Copyright © 2000. Oxford University Press. All rights reserved.

May not be reproduced in any form without permission from the publisher, except fair uses permitted
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 OXFORD MATHEMATICAL MONOGRAPHS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Series Editors

J. M. BALL E. M. FRIEDLANDER W. T. GOWERS

N. J. HITCHIN I. G. MACDONALD L. NIRENBERG
R. PENROSE J. T. STUART
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 OXFORD MATHEMATICAL MONOGRAPHS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A. Belleni-Moranti: Applied semigroups and evolution equations

A.M. Arthurs: Complementary variational principles 2nd edition
M. Rosenblum and J. Rovnyak: Hardy classes and operator theory
J.W.P. Hirschfeld: Finite projective spaces of three dimensions
A. Pressley and G. Segal: Loop groups
D.E. Edmunds and W.D. Evans: Spectral theory and differential operators
Wang Jianhua: The theory of games
S. Omatu and J.H. Seinfeld: Distributed parameter systems: theory and applications
J. Hilgert, K.H. Hofmann, and J.D. Lawson: Lie groups, convex cones, and semigroups
S. Dineen: The schwarz lemma
S.K. Donaldson and P.B. Kronheimer: The geometry of four-manifolds
D.W. Robinson: Elliptic operators and Lie groups
A.G. Werschulz: The computational complexity of differential and integral equations
L. Evens: Cohomology of groups
G. Effinger and D.R. Hayes: Additive number theory of polynomials
J.W.P. Hirschfeld and J.A. Thas: General Galois geometries
P.N. Hoffman and J.F. Humpherys: Projective representations of the symmetric groups
I. Gyori and G. Ladas: The oscillation theory of delay differential equations
3. Heinonen, T. Kilpelainen, and O. Martio: Non-linear potential theory
B. Amberg, S. Franciosi, and F. de Giovanni: Products of groups
M.E. Gurtin: Thermomechanics of evolving phase boundaries in the plane
I. Ionescu and M. Sofonea: Functional and numerical methods in viscoplasticity
N. Woodhouse: Geometric quantization 2nd edition
U. Grenander: General pattern theory
J. Faraut and A. Koranyi: Analysis on symmetric cones
I.G. Macdonald: Symmetric functions and Hall polynomials 2nd edition
B.L.R. Shawyer and B.B. Watson: Borel's methods of summability
M. Holschneider: Wavelets: an analysis tool
Jacques Thevenaz: G-algebras and modular representation theory
Hans-Joachim Baues: Homotopy type and homology
P.D.D'Eath: Black holes: gravitational interactions
R. Lowen: Approach spaces: the missing link in the topology-uniformity-metric traid
Nguyen Dinh Cong: Topological dynamics of random dynamical systems
J.W.P. Hirschfeld: Projective geometries over finite fields 2nd edition
K. Matsuzaki and M. Taniguchi: Hyperbolic manifolds and Kleinian groups
David E. Evans and Yasuyuki Kawahigashi: Quantum symmetries on operator algebras
Norbert Klingen: Arithmetical similarities: prime decomposition and finite group theory
Isabelle Catto, Claude Le Bris, and Pierre-Louis Lions: The mathematical theory of thermo-
dynamic limits: Thomas-Fermi type models
D. McDuff and D. Salamon: Introduction to symplectic topology 2nd edition
under U.S. or applicable copyright law.

William M. Goldman: Complex hyberbolic geometry

Charles J. Colbourn and Alexander Rosa: Triple systems
V.A. Kozlov, V.G. Maz'ya and A.B. Movchan: Asymptotic analysis of fields in multi-structures
Gerard A. Maugin: Nonlinear waves in elastic crystals
George Dassios and Ralph Kleinman: Low frequency scattering
Gerald W. Johnson and Michel L. Lapidus: The Feynman integral and
Feynman's operational calculus
W. Lay and S.Y. Slavyanov: Special functions: A unified theory based on singularities
D. Joyce: Compact manifolds with special holonomy
A. Carbone and S. Semmes: A graphic apology for symmetry and implicitness
Johann Boos: Classical and modern methods in summability
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
Nigel Higson and John Roe: Analytic K-Homology
DISTRITAL FRANCISCO JOSE DE CALDAS
S. Semmes:
AN: 98476Some novel types
; Johnson, of fractal
Gerald geometry
W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Tadeusz IwaniecCalculus
Operational and Gaven Martin: Geometric function theory and nonlinear analysis
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The Feynman Integral and

Feynman's Operational Calculus
Gerald W. Johnson
Department of Mathematics and Statistics
University of Nebraska-Lincoln

Michel L. Lapidus
Department of Mathematics
University of California, Riverside
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
CLA
DISTRITAL FRANCISCO JOSERDEE NDON PRESS • OXFORD
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus 2000
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

OXFORD
UNIVERSITY PRESS
Great Clarendon Street, Oxford 0x2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Bangkok Buenos Aires Cape Town Chennai
Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata
Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
Sao Paulo Shanghai Singapore Taipei Tokyo Toronto
with an associated company in Berlin
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© G. W. Johnson and M. L. Lapidus, 2000
The moral rights of the authors have been asserted
Database right Oxford University Press (maker)
First published 2000
First published in paperback 2002
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose this same condition on any acquirer
A catalogue record for this book is available from the British Library
Library of Congress Cataloging in Publication Data
Johnson, Gerald W., 1939-
The Feynman integral and Feynman's operational calculus/Gerald
W. Johnson, Michel L. Lapidus.
p. cm.—(Oxford mathematical monographs)
Includes bibliography references and indexes.
under U.S. or applicable copyright law.

1. Feynman integrals. 2. Calculus, Operational. I. Lapidus,

Michel L. (Michel Laurent), 1956-. II. Title. III. Series.
QA312.J54 2000 515'.43—dc21 99-42820
ISBN 0 19 851572 3 (Pbk)
Typeset by
Newgen Imaging Systems (P) Ltd., Chennai, India
Printed in Great Britain by
Bookcraft (Bath) Ltd,
Midsomer Norton, Avon

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

To Joan, my love and best friend, and to our family:

Tom, Lisa, Caitlin, Carly and Hannah;
Greg, Melissa and Sarah;
Katie; Jenny and Doug

Gerald W. Johnson
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

To my parents,
Myriam and Serge Lapidus

To my wife and love,

Odile Lapidus

To my children,
Julie and Michael

To my sisters and their families,

Sylvie Hanus and Muriel Attia

For their love, teaching,

under U.S. or applicable copyright law.

encouragement, and inspiration

Michel L. Lapidus

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

PREFACE

This book is directed primarily to mathematicians and mathematical physicists, but also to theo-
retical physicists and to other scientists with an interest in quantum theory. One of our purposes in
writing this book on the beautiful and closely related topics of the Feynman integral and Feynman's
operational calculus is to make these subjects accessible to a wider audience, including graduate
students. Accordingly, much of the necessary background material is provided within; we call
the reader's attention especially to Chapters 3, 4, 6, 9 and 10 in the table of contents. Chapter
7 also consists, in a sense, of background material, but it deals with the heuristic ideas that led
to the Feynman integral and with the difficulties that arise from attempts to make this subject
mathematically rigorous. Of course, many potential readers will know a significant portion of the
background information and will therefore be able to go quickly over the corresponding parts of
the book.
Both authors have taught courses in Lincoln and Riverside, respectively, over the material of
this book as it was being developed and refined. Also, both of us have given lectures, sometimes
series of lectures (or short courses), on these subjects in many places around the world. Our
experience suggests that it takes about three to four semesters to go through Chapters 1 to 19. The
material divides rather naturally into Chapters 1-13 and Chapters 14-19, although there is a great
deal of overlap and cross-referencing between these two parts of the book. Many of the listeners
at the courses or conference talks mentioned above have been helpful to us by asking thoughtful
questions or by making insightful comments, and perhaps most useful of all, by helping us to
maintain our enthusiasm for this long-term endeavor.

Gerald W. Johnson and Michel L. Lapidus, May 1999

under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

ACKNOWLEDGEMENTS
(joint and individual)

The strong scientific influence of the physicist Richard Feynman will be apparent to the readers of
this book, especially in the discussion of his imaginative heuristic ideas in Chapters 7 and 14 in
connection with the Feynman integral and Feynman's operational calculus, respectively.
Mathematicians and mathematical physicists whose work has had a major impact on several
aspects of this book include Robert Cameron, Mark Kac, Tosio Kato and Edward Nelson.

Acknowledgements by Gerald W. Johnson

I have been fortunate enough to have many positive influences in my scientific life, but it is proba-
bly not reasonable to name them all. The following people have been mentors and/or collaborators
as well as friends over an extended period: David Skoug, my long-term colleague and collab-
orator at the University of Nebraska at Lincoln (UNL); Robert Cameron and David Storvick;
Gopinath Kallianpur; Kun Soo Chang; Jesus Gil de Lamadrid, my Ph.D. advisor at the University
of Minnesota and mentor even long after I had switched areas.
The following Ph.D. students and postdoctoral visitors at UNL have helped with the proof-
reading of the manuscript and sometimes with useful comments regarding the presentation of the
material: Byung Moo Ann, Lisa Johnson, Jeong Gyoo Kim, Jung Won Ko, Jung Ah Lim, Lance
Nielsen, Yeon-Hee Park, Tristan Reyes and Troy Riggs.
Finally, I would like to acknowledge the support of several institutions and individuals going
back to the years when this "book" was only a vague plan and some lecture notes from graduate
classes at UNL and from talks given elsewhere (needless to say, this plan was later substantially
modified and merged with that of my co-author): University of Erlangen, Germany (Dietrich
Kolzow); University of North Carolina, Chapel Hill, Center for Stochastic Processes (Gopinath
Kallianpur); University of Sherbrooke, Quebec (Pedro Morales), University of Clermont-Ferrand,
France (Albert Badrikian), Yonsei University, Seoul, Korea (Kun Soo Chang); University of
Minnesota (Robert Cameron, David Storvick and Jesus Gil de Lamadrid); University of Georgia,
Athens (my co-author); Bielefeld-Bochum Stochastic in Bielefeld, Germany (Sergio Albeverio,
Ph. Blanchard, Ludwig Streit); University of Missouri, Columbia (Brian DeFacio); University of
California, Riverside (my co-author); University of New South Wales, Sydney, Australia (Brian
Jefferies); University of Warsaw, Poland (W. Chojnacki); Institute for Applied Mathematics,
under U.S. or applicable copyright law.

Chinese Academy of Sciences, Beijing (Zhiming Ma, J. A. Yan); and finally, of course, University
of Nebraska-Lincoln (David Skoug).

Gerald W. Johnson, University of Nebraska-Lincoln

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ix
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Acknowledgements by Michel L. Lapidus

I would like to acknowledge my indebtedness to the following mathematicians, mathematical
physicists and physicists who have played an important role in my formative years as a research
mathematician.
Gustave Choquet, for opening up for me the wonderful world of analysis and geometry during
my student years in Paris and for remaining a close advisor and gentle critic throughout my
scientific life.
Yvonne Choquet-Bruhat, who has directed my very first research memoir (dealing with general
relativity and differential topology) and who has provided me with kind guidance and direction at
a time of personal hardship.
Haim Brezis, for guiding some of my first steps into research and for being my advisor both for
my Ph.D. Dissertation [Lal], my These de Doctorat d'Etates Sciences [Lal3], and my Habilitation
(to direct research), all at the Universite Pierre et Marie Curie (Paris VI). He has set high ethical
and scientific standards by his own example which I have always tried to emulate.
Tosio Kato and Paul Chernoff, for warmly welcoming me at the University of California,
Berkeley, and for providing guidance and inspiration during my first few years of academic life in
the United States.
Isadore Singer, whose weekly three hour long Gauge Theory (and Mathematical Physics)
Seminar at UC Berkeley in the late 1970s and the early 1980s has been an enthralling intellec-
tual and cultural experience for me, in particular by transcending the traditional divides between
mathematics and physics and by helping to build a very useful dictionary between these two
disciplines.
Mark Kac, of whom I was fortunate to be a junior colleague during the first half of the 1980s
in Los Angeles and who has been for me both a close friend and advisor as well as a scientific
mentor until his death towards the end of 1984.
Richard Feynman, whom I have had the privilege to know at the California Institute of Tech-
nology in Pasadena from 1981 until shortly before his death in 1988. I am grateful to him for
listening to some of my ideas and results on the 'Feynman integral' (even by attending, apparently
for the first time in Caltech, a Mathematics Colloquium which I gave on this subject in 1982), as
well as for suggesting to me to read his 1951 paper on the operational calculus [Fey8] and for
strongly encouraging me to develop a mathematical theory justifying and extending it.
Alain Connes, for the depth and intrinsic beauty of his work which I have watcned develop
since my student years, as well as for many friendly and enlightening conversations about much
of mathematics over the last few years.
My long-time friend and mentor, Moshe Flato, who passed away a few months ago and to
whom I would have so much liked to give one of the first copies of this book. From him, I have
under U.S. or applicable copyright law.

learned more physics and mathematical physics than from anybody else. I will always remember
Moshe fondly for his wit, breadth and generosity.
To all of them, as well as to many colleagues and researchers throughout the world with whom
I have had the pleasure to collaborate or to interact, I wish to pay homage and give my heartfelt
thanks.
In addition, I am grateful to a number of graduate students for attending my lectures or seminars
on or related to various parts of the material presented in this book and for providing me with
their feedback and comments. Among them, I would like to mention, in particular, Christina He,
Derek Dreier, Cheryl Griffith, Piotr Hebda, Peiqing Jiang, Lior Kadosh, Sasa Kresic-Juric, Luong
Le, EBSCO
KathyPublishing
Nabours:and Trieu
eBook Nguyen.
Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 X
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

On the more personal side, I would like to thank my wife, Odile, and my children, Julie and
Michael, for their patience and understanding, and above all, for not voicing too loudly their doubts
and concerns at times when they would have been fully justified to do so. Finishing up within the
same year two books on completely different subjects (the present one, which has been a more
than ten year enterprise, and the research monograph [La-vF2], which is the realization of another
one of my long-term dreams) has certainly not been easy for me, but it was clearly even less so for
them who had to maintain their support while watching in disbelief. I promise them not to begin
writing another book... at least for a while.
Finally, it is a pleasure to also acknowledge the support of many universities and research
institutes throughout the years during which this book was written or the research leading to it was
being developed. In particular, the University of California at Berkeley, the University of Southern
California in Los Angeles, the Mathematical Sciences Research Institute (MSRI) in Berkeley (on
many occasions from the mid-1980s through the late 1990s), the University of Georgia in Athens
(at which I have taught my first graduate course on a very preliminary version of the beginning of
this material in the late 1980s), Yale University in New Haven, Connecticut, and the University of
California at Riverside. Furthermore, I would like to express my deep gratitude to the Institut des
Hautes Etudes Scientifiques (IHES) in Bures-sur-Yvette, near Paris, France (of which I have been a
very frequent visitor during the last seven years), the Fields Institute for Research in Mathematical
Sciences in Toronto (and previously in Waterloo), Canada, the Erwin Schroedinger International
Institute for Mathematical Physics in Vienna, Austria, as well as the Isaac Newton Institute for
Mathematical Sciences of the University of Cambridge, England, where some of the later chapters
of this book were completed.
Last but not least, I wish to acknowledge the financial support of several research foundations,
including especially the U.S. National Science Foundation (NSF) which has supported my research
for the past fifteen years and, more recently, the Research Foundation of the Academic Senate of
the University of California.
Michel L. Lapidus, University of California, Riverside

Both of us would like to thank Mrs. Jan Carter for an excellent job of typing a difficult
manuscript and for remaining in good spirits against all odds.
We would also like to thank the Mathematics and Physical Sciences Editors at Oxford Uni-
versity Press, Elizabeth Johnston, Managing Editor, and Julia Tompson, Development Editor, for
helpful advice and encouragement as well as careful handling of the manuscript.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

CONTENTS

1 Introduction 1
1.1 General introductory comments 1
Feynman's path integral 1
Feynman's operational calculus 4
Feynman's operational calculus via the Feynman and
Wiener integrals 5
Feynman's operational calculus and evolution equations 7
Further work on or related to the Feynman integral: Chapter 20 8
1.2 Recurring themes and their connections with the Feynman integral and
Feynman's operational calculus 8
Product formulas and applications to the Feynman integral 8
Feynman-Kac formula: Analytic continuation in time and mass 10
The role of operator theory 12
Connections between the Feynman-Kac and Trotter product formulas 13
Evolution equations 13
Functions of noncommuting operators 15
Time-ordered perturbation series 15
The use of measures 16
1.3 Relationship with the motivating physical theories: background and
quantum-mechanical models 17
Physical background 17
Highly singular potentials 18
Time-dependent potentials 19
Phenomenological models: complex and nonlocal potentials 19
Prerequisites, new material, and organization of the book 21
2 The physical phenomenon of Brownian motion 24
2.1 A brief historical sketch 24
under U.S. or applicable copyright law.

2.2 Einstein's probabilistic formula 28

3 Wiener measure 31
3.1 There is no reasonable translation invariant measure on Wiener space 32
3.2 Construction of Wiener measure 34
3.3 Wiener's integration formula and applications 42
Finitely based functions 43
Applications 45
Axiomatic description of the Wiener process 51
3.4 Nondifferentiability of Wiener paths
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
51
DISTRITAL dFRANCISCO
-dimensional Wiener
JOSE DE CALDAS measure and Wiener process 57
3.598476 Appendix:
AN: Converse
; Johnson, Gerald measurability
W., Lapidus, results
Michel L..; The Feynman Integral and Feynman's 57
Operational Calculus
3.6 Appendix: B(X x Y) = B(X) ® B(Y) 60
Account: ns000601
 xii CONTENTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

4 Scaling in Wiener space and the analytic Feynman integral 62

4.1 Quadratic variation of Wiener paths 63
4.2 Scale change in Wiener space 67
4.3 Translation pathologies 74
4.4 Scale-invariant measurable functions 77
4.5 The scalar-valued analytic Feynman integral 79
4.6 The nonexistence of Feynman's "measure" 82
4.7 Appendix: Some useful Gaussian-type integrals 85
4.8 Appendix: Proof of formula (4.2.3a) 87
5 Stochastic processes and the Wiener process 89
5.1 Stochastic processes and probability measures on function spaces 89
5.2 The Kolmogorov consistency theorem 90
5.3 Two realizations of the Wiener process 92
6 Quantum dynamics and the Schrodinger equation 94
6.1 Hamiltonian approach to quantum dynamics 94
6.2 Transition amplitudes and measurement 95
6.3 The Heisenberg uncertainty principle 96
6.4 Hamiltonian for a system of particles 97
7 The Feynman integral: heuristic ideas and mathematical difficulties 99
7.1 Introduction 99
7.2 Feynman's formula 101
Connections with classical mechanics: The method of stationary pha 105
7.3 Heuristic derivation of the Schrodinger equation 106
7.4 Feynman's approximation formula 109
7.5 Nelson's approach via the Trotter product formula 111
The Trotter product formula 114
7.6 The approach via analytic continuation 115
8 Semigroups of operators: an informal introduction 121
9 Linear semigroups of operators 127
9.1 Infinitesimal generator 127
Integral equation 129
Evolution equation 130
Closed unbounded operators 130
9.2 Examples of semigroups and their generators 134
under U.S. or applicable copyright law.

The translation semigroup 134

The heat semigroup 135
The Poisson semigroup 137
9.3 The resolvent 138
9.4 Generation theorems 140
The Hille-Yosida theorem 140
Dissipative operators and the Lumer-Phillips theorem 141
9.5 Uniformly continuous and weakly continuous semigroups 143
9.6 Self-adjoint operators, unitary groups and Stone's theorem 144
9.7 Publishing
EBSCO Perturbation
: eBooktheorems
Collection (EBSCOhost) - printed on 6/8/2017 5:02PM via UNIVERSIDAD 147
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONTENTS xiii
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

10 Unbounded self-adjoint operators and quadratic forms 152

10.1 Spectral theorem for unbounded self-adjoint operators 153
Multiplication operators 153
Three useful forms of the spectral theorem 155
10.2 Applications of the spectral theorem 161
The free Hamiltonian HO 161
The heat semigroup and unitary group 163
Standard cores for the free Hamiltonian 171
Imaginary resolvents 174
10.3 Representation theorems for unbounded quadratic forms 176
Basic definitions and properties 176
Representation theorems for quadratic forms 183
The form sum of operators 190
10.4 Conditions on the potential V for Ho-form boundedness 193
11 Product formulas with applications to the Feynman integral 197
11.1 Trotter and Chernoff product formulas 198
Product formula for unitary groups 203
11.2 Feynman integral via the Trotter product formula 204
Criteria for essential self-adjointness of positive operators 204
A brief outline of distribution theory 205
Kato's distributional inequality 206
Essential self-adjointness of the Hamiltonian H = HQ + V 210
Conditions on the potential V for Hg-operator boundedness 215
Feynman integral via the Trotter product formula for 217
unitary groups
11.3 Product formula for imaginary resolvents 220
Hypotheses and statement of the main result 220
Proof of the product formula 222
Consequences, extensions and open problems 229
11.4 Application to the modified Feynman integral 232
Modified Feynman integral and Schrodinger equation 233
with singular potential
Extensions: Riemannian manifolds and magnetic vector potentials 241
11.5 Dominated convergence theorem for the modified Feynman integral 245
Preliminaries 247
Perturbation of form sums of self-adjoint operators 248
under U.S. or applicable copyright law.

Application to a general dominated convergence theorem 252

for Feynman integrals
11.6 The modified Feynman integral for complex potentials 259
Product formula for imaginary resolvents of normal operators 260
Application to dissipative quantum systems 263
11.7 Appendix: Extended Vitali's theorem with application to unitary groups 266
Extension of Vitali's theorem for sequences of analytic functions 266
Analytic continuation and product formula for unitary groups 269
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 xiv CONTENTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

12 The Feynman-Kac formula 272

12.1 The Feynman-Kac formula, the heat equation and the
Wiener integral 273
12.2 Proof of the Feynman-Kac formula 276
Bounded potentials 276
Monotone convergence theorems for forms and integrals 282
Unbounded potentials 284
12.3 Consequences 290
13 Analytic-in-time or -mass operator-valued Feynman integrals 293
13.1 Introduction 293
13.2 The analytic-in-time operator-valued Feynman integral 298
13.3 Proof of existence 300
13.4 The Feynman integrals compared with one another and with the
unitary group. Application to stability theorems 303
13.5 The analytic-in-mass operator-valued Feynman integral 308
Definition of the analytic-in-mass operator-valued Feynman integral 309
Nelson's results 312
Haugsby's result for time-dependent, complex-valued potentials 321
Further extensions via a product formula for semigroups 323
13.6 The analytic-in-mass modified Feynman integral 331
Existence of the analytic-in-mass modified Feynman integral 334
Product formula for resolvents: The case of imaginary mass 337
Comparison with other analytic-in-mass Feynman integrals 345
Highly singular central potentials—the attractive inverse-square potential 346
13.7 The analytic-in-time operator-valued Feynman integral via additive
functionals of Brownian motion 358
Introductory remarks 359
The parallel with Section 13.3 359
Generalized signed measures 361
The generalized Kato class 361
Capacity on Rd 362
Smooth measures 363
Positive continuous additive functionals of Brownian motion 364
The relationship between smooth measures and PCAFs 365
The analytic-in-time operator-valued Feynman integral exists for
under U.S. or applicable copyright law.

H = u+ — u- e S — GKd 367
Examples 368
14 Feynman's operational calculus for noncommuting
operators: an introduction 374
14.1 Functions of operators 375
14.2 The rules for Feynman's operational calculus 376
Feynman's time-ordering convention 377
Feynman's heuristic rules 377
Two elementary examples 378
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONTENTS xv
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

14.3 Time-ordered perturbation series 383

Perturbation series via Feynman's operational calculus 383
Perturbation series via a path integral 389
The origins of Feynman's operational calculus 393
14.4 Making Feynman's operational calculus rigorous 394
/. Rigor via path integrals 394
II. Well-defined and useful formulas arrived at via Feynman's
heuristic rules 395
III. A general theory of Feynman's operational calculus with
computations which are rigorous at each stage 396
14.5 Feynman's operational calculus via Wiener and Feynman integrals:
Comments on Chapters 15-18 397

15 Generalized Dyson series, the Feynman integral and Feynman's

operational calculus 404
15.1 Introduction 404
15.2 The analytic operator-valued Feynman integral 407
Notation and definitions 407
The analytic (in mass) operator-valued Feynman integral K t ( . ) 410
Preliminary results 413
15.3 A simple generalized Dyson series (n = u + wdT) 416
The classical Dyson series 424
15.4 Generalized Dyson series: The general case 426
15.5 Disentangling via perturbation expansions: Examples 434
A single measure and potential 435
Several measures and potentials 442
15.6 Generalized Feynman diagrams 446
15.7 Commutative Banach algebras of functionals:
The disentangling algebras 451
The disentangling algebras A, 452
The time-reversal map on At and the natural physical ordering 455
Connections with Feynman's operational calculus 459
under U.S. or applicable copyright law.

16 Stability results 462

16.1 Stability in the potentials 462
16.2 Stability in the measures 464

17 The Feynman-Kac formula with a Lebesgue-Stieltjes

measure and Feynman's operational calculus 477
17.1 Introduction 477
Notation and hypotheses 478
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 xvi CONTENTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

17.2 The Feynman-Kac formula with a Lebesgue-Stieltjes measure: Finitely

supported discrete part v 480
Integral equation (integrated form of the evolution equation) 480
Differential equation (differential form of the evolution equation) 481
Discontinuities (in time) of the solution 482
Propagator and explicit solution 483
17.3 Derivation of the integral equation in a simple case (77 = u + wdT) 486
Sketch of the proof when v is finitely supported 495
17.4 Discontinuities of the solution to the evolution equation 496
The time discontinuities 496
Differential equation and change of initial condition 497
17.5 Explicit solution and physical interpretations 499
Continuous measure: Uniqueness of the solution 500
Measure with finitely supported discrete part: Propagator and explicit
solution 501
Physical interpretations in the quantum-mechanical case 503
Physical interpretations in the diffusion case 504
Further connections with Feynman's operational calculus 505
17.6 The Feynman-Kac formula with a Lebesgue-Stieltjes measure: The general
case (arbitrary measure n) 507
Integral equation (integrated form of the evolution equation) 507
Basic properties of the solution to the integral equation 509
Quantum-mechanical case: Reformulation in the interaction
(or Dirac) picture 511
Product integral representation of the solution 514
Distributional differential equation (true differential form of the
evolution equation) 517
Unitary propagators 519
Scattering matrix and improper product integral 520
Sketch of the proof of the integral equation 521
18 Noncommutative operations on Wiener functionals, disentangling
algebras and Feynman's operational calculus 530
18.1 Introduction 530
18.2 Preliminaries: maps, measures and measurability 532
18.3 The noncommutative operations * and + 535
540
under U.S. or applicable copyright law.

18.4 The functional integrals K t ( . ) and the operations * and +

18.5 The disentangling algebras At, the operations * and +, and the
disentangling process 544
Examples: Trigonometric, binomial and exponential formulas 552
18.6 Appendix: Quantization, axiomatic Feynman's operational calculus, and
generalized functional integral 553
Algebraic and analytic axioms 554
Consequences of the axioms 556
Examples: the disentangling algebras and analytic Feynman integrals 559
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONTENTS xvii
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

19 Feynman's operational calculus and evolution equations 562

19.1 Introduction and hypotheses 562
Feynman's operational calculus as a generalized path integral 562
Exponentials of sums of noncommuting operators 563
Disentangling exponentials of sums via perturbation series 563
Local and nonlocal potentials 565
Hypotheses 566
19.2 Disentangling exp{-tx + f' B ( s ) u(ds)} 568
19.3 Disentangling exp{-ta + ft B1(s)u1 (ds) + ... + f1 Bn(s)un (ds)} 573
19.4 Convergence of the disentangled series 581
19.5 The evolution equation 587
19.6 Uniqueness of the solution to the evolution equation 596
19.7 Further examples of the disentangling process 599
Nonlocal potentials relevant to phenomenological nuclear theory 604

20 Further work on or related to the Feynman integral 609

20.1 Transform approaches to the Feynman integral. References to further
approaches 609
A. The Fresnel integral and other transform approaches to the
Feynman integral 610
The Fresnel integral 610
Properties of the Fresnel integral 611
An approach to the Feynman integral via the Fresnel integral 613
Advantages and disadvantages of Fresnel integral approaches to the
Feynman integral 613
The Feynman map 615
The Poisson process and transforms 616
A "Fresnel integral" on classical Wiener space 616
The Banach algebras S and F(H\) are the same 620
Consequences of the close relationship between S and F (H1) 621
More functions in F(H1) 622
A unified theory of Fresnel integrals: Introductory remarks 623
Background material 624
under U.S. or applicable copyright law.

A unified theory of Fresnel integrals (continued) 626

The Fresnel classes along with quadratic forms 629
The classes Gq (H) and Gq (B) 630
Quadratic forms extended 631
Functions in the Fresnel class of an abstract Wiener space: Examples of
abstract Wiener spaces 632
Fourier-Feynman transforms, convolution, and the first variation for
functions in S 636
B. References to further approaches to the Feynman integral 636
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 xviii CONTENTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

20.2 The influence of heuristic Feynman integrals on contemporary

mathematics and physics: Some examples 637
The heuristic Feynman path integral 638
A. Knot invariants and low-dimensional topology 639
The Jones polynomial invariant for knots and links 639
Witten's topological invariants via Feynman path integrals 641
Further developments: Vassiliev invariants and the
Kontsevich integral 654
B. Further comments and references on subjects related to the
Feynman integral 659
Supersymmetric Feynman path integrals and the Atiyah-Singer index
theorem 659
Deformation quantization: Star products and perturbation series 674
Gauge field theory and Feynman path integrals 682
String theory, Feynman-Polyakov integrals, and dualities 688
What lies ahead? Towards a geometrization of Feynman path
integrals? 695
References 697
Index of symbols 745
Author index 750
Subject index 756
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

1
INTRODUCTION

The main purpose of this book is to provide a mathematical treatment of the Feynman
path integral and the related subject of Feynman's operational calculus for noncommuting
operators. The former subject is more widely known than the latter and has the reputation
of being a formidable and rather elusive mathematical topic.
We will keep this introductory chapter, especially Section 1.1, nontechnical and
relatively brief as far as possible. A detailed table of contents is provided and additional
introductory chapters are included in the book in appropriate places. The main two are
Chapters 7 and 14, dealing, respectively, with the first and second subjects:
Chapter 7, entitled "The Feynman integral: Heuristic ideas and mathematical diffi-
culties", provides an introduction to quantum theory mainly from the perspective of the
physicist Richard Feynman. Further, it points out why the Feynman "integral" is a dif-
ficult subject and shows how Feynman's ideas have led to the mathematical approaches
to the Feynman integral which are used in Chapters 11-13 and 15-18.
Chapter 14 provides an introduction to Feynman's operational calculus for noncom-
muting operators, the subject of Chapters 15-19, and indicates how the Feynman integral
and Feynman's operational calculus are related both in the present theory and in their
historical development.

1.1 General introductory comments

Feynman's path integral
I find Feynman's formula to be very beautiful. It connects the quantum mechanical propagator,
which is a twentieth-century concept, with the classical mechanics of Newton and Lagrange in a
uniquely compelling way.
under U.S. or applicable copyright law.

Mark Kac, 1984 [Kac5, p. 116]

Bohr got up and said: "Already in 1925, 1926, we knew that the classical idea of a trajectory or a
path is not legitimate in quantum mechanics; one could not talk about the trajectory of an electron
in the atom, because it was something not observable." In other words, he was telling me about
the uncertainty principle. It became clear to me that there was no communication between what I
was trying to say and [what] they were thinking. Bohr thought that I didn't know the uncertainty
principle, and was actually not doing quantum mechanics right either. He didn't understand at all
what I was saying. I got a terrible feeling of resignation. I said to myself, "I'll just have to write it
all down and publish it, so that they can read it and study it, because I know it's right! That's all
there is toPublishing
EBSCO it. : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO
Richard P. Feynman, JOSEreminiscing
DE CALDAS about the 1948 Pocono conference.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus (Quoted in [Me, p. 248].)
Account: ns000601
 2 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We begin with Feynman's famous heuristic formula [Fey 1,2] for the evolution of a
nonrelativistic quantum system:

where i = \/—I. We will make some comments about this formula here, but a much
more thorough discussion will be given in Chapter 7.
In (1.1.1), C0,tv is the space of all real-valued (more generally, Rd-valued) continuous
functions x on [0, t] such that x(0) = u and x(t) = v. Further, Dx represents a measure
on C0,v which weighs all paths x equally (in much the same way as Lebesgue measure
weighs all points in R equally), h is Planck's constant divided by 2n, and S(x) is the
action integral associated with the path x; that is,

The integrand in (1.1.2) is the Lagrangian; it equals, for each s in the time interval [0, t],
the kinetic energy minus the potential energy at the point x(s).
Note that the potential V in (1.1.2) is real-valued, so that the integrand in (1.1.1) has
a constant absolute value of one. Hence, it is the net interference effect as x ranges over
the space of paths that determines the value of the oscillatory integral.
Feynman's ideas on the path integral (or "sum over histories") were ingenious and
have had far-reaching consequences in many parts of physics, and more recently, of
mathematics as well. At first, however, they seemed "crazy" to many physicists, including
some famous ones (see [Me, §2.4]). Paths—and concepts that depend on paths, such as
the Lagrangian and the action integral—play a crucial role in Feynman's formulation,
whereas they had been "banned" (in light of the Heisenberg uncertainty principle) from
the standard Hamiltonian approach to quantum dynamics (see Chapter 6).
The formula (1.1.1) seems hopeless at first to most mathematicians who come in
contact with it. The "integral" in (1.1.1) is over a space of functions x "most" of which
are nowhere differentiable, and yet the formula for the action S(x) in (1.1.2) involves
calculating the derivative of x. Further, there is a mathematical theorem which implies
under U.S. or applicable copyright law.

that there is no countably additive measure on C0,t which weighs all paths equally. (See
Section 3.1 for a closely related result.)
We should add that Feynman had some awareness of the mathematical difficulties
just described above and concentrated throughout much of [Fey2] on a second approach
(see Section 7.4) that begins with a discretization of the time interval [0, t]. (It enabled
him, in particular, to replace the normalization constant K—which is ill-defined and for
all practical purposes, infinite—by a suitable sequence of finite normalization constants.)
This alternative approach involves fewer but still substantial mathematical difficulties.
The path integral of Feynman is not a Lebesgue integral; indeed, there is no "Feynman
measure" (see Section
EBSCO Publishing : eBook4.6, especially
Collection Theorem
(EBSCOhost) 4.6.1).
- printed At least
on 6/8/2017 forPMfunctions
5:02 of physical
via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
interest, conditional convergence—instead of absolute convergence (as in the Lebesgue
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERAL INTRODUCTORY COMMENTS 3
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

theory)—is at the heart of the matter. Additionally, since the domain of integration of this
oscillatory "integral" is a set of paths, the subject is intrinsically infinite dimensional.
(Physically, the cancellation effects caused by the oscillatory nature of the Feynman
integral correspond to interference effects between quantum-mechanical matter waves.)
The Feynman integral has been approached from many different points of view by
mathematicians and physicists with varied background and interests. The resulting diver-
sity has led to many different definitions of "the" Feynman integral. In this book, we
address several (certainly not all) of these approaches in a setting appropriate for nonrela-
tivistic quantum mechanics. In each of the cases considered, the existence of the Feynman
integral is established under very general assumptions. The different approaches have
their own domain of validity as well as their own strengths and weaknesses, as will be
discussed further on in the book, especially in Chapters 11 and 13. However, under more
restrictive but still quite general hypotheses, we will show that there is far more agree-
ment than seems to have been previously realized between three of these approaches
to the Feynman integral and the standard Hamiltonian approach to quantum dynamics.
(See Section 13.4.)
Results on the Feynman integral for highly singular potentials are given in Chapters
11-13. Chapter 7, which was mentioned earlier, is crucial to an understanding of the
Feynman integral. There, the physical background for nonrelativistic quantum mechanics
is discussed from Feynman's point of view along with the way in which his ideas on the
subject have led to several of the definitions of the Feynman integral which are used in
Chapters 11-13. (Chapter 6 provides an extremely brief discussion of a few of the ideas
which are common both to the usual Hamiltonian approach to quantum dynamics and
to Feynman's approach.)
We close this part of the general introductory comments by providing more specific
information on some issues that are central to the subject matter of this book through
Chapter 13.
The following are shortcomings of many of the mathematical theories of the Feynman
integral which are often pointed out:
(1) The existence theory is not sufficiently general. In particular, many of the standard
real-valued, time independent potentials (V : Rd —> R) which are used in model-
ing quantum systems are singular (for example, the attractive Coulomb potential)
and do not fit within the theory.
under U.S. or applicable copyright law.

(2) Not much information is given about how the various approaches to "the" Feynman
integral are related to one another or to the unitary group which gives the evolution
of the quantum systems in the standard approach to quantum dynamics.
(3) There is a shortage of satisfactory limiting theorems. Indeed, in some cases, no
such theorems are available, while in others, the results do not seem natural from
a physical point of view.
One of the strong points of the work here is that we give quite satisfactory responses
to all three of these objections, especially for three of the four approaches to the
Feynman integral: which
EBSCO Publishing are developed
eBook Collection in- detail
(EBSCOhost) printed in
on this book.
6/8/2017 5:02 The
PM viaFeynman
UNIVERSIDADintegral
DISTRITAL FRANCISCO JOSE DE CALDAS
defined via the Trotter product formula is shown to exist under very general conditions in
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 4 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 11.2.22. Both the modified Feynman integral and the analytic-in-time operator-
valued Feynman integral are shown to exist under even more general conditions in Corol-
lary 11.4.5 and Theorem 13.3.1, respectively. Further, under the common conditions for
their existence in the corollary and theorem just referred to, the modified Feynman inte-
gral and the analytic-in-time operator-valued Feynman integral not only exist but agree
with each other and with the unitary group, as is shown in Corollary 13.4.1. Under the
somewhat more restrictive conditions of Corollary 11.2.22, we will see in Corollary
13.4.2 that the Feynman integral via the Trotter Product Formula can be added to the list
so that all three of these Feynman integrals exist and agree with one another and with the
unitary group associated with the usual Hamiltonian approach to quantum dynamics.
Our limiting theorems for the three approaches to the Feynman integral referred
to above are "dominated-type" convergence theorems. Since cancellation effects are
intrinsic to the Feynman integral, there cannot be dominated convergence theorems in
this subject that exactly parallel the Lebesgue dominated convergence theorem. However,
in the most frequently used models in nonrelativistic quantum mechanics, it is only the
potential energy function that may vary and our assumptions are that the sequence of
functions (V m ) is pointwise convergent (Lebesgue almost everywhere) and "dominated"
in an appropriate sense (see (11.5.20) and (11.5.21) for example). The result for the
modified Feynman integral, Theorem 11.5.19 [Lal2], is the key. The corresponding
result for the analytic in time operator-valued Feynman integral, Corollary 13.4.3, is an
easy corollary of Theorem 11.5.19 and Corollary 13.4.1. The convergence result for the
Feynman integral via the Trotter product formula, Corollary 13.4.6, rests on Theorem
11.5.19 and Corollary 13.4.2 but also on some further considerations.
Although it is not especially difficult, Section 13.4 is quite pleasing because it brings
together all of the positive results associated with items (l)-(3). (Note that we have omit-
ted from the present discussion the analytic-in-mass operator-valued Feynman integral
as studied in Sections 13.5 and 13.6. This material is interesting in its own right, but it
is not readily compared with the three approaches above.)
The questions raised in (l)-(3) above are clearly central to the mathematical theory
of the Feynman integral, but the answers provided in this book are not the only possible
ones. Moreover, there are other important issues besides those implicit in (l)-(3). For
example, the method of stationary phase is one of the heuristically appealing features of
the Feynman path integral (see Chapter 7) but is not discussed rigorously anywhere in
this book. However, it has been justified in the context of the Fresnel integral approach
under U.S. or applicable copyright law.

to the Feynman integral (see, for example, [AlHo2, Rez, AlBrl]).

Feynman's operational calculus

We turn now to the second topic in the title of this book, Feynman's operational calculus
for noncommuting operators. A fuller introduction to this topic is given in Chapter 14.
It is easy to form functions of operators if the operators commute with one another.
However, the subject becomes far more difficult when the operators fail to commute.
Motivated by problems arising in quantum mechanics and quantum electrodynamics,
Feynman ([Fey8],: 1951)
EBSCO Publishing gave heuristic
eBook Collection "rules"
(EBSCOhost) for forming
- printed on 6/8/2017functions
5:02 PM viaofUNIVERSIDAD
noncommuting
operators.
DISTRITALOne of these
FRANCISCO JOSE "rules"
DE CALDASsays to treat the operators as though they commuted, once
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERAL INTRODUCTORY COMMENTS 5
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

a suitable time-ordering convention has been adopted. For example, Feynman writes such
"equalities" as

even when A and B fail to commute.

The process of appropriately restoring the conventional ordering of the operators
after the use of "equalities" such as in (1.1.3) above is referred to as "disentangling".
This "disentangling process" is central to Feynman's operational calculus.
Feynman's "rules", as strange as they may seem, have led to useful results, notably
the time-ordered perturbation series (or Dyson series) of quantum theory.
Feynman's work on his operational calculus is far from mathematically rigorous,
as he himself noted. One of the challenges to mathematicians is to suitably interpret
Feynman's ideas and to put them on a firm mathematical basis. Our work in Chapters
15-18 and in Chapter 19, respectively, discusses two ways of carrying this out and also
further develops the subject in several directions.
What led Feynman to his operational calculus? He wanted a path "integral" in order
to calculate perturbation series in quantum electrodynamics, but he had no such integral
in that setting. His operational calculus was motivated initially by a desire to find methods
of calculation which would generalize those which could be carried out in nonrelativistic
quantum mechanics via his path "integral".
The operational calculus for noncommuting operators which Feynman discovered
generalizes some aspects of path integration. This suggests that in settings where math-
ematically rigorous path integrals are available, it might be possible to use such integrals
to interpret and make rigorous Feynman's operational calculus. Indeed, this is what we
do in Chapters 15-18 using the Wiener and Feynman path integrals.

Feynman's operational calculus via the Feynman and Wiener integrals

Feynman's operational calculus, the Feynman integral and the Wiener integral all come
together in Chapters 15-18 as well as in Sections 14.3-14.5. Chapters 15, 16 and 18 are
based on joint work of the authors; much of this material can be found in [JoLal] and
[JoLa4], respectively. Chapter 17 is adapted from the following papers of the second
author [Lal5, Lal8, Lal6].
The Wiener process (or Brownian motion) does not appear in the title of this book,
under U.S. or applicable copyright law.

but it—along with the associated Wiener measure and integral—appears repeatedly in
this work. It plays an especially important role in Chapters 7 and 12-18. Chapters 3 and
4 present the information that we will need about Wiener measure from an analyst's
point of view. A short Chapter 2 discusses physical Brownian motion and relates it to its
mathematical model, the Wiener process. In Chapter 5, another short chapter, we give a
very brief discussion of a more probabilistic approach to the Wiener process.
The main emphasis in Chapters 15-18 is on using the Feynman and Wiener inte-
grals to study Feynman's operational calculus in the quantum-mechanical and diffusion
(alternatively, heat or probabilistic) settings, respectively. However, many of the results
EBSCO Publishing
in Chapters 15-18 : eBook
haveCollection (EBSCOhost)
an interest of their- printed
own ason contributions
6/8/2017 5:02 PM to
viathe
UNIVERSIDAD
Feynman and
DISTRITAL FRANCISCO JOSE DE CALDAS
Wiener integrals, apart from their connection with Feynman's operational
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
calculus.
Operational Calculus
Account: ns000601
 6 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We will now describe more precisely than above our approach to the operational
calculus in this context. A more detailed overview of Chapters 15-18 is provided in
Chapter 14, especially in Sections 14.3-14.5.
The functions on the space of continuous paths on [0, t ] that are Wiener and Feynman
integrated in Chapters 15-18 belong, for each time t > 0, to the "disentangling algebra"
At. This commutative Banach algebra consists of certain infinite sums of finite products
of functions of the form

where 0 (often thought of as a time-dependent potential) is a complex-valued function

on [0,t) x Rd and n is a bounded Borel measure on [0, t). The function exp(F) is an
important example of a function in At. (It is called the "Feynman-Kac functional with
Lebesgue-Stieltjes measure" n; see Chapter 17. More generally, the elements of At will
often be referred to as "Wiener functionals" in Chapters 14-18.)
The operator-valued path integral of F E At is denoted Kt (F). For X > 0 (the
diffusion case), Kt (F) is defined as a Wiener integral and then extended first via ana-
lytic continuation in A to C+, the open right half-plane, and then via continuity to
C+ := C+\{0}. When A is purely imaginary (the quantum-mechanical case), K t ( F ) is
the "Feynman integral" of F. [This is the analytic (in mass) operator-valued Feynman
integral of F; see Definition 15.2.1 for a more precise statement.]
The disentangling process is carried out in Chapters 15-18 by calculating the path
integral K t ( F ) for A > 0 and then extending the result to A 6 C+. One need not invoke
Feynman's "rules" explicitly in this setting; the necessary time-ordering is done naturally
(but not always easily) while calculating the functional integrals.
The disentangled operators K t ( F ) are expressed as time-ordered perturbation expan-
sions or "generalized Dyson series". Generalized Feynman diagrams (see Section 15.6)
provide a visual aid for keeping track of the terms of a generalized Dyson series. (These
diagrams can be complicated in their own right but they generalize the simple diagrams
of nonrelativistic quantum mechanics and not those of quantum electrodynamics.)
The work in Chapters 15-18 (and also in Chapter 19) not only interprets Feynman's
ideas and makes them rigorous, but also extends them in several different ways. Non-
commutative operations * and + on the family of disentangling algebras [At}t>o are
introduced in Chapter 18. They can be thought of as a noncommutative multiplication
under U.S. or applicable copyright law.

and addition, respectively, on the space of Wiener functionals; see Section 18.3. Such
operations—introduced by the authors in [JoLa3,4]—were not envisioned by Feynman
but they fit nicely into the operational calculus in various ways. If F € At1 and G E At2,
then we know that the operators Kt (F) and K t ( G ) can be disentangled via general-
ized Dyson series. It is natural to ask if the product of Kt (F) and K t ( G ) can also be
disentangled. It can; in fact (Theorem 18.5.6 and Corollary 18.5.7), F * G E Atl+t2 and
for all A 6 C+,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERAL INTRODUCTORY COMMENTS 7
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Since we can show that

on the level of the functionals, we immediately deduce from (1.1.5) that, on the level of
the operators,

Note that (1.1.6) formally resembles Feynman's paradoxical formula (1.1.3) but involves
the noncommutative operations * and + on the disentangling algebras.
The family of commutative disentangling algebras {At }t >o—equipped with the non-
commutative operations * and + along with the (operator-valued, analytic-in-mass)
Feynman integrals Kt (.)—forms a rich interlocking algebraic and analytic structure that
enables us to explore more deeply the noncommutative aspects of Feynman's operational
calculus.
Our systematic use of measures as in (1.1.4) contributes significantly to the richness
of Feynman's operational calculus. Different measures can provide different directions
for disentangling. For example, what is one exponential function of a sum of commuting
operators becomes infinitely many different exponential functions of a sum of noncom-
muting operators. This leads in Chapter 17, entitled "The Feynman-Kac formula with
a Lebesgue-Stieltjes measure and Feynman's operational calculus" and based on work
of Lapidus in [La 14-18], to the solution of a wide variety of evolution equations which
can incorporate both discrete and continuous phenomena.
Feynman's operational calculus and evolution equations
Another approach to Feynman's operational calculus is considered in Chapter 19, based
on joint work of the authors with Brian DeFacio ([dFJoLal] and especially [dFJoLa2]).
The setting is much more general than in Chapters 15-18, but, on the other hand, attention
is focused almost exclusively on exponentials of sums of noncommuting operators. In
[Fey8] and in the papers which led up to it, the emphasis was also on such exponential
functions. This particular focus came from Feynman's desire to calculate formulas for
the evolution of physical systems.
The operators that appeared as the arguments of the exponential function in
Feynman's work were associated with the different forces involved in the physical prob-
under U.S. or applicable copyright law.

lem. Feynman seemed to have complete confidence that applying his "rules" to such
exponential expressions would yield a formula for the evolution of the physical system
at hand. The main results of Chapter 19, Theorems 19.5.1 and 19.6.1, justify (in a math-
ematical sense) Feynman's confidence (under a certain rather general set of hypotheses)
by showing that the disentangled exponential expression gives the unique solution to the
associated evolution equation. Our method is to use Feynman's heuristic ideas to "dis-
entangle" the exponential expression; we then prove that the disentangled expression
makes sense and satisfies the evolution equation.
We hope that the combination of some simple examples of disentangling found in
EBSCO Publishing
Chapter : eBook
14, the more Collection (EBSCOhost)
complicated - printed
calculations fromonChapter
6/8/2017 19
5:02that
PM via
wereUNIVERSIDAD
just referred to
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 8 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

above, along with some additional examples that are provided in Section 19.7, will help
to clarify Feynman's heuristic "rules" for the reader. Chapters 15-18 will also be helpful
in this regard. Although the disentangling is carried out in these chapters in the process
of calculating the Wiener and Feynman integrals, one can see clearly the connections
with Feynman's time-ordering ideas both in the details of the calculations and in the
resulting answers.
Further work on or related to the Feynman integral: Chapter 20
Chapter 20, our last chapter, has a very different character from the rest of this book.
Our main focus in regard to the Feynman integral will be on operator-valued approaches.
However, in Section 20.1, we will give a brief expository account (without proofs) of
scalar-valued approaches to the Feynman integral which involve "transform assump-
tions". A great deal of work on the Feynman integral has been along these lines since
the 1976 monograph of Albeverio and Hoegh-Krohn [AlHol] on the "Fresnel integral".
In Section 20.2, our main concern is with the connections between the "heuristic
Feynman integral" and a variety of further topics in contemporary mathematics and
physics. The greatest emphasis will be on Section 20.2.A where we discuss Witten's
heuristic Feynman integral [Wit14] and its influence on the subjects of knot theory and
low-dimensional topology. In Section 20.2.B, we briefly discuss the relationship between
heuristic path integrals and four additional topics: The Atiyah-Singer index theorem,
deformation quantization, gauge field theory, and string theory. We should stress that the
mathematical existence of the "Feynman integrals" used in Section 20.2 has usually not
been established. We should also caution the reader that the authors are far from being
experts on the subjects involved in Section 20.2.
Given its special nature, Chapter 20 will be excluded from our discussion in the
remainder of this introduction.
Section 1.1, with the exception of its last subsection, has been a brief introduction
to the main topics of this book. Next we turn to a discussion of some of the themes
that are repeated in several places in this work. An ordered (rather than thematic) and
quite detailed list of the topics treated in this book can be found in the list of contents;
the latter has been written partly with this goal in mind. Section 1.2 below is somewhat
more technical than Section 1.1. Depending on their background, some readers may wish
initially to go over parts of this material quickly and then return to it at a later time.
under U.S. or applicable copyright law.

1.2 Recurring themes and their connections with the Feynman integral and
Feynman's operational calculus
There are a number of subjects related to those in the title of this book which will play
an important role and will reappear frequently; the Wiener process has already been
mentioned in this connection. Product formulas, such as the Trotter Product Formula
and the product formula for imaginary resolvents discussed in detail in Chapter 11,
certainly fall into this category as well.
Product
EBSCO formulas
Publishing : and
eBookapplications to the Feynman
Collection (EBSCOhost) integral
- printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
Perhaps the approach to the Feynman integral which is most straightforwardly motivated
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
by Feynman's original paper ([Fey2], 1948), is the approach using the Trotter product
Operational Calculus
Account: ns000601
 RECURRING THEMES 9
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

formula. It is Trotter's formula for the case of unitary groups that is used. Ignoring
some technicalities, this result says that if A and B are (unbounded, noncommuting)
self-adjoint operators on a Hilbert space H and if A + B is essentially self-adjoint (i.e.,
if it has a unique self-adjoint extension), then

where here, by the operator A + B on the left-hand side of (1.2.1), we mean the unique
self-adjoint extension of the algebraic sum A + B.
When (1.2.1) is applied to the Feynman integral, the Hilbert space H will be L2(Rd),
and we will take, after normalizing the physical constants, A = —1A = HO (the free
Hamiltonian), where A denotes the Laplacian on Rd. Further, we will let B = V, the
operator of multiplication by the potential energy function. (The "potential" V : Rd ->
R is a suitable real-valued function on R d .) Finally, we let H = A + B = H0 + V
denote the Hamiltonian or energy operator associated with V. Then, when applied to an
appropriate wave function o, the left-hand side of (1.2.1), namely, w ( t , •) := e-it H p,
yields the unique solution of the Schrodinger equation

with initial state W (0, •) = <P in the domain of H.

The approach to the Feynman integral via the Trotter product formula is the first of two
approaches which appeared in Nelson's paper ([Nel], 1964). An informal explanation
of the connection between [Fey2] and the Trotter product formula is given in Sections
7.2 and 7.5 and a precise discussion with proofs of a result which is more general than
the one in [Nel] appears in Sections 11.1 and 11.2.
Inspired by the product formula of Trotter and the work of Nelson, Lapidus found
[La11] a "product formula for imaginary resolvents" and used it to define and establish
the existence of the "modified Feynman integral". The result of Lapidus goes well beyond
the case where A + B is essentially self-adjoint; in fact, his formula involves the "form
sum" A + B of the operators A and B. Also, the unitary operators e-1 t A and e~'«B on
under U.S. or applicable copyright law.

the right-hand side of (1.2.1) are replaced by the imaginary resolvents [/ + i(t/n)A]~l
and [/ + i ( t / n ) B ] - 1 , respectively. Thus, we have the "product formula for imaginary
resolvents" (see Section 11.3, especially Theorem 11.3.1 and Corollary 11.3.7):

(If A + B is essentially self-adjoint, as in the hypothesis of the product formula for unitary
EBSCO (1.2.1),
groups Publishing : eBook
then the Collection
form sum(EBSCOhost) - printed onwith
A + B coincides 6/8/2017
A +5:02
B, PM
theviaunique
UNIVERSIDAD
self-adjoint
DISTRITAL FRANCISCO JOSE DE CALDAS
extension of the algebraic sum A + B—and so the left-hand side of (1.2.3) coincides
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
withOperational
that of (1.2.1);
Calculus see Proposition 11.2.10(ii).)
Account: ns000601
 10 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

When the product formula (1.2.3) is applied to define and establish the convergence of
the modified Feynman integral, we obtain much as before a solution to the Schrodinger
equation, but now with the Hamiltonian given by the form sum of H0 and V; i.e.,
H = H0 + V. (See Section 11.4, including Definition 11.4.4.)
In the setting we have been considering, the potential is a real-valued and time-
independent function V and the Hamiltonian is obtained by "adding" V to H0, with the
sum allowed to be a form sum. The maximum domain of validity for V in this framework
is (as we will see in Section 11.4) exactly the same for the modified Feynman integral as
it is in the Hamiltonian approach to quantum dynamics. Further, this maximum domain
of validity is physically natural; the "form domain" of the Hamiltonian H = HO + V
consists precisely of those functions o e L 2 (R d ) which have finite total (i.e., kinetic +
potential) energy. Looking ahead and considering the same setting, the maximal domain
of validity for V in the case of the analytic-in-time operator-valued Feynman integral
agrees with the other two. It should be added, however, that the modified Feynman
integral extends nicely to the case of complex potentials V (see Section 11.6) whereas a
corresponding theorem has not been proved (and may not be true) for the analytic-in-time
operator-valued Feynman integral considered in Sections 13.3 and 13.4.
An advantage of the generality of the modified Feynman integral is that it allowed
Lapidus to obtain in [Lal2] a very general stability theorem (relative to the potential) and
to deduce a "dominated-type convergence theorem" appropriate for this context. (See
Section 11.5.)
The results leading to the definition of the "Feynman integral via TPF" [Nel] are
discussed in Section 11.2, while those concerning the "modified Feynman integral"
[Lal-2, La6-13] and various extensions of its definition (notably, to C-valued potentials
[BivLa]) are presented in Sections 11.3-11.6. In addition, we mention that Sections 13.5
and 13.6, respectively, describe analytic (in mass) versions of these two approaches to
the Feynman integral. Product formulas of various types—not themselves consequences
of (1.2.1) or of (1.2.3)—also play a prominent role in these sections.

Feynman-Kac formula: Analytic continuation in time and mass

Mark Kac heard Feynman speak about his path integral in 1947. Kac realized that
if time t in Feynman's formula is replaced by —it ("imaginary time" from the per-
spective of quantum physics), then the expression involved before the limit is taken is
equal to a Wiener integral, a true integral in the Lebesgue sense with respect to the
under U.S. or applicable copyright law.

countably additive Wiener measure m. The powerful results of the Lebesgue theory
of integration can then be used to rigorously justify the calculation of the limit. One
outcome of all this is the famous Feynman-Kac formula. (A detailed proof of a very
general version of the result is given in Chapter 12, based on work of B. Simon in
[Si9].)
Kac's discovery expresses the solution to the heat equation as a certain Wiener
integral. More precisely, if the "Feynman-Kac functional" F is given by

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 RECURRING THEMES 11
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

then for t > 0 and £ e Rd, we have

where m denotes Wiener measure on the space Ct of continuous paths x such that
x(0) = 0. The left-hand side of (1.2.5) yields the unique solution, u(t, •) = e~'H(p, at
time t > 0 of the heat (or diffusion) equation

with initial condition u(0, •) = p. Here, as before, H = H0 + V, with HO = -1A.

Note, however, that we now use the heat semigroup e-tH to represent the solution to the
heat equation (1.2.6) whereas we have used earlier the unitary group e~"H to represent
the solution to the Schrodinger equation (1.2.2).
The "Feynman-Kac formula" (1.2.5) has been extremely useful for a variety of
purposes, both in mathematics and in physics (see Section 7.6 for a brief discussion of this
along with some references), but it does not by itself resolve the problem of making sense
of the Feynman integral since the change from t to —it takes us from quantum theory
and the Schrodinger equation to the heat equation. The Feynman-Kac formula does,
however, suggest an approach to the Feynman integral. Start with imaginary time and the
theoretically powerful Wiener integral and define the Feynman integral by analytically
continuing to real time. Indeed, operator-valued analytic continuation in time is another
of the approaches to the Feynman integral which will be discussed in detail in this book.
These results on the analytic-in-time Feynman integral (at the level of generality found
here) are due to Johnson [Jo6] and are the subject of Sections 13.2 and 13.3. We should
mention that what is imaginary time from the point of view of quantum theory is real
time from the perspective of the heat equation. We shall adopt the latter point of view
in Chapter 13 (Sections 13.2, 13.3 and 13.7) and analytically continue from real time to
purely imaginary time—going in the process from the Wiener integral to the Feynman
integral.
We remark that Section 13.7 gives a brief discussion of an extension (see [AIJoMa])
under U.S. or applicable copyright law.

of the analytic-in-time operator-valued Feynman integral which is based on the theory

of "additive functionals of Brownian motion" (see [Fuk, FukOT]) and Feynman-Kac
formulas in which, for example, the potential V can be replaced by a suitable measure
on Rd.
The last of the approaches to the Feynman integral which will be treated in detail
in this book is operator-valued analytic continuation in mass. Again, one starts with
the Wiener integral but this time, the analytic continuation is in a mass parameter (or
alternately, in a variance parameter). The connection between Feynman's ideas and the
approach to the Feynman integral via operator-valued analytic continuation in mass is
discussed in an informal
EBSCO Publishing way in (EBSCOhost)
: eBook Collection Section 7.6, with on
- printed the6/8/2017
approach
5:02 via theUNIVERSIDAD
PM via Trotter product
DISTRITAL FRANCISCO JOSE DE CALDAS
formula serving to link the two.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 12 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The precise discussion of the analytic-in-mass operator-valued Feynman integral

is given in Section 13.5. The crucial starting point for this work is Nelson's second
approach developed in [Nel]. An earlier paper by Cameron ([Cal], 1960) used scalar-
valued analytic continuation in mass; the key contribution of [Cal] was the proof that
there is no countably additive "Wiener measure" with a complex variance parameter
(see Theorem 4.6.1). This result corrected an error in [GelYag], an interesting and even
earlier paper which used analytic continuation.
Various extensions of Nelson's results are given in Sections 13.5 and 13.6. Among
them, the reader will find hybrids which combine a suitable product formula with analytic
continuation in mass. A comparison of the resulting analytic in mass Feynman integrals
within their common domain of validity is provided towards the end of Section 13.6.
We remind the reader that the analytic-in-mass operator-valued Feynman integral
will also be used in Chapters 15-18. Unlike the approaches in Chapter 13 via analytic
continuation in mass, this Feynman integral exists for every (rather than Lebesgue almost
every) value of the mass parameter. The class of functionals treated in Chapters 15-18
is, in some respects, much larger than in Chapter 13. However, in Chapters 15-18, no
attempt is made to deal with potential functions with strong spatial singularities.
There are four different versions of the analytic-in-mass Feynman integral discussed
in this book, as was alluded to above; in addition, three other approaches to the Feynman
integral have already been discussed in this introduction. In the next two paragraphs, we
indicate briefly what these are and where they are to be found.
The approaches to the Feynman integral that are discussed at any length in this
book are all operator-valued. (Recall that we are not taking Chapter 20 into account in
our present discussion.) Two of the analytic-in-mass approaches start from the Wiener
integral when the mass parameter is real. One of these is discussed in the first part of
Section 13.5; the other, which has quite different features, is defined in Section 15.2
and used throughout Chapters 15-18. The last two begin with product formulas for
semigroups (in Section 13.5) and resolvents (in Section 13.6) to yield analytic-in-mass
versions of the Feynman integral via TPF ([Kat7, BivPi]) and of the modified Feynman
integral [BivLa], respectively.
The Feynman integral defined via the Trotter product formula for unitary groups
is discussed in Section 11.2 and the modified Feynman integral (defined via a product
formula for imaginary resolvents established in Section 11.3) is treated in Sections 11.4-
11.6. Finally, the analytic-in-time Feynman integral appears in Sections 13.2 and 13.3,
under U.S. or applicable copyright law.

with an extension given in Section 13.7.

The role of operator theory

As mentioned above, the approaches to the Feynman integral that will be discussed in
detail in this book are all operator-valued. Further, there is always at least one unbounded
operator involved; much of the time, it is H0 = —1A, the free Hamiltonian, although
various physically meaningful substitutes for HO are allowed in Sections 11.4, 11.6,
and Sections 13.5-13.6, and more abstract generators are considered in Chapters 11
andEBSCO
19. In Sections
Publishing 11.2-11.5,
: eBook CollectionChapter 12,- Sections
(EBSCOhost) 13.2-13.4
printed on 6/8/2017 5:02 and 13.7,
PM via the theory of
UNIVERSIDAD
(notDISTRITAL
necessarily bounded)
FRANCISCO self-adjoint operators and functions of them is sufficient for
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 RECURRING THEMES 13
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

our needs. These needs include various forms of the spectral theorem for unbounded
self-adjoint operators as well as basic results about unbounded quadratic forms and form
sums of operators. This background material is provided in Chapter 10 which is titled
"Unbounded self-adjoint operators and quadratic forms". (See also Section 9.6 for intro-
ductory material on unbounded self-adjoint operators and the associated semigroups.)
The spectral theorem enables us to define the functions e-itH (the unitary group) and,
if the spectrum of the self-adjoint operator H is bounded from below, the (self-adjoint)
semigroup e-tH. For us, in most applications, H is the Hamiltonian (or energy operator),
a suitable self-adjoint extension of H0 + V, where V is the potential. (More specifically,
in Section 11.2, H is the unique self-adjoint extension of H0 + V, and, more generally,
it is the form sum of HO and V in Sections 11.3-11.5, Chapter 12, Sections 13.2-13.3
and 13.7.)
Self-adjoint operators—and the associated unitary groups or self-adjoint semi-
groups—are not adequate for everything that we will do. Strongly continuous (or (C0))
semigroups of operators will be discussed in Chapter 9 (and in the brief and informal
chapter that precedes it). Such semigroups (not necessarily associated with self-adjoint
operators) will be used in Sections 11.1, 11.6, 13.5, 13.6, parts of Chapter 14 and through-
out Chapter 19. They will also frequently be present in Chapters 15-18 but will be used
in a more straightforward way there.
Connections between the Feynman-Kac and Trotter product formulas
The Feynman-Kac and Trotter product formulas have already been discussed above,
but there are additional places in the book where these related formulas or variations
of them appear. The Trotter product formula is the main tool in the crucial first step
of the proof of the Feynman-Kac formula in Chapter 12. A variation of the Feynman-
Kac formula, the "Feynman-Kac formula with a Lebesgue-Stieltjes measure", is—along
with its connection with Feynman's operational calculus—the topic of Chapter 17, which
describes part of the work in [La14-18]. A related product integral, a relative of the Trotter
product formula, is discussed in Section 17.6 [Lal8,16]. Example 16.2.7 (in conjunction
with Example 15.5.5) looks at the relationship between the Trotter product formula
and the Feynman-Kac formula from the point of view of weak (or vague) convergence
of measures. This broad perspective is informative even though the results are far less
general than those proved in Chapters 11 and 12. A version of the Feynman-Kac formula
which substantially extends the one in Chapter 12 is discussed briefly in Section 13.7.
under U.S. or applicable copyright law.

There, for example, the potential energy function can be replaced by certain measures (in
the space rather than in the time variable, as in Chapter 17) which are singular with respect
to Lebesgue measure. Finally, a Feynman-Kac formula for certain complex potentials
is contained in the work presented in Sections 13.5 and 13.6.
Evolution equations
A fundamental concept of quantum mechanics is a quantity called the propagator, and the standard
way of finding it (in the non-relativistic case) is by solving the Schrodinger equation. Feynman
found another way based on what became known as the Feynman path integral or "the sum over
EBSCO Publishing
histories" ... : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS Mark Kac, 1984 [Kac5, p. 116]
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 14 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The evolution of physical systems concerns us throughout this book, so it is not sur-
prising that the subject of evolution equations is another recurring theme. Our point
of view (following Feynman) is not, however, the usual one. Typically, the evolu-
tion equation comes first and is regarded as the model for the physical system. One
then looks for a method to solve the evolution equation and the solution gives the
evolving state. Our deviation from this point of view is perhaps seen most clearly in
Chapter 19. The idea there is: Given the forces involved in the problem, write down
and then "disentangle" the exponential of a sum of integrals (from, say, 0 to t) of
associated time-ordered operators (see (19.4.8)). The resulting time-ordered perturba-
tion series (see (19.3.14)) should give the evolution of the physical system. Of course,
it is of mathematical and physical interest to know if this series solves some related
evolution equation. Theorem 19.5.1 shows that this is so under a quite general set of
assumptions.
As remarked earlier in this introduction, the approach to quantum dynamics provided
by "the" Feynman path integral differs in several ways from the standard Hamiltonian
approach. The point we wish to make here is that the path integral itself should give the
evolving state. No evolution equation is needed ahead of time. Of course, it is of interest
to know conditions under which the evolving state given by the Feynman integral satisfies
the Schrodinger equation or some variation of it.
The different specific approaches to the Feynman integral discussed in this book have
differing relationships with the standard Hamiltonian approach to quantum dynamics.
Our first comments along these lines pertain to Chapter 17. Recall that in Chapters
15-18, the potentials can be time-dependent and complex-valued but are not allowed to
have strong singularities in the space variables. If we take the appropriate Wiener integral
involving the usual Feynman-Kac functional e x . p { F 0 ( x ) } , where F0,i is given by (1.1.4)
and / is Lebesgue measure on the time interval (0, t), we obtain a function of time and
space which describes the evolution of a distribution of heat. By analytically continuing
in mass (and making an adjustment in the potential), we arrive at a function giving
a quantum evolution. These time evolutions are solutions to the heat and Schrodinger
equations, respectively. In Chapter 17, we replace the Feynman-Kac functional exp{ F0,l }
by the Feynman-Kac functional exp{F0,n} (where F0,n is given by (1.1.4) and n is a
Lebesgue-Stieltjes measure) and follow the procedure above. We show first that the
resulting evolutions involve an interesting variety of discrete and continuous phenomena
and then also that they are solutions to correspondingly adjusted versions of heat and
under U.S. or applicable copyright law.

Schrodinger equations which are quite different from the usual ones (see especially
Sections 17.2 and 17.6).
Even though Feynman's approach to quantum dynamics does not depend a priori
on the usual one, the method of proof for three of the specific approaches discussed
in this book, the Feynman integral via the Trotter product formula (Section 11.2), the
modified Feynman integral (Sections 11.3 and 11.4), and the operator-valued analytic-
in-time Feynman integral (Sections 13.3 and 13.7), not only depend heavily on operator-
theoretic results but also on the existence of the unitary group as established in the
standard Hamiltonian approach. [In the case of the modified Feynman integral with
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
complex (rather
DISTRITAL than
FRANCISCO real)
JOSE potential studied in Section 11.6 ([BivLa]), the Schrodinger
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 RECURRING THEMES 15
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

operator must be defined appropriately and the associated time evolution is dissipative
but in general, not unitary.]
The situation is quite different for the analytic-in-mass operator-valued Feynman
integral, whether you begin on the real line with a Wiener integral (Section 13.5) or with
product formulas (Section 13.6 and the last part of Section 13.5). Although operator
techniques are still heavily involved, they are not the ones based on self-adjointness that
are used commonly in quantum mechanics. Moreover, knowledge of the existence of
the unitary group from the usual approach to quantum dynamics is not needed in the
proof. In fact, for extremely singular potentials (see Examples 13.6.13 and 13.6.18), the
analytic-in-mass operator-valued Feynman integral exists but the Hamiltonian approach
does not, at least not in an unambiguous way.

Functions of noncommuting operators

The formation of functions of noncommuting operators is a theme which is implicit
in the title of this book and which is of direct concern to us throughout Chapters 14-
19. Although it is less obvious, the same subject is also involved in Chapters 6-13. For
example, if A and B are commuting self-adjoint operators, there is no need for the Trotter
product formula (1.2.1); we simply have e - i t ( A + B ) = e - i t A e - i t B . The Trotter product
formula has sometimes been referred to as the noncommutative exponential law. (In light
of our later work, especially in Chapters 17 and 19, it would be more accurate to describe
it as an especially important example but just one of many noncommutative exponential
laws.) The spirit of the theory of semigroups of operators is that it is the theory of forming
the "exponential function" of operators. In practice for us (and in general), the operator
to be "exponentiated" is often of the form A + B, where A and B do not commute.
The Feynman-Kac formula expresses the heat semigroup e~'H = e~'<-H°+v^ (where
HO = —1A is the free Hamiltonian and V is the operator of multiplication by the
potential V) as a certain Wiener integral. (See equations (1.2.5) and (1.2.4) above.) In
some sense, this formula can be thought of as providing a way to handle the fact that the
operators HO and V do not commute.

Time-ordered perturbation series

In [Fey8] and in the work in this book on Feynman's operational calculus, the disen-
tangled functions of operators are more often than not expressed as time-ordered per-
turbation series. In Chapters 14-19, such series appear repeatedly. They are most often
under U.S. or applicable copyright law.

referred to as generalized Dyson series in Chapters 15-18. Indeed, special cases of the
perturbation series in all of Chapters 14-19 coincide with the classical Dyson series of
nonrelativistic quantum mechanics.
In Chapter 15 (and then throughout Chapters 16-18), our generalized Dyson series
play a crucial role in defining the operators K[(F) involved, especially in the quantum-
mechanical (or Feynman) case where a bona fide path integral (such as the Wiener
integral) is no longer available. In turn, these perturbation expansions—which can be
thought of, in some extended sense, as providing a "sum over all possible histories" of a
quantum particle—are
EBSCO Publishing : eBook very helpful
Collection mathematical
(EBSCOhost) - printed tools and enable
on 6/8/2017 us to
5:02 PM via derive various
UNIVERSIDAD
properties
DISTRITALwith relative
FRANCISCO ease.
JOSE DE (See, for example, Sections 15.3, 15.5, and Chapter 16.)
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 16 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

When F is an exponential functional (see Chapter 17), they also play a key role in deriving
the evolution equation (either in differential or integral form) satisfied by t i->- K[(F).
(This is especially true in the quantum-mechanical case.)
[For simplicity, we will limit ourselves here to the setting of Chapters 15-18. We point
out, however, that despite certain differences due to the generality of the assumptions
made in Chapter 19 and the absence of any kind of path integral in that framework, our
above comments regarding the definition of the operators involved and the derivation of
a corresponding evolution equation remain valid in the setting of Chapter 19 as well.]
At this point, it may be helpful to recall from our discussion in Section 1.1 that in
Chapter 15, given a Wiener functional F in the disentangling algebra At, we associate
with it an operator K l ( F ) , called the analytic (in mass) Feynman integral of F, which
can be disentangled via a generalized Dyson series. [Briefly, the bounded linear operator
K((F) is defined as a genuine Wiener (path) integral in the diffusion case when A is real,
and then, for complex A, by analytic continuation followed by passage to the limit along
the imaginary axis of the resulting perturbation expansion.] Consequently, the time-
ordered perturbation series for K t ( F ) has the same general expression as a function of
the parameter A both in the diffusion (or probabilistic) case (A real and positive) and
in the quantum-mechanical (or Feynman) case (A purely imaginary and nonzero). This
fact enables us to deal with these two situations in parallel in much of Chapters 15-18.
(Notable exceptions occur in Sections 16.2 and 17.6.)
There is a last general comment that we wish to make about the "disentangling"
provided by our generalized Dyson series: It is not necessarily unique; indeed, a given
operator K t ( F ) can be represented in many different ways via a time-ordered perturba-
tion series, some of which may be more suitable than others in a given situation. (See
especially Section 15.5 for various examples; see also, for instance, Section 17.6.) We
stress that in spite of this fact, the operator K t ( F ) associated with a function F in the
"disentangling algebra" A, is always defined uniquely (and hence unambiguously). In
some suitable sense, the mapping Kt (defined in Section 15.7) can be thought of as a
quantization map from the commutative disentangling algebra At to a noncommutative
algebra of (bounded linear) operators. [See especially Chapter 18 (including Appendix
18.6), where the action of the noncommutative operations * and + on the family of
disentangling algebras {At : t > 0} is taken into account.]

The use of measures

under U.S. or applicable copyright law.

Measures and their associated integrals enter into this book in various ways. We mention
some of these here and emphasize those which are less widely familiar but will be
especially important to us.
Two of the definitions of "the" Feynman integral that are stressed in this book start
with the Wiener integral. Our purpose in Chapter 3 is to give the reader who is not
acquainted with Wiener measure some idea of how it can be constructed and some
familiarity with the properties of the Wiener process that will be needed in subsequent
work. The construction follows the pattern of Lebesgue measure on the line, a topic
familiar to most :mathematicians.
EBSCO Publishing It begins- printed
eBook Collection (EBSCOhost) with the definition
on 6/8/2017 ofvia
5:02 PM theUNIVERSIDAD
measure of an
"interval"
DISTRITALand ends JOSE
FRANCISCO withDEanCALDAS
application of the Caratheodory extension theorem.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 RELATIONSHIP WITH THE MOTIVATING PHYSICAL THEORIES 17
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We expect that many potential readers will be familiar with the results of Chapter 3
and with Levy's quadratic variation law which is the subject of Section 4.1. However,
we anticipate a much lower degree of familiarity with most of the rest of Chapter 4
which deals with such topics as the family of scaled Wiener measures {ma : a > 0},
scale-invariant measurability [JoSk7] and the refined equivalence classes of functions
that are needed for a careful discussion of the Feynman integral obtained via analytic
continuation in mass. This definition of the Feynman integral will concern us in Section
13.5 (the second approach in [Nel]) as well as throughout Chapters 15-18.
Measures on subintervals of K (Lebesgue-Stieltjes measures) are used systemat-
ically throughout Chapters 14—19 in connection with Feynman's operational calculus.
They serve not only to assign weights but also to time-order the integrands which are
usually (perhaps after some preliminary steps) products of noncommuting operators. The
measures give directions for "disentangling", and a different set of measures can yield
very different results. The first few pages of Section 14.2 (through Example 14.2.1) can
be read independently of all of the earlier material in this book and will provide the reader
with a discussion of Feynman's heuristic "rules" and an extremely simple example of
the points made above.

1.3 Relationship with the motivating physical theories: background and

quantum-mechanical models
What does this book have to say about the physical theories which motivate it? The
reader will not find here applications to concrete and detailed physical problems of the
mathematical results contained within. However, in certain respects, we do discuss in a
number of places related physical theories and especially quantum mechanics.

Physical background
A discussion of the relevant physical background is provided in key places. Most import-
antly, Feynman's way of looking at quantum mechanics and his path integral and how
this has led to the approaches to the Feynman integral found in this book is the subject of
Chapter 7. Chapter 6 contains an extremely brief discussion of some parts of the usual
Hamiltonian approach to quantum dynamics; this chapter is included partly for the sake
of contrast but also because the two approaches have, of course, some common features.
It seems to us that it is difficult to get an appreciation for the mathematics of the Feyman
under U.S. or applicable copyright law.

integral without at least some understanding of the physical background.

As noted earlier, this book contains a good deal of information about the Wiener
integral (see Chapters 2-5, 7 and 12-18). Much of this material, apart from Chapter 3,
Section 4.1 and Chapter 5, is not the standard fare but consists of special topics related to
the two items in the title of this book. Chapter 2 discusses the character of physical Brown-
ian motion and the way in which that led Norbert Wiener, through the work of Brown,
Einstein and Perrin, to what is now known as the Wiener process, the mathematical
model of Brownian motion.
Chapter 14 is an introduction to Feynman's operational calculus. Some discussion
of the physical
EBSCO problems
Publishing that led (EBSCOhost)
: eBook Collection Feynman -toprinted
this calculus can
on 6/8/2017 bePMfound
5:02 there, but much
via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
less than one might guess. Why is that?
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 18 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The primary purpose of the paper [Fey8], "An operator calculus having applications
in quantum electrodynamics", was to present the ideas and rules which Feynman had
developed in connection with [Fey5-7] for forming functions of noncommuting oper-
ators. While most of the examples in [Fey8] are from quantum theory, Feynman was
well aware that he had found a computational technique with implications beyond that
particular setting. [In fact, this point was stressed repeatedly by Richard Feynman him-
self in a number of conversations with the second-named author (M. L. L.), during the
first of which (in about 1981) Feynman mentioned his paper [Fey8] on the subject and
urged M. L. L. to develop his operational calculus and to put it on a firm mathematical
basis.] Chapter 14 is an exposition of these mathematical (but not mathematically rigor-
ous) ideas of Feynman and how they will be interpreted, extended and developed with
mathematical rigor in Chapters 15-19.
[The reader may be aware of Feynman's sometimes negative comments about some
of the mathematicians' musings (see, for example, [Fey 16,17]). However, he/she may
wish to contrast this impression with Feynman's comments in [Fey8, p. 108] regarding
the need for mathematical rigor and for further mathematical exploration of his "operator
calculus". (See the second quote from [Fey8] at the very beginning of Chapter 14, which
is in complete agreement with the second author's conversations with Feynman.) Perhaps
it is appropriate at this point to add two more personal recollections. When asked by a
physics Ph.D. student how much mathematics he needed to learn, Feynman answered
without hesitation: "As much as possible." (This was witnessed by the second author in
Los Angeles in 1981.) Finally, and to give a more balanced view, when during his 1983
UCLA public lectures for a general scientifically curious audience (of which his book
QED, [Fey15], is an edited version), he was asked what were the relationships between
mathematicians and physicists, he began his answer (approximately) as follows: "They
are very good friends, but they do not consider the same problems, and they do not have
the same point of view. The mathematician looks at a very broad area and is interested
in everything related to it. The physicist, on the other hand, who is interested in certain
specific questions, can go much further in some particular directions..."]
The discussion of physical background and physical interpretation of results goes
beyond the introductory chapters mentioned above. It can be found in various places
throughout the book. We mention Chapters 11, 13, 15, 16 and especially, Chapters 17
and 19.
under U.S. or applicable copyright law.

Highly singular potentials

A variety of quantum-mechanical models are discussed in this book. These include in
Chapters 11 and 13 highly singular potentials V and the standard Hamiltonian

In (1.3.1), V denotes the operator of multiplication by a time independent, real-valued

potential energy function V. [The precise form of H when the mass m and H = (Planck's
constant)/2n are not normalized is given in (6.4.1). For the case of an N-particle system
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
where the jth
DISTRITAL particle
FRANCISCO JOSEhas
DE mass
CALDAS mj, j = 1 , . . . , N, see (6.4.2).] The inclusion of highly
singular potentials
AN: 98476 ; Johnson,inGerald
the approaches to theL..;
W., Lapidus, Michel Feynman integral
The Feynman discussed
Integral in Chapters 11
and Feynman's
Operational Calculus
Account: ns000601
 RELATIONSHIP WITH THE MOTIVATING PHYSICAL THEORIES 19
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and 13 is a major advantage of those approaches. Some of the most basic potentials of
quantum mechanics such as the Coulomb potential are singular in the space variables.
(See [FrLdSp] for a detailed account from a physicist's point of view of the role of
singular potentials in quantum theory.)
A discussion of highly singular central potentials is provided in Example 11.4.7 and
pursued in Example 13.6.13. The interesting special case of the inverse-square potential
is treated in Example 13.6.18.
We give in Example 11.4.12 and in parts of Sections 13.5 and 13.6 a brief discussion
of a refined and highly singular Hamiltonian which is obtained by supplementing H in
(1.3.1) by a magnetic vector potential. This corresponds to the Schrodinger equation asso-
ciated with a magnetic as well as an electric field. Further, in Example 11.4.10, we con-
sider the case where a d-dimensional Riemannian manifold replaces Euclidean space Rd.
Time-dependent potentials
The operator-valued Feynman integral used in Chapters 15-18 is defined via analytic
continuation in mass. In those chapters, the emphasis is on Feynman's operational cal-
culus and, in particular, on disentangling via time-ordered perturbation series by using
the Wiener and Feynman integrals. The "potentials" allowed there are very general
in most respects; they can be time-dependent and complex-valued and no smoothness
assumptions are made. However, they are required to be essentially bounded in the space
variables; that is, no spatial singularities are permitted. (Hence, for instance, the Coulomb
potential is not allowed in this setting since it has a blow-up singularity at the origin.)
Potentials which are bounded and may be time-dependent appear in various places
in the physics literature. Forces that are under the control of an experimenter provide a
natural source of examples of potentials that are both time-dependent and bounded.
It is not just the potentials $ that influence the possible physical models in Chapters
15-18, but also the Lebesgue-Stieltjes measure n as in (1.1.4). These measures determine
the disentangling (as noted earlier) and, when combined appropriately with an exponen-
tial function, determine the evolution of an associated physical system (see Chapter 17).
[We refer, in particular, to Section 17.5 for possible physical interpretations of the corres-
ponding results both in the quantum-mechanical (or Feynman) case and in the diffusion
(or probabilistic) case.] The fact that such measures may have continuous and/or discrete
parts allows us to study both continuous and discrete phenomena and their relationships
with one another. This considerably broadens our approach to Feynman's operational
under U.S. or applicable copyright law.

calculus via Wiener and Feynman path integrals in Chapters 15-18. Mathematically, it
also gives a rich combinatorial structure to the time-ordered perturbation expansions (or
generalized Dyson series) and the associated generalized Feynman graphs introduced in
Chapter 15 and used throughout the above chapters.
A brief discussion is given in Section 13.5 of Haugsby's extension of Nelson's second
approach to the Feynman integral. This is the only place in the book where potentials
are treated which can be both singular in the space variables and time-dependent.
Phenomenological models: complex and nonlocal potentials
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
We DISTRITAL
are alsoFRANCISCO
able toJOSE
treat certain phenomenological models. By a phenomenological
DE CALDAS
model, we mean
AN: 98476 oneGerald
; Johnson, that does not arise
W., Lapidus, from
Michel L..;the
Thebasic principles
Feynman of Feynman's
Integral and quantum mechanics
Operational Calculus
Account: ns000601
 20 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

but has, nevertheless, been found useful in modeling certain quantum systems. We have
already mentioned complex potentials above. Such potentials are used in modeling dis-
sipative (or open) quantum systems. An extensive discussion of this topic—including
its strengths and weaknesses and its relationship with "the" Feynman integral—can be
found in Exner's book [Ex], Open Quantum Systems and the Feynman Integral. Com-
plex potentials are permitted in some of the results in Chapters 11 and 13 (see especially
Sections 11.6 and 13.6, as well as the end of Section 13.5) and in nearly all of the results
in Chapters 15-18. The setting of Chapter 19 is more general, but operators of multipli-
cation by a potential can be considered, and, when they are, the potentials involved can
be both time-dependent and complex-valued.
Chapter 19 deals with time-dependent families {B(s) :0 < s < 00} of bounded oper-
ators on a Hilbert space. (A strongly continuous semigroup of operators on the Hilbert
space and the generator of that semigroup are also involved but are not particularly
relevant to the present comments.) Nonlocal potentials are used phenomenologically in
many body problems in several areas of quantum physics (see [Tab, ChSa, Mc] and the
relevant references therein). The operator is an integral operator whose kernel V(x, y)
(or V(s;x, y) if we have time-dependence) is referred to as a "nonlocal potential". It
is nonlocal in that this "potential" does not depend on one sharp choice for the space
coordinates (see formula (19.7.15)). Such nonlocal potentials are used, for example, in
nuclear physics where the kernels used to model various situations are surprisingly sim-
ple; they are, in practice, separable kernels of finite (and low) rank (see Example 19.7.5).
Finally, we mention that some of the highly singular potentials discussed just above
and treated in Section 11.4 and Sections 13.5-13.6 can also be viewed as providing
suitable phenomenological models for certain problems occurring in quantum physics
or in molecular chemistry. (See, for example, [LL, Nel, FrLdSp].) For instance, the
attractive inverse-square potential (Example 13.6.18) and more generally, highly singu-
lar attractive or repulsive central potentials (as in Examples 11.4.7 and 13.6.13), can be
used to model problems occurring in quantum field theory or in polymer physics. They
are often considered as "nonphysical" or only of academic interest because, in particular,
they may lead (as in Example 13.6.18) to nonunitary evolutions and thus to Schrodinger
operators which are no longer self-adjoint—in contradiction with one of the basic tenets
of standard Hamiltonian quantum mechanics. (This unusual aspect is apprehended natu-
rally within the context of the various approaches to analytic-in-mass Feynman integrals
discussed in Sections 13.5 and 13.6; see [Nel, Kat7, BivPi, BivLa].) Actually, the sit-
under U.S. or applicable copyright law.

uation is somewhat more complicated than that and a suitable dose of pragmatism is
needed to decide which model (whether of Feynman type or of Hamiltonian type, say) is
most appropriate for a given physical situation; see, for instance, [Case, R, FrLdSp] and
Example 13.6.18. In spite of these drawbacks, it can be argued convincingly that such
highly singular potentials provide better approximate (or "phenomenological") models
of suitable physical systems than their more regular counterparts. (See especially the
review article [FrLdSp] as well as, for example, [LL, Nel, PariZi, MarPari] and the
relevant references therein.)
In closing the main part of this introduction, we briefly return to Chapter 20 which,
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
as was mentioned
DISTRITAL FRANCISCO earlier, is of a very different nature than the rest of this book. We
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 RELATIONSHIP WITH THE MOTIVATING PHYSICAL THEORIES 21
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

recall, in particular, that in Section 20.2, we discuss some of the relationships between
heuristic Feynman-type integrals (as well as aspects of Feynman's operational calcu-
lus) and a variety of subjects from contemporary physics (or mathematics). In addi-
tion, in Section 20.2.A, several mathematical or physical models inspired by quantum
field theory (specifically, "Chern-Simons gauge theory" defined in terms of a heuristic
Feynman-Witten functional integral [Wit 14]), are used to gain insight into (and extend)
the celebrated Jones polynomial, along with other topological invariants that are central
to modern knot theory and low-dimensional topology.
We hope that despite its relative brevity, Section 20.2 will prove helpful to a reader
interested in getting a sense of the fascinating interplay between heuristic Feynman path
integrals and a number of topics lying at the border of mathematics and theoretical
physics.

Prerequisites, new material, and organization of the book

We end this introduction by making some specific comments about the content and the
structure of this book, along the lines suggested in the title of the present italicized
subsection.
As was mentioned in the preface and further explained earlier on in this chapter,
much of the background material needed for the main part of this book is provided here;
see especially Chapters 3,4, 6, 8-10, along with Chapter 7.
Detailed proofs—based mainly on the background material just referred to—are
given for nearly all the theorems which deal with the main topics of this book. Most of
the exceptions come in Sections 13.6, 13.7 and in the last part of Section 13.5, as well
as in Section 17.6.
The reader will see that proofs are provided even for a good portion of the background
material itself; see, in particular, Sections 3.1-3.4, 4.1-4.2, 4.5-4.6, and 10.2-10.3. We
remark that if the reader is willing to forego the proofs in Sections 11.6,13.5 and 13.6, then
the operator-theoretic background needed for the book (especially through Chapter 13)
is reduced to the information about self-adjoint operators and quadratic forms found
in Section 9.6 and in Chapter 10 plus relatively few basic facts about semigroups of
operators provided in Chapters 8 and 9.
We should mention that the comments in the preceding paragraphs do not apply to
Chapter 20; no attempt there is made to supply proofs. (In the case of Section 20.2,
under U.S. or applicable copyright law.

in which much of the material connected with Feynman path integrals is of a heuristic
nature, rigorous mathematical proofs are usually not known.)
The Lebesgue theory of measure and integration is employed in many places in
this book. Precise references are typically given for the results used, but no systematic
presentation of measure-theoretic prerequisites is provided. Brief discussions of this
subject can be found in the books [BkExH, Appendix A, pp. 531-544], [Ru2, pp. 5-75]
and [ReSi1, pp. 12-26]. Fuller treatments are given in many places, for example, in [Roy,
Fol2, Coh, Du].
The list of references provided at the end of this book is extensive but certainly not
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
complete.
DISTRITAL (We noteJOSE
FRANCISCO thatDE aCALDAS
significant fraction of the references is connected in some
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 22 INTRODUCTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

way with Chapter 20, which deals with a broad selection of topics.) When the references
are given in the main body of the text, they are typically presented in enough detail so
that the relevant material can be easily located. The topics discussed in this book are
interrelated in a variety of ways; we try to keep track of these relationships by systematic
cross-referencing.
A substantial part of the material in this book other than the background material has
appeared previously only in the research literature and, in a number of cases, only in the
recent research literature. A few results that will play a prominent role come from sources
that are not widely available. The primary example of that is the Sherbrooke monograph
of the first author [Jo6] which plays a key role in the operator-valued analytic continuation
in time results in Chapter 13.
There is a significant amount of novelty to the exposition in several places. For exam-
ple, in Section 4.6, we discuss in detail what is meant by the "nonexistence of Feynman's
measure" as well as related issues. Chapter 14 is an introduction to Feynman's opera-
tional calculus for noncommuting operators. This subject extends certain aspects of the
Feynman integral, a fact that does not seem to be widely understood in the mathematical
literature. We explain this in some detail in Chapter 14, and the idea is developed further
in Chapters 15-19.
It will be clear to the reader of this book that the research interests of the authors
have influenced much of the content. However, the influence went the other way as well;
the desire to fill in missing pieces of the book directed some of our research in recent
years. (For instance, reference [dFJoLa2]—on which Chapter 19 is based—was very
much written with our book in mind.) A portion of that work is new here. We wish to
call the reader's attention to a few such items.
In Section 13.4, we show that under rather general conditions, three of the four
approaches that are discussed in this book are closely related. Most of this material is
new.
The last part of Section 13.5 and nearly all of Section 13.6 deal with product formulas
and operator-valued analytic continuation in mass from such formulas (rather than from
the Wiener integral). A good portion of this material is new as well, particularly with
regard to the explicit connections with the Feynman integral. Further, a comparison of
the various analytic-in-mass Feynman integrals is provided, along with related results;
see Theorems 13.6.10 and 13.6.11, along with Corollary 13.5.18. In addition, a detailed
treatment of highly singular central potentials is given in this context; see Examples
under U.S. or applicable copyright law.

13.6.13 and 13.6.18.

In Section 15.7, a "time-reversal map" is introduced and studied for our disentangling
algebras in Chapters 15 and 18; see Definition 15.7.5, Theorem 15.7.6 and Corollary
15.7.8. This enables us, in particular, to clarify the connections with the usual "physical
ordering" in the context of Feynman's operational calculus. In turn, these changes have
repercussions in Chapter 17. We expect that more work will be done along related lines
in the future.
New examples are provided in several places. We mention one, Example 15.5.3, that
may be of particular interest. This example involves the purely continuous but singular
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
measure associated
DISTRITAL withDEthe
FRANCISCO JOSE Cantor function.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 RELATIONSHIP WITH THE MOTIVATING PHYSICAL THEORIES 23
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

At the end of Chapter 16, we indicate how results from our earlier work (contained
in [JoLal j) on "stability" in the measures can be extended. We carry this out in detail in
one case (see Proposition 16.2.14) and indicate how to go further in another case.
In Remark 11.5.15(d), we point out that the requirements on the negative part of
the dominating "potential" in Theorem 11.5.13 (from [Lal2]) can be reduced to mem-
bership in the "Kato class" of functions on Rd. (Theorem 11.5.13 is a dominated-type
convergence theorem for the modified Feynman integral; it is the subject of Section 11.5
and is also applied in Chapter 13 to other approaches to the Feynman integral.)
Additional work on Feynman's operational calculus by Brian Jefferies and the first
author ([JeJo]) is discussed briefly in III of Section 14.4. That work provides a nice
supplement to the treatment given in Chapters 15-19 of Feynman's operational calculus
(and based on [JoLal-4, Lal4-18, dFJoLal,2]). However, some aspects of the new
material still need further development and so a fuller discussion of this topic could not
be included in this book.
Several exercises or problems are proposed throughout the book. They are of vary-
ing degrees of difficulty. Typically, the exercises are mainly intended to illustrate a new
concept, apply a new technique, or supplement some material in the text. Most of them
should be accessible to graduate students. However, in a few instances, some of the pro-
posed problems are extremely difficult and not yet solved in the literature (e.g. Problem
11.3.9). In other cases, they correspond to results already published but the proof of
which is not discussed fully in the book (e.g. Problem 17.3.6 or 17.6.28). In addition, a
few open-ended problems—the precise interpretation or formulation of which is left to
the reader-are provided either formally (e.g. Problem 17.6.31) or in various comments
or remarks scattered in the text. When appropriate, we have usually indicated the nature
or the difficulty of the problem at hand.
The numbering system used in this book is straightforward. For example, Theorem
11.5.13 is the thirteenth numbered item in Section 5 of Chapter 11; a similar comment
applies to the numbering of equations. Further, Section 15.4 is the fourth section of
Chapter 15. Frequently, unnumbered subsections with italicized headings are used within
a given section in order to delineate or highlight certain topics.
Indexes for symbols or notation, authors, and subjects are provided just after the
bibliography. Along with the detailed list of contents, we hope that they will prove to be
a useful guide to the reader throughout this book.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

THE PHYSICAL PHENOMENON OF BROWNIAN MOTION

While examining the form of these particles immersed in water, I observed many of them very
evidently in motion; their motion consisting not only of a change of place in the fluid, manifested
by alterations of their relative positions, but also not infrequently of a change in form of the particle
itself; a contraction or curvature taking place repeatedly about the middle of one side, accompanied
by a corresponding swelling or convexity on the opposite side of the particle. In a few instances,
the particle was seen to turn on its longer axis. These motions were such as to satisfy me, after
frequently repeated observations, that they arose neither from currents in the fluid, nor from its
gradual evaporation, but belonged to the particle itself.
Robert Brown, 1827

In this paper it will be shown that according to the molecular-kinetic theory of heat, bodies of
microscopically-visible size suspended in a liquid will perform movements of such magnitude
that they can be easily observed in a microscope, on account of the molecular motions of heat.
It is possible that the movements to be discussed here are identical with the so-called "Brownian
molecular motion"; however, the information available to me regarding the latter is so lacking in
precision that I can form no judgment in the matter.
Albert Einstein, 1905 [Ei, p. 1]

We will still stay within the realm of experimental reality if, putting our eye on a microscope, we
observe the Brownian motion which agitates every small particle suspended in a fluid. In order to
obtain a tangent to its trajectory, we should find a limit, at least approximatively, to the direction
of the line which joins the positions of this particle at two successive instants very close to each
other. But, as long as one can perform the experiment, this direction varies wildly when we let the
duration that separates these two instants decrease more and more. So that what is suggested by
this study to the observer without prejudice, is again the function without derivative, and not at all
the curve with a tangent.
Jean Perrin, 1913 [Per, p. 27]
under U.S. or applicable copyright law.

2.1 A brief historical sketch

The distinguished English botanist Robert Brown made the first careful study of
"Brownian motion". In 1827, he was investigating the fertilization process in a cer-
tain species of flower. While looking at the pollen in water through a microscope, he
observed small particles in "rapid oscillatory motion". He examined the pollen of other
species with similar results. He first hypothesized that the motion was particular to the
male sexual cell of plants and next that the motion involved living matter. His experi-
ments with inorganic material showed that both of these are wrong. What does cause the
EBSCO Publishing
motion? Brown did : eBook Collection (EBSCOhost)
an experiment - printedsome
which refutes on 6/8/2017 5:02 PM via
explanations UNIVERSIDAD
that were put forth
DISTRITAL FRANCISCO JOSE DE CALDAS
wellAN:after his study; he immersed a drop of water containing one particle in oil and looked
98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
at itOperational
under theCalculus
microscope. The motion is observed as before. It is clear then for one thing
Account: ns000601
 A BRIEF HISTORICAL SKETCH 25
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

that the source of the motion is not a force of attraction between the particles. Brown also
noted that when several particles are present, they appear to move independently of one
another even when they are very close together. Brown's conjecture, carefully labeled
as such, was that matter is composed of small particles, which he calls active molecules,
which exhibit a rapid, irregular motion having its origin in the particles themselves and
not in the surrounding fluid. Considerably later, it became clear that the surrounding
fluid is the source of the motion.
A brief discussion of the work between Brown and Einstein is given in [Ne2,
pp. 11-13] and by Furth in Note 1 of [Ei, pp. 86-88]. These make interesting read-
ing and illustrate well that progress in science does not always proceed linearly. We
restrict our attention to a few positive contributions. Cantoni and Oehl found in 1865
that the movement continued unchanged for a year. S. Exner observed in 1867 that the
motion is most rapid with the smallest particles and is increased by light and heat rays.
In 1877, Delsaux expressed the now commonly accepted idea that Brownian motion is
caused by the impacts of the molecules of the liquid on the particles. Guoy found in
1888 that the motion becomes more lively as the viscosity of the fluid is decreased and
that, on the other hand, a strong electromagnetic field has no effect. Like Delsaux, Guoy
attributed the motion of the particle to the molecular motions of the fluid.
The main points made by observing Brownian motion are as follows:
(1) The path followed by the particle is continuous but extremely jagged.
(2) The particles move independently of one another.
(3) The motion is more active the smaller the particles.
(4) The composition and density of the particles have no effect.
(5) The motion is more active the less viscous the fluid.
(6) The motion is more active the higher the temperature.
(7) The motion never ceases.
It has been generally accepted for a long while now that the source of the motion is
the bombardment of the particle by the molecules of the surrounding fluid. The following
remark addresses the extent to which (1)-(7) above are consistent with this explanation.
Remark 2.1.1 (a) It is certainly reasonable that a particle under molecular bombard-
under U.S. or applicable copyright law.

ment would follow a continuous path. The extreme jaggedness of the path seems less easy
to account for but is plausible given the random and almost continuous bombardment
of the particle.
(b) Items (2) and (7) above are quite consistent with the molecular-kinetic theory.
(c) It is not immediately evident that (3) is consistent. How does one attempt to
explain it? Individual molecular hits do not produce observable motion. Such motion
occurs when there is a preponderance of hits in one or the other direction. The larger
the particle, the less often will the hits be sufficiently unbalanced to produce observable
motion.
(d) IfPublishing
EBSCO the motion comes
: eBook from (EBSCOhost)
Collection the bombardment
- printed onof6/8/2017
the particles by UNIVERSIDAD
5:02 PM via the molecules of
DISTRITAL FRANCISCO JOSE DE CALDAS
the fluid, is it reasonable that the density of the particles has no effect as (4) asserts? Not
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 26 THE PHYSICAL PHENOMENON OF BROWNIAN MOTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

entirely it seems. What probably happens is that within the range of densities, viscosi-
ties, etc. where Brownian motion is observed, (4) is approximately true. Relatively high
viscosity is present when Brownian motion is observed and so the velocity caused by a
bump (or bumps) is quickly damped out. This will also be relevant to our discussion of
"independent increments" in the next section.
(e) The viscosity of a fluid is a measure of internal friction and so it affects the ease
with which a particle moves through the fluid. As such, (5) is not surprising.
(f) Since increasing the temperature speeds up molecular action, (6) is certainly
consistent with the molecular-kinetic theory.
The quantitative theory of Brownian motion began with Albert Einstein. His five
papers on the subject between 1905 and 1908 made careful quantitative predictions
based on the molecular-kinetic theory of heat.
Einstein's first paper in 1905 was a prediction motivated by the molecular-kinetic
theory that a phenomenon with properties similar to Brownian motion ought to be observ-
able in nature. He was apparently just becoming aware of the earlier studies on Brownian
motion. He says, "It is possible that the movements to be discussed here are identical with
the so-called 'Brownian molecular motion'; however, the information available to me
regarding the latter is so lacking in precision that I can form no judgment in the matter."
On the other hand, his second paper in 1906 is titled "On the theory of the Brownian
movement". See [Ei], where his five papers on Brownian motion are reprinted. [In 1905,
Einstein formulated the special theory of relativity, studied the photoelectric effect and
wrote his first paper on Brownian motion—one of the most productive years in the history
of science; see [Srt], where his five 1905 papers are reprinted.]
Let p = p(u, t) be the probability density that a Brownian particle starting at the
origin at time 0 is at u at time t. Through physical reasoning, Einstein derived the diffusion
equation

where D is a positive constant called the coefficient of diffusion. A solution to this

equation is given by
under U.S. or applicable copyright law.

Hence, the probability that the particle is in say a cube E at time t

Remark 2.1.2 (a) That Einstein expressed position at time t in probabilistic terms was
not surprising since this is the framework of the molecular-kinetic theory. There are too
many particles involved to use Newtonian mechanics effectively.
(b) InPublishing
EBSCO his 1923 paperCollection
: eBook Differential space
(EBSCOhost) ([Wil],on reprinted
- printed inPM[Wi3,
6/8/2017 5:02 pp. 455-98]),
via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
Wiener obtained the formula (2.1.1) quite differently. He combined relatively elementary
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 A BRIEF HISTORICAL SKETCH 27
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

physical considerations with the central limit theorem to obtain (2.1.1). He does not
specifically invoke the central limit theorem, but it is clear that he had it in mind. We will
give a closely related discussion in the next section.
Contained in the formula (2.1.1) is the information that the mean square displacement
of a particle in time t is 2Dt. We will return to this fact shortly.
Assuming that the particles are spheres of radius a, Einstein also derived the formula

where n is the coefficient of viscosity, T is the temperature and k is Boltzmann's constant.

Note how points (3)-(6) earlier are reflected in the formula (2.1.1). [Einstein's papers
contain much more, but the key for us will be (2.1.1).]
In 1909, the French physicist Jean Perrin observed that the Brownian trajectories
appear to have no tangents and mentions the nowhere differentiable curves of Weierstrass.
This comment led to a beautiful theorem of Norbert Wiener.
Wiener [Wil] quoting Perrin [Per]: "One realizes from such examples how near the
mathematicians are to the truth in refusing, by a logical instinct, to admit the pretended
geometrical demonstrations, which are regarded as experimental evidence for the exist-
ence of a tangent at each point of a curve."
Are the paths traced out by the physical Brownian particles actually nondifferen-
tiable? We suppose not. Consider what is likely to happen to the difference quotient
h when h is orders of magnitude smaller than the average time between molecu-
lar hits. On the other hand, the paths of the Wiener process, the mathematical model of
Brownian motion, will turn out to be nondifferentiable. We remark that there is another
model for Brownian motion, due to Ornstein and Uhlenbeck, such that the associated
paths are differentiable [Ne2].
The Brownian motion studies played an important role in supporting the atomistic
theory at a time when this theory was much in doubt. In Einstein's formula (2.1.2), T
and n can be calculated and it can be arranged that the particles are all spheres of the
same radius a. Further, the mean square displacement 2Dt (see above) at time t = 1 can
be estimated statistically. The idea is simple: Observe the paths X j ( . ) , j = 1, . . . ,n, in
R3 of n independent Brownian particles starting at the origin at time 0 and record their
positions when t = 1. We then have
under U.S. or applicable copyright law.

Hence one arrives at a statistical estimate for Boltzman's constant k.

andEBSCO
so also for Avogadro's
Publishing number
: eBook Collection NA = -R/k,
(EBSCOhost) printedwhere R is 5:02
on 6/8/2017 the PM
universal gas constant.
via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
This was done by Perrin (and others) and the value was sufficiently close to the value
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 28 THE PHYSICAL PHENOMENON OF BROWNIAN MOTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

obtained by another quite different approach to be regarded as a significant confirmation

of the atomistic theory. (Recall that Avogadro's number, NA = 6 x 1023, is defined as
the number of atoms or molecules in a mole of substance.) Perrin received the Nobel
prize in 1926 in recognition of his work on Brownian motion.
Remark 2.1.3 Many of the conclusions drawn by Einstein concerning Brownian motion
were reached at about the same time by the Polish physicist, M. Smoluchowski [Smo].
Even earlier, L. Bachelier in his 1900 dissertation [Bac] arrived heuristically at some
of the same mathematical formulas. Bachelier's study was motivated by a very differ-
ent problem; he was attempting to analyze the French stock market. It was primarily
Einstein's work that influenced Norbert Wiener's later brilliant study of "Brownian
motion" ([Wil,2], [Wi3, §IC]).
We note that much of the material in this section is adapted from d'Abro [dA,
pp. 411-415], Einstein [Ei] (including the notes by Furth on pp. 86-88), and especially
from Sections 2 and 3 of Nelson's book Dynamical Theories of Brownian Motion [Ne2].

2.2 Einstein's probabilistic formula

The starting point in the introduction of Wiener measure is Einstein's probabilistic for-
mula (2.1.1) which we now recall:
The probability that a particle which starts at the origin at time 0 is in a (Lebesgue
measurable) set E at time t

In order to simplify matters, we concern ourselves for now with the motion of the particle
in a single direction. Further, we normalize the diffusion coefficient, taking 2D = 1, and
we take the set E (E c R now) to be the interval (a, B]. Formula (2.1.1) then becomes:
The probability that a particle which starts at the origin at time 0 is in (a, ft] at time t

Letting x = x ( t ) be the location of the particle at time t, we can rewrite (2.2.2) as

under U.S. or applicable copyright law.

