range forces will be
ignored. These will be treated in Chapter VI.
The first task is to characterize the single particle space, the subspace
(1)
'H C 7-t containing all state vectors which describe one particle, alone in
the world. The answer is suggested by the analysis of Wigner, discussed in
1.3, together with axiom A. In the reduction of the representation U(a, a) of
93 some irreducible representations U,,,;, 9i should occur with a discrete weight.
More simply, the spectrum of the operator M 2 = P1ZP/, should have a discrete
part. WO ) is the subspace belonging to the discrete part of the spectrum. The
masses of particles appearing in the theory are the discrete eigenvalues of M.
Their spins follow from a complete reduction of the representation U(a, a) in
the subspace 7- ( ' ) . The postulates of section 1 do not guarantee that there is
any discrete part of the spectrum of the mass operator. For the moment we
shall just add the assumption that 1t( 1) is not empty and postpone a deeper
discussion of the question "when does a field theory describe particles" to later
chapters.
The continuous part of the mass spectrum corresponds to the states of more
than one particle. The disentangling of this part in physical terms is no longer
a purely group theoretical reduction problem. We take as a guide the general
quantum theory of collision processes as described by Brenig and Haag [Bren
591. Let us take a complete orthonormal basis W,,„ in 7 -t( ') . The index i shall
denote the species of particle, thus
E ®71 1) .
Ni 1) (II.3.1)

The symbol E® denotes the direct sum of Hilbert spaces. The index a labels
basis vectors in 7-t ' . Our description is in the Heisenberg picture. So W,c de-
scribes the state "sub specie aeternitatis"; we may assign to it, as in (1.3.29),
a wave function in space-time obeying the Klein-Gordon equation. This gives a
rough picture of where one can find the particle at any given time.' Since W,,,„
is a normalized state the wave function is localized essentially in some finite re-
gion of space at time t = O. For large positive or negative times it will move to
infinity and spread out. We expect now, on physical grounds, that an arbitrary
1 That
this wave function cannot be precisely interpreted as a probability amplitude for
finding the particle at time s° at the position x has already been pointed out in section I.3.
We might use the Newton-Wigner wave function but this is related to the wave function
(I.3.29) by convolution with an integral kernel K(x — x') which is practically zero as soon as
Ix — x I exceeds a few times the Compton wave length of the particle. If the particle is not
of zero mass this convolution does not change the qualitative picture when we deal with the
large distance behaviour. So, for our purpose, we may ignore this problem.
 11.3 Physical Interpretation in Terms of Particles 77

state, if viewed at a sufficiently early or late time, can be described in terms of

configurations of far separated particles. To express this mathematically let us
first introduce some qualitative concepts which will be made precise at a later
stage. We say that a state is localized at time t in a certain space volume V
if it is orthogonal to the vacuum but "looks like the vacuum" for observations
at this time in the spatial complement of V. Example: a single particle state
whose wave function at the time in question has its support (essentially) in V.
Suppose the state vectors W 1 , W2 describe states which at some particular time
t are localized in separated space regions V1, V2. Then it should be meaningful
to define a "product state vector"

=W1 ®t ^2 (1I.3.2)

which has two localization centers at time t, corresponding to W1 and W2 respec-

tively. Note that, although in general no tensor product between vectors of 1.1
with values in 71 is defined, such a product becomes meaningful between states
which are localized far apart at a particular time. The product refers to and
depends on this time. We use it to define asymptotic products of single particle
states, where asymptotic refers to times t —> ±oo. Let us write a single index A
instead of the double index i, a to label the basis in H(l) • We claim that

Cri ...a„ = lim Wal ®t Wa2 ®t • • •^/A n - l im Wa i ,...an (t) (I1.3.3)

and
^ut = li w ®t W = li (II.3.4)
^n • m t ag ®t An
t• 'Al, An (t)

are well defined (Heisenberg) state vectors. We shall write in Dirac notation for
(11.3.3) simply IA i , • • • AO'. A formal proof of this claim will be given in the
next section. For the moment we should be content with a qualitative argument.
If WA and WAS are single particle states whose momentum space wave functions
have disjoint supports, then at large positive or negative times the essential
supports of their position space wave functions will be far apart. Therefore the
product ® t will be defined between them for any large ItI If t 1 and t2 are large
and have the same sign then the change in the meaning of the product at the
two times is due to the interaction of the particles in the time interval (t1, t2].
It is the difference between composing the wave packets IA) and IA') to a two
particle state at t i letting it develop then under the full dynamics to time t 2 , and
,

the composition of these single particle wave packets (which develop without
knowledge of the presence of the other particle) to a two particle state directly
at t 2 . We expect that the interaction decreases to zero as the separation of the
localization centers increases to infinity. Therefore the sequence of state vectors
on the right hand side of (1I.3.3) or (I1.3.4) will converge strongly to a limit
vector as t —> ±oo. 2 If the supports of the momentum space wave functions are
not disjoint the argument can still be carried through due to the spreading of
2This means that the sequence of Hilbert space vectors W(t) forms a Cauchy sequence in
the norm topology of vectors, so IF W — W(t) 11—> O.
78 II. General Quantum Field Theory

the wave functions in position space. If one divides space into large cubes of
fixed edge length R then in a single particle state the probability w(t) of finding
the particle in a particular cube at time t decreases proportional to t-3 ; the
number N(t) of cubes in which one may find the particle increases proportional
to t 3 . The probability of finding two particles in the same cube or in adjoining
cubes decreases like N(t) • w(t) 2 ti t -3 .
The decrease of the interaction with increasing separation can be related
to the decrease of the correlation functions W T for large space-like distances
and this, in turn, can be derived from the general postulates if the energy-
momentum spectrum of the theory has a gap between the vacuum and the
lowest-lying single particle state (see section 4). All the above arguments refer,
of course, to particles with non-vanishing mass. For the collision theory in the
case of a theory without mass gap see Chapter VI. One shows now that the
states 1a 1 , • • • )1n ) have the same orthogonality relations as the multiparticle
states of a free theory. Specifically, the product WA ®t WA ' is commutative if a and
A' refer to two bosons or to one boson and one fermion, it is anticommutative
if both indices refer to fermions. Thus we get for instance

in (A1A21A1A2) in = (X11a1)(A21a2) ± (X11a2)(A21a1). (11.3.5)

Furthermore, these states transform under U(a, a) as if the theory were free i.e.

U(a, a)1a1, ... an )' = IA /1 , • • • A,,) 111 with IA;) = U(a, 0)P9). (II.3.6)

Thus these states span a Fock space as in (I.3.51), (I.3.52) with the tensor
product (g in = lim ®t and the action (I1.3.6) of the Poincaré group on it. The
equivalence class of the representation of s3 in this Fock space is, as in the case
of free fields given by

i
(II.3.7)
UF,z (u j ) sA
® .

The interaction can not be seen from the representation class of U(a, a).
The following picture of a 2-particle scattering process in classical mechan-
ics is very instructive to illustrate the existence of two basis systems in 7-f in
which the representation U(a, a) looks like the one for fr ee field theory. The
full lines describe the actual trajectories of the two particles in space (we have
indicated times t i by markings on the trajectories). Together the two full lines
correspond to a Heisenberg state of the 2-particle system; they give the com-
plete history. Each trajectory has two asymptotes, drawn as dotted lines. They
represent straight line motions with constant velocities and refer to the asymp-
totic times t —> ±oo. Each asymptote corresponds to a Heisenberg state of a
single particle alone in the world. The Heisenberg state of the 2-particle sys-
tem can be characterized by giving its asymptotes at t —> —oo (i.e. in classical
mechanics by giving the asymptotic initial velocities and impact parameters).
 I1.3 Physical Interpretation in Terms of Particles 79

t2

-oc

+ 00

Fig. II.3.1.

Equally well, the state is characterized by giving the asymptotes at t -* +oo.

A Poincaré transformation shifts the trajectories in space-time but each new
asymptote is just the shifted old asymptote, unaffected by the presence of a
second particle.
From this picture we can also read off the time dependence of the product
composition symbol ® t

el. (W1 ®t W2) = ( e HT w )

® t+r (eiHrw) (II.3.8)

or
^1
® t+r T, = eiHr (e_HTWi) ®t (e_iHrw2) . (II.3.9)
2

The basic idea of this approach to collision theory is due to Ekstein,

[Ek 56a and b]. It was expanded by Brenig and Haag [Bren 59]. Let us compare
it with the more familiar description of single channel scattering theory in quan-
tum mechanics. There the motion of particles without interaction is described
by a "free Hamiltonian" H0 . Fixing t once and for all, say as t = 0, and choosing
the wi as single particle states we may write (II.3.9) as 3

w2 = eiHre—iHor (w1 ®O w2) (11.3.10)

w1 ®r

since the inner bracket in (II.3.9) describes the motion of each particle in the
absence of the other. Thus
3 This requires the definition of the product also for states which are not localized far apart
at the time t = 0, but this is no problem; it means only that a certain freedom remains in
the choice of the definition of this product. In wave mechanics one takes the tensor product
of the wave functions at the time t = 0.
80 II. General Quantum Field Theory

W1®in W2 = llm ®T W2 =52 (W1® tl W2) (11.3.11)

T -+--00
-

where fr is the Moller operator, the limit of the operator eixte-iH°t for
t — —oo. The advantage of (1I.3.3) over (II.3.11) is, that we do not have
to refer to a splitting of H into a free part and an interaction term. This is
already useful in nonrelativistic quantum mechanics if one has to deal with
"bound states" ("composite particles") because then one has to take a different
splitting of H for each channel. This cannot be done without abandoning the
permutation symmetry, so one has to be careful to take the Pauli principle cor-
rectly into account if one uses the splitting method. In the field theoretic case,
even in situations without "composite particles", the separation of "persistent
interactions" from interactions which vanish in states where particles are far
apart is not directly visible in the form of the Hamiltonian. This leads to the
need for renormalization of parameters like mass and charge in (II.2.32) (see
the remarks at the end of subsection II.2.4) .
To sum up: given the existence of single particle states and the definition of
the product ® t we can construct in 1 -1 the state vectors of specific configurations
of incoming or of outgoing particles. The above qualitative discussion indicates
that this construction needs only the postulates of section 1 and the assumption
of a mass gap in the Po-spectrum. This will be corroborated in the next section.
We shall assume in addition that all states can be interpreted in terms of
incoming particle configurations, in other words, that the IA 1i • • .) i n span the
whole of 7-i, that they provide a complete system of basis vectors. The same
should be true, of course, for the vectors I),, • ..)out too. This assumption, called
asymptotic completeness is not a consequence of the postulates of section 1. In
chapter VI we shall discuss natural postulates needed for this as well as for a
deeper understanding of the notion of particles and collision processes in the
case where the continuum of the energy spectrum begins at zero.

11.3.2 Asymptotic Fields. S-Matrix

We assume asymptotic completeness. Then 7-i is isomorphic to the Fock space

generated by means of the tensor product ® in from 7- 01} . We may therefore
introduce annihilation and creation operators a, am as in section 3 of chapter I.
The representation of q3 respects this tensor structure i.e.

U(a, a)(W1 ®in W2) = U(a, a)W1 ®in U(a, a)W2, (II.3.12)
and in 1 4 1) it is given by Un„,s, . We may therefore combine the creation and
-

annihilation operators of incoming particles for each species of particle to a

covariant free field 0!n(x) as described in the subsection I.5.3. For spin zero we
have a scalar field, for spin 1/2 a Majorana or Dirac field (depending on the
existence of an antiparticle). For higher spin there is some arbitrariness in the
choice as can be seen from the discussion in I.3. and 1.5.
The same construction can, of course, be made with respect to the product
®out, leading to fields O aut(x)
 IL3 Physical Interpretation in Terms of Particles 81

Since the bases I.\i, • • • An) in and I ) i, • • • ))ou t have the same orthogonality
relations they are related by a unitary operator S:

I a1, • • • )tn/in = SI A1, • • • An)out' (II.3.13)

or, in terms of the incoming and outgoing fields,

Oan(x) = SO°ut(x)S-1 (II.3.14)

The operator S commutes with the Poincaré operators because 45in and Oout
transform in the same way,

U(a, a)SU(a, a) -1 = S. (II.3.15)

The quantities of experimental interest are the matrix elements of S in the

basis of outgoing particle configurations. They give the probability amplitude
for finding a specific outgoing configuration when the incoming configuration is
known. Due to the unitarity of S these matrix elements are also equal to the
matrix elements in the basis of incoming configurations
i p1/41, ...An ) i n
out 01/4 '1, Am %1m 1, . . . A n \out
, IsI^
...

/
`
(II.3.16)
in (A'' , • . • )^m ISIai, • • • An) in

Therefore we shall omit the labels in or out in writing the S-matrix elements. In
almost all applications one will have only two particles in the initial state. One
uses the improper basis of momentum eigenstates in 1/( 1) , so the translation
covariance of S is simply expressed by writing

(p1, .. .pen is — 11 Ipi,P2) = 6(4)(p1 +p2 — Epi)(1, .. .pm lTlpi,p2). (II.3.17)

We have suppressed the spin indices. The "T-matrix-element" is a Lorentz in-

variant function of the momenta and the spin indices. Its absolute square mul-
tiplied with a kinematical factor, depending on the relative velocities of the
incoming particles and the choice of the independent variables describing the
final state, gives the differential cross section for the process
,
P1, p2 --^ p 1 , . • • pm.

11.3.3 LSZ-Formalism

Returning now to the question raised at the beginning of this section one answer
suggests itself in analogy to Fig. II.3.1.. The basic local fields 45 (x) in terms of
which the theory is formulated and to which the postulates of section 1 are
supposed to apply, become asymptotically close to incoming fields when their
time argument tends to —oo and to outgoing fields as s0 ---> +oo. In other
words, the fields serve to interpolate between lin and 0"t and this establishes
82 IL General Quantum Field Theory

the connection with the physical interpretation in terms of particles and collision
cross sections. Of course this interpretation can be fully adequate only if we have
a one to one correspondence between basic fields and species of stable particles.
In the general case we need an interpretation which does not rely on field-
particle dualism. This will be given in section 4. The purpose of the remainder
of this section is to give a precise meaning to the intuitive assumptions

q
5in( x ) r-
t --■ - co
(x)
+^ oout ( x )
(1I.3.18)

in the case of a one to one correspondence of fields and particles and to derive
from it an algorithm relating S-matrix elements to 'r-functions.
In order not to burden the discussion with too many indices we shall consider
in the following the simplest case, where we have to deal with only one species
of particle which has zero spin and nonvanishing mass and where we have one
basic (scalar, Hermitean) field 0 in the theory. The generalization to particles
with spin 1/2 and Dirac fields is obvious. For more general cases we refer to
section 4.
Let cp denote a single particle state, cp(x) its wave function which is a posi-
tive energy solution of the Klein-Gordon equation. The corresponding creation
operator for incoming configurations is then given by (see (1.5.9))

a * i n((p) = 2 in(x) acc( X) ^

f o_t {

a p in (x) cP(x) d3x, (I1.3.19)

where the time t can be arbitrarily chosen. Consider the analogous expression

00,o; t) = i f o_t {(x) . (x) — ( —

ax0 0(x)) cp(x) d 3x. (II.3.20)

Since the actual field does not obey the Klein-Gordon equation if there is any
interaction (though cp does) the operator (II.3.20) will depend on the time of
integration. 4 In fact one has

8( t)
t (Ko)cpd3x (11.3.21)
at — —Z f 0_
where we have used the abbreviation K for the Klein-Gordon operator

K = ❑ + m2 . (I1.3.22)

Lehmann, Symanzik, Zimmermann [Leh 55] proposed as the precise form

of the asymptotic condition, replacing (II.3.18)

(W11a *in (t0A) t oo (W1l 0 (40 ; t)lW2) +oo (WI la*out((p)1W2) (II.3.23)
4 We mentioned that is not an operator valued distribution at a sharp time but needs
smearing out with a smooth function also in time. Therefore the expression (II.3.20) is not
an operator but only a sesquiliner form. But this could easily be remedied by an additional
averaging over a small time interval and moreover we shall only use matrix elements of 45((p; t)
in the following. So we can forget this complication.
 II.3 Physical Interpretation in Terms of Particles 83

for all state vectors W 1 , W2 belonging to a dense domain which contains the
vectors generated from the vacuum by repeated application of the operators
a*in and a*°ut with smooth wave functions. We may also express this by saying
that 0(cp; t) converges weakly on the mentioned domain to a* in(cp), respectively
a*out((p) as — Too.
t
In (IL3.23) the actual field is normalized so that it has the same matrix
elements between the vacuum and the single particle states as the incoming
field. From the general postulates of section 1 we have, with this normalzation

w2 ( x 1, x2) = id+(x 1 — x2; m) + i f d+(x1 — x2; K)dp(K) (11.3.24)

where dp is a positive measure with support in the interval [2m, oo]. (see
Lehmann [Leh 54]). The vacuum expectation value of the equal time commu-
tator of 2r(x) =4900(x) with 45 is

[ir(x1), 0(x2)] _ — ib(3) (xi — X2)( 1 + f dp(i))

which differs from the canonical commutation relation by the factor in the
bracket, which is the (divergent) factor Z -1 mentioned in section 2.4.
From the asymptotic relations (II.3.23) one deduces the very useful reduc-
tion formulas
out ((p1, . . .
(Pik I T4) (X1) • • • 0 (Xm) I (pi , • • • M in

=i f IK y out (v)1, . . . (Pk ITO(X1) • .

• 45 (x7,) 0(y) IcP1, ... (p1-1) in } (p[(y)d4y
(II.3.25)
f
= i {K0t1t(,... (10,k-iI T1(z1) • ^(zm) c(y)I 1
(p1, ... (P1) in } (Pk(y)d4 y
J (II.3.26)
Here Ky = ❑y + m 2 is the Klein-Gordon operator with respect to y and we have
assumed that an orthogonal system of single particle states is used and that the
incoming and outgoing configurations contain no common state.
The reduction formulas allow one in a very elegant manner to reduce the
number of particles in the incoming or outgoing configuration by adding one
more field under the T-product. By repeated application we thus convert an S-
matrix element to an expression in terms of T-functions. Using the momentum
space basis we get, if the final momenta are all different from the initial momenta

1I' 1, • • .P k IsIPI, • • . Pi )

= ( — i) k+d 11(Pj — m2 )(Pi — m2 009 1, • • .Pk, — pi, • • . — P[) (11.3.27)

where T is the Fourier transform of T. Looking at the Feynman rules for com-
puting the functions T (see section 2.4) we see that these functions have poles
at q2 = m2 coming from the propagators of external lines. The factors ( p2 — m2 )
in (11.3.27) cancel these. The S-matrix elements are the T-functions with am-
putated external lines, taken for momenta which lie all on the mass shell (the
hyperboloid p2 = m2).
84 II. General Quantum Field Theory

11.4 General Collision Theory

We now want to divorce the treatment of collision processes from the assumption
of a field-particle duality. Specifically we want to cover situations where particles
occur which are not related to one of the basic fields via an asymptotic condi-
tion (11.3.23) ("composite particles") or where basic fields occur which have no
counterparts among the species of observable particles (e.g. quark fields). The
physical interpretation of the quantum fields will not be primarily attached to
particles but to local operations. Specifically an operator O(f) (a basic field
smeared out with a test function f) represents a physical operation performed
on the system within the space-time region given by the support of f. 1 Naively
speaking, the argument x of a basic field has direct physical significance. It
marks the point where 0(x) applied to a state produces a change. This sounds
obvious and in fact one may regard it as the reason for axiom E. However one
should remember that the argument x in 45''(x) does not have this direct physi-
cal meaning though Vn, being formally a free covariant field, also satisfies axiom
E. Also it is not obvious at this stage that such a purely space-time-geometric
interpretation of the basic fields will suffice to analyze the phenomenological
consequences of the theory since it introduces no distinction between the dif-
ferent types of fields occurring. The suggested interpretation suffices indeed to
construct the states Pa l , • • •)'n and [A i , • . •)out This will be demonstrated un-
der some simplifying assumptions in this section and discussed more fully in
Chapter VI.

II.4.1 Polynomial Algebras of Fields. Almost Local Operators

An operator

Q=E J f(x i ,. . .)ç 1 ).... 0(xn) H d4x i (II.4.1)

will be called local (with localization region 0) if all f n have their support
in 0 Xn (i.e. vanish whenever one of the points x i lies outside of 0). The set
of operators with localization region 0 generates an algebra, the polynomial
algebra A(0) of the region O. Our interpretation of the theory will be based
on the assertion that the elements of A(0) correspond to physical operations
performed in O [Haag 57].
The operator (II.4.1) will be called almost local if all f" belong to S 4' (the
relevant property in this context being that they decrease faster than any power
if any point xi moves to infinity). We denote by Q(x) the translated operator
Q(x) = U(x)QU -1 (x). (II.4.2)
Of course if Q is almost local so is Q(x) but if we consider very large translations
then Q(x) may be regarded as representing an operation centered around the
1 0f course
0 and f are to be understood generically in the sense of (II.1.7) We do not
explicitly write the indices distinguishing the types and components of the fields.
 II.4 General Collision Theory 85

point x. The extension of the support of f' in (11.4.1) becomes then irrelevant.
In particular one has

Lemma 4.1.1
Equal time commutators 2 of almost local operators tend fast to zero with in-
creasing separation. In application on the vacuum we have

I [Q1(t, xl), Q2(t, x2)JIQ) II < cNlxi — x21 -N , (11.4.3)

for any positive N.

To justify (I1.4.3) note that if the Q are localized in some finite region then
axiom E implies that the commutator is zero as soon as 'x i — x2 1 exceeds a cer-
tain finite distance. Almost local operators can be approximated by local ones
with support in a 4-dimensional cube of edge length d so that the mistake on the
left hand side of (11.4.3) decreases faster than any power of d as d — oo. For this
latter claim it suffices to assume that the Wightman distributions are tempered.

The assumptions we want to add now to the axioms of section 1 are

a) The mass operator M = (Pti P4 ) 1 '2 has some discrete eigenvalues. In
other words, the single particle subspace 'H(l) is not empty.
(3) In the subspace orthogonal to the vacuum the spectrum of M has a lower
bound mo > O.
-y) For any single particle state W with smooth momentum space wave
function3 cp(p) there exists an almost local operator Q generating it from the
vacuum
W = Qr2. (II.4.4)
It should be understood that none of the assumptions a) — y) nor all of
the axioms of section 1 are of fundamental significance. 4
In the following let Q denote any almost local operator. The basic tools in
the subsequent analysis are

Lemma 4.1.2
The improper operator
Q(p) = f Q(x)e -vxd 4x (11.4.5)
changes the momentum of a state precisely by p. Equivalently, with
2 respectively anticommutators in the case of fermionic operators.
3 We suppressed the indices indicating the type of particle and the spin component. co(p)
may be considered as a function of the 3-momentum since the energy is then determined
by the mass. Smooth shall mean that cp is infinitely often differentiable with respect to the
components of p. This implies that the state is, roughly speaking, localized at finite times in
a finite part of space. More precisely, that the position space wave function at finite times
decreases faster than any power as x goes to infinity. Actually y) is not really needed (see
[Ara 67]). We assumed it here only for convenience.
4 We ignore the infrared problems which invalidate assumptions (3) and y), we ignore as-
pects of local gauge theories where some of the axioms have to be modified.
86 II. General Quantum Field Theory

Q(f) = f Q(x)f (x)d 4x

we have
(W21Q(f)IWi) = O if (E1 + A) n E2 is empty (1I.4.6)
where the momentum space regions E1 are the spectral supports of Wi (in the
spectral decomposition with respect to Pu) and d is the support of

i(P) = f Px f (x)d4x.

The proof of this lemma follows exactly the line of argument leading from
(II.2.4) to (1I.2.8).

We consider now the vacuum expectation value of products of shifted al-

most local operators, expressions of the form (S2JQ 1 (xi ) • • • Qn (xn) I,fl). The cor-
responding truncated expectation values will be denoted by (• • •) T . They are
defined as in section 2, taking the Qi instead of the basic fields, ignoring the
internal structure of each Qi . Then one has

Lemma 4.1.3 [Ruelle 62]

Let xk = (t, xk ) be a configuration of points at equal time and define R =
maxjxi xkl. Then, for any fixed set of almost local operators {Qk} and for
—

i,k
any positive N there is a constant AN such that

(Q1(x1) . • . Qm(xm))T < ANR -N . (II.4.7)

This gives a uniform bound for the truncated functions in terms of the
radius R of the configuration.
The proof given by Ruelle for this crucial lemma is rather complicated. We
shall describe here the essential ideas used in it and refer the reader for a wa-
tertight proof to Ruelle's original paper.

First observation: Due to the assumption of a mass gap (assumption (3)) we

may, with the help of lemma 4.1.2, decompose an almost local operator into
three parts
Q = Q - + Q+ + Q° (11.4.8)
where
Q- fl = 0, Q+* fl = 0, Q° ,(2 = (flIQlS2),(2, (11.4.9)
and each part Qi (i = +, —, 0) is still almost local. To see this let us take for Q
the n -th term in (II.4.1) and define

Q(p) — (2/0-4 j nx1 — y, . . . xm — y)d)(x1 ) . . . 0(xn )e - :md4y IT d4xi. (II.4.10)

 II.4 General Collision Theory 87

Then

Q = f Q(p)d4p = f(F+ (p) + F_(p) + Fo(p)} Q(p)dip (11.4.1 1)

where we have made a smooth partition of unity

F+ +F_ +F ° = 1,
F+(p) = 0 for p° < a,
F_(p) = 0 for p° > —a, (I1.4.12)
Fo(p) = 0 for Ip° 1 > b,
0< a<b <mo.

The three parts of Q arising from this partition satisfy (II.4.9) according
to lemma 4.1.2 because of the spectral assumption f3). Furthermore each Qi
(i = +, —, 0) is still almost local: the functions

fi(x1, • • • xn) = f f (x1 y, ... xn im„Fi(p)d4yd4p

y)e
-

— —

decrease faster than any power for large arguments because the Fi (p) are in-
finitely often differentiable. Thus the decomposition (1I.4.8) leads to almost
local operators of which the first and the adjoint of the second annihilate the
vacuum whilst the third reproduces the vacuum up to a factor.

Second observation: For the case of two factors i.e. for n = 2 in (II.4.7) lemma
4.1.3 follows directly from the above observations. We have

(QIQ1(xi)Q2(x2)1(2)

= (QI (Qj(xi) + Q1(xi)) (Q(x2) + Q2(x2)) In).

For large lxi — x 2 1 we can commute Q2 to the left and QT to the right where
they give zero. The remaining term is

(QIQ ? (xi ) Q2( x 2 )IQ) = (QIQ1( x i)IQ)(Q I Q2(x 2) IQ).

The neglected terms from the commutators decrease faster than any power with
lxi — x21 due to lemma 4.1.1.
For n > 2 one observes that for a configuration of "diameter" R as defined
in lemma 4.1.3. there is a subdivision of the points into two subsets so that the
spatial distance of any point in one subset from any point in the other subset
is at least R/n. One may then apply the technique described above recursively.

II.4.2 Construction of Asymptotic Particle States

We choose now in HO ) for each species of particle (index i) and each spin
orientation (index a) a reference state vector ^ ia . Its momentum space wave
88 II. General Quantum Field Theory

function îp4,« (p) shall be smooth and nowhere vanishing on the mass hyper-
boloid. According to assumption 'y) we can pick an almost local operator qi,«
so that
q4,«Q = (27r) -3/2Wî,« . (I1.4.13)
Let
hi (x) = ( 27) -3/2 f hi (P)e-7Pxdlti(P) (11.4.14)
be a positive energy solution of the Klein-Gordon equation to mass m 4 and
define

qi «(hii t)
'
= i Jfl-t { i a (1) / hi(1) - ( qt,Q (x) h4(x) d3 x.
) (I1.4.15)

Then g4,4 (hi; t)Q is a single particle state (of the same species and spin orien-
tation) with momentum space wave function
42(p) -= tPi,« (P)hi (P). (I1.4.16)

This state is independent of the choice of t.

We shall now drop the indices i, a and write for instance q 1 instead of q41,«î .
Under the assumptions a) - -y) (and the standard axioms) one arrives at the
following conclusion which we state as a theorem. Part a) gives the construc-
tion of states corresponding to specific configurations of incoming or outgoing
particles. Part b) gives the orthogonality relations, part c) the transformation
properties of these states.

Theorem 4.2.1 [Haag 58]

Let qk (hk; t) be defined as in (11.4.15) with hk (p) smooth. Then
a) The sequence of state vectors
W(t) = ql (h1 ; t).. . gn (hn ; t),(l (I1.4.17)

converges strongly for t — ±oo. The limit states have the physical interpretation

Win = lim ^ t = in • • . ®in Wn ` IV17 • • • Vn)in

( ) (11.4.18)
r ut =_ lim W(t) =
WI (gout ... ®ou
t^n = ioi ,
•.. (Pn
}out
(I1.4.19)
t,+00
t-+
(see section 3). Here the Wk are single particle states whose type and spin orien-
tation follows from the choice of qk and whose momentum space wave function
is
Sok(p) = cpk (p)hk (p)• (I1.4.20)
b) The scalar product of vectors Win is

inOp1l . . . 49nl(P17 . . . Vn)in = E( -1 ) s ^(^pPkl (pk) (I1.4.21)

l P k
 II.4 General Collision Theory 89

where P runs through all permutations of the indices (1, • • • n) and s is even
or odd depending on the signature of the permutation of the Fermi factors.
The scalar product between states with different numbers of incoming particles
vanishes.
The same relation holds if the symbol in is replaced by out.
c) The Poincaré operators U(a, a) respect the product structure introduced
by ®sn and ®out i.e.

U(a, a){W1 ®'sn W2 • • .} = U(a, a)Wi ®`n U(a, a)W2 • • - (II.4.22)

with the same relations if ®'n is replaced by ®out

Proof. The proof follows the line of argument sketched in section 3 with W1® t W2
1 (hi i t)q2(h2; t)!?. The decrease of interaction at large distances is realizdbyg
replaced by the decrease of the truncated vacuum expectation values of products
of almost local operators which is assured by lemma 4.1.3.
We first need the asymptotic form of single particle wave functions at large
times. Let

.f (tf x) = f f (p) eXP i(px — €Pt) ^ ^

P = (p2 + m2 )
1/2

^ (II.4.23)
pP ^

with f a smooth function, decreasing fast for jpj — oo. The asymptotic form
for large jti can be obtained by applying the method of stationary phase to the
integral (II.4.23). Since the phase px ept is rapidly varying as a function of p
—

when t (and possibly also x) is large, the dominant contribution to the integral
comes from the point

p = mv(1 -- v2)-1/2; v = x/t (II.4.24)

where the phase is stationary. One finds the asymptotic expansion

f (t, x) = const. t -312 exp(—im'y -I t) (y3h'2j(myv) + O(t-1 )) . (II.4.25)

r y = (1 — v2 ) -1/2 ; v = x/t.

Whilst (II.4.25) describes the general asymptotic behavior of smooth wave

functions a much stronger statement can be made for those directions in veloc-
ity space which lie outside the support of f .

Lemma 4.2.2 [Ruelle 62]

Let f (x) be as in (II.4.23), Z its support in velocity space i.e.
E7 = {v = p/p° : p E supp f } and Ll any open set containing Z. Then

(i) for y 0 14 one has I f (t, vt)j < CN(1 + Ivj)_ N tj -N (11.4.26)
90 II. General Quantum Field Theory

for arbitrary positive N.

I f (t, vt) I < Cl t I

—3/2
(ii) for y E 14 one has (II.4.27)

Next we consider scalar products between vectors obtained by applying

several operators of the form

Qf (t) = f o _t Q(x)f (x)d 3 x (1I.4.28)

to the vacuum. Q shall be almost local with (01 QI (l ) = 0, f a smooth single

particle wave function and all times are taken equal.

Lemma 4.2.3
Under the conditions stated immediately above one has

(QIQ f,(t)* • • •Q f;,(t)*Q fn(t) ... Q fi(t)ÎQ)

bmn i (-1)3
pairings
H
pairs
( Q I (2fk (t)*Qf, IQ). (1I.4.29)

The difference between the right and the left hand sides of (11.4.29) decreases
in general like I t I ` 3/ 2 If the supports of the functions fk in velocity space as
well as those of the functions fk are disjoint the decrease of the difference is
faster than any inverse power of ItI.

In (II.4.29) the symbol Q is meant generically. The almost local operators

may be different in each factor. The proof of this lemma is a rather straightfor-
ward application of lemmas 4.1.3 and 4.2.2 or equation (11.4.25) . We make a
cluster decomposition of (QIQ*(y l ) • • • Q *(ym )Q(x i ) • • • Q(x n )1,fQ) and insert the
estimate (I1.4.25) for the wave functions. Numbering the clusters by an index
r (r = 1, ... N) we denote the first point in the cluster r by X, use for the
other points in this cluster the relative coordinates ti = xi -- X,.. The truncated
expectation values are, by lemma 4.1.3, fast decreasing functions of the e. They
are independent of the X and of t. Therefore the integration over the e can be
performed. It gives for cluster r with n r factors Q*, m,. factors Q a contribution
of the form
3
CI t I- /2(nr+mr) (II.4.30)
Changing the remaining integration variables from X,. to V,. = Xr It we
obtain finally as the contribution to (11.4.29) from a clustering with N clusters
Cr' It I3N- 3 /2(ra +m) (II.4.31)

Note that the V,. are restricted to the balls IV,.I < 1, the boundary corre-
I.
sponding to the momenta p ---* oo in Therefore the integrations over V,. only
produce a finite factor independent of t.
 II.4 General Collision Theory 91

Since ([Link]) = 0 each cluster must contain at least two points. For all
clusterings which contain a cluster with more than 2 points N < 1/2(n+m) and
therefore (11.4.31) decreases at least as ItI-3t2. Thus we can get a nonvanishing
contribution for Iti —> oo only in the case of a complete pairing. Furthermore
we get a nonvanishing contribution only if each cluster pairs a factor Q* with a
factor Q. A cluster with two factors Q contributes an expression

f w(p)fi(p)fk( — p)e
-i(el+E2)t d3
p

where iv is the 3-dimensional Fourier transform of

w(xx — x2 ) _ (QI Q1(t, x1)Q2(t, x2iQ),

and therefore a smooth function of p due to the fast vanishing of w for large
Ix^ — x2 I. The momentum space wave functions in the integral are smooth by
assumption and the exponential factor is a fast oscillating function of p when t is
large. Therefore the integral decreases faster than any power of VI for Iti — oo.
A complete pairing between factors Q* and Q is possible only for n = m
and yields then the right hand side of (1I.4.29). We may note that for n = m
the difference between the left and the right hand side of (11.4.29) decreases at
least as ItI -3 , coming from a clustering where at least two clusters contain three
factors. In the case of disjoint supports of the momentum space wave functions
any clustering which is not contained on the right hand side of (II.4.29) gives
a contribution vanishing faster than any power due to part (i) of lemma 4.2.2.

Corollary 4.2.4
Let Q f(t) be as in lemma 4.2.3, k = 1, • • • n and P a permutation sending k to
Pk. Consider the vectors

W = fl Qfk(t)Q and W' ( -1 ) 8 fl QfPk(t)Q

where s is the number of transpositions of Fermi factors arising from the per-
mutation. Then as Iti — o0

II qr( — tir(' II< cltl — " 2 (II.4.32)

To see this note that

II w — w 11 2 = (WiW) — (wIw') — (w ' Iw) +

'

Each of the terms is of the form (II.4.29) with m = n and they all give the same
dominant contribution (the right hand side of (II.4.29)) because the pairings
are independent of the order of factors. Therefore the dominant contributions
cancel and the remainder decreases like ItI -3

After these preliminaries the proof of the theorem is quick.

We first show that II aW(t)/at iI < cltf --312 for large Ito. This implies the strong
convergence of W(t) with
92 IL General Quantum Field Theory

in -°°
i^ W" W(t) II < 2 c ptI -1 /2 for Iti -+ oo; ex = for t -+
l +oo out
(II.4.33)
Why does 1' 8W(t)/at I! decrease for large Its? Differentiating W(t) we get a
sum of terms of the same form as (I1.4.17) with one of the q(h; t) replaced by
a/at q(h; t). We commute aq/ât to the right till it stands directly in front of Q
where it gives zero because q(h; t) Q, as a single particle state, is independent of
t. According to the above corollary this change of the order of factors introduces
only a difference of order itL -312 in the norm.
After the convergence of W(t) is established part b) of the theorem follows
directly from (II.4.29).

Coming to part c) of the theorem we remark that

U(a; a)qi ... %i Sl = ql ... qnQ, (I1.4.34)

gk = U(a; a)gkU(a; a) -1 . (II.4.35)

Now
q' Q = Wk = U(a; a)Wk (II.4.36)
is the transformed single particle state. For translations and rotations qk is
again of the form (II.4.15) and the claim c) follows directly from part a) of
the theorem. In the case of proper Lorentz transformations the verification of
(II.4.22) needs a little more care since (II.4.35) will then involve almost local
operators at different times. But the relevant configurations of points will still
have large space-like separation and lemma 4.1.3 is valid in any Lorentz frame.
This concludes the proof of the theorem apart from the claim that the limit
states (11.4.18), (I1.4.19) must be interpreted as states with specified configu-
rations of incoming or outgoing particles. Although this is rather evident from
the qualitative discussion in section 3 we shall illustrate it by simulating the
experimental procedure of particle detection more directly in the mathematical
formalism.

II.4.3 Coincidence Arr an gements of Detectors

Let us simulate a detector, centered near the point s = 0, by a positive operator

C with CSI = 0, localized (essentially) in some finite region around the origin. 5
i,•• Acoindeargmtfsuchdeor,natpisx
which lie far space-like separated from each other will then be represented by the
product C(x i ) • • • C(x„L ). We take the times equal and put x i = (t, vit), v i v i .
Wearintsdhxpecaonvluftisrdchea(I1.47)
'Due to the Reeh-Schlieder theorem (see next section) C cannot be strictly localized in a
finite region if we demand CSI = O. We must either take it as almost local or allow a small
response rate in the vacuum.
 II.4 General Collision Theory 93

For simplicity we choose states which are composed of single particle states
whose momentum space wave functions have disjoint support. We thus choose
the hi in (II.4.17) so that

supphi n supphj =O for i# j. (I1.4.37)

This is no serious restriction for most purposes. It just excludes thresholds

where two or more particles with the same momentum occur in a configuration.
It brings the advantage that the convergence of W(t) to wex is much faster than
in (1I.4.32), namely
I Wex II< CNIti -N (II.4.38)
for arbitrary positive N. Applying the familiar method we find that the con-
tribution of a cluster decreases faster than any power of (ti in any one of the
following circumstances:

(i) the cluster contains more than one factor C (because v i # vi ),

(ii) it contains more than one factor q or more than one factor q* or q and q*
in one cluster are not associated with the same wave function (because of
(II.4.37),

(iii) C is not accompanied by a factor q on the right and q* on the left. (Since
= 0 and C = C* we get zero if C stands immediately adjoining the Cf2
vacuum) .

Thus one gets for any positive N

t^ N (1 (t)I C(x 1 ) • • • C(x, n )IW(t)) —> 0 if m > n (II.4.39)

and unless each vi is in the velocity support of one of the hk.

If vi E E(N) for i = 1, • • m and m < n we get

m n
(W(t)IC(x i ) ... C(x,01W (t)) —> 11(W IC(xi )IWi )fi (WOO. (II.4.40)
1 ,n+1

A factor (Wi lC(x)1Wi ) gives the response of a detector centered near the point
x in the single particle state A. Of course each such factor decreases like t--3
because the spreading of the wave function makes the probability of meeting
the particle in some space region of fixed extension decrease. So (II.4.40) can
be used only for large finite t, not for infinite times; this corresponds to the
experimental situation of detector arrangements.

Remark: The preceding argument corroborates the physical interpretation of

the states W'. At the same time (I1.4.39), (11.4.40) give the most natural defi-
nition of what is meant by particles. A state with n incoming particles is defined
as a state which, at all times earlier than some time t, gives no signal in any
94 II. General Quantum Field Theory

coincidence arrangement of more than n far separated detectors (mathemati-

cally expressed in (1I.4.39)), but does give a signal in some n-fold coincidence
arrangements (Equ. (1I.4.40)). This criterion is applicable even if one has to
abandon the association of discrete eigenvalues of the mass operator with single
particle states as one is forced to do in a serious treatment of infrared prob-
lems caused by photons because a charged particle alone in the world cannot be
sharply distinguished from one where it is accompanied by some soft photons.

II.4.4 Generalized LSZ-Formalism

To close this section we want to establish the bridge to the LSZ-formalism and
its generalizations. We compute matrix elements between states W(t), W' (t) of
the form (I1.4.17) for an almost local operator A(x) centered around the point
= (t, vt) where, unlike the case of detectors, neither A nor A* annihilates the
vacuum but only the vacuum expectation value of A is required to vanish and,
unlike (I1.4.13) A,f2 is not a single particle state. Let W(t) contain n factors,
(t) m factors and assume that in each of these states the wave functions h i
have disjoint velocity support. With the technique, boringly familiar by now,
we find that the only terms which do not decrease faster than any power for
large iti come from clusterings where A is accompanied by one factor q or q* or
by one q* and one q. This leads to

>(Q A(x) ko i) (Wi (çoO W n)

(W ' (t) I A(x) I W (t))
j )W'iniWin)
+ E(vj iA(x)IQ)(ain(ç
^', in in (vi)win )
(II.4.41)
+ [r( vjIA(x)I(pi)(ain(v)^ ia ,
i, j

where
^sn W 'n in
= Rol, • • • Son )in; = I49 X, . . . 49m)
The first two terms on the right hand side decrease like 1t1 -312 ; they are re-
spectively positive energy and negative energy solutions of the Klein-Gordon
equation to the mass associated with the wave function (pi resp. cpj . The last
term decreases like 1t1 -3 . To pick out matrix elements of ain* for a certain particle
type, say of mass m, we choose an A which has nonvanishing matrix elements
between the vacuum and single particle states of this type. If f is a positive en-
ergy solution of the Klein-Gordon equation to mass m and (compare (1I.3.20))

A(f ; t) = i f A(x) ax° f (x) — ( /_A(x) )

f (x) d 3 x, xo = t

then we get from (I1.4.41)

(W' (t)IA(f; t)ÎW (t)) (iPiA(f; t)I)(W' lain* (ipj)1W ) (1I.4.42)

 II.4 General Collision Theory 95

because the first term in (H.4.41) is cancelled by the integration (since f and
(f2IA(x)1(p i ) are wave functions with the same sign of the energy), the second
is the term written down; it is independent of t. The last term goes to zero
in the limit. We may omit writing the vectors W and W' in (II.4.42) if we use
a complete orthonormal basis of single particle states of mass m, for instance
1p, s) where p is the momentum, s the spin orientation. Then (H.4.42) can be
written as

A(f ; t) --+ a'n* (cp) as t -' —oo

with (II.4.43)
Ço9(p) = 0, s Al Q)f(p) ,
where the arrow is understood to mean weak convergence (convergence of matrix
elements) on a certain domain.
In the LSZ-formalism, described in section 3, we were dealing with a spin-
less particle and took for A the basic (scalar) field operator at the point s = O.
The matrix element (plAl,fQ) is then a constant. So cp and f can be identified if
the normalization of the field is appropriately chosen. It is now evident how the
"weak convergence method" (the generalized LSZ-formalism) can be extended
to cover situations with composite particles. By these we mean here that the
matrix elements of all basic fields from the vacuum to the single particle states
in question vanish. One simply must look for some almost local operator A
with nonvanishing matrix elements between the vacuum and these single par-
ticle states. Such operators are usually easy to find among the polynomials of
the basic fields. This "weak convergence method" applied to the bound state
problem was developed by Nishijima [Nish 58] and Zimmermann [Zi 58].

Remarks: In the proof of the main theorem 4.2.1 given in [Haag 58] the fast
decrease of truncated functions in space-like direction was introduced as an ad-
ditional assumption, made plausible on physical grounds, with the hope that
this assumption could be shown to be a consequence of the axioms. Several
authors obtained partial results in this direction. The situation was completely
clarified by the work of Ruelle. He showed that if there is a lowest mass then
the fast decrease of truncated functions follows (lemma 4.1.3). Ruelle also made
all the estimates involved in the proof watertight. In particular he avoided the
use of the asymptotic expansion (I1.4.25) and replaced it by lemma 4.2.2. The
truncated Wightman functions or, alternatively, the left hand side of (II.4.7) in
the case when the Qk are local instead of almost local, decrease proportional to
e-mR . This was proved by Araki, Hepp and Ruelle [Ara 62b]. An estimate inde-
pendent of the type of the support regions was given by Fredenhagen [Fred 85b].
He obtained the estimate

l(fljABlfl) — (f-2lAlrl)(r2lBr2)1
6 If there are several types of particles with the same mass then s must also include the
index distinguishing these types.
96 II. General Quantum Field Theory

^ e -mR {ii A* S2 ii i^ ASl ^^ ^i B* f2 BO 11} 1/2 , (1I.4.44)

where m is the smallest mass, R the largest distance so that eimAe-iHt still
commutes with B.
The discussion of the detector arrangements is taken from [Ara 67] and will
be resumed in chapter VI. The relation to the generalized LSZ-formalism is
discussed in [Hepp 65] and [Ara 67].

II.5 Some Consequences of the Postulates

II.5.1 CPT-Operator. Spin-Statistics-Theorem

In discussing the analyticity of Wightman functions wn(z l , • • • zn ) two important

facts were mentioned in section 2. First, if z1 , • • zn is in the analyticity domain
and A any complex matrix satisfying (II.2.11) then 4, • • • zn with z k = Az k
is in the analyticity domain and wn(zx, • • • zn) may be found by applying (an
extension of) the transformation law (II.1.9). Secondly in the complex Lorentz
group the space-time reflection xµ —> —xo can be connected by a continuous
path to the identity.
Let us look at the extension of (II.1.9) to the complex Lorentz group. The
relation between •L and SL(2,C) was established by writing a real 4-vector as
the matrix (1.2.14) and its Lorentz transform by (I.2.16). If we keep (I.2.14)
for complex 4-vectors and allow complex Lorentz transformation then (I.2.16)
is generalized to
V = aVQT (II.5.1)
where a and ,Q are independent matrices from SL(2,C). The special case of
real Lorentz transformations is /3 = â. The covering group of the (connected
part of the) complex Lorentz group is thus given by SL(2,C) x SL(2,C) i.e its
elements are pairs (a, (3). The complex Lorentz matrix A determines the pair
up to an overall sign. Considering now transformations of spinors we see that
the generalization (II.5.1) just means that a spinor cp r (with undotted index)
is transformed with the matrix a while a "conjugate spinor" cp,. (dotted index)
is transformed by (3 instead of â. From this the transformation properties of
spinors of higher rank with several undotted and dotted indices follow. The
space-time reflection is represented by the pair (a, /3) = (1, —1) or by (-1,1) .
To connect it to the identity we may, choosing the first alternative, pick a
continuous path in SL(2,C) leading from /3 = 1 to 0 _ —1 (e.g. a rotation by
360 degrees) while keeping a fixed at 1. This will leave a spinor with undotted
index unchanged and give a factor —1 for a spinor with dotted index. For
a spinor of higher rank with n undotted and m dotted indices we will pick
up a factor (-1)m. It suffices to consider such "pure" spinorial fields since all
covariant fields are (direct) sums of such.
This transformation law gives the following relation between Wightman
functions
 11.5 Some Consequences of the Postulates 97

wn(zl, • • . zn ) = (-1) Mwn(—zl, ... — zn ) (11.5.2)

where M is the total number of dotted indices of all fields occurring in wn.
Each zk is attached to a specific component of a specific basic field. If we would
let the imaginary parts of the zk go to zero in (II.5.2) we would not get a
relation between the Wightman distributions with real arguments because the
imaginary parts of the difference vectors of successive points have opposite signs
on the two sides. The relation holds, however for Jost points and there, due to
the local commutativity axiom E we may change the order and write ([Jost 57] )

wn (zi, ... z n ) = (-1) P(-1) M wn(—z n , ... — z1 ). (II.5.3)

The first sign comes from the interchanges of anticommuting fields needed in
the reversal of the order. Now, if zr, ... z n is in the primitive domain so is
—z n , ... — zl and therefore we may let the imaginary parts go to zero and retain
a relation between Wightman distributions valid for all real points. Specializing
(II.5.3) to n = 2 with 01 = 0, 02 = 0* we get

( 0 1 0 (x) 0* (y)If2 ) = (-1)P+M(QI0*(—y)P(—x)I(2), (II.5.4)

and after integration with test functions f (x) • 7(y)

II c* (f)Q 112= (-1)P+M II ÇP(f),rl 112

with f(x) = f (—x). So P + M must be even. 1 In the case considered M is the

sum of the numbers of dotted and undotted indices of 0. It is even for true
representations and odd for double valued representations. P is even if 0 com-
mutes with 0*, odd if it anticommutes (at space-like distances). This suggests
that we can make the Bose-Fermi alternative in axiom E more precise.

Definition 5.1.1
0 is called a Bose field if

[0(x), 0*(y)] = 0 for (x — y) 2 < 0, (II.5.5)

and it is called a Fermi field if

[Cx), ^* (y)] + = 0 for (x — y) 2 < O. (1I.5.6)

We can then express the conclusion from (II.5.4) as the

"Spin-Statistics" Theorem 5.1.2

Fields belonging to integer spin representations are Bose fields, fields belonging
to half odd integer spin representations are Fermi fields.

1 The other possibility would be the vanishing of sPQ and Q. But this can be excluded
because it would lead to sP = 0 due to theorem 5.3.2 below.
98 II. General Quantum Field Theory

To evaluate the factor (--1) P in (II.5.3) for general configurations of fields

we need to know the (space-like) commutation relations between different fields.
One has the following general restriction:

Lemma 5.1.3 [Dell' Ant 61]

If [01(x), C (y)]± = 0 for (x — y)2 < 0

then also [C(x), 0 2 (y)4 = 0 for (x — y) 2 < O.

Proof.

(fl1C(f)V2`(g)^2(g)C(f)1f 2) = ti 02(g)01(f)IQ) 112 > O.

If the supports of f and g lie space-like with a minimal separation R then the
above expression equals

u(Ql cP (f) 0 1(f)C(g) 452(g)IQ)

u(fliC(f)1(f)I QQ)(Q102(g).P 2 (g)1Q)

-I

R--■ o0

= a 11 C(f)f2 11 2 II'P2(g)Q 11 2 .
Here u = ±1 if 02(g).P2(g),4Pi(f)]T = O. The positivity of the norms squared
implies u = +1 which is the claim of the lemma. D

This suggests the following convention.

Normal Commutation Relations. At space-like distances any two Bose

fields commute, any Bose field commutes with any Fermi field and any two
Fermi fields anticommute.

Comment. We called this a convention because it can be shown [Ara 61] that
although different possibilities exist they can always be transformed to the "nor-
mal" case by a redefinition of the fields in a way which does not change the phys-
ical consequences. This is related to superselection rules which will be discussed
in a more general context in detail in chapter IV.
Assuming now normal commutation relations we have in (II.5.3)

P = F(F — 1)
21

where F is the number of Fermi fields in the brackets. For non vanishing Wight-
man functions F must be even, so

(-1) P = (-1) F/2 = 2F .

Then we get from (II.5.3) the theorem

 II.5 Some Consequences of the Postulates 99

CPT-theorem 5.1.4 [.lost 57]

a) There exists an antiunitary operator e uniquely defined by

645 (x)6 -1 = q5c (x) = (-1) m (—i) F P*(—x), (I1,5.7)

eft = Q. (II.5.8)

_ 0 if 0 is a Bose field
F—
1 if 0 is a Fermi field,

m is the number of dotted spinor indices of 0.

b)
eU(a; a) = U(—a; a)9. (1I.5.9)
c) 02 commutes with all Bose fields and anticommutes with all Fermi fields.
Alternatively stated

0 2 _ +1 on the subspace of integer spin states

—1 on the subspace of half odd integer spin states.

d) (9 transforms a single particle state of species i into a single particle state

of species i the antiparticle of i. It is not excluded that particle and antiparticle
,

are identical. If they are different then we may use bases 1pAi) in 7-11 ,É and +pai)
in Hl É and where A denotes an array of undotted and dotted spinor indices
A = 7'1 r2 ... , Sl, s2 ... (ri = 1, 2; sÉ = 1, 2); A = ?"1, 7'2... , sl, s2 .. • so that
,

elpAi) = (-1)mfpai) (II.5.10)

with m the number of dotted indices in A. 2

e)
elvl,...çon)out = )in (1I.5.11)
.,.(Prci
with
14) = ekPk).
The wave function of the charge conjugate single particle state cpc is related to
that of cp by
(P.1i(P) = ( -1 )m7)ai(P)• (II.5.12)
For the S-matrix elements this gives
2 If particle and antiparticle are identical then we may still keep (11.5.10) but then the two
bases refer to the same Hilbert space 7 -fi,= and we have to give the relation between them. In
the example of a Majorana particle

Ipr) prâ'E e s l ps)

where pT8 is the spinorial form of the 4-vector p i.e. the matrix 13= (po + per) re .
100 II. General Quantum Field Theory

((pi,
out ' ' ... in _ out C .. ^nC ^1'C , .. . 'C `in
(11.5.13)
^1^ ^n^ — vei,. (P m )

and for the S-operator

es = s-1 e. (II.5.14)
Proof. Equation (II.5.3) remains valid for real points and can then be written,
using the knowledge of the sign factor (-1) P and inserting the definition of 'PC
from (11.5.7) as

WIC (xi)• • • 'Pn.(xn)IQ) " (QOn*(2n) ... *(x1)IQ).

The antilinear operator 8 defined by

... (II.5.15)
641(8) ... n(fn).(2 = (f 1) 'Pn (fn)f2

is therefore antiunitary i.e. it maps 1 1 onto itself and satisfies

-

(8w2Iew1) = (w1Iw2)•

The definition (1I.5.15) for 8 coincides with the definitions (II.5.7), (II.5.8).
Concerning b) one verifies directly that the unitary operator 8U(a; a)8 -1
transforms each field in the same way as the operator U( —a; a). Because of the
completeness axiom F the two operators can differ only by a numerical factor.
But they have the same action on (2, so they coincide. Part c) follows simply
from the definition (II.5.7).
From b) it follows that 8 transforms a state with momentum p and spina to
one with the same momentum but spin Â. For single particle states, associated
with an irreducible representation of i13, this fixes the image vector up to a
phase which depends on the convention adopted for the matrix elements of the
fields between the vacuum and the single particle state vectors. If particle and
antiparticle are different then the only restriction is provided by c) and (II.5.10)
is a consistent choice.
Part e) follows from a), b) together with (I1.4.17), (II.4.18), (11.4.19).

II.5.2 Analyticity of the S-Matrix

The wish for better understanding of the mathematical problems and concep-
tual structure was one of the motivating forces in the development of general
quantum field theory. Another, more pragmatic, motive was provided by the
status of the theory of strong interactions. It had become clear that a pertur-
bation treatment of the Yukawa theory of nucleon and pion fields could not
lead to a quantitative comparison between theory and experiment. Therefore
it became of interest to establish relations between different experimentally
amenable quantities which did not depend on a detailed model but tested only
general basic assumptions of the theory. A prime example was afforded by the
dispersion relations for the forward scattering amplitude. It was recognized that
the axioms of section 1 together with the reduction formulas of section 3 suffice
 11.5 Some Consequences of the Postulates 101

to show that the forward scattering amplitude for pion-nucleon scattering is an

analytic function of the energy in the cut complex plane. One may then repre-
sent this amplitude as a Cauchy integral in terms of the jump across the cut.
The latter, called the "absorptive part" is related to the total cross section. One
so obtains a dispersion relation which gives the forward scattering amplitude at
energy E as an integral involving the total cross section at other energies.
The derivation of these relations from general principles and their experi-
mental verification with significant accuracy has been hailed as one of the tri-
umphs of general quantum field theory. It put a definite end to the conjecture
about the existence of a fundamental length of 10 -13 cm to 10 -14 cm, below
which the causal structure of space-time might become meaningless. 3 It stim-
ulated much effort to derive further analytic properties of the S-matrix from
first principles. These efforts met with some but not with spectacular success.
There is no need to treat this topic here since there exist many extensive and
easily available discussions in the literature. 4
Another strategy, pursued and propagated in particular by G. Chew in the
sixties, was to assume that S-matrix elements in momentum space are as ana-
lytic as they possibly can be, irrespective of what one can prove from quantum
field theory. From a pragmatic point of view this appears eminently reasonable.
From a fundamental point of view the renunciation of microscopic causality in
space-time as envisaged earlier by Heisenberg's idea of a pure S-matrix the-
ory seems to be premature. The subsequent development of elementary particle
physics indicates that the special relativistic space-time concept is viable up to
the highest energies available in present day accelerators.

11.5.3 The Theorem of Reeh and Schlieder

By the same argument which was used in section 2 to derive the analyticity of
Wightman functions one finds that matrix elements

F(x 1 , ... x n ) = (W1.1.(x 1 ) . . . .t.(xn) IW) (II.5.16)

are boundary values of analytic functions in x-space for any W E D if the spectral
support (in energy) of the vector W1 is bounded. The analyticity domain is the
tube

z1 = x1 + 2171, Z2 = (x1 — x2) — î112, ... zn = (xn-1 — xn) — bin (II.5.17)

where all imaginary parts Tlk lie in the forward cone, the real parts being arbi-
trary. It follows then from the "edge of the wedge theorem" that if F vanishes
3 It was this conjecture which, in train of the frustration with quantum field theory in the
late thirties, had led Heisenberg to propose a pure S-matrix theory from which concepts like
fields, space and time are eliminated at the fundamental level. Heisenberg did, however, not
adhere to this point of view for long. I remember him saying in a discussion in 1956: "The
S-matrix is the roof of the theory, not its foundation."
4 See e.g. [Bjorken and Drell 19651 and references quoted there. The dispersion relations
do not imply the existence of point-like fields. Compare [Ep 691, [Buch 851.
102 II. General Quantum Field Theory

within some open region of real configuration space then F vanishes for all
configurations of points in Minkowski space. This has the startling consequence
Theorem 5.3.1 (Reeh and Schlieder, [Reeh 61])
The set of vectors A(0)Q, generated from the vacuum by the polynomial alge-
bra of any open region, is dense in fil.

To see this take W1 = Q and W in the orthogonal complement of A(0)IQ).

Then F = 0 as long as all x i E O. Hence F = 0 identically. Hence W = 0.

Remarks: (i) Intuitively one might have thought that with Q E A(0) the vec-
tor QS2 could be interpreted as representing a state localized in 0, i.e. a state
looking practically like the vacuum with respect to measurements in the causal
complement of O. While, due to the cluster property of Wightman functions,
this is qualitatively true if Q is picked at random in A(0) and measurements at
a sufficiently large space-like distance from O are considered, the theorem tells
us that for any chosen state vector W one can always find an operator Q E A(0)
which, applied to the vacuum, produces a state vector arbitrarily close to W. To
achieve this the operator must judiciously exploit the small but nonvanishing
long distance correlations which exist in the vacuum as a consequence of the
spectral restrictions for energy-momentum in the theory. The theorem shows
that the concept of localized states, if used in a more than qualitative sense,
must be handled with care.
(ii) Obviously in the theorem the vacuum may be replaced by any vector with
bounded energy.
Closely related to theorem 5.3.1 is

Theorem 5.3.2
If 0 has non void causal complement then A(0) does not contain any operator
which annihilates the vacuum (or any other state vector with bounded energy).

11.5.4 Additivity of the Energy-Momentum Spectrum

The simultaneous spectrum of the energy-momentum operators PA is a closed

subset of 4-dimensional p-space. We denote it by Spect P. If d is an open region
of p-space which intersects Spect P then there exists an almost local operator
Q with QS2 I 0 which, in the notation of lemma 4.1.2 , has
supp Q(p) C Q. The state vector W = QS2 has spectral support in A. The
translated vector U(a)W has the same spectral support.

Theorem 5.4.1
If pi and p2 are in Spect P then p i + p2 is in Spect P.

Proof. Take neighborhoods Ai, Q2 of the points p i , p2 respectively and al-

most local operators Q i , Q2 as described above. Then
 IL5 Some Consequences of,the Postulates 103

W21( a) - Q2 (a)Q1.f2 (II.5.18)

(where a is an arbitrary translation 4-vector) will have its spectral support

in Q1 + Q2. We have to show that 11 W21(a) (I does not vanish for every a.
The essential property needed to show this is the general cluster property. For
•oo we get

lim II W2i(a) 11 2 = (QIQ2(a)Q2(a)1Q)(QIQ*Q11Q) =11 W2 11 2 11 WI 11 2 • (II.5.19)

So W21(a) cannot vanish for all a.

Remark: In Ruelle's lemma 4.1.3 the cluster property was proved using some
assumptions about the spectrum, namely the positivity of the energy and the
existence of a mass gap. Under these assumptions it was shown that the rate of
decrease of truncated functions is faster than any inverse power of R. For the
proof of theorem 5.4.1 we do not need to know the rate of decrease. The general
cluster property, demanding only that the truncated functions vanish as the
distances go to infinity, follows without any assumptions about the spectrum
from the commutativity of space-like separated observables, and the existence
of a normalizable, pure, translation invariant state (the vacuum) (see chapters
III and V).
If Spect P is Lorentz invariant (which follows for instance if we accept axiom
Al) then theorem 5.4.1 leaves only the possibilities

a) Spect P C V.
Then Spect P D {p : p° > 0,p2 > M2 for some M}, (II.5.20)
b) As a) with the sign of the energy reversed,
c) Spect P = IR4 .

To see this let p E V+ with p2 = m2 . Varying the vectors P1, p 2 on the hy-
perboloid with mass m the sum P1 + p2 will have a mass ranging continuously
from M = 2m to infinity. This demonstrates the second part of a). If P1 E V+
and P 2 E V- are in the spectrum then, using the second part of a) (resp. b)),
we have vectors with equal mass but opposite energy in the spectrum. By ad-
dition of these we can reach a space-like p with arbitrary negative p 2 . Finally,
if a space-like p is in the spectrum, say with p 2 = —/c2 , then the addition of
Pf = (/p2 k2 , ±p) will lead to vectors in V+ with arbitrary mass as C pl varies
—

from IC to infinity. Choosing the zero components of p t negative we can reach

all mass values in V.
In conclusion we note that the spectral properties and the locality principle
are interwoven. The alternatives a) and b) are physically equivalent. The one
results from the other by changing the convention in the choice of the sign
of the energy. Thus, if any open set in p-space is outside the spectrum, then
Spect P has the "physical shape" demanded by axiom A3 and further restricted
by (I1.5.20). That the locality principle does not rule out the alternative c) has
been demonstated in a model by Doplicher, Regge and Singer [Dopl 68 b}.
104 II. General Quantum Field Theory

II.5.5 Borchers Classes

Let 0 be a field system satisfying the axioms A—F and let be another field
system acting in the same Hilbert space and satisfying A—D with respect to
the same representation U(a, a) of Then 0' is called local relative to if
.

[ (x), .P(y)] = 0 for space-like x — y.

One has

(i) if 0' is local relative to 0 then 0' satisfies axiom E, i.e. it is itself a local
field system.

(ii) Let 0, 0', 0" be complete, local field systems in 11 for given {U(a, a)}. If
0' is local relative to and cP' local relative to 0' then 0" is local relative
to 0.

Thus relative locality for complete, local field systems is transitive and can serve
to define an equivalence relation. We call two such field systems equivalent if
they are local relative to each other. An equivalence class of complete, local
field systems is called a Borchers class because Borchers introduced the notion
of relative locality and derived the essential results in [Borth 60]. From the
physical point of view the terminology "equivalence" is justified because one
finds

(iii) The S-matrix depends only on the Borchers class, not on a specific field
system chosen in this class.

A simple illustration of a Borchers class is available in the case of a free field cp.
In this case the Wick ordered powers : cp(x)n : are fields in the Borchers class of
cp and the Wick polynomials of cp and its derivatives at the same point exhaust
the class [Ep 63], [Schroer, unpublished preprint 63]. Any one of these fields
leads to S = 11.
III. Algebras of Local Observables and Fields

III.1 Review of the Perspective

In quantum physics just as in classical physics the concept of "field" serves to

implement the principle of locality. In particular, a "quantum field" should not
be regarded as being more or less synonymous with a "species of particles".
While it is true that with each type of particle we may associate an "incoming
field" and an "outgoing field", these free fields are just convenient artifacts
and the discussion of collision theory in Chapter II, sections 3 and 4 should
make it clear that the primary physical interpretation of the theory is given in
terms of local operations, not in terms of particles. Specifically, we have used
the basic fields to associate to each open region O in space-time an algebra
.A(0) of operators on Hilbert space, the algebra generated by all «f), the
fields "smeared out" with test functions f having their support in the region O.
We have interpreted the elements of A(0) as representing physical operations
performable within O and we have seen that this interpretation tells us how to
compute collision cross sections once the correspondence

O -- .A(0) (III.1.1)
is known.
This suggests that the net of algebras A, i.e. the correspondence (III.1.1),
constitutes the intrinsic mathematical description of the theory. The mentioned
physical interpretation establishes the tie between space-time and events. The
rôle of "fields" is only to provide a coordinatization of this net of algebras. This
point of view is supported by the observation of Borchers concerning equiva-
lence classes of fields: different choices of a system of fields may yield the same
S-matrix. Extrapolating from that we expect that no physical consequence of
the theory depends on the choice of a specific field system within a Borchers
class or, alternatively speaking, on the way how the net is coordinatized.

Bounded Operators. The algebras constructed from fields in the way de-
scribed above are called polynomial algebras because their elements are obtained
as sums of products .P(ff )cP(f2 ) ... of smeared out fields. In the frame of section
II.1 these elements are unbounded operators. Mathematically the polynomial
algebras are somewhat unwieldy. Since one is dealing with unbounded operators
the questions concerning their domains have to be considered carefully. Since
106 III. Algebras of Local Observables and Fields

one is dealing with operator valued distributions the topology in such an algebra
is injected by the Laurent Schwartz topology of the test function spaces which is
an extremely fine topology, too fine to be of intrinsic significance for the physical
content of the theory. Thus, polynomial algebras of different field systems in the
same Borchers class will not coincide in general. Remembering the discussion in
I.1, in particular the approach by Segal [Seg 47], it appears that, without loss of
generality, we may go over to bounded operators. Given an observable, mathe-
matically represented by a self-adjoint (possibly unbounded) operator, we may
consider instead its family of spectral projectors or the bounded functions of it.
In the case of more general operations, represented by (closeable) unbounded
operators, we can make a polar decomposition yielding an isometric operator
and a positive self-adjoint one which may be spectrally decomposed.
For algebras of bounded operators on a Hilbert space there are several math-
ematically natural topologies and it will become apparent that they also have
physical significance. They lead to the concepts of von Neumann algebras and
concrete C*-algebras. We shall outline some of the basic mathematical theory of
such algebras in the next section. In most of the subsequent discussions we shall
take the algebras A(0) in the correspondence (III.1.1) as algebras of bounded
operators.
The question as to whether the Wightman axioms are equivalent to a theory
formulated in terms of a net of algebras of bounded operators has been the sub-
ject of extensive discussions. One may ask whether an observable field, smeared
out with a test function with compact support and defined on the Wightman
domain, has an extension to a self adjoint operator and whether this extension
is unique (Borchers and Zimmermann, [Borch 64]). Or one may ask under what
conditions the construction of von Neumann algebras by polar decomposition of
operators from polynomial algebras leads to a net respecting the causal struc-
ture. Such questions have been studied by Driessler, Summers and Wichmann
[Driess 86] and by Borchers and Yngvason [Borch 90]. Conversely one may ask
whether, given a net of algebras of bounded operators, one can define fields by a
limiting process, shrinking the regions to points and whether this yields all field
systems in a Borchers class (see [Fred 81b], [Summ 87]). We shall not enter here
into a discussion of these questions but note that the Wightman axioms alone
are not sufficient to guarantee the existence of a net of local algebras of bounded
operators and that, conversely, such a net does not guarantee the existence of a
field system satisfying the Wightman axioms. For most purposes, however, the
difficulties of passing from one frame to the other may be ignored.
To what extent can the axioms of section II.1 be translated into strucure
properties of a net of local algebras?

Axiom A, dealing with the representation of the Poincaré group, is unaffected.

Besides the algebras A(0) there are the representors U(a, a) of the symmetry
transformations and we shall keep the assumptions pertaining to them, most
importantly the existence of a vacuum state and the positivity of the energy.
We shall, however, try to relate these later to algebraic properties of the net.
 III.1 Review of the Perspective 107

Axiom B, introducing the notion of fields, suggests the additivity property

.A (0 1 U 02 = ,A (0 1 ) V A( 0 2 )
) (III.1.2)

where the symbol V on the right hand side shall denote the operator algebra
generated by the two algebras .A(0 i ), i = 1, 2. For a precise formulation see
section 4.

Axiom C (Hermiticity) means that A(0) is an involutive algebra, a

*-algebra: apart from the algebraic operations of forming products and linear
combinations of elements one has within each A(0) the involution A -* A*
which assigns to every element A its Hermitean adjoint A*.

Axiom D, concerning the transformation properties of fields, becomes

U(a, a)A(0)U -1 (a, a) = A(A(a)0 + a). (III.1.3)

The geometric symmetry operations map the algebra of one region onto the
algebra of the transformed region. The detailed transformation laws of tensorial
or spinorial fields from which the algebra may be generated are lost.

Axioms F (completeness) and G ("primitive causality") are transcribed in an

evident way.

Axiom E combines several features abstracted from conventional field theoretic

models. The main principle expressed by it is the causal structure of events.
Two observables associated with space-like separated regions are compatible.
The measurement of one does not disturb the measurement of the other. The
operators representing these observables must commute.

To avoid possible confusion it must be stressed that this has nothing to do

with the discussion around the Einstein-Podolsky-Rosen paradox and Bell's in-
equality. There one is dealing with the joint probability distribution of measure-
ments on two far separated particles coming from a common root e.g. as decay
products of an unstable particle. If a neutral particle decays into two oppositely
charged ones it will surprise nobody that a charge measurement on one of the
decay products suffices to tell us the charge of the other one, no matter how far
away it is. This is due to the correlation resulting from charge conservation, not
to a causal influence between the charge measurements in space-like separated
regions. The total experiment includes, of course, the preparation of the state of
the unstable particle, by which the charge (resp. spin) are fixed. If, instead of the
charge, we consider the angular momentum, the situation becomes indeed more
curious. Instinctively one would like to associate with each of the particles (once
they are sufficiently separated) an "objective", "real" state which determines
the probability of finding a specific result in the subsequent measurement of
the angular momentum component in any chosen direction. As Bell has shown
108 III. Algebras of Local Observables and Fields

[Bell 64] this picture, together with angular momentum conservation, demands
that a certain inequality must hold for the joint probability distributions for
such measurements on the two particles. This inequality is not satisfied in the
quantum mechanical description. Very fine experiments have been performed to
check this inequality. They appear to speak for quantum mechanics and against
the inequality. What is the message of this? It does not relate to a physical
influence propagating faster than light but it illustrates in a particularly drastic
way that the concept of a materially defined "physical system" has to be han-
dled with extreme care. This latter is a mental construct whose correspondence
to "reality" is (sometimes) questionable (compare [d'Espagnat 1979]). We shall
give a thorough discussion of this problem in Chapter VII. Here we note only
that the existence of correlations between far space-like separated events does
not contradict the limitation of causal influences to time-like directions as de-
manded by axiom E.

Unobservable Fields. Returning after this aside to axiom E we may note

that in quantum field theory there occur observable and unobservable fields.
The former generate a net of algebras A 0 (0) in which the causality principle
is expressed by

[A 1 , A2] = 0 if A E Âoh(0 ), i = 1, 2 and 0 1 is space -- like to 02 .

(III.1.4)

In the language of II.1 the observable fields are Bose fields though not all Bose
fields need to be observable. But space-like commutativity is not postulated
with a view of introducing Bose statistics but is a mirror of the causal struc-
ture of space-time. Why does one need unobservable fields at all and what are
their commutation properties? A pragmatic answer to the first question comes
from the existence of superselection rules. It has been recognized very early by
Wigner that one cannot have an unrestricted superposition principle among
the pure physical states. If W1 and W2 are state vectors carrying integer spin
and half odd integer spin, respectively, then the relative phase between W1 and
W2 can have no observable significance because a 360°-rotation (which is the
identity transformation) changes the relative sign between these state vectors.
Similarly one expects that the relative phase between vectors describing states
with different electric charge should be meaningless. This phase is changed by
a gauge transformation which has no effect on the observables. The situation
is described by Wick, Wightman and Wigner [Wick 52] as follows: The Hilbert
space h to which axiom A refers is a direct sum of subspaces 7-lk which we
may call coherent subspaces or superselection sectors. Within each 74 one has
the unrestricted superposition principle whereas phase relations between state
vectors belonging to different sectors are meaningless. 1
'Alternatively speaking, a linear combination of vectors from different sectors does not
represent a pure state but a mixture and only the absolute square of the coefficients are
relevant as the weights of the components.
 III.1 Review of the Perspective 109

Within our present context we may say that the observable algebras Aobs(0)
transform each sector into itself; they do not connect states in different sectors.
The rôle of the unobservable fields is to lead from one sector to a different
one. Unobservable fields transfer some "superselection quantum number", some
"generalized charge". They are "charge carrying fields". Still we may insist that
the theory must be completely fixed by the net .Robs , by the observable algebras
alone. Since these do not connect different sectors we have in each sector a net
of operator algebras Aobslk, the restriction of Aobs to 7`tk. These nets are not
unitarily equivalent but algebraically isomorphic. Moreover any single one of
them must contain all the physically relevant information since different sectors
are distinguished by a global property (total charges). We may change the charge
of a state by adding a charge arbitrarily far away and this will change the
physical situation in any finite part of space-time arbitrarily little. The natural
explanation of the situation is then the following: The intrinsic structure of the
theory is fully characterized by the algebraic relations in the net of observable
algebras. In other words, the basic object is a net of abstract algebras (as opposed
to their representative operator algebras on a Hilbert space) . We choose them
as abstract C*-algebras, a concept which will be discussed in section 2.2. We
shall henceforth denote the (abstract) observable algebra of the region O by
21(0). To emphasize the local point of view we may regard the algebras 21(0)
to be defined only for finite regions.' From these we may define the algebra 9.110c
of "all local observables" as
24oc = U21(0) (111.1.5)
where U denotes the set theoretic union taken over all finite regions 3 and the
C*-algebra
21 _ 240c, (111.1.6)
the completion of 211oc in the norm topology. In physical terms 21 may be called
the algebra of all quasilocal observables; in mathematical terms it is the "C*-
inductive limit" of the system {21(0)}.
Superselection rules arise if the net 21 possesses several inequivalent repre-
sentations by operator algebras acting on a Hilbert space. The divorce of the
basic description of the theory from Hilbert space brings a tremendous addi-
tional freedom. It allows us, for instance, to consider thermodynamic equilib-
rium states corresponding to Gibbs' grand canonical ensembles in infinite space
or, more generally, any kind of distribution of matter and energy extending to
infinity. We shall indeed discuss such states in chapter V but of more direct in-
terest in elementary particle physics are the charge superselection sectors. They
arise if there remain inequivalent representation of 21 even when we restrict the
attention to states which "look like the vacuum" at space-like infinity. We treat
them in chapter IV.
2 By a "finite region" we always understand an open subset of Minkowski space with
compact closure.
3 216c is an algebra because, according to the interpretation, the net must have the isotony
property 2402 ) D 2I(01 ) if 02 D 01 and for any two finite regions there exists another finite
region containing them both.
110 III. Algebras of Local Observables an d Fields

Once we base the theory on abstract algebras we must reconsider the defini-
tion of Poincaré symmetry and axiom A. Poincaré invariance means now that to
a transformation g E q3 there corresponds an automorphism a 9 of the abstract
net with the property
01921(0) = 2t(g0). (III.1.7)
In words: a9 maps the elements of 21(0) onto the elements of the algebra of the
transformed region gO in such a way that all algebraic relations are conserved.
A representation ir of ?t is a homomorphism from the net 2t to a net of op-
erator algebras ir(Qt) i.e. ir assigns to each algebraic element A its "representor"
ir(A), an operator acting in a Hilbert space, in such a way that the mapping
conserves the algebraic relations. Given a representation the automorphism a9
may or may not be implementable. a9 is called implementable in the representa-
tion it if there exists a unitary operator U(g) acting in the representation space
such that
U(g)ir(A)U -1 (g) = ir(a9A). (11I.1.8)
Axiom A can now be replaced by the requirement that the abstract net should
possess an irreducible representation lra in which a 9 is implementable and which
is furthermore distinguished by the feature that U(g) satisfies A2 and A3. We
shall call this representation the vacuum sector of the theory (leaving aside for
the moment the possibility of vacuum degeneracy). The implementability of a9
in a representation depends on global features of the class of states described by
vectors in the representation space because g E i3 acts on all regions, no matter
how far away. Thus axiom A brings in a global aspect whose relation to local
properties of the theory is not well understood. One can, however, formulate the
contents of axiom A in purely algebraic terms and we shall do this in section 3.
The other structural assumptions are simple and natural in the algebraic
language. We collect them here once more:
a) The theory is characterized by a net of abstract C*-algebras

O —> 2t(0) (III.1.9)

where O denotes an open, finitely extended region of Minkowski space.

The self- adjoint elements of 2t(0) are interpreted as observables which
can be measured in the region O.

b) The Poincaré group is realized by a group of automorphisms of the net

(g E 43 --+ a9 ) with the geometric significance (III.1.7).

c) The causality structure is expressed by

(i) 21(0 1 ) commutes with 21(0 2 ) when the two regions lie space-like,

(ii) Let O denote the causal completion of 0 (see section 4) then

9t(6) -- a(o) . (III.1.1 0)

 III.1 Review of the Perspective 111

This stipulates that there is a dynamical law respecting the causal struc-
ture. It corresponds to the hyperbolic propagation character of fields (see,
however, the remarks at the end of section 3).

Remark. The prototype of a causally complete region is a diamond or double

cone. An open diamond is the intersection of the interior of the forward light
cone from a point x1 with the interior of the backward light cone from a point
x2 which lies in the future of x i . The standard diamond, denoted by OR , is
obtained by taking x1 = (—R, 0), X2 = (R,0). It is the causal completion of
the ball I x l< R in the hyperplane z° = O.

The above frame opens the possibility of discussing the intrinsic signifi-
cance of Fermi statistics and possible generalizations since it does not inject the
Bose-Fermi-alternative as a basic assumption, it does not tie it to commutation
relations between unobservable fields. Rather we have now the task of analyz-
ing the superselection structure, the composition law of charges, the effect of a
permutation of identical charges, the construction of unobservable fields. This
will be the topic of chapter IV. The frame also provides a natural approach to
the description of thermal equilibrium states and more general hydrodynamic
states without the need to define a material system enclosed in a box. This will
be discussed in chapter V.

Remarks and references. The proposal to base the theory on a net of local
algebras corresponding to space-time regions originated in [Haag 57]. There the
algebras were thought of to be generated by both observable and unobserv-
able fields and taken as operator algebras acting on Hilbert space. In the first
applications to physical and structural questions polynomial algebras or von
Neumann rings were used according to convenience and the step from the for-
mer to the latter was considered as unproblematic. The main reward was the
deeper understanding of collision theory as described in II.4 which eliminated
the distinction between elementary and composite particles and the assumption
of a correspondence between elementary particles and basic fields. A survey of
the postulates from this point of view was given by Haag and Schroer [Haag
62a]. The development of the theory in terms of nets of von Neumann algebras
of observables is largely due to Araki. He built a fairly self-contained frame on
solid mathematical ground [Ara 62a]. The idea that one should take an abstract
C*-algebra as the basic object and that the representation problem (see H.1) is
irrelevant for physics, was advocated by Segal [Seg 47], [Seg 57]. An adequate
physical interpretation was lacking, however. His proposal that the S-matrix
should be considered as an automorphism of the algebra was not acceptable.
It took some years till it was realized that a net of (abstract) C*-algebras of
observables provided the natural setting for understanding superselection rules
and the rôle of unobservable fields (Haag and Kastler, [Haag 64]).
112 III. Algebras of Local Observables and Fields

III.2 Von Neumann Algebras. C*-Algebras.

W*-Algebras

III.2.1 Algebras of Bounded Operators

The purpose of this section is to provide a brief glossary of mathematical terms,

concepts and theorems which will be subsequently used. Proofs are only sketched
or omitted altogether and the reader is referred to the bibliography at the end
of the section to supplement this minimal exposition.
The set of all bounded, linear operators acting in a Hilbert space N
is denoted by a3(N). A subset S C 93(N) is called an algebra if, with
A, B E S, a, 0 complex numbers also aA + fiB E S and AB E S. It is called
a *-algebra if furthermore with A E S also the adjoint A* belongs to S. We
consider *-subalgebras of 93(H).
For all processes of analysis one needs a topology i.e. a definition of what we
mean by a neighborhood of an element. The most evident topology in 93(H) is
provided by the norm of operators. The statement that A E 93 (H) is bounded
means that
11A11=sup11AW11•11W11 -1 (IH.2.1)
WEN
is finite. On the right hand side II II denotes the norm (length) of vectors in the
Hilbert space, the left hand side is the norm of the operator A. One checks that

IIaAII=IaIIIAII,
IIA+ BII<—IIAII +IIB II,
II AB II—< IIAII'II BII,
IIA * A II= 11A 11 2 .
The first two relations say that 93(H) is a normed linear space.
An c-neighborhood of A is the set of operators B with If A — B II < e. The
topology based on this concept of neighborhoods is called the norm topology
or uniform topology in B(7-0. Closure of a set S in this topology means that
we add to S all elements which are limits of uniformly converging Cauchy se-
quences in S.

Definition 2.1.1
A *-subalgebra of B (H) which is uniformly closed is called a (concrete)
C*-algebra.

Other important topologies in 93(H) are defined by means of seminorms.

Thus we may pick an arbitrary vector W in N and define the "W-seminorm"
of the operator A as the vector norm II AW II. Using these seminorms for all
choices of W one obtains the so called strong operator topology in 93(H). Thus a
sequence of operators A n is strongly convergent if for every W E N the sequence
of vectors Wn = A nW is strongly convergent i.e. if II Wn — Wm I ---r 13 for n, m - oo.
 III.2 Von Neumann Algebras. C*-Algebras. W*-Algebras 113

For topologies which are defined by a system of seminorms it is not enough to

consider only the convergence of sequences. The closure of a set is obtained
by adding the limit points of "generalized sequences" (nets). Still, as a first
orientation, it is useful to look at ordinary sequences.
The weak operator topology is obtained if we use the absolute values of
matrix elements I (01AIW) I between arbitrary state vectors as a system of
seminorms. Thus a sequence of operators converges weakly if all matrix ele-
ments converge. We have encountered weak and strong convergence before in
the context of collision theory.
Still another system of seminorms is provided by ( tr AA I where P is an
arbitrary density matrix. The resulting topology was originally called "ultra-
weak". This is, apart from the norm topology, the most natural one. It is the
weak *-topology induced by the set of "normal states" (see subsection 2.2).
All the mentioned topologies are different. The following examples may
serve as an illustration. Let Wn be a complete, orthonormal basis in 7`t, En
the projector on the subspace spanned by the first n basis vectors and Tn the
operator defined by Tn Wk = Wn+k. Then the sequence of operators E n converges
strongly to the unit operator for n -+ oo because the component of each fixed
vector W in the orthogonal complement of En7-1 tends to zero. This sequence
does, however, not converge uniformly since II En — Em 11= 1 for any n in.
The sequence Tn converges weakly to zero but not strongly.
It is evident that the uniform topology is the strongest, the weak operator
topology the weakest. The weak *- and the strong operator topology cannot be
compared.

Definition 2.1.2
A weakly closed *-subalgebra of 93(N) which contains the unit operator will be
called a von Neumann algebra or (synonymously) a von Neumann ring.

Remark. The requirement that the unit operator shall be contained in it is

not always considered as part of the definition of a von Neumann ring. It is
added here to simplify the formulation of some later theorems.

Quite generally, if a subset is closed in one topology it is automatically closed

in every stronger topology because, with decreasing strength of the topology,
one gets more and more limit points. Thus a von Neumann algebra is also a con-
crete C*-algebra whereas the converse is not true. However von Neumann has
shown that for a *-subalgebra of B(7-t) the closures in the three other mentioned
topologies (strong operator, weak *, weak operator) all coincide. Therefore we
meet only two kinds of topological *-subalgebras of q3(7-() namely concrete Ce-
algebras and von Neumann algebras.

Definition 2.1.3
The commutant of an arbitrary subset S C 93(7-0 is the set of all bounded
operators which commute with every element of S. It is denoted by S'.
114 III. Algebras of Local Observables and Fields

Theorem 2.1.4
Let S C B(7l) be a self adjoint set (i.e. S shall contain with each element also
its adjoint). Then

(i) 5' is a von Neumann ring.

(ii) S' - (S')' is the smallest von Neumann ring containing S.
(iii) Sf' = S'.

Comment. If A and B commute with X and X* then obviously also linear

combinations of A and B, the product AB and the adjoint A* will commute
with X . The unit operator certainly commutes also with X . Thus S' is a
*-algebra with unit. To obtain (i) one must show that S' is closed (strongly
or weakly). If An is a sequence of bounded operators converging strongly to A
and satisfying [As , X] = a then one sees easily that also [A, X] = O. One shows
(not quite trivially) that it suffices to add the limits of strongly convergent se-
quences to obtain the strong closure of a *-algebra in a3(7-l). Statement (ii) is
the famous von Neumann double commutant theorem. We shall not sketch its
proof here. Statement (iii) follows from

S„ _ ((s')') ' = (s") ' = (S)"; S C S"; S' C (S)",

noting that S C T implies T' C S.

We shall from now on use the symbol 7Z to denote von Neumann algebras.
From the theorem it follows that a von Neumann algebra may be characterized
as a self-adjoint subset of 93(7 -i). satisfying

R" = 37- (11I.2.6)

Denoting the smallest von Neumann algebra containing R. and R.2 by R1 V R2,
the largest von Neumann algebra contained in R. and in R.2 by R1 A R.2 one
has
7Z1VR2E(R1 UR2 ) (III.2.7)
and
(R1vR2)'=Ri nR2=Rr n 7Z2. (II1.2.8)
The set of von Neumann algebras on a Hilbert space forms an orthocomple-
mented lattice with respect to the operations A, V, and '.