In words, the right-hand side of (2.2.3) gives the probability that the path traced out by
the particle goes through the gap (a, B] at time t. (See Figure 2.2.1.)
We will have to go far beyond the simple sets of paths involved in (2.2.3) to develop
a satisfactory mathematical theory. Before doing this, however, we wish to give a largely
heuristic discussion of the appropriateness of (2.2.3) as a starting point for a mathe-
matical
EBSCO model of Brownian
Publishing motion.
: eBook Collection Some knowledge
(EBSCOhost) of introductory
- printed on 6/8/2017 5:02 PM viaprobability
UNIVERSIDAD theory
DISTRITAL FRANCISCO JOSE DE CALDAS
will be needed for what immediately follows but will not be necessary as we continue.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 EINSTEIN'S PROBABILISTIC FORMULA 29
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 2.2.1. The Brownian paths passing through the gap (or, B] at time t

, -u2-
The right-hand side of (2.2.3) involves the density function, -4= e 2r, of an N(0, t)
random variable; that is, a random "variable which is normally distributed with mean 0
and variance t. Let us try to get some idea why this density should appear.
Relatively high viscosity is assumed so that the effect of a single molecular hit on
the motion of a physical Brownian particle is quickly damped out. Hence, it is not
unreasonable to assume in the mathematical model of Brownian motion that x(t + s) —
x(t) is independent of x(t) for s > 0; i.e., the path x has independent increments. Also,
for any integer n, we can write

As long as the physical conditions (for example, the temperature) are unchanged as time
goes on, it seems reasonable that the random variables x(jt/n) - x ( ( j — l)t/n), j =
1 , . . .,n, have the same distribution. Thus we see from (2.2.4) that x(t) is written as the
sum of n independent, identically distributed random variables. The presence of a normal
density function is then not surprising in light of the central limit theorem. Further, for
under U.S. or applicable copyright law.

the physical particle, it seems equally likely that x (t) be positive or negative and so a zero
mean is reasonable. Now the variance of a random variable with mean 0 is the expected
(or average) value of its square. Also the variance of the sum of independent random
variables is the sum of the respective variances. Finally, if the physical conditions remain
constant, Var[x(s +t) — x(s)] = E[(x(S + t)— x(s))2] should be a function of t alone.
With these things in mind, we write

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 30 THE PHYSICAL PHENOMENON OF BROWNIAN MOTION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and so

Hence doubling t doubles the variance. Similarly, tripling t triples the variance, etc.
Thus it seems that Var[x(t)] depends linearly on t . Since Var[x(0)] = 0 , we see that
Var[x(t)] = ct. The constant c depends on the diffusion constant and c = 1 is appropriate
here because of our earlier normalization.
Summary: The physical properties of Brownian motion make the appearance of the
N(0, t) density in (2.2.3) (the starting point of the mathematical model) seem quite
reasonable.
Key ideas in the preceding discussion come from Wiener's 1923 paper [Wi1] or from
Lamperti's book [L1,§20].
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:02 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

WIENER MEASURE

Evolution of Mathematics is, by and large, a continuous process and its growth and progress seldom
deviate greatly from the natural historical lines. It is because of this that we tend, in retrospect,
to admire most those developments which though born well outside it have grown to join and to
enrich the mainstream of our science.
It was the great fortune and the great achievement of Norbert Wiener to initiate such a devel-
opment when, in the early twenties, he introduced a measure, now justly bearing his name, in the
space of continuous functions. . . .

In retrospect one can have nothing but admiration for the vision which Wiener had shown when,
almost half a century ago, he had chosen Brownian motion as a subject of study from the point of
view of the theory of integration. To have foreseen, at that time, that an impressive edifice could
be erected in such an esoteric corner of mathematics was a feat of intuition not easily equalled
now or ever.
It was Josiah Willard Gibbs, whom Wiener admired so much, who said that "one of the principal
objects of the theoretical research in any department of knowledge is to find the point of view from
which the subject appears in its greatest simplicity".
Integration in function spaces provided such a point of view over and over again in widely
scattered areas of knowledge and it gave us not only a new way of looking at problems but actually
a new way of thinking about them.
The fate of all great work is to be subsumed; the more attention it attracts the greater the
chances of becoming engulfed in a cascade of generalizations and extensions.
This is especially true today because of a growing tendency to believe that the latest improve-
ment supersedes all that preceded it and that a generalization constitutes a license to subsume.
It is therefore well to repeat that Wiener's contribution to the subject of integration in function
spaces will forever be the greatest because he had the idea first; and should anyone try to attribute
it to luck let him be reminded that it is the deserving ones who are also lucky.
under U.S. or applicable copyright law.

Mark Kac, 1966 [Kac3, pp. 52 and 68]

We give in this chapter an introduction to Wiener measure with a strong emphasis on

those aspects of the subject which are relevant further on in the book. The perspective is
that of an analyst rather than a probabilist. As such, it may serve as a useful introduction
to the fascinating Wiener process for those with a minimal background in probability
theory. Probabilists should browse through the chapter but will find that they can skip
over much of the material.
We should emphasize that our treatment of the Wiener process (or Brownian motion)
EBSCO Publishing
in this chapter : eBook
and Collection
the next is far(EBSCOhost) - printed on
from complete, We 6/8/2017
hope 5:02
that PMit via
willUNIVERSIDAD
stimulate some
DISTRITAL FRANCISCO JOSE DE CALDAS
readers to pursue the subject elsewhere. A few of the many references
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
available are
[Dool, Durr, Calculus
Operational Fre, Hid, L1, StroVa, Wil, Yen].
Account: ns000601
 32 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

3.1 There is no reasonable translation invariant measure on Wiener space

We begin by introducing some terminology and notation which will be used throughout
much of this book.
Definition 3.1.1 Given a positive integer d and real numbers a and b with a < b,
C([a, b], Rd) will denote the space of continuous functions on [a, b] with values in Rd.
Often, in a given discussion, we will have d — 1 or d will be arbitrary but fixed. If it
is clear from the context which range is intended, we will usually write C[a,b]or C a,b
rather than C([a, b], R.d). In fact, later on we will often take a = 0 and will write Cb in
place of C([0, b], Rd). The space of functions x in C([a, b], Rd) such that x(a) = 0 will
be denoted C0([a, b], Rd) (respectively, C0[a, b], Coa,b, Cbo). We will often refer to C%'b
as Wiener space in light of the role that it plays in connection with Wiener measure.
Given x in Ca-b, let

where \\x (t) || denotes the Euclidean norm of x (t) in R rf . It is well known that the function
|| -|| defined in (3.1.1) is a norm on Ca'b and that (Ca-b, \\ • ||) is a Banach space. Since C^
is a closed subspace of C a,b , it follows that (Ca,b , ||.||) is also a Banach space. By the
Weierstrass approximation theorem, the polynomials are dense in C ([a, b], R) and, from
this, it follows that the polynomials with rational coefficients are dense in C([a, b], R).
By considering components, one sees that C([a, b], Rd) is a separable Banach space.
Since every subspace of a separable metric space is separable, Co([a, b], Rd) is separable
as well.
The paths followed by the Wiener process, the mathematical model of Brownian
motion, over the time interval [a, b] will lie in the Banach space CQ'*. Wiener measure
will turn out to be a measure defined on a or-algebra containing the "Borel class" of CQ ' .
(The basic elements of measure theory used in this chapter—as well as later on in the
book—can be found in [Cho3, Coh, Roy, Ru2].)
Definition 3.1.2 Given any topological space T, B = B(T), the Borel class of T, is the
a-algebra generated by the open subsets of T.
It is the Borel class B — B(Ca' b ) of CQ' which presently concerns us.
Recall that Lebesgue measure on Rd is translation invariant; that is, Leb.(E + u) =
under U.S. or applicable copyright law.

Leb.(E) for every Lebesgue measurable subset E of Rd and for every u in Rd. As we
will see later, in formulating the "Feynman integral", Feynman thought in terms of a
translation invariant measure on the space of continuous functions. Wiener measure m
will play a crucial role in two of the approaches to the Feynman integral which we
will treat in some detail. The measure m does have the property of assigning positive
measure (or probability) to every ball in the metric space CQ'*, but it is not translation
invariant. Since our primary goal is a treatment of the Feynman integral, one might
reasonably think that we should be introducing a translation invariant measure on CQ' at
thisEBSCO
pointPublishing
rather than m. Collection
: eBook The trouble is that -there
(EBSCOhost) is no
printed reasonable,
on 6/8/2017 5:02 PMnontrivial translation
via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE bCALDAS
invariant measure on C^' . The following theorem makes that point sufficiently well
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 NO REASONABLE TRANSLATION INVARIANT MEASURE ON WIENER SPACE 33
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

for our purposes and is not hard to prove; however, there are stronger and much more
general theorems in the literature. (Theorem 3.1.5 which we will state below is a special
case of such a result.) Unless otherwise specified, all measures used here will be positive
(and countably additive).
Theorem 3.1.3 If a measure u on (€$' , B(€Q' )) is translation invariant and assigns
finite measure to some nonempty open set, then u is identically 0.
Proof We simplify notation by giving the proof for the case CQ = Co([0, 1], R). Since
u is translation invariant and any r-ball Br(y) is the translate of any other r-ball Br(z)
(in fact, Br(y) = Br(z) + (y — z ) ) , we see that all r-balls have the same u-measure.
(Here, Br (y) denotes the open ball of CQ of center y and radius r.)
Let O be a nonempty open set of finite u-measure. Given any x in O, there exists
ro > 0 such that Bro(x) c. O. By the monotonicity property of measures, u(Bro (x)) <
u ( O ) < oo. Using the translation invariance of u and invoking monotonicity again, we
see that if 0 < r < ro, u(B r (y)) < cc for all y in C01.
Now we define a sequence of functions {xn}in C01 as follows:

It is easy to see that \\xn|| = ro/2 and that \\xn — xm\\ = ro/2 for n = m. It follows that
Bro/4 (xn) n Bro/4(xm) = 0 if n = m and that B ro / 4 (x n ) c Bro(0) for every n.
Hence Bro (0) contains an infinite disjoint sequence of sets {Br0/4 (*n)} all having the
same measure and such that 0 < u(B r o /4(x n )) < oo. Since u(Bro(0)) < oo, we see that
u(B ro /4(x n )) = 0 for every n.
Now C10 is separable and so there exists a countable dense set {yj}^.] • It is easily
checked that CQ = Uj=1, B ro /4(y j ) and so

It follows that n is the identically 0 measure, as claimed. D

under U.S. or applicable copyright law.

Reflections on the proof of Theorem 3.1.3 lead one to suspect that there are related
results of a much more general nature. One such result can be found in the book of
Gelfand and Vilenkin [GelVi, Theorem 4, p. 359]. We will state a special case of this
result.
Definition 3.1.4 Let X be a Banach space and let u be a measure on (X, B(X)). We
say that u, is quasi-invariant provided that A in B(X) with n(A) = 0 implies that
u(A + x) — O for every x in X, Further, we say that u is a -finite if X =U00n=1Bn,with
Bn in B(X) and u (Bn) < oo.
Theorem 3.1.5 Let
EBSCO Publishing X be
: eBook a separable
Academic Collectioninfinite dimensional
(EBSCOhost) - printed onBanach
6/8/2017 space
5:27 PM and
via let u be a
UNIVERSIDAD
measure DISTRITAL
on (X, B(X))FRANCISCO
whichJOSE DE CALDASand quasi-invariant. Then u is identically 0.
is a-finite
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 34 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 3.1.6 If a measure u is translation invariant and a-finite on (€$' ,B(C^' )),
then u is identically 0.

3.2 Construction of Wiener measure

We will now outline the construction of one-dimensional Wiener measure. Except for
notational complication, the construction of cf-dimensional Wiener measure is much
the same, as will be discussed in more detail at the end of this chapter. Hence, unless
otherwise specified, we will assume in the following that d — 1.
We begin by stating another variation of Einstein's probabilistic formula (2.2.3) when
d = 1. Temporarily fix t so that a < t < b and suppose that —oo < a < B < +00.
The normalized and one-dimensional version of Einstein's formula for a particle which
starts at MO at time a is as follows (see Figure 3.2.1):

where

The function p(u, uo, t — a) is the N(U0, t —a) density. The mean MO and variance t — a
seem entirely appropriate here since the particle starts out at MO and the elapsed time is
t -a.
Based on the consideration above, Wiener's problem was to demonstrate the
existence of a countably additive probability measure m = ma,b on C0a,b such that
if a = to < t1 < • • • < tn < b and if cj, Bj are extended real numbers such that
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
FIG. 3.2.1. DISTRITAL
UNIVERSIDAD Paths starting atJOSE
FRANCISCO MO at
DE time
CALDASa and going through the gap (a, B] at time t
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONSTRUCTION OF WIENER MEASURE 35
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FiG. 3.2.2. Wiener paths starting at 0 at time a and going through the gap (a,-, Bj] at
time tj, for j = 1,2,3

Figure 3.2.2 illustrates the set of paths involved in (3.2.3) in the case n — 3. The first
factor in the integrand in (3.2.3) represents, as before, the probability density for being
at u1 at time t\ having started at 0 at time a. The second factor gives the probability
density for moving from u1 at time t1 to u2 at time t2. The appropriateness of this factor
comes in part from our assumption in the mathematical model for Brownian motion that
x(t2) — x(t\) is independent of x(t\).
We will almost always write the right-hand side of (3.2.3) using more abbreviated
notation such as one of the following:
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 36 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Subsets 7 of Coa,b as in the first expression in (3.2.4) will be called intervals and we
will write

The collection of all such intervals (also called "cylinder sets") will be denoted J.
Wiener's idea was to develop, using (3.2.3) (or (3.2.4)) as a starting point, a full
fledged Lebesgue type integral over the (infinite dimensional) space CQ' , thus making
available the powerful results of the Lebesgue theory. Wiener succeeded in this endeavor;
his achievement was a major advance which has had an amazing number of repercussions.
You will appreciate Wiener's contributions ([Wil,2], [Wi3, §IC]) all the more if you keep
in mind that his work was done when the Lebesgue theory was not exactly ancient history
and about a decade before the appearance of the book of Kolmogorov [Kol], which is
regarded as the beginning of probability theory as a separate discipline.
We will proceed to show that the collection T of intervals is a semi-algebra and m
is well-defined and additive on 1. In fact, although we will not carry out the proof, m is
countably additive on J. Thus the Caratheodory extension process ([Roy, Theorem 12.8,
p. 295] or [Con, Theorem 1.3.4, p. 18]) can be applied to obtain a countably additive
measure on a(I), the a-algebra generated by 1. Further, we will show that a (I) =
B(C^'b). Finally, the measure space (C^, B(C^'b), m) can be completed producing the
cr-algebra S1 of Wiener measurable sets and the complete measure space (C^' , S1, m).
There are several ways of introducing Wiener measure (see, for example, [Dool,
Durr, Hid, Kall, Nel, Wil, Yeh]). We choose the approach which we have just out-
lined since almost all mathematicians and mathematical physicists are familiar with the
Caratheodory extension process, at least on M. If one wishes to include the proof of
countable additivity, there are more efficient ways to proceed, and which you prefer is
somewhat a matter of background and taste.
We now turn to carrying out the steps outlined above. We begin by reviewing the
under U.S. or applicable copyright law.

definition of a "semi-algebra".
Definition 3.2.1 A collection R of subsets of a set X is called a semi-algebra if and
only if (a) 0 and X are in H, (b) A and B in K implies that A n B is in R, and (c) the
complement of any set in H is a finite disjoint union of sets in R.
Proposition 3.2.2 The collection T of intervals is a semi-algebra of subsets of C^' .

Proof (a) 0 = [x in C%'b : 1 < x(b) < 1} and C$b = {x in C%'b : -oo < x(b) <
+00} and so 0 and CQ' are in I.
(b) Let
EBSCO 7, J be: in
Publishing J. The
eBook set of
Academic restriction
Collection points -for
(EBSCOhost) 7 n Jonis6/8/2017
printed the union
5:27of
PMthe
via restriction
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
points for 7 and the restriction points for J. The restriction interval for any tj which is
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONSTRUCTION OF WIENER MEASURE 37
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

common to both I and J is the intersection of the restriction intervals. The restriction
interval for a tj which is a restriction point for one of the intervals, say /, but not for the
other, say J, is just the appropriate restriction interval associated with /.
We illustrate the situation in (b) with a simple example. Suppose that a < t\ < t2 <
t3, <b and that

and

(c) Let / be given by (3.2.6). We want to show that CQ' \I can be written as a
finite disjoint union of intervals. We will do this under the assumption that the intervals
( o j , B j ] , j = 1, . . . , n, are all finite and then will comment on the adjustments that
should be made if this is not the case. With the cj and f)j all finite, it is quite clear that
CQ'*\/ is the disjoint union of 2n intervals as follows:

Whenever any of the restriction intervals, say (c2, fa] = (—00, fa], is infinite, then
the key is that the corresponding two intervals in (3.2.7) can be reduced to one. For
example, the two intervals on the right-hand side of (3.2.7) whose restriction points are
exactly t1 and t2 get replaced by the single interval

The proof of the next proposition is a straightforward but rather tedious calculation
which we will just outline. However, the resulting equation, the Chapman-Kolmogorov
under U.S. or applicable copyright law.

equation, is crucial and will be used many times in the sequel. With future applications
in mind, we include a positive parameter A. although A. can be taken equal to 1 for a while.
Proposition 3.2.3 (Chapman-Kolmogorov equation) Let r, s and t be real numbers with
r < s < t and let A, > 0. Then

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 38 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof We begin by considering the exponent of the integrand in (3.2.8), but without the
minus sign. This is a quadratic function of v and, by completing the square, we obtain

where KI and K^ have their obviously intended meaning. Using (3.2.9) and the transla-
tion invariance of the Lebesgue integral, we see that the left-hand side of (3.2.8) equals

where the last equality in (3.2.10) follows from the well-known formula

(See (4.7.1) below.) Finally, some routine algebraic calculations show that

and so (3.2.8) is established. D

The formula (3.2.3) (or (3.2.4)) for m(7) is based on the representation (3.2.6) for /
under U.S. or applicable copyright law.

in terms of the numbers tj, a; and Bj. Such representations are not unique as we will
presently discuss and so there is some question as to whether m(7) is well-defined.
If there is a fixed set of ts involved, say Tn ~ [t\, ...,tn] with a = IQ < t\ < • • • <
tn < b, then it is clear that the numbers oij and f}j uniquely determine the interval /
and conversely. One can however always add t-values without changing /. Suppose, for
example, that s is in (a, b]\Tn. Then 7 can alternatively be written as

ButEBSCO
if one adds further
Publishing ts, the new
: eBook Academic "restrictions"
Collection (EBSCOhost) -must always
printed be —5:27
on 6/8/2017 oo and
PM via+00. Also
oneUNIVERSIDAD
can remove tj whenever
DISTRITAL FRANCISCO ay
JOSE=DE—oo and Bj = +00 but not otherwise.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONSTRUCTION OF WIENER MEASURE 39
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In spite of the ambiguities in the representation of /, we have the following:

Proposition 3.2.4 m is well-defined on I.
Proof There is a minimal representation for / such that for each j, at least one of
cj, Bj is finite. Of course, there is a corresponding expression for m(7). Any alternate
representation for / must involve additional points with the artificial restrictions — oo and
+00. We will show that, in the case of one additional point, the corresponding formula for
m(7) agrees with the formula associated with the minimal representation for I. The case
of N additional points with artificial restrictions can be done by applying the procedure
below N times.
Suppose that the minimal representation of / is given by (3.2.6) and that the extra
point, say s, satisfies tk < s < tk+1. (The argument takes a slightly different and easier
form if s is in (a, t) or (tn, b).) Then

and the corresponding formula for m(7) is

Now the integrand in (3.2.12) is positive and so we may apply Tonelli's theorem and
integrate out with respect to v. By the Chapman-Kolmogorov equation, the result is the
right-hand side of (3.2.3) as desired. d
Exercise 3.2.5 Show that m(C0([a, b], R)) = 1.
under U.S. or applicable copyright law.

Exercise 3.2.6 Define F : C a,b x [a, b] -> R by F(x, t) = x(t). Show that F is a
continuous function on the product space Ca'b x [a, b].
Proposition 3.2.7 (i) If is in I and I=0, then m(7) > 0.
(ii) m is finitely additive on I.
Proof (i) Let 7 be given by (3.2.6). If 7 = 0, then cj < Bj for j = 1, . . . , n. Hence
in formula (3.2.4) for m(7), a positive function is being integrated over a set of positive
Lebesgue measure. Thus m(7) > 0.
EBSCO Publishing
(ii) Let : eBook
7, J and 7U Academic
J be Collection
in I with(EBSCOhost)
7 n J =- printed
0. Byonadding
6/8/2017points
5:27 PMwith
via artificial
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
restrictions, we can always arrange to have 7 and J based on the same set of restriction
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 40 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

points. Accordingly, using the notation of (3.2.6), let / = If((a\, /3\] x • • • x (c n , B n ] )

and J = It((xi, 81 x • • • x (Yn, <5«]) where t — (t\,... ,tn) and a < t\ < • • • tn < b.
Since / n / = 0, we have (a\, fi\] x • • • x (an, /?„] n (yi, S\] x • • • x (/„, 5n] = 0.
Hence we can write

where the last equality holds since

Thus m is finitely additive on J, as desired. D

Remark 3.2.8 (a) In the proof above, since I (J J is assumed to be an interval, there
must be numbers Cj,dj, j = 1, . . . ,n such that (a\, f)\] x • • • x («„, ftn] U (y\, S\] x
• • • x (yn, Sn] = (c\, d\] x • • • x (cn, dn]. However, it is not necessary for the purposes
of the proof to identify the Cj and dj.
(b) It may be helpful to some readers to compare the present construction of Wiener
measure m to the more familiar construction of ordinary (one-dimensional) Lebesgue
measure on R. (See, for example, [Roy, Chapter 3] or [Coh, Chapter 1].) Naturally, in
this simpler situation, X is replaced by the semi-algebra of intervals of IR of the form
(c, d}.
The next step is to show that m is countably additive on the semi-algebra T. (That is,
if {/fc}^! is a sequence of mutually disjoint intervals in I and if / := U£l[ 4 is also in
Z, then m(7) = Y^=\ m (^)-) As indicated earlier, we will just state this result without
proof. The theorem is essentially due to Wiener although he took a somewhat different
approach to these matters.
under U.S. or applicable copyright law.

Theorem 3.2.9 (Wiener) The set function m is countably additive on T and so (by the
Caratheodory extension theorem) has an extension to a countably additive measure on
a (Z), the a-algebra generated by I.
There are several approaches to proving the countable additivity of Wiener measure,
some of which can be found in [Doo, Hid, Kall, Nel, Wil, Yeh]. Nelson's approach [Nel,
Appendix A, pp. 338-340] may be particularly appealing to readers with a functional
analytic background. Kallianpur gives a short proof on pages 13-14 of his book [Kall]
as well as a good deal of related information in Sections 1.2 and 2.1.
NextPublishing
EBSCO we wish:toeBook
show that a(2)
Academic = B(C^'
Collection ). The
(EBSCOhost) following
- printed exercise
on 6/8/2017 will
5:27 PM viahelp us do
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
that and will be useful many times as we continue.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONSTRUCTION OF WIENER MEASURE 41
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Exercise 3.2.10 (a) Let X and Y be sets and let £ be a a-algebra of subsets of X. Also
let f : X -> Y be a function from X to Y. Define A := {B c Y : f - l ( B ) e £}. Show
that A is a a-algebra of subsets of Y. (Clearly, f is £ — A measurable because of the
way A is defined.)
(b) Let X and Y be topological spaces and let f be a continuous function from X to
Y. Show that f is Borel measurable; i.e., f-1 (B) e B(X)for every B & B(Y).
Theorem 3.2.11 a (I) = B(C^b).
Proof Let a < t < b and define P, : C^b ->• R by

It is easy to see that Pt is continuous. (This also follows from Exercise 3.2.6.) Hence, by
Exercise 3.2.10(b), P, is Borel measurable.
Now for an arbitrary interval / e I given by (3.2.6), we can write

It follows from (3.2.15) and the Borel measurability of each Ptj that / is in B(C^b).
Hence I is a subclass of B(C%'b) and so a(T)C. B(CQ~b) as desired.
It remains to show that H(CQ' & ) c a (I), and, for this, it suffices to show that every
open subset of CQ' is in a(T). Now CQ' is a separable metric space and such spaces
have the property that every open subset of them is a countable union of open balls.
Thus it suffices to show that every open ball is in a(T). But every open ball Br(xo) is
the countable union of closed balls; specifically,

Thus it suffices to show that every closed ball, say Bs(yo), is in a (I).
Now pick a countable set of t-values, [t1, t2, . . .}, which is dense in [a, b]. For each
under U.S. or applicable copyright law.

Each KN belongs to a (I) since

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 42 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Hence to finish the proof it suffices to show that

It is clear that B& (yo) c KN for every N and so the proof will be finished if we show
that

Suppose that y £ Bs(yo). Then there exists .y e (a, b] and 8\ > 0 such that \y(s) —
yo(s)\ = S + Si. Let {tjk} be a sequence from the dense set [t\,t2,...} such that tjk -*• s.
Then y(tjk) -> y ( s ) and yo(tjk) -*• yo(s). Hence there exists k such that \y(tjk) —
yo(tjk)\ > 8 + (Si/2). Thus y £ KN for any N such that tjk e (t\, . . ., tN}. Therefore

The Caratheodory process actually carries us beyond the a -algebra generated by I

to its completion with respect to m; i.e., to the <r-algebra of Wiener measurable sets. Our
next theorem summarizes Theorems 3.2.9 and 3.2.11, Exercise 3.2.5 and the fact just
mentioned.
Theorem 3.2.12 m is a complete, countably additive probability measure on the
a-algebra S\ of Wiener measurable subsets of C^' . In fact, (C%' , S1, m) is precisely
the completion of (C°'b, B(C^'b), m).
Remark 3.2.13 It will be helpful for later purposes to recall two properties of the com-
plete space (CQ* , S1, m): (i) If A e S1 and m(A) = 0, then every subset E of A is in
S1 andm(E) = 0. (ii) S1 consists of all sets of the form BUN, where B e B(C^'b) and
N is a subset of NO with NO e B(C%'b) and satisfying m(No) = 0.
We now have available the full power of the abstract Lebesgue theory of integrals as
we continue to study and make use of the space (€$' , S1, m).

3.3 Wiener's integration formula and applications

under U.S. or applicable copyright law.

We will for now fix t = (t1 , . . . ,tn) such that a < t1 <. . .< tn< b. Given any subset
E of Rn, we let

It is clear that with ? fixed, the sets E and Jf (E) determine one another uniquely.

Note that for any subset E of M"

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 WIENER'S INTEGRATION FORMULA AND APPLICATIONS 43
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proposition 3.3.1 The function Ptl tn is continuous and Borel measurable. Thus
Jj(B) e B(C°'b) for every B e B(R").
Proof It is easy to see that Ptl ,...,tn is continuous and so, by Exercise 3.2.10(b), it is Borel
measurable. Hence, for any Borel measurable subset B of R", J^(B) = Pt~^^tn(B) e

Finitely based functions

We wish to have a formula for integrals of finitely based functions:

The abstract change of variable theorem [Coh, Proposition 2.6.5, p. 82] will be helpful
in this regard and we now review it.
Let (X, A, /z) be a measure space, let (Y, C) be a measurable space, and let g : X -»•
Y be A — C measurable; that is, g is measurable from (X, .4) to (Y, C). Define /z o g~l
on C by (/z o g ~ l ) ( C ) = fJ,(g~l(C)). It is easy to show that /z o g~l is a measure on C;
it is called the image measure or, more fully, the image of /x under g.
Theorem 3.3.2 (Change of variable theorem) Let (X, A, (J.) be a measure space, let
(Y, C) be a measurable space, and let g : X -»• Y be A - C measurable. Let f be an
extended ^.-valued C measurable function on Y.
Then f is integrable with respect to the image measure n, o g~l if and only if f o g
is ii-integrable. If these functions are integrable, then

Further, (3.3.4) holds in the strong sense that if either side is defined (even if it is not
finite), then the other side is defined and agrees with it.
Proposition 3.3.3 Let B e B(Rn). Then
under U.S. or applicable copyright law.

Proof First note that Jj(B) e B(C$b) by Proposition 3.3.1 and so the left-hand side
of (3.3.5) makes sense. Also the function Wn(f, •) given by (3.2.5) is positive and
/Rn Wn(t, U)dU = 1. Hence, the measure

induced by the density function Wn (?, •) and Lebesgue measure is a probability measure
on EBSCO Publishing
the class : eBook of
Leb.(M") Academic Collection
Lebesgue (EBSCOhost) subsets
measurable - printedofon R".
6/8/2017 5:27 PMby
Further, via(3.2.3) and
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
(3.3.3), v (-) agrees with the image measure m o P~' _ (•) = m(Jj(-)) on the intervals
AN: 98476n ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman (nIntegral and Feynman's
Operational Calculus
Account: ns000601
 44 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(«i, B1] x • • • x («„, f)n], a collection of subsets of R" which generates B(M") [Coh,
Proposition 1.1.4(c), p. 5]. It thus follows [Coh, Corollary 1.6.2, p. 45] that

on B(E"). The equality (3.3.5) now follows. D

Corollary 3.3.4 The function Ptl ,„ is Si(CQ'b) - Leb.(M") measurable. Thus
/?(£) e Si (Cg'6) for every £ e Leb.(R") and, further,

Proof First suppose that Leb.(A^) — 0. Then, as follows from [Roy, Proposition 12.7,
p. 294], N c NO, where N0 6 B(W) and Leb.(AT0) = 0. Thus, by Proposition 3.3.1,
p
t^...,tn (^o) e B(C^'b). Hence by Proposition 3.3.3 (and formula (3.3.3)),

But Jj(N) c Jj(No) and so, by the formula above and the completeness of (CQ'*, S1, m),
we see that Jr(N) e S1 and m(Jj(N)) = 0 = fN Wn(t, U)dU.
Now suppose that £ is an arbitrary set in Leb.(R"). Then [Roy, Proposition 12.7,
p. 294] E = BUN, where B e fi(R") and Leb.(^V) = 0. Since

we see from Proposition 3.3.1 and the preceding part of the proof that J?(E) e S\.
Further
under U.S. or applicable copyright law.

as desired. D
Theorem 3.3.5 (Wiener's integration formula) Let f : R" —> R be Lebesgue measur-
able. Then, with Wn(7, U) given by (3.2.5), we have

where the equality is in the strong sense that if either side of (3.3.9) is defined (in the
sense of the
EBSCO Lebesgue
Publishing theory),
: eBook Academicwhether
Collectionfinite or infinite,
(EBSCOhost) thenonso
- printed is the 5:27
6/8/2017 otherPM side
via and they
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
agree.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 WIENER'S INTEGRATION FORMULA AND APPLICATIONS 45
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof Using the change of variable theorem (Theorem 3.3.2), we can write

where the last equality comes from Corollary 3.3.4 which tells us that the image measure
m o Pr~ tn is the same as the measure given by the density Wn(t, U) and Lebesgue
measure on R". D
Remark 3.3.6 (a) The right-hand side of (3.3.9), and hence also the left-hand side,
is finite for many functions f which grow rapidly at oo. On the other hand, if f has
a singularity at a point (u\,..., ~u~^) e W which is not LI with respect to Lebesgue
measure, the function f(-)Wn(t, •) has a non-L1 singularity at the same point.
(b) Formula (3.3.9) reduces an integral over the infinite dimensional path space CQ'
to an ordinary Lebesgue integral over the finite-dimensional space K". Such a reduction
is not possible in general; it works here because the function x H>• / (x(t\),..., x(tn))
depends only on the value of the path x at n points.
In spite of its finite-dimensional nature, (3.3.9) is the starting point for various Wiener
integration formulas, for example, the Cameron-Martin translation theorem [CaMal],
which are truly infinite dimensional in character. One begins by applying (3.3.9) where
larger and larger finite subsets of a dense set of f-values are involved.
Applications
Next we turn to some examples illustrating the use of the formula given in Theorem 3.3.5.
Proposition 3.3.7 Let t satisfying a < t < b be fixed. Then

Proof The function /(MI) := u\ is Lebesgue measurable and u\W\(t, u\) is Lebesgue
integrable. It follows from Theorem 3.3.5 that Pt(x) is S\-measurable and Wiener inte-
under U.S. or applicable copyright law.

grable and that

where the last equality in (3.3.12) holds since the integrand in the second expression in
(3.3.12) is an odd function. n
Remark 3.3.8 (a) We know from Corollary 3.3.4 that P, is Si-measurable. We men-
tioned
EBSCOitPublishing
in the proof above
: eBook to emphasize
Academic that Wiener
Collection (EBSCOhost) measurability
- printed on 6/8/2017 is a part
5:27 of what can
PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
be concluded from Theorem 3.3.5.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 46 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(b) In the notation of expected (or average) values, (3.3.11) becomes

Formula (3.3.13) says, in the language of stochastic processes, that the mean function
of the Wiener process is 0. Our next application of Theorem 3.3.5 is a calculation of the
"covariance function", Cov(x(t\,), x ( t 2 ) ) , of the Wiener process. We will carry out this
computation in some detail but will often abbreviate similar arguments as we continue.
Proposition 3.3.9 Let t1, ti belong to (a, b] with t\ = t2. Then F(x) := x(t1) • x(t2) is
a Wiener integrable function on C^b and

Proof Suppose for definiteness that a < t\ < t1 < b. It suffices to show that

Let f(u1, u2) = u1 • u2- Since / is Lebesgue measurable and f(-)W2(f, •) is Lebesgue
integrable over R2, we have that F is Wiener integrable and

Applying these facts to the right-hand side of (3.3.16) and then using the Fubini theorem,
we obtain
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 WIENER'S INTEGRATION FORMULA AND APPLICATIONS 47
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the last equality in (3.3.17) results from the fact that V2 K-»- V2e~v* is an odd func-
tion. Now using the earlier integration formula (3.2.11) as well as the formula (established
in (4.7.2) of Appendix 4.7)

we can finally write

as claimed.
A simpler calculation than the above establishes the following:
Exercise 3.3.10 Let t be such that a < t <b. Then

Remark 3.3.11 The variance of a random variable X is defined as Var(X) :— E(X —

Mx)2. where fj,x = E(X) is the average value of the random variable. By Proposition
3.3.7, E(x(0) = 0 and so (3.3.19) tells us that Var(;c(r)) = t - a. Similarly, the
covariance of two random variables X and Y is defined by Cov(X, Y) := E[(X —
Mx)(y~M'J')]' Hence Proposition 3.3.9 tells us that Cov(jc(fi), x f a ) ) = min{fi, 12}— a.
If X and Y are separable metric spaces, then the Borel class of X x Y equals the
product a -algebra B(X) <8> B(Y), where

B(X) <g> B(Y) := a({Bi x B2 : Si e B(X) and B2 6 B ( Y ) } ) . (3.3.20)

This fact will be proved in Appendix 3.6 and will be helpful in connection with our next
under U.S. or applicable copyright law.

proposition.
Proposition 3.3.12 Define F : C^'h -»- E by

Then F is Wiener integrable and

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 48 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof The map (x, t) (-»• x(t) is continuous by Exercise 3.2.6, Thus the function
G(x,t) := x2(t) is continuous and so B(CQ<b x [a, &])-measurable. But Cg' b and
[a, b] are both separable metric spaces and so B(Cg'fo x [a, b]) = B(CQ'b) <g> B([a, b]).
Hence G is B(C°'h) ® B([a, b])-measurable and so is certainly S1 (C°'b) <8> Leb.([a, b})-
measurable. Using the facts just noted, the Tonelli theorem, and Exercise 3.3.10, we can
write

and thus (3.3.22) is established. D

Note that there are no finite set of f-values, a < t\ <•••<(„< b, such that the
vector (x(t\),..., x(tn)) determines the function F; i.e. the function denned in (3.3.21)
is not "finitely based". Also note that F is continuous and nonnegative and so it is clear
from the start that the left-hand side of (3.3.22) is defined.
Exercise 3.3.13 Let a < t1 < t2 < t3 < b. Calculate

Exercise 3.3.14 Let p > 0 and f e R be given and suppose that 0 : [a, b] x E -> R is
bounded and continuous. Further, let

for x € C^'. Show that F e Ll(C^' , m) and find an expression for fca,h Fp^(x)dm(x)
under U.S. or applicable copyright law.

in terms of ordinary Lebesgue integrals.

Exercise 3.3.15 Show that

Remark 3.3.16 Exercises 3.3.14 and 3.3.15 preview calculations that will appear fur-
therEBSCO
on. Publishing
We will eventually see that
: eBook Academic the formula
Collection that- comes
(EBSCOhost) out6/8/2017
printed on of Exercise
5:27 PM3.3.14
via is valid
UNIVERSIDAD DISTRITAL FRANCISCO
with much weaker hypotheses on 6. JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 WIENER'S INTEGRATION FORMULA AND APPLICATIONS 49
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Let (£2, A, P) be a probability space and let X : £2 —> R be a random variable (i.e.
a measurable function) on £2. Recall that the distribution function FX of X is defined for
every /3 e R by

If X, : £2 -> R, y = 1 , . . . , n, are n random variables, then the joint distribution

function FXl *„ of X i , . . . , Xn is defined for all (A,..., $,) € R" by
Fx1,...,xn(B1, . . . . ., Bn) := P({a> e S2 : X1(o>) < B , . . . , X«(o>) < #,}). (3.3.26)
Our last two examples (for the present) will deal with the calculation of certain special
distribution functions. These examples are related to one another and to a discussion
which will soon follow.
Proposition 3.3.17 Let a < t\ <t2<b and let B be any real number. Then

where FX(t2)-x(t\) denotes the distribution function of the random variable G(x) :=
x(t 2 )-x(t1).
In the language of probability theory, (3.3.27) says that the random variable G is
distributed normally with mean 0 and variance t2 — t1; i.e. G ~ N ( 0 , t2 — t1).
Further, since the distribution of G does not depend on t[ and ti but only on ti — t\,
one says that the Wiener process has stationary increments.
Proof First note that
under U.S. or applicable copyright law.

It is easy to check that the linear change of variables t>i = u\, V2 = u^ — u\ has Jacobian
equal to 1 and carries the region B onto R x (—oo, /J]. We see from this, the Fubini
theorem and (3.2.11) that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD
which DISTRITAL result.
is the desired FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 50 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Recall that the random variables X\,..., Xn are said to be independent if and only
if

and so the random variables x(t\), x (t2) — x(t\), . . . , x(tn) — x(tn-\) are independent.
Proof We limit our attention to the case n = 2 which conveys the general idea rather
well. By (3.3.8) we can write

where D := {(u1, u2) € K2 : —oo < u1 < B1, —oo < u2 — u1 < fa}. The same
change of variables as in the proof of Proposition 3.3.17 carries the region D onto
(-00, p\\ x (—00, fa}. Thus using the Fubini theorem and Proposition 3.3.17, we have
under U.S. or applicable copyright law.

as desired. n
Remark 3.3.19 Our last two propositions dealt with distribution functions of random
variables. However,
EBSCO Publishing theAcademic
: eBook distribution measure
Collection (or,- simply,
(EBSCOhost) printed onthe distribution
6/8/2017 of the ran-
5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
dom variable) is the more fundamental object. The distribution measure of the random
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 NONDIFFERENTIABILITY OF WIENER PATHS 51
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

variable X on the probability space (Si, A, P) is just the image measure P o X~l
on R. The joint distribution of random variables X\,..., Xn is the image measure
Po(Xi,..., Xn)~l on Rn, where (X1, . . . , Xn) denotes the Rn-valued random variable
which maps co to (X\ ( a > ) , . . . , Xn(a>)). Note that the distribution function is expressed
simply in terms of the distribution measure: Fx(fi) = (PoX~l)(—oo, ft]. Also, since the
intervals of the form (—00, /J], ft G K, are a generating class for B(R), the distribution
function FX completely determines the distribution P o X~'. Similar remarks apply to
the joint distribution; in particular,

The condition (3.3.30) for the independence of the random variables X\, . . ., Xn, is
equivalent to the assertion that the distribution measures satisfy the relationship

(Equation (3.3.33) corresponds closely to the intuitive idea of the independence of

X 1 , . . . , Xn, as the reader familiar with elementary probability theory can verify.)
Axiomatic description of the Wiener process
In books on probability theory or stochastic processes, for example [KalKar, L2, Wil],
a "stochastic Wiener process" is often defined in terms of axiomatic properties. One
choice of the list of properties is as follows: a (standard) Wiener process on a probability
space (£2, A, P) is a family of random variables {xt : a < t < b] satisfying:

Our set-up provides one particular realization of these properties. For us, (£2, A, P) =
(C^'b, S1, m) and x t (w) is replaced by x(t) where x plays a dual role as a path (or
function on [a, b]) and as an element of £2. Properties (i) and (iv) in (3.3.34) are satisfied
in our setting for every a> since our space £2 = CQ' consists entirely of continuous
functions which are 0 at a. Propositions 3.3.18 and 3.3.17 assure us that properties (ii)
and (iii) hold.
under U.S. or applicable copyright law.

Remark 3.3.20 (a) The time interval involved above need not be finite. The intervals
[0, +00) and [0, b] are the most commonly used.
(b) When alternatives to the list of axioms in (3.3.34) are given, they often include
the covariance function. Proposition 3.3.9 and Remark 3.3.11 identified this function
for us.
3.4 Nondifferentiability of Wiener paths
Wiener paths are, with probability 1 (that is, a.s.), everywhere continuous but nowhere
differentiable. In:light
EBSCO Publishing eBook of the remarks
Academic Collectionof(EBSCOhost)
Perrin and the earlier
- printed discussion
on 6/8/2017 of our math-
5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
ematical model of physical Brownian motion, the nowhere differentiable nature of the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 52 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Wiener paths is not a complete surprise. Recall, however, that it was long believed that
a continuous function must be differentiable except possibly on some "small" subset
of the interval in question. Of course, the famous example of Weierstrass [Fal, §11.1]
showed that this is not true, but the result for Wiener paths shows that everywhere con-
tinuous, nowhere differentiable functions exist in great abundance. To put it somewhat
facetiously, if instead of constructing his function, Weierstrass had selected at random
a function from the "urn" C^'b of continuous functions, he would have been incredibly
unlucky, in the sense of Wiener measure, to pick a function that was differentiable at
even a single point.
The result that we will actually prove in this section is that Wiener paths are, with
probability 1, differentiable at most on a set of Lebesgue measure 0. This is easier to
establish and makes the point sufficiently well for our purposes. An important corollary
is that Wiener paths are almost surely not of bounded variation on any subinterval.
We will see later that the nondifferentiability (a.s.) of the Wiener paths has implica-
tions for the theory of the Feynman integral.
We begin by introducing some sets that will be useful to us. Given h > 0, 0<y < l,
and t, t' in [a, b], let

Further, let

and

Lemma 3.4.1 The sets CYh(t, t'), CYh(t), andCYh in (3.4.1)—(3.4.3) above are all dosed
subsets o/Co' fc and so are in B(C^b).
under U.S. or applicable copyright law.

Proof Let [xm] be a sequence in CYh(t, t') such that \\xm — x\\ -> 0 as m -> oo. Since
xm e CYh(t,t'),xm(t) - xm(t') e [-h\t -t'y,h\t - t'\Y] for m = 1 , 2 , . . . . Hence
x(t)-x(t') = lim [jc m (0--tm(«')]isalsointheclosedinterval[-/i|r-r / |> / ,ft|r-f'| )/ ].
m—»oo
Thus CYh(t, t') is a closed subset of CQ' & . Since the intersection of an arbitrary family of
closed sets is again closed, it follows that CYh (t) and Cvh are also closed. D
Lemma 3.4.2 For t ^ t', we have the inequality

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 NONDIFFERENTIABILITY OF WIENER PATHS 53
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof We may assume without loss of generality that t' > t. We will also assume that
a < t. The case a = t is easier as the reader can check. From (3.3.8) of Corollary 3.3.4
and (3.4.1) above, we see that

where

Hence (3.4.4) holds. D

Lemma 3.4.3 If y > 3, then m(C^(t)) = 0 and so, of course, m(C£) = 0.
Proof Let {tk} be a sequence of points in [a, b] which are distinct from / and are such
that tk —*• t as k —> oo. By (3.4.2) and Lemma 3.4.2, we have
under U.S. or applicable copyright law.

for all k. Since the last expression in (3.4.7) goes to 0 as A: -» oo, we see that
m(Cyh(t))=0. D
Although it is not the main point that we are driving towards, we can easily prove
a corollary of Lemma 3.4.3 which is of some interest. Let 0 < y < 1. Recall that a
function x : [a, b] -» R is said to be Holder continuous of order y if and only if there
exists a positive constant h such that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
for UNIVERSIDAD
all t, t' in DISTRITAL
[a, b]. FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 54 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 3.4.4 Let | < y <1.The set of functions in C^'b which are Holder continu-
ous of order y is a Borel subset of C^ which has Wiener measure (or probability) 0. In
other words, almost surely, a Wiener path is not Holder continuous of order y > j (and
hence cannot be continuously differentiable).
Proof The result follows from Lemma 3.4.3 and the equality

We simply state one further related result with a rather different conclusion. (For a
proof, see, for example, [Si9, p. 45].)
Theorem 3.4.5 Let 0 < y < |. The Wiener measure of the set of functions in Cg'
which are Holder continuous of order y is 1. In other words, a Wiener path is almost
surely Holder continuous of order y < j.
The next lemma returns us to the main line of our development.
Lemma 3.4.6 Let a < t < b and let D, := {x e C^b : x'(t) exists}. Then m(D,) = 0.
(Here, x'(t) denotes the ordinary derivative of x at t in (a, b); further, if t = a or t — b,
the appropriate one-sided derivative is used in the definition of Dt.)
Proof We begin by showing that

where C^(t) is defined by (3.4.2) with y = 1. If x e D,, then the difference quotient
*frf}~*^ is bounded for t' € [a, b]\{t}. Hence there exists a positive integer h such
that \x(t') - x(t)\ < h\t' - t\ for every /' € [a, b]; that is, x e C^(f). Thus (3.4.9) is
established. From (3.4.9) and Lemma 3.4.3 we obtain
under U.S. or applicable copyright law.

and so the proof is complete.

We come now to the main theorem of this section.
Theorem 3.4.7 Paths x e CQ' are, with probability I, differentiable at most on a subset
of [a, b] ofLebesgue measure 0.
Proof Define F : C°'b x [a, b] -» R by

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 NONDIFFERENTIABILITY OF WIENER PATHS 55
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A major part of the proof will be devoted to showing that F is measurable as a function
of x and t. For this, it suffices to show that the set

is of measure 0 with respect to the completed product measure tn x Leb. Let

where it is convenient to take x(s) = x(b) for s > b. Now each /„ is continuous as a
function of x and t and so is certainly measurable. Let

G* is measurable since it is the set where a sequence of measurable functions has a

finite limit. Since G c G*, to show that (m x Leb.)(G) = 0, it suffices to show that
(m x Leb.)(G*> = 0. But, by the Fubini theorem,

where G*t = {x e C%b : (x, t) e G*} is the f-section of G*. We will show that m(G*) =
0 for all a < t < b, and from (3.4.14), it will then follow that (m x Leb.)(G*) = 0 as
desired.
Let a < t < b. Of course,

For every positive integer h, let

Clearly G* C (J°i, Kh(t). Hence to show that tn(Gf) = 0, it suffices to show that
m(Kh(t)) = 0 for h = 1,2, ....But
under U.S. or applicable copyright law.

Using (3.4.15) and Lemma 3.4.2, we see that for every n ,

Since
EBSCOthe last expression
Publishing in Collection
: eBook Academic (3.4.16) (EBSCOhost)
converges- to 0 ason «6/8/2017
printed —»•5:27oo, it follows that
PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
m(Kh(t)) = 0 as desired.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 56 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Now that we have the measurability of the function F given by (3.4.10), we can
apply Fubini's theorem and write

But, in the notation of Lemma 3.4.6, F(x, t) — XD (•*)• Hence from (3.4.17) and Lemma
3.4.6 we obtain

From (3.4.18) it follows that for m-almost every x, fa F(x, t)dt = 0. Hence for m-
almost every x, F(x, 0=0 for Leb.-almost every t. It follows that for m-almost every
x, the derivative x'(t) exists at most on a set of Lebesgue measure 0 as claimed in the
theorem. n
We can now easily prove the following corollary.
Corollary 3.4.8 Wiener paths are, with probability 1, of unbounded variation on every
subinterval of [a, b}.
Proof If a function x is of bounded variation on any interval, then it is differentiable
Lebesgue almost everywhere on that interval [Roy, Corollary 5.6, p. 104]. Hence

But the right-hand side of (3.4.19) has Wiener measure 0 and so the left-hand side does
as well. D
Remark 3.4.9 (a) The paths of physical Brownian motion are wildly varying, but it
seems most unlikely that they have infinite variation on every time interval, no matter
under U.S. or applicable copyright law.

how small.
(b) A good deal is known about the properties possessed, with probability 1, by
Wiener paths. (See [Durr, Fre, Tayl3], for example.) Many of these results are math-
ematically appealing and definitely nontrivial. We will not need much more along these
lines, however, and will discuss what we do need as the occasion arises.
We just state (for d = 1) two further sample path results below. These theorems
emphasize the gap between a typical Wiener path and the paths (or functions) which we
ask students to deal with in elementary mathematics.
Theorem 3.4.10 :With
EBSCO Publishing eBookprobability 1, the local
Academic Collection maxima
(EBSCOhost) of a Wiener
- printed path
on 6/8/2017 form
5:27 a countable
PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
dense set.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: CONVERSE MEASURABILITY RESULTS 57
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 3.4.11 With probability \, the zero set, Z(jc) := {t € [a, b] : x(t) = 0}, of a
Wiener path x is perfect (i.e. closed and without isolated points) and so is uncountable,
but has Lebesgue measure 0.
The exercise which we are about to state asks you to prove the last assertion in
Theorem 3.4.11.
Exercise 3.4.12 Show that, with probability 1, the zero set of a Wiener path has Lebesgue
measure 0.
Finally, we mention that the "fractal" properties of Wiener paths have been analyzed
in great detail; see, e.g., ([benA], [Fal, §16.1], [leGa], [Mand, Chapter 25], [Tayll-3]).
d-dimensional Wiener measure and Wiener process
For notational simplicity, we have outlined in this chapter the construction of one-
dimensional Wiener measure m and studied, in particular, some of the sample path
properties of the associated one-dimensional Wiener process. When d > 2, the con-
struction of d-dimensional Wiener measure (still denoted by m)—and hence of the asso-
ciated d-dimensional Wiener process, the mathematical model of Brownian motion in
W1—is entirely analogous. Alternatively, the d-dimensional Wiener measure can be
defined as the product of d copies of one-dimensional Wiener measure; of course, m
(or its completion) is now viewed as a probability measure on (C^' , B(C^' )) (or on
(C^'b, SO), where C^ = Co([a, b], Rd). Further, the d-dimensional Wiener process
can be viewed as the product of d independent copies of the one-dimensional Wiener
process.
All the theorems of this chapter (with the exception of Theorems 3.4.10 and 3.4.11)—
as well as of Chapters 4 and 5 below, where we mostly work in dimension d = 1 for
notational convenience—remain valid when d > 2, with the obvious adjustments. In the
rest of this book, the dimension should be clear from the context.

3.5 Appendix: Converse measurability results

In this appendix, we will establish converses to measurability results given in Section
3.3. Fix ti, ...,tn such that a = to < t\ < • • • < tn < b and let Pr, ,n be given by
(3.3.2). We saw in Proposition 3.3.1 that if B e B(E"), then P,"1 tn(B) e B(C^'b). In
under U.S. or applicable copyright law.

fact, the converse holds as well.

Proposition 3.5.1 P^...,tn (B) e B(CQtb) if and only if B e B(E").
Proof We need only show that P^1 ..,,„(#) 6 B(C°'b) implies that B € B(W).
Define //,,,...,,„ : M" -» C^b as the polygonal path in C^b with vertices at (a,0),
(t\, MI), ..., (tn,un) and (b, un); that is,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 58 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where HO = 0. Since t\,..., tn will be fixed throughout, we simplify notation and write
P and H in place of />,, ,n and //»,,...,/„, respectively. The function H is easily seen to
be continuous and so, by Exercise 3.2.10(b) and the present assumption, it follows that
H~l (P-1 (B)) belongs to H(K"). Thus to finish the proof it suffices to show that

Let (MI, . . . , un) belong to the left-hand side of (3.5.2). Then //(MI, . . . , « „ ) 6
P~ 1 (fi)andso

However, (u\,...,un) — (H(u\,..., un)(t\),..., H(u\,..., «„)(*„)) and therefore,

(MI , . . . , un) e B as we wished to show.
Conversely, let (MI , . . . , un) e B. We need to show that H (MI , . . . , un) € P~l(B);
that is,

This is so, however, under the present assumption since (u\,...,un) =

(//(MI, . ..,un)(t\),..., //(MI, .. .,Un)(tn)). Hence the result is established. D
The converse measurability results in Proposition 3.5.1 (as well as Theorem 3.5.2
below) will be used, in particular, in Proposition 4.2.12.
We know by Corollary 3.3.4 that Pt~,l..,ta (E) is Wiener measurable provided that E is
a Lebesgue measurable subset of R". Is the converse valid? This question was posed by
Robert Cameron in his lectures at the University of Minnesota and was solved but never
published by Fulton Koehler, a colleague who was attending Cameron's course. Koehler's
argument was quite involved. The simpler proof below was given by Siegfried Graf after
the first author mentioned Koehler's theorem during some lectures at the University of
Erlangen in 1980. The key to Graf's proof is the fact that m, as a finite measure on the
complete separable metric space C^'b, is regular (see, e.g., [Con, Proposition 8.1.10,
p. 258]).
Theorem 3.5.2 Let E c W. Then P,~^tn(E) is Wiener measurable if and only if E is
Lebesgue measurable.
Proof We need only prove one direction. Accordingly, assume that P,~l <tn (E) is Wiener
under U.S. or applicable copyright law.

measurable. We begin the proof with a string of three equalities. The first of these follows
from our assumption and the regularity of m and the other two will be explained below.
We will simplify notation by writing P rather than /*/, ,...,rn - We have successively:

We EBSCO
will Publishing
establish:the second
eBook Academicand third equalities
Collection (EBSCOhost) in (3.5.3)on by
- printed showing
6/8/2017 inequalities
5:27 PM via in
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
both directions. The relationship < corresponding to the second equality in (3.5.3) is
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: CONVERSE MEASURABILITY RESULTS 59
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

clear since K c P ~ l ( P ( K ) ) ; the inequality > holds since K c P~l(E) implies that
P(/T)C EmdsoP-l(P(K)) c ?-'(£).
The sets P(AT) appearing on the left-hand side of the third equality in (3.5.3) are
compact subsets of E since K is compact and P is continuous; the inequality < follows
immediately. The inequality > follows from the fact that P~l(L) C P~~l(E) and from
the first two equalities in (3.5.3).
As we saw in the proof of Proposition 3.3.3, the image probability measure m o P"1
equals the probability measure v defined on B(R") by

Further, since v is mutually absolutely continuous with respect to Lebesgue measure on

R", the a -algebra associated with the completion of v is precisely Leb.(Rn). We can
now summarize the results of (3.5.3) by writing

In order to complete the proof, it will be helpful to recall some facts about the inner
measure v+ associated with v. By definition [Coh, p. 39], v* is given by

for any subset A of R". Since v is regular, for F e Leb.(R") we have

Putting (3.5.6) and (3.5.7) together, we obtain

Combining (3.5.5) and (3.5.8), we see that for any subset E of R" for which P"1 (E) e
<Si, we can write

Now if E is such that P~l(E) e S}, then P~l(Ec) = P~l(E)c e Si also. (Here, Ec
denotes the complement of E in R".) Hence, by (3.5.8),
under U.S. or applicable copyright law.

From (3.5.9) and (3.5.10) and the fact that m(Co'fo) = 1, we obtain

Further, every subset A of R" satisfies [Coh, Problem 7, p. 42]

c c
It follows from (3.5.11)
EBSCO Publishing and (3.5.12)
: eBook Academic Collection(with A = - E)
(EBSCOhost) thatonv*(E
printed ) =5:27v*(E
6/8/2017 PM via). But this
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS c
implies [Coh, Proposition 1.5.5, p. 39] that E , and hence also E, is v-measurable. Since
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 60 WIENER MEASURE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

the v-measurable sets coincide with Leb.(R") as noted earlier, we see that E e Leb.(M")
as desired. D
We now easily obtain the following two corollaries.
Corollary 3.5.3 There exist subsets ofC^b which are Wiener measurable but not Borel
measurable.
Proof Let a < t < b and take E e Leb.(R)\6(R). By Theorem 3.5.2 and Proposition
3.5.1, Pt~l(E) € Si\B(C%'b). n
Corollary 3.5.4 There exist subsets ofC^'b which are not Wiener measurable.
Proof Let a < t < b and take a subset E of K which is not Lebesgue measurable. Then,
by Theorem 3.5.2, P~l (E) i S\. Q
Remark 3.5.5 As far as we know, the first discussion of Theorem 3.5.2m the literature
appeared in a paper of David Skoug [Sk] where Yeh-Wiener (or two-parameter Wiener)
space was the primary object of interest. A quite general result along the lines of Theorem
3.5.2 has been established by Chang and Ryu [ChanRyl].

3.6 Appendix: B(X x Y) = B(X) ® B(Y)

Let X and Y be topological spaces. Throughout this discussion we assume that X x Y
has the product topology.
Proposition 3.6.1 Let X and Y be topological spaces. Then

Proof Let n\ and n2 be the projection maps from X x Y onto X and Y respectively.
Clearly, it\ and n2 are continuous. (Indeed, the product topology on X x Y can be
described as the weakest topology on X x Y making n\ and n2 continuous.) Hence n\
and 7T2 are Borel measurable.
SinceB(X)®B(Y) := a([BixB2 : BI e S(X)andB2 e B(y)}),(3.6.1)willfollow
if we show that BI x B2 e B(X x Y) for every B\ e B(X) and B2 e B(Y). Accordingly,
let BI e B(X) and let B2 e B(Y). Then BI x Y = n^(Bi) and X x B2 = n~l(B2) are
both in B(XxY). Therefore BI x B2 = (Bi xY)C\(XxB2) =srf 1 (Bi)nn- 2 ~ 1 (fl 2 ) e
under U.S. or applicable copyright law.

B(X x Y), as desired. D

Proposition 3.6.2 If X and Y are second countable topological spaces, then

Proof By Proposition 3.6.1, it suffices to show that

Since
EBSCOXPublishing
and Y :are second
eBook Academiccountable, there are- printed
Collection (EBSCOhost) countable bases 5:27
on 6/8/2017 [U\,PM U
via2,...} and
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
{Vj, V2,...} for X and Y respectively. It is well known then (and easy to prove) that
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: B(X x Y) = B(X) ® B(K) 61
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

{£/y x Vk : j = 1, 2 , . . . , k = 1, 2 , . . . } is a basis for the product topology on X x F.

Now we write

where only the first containment in (3.6.4) needs commenting on. Since {Uj x Vk} is a
base for the product topology on X x Y, every open subset of X x Y can be written as
a countable union of sets of the form Uj x Vk.lt follows that

and so the proof is complete.

Corollary 3.6.3 IfX and Y are separable metric spaces, then

Proof It is well known that every separable metric space is second countable and so the
claim made here follows immediately from Proposition 3.6.2. D
Remark 3.6.4 The equality B(X) <8> B(F) = B(X x F) is in fact true for Souslin spaces,
a much larger class of topological spaces (see, e.g., [Coh] or [Schwl]). However, it is
not true for arbitrary topological spaces X and Y.
Some further information related to the topic of this appendix can be found in [Bau,
Problem 2, p. 384].
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

4
SCALING IN WIENER SPACE AND THE ANALYTIC
FEYNMAN INTEGRAL

Notre imagination se lassera plutot de concevoir que la nature de fournir.

[Our imagination will rather grow weary of conceiving than nature of providing.]
Paul Levy, citing Leibniz, and discussing the complexity of Brownian paths. (French text
quoted in [benA, p. 8].)

Mathematicians, however, have well understood the lack of rigor of such considerations, said to
be geometric, and how, for example, it is childish to want to prove, by simply drawing a curve, that
every continuous function admits a derivative. If the differentiable functions are the simplest, the
easiest to deal with, they are nevertheless the exception; or if one prefers a geometric language, the
curves which do not have a tangent are the rule, and the nicely regular curves, such as the circle,
are quite interesting, but very particular cases.
Jean Perrin, 1913 [Per, p. 25]

There are very interesting mathematical problems involved in the attempt to avoid the subdivision
and limiting processes. Some sort of complex measure (italics added) is being associated with the
space of functions x(t). Finite results can be obtained under unexpected circumstances because the
measure is not positive everywhere, but the contributions from most of the paths largely cancel out.
Richard P. Feynman, 1948 [Fey2, p. 372]

Let CT > 0. The simple transformation of scale change, x i-»- ax, in CQ'* has some
surprisingly pathological features with respect to Wiener measure m. When a is taken
as fixed, as it is in most situations, these features do not cause any particular concerns.
However, one of the approaches to the Feynman integral which will most interest us
involves analytic continuation in the scaling parameter; in that setting, scalings by all
positive numbers cr will be involved and we will need to exercise caution.
under U.S. or applicable copyright law.

The basis of the results in Sections 4.2-4.5 is Levy's famous quadratic variation
theorem, established in Section 4.1. This theorem is well known to probabilists but the
content of Sections 4.2^.5—which discuss consequences of this result for scaling—is
less well known. In Section 4.5, we define the scalar-valued analytic Feynman integral
and explain the relevance of the earlier material for this topic. In particular, we will see
in Example 4.5.3 that the usual equivalence relation for functions on Wiener space (i.e.,
equality m-a.e.) is not adequate for our purposes. A more refined equivalence, introduced
in Definition 4.5.5, is shown to have the appropriate properties in Theorem 4.5.7.
We discuss
EBSCO in :Section
Publishing 4.6 theCollection
eBook Academic nonexistence of Feynman's
(EBSCOhost) - printed on"measure "; specifically,
6/8/2017 5:27 PM via we
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
show in Theorem 4.6.1 that the natural analogue of Wiener measure with complex vari-
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
anceOperational
parameter a (a = X ~ 2) is not a countably additive (complex) measure. Section 4.6
Calculus
Account: ns000601
 QUADRATIC VARIATION OF WIENER PATHS 63
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

fits well with the rest of this chapter but is also directly relevant to Chapter 7, where the
Feynman integral will be introduced heuristically.
For notational simplicity, we will assume that d = 1 throughout this chapter although
all the results presented here are valid in any dimension d > 1.

4.1 Quadratic variation of Wiener paths

The key to the results of this section is Levy's 1940 theorem on the quadratic variation
of Wiener paths ([Levl], [Lev2, Chapter VI)). We w^ll prove the special case which
was discovered later (1947) but independently by Cameron and Martin [CaMa3] in
connection with their investigation of change of variable formulas in Wiener space. Our
proof, an "analyst's proof", will essentially follow [CaMa3]. For a "probabilist's proof"
of the general result, see, for example, [Tuc, p. 243].
The quadratic variation result is of independent interest and we will discuss some
aspects of it which have no particular relevance to the Feynman integral.
Given a partition n of [a,b],a = to<t\ < • • • < & = b, and x € CQ'*, let

In the case where FI divides [a, b] into k subintervals of equal length, we will simply
write Sk(x) instead of Sn(x). It will be shown in Theorem 4.1.2 below that

almost surely.
Remark 4.1.1 To get some perspective on formula (4.1.2), note that if x € CQ'
is nice enough to satisfy, say, a Lipschitz condition with Lipschitz constant K, then
limn-».oo Sy>(x) = 0 since
under U.S. or applicable copyright law.

as n —> oo.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
Theorem 4.1.2 Wiener paths x e C^b satisfy formula (4.1.2) almost surely.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 64 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof We restrict attention to the case [a, b] = [0,1]. This will simplify the notation
without making any essential difference in the proof. We will begin by showing that

from which it will immediately follow that

Since [a, b] = [0, 1], given any positive integer k, we have tj = {, j = 0, 1 , . . . , k

and

Let

We know from Proposition 3.3.17 that x(tj)-x(tj-\) ~ N(0, tj-tj-i), j = 1 , . . . , k.

It follows from this and (4.1.5) that

To evaluate Ik it remains to calculate /c«,* S^(x)dm(x). Using Theorem 3.3.5 and

under U.S. or applicable copyright law.

the fact that tj — tj-\ = l/k, j = 1,..., k, we can write

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 QUADRATIC VARIATION OF WIENER PATHS 65
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

J
where MQ '•— 0- The change of variables vj — /TJT[ , j = 1, • • •, k, has Jacobian
(l/k)k/2, and so we obtain

where the next to last equality comes from formulas (4.7.1)-(4.7.3) of Appendix 4.7 to
this chapter. Combining (4.1.6), (4.1.7) and (4.1.9), we now see that
under U.S. or applicable copyright law.

as claimed in (4.1.3).
It immediately follows from (4.1.3) that Sk — l —>• 0 in the L2-norm on Wiener space,
but this does not yet give us the conclusion we seek since L2-convergence does not in
general imply almost sure convergence. However, "fast enough" L2-convergence does
imply convergence almost surely; the rest of the proof will consist of showing this in our
present setting. (A general result of this type will be stated in Exercise 4.1.3 below.)
Formula (4.1.3) yields (4.1.4) immediately as observed earlier. Now let

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 66 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We must have

since, if not,

which contradicts (4.1.4).

Let

Then, by (4.1.12),

where K is twice the sum of the convergent geometric series £]tlo ,3 Lt •

Now for x e C2'*\Fn = n£ln ££, where Eck denotes the complement of Ek in CQ'*,
it follows from (4.1.11) that | S2* Cx) - 11 < 2~/:/3 for it = n, n + 1,.... Hence, if there
is an n such that x £ Fn, then limjt_>oc ^W = !• Thus limt_>.oo ^(jc) = 1 except
possibly on n£L, F&. Thus it suffices to show that tnCn^lj F*) = 0. But, for every n, we
have, by (4.1.13),

The result follows from (4.1.14) since (£/2 n / 3 ) ->• 0 as n -> oo.
under U.S. or applicable copyright law.

Exercise 4.1.3 Let (n, A, P) be a probability space and let X, Xn, n = 1, 2, . . . , be

(R-valued) random variables on £2 such that

Show that X n ((w) ->• X (w) P-a.s. Can you get the same conclusion if P is just required
to be a finite measure? What if the number 2 in (4.1.15) is replaced by some number
r >0?
Remark 4.1.4 (a): eBook
EBSCO Publishing If weAcademic
had given Exercise
Collection 4.1.3
(EBSCOhost) - before the6/8/2017
printed on proof 5:27
of Theorem
PM via 4.1.2,
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
we would have been done with the proof of Theorem 4.1.2 as soon as (4.1.4) was
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 SCALE CHANGE IN WIENER SPACE 67
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

established. Note, however, that Exercise 4.1.3 does not allow us to conclude from (4.1.3)
that Sk -> 1 a.s.
(b) In the notation of probability theory, (4.1.15) is written more compactly as

(c) The results in Exercise 4.1.3, in our opinion, deserve to be better known than they
seem to be.
Exercise 4.1.5 Use Theorem 4.1.2 to show that the set of functions in C^'b which are of
bounded variation on [a, b] is a set of m-measure 0. (Corollary 3.4.8 already gave us a
stronger conclusion than is provided by this exercise, but it is somewhat informative to
see that Theorem 4.1.2 is related to these facts.)
Next we state without proof the stronger quadratic variation result which we men-
tioned earlier and which is due to Levy. (See [Levl], [Lev2, Chapter VI].)
Theorem 4.1.6 Let {FI^} be any nested sequence of partitions of [a, b] whose norm
approaches 0. Then Wiener paths x almost surely satisfy

where Snk is given by (4.1.1).

The nature of the sums in (4.1.1) makes the term "quadratic variation" seem reason-
able. However, a more straightforward extension of the usual definition of total variation
to the quadratic case would be

where x is a function from [a, b] to R and where the supremum is taken over all partitions
n of [a, b], a = to < t\ < • • • < tk — b. In light of Theorem 4.1.6, it may seem
surprising that Wiener paths x almost surely satisfy Quad.Var.(jc) = +00. This follows
from the work in [Fre, p. 48].
under U.S. or applicable copyright law.

4.2 Scale change in Wiener space

Much of the next four sections of this chapter is adapted from the 1979 paper of Johnson
and Skoug [JoSk7] which discusses and expands on results of Cameron and Martin
[CaMa3] and Cameron [Ca2], as well as relates them to various results in the more
recent literature. We turn now to our main concern in this chapter, the implications of
Theorem 4.1.2 for scale change transformations in Wiener space.
Given a > 0, let

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 68 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where

We will use m \ rather than m, at least in the remainder of this chapter, to denote (standard)
Wiener measure on C^'b. Let ma (a > 0) be the image measure

(By this, we mean that ma is the image measure of m1 by the continuous map <pa :
CQ'& -> CQ' given by (pa(x) = ax.) The measure ma is certainly defined on B(C^' );
in fact, ma(B) — mi(ff~ 1 fl), for B e B(C^'b). (Before long, we will consider its
completion as well.) In probabilistic terms, the space (Co' 6 ,nv) is a realization of
the Wiener process with variance parameter a2; that is, axioms (i)-(iv) of (3.3.34) are
unchanged except that (iii) becomes

The properties (i), (ii), (iii)CT and (iv) are readily verified for the space (CQ'*, ma); in
fact, (i) and (iv) are immediate simply because C^'b is our probability space. We will
show in Appendix 4.8 to this chapter that (iii)CT holds.
We go next to some elementary but rather informative propositions. The first of these
tells us that the sets £2ff are disjoint from one another but simply related and that ma is
concentrated on £2CT.
Proposition 4.2.1 (i) The set Q,j is Borel measurable for every a > 0.
(ii) Given any a\, ai> 0, we have

and, in particular, for every a > 0,

(iii) ma(Qa) = 1 for every a > 0.

(iv) If a\ ^ 02 (a\, 02 > 0), then J2ffl D Qa2 — 0 so that the measures mai and nv2
are mutually orthogonal.
under U.S. or applicable copyright law.

Proof (i) $2" is continuous and hence Borel measurable for every n. It follows from
(4.2.1) that f2CT is Borel measurable for every a > 0.
(ii) This is an immediate consequence of (4.2.1) and (4.2.2).
(iii) By Theorem 4.1.2, m1(£2i) = 1. Using this fact as well as (4.2.5) and (4.2.3),
we can write

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
(iv) This follows
UNIVERSIDAD DISTRITAL immediately from
FRANCISCO JOSE DE (4.2.1) and (iii).
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 SCALE CHANGE IN WIENER SPACE 69
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 4.2.2 In light of (iii), we see that £la is a set of full ma-measure. We will say
that nv is concentrated on £la. However, we will not refer to &a as the "support" of
trie, since there is a rather standard definition of support [Schwl, p. 40] to which the
present situation does not correspond. (In fact, Supp(ma) = C^'b for every a > 0.)
The following definitions will be important to us as we continue this chapter and also
later in connection with the analytic Feynman integral.
Definition 4.2.3 A subset A of CQ ' is said to be scale-invariant measurable provided
a A € Si for all a > 0. A scale-invariant measurable set N is said to be scale-invariant
null provided that m\ (aN) = Ofor all a > 0. A property which holds for all x e CQ'&
except for a scale-invariant null set is said to hold scale-invariant almost everywhere
(briefly, s-a.e.). The classes of scale-invariant measurable and scale-invariant null sets
will be denoted S and J\T, respectively.
Remark 4.2.4 The definition just given may well seem most peculiar to many readers.
If A e S\, one would expect it to follow that, for example, 2A € 5i. We will see below,
however, that this does notfollow; multiplication by 2 is a nonmeasurable transformation
on Wiener space (Cg' , Si, mj)-
Definition 4.2.5 We let Sa be the a -algebra obtained by completing the measure space
(Cp1*, B(CO'*), nv) and we letMa denote the class of ma-null sets.
We show in our next proposition that Sa = a S1 and that nv(,E) = m1(a~ l E) for
every E e Sa.
Proposition 4.2.6 (i) N e No if and only if a~lN € N1; equivalently, Na = oN1.
(ii) E & Sa if and only if a~lE 6 S1; equivalently, Sa = a S1.
(iii) We have ma(E) = mi (a~l E) for every E e Sa.
Proof (i) We first show that N € Na implies that a ~l N € A/i. Let N € Ma. Then there
exists a Borel set M such that N c M and mCT (M) = 0. By definition of nv. mo (M) =
m\(a~lM). Thus a~lM is a Borel set which is mi-null. But a~lN c &~1M and so
a ~l N e .A/i as claimed. A similar argument, which we leave to the reader, shows that
if o--lN eM.thenW eMa.
Next we show that Ma C a MI. Let N e Ma. By the above, a~lN e A/i- Thus
N = a(a~lN) e aj\f\. Similarly, one can easily show that aN1 c A/"CT.
under U.S. or applicable copyright law.

(ii) We will carry out the proof that a~lE e S1 implies that E e Sa and leave
the rest to the reader. Accordingly, let a ~l E e S1. Then there exists a Borel set B and
N e M such that a~lE = BUN.By (i), crN €<r.A/i = Ma. Of course, oB is Borel,
and so we can write

Thus E e Sa as desired.
(iii) Let £ e Sa. Then E = B U N, where B is Borel and W e Na. Then
l l
mo(E) = mo(B
EBSCO Publishing U N)
: eBook = Collection
Academic ma(B) (EBSCOhost)
= m1(a~ B) =
- printed m1(a~5:27
on 6/8/2017 B PMUviaa ~ 1 N ) =
l
UNIVERSIDAD DISTRITAL FRANCISCOlE).Thusm
JOSE DE CALDAS(E) = m i ( a ~ l E ) as claimed.
mi(cr~ (BL)N)) = mi(o— a O
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 70 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Next we show that the scale-invariant measurable sets form a a -algebra and that a
set is scale-invariant measurable if and only if it is ma -measurable for every a > 0.
Proposition 4.2.7 (i) S is a a-algebra; in fact, S = C\a>o^"-
di)M=r}a>0M<,.
Proof (i) Let A e S. By Definition 4.2.3, a~l A e S1 for every a > 0. Thus, by (ii) of
Proposition 4.2.6, A = a(cr~l A) e crS\ = Sa for everyCT> 0. Hence 5 c n<r>o Sa-
Conversely, suppose that A e f\>o ^ • Then A e Sa = crSi for every a > 0. Thus
a"1 A € .Si for every a > 0. But then A e 5. Hence n^o^V £ 5. It now follows
that 5 = Ho-s-o'Sff an<^ *ha* <S is a cr-algebra since the intersection of a -algebras is a
CT-algebra.
We leave the proof of (ii) for the next exercise. O
Exercise 4.2.8 Prove (ii) of Proposition 4.2.7.
Now we show that a set is scale-invariant measurable if and only if its intersection
with every &(, is an ma-measurable set.
Proposition 4.2.9 (i) E e S if and only if E D fiff € Sa for every a > 0.
(ii) N € M if and only if Nrt£la e Afa for every a > 0.
Proof (i) Suppose that E e S and letCT> 0 be given. By Proposition 4.2.7, E € Sa
and, by Proposition 4.2.1, £2a is a Borel set. Therefore, E n £2CT € «Scr.
Conversely, suppose that E r\Sla e Sa for every cr > 0. To show that E e 5 it
suffices, by Proposition 4.2.7, to show that E £ Sa for every a > 0. But, by (iii) of
Proposition 4.2.1, Co'fe\atf is nv-null and so E *= (£nS2 < T )U(£n(Co'*\n ( r )) e £„.
We leave the proof of (ii) to the reader. D
Our next result shows rather well what scale-invariant measurable sets and scale-
invariant null sets are like and how they compare to Wiener measurable sets and Wiener
null sets, respectively.
Theorem 4.2.10 (i) £ e 5 if and only if E has the form

where each Ea is an ma-measurable subset of £2CT and L is an arbitrary subset of

under U.S. or applicable copyright law.

c b
o' \ Ua>o ®a- Further, for E written as in (4.2.6), we have

for all a > 0.

(ii) N e A/* if and only ifN has the form

where
EBSCO each Nf, : is
Publishing anAcademic
eBook m^-null subset
Collection of &„ - and
(EBSCOhost) printedL onis6/8/2017
an arbitrary
5:27 PM via subset of
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
C*\U>o^-
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 SCALE CHANGE IN WIENER SPACE 71
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof (i) Suppose that E is in 5. Let Ea :— E n £2CT for every a > 0 and let £ :=
E n (CQ'*\ U<r>o ^2<f )• By Proposition 4.2.9, Ea is an ma-measurable subset of £2a for
every a > 0. Also it is clear that L is a subset of Cg'fc\ Ucr>o ^- The decomposition
(4.2.6) now follows since, by (iv) of Proposition 4.2.1, the union

is a painvise disjoint union.

Conversely, suppose that E has the form (4.2.6). To show that E e S it suffices by
Proposition 4.2.9(i) to show that E D Sla e Sa for every CT > 0. But, from (4.2.6),
E n fiff = Ea and Ea 6 Sa. Hence E is in <Sff as desired. The formula nta(£') =
m^ (Ea) follows from the fact, established in Proposition 4.2.1, that mCT is concentrated
on Qa.
(ii) Suppose that N e A/". Then N e S and so can be written in the form N =
(U<r>o Na)\JL, where each Na is in 5a and L c Cg' fe \ U(7>o *V We only need to
show that mCT (Na) = 0. But N e N implies, by Proposition 4.2.7, that N e Ma. Hence,
by (4.2.7), 0 = ma(N) = m.a(Na) as desired.
The converse of (ii) follows from the representation (4.2.8) and from the formula
(4.2.7). D
Remark 4.2.11 (a) It is noted in the proof of Theorem 4.2.10 that the union on the
right-hand side of (4.2.9) is a painvise disjoint union. It follows that the union in (4.2.6)
is a painvise disjoint union.
(b) Theorem 4.2.10 shows strikingly that there are many more Wiener measurable sets
(i. e. sets in S\) than there are scale-invariant measurable sets. It is quite clear that a set E
is Wiener measurable if and only if it has the form E\ UL, where E\ is an mi-measurable
subset of £l\ and L is an arbitrary subset o/Uo-so^i ^CT U (^o \Ucr>o^< 7 )- The
reader should compare this with (4.2.6). The set E is Wiener null if and only ifE] in the
decomposition E = E\ U L is an mi-null subset of£l\.
Our next proposition compares the a-algebras B(C^' ), S and Sag for anyCTQ> 0.
Proposition 4.2.12 For every a0 > 0, we have B(£^b) CSC Sao.
under U.S. or applicable copyright law.

Proof The containments are clear since B(C^'b) c Sa for every er > 0 and since, by
Proposition 4.2.7, S = (~}a>0Sa.
Let a < t < b and let G e Leb.(R)\S(R). By Proposition 3.5.1, Pt~l(G) = (x 6
C0'fe : x(t) € G} i B(Cg'b). However, for every a > 0, multiplication by a is a
Leb.(R) - Leb.(R) measurable map from K to E. Hence aG € Leb.(E) for every a > 0.
Since aP~l(G) = Pt~l(aG), it follows from Theorem 3.5.2 that aP,~l(G) e Si for
every a > 0. Thus P,~l(G) 6 S, and so now we know that P,~l(G) 6 <S\B(C0'*).
Now let G be a subset of K which is not Lebesgue measurable and take a\ > 0
EBSCO Publishing l - printed on 6/8/2017 5:27 PM via l
such that a\ ± : cr eBook Academic Collection (EBSCOhost)
0. By Theorem 3.5.2, Pt~ (G) £ Si. Therefore, <?iP~ (G) i
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS l
a\Si. But, by Proposition 4.2.6, ai<Si = S Hence 0iP,~ (G) £
ar Feynman Integral and Feynman's
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The
S^ and so
Operational Calculus
Account: ns000601
 72 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

E := a\P,~l(G) D &ai £ Sai. Thus E £ S. However, E n S2CTo = 0 and so £ 6 £„.„.

Therefore £ e <SCTO \5 as we wished to show. O
Remark 4.2.13 (a) In showing that the containment S C Sag in the preceding propos-
ition is proper, we have actually proved more than was claimed. We have shown that
given any (TO, a\ > 0 with a\ ^CTO>there exists E e Sa{ \SaQ.
(b) In Proposition 4.2.12, we fixed a single point t,a < t < b, and worked with
Pt~ (G) where G c R. We could just as well have taken a < t\ < • • • < tn < b and
G C R" and worked with /)^,...,/n (G) since Proposition 3.5.1 and Theorem 3.5.2 apply
to this situation just as well.
The exercise that we are about to state is easy, but makes a point worth noting.
Exercise4.2.14 Let a < t\ < • • • < tn < b and take G e Leb.(R"). Show that
^1*«?)6S.
Let £ be a scale-invariant measurable subset of CQ' . What can be said about the
dependence of m i (cr £) onCT? Answer: Absolutely nothing. A striking result of Cameron
and Martin [CaMa3] shows that m\(aE) can vary with a in an arbitrarily nasty way.
Further, using Theorem 4.2.10, this result is not hard to prove as we now show.
Corollary 4.2.15 Let f be an arbitrary function (nonmeasurable, for example) from
(0, +00) to [0,1]. Then there exists E s S such that m\(aE) = f (a) for all a > 0.
Proof First we claim that for every a > 0 there exists Ea c £2CT such that £<j e Sa and
ma(Ea) = /(cr^ 1 ). If f ( c r ~ l ) = 0, simply take Ea = 0. Otherwise, begin by taking
va e (—00, +00] such that

Now let Aa := P^l((-oo, va)). By (4.2.10), mi(A C T ) = /(cr"1). Further, if we let

Fa := Aan ft], Fa e B(CQ'b) c 5] tm&m\(Fa) = f (a~l).Finally,take£CT := aFa.
Then £CT e aS\ — Sa and Ea c Qa since Fa c Q,\. Also, we can write

as desired.
under U.S. or applicable copyright law.

It is easy to finish. Let £ := U <T>0 £ (7 - Then £ 6 S by Theorem 4.2.10, and, by

(iii) of Proposition 4.2.6 and (4.2.7) of Theorem 4.2.10, we have

and so the proof is complete. d

Multiplication by any positive number other than 1 fails to preserve Wiener measur-
ability as we now show.
Theorem 4.2.16 :There
EBSCO Publishing exists aCollection
eBook Academic subset H of CQ' -which
(EBSCOhost) printed is
onnot Wiener
6/8/2017 5:27 measurable
PM via but
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
such that a H is Wiener measurable (in fact null) for every a > 0 except a = I. It follows
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 SCALE CHANGE IN WIENER SPACE 73
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

that the sets aH are Wiener measurable (0 < a ^ 1) but a l(aH) = H is not Wiener
measurable.
Proof We need only prove the first assertion since the second follows from it imme-
diately. Let M be any subset of CQ' which is not Wiener measurable. For example,
by Theorem 3.5.2, we can take M = P^ (E), where E is a subset of K which is not
Lebesgue measurable. Let H := M n QI. Since M = H U (M n £2C{) and M n Q£
is mi-null, it follows that H is a subset of fij which is not Wiener measurable. Now
for anyCT(0 < CT / 1), aH c. a£l\ = £i<? and hence is mi-null and so is certainly
mi-measurable as claimed. D
Theorem 4.2.16 centers on mi-measurability or its absence but a similar result holds
where mi is replaced by ma, a > 0.
Corollary 4.2.17 LetCTQ> 0 be given. There exists a subset H ofC$' which is not
mao -measurable but such that a H is mffo -measurable (in fact mao -null) for every a > 0
except a = 1. It follows that the sets aH are m^-measurable (0 < a ^ 1) but
a~^(aH) — H is notmao-measurable.
Proof Take H := M D £lao. Otherwise the proof is the same as that of Theorem 4.2.16.
a
The sets Qa, a > 0, depend on the particular sequence of partitions of [a,b] that
we choose. If n = {111, l\i,...} denotes another nested sequence of partitions whose
norms go to 0, we may let

Essentially because of Theorem 4.1.6, Levy's theorem, all of the results obtained up to
this point go through with changes in notation where appropriate. Note however that
S<T, m CT , S and A/" are all independent of the sequence of partitions.
A set E in S now has two decompositions according to the two versions of Theorem
4.2.10:
under U.S. or applicable copyright law.

where E™ = E n Sl™ and Ln = E n (Cg'*\ |J(T>o ^)• How do these two decomposi-
tions relate to one another? Our next result shows that they agree up to a scale-invariant
null set.
Proposition 4.2.18 Let E be a subset ofC^' which is scale-invariant measurable. Then
the two decompositions of E given by (4.2.13) and corresponding respectively to our
original sequence of partitions and to n have the property that the set

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 74 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

is scale-invariant null. (Here, A A B := (A\B)U(B\A) denotes the symmetric difference

of the subsets A and B ofC^.)
Proof First note that for all a > 0,

Hence, by Theorem 4.2.10, the set \Ja>0(Ea\E^) u (L\Ln) is scale-invariant null.

A symmetric argument shows that the set U CT>0 (£"\£ CT ) U (L n \L) is scale-invariant
null. Since

the proof is complete.

4.3 Translation pathologies

Cameron showed in [Ca2] that the scale change pathologies in Wiener space imply
certain pathologies for translation maps. Further negative and some positive results were
developed in Section 3 of [JoSk?]. We simply state some of these results; proofs can be
found in [JoSk?].
Our first proposition was used in [JoSk7] to obtain others which are of more immedi-
ate interest here. We state it because it is of some interest in its own right and because the
under U.S. or applicable copyright law.

equivalence of (i) and (ii), which is easily established, is used in the proof of Proposition
4.4.3 below.
Proposition 4.3.1 Let f be an ^.-valued junction on C^b and let a and p be positive
numbers. The following assertions are equivalent:
(i) /((<72 + p2) 2z) is an m\ -measurable function ofz.
(ii) f ( z ) is an m i -measurable function ofz.
(<T 2 +p 2 )J

(iii) f(x + y) is an rtv x mp-measurable function ofx and y.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
(iv) f(ax DISTRITAL
UNIVERSIDAD + py) is an m\ JOSE
FRANCISCO x mi-measurable
DE CALDAS function ofx and y.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TRANSLATION PATHOLOGIES 75
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Further, if any one, hence all, o/(i)-(iv) hold, then

where the equalities are in the sense that if either side is defined, then so is the other and
equality holds.
Proposition 4.3.2 Let a and p be positive numbers and let E e S i. Then E + y
(<7 2 +p 2 ) 2
and E — y are in Sa for mp-a.e. y and m CT (E + y) andma(E — y) are mp-measurable
functions ofy. Similarly, E + x and E — x are in Sp for nv -a.e. x and m,p(E + x) and
mp(E — x) are ma-measurable functions ofx. Furthermore,

Cameron showed in [Ca2] that there exists E e S\ such that for m\-a.e. y, E + y <£
S\. The first assertion in the conclusion of Proposition 4.3.2 gives instead a positive
result about the measurability of translates. We now state as a corollary a special case
of this positive result which is especially closely related to Cameron's negative result.
Corollary 4.3.3 Let E be in S^. Then E + y & S\ for m\-a.e. y.
under U.S. or applicable copyright law.

The next corollary is just Cameron's negative result when a = p — 1.

Corollary 4.3.4 Let a and p be positive numbers. The translation map Ty from
(CQ'h,Sp,mp} to (c£' ,Sa,ma} fails to preserve measurability for mp -a.e. y. (Of
course, Ty : CQ' —> CQ' is defined by Tyx := x + y.)
Corollary 4.3.5 For m\ x mi -a.e. (x,y), x + y € £2^. More generally,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 76 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In contrast, if r > 0 but T ^ (<r2 + p 2 )2,

Remark 4.3.6 Given two finite measures n and v on (C^'b ,B(C^'b)), ^ * v, the
convolution of ^ and v, is the measure on (C^' , B(C^' )) given by

Part of the content of '(4.3.2) from Proposition 4.3.2 is that

The last two corollaries that we will state in the present discussion involve translating
scale-invariant measurable and scale-invariant null sets, respectively.
Corollary 4.3.7 Let E be scale-invariant measurable. Then for every a > 0, E + y e
Sa with the exception of at most a scale-invariant null set ofy-values.
Corollary 4.3.8 Let N be scale-invariant null. Then for each a > 0, N + y e Na with
the exception of at most a scale-invariant null set ofy-values.
The following questions were asked in [JoSk7, pp. 164-165] but have never been
completely answered as far as we know: (i) Let E e S. Is it true that E + y e S for
5-a.e. y? (ii) Let NeN.lsN + yeAf true for s-a.e. y? Ryu [Ry] has recently given a
partial answer to these questions.
Remark 4.3.9 (a) Our discussion of the scaling and translation transformations on
Wiener space has included some positive results but the emphasis has been on the nega-
tive. This gives a somewhat distorted perspective. For example, in the 1940s, Cameron
and Martin computed the Radon-Nikodym derivatives associated with some transform-
ations where the image measures were absolutely continuous with respect to m\. The
resulting change of variable formulas, notably the Cameron-Martin translation theorem
[CaMal] and their linear as well as nonlinear change of variable results [CaMa2,5],
under U.S. or applicable copyright law.

have been extended in a variety of ways and have proved very useful. It turns out that
there is a small (m1-measure 0) but important set of translators for which the translation
map from (C^' , S\, m1) to itself is measurable.
(b) We restricted attention to a > 0 in our discussion ofscalings, but similar results
hold for negative a except for a = — 1. In that case, the transformation is measurable
and one can show that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
In fact, formula
UNIVERSIDAD (4.3.7)
DISTRITAL is involved
FRANCISCO JOSE DEin the proof of Proposition 4.3.2.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 SCALE-INVARIANT MEASURABLE FUNCTIONS 77
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

4.4 Scale-invariant measurable functions

Our next topic is scale-invariant measurable functions. We discuss only real-valued
functions but all the results hold for complex-valued functions as well. As we continue,
we will primarily be interested in functions which are s-a.e. defined, but we do not require
this immediately.
Definition 4.4.1 Given a > 0, let FQ (resp., fs>^Sa) denote the class of all
^.-valued functions which are defined on a Borel measurable (resp., S-measurable,
Sa-measurable) subset of CQ' and which are measurable with respect to the
a-algebra B(C^' ) (resp., S,Sa). Functions in Fg are said to be scale-invariant
measurable.
The next three propositions follow from Propositions 4.2.7, 4.3.1, and 4.2.12,
respectively.
Proposition 4.4.2
Proposition 4.4.3 Let F be an R-valued function defined on a subset DofC^b. Given
a > 0, let Fa be defined on a~^D by Fa(x) := F(ax). Then (i) Fa is in fsl if and
only if F is in Fsa- Also (ii) F is in Fs if and only if Fa is in FS1 for every a > 0.
Proposition 4.4.4 For every
The theorem to follow provides a useful necessary and sufficient condition for a
function F to be s-a.e. defined and scale-invariant measurable.
Theorem 4.4.5 Let F be an R-valued function with domain D C CQ' . Then F is s-a.e.
defined and in FS if and only if, for each a > 0, the function F(o) defined by restricting
F to D n Q.a is ma-a.e. defined and in fsa.
Proof Suppose that F is s-a.e. defined and in Fg. Then by Theorem 4.2.10, F is defined
except on a scale-invariant null set
under U.S. or applicable copyright law.

where Na C &a with ma(Na) = 0 and L C Cg'6\ U <T>0 &a. Hence F(a) is defined
on £la except on Na. Thus /r(cr) is ma-a.e. defined. Now F e FS by assumption and
so, given a Borel subset B of R, F~\B) e S = r\a>oS"- Thus (F(°Tl(B) =
F~l(B) n Sla & Sa for every a > 0. Hence F (a) € fga for every a > 0 as desired.
Conversely, suppose that for each a > 0, F^' is defined except on an ma-null
subset Na of £2(j and F^ e J:$a. Hence F must be defined except on some subset of
the scale-invariant null set

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 78 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Thus Co' & \D is scale-invariant null and F is s-a.e. defined. Now let B € B(R). To show
that F~l(B) e S, it suffices, by Proposition 4.2.9, to show that F~l(B) n £2CT 6 <SCT
for every CT > 0. But this is so since F~l(B) n S2a = ( F ( a ) ) ~ l ( B ) which is in Sa
as desired. n
We will see in various places as we continue that there are many useful functions
which are .s-a.e. defined and scale-invariant measurable but not necessarily Borel meas-
urable. The following easily proved proposition identifies such a class of functions; also,
many further such functions are generated by this class.
Proposition 4.4.6 Let a < t\ < • • • < tn < b and let f be an R-valued Lebesgue
measurable function which is Leb.-a.e. defined on Rn. Define F : Cg' -> R by

Then F is s-a.e. defined and scale-invariant measurable.

We finish this section with an example indicating why care may be necessary when
dealing with scale changes in Wiener space. Further such examples will be given in the
next section.
If F is a bounded Borel measurable function on R and if [an}, a are positive numbers
such that on —> a as n —»• oo, then it is not difficult to show that f£ F(onx)dx —*•
f£ F(ax)dx. It is tempting to think that a similar result would be true on Wiener space;
this is not so as the following example shows.
Example 4.4.7 Let a > 0 be given and suppose that \an} is a sequence of positive
numbers such that an ^ a for all n and an —>• a. Let

The function F is Borel measurable since, by Proposition 4.2.1, QCT is Borel measur-
able. Further, F is clearly bounded. Now x is in £2i for mi-a.e. x and so anx is in
on$l\ — &<,„ c CQ'fc\J2CT for mi-a.e. x. Thus fca.b F(anx)dm\(x) = 1 for every n.
On the other hand, frLa,b F(ox)dm\(x) — 0. Certainly then, (cra,b F(anx)dm\(x) -/*
o o
La.b F(ax)dm\(x).
"-o
Remark 4.4.8 (a) Certain difficulties associated with scaling in Wiener space can be
under U.S. or applicable copyright law.

avoided by restricting attention to Borel sets and Borel measurable functions. As an

illustration, the scaling maps, while not Wiener measurable by Theorem 4.2.16, certainly
are Borel measurable. It is important to realize, however, that not all the difficulties
disappear when the sets and functions involved are Borel measurable. Example 4.4.7
above already shows this and we will see further examples as we continue,
(b) Under the assumption that F is continuous s-a.e., one can show that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
as UNIVERSIDAD
n —> oo, DISTRITAL
where we are assuming
FRANCISCO as above that an —*• a.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE SCALAR-VALUED ANALYTIC FEYNMAN INTEGRAL 79
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

4.5 The scalar-valued analytic Feynman integral

The operator-valued analytic Feynman integral will be among our main concerns and
scaling in Wiener space will play a role in that context. (See especially Definition 15.2.1
and Chapters 15-18 below; see also Section 13.5 for a somewhat different approach.)
However, at present, we will just introduce the scalar-valued analytic Feynman integral
and give a few results which indicate why scaling considerations are essential. Motivation
for the Feynman integral will be provided later on in Chapter 7; for now we will settle
for the formal definition of this one version of "the" Feynman integral.
Definition 4.5.1 Let F be a function on CQ' such that for every A. > 0,
f a,b F(X~l/2x)dm\(x) exists. If there exists a function o/A which is analytic in the
c
right-half complex plane C+ := {A. 6 C : Re A > 0} and which agrees with
La.t, F(X-V2x)dmi(x) for A > 0, this function is denoted fa"^ F(x)dmi(x) and
0 CQ

is called the (scalar-valued) analytic Wiener integral of F with parameter A.. If, in addi-
tion, for q e R (q ^ 0), the limit,

exists, this limit is called the (scalar-valued) analytic Feynman integral of F with param-
q
eter q and is denoted $™/ c
F(x)dm\ (x).
o'
Remark 4.5.2 We restrict our attention here to comments related to scaling and post-
pone discussion of other aspects of Definition 4.5.1. As has often been done in the liter-
ature, we have phrased the definition above so as to minimize the reference to scaling.
Note however that when we require /£«,» F(X.~l/2x)dm\ (x) to exist for every A. > 0, we
are implicitly insisting that F be s-a.e. defined and scale-invariant measurable. Further,
as we will soon see, the equivalence classes of functions appropriate to Definition 4.5.1
are not those associated with equality m\-a.e. as one might expect.
Example 4.5.3 Let G : C^b -»• M be identically 0. Then, for every A > 0,
fra.b G(*.-l/2x)dm}(x) = 0,and so f^^G(x)dm[(x) = OandfmJ9G(x)dmi(x) =
o C0' C0'
0 for all A. e C+ and for all q e E (R = 0). On the other hand, let F := Xo where
under U.S. or applicable copyright law.

"<TO
ao 7^ 1. (Recall that XA denotes the characteristic function of A.) Since F = 0 on
£2i, F = G mi-a.e. (In fact, F = G mff-a.e. for every a > 0 except a = ao). Now

Certainly then the function on the left-hand side of (4.5.1) does not have an analytic
continuation to C+ and so the analytic Wiener and analytic Feynman integrals of F fail
to exist. The point: we
EBSCO Publishing wish
eBook to emphasize
Academic here is that
Collection (EBSCOhost) F andonG6/8/2017
- printed are equivalent
5:27 PM viain the usual
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
sense of functions on Wiener space (i.e. equal m\-a.e.) and that the analytic Feynman
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 80 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

integral of G exists but the analytic Feynman integral of F does not. Note that this occurs
in spite of the fact that F and G are both Bore I measurable.
In Example 4.5.3, one might object that the left-hand side of (4.5.1) can, with the
exception of a single removable singularity at A. = a0~~2, be analytically continued to
the zero function on C. However, if we take {oj : j = 1, 2 , . . . } to be a dense subset
of (0, +oc) such that 1 $ {<TJ : j = 1, 2,...} and let F := XE with E '•= IJ/li ^'
then F is still Borel measurable and F = G mi-a.e. However, now the left-hand side
of (4.5.1) is discontinuous at all points of (0, +00) and so does not have an analytic
continuation to C+ in any reasonable sense.
Exercise 4.5.4 Find a scale-invariant measurable function F such that F = G mi -a.e.
(G = 0 as above), f™^
c
F(x)dmi(x) and fac"/q F(x)dmi(x) exist for all X e C+
o' o'
and q 6 R (q = 0) but

The basic problem in the examples we have just been discussing is that the usual
equivalence relation for functions on Wiener space (equality m1 -a.e.) is not sufficiently
refined. We now introduce the appropriate equivalence relation.
Definition 4.5.5 Let F and G be two functions on Wiener space. We say that F is
equivalent to G and write F ~ G provided that F = G s-a.e. on CQ' .
The following proposition is easily proved.
Proposition 4.5.6 (i) The relation ~ is an equivalence relation.
(ii) F ~ G if and only if, for every cr > 0, F = G ma-a.e. Thus, for every a > 0,
~ is a refinement of the equivalence relation associated with ma.
(iii) F ~ G if and only if for every a > 0, F(ax) = G(ax) m\-a.e.
The next result is very simple, but the ideas are sufficiently important to us that we
designate it as a theorem.
Theorem 4.5.7 Let F and G be functions on Wiener space such that F = G s-a.e. and
suppose that the analytic Wiener integral of G with parameter X exists for all X € C+.
under U.S. or applicable copyright law.

Then the analytic Wiener integral of F with parameter X exists for all X e C+ and we
have

If, further, the analytic Feynman integral of G with parameter q exists (q e R, q = 0),
then the analytic Feynman integral of F with parameter q exists and we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE SCALAR-VALUED ANALYTIC FEYNMAN INTEGRAL 81
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof The assumptions imply that fca,h G(A. l^x)dm.\ (x) exists for every A. > 0.
Further, since F = G .s-a.e., fca,b F(k~l/2x)dm\(x) exists and

for every A. > 0. The result follows immediately.

When it exists, fa"^ F(x)dm\ (x) is an analytic function on C+, and the Feynman
"'
integrals f ab9 F(x)dm\ (x), when they exist, are the boundary values of this analytic
C
P
function. It is natural to ask if arbitrary analytic functions and their boundary values can
arise as analytic Wiener and Feynman integrals. They can, as our next result shows. In
this result, the function F on Wiener space may well be complex-valued. Our earlier
results have been stated for R-valued functions but are valid for C-valued functions as
well, as is easily checked.
Proposition 4.5.8 Given an arbitrary analytic function g on C+, there exists a scale-
invariant measurable function F on CQ' such that for all X in C+,

Proof Let F(x) = g(A.) for x e %-i/2, A. > 0 and let F(x) = 0 for x e
C'o' fc \(U<r>o ^")- The function F is scale-invariant measurable by Theorem 4.4.5. Fur-
ther, since AT1/2 £2i = £2 A -i/2, we have for every A. > 0,
under U.S. or applicable copyright law.

from which (4.5.5) follows.

An argument such as in the proof of Proposition 4.5.8 can be used to establish the
final proposition in this section.
Proposition 4.5.9 Given an arbitrary function f on (0, +00) (nonmeasurable, for
example), there exists a scale-invariant measurable function F such that for all A. > 0,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 82 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

4.6 The nonexistence of Feynman's "measure"

Next we turn to a discussion which is helpful in appreciating why analytic continuation is
used in Definition 4.5.1. Our comments will lead us into some consideration of the nature
of the Feynman integral. In particular, we shall encounter for the first time one of the
basic complexities of the subject; namely, the nonexistence (at least in a simple-minded
way) of "the" mythical Feynman "measure". The reader may find it helpful to review
this material later in connection with Chapter 7 where a rather detailed discussion of the
heuristic ideas behind the Feynman integral will be given.
Recall that in the definition of the scalar-valued analytic Wiener and Feynman inte-
grals of a function F on CQ' , we began by considering the Wiener integral

for A, > 0. Using Theorem 3.3.2, the abstract change of variable theorem, (4.6.1) can
just as well be rewritten as the integral of F with respect to the scaled measure mA-1/2;
namely, it is equal to

Why do we analytically continue (4.6.1) or, equivalently, (4.6.2) to A. e C+? Why not
just use the integral in (4.6.2) where m\-1/2 could be (one might reasonably hope) a
C-valued "Wiener measure" with "variance" parameter 1/A. (in C+\R+) built up from
finite-dimensional "distributions", much as in the case of the normalized Wiener measure
mi = m in Section 3.2. The idea is tempting; indeed, in one of the early papers on
the Feynman integral, Gelfand and Yaglom asserted incorrectly that this could be done
([GelYag, p. 58], reprinted in [Gel, p. 544]). In fact, if one fixes 1 — ( t 1 , . . . , t n ) , with a <
t\ < • • • < tn < b, and considers the set function u.^ defined on the a -algebra
B; := (Jr(B) : B e B ( R n ) } by

where (much as in (3.2.5))

under U.S. or applicable copyright law.

with MO := 0, then fj.jA is a countably additive, C-valued measure. We remind the reader
of the notation in Section 3.3 (see (3.3.1)) according to which

EBSCO Publishing
In order : eBook Academic
to continue Collection it
our discussion, (EBSCOhost)
is helpful- printed on 6/8/2017
to calculate 5:27 PM
the total via
variation [Coh,
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
p. 126] || ; xJohnson,
AN: 98476 II of the measure
Gerald ^-^ Michel
W., Lapidus, actingL..;
onThe
Bf.Feynman
(TheIntegral
interested reader can find the
and Feynman's
Operational Calculus
Account: ns000601
 THE NONEXISTENCE OF FEYNMAN'S "MEASURE" 83
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

definition of the total variation of a (complex) measure in Section 15.2 (item F), just
above Equation (15.2.3); see also [Ru2, Chap. 6].) Now J-j(B1),..., 7-t, (Bk) is a partition
of C0ab if and only if B1,..., Bk is a partition of Rn. With this in mind, it is not hard to
see that||u-t,n||= ||v-t,n|| where v-t,y is the measure on J 3 ( R n ) given by

But the total variation of a measure defined in terms of an L1-function as in (4.6.6) is

just the Z1-norm of the function. Thus we have

where the last equality follows from the well known formula (4.7.1).
Now letting n and 7 vary, it follows from (4.6.7) that if the measures {v-t,n}
were to serve as the finite-dimensional "distributions" of a countably additive,
C-valued measure on the a-algebra generated by all the J-t (B)s (i.e. onB(C0a,b)), then this
measure would necessarily have infinite total variation. (Note that since A e C+ \ R+,
(|A.|/Re A)n/2 -> oo as n -> oo.) But, on the contrary, it is well known ([Ru2, Theo-
rem 6.4, p. 118] or [Coh, Proposition 4.1.6, p. 126]) that a countably additive, C-valued
measure must have finite total variation.
We summarize what has been proven above in the following theorem, which is due
to Cameron and which appeared in his first paper [Cal, p. 126] on the Feynman integral.
under U.S. or applicable copyright law.

Theorem 4.6.1 Let A e C+\R+. There is no countably additive, C-valued measure on

B(C0ab ) whose value at each cylinderset J->t(B) is given by the right-hand side of (4.6.3).
Remark 4.6.2 (a) Various further questions can be asked and results given related to
the preceding theorem. Theorem 4.6.1 is the only result which we will state formally, but
we will make some further comments.
(b) The collection [J-t(B)} of all cylinder sets as n, t and B vary (B € B(W)) forms
an algebra which is a subalgebra of the a-algebra B(CQ0ab). Is it possible that (4.6.3)
defines a countably additive, C-valued measure on this algebra ? No. The Caratheodory
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
extension theorem
UNIVERSIDAD ([Roy,
DISTRITAL Theorem
FRANCISCO JOSE DE12.8, p. 295] or [Coh, Theorem 1.3.4, p. 18]) assures
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 84 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

us that such a measure on the algebra could be extended to a countably additive, C-valued
measure on B(C^' ), a possibility ruled out by Theorem 4.6.1.
(c) In connection with the discussion of finite-dimensional "distributions" above (see
(4.6.6)), the reader may wish to review Chapter 5 below and the special case associated
with the standard Wiener measure m1 = m in Section 3.2.
(d) The paper of Gelfand and Yaglom seems to be best known for the error [GelYag, p.
58] mentioned above. This is unfortunate since [GelYag] still makes interesting reading
for its positive aspects.
Our discussion so far dealt with A e C+\M+; in this case, (4.6.3) defines an entirely
reasonable countably additive, C-valued measure for fixed t, but difficulties arise as
n -> oo. The "Feynman case" actually corresponds to purely imaginary A; here there are
difficulties even for fixed 1. Let us illustrate this with the specific choice X = -i,~t — t\.
Formally, (4.6.3) then becomes

where B is to range over B(R). Even though (4.6.8) does make sense for bounded Borel
sets, it is not defined for, say,

and it certainly does not give a countably additive, C-valued measure.

Note that the integral in (4.6.8) does exist, as an improper Riemann integral, for
certain unbounded Borel sets such as B = R. (In this latter case, it is a Fresnel integral;
see formula (4.7.10) of Appendix 4.7 below.) When this happens, it is because of the
highly oscillatory nature of the integrand. In contrast, for A. > 0, the integrand in
under U.S. or applicable copyright law.

is a probability density and is not at all oscillatory. For a > 0 and ft ^ 0 so that
A. — a + ip 6 C + \R+, the integrand in (4.6.9) becomes partially oscillatory in character,
and the oscillations become more pronounced as A. approaches the imaginary axis, i.e.,
as a -> 0+. In fact, it follows from the n = 1 case of (4.6.7) that the total variation
[|A|/Re A.]1/2 of the measure in (4.6.9) goes to oo as a -> 0+. Recall from (4.6.7) that
for fixed X e C + \M+, the total variation of the measure goes to oo as the number n of
partition points goes to oo.
We finish this discussion with remarks which come out of what was said above but
EBSCOalso
which Publishing : eBookto
look ahead Academic Collection
the rather (EBSCOhost)
detailed - printed
heuristic on 6/8/2017
discussion of the5:27 PM via
Feynman integral
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
which will ;beJohnson,
AN: 98476 givenGerald
in Chapter 7: The
W., Lapidus, reader
Michel canFeynman
L..; The get some idea
Integral and of the importance of
Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: SOME USEFUL GAUSSIAN-TYPE INTEGRALS 85
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

oscillations for the Feynman integral and of how far that integral is from a Lebesgue
integral by realizing that in the Feynman setting, we have a = 0 and n = oo.
Remark 4.6.3 (a) We do not mean to imply in the preceding paragraph that the Feynman
integral cannot be thought of as an integral with respect to some kind of "measure ".
However, we wish to emphasize that any definition of the Feynmann integral which is
to encompass the integrands of physical interest must take into account the oscillatory
nature of those integrands and the resulting cancellation effects; "the " Feynman integral
is not (at least in a straightforward way) a Lebesgue integral.
(b) The reader may also wish to consult Section 5.1 of Exner"s book [Ex] which has
some overlap with the material above.

4.7 Appendix: Some useful Gaussian-type integrals

We have already found the following two formulas useful:

Later on, we will also use the formula

Here we will discuss briefly and without formal proof the derivation of the general
formula

where (2k - 1)!! := (2k - 1)(2k - 3)(2k - 5 ) . . . 5 • 3 • 1.

Before giving the derivation, we note that for any k = 0, 1, 2,. . . , we have the
under U.S. or applicable copyright law.

formula

since the integrand in (4.7.5) is an odd function. Further, using (4.7.4) and (4.7.5), we
can explicitly calculate the integral

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via
where p is any
UNIVERSIDAD polynomial.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 86 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 4.7.1 None of the Integrals (4.7.1)-(4.7.4) can be calculated by finding an

elementary function whose derivative is the appropriate integrand since, in fact, in each
case, there is no such elementary function. The integrands in (4.7.5) do have antideriva-
tives among the elementary functions, but there was no need for us to find them since
(4.7.5) was clear without this.
Formula (4.7.1) is among the exercises in many calculus books and is derived by
squaring the left-hand side of (4.7.1) and then writing in polar coordinates the double
integral so obtained. Formula (4.7.2) follows from (4.7.1) by differentiating both sides
of (4.7.1) with respect to the parameter a. It is not difficult to justify differentiating under
the integral sign, but we will simply do the calculation:

and so

from which (4.7.2) follows immediately. Differentiating (4.7.2) we get (4.7.3), etc. Let
us look at the induction step. Suppose that (4.7.4) holds for an integer k. Differentiating
both sides of (4.7.4), we obtain

from which it follows that

But (4.7.7) is just formula (4.7.4) for the integer k + 1, as desired.

Remark 4.7.2 One cannot only calculate integrals of the form (4.7.6) using (4.7.4) and
(4.7.5), but also any integral of the form
under U.S. or applicable copyright law.

where p ( v 1 , . . . ,vn) is a polynomial in n variables and aj > 0 for j = 1 , . . . n.

Exercise 4.7.3 Let a < t\ < • • • < tn < b. The Wiener integral

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:27 PM via

withUNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS

p a polynomial in n variables, can always be explicitly calculated. Explain
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
why.
Operational Calculus
Account: ns000601
 APPENDIX: PROOF OF FORMULA (4.2.3a) 87
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 4.7.4 By analytic continuation, it is easy to check thatformulas (4.7.1)-(4.7.5)

remain valid for any complex number a with Re a2 > 0.
The following exercise was already used in Section 4.6 above and will be of further
use in our discussion of the Feymann integral in Chapter 7, especially in Section 7.3.
Exercise 4.7.5 (a) Show that

The (convergent) improper Riemann integral in (4.7.9) is often called a Fresnel integral
in the literature, after the French physicist Augustin Fresnel, who introduced it in his
study of the diffraction of light in geometrical optics.
[Hint: One may establish (4.7.9) by applying the residue theoremfrom complex analysis. J
Deduce from (4.7.9) that formula (4.7.1) remains valid (not only for a with Re a2 >
0, in agreement with Remark 4.7.4, but also) for a2 purely imaginary; namely, if a2 :=
—ia, then

for a > 0, a e R. (A similar formula clearly holds for a <0.)

(b) Do formulas (4.7.2)-(4.7.5) still hold for a2 purely imaginary? In particular,
explain what is wrong with the following argument: Since the integrand is odd, we
have /R v eiav2 = 0; thus, (4.7.2) holds for a2 = -ia € z'R. (Of course, essentially
by definition, this argument could be straightened out by using the notion of Cauchy's
principal value of an integral.)

4.8 Appendix: Proof of formula (4.2.3a)

We wish to show that property (iii)a stated in (4.2.3a) holds in the space (C0a,b, m a );
that is, x(t2) — x ( t 1 ) ~ N(0, a 2 (t 2 — t 1 ))- It is sufficient to show that, given any B > 0,
under U.S. or applicable copyright law.

since the integrand in (4.8.1) is the N(0, a2(t2 — t1)) density function. Recalling that
ma = m1 o a - 1 , we can write

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 88 SCALING IN WIENER SPACE
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the next to last equality in (4.8.2) comes from the fact that x(t2) — x(t\) ~
N(0, t2 — t1) in the space (Cg' fc , m1) and the last equality results from the change of
variables v = alu. Thus (4.8.1) is established, as desired.
We remark that the independence property (ii) of (3.3.34) can be established in the
space (CQ' , m CT ) in a similar way.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

STOCHASTIC PROCESSES AND THE WIENER PROCESS

We would like to relate our treatment of the Wiener process to the larger framework
of stochastic processes. Actually, our discussion will center around two theorems of
Kolmogorov, the first being the "Kolmogorov consistency theorem". The discussion
will be brief and no proofs will be given. A much more thorough discussion of stochastic
processes can be found in many places, for example in [Dool, Fuk, L2, Sh, StroVa,
Wil, Yeh].

5.1 Stochastic processes and probability measures on function spaces

Definition 5.1.1 Let T be a subset of R and let (Q., A, P) be a probability space. A
stochastic process with parameter set T and underlying probability space (Q, A, P) is
a function x : T x £1 —> R such that x(t, •) is a random variable (i.e. a measurable
function) for every t e T.
Remark 5.1.2 The most common choices for T are [a, b], [0, +00), {1, 2, 3,...} , K
and {..., -2, -1, 0, 1, 2,...}. We will fix T = [a, b] for our discussion. When T =
{1, 2, 3,...}, one often speaks of a sequence of random variables. This type of "stochastic
process" is studied even in introductory courses in statistics and probability theory.
The theory of stochastic processes is intimately related to the study of probability
measures on spaces of functions. We wish to begin by making that connection. The func-
tion space for much of what we will have to say will be RT, the space of all K-valued
functions on T — [a, b]. (This is, in fact, too large a space for more advanced consider-
ations, a point we will comment on towards the end of this chapter.)
Let n be a positive integer and suppose that a < t\ < • • • < tn < b and that, for
j = 1 , . . . , n, — oo < aJ < Bj < +00. We take I to be the collection of all subsets
under U.S. or applicable copyright law.

I of RT of the form

Such sets / are called intervals in RT. Much as in our earlier setting, we could show
that 1 is a semi-algebra of subsets of RT, in the sense of Definition 3.2.1. Let T denote
the EBSCO
cr-algebra generated by 1.
Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
We firstFRANCISCO
DISTRITAL indicate JOSEhow a given stochastic process x : T x £2 —> R gives rise
DE CALDAS
to aAN:probability
98476 ; Johnson, Gerald W.,
measure on Lapidus, Michel L..;
the function space RT equipped
The Feynman Integral and Feynman's
with the cr-algebra T.
Operational Calculus
Account: ns000601
 90 STOCHASTIC PROCESSES AND THE WIENER PROCESS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Define u on J by

where/isgivenby(5.1.1).Thisdefinitionof/*makessensebecausejc(t1, .), ...,x(tn, •)

are random variables on (£1, A, P). Since P is countably additive, it is not hard to show
that u is countably additive on I. We can now apply the Caratheodory extension theorem
and extend /u, to a countably additive probability measure on a (X) = J-. In summary,
the stochastic process x gives rise to a probability measure u on the function space RT
equipped with the a -algebra f.
In the converse direction, a probability measure u on (R T , F) easily gives rise to a
stochastic process x with parameter set T and underlying probability space (R T , F, u,).
Given u on (R T , f), simply let x(t, f) := f(t). It is easy to see that A: (t, .) is a random
variable on (R T , F, u,) for all t e T.
Remark 5.1.3 If one begins with (R T , F, u) and induces a stochastic process as above
and then uses this stochastic process to induce a measure on (R T , F), you are back
where you started (i.e. with (R T , F, fJ.)). On the other hand, if one starts with a stochastic
process x : [a, b] x Q —> K and induces (RT ,f, /A) and then uses ( R T , J, u)to induce
a stochastic process XQ : [a, b] x RT —» R, the new stochastic process is related to but
need not be the same as the original process x. In fact, for the most studied stochastic
processes, Q, is a space of functions on T, but, typically, £2 is much smaller than RT. In
our earlier work, for example, we had £2 = C0 .

5.2 The Kolmogorov consistency theorem

Rather than beginning with either a stochastic process or a probability measure on
(R T , J"), one typically starts with more primitive information about the "probability"
of finitely based sets. Recall that this was Wiener's starting point (see Section 3.2). One
then wishes to extend this to a full fledged probability measure so as to have available
the power of Lebesgue theory. According to a famous theorem of Kolmogorov which
appeared in his fundamental 1933 monograph [Kol], this can be done under minimal
conditions on the initial "probabilities". We next wish to lead up to and then state the
Kolmogorov consistency theorem. (See, for example, [Yeh, Theorem 2.4, p. 14].)
Temporarily suppose that we have a stochastic process and so a probability measure
under U.S. or applicable copyright law.

fj, on (R T , F). Given any finite sequence t1,... ,tn from T — [a, b], not necessarily
arranged in increasing order, we define fj,^-—'" acting on the interval («i, $1] x • • • x
(a n ,£ n ] in Rn by

It is easy to see that if t is any permutation of the integers 1 , . . . , n, then

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE KOLMOGOROV CONSISTENCY THEOREM 91
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and, if t e [a,b] \ {ti, . . . ,t n },

Further, it is not hard to show that each set function ut1,. . . .,tn is countably additive on the
semi-algebra of intervals in Rn and so has a countably additive extension to the cr-algebra
generated by the intervals; i.e. to B(R n ). This family of measures is called the family of
finite-dimensional distributions of u (or of the stochastic process associated with u).
In summary: Given (R T , .F, u),we get a mapping carrying finite sequences t1, . . . ,tn
from [a,b] into probability measures on B ( R n ) such that the consistency conditions
(5.2.2) and (5.2.3) are satisfied.
We are now ready to state Kolmogorov's result which starts with the family of
measures {ut1, . . . ,tn} and produces the measure u, on (RT, f).
Theorem 5.2.1 (Kolmogorov consistency theorem) Suppose that for every finite
sequence t1, . . . , tn from[a, b], we have a probability measure ut1,. . . ,tn on B ( R n . Further
suppose that the family {ut1,. . .,tn} satisfies the consistency conditions (5.2.2) and (5.2.3).
Then there exists a probability measure u on (R T , F-) such that the finite-dimensional
distributions of u are the given ones.
Remark 5.2.2 The Kolmogorov consistency theorem can be applied to the family of
finite-dimensional Borel measures singled out by Einstein and Wiener. Doing this pro-
duces a countably additive measure w on (R7, F) which is a version of the Wiener
process.
The probability space (RT, T, w) is much less useful than (Co a,b , B(Co a,b ), m1)
roughly because R7 is too large and F is too small. Further on we will provide more
information about how (R r , :F, w) and (Cg'b, B(CQ' & ), mi) are related. Next, however,
we state some results which indicate the deficiencies of F.
Given any subset S of T, we think of Rs as equipped with the product topology.
Recall that sets of the form
under U.S. or applicable copyright law.

where { s 1 , . . . , sn] is a finite subset of 5 and U\,..., Un are open subsets of R, form a
basis for the product topology on Rs. Given a countable subset 5 of T and B e B(RS),
let

where f\s denotes the restriction of / to S.

Proposition 5.2.3 Let Q := {J(S; B) : S is a countable subset of T and B e B(RS)}.
Then Q is a a-algebra of subsets of RT and F- c Q.
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
We nowFRANCISCO
DISTRITAL deduce JOSE
easily the following corollaries.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 92 STOCHASTIC PROCESSES AND THE WIENER PROCESS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 5.2.4C0a,bwF?.
Corollary 5.2.5 Let g e RT. Then the one element set [g] is not in T. In fact, if Fef
and F = 0, then card (F) = cc, where c denotes the cardinality o/R.
One can see from Proposition 5.2.3 and Corollaries 5.2.4 and 5.2.5 that many natural
subsets of M.T are not measurable with respect to the space (R T , T). It is not surprising
then that various natural functions are not J'-measurable. The next corollary provides
one illustration of this.
Corollary 5.2.6 The function G : RT ->• (-00, +00] defined by G(f) := sup{f(t) :
t e T] is not F-measurable.

5.3 Two realizations of the Wiener process

We now turn to the relationship between the two versions (C0a,b, B(C0 a,b ),m1) and
(RT, F, w) of the Wiener process. We begin this by stating a theorem of Doob which
is of interest in its own right. For discussion of this result and of related matters, see
Sections 27 and 28 of [Yeh], especially Lemma 27.5.
Theorem 5.3.1 (Doob) Let (n, A, P) be a probability space and let no be a subset of
n such that P * ( n o ) = 1. Define P0 on the a-algebra A n n0 : = [A n no : A e ,4} by
Po(A n no) = P (A), for A e A
Then PQ is well-defined and (no An no, Po ) is a probability space.
Note that the set no in Theorem 5.3.1 is not required to be A-measurable; in fact,
that is the main point of the result.
Recall that the outer measure P* associated with P is defined by

for any subset A of n. (See, for instance, [Con, §1.3].)

Theorem 5.3.2 w*(C% a,b ) = 1 and so (by Theorem 5.3.1) w0(F n C0a,b) := w(F) is
a probability measure on the a-algebra F n C0ab . Further, F n C0a,b =B(C0a,b) and
wo = wo on this a-algebra. In summary, (C0a,b, B(C0a'b), m1) = (C0a,b, F n C0a,b, W0).
under U.S. or applicable copyright law.

It is not necessary for our discussion, but we remark thatw*;(C0a,b)= 0. [Of course,
here as well as in Theorem 5.3.2, w* (resp., w*) denotes the outer (resp., inner) measure
associated with w.]
Theorem 5.3.2 shows the relationship between (C0a,b, B(C0 a,b ), m1) and the version
of the Wiener process provided by the Kolmogorov consistency theorem. One might
think that the method we have been describing to produce(C0a,b, FnC0a,b , W0) actually
provides a proof of the countable additivity of WQ = m1. This is not the case because
the proof thatw*(C0a,b) = 1 makes use of the countable additivity of m1. Can the
EBSCO Publishing
Kolmogorov : eBook be
approach Collection
used to(EBSCOhost)
obtain the- printed on 6/8/2017
countable 5:33 PM
additivity of via UNIVERSIDAD
Wiener measure on
DISTRITAL FRANCISCO JOSE DE CALDAS
(C0a,b, B(C0a,b))?
AN: 98476 It can
; Johnson, Geraldif W.,
it isLapidus,
supplemented
Michel L..; by
The another result and
Feynman Integral of Kolmogorov,
Feynman's which
Operational Calculus
Account: ns000601
 TWO REALIZATIONS OF THE WIENER PROCESS 93
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

we now state. (See, for example, [Kal1, Theorem 1.2.2, p. 9] as well as [Yeh, §I.6] for a
closely related statement.)
Theorem 5.3.3 (Kolmogomv) Suppose thatfor each finite sequence t1,..., tn from T =
[a, b] we have a probability measure ut1,... tn B(Rn) such that the family {ut1,...,tn}
satisfies the consistency conditions (5.2.2) and (5.2.3). Suppose further that there exist
positive constants S, €, and K such that for all t1, t2 e T,

Then there exists a probability measure u, on (Ca'b,B(Ca'b)) whose finite-

dimensional distributions are the given ones.
Remark 5.3.4 (a) Even without (5.3.2), the Kolmogomv consistency theorem (Theorem
5.2.1) can be applied, yielding a measure on (R.T, J-). The inequality (5.3.2) yields the
additional important piece of information that the measure can be taken on the space
(Ca'b, B(Ca-b)).
(b) In the case of the finite-dimensional probability distributions considered by
Einstein and Wiener, the inequality (5.3.2) is satisfied with 8 = 4, e = 1, and K = 3.
We finish this chapter with an example of a consistent family of finite-dimensional
distributions for which the corresponding probability measure /z is not concentrated on
(Ca-b, B(Ca'b)). We will simply state the example and settle for a remark which gives
some clue why the example has the asserted property.
2
For each t € [a, b] and B e B(R), let n'(B) := fB -—e^du. Now, given any
finite sequence t\,...,tn from [a, b] and B e B(E"), let ^•—'"(B) = (/i'< x • • • x
Htn)(B). It is easy to check that this family {/j,'J '"} of probability measures satisfies
the consistency conditions (5.2.2) and (5.2.3), and so, by the Kolmogorov consistency
theorem, a corresponding measure /u, on (R r , ^F) is determined. Why is it that /z is not,
in this case, concentrated on C a,b ? The measures fj.'^-—'" have been defined so that
{x(t) : t 6 [a, b]} is an uncountable family of independent random variables. However,
if x(t) and x(s) are independent no matter how close t and s are, it seems reasonable
that x is likely to be discontinuous.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

6
QUANTUM DYNAMICS AND THE SCHRODINGER EQUATION

At the moment I am struggling with a new atomic theory. If only I knew more mathematics!
Erwin Schrodinger, letter to Willy Wien, 27 December 1925

I am so happy to have escaped the terrible mechanics ... which I never really understood.
Now everything is linear, everything can be superposed.
Erwin Schrodinger, letter to Willy Wien, 22 February 1926

You are only going to spoil Heisenberg's physical ideas with your futile mathematics.
Wolfgang Pauli to Max Born, 19 July 1925

Heisenberg's new work, which will very soon appear, looks very mystical, but it is certainly
correct and deep.
Max Born, letter to Albert Einstein, 15 July 1925

Just now I am teaching the foundations of poor deceased mechanics, which is so beautiful.
What will her successor look like? With that question I torment myself incessantly.
Albert Einstein, letter to Heinrich Zangger, 14 November 1911

[All the above quotations can be found in [Han], pages 5, 48, 63, 179 and 246, respectively.]

We give in this section a brief discussion of some of the ingredients in the usual approach
to quantum dynamics via the Schrodinger equation and operator theory. Our discussion
will be far from complete and is intended primarily to permit comparison with the
conceptually very different approach to quantum dynamics provided by the Feynman
integral. Our treatment of the Feynman integral itself will begin in the next chapter with
a discussion of the heuristic ideas behind the subject. Neither this chapter nor the next
is intended to be mathematically rigorous.
under U.S. or applicable copyright law.

6.1 Hamiltonian approach to quantum dynamics

Consider a single (nonrelativistic) quantum particle moving in "configuration space" R.
(We could just as well consider a particle or system of particles moving in R3.) The
state of the quantum particle at time t is given by a probability amplitude, that is, by a
C-valued function \jr(t, •) € L 2 (R) such that

Given a subset A of R, the probability that the particle is in A at time t is given by

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TRANSITION AMPLITUDES AND MEASUREMENT 95
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We can see already that the framework in quantum mechanics is quite different than in
classical Newtonian mechanics. The state of a classical particle moving in R is specified
at any time t by two numbers giving, respectively, the position and the velocity of the
particle at that time. In contrast, the state of the quantum particle at time t is given by
an element on the unit sphere of the Hilbert space L 2 (R). Further, the quantum state
only gives us probabilistic information about the location of the particles whereas the
classical state tells us precisely where the particle is.
The basic problem of quantum dynamics: Given the forces involved in the problem
specified by a potential energy function V : R -» R (called the "potentiaF', in short) and
given the initial state of the panicle <£(£) = ^(0, £). find the evolution of the state in
time; i.e. find ty(t, £) for all t e R. The function i]s(t, f) should satisfy Schrodinger's
equation, the dynamical law.

where m is the mass of the particle and h is Planck's constant h divided by 2n. We wish
to rewrite (6.1.3) in operator notation. Let the formal Hamiltonian or energy operator H
be given by

where V denotes the operator of multiplication by the potential energy function V; that
is, (V<p)(l=) = V(f)<»($), for 6 I 2 (R).
Then (6.1.3) can be rewritten as

It is easy to solve (6.1.5) formally, obtaining

It can be proved rigorously that (6.1.6) makes sense and gives a solution to (6.1.5) under
rather general conditions on the "potential" V. There is an extensive literature on this
under U.S. or applicable copyright law.

topic (see, for example, [CyFKSi, Gol, Kat8, ReSi2, vC]), some of which we will have
occasion to discuss further on.

6.2 Transition amplitudes and measurement

Feynrnan usually writes his formulas for transition amplitudes VKO, t', u, i-) rather than
for the amplitudes ^(t, f) themselves. The complex number VKO, t; u, ) is described
as the amplitude to go from u at time 0 to £ at time t. There is a simple formula for the
probability amplitude in terms of the transition amplitude and the initial state (/>:

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 96 QUANTUM DYNAMICS AND THE SCHRODINGER EQUATION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In words, the amplitude for being at f at time t is the "sum" (integral, actually) over all
u of the amplitude for being at u at time 0 times the amplitude of transition from u to £
between times 0 and t.
If no "measurement" of position is made over the time interval from 0 to t, transition
amplitudes combine according to the formula

where 0 < s < t. The reader familiar with probability theory should note the striking
similarity between (6.2.2) and the formula for combining the transition density func-
tions of a Markov process. Indeed, the formula for transition densities is written exactly
like (6.2.2). In spite of this, the situations are different in a crucial way. The transition
amplitudes are C-valued functions and (6.2.2) involves cancellation effects which can-
not possibly be present with the necessarily nonnegative transition densities of a Markov
process.
Since, according to (6.1.2), probabilities are finally computed using the probability
density functions.|V(t,.)|2, why not work directly with these probability densities? The
answer is that the cancellation effects involved in (6.2.1) and (6.2.2) are absolutely
crucial to quantum phenomena, whether approached from the Schrodinger equation-
operator theory point of view or from Feynman's point of view. For example, if there is
no "measurement" of position over the time interval from 0 to t, the formula

is false. Actually, things are much stranger than we have indicated so far.
If there is a "measurement" at time s, (6.2.3) is true and (6.2.2) is false. This peculiar
state of affairs is tied up with the dual wave-particle nature of matter and the Heisenberg
uncertainty principle. We note that the change in the quantum system brought about by
a "measurement" is often referred to as "reduction of the wave packet" or "collapse of
the wave function" [Jam, vN].

6.3 The Heisenberg uncertainty principle

The Heisenberg uncertainty principle tells us that we cannot simultaneously determine
the position and momentum of a quantum particle to an arbitrary degree of accuracy.
under U.S. or applicable copyright law.

(Recall that momentum = (mass)(velocity).) Further, this lack of precision is not simply
due to limitations in the measuring devices, but rather is built into the theory itself. The
result says that
(the standard deviation of a position measurement)
((the standard deviation of a momentum measurement)

Thus as we measure position more and more precisely, our ability to determine momen-
tumEBSCO
(equivalently,
Publishing : velocity) diminishes
eBook Collection accordingly.
(EBSCOhost) In6/8/2017
- printed on fact, if 5:33
we know
PM via the exact location
UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
of aAN:
quantum particle, it is not possible to say anything precise about its momentum.
98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 HAMILTONIAN FOR A SYSTEM OF PARTICLES 97
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We turn next to a brief discussion of some of the implications of the uncertainty

principle. The reader who knows nothing about quantum mechanics but recalls the pre-
eminent role played by the path of a particle in classical mechanics (even in elementary
calculus) may have been surprised that there has been no mention of paths in our dis-
cussion of the usual approach to quantum dynamics. This is not due simply to our brief
treatment; to put it perhaps overdramatically, paths were banned from quantum mechan-
ics. When the classical path is known, the exact position and velocity (and so momentum)
can be found at any time t. But such information is forbidden to us in quantum mechanics
by the Heisenberg uncertainty principle. We will see in the next section that Feynman's
approach restores the path to a prominent, although altered, role in mechanics on the
quantum level. This permits the restoration of further important concepts from classical
mechanics that depend on the path.
The constant | appearing on theright-handside of (6.3.1) would not forbid classical
paths if it were equal to 0. This suggests that quantum mechanics should go over to
classical mechanics as H -> 0. In fact, there is an appealing heuristic argument to
that effect in the framework of the Feynman integral. We should mention that h is
extremely close to 0 compared to scales of interest in classical physics; namely, h =
1.054 x 10-34 J s (= 1.054 x 10-27 erg s). This tiny value of h is intimately connected
with the fact that quantum effects are not noticeable in classical situations.
We note that h is equal to h/2n, where h is Planck's constant; it is often referred to
as the quantum of action in the physics literature.
We saw in the concept of state that quantum mechanics involves probability theory
right from the beginning. As we can see from (6.1.2), there is a random variable associated
with the position of the particle at any time t. Although we have not directly discussed
it, there is also a random variable connected with the momentum of the particle at
t. However, because of the Heisenberg uncertainty principle and inequality (6.3.1) in
particular, the position and momentum cannot have a joint distribution. Why should this
be so? We give a remark which provides some clue. If there were a joint distribution,
then fixing a precise value of momentum would force the distribution function P(X <
x) of the random variable X associated with position to be a uniform distribution on
R. However, no such distribution exists. The role of probability theory in quantum
mechanics has been and continues to be a subject of much interest; see, for example,
[Ne5] and [B1CZ, esp. pp. 12-15] for a more precise discussion related to the above as
well as various further results, remarks, and references.
under U.S. or applicable copyright law.

6.4 Hamiltonian for a system of particles

So far, we have considered a single quantum particle moving in E. We finish this chapter
by giving the form of the Hamiltonian operator H (see (6.1.4)) and the Schrodinger
equation (6.1.5) in more complicated situations. In the case of a single particle moving
in R3, the sum of the second partials with respect to each of the three space variables,
that is, the Laplacian A, replaces ~j in (6.1.4). We have

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 98 QUANTUM DYNAMICS AND THE SCHRODINGER EQUATION
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the potential V is now a function of three variables. The form (6.1.5) of the
Schrodinger equation does not change but H is now given by (6.4.1) and for each t, the
state \ f f ( . , t ) is on the unit sphere of L2(R3).
If we are dealing with a system of N particles each of which is moving in R3, the
configuration space is taken to be R3/v and

where mj is the mass of the jth particle,

and V is a function on R3n which takes into account the interactions between the particles
as well as the influence of external forces. The form (6.1.5) of the Schrodinger equation
still remains the same, but the state \fr(-,t) of the quantum system at time t is now an
element of the unit sphere of L 2 (R 3N ).

There are many books on the standard approach to quantum mechanics with varying
degrees of mathematical rigor. Here are several references which may be of some help to
the reader [BkExH, Cass, CoTDL, Han, LL, Pru, ReSil-4, Sud, Th]. In addition, some
of the early papers on or leading to quantum mechanics are collected in [vW].
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

THE FEYNMAN INTEGRAL: HEURISTIC IDEAS AND

MATHEMATICAL DIFFICULTIES

These curious mathematical problems are sidestepped by the subdivision process. However, one
feels like Cavalieri must have felt calculating the volume of a pyramid before the invention of
calculus.
Richard P. Feynman, 1948 [Fey2, p. 372]

The occurrence of i (which is essential for Quantum Mechanics!) makes manipulations with
integrals like [formulas (7.4.2) and (7.4.3) below] extremely tricky.
First the existence of the limit is, in general, difficult to establish, and secondly it is awkward
to be dependent on a particular discretization of the action integral. Still the intuitive appeal of
Feynman's definition is enormous, and it gave birth to a dream that perhaps all of Physics could
be rewritten in terms of "sums over histories".
MarkKac, 1980 [Kac4,p. 51]

The role of action in quantum physics was clarified in an extraordinary fashion by Richard
Feynman, who based his discussion on earlier work of Dirac. In the absence of a fundamental
quantum theory we are forced to formulate his ideas as a recipe for "quantization", i.e., transition
from the classical description of some physical system to the quantum description. According to
this recipe, one imagines that each classical history 0 contributes to the quantum description of
the history of the system, but it does so with its complex weight (phase factor) eiS(o). (The action
[5(0)] is, of course, measured in units of h.)
Yu I. Manin, 1979 [Manil, p. 92]

Physics is where Action is.

Author unknown. (Quoted in [Manil, p. 88].)
under U.S. or applicable copyright law.

7.1 Introduction
The Feynman integral was introduced by Richard Feynman in 1948 in his seminal paper
"Space-time approach to non-relativistic quantum mechanics" [Fey2]. Feynman's paper
has had a major influence directly or indirectly on several areas of mathematics and
physics. Our main purpose in this chapter is to discuss some of Feynman's heuristic
ideas and the mathematical difficulties with them. A secondary purpose is to indicate
briefly some of the approaches to the mathematically rigorous theory of the Feynman
integral which have arisen rather directly from Feynman's paper; we will study these
approaches more thoroughly as we continue.
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
We begin
DISTRITAL with introductory
FRANCISCO JOSE DE CALDAS remarks which will be somewhat vague but which we
hopeAN:will
98476provide some
; Johnson, Geraldorientation
W., Lapidus, to the L..;
Michel subject. Feynman's
The Feynman Integral formulation
and Feynman's of the basic
Operational Calculus
Account: ns000601
 100 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

principles of quantum mechanics is very different from the usual formulation in terms of
the Schrodinger equation and operator theory. This difference already appears in the first
three words in the title of Feynman's paper: "Space-time approach ... ". Feynman was
thinking of space coordinates as a function of the time coordinate; that is, he was thinking
in terms of paths. Recall from the last chapter that paths were banned from quantum
mechanics in the standard approach to the subject. Along with the reinstated path came
related concepts from classical mechanics such as the Lagrangian, the action integral,
and the principle of least action. Here is a one sentence summary of what Feynman
did: He expressed the evolution of a quantum particle in terms of a heuristically defined
integral over the space of paths along which the particle might travel.
The degree of interest in the Feynman integral has varied considerably since 1948, but
the interest has endured and is probably now at its peak. The integral has a strong heuris-
tic appeal and possesses pleasant formal transformation properties. The relationship
between classical and quantum mechanics seems clearer; indeed, the fact that quantum
mechanics passes over into classical mechanics as Planck's constant goes to 0 can be
obtained heuristically from an appealing stationary phase argument ([Fey 1], [Fey2, §7])
as we will discuss below. Some form or another of the Feynman integral, usually not
mathematically rigorous, is used these days in many books and articles on theoretical
physics. Indeed, the Feynman integral is at least formally applicable in some cases where
it is not clear how to use the standard approach or the standard approach is unwieldy. The
following references give a sample of the recent physical literature on or involving the
Feynman integral [At2, At5-8, Bae, BrBh, Fa2, FaSl, GibHaw, GliJa, GreSWit, Gut2,
GutlKS, Haw, ItzZu, Kak, Kon2,4,8, LuRSS, Manil,2, Polc3, Polyl-3, Pop, RaSinW,
Ram, Riv, Roz2,3, Schu, Sinl,2, We 1,2, Whe, Wig, Witl-15, Zee]. They provide, in
particular, various applications of Feynman path integrals and closely related functional
integrals to applied physics and chemistry [GutlKS, LuRSS, Schu, Wig], including optics
and polymer science, as well as to theoretical physics, including quantum mechanics,
statistical physics, quantum field theory, gauge theory, string theory and quantum grav-
ity [At2, At5-8, Bae, BrBh, FaSl, GibHaw, GliJa, GreSWit, Gut2, GutlKS, Haw, ItzZu,
Kak, Kon2,4,8, LuRSS, Manil-2, Polc3, Polyl-3, Pop, RaSinW, Ram, Riv, Roz2,3,
Schu, Sin 1,2, Wei,2, Whe, Witl-15, Zee]. Some of the influential work of Edward
Witten [Witl-15], which is relevant to many areas of quantum physics and relies on
the heuristic use of a Feynman-type integral, is also included. Aspects of Witten's work
(particularly [Wit4, Wit7 and Wit12-14]) are briefly discussed in [At8, Fa2, Jone6]
under U.S. or applicable copyright law.

and, in more detail, in [At2, At6,7, Benn, Kau8,9]. We will discuss especially parts of
[Witl3, 14] in Section 20.2.A below, connecting (heuristic) Feynman path integrals and
knot theory or low-dimensional topology. Some other examples of the use of heuristic
Feynman integrals in contemporary physics (and mathematics) will be briefly discussed
in Section 20.2.B.
The Feynman integral is difficult to deal with in a mathematically precise way. In fact,
two of the three words in the phrase "the Feynman integral" are extremely misleading.
The word "integral" calls to the mind of most mathematicians the Lebesgue theory which
is aEBSCO
theory of absolute
Publishing : eBook convergence that includes
Collection (EBSCOhost) - printedseveral powerful
on 6/8/2017 5:33 PM theorems permitting
via UNIVERSIDAD
the interchange of limit processes. In contrast, when the Feynman integral exists in
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN'S FORMULA 101
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

cases of physical interest, it tends to be because of the highly oscillatory nature of the
appropriate integrands and the resulting cancellation effects. Further, the interchange of
limit processes is typically difficult to justify.
The word "the" in the phrase "the Feynman integral" is more misleading than the word
"integral". There have been almost as many approaches to "the" Feynman integral as
there have been mathematicians who have worked on the subject. Much of the research
since the influential monograph of Albeverio and Hoegh-Krohn in 1976 [AlHol] has
involved, in one way or another, transform assumptions on, say, the potential V and
the initial probability amplitude <p. It is now known that there is a good deal of unity
under transform assumptions (see [Jol, KalBr] and especially [KalKanK]), as will be
discussed in Section 20.1.A, but, otherwise, surprisingly little is known about how the
various Feynman integrals are related to one another. Further, no one theory seems to
be best in all respects. (Two results in this book, Corollaries 13.3.1 and 13.3.2, show
that under very general conditions, certain operator-theoretic approaches to the Feynman
integral agree with one another and with the standard Hamiltonian approach to quantum
mechanics.)
Communication between researchers working on the Feynman integral has often
been difficult. People come to the subject with diverse backgrounds such as theoretical
physics, mathematical physics, probability theory, operator theory, partial differential
equations, and classical or functional analysis. What is well known or even trivial for
one may be difficult for another. Even notational differences, while not a serious obstacle
taken individually, contribute to the problem when regarded collectively.
The intrinsic difficulty of the subject along with the multiplicity of definitions and
notation have made the Feynman integral difficult to learn. Where should one begin?
We think that it is best to begin by trying to get some feeling for the key heuristic ideas,
some of which we will discuss below. A good part of what we will do is based on
Feynman's original paper [Fey2] which is still well worth reading. It is not easy to read,
perhaps especially for mathematicians. Feynman had excellent mathematical instincts;
a careful reading of his paper shows that he had a good feeling for where the main
mathematical obstacles were to be found. However, he was not concerned with proving
theorems, and he frequently used mathematical words in a manner that is less precise
than mathematicians prefer. We are ready now to discuss Feynman's formula for the
transition amplitude.
under U.S. or applicable copyright law.

7.2 Feynman's formula

Feynman's beautiful heuristic expression for the transition amplitude in terms of a path
integral is given by

For the notion of transition amplitude, see the discussion associated with formula (6.2.2)
in the previous chapter. In (7.2.1), €„',[ is the space of paths x in €"•' such that;t(0) = u
andEBSCO
x ( t ) Publishing
= v. The : eBook Collection
symbol T> is(EBSCOhost)
supposed- to
printed on 6/8/2017
denote 5:33 PM
a measure via UNIVERSIDAD
which weighs all paths
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 102 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

equally, and K is to be an "appropriate" normalizing constant. Further, S(x) is the

classical action associated with the path x; that is,

where m is the mass of the particle and V is the potential energy associated with the
problem. The integrand in (7.2.2) is the classical Lagrangian associated with the motion
of a particle under the influence of the potential V. The reader should note the prominent
role played right from the start by classical riotions which had disappeared from quantum
mechanics.

There are serious mathematical difficulties associated with (7.2.1), a fact of which
Feynman was largely aware at least on an intuitive level:
(a) There is no countably additive measure T> on €„][ which weighs all paths equally.
Recall that we discussed closely related facts in Section 3.1; see, in particular, Theorems
3.1.3 and 3.1.5. (Those theorems do not precisely fit the present set-up because C£,'(, is
not a linear space.)
(b) The "normalizing constant" K was not specified by Feynman at this stage. (How-
ever, an appropriate constant was specified by Feynman at each stage of his approximation
formula which we will discuss in Section 7.4 below.)
(c) The function S(x) fails to be defined on much of C£(, since it involves the
derivative of x and C£;(, consists of continuous paths.
(d) Even if T> did exist as a countably additive measure evenly spread over Cu',v,
there would still be problems. The space Cul'v is not compact and so its measure would
be infinite and the integrand in (7.2.1) would not be absolutely integrable.

In spite of the fact that formula (7.2.1) is difficult to make sense of in a precise way,
it represents sharp insight on Feynman's part and has led to interesting results in various
directions. Where did it come from? How is it that paths and action integrals are present in
mechanics on the quantum level when it seemed that they had disappeared? In particular,
where does the integrand exp {fa S(x)} in (7.2.1) come from? We would like at least some
rough heuristic clues.
Feynman says in the introduction of [Fey2] that his formulation was suggested by
under U.S. or applicable copyright law.

remarks of Dirac ([Dirl], [Dir2, esp. §32]) concerning the relation of the classical action
to quantum mechanics. In [Fey2, p. 374], the formula

appears. It is to give the development of the probability amplitude over the small (or
infinitesimal) time interval from £ to s + e. Here, S(uj+1, uj) is the action integral
involved in going from uj at time s to uj+1 at time s + f along the classical path under
the EBSCO Publishing
influence : eBook
of the Collection
potential V. (EBSCOhost)
(Of course,- printed
S(uj+\,on 6/8/2017
Uj) also5:33 PM via UNIVERSIDAD
depends on 5 and e; see
DISTRITAL FRANCISCO JOSE DE CALDAS
(7.3.1).)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman'sus try to get
We will indicate below where (7.2.3) comes from, but first let
Operational Calculus
Account: ns000601
 FEYNMAN'S FORMULA 103
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

some idea how (7.2.1) might arise from (7.2.3) if (7.2.3) is iterated over the continuum
of "instants" from 0 to t. It is helpful to begin by seeing what happens when (7.2.3) is
applied twice:

In light of (7.2.4), if we think of starting from (0, u) and iterating (7.2.3) from instant to
instant until we reach (t, v), it seems somewhat plausible that we might obtain a formula
like (7.2.1). The (fictional) measure T> obtained from the iteration should be the product
of a continuum of copies of Lebesgue measure; that is,

The space of integration after the iteration should be roughly RM'^. by which we mean the
space of all functions on [0, t ] which begin at u at time 0 and end at v at time t. Further on
in his paper [Fey2, p. 376], Feynman discusses the reasonableness of restricting attention
to the continuous paths €%',[• It is somewhat plausible that the integrand in (7.2.1) could
be obtained from the integrand in (7.2.3) after iteration because the former involves
the exponentials of an action integral over (0, t) and the latter involves the product of
exponentials of action integrals over the time intervals (0, e), (c, 2e),..., (t — 6, t).
Even if one swallows the large dose of rough heuristics just discussed, we are still left
with the question of where (7.2.3) comes from. It is an analogue of Huygens' principle.
Huygens' principle of wave optics relates wave optics to geometrical (or ray) optics.
Formula (7.2.3) is Huygens' principle for wave (i.e. quantum) mechanics: If the ampli-
tude of the wave \/f is known for all uj at time s, its value at uj+1 at time s + e is a sum
of contributions from all points uj at time s; each such contribution is w (s, uj) changed
(or delayed) in phase by an amount proportional to the action it would require to get
under U.S. or applicable copyright law.

from (s, uj) to (s + e, u j + i ) along the path of least action of classical mechanics. (See
Figure 7.2.1.)
We indicate the analogy between Huygens' principle in wave optics and in wave
mechanics a little more fully in Table 7.2.1.
Remark 7.2.1 (a) Note that (7.2.3) as a "sum of contributions from all points uj at
time s" will involve cancellation (or interference) effects typical of wave motion of all
kinds.
(b) Feynman comments [Fey2, p. 377] that Huygens' principle actually works better
in wave mechanics than it does in wave optics where Kirchhoff's modification is needed.
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
Kirchhoff's modification
DISTRITAL FRANCISCO JOSE DErequires
CALDAS that both the amplitude and its time derivative be known.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 104 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 7.2.1. Huygens' principle for wave mechanics: The amplitude at point uj+1 at time
i + e is the sum of all contributions from all points HJ at time s

TABLE 7.2.1. Wave mechanics relates to classical mechanics as wave

optics relates to geometrical optics
Optics Mechanics
1. Wave optics 1. Wave (or quantum) mechanics
2. Geometrical (or ray) optics 2. Classical (or geometrical) mechanics
3. Principle of least time 3. Principle of least action
4. Path of least time 4. Path of least action (or classical path)

This is associated with the fact that the differential equation for wave optics is second
order in time. On the other hand, the Schrodinger equation is first order in time and,
correspondingly, it is enough to know the amplitude.
(c) We refer, for example, to [BorWoJfor a thorough discussion of the basic physical
principles of optics referred to here.
(d) Feynman never stated his formula in one place in [Fey2] or even in [FeyHi] in
quite the form (7.2.1) although formula (2-25), p. 34 of the book of Feynman and Hibbs
[FeyHi] comes close. Feynman mentions various possibilities, but it seems clear from
the text of [Fey2] that (7.2.1) and an approximation formula to be discussed in Section
7.4 are what he primarily had in mind. Formula (7.2.3) is explicitly stated in [Fey2], as
previously noted.
In Feynman's formulation, previously discarded concepts from classical mechanics,
the path and the action integral, are used in an essential way to describe the evolution of
under U.S. or applicable copyright law.

a quantum particle. The principle of least action is perhaps even more crucial here than
in classical mechanics. (See [Fey2], as well as Feynman's 1942 Ph.D. Dissertation The
Principle of Least Action in Quantum Mechanics [Fey l].)
We refer the interested reader to Mehra's book [Me, esp. Chapter 6] for a detailed
history of the so-called "Hamilton's principle of least action" and for a thorough analy-
sis of Feynman's (and Dirac's) contributions to this subject. In particular, after having
discussed the fundamental importance for much of contemporary physics of Hamilton's
principle of least action (extended in a straightforward way to systems with infinitely
many degrees
EBSCO of freedom),
Publishing Mehra(EBSCOhost)
: eBook Collection raises the- question
printed onas to what
6/8/2017 liesPM behind
5:33 this principle.
via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
He AN:
then states [Me, p. 129] (the full quote will be provided towards the end of this book
98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN'S FORMULA 105
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

in Section 20.2.B below): "The physicists before P. A. M. Dirac and Richard Feynman
were not able to answer this question, because the principle of least action is perhaps the
most basic principle of classical physics, but its full significance lies in quantum physics;
one can find it only in Feynman's path integral formulation of quantum mechanics."

Connections with classical mechanics: The method of stationary phase

An appealing heuristic argument using (7.2.1) "explains" how quantum mechanics passes
into classical mechanics as H —> 0. We sketch the idea roughly; for a more detailed
heuristic discussion, see [Fey2, §7] and [FeyHi, §2-3].
First note that, in terms of the absolute value of the integrand of (7.2.1), all paths
contribute equally and the classical path x is indistinguishable from any other path.
However, x will be a stationary point for the function S. (In most cases, x minimizes the
function 5; that is, x is the path of least action.) Hence the derivative SS, a functional
derivative actually, satisfies (8S)(x) = 0. Let us contrast x with a path y such that
(8S)(y) = 0. Consider a small set of paths about y such as in Figure 7.2.2. Let y\ and
yi be two paths in the set Since (8S)(y) = 0, we expect to have $(y1) ^ S(y2)- Even
if the real numbers S(y1) and S(y2) are close together on a classical scale, the phases
\S(y\) and jjSO^) will typically be enormously further apart. Thus exp { S ( y 1 ) } and
exp {^(yi)} will be points on the unit circle in the complex plane which need not be
close together. If we add all the contributions made to the integrand by the small set
of paths about y, then (roughly) the net contribution will be approximately 0, since, if
one path makes a certain contribution, another one close by on a classical scale will
make a canceling contribution. On the other hand, if y in Figure 7.2.2 is replaced by
the classical path x, changes in %S(x) are relatively small as x varies over the small set
about x. Hence the phases combine constructively, very little cancellation occurs, and
most of the contribution to the integral (7.2.1) is found here. As h -> 0, the paths that
contribute in a significant way to the integral tend to cluster in a smaller and smaller set
around the classical path x, and, in the limit, the "weight" is concentrated entirely on x.
This appealing heuristic discussion of the passage from quantum to classical mechanics
describes the method of stationary phase as applied to this particular infinite dimensional
setting. Rigorous treatments of the method of stationary phase in finite-dimensional
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
FIG. 7.2.2. Paths concentrated near the path y
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 106 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

situations can be found, for example, in [GuiSbg, Fel, H6r2, Hor3, §7.7, Hor4]. The
infinite dimensional case discussed above has been extremely difficult to prove and will
not be discussed rigorously in this book. Rigorous work on or closely related to this
important topic under rather strong assumptions can be found in the papers of Albeverio
and Hoegh-Krohn [AlHo3,4], Truman [Trul], Elworthy and Truman [ElTr], Rezende
[Rez], and Albeverio and Brzezniak [AlBr].
We wish to discuss in the next section the connection between the path integral
and the Schrodinger equation. However, this is perhaps an appropriate place to remark
briefly on how "the" Feynman integral is presently being used in a variety of places in
theoretical physics. What is needed is a set P of "virtual paths", an action functional 5,
and an appropriate "measure" V. The "formula" for the Feynman integral then looks
much like (7.2.1):

Examples of the above can be found in several places. The symmetries of particular
physical theories show up as symmetries of the action functional S and result in pleasant
formal transformation properties of the "integral" (7.2.6). The method of stationary
phase continues to play an important role in relating the "quantum" theory and the
corresponding "classical" theory. The interested reader with a limited background in
physics might begin with the expository book of Zee [Zee, §7 and pp. 142-144 of §10]
or the expository but much more advanced paper of Manin [ManiS, esp. pp. 63-65].
We note that various uses in mathematics and/or physics of heuristic Feynman inte-
grals of the type (7.2.6) (or of more general functional integrals) will be discussed in
Section 20.2.

7.3 Heuristic derivation of the Schrodinger equation

We wish to present Feynman's heuristic argument that the path integral provides a solu-
tion to Schrodinger's equation [Fey2, pp. 374-376]. We will actually work with formula
(7.2.3) where, following Feynman, we will see what choice to make for L towards the
end of the discussion. Feynman makes various interesting comments along the way but
we will not include all of these. We will, however, "see" in the course of the discussion
why the space of integration in (7.2.1) can be taken as C^v rather than R2',«-
First of all, recall that S(uj+\, Uj)is to be the action integral involved in going during
under U.S. or applicable copyright law.

the infinitesimal time interval € from (s, uJ) to (s + e, uj+i) along the classical path
under the influence of the potential V. Thus

where x is the classical path just mentioned. Feynman approximates S ( u j + 1 , Uj) as

follows:
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 HEURISTIC DERIVATION OF THE SCHRODINGER EQUATION 107
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In (7.3.2) and as we continue, we use equal signs for approximations as well as for true
equalities.
Now, substituting the right-hand side of (7.3.2) in (7.2.3), we obtain

Let us make the substitution f = «7 — Uj+\ and, for convenience, let u = Uj+\. Then
(7.3.3) becomes

[For ijs e L2(R), the right-hand side of (7.3.4) need not exist as a Lebesgue integral, but
sense can be made of it as we will have occasion to discuss later on.]

the function exp ( ^- J oscillates extremely rapidly around the unit circle in the complex
plane, causing almost complete cancellation. Thus the only substantial contribution to
the integral (7.3.4) comes from a small interval of f s around 0 satisfying, say, (7.3.5).
[It is interesting that on an interval of length €, values of § = Uj — MJ+I on the order
of */€ need to be taken into account (see (7.3.5)). Feynman comments on this explicitly
on page 376 of [Fey2] and mentions paths that are continuous but not differentiable
and Brownian motion. (We note that at least heuristically, (7.3.5) can be "justified"
more fully by applying the classical method of stationary phase for finite-dimensional
integrals; see [Fey2].)] Since only small £ contribute in a nontrivial way, tfr(s, u + e)
from the integrand in (7.3.4) can be expanded as a Taylor series in £ about 0. [Throughout
under U.S. or applicable copyright law.

this entire discussion we make no claim of mathematical rigor, but the assertion of the
existence of a Taylor series expansion when w need only be square-integrable particularly
goes against a mathematician's sensibilities.] Using this expansion to replace w (s, u + f)
in (7.3.4), we obtain

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 108 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Next we integrate out the first four terms on the right-hand side of (7.3.6) using the
following four definite integrals (which clearly do not exist as Lebesgue integrals):

[The first of these oscillatory integrals, a "Fresnel integral", exists as an improper Rie-
mann integral and can be computed by contour integration. (See Exercise 4.7.5, especially
formula (4.7.10), in Appendix 4.7 above.) The second and fourth can be interpreted as
principal value integrals. In view of (4.7.2) and Remark 4.7.4 in Appendix 4.7, the value
of the third can be obtained by replacing i by —8 + i, with & > 0, and formally passing
to the limit as S —> 0+. (In (7.3.7) and (7.3.9), we use the analytic determination of ^/z
which is positive for z > 0; so that, for example, i 1 / 2 = (l+i)/\/2.) We stress, however,
that none of the integrals in (7.3.8)-(7.3.10) exists as an improper Riemann integral.]
Thus (7.3.6) yields

where the terms omitted from the series in (7.3.11) are all of order 2 or higher in e.
Next we expand \j/(s + e, u) to the first order in e and set the result equal to the
right-hand side of (7.3.11), obtaining:
under U.S. or applicable copyright law.

In order that both sides of (7.3.12) agree to order zero in €, we must take

Now expanding the exponential containing V in (7.3.12) to the first order in e, we

obtain

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN'S APPROXIMATION FORMULA 109
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Multiplying out on the right-hand side of (7.3.14) but dropping the second order term
in e, we obtain

Canceling ijs(s, u) from both sides of (7.3.15), we get

and this is just Schrodinger's equation as stated previously in formula (6.1.3).

7.4 Feynman's approximation formula

As mentioned earlier, Feynman was aware, at least in part, of the difficulties associated
with formula (7.2.1). He tried to avoid the difficulties by replacing (7.2.1) by a limit of
finite-dimensional integrals, as we now explain.
Divide [0, t] into n equal parts each of length t/n. Consider the piecewise linear
functions x which have vertices only at the points corresponding to tj := j(t/n),
j — 1 , . . . , n — 1, and which satisfy x(0) = u and x(t) = v. Figure 7.4.1 shows a
typical path in the case n = 4. Note that the action S(x) given by (7.2.2) is defined for
such a path. Letting Xj = x ( j t / n ) , we obtain

where the last step in (7.4.1) is an approximation rather than an equality.

under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
FIG. 7.4.1. A polygonal path for n = 4 going from u to v in time t
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 110 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Feynman takes as his "approximation" to w, t; u, v):

where XQ = u and xn = v. (Here, we set V^7 = e -in/4 = (1 — i)/\/2.) He then

"defines" w(0, f, u, v) by

Next let us comment about (7.4.2) and (7.4.3) and their relationship with Feynman's
formula (7.2.1).
(i) Comparing (7.2.1) and (7.4.2) in light of the calculation (7.4.1) of S(x) for a
typical piecewise linear path x, we see where the integrand in (7.4.2) comes from.
(ii) The fictional measure D> from (7.2.1) has been replaced by a true translation
invariant measure (or Haar measure), namely, (n — 1)-dimensional Lebesgue measure,
in the approximation (7.4.2).
(iii) Formula (7.2.1) envisioned integrating over all paths in Cu,vo,t. Does that idea
remain in any sense in (7.4.2) and (7.4.3)? Yes. Note that we integrate over all possible
values of x1, . . . , xn-\ in (7.4.2) and then take finer and finer partitions of [0, t] in
(7.4.3). The collection of such polygonal paths is dense in C0,tu,v in the sense of uniform
convergence.
(iv) The constant K which was not identified in (7.2.1) has been replaced by a specific
constant in the formula for wn; this new constant is of the form 1/Ln, with L given by
(7.3.13) and e = t/n.
The situation is much better, but there are still serious difficulties with (7.4.2) and
(7.4.3):
under U.S. or applicable copyright law.

(a) If V is R-valued (as it is in the most standard problems), the integrand in (7.4.2)
has constant absolute value 1, and the integral does not exist as an absolutely convergent
Lebesgue integral.
(b) Even when the objection just raised can be satisfactorily resolved, it is not at all
clear when and in what sense the limit in (7.4.3) exists.
The hope is to take advantage of the highly oscillatory nature of the integrand in
(7.4.2) to make sense of the integral and then to investigate conditions under which
the limit in (7.4.3) exists in some reasonable way. Indeed, one approach or family of
approaches to the Feynman integral consists of trying to do this in a more or less straight-
EBSCO Publishing
forward manner. We : eBook
willCollection
discuss (EBSCOhost)
this somewhat- printed on 6/8/2017
further in the5:33
nextPMsection
via UNIVERSIDAD
and then with
DISTRITAL FRANCISCO JOSE DE CALDAS
moreAN:detail
98476 ;and greater
Johnson, precision
Gerald laterMichel
W., Lapidus, on, especially in Chapter
L..; The Feynman 11.Feynman's
Integral and
Operational Calculus
Account: ns000601
 NELSON'S APPROACH VIA THE TROTTER PRODUCT FORMULA 111
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Recalling from (6.2.1) that the probability amplitude TJr(t, v) is related to the initial
probability amplitude (j>(u) and the transition amplitude i/f(Q, t; u, v) by the formula

suggests how to alter Feynman's formula (7.4.2) to give an approximation ifn(t, v) for
if(t,v):

where xn = v as before. Of course, we then take

It is formulas (7.4.4) and (7.4.5) that we will actually work with in the next section.
We have heard it said that the approach represented by (7.4.2) and (7.4.3) (alterna-
tively, by (7.4.4) and (7.4.5)) is what Feynman had in mind and so what mathematicians
should concentrate on. It is interesting that this is strikingly different from the point of
view expressed by Feynman [Fey2, p. 384 and p. 371]: "At present, it [i.e. essentially
(7.4.4) and (7.4.5)] requires an unnatural and cumbersome subdivision of the time inter-
val to make the meaning of the equations clear.... One needs, in addition, an appropriate
measure for the space of the argument functions x(t) of the functionals." Also, "These
curious mathematical problems are sidestepped by the subdivision process. However,
one feels as Cavalieri must have felt calculating the volume of a pyramid before the
invention of calculus" In fact, the approach to be discussed in the next section via (7.4.4)
and (7.4.5) and the "Trotter product formula" is one of the best known and, in some
ways, most appealing approaches to the Feynman integral. However, we see from the
quotes above that it was certainly not the unique approach envisioned by Feynman.

7.5 Nelson's approach via the Trotter product formula

under U.S. or applicable copyright law.

In this section and the next, we will show how certain approaches to the mathematical
theory of the Feynman integral have come rather naturally from Feynman's heuristic
formulations. Our present goal is just to see how these approaches arise; further on, we
will pursue them. We will proceed in this section by (i) stating a sample theorem, (ii)
pointing out its connection with Feynman's approximation formula, and (iii) discussing
briefly the Trotter product formula and its relevance to the theorem.
For simplicity, we restrict our attention in this chapter to the Trotter product formula
(for unitary groups) but later on, we will also discuss more general results using a
"product formula for imaginary resolvents", obtained by Lapidus in [Lal 1], and use it to
prove the existence of the "modified Feynman integral" (also introduced by the second
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
author [Lal,2,
DISTRITAL 6, 7],JOSE
FRANCISCO andDEfurther
CALDAS developed in [La8-13, BivLa]) under significantly more
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 112 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

general conditions than can be done by using the Trotter product formula. (See especially
Sections 11.3 and 11.4, as well as Sections 11.6 and 13.6; also see Remark 7.5.3(c) below.)
The best known and, in some respects, the nicest way of making precise mathematics
out of Feynman's approximation approach is by means of the Trotter product formula.
This was the first of two approaches used by Edward Nelson in his beautiful 1964 paper
[Nel], Feynman integrals and the Schrodinger equation. Nelson's first approach—to be
treated at greater length in Chapter 11 (Section 11.2)—depended not only on Trotter's
formula, which we will discuss briefly near the end of this section and in more detail
in Section 11.1, but also on a pivotal 1951 result of Kato [Katl] on the essential self-
adjointness of the energy operator (or "Hamiltonian") H, where H is defined by

with A denoting the Laplacian acting on functions on Rd. Here, H is the extension
to d dimensions of the operator appearing in (6.1.4). Alternatively, H agrees with the
right-hand side of (6.4.2) if d = 3N and m \ = m2 = • • • = mN. The energy operator
has been intensively studied [ReSi2] and, in particular, is now known to be essentially
self-adjoint under much more general conditions on the potential V than were known
at the time of Nelson's paper. (See, in particular, [Kat2], to be discussed towards the
beginning of Section 11.2.) One obtains the theorem which we will soon state by using
Nelson's ideas, one of the new results insuring essential self-adjointness of H, and the
Trotter product formula.
The difficulties that arise in Feynman's approximation formula (7.4.4) because the
integrand is not Lebesgue integrable are taken care of in large part by using "integrals in
the mean" (just as in the theory of the L2 Fourier-Plancherel transform). Let us review
the concept of "integral in the mean" in the setting in which we will use it here and
further on.
Let g be a C-valued function in L2(Rd). Let k : Rd x Rd -> C be such that, for every
M > 0, (i) *(-, |) is in Ll([-M, M]d) for almost every £, and (ii) f{_M M]d k(w, $)dw
is in L 2 (R d ) as a function of f. We say that g is the integral in the mean of k and, abusing
notation, write simply,
under U.S. or applicable copyright law.

provided that, as M — oo,

Recall that a measurable function f : Rd -> R (or C) is said to be in L2oc(Rd)

provided that it is square-integrable over every ball in Rd of finite radius.
Given V : Rd -» R, we will write V = V+ — V- for the usual decomposition of
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
V into its positive and negative parts. Recall that V+(u) — max(V(u), 0) and V-(«) =
DISTRITAL FRANCISCO JOSE DE CALDASd
max(- V(u), 0). Finally,
AN: 98476 ; Johnson, GeraldLp
W.,( RLapidus,
) + L°° ( R d )L..;
Michel will
Thedenote
Feynman the vector
Integral and space of all C-valued
Feynman's
Operational Calculus
Account: ns000601
 NELSON'S APPROACH VIA THE TROTTER PRODUCT FORMULA 113
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

functions on Rd which can be written as the sum of an Lp-function and an essentially

bounded function.
The theorem which we are about to give will be stated precisely but there is further
related information whose discussion we postpone until Section 11.2.
Theorem 7.5.1 Let V : Rd -> R and suppose that V+ is in L2loc (Rd) and V- is in
L P ( R d ) + L°°(R d ), where the restriction on p is dimension dependent as follows:
p = 2 if d < 3, p>2ifd = 4, and p > d/2 i fd > 5. Let 0 € L 2 (R d ).
Then, for every t > 0, the integrals on the right-hand side of the following formula
exist for almost every v € Rd when interpreted in the mean and define a function which
is in Lw(Rd) as a function of v:

where xn := v. Further, there is a function w : (0, +00) x R -> C such that w(t, •) e
L 2 ( R d ) f o r every t > 0 and, as n —>• oo,

Formula (7.5.4) differs from Feynman's formula (7.4.4) only in that (7.4.4) was, for
convenience, restricted to the case d = 1, and (7.5.4) is written as an n-fold iterated
integral instead of as a multiple integral.
Note that, in contrast to (7.4.5), the sense in which wn converges to w is clearly
under U.S. or applicable copyright law.

identified in (7.5.5). Further, the mode of convergence seems natural in light of the
fact that probabilities are, in the last analysis, computed using the probability densities
1W (t, .)|2.
Although we do not wish to discuss the matter in detail here, we mention that the
hypotheses of Theorem 7.5.1 are already sufficiently general to cover many of the poten-
tials of interest in atomic physics (such as harmonic oscillators and attractive or repulsive
Coulomb potentials). The positive part of the potential V+ can have an arbitrary singu-
larity at oo as well as finite range singularities as long as they are square-integrable. The
assumptions on V- are more restrictive; V-(x) —> oo as ||jc|| -» oo is not permitted
andEBSCO
finitePublishing
range singularities must be pth power integrable, with p as in the statement
: eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
of Theorem 7.5.1. JOSE DE CALDAS
DISTRITAL FRANCISCO
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 114 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The Trotter product formula

What is the Trotter product formula and how is it relevant to Theorem 7.5.1? Trotter's
original result [Tro2] appeared in 1959 and has since been extended in various directions.
(See, for example, [BergSlo, BivLa, BivPi, Brel, BrePa, Cherl,2, DoFril-3, ExK, Farl,
Fri, Gol, Kat4,5, Kat7, KatMasu, Kur, Lal-11, Lal3, Mars, NdZag, Nel, ReSil, Schl,2,
Sem, Sve, Zag].) We state a special case of Trotter's original result [Tro2] (relying on
[Tro 1 ]) which is sufficient for our present discussion. The operator theory involved in the
statement of the theorem will be discussed further on in Chapters 8-10, while Trotter's
formula and its application to the Feynman integral will be treated in more detail in
Chapter 11 (Sections 11.1 and 11.2). The reader who is not familiar with this type of
operator theory should try at present just to get some idea of the nature of Trotter's
formula.
Theorem 7.5.2 If A and B are (not necessarily bounded and not necessarily commuting)
self-adjoint operators on a Hilbert space H. and if A + B is essentially self-adjoint (i.e.
it has a unique self-adjoint extension, necessarily its closure A + B) on the intersection
of the domain of A with the domain of B, then for all u e Ti,

uniformly in t on all bounded subsets of R.

The Trotter product formula for unitary groups (7.5.6) may appear strange if you
are not already somewhat familiar with it or related formulas. It expresses the operator
e-it(A+B)^ , which looks rather simple, as the limit of expressions which appear more
complicated. In fact, the left-hand side of (7.5.6), the exponential of the sum of (usually)
noncommuting operators, appears simpler than it is. In practice, explicit and tractable
analytic expressions may be known for

indeed, this is the case in the situation which concerns us.

To make the connection between Theorems 7.5.1 and 7.5.2, take H = L2(Ed), A =
-2^A, andB = ^V;then
under U.S. or applicable copyright law.

where H is the energy operator (also called the Schrodinger operator or Hamiltonian)
given by (7.5.1). Now the exponential of the operator of multiplication by — ^ V is the
operator of multiplication by

ThisEBSCO Publishing
is the : eBook Collection
first operator that acts(EBSCOhost) - printed
on the initial stateon06/8/2017
on the5:33 PM via UNIVERSIDAD
right-hand side of (7.5.4).
DISTRITAL FRANCISCO JOSE DE CALDAS
TheAN:next operator to act in (7.5.4) is the convolution-type integral operator associated
98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE APPROACH VIA ANALYTIC CONTINUATION 115
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

with integration with respect to Xo. It is well-known ([ReSi2, p. 59] or Theorem 10.2.7
•L h A
below) that this is the operator el« 2m , which is the operator corresponding to the free
evolution of the quantum system over a time interval of length t/n. A careful look at the
rest of the right-hand side of (7.5.4) shows that it is the n-fold iteration of the product
•iJ|-A _LLy
el n 2m e hn , just as it should be to fit the expression on the right-hand side of (7.5.6).
(See Section 11.2 for further related information.)
Remark 7.5.3 (a) Formula (7.5.6) expresses e~i!(A+B) (n terms ofe-1-1/nAande-il/Aand e-i1/B
If A and B commute, the relationship is much simpler; the familiar exponential law,

holds. Formula (7.5.6) (or some variation of it) is sometimes referred to as the non-
commutative exponential law. (For simplicity, we ignore in this comment the distinction
between A + B and its closure A + B.)
(b) Even in rather elementary mathematics, for example, in solving a system of lin-
ear ordinary differential equations with constant coefficients, one sees the advantage
of forming functions of operators. The problem of forming functions of two or more
noncommuting operators will eventually be a major concern of ours, most explicitly in
Chapters 14 and 19, but also in Chapters 15-18. Formula (7.5.6) deals with exponen-
tating the sum of noncommuting operators and is related to this subject.
(c) There are alternatives to the right-hand side of (7.5.6). One of these is intimately
connected with "the modified Feynman integral" of Lapidus [Lal,2, La6-13], a topic
which we will also discuss in some detail in Chapter 11 (Sections 11.3 and 11.4). One
advantage of the latter approach to the Feynman integral—based on a "product formula
for imaginary resolvents" [La2, La6, Lall] (Section 11.3) rather than for unitary groups
as above—is that it will enable us to deal with very general potentials (Section 11.4) and
that it possesses very pleasant stability properties [Lal2] (Section 11.5). [See also the
results from [BivLa] (discussed in Section 11.6 and, in a related setting, in Section 13.6)
which allow us, in particular, to extend its definition to singular complex (rather than
real) potentials.] The Trotter product formula and relatives of it will appear in various
ways as we continue.
(d) One can show, under the hypotheses of Theorem 7.5.1, that the function w ( t , v)
whose existence is established by use of Theorem 7.5.2 satisfies the Schrodinger equation
under U.S. or applicable copyright law.

in an appropriate sense.

7.6 The approach via analytic continuation

During his lecture Feynman sketched the derivation of his formula and I was struck by the similarity
of his steps to those I had encountered in my work. In a few days I had my version of the formula,
although it took some time to complete a rigorous proof.
Mark Kac, 1984 [Kac5, p. 116]

Mark Kac attended a physics colloquium at Cornell in 1947 at which Feynman

spoke about his path space integral. Kac realized that there is a connection between
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
the DISTRITAL
elusive FRANCISCO
Feynman JOSEintegral
DE CALDASand the well-defined and analytically powerful Wiener
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 116 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

integral. We will describe Kac's insight in this section and indicate briefly some of
the mathematics to which it has led, especially approaches to the Feynman integral
via analytic continuation from the Wiener integral. We will pursue this last topic in
Chapters 13 and Chapters 15-18.
We have already seen the connection between Feynman's ideas and formula (7.5.4)
and so we will start from there. Kac saw that if the time parameter t in (7.5.4) is replaced
by —it ("imaginary time" from the point of view of quantum physics), then (7.5.4) can
be written as a Wiener integral and the Lebesgue theory can be used to deal with the
limit as N -> oo. Kac's original result [Kac 1,2], the "Feynman-Kac Formula", will
be discussed in Chapter 12 in a very general form; its use for obtaining the Feynman
integral via analytic continuation in time will be one of the main subjects of Chapter 13.
(See Sections 13.2-13.3 and 13.7.) Alternatively, one can complexify the mass rather
than the time, and it is this approach which we will discuss in most detail below and
which will be used in Section 14.5 and in Chapters 15-18. (See also Sections 13.5 and
13.6 for somewhat different approaches.)
We return to the case d = 1, for convenience, and we replace m in (7.5.4) with
i Kk, where A. > 0. The revised formula, which is no longer actually an approximating
formula for the quantum-mechanical probability amplitude, becomes
under U.S. or applicable copyright law.

where xn := v.
A close look at the right-hand side of (7.6.1) especially in the case A. = 1 indicates
that it is the type of expression which arose as the Wiener integral of a finitely based
function on Wiener space. In fact, if we let x e C0'' = C'0, v e R, and apply the Wiener
integration formula (3.3.9) to the (finitely based) function G defined on x + v e Ct by

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE APPROACH VIA ANALYTIC CONTINUATION 117
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

we obtain

where u-\ := 0 and m denotes Wiener measure defined in Chapter 3.

Letting WQ = un-1, w1 = un-2,..., Wn-2 = u1, wn-1 = u0, we see that

where wn := 0.
Next letting XQ = W0 + v, x\ = w\ + v,..., x.n-i = wn-2 + v, xn-\ = wn-\ + v,
we obtain
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
where now FRANCISCO
DISTRITAL xn := v.JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 118 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Comparing (7.6.5) with (7.6.1), we see that

Proceeding as above in (7.6.2) through (7.6.6) but with x replaced by A. l/2x (the
Wiener process with variance parameter A - 1 ) , we obtain

that is,

The argument of the exponential in the third expression in (7.6.8) is, under reasonable
conditions, a Riemann sum approximation for the integral

In light of this and in light of the presence of the countably additive measure m in (7.6.8),
under U.S. or applicable copyright law.

it seems plausible that

The fact that the limit in (7.6.9) exists and equals the Wiener integral appearing there can
be justified under even more general conditions than the preceding discussion might lead
one to think. This is, at least in part, because one only needs to show that the integrand of
the EBSCO
WienerPublishing : eBook
integral Collection
in (7.6.8) (EBSCOhost)
converges - printed
to the on 6/8/2017
integrand of the5:33 PM via integral
Wiener UNIVERSIDAD
in (7.6.9)
DISTRITAL FRANCISCO JOSE DE CALDAS
for AN:
m x98476
Leb.-almost everyW.,(x,Lapidus,
; Johnson, Gerald v) in C' Q x R.
Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE APPROACH VIA ANALYTIC CONTINUATION 119
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 7.6.1 We have discussed the connection between Feynman's approximation

formula and the Wiener integral in (7.6.9) in detail primarily because we regard it as
a crucial insight but also because it is a bit tricky. Note in particular that $ appears in
the Wiener integral in (7.6.9) associated with the final time t. This reverses the physical
order since <f>, as the initial probability amplitude, is naturally associated with time 0,
The reversal, from the point of view of quantum mechanics, of the time-ordering by the
Wiener process will turn up in other places as we continue. (See especially Chapters 15
and 17.)
The insights of Kac have led to an enormous amount of work in various directions.
If the mass in the Schrodinger equation is replaced by z'fiA., one obtains a form of the
heat equation. This along with (7.6.3) and the work of Feynman relating the Feynman
integral to the solution of the Schrodinger equation (see Section 7.3, especially formula
(7.3.16)) suggested to Kac [Kacl,2] that one should be able to express the solution to
the heat equation as a Wiener integral. Let ^ (l> v) b£ tne solution to the heat equation
with initial data </> corresponding to a particular choice of the positive number A.. One
obtains, under rather general conditions, the formula

which is a version of the Feynman-Kac formula. Note that the imaginary unit i appears
on the right-hand side of (7.6.10). This has happened because we complexified the mass
rather than the time. In order to obtain an integrand in (7.6.10) suitable for a heat flow
problem, one must then take V purely imaginary.
The work of Kac expressing the solution of the heat equation as a Wiener integral led
to much further work on the same topic and to applications in statistical mechanics and
quantum statistical physics as well as to results expressing the solution of other parabolic
partial differential equations as functional integrals with respect to probability measures
associated with corresponding Markov processes. See, for example, the books of Bratteli
and Robinson [BraRo], Freidlin [Frei], Glimm and Jaffe [GliJa], Simon [Si9], as well as
of Varadhan [Va] and the expository article of Orey [Or].
When we complexify the time (or the mass), we remove ourselves from the quantum-
under U.S. or applicable copyright law.

mechanical setting. However, various questions central to quantum theory, notably spec-
tral questions, have been effectively dealt with using imaginary time techniques and the
Wiener integral. Much of this work is described in the books of Simon [Si9] and Glimm
and Jaffe [GliJa].
Most importantly, from the point of view of our work, Kac's paper suggested another
approach to the Feynman integral. Begin with the well-defined and analytically tractable
Wiener integral on the right-hand side of (7.6.9). Under general conditions on V and
with 0 in L2(R), this makes sense for all A > 0. One then attempts to pass to A. purely
imaginary, the quantum-mechanical case, by first analytically continuing to the right half-
plane and then taking boundary values as A. approaches the imaginary axis. There are a
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
variety of ways
DISTRITAL to analytically
FRANCISCO continue. For example, one may regard the right-hand side
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 120 INTRODUCTION TO THE FEYNMAN INTEGRAL
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

of (7.6.9) as a number and analytically continue in a scalar-valued sense. Instead, as we

will do further on, one can analytically continue in an operator-valued sense, regarding
the Wiener integral in (7.6.9) as a bounded operator acting on functions <j> in L2(R). We
will also discuss later operator-valued analytic continuation in time (see Sections 13.2-
13.3 and 13.7). Since functionals F on C' other than the standard quantum-mechanical
functional £,

are of interest (see, for example, [CaStl,3], [JoSk2,6], [JoLal,4], [Lal5,18]), we will
usually start our discussions with the Wiener integral

rather than with the special case F = E in formulas (7.6.9) and (7.6.11). Various
examples of nonlinear functionals will be studied in Chapters 15-18.
We have been discussing the influence of Feynman's path integral and the insights
of Kac on mathematics and mathematical physics. Feynman's path integral also had a
major impact on theoretical physics. Between 1948 and 1951, Feynman's work on the
path integral (1948), quantum electrodynamics including the famous Feynman diagrams
(1949 and 1950), and on the operator calculus for noncommuting operators (1951), all
appeared. As Silvan Schweber says in his paper "Feynman and the visualization of
space-time processes" [Schwe2, p. 504], "These papers, on the space-time approach to
nonrelativistic quantum mechanics, on quantum electrodynamics, and on his operator
calculus, must surely be placed near the top of any list of the most seminal and influential
papers in theoretical physics during the twentieth century." Although the work on quan-
tum electrodynamics and the operator calculus did not formally depend on Feynman's
path integral, it did depend heavily on the manner of thinking established there. Quoting
Schweber again [Schwe2, p. 502], "In any case, Feynman never felt order-by-order was
anything but an approximation to the 'thing' and the 'thing' was the path integral." An
enormous amount of later work in theoretical physics has been heavily influenced in turn
by these papers of Feynman.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

8
SEMIGROUPS OF OPERATORS:
AN INFORMAL INTRODUCTION

Next we wish to provide in Chapters 8-10 some background in operator theory. We will
begin by discussing "strongly continuous one-parameter semigroups of bounded linear
operators on a Banach space X"; that is, "(Co) semigroups". We will not need this level
of generality for some time, but it will eventually be useful to us. Much of what we
will have to say would need to be said even if we restricted the level of generality. Our
discussion of (Co) semigroups is based largely in Chapter 8 and parts of Chapter 9 on
Jerome Goldstein's nice book [Gol], Semigroups of Linear Operators and Applications.
We will start in the present chapter with an informal discussion of semigroups of
operators and then turn to a more systematic presentation in Chapter 9. Let X be a
Banach space and let A be a linear map from its domain D(A), a subspace of X, to X,
We indicate some of this briefly with the notation, A : D(A) c X -> X. Consider the
evolution equation along with initial data,

The statement that j-,u(t) = Au(t) means that u(t) € D(A) for every t > 0 and

Ignoring all difficulties for now and solving (8.1.1) purely formally, one obtains

The formula (8.1.3) cannot be taken entirely seriously, but it does suggest some intuition
under U.S. or applicable copyright law.

about solutions to (8.1.1) which is correct under assumptions to be specified later.

The setting of (8.1.1) can be further generalized. The space X could be a linear
topological space or a differentiable manifold; further, certain kinds of nonlinear opera-
tors A can be dealt with. However, even restricting attention to the framework above, a
surprising variety of problems can be cast in the form (8.1.1). We give some examples
beginning with the two which will be of most interest to us. At least for now, we choose
units to make the physical constants disappear from sight.
Example 8.1.1 (The heat equation) f£(f,Jt) = (A0)(r,jc) - V(x)<t>(t,x),(/>(Q,x) =
</>o(x), t >0,x eK d .
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
In this case,
DISTRITAL weJOSE
FRANCISCO take
DE A = A - V, u(t) = <f>(t, •), w(0) = </>(0, •) = <fo, with V
CALDAS
AN: 98476 R-valued
a suitable ; Johnson, Gerald
function on Kd, Michel
W., Lapidus, L..; The
and there areFeynman Integral
various and Feynman's
possibilities for the Banach
Operational Calculus
Account: ns000601
 122 INFORMAL INTRODUCTION TO SEMIGROUPS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

space X such as L 2 (Rd); LP(Rd) (1 < p < oo); or BUC(R d ) (the space of bounded,
uniformly continuous functions on Rd). With these choices, we can at least formally
identify the heat equation with the evolution equation (8.1.1). The solution to the heat
equation according to (8.1.3) is then u(t) - e t ( A - V ) </>o = e~ f( ~ A+v Vo, or

where H = — A + V is the energy operator or "Hamiltonian".

Example 8.1.2 (The Schrodinger equation) ^-(t, x) = i[(&i/r)(t, x) - V(x)\/r(t, *)],
tfr(0, x) = $o(x), t > 0, x € Rd.
Here we take A = i(A - V), u(t) = ^(t, •), w(0) = Vo and X = L 2 (R d ). The
Schrodinger equation can now be "identified" with (8.1.1), and the solution to the equa-
tion following (8.1.3) is u(t) = e"<A-vVo or

where H = - A + V as before.
In contrast with Example 8.1.1, we may allow t e R here, as will be seen later on.
Note that in Examples 8.1.1 and 8.1.2, higher-dimensional problems have been
rewritten as formally simpler one-dimensional problems. This is a recurring theme in
the applications of semigroups of operators.
We give a rather brief treatment of two further examples in order to help the reader get
some appreciation for the range of problems that can be regarded as evolution equations.
Example 8.1.3 (Mixed initial-boundary value problem) Let Q be a bounded domain in
Rd with a "nice" boundary 3£2. A first glance at the evolution equation (8.1.1) might
lead one to think that its applications are restricted to initial-value (or Cauchy) problems.
However, the present example concerns a mixed initial-boundary value problem for the
heat equation. We seek a function w = w(t, x) defined on [0, +00) x & such that
under U.S. or applicable copyright law.

(We also need f ( x ) = 0 for x 6 3 £2 for consistency.)

We take u(t) = w(t,-) much as before and also let X = C(?2). (X = Lp(£l), p > 1,
is also a possibility.) Let A = A and take

D(A):= {v e C(Q): v is twice differentiable, Av 6 C(£2),

and v(x) — 0 for each x € 3fi}.
This set-up puts the problem in the form (8.1.1). Note that the boundary condition
EBSCO Publishing
is absorbed : eBook
into the Collection
domain (EBSCOhost) -ofprinted
of definition on 6/8/2017A5:33
the operator andPMthe
via requirement
UNIVERSIDAD that
DISTRITAL FRANCISCO JOSE DE CALDAS
u(t) AN:e 98476
D(A); Johnson, > 0.W., Lapidus, Michel L..; The Feynman Integral and Feynman's
for all tGerald
Operational Calculus
Account: ns000601
 INFORMAL INTRODUCTION TO SEMIGROUPS 123
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 8.1.1. The domain fil and its boundary 9 & for d = 2

Example 8.1.4 (The wave equation) The evolution equation (8.1.1) appears to be limited
to differential equations which are first order in time; but, in fact, that is not so. This
example involves a wave equation which is second order in time:

The space X will be taken as some space of pairs of functions on Rd. We set

With this choice,

Hence (8.1.1) becomes

under U.S. or applicable copyright law.

and this is just a restatement of the initial-value problem for the wave equation given
above.
Further examples can be found in [Gol]. When the initial-value problem (8.1.1),

is used to model a physical problem, one would like this mathematical model to have
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
the DISTRITAL
following basic JOSE
FRANCISCO properties:
DE CALDAS (i) It has a solution, (ii) The solution is unique, (iii) The
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 124 INFORMAL INTRODUCTION TO SEMIGROUPS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

solution depends continuously on the initial data / and on the operator A. When (i)-(iii)
are satisfied, the problem is said to be well posed. (Note that, consistent with the informal
nature of this chapter, our "definition" of "well posed" is not entirely precise.)
Given a Banach space X, £(X) will denote the space of bounded linear operators
on X. Motivated in part by the intuitive formula (8.1.3), u(t) = etA f, for the solution
of (8.1.1) and writing T(t) instead of e'A (t > 0), we make the following definitions.
We give precise definitions here since we are in a position to do so and it will eliminate
repetition further on.
Definition 8.1.5 A family of operators [T(t): 0 < t < +00} in C(X) is called a strongly
continuous semigroup of operators or, briefly, a (Co) semigroup on X if and only if
(i) T(t + s) = T ( t ) T ( s ) for each t,s£ [0, +00),
(ii) T(0) = /,
(iii) T(-)f e C([0, +00), X) for each f € X.
Here, / is the identity operator on X and C([0, +00), X) denotes the space of con-
tinuous functions from [0, +00) to X. Further, (iii) means that for each fixed / e X, the
map t )-» T(t)/ is continuous (in the norm of X) from [0, +00) to X. This is called the
strong (operator) continuity of the semigroup (T(t)}.

Definition 8.1.6 A (Co) contraction semigroup on X is a (Co) semigroup [T(t)} on X

such that \\T(t)\\ < I for every t e [0, +00).

Definition 8.1.7 Let T = (T(t)} be a (Co) semigroup on X. The (infinitesimal) gener-

ator A of T is defined by the formula

where the domain D(A) is the set of all f e X such that the limit defined above exists
(in X).
Note that in (8.1.6) [or (8.1.2)], one should really take the limit as t -> 0+. In the
sequel, it will be clear from the context when we are making this abuse of notation.
We return now to our more informal discussion.
under U.S. or applicable copyright law.

Theorem 8.1.8 (Well posedness theorem) The initial-value problem (8.1.1) is "well
posed" if and only if A is the generator of a (Co) semigroup [T(t)}. In this case, the
unique solution of (8.1.1) is given by u(t) = T(t) f for all f e D(A).
In light of Theorem 8.1.8, it is natural to ask which operators A generate (Co)
semigroups. In order to simplify the discussion somewhat at this point, we ask: Which
operators A generate (Co) contraction semigroups?
Given a number a 6 C, consider the Laplace transform formula

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 INFORMAL INTRODUCTION TO SEMIGROUPS 125
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

which is valid for all A e R such that A. > Re a. This formula suggests the operator
equation

which turns out to be valid for all X > 0 and all / e X.

Remark 8.1.9 /?(A.) = #(A.; A) := (A./ - A)"1 is called the resolvent operator of A.
Equation (8.1.7) tells us that /?(A; A) is the Laplace transform of the semigroup {T(t)}.
Given / e X and A > 0, we can, at least formally, do the following calculation:

This suggests that for every A. > 0, (A. — A) ' := (A./ — A) Ms an everywhere defined
bounded operator on X with operator norm less than or equal to £; hence

for every A. > 0, A,/ — A maps D(A) onto X and, for all / € X,

Theorem 8.1.10 (Hille—Yosida generation theorem) A linear operator A : D(A) c

X -*• X generates a (Co) contraction semigroup if and only if D(A) is dense in X and
property (8.1.8) holds.
(Theorem 8.1.10 will be stated with more detail and precision in Chapter 9; see
Section 9.4.)
There are a variety of theorems which say that if an operator A generates a (Co) semi-
group and if B satisfies certain conditions, then A + B generates a (Co) semigroup. Since
it is not always easy to tell if an operator generates a (Co) semigroup, such perturbation
results are of considerable importance. We state one such theorem which is relatively
simple. We can state it precisely at this point and so we will not need to repeat it further
under U.S. or applicable copyright law.

on. [Here and thereafter, we denote by C(X) the space of bounded (i.e. continuous)
operators from X to itself.]
Theorem 8.1.11 (A perturbation theorem) If a linear operator A : D(A) C X - X
generates a (Co) semigroup on the Banach space X and if B e C(X), then A + B also
generates a (Co) semigroup on X.
The next result, which will be discussed more fully further on, assures us that the
solution of the evolution equation (8.1.1) depends continuously on the generator.
Theorem 8.1.12 (Approximation theorem; Trotter-Kato-Neveu-Chernoff) For n =
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
0, 1DISTRITAL
, 2 , . . . , FRANCISCO
let An generate a (Co) contraction semigroup Tn on the Banach space X.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 126 INFORMAL INTRODUCTION TO SEMIGROUPS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then the following are equivalent:

(i) lim Tn(t)f = T0(t)f for every f e X and t > 0.
n-»oo
(ii) lim (k - An)-1 f = (A. - A0)-1 f
for every f eX and A. > 0 .
n—»oo
Moreover, a sufficient condition for (i) to hold is that for every n > 1,

£>(Ao) £ D(A n ) and lim An/ = A0/ far every f e D(A0).

n—oo

Next we give an example (neither detailed nor precise) which we hope will give the
reader at least an idea why perturbation and approximation theorems are important.
Let A be a second order partial differential operator of the form

where <t> is taken from some Banach space of functions on Rd, and suppose it is known
that A generates a (Co) semigroup. Let B be the first order partial differential operator

Under certain conditions, a variation of Theorem 8.1.11 can be used to show that A + B
generates a (Co) semigroup. Also, under appropriate conditions, the unique solution u
of the evolution equation ^ = (A + B)u, u(0) = f, can be shown by Theorem 8.1.12
to depend continuously on the coefficient functions av, aaft (0< y < d, 1 < a, ft < d).
We close this introductory chapter by simply remarking that there is an intimate
relationship between the theory of semigroups of operators and the theory of Markov
processes. This connection is sketched in Goldstein's book [Gol] and is developed in
detail in a variety of places; see, for example, [BluG, Frei, Fuk, FukOT, L2, Va].
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

9
LINEAR SEMIGROUPS OF OPERATORS

9.1 Infinitesimal generator

We turn now to a more consistently precise discussion of (Co) semigroups on X. Given
such a semigroup { 7 ( t } := {T (t): t : t > 0), recall from Chapter 8 that the generator
A : D(A) c X -» X is defined by

where D(A) is defined as the set of all / 6 X for which the limit exists in the norm on
X, Our first proposition is trivial.
Proposition 9.1.1 Given a (Co) semigroup {7X0} °n X with associated generator A,
D(A) is a linear subspace of X and A is a linear map from D(A) to X.
We will want to be able to integrate functions h : [ a , b]-+ X. Sometimes a Riemann-
type integral will be good enough. Such an integral is easily defined and has many of the
properties familiar from ordinary Riemann integration. The integral fa h(s)ds is defined
by taking the limit in the Banach space X of the usual Riemann sums, J^-i h(sj)(sj -
Sj- i), as the norm of the partition goes to 0. We list some of the basic properties:
(i) The class ft = R([a, b], X) of Riemann integrable functions is a vector space
and the Riemann integral acts linearly on ft.
(ii) For h g ft, ||/bah(s)ds||</s|| <! b a l\h(s)\\ds.
(iii) If/z <= C([a, b], X), then h 6 ft.
(iv) f* [ajACO] ds = h(b) -h(a) for h e C1 ([a, b], X), the space of functions with
under U.S. or applicable copyright law.

one continuous derivative on [a,b\.

(v) If I e £(X) and if h e ft, then L o A e ft and L(f* h(s)ds) = jba L(h(s))ds.
Finally, we mention that the improper Riemann integral is defined just as it is
for R-valued functions except that the limit is in the Banach space X. For example,
if h is Riemann integrable on every bounded interval [0, b], then /0+0° h(s)ds :=
lim /Q h(s)ds whenever the limit exists.
b—*+oo
There are now many books treating the theory of semigroups of operators; see, for
example, [Brel, DautLio, Dav, DunScl,2, Gol, HilPh, Kat8, KuSe, Paz, ReSil, vC, Yo].
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
OurDISTRITAL
discussion of (Co)
FRANCISCO JOSE semigroups
DE CALDAS will include very few proofs; however, we develop the
firstAN:important issue rigorously
98476 ; Johnson, in Proposition
Gerald W., Lapidus, Michel L..;9.1.2 and Theorems
The Feynman Integral and9.1.3 and 9.1.4 below.
Feynman's
Operational Calculus
Account: ns000601
 128 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proposition 9.1.2 Let [ T ( t ) } b e a (Co) semigroup on X with generator A. Then D(A)

is dense in X and is invariant under T(t); that is, for every t > 0,

Further, A commutes with T(t) in the sense that for every f e D(A) and t > 0,

Proof Let / e X and set ft = /0' T ( s ) f d s . The definition of // makes sense, since
T(-)f e C([0, +00), X) and so is locally Riemann integrable. (Recall from Chapter 8
that C([0, +00), X) denotes the space of continuous functions from [0, +00) to X.)
Using (v) (from the basic properties of the Riemann integral), the semigroup property,
and an easily justified change of variables, we can write for any h > 0,

Using (9.1.1), it is not hard to see that

From the definition of A and D(A), it now follows that ft E (A) and
under U.S. or applicable copyright law.

Now \ft = \ /0' T(s)fds -» T (0) f = f as t -> 0. Thus D(A) is dense in X.

Next suppose that / e D(A) and t > 0. Then, since all the elements of the semigroup
commute with one another, we can write

It follows that T (t)f

EBSCO Publishing E D(A)
: eBook as we
Collection wished -toprinted
(EBSCOhost) show,onand, in addition,
6/8/2017 AT(t)f
5:33 PM via = T(t)Af
UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
as claimed.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
D
Operational Calculus
Account: ns000601
 INFINITESIMAL GENERATOR 129
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Integral equation
Next we show that T(t)f satisfies a certain integral equation for f e D(A). The integral
equation (9.1.3) in Theorem 9.1.3 below is closely related to the evolution equation
du/dt = Au in Theorem 9.1.4, as we will see in the proof of Theorem 9.1.4. We will
sometimes refer to the integral equation itself as the evolution equation or as the integrated
form of the evolution equation.
Theorem 9.1.3 Let {T(t)} be a (Co) semigroup on X with generator A. Then for f €
D(A) and any t > 0,

Proof Let / E D(A) and set

We will show first that

Now
under U.S. or applicable copyright law.

Using (9.1.6) and the fact that / e D(A) implies that T(t)f e D(A) (Proposition
9.1.2), we see that

where the last equality in (9.1.7) follows from the commutativity established in
Proposition 9.1.2. The assertion in (9.1.5) now follows.
+
SinceG(t+h)-G(t)/h->o in norm->as h -> 0 , it follows that x* [1/h(G(t+h)-G(t))] -*
EBSCO Publishing
+ : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
0 asDISTRITAL
h -» 0FRANCISCO
for everyJOSEx*
DE eCALDAS
X* (where X* is the dual of the Banach space X). Thus we see
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 130 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

that for every x * e X*, the scalar-valued function x * oG has right-hand derivative equal to
0 forevery t > 0. Since also (x* oG)(0) = x*[7'(0)/'-/'-/0° T(s)Afds] = Jt*(0) = 0,
it follows from a refinement of the mean value theorem [Choi, Theorem 15.4, p. 155]
that x* o G = 0 on [0, +00). But, by the Hahn-Banach theorem [Ru2, Theorems 5.16
and 5.20, pp. 104 and 107], it then follows that G = 0 on [0, +00). Hence the integral
equation (9.1.3) is satisfied, as desired.
Evolution equation
We show next that a (Co) semigroup with generator A provides the solution to the
evolution equation = AM, «(0) = /.
Theorem 9.1.4 Let { T ( t ) } be a (Co) semigroup on X with generator A and let f 6
D(A). Then u(t) :— T(t)f is continuously differentiable on [0,+ 00), satisfies the
initial condition u(0) = f, and, for any t > 0, ^ = AM.
Proof First note that M(0) = T(0)f — f and so the initial condition is satisfied. In
the proof of Theorem 9.1.3, we took the derivative from the right, D+, of G(t) :=
T(t)f - f - /0' T(s)Afds. A review of (9.1.6) and (9.1.7) will reveal that, in the
process, we showed that for t > 0,

Now we need to calculate D~u(t) for any t > 0. Using the integral equation (9.1.3)
from Theorem 9.1.3, the continuity of T ( . ) A f , and the commutativity established in
Proposition 9.1.2, respectively, to justify the last three equalities, we can write

We see from (9.1.8) and (9.1.9) that ^ = AM as desired.

Finally, since ^ = AT(t)f = T ( t ) A f , it follows that ^ is continuous in t and so
u(t) is continuously differentiable.
under U.S. or applicable copyright law.

Our presentation of the proofs above closely follows that in Davies' interesting book
[Dav].
Closed unbounded operators
We will continue with a series of definitions and results revolving around unbounded
operators and (Co) semigroups. Many of the operator results are of interest apart from any
connection with semigroup theory. We will consider linear operators A whose domain
D(A) is a subspace of a Banach space X and which take values in the same Banach space
X. EBSCO
In many cases,: eBook
Publishing the range of A(EBSCOhost)
Collection could just as well
- printed on be a different
6/8/2017 Banach
5:33 PM via space Y. In
UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
fact,AN:some of the results are true in settings more general than the Banach space setting.
98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 INFINITESIMAL GENERATOR 131
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Until further notice, the symbol X will denote a Banach space. Also, when we say
"operator ", we will mean "linear operator ". Finally, we will understand that the domain,
D(A), of an operator A is a linear subspace of X.
Definition 9.1.5 If A : D(A) C X —>• X is an operator, we say that A is closed if and
only if whenever {/„} is a sequence from D(A) with |[ /„ — /1| -*• 0 and || Afn —g\\ ->• 0,
then / € D(A) and Af = g.
Definition 9.1.6 Let A : D(A) C X -> X be an operator. Then

is called the graph of A.

It is easy to see that Q(A) is a subspace of X x X. Since X x X is a Banach space
under the norm ||(/, g)\\ = \\f\\ + \\g\\, Q(A) is certainly a normed linear space under
that norm.
Proposition 9.1.7 An operator A : D(A) C X -* X is closed if and only ifQ(A) is a
closed subspace of X x X.
The function HI/HI := ||/|| + II Af\\ is a norm on D(A). It is called the graph norm
of A. It is easily seen that A e C(D(A), X) when D(A) is equipped with its graph norm.
[Here, given two normed linear spaces X and Y, C(Y, X) denotes the space of bounded
linear operators from Y to X; in particular, £(X) — C(X, X) as before.]
Proposition 9.1.8 A is closed if and only if (D( A), \\\ • |||) is complete.
Remark 9.1.9 By the closed graph theorem [Roy, Theorem 10.12, p. 231], if A is closed
and D(A) = X, then A € C(X).
The next proposition gives one connection between the concepts above and semi-
groups of operators.
Proposition 9.1.10 The generator A of a (Co) semigroup {T(t)} is closed.
We will prove Proposition 9.1.10. First we establish a lemma which is of interest in
its own right.
Lemma 9.1.11 Suppose that (T(t)} is a (Co) semigroup and that [a, b] is any compact
under U.S. or applicable copyright law.

subinterval of [0, +00). Then there is a constant M > 0 such that

for all t e [a,b].

Proof Since (T(t)} is a (Co) semigroup, for every f e X the function s H> T(s)f
is a continuous function from [0, +00) to X. Since the norm is a continuous function
on X, it follows that s \-> \\T(s)f\\ is an M-valued continuous function. Hence there
exists a constant M/ > 0 such that \\T(s)f\\ < M for all s 6 [a, b]; that is, {T(s) :
EBSCO Publishing : eBook Collection (EBSCOhost) - printed onf 6/8/2017 5:33 PM via UNIVERSIDAD
a <DISTRITAL
s < b]FRANCISCO
is strongly
JOSE DEbounded.
CALDAS By the uniform boundedness theorem [Roy, p. 232],
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 132 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(T(s) : a < s < b] is norm bounded; that is, there exists a constant M > 0 such that
\\T(s)\\<M for all s e [ a , b ] ,
Proof of Proposition 9.1.10 Let {fn} be a sequence from D(A) and suppose that fn ->
/ and Afn -» g. We wish to show that f e D(A) and Af = g. Given any to > 0, there
exists, by Lemma 9.1.11, a constant M > 0 such that ||r(s)|| < M for all * 6 [0, to].
Now for any t e [0, t0], the boundedness of T(t) and the integral equation (9.1.3)
from Theorem 9.1.3 allow us to write

where the last equality in (9.1.12) will be justified below. We have

From (9.1.13), (9.1.12) now follows. Next using (9.1.12) and the strong continuity of
the semigroup, we obtain

By definition of the generator, we now see that f e D(A) and Af = g as we wished

to show. .
Proposition 9.1.12 Let (T(t)} be a (Co) semigroup with generator A. Then the space
(D(A), ||| . |||) is a Banach space, and, further, the restriction of T(t) to D(A) is a (Co)
under U.S. or applicable copyright law.

semigroup on (D(A), ||| . |||).

Our next theorem shows that a (Co) semigroup is uniquely determined by its
generator.
Theorem 9.1.13 If {T(t)} and {S(t)} are two (Co) semigroups having the same gener-
ator A, then T(t) = S(t) for all t > 0.
Theorem 9.1.14 Let A be the generator of a (Co) semigroup {T(t)} and let f e D(A).
Then the initial-value problem (or Cauchy problem)
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 INFINITESIMAL GENERATOR 133
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

is (not only solvable but is) uniquely solvable. Moreover, the unique solution of (9.1.15)
is given by u(t) = T(t) f for t e [0, +00).
Remark 9.1.15 (a) It is the uniqueness assertion that is new in Theorem 9.1.14. We
already knew from Theorem 9.1.4 that (9.1.15) has a solution and that it is given
by the semigroup.
(b) We will discuss further on, more precisely than we did earlier, which operators A
are the generators of (Co) semigroups.
Typically, the precise domain of the generator of a (Co) semigroup is difficult to char-
acterize. It is easier instead to work on a smaller subspace called a "core" of the operator.
Our discussion for a while is of interest apart from its connection with semigroups of
operators.
Definition 9.1.16 Let A be a closed operator. We say that a subspace D c D(A) is a
core/or A if and only if for all f & D(A) there exists a sequence {fn}00n=1 in D such that
\\fn - f\\ -* 0 and \\Afn - Af\\ ^Oasn^oo.
It is easy to show the following:
Proposition 9.1.17 A subspace D of D(A) is a core for the closed operator A if and
only if it is dense in the Banach space (D(A), \\\ • |||).
Definition 9.1.18 An operator A : D(A) c X -> X is called closable if and
only if whenever {fn} 00 n [ is a sequence from D(A) such that \\fn\\ —> 0 and
\\Afn - g\\ -> 0, then g = 0.
An operator B : D(B) c X —> X is an extension of A means, of course, that
D(A) c D(B) and Bf = Af for all / € D(A). One often writes B 3 A to indicate
that B is an extension of A. If A and B are operators, it is clear that B 2 A if and only
if£(B)20(A).
Proposition 9.1.19 If an operator A : D(A) C X -> X is closable, then it has a least
closed extension, denoted A and called the closure of A. Further, Q(A) = Q(A).
Proposition 9.1.20 (i) An operator A : D(A) C X -> X is closable if and only if A
has a closed extension.
(ii) If A is closable and A is its closure, then D (A) is a core for A.
under U.S. or applicable copyright law.

The next proposition gives three basic facts about closed operators. A special case
of the third was already noted in Remark 9.1.9.
Proposition 9.1.21 Let A : D(A) c X -> X be a closed operator.
(i) If B e £(X), then A + B, with D(A + B) = D(A), is closed.
(ii) If A is injective (i.e. one-to-one), then A1 is closed. Rephrased: If A-1 exists
as a function, then it is closed.
(iii) If D(A) is closed in X, then A e £(D(A), X), where D(A) is equipped with the
norm of the Banach space X. [This follows from the closed graph theorem.]
EBSCO Publishing
Proposition : eBook
9.1.22 Let Collection (EBSCOhost)
A e £(D(A), X). -Then
printed on closed
A is 6/8/2017if 5:33
andPMonly
via if
UNIVERSIDAD
D(A) isclosed.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 134 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 9.1.23 Let A be the generator of a (Co) semigroup ( T ( t ) } . If a subspace

D C D(A) is dense in X under the original norm on X and if D is invariant under the
semigroup, that is, T(t)D C D for all t > 0, then D is a core for A.

9.2 Examples of semigroups and their generators

Next we give, without proof, three examples of semigroups and their generators. It is the
second example that will particularly concern us as we continue.
The translation semigroup
Example 9.2.1 Let X = L 2 (R) and, for every t > 0 and f e L2(R), let

Then { T ( t ) : t > 0} is a (Co) contraction semigroup. In fact, it is clear that 7X0) = I

and that T(t)T(s) = T(t + s). Also, it is easy to show that \\T(t)f ||2 = || f ||2 for every
t > 0 and so each T(t) is certainly a contraction. Next we do a calculation without
proof which will give us the correct form of the generator. For "sufficiently smooth"
functions /,

as t -» 0+. This suggests that the generator A is the first derivative operator; i.e.

Indeed, (9.2.1) gives the solution of the evolution equation fj = ff, "(0, •) = /(•)•
The reader should keep in mind that the operator A is not completely identified until its
domain is given. Often, however, it is sufficient and more convenient to specify a core.
We remark that the space D(R) of infinitely differentiable functions on R of compact
support is a core for the generator A of { T ( t ) } and, on that core, A is given by (9.2.3).
Remark 9.2.2 (a) The operators T(t) in Example 9.2.1 above are actually unitary
operators on the Hilbert space L2(R); also, T(t) makes sense for negative t and
{ T ( t ) : t e R} is a group of unitary operators, in the sense of Definition 9.6.9. Further,
various Banach spaces other than L2(R) can be considered such as L P (R) (1 < p <
under U.S. or applicable copyright law.

oo), C[—oo, +00], Co(R) (the space of continuous functions on R which vanish at oo),
and BUC(R) (the space of bounded uniformly continuous functions on R). In that case,
{T(t) : t e R} is a group of isometries of the corresponding Banach space.
(b) Another variation of Example 9.2.1 is obtained by replacing (9.2.1), for fixed
v € R, by

This group of operators models the deterministic physical phenomenon of motion in a

straight line at constant
EBSCO Publishing velocity (EBSCOhost)
: eBook Collection v. In this -case, A on—6/8/2017
printed v j^ and
5:33the appropriate
PM via UNIVERSIDAD partial
DISTRITAL FRANCISCO JOSE DE CALDAS
differential equation is ff = vjjf, w(0, •) = /(•).
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 EXAMPLES OF SEMIGROUPS AND THEIR GENERATORS 135
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The heat semigroup

Example 9.2.3 Let X = L2(E) and, for t > 0 and / e L2(M), let

Also let F(0) = I. The semigroup property T ( t ) T ( s ) = T(t + s) follows from

Proposition 3.2.3, the Chapman-Kolmogorov equation. One can also show that for every
/ e L2(K) and for every t > 0, ||T (t) f ||2 < || f ||2 and, further, that || T (t) f - f||2 -f| 2+0
as t -» 0+. Thus (T (t) : t > 0} is a (Co) contraction semigroup. Next we do a cal-
culation, again without proof, which will give us the form of the generator A. Let
/ eC00OO(R) = (X>(R)), the space of infinitely differentiable functions with compact
support in R. Using Taylor's theorem, we can write

where y is between x and y. Now / € C^(R) implies that |/(4)| is bounded by some
number, say M. Then
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 136 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the next to last equality in (9.2.6) comes from formula (4.7.2) of Appendix 4.7.
From (9.2.6), the bound on |/'4'|, and formula (4.7.3), as well as the fact that / has
compact support, we can write

as t -»• 0+. It now looks quite plausible from (9.2.7) that the generator A of the semigroup
in (9.2.4) satisfies

at least for functions f e C^(M). This is in fact the case, and C^(R) is a core for the
generator A. Thus the semigroup in (9.2.4) provides a solution for the heat equation

and, hence, the name heat semigroup.

Remark 9.2.4 (a) The presence of the normal density function with mean x and variance
t on the right-hand side of (9.2.4) as well as the use of the Chapman-Kolmogorov
equation above to verify the semigroup property suggest that the heat semigroup and
its generator are related to the Wiener process. In fact, using the Wiener integration
formula (3.3.9) from Theorem 3.3.5 (and recalling that CQ := C0'r, in the notation of
under U.S. or applicable copyright law.

Chapter 3), the reader can easily check that

A great deal more can be said [BluG, Frei, L2, Si9, Va] about the relationship between
the Wiener process, the heat semigroup and its generator. We will pursue certain aspects
of this as we continue.
(b) The exact domain of the generator of the heat semigroup is known. It is the
EBSCO Publishing : eBook Collection 2
collection of all functions f in L(EBSCOhost)
(R) which- printed
have twoon 6/8/2017 5:33 PM via derivatives"
"distributional UNIVERSIDAD both
DISTRITAL FRANCISCO JOSE2 DE CALDAS 2
of which belong
AN: 98476 to LGerald
; Johnson, (R);W.,inLapidus,
other Michel
words,L..;
it is
Thethe Sobolev
Feynman space
Integral H (K). In addition
and Feynman's
Operational Calculus
Account: ns000601
 EXAMPLES OF SEMIGROUPS AND THEIR GENERATORS 137
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

to CQQ(R) which was mentioned above, another useful core for the generator is the
Schwartz space S = S(R) of infinitely differentiate functions on R of rapid decrease.
Suchfunctions f are required to satisfy \x\m f^(x) —» Oo?|jr| —»• ooforallnonnegative
integers m andn. (Analogous comments apply when R is replaced by M.d, as in (c) below.)
(c) In L 2 (M d ), the heat semigroup is given by

and its generator is 1/2 times the Laplacian; that is,

(See Theorem 10.2.6 below.) The semigroup then gives the solution to the heat (or
diffusion) equation in Rd:

The generator in (9.2.12) is often written as

Also one usually writes e tH° rather than T(t)for the semigroup. The exponential can
be defined apart from the theory of (Co) semigroups in this case via the spectral theory
since the "free Hamiltonian" H0 is a self-adjoint operator. We will discuss these matters
somewhat further on in Chapter 10 (Sections 10.1 and 10.2). In place of (9.2.11) and
(9.2.10), one writes

for all f e L 2 (R d ).
(d) There are other Banach spaces on which the formula (9.2.4) defines a (Co)
semigroup: LP(R) (1 < p < oo), BUC(K), and C[-oo, +00].
under U.S. or applicable copyright law.

The Poisson semigroup

Example 9.2.5 Let X = BUC(R) and, for t > 0 and / €BUC (R), let

It can be shown that { T ( t ) } is a (Co) contraction semigroup on X whose generator A is

the difference operator given by

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
In this case,FRANCISCO
DISTRITAL A € £(X)JOSE and ||A|| < 2|A.|.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 138 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The heat semigroup and the Poisson semigroup (Examples 9.2.3 and 9.2.5, respec-
tively), have more in common than may be immediately apparent. Taking d = 1, we can
rewrite (9.2.15) as

Also, (9.2.16) with A. = u — 1 can be rewritten as

where (H1 f ) ( x ) := f(x) - f(x — 1). Formula (9.2.15a) inVolves integration with
_1 -V
respect to the probability measure (2nt) ^e"S~dy (normal distribution with mean 0 and
variance t), whereas (9.2.16a) involves integration with respect to the discrete probability
^
measure with mass e~' '^ at k, where k =0,1,2,... (Poisson distribution with intensity
1 [Kan, Chapter 4]). Also, we saw in (9.2.10) that the heat semigroup can be written as
a path space integral,

for all f e L2(Rd). It is also true that (9.2.16a) can be written as a path space integral in a
similar manner. Further, as we will see in our discussion of the Feynman-Kac formula in
Chapter 12, (9.2.10a) can be greatly extended. An analogous extension is again possible
in the Poisson case.
Our main purpose in comparing the heat and Poisson semigroups is to indicate that
the use of path integrals is not restricted to the Wiener process. Indeed, such integrals can
be employed along with semigroup methods in connection with a variety of stochastic
processes [Frei, Fuk, FukOT, L2, MaRo, Si9, Va].

9.3 The resolvent

We now return to the abstract theory of (Co) semigroups.
under U.S. or applicable copyright law.

Definition 9.3.1 Let A : D(A) c X -+ X be an operator on X. The resolvent set of A

is

Here, as before, I denotes the identity operator on X. Further, we will often write
X — A instead of AI — A.
EBSCO Publishing
Proposition 9.3.2: eBook
(i) IfCollection
p(A) +(EBSCOhost)
0, then A- isprinted on 6/8/2017 5:33 PM via UNIVERSIDAD
closed.
DISTRITAL FRANCISCO JOSE DE CALDAS
(ii) If A is closed, then p(A) = {A e C : XI - A : D(A) -» X is bijective}.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE RESOLVENT 139
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof (i) Suppose that X e p(A).Then(X-A)- 1 € £(X)andso(y-A)- 1 is obviously

closed. Hence, by Proposition 9.1.21 (ii), X—A is closed. It follows from (i) of Proposition
9.1.21 that X - A - X = -A is closed. Thus A is closed.
(ii) Suppose that XI - A is bijective and that A is closed. By Proposition 9.1.21(i),
XI — A is closed. Hence, from Proposition 9.1.21(ii), (y. — A)-1 is closed. The closed
graph theorem then implies that (X — A)-1 6 C(X).
Definition 9.3.3 a (A) := C \p(A) is called the spectrum of A.
Definition 9.3.4 If X e p(A), the operator

is called the resolvent (operator) of A.