Isomorphies. Let A1, A2 be two *-algebras of bounded operators acting on

the Hilbert spaces 7h1, 7 .6, respectively. An isomorphism between A l and A2
means that there exists a bijective map from A l to A2 (a one to one correspon-
dence between elements) which conserves the relevant structure. In our case
three possibilities are of interest:

1) Algebraic isomorphism. We require only that the algebraic structure (lin-

ear combinations, products, adjoints) is conserved.
 II1.2 Von Neumann Algebras. C*-Algebras. W*-Algebras 115

2) "Complete isomorphism". The algebraic and topological structures are

conserved.

3) "Spatial isomorphism" or unitary equivalence

A2 = VA1V-1 (III.2.9)

where V is a unitary map from H 1 onto N2.

It is an interesting fact that in the cases of C*-algebras and von Neumann alge-
bras algebraic isomorphism entails complete isomorphism. For two algebraically
isomorphic C*-algebras also the norms of corresponding elements agree and for
two algebraically isomorphic von Neumann algebras the norms and the weak
*-topologies agree. Two algebraically isomorphic von Neumann algebras are also
called quasi equivalent. Spatial isomorphism is more restrictive.

Reduction. Suppose that R, acting in N, transforms some proper subspace

H1 into itself. Then the projector P1 on this invariant subspace belongs to R'
and the restriction of R to N1 is

RI = P1 RP1 = P1 R = RP1 . (II1.2.10)

It is a von Neumann algebra in N1. The map

AER--+PI A (I11.2.1 1)

is a homomorphism conserving the algebraic and topological structure. One says

that the decomposition of N into H 1 and its orthogonal complement gives a re-
duction of R, or that R 1 is a subrepresentation of R. R 1 will be quasi-equivalent
to R if (11I.2.11) maps no non-zero element of R to zero. Let 3 be the set of
elements of R mapped to zero by (1E2.1 1) i.e. 3 = {A e R : PI A = 0}. Obvi-
ously 3* = 3 and 3R =1Z3 = 3. Thus 3 is a 2-sided *-ideal in R (a *-subalgebra
which is stable under multiplication with any element of R from left or right)
and 3 is weakly closed. It is not difficult to show (see e.g. [Naimark 1972])

Theorem 2.1.5
A weakly closed 2-sided *-ideal of a von Neumann ring R is always of the form

3= RE0 (III.2.12)

where E0 is a projection operator belonging to the center 3 = RC-1R' of R. E0

principal unit of Z. It satisfies iscaledth

E0 X =XE 0 =X for X e 3. (III.2.13)

The subspace (Il — E0 )N consists of all vectors which are annihilated by every
element of 3.
116 III. Algebras of Local Observables and Fields

Definition 2.1.6
The von Neumann ring R is called a factor if its center is trivial (consists only
of the multiples of the unit element):

3 = 1Z n 7Z' = {aIl} , (III.2.14)

or, equivalently,
RVR' =93(7-1). (11I.2.15)
(11I.2.15) is called a factorization of 93(H).

From the preceding discussion we conclude

Theorem 2.1.7
a) A self adjoint set S C 93(71) is irreducible (transforms no proper subspace
into itself) if S' = {Ail} or, equivalently, S" = 93(7-1) ("Schur's lemma").
b) If R is a factor then its restriction (III.2.10) to any invariant subspace is
quasi-equivalent to R.

Thus a factor contains only one quasi-equivalence class of subrepresenta-

tions. The reduction of a von Neumann algebra into factors is unique and results
from the "diagonalization" of its center. The further reduction into irreducibles
is not unique; it is afforded by the choice of some maximally Abelian subalgebral
in the commutant and its diagonalization.

Classification of Factors. Let R. be a factor. The basic classification the-

ory of Murray and von Neumann starts from a comparison and ordering of the
projectors contained in R.

Definition 2.1.8
Two projectors Pi E R (i =1, 2) are called equivalent with respect to R., in
symbols P1 ti P2 , if there exists an operator V E R. with
P1 = V*V; P2 = VV* . (1II.2.16)
These relations say that V maps the subspace H 1 = P1 7-1 isometrically on the
subspace N2 = P211 and annihilates all vectors of the orthogonal complement
of 7-11. V is therefore called a partial isometry from N 1 to 712. We write Pi > P2
or P2 < P1 if the Pi are not equivalent but there exists a subspace of 7 -11 whose
projector P11 ti P2 (of course this requires also P 11 E R.). One has

Theorem 2.1.9
Let R. be a factor, Pi projectors belonging to R. Then precisely one of the
following relations holds

P1 > P2, P1 N P2, P1 < P2 . (11I.2.17)

'An algebra R.1 C R' satisfying R.', = R l .
 1II.2 Von Neumann Algebras. C*-Algebras. W*-Algebras 117

The proof of this theorem uses the following basic lemma which is of interest
for its own sake.

Lemma 2.1.10
If R is a factor, A E R, B E R' then AB = 0 implies that either A = 0 or
B= 0.

The ordering of projectors in a factor allows the introduction of a relative

dimension for them. Dim P is a positive number (possibly oo) with the proper-
ties

(i)
> >
Dim P1 = Dim P2 if P1 ti P2. (III.2.18)
< <
(ii) If Pi is orthogonal to P2 (i.e. P1 P2 = 0) then

Dim (P1 + P2 ) = Dim P1 + Dim P2 (III.2.19)

(iii)
Dim 0-0. (I11.2.20)
These three properties determine a dimension function on the set of projectors
in a factor uniquely up to normalization. The following alternatives exist:

Ty pe I. R contains minimal projectors. P is called minimal if it is not zero

but R contains no nonzero projector P 1 < P. Normalizing the dimension
function so that it takes the value 1 on minimal projectors, Dim P ranges
through the positive integers up to a maximal value n or up to oo. If
n is finite R. is called of type Imo,, otherwise of type I. The factor I mo. is
quasi-equivalent to the full matrix ring of complex n x n matrices. I,,, is
quasi-equivalent to 93(7-1); in fact one can write 7 as a tensor product
71 irr ® 7 deg so that R = 93(7-li rr ) ® L eg Thus a factor of type I differs
from an irreducible algebra only by an additional multiplicity resulting
from the tensoring with some degeneracy space.

Type II. Dim P ranges through a continuum of values which may either have
an upper bound, in which case one can normalize the values to fill the
closed interval [0, 1] and R is called a factor of type II I. ; or it may be the
whole positive real line, including 0 and oo, and R is said to be of type
II° .

Type III. Dim P takes only the values 0 and oo. All proper projectors in R
have infinite dimension and, if 7-1 is separable (which is the only case with
which we shall be concerned), they are all equivalent.
The Relative Trace. A trace on a von Neumann algebra R assigns to each
positive operator in R a positive number, possibly oo, so that
118 III. Algebras of Local Observables and Fields

traA=atr A; (a> 0),

tr (A + B) = tr A+ tr B
and such that elements which are conjugate by a unitary operator from 7Z have
the same trace:
-I
tr UAU = tr A for all U E 7Z. (III.2.21)
Again, if 7Z is a factor, the trace is uniquely determined by these properties up
to normalization. In the type III case all non zero elements have infinite trace.
In the type I case we may choose the normalization so that minimal projectors
have trace 1. Then we get the customary definition of the trace as the sum of
the diagonal elements of the matrix representing the operator in an orthonormal
basis. In type I , as well as in II,, the trace is finite for some operators, infinite
for others. For factors of type II I the trace is finite for all elements and we may
normalize it by setting tr 11= 1.

Remarks. A finer subdivision of the types II and III has been attained in the past
decades. This development has been an example of a very fruitful interaction
between theoretical physics and mathematics. We shall come to it in chapter
V but mention already here that the local von Neumann algebras relevant in
relativistic quantum physics are of type III.

III.2.2 Abstract Algebras and Their Representations

An abstract algebra, like an abstract group, is just a set in which certain rela-
tions are defined between its elements. In our case the relations are formed by
linear combinations with complex coefficients, an associative (in general non-
commutative) product and an involution (*-operation). These operations obey
the familiar rules, specifically the distributive law for products of sums and,
with A, B elements of the algebra and a E C

A** =
= A; (aA)* = âA*; (A + B)* = A* + B*; (AB)* = B*A*. (11I.2.22)

If the algebra is equipped with a norm l) II then the norm should satisfy at
least (III.2.2), (III.2.3) and (III.2.4) in order to define a reasonable topology.
Furthermore the elements with zero norm should be identified with the zero
element of the algebra

It Ali = 0 implies A = 0. (III.2.23)

If (III.2.23) is not adopted then ll El is called a seminorm. A normed *-

algebra A which is complete in the norm topology (which contains all the limit
points of Cauchy sequences) is called a Banach *-algebra if the norm satifies
(III.2.2) - (1II.2.4) and (II1.2.23). 1f, in addition, the norm satisfies also (III.2.5)
it is called a C*-norm and A an (abstract) C"`-algebra. The requirement (III.2.5)
looks at first sight somewhat ad hoc in the case of abstract algebras. The fact
 I1I.2 Von Neumann Algebras. C*-Algebras. W*-Algebras 119

that the operator norm in 93(71) satisfies it implies, of course, that if A has
any faithful representation by operators in 13(x) then we can equip A with a
C*-norm. Disregarding the question of Hilbert space representations we shall
see that the C*-norm is distinguished and uniquely determined by the algebraic
structure through its relation with the spectrum of elements.

Definition 2.2.1
Let A be an algebra with unit element (not necessarily a Banach algebra). The
spectrum of an element A E A, denoted by Spect (A), is defined as the set of
complex numbers A for which (all - A) -1 does not exist in A. The spectral
radius of A is defined as

g(A) = sup I a I, ). E Spect (A). (III.2.24)

From this definition one obtains by purely algebraic arguments, not involv-
ing any limiting processes

Theorem 2.2.2
Let A be an algebra with unit (not necessarily a Banach algebra) and A, B E A.
(i) If F(x) is a polynomial in the variable x and
A E Spect (A) then F(a) E Spect F(A). (III.2.25)
(ii) If A-1 exists and A E Spect (A) then A -1 E Spect (A -1 ).
(iii) If a E Spect (AB) and a O then a E Spect (BA).
(iv) If A is a * - algebra and
A E Spect (A) then A E Spect (A*).

The demonstration of these basic properties of the spectrum is elementary

(see e.g. [Bratteli and Robinson 1979, section 2.2]).
The spectrum depends only on the set A and the algebraic relations within
it, not on the topology of A. However, if A is a Banach algebra then

p (A) < II A II . (III.2.26)

This follows from the power series expansion

2
(all - A) -1 = a -1 (ll + + + ... (III.2.27)

which is convergent in the norm topology if II A Il< IAI.

One can obtain a precise relation between the spectral radius of A and the
norms of powers of A. The convergence of (III.2.27) is determined not by II A II
but by the behaviour of II An II for large n. Let

r = I An Il l 1 n • (III.2.28)
120 III. Algebras of Loc al Observables and Fields

One shows that

r = lim rn
n-yoo

exists and satisfies

0 <r5IIAII.
The sequence (III.2.27) converges in norm if IAI > r. On the other hand one
shows that if (All — A) -1 would also exist for all I A 1= r then this would require
II An II /r n -+ 0 in contradiction to the definition of r. 2 Thus one has

Theorem 2.2.3
In a Banach algebra with unit the spectral radius of an element A is
g (A)=r(A)^ linl,II An Illin. (III.2.31)

The theorems suggest that, given a Banach *-algebra one may equip it with
another, more natural, norm using the spectral radius of elements. On the one
hand one sees from (III.2.31) that if A is self adjoint and the norm is a C*-norm
(i.e. satisfies (I1I.2.5)) then II A and 11 An '1n do not differ, hence the norm
must be equal to the spectral radius. On the other hand one sees from theo-
rem 2.2.2. that in a Banach *-algebra generated by a single self adjoint element
the polynomials of Ao with complex coefficients form Ao(anlgebriwhc
a dense set) the spectral radius of elements satisfies the conditions (III.2.2) -
(111.2.5). The spectral radius provides a C*-seminorm on such an algebra. It
may happen that o (A) = 0 for some A i 0. This will be the case if

lim II A n II1in= O. (III.2.32)

We may call such elements generalized zero elements. They constitute a closed
ideal 3 C A. If we throw them out (identify them with the zero element) the
quotient algebra A/3 (whose elements are classes of elements in A modulo 3)
can be completed to a C*-algebra with respect to the norm II A II = Q(A). It
turns out that the elements of this C*-algebra are in one to one correspondence
with the complex valued, continuous functions on the spectrum of Ao. A spectral
value of Ao characterizes a maximal ideal in this algebra namely the set of all
functions which vanish at this point.
The last remark is relevant for the adaptation of the discussion to the case
of general commutative Banach *-algebras which leads to Gelfand's theory of
commutative C*-algebras. A closed maximal ideal in such an algebra A is called
an element of the spectrum of the algebra. It corresponds to a set of simulta-
neous spectral values of the elements of A. We denote the spectrum of A (set
of maximal ideals) by K. Gelfand has shown that a natural topology can be
introduced on 1C and that 1C is then a compact Hausdorff space. One has

'See [Bratteli and Robinson 1979, proposition 2.2.2].

 1II.2 Von Neumann Algebras. C* Algebras. W* Algebras
- - 121

Theorem 2.2.4 (Gelfand)

A commutative C*-algebra A is isomorphic to the C*-algebra of complex val-
ued, continuous functions on a compact Hausdorff space 1C, the spectrum of A.
The norm of an element is given by the supremum of the absolute value of the
corresponding function.

For a non-commutative Banach *-algebra one has

Theorem 2.2.5
1f a Banach *-algebra admits a C*-norm then this norm is uniquely determined
as

II A i= e(A) if A and A* commute,

II A II= {o(A*A)} h/2 in general.

The spectrum of an element A in a C*-algebra is already determined by its

spectrum in the smallest C*-algebra containing A.

For the proof see Bratteli and Robinson loc. cit. section 2.2.

The theorems show that the C*-condition (III.2.5) for the norm is necessary
to allow a viable spectral theory. In particular in a C*-algebra the spectrum of
an element A is the same in any C*-subalgebra containing A. This also allows
a simple characterization of positive elements in terms of the *-operation.

Definition 2.2.6
An element A E A is called positive if Spect A C IR+ U {0} (the non negative
reals). The set of positive elements of A is denoted by A.

Theorem 2.2.7
Let A be a C*-algebra. Then

(i) A + is a convex cone i.e.

with A, B E A+ and a, b E IR+one has aA + bB E A+.
(ii) A E A + if and only ifA= B*B with B E A,B#0.
For the proof see [Sakai 1971].
Positive Linear Forms and States.

Definition 2.2.8
(i) A function (p from an algebra A to the complex numbers is called a linear
form over A if

cp(aA +,ûB) = acp(A) +/3cp(B); A, B E A, a,0 E C.

(ii) If A is a Banach algebra we call a linear form cp bounded if

122 III. Algebras of Loc al Observables and Fields

I w(A)I cllAlf•
The lowest choice for the bound c is called the norm of cp,

IIcPll= Asup
EA
1 v(A) IliA II -1 • (III.2.33)

(iii) Let A be a *-algebra with unit, cp a linear form. cp is called

real if cp(A*) = cp(A),

positive if cp(A*A) > 0,
normalized if II V II == 1.
A normalized positive linear form is called a state.

Theorem 2.2.9
A positive linear form over a Banach *-algebra with unit is bounded and

II cp II = cp(Il). (III.2.34)

It satisfies the Schwarz inequality

1 co(AB) 12 < cp(AA*) cp(B*B). (III.2.35)

Remark. Note that a positive linear form over a *-algebra with unit is, a for-
teriori, real. The term state for a normalized positive linear form is, of course,
borrowed from physics and indicates the physical relevance of the concept. If we
assume that the self- adjoint elements of A correspond to observables and the
unit element to the trivial observable having the v alue 1 in any physical state
then a normalized positive linear form w may be interpreted as an expectation
functional over the observables i.e. the mathematical notion of state corresponds
to the physical one.

The Gelfand - Naimark - Segal (GNS) - Construction. The following obser-

vation, elaborated by I.M. Gelfand and M.A. Naimark and by I.E. Segal is of
fundamental importance. Each positive linear form w over a *-algebra A defines
a Hilbert space '1t, and a representation 7rw of A by linear operators acting in
I-LW .
One first notes that A itself is a linear space over the field C and that w
defines an Hermitean scalar product on A by

(AIB)=w(A*B), A, B E A. (III.2.36)

Due to the positivity of w this scalar product is semi-definite i.e.

(AJA) > 0

and by (III.2.35)
 1II.2 Von Neumann Algebras. C*-Algebras. W*-Algebras 123

I (B I A) 12 << (BI B)(AIA).

Therefore the set 3 C A consisting of all elements X E A with w(X*X) = 0 is
a left ideal i.e. a linear subspace of A which is stable under multiplication by
an arbitrary element of A from the left:

X E 3, A EA implies AX E 3. (III.2.37)

We shall call 3 the Gelfand ideal of the state. The set of classes A/3 is thus a
pre-Hilbert space, a linear space equipped with an Hermitean, positive definite
scalar product. A vector W in this space corresponds to an equivalence class
modulo 3 of algebraic elements

W={A+ 3}, - A E A.

We denote the class of the element A by [A]. The scalar product (III.2.36) does
not depend on the choice of the algebraic element within one class and, if A is
not in the class of the zero element

(wIw) =II w II 2 > 0. (III.2.38)

7-L is obtained by the completion of A/3 in the norm topology given by

(III.2.38). The product in A defines an action of A on the vectors in A/3; it
associates to each AEA a linear operator 7ru,(A), defined on the dense domain
A/3 c 7-1, by
71-u,(A)W = [AB] if W = [B]. (III.2.39)
If A is a Banach *-algebra then the domain of definition of 7r,(A) can be ex-
tended to the whole of 7- W by continuity.

Theorem 2.2.10
If A is a C*-algebra with unit and w any state such that 7r, is a faithful rep-
resentation (i.e. 7r u,(A) 0 for A 0) then the operator norm of 7r u,(A) equals
the C*-norm of A. (Uniqueness of the C*-norm).

A representation 7r is called cyclic if there exists a vector S? in the represen-

tation space 7-L such that 7r(A).fl is dense in 7-. S? is called a cyclic vector. It is
clear that, if A has a unit, the GNS-construction provides a cyclic representation
with the cyclic vector .fl corresponding to the class of the unit element

f2 — [Il].

Moreover
w(A) = (0I7rw(A) I f2). (III.2.40)
So the expectation value is expressed in a form familiar from quantum mechan-
ics. We may say that the state w is represented by the state vector 12 in 7 -L.
Any vector W E 7-u defines a state
124 III. Algebras of Local Observables and Fields

w,k( A) = (wI rw( A ) Iw), (I1I.2.41)

which may be approximated by w(B*AB) with B E A since W can be approxi-

mated by irw (B)Q.
The GNS-construction shows that states over an abstract *-algebra come in
families. One state w determines a family of states, the states of form (I1I.2.41)
with W E 7-L,. We call these the vector states of the representation irw . More
generally we may consider the states

we (A) = tr pr-,(A) (I1I.2.42)

with o a positive trace class operator in /3(7-G). The set of all the states
(11I.2.42) we call the folium of the representation 7r, or the set of normal states
of the von Neumann algebra ir,(A)". It is, of course, determined by w. Clearly
one may now take a state w' in this folium and make again the GNS-construction
starting from w ' . Then w' which was represented as a density matrix (positive
trace class operator) in 7, is now represented as a vector in 7 -h; . To understand
the relation between these different ways of representing the algebra and the
states we have to take a closer look at the dual space of A.
We begin with some generalities. Let V be a Banach space over C. The dual
space, denoted by V*, is the set of all bounded, complex valued linear forms on
V . V* is again a Banach space with respect to the natural norm

I4 = sup I 92(X) I 11 X Il
XEV
, ^ E V* . (III.2.43)

Besides the norm topology there is another important topology on V*. Choosing
any finite set X 1 , ... X n of elements of V one can define a seminorm on V* by

lcp ^ ^x,...x^ = sup I ço(Xk) I, (III.2.44)

k=1...n
and a corresponding neighborhood N(E; X1 , ... Xn ) of the origin in V* by

N(E; X 1 , ... Xn) _ cp E V * = ?Z cp 2 ^xl ...xn < E}

{
,

={cp EV*:I cp(X k }1<e for k=1,...n}.

Such neighborhoods for arbitrary finite n and arbitrary choice of the ele-
ments Xk define the weak topology in V* induced by V (also called the weak
*-topology). In particular, a generalized sequence (p c, converges weakly to cp if

Va(X) — ço(X) I-- ^^ 4 for every X EV. (II1.2.45)

The following theorem will be used on several occasions.

Theorem 2.2.11 (Alaoglu)

The closed unit ball in V* (i.e. the set Iv E V* : II cp It < 1} is compact in the
weak topology induced by V.
See [Dunford and Schwartz 1958].
 III.2 Von Neumann Algebras. C*-Algebras. W*-Algebras 125

We are interested first in the case where V is a C*-algebra A. The dual

space A* 3 has a distinguished convex cone A*+, the set of positive linear forms
on A. The states over A are the elements of A*+ with norm 1. One has (see
[Sakai 1971])

Theorem 2.2.12
The folium of a representation and the set of vector states of a representation
are norm closed subsets of A*+.

On the other hand one has

Theorem 2.2.13 [Fell 60]

The folium of a faithful representation of a C*-algebra is weakly dense in the
set of all states.

Remark. Since in physics we can never perform infinitely many experiments

and since each experiment has a limited accuracy we can, by monitoring a
state, never determine more than a weak neighborhood in A*+. By Fell's theo-
rem this means that we cannot find out in which folium the state lies. Still there
are reasons for considering more than one equivalence class of representations
of A. They may come from idealizations of the problem (injecting an infinite
amount of information which we do not really have in order to simplify the
problem). Examples will be seen in chapters IV and V. And there may also be
reasons for excluding some folia as "physically not realizable" (see the remarks
at the end of this section and the assumption of local definiteness in section 3).

Definition 2.2.14
Given two representations 7r 1 , 7r2 acting in the Hilbert spaces 7-ll , 7-12 respec-
tively. A bounded linear operator R from 7-1 1 to 7-2 is called an intertwiner
(from 7r1 to 7r2 ) if

R 7r1 (A) = 7r2 (A) R for all A E A. (I11.2.46)

Its adjoint R* is uniquely defined as a bounded linear operator from 7 -l2 to 7-

by
(W1IR * IW2) = (RIN1W2), Wk E ?1k, k = 1, 2. (III.2.47)
It is an intertwiner in the opposite direction

R *7r2 (A) = 7r1 (A)R*. (III.2.48)

An intertwiner decomposes each of the two representation spaces into two in-
variant orthogonal subpaces 74 = 7H'k ® 7-k, k = 1, 2. I-4 is the kernel of R,
3 Some confusion may arise because the symbol C7' (with C7 c A) is sometimes used to
denote the set of all the adjoints of the elements of C7. Hopefully the different meaning of the
star will always be evident from the context.
126 III. Algebras of Local Observables and Fields

the set of vectors in 7i 1 which is annihilated by R; NZ is the closure of the

range of R (set of image vectors). The restriction of R to 1-li is a non-singular
linear operator from 7-1i onto 7-6 which intertwines from the restriction of 7r 1 to
Nil to the restriction of 7r 2 to 7-12. This means that these two subrepresentations
are unitarily equivalent because one can make a polar decomposition

R = V(R*R) 112 ;

V is a unitary map from 7-i onto 7-6 which also intertwines the subrepresenta-
tions since R*R is in the commutant of 7r 1 .

Definition 2.2.15
Two representations are called disjoint if they contain no subrepresentations
which are unitarily equivalent i.e. if the set of (non zero) intertwiners is empty.
Two representations are called quasi equivalent if every subrepresentation of the
first contains a representation which is unitarily equivalent to a subrepresenta-
tion of the second. A representation is called primary if it is quasi-equivalent to
each of its subrepresentations. A state is called primary if it leads to a primary
representation by the GNS-construction.

Note that the set of intertwiners of a representation 71 - with itself is a

von Neumann algebra namely the commutant ir (A)'. The notion of quasi-
equivalence defined here agrees with the one given earlier for von Neumann
algebras. A primary representation is one in which the center 7r(A) / n ir (A)" is
trivial i.e. one in which ir(A)" is a factor.
We mention still

Theorem 2.2.16
The norm distance between any two states belonging to disjoint representations
is 2 (the same as between two density matrices with supports in two orthogonal
subspaces of a Hilbert space).

Purification. For two states w 1 ,w2 and 0 < a < 1

w=awl+(1—A)w2 (1II.2.49)

is again a state, the mixture of w 1 and w2 with weights a and 1 — A. If w can

be written in the form (III.2.49) then we say that w dominates w1 (and w2 ).
Excluding the trivial case w1 = w2 we also call w1 a purification of w One sees
readily that if w dominates wi then the GNS-representation 7r, 1 is unitarily
equivalent to a subrepresentation of 7rw . The intertwiner is given by

R iru,(A) ,f2 = 7rwl (A) Q 1 = 9rw1 (A)R Q. (III.2.50)

This is consistent because w(A*A) = 0 implies wl(A*A) = 0 and it suffices to

define R because Q is a cyclic vector. This leads to
 III.2 Von Neumann Algebras. C*-Algebras. W*-Algebras 127

Theorem 2.2.17
(i) The folium of a representation uniquely determines its quasi-equivalence
class. If w dominates w1 then ir gl is contained in
(ii) The GNS-representation arising from a pure state is irreducible and the
vector states in an irreducible representation are pure.
(iii) Every state in the folium of a primary representation is primary.

The theorem shows that the distinction between vector states and density
matrices has an intrinsic significance only in the case of irreducible representa-
tions where it corresponds to the distinction between pure and impure states.
There remains the question of the existence of "sufficiently many" pure states.
This is answered by

Theorem 2.2.18
For a C*-algebra with unit the set of its states is weakly compact. It has ex-
tremal points, the pure states, and the whole set of states is generated from the
pure states by convex combinations and weak closure.

Remark. The compactness follows from theorem 2.2.11 noting that cp(1) = 1.
If A has no unit it may happen that sequences of states converge weakly
to zero. A typical example, not entirely void of interest for physics, is the
following. Let H = £( 2)(IR) and U(a) be unitary translation operators i.e.
(U(a)')(x) = i(x — a); x, a E IR. Let K be a Hilbert-Schmidt operator,
i.e.
(KIP)(x) = f .K(x, x')(x')dx'

where the integral kernel is a square integrable function of x and x'. Obviously,
for a sequence s = U(a)Ii we have
lifI(2NI Ma ) = 0 (III.2.51)
The norm closure of the set of Hilbert-Schmidt operators is a C*-subalgebra of
930`0 without unit. It is the set of compact operators, a closed 2-sided *-ideal
.3c in 93(1-0. By (111.2.51) the sequence of vector states O a , restricted to 3c,
converges weakly to zero as a — ^^ oo. On the other hand, since (Oa I11Pa ) = 1
the sequence of states over B(H) must have limit points in TN* which are
states on 93(1`l). These states are, however, not normal states of the defining
representation of B(H).

W*-Algebras. A von Neumann algebra also has an abstract counterpart, a

W*-algebra. This is a C*-algebra which, in addition, is the dual space of a
Banach space V over the field of complex numbers. From the point of view
of our application to physics W*-algebras arise naturally if we consider state-
space and its structure rather than the algebra as the primary object. V is
then interpreted as the complex linear span of the space of states (expec-
tation functionals, not state vectors). It is a Banach space with a distin-
guished positive cone V+ of which the elements with norm 1 are the states.
128 III. Algebras of Local Observables and Fields

Its dual space V* (the Banach space of bounded linear forms over V) is then
equipped with a norm topology and the weak topology induced by V as de-
scribed above. By theorem 2.2.11 it is also weakly closed. If a product can
be defined within V* satisfying (II1.2.4), (III.2.5) so that V* becomes a
C*-algebra, with V*+ as its positive cone, then this algebra, equipped with
the weak topology inherited from V is called a W*-algebra. When should one
be able to introduce such a product in V*? This is encoded in the facial struc-
ture of V+ (see Chapter VII). The papers quoted there give a partial answer.
If we write V* = A then V is called the predual of A (which is another way of
saying that A is the dual of V). The predual is denoted by a lower star: one
writes V = A. The set of all bounded linear forms over A, the dual space ,A*,
will be larger than V = A. Thus a W*-algebra has a distinguished folium of
states, the normalized elements of A,4+, . They are called the normal states. If
we use one of the normal states of a W*-algebra in the GNS-construction of a
representation we obtain as the image a weakly closed operator algebra i.e. a
von Neumann ring.
Thus one might also say that a W*-algebra arises from a C*-algebra to-
gether with a distinguished folium of states. It is isomorphic to the closure of
the algebra in the weak topology determined by this folium. The distinction of
a specific folium may be relevant in physics since it is not clear whether one
can or should define the net of C*-algebras of local observables in such a way
that all mathematical states are physically realizable. An example of physically
unrealizable states would be states carrying infinite energy in a finite region.

References. The basic theory of von Neumann algebras was developed be-
tween 1936 and 1943 in a series of papers by F.J. Murray and J. von Neu-
mann in the Annals of Mathematics. C*-algebras originated in the Moscow
school of I.E. Gelfand in the early 40's (see the still highly recommendable
book [Nairnark 1972]). An independent approach was pioneered by I.E. Segal
[Seg 47].
Standard mathematical texts covering this area are [Dixmier 1981], [Dix-
mier 1982], [Sakai 1971], [Pedersen 1979], [Kadison and Ringrose 1983, 1986].
A rather detailed exposition addressed to physicists is contained in [Brat-
teli and Robinson, Vol I, 1979].

111.3 The Net of Algebras of Observables

With the help of the mathematical concepts described in the last section we can
elaborate the frame of local quantum theory in Minkowski space. In this section
and in the next one we shall look at some specific structural properties which
the net of algebras of local observables should (or might) have and describe
some consequences. One objective is to place axiom A into a broader context.
In the endeavour of formulating the assumptions concisely one has to balance
between the need for keeping them general enough to encompass the physically
 III.3 The Net of Algebras of Observables 129

interesting situations and the wish to make them as restrictive as possible. It

should be borne in mind that the theory is not in a closed, final stage; hence
there is some leeway.

III.3.1 Smoothness and Integration. Local Definiteness

and Local Normality

In chapter II we frequently used integrals like f f (x) a x A d 4x with f a smooth

function with fast decrease and a x the automorphism corresponding to a trans-
lation by x. We have to see how such integrals can be defined. Let A E 21.
Since II axA II is independent of x and the norm has the subadditivity property
(II1.2.3) the only question concerns the local integrability. If one has this then

a f A - ff(x)ciAd 4x (11I.3.1)

is defined for continuous £ 1 -functions f as a Bochner integral (the generalization

of the Riemann integral to integrands with values in a Banach space) and one
has
IIafAII<IIAIIII f Ili, (11I.3.2)
where II f II i = f I f (x) I d 4x is the G 1 -norm of f . One has local integrability
(in the sense of the Bochner integral) certainly for those elements A E 21 for
which the function x --0 axA is norm-continuous.

Definition 3.1.1
An element A E 21 is called differentiable if

)C -1 II a ax A — A II converges as a ---> 0. (II1.3.3)

The differential quotients â/âxt`a x Al x=o are then defined and belong again to
21. The element A is called smooth if it is infinitely often differentiable.

We argue now that it is possible and warranted to choose the algebras so that
the action of the translation automorphisms on the elements is continuous in
the norm topology. We shall denote such a choice of the algebras by 21 s .

Assumption 3.1.2 (Smoothness)

We can choose for the algebra of local observables in O a C*-algebra 21s(0)
containing a norm-dense set of smooth elements. Then a xA is continuous in the
norm topology i.e.

II axA — All -- 0 as z -- ^ 0 for any AE2(s. (1II.3.4)

Let us present some arguments supporting this assumption.
Suppose we start from a net {21(0)} not satisfying this assumption. Denote
the representation obtained by the GNS-construction from the vacuum state by
no and take the von Neumann rings
130 III. Algebras of Local Observables and Fields

R(0) = 7ro M(0))"; R. = 7r0(%)". (III.3.5)

Then ax is implementable by a strongly continuous group of unitary operators

U(x) and the energy-momentum operators PI-4 are defined in the representation.
Let xE = PEW be the spectral subspace of P ° corresponding to spectral values
below E and PE the spectral projector. Then matrix elements

( w' I U(x)X U-1 ( x ) I w ) (111.3.6)

are differentiable functions of x for any X E R. if W and belong to xE

because PEPP = PPPE is bounded. For arbitrary vectors W, W' (1II.3.6) is
still continuous since the vectors can be strongly approximated by vectors from
'HE as E —+ oo. Let f(x) be a smooth function with support contained in a
neighborhood 00 of the origin and let A E 21(0 1 ). Then the integral

7r(A f ) a f f (x) 7r(a x A) d 4 x (111.3.7)

is defined in the sense of matrix elements yielding an operator in 1Z(0 1 + Cho)

with norm
I 7r(Af)II<IIAII ii
II1. (11I.3.8)
Furthermore 7r(A f ) is infinitely often differentiable in the norm topology because
the differentiations can be shifted to the smooth function f . Since any open
region 0 can be written as 01 + 00 with suitably chosen, open 0 1 , Cho we see
that there are smooth elements in every 1Z(0) and we may define a (concrete)
C*-algebra 21,s (0) as consisting of smooth elements of R.(0) and their limits in
the norm topology.
In this construction the folium of states in the vacuum representation
was used. What happens if we consider other folia of states? 1f we adhere
strictly to the dogma that all physical information is contained in a net of
abstract C*-algebras then we must regard every mathematical state (normal-
ized positive, linear functional on the net) as a physically realizable state. There
are, however, good reasons for the following principle which suggests that we
should soften the dogma and distinguish a subset of the mathematical states as
the physically realizable states.

Principle 3.1.3 (Local Definiteness)

Let Or, be a directed set of regions shrinking to a point x, i.e.
On+ 1 C 0,6 n on = x. Any two physically realizable states cw ; (i = 1, 2)
will become indistinguishable in restriction to 21(O„,) in the limit n --4 oo.
Specifically
II (w1 -- cv2)I21(0 n ) II--^^ 0 as n —+ oo. (1II.3.9)

This principle is natural if we think of defining the theory in terms of relations

in the small. There should remain no superselection rules if we restrict atten-
tion to sufficiently small regions of space-time. The distinction of two states by
 III.3 The Net of Algebras of Observables 131

means of observations in a very small region requires very high energies and
there remain no observables at a point.' If we believe, in additition, that among
the realizable states we have states which are primary for the algebras of small
regions then, appealing to theorem 2.2.16 we conclude from the principle 3.1.3
that for any two folia of physically allowed states and any point x E M there
will be a neighborhood of x such that the restrictions of the folia to the algebra
of this neighborhood coincide and are primary.

Remark. We motivated the use of abstract algebras by the wish of understand-

ing superselection rules in an intrinsic way, namely as arising from inequivalent
representations of one and the same algebraic structure. This is indeed one good
reason. However, the superselection rules encountered so far (various types of
total charges, thermodynamic-hydrodynamic quantities of the infinite system)
all relate to global aspects of the states and are not effective once we restrict
attention to a finite region. This fits in with the above conclusion and suggests
the following working hypothesis.

Tentative Assumption 3.1.4 (Local Normality)

In restriction to the algebra of a finite, contractible region all physically realiz-
able states belong to a unique primary folium.

This makes the supporting argument for the smoothness assumption 3.1.2
independent of the use of the vacuum representation. We have then the follow-
ing picture. There is the net 2l9(0) on which the automorphisms of translations
act continously and to which we shall refer the notions of states and representa-
tions. On the other side we have the set 6 of physically realizable states. Their
restriction to an algebra 21s(0) will give the set of partial states on 0, denoted
by 6(0). For 01 C 0 there is the natural restriction map

6(n) --4
6((91). (1II.3.10)

It defines an equivalence class in 6(0) with respect to 01, consisting of all

partial states on 0 which have the same restriction to 0 1 . So, in mathematical
terms, the collection {65(0)} is a presheaf. Let E(0) denote the Banach space
over C generated by 63(0). It is the complex linear span of 6(0) completed in
the norm topology inherited from the norm in 6(0) (in precise terms E is a
"base norm space" ). The dual space of E(0) is a W"-algebra which we denote
by 7Z.(0)
7Z(0) = E* (0); E(0) = 7Z(0)* . (111.3.1 1)
For finite, contractible regions R.(0) is isomorphic to the corresponding von
Neumann algebra in the vacuum sector for which we shall use the same symbol.
'The relevance of this principle was recognized in the context of quantum field theory in
curved space-time and used in the discussion of the Hawking temperature by Haag, Narn-
hofer and Stein [Haag 84]. If the metric structure of space-time is not (classically) given but
determined by the prevailing physical state then the possibility of regarding the metric at
each point as a superselection rule may be of interest; see [Fred 87] and [Bann 88].
132 III. Algebras of Local Observables and Fields

The C*-algebra 2ts(0) is weakly dense in R(0). The net {2ts (Q)} is not needed
in the discussion of superselection rules in the next chapters. What is impor-
tant there is that the global algebra 2t must be considered as a C*-algebra, not
a W*-algebra. It is the completion of the union of local algebras in the norm
topology in order to preserve the quasilocal character, irrespective of whether
we take the local algebras as W*- or C*-algebras. Nevertheless a physically dis-
tinguished C*-net {2ts(0)} carrying more information than the net {R(Q)}
may be expected to become important in a finer analysis, building up the the-
ory from information in the small in analogy to the methods of differential
geometry. An example may be the (not yet achieved) understanding of a local
gauge principle directly in the quantum theoretic setting. For the definition of a
physical distinguished net Os one needs, of course, more than just the demand
that the translation automorphisms act continuously on it. It should satisfy for
instance some minimality requirement.

III.3.2 Symmetries and Symmetry Breaking. Vacuum States

Definition 3.2.1
A group Ç is called a symmetry group if there is a realization of C by automor-
phisms of the net:
gEÇ--+ a9 EAut2t (I11.3.12)
with the additional requirement that the image of the algebra of a finite region
under a9 shall be again the algebra of a finite region.

If w is an invariant state, i.e. a normalized, positive linear functional over

21. with
w(a9 A) = w(A) for g E G, A E 0. (I1I.3.13)
then in the GNS-representation built from w the automorphisms ag are cer-
tainly implementable. Denoting by (Q, 7r, i`l) respectively the cyclic vector
corresponding to w, the representation and the Hilbert space on which it acts,
the definition
U(g)r(A)S? = ir(a 9 A),f2 (1II.3.14)
provides a unitary representation g — ^^ U(g) of the group. 2
Each group element has a "geometric part", a point transformation of
Minkowski space conserving the causal structure. This is because we required
that ag transforms the algebra of a finite region into that of another finite re-
gion and a 9 respects inclusion properties of regions and causal structure since

2 One has to check that the definition is consistent i.e. if r(A)f2 = 0 then also ir(a9 A)Q = O.
This follows from the isometry

II U(g)ir(A)f2 II 2 = (Ql ir(a9A)'ir(a9A)V2) = w(agA * A) = w(A"A) = II r(A)S2 112 .

Since S2 is a cyclic vector for r(A) (III.3.14) defines U(g) on a dense domain. The image is
again a dense domain because ag is invertible. So U(g) is extendable to a unitary operator
on 7-1,,,. One easily checks the representation property U(g2)U(gi) = U(g2g1).
 III.3 The Net of Algebras of Observables 133

these manifest themselves in the algebraic structure of the net. G is therefore

a semidirect product of a purely geometric symmetry group and an "internal
symmetry group" which transforms the algebra of each region into itself:

%2[(0) = 240) for g E Jint• (111.3.15)

Since the geometric symmetry group must conserve the causal structure
it could be at most the conformal group. We shall mostly take it to be the
Poincaré group. In any case the geometric symmetry group is locally com-
pact and has an invariant measure dµ(g). We assume that the same is true
for the full symmetry group. Then we can construct g-invariant states in the
following way. Consider an increasing sequence of compact subsets: Sk C Ç,
Sk+1 D Sk, USk = G. Starting from an arbitrary state w we define co, as 3

W k ( A) — µ (Sk) - 1 (agA)d^(g )• (111.3.16)

f W

Obviously Wk is again a state since w(11) = 1 and, using the invariance of the
measure one gets

wk(ag A) = µ( Sk) -1 f k Wk( a 91 9 A) d/(g1 ) = µ(Sk)-1 fSkg912o)

where Sk g is the set resulting from Sk by right translation with g. One has

I wk(agA) — wk(A) I < II A II 1i(Sk)-1 (p(LX) + µ( 4 01

where
,Sk = Sk - (Skg n Sk), dk = Skg — (Skg n Sk)•

By theorem 2.2.18 the set of all states over 21 is weakly compact. Thus the
sequence co, has weak limit points which are states. For the subgroups formed
by space-time translations together with any compact group (but not for the
Poincaré group!) the ratios µ(4k)/µ(S k ) go to zero for fixed g and k ---0 oo. So
one can construct states which are invariant under translations and rotations
by this procedure.
The set of invariant states is evidently a convex subset of state space (mix-
tures of invariant states are invariant); it is weakly closed and therefore weakly
compact (the set of all states being weakly compact). Therefore, by the theorem
of Krein and Milman (see [Dunford and Schwartz 1g58]) all invariant states can
be obtained as convex combinations (m ixtures) of extremal invariant states, i.e.
invariant states which cannot be decomposed any more into a convex combina-
tion of other invariant states. There is, however, no reason yet why an extremal
invariant state should be pure or primary. It may be possible to decompose it
into a mixture of other states which are not invariant. This is the scenario of
spontaneous symmetry breaking.
3 We assume here that ag acts continuously on 2t for the whole symmetry group, not only
for the translation subgroup.
134 III. Algebras of Local Observables and Fields

The spontaneous breakdown of symmetries is a well known phenomenon

in solid state physics or, more generally, in the nonrelativistic quantum theory
of the structure of matter, idealized as an infinitely extended medium with a
finite mean density of particles. Translation invariance may be broken by crys-
talization, rotation invariance by spontaneous magnetization. Galilei invariance
(replacing Lorentz invariance) is always broken if the mean particle density is
finite because a state describing the matter at rest in the reference system and
one describing a flow of matter with uniform velocity are "infinitely different"
and thus lie certainly in disjoint folia.
The mathematical counterpart of these phenomena is the following. Let w
be an extremal invariant state with respect to the symmetry group and 7r,,,
the GNS-representation arising from it. If w is not primary then the center
3 = 7r,,(21)" n 7ru,(21)' is not trivial. We may regard a self adjoint operator
Z E 3 as an idealized observable which commutes with all local observables
(it is typically a question involving the behavior of the state at infinity). Thus
Z may be considered as a classical quantity. If we do not know its value this
may be considered as subjective ignorance. If we decompose the Hilbert space
according to the simultaneous spectrum of the Abelian von Neumann algebra 3

?`t„ = f ?- 4,,dµ(K) (III.3.17)

we get a unique decomposition of w as a convex combination of primary states

W,c (K E IC, the spectrum of 3). 4 Each such primary component may then be
regarded as an optimal objective description of the state of an individual system.
In the case of crystalization the relevant elements of the center are of the form

Z = lim V--1 f ax A f (x) d3x (III.3.18)

v---.00v
where f is a periodic function. In the case of magnetization it is of the same
form with f = 1 and A a component of a vectorial observable (the magnetic
moment per unit volume).
The physical significance of the central decomposition is emphasized by the
following theorem.

Theorem 3.2.2
A primary state w has the cluster property

I w(Acxx B) — w(A)w(a x B) I--> 0 as I xI-- ^ oo. (III.3.19)

Proof. The von Neumann algebra ir„(21)" is the dual of a Banach space. There-
fore the ball ii X li < c, X E ir„(2t)" is weakly compact. A sequence ir„(a x B)
with I x I—^^ co has weak limit points in ir„(2t)". For large space translations the
4 The measure theoretic intricacies of this "central decomposition" and of the decomposi-
tion of invariant states into extremal invariant states are discussed in [Bratteli and Robinson
197g, chapter 4 of Vol. 1.]
 III.3 The Net of Algebras of Observables 135

commutator of a.B with any fixed element of the quasilocal algebra tends to
zero. Therefore the limit elements of ir,,(a.B) lie also in the commutant 7r„(2t)'.
So they lie in the center. If w is primary the center consists only of multiples of
the unit operator. Thus

ir,,(a.B) — w(a,d3)11 —+ 0 (weakly)

which implies (III.3.19).

Let us illustrate this in the example of spontaneous magnetization. A pri-
mary state describes a sharp direction of magnetization. If the observable M(x)
denotes the magnetic moment per unit volume around the point x and w is
primary then w(M(x)) = m(x) 0 but the correlation w (M(x)M(y)) —
m(x)m(y) tends to zero for large I x --- y I. If we take instead the extremal
invariant state w which results from w, as in (1II.3.16), by averaging over the
rotation group then the 1-point function w(M(x)) vanishes and the 2-point
function has infinitely long range correlations.
Let us turn now to the notion of a vacuum state in relativistic, local theory.
There is a strong analogy between quantum field theory and the treatment of
an infinitely extended medium in nonrelativistic quantum theory. In the latter
case we know, however, a priori what we mean by matter density and vacuum
in terms of the mathematical objects from which the theory is constructed. In
our case, where we define the theory by a net of algebras of local observables the
meaning of "absence of matter" is not immediately evident. We can start from
lemma 4.1.2 in chapter II which tells us that for any A E 0. and any function
f (x) whose Fourier transform has support in the momentum space region L
the algebraic element
A(f) = ff(x)aAd 4 x
affects an energy-momentum transfer within L . This suggests that the positive
observables A(f)*A(f) with supp f (p) confined to negative values of the energy
p° register the presence of matter, or, taking Lorentz invariance into account,
that such observables are detectors of matter whenever supp f (p) lies outside
the forward cone V+. In the context of the GNS-construction we saw that a
state w defines a left ideal in the algebra, its Gelfand ideal or annihilator ideal,
consisting of all elements X E 0. for which w(X*X) = O. Conversely, given any
proper, closed left ideal 3 there exists at least one state which has this ideal as
its Gelfand ideal. This fact was used by Doplicher [Dopl 65] to give an algebraic
formulation of axiom A3. We associate with each region 4 of momentum space
the left ideal of 21

3.(4)= {X =BA(f)} with BE2[, A E 21, suppf C (III.3.20)

Definition 3.2.3
A state whose Gelfand ideal (annihilator ideal) contains the "spectral ideal"

3(V+c) = 3(ud), L n V+ = 0 (III.3.21)

136 III. Algebras of Local Observables and Fields

(the union is taken over all momentum space regions which do not intersect the
forward cone) is called a vacuum state.

The existence of a vacuum corresponds to the algebraic property that a(V+c)

is a proper left ideal i.e. that its closure is not the whole algebra 2l.
Obviously so defined vacuum states form a Poincaré invariant set. So one can
obtain translation invariant vacuum states by averaging. The GNS-construction
from a translation invariant vacuum satisfies axiom A3 of chapter II.
One can show now that translation invariance cannot be spontaneously
broken in a vacuum state: the vacuum cannot be a crystal. Furthermore, an
extremal invariant vacuum state is automatically pure. One has

Theorem 3.2.4 [Ara 64b]

Let ?Z be a v. Neumann algebra and U(x) a strongly continuous unitary rep-
resentation of the space-time translation group such that the spectrum of the
generators is contained in the forward cone and

axA = U(x)AU(x) -1 , A E R.
transforms R. into itself. Then
(i) each element of the center 3 - R. n R.' commutes with all U(x).
(ii) If the Hilbert space contains a cyclic vector Q which is invariant under
U(x) then U(x) E R. .

Proof. For any X E 93(x) and any class S-function f (x) we use the ab-
breviations X(x) = U(x)X U(x) -1 , X(f) = f f(x)X(x)d 4x. If Z E 3 then
Z* E 3, Z(f) E 3. We show that for any vector W from a dense set in 7-1, any
Z E 3 and any f such that its Fourier transform f has support in a region LI of
momentum space which excludes the origin, Z(f )W = O. This implies claim (i) of
the theorem, because then, for any 0 E 7-1 and W in this dense set (0I Z(x) W) is
a bounded function whose Fourier transform is a Laurent Schwartz distribution
with point support at p = O. Hence it is a constant and we have

Z(x) = Z, (III.3.22)
which is part (i) of the theorem.
To show that Z(f )W = 0 under the stated conditions we note that for central
elements ZnW = 0 implies ZW = 0 and Z*W = 0 because Z*fZn = (Z*Z)n =
(ZZ*)n and Z*Z is a positive operator. Next, if d o is some compact subset of
V+ then for any p E M except the origin there is a neighborhood 4 such that
for sufficiently large n either the region

4a+nd- {p=q +pl +p2+••• pn:gE Llo, pi ELi for i =1, ...n}
or the region Ll o -- nLI is disjoint from V. Thus for vectors W with spectral
support in 40 either Z(f )nil/ or Z*(f)nW vanish for large n and supp f C LI
 III.3 The Net of Algebras of Observables 137

and therefore Z(f)W = O. Any f with support disjoint from the origin is a sum
of functions to which the above argument applies.
To prove part (ii) we observe that with S E R' and A E R

(QI S(x)AI f2) = (f2IsU(—x)42) = (f2IAS(x)If2) = (9IAU(x)s1f2).

Due to the spectral properties of U(x) the Fourier transform of any matrix
element of U(x) has support in the forward cone. Thus the Fourier transform
of the above function has support in the intersection of the forward and the
backward cone, hence only at the point p = 0 i.e. the function is constant. We
can write this

(W I S(x) I,f2) _ (WISI(2) where W = A*Q.

Since Q is cyclic for R this means S(x)f2 = Sf2 and ultimately S(x) = S. So
S commutes with the translation operators i.e. U(x) E (R' )' = R. ❑
Part (i) of the theorem implies that the central decomposition of a transla-
tion invariant vacuum state respects translation invariance. It leads to transla-
tion invariant primary vacua. Thus there is no breaking of translation invariance
possible in a vacuum state.
From theorem 3.2.2 one obtains immediately

Lemma 3.2.5
Let w be a primary translation invariant state, Q the corresponding cyclic vec-
tor in the GNS-representation 7r,,. Then there is no discrete eigenvector of the
spatial momentum operators P in 7-Lu, apart from the multiples of Q.

Proof. From (III.3.19) together with the invariance of w and the asymptotic
commutativity of o A with fixed elements B, C E 2t5 we get

w(B(c A)C) —* w(BC)w(A).

Since Q is a cyclic vector in the GNS-representation this means that ir w (a„A)

converges weakly to w(A)11 for large space-like translations. Assume a is a
discrete eigenvector of P to eigenvalue p (possibly 0). Then

(0 I 7w(a.A)IQ) = eiPx(S2 1 Iirw (A)I.f2).

If Q' c Q we may take Q' orthogonal to Q. Then the left hand side vanishes
for all A in the limit I x I—, oo, the right hand side does not since Q is cyclic. ❑

From part (ii) of theorem 3.2.4 and lemma 3.2.5 we get

Theorem 3.2.6
A primary vacuum state is pure.

Proof. Any operator from the commutant of 7r w (2t) will transform Q into an-
other eigenvector of Pµ to eigenvalue zero due to (ii) of theorem 3.2.4. Thus
138 III. Algebras of Local Observables and Fields

by lemma 3.2.5 the commutant must transform the ray of fl into itself. The
cyclicity of Q implies then that the commutant can only contain multiples of
the unit operator. ❑

This leaves the question as to whether Lorentz symmetry or internal sym-

metries may be spontaneously broken in vacuum states. For A E 2 the group
relations give
aAax = ax, aA ; x = Ax. (III.3.23)
For g E g;,,t , if the translations commute with the internal symmetries, 5
ag ar = ax ag . (III.3.24)

We note that spontaneous symmetry breaking in a vacuum sector is syn-

onymous with non-invariance of the vacuum state. If the vacuum is invariant
then the automorphism is implementable (equ. (11I.3.14)). If the vacuum is not
invariant the automorphism cannot be implemented according to lemma 3.2.5
together with (III.3.23), (III.3.24).
Consider now the continuous part of g. An element of the Lie algebra gener-
ates a 1-parameter subgroup. The corresponding automorphisms will be denoted
by co,. A E ]R is the parameter, and A = 0 corresponds to the identity. Assume
that there is a dense set 0(0) C 21(0) (norm dense in 2îs(0), weakly dense in
1,(0)) for whose elements a),A are smooth. Then

= 6A, A E U0(0) (III.3.25)

A= 0
defines a derivation 6 on the dense set Z. Translation invariance of the vacuum
and relations (III.3.23) or (III.3.24) give

wo (6ax A) = w o (6A). (III.3.26)

The condition for implementability is

wo (6A) = 0, A E? (III.3.27)

It has been recognized that spontaneous symmetry breaking is usually con-

nected with the existence of zero mass particles ("Goldstone's theorem") [Gold
611. Here we give the following variant of this.
5 This is the "normal situation" . An action of the translations on the internal symmetry
group of the observables can be ruled out in a massive theory if there is sufficient interaction.
See e.g. Coleman and Mandula [Col 671. In a fr ee field theory any Weyl operator W(f) = eb 0( f
gives an internal symmetry for a fr ee bose field q since

afA- W(f)AW*(f)

defines an automorphism transforming each 21(0) into itself because

a f0(g) = 0(g) + c(f,9)Il.

Local gauge transformations and supersymmetries give other examples of an action by trans-
lations on internal symmetries.
 III.3 The Net of Algebras of Observables 139

Theorem 3.2.7
If the vacuum is separated by an energy gap m 0 from the other states in its
folium and if there is a uniform bound

I wo( SA ) I « (R) (II AQ II +IIA* Q II) , A E V(OR), (III.3.28)

within the diamond OR with base radius R centered at the origin such that

lim R-n 'P(R) — 0 as R — oo for some n > 0, (III.3.29)

then one has (III.3.27) i.e. the symmetry is unbroken.

If the symmetry is related to a conserved current then the property

(III.3.28), (I11.3.29) follows. In that case one has, due to locality

SA = i [QR, Al for A E Z(OR)

f
with
QR = B(y)f (y)d4y
where p(x) denotes the zero component of the current density, f (y) = 1 in a
small time slice covering the base of OR and f (y) = 0 outside of some slightly
larger region. This yields

Iwo (SA)I 11 2RII(II AQII+II A*r2 II); A Ev(oR) (III.3.32)

with QE = QRQ,

II QR 11 2 = f(QIQ(x)Q(Y) I Q)7(x)f (y) d 4 xd4 y.

We may choose (Ql e(x)f SZ) = O. The 2-point function decreases fast for increas-
ing space-like separation of the points if the theory has a lowest mass m > 0 so
that one obtains
R3/2 for large R. (1II.3.33)
II QR II C
Thus one has (III.3.28), (III.3.29) for any n > 3/2. This observation was the
starting point of the discussion of the Goldstone theorem by Kastler, Robinson
and Swieca [Kart 66].
To prove the theorem we can now proceed as follows. Specializing in
(III.3.26) ax to time translations a t , multiplying with a smooth function g(t)
and integrating we get

wo (SA) f g(t)dt = wo(6A(g)) (III.3.34)

with
A(g) = f g(t) atA dt. (III.3.35)
One has
140 III. Algebras of Local Observables and Fields

A(g)fl = f g(t)e' [Link] dt = g(H)Afl, (III.3.36)

where g is the Fourier transform of g and H the Hamiltonian. Since 511 = 0 we

can take A so that
wo(A) = 0; A E V(OR) (III.3.37)
Then, if the spectrum of the Hamiltonian in the subspace orthogonal to the
vacuum is confined to values e > m we have

II A(g)0 II <_ Es>up I g(E) 1 11 A^ 11 •

^,
(II1.3.38)

We can use now the freedom in the choice of g. Setting g(t) = T- 1 h(t/T) and
denoting the set of smooth functions h(r) with support in the interval (--1,+1]
and h(0) = 1 by . we get

II A(9)f2 II Ç F(mT) f J All II, (III.3.39)

where
F() = inf sup I h(w) 1 . (III.3.40)
hE.F w>

F is a universal function which decreases exponentially as shown in (Buch 921.

Analogously one gets a bound for IIA(g)*,flll. Applying the estimate (III.3.28)
to the right hand side of (I11.3.34) one obtains

I wo(SA) I < c.P(R + T) F(mT).

Letting T — oo one obtains (III.3.27).
One concludes that if the generator of a 1-parameter group of symmetries
satisfies condition (III.3.28), (1II.3.29) and if there is a mass gap then the
symmetry is implementable (unbroken) in the vacuum sector. The conclusion
may be strengthened to the statement that spontaneous symmetry breaking is
only possible if the mass spectrum contains zero as a discrete eigenvalue. A
discussion using conserved currents is given in [Ez 671. 6
Since we know that there exist photons and neutrinos the P'-spectrum has
no gap; so, strictly speaking, we can conclude nothing from the above theorem
in elementary particle physics. In particular, the masslessness of the photon
could allow a spontaneous breaking of Lorentz invariance even in the vacuum
sector so that there might be no Lorentz invariant (pure) vacuum state. If one
considers an idealized theory of strong interactions only in which one assumes
a mass gap then it follows that the vacuum is automatically Lorentz invariant
and invariant under the continuous part of the internal symmetry group of the
observables. The latter does not include the gauge groups since these do not
act on the observables. But it does include the chiral symmetry expressed by an
(algebraic) conservation law of an axial vector current. The idea that one has
6 Preciseconditions for the occurrence of spontaneous symmetry breaking with or without
the existence of massless particles have been given by Buchholz, Doplicher, Longo and Roberts
[Buch 92].
 II3.3 The Net of Algebras of Observables 141

such a symmetry in the idealized theory and that it is spontaneously broken

implies then that there must be a massless particle in this model, notably the
pion. Since the pion mass is relatively small compared to other hadronic masses
this picture appears reasonable and has led to the idea of a "partially conserved
axial vector current" in the full theory.
In the many body problem the Goldstone theorem does play a relevant rôle
(breaking of Galilei invariance, existence of phonons; see [Swieca 67]). In the
presence of long range forces there is, however, another mechanism for spon-
taneous symmetry breaking. The Bardeen-Cooper-Schrieffer model of super-
conductivity and the Coulomb forces in a plasma provide the prime examples.
Within the algebraic setting the former is discussed in [Haag 62b], the general
situation by Morchio and Strocchi in [Morch 85, 87]. The idea that in relativis-
tic field theory local gauge invariance provides an analogon to long range forces
(irrespective of the presence or absence of a mass gap) and that the sponta-
neous breaking of gauge invariance may in fact, be a mechanism for generating
an energy gap, parallel to the cases of the BCS-model and the plasma, was
forwarded by Nambu and Higgs. It led to the Higgs-Kibble mechanism which
became a central theme in the development of the standard model in elemen-
tary particle physics. We leave this important topic aside here because of the
gaps remaining in our understanding of the principle of local gauge invariance
in quantum physics (compare, however, the localization properties of charges in
the BF-analysis described in Chapter IV, section 3).

III.3.3 Summary of the Structure

To conclude this section let us summarize the essential structure of the the-
ory. We reserve here the symbol 0 for a finite, contractible, open region in
Minkowski space. The discussion above suggests that the following structures
are relevant:

(1) A net of C*-algebras with common unit

0 2ts (0),

with the total C*-algebra (inductive limit)

Z s = U o 2t s ( 0 ).

(The bar denotes the completion in the norm topology).

The action of ax on 21,9 is continuous in the norm topology.

(
i) A set 6 of physical states over 2t s and the complex linear span of Cam,
denoted by Z.
The restrictions of 6 and Z to the local algebras 2ts (0) yield the presheafs
{6(0)}, {Z(0)} of partial states or linear forms respectively.
142 III. Algebras of Loc al Observables and Fields

(iii) The dual of the presheaf E is a net of W*-algebras with common unit

O — R.(0) = E(0)*. (III.3.43)

R.(0) is closed in the weak topology induced by E and 21s(0) is weakly

dense in R.(0).

(iv) The symmetry group C. Its elements have a geometric part, point transfor-
mations of space-time conserving the causal structure. They are realized
by automorphisms of the net satisfying (III.1.7) where 21(0) may be
taken as either R.(0) or 2ts(0). The action of the continuous part of Ç
on 21,E is assumed to be continuous in the norm topology.

Remark. The local normality assumption asserts that for a finite, contractible
region O we have to deal with only one quasi-equivalence class of representa-
tions of 21s(0). Further, that the restriction of any state w E 6 to 21s(0) is
primary and the GNS-construction leads to a von Neumann factor 7r,,(2ls(0))"
which is isomorphic to R.(0). On the other hand 6 will still contain many
disjoint folia of states over the total algebra 21 s . The von Neumann algebras
iç,(U21 s (0))" will not be isomorphic for all states. We call each primary folium
of states over 21s a sector of the theory. For the analysis of the sector structure it
is not relevant whether we start from the net Ms(0) or from R.(0) but the total
algebra must be taken as the C*-inductive limit of the local algebras conforming
with the idea that we want to include only quasilocal elements. Examples of
different folia of physical states are provided by hydrodynamic-thermodynamic
states characterized by non vanishing matter and energy distribution at infinity
(chapter V ) and by the charge superselection sectors in elementary particle
physics (chapter IV).

The stability requirement, formulated in II.1.2 as axiom A2, A3, may be

expressed in algebraic form as the requirement that the left ideal 3(V+c) defined
in (III.3.21) is a proper ideal. The causality requirement, contained in axiom
E of II.1.2, is, of course, expressed by (III.1.4), where AObs may be taken as
either 21s(0) or 72-(0). The remnant of axiom B, the additivity property of A
suggested in (III.1.2), cannot be expected to hold in 21s but will require R (see
next section) . The same will very probably be true for the dynamical law. We
expect that it cannot be formulated within 216 , but requires also the states 6
or, alternatively speaking, the net R. So we should replace (III.1.10) by

R(0 ) = R (0). (III.3.44)

This feature appears also satisfactory in view of general relativity where even
the metric, which is essential for the dynamical equations, depends on the state.
We have seen that the properties listed imply the existence of at least one
pure, translation invariant ground state wo. In the GNS-representation induced
by wo the spectral condition A3 is satisfied. If the theory has a lowest mass
m > 0 then wo is invariant under the full continuous part of g. In particular
 1II.4 The Vacuum Sector 143

w0 is Poincaré invariant, axiom Al will apply and one can expect that the
vacuum state is unique. If there is no mass gap then the Lorentz invariance of
the vacuum is not guaranteed on general grounds.

III.4 The Vacuum Sector

We discuss here some further properties which the von Neumann algebras in the
vacuum sector possess (or might possess), considering also algebras associated
with infinitely extended or not contractible regions.

111.4.1 The Lattice of Causally Complete Regions

Definition 4.1.1
Let M be any set of points in Minkowski space. The causal complement of M
is the set of all points which lie space-like to all points of M. It will be denoted
by M'.
(M')' - M" is called the causal completion of M. M is called causally com-
plete if M" = M.

Proposition 4.1.2
(i) M' is always causally complete. One has
M ' =(M , ) „_ of ) ,_ M,,,
(III.4.1)

(ii) The intersection of two causally complete regions is causally complete.

If Mk = Mk , (k = 1, 2) then

M1 n M2 =(M1 n M2 )"-M1 A M2 . (11I.4.2)

The symbol A is used to indicate that the set is the largest causally complete
set contained in both M1 and M2.
(iii) There exists a lowest bound, denoted by M1 V M2, of the causally
complete sets containing both M 1 and M2. It is given by

M1VM2 = (M1UM2) '1 =(M;nM2) 1 - (111.4.3)

In short: The set of causally complete regions is an orthocomplemented lattice.

The smallest element is 0 (the empty set), the largest element is IR 4 . We denote
this lattice by K.

Proof. (i). (M')" = ((M')')' = (M")'. Since causal completion cannot di-
minish a set one has (M')" D M' and M" D M which yields (M")' C M. Thus
Mi"= M'.
144 III. Algebras of Local Observables and Fields

(ii) and (iii). The set of all points lying space-like to both M 1 and M2 may
be characterized either as M; n M2 or as (M 1 U M2)'. Thus

M; n M2 = (M1 u M2 )'. (III.4.4)

For causally complete sets we may put Mk = Nk, Mk = N and obtain

2
N1 n N2 = (N; U N )'.

This shows, according to (i) that N1 n N2 is causally complete. Taking the causal
complement one gets
M2)" =
(M1 u (M; n M)'.
Clearly this is the smallest causally complete set containing both M 1 and M2 .0

I11.4.2 The von Neumann Rings in the Vacuum Sector

We consider now the von Neumann algebras resulting by the GNS-constrûction

from the vacuum state w o . As long as we believe in the local normality assump-
tion for all finite, contractible regions the von Neumann rings ir o (R.(0)) and the
W*-algebras R(0) will be algebraically and topologically isomorphic. So we can
and shall omit the symbol 70 and take the von Neumann rings in the vacuum
sector as defining the net R. Only for an infinitely extended or topologically
non trivial region U the limitation to the vacuum sector becomes relevant. To
such a region we associate the von Neumann algebra

R(u) = (UocuR( 0))", (III.4.5)

where the union is taken over all finite, contractible regions contained in U.
Note that the double commutant refers to the vacuum representation. So this
object is not universal for all sectors.
We may confine attention to causally complete regions (see (II1.3.44)). One
notes the parallelism between the lattice structure of 1C described in the previ-
ous subsection and that of systems of von Neumann rings where the commutant
gives an orthocomplement in IC. The causality principle relates the orthocom-
plementation in /C with that in the set of von Neumann rings. This is, in fact,
the reason why we used a prime to denote the orthocomplement in .C. This
parallelism makes it tempting to assume that in the vacuum sector, ignoring
now the distinction between finite and infinite regions (as well as distinctions
between other classes of regions), the structural properties of the net can be
strengthened to the

Tentative Postulate 4.2.1

The vacuum sector of the theory is described by a homomorphism from the
orthocomplemented lattice 1C of causally complete regions of Minkowski space
into the lattice of von Neumann rings on a Hilbert space Ii.
 1II.4 The Vacuum Sector 145

This summarizes and extends the previously mentioned properties of the

net R. Specifically then

K R(K) C B(?-0),
R(K z A K2) = R(K2),
R(K1 V K2 ) = R(K1 ) V R(K 2 ),
R(K') = R(KY ,

1Z( 0 ) = {Il}.

It must be realized that postulate 4.2.1 is a considerable extrapolation from

physically motivated assumptions. It is suggested by mathematical simplicity,
motivated by the desire to impose the strongest restrictions which are not in
conflict with known facts. It is not used in this sharp form in the previous
literature nor in the subsequent chapters of this book. If one wants to be more
cautious one can start from considerably weaker requirements. Thus, in the
analysis of superselection sectors in chapter IV we shall assume the duality
relation (I1I.4.9) only for particularly simple regions, the diamonds (double
cones)
This duality relation extends the causality requirement which demands that

R(K i ) C 12,(K) 1 . (111.4.1 1)

If this is satisfied we may ask whether we can add further elements to 1.(K)
so that causal commutativity is still respected in the resulting augmented net.
If this is possible the left hand side in (III.4.11) increases, the right hand
side decreases. Duality means that one can make all local algebras maximal.
(I1I.4.9) was suggested on such grounds by Haag and Schroer, [Haag 62a1. In the
field theoretic setting it is supported by Borchers' argument on "transitivity of
locality" (see II.5) and, more generally, by the work of Bisognano and Wichmann
[Bis 75, 761. Araki proved the duality relation for von Neumann rings associated
with fr ee scalar fields [Ara 63b, 64a] . However most of the supporting evidence
for (III.4.9) relates to simple regions.
The intersection property (III.4.7) has been previously discussed for dis-
joint diamonds where Kl A K2 is empty. In this restricted form it has been
called extended locality and was shown to be a consequence of the standard
requirements by L.J. Landau [Land 69].
The strength of the combination of (II1.4.6) through (III.4.10) can be
illustrated in the following examples. They show that postulate 4.2.1 is violated
in some models constructed from fr ee fields but that it may hold if an interaction
satisfying the local gauge principle is present.
Take the fr ee Dirac field 0 and consider the current density ju,(x) as
the observable field by which the algebras R(0) are generated.' Then take
'Explicitly, start from the polynomial algebras of the smeared out currents in the vacuum
sector and go over to bounded functions of them e.g. by the polar decomposition of the
unbounded operators.
146 III. Algebras of Local Observables and Fields

K = K 1 V K2 where Kk (k = 1, 2) are space-like separated diamonds. K is dis-

connected. Each R(Kk) satisfies the duality relation (II1.4.9). This was proved
in [Dop 69a1. But if R(K) is defined by the additivity relation (III.4.8) then
R(K) does not satisfy duality for the simple reason that objects like ,*(f)/i(g),
where the support of f is in K1 and the support of g in K2, commute with all ele-
ments of R(K') and are not in R(K 1 ) V R(K2 ). One cannot construct ,(x) from
the currents in a neighborhood. Thus we have duality only for connected regions
in this model. But worse than that. Take K1, K2 as two concentric diamonds,
the causal completions of the 3-dimensional balls I x < rk at xo = 0, r1 < r2 .
The quantity Q = f j o (x) f (x)d 3x with smooth f and

1 for I x I< ri
f (x) _ 0 for IxI> r2
(or more precisely an expression like (III.3.31) is the approximate charge in the
smaller region. It commutes with `TL(K 1 ) due to the global gauge invariance of
the currents and with 1 (K2) due to locality. By (III.4.9) it should be contained
in R(K 1 n R(K2 ), and by (III.4.7) in R(K; n K2 ), the causal completion of
)

the spherical shell. This is not the case in the free Dirac theory. The situation is
improved, however, in full quantum electrodynamics because Gauss' law allows
the conversion of Q into an integral over the electric field strength in the region
of the spherical shell. This observation was presented as an argument in favour of
the validity and strength of the unrestricted duality relation [Haag 63]. Postulate
4.2.1 excludes the free Dirac theory and requires the currents to be accompanied
by another field which allows the determination of the charge inside a region
by measurements in a surface layer. Indeed, also the failure of (I11.4.9) for
disconnected regions in the free Dirac theory is remedied in full electrodynamics
because there i* (f)11i( g) is not in the algebra of observables; it is not invariant
under local gauge transformations. Instead of ili*(x),(y) we must now take the
gauge invariant objects (see I.5.34)

o(x, y) = (x) exp ie fxy A µ (x')dx' 1` 0(y) (III.4.12)

as (improper) elements of the observable algebra. A last example is provided

by the Maxwell field. The relation divB = 0 implies that the magnetic flux
through two 2-dimensional surfaces having the same 1-dimensional boundary
(for instance a circle) is the same. We may take small neighborhoods in JR 4 of
the surfaces and call their causal completions K 1 resp. K2. Then one can find an
observable (essentially the magnetic flux through either one of the two surfaces)
which belongs to 1 (K 1 ) and to 'R.(K2), hence by (III.4.7) to 1?..(K 1 n K2) which
is a neighborhood of the common boundary of the two surfaces. In free Maxwell
theory this is not possible if we require the additivity property (1II.4.8). Ig-
noring the small 4-dimensional extension of the regions considered, which was
introduced only to guarantee that the integrals give well defined observables,
the flux is given by the line integral of the vector potential around the (com-
mon) boundary. But the vector potential is not an observable while its line
 II1.4 The Vacuum Sector 147

integral around a closed curve is measurable namely as the magnetic flux. This
latter cannot be constructed from the field strengths in the neighborhood of
the boundary curve. In full electrodynamics we have, however, besides the field
strengths the quantities (111.4.12) and the currents generating the observable
algebra and one may hope to construct from them the line integral of A i, around
a closed curve using only a neighborhood of this curve. Unfortunately a verifi-
cation of this hope with any degree of rigour is difficult due to the very singular
character of the quantities involved. So this remark cannot be taken very seri-
ously.
The discussion in the last paragraph is intended to show that postulate
4.2.1 may be reasonable in physically interesting theories and that it is very
strong. It excludes free theories with non trivial charge structure while it possibly
admits full electrodynamics. The postulate looks mathematically canonical so
that it might be worth while to investigate its consequences for a mathematical
classification of theories. Still, since such a study is lacking we shall subsequently
adopt the cautious point of view and use the duality relation only for diamonds,
the intersection property only for disjoint elements of 1C ("extended locality").
An immediate consequence of postulate 4.2.1 is that R(K) is a factor for any
K E IC. If we adopt the cautious version then we shall only assume that R(K)
is a factor if K is a diamond.
IV. Charges, Global Gauge Groups
and Exchange Symmetry

IV.l Charge Superselection Sectors

By focusing attention on observables and algebras we have kept only those

parts of the axioms of chapter II which have a direct, immediately clear phys-
ical significance. Other parts, reflecting traditional field theoretic formalism,
such as the assumption of a Bose-Fermi alternative, or the founding of the the-
ory on fields transforming according to finite dimensional representations of the
Lorentz group do not appear. The understanding of internal symmetries is also
modified. In the development of elementary particle theory during the past two
decades gauge symmetries have acquired a more and more prominent position
and it may well be that all internal symmetries should properly be regarded as
gauge symmetries, formulated in the field theoretic frame by transformations
of unobservable fields leaving all observables unchanged.' Conversely, if in the
algebraic approach we take the extreme position and claim that the net of ob-
servable algebras defines the theory completely without need for any additional
specification, then we cannot allow any internal symmetry of the observable net
in the sense of (III.3.15) with a9 # id. So this would also lead to the conclusion
that internal symmetries should always be gauge symmetries. The most salient
observable consequence of internal symmetries is the set of charge quantum
numbers which serves to distinguish different species of particles and to charac-
terize their properties. In the case of gauge symmetries these quantum numbers
manifest themselves through the existence of superselection rules for the states
over the observable algebra. The charge structure with its composition laws as
well as the exchange symmetry ("statistics") are encoded in the structure of
the net of algebras of observables. By studying how they are encoded we get a
natural approach to these questions, unbiased by traditional formalism.

Strange Statistics. The question as to whether every particle must be ei-

ther a boson or a fermion and the wish for a deeper understanding of the Pauli
principle have been a recurrent theme of discussion on various levels of the
1 0f course this need not apply in models in which approximate symmetries are idealized
as exact ones.
150 1V. Charges, Global Gauge Groups and Exchange Symmetry

CITED EFED ER:

(a) (b) (c) (d) (e) (f) (g)

Fig. IV.]..]..

theory. One generalization of the Bose-Fermi alternative was suggested by H.S.

Green [Green 53]. He pointed out that certain trilinear commutation relations
of creation-annihilation operators are in harmony with the equation of motion
and transformation properties of free fields. He showed then that this algebra
can be decomposed into disjoint components, "the parastatistics of order p" (p
a natural number) and for each p one has a bosonic and a fermionic version.
As p -4 00 the two versions converge towards "Boltzmann statistics". Physical
consequences of the scheme have been studied by several authors, e.g. Volkov
[Volk 60], Greenberg and Messiah [Greenb 64, 65]. The contents can be most
simply described in the wave mechanical setting where states of n identical
particles are characterized by wave functions (6, ... , fi n ); here stands for
position and spin orientation. In the Hilbert space spanned by such wave func-
tions one has a unitary representation of the permutation group Sn resulting
from permutations of the arguments in the wave function. This representation
may be decomposed into irreducible parts. An irreducible representation of Sn
is labeled by a Young tableau, an array of boxes in rows and columns of non-
increasing length as pictured for the case n = 5 in the figure (IV.1.1). 2 The
length of a row indicates the degree of symmetrizability, the length of a column
the degree of antisymmetrizability. Thus a 5-particle wave function of symmetry
type (c) can be symmetrized in 3 arguments and then again in the remaining 2.
Alternatively it can be antisymmetrized in 2 arguments and then again in two
others but not more. If is an n-particle wave function of para-bosons of order
p then, for arbitrary n, precisely all such symmetry types are admitted which
have not more than p rows (i.e. for which cannot be antisymmetrized in more
than p variables). Correspondingly, for para-fermions of order p the number of
columns is limited to at most p.
A permutation of the order of arguments in a wave function of indistinguish-
able particles leads to the same physical state. Since for p > 1 we encounter
representations of Sn which are more than 1-dimensional this implies that pure
states are now no longer described by rays in Hilbert space but by subspaces
with the dimensionality of the representation of S n belonging to the Young
tableau. The specific limitations for the allowed Young tableaux in parastatis-
2 See any text book on group representations.
 IV.1 Charge Superselection Sectors 151

tics can be traced to the requirement that if some particles are very far away
then (asymptotically) the others can be regarded as an independent system
[Stolt 70], [Lands 67].
One may notice that the situation is analogous to the one encountered
in the treatment of the fine structure of atomic spectra by F. Hund. There,
building up wave functions from products of position space wave functions and
spin functions, the two factors need not be totally antisymmetric. In order to
obtain an antisymmetric total wave function the Young tableau of the spatial
wave function must be the mirror image of the tableau for the spin wave func-
tion. The Young tableau of a spin wave function can have at most two rows
because the electron spin has only two orthogonal states. The tableau of the
spatial wave function can therefore have, at most, two columns. Thus, if the spin
were unobservable, one could say that the electron is a para fermion of order 2.
This suggests that parastatistics of order p can be replaced by ordinary Bose
or Fermi statistics, respectively, if one introduces a hidden degree of freedom
("generalized isospin") with p possible orientations or, alternatively speaking,
if one has a non-Abelian unbroken global gauge group (isospin group). This has
been elaborated by Ohnuki and Kamefuchi [Ohnu 68, 691 and by Drühl, Haag
and Roberts [Drühl 70] .
Other types of exotic statistics are encountered in models on lower dimen-
sional position space. In these the exchange symmetry cannot be described in
terms of representations of the permutation group. Thus the full intrinsic sig-
nificance of exchange symmetry is not seen if one starts with multiparticle wave
functions, claiming that a permutation of the order of arguments leads to the
same state, nor if one bases the discussion on commutation relations between
unobservable quantities.
We shall show in the next sections how exchange symmetry arises from the
causality principle for observables and describe the possible types of exchange
symmetry which ensue.

Charges. Superselection rules arise because the abstract net has inequivalent
Hilbert space representations. But only a small subset of these representations
will concern us in this chapter, namely those whose states have vanishing matter
density at space-like infinity. These are the states of direct interest to elementary
particle physics, corresponding to the idealization that we have empty space far
away. 3 The natural formulation of this idealization would be to say that a state
W is relevant to elementary particle physics if

W (a„B) — ^^ 0 (IV.1.1)

as x goes to space-like infinity when B is any element of the Doplicher ideal

3(V+c) (see section III. Def.3.2.3). The analysis of superselection structure
3 Otherrepresentations, corresponding to thermal equilibrium (without boundaries) will
be discussed in chapter V.
152 IV. Charges, Global Gauge Groups and Exchange Symmetry

within the set of states restricted only by (IV.1.1) has not been carried through.
Borchers proposed instead to study the following set of representations.

Borchers' Selection Criterion (Positive Energy). We consider all rep-

resentations of the observable algebra in which the translation symmetry can
be implemented and the PA-spectrum lies in V.

The relation of this criterion to the idealization of absence of matter at infinity

is qualitatively seen in the following way. If the operators Pt,, can be defined at
all in a representation of then, starting from a state with finite energy we
may lower the energy by taking matter away - i.e. by acting on the state vectors
with quasilocal elements from .(V - ). If there is an infinite reservoir of matter
extending to spatial infinity then this procedure can be repeated indefinitely
often and the energy can have no lower bound. The first systematic attempt to
analyse the superselection structure was made by Borchers [Borch 65a, 65b1 and
based on this criterion. Unfortunately it contained some errors which were hard
to trace but led to conclusions to which counterexamples could be constructed.
An analysis based on Borchers' criterion without any further restrictive assump-
tions is burdened by complications arising from long range forces and infrared
clouds when the mass spectrum has no gap. Therefore the first objective was
to understand the situation in massive theories. With this regime in view Do-
plicher, Haag and Roberts started from the following criterion.'

DHR Selection Criterion. The representations considered are those which,

in restriction to the causal complement of any diamond of sufficiently large di-
ameter, become unitarily equivalent to the vacuum representation. In symbols

7r 121(d,) ti
71-0 121(0') (IV.1.2)

if O is a sufficiently large diamond.

This requirement is closely related to the common (vacuum-like) appearance

with respect to measurements in very distant regions of all the states consid-
ered.

Lemma 1.1
Take a directed sequence of diamonds exhausting space-time i.e. 0, i+1 D On ,
U On = 1R4 . If cv is a state in the folium of a representation satisfying (IV.1.2)
then
k
lira W —wo)io,{ =0. (IV.1.3)

Here wo denotes the vacuum, Sa bo, the restriction of the linear form cp to the
subalgebra Qt(O').
4This provided, in fact, the motivation for [Dopl 69a).
5In the present chapter we refer to this work as DHR 1, 2, 3, 4. This corresponds to
[Dopl 69a, 69b, 71, 74] in the author index.
 IV.1 Charge Superselection Sectors 153

Proof. By (IV.1.2) the folia of 7r and of pro coincide when the states are re-
stricted to an algebra %(O^) and n is large enough. Therefore yo = —to o is
a normal linear form on the von Neumann algebra R(O„) = iro (2t(O;^))". Con-
sider a sequence E R(On) with 1 B,, 11= 1. Since the unit ball of a von
Neumann algebra is compact in the weak *-topology this sequence has weak
limit points in 93(7-1) and each such limit B commutes with all observables.
Since the vacuum representation is irreducible B is a multiple of the identity
and cp(B) = O. ❑

The converse of lemma 1.1 is also true under rather mild assumptions. If the
restrictions of 7r and pro to 24O0 are primary then (IV.1.3) implies that these
restrictions are quasiequivalent for large n. For type III factors quasiequivalence
is the same as unitary equivalence. It will be shown in chapter V that the rel-
evant von Neumann algebras are indeed of type III. We can therefore consider
(IV.1.3) as an alternative formulation of the selection criterion (IV.1.2) .
Condition (IV.1.3) is a much more stringent restriction than (IV.1.1) because
it demands for the states considered a uniform approach of the expectation val-
ues for all observables far away to the vacuum expectation values (irrespective
of the extension of their support region). Thus it excludes from consideration
states with electric charge because, by Gauss' law, the electric charge in a fi-
nite region can be measured by the flux of the field strength through a sphere
of arbitrarily large radius. This is due to the long range character of electric
forces which, in turn, is tied to the vanishing of the photon mass. In a massive
theory forces are expected to decrease exponentially with distance so that the
charges become shielded. However, a careful analysis by Buchholz and Freden-
hagen [Ruch 82a} showed that even in a purely massive theory there can exist
charges accompanied by correlation effects which are discernable at arbitrarily
large distances. They found that if one starts from a sector containing single
particle states but not the vacuum in a massive theory then Borchers' criterion
is equivalent to a requirement of the form (IV.1.2) but with O replaced by C,
aniftelyxdcoarunsmebitlychonpa-kedirt.
It is this modification which allows topological charges. The adjective "topolog-
ical" is appropriate because a particular state does not determine the direction
of the cone which has to be excluded. It is only important that a large sphere
surrounding the charge must be punctured somewhere. We shall describe this
in section 3 (BF-analysis).
The DHR criterion excludes such "topological charges" from consideration.
It is therefore too narrow. It is instructive to see that this criticism goes deeper
and touches some of the pillars of general quantum field theory. The main ar-
gument in favour of (IV.1.2) as a reasonable selection criterion in a massive
theory was the fact that it is implicitly assumed in the standard treatment of
collision processes described in II.4. This, in turn, was based on the axioms
in II.1 but continues to work in a more general setting where the Bose-Fermi
alternative is eliminated from the assumptions. Specifically, let us consider the
frame sketched in III.1. We have a Hilbert space 71, a net of operator algebras
154 IV. Charges, Global Gauge Groups and Exchange Symmetry

.A(0) whose elements are interpreted as representing physical "operations" in

the respective region. We may imagine that it is generated by fields, some of
which are charge carrying and unobservable. Therefore we call A the field net.
We have furthermore a unitary representation of 513 transforming the field net
according to equ. (111.1.3), There is a ground state (vacuum). Inside A(0) we
have the observable part 21(0), concretely realized as an operator algebra on
7-1. The set of all unitary operators which commute with all observables and
represent an internal symmetry of the field net is called the global gauge group
g. The observables are assumed to obey causal commutation relations but the
commutation relations between unobservable quantities are left open. The stan-
dard construction of incoming particle states works if the vacuum expectation
values of fields have the cluster property

limwo (Fi(axF2)F3) = wo(F1F3) wo(F2); F { E Aloe (IV.1.4)

for x moving to space-like infinity and if the fields commute with the observables
at space-like distances. Under these circumstances, as shown in [Dopl 694 the
vector states in R satisfy the criterion (IV.1.2). The decomposition of h with
respect to the center of 21" (which is also the center of Q" because G" = 21')
gives the decomposition into charge superselection sectors. If the DHR-criterion
is too narrow then also the standard collision theory is too narrow. The topo-
logical charges encountered in the BF-analysis cannot be created by quasilocal
operators of a field net.