The resolvent plays an important role in the theory of (Co) semigroups. The next
proposition, although simple, is quite useful.
Proposition 9.3.5 (The resolvent equation) Let A : D(A) c X —> X be a closed
operator on X. Then for all X, u e p(A),

Further (it follows easily from (9.3.2) that) R(X; A) and R(u.; A) commute.
The following result, which does not have a direct counterpart in the theory of non-
linear semigroups [Brel], is extremely useful in passing from semigroups to resolvents.
Theorem 9.3.6 Let A be the generator of a (Co) semigroup (T(t)} satisfying

for all t > 0 and for some constants w > 0 and M > 1. Then a (A) C {£ €C : Re £ <
w}. Moreover, if Ref > w,
under U.S. or applicable copyright law.

In short, the resolvent is the Laplace transform of the semigroup T(t).

Remark 9.3.7 (a) As will be seen in Proposition 9.4.1 below, a (Co) semigroup always
satisfies an inequality of the type (9.3.3).
(b) We implicitly assume in Theorem 9.3.6 that X is a complex Banach space. If X
is only a real Banach space, then the analogous result holds, but only for real
numbers.
(c) The integral in the right-hand side of (9.3.4) converges in the strong oper-
ator topology; i.e. for each f € X, S(£)f := /0 e~^'T(t)fdt is a con-
vergent improper Riemann integral. Note that by (9.3.3), \\e~^'T(t)f\\ <
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
Mexp{-(Re£
DISTRITAL FRANCISCO JOSE- DEco)t}\\f\\,
CALDAS so that S(?) €£(X) and \\Stf )|| < M/(Ref - «).
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 140 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Exercise 9.3.8 (Exponential formula) Under the assumptions of Theorem 9.3.6, show
by induction that if Re f > w and n is an integer > 1, then

Then prove that for all f e X,

uniformly in t on compact subsets of [0, +00). Rephrased:

This is called the exponential formula and provides one justification for the notation
T(t) = etA. (Often, one prefers to write T(t) = e-t B , where A = —B is the generator.)
Remark 9.3.9 Formula (9.3.5) provides a simple way of generating the semigroup by
means of the resolvents. It motivates in part one approach to the Feynman integral based
on a product formula for "imaginary resolvents" ([La6], [Lal1], [Lal3, Part I]) to be
discussed later on in Sections 11.3-11.6.

9.4 Generation theorems

We now turn to the generation theorems which play a crucial role in the theory of (Co)
semigroups. We begin with a simple proposition whose proof we include.
Proposition 9.4.1 Let (T(t)} be a (Co) semigroup on X. Then there exist constants
w > 0 and M > 1 such that for all t > 0,

Proof We know from Lemma 9.1.11 that there exists a constant M > 1 such that
\\T(s)\\ < M for all s € [0, 1]. Let w) = In M so that M = ew. We consider a fixed
t, with t > 0. Let n be the largest integer such that t < n but t > n - 1. By the
semigroup property, T(t) = T(t/n) . T(t/n) T(t/n) = T(t/n)n. Then because
under U.S. or applicable copyright law.

of the inequality \\AB\\ < \\A\\ \\B\\ in L ( X ) , we can write

Thus (9.4.1) is established.

The Hille-Yosida theorem

EBSCO
Next wePublishing
will state: eBook Collection
precisely the (EBSCOhost)
Hille-Yosida- printed on 6/8/2017
theorem 5:33 (Co)
first for PM viacontraction
UNIVERSIDAD semi-
DISTRITAL FRANCISCO JOSE DE CALDAS
groups and ;then
AN: 98476 in general.
Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERATION THEOREMS 141
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 9.4.2 (Hille-Yosida) The operator A is the generator of a (Co) contraction

semigroup if and only if A is closed, densely defined, and for each A. > 0, X e p (A) and

The proof of the following general theorem makes use of the contraction case and
of Proposition 9.4.1 above.
Theorem 9.4.3 (Generation theorem) The operator A is the generator of a (Co) semi-
group { T ( t ) } if and only if A is closed, densely defined, and constants M > 1, w e R
exist such that A. € p(A)for each A. > (a and

whenever A, > u> and n — 1,2,3 In this case, for every t > 0,

Moreover, an equivalent norm \\\ • \\\ exists on X such that { S ( t ) : t > 0} is a (Co)
contraction semigroup on (X, ||| . |||) with generator A — wI, where

Dissipative operators and the Lumer-Phillips theorem

Next we introduce "dissipative operators". Our immediate purpose is to state the Lumer-
Phillips theorem, which is a relative of the Hille-Yosida theorem, but with assumptions
that are easier to verify. The following two definitions may appear a bit cumbersome to
some readers; in fact, as we will see in Remark 9.4.6(a), they take a much simpler form
when X is a Hilbert space, the setting of most interest to us.
Definition 9.4.4 Let X be a Banach space with dual denoted by X*. Given f in X, let

For each f e X, J(f) / 0 as is easily seen from one of the standard corollaries
of the Hahn-Banach theorem [Ru2, Theorem 5.20, p. 107]. Then J is (in general)
under U.S. or applicable copyright law.

a multivalued function called the duality map of X. Let J be a section of J; i.e. let J
be a function, J : X —> X*, such that J ( f ) e J(f) for every f € X. The point is that
J is a true function and not just a multifunction. Such a J exists by the axiom of choice
and is called a duality section.
Definition 9.4.5 An operator A : D(A) c X —> X is called dissipative with respect to
a duality section J if and only if for each f e D(A),

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:33 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 142 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

If we restate (9.4.5) using the symbol (•, •) to denote the bilinear action of X and X* on
one another, it becomes

for every f e D(A). Then A is dissipative if and only if it is dissipative with respect
to some duality section. A dissipative operator A is m-dissipative if and only if p(A) (~l
(0, +00) + 0.
Remark 9.4.6 (a) If X is a Hilbert space, there is no need to consider duality sections
since J ( f ) = J(f) = f for all f e X. In this case, the dual action (or "duality
bracket") (•, •) is induced by the inner product (•, •), and the condition (9.4.6) just
becomes the assertion that for all f e D(A),

(Also see Remark 9.6.15(a) below.) Dissipative operators have been studied most in the
Hilbert space setting.
(b) One says that A is accretive (resp., m-accretivej if — A is dissipative (resp.,
m-dissipative). This alternative terminology is used in much of the work on semigroups
of operators, for example, [Brel, Kat8, Patz].
(c) There are Banach spaces other than Hilbert spaces where the duality map J is
single-valued. This happens, for example, when X = LP(W), p e (1, oo)\{2}.
We now state the Lumer-Phillips theorem. Its proof [Gol, pp. 26-27], which we will
not give, rests in part on the Hille-Yosida Theorem.
Theorem 9.4.7 (Lumer-Phillips theorem) Suppose that A generates a (Co) contraction
semigroup. Then
(i) D (A) is dense in X,
(ii) A is dissipative with respect to every duality section,
(iii) (0, +00) c p(A).
Conversely, if A : D(A) c X —> X satisfies
(i') D(A) is dense in X,
(ii') A is dissipative with respect to some duality section,
under U.S. or applicable copyright law.

(iii') (0,+oo)n/o(A) ^0,

then A generates a (Co) contraction semigroup on X.
Combining the above, we see that A generates a (Co) contraction semigroup if and
only if A is densely defined and m-dissipative.
Definition 9.4.8 A dissipative operator A is called maximal dissipative if and only if A
has no proper dissipative extension.
Proposition 9.4.9: eBook
EBSCO Publishing (i) IfCollection
A is m-dissipative, then Aonis6/8/2017
(EBSCOhost) - printed maximal 5:33dissipative.
PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
(ii) If X is a Hilbert space, A is m-dissipative if and only if A is maximal dissipative.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 UNIFORMLY AND WEAKLY CONTINUOUS SEMIGROUPS 143
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

9.5 Uniformly continuous and weakly continuous semigroups

By definition of a (Co) semigroup (T(t)}, for every / 6 X, T(t)f is continuous in t
with respect to the norm topology on X; that is, T(t) is continuous in t with respect to the
strong operator topology on L ( X ) . It is natural to ask what happens if we require T(t) to
be continuous in the operator norm topology on L(X) or equivalently, to be a uniformly
continuous semigroup. A rough answer is that pleasant things happen but that it throws
us into a setting which is too pleasant for many of the applications of interest. The next
two propositions will make the preceding remarks precise. First we give a definition.
Given A e C,(X) and any t e R, it is clear that the seriesE00n=0(tA)n/n!converges in
operator norm.
Definition 9.5.1 Let A € L(X). We define for any t e R,

Proposition 9.5.2 Given A e C(X), let T(t) := etA for every t > 0. Then (T(t)} is a
(Co) semigroup on X with generator A. Also, T(t) is continuous in operator norm as a
function of t; in particular,

Proposition 9.5.3 Let (T(t)} be a (Co) semigroup on X satisfying (9.5.2). Then the
generator A of{T(t)} belongs to L ( X ) and T(t) = etA for all t > 0.
Remark 9.5.4 Formula (9.5.1) in Definition 9.5.1 actually defines an operator norm
continuous (Co) group of operators, (T(t) : t e R}, on X.
In the definition of a (Co) semigroup {T(t)} (see Definition 8.1.5), we require con-
tinuity in the strong operator topology (in the sense that T ( . ) f e C([0, +00), X) for
each fixed f e X). What happens if we reduce the requirement to continuity in the weak
operator topology? The next result, which is useful in the proof of several of the basic
theorems, tells us that it makes no difference.
under U.S. or applicable copyright law.

Theorem 9.5.5 Let (T(t) : t > 0} be a family of operators in L ( X ) satisfying

(i) T(t + s) = T(t)T(s) for each t, s e [0, +00),
(ii) 7X0) = I,
(iii) 0[T(-)/] € C([0, +00), C) for each f 6 X and(f e X*.
Then T ( . ) f € C([0, +00), X)for each f E X; that is, (T(t)} is a (Co) semigroup
on X.
Remark 9.5.6 Theorem 9.5.5 holds even when (iii) is reduced to the assumption that
0 [ T ( . ) f ] is continuous at 0 for each f € Xand0p e X*; in other words, in the presence
of assumptions (i) and (ii), weak operator continuity of T(.) at 0 implies strong operator
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
continuity on [0, +00).
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 144 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

9.6 Self-adjoint operators, unitary groups and Stone's theorem

Next we turn to the consideration of adjoint operators on a (complex) Hilbert space H
and their relationship with (Co) semigroups on H.
Definition 9.6.1 Let A be a densely defined operator on the Hilbert space H. The vector
g e H. is in the domain of A* if and only if the map f \-+ (Af, g) is a continuous linear
functional on D(A) which we then extend by continuity to D(A) — H. In this case, there
is a unique h e H. such that (Af, g) = (f, h) for all f 6 D(A). We then define A*g = h.
Hence, for f e D(A) and g e D(A*), we can write

The operator A* : D(A*) C H -* H, is called the (Hilbert) adjoint of A.

Our next proposition lists some basic properties of adjoints.
Proposition 9.6.2 (i) If A € C(H), then A* e C(H) and \\A*\\ = \\A\\.
(ii) If A and B are in L(H), then (AB)* = B* A* and (A + B)* = A* + B*.
(iii) If D(A) n D(B) = H, then (A + B)* A* + B*.
Further, if D(A) = H, then
(iv) (a A)* = a A*, for all a e C.
(v) A* is closed.
(vi) A is closable if and only if D(A*) = "H. In this case, A** := (A*)* is well
defined and A = A**; further, (A)* = A*.
(vii) A C B implies B* C A*.
(For a proof of (v) and (vi), see, e.g., [ReSil, Theorem VIII.l, pp. 252-253].)
Theorem 9.6.3 Let A generate a (Co) semigroup (T(t) :t>Q}on'H. Then (T(t)* :
t > 0} is a (Co) semigroup on H (called the adjoint semigroup) whose generator is A*.
Definition 9.6.4 Let B be a densely defined operator on H.. Then
(i) B is symmetric if and only if B C. B*; i.e. for every g e D(B), the map
f i-* (Bf, g) is a continuous linear functional on D(B) so that g e D(B*) and,
further, B*g = Bg. Thus, when B is symmetric, we can write
under U.S. or applicable copyright law.

for all f, g e D(B).

(ii) B is skew-symmetric if and only if B C. —B*.
(iii) B is self-adjoint if and only if B* = B (i.e. B c B* and B* C B).
(iv) B is skew-adjoint if and only if B* = —B.
(v) B is essentially self-adjoint if and only if it has a unique self-adjoint extension,
necessarily its closure B.
EBSCO
Let BPublishing : eBook defined
be a densely Academic operator
Collection on
(EBSCOhost)
Ji. It is-clear
printed on (a)
that 6/8/2017
if B 5:35 PM via
is self-adjoint, then
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
B isAN:symmetric and that
98476 ; Johnson, Gerald(b)W.,if Lapidus,
B is skew-adjoint then
Michel L..; The it is Integral
Feynman skew-symmetric.
and Feynman'sThe converse
Operational Calculus
Account: ns000601
 SELF-ADJOINT OPERATORS AND UNITARY GROUPS 145
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

of each of these assertions is false. In particular, in spite of occasional confusion on this

point among people trying to apply the theory of self-adjoint operators, B symmetric
does not imply that B is self-adjoint. We illustrate this with the following example. For
more details, see [DunSc2, p. 1223].
Example 9.6.5 Let H := L2[0, 1] be the space of C-valued square integrable functions
on [0,1]. Let

where

(Here, f denotes the derivative of / which, by Lebesgue's theorem ([Roy, Corollary

5.12, p. 109] or [Cho2]) exists pointwise almost every where on [0, 1] since / is assumed
to be absolutely continuous.) We define operators A2 and Aa with the same formula
(9.6.3) but by restricting the domain:

Clearly A3, C A 2 C AI. We can see that A3 is symmetric by integrating by parts

[Coh, Corollary 6.3.8, p. 189]. Accordingly, for f, g e D(A3) we can write

and so AT, is symmetric. Next we show that A3, is not self-adjoint by showing that
1D(A\) C D£>(A3). Let g 6 D(A1) so that g is absolutely continuous and g' 6 L2[0, 1].
We must show that the map / H> (A3/, g) from D(A3) to C is continuous. But, much
as in (9.6.7),
under U.S. or applicable copyright law.

and so the desired continuity follows.

When an operator such as AT, is symmetric but not self-adjoint, the hope is to find
a self-adjoint extension. It may well be that no such self-adjoint extension exists (see
[DunSc2, Corollary 13, p. 1230]). However, in the present case, A 2 is a self-adjoint exten-
sion of A3. In fact, a more complete picture of the relationships between the operators
A], A2, A3 defined in (9.6.3) through (9.6.6) is expressed by
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 146 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We note that if an operator A is self-adjoint, then it does not have any proper self-
adjoint extension. This is an easy consequence of Proposition 9.6.2(vii).
The next proposition is trivial but the connection it describes between skew-
adjointness and self-adjointness has some importance.
Proposition 9.6.6 Let B be a densely defined operator on H. Then B is skew-adjoint if
and only if i B is self-adjoint.
Proof Suppose that B* = -B. Then (iB)* = iB* - -iB* = -i(-B) = iB as
claimed. The converse is just as easy.

The following result answers a natural question in view of Theorem 9.6.3 above
and will be revisited in Section 10.1, where further background will be provided. (See
especially Remark 10.1.12(b) and Exercise 10.1.13.)
Theorem 9.6.7 {T(t) : t > 0} is a self-adjoint (Co) semigroup on H (i.e. T(t) is self-
adjoint for each t > 0) if and only if its generator —A is self-adjoint, with A bounded
below (i.e. there exists c e M such that (Af, /) > c||/||2, for all f € D(A)). In this
case, we write T(t) = e~'A, for t > 0.
In particular, (T(t) : t > 0} is a self-adjoint contraction semigroup if and only if A
is self-adjoint and nonnegative (i.e. (Af, f) > 0, for all f e D(A)).
Proof See Exercise 10.1.13.
Proposition 9.6.8 If B e L(H) and B is symmetric, then B is self-adjoint.
Definition 9.6.9 (i) Let U e C(H). Then U is unitary if and only if U-1 exists as an
element of C(H) and U* = U-1.
(U) A (Co) unitary group {T(t) : t 6 M} is a (Co) group of unitary operators.
Hence, in particular, both (T(t) : t > 0) and (T(-t) : t > 0} are (Co) semigroups,
with (necessarily) opposite generators B and —B. The operator B is then called the
generator of the unitary group.
Proposition 9.6.10 U e C(H) is unitary if and only if U is isometric and surjective.
Next we state a theorem of M. H. Stone [Sto] which is crucial to the mathematical
study of quantum mechanics.
Theorem 9.6.11 (Stone) The operator B is the generator of a (Co) unitary group (T(t) :
under U.S. or applicable copyright law.

t 6 R) on H if and only if B is skew-adjoint. In this case, by Proposition 9.6.6, B = —iA,

where A is self-adjoint, and so we write T(t) = e i t A for t e R.
Remark 9.6.12 Much of the usefulness of Theorem 9.6.11 is due to the fact that the
"spectral theorem" is valid for (unbounded) self-adjoint operators. We will discuss the
spectral theorem in Section 10.1 and justify (in Remarks 10.1.12(b) and (a), respectively)
the above notation for self-adjoint semigroups and unitary groups.
We have defined A* only for operators A on a Hilbert space. We now define the
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
adjoint of an operator on a Banach space and state one theorem about (Co) semigroups
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
in this context.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PERTURBATION THEOREMS 147
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Definition 9.6.13 Let A be a densely defined operator on a Banach space X. The

(Banach) adjoint A' is to be an operator on the dual space X* (the space of contin-
uous linear functionals on X). An element (j> is in the domain of A' if and only if there
exists i/r 6 X* such that (Af, <j>} = (f, tfr) for all f e D(A); in this case, A'4> := ij/.
Thus if(f) € D(A') and f e D(A), we can write

Theorem 9.6.14 Let A generate a (Co) semigroup {T(t) : t > 0} on a reflexive Banach
space X. Then the "adjoint semigroup" (T'(t) : t > 0} is a (Co) semigroup on X*
whose generator is A'.
Remark 9.6.15 (a) Let A be an operator on a Hilbert space H. We assume for simplicity
that A is a bounded operator. Since "H is a Banach space, A has a Banach space adjoint
(according to Definition 9.6.13) which we denote here as A'. The Hilbert space adjoint
A* is closely related to but not quite the same as A'. The source of the difference is
that the dual action (-,-) is bilinear whereas the inner product (•, •) is linear in the
first variable but conjugate linear in the second variable. One consequence is that the
property (cA)* = ~cA* from (iv) of Proposition 9.6.2 is replaced by (cA)' = cA'. A
careful discussion of the relationship between the Hilbert space adjoint and the Banach
space adjoint can be found in [ReSil, §V7.2, pp. 185-187]. We caution the reader that
in [ReSil] the inner product is taken to be linear in the second variable and conjugate
linear in the first, in agreement with the physicists' convention.
(b) If X is replaced by a Hilbert space H, then as was seen in Theorem 9.6.3, the
exact analogue of Theorem 9.6.14 holds, except for Banach adjoints replaced by Hilbert
adjoints.
(c) With obvious exceptions such as Theorem 9.6.11, most of the results we discuss
here are true for either real or complex Hilbert spaces, sometimes with minor changes.
Throughout the rest of this book, we will most often work with complex Hilbert spaces
since we are especially interested in connections with quantum mechanics. Probabilists,
in contrast, would usually be interested in Hilbert (or Banach) spaces over the reals.

9.7 Perturbation theorems

We move next to the discussion of some perturbation results.
under U.S. or applicable copyright law.

Theorem 9.7.1 Let A generate a (Co) contraction semigroup. Let B be dissipative with
D(B) 5 D(A). Assume that there are constants 0 < a < 1 and b > 0 such that

Then A + B (with domain D(A)) generates a (Co) contraction semigroup.

If B is a bounded generator as in Theorem 8.1.11, then (9.7.1) is satisfied with a = 0
and b — \\B\\. More interestingly, unbounded operators B can satisfy (9.7.1) as long as
the EBSCO
unboundedness of B is controlled relative to A. In a concrete case, A might be a
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
second order DISTRITAL
UNIVERSIDAD differential operator,
FRANCISCO JOSE DEBCALDAS
a first order differential operator and (9.7.1) some
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 148 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

type of Sobolev inequality. See Section 1.2 of Chapter IV of Kato's book [Kat8] for
examples; further related material can be found in various other places in [Gol], [Kat8]
and [ReSi2].
The next result gives an expression for the perturbed semigroup as a perturbation
series. First we need a definition.
Definition 9.7.2 Let A be the generator of a (Co) semigroup (T(t)} on X. We denote
by P(A) the set of operators B on X satisfying:
(i) D(B) ^ D(A) and B is closed,
(ii) there exists a function K : (0, +00) —>• [0, +00) such that for each t > 0 and
for each f e D(A)

and, further,

Theorem 9.7.3 Let A be the generator of a (Co) semigroup (T(t)} on X and let B e
P(A). Then A + B (with domain D(A)) generates a (Co) semigroup {S(t)} given by the
perturbation series

where S0(0 = T ( t ) and, for n = 0, 1, 2 , . . . and f e X,

Finally, the series in (9.7.4) converges absolutely (in operator norm) and uniformly on
any compact subset of [0, +00).
Remark 9.7.4 The condition (9.7.2) holds if B e C(X), but it can also hold for
unbounded B. What is required by (9.7.2) is that BT(t) be bounded on D(A) for every
t >0.
under U.S. or applicable copyright law.

Since it is related to later work, we will give an expression for Sn in terms of a

multiple integral. We will not give any proof but just do the formal calculations:

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PERTURBATION THEOREMS 149
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and in general, for n = 1, 2, . . . ,

[Compare (9.7.4) and (9.7.8) with some of the simplest perturbation expansions (for
example, (15.3.22) and (15.3.2)-(15.3.4)) appearing in our approach to Feynman's oper-
ational calculus presented in Chapters 14-19. A related brief comment is given in Remark
9.7.6(b) below.]
Theorem 9.7.3 is stated without proof in [Gol, Theorem 6.5, p. 41] and is worked
out in detail in [DunScl, pp. 630-641]. We will state the results of Exercise 2 in [Gol,
p. 42] as an illustration of a situation to which Theorem 9.7.3 can be applied.
Example 9.7.5 Let X = BUC(R) and let {T(t)} be a (Co) contraction semigroup on
X defined by

[This is essentially the heat semigroup from Example 9.2.3 but acting on the Banach
space BUC(K) and with 2t from (9.2.4) replaced by 4t above.] The semigroup (T(t)}
has generator

where

Let

where h e X and
under U.S. or applicable copyright law.

D(B) := [f e X: f is continuously differentiable in a neighborhood of

Then B € P(A) so that, by Theorem 9.7.3, A + B generates a (Co) contraction

semigroup {S(t)} which can be written as a perturbation series (9.7.4).
Remark 9.7.6 (a) One can take K(t) := \\h\\oo/vnt, where K is the function which
must be produced to show that the operator B from (9.7.12) is in P(A).
(b) The series (9.7.4) with Sn defined (EBSCOhost)
EBSCO Publishing : eBook Academic Collection
by (9.7.8)- can be thought of as making sense of
printed on 6/8/2017 5:35 PM via
Feynman 's time-ordered
UNIVERSIDAD operational
DISTRITAL FRANCISCO calculus for noncommuting operators in the present
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 150 LINEAR SEMIGROUPS OF OPERATORS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

setting. The role of time-ordering is most clearly apparent in (9.7.8). If functions of the
noncommuting (or not necessarily the commuting) operators A and B are involved here,
what are these functions? We do not want to take time to explain ourselves precisely here,
but we mention that the function associated with the series is g(A, B) = e'(A+B^ whereas
the function associated with the nth term of the series is gn(A, B) = etA^f-.
Next we define the concept of a "Kato perturbation" and state two final perturbation
results which are easy consequences of Theorems 9.7.1 and 9.6.11.
Definition 9.7.7 Let A : D(A) C X -> X be an operator on X. An operator B on X
is said to be a Kato perturbation of A if and only if D(B) ~3 D(A) and for each a > 0,
there is a constant b — b(a) > 0 such that

for each f € D(A).

It is not hard to show that the class of Kato perturbations is closed under finite linear
combinations. In contrast, the perturbations involved in Theorem 9.7.1 do not have this
pleasant closure property.
Corollary 9.7.8 If A generates a (Co) contraction semigroup and if B is a dissipative
Kato perturbation of A, then A + B (with domain D(A)) generates a (Co) contraction
semigroup.
Corollary 9.7.9 If A is a self-adjoint operator on a complex Hilbert space H and if B
is a symmetric Kato perturbation of A, then A + B (with domain D(A)) is self-adjoint.
Next we give three theorems dealing with the continuous dependence of a semigroup
on its generator.
Theorem9.7.10 Let An (n = 0, 1 , 2 , . . . ) generate a (Co) semigroup (Tn(t)} and
suppose that this family of semigroups satisfies the "stability condition"

where M and u> are independent of n and t. Let D be a core for A0. Assume that for each
/eZ>,
under U.S. or applicable copyright law.

Then lim Tn(t)g = To(t)g for each g e X, where the convergence is uniform for t
n—*oo
in compact subsets of [0, +00).
The next two theorems are important to the theory. (See, for example, [Gol, §1.7] and
[Kat8, §IX.2.5].)
EBSCO Publishing
Theorem 9.7.11 : (Trotter,
eBook Academic Collection
Neveu, Kato, (EBSCOhost)
Chemoff) - Letprinted
An on(n6/8/2017
= 0, 1,5:352 , PM
. . .via
) generate a
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
(Co)
AN:semigroup {Tn (t)}
98476 ; Johnson, satisfying
Gerald theMichel
W., Lapidus, stability
L..; condition
The Feynman (9.7.15) with
Integral and M, u> independent
Feynman's
Operational Calculus
Account: ns000601
 PERTURBATION THEOREMS 151
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

of n, t. Then:
(i) lim Tn(t)f = To(t) f for each t > 0 and f e X implies that lim R ( X ; An)f =
n-oo n—>oo
R(\; A0)f for each f € X, uniformly for X in compact subsets of (w, +00).
(ii) lim R(X; An)f = R(X; A0)/ for each f € X and A. > w implies
n—»oc
lim Tn(t)f = To(t)f for each f E X, uniformly for t in compact subsets
n—*OQ
of [0, +00).
Next we state a result much in the spirit of (ii) of Theorem 9.7.11 but having the
advantage that the limiting generator A0 and semigroup {T0(f)} need not be identified
ahead of time.
Theorem 9.7.12 Let An(n = 1, 2 , . . . ) generate (Co) semigroups [Tn(t)} satisfying
||T n (t)|| < Me wt with M and w independent of t and n. Let A.Q satisfy Re A0 > co, and
suppose that

holds for all f € X and some injective operator R on X having dense range.
Then a semigroup generator AO and its corresponding semigroup {7b(?)} exist such
that R = R(Xo; A0) and lim Tn(t)f = T0(t) f holds for all f e X, uniformly for t in
n -00
compact subsets of [0, +00).
We know from Theorem 9.1.4 that if {7(0} is a (Co) semigroup on X with generator
A and if f e D(A), then u(t) := T(t)f is the solution of the Cauchy problem u'(t) =
AM, w(0) = /. We have not mentioned it earlier, but, in fact, semigroup methods also
allow us to solve the nonhomogeneous Cauchy problem:

Theorem 9.7.13 Let A generate a (C0) semigroup (T(t)} on X and let f e D(A).
Assume that either
(i) g belongs to C([0, +00), X) and takes values in D(A) and, further, Ag e
C([0, +00), X),
or else that
under U.S. or applicable copyright law.

(ii) g e C'([0, +00), X), the space of X-valued continuously differentiable func-
tions on [0, +00).
Then the nonhomogeneous Cauchy problem (9.7.17) has a unique solution u e
C1 ([0, +00), X) with values in D(A) which is given for all t > 0 by

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

10
UNBOUNDED SELF-ADJOINT OPERATORS
AND QUADRATIC FORMS

The object of this book is to present the new quantum mechanics in a unified representation
which, so far as it is possible and useful, is mathematically rigorous. ... [A] presentation of the
mathematical tools necessary for the purposes of this theory will be given, i.e., a theory of Hilbert
spaces and the so-called Hermitian [or self-adjoint] operators. To this end, an accurate introduction
to unbounded operators is also necessary; that is, an extension of the theory beyond its classical
limits (developed by Hilbert and E. Hellinger, F. Riesz, E. Schmidt, O. Toeplitz). The following
may be said regarding the method employed in this type of treatment: as a rule, calculations should
be performed with the operators themselves (which represent physical quantities) and not with the
matrices, which, after the introduction of a (special and arbitrary) coordinate system in Hilbert
space, result from them. This "coordinate free" (i.e., invariant) method, with its strongly geometric
language, possesses noticeable formal advantages.
Dirac in several papers and in his recently published book ([Dir2]), has given a representation
of quantum mechanics which is scarcely to be surpassed in brevity and elegance, and which is,
at the same time, of invariant character. ... The method of Dirac ... in no way satisfies the
requirements of mathematical rigor—not even if these are reduced in a natural and proper fashion
to the extent common elsewhere in theoretical physics. For example, the method adheres to the
fiction that each self-adjoint operator can be put in diagonal form. When this is not actually the
case, this requires the introduction of "improper" functions with self-contradictory properties....
It should be emphasized that the correct structure [of quantum mechanics] need not consist
in a mathematical refinement and explanation of the Dirac method, but rather that it requires a
procedure differing from the very beginning; namely, the reliance on the Hilbert theory of operators.
John von Neumann, 1932 (excerpts from the preface of his book Mathematical Foundations
of Quantum Mechanics [vN])

Up till now, the theory of quadratic (or Hermitian) forms of infinitely many variables has been
developed mainly for a special class ("bounded" forms) (D. Hilbert, Grundziige einer allgemeinem
under U.S. or applicable copyright law.

Theorie der linearen Integralgleichungen; E. Hellinger, Crelles Joum. 136 (1910), 1). But here we
are concerned just with non-bounded forms. We may nevertheless assume that in the main, the
rules run likewise.
M. Born, W. Heisenberg and P. Jordan, 1926 ([BorHJ], reprinted in [vW, p. 351])
The spectral theorem for unbounded, self-adjoint operators (Section 10.1) and the rep-
resentation theorems for unbounded, quadratic forms (Section 10.3) are the main topics
in operator theory that we wish to discuss before returning to the mainstream of our
development. In addition, we also consider in Sections 10.2 and 10.4 various concrete
EBSCOthat
results Publishing : eBook Academic
are directly relevantCollection (EBSCOhost)
to our later work.- More
printedprecisely,
on 6/8/2017 in
5:35 PM via 10.2, we
Section
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS tH0
use AN:
the98476
spectral theorem
; Johnson, GeraldtoW.,
study the Michel
Lapidus, semigroup
L..; The e-
Feynmanand the unitary
Integral group e-itH0, as
and Feynman's
1
Operational Calculus
well as the imaginary resolvent [/ + it H o ] - , associated with the free (or unperturbed)
Account: ns000601
 SPECTRAL THEOREM FOR UNBOUNDED SELF-ADJOINT OPERATORS 153
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Hamiltonian H0 = — 1/2A- Further, in Section 10.4, we use the representation theorems

to define the notion of "form sum" of (suitable) self-adjoint operators and apply this
to the study of the Hamiltonian H — H 0 +V, under appropriate assumptions on the
potential V.

10.1 Spectral theorem for unbounded self-adjoint operators

The spectral theorem for self-adjoint operators and its associated functional calculus are
among the most powerful tools of linear analysis. A great deal has been written on this
subject. The 1064 page second volume of Dunford and Schwartz [DunSc2] is entitled
Spectral Theory: Self Adjoint Operators in Hilbert Space. Further, much useful material
related to this topic can be found in Volumes I and II of Reed and Simon [ReSil, 2].
(Recall that the notion of self-adjointness for unbounded operators was introduced and
briefly discussed in Section 9.6, especially Definition 9.6.4, which the reader may wish
to review at this point.)
Multiplication operators
We start with two propositions which provide a large class of examples of self-adjoint
operators. These examples are of interest in connection with many concrete problems,
but, as we will see soon, there is a more essential reason for interest in them: they are,
in a sense to be made precise, universal.
Let (fi, A, /u) be a finite measure space and let Ti. = L 2 ( u ) = L 2 (£2, A, M) be the
space of C-valued square-integrable functions on Q. Let / be a measurable, M-valued
function on £2 which is finite /^-a.e. Let M — Mf be the operator on L 2 ( u ) of multipli-
cation by the function /; that is,

where

Proposition 10.1.1 Suppose that f is R-valued and belongs to L°°(u). Then Mf is a

bounded self-adjoint operator on L 2 ( u ) and \\Mf\\ — ||/||oo
Exercise 10.1.2 (a) Prove Proposition 10.1.1.
[Hint: First show that Mf € £(L2 (/*)). Then, to establish self-adjointness, it suffices by
under U.S. or applicable copyright law.

Proposition 9.6.8 to prove that Mf is symmetric.]

(b) Further, show its converse; namely, if Mf is bounded, then f e L°°(/L/,).
Proposition 10.1.3 Suppose that f is ^-valued and finite \jL-a.e. Then Mf with domain
given fey (10.1.2) is a self-adjoint operator on L2((j,).
Proof It is easy to show that Mf is symmetric: Let <j>, ty e D(Mf). Then

Hence Mf is symmetric and so Mf c Mt. To prove that Mf is self-adjoint it now

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
suffices to show
UNIVERSIDAD that FRANCISCO
DISTRITAL MJ c JOSEMf. DEAccordingly,
CALDAS let TJT € D(Mp. We must check that
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 154 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

fw e L 2 (u); that is, we D(Mf). For each positive integer N, let

Using the monotone convergence theorem, Proposition 10.1.1, and the fact that TN0 e
D(Mf) to justify the second, fifth, and sixth equalities, respectively, we can write

Using (10.1.5), the fact that TNf is bounded and R-valued, and Proposition 10.1.1, we
can now write

Now using the monotone convergence theorem and (10.1.6) to justify the first and third
equalities, we obtain
under U.S. or applicable copyright law.

Hence fw e L 2 (u) and so we D(Mf) as desired.

Remark 10.1.4 The exact counterpart of Propositions 10.1.3 and 10.1.1 holds for
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
C-valued (rather than R-valued) functions f, provided we substitute "normal" for
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
"self-adjoint" operator
AN: 98476 ; Johnson, inW.,
Gerald theLapidus,
statement
MichelofL..;
these results.Integral
The Feynman In particular, if the function
and Feynman's
Operational Calculus
Account: ns000601
 SPECTRAL THEOREM FOR UNBOUNDED SELF-ADJOINT OPERATORS 155
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

/ is C-valued, then M*f, = M-j and hence Mf is a normal (usually unbounded) oper-
ator; here, f denotes the complex conjugate of f. Further, if in addition, f belongs to
L°°(u), then Mf is a bounded normal operator, with norm given by \\Mf\\ = ||f||oo-
[Note that MfM*f = M*fMf = (M |f|2 ) since f f = f f (= \ f \ 2 ) . ] We will use this
simple extension of Propositions 10.1.1 and 10.1.3 (the proof of which can be found
in [Rul, pp. 303-305]) in several places in the book, particularly in Sections 11.6 and
15.2.H.
Definition 10.1.5 Let(n, A, u) be a measure space and let f : n -> R be a measurable
function. We say that n e R is in the essential range of f if and only if for every
€ > 0, u({w e n : f(w) € (A - E, A + €)}) > 0.
We simply state the next proposition which identifies the spectrum o-(Mf) of the
operator Mf. (See, e.g., [Rul, pp. 273 and 303].)
Proposition 10.1.6 Suppose that f is an R-valued measurable function which is finite
u-a.e. Then a ( M f ) equals the essential range of f. In particular,
(i) a(Mf) c R,
and hence
(ii) C \R c p(Mf).
Also, if in addition f e L°°(pi), then

Remark 10.1.7 If A is any (unbounded) self-adjoint operator, we also have a (A) c R

and henceC\R c p(A) = C\0(A). Moreover, if A isbounded, a(A) C [-||A||, ||A||].
This follows immediately from the spectral theorem (Theorem 10.1.8) and Proposi-
tion 10.1.6. (The definition of the resolvent set p(A) and of the spectrum cr(A) of an
unbounded operator A is provided in Definitions 9.3.1 and 9.3.3, respectively.)
Three useful forms of the spectral theorem
We will now state several theorems which will describe related but somewhat different
aspects of the spectral theory for (unbounded) self-adjoint operators. Our reference here
is [ReSil, §VIII.3, pp. 259-264].
under U.S. or applicable copyright law.

Theorem 10.1.8 (Spectral theorem: multiplication operator form) Let A be a self-

adjoint operator on a separable Hilbert space "H with domain D(A).
Then there exist a finite measure space (Q, A, u), a unitary operator U : H ->
2
L ((u), and a measurable, R-valued function f on n which is finite u-a.e., so that
(i) h e D(A) if and only if f(-)(Uh)(-) e L 2 (/u) (i.e. (Uh)(-) e D(Mf))
and
(ii) If0 6 U[D(A)], then ( U A U - 1 0 ) ( a ) ) = /(o>)<A(w) = (M/</>)(o>).
Remark 10.1.9 The formula in (ii) above can be abbreviated to UAU - 1 = Mf. Also,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
for UNIVERSIDAD
h 6 D( A)DISTRITAL
we can FRANCISCO
write A JOSE
= U-1 Mf U. We say that A is unitarily equivalent to Mf.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 156 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Thus Theorem 10.1.8 says that all self-adjoint operators are multiplication operators—
in the sense that every self-adjoint operator is unitarily equivalent to an operator of
multiplication by an R-valued function on some finite measure space L 2 (u).
It is quite easy to guess correctly how to form functions of the operator of multipli-
cation by /; for example, eitMf = Meitf. Since, by Theorem 10.1.8, every self-adjoint
operator is (unitarily equivalent to) a multiplication operator, this gives us a way of form-
ing functions of an arbitrary self-adjoint operator. The resulting "functional calculus" is
rich and extremely useful.
Definition 10.1.10 Let A be a self-adjoint operator on the Hilbert space H. For any
bounded, C-valued Borel measurable function g on a (A), let

where f and U come from Theorem 10.1.8. We will also write

Theorem 10.1.11 (Spectral theorem: functional calculus form) Let Abe a self-adjoint
operator on a separable Hilbert space H with domain D(A). Then there is a unique map
$ from the bounded Borel measurable functions on a (A) into £.(H) so that
(i) <t> is an algebraic *-homomorphism (i.e. ^(g1+g2) = $(g1) + ®(g2)> ®(cg) —
c<$(g), $(g) = $(g)*, O(g1g2) = *(g1)O(g2)).
(ii) $ is norm continuous; in fact, \\^>(g)\\C(H) < \\8\\oo '•— sup \g(x)\.
x€<r(A)
(iii) g > 0 implies that $(g) > 0. (Note: < > ( g ) > 0 means that ($(g)h, h) > 0 for
all h 6 H.)
(iv) If \ is an eigenvalue for A with eigenvector h (i.e. Ah = \h with h e D(A)),
then g(X) is an eigenvalue for <t>(g) = g(A) with eigenvector h (i.e. $>(g)h =
g(A)h = g(X)h).
(v) Ifgn(x) —*• g(x)forallx e cr(A) and if the sequence {\\gn\\oc} is bounded, then
®(gn) — gn(A) —> g(A) = <t>(g) in the strong operator topology on C(H) (i.e.
for every fixed h € H, gn(A)h —> g(A)h in H, as n —> oo).
(vi) Let [gn] be a sequence of bounded Borel functions on er(A) with gn(x) —> x as
under U.S. or applicable copyright law.

n —> oofor each x € a(A) andwith |g n (x)| < | x | f o r all x and n. Then, for any
h 6 D(A), <b(gn)h = gn(A)h -»• Ah as n -+ oo.
Proof We will prove (iii) and one part of (i) to help the reader get some feel for this
result. First we show that 3>(g1 . g2) = *(g1)O(g2):

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
as claimed.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 SPECTRAL THEOREM FOR UNBOUNDED SELF-ADJOINT OPERATORS 157
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Next we show that g > 0 implies that <J>(g) > 0. Let h e H. We must show that
(O(g)h, h) > 0. (For clarity, we denote by (•, •)« and (• , .) L2(u) the inner products in
H and L 2 (u), respectively.) But

where the final equality comes from the fact that g > 0.
Remark 10.1.12 (a) (Unitary groups) The function gt(u) = e-itu is bounded on all
of R and so is certainly bounded on a(A). Hence <b(gt) = gt(A) = e-itA is defined.
Further, this function of the operator A has the usual properties of exponential functions.
For example, since gs(u)gt(u) — e-itu = e-iltu = e-'(s+t)u = gs+t(u), it follows from
Theorem 10.1. ll(i) that

Moreover, also by Theorem 10.1.ll(i), 3>(g,)* = 4>(g,) l so that <t>(g,) = e~"A

is unitary. Hence {e~l'A : t e M} is a unitary group with generator —iA. Recall that
according to Stone's theorem (Theorem 9.6.11 above), all unitary groups arise in this
manner.
(b) (Self-adjoint semigroups) The function h,(u) = e~lu is bounded on a (A) if
and only ifcr(A) is bounded from below. If the spectrum of the operator A is bounded
below, then T(t) := 4>(/j r ) = e~'A is a (Co) semigroup of operators with generator
—A. Further, since clearly, by Theorem lO.l.ll(i), each operator e~'A is self-adjoint,
[e~~'A : t > 0} is a self-adjoint semigroup (with generator —A). In fact, it follows from
Theorem 9.6.7 (established in Exercise 10.1.13 below) that all self-adjoint semigroups
arise in this way.
Moreover, if in addition, a (A) c [0, +00), then {e~'A : t > 0} is a self-adjoint
contraction semigroup. Conversely, it follows from Theorem 9.6.7 that all self-adjoint
contraction semigroups are obtained in this manner. (See also Exercise 10.1.13.)
under U.S. or applicable copyright law.

(c) Many other useful functions of operators can be formed. For example sin A and
cos A make sense and sin2 A+cos2 A = I. It is important that the functional calculus for
self-adjoint operators extends beyond continuous functions. To illustrate this point, let
g(u) — XE (w) where E € B(K). (Here, XE denotes the characteristic function of E; that
is, x£ (x) — 1 ifx 6 E, and 0 otherwise.) Note that g2 = g and ~g = g. It follows from
these properties and Theorem W.l.ll(i) that <t>(g) — XE(^) satisfies tt|2(g) = <t>(g)
and <t>*(,g) = ^(g); that is, 3>(g) is a self-adjoint projection (and so is an orthogonal
projection onto its range).
(d) Part (v) of Theorem 10.1.11 is sometimes extremely useful. It permits, in the
appropriate setting, a conclusion of strong operator convergence from the pointwise
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
convergence of ordinary
UNIVERSIDAD DISTRITAL functions.
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 158 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(e) Theorem 10.1.11 is essentially Theorem VIII.5 from [ReSil, p. 262] except that
we require the functions g to be bounded just on a (A) rather than on all o/R. This more
general form of the result is relevant to cases of interest. The function ht(u) — e~'u in
part (b) of these remarks is definitely not bounded on IR but usually is bounded on a (A)
in situations of physical interest The fact that the more general theorem is valid can
be seen by examining the proof or by checking various other sources such as [DunSc2,
§X//.27- The functional calculus extends to unbounded functions g(A) of the self-adjoint
operator A as we will see below.
(f) The functional calculus continues to work smoothly for functions of any finite
number of self-adjoint operators as long as the operators commute. As soon as they fail
to commute, this beautiful and useful functional calculus collapses (even in the presence
of self-adjointness). Feynman 's operational calculus [Fey8], to be discussed later on in
Chapters 14-19, is an approach to a functional calculus for noncommuting operators.
(g) Let H and 1C be Hilbert spaces and suppose that U is a unitary operator which
maps H onto K.. Let Abe a linear operator from D(A) (C H) to "H and suppose that
B = UAU~\ where D(B) = {k € 1C: U~lk € D(A)}; i.e. A and B are unitarily
equivalent. Then A and B must share any property which either of them possesses and
which can be formulated purely in terms of the abstract Hilbert space structure. Hence,
when a self-adjoint operator A is unitarily equivalent to a multiplication operator Mf as
in Theorem 10.1.8, the purely Hilbert space properties of A are completely determined
by the corresponding properties of Mf. With this in mind, the unitary operators U and
U~l are often ignored and A is studied just by studying Mf. We will sometimes take
advantage of this point of view in later chapters.
The following problem—which supplements Remark 10.1.12(b) and provides a proof
of Theorem 9.6.7—is of interest in its own right and will also be used later on.
Exercise 10.1.13 (a) Recall that we say that the self-adjoint operator A is bounded
below (or semibounded) and write A > c if there exists c € E such that (A<j>, </>) >
c\\<j)\\2,for all (j> 6 D(A). Show that we then have a (A) c [y A> +00), where

!A
(b) Deduce from (a) that ifA>c, then for all t > 0, e € C(H) and satisfies the
"stability condition "
under U.S. or applicable copyright law.

(c) If follows from (a) and (b) that if A > 0 (i.e. (A</», (/>) > 0 for all (/> € D(A)),
then (i) a (A) C [0, +00] and (U) \\e~~'A\\ < 1, for all t > 0. Establish the converse;
precisely, show that if either one of conditions (i) or (ii) (with e~'A self-adjoint) holds,
then A > 0.
(d) Prove Theorem 9.6.7.
[Hint: In addition to (a) and (b), one can use, for example, Theorem 9.3.6 (or 9.4.2) and
Proposition 9.4.1.]
EBSCO
TherePublishing : eBook Academic
is an alternative Collection
version (EBSCOhost)
of the spectral- printed
theoremon 6/8/2017 5:35 PM via operator
in multiplication
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
form
AN: which
98476 ; we nowGerald
Johnson, present. For ourMichel
W., Lapidus, purposes,
L..; Thethe earlier
Feynman result,
Integral andTheorem
Feynman's 10.1.8, and
Operational Calculus
Account: ns000601
 SPECTRAL THEOREM FOR UNBOUNDED SELF-ADJOINT OPERATORS 159
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 10.1.14 below can be used interchangeably; we will sometimes use the latter
because it is simpler conceptually. (We should mention that the new version plays an
essential role after some further analysis if one wishes to study "multiplicity theory"
and to provide a "complete set of unitary invariants" for a self-adjoint operator A; see
[ReSil, pp. 226-234 and p. 262] or [DunSc2].)
Theorem 10.1.14 (Spectral theorem: multiplication operator form, concrete version)
Let A : D(A) -> H be a self-adjoint operator on a separable Hilbert space H. Then
there exists N e {1, 2 , . . . ; 00} and positive measures m,..., px on (R, B(R)), with
fj.p(a(A)) = Atp(R) (p = l,...,N) and £)p=i /i p (R) < oo, such that A is unitarily
equivalent to the operator of multiplication by x on the Hilbert space ®^=1 L 2 (M, ftp),
the orthogonal direct sum of the Hilbert spaces L 2 (R, up) for p = 1,..., N. More
precisely, A = t/~' BU, where U : Ji —»• ®^_i Z,2(R, Mp) is unitary and we have

and

We emphasize that when N = oo,(<p\,..., <PN) above should be interpreted as the

infinite sequence (<p1, ( p 2 , . . . ) . and appropriate adjustments should be made throughout.
Note that, unlike Theorem 10.1.8, there is no abstract measurable space involved
in the preceding theorem. Also, the multiplication operator is always the operator of
multiplication by the independent variable.
Theorems 10.1.8 and 10.1.11 dealt with the spectral theorem in its multiplication
operator and functional calculus forms, respectively; further, Theorem 10.1.14 dealt
with a slightly more concrete form of Theorem 10.1.8. We still wish to discuss the
"projection-valued measure form". Following [ReSil, pp. 262-264], we will do this in
paragraph style and then summarize the main facts in Theorem 10.1.16 below.
Given any Borel subset B of R, let
under U.S. or applicable copyright law.

where, as before, XB denotes the characteristic function of B.

The family of operators [Pg : B € B(E)} in L(K) has the following properties:
(i) Each PB is an orthogonal projection (i.e. P2 — PB and Pg = PB)-
(ii) Pa = 0, PR = I.
(iii) If B - (J£Li Bn with Bn n Bm = 0 if n ^ m, then PB - Y%L\ ?Bn, where the
limit involved in the infinite series is taken in the strong operator topology.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
(iv) PBlPB2DISTRITAL
UNIVERSIDAD = Pa\r\B 2-
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 160 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof We indicated why (i) holds in Remark 10.1.12(c). Here we will show in addition
why (iv) holds. Note that xBlr\B2 = X fll ' Xs2 • Hence, by Theorem 10.1.1 l(i),

Definition 10.1.15 A family {PB : B e B(R)) satisfying (i)-(iv) above is called a

projection-valued measure.
For any h e H, the set function B i->- (h, Pgh) is a true R-valued measure on B(M)
which we denote by d(h, P\h). (This notation is justified by setting PX := ^(XC-M *])•)
Given h\,hi e H, B i-» (h\,PBhi) is a C-valued measure on B(R) denoted
d(h\, /)i./?2). Given a bounded Borel function g on o(A), one can show that g(A) is
uniquely defined by the formula

In fact, the measures d(h\, P\ti2) are all supported by <r(A) and so we can rewrite
(10.1.12) for each h e U as

One can show that the correspondence g \-> g(A) above agrees with the map 4> from
Theorem 10.1.11; that is, for all h € H,

Now suppose that g is an unbounded C-valued Borel function and let

Then one can show that Dg is dense in H and that an operator g(A) is uniquely defined
on Dg by the formula
under U.S. or applicable copyright law.

We write symbolically

For/z,,/z 2 e D(A),

If gEBSCO
is Publishing
M-valued, : eBook
thenAcademic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
is self-adjoint.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
We now summarize some of the key facts above.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 161
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 10.1.16 (Spectral theorem: projection-valued measure form) Let H be a sep-

arable Hilbert space. There is a one-to-one correspondence between self-adjoint opera-
tors A on H and projection-valued measures {PB : B e B(E.)} on H, the correspondence
being given by

If g is an R-valued Borel function on R, then

defined on Dg (see (10.1.14)), is self-adjoint. If g is bounded on a (A), then g(A) is

bounded and coincides with <t>(g) in Theorem 10.1.11.

10.2 Applications of the spectral theorem

In this section, we use the results from Section 10.1 to establish several facts which will be
crucial to us throughout the rest of this book. In particular, we obtain explicit expressions
for the (free) heat semigroup e~'H° (t > 0), its analytic extension e~zli° (Re z > 0), the
unitary group e~itH° (t e R), as well as the imaginary resolvent [/ + itH<)]~1 (t € R).
We also discuss the domain of the free Hamiltonian HQ and identify two of its standard
cores.
The free Hamiltonian HO
We begin by recalling a few facts about the Fourier transform T. If / e L1 (Rd), the
space of Lebesgue integrable C-valued functions on Rd, then

and Tf is in the space of continuous functions on Rd that vanish at oo. (Here, x • y

denotes the inner product of x and y in Rd.) Similarly, the inverse Fourier transform
JT-1 is given for / e L1 (Rd) by
under U.S. or applicable copyright law.

When F is restricted to L! (Rd) HL2 (Rd), it preserves I2 -norms and can be extended
to a unitary operator on L1 (Rd). This extension is called the Fourier-Plancherel trans-
form (or L2'-Fourier transform) and is also denoted by T. In fact, f and JF~' are
still given by formulas (10.2.1) and (10.2.2) above except that the integrals involved
must be interpreted in the mean (see Section 7.5, especially equation (7.5.3)) when
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
/f fc
aUNIVERSIDAD
iL,' Att^N \\ iL2
2 /mxA
\K. jDISTRITAL (Rd).
FRANCISCO
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 162 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Next we recall the pleasant relationship between the Fourier transform and differen-
tiation. Let f e D = D (Rd), the space of C-valued infinitely differentiable functions
of compact support on Rd. Then

It follows that Hence

and

where 1/2|| . ||2 denotes the operator M —M1/2||y||22 of multiplication by the function
y->-1/2||y||2.Now, the operator M with domain

is an unbounded, nonnegative self-adjoint operator on L2 (Rd) with spectrum a(M)

equal to the essential range of the function 1/2|| . ||2; that is, a(M) = [0, +00) (see
Proposition 10.1.6).
We can use the right-hand side of (10.2.5) to define an extension HO of — 5 A. Simply
take

and let
under U.S. or applicable copyright law.

Since F is a unitary operator, we conclude that H0 is an unbounded, nonnegative self-

adjoint operator on L2 (Rd) wither (Ho) — [0, +00).
Remark 10.2.1 If we interpret the derivatives involved in the distributional sense, then
we can write

over D(Ho). Any function f € L,1 loc(R.d) defines a distribution with D = D (Rd) =
Coo00 (RdPublishing
EBSCO ) as the :associated
eBook Academic Collection
space (EBSCOhost)
of test functions. - printed 2
(Rd) 5:35
SinceonL6/8/2017 C L,PM 1via(Rd), every
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
L2-function
AN: 98476 ;can be differentiated
Johnson, infinitely
Gerald W., Lapidus, oftenTheasFeynman
Michel L..; a distribution.
Integral and(For some discussion
Feynman's
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 163
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

of those and related facts, see Section 11.2.) It can be shown that D(H0) as given in
(10.2.7) can also be described by

and that (10.2.9) then holds as well. The reader familiar with the theory of Sobolev
spaces will recognize that D(H0) = H2 (Rd), the Sobolev space of it-functions with
first and second distributional derivatives also in L2 [Ad, Bre2, Kes].
The heat semigroup and unitary group
The operators e - z H 0 , where z is a complex number with Re z > 0, will play a crucial role
in our development. The spectral theorem (see especially Theorems 10.1.8 and 10.1.11),
the self-adjointness of HO and the formula (10.2.8) tell us immediately that

We know then from (10.2.11) that e-zH0 is an everywhere defined operator on L2 (Rd)
which satisfies

since the function u (-> e-zu is bounded in absolute value by 1 on a (H0) = [0, +00).
It further follows from the spectral theorem (specifically, Theorem 10.1.1 l(i)) that

Formula (10.2.13) can also be argued directly from (10.2.11) as follows:

under U.S. or applicable copyright law.

It is not hard to show that the operator on L2 (Ed) of multiplication by e 2"'"

is strongly continuous as a function pf z for z e C+. (We remind the reader that
C+ := {z e C : Re z > 0}. Further, C+ denotes the closure of C+; namely, C+ :=
{z e C : Re z > 0}.) Since F is unitary it then follows easily from (10.2.11) that
the operator e-zH0 is strongly continuous as a function of
When z is purely imaginary, say z = it, t e R, we see from (10.2.11), (10.2.13) and
(10.2.14) that e-it H0 is a strongly continuous unitary group. Of course, Stone's theorem,
Theorem 9.6.11 (see also Section VIII.4 of [ReSil], especially Theorem VIII.8) tells us
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
thisUNIVERSIDAD
along with the fact
DISTRITAL that —JOSE
FRANCISCO i HODEisCALDAS
the generator of this group.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 164 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

When z — t is real and nonnegative, the function e 21/2||.||2takes values in the interval
(0, 1], and so it follows that the operators e-tHo, t > 0, are nonnegative and self-adjoint.
By(10.2.12), (10.2.13), and(10.2.14), { e -t Ho : t > 0} is a (C0) contraction semigroup.
One can also show that this semigroup has generator — HO.
Formula (10.2.11) is a rather explicit and, as we have already seen, useful formula
for e~zH°. We now want to discuss an even more explicit formula which will be used
many times as we continue. We will need the formula for the Fourier transform of the
Gaussian functione~3/2||.||2, where z e C+. We begin with the case d = 1.
The formula

can be found in many places (see, e.g., [ReSi2, p. 4] or [SteWe]). (Alternately, one can use
the fact that the standard normal probability density function is its own Fourier transform
and then obtain (10.2.15) by a change of scale.) Rewriting (10.2.15) as

one can argue that both sides of (10.2.16) exist and are analytic for z e C+, and hence
their agreement for z > 0 implies their agreement throughout C+. (Note: When z e C+,
we interpret z1/2 so that Arg(z1/2) < n/4.) It is easy to extend the formula (10.2.16) to
Rd and to see that the extension is analytic in z, z e C+:

We will actually need the inverse Fourier transform of the function above. However, we
under U.S. or applicable copyright law.

see from (10.2.1) and (10.2.2) that

Hence (10.2.17) and (10.2.18) yield

In our discussion of explicit formulas for the operators e-Ho, we will use the space
S ofEBSCO Publishing
functions of: rapid
eBook Academic
decreaseCollection
and a (EBSCOhost)
few facts -about
printedthe
on action
6/8/2017 of
5:35F PMon
viathis space.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
Before recalling the definition of S, we review the multi-index notation.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 165
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Let a = (a\,..., Ud) be ad-tuple of nonnegative integers and set |a| = a\-\ ha,/.
Let x = (xi,..., Xd) e Rd. We set

Definition 10.2.2 The space S = S (Rd) of functions of rapid decrease (or Schwartz
space) is the space of infinitely differentiable C-valued functions <p on Rd for which

for all multi-indices a and B.

Note that S is dense in L2 since T> is dense in L2 and T> c S. The function
g- 2II 'II ^ Re j > 0, is an example of a function in S \ D.

Remark 10.2.3 We mention here some central facts about S which we will not need
below. The first is that the functions ||.|| a,B are seminorms and that the vector space
S (Rd) is a Frechet space under the topology given by these seminorms. It is often more
convenient for technical reasons to replace an initial collection of seminorms with an
equivalent directed family. This is done in the present setting by defining for <p e S and
all nonnegative integers k and m,

The dual S' = S' (Rd) of the space S = S (Rd) is called the space of tempered
distributions (or Schwartz distributions) on Rd. A great deal more information about S
andS' can be found in many places including [Schw2, ReSil,2, Fol2J.
The Fourier transform is a beautifully behaved object on S (and also, in fact, on
S'). The following result (which can be found in [ReSi2, Theorems IX.1 and IX.3, pp. 3
and 6]) includes the facts that we will need below. Recall that the convolution of two
under U.S. or applicable copyright law.

functions f and g is defined by

Theorem 10.2.4 The Fourier transform f is a linear bicontinuous bijection from S (Rrf)
to S (Rd). Its inverse map F-1 is the inverse Fourier transform. Further, for every
<p, & € S (Rd),

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 166 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We are now prepared to establish the desired formula for ip e S and z e C+. Using
(10.2.11), Theorem 10.2.4, (10.2.24), and (10.2.19), we can write

In summary then, for <p e S (Rd) and z e C+,

Theorem 10.2.5 (Free complex heat semigroup) Let z € C+ and take cp e L2 (Rd).
Then formula (10.2.25) holds for Leb.-a.e. x e Rd.

Proof The function Ez(u) := (2nzrd/2e~ 2z e S(T&.d) C L1 ( R d ) . Hence, by

Young's inequality [Fol2, p. 232], Ez * <p e L2 and \\EZ * <p\\2 < \\Ez\h\\tp\\2. It follows
easily that the map <p i-> £z * <p is a bounded linear operator on L2 (R d ).
Now take a sequence {<pn} from S such that [\<pn — <p\\2 ->• 0. Then \\e~zfl°<pn -
e~*H°<p\\2 -> 0, and, by our remarks beginning the proof, \\EZ * <pn ~ Ez * <p\\2 -> 0.
By passing to a subsequence if necessary, we can also insure that each of the sequences
of functions above converges Leb.-a.e. to its limit function. We simplify notation by
assuming that {q>n} is such a sequence to begin with. Using the limit just discussed as
well as the version of formula (10.2.25) which has already been proved, we have for
Leb.-a.e. x,
under U.S. or applicable copyright law.

as desired.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
The case z = t > 0 is especially important.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 167
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 10.2.6 (Free heat semigroup) For w e L2 (Rd) and t > 0, the formula

holds for Leb.-a.e. x e Rd. Further, forwp e D(H0) (see (10.2.7) and (10.2.10)) and
t > 0, the function

gives the solution, in the sense discussed in Section 9.1, to the heat equation

with initial data w(.) (and potential V = 0).

Proof The assertion regarding (10.2.26) is just a special case of Theorem 10.2.5. The
claim regarding the heat equation follows from Theorems 9.1.4 and 9.1.14 and the fact
that e t ( - H o ) is a (Co) semigroup on L2 (Rd) with generator -H0.
The semigroup e-tHo, t > 0, is called the (free) heat semigroup. Later, we will also
be interested in "heat semigroups" e-t(H°+v) t > 0, which involve a nonzero potential
(or interaction term) V.
Next, we wish to establish formula (10.2.25) for w e L1 (Rd) n L2 ( R d ) and z = it
purely imaginary but not equal to 0. Recall from (10.2.14) that the operator e-zho is
strongly continuous in C+. Now let {zn} be a sequence in C+ such that zn -> it, t = 0.
By the strong continuity just noted above we have that ||e- ZnH ° w — e-it H °w||2 ->• 0
as n -> oo. By dropping to a subsequence if necessary (which we will not indicate
in our notation), we can also insure that (e~ z n H °W)(x) —> (e-itHo W ( x ) for Leb.-a.e.
x e_R d . On the other hand, since w e L1 and|g-| *-«x-u| 2/2z|<1foranx,u e Rd and
z e C+ \ {0}, we see from the dominated convergence theorem that
under U.S. or applicable copyright law.

Now Theorem 10.2.5 along with the facts discussed above allow us to finish this case
(namely, z = it and w e L1 n L2) by taking limits:

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 168 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Finally, we wish to establish (10.2.25) for z = it and <p € L2 (W1). Actually, when
<p e L2 \ L', it is not exactly the same formula; the integral on the right-hand side must
be interpreted in the mean. We begin by showing that the appropriate integral in the mean
exists.
First note that I k - M | | 2 = (x - u) • (x - u) = \\x\\2 + \\u\\2 -2x • u. Thus,

'ii-ir ,
Now since <p(-)e 2' e L , it has a Fourier-Plancherel transform. Hence, the integral

!M|2
7 r ' i
has the L -function F\ <p(-)e it \ ( x )
->
as its limit in L -norm as r —> oo. Thus, by
i \\x ii2 r i||.||2i
2 d/2
(10.2.31), the L -function(it)~ e 2t F\<p(-)e 2t \(x/t) is the limit in L 2 -normas
r —> oo of the integral

But this is precisely what is meant when one writes

and says that the integral on the right-hand side of (10.2.32) is to be interpreted in the
mean.
Now if we let

then \\(p - <prh ->• 0 as r -+ oo. Hence \\e~itH°(p - e-itH0lVrll2 -* 0 as r -* oo. Also
under U.S. or applicable copyright law.

(10.2.30) holds for the function (pr since <pr 6 L1 n L2. Putting the facts above together,
we can write the following where the limits involved are in L2-norm and the last integral
is interpreted in the mean:

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 169
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In the following theorem, we (i) summarize what we have proved above regarding
the explicit representation of the unitary group e-it H0, and (ii) observe that e-it H 0 (p,
(p 6 D(H0), provides the solution to the Schrodinger equation with initial data <p for the
free particle.
Theorem 10.2.7 (Free unitary group) For any <p e L2 (Rd) and any nonzero t e K,

where l.i.m. indicates that the integral on the right-hand side of (10.2.33) is to be
interpreted in the mean; i.e. it is the L2-limit as r —> oo of the integral over the r-ball
of the same expression. When tp e L1 (R d ) D L2 (R d ), we can write

where the right-hand side of (10.2.34) is an ordinary Lebesgue integral.

Also, fort 6 Randy e D(H0) (see (10.2.7) and (10.2.10)) with \\p\\2 = 1, the
function

gives the solution in the sense discussed in Section 9.1 (see also Definition 9.6.9(ii)) to
the Schrodinger equation

with initial probability amplitude (p for the free particle (i.e. potential energy V = 0)
in Rd.

The function ( 2 n i t ) - d / 2 e 2 r appearing in (10.2.33) and (10.2.34) is often called

the free propagator in the physics literature. (Mathematically, it can be viewed as the
fundamental solution—based at the point x e Rd and evaluated at the time t € R—of the
(distributional) Schrodinger equation (10.2.36), with initial condition <p(u) — S(x — u),
under U.S. or applicable copyright law.

the Dirac distribution at x; see, for example, [Hor3, Tre, Schw2] and [ReSi2, §IX.5].)
Note that while we do not have the representation (10.2.33) when t = 0, the operator
e-it H0 is simply the identity operator in that case.
Finally we remark that some authors take the free Hamiltonian to be two times our
HQ (so that it corresponds to — A instead of — j A). When this is done, t on the right-hand
side of (10.2.33) and (10.2.34) and in many earlier places in this section (as well as in
(10.2.52) below) gets replaced by It.
As mentioned earlier, our explicit formulas for e-z H0 and, in particular, for e-tH0,
willEBSCO
be used many times in the sequel. We illustrate the usefulness of 5:35 these
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017H<>
formulas here
PM via
by UNIVERSIDAD
giving an DISTRITAL
explicit FRANCISCO
asymptotic JOSE formula
DE CALDAS for the free evolution e~" .
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 170 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 10.2.8 LettpeL2 (Mrf). Then

Proof We know that the Fourier transform T is a unitary operator on L2. Using this, it
is not hard to show that the scaled transform t~d/2(F<p)(!j-) is also unitary. Since the
d 2
'INI 2
operator of multiplication by i~ ! e 2t (jF^O) js itself a unitary operator, we see
that the operator Wt defined for y e L2 by

is also unitary. Now we know that e "H° is unitary as well, and so it is not hard to see
that it suffices to establish (10.2.37) for <p in the dense subset S of L2.
Let <p € S. By Theorem 10.2.7 and its proof, especially (10.2.32), we see that

Hence, by (10.2.38) and (10.2.39),

and so
under U.S. or applicable copyright law.

where the inequality in (10.2.41) follows from the estimate

Hence the theorem is established.

EBSCO Publishing
Theorem : eBook
10.2.8 Academic
is given in Collection (EBSCOhost) IX.
[ReSi2, Theorem - printed on 6/8/2017
31, pp. 60-61]5:35 PM via
where its physical
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
interpretation is also Gerald
AN: 98476 ; Johnson, discussed.
W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 171
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Standard cores for the free Hamiltonian

Next we wish to show that the operator Ho|S and H0|D are essentially self-adjoint and that
f/0|S = //o|X> = HQ. (Here, //o|5 and HQ\T> denote the restriction of HO to S and
T>, respectively.) We will make use of two results found in [ReSil, Theorem VIII.3
and Corollary, pp. 256-257]. We now state both the theorem and its corollary. (At this
point, the reader may wish to review Section 9.6, especially Definition 9.6.4(iii), (v),
where the notions of self-adjointness and essential self-adjointness were introduced.)
Theorem 10.2.9 (Criteria for self-adjointness) Let T be a symmetric operator on a
Hilbert space H. Then the following three statements are equivalent:
(i) T is self-adjoint.
(ii) T is closed and Ker(T* ± i) = {0}.
(iii) Ran(T ±i) = H.
Here, Ker(T* ± z) denotes the kernel of T* ± i and Ran(r ± i) denotes the range of
T±i.
In the corollary, we replace "self-adjoint" in (i) with "essentially self-adjoint". The
corresponding changes in (ii) and (iii) turn out to be what one might guess.
Corollary 10.2.10 (Criteria for essential self-adjointness) Let T be a symmetric oper-
ator on a Hilbert space Ti.. Then the following are equivalent:
(i) T is essentially self-adjoint; that is, it has a unique self-adjoint extension
(necessarily its closure T).
(ii) Ker(r* ± i) = {0}.
(iii) Ran(T ± j) are both dense in "H.
We first show that HQ\$ is essentially self-adjoint and then use that fact as a key
element in the proof that Hop has the same property.
Theorem 10.2.11 (i) The operator H0|S " essentially self-adjoint.
(ii) <S is a core for HQ.
Proof (i) By Corollary 10.2.10, it suffices to show that Ran(Ho|S i 0 are both dense
in L2 (Md). Since the argument that we will make does not really depend on the sign of
i, we will just show that Ran(H0|S + 0 is dense in L2. Now, from (10.2.7) and (10.2.8)
under U.S. or applicable copyright law.

we know that

where

Clearly, S c D(//o), and so what we need to show is that the set of functions
{F~l [\ INI 2 'F(f)\ + // : / 6 5} is dense in L2. Since S is dense in L2, it suffices to
show that for any g e S, there is an / e S such that J""1 [| || • ||2 F(f)] -\-iF~1 [F(f)] =
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
g or, equivalently,
UNIVERSIDAD that
DISTRITAL ||2 +DEi )CALDAS
(||| • JOSE
FRANCISCO F ( f ) = f(g). By Theorem 10.2.4, h :- f(g) is
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 172 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

just another element in S. Hence, we need to know that there is f e S such that

Since F maps S onto S (Theorem 10.2.4), it is enough to show that the function

belongs to S. However, since h e S, it is not difficult to convince oneself that the func-
tion in (10.2.43) is infinitely differentiable and satisfies (10.2.21) for all multi-indices
a and ft. [First note that |(||y||2 + 2i)-1 <1/2for all y € Rd. Next, observe that the
derivatives of the function (10.2.43) involve higher and higher powers of ||y ||2 + 2i in
the denominator and polynomials multiplied by h and its derivatives in the numerator.
Thus 2f+2i e S as claimed.] Hence, there exists / e S such that (10.2.42) holds.
Thus Ran(//Q|5 + i) is dense in L2. Since, similarly, Ran(//o|5 — 0 is dense, we see
that //ojS is essentially self-adjoint as claimed.
(ii) Since //o|5 is essentially self-adjoint by (i), we know that #o|S has a unique
self-adjoint extension, necessarily the closure of HQ\S (see Definition 9.6.4(v)). But HQ
is a self-adjoint extension of HQ\<; and so HQ = //o|S- It now follows from the fact
that HQ = HQ is closed (see Theorem 10.2.9 or Proposition 9.6.2) and from Propositon
9.1.20(ii), that S is a core for HQ. D
Next we will prove a result like Theorem 10.2.11 but for //o|D rather than for HQ\$.
We cannot give the same proof as in Theorem 10.2.11 basically because the Fourier
transform does not map D onto D. We will however use Theorem 10.2.11 in the proof
to follow.
Theorem 10.2.12 (i) The operator H0|D is essentially self-adjoint.
(ii) D is a core for H0.
Proof The key to the proof is to show that HQ\TJ 2 HQ\$. Let <p e S. We seek a sequence
{ifn} from V such that \\<pn — tp\\2 ->• 0 and \]Ho<pn - Ho<p\\2 -> 0. Let w e V be such
that w(x) = 1 for \\x\\ < 1 and 0 < w(x) < 1 for all x e Rd. Let
under U.S. or applicable copyright law.

Note that q>n e T> for all n. Also, it is not hard to show that \\ipn — <p\\2 —*• O a s n —»• oo:
First observe that w (^) —>• 1 for all jc e M^. Then we can write

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 173
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the next to last equality follows from the dominated convergence theorem with
4|<p(jc)|2 serving as a dominating integrable function.
An easily verified "product rule" for the Laplacian of smooth functions on Rd tells
us that

Applying (10.2.45) to <pn in (10.2.44), we obtain

Now A<p is a fixed element of S and so our earlier assertion that \\<pn — <p\\2 -» 0
implies that

Further, since the partial derivatives f j r , . . . , fjr- are all in D, there is a uniform bound,
say N, for all of these functions. Hence, we have

and so

Finally, since Aw e D and so is bounded, one has

Putting together (10.2.47)—(10.2.49), we see from (10.2.46) that || A<pn - A<p|| 2 -* 0

as n ->• oo and so \\Ho<pn — Ho<p\h -»• 0. Since we showed above that \\<pn - y>\\2 -» 0,
it follows that //o|Z> 5 //o|S as we wished to show.
under U.S. or applicable copyright law.

Combining Theorem 10.2.11 and the fact that //o|£> => #0|S* we can now easily
show that //op = HQ. Of course,H0|D> <= #o|S and so HO\DT> Q HQ\S = HQ. Also, since
^o|5 Q HO\T>, we have HO = #o|S c //op- Therefore, HQ\-D = HQ as claimed.
The equality HQ\?> = HO and Proposition 9.1.20(ii) tell us that V is a core for HO
so that (ii) of our theorem is established. (Recall that Proposition 9.1.20 states that an
operator A is closable if and only if it has a closed extension, and that, if A is closable
with closure A, then D(A) is a core for A.)
In order to show that Ho|X> is essentially self-adjoint, we will show (see Definition
9.6.4(v)) that it has a unique self-adjoint extension. Since we know that HQ is a self-
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
adjoint extension
UNIVERSIDAD of HQJ£>>
DISTRITAL FRANCISCO^ JOSE
suffices to let B be an arbitrary self-adjoint extension of
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 174 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

//op and show that B = HQ. Such a self-adjoint extension B of //op is certainly closed
since it is self-adjoint (see Theorem 10.2.9 or Proposition 9.6.2). However, since B is a
closed extension of //op, it must certainly contain the smallest closed extension; namely,
B l> (//op)** — HQ\T> = HO. From this and from the self-adjointness of B and HO, we
deduce that

Thus we have both B ^ HO and B c HO; hence B — HO. Our proof of Theorem 10.2.12
is now complete. D
The functions in D(//o) need not possess a classical Laplacian; on the other hand,
D(Ho) falls well short of including all functions in L2 (R rf ). What can be said about
the functions in D(//o)? We simply state one result along these lines [ReSi2, Theorem
IX.28, p. 55] which is closely related to the Sobolev imbedding theorem [Ad, Bre2, Kes].
Theorem 10.2.13 Let y e L2 (Rd) be in D(//0), the domain of the free Hamilto-
nian HO.
(i) If d < 3, then (p is bounded and continuous and for any a > 0, there is a positive
constant b, independent of<p, such that

(ii) Ifd >4andqe [2, 2d/(d - 4)), then <p e L* (Erf) and for any a > 0, there is
a positive constant b—depending only on q,d and a—such that

Imaginary resolvents
We have discussed the operator HO and the exponential functions e~zH° (z e C+) at some
length. In applying the "product formula for imaginary resolvents", Theorem 11.3.1, to
the "modified Feynman integral" ([Lai, 2], [La6-13], [BivLa]) in Sections 11.4-11.6,
we will also need to consider the resolvents [/ + itHo]~l, t e M. The central result that
is needed expresses [/ + it//o]"1 as an integral operator. We will omit the proof, which
under U.S. or applicable copyright law.

can be found in [ReSi2, Theorem IX.29 and Examples 1,2, pp. 57-59].
Theorem 10.2.14 (Free imaginary resolvent) For all t e M, [7 + itHo]~l e
C(L2 (Rd)). Further, for all t ^ 0, [/ + itH0]~l is the integral operator (with ker-
nel the free "Green's function" G(x, j; t)) given for all <p e L2 (Rd) by

EBSCOthe
where Publishing
integral: on
eBook
theAcademic Collection
right-hand side(EBSCOhost)
o/(10.2.52)- printed
existson as
6/8/2017 5:35 PM via
a Lebesgue integral. The
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
function G(x,y\ t) is known explicitly for all nonzero t andd = 1,2,
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
In particular,
Operational Calculus
Account: ns000601
 APPLICATIONS OF THE SPECTRAL THEOREM 175
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

for t > 0, we have for d = I

and for d = 3,

In (10.2.53a), z1/2 denotes the standard analytic determination of the square root
which is positive for z > 0. Moreover, for t < 0 (instead of t > 0), the expression of
G(x, y; t) in (10.2.53) would be the same as above except for (1 — i)t~1/2 replaced by
(1 + i)\t\~1/2 inside the exponentials.
Finally, ford = 1 or 3 (resp., otherwise), the integral in (10.2.52) converges for all
x (resp., Leb.-almost every x) in Rd.
Remark 10.2.15 (a) Theorem 10.2.14 is in much the same spirit as our earlier Theorem
10.2.7 which dealt with the free unitary group; indeed, there are common elements in
the proofs which both make use of the Fourier transform. There are also differences
between Theorems 10.2.7 and 10.2.14. The integrals in (10.2.33) are oscillatory and
hence are interpreted as integrals in the mean when <p e L2 (Mrf) \ L' (Rd), whereas the
integrals in (10.2.52) are true Lebesgue integrals (and hence are absolutely convergent)
for all (p e L2 (Rd). On the other hand, the kernels in (10.2.33) are familiar elementary
functions for all d, but those in (10.2.52) are expressed for d ^ 1 , 3 m terms ofBessel
functions. (See [La7, p. 3], [Ec, Chapter I], [GliJa, Chapter 7], [Sorn, p. 233], as well
as the comment following Exercise 10.2.16(i) below. For clarity, we leave aside here the
case d = 2 where Green's function has a logarithmic singularity.) However, for d > 4,
they have a similar asymptotic behavior. (See [SilO, p. 70].)
(b) For z 6 C+, [/ + zHo]~l is also an integral operator on L2 (Rd) and its kernel
can be computed via entirely analogous formulas. (See Exercise 10.2.16 below.)
(c) When t = 0, [/ + it HQ]~I does not have a representation of the form (10.2.52)
but [I + it//o]~' = I~l = I is extremely simple in this case.
(d) In [Lai, 2, La6-13] and [ReSi2], for example, the free Hamiltonian is taken to
be two times HO; that is, it is chosen to be equal to —A instead of—A A. In this case, as
was mentioned earlier, t on the right-hand side of (10.2.53) must be replaced by 2t.
Exercise 10.2.16 (i) Show that for every z e C+, the complex resolvent [I + zHo]~l is
under U.S. or applicable copyright law.

a bounded integral operator on L2 (Rd) and calculate its kernel (which can be viewed
as a complex Green's function). In particular, for z = t (resp., z = it), with t ^ 0,
deduce the expression of the (standard) Green's function, the kernel of [I +1 HQ] ~l, and
recover the expression obtained in (10.2.53) for the kernel of [I + itHo]~l.
[Advice: In part (i), you may assume that d = 1 or 3, for simplicity. Further, for d = 3,
for instance, you should find that the kernel of [—A + f 2 ]~' is given by (4jr||jt —
y\\)-ie-S\\x-y\\, for all f e C+. By contrast, for d = D + 2 (D > 1), this same kernel
is given for £ € C+ by

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 176 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where wrj denotes the solid angle in D dimensions, F is the gamma function, and HDn
denotes the Bessel function of the third kind (Hankel function) of order D/2. Hence, in
(10.2.52), G(x, y; t) is equal to (2it) -1 times this expression, where f := (1 — z)A/f
or f := (1 + i)/ J\T\, for t > 0 or t < 0, respectively.)]
(ii) Show that the operator-valued function [I + Z//Q]~' is analytic for z € C+ and
strongly continuous for z € C+.
[Hint: The proof of part (ii) parallels that of Theorem 13.3. L]

10.3 Representation theorems for unbounded quadratic forms

Our interest here is in unbounded operators and forms, but let us begin by recalling the
situation with bounded operators. Let H be a Hilbert space over C. If A e £(7i) and if
we let qA '• T~i x H —»• C be defined by q& (<p, iff) = (up, Ai(/), then it is easy to see that q&
is a bounded sesquilinear form; that is, qA is linear in the first variable, conjugate linear
in the second variable, and satisfies \qA(<p, VOI < I I ^ I I I M I I W I f°r all <P< & ^ "H- The
more interesting fact is that the converse holds; that is, if q : H x H, -> C is sesquilinear
and satisfies \q(<p, VOI < C||^||||^|| for some constant C > 0 and all tp, ty 6 H, then
there is a unique operator A 6 £(W) such that q — qA- The converse result depends on
Riesz's lemma from which it follows quite easily [ReSil, p. 44].
Our main goal is a much deeper result which is in the spirit of the converse result above
but involves semibounded forms and self-adjoint operators. Specifically, we will see that
"if q is a closed semibounded sesquilinear form, then it is the form associated with a
unique self-adjoint operator". This result has proved to be important in the mathematical
treatment of quantum theory, as well as in various other subjects, including Dirichlet
forms, Markov processes and the Feynman integral.

Basic definitions and properties

We will let Ti be a Hilbert space over C throughout our discussion.
Definition 10.3.1 A sesquilinear form is a map q : Q(q) x Q(q) —»• C, where Q(q) is a
dense linear subspace of H, called the form domain of q, such that q is linear in the first
variable and conjugate linear in the second variable. If q(tf>, if) = q(ifr, 9) for every
<p,tj/ e Q(q), we say that q is symmetric. In the following, for notational simplicity, we
will often write q(<p) instead of q(<p, (p) when (p e Q(q). If there exists a number c > 0
such that
under U.S. or applicable copyright law.

for <p e Q(q), we say that q is semibounded (or bounded below). If c can be taken equal
to 0, then q is said to be nonnegative.
Remark 10.3.2 (a) When a sesquilinear form q is restricted to the diagonal of Q(q) x
Q (q), it is called a quadratic form. In fact, the quadratic form determines the sesquilinear
form via the "polarization identity"
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 177
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Partly with this in mind, q acting on all of the product space Q(q) x Q(q) is often itself
referred to as a quadratic form.
(b) If q is semibounded with (10.3.1) holding, then q' defined for <p, i/r e Q(q) by

is clearly a nonnegative sesquilinear (or quadratic) form. Because of the simple relation-
ship between q and q', results which can be established for nonnegative quadratic forms
usually carry over easily to semibounded quadraticforms. Hence proofs for semibounded
forms are often given just for nonnegative forms.
(c) One can show [BachN, Theorem 20.13, p. 369] that a sesquilinear form q is
symmetric if and only if q((f>) e R/or all (p 6 Q(q)- It follows from (10.3.1) that a
semibounded sesquilinear form is necessarily symmetric. (7f is implicit in the definition
of semiboundedness that q((p, (p) is real and (10.3.1) holds.)
Recall that an operator A : D(A) —>• H is said to be closed if and only if its graph is
closed in H x H; further, recall that this is equivalent to requiring that D(A) is complete
under the norm \\tp\\A := \\Aq>\\ + \\tp\\ (see Propositions 9.1.7 and 9.1.8). In order to
study the relationship between self-adjointness and semibounded quadratic forms, we
will need to extend the notion of "closed" from operators to semibounded forms.
Let q be a semibounded quadratic (or sesquilinear) form and let c > 0 be such that
(10.3.1) holds for all tp e Q(q). It is not hard to show that the formula

defines an inner product on Q(q). The associated norm is then given by

Definition 10.3.3 A semibounded quadratic form q as above is called closed if and only
if Q(q) is complete under the norm \\ • \\+\ given by (10.3.4). If q is closed and D is a
subspace of Q(q) which is dense in Q(q) in the norm || • ||+i, then D is called a form
core for q.
Our first proposition gives a necessary and sufficient condition for q to be closed.
We will include the simple proof.
under U.S. or applicable copyright law.

Proposition 10.3.4 Let q be a semibounded (and hence symmetric) quadratic form

satisfying (10.3.1) for all <p e Q(q). Then q is closed if and only if whenever {(pn] is a
sequence in Q(q) such that \\<pn — <p\\ -> 0 and q(<pn — <pm) —>• 0 as n, m -» oo, we
have (p e Q(q) and q((pn — (p) —> 0.
Proof Suppose that q is closed and that both \\(pn — <p\\ -> 0 and q(<pn — <pm) -> 0 as
n,m —> oo. Certainly then

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
as n, m -* oo.
UNIVERSIDAD But (Q(q),
DISTRITAL \\ •JOSE
FRANCISCO ||+i)DE is complete and so there exists (po e Q(q) such that
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 178 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Hence \\<pn —(po\\ —>• 0. Since \\<pn — <p\\ —> 0 also, we see that <po = <p and so <p e Q(q)-
Finally, using (10.3.5) again and the fact that <po = <p, it follows that q((pn — (p) —>• 0 as
n —* oo, as we wished to show.
Next we establish the converse. Let {<pn} be a sequence from Q(q) such that
\\<Pn — <Pm\\+i —>• 0 as n, m —> oo. We seek <p e £>(<?) such that \\<pn — <p\\+\ —> 0
as n -> oo. But ||0>n — ^m ||+i -* 0 implies that

as n, m -» oo. Since the sum of the first two terms on the right-hand side of (10.3.6) is
nonnegative,itfollowsthat||<pn — <pm\\ -> Oasn, m ->• oo. Hence, by the completeness of
(H, || -II). there exists <p e H such that ||<pn-<p|| -> 0. Certainly also(c+l)||<p n -<p m || 2 -»
0, and so, from (10.3.6), q(<pn — tpm) -»• 0 as n, m -»• oo. By our present assumption, we
then have that <p e Q(q)andq(<pn — (p) —> Oasn —> oo. Since we already saw that ||<pn —
401| —» 0, we now have that ||<pn — ^||+i —»• 0. Therefore q is closed, as desire
Let A be an (unbounded) self-adjoint operator on H. There is always a quadratic
form qA associated with A, which we now wish to describe. We remark that q& will
not in general be semibounded. We will identify Q(qA) and q^ in terms of a spectral
representation for A. This spectral representation will involve not only the representing
Hilbert space but also H and a unitary operator U from H onto to the representing space.
It will be convenient notationally to suppress reference to U and just act as though H were
the representing space. (For simplicity, we will assume that Ti. is separable throughout
the rest of this discussion.)
Since A : D(A) —> Ti is self-adjoint, by Theorem 10.1.14, A is (unitarily equivalent
to) multiplication by x acting on ®^ =1 L 2 (K, B(E), /J,P), where N may be any element
from the set ( 1 , 2 , . . . ; oo}, each np is a (positive) Borel measure with fjLp(a(A)) =
HP(K), and Y^^=i Mp(K) < oo. In fact D(A), the domain of A, is then given by
under U.S. or applicable copyright law.

and, for ( < p i , . . . , <PN) e D(A),

Recall that when N = oo, ( < p \ , . . . , (pu) should be interpreted as the infinite sequence
(<P\, <P2, • • •) and appropriate further adjustments should be made throughout. (In order
to better understand the notation in (10.3.7) and (10.3.8) above, the reader may wish to
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
review Theorem 10.1.14, the concrete version of the multiplication operator form of the
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
spectral theorem.)
AN: 98476 We
; Johnson, are W.,
Gerald nowLapidus,
ready Michel
to define Q(<?A)
L..; The andIntegral
Feynman qa. and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 179
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Definition 10.3.5 With the setting as described in the last two paragraphs, we let

and, for (<pi,..., <pN), ( V q , . . . , VAT) '" Q(qA),

Then the form q^ is called the quadratic form associated with A, and we sometimes write
Q(qA) = Q(A). The subspace Q(A) is called the form domain of the operator A. It
follows from this definition that q& (<p, T/0 = (<p, Aifr) for all <p in Q(A) = Q(qA) and
all VT in D(A). (See Proposition 10.3.9 below.) For <p, ifr both in Q(A), some authors
still write qA(<p, VO = (<P, Aifr) even though A need not be defined on Q(A). It seems
to us that this notation is sometimes confusing, and we will avoid it.

Remark 10.3.6 By the multiplication operator form of the spectral theorem, Theorem
10.1.8, A is (unitarily equivalent to) the operator of multiplication by some measurable
^-valued function f acting on L2(/j,) = L2(Q,A, f-C), where (Q,^4,/x) is a finite
measure space. Then (10.3.7) through (10.3.10) are replaced respectively by

All the remaining proofs in this chapter could be rewritten with more concise notation
using this formulation. However, the concreteness of (10.3.7)-(10.3.10) has its own
appeal.
Definition 10.3.7 A symmetric operator A : D(A) -> H is said to be semibounded (or
under U.S. or applicable copyright law.

bounded below) if and only if there exists a number c > 0 such that

for all <p e D(A).

We will show in Proposition 10.3.8 below that if a self-adjoint operator A is semi-
bounded then its associated quadratic form qA is closed. We will make use of the spectral
representation of A as discussed above in the proofs of Propositions 10.3.8-10.3.10. The
case N = 1 is notationally simpler but still contains the essential ideas. We will restrict
ourEBSCO
attention to that
Publishing caseAcademic
: eBook in Propositions 10.3.9 and
Collection (EBSCOhost) 10.3.10.
- printed However,
on 6/8/2017 5:35we
PM will
via illustrate
the UNIVERSIDAD
more complicated (at least
DISTRITAL FRANCISCO JOSEnotationally)
DE CALDAS case N = oo in the proof of Proposition
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
10.3.8.
Operational Calculus
Account: ns000601
 180 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proposition 10.3.8 Let Abe a semibounded self-adjoint operator on the Hilbert space
Ti. and let q& be the quadratic form associated with A. Then qA is semibounded and
closed.
Proof Let c > 0 be such that (10.3.15) is satisfied. It follows from Exercise 10.1.13 that

Hence we have, for p = 1 , 2 , . . . ,

Support(/u,p) c a(A) c [—c, oo).
Therefore, (10.3.9) and (10.3.10) of Definition 10.3.5 can be rewritten in this situation as

and, for (<p), ($) e Q(qA),

It follows readily from (10.3.17) and (10.3.15) that qA is semibounded.

Now let (<pn) = [<p", y>2, • • •} be a Cauchy sequence in (Q(qA), II • ll+i)- Then

By (10.3.16), x + c > Oona(A) andsox+ c+ 1 > 1 oner (A), Hence, it follows from
(10.3.20) that ((<?")} is a Cauchy sequence in 0^° =| L 2 (cT(A), (x + c + \)d^p). Thus
there exists (<p) e 0^ =1 L 2 (a(A), (x + c + l)d^p) such that
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 181
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

on [—c, oo) D o(A). Hence (<p) e ®^LlL2(a(A), \x\d/j,p). But this last space is just
Q(qA) by (10.3.18), and so (<p) e Q(qA)- Finally, using (10.3.21) and the fact that
(<P) £ Q(QA)< we see that qA is closed.
In the next proposition, we show that the form domain of any self-adjoint operator A
always contains the operator domain of A. Further, we will see that for every <p e Q (q&)
and ij/ € D(A), qA(<p, ^) is given by the inner product (<p, Aijf). We will continue to
make use of the spectral representation of A (and to omit reference to the unitary operator
from H onto the Hilbert space involved in the spectral representation).
Proposition 10.3.9 Let A : D(A) -» H be a self-adjoint operator on H and let Q(qA)
and qA be given as in Definition 10.3.5. Then D(A) c Q(qA) and for cp e Q(qA) and
\lr € D(A), we have

Proof As mentioned earlier, we will limit our attention in this proof to the case N = \
of the spectral representation. Let <p e L 2 (R, u) be in D(A). Then by (10.3.7) we have
A<p 6 L 2 (R, u), so that

But then certainly,

Since the function x -> \x\ is bounded on [—1, 1], u. is a finite measure and y e
L 2 (R, u), we also have

It follows from (10.3.23) and (10.3.24) that /R \x\\<p(x)\2dn < oo. Hence, the finiteness
condition in (10.3.9) is satisfied and so <p e Q(qA) as we wished to show.
under U.S. or applicable copyright law.

Finally, let <p e Q(qA) and let ^ e D(A). Then using the spectral representation of
A and (10.3.10), we have

as claimed in (10.3.22).
Note that A was not required to be semibounded in Proposition 10.3.9.
Proposition 10.3.10 Let A be self-adjoint and semibounded. Then D (A) is \\ • \\ +1 -dense
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
in Q(q A).
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 182 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof We again limit our attention to the case N = 1 of the spectral representation.
Recall from Proposition 10.3.8 that qA is semibounded, so that || • ||+1 defined by (10.3.4)
is a norm.
Now let p e Q(qA) so that we have /r |^>(x)| 2 du < oo and, by (10.3.9),
/R \x\\<p(x)\2dfj, < oo. Take <pn := <p • X[-«,ni- Then (10.3.7) implies that <pn e D(A)
for every n since

Also \\<pn— <P\\2 = f\xi>n \<p(x)\2dfj. —>• Oasn —> oo. Hence, to finish showing that \\<pn —
<p\\+\ -»• 0, it suffices to show that q&(<pn - <P) ->• 0 as n ->• oo. But/ M \x\\<p(x)\2dn <
oo as noted above and so

as n -*• oo, as desired.

Exercise 10.3.11 Prove Propositions 10.3.9 and 10.3.10 for N = oo.
Proposition 10.3.10 tells us, in the language of Definition 10.3.3, that the domain of
a self-adjoint semibounded operator A is a form core for qA. Our next proposition says
that more is true: Any operator core for such an A is a form core for qA.
Proposition 10.3.12 Let A be self-adjoint and semibounded. Then any operator core
for A is a form core for qA.
Proof First we note that a self-adjoint operator is necessarily closed (see Proposition
9.6.2(v)). Now let DO be an operator core for A (Definition 9.1.16). By Proposition
9.1.17, DO is dense in D(A) under the graph norm
|h|| := |h|| + \\Ah\\ on D(A). We wish to show that DO is dense in Q(qA) under the
|| • ||+1 norm. Since D(A) is dense in Q(qA) under the || • ||+1 norm by Proposition
10.3.10, it suffices to show that D0 is dense in (D(A), || • ||+1).
Let h e D(A). Since DO is dense in D(A) in the graph norm, there exists a sequence
{hn} in DO such that
under U.S. or applicable copyright law.

Now by Proposition 10.3.9, the Cauchy-Schwarz inequality, and (10.3.25a), we see that

as n —> oo. Using (10.3.25), we see immediately that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
Thus DO is aDISTRITAL
UNIVERSIDAD form core for qA,
FRANCISCO JOSEasDEdesired.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 183
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Let D = D(R), the space of infinitely differentiable functions of compact support

on R. Next, we give an example of a quadratic form q on D x D which is positive
and symmetric but for which there is no semibounded, self-adjoint operator A such that
q((p, ifr) = (<p, At/r) for all <p, iff in T>.
Example 10.3.13 Define q : V x T> -*• C by

Then q is easily seen to be & positive, symmetric quadratic form. However, we claim that
(i) q has no closed extension, and (ii) there is no semibounded, self-adjoint operator A
such that q(<p, ty) = (tp, A\jf)for all <p, iff in T>.
Proof Consider a sequence {$„} from T> such that <pn = 1 on [— ~, i], <pn = 0 on
R \ (— |, |), <pn is increasing on [— |, — i) and is decreasing on [i, |]. Clearly \\<pn —
Q\\2 -»• 0 and <?(<?„ - <pm) = \tpn(0) - pm(0)|2 = 0 -» 0 as n -» oo. However,
q(tpn— 0) = l^«(0)|2 = 1 ->• 1 asn -> oo. It now follows immediately from Proposition
10.3.4 that q has no closed extension. Hence, (i) is established.
Now suppose that A is a semibounded, self-adjoint operator such that q(<p, ^r) =
(#>, At/0 for all <p, i/r in T>. Then q immediately has an extension to a semibounded
quadratic form on D(A) x D(A). By Propositions 10.3.8 and 10.3.9, the quadratic form
qA associated with A as in Definition 10.3.5 is a further extension of q which is closed. But
no such closed extension of q exists as we have shown above. Hence (ii) follows. D
Representation theorems for quadratic forms
We are ready for the main theorem in this development.
Theorem 10.3.14 (First representation theorem) Let q be a closed semibounded
quadratic form. Then:
(a) There exists a unique self-adjoint operator A such that q — qA (i-e. such that
q is the quadratic form associated with the self-adjoint operator A, in the sense of
Definition 10.3.5). Hence, by Proposition 10.3.9, D(A) C Q(q) =; Q(A) and

In particular,
under U.S. or applicable copyright law.

(b)Ifc>0 is such that q satisfies q(<p) > -c\\<p\\2 for all <p e Q(q), then the
operator A from the conclusion of part (a) is semibounded with (<p, Ay) > —c\\<f>\\2 for
all <p in D(A). In particular, if the quadratic form is nonnegative (i.e. c = 0), then so is
the operator A.
Proof The fact recorded in (b) is useful but is a trivial consequence of (a): Let <p € D(A).
By Definition 10.3.5 and Proposition 10.3.9, (<p, A<p) = qA(<p) = q(<p) > -c||<p||2.
We now proceed to establish the key claim made in the theorem; namely, the existence
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
of aUNIVERSIDAD
self-adjoint operator
DISTRITAL A such
FRANCISCO JOSE that q = qA- In the spirit of Remark 10.3.2(b), we will
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 184 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

give the proof under the additional assumption that q is nonnegative; i.e. q(<p,<p) > 0
for every ip e Q(q). (In the more general case when q is semibounded, we simply add
c((p, VO to the right-hand side of (10.3.28) below.)
Since the form q is closed as well as nonnegative, we know (see (10.3.3) and
Definition 10.3.3 with c = 0) that the domain Q(q) of the form is a Hilbert space
under the inner product

We will denote this Hilbert space H+1 and we let H-1 denote the space of bounded
conjugate linear functions on H+1. Let j : H ->• H-1 be the linear mapping of H into
H+1 defined by

Each j (VO is clearly conjugate linear. Also, it follows easily from the Cauchy-Schwarz
inequality and the fact that \\<p\\ < \\<p\\+\ for all (p e H+i, that

Hence j (\jf) is a bounded conjugate linear functional on H+\, as claimed. Next, we show
that j is injective. Let TJt\, fa e H and suppose that j ( f a ) = j ( f a ) . Then for every
(p e H+1, we have

But H+1 — Q(q) is dense in H and so ^r\ = fa as claimed. Finally, it is easy to see
from (10.3.31) that j is continuous. In summary, j is a linear, continuous injective map
from H to H-1. Clearly, the identity map / : H+1 -> H is also linear, continuous and
injective.
Since H+1 is a Hilbert space, every element * in H.+1 defines an element 54* of
H-i via the formula

In fact, by the Riesz lemma [ReSil, Theorem II.4, p. 43], B is an isometric isomorphism
under U.S. or applicable copyright law.

ofH+i ontoW-i.
We are now ready to define an operator B which will turn out to be related in a
simple way to the operator A which we seek. Let R(j) denote the range of the imbedding
j : H -»• H-i discussed earlier, let

and define B : D(B) -» H by

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 185
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The following diagram, along with (10.3.32}-(10.3.34), will help us to keep track of
some of the relationships discussed above:

We will need to know that B is densely defined on "H. In this connection, we begin
by showing that R(j) is dense in ~H-\. If this were not so, then there would be A €H*_\
such that X ^ 0 but for all V 6 H

Now T~i+i is a Hilbert space and H*+i = T~t-\. It follows that ti*_^ = ~H+\ and so there
exists <px 6 T~i+i such that

for all T/f e 71., where the second equality in (10.3.37) comes from the definition of j.
Combining (10.3.36) and (10.3.37), we see that

But (10.3.38) is not possible since <p\ ^ 0. It follows that R(j) is dense in H-\ as we
wished to show.
Since R(j) is dense in W_i and B is an isometric isomorphism of H+i onto H-i,
we can conclude immediately that D(B) is || • ||_|_i-dense in 7i+i. Finally, the norm
inequality ||<p|| < ||<p||i, q> 6 Tt+i, and the fact that H+i = Q(q) is norm dense in
H allow us to show easily that D(B) is norm dense in H: Given h e H and e > 0,
first, take h+i e T-L+\ such that \\h - h+\ \\ < e/2. Next take k e D(A) such that
H/.+,-*||+l < e/2. Then
\\h - k\\ < \\h - h+i || + \\h+1 - k\\ < \\h - h+l || + H/r+i - ftll+i < e/2 + e/2 = e
as desired.
Next we show that B is a symmetric operator. Let <p, ty € D(B) (see (10.3.33)) and
recall from (10.3.34) that B = j~l o B. Also recall that j imbeds H into H-1 = H*+l
via (10.3.35) so that j~l carries Bty e R(j) C Ti-i into an element of H. Then we can
under U.S. or applicable copyright law.

write

Now from (10.3.39) and the symmetric nature of the hypotheses on <p and ^, we see that
(Bi/f, <p) = (B\/r)(<p). Using (10.3.39) and this last fact, we obtain

andEBSCO
so we see that B is symmetric as claimed. Thus B is a densely defined symmetric
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
operator.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 186 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Next we show that B is actually self-adjoint. Let C = B~l o j. Since j maps H into
Ji-\ and fi"1 is an isometric isomorphism of "H-\ ontoW+i c 7, it follows that C takes
H into H. Note that C is every where defined and that C — B~loj — (j~loB)~l = B~l.
We claim that C is symmetric. Let h1,h2 e W.SinceC = B ~l, there exist k1, k2 e D(B)
such that h1= Bk1 and hi = Rki- Then since B is symmetric, we have

as claimed.
The Hellinger-Toeplitz theorem [ReSil, p. 84], a simple consequence of the closed
graph theorem, assures us that the everywhere defined symmetric operator C is bounded
and self-adjoint. Also, C is injective since both j and B"1 are injective. Hence by
Theorem 10.1.8, the spectral theorem in multiplication operator form, there is a finite
measure space (£2, A, M) and an R-valued function / e L°°(£2, A, M) such that C is
unitarily equivalent to the operator M/ on L2(£2, A, M) of multiplication by /. Since C
is injective, M/ is also injective and so / ^ 0 u-a.e. Thus 1/f is R-valued u-a.e. and
M1/f is a self-adjoint operator (not necessarily bounded, of course) on L 2 (Q, A, u.). The
spectral theorem applied this time to an unbounded—rather than a bounded—function
of a self-adjoint operator assures us that C-1 is unitarily equivalent to M\/f and so
C"1 = B is self-adjoint.
We define A := B - I with D(A) = D(B). Then A is self-adjoint. Also we see
from (10.3.40) that (<p, fii/0 = ?(<?. VO + (<P, VO for every <p, V e D(B). Hence, for
<p, \lf e D(A),

But, by Proposition 10.3.9, we have that (<p, A\jf) = qA(<P, VO f°r aU <P,ir £ D(B),
where qr^ is the quadratic form associated with the self-adjoint operator A. From this
and (10.3.41), we have

In order to finish the proof of the existence part of the assertion of this theorem, we
need to show that Q(q) = Q(qA) and that q = q& on this common domain.
under U.S. or applicable copyright law.

Q(q) ^ Q(qA)'- Let h € H+\ = Q(q). Recall that in the process of showing above
that D(B) is dense in H, we saw that D(B) is || • ||+1 -dense in H+\. Since D(A) - D(B),
D(A) is then also || • ||+i-dense in H+I. Using this fact, we can take a sequence {an}
from D(A) such that \\an - h\\+i -+ 0.
We see from (10.3.4) that the self-adjoint operator A is nonnegative. For any k e
Q(9A), let

Since qA = q on D(A) x D(A), we see that || • |U,+i = II • ll+i on D(A). Hence we

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
haveUNIVERSIDAD
\\an - aDISTRITAL
m |U,+i FRANCISCO
= \\an -JOSEamDE||+i ->• 0. Since qA is closed by Proposition 10.3.8,
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 187
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

there exists /JQ e HA,+\ '•= Q(QA) such that \\an — AolU.i -*• 0. The proof of the
containment will be finished if we show that ho = h. However, q(an -h) + \\an — h\\2 =
\\an - h\\2+l -> 0 and q(an - h0) + \\an - h0\\2 = \\an - h0\\\v -»• 0. It follows that
ll^n — h|| ->• 0 and \\an — ho\\ ->• 0, and so h — ho as we wished to show.

Q (QA ) £ Q (l) : The proof of this containment is essentially the same as that of the
containment above and we leave it as an exercise for the reader except to remark that the
|| • || A,i-density of D(A) in Q(qA) comes from Proposition 10.3.10.
The two containments assure us that Q(q) — Q(qA) and so it remains to show that
q — qA on their common domain. Let h & Q(q). Take a sequence {an} from D(A)
just as in the first part of the proof so that \\an — /Mil -> 0. From the earlier argument
and the fact that ho above turned out to be h, we see that \\an — A |U, i -*• 0- It follows
that ||an||i -» ||h||i and \\an\\A.\ -* \\h\\A,i. But || • |U = || - ||Ail on D(A) (as noted
previously) and so ||&||i = ||/z|U,i- But then, by the polarization identity (equation
(10.3.2)), q(., •) + (-, •) = qA(-, •) + (-, •) on Q(q) = Q(qA). Hence q = qA as desired.
We will finish the proof of Theorem 10.3.14 by establishing the uniqueness that is
claimed in part (a). Accordingly, let A and B be self-adjoint operators such that qA =
qB = q. Then Q(qA) = Q(qB) and qA = qs on Q(qA) x Q(qA) = Q(qB) x Q(qB).
We need to show that D(A) = D(B) and that A and B agree on this common domain.
We show below that D(A) c D(B*) and B* = A on D(A). Since B is self-adjoint,
it will follow that D(A) c D(B) and B - A on D(A). Further, from the symmetric
nature of the assumptions, we will then have that D(B) c D(A) and A = B on D(B).
Consequently, it will follow that D(A) = D(B) with agreement of A and B on their
common domain.
Thus, it remains only to show that D(A) c D(B*) with B* = A on D(A). Let
(p e D(A). It suffices, by definition of B*, to show that (B^jr, <p) — (\j/, A<p) for every
^ e D(B). (See Definition 9.6.1.) Let ty e D(B). Now we have both D(A) c Q(A)
(= Q(B)) and D(B) C Q(B) (= Q(A)) by Proposition 10.3.9. Using those facts in the
second and fifth equalities below and the fact that <p 6 D(A) and \js € D(B), we can
write
under U.S. or applicable copyright law.

Hence, <p e D(B*) and B*tp — A(p as we wished to show.

As was noted earlier, now that we know that q = qA (in the sense of Definition 10.3.5),
the rest of the claim in Theorem 10.3.14 easily follows from Proposition 10.3.9. D
Theorem 10.3.14 above is often referred to as the "first representation theorem" for
forms. See Kato's book [Kat8, Theorem VI.2.1, p. 322] for additional details and a more
general result. We will also need an additional representation theorem [Kat8, Theorem
VI.2.23, p. 331] which we now give.
Theorem 10.3.15: eBook
EBSCO Publishing (Second representation
Academic theorem)
Collection (EBSCOhost) Let onq 6/8/2017
- printed be a closed nonnegative
5:35 PM via
quadratic form
UNIVERSIDAD and let
DISTRITAL A be the
FRANCISCO JOSEassociated
DE CALDAS nonnegative self-adjoint operator such that
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 188 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

q = qA. (See Theorem 10.3.14.) Then Q(A), the form domain of A (see Definition
10.3.5), Is given by Q(A) := Q(q) = D(Al/2). Moreover,

and so, in particular,

Proof We first note that the existence and uniqueness of the nonnegative self-adjoint
operator A such that q = qA is guaranteed by Theorem 10.3.14, the first representation
theorem.
Now, Theorem 10.3.15 is easy to prove using either the definition of the quadratic
form qA associated with the self-adjoint operator A (see Definition 10.3.5 and (10.3.7)-
(10.3.10)) or the alternative described in Remark 10.3.6 (see (10.3.11)-(10.3.14)). We
will use the latter.
The operator A is (unitarily equivalent to) the operator Mg on L 2 (/z) = L 2 (£2, A, /J.)
of multiplication by the .A-measurable, M-valued function g, where u(£2) < oo and g
is finite /i,-a.e. As earlier, g will denote both the function itself and the operator Mg of
multiplication by g. In the present case, g > 0 since A > 0.
We see from (10.3.13) and (10.3.14) that

and

But since A 1 / 2 = g 1 / 2 , it follows that

and
under U.S. or applicable copyright law.

Comparing (10.3.44) and (10.3.46), we see that Q(A) = D(Al/2). Similarly, from
(10.3.45) and (10.3.47) we obtain (10.3.43).
Exercise 10.3.16 Prove Theorem 10.3.15 using Definition 10.3.5 and formulas (10.3.7)-
(10.3.10).
Given a self-adjoint operator A : D(A) —* 7, we discuss next the operators
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
\A\,UNIVERSIDAD \A\1/2, A+
A+, A-, DISTRITAL , A_JOSE, and
FRANCISCO the relationships between them. Our first
DE CALDAS use of
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 189
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

the results of this discussion will give information about the form domain of A in the
case where A is not necessarily positive, but we will make use of these operators in sev-
eral other places as we continue. (See especially Sections 11.3-11.6, as well as Chapters
12 and 13.)
According to the spectral theorem in the multiplication operator form, Theorem
10.1.8, the self-adjoint operator A is unitarily equivalent to the operator Mf on L2(/u.) =
L 2 (fit, A, n) of multiplication by an R-valued .4-measurable function /, where /K, is a
finite measure. We will frequently suppress reference to the unitary operator U : Ti. ->
L 2 (/z) and just regard A as equal to A//. We will be more formal here and write

where D(A) is related to D(Mf) as follows:

The operators | A \, A+ and A -—called respectively the absolute value, the positive part
and the negative part of A— are then defined uniquely via the spectral theorem by

and

These three self-adjoint operators are all nonnegative since the functions |/|, /+, /_ are
nonnegative. (Here, the functions /+ := max(/, 0) and /_ := max(—/, 0) denote the
positive part and the negative part of /, respectively.) Further, from the relationships
/ = /+ — /- and | /| = /+ + /_, the spectral theorem assures us that

and
under U.S. or applicable copyright law.