The Program and the Results. Having settled for one of the criteria sin-
gling out the states of interest the remaining task is clear. Starting with the
abstract algebra of observables we have to classify the equivalence classes of its
irreducible representations conforming with the criterion. We shall call each of
these a charge superselection sector and the labels distinguishing sectors charge
quantum numbers. This terminology is appropriate if the set of sectors is dis-
crete and this will turn out to be the case in massive theories under physically
transparent conditions. 6 Still the term charge is used in a wide sense. In spe-
cific models it may be baryon number, lepton number, magnitude of generalized
isospin ...
I give a brief preview of the results. Two properties of charge quantum num-
bers follow from the general setting: a composition law and a conjugation. Tied
to this is the exchange symmetry of identical charges ("statistics") . If the theory
is based on Minkowski space (or higher dimensional space times) the total struc-
ture is remarkably simple and not affected by the presence or absence of topo-
logical charges. The charge quantum numbers correspond in one-to-one fashion
to the labels of (equivalence classes of) irreducible representations of a compact
group, the global gauge group. The composition law of charges corresponds to
the tensor product of representations of this group, charge conjugation to the
6 In DHR1 it corresponds to the compactness of the global gauge group which was shown
there to follow from the absence of infinite degeneracy of particle types with equal mass and
the completeness of scattering states (asymptotic completeness).
 IV.1 Charge Superselection Sectors 155

complex conjugate representation. in addition, intrinsically attached to each

type of charge, there is a sign, determining whether the charge is of bosonic or
fermionic nature. If the global gauge group is Abelian then the charge quantum
numbers themselves form a group, the Pontrjagin dual of the gauge group. The
exchange symmetry gives ordinary Bose or Fermi statistics. If it is not Abelian
the composition of two charges leads to a direct sum of several charge sectors
corresponding to the Clebsch-Gordan decomposition of the tensor product of
representations of the gauge group. The statistics in the sectors arising from
n-fold composition of one type of charge may be described in two ways. One
may either say that it is para-Bose (resp. para-Fermi) of order d, where d is the
dimension of the irreducible representation of the gauge group in the charge 1-
sector. Or one may take an extension of the observable algebra, adding "hidden
elements", and consider the non-Abelian part of the gauge group as an internal
symmetry group of this enlarged algebra. The statistics reduces then again to
the ordinary Bose-Fermi alternative. A simple example illustrating these dif-
ferent descriptions is the Yukawa type theory of pions and nucleons neglecting
electromagnetic and weak interactions, the "charge independent theory". There
SU(2)xU(1) is the global gauge group, SU(2) corresponding to charge inde-
pendence, U(1) to baryon number conservation. One is unable to distinguish a
neutron from a proton or the different electric charges of a 7r-meson. Instead one
has a nucleon, belonging to isospin 1/2 as a para-fermion of order 2 and a meson,
belonging to isospin 1, as a para-boson of order 3. Enlarging the observable alge-
bra we come to the usual description with two distinguishable types of nucleons,
three types of mesons, which are ordinary fermions and bosons respectively. If
we do this without giving up charge independence then the SU(2)-group appears
as an internal symmetry group of the (extended) observable algebra and only
the U(1) part remains as a gauge group. Practically, of course, the very fact
that states with different electric charge can be experimentally distinguished
means that the SU(2) symmetry is lost.
In recent years there has been growing interest in models based on 2- or
3-dimensional space times. The motivation has come from various sources: the
existence of solvable or partly solvable models in lower dimensions, the informa-
tion about non-relativistic statistical mechanics at critical points which may be
drawn from conformal invariant field theories in 2-dimensional space-time and
the fact that such a model serves as a point of departure for theories with strings
and superstrings. It is therefore noteworthy that the discussion of the superse-
lection structure and the exchange symmetry is less simple in low dimensions.
The permutation group is replaced by the braid group, the gauge group by a
more complicated structure ("quantum group" or worse). In the presence of
topological charges the richer structure appears already in (2 + 1)-dimensional
space-time, for localizable charges in (1 + 1)-dimensional space-time. We shall
discuss this in section 5 of the present chapter.
156 IV. Charges, Global Gauge Groups and Exchange Symmetry

IV.2 The DHR-Analysis

Although, as mentioned, the DHR-criterion is too restrictive it is worthwhile

to present the DHR-analysis. It illustrates the essential features. Modifications
will be discussed in sections 3 and 5 of this chapter and, concerning infrared
problems, in Chapter VI.
In this section the symbol O shall always denote a diamond (as the
prototype of a finite, causally complete region) and O' its causal comple-
ment. We write RepL for the set of representations of the observable alge-
bra conforming with (IV.1.2). The index L stands for "localizable charges".
R(0) = era (2L(0)") , R(O`) = Ira (2t(O`))" denote the von Neumann algebras
in the vacuum representation. The duality relation

R(O')' = R(0) (IV.2.1)

will play an important rôle. All representations from RepL are locally equivalent
i.e.
7r (24O)) ti 7r0 (21.(0)) ; r E RepL . (IV.2.2)
We meet, in restriction to finite regions, only one equivalence class of represen-
tations, the normal representations of the von Neumann algebras 7Z(0). The
same holds, by (IV.1.2), for the infinite regions O' at least if O is large enough.
So the charge sectors differ essentially only in the way how the representations
of T (0) and TZ(O') are coupled. We may, for our purposes here, take the net
{R.(0)} instead of {2t(0)} as defining the theory, write A instead of ir o (A) and
consider the elements A E UR(0) as operators acting in ? -10 . The algebra of all
approximately local observables 0. is the C*-inductive limit of the net 7Z

2t = U RZ(0) • (IV.2.3)
a
IV.2.1 Localized Morphisms

The language is simplified if we describe all representations in the same Hilbert

space, the space 71 0 on which the vacuum representation is defined by the op-
erator algebras R(0). If 7r E RepL acts on l,r then we may replace it by an
equivalent representation acting on 7 -to
Q(A) = V 7r(A) V -1 ; A E 21. (IV.2.4)
where V is any unitary mapping V : 7-1,r — ^^ ?-lo. Such mappings exist because
no and 7-i are separable Hilbert spaces. By (IV.1.2) we can choose V so that
it establishes the equivalence between 7r and 710 for some O`. This means

g(A) = A for A E 24O'). (IV.2.5)

Note that on the one hand A — ^^ A(A) gives a representation of 2t. and of the net
R. (in the equivalence class of 7r) by operator algebras in 13(7 -‘0 ). On the other
hand we may regard O as a map from 2t into O. with the property
 IV.2 The DHR-Analysis 157

A 77.(01 ) C R(01 ) for 01 D 0 . (IV.2.6)

Property (IV.2.6) is a consequence of the duality relation (IV.2.1). Take A E
R(01). Then by (IV.2.4), (IV.2.5) A(A) commutes with R.(0) since 01 C 0'.
Thus (IV.2.1) implies (IV.2.6). Further, the map A is isometric

II60)11 = 11All • (IV.2.7)

This follows because we can obtain an equivalent representation A1, starting

from a different choice of the region n in the above construction as long as the
radius of the region is large enough. In particular we can choose it space-like to
the support of any element A E 2110c which we consider. Then (IV.2.5) for P i
211p, is norm dense in 21 one sees that A indeedwilmpy(IV.27)forASince
maps 21 into 21. Finally, since A — A(A) is a representation, the map A respects
the algebraic relations. Thus we may regard A as an endomorphism of the C*-
algebra 21 with the additional features (IV.2.5) and (IV.2.6). We call such an
endomorphism of 21 a localized morphism with localization region (or support)
n because A acts trivially in the causal complement of n and transforms R(0)
into itself. We have

Proposition 2.1.1
For any 7r E Rep L and any diamond n of sufficiently large diameter we can
obtain a representation in the equivalence class of 7 by operators acting on no
applying a localized morphism with support 0 to the net R.

Definitions 2.1.2
(i) Two localized morphisms Pi , 02 are called equivalent if the representations
Pi (21), g 2 (21) are unitarily equivalent. We denote the equivalence class of A by [A]
The morphism A is called irreducible if the representation A(21) is irreducible.

(ii) A localized morphism with support in 0 is called transportable if there exist

equivalent morphisms for any support region arising by a Poincaré transforma-
tion from 0. The set of transportable, localized morphisms will be denoted by
4 the subset of irreducible ones by di„, the set of those with support in n by
d(0).
(iii) A unitary U E gtloc = U R.(0) defines a transportable, localized automor-
phism ou by QUA T UAU -1 , A E 21. The set of these (the inner automorphisms
of 211 ) is denoted by I.

Lemma 2.1.3
Two localized morphisms P i , A2 are equivalent if and only if

A2 = QA1! Q E Z (I V.2. $ )
158 IV. Charges, Global Gauge Groups and Exchange Symmetry

Proof. The equivalence means that there exists a unitary U E B(l0) such that
02(A) = Ug1 (A)U -1 . There exists a diamond 0 containing the supports of both
Pi and g2 so that gl and g2 act trivially on 77,(0'). Hence U must commute with
R(0'). By duality U E 77,(0) and aU E T. ❑

The important point about describing representations by morphisms is that

morphisms can be composed. We show that the product of morphisms from
d respects the class division so that we also obtain a product of equivalence
classes.
Notation. Considering g as a representation we have written g(A) for the rep-
resentor of the algebraic element A. Considering g as a mapping it is more
convenient to write pA instead. Since we now have two multiplications, one be-
tween morphisms, one between algebraic elements, some consideration has to
be given to the setting of brackets. We shall omit them in a straight product
of morphisms and algebraic elements. Thus p 2 p1 AB means that the two mor-
phisms are successively applied to the algebraic product of A and B, a bracket
indicates the end of the effect of a morphism. Examples: gAB = (gA) (pB),

g QU A = g UAU -1 = (pU)(gA)(gU -1 ) = aw p A, (IV.2.9)

W=pU.

Lemma 2.1.4
If pt E d, i = 1, 2 then the equivalence class of the product depends only on
the equivalence classes of the factors. Therefore a product of classes is defined

[Q2p1] = [p2][pl] • (IV.2.10)

Proof. If i.e. = auipt then, using (IV.2.9), we have g2g' = aw0201

with W = U2g2U1 • ❑

Next we note that the causal structure carries over from the observables
to the morphisms.

Lemma 2.1.5
Morphisms from 1 commute if their supports lie space-like to each other.

Proof. Let g= E .6(0 0, i = 1, 2 and A E 77.(0) where 0 1 lies spacelike to

02 and d is arbitrary. As indicated in the figure (IV.2.1) we can achieve, by
translating the regions 0 / , 02 respectively to 0 3 , 04 , that 0 is space-like to
both 03 and 04 , that 02 is space-like to a diamond 0 5 containing both 0 1 and
03 and 0 1 is space-like to a diamond 06 containing both 02 and 04 . There
are morphisms g3 = a13101, p4 = aU42 p2 i localized in 03 , 04 respectively and
equivalent to p i , resp. 02 . Then
 IV.2 The DHR-Analysis 159

Fig. IV.2.1.

A4A3A = 0304A = A (IV.2.11)

By (IV.2.1) (duality) the localization of the unitaries which transport the mor-
phisms is given by U42 E R(06), U31 E 17.(O5 ). Therefore A1U42 = U42, A2U31 =
U31; U31U42 = U42U31, and (IV.2.11) implies A2A1A = A1A2A. ❑

As a consequence we have

Proposition 2.1.6
(i) The charge quantum numbers are in one to one correspondence with the
elements of Birr /Z (the set of classes).
(ii) .A is a semigroup, d/Z is an Abelian semigroup.

IV.2:2 Intertwiners and Exchange Symmetry (Statistics)

The concept of intertwiners between non-disjoint representations was described

in II1.2. In our present context, where the representations are obtained by lo-
calized morphisms of a local net, there results some additional structure. We
denote the set of intertwining operators from a morphism A to a morphism A'
by i(A', A). An element of this set is an operator R E 93(N0) such that

(A'A)R = R(AA); A E 21. (IV.2.12)

It follows then from duality that R E 2t10 . We shall use the term intertwiner
for the triple (A', R, A) standing in the relation (IV.2.12) and denote it by R.
The morphism A will be called its source, A' its target. The adjoint

R* = (A, R * , A') (IV.2.13)

is an intertwiner in the opposite direction. If R 1 E i(A1, A), R2 E i(A2, Pi) then

R2 R 1 E i(A2, A). Thus, if the target of R1 coincides with the source of R2, there
is an obvious composition product of these intertwiners

R2 0 R1 = (02) R2R1, A) • (IV.2.14)

160 IV. Charges, Global Gauge Groups and Exchange Symmetry

The product of morphisms introduces another kind of product between

intertwiners. Given Ri E 2 (At, pi ), (Z = 1,2) then one checks that R = R2A2R1 =
(02R1)R2 is an intertwining operator from A2A1 to A2Ai. We write

R2 x R1 = (61 2011, R2 A2R1, A2 AI) • (IV.2.15)

One checks that this cross product is associative, that

(R xS)*= R* x S*, (IV.2.16)

and, if Rt = (4, Ri, p), Ri _ (Ai, Ri, 0i), i = 1,2 then

(R2oR2 ) x (Rio R1 ) = (R.'2 x o (R2 x R1 ) . (I V.2.17)

The cross product inherits causal commutativity if both the sources and the
targets of the factors have causally disjoint supports.

Lemma 2.2.1
Let Ri = (As, Ri pi ), 2 = 1, 2. If the supports of 01, 02 as well as those of Ai, A'2
,

are causally disjoint then

R2xR1=R1xR2. (IV.2.18)

Warning: this lemma does not hold in (1 + 1)-dimensional space-time.

Proof. We have to show that under the stated conditions

Ri(A1R2) = R2(02R1) (IV.2.19)

This is trivially true if for instance

supp Pi = supp Ai = 01; Stipp A2 = Kipp A2 = 02 (IV.2.20)

and 01, 02 are causally disjoint. Consider the change of both sides of (IV.2.19)
when we make a small shift of 0 2 to 03 so that the smallest diamond containing
the supports of both 0 2 and A3 remains space-like to the support of Al (see fig.
(IV.2.2)). This replaces R2 by R3 = R2 o U where

U = (A2, U, A3); e3 = au- P2

and supp U is space-like to supp AI, thus A1U = U. Then
R1 x R3 (Ai A2 , R13, PI N) with
R13 = RIp1R3 = R1A1R2U
= Ri(AIR2)(AIU) = Ri(AIR2)U .

In the same way one gets

R3 X RI = ( A2Ai, R31, 0301) with

R31 = R2 (A2 R1)U
 N.2 The DHR-Analysis , 161

Fig. IV.2.2.

Thus, by the change from 02 to A3 the two sides of (IV.2.19) change by the
same factor U. If they were equal before they remain equal. By a succession of
such small shifts we can change 02 to any morphism in its class with support
space-like to 01 . Then, analogously, we can change 0i by continuous defor-
mation to any morphism in its class with support space-like to 02 and thus
reach the general placement of all the supports specified in the lemma. Note
that in these deformations we always have to keep the support of the moving
morphism space-like to that of the fixed one. In order to reach an arbitrary final
configuration from the initial one we need a space with at least two dimensions
(at fixed time). ❑

The next topic is the exchange symmetry. We start from a charge 6 E dirr/I
and consider the folium 62 of "doubly charged states". To characterize state vec-
tors we pick a reference morphism A in the equivalence class of 6 and a reference
state w = wo o 02 in the folium (i.e. w(A) = w o (02 A)). Other doubly charged
states are obtained by taking other morphisms

Ai= Qut O; i =1,2 (IV.2.21)

in the class and forming

w21 (A) = wo(0201A) . (IV.2.22)
Such a state may be pictured as one where a charge 6 has been placed into the
support regions of Ai and of 02, leaving the vacuum everywhere else. Having
picked the unitary intertwiners from A to 0i in the singly charged sector

Ui = (0i , Ui T A) (IV.2.23)

we get the intertwiner from 0 2 to 0201 as

U2 x U1 = (02011 U21) 0 2 ); U21 - U2AU1 (IV.2.24)

and we get as the natural assignment of a state vector to the state w2 1 in the
representation 02 the vector
162 IV. Charges, Global Gauge Groups and Exchange Symmetry

Q21 = U21 ,f2 (IV.2.25)

where Q denotes the reference state vector corresponding to w o o 02 which is the
vacuum state vector in Ho . Now, if the supports of Ai, 02 are causally disjoint
then, by lemma 2.1.5, w21 = W12 and we might equally well have assigned to
this state the state vector

fl12 = U2 f2 ; U12 = Ui AU2 • (IV.2.26)

The two assignments differ by a unitary Ee = U1 2 U21 whose properties are de-
scribed in the next lemma.

Key Lemma 2.2.2

Let Ai, 02, A be equivalent morphisms from d. So

Ai = au, 0, i = 1, 2, au, E I , (IV.2.27)

and let the support of Al be space-like to that of 02. Then the unitary operator
(the statistics operator)

Ee
(0U2 1)Ul 'U2(pUi) (IV.2.28)

has the following properties

a) commutes with all observables in the representation 0 2 , i.e.

Ee E (02 21)' . (IV.2.29)

b) Ee depends only on A, not on Al, 02 as long as they vary in the class

[A] and have causally disjoint supports. Nor does it depend on the choice of
the unitary operators Ui which implement the inner automorphisms au in the
singly charged sector.
c)
(IV.2. 3 0)
d) If A' = awg then Ee, = aV E Q with V = W(AW).

Warning: parts b) and c) do not hold in 2-dimensional space-time.

Proof. Part a) expresses the fact that Eg is an intertwining operator from 0 2 to

A2 :
Ee = (U1 X U2 )* o (U2 x U,) = (0 2 , Ee, 0 2 ) . (IV.2.31)
Part b). The independence of e Q from the choice of the Ai follows from lemma
2.2.1. Let Ai, p2 be two other morphisms in the class [A] with causally disjoint
supports, related to Ai, 0 2 respectively by the intertwiners

Wi = (Ai Wi, Ai), i = 1,2

,

(see fig. (IV.2.3)). Then the intertwiners from 0 2 to A2A1, respectively to Ai, A2
are
 IV.2 The DHR-Analysis 163

Fig. IV.2.3.

U21 =(W2 o U2) x (W1 o U1) = ( W2 x W1) o (U2 x U1)

U12 = ( W1 o U1) x (W2 o U2) = ( W1 x W2) o (U1 x U2)
(IV.2.32)

where we have used (IV.2.17) . Since the sources as well as the targets of W 1
W2 are causally disjoint one has W2 x W1 = W1 x W2 by lemma 2.2.1. and
Inserting (IV.2.32) in the definition (IV.2.31) for Ep the W-factors cancel and
one obtains Ely = EQ . The independence of Ee from the choice of the intertwining
operators Ui in the representation p follows as the special case pi = pi . A direct
proof of part b), avoiding the use of lemma 2.2.1, is given in DHR3. It should
be kept in mind that the argument depends on the possibility of shifting space-
like separated diamonds continuously to arbitrary other space-like separated
positions without having to cross their causal influence zones. This homotopic
property needs a space-time dimension of at least 3. Part c) follows from b)
because an interchange of p1 and p2 in (W.2.31) changes into e;1 . Part d)
follows from elementary computation. ❑

The generalization of this lemma to the sector [p]1 , the states with n iden-
tical charges, is given in the following theorem.

Theorem 2.2.3
Let p1, ... en be morphisms from A, with causally disjoint supports and all
equivalent to a reference morphism p with

Ut = (pi, Ui, e), 2 = 1, ... , n

intertwiners from p to p i . For each permutation

1 ... n
P=
P(1) ... P(n)
define

Up(P) = Up -1 (1) X Up -1(2) X ...Up-i( n )

(PP-i(1) . • • p p-1( n,), Üp (P), On )

(IV.2.33)
164 IV. Charges, Global Gauge Groups and Exchange Symmetry

and

Eénl (P) = Ué(P) o Ue(e)

= (A n e(en) (P), An) (IV.2.34)

where e denotes the unit element of the permutation group. Then

a) £(on) (P) commutes with all observables in the representation pn, i.e.

£(:)(P) E (An21)' (IV.2.35)

b) £pnl (P) is independent of the choice of Al, ... , An and of the choice of the
intertwining operators Ui as long as the supports of the p ti are causally disjoint.
c) The operators £ o( n) (P) form a unitary representation of the permutation
group:
£(n) (P2)£ g( n) (P1) = £e( n) (P2P1) (TV.2.36)
d) If Tm denotes the transposition of m and m + 1, (m < n),

en) (Tm ) = Iel -1 X£

e X Ip -m-1 (1V.2.37)

where Ip stands for (pr, 1,0r ) and £ e = (p2 , £e , p2 ) is defined in lemma 2.2.2 (the
special case for n = 2); in terms of the operators

£(en)(Tm) = A m-1 £ . (IV.2.38)

Proof. Parts a) and b) are the generalizations of the corresponding statements

in the key lemma from n = 2 to arbitrary n and the proof proceeds in complete
analogy. For part c) pick two permutations P' and P" and set k' = P'(k), k" =
P"(k), k = 1, ... n. An intertwiner from Ak, to Ak „ is
,

*
k" k' = Ak" , Uk" , A ^ A, Uk ' , Ak' •

By lemma 2.2.1 the cross product over all k

V 17-- Xk_l Uk"k ,

is independent of the order of factors since all sources are causally disjoint and
all targets also. Hence V does not change if P' and P" are replaced by P` P"'
and P" P"' respectively. On the other hand we have by (IV.2.17) and (IV.2.33)
v Ue(P,r-1) o Ue (P' -1 )* .
Thus
Ue (P"-1 ) o U 0 (Pi-1 )* = Ue (e) o U(P"P' 1)* . -

With the definition of e(on) , putting P"Pi-1 = P1 , P"' = P2, this gives (IV.2.36).
Part d) is evident from the definitions. ❑
 IV.2 The DHR-Analysis 165

We note that, quite generally, the equivalence class of a product of mor-

phisms does not depend on the order of factors. One can construct natural in-
tertwiners between products of morphisms in different order. This is described
in the following generalization of the preceding theorem.

Proposition 2.2.4
Given ok E d which may be in different equivalence classes; k = 1, ... , n.
Choose n morphisms ok°) with mutually space-like supports, equivalent respec-
tively to pk so that there exist unitary intertwiners Uk = (ok°), Uk, ok). The
cross product of these intertwiners in the order determined by the permutation
P as in (IV.2.33) is denoted by U(P) and

= U*(P) o U(e) . (IV.2.39)

Then
a) e(o1, • , On; P) is independent of the choice of the ok°) and Uk within the
specified lira itat ions.
b) E(o1, • .. , On; P) = Il if the pk have mutually space-like supports.
c) Given intertwiners Rk = (ok, Pk, 0k) then

R(P)oE(os,-••,N;P) =E(oi,..., Ain; P)oR(e) (IV.2.40)

The computations establishing these claims can be found in DHR3.

Discussion. The significance of the operators E e( n) has been previously described

for the case n = 2. Similarly, in the folium of states with n identical charges
we pick a reference state w ° o on and assign to it the state vector Q in the
representation on. The morphisms ok may be interpreted as charge creation
in the support regions 0k. If these regions are causally disjoint then the state
wo 0 op-1(1) ... op-1(n) does not depend on the order of the factors ok i.e. it is
independent of the permutation P. The natural assignment of a state vector to
this state depends, however, on the order "in which the charges are created".
For the above order it is
Wp = UQ (P) * ,f2 . (IV.2.41)
This vector in the representation On may be regarded as the products of the
single charge state vectors [Link] from the representation p, applied in the order
in which the factors Uk appear in U4 (P). The 4,n) change this order

E e( n) (P')Wp = Wp , p . (IV.2.42)

The state is independent of the order because the E(pn) commute with all observ-
ables in the representation pn. These properties of the permutation operators e
'Product in the sense of the symbol ® t used in the collision theory of chapter II. The time
t corresponds now to a space-like surface through the regions Ok
166 W. Charges, Global Gauge Groups and Exchange Symmetry

make them completely analogous to place permutations of product wave func-

tions of n identical particles. By part d) of the theorem the Eel for arbitrary n
can be constructed from EQ of lemma 2.2.2.

Remark. If o is an automorphism (if the image of 21, under A is all of a[) then
An is irreducible and thus all 4) are multiples of the unit operator. This leaves
only the alternative between the completely symmetric representation of the
permutation group (Bose case) and the completely antisymmetric one (Fermi
case); EQ of lemma 2.2.2 is either +1 or —1 and this sign is an intrinsic property
of the charge [A]. In this context the following lemma is of interest.

Lemma 2.2.5
Let A E Birr. The following conditions are equivalent
a) A is an automorphism,
b) g2 is irreducible,
c) E Q = ±11.
d) The representation of the permutation group in the sector p' is either
the completely symmetric or the completely antisymmetric one.
e) The representation o satisfies the duality condition

(o1.(Q'))' = g17.(0) . (IV.2.43)

The proof is rather simple (see DHR3, lemmas 2.2, 2.7 and equation (IV.2.37)
above) . We shall record here only the argument leading from e) to a) . It suffices
to show that A maps RZ(n) onto itself when 0 contains the support of g. In
that case, however, A acts trivially on 77.(0') so one can omit A on the left
hand side of (IV.2.43). Then, using the duality in the vacuum sector, we get
RZ(n) = of (n).
The fact that e) entails a) is of interest because it suggests an alternative
approach to the construction of a global gauge group, simpler than that of
section 4 below and in line with the qualitative arguments at the end of section
1. It should be possible to show that within the set of representations considered
one has "essential duality" namely that there exists an extension of the algebra
of observables which is maximal i.e. satisfies the duality relation. This requires
a generalization of the theorem of Bisognano and Wichmann (see chapter V).
For the extended algebra one has then the Bose-Fermi alternative in all charge
sectors. The automorphism group of the extended algebra which leaves the
observables element-wise invariant is the non-Abelian part of the gauge group.
It may be regarded as an internal symmetry group for the extended algebra. The
parastatistics for the observable algebra arises from the ordinary statistics of
the extended algebra by ignoring the "hidden", unobservable degrees of freedom
as described at the end of section 1.
 IV.2 The DHR Analysis
- 167

IV.2.3 Charge Conjugation, Statistics Parameter

We show next that all morphisms P E d can be obtained as limits of sequences

of inner automorphisms, the charge transfer chains. Let P E 2(0) and take a
sequence Ok, space-like translates of 0 moving to infinity as k — oo. For each
k take a morphism Pk with support in Ok in the equivalence class of p, so
PkA = cruk pA = Uk(PA)Uk 1 (IV.2.44)
with Uk a unitary from R(0) V R(0 k ). If A E 2t1oc then pkA = A for sufficiently
large k. Therefore
pA = lim cru-1A = lim UkAU k (IV.2.45)
k ■ oo k

where the convergence is understood in the uniform topology of 93(74). If p

is an automorphism then also the sequence Uk AUU will converge uniformly to
p-l A. If p is only an endomorphism the convergence of this sequence is not
evident. Under some additional conditions it has been shown that this sequence
converges weakly in 23(7). These conditions, which cover the cases of principal
physical interest are [Ruch 82a]

BF-Conditions 2.3.1
The automorphisms of space-time translations are implement able in the sector
[p]. The energy spectrum is positive, the theory is purely massive and the sec-
tor [p] contains single particle states. In other words, there is an isolated mass
hyperboloid in the Po-spectrum of the sector.

Lemma 2.3.2 2
Let Uk be a sequence of charge transfer operators as described above with Ok
moving to space-like infinity as k —> oo. If p satisfies the BF-conditions then
crUk A converges weakly. The limit is independent of the chain Pk (though not of
the starting point p) and defines a positive map from 2t to 21.:
°QA = w — lim Q Uk A (IV.2.46)
with the properties
(i)
06,A = A for A E WO') ;
0 e2t(01) C 2t(0 1 ) if dl D 0 , (IV.2.47)
(ii) Oe is a left inverse of p, i.e.
OQ ( pA)B( pC) = A(O eB)C , (IV.2.48)
O e ( Il ) = Il. (IV.2.49)

2 1n DHR3 the weak convergence of auk A was not used. The analysis is then more com-
plicated but the conclusions remain essentially the same except that the case of "infinite
statistics" i.e. = 0 in (IV.2.55) can then not be excluded and that in this pathological case
the left inverse is not unique and the conjugate sector cannot be defined.
168 IV. Charges, Global Gauge Groups and Exchange Symmetry

The hard part in the proof of this lemma concerns the weak convergence
and uniqueness. We defer it to the next section. Claim (i) then follows from the
fact that QUk acts trivially on those elements of 2t whose support is causally
disjoint from both 0 and Ok, together with the duality in the vacuum sector.
For part (ii) we note that due to the uniform convergence in (IV.2.45) QUk PA
converges uniformly to A and

Cuk(PA)B(PC) = (crukpA)(clUkB)(cTukpC)

Remark. Oe is not a morphism of 2t. It does not respect the multiplicative

structure i.e. 06,AB (O e A)(0 eB) unless one of the factors belongs to p2t. Pe
is, however, a positive linear map of 2t into 2t. This, together with (IV.2.49)
implies that
wc ([Link]) ZDe (A) (IV.2.50)
is a state. It determines a representation Tr via the GNS-construction. One finds
(see below) that, if P E Airr, satisfies the BF-condition, the automorphisms of
space-time translations are implementable in Tr with an energy-momentum spec-
trum coinciding with that of the representation P and that Tr is irreducible. In
particular this representation satisfies the selection criterion (IV.1.2) and is thus
unitarily equivalent to a representation A —> pA. Obviously this representation
must be interpreted as the charge conjugate to p.
Charge conjugation entails further properties of the statistics operators
E e , E(n) ( P ) • Since ee commutes with p22t, Oee commutes with 4 and is therefore
a multiple of the identity operator if P E Al rr .

Lemma 2.3.3
Associated with each charge e E airy jz there is a number A t , called the statistics
parameter of the charge, given by

Oeee = Atli , (IV.2.51)

where P is any morphism in the class e.

One easily checks the invariance of a t i.e.

00Ee = Oe'ee` if P p'

Similarly one finds

Lemma 2.3.4
With P E Airy , [P] = e one has
,(ems.-1)
E(en) (P) = A(tn) (P) Il. (IV.2.52)

The numbers Ar(P) define, for fixed e and n, a normalized, positive linear
form over the group algebra of the permutation group S n .
 IV.2 The DHR-Analysis 169

Comment. The Een)(P) form a representation of Sn by unitary operators from

2. C 93 (no). Their complex linear combinations form a concrete C*-algebra ho-
momorphic to the abstract group algebra of Sn . The fact that cP is a positive
linear map with 0(1I) = Il implies that ) (} (P) defines a state on this C*-algebra.
From (IV.2.37) this state can be computed explicitly in terms of A 6 which shows
that it depends only on . We give only the results and refer to DHR3 for the
computation.

Lemma 2.3.5
Let Esn}, EP) denote the totally symmetric, respectively the totally antisym-
metric projectors in the group algebra of Sn . Then
A
6( 12) (En )) = n! -1 (1 + A 6 )(1 + 2)t) ... (1 + (n - 1)A 6 ), (IV.2.53)
A n)( Ean)) = n! -1
6( (1 - A)(1 - 2.À) ... (1 - (n - 1)À) . (IV.2.54)

This limits the allowed values of A6 to the set

)% = 0, ±d6 1 where d6 is a natural number. (IV.2.55)

Comment. E s and Ea are positive elements of the group algebra. For any value
of ) outside of the set (IV.2.55) either (IV.2.53) or (IV.2.54) will become neg-
ative for large n.
We shall call the integer d4 the statistics dimension of the charge. It cor-
responds precisely to the order of the parastatistics discussed in section 1. To
show this one notes first that the representation of Sn in the sector Vis quasi-
equivalent to that obtained by the GNS-construction from the state A n} .
For
)t 0 the explicit expression for this state is described in the following lemma.

Lemma 2.3.6
Let Ij be a d-dimensional Hilbert space (d finite) and en the n-fold tensor
product of [j with itself. The natural representation of Sn on en which acts by
permuting the order of factors in product vectors is denoted by 7r; the represen-
tation 7r' is defined as 7r'(P) = sign P . 7r(P). Then

en tr r(P) fo r )^^ > 0 ,

a d -f ,
-1 ^. (IV.2.56)
)471)(P) — d- tr 7r (P) for )% C 0,

The product vectors in en can be antisymmetrized in at most d factors

while there is no limitation for the symmetrization. This yields

Proposition 2.3.7
If ) > 0 the representation of S n in the sector r contains all irreducible rep-
resentations whose Young tableaux have not more than d6 rows, if a < 0 it
contains all those whose Young tableaux have not more than d6 columns.
170 IV. Charges, Global Gauge Groups and Exchange Symmetry

Thus, if the statistics parameter of a charge is positive it means that the

multiply charged states obey para-Bose statistics of order d, if it is negative they
obey para-Fermi statistics of order d. There remains the case at = 0 (or d = oo).
This case, called infinite statistics, can, however, not occur if we consider sectors
in which the translation group is implementable with positive energy spectrum
and where the charge has a conjugate charge characterized by the prop-
erty that a contains the vacuum sector (see the appendix in DHR4). Thus, in
particular, the case of infinite statistics is excluded if the charge satisfies the
BF-conditions. The proof uses an analyticity argument. We shall return to this
in the subsections 2.4, 2.5 and 3.4.
The close relation between the exclusion of infinite statistics and the exis-
tence of charge conjugation is also seen by the following direct construction of
the conjugate sector (see DHR4). Let Ell), Ean) denote the totally symmetric,
respectively totally antisymmetric projectors in the group algebra of S n and,
for 1 At 1= d-1 , d > 1
E = E(ed)(EE(d)); E' = 4d-1)(E(d-1)) if at > 0
E = E(6,d) (EV ) ); E' = e.(ed-1)(E(d-1)) if At < 0 .
In words: in the para -Bose case E and E' are the representors of the antisym-
metrizers in the representation spaces od respectively Pd-1. In the para -Fermi
case they are the representors of the symmetrizers. Then one finds that the
representation of the observable algebra Eod has "statistical dimension" 1. By
lemma 2.2.5 it is thus equivalent to a localized automorphism y. The represen-
tation E od -1 is equivalent to a morphism O' with the same statistics parameter
as o. Since 7 has an inverse one can define
p = ,- 1
(IV.2.57)
and verify that the sector p is indeed the conjugate to o.

Theorem 2.3.8
If B E dirr has finite statistics, [O] = , there exists a unique conjugate charge
= [p] characterized by the property that gro contains the vacuum sector. It is
obtained by (IV.2.57). Moreover pp- contains the vacuum sector precisely once
and
a = . (IV.2.58)
If R is an intertwining operator from the vacuum representation to -pp then
R = sign(XQ ) E(p, o)R (IV.2.59)
is an intertwining operator from the vacuum sector to gp. R may be chosen so
that
R oR = Il; R* pR = Il; R*R = R R = d(g)II . (IV.2.60)
For the proof see DHR4. The relations (IV.2.59), (IV.2.60) hold also when B
is reducible. In that case d is the sum of the dimensions of the irreducible parts
contained in O. If B is irreducible then R is uniquely determined by (IV.2.59),
(IV.2.60) up to a phase.
 IV.2 The DHR Analysis 171

W.2.4 Covariant Sectors and Energy-Momentum Spectrum

Let us assume now that the sector o is covariant i.e. that the covering group of
the Poincaré group is implementable by unitary operators Ue (g), g = (x, a) E
q3. Then the operators
Te (g) = Uo (g)Ue (g) (IV.2.6I)
are charge shifting operators as considered in (IV.2.44). Specializing to transla-
tions g = x we have
Te (x) (pA)T e (x)* = x A (IV.2.62)
where Bx denotes the "shifted" morphism

ex = as Pa -x (IV.2.63)

The key lemma 2.2.2 gives then, with U 1 = II, U2 = TQ (x), supp o C O,

ce = (pTe (x)*)Te (x) if O + x is space-like to O. (IV.2.64)

Applying the left inverse 0 we get for such x

0Te (x) = A eTT (x). (IV.2.65)

In the case of infinite statistics (a e = 0) this would give

OTe (x) = 0 for large space-like x (IV.2.66)

This relation indicates the essential reason why Ae cannot vanish. Combined
with the positivity of the energy in the representations under consideration one
finds, by analyticity arguments that it would entail the identical vanishing of
some two- point functions.
The properties of covariance and finiteness of statistics are conserved if one
takes products of morphisms, subrepresentations or conjugates (see DHR4). In
a massive theory all sectors of interest should be of this type. We list the most
relevant consequences.

Lemma 2.4.1
If P is irreducible then Ue (g) is uniquely determined and
(i) Ue(g) E (p2t) ",
(ii) if R is an intertwiner from O i to O2 then it also intertwines from Ue1 to Ue2 .

We denote the Po-spectrum in the representation B by Spect Ue . The addi-

tivity of the spectrum, discussed at the end of chapter II, now becomes

Theorem 2.4.2
(i) Spect Uoe, D Spect Ue2 + Spect Ue, •
(ii) If Bi and O2 are irreducible and B is a subrepresentation of g2Pi then
also
Spect Ue D Spect Ue1 + Spect Ue2 .
172 IV. Charges, Global Gauge Groups and Exchange Symmetry

The proof of part (i) is a simple adaptation of the proof given for the field
theoretic case in section 5 of chapter II. Part (ii) follows if one uses in addition
part (ü) of lemma 2.4.1. For details see DHR4.
One would expect that in analogy to the field theoretic case one can con-
struct a CPT-operator e, giving an antiunitary mapping between the repre-
sentations B and B and that, as a consequence, Spect Ue and Spect UT, are the
same. It must be remembered, however, that an essential ingredient in Jost's
construction of e was the assumption of fields with a finite number of compo-
nents obeying a transformation law specified in axiom D (section 1 of chapter
II). This assumption is now absent. Still, H. Epstein has shown that the es-
sential features of Jost's conclusions remain valid in a theory based on a net of
local algebras, i.e. in the frame described in section 1 of chapter III [Ep 67]. One
can construct a CPT-operator for the single particle states and for the collision
states constructed from them and one retains the spin-statistics connection. The
method of Epstein was adapted to the present analysis in DHR4. One has

Theorem 2.4.3
a) If the sector contains single particle states with mass m, spin s then the
conjugate sector contains single particle states with the same mass, spin and
multiplicity.
b) Spin-statistics connection:
The sign of the statistics parameter is (-1) 28

IV.2.5 Fields an d Collision Theory

The construction of an S-matrix in the general collision theory, described in

11.4 used charged local fields. It can be applied without change in the algebraic
setting of III.1 but its adaptation to the present analysis, where instead of a
field algebra we have localized morphisms, is less evident. The point is the dis-
tinction between states and state vectors. It is obvious how to construct states
(expectation functionals) corresponding to specific configurations of incoming or
outgoing particles. They determine transition probabilities and collision cross
sections. The S-matrix contains in addition to this also the phases of scalar
products of state vectors. For this we must use the complex linear structure
of the Hilbert spaces of state vectors in the various sectors and the "charged
fields" are a convenient tool to express this. The construction of a field algebra
from the observables and localized morphisms will be discussed in section 4.
For the purposes of collision theory a more primitive concept, called field bun-
dle suffices. We sketch this here because the structure has some interest in its
own right and is needed in cases where a field algebra cannot be constructed
(e.g. section 5) .
We have described all representations by operators acting in the same
Hilbert space 7-t o . Thus, picking a vector in 1-to determines a state on 2t only if
we also specify the representation in which it is to be understood. Let us there-
fore consider as a generalized state vector a pair { O; W} where the first member
 IV.2 The DHR-Analysis 173

is a covariant morphism and the second a vector in fo . Correspondingly we

consider a set 13 of operators acting on these generalized vectors. An element of
B isa pair 13= fg; B} with g E L and B E2t. It acts onW= Id; W}by
BW = { P'p; (Q B)W } . (IV.2.67)

This leads to the associative multiplication law within B

B 2 B1 = {P1P2; (01B2)B1} when Bk = {Pk ; Bk} (IV.2.68)

The action of q3 on these objects is then defined as

U(g){P; W} = {o, Ug (g) W} , g E q3, (IV.2.69)

ay fg; Bi- = {P; Ue(9)BUo(g) -1 } . (IV.2.70)

The description is, of course, redundant because a representation may be re-

placed by an equivalent one without changing the physical significance. This
means that

{P; w} ^' {QUA; Uw}, (IV.2.71)

{ P; B} ^' {auP; UB} (IV.2.72)

when U is a unitary from 2tloc . The reader is invited to check that the appear-
ance of UB rather than cru B on the right hand side of (IV.2.72) is no printing
mistake! One may then consider the set of generalized state vectors and B as
bundles whose base are the equivalence classes [g] and where the unitary group
in 2tloc is the structure group. One can use the equivalence relation to endow t3
with a local structure:

Definition 2.5.1
B = { p; B} E 13(7) if there exists a unitary U E 2tioc such that au p has support
in 0 and UB E 2t(0)

A conjugation is defined by

B = {o;, B} — ) Bt = {e; (QB*)R} (IV.2.73)

where R is an intertwiner from the vacuum sector to pg. Note that in the case
of infinite statistics the conjugation could not be defined and relation (IV.2.66)
would yield
(B.f2 I U(x) I BI?) = 0 (IV.2.74)
if B E B(0) and 0 + x space-like to 0.
This relation could be extended by analyticity to all x and one would arrive at
a contradiction.
With the help of these bundles the construction of state vectors for config-
urations of incoming or outgoing particles can be carried through in analogy to
II.4. For the explicit discussion and details of the algorithm concerning 13 the
reader is referred to DHR4.
174 N. Charges, Global Gauge Groups and Exchange Symmetry

Fig. IV.3.1.

IV.3 The Buchholz-Fredenhagen (BF)-Analysis

IV.3.1 Localized Single Particle States

The DHR-criterion (IV.1.2) aimed at singling out the subset of states with van-
ishing matter density at infinity within a purely massive theory. With the same
aim in mind Buchholz and Fredenhagen start from the consideration of a charge
sector in which the space-time translations are implementable by unitary oper-
ators U(x) and the energy-momentum spectrum is as pictured in fig. (IV.3.1).
It shall contain an isolated mass shell of mass m (single particle states) sepa-
rated by a gap from the remainder of the spectrum which begins at mass values
above M > m. In this situation they construct states with spectral support
on a bounded part of the single particle mass shell which may be regarded as
strongly localized in the following sense. Let rr denote the representation of the
observable algebra for this sector, 1-1 the Hilbert space on which it acts and W
the state vector of such a localized state. Then the effect of a finite translation
on W can be reproduced by the action of an almost local operator on W, or, in
terms of the infinitesimal generators (the energy-momentum operators in this
representation) by

PA W = BM W ; BA = Bµ E rr(2ta.1.)• (IV.3.1)

Here 2ta.1. denotes the almost local part of Qt i.e. the set of elements which can
be approximated by local observables in a diamond of radius r with an error
decreasing in norm faster than any inverse power of r.
To see what is involved in the claim (IV.3.1) let us first suppose that Lorentz
transformations are implementable in the sector considered. Then one can adapt

,) _
the argument of Wigner (chapter I section 3.2) and introduce for the state
vectors of single particles of mass m an improper basis I p, e), with
pp. p, C) ; 1)°
Ip = (p2+m2 ) '12 ,
p, = U (3 (p)) I o,
 IV.3 The Buchholz-Fredenhagen (BF)-Analysis 175

(see equ. (I.3.18), (I.3.20)). labels a basis in the degeneracy space Ij. A general
state vector W in HP ) can be described by a wave function W(p) or, in position
space, by
W(x) = f (p)e=p•X dap
which is determined by its values on the space-like surface x ° = 0. It may be
considered as a function of x taking values in the little Hilbert space l).
An "almost localized" state in 11(1) is naturally characterized as one for
which 11 W(x) 11E) decreases faster than any inverse power as I x 1—> oo. This
corresponds to the Newton-Wigner concept of localization for single particle
states. If W and 0 describe localized states in this sense then ( 0 i U(x) i W)
decreases fast to zero when x moves to space-like infinity. The construction of
such states is illustrated by the following argument. Let 41 , 42 be two disjoint
regions of momentum space intersecting the spectrum only on the mass shell
m. If C is an almost local operator from rr(20 with momentum transfer 4 so
that 4 1 + 4 lies in 4, 0 a state vector with momentum support in 4 1 and
W = CO then we expect that

W(x) = f K(x, x')0(x')d3x' (IV.3.2)

with an integral kernel K with values in 930) and such that 11 K(x, x') (i tends
to zero fast when either x or x' move to infinity. This must be so if we believe
that the notion of localization of Newton and Wigner agrees roughly with the
one determined by the interpretation of local algebras. In fact C, being almost
local, can produce a change essentially only in a finite neighborhood of the
origin, but it must produce a momentum change from 4 1 to 42 and so cannot
depend on the parts of 0(x) with large x. In this way we can produce an almost
localized state W from an arbitrary one by momentum space restrictions using an
almost local operator. Clearly, if W is almost localized and has bounded spectral
support then the same will be true for P'W. The last question in establishing
(IV.3.1) is therefore whether the set of almost local operators in ir(21) is rich
enough to transform any almost local state in W1) into any other such state. If
[j is finite dimensional this poses no problem.
In their analysis [Buch 82a], quoted as BF in this section, the authors dis-
pense with the assumption that the Lorentz symmetry is implementable and
they do not use the Newton-Wigner notion of localization. Their construction
of localized states follows the heuristic idea underlying (IV.3.2), starting from
a single particle state 0 and an almost local operator C with the mentioned
restrictions of their momentum space supports. Their proof that such states
satisfy (IV.3.1) uses only the shape of the spectrum as pictured in fig. (IV.3.1)
and the relation P° = (P2 +m2)1/2 on 70). In fact they derive a stronger result.
Instead of Po one may take in (IV.3.1) any smooth function of the Po. Their
proof is, however, rather technical and we do not reproduce it here.
176 IV. Charges, Global Gauge Groups and Exchange Symmetry

In the following analysis we shall assume that 7r is irreducible' and that

there is a state vector satisfying (IV.3.1).

Theorem 3.1.1
The representation it determines a pure, translationally invariant state coo (vac-
uum) by
w — lim7r(az A) = wo (A)ll; A E 2 (IV.3.3)
as x goes to space-like infinity in any direction. The GNS-representation 7r0
constructed from w 0 has a Po-spectrum consisting of the isolated point p = 0
which is a nondegenerate eigenvalue corresponding to the vacuum state and the
other part contained in p0 >0, p2> (Al - m)2.

Proof. Let A E 21(0) and 0 a diamond centered at the origin. Since the
unit ball of 93(7-O is weakly compact there exist weak limit points of 7r(a x A) as
x moves to space-like infinity. Each such limit is in 7r(21)' and hence, as it was
assumed irreducible, it is a multiple of the identity. The coefficient, denoted by
/0 (A) above, can be calculated if we take the expectation value of (IV.3.3) for
6n arbitrary state vector W E H. For W we choose a vector satisfying (IV.3.1).
Then

a, (w I r(axA) I `1`) = i ( w I [PA, r(azA)]I w)

= i (w I [BA , 7r(ayA)] I W)

Since Bu is almost local and A local, the commutator decreases fast as x moves
to space-like infinity. Integrating from x to y along some path connecting these
points we get

R`1` I r(ayA) I W) — (W I r(azA) I W) I < NO II A II , (IV.3.4)

with r = R — d where R is the minimal (space-like) Lorentz distance from the

origin to an arc connecting x and y and d is the diameter of the diamond 0. h(r)
decreases faster than any inverse power of r. Thus 7r(ax A) converges weakly for
A E 2t1oc and hence also for A E 21 and the limit is independent of the way
how x moves to infinity. (The last statement is, of course, no longer true if we
have only one space dimension. There we might have to distinguish between a
"right hand vacuum " and a "left hand vacuum", a phenomenon characteristic
for solitons).
w0 is a state on 21 since it is a positive, linear functional on 2tl oc with
w0 (1) = 1, it is translation invariant and continuous with respect to the norm
topology of 21.. Therefore we have in the representation 7r0 , arising from coo by
the GNS-construction, certainly an implementation of a r by unitary operators
U0(x) (see equ. (II1.3.14)).
The additivity of the energy-momentum spectrum discussed in II.5 general-
izes to a relation between the spectra of U0 in 7ro and U in 7r. Suppose 0 E 7-1 has
1 This is done only to simplify some arguments. In BF 7 is only assumed to be primary.
 IV.3 The Buchholz-Fredenhagen (BF)-Analysis 177

spectral support (with respect to U) in the momentum space region A and f is

a test function with momentum space support A, fast decreasing in x-space,
A E 2110c , A(f) = f f (x)a x Ad 4x. Then ir(A(f )) I ) has spectral support in
A + A and the same is true for 7r(ay A(f )) I .T.) with arbitrary y. The norm
square of this vector converges to w o (A(f)*A(f)) for I y 1--; oo which does not
vanish for all A if 41 E Spect U0 . Therefore

Spect U D Spect U + Spect Uo . (IV.3.5)

Given Spect U as in fig. (IV.3.1) the spectrum of U0 can therefore consist only
of the point p = 0 and momentum vectors with positive energy and mass above
(M — m). So pro is a vacuum representation with mass gap of (at least) M — m.
Next one shows that w© has the cluster property

wo (A 2 ax A 1 ) --^^ wo (A 2 )wo (A 1 ) as IxI - (I V. 3.6)

Consider the function

F(x, y) = (w I 7r (ay(A2axAi)) I W) (IV.3.7)

for A1 , A2 E 2t(0), y and x space-like to the origin. Take the radial straight line
from y to infinity and its parallel beginning at x + y. The minimal space-like
distance from these lines to the origin shall be R. Then, by the same argument
leading to (IV.3.4) we get

IF(x, y) — wo(A2axA1), < h(r) II A2 I I II Al II (IV.3.8)

where h is a fast decreasing function of r = R — d, independent of A 1 , A2 in

2t(0). On the other hand

lip F(x, y) = (W I 7r(a y A2) I W) wo(Al) 1 ,'„,. wo(A2)wo(A1) • (IV.3.9)

The cluster property of w o implies that wo is the only translation invariant

state in the folium of pr o , so the GNS-vector Q0 is a non-degenerate eigenstate
of Po. Since, by theorem III.3.2.4, U0 (x) is in the weak closure of 7r o (21) and Q0
o the commutant of 7ro (2t) must be trivial i.e. pr o is irreducible, wo iscylforp
is a pure state. ❑

IV.3.2 BF Topological Charges

Consider in a space-like plane the region C (the shaded region in fig. (IV.3.2)).
Starting with a ball around the origin with radius r we draw a straight line
from the origin to infinity and, around the point at distance r' from the ori-
gin on this line we take a ball with radius r + ryr'; ry > O. C is the union of
all these balls for 0 < r' < oo. We denote by 21 ° (C) the relative commutant of
2t(C) i.e. the set of all elements of 91 which commute with every element of 21(C).
 ,oreittill
178 IV. Charges, Global Gauge Groups and Exchange Symmetry

Fig. 1V.3.2.

Lemma 3.2.1
Let w(A) = (W 1 ir(A) W) with W satisfying (W.3.1), w o as defined in theorem
3.1.1 and C as described above. Then the restrictions of the states w, w o to the
relative commutant 21.'(C) satisfy

11P — WO) IT fl C (îV.3.10)

where e does not depend on the direction of the center line of C and vanishes
fast as r —* oo.

[Link] W = U(x)W, By (x) = U(x)A,,U*(x) one obtains from (IV.3.1)

P4,Wx = Bp (x)WW

and from theorem 3.1.1, with e denoting the unit vector in direction of the
center line of C

ôx' (wx' 7r(A) 1 tfry') dr'

^(w —wo)(A) =
Jo

f ( We I [et`Bp(x'), 7r(A)l I We) dr'

with s' = (0, r'e); e = (0, e).
ePBi.,,(x') can be approximated by an element of the algebra of a diamond with
center s' and diameter d' up to an error with norm f (d) where f is a fast
decreasing function. For A E 2ic (C), choosing d' = r + ryr', this implies that the
commutator under the integral above has norm less than 2 II A I! f (r + 'yr').
Thus

1(w — wo)(A)I < II A I f f(r+?r')dr'

< II A I g(r, ,
where g is fast decreasing with r.
 IV.3 The Buchholz - Fredenhagen (BF) - Analysis 179

Definition 3.2.2
Let 0 be a diamond space-like separated from the origin and a E IR 4 an arbi-
trary point. The region

S =a +U), AO; 0 < .1 <oo (IV.3.12)

is called a space-like cone with apex a.

Theorem 3.2.3
Let 7r be a representation of 2t as described above 2 , acting in the Hilbert space
1-1, and (7ro , ft0) the corresponding vacuum representation (according to theo-
rem 3.1.1). Then, for any space-like cone S one has a unitary mapping V from
?-t onto 'Ho such that
V7r(A) = iro (A) V for A E %c (S) . (IV.3.13)

In other words: the restrictions of 7r and 7ro to the relative commutant of the
algebra of any space-like cone are unitarily equivalent.

Proof. If the theorem is true for one position of the apex it will be true for
any apex position since the translations are implementable in the representation
7r. The region C in lemma 3.2.1 is contained in a space-like cone S with apex
a = —ry -l re and central axis along e. Therefore lemma 3.2.1 says that for arbi-
trarily small e > 0 we can choose r sufficiently large so that there is a state w in
the folium of 7r which in restriction to 2te (S) approximates w 0 with an error of
norm less than E. Since the folium of states in a representation is complete in the
norm topology the folium of 7r contains a state w which in restriction to 2t°(S)
coincides with w0 . One shows that under the prevailing circumstances there is
a vector ,f2 E 1-t inducing the state w on 2tC(S) and cyclic for this subalgebra,
i.e.

(S2 7r(A) I Sl) = w(A) = wo (A) for A E 2t° (S) (IV.3.14)

7r(` c (S))b = i-t . (IV.3.15)

Accepting this, the intertwining operator in (IV.3.13) is defined by

V7r(A)S = 7r0 (A)S2 for A E 2tc(S) , (IV.3.16)

where Q is the vacuum state vector in n0. To show that a vector Sl with
the properties (IV.3.14), (IV.3.15) exists and that (IV.3.16) gives a consistent
definition of an operator V mapping 1-t onto 1 -t0 one may use the following spe-
cialization of [Sakai 1971, theorem 2.7.97):

2 The PA-spectrum shall be as in fig. (IV.3.1) and, for simplicity, we also assume
irreducibility.
180 IV . Charges, Global Gauge Groups and Exchange Symmetry

Theorem 3.2.4
Let N be a von Neumann algebra acting on a Hilbert space 7-t, c7.) a normal state
on N. If there exists a vector W E 7-t which is cyclic for N' then there is a vector
Sl E 7-t inducing w. If w is faithful on N and W cyclic for N then ( is cyclic.

For our application N = 7r (2tC (S) )" and W may be taken as the vector
from which the construction started. It was chosen to have bounded spectral
support and is therefore cyclic for both N and N' due to the (generalized)
Reeh-Schlieder theorem. By the same theorem Q is cyclic for 7r o (QLc(S)) and by
a slight variant of this theorem one shows that CO is faithful on N.
Replacing A by AB in (IV.3.16) one sees that V satisfies (IV.3.13). The
isometry of V follows from (IV.3.14).
This completes the proof for the case of cones with an axis e having van-
ishing time component. But lemma 3.2.1 holds in any Lorentz frame. ❑

Comment. Lemma 3.2.1 comes close to the DHR-selection criterion (IV.1.3)

since we can cover the outside of a ball of radius r by the complements of two
regions CC of the shape depicted in fig.(IV.3.2). The step from lemma 3.2.1 to
theorem 3.2.3 is analogous to the one from (IV.1.3) to (IV.1.2). It is therefore
at first sight surprising that one cannot infer (IV.1.3) from lemma 3.2.1 using
additivity (III.4.8) for the local algebras and the cluster property for the states
which, in a massive theory, demands a fast decrease of correlations with increas-
ing distance. The following rudimentary model may show how the difference
between lemma 3.2.1 and equ. (IV.1.3) can arise due to homotopy properties
of regions. Consider an Abelian subnet on 3-dimensional space at a sharp time.
The net shall be generated by unitaries V(f) where f is a real, 3-vector-valued
function of x E 1R3 . V(f) is assigned to the algebra of the support region of f
and
V(fl + f2 ) = V(fl )V(f2 ) . 3 (IV.3.17)
This guarantees additivity because, given an open covering of the support region
of f, we can write f as a sum of functions f with supports in the subregions and
then apply (IV.3.17). Now let

wa (V(f)) =
e-(f'Kf) (IV.3.18)
,

w (V(f)) = e i(h'f) wo (V(f)) , (IV.3.19)

where h is a fixed function with curl h = 0 and the support of div h is contained
in the ball I x l< ro . We may interpret h as the electric field of an external charge
with density A = div h.
The linear combinations of elements V(f) with complex coefficients form a
*-algebra and co o and w define states on it if K is a positive operator in the test
function space. The states have the cluster property if in

(f ,K f) - f f(x')K(x' — x)f(x)d3x'd3x (IV.3.20)

3 We may put V(f) = e:E(f) and consider E as an analogue to the electric field strength.
 IV.3 The Buchholz-Fredenhagen (BF)-Analysis 181

the kernel K is fast decreasing with I x' - - x I. Now (h, f) = 0 if supp f is in the
complement of any cone containing the support of div h. One sees this most
easily if one decomposes f into a longitudinal and a transversal part:

f= grad cp+ curl g. (IV.3.21)

Then, since h is curl free

(h,f)=— f cp div h d3x (IV.3.22)

where cp is given as the solution of the potential equation

dcp = div f

with the boundary condition that ça vanishes at infinity. If the support of div h
(the ball I x l< ro ) is connected to infinity by paths avoiding the support of
div f then cp = 0 in the support region of div h and hence (h, f) = O. Therefore
w and wo coincide on V (f) whenever supp f is in the complement of an arbitrary
cone containing the support of div h. On the other hand let

f(x)= a (IxI) x (IV.3.23)

with a having support in the spherical shell r 1 <I x r2 with r i > ro . Then,
for r < r 1 , ça is a nonvanishing constant cpo = — f a(r)rdr and

w (V(f)) = e -14 `POw0 (V(f)) ; q = f div h d3x . (IV.3.24)

We can choose ri arbitrarily large keeping the value of cpo fixed. So there are
elements in the observable algebra of a spherical shell of arbitrarily large inner
radius for which w© and w differ markedly. If we decompose f in (IV.3.23)
smoothly
f =fi +f2
where fi has its support limited to polar angles T9 < 01 and f2 to polar angles
> '02 (02 < 19i) then

w (V(f,)) = w o (V(fi )) , i = 1, 2 ,

but Iwo (V(f)) — w o (V(f1 )) wo (V(f2 ))I need not be small in spite of the fact that
f1 and f2 have only a small overlap and K(x' — x) is fast decreasing with in-
creasing distance. Suppose as a typical example

(f , K f) = (curl f) 2 d3x
J
Then, for f given by (IV.3.23), wo(V(f)) = 1 because f is curl free. But fi arising
from f by multiplication with a function of 19 which decreases from 1 to 0 in the
interval 192 < z9 < t91, is not curl free. In fact, the closer one takes Vi to'0 2 the
larger becomes (f , K f). Thus, no matter how one makes the decomposition,
w0(V(fi)) < 1.
182 IV. Charges, Global Gauge Groups and Exchange Symmetry

IV.3.3 Composition of Sectors and Exchange Symmetry

Again, as in section 2, it is convenient to replace the representation 7r by an

equivalent representation acting in 7 -t0. Choosing an arbitrary space like cone
S we map '1-t onto 7-to by the unitary operator V of (IV.3.16) and define the
representation A —> QA E B(74) by

QA = Vr(A)V*, A E 2< . (IV.3.25)

Again we may consider the vacuum representation as the defining representation

and omit the symbol 70. The map 0 from 2l to ! (7-to) acts trivially on % c (S)
and, in fact, the triviality extends to the weak closure 21.(S)". Thus 0 maps
R.(S) = 2l(5)" into itself and one may say that 0 is "localized in S". For general
elements A E 2i the image QA will, however, not necessarily belong to 2t. Thus 0
need not be a morphism of 2t and therefore the composition 0201 is not directly
defined because 91A may lie outside the domain of definition of 02. Now, for
A E 2(0), QA will be in the commutant of 2l(S'n0') because, if 01 is space-like
to both S and 0 and B E 2t(01) then

BQA = (0B)(0A) = QBA = QAB = (0A)(0B) = (eA)B .

If we take an element A E 21(0) and two mappings 0k localized respectively
in Sk so that S2 is space-like to both Si and 0 then 0201A = 01A because
Q1 A is in 2t(Sl. n 0')' which is contained in 21(82)' where 02 acts trivially. To
define a composition of two maps 0k which are localized in cones with arbitrary
mutual placement one introduces a suitably placed auxiliary cone. In BF the
auxiliary cone is used to define an extended algebra which is stable under the
action of the maps so that the ek become morphisms of the extended algebra.
A somewhat simpler way to define the composition product was described by
Doplicher and Roberts [Dopl 90). Let Si be the support of Ui and A E 240).
Choose an auxiliary cone with support in S3, space-like to both Si and 0 Then
there is a representation o3 with support in 53 and equivalent to 9 2 with unitary
intertwiner V3:
02B = V3 ( 3B)V; B E 2l .
Since we have
Q3oiA = 01A for A E 24(0)
this suggests the definition

0201A = V3 (0i A)V3 ; A E 21(0) . (IV.3.26)

The location of the auxiliary cone has to vary as the support of A varies. One
shows, however, that (IV.3.26) defines a representation depending only on 01, 02,
independent of the choice of 0 and V3 within the specified limitations.
Once the composition of sectors is defined the analysis of exchange symme-
try, the construction of a field bundle and of collision states proceeds in analogy
to DHR (see subsection 2.5). The discussion becomes somewhat more tedious
 IV.3 The Buchholz-Fredenhagen (BF)-Analysis 183

because one has to verify the independence from the choice of the auxiliary
cone at all steps. We shall not repeat it here but refer to BF and [Dopl 90]. In
a theory based on Minkowski space the conclusions remain the same as those
described in section 2. It is, however, already intuitively clear that now all 4
dimensions are needed since the morphisms are now homotopically no longer
points but half lines extending to infinity and one needs three such lines for the
construction of E. So one needs at least three space-like dimensions to effect the
interchange of two mutually space-like morphisms by continuous shifts without
entering into time-like configurations. Thus the relevance of the braid group
instead of the permutation group (section 5) begins already in 3-dimensional
space-time for BF topological charges.

W.3.4 Charge Conjugation and Absence of Infinite Statistics

Let S be the support cone of A and e a space-like direction so that S + )e

becomes space-like to every diamond as A — +00. We have

Ue (x) (QA)U; (x) = c A, A E 2l .

Consider now the sequence Ue (x)AUp (x), A E 2 with x = ae, A –4 +oo. (Note
that A = iro (A) E 3(74)). Due to the weak compactness of the unit ball of
93(710 ) it has weak limit points in 93(7 -10). Each such limit is in the commutant
of 02t. To see this take A and B E 2t10 . Then

[uQ (x)Au;(x), QB] = UQ(x) [A, oo_ X B] U; (x) .

But o_x B moves out of the support of A as A –' oo. Then g acts trivially and
the commutator vanishes for large enough A. Since A was assumed irreducible
the limit points are multiples of the identity and we can evaluate them by taking
the expectation value in any state. For this we choose the state vector W which
satisfies (IV.3.1). Note that with the present notation PA is the generator of
Up (x) and Bu in (IV.3.1) should be written as pBt, with Bµ E a.l. The same
argument as used in the proof of theorem 3.1.1 shows then that the sequence
converges weakly and one has

Lemma 3.4.1
With e as above

w -- lim Ue (.1e)AU^(.1e) = 7)(A)11; A E % (IV.3.27)

The state 7 satisfies

OW) = (.)0 (A). (IV.3.28)

The relation ([V.3.28) follows directly from theorem 3.1.1. It shows that
is a state in the conjugate sector.
184 IV. Charges, Global Gauge Groups and Exchange Symmetry

We can also construct a left inverse of g. Defining the charge shifting oper-
ator
Te (x) = Uo(x)UU(x) (IV.3.29)
(compare (IV.2.61), (IV.2.62)) we get for large A

(CQO I crTQp, e )A I BS2o/ _ { 0o I (a---aeC * )Ue (ae)AUp(ae)(a--aeB) I 00)

= (QO Ue(ae)(AC*)A(OB)Ue(ae) I S 2oi
A—>oo
^ ((gC*)ApB).
We have used that the vacuum vector S20 is invariant under Uo(x), that a---ae2ttoc
moves outside the support of o so that we can insert the action of A on the re-
spective elements, and finally (IV.3.28). This gives

Lemma 3.4.2

— li m' crTe ^^1e) A = ^(A); A E 2t,

w (IV.3.30)
À 00
(Cflo I 0 (A) I B,f2o) = W ((eC*)AoB) . (IV.3.31)
4;1 is a left inverse of e.

One can use the left inverse cP to define the conjugate sector, the statis-
tics parameter A e (e.e) and show, as in section 2 that the possible values of
Ae are limited to 0, ±d -1 . The final aim is to exclude Ag = 0 (infinite statistics).
This was achieved by Fredenhagen [Fred 814 He shows that A t, = 0 implies
(IV.2.74) for a non void open set of points x E .A4 and that one has sufficient
analyticity to conclude that (IV.2.74) must hold then for all x. This is in conflict
with the positivity of the Hilbert space metric.
A proof of the spin-statistics connection in the case of BF-charges has been
given by Buchholz and Epstein [Buch 85] .

IV.4 Global Gauge Group and Charge Carrying Fields

In the traditional field theoretic approach (which in algebraic guise was summa-
rized in III.1) we deal with a Hilbert space x(u) which contains all superselec-
tion sectors as subspaces (the index u was added to indicate "universal") and
we deal with a larger algebra a (denoted by A in III.1) acting irreducibly on
i-hu) . It contains besides the observables also operators mapping from one co-
herent subspace to a different one ("charge carrying fields"). Furthermore there
is a faithful, strongly continuous representation of a compact group, the global
gauge group Ç, by unitary operators in 93(7 -(u) ), singling out the observables in
a as the invariant elements under the action induced by rr(u)(Ç). In other words
rr(u) (2t) = a n rr (u) (CJ)' . (IV.4.1)
 IV.4 Global Gauge Group and Charge Carrying Fields 185

This poses the question as to whether Ç and the algebraic structure of a

are determined in a natural fashion by the physical content of the theory i.e. by
2l and its superselection structure as described in the last sections. There are
several facets to this question.

Implementation of Endomorphisms. If we start from a representation 7r(2t)

in a Hilbert space 7-t in which each sector occurs at least once (we omit now the
index (u) on 7r and 7-0 then, for g E 4(0) there will exist intertwining operators
V between 7r(2t) and 7r(4):
zinr(A) = 7r(PA)71i; VJ*7r(PA) = 7r(A)0*. (IV.4.2)

We may choose V) as a partial isometry, an isometric mapping from a subspace

E17-t onto a subspace E27-( where the projectors Ei are given by

El = ***; E2 = . (IV.4.3)

The subspaces must be chosen so that each equivalence class of subrepresenta-

tions of 7r(2t)Ei appears in 7r(g2t)E 2 with the same multiplicity. If the operators
V) and z/b' both satisfy (IV.4.2) then we have

V*071- (A) = ir(A)0 *0;! i.e. 1//*0 E 7r(2t)`. (IV.4.4)

Now the commutant 7r(2t)' contains only global quantities, not referring to any
particular region. On the other hand, if g E 4(0) then gA = A for A E 21(0').
Hence V/ and 7/1' commute with ir (2t(0')). This suggests that we associate V) and
z/i' with a field algebra of the region 0;

E Ç(Q); 5(0) c 7r(21.(0'))'. (IV.4.5)

We do not want a(0) to contain any non trivial global quantities. Since z/'*//i E
(0) n 7r(2t)' we should demand
o '* 0 = cIl. (IV.4.6)

The set of operators satisfying (IV.4.2) and (IV.4.6) forms a linear space in
B(7-0. It is in fact a Hilbert space (of operators) since we can define a numerical
sesquilinear positive definite scalar product (V/i' 1 0) by

O'*0 = ( kb ' I 0} Il. (IV.4.7)

Let us denote this Hilbert space by 1) and assume that it has finite dimension
d. The elements of unit length are isometries i.e. their source projector E l = Il.
We may choose an orthonormal basis z/'k in 1):

z/1a z/Jk = Sik l. i k = 1, ... d

, (IV.4.8)

If (and only if) the range projectors V407, add up to Il, i.e.
186 IV. Charges, Global Gauge Groups and Exchange Symmetry

E okok = Il (IV.4.9)

we can implement the endomorphism A in the representation 7r and obtain from

(Iv.4.2)
7(4) = E iper(2)0k• (IV.4.10)
The relations (IV.4.8), (IV.4.9) may be considered as the defining relations of
an abstract *-algebra generated by the bk, 7111. Cuntz has shown that there is a
unique C*-norm on this algebra and that the completion of the algebra in this
norm yields a simple C*-algebra, the Cuntz algebra Od.
One checks that with V E 1l and C E Od

= (abC)0 , (IV.4.11)

where ch is the "inner" endomorphism of Od

C E Os—i abC= E z1)kCzPk. (IV.4.12)

Note that ah depends only on Ij, not on the choice of the basis bk. We call o- h
Od. Thus ah extends the endomorphism p to thethecanoildmrpsf
part of a(0) consisting of the representation of the Cuntz algebra connected
with p. One anticipates that the dimension d of h must be the statistics dimen-
sion of p. To show this one needs the full machinery involving the conjugate
charge and permutation operators (see theorem 4.2 below).

Charges with d = 1. Here the charges form an Abelian group, the conju-
gate charge being the inverse. For each charge 4 - we choose some morphism p
in its class and a representation 7r4 of 91 acting in a Hilbert space 7-te, equivalent
to pO. in 7-t0 . So we have a unitary map from 7 -lo to lie with

7r4 (A) = V4 (p4A)Vj .

It appears natural to form the direct sums

,.,1(u) = ir (u) =
® 7-l; E® 7r (IV.4.13)
g

Then each sector appears in 7-t(u ) exactly once. Since each ir k is irreducible
the commutant of 71-(u)(21.) reduces to multiples of the identity on each 1-4 and
thus consists of multiplication operators by functions F(4). A basis system of
functions, in terms of which general functions may be expanded, is provided by
the characters of the group (harmonic analysis). We recall that a character may
be defined as a 1-dimensional, unitary representation i.e. it is a function XO
with values in T (the unit circle in the complex plane) with the property

X(e2)X(el) = X(6261) • (P1.4.14)

The set of characters of an Abelian group forms itself a group, the so called
"dual group", by
 IV.4 Global Gauge Group and Charge Carrying Fields 187

-1
(x 2x 1) ( e) = x 2(e )x l( e)+ x-1 ( e) = (x (0 ) • (IV.4.15)

We denote it by g. It is the (global) gauge group. By the Pontrjagin duality

theorem the charge group is then the dual group of G. The representation of G
in x(u) is
7r(u) (x)W = x(e)W for W E . (IV.4.16)
Next we construct unitary operators in 13(71(u)) which implement the au-
tomorphisms o E d. Let [o] = C then

i(u) (QA) Inv = 1'eVkret(owA)Veti (W.4.17)

where W is an intertwining operator between oeo and oet

off , o = oeoW . (IV.4.18)

Introducing an operator which lowers the charge by C and acts on lie by

^a I nE, = Vf' Vjerf'f ( W ) (IV.4.19)

we have
71-(u) (eA) = 0 127r (u) (A) ; . (IV.4.20)
The adjoint (and inverse) - raises the charge by C.
The computation of W for arbitrary o and C' and hence the commutation
properties of 00, 4 depend on some conventions. We refer to DHR2 for a gen-
eral discussion of the case of charges with d =1. The result is that there remains
some freedom but that it is always possible to construct a field algebra with
normal commutation relations i.e. such that at space-like distances two fields
carrying fermionic charges (a = —1) anticommute, a field carrying bosonic
charge commutes with all other fields.

Endomorphisms and non Abelian Gauge Group. New aspects occur when
the charge structure corresponds to localized endomorphisms, the general situ-
ation described in section 2.
It was shown there that o E drr leads to parastatistics of some finite order
d in the multiply charged sectors on. The qualitative discussion of section 1
suggests that there is an alternative description in which the net of observable
algebras is embedded in a larger net Qt which still has causal commutativity
but has a non Abelian internal symmetry group go realized by automorphism
g E Go — a 9 E Aut 2t 1 so that the observables are the invariant elements of I/
under a9 . It is then a matter of taste whether one regards It as as the "true"
algebra of observables and Go as an internal symmetry or 2. as the observable
algebra and go as (part of) the global gauge group. The latter point of view is
more appropriate if the symmetry is not broken. The question is now whether
the structure encountered in section 2 can always be interpreted in this way.
li.e. ag transforms each 21(0) into itself.
188 IV. Charges, Global Gauge Groups and Exchange Symmetry

This would mean that the superselection structure determines a group such
that the pure charges C E Airy/T are in one-to-one correspondence with the
equivalence classes of irreducible representations of the group, the composition
e2ei corresponds to the tensor product of the group representations, charge
conjugation to the complex conjugate representation and the decomposition of
WI. into irreducibles mirrors the Clebsch-Gordan decomposition of the tensor
product. In short: the structure of the semigroup d/Z should be recognizable
as the dual object of a group.
To say simply that the answer is affirmative would not do justice to the work
invested by Doplicher and Roberts in the study of this question and the insights
gained through this investigation. It has added a new chapter to the mathemat-
ical theory of group duality. In particular it has shown that a semigroup of
endomorphisms of a C*-algebra together with its intertwiners, possessing pre-
cisely all the properties elaborated in section 2 (apart from the sign of the
statistics parameter), may be regarded in a natural fashion as the (abstract)
dual of a compact group. Further that a concrete group dual is obtained by
embedding the structure in a larger algebra (the algebra 2f. mentioned above or,
ultimately, the field algebra).
It is beyond the scope of this book to present a full account of this work.
I shall outline the central ideas and state (without proof) results. This may
facilitate the study of the original papers for the interested reader.
The natural language is provided by category theory. This need not be a de-
terrent for a theoretical physicist of our days. The basic concepts are simple and
natural. A category consists of "objects" carrying some mathematical structure
(e.g. algebras, representations of a group ...) and "arrows" between objects
corresponding to maps conserving the structure. In the examples relevant here
the objects are Hilbert spaces (possibly equipped with a representation of a
group or algebra). This implies that the set of arrows from object a to object
(3, which will be denoted by i(a, (3) 2 , is a Banach space: linear combinations
of arrows in this set are naturally defined and so is a norm on these arrows.
In any category one has a composition of arrows provided that the target of
the first coincides with the source of the second. If S E i(a, f3), S' E 0, ry)
then S' o S E i(a, ry). Here ll S' o S 1111 S' 1111 S [t; one has an adjoint
S* E O, a) and 11 S* 11=11 S 11,11 S* o S 11-11 S 11 2 . This justifies the name
C*-category for such a structure. It stands in the same relation to a C*-algebra
as a groupoid to a group (the algebraic operations being only defined if the
elements fit together). For each object a the arrows i(a, a) form a C*-algebra.
Its unit will be denoted by ll„. This allows the definition of subobjects and
direct sums of objects. The category is said to have subobjects if, for every pro-
jector E E i(a, a) there is an object ry (called a subobject of a) and an arrow
S E i(a, ry) such that S o S* = 1L) , S* o S = E. The direct sum of two objects
a, 0 is defined if there is an object ry (called the direct sum of a and 0) such
that one has arrows V E i(ry, a), W E i(ry, (3) with V o V* = Ila , W o W* = Ilp
2 In order to conform with standard usage we have now written the source to the left, the
target to the right in the bracket. The opposite convention was used in section 2.
 IV.4 Global Gauge Group and Charge Carrying Fields 189

and V* o V + W* o W = 17 . The C*-categories we deal with here are closed

under direct sums and subobjects.
Two such categories concern us in our context. One is the representation
theory of a compact Lie group G, the "dual of ç". We may take as objects finite
dimensional unitary representations of G and as arrows the homomorphisms
between them. It suffices if the set of objects contains at least one sample of
each equivalence class of representations. We denote this category by Tg. The
other category is the superselection structure discussed in section 2. The objects
are the localized morphisms from d, the arrows are the intertwiners. We denote
it byT21 .
In both cases the categories have the following additional structure. There
is a product of objects which entails an (associative) product of arrows. In Tg
this is the tensor product of representations accompanied by the tensor product
of homomorphisms. In T it is the product of morphisms accompanied by the
cross product of intertwiners. The relations (IV.2.16), (IV.2.17) hold in both
cases. There is a unit object, denoted by t. It is the trivial representation of
G (respectively the vacuum representation or the identity morphism of 2(.). A
category with such a product structure is called a strict monoidal category. The
categories are also symmetric. To each pair of objects a, 0 there is a unitary
e(a, 0) E i(a[3, Oa) interchanging the order of factors of arrows such that for
S E i(a, f3), S' E i(a', (3') one has

E(/3', [3) o (S' x S) = (S x S') o E(a', a) . (IV.4.21)

This leads, as discussed in section 2 to a unitary representation of the permu-

tation group Sn in the algebra i(an, an) for each a.
The final element of structure is the existence of conjugates. To each object
a one can pick an object â and arrows

R E i(t, cxa), R= E (Cx, a) o R E i(t, arc) , (IV.4.22)

with3

(R* x1„)o (11 x R) = 11 ; (R*x IN) o(1l xR)=1l . (IV.4.23)

In the categories Tg and TT the algebra i(t, t) reduces to C (the complex

numbers) . For TT this follows from the irreducibility of the vacuum represen-
tation. Let us now call, for short, a category with all the structure mentioned
above a DR-category. It is a strict monoidal C*-category, closed under subob-
jects and direct sums, symmetric and with conjugates and we shall also require
3 Note that the relations (IV.4.23) differ by the sign of the statistics parameter from
(IV.2.59). This finds its natural expression by saying that T should be regarded as a "Z 2
gradectoy”[Dpl8).Inkeigwthobjcvmenidathbgof
the section we may also concern ourselves first with the subcategory formed by the bosonic
charges and construct go and 2L. This algebra will still have localized automorphisms con-
taining a charge group Z2; in a second step one can then apply the methods for the case
d = 1 to construct the full gauge group G and the field algebra -a. We shall not take up the
Bose-Fermi problem here again.
190 IV. Charges, Global Gauge Groups and Exchange Symmetry

= C. We list some important properties of DR-categories. The proofs may

i(t, t)
be found in [Dopl 89b].

Lemma 4.1
The conjugate is defined up to unitary equivalence i.e. if â 1 and â2 are possi-
ble choices for a conjugate to a and R1, R2 are the respective intertwiners of
(IV.4.22), (IV.4.23) then there is a unitary intertwiner U E i(ce1, GY2) so that

R2=(Ux1 ck )oR1.

Theorem 4.2
There is a dimension function on the objects a — d(a) defined by

d(a) = R* o R E i(t, t)

where R is given by (IV.4.22), (IV.4.23) and d(a) is independent of the choice

of â (due to lemma 4.1). It satisfies

(i) d(ai a2 ) = d(a1)d(a2)

(ii) d(a2 ® a1 ) = d(a2 ) + d(a1 ),

(iii) the possible values of d(a) are the natural numbers.

Definition 4.3
The subobject of ad(cc) corresponding to the completely antisymmetric subspace
will be called the determinant of a and denoted by det a. The objects with de-
terminant t will be called special objects.

Theorem 4.4
(i) det a has dimension 1.

(ii) det (a ®â} = t.

(iii) Every finite set {a 1i ... , an } of objects is "dominated" by a special object,

i.e. all ak (k = 1, ... n) are subobjects of some a with det a = t.
Consider now the symmetric, monoidal C*-category whose objects are the
tensor powers of a d-dimensional Hilbert space lj d (not equipped with any ad-
ditional structure) and their subspaces. The arrows are the isometrics between
these spaces (or, more conveniently, all linear maps between them) . It is ob-
vious how this can be embedded in the Cuntz algebra. f)d corresponds to the
basic Hilbert space fj in Od; 14n to the linear span of products of n elements
E lj C Od ; i(lj® n+k ,®n ) to the linear span of monomials with l + k factors
0 on the left and l factors 0* on the right where l < n, l + k > 0 and k E Z. In
the last correspondence we have identified arrows T and T x Ilhd with the same
element of the Cuntz algebra; then tensoring on the left with 1I corresponds
to the canonical endomorphism of Od.
 IV.4 Glob al Gauge Group and Charge Carrying Fields 191

If Ç is a compact Lie group it has a faithful unitary representation g E Ç —^^

u(g) with determinant 1 in some d-dimensional space and such that every (finite
dimensional) representation of G is equivalent to one contained in some tensor
power of u. The category generated by u (the objects being the tensor powers of
u, their subrepresentations and direct 'sums, with the trivial representation as
the element t; the arrows being homomorphisms of these representations) may
be regarded as a dual of g. We denote it by Tg . It is a DR category in which u
-

is a special object and it is naturally embedded in the Cuntz algebra, equipped .

now with an automorphism group g E Ç ^^ ag E Aut Od where ag is defined

—

by its action on b
agni = u(g)h . (IV.4.24)
The arrows correspond to invariant elements of Od under the action of ag .

Theorem 4.5
The DR-category To generated by a single object P of dimension d and determi-
nant t is isomorphic to Tg for some compact Lie group Ç. Both can be naturally
embedded in Od , equipped with an automorphism group ag (g E Ç) as de-
scribed in (IV.4.24) such that the arrows in To are mapped onto a subalgebra
Og C Od , the fixed points under ag in Od. Let T E Og be the image of the
arrow T under this map then T is also the image of T x 1l whereas 11 0 x T is
mapped on u T (see (IV.4.12)).

Returning to the original problem: we have the observable algebra 21, the set
d of its localized morphisms and the intertwiners, implemented by elements of
21. Considering first the bosonic morphisms dbos (positive statistics parameter)
one has

Theorem 4.6
The structure (21, abos ) can be embedded in a C*-algebra with causal net
structure and equipped with an automorphism group Ç0 such that to each
p E dbos there is a subspace be of isometries in 21 (basis 01, ... , Od) imple-
menting p by
pA = E
qpkAok, A E 2l , (IV.4.25)
and 21. consists of the invariant elements of 2l under the automorphisms a g E go .
If the semigroup of charges is generated by a finite set then go is isomorphic
to a compact Lie group. 4 For the full set (21, d) one can find an embedding
in a C*-algebra .3 with normal Bose-Fermi causal net structure and a compact
automorphism group G such that 21 is the fixed point subalgebra of 3 under a9
and (IV.4.25) holds with i,bk E a.
References The survey given in this section is based on the original papers
[Dopl 88], [Dopl 89a], [Dopl 89b], [Dopl 90].
4ç need not be connected and it may possibly be zero dimensional i.e. discrete.
192 IV. Charges, Global Gauge Groups and Exchange Symmetry

IV.5 Low Dimensional Space-Time

and Braid Group Statistics

Let us follow the steps of the DHR analysis for a quantum field theory in 2-
dimensional space-time. The results of subsection 2.1 remain unchanged but
lemma 2.2.1 and the key lemma 2.2.2 are modified. Consider the set of ordered
pairs (01, 02 ) with space-like separation. If the dimension of space-time is larger
than 2 then this set is connected; we can continuously shift a space-like con-
figuration of two points to any other such configuration without crossing their
causal influence zone. In 2-dimensional space-time we have two disconnected
components. If 0 1 lies to the left of 02 then 01 (s) will have to remain on the
left of O2(s) for any continuous family of pairs Ok(s) with space-like separation.
In the proof of lemma 2.2.1 we used the possibility of continuously moving the
supports of the pair (g i , 02) to the supports of the pair (0i, 02). In 2-dimensional
space-time this is only possible if these (ordered) pairs of supports lie in the
same connectivity component. The consequence for the key lemma is that e a is
not completely independent of the choice of the morphisms 01, 02 but we may
obtain two different operators s o , depending on whether in the construction
(IV.2.27), (IV.2.28) we choose the support of 01 to the left of the support of 02
or to the right. Let us adopt the convention of defining EQ by the first mentioned
choice of the supports. Then one finds that the opposite choice leads to X81 in
(IV.2.28). The relation (IV.2.30) is lost. Therefore ee does not correspond to a
permutation of two elements but generates a braiding of two strands. An illus-
tration is afforded by two strands of hair of a young lady. If they are originally
parallel then the position of the loose end points may be interchanged in two
inequivalent fashions, rotating by 180° around the center line in a clockwise or
in an' anticlockwise sense. E , corresponds to the one, E 1 to the other operation.
Repetition of one of the procedures does not lead back to the original situation
but to the beginning of a braid.
The braid group Bn for n strands is generated by such operations on neigh-
boring strands. Let Qk (k = 1, ... , n — 1) denote the interchange of the ends
of the k-th and (k + 1)-th strands by clockwise rotation. Then, as Artin has
shown, the only independent relations which the uk will satisfy are

QkQk+1Qk — Qk+1QkQk +1 k = 1, . . . , n — 2 (IV.5.1)

aiak = Qkaa if Ii—ki>2. (IV.5.2)

The elements of Bn are then the products of the Qk and their inverses. Since for
n > m the group Bm is naturally embedded in Bn one may define Bc,,, as the
braid group for an unspecified finite number of strands.
Next we look at the adaptation of theorem 2.2.3, the generalization of the
key lemma to the sector pn . In (IV.2.37) we have only to change Tm to am
(corresponding to the choice that supp 0m lies to the left of supp 0m+1) and
obtain as the operator representing am in Bn (see equ. (IV.2.38))

(IV.5. 3 )
 IV.5 Low Dimensional Space-Time and Braid Group Statistics 193

One checks that this respects the group relations (IV.5.1), (IV.5.2):

er(Qm )e(on) (Qk) = er(uk)s (;) (Qm ) for 1 m — k I> 2 (IV.5.4)

and

e^n) (Um) ^^n) (Um+1)^^n) (Um) = E^n) (Um+1)E^n) (Um)E^n) (Um+1) . (IV.5.5)

Thus (IV.5.3) generates a unitary representation of B n . The relation (IV.5.4)

follows directly from the fact that EP commutes with 022t. This is part a) of
lemma 2.2.2 which remains unaffected by the dimensionality of space-time. To
show (IV.5.5) we start from the definition (IV.2.28) for e, which we may simplify
by choosing 02 = 0, U2 = Il to

EP = U1-1 g (IV.5.6)

if supp 01 is chosen in the left space-like complement of supp 0. Then

1 2=U^0
Ee0E e vi ,

( 06 e) E P(g E 0) = (PEP)Ui 1 02 U1
= U1 1 (01eP)(02 U1)

_
= U1-leg(0 2 U1)
ce(PEP)6

where we have used that U1 is an intertwining operator from 0 to 01 , then that

( IV.5.7)

E is localized in supp 0 which is space-like to supp 0 1 , implying 01e, = E and

finally (IV.2.29). Applying 0m_1 to (IV.5.7) yields (IV.5.5).
We note that in (IV.5.3) the right hand side is independent of n, the number
of strands used, as long as n > m. Therefore we get a unitary representation of
B„, which we denote by Fe
b E B„,, EP(b) (IV.5.8)

generated from
e(Um)
= 0m-1EQ (IV.5.9)
The next question concerns the characterization of the representation
(IV.5.8) of BOE, which is associated with an irreducible sector [0]. For this, as
in subsection 2.3, the conjugate sector [P] and the left inverse 00 of 0 are im-
portant. 0e is defined by (IV.2.46) and has the properties (IV.2.47), (IV.2.48),
(IV.2.49). Due to (IV.2.48) and (IV.2.29) 0060 commutes with 0% and thus, for
irreducible o, it is a multiple of the identity

P Q E Q = AQIl, (aP E C) . (IV.5.10)

As in (IV.2.51) the "statistics parameter” a P depends only on the class [0]. But
it may now be a complex number. We decompose it into a phase and an absolute
value
194 IV. Charges, Global Gauge Groups and Exchange Symmetry

T
1

•
•

Fig. IV.5.1.