Now D(Mf) = D(M|/|) = {y & L2(,u) : ftp e L2(/j,)}, and D(Mf±) = {<p e
2 2
L (n) : f±-<p € L (/z){.Sincealso[a> e ft : f+(a>) > 0}r\{ca € ft : /_(«) > 0} = 0,
we see that

and so
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 190 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In similar fashion, we define the following nonnegative self-adjoint operators, the

square root of \A\, A+, and A_, respectively:

and

Since /+/_ = 0, we have |/|1/2 - /+/2 + /1/2, and so we see that

Arguing much as we did in connection with (10.3.53), we have

Now from the spectral theorem (Theorem 10.1.8), the second representation theorem for
forms (Theorem 10.3.15), and (10.3.56), we deduce that Q(A), the form domain of A,
is given by

Finally, from (10.3.9) and (10.3.10), the equation x = x+ - x-, and the second repre-
sentation theorem, we obtain that for <p, i{r 6 Q(A),

In particular,

The form sum of operators

The concept of the "form sum" of two self-adjoint operators will play an important role in
two of our approaches to the Feynman integral (Sections 11.3-11.5 and Sections 13.1-
under U.S. or applicable copyright law.

13.2, respectively), as well as in our dicussion of the Feynman-Kac formula (Chapter

12). The ordinary operator sum (or algebraic sum) of two self-adjoint operators A and
B will also be used and has a straightforward definition:

and

It can happen, however, that D(A) f~l D(B) is not dense in H. Indeed, the extreme
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
caseUNIVERSIDAD
where ADISTRITAL
and B are densely
FRANCISCO JOSEdefined
DE CALDASself-adjoint operators but D(A) n D(B) = {0}
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 REPRESENTATION THEOREMS FOR UNBOUNDED QUADRATIC FORMS 191
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

is possible (see Example 10.3.21 below). The idea is to consider the sum qA + qs
of the forms associated with A and B. Under appropriate hypotheses to be discussed
below, qA + qa will be densely defined, semibounded and closed. Theorem 10.3.14, the
first representation theorem, then assures us that there exists a unique densely defined,
semibounded, self-adjoint operator associated with the form qA + qa- This self-adjoint
operator will be denoted A+B and called the/orm sum of A and B. It can happen in
cases of interest that A+B makes perfectly good sense even though A + B has a domain
which is too small to be useful. Our concern with form sums comes from our desire to
have the Feynman integral defined even when highly singular potential energy functions
are involved as indeed they sometimes are in physical problems.
In most of the concrete applications of form sums, including those which are our
primary concern, the self-adjoint operator A is nonnegative. Hence, we restrict our atten-
tion to that case. Following (10.3.51a), we will write the self-adjoint operator B as
B = o_j- — B~.
Definition 10.3.17 We say that B_ is relatively form bounded with respect to A (briefly,
B- is A-form bounded,) with bound less than 1 if and only if

and there exist positive constants y < 1 and S such that

The infimum of all such positive numbers y is called the A-form bound of B. (Hence it
is strictly less than one.)
Remark 10.3.18 (a) Definition 10.3.17 can easily be extended to the case where A is
a semibounded, self-adjoint operator and B is a symmetric (densely defined) operator.
We simply replace Q(B-) by Q(B) in (10.3.62a) and (10.3.62b) by

(b) When the number Y in (10.3.62b) can be made smaller, it may well be necessary to
take S larger. We stress that the infimum of the numbers y which can be used in (10.3.62b)
need not be a bound itself.
under U.S. or applicable copyright law.

From our earlier discussion, especially the second representation theorem (Theorem
10.3.15), we see that the expressions in Definition 10.3.17 make sense and that (10.3.62b)
can equivalently be written as

We add to the hypothesis of relative boundedness the assumption that Q(A) n Q(B+)
is dense in H. Now the map

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 192 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

<p, i/f e Q(A) n Q(B+), is clearly a sesquilinear form. Further, it follows easily from
(10.3.62b) that this form is bounded below since for <p e Q(A) n Q(B+), we have

We leave it as an exercise for the reader to show that the form qA + qs is also closed.
We can now apply the first representation theorem (Theorem 10.3.14) to produce the
self-adjoint operator A+B. We summarize in the next theorem the central facts coming
from the discussion above.
Theorem 10.3.19 (Form sum of operators) Let A and B be self-adjoint operators on Ti,
with A nonnegative. Suppose further that B_ is A-form bounded with bound less than 1
and that Q(A) D Q(B+) is dense in U.
Then the quadratic form QA+QB given by (10.3.63) is closed and semibounded and so
there exists a unique semibounded self-adjoint operator A+B, called the form sum of A
and B, such that qA+B = qA+qB; further Q(A+B) = Q(A)r\Q(B) = Q(A)r\Q(B+).
The operator A+B is a self-adjoint extension of the algebraic sum A + B. (Recall that
D(A + B) = D(A) n D(B).)
Remark 10.3.20 The density assumption could be eliminated from Theorem 10.3.19 at
the price of working on the Hilbert subspace of H obtained by taking the closure of
Q(A) n Q(B+) in H. A similar comment could be made at several other places in this
book.
When we apply Theorem 10.3.19 below, U will be L1 (Rd), A will be the free
Hamiltonian HO — —1/2 A and B = B+ — B- will be the operator My of multiplication
by the potential energy function (or briefly, potential) V. Thus

and we will need conditions on V+ and V_ which assure us that Q(Ho) n Q(My+) is
dense in L2 (Ed) and that Mv_ is //o-form bounded with bound less than 1. We will
under U.S. or applicable copyright law.

provide such conditions in Section 10.4. (See especially Theorem 10.4.8.)

We close this section by providing a somewhat pathological example ([Cher2],

adapted to quadratic forms in [La6, footnote to pp. 1705-1706]) which shows that in
general, caution must be observed when comparing the operator sum with the form sum.
Example 10.3.21 Let H = L 2 (R).Let A (- H0) be the differential operator -1/2d2/dx2
with domain

EBSCO Publishing
D(A) =: {<peBook
e Academic Collection
7i : <p and <p' are(EBSCOhost)
absolutely - printed on 6/8/2017
continuous, 5:35 ePM Ti].
<p', <p" via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONDITIONS ON THE POTENTIAL V FOR #0-FORM BOUNDEDNESS 193
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(So that || A^vf = \ /+* \<p'(x)\2dx and

Q(A) = D(A 1/2 ) = \<p e 7i : <p is absolutely continuous,^' e H\.)

Let /? = MV (= V) be the operator of multiplication by a nonnegative measurable

function V on R. (So that D(5) = {<? e W : V<p e H] and g(B) = D(B'/ 2 ) =
{<p e Ti. : Vl/2(p € Ti.}.) Assume that V is integrable on R. but nowhere locally square-
integrable; i.e. _/j-0fc, | V(jf)| Jjc = oo for every compact interval [a, b]. The former
condition implies that Q(A) n (2(5) is dense in H (because it contains X>(R)) whereas
the latter insures that D(A) n D(B) = {0}.
Consequently, the operator sum A + B is trivial (being the zero operator with domain
{0}) whereas, by Theorem 10.3.19, the form sum A+B is a well-defined (and hence
densely defined) self-adjoint operator in H.
To obtain a function V satisfying the above hypotheses, take K > 0, K e L 2 (R), but
with K4 not integrable near x = 0. Then, if {rn}'^_l is an enumeration of the rational
numbers, set

where the series in (10.3.66) is convergent in L2-norm. We leave the verifications as an

exercise for the reader.

10.4 Conditions on the potential V for Ho-form boundedness

Let HO be the free Hamiltonian as discussed in Section 10.2. Suppose that V : E.d -> R
is Lebesgue measurable and let V = V+ — V- be the usual decomposition of V into its
positive and negative parts: V+ = max(V, 0) and V_ = max(—V, 0). We will have in
some key theorems to follow (in Chapters 11-13) the assumption that V_ is Ho-form
bounded with bound less than 1. We state in this section various concrete conditions
on V which guarantee its form boundedness. This will enable us to use the results
obtained at the end of Section 10.3, particularly Theorem 10.3.19, to define the form
sum H = H0+V, a natural realization of the quantum-mechanical Hamiltonian, under
very general hypotheses.
under U.S. or applicable copyright law.

The classes Kd in the following definition were introduced by Kato [Kat2] and are
often referred to as the Kato classes. Our main references for this material are [CyFKSi,
§1.2] and Simon's survey article [Sil 1]. (Also, see [Kat8] and [AiSi], as well as the more
recent work [GulKon] quoted in Remark 10.4.9(b) below, along with [Gull].)
Definition 10.4.1 (The Kato class Kd) Let W : W1 -» R be Lebesgue measurable. The
function W belongs to the Kato class K^ if and only if

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 194 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We will also need the following classes of functions in connection with the present
dicussion as well as further on.
Definition 10.4.2 Let W : Rd ->• E be Lebesgue measurable, and let p e [1, oo).
(i) We say that W is in L^c(Rd)-uniformly and write W e tfoc(Rd)u (or simply
^^(L^J if and only if

(ii) We say that W is in (L^K^-strongly uniformly and write W e L^^R4)-

(or simply W € (i-j^.)-) if and only if

Remark 10.4.3 (a) Observe that K\ = L^R),,.

(b) Simple compactness arguments yield the first two containments below:

where L^oc(Wi) is the space of functions W such that fc \W\pdy < oo for every compact
subset G ofRd.
(c) The class (•£-£,<.)- was first introduced by Simader in his work on Schrodinger
operators with singular potentials [Sd].
(d) It is apparent that the classes (££,<.)» and (L^oc)u are related. See (Hi) and (iv)
of Proposition 13.4.5 for some precise information on this point.
The next proposition accounts for our interest in Kd • (At this point, the reader may
wish to review Definition 10.3.17 and Remark 10.3.18 where the notion of Wo-form
boundedness is introduced.)
Proposition 10.4.4 If W e Kd, then W is Ho-form bounded with bound less than 1.
(Actually, W has Ho-form bound 0 in this case.)
under U.S. or applicable copyright law.

A proof of Proposition 10.4.4—based on a characterization of the Kato class

Kd in terms of Brownian motion—can be found in [AiSi] and is sketched in [Sill,
pp. 458-459]. For smaller classes—such as (L\oc)~, for example—Proposition 10.4.4
easily follows from the Sobolev inequalities (e.g., [Ad, Bre2, Kat8, Kes]); see, for
example, [BreKat, pp. 139-140].
We now give some information about Kd and its contents.
Proposition 10.4.5 (a) Kd c L,1^ (E.d)u for all positive integers d.
(b) If we take p > d/2 ford>2 and p = 1 for d = 1, we have
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONDITIONS ON THE POTENTIAL V FOR #0-FORM BOUNDEDNESS 195
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We now briefly comment on the first containment in (10.4.1) above. Recall to begin
with that, by definition, W e LP + L°° if and only if W = Wi + W2, with Wi e LP
and V/2 € L°°. Next, since V in Lp implies that the measure | V(y)\pdy is absolutely
continuous with respect to Lebesgue measure, it is easy to see that Lp c (L^c)~; further,
since L°° is also contained in (L^- and (L^.)- is closed under addition, it follows
that Lp + L°° is contained in (L^oc)~, as desired.
Example 10.4.6 ([CyFKSi, p. 8]) If d > 3 and & € (0, 1], then

is //o-form bounded with bound less than one but is not in Kj. The function V does
belong to Kd if and only if S > 1.
We note that the first two inclusions in (10.4.1) hold for any p e [1, oo). Moreover,
an example showing that the containment (££,<.)« £ (L^u is strict can be found in
[Jo6, pp. 15-16]. Further, Example 2.6 of [JoKim] shows that the first containment in
(10.4.1) is strict. Finally, by Remark 10.4.3(a), the last containment in (10.4.1) is actually
an equality when p = 1.
We state one further pleasant property which is possessed by the classes K^, d —
1,2, . . . .
Proposition 10.4.7 Let d and d' be positive integers with d < d'. Suppose that the
function W : E.d ->• R is in Kd and that T : R.d' ->• Rd is linear and surjective. Then
V(x) := W(Tx) is in Kd<.
The linear map T in Proposition 10.4.7 could be, for example, a projection onto a
subspace or, in connection with an N-particle quantum system (see Section 6.4), we
could have

where p, q £ ( I , . . . , N] with p ^ q.
We close this section by applying the above results to Theorem 10.3.19. Here, H =
2
L ^) and the operators A and B from Theorem 10.3.19becomef/0 and V = V+-V-,
respectively.
under U.S. or applicable copyright law.

Theorem 10.4.8 Let V : Rd ->• R be Lebesgue measurable. We assume that V+ e

.LI'OC (Rd} and that V- is Ho-form bounded with bound less than 1. (We know from
Propositions 10.4.4 and 10.4.5 that this condition on V- is satisfied if V- is in any of
the classes Lp (Rd) + L°° (Rd) c L£C (Rd)~ c L£C (Rd)u c Kd, where p > d/2 if
d>2andp = lifd=l.)
Then the form sum H := Ho+V is well defined and is a semibounded, self-adjoint
operator on L2 (Rd).
p
Remark 10.4.9 (a)
EBSCO Publishing If we
: eBook assume
Academic that (EBSCOhost)
Collection V_ belongs to (L^z
- printed ^ L5:35
on 6/8/2017 +PM L°°,
via then the
conclusion
UNIVERSIDADofDISTRITAL
Proposition 10.4.4
FRANCISCO holds
JOSE DE even when p = d/2 for d > 3 (i.e. d > 2); see,
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 196 SELF-ADJOINT OPERATORS AND QUADRATIC FORMS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

for example, [BreKat, Lemma 2.1 and Remark 2.1, pp. 139-140]. Hence the condition
on p in Theorem 10.4.8 can be relaxed accordingly.
(b) Recently, A. Gulisashvili and M.A. Kon [GulKon] have obtained, in particular, an
elegant potential-theoretic characterization of the Kato class Kd, building upon [Sill,
Proposition A.2.3, p. 454] (obtained in [AiSi]) and comments in [Sill, e.g. p. 458].
Briefly, their result [GulKon, Lemma 2.1] can be stated as follows: Let V € L^ (Rd).
Then V e K,j = Kj (Md) if and only i f \ V \ — k — Ak, where k is a bounded uniformly
continuous function on Rd and where Ak denotes the distributional Laplacian (closely
related to the second Bessel potential) ofk. (Actually, the main theorem in [GulKon]
solves a conjecture ofB. Simon [Sil2] concerning the smoothness of the heat semigroup
e~'H, where H = HQ+V and the potential V is a suitable function in the Kato class Kj.
We note that very recently, this result has been extended to a certain class of singular,
time-dependent potentials in [Gul2].)
Although the definition of the Kato class (given in Definition 10.4.1) appears some-
what technical at first, it may now seem more natural in the light of the above charac-
terization ofKd-
The following exercise will be useful to us later on; it requires from the reader some
knowledge of the theory of Sobolev spaces [Ad, Bre2, Kat8, Kes].
Exercise 10.4.10 (Form domain of the Hamiltonian H = Ho+V)
(a) Show that

and that Q(H0) = H1 (W1), the Sobolev space of functions <p e L2 (M.d) with distribu-
tional gradient v^ in L2 (Rrf); i.e. each of the d components ofVtp lies in L2(]Rrf).
[Hint: If<p e T> (Md), equation (10.4.3) follows from Green's formula (and the fact that
qH0(<p) = (<P, Ho<p)for <p e D(Ho)). It then extends to any <p e H1 (Rd), by density of
T>(M.d)inHl(E.d).]
(b) Deduce from (a) that under the assumptions of Theorem 10.4.8, we have for
H = HO+V, the form sum of HO and V,
under U.S. or applicable copyright law.

for all (p € Q(H).

We see from (10.4.4) that Q(H), theform domain of the quantum-mechanical Hamil-
tonian H, is precisely equal to the set of "quantum states" <p with finite total energy (i.e.
kinetic plus potential energy). In this sense, the form sum H = Ho+V is a natural
realization of the quantum-mechanical Hamiltonian. (See [Ne4, Sil, ReSi2, Kat8].) The
form qn from (10.4.5) is often referred to as the "energy form".
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

11
PRODUCT FORMULAS WITH APPLICATIONS TO THE
FEYNMAN INTEGRAL

Now that we have assembled the necessary tools, we are going to obtain in the next
three chapters (especially in Chapters 11 and 13) some very general existence theorems
for the (operator-valued) Feynman integral. We begin in Chapter 11 by considering two
approaches to the Feynman integral via product formulas associated with semigroups
(or other suitable functions) of operators; namely, the "Feynman integral via the Trotter
product formula" (for unitary groups) and the "modified Feynman integral" (via a product
formula for imaginary resolvents).
The Trotter product formula, you will recall, was introduced in Chapter 7
(Section 7.5). We now prove in Section 11.1 a more abstract form (by means of the
Chernoff product formula [Cherl,2]) and then obtain in Section 11.2 a general existence
theorem for the Feynman integral (via the Trotter product formula for unitary groups)
F'TP(V) introduced by Nelson in [Nel]. The beautiful distributional inequality of Kato
[Kat2] permits us to establish the essential self-adjointness of the Schrodinger operator
HQ + V under more generality (on the potential V) than was known at the time of Nelson's
paper and therefore leads to broader results. (See Section 11.2.)
Interest in the Feynman integral led Lapidus ([Lai 1], [Lal3, Part I]) to obtain a "prod-
uct formula for imaginary resolvents" (rather than for unitary groups, as used in [Nel]).
This formula is established in Section 11.3. Its proof rests in part on quadratic form
techniques and the spectral theorem (as discussed in Chapter 10), as well as on a result
of Chernoff in [Cher2]. When specialized, the product formula for imaginary resolv-
ents yields in turn the existence of the "modified Feynman integral", also introduced by
the second author in [La6, 11]. (See Section 11.4.) Thus the existence of the "modified
Feynman integral" F'M(V) associated with the potential V is established under signifi-
under U.S. or applicable copyright law.

cantly more general assumptions on V than for the Feynman integral via the Trotter
product formula ^ P (V). In particular, essential self-adjointness of the Schrodinger
operator HO + V is no longer required. This enables us to show the convergence of
T1M(V) (essentially) whenever the Hamiltonian (or "energy operator", defined by means
of quadratic forms) is bounded from below, and hence the corresponding Schrodinger
equation (with potential V) is defined without ambiguity.
It is natural to ask whether the Feynman integral is stable under small perturba-
tions in the potential V. This question seemed not to have been directly addressed until
the paper by Johnson [Jo3] which gives a "bounded-type" convergence theorem for
the EBSCO
analytic operator-valued
Publishing Feynman
: eBook Academic Collectionintegral (to- be
(EBSCOhost) studied
printed in Chapters
on 6/8/2017 15-16). The
5:35 PM via
paper of Lapidus
UNIVERSIDAD [La12]
DISTRITAL gives JOSE
FRANCISCO a "dominated-type"
DE CALDAS convergence theorem for the modified
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Feynman integral under far less restrictive conditions. (See Section 11.5.)
Operational Calculus
Account: ns000601
 198 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Finally, in Section 11.6, we discuss extensions of the modified Feynman integral to

the case of complex-valued potentials, following work in [BivLa]. This is of interest in
the study of dissipative (rather than isolated) quantum systems.
In closing this brief overview of Chapter 11, we point out the survey article [FrLdSp]
which gives a number of examples where singular potentials (which play a central role
in large parts of this chapter and in Chapters 12 and 13 below) are used as mathematical
models in phenomenological high energy physics and quantum chemistry. Moreover,
we mention the book [Ex] where complex-valued potentials are discussed extensively
as mathematical models for dissipative quantum-mechanical systems.

11.1 Trotter and Chernoff product formulas

We turn back to the mainstream of our development and proceed to a proof of the Trotter
product formula. Our interest is primarily in the Hilbert space setting; however, we have
assembled the tools necessary to give the proof in the Banach space setting, and hence
we will do so. In fact, we will prove a more general result, the Chernoff product formula,
from which Trotter's theorem will follow quite easily.
Given a positive integer n, it will be helpful to us below to know that

Formula (11.1.1) is an easy consequence of the calculation, found in introductory prob-

ability texts such as [Ros, pp. 54-55], of the mean and variance of a Poisson random
variable Z with rate A. (A > 0). The Poisson distribution (or measure) is concentrated on
the nonnegative integers 0,1, 2 , . . . and

It is shown that the mean and variance of Z are both A.. In particular,
under U.S. or applicable copyright law.

We obtain (11.1.1) by applying (11.1.2) to the case A. = n.

The proof of the Chernoff product formula is based on the following inequality from
[Cherl].
Lemma 11.1.1 (Chernoff's lemma) If L is a bounded linear operator on the Banach
space X such that \\L\\ < 1, then for all f € X and for each nonnegative integer n,

In the following, an operator L in L.(X) such that \\L\\ < 1, as in Lemma 11.1.1, will
be EBSCO Publishing
called a linear :contraction
eBook Academicor,
Collection
briefly, (EBSCOhost) - printed
a contraction on X.on 6/8/2017 5:35 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TROTTER AND CHERNOFF PRODUCT FORMULAS 199
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof of Lemma 11.1.1. We will use the following fact. For any B e C(X) and any
a eR,

Formula (11.1.4) is easily justified:

Now using (11.1.4) with a = n we can write

Using the assumption that \\L\\ < 1, we have for k > n,

Similarly, for k < n, we can write

Using (11.1.6) and (11.1.7) we obtain for all fc = 0, 1, 2 , . . . ,

under U.S. or applicable copyright law.

Using again the fact that \\L\\ < 1, we have for m = 0, 1, 2 , . . . ,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 200 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Now from (11.1.5), (11.1.8) and (11.1.9) and then the Cauchy-Schwarz inequality, we
deduce

where the next to last equality comes from (11.1.1). Hence formula (11.1.3) is established.
D

We are now prepared to prove the Chernoff product formula [Cher 1,2].
Theorem 11.1.2 (Chernoff product formula) Let {F(0}r>o be a family of contractions
on the Banach space X with F(0) = /. Suppose that the derivative F'(Q)f exists for
all f in a subspace D such that the closure A of F'(0)\o generates a (Co) contraction
semigroup {T(t) : t > 0}.
Then for each f e X,

uniformly for t in compact subsets of [0, +oc).

[Of course, we denote by F'(G)\D the restriction to D of the (strong) derivative F'(Q).]
Proof We will carry out the proof except for the uniformity which is claimed in connec-
tion with the limit in (11.1.11). The key steps will be justified by Theorem 9.7.10 and
Lemma 11.1.1. Let AQ = A and, for t > 0, let
under U.S. or applicable copyright law.

The operators F ( t ) are contractions by assumption. The bounded operators An generate

norm continuous (see Proposition 9.5.2), and so certainly strongly continuous, semi-
groups Tn(t) := e'A". We now show that ||r w (OH < 1 for every n and for every t > 0.
Formula (11.1.4) will again be useful to us.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TROTTER AND CHERNOFF PRODUCT FORMULAS 201
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

From (11.1.13) we see that the stability condition involved in the hypotheses of Theorem
9.7.10 is satisfied with M - 1 and w = 0. Applying Theorem 9.7.10 (or Theorem 8.1.12)
we obtain

uniformly for t in compact subsets of [0, +co). Now Lemma 11.1.1 and the existence
of F'(0)f for / e D allow us to write

as n -»• oo, for all / e D. Combining (11.1.14) and (11.1.15), we see that for every
/ e D, \\Fn(t/n)f - T ( t ) f \ \ ->• 0 as n -> oo. Since D is dense in X, it follows that
for all / e X

as we promised to show. D
Remark 11.1.3 The reader should note that the family of contractions {F(t)} involved
in the Chernoff product formula is not required to form a semigroup.
As mentioned earlier, the Trotter product formula [Tro2] follows easily from the
Chernoff product formula.
Theorem 11.1.4 (Trotter product formula) Suppose that A, B and A + B generate (Co)
contraction semigroups {S/i(0}f>o- {Se(/)}(>o, [T(t)}t>o on the Banach space X. Then

for each f e X, where the convergence (is in the norm on "H and) is uniform for t in
compact subsets of [0, +00).
under U.S. or applicable copyright law.

The formula (11.1.17) is often written instead as

We caution the reader that the exponential notation in (11.1.17')—although fre-

quently used—can be misleading because, as we have seen in Chapters 8 and 9, semi-
groups do not possess all of the properties of the numerical exponential function. When
the generators of the semigroups are self-adjoint (or skew-adjoint as in Corollary 11.1.6
below), the richness of the associated functional calculus (Section 10.1) can be used to
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
more fully justify
UNIVERSIDAD theFRANCISCO
DISTRITAL exponential
JOSE DEnotation.
CALDAS (See especially Remark 10.1.12(a), (b).)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 202 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We note that the limit in (11.1.17') (or (11.1.17)) is also sometimes denoted by

and referred to as the "strong operator limit" or "limit in the strong operator topology of
£(H)".

Proof of Theorem 11.1.4 We will apply Theorem 11.1.2 with D := D(A + B) =

D(A) n D(B) and

Since SA(t) and SB(t) are contractions, F(t) is certainly a contraction. Also F(0) =
SA(0)SB(0) = I2 = I. Now, for / € D, we can write

as t -*• 0+. (To justify the first limit in the last line of (11.1.19), we use the fact that
IISxWII < !•) Thus the strong derivative F'(0) exists and equals A + B on D =
D(A + B). Hence, from this and one of our assumptions, we have that the closure of
F'(0)|o, that is, A + B, generates the (Co) contraction semigroup {T(t)}. Now applying
the Chernoff product formula (11.1.11) in the present situation, we obtain

as desired.
under U.S. or applicable copyright law.

Remark 11.1.5 (a) In view of the Chernoff product formula (Theorem 11.1.2) the Trotter
product formula (Theorem 11.1.4) extends to the case where there exists a subspace D of
D(A) n D(B) such that (A + B)\D generates a (Co) contraction semigroup. The proof
is the same; it suffices to replace A + B by (A + B)\D.
(b) Formula (11.1.17') is sometimes referred to as the Trotter-Lie Product Formula
in the literature; see, e.g., ([Cher2], [Kat4], [La1, 5,6], [La13, Part I]). Indeed, in about
1875, Sophus Lie established the finite-dimensional case of(11.1.17'), which corresponds
to smooth vector fields generating one-parameter groups of matrices associated with
the solutions of differential equations. (Of course, in this situation, there is no need to
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
distinguish between A + B and A + B.)
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TROTTER AND CHERNOFF PRODUCT FORMULAS 203
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Product formula for unitary groups

Corollary 11.1.6 (Trotter product formula for unitary groups) Let A and B be self-
adjoint operators on a complex Hilbert space H, and suppose that there exists a subspace
D of D(A) n D(B) such that (A + B)\D is essentially self-adjoint; i.e. C = (A + B)\D
is self-adjoint.
Then for every u e H,

uniformly in t on all bounded subsets of R.

Proof This is an immediate consequence of Theorem 11.1.4, Remark 11.1.5(a), and the
relationship between self-adjoint and skew-adjoint operators. Note, in particular, that the
skew-adjoint operators ±i A, ±iB and ±i'C = ±i(A + B)\D generate (Co) contraction
semigroups. (See Section 9.6, especially Definition 9.6.4 and Theorem 9.6.11, Stone's
theorem, according to which e"(±A), e"(±B) and e't(±C)(f > 0) are (Co) semigroups of
unitary operators.)
We discussed the relevance of the Trotter product formula to the Feynman integral
in Section 7.5. The reader should review that material at this point and might also wish
to consult the book of Goldstein [Gol, §8.13, pp. 54-55]. We remark that the Trotter
(or Chernoff) product formula is also useful in other contexts. For example, one can
derive from it the Central limit theorem (see [Tro3] and [Gol, pp. 51-53]). We will
discuss several additional applications of product formulas to diffusion theory or quantum
physics throughout the rest of this book.
Further, the Trotter (or Chernoff) product formula—along with its nonlinear exten-
sions (see Remark 11.1.9(b) below) and various generalizations—provides useful numer-
ical algorithms for solving certain (linear or nonlinear) partial differential equations and
variational inequalities; see, e.g., [ChorHMM], [KatMasu], [La5], [Lal3, Part I]) and
the references therein.
Exercise 11.1.7 Let F(t) be given by (11.1.18). In cases of interest, [F(t)},>o is not a
semigroup. Give a condition under which {F(t)} does form a semigroup.
We shall also need (for example, in Step 2 of the proof of Theorem 11.3.1) a more
under U.S. or applicable copyright law.

general version of Theorem 11.1.2, also due to Chernoff [Cher2, Theorem 1.1, p. 4],
which we now state without proof.
Theorem 11.1.8 (Chernoff's theorem) Let {F(t)}t>o be a family of contractions on the
Banach space X such that F(0) = /. Suppose that there is a (Co) contraction semigroup
{T(t) : t > 0} generated by the m-dissipative operator — C, such that for one (and hence
for all) A. > 0, we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
all f eX.DISTRITAL FRANCISCO JOSE DE CALDAS
for UNIVERSIDAD
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 204 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then for all f € X,

uniformly for t in compact subsets of [0, +00).

(Note that following the convention in Exercise 9.3.8, we can write T(t) = e~1c.)
Remark 11.1.9 (a) In the statement of Theorem 11.1.8, the word "m-dissipative" can
be omitted in view of Theorem 9.4.7; moreover, under suitable assumptions, the word
"semigroup" can be omitted as well (see [Cher2, Theorem 2.5.3, p. 28]). Finally, under
the hypotheses of the theorem, the conclusion (11.1.23) not only follows from (11.1.22)
but is equivalent to it (this is essentially a consequence of Theorem 9.7.11 or of [Kat8,
Theorem IX.3.6, p. 513]); see [Cher2, Theorem 1.1].
(b) Chernoff's theorem has been extended in a number of ways; in particular, it was
generalized to nonlinear semigroups by Brezis and Pazy [BrePaz]. (See also Brezis' book
[Brel, Theorem 4.3, p. 124], as well as [Reil,2] and the relevant references therein.)
Exercise 11.1.10 Show that the Chernoff product formula (Theorem 11.1.2) follows
from Theorem 11.1.8.

11.2 Feynman integral via the Trotter product formula

Feynman's approximation formula was discussed in Section 7.4, and we saw in Section
7.5 how this leads to Nelson's first approach to the Feynman integral via the Trotter prod-
uct formula. Consistent with the tone of Chapter 7, proofs were not given. However, con-
siderable background information in operator theory has now been provided in Chapters
8-10, and a quite general version of the Trotter product formula, Theorem 11.1.4, has
been proved; further, its application to unitary groups has been provided in Corollary
11.1.6. Hence, we are now in a position to give a much more detailed and precise math-
ematical discussion.
As mentioned in the introduction to Section 7.5, Nelson's result depended not only
on the Trotter product formula but also on the 1951 result of Kato [Katl] insuring the
essential self-adjointness of the (formal) energy operator H, where
under U.S. or applicable copyright law.

(In (11.2.1), we have simplified (7.5.1) by taking the physical constants h and m equal
to 1.) The energy operator (also called the "Schrodinger operator" or "Hamiltonian") is
now known to be essentially self-adjoint under far more general conditions than were
known at the time of Nelson's paper. We will discuss these matters below and will carry
out the part of the proof which seems to us to be the most novel and interesting. The
argument will revolve around a clever distributional inequality due to Kato [Kat2].
Criteria for essential self-adjointness of positive operators
WeEBSCO
beginPublishing
by stating an important abstract criterion for essential self-adjointness [ReSi2,
: eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
Theorem X.26,
UNIVERSIDAD p. 182]
DISTRITAL which
FRANCISCO JOSEwill be helpful to us in applying "Kato's inequality"
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 205
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(Theorem 11.2.3 below) to establish the essential self-adjointness of H, under suitable

hypotheses on the "potential" V. (See the proof of Theorem 11.2.5 below.) In the result
which we are about to state, it is only the fact that (ii) implies (i) that will be needed in
this section. (For the notion of symmetric and essentially self-adjoint operators, we refer
back to Definition 9.6.4.)
Theorem 11.2.1 (Criteria/or essential self-adjointness of positive operators) Let Ti. be
a Hilbert space and let Abe a symmetric operator on H which is also strictly positive;
i.e. there exists c > 0 such that (A<p, <p) > c((p, (p) for all <p 6 D(A). Then the following
are equivalent:
(i) A is essentially self-adjoint; that is, it has a unique self-adjoint extension
(necessarily its closure A).
(ii) Ker(A*) = {0}.
(iii) Ran(A) is dense.
(iv) A has only one semibounded self-adjoint extension.
Remark 11.2.2 (a) The above theorem obviously applies to semibounded operators (see
Section 10.3) A by replacing A by A+c, where c is a sufficiently large positive constant,
(b) Theorem 11.2.1 is related to an earlier criterion for essential self-adjointness,
Corollary 10.2.10, where semiboundedness of A was not assumed. Note, however, the dif-
ferences between the statements of Theorem 11.2.1 and Corollary 10.2.10. In particular,
in part (iv) of Theorem 11.2.1, we assume that A has a unique semibounded self-adjoint
extension. [Contrast this with part (i) of Corollary 10.2.10 (or of Theorem 11.2.1).] This
is not immediately equivalent to the essential self-adjointness of A because, in general,
a symmetric semibounded operator can have a self-adjoint extension that is not semi-
bounded. (See the comment following the proposition on page 179 of [ReSi2], as well
as Problem 26, page 341 in [ReSi2J.)

A brief outline of distribution theory

We now give an extremely brief summary of the parts of distribution theory which will
be helpful to us, especially in connection with Kato's inequality. A detailed treatment
of distribution theory can be found in many places, for example, [Schw2, ReSil,2,
H6r3]. The space of test functions will be T>(R.d), which we will denote by V throughout
this section. (Recall that T>, also sometimes denoted C^(Krf ) in earlier chapters, is the
under U.S. or applicable copyright law.

space of C-valued infinitely differentiable functions with compact support in Rd.) The
appropriate topology on Rrf is an inductive limit topology under which Z> is not metrizable
[ReSil, p. 147]. However, the convergence of sequences in T) is all that we will need,
and that is easily described. Let (<£>„} and <p belong to T>; then <pn converges to <p if and
only if there exists a compact set K such that all the functions involved have support
inside K and the derivative Datpn converges uniformly to Daip for each multi-index a
of nonnegative integers. (Here, as before, D"q> — —a3, " v a, for a = (a\,..., a^) and
3*, •••dXj
\a\ =«[ + ---+ad.)
The space of distributions T>' is the space of all continuous linear functionals on T>.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
Given {Tn} and
UNIVERSIDAD T from
DISTRITAL D', Tn JOSE
FRANCISCO converges
DE CALDASto T if and only if Tn(<p) ->• T(ip) for all <p e V.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 206 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A key fact about distributions is that they possess partial derivatives of all orders which
remain in the space D'. The am derivative of T where a is a multi-index acts on tp e D
as follows:

Further, differentiation is a continuous operation on V. Our concern here will be with

the distributional Laplacian A = -^r3jc,
-^-? acting on functions u e L1 which have
+ • • • + 3x
d

the property that the distribution AM is again in L2. (See the earlier related discussion in
Section 10.2 as well as in the next paragraph.)
The space L^ = Lj^R^) of C-valued functions u on Rd which are integrable over
every compact subset of R* (or "locally integrable", in short) is continuously imbedded
as a subspace of V via the formula

In particular, if Tv = Tw, then v = w Leb.-a.e. [H6r3, Theorem 1.2.5, p. 15], and so

v = w as elements of L'OC. Further, it is easy to show that if vn -» v in L|OC (i.e., for
every compact K c R rf , we have fK \vn — v\dx —> 0), then TVn —> Tv in V'. In the
following, we will write i; instead of Tv when no ambiguity may arise.
Various easily established facts about L'OC and L2OC will be used below, most often
without comment. For example, if « e L2 and V e L2OC, then uV e L,1^ (by the
Cauchy-Schwarz inequality).
We say that T e V is positive and write T > 0 if and only if Tip > 0 whenever
<f> e "D is nonnegative. Given two distributions T\ and T^, we write T2 > T\ if and only if
T2 — T\> 0. It is known that a positive distribution T actually corresponds to a positive
measure [H6r3, Theorem 2.1.7, p. 38]; that is, there exists a measure /u : B(Rd) ->
[0, +00) which is finite on compact sets and satisfies

for all (p e £>.

Kato 's distributional inequality
under U.S. or applicable copyright law.

We come now to a key step, the beautiful distributional inequality of Kato [Kat2,
Lemma A, p. 138].
Theorem 11.2.3 (Kato's inequality) Let u e L^ be such that its distributional Lapla-
cian Aw e L'^. Define

Of course, sgn u e L°°, so that sgn u, (sgn u) AM, and Re[(sgn M)A«] are all in L^
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
andUNIVERSIDAD
hence areDISTRITAL
all distributions.
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 207
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then we have the following distributional inequality:

Proof We will proceed in two steps.

Step 1: We suppose for now that u e C°°(Ed), the space of C-valued infinitely
differentiable functions on R rf , and we let

Taking the gradient on both sides of the equality

we obtain

Hence

Using the fact that \u\ < \uf\ and (11.2.6), we obtain

It follows that

(One may worry about the case u(x) — 0; but then it suffices to replace ubyu + l.)
Now

Applying (11.2.8) to both sides of (11.2.6), we have

Hence
under U.S. or applicable copyright law.

Next from (11.2.10) and (11.2.7), we deduce that

Thus

pointwise and so in a distributional sense. Now letting

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 208 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and using (11.2.12), we obtain (for all u e C°°(Rd))

In Step 2, we will use the inequality (11.2.13) and suitable approximation procedures
to finish the proof of Kato's inequality.
Step 2: Now we assume that u and AM are in L,1^. Let jg be an approximate iden-
tity (or "mollifier"); i.e. for S > 0, let js(x) := j(x/S)S~d where ; > 0, j € V,
fRd j(x)dx = 1, and the support of j is the closed unit ball of W1. Let

where the convolution * is defined as in (10.2.23). Since us € C°°(lRrf) [H6r3, Stel,

SteWe], we deduce from (11.2.13) that

It is well known and not difficult to check that us -> u in L11oc(IRrf). (See, for example,
the proof of Theorem 1.18 in [Ste We, p. 10].)Since AM € Z,,1^ and A(« 6 ) = (AM)*/? -
(Aw) 5 (by [H6r3, Theorem 4.1.1, p. 88]), we also have Aw"5 -> AM in L^oc. By passing
to a subsequence {Sp} (using a diagonalization argument) we also get convergence a.e.
of AUSP to Aw and of USP to u; in this connection, we will suppress reference to the
subsequence and simply write AM^ —> AM and us —> u a.e. (Here and thereafter, we
write a.e. instead of Leb.-a.e. when no ambiguity may arise.)
Now, for a fixed e > 0, we have (in view of (11.2.5) and (11.2.14))

where the convergence in (11.2.16) holds almost everywhere. Note also that both
|sgn e (M (S )| and ]sgn e (M)| are bounded by 1 for all S and €. Now take y € V and let
K = supp(^). Of course, K is compact. Using the fact that \\Aus - AM|| L I ( A -) -»• 0 as
under U.S. or applicable copyright law.

well as (11.2.16) and the dominated convergence theorem, we can write

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 209
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Thus

in the sense of distributions. It then follows easily from (11.2.17) that

in the sense of distributions.

We have just shown above that for u such that both u and AM are in L1loc, the right-
hand side of (11.2.15) converges in the sense of distributions to the right-hand side of
(11.2.13) as 8 ->• 0+. Our next goal is to show that the left-hand side of (11.2.15)
converges in the same sense to the left-hand side of (11.2.13). As noted earlier, us —»• u
in L\0(:. Using this, we can write, for any compact subset K of R rf ,

(Note that in (11.2.19), we have used the norm \ \ ( z \ , 12) \\p(c2) '•- (kl I2 + Iz2l 2 ) 1 / 2 , for
< Z 1 , Z 2 > €C 2 . )
We see from (11.2.19) that
under U.S. or applicable copyright law.

Hence (us)e -> uf in £>''. Finally, since differentiation is continuous in T>', we have

Now for u € L'OC such that AM e L^, the distributional inequality

follows from (11.2.15) via a limiting argument using (11.2.18) and (11.2.21).
+
Finally
EBSCO we let: €eBook
Publishing -» 0Academic
. Clearly sgn€(u)
Collection -> sgn(w)
(EBSCOhost) pointwise
- printed with
on 6/8/2017 5:38a PM
uniform
via bound
of 1, so that the
UNIVERSIDAD right-hand
DISTRITAL sideJOSE
FRANCISCO of DE
(11.2.22)
CALDAS converges in the sense of distributions to the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 210 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

right-hand side of (11.2.4). Further, we clearly have uf -»• \u\ a.e. Since we also have
u€ < (|«|2 + 1)1/2 < |M| + 1 e L,1^, we get uf -> \u\ in L^. Therefore «t ->• |w|
in the sense of distributions and so AM€ —>• A|w| in the same sense. The distributional
inequality (11.2.4) is now established and the proof is complete. D
Although we shall not use this fact here, we note that more abstract (operator-
theoretic) versions of Kato's inequality have been obtained, especially for the generators
of positive (i.e., positivity preserving) semigroups. See [Si4, 7], as well as, for example,
([Are], [La5, §5.3], [Na], [NaUh]) and the references therein. We will use, however, a
variant and extension of Kato's inequality in Section 13.6 and would also use it in the
proof (not provided there) of one result in Section 13.5. (See Lemma 13.6.6.)

Essential self-adjointness of the Hamiltonian H = HQ + V

We will use the fact that (ii) implies (i) in Theorem 11.2.1 to show in Theorem 11.2.5
below that — A + V is essentially self-adjoint when V is nonnegative and in L2loc. In that
connection, we will show that if u e £>[(- A + V + !)?£,] and [(- A + V + I)\T>]*U = 0,
then u — 0. In order to carry this out we will need the following lemma.
Lemma 11.2.4 Let V e L2oc(Rd) be such that V > 0 Leb.-a.e. Ifu e D[(-A + V +
l)|p]*. then both [(—A + V + 1)|£>]*« and (—A + V + 1)« make sense as distributions
and

as elements ofT>'.
Proof We show to begin with that -A + V + 1 is defined on £> with values in L 2 (R d ).
Since — A + 1 maps Z> into T>, it suffices to prove that V<p € L2 for every <p e V.
Fix (p e V and let K be the support of cp. Of course, K is compact. We wish to show
that fK V2\<p\2dx < oo. But V e L2(K) since V e L2^. Also \<p\2 e L°°(K) and so
V2(p2 e Ll(K). Thus fK V2\(p\2dx < oo as we wished to show.
We saw in Section 10.2 that HO = —\ A is self-adjoint on £>(#o) and that Z> c
D(Ho). Thus —A is a symmetric operator on V. It is easy to show that V + 1 is a
symmetric operator on T>, and so -A + V + 1 is itself a symmetric operator on T>.
Recall also (Definition 9.6.4(i)) that the adjoint of a symmetric operator extends the
operator. Thus
under U.S. or applicable copyright law.

Next we claim that (— A + V + \)u makes sense as a distribution for u e L2 and so

certainly for u 6 D[(- A + V + !)*£>] c L2. First note that any u e L2 is in L2OC c LJ^
and so u e T>'. Since — A + 1 maps T>' into T>', it suffices to show that Vu 6 L^. and
so can be regarded as a distribution. But V e L2OC and u €L2loc implies that Vu e L*oc,
as observed earlier.
Now[(—A+V4-l)| / z?]*MmakessenseasadistributionforM e D[(— A + V + l)\x>*]
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
simply because
UNIVERSIDAD [(-A FRANCISCO
DISTRITAL + V + JOSE
l)p]*w 6 L2 c L2oc c L,1^.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 211
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We are now prepared to establish (11.2.23), the main point of this lemma. Let u e
D[(-A + V + 1)|£>*] and take a sequence [un] from D> such that \\un - «||2 -*• 0. We
claim that

and

Since (-A + V + 1) and [(- A + V + l)|i>]* agree on T>, it will follow immediately
from (i) and (ii) that the L2-function [(—A + V + l)|i>]*w (which is also a distribution)
equals the distribution (—A + V + I)H in the distributional sense and so the proof of
this lemma will be complete.
(i): Fix tp e D. Then

where the second step follows from the symmetry of (—A + V + l ) o n £ > and the
fourth comes from the definition of the adjoint (see Definition 9.6.1) and the fact that
M € D[(-A + V + l)|u*] and <p e T). Hence (i) is justified.
(ii): We first claim that \\un — u\\2 -> 0 implies that un -> u in the distributional
sense. Fix <p e T> and let K — supp(^)). Then

as needed. It follows immediately that (-A + l)un -»• ((—A + \)u in a distributional
sense.
It remains only to show that for <p, K as above, fK(un—u)V<pdx -> Oasn —> oo.But
under U.S. or applicable copyright law.

where the last assertion in (11.2.24) follows from the fact that \\un — u\\2 ->• 0 and
fK V2\y>\2dx < ll^ll^j/jf V2dx < co. This finishes the proof of (ii) and so, as noted
above, the proof of the lemma. D
Note that it follows from the preceding lemma that if u e £>[(—A + V + l)|i>*],
then the distribution (-A + V + l)u lies in L2(E.d). In fact, it can be shown [Kat2,
p. 136] that D[(-A + V + l)|i>*] consists exactly of those functions u e L2(Efl!) such
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
thatUNIVERSIDAD
(-A + VDISTRITAL 6 L2. JOSE DE CALDAS
+ 1)«FRANCISCO
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 212 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We are now ready to show that if V e L2locwith V > 0, then (—A+V)|£> is essentially
self-adjoint. Following this, we will state a result—also due to Kato in [Kat2,3]—proving
the essential self-adjointness of (—A + V)\-p, where V — V+ — V- is an R-valued
potential with V+ e L2loC and with V- "relatively operator bounded with respect to HQ
with relative operator bound less than 1."
Theorem 11.2.5 (Kato) Let V e £ 2 oc (R rf ) with V > 0 pointwise. Then -A + V is
essentially self-adjoint on T>.
Proof It is easy to see that - A + V is essentially self-adjoint on V if and only if — A +
V + 1 is essentially self-adjoint on £>. We will work with the operator (-A + V + 1)|£>
since it has the advantage of being strictly positive. [In fact, ((—A + V + l)(f>, ip) > \\ip\\2
for all (p e V.] According to the implication (ii) =>• (i) in Theorem 11.2.1, it suffices to
show that if u e Z)[(-A + V + l)jx>*] and

thenw = 0. Now by Lemma 11.2.4, it suffices to show that if u e D[(-A + V + 1)|£>*]

and

then H = 0. Since u e L2 c L2oc and V e L2OC, both u and Vu are in L,1^. Thus, by
(11.2.26),

Hence, the hypotheses of Theorem 11.2.3, Kato's inequality, are satisfied and so we have
A|M| > Re [(sgn u) AM] in the distributional sense. Thus, we can write

in the distributional sense. In particular, we see that

under U.S. or applicable copyright law.

Let js be an approximate identity as in Step 2 of the proof of Kato's inequality.

[For S > 0, take js(x) := j(x/S)S~d where j > 0, j e £>, /R(/ j(x)dx = 1, and the
support of j is the closed unit ball of Md.] Let v := \u\ and i/ := v * 7'^. Now v e L2
since u e L2. Also vs e L2 since it is the convolution of the L1 function js with the
L2 function v [Fol2, (8.7), p. 232]. Further, vs is infinitely differentiate (since it is the
convolution of the locally integrable function, and hence distribution, v with j$ e T>)
Aw"5Publishing
andEBSCO = v * Aj [H6r3, Theorem 4.1.1, p. 88]. In addition, At) 6 e L2 since v e L2
: 5eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
1
Ay's e Z>DISTRITAL
andUNIVERSIDAD c L . FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 213
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We have seen above that both vs and Ai/ are in L2. Recalling from (10.2.10) that
D(Ho) = [w € L2 : Aw € L2 in the sense of distributions},
we see that vs e D(Ho). (The fact that HO equals — ^A instead of -A clearly does not
matter for the present discussion.) Since HQ is a nonnegative operator, we have

Now since A |« | is by (11.2.28) a positive distribution, there is (see (11.2.3)) a measure

fj. : B(R.d) -> [0, +00] which is finite on compact sets such that the action of the
distribution A|«| on any (p e T> is given by

Further, At/ = A(|w| * y',$) = A(|w|) * jg and At/5 is infinitely differentiable [H6r3,
Theorem 4.1.1, p. 88]. Hence, by (11.2.30),

so that At;5 is pointwise nonnegative. (Clearly, whether A \u \ is viewed as a distribution or

as a measure does not affect the value of the convolution product here.) Since vs = \ u \ * js
is nonnegative, we see that (vs, At/) > 0, and so

Combining (11.2.29) and (11.2.31), we have that

Now vs = \u\*js is infinitely differentiable with all of its derivatives in L2 (Davs —

\u\ * Dajs is the convolution of the function Daj$ e T> c L1 with \u\ 6 L 2 ). Hence we
have the formula

and so we see from (11.2.32) and (11.2.33) that Vvs = 0 a.e. In fact, since the first partials
under U.S. or applicable copyright law.

of v5 are continuous, (Vvs)(x) = 0 for all x e K rf . Thus vs is a constant function, and,

since the only constant function in L2 is the zero function, t/ = 0.
Finally, vs -» v a.e. as S —> 0+ and so v = u \ = 0 a.e. Thus u = 0 as we wished
to show. D
Remark 11.2.6 It can be shown ([Kat2] or [BreKat]) that the exact domain of the
unique self-adjoint extension of(— A + V)\v where V > 0 and V e L^OC(R'') is
{u € L 2 (M d ) : Vu is in Lloc(Rd) and the distribution (-A + V)u is in L2(R.d)}.
We assumed
EBSCO Publishing in the preceding
: eBook theorem
Academic Collection that V ->printed
(EBSCOhost) 0 pointwise.
on 6/8/2017It5:38
suffices
PM via to assume
thatUNIVERSIDAD
V is bounded below.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 214 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 11.2.7 Let V e L^oc(Rd) and let c be a nonnegative constant such that
V > — c pointwise a.e. Then — A + V is essentially self-adjoint on T>.
Proof The preceding theorem shows that — A + (V + c) is essentially self-adjoint on T>.
One can then show easily that - A + V = — A + ( V + c ) - c has a unique self-adjoint
extension and so is also essentially self-adjoint. a
It is easy to give many examples of functions V which satisfy the conditions of the
corollary. One class of examples follows.
Example 11.2.8 The operator - A + P ( x \ , . . . , xj) is essentially self-adjoint on T>(Rd),
where P is any polynomial with coefficients in R which is bounded below. If d = 1,
P can be any polynomial of even degree whose highest powered term has a positive
coefficient.
We next turn to the statement and some discussion of an essential self-adjointness
result where a certain class of potentials having an unbounded negative part is permitted.
Definition 11.2.9 Let A be self-adjoint and Bbea (densely defined) symmetric operator
on a complex Hilbert space Ji. We say that B is relatively operator bounded with respect
to A (briefly, B is A-operator bounded) with bound less than 1 if and only if

and there exist positive constants a < 1 and b such that

The infimum of all such numbers a is called the A-operator bound of B.

We note that the analogue of Remark 10.3.18(b) holds. In particular, the infimum
need not be a bound itself.
The proposition to follow will eventually permit us to compare different approaches
to the Feynman integral; particularly, the one discussed at the end of the present section
and the "modified Feynman integral" [Lai 1] defined in Section 11.4 below. (We use here
some of the notation from Sections 10.2 and 10.3.)
Proposition 11.2.10 Let A and B be self-adjoint operators on H, with A nonnegative
and Q(A) H Q(B+) dense in H. Then:
under U.S. or applicable copyright law.

(i)IfB- is A-operator bounded with bound less than 1 (as in Definition 11.2.9), then
it is also A-form bounded with bound less than 1 (in the sense of Definition 10.3.17).
(ii) Assume that B- is A-form bounded with bound less than 1, so that (by Theorem
10.3.19) A+B is well-defined. If A + B, restricted to a subspace D, is essentially self-
adjoint , then its unique self-adjoint extension (A + B)\D, coincides with A + B, the
form sum of A and B.
We note that part (ii) follows from Theorem 10.3.19 since the latter implies that
A+B is a self-adjoint extension of (A + B)\D; hence A+B must necessarily coincide
withEBSCO
(A Publishing
+ B)\D, :the unique
eBook self-adjoint
Academic extension -ofprinted
Collection (EBSCOhost) (A + onfi)|o- The
6/8/2017 proof
5:38 of (i) is left
PM via
as an exerciseDISTRITAL
UNIVERSIDAD for the FRANCISCO
reader. JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 215
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The following extension (also given in [Kat2]) of Kato's essential self-adjointness

theorem (Theorem 11.2.5) enables us to deal with certain potentials V which are
unbounded below. (Compare Corollary 11.2.7.)
Theorem 11.2.11 (Kato) Let V = V+ — V- be an R-valued Lebesgue measurable
function on Rd such that V+ 6 £j!oc(Rd) and V- is Ho-operator bounded with bound
less than 1.
Then HO + V is essentially self-adjoint on T> = T>(Rd).
Remark 11.2.12 (a) The conclusion of Theorem 11.2.11 still holds if V - V\ + Vz,
with V\ > 0, Vi e LJ!oc(R<i), and Vi assumed to be Ho-operator bounded with bound
less than 1.
(b) Theorem 11.2.11 is stated here as in [ReSi2, Theorem X. 29, p. 185] rather than as
in [Kat2, Theorem, p. 137]; it is a bit more general than in [Kat2] where the assumption
of "Ho-operator boundedness" was replaced by concrete conditions on the potential of
the type discussed below. Its proof—which can be found in [ReSi2, pp. 185-186]—is
analogous to that of Theorem 11.2.5 except for one additional argument concerning the
resolvent involved (see also [Kat2, pp. 140-141]).
(c) Earlier essential self-adjointness results can be found in ([IkKa], [Si2]), as well
as in Martin Schechter's book [Sche] and the references therein. Related results and
further extensions of Theorem 11.2.5 can be found in [ReSi2, Chapter X].
(d) For notational convenience, we have worked above with — A instead of the free
Hamiltonian HO = — 3 A. Note, however, that Theorems 11.2.5 and 11.2.11 have an
obvious counterpart for the essential self-adjointness of Ho + V = — j A + V (instead
of — A + V), under the same assumptions on the potential V. In the remainder of this
section, we will use without further comment this version of Theorems 11.2.5 and 11.2.11.
Conditions on the potential V for Ho-operator boundedness
What kinds of functions satisfy the conditions imposed on V_ in the preceding theorem?
We wish to discuss this next. Our main reference for this material is [CyFKSi, §1.2].
The classes Sd in the definition to follow were originally introduced by Stummel in
[Stu].
Note the similarity between the definition of Sd and that of Kd, introduced in Defi-
nition 10.4.1 to study Ho-form (rather than Ho-operator) boundedness. As we shall see,
Sj and Kj are different classes of functions but many of their properties are analogous.
under U.S. or applicable copyright law.

Definition 11.2.13 Let W : Rd -* R be Lebesgue measurable. The function W belongs

to Sd if and only if

Note that for d < 3, Sd = L^ oc (R d ) H . Recall that the space L£c(Rd)H has been
introduced in Definition
EBSCO Publishing 10.4.2(i).
: eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
The next DISTRITAL
UNIVERSIDAD proposition accounts
FRANCISCO JOSE DE for our interest in Sd.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 216 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proposition 11.2.14 If W € Sj, then W is Ho-operator bounded with bound less than 1.
(In fact, W has Ho-operator bound 0.)
We give some further information about Sj. and its contents.
Proposition 11.2.15 (a) We have 84 c L^ oc (M d ) u /or all positive integers d.
(b) If we take p > d/2for d > 4 and p = 2 for d < 3, then

Proof (a) For d < 3, the containment is trivially true and is actually an equality. We will
work out the case d > 5', the remaining case, d = 4, can be done in a similar manner.
Pick any number r > 0. It is easy to show that W e L^ oc (E d ) H if and only if

Suppose now that W is not in L^ oc (M d ) M . Let r e (0,1] be fixed (for now). We see from
(11.2.36) that there exists a sequence {x^} in Rd such that

Since 0 < r < 1, it follows that

Hence,

We see then that

under U.S. or applicable copyright law.

Thus W is certainly not in Sj.

We will omit the proof of (b) except for the observation that the first containment is
trivial.
Example 11.2.16 ([CyFKSi, p. 8]) If d > 5 and S e (0, 1/2], the function

is #o-operator bounded with bound less than one but is not in Sj. The function W does
belong to Sd if and only if S > 1/2. If we compare this example with Example 10.4.6
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
taking d > 5 DISTRITAL
UNIVERSIDAD and S e FRANCISCO
(1/2, 1],JOSE
weDEsee that Sj is not a subset of K^.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 217
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 11.2.17 (a) Note that Definition 11.2.13 makes perfectly good sense for
C-valued functions. However, our main interest in these definitions is in connection
with self-adjointness results and so we have kept W real-valued.
(b) The analogue of Proposition 10.4.7 still holds for the class Sd.
Feynman integral via the Trotter product formula for unitary groups
We are now prepared to state and prove the key theorem establishing the existence of
the Feynman integral via the Trotter product formula. We will actually postpone formal
reference to "the Feynman integral" until (Definition 11.2.21 and) Corollary 11.2.22.
However, the reader will recall (perhaps after reviewing Sections 7.4 and 7.5) that the
following result is related in a rather straightforward way to Feynman's original paper
[Fey2]. The proof below will not be difficult since it is simply a matter of applying
results which we now have at hand. Unlike our preliminary discussion in Chapter 7 (see
especially Theorem 7.5.1), we will simplify notation by taking the physical constants m
and h equal to one.
We begin with a lemma which is concerned just with the nth Trotter product and not
with its limit. Note that the assumptions on V in the lemma are minimal.
Lemma 11.2.18 Let V : R.d —>• K be Lebesgue measurable and finite Leb.-a.e. Then
for every t e R and y> 6 L 2 (R rf ) and for Leb.-a.e. v, we have the equality

where xn := v and where the iterated integrals on the right-hand side of (11.2.38) exist
under U.S. or applicable copyright law.

for Leb.-a.e. v e Kd when interpreted in the mean (in the sense of Theorem 10.2.7) and
define a function which is in L 2 (M rf ) as a function of v.
Proof The operator e~'"v is the operator of multiplication by the function e~'«v, and
this is the first operator that acts on (p on the right-hand side of (11.2.38). According to
the left-hand side of (11.2.38), the operator e~~'*H° should now act on e~'"V( 'V(0- But
we know from Theorem 10.2.7 that e~'nHa is precisely the second operator that acts on
the right-hand side of (11.2.38), where the integral involved should be interpreted in the
mean (at least when the function it is acting on is in Z,2 \ L 1 ). (See especially formula
EBSCO Publishing : eBook AcademicHCollection
v (EBSCOhost) - printed on 6/8/2017 5:38 PM via
(10.2.33).)
UNIVERSIDADThe product
DISTRITAL e~'nJOSE
FRANCISCO °e~'" should now be iterated n times according to the
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 218 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

left-hand side of (11.2.38). In summary, the right-hand side of (11.2.38) is exactly the
nth Trotter product (e~l«H°e~'«v)" acting on <p and evaluated at v := xn.
Theorem 11.2.19 Let V : Rd -*• R be Lebesgue measurable and such that V+ is in
L2oc(Rd) and V- is Ho-operator bounded with bound less than 1 (in the sense of
Definition 11.2.9). Also, let<p e L2(E.d).
Then, for every f e K,

defines a function which is in L 2 (R rf ) and is given for almost every v e E.d by the
integrals (interpreted in the mean) on the right-hand side of (11.2.38).
Further, there is a function \jf : (-00, +00) x E.d -> C such that ijr(t, •) e L 2 (R d )
for every t e R and

where the limit is uniform in t on all bounded subsets o/R. In fact, the function ^r(t, •)
is given by the action of the unitary group e~"H on <p; that is,

where H := (Ho + V)\x> is the unique self-adjoint extension of the essentially self-
adjoint operator (Ho + V}\x>- (We know from Proposition 11.2.10 that we must also
have H = HQ + V, the form sum of HQ and V.)
Finally, for <p e D(H), ijr(t,v) is the unique solution (in the semigroup sense as
discussed in Chapter 9) of the Schrodinger equation

with initial state (p.

Proof We will apply Corollary 11.1.6, the Trotter product formula for unitary groups,
with H = L 2 (R d ), D = T> = V(Rd), A = H0 and B = V, the operator of multiplica-
under U.S. or applicable copyright law.

tion by the R-valued function V. Now, HQ is self-adjoint as we saw in Section 10.2 (just
below Equation (10.2.8)), and, of course, V is self-adjoint as well (by Proposition 10.1.3).
Under the present hypotheses on V, (Ho + V)|£> = ( — j A + V)\x> is essentially self-
adjoint by Theorem 11.2.11. (See Remark 11.2.12(d).) We let C = H = (H0 + V) |X >.
Thus we see that the hypotheses of Corollary 11.1.6 are satisfied in our present setting.
It follows from (11.1.21) that for all <p e L 2 (M rf )

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
2 d
where the limit
UNIVERSIDAD is in the
DISTRITAL norm on
FRANCISCO JOSELDE(R ) and is uniform in t on all bounded subsets of M.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN INTEGRAL VIA THE TROTTER PRODUCT FORMULA 219
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Now we simply rewrite the nth Trotter product on the right-hand side of (11.2.43) by
means of Lemma 11.2.18. It follows (for Leb.-a.e. v) that i/rn(t, v) defined by (11.2.39)
is given by the right-hand side of (11.2.38).
Further, the fact that the limit of irn(t, v) exists as n -> oo in the sense claimed in
the theorem and equals the unitary operator e~'tH acting on <p now comes as well from
(11.2.43).
Finally, the fact that \jf(t, v) = (e~ltH<p)(v) gives the unique solution (in the
semigroup sense) of the Schrodinger equation (11.2.42) when the initial state <p is in
D(H) = (D[(H0 + V)p]) follows from Theorems 9.1.4 and 9.1.14. D

Remark 11.2.20 (a) Recall from Proposition 11.2.14 that if V- € Sd, then it satisfies the
conditions of the preceding theorem. Also, by Proposition 11.2.15, Lp(Rd) + L°°(Rd) c
Lfoc(R.d)u c Sd where p = 2 for d < 3 and p > d/2for d > 4. [If V_ e Lp(Rd) +
L°°(R.d), as in Theorem 7.5.1, then the conclusion of Proposition 11.2.14 holds even if
we take p = d/2ford > 5 (i.e. d > 4)J
(b) Theorems 7.5.1 and 11.2.19 are closely related, but the latter was not only proved
but also is considerably more detailed. In addition, the conditions on V- are more general
in Theorem 11.2.19; we assumed in Theorem 7.5.1 that V- e Lp(Rd) + Z.°°(Rd), with
p as in (a) above, and avoided discussion of relative operator boundedness and the
classes Sd.
(c) The conditions on the function V in Theorem 11.2.19 permit a much broader
class of potentials than were allowed in Nelson's 1964 paper [Nel], Nelson required
that V : Rd -> R be a finite sum of functions in Lp (W*)for values of p satisfying p > 2
and p > d/2 and functions which "after an inhomogeneous linear change of variables
are of theform W ( x 1 , . . . , Xj)where j < d and W is in Lp (Rd) for values of p satisfy ing
p > 2 and p > d/2." It is Section 1 and Appendix B, especially Theorem 8, of [Nel] that
are related to our present considerations and which the reader may wish to consult. The
increased generality of the assumptions in Theorem 11.2.19 is particularly striking for
the positive pan of the potential since it is only required that V+ e L^oc(Rd). However,
we should note that Nelson's paper was already sufficiently general to accommodate the
case of a finite number of particles with Coulomb interactions.
The Feynman integral via the Trotter product formula (briefly, TPF) is probably
the most widely known of the many approaches to this subject. Perhaps because of
under U.S. or applicable copyright law.

this and because of the close conceptual connection with the approximation approach
in Feynman's original paper (see Section 7.4), this version of the Feynman integral is
often referred to as simply "the Feynman integral". We will refer to it, however, as the
Feynman integral via TPF and write f'TP(V) for this Feynman integral associated with
the potential V.
Definition 11.2.21 Let V : Rd ->• R be Lebesgue measurable and finite Leb.-a.e. Note
that the operator (e~'"H°e~'"V)n on L2(K.d) makes sense for every positive integer
n and that for every (p € L 2 (R d ) and Leb.-a.e. v, [(e~'"Hoe^'nV)n(p](v) equals the
expression on the right-hand side of (11.2.38) with the integrals interpreted in
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
the mean
when necessary.
UNIVERSIDAD The Feynman
DISTRITAL integral
FRANCISCO JOSE via TPF associated with the potential V, denoted
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 220 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

by FjpW), is the strong operator limit

in £(L2(Rd)) whenever this limit exists. (That is,

for all ip eL 2 (R J ).j

The following corollary restates much of Theorem 11.2.19 in the language just
introduced.
Corollary 11.2.22 (Existence of the Feynman integral via the Trotter product formula)
Let V : Kd -* R be Lebesgue measurable and such that V+ e £2oc(lRd) and V- is
relatively operator bounded with respect to HQ with relative bound less than 1.
Then F'Tp(V) exists for all t &R, where the limit in (U.2.44) is, for all <p e L 2 (R d ),
uniform in t on bounded subsets of R. Further, for all t e R,

that is, f'TP(V) agrees with the unitary group which specifies the dynamics in the
standard approach to quantum mechanics. Finally, for (p 6 D(H), ^^(V)^ is the
unique solution (in the semigroup sense) of the Schrodinger equation (11.2.42) with
initial state (p.

11.3 Product formula for imaginary resolvents

We present in this section the results of Lapidus [Lai 1]—building upon his earlier work
in [La1, 2, La6-9]—which establish a product formula for the "imaginary resolvents"
of (suitable noncommuting and unbounded) self-adjoint operators. The main results
(Theorem 11.3.1, as well as its consequences, Corollaries 11.3.5 and 11.3.7) will be key
to defining and proving the convergence of the "modified Feynman integral" (also intro-
duced by the second author in [La1, 2, La6-ll,Lal3]) for very general singular potentials.
under U.S. or applicable copyright law.

(See Section 11.4.) They will also be essential to establish "dominated-type convergence
theorems" [Lal2] for this approach to the Feynman integral. (See Section 11.5.)

Hypotheses and statement of the main result

We next state the assumptions made in much of this section. Let H be a complex Hilbert
space, with inner product denoted by (•, •) and corresponding norm || • ||. Let A, B be
(typically unbounded) self-adjoint operators on H, with A nonnegative. Let B+ and
B- be the positive and negative parts of B defined through the spectral theorem as in
Section 10.3; then B+ and B- are nonnegative self-adjoint operators and B — B+ — B-.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
Assume that DISTRITAL
UNIVERSIDAD B_ is relatively/orm bounded
FRANCISCO JOSE DE CALDAS with respect to A (in short, A-form bounded)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PRODUCT FORMULA FOR IMAGINARY RESOLVENTS 221
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

with relative bound less than 1 (see Definition 10.3.17); i.e.

and there exist positive constants y < 1 and S such that

Assume, for simplicity, that Q(A) n Q(B+) (= Q(A) n Q(B) by (11.3.la)) is dense
in H. (See part (i) of Corollary 11.3.7 below for the general case.) Recall then from
Section 10.3 that the quadratic form

is bounded from below and closed, and so there exists a unique (bounded from below)
self-adjoint operator A+B associated with this quadratic form. (See Theorem 10.3.19.)
The operator A+B is called the form sum of A and B.
We are now ready to state the main result obtained by Lapidus in [Lai 1]. (See [Lai 1,
Theorem l,p. 263].)
Theorem 11.3.1 (Product formula for imaginary resolvents) Under the above hypothe-
ses, we have for all u e "H,

uniformly in t on compact (or equivalently, bounded) subsets of R.

We remark that with the notation introduced in (11.1.17"), we can restate (11.3.2)
more briefly as follows:

We will sometimes find it convenient to use (11.3.2') rather than (11.3.2). (See espe-
cially Section 11.5 below.)
The following comment may help a reader (who is not familiar with operator theory)
under U.S. or applicable copyright law.

understand the statement (as well as the proof given below) of Theorem 11.3.1. [Our
remark deals with the "imaginary resolvent" [/ + i t B ] ~ ] (in the terminology of [Lal,2,
La6-13]) but, of course, an entirely analogous comment applies to justify the existence
of the "imaginary resolvent" [7 + itA]~l of the unbounded self-adjoint operator A.]
Note that since the operator B is self-adjoint, its spectrum cr(B) is contained in R
(see Remark 10.1.7), and so its resolvent set p(B) := C \ a(B) contains the imaginary
axis j'R; hence, by Definition 9.3.1, the imaginary resolvent [1 + itB]~^ exists (as an
element of C(H)~) for all t e R. (Clearly, for t = 0, it is equal to /, the identity operator
on H, while for t ^ 0, it is equal to (i/t)R((i/t); B), in the notation of Definition 9.3.4.)
An EBSCO
important fact to keep in mind (in the proof of Theorem 11.3.1 below) is that it can
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
alsoUNIVERSIDAD
be obtained via the
DISTRITAL functional
FRANCISCO JOSE DEcalculus
CALDAS associated with the unbounded self-adjoint
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 222 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

operatorB(Theoreml0.1.11);namely,[I+itB] -1 = g(B),where g(x) :— (1+itx) - 1

for x e a(B) c R. (At this point, it may be helpful to some readers to briefly review
the various forms of the spectral theorem for unbounded self-adjoint operators discussed
in Section 10.1. Indeed, they will be used extensively in the proofs given below; see
especially Theorems 10.1.8, 10.1.11 and 10.1.14.)
Proof of the product formula
We shall need two preparatory lemmas before proving Theorem 11.3.1.
Lemma 11.3.2 Let B be a self-adjoint operator on H with positive part B+and negative
part B-. Then, for all t > 0, we have

where

are nonnegative, bounded self-adjoint (i.e. Hermitian) operators on H.

Moreover, as t -> 0+,

and

Proof A straightforward computation yields

The first part follows since B = B+ — B-.

Now by Theorem 10.1.8, the multiplication operator form of the spectral theorem,
we may assume that B = Mg, the operator on L 2 (u,) of multiplication by the R-valued
function g. Following standard notation, we use g to denote the operator Mg as well as
under U.S. or applicable copyright law.

the function itself. Let f e L 2 (u). Note that by (10.1.2) (or (10.3.11) and (10.3.13)),
/ € Q(B) = D(|B11/2) if and only if |g|1/2|f| e L 2 (u). Similarly, f e Q(B±) =
D(Bi /2 )ifandonlyifgi /2 |/| E= L 2 (u,). Thus, in particular, Q(B) = Q(B+)n Q(B-).
(Also see equation (10.3.57).) Further, one easily sees that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PRODUCT FORMULA FOR IMAGINARY RESOLVENTS 223
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(After some elementary manipulations, the first inequality in (11.3.4) follows from the
fact that 0 < x2 + t2x4 + t4x6 for all x, t in R. The last two inequalities are immediate.)
[Actually, all three inequalities can be obtained, up to a multiplicative constant, by
observing that the function x H-> x-1 (1 — r(x)) is bounded on all of R, where r(x) :=
(1 + i x ) - 1 ; note that |T| < 1, r(0) = 1 and r'(0) = -i.]
Taking f e Q(B) in the first inequality in (11.3.4), (11.3.3a) now follows from
the Lebesgue dominated convergence theorem (applied in L 2 (u)). Similarly, (11.3.3b)
[resp., (11.3.3c)] follows from the second (resp., third) inequality in (11.3.4) by taking
f € Q(B+) [resp., f e Q(B_)]. O
Lemma 11.3.3 Let {Ut}t>o, {Vt]t>o be families of bounded operators on a Banach
space X such that U, exists as a possibly unbounded operator. Assume that || Vt || < 1
and that there is a constant k > 0 such that

Then, for sufficiently small t, [U, — V t ] - 1 is a bounded operator on all of X and

Moreover, if Ut v —> v as t —> 0+, then

Proof Since ||U,V,|| < ||Ut|| ||V,|| < 1 - kt + o(t) < 1 for small t, it is well-known
[Ru2, Theorem 18.3, p. 357] that [/ — Ut V t ] - 1 is a bounded operator defined on all of
X and has a Neumann series expansion in £(X):

The operator Ut-1 may be unbounded and hence U-1 — V, is only defined on the
subspace D ( U - 1 ) . Let us show that U, — Vt is invertible, Clearly
under U.S. or applicable copyright law.

since the equality U, — V, = U, (I — U1Vt) between unbounded operators holds.

Similarly, if J denotes the inclusion mapping of D ( U - 1 ) into X,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
Hence U-1 -DISTRITAL
UNIVERSIDAD V, is invertible andDE(11.3.5)
FRANCISCO JOSE CALDAS holds.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 224 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Observe that

[Here, 0(1) (resp., o(l)) denotes a function which is bounded (resp., vanishes) as t ->
0+.] This follows since

Now, by (11.3.5),

Thus (11.3.6) follows immediately from (11.3.8). D

Remark 11.3.4 Lemma 11.3.3 is due to Kato [Kat7, Lemma 4.1, p. 112]. Originally,
Kato stated his lemma in a slightly different form, assuming instead that || Ut || < e-1,
since he was primarily interested in the behavior of semigroups and not of resolvents.
We are now ready for the proof of Theorem 11.3.1.
Proof of Theorem 11.3.1 Fix A > 0. For t > 0, set

and

Note that Lemma 11.3.3 applies to the families {U t } t>0 and {V,} t>0 since

Clearly,
under U.S. or applicable copyright law.

since, by (11.3.5) and (11.3.7),

Note further that U-1 = I +1 (A + i A) and that Ran(W t,y )on=6/8/2017

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed
D ( U - 15:38
) =PM D(A).
via
Hence
Wt,yUNIVERSIDAD
E D(A)for allFRANCISCO
DISTRITAL u E H. JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PRODUCT FORMULA FOR IMAGINARY RESOLVENTS 225
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The remainder of the proof is divided into two steps. In the first step, we shall hold
the parameter A fixed; we shall let it tend to zero at the end of the second step.
Step 1: Let A > 0. We will show that for every v € H, Wt,v -> [A + i C ] - 1 v as
t -> 0+, where we have set C = A+B.
To see this, fix v e H and put wt = Wt,y v. As noted above, wt e D(A). Thus w, e
Q(B-) since, by (11.3.la), D(A) c Q(A) c g(B_). Moreover, from the definition
of Wa,

Thus

Also, we have

and

where RB, , Ig+ and IB- are the nonnegative, bounded self-adjoint operators defined in
Lemma 11.3.2.
Taking the inner product of v against wt and combining (11.3.10) and (11.3.11), we
see that

Therefore

By (11.3.9),
under U.S. or applicable copyright law.

Hence by the Cauchy-Schwarz inequality,

Recall that IB- = B-[I+ t 2 B 2 ] - 1 . Thus by the spectral theorem (Theorem 10.1.8), we
see much as in the proof of Lemma 11.3.2 that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
(Indeed, (11.3.15)
UNIVERSIDAD follows
DISTRITAL by JOSE
FRANCISCO using
DE the third inequality in (11.3.4) but for f 0 Q(B-).)
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 226 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Applying the key observation (11.3.15) to £ = wt and recalling (11.3.1b), we obtain

(As was noted earlier, we have E = wt e Q(A) c Q(B-).) Consequently

and, by (11.3.12),

Using (11.3.13) and (11.3.14), we deduce that

Since y < 1, it follows that Al/2wt and I 1B/ 2 +w, are bounded as t -> 0+. By (11.3.16)
t
1/2
(and (11.3.13)), I B J wt is bounded as well. In view of (11.3.12) and (11.3.14), it is
t
immediate that wt and RB' wt are also bounded. Thus there exists a sequence of positive
numbers [tn] (tn -> 0+) along which

(for some vectors w, a, a, B+ in H), where the symbol -^ denotes weak convergence
in the Hilbert space H. (Here and in the next few lines, it is implicitly understood that
the limits are taken along the sequence {tn}. Further, recall that, for example, wt —> w
means that for every fixed y e H, ( w t , y) -> (w, y) in C.)
Let y e Q(A). Equation (11.3.17) implies that

[Here and hereafter, we also use the following easily verified fact. If wt -> w weakly in
H and pt —> p strongly in (the norm of) W, then (wt, pt) -> (w, p) in C.] Since A 1 / 2
is self-adjoint, it follows that w € Q(A) and a = Al/2w.
under U.S. or applicable copyright law.

Similarly, let y e Q(B) - Q(B+) n Q(B-). Then

where we have used (11.3.3a) in Lemma 11.3.2 and (11.3.17). Since Q(B) is dense in
H, it follows that a = 0.
Lety e Q(B+). By (11.3.3b) in Lemma 11.3.2 and (11.3.17),

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
Hence w e Q(B
UNIVERSIDAD +) and
DISTRITAL Bl+DE/2w.
0+ =JOSE
FRANCISCO CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PRODUCT FORMULA FOR IMAGINARY RESOLVENTS 227
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In the same manner, we see by using (11.3.3c) in Lemma 11.3.2 that ft- = B_i w.
Since Q(C) = Q(A) n <2(B+), we may summarize the above discussion as follows:

Now let y e D(C). Note that D(C) c. Q (C) C Q(B-) n Q(B+), since Q(A) C
Q(B-)by(11.3.1a). Then

where we have used successively (11.3.10), (11.3.11) and, after passing to the limit
along [tn], (11.3.17), (11.3.18) and (11.3.3) in Lemma 11.3.2. Hence we have ( i - 1 ( v -
Xw), y) = (w, Cy) for every y e D(C). Since C is self-adjoint, it follows that
w € D(C),

and so

The limits in (11.3.17) are, therefore, independent of the sequence {tn}. By a standard
compactness argument (namely, the existence of a weak limit point for a sequence in
a weakly compact set implies the existence of its weak limit), it follows that weak
convergence holds as t -> 0+ in (11.3.17). In particular, w, -> w and RB' w, -^ 0,
where w is given by (11.3.19b).
Passing to the limit as t -»• 0+ in the real part of (11.3.12) and using (11.3.19a), we
under U.S. or applicable copyright law.

see that

Consequently,

Since X > 0, it follows in particular that w, -»• w (in the norm of Ti.), as required. This
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
completes theDISTRITAL
UNIVERSIDAD proof ofFRANCISCO
Step 1.JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 228 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Step 2: By Step 1 and equation (11.3.6) of Lemma 11.3.3, we have for every A. > 0
(with A. still fixed) and v e H,

Set F(t) = UtVt. Note that ||(1 + Xt)F(t)\\ < 1. (This follows since ||F(r)|| <
IIU, || || V, || < (1 + X t ) - 1 ) Furthermore,

and (11.3.20) yield for every v e H,

Now we apply Theorem 11.1.8, Chernoff's theorem, to the family of contractions

{(1 + kt)F(t)}t>o and to the m-dissipative operator — iC (see Definition 9.4.5 and
Theorem 9.4.7) which generates the contraction semigroup {e-itc<c : t > 0}. This along
with (11.3.21) implies that for all u e U,

uniformly for t € [0, T], for all T > 0.

Fix T > 0 and t with 0 < t < T. Clearly,

At this stage, we recall that Ut, and hence also F(t), depends on A.. Now

In the next to last inequality, we have used the resolvent equation (see Proposition 9.3.5):
under U.S. or applicable copyright law.

In view of (11.3.23) and (11.3.24),

Now an-bn = En-1 a n - 1 - k (a-b)b k for arbitrary a ,b in L(H).Hence' if ||a||<1

and || b|| < 1, we have
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PRODUCT FORMULA FOR IMAGINARY RESOLVENTS 229
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

With the obvious notation, (11.3.25) and (11.3.26) yield

It now suffices to let X -> 0+ in (11.3.27) to deduce from (11.3.22) that (11.3.2) holds,
uniformly for t e [0, T] for all T > 0. (Recall that C = A + B.)
By replacing A, B with their negatives, we would reach the same conclusion for
t € [-T, 0]. This completes the proof of Step 2 and of Theorem 11.3.1. D

Consequences, extensions and open problems

We now state an immediate—but important (see Remark 11.3.6)—corollary of Theorem
11.3.1.
Corollary 11.3.5 (Arbitrary nonnegative self-adjoint operators) Let A and B be arbi-
trary nonnegative self-adjoint operators on the complex Hilbert space H. Assume, for
simplicity, that their common form domain, Q(A) n Q(B), is dense in H. (See Corollary
11.3.7(ii) below for the general case.)
Then, the form sum A + B is a well-defined (and hence densely defined) self-adjoint
operator on H, and we have (with the notation introduced in (11.1.17"))

uniformly in t on compact subsets of R.

Remark 11.3.6 (a) Corollary 11.3.5 (as well as part (ii) of Corollary 11.3.7 below) was
first obtained by the second author in [La6, Theorem5.1,p. 1720]. (See also [LaL,2 ].) Its
original proof, however, was more involved than the proof of the more general "product
formula for imaginary resolvents" statedin Theorem 11.3.1. It was using, in particular,
techniques from papers by Kato [Kat5] and Ichinose [Icl]; the latter paper was also
motivated by problems connected with the Feynman integral.
(b) Under the assumptions of Corollary 11.3.5 (and, in fact, under the more general
hypotheses of part (ii) of Corollary 11.3.7 below), Kato [Kat4,5] has obtained a "product
formula for an arbitrary pair of self-adjoint semigroups"; namely,
under U.S. or applicable copyright law.

as well as

uniformly in t on compact subsets of [0, +00). [Note that we now deal with self-adjoint
(contraction) semigroups (by Theorem 9.6.7, since A and B are nonnegative) or real
(rather than imaginary) resolvents in (11.3.29) or (11.3.30) respectively.] More pre-
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
cisely [Kat5],DISTRITAL
UNIVERSIDAD under the hypotheses
FRANCISCO of Corollary 11.3.7(ii) below, the right-hand side of
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 230 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(11.3.29) or (11.3.30) must be replaced by e~t ( - A+B) n, where U denotes the projection
onto the closed subspace H1.— Q(A) n Q(B). These results have been extended to
nonlinear semigroups generated by the subdifferentials of convex functionals by Kato
and Masuda [KatMasu], except that the conclusion of [KatMasu] in the linear case is
somewhat weaker when Q(A) n Q(B) is not assumed to be densely defined. [In the non-
linear setting (proper, lower semicontinuous) convex functionals are a natural analogue
of (nonnegative) quadratic forms.]
Both the results in [Kat5] and [KatMasu] have been further extended by the second
author [La5] (and [Lal,3]) to "average formulas " (coined by analogy with the arithmetic
mean and the geometric mean, in particular). Hence, for instance, (11.3.29) becomes

Similarly, the conclusion of the counterpart of Corollary 11.3.5—obtained in [La6,

§4, esp. Theorem 4.1 and Corollary 4.1, pp. 1710 and 1717]—states that

(Actually, it also extends to an arbitrary finite number of nonnegative self-adjoint oper-

ators; see [La6, Corollary 4.1, p. 1717].)
We note that none of formulas (11.3.29)-(11.3.32) holds (or even makes sense) when
B has an unbounded negative part, as in Theorem 11.3.1 (or Corollary 11.3.7(i) below).
We will further discuss this point in Remark 11.7.5.
(c) Even when A and B are arbitrary nonnegative self-adjoint operators (with a
dense common form domain)—and let alone under the more general assumptions of
Theorem 11.3.1 (or Corollary 11.3.7 below)-the counterpart for unitary groups of the
product formula (11.3.28) is not known. Recall that in Corollary 11.1.6, the "product
formula for unitary groups", we had assumed that A + B, the algebraic sum of A and
B defined on D(A) n D(B), is essentially self-adjoint; that is, A + B has a unique
self-adjoint extension, necessarily its closure A + B; further, by Proposition 11.2.10(ii),
we must have A + B = A + B, the form sum of A and B, under the assumptions of
Theorem 11.3.1 (or Corollary 11.3.5).
The problem of extending the product formula for unitary groups (Corollary 11.1.6) to
under U.S. or applicable copyright law.

more general situations (such as those considered in the present section) is an extremely
difficult one which has been intensively investigated in the 1970s and the early 1980s;
see, in particular, the works of Paris [Far 1], Friedman [Fri], Chemoff[Cher2], Ichinose
[Icl], Lapidus [Lal,2, La4, La6—11], and the references therein. To our knowledge, it
still remains an open problem (which we will formally state in Problem 11.3.9 below).
It is, of course, motivated 2 by Nelson's approach to the Feynman integral via the Trotter
product formula [Nel] described in Section 11.2 above.
(d) Recall from Example 10.3.21 that there exist nonnegative self-adjoint operators A
andEBSCOB Publishing
in H ::= L (R)suchthatQ(A)r\Q(B)isdensein'HwhereasD(A)r\D(B) = {0}.
eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
Consequently, their form
UNIVERSIDAD DISTRITAL sumJOSE
FRANCISCO A +DE BCALDAS
is well defined (and, in particular, densely defined
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PRODUCT FORMULA FOR IMAGINARY RESOLVENTS 231
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

in H), whereas their algebraic sum A + B (with domain D(A) D D(B) = {0}) is trivial
(and hence certainly not essentially self-adjoint). In this situation, the product formula
for imaginary resolvents (Theorem 11.3.1 or Corollary 11.3.5) still provides meaningful
information although the product formula for unitary groups (Corollary 11.1.6)—at least
in its current form (see (c) above)—is useless.
Corollary 11.3.7 Let A be an arbitrary nonnegative self-adjoint operator and let B =
B+ — B- be a self-adjoint operator with an arbitrary positive part B+ and a negative
part B- which is A -form bounded with bound less than 1. (Note that we no longer require
that Q(A) D Q(B+) is dense in W.)
(i) Then the form sum A + B can be viewed as a semibounded self-adjoint operator
on the closed subspace H1 := Q(A) n Q(B+) c H and, for all u E H1,we have

uniformly in t on compact subsets of R

(ii) Hence, in particular, if in Corollary 11.3.5 above, we only suppose that A and
B are arbitrary nonnegative self-adjoint operators on H (without assuming the density
of Q(A) n Q(B)) then the conclusion (11.3.28) must be replaced by (11.3.33), with u
in H1.
Proof (i) This follows from the proof of Theorem 11.3.1 and from refinements of the
classical approximation theorems which can be found in ([Kat5], [Kat8, §IX.3]).
(ii) is an obvious consequence of (i); note that B+ = B in this case. D
Remark 11.3.8 Observe that if Q(A) n Q(B+) (which equals Q(A) n Q(B) since
Q(B+) c Q(A)) is dense in H, then H1 = Q(A) n Q(B) = H, and hence pan
(i) (resp., (ii)) of Corollary 11.3.7 yields Theorem 11.3.1 (resp., Corollary 11.3.5) above.
The following open problem is unusually challenging—as was noted in Remark
11.3.6(c)—and would have important consequences for Nelson's approach to the
Feynman integral via the Trotter product formula (discussed in Section 11.2 above).
In this form, it was stated in [La6, p. 1738] for the special case (ii), and then extended
in [Lall] (or [Lal3, Part I]) to the more general situation considered in case (i). (The
answer is conjectured to be affirmative in [La6, §7, Conjecture, pp. 1737-1738].) Despite
attempts by many investigators, it does not seem to have yet been resolved.
under U.S. or applicable copyright law.

Problem 11.3.9 (General product formula for unitary groups?)

(i) Is it true that for (noncommuting, unbounded) self-adjoint operators A and B
satisfying the assumptions of Theorem 11.3.1, we have

where A + B denotes the form sum of A and B?

(ii) Does (11.3.34) hold under the more restrictive assumptions of Corollary 11.3.5,
thatEBSCO
is, for an arbitrary pair of nonnegative self-adjoint operators (with Q(A) n Q(B)
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
dense in H?DISTRITAL FRANCISCO JOSE DE CALDAS
UNIVERSIDAD
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 232 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 11.3.10 (a) The point is, of course, that in contrast to the usual Trotter product
formula for unitary groups (Corollary 11.1.6), we no longer assume that the algebraic
sum A + B, defined on D(A) n D(B), is essentially self-adjoint.
(b) It is shown in [La6, Proposition 3.1, p. 1706] (Proposition 11.7.3 below)—by
means, in particular, of results of Chernoff[Cher2] and Kato [Kat5], aswell as of Vitali's
theorem [HilPh, Theorem 3.14.1, p. 104] (Theorem 11.7.1(i) below) for sequences of
analytic functions—that if the (strong operator) limit on the left-hand side of (11.3.34)
exists (with A and B nonnegative, but otherwise arbitrary), then Q(A) D Q(B) is dense
in H, the limit in (11.3.34) is necessarily a unitary group, say e~itc, and in addition,
we must have C = A + B, the form sum of A and B. [See also [La6, Proposition 3.2,
p. 1708] (Proposition 11.7.4 below) for some positive results.] We note that this partly
justifies the conclusion (11.3.33) obtained in Corollary 11.3.7. Hence, the problem (in
this situation) is not to identify the limit but rather to prove its existence.
We refer to Appendix 11.7 at the end of this chapter for a more thorough discussion
of the results of [La6] alluded to in Remark 11.3.10(b), as well as of the closely related
issue of the "analytic continuation" of formulas such as (11.3.29) into (rather) weak
forms of (11.3.34). (See, in particular, Proposition 11.7.4.)

11.4 Application to the modified Feynman integral

In this section, we will apply the above product formula for imaginary resolvents
(Theorem 11.3.1) from [Lall] in order to define the "modified Feynman integral",
FM(V), introduced by Lapidus in [Lal,2, La6-ll, Lal3] (and especially [Lall] in
the present generality). More precisely, we will apply, in particular, Theorem 11.3.1 to
the self-adjoint operators A = HQ = —5 A and B = V, the operator of multiplication
by a suitable (but possibly very singular) "potential" function V : R.d —> R, acting in
the complex Hilbert space H = L2(Rd).
This will enable us to establish the convergence of the "modified Feynman integral"
F'M(V) in (essentially) the most general case when the energy operator (or Hamiltonian)
admits a self-adjoint realization that is bounded from below. (In particular, in contrast
to the Feynman integral via the Trotter product formula F t T p ( V ) studied in Section
11.2, we will no longer have to require that H0 + V is essentially self-adjoint. Compare
under U.S. or applicable copyright law.

Theorems 11.2.19 and 11.4.2.) Aside from a few exceptions, this covers all the situations
of interest in the applications. We will address, however, in Chapter 13, the case when
the energy operator is unbounded from below. (See Sections 13.4 and 13.5, which deal
with the appropriate extensions of the definition of fTp(V) and fM(V), respectively.)
Towards the end of this section, we will also briefly discuss the "modified Feynman
integral" on a smooth Riemannian manifold (instead of Rd) and the case of the
Schrodinger equation with singular scalar as well as magnetic vector potentials. (See
Examples 11.4.10 and 11.4.12, respectively.)
Finally, we mention that we will only consider here the case when the scalar potential
V EBSCO
is real-valued—and hence the corresponding quantum system is conversative (or
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
isolated), which
UNIVERSIDAD is theFRANCISCO
DISTRITAL situation that
JOSE DE is traditionally considered in mathematical physics.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATION TO THE MODIFIED FEYNMAN INTEGRAL 233
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The case when V is complex-valued—and hence the quantum system is dissipative (or
"open" [Ex])—will be dealt with in Section 11.6 (as well as in Section 13.5).
Modified Feynman integral and Schrodinger equation with singular potential
Let HQ = —2A denote the free Hamiltonian acting in L 2 (R d ), as defined in Section
10.2.
We first recall from Section 10.2 some facts concerning the imaginary resolvent
[I + it H0]-1 of HO, with t € R. Let G = G(x, y; t) : Rd x Rd x R -> C denote the
free Green's function defined in Theorem 10.2.14; that is, for t £ R, t = 0, G( . , .; t) is
the integral kernel of the convolution operator [/ + it Ho]-1 acting in L 2 (R d ). Hence,
for all <p € L 2 (R d ) and for Leb.-a.e. x e Rd, we have

where the integral on the right-hand side is an ordinary Lebesgue integral (and hence
converges absolutely). (Of course, for t = 0, [/ + it Ho]-1= I is simply the identity
operator, which we continue to write in the form (11.4.1).) In the present case of Euclidean
space, G(x, y; t) is (for fixed t) a function of x — y alone. Further, by Theorem 10.2.14,
when t > 0, we have for d = 1

and for d = 3,

In (11.4.2a) and in the following, we use the analytic determination of ^/z which is
positive for z > 0; so that, for example, (—i) 1 / 2 = (1 — i>/V2- (For t < 0, we must
replace in (11.4.2) (1 — i ) / t by (1 + i)/^/\f\ inside the exponentials; see Theorem
10.2.14. Therefore, the formulas in (11.4.5) below will have an obvious counterpart for
t <0.)
We now proceed much as at the end of Section 11.2 (beginning with Lemma 11.2.18),
except that unitary groups are replaced by imaginary resolvents. We will also point out
under U.S. or applicable copyright law.

differences with that situation along the way.

Lemma 11.4.1 Let V : Rd -> R be Lebesgue measurable and finite Leb.-a.e. Then for
every t e R and (p e L2(Rd) and for Leb.-a.e. v e Rd, we have the equalities

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 234 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where xn := v and where the n-fold iterated integrals in the first equality of (11.4.3)
as well as the multiple integral in the second equality of (11.4.3) exist for Leb.-a.e.
v € ~Rd as ordinary Lebesgue integrals, and define a function which is in L 2 (R d ) as a
function of v.
Here, G n (xo, x1, . . . , xn; t) is the nth iterated kernel of the convolution operator
[I + i(t/n)H0] - 1 given by (11.4.1) (with t replaced by t/n):

In particular, it follows from (11.4.2)-(11.4.4) that when t > 0, we have ford — 1

and for d = 3 (the case of a single quantum particle moving in R3),

Moreover, for t < 0 (instead of t > 0), the expression of Qn(xo, x1, . . . , xn; t) in
(11.4.5) is the same as above except for (1 — i)/V?/« replaced by (1 + i)/T/\t\Jn
inside the exponentials.
under U.S. or applicable copyright law.

Proof Fix p € L 2 (R d ). The operator [/ + i(t/n) V]-1 is the operator of multiplication

by the function (1 + i (t/n) V ) - 1 , and this is the first operator that acts on p on the right-
hand side of the first equality of (11.4.3). According to the left-hand side of (11.4.3), the
integral operator [I+ i ( t / n ) H o ] - 1 must now act on [l+i(t/n)V(-)]~V(-)-By (11.4.1)
(Theorem 10.2.14), we have for all f e L2(Rd) and Leb.-a.e. w € Rd,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATION TO THE MODIFIED FEYNMAN INTEGRAL 235
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the integral is an ordinary Lebesgue integral. (Note that since V is M-valued,
\(l+i(t/n)V(y))-1\< 1.)
Iterating (11.4.6) n times, we see that the n-fold iterated integral on the right-hand
side of the first equality of (11.4.3) coincides precisely with the nth product ([/ +
i(t/n)Ho]~l [I + i(t/n)V]~l)n acting on q and evaluated at v := xn.
The second equality of (11.4.3) follows from the Fubini theorem and the rapid decay
at infinity of G(x, y; t), viewed as a function of \\x — y\\ alone. D
We can now state the main result of this section (see also Corollary 11.4.5 below),
which establishes the existence of the "modified Feynman integral" (Definition 11.4.4
below) for a rather large class of potentials V; essentially, the largest class for which
the Hamiltonian H (or rather, the associated "energy functional" or quadratic form) is
bounded from below and hence the Schrodinger equation can be defined without ambi-
guity. (At this point, the reader may wish to review some of the material in Sections 10.3
and 10.4 concerning quadratic forms and their applications; see also Remark 11.4.3(d)
below.)
Theorem 11.4.2 Let V : Rd -> R be Lebesgue measurable and such that V+ is in
L 1 ( R d ) and V_ is Ho-form bounded with bound less than 1 (in the sense of Defi-
nition 10.3.17). Also, let p € L2(Rd).
Then, for every t e R,

defines a function which is in L 2 (R d ) and is given for almost every v e Rd by the

finite-dimensional Lebesgue integrals on the right-hand side of (11.4.3).
Further, there is a function ty : (-00, +00) x Rd -> C such that i/f(t, .) e L2(Rd)
for every t 6 R and

where the limit is uniform in t on all bounded subsets of R. In fact, the function i(r(t,. )
is given by the action of the unitary group eith on p; that is,
under U.S. or applicable copyright law.

where H := HO + V is the form sum of the self-adjoint operators HO and V. [Recall

from Theorem 10.3.14 and Exercise 10.4.10 that H is the unique (boundedfrom below)
self-adjoint operator acting in L2(Rd) and associated with the semibounded quadratic
form (or "energy functional") q(u) := 1/2 fRd ||Vu||2 + fRd V|u| 2 .]
Finally, for p e D(H), fy(t, v) is the unique solution (in the semigroup sense dis-
cussed in Chapter 9) of the Schrodinser equation

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
withUNIVERSIDAD
initial state p.
DISTRITALFRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 236 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof We apply Theorem 11.3.1, the product formula for imaginary resolvents, with
H = L2(Rd), A = H0 and B = V, the operator of multiplication by the R-valued
function V. As we know, HQ is self-adjoint (see Section 10.2) and V is self-adjoint
as well (by Proposition 10.1.3). Further, HO is nonnegative and, by assumption, (the
multiplication operator by) V_ is Ho-form bounded with bound less than 1. Hence the
hypotheses of Theorem 11.3.1 are satisfied and so for all p e L2(Rd)

where the limit is in the norm on L 2 (R d ) and is uniform in t on all bounded subsets of
R. (Here, in the notation of Theorem 11.3.1, we have let C - A + B - HQ + V = H.)
We now simply use Lemma 11.4.1 to rewrite the nth Trotter-like product on the
right-hand side of (11.4.11). It follows that (for Leb.-a.e. v) on(t, v) defined by (11.4.7)
is given by either one of the Lebesgue integrals on the right-hand side of (11.4.3).
Further, the fact that the limit of 4 n ( f , v) as n -> oo exists in the sense claimed in
the theorem and equals ( e - i t H p ) ( v ) also comes from (11.4.11).
Finally, since H = HO + V is self-adjoint, the fact that (t, v) = ( e i t h p ) ( v ) yields
the unique solution (in the semigroup sense) of the Schrodinger equation (11.4.10) when
the initial state q is in D(H) follows from Theorems 9.1.4 and 9.1.14 (as well as from
Section 9.6). D
Remark 11.4.3 (a) Recall from Proposition 10.4.4 that if V- belongs to the Kato class
Kd, then it satisfies the conditions of the preceding theorem. Also, by Proposition
10.4.5(b), LP(Rd) + L°°(R d ) c L p . ( R d ) y c L,poc(Rd)u c Kd, where p = 1 for
d = 1 and p > d/2 for d > 2. Further, by Remark 10.4.9(a), we can allow p = d/2
when d > 3 if V_ belongs to the smaller class (Lf oc (R d ))u u Lp(Rd) + L°°(Rd). (We
refer to Section 10.4 for the notation used here.)
(b) The fact that V+ belongs to L1oc(Rd) can be greatly relaxed provided we do not
require the quadratic form q and the operators involved to be densely defined in H. =
L2(E.d). Indeed, if we assume that V+ : Rd -> [0, +00] is Lebesgue measurable (but
otherwise arbitrary), then under the same hypotheses as above on V-, H = HO + V is a
self-adjoint operator on H1:.= Q(q) (rather than H and the counterpart of (11.4.8)—
(11.4.10) holds when restricted to (p in H1. (In the analogue of (11.4.10), we would
still require that p E D(H) but now D(H) c H1.) This follows from Corollary 11.3.7.
under U.S. or applicable copyright law.

[Naturally, according to an observation made later in [Jo6] in a related context, we

can allow V+ e L1loc(Rd\G), where G is an arbitrary closed subset of Rd of Lebesgue
measure 0, without changing the fact that H is densely defined in H = L 2 (R. d ); this
follows since then, L2(Rd) = L2(Rd\G).]
(c) Recall that in Example 10.3.21, we have discussed a very singular positive poten-
tial V such that the form sum Ho+V is densely defined in L 2 (R d ) but the domain of the
algebraic sum HO + V is reduced to {0}. In this case, the familiar space D = D ( R d )
is contained in the form core. It is possible, however, to have a form core that is much
different than the usual ones but such that the form sum is still densely defined. In Exam-
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
ple UNIVERSIDAD
13.7.20 and Remark 13.7.21, we will discuss a highly singular potential V such
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATION TO THE MODIFIED FEYNMAN INTEGRAL 237
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

that Ho+V is densely defined but no nonzero continuous function can be in any of its
form cores. (In fact, for every p > 0, V is nowhere p-th power integrable; in particular,
V E Lfa. (R d ).) Theorem 5.7 of [AlMal] implies that there is a form corefor Ho+V such
that all the functions in the form core are bounded, compactly supported and "quasi-
continuous." This example and the theorem of Alheverio and Ma rest on fine properties
of Brownian motion, some of which will be discussed in Section 13.7.
(d) We will not discuss here the case of "potentials" (or "interactions") which are
given by (suitable) measures u rather than by functions V (on Rd). This will be the object
of Section 13.7 (based on the later work of Albeverio, Johnson and Ma in [AlJoMa]), in
the context of the "analytic in time (operator-valued) Feynman integral". However, we
mention that under the assumptions of the main result of Section 13.7, Theorem 13.7.16,
the product formula for imaginary resolvents of [La] 1] (Theorem 11.3.1) can still be
applied (since its hypotheses hold, by Remark 13.7.24(b)), and hence the "modified
Feynman integral" still exists for "interaction measures" u, = u+ — u- as in Theorem
13.7.16. [Inparticular, u+ is a "smooth measure" (e.g. the area measure on the surface
of a sphere in R3) and u_ is in the analogue of the Kato class for measures; see Section
13.7 for the relevant terminology.] We note that the Hamiltonian H = HQ+u—as well
as the associated quadratic form (or "Dirichlet form")—is still bounded from below in
that case. We point out, however, that in order to obtain a complete analogue of Theorem
11.4.2 (and of Corollary 11.4.5 below) in this broaderframework, there remains to obtain
a concrete expression for the action in L 2 (R d ) of the resolvent operator [I + it u , ] - 1 ;
see Remark 13.7.24(b).
Definition 11.4.4 (Modified Feynman integral) Let V : Rd -> R be Lebesgue measur-
able. Note that the operator ([I+i(t/n)Ho]~l[I+i(t/n)V]~1)" on L 2 (R. d ) makes sense
for every positive integer n and that for every cp e L 2 (R d ) and Leb.-a.e. v € Rd, the
number [([I + i ( t / n ) H o ] - 1 [I + i ( t / n ) V ] - 1 ) n p ] ( u ) equals either one of the expressions
on the right-hand side of (11.4.3), with the integrals interpreted as ordinary Lebesgue
integrals. The modified Feynman integral associated with the potential V, denoted by
Ft m(V), is the strong operator limit

in £(L2(Rd)) whenever this limit exists. (That is,

under U.S. or applicable copyright law.

for all p € L2(Rd).)