111.e = we a d e 1 ; ^
w=1, de 1, ( IV.5.11)

and call w e the "statistics phase", de the "statistics dimension". In contrast to

(IV.2.55) w e is not limited to the values ±1 and de need not be an integer.
In the simplest case, when P is an automorphism, ee is itself a multiple of
the identity, the representation ee of Boo is 1-dimensional, de = 1 but we may
be any phase factor. We do not have the Bose-Fermi alternative. Models with
particles carrying such a charge have probably first been described by Streater
and Wilde [Streat 70]. The term "anyon" was introduced for such a particle
in [Wilc 82] to emphasize the fact that such particles may carry any statistics
phase. For particles whose statistics dimension differs from 1 (i.e. where P is
an endomorphism) the term "plekton" (for Greek "braided") was suggested by
Fredenhagen, Rehren, Schroer in [Fred 89a].
At this stage it may be remarked that in 3-dimensional space-time we meet
the same situation if instead of the compactly localized charges of DHR we
consider the string-like localized charges of the BF-analysis. This results from
the fact that in order to define the composition of two charges localized in cones
Si , 82 we need an auxiliary cone Sr which is space-like to both S i and 82 (see
section 3). There are two disconnected possibilities for the choice of Sr (see
fig. (IV.5.1) where we have represented the cones by their center lines in the
space-like plane s0 = 0 and marked two inequivalent choices of Sr by a dotted
and a dashed line respectively.)
As far as the statistics is concerned one arrives at the same conclusions as
in the case of compactly localized charges in 2-dimensional space-time. There
is, however, one further feature. In 3-dimensional space-time we may have spin.
Adapting the analysis by Wigner (subsection 1.3.2) we have to look at the sta-
bility group of a momentum vector. This is, in the case of a massive particle, the
Abelian rotation group in a plane which has IR as its covering group. An irre-
ducible representation of this is 1-dimensional and given by s E IR -f ei'x where
s is an arbitrary real number labelling the representation and corresponding to
the spin. There is a relation between the spin s and the statistics phase
 IV.5 Low Dimensional Space-Time and Braid Group Statistics 195

e 4wis = c4j 2 (IV.5.12)

and an addition theorem for the spin of plektons with statistics phases
spins s i , s2 combining to a pure state with statistics phase w, spin s
e 27ris cv
e 2ni(51+52) = w 1 • w2 '
(IV.5.13)

For the derivation see [Fred 89b], [Frbh 89] .

Continuing with the discussion of statistics we can define a complex val-

ued function yo on Boo using the definitions (IV.5.8), (IV.5.9), setting

li^ ^^ E e (b) = Sp(b) Il . (IV.5.14)

This is analogous to (IV.2.52). We have used the fact that any b E B,„, belongs
to some Bm for finite m. If the power n of 45e in (IV.5.14) exceeds m then by
(IV.2.49) ns&° is a multiple of the identity and does not change any more if n is
increased. So we become in the limit independent of the number of strands in b.
The function cp defines a state on the group algebra of Boo and it has properties
analogous to lemma 2.3.6.

Sp(b i b2 ) — Sp( b 2b i ) (IV.5.15)

and, if b1 E Bn and b2 is generated by the Qk with k > n,

cp(bi b2 ) = 40 (b1)40 (b2) • (IV.5.16)

For the derivation see [Fred 89a]. A function on Boo with these properties is
called a "strong Markov trace".
The representation eg is obtained from ço by the GNS-construction. Similar
to proposition 2.3.7 the statistics dimension de limits the irreducible components
which can occur in the restriction of E° to B r,.
The finite dimensional irreducible representations of Bn have not yet been
classified. However, an important advance in this direction has been initiated by
the work of Jones [Jones 83] . He considered inclusions of von Neumann factors
of type II i , showed that such an inclusion defines a number with remarkable
invariance properties, called the "index of the inclusion" and he gave a formula
for the possible values this index can take. Kosaki as well as Pimsner and Popa
generalized the notion of the Jones index to inclusions of arbitrary factors [Kos
86], [Pim 86]. Longo showed that in the case of the inclusion eR(n) C R(0)
which we discussed above the Jones index is just the square of the statistics
dimension [Longo 89]
Ind p = d2g . (IV.5.17)
The recognition by Jones, Ocneanu and others in the mid 80's that the theory
of W*-inclusions leads to Markov traces on Boo and to invariants for knots and
links spurred a considerable mathematical activity in this area. One interesting
196 IV. Charges, Global Gauge Groups and Exchange Symmetry

result is that the allowed values for the statistics dimension consist of a discrete
spectrum in the interval between 1 and 2 with an accumulation point at 2 and
a subset of [2, 00). 1 The discrete spectrum below 2 is given by

7r ; m a positive integer and m > 3 .

d = 2 cos — (IV.5.18)
m
For these values of d the Markov trace is uniquely determined by the statistics
parameter A.
There are cross connections to several other much discussed topics both in
mathematics and physics. Among them are "quantum groups" and the Yang-
Baxter equations in models of statistical mechanics and exactly solvable quan-
tum field theory models in 2-dimensional space-time. For reviews and guides to
the literature see [de Vega 891, [Fr6h 88], [Fadd 90].
Choosing one representative morphism pa in each class of irreducible mor-
phisms the composition law of the charges can be written as

pa p7 = ENâppp, (IV:5.19)

where NIg is the multiplicity with which the sector [Pp] occurs in the decom-
position of the product Aae.t. N7Q is a non negative integer and N7 is called
the fusion matrix. Keeping 8 7 = p fixed one considers diagrams with the p ,
as vertices and paths between them allowed by the fusion rules, (see [Ver 88]).
For each pair a, /3 we have N7p edges and a general path from an initial vertex
to a final vertex is composed of such edges. The set of intertwiners from pP to
per p is an N2:0-dimensional linear space in which an hermitean scalar product is
defined because, if T1 and T2 are two such intertwiners then the product TM
is a multiple of the identity and we may set

777'2 = (T1 T2) Il . (IV.5.20)

Picking an orthonormal basis in this space of intertwiners and picturing the

basis elements as the edges in the diagrammatic description we get for each
path e of length n, starting at Pa and ending at pb an intertwiner from pb
to pa pa, namely the product of the intertwiners associated with the edges. One
has an action of the braid group Bn on the set of intertwiners {T£} belonging
to paths of length n from a to S. s (b) intertwines from en to pn; so pae9 (b)
intertwines from Papa to pp' and (p,„Eo(b) )T from pb to pa pn. Thus

(paep (b)) TT = E41 (b)77q . (IV.5.21)

Such R-matrices, labelled by paths, have previously been encountered in models

of 2-dimensional conformal invariant quantum field theories. They are represen-
tation matrices of BŒ, and as such satisfy the group relations (IV.5.1), (IV.5.2).
'Recently Longo showed that there must be a gap in the spectrum above d = 2 [Longo
92]. The spectrum of d below V6 has been completely determined by Rehren [Rehr 95].
 N.5 Low Dimensional Space-Time and Braid Group Statistics 197

One approach to the construction of charge carrying fields starts from a

Hilbert space
7-l =E ®li« , (IV.5.22)
where each 7-l a can be identified with no but carries the representation oa2t of
the observable algebra. Then one introduces operators, called "exchange fields"
or "generalized vertex operators" leading from 7-1a to lip and induced by a
morphism o7 as in the composition (IV.5.19). One defines localization for such
exchange fields in the same way as it was done for the elements of the "field
bundle" in subsection 2.5. Products of space-like separated exchange fields sat-
isfy commutation relations in which the R-matrices enter. This structure is the
"exchange algebra" of Rehren and Schroer [Rehr 89]. It corresponds to the field
bundle rather than to the field algebra described in the last section. One may
ask whether a more economical net of algebras a(0) can be constructed so that
there are no non trivial common elements of 3(O1 ) and a(O2 ) for disjoint re-
gions and one has (IV.4.6). In the last section this succeeded by taking instead
of (IV.5.22) each sector with a suitable multiplicity. Here it seems improbable
that this can be achieved (see [Longo 95], [Fred 92]).
There remains the question of finding the analogue of the gauge group. This
problem has not yet been satisfactorily resolved. Although the representation
theory of a "quantum group" as defined by Drinfeld [Drin 87] or Woronowicz
[War 87] shares some features with the charge structure encountered in braid
group statistics there seem to be some differences. So the dual object of the
superselection structure in 2-dimensional field theories is not yet known. In the
situation discussed in [Longo 95] it is a "paragroup" in the sense of Ocneanu
[Ocn 88]. Other examples are discussed by Buchholz, Mack, Todorov [Buch
88], Mack and Schomerus [Mack 90, 91a, 91b], Frohlich and Kerler [Frei 91],
Hadjiivanov, Paunov and Todorov [Hadj 90], Majid [Maj 89], Rehren [Rehr 91].
The classification of braid group statistics remains largely open. An inter-
esting treatment of the relation between spin and statistics which emphasizes
the algebraic aspect is due to Guido and Longo [Guido 95a,b].
V. Thermal States
and Modular Automorphisms

V.1 Gibbs Ensembles, Thermodynamic Limit,

KMS-Condition

V.1.1 Introduction

In chapter IV we focused attention on states which, with respect to local obser-

vations, differ from the vacuum essentially only in some finite space region at a
given time. This is an appropriate idealization for elementary particle physics
but not for the study of properties of bulk matter, the "many body problem"
(many ti 1024 ) There the simplification leading to the deduction of laws of
thermodynamics and hydrodynamics from statistical physics is achieved by the
idealization that matter fills all space with finite density and, instead of the
vacuum, we have as the simplest states of interest the thermodynamic equilib-
rium states. To avoid misinterpretation: of course the realistic material systems
in which we are interested have finite extension and the standard approach in
statistical mechanics starts with a system of N particles enclosed in a box of
volume V with total energy E. But in order to give concepts like "temperature",
"phase transition" an unambiguous meaning the thermodynamic limit N —t oo,
V —' oo, E —> oo with N/V and E/V finite must be performed (or is implic-
itly understood). By basing the theory on the algebra 2( of local observables
with its net structure we have the advantage that the box is dispensable. Equi-
librium states in the thermodynamic limit are good states over 2.1 and can be
directly characterized. In the terminology of thermodynamics the elements of
21 are "intensive quantities".
To arrive at such a description of equilibrium states let us recall the rules
for computing them in non-relativistic quantum mechanics for a 1-component
system (a system containing only one type of particles) confined to a box of
volume V. If the total number of particles is given one considers the Hilbert
space NN of totally antisymmetric (resp. totally symmetric) N-particle wave
functions. Choosing suitable boundary conditions on the walls of the box one
has a Hamilton operator H. A general state is described by a density matrix
P (a positive operator with trace 1); the expectation value of an observable
A E 93(1-(N ) is given by
200 V. Thermal States and Modular Automorphisms

w(A) = tr QA. (V.1.1)

To an equilibrium state with inverse temperature 3 = (kT) -1 (k is the Boltz-
mann constant, T the absolute temperature) one assigns g = gp
go -011 ; Z = tr e -
= Z -1 e ' 3H . (V.1.2)

This is the adaptation of Gibbs' canonical ensemble to quantum mechanics.

Note that e —)311 is a trace class operator in the case of finite V since H then has
discrete spectrum with a level density increasing less than exponentially for large
energies and H is positive. The internal energy in the sense of thermodynamics
is then given by
E = tr gQH. (V.1.3)
One may remark that if, instead of /3, one prescribes E one may choose for
A other positive functions of H. Putting o = F(H) with tr F(H) = 1 and
tr F(H)H = E one observes that the shape of the function F plays no impor-
tant rôle. The "microcanonical ensemble" where F is chosen to have support
in a neighborhood of E is a familiar example. The difference between the con-
sequences of different choices disappear in the thermodynamic limit if E/V is
related to 0 by (V.1.3). The canonical ensemble is, however, the most conve-
nient choice for F in computations.
Even more convenient, with the thermodynamic limit in view, is the grand
canonical ensemble. Here we do not fix the number N but consider the Fock
space '1-IF — ® N o1-lN. The "observable" N (the particle number) is then an
operator in NF. One considers then the algebra of operations, generated from
bosonic and fermionic creation and annihilation operators. We shall denote it
again by A to distinguish it from the algebra of observables ?t. With H denoting
the Hamiltonian in 7-(F one takes o as
6,04, —
— G -ie-Q(H-µN)., G — tr e -13(11-1-LN) .
— (V.1.4)

,(3 is again the inverse temperature, p is the "chemical potential". By

E = tr oa,µ H; (N) = tr op,µ N (V.1.5)

the expectation values of energy and particle number become functions of /3, p
and V. In the thermodynamic limit V -1 E, V -1 (N) and e = V-1 In G become
functions of 13 and p alone. One finds e = Op, where p is the pressure. All ther-
modynamic quantities may be obtained from ® by differentiation with respect
to ,3 and u. As p increases the matter density increases. For fermions p can take
all values beween —oo and +oo, whereas for bosons p < 0. 1
The Gibbs states (V.1.1), (V.1.2) have a property first pointed out by Kubo
[Kubo 57] and used to define "thermodynamic Green's functions" by Martin and
Schwinger [Mart 59]. Many applications are described in [Kadanoff and Baym
1962]. If A is any observable then
'For the preceeding claims see any standard text book on statistical mechanics.
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 201

at ( A ) = eiHtAe-ixt (V.1.6)

is its time translate. Due to the invariance of the trace under cyclic permutations
one has for A, B E B(n) and wp defined by (V.1.1), (V.1.2)

wp((at A)B) = Z -l tre -px eixt Ae -iHt B = Z- l trBe iH(t+ ip) Ae -iHt
iH(t +i0) Ae -iH(t+0) ).
o (Be =w
So
wp((at A)B) = wp(Bat+ipA), (V.1.7)
where we have written, replacing t by a complex variable z
azA = e iHzA e —iHz (V.1.8)

Introducing for each pair A, B E 93(n) the two functions of z

FÂB(z) = wp(B(azA)) (V.1.9)

Ga,B(z) = wp((azA)B)

we see that with z = t +

FrB (z) = Z-ltr BeiHte-yHAe-ixte-(3-7)H

is an analytic function of z in the strip

0< <y<,B, -y=Imz, (V.1.10)

because He - H« is a trace class operator for a > 0. For real values of z, F and
G are bounded, continuous functions of t and one obtains G(t) as the boundary
value of F(z) for z — t + i/3.

GAp B (t) = F f (t + i18). (V.1.11)

The same relation results in the case of the grand canonical ensemble, replacing
0 by 0, µ and H by H(µ), at in (V.1.6) by at , 21. by A, where
H(µ) = H -- µN, at A = eiH(OtAe-iH(µ)t. (V.1.12)

The importance of relation (V.1.11) stems from the fact that it survives the
thermodynamic limit. Specifically we may regard A and B as local quantities,
as the state (normalized, positive, linear form over A) corresponding to
equilibrium with inverse temperature ,Q, chemical potential in unlimited space.
We consider H and N in (V.1.4) as generators of symmetries which are realized
by automorphism groups on A, namely the time translations a t and the U(1)-
gauge transformations
-iN `p.
ryv,A = eiN`PAe (V.1.13)

H(µ) is an element in the Lie algebra of the symmetry group and

202 V. Thermal States and Modular Automorphisms

at = ary_, t . (V.1.14)

We define, for each pair A, B E A the functions

F11,1(t) = wp, N,(Bai A) — wQ (A)w^(B) , (V.1.15)

Gâ;B(t) = wQ,/,((at A)B) — wa, p (A)wo,i,,(B). (V.1.16)

Then the above discussion suggests that F(t) is the boundary value on the real
axis of a function F(z), analytic in the strip (V.1.10) and that G(t) is obtained
as the boundary value for Im z = ,@ as in (V.1.11).
With this adaptation to the algebraic setting of the work of Kubo and of
Martin and Schwinger quoted above Haag, Hugenholtz and Winnink [Haag 67]
postulated the relation
GÂ;B ( t ) = FÂ,B ( t + zQ), (V.1.17)
with the definitions (V.1.15), (V.1.16) and the analyticity requirement for F as
the defining property of an equilibrium state with parameters Q, µ and called it
the KMS-condition (for Kubo, Martin, Schwinger). This paper will be referred
to in the following as HHW. The KMS-condition implies that w )3,0 is an invariant
state with respect to at 2

wQ, p (at A) = wQ,µ (A) . (V.1.18)

The subtraction of w(A)w(B) in (V.1.15), (V.1.16) (which could be omitted)

takes away the uncorrelated part in the expectation value of the product. This
has the advantage that then we may also assume in the infinite volume limit

lira F(t) = lim G(t) = 0, (V.1.19)

t--^foo

because in a system without boundaries the correlations between local quantities

at different times are expected to tend to zero as the time difference increases
to infinity. 3
An alternative characterization of the KMS-condition is the following; (for
details of the argument see HHW).

2 Putting B =1, t = 0 one has wa,p (a:A) = ws, µ (A). Due to the analyticity and bound-
edness in the strip 0 < 1m z < Q wp, t,(aµA) is then a periodic function in the direction of the
imaginary a xis, hence analytic in the whole complex plane and bounded. Thus, by Liouville's
theorem, it is constant and one has (V.1.18).
3 For a system enclosed in a box this will not be so because the Hamiltonian then has a
discrete spectrum and so FA, B (t) will generically be an almost periodic function. In classical
mechanics this corresponds to the Poincaré recurrence. In contrast, in an unbounded medium
the effect of a loc al operation will spread and dissipate so that it is not felt in a finite region
after a very long time.
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 203

Lemma 1.1.1
The KMS-condition is equivalent to the requirement that the Fourier trans-
forms of the functions F and G, defined in (V.1.15), (V.1.16) are related by a
Boltzmann factor:
G(E) = e-p£F(e). (V.1.20)
Here the Fourier transform is taken as

F(e) = f F(t)e -t£t dt

-

and considered as a distribution over smooth test functions f (e) with compact
support.

V.1.2 Equivalence of KMS-Condition

and Gibbs Canonical Ensembles for Finite V

Does the KMS-condition suffice to characterize an equilibrium state? Let us

first look at the case of the system enclosed in a box. The standard way to
describe such a system with an arbitrary number of (identical) particles in non
relativistic quantum theory uses creation and annihilation operators acting in
a Fock space Î'lF.
To fix the ideas' let us think of spinless bosons interacting by 2-body forces
given by a potential V(x — x) . Denoting by a (x) the (improper) annihilation
operator of a particle at the point x E V (the region of RR3 enclosed by the box),
by a*(x) the creation operator, we have the commutation relations

[a(x), a*(x')] = 6 3 (x — x'); [a(x), a(3 )] = 0. (V.1.21)

We have a vacuum state vector S? E xF characterized by

a(x),fl = 0 for all x E V. (V.1.22)

The Hamiltonian H and the particle number operator N are formally given by

H= f (Va*)(Va)d3x + 2 f a*(x')a*(x)V(x — x)a(x)a(x )d3 xd3x' ,

2m
(V.1.23)
N f a*a d3x. (V.1.24)
In order to make them well defined self adjoint operators the formal expressions
have to be supplemented by boundary conditions on the walls of the box defining
the domains of H and N. These conditions reflect the way in which the walls
influence the system. They have no counterpart in the case of the infinite system.
The question of the sensitivity of the thermodynamic limit to the choice of the
boundary conditions will concern us later. The operators H, N and H(lc) =
4 Thereader insisting on mathematical rigour is invited to skip this passage. It could be
made precise but the effort would only serve to deviate attention from the essentials.
204 V. Thermal States and Modular Automorphisms

H — AN yield the automorphism groups a t , 74, and at acting on 13(liF) as

described before.
For the finite system we can phrase the question of the equivalence be-
tween the KMS-condition and the Gibbs prescription as follows. Given the 1-
parametric automorphism group (Ili'. Does the requirement that w is a normal
state on B(?-1F ) satisfying the KMS-condition imply that w is given by (V.1.1),
(V.1.4)? The answer is yes. First, any normal state on 93(flF) is described by a
density matrix as in (V.1.1). Secondly, all automorphisms of Q3(NF) are inner;
so at defines an implementing unitary UA(t) E (1-( F) up to a phase factor
and, given adequate continuity with respect to t, it defines a generator H(i) up
to an additive constant which plays no rôle in (V.1.4). If we insert in (V.1.15),
(V.1.16) for A an invariant element i.e. one which commutes with all Uµ(t) then
FA ,B(Z) and GA,B(z) are independent of z and the KMS-condition says that F
and G are equal in that case, which implies for the density matrix
tr [o, A]B = 0 for all B E 93(1-lF)

i.e. [o, A] = 0 if A commutes with H(A). Thus A E {Ut UI (t)}" and A must
be a bounded function of H(A). That this function is of the form c e -01(u)
follows then by choosing for A and B operators which have nonvanishing matrix
elements only between two vectors W1, W2 which are simultaneous eigenvectors
of H(µ) and g. Note that since A is a trace class operator it must have discrete
spectrum. This implies, incidentally, that a KMS-state over 93 (l-lF) for positive
Q exists only if H(A) has discrete spectrum, bounded below, with a density of
eigenvalues increasing less than ei 36 for energy E -i oo.

V.1.3 The Arguments for Gibbs Ensembles

The line of argument justifying Gibbs ensembles as the proper characterization

of equilibrium for finite systems is the following. An equilibrium state is sta-
tionary. Therefore the density matrix (or, in the classical case the distribution
function in phase space) should be a function of the constants of motion. There
are some constants of motion we know a priori; they relate to the symmetries
(see section 4 of Chapter I). In the non relativistic case for a system with a single
species of particles in a fixed box there remain among those only two: energy
and particle number.' The essential question is whether there are no "hidden"
constants of motion, stationary quantities not tied to generally known sym-
metries, which could enter in the distinction of equilibrium states. It was the
original objective of ergodic theory to show that this worry can be excluded for
"almost all" dynamical systems. In classical systems the situation is, however,
much less simple. I refer to the famous KAM-theorem. 6 For infinitely extended
quantum systems we shall come back to this problem in Section 3.
5 Thebox destroys spatial translation invariance, Galilei invariance and also rotation invari-
ance unless it is of special symmetric shape. In some problems (rotating stars, spin systems..)
a non vanishing angular momentum must, of course, be considered.
6 Kolmogorov, Arnold, Moser. See e.g. the book by [Arnold 1978] and references therein.
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 205

Assuming that there are no "hidden" constants of motion one concludes that
the density matrix for an equilibrium state of a 1-component system must be a
function of H and N. Which function? As mentioned earlier the thermodynamic
consequences are not sensitive to the choice of this function (within reasonable
limits) as long as it yields the right expectation values for the energy density u
and the particle number density n:
u = V-1 tr pH; n = V-1 tr QN. (V.1.25)
To describe equilibrium states for different values of u and n one needs a 2-
parametric family of functions of H and N (for fixed V). The grand canonical
ensemble (V.1.4) parametrized by 0 and µ, the canonical ensemble (V.1.2),
parametrized by 0 and n, the microcanonical ensemble, parametrized by u and n
are the standard choices. They, as well as other reasonable choices of the function
of H and N, all lead to the same family of states over the local algebras in the
thermodynamic limit, namely to KMS-states with respect to a 1-parameter
subgroup of the automorphism group of symmetries, which in the case of the
1-component system is generated by a t of (V.1.6) and 'yip of (V.1.13). In section
3 we shall give direct justifications (independent of the above arguments) for
the claim that in an infinite medium the KMS-condition is the appropriate
characterization of equilibrium.
Another road to the Gibbs ensembles starts from the principle of maximal
entropy. For a state over 93(H), described by a density matrix o, the entropy is
defined as
S = —tr oln o. (V.1.26)
It is some measure for the lack of information, the impurity of the state. ? An
equilibrium state may be defined as one having maximal entropy among the
states with the same (prescribed) expectation values for energy and particle
number. An infinitesimal variation of o by So changes S by
SS = ---tr So(ln o +1).
Note that in spite of the fact that Sp need not commute with o one has the
formula Str F(o) = trSQF' (o) which may be justified for instance by using first
order Schrödinger perturbation theory for the eigenvalues of F(o +- So). Taking
the auxiliary conditions Str o = 0, Str off = 0, Str oN = 0 into account by
Lagrange's multiplier method the extremality requirement for S yields
trSo(lno+1 +a+ /3H + vN) = 0
for essentially arbitrary 4. 8 The multipliers a, , 3 and v are numerical constants.
So S is extremal under the stated auxiliary conditions if and only if the factor
a pure state S = 0. For a mixture of n pure states corresponding to linearly inde-
7 For
pendent state vectors we have 0 < S < ln n and the maximum value of S is attained for
g= n-'Pn where Pn is the projection operator on the subspace spanned by the state vectors.
The state of maximal entropy is the one which prefers no direction in the accessible subspace.

8 The only remaining restriction for 6p is that P + E o p should remain positive for E -^^ O. If
P is not on the boundary of the convex set of states this gives no restriction for Op.
206 V. Thermal States and Modular Automorphisms

of b p under the trace vanishes. This means

io = c e -OH-vN
which is the grand canonical ensemble.

Comment. Before proceeding further let us comment on the terminology used

in connection with (V.1.13). We have tied the conservation of particle number
to the invariance under a U(1) gauge group. If the term "gauge group" is un-
derstood in the sense of Chapter IV this elevates the conservation of particle
number to a superselection rule. The algebra A then corresponds to the "field
algebra" whereas the "algebra of observâbles" is the subalgebra of "gauge in-
variant elements", i.e. operations which do not change the number of particles.
In the case of a system with several components 9 the "gauge group" is a direct
product of U(1)-groups, one factor for each component. Correspondingly we
need for the parametrization of equilibrium states a chemical potential for each
component. This illustrates that we are dealing here with an approximation, an
idealization appropriate to a specific regime, since there may be reactions which
are not forbidden in principle but strongly inhibited at the energies considered.
If we would have to take the possibility of nuclear reactions into account we
would be left with three components, the total electric charge, the total baryon
number and the lepton number. Strictly speaking the conservation laws do not
refer to particle number but to charges and this is indeed tied to the invariance
under a global gauge group as discussed in Chapter IV. One simple and well
known illustration of the importance of this distinction is the case of a "pho-
ton gas". There is no charge associated with a photon, no superselection rule
between the vacuum and states with photons. Therefore one has no chemical
potential and there is only the single parameter /3 needed to distinguish equi-
librium states of a photon gas, a fact of paramount historical importance for
the discovery of Planck's constant in black body radiation.
Still, also in a regime where some reactions which are possible in principle
are so strongly inhibited that one can treat them as forbidden, it is appropriate
to consider the relevant approximate conservation laws as tied to a gauge group
and restrict attention to the "obse rvable algebra" composed of the gauge in-
variant elements. The rôle of the chemical potentials as a distinguishing feature
for equilibrium states on the obse rvable algebra must then be understood in
analogy to the appearance of charge superselection sectors in Chapter IV. In
the direct characterization of equilibrium states of an infinite medium which
does not start from the Gibbs Ansatz for finite systems this is indeed necessary
and we shall deal with this in subsection 3.4.
As mentioned before the advantage of the KMS-condition over the Gibbs
prescription (V.1.1), (V.1.4) is that it carries over to the infinitely extended
medium. There w0,0 can no longer be described by a density operator in 7-1F but
9 a system composed of different species of particles where we count as components only a
basic set of chemically independent species, omitting to count those which can be reached by
chemical reactions from the basic set.
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 207

as a state over the abstract C*-algebra A, equipped with its local net structure
and the automorphism groups describing the symmetries. This eliminates the
box, the boundary conditions, the consideration of the thermodynamic limit and
applies with equal ease to the relativistic case. An equilibrium state is charac-
terized as one satisfying the KMS-condition with respect to a 1-parameter sub-
group of automorphisms whose geometric correspondence in Minkowski space
is a time-like motion for all points.' °

V.1.4 The Representation Induced by a KMS-State

Let w be a KMS-state over A. It is natural to look at the representation 7r of A

resulting from w by the GNS-construction. The states in this folium are those
which deviate from the equilibrium state w only by essentially local excitations.
As shown in HHW the representation 7r has some remarkable properties. It
is instructive to exhibit them first for the case of the system in a box where
A = 93(f F) and w is given by (V.1.1), (V.1.4). We drop the indices ,/3, p
during this discussion. The density operator Q is positive. Therefore
1 /2 (V.1.27)
ko = P

is well defined and of Hilbert-Schmidt class, that is in the set

{k : tr k*k < cc, k E 93 (H F) }. We denote this set here by H because it is
a Hilbert space with respect to the scalar product

(Kilo) = tr k* , (V.1.28)

complete with respect to the norm 11 IIH= (tr * k)1/2. It is also a *-algebra
(though not a C*-algebra because ^^ K 11H is not the operator norm of ic) and it
is a 2-sided ideal in Z(fF):

kE H and A E 93(7-IF) implies AK E H and kA E H. (V.1.29)

Since g has finite trace, K o E H and since all spectral values of p and hence of
ico are non vanishing, we have

A,co # 0; koA 0 0; w(A*A) T tr QA* A 0 0 for A E 93 (7iF), A# 0.

(V.1.30)
Thus w is a faithful state; it has no Gelfand ideal. Consider now the following
representation of A = 93(HF) by operators acting on H.

7r1(A)IK) = IA,c); K E 7-1, A E 93{7-lF). (V.1.31)

'°This limits the relevant symmetries to a combination of positive time-like translations
and internal symmetries. If one takes a generator in the Lie algebra of symmetries which
brings in rotations and Lorentz transformations then the orbits of points are not positive
time-like everywhere and global KMS-states with respect to such subgroups will either not
exist at all or have no physical significance. There is, however, the interesting case of causally
complete subregions, invariant under such groups which we shall meet in connection with the
Bisognano-Wichmann theorem and the Hawking temperature of black holes.
208 V. Thermal States and Modular Automorphisms

Here we used for distinction the notation l,c) if the Hilbert-Schmidt operator 'c
is to be considered as a vector of the Hilbert space H. One has

( ^so lirl(A)f moo) = w(A). (V.1.32)

Due to (V.1.30) k o) is a cyclic and separating vector for the representation 7r 1 .

From (V.1.31) and (V.1.32) one recognizes that it is isomorphic to the GNS-
representation induced by w and that lic a) is the state vector corresponding to
w. We shall therefore drop the index l and write 7r instead of 7r 1 .
There is another natural mapping A E 93 (7-Hp) —; 7r r (A) E Z (H) :

7r(A)1K) = kA*). (V.1.33)

It gives a conjugate linear representation of A, namely

7rr (AB) = 7rr (A)7rr (B); 7rr(A*) = ( 7rr(A)) * ; 7r r (cA) = C7rr (A). (V.1.34)

From the definitions one sees directly that 7r r (A) commutes with 7r(B) for any
A, B E A. One finds

Theorem 1.4.1
(i) The operator norms of 7r(A) and 7r r (A) are equal.

11 7r(A) I1=11 hr(A)11 • (V.1.35)

(ii) The commutant of 7r(A) is precisely the weak closure of 7rr (A)

( 7r(A)) ' = ( 7rr(A)) " . (V.1.36)

(iii) Co is a cyclic vector for both 7r(A) and 7rr (A) and

w(A) (icol 7r(A)1ico) = (lc a l7rr (A)lfka ). (V.1.37)

(iv) 7r and 7rr are transformed into each other by an antiunitary operator J
defined by
J lk) (V.1.38)
One has
J7r(A)J = 7rr (A); J2 = Il; Jlko) = ko ). (V.1.39)

We omit the proof which is simple and may be found in HHW.

Since w is invariant under at we can implement this automorphism group in
the GNS-representation by unitary operators UP(t) in the standard way putting

UP(t)7r(AN) = 7r(4 A)lico ). (V.1.40)

Similarly, if w is invariant under a t and ty i, separately, we have unitaries

U(t), V ((p) implementing these automorphisms. The generators of U, V, Wz
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 209

may be regarded as the Hamiltonian H, the particle number operator N and

H(µ) = H — AN, which all are operators acting in H and satisfy

Hiko ) = 0; NIko ) = 0. (V.1.41)

These operators are, however, clearly not the representors of H, N considered

in (V.1.23), (V.1.24). Rather, if we denote the latter by HF NF and regard ,

UF(t) = exp iHFt as elements of A then

U(t) = 7r(UF (t))7rr(UF (t)), (V.1.42)

or, symbolically
H = 7r(HF) — 7rr(HF), (V.1,43)
with analogous expressions for U P , H(µ) and for V (cp), N. Since 7rr (,A) com-
mutes with 7r(A) the second factor in (V.1.42) has no effect on the automor-
phism and

atr (atA) t=0

= i[H, 7r(A)] = i[7r(HF), 7r(A)]. (V.1.44)

The subtraction of 7rr (HF) in (V.1.43) becomes, however, essential in the ther-
modynamic limit. The total energy operator 7r(HF) becomes meaningless in the
limit. Its expectation value and its fluctuations become infinite as V oc.
The second term in (V.1.43) cancels the infinities so that the equilibrium state
vector becomes an eigenvector of H to eigenvalue zero as seen in (V.1.41). The
mechanism for this can be seen by noting that
!cp = Z-1/2 e -0 /2 HF .

We have
r(HF)I moo) = Z- 1 / 2 IHFe-0/ 2
7 HF)
- W,
Whereas, according to (V.1.43)
e-0/2
Hlno) = Z-1/2 1 [HFe HF ] ) —
0.

The norm of W increases to infinity as the volume V -3 oc. We also see 11 that
e -012 H = I KOA) = JO*
O NO)

This gives the relation

7r(A * )1ko) = Je -"3/2117 ( 11)1 60), (V.1.45)

which also survives the thermodynamic limit and whose key rôle will become
apparent in the mathematical theory of Tomita and Takesaki described below.
'Note that due to the conjugate linearity of 7r,. we get from (V.1.42) for imaginary values
of t
e -1/20H _ 7r(e -1 / 2i3HF )7rr (e l / 2QHF ) (V.1.42a)
210 V. Thermal States and Modular Automorphisms

Summing up: for the system in a box we have two equivalent descriptions
of equilibrium states. The traditional one uses an irreducible representation
of A in HF (in fact we have taken A = 93(HF)). The impure state wa,u is
then described by the density operator (V.1.4). The other uses a reducible
representation it in H. The state wp, µ is then described by a vector in 1-1, an
eigenvector of the modified Hamiltonian H(µ). The commutant of 7r is obtained
from 7r by conjugation with an antiunitary operator J (see (V.1.36), (V.1.39)).
The first description is no longer possible in the thermodynamic limit but the
second one survives and the essential features remain.
Let Vn be a set of increasing volumes exhausting space

V11+1 D V,"; UVn = IR 3 .

For each box Vn we have a Fock space f(Fn) , operators Hn , Nn defining au-
tomorphisms cxtn) , in) , cx/`'(n) of An = B(fF ) ) and statistical operators 44)
as described before. The algebra A m, is naturally identified as a subalgebra of
An when n > m and the inductive limit of this directed set of algebras (as a
C*-algebra) is denoted by A. Then one has

Theorem 1.4.2
Assume that

a) for each positive ,8 and a certain range of values of µ (independent of n)

pt) is a trace class operator on HF),
b)
li at'(n)(A) — cxt ,(m)(A) ii —' 0 as n, m —4 00 (V.1.46)

for any A E Ak and any t E IR (A and t being kept fixed in the limiting
process),

c) The numerical sequence w(A) = tr pSn;A, A E Ak converges as n ---)

12
00.

Then

(i)
ce A = lim (en) A (V.1.47)
defines an automorphism group on A,

wp,u (A) = lim wt) (A) (V.1.48)

defines a state on A which satisfies the KMS-condition for cxµ (with pa-
rameter ,Q).
12
This assumption concerns only the uniqueness of the limit state. It may be dropped and,
in fact it does not hold in many cases of interest (see the comments below).
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 211

(ii) The GNS-representation 7r of A induced by w134, has properties analogous

to those listed in theorem 1.4.1. Specifically there is a self-adjoint operator
H(µ) and an antiunitary operator J on the representation space H defined
by
7r(o A) = eiH(u)tI(A)e-iH(u)t; HOc)S2 = 0, (V.1.49)
7r(A*)S2 = Je-1312 H(I`) 7r(A)S2; J,f2 = S2, (V.1.50)
where S2 E H is the vector representing w13,.

J = J-1 . (V.1.51)

Defining a conjugate representation 71,. by

7r,.(A) = J7r(A)J, (V.1.52)

one has (V.1.35) through (V.1.37) with K o replaced by Q.

The proof can be skipped here. It is obtained from the stated assumptions in a
straightforward way using lemma 1.1.1. For details see HHW. Of particular in-
terest will be part (ii) which establishes the connection between thermodynamic
equilibrium states and the Tomita-Takesaki theory of modular automorphisms
of von Neumann algebras, a mathematical structure which will be discussed in
the next section.

Comments. Conditions for the interparticle forces which guarantee the assump-
tions a) and b) have been extensively studied. The first concerns "saturation".
ForfixedvlumthgnsaeryofN-pticlsemhoudn
drop towards --oo faster than linearly with increasing N. Assumption b) means
that the effect of the boundary of the box on the time translation of a well
localized quantity during a fixed time interval should become negligible when
the box is taken sufficiently large. The worst case for both a) and b) is that of
long range, attractive forces. For certain classes of models the validity of the
assumptions has been rigorously established. They include lattice systems with
short range interaction (see [Streat 68], [Ruelle 1969], [Bratteli and Robinson,
Vol. II 1981]) and Galilei invariant theories in which the interaction decreases
not only with the distance but also with the relative momentum of the particles
(see Narnhofer and Thirring [Narn 91]). For Coulomb forces between an equal
number of positive and negative charges saturation has been proved by Dyson
and Lenard [Dy 67, 68] and, with significantly improved estimates, by Lieb and
Thirring [Lieb 75 a,b]. See also [Thirring, Vol.4 1988]. Thermodynamics in cases
of non saturating forces has been discussed by Hertel, Narnhofer and Thirring
[Hert 72], Narnhofer and Sewell [Narn 81].
Granted the assumptions a) and b) the problem of the existence of the
thermodynamic limit (assumption c)) is essentially a question of choosing a
consistent set of boundary conditions as the size of the box increases. This
relates to the question of uniqueness of KMS-states with given parameters 0, µ.
We know that at a phase transition point or in connection with spontaneous
212 V. Thermal States and Modular Automorphisms

breaking of a symmetry (e.g. spontaneous magnetization) the KMS-state cannot

be unique and this has its counterpart in the sensitivity of the states w o( ;u on
the boundary conditions, on the precise definition of the box Hamiltonian. This
leads to the next topic, the decomposition theory of KMS-states in the infinite
medium.

V.1.5 Phases, Symmetry Breaking

and Decomposition of KMS-States

Let K 0 be the set of KMS-states over the C*-algebra 21 with respect to an au-
tomorphism group at and parameter 0. M ixtures of KMS-states with fixed at
and /3 again satisfy the KMS-condition. So K 0 is a convex set. Furthermore it
is closed in the weak topology induced by % in state space and hence weakly
compact. Therefore Kp has extremal elements (states which are not decompos-
able as m ix tures of other states in K 0 ). General states in Ka may be regarded
as mixtures of the extremal states in Kp. The basic fact is now

Theorem 1.5.1
(i) An extremal state of Kp is primary; a primary KMS-state is extremal.

(ii) K3 is a simplex i.e. the decomposition of a general state in Ka into ex-

tremals is unique.

Proof. Part (ii) follows from (i) because the central decomposition of a state is
unique. It is obtained by considering the restriction of the state w on the center
of the von Neumann algebra irw (%)". The center is Abelian and the state space
of an Abelian algebra is a simplex (the situation of classical physics).
For part (i) we may start from a state w in the interior of the convex set
Ka and look at the GNS-representation induced by it, obtaining thus a Hilbert
space x, a von Neumann algebra 1Z with a cyclic, separating vector Q and a
group of unitaries U(t) implementing the automorphisms a t . Any other state
wl E K03 is then dominated by w and can be obtained as

wl (A) = w(AT') (V.1.53)

where Ti is a positive operator in the commutant i.e. T E 1Z'. The requirement

'

that wl is again in Ka implies that Ti must commute with the unitaries U(t)
because, with A, B E 1Z

(AQIU(t)T'U*(t)IBSQ) = (QIU(t)T'U*(t)A* BIQ)

= (OFT U(—t)A* BU* (—t)IQ) = wi (a - t A*B) = wi (A*B) = (A,fQIT'IBf2) .

Here we have used the fact that U(t)T'U*(t) as well as T' belong to the com-
mutant, that U(t)Q = Q and that w 1 , being a KMS-state with respect to at , is
invariant under a t . So we get
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 213

U(t)T'U *(t) = (V.1.54)

because Rf2 is dense in f.

The proof is completed now by the following lemma

Lemma 1.5.2
Let (7-1, R, S2, U(t)) be the representation induced by a KMS-state w with the
notation as defined above and J the conjugation defined in (V.1.50). If T ' E R'
commutes with the unitaries U(t) then
(i) T = JT'J E R also commutes with the U(t) and

T' (2 = TO; wi (A) - w(AT') = w(AT). (V.1.55)

(ii) If furthermore the state co l , defined by (V.1.53) is in Kp then

T =T' E R 7Z = 3. (V.1.56)

Proof. Part (i) follows directly from (V.1.50), (V.1.52). For part (ii) We use
the KMS-condition for both w and col in the form

wk(AB) = wk((a_ 0 B)A); wk E K 0 , A, B E R. (V.1.57)

One has, with A, B, C, T E R

w(ATBC) = w((a__ ipBC)AT) = w l ((a_.0B)(a_ i0C)A)

= wl ((a_ i0 C)AB) = w((a_ ipC)ABT) = w(ABTC).

Thus the matrix elements of [T, B] between a dense set of state vectors vanish
for all B E R i.e. T E R'. So T is in the center 3 = R n R'. ❑

The theorem suggests the physical interpretation that an extremal KMS-

state should correspond to a pure thermodynamic phase. Ko consists of more
than one element if and only if there are several coexisting phases possible at
the parameter value /3. The term "phase" used in this context needs some qual-
ification. If we think of the example of a liquid and gaseous phase which may
coexist at special v al ues of /3 and p (the latter entering in the choice of the
automorphism group at ) then, at these parameter values there is also an equi-
librium state in which we have a phase boundary, say the plane x3 = 0, such
that the half space x 3 > 0 is occupied by the gas, the half space x 3 < 0 by the
liquid. This is not a statistical mixture of two homogeneous states (gas, resp.
liquid in all space) but again an extremal state of K. Thus the pure phases in
the thermodynamic sense should be understood as the macroscopically homo-
geneous, extremal KMS-states. In the example we have only two homogeneous,
extremal KMS-states (at the special values of /3, µ) but an infinity of inhomo-
geneous extremal KMS-states at these parameter values. The interior points of
the convex set Ka result if we average the latter over the position of the phase
boundary. The existence of two distinct homogeneous extremal KMS-states is
214 V. Thermal States and Modular Automorphisms

not related in this example to any symmetry but analogous to an accidental

degeneracy, yielding the same value for the pressure in two entirely different
configurations. Therefore it is not expected to happen for generic values of the
equilibrium parameters but at most on a submanifold of these of codimension
1. This leads to Gibbs' phase rule for the maximal number of coexisting phases
of this type (compare the introduction by Wightman to [Israel 1979]).
Another reason for the appearance of several phases is spontaneous sym-
metry breaking. Clearly, if -y is an automorphism which commutes with a t then
the transform under -y of an extremal KMS-state i.e. the state w`, defined as
w` (A) = w('A), is again an extremal KMS-state to the same parameter val-
ues. If w' # w one says that the symmetry is spontaneously broken and then
KQ consists of more than one point. This phenomenon, being not "accidental"
may occur in a subset of the manifold of equilibrium parameters with the full
dimensionality.
We cannot enter into a discussion of conditions for phase transitions and
spontaneous symmetry breaking in states with finite matter density. This fasci-
nating subject is so rich that any serious attempt to include it would unbalance
this book. It has been a central topic in statistical mechanics for decades. Some
results using the algebraic approach and the KMS-condition are given in the
books [Ruelle 1969], [Bratteli and Robinson, Vol. II, 1981], [Sewell 19891.
A relativistic version of the KMS-condition which involves all space-time
variables and is stronger than the standard form has been proposed by Bros and
Buchholz [Bros 94]. For finite temperatures Lorentz invariance is spontaneously
broken [Ojima 86].

V.1.6 Extremality and Autocorrelation Inequalities

The maximal entropy principle for equilibrium states of a system in a finite

volume can be cast in another form.

Proposition 1.6.1
The grand canonical ensemble to parameters J3, µ is distinguished among all
density matrices p as the one yielding the maximal value for the functional

P,,u(p) = /3- 'S(p) — tr pH(µ); H(1i) = H — µ1V. (V.1.58)

The maximal value of 0, achieved for p = p0,A , equals /3 -1 1n G = pV where G

is the grand partition function, p the equilibrium pressure.

Proof. By (V.1.4)
/3H(µ) = — In pa,i, — ln Go,u Il.
So
In Go,u — /3 00,u (p) = tr (pin p — gin pa, p ), (V.1.59)
0 a,^(pa,µ) -=_ Q i ln Gp,u• (V.1.60)
 V.1 Gibbs Ensembles, Thermodynamic Limit, KMS-Condition 215

Now we note that for a positive trace class operator B and any unit vector W
one has
(WI In BMW) < In (W I BMW), (V.1.61)
where we do not exclude the value —oo for the left hand side. This is seen by
writing for B its spectral resolution B = E bi l Wi) (W I . Then

(WIln BMW) = Ai In bi (W I BMW) = > a i bi .

, (V.1.62)

ai = (wiiw)I 2 >0, Eai =1. (v.1.63)

Since the function In x is convex one has

E Ai In b i < In E A i bi,
which gives the inequality (V.1.61).
Next, let A be a positive trace class operator with the spectral resolution
A = E ai I 0i) (0 i I; thus a i > 0 and {0i } a complete, orthonormal system. Then,
using (V.1.61)

trA 1n B < Ea i ln (0i1B10i),

trA1nA—trA1n B> ai (lnai —ln(0i IBI0i )).

Now, for positive numbers x and y

x(lnx — lny) >x— y.

Thus we get, for positive trace class operators A, B the inequality

trAlnA — trA1nB > trA — trB. (V.1.64)

In the derivation of (V.1.64) we have interchanged the order of summation of

infinite series and this is legitimate only if the series are absolutely convergent
which means that also A In A and A In B should be trace class operators. Putting
A = A and B = Ap,u (V.1.64) with (V.1.59), (V.1.60) gives

0 0,u(Aa,u) — f0,µ(A) ? 0. (V.1.65)

This is the claim of the proposition. ❑

The practical advantage of proposition 1.6.1 as compared to the principle

of maximal entropy is that the maximality of 0 0,4 is no longer subject to aux-
iliary conditions. It gives a powerful tool for the derivation of inequalities for
the functions c,)p, u (A*a ^,A), ry E [0, 8] i.e. the functions FA ,A .(z) of equation
(V.1.15) with imaginary argument. Such functions are called autocorrelation
functions (of A) and the inequalities play an important rôle in the study of
phase transitions. The derivation and applications of such inequalities are ex-
tensively discussed in [Bratteli and Robinson, Vol. II, 1981]. We shall collect
two of them in the following theorem and note that - in contradistinction to
216 V. Thermal States and Modular Automorphisms

proposition 1.6.1 - they remain valid in the case of an infinite medium; they
may be regarded as a replacement of the extremality principles in this case and,
moreover, they give an alternative characterization of KMS-states. The follow-
ing theorem evolved from the work of Roepstorff, Araki and Sewell, Fannes and
Verbeure; [Roep 76, 77], [Ara 77a], [Faun 77a and b]. For a proof and further
references see Bratteli and Robinson loc. cit..

Theorem 1.6.2
Let 2( be a C*-algebra, at a 1-parameter subgroup of automorphisms, 6 the
generator of at i.e. SA = a&tA/atI t =0 • Note that S is a (possibly unbounded)
derivation, defined on a dense domain D(S) C 2i so that SA E 2I for A E D(6).
Let w be a state on 2 and put

u = w(A*A), v = w(AA*). (V.1.66)

Then the following conditions are equivalent

a) w is a KMS-state with respect to the group at with parameter (;

b) —iI3w(A*SA) > u Inv for all A E D(S); (V.1.67)
c) 0-1 fô w(A*a iy A)d-y < (u — v)(ln u — lnv)` 1 (V.1.68)
for all A E D(6).

In short: the inequalities (V.1.67), (V.1.68) are consequences of the KMS-

condition and either one of these inequalities implies the KMS-condition.

V.2 Modular Automorphisms

an d Modular Conjugation

This section is a mathematical digression. It is devoted to the Tomita Takesaki

theorem and consequences thereof. This theorem is a beautiful example of
"prestabilized harmony" between physics and mathematics. On the one hand
it is intimately related to the KMS-condition. On the other hand it initiated a
significant advance in the classification theory of von Neumann algebras and led
to powerful computational techniques. The results will be used in the discussion
of physical questions in the subsequent sections.

V.2.1 The Tomita-Takesaki Theorem

It was known before that any W*-algebra which has a faithful representation
on a separable Hilbert space is (algebraically and topologically) isomorphic to
a von Neumann algebra R in "standard form". "Standard" means that R acts
on a Hilbert space H possessing a cyclic and separating vector Q. In other
words the domains D = 7W C H and D' = 7Z' ,f2 C 7-t are both dense in 7-t. If
 V.2 Modular Automorphisms and Modular Conjugation 217

(1Z, 1l, (2) is standard then one may consider the conjugate linear operator S
from D to D
SAf2 = A*,f2; A E R. (V.2.1)
Similarly one has a conjugate linear operator F from D' to D'

FA' f2 = A'*,fl; A' E 1Z'. (V.2.2)

One finds that S and F are closeable. This means that if Wn is a sequence in D
converging strongly to W E 1l and if at the same time SW F,, converges strongly
to some vector 0 E ? l then 0 is uniquely determined by W and we may extend
the domain of S by defining SW = 0. Otherwise put, if Wn E D, !! Wn 11—* 0 and
SW„ converges strongly then lim SWn = O. We use the letters S and F for the
closures of the operators defined by (V.2.1), (V.2.2). A closed operator has a
unique polar decomposition which we write

S = Je] 1 / 2 . (V.2.3)

.6 is a self adjoint, positive operator (unbounded in general) and J is a conjugate

linear isometry, in our case J is an antiunitary operator, since D is dense in 1-1.
From the definitions one finds the relations

FS = d; SF. d-1 ; (V.2.4)

J=J* =J-1 ; (V.2.5)

J41I2 As? = A` 112 JA,f2- = A * (2; (V.2.6)
Afl=.fQ; J(2=,fl. (V.2.7)
Defining the 1-parameter unitary group

U(t) = zut (V.2.8)

one has
JAJ = A -1 ; JU(t)J = U(t). (V.2.9)
The key theorem is now

Theorem 2.1.1 (Tomita-Takesaki).

Let R be a von Neumann algebra in standard form, .0 a cyclic .and separating
vector and 4 J, U(t) as defined above. Then

JRJ = 1Z , (V.2.10)

U(t)RU* (t) = R,
(V.2.11)
U(t)R' U* (t) = R'
for all real t.
218 V. Thermal States and Modular Automorphisms

Thus the map ut defined by

6t A - U(t)AU*(t), A E ?Z (V.2.12)

is an automorphism group of R. It is called the group of modular automorphisms

of the state to on the algebra ?Z. Correspondingly J is called the modular con-
jugation and d the modular operator of (R., ,f2).

We shall skip the proof of these claims. It can be found in many books e.g.
[Takesaki, 19701, [Bratteli and Robinson, Vol. I, 1979], [Kadison and Ringrose,
Vol. II, 1987], [Stratila and Zsido 1978], [Stratila 1981]. Let us try to elucidate
the significance of the theorem.
Writing
= e-K , thus U(t) = UK, (V.2.13)
we call K the "modular Hamiltonian". K is a self adjoint operator whose spec-
trum will extend in general from —oo to +oo and which has Q as an eigenvector
to eigenvalue zero. The analogy of equations (V.2.1) through (V.2.12) with the
relations discussed in theorems 1.4.1 and 1.4.2 in connection with the represen-
tation induced by a KMS-state is striking. Indeed, a simple calculation shows
that w satisfies the KMS-condition with respect to the automorphism group of
for the parameter value ,Q = 1 :

_
—

w ((QtA)B) -~ (.f21 AU*(t)Bl.f2) = (J[] 1 /2 Af21 U*( t )JA 112 B*.(2)

= (J.6112 A.f2I JU*(t)av 2 B*S2) (U*(t)[]1/2 B*Q14 1 /2 A,f2)

= (.f21 B.6 1/2 U(t) 41/2 A.f2)
(Q1 Be -K- iKt AlQ) _ (.Q1 Ba t_fl A(2). (V.2.14)
We have used U(z),f2 = Q, relations (V.2.6) and (V.2.9) and the antiunitarity
of J, i.e. (JWtJO) = (01W). Comparison with (V.1.7) shows that (V.2.14) is
the KMS-condition if we put
ut = o _pt . (V.2.15)
The (unfortunate) change of sign of the parameter results from the definition
(V.2.8), (V.2.12) due to which the parameter of the modular group becomes
proportional to negative time in statistical mechanics.
The striking conclusion for physics is:
An equilibrium state with inverse temperature 0 may be characterized as a faith-
ful state over the observable algebra whose modular automorphism group a,- (as
a group) is the time translation group, the parameter r being related to the
time t by t = —07 - . The extension of the state to the algebra of operations, the
field algebra in the sense of Chapter IV, is a faithful state whose modular auto-
morphism group is a 1-parameter subgroup of the symmetries composed of time
translations and gauge transformations.
One salient feature of the KMS-property, and therefore also of modular
automorphisms, is the fact that it generalizes the positive energy requirement.
 V.2 Modular Automorphisms and Modular Conjugation 219

While the latter corresponds to the analyticity of w(Ba tA) in the whole upper
half plane Im t > 0 (Chapter II.2) we have, for an equilibrium state with inverse
temperature /3, analyticity in the strip 0 < Im t < ,Q and for a faithful state
with modular automorphism group a t analyticity of w(BaT A) in the strip —1 <
Im r < 0. The modular Hamiltonian (V.2.13) is not positive but, with respect
to 1Z, its negative part is "suppressed" while, with respect to R.', its positive
part is suppressed in the following sense: Let ES» be the spectral projection of
K for the interval [—oo, —# ], EK+ ) the spectral projector for the interval [tc, oo],
taking c positive in both cases. Then

I E» 4 I I < e-k ; II E^ L T ' I I < e-k ; JES+1 J = (V.2.16)

Therefore we get from (V.2.6)

' "2 II
I E,(»AQ II < e -

A* Q II < e-x/2 II A II, for A E 7Z. (V.2.17)

Similarly
I El+)A'.f2 II < e-' 2 II A' II, for A ' E 7Z . (V2.18)
Thus any vector in 1Z,f2 has exponentially decreasing components in the negative
part of the spectrum of K. The vectors in [Link] are in the domain of L\ for
0 < a < 1/2, the vectors in 1Z's2 are in the domain of for 0 < a < 1/2.

V.2.2 Vector Representatives of States. Convex Cones in ? -t

Another important fact is that any normal state cp of a W*-algebra has a rep-
resentative state vector 0 E x in the standard form representation

cp(A) = (0}AIP); P E 7-t, A E 7Z. (V.2.19)

Thus the folium of normal states is already provided by the vector states in this
representation and we do not have to resort to density matrices. Of course the
correspondence cp —* 0 is one too many. One has

Lemma 2.2.1
Two vectors 0 1 , 02 are representatives of the same state iff there is a partial
isometry V E 1Z' from a subspace E1 1-t to a subspace E27-t such that

02

This means, more explicitly, there are projectors E a E 7Z (i = 1, 2) satisfying

'

E1 = V* V; E2 = VV * ; V E R.. (V.2.20)

The subspaces Et 7-t are the "cyclic components" of 0 t i.e.

=
220 V. Thermal States and Modular Automorphisms

One can make the correspondence cp —) 0 unique if one subjects 45 to further

conditions. There are the two important relations

421Ad U BI,f2) = (,f2IB211-v 40); 0 < u < 1, A, B E R. (V.2.21)

(,f2lA*A4 112 B*BI,f2) > 0. (V.2.22)
The first follows from

(,f2IAZVBI,f2) = (d -1 /2 JA,f2l.6U1d -1 /2 JB*0) = (JAf2IJ.6 1-vB*I,f2)

= (4 1-U B*Q1 Aft) = (QI Bd 1-v AI ,f2).

The second by

(01A*A4 1/2 B*BI,f2) _ (0IA*AJB*B(2) _ (.f2IX*X1.f2) > 0

with
X = A j(B)
where we have used the abbreviation

j(B) - JBJ, (V.2.23)

and the fact that j maps R. on the commutant so that we can shift j(B*) to
the left.
One introduces now for each a in the interval [0, 1/21 a convex cone of
vectors in f
Pn-Qa^+10), 0 <a<2, (V.2.24)
where WE is the set of positive elements of R. and the bar denotes the clo-
sure in the norm topology of Hilbert space. Equation (V.2.22) shows that for
W E Pn, 45 E P1,1
2--a we have (WIC) > 0. A closer study shows that P12-a is
precisely the polar (or dual) cone to Pn i.e. the set

P1^ 2-a = {0 : (01!) > 0 for all W E Pn}. (V.2.25)

Since
d°`A*A1,(2) = 2,a-1/2 j(A* A)1Q)
the cones may also be obtained as
Pa = Qa-1/2R'+IQ) (V.2.26)
n —

In particular
P il1 2 =R '+I,n). (V.2.27)
With these definitions one obtains now the following theorems which evolved
mainly through the work of Araki and Connes. We state them without giving
the proofs. They are given in the book by Bratteli and Robinson Vol. II where
also the references to the original papers may be found.
 V.2 Modular Automorphisms and Modular Conjugation 221

Theorem 2.2.2
Every normal state on R. has precisely one vector representative in each of the
cones Pn for 0 < a < 1/4. A state cp which is dominated by w in the sense
Aw --- cp > 0 for some A > 0 has a (unique) vector representative also in the
cones with 1/4 < a < 1/2.

Of special significance is the cone Pp, called the natural cone, which we
shall denote simply by PQ. It is self dual. We list the main properties of Po in
the next theorem. We shall omit indicating the reference state w and its vector
representative Q to which the objects J, d, U(t), P are related.

Theorem 2.2.3
Let W, 0 E P. Then

(i) (WI4') = (45 IW) ? 0, (V.2.28)

(ii) U(t)W E P, (V.2.29)
(iii) JW E P. (V.2.30)

Conversely, if W E 7-1 and JW = W then

W = W+ — W_ with W+ E P, W_ E P and (W+ Icl!) = O. (V.2.31)

(iv) P may also be characterized as the closure of the set

Aj(A)1Q), A E R. . (V.2.32)

(v) If 0 E P is separating for R then it is also cyclic for 7Z and vice versa.
(vi) Let 01 , 02 be the (unique) vector representatives of the states (pi, cp2 then

II 01--02 11 2< II (Pi -(P211 2< II 'P1 - 'P2 II II 'P1+ 42 II . (V.2.33)

Most remarkable is the property (vi). It means that the map from the set
of normal states over 7Z onto the set of representative vectors in P is a home-
omorphism with respect to the norm topology in state space on the one side
and the Hilbert space norm topology on P. Of course, if two state vectors are
close together then also the corresponding states will be close together. This
is the second of the inequalities (V.2.33); it does not need P. The converse,
however, the fact that closeness of states implies the closeness of their vector
representatives is special to the natural cone P. It does not hold if the vector
representatives are chosen on any one of the cones Pc' with a 1/4.
Another remarkable consequence of theorem 2.2.3 is

Theorem 2.2.4
Every automorphism a of a W*-algebra is implementable in a standard form
representation by a unitary U(a) i.e.
222 V. Thermal States and Modular Automorphisms

aA = U(a)AU*(a); A E R, U(a) unitary from z(7-t). (V.2.34)

U(a) may be chosen so that

U(a)P = P; JU{a) = U(a}J. (V.2.35)

Comment. Theorem 2.2.4 shows that a W*-algebra is a very rigid object. If an

automorphism of a C*-algebra 2i cannot be implemented in a representation 7r
then it cannot be extended from 2L to the von Neumann algebra ir(2t)".
Let us describe the construction of U(a) implementing the automorphism
a as stated in theorem 2.2.4. The state w' defined as

w'(A) = w(a -1 A)

is faithful. Hence it has a vector representative W E PD which is separating for

R and thus, by theorem 2.2.3 (v), also cyclic. Set

U(a)Al,fl) = (aA)IW) . (V.2.36)

Due to the cyclic and separating property of both 9 and W (V.2.36) defines
U(a) on a dense domain and the range is dense too. Then

II U(a)A19) 112=11 aAIW) 11 2= w'(aA*A) = w(A*A) —11 AIQ) 11 2 -

Therefore U(a) can be extended to a unitary operator on 7 -t. From the definition
(V.2.36) one checks the relations (V.2.34), (V.2.35).

V.2.3 Relative Modular Operators and Radon-Nikodym Derivatives

We study next the dependence of modular automorphisms on the reference

state. Let w, w' be faithful normal states and Q, Q' vector representatives
which are cyclic and separating for R. Then one can define the conjugate linear
operator S n by
S1',QAI Q) = A* I.Q' ); A E R. (V.2.37)
It is closeable and hence has a polar decomposition
1/2 (V.2.38)
S',n = J'42(d2',r2)
The positive operator .6 .0, ,f2 is called the relative modular operator of the pair
Q', Q. If a is on the natural cone of S2 then

Jo, 1(2 = J,r242 = J. (V.2.39)

Defining the unitaries

Ur2',.o(t) = (dn',.(2)` t (V.2.40)
one finds for any pair of cyclic and separating vectors
 V.2 Modular Automorphisms and Modular Conjugation 223

UQ# ,n (t)AUU, n (t) = Un, (t) AWL, (t); A E 7Z, (V.2.41)

Un, n (t)A UQ' n (t) = UQ(t)A' UU(t); A' E R.' (V.2.42)

where U2(t) = UQ,n(t) denotes the unitary defined in (V.2.8). Defining for a
triple of such vectors

Up/ ,n (t) = U ',n„ (t)Un,n" (t), (V.2.43)

, ( ),
U n, (t) U^„ n t
z^ n, n (t) - ^„ (V.2.44)

one sees from (V.2.41), (V.2.42) that ?La, n commutes with R.' and does not
depend on Q" (as indicated in the notation). Thus

un',n(t) E R, (V.2.45)

un',n(t) E 11. (V.2.46)

Furthermore the unitaries (V.2.43), (V.2.44) are independent of the choice of
the vector representatives of the states w, w . In their dependence on t they are
'

strongly continuous.
To illustrate the significance, let us look at the example afforded by Gibbs
states of a finitely extended system as described in section 1. For such a state
with density matrix P the modular operator factors into a part from R and one
from R.'
An = Po .?(Qnl). (V.2.47)

Writing
H (V.2.48)
Pn = e
-,^

we see from (V.2.43) that all the operators UQ'n, .6 12, 42 also factor. We get

Un',n(t) = e-itHn ^ e it j(H0) 3 (V.2.49)

-itHn^ eitHn
un,, n(t) = e , (V.2.50)
a form familiar from collision theory in quantum mechanics. Introducing the
relative Hamiltonian
hn' n = H n , — HD (V.2.51)
one may compute Un, n and en, by standard perturbation expansion with re-
spect to h n' n as described in subsection 2.4 of Chapter II.
The special feature here, the factorization of 4 corresponds to the fact
that in this case the modular automorphisms 611' are inner, i.e. they may be
implemented by the unitaries exp(—iH n t) E R.. This feature is lost in general,
for instance already in the thermodynamic limit of Gibbs states. Still there
always remain the unitaries n(2, 42 (0 E R. . Relation (V.2.50) is replaced by the
cocycle identity

un' ,n (tl + t2) = u n ',n( tl ) ( o u Q' Q (t2 ) ) (V.2.52)

224 V. Thermal States and Modular Automorphisms

from which, if of is given, u n' ,n (t) may be computed by perturbation expansion

with respect to the relative Hamiltonian

a
hn' ,^a = at (V.2.53)
,9 t= o

Thus any pair of faithful, normal states on 1Z determines a cocycle of unitaries

in 1Z. If a is inner then this cocycle is of the form (V.2.50); it is a coboundary
in the terminology of group cohomology.
The cocycle u n, n (t) which characterizes the relation between the two states
is called the Radon-Nikodym
) cocycle and is written as

(Dw' : Dw)(t) , n (t) E 1Z. (V.2.54)

Besides the cocycle identity (V.2.52) it satisfies the chain rule

(Dwi : Dw 2 )(t) (Dw 2 : Dw3)(t) = (Dwi : Dw3 )(t), (V.2.55)

has the intertwining property

(Dw i : Dw2 )(t) (Qt 2 A) = (or A) (Dw i : Dw2 )(t), (V.2.56)

and the initial condition

(Dw1 : Dw2 )(0) = 11. (V.2.57)

V.2.4 Classification of Factors

If, for a particular value of t and a particular faithful, normal state w the modular
automorphism of is inner (i.e. can be implemented by a unitary from 1Z, which
implies that Un(t) factors into a part from 1Z and one from 7Z') then (V.2.56)
shows that for any other normal state w also is inner. Therefore the set of
t-values
T = {t : Qt is inner} (V.2.58)
is a property of 1Z, independent of the choice of w. If 1Z is not a factor then T is
the intersection of the sets T,, attached to the factors 1Z,, occurring in the central
decomposition of 1Z. Therefore T does not have much information unless 1Z is a
factor. We specialize to this now. Obviously always 0 E T and, with t1, t 2 E T
also t1 ± t2 E T. So T is a subgroup of IR, the group of real numbers (in additive
notation) .
There are three "simple" possibilities for T:

a) T = Ift,
b) T= nto , n E71,
c) T = {0}.
 V.2 Modular Automorphisms and Modular Conjugation 225

But T could also be some dense subset of IR since, so far, we have not equipped
T with a topology. But, as Connes has shown, T is related to the spectrum of
the modular operators d„, [Conn 73, 74]. Defining the spectral invariant

S(R) = l,Spect4 , (V.2.59)

(where w ranges over all normal states of R) and the set

1'(R) = {a E IR : eiat = 1 for all t E T}. (V.2.60)

Cannes showed that

1'(R) D ln(S(R) \ 0) (V.2.61)
and that S(R) \ 4 is a closed subgroup of the multiplicative group of positive
real numbers. The three cases listed above correspond to
a) S(R) = {1},
b) S(R) = {0 u .V }, n E 71; 0 < \ < 1,
c) S(R) is the set of all non — negative reals.
The remaining possibility is
d) S(R) = {0, 11.

In case a) one can define a semifinite trace over R [ Takesaki 1971. If o

is inner for all t then Z factors as in (V.2.47) where P is a selfadjoint opera-
tor affiliated with R and commuting with d. This yields the following formal
relations (provided that the expressions occurring exist)

w ( o°`A) = w (Q ° Ap') , (V.2.62)

w (Q1AB) = (,(QI Aj (g- 1 ) BIQ) = (,f2IAdP - 1 B1Q) = (Q10- 1 BAI,fl),

(V.2.63)
where we have used (V.2.20) and finally (V.2.21). Setting

trA = w (g-1 A) = w (Q_1/2AQ_1/2) (V.2.64)

one sees that this defines a positive linear functional over a dense set in R with
the property
tr AB = tr BA.
So (V.2.64) defines a semifinite trace. Therefore case a) contains the factors
of type I and II in von Neumann's classification. Case b) corresponds to the
Powers factors [Pow 67]. They are denoted as type IIIA (0 < a < 1). Case c)
are the factors denoted as type III i . Case d) is denoted as type III ° (one cannot
quarrel with notations).
We shall see that case c) (type II h ) is of paramount interest for us. The
representation of the quasilocal algebra induced by an equilibrium state as well
as the local algebras of relativistic theory in the vacuum sector are of type
226 V. Thermal States and Modular Automorphisms

IIIi . In fact we shall see in Section 6 that these algebras are isomorphic as
W*-algebras to the (unique) hyperfinite factor of type III i .
The special feature of type III I is that in this case the spectrum of each 44,
is maximal (all of 1R+ ). S(R.) is already given by the spectrum of the modular
operator of any single state. If one can exhibit a faithful primary state w such
that Spect4, is minimal in the folium i.e. that S(R.) is already given by SpectL
then the possibilities are very limited. One can exclude case d) because it would
mean that zero is an isolated, discrete eigenvalue. Furthermore case a) would
mean then that 44, = Il i.e. that (.0 is a trace state. This special situation prevails
only for equilibrium states at infinite temperature (@ = 0). In this limiting case
7Z is of type II I . Apart from this and the very special case of Powers factors
this leaves as the generic situation only type III I .
In this context the following observation of Araki [Ara 72] and Stormer
[Storm 72] is important.

Theorem 2.4.1
Let 2t be a C*-algebra which is asymptotically Abelian with respect to anrauto-
morphism group, w an invariant, primary, faithful state over 2t, iÎ41 the GNS-
representation of 2t induced by w and 7Z = 7r w (2t)". If the extension of w to R.
is faithful then
Spect441 = S(R). (V.2.65)

Comment. Asymptotically Abelian shall mean that there is an automorphism

a of 2t such that

jf [B, an A] II-- 0 as n --* oo; A, B E R. (V.2.66)

Invariance of w:
w(aA) = w(A). (V.2.67)
Typically, if 2t is the quasilocal algebra and a corresponds to space translations
by some 3-vector then (V.2.66) is satisfied. If w is an extremal. KMS-state for
some finite temperature which is homogeneous (or at least invariant under some
subgroup of translations as in the case of a crystal) then we have (V.2.67) and
w is also faithful on R.
To prove the theorem one has to show that for any state w' in the folium of
w the spectrum of /& contains the spectrum of 4 4,. This results from the fact
that under large translations an (n — ^^ oo) every state in the folium tends to a
multiple of w. The arguments are reminiscent of the discussion of the structure of
the energy-momentum spectrum (theorem 5.4.1 in Chapter II ). In fact one can
see rather directly that Spect4 41 is a group and thus, since the spectrum is closed
and 4k1 is positive, a closed subgroup of the non-negative reals. Namely, we know
that 1 E Spect441 that with a E Spect4„, also a -1 E Spect4„,, (this follows
from (V.2.9)). It remains to show that SpectA„, is closed under multiplication.
This corresponds to the additivity of the spectrum of the modular Hamiltonian
K. The faithfulness of w and its invariance under a imply that a commutes with
 V.3 Direct Characterization of Equilibrium States 227

the modular automorphism group. Therefore the additivity of the K-spectrum

can be established with the technique used in the proof of theorem 5.4.1 in
Chapter IL

V.3 Direct Characterization of Equilibrium States

V.3.1 Introduction

In the last section we argued that an equilibrium state of an infinitely extended

medium can be characterized as a state over the algebra of local operations
satisfying the KMS-condition at some value of /3 for some 1-parameter subgroup
of automorphisms generated by time translations and gauge transformations.
Obviously this characterization is far from "first principles". It resulted from
the Gibbs Ansatz by passing to the thermodynamic limit. The Gibbs Ansatz in
turn was justified by combining the requirement of stationarity of the state with
some assumptions about the ergodicity of the dynamics or, alternatively, by the
principle of maximal entropy for states having specified expectation values for
the relevant constants of motion.
The objective of the present section is to give a direct and natural definition
of equilibrium for the infinitely extended medium and derive the KMS-condition
from it. Again we shall have two roads. The first uses a concept of dynamical
stability, the other, starting from the second law of thermodynamics, the notion
of passivity. Both lead to the result that an equilibrium state is a KMS-state over
the algebra of observables with respect to the time translation automorphism
group.
There remains the second task, to elucidate the rôle of the chemical poten-
tials. We shall see that the existence of charge quantum numbers in the sense
of Chapter IV implies, generically, the existence of a continuum of different
KMS-states over the observable algebra with respect to a t at the same value of
the temperature; each such state has an extension to the algebra of operations
(the field algebra in the sense of Chapter IV) and these extensions are KMS-
states with respect to a modified automorphism group at over the algebra of
operations.

V.3.2 Stability

Let 21 be the C*-algebra of quasilocal observables and a t the automorphism

group representing time translations. A (pure phase) equilibrium state is cer-
tainly a primary state over 21 which is stationary (invariant under time trans-
lations)
w(at A) = cv(A). (V.3.1)
But this alone may not suffice to distinguish equilibrium from other states. An
equilibrium state should not be too sensitive to small changes in the dynamical
228 V. Thermal States and Modular Automorphisms

law (adding a few grains of dust). This old idea was formalized by Haag, Kastler
and Trych Pohlmeyer [Haag 74] as a requirement of "dynamical stability". Sup-
pose the dynamical law is slightly changed, corresponding to a change of the
Hamiltonian by ) h, where the "coupling constant" A shall ultimately tend to
zero and h is a fixed element of 21. This is expressed by replacing a t by 4h
defined by
l ath A = ia[h, A], (V.3.2)
dtat t=o
ah . a Ah
t+t i = aah
aAh ta t ' = id. (V.3)
t t^o
f
From these relations one can compute 4h in terms of a t and h, for instance by a
norm convergent perturbation expansion as described in [Bratteli and Robinson
1981]. Introducing the cocycle of automorphisms (compare (V.2.52))

f t h = ct-laAh
n a (V.3.4)

one has
[h, athA], (V.3.5)
1
— —iA (cx) [h, at A]. (V.3.6)
If co is a primary state, stationary with respect to the dynamical law a t ,
then we say that w is dynamically stable with respect to the perturbation h
if, for sufficiently small A, there exists a state coA h in the primary folium
of co which is stationary with respect to the perturbed dynamics 4h and
such that the family co ati depends continuously on A and tends to w for
A —> O. To simplify the arguments we shall even require differentiability. Thus
ço(A) = d/dAwXh (A) a-o shall be a normal linear form in the folium of co. With
this terminology we shall define a (pure phase) equilibrium state as a primary
state over 21 which is

i) stationary with respect to at ,

ii) stable with respect to any perturbation of the dynamics by h E D, where

D is a norm dense set in 21.

This definition is effective in the case of an infinite medium, not for a system
bounded by walls In the latter case things are more complicated due to recur-
rences caused by reflection from the walls. This is analogous to the situation
in collision theory. The concept of an S-matrix can be introduced only if in-
finitely extended space is available. Otherwise, though the effects of walls far
away may be insignificant, they prevent this idealization. In fact, the analogy
is very close. The essential rôle in the subsequent arguments will be played by
asymptotic Abelianness of 21 with respect to time translations. The physical
picture for this is the following. The effect of an operation in a finite space
region at time to on the outcome of later observations in another fixed, finite
 V.3 Direct Characterization of Equilibrium States 229

space region should decrease in strength as the time difference increases. Thus
the commutator [A, a tB] should tend to zero as t -+ oo. Specifically we shall
base the discussion on

Assumption 3.2.1 (C (') Asymptotic Abelianness in Time).

There is a norm dense subset D C 21 such that II [A, at B] II is an absolutely
integrable function of t when A, B E D.

Comments. According to the intuitive picture one should think of D as the

set of sharply localized observables. For a system on non-interacting particles
II [A, atB] II decreases generically like ItI-3/2 and the same can be expected in
case of purely repulsive interactions. The assumption imposes limitations on the
the type of attractive interactions (e.g. saturation properties as mentioned at
the end of section 1) and it may be that the assumption does not cover all cases
of physical interest. We shall not discuss the possibility of replacing assumption
3.2.1 by weaker conditions. Some elaborations on the status of the assumption
are given in [Bratteti and Robinson 1981]. Note that the assumption concerns
an algebraic property, a property of the system not referring to any particular
state. It implies that for every primary state w the correlation functions tend
to zero for large time differences. Thus, if w is primary and stationary and as-
sumption 3.2.1 holds, one conludes (by the same reasoning as in the proof of
theorem 3.2.2 of Chapter III) that

Ic.4)(Aa t B)I - w(A)u)(B)I O as it! -

^^ oo. (V.3.7)

This does not specify the rate of decrease. The system may possess different
types of equilibrium states (e.g. phase transition points of different order) in
which the correlation functions have different rates of decrease for large sepa-
ration in space and time and only the differences of the correlation functions
F - G are absolutely integrable.
From (V.3.6) we get for a state which is invariant under aah
t
((h)_1
(vah A)) = c2h (A) -- ia f (211 ([h, at, AD dt'. (V.3.8)

By assumption 3.2.1 the limits for t -^^ ±oo exist for A E D. The normalized
positive linear forms on D
1A
W± (A) = lim (.1) ah ( (Om') ) (V.3.9)
t^f0o

define, by extension, states on 21. Since

(ith)-1 at' = ^h ah 1
(V.3.10)
t' (Qt +t' )

the states w ± are invariant under at i.e.

c.)f (at A) = cl)±(A). (V.3.11)

230 V. Thermal States and Modular Automorphisms

Now we argue that due to the dynamical stability requirement the states
w+ , w_, wrh, w all must lie in the same primary folium for sufficiently small
since two primary states with norm distance less than 2 lie in the same folium
by theorem 2.2.16 of Chapter III. On the other hand a primary folium can
contain at most one state which is invariant under an automorphism group with
respect to which 2L is aysymptotically Abelian. This follows from the argument
establishing lemma 3.2.5 in Chapter III. Therefore we conclude

w+ =w_=w. (V.3.12)

Combining (V.3.8), (V.3.9), (V.3.12) and II wah - w O we get the following

expression of the stability requirement

j_00 CA) ([11 , cttAD dt ---- 0, h,A E D. (V.3.13)

Remarks. (i) Narnhofer and Thirring [Narn 82] showed that the condition
(V.3.13) may also be interpreted as the condition for adiabatic invariance of
w (instead of dynamical stability). One considers the effect on w when the dy-
namical law is changed by a local perturbation which is slowly switched on and,
as t oo, slowly switched off again. Adiabatic invariance means that the state
returns to its original form at the end of this procedure.
(ii) It is also instructive to compare (V.3.13) with the treatment of a
metastable state in quantum mechanics. One standard approach starts from an
approximate Hamiltonian H0 which has a discrete eigenvalue inside a continuous
part of the spectrum. Computing the spectral projectors of the true Hamilto-
nian H by perturbation expansion with respect to the difference h — H Ho —

one finds generically that the discrete eigenvalue disappears; it is dissolved in

the continuum. A discrete eigenvalue survives only if h is such that (V.3.13)
holds for sufficiently many A E 93(7 -t) (where w denotes the original discrete
eigenstate of Ho and at A = eiHOtAe_ 0t). In our case the equilibrium state of an
infinitely extended medium is a discrete eigenstate of the effective Hamiltonian
in the representation 7r u, and it should remain generically a discrete eigenstate
if the dynamics is changed slightly in some finite region. Hence the requirement
of (V.3.13) for all h E D.

In the notation of (V.1.9), (V.1.20) condition (V.3.13) becomes

FA,h(0) = GA,h(0), (V.3.14)

i.e. it is the (Fourier transform of the) KMS-condition at the value c = 0. It is

now very remarkable that (V.3.13) together with the requirements that w shall
be primary and stationary implies already that w is a KMS-state with respect
to at for some value of the inverse temperature 0. The proof presented here
uses the assumption 3.2.1 and combines arguments from [Haag 74] with the
method described in [Haag 77, appendix] . For more details and variants of the
 V.3 Direct Characterization of Equilibrium States 231

assumptions see [Bratteli and Robinson 1981]. Let us set for h and A in (V.3.13)
respectively
h = hl aT h2 , A = Aicx A2 .
This is legitimate for finite T since h, A E D when hk, Ak E D (k = 1, 2). We
obtain

0 = J w(I + II + III + IV )dt, (V.3.15)

I = hi [aTh2, atAi]at+TA2,
II = hi(atAi)[a7h2, at+TA2], (V.3.16)
III = [h1, atAi1(at+TA2)aTh2,
IV = (atA1)[hi, at+TA2]aTh2.
All four integrals are finite, continuous functions of T and we may consider their
limits as T -> oo. For the contribution of I we get, setting t = T + t' and using
the invariance of w

T
((_T h1)[h2, at' Al]a t , +T A2 dti + II h1 II II A 2 I! 1 I>Tll[h2, a' dt'
Ir
In the first term we may take the limit T -> oo under the integral sign. For pri-
mary w it approaches (due to asymptotic Abelianness and the cluster property)

w (h1)w(A2) f [h2i at' A 1 ]dt'.

T

Thus it vanishes in the limit T —* oo due to (V.3.13). The second term also
apprôaches zero as T — ^^ oo since the integrand is an GM-function. Thus
lim f w(I)dt = O as T —^^ oo and the same is true for lim f w(IV)dt. The norms
of II and III are bounded by L (1)-functions of t which are independent of T.
Hence the limit may be taken under the integral sign. This yields

lim ^ f
T^00 00
CO(/' + III)dt

= f (w(hiatAi)w([h2, atA2]) + w([hi, atAi])w(atA2)h2)) dt

= f (Fi (t)F2 (t) — G l (t)G2( t )) dt,

with the abbreviations

Fk - FAk,hk; Gk = GAk,hk • (V.3.17)

Due to (V.3.13) we may subtract in F and G the uncorrelated parts (see the
remarks in connection with (V.1.19)). (V.3.15) becomes

f (F1(t)F2 (t) — G 1 (t)G2(t)) dt = 0 (V.3.18)

232 V. Thermal States and Modular Automorphisms

and, repeating the process with three factors we get

f (Fl (t)F2 (t)F3 (t) — G l (t)G2 (t)G3 (t)) dt = 0. (V.3.19)

We can write (V.3.18) as

f Pi ( —e)F2 (e)de = fÔi (_e)ã2 (e)de, (V.3.20)

and

FA,h(t) _ (fljhU(t)AjQ) — (flI hI Q)(QI AI Q), U(t) = etxt, (V.3.21)

where U(t) implements a t . Applying the standard technique (see Chapter II,
lemma 4.1.2) we can choose A2, h2 so that F2 is non vanishing only in an-

arbitrary small interval around a chosen point e o . For this purpose set

A2 =f f ( O at, Bd e , h2 = A2, (V.3.22)

-4E 't' d'.

f (t) = f f(e )e t; (V.3.23)

Then F2(e) and 62(e) vanish if e is outside the support of f. If co is in the spec-
trum of H, we can choose B so that F2 is non-vanishing inside the support of
f. The choice (V.3.22) makes F2 (e) positive; we can arrange that f F2(e)de = 1
and shrink the support. In other words we can, by suitable choice of B and f
let F2 approximate a 8-function arbitrarily well. Simultaneously 62 (e) will ap-
proach a multiple of a 6-function (possibly 0). Assuming that F l is continuous'
we conclude from (V.3.20), for ea E Spect H

FAi,h, (—eo) = CGAi,hi ( — eo),

where c results from the ratio of the integrals of 6 2 and F2 in the limit. The
essential point is that c is independent of A l and h 1 but will depend on c o since
it results from the choice of A2, h2 as described above. If both co and —Co are in
the spectrum of H then, for suitable A 1 , hl , both F1 and G, are nonvanishing.
We have

FA,h(e) = (e)b- A,h(e) if e and — e are in Spect H. (V.3.24)

From the definition of F and G follows

FA,B ( C) = GB ,A ( —e) , (V.3.25)

1 This will be the case if the correlation function F1 (t) and not only the difference F — G
is square integrable. The conclusion of the subsequent argument remains valid under weaker
assumptions. For instance, if for certain states (such as at a critical point) the correlation
functions should decrease more slowly at infinity then one can expect that derivatives become
LO ) -functions and so F(e) is continuous with the exception of the point e = O. This does not
affect the main conclusion.
 V.3 Direct Characterization of Equilibrium States 233

which leads to
C—e) = 0(e) -1 . (V.3.26)
Further information about the function Ce) is obtained if one applies the same
technique once more to the relation (V.3.19). This yields

0(6)0(e r — e) = 0(e'), (V.3.27)

provided that ±e, ±e' and ±(e' — c) are in the spectrum. Therefore, if Spect H
is the whole real line

45(e) = de for some value of Q. (V.3.28)

This is the KMS-condition.

Next we show that assumption 3.2.1 implies that Spect H is either all of IR
or it is one-sided i.e. a subset of {0 U IR+ } or of {0 U IRT}. One has

Lemma 3.2.2
Assume 3.2.1. Then in a primary folium

(i) el, e2 E Spect H implies el - e 2 E Spect H.

(ii) There can be no isolated point in spect H besides (possibly) e = O.

Proof. Part (i) is the frequently encountered additivity of the spectrum which
follows from the assumption as in theorem 5.4.1 of Chapter II. Part (ii) fol-
lows from the argument proving lemma 3.2.5 in Chapter III, using asymptotic
Abelianess for time translations instead of space translations.

Theorem 3.2.3
If co is a primary, stationary state and assumption 3.2.1 holds then there remain
the following three alternatives for Spect H

a) Spect H = IR;

b) Spect H C {0 U IR +}, co is a ground state;

b) Spect H C {0 U IR - }, w is a ceiling state.

Proof. We must only show that if Spect H contains two values of opposite sign,
say a and b (a > 0, b > 0) then every point in R is in Spect H. By part (i)
—

of lemma 3.2.2 na — nib E Spect H for all positive integers n, Tn. Since Spect H
is closed, part (ii) of the lemma tells us that we may assume that a/b is not a
rational number. Then the set na — nib is dense in IR and Spect H = IR..

We have finally as a consequence of assumption 3.2.1

Theorem 3.2.4
A stationary, primary state w which is dynamically stable satisfies the KMS-
condition with respect to a t for some value of the parameter 0 unless it is a
234 V. Thermal States and Modular Automorphisms

ground state (which may be regarded as the limiting case for )3 = oo) or a
ceiling state (the limiting case )3 = —oc).