We next restate in the above language much ofTheorem 11.4.2.
Corollary 11.4.5 (Existence of the modified Feynman integral for a semibounded
Hamiltonian with highly singular potential) Let V : R.d —> R be Lebesgue measurable
and such that V+ e L 1 o c (R d ) and V- is relatively form bounded with respect to HQ with
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
relative bound
UNIVERSIDAD less than
DISTRITAL 1. (This
FRANCISCO is CALDAS
JOSE DE the case, for example, if V_ belongs to any of the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 238 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

classes LP(E.d) + L°°(E.d) c Lfoc(Rd)z c Lf oc (R d ) u c Kd, where p = 1 f o r d = 1

and p > d/2ford > 2; see also Remark 11.4.3(a) above.)
Then FM(V) exists for all t e R, where the limit in (11.4.12') is, for all p € L 2 (R. d ),
uniform in t on bounded subsets of R Further, for all t € R,

that is, FM (V) agrees with the unitary group which specifies the dynamics in the standard
approach to quantum mechanics. Finally, for p € D(H) and t e R, J-'M(V)p is the
unique solution (in the semigroup sense) at time t of the Schrodinger equation (11.4.10)
associated with the semibounded Hamiltonian H and with initial state (p.
Remark 11.4.6 In Theorem 11.4.2 (and Corollary 11.4.5), our assumptions on both the
positive part and the negative part of the potential V are significantly more general than
those made in Theorem 11.2.19 (and Corollary 11.2.22) which guaranteed the existence
of Ft (V), the Feynman integral (associated with V) via the Trotter product formula.
Indeed, if V+ belongs to L^oc (Rd), then it obviously belongs to L1loc (Rd) (see also Remark
11.4.3(b) according to which V+ can be an arbitrary nonnegative measurable function);
moreover, it follows from Proposition 11.2.10(i) that if V- is Ho-operator bounded with
bound less than 1, then it is also Ho-form bounded with bound less than 1 (but, of
course, the converse is not true in general). Hence, under the hypotheses of Theorem
11.2.19 (or Corollary 11.2.22), both F t T P ( V ) and f'M(V) exist (and coincide with the
unitary group e - i t H , since then, by Proposition 11.2.10(ii) and Theorem 11.2.11, H =
(Ho + V)\D = HO + V). We will pursue this comparison early in Section 13.4. [Recall
that F'Tp(V) is only known to exist when the algebraic sum HO + V is essentially
self-adjoint since it requires the convergence of the Trotter product formula for unitary
groups (Corollary 11.1.6) rather than for imaginary resolvents (Theorem 11.3.1), as for
F'M(V); see Problem 11.3.9.]
The following example provides a simple situation where fM(V) exists although
HQ + V has many self-adjoint extensions (and hence is not essentially self-adjoint).
Example 11.4.7 (Strongly singular central potential) Assume that d = 3, for simplicity.
Set
under U.S. or applicable copyright law.

where \\x \\ denotes the length of AC e R3 \ (0) and a, B are positive constants. We are thus
considering a single (nonrelativistic) quantum-mechanical particle moving in R3 under
the influence of the attractive central potential V. (Here, V is said to be attractive if V
is negative, and repulsive if V is positive.) The larger the exponent a, the more singular
and attractive V is. (Note that for a = 1, V is just the standard Coulomb potential which
plays a crucial role in quantum mechanics.)
Then, for anyB,Ho+V is essentially self-adjoint on C^(R3\{0}) as long asa < 3/2.
Hence, in region (I) of Figure 11.4.1 (0 < a < 3/2), both Fp(V) and f'M(V) exist
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
(andUNIVERSIDAD
coincide). However,
DISTRITAL forJOSE
FRANCISCO arbitrary ft, the hypothesis that V_(= \V\) is Ho-form
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATION TO THE MODIFIED FEYNMAN INTEGRAL 239
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

bounded with bound less than 1 holds for all a < 2 and, in the limiting case where
a = 2, for B < 1/4. (See [Far2, Example, pp. 96-97] and [ReSi2, pp. 169-170], [Kat8,
Example VII.4.15, p. 402]; note for comparison purposes that in the present situation,
we have HO = — 1 A rather than —A, which explains the factor 1/2 in the definition of
the potential V in (11.4.14).) For 3/2 < a < 2 or a = 2 with ft < 1/4 (region (II) of
Figure 11.4.1), HQ + V need not be essentially self-adjoint and hence one does not know
whether it is possible to use Nelson's definition of the Feynman integral, F t T P ( V ) . (For
a = 2 and ft > 0 arbitrary, for example, Ho + V has many self-adjoint extensions and
hence is certainly not essentially self-adjoint; this follows, e.g., from [Far2, Cases 2 and
3, p. 97] or [Kat8, footnote 1, p. 403] by passing to spherical coordinates.) Nevertheless,
the above modified Feynman integral F ( V ) converges because Theorem 11.4.2 (or
Corollary 11.4.5) applies to this situation. (See Figure 11.4.1.)
In region (II) of Figure 11.4.1, we have J^M(V) = e~itH, where H = H0 + V, the
form sum of HQ and V (or the Friedrichs extension of HQ+V). It is natural, then, to wonder
whether the modified Feynman integral selects the "correct" self-adjoint extension of
H0 + V (restricted to C^(Rd) \ {0}). This is so, however, since H is the self-adjoint
operator associated with the semibounded energy functional q, as was explained in
Exercise 10.4.10 and is discussed at greater length in ([Nel,4], [Kat8], [Sil], [ReSil,2]).
On the other hand, when a > 2 or a = 2 and B > 1 /4 (region (III) of Figure 11.4.1), q
is no longer bounded from below and so the form sum HO + V is not well-defined; hence
neither F' (V) nor ^(V) exists. We will discuss situations of this type in Chapter
13 where TP extensions of the definitions of J^p(-) and ^/(-) (based on a method due
to Nelson [Nel] and adapted to the modified Feynman integral in [BivLa, §3]) will be
considered. [See Sections 13.5 and 13.6, respectively. Further, for the situation of the
present example corresponding to region (III) of Figure 11.4.1, we refer to Example
13.6.13 and—for the interesting special case of the attractive inverse-square potential
(or = 2 and B = 1)—to Example 13.6.18 below.]
This example extends to strongly repulsive (rather than attractive) potentials, where
we now assume that V(x) = +p/2\\x\\a (rather than -B/2\\x\\ a , as in (11.4.14)), for
x 6 R3 \ {0} and with a, B positive constants. Then, for instance, in the limiting case when
a = 2, the symmetric operator HO + V is essentially self-adjoint (on C^(R3 \ {0})—and
hence F ( V ) is presently known to exist—if and only if B > 3/4. (See, e.g., [Far2,
pp. 96-97] or [Kat8, footnote 1, p. 403]; in the terminology of differential equations,
the case B > 3/4 or B < 3/4 corresponds to the limit point or the limit circle case,
under U.S. or applicable copyright law.

respectively.) In contrast, f'M(V) exists for all values of B > 0. Actually, in view of
Remark 11.4.3(b) (or 12.1.3(c)), f'M (V), the modified Feynman integral associated with
the positive potential V (= V+), exists for all values of (a, B), with a > 0 and ft > 0.
We remark that for notational simplicity, we have only taken into account the radial
variable r := \\x\\, but ignored the angular variables in the present discussion. The minor
adjustments needed for a more detailed discussion will be briefly explained in Example
13.6.18, especially in Remark 13.6.19.
Finally, we note that Example 11.4.7 also extends to multiparticle Hamiltonians (see
Section 6.4 above and [ReSi2, Theorem X. 19, p. 170]). Further, it is of interest in quantum
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
chemistry, especially in the phenomenological study of certain large molecules. In this
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 240 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 11.4.1. The central potential V(x) = B/2\\x\\ a , x e R3 \ {0}

regard, we point out the very interesting review article [FrLdSp] by Frank, Land, and
Spector surveying (through the late 1960's), in particular, the various uses of singular
potentials as mathematical models in quantum physics (e.g. Regge poles formalism), as
well as their applications to molecular physics and high energy phenomenology.
Remark 11.4.8 (a) The free unitary group e~ itH ° is a generalized integral operator
in L2(Rd). (See, e.g., [Hor3,4, Ste2].) In fact, it acts as an ordinary integral operator,
with kernel (2nit)~ d / 2 e'|| x - y | | 2 /2t, when restricted to a (dense) subspace of functions
decreasing sufficiently fast at infinity, such as the Schwartz space S(Rd), for instance
(or the larger space L1 (Rd) D L2(Rd)). It can then be extended by continuity to all of
L 2 (R d ). (See Section 10.2, especially Theorem 10.2.7, as well as the beautiful discussion
in [Kat8, §IX.1.8, pp. 495-496].) The free imaginary resolvent [I + it Ho]-1. however,
is an ordinary integral operator in L2(Rd). Consequently, for any p 6 L 2 (R d ), the
under U.S. or applicable copyright law.

integrals in (11.4.3) are Lebesgue integrals and can be viewed either as n-fold iterated
or multiple integrals; in particular, they do not have to be interpreted in the mean or some
other special sense, as is the case in (11.2.38) with the definition of F t j p ( V ) . (See also
Remark 10.2.15(a). On the other hand, an advantage of (11.2.38) over (11.4.3) is that
the expression is given by the same familiar exponential function for all dimensions.)
(b)As was pointed out by Mark Kac when he was presented by the second author with
the above results, the definition of the modified Feynman integral (in Theorem 11.4.2 or
Corollary 11.4.5) provides a "natural summation procedure " for the Feynman integral.
(c) (The subsequent comments are purely heuristic and do not pertain to the domain
EBSCO
of rigorousPublishing : eBook Academic
mathematics.) In some Collection
sense,(EBSCOhost)
from the- point
printedof
on view
6/8/2017 5:38 integral
of the PM via kernel,
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATION TO THE MODIFIED FEYNMAN INTEGRAL 241
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

the imaginary resolvent [I + itHo]~l shares features with the unitary group e~ itH °
(see (a) above and (10.2.33)) as well as with the heat semigroup e~'H° (see (10.2.26)).
When d = 1, for instance, the kernel G(x, y, t) of the operator [I + it Ho]~' is not
only oscillatory but also decays exponentially fast for large values of \x — y|/\/7 (see
(ll.4.2a)). Moreover, intuitively, in the last equality of (11.4.3), the factor ["]"=1 (1 +
;' (t/n)V (xj))~l can be interpreted as an approximation to the "exponential functional"
exp{—i f0 V ( w ( s ) ) d s } , where w is a continuous path going from XQ to xn — v within the
time interval [0, t]. Accordingly (and still assuming that d = 1 for simplicity), one can
argue that the "important "paths in the finite-dimensional integrals in (11.4.3) are Holder
continuous of order at most 1/2, as is the case of the "typical" Brownian paths. (Compare
with Corollary 3.4.4 and Theorem 3.4.5.) Note that in order to reach a similar conclusion,
Feynman had to use (formally) an argument related to the method of stationary phase;
see [Fey2, pp. 375-376] and the comments following formula (7.3.5) in Section 7.3.
The following exercise supplements the above remarks and may help the reader—as
it did the second author—to motivate and understand intuitively the modified Feynman
integral.
Exercise 11.4.9 Go over the heuristic derivation of the Schrodinger equation (when
d — 1) given in Section 7.3 above ([Fey2, pp. 374-376]) by using, instead of the function
exp{'(*^y) }, the function exp{—(1 — i ) ^ j j ^ } [and hence, essentially, by replacing the
kernel ofe~itH° given by (10.2.33) by that of [1 + itH0]~l (when d = 1 and t > 0)
given by (11.4.2a)].
More specifically, determine in the process the analogue of the constant L in (7.3.3)
and identify it with the pre-factor in (11.4.2a).
[Advice: Recall that (—j)1/2 = (1 - OA/2 and calculate the Lebesgue integrals

This should enable you to carry out the analogue of the computation performed in Section
7.3 by means of ordinary Lebesgue integrals (and hence absolutely convergent integrals)
over R rather than oscillatory "integrals " which need not be convergent in the sense of
either Riemann or Lebesgue (see the comments following Equations (7.3.7)-(7.3.10)).]
under U.S. or applicable copyright law.

Extensions: Riemannian manifolds and magnetic vector potentials

We close this section by considering two additional examples: (i) the more geometric
setting of curved differentiable manifolds (Example 11.4.10), and (ii) the case of a quan-
tum particle moving in a magnetic as well as electric fields (Example 11.4.12). (Nothing
in the rest of this book—except, in the case of Example 11.4.12, part of Section 13.6 and
the end of Section 13.5—depends on these examples, and so one or both of them can be
skipped on a first reading. On the other hand, each of them identifies a direction in which
further work related to other topics in this book might well be successfully pursued.)
Example 11.4.10 (Modified Feynman integral on Riemannian manifolds) Let Md be
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
a complete,
UNIVERSIDAD smooth
DISTRITAL (d-dimensional) Riemannian manifold. (Recall that a Riemannian
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 242 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

manifold is complete if, for example, it is compact or is a closed submanifold of Euclidean

space Rd, such as Rd itself; see, e.g., [Chav2, §1.7, esp. Theorem 1.10, p. 26] or [Chob].)
Let A — div(grad) be the Laplace-Beltrami operator acting in H = L2(Md); here, the
inner product in H is taken with respect to the Riemannian volume of Md. Then, HO :=
— 2 A is a nonnegative self-adjoint operator on H and is (by definition) equal to the closure
of its restriction to D ( M d ) = C^(Md) (the space of smooth functions with compact
support in Md). It is to guarantee this latter (essential self-adjointness) property that the
assumption of (geodesic) completeness is required. (We recommend Strichartz's nice
article [Stril] as a simple introduction to the subject of the Laplacian on a Riemannian
manifold; other references include [OsWe], [Chavl] and [Chob-dWm, Chapter V].)
For real t = 0, [1 + itHo] - 1 is an integral operator acting on H, with kernel still
denoted by G(x, y; t). Let V be a (measurable) real-valued function on Md with positive
part V+ and negative part V- satisfying a condition analogous to that of Theorem 11.4.2.
This will be the case, for instance, if V+ e L^M4) and V- e L?(Md) + L°°(Md),
where p = 1 if d = 1, p > 1 if d = 2 and p > d/2 if d > 3, and if, in addition, the
Sobolev inequalities hold in D ( M d ) . In fact, the proof of this statement is the same as in
the Euclidean case since it relies on the Sobolev inequalities. (The Sobolev inequalities
hold for functions in V(Md) if, for example, Md has nonnegative Ricci curvature and if
the volume of the geodesic balls of radius r grows like rd at infinity. See, for example,
[Yaul, p. 10] or [Au]; also see [Heb, esp. Proposition 3.6, p. 59] for a more detailed
discussion and for further information.)
Under these assumptions, Theorems 11.3.1 can be applied and so Theorem 11.4.2,
Definition 11.4.4 (of the modified Feynman integral) and Corollary 11.4.5 extend easily
to this situation provided, of course, that the integration in (11.4.3) is now performed with
respect to the Riemannian volume of Md. (Naturally, the Hamiltonian H — HO + V is
still defined as the form sum of HO and V.)
Remark 11.4.11 (a) In general, however, G(x, y; t) does not have an explicit expres-
sion, in contrast to the Euclidean case. Hence, for example, formulas (11.4.2) and(11.4.5)
do not have an exact counterpart in this setting. (An analogous comment would be true if
we worked with the unitary group e~"H° instead of the imaginary resolvent [I +itHo\~l,
as in the Feynman integral via TPF.)
(b) In view of part (a) and Remark 11.4.8(c), it is noteworthy that precise kernel
estimates are now available in a rather broad context. In particular, if the Ricci curvature
under U.S. or applicable copyright law.

of the Riemannian manifold Md is bounded from below, then the kernel G(x, y ; t ) exhibits
an asymptotic behavior roughly similar to that of its Euclidean counterpart. (See [CheGT,
§0.27 for a precise statement of this fact.)

Example 11.4.12 (Schrodinger equation with singular magnetic vector potential) For
simplicity, we again work in Rd (with d > 1), rather than in more general curved
Riemannian manifolds as in the previous example.
Let a = (a1,a2, . . . ,ad) € (LjJoc(Rd))d be a (locally square-integrable) vector-
valued function on Rd; that is, each of its components aj (1 < j < d) is real-valued and
2
in LEBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
(Rd).
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPLICATION TO THE MODIFIED FEYNMAN INTEGRAL 243
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Set H = L2(E.d), equipped with it usual norm || • \\2 = \\ . \\L2 Rd)- Consider the
(nonnegative, closed) quadratic form

with domain

where V = (9i, . . . , 9j) = ( d / d x 1 , . . . , 3/dxd) denotes the distributional gradient on

Rd and i = V=T.
Let A = Ag = — j(V - ia)2 be the (nonnegative) self-adjoint operator associated
with/I^ (viaTheorem 10.3.15).
Now, let V be a measurable real-valued function on Rd such that V+ = max( V, 0) is
in L]1oc(Rd) and V_ = max(— V, 0) is A-form bounded with bound less than 1. Further,
let B = V be the operator of multiplication by V acting in L 2 (R d ).
Then C — A + B, the form sum of A and B, is the semibounded self-adjoint operator
associated with the (semibounded, closed) quadratic form

with domain

where we have used (11.4.15). (For more details, see ([Kat7], [ReSi2].) We write
under U.S. or applicable copyright law.

Physically, H = H^y represents the Hamiltonian or energy operator with scalar poten-
tial V and magnetic vector potential a.
In view of (11.4.16), H is a natural (bounded from below) self-adjoint realization
of the symmetric operator — 1(V — ia)2 + V. (Of course, when a — 0, we have A —
— 1 A = HO and so H = HQ + V as in the rest of this section.)
We can now apply Theorem 11.3.1 and deduce from it that the unitary group e~"H
is given by (11.3.2); which enables us to write the solutions of the Schrodinger equation
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
-jj- UNIVERSIDAD
= — iHty DISTRITAL
(t e R)FRANCISCO
in termsJOSE
of DE
expressions
CALDAS involving a and V separately.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 244 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We note that we could work here on an arbitrary (nonempty) open subset Q of Rd

instead of Rd itself. (We would then have to impose, for example, Dirichlet boundary
conditions, corresponding to a particle constrained to remain in Q.) To avoid repetitions,
however, we will postpone doing this until the latter part of Section 11.6, where the
results and definitions of this section are extended to include Schrodinger operators with
complex-valued scalar potentials.
Remark 11.4.13 (a) If we wanted to apply Corollary 11.1.6 (the Trotter productformula
for unitary groups) instead of Theorem 11.3.1, we would have to assume that — 1(V —
ia)2 + V is essentially self-adjoint on D(R d ). This is known to be true under the more
restrictive hypotheses a e (L4 oc (R d )) d and V-a :=diva = 0 (as well as V+ e L 2 Q C (R d )
and V_ is A-operator bounded with bound less than 1); see the work of Leinfelder
and Simader [LeiSd] which extends Kato's essential self-adjointness results [Kat2,3]
discussed in part in Section 11.2 (Theorem 11.2.11).
(b) Some readers may wonder where the definition (11.4.17) of the Hamiltonian with
magnetic vector potential comes from. Physically, when d = 3, a = a(x) represents the
magnetic vector potential (associated classically with the magnetic field B = V x a :=
curia) and V represents the scalar potential (associated classically with the electric field
£ = — VV := —grad V) acting upon a single quantum-mechanical particle moving in
R3. The formal expression for the quantum-mechanical Hamiltonian,

can then be easily deduced from the "correspondence principle " between classical and
quantum mechanics. (See, for example, [ReSi2, Example 5, p. 173].) Further, in physical
units such that H = 1, H can be written formally as

where m and e denote the mass and the charge of the quantum particle, respectively, and
where c is the speed of light.
(c) Example 11.4.12 above extends naturally to an N-body problem corresponding
to a system of N charged particles (of mass mk and charge ek, k = 1, . . . , N); with the
under U.S. or applicable copyright law.

obvious notation, the formal expression for the Hamiltonian is then given by

(In ordinary units, we must replace V and Vk by h V and HVk, in (11.4.18) and (11.4.19),
respectively.)
Exercise 11.4.14 Write down the quadratic form corresponding to (11.4.19) and give
(under suitable hypotheses) a precise definition of the N-body Hamiltonian with magnetic
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
vector potential
UNIVERSIDAD discussed
DISTRITAL in Remark
FRANCISCO 11.4.13(c).
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DOMINATED CONVERGENCE AND THE MODIFIED FEYNMAN INTEGRAL 245
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

11.5 Dominated convergence theorem for the modified Feynman integral

The purpose of this section is to show that the "modified Feynman integral" from [Lai ,2,
La6-ll, Lal3]—introduced and discussed in Section 11.4—presents the advantage of
enjoying very satisfactory stability properties; in fact, close to the optimal ones that could
be expected from any definition of the Feynman integral.
The results of this section are due to Lapidus in [Lal2]. Our approach is operator-
theoretic in nature. Following [La 12], we first obtain an abstract perturbation theorem for
unitary groups generated by form sums of operators (Theorem 11.5.7). We then combine
this result with certain analytic techniques in order to derive a dominated convergence
theorem for Feynman-type path integrals (Theorem 11.5.13). In applying the latter to
the modified Feynman integral, we use in an essential way the fact that the "product
formula for imaginary resolvents" from [La1 1 ] (Theorem 11.3.1, established in Section
11.3 above) holds in a rather general context. This enables us, in particular, to consider
perturbations with "arbitrary" positive part and with unbounded negative part, in the
sense of quadratic forms.
More precisely, in a special case, the main result of this section is the following
"dominated-type convergence theorem" (see Theorems 11.5.13 and 11.5.19 for a more
general statement):
Theorem 11.5.1 (Dominated convergence theorem for the modified Feynman integral:
special case) Let V,Vm,m — 1,2, . . . ,be Lebesgue measurable real-valued functions
on Rd. Let Vm,+ — max(Vm,Q) and V m ,_ — max(-V m ,0) denote respectively the
positive and negative part of Vm. Assume that " Vm converges to V dominatedly ", in the
following sense:
(a) Vm -> V Leb.-a.e. in Rd,
(b) V m ,+ < U for some U e L1loc
(c) Vm,- < W for some W e LP + L°°,
where p = 1 if d = 1 and p > d/2 if d > 2.
Let H = Ho + V and Hm = HO + Vm be the Hamiltonian [or Schrodinger operator,
acting in L2 (Rd)] associated with V and Vm, respectively, where as before, HQ = — 5 A
denotes the free Hamiltonian acting in L 2 (R d ). Then we have asm —^ oo:
under U.S. or applicable copyright law.

and

where the limit in (11.5. la) [resp., (11.5.1i>)] holds in the strong operator sense in
L(L 2 (R d )) and is uniform in t on compact subsets of R [resp., [0, +00)].
Moreover, the "modified Feynman integrals", f \ f ( V ) and F'M(Vm), associated with
V and Vm, respectively, exist (i.e. are well defined, in the sense of Definition 11.4.4) and
for all t 6 R
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 246 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

thatis, for all p e L2(Rdl

in the norm of L2(Rd).

More specifically, we have

Note that the last equality in (11.5.3) indicates that the limits in m and n, occurring
in the definition of the modified Feynman integral (Definition 11.4.4) associated with
the "potentials" Vm, can be interchanged.
In the actual statement of Theorem 11.5.13, it is assumed that W belongs to the
larger space (£loc)u, the set of all functions which are "locally strongly uniformly in
Lp" (see Definition 10.4.2(ii)). [Naturally, in our present context, the "domination"
refers to the potential and not to the corresponding "functional" in the integrand of a
(formal) Feynman path integral.] This enables us to obtain a theorem in which no a priori
assumption is made on the limiting function V, as should be the case with a "dominated
convergence theorem". [More generally, we will slightly extend the result of [Lal2] by
allowing W to be in the broader Kato class Kd for any d > 1 (see Definition 10.4.1 and
Remark 11.5.15(d)).]
The work in [La 12] was motivated in part by a private query of the first author con-
cerning his "bounded convergence theorem" for the "analytic operator-valued Feynman
integral" [Jo3, Theorem 2]. (As far as we know, the paper of Johnson [Jo3] was the first to
directly address in any detail the question of stability of Feynman integrals under small
changes in the potential.) The second author answered this query in the latter part of his
article [Lal2, Corollary 4.2] and obtained some related results [Lal2, Proposition 4.1]
pointing out the usefulness of operator-theoretic methods in this context. [We will take
advantage of this in Chapters 15 and 16. Moreover, in Section 16.1 (Theorems 16.1.1
and 16.1.2), based on later work in [JoLal], we will significantly extend the stability
under U.S. or applicable copyright law.

results of [Jo3] to a broad class of Wiener functionals.]

The rest of this section is organized as follows. We first recall some definitions and
results from operator theory that will be useful to us here. We then present an abstract
perturbation theorem for the form sum of self-adjoint operators (Theorem 11.5.7), and
combine it with Theorem 11.3.1 above to justify an interchange of limits in the cor-
responding product formulas for imaginary resolvents. Finally, we establish a general
dominated-type convergence theorem for Feynman integrals (Theorem 11.5.13) and dis-
cuss its applications to the modified Feynman integral (Theorem 11.5.19).
We note that these results (or a special case thereof) will also be applied to other
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
approaches
UNIVERSIDADto the Feynman
DISTRITAL FRANCISCO integral towards the end of Section 13.4; see especially
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DOMINATED CONVERGENCE AND THE MODIFIED FEYNMAN INTEGRAL 247
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollaries 13.4.3 and 13.4.6 (from [Jo6] and [JoKim], respectively), dealing with
dominated-type convergence theorems for the analytic in time Feynman integral and
the Feynman integral via TPF.

Preliminaries
Our setting for the abstract results is much the same as that of Section 11.3. Let H. be
a complex Hilbert space equipped with the inner product (., .) and the associated norm
|| • ||. The arrow -> (resp., ->) indicates strong (resp., weak) convergence in H. If T
is an unbounded self-adjoint operator in H, we denote as before by T+ (resp., 71) its
positive (resp., negative) part, defined through the spectral theorem (Theorem 10.1. 11),
and we let Q(T) = Q(T+) n Q(T-) be its form domain. (See Section 10.3.) If T is
nonnegative, Q(T) = D ( T 1 / 2 ) , while if T is bounded from below (say, T + al > 0,
for some a s R), then Q(T) :- Q(T + al).
We will need a counterpart for semibounded operators of the notion of form core for
semibounded quadratic forms introduced in Definition 10.3.3.
Definition 11.5.2 Let T be a self-adjoint operator on Ti. which is bounded from below
(say, T + al > 0, for some a € R). Then a subspace D of Q(T) is said to be a form
core for T if and only if it is dense in the space Q(T) equipped with the Hilbert norm

that is, D is a form core (in the sense of Definition 10.3.3) for the nonnegative quadratic
formx H-> ||(T +aI) 1/2x || 2 , with domain Q(T).
Exercise 11.5.3 Let T be as in Definition 11.5.2. Show that the linear operator (T +
aI) 1 / 2 is continuous from (Q(T), ||| . |||1) to (H, \\ • ||).
We will also need a suitable notion of convergence for unbounded self-adjoint oper-
ators. (See, for example, [ReSil, §VIII.7] for a number of additional properties.) This
notion will be used again (in a slightly different form) in Section 12.2.
Definition 11.5.4 Let B, Bm, m = 1,2, . . . , be (unbounded) self-adjoint operators in
H. We say that {B m } m = 1 converges to B in the strong resolvent sense if and only if
under U.S. or applicable copyright law.

that is, [I + i B m ] - 1 x -+ [I + iB] - 1 x, for all x e H.

In this section, unless otherwise specified, all limits involving operators will be taken
in the strong operator sense (that is, the topology of pointwise convergence on H) and
will be denoted by -> or, equivalently, by s - lim.
Finally, we reformulate and specify—in the special case of self-adjoint operators—
Theorem 9.7.11 (or 8 .1.12), a basic result in the approximation theory of semigroups of
operators. (For the present statement, see [Kat8, Theorem IX.2.16, p. 504] and [KuSe,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
Theorem 11.4,
UNIVERSIDAD p. 312FRANCISCO
DISTRITAL and Corollary 11.4.1, p. 315] or [ReSil, Theorem VIII.21, p. 287].)
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 248 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 11.5.5 Let H,Hm, m = 1,2, . . . , be self-adjoint operators in H. Then the

following statements are equivalent:
(a) {H m } m = 1 converges to H in the strong resolvent sense.
(b) e~itHm -> e~"H strongly, for all t €R.
(c) [/ + iXHm]~l -> [I + i ^ H ] - 1 strongly, for all y e R.
(d) The same as (b), uniformly in t on compact subsets of R.
If, in addition, the operators Hm and H are uniformly bounded from below, then
(a) implies:
(e) e~'Hm -> e~'H strongly, uniformly in t on compact subsets of [0, +00).
Perturbation of form sums of self-adjoint operators
We establish here a simple abstract perturbation theorem for the form sum of (suit-
able) unbounded self-adjoint operators (see Theorem 11.5.7 and Corollary 11.5.8 below,
obtained in [Lal2]), and then apply it (in conjunction with Theorem 11.3.1 above, from
[Lai 1]) to justify the interchange of limits associated with the product formulas for the
imaginary resolvents of these operators.
We now present the aforementioned perturbation theorem that will be applied later on
in this section. We do not strive here for maximal generality since we are mainly interested
in the applications to convergence theorems for the modified Feynman integral. We first
state our assumptions for this theorem.
Hypotheses. Let A, B, Bm, m = 1, 2, . . . , be self-adjoint operators on H. Assume
that A is nonnegative. Suppose for simplicity that Q(A) n Q(B+) is dense in H. Assume
further that the following conditions (CO, ( C 2 ) are satisfied:
(CO For each m > 1, Q(A) c Q(B-) n Q(Bm,-) and

for all x 6 Q(A), for some positive constants y < 1 and S independent of m and x.
We then set H = A + B and Hm = A + Bm. (Note that by Theorem 10.3.19, H and
Hm are semibounded self-adjoint operators.)
( C 2 ) There is a form core D for B+, B- and H (in the sense of Definition 11.5.2)
such that for all x e D,
under U.S. or applicable copyright law.

and

as m —> oo.
Here and hereafter, we denote by 5m,+ (resp., 5 m ,_) the positive (resp., negative)
part of Bm and by B^+ (resp., Bm'_) the square root of Bm,+ (resp., B m , _).
Remark 11.5.6 Condition (C\) implies that B- and Bm,-on 6/8/2017
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed
are A-form bounded (Def-
5:38 PM via
inition 10.3.17)
UNIVERSIDAD withFRANCISCO
DISTRITAL uniformJOSEbound y less than one. Furthermore, conditions (CO
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DOMINATED CONVERGENCE AND THE MODIFIED FEYNMAN INTEGRAL 249
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and (C2) imply that D c Q(H) n Q(Hm); in particular, Q(H) n Q(Hm) is dense
in "H. It thus follows from Theorem 10.3,19 that the form sums H = A + B and
Hm — A + Bm (m = 1 , 2 , . . . ) are well-defined (and necessarily densely defined)
semibounded self-adjoint operators on H, with uniform lower bound —8.
We can now state the following perturbation theorem of the Kato-Rellich type, which
is obtained in [Lai2, Theorem 3.1 (and Corollary 3.1), p. 43].
Theorem 11.5.7 (Perturbation theorem for form sums of self-adjoint operators) Under
the hypotheses just stated:
{/fm}^L] converges to H in the strong resolvent sense. (11.5.4)
In view of Theorem 11.5.5, the next statement follows immediately from Theorem
11.5.7. Recall from Remark 11.5.6 that the stability condition (CO implies that the
operators Hm and H are uniformly bounded from below, so that part (e) of Theorem
11.5.5 can also be used here.
Corollary 11.5.8 Under the assumptions of Theorem 11.5.7, we have

and

strongly as m -> oo.

Moreover, the convergence in (ll.S.Sa) [resp., (11.5.5b)] is uniform in t on compact
subsets o/R [resp., [0, +00)].
Proof of Theorem 11.5.7 Fix v e «. Set wm = [1 + iHmr{v. Then wm e D(Hm)
and

Clearly,

Consequently,
under U.S. or applicable copyright law.

By (11.5.6) and the Cauchy-Schwarz inequality,

Since wm e D(Hm) c Q(Hm) = Q(A) n Q(Bm,+), condition (d) implies that

Taking the imaginary part of (11.5.8) and using (11.5.9), we see that
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 250 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Since y < 1, it follows that {Al/2wm} and {B^+wm} are bounded sequences. (Note
that we have used here the fact that S is independent of m.) According to (11.5.6) and
(11.5.10), so is {Bm' _wm}. Hence there exist vectors w, a, b+ and b- inH such that

along some subsequence {rrij} —> oo. (Here and in the next few lines, it is implicitly
understood that the limits are taken along {mj}.)
We claim (much as in Step 1 of the proof of Theorem 11.3.1) that

In fact, let y e Q(A). Then, by (11.5.11),

Hence w € Q(A) and a = Al/2w. (Recall that Q(A) = D(A1/2) and A1/2 is self-
adjoint.)
Now, for all y e D,

where we have used (11.5.11) and condition (€2).

Since D is a form core for B+, it follows by continuity (see Exercise 11.5.3) that

Hence w € Q(B+) and b+ - B^w.

Similarly, using (11.5.11) and (C2), we see that b- = B 1/2 _w. Since Q(H) =
Q(A) n Q(B+), this concludes the proof of (11.5.12).
Next, let y e D. Then
under U.S. or applicable copyright law.

where we have used successively (11.5.7), and after passing to the limit along {mj},
(11.5.11), (11.5.12) and condition (C2). Here, we denote by q the sesquilinear form with
which H is associated. Hence q(w, y) = ((v — w)/i, y), for all y 6 D. Since, by (C2),
D is a form core for H, it follows from the first representation theorem for unbounded
quadratic forms (Theorem 10.3.14) combined with a simple continuity argument that
w e £>(//) and Hw = (v - w)/i. Thus
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DOMINATED CONVERGENCE AND THE MODIFIED FEYNMAN INTEGRAL 251
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and so

In view of (11.5.13b), the limit w in (11.5.11) does not depend on the subsequence
{wmj}. By a standard compactness argument, it follows that wm —>• if as m —»• oo.
Hence by (11.5.8) and (11.5.13a),

which in turn implies that || wm — w ||2 -> 0 and so wm -+ w,as required. This concludes
the proof of Theorem 11.5.7. D
It is natural to ask whether the limiting procedures occurring in the statements of
Theorem 11.3.1 and Theorem 11.5.7 can be combined or even interchanged. The fol-
lowing result answers this question in the affirmative. It will be applied to the modified
Feynman integral at the end of this section (see Theorem 11.5.19).
Corollary 11.5.9 (Interchange of limits) Assume that the hypotheses of Theorem 11.5.7
hold and {Bm}™=} converges to B in the strong resolvent sense. Then, for all t e R, we
have

Proof We shall show by repeated use of Theorem 11.3.1 (the product formula for imag-
inary resolvents) and Theorem 11.5.7 (the perturbation theorem for form sums) that both
double limits exist and are equal to the same operator, namely e~lt(-A+B\ As before, we
set H = A + B and Hm = A + Bm. Also, we fix t e R.
By Theorem 11.3.1, we have for each fixed m,
under U.S. or applicable copyright law.

Further, by Theorem 11.5.7 (or more precisely, by equation (11.5.5 a) of Corollary 11.5.8),

Hence the first double limit occurring in the statement of Corollary 11.5.9 exists and
equals e~'tfi.
On the other hand, for each fixed n, our additional assumption in Corollary 11.5.9
(about the strong resolvent convergence of {Bm}^=1 to B) implies that
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 252 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Since all the resolvents involved are contractions, it follows by an immediate induc-
tion that, for fixed n,

Now, applying Theorem 11.3.1 once more, we see that

Consequently, the second double limit in Corollary 11.5.9 exists and equals e "H. Hence
the conclusion. D
Remark 11.5.10 (a) It will be very easy In the applications to verify that {Bm}m=1
converges to B in the strong resolvent sense (see the end of the proof of Theorem 11.5.13).
(b) The analogue of Corollary 11.5.9, obtained by replacing imaginary resolvents by
unitary groups in its statement, would follow from Theorem 11.5.7 in a similar way if it
were known that the Trotter product formula for unitary groups (Corollary 11.1.6) held
under the assumptions of Theorem 11.3.1. Recall, however, that even for unitary groups
generated by nonnegative self-adjoint operators, this problem remain open. (See Remark
11.3.6(c) and Problem 11.3.9, as well as [La6] and [Lai 1].) For the same reason, it is not
known whether Theorem 11.5.13 below can be applied in its full generality to Nelson's
definition of the Feynman integral (via TPF) introduced in [Nel] and discussed in detail
in Section 11.2.

Application to a general dominated convergence theorem for Feynman integrals

It will be convenient to work with the class of functions (Lploc)u — (Lploc(Rd))u which
was introduced in Definition 10.4.2(ii). This class was originally considered by Simader
[Sd] in the context of Schrodinger operators.
Loosely speaking, the elements of (Lpc)u are the functions which are "locally
strongly uniformly in Lp". Indeed, recall from Definition 10.4.2(ii) that f E (Lploc)u
if and only if f ||x _ y||<r | f ( y ) | p d y -> 0 as r -> 0+, uniformly in x e Rd.
The following facts can easily be deduced from Definition 10.4.2(ii) (and Proposition
under U.S. or applicable copyright law.

10.4.5(b)). [Unless otherwise specified, the exponents p, q belong to [1, oc); further, we
write, for example, LP instead of LP(Rd).]

where Kd denotes: the

EBSCO Publishing
Kato class in dimension d (Definition 10.4.1). (For the containment
eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
of (11.5.14b), we assume that
UNIVERSIDAD DISTRITAL FRANCISCO p DE>CALDAS
JOSE d/2 if d > 2 and p = 1 if d = 1.) Of course,
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DOMINATED CONVERGENCE AND THE MODIFIED FEYNMAN INTEGRAL 253
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(L^n c Lploc. In particular, it follows that (as noted before)

The inequality 0 < f < g, with / Lebesgue measurable and

In the next two statements, (11.5.17) and (11.5.18), we assume that p > d/2 if d >
3, p > 1 if d = 2 and p= 1 if d= 1.

Set f+ = max(/, 0) and /_ = max(-f, 0). Let f+ 6 L^ and /_ e (ifoc)u. Then,

for any r > 0, we have (unless otherwise specified, all integrations are taken over R d ):

where C (resp., Cr) is a universal constant depending only on d (resp., on d and r)

but not on f_, and where Br(x) denotes the ball in Rd of center x and radius r. For a
proof, based on the Sobolev inequalities, see, e.g., [BreKat, Lemma 2.1 and Remark
2.1, pp. 139-140]. Here, we denote by HQ (the Sobolev space) the completion of
D = D(R d ) (= C^(Rd))—the space of infinitely differentiable functions with compact
support in Rd—under the Hilbert norm |||M|||I = (||Vu||2;2 + || M ||2 2 ) 1/2 .
Let HO = -5 A be the free Hamiltonian acting in H = L2(Rd) and let / be as in
(11.5.18). In the following, we will (as before) use the same notation for the function
/ and the corresponding multiplication operator by / in H. The form sum HQ + f is
a well-defined self-adjoint operator in "H. In fact, it suffices to choose r so small that
C supxeRd ||f _||L p (B r (x)) < 1 for inequality (10.3.62b) to hold in Definition 10.3.17
(see, e.g., [BreKat, Remark 2.1, p. 140]). Naturally, here, A = H0, B = f, B+ - f+
and B_ = f _ .
Remark 11.5.11 (a) The properties (11.5.15), (11.5.16) and (11.5.18) will enable us
under U.S. or applicable copyright law.

to state a "dominated convergence theorem" (Theorem 11.5.13) in which no a priori

assumption is made on the limiting function.
(b) In view of (11.5.17), Theorem 11.5.13 below will apply to most situations encoun-
tered in the applications. Further, (11.5.17) implies that Theorem 11.5.1 stated at the
beginning of this section follows from Theorem 11.5.13.
As a matter of fact, any space of functions enjoying properties similar to those of
Simader's space (L^c)u (especially the counterpart of (11.5.15), (11.5.16) and (11.5.18))
could be used to formulate the statement of Theorem 11.5.13 presented below. (See, in
particular, Remark
EBSCO Publishing 11.5.15(d)
: eBook below which
Academic Collection will enable
(EBSCOhost) - printed us
on to replace
6/8/2017 5:38 (L^y
PM via (with p
satisfying (11.5.19))
UNIVERSIDAD DISTRITAL by the Kato
FRANCISCO JOSE class Kd in hypothesis (V 2 ).)
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 254 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We now state our assumptions:

Hypotheses. Let V,Vm, m = 1, 2, . . . , be measurable functions on Rd. Set H =
L2(Rd), A = Ho = -1 A, B = V and Bm = Vm. Assume that

Assume further that conditions (V1) and (V2) hold:

(V1) Vm —> V Leb. almost everywhere (a.e.) in Rd.
(V2) There exist U e L^ and W e (Lfoc)u such that for all m>\,

and

Remark 11.5.12 By (11.5.15) and (11.5.16), the above assumptions imply that
V+, Vm,+ 6 L,1^ and V_, V m ,_ € (LJ^)jr; note that V+ < U and V_ < W.

In view of the previous remark and Theorem 10.4.8, we may set H = HO + V and
Hm = HO + Vm, and consider them as unbounded operators on H. We can now state the
main result of this section [Lal2, Theorem 4.1, p. 52], which we will apply afterwards
to the modified Feynman integral (see Theorem 11.5.19).
Theorem 11.5.13 (Dominated-type convergence theorem for Feynman integrals) With
the above notation and assumptions, the conclusion of Theorem 11.5.7, as well as of
Corollaries 11.5.8 and 11.5.9, holds. In particular, for every fixed p e L 2 (R d ), e ~ i t H p
converges to e~itH p in L2(Rd), uniformly in t on all bounded subsets of R.
Proof of Theorem 11.5.13 We must check that Theorem 11.5.7 applies to this situation.
For all u € H1 = Q(A) and r > 0, we have:
under U.S. or applicable copyright law.

where we have applied inequality (11.5.18) to W 6 (Lpk,c)u By choosing r so small

that y = C(sup xeRd ||W||Lp(B r (x))) < 1, we see that condition (C1) holds with S =
CCr sup xeRd ||W||LP(Br(x)). Here and in the rest of this section, ||W||tp (instead of
|| W||p) often denotes the L^-norm of W on Rd; further, ||W||z,p(Br^)) denotes the Lp-
norm of W on Br (x), the open Euclidean ball of radius r and center x in Rd.
Next, we check that condition (€2) is satisfied with D := D = D(R d ). Since, for
u e D, |V1/2+u|< U1/2\u\, withW 1/2 |u| e L2 and V1/2+u -> V1/2+u a.e., it follows
from the Lebesgue dominated convergence theorem that
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:38 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DOMINATED CONVERGENCE AND THE MODIFIED FEYNMAN INTEGRAL 255
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Similarly, \V^u\ < \V^2\u\ with W'^lwl e L2 [in fact, W 6 L^ by (11.5.15)] and
Vm _« —> V_ M a.e. Hence, again by dominated convergence,

Moreover, D = Z> is a form core for H by a theorem of Simon (see [Si8] and [Sil 1,
Theorems A.2.8 or B. 1.5, p. 460 or 463]) which extends to form cores a theorem of Kato
on operator cores (see [Kat2] and [Sil 1, Theorem A.2.9, p. 460]). [This is the only place
where we have to require that p > d/2 for d > 3, Note that, in view of the remark
at the top of page 464 in [Sil 1], Simon's theorem applies here since (L^z C Kj by
(11.5.14b). (Further, since the latter theorem is valid if V+ € L1loc and V_ e Kj, the
generalization of Theorem 11.5.13 provided in Remark 11.5.15(d) below will not pose
any problem here.)]
Finally, it remains to check that D = D is a form core for B+ = V+ and B- = V_;
since V+ and V- belong to L^, this is an immediate consequence of the following
simple lemma:
Lemma 11.5.14 Let f € L^ with f > 0. Then V is a form core (in the sense of
Definition 11.5.2) for the operator of multiplication by f in L2.
Proof of Lemma 11.5.14 Recall that u e Q(f) if and only if u e L2 and f 1 / 2 u e L2.
Observe that a subspace E of Q(f) is a form core for / if for all u 6 L2, there
exists a sequence [un] in E such that un -> u in L2 and f 1 / 2 u n -> f l/2 u in L2 (see
Definition 11.5.2). The result is now obtained by using a series of standard approximation
procedures analogous to those involved in the proof of the Feynman-Kac formula in the
general case (Theorem 12.1.1) to be discussed in Chapter 12 (Section 12.2). (Compare
[Sill, Proof of Theorem A.2.8 (Theorem B.I.5), pp. 463-464].)
Step 1: E\ := {u e Q(f) : u has a compact support) is a form core for /.
In fact, let 0 be a continuous function with compact support in Rd and such that
0(0) — 1. For M € Q(f), define un by un(x) = 0(x/n)u(x). Clearly, un has compact
support and |un| < ||0||ool"|. Hence |/ 1/2 w n ] < ||0||ool/ 1/2 «l- In particular, «„ e £1.
Because wn -> M a.e., we also have that f 1/2 u n -> f1/2 u a.e. Since both M and f 1/2 u
belong to L2, it follows from the Lebesgue dominated convergence theorem that un -> u
and / 1/2 M n -» /'/ 2 M in L2.
Step 2: EI := E\ n L°° is a form core for /.
under U.S. or applicable copyright law.

Let M e E1. By separating the real and imaginary parts of M, we may assume that u
is real-valued. Define un by

Clearly, «„ is bounded and has compact support. Since \un\ < \u\ and un -> u a.e., it
follows as above that un -> u and f 1/2 w -> f 1/2u in L2. Since {un}lies in E2, the
EBSCO Publishing : eBook Collection (EBSCOhost) n- printed on 6/8/2017 5:41 PM via UNIVERSIDAD
claim resultsFRANCISCO
DISTRITAL from Step 1. CALDAS
JOSE DE
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 256 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Step 3:Disa form core for /.

In fact, let u € £7- Let {j») be an approximate identity on Rd (i.e. jn e £>, jn > 0,
Supp jn C Bn-\ (0) and / _/„ = 1, where Supp jn denotes the support of jn and Bn-\ (0)
is the closed ball of center 0 and radius n~ ] ). Let un = jn * u be the convolution of
jn with u. It is well known that since u € L2, un € V and un -H> u in L2. (See, e.g.,
[Stel, Theorem 2 (the p = 2 case), pp. 62-64].) By possibly considering a subsequence,
we may assume that un -> u a,e. and hence that fl/2un -»• /^ 2 M a.e. Moreover,
Supp «„ c Supp u + Supp ;'„ c K, where A' := Suppw + B\(0); note that K is
compact because w has compact support and 61 (0) is closed. Since u e L°°, we see that

Thus \un\ < \\u\\ L°°XKi where XK denotes the characteristic function of AT. This implies
that

Note-that f l / 2 X K e L2 since / e L,1^. By (11.5.22), it follows from the dominated

convergence theorem that f^2un —> fl/2u. In view of Step 2, this concludes the proof
of Lemma 11.5.14. n
We have just shown that the assumptions of Theorem 11.5.7 are satisfied. To
conclude, we note that the hypotheses of Corollary 11.5.9 hold since condition (Vi)
alone implies that {Vm}^=1 converges to V in the strong resolvent sense. Indeed,
[/ + j Vm]~l = (1 + iVm)~l (the multiplication operator), and similarly for V. Fix
u e L2; then |(1 + iVm)~lu\ < l«l and (1 + iVm)~lu -» (1 + iV)~lu a.e. in Rd.
Hence (I + iVm)~lu ->• (1 + iV)~lu in L2, by a final application of the dominated
convergence theorem. The proof of Theorem 11.5.13 is now complete. n
Remark 11.5.15 (a) It follows from the proof of Theorem 11.5.13 that, in condition
(Vi), the assumption Vm,+ < U for some U € L\oc can be replaced by the following
weaker hypothesis:
under U.S. or applicable copyright law.

Indeed, for v 6 V, /(V m ,+ - V+)\v\2 -+ 0 since \v\2 = vv € T>.

(b) The fact that, for nonnegative potentials Vm satisfying (11.5.23), Hm converges
to H in the strong resolvent sense, also follows from the works of Simon [Si8, Theorem
4.1, p. 44] and Kato [Kat6].
We note, without going into the details, that Theorem 11.5.7 can be generalized in
order for Theorem 11.5.13 to encompass the case of Schrodinger operators with magnetic
vector potentials studied in Example 11.4.12. More precisely, let the "scalar potentials"
V, Vm be as in Theorem 11.5.13 and let the "vector potentials" a, am be W1-valued
functions on Rd such that am -> a in L2^. Then the counterpart of Theorem 11.5.13
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
holds with H
DISTRITAL := -|(V
FRANCISCO ia)2 + V and Hm := -i(V - iam)2 + Vm.
- CALDAS
JOSE DE
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DOMINATED CONVERGENCE AND THE MODIFIED FEYNMAN INTEGRAL 257
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(c) It is noteworthy that a "monotone convergence theorem" for quadratic forms is

also available in this context. (See [Kat8, § V//7.3 and VIII.4, pp. 459-^62] and [ReSil,
Supplement to VIII. 7, pp. 372-377], as well as Theorems 12.2.2 and 12.2.4 below.) This
fact will be useful in the proof of the Feynman—Kac formula given in Chapter 12.
(d) Close inspection of the beginning of the proof of Theorem 11.5.13 shows that
in condition (V2) we could replace the hypothesis that W e (L^ l o c ) by the weaker
assumption that W 6 Kd, where Kd denotes the Kato class in dimension d, with d > 1;
see Definition 10.4.1. (This fact was not observed in [Lal2].) Indeed, it follows from
Definition 10.4.1 and (the proof of) Proposition 10.4.4 that if for all m,Q< Vm_ < W,
with W e Kd, then Vm_ e Kd and V m ,_ is H0-form bounded with uniform bound less
than one (i.e. with uniform constants y < 1 and S > 0 in (CO). Hence condition (C\)
of Theorem 11.5.7 still holds in this situation. (See the proof of Proposition 10.4.4 given
in [AiSi] and sketched in [Sill, pp. 458—459]; alternatively, see the explicit estimates
obtained in [Gull,2].)
We now give examples of situations where Theorem 11.5.13 applies.
Example 11.5.16 Let V be such that V+ e Llloc and V_ e (Lfoc)u (in view of (11.5.17),
this is the case, for instance, if V_ e Lp + L°°), where p is given by (11.5.19). (More
generally, by Remark 11.5.15(d) above, we could assume instead that V_ belongs to
Kd.) Consider Vm obtained by truncation as follows:

Then Theorem 11.5.13 clearly applies to this setting.

Example 11.5.17 Let d > 2. Consider the function p defined on (0, +00) by p(r) =
P/2ra, where a, B are positive constants and a < 2. Define V, Vm, for m = 1,2, . . . ,
as follows. Assume that V+ e L^ and V-(x) < P(||x||), where |x| denotes the length
of x e Rd. Set
under U.S. or applicable copyright law.

Assume further that Vm+ —> V+ in L^oc. Let Vm := V m, + — V m ,-. By using the
monotonicity of p, it is then easy to check that Theorem 11.5.13 applies (see Remark
11.5.15(a)); note that, for W(x) = p (||x||), we have W 6 L^ + L00 with p e (d/2,d/d)
satisfying (11.5.19). We refer to Example 11.4.7 (as well as [ReSi2, pp. 161, 169-170]
and [Far2, Example, pp. 96-97]) for a motivation of the present example. Note that
one could not (at least in a straightforward way) make use of a monotone convergence
theorem here.
Remark 11.5.18 (a) By possibly passing to a subsequence, it is easy, in principle, to ver-
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
ify condition (V2) in Theorem 11.5.13. Indeed, recall thefollowing fact: For p e [1, +00),
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 258 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

let {gm} be a sequence in Lp and g e Lp such that \\gm — g\\ip -> 0. Then there exists
a subsequence {gmj} and a function h e Lp such that gmj —> g a.e. and \gmj \ < h for
all j (see, e.g., [Bre2, Theorem IV.4, p. 58]). If, in addition, ||gm — g|| Lp tends to zero
fast enough—in the sense that Em=1 ||8m — g\\L? < oo—then the subsequence {gmj}
can be replaced by the entire sequence {gm} (see, e.g., [Des, Proposition 2.2, p. 94]).
(b) An extension of Theorem 11.5.13 [Lal2, Theorem 4.1] (or rather, of the semi-
group convergence (ll.S.lb) obtained in Theorem 11.5.1) was obtained by Voigt [Vo2]
in the more general context of "absorption semigroups"; that is, of Schrodinger-type
semigroups (generated by the infinitesimal generator of a positive (Co) semigroup [Na]
perturbed by a suitable potential function) acting in L p (u), where 1 < p < oo and
H is a measure on some measurable space (X, A). (See Theorem 3.5 and Corollary
3.6, pp. 126-127, as well as Example 4.3 in [Vo2].) The counterpart in [Vo2, Corollary
3.6] of condition (V\) and (V%) above is still assumed; in particular, in the analogue
of (V2), different domination hypotheses are made on the positive and negative parts
of the approximating potentials Vm (namely, "admissibility" and "regularity", in the
sense of [Vo2]). The work of Voigt in [Vo2] was motivated by his earlier work [Vol] on
"absorption semigroups", as well as by that of Lapidus in [Lal2] on "dominated-type
convergence theorems ".
We now conclude this section by applying the previous results to the modified
Feynman integral f M (.). We briefly recall from Section 11.4 the definition of the latter.
(We use more concise notation than in Definition 11.4.4 and the corresponding Theorem
11.4.2 or Corollary 11.4.5.) Let V, Vm, m = 1, 2, . . . , be as in Theorem 11.5.13 and let
(p € L2 = L2(Rd). Then, for every t e R and a.e. v =: xn 6 Rd,

where the limits as n -> oo hold in L2. Here, Gn (XQ, . . . ,xn,t; V) denotes the integrand
of (11.4.3); namely,
under U.S. or applicable copyright law.

with Gn given by (11.4.4).

In this manner, the solution at time t, e ~ i t H p , of the Schrodinger equation

with initial condition fy(., 0) = (p € D(H) c L2, is represented by a sequential limit

of (Lebesgue-convergent) finite-dimensional integrals. Naturally, f'M(Vm) is defined in
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
the DISTRITAL
same way, except
FRANCISCO forDE VCALDAS
JOSE replaced by Vm.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE MODIFIED FEYNMAN INTEGRAL FOR COMPLEX POTENTIALS 259
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We can now combine Theorem 11.5.13 and Corollary 11.5.9, as well as Theorem
11.4.2 (or Corollary 11.4.5), in order to obtain the desired stability result, and thereby
complete the proof of Theorem 11.5.1 as well as of the more general theorem below.
Theorem 11.5.19 (Dominated convergence theorem far the modified Feynman integral)
If "Vm converges to V dominatedly"(i.e. in the sense of Theorem 11.5.13 or, more gen-
erally, of Remarks 11.5.15(a)) and 11.5,15(d)), then the "modified Feynman integrals"
FM(V), F'M(Vm) associated with V, Vm are well-defined and for all p € L2 and all
t e R:

where the above limits hold in the norm of L2 = Z,2 (Rd) and are uniform in t on compact
subsets 0f R. Here, Gn is given as in (11.5.25).
Exercise 11.5.20 Show that if pm —> p in L2, then

and so formula (11.5.26) holds, except for p replaced by pm on the right-hand side of
the five equalities.
Remark 11.5.21 We close this section by recalling from Remark 11.5.10(b) that we
cannot deduce from Theorem 11.5.13 an entirely analogous dominated convergence
theorem for Nelson's definition of the Feynman integral f T P ( . ) (Section 11.2, espe-
cially Definition 11.2.21 and Corollary 11.2.22) since we do not have at our disposal
a sufficiently general existence theorem for this definition. (See Remark 11.5.10(b) and
under U.S. or applicable copyright law.

Problem 11.3.9 for a more thorough discussion of this point.) However, we note that the
present dominated-type convergence theorem (Theorem 11.5.1 or 11.5.19 above, from
[Lal 2])—along with the Trotter product formula for unitary groups (Corollary 11.1.6)—
will be used in Section 13.4 to deduce a weaker result for ft-p(.); see Corollary 13.4.6
(from [JoKim]),

11.6 The modified Feynman integral for complex potentials

In this section, we present results of Bivar-Weinholtz and Lapidus [BivLa] extending
the EBSCO
earlier results of Lapidus [Lall] discussed in Sections 11.3 and 11.4 above. More
Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
precisely,
DISTRITALwe obtainJOSE
FRANCISCO an DE
abstract
CALDAS formula for the imaginary resolvents of normal (rather
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 260 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

than self-adjoint) operators. We then deduce from it the convergence of the modified
Feynman integral with a highly singular complex (rather than real) potential V. This is
of current interest in the study of the Schrodinger equation associated with dissipative
(or "open") quantum systems [Ex]. Specifically, the assumptions on the real part of the
potential, Re V, are exactly the same as in Section 11.4. Furthermore, we assume for
the imaginary part that Im V e L11oc(Rd) and Im V < 0 (a mathematical condition
akin to "dissipativity of the energy")- Quantum-mechanically, this will imply for the
nonisolated system a "loss ofunitarity" and hence a "loss of conservation of probability"
(compare with the discussion in Section 6.1). (As in Section 11.4, singular magnetic
vector potentials are also allowed.)

Product formula for imaginary resolvents of normal operators

Let T be a (not necessarily bounded) normal operator in a (complex) Hilbert space H:
that is, T is a densely defined closed operator on U such that T*T - TT*, where T*
denotes the (Hilbert) adjoint of T, as given by Definition 9.6.1. (See [Rul, p. 348].) Then
we can write

with T1, T2 self-adjoint operators on H. Here, T1, TI are respectively the real and minus
the imaginary parts of T; they are defined by means of the operational calculus given by
the spectral theorem for normal operators, with the functions Re z and Im z defined on
the spectrum a(T) c C. (We note that so far, the spectral theorem and the associated
functional calculus have been discussed only for (unbounded) self-adjoint operators; see
Section 10.1. However, they can be extended from self-adjoint to normal operators; see,
e.g., pages 348-355 in Rudin's book [Rul]. We will use this fact throughout this section.)
Since D(T) = D(T*), it easily follows from Definition 9.6.1 that

If, in addition, we suppose the operators T1, T2 to be nonnegative, then —iT is seen to be
m-dissipative (or equivalently, by Remark 9.4.6(b), iT is m-accretive) and, hence, by
the Lumer-Phillips theorem (Theorem 9.4.7), it generates a (Co) contraction semigroup
{e~l'T}t>o on H. In view of [Paz, Corollary 4.4, p. 15], this follows from the obvious fact
under U.S. or applicable copyright law.

that both — iT and — (iT)* are dissipative (see Remark 9.4.6(a)). (We refer to Section
9.4 for the basic facts concerning dissipative operators.) We shall denote by

the form domain of T; of course, | T |1/2 is also defined by means of the spectral theorem.
If now A = A] — i A2, B = B\ — iB2 are normal (possibly unbounded) operators
on H with Aj, Bj nonnegative self-adjoint (j = I, 2) and

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE MODIFIED FEYNMAN INTEGRAL FOR COMPLEX POTENTIALS 261
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

we can define the form sum A + B of A and B as the operator in H associated with the
sesquilinear form

for u, v e Q(s) := Q. Naturally, the quadratic form associated with s is no longer

semibounded (since, for instance, s(u) = s(u, u) may not be real); nevertheless, (part
of) the first representation theorem (Theorem 10.3.14) can be adapted to the present
situation, as we now briefly explain. By definition, u e D(A + B) if and only if the map
v -> s(u, v) is continuous on Q c H with respect to the norm topology on H. Then,
(A + B)u is the unique vector v in H (given by the Riesz representation theorem) such
that

By (11.6.1) and (11.6.2), — i(A + B) is obviously dissipative; further, one can easily
show by using the Lax-Milgram lemma (e.g., [Bre2, Corollary V.8, p. 84]) that it is in
fact m-dissipative. We denote by {e~"^A+B)}t>o the (Co) semigroup which it generates.
The same conclusions still hold if one replaces the hypothesis B\ nonnegative by
a weaker assumption on its negative part B_. (Henceforth, we write B1 = B+ — B-,
where B+, B- are respectively the positive and negative parts of the self-adjoint operator
B1, given by the spectral theorem.) Namely, we shall assume, by analogy with Theorem
11.3.1, that B- is relatively form bounded with respect to the nonnegative operator A1
with relative bound less than 1; i.e. the counterpart of (11.3.1) holds, with A replaced
by A1.
In this case, in the definition (11.6.1) of the sesquilinear form 5, one has of course to
replace the term ( B 1 / 2 u , B1 1/2 v) by (B+1/2u, 5+1/2u) - (Bl_/2u, Bl_/2v).
We can now state the following abstract result, which extends Theorem 11.3.1 from
self-adjoint to normal operators. (In summary, in this extended product formula, we
assume suitable sign conditions on the real and imaginary parts of the normal operators,
as well as relative form-boundedness of the negative real part of one of the operators
with respect to the real part of the other.)
under U.S. or applicable copyright law.

Theorem 11.6.1 (Productformula for imaginary resolvents of normal operators) Under

the above hypotheses,

for all u € H, uniformly in t on bounded subsets 0f [0, +00).

Proof The proof parallels that of Theorem 11.3.1 and we shall only indicate the changes
to be made in the argument. Fix A > 0, v e W; for t > 0, set
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 262 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and

We will show that wt -> [A. 4- iC]~lv as f -> 0+, where C := A + B. (This
will establish the counterpart of Step 1 in the proof of Theorem 11.3.1.) We have wt &
D(A) C Q(A) C Q(B-), and, by definition,

By using the spectral theorem for normal operators [Rul, pp. 348-355], we obtain

where /?g,, IB>, IB+ and IB- are nonnegative bounded self-adjoint operators given by

and

(Note that B\ and 82 commute since they are functions of the same operator B.)
Taking the inner product of (11.6.3) against wt, we then easily check that

hence, also ||A2/2iu»||, ||/?^2if/|| and \\Ilg,2w,\\ are bounded independently of t. This,
along with the observation (already made in equation (11.3.15), but now a consequence
of the spectral theorem for normal rather than self-adjoint operators) that

allows us to apply the counterpart of (11.3.1) (with A replaced by AI) to deduce-

much as was done following (11.3.15)—that ||7B4.w,|| and ||A,' w<|| are also bounded
r
independently of t. We can then pass to the limit, weakly in H, along a sequence tn -> 0;
now, wt tends to a certain w, and to identify the limits of the other bounded sequences,
under U.S. or applicable copyright law.

we can use the spectral and Lebesgue theorems We first obtain, in particular, that w e
Q(A) n Q(B) = Q. Note also that by the m-dissipativity of — i B , we have

so that, for z = x + iy e a(B), we obtain

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE MODIFIED FEYNMAN INTEGRAL FOR COMPLEX POTENTIALS 263
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Hence, in fact, for every fixed e € Q(B),

from which we easily deduce now that, along [tn],

Next, let y e Q = Q(A) D Q(B); the inner product of (11.6.3) with y gives, upon
passing to the limit in the second member,

So that, by definition of C = A + B, w e D(C) and

It follows that Wt tends weakly to w = (A + i C ) - 1 v . Norm convergence can be deduced

much as in the lines following (11.3.19). This completes the proof of the analogue of
Step 1 in the proof of Theorem 11.3.1.
The rest of the proof can be carried out just as in Step 2 of the proof of Theorem
11.3.1, and so we omit it. D
Remark 11.6.2 (a) A corollary analogous to Corollary 11.3.7—in which the form
domain Q = Q(C) is not necessarily densely defined in H—is also available. Then
the limit in Theorem 11.6.1 only holds for all u e H := Q.
(b) In contrast to what happens in Theorem 11.3.1, the limiting (one-parameter)
semigroup in Theorem 11.6.1, [e~"c — e~"^A+B^}t>o, is not a group in general and the
operators e~ltc are (usually) not unitary for t > 0.
(c) Even in the case of nonegative self-adjoint operators [Lai,2, La6], let alone
in the case studied in [Lall] (Section 11.3 above) or the present situation of normal
under U.S. or applicable copyright law.

operators, it is not known whether one can replace the "imaginary resolvents " by the
corresponding semigroups in the statement of Theorem 11.6.1; see Problem 11.3.9.

Application to dissipative quantum systems

We specialize now to ~H = L2(£2), where £2 is a nonempty open subset of R^with
d > 1. Let A = HO = — ^ A be the free Hamiltonian acting in L 2 (£2), where A
denotes the Dirichlet Laplacian on S2. (When £2 — Rd, HQ is the usual free Hamiltonian
on K d .) Further, let B be the multiplication operator by a complex-valued function
V EBSCO
= q+Publishing
— <7_ — : iq',
eBookwhere q+(EBSCOhost)
Collection ,q_, q' are real nonnegative
- printed functions
on 6/8/2017 5:41 in £^(£2) and
PM via UNIVERSIDAD
q- DISTRITAL
is relatively Ho-form
FRANCISCO JOSE DE bounded
CALDAS with bound less than 1. In this case, the form sum
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 264 PRODUCT FORMULAS AND PEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A + B extends, hence coincides with, the natural realization H of — ^ A + V(x), with

domain

(Here, as before, HQ(&) denotes the Sobolev space defined as the closure of £>(£2)
with respect to the Hilbert norm ||«||#i(Q) := (\\u\\2 + IIVM\\ 2 1 ) l / 2 \ see, e.g., [Kes].)
In fact, H -\r X is known to be m-accretive (or equivalently, —(H + A.) is known to be
m-dissipative) for some A. > 0 (see, e.g., [BreKat]). In the case when fi = Rd, with
some supplementary assumptions on q-, one can replace "M e H^(QY'by"u e L 2 (£2)"
in the definition of the domain (see [BreKat]). (The hypothesis q' > 0 is not required
for the results of Brezis and Kato [BreKat] to hold and a somewhat stronger hypothesis
is assumed on q-, but then it is the closure of the operator H + A. that is known to
be m-accretive. In the present case when q' has a sign, however, even with the stated
hypothesis on q-, the methods of [BreKat] easily adapt to show that H + A. itself is
m-accretive with D(H) C D(A + B).)
We can then apply Theorem 11.6.1 to A and B, which gives us a representation of
the solutions of the Schrodinger equation

corresponding to the "energy operator" H (also called the "Schrodinger operator" in

[BreKat] or the "pseudo-Hamiltonian" in [Ex]). [We refer to Exner's book [Ex] for
a description of the properties and physical relevance of such operators to the study
of "open" (that is, nonisolated or dissipative) quantum systems.] When £2 = E.d, this
gives us an explicit representation of e~ltH as a limit of iterated integral operators that
generalizes the modified Feynman integral from [Lai,2, La6-13] (introduced for real
potentials in Section 11.4 above) to the case of a complex potential with a highly singular
(merely L,1^) negative imaginary part.
We summarize the above discussion (when Q, = R rf ) in the following theorem (from
[BivLa]).
under U.S. or applicable copyright law.

Theorem 11.6.3 (Modified Feynman integral with highly singular complex potential)
Let V : Rd —> C be such that V = q+ — q- — iq', where q+, q- and q' are nonnegative
functions in L^. (Rd). Assume further that q- is Ho-form bounded with bound less than 1.
Then the analogue of Theorem 11.4.2 holds (for all t > 0 instead of for all t 6 K).
Hence the "modified Feynman integral" J-'M(V) (given as in Definition 11.4.4 except
for V complex-valued and t > 0) exists for all t > 0, where the limit in (11.4.12') is, for
all <p e L2(M.d), uniform in t on bounded subsets of[0, +00). Moreover, for all t > 0,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
where H is FRANCISCO
DISTRITAL the "Schrodinger operator" (or "pseudo-Hamiltonian") associated with the
JOSE DE CALDAS
complex potential V, as definedLapidus,
AN: 98476 ; Johnson, Gerald W., Michel
above. L..; for
Finally, The Feynman Integral
<p e D(H) and Feynman's
andt > 0, J-'M(V)<p is the
Operational Calculus
Account: ns000601
 THE MODIFIED FEYNMAN INTEGRAL FOR COMPLEX POTENTIALS 265
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

unique solution (in the semigroup sense) at time t of the Schrodinger equation (11.6.6)
with initial state ^r(0, •) = <f>.
Remark 11.6.4 (a) Recall from Remark 11.6.2(b) that {e~<tfi}t>Q is (in general) no
longer a group but a semigroup and does not consist of unitary operators. Mathemati-
cally, this "loss of unitarity" is a reflection of the time-irreversibility of the "Schrodinger
equation " (11.6.6) associated with a (nonreal) complex potential V; further, physically, it
corresponds to a "loss of conservation of (quantum) probability " and the fact that when
it exists, the classical counterpart of an open (i.e., nonisolated) quantum system [Ex]
has nonconstant total energy. (Compare with the discussion in Chapter 6, especially
in Section 6.1, which dealt with the standard case of a quantum-mechanical particle
moving under the influence of a real potential. Then, of course, the quantum probability
was conserved due to the unitarity of the time evolution associated with the Schrodinger
equation.)
(b) Of course, in ordinary quantum mechanics, the observables are represented by
self-adjoint (rather than normal) operators. Nevertheless, complex potentials provide
useful approximate models for dissipative quantum systems. We refer to [Ex] and the
relevant references therein for a more thorough discussion of this point and of related
physical issues.
We close this section by somewhat more technical considerations dealing with gen-
eralized Schrodinger operators with magnetic vector potential a and complex scalar
potential V. (We refer to Example 11.4.12 for the simpler case where 2{£/t} is the
identity matrix and V is real-valued on £2 = Rd.) Let us again assume that £2 is an arbi-
trary (nonempty) open subset of Rd. Then, in the above discussion, we can also replace
HO = — | A by a more general (second order) elliptic operator L with a vector potential;
formally,

where [bjk}dsk=\ is a symmetric, uniformly elliptic matrix of real-valued bounded

measurable functions on £l and a = (a\, . . . , a^) is a d-dimensional vector of L^oc
real-valued functions on Q (see Example 11.4.12). More precisely, L is the nonnegative
self-adjoint operator associated with the closure of the minimal energy form correspond-
under U.S. or applicable copyright law.

ing to (11.6.8) (i.e. the closure h of the form defined on T>(&) = Cgg(ft)). The form
domain of L is

and L is the maximal restriction, as an operator on H, := L2(£i), of the operator

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
WeDISTRITAL
refer toFRANCISCO
the work JOSEofDEBivar-Weinholtz
CALDAS [Bivl-3] for the definition of such operators
AN: 98476 ; Johnson, Gerald W., Lapidus,
and for the properties that enable us to apply Michel L..;Theorem
The Feynman11.6.1
Integralinand Feynman's
this case.
Operational Calculus
Account: ns000601
 266 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Recall from Example 11.4.12 that when £2 = E.d and {bjk} = ^ I, then L = - ~ (V -
ia) 2 ; further, if, in addition, V is real-valued, then H = — j(V — ia)2+V, the form sum
of L and V, as in Example 11.4.12.
Remark 11.6.5 It is natural to wonder whether the stability results from [Lal2] dis-
cussed in the previous section can be extended to complex potentials. (See the query in
[BivLa, Remark 2, p. 455].) Recently, such a "dominated-type convergence theorem"
extending part of [Lai2] and [Vo2] (along the lines of Remark 11.5.18(b)) has been
obtained in [LisM]. Combined with Theorem 11.6.1 (or Theorem 11.6.3) from [BivLa],
and (an easy generalization of the proof of) Corollary 11.5.9, it would provide an ana-
logue of Theorems 11.5.13 and 11.5.19 above (from [Lal2]) to the present context of
("absorption ") semigroups generated by Schrodinger operators with singular C-valued
scalar potentials (and magnetic vector potentials).

11.7 Appendix: Extended Vitali's theorem with application to unitary groups

Extension of Vitali's theorem for sequences of analytic functions
The following theorem [La6, Theorem, p. 1739] extends Vitali's classical theorem
[HilPh, Theorem 3.14.1, p. 104] and is abstracted from ideas of Babbitt [Babl-3] fur-
ther developed by Feldman [Pel, p. 261] in the context of the analytic (in time) Feynman
integral (Sections 13.1 and 13.2), and later used by Chernoff [Cher2, Remark, pp. 90-
91] and then the second author [La6, §3] in the context of product formulas for unitary
groups of operators. (See the latter part of this appendix.) We will also use Theorem
11.7.1 in Chapter 16 (Proposition 16.2.4) when discussing a result from [JoLal]. The
classical theorem of Vitali (part (i) of Theorem 11.7.1) will be used in several places
in this book, especially in Chapter 13 and Chapters 16-17. [In all these cases, we will
use the more general situation where the functions /„ are operator-valued (rather than
C-valued) analytic functions, as discussed in Remark 11.7.2(c).]
We will follow here the exposition given in the appendix, pages 1738-1741, of [La6].
Theorem 11.7.1 (i) (Vitali's classical theorem) Let D be a domain ofC, with closure
D and boundary F. Let {/n}^l0 ^e a sequence of analytic functions on D, continuous
on D, and uniformly bounded on D (say, \fn(z)\ < M for all n > 0 and all z in D).
Assume that limn_,.oo fn (z) exists for all z in a subset of D having a limit point in D.
Then lim n_.oo fn (z) exists for all z in D, uniformly on compact subsets of D. Further,
under U.S. or applicable copyright law.

the limit, denoted by f ( z ) , is analytic in D.

(ii) (Extended Vitali theorem) Suppose, for example, that D = C+ = {z € C :
Rez > 0), the open right half plane, and so D = C+ = {z e C : Rez > 0} and
F = {z 6 C : Rez = 0), the imaginary axis equipped with its natural Lebesgue
measure. Assume, in addition to the hypotheses of part (i), that the limit f extends
continuously to D. Then

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
all p e FRANCISCO
for DISTRITAL L1 (F). JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: EXTENDED VITALFS THEOREM AND UNITARY GROUPS 267
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof (i) This is just Vitali's theorem [HilPh, Theorem 3.14.1, p. 104], the proof of
which can be found in [HilPh]. (Actually, in part (i), the functions fn need not be defined
on the boundary F; see Remark 11.7.2(b) below.)
(ii) It now remains to establish the extended Vitali theorem. For notational simplicity,
we may assume (by means of a conformal equivalence) that D is the open unit disk, and
hence F is the unit circle equipped with its natural (Haar) measure.
The proof may be summarized as follows. Formula (11.7.1) holds for ip = Pr, the
Poisson kernel of parameter r (with 0 < r < 1). It is then shown to hold for all <p e L'(F)
by means of a density argument.
More precisely, consider the Poisson kernel of parameter r,

Given tp e L 1 ( F ) , we set

It will be convenient to use the notation p(9) or p(e i 0 ) interchangeably.

The following properties will be useful:
(a)||Pr||Li = l.
(b)||P r *<0-«0|| L i ( r ) ->0,asr-> 1, p e L 1 (T).
(c) f(Pr * p)(eW(9)d9 = fp(0)(P r * rlf)(6)de, <p e L'(r), * 6 L°°(F).
In the proof of (c), the point is that Pr is even:

If g is a continuous function on D that is analytic in D, then it is the Poisson integral

of its restriction to the boundary F (see, e.g., [Ru2, Chapter 1]):
under U.S. or applicable copyright law.

Thus we can write for any <p e L1 (F):

where
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 268 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and

where we have used property (c) in the last equality.

Clearly,

Hence, in view of property (b), we can choose r close enough to 1 so that An<r is small
(independently of n). Then, for such fixed r, Bnj tends to zero as n —>• oo. [This follows
from part (i) of the theorem (Vitali's classical theorem) and from the Lebesgue dominated
convergence theorem.] In view of (11.7.2), we thus deduce that (11.7.1) holds for every
(p in Ll(T), as desired. D
Remark 11.7.2 (a) In general, in part (ii), nothing more precise can be said about the
convergence on the boundary. Take, for example, fn(z) = zn and f ( z ) — 0,for \z\ < 1.
Then the hypotheses of Theorem 11.7.1 are satisfied in this case (with D the open unit
disk). However, {fn(z}} never converges to f ( z ) , f o r z\ — 1. Moreover, [fn(z)} diverges
for "almost every" value ofz in F.
(b) It would suffice in part (i) of Theorem 11.7.1 to assume that the functions fn
are analytic in D and uniformly bounded in D (see [HilPh]). Further, part (ii) clearly
applies to other domains of C, besides the open half plane or the open unit disk. How-
ever, we will not need these more general forms of Theorem 11.7.1 in the applications.
(c) Theorem 11.7.1 remains valid (with the exact same proof) for functions fn which
are vector-valued (with range in a complex Banach space E) or operator-valued (with
values in £(£), the space of bounded linear operators on E). In this last case, the
assumptions of continuity must be replaced by "strong continuity", and the limit in
(11.7.1) (or in the statement of part (i)) holds in the strong operator topology. (Further,
under U.S. or applicable copyright law.

the hypothesis that {fn }^L0 is uniformly bounded must be interpreted in the norm of E or
L(E), respectively. Also, L1 (F) still denotes the space of C-valued integrable functions
on Y.) Recall that a function g : D C C —> E (resp., C.(E)) is said to be analytic if
(and only if) for all continuous linear functionals L on E, the map z -> L, g(z) >
(resp., z -> < L, g(z)u >, for any fixed vector u in E) is an ordinary C-valued analytic
function on D. (See [HilPh, Definition 3.10, pp. 92-93]; here, < •, • > denotes the
duality bracket between E and its dual.) In addition, all the natural notions of analyticity
(a priori, two for E-valued functions and three for L(E)-valuedfunctions) coincide. (See
[HilPh, Theorem 3.10.1, p. 93].) This latter fact follows from the uniform boundedness
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
principle.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: EXTENDED VITALI'S THEOREM AND UNITARY GROUPS 269
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Analytic continuation and product formula for unitary groups

Our aim here is to present some results from [La6, §3]—extending results of Chernoffin
[Cher2] and completing our discussion at the end of Section 11.3—which point out the
limitations of the analogy between the "real" case, corresponding to self-adjoint semi-
groups, and the "purely imaginary" case, corresponding to unitary groups. (In physics,
these would be called the "imaginary time" case and the "real time" case, respectively.)
By analytic continuation (Theorem 11.7.1 (i)), we draw some preliminary information
from results of Kato [Kat5], who has obtained a product formula for an arbitrary pair of
self-adjoint contraction semigroups. (See Proposition 11.7.3.) We then deduce from the
extended Vitali theorem (Theorem 11.7.1(ii)) a very weak form of the product formula
for unitary groups in this context. (See Proposition 11.7.4.)
Let H be a complex Hilbert space. Recall that if A is a nonnegative self-adjoint
operator in H, the semigroup [e~zA : z e C+} is analytic for Rez > 0 and strongly
continuous for Rez > 0. (See, for example, [Kat8, §IX.1.6] and [Gol, HilPh, ReSi2,
Yo]. This fact can be proved much like Theorem 13.3.1) Here, e~zA is defined by means
of the functional calculus (Theorem 10.1.11) since the function e~zx is bounded for x
in a (A) c [0,+00).
The next proposition ([La6, Proposition 3.1, p. 1706]) supplements [Cher2, Theorem
7.2, p. 84].
Proposition 11.7.3 Let A, B be arbitrary nonnegative self-adjoint operators on H.
Assume that

exists for all t e R.

Then Q(A)n Q(B) (= D ( A 1 / 2 r ) D ( B 1 / 2 ) ) is dense in H and U(t) coincides with the
unitary group generated by A-}-B, the form sum of A and B; namely, U(t) = e-"<A+B\
for all t € R
Proof According to a theorem of Chernoff [Cher2, Corollary 2.5.4, p. 32], U(t) is a
(strongly continuous) unitary group of operators on 7i. By Stone's theorem (Theorem
9.6.11), we then know that there exists a self-adjoint operator C on ~H such that U(t) —
e-itc, for all t e E.
under U.S. or applicable copyright law.

By the theorem of Kato mentioned above [Kat5] and briefly discussed in Remark
11.3.6(b), we have

where fl denotes the orthogonal projection of ft onto Ti.\ := Q(A) n Q(B).

Now, the (L(H)-valued) function Un(z) := ( e - (z/n)A e -W B ) n is analytic for
RezEBSCO
> Publishing
0 and strongly continuous for Rez > 0. Moreover, the sequence {t/ }£L0 is
: eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDADn
uniformly
DISTRITALbounded for Re
FRANCISCO JOSE z > 0 (since || Un (z) || < 1 for Re z > 0). By applying Vitali's
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 270 PRODUCT FORMULAS AND FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

classical theorem (Theorem 11.7.1(1) and Remark 11.7.2(c)) as well as the principle of
analytic continuation (on the right-hand side), we deduce from (11.7.4) that

The functions U(z) and e zC are analytic for Rez > 0 and bounded, strongly
continuous for Re z > 0. Since, in addition, they coincide on the boundary, Re z = 0,
they must also coincide in the entire right half-plane Re z > 0. For z — t (with t > 0),
this fact combined with (11.7.5) (or (11.7.4)) yields

In particular,

Thus Q(A) n Q(B) is dense in H, A + B is a self-adjoint operator on U (= U\)

tC t(A+B)
and e~ — e~ , for all t > 0. These two semigroups must then have the same
infinitesimal generators. Whence C = A + B, as required. D
As we pointed out at the end of Section 11.3, the above proposition shows that the
difficulty with Problem 11.3.9 (concerning a general product formula for unitary groups)
is no longer to identify the limit in (11.3.34), but rather to justify its existence.
Finally, we close this appendix by providing some additional information regarding
(part (ii) of) Problem 11.3.9. (See [La6, Proposition 3.2, p. 1708] which extends the
corresponding result in [Cher2, p. 90].)
Proposition 11.7.4 Let A, B be arbitrary nonnegative self-adjoint operators in H. Let
n be the orthogonal projection of H onto its closed subspace H1 := Q(A) n Q(B).
Then

uniformly on compact subsets of the open right half-plane Re z > 0. Moreover, we have
under U.S. or applicable copyright law.

for all p e L l (R).

Of course, if Q(A + B) — Q(A) n Q(B) is dense in H., then we may omit n on the
right-hand side of formulas (11.7.7) and (11.7.8).
Proof In view of (Remark 11.7.2(c) and) the main result in [Kat5] recalled in (11.7.4)
above, (11.7.7) follows from the classical Vitali theorem (Theorem 11.7.1(i)), while
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
(11.7.8) follows
DISTRITAL from
FRANCISCO JOSEthe
DE extended
CALDAS Vitali theorem (Theorem 11.7.l(ii)). n
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: EXTENDED VITALI'S THEOREM AND UNITARY GROUPS 271
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Aside from the very weak type of convergence obtained in (11.7.8) above, it appears
to be difficult to obtain more precise information in this way about Problem 11.3.9.
In conclusion, it does not seem possible to deduce directly by analytic continuation
the existence of

from that of

where t > 0. We will see, however, in Chapter 13 as well as in Chapters 15-18, that
various kinds of analytic continuation procedures are used in defining (and then studying)
certain approaches to the Feynman integral.
Our final comment points out another difference between self-adjoint semigroups
and unitary groups in this context. In hindsight, it is quite natural since, in concrete
settings, they correspond to very different physical situations (such as the diffusion case
and the quantum-mechanical case, respectively).
Remark 11.7.5 Note that we assume throughout that the operators A and B are both
nonnegative. This corresponds to the assumptions of Problem 11.3.9(ii) (or Corollaries
11.3.5 and 11.3.7(ii)) rather than of Problem 113.9(1) (or Theorem 11.3.1 and Corollary
11.3.7(i)), where B- is allowed to be unbounded. Indeed, if we assume instead that
B = B+ — B- is unbounded from below (or equivalently, that its negative part
B- is unbounded), then the product formula for self-adjoint semigroups given in
(11.7.4) no longer makes sense. This is so because then, for each t > 0, the oper-
ator e-tB = e -tB+ e' B - is unbounded (as can be seen, for example, by applying the
multiplication operator form of the spectral theorem (Theorem 10.1.8) in conjunction
with Exercise 10.1.2(b)). Thus, obviously, in this situation, we cannot deduce (after ana-
lytic continuation) results about a (possible) product formula for unitary groups from a
corresponding formula for self-adjoint semigroups.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

12

THE FEYNMAN-KAC FORMULA

Be this as it may, the formal analogy between [formula (7.4.2) above] and integrals appearing in
Wiener's theory is striking and since Wiener's theory is rigorously founded Feynman's heuristic
connection between the Schrodinger equation and the path integral can be made into an unassailable
theorem [namely, the Feynman-Kac formula].

Mark Kac, 1980 [Kac4, p. 51]

In its various guises the F-K [i.e., Feynman-Kac] formula is ubiquitous throughout much of
quantum physics on the one hand and probability theory on the other. It is probably safe to say
that I am better and more widely known for being the K in the F-K formula than for anything else
I have done during my scientific career....
The differential equation my formula solves is nineteenth-century, while the Wiener integral is
a creation of the twentieth. It turned out that many properties of the solution which were difficult
and even opaque from the nineteenth-century point of view became simple and transparent when
looked upon from the point of view of the twentieth. The formula and its many close and distant
relatives have proven to be useful and sometimes powerful tools in a variety of problems. I got
tremendous mileage out of them and so did many others. Hardly a month passes without someone
discovering yet another application.

Mark Kac, 1984 [Kac5, pp. 115-116]

It is not true that these problems all require functional integration for their solution (although, at
the present moment, some of them have only been solved with such methods), but they all share
the property of being problems with "obvious'-' answers and with elegant, conceptually "simple"
solutions in terms of the tools we shall develop here. Once the reader has understood these methods
and solutions, he will probably have little trouble giving a "word-by-word translation" into a
solution that never makes mention of functional integration but rather exploits the Trotter product
formula and the fact that e'A is an integral operator with a positive kernel. That is, there is a sense,
under U.S. or applicable copyright law.

somewhat analogous to the sense in which the Riemann integral is a systematized limit of sums,
in which the Feynman-Kac formula is a systematic expression of the Trotter product formula and
positivity of <?' A . In part, the point of functional integration is a less cumbersome notation, but
there is a larger point: like any other successful language, its existence tends to lead us to different
and very special ways of thinking.

Barry Simon, 1979 [Si9, p. 4]

The Feynman-Kac formula, which expresses the heat semigroup e~t(H°+ V) in terms of
a Wiener integral, is one of the most influential results in the theory of path integration.
Kac's formula does not deal directly with quantum mechanics although it is related
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
to that subject
DISTRITAL and JOSE
FRANCISCO indeed was inspired by Feynman's path "integral". These points
DE CALDAS
wereAN:discussed
98476 ; Johnson, Geralddetail
in some W., Lapidus, Michel L..;
in Section 7.6,Thewhere
FeynmananIntegral and Feynman's
informal introduction to the
Operational Calculus
Account: ns000601
 THE FEYNMAN-KAC FORMULA AND THE HEAT EQUATION 273
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Feynman-Kac Formula via the Trotter product formula as well as a brief account of the
beginnings of the subject were also provided. Applications of Feynman-Kac formulas
in various settings have been given in a broad range of areas such as, for example,
diffusion equations, the spectral theory of Schrodinger operators, quantum mechanics,
statistical physics and quantum field theory [BraRo, Car, Frei, GliJa, ReSi2, Si9, Va].
In addition to these references, Kac's original papers [Kacl,2] and his later expository
writings, notably his paper in the Wiener Memorial Volume [Kac3] and his 1980 Fermi
lectures Integration in Function Spaces and Some of its Applications [Kac4], make very
interesting reading.

12.1 The Feynman-Kac formula, the heat equation and the Wiener integral
We first state the hypotheses under which we will establish the Feynman-Kac formula.
Let V : W* —> M be a measurable function, with positive part V+ = max(V, 0) and
negative part V_ = max(— V, 0), and let //o = —5 A. We will assume that V+ e
Lioc(Rd) and that V_ is Ho-form bounded with bound less than 1 (see Definition 10.3.17);
that is,

and there exist constants 0 < y < 1 and S > 0 such that

[In this chapter, as before, we often abbreviate /Rrf \q>(y)\^dy by f |^|2; we also write
||V</>|| 2 for the square of the magnitude of the (distributional) gradient of <p. Further, as
in Sections 10.3, Q(Ho) denotes the form domain of the operator //o and, since HO is
nonnegative, the second representation theorem for forms (Theorem 10.3.15), implies
that Q(H0) = D(#01/2).] Recall from Section 10.4 (or 10.3) that the quadratic form

with domain Q(q) := Q(H0) f~l Q(V+),

under U.S. or applicable copyright law.

is semibounded and closed. Consequently, by Theorem 10.4.8 (or 10.3.19), there exists
a unique semibounded self-adjoint operator, denoted by H := Ho + V (and called
the form sum of HQ and V), associated with this quadratic form. In the mathematical
physics literature dealing with quantum theory, H = HO + V is often referred to as
the Hamiltonian with potential (or interaction term) V. As we will see below, in the
Feynman-Kac formula (12.1.4), the contribution of the free Hamiltonian HO will be
absorbed in the term involving Wiener measure m.
Of course, if V belongs to L°°(Rd), then we simply have H = H0 + V, the ordinary
sumEBSCO
of the operators
Publishing HOCollection
: eBook and V, with domain
(EBSCOhost) D(H)on =6/8/2017
- printed D(Ho);5:41further, H is essentially
PM via UNIVERSIDAD
d d
self-adjoint on V(RJOSE
DISTRITAL FRANCISCO ) =DE C^(R
CALDAS ) in that case.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 274 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We saw in Section 10.4 (Proposition 10.4.4) that a (nonnegative) function V_ which

is in any of the classes

satisfies the condition (12.1.1) required in Theorem 12.1.1 below; in (12.1.3), we assume
that p > d/2 if d > 2 or p = 1 if d - 1. (Also see Remark 10.4.9(a).) In that case, H is
a natural realization of -± A + V(£) in L2(Rd). (Recall that the definition of the Kato
class Kd was given in Definition 10.4.1.)
We are now prepared to state the main result of this chapter, the Feynman-Kac
formula. This formula, under the conditions given in Theorem 12.1.1 below, is stated
and proved in Simon's book [Si9], Functional Integration and Quantum Physics. (See
[Si9, Theorem 6.2, p. 51].) Our proof will fjllow the outline in [Si9] but is much more
detailed. The result that we establish is slightly different than the theorem in [Si9], which
will be given later in Corollary 12.3.3.
Theorem 12.1.1 (The Feynman-Kac formula) Let V : Rd -+ M be such that V+ e
L11oc(]Rd) and V- is Ho-form bounded with bound less than 1. Then the heat semigroup
e~'H is given for every t > 0 by the formula

where iff e L2(l&d), £ 6 R rf , C'0 is the space of"K." -valued continuous functions x on
[0, t] such that x(Q) =0 (i.e. Wiener space) and m is the associated Wiener measure (as
defined in Chapter 3). The equality (12.1.4) is in the sense of L?-functions; in particular,
for each 1/r 6 L 2 (R d ), the two sides o/(12.1.4) are equal for Leb.-a.e. % € Ed.
Once Theorem 12.1.1 is established, it follows immediately from Theorem 9.1.4
(connecting a semigroup of operators to the evolution equation associated with its
infinitesimal generator) that the solution to the heat equation associated with the Hamil-
tonian H — — j A - j - V = HQ + V can be represented via the Feynman-Kac formula
(12.1.4), as desired. We formally state this key result as a corollary of Theorem 12.1.1.
Corollary 12.1.2 (Heat equation and the Feynman-Kac formula) The solution
under U.S. or applicable copyright law.

of the heat (or diffusion) equation

is given at time t > 0 and (a.e.) point t- € M.d by the Wiener integral on the right-hand
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
sideDISTRITAL
of '(12.1.4).
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE FEYNMAN-KAC FORMULA AND THE HEAT EQUATION 275
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 12.1.3 (a) In probabilistic notation, (12.1.4) becomes

where E (resp., E%) denotes expectation with respect to Wiener paths starting at
0 (resp., £) at time t = 0.
(b) The reader will note that the assumptions on the potential V = V+ — V_ in Theo-
rem 12.1.1 (or Corollary 12.1.2) are the same as in Theorem 11.4.2 (or Corollary 11.4.5)
which guarantees the existence of the modified Feynman integral T1M (V) in this setting.
Recall, however, that Theorem 11.4.2—which follows from a product formula for imag-
inary resolvents (Theorem 11.3.1)—expresses as a limit of suitable finite-dimensional
integrals the solution to the Schrodinger (rather than the heat) equation associated with
the quantum-mechanical Hamiltonian H = Ho + V. (That is, in physicists" terminol-
ogy, it deals with the "real time" case rather than the "imaginary time" case.)
(c) The hypothesis V+ e L$oc(R.d) can be weakened to V+ € L/^K^G), where G
is a closed subset of Rd of Lebesgue measure 0; indeed, it is not hard to show that q is still
densely defined in that case, with D(Rd \ G) dense in L2(Rd) and satisfying D(Rd \ G) C
Q(Ho)r\Q(V+) = Q(q). Actually, onecanallow V+ to be any (nonnegative) measurable
function on Rd, provided that the semigroup e~tH is regarded as acting in the Hilbert
space Q(q), the closure of Q(q) in L2(Rd). (We note that a direct probabilistic proof of
the Feynman-Kac formula for arbitrary nonnegative, measurable potentials was given
by McKean in [McK].)
(d) Under the general hypotheses of the theorem, it is not true that the integral
f0 V(x(s) + $-)ds exists for all % and x. However, it follows from Theorem 12.1.1 that
for Leb.-a.e. f, this integral is defined for m-a.e. x. (See Corollary 12.3.1 below for a
more precise statement.)
Now we turn to a measure-theoretic lemma which will be useful in the proof of the
Feynman-Kac formula as well as later on in the book. Let E : [0, t] x C'0 -> Rd be the
evaluation map
under U.S. or applicable copyright law.

Lemma 12.1.4 Let N be a subset of Rd of Lebesgue measure 0. Then for m-a.e. x € C'0,
we have Leb.({s e [0, t] : x(s) € N}) = 0.
In other words, m-almost surely, the "occupancy time" of the Wiener path x in N
is zero.
Proof Since every set of Lebesgue measure 0 is contained in a Borel set of Lebesgue
measure 0 (as follows from [Roy, Proposition 12.7, p. 294]), it suffices to establish this
result for a Lebesgue null Borel set N. By Exercise 3.2.6, E is continuous and so E ~ 1 ( N )
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
is aDISTRITAL
Borel subset of [0,
FRANCISCO JOSEt]DExCALDAS
CQ. Now if XF denotes the characteristic function of the set F,
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 276 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

By (12.1.7) and the Fubini theorem,

Hence, for m-a.e. x in C'0, f0' X E ~ I ( N ) ( S ^ X ^ S ~ 0- Thus, for m-a.e. x in C'0, 0 =

XE~I(N)(S> x) = X{se.[Q,t\.x(s)eN}(s) for Leb.-a.e. s e [0, t]; that is, for m-a.e. x in C'Q,
Leb.fs e [0, t] : x(s) e N] = 0 as we wished to show. . a
Corollary 12.1.5 Let N e B(Rd) be such that N has Lebesgue measure 0. Then for
every £ e IRrf it is the case that for m-a.e. x e CQ,

Leb.({s € [0, f] : *(s) +1 € N}) = 0.

Proof Fix f e Md. Then A? - £ e S(lRrf) and Leb.(N - ?) = 0. Now simply apply
Lemma 12.1.4 with Af replaced by N — t-. D

12.2 Proof of the Feynman-Kac formula

We will prove Theorem 12.1.1 in four steps: First, for V continuous (with compact
support) in Step 1, and then V bounded in Step 2; second, for a general potential satisfying
the hypotheses of Theorem 12.1.1, in Steps 3 and 4.

Bounded potentials
Step I: V is a continuousfunction of compact support. (One could begin with an infinitely
differentiable function of compact support, but the extra assumptions would not simplify
the proof of this step.) Recall that the operator My of multiplication by V is a self-adjoint
under U.S. or applicable copyright law.

operator, and that the operator e~tMv = e~'v. Further recall that HQ is a nonnegative
self-adjoint operator and that the contraction semigroup {e~'H° : t > 0} is given on
L2(Rd) by (see formula (10.2.26) of Theorem 10.2.6)

Thus

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF THE FEYNMAN-KAC FORMULA 277
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and so the nth Trotter product is given by

where in the last expression of (12.2.3) we take e = UQ and we identify (Rd)" with Rdn.
Letting Vj = Uj + e, j — 1, . . . , n, we obtain for every integer n > 1:

where MO = 0 in (12.2.4). Now from (12.2.4) and Theorem 3.3.5 (a basic Wiener inte-
gration formula for finitely based functionals) we can write
under U.S. or applicable copyright law.

We would now like to apply Theorem 11.1.4, the Trotter product formula, to
( e -(t/n)Hoe e- (t/n)V)n However, Theorem 11.1.4 is not quite directly applicable since
(for s > 0) the operator e~sV need not be a contraction. But, by Proposition 10.1.1, the
norm of this operator equals ||e~sV/(^ ||oo which is less than or equal to e~ s " v "°°. Thus
the semigroup e~s(v+]l\v^ = e^s^v^aoe~sV is a contraction semigroup. Using this,
s V|l
the EBSCO
fact that the operators
Publishing of multiplication
: eBook Collection (EBSCOhost) -by the constant
printed function
on 6/8/2017 e~~UNIVERSIDAD
5:41 PM via N °° commutes
DISTRITAL FRANCISCO JOSE DE 2
CALDAS d
with every operator in L(L (R )), and (a simple case of) the Trotter product formula
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 278 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(Theorem 11.1.4), we can write

for every i/f 6 L 2 (R d ), where the limit (as n -> oo) is in the norm on L 2 (R d ). Of course,
it follows immediately from (12.2.6) that

with the limit holding in the norm || • \\2.

[Note that the Trotter product formula (Theorem 11.1.4) obviously applies here.
Indeed, let W := V + ||V||oo; then, since W is bounded and nonnegative, so is
the multiplication operator by W. Hence the algebraic sum HQ + W (with domain
D(Ho)) is itself self-adjoint and nonnegative. It follows that —(Ho + W) is
w-dissipative (see also Remark 10.1.12(b)) and thus Theorem 11.1.4 can be applied
to yield (12.2.6). Finally, observe that H0 + W = H0+ W = H0 + W = H + || V||oo-
(The above argument could be simplified somewhat by observing that if the Feynman-
Kac formula (12.1.4) holds for a given potential V, then it also holds for its translates
V + c,for any constant c. Hence we could have assumed without loss of generality that
V > 0 and deduced directly (12.2.7) from Theorem 11.1.4.).]
It follows from the convergence just established that there exists a subsequence { n p }
for which the convergence in (12.2.7) holds for Leb.-a.e. e e R.d. Since the paths x are
continuous and since V is continuous under our present assumptions,
under U.S. or applicable copyright law.

for all x e C'0 and f e Rd. (This is the Riemann sum approximation procedure alluded
to in Section 7.6, just below Equation (7.6.8).) Thus, for every f e W1, we have as
p —»• oo:

for m-a.e. x (specifically, for every x such that ty(x(t} + f) is defined).

NowPublishing
EBSCO (12.2.7) : and
eBook(12.2.9)
Collectionwill allow -usprinted
(EBSCOhost) to finish the proof
on 6/8/2017 ofviatheUNIVERSIDAD
5:41 PM Feynman-Kac
DISTRITAL
formula FRANCISCO
under JOSE DE CALDAS of Step 1 by taking the limit on both sides of (12.2.5)
the assumptions
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF THE FEYNMAN-KAC FORMULA 279
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(with n replaced by np) as p -> oo provided that we can justify an application of the
dominated convergence theorem on the right-hand side. But

and by Theorem 3.3.5 again (or, more specifically, by formula (9.2.15) with f := \ifr\ e
L2(Rd)),

Step 2: V e L°°(Rd). We will use Step 1 and a limiting argument to establish the
Feynman-Kac formula for V & L°°(Rd). With this in mind, we want a sequence {Vn}
from T>(Rd) such that | Vn (it) \ < || V || oo for every positive integer and every u € Rd, and,
in addition, Vn(u) —> V(u) for Leb.-a.e. u e Rd. We indicate in some detail how such a
sequence can be produced by means of an approximate identity although we will not give
a complete proof. Take a function J e D(R d ) such that J > 0, J(0) > 0, J ( u ) = J ( v )
when ||u|| = ||v||, J ( u ) > J ( v ) for ||u|| < ||u||, and /R</ J(u)du = 1. (The graph of
a 1-dimensional version of such a function /* appears in Figure 12.2.1. Let K(u) :=
y*(||w||2) and then take J = cK, with the constant c chosen so that fRd J(u)du = 1.)
Now for n = 1 , 2 , . . . , let J\(u) = ndJ(nu), Note that J\_ e T>(Wi) and that
« «
fyi.d JL (")^M = 1 for n = 1, 2, — Next we let
n
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
FIG. JOSE
DISTRITAL FRANCISCO 12.2.1. The graph of the function J* (when d = \)
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 280 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Clearly, for every positive integer n and every u e Md,

Also, Wn is infinitely differentiable ([H6r3, Theorem 4.1.1, p. 88] or [Stel, Theorem 2,

part (b) (the p = oo case), pp. 62-64]) and Wn(u) -> V(u) for Leb.-a.e. u. It need not
be the case, however, that Wn is compactly supported. In order to remedy this, we begin
by taking 0 e D(R d ) such that 0 < 0(w) < 1 for all u, 0(0) = 1, and fRd 0(u)du = 1.
Now let

It is easy to see that Vn has the pleasant properties possessed by Wn and, in addition, Vn
is compactly supported.
Summarizing, we now have a sequence {Vn}£lj from T>(Rd) such that |V n (w)| <
|| V||oo for every u e M.d and for every n, and, in addition, Vn(u) -» V(w) for Leb.-a.e.
u eR d .
LetHn = Ho + Vn,n = 1,2, We wish to apply Theorem 9.7.10, an earlier result
which dealt with the continuous dependence of a semigroup on its generator. From the
third paragraph preceding the statement of the present theorem, we know that each Hn as
well as H — HO + V is self-adjoint on D(Ho) and is essentially self-adjoint on P(Md).
Further, all of these operators are uniformly bounded below; specifically, we have in the
sense of self-adjoint operators,

as we now explain. Since by construction of the function Vn, we have Vn(u) > — ||V||oc
pointwise, we can write for any 4> £ D(Ho):
under U.S. or applicable copyright law.

and similarly for H. Hence (12.2.14) holds.

Now(12.2.14) [and the spectral theorem (Theorem 10.1.ll(ii) to be precise)] easily
imply (see Exercise 10.1.13(b)) that

Inequality (12.2.15)
EBSCO Publishing shows
: eBook that the
Collection "stability
(EBSCOhost) condition"
- printed involved
on 6/8/2017 5:41 PM in
viathe hypotheses of
UNIVERSIDAD
Theorem 9.7.10
DISTRITAL is satisfied
FRANCISCO in our present setting.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF THE FEYNMAN-KAC FORMULA 281
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In order to finish showing that Theorem 9.7.10 can be applied, it suffices to prove
that for every <j> eV(Rd),

(Recall from Theorem 10.2.12 that Z>(Rd) is an operator core for H since H is essentially
self-adjoint on P(Erf).) But \\(H0 + Vn)4> - (Ho + V)(p\\2 = \\Vn<j> - V<j>\\2, and so it
suffices to show that, for </> e P(R rf ),

In fact, we will show that (12.2.17) holds for any <j> e L 2 (E rf ). Accordingly,

where the second equality in (12.2.18) comes from the dominated convergence theorem
(since \Vn(u) - V(w)| 2 |</>(w)l 2 < 4||V|| 2 0 i^(M)| 2 € L 1 ^)) and the last equality from
the fact that Vn(u) -> V(u) for Leb.-a.e. u e Rd.
We now have the hypotheses of Theorem 9.7.10 satisfied and so, for every t/r e
Lr 2fmd\
(M. ),

as n -> oo. [In fact, Theorem 9.7.10 assures us that the convergence in (12.2.19) is
uniform for t in compact subsets of [0, +00). We do not need this uniformity at present
but it is sometimes a useful fact.]
Now let N be the Borel subset of W1 of Lebesgue measure 0 such that for u e
N, Vn(u) -» V(u). By Corollary 12.1.5, for every £ e E.d it is the case that for m-a.e.
x e CQ, x(s) +%eNforat most a subset of [0, t] of Lebesgue measure 0. Hence for
a.e. s in [0, t],
under U.S. or applicable copyright law.

Since \Vn(u)\ < ||V||oo for all n and all M, (12.2.20) and the bounded convergence
theorem imply that for every £ e Rd it is the case that for m-a.e. x e CQ,

Thus for every E and m-a.e. x, we have

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 282 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and so for Leb.-a.e. £ and m-a.e. x,

as n —> oo. Since the function x -> e t||V||00 |i/r(jt(0 + £)| is a dominating function for
the left-hand side of (12.2.23) which is integrable with respect to m as we saw earlier in
(12.2.11), we deduce from the dominated convergence theorem that for Leb.-a.e. f e Rd,

Now by (12.2.19), we know that we can pass to a subsequence {np} such that for
Leb.-a.e. £,

Thus by combining (12.2.25), Step 1 of the proof, and (12.2.24), we can write

Hence the Feynman-Kac formula is established for the case involved in Step 2.

Monotone convergence theorems for forms and integrals

We next wish to consider the general case when V+ e L,1^ and V- is H0-form bounded
with bound less than 1. In order to deal with this case, we provide some operator-theoretic
background regarding monotone-type convergence theorems for quadratic forms. We
under U.S. or applicable copyright law.

restrict our attention to a Hilbert space H. (In Steps 3 and 4 below, H will be the
space L 2 (R d ), equipped with its natural norm || • ||2 = || • IIz, 2 ^) (Rd) inner product
(.,.) = (•, •)L 2 (Rd).) At this point, some readers may want to review the introductory
material on unbounded quadratic forms provided in Section 10.3. Recall that we denote
by Q(q1) the domain of the quadratic form q\ and write q\ (p) instead of q\ (p, p).
Our treatment will follow [ReSil, Supplement to §VIII.7, pp. 372-377], where the
proof of Theorems 12.2.2 and 12.2.4 below can be found. (Also see [Kat8, §VIII.3 and
§VIII.4, pp. 459-462], as well as [Si5,6] and the references therein.)
Definition 12.2.1 Let q\ and q2 be quadratic forms on a (complex) Hilbert space H.
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
We DISTRITAL
say that FRANCISCO
q\ <q2JOSEi fDE
Q (CALDAS
q 2 ) Q(q1) and q 1 ( p ) < q 2 ( p ) for all <p € Q(q1).
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF THE FEYNMAN-KAC FORMULA 283
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A moment's reflection will convince the reader that the inclusion Gfe) =2 (q1) is
natural in the previous definition.
We can now state the first convergence theorem, which will be used in Step 4.
Theorem 12.2.2 (Monotone convergence theorem for nondecreasing forms) Let
q1, q2,... , q n , . . . , be a nondecreasing sequence of closed, nonnegative forms:

Let

Then, if 2(qoo) is dense in H, the quadratic form qoo with domain Q(qoo) and defined
by

is closed, Moreover, if Tn (resp., T) is the nonnegative self-adjoint operator associated

with the form qn (resp., qoo), then as n -> oo,

[/ + T n ]-1 —> [I + T] in the strong operator topology (or strongly, in short);

(12.2.29a)

that is, for all p e H,

To state the second result, we need some additional preliminaries. We say that the
form q2 is closable if it admits a closed extension. In that case, q2 has a smallest (in the
sense of Definition 12.2.1) closed extension q2, called the closure of q2.
Proposition 12.2.3 Let q2 be a nonnegative quadratic form on H. Then q2 admits a
under U.S. or applicable copyright law.

largest closable form (q2)r, that is smaller than q2_.

We are now prepared to state the second convergence result, which will be used in
Step 3 below. (In that case, the limiting form qoo will be closed and hence we shall simply
have qoo = (qoo)r = (qoo)r.)

Theorem 12.2.4 (Monotone convergence theorem for nonincreasing forms) Let

q1, q2,..., qn,..., be a nonincreasing sequence of closed, nonnegative quadratic forms:

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 284 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Let Q(qoo) = Uoo n=1 Q(qn) and consider the form qoo with domain Q(qoo) and
defined by

Let Tn be the (nonnegative self-adjoint) operator associated with the form qn and let T
be the (nonnegative self-adjoint) operator associated with the closure of (q00)r (where
(qoo)r is defined in Proposition 12.2.3). Then as n —> oo,

Remark 12.2.5 (a) In the literature, one often refers to the convergence in (12.2.29)
and (12.2.31) as "strong resolvent convergence" of Tn to T.
(b) To avoid unnecessary complications, we have assumed implicitly that the
quadratic forms involved are densely defined.
Finally, we shall need the following simple extension of the ordinary monotone
convergence theorem for integrals; it is proved much like the standard theorem. (See, for
example, Exercise 6, page 89 and the remark at the bottom of page 92 in [Roy].)
Theorem 12.2.6 (Generalized monotone convergence theorem for integrals) Let
(X, A, u) be a measure space with u a positive measure. Let {fn} be a sequence of
measurable real-valued functions that is bounded below (resp., above) by an integrable
function g. Suppose further that fn —> f u-a.e.as n -» oo and that for each n, fn < f
(resp., fn > /) u,-a.e. Then

Remark 12.2.7 Note that the sequence {fn} in Theorem 12.2.6 is not assumed to be
monotonic. When we apply Theorem 12.2.6 in Steps 3 and 4 below, however, we shall
havefn| f (resp., f n f f).

Unbounded potentials
Let V satisfy the hypotheses of Theorem 12.1.1; hence V+ e L,1loc(Rd) and V- is
under U.S. or applicable copyright law.

Ho-form bounded with bound less than 1. Given positive integers n and m, we define
d)
Vn,m ee LLoo(R
\7 bu
(K ) by

We will now complete the proof of Theorem 12.1.1 in two additional steps. In Step 3, we
will keep m fixed and let n -» oo (after having applied Step 2 to the bounded potential
V n,m ). Finally,
EBSCO Publishingin: eBook
Step Collection
4, we will let m —>
(EBSCOhost) oo and
- printed conclude
on 6/8/2017 5:41 that
PM viathe Feynman-Kac
UNIVERSIDAD
formula (12.1.4)
DISTRITAL holds
FRANCISCO JOSE for the potential V.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF THE FEYNMAN-KAC FORMULA 285
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Step 3: Define the quadratic form qn,m by

Note that qn,m > 0. Indeed, since V n , m — V n , m + — Vn,m, - > — V-, where V n,m, ±
denotes the positive (negative) part of Vn m, we have for <p e Q(Ho) = Q(Ho) n Q(V-)
(by (12.1.la)):

where we have used hypothesis (12.1.1b) in the second inequality.

Clearly, the form qn,m is closed as the sum of a closed form (<p i-> ^ / IIV<p|| 2 —
(Hocp, <p), <p e Q(Ho)) and a bounded one. Further, since (for fixed m)

we have

in the sense of quadratic forms (see Definition 12.2.1). [Note that Q(qn,m) — Q(Ho)
for all n, so that the inclusion between domains required by Definition 12.2.1 is trivially
satisfied.]
Let Tn,m be the nonnegative self-adjoint operator associated with the closed, nonneg-
ative form qn,m (see Theorem 10.3.15). Clearly, Tn,m = HO + (8 + Vn<m), with domain
D(Tn,m) = D(//o); in other words, Tn,m is simply the algebraic sum of HO and of the
bounded operator 8 + Vn^m.
Let q00,m be the quadratic form defined in Theorem 12.2.4; that is,
under U.S. or applicable copyright law.

We now show that

where

Indeed, since Vn,:m eBook

EBSCO Publishing ], VCollection
m as n -> oc, (12.2.35)
(EBSCOhost) - printedfollows from
on 6/8/2017 5:41(12.2.34) and (12.2.33)
PM via UNIVERSIDAD
by use of Theorem
DISTRITAL FRANCISCO 12.2.6, the generalized monotone convergence theorem for integrals.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 286 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

More precisely, fix <p € Q(H0) c L2(Rd) and m > 1. Then by (12.2.32), Vi,m < m and
so Vi,m M2 < m\ip\2 e Ll(K.d); since in addition, V n > m |^| 2 | Vm\<p\2, we deduce from
Theorem 12.2.6 that / Vn<m \<p\2 J, / Vm |<p|2 as n -> oo and hence, in light of (12.2.33)
and (12.2.34), we conclude that (12.2.35) holds.
Let Vm, _ be the negative part of Vm. Since by (12.2.36), V m, - = V_ is Ho-form
bounded with bound less than 1, the nonnegative form qoo,m is closed (see Theorem
10.3.19). Hence, in the notation of Theorem 12.2.4, (qoo,m)r = qoo,m. Let Tm be the
(nonnegative) self-adjoint operator associated with q00,m. Note that by Theorem 10.3.19,
Tm = H0 + (S + Vm), the form sum of the operators HO and S + Vm.
By Theorem 12.2.4, the monotone convergence theorem for nonincreasing forms,
[I + Tn,m]~} -> U + Tm]-] strongly as n -> oo. By Theorem 9.7.11 (or 8.1.12), a key
approximation theorem for semigroups, it follows that for all t > 0, e-tTn,m -> e-tTm
strongly as n —> oo; namely for every \fr e L 2 (R d ),

Let € L°°(Rd) and t > 0. Since F n,m = HO + (S + Vn,m), with Vw,m e L°°, we
may apply Step 2, the Feynman-Kac formula for bounded potentials, to deduce that for
Leb.-a.e. $ € Rd,

Next we turn to an examination, under the assumption of Step 3, of the Wiener

integral side of the Feynman-Kac formula (12.2.38). Since V n,m Vm Leb.-a.e. as
n -> oo, —Vn,m t ~Vm Leb.-a.e. as n -» oo. Hence by Corollary 12.1.5, we deduce
as in Step 2 that for every £ and m-a.e. x

for Leb.-a.e. .s e [0, t]. Thus by a second application of Theorem 12.2.6,

under U.S. or applicable copyright law.

for every £ and m-a.e. jc. (Note that —m < — V l , m (x(s)+f) and Q |— m\ds — mt < oo;
so that, in view of (12.2.39), Theorem 12.2.6 can be applied to yield (12.2.40).) It follows
immediately from (12.2.40) that for every f and m-a.e. x,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF THE FEYNMAN-KAC FORMULA 287
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Hence for <£ e L2(Wt) with tj> > 0, we have, for Leb.-a.e. £,

(To deduce (12.2.42) from (12.2.41), it suffices to apply the ordinary monotone conver-
gence theorem since all the functions involved are nonnegative.)
Now, by (12.2.37), we can pass to a subsequence {np} such that for Leb.-a.e. £,

Then by combining (12.2.42), the Feynman-Kac formula (12.2.38) obtained in Step 2,

and (12.2.43), we see that for Leb.-a.e. f,

We next make some comments which will serve to both help clarify equation
(12.2.44) and be useful to us in Step 4 (as well as to establish Corollary 12.3.1 below).
The last expression in (12.2.44) is in L2(Ed) as a function of £ and so the same is true
of the first expression. Hence for Leb.-a.e. £, the integral (in the first expression) over
Ct exists. Also, the integrand (which is > 0 here since <j> > 0) must be finite for m-a.e.
x. Thus (taking the case (p > 0), we see that
under U.S. or applicable copyright law.

Note that the symbol /0'(<5 + V m (jf(.s)+ £))</$• is defined since by (12.2.36), Vm < m and
so /Q'(S + Vm,+(x(s) + $))ds <St + mt<oo. Therefore /0'(<5 + Vm(x(s) + %))ds =
—oo only for a set of paths of m-measure zero. Hence for Leb.-a.e. £, the function
s M>- 8 + Vm(x(s) + |) is integrable on [0, t]. We can thus conclude that for Leb.-a.e.
fcfc
£ a Rd
10*^ ,

2 d
We now
EBSCO complete
Publishing the
: eBook proof of(EBSCOhost)
Collection Step 3. For an arbitrary
- printed (complex-valued)
on 6/8/2017 ^ e L (E. ),
5:41 PM via UNIVERSIDAD
we DISTRITAL
can writeFRANCISCO JOSE —
V = (<t>\ DE 4>2)
CALDAS
+ *(</>3 — $4) with each <j>}• (j = 1 , . . . , 4) a nonnegative
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 288 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

function in L2(Rrd); here, <j>\ — 02 is the decomposition of Re rjf into its positive and
negative parts and cfe -$4 is the decomposition of Im i/s into its positive and negative parts.
By (12.2.44), the Feynman-Kac formula holds for each of the nonnegative functions (j>j
and so, by linearity, also holds for arbitrary iff € L 2 (R d ); that is for every m > 1 and
ir e L2(Ed), we have for Leb.-a.e. £,

This concludes Step 3.

Step 4: In this step, we let m -> oo and complete the proof of Theorem 12.1.1. Since
clearly Vm f V by (12.2.36), it follows from (12.2.35) that

in the sense of quadratic forms. Let q00 be the limiting nonnegative form as in Theorem
12.2.2; that is,

In view of definitions (12.2.35) and (12.2.46a), it follows from the generalized mono-
tone convergence theorem for integrals (Theorem 12.2.6) that for tp e Q(H0),
under U.S. or applicable copyright law.

where the quadratic form q is defined by (12.1.2). The use of Theorem 12.2.6 can be
justified as we now briefly explain. Since Vm, __ = V_, (<5-V_)|<p| 2 < (5+y m )|<p| 2 for all
m; but for Q(H0) c Q(V-) (by (12.1. la)), V_|<p| 2 € Ll and so (S- V-)\<p\2 e L1.
Hence by the first part of Theorem 12.2.2,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
Thus (12.2.46a)
DISTRITAL yields
FRANCISCO JOSE (12.2.47).
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF THE FEYNMAN-KAC FORMULA 289
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

if and only if

Therefore, (12.2.46b) yields

It thus follows from (12.2.47) and (12.2.48) that

Consequently, if T is the nonnegative self-adjoint operator associated with q00,

where H = H0 + V is the Hamiltonian appearing in the statement of this theorem.

We can now apply the second part of Theorem 12.2.2, the monotone convergence
theorem for a nondecreasing sequence of closed forms {qoo,m}oom=1, and conclude that
[I + T m ]-1 -» [/ + 7T1 strongly as m -> oo. (Note that by (12.2.48), Q(qoo) is dense
in L 2 (R d ) as required by the hypotheses of Theorem 12.2.2.) By Theorem 9.7.11 (or
8.1.12), the Trotter-Kato-Neveu-Chernoff theorem, we thus deduce (as in Step 3) that
for all t > 0,

e-tTm -> e-tT in the strong operator topology as m -> oo. (12.2.49)

Fix <j) > 0, (j> e L2(Rd). By (12.2.49), there exists a subsequence {mp} such that for
Leb.-a.e. f in Rd,
under U.S. or applicable copyright law.

We next apply the Feynman-Kac formula (12.2.45) obtained in Step 3; namely, for
each m > 1 and 0 e L 2 (R d ),

Assume for now :that

EBSCO Publishing <t>Collection
eBook > 0. Note that since
(EBSCOhost) 0 > 0,
- printed the right-hand
on 6/8/2017 side
5:41 PM via of (12.2.51) is
UNIVERSIDAD
nonnegative and hence
DISTRITAL FRANCISCO JOSE so is the left-hand side. Let Z = Z ( j ) be the set of points f in
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 290 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Rd such that (12.2.50) holds (pointwise) and both sides of (12.2.51) exist and are finite,
for all m > 1. Then clearly, R d \Z has Lebesgue measure 0.
Fix f e Z. According to the comment following equation (12.2.44), we then know
that for m-a.e. .tin Ct, the functions H» Vm(x(s)+%) is integrable on the interval [0, t].
Therefore, since — Vm —V, we may apply Theorem 12.2.6, the generalized monotone
convergence theorem for integrals, to deduce that

Note that for m-a.e. x e Ct 0, the integral on the right-hand side of (12.2.52) is either —oo
or finite and thus, its counterpart in (12.2.53) is either 0 or > 0, but in any case finite.
(One minor point: In view of Corollary 12.1.5, we may assume that £ € Z implies that
<p(x(t) + £) is well defined for m-a.e. x.)
By Theorem 12.2.6, the right-hand side of (12.2.51) converges to

[Observe that by our choice of Z, the integrand on the right-hand side of (12.2.51) is a
(nonnegative) integrable function of x over Ct.] Thus by passing to the limit along the
subsequence {mp} on the left-hand side of (12.2.51), we deduce from (12.2.50) that
under U.S. or applicable copyright law.

as desired. By the linearity of both sides of (12.2.54), the Feynman-Kac formula (12.1.4)
follows for any ^ e L 2 (R d ). This concludes the proof of Theorem 12.1.1.
Exercise 12.2.8 Try to prove Theorem 12.1.1 by interchanging Steps 3 and 4; that is, by
first letting m —» oo and then letting n —» oo.

12.3 Consequences
We close this chapter by giving a few simple corollaries of the Feynman-Kac formula
stated
EBSCOinPublishing
Theorem: eBook
12.1.1 (or Corollary
Collection (EBSCOhost)12.1.2).
- printedIn
on each of 5:41
6/8/2017 these corollaries,
PM via UNIVERSIDAD we will
implicitly
DISTRITALassume
FRANCISCOthat
JOSEthe hypotheses of Theorem 12.1.1 are satisfied.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 CONSEQUENCES 291
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 12.3.1 For Leb.-a.e. % in Rd, the integral /0' V(x(s) + f )ds exists and is
either finite or oo for m-a.e. x in Ct0; i.e.

In particular, if V is bounded above, then for Leb.-a.e. £ in Rd, the function s -»

V(x(s) + £) is integrable on the interval [0, t] for m-a.e. x in C'0.
Exercise 12.3.2 Prove Corollary 12.3.1.
[Hint: The proof is in the spirit of the comment following formula (12.2.44) in Step 3,
except that the potential need not be bounded above here.]
The next corollary yields the version of the Feynman-Kac formula that is given in
[Si9, Theorems 6.1 and 6.2, pp. 49 and 51].
Corollary 12.3.3 Let p, $ e L 2 (R d ). Then

where (•, •) denotes the inner product in L 2 (R d ).

Proof By the Feynman-Kac formula (12.1.4),
under U.S. or applicable copyright law.

then by the Fubini theorem, this last iterated integral is equal to the right-hand side of
(12.3.2), as desired.
We next briefly justify the above use of the Fubini theorem in (12.3.3). Assume both
(p, $ > 0. Hence the first integral in (12.3.3) is finite and thus so is the iterated integral in
(12.3.3). Therefore the Fubini-Tonelli theorem yields formula (12.3.2) in this case. For
general complex-valued (p, fy in L 2 (R V ), we complete the argument by use of linearity.

Remark 12.3.4 The measure Leb. x m on Rd x C'0 appearing in (12.3.2) is referred to

by Simon [Si9, p. 38] as "Wiener measure". For us, Wiener measure is the probability
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
measure m FRANCISCO
DISTRITAL acting onJOSEC'DE
0. (See
CALDASChapter 3.)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 292 THE FEYNMAN-KAC FORMULA
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 12.3.5 The heat semigroup [e-tH = e - t ( H o + V ) ; t > 0} is positivity pre-

serving; that is, for all t > 0, e-tH t(> > 0 Leb.-a.e. whenever <p > 0 Leb.-a.e. with
<p 6 L 2 (R d ). Hence the solution of the heat equation (in the sense of semigroup theory
as discussed in Chapter 9)

when <p > 0 (< 6 D(H)) is nonnegative for all times: u(•, t) > 0 for all t > 0.
Proof This is an immediate consequence of the Feynman-Kac formula (12.1.4) and of
Corollary 12.1.2.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

13
ANALYTIC-IN-TIME OR -MASS OPERATOR-VALUED
FEYNMAN INTEGRALS

13.1 Introduction
In this chapter, we define and study the existence of analytic-in-time or -mass operator-
valued Feynman integrals. Sections 13.2, 13.3 and 13.7 will deal with analytic continu-
ation in time, whereas Sections 13.5 and 13.6 will be concerned with analytic continuation
in mass. In 13.2, 13.3, 13.7 and much of 13.5, the analytic continuation will be from an
appropriate Wiener integral (as is generally understood when "the" analytic Feynman
integral is discussed). However, in the last part of 13.5 and in 13.6, the analytic continu-
ation in mass will start from an operator-valued function of mass (as well as time) which is
obtained from a product formula. The product formula is of the "Trotter type" in Section
13.5 but does not follow from the Trotter product formula in Section 11.1. In Section
13.6, the product formulas are of the "resolvent type", but again do not follow from the
product formulas for the resolvents of self-adjoint or normal operators given in Sections
11.3 and 11.6, respectively. We will refer to the integrals obtained by analytic continu-
ation in mass from appropriate product formulas as analytic-in-mass Feynman integrals.
This seems to us to be reasonable terminology even though it has been restricted in the
past to cases where the starting point was a true integral with respect to Wiener measure.
Our notation in Section 13.6 and in the last part of Section 13.5 distinguishes between the
different starting points. The former deals with the "analytic-in-mass modified Feynman
integral", ^^(V), from [BivLa], while the latter deals with extensions .?Y'£ian(V),
based on [Kat7, BivPi, Biv3], of Nelson's work [Nel] on the "analytic-in-mass operator-
valued Feynman integral", K^(F-iv)-
The realization that there is a connection between Feynman's path integral and the
Wiener integral, a true Lebesgue integral with respect to the countably additive Wiener
under U.S. or applicable copyright law.

measure, is due to Mark Kac [Kacl,2]. Kac's insight not only produced the Feynman-
Kac formula which was discussed in a very general setting in the previous chapter, but
has also suggested approaches to "the" Feynman integral via analytic continuation from
the Wiener integral. These matters were discussed informally in Section 7.6 where some
background material and related references were also provided.
The discussion in Section 7.6 focused primarily on operator-valued analytic continu-
ation in mass from the Wiener integral. This approach will play a prominent role in
Section 13.5. It will also be used throughout Chapters 15-18, but the goal of those chap-
tersEBSCO
is quite different from the goal of Section 13.5 and so requires different techniques
Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
andDISTRITAL
a more stringent definition
FRANCISCO JOSE DE CALDASof the analytic in mass operator-valued Feynman integral
(compare Definition 15.2.1W.,with
AN: 98476 ; Johnson, Gerald Lapidus, Michel L..;
Definitions The Feynman
13.5.1 Integral and Feynman's
and 13.5.1')-
Operational Calculus
Account: ns000601
 294 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We will show in Theorem 13.3.1 that the analytic-in-time operator-valued Feynman

integral Jit ( F v) (see (13.2.1) and Definition 13.2.1) exists and agrees with the unitary
group e-"(Ho + V) for all t in K whenever V = V+ — V_ satisfies the following two
assumptions: (i) V+ 6 L,1loc(Rd). (In fact, it is sufficient that V+ e L/^OR^G), where
G is a closed subset of Rd of Lebesgue measure 0.) (ii) V_ is H0-form bounded with
relative form bound less than 1.
We note that the existence of the modified Feynman integral and its agreement with
the unitary group was established in Corollary 11.4.5 under precisely the same assump-
tion on V as in (i) and (ii) above. This permits us to conclude easily (Corollary 13.4.1) that
under these same very general assumptions, the modified Feynman integral, the analytic-
in-time operator-valued Feynman integral, and the unitary group associated with the stan-
dard Hamiltonian approach to quantum dynamics all exist and agree. Further, under the
still quite general assumptions used in Theorem 11.2.19 to establish the existence of the
Feynman integral defined via the Trotter product formula (namely, the assumptions that
(iii) V+ 6 L2loc(Rd) and (iv) V- is H0-operator bounded with relative operator bound
less than 1), we conclude (in Corollary 13.4.2) that all three of the approaches to the
Feynman integral via the Trotter product formula, the modified Feynman integral, and
the analytic-in-time operator-valued Feynman integral exist and agree with one another
and with the unitary group. Still further, these consistency results can be combined with
the "dominated-type" convergence theorem of Lapidus for the modified Feynman inte-
gral, Theorem 11.5.19, arid some additional considerations to establish dominated-type
convergence theorems for both the analytic-in-time operator-valued Feynman integral
and the Feynman integral defined via the Trotter product formula (see Corollaries 13.4.3
and 13.4.6, respectively).
Section 13.4 is particularly informative even though the proofs found there are not
especially difficult. Indeed, the combination of new results and the review of earlier
results provides an answer to some of the common objections made to mathematical
theories of "the" Feynman integral (see items 1-3 along with the related discussion in
Section 1.1 of Chapter 1, the introduction to this book).
Theorem 13.3.1—which establishes the existence of the analytic-in-time operator-
valued Feynman integral—rests on some key results discussed earlier in this book:
Theorem 10.3.14, the first representation theorem for quadratic forms; Theorem 10.3.19,
an abstract result which gives conditions under which the form sum of self-adjoint oper-
ators is self-adjoint; Theorem 10.4.8, which insures that HO + V is self-adjoint and
under U.S. or applicable copyright law.

bounded below under the conditions on V discussed earlier; and Theorem 12.1.1, the
Feynman-Kac formula. With these results in hand, the proof of Theorem 13.3.1 is not
difficult. It depends primarily on consequences of the spectral theorem.
The 1963 papers of Babbitt [Bab2] and Feldman [Fel] seem to have been the first
mathematically rigorous papers on the Feynman integral which used analytic continu-
ation in time. (A 1960 paper of Cameron [Cal] used scalar-valued analytic continuation
in mass.)
Theorem 13.3.1 is due to Johnson [Jo6]. Before [Jo6], there seems to have been a
rather widespread opinion among "Feynman integrators" that the class of potentials V
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
for DISTRITAL
which the analytic
FRANCISCO JOSEin
DE time
CALDASFeynman integral exists is extremely limited. The work
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 INTRODUCTION 295
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

on [Jo6] started with the idea of understanding and explaining this opinion. Instead a
positive result, Theorem 13.3.1, was proved (Theorem 6.1 of [Jo6]).
The 1963 paper of Feldman [Fel] which was mentioned above dealt with very general
nonnegative potentials. In retrospect, this paper should have had more influence than it
did. However, [Fel] is quite technical and does not apply to negative potentials such as
the attractive Coulomb potential, the most basic potential of atomic physics.

A number of variations of "the" Feynman integral defined via operator-valued ana-

lytic continuation in mass are discussed in Sections 13.5 and 13.6. Before commenting
on the individual results, we emphasize some features that these approaches have in
common and mention a few differences. The reader can consult the sections themselves
for further discussion of the differences.
A major strength shared by all of these approaches is that the potentials are allowed
to have arbitrarily strong singularities independent of sign. A serious weakness is that
the existence of the strong operator limits depends on a nontangential approach to the
imaginary axis and, even then, these limits are shown to exist only for Leb.-a.e. value on
that axis.
The proofs of the results (some of which are just outlined or remarked on, or even
omitted) have major differences but also have some common elements. The concept of
(Newtonian) capacity plays a key role in all of these approaches. Throughout Sections
13.5 and 13.6, with the exception of Haugsby's extension (Theorem 13.5.9) of Nelson's
results, the arbitrary singularities can occur on a closed set of capacity 0. In the approach
of Nelson [Nel], this set of capacity 0 is the only place where finite range singularities
are allowed to occur. In contrast, in Section 13.6 and near the end of 13.5, other strong
(but not arbitrary) finite range singularities are permitted.
Product formulas play a role in each of these approaches. A link is made with the
Wiener integral in [Ne 1 ] and [Hau]. This is used to give a rather easy proof of the product
formula in the case of [Nel]. The "product formula" in [Hau] is not discussed below
but we remark that it is harder to prove both because of more general hypotheses and
because arbitrary partitions of the time interval are considered and the limit is taken as
the norm of the partition goes to zero. The product formula (13.5.41) used in the last
part of Section 13.5 is due to Kato [Kat7, p. 107]; its proof depends in part on the deep
distributional inequality (11.2.4) which is also due to Kato. A product formula due to
Bivar-Weinholtz and Lapidus [BivLa] (see (13.6.1)) is key to Section 13.6. It is used, in
under U.S. or applicable copyright law.

particular, to establish the existence of the "analytic-in-mass modified Feynman integral"

under very general hypotheses.
All of the approaches in Sections 13.5 and 13.6 make crucial use of operator-valued
versions of three beautiful results from complex analysis: Vitali's convergence theorem
(discussed earlier in Theorem 11.7.1 and reviewed in the proof of Theorem 13.5.6),
Poisson's representation theorem and the Fatou-Privaloff (or edge of the wedge) theorem
(both reviewed in the proof of Theorem 13.5.7).
Another common thread that runs through Sections 13.5 and 13.6 (and contrasts
with the approaches in Chapter 11 and in Sections 13.2-13.4 and 13.7) is that there
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
is no guarantee
DISTRITAL that
FRANCISCO JOSEthe evolving system is described by a unitary group. (It is
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 296 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

always described by a contraction semigroup, but in general—for instance, for highly

singular attractive central potentials—the time evolution is not reversible; see Example
13.6.18.)

We turn next to a brief discussion of some of the results in 13.5 and 13.6. The
earliest mathematically rigorous papers on "the" Feynman integral defined via analytic
continuation in mass were those of Cameron [Cal, 1960] and Nelson [Nel, 1964]. It is
the beautiful results of Nelson that will be treated most completely in Section 13.5. (See
the beginning of Section 13.5 for additional comments on the early papers [Cal, Nel]
and the even earlier 1956 paper [GelYag] of Gelfand and Yaglom.)
The hypothesis on the potential V in Nelson's results, Theorems 13.5.4, 13.5.6 and
13.5.7, is easily described: There is a closed subset J of Rd of capacity 0 such that V is
continuous and R-valued on Kd \ J. Note that there is no separate mention of the positive
and negative parts of V as there was in the results on the Feynman integral in Chapter
11 and as there will be in Sections 13.3, 13.4 and 13.7 of this chapter. The hypothesis
just stated above (which is the same as (13.5.5) below) permits V to have arbitrarily
strong singularities at oo and on J but requires continuity and so permits no finite range
singularities on the open set R d \J.
Haugsby's extension [Hau], Theorem 13.5.9, of Nelson's results comes next in
Section 13.5. Arbitrarily strong singularities of the potential V are still allowed much as in
Nelson's theorems. Further, V is permitted to be complex-valued and time-dependent.
In addition, Nelson's continuity assumption is considerably weakened (although not
removed entirely). Theorem 13.5.9 is the only result in this book in which the potential
can be time-dependent and also have strong spatial singularities. Haugsby's theorem
does not extend all aspects of Nelson's work. For example, one cannot expect the semi-
group property to be satisfied when V is time-dependent. Thus the information about
the semigroups from Nelson's results does not carry over to Haugsby's setting.
As part of his work on the analytic in mass Feynman integral in [Nel], Nelson
established some interesting operator-theoretic results for the Schrodinger operator with
complex potential. The Wiener integral was naturally involved in this work (see Defini-
tions 13.5.1 and 13.5.1') and the Lebesgue integral with respect to Wiener measure was
used in a crucial way to establish a Trotter-type product formula (see (13.5.11) below).
Kato showed in [Kat7] that the operator-theoretic aspects of Nelson's work including
the product formula can be obtained and significantly extended by means of operator-
under U.S. or applicable copyright law.

theoretic methods and without the use of the Wiener integral. It is worth noting that Kato
precisely identified the domain of the generator of the semigroup which gives the evo-
lution. Nelson's result contained only partial information in this direction; see Remark
13.5.5(a). (In fact, Nelson did not need to know the precise domain of the generator to
accomplish his main purpose.)
Starting from the Trotter-type product formula of Kato (13.5.41), one can analytically
continue in mass much as Nelson did and obtain "the" operator-valued Feynman integral
in Theorem 13.5.16 below. Kato did not do this although it certainly seems that he could
have. (Of course, operator theory was his main area of interest and expertise. It is also
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
possible that he did not feel entirely comfortable with the elusive Feynman integral.)
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 INTRODUCTION 297
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

What are the conditions on the potential V in Theorem 13.5.16 (the theorem which
stems from [Kat7]) and how do they compare with the assumptions (13.5.5) in Nelson's
results? Recall that Nelson assumed the existence of a closed subset J of Rd of capacity
0 such that V is continuous and R-valued on Wd\J. Theorem 13.5.16 allows V to
be C-valued but assumes the dissipativity condition, Im V < 0. Also, much less than
continuity is required of Von Rd\J, namely that V e L r l o c (R d \J) where r = 2d/(d+2)
for d > 3, r > 1 for d = 2 and r = 1 for d = 1.
Corollary 13.5.18 is a consistency result. It tells us that when the hypothesis (13.5.5)
of Nelson's results are satisfied, then the assumptions of Theorem 13.5.16 are also sat-
isfied and the two Feynman integrals agree. Moreover, this corollary takes advantage
of both results by identifying precisely the domain of the generator and making the
connection with the Wiener integral.
Kato's results in [Kat7] have been extended in various ways by Brezis and Kato
[BreKat], Bivar-Weinholtz [Bivl-3], Bivar-Weinholtz and Piraux [BivPi], and Bivar-
Weinholtz and Lapidus [BivLa]. Theorem 13.5.22 summarizes consequences (of all but
the last) of these papers for the analytic in mass operator-valued Feynman integral where
the analytic continuation starts from a Trotter product type formula. Part (a) of Theorem
13.5.22 ([Biv3], generalizing [BivPi]) extends Theorem 13.5.16 in that the potential
V is only required to be in L1 loc (R d \J) instead of in Lrloc(Rd\7), where J is again a
closed subset of Rd of capacity 0. Both a scalar and a magnetic vector potential are
considered in part (b) of Theorem 13.5.22. (However, from a mathematical point of
view, the assumptions made on the magnetic potential are not as natural as those made
in Section 13.6.)
The analytic-in-mass modified Feynman integral is considered in Section 13.6. The
existence of such integrals is established in Theorem 13.6.4 and relies on a product
formula for resolvents given in Theorem 13.6.1. The product formula is itself a special
case of a more general result, Theorem 13.6.7. These results on the analytic in mass
Feynman integral are based on work of Bivar-Weinholtz and Lapidus in [BivLa, §3].
The potential V is assumed to be in L1 loc (R d \J), where J is a closed subset of Rd
of capacity 0. Also, V is allowed to be C-valued provided the dissipativity condition
Im V < 0 is satisfied. Further, the Schrodinger operator may involve a magnetic vector
potential a whose components are required to be in L2loc(Rd\J). The proof of Theorem
13.6.7, on which the results just mentioned depend, involves techniques from the theory
of elliptic partial differential equations which are not used elsewhere in this book. It also
under U.S. or applicable copyright law.

relies on a suitable version and extension of Kato's distributional inequality (see Lemma
13.6.6).
Theorems 13.6.10 and 13.6.11 are comparison results for analytic-in-mass, operator-
valued Feynman integrals which are in some respects analogous to the earlier comparison
results, Corollaries 13.4.2 and 13.4.1. These theorems stress the unity of these analytic-
in-mass Feynman integrals within the intersection of their domains of validity.
We conclude Section 13.6 by considering highly singular central potentials with
the emphasis on potentials of the form ^j-, where r is the distance from the origin in
R3; see Example 13.6.13. Such potentials are used, for instance, as phenomenological
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
models for FRANCISCO
DISTRITAL molecular JOSEinteractions
DE CALDAS and aspects of quantum field theory. The attractive
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 298 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

inverse-square potentials ^f- are studied in Example 13.6.18 and have some particularly
interesting features.

In Section 13.7, we will discuss recent results of Albeverio, Johnson and Ma

[AUoMa] which extend substantially the already very general work in Section 13.3
on the analytic-in-time operator-valued Feynman integral. We have not provided in this
book the necessary prerequisites for a detailed and mathematically rigorous treatment of
[ AUoMa] and so we will settle in Section 13.7 for a discussion intended to give the reader
some understanding of the results. The missing prerequisites include smooth measures /A
on Rd (see Definition 13.7.8) and the associated positive continuous additive functionals
Aul of Brownian motion (see Definition 13.7.11) as well as operators Hu := H o + u ,
where the usual potential V can be replaced by the measure u (see 3' and 4' in Section
13.7). These ingredients come together in a Feynman-Kac formula

which extends the Feynman-Kac formula (12.1.4) from Chapter 12.

The measure u. plays the role of a "potential" but u need not be absolutely continuous
with respect to Lebesgue measure in Rd. As a consequence, there need not be a potential
V in the usual sense of the word. Area measure on the surface of a sphere in R3 is
an example which is of physical interest and satisfies the assumptions of the results in
Section 13.7 but not the assumptions of Theorem 13.3.1.
The results in Section 13.7 provide additional information about the analytic in
time operator-valued Feynman integral even when the "potentials" u are absolutely
continuous withrespect to Lebesgue measure in Rd;i.e., u ( d x ) — Vu(x)dx as in Section
13.3. For instance, the positive potential V = Vu in Example 13.7.20 is nowhere locally
integrable. This implies that the form domain of HO + V does not contain any nonzero
continuous function. In spite of this, Ho + V is densely defined on L2(Rd) and J i t ( F y )
agrees with the unitary group e-it (H0 + V) for every t in R.
The reader should compare the results in Section 13.7 with the analytic continuation
in mass results in Sections 13.5 and 13.6 where the limits as the imaginary axis is
approached are taken nontangentially and still may only exist for Leb.-a.e. value on the
imaginary axis. On the other hand, it is only in Sections 13.5 and 13.6 that arbitrarily
strong singularities are permitted which are independent of sign. In all of the other
approaches to the Feynman integral in Chapters 11 and 13, the assumptions on V_ are
under U.S. or applicable copyright law.

more restrictive than those on V+. (It may be useful here to recall that physically, V_
and V+ represent, respectively, the attractive and repulsive part of the potential V.)
Section 13.7 is the only place in Chapters 11 and 13 where potentials V = V+ — V-
are allowed to be replaced by certain measures u = u+ — u-. However, it seems likely
that this could be done for the modified Feynman integral as well (with the aid of the
results of Section 11.4 and of parts of Section 13.7, see Remark 13.7.24(b)).

13.2 The analytic-in-time operator-valued Feynman integral

Recall that Ct = C([0, t], Rd) denotes the space of continuous Revalued functions
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
d
on DISTRITAL
[0, t]. Let V : RJOSE
FRANCISCO ->DE CALDAS
R be Lebesgue measurable. (In Sections 13.5 and 13.6, the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME OPERATOR-VALUED FEYNMAN INTEGRAL 299
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

"potential" V will be allowed to be C-valued.) Our primary interest in this section (as in
the following one) is in functions F = FV on C' which are of the form

(Without further assumptions on V, the subset of C' on which FV is actually denned may
be very small.) However, as we will see further on, there are reasons to be interested in
functions which are not necessarily of the form (13.2.1), and so we will give the definition
of the analytic in time operator-valued Feynman integral without the assumption that F
has the special form above. [Recall from Section 10.2 that C+ (resp., C+ ) denotes the
set of complex numbers z with Re z > 0 (resp., Re z > 0).]

The operator-valued function space integral J t ( F ) exists if and only if (13.2.2) defines
J t ( F ) as an element of L(L 2 (R d )). If J t ( F ) exists for every t > 0 and, in addition,
has an extension (necessarily unique) as a function of t to an operator-valued analytic
function on C+ and a strongly continuous function on C+, we say that J t ( F ) exists for all
t E C+. When t is purely imaginary, Jt (F) is called the analytic in time operator-valued
Feynman integral.
We now make several remarks concerning the definition just given. The comments
in (a) below are most important since they describe our present concerns more fully and
indicate how they are related to the last chapter and to previous operator-theoretic matters.
Remark 13.2.2 (a) Note that when F — FV is given by (13.2.1), the integrand on the
right-hand side of (13.2.2) has exactly the same form as in the Feynman—Kac formula,
Theorem 12.1.1. Hence, if the potential V satisfies the hypotheses of Theorem 12.1.1,
the Wiener integral in (13.2.2) is equal for Leb.-a.e. E to the operator e-tH from the heat
semigroup { e - t H : t > 0} acting on the function W and evaluated at E. Thus, J t ( F v )
is certainly defined for all t > 0 under these circumstances. Using the comments just
made as a starting point, we will show that J t ( F y ) exists for every t E C+ and equals,
for t purely imaginary (the Feynman integral case), the unitary group given by the usual
under U.S. or applicable copyright law.

operator-theoretic approach to dynamics. Hence, writing the parameter in the purely

imaginary case as t — ito, to € R, we will have J i t o ( F ) — e~it0H or, with a more
complete notation,

(b) We caution the reader that what is seen as imaginary time above when we start
with the heat semigroup, the Wiener integral and t > 0 but end up in the quantum setting
with t = —ito, to E R, is seen as real time by the quantum physicists. Indeed, from their
perspective, it is the methods that use the heat semigroup and the Wiener integral to
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
study quantum
DISTRITAL problems
FRANCISCO JOSE DE that are referred to as "imaginary time techniques".
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 300 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(c) Definition 13.2.1 involves the concept of analyticity of an L(L 2 (Rd))-valued

function. But what type of analyticity is meant? Fortunately, as was briefly discussed in
Appendix 11.7, the three concepts of analyticity with respect to the operator norm, strong
operator or weak operator topologies on L(L 2 (Rd)) are all equivalent [HilPh, §3.70,
esp. Theorem 3.10.1, p. 93] and so any of these can be used. (See Remark 11.7.2(c) above.)
(d) As indicated earlier, our main interest in Sections 13.2-13.4 and in 13.7 is in func-
tions of the form (13.2.1) when V : Rd —> R. Potentials V which are C-valued and/or
time-dependent are of physical interest (see [Ex], for example) and will be discussed
further on, primarily in connection with the analytic-in-mass operator-valued Feynman
integral. (See especially Section 13.5 and Chapters 15-18. See also Section 11.6 and
Section 13.6 where highly singular complex potentials are discussed in connection with
the modified Feynman integral.) We note that Definition 13.2.1 above does make sense
for such functions (as well as for certain functions which are very different from the
"Feynman-Kac functional" in (13.2.1)).
(e) Definition 13.2.1 was phrased in such a way that it has an all or nothing character
off the positive real axis. (It is in fact very analogous to the definition of the analytic-in-
mass operator-valued Feynman integral adopted in [JoLal, Definition 0.1, p. 10] and
to be discussed later on in the book; see Definition 15.2.1.) This fits the theorem that
we are working towards since the existence of J t ( F ) will be established in this strong
sense. However, one may phrase the definition so that, for example, the question of the
existence of strong limits as the imaginary axis is approached is examined at each point.
(f) There are many possible variations in the definition of "the" analytic Feynman
integral. We have already mentioned that we will consider the analytic-in-mass operator-
valued Feynman integral later on. We gave a limited discussion earlier in Section 4.5
of a scalar-valued analytic Feynman integral (see Definition 4.5.1). Even within the
operator-valued approach, the limits as the imaginary axis is approached can be taken
in the weak operator topology on L(L 2 (R d )) rather than in the strong operator topol-
ogy; further, the analytic continuation can be taken through some subset of C which is
more restrictive than C+.

13.3 Proof of existence

We are now ready to state and prove the theorem which insures the existence of the
analytic-in-time operator-valued Feynman integral for a large class of potentials V. We
will also see that this Feynman integral agrees for the same class of potentials with the
under U.S. or applicable copyright law.

unitary group which specifies the dynamics in the standard operator-theoretic approach
to quantum mechanics.
Theorem 13.3.1 Let V : Rd -> R be such that V+ E Llloc(Rd) and V- is H0-form
bounded with bound less than 1. Also, for any t > 0, let Fv : Ct —>• R be given by
(13.2.1).
Then J t ( F v ) (as in Definition 13.2.1) exists for every t E C+ and is analytic in
C+. For t > 0, J t ( F v ) is given by the right-hand side of (12.1.4), the Feynman-Kac
formula. For all t E C+,
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PROOF OF EXISTENCE 301
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where e-t(H0+V) is given meaning via the spectral theorem applied to the self-adjoint
operator H = H0 + V. In particular, the analytic-in-time operator-valued Feynman
integral J i t o ( F v ) exists for every to E R and we have

where {e-itoH : t0 E R} is the group of unitary operators given by the usual operator-
theoretic approach to quantum dynamics.
Finally, for every y E D(Ho + V), W(to, E) := ( J i t o ( F v ) ( p ) ( % ) is the unique solu-
tion (in the semigroup sense as discussed in Chapter 9) of the Schrodinger equation

with initial state y.

Remark 13.3.2 We remind the reader that we have previously discussed in Section 10.4
conditions which insure that the function V- satisfies the hypothesis imposed on it above.
The least restrictive of these sufficient conditions is that V- belongs to the Kato class
Kd (see Definition 10.4.1). This particular class of functions is mentioned here for the
purpose of a later comparison of Theorem 13.7.16 with the theorem immediately above.
We further recall that under the assumptions of Theorem 13.3.1, the operator H0 + V is
self-adjoint and semibounded below. We forego any further review since the salient facts
were discussed briefly at the beginning of Section 12.1.
Proof of Theorem 13.3.1 In light of Definition 13.2.1 and the Feynman-Kac formula,
Theorem 12.1.1, the key is to show that the operator-valued function T(t) := e-tH is
defined and strongly continuous for all t E C+. (See Remark 13.2.2(a).) It will follow
immediately that J t ( F v ) exists for all t E C+ and (13.3.1) holds. Restricting / to the
imaginary axis, say t — ito, to 6 R, we then have (13.3.2); so that J i t o ( F V ) = e-itoH is
the unitary group from the usual Hamiltonian approach to quantum dynamics. From this
it follows that W(to, E) := (•/""(FV)<")(£) is, for every (y E D(H), the unique solution
(in the semigroup sense) of the Schrodinger equation (13.3.3) with initial state (p.
Establishing that T(t) = e-tH is strongly continuous in C+ and analytic in C+ is
simply a matter of combining some consequences of the spectral theorem with standard
under U.S. or applicable copyright law.

arguments as we will now show. Since H is bounded below, —E:= inf o (H) is greater
than —oo. Because of this, it is easy to check that the function/, : a(H) —»• Cdefinedby
f t ( u ) :— e-tu is a bounded function of u for any t E C+. Using the functional calculus
which arises from the spectral theorem, Theorem 10.1.1 l(ii), we see that T(t) = e~t H
is defined and belongs to L ( L 2 ( R d ) ) for all t E C+.
Next, let {tn} be a sequence in C+ such that tn —> t. Then ftn(u) -> f t ( u ) for all
u E a ( H ) . Further, the sequence {11ftn1100}is bounded, where 11ftn11 oo = sup{|ftn(u)| :
u E a(H)}. Using part (v) of Theorem 10.1.11, we have that ftn(H) -> f t ( H ) strongly;
that is, T(tn) = e~tnH —> e~t H = T(t) in the strong operator topology as n - oo.
It remains
EBSCO only
Publishing to show
: eBook that(EBSCOhost)
Collection - t H is onanalytic
T(t) = -e printed 6/8/2017 in
5:41C+. In UNIVERSIDAD
PM via light of Remark
13.2.2(c),
DISTRITALitFRANCISCO
is enough to DE
JOSE show that(e~tH W, p) is analytic in C+ for every W, y E L 2 (R d ).
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 302 ANALYTIC-IN-TIME OK -MASS FF.YNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In fact, by polarization, it suffices to show that ( c ~ t H p, p) is analytic for every p E

L - ( R d ) . For the rest of the argument, we fix p E L 2 ( R d ) ; we may as well even assume
that \ \ p \ \ 2 — 1. and we do so since it is mildly convenient.
Let P be the projection-valued measure associated wit h the sell-adjoint operator H.
See Definition 10.1.5, Theorem 10.1.16 and the associated discussion for the information
about P which we w i l l need. The measure up.p defined for any Borel subset B of R by
u p . p ( B ) : = ( p . P ( B ) p ) i s a probability measure such that u p . p ( a ( H ) ) = I . Further,
by the projection-valued measure form of the spectral theorem (Theorem 10.1.16).

We know from an earlier part of the proof that e-tH is strongly continuous on C + .
Certainly, then ( e ~ t H p.p) is continuous on C_|.. From Morera's theorem of complex
analysis [Ru2, Theorem 10.17, p. 208], the desired analyticity w i l l follow if we can
show that

for any triangle F in C+. We now give the argument which establishes (13.3.5) and then
comment on the steps:

The first equality in (13.3.6) follows from (13.3.4): the third equality comes from the
Cauchy integral theorem since the scalar-valued function e-tH is certainly an analytic
function of t for t E C+ . The second equality is a consequence of the Fubini theorem. In
order to see that the Fubini theorem is applicable, observe that the now fixed triangular
contour F is bounded and is bounded away from the left half-plane. Hence, there is a
bound for e=tu, where t runs over P and u runs over [ —E. oo). Since up.p is a probability
under U.S. or applicable copyright law.

measure and the integration over the contour is, alter parameterization. integration with
respect to a measure of f i n i t e total variation, the use of the Fubini theorem is justified
and the proof is now complete. D
Remark 13.3.3 (a) Recall from Remark 12.1.3(c) that the hypothesis V+ E L l l o c ( R d )
can he weakened to V+ E L l l o c ( R d \ G ) , where C is a closed subset of Rd of Lebesgue
measure 0. The same change can he made in the theorem which we just finished proving
(ax was done in /Jo6/). Further, since extremelv singular potentials are of interest in
quantum physics, this change has some significance. While C is a small set in terms of
d-dimensional Lehesgne measure, it could he, for example, a (d — 1)-diniensional sub-
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
space of Rd FRANCISCO
DISTRITAL , a ratherJOSE
large set on which to permit arbitrary singularities. (This comment
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 COMPARISON OF FEYNMAN INTEGRALS 303
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

also applies lo the modified Feynman integral studied in Sections 11.4-11.6 us well as
to the Feynmun-Kac formula discussed in Chapter 12.)
(b) We have proved directly that under the assumptions of Theorem 13.3.1, the semi-
group {e~t H : t E C+ } is a hounded holomorphic semigroup (of angle n/2), and hence
is analytic in C+, the open right-half plane. We could have used instead the general the-
ory of holomorphic semigroups to establish this fact. /See / KalH, §IX.1.6] or [ReSi2, pp.
248-2571 ax well as jGol, HilPh, Yo] for the theory of holomorphic (or analytic) semi-
groups. I However, the more general results do not by themselves yield strong continuity
on the imaginarv axis, the Fevmnan case.

13.4 The Feynman integrals compared with one another and with the
unitary group. Application to stability theorems
This section interrelates and expands upon the most central results in this book regarding
existence and stability theorems for "the" Feynman integral. More specifically, we are
now in a position to show that the approaches to the Fey nrnan integral studied so far in this
hook (see Corollary 1 1.2.22 of Theorem 1 1.2.19, Corollary I 1.4.5 of Theorem 1 1.4.2,
and Theorem 13.3.1) are closely related to one another and lo the unitary group from
Ihe standard approach to quantum mechanics. Further, these relationships along with
Lapidus' dominated convergence theorem for the modified Feynman integral, Theorem
1 1.5.19, allow us to establish closely related convergence (or stability) results for the
analytic-in-time operator-valued Feynman integral and for the Feynman integral defined
via the Trotter product formula.
The reader may have noted that the assumptions on V in Theorem 13.3.1 coincide
exactly with the assumptions in Theorem 11.4.2 (or Corollary 1 1.4.5) which guaranteed
the existence of the modified Feynman integral discussed in Section I 1.4. These two
theorems tell us that the modified Feynman integral and the analytic-in-time operator-
valued Feynman integral both agree with the unitary group under the same assumptions.
It follows immediately that these two versions of the Feynman integral must agree with
each other. We state this formally as a corollary of Theorems 13.3.1 and 1 1.4.2.
Corollary 13.4.1 Let V : Rd -> R he such that V+ E Llloc(Rd) and V- is Ho-form
hounded with hound less than I. Also, for any t > 0, let Fv : Ct —> R he given by
(13.2.1).
Then the unitary group e ~ i t ( H o + V). the modified Feynman integral F l M ( V ) and the
under U.S. or applicable copyright law.

analytic-in-time operator-valued Feynman integral .1" ( F v ) all exist for every t E R,

and we have

We can give a somewhat less general related result which also involves the Feynman
integral defined via the Trotter product formula (TPF), obtained in Theorem 11.2.19 and
Corollary 1 1.2.22. The conditions will be such that all three versions of the Feynman
integral exist and agree with the unitary group.
Corollary 13.4.2 Lei V : Rd -> R he such that V, E L2loc(Rd) and V- is H0-operalor
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
hounded with
DISTRITAL houndJOSE
FRANCISCO lessDE than
CALDASI.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 304 ANAI.YTIC-IN-TIMH OR -MASS F H Y N M A N INTHGRAl.S
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then (H0 + V )iD is essentially self-adjoint on D = D ( R d ) . Denoting the closure

(H0, + V )ID by H. we have that the unitary group e itH, the modified Feynman integral
.F1M(V). the Feynman integral F 1 T P , ( V ) via TPF, and the analytic-in-time operator-
valiied Feynman integral Jit ( F v ) all exist for every t E R, and

Proof Under the hypotheses of" Corollary 13.4.2. V_ is M0-form bounded wit h hound
less than 1 (see Proposition 1 1.2.10(1)); further, by Proposition 1 1 . 2 . 1 0 ( i i ) . we also have
H = HO + V, the form sum of M) and V. Therefore. Corollary 13.4.2 follows from
Corollary 1 1.2.22 and Corollary 13.4.1.

Recall from Remark 11.4.3(a) that (by Propositions 10.4.4 and 10.4.5(b)) V satisfies
the condition of Corollary 13.4.1 if it belongs to any of the elasses

where p = 1 for d — I and p > d/2 for d > 2. Also, if V. belongs to either of the
classes L p ( R d ) + L~ (Rd) cLoloc( R d ) u . then we can allow p = d/2 when d > 3.
Further recall that (by Theorem 11.2.1 I. Propositions 11.2.14 and 1 1.2.15(b)) V
satisfies the condition of Corollary I 3.4.2 if it belongs to any of the classes

where p — 2 for d < 3 and p > d/2 for d > 4. (Here and in the following, the spaces
Lploc (Rd)u and Lploc are given as in D e f i n i t i o n 10.4.2.)
Once Theorem 13.3.1 was established (Theorem 6.1 of | Jo6|), it was observed (Corol-
lary 6.2 of |Jo6|) that the connection between F 1 M ( V ) and Jit(Fv) (Corollary 1 3 . 4 . 1 1
and Lapidus' dominated convergence theorem. Theorem 1 1.5.19-, easily yield the fol-
lowing result. [ F r o m now on. as in Section 1 1 . 5 .Vm,+ (resp.. Vm, ) denotes the positive
(resp., negative) part of V m . 1
Corollary 13.4.3 (Dominated convergence theorem for the analytic-in-time Feynman
integral) Let V. Vm. m = 1,2 he Lebesgue measurable real-valued functions on
Rd . Assume that "Vm converges to V dominatedly" in the following sense:
under U.S. or applicable copyright law.

(a) Vm -^ V Leb.-a.c. in Rd.

(b) Vm.+ < U for some U E L l l o c ( R d ) ,
(c) Vm.- < W for some W E Lploc(Rd)u.
where p — 1 if d = I and p is any number in the interval ( d / 2 . oo) for d > 2.

Then Jit (Fv) and J i t ( F v m ) . m = 1.2 all exist and

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
in the strong operator topology, uniformly in t on compact subsets of R.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 COMPARISON OF FEYNMAN INTEGRALS 305
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof Assumptions (a) and (b) imply that all of V+ and V m, +, m = 1, 2 , . . . , are dom-
inated by U. Also, (a) and (c) imply that all of V_ and Vm, _, m = 1, 2 , . . . , are dominated
by W. Then Corollary 13.4.1 along with Propositions 10.4.4 and 10.4.5 imply that all
of F t M ( V ) , J i t ( F v ) , F t M (V m ) and Jit(Vm), m = 1 , 2 , . . . exist, and we have the equal-
ities Ft M (V) = J i t ( F v ) and Ft M (V m ) = Jit(Fvm), = 1, 2 , . . . . Further, by Theorem
11.5.19 (and Theorem 11.5.13), Ft M (V m ) -> Ft M (V) in the strong operator topology,
uniformly in t on compact subsets of R. The assertion (13.4.4) now follows immediately.
D

Remark 13.4.4 (a) Recall from Section 11.5 [La 12] that an entirely analogous dom-
inated convergence theorem holds for the modified Feynman integral. (See, in particular,
Theorems 11.5.19 and 11.5.13.)
(b) After the publication of [Lal2], the second author realized that the class of
functions LPloc(Rd);; could be replaced in his dominated convergence theorem by the
larger class Kd (see Remark 11.5.15(d)). A corresponding improvement can be made in
Corollary 13.4.3; the only adjustment in the proof above is that Proposition 10.4.5 is no
longer needed.
The final result in this section depends mainly on Lapidus' convergence theorem
and on Corollary 13.4.2. We begin with a proposition which will help us with the proof
and will also clarify some related issues. Proposition 13.4.5 and Corollary 13.4.6 can be
found in [JoKim]. Items (i), (ii) and the first containment in (13.4.8) of the proposition
are known. The second containment in (13.4.8) and (iv) may not be known.
Proposition 13.4.5 Let d be any positive integer and let p and q be real numbers
satisfying 1 < q < p < oo. Further, let W be any R-valued, Lebesgue measurable
function on Rd. Then we have the following:
(i) For any x E Rd and r > 0,

where Br(x) is the ball of radius r centered at x in Rd and F is the gamma function. It
follows immediately that
under U.S. or applicable copyright law.

(ii) Further,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 306 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof (i) The Lp-norm on a probability space is an increasing function of p. Hence,

It follows that

Thus (13.4.5) follows since

(ii) Let W e L^CM^)?. By definition, this means that

Hence, by (13.4.5),

Thus W E Lqloc(Rd)u as claimed.

(iii) The first containment in (13.4.8) has already been noted in (10.4.1). See also
the paragraph immediately following Example 10.4.6 where it is noted that the first
containment in (13.4.8) holds for all p E [1, oo).
It is the second containment in (13.4.8) which mainly concerns us. Let W E
Lploc(Rd)u.Then supxERd \\W\\iP(Br(x)) < °°- Also, since q < p,
under U.S. or applicable copyright law.

Hence, under the present assumptions, the (adjusted) argument in (13.4.10) still shows
that WELqloc3(Rd)u.
(iv) The fact that the left-hand side of (13.4.9) is contained in the right-hand side
follows immediately from the first containment in (13.4.8). Finally, let W E Lploc(Rd)u
for some p > c > 1. Choose q such that p > q > c. By the second containment in
d
(13.4.8), we see that
EBSCO Publishing WCollection
: eBook e Lqloc(Rd);r and -soprinted
(EBSCOhost) W EonU6/8/2017
p>c Lploc(R
5:41 PM )u, as we wished to
via UNIVERSIDAD
show.
DISTRITAL FRANCISCO JOSE DE CALDAS D
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 COMPARISON OF FEYNMAN INTEGRALS 307
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 13.4.6 (Dominated convergence theorem for the Feynman integral defined
via the Trotter product formula) Let V,Vm,m = 1,2,..., be Lebesgue measurable
real-valued functions on Rd. Assume that "Vm converges to V dominatedly" in the
following sense:

where p — 2 for d = 1, 2, 3 and p is any number in the interval (d/2, oo)for d > 4.
Then F'TP(V) and Ft T p (V m ),m = 1 , 2 , . . . , all exist and

in the strong operator topology, uniformly in t on compact subsets of R.

Proof The domination assumption (b) insures that each V m ,+ belongs to L2loc(Rd).
Combining (b) and the pointwise convergence from (a), we see that V+ is also in Lloc (Rd).
Combining (c) and (a), we see that each V m , -,m = 1,2,..., and V- are dominated by
W and so all belong to Lploc(Rd)u, where the number p depends on the dimension d as
indicated in (c) above. It then follows from Corollary 11.2.22 and Propositions 11.2.14
and 11.2.15 that the Feynman integrals Ft TP (V) and Trp(V m ), m = 1,2,..., all exist
and agree with the unitary groups e~itH and e~it Hm , m — 1, 2 , . . . , respectively. We also
recall for use below that H and Hm,m = 1, 2,..., are all the closures of the essentially
self-adjoint operators (Ho + V)ID and (Ho + Vm)ID, m = 1 , 2 , . . . , respectively (see
Theorem 11.2.11); i.e.

Next, we recall that L21oc(Rd) c L11OC(Rd). Hencb, our dominating function U satis-
fies the hypothesis (11.5.20) of (V2) of the dominated convergence theorem, Theorem
11.5.13, for the modified Feynman integral.
We also claim that W satisfies the hypothesis (11.5.21) of (V2) of the same theorem.
Comparing the hypothesis just referred to with our assumption (c), we see that it suffices
to show that for d = 1, d = 2, d = 3 and d > 4, respectively, we have:
under U.S. or applicable copyright law.

But Proposition 13.4.5(iii) tells us that (i) holds and that (ii) holds with, say, p = 7/4.
The same part of Proposition 13.4.5 yields (iii) above where we can, for example, again
takeEBSCO
p =Publishing
7/4. Part (iv) of Proposition 13.4.5 yields (iv) above where c is taken as d/2.
: eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
Thus W satisfies
DISTRITAL theJOSE
FRANCISCO desired hypothesis.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 308 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Now by the second author's dominated convergence theorem (see Theorems 11.5.13
and 11.5.19), it follows that F t M ( V m ) ->• Ft M (V); in fact,

in the strong operator topology as m ->• oo, uniformly in t on compact subsets of R.

But since (Ho + V)ID is essentially self-adjoint, we know by Proposition 11.2.10(ii) that
H := (Ho + V)ID (see (13.4.12)) agrees with the form sum HO + V. In the same way,
Hm := (Ho + V m )ID = H0 + Vm, for m = 1, 2 , . . . . Hence, it follows that

in the strong operator topology as m —> oo, uniformly in; on compact subsets of R. But
by Corollary 13A.2, F t T p ( V ) = F t M (V) and F t T p ( V m ) = F t M ( V m ) , and so by (13.4.14),
we have

in the strong operator topology as m ->• oo, uniformly in t on compact subsets of R, as

we wished to show. Q
Remark 13.4.7 (a) Since L P ( R d ) + L°°(Rd) c Lploc(Rd)u for 1 < p < oo (see
Proposition 10.4.5(b) and the paragraph following Example 10.4.6), we see that W E
LP(Rd) + Loo (Rd) implies that the assumption on W in (c) of Corollary 13.4.6 holds.
Here, p depends on the dimension d exactly as in (c) of Corollary 13.4.6.
(b) As noted in Remark 13.4.4 above, the second author's dominated convergence
theorem [Lal2] and so Corollary 13.4.3 can be strengthened by replacing the class
Lploc(Rd)u by the larger class Kd. A related improvement in Corollary 13.4.6 can be
made by replacingLploc(Rd)u by Sd n Kd (see Definition 11.2.13 and recall from Prop-
ositions 10.4.5(b) and 11.2.15(b) that LPloc(Rd)u c Sd n Kd for appropriately chosen
p.) One might guess that Sd is a subset of Kd so that Kd could be replaced by Sd
rather than Sd n Kd. However, Example 11.2.16 shows that Sd is not in general a subset
of Kd.

13.5 The analytic-in-mass operator-valued Feynman integral

The earliest mathematical papers on "the" Feynman integral defined this elusive "inte-
under U.S. or applicable copyright law.

gral" via the analytic continuation in mass of a Wiener integral. The paper of Gelfand and
Yaglom [GelYag] was published in Russian in 1956 and an English translation appeared
in 1960. Their paper is, as we mentioned in Remark 4.6.2(d), well worth reading for its
positive aspects but is best known for an error involving the analytic continuation pro-
cess. (See the discussion in Section 4.6.) The first mathematically rigorous work using
analytic continuation was, as far as we know, done by Cameron [Cal] in 1960 where
the analytic continuation was scalar-valued and in the mass parameter. In that paper,
Cameron pointed out the error in [GelYag] and also gave a positive result under what
nowEBSCO
seems like extremely restrictive hypotheses. (Two early papers [Bab2, Pel] on the
Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
analytic-in-time operator-valued
DISTRITAL FRANCISCO JOSE DE CALDAS Feynman integral appeared in 1963.)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 309
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Our main concern in this section is the beautiful theorem of Nelson on the analytic-
in-mass operator-valued Feynman integral. This result was the focal point of the paper
[Nel] in which additionally the approach to the Feynman integral via the Trotter product
formula was first treated with mathematical rigor (see Sections 11.1 and 11.2). We will
not establish all of Nelson's assertions but we will give a complete proof of the existence
of the analytic-in-mass operator-valued Feynman integral (in the sense of Definition
13.5.1' below) for almost every value of the mass parameter.
Recall that in the approaches to the Feynman integral in Sections 11.2, 11.4, 11.6
and 13.3, the assumptions on V_, the negative part of the potential, were (although quite
reasonable) always stronger than the assumptions on V+. A strength of Nelson's result
is that it allows certain potentials with arbitrarily strong singularities regardless of sign.
(The results to be discussed in Section 13.6 also have this feature.)
Haugsby extended the work of Nelson in his 1972 thesis [Hau] by weakening the
continuity requirement on V and by allowing V to be time-dependent and complex-
valued. Both of [Nel, Hau] have a serious shortcoming which we will discuss below.
(See, in particular, Remarks 13.5.8(a) and 13.5.5(b).) This shortcoming is not shared by
the approaches to the Feynman integral which were considered earlier. We will just state
Haugsby's theorem and give a brief discussion of it.
Unfortunately, Haugsby's thesis has never been published. However, the first author
has stated and briefly discussed the main theorem of [Hau] in [Jo6, pp. 38-50] and in
[Jo7, pp. 15-20]. Part of the present section is based on [Nel, Hau].
In the last part of this section, titled "Further extensions via a product formula for
semigroups", we will state (essentially without proof) later results by Kato [Kat7] and
then discuss their consequences for a suitable version of the analytic-in-mass Feynman
integral. These consequences were not considered in [Kat7], Kato's results (and their later
generalization by Bivar-Weinholtz and Piraux [BivPi, Biv3] also stated at the end of this
section) reprove and extend the operator-theoretic part of Nelson's results without using
Wiener integrals. Instead, they are based on a suitable product formula for semigroups
which is used as a substitute for a generalized Feynman-Kac formula with imaginary
potential.

Definition of the analytic-in-mass operator-valued Feynman integral

Let C and C+ denote, respectively, the complex numbers and the complex numbers
with positive real part; further, let C+ denote C+ with the origin of the complex plane
under U.S. or applicable copyright law.

removed.
We fix a positive integer d and also fix t > 0. Our definition will be given for rather
general functions F : Ct = C([0, t], Rd) —>• C, but our interest in this section will be
in functions of the form

where V : [0, t] x Rd -> C. In fact, most of our discussion will concern Nelson's
theorem in which the "potential" V is time independent so that F has the form indicated
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
in (13.2.1). We remark
DISTRITAL FRANCISCO JOSE that in Chapters 15-18, the functions F that will be considered
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 310 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

will not be restricted to the form (13.5.1). In fact, certain Banach algebras of functions
At,t > 0, will play an important role in those chapters.
Definition 13.5.1 Fix t > 0 and let F be a function from Ct to C. Given n > 0, W E
L2(Rd) and E E Rd, consider the expression

The operator-valued function space integral exists for A. > 0 if (13.5.2) defines Ktn(F)
as an element of L(L 2 (R d )). If Ktn(F) exists for every A > 0 and, in addition, has an
extension (necessarily unique) to an L(L 2 (R d ))-valued function of n which is analytic
in C+, we say that Ktn (F) exists for all n E C+. Finally, for no = —iqo purely imaginary
(qo = 0), we say that Ktno (F) exists provided that the limit, limn->n0 Ktn(F), where A
approaches A.Q through C+, taken in the strong operator topology, exists and equals the
bounded linear operator Ktno (F). We write briefly,

and we callKtno(F) = Kt-iqo (F) the analytic-in-mass operator- valued Feynman integral
of F with parameter n0.
A number of variations of the preceding definition can be given. In fact, one of
these will be the definition that is actually used in the theorems of Nelson and Haugsby.
There will be no change from Definition 13.5.1 for n E C+ but, for n0 = —iqo E
C+\C+, we will require only that

in the strong operator topology as n approaches n0 "nontangentially" through C+. We will

see in the following paragraph and Definition 13.5.1' what "nontangentially" refers to.
Let a e (0, n/2) and let n0 = — iqo purely imaginary (qo / 0) be given. The symbol
Wa (A-o) will denote the wedge in C+ with vertex at no and with both sides making the
angle a with the imaginary axis.
under U.S. or applicable copyright law.

Definition 13.5.1' Suppose that Ktn(F) exists in the sense of Definition 13.5.1 far all
n E C+. Let n0 = — iqo be purely imaginary with qo = 0. We say that Ktn0 (F) exists
provided that for every a 6 (0, n/2),

as A — n0 through the wedge Wa (n0) and where the limit is taken in the strong operator
topology on L(L 2 (R d )). We will again refer to Ktn0 (F) as the analytic-in-mass operator-
valued Feynman integral with parameter n0, but, when confusion might arise, we will
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:41 PM via UNIVERSIDAD
addDISTRITAL
the phrase "in the
FRANCISCO JOSE sense of Definition 13.5.1".
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 311
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Clearly, the existence of the analytic-in-mass operator-valued Feynman integral with

parameter n0 = — iqo in the sense of Definition 13.5.1 implies its existence in the sense
of Definition 13.5.1'. There will be a third definition, Definition 15.2.1, in Chapter 15. In
that case, the Wiener integral on the right-hand side of (13.5.2) will be required to extend
to an L(L2(Rd))-valued function which is strongly continuous in C+ and analytic in C+.
The existence of the Feynman integral in the sense of Definition 15.2.1 certainly implies
its existence for all n0 = -iqo in the sense of both Definitions 13.5.1 and 13.5.1'. Since
Definition 15.2.1 will be used throughout Chapters 15-18, we will refer to it there as
simply the analytic operator-valued Feynman integral. (We may sometimes even drop
the words "operator-valued".)
Definition 13.5.1 has been stated (as has often been done in the literature) so as
to minimize reference to the somewhat subtle issues connected with scaling in Wiener
space. This topic and its implications for the scalar-valued analytic Feynman integral
has been discussed in detail in Chapter 4; Sections 4.5 and 4.6 are especially relevant to
our present considerations. Analogous issues for the analytic-in-mass operator-valued
Feynman integral will be treated in part I of Section 15.2. With that in mind, we will
settle here for a few remarks.
Remark 13.5.2 (a) Using the change of variable theorem, Theorem 3.3.2, we can rewrite
the integral in (13.5.2) as

(The reader may wish to review the definition of ma, Definition 4.2.3, and to compare
(4.6.1) with (4.6.2).) Thus we see thatjust the existence of the integral in (13.5.2) for every
n > 0 means that for every one of the mutually singular measures on Ct0 in the collection
{ma : a > 0} and for Leb.-a.e. E, F(x + E) must be measurable and integrable with
respect to ma; in particular, for Leb.-a.e. E, F ( ( . ) + E) is scale-invariant measurable.
Further, much as in Section 4.5, the proper equivalence relation for functions F(x + E)
and G(x + E) is not simply equality mx Leb.-a.e. but rather equality ma x Leb.-a.e. for
every a > 0.
(b) Our starting point as well as Haugsby 's is the integral (13.5.2) for X > 0. Nelson's
starting point is essentially the same although that is not apparent at first glance. Nelson
under U.S. or applicable copyright law.

begins in [Nel, §27 with a diffusion constant D — ^, where m is purely imaginary

with Im m > 0. Next, Nelson integrates with respect to the probability measure which
corresponds to the diffusion constant D > 0 and to continuous paths which start at E at
time 0. The vector E appears in Nelson's notation but the diffusion constant D does not.
No mention is made of equivalence classes of functions, but since D will (for m purely
imaginary and Im m > 0) run over all positive numbers, the appropriate equivalence is
equality ma x Leb.-a.e. for every a > 0.
(c) Nelson eventually analytically continues to real m, with m nonzero. Throughout,
his m and our A. are related by the simple equation n = — im. Comparing formula (2.1.2)
for EBSCO
the diffusion
Publishingconstant D, weCollection
: eBook Academic see that (when m >- printed
(EBSCOhost) 0) Nelson absorbs
on 6/8/2017 5:43the
PM mass
via m of the
quantum-mechanical particle JOSE
UNIVERSIDAD DISTRITAL FRANCISCO into,
DE say, the coefficient of viscosity in Einstein's formula.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 312 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(d) Recall from Chapter 6 (especially Equation (6.4.1)) that for a single quantum-
mechanical panicle of mass m moving in Rd, the Hamiltonian is given formally by

where A denotes the Laplacian in Rd. Note that the mass is involved with the Laplacian,
but not with the potential V. The reader may wish to keep this in mind when following
the rest of this section.
We add two further comments unrelated to the issue of scaling.
Remark 13.5.3 (a) We remind the reader that the concepts of analyticity of L ( L 2 ( R d ) ) -
valuedfunctions are equivalentfor the operator norm, strong operator and weak operator
topologies on L ( L 2 ( R d ) ) (see Remark 13.2.2(c)).
(b) What accounts for the fact that we will be able to establish the existence of the
analytic in mass operator-valued Feynman integral under more stringent requirements
in Chapters 15-18? Briefly, our goals in those chapters are different than in Chapters
10-13, and we will not attempt there to deal with highly singular "potentials".
Nelson's results
We will analytically continue in mass rather than in time. This change from Sections
13.2 and 13.3 will have both advantages and disadvantages (see Remarks 13.5.5(b) and
13.5.8(a) below).
It is the potential V on Rd that will vary from problem to problem and so it is on V
that we must place our hypothesis.
Hypothesis on V: There is a closed subset J of Rd of capacity 0 such that V is
continuous and real-valued on Rd\ J. (13.5.5)
The concept of (Newtonian) capacity will be defined (see formula (13.7.1) and Def-
inition 13.7.5) and briefly discussed in Section 13.7. For now, we just list the properties
of capacity that will be helpful to us in the present section. The capacity of a subset A
of Rd will be denoted Cap(A).
(13.5.6)
(1) Cap(W) = 0 implies that Leb.(AO = 0, but not conversely.
under U.S. or applicable copyright law.

(2) If Cap(N) = 0, then for every n > 0 (equivalently, for every diffusion constant
D > 0) and for every E E Rd\N, the path n - 1 / 2 x ( s ) + E, 0 < s < oo, misses N for
m-a.e. x [Nel, Theorem 5 and following paragraph]. This implies that the following is
true for any V satisfying (13.5.5): For every t > 0, n. > 0 and for every E E Rd\J, the
function V ( n - l / 2 x ( s ) + E), 0 < s < t, is continuous and real-valued for m-a.e. x €Cto.
(3) The only subset of R of capacity 0 is the empty set, so that this concept will not be
helpful to us for d = 1. On the other hand, every countable set has capacity 0 for d > 2.
Thus, for example, a potential V : R.d ->• M satisfying (13.5.5) can have arbitrarily strong
singularities independent of sign at oo and at any finite subset of Rd, d > 2. Larger sets
of singularities
EBSCO Publishingare possible
: eBook as dCollection
Academic increases. Any line- printed
(EBSCOhost) or lineonsegment
6/8/2017 has
5:43capacity
PM via 0 in Rd
for UNIVERSIDAD
d > 3. In DISTRITAL
fact, anyFRANCISCO
subset JOSE
of a DE(dCALDAS
— 2)-dimensional affine subspace has capacity 0 in
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 313
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Rd. Also, such facts are by no means limited to flat subsets; any compact Riemannian
submanifold of Rd with dimension less than or equal to d - 2 is a set of capacity 0 in
Rd. (See, e.g., [RauTay, Example 3, p. 35].)
Item (3) just above was included in order to give the reader a better feeling for the
implications of the theorems in this section as well as in Sections 13.6 and 13.7.
It will be convenient for us in this section and throughout Chapters 15-18 to restate
our formulas for the operators [ e - z H 0 : Re z > 0} (see Theorems 10.2.5-10.2.7 and
especially formulas (10.2.25), (10.2.26), and (10.2.33)) in a slightly different form. For
s > 0 and n E C+, we write

with the convention that e -0(H0/n) = /. A brief review of the basic facts about these
operators cast in the notation of (13.5.7) will be given in part £ of Section 15.2. The
reader may find it helpful to consult that material. We settle for recalling here that when
in the mean (see (10.2.33)).
We now make further comments which will help us with the proof of Theorem 13.5.4
below. We will assume that our potential V is time independent and satisfies (13.5.5).
It follows from (1) of (13.5.6) that V is R-valued Leb.-a.e. on Rd. Thus the operator
V = MV of multiplication by V is self-adjoint (see Proposition 10.1.3) and the operator
e~isV is unitary for all s e R.
In the notation of (13.5.1), it is the function F_iiv that will concern us. Note that for
y E Ct,

The formula (13.5.2) becomes in this case for any W E L 2 (R d )

under U.S. or applicable copyright law.

Taking advantage of Remark 13.5.2(a), we can rewrite (13.5.9) as

WeEBSCO
will give Nelson's results in three parts, beginning with the case n, > 0. (See Theorems
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
13.5.4, 13.5.6DISTRITAL
UNIVERSIDAD and 13.5.7.) The
FRANCISCO JOSEtheorems
DE CALDAS will be stated in full but not all the assertions
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 314 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

will be proved. However, the three results taken together will provide a complete and
rather detailed proof that under the assumption that V satisfies (13.5.5), the analytic in
mass operator-valued Feynman integral Kt-iqo (F-iv) exists in the sense of Definition
13.5.1' for Leb.-a.e. qo in M.
Theorem 13.5.4 (Imaginary mass) Let n > 0 be given and suppose that the potential
V satisfies (13.5.5). Then, for all t > 0 and W E L2(Rd), we have

where the limit is in the norm on L 2 (Rd). The sense of (13.5.11) is that the operators
(e~"(-H°^'le~l nv)n in £(L2(Rd)) converge in the strong operator topology to the oper-
ator K'^(F-iv), where the action of this last operator is given for all TJS in L 2 (Rd)
Leb.-a.e. by the right-hand side of (13.5.9).
Finally, for each A. > 0, the operators [K[(F-iv) : t > 0}form a (Co) contraction
semigroup on H = L2(Rd).
Proof Let A. > 0, t > 0 and E E Rd\ J be given. By (2) of (13.5.6) (which uses in an
essential way the fact that J has capacity 0), the function V ( n - 1 / 2 x ( s ) + E)> 0 < 5 < t,
is continuous and R-valued for m-a.e. x € C0. Thus the integral in the argument of the
exponential in (13.5.9) makes sense for m-a.e. x. Further, the integrand in (13.5.9) is
bounded by |W(n-1/2x (t) + E)|. By applying the Wiener integration formula in Theorem
3.3.5 and then the explicit formula for the free heat semigroup in Theorem 10.2.6, we
obtain

But the right-hand side of (13.5.12) belongs to L2(Rd) as a function of E (see (10.2.12))
and so is finite for Leb.-a.e. E. (In fact, more is true. The right-hand side of (13.5.12) is
the convolution of two L2-functions (see (10.2.26)) and so is easily seen to be bounded
and continuous on Rd.)
Because of the continuity of V(n-1/2x(s) + E) as a function of s for m-a.e. x, the
integral in the equality to follow can be regarded as a Riemann integral and so can be
written as a limit of Riemann sums for m-a.e. jc:
under U.S. or applicable copyright law.

Using (13.5.13) and the Lebesgue dominated convergence theorem with |W ( n - 1 / 2 x ( t ) +

E)| serving as a dominating integrable function, we obtain (for E E J)

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 315
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Now, the integrand on the right-hand side of (13.5.14) depends only on the value of the
Wiener path x at the n times ^ . ^ . • • - . ^ = ^ Hence the integral can be calculated using
Theorem 3.3.5, Wiener's integration formula. We have done closely related computations
earlier (see (7.6.1)-{7.6.9) as well as the explicit formula (10.2.26) for the free heat
semigroup) and so we will just use the result here. Doing so, (13.5.14) becomes for
E E J,

Since (by (1) of (13.5.6)) the set J of capacity 0 necessarily has Lebesgue measure 0,
(13.5.15) tells us that the sequence of Trotter-like products

each of which belongs to L2(Rd), converges Leb.-a.e. to the L2-function Ktn(F-iv)W

We want to show that

First, it is easy to deduce from (13.5.7) that for n > 0, t > 0 and any p E L2(Rd)

We now wish to show that for all integers n > 1,

We will write out the case n = 2. (The same ideas are involved in the proof for general n.)
under U.S. or applicable copyright law.

where the inequalities follow from (13.5.17).

Now from (13.5.18), we have |Wn - Wn| < 2e-t(H0/n) |W| and so

But Wn — Wni -> 0 for Leb.-a.e. E since both Wn and Wni converge Leb.-a.e. to
Ktn(F-iv)W. Thus by the Lebesgue dominated convergence theorem, we see that {W n }
is a Cauchy sequence in L 2 (R d ). Hence {W n } converges in L2-norm to some element in
L 2 (R d
EBSCO) Publishing
and, since we Academic
: eBook have seen that W
Collection n -> Ktn(F-
(EBSCOhost) iV )Won Leb.-a.e.,
- printed 6/8/2017 5:43(13.5.16)
PM via follows.
ButUNIVERSIDAD
(13.5.16)DISTRITAL
is just the desired
FRANCISCO JOSEequality
DE CALDAS (13.5.11) in different notation.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 316 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Since the operators e-s(H0/n), s > 0, N > 0, are all contractions (see (10,2.12)) and
since the operators of multiplication by e - i s V , s > 0, are clearly unitary, it follows that
the operators

are all contractions. Further, it is easy to show that the strong operator limit of contraction
operators is also a contraction. Therefore, it follows from (13.5.16) that the operators
Ktn(F-iv), t > 0, n > 0, are contractions.
Nelson goes on to show that Ktn(F_iv) is strong operator continuous as a function
of t for t E [0, +00) and that this family of operators satisfies the semigroup property

(In view of (13.5.11), the latter property is easy to establish.) Thus Theorem 13.5.4
follows. D
Remark 13.5.5 (a) Since {Ktn(F-iv) : t > 0} is a (Co) semigroup, it has an infinitesimal
generator. (Recall that much of Chapters 8 and 9 revolved in one way or another around
this concept.) Nelson proves that every function p in the domain of the generator has
finite gradient. More precise information is now known about this domain, as will be
discussed further on. (See formula (13.5.42b), Corollary 13.5.18 and Remark 13.5. J9(a)
below.)
(b) Some of the differences in techniques between analytically continuing in mass as
opposed to time have an elementary source. Let the potential V be R-valued and time
independent. In the method of analytic continuation in time, we start with the semigroups
H0
e-sH0 and e-sV and analytically continue through C+ to the unitary groups e~is and
isV
e~ . We begin with the pair of semigroups for diffusion (or heat) problems and end
with the pair of unitary groups for quantum-mechanical problems. In contrast, when we
analytically continue in mass, we begin with e~ s(H0 /n) (n > 0) and e~'sV and end with
g-isH0 and e-isV Analytically continuing in A. has no effect on e~isv. If we want to end
with e~is V , the right thing for (standard) quantum problems, e~isV has to be there to
begin with. Of course, what we start with is then not the right thing for diffusion problems.
This will not concern us much in this chapter as our focus is on the quantum-mechanical
setting. However, we will use analytic continuation in mass throughout Chapters 15-18,
under U.S. or applicable copyright law.

and in those chapters we will often wish to draw conclusions about both the diffusion
and quantum-mechanical situations. In order to cover the diffusion setting, we will have
to start with V purely imaginary so that —isV will be real.
The presence of e~isV with V real right from the start in the analytic continuation in
mass framework has advantages and disadvantages. The function e~isV—and thus also
the functional exp[—i ft0 V(x(s) +E)ds]—has constant absolute value 1, independently
of the sign of V and no matter how large | V \ becomes. This makes the limiting argu-
ments involving the Wiener integral in the proof of Theorem 13.5.4 rather elementary. In
particular, it is easy to find dominating integrable functions. Also, the special care with
the EBSCO
negative part of the potential that was needed in the analytic continuation in time
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
approach wasDISTRITAL
UNIVERSIDAD not needed above.
FRANCISCO OnCALDAS
JOSE DE the negative side, HO + iV is not self-adjoint and
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 317
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

this means that arguments that were based on the self-adjointness of H0 + V in Section
13.3 are not available here.
(c) The reader may wonder why Nelson did not simply apply the Trotter product
formula (Theorem 11.1.4) to deal with the limit of the functions Wn in (13.5.16). In
fact, this formula does not apply here under the present assumptions because in general,
HO + i V need not be "essentially m-accretive" on D(Ho) n D ( i V ) ; that is, —(Ho + iV)
need not generate a (Co) contraction semigroup.
We now proceed to the second case of Nelson's results, namely, A E C+. Theorem
13.5.6 below says, roughly, that Theorem 13.5.4 continues to hold for all A E C+. There
is a crucial difference, however; the limit on the right-hand side of (13.5.11) will again
be shown to exist but, in this case, will be used to define the operator on the left-hand side
of (13.5.11). No claim will be made that the formula (13.5.9), or equivalently (13.5.10),
holds for A E C+; in fact, there can be no such formula (in any straightforward sense) for
n E C+\(0, +00) since there is no countably additive, complex-valued Wiener measure
with such a variance parameter (see Section 4.6 and especially Theorem 4.6.1).
Theorem 13.5.6 (Complex mass) Let A E C+ and suppose that V satisfies (13.5.5).
Then, for all t > 0, the sequence (e~n ( H ° / n ) e~il/n) n from L ( L 2 ( R d ) ) converges in the
strong operator topology to an operator in £(L 2 (R d )) which we will denote K^(F-iv).
We write, for all W E L 2 (R d ),

where the limit in (13.5.19) is in the norm topology on L2(Rd). Further, the operator-
valued function Ktn(F-iv) of n is analytic in C+ and agrees with the operator Ktn(F-iv)
in (13.5.9) for n > 0. Hence Ktn(F- iv ) exists in the sense of Definition 13.5.1 (or 13.5.1')
for all n E C+.
Finally, for each n E C+, the operators {Ktn(F-iv) : t > 0} form a (Co) contraction
semigroup on H = L2(Rd).
Proof We will prove all but one of the claims made in the statement of the theorem.
Note to begin with that e~tH° is analytic in C+ as an operator-valued function of t.
This is a consequence of Theorem 13.3.1 in the simple case where V = 0. (The reader
should consult the proof as well as the statement of Theorem 13.3.1.) It follows that
under U.S. or applicable copyright law.

H
e -s(H 0 /n) is anaiytic as a function of A in C+. Thus the products (e~n( o/We-' ^ V)« are

analytic operator-valued functions of A throughout C+. Further, the operators e~« ( H °/n )
are all contractions for n E C+ (see (10.2.12)) and so the above-mentioned products are
contractions as well. Summarizing, the sequence of operator-valued functions of A,

is analytic in C+ and bounded in norm by 1. Further, we know from Theorem 13.5.4 that
for all n > 0, f n (A) -> Ktn(F-iv) in the strong operator topology as n -» oo. All but one
of the conclusions of our theorem now follow immediately or at least easily from the Vitali
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
convergence theoremFRANCISCO
UNIVERSIDAD DISTRITAL for operator-valued
JOSE DE CALDAS analytic functions [HilPh, Theorem 3.14.1,
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 318 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

p. 104] (see also Theorem 11.7.1 (i) and Remarks 11.7.2(b), (c) in Appendix 11.7 above
for the corresponding classical Vitali theorem).
The version of Vitali's theorem which we will apply says the following: Let {fn ( n ) } be
a sequence of analytic operator-valued functions defined on a domain D in the complex
plane and suppose that ||fn(n)|| < M for all n and all n E D. Finally, suppose that
limn-oo fn (n) exists in the strong operator topology for all A in a subset DO of D, where
DO has a limit point in D. Then limn_»oo fn (z) exists in the strong operator topology for
all n E D, uniformly on all compact subsets of D. Further, the limit, denoted by f(n),
is analytic in D and satisfies ||f(n)|| < M for all X 6 D.
For us, /„(>.) is given by (13.5.20), D = C+, M = 1 and D0 = (0, +00). The
existence of the limit (13.5.19) in the strong operator topology for all A, 6 C+ and the
analyticity of the limit function Ktn(F-iv) throughout C+ follow immediately as does
the existence of Ktn(F-iv) in the sense of Definition 13.5.1 for all n E C+. The inequality

also comes immediately from Vitali's theorem.

Now, by Theorem 13.5.4, we have the semigroup identity

for all n > 0. It then extends to all n e C+ by analytically continuing (for fixed s, r > 0)
all three expressions in this identity.
The only assertion from the theorem that we will not prove is the strong operator
continuity for all n E C+ of the semigroup {Ktn((F_iv) : 0 < t < 00}. O
We are now prepared to establish the existence in the sense of Definition 13.5.1' of
the analytic in mass operator-valued Feynman integral with parameter b0 = —iqo for
Leb.-a.e. qo 6 R. As before, we will not give a complete proof of the theorem to follow.
Our presentation of this third case, n purely imaginary, of Nelson's results is influenced
somewhat by the proof of Theorem 3 in [JoSkl].
[The reader may wish to review Remarks 13.5.2(b),(c) to see why X purely imaginary
corresponds to a real mass parameter.]
Theorem 13.5.7 (Real mass) We suppose that V satisfies (13.5.5). Then for Leb.-a.e.
qo in R., qo = 0, the analytic-in-mass operator-valued Feynman integral with parameter
under U.S. or applicable copyright law.

n0 = — iqo exists in the sense of Definition 13.5.1'; that is, for every t > 0 and a e
(0,7T/2), Ktn(F-iv) converges in the strong operator topology as n —> n0 through
the wedge Wa(no) to an operator in L(L 2 (R d )) which we will denote Ktn0(F-iv) or
Kt-iqo (F-iv). When this happens, we write, for every W E L 2 ( R d ) ,

and understand that A. can approach n0 = — iqo through any of the wedges Wa(A.o).
Finally, for every A-o — —iqo for which K'_. (F_,-v) exists, the operators
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
(K^jUNIVERSIDAD
(F-iv)DISTRITAL
: t > 0}form
FRANCISCOaJOSE
(Co)DE contraction
CALDAS semigroup on L2^).
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 319
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof Fix t > 0. One of the keys to the proof is the use of the Fatou-Privaloff theorem
[Ru2, Theorem 11.23, p. 244] which says (for the domain that concerns us) that a bounded
analytic function on C+ has a nontangential limit almost everywhere on the real axis.
Since Ktn[(F-iv) is analytic in C+ as a function of A. by Theorem 13.5.6, we know that
(K[(F-iv) W, W) is a scalar-valued analytic function on C+ for every W, p in L 2 (R d ).
Further, this function is bounded by ||W|| ||p|| since each of the operators Ktn(F-iv) is
a contraction. Hence, by the Fatou-Privaloff theorem, there exists a Lebesgue null set
N(W, p) of real numbers such that if qo E N ( W , p) then (Ktn(F-iv)W, p) converges as
n —> — iqo nontangentially through C+. (During the rest of the proof, limits as n - — iqo
are to be interpreted as taken nontangentially through C+.) Now the separability of
L2(Rd) and the boundedness of {Ktn((F- iV ) : n E C+} allows us to find a null set
N independent of W and p such that if qo E N, then (Ktn(F-iv)W, p) converges as
n -> -iqo, for all W, p E. L2(Rd). The completeness of £(L2(Rd)) in the weak operator
topology insures us that for each qo E N, there is an operator in L(L 2 (R d )) which we
shall denote Kt_. (F_iv) such that as n -> — iqo

It follows from (13.5.23) that the operators Kt_iqo (F_,iv), qo E N, are contractions
since all the operators Ktn(F-iv) are contractions and

We wish to show that for qo E N,

Ktn[(F-iV) -» Kli^F-iv) (strong operator topology), (13.5.24)

Now for A. € C+, Ktn(F-iv)W is weakly measurable in n since it is analytic in A.. Hence,
Kt-iq (F-iv)W is a weakly measurable function of q. But L 2 (Rd) is separable and so
Kt_iq (F-iv) W is strongly measurable [HilPh, Corollary 2, p. 73]. Then it also follows
that||Kt-iq(F-iv)WII is measurable as a function of q, by [HilPh, Theorem 3.5.2].
under U.S. or applicable copyright law.

Next, for each A = u - vi E C+, the L2(Rd)-valued function

is Bochner integrable (by [HilPh, Theorem 3.7.4]) since

Note thatPublishing
EBSCO the integrand
: eBook in the middle
Academic expression
Collection (EBSCOhost)of
- (13.5.27) is the Poisson
printed on 6/8/2017 kernel (with
5:43 PM via
parameters
UNIVERSIDADu DISTRITAL
and v) for the region
FRANCISCO JOSE DEC+ (see [Hil]). This integrand is nonnegative and has
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 320 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

L1(1K) norm 1 as a function of q. [The Poisson kernel and the Poisson representation
theorem (used below) are for C+ and its boundary. We used the (conformally equivalent)
Poisson kernel and Poisson representation theorem for the open unit disk and its boundary
intheproof of Theorem 11.7.1(ii).]
Now that we know from (13.5.27) that g(q) is Bochner integrable, we are ready to
establish the (operator-valued) Poisson formula

which relates the "values" of the function on the boundary to its values on C+. Since both
sides of (13.5.28) are in L 2 (R rf ), it suffices to show that we get equality when we take
the inner product of the two sides with an arbitrary p € L2. But closed linear operators
and so certainly bounded linear functionals can be taken inside Bochner integrals [HilPh,
Theorem 3.7.12 and following comment], and thus it suffices to show that

However, (13.5.29) follows immediately from the classical scalar-valued Poisson repre-
sentation formula [Hil, p. 455]. Hence, we have now proved (13.5.28).
Using (13.5.28) and a simple inequality for Bochner integrals [HilPh, Theorem 3.7.6],
we have for every X = u — vi e C+,

Now using the fact that || Kt_. (F-iv)W || is a bounded, real-valued measurable function
of q and [Hil, Lemma 19.2.1], we have for Leb.-a.e. qo E R,

Combining (13.5.30) and (13.5.31), we see that for Leb.-a.e. go.

under U.S. or applicable copyright law.

Enlarging if necessary the null set associated with (13.5.23) to reflect the step just above,
we can also insure that for qo E N,

as A = u — vi —> — iqo. But when weak convergence is present in a Hilbert space, one
always has the norm of the limit less than or equal to the lower limit of the norms [Kat8,
Eq. (1.26), p. 137]. Thus, in our case,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 321
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Combining (13.5.32) and (13.5.34), we see that for q0 $ N,

But this last equality and the weak convergence in (13.5.33) implies the convergence in
the norm of L 2 (R d ) (see [BkExH, Proposition 2.1.5, p. 43]) that we wished to establish
(see (13.5.24) and (13.5.22)).
This will finish the proof for us, but Nelson goes on to show that the exceptional set
N of qs can be chosen independently of t and that {K'_f (F_,-y) '• t > 0} is a (Co)
semigroup for every qo £ N. D
Remark 13.5.8 (a) The existence of K'^ (F-iv) only for Leb.-a.e. q in Theorem 13.5.7
is a serious weakness of Nelson's beautiful results. (Actually, all the approaches to the
Feynman integral discussed in this section and the next depend on the Fatou-Privaloff
theorem and so have this same shortcoming.) Given a particular A.Q = —iqo, one does
not know whether K'^ (F_/v) exists or not. Further, even if it does exist, the possibility
that a minute change in XQ may cause it not to exist is certainly not satisfying.
(b) Can the existence of K t _ . ( F - i v ) only for almost every (nonzero real) q be
used to perform the basic computation of quantum mechanics? Perhaps. We engage in
some speculation which might provide a way of avoiding the difficulties. Suppose, for
simplicity, that we are interested in a single quantum particle moving in R3 under the
influence of the potential V. Let Abe a measurable subset o/R3. The probability that
the particle is in A at time t is ordinarily calculated by

where W is the initial probability amplitude and q is the mass parameter. The trouble is
that for a fixed qo, we have no guarantee that the expression in (13.5.36) makes sense.
However, for any e > 0, it does make sense to average (13.5.36) in some way over the
interval (qo — e, qo + e). For example, we may consider

The expression in (13.5.37) seems to have some heuristic appeal especially if E is taken
under U.S. or applicable copyright law.

so small that q is not known with better than e-accuracy anyway.

Haugsby's result for time-dependent, complex-valued potentials

Nelson considers potentials V : Rd -> M which take finite real values except possibly on
a closed set J of capacity 0 (see (13.5.5)). It is such time independent R-valued potentials
that are appropriate for the standard problems of quantum mechanics. However, poten-
tials that are time-dependent or complex-valued are also of interest. Time-dependent
potentials arise naturally when the force which the potential represents varies with time.
OneEBSCO
might have, :for
Publishing example,
eBook V = V\ (EBSCOhost)
Academic Collection + V2, where Vi is
- printed on a time-dependent
6/8/2017 5:43 PM via potential
suchUNIVERSIDAD
that the DISTRITAL
associated forceJOSE
FRANCISCO is under the control of an experimenter. Complex-valued
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 322 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

potentials were discussed at some length in Section 11.6. We just mention here that such
potentials are used to provide phenomenological models for "open quantum systems"
such as decay problems where attention is restricted to the material that remains and the
decay products being emitted are ignored (see [Ex, BkExH] for example). Haugsby's
theorem permits potentials that are time-dependent and/or C-valued and can also have
the same kind of strong singularities allowed in Nelson's results.
We are now prepared to state the theorem of Haugsby [Hau].
Theorem 13.5.9 Let V : [0, t] x Rd -* C be Lebesgue measurable and let T be a
subset of Rd of capacity 0. Further let V satisfy the following:

(i) Re V(s, u) < Bfor all (s, u) € [0, t] x Rd, where B is a real constant.
(ii) V(s, u) is bounded on every compact set disjoint from [0, t ] x P.
(iii) V(s, u) is continuous almost everywhere on [0, t] x A, where A is an open subset
of Rd satisfying at least one of the conditions (c1) P c A or (c2) P c A and
A\A has Lebesgue measure 0.

Then, for F = Fv given by (13.5.1), Ktn(FV) exists for all X € C+ and Ktn0(F v ) =
Kt-iq0 (Fv) exists in the sense of Definition 13.5.1' for Leb.-a.e. qo € R.
[We remark that our present notation involving the "potential" V and the associ-
ated functional F = FV are somewhat different than in the rest of this section and in
Section 13.6.]
A check of the conclusions of Theorems 13.5.4, 13.5.6, and 13.5.7 will show that
Nelson's results include several assertions that have no counterpart in the theorem just
above. However, the following simple proposition shows that if attention is restricted to
the existence of Ktn[(Fv), n € C+, then Haugsby's theorem implies Nelson's theorem.
Proposition 13.5.10 If the potential V\ satisfies the hypothesis (13.5.5) of Nelson's
theorem, then the corresponding potential V(s, u) = —iV\(u) satisfies the hypotheses
of Theorem 13.5.9.
Proof The real part of V(s, u) — —iV\(u) is identically 0 and so (i) of Theorem 13.5.9
is certainly satisfied. The set of capacity 0, call it F here, in (13.5.5) is required to be
closed and V\ is required to be continuous on R d \F. Clearly then V(s, u) — —i V\ (u) is
under U.S. or applicable copyright law.

bounded on every compact set disjoint from [0, t] x F; that is, on every compact subset
of [0, t] x (R d \F). Thus (ii) is established.
If V\ (u) satisfies (13.5.5), then V(s, u) = -i V\ (w) is continuous on [0, t] x (R<V).
Hence condition (c\) of (iii) is satisfied where F = F = / and A = Rd. (Recall from
(1) of (13.5.6) that Cap(J) = 0 implies that J has Lebesgue measure 0.) D
Remark 13.5.11 (a) Hypothesis (i) of Theorem 13.5.9 seems not to be a serious restric-
tion from a physical point of view. In the most standard problems, Re V(s, u) =
Re[—i V\ (u)} — 0 and in dissipative systems, Re V(s, u) < 0.
(b) One should think of A as a "small" open set containing P in (c1) of (iii) or,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
alternatively, containing
UNIVERSIDAD DISTRITAL F inJOSE
FRANCISCO (c2).
DE From
CALDAS that perspective, the assumption of continuity
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 323
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

almost everywhere on [0, t] x Q, is considerably less restrictive than the assumption in

(13.5.5) of continuity (of the time independent potential V\) on E.d\J.
(c) Haugsby's proof in [Hau] is clearly inspired by that of Nelson in [Nel] in a
number of key places such as the use of (i) the Wiener integral, a true integral with
respect to a countably additive measure, for X > 0, as well as (ii) three powerful
theorems of complex analysis, the theorems of Vitali, Fatou-Privaloff, and the Poisson
representation theorem.
(d) The function F-w given by (13.5.8) has constant absolute value 1. This fact was
quite useful in the proof of Nelson's results, but no longer holds in Haugsby 's setting for
the function Fv(y) given by (13.5.1). This makes less difference in the proof than one
might think since we have from assumption (i) of Theorem 13.5.9,

(e) In the proof of Nelson's theorem we had, for every t > 0, A > 0 and f E Rd\ J,
the continuity of V ( n ~ l / 2 x ( s ) + E) as a function of s, 0 < 5 < t, for m-a.e. x E Ct0.
This does not hold in Haugsby's setting and this fact adds technical difficulties to the
first part (A > 0) of Haugsby's proof.
(f) The "potentials" treated in Chapters 15-18 will be allowed to be time-dependent
and/or complex-valued as in Haugsby's theorem, but they will not be permitted to have
spatial singularities. It should be added that these latter chapters focus on perturbation
series and Feynman 's operational calculus for noncommuting operators and have goals
which are quite different from those of Chapters 11-13.
Further extensions via a product formula for semigroups
Let W : Rd —> C be a suitable (measurable) complex-valued function. Further, let C be a
suitable m-accretive "realization" of the "Schrodinger operator" — 1 / 2 A + W ( y ) (y E Rd).
As we have seen in the first part of this section, a key step in the proof of Nelson's result
in [Nel] consisted in establishing the following Trotter-type product formula:

where { T ( t ) := e- t C : t > 0} denotes the (Co) contraction semigroup generated by

—C. The convergence in (13.5.38) is assumed to hold in the strong operator topology
under U.S. or applicable copyright law.

for all t > 0. (See Theorem 13.5.4 and its proof, with the obvious changes of notation;
in particular, with our previous notation, we have W = —i V, where V denotes the
"potential".)
We stress that (13.5.38) does not follow from the classical Trotter product for-
mula (Theorem 11.1.4) or from Chernoff's generalization (Theorem 11.1.8) because
the hypotheses of these theorems are clearly not satisfied under the hypotheses to be
made on W. See (13.5.39) below.
Recall that Nelson [Nel] established (13.5.38) under the assumption that Re W = 0
andEBSCO
W isPublishing
continuous on Rd\J, where J c Rd is a closed set of capacity 0, but without
: eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
anyUNIVERSIDAD
further restriction on theJOSE
DISTRITAL FRANCISCO behavior
DE CALDASof W near J. The proof of (13.5.38) given in
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 324 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

[Nel] (as well as that of Theorem 13.5.4 above) uses the Wiener path integral in an
essential manner and in fact, deduces the existence of the product formula in (13.5.38)
from that of a corresponding Feynman-Kac formula with potential W (assumed to be
purely imaginary, in [Nel]). We note that even though Nelson showed that the limiting
semigroup T(t) in (13.5.38) is a (Co) contraction semigroup, he did not precisely identify
the domain of its generator — C. (See Remark 13.5.5(a) above.)
In a remarkable paper [Kat7], Kato has extended the operator-theoretic aspects of
Nelson's results in several directions and has exactly determined the domain of the
generator -C. (See (13.5.40) below, as well as (13.5.42) in Remark 13.5.15(b).) His
extension—obtained by means of operator-theoretic methods and without the use of the
Wiener integral—had the following goals:
(i) Relax the conditions on the complex-valued potential W (with Re W < 0, say).
(ii) Allow for stronger singularities on the imaginary part of W (and hence on the
real part of the potential V, if we set W = —iV).
[In view of Lemma 13.5.14 below, this is achieved by working with a suitable
Schrodinger operator on the open set £2 — Rd\J, where J is as above.]
(iii) Allow for the free Hamiltonian HQ = —1/2Ato be replaced by a suitable sec-
ond order elliptic differential operator, with variable and possibly discontinuous
coefficients.
It should be noted that Kato did not apply his results to the Feynman integral, as we
will do further on in Theorem 13.5.16.
We will now discuss (essentially without proof) some of these extensions, and then
mention some of their later generalizations obtained in [BivPi, Bivl-3]. [The correspond-
ing (but more general) results for the analytic (in mass) modified Feynman integral—
obtained by Bivar-Weinholtz and the second author in [BivLa] via a suitable product
formula for imaginary resolvents—will be discussed in Section 13.6. The interested
reader may wish to consult that section; especially the proof of Theorem 13.6.7, to have
a more precise idea of the techniques used in proving the results below.]
Let A C Rd be an arbitrary open set and put H = L 2 (A). Denote by HO =-1/2A
the (normalized) Dirichlet Laplacian on A, acting in H. (Recall that we have already
considered this choice of free Hamiltonian towards the end of Section 11.6.) Further,
let W : A —»• C satisfy the following conditions. (These assumptions have later been
relaxed in [BivPi,Biv3], as will be briefly discussed at the end of this section.)
under U.S. or applicable copyright law.

Hypotheses on W:

and

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 325
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Define the (linear, unbounded) operator C in H = L2(A) by

with domain

Remark 13.5.12 In (13.5.40b), much as in our discussion of Kato's inequality in Section

11.2, —1/2Au+ Wu is considered a priori as a distribution in D'(A). Further, as in
Section 11.6, the Hilbert space H10 (ft) is defined as the closure in H1 (A) of D(A) =
Coo00(A). Here, the Sobolev space H 1 ( A ) (used in Lemma 13.5.14 below) denotes the
set of functions u in L 2 (A) with distributional gradient VM in L2(A); i.e. each of the d
components of Vu lies in L 2 (A). Recall that H1(A) is a Hilbert space when equipped
with the norm

(See, for example, [Ad, Bre2, Kes].)

We can now state Kato's main result [Kat7, Theorems I and II, p. 107]:
Theorem 13.5.13 Assume that W satisfies hypotheses (H1) and(Hi) statedin (13.5.39).
Then the operator C defined by (13.5.40) is m-accretive (i.e. — C is m-dissipative by
Remark 9.4.6(b)) and therefore, by Theorem 9.4.7, defines a (Co) contraction semigroup
(T(t) — e~tc : t > 0}. Further, this semigroup is given by the Trotter-like product
formula (13.5.38); namely, for all t > 0 and W E L 2 (A), we have

where the limit is in the norm on L2(A) and holds uniformly for t in bounded subsets of
[0, +00).
The proof of Theorem 13.5.13 makes essential use of Kato's distributional inequal-
ity (Theorem 11.2.3), as well as of operator-theoretic and semigroup methods. It also
involves a well known lemma of Stampacchia [Stam] concerning the Sobolev space
under U.S. or applicable copyright law.

H1 (ft), and a lemma of Kato [Kat7, Lemma 4.1, p. 112] which we have already used in
Section 11.3; see Lemma 11.3.3 above. We refer to Sections 3 and 4 of [Kat7] for the
complete proof.
Next, in order to connect the present situation to our earlier setting, we recall a very
useful potential-theoretic lemma [Kat7, Lemma 2.6, p. 106], that seems to have first been
established in this form in [HorLio]. (See also [RauTay, Lemma 2.1 and Proposition 2.2,
pp. 34 and 36] for a closely related statement; the latter result relies on Carleson's results
in [Carl].)
Lemma 13.5.14 Let J be a closed subset of Rd and set A = R d \J. Then Hl0 (A) =
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
H 1 (UNIVERSIDAD
R d ) if (and only FRANCISCO
DISTRITAL if) J hasJOSE
(Newtonian)
DE CALDAS capacity 0.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 326 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 13.5.15 (a) The property expressed by Lemma 13.5.14 can be thought of as
a potential-theoretic counterpart of the fact (used in the proof of Theorem 13.5.4) that
almost surely, Brownian paths do not "hit'' a set of capacity 0; see (2) of (13.5.6) for a
precise statement.
(b) It follows from (13.5.40) (where we have set W :— -i V) and Lemma 13.5.14 that
for each n > 0, the operator Cn in part (I) of Theorem 13.5.16 below is also defined by

with domain

Note that with our earlier notation in (13.5.40), we have C\ — C.

We can now deduce the following result from Theorem 13.5.13 and Lemma 13.5.14
(for part (I)), as well as from the methods of proof of Theorems 13.5.6 or 13.5.7 (for part
(II) or (III), respectively). We note that neither Theorem 13.5.16 nor Corollary 13.5.18
below were stated in [Kat7] because Kato was primarily interested in establishing the
product formula (13.5.38) and did not discuss its consequences for the analytic (in mass)
operator-valued Feynman integral.
Theorem 13.5.16 Let J C Rd be a closed set of capacity 0. Further, let V : Rd ->• C
be a measurable function such that V € Lrloc(Rd\J) (i.e. V € L r ( X ) for every compact
set X C R d \ J ) , where the exponent r is as in hypothesis (H2) of (13.5.39). Assume,
in addition, that V satisfies the "dissipativity condition" Im V < 0. [Of course, the
case where V is real-valued—and hence Im V = 0—is that of most interest in ordinary
quantum mechanics.]
Then:
(I) (Imaginary mass) Let n > 0 be given. Then, for every t > 0 and W 6 L2(Rd),
we have

where Cn is the m-accretive operator defined just as in (13.5.40), except with —iV
under U.S. or applicable copyright law.

instead of W and with — ^ instead of — j. (Equivalently, by Remark 13.5.15(b) above,

d is defined by (13.5.42).)
(II) (Complex mass) Let A. 6 C+. Then the limit on the right-hand side o/(13.5.43)
continues to hold and defines a (Co) contraction semigroup, denoted T^(t) (t > 0).
Further, for all t > 0, the function X M- 7\(r) is analytic from C+ to £(L2(Rd)).
(III) (Real mass) For Leb.-a.e. q^'« R> <?o 7= 0, the function X t->- 7X(t) has a (strong)
nontangential limit (in the sense of Definition 13.5.1') as n approaches the parameter
n0 = — iqo; that is, for every t > 0 and a E (0, n/2), Tn(t) converges in the strong
operator topology as n -»• n0 through the wedge W (no) to an operator in L(L2(Rd)),
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - aprinted on 6/8/2017 5:43 PM via
which we willDISTRITAL
UNIVERSIDAD still denote 7n0JOSE
FRANCISCO (t) or T-iqo(t).
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 327
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Finally, for every A.Q = —iqoforwhich T-iq0(t) exists, the operators {T-iq0(t) : t >
0}form a (Co) contraction semigroup on L 2 (R d ).
We will say that the "analytic in mass Feynman integral" associated with V exists
(see Remark 13.5.17(a) below) and denote it byF t , n , T p a n ( V ) ;so thatF t , n T p a n ( V )= Tn(t)
for all n € C+ and for Leb.-a.e. nonzero purely imaginary n.
Proof (I) In view of Lemma 13.5.14 and Remark 13.5.15(b), part (I) (with n = 1)
follows from Theorem 13.5.13, specialized to A := Rd\J and W := -iV;so that, in
particular, Im V = Re W < 0. (Naturally, the case where X > 0 is arbitrary amounts to
nothing more than replacing — y by ^ in the proof of Theorem 13.5.13.)
Note that by (1) of (13.5.6), J has Lebesgue measure 0 and hence H = L 2 (A) =
L (Rd).
2

(II) Part (II) follows from part (I) exactly as Theorem 13.5.6 was deduced from
Theorem 13.5.4, by analytic continuation via an application of Vitali's convergence
theorem (see Theorem 11.7.1(1) and Remarks 11.7.2(b),(c)).
Note that for all nonzero n E C+ and t > 0, we have||e-t(H0/n)||< 1 and ||e~ itv || <
1, because Im V < 0.
(III) Finally, part (III) follows from part (II) much in the same way as Theorem
13.5.7 was deduced from Theorem 13.5.6 by using, in particular, the classical Fatou-
Privaloff theorem [Ru2, p. 244] regarding the boundary value of a bounded analytic
function. D
Remark 13.5.17 (a) Let V : Rd -> C be a measurable function. Then, by analogy
with Definitions 13.5.1 and 13.5.1', the "analytic-in-mass Feynman integral via TPF"
associated with V is said to exist and is denoted byF t , n T p , a n ( V )provided that the following
conditions are satisfied: (i) The product formula for semigroups (13.5.43) holds for all
n > 0. (ii) It continues to hold for all n E C+ and defines Tn(t) E L(L2(Rd)) which
depends analytically on n € C+. (Hi) Finally, for Leb.-a.e. nonzero purely imaginary
n0, Tn(t) admits a nontangential limit (as in part (HI) of Theorem 13.5.16 and denoted
Tn0 (t)), as A, approaches n0 through C+. (The reader may wish to compare the definition
of Ft,n T P,an(.) with that of the "Feynman integral via TPF" given in Section 11.2; see
Definition 11.2.21.) We note that even though we use the term "TPF" in the above def-
inition, the product formula (13.5.43) does not follow from the classical Trotter product
formula (Theorem 11.1.4).
(b) Of course—except for the claim (madeinTheoreml3.5.4)that for each n > 0, the
under U.S. or applicable copyright law.

operator Tn (t) is given by the Wiener path integral Ktn(F-iv) in (13.5.9) (see (c) below)—
Nelson's original result (given in Theorems 13.5.4, 13.5.6 and 13.5.7 above) follows at
once from Theorem 13.5.16 by taking V real-valued and continuous on R d \J. Clearly,
in the case of time independent potentials and uniform partitions (along with minor
notational changes), the same is true of Haugsby's generalization [Hau] of Nelson's
result (see Theorem 13.5.9 and Proposition 13.5.10).
(c) Note that—unlike in Theorem 13.5.4—we did not claim in pan (I) of Theorem
13.5.16 above that for n > 0, Tn(t) is given by an extended Feynman-Kac formula, with
(complex) potential W := -iV, as in (13.5.9) and (13.5.11) with Tn(t) := Ktn(F- iV ).
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
(See, however,
UNIVERSIDAD Corollary
DISTRITAL 13.5.18
FRANCISCO below.)
JOSE DE CALDAS Actually, it would be interesting to determine
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 328 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

whether a counterpart of the extended " Feynman-Kac formula" (13.5.9) holds for all
X > 0 under the hypotheses made in (13.5.39)—and, more generally, under the assump-
tions made at the end of this section as well as in Section 13.6 below. [Perturbation
theorems and interchange of limits arguments of the type of [Lai2]—as discussed in
Section 11.5 (and recently extended to C-valued potentials in [LisM])—may be useful
to tackle this question; see Remark 11.6.5.]
(d) Strictly speaking, according to the above comment made in (c), we cannot refer
to Theorem 13.5.16 as establishing the existence of the analytic in mass operator-valued
Feynman integral (in the sense of Definition 13.5.1'). (Compare Theorem 13.5.7.) How-
ever, in view of the product formula (13.5.43)—which is shown in Theorem 13.5.16 to
holdfor all A. 6 C+ as well as for Leb.-a.e. A = —iq (q E R, q = 0)—it is natural in the
present context to extend Definition 13.5.1' and to think of Tn(t) as the counterpart of
Ktn(F-iv), the analytic-in-mass operator-valued Feynman integral associated with the
potential V.
(e) Note that in Theorems 13.5.4, 13.5.6 and 13.5.7 ([Nel]), the potential V was
assumed to be real-valued, whereas it is now allowed to be complex-valued provided it
satisfies (13.5.39), and particularly the "dissipativity condition" (H\). We will further
relax these conditions at the end of this section—and even more so, in Section 13.6 below
in the context of the analytic (in mass) modified Feynman integral [BivLa]. For now, we
simply note that, obviously, we can replace the dissipativity condition Im V < 0 by
Im V < B, for any real constant ft (as in Theorem 13.5.9). The only changes in the
statement of Theorem 13.5.16 are then that a suitable translate of C is m-accretive and
that the (Co) semigroup {Tn(t)}t>0 is no longer composed of contractions.
(f) Just as in Theorem 13.5.7, in part (HI) of Theorem 13.5.16, the semigroup
{T_iqo (t)}t>0 is in general not a group, let alone a unitary group, and can only exist
(in the case of very singular potentials) for Leb.-a.e. nonzero real qo- In particular, when
it exists, the m-accretive operator C_((?0 is usually not skew-adjoint—and hence the
operator |C_iqo is not self-adjoint. Nevertheless, it follows from semigroup theory (see
Theorem 9.1.14) that for p E D(C_iqo), W(t,.) := T-iqo(t)<p satisfies the evolution
equation
under U.S. or applicable copyright law.

with initial condition iff(0, .) = p; here, formally, —C-iqo W can be interpreted as

"—i(—A\l//2qo + V (x)W)". (It can be checked that—with this interpretation—equation
(13.5.44) also holds in the distributional sense.) Note that if qo is positive, it can be
thought of as the "mass" of the associated particle in the "formal realization" of the
"Schrodinger operator" — A/2go + V(x).
The following corollary of Theorem 13.5.16 specifies Nelson's result (discussed in
Theorems 13.5.4,13.5.6 and 13.5.7 above), as well as Theorem 13.5.16 itself; see Remark
13.5.19 below. It also shows the unity between the approach via a product formula for
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
semigroups
UNIVERSIDADand the analytic
DISTRITAL FRANCISCO(in
JOSEmass) Feynman integral.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS OPERATOR-VALUED FEYNMAN INTEGRALS 329
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Corollary 13.5.18 Suppose that V satisfies hypothesis (13.5.5) of Theorems 13.5.4,

13.5.6 and 13.5.7. Then the assumptions of Theorem 13.5.16 are also satisfied and for
each A E C+,

In particular, for every X > 0, the generator of the semigroup [K^(F-iv) '• t > 0} is
equalto —Cx, where C\ isthem-accretiveoperator given by (13.5.42a) and with domain
as in (13.5.42b). Further, still for A > 0, 7X(t) = e~tCx is given by the generalized
Feynman-Kac formula (13.5.9), with potential -iV. More precisely, for all k > 0 and
W E L 2 (Rd), Tn(t)W = e~tcnW is given by the Wiener path integral on the right-hand
side of (13.5.9).
Moreover, for Leb.-a.e. nonzero real qo, the analytic in mass operator-valued
Feynman integral (with parameter no = —iqo) %*_• (F-iv), exists (in the sense of
Definition 13.5.1') and coincides with TL,-iqo(t). More precisely, for a given nonzero qo
in R, Kt-iqo(F-iv) exists if and only if T-iqo(t) exists, and then we have Kt-iqo (F_iv) =
T-iqo(t).
In particular, under the above assumptions, the analytic-in-mass operator-valued
Feynman integral, Ktn(F-iv), and the operator Tn(t) coincide whenever they exist for
n E C+, n = 0.
Proof For n > 0, (13.5.42) holds because by Theorem 13.5.4 and part (I) of Theorem
13.5.16 both Ktn(F-iv) and Tn(t) are given by the same product formula (13.5.38), and
hence are the limit of the same sequence of operators. In view of Theorem 13.5.16 along
with Theorems 13.5.4, 13.5.6 and 13.5.7, the result then follows by the uniqueness of
the analytic continuation and of the boundary value of a bounded analytic function.
Note that the claim about the generator of {Ktn[(F_,- v )}t>o for A > 0 follows from part
(I) of Theorem 13.5.16, while that about the extended Feynman-Kac formula (13.5.9)
follows from Theorem 13.5.4. D
Remark 13.5.19 (a) Corollary 13.5.18 specifies Nelson's result in [Nel], especially
Theorem 13.5.4 above, because for A > 0 (that is, for "purely imaginary mass"
in the terminology of [Nel]), it identifies the infinitesimal generator —C\ of the
semigroup {Tn(y) = e~itCn : t > 0). In particular, in view of (13.5.42b),
under U.S. or applicable copyright law.

it provides a precise description—in the language of distributions or generalized

(Sobolev) functions—of the domain D(Cn) — D(—Cx) of this generator. Recall from
Remark 13.5.5(a) that according to [Nel], we only knew that D(Cn) c H1(R d );
that is, given u in D(Cx), then its distributional gradient VM belongs to L2(Rd)
(compare (13.5.42b)).
(b) Further, Corollary 13.5.18 supplements part (I) of Theorem 13.5.16 (or equiva-
lently, for A = 1, Theorem 13.5.13 above from [Kat7]) because for A > 0, it provides
a Wiener integral representation o f T ^ ( t ) W = e~tCn W for each W in L 2 (E rf ), at least
under the assumption made in [Nel]; see (13.5.10). This fact does not seem to have been
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
noted in [Kat7]
UNIVERSIDAD or in FRANCISCO
DISTRITAL later papers
JOSE DEon this subject.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 330 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Kato's results in [Kat7] have been extended in several directions:

(i) First, the study of Schrodinger operators with highly singular complex potentials
begun in [Kat7] has been pursued by Brezis and Kato in [BreKat], where such operators
-5 A + W—or rather, their m -accretive realization C in H = L 2 (A) := L 2 (A; C), with
Dirichlet boundary conditions—have been defined and their domain D(C) analyzed
under very general assumptions on the "potential" W : A —> C, with A an arbitrary
open subset of Rd. (As in the proof of Theorem 13.5.13 given in [Kat7], a version
of Kato's inequality, Theorem 11.2.3, also plays a crucial role in [BreKat].) We will
implicitly use these results (and further extensions obtained in [Bivl-3, BivPi, BivLa])
in the statement of Theorem 13.5.22 below, as well as in the next section where the
results of [BivLa] will be discussed.
(ii) Second, the product formula for semigroups (13.5.38) has been established under
progressively more general conditions on the potential W by Bivar-Weinholtz and Piraux
in ([BivPi], [Bivl,3]). (We note that no such product formula was derived in [BreKat].)
In particular, (13.5.38)—and, more precisely, the analogue of Theorem 13.5.13—has
been obtained for the C-valued function W. Here, we assume that —W = q' — p + iq,
with q', p, q : A -> R, q', p > 0, W E Llloc(A) and Re W = -q' + p < p; further,
we suppose that either p E L°°(A) or else that the following condition (13.5.46) holds:

and

Remark 13.5.20 (a) For simplicity, the reader may well wish to assume in the follow-
ing that p = 0, as in most applications. Then, in that case, condition (13.5.46b) is
unnecessary.
(b) When p = 0 (or, more generally, when p E L°°(&)), we must replace HO
by H0-p := HO + (-p) and -W by q' + iq in the left-hand side of (13.5.38) and
(13.5.41). Indeed, otherwise, the semigroup etw would consist of unbounded operators
under U.S. or applicable copyright law.

for all t > 0.