Comment. One should take note that no information about the sign of ,Q is pro-
vided by theorem 3.2.4. On the other hand, given a dynamical system (21., at ),
the set of values {l} for which KMS-states exist is determined. It appears that
one fundamental property of the net of local algebras and their geometric auto-
morphisms must be that no KMS-states to negative values of ,Q exist. This may
be regarded as an extension or generalization of axiom A3 in Chapter II which
plays such an essential rôle in quantum field theory. Compare also the concept
of passivity discussed in the next subsection. It is probable that this feature can
be reduced to a principle concerning the structure of the net in the small (local
definiteness, together with a characterization of the germ of the theory such as
attempted via a concept of "local stability" by Haag, Narnhofer, Stein [Haag
84]) .
What is the significance of the parameter ,3 (the inverse temperature in
physics) in this context? From
w Ah (A) = lim c.,1 (flÀhA) (V.3.29)
t oo
^

and (V.3.5) we get

oo oo
Wah(A) =if coo ([h, atA]) dt = i f (FA,h(t) — GA,h(t)) dt.
da a=o
Due to the analytic properties of F(z) this may be replaced by an integral along
the imaginary axis from O to if3

d^ W ah (A) = — f (FA,h(i'Y)) 0. (V.3.30)

a—d
Therefore the change of the state due to a small local perturbation of the dy-
namics is - at least for small Q (high temperature) - proportional to i. One may
regard ,Q as a stability parameter: The sensitivity of an equilibrium state to a
change in the dynamical law decreases with decreasing fi. At infinite tempera-
ture (3 = 0) the state is completely insensitive.
Let us add one last remark to the topic of stability and equilibrium. It is
intuitively suggested that an equilibrium state which is not at a phase transition
point possesses higher stability. One expects that it will satisfy (V.3.13) even
for homogeneous perturbations i.e. even if h is no longer in 2t. Therefore it is of
interest to study the degree of stability of different states, considering changes
of the dynamics of the form

h = f .f (x)oz h dix,
where h' E 21 and lim Ixri f (x) I= O as 'xi — ^^ oo. For n > 3 we have essentially
local perturbations, n = O allows almost homogeneous perturbations. The value
 V.3 Direct Characterization of Equilibrium States 235

of n for which the stability breaks down relates to the rate of decrease of cor-
relation functions in space-time and is one characteristic of different types of
phase transitions.

V.3.3 Passivity

Another direct characterization of equilibrium states was given by Pusz and

Woronowicz [Pusz 78]. It implies that an equilibrium state is a KMS-state over
21 (with respect to time translation) for some non-negative value of the param-
eter,3. One may start from the observation

Lemma 3.3.1
Let w be a KMS-state over the C*-algebra 21 with respect to the automorphism
group at and parameter ,8. Denote by

ôA = (V.3.31)
dt atA t -o
the infinitesimal generator of a t , defined on a dense domain D(5) E 21. 2 Then,
for any unitary U e D(6) and for any self- adjoint A E D(5) one has

>0 for Q>0

—2W(U*ôU) (V.3.32)
< 0 for Q<0

>0 for 0>0

—iw(A6A) {—< (V.3.33)
0 for 3<0

Proof. Take the GNS-representation (7r w , Q, 7-1) in which a t is implemented

by the unitaries U(t) = eiHt and

7r,(6A) = i[H, irw (A)], A E 21, (V.3.34)

H,f2 = 0. (V.3.35)
The modular operator of 0 is d = e-QH . The basic relation following from the
KMS-property (see (V.1.50)) is
-
Je p/ 211 7rw (A)I0) = 7rw(A * )IQ) • (V.3.36)

For unitary U E 21 this yields

1 = (rw(U*)QI7rw(U*)Q) = ( 7r (U)QJe -13H1w(U) (2 )

or
(QI 7w(U * ) (11 — e-13H) irw(U) I S 2 ) = 0. (V.3.37)
Since, for a real variable x //
26 is an unbounded derivation on 21.
236 V. Thermal States and Modular Automorphisms

-x
x > 1—e
we get for U E D(5), using (V.3.35)

(Q 17rw (U*)/3H7rw (U) I Q) = — iI(QI rw(U * 5U)I Q) > 0,

— i/3w(U *SU) > 0. (V.3.38)

For self- adjoint A E D(8) one finds similarly

— i,dw(ASA) > 0. (V.3.39)

There is a converse to this lemma.

Theorem 3.3.2
If 21 is asymptotically Abelian with respect to a t and w is a primary state
satisfying either

— iw(U*8U) > 0 for every unitary U E D(6) (V.3.40)

or
— iw(ASA) > 0 for every self — adjoint A e D(S) (V.3.41)
then w is either a ground state or it is a KMS-state with respect to at for some
positive value of Q.

Note that (V.1.67) for A* = A reduces to

sign 8 ( iw(A5A)) > 0.

— (V.3.42)

So the theorem tells that in the case where 21 is asymptotically Abelian under
at and w is primary this special case of (V.1.67) suffices to prove that w is a
KMS-state (including the limiting situation oo).

Definition 3.3.3
A primary state w satisfying either (V.3.40) or (V.3.41) is called a passive state.

Let us make a few remarks concerning the proof of theorem 3.3.2. For a com-
plete proof the reader is referred to the original paper by Pusz and Woronowicz
or to volume II of Bratteli and Robinson.
First, one sees easily that (V.3.40) implies (V.3.41) by using for the unitaries
a family ejeA and expanding in powers of e. Next one sees that (V.3.41) implies
that w is stationary: If A* = A then (A 6A)* = 8A A and, if w(A 6A) is purely
imaginary then w(A 8A)* _ —w(ASA). So one has in this case

w(A 8A) + w(SA A) - w(6A 2 ) = 0,

which implies w(a t A2 ) = w(A 2 ). Since linear combinations of positive elements

in D(6) are dense in 2E one has
 V.3 Direct Characterization of Equilibrium States 237

w(at B) = w(B) for all B E 2t. (V.3.43)

Next one forms the GNS-representation induced by w yielding

(7r, Q, 1-1, U(t), H) with the relations (V.3.34), (V.3.35). By theorem 3.2.3
w may either be a ground state or Spect H = IR. The third possibility, that w
may be a ceiling state, is excluded by (V.3.41). If Spect H = R then one shows
that w must be faithful. So one can_ apply the machinery of the Tomita-Takesaki
theory. From (V.3.43) it follows that U(t) commutes with the modular operator
L. The last step is to show that then, under the assumptions of the theorem, the
modular Hamiltonian K must be a multiple of H (possibly zero). This problem
is similar to the one discussed in the last subsection (with dynamical stabil-
ity replaced by passivity) . We shall not reproduce here the somewhat lengthy
argument.
The most interesting aspect of theorem 3.3.2 is its relation to the second law
of thermodynamics. Suppose we can change the dynamical law, e.g. by applying
external fields in a controlled way. Studying the behaviour of the system under
different dynamical laws we should generalize the concept of "observable" to
that of "observation procedure" . 3 Recalling the discussion at the beginning of
section 3 in Chapter I, a procedure specifies on the one hand the intrinsic
construction of the measuring apparatus (the workshop drawing) and on the
other hand the placement of the apparatus within the space-time reference
frame of the laboratory. For the present purpose we shall combine the intrinsic
construction and all placement parameters with the exception of the time of
measurement into one symbol A but specify the time separately. Thus (A, t)
denotes a procedure and A will be mathematically represented by a self-adjoint
element of an algebra % which we might call the kinematical algebra and which
we may also identify with (2t, 0), the measurement procedures at t = O. Given
a dynamical law one has an equivalence between procedures at different times:
To every procedure (A, t 1 ) there is a procedure (B, t 2 ) so that in every state
the measuring results obtained by either one of the two procedures coincide.
This may be expressed by saying that a dynamical law defines a 1-parameter
family Tt of automorphisms of 2t and we can represent the procedure (A, t)
by the algebraic element TL A E 2t. If the dynamical law is not homogeneous in
time then Tt is not a group. We shall start, however, from a basic homogeneous
dynamical law T°, denoted as before by a t (absence of external fields) and
consider changes of this law of the form

d rt A = Tt (6A + i[h(t), A]) ; h(t) E 2t, A E D(6),

— (V.3.44)
dt
where b is the generator of a t (see lemma 3.3.1).

Remark. In the case of a finite system we could write

6A = i[Ho , A], (V.3.45)
3 Thatthis distinction is important was emphasized by H. Ekstein in connection with his
concept of "presymmetry" [Ek 69j.
238 V. Thermal States and Modular Automorphisms

where Ho is affiliated with 2t" and interpreted as the energy operator of the
unperturbed system. Further, writing

TtA = U(t)AU*(t), (V.3.46)

^t U(t) = i(TtH(t))U(t) = (OH (t); dt TtA = irt [H(t), Al, (V.3.47)

H(t) = Ho + h(t). (V.3.48)
The operator T t H(t), representing the procedure (H(t), t), is the energy oper-
ator for the system at time t including the then prevailing external influence
(h(t), t). If
h(t) = 0 for t < 0 and t > T (V.3.49)
then the energy transferred to the system in the time interval during which the
external influence was acting is
T
aE = TTHo — H o = Jo dtTtHodt. (V.3.50)

This can be rewritten in another form. By (V.3.46), (V.3.47)

—0 (V.3.51)
^tTtH(t,) t' = t = dtTtHo + ^t Tt h(t,) t' =t

which is the adaptation of energy conservation to the case of time dependent in-
fluences. The energy transfer (V.3.50) can then be written, taking the condition
(V.3.49) into account

aE=—f oT
( i Tt )
h(t)dt= f
T d
dt. (V.3.52)

On the other hand, defining

rim= U(t)e- 1Hot (V.3.53)

(V.3.50) gives

dE = r(T)Hor(T)* — Ho = _ir(T)61 (T)*. (V.3.54)

In the infinite system Ho and U(t) cannot be defined but r(t) remains a well
defined unitary in fit, given by the differential equation

d r (t) = ir(t)ath(t), (V.3.55)

with initial condition iv). 11. The above argument remains valid if T is finite
and h(t) is affiliated with the algebra of a bounded region when we replace Ho
by the integral of the energy density over a sufficiently large region. So we can
write for the energy transfer
 V.3 Direct Characterization of Equilibrium States 239

dE = —iP(T)bP(T)* = f
T Ttd^tdtt) (V.3.56)
cit.
This justifies the term "passive" in the definition 3.3.3. We have

Proposition 3.3.4
A state w is passive if, for any smooth family of self- adjoint elements h(t) E 21
satisfying (V.3.49)

dE =wJoT Tt d dtt ) dt > 0. (V.3.57)

If we interpret the change of the dynamical law from a t to Tt during the

interval 0 < t _< T as a cyclic process caused by external forces we see that from
a passive state we cannot extract energy by such a cyclic process.
One may then identify the equilibrium states of the dynamical system (2t, a t )
as the passive states. By theorem 3.3.2. the passive states are the KMS-states
with respect to at and non-negative parameter a (including the limiting case
13 = oo, the ground state).

V.3.4 Chemical Potential

The characterization of equilibrium states by either one of the two properties:

dynamical stability or passivity led to the conclusion that an equilibrium state
is a KMS-state of the observable algebra 21 with respect to time translations
at . On the other hand, starting from the grand canonical ensemble for a finite
system, the thermodynamic limit leads to a KMS-state over the "algebra of
operations", the "field algebra" A with respect to a 1-parameter subgroup of
time translations and gauge transformations. If we consider for simplicity here a
1-component system then this subgroup is characterized by the two parameters
0, p (see (V.1.14)). Of course, since the observables are invariant under ry w , the
state restricted to 21 is a KMS-state with respect to a t and parameter 0. But
it is clear that the states wo,j,, over A with different values of p yield different
states over 2t by this restriction. The problem of extending (a t , 0)-KMS-states
over 21 to states over A has been considered by Kastler [Kart 76], Araki and
Kishimoto [Ara 77b], and most completely by Araki, Haag, Kastler, Takesaki
[Ara 77c] which is the main source of the subsequent discussion.
Focusing attention on the observable algebra and ignoring for the moment
the embedding algebra A, the appearance of a distinguishing parameter p within
the set of (a t , ,Q)-KMS-states over t has the same root as the appearance of
charge quantum numbers in the set of states satisfying one of the selection
criteria of Chapter IV. It is the existence of localized morphisms of 21 which are
not inner. In our simplest example (Abelian gauge group) we are dealing with
an equivalence class of automorphisms modulo the inner automorphisms.
Let p be an automorphism in the class which adds one unit of charge. If w
is a state over 2. then
240 V. Thermal States and Modular Automorphisms

we (A) = w(pA) (V.3.58)

defines a state differing from w by one additional charge localized in some finite
volume at time zero.
Ot = atOat 1 (V.3.59)
describes the automorphism shifted by a time translation i.e. the adding of one
unit of charge the same way at a later time t. O and Ot are in the same class; so
they are related by an inner automorphism 4 and we can write

Pt = , Ad vt , or explicitly OtA = O(v t Avt ), (V.3.60)

where vt is a unitary element of 2t. Ad vt is determined by O and a t . Correspond-

ingly, vt is determined up to a phase factor which can be fixed by a convention.
First, by (V.3.59), we see that we can choose the family v t so that it satisfies
the cocycle identity
vt+t' = vt atvt', (V.3.61)
or, equivalently, the differential equation

d _ vt at h, h —_ d
i (V.3.62)
dt vt — —2 dt vt t=o
Here h is a self-adjoint, uniquely determined by O and at up to cil where c is an
arbitrary real constant. This fixes v t up to a factor eict. The important point is,
however, that once we have adopted a convention for choosing c this cannot be
changed any more in the subsequent consideration of equilibrium states with
different chemical potential.
Next, let w be a primary (a t , 1L3)-KMS-state over 2[ and w e as in (V.3.58).
In contradistinction to Chapter IV, where we discussed states with vanishing
matter density at infinity and the addition of a charge led to a different su-
perselection sector, now w describes a state which has already infinitely many
particles and the addition of one particle s is not a drastic change; it does not lead
out of the folium. Therefore w e is again a vector state in the GNS-representation
of t. We may compare the modular automorphism groups at and of of the
states w and w e. One sees easily from the defining relations (V.2.1) through
(V.2.12) that
- (V.3.63)
of = O 1 Qt p.
The Radon-Nikodym cocycle

Wt - Dw e : Dw (V.3.64)

is a family of unitaries in i,,(2l)" and the intertwining property (V.2.56) gives

4 Concerning the notation: in Chapter IV the inner automorphism induced by a unitary U
was denoted by au. Here we shall write Ad U instead of au in order to avoid confusion with
modular automorphisms.
5 In the non-relativistic context the "charge" we talk about is synonymous with particle
number.
 V.3 Direct Characterization of Equilibrium States 241

of _ (Ad Wt )at , (V.3.65)

or, by (V.3.63)
at out 1 = o Ad Wt . (V.3.66)
But the modular group of w is

Qt = a-p t . (V.3.67)

So, by (V.3.59)
o--pt = o Ad Wt . (V.3.68)
This implies that Ad Wt is actually an inner automorphism of 2[ and the com-
parison with (V.3.60) gives

Ad Wt = Ad v_pt . (V.3.69)

Since the center of ir w (2i)" consists only of multiples of the identity W t and
7rw (v_pt ) agree up to a phase factor and, since Wt satisfies the cocycle relation
(V.2.52), this factor is a function eiat. So we can write

Wt = ez ) 4t7rw (v-pt ), (V.3.70)

where the parameter µ depends on the state w.

This parameter is the chemical potential (up to a constant c which depends
on the convention used in fixing vt ). This is seen by looking at the extension of
w to a state D over the field algebra A. Let us suppose that this extension is an
(at , ,Q)-KMS-state (see (V.1.14)) so that the modular group of D is

Qt = a-pt'Yppt• (V.3.71)

The extension P of the charge creation automorphism to A is implemented by a

unitary u E A. For distinction we shall denote the elements of A by lower case
letters a, b,... , those of 2< by capital letters A, B, .... Then

"da = uau*, a E A, u E A, (V.3.72)

with
ry,p u = ei'Pu. (V.3.73)
As above we have
-dt = agar 1 = ô Ad vt , (V.3.74)
and find from (V.3.72) that vt can now be chosen as

vt = u*a t u. (V.3.75)

The modular automorphism groups of D e and D are related as in (V.3.63) and

the Radon-Nikodym cocycle is now

Wt = Me : DD = u * a't u, (V.3.76)
242 V. Thermal States and Modular Automorphisms

which becomes, inserting (V.3.71) and (V.3.73), according to (V.3.75)

Wt = e 2Qµt y Qt .
- (V.3.77)

We have omitted writing the symbol iw since all the objects are understood
to be taken in this GNS-representation. The restriction of (V.3.77) to 2i just
means that we drop the marking - and then (V.3.77) becomes (V.3.70).
One should still understand more directly why and how the extension of
an (a t , ,Q)-KMS-state w over % leads to a state CD over A which is a KMS-
state with respect to a modified automorphism group. Let us start from the
following setting. A is the C*-algebra of quasilocal operations which may be
constructed from the observable algebra % and its localized morphisms as de-
scribed in Chapter IV. The full symmetry group of A consists of the geometric
symmetries and a (global) gauge group Ç. Of the former we shall use here only
the time translations a t and the powers of some automorphism T with respect
to which A is asymptotically Abelian 6

lim [a, Tn b] = 0.
n-4.00
(V.3.78)

T may be interpreted as a space translation or, under the assumption 3.2.1, also
as a time translation. We shall further assume that Ç is a compact Lie group
(possibly non Abelian). It is represented by automorphisms of A denoted by
`y9, g E Ç which commute with a t and T. We shall consider states over % which
are extremal T-invariant i.e.

w(TA) = w(A),

and such that w is not a m ix ture (convex combination) of other T-invariant

states. The asymptotic Abelianness of ? with respect to T implies that extremal
T-invariance is equivalent to "weak clustering" (see Doplicher, Kadison, Kastler,
Robinson [Dopl 67]):

N^^ (2N) -1 E w(aTnb) = w(a)w(b).

N
(V.3.79)
—N

This is the main structural property used in the following.

Theorem 3.4.1
An extremal T-invariant state over ? has extensions to extremal T-invariant
states over A. Two such extensions (p i , tp2 can differ only by a gauge transfor-
mation, i.e. there is an element g E Ç such that

SP2 (a) = W1(y9a) • (V.3.80)

6 Since A contains fermionic and bosonic elements the commutator in (V.3.78) should be
replaced by the "graded commutator" [a, b]±. This can be taken into account by a straight-
forward adaptation of the subsequent argument. We shall omit it here; it is given in [Ara
77c] .
 V.3 Direct Characterization of Equilibrium States 243

The first part of the theorem, the existence of extremal r-invariant exten-
sions follows from general abstract arguments (Hahn-Banach theorem, averag-
ing over the group r and Krein-Milman theorem). The uniqueness of such an
extension up to a gauge transformation follows from the consideration of the
set of continuous functions over C obtained from a state cP by

fa(g) = cp(-y ga); a E A. (V.3.81)

Obviously this set (for a ranging through A and fixed cp) is a linear space,
containing with every function also its complex conjugate. We define the norm
of f as the supremum of I f (g) I and denote the completion of this set in the
norm topology by Cy,. If yo is weakly clustering then (V.3.79) implies that with
f, f E C, also the product f f E C ,. So C, is an Abelian C*-algebra. It may be
^

that Sp is invariant under some subgroup of C g which will be called the stability
group or the stabilizer of yo and denoted by Ç,. By the definition (V.3.81) then

fa(g) = fa (hg) for all h EÇ .

, (V.3.82)

So the functions in C, are constant on each right coset of C w and C, may be

regarded as the C*-algebra of all continuous functions of the right cosets. Now,
let (pi and Sp2 be two extremal r-invariant extensions of the state w on t. For
any a, b E A
A n (a, b) = J(yga*rnb)dp(g) (V.3.83)

is gauge invariant and hence belongs to 2.. In (V.3.83) denotes the Haar
measure on g. So (p i (A n (a b)) = Sp2 (A n (a, b)) . Taking the limit n — oo and
,

using (V.3.79) one gets

J 7(1)(g)f1)(g)dj(g)
=
J 7(2)(g)f(2)(g)d/L(g)
, (V.3.84)

where Pi) are defined as in (V.3.81) with Spy replacing cp. Furthermore the map
V from Co onto Co given by

V f âi^ = f a2) (V.3.85)

respects the product structure. It gives an isomorphism of the C*-algebras.

Defining right and left translations of functions on Ç respectively by

(psf)(9) = f(g 1 9)1 (a9f) (9) = f(g9 1 ) (V.3.86)

one sees that V commutes with all right translations. One can show that an
isomorphism of two C*-subalgebras of the continuous functions on Ç which
commutes with all right translations is given by a left translation (see appendix
A in [Ara 77c]). Thus V = a(g) for some element g E Ç. This is precisely
the claim of (V.3.80). It also follows that the stability subgroups of (p i and Sot
are related through conjugation with this element g. In physical language the
stability group is the unbroken part of the gauge group for the extended state.
We see that it is determined by the state w on 2[ uniquely up to a conjugation.
244 V. Thermal States and Modular Automorphisms

Theorem 3.4.2
Let w be an a t -invariant, extremal 7-invariant state on 2t and cp an extremal
7-invariant extension to A. Then there exists a continuous one-parameter sub-
group of g
tE1R—E(t)EÇ
such that cp is invariant under the modified time translation
a t = at`YE(t) • (V.3.87)

This is a simple consequence of theorem 3.4.1. Since w = w o a t the states

Sp and cp o at are two (possibly different) extensions of w. Therefore they are
related as in (V.3.80) where now, of course, g will depend on t.

The most interesting question concerns the extension of an extremal (a t , ,Q)-

KMS-state w over 21.. We shall not reproduce here the rather lengthy arguments
establishing the properties of the extensions in full generality. They can be found
in [Ara 77c]. Instead we shall comment on the main conclusions. The state w
is at-invariant and, since it is also primary it is an extremal a t-invariant state.
Under assumption 3.2.1. we can use the time translation group instead of the
group Tn . Theorem 3.4.1. holds also for this case and tells us that there is an
extremal a t-invariant extension cp to A. It is unique up to gauge and has a
stability group Ç,,, c Ç (the unbroken part of the gauge symmetry). It may
happen that C4, has a subgroup NÇ called the asymmetry group with respect
to which yo is a ground state or ceiling state. This means the following. Since
cp is Ç,-invariant we have in the GNS-representation (ir,p (A), ?-(0, 1°)) an im-
plementation of 1(9 for g E Ç, by unitaries leaving the state vector 10) (the
vector representative of Sp) invariant. Their generators are self-adjoint operators
on fl, representing elements of the Lie algebra of ÇÇ ,. The spectra of these
generators are additive semigroups. This follows from the T- (or at )-asymptotic
Abelianness by essentially the same argument which establishes the additivity
of the energy-momentum spectrum (Chapter II, theorem 5.4.1). So, if N is such
a generator, its spectrum can either be one-sided (confined to non-negative or
to non-positive reals) or it may extend from —oo to +oo. The asymmetry group
N: is that part of Ç,p whose Lie algebra is represented by operators with a
one-sided spectrum. In statistical mechanics this part does not play a rôle. It
just means that the components (charges) associated with these directions in Ç
are absent, their chemical potential is —oo. One may eliminate the asymmetry
group by considering instead of A the subalgebra Al of invariant elements under
N: (fixed point algebra for 4) and instead of gip the quotient group C,/N„.
We shall therefore formulate the conclusion only for the case where N ip is absent.

Theorem 3.4.3
Let w be an extremal (a t , 0)-KMS-state over 2t, v) an extremal a t-invariant ex-
tension to A (unique up to gauge), Ç , the stabilizer (unbroken part of Ç in the
state (p). If the asymmetry group NÇ is trivial then cp is an (at, ,Q)-KMS-state
where
 V.4 Modular Automorphisms of Local Algebras 245

at = at -YE(t) (V.3.88)
and e(t) is a one-parameter subgroup in the center of Ç ,.

Note that if the gauge symmetry is completely broken or if the Lie algebra
of g„ has trivial center then at = at Note also that Ç , and the element in the
center of the Lie algebra of Ç, defining E(t) are already essentially determined
by w. For two different extensions g,,,2 = gÇw1 g -1 where g is some fixed element
of C. The state Sp need not necessarily be an extremal KMS-state. If not then its
extremal KMS-components are also extensions of w which are then, however,
not at-invariant.

V.4 Modular Automorphisms of Local Algebras

Let us return to the consideration of the vacuum sector of a relativistic local

theory. We use the concepts and notation of section III.4. The theorem of Reeh
and Schlieder (subsection II.5.3) says that the state vector .f2 of the vacuum is
cyclic and separating for any algebra R.(0) as long as the causal complement of
0 contains a non-void open set. Thus we can apply the Tomita-Takesaki theory.
The vacuum defines a modular operator a(0), a modular conjugation J(0) and
a modular automorphism group o (0) for any such R(0). What is the meaning
of these objects? We know that a.,(0) maps R.(0) onto itself, J(0) maps R(0)
onto RZ(0)'. But beyond this we have, in general, no geometric interpretation;
the image under a,-(0) or J(0) of RZ(O 1 ) with 0 1 C 0 is not, in general, some
local subalgebra. There are, however, interesting special cases where a.- and J
do have a direct geometric meaning.

V.4.1 The Bisognano-Wichmann Theorem

Let W denote the wedge in Minkowski space

W = {x E .M : x 1 > (x° I; x2 , x3 arbitrary}. (V.4.1)

The 1-parameter subgroup A(s) of special Lorentz transformations ( "boosts" in

the x 1 -direction)
cosh s sinh s 0 0, \
sinh s cosh s 0 0,
A(s) = V.4.2
0 0 1 0 ( )
0 0 0 1
transforms W into itself. The transformation law of points may also be regarded
as a relation in the Poincaré group. If U(A(s)), U(x) are the unitary operators
implementing respectively the boosts (V.4.2) and the space-time translations
then
U(A(s))U(x)U(A(s)) -1 = U(x(s)), (V.4.3)
246 V. Thermal States and Modular Automorphisms

where
x° (s) = x° cosh s + x 1 sinh s,
x l (s) = x° sinh s + xl cosh s, (V.4.4)
xr(s) = xr for r = 2, 3.
This relation may be extended to complex values of the parameter s. Setting

s = a + iµ (v.4.5)

with A, p real we have

xo(s) = x° (A) cos p + i sin p (x° sinh A + x1 cosh A),
x 1 (s) = x 1 (A) cos p + i sin p (1° cosh A + x 1 sinh A), (V.4.6)
xr(s) = xr for r = 2, 3.
Introducing the generators, K for the boosts, Po for the translations

U(A(s)) = exp iKs; U(x) = exp iPoxµ (V.4.7)

we get
exp iK(A + iµ ) U(x) exp —iK(a + iµ )
= U(cos p x(a)) exp {— sin µ(PN ° — Pl n l )}, (V.4.8)
with
n ° = x° sinh A + x 1 cosh A,
ril = x° cosh A + x 1 sinh A, (V.4.9)
rir = 0 for r = 2, 3.
One notes that for x E W rj (A) is a positive time-like vector for all A. Therefore,
due to the spectrum condition for energy-momentum in the vacuum sector, the
second factor on the right hand side of (V.4.8) provides an exponential damping
as long as
0<u< 7r; x E W. (V.4.10)
The first factor is unitary. We conclude that under the condition (V.4.10) the left
hand side of (V.4.8) is a bounded operator and moreover an analytic function
of the complex variable s in the strip 0 < Im s < 'Tr.
For p = 7r, A = 0 we have

11(i7r)x — x',

x'u _ — xµ for p = 0, 1
(V.4.11)
x'` for p = 2, 3.
We can invert the signs of xr, by a spatial rotation through an angle of it around
the 1-axis. Denoting this by R i (7r) we get

Ri(ir)A(ir)x = —x. (V.4.12)

Using the CPT-operator e of Chapter II, theorem 5.1.4 we define the conjugate
linear operator S by
 V.4 Modular Automorphisms of Local Algebras 247

S = &U(Ri(7r))U(A(i7r)), (V.4.13)
and obtain from (1I.5.9)
SU(x)S -1 = U(x). (V.4.14)
Let us evaluate the transformation law of a covariant field 0 under S. If we
use spinorial notation and consider a field with n undotted, m dotted indices
then a boost A(A) acts on each undotted index by the matrix exp (1/2 )(a 1 ), a
rotation R1 ((IQ) by exp (1/2 icpo-i ), where al denotes the first Pauli matrix. So the
product Ri (7r)A(iir) brings a factor exp(ira i ) = —1 for each undotted index.
For dotted indices we have the complex conjugate transformation matrices (for
real A). The product Ri(7r)A(iir) gives +1. Altogether

U ( Ri (7r)) U ( A (i7r))0,.(0)U(A(i7r)) -1 U(Ri (7r)) -1 = (-1)n0,.(0), (V.4.15)

where the polyindex r consists of n undotted, m dotted spinor indices. Together

with (I1.5.7) this yields for observable fields (where n + m is even, F = 0)

5ÇPr (0)S -1 = ;(0). (V.4.16)

Now we consider the (vector valued) distribution

0(x 1 )0(x 2 )- • • 0(xn )I(2) = U( xi )0(0)U( x2 — x1 )0(0) ... U(x. — xn-1)0(0)1.0)•

Applying (V.4.14), (V.4.16) we obtain formally

S0(xi)Cx2) ... ÇP(x n )lQ) = 0*(xi) ...0*(xn)IQ).

We must remember, however, that on the way we used (V.4.8) and this gives
an analytic continuation to complex values of s only if each of the arguments
x i , (x2 — x 1 ), ... (x, — xn_ i ) lies in W. For such configurations of points all
xi lie space-like to each other. So one can invert the order of factors on the
right hand side and one must do so to approach the boundary of the analyticity
domain in the right manner (see the corresponding discussion in the proof of
the CPT-theorem in Chapter II). We obtain

S0(x 1 ) ... 0 (x n) I.f2) = (1(xi) ... Cxn)) * III). (V.4.17)

It remains to show that the special configurations of points used above suffice
to conclude from the analyticit -operties that (V.4.17) holds for all configu-
rations of points in W. Then one has

SAlQ) = A*I ,0) (V.4.18)

for all A in the polynomial algebra P(W), generated by fields smeared with test
functions with support in W. One must show further that the domain P(W)If?)
is a "core" for S i.e. that (V.4.18) for A E P(W) uniquely determines S. For the
technically quite tedious proof of these claims we refer to the original papers by
Bisognano and Wichmann [Bis 75, 76]. Accepting this one knows that (V.4.18)
248 V. Thermal States and Modular Automorphisms

holds also for TZ(W ), the von Neumann algebra with which the fields in W are
affiliated and one has finally

Theorem 4.1.1 (Bisognano and Wichmann)

If 7Z(W) has a system of affiliated observable fields satisfying the axioms of
Chapter II then the modular conjugation for the vacuum state is

J(W) = ®U(Rl (rr)), (V.4.19)

the modular operator is

d(W) _ e-27rx 1 (V.4.20)
the modular automorphisms ut act geometrically as the boosts (V.4.4) with
s = ---2irt.

Remarks. 1) Since the relations are purely geometrical one may wonder whether
the theorem cannot be established by considering only the net 7Z(0) without
reference to a generating set of fields. This would demand to show that the oper-
ator J(W), as determined from the Tomita-Takesaki theorem, has the geometric
significance
J7Z(0)J = TZ(r0), (V.4.21)
where r is the reflection (V.4.11) This would also provide a proof of the CPT-
theorem in the algebraic setting, defining ® by (V.4.19). 1 Besides (V.4.21) one
would have to prove the KMS-condition for the functions
FA B(t) = wo(Be27rixtAe-27rixt),

2 ^riKt Ae -2zrBKt
GA,B(t) = wo(e B), A, B E TZ(W).
Looking at the argument above we see that the essential idea is to generate the
algebra by translates a 2 A with x E W and A E 7Z(0o ) where 00 is a small
neighborhood of the origin. The problem is then to control the growth of

Ae K1'; A E R,(00),0 < p,< 7r. (V.4.22)

If 00 shrinks to a point and A is replaced by a field 0(0) transforming according

to some finite dimensional representation of the Lorentz group then this is
answered by (V.4.15). But we lack a study of the behaviour of (V.4.22) for
small finite 00 .
2) The trajectory of the point x = (0, o, 0, 0) under the boosts (V.4.2)
can be interpreted as a uniformly accelerated motion with acceleration

b = 0-1

For an observer moving on this trajectory and using his proper time r = gs as
time coordinate the operator H = p1 K generates time translations in his body
Significant progress towards this goal has recently been made in Porch 92, 951.
 V.4 Modular Automorphisms of Local Algebras 249

fixed coordinate system. According to theorem 4.1.1 the vacuum state looks to
him like a thermal state with temperature
b
T = (27r0 -1 = (V.4.23)
27r
—.

This very interesting result is closely related to the Hawking temperature of

a black hole [Hawk 75]. In fact, the wedge W in Minkowski space, populated
by observers which are constrained to stay always in W, provides the simplest
example of a horizon (see Rindler [Rind 66]). If such an observer sends a signal
to a region outside the wedge neither he nor his fellow travellers can receive an
answer back. In this analogy between the wedge W, considered as the "Rindler
universe", and a (permanent) black hole the Bisognano-Wichmann temperature
(V.4.23) corresponds to the Hawking temperature. Motivated by Hawking's pa-
per Unruh discussed the response of a uniformly accelerated detector to the
vacuum state of a quantum field in Minkowski space [Unr 76]; see also [Day
75], [de Witt 80]. Unruh idealized the detector as a collection of "molecules"
(quantum mechanical systems with a discrete energy spectrum and negligeable
spatial extension) coupled with the quantum field by an interaction Hamilto-
nian proportional to 0(x) where x denotes the moving point on the trajectory.
His result can be expressed in the following way: if the quantum field is in the
vacuum state then the population of the energy levels of the molecules adjusts
to the same distribution which the system at rest would have in a surround-
ing thermal bath at temperature (V.4.23). We shall return to the problem of
quantum field theory in the presence of a gravitational background field (curved
space-time) and the equivalence principle in quantum physics in Chapter VIII.

V.4.2 Conformal Invariance and the Theorem of Hislop and Longo

In general the wedge regions are the only ones for which the modular automor-
phisms (induced by the vacuum state) correspond to point transformations in
Minkowski space. However, if the theory is conformally invariant there are wider
classes of regions for which the modular automorphisms act geometrically. They
include, as the most important case, the diamonds [Hisl 82].
The basic observation is most easily seen in the formalism described in
subsection I.2.1 of where the conformal group is realized by pseudoorthogonal
transformations in a 6-dimensional space with coordinates ea. Consider rota-
tions in the e 1 — 64 -plane

T41 (c9)e = e(cP),

e4 ((p) _ e4 cos cP + e 1 sin cp,
(V.4.24)
e1(cP) = —e4 sin cP + e 1 cos cp,
ea((p) = Ca for a 1, 4.
One verifies that if the original point x(0)

xp,(0) = e.t(e 5 -1
) ,

ji
= 0, 1, 2, 3 (V.4.25)
250 V. Thermal States and Modular Automorphisms

lies inside the wedge W then x' (cp) follows a continuous path, never reaching
infinity for 0 < cp < 7r. For cp = 7r/2 the wedge is mapped onto the diamond K 1
whose vertices are the points x ° = +1, x = O.

T41 (2) (V.4.26)

W— Ki
Kl = {x : lx° 1 + < 1}. (V.4.27)
To see this it is instructive, though somewhat tedious, to follow the orbits
xlt ((p) of the boundary points of W in detail. A quicker, though somewhat blind,
way is to notice that

(64 + 5 1
(0((p) + e5(cp)) ) = cos cp + x l sin cp + (1 — (x, x)) (1 — cos cp)

cannot vanish for 0 < cp < 7r since, for x E W, z 1 > 0 and (x, x) < 0. Thus the
orbits cannot go to infinity. Then one checks that the inequalities characterizing
the wedge
x ; > Ix° 1
go over for cp = 7r/2 into the inequalities characterizing the diamond K 1

^
x° (-7 \ + x
2 Gr) E <1 .
The modular transformations of W are the boosts (V.4.4) which are "pseudo-
rotations" in the 61— 6° plane.

T10 (s)C = V(s),

6 ° (s) = 6° cosh s + 61 sinh s, (V.4.28)
e1 (s) = 6° sinh s + 61 cosh s,
Ms) = 6' for a # 0, 1.

This suggests that the modular group of K l corresponds to the conformal trans-
formations
( -1
T(s) = T41 (i) (V.4.29)
T lo(s)T41 To4(s).
The orbit of a point under these transformations is

T04 (s)C = C(s),

6 ° (s) = 6° cosh s + 4 sinh s, (V.4.30)
4 (s) = 6° sinh s + 64 cosh s,6
Ca(s) = Ca for a 0 0, 4,
or, in x-space

x°(8 ) = N(s) -1 (x ° cosh s + 2 {1 + (x, x)} sinh s) ,

x'(s) = N(s) -l x' i = 1, 2, 3,
, (V.4.31)
N(s) = (x° sinh s + â {1 + (x, x)} cosh s + â {1 — (x, x) }) .
 V.4 Modular Automorphisms of Local Algebras 251

This can be simplified by introducing

x + =x° +lxI, x_ = x ° —lxI. (V.4.32)

The denominator N(s) factors into

l es ((1 + x+ ) + e -s (1 — x +)J ((1 + x_) + e -s(1 — x_))

and (V.4.31) becomes

x±(s) = ((1 + x±) — e -s (1 — x±)) ((1 + x±) + e-3 (1 — x±)} 1 (V.4.33)

The modular conjugation for W is the inversion

x i — > - xl ; x° — > -x° ; x r — ^ xr ; for r = 2, 3.

Conjugation with T 14 (7r/2) leads to the inversion map for K 1

xi —^^ —x'(x, x) -1 , i = 1, 2, 3; x° x° (x, x) -1 (V.4.34)

Theorem 4.2.1 (Hislop and Longo)

In a free, massless field theory the modular automorphism group ar (Ki) of
the diamond region K1 induced by the vacuum state has the geometric action
(V.4.33) (with the parameter s replaced by —27r1- ). The modular conjugation
J(Ki) has the geometric action (V.4.34).

Proof. The above discussion shows that theorem 4.2.1. follows from the
Bisognano-Wichman theorem if the theory is conformally invariant in the fol-
lowing sense
(i) algebraically: given any finitely extended region 0 there is a neighbor-
hood of the identity N in the conformal group such that for g E N one has an
algebraic isomorphism a9 from R.(0) to TZ(gO) respecting the net structure i.e.

a9TZ(0 i ) = R.(gO i ) for any 01 C 0, g E N. (V.4.35)

In field theory this means that we must have a transformation law for fields
under the conformal group conserving the algebraic relations (commutation
relations and equation of motion) and that, for group elements sufficiently close
to the identity, the argument of the field is transformed thereby according to
the geometric action of the conformal group on Minkowski space as

x —^^ gx. (V.4.36)

It is clear that (V.4.36) cannot be required globally for all pairs g, x since
conformal transformations do not give globally defined maps from M to M.
In the present context one can use the conformal invariance only for orbits of
points which do not pass through infinity under the conformal transformations
applied.
252 V. Thermal States and Modular Automorphisms

(ii) The vacuum state must be invariant under the conformal group

wo(a9A) = w° (A) (V.4.37)

whenever agA is defined. In other words, the conformal invariance should not
be spontaneously broken in the vacuum sector. Then ag can be implemented
by unitaries Ug in the usual fashion

UgAJQ) = a gAl fl). (V.4.38)

For a massless free theory both items (i) and (ii) can be established. See e.g.
Swieca and Volkel [Swiec 73), Schroer and Swieca [Schroer 74) where the uni-
taries Ug and the limitations in their geometric interpretation are studied in
detail. In the case of a scalar field 45 the transformation law is

(ag 45)(x) = N(9, x) -1 'P(gx), ( V .4.3 9)

where N is the factor given in (I.2.46). El

Another interesting result in this context is

Theorem 4.2.2 [Such 78]

In a fr ee, massless theory the modular automorphism group aD (V+) induced by
the vacuum state on the algebra of the forward light cone corresponds geomet-
rically to the dilations
(T) = e-2"xµr
xth (V.4.40)
the modular conjugation to
X µ —> — x t `. (V.4.41)
Proof. We start from the observation that rotations in the E° - e 5 -plane

T05 ((P)e = ((P),

°((P) = e° cos (09+ e5 sin cp,
(V.4.42)
e5 ((P) = -e° sin (P + e5 cos (P,
ea(SO) _ ea for a # 0, 5
transform the diamond K 1 to the forward light cone V+ for the angle (P = 7/2
and that the orbit of points from K 1 is continuous in the interval 0 < cp < ic.
Thus, if the theory is conformally invariant in the sense described in the proof
of theorem 4.2.1, then the modular structure for V+ results from that for K 1
05 (7r/2) yielding instead of the orbits (V.4.30) the orbits byconjugatiwhT

e (s) = T 54 (s)e,
et`(s) = et' for µ = 0, 1, 2, 3,
e 4 (s) + e(s) = e-s (e4 + e),
(V.4.43)

e4 (s) e(s)
- - e9 (e4
 V.4 Modular Automorphisms of Local Algebras 253

Tt
1
I

Fig. V.4.1.

This establishes (V.4.40) (putting s = —27rr as in theorem 4.2.1.). In the same

way one obtains (V.4.41) from (V.4.34) . ❑

Remark. The inversion (V.4.34) maps K 1 onto K;, the shaded region in
Fig V.4.1.. The inversion (V.4.41) maps V+ onto V. If in a theory the modular
conjugations for 7Z(K1) and TL(V+) are as demanded by theorems 4.2.1. and
4.2.2. then
R(K l )' = Te(Kl ), (V.4.44)
77,(V+) 1 = TL(V - ). (V.4.45)
Since Ki and V - are not the causal complements of K1 and V+, respectively,
we do not have the relation TL(O") = R(0) (Chapter 111.5) in such theories.
This is due to the fact that observables commute here not only for space-like
separation but also for time-like separation; the nonvanishing of commutators is
restricted to light-like directions. There are indications that the theorems have
a partial converse, namely that the vanishing of commutators in both space-like
and time-like directions characterizes free massless theories. (See for instance
Buchholz and Fredenhagen [Buch 77a, c]) .
254 V. Thermal States and Modular Automorphisms

V.5 Phase Space, Nuclearity, Split Property,

Local Equilibrium

V.5.1 Introduction

The principles formulated in Chapter III do not yet entail such basic features
of experience as the existence of particles and a reasonable thermodynamic
behaviour. The same applies, of course, to the axioms in Chapter II.1. We
may recall that in Chapter II the occurrence of particles was put in "by hand"
through the requirement that the mass operator PoPA should have some discrete
eigenvalues. If, furthermore, the masses are nonzero then this, in conjunction
with the listed principles suffices to develop a (possibly incomplete) particle
interpretation and determine the collision cross sections. Similarly, while we
have defined what we mean by a thermodynamic equilibrium state, the existence
of such states in the theory is not guaranteed by the principles stated so far.
Both questions have a common root. We must consider the analogue of
classical phase space volumes i.e. the part of state space corresponding to a
simultaneous limitation of energy and space volume. Loosely speaking, finite
volumes in classical phase space should correspond to finite dimensional parts
of state space in quantum physics. Starting from the vacuum representation this
idea may be implemented in the following way.
Let Kr be the diamond whose base is a ball with radius r at time t = O. We
wish to consider the set of states which can be called "essentially localized" in
Kr. While the localization of observables is a fundamental concept in our frame
the localization of states is a less clear cut notion. It needs the vacuum as a
reference state. [Knight 61] and [Licht 63] defined strictly localized states in a
region O as those states which give the same expectation values as the vacuum
for all measurements in the causal complement of O. The corresponding state
vectors are then given by
W = W IQ) with W*W = ii, W E A(0), (V.5.1)
i.e. states which are generated by an isometric operator from the field algebra
of the region applied to the vacuum. Indeed, for such states
(CAP) = (QIAIQ) for A E TZ(O'). (V.5.2)
A (superficial) paradox arises from the Reeh-Schlieder-theorem. Since A(0)IQ)
is dense in f the linear space spanned by the vectors of the form (V.5.1) is
all of H. So we can approximate any state vector by linear combinations of
vectors describing states "strictly localized" in some region O. This is due to
the correlations between distant observables present in the vacuum state which,
though decreasing fast with increasing distance, never vanish. By exploiting
them judiciously one may, by an operation in Kr on the vacuum, create a local-
ization center far away from Kr . But at what cost? It suffices for the following
to restrict attention to the vacuum sector (uncharged states) and consider state
vectors of the form
 V.5 Phase Space, Nuclearity, Split Property, Local Equilibrium 255

W = AI S) with A E R(Kr). (V.5.3)

The ratio of cost vs. effect may be measured by the quantity
A
CA = (V.5.4)
II A I ^ IIII'
the norm of the operator corresponding to the expenditure, the length of the
resulting vector AI fl) to the achieved effect. It was shown by Haag and Swieca
[Haag 65] that the states of the form (V.5.3) can indeed be interpreted as states
which are "essentially localized" in Kr provide that c A < c where the choice of c
determines the degree of tolerated delocalization which grows with c. The case
(V.5.1) ("strict localization") corresponds to c = 1, the minimal value possible.
With these considerations in mind we are now able to "measure" phase space
volumes. The latter correspond to subsets of K

ME,,. = PER(1)(K")I Q), (V.5.5)

where PE is the spectral projector of the Hamiltonian to the interval [0, E] and
RP ) denotes the unit ball in R

Te 1) (0) = {A E R( 0) : II A II < 11. (V.5.6)

Again, the linear span of ME,r has no memory of the localization region but
the length of vectors in ME ,,. decreases fast with increasing delocalization. The
intuitive argument at the beginning of this section leads to the requirement
that ME,r should be "essentially finite dimensional". More precisely: there
should be an ascending sequence of finite dimensional subspaces lid C '
(d denoting the dimension) and for each c > 0 a dimension d(e) such that
the vectors in ME ,,. orthogonal to 1 -td(e) have length less than e. This provided
the motivation for the

Compactness requirement 5.1.1 [Haag-Swieca]

The set ME,,. is compact in the norm topology of H.

It was recognized by Buchholz and Wichmann [Buch 86a] that the same
intuitive picture leads to a much stronger requirement and that the estimates
of the dependence of d(e) on E and r given in [Haag 65] can be considerably
improved Instead of the sharp cut-off in energy Buchholz and Wichmann use
an exponential damping and replace the requirement 5.1.1 by the

Nuclearity requirement 5.1.2 [Buchholz-Wichmann]

The set
Np,r = e —aHR(I) (Kr ) I Q) (V.5.7)
is a nuclear set in H for any ,Q > O. The nuclearity index v1,,. is bounded by
vo,,. < exp cr 30-n (V.5.8)

(for large r, small ,Q) with c, n some positive constants.

256 V. Thermal States and Modular Automorphisms

In the present context the nuclearity of JV ,,. means that there is a positive
trace class operator T,3,,. such that

J1Ï,3,r C TQ,m(i) ; 7-61) the unit ball in R. (V.5.9)

The nuclearity index is defined as

= inf tr T,3 , ,., (V.5.10)

the infimum being taken over all trace class operators satisfying (V.5.9).
The status of this requirement may be assessed from the following com-
ments.

1) It has been tested in free field theories. There it holds if and only if the
mass spectrum of particles satisfies

E fi exp —/3mi < oo for all 0 > 0, (V.5.11)

where ft is the multiplicity of the mass value m i including a factor (28+1)

for spin orientations. It is worthwhile to note that it still holds if some
fields are massless. We shall not reproduce these computations here but
refer the critical reader to Buchholz and Junglas [Buch 86b], Buchholz
and Jacobi [Buch 87b] .

2) The requirement is necessary and sufficient to ensure "normal thermody-

namic properties", namely the existence of KMS-states for all positive Q
for the infinite system and for (properly definable) finitely extended parts.
For the latter the nuclearity index is directly the partition function [Buch
86a], [Jung 87], [Buch 89]. The r-dependence of u13,,. in the estimate (V.5.8)
expresses the proportionality of the free energy to the volume for large r.
The dependence of v on ,Q gives the relation between energy and tem-
perature in the high temperature region. It is determined by the density
of energy levels in the asymptotic region. The possibility that the level
density might increase stronger than exponentially, thereby giving rise to
a maximal temperature [Hagedorn 67] can be accommodated by a modi-
fication of the estimate (V.5.8), replacing the power ,Q -n in the exponent
by a function «(,Q) which becomes infinite at the Hagedorn temperature.

The definition of finitely extended "subsystems" is tied to a property of

the net of local algebras first conjectured by Borchers and studied in detail by
Doplicher and Longo [Dop1 84a], the "split property". The local algebras 17,(0)
are not of type I i.e. not isomorphic to the algebra 93(7 0) of some Hilbert space.
This is the ultimate reason why it is not possible to define a (linear) subspace
R0 C ?-1 which can be considered to correspond to the states "localized" in C^
and why we have to take recourse to constructions like (V.5.5) or (V.5.7) for
this purpose. However, given two concentric (standard) diamonds with radii rk
where r2 > r 1 , the split property asserts that there exists a type-I-factor J1Î such
that
 V.5 Phase Space, Nuclearity, Split Property, Local Equilibrium 257

1Z(K1) C N C 7Z(K2). (V.5.12)

It was shown in [Buch 86a] that this property follows from the nuclearity re-
quirement 5.1.2 (at least if r 2 — r1 is sufficiently large). Subsequently Buchholz,
D'Antoni and Fredenhagen [Buch 87c] showed that it follows for any pair r 2 , r1
2 > r 1 if the nuclearity index is bounded by the estimate (V.5.8). If the withr
nuclearity index diverges at some finite temperature then the minimal separa-
tion r2 — r1 needed for the split is just the inverse Hagedorn temperature. In the
next subsection we shall outline this argument and mention some consequences
of the split property.
In a further development of this analysis Buchholz, D'Antoni and Longo
showed that the nuclearity requirement can be formulated without using the
Hamiltonian, working instead with the modular operator of TZ(K r ). Thereby
the requirement becomes a purely local one [Buch 90a].
Another variant of phase space conditions stems from a remark due to
Fredenhagen and Hertel (unpublished preprint 1979). Instead of starting from
localized states and then damping the energy one may start from the set of linear
forms over the quasilocal algebra 2t which have energy below some bound E
(or are exponentially damped) and consider their restrictions to a local algebra
2t(K r ). Again one expects that the resulting set of linear forms over 2t(K r )
must be nuclear. This approach is closer to the basic structure of the theory.
One may note that an energy bound for states over 2t can be expressed as a
condition on their annihilator ideal (see the discussion around definition 3.2.3 of
Chapter III). For a comparison of various phase space conditions see Buchholz
and Porrmann [Buch 90c].
Technically it is somewhat more convenient to work with nuclear maps be-
tween Banach spaces rather than with nuclear sets in one space. The latter are
then just the images of the unit ball under the former.

Definition 5.1.3
Let E and .F be Banach spaces and 0 a bounded linear map from E into .F,

I e I1= sup{II e(E) II, E E e, II E

Generalizing the notion of a dimension function d(E), mentioned in the sequel

of (V.5.6), one defines the "E-content" N(E) of the map as the maximal number
of elements Ek E E (1) (the unit ball of E) such that the distance between any
pair of their images exceeds E

I eEa — eEkII> e; i k, II Ek I < 1.

e is called compact if N(E) is finite for every E > O.
If the growth of N(E) is bounded by
N(E) < exp (E -q) as E —> 0 (V.5.13)
then the smallest (positive) number q for which this holds is called the order of
the map e.
258 V. Thermal States and Modular Automorphisms

0 is called a nuclear map if it can be written as

OE cpi (E)Fi (V.5.14)

with
EII(Pi (IIFi II< oc) (V.5.15)
where (pi are bounded linear forms on E and Fi E .F. We define the nuclearity
index (or G( 1 )-norm) of e as

IIO11 1 = inf çoiIIIIFiII (V.5.16)

where the infimum is taken over all realizations of 0 in the form (V.5.14).
Similarly one defines p-nuclearity, replacing (V.5.16) by

P P 1/P
I e lip= inf (EII(Pi II IIFiII ^ (V.5.17)

where p is some positive number.

Comment. For the relation between the order of a map and nuclearity see
[Pietsch 1972]. We only mention that a map of order q < 1/2 is certainly
p-nuclear with p < q/(1 — 2q).

V.5.2 Nuclearity and Split Property

The split property may be viewed as a sharpening of the locality principle. If

01 and 02 are space-like separated then the algebras R(O 1 ) and R(02) should
not only commute but be "statistically independent". This means first that the
partial states are uncoupled: if cPk is an arbitrary pair of normal states over
R(Ok ), (k = 1, 2) then there exists an extension to a normal state cp over
R.(01 ) V R.(0 2 ). It means furthermore, that this extension can be so chosen
that there are no correlations between the measuring results in 01 and 02 in
the state cp i.e.

cp(AlA2) = (Pi (A1)(P2(A2) for Ah E R(C'k), (V.5.18)

and that for faithful states (ph the extension remains faithful over

R(01 ) V R(02 ).

Irrespective of the physical interpretation it is a natural mathematical

question to ask under what conditions two commuting von Neumann algebras
R1, R2 acting on the same Hilbert space 71 are statistically independent. One
prerequisite is "algebraic independence". This may be reduced to the condition
that A l A2 = 0 with Ak E Rh implies that either A l = 0 or A2 = O. If R1
and R2 are factors then the algebraic independence is automatic (Chapter III,
lemma 2.1.10). In the subsequent discussion we shall always assume that the
Rk are factors. This covers the cases of principal physical interest and though
 V.5 Phase Space, Nuclearity, Sp li t Property, Local Equilibrium 259

this assumption could be avoided, it simplifies some formulations and argu-

ments. Starting now from the algebraic (not yet topological) tensor product
C of Ri, TZ2 consisting of finite linear combinations of products A 1 A2 with
Ak E Rk we have the representation of C acting on 7-1 which we denote by 7r:

7r(A1A2) = A1A2. (V.5.19)

If R 1 and R2 are algebraically independent then we have another representation,

denoted by 7rr , acting on 71® N, generated by

7rP (A1A2) = A 1 ® A2, (V.5.20)

(A 1 acting on the first factor N in the tensor product, A2 on the second). One has

Proposition 5.2.1
Let R, R. be factors acting on N and R C R Let S2 E N be a cyclic and sep-
arating vector for R., R and R A R' In this situation the following conditions
are equivalent
(i) R and R' are statistically independent,
(ii) there exists a vector i E N, cyclic and separating for R V R' satisfying

(77IAB'I77) = (Q01Q)(Q1B'1(2) for A E R, B' E R', (V.5.21)

(iii) there is a unitary operator W from 7-L to N ON such that

Wlr(AB')W* = irp(AB'); A E R., B' E R', (V.5.22)

(iv) there exists an intermediate type-I-factor J1î:

RcNcR. (V.5.23)

If any of these conditions is satisfied the triple (R, R, S2) defines a "standard
split inclusion" in the terminology of [Dopl 84a]. Q is called a standard vector
for the split because it is cyclic and separating for R, R.' and R V R.'. The
existence of a product state vector n in the case of a free field has first been
shown in [Buch 74].

Proof. (i)— (ii). The state w induced by 0 on the various subalgebras un-
der consideration is faithful on R and R and thus, by assumption, there exists
a faithful, normal product state wp on R. V R . Since S2 is cyclic and separating
on R. V 7Z there is a vector representative 77 E 7-E of wp (theorem 2.2.2). The
faithfulness of wp means that i is separating. We may choose ri on the natural
cone of fl and then, by theorem 2.2.3 n is also cyclic.
(ii)—(iii). The map W : 71 -4 71® 71 defined by

W AB' In) = Alf?) ® B'0) for A E R, B' E 7i (V.5.24)

is densely defined and has dense range. From (V.5.21) it follows that W is
isometric. Hence it extends to a unitary map.
260 V. Thermal States and Modular Automorphisms

(iii)—>(iv). The algebra 23(71) ® il is a type-I factor on f ® H. Its image

N= W*B(71)®1lW (V.5.25)

is a type-I-factor on 7-1 which contains

R= WIZ ®i1W (V.5.26)

and commutes with

R = W *Il ® TZ W (V.5.27)
and hence is contained in Ti.
(iv)-->(i). An infinite type-I-factor with infinite commutant is unitarily
equivalent to 93(7-( ® 11 on 71 ® 7-H. Thus we have (V.5.25). If we have fur-
thermore a standard vector S2 and the inclusion relations (V.5.23) then the
identifications (V.5.26), (V.5.27) and

7 1= W* 1 0 ) Oka), wp(C) = (nrin), C E R V TZ (V.5.28)

show that R and R are statistically independent. ❑

Our main objective in this subsection is to show that the split prop-
erty follows from the nuclearity requirement 5.1.2 We start from the following
observation. Let K1, K2 be concentric standard diamonds with radii r 1 , r2
1 = 6 > 0, A E R (K1), B' E R (K2 )'. Then respctivly,wh2—r
[at A, B'j = 0 for jtj < 6. (V.5.29)

Lemma 5.2.2 (Buchholz, D'Antoni, Fredenhagen)

If A and B' satisfy (V.5.29) then one can construct a continuous function f of
a real variable such that
(.01AB'I0) _ (S2IAf(H)B'IQ) + (SQIB f (H)AIS2) (V.5.30)
and which furthermore has "almost exponential decrease":
lim (exp jEjN) if (E)j = 0 for 0 < n < 1. (V.5.31)
lEI—,00

Proof. Consider the following two functions

h+(t) = w ((c_tA)B') = (S2jAeixtB'jS2), (V.5.32) .

h_(t) = w (B'û_A) = Wig Cilit AjS2) (V.5.33)

Due to the positivity of the spectrum of the Hamiltonian h+ is the boundary
value of an analytic function in the upper half plane, h_ is the boundary value
of an analytic function in the lower half plane and, due to (V.5.29) the boundary
 V.5 Phase Space, Nuclearity, Split Property, Local Equilibrium 261

values coincide in the interval —6 < t < 6. So there is a function h(z), analytic
in the cut complex t-plane, the cuts running along the real axis from —oo to
—6 and from 6 to +oo such that h has the boundary values h+ respectively h_
as the cuts are approached from above or below. We have

h(0) = (S2I AB' IS2). (V.5.34)

By the transformation
2Tw
'IV —^ z= (V.5.35)
w2 + 1
the interior of the unit circle in the complex w-plane is mapped onto the
complex z-plane with cuts along the real axis from —oo to — T and from +T
to +oo, the upper half plane corresponding to the upper half disc IwI < 1,
0 < cp - arg w < 7r , the lower half plane to the lower half disc. Evaluation of
h(0) by a Cauchy integral around the unit circle in the w-plane
21r
h ((2re)(1 + e24`A)-1) dcp
h(0) =1 j
yields in terms of vacuum expectation values

27r(,f2IAB' IQ)

= f dco ((QiAexp (iHT/ cos (p) B'IS2) + (.f2IB' exp (iHT/ cos (p) AP)).
(V.5.36)
This relation holds for all T with 0 < T < 6. If we multiply it with a smooth
function g(r) whose support is in the interval [0, 6] and integrate over T we get

(.0 I AB' f2 ) = (QI Af(H)B ' IQ) + (QIB f (H)AI S2)

with

27r f (E) = g(0)-1 f Eço ) where

dcpg ( cos g(E) = f g(T)e 4ET dT. (V.5.37)

Now, as Jaffe has shown [Jaffe 67], there exist smooth functions g with the
indicated support whose Fourier transforms decrease almost exponentially (in
the sense of (V.5.31) and do not vanish at E = O. Picking such a function g one
sees that f is continuous and satisfies the claims of the lemma. ❑

Now, the required nuclearity of the map

®o : R(K,.) — ^^ x ; ê A = e - '311 AI S2); A E R(K r) (V.5.38)

with a nuclearity index bounded by

n
vo < exp( 00 —
f (V.5.39)
262 V. Thermal States and Modular Automorphisms

implies the nuclearity of the map

Of : R(Kr) -4 H; Of A = f (H)AIQ); A E R(Kr) (V.5.40)

when f satisfies the conditions of lemma 5.2.2. To see this write

of = > ek; ek = Pkf (H)e 311 ê0, (V.5.41)

with Pk the spectral projector of H for the interval [k — 1, k]. e k is obviously

nuclear and its index satisfies
n
I1 Ok Iii< fk exp fik + (3° )
(
fk = sup I f (E) I for k-1 <E<k.
(V.5.42)
Since this holds for any positive value of ,3 we pick ,3 for each k so that the
exponent in (V.5.42) becomes minimal. This happens for ,3 = ,3o (n/fio k) 1 /n+1 ,
yielding
11 ek II i < fk exp (Qok) nin+1
,

where the constant en < 2 for all n > 0. Thus

11 e II ekI11< 00 .
From the nuclearity of 01 it follows that we have

f (H)AI.f2) - O f A = cpl (A) IW4 ); A E TZ(K r ) (V.5.43)

where A E 7l cps is a linear form on R(Kr ) and

,
E II cpti 11 11 A 11<
oc. In fact,
since Of gives a map from a von Neumann algebra to normal states over it, the
linear forms cp s can be taken in 'R., the predual of R, i.e. the cpt are normal
linear functionals over R(K,.). Then

(,f2 I B ' f (H)AIO) = E çoti(A) (QI B' I WW)

with an analogous expression for (QI Af (H)B' I ,f2). Putting r = r 1 we see by
(V.5.30) that w is an absolutely convergent sum of factorizing normal linear
forms over the algebraic tensor product C of R(K r,) with R(Kr2 )' and therefore
w is a normal state in the representation 7rp. This shows that the representa-
tions it and 7rp are not disjoint. Actually one finds that it and 7rp are unitarily
equivalent. We shall not prove this here but refer the sceptical reader to [Buch
87c, 91b] for this part of the argument.
Summing up we have

Theorem 5.2.3
If 00 is nuclear with index bounded by

v < exp c ,3'

(for some positive constants c and n) then we have the split property and Q is
a standard vector.
 V.5 Phase Space, Nuclearity, Split Property, Local Equilibrium 263

V.5.3 Open Subsystems

The split property provides a tool for constructing some analogue of finitely
extended subsystems. Instead of one box we have two diamonds Kk with K2 D
K1 ; instead of the ground state in the box we have a product vector i satisfying
(V.5.21). The three objects K1, K2, 77 define the split. They determine W by
(V.5.24) and N by (V.5.25) Let us use the single symbol A for a split. There is
a (linear) subspace RA C f

'HA = NI r» = W * (1-1 ® (2 )W, (V.5.44)

whose vectors correspond to strictly localized states in K2 in the sense of (V.5.2)

(though they do not exhaust them). The projector on RA is

PA = W * (11® P,Q)W E N' , (V.5.45)

with Pa the 1-dimensional projector on ,f2 in 7 -l. One notes that there is an
isometric operator V E 7Z(K2 ) such that

VI.0) = In). (V.5.46)

Indeed, since Q is cyclic and separating for R(K 2 )', setting

VB'1,0) = B In)' for B' E R(K2) ' (V.5.47)

defines V consistently on a dense set of vectors. Relation (V.5.21) implies that

V is isometric on its domain of definition and hence extends to an isometric
operator on R. From (V.5.47) it follows directly that V commutes with any
C' E RZ(K2)' and thus V E R.(K2 ).
This implies that every unit vector in can be obtained by applying
an operator with unit norm from 7Z(K2 ) to the vacuum. Therefore, by the
nuclearity condition 5.1.2, exp —,QH maps the unit ball of H A onto a nuclear set
in 7-1 i.e. exp —,QH PA is a trace class operator and one may define the partition
function for A (to inverse temperature ,Q) as

Zo,A = tr PA e -pH PA . ( V.5.48)

[Jungl 87]. Then, with r2 the radius of K2

43,A VO,r2 (V.5.49)

(compare (V.5.8), (V.5.10)). For thermodynamic arguments (large r 2 and fixed

r2 — r 1 ) one may identify the nuclearity index vp, r itself with the partition
function as originally suggested in [Buch 86a].
To carry the analogy with the treatment of finite systems in non-relativistic
quantum statistical mechanics further one would like to define a "box Hamilto-
nian", charge operators for the box etc. (the box being replaced by a split A). A
first step towards this is the following observation by Buchholz, Doplicher and
Longo [Buch 86c] :
264 V. Thermal States and Modular Automorphisms

Observation 5.3.1
Let G be an unbroken symmetry group of the theory. Thus we have automor-
phisms ag , g E G with geometric action'

a9R(0) = R(gO)

which are implemented by unitaries U(g) E 93(n) leaving the vacuum invariant

U(g)IS)=fQ).

Then there is also a "local representation" U4 (g) of G, given by

UA (g) = W*(U(g) ® 1l)W. (V.5.50)

Besides being a unitary representation of G the UA satisfy

UA(g) E M; Un(g)117) = 1i), (V.5.51)

and, for A E R(K r) with r < r 1 and g sufficiently close to the identity

Un(g)AUn(g) -1 = a9 A. (V.5.52)

(V.5.52) holds if there is a path from the identity to g such that g Kr stays in
K1 for all g' along this path.
Since PA E .N4 we may equally well take the restriction of U4 to 7-h

Ust = UAPA = PnUnPn, (V.5.53)

which also gives a representation of G.

Remark. In field theory the generators of symmetries can be written as integrals

of densities over 3-space at a given time. These densities, e.g. components of the
energy momentum tensor or conserved currents are Wightman fields. Doplicher
suggested that the split property may be used to construct these fields from the
local algebras [Dopl 82]. An attempt to realize this may start from observation
5.3.1 [Buch 86c]. There remain, however, some unresolved difficulties. One re-
sults from the fact that the subspace 7 -14 contains infinitely many orthogonal
states which with respect to observations in K 1 look almost like the vacuum
(with respect to K2 they look exactly like the vacuum). Typical examples of
these are state vectors W * (T f (2) ®i S2)) with T a unitary localized far away from
K2. Such states may be regarded as excitations on the surface of A. There are
too many of them with arbitrarily high energy. Therefore Ht can not yet be
regarded as the box Hamiltonian.
1 In the case of gauge transformations and charges we would have to start here from the
algebra of operations, the field algebra A instead of the observable algebra.
 V.5 Phase Space, Nuclearity, Split Property, Local Equilibrium 265

V.5.4 Modular Nuclearity

The nuclearity condition really concerns the local structure of the theory. This
is not evident from the formulation 5.1.2 which uses the global Hamiltonian
and global vacuum state. The work of Buchholz, D'Antoni and Longo [Bu 90a]
shows how 5.1.2 can be replaced by a local condition. Let K 1 , K2 be concentric
diamonds as above with r2 -- r 1 = S > 0, w a faithful state on 7Z(K 2 ), not nec-
essarily the vacuum. Take the GNS representation of R(K2 ) generated from w.
-

9-1, a cyclic, separating vector S2 E f and the associatedItgivesaHlbrpc

Tomita-Takesaki operators 4 and J. Put
-1
R = (1 + 4-1 /2 ) . (V.5.54)

Then we have

Modular Nuclearity Condition 5.4.1

The map E : R(K1 ) ---0 71 defined by

5A = RAID), A E R(K1) (V.5.55)

is nuclear of some order q.

The map " is closely related to the maps "), defined as

Z.-7),A = 4^ AID), A E 7 (K 1), 0 < a < 2 . (V.5.56)

One notes that the order q), of SA is symmetric around a = 1/4 since

4ÀA1S2) = J4 1 /2-a A*I,f2).

We shall be interested in Ex for a < 1/4 and write ^.#, q# for the case a = 1/4.
One has

Proposition 5.4.2

q), = ( 4A) -1 q#; 2 q# < q < q#, 0 < a < 4 . (V.5.57)

Comment. From the discussion of simple models it appears that the orders
q, qa may be assumed to be zero i.e. the E-content of the maps increases less
than the exponential of any power of E -1 for E — ^^ 0.

Proof. The comparison of the orders q), for different values of a follows from
the simple lemma

Lemma 5.4.3
If 0,, = Tke1 is a family of maps from the Banach space E into a Hilbert space
266 V. Thermal States and Modular Automorphisms

where T is a positive operator whose domain contains the range of 0 1 then for
O < #c <1 the order q,ç of 0,ç satisfies qk _ i-1 qi .

The estimate q < q# in (V.5.57) follows from R (A1/4 + A -1/ 4)-1 A114

and the fact that (4 1/4 + A-1/4) -1 is bounded. To establish the upper bound
for q# one considers the map S * from R(Ki ) to the predual of R(K 2 )' (the set
of normal linear forms on R(1(2) 1 ):

(E*A)(B' ) = (QIB'AIQ); A E R(K i ), B' E R.(K2) ' . (V.5.58)

From the identity

II Al/4A19) 1 1 2= (Qlj(A)AIQ) = ( 5*A)(jA) < II A 11 11 EF..A 11

it follows that the e-contents N#(c), N * (c) of c# and E* satisfy

N#(c) < N* (e2 /2) and therefore q# < 2q* . On the other hand

A-1/2)-1 M2)
(F.*A)(B ' ) _ (QI B'AI c2) = (( 1 + A -1 / 2) B'*S2I (1 +

= (B'* + j (B').flI A) <_ 2 II B' II IJ A II .

So q. <q. ❑

The modular nuclearity condition for the vacuum state and the energy
nuclearity 5.1.2 are closely related. One has

Proposition 5.4.4
Let Q be the vacuum state vector, 0p : R(Ki) -4 n

0pA = e -p"AI,f2), A E R(Ki), (V.5.59)

qp the order of the map 0 0 . Then it follows from requirement 5.1.2 that q# < 40
for 0 = 6/c where 6 = r2 — r1 and c is a numerical constant < 5.

Proof. With A E R(Ki), S JA1/2 the modular operators referring to

K2 we have
h+(t) 2 (0I U(t)SAI (2) = (IIU(t)A*Is2).
For 0 in the domain of S* (or, equivalently, in the domain of A -1 / 2 )

h_(t) - (0ISU(t)AI Q) = (U(t)Af2I S*P) = (Q A*U(—t)IS*1P).

Due to the positivity of the Hamiltonian h+(t) is the boundary value on the
real axis of an analytic function in the upper half complex t-plane, h_ (t)
similarly the boundary value of an analytic function in the lower half plane. For
1 tI < 6 = r2 - r1 we have a t A E R(K2 ) and therefore

h_(t) = (OIat A*IQ) = h+ (t) for iti < 6.

 V.6 The Universal Type of Local Algebras 267

We can then apply the technique of lemma 5.2.2. and obtain

(0 I SAI Q) = (0 I f(H)SAl Q) + (0 I sf(H)AI(2).

Since ,A -1 /2 R is a bounded operator, putting 0 = RW ensures that 0 is in the

domain of S* for any W E H. Thus we obtain

RSAI S2) = R f (H)SAP) + RS f (H)AI S2),

and, since RS = JR and SAID) = A* I ,f2),

RAI S2) = Rf (H)AI S2) + JRf (H)A*I S2); A E R(K1).

Here f is any function constructed as in (V.5.37) and it has the properties de-
scribed in lemma 5.2.2. The nuclearity of 9f discussed in the sequel of (V.5.40)
implies then the nuclearity of S as defined in (V.5.55). The comparison of the
orders of the maps S. and 9p is achieved by means of the following lemma.

Lemma 5.4.5
There is a numerical constant c > 0 such that for (3 < (r2 - ri )/c and any
W E 7-1
inf II f(H)IW) II<—
k--1 II e--sHIW)
Il k . II Ic) 11 1-k ,
where the infimum is taken over all choices of f satisfying the required condi-
tions and k = (l. + c(3/(r2— r1)) -1

This leads to the upper bound for q# stated in proposition 5.4.4. We refe.
for this computation to the original paper [Buch 90a]. ❑

V.6 The Universal Type of Local Algebras

We claim now that the algebra R(K) of a diamond is (as a W*-algebra) iso-
morphic to a unique mathematical object: the hyperfinite factor of type III 1 .
This means that physical information distinguishing different theories or dif-
ferent sizes of K is not contained in the algebraic structure or topology of an
individual algebra R(K). The information comes from the relation between the
algebras of different regions, from the net. The universality of R(K) may be
seen as analogous to the situation in quantum mechanics where we can associate
to each system or subsystem an algebra of type I, i.e. an algebra isomorphic
to the set of all bounded operators on a Hilbert space. The change from the
materially defined systems in mechanics to "open subsystems" corresponding
to sharply defined regions in space-time in a relativistic local theory forces the
change in the nature of the algebras from type I to type M i . The fact that
there is only one hyperfinite factor of type III 1 up to W*-isomorphy has been
proved by Haagerup [Haager 87] based on work by Connes.
268 V. Thermal States and Modular Automorphisms

A von Neumann algebra R. is called hyperfinite if it is the weak closure of

an ascending sequence of finite dimensional algebras, in short if it is the W*-
inductive limit of finite dimensional algebras. One should be aware of the fact
that a subalgebra of a hyperfinite algebra is not necessarily again hyperfinite.
But the W*-inductive limit of an ascending sequence of hyperfinite algebras
is hyperfinite. Clearly the factor I mo , the set of all bounded operators on a
separable Hilbert space, is hyperfinite. It is the inductive limit of the matrix
algebras of finite dimensional subspaces. The hyperfiniteness of R(K) follows
then from the split property provided that this property holds for any pair of
concentric diamonds with radii r 2 > r i for arbitrarily small r2 — r i since R(K2)
is the inductive limit of the intermediate type I factors as r i tends to r2. We
have seen that this in turn is a consequence of the nuclearity assumption if the
growth of the nuclearity index with the temperature is bounded by (V.5.8) i.e.
if the level density does not increase too fast with the energy (no Hagedorn
temperature).
Reasons for demanding that R(K) should be a factor have been discussed
earlier (see e.g. section 5 of Chapter III). Perhaps one should keep in 'mind
that a more thorough study of this question is desirable. At present we see
no indication suggesting that local W*-algebras might have a non trivial center.
However, if one wants to envisage this possibility then the claim at the beginning
of this section has to be replaced by the more cautious statement: in the central
decomposition of R(K) only factors of type III i can appear.
The conviction that R.(K) must be of type III dates from the work of Araki
[Ara 64c] who established this in a theory of free fields by explicit computation.
It was recognized by Driessler [Driess 77] that in a theory which is invariant
under dilatations there is a very simple argument showing that the relative
dimension function of projectors in R(K) can only take the values 0 and oo. We
recall that this was von Neumann's original definition of type III (see subsection
2.1 of Chapter III). Using theorem 2.4.1 of the present Chapter, one can sharpen
this and say that the type must be III i in this case. A dilatation by a scale factor
a maps R(K) onto R(ÀK). For a —^^ 0 R(K) is asymptotically Abelian with
respect to this action since the region 05, = (AK)' n K, the intersection of the
causal complement of AK with K grows with decreasing a and the W*-inductive
limit of R(Oa) for a — ^^ 0 is R(K).
Of course we cannot have (unbroken) dilatation invariance if there are non-
vanishing masses. But there are good reasons to believe that a realistic theory
should become conformally invariant in the short distance limit. Among them
we may count the dominant rôle of chirality in the standard model of elemen-
tary particle physics. The existence of an "ultraviolet fixed point", of a scaling
limit, also plays a crucial rôle for quantum field theory in curved space-time
(see Chapter VIII, section 3). In our context this relates to the question of the
geometric significance of the modular automorphisms of R(K). If the theory has
a scaling limit then the action should be approximately geometric in a well de-
fined sense. This question was studied by Fredenhagen [Fred 85a]. We describe
the main results but refer for the proofs to the original paper.
 V.6 The Universal Type of Local Algebras 269

First, let Rk (k = 1, 2) be von Neumann algebras with a common cyclic

and separating vector Q, Sk the Tomita operators defined in (V.2.1), ,Ak the
modular operators and (A k) the modular automorphisms. If R2 D R1, the do-
main of S2 contains the domain of Si and in restriction to the domain of S i
2 and Si are equal. This implies some relations between the thewoprasS
modular operators. In particular one has

Proposition 6.1
Let R2 D R1 and A E Ri such that a 2) A E Ri for Itl < T. Then

II {( 1 + 4 2 ) -1 - (1+ 4 1 ) -1 7 AO II

< 2 tanh_ 1 (e - ") { I AO 11 2 + II A* Q II 2 } 1/2. (V.6.1)

TF

One applies this estimate to the case where R2 is the algebra of the wedge
region W (see (V.4.1)) in the vacuum sector, R 1 is the algebra of a diamond K
tangent to the wedge at the origin, Q is the vacuum state vector. As A decreases
AK will contract to a small subdiamond of the same nature, ultimately to the
origin. Since we know the geometric significance of the modular automorphisms
of the wedge algebra we can estimate the parameter range of t for which an el-
ement of R(AK) will stay in R(K) under the action of W ). Then proposition
6.1. leads to

Proposition 6.2
For each f E £( ') (IR) and 0 < A < 1 there exists a constant cf(À) such that

II f dt f (t) { z - 41 A,f211 2 < cf(A) {II AQ 112 + 11 A* Q 11 2 } (V.6.2)

for all A E R(AK) and one has

c f(A) -> 0 as A -; 0. (V.6.3)
The constant c f neither depends on the details of the theory nor on the size of K.

This shows that the action of all° on the elements of R(AK) becomes al-
most geometric for small A; it approaches the action of Qtr) . This can be used
to show that if the theory has a scaling limit then the spectra of , K' and - w
1Z(K) we need information on [Link]
the spectra of modular operators for all normal states, not only for the vacuum.
To obtain this Fredenhagen proves the following variant of the theorem 2.4.1.

Proposition 6.3
Let w be a faithful normal state on a von Neumann algebra R and A, its mod-
ular operator. Then a necessary and sufficient condition for the number k E IR+
270 V. Thermal States and Modular Automorphisms

to be in the spectrum of d,, is that for each e > 0 there is some A E R with
w(A*A)= 1 such that for all B E R
1 2
iw(AB) — kw(BA)1 < Œ fw(B *B) + kw(BB*)} / . (V.6.4)

He then proceeds to show that in a theory which has a scaling limit this
criterion is satisfied for all values of k E IR+ . So the Connes invariant is maximal,
the type is III1.
VI. Particles. Completeness
of the Particle Picture

VI.1 Detectors, Coincidence Arrangements,

Cross Sections

VI.1.1 Generalities

In section 1 of Chapter III we claimed that the theory is fully characterized

once the abstract net of observable algebras 240) is known. This means that
all physical consequences must be derivable from this net. For comparison with
experiments in high energy physics the most interesting ones concern the types
of particles which occur and their collision cross sections. What is a particle?
In section 3 of Chapter I we argued that the state vectors in an irreducible
representation of 13 describe a single particle, a particle alone in the world. This
was taken as the starting point for the discussion of the particle aspects of the
theory in Chapter II, sections 3 and 4. Within the Hilbert space setting provided
by the Wightman axioms the single particle subspace f(') was defined as the
part belonging to the discrete spectrum of the mass operator M = (P, Pui)1I2 .
From 71{1 M we constructed the states of asymptotic particle configurations by
introducing a product composition ®t between state vectors which becomes
unambiguous and has an obvious physical interpretation when it is applied to
states which at time t are localized in far separated regions. It was then shown
that this condition is satisfied at asymptotic times for any subset of single
particle states and that ®t converges for t —+ ±co to a composition ®out, ®in,
respectively of the state vectors of H('). The proof depended on the assumption
of a fast decrease of correlation functions with increasing space-like separation.
This assumption in turn was shown to be a consequence of locality and positivity
of energy provided that the vacuum was clearly separated from all other states
by a mass gap. When compared to the claim at the beginning of this section
the results and methods' of Chapter II fall short on several counts.
(i) They make essential use of unobservable (charge carrying) fields. Al-
ternatively put, the Hilbert space used must contain all superselection sectors.
Though this is no objection in principle to the claim that the net of observables
1 With the exception of the approach by Araki and Haag [Ara 67] which was briefly de-
scribed in II.4.3. This approach will be carried further below.
272 VI. Particles. Completeness of the Particle Picture

suffices for the construction of collision states because the unobservable fields
may be constructed with the methods of Chapter IV, it indicates that a more
direct approach, closer to experimental procedure, is warranted.
(ii) The existence of a discrete part of the mass spectrum and the com-
pleteness of the particle picture (f in = fout = ?`l) was put in by assumption,
unrelated to the properties of the local net.
(iii) The method fails if there is no mass gap. It may be argued that in the
regime of high energy collisions the additional complications connected with the
vanishing of the photon mass (slow decrease of the correlations in the vacuum
state, infrared photon clouds) are of peripheral interest and that their effect can
be dealt with once collision theory in a purely massive model is understood. Yet
it is unsatisfactory to base the approach on a simplification which we know to
be not true in reality. We should take this also as an indication for the need to
find a more natural approach.
Taking up the suggestion from item (i) we note that experimentally all
information comes from the use of detectors and coincidence arrangements of
detectors. The essential features used are that a detector is a macroscopically
well localized positive observable which gives no signal in the vacuum state. In
the mathematical set up of the theory the two requirements cannot be strictly
reconciled due to the Reeh-Schlieder theorem; the algebra R(0) of a strictly
finite region does not contain positive operators with vanishing vacuum expec-
tation value. However we may represent a detector centered at the origin by a
positive, almost local element C of the algebra of observables, sufficiently well
approximated in norm by an element C,. E 7Z(K,.). Specifically

C E %+; 11C 11= 1; w o (C) 0,

and, for any prescribed (arbitrarily small) inaccuracy e > 0 there is a radius r
such that
C — Cr 11 < e for some Cr E 7Z+ (Kr ). (VI.1.2)
We may consider r for given e as the essential size of the detector. A detector
centered at the point x is then, of course, represented by

C(x) = ax C, (VI.1.3)

and a coincidence arrangement of n detectors Ck at time t is described by the

product
D,,, = Ci(xi) ... C, (x,2 ); xk = (t, xk) (VI.1.4)
with xti — xk I> R where R is large compared to the essential extensions rk of
the detectors Ck. If in a state w we have w(C) = p with p large compared to e
we know that the state deviates noticeably from the vacuum in the region Kr ,
it has a local excitation. If w(Dn ) differs significantly from zero we know that
w has at least n coexisting localization centers at time t.
Before proceeding further we should consider a question which will be un-
doubtedly on the mind of one who strives for a precise correspondence between
 VL1 Detectors, Coincidence Arrangements, Cross Sections 273

experimental procedure and mathematical representation. Suppose an experi-

mentalist shows you an instrument saying that this is a very good muon detec-
tor. He gives you the engineering manual which describes how it is constructed.
Then he says with an innocent smile: can you work out the algebraic element
which corresponds to this manual in the mathematical description? Conversely,
given C satisfying (VI.1.1), (VI.1.2), how can one construct a corresponding
instrument? The answer is: we do not want to bother and we need not know
in such detail. On the experimental side the development of an apparatus opti-
mally suited for a particular purpose (highly selective sensitivity) is a lengthy
process involving trial and error. The same situation prevails on the theoretical
side. The conditions (VI.1.1) and (VI.1.2) are so weak that they allow infinitely
many algebraic elements. Most of them will correspond to poor detectors if one
has a specific task in view. But theoretical guidance for choosing suitable ones
exists. We may choose C of the form

C = L*L, (VI.1.5)

where L E 2(S has energy-momentum transfer limited to a prescribed region 4

in p-space in the complement of the closed forward cone i.e.

n V+ = 0,
f L(x)f(x) d 4x = 0 for supp f c d° ; d (VI.1.6)

where f denotes the Fourier transform of f and L(x) = ax L. In addition L shall

be essentially contained in 7Z(K r ) in the sense of (VI.1.2). Of course, if q is the
diameter of 4 this is possible only for

rq »1.
One can define the effective volume of the detector as

vc - 3 c IIf [L(x),L LII ] II d3x ; ^° = 0

r =
II
. (VI.1.7)

One should note that we can choose L in the observable algebra even if we want
to register particles which carry some charge and have a sharp mass. For this
purpose we have to choose a space-like. Then L can have nonvanishing matrix
elements between state vectors in the charged sector on a mass hyperboloid. We
shall furthermore choose d to be sufficiently separated from V+. Then not only
the vacuum but all states with energy below some value 6 will be annihilated
by L and thus C registers only excitations with energy above 6 in the volume
Vc. 6 fixes a boundary between "hard excitations" which are registered and
"soft excitations" which are not registered. Without loss of generality we may
restrict attention to states which carry a total energy below some value E. Then
the compactness requirement in the form suggested by Fredenhagen and Hertel
(see Chapter V, subsection 5.1) suggests that a finite number of different Ck will
suffice to determine the partial state in the detector region (i.e. the restriction
of w to R(Kr ) up to any desired accuracy in the norm topology of TZ(K r ) * ). The
274 VI. Particles. Completeness of the Particle Picture

sensitivity of a detector to specific states has to be tested by monitoring. Once

the single particle states of various types are known (see below) it is a finite
problem to choose detectors, selectively sensitive to a specific type. Furthermore,
the supports dk of the chosen Lk will then also give information about the
energy-momentum content. We can distinguish local excitations which carry
different momenta. In this way we can ultimately deduce all information from
the probabilities for response of geometric arrangements of detectors, knowing
a priori not more than that each of them responds to some local excitation
carrying some approximate momentum.
Let us reconsider now our understanding of the particle picture. We call a
state w at most n-fold localized at time t if it cannot trigger any (n + 1)-fold
coincidence arrangement at this time. We shall express this as the demand that 2

Jlx;-xkj>R
wf e1 (t, X1) . . . Cn+1(t, Xn +1) }d 3 d3 < ED> (VI.1.8)

for any choice of the detectors conforming with the mentioned requirements.
Here we may first fix the tolerated background probability eD , choose the sizes
rk of the detectors and the separation distance R which has to be large compared
to every r k . If we use preexisting information about the energy bounds of the
state w then a finite number of different choices of the detectors Ck will suffice in
the test (VI.1.8). Clearly the choice of rk and, more significantly, of R determines
what we regard as a single localization center. We are interested in the regime
of "small particle physics" where we expect to encounter matter concentrations
of small intrinsic extension, not rocks or other large chunks of cohesive matter.
This will already be ensured by the energy bounds on the state since we know
from experience that with growing extension of cohesive matter also the mass
will increase. So for each energy bound E there is a radius RE which bounds the
intrinsic size of possible chunks of cohesive matter. This empirical knowledge
must, of course, also follow from the theory. We have included it in the property
(VI.1.13) below which implies that, given the total maximal energy, we can
choose R sufficiently large so that a signal from an n-fold coincidence implies
that there are at least n distinct coexisting localization centers at the respective
time.
If w satisfies (VI.1.8) and has no component with less than n localization
centers i.e.
w 0 awn-1+(1— A)w'; for a# 0,
where wn_1 is a state satisfying (VI.1.8) with n replaced by n — 1, then we say
that w is exactly n-fold localized at the time.

Remarks. (i) Since we are integrating in (VI.1.8) over infinitely extended space
one might worry that the left hand side could become infinite and the condition
empty. Indeed f C(x)d 3x is not an element of the quasilocal algebra and it is
2 lnsteadof the integral one might, at this stage, equally well take the supremum of the
integrand for J x i — xk I> R as the relevant quantity. We shall see, however, that (VI.1.8) is
the most convenient starting point for the analysis. See remark (i) below.
 VI.1 Detectors, Coincidence Arrangements, Cross Sections 275

unbounded. However, within the subset of states with total energy below E
there is a uniform bound

w (f C(x)d 3 x) < ( E. (VI.1.9)

This key lemma has been proved by Buchholz (lemma 2.2. in [Buch 90b]). The
only feature needed besides locality is the restriction that we are dealing with
states in a sector in which the translations can be implemented by unitaries
satisfying the spectrum condition. An intuitive argument runs as follows. The
response of the detector indicates that there is an energy larger than S in the
volume Vc around the point x. So the expectation value of the energy density
at time t at x is larger than 6/V c w(C(x)) and the total energy of the state
larger than bIVC f w (C(x)d 3 x) .
(ii) Instead of C(x 1 ) ... C(xn+1 ) in (VI.1.8) we may equally well use

W (L* (xi) ... L*(x n+1)L(x n+1 ) ... L(x1)) d 3 xi . . . d3x n+1 < eD
Li - xk I >R
(VI.1.10)
since the commutation of a factor L(x) with L(x + y) or L*(x + y) introduces
only a small change when 1 y 1> R, a change which remains small even when
integrated over y. This shows that for states with energy below E the maximal
number of localization centers stays uniformly bounded for all times by
E
nmax = b. . ( VI.1.11)

We may generalize (VI.1.9) to

0 < fw(L* (xl)... L*(x fl+l )L(x fl+l )...L(xl))Hd 3 xk

< Vc+lnmax(nmax — 1) . . . X (nmax — n). (VI.1.12)

VI.1.2 Asymptotic Particle Configurations

A state describing a single stable particle, alone in the world, can be char-
acterized as a state which is singly localized at all times; it cannot trigger a
two-fold coincidence at any time. Actually, since we consider detectors which
have a threshold S as the minimal energy needed for a response, we are talking
here about a single "hard" particle. It may be accompanied by an unregistered
background whose energy density exceeds nowhere the threshold level 6/VC. We
note, however, that "hard" does not mean massive. Photons of energy beyond
6 are included.
For an intuitive picture it is helpful to introduce an (unbounded, non local)
observable N(t), the localization number at time t which assigns the value n to
a state wich is n-fold localized at time t. Mathematically it is a linear form over
276 VI. Particles. Completeness of the Particle Picture

a domain in 21* which contains the states of bounded energy. More intuitively it
may be considered as an unbounded operator with non negative integer eigen-
values in a Hilbert space. Of course N(t) depends on the choice of R, 6D and of
the r k . But we expect that this becomes irrelevant for large times. We expect
that with increasing time the configuration expands, its diameter growing lin-
early with time. The probability densities w (C(t, xi ) ... C(t, xn+z )) then must
decrease correspondingly because the integral over the x k stays bounded. In
other words the theory must have the property that for sufficiently large times
t > T (depending on w)

IRE < 'xi -x2 l <R(t)

W (Ci ( t, X l) C ( t, X2)) d3X1d 3 12 < 6

if we take for any 0<k<1

R(t) < const. t In .

The essential part of the coincidence integrals (VI.1.8) for large times is expected
to come from separations R > R(t) so that we may let R grow with I t I in
the definition of the localization number N(t) and, correspondingly let ED tend
to zero. This means that the only parameter which remains relevant for the
definition of N(t) at large times is 8. Note that (VI.1.13) contains also the
(expected) property that the intrinsic extension of cohesive matter is bounded
by RE independently of the time.
If (VI.1.13) is satisfied we can expect that for sufficiently large times the
individual localization centers in an n-fold localized state will no longer influence
each other. Since the maximal localization number is uniformly bounded by
(VI.1.11) this means that each asymptotic localization center will be a stable,
hard particle (possibly accompanied by an unregistered background with energy
density below (5/VC). In the limit N(t) will become the asymptotic particle
number. More precisely, we expect that

(AO) = f w(ax L *AL) d3x, x _ (t, x), A E 7Z(0) (VI.1.15)

should converge for I t I -i oo and, if

lim II got II= lim f W(a x L*L)d 3 x 0 ( V I. 1. 1 6)

we anticipate that the limit of (VI.1.15) gives (apart from normalization) the
expectation value of A in a state which is permanently singly localized i.e. the
limit describes a mixture of stable single particle states.
Starting from the qualitative picture described above Buchholz drew up the
following strategy for the analysis of the particle content of the theory [Buch
87a] .
1) One wants to show the convergence (VI.1.15) and the existence of a
nonvanishing limit for some choices of L. The set of nonvanishing limits gives the
 VI.1 Detectors, Coincidence Arrangements, Cross Sections 277

particle content. We may note that closely related to the question of convergence
is the property (VI.1.13).
2) In order to obtain a mathematically well defined object which can re-
place the "single particle subspace" H (1) of Chapter II one must remove the
arbitrary threshold b. If the theory has no mass gap then this step may demand
a generalization of the notion of state for a sharply defined particle. This is
no tragedy. Instead of states we must consider weights on the algebra 2t. This
is mathematically well defined. In our context a weight may be regarded as a
positive linear form on the subalgebra C of detectors. A precise definition will be
given later. C does not contain the unit element and therefore a weight cannot
be normalized in the standard way. The single particle weights are the limit
elements for x° —p oc of

f w (axC) d3x; CEC (VI.1.17)

for states of bounded energy (considered as positive linear forms over C).
3) One studies the decomposition of single particle weights into pure com-
ponents. This replaces the familiar decomposition of 7 -t(1) into irreducible parts.
It turns out that the decomposition leads to pure single particle weights which
have sharp momentum, mass and spin and allow the distinction of different
particle types occurring in the theory.
4) Using a pure single particle weight (instead of a state) in the GNS-
construction one obtains a representation of C. One shows then that under the
standard assumptions for the theory this representation can be extended to a
representation of 21 on a separable Hilbert space which is locally normal i.e.
equivalent to the vacuum representation when restricted to the subalgebra of a
finite region. The pure weight itself appears as an improper state vector, like
a plane wave in quantum mechanics, which we denote by I p, a). Here p is
the energy-momentum of the weight, a combines the remaining classification
parameters i.e. spin and particle type. One would like to show that for fixed p
the index a can run only through a finite number of values and also that the
set of possible mass values m = (pop,4 ) 1/2 is a discrete set.
5) The functions

C—+ro(C;p)=(p, p, a) (VI.1.18)

give the sensitivity of the detector C. If there is only a finite number of particle
types in a finite mass interval one can then construct special elements Cry which
are sensitive only for a particular value of the index 7:

ra0( 01 ; p) = (5 (5 yC"(p). (VI.1.19)

With the help of these one can find the momentum-space densities Ayu t (p), oy (p)
at asymptotic times ±oo for any state w by the formula

f -10d3p ^^ lim fw (Ch ) (t, x)} h ( t) d3x. (VI.1.20)

C (ry) (p)w ((p)) h p t f^
278 VI. Particles. Completeness of the Particle Picture

Here ± stands for out, in, respectively and h is any smooth function. Letting
h tend towards a 6-function in velocity space one gets the probability density
for finding a particle of type 'y with the corresponding asymptotic momentum
in the state w. The formula (VI.1.20) was derived in [Ara 671 under much
more restrictive assumptions. As outlined there, it allows the determination
of the collision cross sections. In the present context (VI.1.20) follows once
one can identify the asymptotic velocity of a particle with p/p°. To obtain a
useful algorithm for the computation of the cross sections one has to devise
an efficient way of filtering out states with prescribed incoming configurations.
One way of doing this, using only the vacuum expectation values of observables,
has been devised by Buchholz and Stein [Stein 891, [Buch 91a}. It is necessarily
more tedious than the methods described in Chapter II and it gives the cross
sections, not the S-matrix. But it is clear that a price has to be paid if one
does not restrict attention to theories with a mass gap and wants to avoid using
charge carrying fields. The price is remarkably low compared to the difficulties
incurred if one really wants a precise formula for hadron cross sections in QCD
in terms of Green's functions of quark fields.
6) There remains the question of completeness of the particle picture. This
is the question whether a state w is completely specified by the knowledge of its
asymptotic particle content. This may again be reduced to the convergence of
an expression like (VI.1.17). If there is an energy-momentum tensor so that the
total energy can be expressed as a space integral over a local density then one
can take instead of C in (VI.1.17) the (suitably smeared out) energy density.
The convergence of this implies that the total energy can be accounted for by the
contributions from the single particle weights in the asymptotic configuration.
How much of this program has been achieved to date? If one focuses on the
aim of relating all the features mentioned to simple structural properties of the
theory which, apart from locality, spectrum condition and Poincarè symmetry
will have to include nuclearity and some aspects of the dynamical law (e.g.
asymptotic Abelianness of 2f with respect to a t ) then there remain at present
some gaps. There are plausibility arguments indicating that these gaps can be
filled. The precise conditions which are necessary and sufficient to ensure the
convergence of (VI.1.15) are not yet known. One knows that (VI.1.15) has weak
limit points and that any one of these is a single particle weight. The first
problem appears therefore to find lower bounds for (VI.1.15) which exclude
the vanishing of all limits. If one has an energy-momentum tensor then the
argument indicated under item 6) of the strategy can serve this purpose. The
decomposition theory of single particle weights and the analysis of the properties
of pure single particle weights has been carried through. We shall describe this
in the next section. Still it is an essential and not yet accomplished task to
show the convergence of (V1.1.15) because, if there are several limit points, one
cannot derive the crucial formula (VI.1.20) on which the determination of cross
sections depends.
 VI.2 The Particle Content 279

VI.2 The Particle Content

VI.2.1 Particles and Infraparticles

In the last section a single particle state was defined as a state which is perma-
nently singly localized. How does this geometric characterization relate to the
more common one which associates single particle states with the discrete part
of the mass spectrum? For the case of a theory with a minimal nonvanishing
mass this has been discussed by Enss. He finds that in this case the two char-
acterizations are equivalent if the theory satisfies a compactness requirement
[Enss 75]. We indicate the argument.
In a purely massive theory we can implement the Poincaré symmetry by
unitary operators in the Hilbert space containing all sectors and we can split
off the center of mass motion in the subspace orthogonal to the vacuum as
described in Chapter I, subsection 3.4 Thus we can write

h = 7-lc ®h, (VI.2.1)

where T(c = ,C( 2 )(1R3 ) describes the center of mass motion and h takes care
of all the remaining degrees of freedom. A general state vector if in N can be
written as a function on 1R3 with values in h

p E 1R3--^ W(p)Eh•

Here p is the center of mass momentum and

(W I W) = f (W(p) I W(p)) dap.

If W has total energy below E and is localized at some time, say at t = 0, in

some given region then the compactness requirement says that there is, for each
p a finite dimensional subspace 1) /v (p) C h such that W(p) can be approximated
sufficiently well by a vector in 11N (p) (N denoting the dimension). Moreover, if
we shift the localization region by a spatial translation the subspace 13 N (p) does
not change since such a translation multiplies W(p) only by a phase factor. So
one concludes that the components of singly localized states (for a fixed choice of
R) at time t = 0 lie essentially in 4 N (p) . For each time t we have such subspaces
4N (p, t) C fj characterizing the states which are singly localized at the respective
time and they result from 11 N (p) by application of eixt since the Hamiltonian
H commutes with the space translations. The p-component of a permanently
singly localized state must lie in the intersection of all these subspaces. If there
are permanently singly localized states this intersection cannot be empty. So for
each p there must be a finite dimensional subspace of 1) which is almost stable
under the mass operator. The mass operator M must have a discrete part in its
spectrum. The converse, namely that a subspace of N-N, belonging to a discrete
eigenvalue of M, contains only states which are permanently singly localized
follows from the discussion in Chapter II, section 4 in theories with mass gap.
280 VI. Particles. Completeness of the Particle Picture

If the theory has no mass gap the above argument fails on two counts. For
mass zero states we cannot separate off the center of mass motion; in addition
the Poincarè symmetry may not be implementable in all sectors. It may then
be no longer true that the existence of a particle can be recognized by the
appearance of a discrete eigenvalue of M. In quantum electrodynamics the 1-
electron states do not belong to an eigenspace of M; the electron mass is the
lower bound of the mass spectrum in the sector of charge 1 but the hyperboloid
p2 = m2 carries zero weight in the spectral decomposition. This is one aspect
of the "infrared problem" in QED. It was worked out by Schroer in a simplified
model and he coined the term infraparticle for a particle like the electron which
is not associated to a discrete eigenvalue of M [Schroer 63].
In the standard field theoretic approach this aspect concerns the nature
of the singularity of the Feynman amplitudes at p2 = m 2 . Unfortunately even
the nature of the singularity depends on gauge conventions. So it is difficult to
extract the information about the spectrum of M in the physical Hilbert space.
See [Kibble 68 a, b], Faddeev and Kulish [Fadd 71]. Focusing attention on the
observables the infraparticle aspect and the spontaneous breaking of the Lorentz
symmetry has been taken up again by Frohlich, Morchio and Strocchi [Fr h 79a,
b] and by Buchholz who showed that the absence of a discrete eigenvalue of M
for states with an electric charge is a direct consequence of Gauss' law [Buch
86d]. We sketch this argument.
Let Fw"(x) denote the electromagnetic field, considered as an operator val-
ued distribution acting in the Hilbert space of a primary representation of
the observable algebra in which the translations are implementable with Po-
spectrum in V + . Pick a test function f with compact support in a region space-
like to the origin and scale it:

fR(x) = R -2 f (R -1 x) (VI.2.4)

Then, for large R the (unbounded) observable F'v(fR ) may be regarded as

some weighted average of the flux through large spheres around the origin.
Specifically, if we take in polar coordinates

f (x) = fi (t, r)f2(V, (P)

then
(f R ) = fP'(Rt,Rr)f i (t,r)dtdr, (VI.2.5)

where

VL(t , ri ) = J Fi"(t/ , rf , , p) f2 (V, (p)r 12 sinz9 d9dco

( (VI.2.6)

is the flux through a sphere of radius r' at time t , averaged over the angles
with the weight function 12 . The essential input for the subsequent argument is
the claim that for all states of interest in elementary particle physics the limit
of the expectation value
 VI.2 The Particle Content 281

(VI.2.7)
R— oo w (FP v(fR)) - fuU(f
^
)

exists, does not vanish for all f in charged states and that the fluctuation stays
bounded as R * 00
—

('(m)) < oo. (VI.2.8)

The justification of these claims relies on Gauss' law, due to which we should
be able to measure the charge of a state by the (space-like asymptotic) flux
of the electric field. In charged states the expectation value of Fuv(x) at large
distances should deviate sufficiently from the vacuum expectation value to yield
a nonvanishing limit as demanded in (VI.2.7) but the fluctuations should re-
main bounded. The fluctuations (VI.2.8) in the vacuum state can be estimated
using standard properties of Wightman functions. They involve the correlation
between FA' (x) and Pm' (y) for very large space-like x and y. One does not
expect that this estimate is significantly affected by the presence of charges in
finite regions.
Granted (VL2.7), (VI.2.8) one notes that in a primary folium Fuv(fR ) con-
verges weakly to a multiple of the identity (on a dense domain) since the limit is
affiliated with the center (compare theorem 3.2.2 in Chapter III) . Thus f µv (f )
is independent of the state w in the folium. Now one considers the commutator
of the mass operator with Fu". One has

(VI.2.9)
[M2' Fl `v (fR)] = (Pe FPv ((f,e)R) + FAv ((f,e)R)P) ,

where f, e = ae f. If w is a state with sharp mass the expectation value of the left
hand side vanishes and, if it is a state with bounded energy, we can use (VI.2.7)
to evaluate the limit of the bracket on the right hand side. So we obtain for
such states
w(P)ce = 0 for cc, = f "'(f,e ). (VI.2.10)
Since co cannot vanish for all choices of f and of indices jc y the momentum
,

spectrum of states with sharp mass is restricted to a subset of the mass hyper-
boloid which has lower dimension i.e. the directions of the spatial momentum
would have to be restricted for such states. This is impossible for normalizable
states in the folia under consideration.
So one can conclude that at least all electrically charged particles are infra-
particles. They do not correspond to a discrete eigenvalue of M; moreover one
finds that the Lorentz symmetry is not implementable in a sector of states with
nonvanishing electric charge. The Lorentz symmetry is spontaneaously broken.
Nevertheless one can attribute a sharp mass to a charged particle and one has
a well defined discrete set of mass values for the particle types occurring in the
theory. The point is only that there remains no (normalizable) charged state
which is permanently singly localized if we let the threshold of the allowed de-
tectors tend to zero. There remain "improper single particle states", weights
up corresponding to an electron with sharp energy-momentum p; the values of
p which occur fill the mass hyperboloid p2 = m2 but the representations of 21.
which are induced by weights with different p are inequivalent. One may say
282 VI. Particles. Completeness of the Particle Picture

that the velocity of the electron gives a superselection rule. This has a simple
physical reason. An electrically charged particle moving with constant veloc-
ity is accompanied by an electromagnetic field (e.g. the Lorentz transformed
Coulomb field of a particle at rest). For different velocities the flux of this field
through some segment of a sphere with arbitrarily large radius will be different
(and nonvanishing). The asymptotic fl ux cannot be changed by the action of
any element of the quasilocal algebra.
As a consequence of this superselection rule no coherent superpositions of
weights Qp with different values of p are possible, no wave packets correspond-
ing to normalizable strict 1-electron states can be formed. Yet electron inter-
ference is a salient fact which is explained in quantum mechanics by applying
the superposition principle to 1-electron wave packets. Obviously this quantum
mechanical idealization is good enough for the discussion of electron interfer-
ence experiments in spite of the fact that QED tells us that, strictly speaking,
there are no such coherent wave packets. The seeming paradox may serve as a
warning against overrating the significance of idealizations in the mathematical
description of a physical situation. The reader is encouraged to work out how
the quantum mechanical description of an electron interference experiment can
be justified within the field theoretic setting. Here we only remark that such an
experiment concerns the partial state in a finite space-time region and that the
initial information we have about it is only up to some background with energy
density below some threshold. The phenomenon studied must be insensitive to
this ignorance. Thus also the soft electromagnetic radiation which is necessarily
generated by the interaction of the primary electron beam with a diffracting
crystal and external electromagnetic fields changing the electron velocity may
be ignored. It causes an uncertainty of the quantum mechanical wave function
which remains irrelevant for the phenomenon. In cosmological applications of
quantum field theory one should, however, be careful to take the infraparticle
aspect into account.

VL2.2 Single Particle Weights an d Their Decomposition

The following analysis, relying on (partly unpublished) arguments and results of

Buchholz and Porrmann [Buch 91a, 941 is based on the geometric definition of
the particle concept and applies therefore equally to particles and infraparticles.
It covers essentially the items 2) - 4) of the strategy outlined at the end of
subsection 1.2. We state here the main results and give some comments and
sketches of proofs.
A representation of Zs in which the translation group is implementable
with PP-spectrum contained in V + will be called, for short, a positive energy
representation. In the following 6 denotes the set of all physically allowed states
which generate via the GNS-construction a positive energy representation and
SE the subset of states with energy below E. The unitaries implementing the
translations in a positive energy representation can be fixed uniquely by the
following theorem.
 VL2 The Particle Content 283

Theorem 2.2.1
Let 7r be a positive energy representation of is and 7r(%5)" = /Z. Then:
a) There is a choice of the unitary group U(x) implementing the translation
automorphisms ax
7r(ax A) = U(x)7(A)U(x)*
such that
U(x) E 7Z. (VI.2.11)
This fixes U(x) up to unitaries in the center R. n 7Z .
b) There is a choice of U(x) E R. such that the Pa-spectrum has a Lorentz
invariant lower boundary in each subspace of 7l which reduces R. i.e. in each
P1-( when P is a projector from R.Q .

Properties a) and b) fix U(x) uniquely. We call the representation x —; U(x)

having these properties the canonical implementation of ax .

Part a) of the theorem is due to Borchers [Borch 661, part b) to Borchers and
Buchholz [Borch 85]. We shall not give the proof.
Let 4 c t5 be the set of almost local annihilators and G the left ideal in
2.s, algebraically generated from Co. Its elements are finite linear combinations
of elements of the form AL 0 with A E 21s, Lo E 4. Thus G is a subset of the
Doplicher ideal (III.3.21) and differs from it only by the additional requirement
that La shall be almost local and G is not completed in the norm topology. Let
G* be the set of adjoints of G and

C = G*G (VI.2.12)

the set of finite linear combinations of elements L7L 2 with Lk E G. C is a *-

algebra, again, of course, not norm closed.

Proposition 2.2.2
(i) For C E C one has

qE (C) - sup f I w(C(x)) I di x < oc, (VI.2.13)

where the supremum is taken over all w E 8E. qE is a seminorm on C and it is

invariant under translations in space and time

gE(axC) = gE(C)-

(ii) For A E 2iS and Lk E G

1/2
qE(LiAL2) <— II A II (gE([Link])gE(L2L2))

The proposition is a consequence of (VI.1.9). Let us consider now, with w E

the sequence of positive linear forms on C
284 VI. Particles. Completeness of the Particle Picture

t+a^t) ,
Wt(C) = b(t) -1 f
w(C(t x))d t d3x,
, (VI.2.16)

where b(t) increases monotonously to infinity as t —* oo. The time averaging

may be unnecessary since, due to the intuitive reasoning of section 1, we hope
that f w(C(t, x))d3x itself converges for t —+ oo. But it helps in the subsequent
argument. Irrespective of the question of convergence the sequence {w t (C) } has
limit points as t — oo since it is uniformly bounded by qE (C) due to proposition
2.2.2. In fact the sequence of forms {w t } has weak limit points in the space C * ,
the topological dual of the space C. 1 These weak limit points are positive linear
forms over C and one can study their properties. The worst possibility is that all
limit points of (VI.2.16) are zero for any w E CSE. If this happens then the theory
has no particle content. Every state which is localized at some finite time will
then dissolve ultimately and the total probability of any detector signal at very
late times goes to zero even if all of space is paved with detectors. As mentioned
above, given a proper local formulation of the conservation laws (e.g. existence
of an energy-momentum tensor as a Wightman field associated with the local
algebras) one can eliminate this possibility.
Let WE denote the set of nonvanishing limit points in C* of (VI.2.16) as w
ranges through BE and put W = UWE. One has

Proposition 2.2.3
Let a E W. Then

(i) Q is a positive linear form on C, a weight on 2ts and

a(C*C) < qE(C*C) for a E WE.

(ii) a is translation invariant

Q(ax C) = Q(C).

(iii) For g running through the Poincaré group a9 (C) - a(agC) and a(C1 ag C2 )
are continuous functions of g.

(iv) For a E WE the Fourier transform of a(C1 ax C2 ) has support in

V+ — ,AE where AiE = {p E V+ : p° < E}.

f
(v)

j a(Cic C2) I d3x < oo. (VI.2.19)

'This follows from a slight generalization of theorem 2.2.11 of Chapter III (Alaoglu's
theorem). The natural topology on C is given by the family of seminorms {qE}. C denotes the
closure of C in this topology. Since E is arbitrary the fact that qE is only a seminorm is not
relevant.
 VI.2 The Particle Content 285

Comments. The claims (i) and (ii) are evident from the definition of a. For
the invariance of a under time translations the averaging over time in the def-
inition (VI.2.16) is used. For the proof of the continuity properties we refer
to [Buch 91b]. Concerning (iv) one observes that f e -'Pxax Cd4 x transfers an
energy-momentum p and therefore, if w E CSE,

f e -apxw(C1 ax C2 )d4x = 0 unless p + AE C V. (VI.2.20)

One has to show then that (VI.2.20) survives the passage from w to a.
Most significant is (v). The estimate (VI.2.19) relates to the intuitively
expected property (VI.1.13). It means that a is singly localized and, since it
is also stationary, it is permanently singly localized. Thus it may rightfully be
called a "single particle weight" (an improper single particle state). The proof
of (v) follows from a judicious application of proposition 2.2.2 and locality.
The next step is to use a for the GNS-construction of a representation of
2L Since a is a positive linear form over C one may follow the standard GNS-
procedure to obtain a Hilbert space rl^ regarding C as a linear space equipped
,

with the scalar product

(C2 I C1 ) = a(C2 C1 ), (VI.2.21)
dividing out the Gelfand ideal and completing in the topology provided by the
norm II IC) II= (a(C*C)) 1/2 . The facts that a is not a state but a weight and
that C is not closed in the norm topology of 2t do not affect this construction.
Furthermore, since C is a left modul of 2t we can obtain a representation of 2t
acting in 71, by
,r„,(A) I C) =1 AC). (VI.2.22)
One checks that 7r, is consistently defined by (VI.2.22) and that

II 7ra(A)II<_I1AII •
(VI.2.23)

The representation Ir e,. should not be considered as exotic. If 6E satisfies the

compactness criterion in the sense of Fredenhagen and Hertel then 7r, is locally
normal i.e. in restriction to the algebra of a finite region it is quasiequivalent to
Ira, where w is the state from which a is obtained by the limit of (VI.2.16). We
may also assume that the Hilbert space is separable. Only there is no vector
representative of a in 7-i„, since a is not a state.
The translation invariance of a implies that one can also implement the
translations by unitary operators U{°) (x) defined by

UP) (x) C) = I ax C), (VI.2.24)

and find, due to item (iv) of proposition 2.2.3 that for a E WE the spectrum of
(the generators of) UP) (x) is contained in V+ — dE. We have added the upper
index (0) to to indicate that this is not the "canonical" implementation of
the translations in Ir e . We can shift the spectrum by multiplying Uv.°) (x) with
a factor e'qx and, due to the known lower bound, achieve that the spectrum is
286 VI. Particles. Completeness of the Particle Picture

moved inside V+ choosing a sufficiently large positive time-like q. Thus we can

apply theorem 2.2.1 and obtain a unique canonical implementation, denoted by
UU (x), with the properties listed in the theorem.
We look now at the decomposition of the representation 7r, or, alternatively
put, at the decomposition of weights in W. W is a convex set of translationally
invariant weights, and a will, in general be in the interior, i.e. it will be a
mixture. Any orthogonal family of projectors in the commutant 7r„(%)' will
decompose the Hilbert space into subspaces invariant under 7r,(2t) and UU (x)
because the latter are in 7r,(2t)" due to theorem 2.2.1. Using the projectors in
a maximal Abelian subalgebra of the commutant we get a decomposition of 7r,
into irreducible representations and, correspondingly, a decomposition of a into
pure single particle weights az

a = fo• dj.t(z).• (VI.2.25)

We have used z as the label distinguishing the different pure single particle
weights arising from a finest decomposition of limit points of (VI.2.16).

Proposition 2.2.4
Let az be a pure single particle weight. Then
(i) the generated representation 7, of 2t is an irreducible, positive energy,
locally normal representation;
(ii) the canonical implementation of the translations is given by
Uz (x) C) z = eiPx I az C) z , (VI.2.26)

where the 4-momentum p E V+ is uniquely determined by z.

Comments. Claim (i) is rather evident from the above discussion. For claim (ii)
we note that, since az retains the translation invariance (VI.2.18) of a we ob-
tain an implementation Uz°) (x) just as in (VI.2.24). Furthermore Uz (x)Uz °i(x) -1
must commute with 7r z (2t) since both choices of unitaries implement the trans-
lations. Since 7rz is irreducible this quotient must be a 1-dimensional represen-
tation of the translation group i.e. it is e iPx for some 4-vector p which lies in the
closed forward cone due to the positive energy property.
One may note that for a E W and C E C
w,c(A) = (C 1 ir0.(A) I C) (VI.2.27)
is a (proper) state over 2t and that the weight a corresponds to the improper
vector Ill), (an unbounded linear form on 7l). The momentum p occurring in
(VI.2.26) is, in physical terms, the sharp momentum of the pure single particle
weight az which one may regard as an idealized eigenstate of the momentum
space operators Pu to spectral value p just like a plane wave in quantum me-
chanics.
From proposition 2.2.3, item (iii), one conludes that if a is a pure single
particle weight to momentum p then aA is one to momentum Ap. Two possibili-
ties exist then. Either the representations 7r, and 7raA are equivalent or they are
 VI.2 The Particle Content 287

disjoint. In the first case one has an ordinary particle and can form wave packets
by coherent superposition f (p(p) I p)d1.c(p) where dp(p) is the Lorentz invariant
measure concentrated on the mass hyperboloid, cp is a wave function and I p)
denotes the improper GNS-vector II). This wave packet is then a normalizable
state vector in 1i. In the second case we have an infraparticle. The velocity p/p°
is a superselection quantity. No normalizable single particle state of this type
exists. The Lorentz symmetry is spontaneously broken.
Special consideration must be given to the Lorentz transformations which
leave p unchanged, the "little group" or "stability group" C p of p. For p2 > 0
this group is isomorphic to the 3-dimensional rotation group. Then there are
(almost local) elements C E C which are invariant under Cr . Under reasonable
assumptions (see below) it follows that this symmetry is not spontaneously
broken i.e. there is a projective unitary representation implementing C p in f..
This allows an adaptation of Wigner's arguments described in Chapter I, section
3, showing that a pure, massive single particle weight may be chosen to have
sharp integer or half integer value of the spin. The same argument may be
applied in the case of mass zero to the 1-parametric subgroup of g r given by
the rotations around the direction of the spatial momentum. This leads to a
sharp helicity of such pure, massless single particle weights. However it can
no longer be inferred that this helicity is restricted to integer or half integer
values. Summing up: The particle content of the theory is given by W, the non
vanishing limit points of (VI.2.16) as t oo for all states of bounded energy.
To each extremal element of W (pure single particle weight) there is a sharp
momentum p E V + and we may restrict attention to those having a definite spin
(resp. helicity). We may consider the Lorentz invariant set {a fl} arising from a
pure single particle weight a by Lorentz transformations as corresponding to a
particle type. The Lorentz symmetry may be spontaneously broken in ir e . Then
the above set contains the incoherent pure weights of an infraparticle.
One would like to show, using again some version of the compactness or
nuclearity assumption, that there is only a finite number of distinct particle
types for fixed mass. This is the "reasonable assumption" mentioned above. A
proof of this does not yet exist.

VI.2.3 Further Remarks on the Particle Picture

and Its Completeness

The strategy described at the end of subsection 1.2 and the part of it carried
through to date (section 2.2) give the most clear cut definition of what we mean
by the particle content of the theory from the point of view of observables,
taking into account all comp lications arising from superselection rules, infrared
problems, long range correlations. It will still require some hard work to close
the remaining gaps in this program but there is little doubt that this can be
achieved and will result in a good understanding of how various structural
properties of the net {2(0)1 relate to different aspects of the particle picture.
288 VI. Particles. Completeness of the Particle Picture

Leaving aside, for the moment, the question of convergence of (VI.1.15) or

(VI.2.16) the simple and most important qualitative conclusion is the following.
If one considers the partial state in a region KR(x) (the diamond with radius
R and center x) then, for any state w E CSE, as x moves to time-like infinity
x = (t, vt) with y (< 1 the state will look with overwhelming probability
like the vacuum state in this region, jJ (w — coo) ^ (xR(x)) 0 as I t o0
Furthermore the deviation from the vacuum, registered by the elements C(x)
with C E C, will, at least for almost all y with I y < 1, decrease like I t
For large times ( t1 3 w(C(x)) will look like (the restriction of) a single particle
weight in the region.
The convergence of

(VI.2.28)
t3
J w (C(t, vt)) h(v)d 321
when the support of h is in the interior of the ball ( y (< 1 concerns the
rescattering problem: can a bounded number of particles in unbounded space
rescatter infinitely often or will there emerge ultimately a configuration where
no further encounter is possible? This problem has been studied extensively in
non-relativistic quantum mechanics (see. e.g. [Enss 83], [Mour 79], [Sigal 1988],
[Graf 90] and references given there). In purely massive quantum field theory
satisfying the Wightman axioms Bros and Iagolnitzer [Bros 88], [Iag 87] have
studied its relation to the analytic structure of n-point functions.
Let us try to describe the main intuitive ingredient of this problem. If we
have a state which at time t describes n far separated particles of definite mass
localized in respective neighborhoods of the points xk then the probability for
a collision at time t' around a space point y will be determined by the momen-
tum space probabilities in each of the single particle states around the values pk
which are calculated classically from the needed velocities v k = (y — xk )/ (t' t)
—

and the masses of the particles. Only if there is a significant probability of

these momenta for at least two particles we can expect a collision at (t', y).
Of course this argument neglects the interparticle forces but, if I xk — x i I and
t' t are large and the forces not of too long range then the modification is
—

not significant. In a collision process with center (t', y) one has conservation
of energy-momentum. This limits the possible spatial configurations of the re-
action products at subsequent times. Extrapolating from this argument, once
we know that a state w of bounded energy is n-fold localized at time t with
large distances between the individual localization centers then we can hope
to decompose w approximately into a convex combination (mixture) of compo-
nents co), where each a corresponds to a subsequent history (or fate) during a
(large) time interval [t, t +71, each history being characterized by a finite set
of space-time points yz marking the centers of events (collision centers during
the interval) together with causal ties between events (a tie corresponding to
a particle with momentum geometrically determined by the space-time points
it connects and its mass). In other words, a history in the time interval can be
pictured like a Feynman graph with the vertices marking points in space-time,
the lines momenta, with momentum conservation at every vertex but with the
 VI.3 The Physical State Space of Quantum Electrodynamics 289

further restriction that the mass value of each momentum must be that of a
stable physical particle. This reduces the problem of convergence to a classical
one (a billard ball problem in which the balls may fracture or recombine) and
to which the method of Hunziker [Hunz 68] can be applied, showing that even
as 7 —p o0 only a finite number of events are possible. In a purely massive
quantum field theory this picture is supported by Symanzik's structure anal-
ysis mentioned in section 2.5 of Chapter II. Within the scope of renormalized
perturbation theory in quantum electrodynamics the problem of "asymptotic
completeness" (completeness of the particle picture) has recently been studied
by Steinmann [Steinm 91] but the comparison of his results with the geometric
picture remains to be done.

VI.3 The Physical State Space

of Quantum Electrodynamics

One consequence of the possible appearance of infrared clouds is that in

quantum electrodynamics there are innumerably many sectors satisfying the
Borchers selection criterion even for states with the same charge. It is then con-
venient to make a coarser distinction and combine all sectors which coincide in
their information about the content of outgoing (or incoming) massive particles
in their respective folia of states into one class. Since electrically charged parti-
cles are massive such a class fixes in particular the electric charge and for this
reason it was called a charge class in [Buch 82b]. We shall keep this terminology
here though these classes still give a much finer distinction than that provided
by the charge quantum numbers.
As in Chapter IV we consider only states with vanishing energy density
at space-like infinity. This is guaranteed by restricting attention to "positive
energy representations" in the sense of subsection 2.2. Let (V+ + a) denote
the open positive light cone with apex a in position space. A massive particle
will ultimately (as t ^^ oo) enter inside this cone, no matter how the apex
—

a is chosen since it moves slower than light. Similarly, an incoming massive

particle can be detected in any cone V - + a. Thus, if 7r is an arbitrary positive
energy representation of the net of observable algebras, all information about
outgoing massive particles in its folium of states can be obtained from the
part 7r(2t(V+ + a))" and the information about incoming massive particles from
7r(2t(V - + a))" for any a. Let us denote the centers of these algebras by 3a+, 3a— ,
respectively and the center of the total algebra by 3, Thus

3Q = 7r(2i(V± + a))" n 7r(2.(V ± + a))'; 3 = 7r(21)" n 7r(20'. (VI.3.1)

One finds

Theorem 3.1 [Buch 82b]

(i) Every element Z E 3¢ is invariant under translations

290 VI. Particles. Completeness of the Particle Picture

Z(x) = U(x)ZU(x) —i = Z. (VI.3.2)

(ii) 3Q is independent of a. We write

31- = 3+. (VI.3.3)

(iii)
3+ c 3. (VI.3.4)
The same statements hold for the elements of 3;.

Proof. Part (ii) follows immediately from (i), part (iii) almost immediately:
For A E 240), 0 C V+ + a and Z E 3Q we get from (i)

[Z(x), 7r(A)] = [Z, 7r(A)] = O.

Thus, for any x, also [Z,71- (21(0 + x))] = 0 and, since the algebras 7r(240 + x))
generate 7(21), Z commutes with 7r(2î). So Z E 3.
It remains to prove part (i). Let n be any positive time-like 4-vector. We
show that U(tn)3â U(tn) -i commutes with 3,4- for any real t. For t positive
(V+ + a + tn) C (V+ + a). For negative times (V+ + a + tn) D (V+ + a) which
implies U(tn)7r(2î(V++a))'U(tn)-i C 7r(24V++aV. So by the definition of 3+,
we have the claimed commutativity for both positive and negative t. Replacing
a by a + t'n we see that the algebra

= V U(tn)3â U(tn) -i

is Abelian. is also stable under the 1-parametric translation group U(tn)

which has a generator with positive spectrum. So theorem 2.2.1 can be applied
to , telling us that apart from U(tn) there is another implementation of these
translations by unitaries V(tn) which belong to e. Since E is Abelian each
element of this algebra is invariant under translations by tn and the same holds
a forteriori for the elements of 3a+ . By choosing different directions of n we can
generate the whole translation group and thus verily the claim (i).
The same argument applies, of course, if we replace the future directed cones
V+ by past directed cones V. So the theorem holds also if 3+ is replaced by
3- . ❑

Given a general positive energy representation 7r we may decompose it

with respect to the part of the center 3+ V 3- . This leads to representations
which are primary on each 21(V+ + a) and on each 2l(V - + a) and in which
the translations are still implementable (with positive energy). Since we expect
that 3 is larger than 3+ V 3 - such a representation may still be decomposed
into primary representations of the total algebra. However, under reasonable
assumptions this further decomposition will lead to representations whose re-
strictions to 21(V+ + a) and to 21(V - + a) are equivalent and therefore do not
differ in any superselection quantum number relating to massive particles. For
 VL3 The Physical State Space of Quantum Electrodynamics 291

this reasons the term "charge class" was introduced for the set of representa-
tions contained in a (positive energy) representation in which 3+ v3 - is trivial.

Asymptotic Photon Fields. What distinguishes different irreducible repre-

sentations of the total algebra within one charge class? There are the different
infrared clouds of incoming or outgoing photons. To describe them one would
like to define asymptotic creation and annihilation operators for photons. They
have been constructed at least for certain classes of representations by a suitable
modification of the LSZ-procedure described in subsection 3.3 of Chapter II. For
the representation generated from the vacuum state this is treated in [Buch 774
The photon is no infraparticle; therefore one will have a subspace 'H l belonging
to the sharp mass zero describing the states of a single photon in the vacuum
sector. Picking a smooth, but otherwise arbitrary element A of the observable
algebra 21 one considers angular averages of translates of A in positive light-like
directions
At = -2t f (â0A) (t, te) sin /, d19dço (VI.3.5)

where e is a unit vector in 3-space with polar angles '9, ça. Applying this to the
vacuum state vector Q one gets

iro(At) I .t) =1 P I- 1 (eit(H_) _ eit(H+IPi)) HA 10). (VI.3.6)

Here P denotes the 3-momentum, H the Hamiltonian. Averaging At over a time

interval 0(t) which grows slowly to infinity as t -+ oo (such that t-1 0(t) -> 0)
we put
c+A(t) ,
At = 13(t)-1 f A t, dt , (VI.3.7)
t
and find that 7ro(.4 t ) I Q) converges strongly

7ro(At ) 1 Sl) - ^^ Pgro(A) I (2), (VI.3.8)

where P1 is the projector on the 1-photon subspace V. From the space-time

support properties ob A t and the Reeh-Schlieder theorem it follows then that
t ) converges strongly on a dense domain in Î-l. It defines an operator 7ro(A
Aout = t ^^ro (.4t). (VI. 3. 9 )
1,
The energy-momentum transfer of Rout is confined to light-like vectors. One can
use these operators to construct a Wightman field F,°,ut which behaves in every
respect like a fr ee Maxwell field. This field defines a net °ut (0) of local algebras.
Of course, though F,°Ût(x) is formally a local field it should not be interpreted
as an (idealized) local observable at the point x. It has this significance only for
asymptotic observations. (See fig. 11.3.1 in Chapter II). Still there remain some
geometric relationships between the net .Foot and the net of local observables.
By construction we have

Plut(V+) C 7oMV + ))". (VI.3.10)

292 VI. Particles. Completeness of the Particle Picture

Furthermore, if x E V - and n is a positive light-like vector then x+ to becomes

space-like to any point in V+ for sufficiently large t. Since Fi°,,t (x) propagates
only in light-like directions ("Huygen's principle") we have
.Fout(V-) C 7ro(2L(V+))'. (VI.3.11)
The latter relation can be strengthened to
.Pout (V - ) = 7ro (21(V+))'. (VI.3.12)
So one has
(70(21(V +)) v Fout(V_))" = 93(n). (VI.3.13)
The missing information in 7ro (Q1(V+))" is supplied by the radiation field .Fout
For the detailed arguments and proofs we refer to [Buch 82b] where it is also
shown how to construct different representations of 2t in each charge class,
called "infrared minimal" because their folia contain a state which coincides
with the vacuum in restriction to a chosen space-like cone (compare section
3 of Chapter IV). For these one can construct radiation fields F in and Fo"t
Itshould,wevrnbifdthaesnrmilchages
exhaust all possibilities or even that they contain all physical states which might
be of interest. Kraus, Polley and Reents [Kraus 77] have described other states,
obtained from an "infravacuum", which have some advantages for the treatment
of the infrared problem arising in collision theory of charged particles. They
contain a sufficiently chaotic background radiation field so that the additional
infrared photons, produced in the bremsstrahlung can be considered as a small
perturbation which does not change the equivalence class of the representation.
Let us now consider the asymptotic observables for charged particles. Ac-
cording to the discussion in the last section they should reduce to the algebra
generated by the momenta and spins of these particles. If we disregard spin
they form an Abelian algebra in contrast to the asymptotic observables for un-
charged particles which are generated by the non-Abelian algebra of free fields
out associated to the various particle types (compare [Ft-6h 79b]). The values
of the particle momenta are not fixed by the label of the class, but a change
in asymptotic momentum must be accompanied by a change in the radiation
field.
Finally, let us look once more at the implementation U(x) of the translations
in a representation of the observable algebra in which ,Put can be defined. (e.g.
an infrared minimal representation). Since pout is stable under translations
there exist, by theorem 2.2.1, unitary operators V o"t(x) E .out" implementing
the translation group on Pou t . We may split U(x) into
U(x) = Vout(x )U4t( x ) , (VI.3.14)
with UM t E .Pout ' and interpret Ur as the translations acting only on the
massive outgoing particles. Then one finds that in the spectrum of Pk' the
mass values of all massive particles (charged or uncharged) appear as discrete
eigenvalues (see [Frdh 79b], [Buch 82b]) whereas in the spectrum of U(x) the
charged particle masses appear only as lower boundaries.
VII. Principles and Lessons
of Quantum Physics. A Review
of Interpretations, Mathematical Formalism,
and Perspectives

Introduction

In the previous chapters we used the standard language of quantum theory.

We have extensively used the terms "observable" in the sense of Dirac's book
of 1930 and "state" in the sense of von Neumann's book of 1932. This lan-
guage implies an essential resignation as compared to the "classical" ideal of
the scope of physics prevailing before the advent of quantum mechanics. Emi-
nent physicists, among them Planck, Einstein, and Schriidinger did not accept
this resignation as unavoidable. Over 60 years have passed since the great de-
bates. We have witnessed fundamental new discoveries, an enormous growth of
knowledge, and the development of the theory from non-relativistic quantum
mechanics to relativistic quantum field theories with applications ranging from
laboratory experiments in high energy physics to cosmology. Yet, with a few
minor modifications and changes in emphasis the standard language remained
viable and no alternative of comparable usefulness has been created. Neverthe-
less some measure of dissatisfaction or at least an uneasiness that something
is missing in the standard picture persisted throughout the years and is even
more acutely felt today. The large amount of literature devoted to the inter-
pretation of quantum mechanics bears witness to this. It therefore seems useful
to review the main arguments advanced during this long-lasting dispute, try-
ing to sharpen the points at issue, and to assess the strengths and weaknesses
of various points. This will be attempted in section VII.1. Another matter is
the specific mathematical structure of quantum physics. Much work has been
devoted to understanding the rôle of Hilbert space and complex algebras from
some deeper-lying operational principles. A brief sketch of such endeavors will
be given in section VII.2. Finally, in section VII.3, I want to present a point
of view which I call the "evolutionary picture". It replaces "measurement re-
sult" by the more general notion of "event" and may be regarded either as
a different idealization from the one implied by the Bohr—Heisenberg cut or,
294 VII. Principles and Lessons of Quantum Physics

more ambitiously, as part of the conceptual structure of a wider theory to be

developed.

The road to disharmony is paved by

efforts to clarify misunderstandings

VII.1 The Copenhagen Spirit. Criticisms, Elaborations

We must bear in mind that the mathematical formalism of quantum mechanics

and its first remarkable successes in physical applications existed prior to the
philosophical discussions concerning the interpretation and proper understand-
ing of the lessons. The point of view resulting from the intense struggle with
this problem in the circle of Niels Bohr was called by Heisenberg the "Copen-
hagen Spirit", presumably to contrast it with the position held by Einstein. The
first and most obvious fundamental lesson appeared to be the indeterminacy of
the laws of nature (originally called acausal behavior of atomic objects). This
was seen as an inescapable consequence of phenomena like radioactive decay,
quantum jumps between stationary states of an atom, and diffraction experi-
ments. In the last of these an individual electron or photon produces a sharply
localized effect (e.g., a dot on a photographic plate) whereas the distribution
of these effects shows an interference pattern that depends on the experimental
arrangement in a way which cannot be reconciled with the assumption that each
particle follows a specific trajectory in space, but which is easily described in
a wave picture. In fact, it was this contrast between localization of effects and
spreading of waves which led Max Born to the probability interpretation of the
Schroedinger wave function. Let us define indeterminacy as the claim that the
optimal attainable knowledge of the past does not enable us to predict future
behavior with certainty. This leaves open the question of whether God knows
the future, a question that lies outside the scope of physics. After the formula-
tion of Heisenberg's uncertainty relations, the closer analysis of indeterminacy
by the discussion of many thought experiments led to a much deeper negative
statement: it is not possible to assume that an electron has, at a particular
instant of time, any position in space; in other words, the concept of position
at a given time is not a meaningful attribute of the electron. Rather, "position"
is an attribute of the interaction between the electron and a suitable detection
device. More generally, a phenomenon which we observe does not reveal a prop-
erty of the "atomic object". The phenomenon is created in the act of detection.
This fundamental point is somewhat veiled in the standard language. If we say
that some "observable" of a quantum object is measured this suggests that
there is some corresponding property of the object which may have different
numerical values in the individual case and the purpose of the experiment is to
determine this value. By contrast Bohr stresses that the measurement result is
a property of the compound system of measuring device plus object and that
 VIL1 The Copenhagen Spirit. Criticisms, Elaborations 295

the full description of the experimental arrangement is an essential part of the

definition of the phenomenon. Thus, Max Born's probability for the "position of
a particle" within some region of space should be understood as the probability
of an effect if a detector is placed in this region.

Niels Bohr's Epistemological Considerations. "We must be able to tell our

friends what we have done and what we have learned". This key sentence of Bohr
alludes to several facts. First, our knowledge about the physical world derives
from observation; in physics observations usually involve planned experiments.
The description of such an experiment must be given in unequivocal terms; the
stated conditions and the results must be reproducible by others. Bohr concludes
that, no matter how sophisticated and abstract the technical language of physics
may become, we may not transcend the language of classical physics in the
description of the experiment. Why "classical"? Planck's quantum of action
introduces a discrete element into physics which implies some discontinuity. In
particular, if we want to control or verify the specified conditions this will involve
some interference which will change the conditions slightly in an uncontrollable
way due to the indeterminacy expressed in the uncertainty relations. Therefore
Bohr argues that the acquisition of knowledge about an atomic object demands
a dichotomy. On the one hand there is the experimental arrangement which must
be described in classical terms. On the other, there is the atomic object. The
observed phenomena depend on both. They give only indirect information about
the object and we cannot speak about the behavior of the object independently
of the means of observation.
Generalizing this dichotomy Heisenberg argued that we cannot avoid intro-
ducing a (notional) cut between the "physical system" we want to study and
the means of observation. The cut may be shifted but not eliminated. This pre-
cludes in particular the consideration of the whole universe as a physical system
in the sense of the formalism of quantum theory (see the remarks on page 3
of this book). Bohr asks us to remember "that in the drama of nature we are
both actors and spectators". This fitted well with Bohr's deepest conviction,
the principle of complementarity: there are many aspects of nature. We may
study one by a particular experimental arrangement. This precludes the use of
another arrangement which would be needed to study another aspect. The more
precisely we want to grasp one aspect the more the other escapes us. There-
fore it is folly to try to catch reality by one complete and precise mathematical
counterfeit. I remember a conversation in 1953 in which Bohr expressed his
dismay at my juvenile criticism saying: "Of course you can change the math-
ematics. But this will not affect the essential lessons we learned". Still, ,T
accept the general lesson there remains the question of how to proceed from
these qualitative considerations to a quantitative prediction for a specific ex-
periment. How do we translate the description of an experimental arrangement
into mathematical symbols? For this Bohr relied on the correspondence princi-
ple which establishes a close connection between models in classical mechanics
and quantum mechanics. With the enormous change in the scope of problems
296 VII. Principles and Lessons of Quantum Physics

to which quantum physics addresses itself today, this question must be seriously
reconsidered. Perhaps the word "classical" has been overemphasized both in the
epistemological analysis and in the reliance on the correspondence principle. We
shall return to this below.

Criticisms and Elaborations. One of the points of Einstein's criticism (not

the most relevant one) concerned indeterminacy. He was deeply convinced that
the fundamental laws of nature cannot be probabilistic ("God does not play at
dice"). One cannot argue about convictions or the ways of God. The opposite
conviction, namely that strictly deterministic laws, reducing the world to clock-
work, would be a nightmare and cannot possibly be the last word, appears to
me more attractive. Of course, metaphysical beliefs — one way or another — do
not provide a basis for discussions of physical theory. In judging the merits of
attempts to overcome the apparent indeterminacy within the realm of physics
by the assumption of hidden degrees of freedom and quantum forces, the cri-
teria must be either the ability of the proposal to suggest finer experiments
or its convincing naturalness, explaining a wide range of phenomena with few
assumptions. The material addressed must at least include spin correlations,
the exclusion principle, particle transmutations, and, ultimately, also the more
subtle effects treated in quantum field theory. The existing proposals seem to
be far from meeting these demands and do not look promising to me. Therefore
I shall not discuss this criticism further but accept indeterminacy as a feature
of the laws of nature.
The deeper problem which disturbed Einstein concerns the question of "re-
ality". In past centuries physicists proceeded from the assumption of a real
outside world which exists separate from our consciousness; the task of physics
was precisely to describe this world, to discover the "laws of nature" which
governed it. Quite a different picture was drawn by philosophers for whom the
mind—body problem was a central issue. There was the recognition of the pri-
macy of consciousness with conclusions ranging from the extreme idealism of
Berkeley to Kant's distinction between "things as such" about which we can
know nothing and "things of appearance" which are partly shaped by a priori
given faculties of our mind.
Prior to the advent of quantum mechanics, the ideas of positivism, demand-
ing the elimination of all concepts in science not directly related to observations,
had exerted some influence on the attitude of physicists. It was natural that
positivists should consider the Copenhagen Spirit as a corroboration of their
doctrine. Niels Bohr did not share this view. On the one hand the abstract
mathematics of quantum mechanics involves concepts far removed from direct
perception. But one should also note an interesting aspect of Bohr's view on
complementarily: he saw it as a generalization of the relativity principle, in the
sense that different observers can catch different aspects of reality, depending
now on their choice rather than their position; the totality of these aspects which
may be "symbolically" united by the mathematical formalism constitutes a full
 VIII. The Copenhagen Spirit. Criticisms, Elaborations 297

picture (of nature). An analogy might be the use of charts in the description of
a manifold.
The discussions on the interpretation of quantum mechanics raised doubts
about the separability of the laws of physics from processes in the human mind
and led to a broad spectrum of opinions.
It should be understood that our discussion here cannot concern the mer-
its of different philosophical positions but only the question of whether the
discoveries in the atomic and subatomic regime force us to reject the assump-
tion that physics is dealing with a real outside world whose laws can be stated
without reference to human consciousness. In other words, does our inability
in "ascribing conventional physical attributes to atomic objects" and the "im-
possibility of any sharp separation between the behavior of atomic objects and
the interaction with the measuring instruments which serve to define the con-
ditions under which the phenomena appear" [Bohr 49] imply that we have to
include into physical considerations mental aspects like consciousness, intelli-
gence in the planning of an experiment, and the free will in deciding on its
execution? I think that this can be answered by a clear no. l The raw mate-
rial of quantum physics which the theory tries to order consists of facts which
can be documented, such as dots on a photographic plate. The (classically de-
scribed) experimental arrangement is also a "physical system", not endowed
with consciousness; the total experiment can be automatically registered and
documented as a computer print-out. Of course it is logically possible to say
that only the last step in the observation procedure, namely the studying of
the print-out by a human observer with subsequent consciousness establishes a
fact. But since the quantum mechanical uncertainties are practically negligible
in this step such a position cannot help in the solution of the riddles with which
the fathers of quantum mechanics were confronted. It may be objected that
in the formalism of quantum mechanics the notion of "fact" does not appear
and is replaced by "measurement result". Indeed one may claim that there are
no precisely definable facts, no absolutely reliable documents. But this is not
remedied by replacing the word "fact" by "measurement result" and shifting
the burden to the judgment of consciousness. The unreliability of the memory
of impressions far exceeds that of a registered document. Rather, to the extent
to which the objection is relevant, it may indicate that a physical theory can
never reach an absolutely precise picture of nature and that at each step of the
'In his "Remarks on the mind-body question" [Wig 63] Wigner writes: "If one formulates
the laws of quantum mechanics in terms of probabilities of impressions, these are ipso facto
the primary concepts with which one deals" and: "The principal argument is that thought
processes and consciousness are the primary concepts, that our knowledge of the external
world is the content of our consciousness and that consciousness therefore cannot be denied".
To avoid misunderstanding my above "no" I should stress that I fully agree with the second
sentence and do not dispute the logic of the first. What I want to assert is that in resolving
the problems implied by the inabilities and impossibilities emphasized by Bohr it is neither
necessary nor even helpful to include consciousness in the description of quantum physical
laws.
298 VII. Principles and Lessons of Quantum Physics

development certain idealizations or, if one wishes, asymptotic notions have to

be used.
Stepping down from lofty generalities to a practical example, let us con-
sider a typical experiment in high energy physics. It yields a computer record
of many individual runs, which is subsequently studied by various teams of
experts. Hopefully they will agree that one such document may be described
in the words: A positron and an electron in the storage ring collided and pro-
duced a multitude of particles, concentrated mainly in two narrow jets (event
number 1). These jets contained a B, B-pair, identified by subsequent decays
B --* D ---r K —* ... (events 2, 3, ...) and corresponding decays of B. If we trust
the experts we must accept, as raw material for the theory, that not only the
computer records produced by electric discharges in macroscopic detectors but
also the mentioned subatomic events are facts. The account is told in a way
which can be understood by our friends. The appearance of words unknown
some decades ago and not in the vocabulary of classical physics does not af-
fect the essence of Bohr's analysis but may be a warning against taking some
formulations too literally. In conclusion I believe that the phenomena to which
quantum theory applies may be appropriately described in an "as if" realism,
where facts, whether observed or not, are assumed to exist and constitute the
cornerstones of the theory. It should not be taken for granted that the notion of
"fact" is synonymous with "macroscopic change". The definition of what consti-
tutes a fact may involve idealizations depending on the regime considered and
the precision demanded, and it may change with the development of physical
knowledge. But I do not see any essential conflict between quantum physics
and that part of Einstein's creed which he poetically expressed in the words:
"Out yonder there was this huge world, which exists independently of us human
beings and which stands before us like a great eternal riddle, at least partially
accessible to our inspection."
We must now come back to the important issue of our inability to "ascribe
conventional physical attributes to atomic objects". First we must ask what
we mean by an atomic object or, more generally, by a "physical system" . The
relevance of this question is brought into sharp focus in quantum field theory.
From previous chapters of this book it is evidently not obvious how to achieve a
division of the world into parts to which one can assign individuality. We have
moved far away from Maxwell's ode on atoms: "Though in the course of ages
catastrophes have occurred ... the foundation stones of the material universe
remain unbroken and unworn. They continue this day as they were created —
perfect in number and measure and weight.i 2 Instead we used a division ac-
cording to regions in space-time. This leads in general to open systems. Under
special circumstances we can come from there to the materially defined systems
of quantum mechanics, claiming for instance that in some large region of space-
time we have precisely an electron and a proton whose ties to the rest of the
world may be neglected. One of the essential elements in singling out such a ma-
terial system and assigning to it an individual, independent (at least temporary)
2 Quoted from the book "Subtle is the Lord" by A. Pais.
 VIL1 The Copenhagen Spirit. Criticisms, Elaborations 299

existence is its isolation, i.e., the requirement that in a large neighborhood we

have a vacuum-like situation. However this is not always enough. Suppose that
in the course of time the electron and the proton, originally close together, be-
come widely separated, still remaining isolated. Then we might want to consider
them as separate individuals and wish to assign to each of them some notion
of "state" (some attribute). The formalism of quantum mechanics, supported
by experiments, tells us that this is not always possible. There may be per-
sistent correlations of a non-classical character (in the experiments concerned
with the EPR-phenomenon they are spin-correlations). It should be noted that
persistent correlations occur also in classical mechanics; there they do not put
the assumption of individual existence of the partners in doubt. The conserva-
tion of angular momentum in the classical mechanics of a two-particle system
implies that if we know the 3-vector of total angular momentum at one time
then the measurement of the angular momentum of one particle much later
tells us exactly what the angular momentum of the other particle must be, no
matter how far apart the particles are then. In quantum theory, however, the
angular momentum is not a numerical 3-vector. The two-particle wave function
belonging to angular momentum zero is not a product of single particle wave
functions but a linear combination of such — which, moreover, can be chosen in
different ways. Thus even if we envisage "unconventional" attributes for each
particle, such as wave functions, this will not suffice to describe the situation.
In other words, in spite of the large separation of the particles they cannot be
regarded as individuals in any realistic sense as long as there remain correla-
tions of this kind, conserving a common property due to some past encounter.
This indicates that in the endeavor to divide the world into parts, the "facts"
(e.g., events marking approximately some space-time point) are more basic than
"quantum objects". The latter notion corresponds to causal links between the
events. We shall return to this in section VII.3.
Can quantum theory stand on its own feet or does it need some props in-
volving a knowledge of classical theory? There are several facets to this question.
For instance: given the formalism of quantum electrodynamics (supplemented
by nuclear physics) together with the interpretation of the basic symbols in
terms of local observables and the general rules relating to states, observables,
and probabilities, can we derive the existence of states describing the behavior
of billiard balls, electric power generators, etc.? Can we retrieve the body of
knowledge collected in the text books of classical physics and show that it re-
sults as an approximation from the given quantum theory? This is a tough task
but the strategy is clear. One has to go over to a "coarse-grained description"
defining suitable "collective coordinates" which can be related to classical quan-
tities. Then one must show that for these quantities one obtains an adequate
approximation to the previously known deterministic laws. As an instructive
example in quantum mechanics Hepp and Lieb [Hepp 73] have treated the c ase
of a pendulum regarded as an essentially rigid body consisting of many atoms.
There is one collective coordinate, the angle B between the pendulum and the
vertical. If one starts from a quantum state in which this angle and its angular
300 VIL Principles and Lessons of Quantum Physics

velocity are prescribed (which is approximately possible because of the large

mass) and the internal state of the solid is the ground state, then during the
motion there will be a transfer of energy from the macroscopic energy of the
collective motion to the internal degrees of freedom and it may be shown that,
due to the large number of atoms, the resulting time dependence of 8 follows the
deterministic equations of motion of a damped harmonic oscillator. Moreover,
if one starts with a state which is the coherent superposition of states corre-
sponding to different initial values of 8, then the phase relation is very soon
lost by the interaction with the internal degrees of freedom. More generally, for
"almost" any initial state the so-called reduced density matrix for the motion of
B (Le., the description disregarding the situation of individual atoms) becomes
diagonal in the 0-representation after a very short time. So, with respect to 8
any ensemble can be considered practically as a classical distribution of har-
monic oscillators with different initial conditions. There remain no interference
effects. This decoherence of collective degrees of freedom appears to be a rather
general feature. It is essential in establishing that the quantum theory allows
us to recover approximately classical behavior under suitable circumstances. A
more elaborate discussion of the derivation of classical properties from quantum
theory and of decoherence may be found in the book by Omnès [Omnès 1994]
where references are also given to other work.
What can one learn from this concerning the problem of measurement in
quantum theory? A measuring instrument for "quantum objects" is a very spe-
cial system. Typically it has several stable or metastable states, close together
in energy but differing strongly in their macroscopic appearance. Treating the
measurement as a collision process between the object and such an instrument
we may distinguish three different types of degrees of freedom: firstly, those
relating to the object, secondly, the macroscopic (collective) variables of the
instrument and, finally its internal degrees of freedom. One may assume that
initially one has a product state gi ®g2 ®g3 where o 3 could be the ground state
or equilibrium state at some temperature of the internal degrees of freedom,
02 is adequately characterized by some numerical values of the macroscopic
variables which we denote summarily by ao. Believing in the effectiveness of
decoherence the final state will be a m ix ture of states, each of which belongs
to some value a of the macroscopic variables (the measurement results). Thus
it defines a probability p(a; Q i ) for the occurrence of the result a which de-
pends, as indicated, on the initial state of the object. If the object survives we
shall furthermore have a correlation between a and the final state of the object,
which, if we are not interested in the final state of the internal motion, can
be described by a density matrix pi depending in general on the measurement
result a and the initial state of the object p i . Up to here we have only used the
arguments described above. Now, for the ideal maximal measurements axioma-
tized by Dirac and von Neumann we should find that gi does not depend on p 1 .
In its dependence on a it should assign to each a a pure state from some family
determined by the instrument. This will be the case if there exists a special
orthogonal family of initial states I'„,) for which the interaction process with
 VII.1 The Copenhagen Spirit. Criticisms, Elaborations 301

the instrument leads to a strict (deterministic) correlation between n and the

measurement result a. In the mathematical formalism this family corresponds
to the eigenstates of the observable A representing the instrument. We refer here
to the ideal case where the measuring device is assumed to define a self-adjoint
operator acting in the Hilbert space of the pure states of the object. Denoting
by an the result attached with certainty to the initial state kI n,) one obtains
for a general pure initial state 0 = > coP n the probability p(an ; ) = IcnI2 and
also the prediction that the final state of the object which is coupled to the
result an is just On . For a realistic instrument we cannot expect such a sharp
correspondence between a and a specific initial and final state of the object.
But there are good instruments for which this is at least approximately true.
For them the final state of the object is sharply fixed by the measuring re-
sult a whereas the information from the initial state is largely forgotten in the
subensemble associated with a fixed value of a. It enters only in the fraction of
systems in this subensemble which is given by the probability p(a; V)). Focusing
future attention on one such subensemble, say on the one characterized by a n ,
and ignoring the rest, we have an apparently discontinuous change of the state
of the system from to O n which is often called the "reduction" or "collapse"
of the wave function. It is not due to any physical effect but to our method
of book keeping, in going over from the original ensemble to the conditional
probabilities in a subensemble associated with a specific measurement result.
Of course the realization of a measurement result as a fact and the possibility
of using it for the characterization of a subensemble of objects for subsequent
study is not fully reconcilable with the quantum theoretic description of the
dynamics of the total system of object plus apparatus. In the above discussion
it is attributed to the decoherence effect and to the additional assumption that
in each individual case one value of a must be realized, i.e., it becomes a fact.
We may ask whether there is a generic reason for the special feature of
"good instruments". Looking at typical detectors like a photographic plate or a
bubble chamber we note that the macroscopic change consists in a local change
in some small part of it. The distinction between measurement results is given
by the position at which this change occurs. Then we can understand the exis-
tence of a preferred family of initial states of the object. A wave packet which
is sharply localized at the time of arrival can produce an effect only in the
neighborhood of the place where it meets the detector; the final state of the
object after detection will again be a reasonably sharply localized wave packet
emanating from the region where the effect was produced. Perhaps we may take
this as an indication that in the measurement problem we should distinguish
between the primary phenomenon due to some irreversible event to which we
can assign a reasonably well-defined region in space-time and the accessories in
the experimental arrangement (like external fields, shields ...) which serve to
define the possibility for such events.
This has a bearing on several other questions. First it suggests that not ev-
ery self-adjoint operator can be realized by a measuring instrument. Secondly,
though position and momentum variables appear on an equal footing in the for-
302 VII. Principles and Lessons of Quantum Physics

malism of quantum mechanics, there is a hierarchy with respect to measurement

procedure. A momentum measurement ultimately also demands the detection
of localized effects, irrespective of whether we use the relation of momentum to
velocity, or to wave number, or appeal to its conservation laws. So the primary
phenomena are again localized events. The precision depends on the control of
the situation in a large region of space-time.
Finally we must come back to the correspondence principle. If we mean by
"correspondence" merely that in the realm of electrodynamics and mechanics
we can retrieve the laws of the classical theory as an approximation valid for
states with high quantum numbers then we do not leave the topic of the previous
discussion. There is, however, also the amazing parallelism of the mathemati-
cal structure of the theories (Poisson brackets corresponding to commutators,
conservation of the form of equations of motion). This has led to the notion
of "quantization" of a classical model as a well-defined procedure, regarded
as forming in some way an indispensable background of the quantum theory.
Related to this there is the problem mentioned in connection with the episte-
mological considerations of Bohr: How do we achieve the translation between
the description of an experimental arrangement and the mathematical symbols
of quantum theory? Is it necessary for this purpose to know a "corresponding"
classical model theory? Looking at modern models developed in high energy
physics such as quantum chromodynamics, it is true that they do have a corre-
sponding classical counterpart, namely the classical field theory with the same
form of the Lagrangean. But this is not of much help for the translation be-
tween the description of experimental hardware and the mathematical symbols
of the theory and it is questionable whether such a formal correspondence is
a relevant aspect in a fully developed quantum theory. However, even in this
regime there remains one classical feature which is used in an essential way: the
assumption of a given space-time continuum. In the preceding chapters we tried
to demonstrate how this last classical relic can be used in developing a detailed
interpretation of phenomena. The situation is not essentially changed if we go
over to curved space-time, as long as the gravitational field can be treated as a
"classical" background field. In quantum gravity, however, this last relic is lost
and this leads to conceptual problems which are unsolved today.
Summarizing the previous discussion, it appears that the quantum theory
of a specific regime in physics (general formalism, specification of the degrees
of freedom, their commutation relations, equations of motion and their relation
to space-time) constitutes a physical theory which is complete and satisfactory
in almost all respects. There remains only one feature which needs additional
careful consideration. The statistical predictions are verified by the study of
many individual cases. Each of them yields a measurement result, regarded as
the transition from a possibility to a fact. If we do not want to attribute this
transition to a change in the consciousness of an observer, then we must say that
the realization of individual facts out of a large number of possibilities is a basic
feature in nature which is not explained by decoherence but has to be either
separately postulated or explained in a picture of nature using new concepts
 VII.1 The Copenhagen Spirit. Criticisms, Elaborations 303

(see section VII.3). In this we should, however, keep in mind that quantum
theory is not only concerned with statistical predictions but also with precise
statements about structural properties, for instance the internal wave function
of an atom in its ground state which, in contrast to the wave function of the
center of mass motion, may be directly regarded as an element of reality.

The Time Reflection Asymmetry of the Statistical Conclusions in

Quantum Theory. The standard formulas for probabilities in quantum theory,
as given for instance in the books by Dirac or von Neumann, refer to a given
"arrow of time". They are not symmetric with respect to time reversal. Let us
illustrate this by a simple argument. Suppose in some ensemble we measure
successively two maximal observables with discrete spectra A l and A2. The
times of measurement shall be t 1 and t2 with t2 > t 1 . We use the Heisenberg
picture and denote the orthonormal systems of eigenstates respectively by 11)k,
V a . The number of cases in which the results k and a are registered shall be
Nk a . The ensemble consists of N individuals. We can look at the conditional
probabilities in the sample

Pk-+a = Nka/Nk ; Nk = Ea
Nka ,
(VII.1.1 )

and
Pa-+k = Nka/Na; Na = Jk Nka . (VII.1.2)

Equation (VII.1.1) is the relative frequency (probability) that, given the result k
in the first measurement, we observe a in the second. Equation (VII.1.2) gives
the probability that, if a is observed in the second measurement, it followed
result k in the first. Now the standard formula in quantum mechanics says that

Pk-.c = 1(40a I V4)1 2 , (VII. 1.3 )

irrespective of the ensemble studied. If this is true then

Pa-+k -- Pk- ■ aNk/Na

depends on the ensemble. In other words, we need more information if we want

to make probability statements about a past situation.
In using the term "ensemble" we must, of course, recognize that this does
not mean an arbitrary collection of individuals. The common feature which is
shared by all the individuals of which it is composed, the criterion used in ad-
mitting an individual as a member of it, must be so clearly stated that it enables
"our friends" to check any statistical regularities we claim to have found in it.
Also we shall not speak of an ensemble unless there exist such regularities. In
other words, given the criteria defining the ensemble there should be proba-
bilities for specific occurrences which can be verified. The time asymmetry in
the basic probability formula results from the claim that it is possible to de-
fine an ensemble entirely by specifications preceding the first measurement. In
304 VII. Principles and Lessons of Quantum Physics

an indeterministic theory it is then impossible to define an ensemble entirely

by "post-selective" specifications. In Bohr's discussion the time asymmetry ap-
pears as obvious. For instance: "The irreversible amplification effects on which
the registration of the existence of atomic objects depends reminds us of the
essential irreversibility inherent in the very concept of observation" [Bohr 581.
We might also speak about the essential irreversibility inherent in the notion of
fact. The above-mentioned asymmetry is not due to the "psychological arrow
of time" nor to any interdiction of statistical statements about the past. The
possibility of registration of sequences of events in a document whose analysis is
not believed to produce any changes allows us to include some "post-selection
criteria" in the definition of an ensemble as discussed by Aharonov, Bergmann
and Lebowitz [Ahar 64] . We can enhance the situation considered at the begin-
ning by adding measurements A at time t < t 1 and B at time t' > t2, study the
statistics of the A 1 , A2 measurements in an ensemble selected by having yielded
the value "a" in the A-measurement and "b" in the B-measurement. Since the
first specification already defines an ensemble, the additional requirement of b
just selects a subensemble. The probability of obtaining the results k, a given a
and b can be written

p(k, a; a, b) = N-1 w(a, k, a, b) ; N = E w(a, k', â , b) , (VII.1.4a)

k' a'

w(a, k, a, b) = tr PbPaPk PaPk PaPb , (VII.1.4b)

where the P's are 1-dimensional projectors corresponding to the indicated prop-
erties and the sum over k', a' runs over a complete orthonormal family of eigen-
states of A 1 , A2. Due to the invariance of the trace under cyclic permutations
we have

w(a, k, a, b) = w(b, a, k, a) ; p(k, a; a, b) = p(a, k; b, a) . (VII.1.5)

Anyone who believes in the dogma that all basic laws of nature must be in-
variant under time reversal may rejoice here. They should, however, recall that
this formula applies only to very specially defined ensembles excluding most
applications in quantum theory and that it is of rather modest value for the
purpose of obtaining retrodictive probability statements for quantum events.
An intriguing view of "quantum past" has been described by Dyson [Dy 92] .

VII.2 The Mathematical Formalism

In the first chapters of his book (1930 edition) Dirac erects an edifice which,
purely on the basis of architectural beauty, could rank among the great master-
pieces of human creations. No traces of the dirt in the previous discussions of
measurements remain visible. It may be appropriate to quote some lines: "We
introduce some symbols which we say denote physical things such as states of a
system or dynamic variables. These symbols we shall use in algebraic analysis
 VIL2 The Mathematical Formalism 305

in accordance with certain axioms which will be laid down. To complete the
theory we require laws by which any physical conditions may be expressed by
equations between the symbols and by which, conversely, physical results may
be inferred from equations between the symbols." The central place among the
axioms is held by the superposition principle, abstracted from interference phe-
nomena but carefully shifted to the realm of symbols for states and deprived
of direct operational interpretation. It leads to the consideration of complex
linear spaces, ultimately Hilbert spaces. The symbols for dynamic variables,
considered by themselves, obey the axioms of abstract *-algebras. In relation
to the state symbols they appear as linear operators acting on Hilbert space.
Many authors in later years have tried to analyse whether these axioms can
be replaced by principles with more direct operational significance, and how
they are embedded in a wider setting, etc. Such endeavors are sometimes clas-
sified as "uninteresting for physics". But, apart from their obvious relation to
many interesting mathematical questions, we must also consider the possibility
that even the basic conceptual-mathematical structure of quantum physics as
we know it may be superseded by a more general theory. It is then relevant
to understand the cross connections between various elements of the theory.
Of course it is quite impossible to present here any comprehensive review. We
can only indicate a few lines of approach, mention some results, and add a few
remarks.
Focusing on the statistical aspect we may start from the general notion
of "state" as representing an ensemble of individual systems, characterized by
some selection criterion which suffices to guarantee statistical regularities in
subsequent observations. Within the set of such "states of a system" we have the
operationally defined process of mixing which implies that this set is a convex
body. As described in section I.1, this may be used to define a real linear space
V with a distinguished positive cone V(+) and a distinguished linear form e (an
element of the dual space) which defines a norm on the elements of V (+) by

IIwII = (e, w) ; w E 0 +) . (VIL2.1)

Since V ( +) is a pointed cone, every element of V can be written (in many ways)
as
cp =wl — w2; cpEV, wi EV (+) . (VII.2.2)
One can then define the norm of (p by I I VI I = (I I wl I I + 11w2 I I) where the
infimum is taken over all possible decompositions. Thereby V becomes a real
Banach space. Due to its special construction it is called a "base norm space".
A convex cone is characterized by its facial structure. A face F C V ( +)
is a convex subcone stable under purification, i.e., if a normalized state w E
F is a m ix ture as in (I.1.11) then w l and w2 also shall belong to F. A pure
state is an extremal point of the convex body, i.e., one for which no non-trivial
decomposition of the form (I.1.11) is possible. The ray defined by a pure state
is a 1-dimensional face.
306 VII. Principles and Lessons of Quantum Physics

Assumption 1. There exist pure states (extremals with respect to the convex
structure). Any state can be obtained (in many ways) as a mixture of pure
states:
w= Aiai; A i > 0, Eati =
1. (VII.2.3)
We use symbols a, Q generically to denote pure, normalized states. In (VII.2.3)
w is a normalized general state.
Among the measurements which can be made on a state (on each individ-
ual of the ensemble) we consider the "propositions" (in Mackey's terminology
"questions") which can only have two alternative outcomes "yes" or "no". Ap-
plying a proposition E to a state w we get a non-negative number p(E; w), the
probability of the yes-answer. To proceed with the analysis we need

Assumption 2. Any proposition E can be applied to any state w.

This allows us to consider a proposition E as a linear form over V, an
element of the dual space V* (in fact of the positive cone V* ( +)) and write

p(E; w) = (E, w) . (VII.2.4)

Comment. There may be good reasons to believe that this assumption cannot
hold in nature (see section VIII.1). One should be aware that a serious restric-
tion here will demand a departure from the standard mathematics of quantum
theory.

In the set of propositions we have a natural partial ordering. We may call

E2 < El if (E2, co) < (El co) for all w E V ( +) . The special element e E V( +) may
,

be regarded as the trivial proposition for which every state gives the yes-answer.
It is the largest element in the mentioned partial ordering. Next we make

Assumption 3. There exist finest propositions ("atoms" which are minimal

with respect to the mentioned partial ordering). Every non-minimal proposition
contains some finest proposition below it.

Comment. Assumptions 1 and 3 correspond to the "type I" situation (compare

sections III.2 and V.2). This is adequate for quantum mechanics; finer topolog-
ical considerations need not concern us here.
While the previous assumptions (though by no means harmless) are of a
fairly general nature, the next one, establishing a close relation between V and
V*, is one of the very specific pillars of the architecture.

Assumption 4. There is a one-to-one correspondence between pure states and

finest propositions. Denoting the proposition corresponding to the pure state a
by g(a) one has
(g(a), a) = 1, (VII.2.5)
i.e., in the ensemble a the finest proposition g(a) yields with certainty the
answer yes. Further, the probability function is symmetric
 VII.2 The Mathematical Formalism 307

(9(a), 0) = CO), a) . (VII.2.6)

As a natural generalization we add:

There is a one-to-one correspondence between the faces of V ( + ) and propositions.
For the general proposition E the corresponding face FE consists of all states
which give with certainty the yes-answer to the proposition E. This implies that
to every face F there is a complementary face F' corresponding to proposition
e — E.

Consequences.

1) Due to (VII.2.2) and assumption I one can extend the map a —p g(a) to
a linear map from V into V*. This, together with (VII.2.6), implies that
we have a real-valued scalar product in V defined by

((Pi I (P2) = (9(V1), tP2) = (W2 1 WO ; (P= E V. (VII.2.7)

This scalar product is non-negative between states and vanishes identically

if the states belong to complementary faces. Thus one may say that V(+ )
V, equipped with the scalar isaelf-porcnthaiespc
product (VII.2.7).

2) The set of faces is an orthocomplemented lattice. Since any subset of

states generates a face by convex combinations and purification, we can
define the face F1 V F2 as the face generated by the union of the states
contained in F1 and F2 and F1 A F2 by the intersection. F —> F' gives an
orthocomplementation:

(F1 V F2 )' = F1 A F . z (VII.2.8)

Thus by assumption 4 the set of propositions is also an orthocomple-

mented lattice.

Following Birkhoff and von Neumann [Birk 36] the operations A, V,' are
interpreted as corresponding to the logical operations of "and", "or", and nega-
tion. This structure in the set of propositions has accordingly been called a
quantum logic. The question is then which additional "axioms" are needed to
show that the lattice of propositions is isomorphic to the set of orthogonal
projectors of a Hilbert space. The required properties of the lattice are "semi-
modularity" and "orthomodularity". Various choices of postulates leading to
this have been given. See, e.g., [Pir 64], [Jauch 69]. The relation of orthocom-
plemented modular lattices to projective geometry and quantum mechanics is
elaborated in [Varadarajan 1968]. A generalization of the standard algebraic
structure, proposed in [lord 34] led to "Jordan algebras", a topic which at-
tracted considerable mathematical interest. For a review see, e.g., [Braun and
Koecher 1966].
308 VII. Principles and Lessons of Quantum Physics

Comments. The relation (VII.2.6) expresses a symmetry between "state prepar-

ing instruments" and "analyzing instruments" and is thus related to time-
reversal invariance. In Dirac's notation it is visible in the use of two different
types of symbols for pure states, the "bras" and the "kets". Of course (VII.2.6)
is the absolute square of the scalar product in Hilbert space.
Instead of looking for additional axioms concerning the lattice of proposi-
tions, one may classify the possible structures of self-polar cones. This has been
a subject of considerable interest in pure mathematics (for a review see, e.g.,
[Iochum 1984]). To translate the problem into algebraic language one considers
the set of linear transformations of V which transform V(+) into itself. They have
a physical counterpart, namely the operations by physical instruments which
do not destroy all systems in the ensemble but leave a final ensemble after
the interaction of the instrument with the members of the initial ensemble. In
short, they produce a change of state. The reason why such operations are rep-
resented by linear transformations of V is that they will conserve the weights in
convex combinations: T E a iwti = E ) Twti if A > 0, E a ti = 1, = 1, which
may be naturally extended to non-normalized states by Taw = aTw. Parallel
to the decomposition theory of von Neumann algebras one has here a "cen-
tral decomposition" of V into primary components. We shall not address here
the fine topological distinctions which arise in the case of infinite-dimensional
V but focus only on the algebraic situation for finite-dimensional subspaces
V of V whose intersection with 0+ ) gives a face V^+} generated by a finite
number n of pure states. Correspondingly we restrict attention to the opera-
tions transforming VP- ) into itself. Then the essential property needed is the
"homogeneity", i.e. the feature that any interior point of V+ ) can be reached
,

from any other such point by an operation. For homogeneous, self-polar, finite-
dimensional cones there is a representation theorem (see, e.g., [Vin 65]) stating
that its faces are in one-to-one correspondence with the subspaces of a Hilbert
space over the field of either real numbers or complex numbers or quaternions;
in particular the 1-dimensional subspaces (the rays) in this Hilbert space cor-
respond to the pure states. The difference between the three alternatives shows
up in the faces generated by two pure states. In the complex case V2 is 4-
dimensional. The scalar product (VII.2.7) in it is the Lorentz scalar product,
so V2 is isomorphic to Minkowski space and V(+) to the forward light cone.
The description in terms of a 2-dimensional complex Hilbert space for the pure
states is given by the spinorial representation of the Lorentz group. This sur-
prising parallel between the simplest non-trivial faces in quantum mechanics
and relativistic space-time has induced speculations about a deeper connection,
such as von Weizsicker's notion of "Ur" (see, e.g., [von Weizsdcker 1988] and
the ideas of Finkelstein on space-time code [Fink68]. Unfortunately this has so
far not led to a viable theory. For the quaternionic case see, e.g., [Fink 62]. For
an argument in favor of the complex case see [Ara 80].
Among the vast literature devoted to the questions discussed in this section
let me mention [Ludwig 1987], [Alfsen 1971], [Alfsen and Shultz 1976], [Alf
 VII.3 The Evolutionary Picture 309

78, 80]; [Miel 69, 74] (theory of filters); [Foul 60], [Pool 68a, 68b] (Baer *-
semigroups), [Conn74], [Be lli 78] (derivations defined by faces).

Final Comments. It may be worthwhile to draw attention to two aspects. First,

the appearance of the linear space V is very general and essentially only due to
the statistical setting. The possibility of considering mixed ensembles leads to
the convex structure of state space which is extended to a linear structure in an
embedding space. Thus this type of linearity is not due to any approximation to
a nonlinear structure and has no deep physical basis (apart from the existence of
ensembles with statistical regularities). The structure of V and V* arising from
the assumptions 1 4 is obviously of the nature of an idealization, distinguished
-

by demands for simplicity and symmetry and justified by its success in a wide
field of applications. In spite of its striking beauty, emphasized by Dirac, we
cannot expect that this specific structure will remain the framework of physical
theories for ever.

VII.3 The Evolutionary Picture

We maintained above that the findings of physics (including quantum physics)

can and should be stated in the picture of an "as if" realism, that physics by
its very method does not transcend the dividing line between an assumed real
outside world and the world of the mind with its impressions, emotions, ....
The surprising behavior of "atomic objects" shows that the "elements of real-
ity" cannot be those in which classical physics believed. What can they be? Let
us start with quantum mechanics where we assume that it is possible to define
a single electron as a "physical system". One says that we can — if we want —
measure the position of the electron at a chosen time. Then one adds that it
would lead to a host of difficulties to believe that the electron has a definite
position when it is not measured. If we look without bias at the experiments
alluded to, we must recognize that the "element of reality" in the position mea-
surement is a dot on a photographic plate or a flash from a scintillation screen.
These are indeed not properties of the electron but properties of the interaction
process of the electron with another system. They are "events" characterized by
a reasonably sharp position in space and in time. Other experiments showing
interference effects, etc., indicate that we should not believe that the electron by
itself has any position at a given time. The natural conclusion is that "position"
is not an attribute of the electron whereas space-time position is an attribute
of a certain class of events. What about the electron itself? We experience it as
a causal tie or link between two events, its "birth" in the electron source and
its death (or transmutation) in the interaction with the detector. We can dis-
tinguish different types of causal links, the simplest ones being stable particles.
Avoiding the poorly defined distinction between elementary particles and com-
posite ones we use the term "particle" for any stable, essentially rigid, structure
(nucleus, atom, crystal, ... ). It corresponds to the "permanently singly local-
310 VIL Principles and Lessons of Quantum Physics

ized states" of Chapter VI. A causal link also has attributes which we can regard
as real. They are structural properties. In the case of the electron they are for
us today only its mass, electric charge, magnitude of spin, and magnetic mo-
ment. For the next generation of high energy physicists they may include some
internal structure. We know empirically that stable particles can be isolated
and that each type has a specific internal structure which can be regarded as a
true attribute, an "element of reality". In fact it is one of the central achieve-
ments of quantum theory that it can determine the possible structures of stable
bodies. The ground state wave function in the center of mass system provides
this information in quantum mechanics. On the other hand, the wave function
for the center of mass motion should not be considered as an attribute of an
individual nor as an element of reality. It refers to probability predictions for
the occurrence of future events.
Let us anticipate some objections. First, why focus on stable particles? This
relates to the "division problem" mentioned earlier. To isolate some part of the
universe which can be considered as independent of the rest we must resort to
idealizations. In order to recognize simple laws we begin with the simplest situ-
ations: a system of a few particles isolated in a large volume for a long time. We
know that such an idealized situation can be realized to a good approximation
though never perfectly (we cannot reach absolute zero temperature or provide
perfect shielding, etc.). One may, of course, also consider an "open system". But
then one has to specify the boundary conditions defining its interaction with the
rest of the world and this is obviously a much less simple task. Let us look at the
idealized situation where we have initially just two clearly separated particles,
an electron and a crystal. Experience tells us that after a possible collision we
find a final situation which can be described in each individual case as a system
of several particles, say ni photons, n2 electrons, and a residual crystal, possi-
bly electrically charged. Furthermore, we can attribute to each such particle an
approximate momentum. The theory reproduces these findings in the following
way. It cannot describe individual processes but treats a statistical ensemble
of them (irrespective of whether we claim to know a pure state wave function
or only a statistical matrix). If we start with a pure state then the final state
will again be described by a wave function. It is a sum of terms of which each
corresponds to one of the final situations described above. This wave function
contains in addition the phase relations between the various terms in the sum.
But we also know (from the theory) that this becomes unobservable in principle
in the idealization in which "final" refers to the limit t — ^^ oo. Thus, effectively,
the final ensemble is a mixture of subensembles, each corresponding to one of
the situations encountered experimentally in an individual case. The theory
suggests, however, that for finite times after the collision there should be possi-
ble experiments which determine the phase relation between different "outgoing
channels" or different momentum configurations. One may argue then that the
individual process cannot be considered as completed at finite times. But then
one must look seriously at the question of how the measurement of the phase
relation in the outgoing wave function, say one microsecond after the collision,
 VII.3 The Evolutionary Picture 311

can be realized. Except in the case of existence of metastable particles with

comparable lifetime, the interference between different final channels has never
been considered and we may wait patiently for the proposal of such an exper-
iment. In the case of long-lived metastable states, the processes of formation
and decay should be regarded as distinct events. To exhibit interference be-
tween different momentum configurations in one channel one needs equipment
which is very extended in space with precisely controlled external fields. The
presence of such equipment conflicts (almost by definition) with the assumed
isolation of the originally defined system in a sufficiently large, prescribed re-
gion of space-time. This reminds us of Bohr's demand for a full description of
the experimental arrangement before we speak of phenomena. However, there
is a change of emphasis. The relevant feature needed is the idealization involved
in separating an individual system from the rest of the world. In the regime
to which local quantum physics applies and limitations due to global circum-
stances beyond our control may be ignored, there is no limit in principle to the
precision with which the definition of such an idealized system can be achieved
and it has a precise sense to say that under such circumstances a certain set of
possible sequences of real events exists from which precisely one is realized in
the individual case s
Another frequently voiced opinion is that macroscopic amplification is es-
sential for the concept of a real event. Indeed, this would be so if we wanted to
equate the notion of event with that of a document which we can show to our
friends. One might argue that there are no facts unless they can be documented.
But this is an unduly rigid position rarely respected by the practising physicist
in quest of understanding and expansion of knowledge. What is important is
that the facts postulated in a theory fit naturally to documented facts. In that
respect the above discussion shows that in the presently available theory the
notion of event or fact indeed needs some idealization but that this does not
mean complex systems or macroscopic amplification but rather isolation in a
sufficiently large space-time volume. Then a collision process between particles
(elementary or composite), leading from an initial configuration to a final con-
figuration and fixing some approximate position in space-time, may be regarded
as a closed event, a fact. The event can be assumed to exist whether we observe
it or not and obviously only few events will be documented. The sharpness
of the mentioned space-time position depends, of course, on the nature of the
event. For low-energy reactions it will be very fuzzy, occupying a large region; in
high-energy reactions we can approach quite sharp localization. We shall return
to this aspect later.
So far the only deviation from the orthodox view has been to replace the
notion of "measurement result" by the more general notion of "event", which is
3 This language has some (superficial) similarity with the "consistent history approach"
[Gruff 84], [Omnès 1994], [Gell 90]. But when we talk here about real unregistered events we do
not mean an arbitrary choice of sequences of projection operators satisfying the "consistency
conditions" with respect to the recorded events, but a very specific set, determined by the
properly idealized physical situation.
312 VII. Principles and Lessons of Quantum Physics

considered as a fact independent of the presence of an observer. This, however,

has important consequences. An event is irreversible. It is the transition from
a possibility to a fact. We are raised in the belief that the fundamental laws of
physics are invariant under time reversal, that they do not stipulate an arrow
of time. What is the evidence which led to this belief, which by now is so firmly
entrenched? Could it possibly be a prejudice? There is the manifest invariance
under time reflection of the basic dynamic equations in classical physics. This
symmetry is somewhat clouded by the question about the appropriate boundary
conditions. In usual applications we use the retarded solution of the Maxwell
equations ("Sommerfeld's Ausstrahlungsbedingung", radiation damping). This
asymmetry arises because we assume that we know the past and not the future.
So it may be considered as "anthropomorphic" or, if one wishes, a consequence
of the "psychological arrow of time" . Another basic piece of experience is the
apparent irreversibility of all natural processes, leading to the second law of
thermodynamics. It is interesting to note that Planck, as a firm believer in the
exact validity of the entropy principle under all circumstances, was motivated
in his studies of black-body radiation by the wish to put the irreversibility of
radiation processes into evidence. Instead of showing this he discovered the
quantum of action and it is slightly ironical that he had to use Boltzmann's
probabilistic interpretation of entropy in his arguments. It was thus the success
of statistical mechanics in explaining the thermodynamic laws starting from a
time-reversal-invariant microscopic theory which convinced most physicists that
the apparent irreversibility was a feature of the macroscopic world resulting from
more basic laws which did not prefer a direction of time.
Let us recall the essential point in the arguments of Boltzmann and Gibbs
leading from reversible microscopic laws to irreversibility in a coarse-grained
description. Different macroscopic states have extremely different statistical
weights or probabilities, roughly defined as the number of microscopic configu-
rations leading to the same coarse-grained appearance. Among the macrostates
— which can be reached from a particular one according to the dynamic law — the
equilibrium state has the highest weight. Therefore, starting from a state with
low weight (far away from equilibrium) there is an overwhelming probability
that it develops into a state of higher weight, i.e., moves closer to equilibrium.
So far the argument does not imply any preference for one time direction. It
would also be overwhelmingly probable that the earlier state was one of higher
weight, i.e., closer to equilibrium. In a laboratory experiment the asymmetry is
again introduced by the psychological arrow of time. There is an initial time at
which the experimenter creates a state far away from equilibrium because he
wants to find something interesting. The various arrows of time are discussed
in [Zeh 1984]. The "biological arrow" and with it the "psychological arrow" is
dominated by the existence of the solar system. We may ask next why the astro-
nomical situation is as it is, why in astrophysics the "thermodynamic arrow" at
any place is obtained by continuous transport of the arrow from another place.
In the approach of statistical mechanics to entropy production the irreversible
Boltzmann equation is derived from reversible microscopic laws by adding some
 VII.3 The Evolutionary Picture 313

natural assumptions about the initial state. The time-reversed situation, where
we apply the corresponding conditions to the final state and calculate back-
wards in time, is excluded as obviously absurd. As a reason for this asymmetry
one can appeal to the cosmological conditions, i.e., the expansion of the uni-
verse. One may accept this latter as an unexplained fact and speculate that
there might be a phase in which the universe contracts and the thermodynamic
arrow turns around. But such questions are clearly beyond the scope of our
discussion here.
Our question is rather: why should there be any difference between quan-
tum physics and classical physics with respect to the status of irreversibility? A
short answer is that quantum physics introduces an element of discreteness man-
ifested in the existence of stable structures and the "indivisibility of a quantum
process". This is closely tied to indeterminacy. The future is open, not precisely
determined by the past. Though some remnant of time-reversal symmetry per-
sists in quantum mechanics and quantum field theory, there is the asymmetry
of the basic statements discussed at the end of section VII.1. In Bohr's view
this "reminds us of the essential irreversibility inherent in the very concept of
observation". In other words it is tied to the psychological arrow of time. But
if we do not want to place the concept of observation into the center of physics
we must ask ourselves: what would be the natural picture if we claim that there
are discrete, real events, i.e., random, irreversible choices in nature?
Starting from this question we come almost unavoidably to an evolution-
ary picture of physics. There is an evolving pattern of events. At any stage the
past consists of the part which has been realized, the future is open and allows
possibilities for new events. Altogether we have a growing graph or, using an-
other mathematical language, a growing category whose objects are the events
and whose (directed) arrows are the causal links. We shall assume further that
the relation to space-time is provided by the events. Each event marks roughly
a region in space-time, the extension and sharpness of which depends on the
nature of the event. No independent localization properties of the links are as-
sumed. This corresponds to the earlier remark that the simplest type of causal
link is a particle and that "position" is not a real attribute of a particle. In fact
it is essential for the understanding of many apparent paradoxes in quantum
physics (including the EPR-effect) that a causal link becomes real only after
both the source and the target events have been realized. Before this it remains
a "potential link", analogous to a free valence bond searching for a partner. For
example, after the source event of emission of a particle we represent it by a
roughly spherical wave function. This should not be interpreted as relating to
the probability for the changing position of a point-like particle but rather to
the probability for the space-time location of the collision center in a subsequent
event. Only after the realization of this target event may we (retrospectively)
assign an approximate world line and incoming momentum to the particle. Let
us assume here for simplicity that customary space-time, in which patterns of
events and links can be embedded, has been independently defined. A pattern
of events and links prior to a given time is a history.
314 VII. Principles and Lessons of Quantum Physics

The quantum laws concern two aspects. On the one hand they must deter-
mine the intrinsic structure of links and events (for instance the internal wave
functions or structure functions of particles). On the other hand they must give
probability laws for the formation of specific patterns, including the positions of
collision centers. The formulation of such laws in the evolutionary picture is an
unaccomplished task. The existing theory provides some guidance but it is only
in the simplest situations (the case of extremely low density of matter) that
a reasonably clean definition of events and links is visible and a comparison
with the standard formulation has been attempted. Let us sketch a strongly
simplified version of this which shows some essential aspects. To each type of
link a (here a type of particle) we have an associated Hilbert space 7la and we
may consider all the subsequently mentioned spaces as subspaces of the Fock
space generated from the -lc, of all types. Consider for simplicity "maximal"
events (corresponding to the strongest possible decisions). They specify their
backward links completely. If the event has two backward links of types a and
a then it selects a specific product vector gyp. ®()ap E 71, ® 7-la and transforms
it to a vector in the tensor product space 711, 0 n ® ... corresponding to the
outgoing channel 4 . This vector, however, is not a product vector but a linear
combination of such. Its expansion into a sum of product vectors depends on
a choice of bases in the factor spaces. The selection of a particular product
vector is realized only by the subsequent events, since links become established
only after both source and target event are realized. A space-like surface not
passing through any event defines a "subjective past" consisting of the pattern
of all earlier events. Among these events there are saturated ones for which all
forward links are absorbed by some other event inside this subjective past and
there are others still having free valence links for the formation of future events.
To such a subjective past we associate a state which summarizes the probability
predictions for possible extensions of the pattern to the future. In our simpli-
fied picture the state depends only on the subpattern of the unsaturated past
events. As the space-like surface is shifted to the future, the associated state
changes as new events appear. This change, analogous to the "reduction of the
wave packet", corresponds to the transition from a possibility to a fact. Let us
illustrate this in the example of Fig. VII.3.1 in which the wavy line indicates the
chosen space-like surface. We are interested in the extension of the past history
by the pattern of events 4 and 5 and the newly established links. The temporal
order of 1, 2,3 is irrelevant but it is assumed that no other events of the past
can play a rôle for the events linked to 3. Events 1, 2, 3 fix unit vectors (not
products)

0 1 E hi ®711; 02 E 7l6 ® 7-c2; 03 E 7 c, ®Hp. (VII.3.1)

Events 4 and 5 are represented by the rank-1 operators (in Dirac notation)

C 04) (çca ® 40-ri, e 05) (Po ® S0 61, (VII.3.2)

4 We have made the further simplifying assumption that the choice of a specific outgoing
channel is included in the characteristics of the event.
 VII.3 The Evolutionary Picture 315

5
I'
II

t /
1/
d
3

Fig. VII.3.1.

where the cps are specific unit vectors in the subspaces 1I ), (A = a, Q, 7, 5) and
Al , T. 5 unit vectors in the tensor product spaces of the new outgoing channels.
The constants c, c', together with the selection of the backward ties, i.e., the
vectors (... I, determine the probability of a single event. Thus the probability
of event 4 is obtained by applying the first operator of (VII.3.2) to c 1 ®03. This
yields a vector whose squared length gives this probability. To obtain the joint
probability of events 4 and 5 we have to apply both operators of (VII.3.2) to
01® 0 2 ®cP3 and square the length of the resulting vector. This joint probability
shows correlations even though these events may lie space-like to each other.
They are due to the fact that the two events have backward causal links to a
common source (event 3). Moreover the vector 03, determined by event 3, does
not specify a product vector cpc, ® cps before both events are realized and thus it
is not possible to assign individual "states" to the not yet established links. It is
this feature which distinguishes the joint probability of events from the case of
classical correlations which would result if we could assume an individual state
for each link (possibly unknown) and then consider correlations between these
states of the links. A prime example is the EPR-phenomenon.
We consider the decay of a spin-zero particle into two spin-2 particles fol-
lowed by a measurement of the spin orientation of the two particles with respect
to respective directions el , e2 prescribed by Stern—Gerlach magnets. This may
be idealized as the situation pictured in Fig. VII.3.1 where events 1 and 2 corre-
spond to the setting of the Stern—Gerlach magnets, event 3 to the decay process,
and events 4 and 5 to the interaction between the decay particles and the two
Stern—Gerlach arrangements, each allowing only a binary decision whose results
are denoted by + or —. Since the events 1 and 2 concern the setting of clas-
316 VII. Principles and Lessons of Quantum Physics

sical apparatus, the links 'y and 6 are already fixed by these events and may
be characterized by the directions e l , e2 . Disregarding the motion in space and
focusing only on the spin, the vector 03 is the unique singlet state in the Hilbert
space of 2-particle spin states. If 4 is the event with outcome + then ço„ is re-
alized as the single particle state (p + (e1 ) (spin oriented in the +e 1 direction).
Since the arrangement is such that we are sure that one of the results + or —
must happen, the constants c and c' are equal to 1. The prescription described
above for finding the joint probabilities thus leads to the well - known quantum-
mechanical expressions. By contrast, suppose we assume that after the event 3
each particle has a definite state, characterized by some variables C so that the
ensemble may be described by a distribution function A(Ci, e2 ) and there is a
probability p +(e, C) that in the state C we get the answer + if the orientation of
the magnet is in direction e, then the joint probability of the result ++ in the
settings el, e2 will be

w++ (el, e2) = f P+(ei Ci)P+(e2,C2)o(Ci,C2)diz(Ci,C2), (VII.3.3)

with corresponding expressions if we replace some of the + signs by — signs.

We know that the p+ and A are non-negative, that p + (e, C) + p_(e, C) = 1 and,
due to the experimentally observed conservation law in each individual case,
w++ (e, e) = w__(e, e) = O. Therefore, if e, e is a pair for which the distribution
function A does not vanish, then

p+(e, e)p+(e , C') = P - (e, C)p - (e, C') = 0 .

If neither p+ (e, C) nor p_(e, C) vanishes, then both p+ and p_ must vanish
for the setting e at C'. This is impossible because their sum must be 1. So
we cônclude that contributions to (VII.3.3) can come only from states where
p+ (el, Ci) equals 0 or 1, i.e., where we have a deterministic coupling between
the state e and the result of the measurement in any direction. This is the case
in which C can be regarded as a "hidden variable" restituting determinism. It
is the case for which Bell derived his famous inequality [Bell 64]. For a simple
derivation see [Wig 78] . This inequality disagrees with the quantum-mechanical
prediction and, according to common belief, also with the experimental results.
Our discussion shows that not only is it impossible to assume hidden variables
but that even the assumption of individual states of the two particles after event
3 and before events 4 and 5 is not tenable.
The decision that one possible pattern of events should be realized may be
regarded as a free choice of nature, limited only by the probability assignment.
The amount of freedom thus accorded to nature is larger than in the standard
view where the experimenter forces nature to decide only on the answer to a
proposed question. It must be stressed, however, that also in the standard use of
quantum-theoretical formalism the element of free choice by nature cannot be
eliminated. It is merely pushed into the background by focusing on ensembles
instead of individual cases. Thus one may derive from the dynamic law governing
the time development of "states" (representing ensembles) that, in the case of
 VII.3 The Evolutionary Picture 317

complex systems, the density matrix very rapidly becomes effectively diagonal
in suitably chosen collective coordinates whatever the initial state may have
been. "Effectively" means that in no realistic experiment will the ofd diagonal
terms play a rôle ("decoherence"). One concludes then that this final ensemble
may be thought of as a mixture of subensembles in each of which the collective
coordinates have specific values. This is perfectly correct as far as statistical
predictions for subsequent measurements are concerned. However, it does not
explain the fact that in each individual case nature has decided on one specific
set of values (e.g., the position of a dot on a photographic plate), a decision not
controlled by the experimenter and not described by the time development of
the density matrix. A striking example of the ambiguities involved in the step
from the statistics of an ensemble to conclusions about individual cases will
be discussed below. It is interesting to note that Dirac advanced the idea of a
free choice of nature in 1927 at the 5th Solvay Congress. Bohr was not happy
with this formulation at that time but used it in his later writings with some
reservations.

Comparison with St an dard Procedure. Consider the interaction of a cos-

mic ray µ-meson with a body of superheated water. The phenomenon observed
will be a string of bubbles which we attribute to "elementary" quantum pro-
cesses
+atom — ^ µ + ion +e+ , (VII.3.4)
each such process creating a localized disturbance which acts as a germ for va-
porization. This verbal description corresponds to the event picture in which we
idealize (VII.3.4) as a closed process which can be separated to a good approxi-
mation from the subsequent amplification. We would further like to assume that
the localization of each bubble is not primarily due to the amplification process
but that it is an attribute of the event (VII.3.4). In fact it should be much more
sharply definable than the extension of a bubble. What can we learn about this
from the conventional treatment of an isolated event of this sort?
We have a Fock space of incoming particles. The initial state is described as
a tensor product of two single-particle wave functions of the respective center of
mass motion (we treat the atom as a single particle). The final state is described
as a vector in Fock space resulting from the application of the S-matrix to this
tensor product. It is a sum of terms describing the different channels. We write
as usual S = 11+ iT and, for a particular final channel (suppressing spin indices)

„l
,out (pl ...p/n) Pin; plrp2)/Mpl)O2 (p2) 64 (E Pik —Epk)dµ(p1)dµ(p2) ,
(VII.3.5)
with dµ(pi ) _ &(p? -- m ?)®(pN4p.
Using
= (270 -4 f esxgd4x
&4(Q)
and noting that eix > Pk represents in any channel just the space-time translation
by x in the Fock space of outgoing particles (similarly eix> P1 for the incoming
318 VII. Principles and Lessons of Quantum Physics

particles) we may write (VII.3.5) in vector notation as

wout =! Tzwind4x

regarding this as a mapping in Fock space where

`fix = U(x)TU -; (x)

is the translation by the 4-vector x of an operator T whose matrix elements are

the functions T(g; pk). The latter are needed only on the subspace of momentum
conservation and their extension away from there is arbitrary. They are smooth
functions of the momenta except at thresholds. We can choose the extensions so
that these functions are smooth (except at thresholds). Then Tx is essentially
a local operator centered around x. The localization of Tx will be poor in the
case of long range forces or "weak processes" like soft photon emission and, to
a lesser degree, also because of the threshold singularities. Let us leave aside
these problems and focus on hard inelastic events. The characteristics of an
event include the nature of the backward links, i.e., the charges, mass and
spin values, and internal structure of the incoming particles and, although they
should not include detailed information about forward links since these are
fixed only in subsequent events, we may include in our case the choice of a
specific final channel and even some rough specification of the momenta of
outgoing particles since this concerns mutually exclusive possibilities, provided
the isolation is adequate. We demanded that we should be able to attribute a
sharply defined space-time region to the event. This is not yet provided by the
sharpness of localization of Tx (which corresponds roughly to the range of the
interaction) but requires that if we make a cell division in x-space, writing

J Tx dx = Tk ; ^k = J Tx9k(x)dx ;
`
^
9k(x) =1;
k 'n ,
T W

with the function gk having support in the cell k, then, for appropriately chosen
cell division, the individual terms Wk may be considered as describing (incoher-
ent) alternatives, one of which is selected by nature in the individual case. By
contrast, believing in the absolute validity of the quantum theoretical formal-
ism, one concludes that the phase relation of different Wk can be made evident
or, in other words, that the needed size of the cells depends strongly on far away
circumstances surrounding the process, not only on the event itself (i.e., on the
presence of instruments which are far away at the time of the event). To assess
the significance of this difference we have to study the statistics of an ensemble
of such processes followed by subsequent measurements on the final state. The
relevant test experiment is a very precise control of the energy-momentum of
all initial and final particles. The assumption of an extension a„ of the event
in the v-direction implies a limitation in the control of the momentum balance
.61:), of order h/ay . This raises the question of how precisely the relevant part of
past history can be controlled in all samples of the ensemble. Here the following
consideration may be instructive. If the overlap region of the wave functions of
 VII.3 The Evolutionary Picture 319

incoming particles were sufficiently narrow, then only a single term, Wk, would
occur. But this is usually not the case. Consider the opposite extreme where
we take the initial state of the atom in (VII.3.4) as an equilibrium state in a
large vessel so that its position is almost unknown. If 0 is the inverse tempera-
ture the state can be described by a density matrix diagonal in the momentum
representation, given (non-relativistically) by

(P'IPIP) = 53 (p' .. P)e_ 2/2m

(we have disregarded the normalization which involves the size of the volume).
Now we note that precisely the same density matrix also arises as a mixture of
Gaussian wave packets, minimal at some time t, with width

A = h(0/2m)1/2 ,

distributed with uniform density in space and time. Numerically, taking m as

the proton m as s, this gives at a temperature of 1 K a value A = 2 x 10-7 cm.
Thus it does not make any difference for the statistics of any subsequent exper-
iment whether we as sume that the initial state is built up from plane waves or
from localized packets of size A. The origin of this ambiguity is, of course, the
non-uniqueness of the decomposition of an impure state and we see here that
we cannot confine attention to decompositions into mutually orthogonal states
because we considered mixtures of packets which overlap and are minimal at
different times. We are reminded again of the fact that the study of statistics
in an ensemble allows widely different pictures for the individual c as e.
Still, there is no known law of nature which would prevent the control of the
momenta of incoming particles and the measurement of the momenta of outgo-
ing particles with arbitrary precision. Such an overall high precision experiment
would be, in the standard language, the one complementary to the well-known
high-energy experiments where we see by inspection in the individual c as e the
existence of a collision center from which the tracks of particles emerge. The
precision in the definition of this collision center may not be great, but it is
much sharper than the controlled localization of the incoming particles. Thus
here is a question indicating a difference between the evolutionary picture and
standard formalism. It should be settled by experiments but this appears at
present to be beyond feasibility because we would need precision measurements
on all initial and final particles and the precision attainable will presumably be
relative to the total momenta involved and therefore not effective for the high
energy reactions in which we expect the localization to be sharp.
Let us turn now to patterns of events and links in the low density situation.
A link, corresponding to a particle, is mathematically described by an irre-
ducible representation of the total symmetry group which is the direct product
of a global gauge group with the Poincaré group. The vectors in this representa-
tion space give the charge quantum numbers and a wave function for the center
of m as s motion and spin orientation. The event is described by a reducible rep-
resentation resulting from the tensor product of the irreducible representations
320 VII. Principles and Lessons of Quantum Physics

associated with the backward links, followed by "quasiprojection" by an oper-

ator Tk. After the event this representation is decomposed again into a sum
of tensor products of irreducible representations, each term corresponding to a
specific channel of outgoing particles which furnish possible links to subsequent
events. A new event is realized by the fusion (tensor product) of such links
originating from different past events. We have been careful so far to speak
of representations, not of vectors in the representation spaces. The reason is
that, in contrast to the simplified picture described before, Tk is not a rank 1
operator and we can only include so much information about backward links
as corresponds to the characteristics we can attribute to events. These include
the approximate momenta determined retrospectively from the location of the
source event. The T-functions in (VII.3.5) do not factor in the variables of out-
going particles. This means that we cannot attribute a specific single-particle
state to a free valence link and this implies in turn that we cannot treat the
probabilities of the formation of subsequent patterns as a classical stochastic
process. While this complication is not very relevant for position patterns in
the case where the mean free path is very large compared to the unsharpness
of localization of events so that all momenta can be taken as quite well-defined
though unknown, there is no corresponding mechanism providing a specification
of the state of spin orientation of the individual particle. This is demonstrated
by the experiments concerning the EPR-effect for spin.

Concluding Remarks. The conceptual structure proposed above incorporates

the essential message of quantum physics and does not seem to be at odds with
known experimental findings. At the present stage it is not clear whether this
structure should be regarded only as an idealization suitable in a certain regime
of phenomena or whether a fundamental theory based on this picture can be
developed. This would demand a more general definition of events and links,
in other words a deeper understanding of the "division problem". It might
demand a finer division of "decisions of nature", related to the quantum of
action rather than to collision processes between stable structures. The relation
of events to space-time must be clarified. It is here that some differences from
the standard formalism will be manifested. One of the factors in favor of the
picture presented is precisely this point. It seems ultimately unsatisfactory to
accept space-time as a given arena in which physics has to play. This feature
persists even in general relativity where a 4-dimensional space-time continuum is
a priori assumed and only its metric structure depends on the physical situation.
In particular, in the absence of all matter and all events there would still remain
this continuum, void of significance. This aspect was one of the factors that
motivated the author to introduce the notion of "event" as a basic concept
with the ultimate aim of understanding space-time geometry as the relations
between events [Haag 90a] . The other motivation was, of course, the desire to
separate the laws of quantum physics from the presence of an observer [Haag
90b]. In this respect it appears that theorists discussing quantum processes
inside a star or in the early universe necessarily transcend Bohr's epistemology.
 VII.3 The Evolutionary Picture 321

Usually the orthodox interpretation is then silently ignored but there are some
efforts to build a rational bridge from the standard formalism to such areas of
physical theory, most prominently the work by Gell-Mann and Hartle [Gell 90,
94]. It uses the concept of "consistent histories" introduced by Griffiths [Gruff
84] and extended by Omnès [Omnès 1994]. One criticism of this concept is that
consistent histories embodying some established facts are highly non-unique.
This led Omnès to the distinction between "reliable properties" and truth.
Still another motivation comes from the following consideration. The gen-
eral mathematical structure of standard quantum theory is extremely flexible.
Its connection to physical phenomena depends on our ability to translate the
description of circumstances (e.g., experimental apparatus) to a specification
of operators in Hilbert space. Apart from the case of very simple systems, the
success in this endeavor is due to the fact that for most purposes no precise
mathematical specification is needed. Thus, for the treatment of collision pro-
cesses in quantum field theory it suffices to give a division of "all" observables
into subsets which relate to specified space-time regions. However, in addition
to this classification of observables one uses the postulate of strict relativistic
causality. Some consequences of this postulate have been verified by the check
of dispersion relations to regions with an extension far below 10 -13 cm. On the
other hand it seems highly unlikely that the construction of an instrument of
intrinsic size of, say, 10 -15 cm and the control of its placement to such an accu-
racy could be possible even in principle, i.e., that we may assume the existence
of such observables. But it is not unlikely that we can attribute to high energy
events a localization of this order of magnitude though we have no means of
verifying this in the individual case. Thus the indirect check by means of dis-
persion relations could be explained by the existence of sharply localized events
rather than sharply localized observables.
The realization of a specific result in each individual measurement has been
recognized by many authors as a challenge to the theory of measurement which
cannot be explained using only the dynamic law of quantum theory applied to
the interaction of a quantum system with a macroscopic device but needs an
additional postulate. In the words of Omnès this is "a law of nature unlike any
other". In a series of papers Blanchard and Jadczyk suggested a formalism in
which irreversibility is introduced in the dynamics of the coupling of a quan-
tum system with a classical one and thereby obtained a (phenomenological)
description of this aspect of measurements (see, e.g., [Blanch 93, 95]).
The evolutionary understanding of reality was proposed many years ago by
A.N. Whitehead [Whitehead 1929]. His writings have influenced philosophers
• and theologians, but few physicists. A notable exception are the papers by H.P.
Stapp in which he outlines a theory of events having many features in common
with the evolutionary picture described above [Stapp 77, 79]. It is a pity that
these seminal papers did not receive the attention they deserve and unfortunate
that I became aware of them too late to incorporate an adequate discussion
of this work. The first two postulates in [Stapp 77] are identical with those
underlying the evolutionary picture. Differences in views concern his postulate 3
322 VII. Principles and Lessons of Quantum Physics

(momentum conservation) and the meaning of causal independence. Especially

the discussion of the EPR-effect in [Stapp 79] differs from the treatment above
and leads to a different assessment of the lessons. In physics D. Finkelstein
suggested an approach to the space-time problem based on similar concepts
[Fink 68]. C.F. von Weizsacker tried for many years to draw attention to the
fundamental difference between facts as related to the past and possibilities as
related to the future and argued that for this reason the statistical statements
in physics must always be future directed [Weiz 73].
Whatever the ultimate fate of these ideas, we should recognize that the
standard formalism of quantum physics is not sacrosanct and will probably be
modified in future theories. With regard to the interpretation there is no basic
disagreement with the epistemological analysis of Niels Bohr but an appeal to
accept that physical theory always transcends the realm of experience, intro-
ducing concepts which can never be directly verified by experience though they
must be compatible with it.
VIII. Retrospective and Outlook

VIII.1 Algebraic Approach

vs. Euclidean Quantum Field Theory

The virtues of the algebraic approach are

1) Naturalness of the conceptual structure; directness and unambiguity of
the physical interpretation.
2) The sometimes amazing harmony between physical questions and mathe-
matical structures. The theory of von Neumann algebras and C*-algebras offers
an adequate language for the concise formulation of the principle of locality in
special relativistic quantum physics and tools for further development of the
physical theory. But more striking are cases where the discussion of physical
questions and the development of a mathematical structure proceed parallel,
ignorant of each other and motivated by completely disjoint objectives, till at
some time the close ties between them are noticed and a mutually fruitful in-
teraction between physicists and mathematicians sets in. One example in our
context is the formulation of equilibrium quantum statistical mechanics in the
algebraic setting and the inception of the Tomita-Takesaki theory of modu-
lar automorphisms. Another one is the correspondence between the statistics
parameter and the Jones index.
3) The fact that a few principles stake out a vast territory in a qualitatively
correct manner. The essential principles are locality in the sense of special rel-
ativity, stability (positivity of the energy), nuclearity. The consequences range
from the emergence of the particle picture with all its ramifications (statistics,
charge structure, collision theory) to a natural characterization of equilibrium
states and the second law of thermodynamics.
4) A rather detailed understanding of the anatomy of the theory with the
recognition of those points where further structural elements are needed and
where modifications are demanded if one aspires to a synthesis with the princi-
ples of general relativity.
The weakness of the algebraic approach is that it has not enabled us yet
to construct or even characterize a specific theory. It has given a frame and a
language not a theory. The many qualitative consequences are all in agreement
with experience but there are few quantitative consequences. Some steps to-
wards a distinction of specific theories within the general frame, involving the
notions of "germs" and "scaling" will be mentioned at the end of this section.
324 VIII. Retrospective and Outlook

Quite another distribution of strengths and weaknesses is met in the ap-

proach via the Feynman path integral in the Euclidean domain. This has been
by far the most widely used approach in the past two decades. Here one starts
from an explicit expression for the Lagrangean in terms of fields. It defines the
action
W = f £(x)d 4x (VIII.1.1)
(taken over all space-time) as a functional of classical field distributions, denoted
collectively by 0. One argues that the generating functional of the hierarchy of
Schwinger functions is given by the functional Fourier transform of e -w when
the time coordinate is taken as purely imaginary.' From the Schwinger functions
physical information is obtained using the Osterwalder-Schrader theorem and
the interpretation of T-functions in the setting of general quantum field theory
as described in Chapter II.
The path integral thus provides a method of quantization of a classical
model. It replaces the canonical formalism and is of much greater elegance
in relativistic field theory. While in quantum mechanics, where Feynman first
presented the path integral as a method of solving the Schrödinger equation, the
scheme is equivalent to canonical quantization, this equivalence is not so clear
in relativistic gauge field theories. The path integral bypasses Hilbert space
and local algebras. To establish the connection to the latter one needs the
Osterwalder-Schrader reflection positivity and growth estimates. This has been
established in simple models but it is not clear whether one really wants it in its
full strength. The path integral is more directly related to T-functions than to
Wightman functions. One gets the expectations of products of local quantities in
temporal order in a distinguished state (the vacuum). It is conceivable that the
time ordered product plays a more b as ic rôle in the foundation of the theory than
the non commutative, associative product of our algebras. The following remarks
may give some indications in this direction. If we look at the efforts to relate the
Hilbert space formalism of quantum mechanics to natural operational principles
it appears from the discussion at the end of VII.2 that one central assumption
is that any observable can be measured in any state. Since, clearly, we cannot
make a measurement before the state is prepared this assumption is reasonable
only if we distinguish between observables and observation procedures and if
we can choose for each observable an observation procedure at an arbitrary
time. This means that one needs a dynamical law which identifies procedures
at different times as the same observable, and this law should not dependent
on the state. General relativity indicates that the dynamical law must involve
the state to some extent. The relevance of this within the context of special
relativistic quantum physics is open, but we may take it as an indication that
the dynamical law cannot be regarded as an algebraic relation in 2t5 (see the
remarks in subsection III.3.3) but arises on the level of von Neumann algebras
'Here Fermi fields (classical spinorial fields) and Bose fields have to be treated differ-
ently. The integration over the Fermi fields (Berezin integration) is the simpler one and can
be carried through in closed analytic form if the Fermi fields occur only bilinearly in the
Lagrangean.
 VIII.1 Algebraic Approach vs. Euclidean Quantum Field Theory 325

and therefore needs at least the weak topology induces by states. If we have
local normality then, in restriction to finite diamonds, all allowed states lead
to the same primary representation and we can get algebraic relations between
observables at different times as implied by (I11.3.44).
In the quantization of a classical field theory by the path integral one meets
again problems familiar from canonical quantization. The devil dwells, as usual,
in details. The counterpart of the representation problem mentioned in section
II.1 is the flexibility in the definition of the functional integral. This is either
handled by splitting e- Iv into two parts, the first corresponding to a known
reference measure, the second to a deviation from it. If the reference measure is
taken as a Gaussian measure, correponding to a free field theory, then the path
integral yields the "magic formula" of Gell-Mann and Low. There remains the
renormalization problem and, in gauge theories, the handling of constraints due
to redundant degrees of freedom. An alternative method, developed in construc-
tive quantum field theory and in the lattice approach to gauge theories, starts
from a problem with a finite number of degrees of freedom by introducing an
ultraviolet cut-off (replacing the continuum of space or space-time by a discrete
lattice) and confining the system to a finite box. Then one can (in principle)
evaluate the functional integrals and study the dependence of the Schwinger
functions on the ultraviolet cut-off, the box volume and on the adjustable pa-
rameters of the model. Renormalization is then replaced by showing that the
adjustable parameters can be chosen as functions of the cut-offs in such a way
that one obtains a well defined limit for the Schwinger functions when the lattice
spacing tends to zero and the box volume to infinity.
The primary question concerns, of course, the choice of the Lagrangean. In
finding out which fields are needed there is some guidance from phenomenology
of high energy physics. However this is not enough as exemplified by the meson
theories of strong interactions and the Fermi type theories of weak interactions
which dominated the scene through some decades. The most valuable theoretical
clue in the search for a specific theory has been the local gauge principle. It is
the generalization from the Maxwell-Dirac system (considered as a classical field
theory) to the Yang-Mills theories with non Abelian local gauge groups. This
principle is well understood and natural on the classical level and, given the
group, it leaves little freedom in the choice of the Lagrangean. Since we do
not know yet how to formulate this principle directly in quantum physics the
characterization of the theory by a classical model has remained indispensable
for heuristic guidance in the search for a good theory. It may be considered as
the central challenge to the algebraic approach to incorporate the local gauge
principle into its conceptual frame.
In conclusion, it seems to me unwise to limit attention to one of the two
approaches. The Lagrangean and the Feynman path integral are at present in-
dispensable tools in the characterization and study of a specific theory. Together
with the local gauge principle they pose questions which in the algebraic ap-
proach are not understood and should be tackled. On the other hand, basing
the theory on a Lagrangean and the path integral only, is certainly too narrow.
326 VIII. Retrospective and Outlook

Many aspects discussed in the previous chapters are not easily accessible there
or not visible at all. The conceptual basis is too weak and needs insights from
other sources even to arrive at the physical consequences of a specific model.
The division of the task of characterizing the theory into two steps: the formu-
lation of a classical model and its subsequent quantization cannot be ultimately
satisfactory. We need a synthesis of the knowledge gained in the different ap-
proaches.

Germs. We may take it as the central message of Quantum Field Theory that
all information characterizing the theory is strictly local i.e. expressed in the
structure of the theory in an arbitrarily small neighborhood of a point. For in-
stance in the traditional approach the theory is characterized by the Lagrangean
density. Since the quantities associated with a point are very singular objects it
is advisable to consider neighborhoods. This means that instead of a fiber bun-
dle one has to work with a sheaf. The needed information consists then of two
parts: first the description of the germs (of a presheaf), secondly the rules for
joining the germs to obtain the theory in a finite region. We shall only address
the first part here.
As J.E. Roberts first pointed out 2 the notions of a presheaf in state space and
its germs are naturally related to the net structure of the algebras. Consider for
each algebra R(0) the set of (normal) linear forms on it. It is a Banach space
E(0). The subset of positive, normalized forms are the states. The algebra
R(0), considered as a Banach space, is the dual space of E(0). To simplify the
language let us denote by R the algebra of the largest region considered and
by E the set of linear forms on it. In E we have a natural restriction map. If
01 C O then R(0 1 ) is a subalgebra of R(0); hence a form cp E E(0) has a
restriction to R(O 1 ) thus yielding an element of E(01). We may also regard an
element of E(0) as an equivalence class of elements in E, namely the class of
all forms on E which coincide on R.(0). In this way we may pass to the limit
of a point, defining two forms to be equivalent with respect to the point x

(Pi ti 402 (VIII.1.2)

if there exists any neighborhood of x on which the restrictions of cp 1 and cp2

coincide. Such an equivalence class will be called a germ at x and the set of all
germs at x is a linear space Ex . It is, however, no longer a Banach space. Let
us choose a 1-parameter family of neighborhoods dr of x, e.g. standard double
cones with base radius r centered at x. Denote the norm of the restriction of a
form cp to R(Or ) by IRoll (r). This is a bounded function of r, non-increasing as
r — O. The equivalence class [cp] x of forms with respect to the point x (the germ
of (p) has a characteristic, common to all its elements, the germ of the function
i^cPf^(r)•
The next question is how to obtain a tractable description of Ex by which
[Link]
2 Private communication around 1985
 VIII.1 Algebraic Approach vs. Euclidean Quantum Field Theory 327

[Haag 96] one may appeal to the compactness or nuclearity properties (see V.5).
Roughly speaking this property means that finite parts of phase space should
correspond to finite dimensional subspaces in H. Here phase space is understood
as arising from a simultaneous restriction of the total energy and the space-
time volume considered. Let EE denote the subspace of forms which arise from
matrix elements of the observables between state vectors in HE, the subspace
of H with energy below E. Correspondingly we have EE (0), the restriction of
these forms to TZ(0) and EE (x) their germs at x. The essential substance of
the compactness requirement is that for any chosen accuracy E there is a finite-
dimensional subspace 4) (0), with its dimensionality n increasing with E and
r, such that for any ço E EE(0) there is an approximate cp E E(;} (0) satisfying

Hy — foII < EIIvII. (VIII.1.3)

As E is decreased n will increase. Since n is an integer, however, this will hap-

pen in discrete steps. For our purposes we shall adopt the following version of
the compactness requirement whose relation to other formulations will not be
discussed.
For fixed E there is a sequence of positive, smooth functions Ek(r), k =
0,1,2... with eo = 1 and ek+ilek vanishing at r = 0. For each k there is
a subspace Nk of E E (x) consisting of those germs for which IIkPII(r) vanishes
stronger than Ek at r = 0. There is a natural number nk giving the dimension
of the quotient space EE(x)/Nk. The n k and the germs of the functions E k are
characteristic of the theory.
We take the ek to be optimally chosen so that 1140 1I (r) = Ek is attained
for some ço E Nk_i. To coordinatize the germs in EE we choose an ordered
basis. A k ,i (r) E TR(n,.) with IIAk,III = 1 such that Iço(Ak,d(r)I = Ek(r)IIcPII for
p E EE /Nk, I = 1,.... (nk — n k _ 1 ). It follows that lim,._+o Ak,a(r)EÇ 1 (r) exists
as a bounded sesquilinear form on HE which might be called a point-like field.
However, since we know so far nothing about the dependence of the structure
on the energy bound E this object could change with increasing E.
In the case of a dilatation invariant theory the situation is much simpler. On
the one hand there are energy independent "scaling orbits" Ak,a(r) provided by
the dilatations. Further, the Ek (leading to the same numbers nk for all energies)
are powers
Ek (E, r) _ (Errk. (VIII.1.4)
Therefore point-like fields

45 k,i = lim Ak, d (r)r-ik (VIII.1.5)

can be defined as sesquilinear forms on any HE Their bound in HE increases in

proportion to E'rk . So the analysis of Fredenhagen and Hertel [Fred 81b] applies
which shows that the Oki are Wightman fields which furthermore have specific
dimensions 7k. The set of these is the Borchers class of observable fields in the
theory and 'yk gives an ordering in this set. Of course, with 0 also the derivatives
328 VIII. Retrospective and Outlook

(V will appear in this set and one can also define a product of such fields as
(the dual of) the germs of
r {ryl+72)(p(A1(r)A2 (VIII.1.6)
(r)),
-

where the indices 1, 2 stand for k 1 , ll and k2 , l2. Since A 1 A2 E 1Z.(Or ) and has
norm bounded by 1 only the values of cp on the basis elements Ak ,a for low k will
-72
be important and give a contribution proportional to r7' -7 1 . Symbolically
one can express this as an operator product expansion

0102 = > c k ,a (r)ok (VIII.1.7)

where the coefficient functions ck ,a behave like rryk -7 1 -72 .

In the physically interesting case of asymptotic dilatation invariance it ap-
pears reasonable to expect that the essential aspects of this structure remain
valid, at least for low k-values. Recently Buchholz and Verch [Buch 95] have
formulated ideas of a renormalization group analysis in the algebraic language.
Perhaps their approach can provide a starting point for a rigorous discussion of
the germs in a theory with asymptotic dilatation invariance.
The accuracy functions ek and corresponding dimensionalities give some
information about the theory. If, in addition, one can define pointlike fields and
an operator product decomposition (which one might hope for in the case of
an asymptotically dilatation invariant theory) then one might also obtain some
relations between the Oki, corresponding to field equations. One can, however,
not expect that this gives complete information about the dynamics. There
remains the problem of joining the germs. One idea concerning this has been
suggested in [Fred 87]. There the additional problem of unobservable fields and
gauge invariance has not been considered. Perhaps the analogue of the gauge
connections of the classical theory will enter in the prescription for joining
germs of neighboring points, involving degrees of freedom not contained in the
individual germ (at least not in low order).

Scaling. Scaling considerations have been a valuable tool in many areas, among
them the analysis of the short distance behavior in Quantum Field Theory (see
e.g. [Sy 72]). In the field theoretic setting one may start from the transformations

0a(x) = N(a)'(Ax), a < a < 1, (VIII.1.8)

where 0 denotes a basic field and N is an appropriately chosen numerical func-

tion depending on 0. In the algebraic setting one has the problem thet (VIII.1.8)
does not define a mapping from A(0) onto A(AO) unless the theory is dilata-
tion invariant because, due to the field equations, an algebraic element can be
expressed in many different ways as a functional of the basic fields and each such
way gives a different image under the transformation since the field equations
change with A. To deal with this Buchholz and Verch [Buch 95] introduce an
algebra A whose elements are orbits A(A) in the quasilocal algebra A which are
restricted by two conditions
 VIII.2 Supersymmetry 329

(i) if A(1) E A(0) then A(A) E A(AO).

(ii) if the momentum transfer of A(1) is contained in a region 4 of p-space

then the momentum transfer of A(A) shall be contained in the region

The authors show that these two restrictions suffice to translate the field theo-
retic discussion based on (VIII.1.8) to the algebraic approach. In particular one
can characterize the different possibilities for the short distance behavior. Of
special interest among these are the cases in which there exists a scaling limit
for ). — 0, leading to a dilatation invariant limit theory, in particular the case
of "asymptotic freedom" in which the scaling limit for the observables reduces
to a free massless theory. One may regard the scaling limit as the theory in
tangent space (compare section VIII.2). Since the scaling limit has higher sym-
metry than the original theory one may expect that the superselection analysis
for the tangent space theory along the lines of Chapter IV leads to a larger
gauge group. Buchholz suggested that in this way the significance of a, local
gauge group in the theory should be understood. For example in QCD the color
group is not visible in the global structure due to confinement but appears in
the tangent space theory [Buck 94, 961.

VIII.2 Supersymmetry

This prominent topic of the past fifteen years has not been mentioned in the
previous chapters. Though its relevance for the description of physical phenom-
ena is not clear at this stage there are several features indicating that it may
contain essential clues for the future development of the theory.
One of them relates to the desire for a unification of geometric and internal
symmetries. In the setting of 11I.3.2 the continuous symmetries form a Lie group
Ç which is a semidirect product of the geometric symmetry group ggeo which
must be a subgroup of the conformal group and contains the Poincaré group
and an internal symmetry group gist which transforms each ,A(0) into itself. 1
g E C has a uniquely determined geometric part (i.e. that Thefact
there is a projection from Ç to ggeo) implies that gint is an invariant subgroup
of Ç and we can write

g (h, k) with h E ginti k E ggeo, (VIII.2.1)

with multiplication law

(h, k)(h', k') _ (hakh' , kk'), (VIII.2.2)

where ak is some action of ÿg eo on Çint. The Lie algebra of Çint must therefore
be a representation space of Ggeo, in particular of the Poincaré group. In the
1 We consider symmetries of the field algebra A, remembering that by the remarks in IV.1.1
and IV.1.3 we should not expect any exact internal symmetries of the observable algebra %.
330 VIII. Retrospective and Outlook

mid sixties the question of whether there can be a non trivial action of ÇÇgeo on
oint was studied and, under some additional physically reasonable assumptions,
it was answered by "no" (see e.g. [O' Rail 65), [Mich 65), [Cole 67), [Orz 70),
[Garb 78a,b)). There can be no generators of Çlnt which do not commute with
the translations or which have any vectorial or tensorial transformation char-
acter under the Lorentz group. The conclusion was that Ç can only be a direct
product of Çgeo and ÿ1m ; there is no interplay between the geometric and the
internal part. The aspect changes when one considers a "super Lie algebra"
or, in mathematical terms, a "Z2-graded Lie algebra". Just as an ordinary Lie
algebra it is a linear space equipped with a bracket operation but it has two
distinct subspaces: the even elements (grade 0) and the odd elements (grade 1).
Denoting the grade of element a by s(a) the bracket has the property
[a, b ] = (-1 )e(a)8(6)+1[b,a] (VIII.2.3)

and satisfies the generalized Jacobi identity

8(c)8(a)+8(c)9(b)
[a , [b ,
c]]+(_1)8(a)8(0+e(a)8(c)[b,

[c, a]]+( -1 ) [c, [a, b]] = 0. (VIII.2.4)

The sign factors by which (VIII.2.4) differs from the usual Jacobi identity can
be read off from (VIII.2.3) . The cyclic permutations of the elements a, b, c can
be obtained by two transpositions and each transposition must be accompanied
by the sign factor in (VIII.2.3). Thus, from the order a, b, c to the order b, c, a
we have the transposition a b followed by a 4-4 c which together yield a
sign (—I)3(a)8(b)+8(a)8(c) . In the physical context the odd elements correspond
to fermionic, the even elements to bosonic operators and the bracket is the
anticommutator or the commutator, respectively. Such a fusion of commutation
and anticommutation relations into a unified structure was proposed by Berezin
and Kac [Bere 70]. The fascinating aspects of this structure became apparent
when Wess and Zumino introduced a super Lie algebra containing the generators
of the Poincaré group in which the energy momentum operators were given
by the bracket of fermionic generators, [Wess 74]. The "baby model" of this
algebra involved only the components Q,. (r = 1, 2) of a "spinorial charge" , its
Hermitean conjugate Q; , the generators PP and M g, of translations and Lorentz
transformations, respectively. Here we used the van der Waerden notation for
2-component spinors described in subsection I.2.1. The commutation relations
of M ,, with the other quantities are fixed by the requirement that the Q,. and
^

Q,'. transform as undotted and dotted spinors, respectively. Writing the 4-vector
PA as a rank 2 spinor PT.; the remaining commutation relations are

[QT, Q3]+ = Pis; [Q,., Q8]+ = 0, (V III.2.5)

[Qr, Pµ ] = 0, (VI I I.2.6)

together with those resulting by Hermitean conjugation. The suffix + indicating
the anticommutator has been written only for emphasis. In the graded Lie
algebra [ , ]+ is just the bracket between odd elements.
 VIIL2 Supersymmetry 331

One remarkable feature is the appearance of spinorial charges which extend

the geometric symmetries in a non trivial way. Applying relations (VIII.2.5),
(VIII.2.6) to single particle states with mass m 0 we obtain a multiplet of
particles with equal mass but different spin. The spin content of such a multiplet
is easily evaluated looking at the restriction of the commutation relations to
the "little Hilbert space" (degeneracy space in the center of mass system where
= (m, 0) and thus Prs is the numerical matrix mll). There (VIII.2.5) just P^
says that m -1 2Qr, m-1^2 Qr satisfy the canonical anticommutations relations.
We have a system of two pairs of fermionic creation-annihilation operators in
the degeneracy space. Their relation to the stability subgroup of the Lorentz
group, the 3-dimensional rotation group, is

[Mk,QT ] = —2 >( Qk)rs (%, 's, k = 1, 2, 3; r, s = 1, 2. (VIII.2.7)

We have used the usual notation for the angular momentum operators i.e. M3
M12. Q* and Q2 raise the 3-rd component of the spin by 1/2, Q2insteadof
Q1 lower it by 1/2. The degeneracy space is spanned by four vectors: a and
bottom state lb), annihilated by the Q,. and belonging to spin 0, the doublet
QTIb) belonging to spin 1/2, and a top state It) = QIQ2Ib), again with spin O.
The most significant aspect of supersymmetry is, however, the expression
for the energy operator given by (VIII.2.5)

P° = Q1Qi + Q2 Q2 . (VIII.2.8)

The right hand side is a manifestly positive operator. Thus the positivity of the
energy, one of the central pillars of the theory, is automatically included in the
supersymmetry relations. This is reminiscent of Dirac's original idea to draw
the square root of the relativistic Hamiltonian. The incredible fruitfulness of
this idea is well known to all of us.
The baby model can be embellished by taking several pairs of spinorial
generators Qr (L = 1, • • N) which are transformed into each other by an
internal symmetry group Ç. The possible commutation relations are severely
restricted. Adapting the argument of [Cole 67] to the case at hand one finds
that in a massive theory the unbroken part of the super Lie algebra can only
contain the generators of j3 and of gint besides the QL , QL* and that q3 and
G;nt still commute with each other. The graded sub-(Lie)-algebra generated from
the QL, QL* contains only the PP and possibly "central charges" of Cu nt which
commute with all other elements in the algebra. In particular the Lorentz group
and the non central internal symmetries are not generated from the spinorial
charges. The relations (VIII.2.5) are replaced by
L M*
[Qr ' Qs ]+ — SLM Prs. (VIII.2.9)

The spinorial generators commute with the translations. The spin content of
supermultiplets in this case is discussed in detail in [Lopuszanski 1991.]
A much more interesting structure emerges in the massless case with con-
formal invariance. Then we may have two types of spinorial charges, denoted by
332 VIII. Retrospective and Outlook

QL and Q,L. Together with their Hermitean conjugates they generate a bosonic
symmetry group which is the direct product of the conformal group and an
internal symmetry group U(N). The explicit form of these relations and their
derivation was given by Haag, Lopuszanski and Sohnius [Haag 75] .
So far the discussion concerned the global charges. In a local theory the
elements of the super Lie algebra should arise as integrals of local densities over
a space-like surface, in the case of the spinorial charges as integrals of local, rank
3 spinor fields. The commutation relations of these fields are the analogue of the
current algebra in the bosonic case. (VIII.2.5) suggests that the bracket of the
rank 3 spinor fields must lead to the energy momentum tensor which is related
to diffeomorphisms of space-time in an analogous way as the electric current
is related to local gauge transformations. Thus, in conjunction with a local
gauge principle one may hope to be led to an approach to a quantum theory of
gravity. It is beyond the scope of this book to discuss the very extensive work
which has been done in this direction, called supergravity. A survey is given in
[ Wess and Bagger 1984 See also the reprint volumes [Salam and Sezgin 1989],
[Ferrara 1981.

VIII.3 The Challenge from General Relativity

VIII.3.1 Introduction

The synthesis between the tenets of quantum physics and those of general rela-
tivity has remained an unsolved problem for over sixty years. The early discus-
sions between Einstein and Bohr on physical reality may well be seen as an omen
forshadowing the difficulties encountered. Local quantum physics as described
in the previous chapters accepts space-time, including its causal structure, as
a given arena in which physics is staged. One may say that the space-time
coordinates, tied to a reference system established by an observer, are classi-
cal quantities belonging to the observer side of the Heisenberg cut. The global
structure of space-time is needed for the commutation relations between observ-
ables, in particular for the causal commutativity at (arbitrarily large) space-like
separations. The local metric structure is needed in the formulation of dynam-
ical laws in quantum field theory. Translation invariance is necessary for the
definition of energy-momentum which, in turn, is central for the formulation of
stability and nuclearity. Now, if we wish to include gravity and understand it
as a modification of the metric field depending on the matter distribution and
hence on the prevailing physical state, we can distinguish three levels.
On the first we still retain space-time equipped with a classical metric field,
attached to the observer side of the Heisenberg cut, but abandon the specializa-
tion to Minkowski space and allow metric fields with curvature (see subsection
I.2.3). At this level much of the previously described structure can be taken
over. We can regard the theory still to be defined in terms of a net .4(0) of
*-algebras associated with open regions of the space-time manifold M and sat-
 VIII.3 The Challenge from General Relativity 333

isfying causal commutativity. This level is appropriately called quantum field

theory (resp. local quantum physics) in curved space-time. The main problem
which has to be resolved there is caused by the absence of translation invariance
(possibly of all geometric symmetries). Thus we do not have axiom A of the
Wightman frame and must find another formulation of stability. This can be
done and there are some interesting consequences which will be discussed in the
next subsections.
On the second level one may still retain a 4-dimensional space-time contin-
uum as an ordering structure for the net of observables but assume no longer
that it is equipped with a (classical) metric field. This level may be called (semi-
conventional) quantum gravity. The problem here is to generalize manifold the-
ory to build an object which we might call a manifold with an event structure
over it from its germs. The macroscopic metric, being classical, must then be
understood in analogy to spontaneous symmetry breaking. It is determined by
the prevailing physical state which breaks the diffeomorphism invariance and
refers to collective degrees of freedom in regions large compared to the Planck
length. Much work has been devoted to the program of quantizing Einstein's
theory of gravity either with or without inclusion of matter fields and gauge
fields. Important progress has been made in this approach by formulating the
theory in terms of spinorial variables replacing the metric tensor gu,, [Asht 87).
For further developments along this line and references see [Ashtekar 19911.
On the third level, aiming at the famous TOE (theory of everything), the
notion of space-time itself is abandoned as a primary concept. To provide the
mathematics which might be needed in this venture several ideas have been
put forward in the past decade: non commutative geometry, quantum spaces
and quantum groups, superstring theory. Conceptually one has to overcome the
Heisenberg cut and develop a theory of real event structures. We cannot enter
this level in this book and shall only offer a few remarks in subsection 3.4.

VIII.3.2 Quantum Field Theory in Curved Space-Time

Assume that a classical gravitational background field is given. The large masses
from which it originated shall not concern us. Mathematically we have a given 4-
dimensional manifold M with a given pseudo-Riemannian (hyperbolic) metric.
In a chart (coordinate system within some region) the latter is described by
a classical metric field gov (x). We want to define the net A(0) on the space-
time manifold M. To focus on a tractable example we consider the case where
the net arises from a Wightman field obeying a linear, generally covariant field
equation, resulting from the field equations of the Minkowski space theory by
replacing the derivatives 8N, by the covariant derivatives Do. In the case of a
scalar field the Klein-Gordon equation is replaced by

(Dg + m2 )45 0 (VIII.3.1)

where
❑ g = 9^ -1I2 a^191 k,2aµ (VIII.3.2)
334 VIII. Retrospective and Outlook

and Igo denotes the absolute value of the determinant of the metric tensor gi ,1.
The commutation relations are

[.P(x), P(y)] = id(x, y)Il, (VIII.3.3)

where 4 is the difference between the retarded and the advanced Green's func-
tion of the Klein-Gordon equation. It is a uniquely defined distribution on
M x M provided M is globally hyperbolic.
Let us start from a more general setting. We consider first a larger net of
algebras which contains only the kinematical information that we are dealing
with a scalar field. We shall call it the net of Borchers-Uhlmann algebras and
denote it by Au since it was first mentioned in [Borch 62], [Uhl 62). For a region
0 C M Au(0) is the tensor algebra of C°°-test functions with support in 0.
It is the direct sum of monomial parts

,Au = ® o.AÛ (VIII.3.4)

where 4 consists of functions of n points. The unit element is the number 1

in the zero grade part. The general element is a hierarchy of functions

F= (VIII.3.5)

of which only a finite number are non vanishing. Addition is grade-wise. The
algebraic product is the (not symmetrized) tensor product. With Fl E AÛ,
F2 E.A7), F1F2EAÛ m :

(F1F2)(x1 • • . x n+m) = f^n) ( xl . • • xn) f2m)( xn +l . • • xn+m). (VIII.3.6)

An involution is defined by reversing the order of arguments plus complex con-

jugation. F F* corresponds to f (n) (x1... xn )
— in) (xn ... xi). Thus A u is a
*-algebra. Moreover there is a natural action of the local diffeomorphism group
of M on Au. If g is a diffeomorphism mapping 01 on 02 then ag is given by l
g-lxn); xk
(c g i) (x1 . . . xn) = f (n) (g-1 xi . . . E 02. (VIIL3.7)

It maps f(n ) E rt (01) to ag f (n) E 4(02 ). A state w is a normalized positive

linear form on Au. It may be characterized by a hierarchy of distributions win)} {

where win) E D'(Mn) and

w(F) = E w(n) ( Pn)). (VIII.3.8)

l One could consider a different transformation law, putting on the right hand side of
(VIII.3.7) a factor fi JP(xk) where J is the Jacobian of the transformation g. The value of
p determines whether 0 is a scalar or a scalar weight of some order. In fact, if we omit this
factor and want 0 to be a scalar then we would have to understand 0(f (1) ) to mean formally
f 0(x)fi ll (x)Igl 1 /2 d4 x. But this will not be relevant for the discussion in this subsection.
 VIII.3 The Challenge from General Relativity 335

The hierarchy must satisfy the positivity condition

w(F*F) > 0 for all F (VIII.3.9)

and the normalization

w(o) = 1. (VIII.3.10)
Of course, ,AU does not carry much information. To inject the dynamical
law and the commutation relations we have to divide A U by an ideal 3, the
elements which are zero due to (VIII.3.1), (VIII.3.3). The dynamical ideal, for
instance, is generated from the elements in the grade-1 part of the form

(©g + m2)*f (1),

resulting from (VIII.3.1) by averaging with a test function. Still, A = A u /3
does not determine the germ of the theory. There remain uncountably many
inequivalent representations of A U (0) for arbitrarily small O. This follows, as
in the case of the Minkowski space theory, from the arguments in II.1 'since
the field equations help only to reduce the algebra to the algebra of 0 and its
time derivative on a space-like surface and then we remain with the canonical
commutation relations. In Minkowski space this problem was handled by re-
ferring to a distinguished state, the vacuum, and subsequently assuming local
normality of all physically allowed representations with respect to the vacuum
representation. The characterization of the vacuum involves, however, global
aspects and, in the case of curved space-time it is not evident how to select a
distinguished state [Full 73]. Insisting on the principle of local definiteness we
have to characterize a unique folium of physically allowed states for sufficiently
small O. For the free theory the first proposal in this direction was to require
that the 2-point function should have the Hadamard form [Adl 77, 78], [Wald
77, 78], [Full 81a, b]

w (2) (x1 i x2) = uQ-i + v ln I Q l + w, (VIII.3.11)

where a(x i , 52 ) is the square of the geodesic distance between x i and x2 and
u, y, w are smooth functions of x i and x2. This was required to hold in a suf-
ficiently small neighborhood of an arbitrary point x. Due to the field equations
and commutation relations u and u are uniquely determined by the geometry in
the neighborhood of x i.e. by the g,,,, and their derivatives. Only the last term,
the function w, can depend on the individual state in the folium. Important
progress in understanding the significance of the Hadamard form for w (2) in a
free theory is due to Radzikowski [Radz 92], who related it to Heirmander's con-
cept of wave front sets and "microlocal analysis". This leads also to a definition
of Wick ploynomials. We shall return to this below. In [Full 81a, b] it was shown
that if the metric is globally hyperbolic and regular then the Hadamard form
persists at later times if it is satisfied in the vicinity of some space-like surface.
Another approach which is not confined to free theories and sheds some
light on the relation between the sho rt distance behaviour of states and the
336 VIII. Retrospective and Outlook

dynamical law assumes that the partial state in the neighborhood of any point
has a scaling limit. This allows the reduction of the theory to the tangent space
of a point. One finds that the tangent space theory has translation invariance
and a distinguished translation invariant state which results as the scaling limit
of an arbitrary physical state of the full theory [Haag 84], [Fred 87]. To describe
the procedure we use a coordinate system in the neighborhood O of the point
x to which we want to contract but the conclusions are intrinsic. Consider a
vector field X which vanishes at x and is of the form

XP(y) = (yµ — st-`) + O(y — x) 2 . (VIII.3.12)

Note that this is a coordinate independent restriction. The form of Xo(y) will
differ in different charts only by terms which are at least quadratic in (y — x).
Consider next the orbits of points in O under the motion

dy0 (a)
^-1XP(y(A)); e(1) = y E O; A E [0, 1]. (VIII.3.13)
da

y(A) will move to x as A —* 0; so (VIII.3.13) contracts the neighborhood to the

point x. The solutions of (VIII.3.13) define a semigroup of local diffeomorphisms
gx(A)
gx (A)y = y(A), (VIII.3.14)

and, at A = 0, the parametrized curve y(A) has a well defined tangent vector
which we denote by nx y, considering rix as a diffeomorphism from O into the
tangent space at x. One has

nxgx(A) = Ar1x• (VIII.3.15)

Corresponding to the diffeomorphism gx (A) we have a morphism o (gx (A)) from

Au(0) to A U (gx (A)O), defined by (VIII.3.7) with g replaced by gx (A). Simi-
larly we have a morphism a (nx ) from Au (d) to the tensor algebra Au of test
functions on the tangent space Tx with support in O = nxO

(c (iix ) f) (z 1 . • • zn ) = f (n) • . . yn ) with zk = rlxyk. (VIII.3.16)

We have denoted the coordinates in Tx by z and shall denote objects relating

to tangent space by a roof °. By (VIII.3.15) one has
^
a( 71x)a (gx(A)) ( ( 71x 1 ) = â(a) = a(D(a)) (VIII.3.17)

where D(A) is the dilation in tangent space:

((A)J) (z1 . . • zn ) = f(a -1 z1 • • • a -1 zn ). (VIII.3.18)

 VIII.3 The Challenge from General Relativity 337

Definition 3.2.1
A state w on Au(0) is said to have a scaling limit with respect to the contracting
vector field X if there exists a scaling function N(A) (positive, monotone) such
that for all natural numbers n and all f (n) E D(On ) lima_,0 N(A)nw(a(gx(a)) f (n) )
exists and is nonvanishing for some f (n) .

Since a(rjx ) is a positive linear map we may regard the limit as a state on
Au , the tensor algebra of test functions in tangent space, and write
(a(gx(A))a(1)f) = lim
Wx (f(n) ) = lim N(A)nw N(A)nw (a&(.X) f) .

(VIII.3.19)
The last form follows from (VIII.3.17), (VIII.3.18) and has the advantage that
it is defined on all of Au since the support of â(A)f (n) will move inside (5 for
sufficiently small A whenever fin) has compact support. The map on tangent
space, introduced by the factor a(rjX 1 ) is essential for the first item in the fol-
lowing theorem which lists the basic properties of the scaling limit.

Theorem 3.2.2
Assume that w has a scaling limit at x with respect to the contracting vector
field X. Then:
(i) The limit state (VIIL3.19) is independent of the choice of the vector field
X within the class (VIII.3.12). It depends only on the base point x and
we shall therefore denote it by (Dz•

(ii) The scaling function is "almost a power" i.e.

(VIII.3.20)
li ^
ô N (a )^ ) = Aa'
We shall call d = 4 + a the canonical dimension of the field
(iii) The limit state has the scaling behaviour
A- nazDz (pn) ). (VIII.3.21)
wz(a(A) 1(n)) =

(iv) If w is primary3 then the scaling limit exists for a dense set of states in
the folium of w and it leads to the same limit state W z for all of them.
(v) If the approach to the limit is uniform for x varying in some neighborhood
(see (VIII.3.25) below) then the limit state is invariant under translations
in tangent space i.e.

wz(a(Ta)F) = Dx (F) where (TaPn))(zi ... zn ) = P n) (zl — a ... zn — a).

(VIII.3.22)
2 Negativevalues of d would mean that the w (n) have no singularity at coincident points.
3 Thisrefers to the von Neumann algebra determined by the GNS-representation 7r u, of
Au (0). The transition from the unbounded operators ir,(F) to bounded operators is afforded
by polar decomposition as indicated in section III.1.
338 VIII. Retrospective and Outlook

The proof of the theorem is given in [Fred 87]. Here we shall only describe
the additional assumption which is needed to establish the property (v) which
is crucial for the following. The basic relation from which the argument starts is
most easily seen if we choose the bases in the tangent spaces as corresponding
to the coordinate axes and choose the contracting vector field to a base point
x so that the term of order (y x) 2 is absent. Then
—

gx(A)y -- z + A(y — x); rizy = y — x, (VIII.3.23)

and thus

9x(À) 11x 1 T (a) = 9x+Aa(a)i1x+Aa = 9x(A)77x 1 +10(A). (VIII.3.24)

The validity of (v) depends on the possibility of neglecting the term of order a
on the right hand side of (VIII.3.24) in the limit a — O. The precise condition
for this is the existence of a bound, uniform in x within some neighborhood, for
the difference between the limit and the scaling value at finite small A

Iwx (A; F) — lint) Lux (A; F)] < R(A; F) for x E 0, (VIII.3.25)

where wz (a; F) denotes the expression in definition 3.2.1.

From Wx we get by the GNS-construction a Hilbert space representation
of the tangent space algebra A u in which iDx is represented by a state vector
Is). The translation invariance of the limit state implies that we also have
a representation of the translation group by unitary operators U(a) acting in
this Hilbert space with IS) as an invariant vector. This allows the definition
of energy-momentum operators in the tangent space theory and therefore the
formulation of a principle of local stability.

Postulate 3.2.3 (Local Stability)

The energy-momentum spectrum in the tangent space theory at z is confined
to the positive cone V + (defined by the metric gµ„(x) at the point z).

The causal commutation relations of the theory on M translate to causal com-

mutation relations in the tangent space theory. It seems probable that must
also be Lorentz invariant, though the necessary conditions have not been inves-
tigated. Thus the tangent space theory is expected to satisfy the axioms A-E of
a quantum field theory in Minkowski space. Moreover, if the dimension of is
d = 1 (which in the case of the free field follows from the canonical commutation
relations (VIII.3.3) and remains to be true if the theory is asymptotically free in
the short distance limit) then the 2-point function in the tangent space theory
is determined to be that of a massless free field theory in Minkowski space and
then it follows from standard arguments (using the positivity (VIII.3.9) and
causality) that the tangent space theory reduces to the theory of a massless free
field in Minkowski space.
The dependence of wx on the base point x in a fixed coordinatization of M
and of the tangent bundle is given by the affine connection:
 VIII.3 The Challenge from General Relativity 339

VAwx = 0, (VIII.3.26)

vu = aa^ - F^S K' , (VIII.3.27)

Ketr3 D (.f) = wx(z81), (VIII.3.28)

k
where F are the Christoffel symbols.
For the 2-point Wightman function of the original state w one obtains the
short distance behaviour

w (2) (x1, x2) = Q - 1 + dw(2) (VIII.3.29)

where .6w (2) must be less singular than of order 2 in the coordinates. Thus the
leading singularity agrees with that of the Hadamard form though (VIII.3.29)
would still allow a singularity in .w(2) stronger than logarithmic for general
physically allowed states. Here it must be remarked that (VIII.3.29) is not
sufficient to guarantee the principle of local definiteness whereas the Hadamard
form suffices for this purpose [Verch 94] (see also [Lüd 90] and the literature
quoted there).
Let us return now to Radzikowski's proposal of characterizing the class of
allowed states in terms of the wave fr ont sets of Wightman distributions. This
contains more information than the scaling limit; it relates to the germs, not
only to the tangent space theory. Moreover it allows a stronger formulation of
the spectrum condition, replacing Wightman's axiom A at least for free fields.
An additional bonus is that one can define Wick polynomials of fr ee fields as in
the Minkowski space theory.
The wave front set of a distribution u, denoted by W Fu, is a refinement
of the notion of singular support, lifting it from base space to the cotangent
bundle. It is a local notion in the sense that only the behavior of u in an
arbitrarily small neighborhood of a point in base space is relevant. Note, how-
ever, the for u = On ) the base space is 4n-dimensional i.e. we consider points
X = x i ... x,,, in configuration space. Introducing a chart we can first "localize"
the distribution u by multiplying it with a smooth function h with support in
some neighborhood U of X and then take the ordinary Fourier transform of
uh, yielding a smooth function uh(e) in "momentum space". The pair X, 6 is
called a regular directed point if there exists some neighborhood U of X and
some conic neighborhood C of 6 such that for every smooth function h with
support in U the function (1 + lel) N UhO stays bounded in q for all N > 0.
means that with 6' E q also A ' is contained in C£ for all a > 0. ForConic
the distinction between regular directed points and others only arbitrarily small
neighborhoods of X are relevant. Therefore the definition does not depend on
the chart and the pairs X,6 may be regarded as points in the cotangent bundle.

Definition 3.2.4
The wave front set WFu of the distribution u in M is the complement of the
340 VIII. Retrospective and Outlook

set of all regular directed points in the cotangent bundle T*M, excluding the
trivial point e = 0.

One has

Theorem 3.2.5 [Radz 92]

Let cP be a free field on the globally hyperbolic manifold M, satisfying (VIII.3.1),
(VIII.3.3) and w a quasifree state. The the following conditions are equivalent

(i) the 2-point function w( 2) has the wave front set

WFw(2) = {x 1 , k1; x2, _k 2 } with x1, k1 ^' x2, k2; k° > 0.

where x, k ti x, k means that there is a lightlike geodesic 'y from x1 to x2

with cotangent vectors k 1 and k2 respectively.

(ii) w(2) has the Hadamard form.

Since for a quasifree state all truncated functions for n 2 vanish the 2-point
function suffices to define the state. The support properties of the On) in mo-
mentum space are such that pointwise products of these distributions are well
defined. As in Minkowski space Wick polynomials of the free field can be defined
[Brun 95]. A construction of Hadamard states has been given by Junkers [Junk
95].
The notion of wave front sets may be applied equally well to the On) of an
interacting theory. Thus it is natural to try to formulate the general spectrum
condition in terms of wave fr ont sets. In an interacting theory it is, however,
not clear whether the singular support is confined to light-like or even time-like
separation of the points. Brunetti, Fredenhagen and Kohler [Brun 95] suggested
the following prescription. Consider a graph Çn whose vertices represent points
in the set {x1 , ... , xn} and whose edges represent connections between (a subset
of) pairs xi , x3 by smooth paths from xi to xi . To every edge e one assigns a
covariantly constant time like covector k(e) which is future directed if i < j but
-

not related to the tangent vector of the path. In g„, there appears with every
edge e also the inverse e -1 which carries the momentum k(e -1 ) = k(e). The
—

proposed condition is then

Proposed Condition 3.2.6 ("Microlocal spectrum condition")

The wave front set of 'Ong is contained in the set {x 1 , k1 i ... , xn , kn } for which
there exists a graph Çn as described above with k i = > k(e ),) where the sum
runs over all edges which have the point x i as their source. (The trivial momen-
tum configuration k 1 = ... kn = 0 is excluded).

Probably the condition is not optimal but the fact that the k i are all time-
like and the distinction between future- and past direction is determined by
the sequence of the index of x i implies that pointwise multiplication of the
distribution is defined.
 VIII.3 The Challenge from General Relativity 341

Summing up: To make quantum field theory in curved space-time well de-
fined (to the same extent as a corresponding theory in Minkowski space) one
needs a specification of the set of physically allowed states as demanded by the
principle of local definiteness. This is partially achieved if one requires that the
states have a scaling limit. This gives, for every point in M, a reduced theory
in the tangent space which allows a formulation of local stability. The tangent
space theories are typically expected to be isomorphic to a free, massless, local
theory in Minkowski space. The transport of the tangent space theories from
one base point to another is governed by (VIII.3.26) to (VIII.3.28). The scaling
limit is, however, not yet sufficient to characterize the germ of the theory. The
requirements on the wave front set are stronger and yield this information. It
should be borne in mind that these restrictive conditions do not characterize a
specific state but a folium where they apply to a dense set of states. There is,
in general, no distinguished state corresponding to the vacuum.

VIII.3.3 Hawking Temperature and Hawking Radiation

Classsical general relativity leads to the conclusion that very massive stars ulti-
mately end by gravitational collapse, leading at some stage to the formation of
a black hole from whose inside no signal can reach an outside observer. Further-
more, for the outside world the black hole has some aspects of a thermodynamic
system in equilibrium ("no hair theorems", entropy) [Hawking and Ellis 1980],
[Bek 73]. In a seminal paper Hawking argued that a black hole has a surface
temperature
T = h(87MG) -1 (VIII.3.30)

which can be understood by considering quantum field theory in the curved

space-time due to the gravitational field of the collapsing star. Here M is the
mass of the star, G the gravitational constant [Hawk 75]. We have written
Planck's constant explicitly since it is one of the surprising aspects that here
for the first time one sees a nontrivial interplay between quantum physics and
general relativity.
Hawking's first computation related the (partial) state of the quantum field
at late time far away from the black hole to the state at an early time before the
star had begun to callapse. Up to this time the metric was practically static. For
a globally static metric one has a well defined ground state of the quantum field
and one may assume that at very early times the field is in the ground state
of the static metric. For later times the state is then determined by the field
equations. Hawking used a geometric optics approximation to solve this initial
value problem in the case of the free field equations (VIII.3.1) in the metric pro-
vided by a classical model of stellar collapse. He then found that at late times
the state describes an outgoing thermal radiation corresponding to the temper-
ature (VIII.3.30). Though there is now little doubt that this computation yields
the essential features of the state at late times there remained some uneasiness
342 VIII. Retrospective and Outlook

concerning the corrections and the wish for a more direct understanding of the
claimed universality of the effect.
Several subsequent papers addressed themselves to the simpler situation
of a (spherically symmetric) eternal black hole whose outside region can be
described as Schwarzschild space. For a review see e.g. [Kay 87]. In standard
spherical coordinates r, 0, yo the black hole radius is

r° = 2MG, (VIII.3.31)

and the metric for r > r °

r l 1
ds2 = ( 1 — rD )dt2 — (i. — T ] dr2 —r2 (d19 2 + sin2 z9 d(p2 ). (VIII.3.32)
\ ///
This metric is static in the Schwarzschild time t but the fact that at T = TO
the metric components g,.,. = oo, gtt = 0 means that T = r° is a horizon and
t is not a good coordinate there. There are other coordinate systems which
extend Schwarzschild space to the interior of the black hole. If t is expressed in
such coordinates it becomes ambiguous on the horizon. There is an analogy of
Schwarzschild space with the Rindler space W which may be described as the
wedge in Minkowski space

xi > ix° l; x° , x 2 , x3 arbitrary. (VIII.3.33)

There the boundary x 1 = 1x° 1 is a horizon. An observer in W, sending a signal

to the Minkowski region x 1 < Ix° I can never get an echo back if he is confined
to move within W. The orbit

x° (t) = p Binh t,
(VIII.3.34)
x l (t) = p cosh t
is a uniformly accelerated motion in Minkowski space and pt is the proper time
on this orbit. One can coordinatize W by p, t and one recognizes that the
"time translation" t --> t + a in these coordinates is a Lorentz transformation in
Minkowski space, hence a symmetry. Therefore the metric in W is static with
respect to t. This Rindler time t is analogous to the Schwarzschild time. Remem-
bering the Bisognano-Wichmann theorem V.4.1.1 we know that the Minkowski
vacuum is a thermal state in W with respect to Rindler time translations. The
temperature (270 -1 is analogous to the Hawking temperature and agrees with
it if the accelerations are properly compared. It is amusing to recall that the
papers [Bis 76] and [Hawk 75] resulted from completely disjoint motivations
and the striking parallelism was first noticed in [Sew 80]. Earlier Unruh had
presented arguments showing that a linearly accelerated detector in Minkowski
space on the orbit (VIII.3.34) should respond to the vacuum state similar as one
at rest in a medium filled with Planck radiation at the temperature (27r0) -1 ,
[Unr 76], compare also [Day 75]. This is the Bisognano-Wichmann temperature
scaled to the proper time of the detector. Although these analogies are sugges-
tive they do not suffice to understand the case of an eternal spherical black hole
 VIII.3 The Challenge from General Relativity 343

because in the Rindler case we have more information. Besides the Rindler time
translation there is (on the extension of W to Minkowski space) another time-
like Killing vector field, our usual time translation, and this is used in defining
the vacuum as a reference state. Only in the neighborhood of the horizon the
analogy between Rindler and Schwarzschild is good. This suggests that the dis-
cussion should focus on the local situation of the state in the neighborhood
of the horizon. Furthermore the formation of the black hole by stellar collapse
should not be ignored since a permanent black hole is a physically unrealistic
mental construct.
So we shall return to the case of a spherically symmetric collapse, pay-
ing special attention to the state near the horizon after the black hole forma-
tion. Outside of the star and the black hole one has the Schwarzschild metric
(VIII.3.32). This follows from Birkhoff's theorem which says that a spherically
symmetric metric in a part of space without matter (and hence vanishing Ricci
tensor) is always of the form (VIII.3.32); it has a time-like Killing vector field
defining the Schwarzschild time even if the metric is not static in the total space.
In fig. VIII.3.1. we have used a time coordinate T which remains meaningful on
the horizon. The collapsing star is indicated by the horizontally shaded region,
the black hole by diagonal shading. At T = 0 the radius of the star crosses the
Schwarzschild radius r o , the horizon begins. Also indicated are outgoing light
rays from the neighborhood of the crossing point and the surface t = O. In the
region
T > 0; r > TO, (VII1.3.35)
it will be convenient sometimes to use the "tortoise coordinate"
T
r* -= r + roln ( — 1 ^ (VIII.3.36)
ro `
344 VIII. Retrospective and Outlook

instead of r. The horizon is then moved to r* —oo. For large r the difference
between r* and r is not significant. The time coordinate r used in the figure
was chosen as
T = t + r* — r, (VIII.3.37)
noting that t + r* stays finite on the horizon. We are interested in the partial
state in a region O which is far away from the black hole at a very late time,
centered near a point r = R, t = T so that

T » R; R » ro . (VIII.3.38)

Specifically we are interested in the response of a detector placed in this region.

We may represent it by the observable

C = Q* Q; Q = 4, (h), (VIII.3.39)

with appropriately chosen test function h having support in O. The computation

of the expectation value w(C) is rendered possible by the following remarks [Fred
90]
1) If f is a c-number solution of the covariant Klein-Gordon equation
(VIII.3.1) and 0 satisfies (VIII.3.1) then the integral over a space-like surface
E
Lp(x) aaf (x) as (û) f (x) deg` 0 f (VIII.3.40)

is independent of E. Here deg` is the surface element. Q can be written in the

form (VIII.3.40) with
f (t, x) = f f t' (t, x)dt'
where ft' is the solution of (VIII.3.1) with the Cauchy data

aft (t, x)
ft , (t, x) t=t, = 0;
at = h(t , x) .

t=t,

So ft differs from zero only if t' is in the t-support of h. To verify this note that
in the neighborhood of O, where r and t are very large ro /r is negligible and for
a surface t = const. the surface element de is adequately given in Schwarzschild
coordinates by do-4 = (dv, 0, 0, 0) where dv is the spatial volume element. Then
P f becomes
, x) a ft^^, x) ao^t, x)
f dt' ((t ft, (t, x) dv.

Since ft' is a solution of (VIII.3.1) we can use again the independence of the inte-
gration surface and put t = t'. Then by (VIII.3.42) the above expression reduces
to f 0 (t, x)h(t, x)d 4x = 0(h). If we now want to apply (VIII.3.40) choosing for
E the surface r = 0 we have to solve the Cauchy problem for the solution ft'
of (VIII.3.1) with initial data given by (VIII.3.42) down to the surface r = 0.
Given this we can express w(C) in terms of the 2-point function w (0(x 1 )0(x 2 ))
near the plane T = 0.
 VIII.3 The Challenge from General Relativity 345

2) The solution of (VIII.3.1) with such initial data has, for T > 0, its
support outside the horizon and the star. So we need only the Schwarzschild
metric. Equation (VIII.3.1) gives then, after separating off the angular part,
using r* instead of r, putting ft, = r-1 Y1, 7T1,(t9, (p)1it(r*)
82 a2
Il = 0. (VIII.3.43)
ât2 ar*2 + V (r*)
The shape of the "barrier" V is indicated in fig. VIII.3.2. The height is of
order ((1 + 1/2)/r o ) 2 , the width of order r o . A wave packet moving according
to (VIII.3.43), having its support at t = T around r = R with the relations
(VIII.3.38), will at T = 0 be split into two clearly distinct parts: the first, having
penetrated the barrier, is centered at very large negative values of r* i.e. very
close to the horizon; the other will be at very large positive values of r (or r*).
This is intuitively understood by looking at the analogous barrier penetration
problem in Schrödinger wave mechanics. It is discussed by Dimock and Kay
[Dim 87]. Refined estimates [Fred 90] show that around T = 0 we have
[(r* )—++Y'-+ 4
with supp '+ = [a(T), oo], supp 1p_ _ [—oo, —a(T)] and d tending to zero
uniformly with all its derivatives as T —> oo and a(T) —4 oo for T —> oo.
3) For very large T the counting rate w(C) is thus related to the 2-point
function around T 0 by a term involving only 0_, one involving only 0+
and a cross term which will, however, tend to zero since 0 21 decreases fast for
large space-like separation of the points. The term with 0+ relates to signals
received by outward directed detectors (looking away from the star); this will
be not much affected by the presence of the black hole and we shall not discuss
it. The term with 0_ involves 0 21 only in a short distance neighborhood of the
crossing point T = 0, r = ro . In fact, it gives a scaling limit of w at this point.
So, if we believe that all physically allowed states have the same scaling limit
and that this is obtained by putting w( 2) = Q-1 , we can evaluate w(C) (in the
limit T —> oo). The result is (for T — ^^ oo)
.0 -1
w(C) = f IDr (e)126^1 (es'e — 1) lhi,,n,(e)1 2de. (VIII.3.44)
1,m c°

Fig. VIII.3.2.
346 VIII. Retrospective and Outlook

Here I DI 2 is the barrier penetration factor,

hi,m (e) = const. f h(x)Y, 1T,(z9, v)ew(t- r *) d4x, (VIII.3.45)

0 = 47rro . (VIII.3.46)
To ensure that C represents a detector h has to be chosen so that h(e) has
support only for positive e (see Chapter VI). Then Ihi,, n (e)I 2 is the sensitivity of
the detector to quanta of energy e, angular momentum 1, m. (VIII.3.46) shows
that the asymptotic counting rate is the one produced by an outgoing radiation
of temperature (4irro) -1 , modified by the barrier penetration effect.
The result (VIII.3.44) to (VIII.3.46) is just a corroberation of Hawking's
original prediction. The derivation presented above shows that the effect is in-
dependent of the details of past history up to the formation of the black hole
and relates to the short distance limit of the state at the surface of the black hole
which is — by the assumption of local definiteness — the same for all physically
allowed states. In fact, in the idealization used, the radiation originates from the
surface of the horizon at the moment of its formation (r = 0) and at large times
there is a steady flow of outgoing radiation, not decreasing as T —> oo. This con-
flicts with energy conservation. The energy radiated away must be compensated
by a loss of mass of the star inside the black hole ("black hole evaporation").
This in turn leads to a shrinking of the Schwarzschild radius with increasing T
and thereby the causal structure is changed so that the radiation arriving at
(T, R) relates to points on the horizon at times r(T), increasing with T; thus
the black hole surface at all times (not only at 'r = 0) is responsible. To turn
this qualitative argument into a quantitative one we must determine the back
reaction on the metric due to the energy distribution in the state of the quan-
tum field. The natural approach to this problem would be to use the Einstein
equations (I.2.94), replacing in the outside region T o, (x) by cv (TN,,(0; x)) , the
expectation value of the energy-momentum density of the field in the pre-
vailing state cv. This has led to an extensive literature attempting to define a
"renormalized energy-momentum tensor", starting from the classical expression
for To,(P; x) and the Hadamard form of the 2-point function of the state. It has
been shown that To„ can be defined as an operator valued distribution in the
Wightman sense uniquely up to an unknown c-number function but it is pre-
cisely this c-number function which we need here. For a survey see [ Wald 1984].
The possibility of defining a Wick product may help in resolving this ambiguity.
Let us add a last remark concerning the universality (model independence).
While it appears that the Bekenstein entropy and the Hawking temperature are
completely model independent these quantities do not have direct observational
significance since they concern the black hole surface in relation to the vector
field of Schwarzschild time translation which should properly be regarded as
an asymptotic symmetry. In the transition region between the surface and the
observer the dynamics of the quantum field plays a rôle. In the fr ee field model
it produces the barrier penetration factor. If we envisage instead a realistic
quantum field theory, say the standard model of elementary particles, some of
 VIII.3 The Challenge from General Relativity 347

the above arguments can be adapted. But the determination of the change of
the state in the transition region is a highly non trivial problem in quantum
field theory which cannot be circumvented by thermodynamical arguments.

VIII.3.4 A Few Remarks on Qu antum Gravity

The gravitational constant G, together with h and c, determines a mass, the

Planck mass
hC 112
MP = G = 1 0 ls mproton • (VIII.3.47)

It is the mass for which the Compton wave length and (half of) the Schwarzschild
radius become equal. The corresponding length, the Planck length is
112
h = Gh
^ =10-33 cm.
lp = (VIIL3.48)
Mpc

It appears clear that at distances of the order of 1p a classical metric has no place
and probably the picture of space-time as a 4-dimensional continuum becomes
unreasonable. On the other hand it seems that classical space-time, equipped
with a (classical) causal and metric structure is well established at distances
above 10 -16 cm 4 and that in this regime the principles of local quantum physics,
described in the earlier chapters apply well. The modifications at larger scales,
related to space-time curvature, can be treated in a semiclassical manner along
the lines indicated in subsections 3.2, 3.3. In this treatment there remains still
the problem of a self-consistent description of the back reaction, hinging on the
definition of the expectation value of the energy-momentum tensor in a given
state of the quantum fields.
The crux of the interface of quantum physics and gravitation is, however,
the short distance regime. There, below 10 101p we have no direct guidance from
experiment and cannot expect any. We can speculate and try to produce a
scheme whose merits can be tested by establishing contact with extrapolations
from high energy physics and ultimately by an understanding of the relation of
mass scales in particle physics to the Planck mass. Much work and ingenuity
has been devoted to this quest. We mentioned some lines of approach at the end
of subsection 3.1. One recent proposal [Dopl 951 is to define a non commutative
geometry of x-space by stipulating commutation relations between the coordi-
nates involving the Planck length and restricting the simultaneous precision of
the measurability in different directions. Space-time regions are then replaced
by states over the algebra of the xp .
In a minimal adaptation of the algebraic approach and the locality principle
one could keep the idea of a net of algebras which, however, should be labelled
now by elements of a partially ordered set .0 (instead of regions in IR4). .0

4 This bound is not meant to have physical significance. It just reflects ignorance and might
be pushed down by several orders of magnitude.
348 VIII. Retrospective and Outlook

could be atomic, with minimal elements (atoms) replacing microcells in space-

time. If one dislikes the somewhat artificial picture of arranging these atoms in a
physical lattice with lattice distance 1p in IR4 then a notion of neighborhood can
be introduced, for instance, by requiring that to any atom a there exist two other
atoms a 1 and a2 (neighbors of a) such that b E L with b > a 1 , b > a 2 implies
b > a. If L is a lattice in the mathematical sense (equipped with operations V
and A) then this gives the relation (a 1 Va2)Aa = a, which implies a non-Boolean
structure. In its coarse grained structure L should go over into the Boolean
lattice of regions in IR 4 . The algebras associated to atoms should naturally be
chosen to be all isomorphic to a finite dimensional algebra. Since no work based
on such a picture has yet been done the preceding remarks should be understood
as "an artist's impression of a scenario". The judgment of its worth must be
left to the future.
Bibliography

Alfsen, E. M. Compact convex sets and boundary integrals. Ergebnisse Math.

57, Berlin: Springer 1971
Alfsen, E. M. and Shultz, F. W. Non commutative spectral theory for affine
function spaces on convex sets. Memoirs of the A. M. S. 172, 1976

Arnold, V. I. Mathematical methods of classical mechanics. New York: Springer

1978

Baumgartei, H. Operatoralgebraic Methods in Quantum Field Theory. Berlin:

Akademie Verl. 1995

Baumgartel, H. und Wollenberg, M. Causal Nets of Operator Algebras. - Math-

ematical Aspects of Algebraic Quantum Field Theory. Berlin: Akademie
Verl. 1992

Bell, J. S. Collected papers in quantum mechanics. Speakable and unspeakable

in quantum mechnics. Cambridge Univ. Press 1987

Bergmann, P. G. Introduction to the theory of relativity. Dover: Prentice-Hall

1976

Birrell, N. D. and Davies, P. C. W. Quantum fields in curved space. Cambridge

Univ. Press 1982

Bjorken, J. D. and Drell, S. D. Relativistic quantum fields. New York: McGraw-

Hill 1965

Bogolubov, N. N. and Shirkov, D. V. Introduction to the theory of quantized

fields. London: Interscience 1958

Bogolubov, N. N., Logunov, A. A. and Todorov, L T. Introduction to axiomatic

quantum field theory. Reading, Mass.: Benjamin 1975

Bogolubov, N. N., Logunov, A. A., Oksak, A. I. and Todorov, I. T. General

principles of quantum field theory. Dordrecht: Kluwer 1990

Bohr, N. Atomphysik und menschliche Erkenntnis. Braunschweig: Vieweg Verl.

1985
350 Bibliography

Bratteli, O. and Robinson, D. W. Operator algebras and quantum statistical

mechanics I. C*- and W* -algebras, symmetry groups, decomposition of
states. New York: Springer 1979

Bratteli, O. and Robinson, D. W. Operator algebras and quantum statistical

mechanics II. Equilibrium states models in quantum statistical mechanics.
New York: Springer 1981
Braun, H. and Koecher, M. Jordan algebras. Berlin: Springer 1966

Dixmier, J. Von Neumann algebras. Amsterdam: North-Holland 1981

Dixmier, J. C* -algebras. Amsterdam: North-Holland 1982

Dunford, N. and Schwartz, J. T. Linear operators. Part I: General theory. New

York: Interscience Publishers 1958

Emch, G. E. Algebraic methods in statistical mechanics and quantum field

theory. New York: Wiley 1972

d'Espagnat, B. Reality and the physicist, Cambridge: Cambridge Univ. Press

1989
Ferrara, S. (ed.) Supersymmetry. Vol 1 and 2. Amsterdam: North Holland,
World Scientific 1987

Feynman, R.P. and Hibbs, A. R. Quantum mechanics and path integrals. New
York: McGraw-Hill 1965

Friedrichs, K. O. Mathematical aspects of the quantum theory of fields. New

York: Intersciences Publishers 1953
Gelfand, I. M. and Shilov, G. E. Generalized functions. Vol.I-III. New York:
Academic Press 1964

Gelfand, L M. and Vilenkin, N. Generalized functions. [Link]. New York:

Academic Press 1964

Glimm, J. and Jaffe, A. Quantum physics. A functional integral point of view.

Berlin: Springer 1987
Hamermesh, M. Group theory and its application to physical problems. Read-
ing: Addison-Wesley 1964

Hawking, S. W. and Ellis, G. F. R. The large scale structure of space-time.

Cambridge Univ. Press 1980
Heitler, W. The quantum theory of radiation. London: Oxford University Press
1936

Hepp, H. Théorie de la renormalisation. Berlin: Springer 1968

 Bibliography 351

Horuzhy, S. S. Introduction to algebraic quantum field theory. Dordrecht: Riedel

1989

Iochum, B. Cônes autopolaires et algèbres de Jordan. Lecture Notes in Math-

ematics, Heidelberg: Springer 1984

Israel, R. B. Convexity in the theory of lattice gases. Princeton: Univ. Press

1979

Itzykson, C. and Zuber, J.-B. Quantum field theory. New York: McGraw-Hill
1980

Jauch, J. M. and Rohrlich, F. T. The theory of photons and electrons. The

relativistic quantum field theory of charged particles with spin one-half.
2nd, expanded ed. Berlin: Springer 1976

Jost, R. The general theory of quantized fields. Prividence, R. I.: American

Math. Soc. 1965

Kadanoff, L. P. and Baym, G. Quantum statistical mechanics. New York: Ben-

jamin 1962

Kadison, R. V. and Ringrose, J. R. Fundamentals of the theory of operator

algebras. Vol I: Elementary theory. New York: Academic Press 1983

Kadison, R. V. and Ringrose, J. R. Fundamentals of the theory of operator

algebras. Vol II: Advanced theory. Orlando: Academic Press 1986

Kastler, D. (ed.) The algebraic theory of superselection sectors. Singapore,

• World Scientific 1990

Ludwig, G. An axiomatic basis for quantum mechanics. Heidelberg: Springer

1987

Lopuszanski, J. An introduction to symmetry and supersymmetry in quantum

field theory. Singapore: World Scientfic 1991

Moser, J. Dynamical systems. Theory and applications. Springer: New York

1975

Naimark, M. A. Linear representations of the Lorentz group. Oxford: Pergamon

Press 1964

Naimark, M. A. Normed rings. Groningen: P. Noordhoff 1972

Ohnuki, Y. Unitary representations of the Poincare group and relativistic wave

equations. Singapore: World Scientific 1988

Omnès, R. The Interpretation of Quantum Mechanics. Princeton: Princeton

University Press 1994
352 Bibliography

Pedersen, G. K. C* -algebras and their automorphism groups. London: Aca-

demic Press 1979

Rohrlich F. T. Classical charged particles. Reading, Mass.: Addison-Wesley

1965

Ruelle, D. Statistical physics. New York: Benjamin 1969

Salam, A. and Sezgin, E. (eds.) Supergravities in diverse dimensions. Vol. 1
and 2. Amsterdam: North Holland, World Scientific 1989

Sakai, S. C* -algebras and W* -algebras. Berlin: Springer 1971

Schilpp, P. A. (ed.) Albert Einstein: Philosopher-Scientist. Evanston: The Li-

brary of Living Philosophers 1949

Schwartz, L. Théorie des Distributions. Paris: Hermann 1957

Sewell, G. L. Quantum theory of collective phenomena. Oxford Science Publ.,

Clarendon Press 1989

Sigal, I. M. Scattering theory for many-body quantum mechanical systems.

Lecture Notes in Mathematics Vol. 1011, Springer: Heidelberg 1983

Steinmann, O. Perturbative expansions in axiomatic field theory. Berlin: Sprin-

ger 1971

Stratila, S. Modular theory in operator algebras. Ed. Acad. Bukarest 1981

Stratila, S. and Zsido, L. Lectures on von Neumann algebras. Turnbridge Wells:

Abacus Press 1979

Streater, R. F. and Wightman, A. S. PCT, Spin and statistics and all that.
New York: Benjamin 1964

Takesaki, M. Tomita's theory of modular Hilbert algebras and its application.

Lecture Notes in Mathematics. Berlin: Springer 1970

Takesaki, M. Theory of Operator Algebras. Berlin: Springer 1979

Thirring, W. A course in mathematical physics, Vol. 1-4. Heidelberg: Springer

1981

Varadarajan, V. S. Geometry of quantum theory. New York: Van Nostrand

1968

Van der Waerden, B. Die gruppentheoretische Methode in der Quanten-

mechanik. Berlin: Springer 1932

Wald, R. M. General relativity. Chicago: Univ. of Chicago Press 1984

 Bibliography 353

Warner, F. Foundations of differentiable manifolds and Lie groups. Glenview,

Ill.: Scott, Foresman and Co. 1971

Weinberg, S. Gravitation and cosmology: principles and application of the gen-

eral theory of relativity. New York: Wiley 1972

v. Weizsâcker, C. F. Aufbau der Physik. Munchen: Hanser Verl. 1985

Wentzel, G. Quantum theory of fields. New York: Intersciences Publishers 1949

Wess, J. and Bagger, J. Supersymmetry and supergravity. Princeton: Princeton

Univ. Press 1983

Wheeler, J. A. and Zurek, W. H. (eds.) Quantum Theory and Measurement.

Princeton: Princeton University Press 1983
Whitehead, A. N. Process and Reality, original edition Cambridge 1929

Wigner, E. P. Group theory and its application to the quantum mechanics of

atomic spectra. New York: Academic Press 1959

Wigner, E. P. Symmetries and Reflections. Bloomington: Indiana Univ. Press

1967
Zeh, H. D. Die Physik der Zeitrechnung. Heidelberg: Springer 1984
Author Index and References

Adler, S.
[Adl 77} - Liebermann, J. and Ng, Y. J. Regularisation of the
stress-energy tensor for massless vector particles propa-
gating in a general background metric. Ann. of Phys. 106,
279 (1977) VIII.3.2
[Adl 78] - Liebermann, J. and Ng, Y. J. Trace anomaly of the
stress-energy tensor for massless vector particles propa-
gating in a general background metric. Ann. of Phys. 113,
294 (1978) VIII.3.2
Aharonov, Y.
[Ahar 59] - and Bohm, D. Significance of electromagnetic
potentials in the quantum theory. Phys. Rev. 115, 485
(1959) 1.5.4
[Ahar 64] - Bergmann, P. G. and Lebowitz, J. L. Time sym-
metry in the quantum process of measurements. Phys.
Rev. 134 B, 1410 (1964) VII.1
Alfsen, E. M.
See bibliography
[Alf 781 - and Shultz, F. W. State space of Jordan-Algebras.
Acta Math. 140, 155 (1978) VII.2
[Alf 80] - Hanche -Olsen, H. and Shultz, F. W. State space of
C* -Algebras. Acta Math. 144, 267 (1980) VII.2
Araki, H.
[Ara 611 Connection of spin with commutation rules. J. Math.
Phys. 2, 267 (1961) I1.5.1, IV.4
[Ara 62a1 Einfiihrung in die axiomatische Quantenfeldtheorie.
Lecture Notes ETH Zürich, Hepp, K. (ed.) (1962)
[Ara 62b1- Hepp, K. and Ruelle, D. On the asymptotic behav-
ior of Wightman functions in space-like directions. Helv.
Phys. Acta. 35, 164 (1962) II.4.4
[Ara 63a1 and Woods, E. J. Representations of the canonical
-

commutation relations describing a nonrelativistic infinite

free Bose gas. J. Math. Phys. 4, 637 (1963)
356 Author Index and References

[Ara 63b] A lattice of von Neumann algebras associated with

the quantum theory of a free Bose field. J. Math. Phys. 4,
1343 (1963) 111.4.2
[Ara 64a1 Von Neumann algebras of local observables for free
scalar field. J. Math. Phys.. 5, 1 (1964) III.4.2
[Ara 64b] On the algebra of all local observables. Prog. The-
oret. Phys. 32, 844 (1964) 1II.3.2
[Ara 64c] Type of von Neumann algebras associated to the free
field. Prog. Theoret. Phys. 32, 956 (1964)
[Ara 67] — and Haag, R. Collision cross sections in terms of II.4.1,1I.4.4,
local observables. Commun. Math. Phys. 4, 77 (1967) VI.1.1
[Ara 72] Remarks on spectra of modular operators of von Neu-
mann algebras. Commun. Math. Phys. 28, 267 (1972) V.2.4
[Ara 77a] — and Sewell, G. L. KMS conditions and local ther-
modynamic stability of quantum lattics systems. Commun.
Math. Phys. 52, 103 (1977) V.1.6
[Ara 77b] — and Kishimoto, A. Symmetry and equilibrium
states. Commun. Math. Phys. 52, 211 (1977) V.3.4
[Ara 77cj — Haag, R., Kastler, D. and Takesaki, M. Extension
of KMS states and chemical potential. Commun. Math.
Phys. 53, 97 (1977) V.3.4
[Ara 77d] Relative entropy of states of von Neumann algebras.
Publ. RIMS, Kyoto Univ. 13, 173 (1977)
[Ara 80] On a characterization of the state space of quantum
mechanics. Commun. Math. Phys. 75, 1 (1980) VII.2
[Ara 82] — and Yamagami, S. On quasi—equivalence of quasi-
free states of the canonical commutation relations. Pubi.
RIMS, Kyoto Univ. 18, 238 (1972)
Arnold, V. I.
See bibliography
Ashtekar, A.
See bibliography
[Asht 87] New Hamiltonian formulation of general relativity.
Phys. Rev. D36, 1587 (1987) VIII.3.1
Bagger, J.
See bibliography: Wess
Bakamjian, B.
[Bakam 53] — and Thomas, L. H. Relativistic particle dynam-
ics II. Phys. Rev. 92, 1300 (1953) 1.3.4
Bannier, U.
[Ban 88] Allgemein kovariante algebraische Quantenfeldthe-
orie und Rekonstruktion von Raum-Zeit. Ph. D. Thesis,
Hamburg University (1988) 1II.3.1
 Author Index and References 357

Bargmann, V.
[Barg 47] Irreducible unitary representations of the Lorentz
group. Ann. Math. 48, 568 (1947) I.3.2
[Barg 54] On unitary ray representations of continous groups.
Ann. Math. 59, 1 (1954) I.3.1
Baumann, K.
See [Schmidt 56]
Baumgartel, H.
[Baumg 89] Quasilocal algebras over index sets with a minimal
condition. Operator theory: Adv. and appl. 41, 43 (1989)
Baym, G.
See bibliography: Kadanoff
Bekenstein, J. D.
[Bek 73] Black holes and entropy. Phys. Rev. D7, 2333 (1973) VIII.3.3
Bell, J. S.
See bibliography
[Bell 64] On the Einstein-Podolsky-Rosen paradox. Physics 1,
195 (1964) III.1 VII.3
,

Béllissard, J.
[Belli 78] — and lochum, B. Homogeneous self dual cones
versus Jordan algebras, the theory revisited. Ann. Inst.
Fourier, Grenoble 28, 27 (1978) VII.2
Berezin, F.
[Bere 70] and Kac, G. Mat. Sbornik 82, 331 (1970) (in rus -
sian) VIII.2
Bergmann, P. G.
See bibliography
See [Ahar 84]
Birkhoff, G.
[Birk 36] and von Neumann, J. The logic of quantum mechan-
ics. Ann. Math. 37, 823 (1936) VII.2
Birrell, N. D.
See bibliography
Bisognano, J. J.
[Bis 75] — and Wichmann, E.H. On the duality condition for III.4.2,
a Hermitean scalar field. J. Math. Phys. 16, 985 (1975) V.4.1
[Bis 76] — and Wichmann, E.H. On the duality condition for III.4.2,
quantum fields. J. Math. Phys. 17, 303 (1976) V.4.1
Bjorken, J. D.
See bibliography
Blanchard, Ph.
[Blanch 93] — and Jadzyk, A. On the interaction between clas-
sical and quantum systems. Phys. Lett. A 175, 157 (1993) VII.3
358 Author Index and References

[Blanch 95] - and Jadzyk, A. Event enhanced quantum theory

and piecewise deterministic dynamics. Ann. der Physik 4,
583 (1995) V11.3
Bleuler, K.
[Bleu 50] Eine neue Methode zur Behandlung der longitudi-
nales und skalaren Photonen. Helv. Phys. Acta 23, 567
(1950) I.5.3
Bogolubov, N. N.
See bibliography
[Bog 57] - and Parasiuk, O. S. Uber die Multiplikation der
Kausalfunktionen in der Quantentheorie der Felder. Acta
Math. 97, 227 (1957) I1.2.4
[Bog 581 A new method in the theory of superconductivity I.
Soy hys. JETP 7, 41 (1958) II.1.1
Bohm, D.
See [Ahar 59]
Bohr, N.
See bibliography
[Bohr 49] Discussions with Einstein. In: Schilpp: Albert Ein-
stein: Philosopher-Scientist. The Library of Living Philo-
sophers 1949. VII.1
Bongaarts, P. J. M.
[Bon 70] The electron-position field coupled to external elec-
tromognetic potentials as an elementary C* -algebra the-
ory. Ann. Phys. 56, 108 (1970)
[Bon 77] Maxwell's equations in axiomatic quantum field the-
ory. I. Field tensor and potentials. J. Math. Phys. 18,
1510 (197 )
[Bon 82] Maxwell's equations in axiomatic quantum field the-
ory. II. Covariant and noncovariant gauges. J. Math.
Phys. 23, 1818 (1982)
Borchers, H. J.
[Borch 60] Uber die Mannigfaltigkeit der interpolierenden
Felder zu einer kausalen S-Matrix. Nuovo Cimento 15,
784 (1960) II.5.5
[Borch 62] On the structure of the algebra of the field opera-
tors. Nuovo Cimento 24, 214 (1962) VIII.3.2
[Borch 63] - Haag, R. and Schroer, B. The vacuum state in
quantum field theory. Nuovo Cimento 29, 148 (1963)
[Borth 64] -- and Zimmermann, W. On the self-adjointness of
field operators. Nuovo Cimento 31, 1074 (1964) III.1
[Borch 65a] On The vacuum state in quantum field theory II.
Commun. Math. Phys. 1, 57 (1965) IV.1
[Borth 65b] Local rings and the connection of spin with statis-
tics. Commun. Math. Phys. 1, 291 (1965) IV.1
 Author Index and References 359

[Borch 66] Energy and momentum as observables in quantum

field theory. Commun. Math. Phys. 2, 49 (1966) VI.2.2
[Borch 67] A remark on a theorem of B. Misra. Commun.
Math. Phys. 4, 315 (1967)
[Borch 69a] On the implementability of automorphism groups.
Commun. Math. Phys. 14, 305 (1969)
[Borch 69b1 On groups of automorphisms with semibounded
spectrum. In: Systèmes à un nombre infini de degrés de
liberté. CNRS, Paris (1969)
[Borch 721 — and Hegerfeldt, G. C. The structure of space—
time transformations. Commun. Math. Phys. 28, 259
(1972)
[Borch 84] Translation group and spectrum condition. Com-
mun. Math. Phys. 96, 1 (1984)
[Borch 85] -- and Buchholz, D. The energy—momentum spec-
trum in local field theories with broken Lorentz symmetry.
Commun. Math. Phys. 79, 169 (1985) VI.2.2 •
[Borch 85a4 Locality and covariance of the spectrum. Fizika
17, 289 (1985)
[Borch 90] — and Yngvason, J. Positivity of Wightman func-
tionals and the existence of local nets. Commun. Math.
Phys. 127, 607 (1990) III.1
[Borch 92] CTP theorem in two—dimensional theories of local
observables. Commun. Math. Phys. 143, 315 (1992) V.4.1
[Borch 95] On the use of modular groups in quantum field
theory. Ann. Inst. Henri Poincaré 63, 331 (1995) V.4.1
Borek, R.
[Borek 85] Representations of the current algebra of a charged
massless Dirac field. J. Math. Phys. 26, 339 (1985)
Bratteli, O.
See bibliography
Braun, H.
See bibliography
Brenig, W.
[Bren 59] — and Haag, R. Allgemeine Quantentheorie der
Stossprozesse. Fortschr. Phys. 7, 183 (1959) II.3.1
Bros J.
[Bros 76] — Buchholz, D. and Glaser, V. Constants of motion
in local quantum field theory. Commun. Math. Phys. 50,
11 (1976)
[Bros 88] — and Iagolnitzer, D. Two particle asymptotic com-
pleteness in weakly coupled quantum field theories. Com-
mun. Math. Phys. 119, 331 (1988) VI.2.3
[Bros 94] — and Buchholz, D. Towards a relativistic KMS-
condition. Nucl. Phys. B 429, 291 (1994) V.1.5
360 Author Index and References

Brunetti, R.
[Brun 95] — Fredenhagen, K. and Köhler, M. The microlocal
spectrum condition and Wick polynomials of free fields on
curved spacetimes. Commun. Math. Phys. to appear 1996 VIII.3.2
Buchholz, D.
See [Bros 76], [Bor 85 ]
[Buch 74] Product states for local algebras. Commun. Math.
Phys. 36, 287 (1974) V.5.2
[Buch 77a1 — and Fredenhagen, K. Dilations and interaction.
J. Math. Phys. 18, 1107 (1977) V.4.2
[Buch 77b] Collision theory for massless bosons. Commun.
Math. Phys. 52, 147 (1977) VI.3
[Buch 77c] — and Fredenhagen, K. A note on the inverse scat-
tering problem in quantum field theory. Commun. Math.
Phys. 56, 91 (1977) V.4.2
[Buch 82a] — and Fredenhagen, K. Locality and the structure IV.1, IV.1,
of particle states. Commun. Math. Phys. 84, 1 (1982) IV.3
[Buch 82b1 The physical state space of quantum electrodynam-
ics. Commun. Math. Phys. 85, 49 (1982) VI.3
[Buch 85] — and Epstein, H. Spin and statistics of quantum
topological charges. Fysica 17, 329 (1985) IV.3.4
[Buch 86a] — and Wichmann, E. Causal independence and the
energy—level density of states in local quantum field theory.
Commun. Math. Phys. 106, 321 (1986) V.5
[Buch 86b] — and Junglas, P. Local properties of equilibrium
states and the particle spectrum in quantum field theory.
Lett. Math. Phys. 11, 51 (1986) V.5.1
[Buch 86c] — Doplicher, S. and Longo, R. On Noether's theo-
rem in quantum field theory. Ann. Phys. 170, 1 (1986) V.5.3
[Buch 86d] Gauss' law and the infraparticle problem. Phys.
Lett. B174, 331 (1986) VI.2.1
[Buch 87a] On particles, infraparticles and the problem of
asymoptotic completeness. Proc. TAMP Conf. Marseille, VI.1.1,
World Scientific 1987 VI.1.2
[Buch 87b] — and Jacobi, P. On the nuclearity condition for
massless fields. Lett. Math. Phys. 13, 313 (1987) V.5.1
[Buch 87c] — D'Antoni, C. and Fredenhagen, K. The universal
structure of local algebras. Commun. Math. Phys. 111,
123 (1987) V.5.1, V.5.2
[Buch 88] — Mack, G. and Todorov, I. The current algebra on
the circle as a germ of local field theories. Nucl. Phys. B
(Proc. Suppl. 5B), 20 (1988) IV.5
[Buch 89] — and Junglas, P. On the existence of equilib-
rium states in local quantum field theory. Commun. Math.
Phys. 121, 255 (1989) V.5.1
 Author Index and References 361

[Buch 90a] — D'Antoni, C. and Longo, R. Nuclear maps and

modular structure II. Commun. Math. Phys. 129, 115
(1990) V.5.1, 'V.5.4
[Buch 90b1 Harmonic analysis of local operators. Commun.
Math. Phys. 129, 631 (1990) VI.1.1
[Buch 90c] — and Porrmann, M. How small is the phase space
in quantum field theory? Ann. Inst. H. Poincare 52, 237
(1990) V.5.1
[Buch 91a] — Porrmann, M. and Stein, U. Dirac versus
Wigner: Towards a universal particle concept in local
quantum field theory. Phys. Lett. 267 B, 377 (1991) VI.1.2
[Buch 91b] — and Yngvason, J. Generalized nuclearity condi-
tions and the split property in quantum field theory. Lett.
Math. Phys. 23, 159 (1991) V.5.2
[Buch 92] —Doplicher, S., Longo, R. and Roberts, J. E. A
new look at Goldstone's theorem. Rev. Math. Phys. Spe-
cial Issue, 49 (1992) III.3.2
[Buch 94] On the manifestation of panicles. In: Proc. Beer
Sheva Conf. 1993 Math. Phys. towards the 21st century,
A. N. Sen and A. Gersten (eds.), Ben Gurion of the Negev VI.2.2,
Press 1994 VIII.1
[Buch 95] — and Verch, R. Scaling algebras and renormaliza-
tion group in algebraic quantum field theory. Rev. Math.
Phys. 7, 1195 (1995) VIII.1
Carey, A. L.
[Car 77] — Gaffney, J. M. and Hurst, C. A. A C* —algebraic for-
mulation of the quantization of the electromagnetic field.
J. Math. Phys. 18, 629 (1977)
[Car 78] — Gaffney, J. M. and Hurst, C. A. A C* —algebraic
formulation of gauge transformations of the second kind
for the electromagnetic field. Rep. Math. Phys. 13, 419
(1978)
Coester, F.
[Coest 60] — and Haag, R. Representation of states on a
field theory with canonical variables. Phys. Rev. 117, 1137
(1960) II.1.1
[Coest 65] Scattering theory for relativistic particles. Helv.
Phys. Acta. 38, 7 (1965) I.3.4
Colemann, S.
[Col 67] — and Mandula, J. All possible symmetries of the II1.3.2,
S-Matrix. Phys. Rev. 159, 1251 (1967) VII.2
Connes, A.
[Conn 73] Une classification des facteurs de type III. Ann.
Sci. Ecole Norm. Sup. 6, 133 (1973) V.2.4
362 Author Index and References

[Conn 74] Caractérisation des espaces vectoriels ordonnés

sous-jacents aux algèbres de von Neumann. Ann. Inst.
Fourier, Grenoble 24(4), 121 (1974) VI1.2, V.2.4
Cuntz, J.
[Cuntz 77] Simple C* - algebras generated by isometrics. Com-
mun. Math. Phys. 57, 173 (1977) IV.4
D'Antoni, C.
See [Buch 87c, 90a]
Davies, E. B.
[Day 70] — and Lewis, J. T. An operational approach to quan-
tum propability. Commun. Math. Phys. 17, 239 (1970)
Davies, P. C. W.
See bibliography: Birrell
[Day 75] Scalar particle production in Schwarzschild and V.4.1,
Rindler metrics. J. Phys. A8, 609 (1975) VII.3.3
Dell'Antonio, C.
[Dell'Ant 61] On the connection of spin with statistics. Ann.
Phys. 16, 153 (1961) II.5.1
Dimmock, J.
[Dim 87] — and Kay, B. S. Classical and quantum scatter-
ing theory for linear scalar on the Schwarzschild metric I.
Ann. of Phys. 175, 366 (1987) VIII.3.3
Dirac, P. A. M.
[Dir 38] Classical theory of radiating electrons. Proc. Roy.
Soc. (London) A 167, 148 (1938) I.2.2
Dixmier, J.
See bibliography
Doplicher, S.
See [Buch 86c, 92]
[Dopl 65] An algebraic spectrum condition. Commun. Math.
Phys. 1, 1 (1965) I1I.3.2
[Dopl 66] — Kastler, D. and Robinson, D. W. Covariance al-
gebras in field theory and statistical mechanics. Commun.
Math. Phys. 3, 1 (1966)
[Dopl 67] — Kadison, R. V., Kastler, D. and Robinson, D. W.
Asymptotically Abelian systems. Commun. Math. Phys. 6,
101 (1967) V.3.4
[Dopl 68a] — and Kastler, D. Ergodic states in a non commu-
tative ergodic theory. Commun. Math. Phys. 7, 1 (1968)
[Dopl 68b] -- Regge, T. and Singer I. M. A geometrical model
showing the independence of locality and positivity of the
energy. Commun. Math. Phys. 7, 51 (1968) I1.5.4
[Dopl 69a] — Haag, R. and Roberts, J. E. Fields, observables
and gauge transformations I. Commun. Math. Phys. 13,
1 (1969) IV.1
 Author Index and References 363

[Dopl 69b] — Haag, R. and Roberts, J. E. Fields, observables

and gauge transformations II. Commun. Math. Phys. 15, IV.1, IV.2,
173 (1969) IV.4
[Dopl 71] — Haag, R. and Roberts, J. E. Local observables and
particle statistics I. Commun. Math. Phys. 23, 199 (1971) IV.1, IV.2
[Dopl 74] — Haag, R. and Roberts, J. E. Local observables and
particle statistics II. Commun. Math. Phys. 35, 49 (1974) IV.1, IV.2
[Dopl 82] Local aspects of superselection rules. Commun.
Math. Phys. 84, 105 (1982) V.5.3
[Dopl 83] — and Longo, R. Local aspects of superselection rules
II. Commun. Math. Phys. 88, 399 (1983) V.5.3
[Dopl 84a] — and Longo, R. Standard and split inclusions of
von Neumann algebras. Invent. Math. 73, 493 (1984) V.5.1, V.5.2
[Dopl 84b] — and Roberts, J. E. Compact Lie groups asso-
ciated with C* —Algebras. Bull. Am. Math. Soc. 11, 333
(1984)
[Dopl 88] — and Roberts, J. E. Compact group actions on
C* -algebras. J. Operator Theory 19, 283 (1988) IV.4
[Dopl 89a1 — and Roberts, J. E. Endomorphisms of C*-
algebras, cross products and duality for compact groups.
Ann. of Math. 130, 75 (1989) IV.4
[Dopl 89b] — and Roberts, J. E. Monoidal C* -categories and
a new duality theory for compact groups. Invent. Math.
98, 157 (1989) IV.4
[Dopl 90] — and Roberts, J. E. Why there is a field algebra
with a compact gauge group describing the superselection
structure in particle physics. Commun. Math. Phys. 131,
51 (1990) IV.3.3, IV.4
[Dopl 95] — Fredenhagen, K. and Roberts, J. E. The struc-
ture of spacetime at the Planck scale and quantum fields.
Commun. Math. Phys. 172, 187 (1995) VIII.3.4
Drell, S. D.
See bibliography: Bjorken
Driessler, W.
[Driess 74] On the type of local algebras in quantum field the-
ory. Commun. Math. Phys. 53, 295 (1974) V.6
[Driess 79] Duality and the absence of locally generated super-
selection sectors for CCR—type algebras. Commun. Math.
Phys. 70, 213 (1979)
[Driess 86] — Summers, S. J. and Wichmann, E. H. On the
connection between quantum fields and von Neumann al-
gebras of local operators. Commun. Math. Phys. 105, 49
(1986) II1.1
364 Author Index and References

Drinfeld, V. G.
[Drin 87] Quantum Groups. In Proc. ICM, Berkeley 1986,
Academic Press 1987 IV.5
Drühl, K.
[Drühl 70] — Haag, R. and Roberts, J. E. On parastatistics.
Commun. Math. Phys. 18, 204 (1970) IV.1
Dunford, N.
See bibliography
Dyson, F. J.
[Dy 49] The S-Matrix in quantum electrodynamics. Phys.
Rev. 75, 1736 (1949) I.5.5, II.2.4
[Dy 67] — and Lenard, A. Stability of matter I. J. Math. Phys.
8, 423 (1967) V.1.4
[Dy 68] — and Lenard, A. Stability of matter II. J. Math.
Phys. 9, 698 (1968) V.1.4
[Dy 92] Quantum Past: The Limitations of Quantum Theory.
Schrödinger Lecture May 92, Imperial College London.
unpublished VI I.1
Eckmann, J. P.
[Eck 73] — and Osterwalder, K. An application of Tomita's
theory of modular Hilbert algebras: duality for free Bose
algebras. Funct. An. 13, 1 (1973)
Edwards, C. M.
[Edw 70] The operational approach to algebraic quantum field
theory. Commun. Math. Phys. 16, 207 (1970)
Ekstein, H.
[Ek 56a] Theory of time dependent scattering for multichannel
processes. Phys. Rev. 101, 880 (1956) II.3.1
[Ek 56b] Scattering in field theory. Nuovo Cimento 4, 1017
(1956) II.3.1
[Ek 69] Presymmetry II. Phys. Rev. 184, 1315 (1969) I.1, V.3.3, VIII.1
Ellis, G. F. R.
See bibliography: Hawking
Emch, G. E.
See bibliography
Enss, V.
[Enss 75] Characterisation of particles by means of local ob-
servables. Commun. Math. Phys. 45, 35 (1975) VI.2.1
[Enss 83] Asymptotic observables on scattering states. Com-
mun. Math. Phys. 89, 245 (1983) VI.2.3
Epstein, H.
See [Buch 85]
[Ep 63] On the Borchers class of a free field. Nuovo Cimento
27, 886 (1963) II.5.5
 Author Index and References 365

[Ep 67] CTP-Invariance of the S-Matrix in a theory of local

observables. J. Math. Phys. 8, 750 (1967) IV.2.4
[Ep 69] — Glaser, V. and Martin, A. Polynomial behaviour
of scattering amplitudes at fixed momentum transfer for
theories with local observables. Commun. Math. Phys. 13,
257 (1969) 11.5.2
[Ep 71] — and Glaser, V. Le role de la localité dans la renor-
malisation pertubative en théorie quantique des champs.
In Statistical Mechanics and Quantum Field Theory.
Stora, R. and DeWitt, C. (eds.), London: Gordon and
Breach (1971) II.2.3, II.2.4
d'Espagnat, B.
See bibliography
Evans, D. E.
[Ev 80] A review of semigroups of completely positive maps.
In: Mathematical problems in theoretical physics. Oster-
walder, K. (ed.), Berlin: Springer 1980
Ezawa, H.
[Ez 67] — and Swieca, J. A. Spontaneous breakdown of sym-
metries and zero-mass states. Commun. Math. Phys. 5,
330 (1967) III.3.2
Faddeev, L.
[Fadd 71] — and Kulish, P. P. Asymptotic conditions and
infrared divergences in quantum electrodynamics. Theor.
Math. Phys. 4, 745 (1971) VI.2.1
[Fadd 90] Integrable models for quantum groups. In: Fields
and particles. Proc. Schladming 1990, Mitter, H., Schwei-
ger, W. (eds.), Springer: Berlin 1990 IV.5
Fannes, M.
[Faun 77a] — and Verbeure, A. Correlation inequalities and
equilibrium states. Commun. Math. Phys. 55, 125 (1977) V.1.6
[Faun 77b] — and Verbeure, A. Correlation inequalities and
equilibrium states II. Commun. Math. Phys. 57, 165
(1977) V.1.6
Federbush, P.
[Fed 68] Local operator algebras in the presence of superselec-
tion rules. J. Math. Phys. 9, 1718 (1968)
Fell, J. M. G.
[Fell 60] The dual spaces of C* -algebras. Trans. Amer. Math.
Soc. 94, 365 (1960) III.2.2
Ferrara, S.
See bibliography
Feynman, R.
See bibliography
366 Author Index and References

[Feyn 48] Space-time approach to non-relativistic quantum

mechanics. Rev. Mod. Phys. 20, 367 (1948) 1.4, VII.1
Finkelstein, D.
[Fink 62] - Jauch, J.M., Schiminovich, S. and Speiser, D.
Foundations of quaternion quantum mechanics. J. Math.
Phys. 3, 207 (1962) VI1.2
[Fink 68] Space-time code. Phys. Rev. 184, 1261 (1968) VII.3
Foulis, D. J.
[Foul 60] Baer * semigroups. Proc. Amer. Math. Soc. 11, 648
(1960) V11.2
Fredenhagen, K.
See [Brun 96], [Buch 77a, 77c, 82a, 87c], [Dopl 95]
[Fred 81a] On the existence of antiparticles. Commun. Math.
Phys. 79, 141 (1981) IV.3.4
[Fred 81b] - and Hertel, J. Local algebras of observables and
pointlike localized fields. Commun. Math. Phys. 80, 555
(1981) 111.1
[Fred 85a] On the modular structure of local algebras of ob-
servables. Commun. Math. Phys. 97, 79 (1985) V.6
[Fred 85b] A remark on the cluster theorem. Commun. Math.
Phys. 97, 461 (1985) II.4.4
[Fred 87] - and Haag, R. Generally covariant quantum field
theory and scaling limits. Commun. Math. Phys. 108, 91 111.3.1,
(1987) VI11.3.2
[Fred 89a] - Rehren, K. H. and Schroer, B. Superselection
sectors with braid group statistics and exchange algebras
I. Commun. Math. Phys. 125, 201 (1989) IV.5
[Fred 89b] Structure of superselection sectors in low di-
mensional quantum field theory. Proceedings Lake Tahoe
Conf. 1989. Plenum Publ. Corp. 1990 IV.5
[Fred 90] - and Haag, R. On the derivation of Hawking radi-
ation associated with the formation of a black hole. Com-
mun. Math. Phys. 127, 273 (1990) V1II.3.3
[Fred 92] - Rehren, H. and Schroer, B. Superselection sec-
tors with braid group statistics and exchange algebras II.
Geometric aspects and conformal covariance. Rev. Math.
Phys. Special issue 1992, 113 (1992) IV.5
Friedrichs, K. O.
See bibliography
Frohlich, J.
[Fish 79a] - Morchio, G. and Strocchi, F. Infrared problem
and spontaneous breaking of the Lorentz group in QED.
Phys. Lett. 89B, 61 (1979) VI.2.1
 Author Index and References 367

[Froh 79b] - Morchio, G. and Strocchi, F. Charged sectors

and scattering states in quantum electrodynamics. Ann.
of Phys. 119, 241 (1979) VL2.1, VL3
[Freih 79c] The charged sectors of quantum electrodynamics in
a framework of local observables. Commun. Math. Phys.
66, 223 (1979) VI.3
[Friih 88] Statistics of fields. The Yang-Baxter equation and
the theory of knots and links. In: Nonperturbative quan-
tum field theory. G. `t Hooft, A. Jaffe, G. Mack, P. K.
Mitter and R. Stora (eds.), Plenum Publishing Corpora-
tion(1988) IV.5
[Froh 89] - and Marchetti, P. A. Quantum field theories of
vortices and anyons. Commun. Math. Phys. 121, 177
(1989) IV.5
[Frôh 90] - and Gabbiani, F. Braid statistics in local quantum
theory. Rev. Math. Phys. 2, 251 (1990) IV.5
[Frôh 91] - and Kerler, T. Zürich, preprint 1991 IV.5
Fulling, S. A.
[Full 73] Non uniqueness of canonical field quantization in
Riemannian space-time. Phys. Rev. D7, 2850 (1973) VIII.3.2
[Full 81a] - Sweeny, M. and Wald, R. M. Singularity structure
of the two point function in quantum field theory in curved
space time. Commun. Math. Phys. 63, 257 (1981) VIII.3.2
[Full 81b] - Narcowich, F. J. and Wald, R. M. Singularity
structure of the two-point function in quantum field theory
in curved space time II. Ann. of Phys. 136, 243 (1981) VIII.3.2
Gabbiani, F.
See [Frdh 90]
Gaffney, J. M.
See [Car 77, 78]
Garber, W. D.
[Garb 78a] - and Reeh, H. Nontranslationally covariant cur-
rents and associated symmetry generators. J. Math. Phys.
19, 597 (1978) VIII.2
[Garb 78b] - and Reeh, H. Nontranslationally covariant cur-
rents and symmetries of the S-Matrix. J. Math. Phys. 19,
985 (1978) VIII.2
Garding, L.
[Gard 54a] - and Wightman, A.S. Representations of the an-
ticommutation relations. Proc. Nat. Acad. Sci. 40, 617
(1954) 1I.1.1
[Gard 54b] - and Wightman, A.S. Representations of the com-
mutation relations. Proc. Nat. Acad. Sci. 40, 622 (1954) II.1.1
Gelfand, I. M.
See bibliography
368 Author Index and References

Cell-Mann, M.
[Cell 51] - and Low, F. Bound states in quantum field theory.
Phys. Rev. 84, 350 (1951) II.2.4
[Gell 90] - and Hartle, J. B. In: Complexity, Entropy and the
Physics of Information. W. Zurek (ed.). Reading: Addison
Wesley 1990, and Proc. of the 25th International Conf. on
High Energy Physics, Singapore 1990. K. K. Phua and Y.
Yamaguchi (eds.), Singapore: World Scientific 1990
[Gell 94] - and Hartle, J. B. Proc of the NATO- Workshop
"On the physical origins of time asymmetry". J. Halliwell,
J. Perez-Mercader and W. Zurek (eds.), Cambridge Univ.
Press 1994 VII.3
Glaser, V.
See [Ep 69, 71], [Bros 76]
[Glas 57] - Lehmann, H. and Zimmermann, W. Field op-
erators and retarded functions. Nuovo Cimento 6, 1122
(1957) II.2.6
Glimm, J.
See bibliography
[Glimm 61] Type I C* -algebras. Ann. Math. 73, 572 (1961) II.1.1
Goldstone, J.
[Gold 61] Field theories with "superconductor" solutions.
Nuovo Cimento 19, (1961) 1II.3.2
Graf, G. M.
[Graf 90] Asymptotic completeness for N-body short range
quantum systems: a new proof. Commun. Math. Phys.
132, 73 (1990) VI.2.3
Green, H. S.
[Green 53] A generalized method of field quantization. Phys.
Rev, 90, 270 (1953) I.3.4, IV.1
Greenberg, O. W.
[Greenb 64] - and Messiah, A. M. L. Symmetrization postulate
and its experimental foundation. Phys. Rev. 136B, 248
(1964) IV.1
[Greenb 65] - and Messiah, A. M. L. Selection rules for
parafields and the absence of para particles in nature.
Phys. Rev. 138B, 1155 (1965) IV.1
Grifiths, R. B.
[Gruff 84] Consistent histories. J. Stat. Phys. 36, 219 (1984) VII.3
Guido, D.
[Guido 95a] - and Longo, R. An algebraic statistics theorem.
Commun. Math. Phys. 172, 517 (1995) IV.5
[Guido 95b] - and Longo, R. The conformal spin and statistics
theorem. Preprint Dipart. Mat. Univ. Rome Tor Vergata
1995 IV.5
 Author Index and References 369

Gunson, J.
[Gun 67] On the algebraic structure of quantum mechnanics.
Commun. Math. Phys. 6, 262 (1967)
Gupta, S. N.
[Gup 50] Theory of longitudinal photons in quantum electro-
dynamics. Proc. Phys. Soc. (London) A 63, 681 (1950) I.5.3
Haag, R.
See [Bren 59], [Coest 60], [Borch 63], [Ara 67], [Dopl 69a,
69b], [Drühl 70], [Dopl 71, 74], [Ara 77c], [Fred 87, 90]
[Haag 54] Lecture Notes Copenhagen, CERN T/RH1 53/54 I1.1.1, II.1.2
[Haag 55a] On quantum field theories. Dan. Mat. Fys. Medd.
29 no. 12 (1955) II.1.1, II.1.2
[Haag 55b] Die Selbstwechselwirkung des Elektrons. Z. Natur-
forsch. 10a, 752 (1955) 1.2.2
[Haag 57] Discussion des "axiomes" et des propriétés asymp-
totiques d'une théorie des champs locale avec particules
composées. In: Les problèmes mathématique de la théorie
quantique des champs (Lille 1957). Paris: Centre National
de la Recherche Scientifique 1959 II.4.1, II1.1
[Haag 58] Quantum field theories with composite particles and
asymptotic conditions. Phys. Rev. 112, 669 (1958) I1.4.2, 1I.4.4
[Haag 62a] -- and Schroer, B. Postulates of quantum field the-
ory. J. Math. Phys. 3, 248 (1962) II1.1, 11I.4.2
[Haag 62b] The mathematical structure of the Bardeen—Coo-
per— Schrieffer model. Nuovo Cimento 25, 287 (1962) III.3.2
[Haag 63] Bemerkungen zum Nahwirkungsprinzip in der
Quantenphysik. Ann. d. Physik 11, 29 (1963) III.4.2
[Haag 64] — and Kastler, D. An algebraic approach to quantum
field theory. J. Math. Phys. 5, 848 (1964) I.1, II1.1
[Haag 65] — and Swieca, J. A. When does a quantum field
theory describe particles? Commun. Math. Phys. 1, 308
(1965) V.5.1
[Haag 67] — Hugenholtz, N. M. and Winnink, M. On the equi-
librium states in quantum statistical mechanics. Commun.
Math. Phys. 5, 215 (1967) V.1.1
[Haag 70] — Kadison, R. V. and Kastler, D. Nets of C*_
alge bras and classification of states. Commun. Math.
Phys. 16, 81 (1970)
[Haag 73] — Kadison, R. V. and Kastler, D. Asymptotic orbit
of states of a free Fermi gas. Commun. Math. Phys. 33,
1 (1973)
[Haag 74] — Kastler, D. and Trych—Pohlmeyer, E. B. Stabil-
ity and equilibrium states. Commun. Math. Phys. 38, 173
(1974) V.3.2
370 Author Index and References

[Haag 75] -- Lopuszanski, J. and Sohnius, M. All possible gen-

erators of supersymmetries of the S—Matrix. Nucl. Phys.
B 88, 257 (1975) VIIL2
[Haag 77] — and Trych—Pohlmeyer, E. B. Stability proper-
ties of equilibrium states. Commun. Math. Phys. 56, 214
(1977) V.3.2
[Haag 84] — Narnhofer, H. and Stein, U. On quantum field the- III.3.1,
ories in gravitational background. Commun. Math. Phys. V.3.2,
94, 219 (1984) VIII.3.2
[Haag 90a] Thoughts on the synthesis of quantum physics and
general relativity and the role of space-time. Nucl. Phys.
B (Proc. Suppl.) 18B, 135 (1990) VII.3
[Haag 90b] Fundamental irreversibility and the concept of I.1, VII.3,
events. Commun. Math. Phys. 132, 245 (1990) VIII.3
[Haag 93] Events, histories, ireversibility. In: Proc. Conf. on
Quantum Control and Measurement. H. Ezawa and Y.
Murayama (eds.), Elsevier Science Publ. 1993 VII.3
[Haag 96] — and Ojima, I. On the problem of defining a specific
theory within the frame of local quantum physics. RIMS
preprint, Kyoto Univ. 1996 VIII.1
Haagerup, U.
[Haager 87] Connes' bicentralizer problem and uniqueness of
the injective factor of type 11I I . Acta Math. 158, 95 (1987) V.6
Hadjiivanov, L. K.
[Hadj 90] — Paunov, R. R. and Todorov, I.T. Quantum group
extended chiral p—models. Nucl. Phys. B 356, 387 (1991) IV.5
Hagedorn, R.
[Hagedorn 67] On the hadronic mass spectrum. Nuovo Ci-
mento 52A, 1336 (1967) V.5.1
Hamermesh, M.
See bibliography
Hanche-Olsen, H.
See [Alf 80]
Hartle, J. B.
See [Gell 90, 95]
Hawking, S. W.
See bibliography
[Hawk 76] Particle creation by black holes. Commun. Math. V.4.1,
Phys. 43, 199 (1975) VIII.3.3
Hegerfeldt, G. C.
See Borch 72
Heitler, W.
See bibliography
Hepp, K.
See [Ara 62b], bibliography
 Author Index and References 371

[Hepp 65] On the connection between the LSZ and Wightman

quantum field theory. Commun. Math. Phys. 1, 95 (1965) II.4.4
[Hepp 72] Quantum theory of measurement and macroscopic
observables. Helv. Phys. Acta 45, 237 (1972)
[Hepp 73] - and Lieb, E. H. Helv. Phys. Acta 46, 573 (1973) VII.1
Hertel, J.
See [Fred 81]
Hertel, P.
[Hert 72] - Narnhofer, H. and Thirring, W. Thermodynamic
functions for Fermions with gravostatic and electrostatic
interactions. Commun. Math. Phys. 28, 159 (1972) V.1.4
Hibbs, A. R.
See bibliography: Feynman
Hislop, P. D.
[Hisl 82] - and Longo, R. Modular structure of the local alge-
bras associated with the free massless scalar field theory.
Commun. Math. Phys. 84, 71 (1982) V.4.2
Horuzhy, S. S.
See [Poli 731, bibliography
van Hove, L.
[Hove 52] Les difficultés de divergences pour un modèle par-
ticulier de champs quantifié. Physica 18, 159 (1952) II.1.1
Hugenholtz, N. M.
See [Haag 67]
[Hug 671 On the factor type of equilibrium states in quantum
statistical mechanics. Commun. Math. Phys. 6, 189 (1967)
[Hug 75] - and Kadison, R. V. Automorphisms and quasifree
states of the CAR-algebra. Commun. Math. Phys. 43, 181
(1975)
Hunziker, W.
[Hunz 68] S-matrix in classical mechanics. Commun. Math.
Phys. 8, 282 (1968) VI.2.3
Hurst, C. A.
See [Car 77, 78]
Iagolnitzer, D.
See [Bros 88]
[Iag 87] Asymptotic completeness and multiparticle analysis
in quantum field theories. Commun. Math. Phys. 110, 51
(1987) VI.2.3
Iochum, B.
See [Belli 78], bibliography
Israel, R. B.
See bibliography
Itzykson, C.
See bibliography
372 Author Index and References

Jacobi, P.
See [Buch 87b]
Jadczyk, A. Z.
See [Blanch 93, 95]
[Jad 69] On spectrum of internal symmetries of the algebraic
quantum field theory. Commun. Math. Phys. 12, 58 (1969)
Jaffe, A.
See bibliography: Glimm
[Jaffe 67] High energy behavior in quantum field theory I.
Strictly localizable fields. Rev. Phys. 158, 1454 (1967) V.5.2
Jauch, J. M.
See [Fink 62], bibliography
[Jauch 69] — and Piron, C. On the structure of quantal propo-
sition systems. Helv. Phys. Acta 42, 842 (1969) VII.2
Jones, V. F. R.
[Jones 83] Index for subfactors. Invent. Math. 72, 1 (1983) IV.5
Jordan, P.
[Jord 34] — von Neumann, J. and Wigner, E. P. On an alge-
braic generalization of the quantum mechanical formula-
tion. Ann. Math. 35, 29 (1934) VII.2
Jost, R.
See bibliography
[Jost 57] Eine Bemerkung zum CTP-Theorem. Helv. Phys.
Acta. 30, 409 (1957) 11.5.1
Junglas, P.
See [Buch 86b, 89]
[Jung 87] Therrnodynamisches Gleichgewicht und Energie-
spektrum in der Quantenfeldtheorie. Thesis, University
Hamburg 1987 V.5.1
Junker, W.
[Junk 95] Adiabatic vacua and Hadamard states for scalar
quantum fields on curved spacetime. Thesis Hamburg 1995 VIII.3.2
Kac, G.
See [Bere 70]
Kadanoff, L. P.
See bibliography
Kadison, R. V.
See [Dopl 67], [Haag 70, 73], [Hug 75], bibliography
[Kad 65] Transformation of states in operator theory and dy-
namics. Topology 3, Suppl. 2, 177 (1965)
[Kad 67] The enërgy—momentum spectrum of quantum fields.
Commun. Math. Phys. 4, 258 (1967)
Kahn, B.
[Kahn 38] and Uhlenbeck, G. E. On the theory of condensa-
tion. Physica 5, 399 (1938) 1I.2.2
 Author Index and References 373

Kamefuchi, S.
See [Ohnu 68], [Ohnu 69]
Kastler, D.
See [Haag 64], [DopI 66, 67, 68a], [Haag 70, 73, 74], [Ara 77c],
bibliography
[Kest 66] - Robinson, D.W. and Swieca, J. A. Conserved
currents and associate symmetries; Goldstone's theorem.
Commun. Math. Phys. 2, 108 (1966) III.3.2
[Kast 66] - and Robinson, D.W. Invariant states in statistical
mechanics. Commun. Math. Phys. 3, 151 (1966)
[Kast 72] - Loupias, G., Mebkhout, M. , and Michel, L. Cen-
tral decomposition of invariant states. Commun. Math.
Phys. 2'T, 192 (1972)
[Kast 76] Equilibrium states of matter and operator algebras.
Sympos. Math. 20, 49 (1976) V.3.4
[Kast 79] - and Takesaki, M. Group duality and KMS-states.
Commun. Math. Phys. 70, 193 (1979)
[Kast 82] - and Takesaki, M. Group duality and KMS-states
II. Commun. Math. Phys. 85, 155 (1982)
Kay, B.
See [Dim 87]
[Kay 87] - and Wald, R. M. Some recent developments related
to the Hawking effect. In: Proceedings Int. Conf. on Dif-
ferential Geom. Methods in Theoret. Phys., Doebner, H.
D. and Hennig, J. D. (ed.)., Singapore: World Scientific VIII.3.2,
1987 VIII.3.3
Kerler, T.
See [Fra 91]
Kibble, T. W. B.
[Kibble 68a] Mass shell singularities of Green's functions.
Phys. Rev. 173, 1527 (1968) VI.2.1
[Kibble 68b] Asymptotic states. Phys. Rev. 174, 1882 (1968) VI.2.1
Kishimoto, A.
See [Ara 77b]
Knight, J. M.
[Knight 61] Strict localisation in quantum field theory. J.
Math. Phys. 2, 459 (1961) V.5.1
Koecher, M.
See bibliography: Braun
Köhler, M.
[Kohl 95] New examples for Wightman fields on a manifold.
Class. Quantum Gray. 12, 1413 (1995) VIII.3.2
Kosaki, H.
[Kos 86] Extension of Jones' theory on index to arbitrary fac-
tors. J. Func. Anal. 66, 123 (1986) IV.S
374 Author Index and References

Kraus, K.
[Kraus 77] — Polley, L. and Reents, G. Models for infrared
dynamics I. Classical currents. Ann. Inst. Henri Poincare
26, 109 (1977) VI.3
Kubo, R.
[Kubo 57] Statistical mechanical theory of irreversible pro-
cesses L J. Math. Soc. Japan 12, 570 (1957) V.1.1
Kulish, P. P.
See [Fadd 71]
Landau, L. J.
[Land 69] A note on extended locality. Commun. Math. Phys.
135, 246 (1969) 1I1.4.2
[Land 87] On the violation of Bell's inequality in quantum
theory. Phys. Lett. 120, 54 (1987)
Landshoff, P. V.
[Lands 67] — and Stapp, H. P. Parastatistics and a unified the-
ory of identical particles. Ann. Phys. (N.Y.) 45, 72 (1967) IV.1
Laporte, O.
[Lap 31] — and Uhlenbeck, G. E. Application of spinor anal-
ysis fo the Maxwell and Dirac equations. Phys. Rev. 37,
1380 (1931) 1.2.1
Lebowitz, J. L.
See [Ahar 64]
Lehmann, H.
See [Glas 57]
[Leh 54] Ober Eigenschaften von Ausbreitungsfunktionen und
Renormierungskonstanten quantisierter Felder. Nuovo Ci-
mento 11, 342 (1954) 11.3.3
[Leh 55] — Symanzik, K. and Zimmermann, W. Zur For-
mulierung quantisierter Feldtheorien. Nuovo Cimento 1,
425 (1955) 11.1.2, 11.3.3
[Leh 57] — Symanzik, K. and Zimmermann, W. On the for-
mulation of quantized field theories II. Nuovo Cimento 6,
320 (1957) I1.2.6
Lenard, A.
See [Dy 67, 68]
Lewis, J. T.
See [Car 77, 78]
Licht, A. L.
[Licht 63] Strict localisation. J. Math. Phys. 4,1443 (1963) V.5.1
Lieb, E. H.
See [Hepp 73]
[Lieb 75a] — and Thirring, W. Bound for the kinetic energy of
fermions wich proves the stability of matter. Phys. Rev.
 Author Index and References 375

Lett. 35, 687 (1975) and erratum Phys. Rev. Lett. 35,
1116 (1975) V.1.4
Liebermann, J.
See [Adl 77, 78]
Logunov, A. A.
See bibliography: Bogolubov
Longo, R.
See [Hisl 82], [Dopl 83, 84], [Buch 86c, 90a, 92], [Guido 95a,
95b]
[Longo 89] Index of subfactors and statistics of quantum
fields. Commun. Math. Phys. 126, 217 (1989) IV.5
[Longo 90] Index of subfactors and statistics of quantum fields
II. Commun. Math. Phys. 130, 285 (1990) IV.5
[Longo 92] Minimal index and braided subfactors. J. Funct.
Anal. 101, 98 (1992) IV.5
[Longo 95] — and Rehren, H. Nets of subfactors. Rev. Math.
Phys. 7, 567 (1995) IV.5
Lopuszanski, J.
See [Haag 75], bibliography
Loupias, G.
See [Kart 72]
Low, F.
See [Gell 51]
Ludwig, G.
See bibliography
Lüders, Ch.
[Lad 90] — and Roberts, J. E. Local quasiequivalence and
adiabatic vasuum states. Commun. Math. Phys. 134, 29
(1990) VIIL3.2
Mack, G.
See [Buch 88]
[Mack 90] — and Schomerus, V. Conformal field algebras with
quantum symmetry from the theory of supers election sec-
tors. Commun. Math. Phys. 134, 139 (1990) IV. 5
[Mack 91a] — and Schomerus, V. Quasi Hopf symmetry in
quantum theory. Nucl. Phys. B 370, 185 (1992) IV. 5
[Mack 91b] — and Schomerus, V. Quasi quantum group sym-
metry and local braid relations in conformal Ising model.
Phys. Lett. 267 B, 207 (1991) IV. 5
Mackey, G. W.
[Mackey 57] Borel structure in groups and their duals. Trans.
Amer. Math. Soc. 85, 134 (1957) II.1.1
Magid, S.
[Maj 89] Quasitriangular Hopf algebras and Yang Baxter
equation. Int. J. Mod. Phys. A 5, 1 (1990) IV.5
376 Author Index and References

Maison, D.
[Maison 71] — and Reeh, H. Properties of conserved local cur-
rents and symmetry transformations. Nuovo Cimento 1A,
78 (1971)
Mandula, J.
See [Col 67]
Marchetti, P. A.
See [Frdh 89]
Martin, A.
See [Ep 69]
Martin, P. C.
[Mart 59] -- and Schwinger, J. Theory of many particle sys-
tems: I. Phys. Rev. 115, 1342 (1959) V.1.1
Mayer, J. E.
[May 37] The statistical dynamics of condensing systems I. J.
Chem. Phys. 5, 67 (1937) 11.2.2
Mebkhout, M.
See [Kast 72]
Messiah, A. M. L.
See [Greenb 64, 65]
Michel, L.
See [Kast 72]
[Mich 65] Relations between internal symmetry and relativis-
tic invariance. Phys. Rev. 137B, 405 (1965) VIII.2
Mielnik, B.
[Miel 69] Theory of filters. Commun. Math. Phys. 15, 1 (1969) VII.2
[Miel 74] Generalized quantum mechanics. Commun. Math.
Phys. 37, 221 (1974) VII.2
Misra, B.
[Mis 65] On representations of Haag fields. Helv. Phys. Acta.
38, 189 (1965)
Morchio, G.
See [Frdh 79a, 79b]
[Morch 85] — and Strocchi, F. Spontaneous symmetry breaking
and energy gap generated by variables at infinity. Com-
mun. Math. Phys. 99, 153 (1985) I11.3.2
[Morch 87] — and Strocchi, F. Mathematical structure for long
range dynamics and symmetry breaking. J. Math. Phys.
28, 622 (1987) 111.3.2
Moser, J.
See bibliography
Mourre, E.
[Mour 79] Link between the geometric and the spectral trans-
formation approaches in scattering theory. Commun. Math.
Phys. 68, 91 (1979) I.1
 Author Index and References 377

Naimark, M. A.
See bibliography
Narcowich, F. J.
See [Full 81a, 81b]
Narnhofer, H.
See [Hert 72], [Haag 84]
[Narn 81] — and Sewell, W. Vlasov hydrodynamics of a
quantum mechanical model. Commun. Math. Phys. 79,
9 (1981) V.1.4
[Narn 82] — and Thirring, W. On the adiabatic theorem in
quantum statistical mechanics. Phys. Rev. A26/6, 3646
(1982) V.3.2
[Narn 91] — and Thirring, W. Gallilei invariant quantum field
theories with pair interaction (review). Int. J. Mod. Phys.
A6, 2937 (1991) V.1.4
von Neumann, J.
See [Jord 34] and [Birk 36]
Newton, T. D.
[New 49] — and Wigner, E. P. Localized states for elementary
systems. Rev. Mod. Phys. 21, 400 (1949) 1.3.3
Ng, Y. J.
See [Adl 77, 78]
Nishijima, K.
[Nish 57] On the asymptotic condition in quantum field the-
ory. Prog. Theor. Phys. 17, 765 (1957) 11.2.6
[Nish 58] Formulation of field theory of composite particles.
Phys. Rev. 111, 995 (1958) 11.4.4
Ocneanu, A.
[Ocn 88] Paragroups. In: Lecture note series 135. Operator
algebras and applications Vol. 2, p. 119. Cambridge 1988 IV.5
Ohnuki, Y.
See bibliography
[Ohnu 68] — and Kamefuchi, S. Some general properties of
para-Fermi field theory. Phys. Rev. 170, 1279 (1968) IV.1
[Ohnu 69] — and Kamefuchi, S. Wavefunctions of identical
particles. Ann. Phys. (N.Y.) 51, 337 (1969) IV.1
Ojima, I.
See [Haag 96]
[Oj 86] Lorentz invariance vs. temperature. Lett. Math. Phys.
11, 73 (1986) V.1.5
Oksak, A. I.
See bibliography: Bogolubov
Omnes, R.
[Omn 90] From Hilbert space to common sence. Ann. of Phys.
201, 357 (1990) VI.2.3
378 Author Index and References

O'Raifeartaigh, L.
[O'Raif 65] Mass differences and Lie algebras of finite order.
Phys. Rev. Lett. 14, 575 (1965) VIII.2
Orzalesi, C.
[Orz 70] Charges as generators of symmetry transformations
in quantum field theory. Rev. Mod. Phys. 42, 381 (1970) VIII.2
Osterwalder, K.
See [Eck 73]
[Ost 73] — and Schrader, R. Axioms for Euclidean Green's
functions, I. Commun. Math. Phys. 31, 83 (1973) II.2.7
[Ost 75] — and Schrader, R. Axioms for Euclidean Green's
functions, II. Commun. Math. Phys. 41, 281 (1975) II.2.7
Parasiuk, O. S.
See [Bog 57]
Paunov, R. R.
See [Hadj 90]
Pedersen, G. K.
See bibliography
Peierls, R.
[Pei 52] The commutation laws of relativistic field theory.
Proc. Roy. Soc. (London), A 214, 143 (1952) I.4
Penrose, R.
[Pen 67] Twistor algebra. J. Math. Phys. 8, 345 (1967) I.2.1
Pimsner, M.
[Pim 86] — and Popa, S. Entropy and index for subfactors.
Ann. Sci. Ecole Norm. Sup. 19, 57 (1986) IV.5
Piron, C.
See [Jauch 69]
[Pir 64] Axiomatique quantique. Helv. Phys. Acta 37, 439
(1964) I.1
Pohlmeyer, K.
[Pohl 72] The equation curl W = 0 in quantum field theory.
Commun. Math. Phys. 25, 73 (1972)
Polivanov, M. K.
[Poli 73] — Sushko, V. N. and Horuzhy, S. S. Axioms of ob-
servable algebras and the field concept. Theor. Math. Phys.
16, 629 (1973)
Polley, L.
See [Kraus 77]
Pool, J. C. T.
[Pool 68a] Baer * semigroups and the logic of quantum me-
chanics. Commun. Math. Phys. 9, 118 (1968) VII.2
[Pool 68b] Semimodularity and the logic of quantum mechan-
ics. Commun. Math. Phys. 9, 212 (1968) VII.2
 Author Index and References 379

Popa, S.
See [Pim 86]
Porrmann, M.
See [Buch 90c, 91a]
Powers, R.
[Powers 67] Representation of uniformly hyperfinite algebras
and associated von Neumann rings. Ann. Math. 86, 138
(1967) V.2.4
Pusz, W.
[Pusz 78] and Woronowicz, S. L. Passive states and KMS
states for general quantum systems. Commun. Math.
Phys. 58, 273 (1978) V.3.3
Radzikowski, M. J.
[Radz 92] The Hadamard Condition and Kay's conjecture in
axiomatic quantum field theory on curved spacetime. PhD
thesis, Princeton Univ. 1992 V1H.3.2
Reeh, H.
See [Maison 711, [Garb 78a, 78b]
[Reeh 61] — and Schlieder, S. Bemerkungen zur Unitdriiquiva-
lenz von Lorentzinvarianten Feldern. Nuovo Cimento 22,
1051 (1961) 11.5.2
Reents, G.
See [Kraus 771
Regge, T.
See [Dopl 68]
Rehren, K. H.
See [Fred 89a, 92], [Longo 95]
[Rehr 88] Locality of conformal fields in two dimensions: Ex-
change algebras on the light cone. Commun. Math. Phys.
116, 675 (1988) IV.5
[Rehr 89] — and Schroer, B. Einstein causality and Artin
braids. Nucl. Phys. B312, 715 (1989) IV.5
[Rehr 91] Field Operators for anyons and plektons. Commun.
Math. Phys. 145, 123 (1992) IV.5
[Rehr 95] On the range of the index of subfactors. J. Funct.
Anal. 134, 183 (1995) IV.5
Rindler, W.
[Rind 66] Kruskal space and the uniformly accelerated frame.
Am. J. Phys. 34, 1174 (1966) IV.4.1
Ringrose, J. R.
See bibliography: Kadison
Roberts, J. E.
See [Buch 92], [Dopl 69a, 69b, 71, 74, 84b, 88, 89, 90, 95],
[Drühl 70], [Lüd 90]
380 Author Index and References

Robinson, D. W.
See [Kast 66, 66a], [Dopl 67], bibliography: Bratteli
Roepstorff, D. W.
[Roep 76] Correlation inequalities in quantum statistical me-
chanics and their application in the Kondo problem. Com-
mun. Math. Phys. 46, 253 (1976) V.1.6
[Roep 77] A stronger version of Bogolubov's inequality and the
Heisenberg model. Commun. Math. Phys. 53, 143 (1977) V.1.6
Rohrlich, F. T.
See bibliography: Jauch and Rohrlich, Rohrlich
[Rohr 61] The equation of motion of classical charges. Ann.
Phys. (N.Y.) 13, 93 (1961) 1.2.2
Roos, H.
[Roos 70] Independence of local algebras in quantum field the-
ory. Commun. Math. Phys. 16, 238 (1970)
Ruelle, D.
See [Ara 62b], bibliography
[Rue 61] Connection between Wightman functions and Green
functions in p-space. Nuovo Cimento 19, 356 (1961) 11.2.7
[Rue 62] On the asymptotic condition in quantum field theory.
Helv. Phys. Acta. 35, 147 (1962) 1I.4.1, 11.4.2
[Rue 66] States of physical systems. Commun. Math. Phys.
3, 133 (1966)
Sadowski, P.
[Sad 71] — and Woronowicz, S. L. Total sets in quantum field
theory. Rep. Math. Phys. 2, 113 (1971)
Sakai, S.
See bibliography
Salam, A.
See bibliography
Schimonovich, S.
See [Fink 62]
Schlieder, S.
See [Reeh 61]
Schmidt, W.
[Schmidt 56] and Baumann, K. Quantentheorie der Felder als
Distributionstheorie. Nuovo Cimento 4, 860 (1956) II.1.2
Schomerus, V.
See [Mack 90, 91a, 91b]
Schrader, R.
See [Ost 73, 75]
Schroer, B.
See [Haag 62a], [Borth 63], [Fred 89a, 92], [Rehr 89]
[Schroer 63] Infrateilchen in der Quantenfeldtheorie. Fortschr.
Phys. 173, 1527 (1963) V1.2.1
 Author Index and References 381

[Schroer 74] — and Swieca,J. A. Conformal transformations

for quantized fields. Phys. Rev. D 10, 480 (1974) V.4.2
Schwartz, L.
See bibliography
Schwinger, J.
See [Mart 59]
[Schwing 59] Euclidean quantum electrodynamics. Phys. Rev.
115, 721 (1959) 11.2.7
Segal, 1. E.
[Seg 47] Postulates for general quantum mechanics. Ann. 1.1,
Math. 48, 930 (1947) 11I.1, II1.2.2
[Seg 57] Caractérisation mathématique des observables en
théorie quantique des champs et ses conséquences pour la
structure des particules libres. In: Les problèmes mathé-
matique de la théorie quantique des champs (Lille 1957).
Paris: Centre National de la Recherche Scientifique 1959 11I.1
Sewell, G. L.
See [Ara 77], [Narn 81], bibliography
[Sew 80] Relativity of Temperature and the Hawking—Effekt.
Phys. Rev. Lett. 79A, 23 (1980) V11I.3.3
Sezgin, E.
See bibliography
Shilov, G. E.
See bibliography: Gelfand
Shirkov, D. V.
See bibliography: Bogolubov
Shultz, F. W.
See [Alf 78], [Alf 80], bibliography: Alfsen
Sigal, I. M.
See bibliography
Singer, I. M.
See [Dopl 68]
Sohnius, M.
See [Haag 75]
Speiser, D.
See [Fink 62]
Stapp, H. P.
See [Lands 67]
[Stapp 77] Theory of Reality, Found. of Phys. 7, 313 (1977)
[Stapp 79] Whiteheadian Approach to Quantum Theory and
Generalized Bell's Theorem. Found. of Phys. 9, 1 (1979)
Stein, U.
See [Haag 84], [Buch 91a]
[Stein 89] Thesis, Hamburg Univ. 1989 V1.1.2
382 Author Index and References

Steinmann, O.
See bibliography
[Steinm 91] Asymptotic completeness in QED. 1. Quasilocal
states. Nucl. Phys. B350, 355 (1991) VI.2.3
Stôrmer, E.
[Storm 72] Spectra of states and asyptotically abelian C-
algebras. Commun. Math. Phys. 28, 279 (1972) V.2.4
Stolt, R. H.
[Stolt 70] and Taylor, J. R. Classification of paraparticles.
Phys. Rev. D1, 2226 (1970) IV.1
Stratila, S.
See bibliography
Streater, R. W.
See bibliography
[Streat 68] On certain non-relativistic quantized fields. Com-
mun. Math. Phys. 7, 93 (1968) V.1.4
[Streat 70] and Wilde, I. F. Fermion states of a Boson field.
Nucl. Phys. B24, 561 (1970) IV.5
Strocchi, F.
See [Frôh 79a, 79b], [Morch 85, 87]
Summers, S. J.
See [Driess 86]
[Summ 87a] From algebras of local observables to quantum
fields: generalized H bounds. Helv. Phys. Acta 60, 1004
(1987)
[Summ 87b] — and Werner, R. Maximal violation of Bell's
inequalities is generic in quantum field theory. Commun.
Math. Phys. 110, 247 (1987)
Sushko, V. N.
See [Poli 73]
Sweeny, M.
See [Full 81a, 81b]
Swieca, J. A.
See [Haag 65], [Kast 66], [Ez 67], [Schroer 74]
[Swieca 67] Range of forces and broken symmetries in many-
body systems. Commun. Math. Phys. 4, 1 (1967) 111.3.2
[Swieca 73] — and Voelkel, A. H. Remarks on conformal in-
variance. Commun. Math. Phys. 29, 319 (1973) V.4.2
Symanzik, K.
See [Leh 55], [Leh 57]
[Sy 66a] Euclidean quantum field theory I. Equation for a
scalar model. J. Math. Phys. 7, 510 (1966) 11.2.7
[Sy 66b] Euclidean quantum field theory. In: Local quantum
theory. Scuola Internaz. di Fisica "Enrico Fermi", Corso
[Link], R. (ed.), New York: Acadamic Press 1969 II.2.7
 Author Index an d References 383

[Sy 67] Many particle structure of Green's functions. In: Sym-

posia on Theoretical Physics, Vol. 3. Ramakrishnan, A.
(ed.), New York: Plenum Press 1967 11.2.5
[Sy 70] Small distance behaviour in field theory and power
counting. Commun. Math. Phys. 18, 227 (1970) 11.2.4
[Sy 72] Small distance behavior anlysis and Wilson expansion.
Commun. Math. Phys. 23, 49 (1971) 11.2.4
Takesaki, M.
See [Ara 77c], [Kast 79], [Kast 82], bibliography
Taylor, J. R.
See [Stolt 701
Thirring, W.
See [Hert 72], [Lieb 754 [Narn 82, 91], bibliography
Thomas, L. H.
See [Baka 531
Todorov, I. T.
See [Buch 88], [Hadj 90], bibliography: Bogolubov
Trych—Pohlmeyer, E. B.
See [Haag 74, 77]
Uhlenbeck, G. E.
See [Kahn 38], [Lap 31]
Uhlmann, A.
[Uhl 62] Uber die Definition der Quantenfelder nach Wight-
man und Haag. Wiss. Zeits. Karl Marx Univ. 11, 213
(1962) VII1.3.2
Unruh, W. G.
[Unr 76] Notes on black hole evaporation, Phys. Rev. D14, V.4.1,
870 (1976) VII1.3.3
Ursell, H. P.
[Urs 27] The evaluation of Gibbs' phase-integral for imperfect
gases. Proc. Cambr. Phil. Soc. 23, 685 (1927) 11.2.2
Varadarajan, V. S.
See bibliography
de Vega, H. J.
[de Vega 90] Integrable theories, Yang-Baxter algebras and
quantum groups. Advanced Studies in Pure Mathematics
19, 567 (1989) IV.5
Verbeure, A.
See [Faun 77a, 77b]
Verch, R.
See [Buch 96]
[Verch 94] Local definiteness, primarity and quasiequivalence
of quasifree Hadamard quantum states in curved space-
time. Commun. Math. Phys. 160, 507 (1994) V111.3.2
384 Author Index and References

Verlinde, E.
[Ver 88] Fusion rules and modular transformations in 2D con-
formal field theory. Nucl. Phys. B3OO, 360 (1988) IV.5
Vilenkin, N.
See bibliography: Gelfand
Vinberg, E. B.
[Vin 65] The structure of the group of automorphisms of a
homogeneous convex cone. Trans. Mosc. Math. Soc. 13,
63 (1965) I.1
Voelkel, A. H.
See [Swieca 73]
Volkov, D. V.
[Volk 65] S-Matrix in the generalized quantisation method.
Soy. Phys. JETP 11, 375 (1965) IV.1
Van der Waerden, B.
See bibliography
Wald, R. M.
See [Full 81a, 81b1, [Kay 87], bibliography
[Wald 77] The back reaction effect in particle creation in
curved spacetime. Commun. Math Phys. 54, 1 (1977) VIII.3.2
[Wald 78] Trace anomaly of a conformally invariant quantum
field theory. Rev. Phys. D17, 1477 (1978) VIII.3.2
Warner, F.
See bibliography
Weinberg, S.
See bibliography
v. Weizsàcker, C. F.
See bibliography
[Weiz 73] Probability and quantum mechanics. Br. J. Phil.
Sci. 24, 321 (1973) VII.3
Wentzel, G..
See bibliography
Wenzl, H.
[Wenzl 88] Hecke algebras of type A n, and subfactors. Invent.
Math. 92, 349 (1988) IV.5
Werner, R.
See [Summ 87b]
Wess, J.
See bibliography
[Wess 74] — and Zumino, B. Supergauge transformations in
four dimensions. Nucl. Phys. B 70, 39 (1974) VIII.2
Wheeler, J. A.
See bibliography
Wichmann, E. H.
See [Bis 75, 76], [Buch 86], [Driess 86]
 Author Index and References 385

Wick, J. C.
[Wick 50] The evaluation of the collision matrix. Phys. Rev.
80, 268 (1950) I.5.2
[Wick 52] — Wightman, A. S. and Wigner, E. H. The intrinsic
parity of elementary particles. Phys. Rev. 88, 101 (1952) III.1
Wightman, A. S.
See [Wick 52], [Gard 54a, 54b], bibliography: Streater
[Wight 56] Quantum field theories in terms of vacuum expec-
tation values. Phys. Rev. 101, 860 (1956) 11.2.1
[Wight 57] Quelque problèmes mathématique de la théorie
quantique relativiste. In: Lecture notes, Faculté des Sci-
ences, Univ. de Paris 1957 and in: Les problèmes mathé-
matique de la théorie quantique des champs (Lille 1957).
Paris: Centre National de la Recherche Scientifique 1959 11.1.2
[Wight 64] — and Garding,L. Fields as operator valued distri-
butions in relativistic quantum theory. Ark. Fys. 23, Nr.
13 (1964) 11.1.2
Wigner, E. P.
See [Jord 34], [New 49], [Wick 52], bibliography
[Wig 39] On unitary representations of the inhomogeneous
Lorentz group. Ann. Math. 40, 149 (1939) I.3.1, I.3.2
[Wig 63] Remarks on the mind-body question. In: Symmetries
and Reflections. Bloomington: Indiana Univ. Press 1967 VII.1
[Wig 78] Interpretation of Quantum Mechanics. In: Quantum
Theory of Measurement. J. A. Wheeler and W. H. Zurek
(eds.) VII.3
Wilczek, F.
[Wilt 82] Quantum mechanics of fractional spin particles.
Phys. Rev. Lett 49, 957(1982) IV.5
Wilde, I. F.
See [Streat 70]
Winnink, M.
See [Haag 67]
de Witt, B. S.
[de Witt 80] Quantum gravity: the new synthesis. In: General
Relativity. Hawking, S. W. and Israel, W. (eds.) V.4.1
Wollenberg, M.
[Wol 85] On the relation between quantum fields and local
algebras of observables. Rep. Math. Phys. 22, 409 (1985)
Woods, E. J.
See [Ara 63b]
Woronowicz, S. I.
See [Sad 71], [Pusz 78]
[Wor 87] Compact matrix pseudogroups. Commun. Math.
Phys. 111, 613 (1987) IV.5
386 Author Index and References

Yamagami, S.
See [Ara 82]
Yngvason, J.
See [Bor 91]
Zimmermann, W.
See [Leh 55], [Leh 57], [Glas 57], [Bor 64]
[Zi 581 On the bound state problem in quantum field theory.
Nuovo Cimento 10, 597 (1958) 11.4.4
[Zi 70] In: Brandeis Univ. Summer Inst. in Theoret. Phys.,
Vol. 1, Deser, Grisam, Pendleton (eds.), MIT—Press 1970 11.2.4
Zsido, L.
See bibliography: Stratila
Zuber, J. — B.
See bibliography: Itzykson
Zumino, B.
See [Wess 74]
Zurek, W. H.
See bibliography
Subject Index

Abstract algebra 109, 118 commutation relations 2, 54

Action principle 39 implementation of translations
Affine connection 23 283
Algebra dimension of field 337
abstract 109, 118 Causal completion 143
Banach 118 Causal link 309
C*- 112 Causality 9, 57
von Neumann 113 Center 116
of observables 109 Center of mass 38
polynomial 84, 105 Charge conjugation 99, 167ff, 183
W*- 127 Charge quantum numbers 154
Almost local 84 Chemical potential 200, 239ff
Analyticity 60ff Classical approximation 299
Annihilation operator 37 Cluster 61
Antiparticle 48, 99 Cluster property 103, 134
Antiunitary 27 Coboundary 19
Asymptotically Abelian 226, 229 Cocycle 19
Asymptotic completeness 80, 271ff Coherent subspace 108
Autocorrelation inequalities 214 Cohomology 19
Axiomatic quantum field theory 58 Coincidence arrangement 92, 271
Collective coordinates 299
Banach algebra 118 Collision theory 76ff, 84ff
Bekenstein entropy 346 Compactness requirement 255
Bell's inequality 107, 316 Complementarity 295
BF-localization 174ff Conformal group 13
Bisognano-Wichmann theorem 245 Conformal invariance 249
Black hole evaporation 346 Connection, affine 23
Boost 245 Convex
Borchers class 104 body 7
Borchers-Uhlmann algebra 334 cone 7, 308
Bose-Fermi alternative 35, 97, 149ff Correlation functions 62
Bose statistics 35 Correspondence principle 295
Braid group 192 Cotangent
vector 11
C*-algebra 106, 112 bundle 11, 22
Canonical Covariant differentiation 24
anticommutation relations 37, 54 Covering group 12
388 Subject Index

CPT Feynman diagrams 67

operator 96, 99 Feynman path integral 42, 324
theorem 99 Field algebra 184
Creation operator 37 Field bundle 172
Cross sections 81, 278 Field-particle duality 46
Cuntz algebra 186 Fock space 36
Curvature 24 Folium of states 124
Cyclic 123 Fredenhagen-Hertel condition 257
Free field 46
Decoherence 300
Density operator 4 Gauge group (global) 184ff
Detector 92, 271 non Abelian 187
DHR-analysis 156ff Gauge invariance (local) 21, 50
Diamond 111 Gauge principle (local) 325
Dimension, relative 117 Gauge transformation 50
Dirac equation 33 Gelfand ideal 123
Disjoint representations 126 Gell-Mann and Low formula 65
Dispersion relation 101 General covariance 22
Distributions 56 General relativity 21ff
Doplicher ideal 135 GNS-construction 122
Double cone 111 Geodesic 39
DR-category 189 Germ 326ff
Duality relation 145 Germ of theory 335, 341
Dynamical law 58 Gibbs ensembles 199ff
Dynamical stability 227 Gibbs phase rule 214
Graded Lie algebra 330
Einstein equations 25
Einstein causality 9 Haag's theorem 55
Einstein-Podolsky-Rosen paradox 107, Hadamard form 335
316 Hagedorn temperature 256
Endomorphism, implementation of 185 Hawking temperature 248, 341ff
Energy-momentum Heisenberg cut 3, 295
operators 56 Helicity 31
spectrum 56, 102 Hislop-Longo theorem 251
tensor 24, 25 History 311, 313, 321
Entropy 205, 214
Equilibrium 199ff Implementable 110
Exchange algebra 197 Incoming fields 80
Expectation value 5, 7 Indeterminacy 294
Extremal invariant state 133 Infraparticle 280
Extremal KMS-state 212 Intermediate type I factor 256
Event 2, 3, 289, 309ff Interpretation of quantum mechanics
2
Face 7 Intertwiner 159
Factor 116 Irreducible 28
Factor type 117, 118, 224ff Irreversibility 312
Fermi statistics 35 Isomorphism 115
Feynman amplitudes 66 Isoparallel 23, 39
 Subject Index 389

Jones index 195 Nuclearity requirement 255

Jordan algebra 307 Nuclearity, modular 265
Jost points 61
Observables 1ff, 25
KMS-condition 199ff Observation procedure 7, 324
Observer 3
Lagrangean 39 Open systems 263
Linked cluster theorem 63 Operation 84
Little Hilbert space 30 Operator valued distribution 56
Local Osterwalder-Schrader theorem 73
definiteness 130 Outgoing fields 80
gauge invariance 21
gauge transformation 50 Parastatistics 35, 150
normality 131 Parallel transport 23
stability 234, 338 Particle 75ff, 271ff
Locality 7 localization of 33
Localizability of charges 153, 177ff single 32
Localization number 275 Passivity 235ff
Localized state 102, 174, 254 Pauli principle 35
Localized morphism 156 Perturbation theory 51, 65ff
Lorentz force 20 Phase, thermodynamic 212
Lorentz group 10ff Phase space 254ff
LSZ-formalism 81 Planck length 347
Plekton 194
Majorana field 48 Poincaré group 8ff
Measurement 3, 300 representations of 28ff
Microlocal Poisson bracket, Peierls' method 40, 41
analysis 335 Polynomial algebras 84, 105
spectrum condition 335, 340 Primary representation 126
Mixture of states 7 Primary state 126, 134
Modular Primitive causality 58
automorphism 217 Projective representation 27
conjugation 217 Projector 4
Hamiltonian 218 Proposition 4
nuclearity 265 Pure state 4
operator 217
Quantization 6, 302
Nahwirkungsprinzip 7 Quantum electrodynamics 48, 66ff
Natural cone 221 Quantum field 42f, 56
Net of algebras 105 fre e 46
von Neumann algebra 106, 113 local 46
von Neumann ring 113 Quantum group 197
Newton- Wigner localization 32 Quantum logic 307
Noether's theorem 42 Quantum mechanics 1ff
Normal order 45 Quasiequivalent 126
Normal states 128
Nuclear maps 257 Radiation reaction 21
Nuclearity index 256 Radon-Nikodym cocycle 224
390 Subject Index

Reality 298 Statistics

Reconstruction theorem 59 dimension 169, 194
Reduction formulas 83 operator 162, 192
Reeh-Schlieder theorem 101 parameter 168, 194
Reflection positivity 75 phase 194
r-functions 73 , infinite 170, 183
Relative modular operator 222 Supergravity 332
Relativity, special 8 Superposition principle 5, 108, 305
Relativity, general 21ff Superselection rules 108
Renormalization 65ff Superselection sector 108, 154
Retarded functions 73 Supersymmetry 329ff
Rindler space 342 Symmetry 26, 41, 132
Symmetry breaking 132ff, 212
Saturation 211 System 3
Scaling 328f
Scaling limit 329, 336 Tangent bundle 11, 22
Schwarzschild metric 343 Tangent space 11, 22, 336
Schwinger functions 73ff 7-functions 64
Second law of thermodynamics 237 Thermal equilibrium 199ff
Second quantization 46 Thermodynamic limit 199
Sector 154 Thermodynamics, second law 237
Single particle states 31 Time slice axiom 58
S-matrix 80 Tomita-Takesaki theorem 217
analyticity of 100 Truncated functions 62
Spectrum 119 Twistor 17
Spin 31 Type of factors 117, 118, 224
Spin-statistics theorem 97 Universal type 267
Spinor 12
Split property 256, 258 Vacuum 56, 137
Stability, dynamical 227ff Vector state 124
Stability, local 338 Vertex functions 71
Stability group 30
State 3, 26, 122 Wave front set 339
pure 4 Weight 278, 282
mixed 4 Wick product 45
primary 126, 134 Wightman axioms 56ff
Statistical independence 258 Wightman functions 59ff

Yang-Baxter equation 196