In [Bivl-3], a magnetic vector potential a — (a1,. . . , ad) € (L/oc(A; R))d with
divergence V.a E L2 loc(A) is also allowed, with suitable modifications. [In particular,
we must replace the free Hamiltonian H0 by H0,a = — i (V — ia)2 as defined (when
A = Rd) in Example 11.4.12 (see (11.4.17) with V = 0) and, at the end of Section
11.6, in (11.6.8) under more general conditions than here.] The case when a = 0 was
obtained earlier in [BivPi].
Remark 13.5.21 We will see in Section 13.6 that—according to the work in [BivLa]—
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
"the" naturalDISTRITAL
UNIVERSIDAD conditions bothJOSEonDEthe
FRANCISCO complex-valued scalar potential W (exactly the
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 331
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

same as in [BreKat]) and the vector potential a (for example, a in (L2loc(A))d ratherthan
in (L4loc(A)) d ) can be assumed provided we replace the semigroups by the corresponding
resolvents on the left-hand side of (13.5.38).
By specializing to the case when A = Rd \ J, where / is a closed set of capacity
0, and by letting W — -iV, with V : Rd -> C (and with p = 0, for simplicity),
we obtain the following extension of Theorems 13.5.13 and 13.5.16, which was not
explicitly stated in the abovementioned references [BivPi, Bivl-3]. (To obtain Theorem
13.5.22, we use—just as in deducing Theorem 13.5.16 from Theorem 13.5.13—Lemma
13.5.14 and the fact that H := L2(Rd) = L2 (Rd\J), since J has Lebesgue measure 0.)
Theorem 13.5.22 Let V : Rd -»• C be a measurable function and let J c Rd be a
closed set of capacity 0.
(a) (Scalar potential) Then the counterpart of Theorems 13.5.13 and 13.5.16 holds
under the weaker assumption that V E L1loc\ J) and Im V < 0. It follows that
Ft,nTP,an(V), the analytic in mass Feynman integral viaTPF associated with the potential
V (as defined in Remark 13.5.17(a) above), exists under these hypotheses for every
n E C+ and for Leb.-a.e. nonzero n0 = — iqo.
(b) (Scalar and magnetic vector potentials) In addition, we can replace the free
Hamiltonian HO = — 3 A by Ho,5 = — j(V — ia)2, where the "magnetic vector poten-
tial" a = (ai,..., ad) : Rd -» Rd satisfies both a € (L4loc(Rd \ J; R)) d and V • a E
L2loc(Rd\J). The domain of C (or more generally Cn, with n > 0) is then defined
appropriately, in the same way as in part (b) of Theorem 13.6.4 below.
We close this section by posing a problem which does not seem to have been addressed
in the literature. (A possible way of tackling it was suggested in Remark 13.5.17(c)
above.)
Problem 13.5.23 (Generalized Feynman-Kac formula with complex, singular poten-
tial) Determine for which complex-valued potentials V the "generalized Feynman-Kac
formula" (13.5.9) holds. Specifically, does it hold under the assumptions of Theorem
13.5.13 and more generally, of Theorem 13.5.22 (with a = 0)? In particular, does it hold
for V real-valued on Rd and V €Llloc(Rd \ J), with J as abovel
Looking ahead at Section 13.6, does it hold under the more general assumptions of
under U.S. or applicable copyright law.

Theorem 13.6.1 below"?

A reader familiar with the Brownian bridge stochastic process (see, for example,
[Si9, pp. 40-41 and §V. 15]) may try to answer the same questions as in Problem 13.5.23
in the case when a = 0; that is, in the presence of a magnetic field.

13.6 The analytic-in-mass modified Feynman integral

In this section, we discuss joint results of Bivar-Weinholtz and Lapidus—obtained in
[BivLa, §3] and concerning an extension of the "modified Feynman integral" F t M ( . )
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
([Lal,2, La6-13],
UNIVERSIDAD [BivLa,
DISTRITAL §1-2])
FRANCISCO discussed
JOSE DE CALDAS in Sections 11.4-11.6 of this book. (See, in
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 332 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

particular, Definition 11.4.4 where Ft M (-) was formally defined.) We will call here this
extension the "analytic-in-mass modified Feynman integral" and denote it by F t,n M,an (-)-
Recall that in Section 11.4 (based on the results of [Lall] presented in Section 11.3),
the existence of the "modified Feynman integral" F t M ( V ) was established under (essen-
tially) the most general assumptions for which the quantum-mechanical Hamiltonian
H = — j A + V = HO+V can be defined without ambiguity (and hence the asso-
ciated energy functional, 5 f ||Vu||2 + f V|u| 2 , is a bounded from below quadratic
form). Namely, V : Rd -> R is measurable, V = V+ - V_, with V+, V_ > 0,
V+ E Llloc(Rd) (or more generally, V+ is an arbitrary, nonnegative measurable func-
tion) and—most importantly from the present point of view—V_ is H0-form bounded
with bound less than 1 (for example, V_ E Kd(Rd) or, in particular, by Remark 11.4.3,
V_ € Lr(R d ) + Loo(Rd), with r > d/2 if d > 2, r > d/2 if d = 2 and r = 1 if
d = 1); see Theorem 11.4.2 and Corollary 11.4.5. Of course, under these conditions,
the Hamiltonian H is a self-adjoint operator and is bounded from below. In addition, in
Section 11.4 (Example 11.4.12), a magnetic vector potential a = ( a 1 , . . . , ad) was also
allowed under the minimal hypothesis that a belongs to (L2loc(Rd; R ) ) d . [Naturally, the
"free Hamiltonian" H0 must then be replaced by H0,2 = -1/2(V — ia)2 (called Aa in
Example 11.4.12); see (11.4.17) with V := 0.]
Moreover, recall that in Section 11.6—based on the results of [BivLa, § 1 and §2]
and motivated in part by a mathematical model of dissipative quantum systems—we
have allowed V to be complex-valued; in particular, the real part of V satisfied the
same assumptions as in Section 11.4 while its imaginary part satisfied the "dissipativity
condition" Irn V < 0; see Theorem 11.6.3. In addition, a vector potential a was also
allowed in Section 11.6 under the same hypothesis on a as in Section 11.4; see the end
of Section 11.6.

The purpose of the present section is twofold:

(i) Firstly, we discuss the consequences of the main results of [BivLa, §3] for the
"analytic in mass modified Feynman integral" F t , n M , a n ( V ) associated with a potential
V; see Theorems 13.6.1 and 13.6.4 below. Our hypotheses will be even less stringent
than those made in Sections 11.4 and 11.6. In particular, the negative part of Re V
will no longer be restricted significantly. Actually, our assumptions will be completely
symmetric in the positive and the negative parts of Re V. Specifically, Re V will be
assumed to be an arbitrary locally integrable function off a closed set of capacity 0
under U.S. or applicable copyright law.

in Rd,without any sign restriction, and to have arbitrarily strong singularities on the
"exceptional set" J or at infinity; see Theorem 13.6.4. It will follow that when V is
real-valued, as is the case in standard quantum mechanics, the aforementioned energy
functional, j / ||V«||2 + f V\u\2, can take arbitrarily large negative values and hence
is usually no longer bounded from below. The price to pay for this increased generality
will be —much as in Nelson's second approach to the Feynman integral discussed in the
previous section—that Ft,nM,an (V) will only exist for almost every value of the parameter A.
on the imaginary axis—and hence, for Leb.-a.e. value of the "mass" of the corresponding
quantum particle; see part (III) of Theorem 13.6.4 below. Further, as in the previous
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
section or in DISTRITAL
UNIVERSIDAD the study of complex
FRANCISCO JOSE DE potentials
CALDAS made in Section 11.6, the corresponding
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 333
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

time evolution will no longer be reversible. Mathematically, this is translated into the
fact that the associated (Co) semigroup Tn(t) = e~tCn(t > 0) is usually not a group
and hence that a suitable translate of Cn is an w-accretive operator but is no longer
skew-adjoint, even for A purely imaginary. Here again, a vector potential a will also
be allowed under significantly weaker hypotheses on a than in Sections 11.4 or 11.6.
Specifically, the components of a will be allowed to have arbitrarily strong singularities
on the exceptional set J or at infinity, and to be any locally square-integrable functions
off 7; see part (b) of Theorems 13.6.1 and 13.6.4.
(ii) Secondly, we represent the semigroup associated with a Schrodinger operator
with highly singular complex potential via a product formula for the resolvent of the
free Hamiltonian (or a suitable substitute thereof) and the imaginary resolvent of (the
multiplication operator by) V; see Theorem 13.6.7 below, from [BivLa, §3]. [Recall
that in Sections 11.4, where V was assumed to be real-valued, we have used instead
(with A = HO and B = V) a product formula for the imaginary resolvents of two
self-adjoint operators A and B obtained in [Lall]; see Theorem 11.3.1. Similarly, in
Section 11.6, where V was assumed to be complex-valued, we have used Theorem 11.6.1
(from [BivLa, §1] ) which extends from self-adjoint to normal operators the product
formula for imaginary resolvents established in Section 11.3.] We will be able to deal in
Theorem 13.6.7 with the same class of singular complex-valued scalar potentials V as
that considered by Brezis and Kato in [BreKat]. Further, as before, we will allow for a
singular magnetic vector potential a of the same type as in (i). More generally, we will
show (as in Section 11.6) that the "free Hamiltonian" can be replaced by a uniformly
elliptic, second order operator with variable and possibly discontinuous coefficients. A
special case of the product formula obtained in Theorem 13.6.7 yields Theorem 13.6.1
discussed in (i), and therefore establishes (when a := 0) the existence of Ft,nM,an( V), the
"analytic-in-rnass modified Feynman integral" associated with V.
The proof (and even the statement) of Theorem 13.6.7 is somewhat technical, and
therefore some readers may wish to omit it on a first reading. However, we do include
it here because—given the material discussed in Sections 11.6 and 11.3—it can be pre-
sented reasonably concisely. Moreover, it makes use of (a suitable version and extension
of) Kato's distributional inequality (Theorem 11.2.3), as well as of techniques from
the theory of elliptic partial differential equations not previously encountered in this
book. It also gives the flavor of the proof of Theorems 13.5.13 and 13.5.22 (regarding
under U.S. or applicable copyright law.

the existence of the "analytic in mass Feynman integral via TPF") that was omitted in
Section 13.5.
We will continue Section 13.6 by comparing the various approaches to the Feynman
integral via analytic continuation in mass considered in this and the previous section.
We will stress, in particular, the unity between these approaches, within the intersection
of their domains of validity; see Theorems 13.6.10 and 13.6.11. This material was not
included in [BivLa].
Finally, we will illustrate these results by discussing the case of highly singular central
potentials, and in particular, the inverse-square potential; see Examples 13.6.13 and
13.6.18. Even though
EBSCO Publishing such
: eBook potentials
Academic are(EBSCOhost)
Collection so singular that they
- printed possess5:43
on 6/8/2017 certain
PM viaunphysical
features (suchDISTRITAL
UNIVERSIDAD as unbounded
FRANCISCOfrom below
JOSE DE CALDAStotal energy), they present some very interesting
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 334 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

mathematical challenges (see, for example, [Nel, Case, R]) and also provide useful
phenomenological models for various situations in molecular chemistry and in quantum
physics. (See especially the survey article [FrLdSp] as well as the more recent physical
references [MarPari, PariZi, GupRaj, HenRajl,2].)

Existence of the analytic-in-mass modified Feynman integral

We first state in Theorem 13.6.1 a product formula for resolvents (one of which is
imaginary) that is the key to deduce the existence of F t,n M,an (V); see Theorem 13.6.4. As
was mentioned above, in view of Lemma 13.5.14, it is a special case of the main result
of [BivLa, §3], Theorem 13.6.7, which will be stated and established below; see Remark
13.6.9.
Theorem 13.6.1 Let J C. Rd be a closed set of (Newtonian) capacity 0 in Rd. Let
V : Rd ->• C be a measurable function such that V E Llloc (Rd\ J); i.e. V is integrable
on every compact subset of Rd\J. Further, assume that the "dissipativity condition "
Im V < 0 is satisfied. [The case where V is real-valued—and so Im V = 0—is that of
most interest in quantum mechanics.]
(a) (Scalar potential) Then there is a (Co) contraction semigroup ( T ( t ) = e~tc : t >
0} such that the following product formula for resolvents (one of which is imaginary)
holds: For all t > 0 and every W E L2(Rd), we have

where the limit is in the norm on L2(Rd) and holds uniformly for t in bounded subsets of
[0, +00). Here, H0 = — jA denotes the standard free Hamiltonian acting in L 2 (R d ).
Further, the m-accretive operator C is a suitable m-accretive realization in H =
L (R d ) of the Schwdinger operator -^A + V(x), x E Rd; more specifically, C is
2

given by Theorem 13.6.2 below.

(b) (Scalar and magnetic vector potentials) Assume, in addition, that a = ( a 1 , . . . ,ad)
is a measurable, vector-valued function on Rd belonging to the space(L2loc(Rd \J; R)) d .
Then the analogue of the product formula (13.6.1) holds provided we replace HO = — A
by H0, a = -1/2(V — ia)2, the free Hamiltonian with magnetic potential a. (Here, H0,a
is acting in L2(Rd) and is defined as in Example 11.4.12, by (11.4.17) with V := 0, or
under U.S. or applicable copyright law.

equivalently, in view of Lemma 13.5.14, as at the end of Section 11.6 on the open set
A := Rd \ J.) Moreover, the definition of C — C% is modified accordingly. (See Remark
13.6.9 and the discussion preceding the statement of Theorem 13.6.7.)
We next specify the m-accretive operator C occurring in Theorem 13.6.1 (a) or equiv-
alently, the m-dissipative operator —C which generates the (Co) contraction semigroup
( T ( t ) = e-tc : t > 0} in (13.6.1).
Define the (linear, unbounded) operator E in H. := L2(Rd) by

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 335
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

with domain

Theorem 13.6.2 Under the hypotheses of Theorem 13.6.l(a), the operator E defined
by (13.6.2) is closable in H = L 2 (Rd) and its closure E coincides with the m-accretive
operator C introduced in Theorem 13.6.1(a); in short, C = E.
Further, for u € D(C), we have u E Hl(Rd) and

where (., .)L2(Rd) denotes the inner product in L 2 (R d ).

Proof This follows from [BreKat, Theorem 3.1, p. 146] and from Lemma 13.5.14 applied
to the open set A := Rd \ /. D
Remark 13.6.3 In order to state Theorem 13.6.4 below, we will also need the following
variant of Theorem 13.6.2. Given n > 0, let En be defined by (13.6.2), except with — ^
instead of — j in (13.6.2a) and (13.6.2b). Then En is closable and Cn := En is an
m-accretive operator. (Observe that with the notation of Theorems 13.6.2 and 13.6.1,
we have E\ = E and C\ = C.) Naturally, the quadratic form (or "energy functional")
\ f \\Vu\\2 + f V\u\2 is then replaced by ^ / \\Vu\\2 + f V\u\2, so that the displayed
equation in Theorem 13.6.2 must be modified accordingly.
We are now ready to state (more explicitly than in [BivLa]) the main existence
theorem (Theorem 13.6.4) for Ft,nM,an (.)- It is the counterpart in our present context of
Theorems 13.5.16 and 13.5.22, which dealt with the existence of Ft,nM,an(.).
Note, however, that the hypotheses of Theorem 13.6.4 on the scalar potential V are
weaker than in Theorem 13.5.16 (where a was set equal to 0), while those on the vector
potential a are weaker than in Theorem 13.5.22, where a was assumed, in particular, to
be in (L4loc(Rd \ J))d rather than in the smaller and more natural space (L2oc(Rd \ J ) ) d .
[The former condition on a is related to essential self-adjointness results in [LeiSd] (see
Remark 11.4.13(a)) for Hamiltonians with magnetic vector potentials, whereas the latter
is that expected from the more general point of view of quadratic forms (see Example
under U.S. or applicable copyright law.

11.4.12).] Moreover, even though the hypotheses of Theorem 13.6.4 are exactly the same
as in Theorem 13.6.1 above, we will repeat them for clarity of exposition.
Theorem 13.6.4 (Existence of the analytic-in-mass modified Feynman integral, for a
highly singular complex potential with "unrestricted" real part) Let V : Rd —»• C
be a measurable function such that V E L1loc(Rd\J), where J C Rd is a closed
set of (Newtonian) capacity 0 in Rd. Assume further that V satisfies the "dissipativ-
ity condition" Im V < 0. [Recall that the case where V is real-valued—and hence
Im V = 0--corresponds to the situation of ordinary quantum mechanics. In that case,
V is allowed to be any real-valued locally integrable function off J, without any sign
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
restriction,
UNIVERSIDADand can have
DISTRITAL arbitrary
FRANCISCO JOSE DE singularities
CALDAS on the "exceptional set" J.]
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 336 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(a) (Scalar potential) Then:

(I) (Imaginary mass) Let A. > 0 be given. Then, for every t > 0 and W E L2(Rd),
we have

where Cn is the m-accretive operator defined just as in Theorem 13.6.2, except with
— 2^ instead of — y. (See Remark 13.6.3 above.) Here, H0 denotes the standard free
Hamiltonian acting in L 2 (Rd).
(II) (Complex mass) Let A. 6 C+. Then the limit on the right-hand side of (13.6.3)
continues to hold and defines a (Co) contraction semigroup, still denoted Tn(t) (t > 0).
Further, for all t > 0, the function n Tn(t) is analytic from C+ to L(L 2 (R d )).
(III) (Real mass) For Lebesgue almost every nonzero purely imaginary number no,
the operator-valued function n > Tn(t) has a (strong) nontangential limit (in the sense
of Definition 13.5.1'), when X approaches n0 through C+; that is, for every t > 0 and
a 6 (0, n/2), 7n(t) converges in the strong operator topology as n —>n0through the
wedge Wa(n0) to an operator in L ( L 2 ( R d ) ) , which we still denote Tn 0 (t).
We will say that the "analytic-in-mass modified Feynman integral" associated with
the (scalar) potential V exists (see Remark 13.6.5(a) below) and denote it by F t , n M , a n (V);
so that F t , n M , a n ( V ) = Tn(t) for all n E C+ and for Leb.-a.e. nonzero purely imaginary A..
Finally, for every nonzero purely imaginary number A.Q such that Tn 0 (t) exists, the
operators (T^Q(t) : t > 0} form a (Co) contraction semigroup on H = L 2 (R d ), with
infinitesimal generator denoted —Cn0 . Moreover, if we write A.Q = —imo (with mo real
and nonzero), then for every fixed <p e D(Cx0), W(t, •) := Tn(t)<p — e~ t c n 0 (p is the
unique solution (in the semigroup or in the distributional sense) of "the Schrodinger
equation" (with real "mass" mo)

with initial state W(0, •) = (p.

(b) (Scalar and magnetic vector potentials) Assume, in addition, that a = ( a 1 , . . . , ad)
under U.S. or applicable copyright law.

is a vector-valued function on Rd belonging to (L 2 l o c (Rd\J; R))d. Then the counterpart

of (I) through (III) holds, provided that, in particular, we replace the free Hamiltonian
H0 = — 5 A by the nonnegative self-adjoint operator H0,a — — j(V — ia)2 (acting in
L 2 (R d ) and defined as in Example 11.4.12 by (11.14.17) with V :— 0 or equivalently, in
view of Lemma 13.5.14, as at the end of Section 11.6 with A := Rd \ J). It follows that
the "analytic-in-mass modified Feynman integral", associated with the scalar potential
V and the magnetic vector potential a, exists under these hypotheses (for every n E C+
and for Leb.-a.e. nonzero n0 = —iqo).
Proof This follows from Theorem 13.6.1 (and Theorem 13.6.2) in the same way as
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
Theorem 13.5.16
UNIVERSIDAD followed
DISTRITAL from
FRANCISCO JOSETheorem
DE CALDAS 13.5.13.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYT1C-IN-MASS MODIFIED FEYNMAN INTEGRAL 337
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(I) More precisely, part (I) (with X = 1) follows from Theorems 13.6.1 and 13.6.2.
[Of course, for X > 0, it suffices to replace — y by — ^ in case (a), and more generally,
-j(V - ia)2 by -i(V - ia)2 in case (b).]
(II) Part (II) follows from part (I) exactly as Theorem 13.5.6 followed from Theorem
13.5.4 via analytic continuation by an application of an operator-valued version of Vitali's
convergence theorem (see Theorem 11.7.1(1) and Remarks 11.7.2(b),(c)). Note that for
all nonzero A. eC+and all f > 0,wehave||[/+f(#oA)r 1 || < 1 and||[/-MWr'|| < 1
because Im V < 0.
(III) Finally, part (III) is a consequence of part (II). It is obtained in much the same
way as Theorem 13.5.7 was deduced from Theorem 13.5.6, by using an operator-valued
version of the classical Fatou-Privaloff theorem [Ru2, Theorem 11.23, p. 244] for the
boundary value of a bounded analytic function. (We leave it as an exercise for the reader
to prove the statement regarding the distributional Schrodinger equation (13.6.4).) D
Remark 13.6.5 (a) Let V : Rd ->• C be a measurable function. Then ^^(V), the
"analytic-in-mass modified Feynman integral" associated with the potential V, is defined
exactly like F t,n M,an (^) in Remark 13.5.17(a) above, except with the product formula
for semigroups (13.5.43) replaced by (13.6.3). Hence, the product formula for resolv-
ents (13.6.3) must hold for all X > 0 (as in part (1) of Theorem 13.6,4); further, it
must continue to hold for all A e C+ and define Tn(t) e £(L2(Rd)), which depends
analytically on n e C+ (as in part (II)). Finally, for Leb.-a.e. nonzero purely imag-
inary n0, Tn(t) is required (as in part (III)) to have a nontangential limit as n — n0
through C+.
(b) In part (I) of Theorem 13.6.4, the fact thatfor fixed X > 0, {7x(0 = e~'c^ : t > 0}
is a (Co) contraction semigroup on L2(Wi) follows from Theorem 13.6.1 (and thus from
Theorem 13.6.7 below) since by Remark 13.6.3, — CA. is m-dissipative on L2(E.d) and
hence generates a (Co) contraction semigroup, by the Lumer-Phillips theorem reviewed
in Section 9.4. Moreover, the corresponding statement regarding {71(0 '• t > 0} in parts
(II) and (III) of Theorem 13.6.4 follows much as in [Nel], as was explained in the proof
of Theorems 13.5.6 and 13.5.7, respectively. (We note that a similar comment could be
made about Theorems 13.5.16 and 13.5.22.)

Product formula for resolvents: The case of imaginary mass

We will now precisely state and then establish the product formula for resolvents obtained
under U.S. or applicable copyright law.

in [BivLa, §3]. It corresponds to the (generalized) Schrodinger equation with singular

magnetic and complex scalar potentials (denoted a and V, respectively), as well as with
"imaginary mass" (in the above sense). When specialized in the obvious way, it yields
the key part of Theorem 13.6.4 above; namely, part (I) corresponding to the case of
"imaginary mass". (See Remark 13.6.9 below.)
At first, it will be convenient to contrast our hypotheses and notation with those
of Section 11.6 [BivLa, §1 and §2] which dealt with the "modified Feynman integral"
with a singular complex potential (as well as with a magnetic vector potential), defined
via a product formula for imaginary resolvents (Theorem 11.6.1 [BivLa, §1], extending
Theorem 11.3.1 [Lai
EBSCO Publishing : eBook1Academic
] from Collection
self-adjoint to normal
(EBSCOhost) operators).
- printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 338 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

With the notation of Section 11.6 (see especially Theorems 11.6.1 and 11.6.3), we
can take A = —iAi = i^ (where A denotes the Laplacian with Dirichlet boundary
conditions on an open subset Q of Rd) and B — V (the multiplication operator with the
scalar potential V — q+ — q- — iq'). But then, the hypotheses of Theorem 11.6.1 would
imply that ^_ is essentially bounded, since now A\ = 0. In this case, however, we can
modify the proof of Theorem 13.6.1 to obtain the analogous result by assuming only that
#_ belongs to L]1OC(S2); i.e. the real part of the potential can be any locally summable
function on £2. [As the reader will see in the proof of Theorem 13.6.7 below, a number
of very significant changes are required, the most interesting of which involve a suitable
version of Kato's distributional inequality in this context (Lemma 13.6.6) and the use
of results concerning the weak (or distributional) solutions of an elliptic equation. (For
notational simplicity, we consider here the case when A. = 1; otherwise, it suffices, of
course, to replace A by A."1 A, with X > 0.)]
We can also add a nonnegative imaginary part p to the potential, with some bound-
edness hypothesis relatively to — ^A, provided that, in the product formula, p appears
in the same resolvent factor as — 5 A; see the left-hand side of the displayed equation in
part (i) of Theorem 13.6.7 below. (Otherwise, the resolvent associated with the potential
would, in general, be unbounded.) Finally, as was done at the end of Section 11.6, we can
replace — j A by an arbitrary second order, uniformly elliptic operator L. Under these
general hypotheses, we can no longer define "A + B" as a form sum in a suitable way,
but we can consider an m-accretive realization of "i(A + B)", defined in [BreKat] for
the case A = '-£• and in [Bivl,3] for A = —iL.
We shall make the same assumptions on the operator L = L (a) as at the end of Section
11.6. Namely, let [bjk }d.k=l be a symmetric, uniformly elliptic matrix of real-valued Loo-
functions on A, and let a = ( a 1 , . . . , ad) e (L 2 O C (A; R)) d denote the magnetic vector
potential.
Then, formally, L is defined by equation (11.6.8):
under U.S. or applicable copyright law.

More precisely, L is defined as the maximal restriction of the operator L (viewed as

taking values in the space of distributions D'(£2)) defined just after equation (11.6.8).
[Assume that [bjk] = j/, where / denotes the identity matrix of order d. Then we
simply have L = L(a) = — j(V — ia)2 = //o,3> with Dirichlet boundary conditions
on A, much as in Example 11.4.12 where we had assumed that A = Rd. Of course,
if in addition, £2 = Rd and a = 6, then L = L(0) = -^A = HO, the standard free
Hamiltonian acting on L2(Rd). More generally, although this may seem less obvious to
some readers, it follows from Lemma 13.5.14 that if A = Rd\J, with J a closed set of
capacity 0 in Rd, then L = L(a) = -j(V - ia) acting in all of L2(Rd); in particular,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
L =UNIVERSIDAD
L(0) coincides
DISTRITALwith the free
FRANCISCO Hamiltonian
JOSE DE CALDAS HQ acting in L2(R.d).]
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 339
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Moreover, we will make the following hypotheses on the complex-valued scalar

potential, which are exactly the same as in [BreKat]:

where

q,p,q' are real-valued functions, with p, q'>Q, (13.6.6b)

and

when d > 2, for some arbitrarily small e > 0. (13.6.6d)

[Caution: In (13.6.6a), (13.6.6b) and in the following, the function denoted by q' is
not the derivative of q.]
We note that hypothesis (13.6.6c) can be replaced by the weaker assumption that
P ^loc^-* 's infinitesimally form bounded with respect to — ^ A; i.e. for every fixed
e

y > 0, there exists 8(y) > 0 such that

Since \\V\u\\\ < ||(V - ia)u\\, for all u e Q(L) (cf. [LeiSd] or [Biv3]), this hypothesis
implies that we can define the form sum L + (—p + 8) = L — p + 5 as a positive self-
adjoint operator with form domain Q(L), where we have set S = S(a) (in the notation
of (13.6.6c')) and where a denotes the ellipticity constant of {bj k } d j , k = 1 . (The operator
L — p + S is the maximal restriction to an operator on H := L 2 (Q) of L — p + S e
C(Q(L), Q(L)'); see [Biv3, II, §2].)
Then it is known that the operator iG such that
under U.S. or applicable copyright law.

is closable, its closure iG being such that iG+8 is m-accretive', see [BreKat] for the case
L = —5 A, and [Biv3] for the general case. In the adaptation of the proof in [BreKat]
to a general L, the following lemma —which is an operator-theoretic version ofKato's
inequality (Theorem 11.2.3), extended to generalized Schrodinger operators with scalar
as well as magnetic vector potentials—plays a crucial role, and it will also be used later
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
on in the proof
UNIVERSIDAD of Theorem
DISTRITAL FRANCISCO13.6.7.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 340 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Lemma 13.6.6 Under the above hypotheses, let g e L°°(£2) with a nonnegative real
pan, and let u e Ji = L2(tt). Then, for all K > S :

Proof of Lemma 13.6.6 This lemma is closely related to Lemma 6 of [LeiSd], the proof
of which can be easily adapted to this situation (see [Biv3, Lemma 2.1, p. 28]). We just
note that in [LeiSd], although g is (real-valued and) nonnegative and p = 0, g is only
assumed to be in Llloc(A). Since we suppose here that g E Loo(A), we do not need
to start with u in Loo(A) and hence the proof does not require the use of Lemma 4 in
[LeiSd]. D
The solution of the generalized Schrodinger equation with imaginary mass and sin-
gular complex scalar (or electric) potential V, along with singular magnetic vector
potential a,

is given by the semigroup e l'c, for which one has the following product formula
([BivLa, Theorem 2, p. 457]).
[For the reader's convenience, in the statement of Theorem 13.6.7 and in Remark
13.6.9 following its proof, we will explain in much greater detail than in [BivLa] how
to deduce Theorem 13.6.1—and thus also Theorem 13.6.4 establishing the existence of
the analytic in mass modified Feynman integral—from this product formula.]
Theorem 13.6.7 (Product formula for resolvents: the case of imaginary mass)
(i) Under the above hypotheses, we have as n —*• oo :

for all u e Ti. = L2(Q), uniformly in t on bounded subsets of [0, +00).

(U) Assume in addition that p = 0, so that (by hypotheses (13.6.6a) and (13.6.6b))
V satisfies the "dissipativity condition" Im V < 0. Then, when (by letting {bjk} = \1)
the uniformly elliptic operator L = L(a) is replaced by //o,3 — —\(^~ ia)2 and
in particular (when a = 0) by HQ = —\&—as is the case in Theorem 13.6.1—the
under U.S. or applicable copyright law.

conclusion of part (i) of Theorem 13.6.7 still holds, except that we can let S = 0 in the
above statement of the product formula.
Moreover, if we further specialize to the situation of Theorem 13.6.1, where V is
a complex-valued function on R1 which is locally integrable off a closed set J c Rd
of capacity 0, then the above product formula (with S — 0) yields the product formula
of Theorem 13.6.1; namely, formula (13.6.1) in the case when a = 0, as in Theorem
13.6. l(a), or its counterpart (with H0 replaced by H0,a = — ^(^ — ia)2) in the presence
of a magnetic field, as in Theorem 13.6. l(b). Consequently, Theorem 13.6.1 (and hence
also Theorem 13.6.4) follows from this special case of the present theorem. (See Remark
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
13.6.9 below,DISTRITAL
UNIVERSIDAD along with the comment
FRANCISCO preceding the statement of (l3.6.6a) above.)
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 341
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 13.6.8 (a) Since iG + 8 is m-accretive, the limiting semigroup appearing in

Theorem 13.6.7,

is well-defined as a (Co) semigroup of contractions on H = L2(A).

(b) We have adopted notation that facilitates the comparison of the statement of
Theorems 13.6.7 and 11.6.1.
(c) In the proof of Theorem 13.6.7 given below, we will also (in so far as possible)
use notation that is similar to that of Theorem 11.6.1 (as well as of Theorem 11.3.1).
Two notable exceptions are the following: (i) We denote by G the operator which used
to be called C in the proof of Theorems 11.6.1 and 11.3.1. (ii) In order to avoid possible
confusion with the "mass parameter", what used to be called X in the proof of Theorems
11.6.1 and 11.3.1 is now denoted K.
Proof of Theorem 13.6.7 We will prove part (i) of the theorem since as will be explained
in Remark 13.6.9, part (ii) is a corollary of Lemma 13.5.14, Theorem 13.6.2 and the proof
of part (i).
We will stress the new features of this proof [based—as was alluded to above—on a
suitable version in this context of Kato's distributional inequality (Lemma 13.6.6) and
on regularity properties of the weak solutions of elliptic partial differential equations]
and will omit the parts that are similar to the proof of Theorems 11.3.1 and 11.6.1.
Fix K > 0 and v € H = L 2 (fi). We shall use the notation of Theorem 11.6.1 with

As will be indicated at the end of the proof, a density argument enables us to assume
that v E L 2 (A) n L°°(A). For such a fixed v in L 2 (A) n Loo(A), we now show that

as t ->• 0+, weakly in H. Here, wt is defined as in the proof of Theorem 11.6.1 (just
above equation (11.6.3)), except with n replaced by K. [More precisely, as in the proof
of Theorem 11.3.1 or 11.6.1, we first show that the limit in (13.6.7) holds along a
sequence {tn}—> 0+, and then deduce from a compactness argument that (13.6.7) holds
as t ->• 0+.]
under U.S. or applicable copyright law.

Fix t > 0. We have

but now wt is the solution of the following elliptic equation:

with

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 342 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and

[Note that like their counterpart in the proof of Theorem 11.6.1, Rt, Iq', and Iq, are
bounded self-adjoint (i.e. Hermitian) and nonnegative operators on W.]
As in the proof of Theorem 11.6.1, we can easily bound

independently of t; and so we can justify that along a sequence tn -> 0+, w, converges
weakly in H to

with

In particular, q'w e £^(£2). (Recall that q' does not stand for the derivative of q.)
If we multiply (13.6.8) by a test function

and integrate on £2, we can pass to the limit on the right-hand side of the equation obtained
in this fashion, much as in the proof of Theorem 11.6.1. We point out the differences in
the argument. For the term involving Rt := RB,, one has to use

a consequence of the Lebesgue dominated convergence theorem and of (the counterpart

of) estimate (11.6.4), combined with the hypothesis V e £^(£2). For the last term,
under U.S. or applicable copyright law.

we observe that by an additional application of Lebesgue's dominated convergence

theorem, we have

either in L 1+£ (Q) or in L'(£2), according to the hypotheses on q or p (cf. (13.6.6d)).

To pass to the limit in this term, it is then sufficient to obtain a bound on w, either
in L f l + e ) (Supp tp) or in L°°(Supp (p), independently of t. (Here, Supp p denotes the
support
EBSCO of (p and (1
Publishing + e)'Academic
: eBook = 1 +Collection
e~' denotes the conjugate
(EBSCOhost) exponent
- printed on of 1PM+ via
6/8/2017 5:43 e.) This can
be achieved
UNIVERSIDADby using FRANCISCO
DISTRITAL Lemma JOSE 13.6.6 with g :— Rt + Ig> + ilqt.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 343
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

[Since we are now in a concrete situation, we can think of Rt, Iq>, and Iqi as mul-
tiplication operators by (essentially) bounded functions. With the usual identification
between a multiplication operator and the associated function, it is then immediate to
check from the definitions above that g belongs to L°°(£2).]
More precisely, a simple algebraic manipulation based on (13.6.8) enables us to apply
Lemma 13.6.6 with the above choice of g to obtain (since L = L(a)):

Thus, we deduce that

where ^ is the unique solution in Hl0 (A) of the elliptic equation

But now, hypothesis (13.6.6c') and an easy generalization of [BreKat, Theorem 2.3,
p. 143] ensure (by a "bootstrap argument") that

[In order to apply this extension of the above-mentioned theorem in [BreKat], we use here
the assumption that v belongs to L2(A) n L°°(fi).] Hence, in particular, w, is uniformly
bounded in L(1+£)'(£2), which allows us to conclude in the case when q e L110^e(fi).
In the alternative hypothesis p e Lloc e(£2), we can apply standard elliptic local
regularity (see, for example, [GigTr]) to equation (13.6.12) to conclude that W E Looloc(ft).
Hence w, is uniformly bounded in L°°(supp (p); so, once again we can pass to the limit
in (13.6.11), which guarantees, <p e CQQ (£2) being arbitrary, that

From (13.6.6a), (13.6.6c'), (13.6.10) and (13.6.13), we deduce that

under U.S. or applicable copyright law.

We can then pass to the limit in (13.6.8) in the sense of distributions, which gives for all
<p E Coo00(A):

where h denotes the quadratic form associated with the nonnegative self-adjoint operator
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
L (see the paragraph
UNIVERSIDAD following
DISTRITAL FRANCISCO JOSE equation
DE CALDAS (11.6.8)). [Here, (•, •} denotes the "duality
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 344 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

bracket" between the space of distributions, D'(A), and the space of "test functions",
2?(A) = Coo00(A).]
By definition of L (see the comments following equation (11.6.8)), we deduce from
(13.6.15) the following equality between distributions (i.e. in D'(A)):

(Recall from (13.6.6a) that V = q + i(p — q').) We thus have, in particular,

In view of the definition of the operator G given just before Lemma 13.6.6, equations
(13.6.14) and (13.6.16) now imply that

It thus follows that along a sequence {tn} of positive numbers tending to zero, wt tends
weakly (in H = L 2 (A)) to

[Note that since iG + 8 — i(G — iS) is w-accretive (see Remark 13.6.8(a) above), its
resolvent [K + i(G — iS)]~l is a well-defined bounded (contraction) operator on H, for
every k > 0.] By (weak) compactness, we deduce that wt tends weakly (in H) to w as
t —>• 0+, as claimed. Moreover, using the same argument as at the end of Step 1 of the
proof of Theorem 11.3.1, we can easily conclude that in fact, wt tends strongly (in Ti.)
to w as t —>• 0+; i.e. ||wt — w||2 —> 0, as t —>• 0+.
Recall that we have assumed so far that v belongs to L 2 (A) n L°°(J2). However,
the density of L 2 (A) n Loo(A) in L 2 (A) = H and the boundedness of the operators
involved now enable us to reach the same conclusion for an arbitrary v E L 2 (A), as
required. We thus obtain the exact counterpart of Step 1 of Theorem 11.3.1 (or 11.6.1).
The rest of the proof of part (i) of Theorem 13.6.7 can be carried out exactly as
in Step 2 of Theorem 11.3.1 above (from [Lall]). [Recall that this part of the proof
of Theorem 11.6.1 (from [BivLa]) was also identical with that of Step 2 of Theorem
under U.S. or applicable copyright law.

11.3.1.] D
Our next remark explains how to deduce part (ii) of Theorem 13.6.7 from part (i),
and consequently how to view Theorem 13.6.1 (and hence also Theorem 13.6.4) as a
corollary of Theorem 13.6.7. (This reduction relies on Lemma 13.5.14 and Theorem
13.6.2.) It may be helpful to begin by reviewing the comment preceding the statement
of hypothesis (13.6.6a).
Remark 13.6.9 First, assume that {bjk} —1/2Iin the definition of the operator L, where
I isEBSCO
the identity matrix, and let p = 0 in the definition of the potential V, as in part (ii) of
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
Theorem 13.6.7.
UNIVERSIDAD Then,FRANCISCO
DISTRITAL a brief JOSE
inspection
DE CALDASof the above proof of part (i) of Theorem 13.6.7
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 345
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

reveals that we can let S =0 throughout the proof, and hence also in the statement of the
product formula. Indeed, under the present hypotheses, we no longer need to introduce
the constant of uniform ellipticity of L and thus we can put S = 0 in the counterpart
of (13.6.6c'). (It is only the existence of a positive imaginary part p that forces us in
general to introduce a nonzero constant S on the left-hand side of the product formula
in Theorem 13.6.7.)
Next, assume in addition that the hypotheses of Theorem 13.6.1 are satisfied, as in the
secondpartofthestatementofTheoreml3.6.7(ii);sothatwehave V : Rd — C, Im V <
0, and V E L1loc (Rd\J), where J C Rd is a closed set of capacity 0. Then we can set A =
R d \J and apply Lemma 13.5.14 to deduce that L(a) = -1/2(V - ia)2 = H0,a (and, in
particular, L(0) = —1/2A= H0) acting in all of L 2 (Rd); see the last comment preceding
the statement of (13.6.6a). (Note that V can be written in the form V = q + i (—q')> with
q' > 0, as in hypotheses (13.6.6a) and (13.6.6b), because it satisfies the "dissipativity
condition" Im V = —q' < 0.) Consequently, in light of Theorem 13.6.2, we deduce that
Theorem 13.6.1 holds. In particular, when a = 0, as in part (a) of Theorem 13.6.1, the
product formula (13.6.1) holds, while when a ^ 0, as in part (b) of Theorem 13.6.1, the
counterpart of (13.6.1) holds with HQ replaced by HQJ = — j(V — ia)2.

Comparison with other analytic-in-mass Feynman integrals

We now discuss material not included in [BivLa] but which is helpful in clarifying further
the relationships between the different notions of analytic in mass (operator-valued)
Feynman integrals considered in this and the previous section.
As we have seen above, the analytic-in-mass modified Feynman integral is (so
far, at least) the most general approach in this context, especially if we allow for
singular C-valued scalar potentials V and/or for singular magnetic vector potentials
a. However, for our present purpose^, we will now focus our attention on R-valued
scalar potentials (as well as let 3 = 0) and make more restrictive assumptions on the
function V.
Our first comparison theorem supplements Corollary 13.5.18 and shows the unity—
in the intersection of their domains of validity—between the various approaches to the
Feynman integral considered in Sections 13.5 and 13.6. It follows from the results of
[Nel], [Kat7], and [BivLa, §3] discussed in those sections.
[Recall that the relevant definition of K t ( F - i v ) , with F-JV as in (13.5.8), is given
under U.S. or applicable copyright law.

in Definition 13.5.1', while that of Ft,P, an (V) (resp., F t , a n ( V ) ) is given in Theorem

13.5.16 and Remark 13.5.17(a) (resp., Theorem 13.6.4(a) and Remark 13.6.5(a)).]
Theorem 13.6.10 Let V be a real-valued function on Rd that is continuous off a closed
set of capacity 0 in Md. Then the analytic-in-mass operator-valued Feynman integral,
K t ( F - i v ) , associated with the functional F-iv, the analytic-in-mass Feynman integral
via TPF, Ft,p an (V), and the analytic-in-mass modified Feynman integral, Ft, an (V),
associated with the potential V, all exist and coincide.
More precisely, for all t > 0,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 346 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

for every A e C+ and for Lebesgue almost every nonzero purely imaginary A (say, for
Leb.-a.e. A — —iqo, with go e R and <?o 7^ 0). //ere, 7x(f) is defined as in Theorem
13.5.16. Moreover, the equality (13.6.17) holds for every nonzero purely imaginary A
such that one (and hence all) the operators in (13.6.17) exists. Also, for such fixed
A, {7x(0 : t > 0} is a (Co) contraction semigroup.
Finally, for A, > 0, each of the operators in (13.6.17) is also represented by a
"generalized Feynman-Kac formula"; that is, when applied to an arbitrary vector W in
L 2 (R d ), it is equal to the Wienerintegral onthe right-hand side of (13.5.9). Further, still
for A. > 0, we have T\(t) — e~t Ck for all t > 0, where Cx is the m-accretive operator
defined by (13.5.42) and, in particular, with domain D(C\) given by (13.5.42b).
Proof This follows from Corollary 13.5.18 (itself a consequence of Theorems 13.5.4,
13.5.6, 13.5.7 and 13.5.16) along with part (a) of Theorem 13.6.4. D
More generally, as our second comparison theorem shows, the analytic-in-mass
Feynman integral via TPF and the analytic-in-mass modified Feynman integral agree
with each other under minimal assumptions on V.
Theorem 13.6.11 Let V be a real-valued function on Rd that is locally integrable off a
closed set of capacity 0 in Rd. Then for all t >0,

for every A. e C+ and for Lebesgue almost every nonzero purely imaginary A,. Moreover,
the equality (13.6.18) holds for every nonzero purely imaginary A such that one (and
hence all) of the operators in (13.6.18) exists.
Proof This follows from Theorem 13.5.22(a) (from [BivPi, Biv3]) and Theorem
13.6.4(a) (from [BivLa, §3]). D
Remark 13.6.12 Note that, in Theorem 13.6.11, there is no longer the claim that
for A > 0, the generalized Feynman-Kac formula holds; see Remark 13.5.17(c) and
Problem 13.5.23. Instead, for A > 0, F^p ^(V) (resp., ^ X a n (V)) is defined as a limit
of Trotter-like products involving, when V is real-valued, a semigroup and a unitary
group (respectively, a resolvent and an imaginary resolvent); see equations (13.5.43)
and (13.6.3), respectively.
under U.S. or applicable copyright law.

Highly singular central potentials—the attractive inverse-square potential

This example suggests that, for highly attractive singular potentials, the solution of the Schrb'dinger
equation obtained by [analytic in mass] Feynman integrals as developed here is physically relevant
even though the operators may not be unitary. If the energy is not bounded below, the potential
may produce collisions as well as scattering.

Edward Nelson, 1964 [Nel, p. 338]

Because of the problems concerning the physical interpretation of attractive singular potentials,
physicists concluded for a long time that no significance could be given to any singular potential as
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
regards the singularity
UNIVERSIDAD DISTRITAL at the center
FRANCISCO JOSEof
DEforce.
CALDASFurthermore the mathematical difficulties made the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 347
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

problem of understanding singular potentials even more formidable. But it was finally realized that
many aspects of singular potentials are physically meaningful, and that certain of their properties
are relevant. In addition, mathematical techniques were developed for handling the calculational
problems of singular potentials.
W. M. Frank, D. J. Land and R. M. Spector, 1971 [FrLdSp, p. 37]

Predazzi and Regge ([PreReg], 1962) were among the first to realize the usefulness of singular
potentials as a formal "laboratory" for investigating many physical and mathematical ideas. They
argued that physical interactions among particles in the real world are very likely highly singular
in character. The study of regular potentials is unlikely to reflect this situation. Rather, the singular
potentials, with their strong repulsion and lack of analyticity in the coupling constant, would be
more likely to shed some reliable light on the physics of strong interactions. Others, as we shall
see in this section, have used this argument in one variation or another to justify the examination
of various features of singular potentials.

W. M. Frank, D. J. Land and R. M. Spector, 1971 [FrLdSp, p. 74]

Let us assume that d = 3, for simplicity. Physically, this corresponds to a single

nonrelativistic quantum particle moving under the influence of the scalar potential V.
(Analogous examples can be given in every dimension d > 3 as well as for multiparticle
systems.) Since a point has capacity 0 in R3 (see (3) of (13.5.6) above), in order for
Theorem 13.6.10 (resp., 13.6.11) to apply, a potential V can have an arbitrarily strong
singularity at the origin, say, provided V is continuous (resp., locally integrable) except
at the origin in R3.
We will now complete from our present perspective the study of highly singular
(attractive) central potentials ( V ( x ) = -fi/ra, r = \\x\\, x € R3\{0}) made earlier
in Example 11.4.7 from the point of view of the (standard) modified Feynman integral.
(See Example 13.6.13.) We will then focus our attention on the very interesting special
case of the attractive inverse-square potential (V(x) = —1/r 2 ). (See Example 13.6.18.)
Example 13.6.13 (Highly singular attractive central potentials) Let us assume, as in
Example 11.4.7, that V is an attractive inverse-power potential:
under U.S. or applicable copyright law.

where r := \\x\\ denotes the length of x e R3\{0} and a, ft are given positive constants.
Schematically, one can think of an electron rotating around (or spiraling down to)
the nucleus of an atom (located at the origin of R3). The parameter ft is then proportional
to the electric charge of the electron while a represents the strength of the interaction.
Clearly, the larger a, the stronger the attraction between the electron and the positively
charged proton of the nucleus, and hence the greater the singularity of the potential at
the origin.
Even though such situations are often described as nonphysical in the literature, they
are EBSCO
usedPublishing
to model: eBook
interesting physical problems. The survey article [FrLdSp] gives a
Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
lengthy discussion
UNIVERSIDAD of the
DISTRITAL role played
FRANCISCO JOSE DE by highly singular potentials in the physics literature.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 353
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where K2 := | + (j(j + 1) —ftmo)(K€C) andu(t, •) is required to be in L2(0, +00)

for each fixed time t. [Here, in the second equality of (13.6.30), we have used the fact
that 2m0V = -ftm0/r2, by (13.6.24).]
Case (1) above and (2) or (3) below corresponds, respectively, to K2 > 0, K2 = 0 or
K2 <0.
We can then consider the time independent differential equation corresponding to
(13.6.30), along with the associated indicial equation. (One recognizes a form of Bessel 's
equation with parameter K.) At least in the present case (I), the necessary argument is
standard and can be supplied easily. (For cases (1)-(3), see [Case, §11],[Mee], [Nel,
p. 338] or [R].)
(b) According to the discussion in (a), for a general azimuthal quantum number
(or spin) j E N, the "critical value" of ft is given by ftmo = \ + j(j + I) (instead
of ^). Hence, the "subcritical", "critical" or "supercritical" case (case (1), (2), or (3),
respectively) corresponds to ftmo being less than, equal to or greater than, respectively,
l + yO' + l)-
For notational simplicity, we assume implicitly throughout much of this discussion
that j — 0, so that ft = ^— is the critical value. [Alternatively, one does not lose any
essential information by working instead in the one-dimensional situation; namely, the
half-line (0, +00) with a suitable boundary condition at r — QJ The interested reader
should not have any difficulty in restoring the correct value of the parameter; see, for
example, Remark 13.6.22(a) below.
(2) For ft = 1/4 (or rather, for ftmo = 1/4), the potential energy is no longer
bounded from below in terms of the total energy [Nel, p. 338] and so the form
sum of HO and V is not defined; also, neither are F'M(V), J"(FV) nor FTP(V).
However, a suitable "cut-off regularization" (see [R, §4] and Remark 13.6.20 below)
provides a reasonable definition of the Schrodinger operator in this situation. By
[Nel] and [R], the resulting operator still gives rise to a unitary evolution and coin-
cides (for Leb.-a.e. value of mo) with the operator }C_,-mo defined in Theorem
13.6.14.
Remark 13.6.20 Briefly, the above regularization can be described as follows. Given
R > 0, one considers the cut-off potential defined by VK(X) = —ft/2r2 for r > R and
VK(X) = —ft/2R2for 0 < r < R. One then shows that in the limit when R -»• 0+, one
recovers the operator provided by the standard differential equation approach discussed
under U.S. or applicable copyright law.

in Remark 13.6.19 above; see [R, p. 545]. This regularization procedure also works (for
j ^ OJ in the previous case (1), but fails in the highly singular case (3) considered below;
see [R, %4] or [Mee, §57.
(3) We now come to the most interesting case from the present perspective; namely,
that when ft > 1/4 (or rather, ftmo > 1/4). Then, it is known that the above total energy
functional is unbounded from below; so that there exists a sequence of normalized L2-
eigenfunctions [pn}'^=^ (or bound states) with energy tending to — oc as n —> oo. As in
case (2), neither the form sum of HO and V nor the Friedrichs extension of — ^- + V is
EBSCO Publishing
defined; : eBook
also, neither Ft MAcademic
(V), JCollection (EBSCOhost)
i t ( F v ) , nor F t p ^ V-)printed
exists.onHowever,
6/8/2017 5:43 PM via
it has been shown
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 348 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Both attractive and repulsive potentials are considered there, with the emphasis on the
attractive case. Sections I, V and VI of [FrLdSp] will give the reader a quick introduction
to the ideas of that paper along with some of the many specific examples to be found in
[FrLdSp].
We note that the inverse-square potential (a = 2 and ft = 1) is discussed in physical
terms in [LL] and [Nel]. (The quote that begins this subsection is from [Nel].) This
potential will be discussed further in Example 13.6.18.
Of course, the standard interaction between an electron and a nucleus is that due to
an attractive Coulomb potential (a = 1 : V(x) — —ft/r, with ft > 0). However, more
complicated physical situations naturally involve highly singular central potentials of the
type described here (as well as other central potentials that can be dealt with by means
of the theorems mentioned above). As a first example, we mention that the electrostatic
interaction between a charge Z and an induced dipole of polarizability p is described by
the potential V(x) = -^Zpe2r~4, where e denotes the electron charge; see [FrLdSp,
p. 36]. (In contrast, recall that the electrostatic interaction between two charges Z\ and
Z2 is described by the classical Coulomb potential V(x) = Z1Z2e 2 r~ l .) Moreover, in
physical chemistry, intramolecular interactions are often described phenomenologically
by means of singular potentials; see, for example, [FrLdSp, pp. 37-38 and esp. § V. A] and
the references therein for a discussion of several models of interactions between (polar
and/or nonpolar) molecules, including the Lennard-Jones potential, V(x) — ar~12 —
br~6 (a,b > 0) and its generalizations. (Note that this potential can be written as the sum
of a repulsive and of an attractive potential.) See also [MarPari] and the references therein
for a recent discussion of singular potentials in connection with the study of long-range
interactions between polymers (or macromolecules), of interest in biophysics.

As we have seen in Example 11.4.7, in the usual situations considered in the mathe-
matical physics literature, the Hamiltonian is always bounded from below and thus we
can always use the (standard) modified Feynman integral FtM (V) defined in Section 11.4
or the analytic-in-time operator-valued Feynman integral from Sections 13.2 and 13.3.
Further, it is frequently the case that - ^ A -I- V is essentially self-adjoint, and hence
we often can also use the (standard) Feynman integral via TPF F t p ( V ) defined in
Section 11.2.
More precisely, in the present situation, if a < 3/2 (for example, for the Coulomb
potential where a = 1), then, by Corollary 13.4.2, all of Tt M (V), J i t ( F v ) and F t T P ( V )
under U.S. or applicable copyright law.

exist and coincide with e~it H , where the Hamiltonian H is the unique self-adjoint exten-
sion of — j A + V; see region (I) of Figure 11.4.1. (Recall that the symmetric operator
- j A + V, defined on C£°(R3\{0}), is essentially self-adjoint in that situation.)
On the other hand, for stronger interactions (region (II) of Figure 11.4.1)—
specifically,when3/2 < a < 2or(a =-2and/3 < 1/4)—only Ft M (V)and Jit(F v )can
be used since, as was discussed in Example 11.4.7, we do not know whether F t p ( V )
exists. Then (in both regions (I) and (II) of Figure 11.4.1), by Corollary 13.4.1, we have
Ft M (V) = Jit(Fv) = e-itH, where H = H0 + V denotes the form sum of H0 = -\ A
andEBSCO
V—which, in this case, coincides with the Friedrichs extension [Kat8, §VI.3] of the
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
symmetric operator
UNIVERSIDAD DISTRITAL— j A + V.
FRANCISCO JOSERecall from Example 11.4.7 that the energy functional
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 349
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

2 /R3 II^ M II 2 + /RS ^M2 is bounded from below in this situation. (We stress that in a
notation compatible with that introduced in Theorem 13.6.14 below, we should replace
4A by -2^A and 5 /R3 IIV«il 2 by ^ /R3 y v w f , form0 > 0.)
We now consider the only case not treated in Example 11.4.7; that of even more
singular potentials, for which none of F t M ( V ) , Jit(Fv) or Ft P (V) is defined. (See
Remark 13.6.15 below.) Namely, let us assume that either a > 2 or (a = 2 and ft > |);
see region (III) of Figure 11.4.1. Then, leaving aside the border-line case when (a = 2
and ft — |), the above-mentioned energy functional is no longer bounded from below.
This is so, in particular, for the attractive inverse-square potential (a = 2 and ft = 1), to
be considered in Example 13.6.18 below.
Since the central potential V is continuous away from the origin and—as was noted
earlier—a single point has capacity zero in R3, the following result is just a corollary
of Theorem 13.6.10 (as well as of Corollary 13.5.18 and Theorem 13.6.4 that led to it).
Hence we will not repeat the entire statement of Theorem 13.6.10 but we will recall some
of the main points and emphasize some new ones. Also, we will introduce a dimensional
(mass) parameter mo (but continue to set h = 1).
Theorem 13.6.14 (Highly singular attractive central potentials) Let the attractive cen-
tral potential V be given by (13.6.19), where a and ft are arbitrary positive constants.
Then the conclusion of Theorem 13.6.10 (and, a fortiori, of Theorem 13.6.11) holds
for this potential. In particular, the analytic-in-tnass operator-valued Feynman integral
K[(F-iv), the analytic-in-mass Feynman integral via TPF Ftp,an(V), and the analy'tic-
in-mass modified Feynman integral Ft,M (V) associated with V, all exist and coincide
(for every A. e C+ and for Leb.-a.e. nonzero purely imaginary A.):

Further, set X = —imo, where mo > 0 can be thought of as the "mass" of

the quantum particle. Then there exists a subset N C (0, +00) of (one-dimensional)
Lebesgue measure 0 such that for all fixed mo € R\N, the family of bounded operators
{T-imo(t) : t > 0} is a (Co) contraction semigroup on H = L2(R3), with generator an
m-dissipative operator denoted C-imo. (We write T-imo(t) — e~ imo for t > 0.)
Moreover, still for fixed mo 6 (0,+oo)\N, "the' formal "Schrodinger equation"
with potential V
under U.S. or applicable copyright law.

with initial condition W(0, •) = <p (<p e £>(C_;mo)), is solved at time t by T-tmo(t)<p =
e~'c~imo<p, where 7L;mo(0 is defined by (13.6.20). [Here, N has been chosen so that
for every <p E L2(R3), all the various analytic-in-mass Feynman integrals occurring
in (13.6.20) exist and (hence) coincide for X. :— —imo, with mo € (0, +oo)\N, when
applied to the "state" (p.]
Remark 13.6.15 (a) We stress that—in region (111) of Figure 11.4.1 (i.e. for a > 2
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
or for (a = DISTRITAL
UNIVERSIDAD 2 and ftFRANCISCO
> 1/4)—not only is the solution of "the" Schrodinger equation
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 350 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(13.6.21) provided by Theorem 13.6.14, but the "Schrodinger operator" with potential V
(and so the "Schrodinger equation" itself) is given meaning via the use of any one of the
above versions of the analytic-in-mass Feynman integral. (This fact will be illustrated in
a striking manner in Example 13.6.18 below.)
(b) Of course, as was already pointed out earlier (for instance, in Remark 13.5.8(a)),
a significant drawback of the method of analytic continuation in mass is that in general,
we do not know a priori for which values of the mass parameter mo the Schrodinger
equation (with mass mo) is defined, whereas a quantum particle (like an electron, for
example) has a well defined mass.
(c) As was also noted earlier, the "Schrodinger equation" (13.6.21) is in general no
longer time-reversible and the associated time-evolution is no longer unitary. Further,
if we want to define the Schrodinger equation (13.6.21) for t < 0 (rather than for
t > 0), we must first define K[(F-iv), .7>'p an^) and ^'M ^V^> f°r K < ° (rather than
for A. > 0), then analytically continue to the left half-plane Re A. < 0 (instead of the
right half-plane Re A. > OJ, and finally take the boundary value along the punctured
imaginary axis (Re A = 0, A. ^ OJ. (Recall that A. and mo are connected throughout by
the relation A = — imo-) It can be checked that for mo ^ (0, +oo)\N, we would then
obtain another (Co) contraction semigroup, {7}mo(0 : t > 0), the adjoint semigroup of
{T-imo(t) : t > 0} [inthe sense of'Theorem 9.6.14, except for Hilbert rather than Banach
adjoints (see Remark 9.6.15(b))], and thus with generator —C*_im , the (Hilbert) adjoint
of—C-imo. (An analogous comment applies to all the versions of the analytic-in-mass
Feynman integral discussed in Sections 13.5. and 13.6; see, for example, the comment
preceding Section 5 in [Nel, p. 336]for the case of K[(F-iv).)
Exercise 13.6.16 Let V be given by (13.6.19), as above, and let N be the Lebesgue null
subset o/(0, +00) defined in Theorem 13.6.14.
(a) (Regions (I) and (11) of Figure 11.4.1.) Assume that a < 2 or that (a = 2 and
ft < 1/4), so that by Example 11.4.7, the (standard) modified Feynman integral ^^(V)
exists for all values of mo > 0. Then show that for all positive mo E N,

It follows that in this case, the semigroup {T_,-mo(?) = e tc-'mo : t > 0} is actually
a unitary group and that Hmo := |C_,-mo is self-adjoint and equal to — ^- + V, the
form sum of—^- and V. Further, it follows from Corollary 13.4.1 that the operators in
under U.S. or applicable copyright law.

(13.6.22) are equal to the analytic-in-time operator-valued Feynman integral J i t ( F v ) .

(b) (Region (I) of Figure 11.4.1.) Assume in addition that a < 3/2, so that by
Example 11.4.7, the (standard) Feynman integral via T PF FTp(V) exists for all values
of mo > 0. Then show that for all positive mo £ N,

It follows that in this case, Hmo(= — ^- + V) = — ^- + V, the unique self-adjoint

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
extension of the
UNIVERSIDAD symmetric
DISTRITAL operator
FRANCISCO ~^- + V.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 351
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 13.6.17 (a) Of course, in (13.6.22) and (13.6.23), ^(V), J^M(V) and
J"(Fv) correspond to the positive mass parameter mo (that is, to the operator — ^-
instead of — y in Definitions 11.2.21 and 11.4.4, respectively). [Earlier, there was no
need to introduce a new notation because, as can be easily checked, each of these notions
exists either for every mo > 0 or else not at all.]
(b) (Repulsive central potentials) Some readers may wonder why we have not con-
sidered here the repulsive (rather than attractive) central potential V(x) = +ft/ra
(rather than —f}/ra), with a, ft > 0. In fact, in that case—as was discussed at the end
of Example 11.4.7—the (standard) modified Feynman integral exists for all values of
mo > 0 and for every pair (a, /?) with or, /8 > 0, since V is a positive potential. Fur-
ther, the same assertion holds for the analytic-in-time operator-valued Feynman integral
J''(Fv); see Theorem 13.3.1 and Remark 13.3.3(a). Hence, there is no need to consider
analytic-in-mass Feynman integrals in this situation. However, Theorem 13.6.10 can still
be applied, and it can be checked (much as in Exercise 13.6.16 above) that for Leb-a.e.
m0 € (0, +00), Ft,an(V) exists and coincides with both Ft M (V) and Jit(Fv).
We point out, however, that—even though they are mathematically easier to deal with
from our present perspective—such singular repulsive potentials are quite useful in var-
ious applications, particularly in molecular chemistry and aspects of phenomenological
quantum field theory; see, for example, [FdLdSp, §IV.A and §V], [MarPari], [PariZi],
and the relevant references therein.
Example 13.6.18 (Attractive inverse-square potential) Let us now specialize the discus-
sion in Example 13.6.13 by assuming that V is the attractive inverse-square potential
(a = 2) with positive parameter ft; thus

In addition to the properties common to all attractive inverse-power central potentials

considered in Example 13.6.13, this example possesses specific features of great interest,
as we will soon see, especially in Remark 13.6.19 (3) and Theorem 13.6.21 below.

(1) As we know from either Example 11.4.7 or Example 13.6.13, if ft < 1/4 (or
rather, since the positive mass parameter WQ is now taken into account, if fmo < 1 /4),
under U.S. or applicable copyright law.

the total energy functional ^- / IIv «ll 2 + / V|w| 2 is bounded from below and there
is a distinguished self-adjoint extension of the symmetric operator S := — j~—h V;
namely, the form sum H = ~^~ 4- V. As was discussed previously, the Hamiltonian
H—defined as a form sum (or equivalently here, as a Friedrichs extension of S)—is
the physically natural self-adjoint realization of S in this case, even though S is not
essentially self-adjoint. (Recall that this situation corresponds to region (II) of Figure
11.4.1.) Then, for all values of the mass parameter mo > 0, the Schrodinger equation

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 352 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

with initial state (p e D(H) c L 2 (K 3 ), is defined unambiguously and its unique solution
at time t e R is given by

where F'M(V) is the (standard) modified Feynman integral and J"(Fy) is the analytic-
in-time operator-valued Feynman integral associated with the potential V. (Recall that
FjpCV) is not known to exist in this case.)
Remark 13.6.19 (a) It may be helpful for some readers to provide a bit more techni-
cal information about the standard differential equation approach to such problems.
(For more details, see, for example, [Case], [Nel,§5], [R], [FrLdSp], and the relevant
references therein.) First, by passing to spherical coordinates, we write

where "®" denotes the tensor product of Hilbert spaces and where S2 = {x e
R3 : \\x\\ = 1} is the unit sphere in M3. Then, we have the "spherical harmonics
decomposition ":

where Bj is the jth eigenspace of the square of the "angular momentum operator" (which
has eigenvalues j (j + 1), for j = 0, 1 , . . . ) ; see, for example, [Han, Chapter 8], [LL,
§32 and §357, or [Nel, §57. (The integer j is often referred to as the azimuthal quantum
number or spin, in more modem terminology; for simplicity, we do not discuss here
half-integer spins, which would require enlarging the Hilbert space of wave functions.
In any case, in the following discussion, the specific value of j is irrelevant.)
Putting together (13.6.27) and (13.6.28), we obtain the orthogonal sum
decomposition:

where
under U.S. or applicable copyright law.

Next, after making the change of dependent variable u(r) = rWrad(r), where Wrad
denotes the radial component of a separable solution W of the (formal) Schrodinger
equation decomposed according to (13.6.29), we are led to solving (for each fixed j e
N = {0,1,...}):

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 354 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

by Case that for every fixed mo > 0, the symmetric operator

has a one-parameter family of self-adjoint extensions {Sp}peT, see [Case, §11]. [Here,
T = (Sl)®(= [0, 2;r)N) can be viewed as an infinite dimensional torus (while Sl is the
unit circle or 1-torus) and the parameter p = (Pj)jeli corresponds in each component
indexed by j e N to a choice of boundary condition at the singular point r = 0 [R, p. 545]
or to a phase factor PJ common to all the eigenfunctions of a particular self-adjoint
extension [Case, pp. 798-799]. (See also Meetz's work [Mee] which makes explicit use
in this situation of von Neumann's method [Sto, Theorem 10.20] for determining the
self-adjoint extensions of a symmetric operator.)] A remarkable fact also due to Case is
that these operators [Sp : p e T) constitute precisely all the self-adjoint extensions of 5
(see [Case, §11] or [R]). However, there is a priori no physically natural way to choose
between these various extensions. In the following, given p e T, we will denote by
{Up(t)}t£K the unitary group {Up(t) = e~"s» : t e R} associated with the self-adjoint
operator Sp.
On the other hand, we know from Theorem 13.6.14 that in this case, the various
forms of the analytic-in-mass Feynman integral exist and coincide for Leb.-a.e. mo > 0.
More precisely, there exists a Lebesgue null set N c (0, +00) such that for each fixed
mo e (0, +oo)\N, a suitable "Schrodinger operator" jC_/ mo is well defined and the
unique solution at time t > 0 of the corresponding "Schrodinger equation" (13.6.25)
with initial state p 6 D(C_,-imo) c L2(R3) is given by (13.6.26) (or, equivalently, by
(13.6.31) below). Recall that C_imo is an m-dissipative operator and hence generates a
contraction semigroup [e~ imo }(>Q, but that ]-C_j mo is not self-adjoint and hence does
not give rise to a unitary evolution.
To summarize the situation, "the most striking difference [between the above two
approaches] is that Case ([Case]) finds the [evolution] operators to be unitary but not
unique whereas Nelson ([Nel]) finds them to be unique but not unitary" [R, p. 544]. It is
then natural to wonder whether these two very different points of view can be reconciled.
In fact, in an intriguing work, Radin [R] has established the precise relationship between
them. In a nutshell, he has shown that "Nelson's nonunitary solution is a simple (time
independent) average over Case's family of unitary solutions " [R, p. 544]. Without going
under U.S. or applicable copyright law.

into the details, we will now give a somewhat more precise statement of Radin's main
results.
The following theorem is a corollary of Theorem 13.6.14 above and of [R, Proposi-
tion, p. 547].
Theorem 13.6.21 Fixmo in(0, +oo)\N such that m0 ft > 1/4, where N is the Lebesgue
null set given by Theorem 13.6.14. Then the nonunitary evolution T(t) — e~~'c-'mo (t >
0) on H = L2(R3) for the "Schrodinger equation"

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 355
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

defined in Theorem 13.6.14 ([Nel], [Kat7], [BivPi], [BivLa]) is a time-independent

average of the unitary evolution operators Up(t) = e~ltSf> obtained by the classical
method discussed by Case in [Case].
More specifically, there is a (time-independent, regular Borel) probability measure
P on T = (51)N such that for each fixed t > 0,

for all <p e H = L2(K3). (See Remark 13.6.22(a) below for a more precise statement.)
Recall that by Theorem 13.6.14,

for all < p e H , and that for (p e £>(C_,-mo) c H, T(t)<p is the unique solution w(t; •) of
(13.6.31) with initial condition T/r(0, •) = (p.
Remark 13.6.22 (a) In the notation and terminology introduced in Remark 13.6.19, a
more accurate statement of the result obtained in [R] can be given as follows: Fix the
azimuthal quantum number j e N(see[LL, §32]) and assume that ftm® > |+,/'(./+ !)•
Further, let (T(J}(t)} (resp., [U$(t)}) denote the restriction of {T(t)) (resp., (Up(t)})
to the eigenspace "Hj. Then there exists a probability measure P(j) on the unit circle (or
1-torus) T(j) := S1 such that the counterpart of equation (13.6.32) holds; namely,

for all (p e Hj (that is, physically, for all states of azimuthal quantum number j or
equivalently, of angular momentum ^/j(j + I)). [The construction of [R] is valid for
every value of ft > 0 (whether critical, subcritical or supercritical). Thus, it is not
necessary to specify the value of ft in Theorem 13.6.21 or in the above comments.
However, for ft subcritical (i.e. niQft < \ + j(j + l)j, all the self-adjoint extensions
{Sp}p coincide with each other and hence so do all the unitary groups {Up(-)}p; in that
case, it follows that, as expected, we recover the known conclusion T(t) = Up(t) and
thus T^(t) == U(£(t) for all t.]
under U.S. or applicable copyright law.

(b) Although it was first discovered in the present situation [R], the above "averaging
phenomenon " is not unique to the Schrodinger equation with inverse-square potential.
Indeed, an abstract counterpart of Equation (13.6.32) has later been obtained by Powers
and Rodin in [PowR] for a large class of boundary value and Cauchy problems. It makes
use, in particular, of Choquet's beautiful integral representation theory, a far-reaching
generalization of the Krein-Milman Theorem ([Cho3, Vol. II, Theorem 25.12, p. 105]) for
the extreme points of a weakly compact convex set in an infinite dimensional topological
vector space; see Chapter 6 in Volume II of [Cho3], as well as [ChoMey] and [Phe].
In the following difficult problem, the operator T(t) is defined by K'imo(Fiv) for
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
t >UNIVERSIDAD
0 and by DISTRITAL
K~'imo(F iV) forJOSE
FRANCISCO t <DE0,CALDAS
as indicated in Remark 13.6.15(c). We caution the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 356 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

reader that to our knowledge, parts of this problem are still open. (We are grateful to
Brian Jefferies for pointing out an inconsistency in the original formulation of Problem
13.6.23.)
Problem 13.6.23 Assume that V(x) = -fir^2, with ft > 0 and /3|mol > 1/4 (as in
case (3)).
(a) Show that the time evolution T(t) contracts distances in L 2 (R 3 ) but usually does
not preserve them. Conclude that (T(t) : t e R} cannot form a unitary group.
(b) Give a plausible physical interpretation for the results obtained in (a).
[Hint: With regard to questions (a) and (b), you may wish to consult [Nel]. We
caution the reader, however, that there may be more that one possible answer to
question (b) and that there does not seem to be a consensus among mathematical
physicists as to which one is "correct".]
(c) Investigate whether the analytic-in-mass Feynman integral of V can exist for all
(nonzero) real values of the mass parameter mo.
(d) Answer the analogue of (a)-(c) for the analytic-in-mass modified Feynman
integral.
(e) Finally, answer the counterpart of(a)-(d)for other attractive central potentials
(including those considered in Example 13.6.13), as well as for more general
highly singular potentials V (as discussed in Section 13.5 and in Section 13.6,
respectively).
We close this section by making a few additional comments of either a physical or
a mathematical nature. The last two of those are rather open-ended and speculative but
may point to some interesting future research directions in this area.
Remark 13.6.24 (a) Naturally, a careful physicist or theoretical chemist—when con-
fronted with the dilemma of choosing between the Hamiltonian approach ([Case, Mee])
or the analytic-in-mass Feynman integral approach discussed in the present example
(and the previous one)—will not decide on purely mathematical or aesthetical grounds.
Instead, the choice will be dictated by more pragmatic and physical considerations based
on the particular features of the physical problem at hand.
(b) Actually, there is a "third" approach to highly singular potentials more recently
discussed in the physics literature. It is based on renormalization techniques and seems
to strike a middle ground between the standard Hamiltonian approach (where "unique-
under U.S. or applicable copyright law.

ness" is lost) and the analytic-in-mass Feynman integral approaches (where unitarity of
the time evolution is lost). See the paper by Gupta and Rajeev [GupRaj] and the relevant
references therein, including [PariZi]; see also the papers by Henderson and Rajeev
[HenRajl, 2] for a closely related work. (The second author is grateful to Teaman Tiirgut
for pointing out these papers and for related conversations on the material of the present
remark and on that of its continuation in Section 13.7, Remark 13.7.18(d) below.) This
third approach begins with the observation that in the above-mentioned approaches,
the ground state energy is physically meaningless (or equal to — ooj. In a nutshell, it
consists in introducing a cut-off a > 0 (so that r = \\x\\ > a instead ofr > 0) and in
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
letting /8 = fi(a)
UNIVERSIDAD varyFRANCISCO
DISTRITAL with a JOSE
in the definition (13.6.24) of the inverse-square potential.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-MASS MODIFIED FEYNMAN INTEGRAL 357
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The suitable dependence of ft on a (or "renormalization" of the parameter ft) is then

determined by requiring that the ground state energy be finite and independent of a (as
well as by imposing Dirichlet boundary conditions on the sphere r = ||jt|| = a); this
amounts to implementing Wilson's renormalization scheme [Wits] in the context of quan-
tum mechanics [GupRaj]. For the moment, this intriguing approach is far from being as
mathematically developed as the other ones. It would be very interesting to investigate it
further and, if possible, to relate it with the approaches discussed in the present example.
(The authors of [GupRaj, HenRajl,2]—who are motivated in part by the study of the
strong force in quantum field theory, namely quantum chromodynamics (QCD)—do not
seem to be aware of Nelson's approach via the analytic-in-mass Feynman integral or of
its later extensions presented in Sections 13.5 and 13.6.) Along similar lines, it is natu-
ral to wonder whether the stochastic process constructed in [HenRajl,2] in association
with the Dirac measure in two dimensions (i.e., the potential V(x) = S(x) in R2) has a
suitable counterpart in the present more difficult case when V(x) = —l/r2 in R3. We
will return to this and related questions in Remark 13.7.18(d) below.
(c) To put it a bit dramatically, a quantum particle submitted to a very singular
attractive inverse-square potential (V(x) = —ft/r2, with ft > 1/4 as in case (3) above)
is confronted with the following rather unpleasant alternative: (i) Either to "commit
suicide " by "choosing " a nonunitary time evolution eventually leading to a collision with
the center of attraction [LL, Nel], if one of the analytic-in-mass Feynman approaches
is used, (ii) Or else to choose a privileged direction and hence "break the symmetry " of
the system, if the Hamiltonian approach is used (which consists in selecting a particular
self-adjoint extension HO + V, and hence a suitable boundary condition at the origin).
However, no such choice is necessary if the potential is less singular (namely, if ft < 1/4,
as in case (1) above) since then the total energy is bounded from below and so the natural
Hamiltonian Ho + V (the form sum of HO and V) can be used unambiguously. [Similarly,
when ft = 1/4 (as in case (2)), there is also a unique choice of Schrodinger operator.]
Note that the analytic-in-mass Feynman integral approaches have the advantage of
not breaking the symmetry of the quantum system (whatever the value of ft is) since by
Theorem 13.6.21 [R], they amount to a suitable averaging over all possible self-adjoint
extensions of HO + V. On the other hand, they lead to the eventual destruction of the
existing system.
It would be interesting to investigate the relationships between the present situation
under U.S. or applicable copyright law.

and the "spontaneous symmetry breaking phenomena" studied by Bost and Connes in
[BosCon 1,2] (see also [Con2, § V. 11 ]), in connection with the Riemann zeta function and
following in part earlier suggestions made by the physicist B. Julia in [Ju]. In the above
example of the inverse-square potential, the "spontaneous symmetry breaking" occurs
for values of ft greater than the critical value fto := 1/4; then, the natural rotational
symmetry of the quantum system (that is, the invariance under the orthogonal group
50(3)) is broken. Correspondingly, in [BosCon] (and with an appropriate choice of
notation), for ft < fto, there is a unique KMS^ state (and the associated factor is of type
IIIi), whereas for ft > fto, there is a continuous family of KMSp states, indexed by
the EBSCO
symmetry group
Publishing of the
: eBook system
Academic (and the
Collection associated
(EBSCOhost) factors
- printed are then
on 6/8/2017 5:43of
PMtype
via IQO). The
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
spontaneous symmetry breaking occurs at "low temperatures", corresponding to ft > /Jo-
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 358 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(We refer, for example, to [Con 1,2] for the terminology from the theory of operator
algebras adopted here. Physically, in quantum statistical mechanics, given ft > 0, the
"KMS0 states" can be thought of as the possible phases of the system at "temperature"
T = f ) ~ l , and a "phase transition" occurs at the critical temperature TO — fi^1.)
To summarize the above discussion, we conjecture (in a work in preparation by the
second author [La21]) that a "phase transition with spontaneous symmetry breaking"
(in the sense of [BosCon]) does occur at the "critical value" ft = PQ. In particular, for
ft > PQ (i.e. at low temperature), the singularity of the interacting potential V(x) =
—p/r2 is sufficiently strong to maintain the system in "thermodynamic equilibrium" in
an asymmetric state, whereas for ft < PQ (i-e. at high temperature), it is mild enough
for "the disorder associated with high temperatures [to cause] a unique homogeneous
phase" ([Lac, p. 332]). (Naturally, since the models studied in [BosCon] and in the
present example are rather different, the detailed structure of the KMS/j states cannot
be expected to be the same.) Further, partly in light of [PowR] (see Remark 13.6.22(b)
above), we expect that a similar conjecture can be made about many other highly singular
attractive potentials.
Pursuing this analogy one step further, one may then wonder whether, in the setting
of [BosCon], a suitable analytic continuation—as in the approaches via the analytic in
mass Feynman integral studied in Sections 13.5 and 13.6—would provide an appropri-
ate average of all the KMSy3 states when P > fa. In addition, one may ask what the
mathematical or physical meaning of this "average state" might be in this situation. (It
should be noted that, in general, this "average state" will no longer be a "state" in the
mathematical sense of the term used, for example, in the theory of operator algebras.)
It may also be worthwhile to investigate the possible connections between these
questions and aspects of the theory of "complex dimensions of fractals" and the study of
geometric, spectral and arithmetic zeta functions conducted by Lapidus and his collab-
orators, Carl Pomerance, Helmut Maier, Christina He, and Machiel van Frankenhuysen,
in [La22-26, LaPoml-3, LaMail,2, HeLal,2] and especially in [La-vFl-4]. See, in
particular, the comments regarding mathematical "phase transitions", "complex dimen-
sions" and the Riemann hypothesis, made in [La23, p. 176] and [La24, pp. 146-147,
esp. Question 2.6, p. 147], in reference to the work of [LaMail,2] which provides a geo-
metric characterization of the Riemann hypothesis in terms of a natural inverse spectral
problem for fractal strings. We note that recently, the latter work has been significantly
extended and put in a more conceptual framework in the research monograph [La-vF2]
under U.S. or applicable copyright law.

(announced in [La-vFl]). (See also [La27].)

We leave the possible investigation of these and related problems to future work.

13.7 The analytic-in-time operator-valued Feynman integral via

additive functionals of Brownian motion
Our goal in this section is to give the reader some idea about recent results of Albeverio,
Johnson and Ma [AlJoMa], even though the background information necessary for a
precise understanding has not been supplied in this book. Most of the missing mate-
rialEBSCO
involves the Wiener
Publishing (or Brownian
: eBook Academic Collectionmotion) process
(EBSCOhost) andon includes
- printed smooth
6/8/2017 5:43 PM via measures,
positive continuous
UNIVERSIDAD DISTRITALadditive
FRANCISCOfunctionals of Brownian motion (in the sense of Fukushima
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 359
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

[Fuk, FukOT]), a generalized Feynman-Kac formula, and the connection between these
three subjects.
Introductory remarks
The results in [AUoMa] are considerably more general than those in Section 13.3, as we
will see in Examples 13.7.17, 13.7.19, 13.7.20 and 13.7.23 below. Nevertheless, if we
restrict V_, the negative part of the potential V, to be in the Kato class Kj (see Section
10.4) and make a corresponding restriction in the setting of [AUoMa] to the "generalized
Kato class" GKj (see Definition 13.7.3 below), then there is a close parallel between
the two sets of ideas which should help the reader gain some insight into the more
recent work. We will begin by pointing out this parallel below—initially with minimal
explanation of the new concepts and facts. After that, we will discuss briefly several of
the ideas used in [AUoMa] and follow with the examples mentioned above which will
illustrate contrasts and connections between Section 13.3 and [AUoMa].
Finally, we will use a more general result from [AUoMa] and give a related example,
Example 13.7.23 below, where the positive part of a highly oscillatory and singular
potential helps to "control" the negative part. As indicated in [AlJoMa, p. 292], it is
likely that even more general existence theorems for the analytic (in time) operator-
valued Feynman integral can be obtained by using Theorem 3.4.4 of that paper. However,
we will not pursue that topic here.
The paper [AUoMa] is based on deep results of Albeverio and Ma [AlMal,2] and
Blanchard and Ma [BlMal,2], the influential book of Fukushima [Fuk], and ideas
of Johnson [Jo6]. While reading this section, one might find it helpful to consult
Chapter 3 (basic information about the Wiener process), Section 10.3 (quadratic forms),
Section 10.4 (the Kato class), Section 12.1 (the statement of the Feynman-Kac for-
mula) and Sections 13.2 and 13.3 on the analytic-in-time operator-valued Feynman
integral.
The reader will note the symbol C for the complex numbers in several places below.
We will not concern ourselves with this, but we mention that the results in [AlMal,2,
BIMal ,2, Fuk] are all obtained for Hilbert spaces over K; however, [AUoMa] was written
with quantum mechanics and the Feynman integral in mind, and so it was necessary to
have C as the scalar field. This required complexifying (roughly speaking) everything
in sight.
under U.S. or applicable copyright law.

The parallel with Section 13.3

We now begin drawing the parallel between the "objects" involved in our earlier results
in Sections 13.2 and 13.3 (numbers 1-6 below) and their replacements (numbers 1'-6').
The reader should keep in mind that in both 1 and 1', the conditions on the negative part
of the "potential" do not cover all that is known.
1. [Earlier potential] V or Vdu; V = V+ - V-, where V+ 6L1loc(Rd/G)and
V- 6 Kd- (Here, G is a closed subset of Md of Lebesgue measure 0.)
1'. [Replacement] /A, a "generalized signed measure"; ^ = /A+ — /u_, where /JL+ 6 5,
theEBSCO
classPublishing
of "smooth measures",
: eBook and /u—(EBSCOhost)
Academic Collection e GKj C- printed
S. (Here, /^ is not
on 6/8/2017 necessarily
5:43 PM via a true
signed measure
UNIVERSIDAD sinceFRANCISCO
DISTRITAL both ^+ JOSE
andDE ,u,_ are permitted to be infinite measures. A key part
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 360 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

of what is required to be a smooth measure is that 0 measure be assigned to all sets of

capacity 0.)
2. [Earlier additive functional] A v ( y ) = ft 0 V(y(s))ds = ftV+(y(s))ds -
/0' V-(y(s))ds, where y € C([0, +00), Rd).
[Note that the integral involving V+ is positive, possesses continuity properties as
a function of y (because of the convergence theorems of the Lebesgue theory), and is
additive as a function of t. Of course, the same observations hold for the integral involving
V-. The integrals involving V+ and V- are not the only "positive continuous additive
functionals" (PCAFs); see Definition 13.7.11 below as well as [Fuk, FukOT].]
2'. [Replacement] A? (y) = A?+ (y) - A?' (y), where Au+ and Au~ are the PCAFs
associated with the smooth measures u+ and u_, respectively.
3. [Earlier Hamiltonian] H^ = HQ+V, the form sum of HO = -\ A and V. Here,
H£ is the self-adjoint operator associated with the quadratic form £y = £c + q^. (The
quadratic forms £c and q^ have been used earlier and are denned in item 4 below.)
3'. [Replacement] H£ = HQ + M» the form sum of HQ and /z. Here, H£ is the self-
adjoint operator associated with the quadratic form €^ = £c +q^. (The quadratic from
qc is denned in (b) of 4' below.)
4. [Earlier quadratic forms] (a)£c(i/r, <p) :— \ fRd Vfi'V<pdu, the classical Dirichlet
form associated with — 5 A. Q(£c) = Hl(Rd, C), the Sobolev space of functions i/r e
L2(Rd, C) with distributional gradient Vi/r whose components are in L2(Rd, C). (Recall
that Q is used to denote the domain of a quadratic form.)
(b) q$W, <p) := /Rrf i/syVdu, with Q(q$) = L 2 (M d , C) n L 2 (E rf , \V\du, C).
4'. [Replacements] (a) No change.
(b) ^(i/r, ^) := /R</ tvrtdu), with Q(«C) = L 2 (R d , C) n L 2 (M d , |/i|, C), where
iMl =M+ + M--
5. [Earlier] Feynman-Kac formula:

where Ef denotes integration with respect to the standard Wiener measure m* associated
with continuous paths y which start at | e Rd at time 0.
under U.S. or applicable copyright law.

5'. [Replacement] extended Feynman-Kac formula:

This formula can be found in [BIMal, §4] under closely related hypotheses.
6. [Earlier] Theorem (briefly stated). The analy'tic-in-time operator-valued Feynman
integral J l t ( F y ) exists for every t € R and we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 361
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

6'. [Replacement] Theorem (briefly stated). The analytic-in-time operator-valued

Feynman integral Jit (Fu ) exists for every t € R and we have

Recall that the analytic-in-time operator-valued Feynman integral was defined by

starting with the expression on the right-hand side of the equation in 5 above and analyt-
ically continuing to an operator-valued function of t which is analytic in C+ and strongly
continuous in C+. (The precise definition is given in Section 13.2.) The same procedure
is used in the new situation except that one begins with the right-hand side of the equation
in 5'. (For more precision, see Definition 2.2.3 of [AlJoMa].)
We turn now to the definition of the terms used in l'-6' above, beginning with the
concept of a "generalized signed measure" (see 1'). Our definition here is equivalent
to—but different from—the one given in [AlJoMa, p. 271].

Generalized signed measures

Definition 13.7.1 Let fj,+ and /z_ be a-finite measures on B = B(Rd), the a-algebra of
Borel subsets o/Rd, and suppose that there exists B e B such that /^+ (Bc) = /u,_ (B) =
0, where Bc denotes the complement of B in R rf . Then /j. — /z+ — ^i_ is a generalized
signed measure on B.
Remark 13.7.2 (a) The set function n, is not necessarily defined on all ofB; it is defined
precisely for those E e B such that at least one of the inequalities At+(6 n E) <
oo, M_(0 C D E) < oo holds.
(b) The total variation |/LI| := ju+ + /^_ is a a-finite measure defined on all ofB.

The generalized Kato class

The Kato class discussed earlier (see Definition 10.4.1) consists of functions on R.d.
The "generalized Kato class" (see 1') is a straightforward extension of this concept to
measures.
Definition 13.7.3 A positive measure /u, on B(Rd) is said to be in GKj, the generalized
Kato class, if and only if
under U.S. or applicable copyright law.

Remark 13.7.4 (a)

EBSCO Publishing It isAcademic
: eBook not hard to show
Collection that /j, €- printed
(EBSCOhost) GKj implies that5:43
on 6/8/2017 ^ PM
is locally
via finite;
d
thatUNIVERSIDAD
is, n(K) DISTRITAL
< oo/or all compact
FRANCISCO JOSE DE subsets
CALDAS K ofR .
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 362 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(b) The Junction V on Rd belongs to Kj, the Kato class of functions on E.d, if and
only if the measure \ V \du belongs to GKj. Thus Kj may be regarded in a natural way as
a subset of GKd. However, Kd is not all of GKj, as we now show by giving an example
of a measure K which is in GK3 but which is not absolutely continuous with respect to
three-dimensional Lebesgue measure.
Let K be two-dimensional Lebesgue measure on the (u1, u2)-plane in R3 =
{u1, u2, u3) : uj e R for j = 1, 2, 3} and extend K to all of R3 by letting K be
the zero measure on the complement of the (u1, u2)-plane. Since the support of K is
the (u1, u2)-plane, a set of three-dimensional Lebesgue measure 0, K is certainly not
absolutely continuous with respect to three-dimensional Lebesgue measure. We claim
however that K e GK3.
Let a > 0 be fixed for now. Thus we can write:

where the next to last equality comes from changing to polar coordinates. Letting a —»
0+, we see from Definition 13.7.3 and the argument just above that K e GKj,, as claimed.

Capacity on W1
We used the concept of "capacity 0" in the preceding two sections even though we had
not yet given the definition of the (Newtonian) capacity of a subset A of Rd. We present
this definition now. First, for ^, <p in Q(£c) = Hl (Rd, C), we let
under U.S. or applicable copyright law.

where (•, •) denotes the inner product in L2(Rd, C).

Definition 13.7.5 Given an open subset U of Rd, we define Cap(U), the capacity of U,
by the formula

Then, for any subset A of Rd, we define the capacity of A as

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
Cap(A)
UNIVERSIDAD DISTRITAL :=JOSE
FRANCISCO inf{Cap(C7)
DE CALDAS : A C U with U open}. (13.7.3)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 363
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Certain facts about capacity that were needed earlier were stated in (1 )-(3) of (13.5.6).
The reader may wish to review those items at this point. Some properties of the set
function Cap(-) are given in the following theorem, the first three parts of which are
proved, for example, in [Fuk, Theorem 3.1.1].
Theorem 13.7.6 The set function Cap(-) is a Choquet capacity; that is:
(i) If A c B, then Cap(A) < Cap(B).
(ii) If {An}oo is an increasing sequence of subsets of Rd, then

(iii) If {An}oo is a decreasing sequence of compact subsets of Rd, then

Further:
(iv) If{An}^_^ is a sequence of subsets of W*, then

Remark 13.7.7 (a) For all dimensions d, if A is the surface of a ball of positive radius,
then Leb.(A) = 0 whereas Cap(A) > 0. Thus we see that strict inequality can hold in
(v) above. Also we see that Leb.(A) = 0 does not imply that Cap(A) = 0, a fact that
was already noted in (1) of (13.5.6).
(b) Physically, the Newtonian capacity of a set ("conductor") A C Rd corresponds
to its "electrostatic capacity". Further, in (13.7.1) or (13.7.2), £f(^, VO can be thought
under U.S. or applicable copyright law.

of as an "electrostatic energy". Naturally, the analogy with electrostatics has played a

key role in the development of potential theory and, in particular, of capacity theory.
(c) Further information about the theory of Choquet capacities can be found in
[Cho2] and in Section 9 of Chapter 3 of [Cho3, Vol. I]. The relationships between
the theory of (Newtonian or more generally, Choquet) capacities, potential theory and
Brownian motion are explored at length in Doob's treatise [Doo2].

Smooth measures
In each of the theorems in [AlJoMa] which insure the existence of the analytic-in-time
operator-valued
EBSCO Publishing Feynman integral
: eBook Academic J i t ( F(EBSCOhost)
Collection u ) , we require
- printedat
on least that5:43both
6/8/2017 parts of the
PM via
UNIVERSIDAD uDISTRITAL
"potential" = u+ FRANCISCO
— u- are JOSE "smooth
DE CALDAS measures". [In particular, the theorem in 6'
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 364 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

above assumes that u+ is smooth and that u- belongs to GKd, a subset of the set S of
smooth measures (see 1').] Thus, "smooth measures" are at the heart of the main results
of [AlJoMa].
Definition 13.7.8 A positive measure u on B ( R d ) is said to be smooth if and only if
C € B(R d ) and Cap(C) = 0 implies that u-(C) — 0 and if there exists an increasing
sequence {Kn}oo=l of compact sets (in general depending on /u) such that
(i) (i(Kn) < oo for n = 1 , 2 , . . . ,
and
(ii) limn_,.oo Czp(K\Kn) = 0, for all compact subsets K of Rd.
Remark 13.7.9 (a) The assertion that C e B(Rd) and Cap(C) = Q implies that ^(C) =
0 is often described by saying that fj, does not charge sets of capacity 0.
(b) Any locally finite measure /u. on B(R d ) which does not charge sets of capacity 0
is smooth; in fact, if we take Kn to be the closed ball of radius n centered at 0, this one
sequence {Kn} satisfies (i) and (ii) of Definition 13.7.8 for all such measures p,.
(c) A measure JJL which does not charge sets of capacity 0 may fail to be locally finite
and still be smooth. Example 13.7.19 gives a simple example of this (when d > 2 and
a > d). The measure there is smooth although it is not locally finite since u ( U ) = +00
for all nonempty open subsets U of Rd which contain the origin. Example 13.7.20 is a
more extreme illustration; the measure there is nowhere locally finite but is still smooth.
(d) Every fj, e GKj is smooth, as was already noted in 1'; i.e. GKd c S.
Definition 13.7.10 A generalized signed measure (Ji = n+ — /i_ (see Definition 13.7.1)
will be called a generalized signed smooth measure if and only if both /u,+ and ^~
are smooth measures. The family of all generalized signed smooth measures will be
denoted S — S.

Positive continuous additive functionals of Brownian motion

We turn now to the topic of "positive continuous additive functionals of Brownian
motion". We have not provided the background necessary for a thorough discussion
of this subject and its relationship with smooth measures, and so we will settle here for a
brief description of the most relevant facts. More information can be found in [AlJoMa,
Fuk, FukOT] as well as in [AlBlMa, AlMal,2, BlMal,2].
Let C([0, +00), Rd) denote the space of continuous, Rd-valued functions on the
under U.S. or applicable copyright law.

interval [0, +00). Given s e [0, +00), we define (much as in (3.3.2)) the operator

by

Now for any t e [0, +00), we let

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 365
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

that is, F0 is the smallest a-algebra of subsets of C([0, +00), Rd) containing all of
the sets P~l(B), where s varies over the interval [0, t] and B ranges over B(R d ). The
a -algebra F0 can also be described as the smallest a-algebra making all of the functions
Ps,0 < s <t, measurable. In similar fashion, we let

We have the following containments: For 0 < s < t < oo,

where C([0, +.00), Rd) is equipped with the topology of uniform convergence on com-
pact subsets of [0, +00) and, as before, B(C([0, +00), Rd)) denotes the Borel class of
C([0, +00), Rd). The first two containments in (13.7.7) follow immediately from the
definitions in (13.7.5) and (13.7.6), and the third follows from the fact that each of the
functions Ps in (13.7.4) is continuous.
We actually enlarge the a-algebras F0 and F° by a type of "universal completion"
that is appropriate to this setting. (See, for example, [AUoMa, p. 279].) These completed
a-algebras Ft and T, respectively, satisfy

We are now ready to define a positive continuous additive functional (in the sense
ofFukushima [Fuk]) of Brownian motion. Given f e Rd, we let m^ denote the Wiener
measure whose support is the set of y 6 C([0, +00), Rd) such that >>(0) = E. As
before, the expectation (or integral) with respect to the probability measure mE will be
denoted E|.
Definition 13.7.11 A function (or process) A : [0,+00) x C([0, +00), Rd) -+ R
is called a positive continuous additive functional (PCAF) if and only if At(-) is Tt-
measurable for each t > 0 and there exists A € F (called a defining set for A) and
N e B(Rd) (called an exceptional set for A) satisfying the following properties:
(i) Cap(N) = 0 and m(A) = 1 for all t- e Rd\N.
(ii) T,y e A for all y 6 A, where (Tty)(s) := y(t +s) for 0 < s < oo.
(iii) For each y e A, the function t-> At(y) is continuous, increasing, and vanishes
at 0 (i.e. A0(y) = 0); furthermore, it is additive in the sense that
under U.S. or applicable copyright law.

Remark 13.7.12 The measurability assumption that is made on A in Definition 13.7.11

is usually referred to in the literature on stochastic processes by saying that "(At) is
(f,)-adapted."
The relationship between smooth measures and PCAFs
There is, as we will see in Theorem 13.7.15 below, a one-to-one correspondence between
smooth
EBSCO measures
Publishing :and certain
eBook equivalence
Academic classes of- PCAFs
Collection (EBSCOhost) printed on(in the sense
6/8/2017 of via
5:43 PM Fukushima).
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
The next definition specifies the nature of the equivalence.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 366 ANALYTIC-IN-TIME OR-MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Definition 13.7.13 Two PCAFs A and B (of Brownian motion) are equivalent if and
only if they share a common defining set A on which they agree.
The correspondence between a smooth measure and its associated equivalence class
of PCAFs is given (in general) by the implicit equation (13.7.15) in Theorem 13.7.15
below. We try to give the reader some insight into this relationship by showing that the
formula holds for the familiar PCAFs of the form

where V is nonnegative, Borel measurable and (for simplicity) bounded. With Av as

just described, the inner integral on the left-hand side of (13.7.15) below satisfies (see
Problem 14, page 50 of [Fol2]) for y e C([0, +00), R rf ):

where/ : [0, +oo)xR d —* R is bounded, nonnegative and Borel measurable as required

in the statement of Theorem 13.7.15. By first using (13.7.11) and then the Fubini theorem
and the Wiener integration formula (Theorem 3.3.5), respectively in the third and fifth
equalities below, we can write
under U.S. or applicable copyright law.

where /z (or ^v) is given by the formula

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:43 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 367
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Hence, the relationship (13.7.15) below is established for all functions / as described
above in the special case where the additive functional and the measure u, are given,
respectively, by (13.7.10) and (13.7.13).
Remark 13.7.14 (a) Since V is bounded, it is clear that the measure /j, — Vdv in
(13.7.13) is locally finite. Hence, by Remark 13.7.9(b), to show that IJL is smooth, itsuffices
to show that fj. does not charge sets of capacity 0; that is, B e B(E.d) and Cap(fi) = 0
implies thatfj,(B) = 0. But this is true since Cap(fi) = 0 implies (see Theorem 13.7.6(v))
that Leb. (B) = 0, which implies in turn (by (13.7.13)) that /j.(B) = 0. Thus /z is smooth.
It is also easy to see that A^ is a PCAF under the present assumptions on V.
Indeed, taking the defining set A and the exceptional set N from Definition 13.7.11
to be C([0, +00), Rd) and 0 (the empty set), respectively, the reader can easily verify
this.
(b) The relationship between V and /u that is established in (13.7.12), namely,

actually holds for an arbitrary nonnegative Borel measurable function V : Rd —»•

[0, +00] and an associated measure given by (13.7.13). In fact, the argument changes
very little. Equation (13.7.11) continues to hold and the Tonelli theorem (seepage 65 of
[Fol2]) rather than the Fubini theorem needs to be used to justify the third equality in
(13.7.12). However, we do not claim for Vs this general that Avt given by (13.7.10) is
necessarily a PCAF nor that n given by (13.7.13) is necessarily a smooth measure.
We are now ready to state the general result giving the connection between smooth
measures and positive continuous additive functionals.
Theorem 13.7.15 For each n e 5, there is a PCAF AM with exceptional set N such
that the formula

holds for every bounded, nonnegative Borel measurable function f on [0, +00) x Rd
and for every f e R d \N. Moreover, A.^ is unique in the sense that if another PCAF B
under U.S. or applicable copyright law.

satisfies (13.7.15), then A^ and B are equivalent (see Definition 13.7.13).

Finally, given any PCAF A, there is a measure jj.eS such that A = AM.
The analytic-in-time operator-valued Feynman integral exists for
V = M+ - n- e S - GKd
Now we restate (with a little more detail) the theorem from 6' which assures us of the
existence of the analytic-in-time operator-valued Feynman integral Jl'(F^) for certain
generalized signed measures /n = /j>+ — /z_.
Theorem 13.7.16 Let n = /z+ — /^_ be a generalized signed measure belonging
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
to SUNIVERSIDAD
- GKdDISTRITAL
(i-e. FRANCISCO
M+ € JOSE 5 and /z_ e GKd), and let F M (v) = e~A?^\ where
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 368 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A? = A?+—A?~ and A?+ and A?~ are the PCAFs associated with the smooth measures
jU,+ and /n_, respectively. Then the analytic-in-time operator-valued Feynman integral
J"(Fn) exists and we have

for all t e R.

Examples
Near the end of this section, we will make brief comments about results from [AlJoMa]
which are more general than Theorem 13.7.16. In particular, we will give an example
where Jlt(Fll) exists even though /u_ ^ GKd and, in fact, is not even locally finite.
However, for now, we give examples (or families of examples) of positive measures
which are in GKj or are at least in S. Any choice then of a generalized signed measure
/A = /i+ — /i_ such that fi+ e 5 and /j.- e GKd produces an example where Theorem
13.7.16 is applicable. We remind the reader that Kd c GKd c S (see Remarks 13.7.4(b)
and 13.7.9(d), respectively), where Kd is the ordinary Kato class (see Definition 10.4.1)
of R-valued functions on Rd.
Example 13.7.17 Take d > 3 and let M be a (d — 1)-dimensional (smooth) submanifold
of Rd. We take K = KM to be the Riemannian volume measure on M and extend K to
R d \M by letting K(R d \M) = 0. [In the literature on smooth measures, somewhat by
abuse of language, K = KM is often called the "^-function" (or "Dirac measure")
supported on M, and is denoted by SM.] One can show that K e GKd. In fact, we
established a very special case of this in Remark 13.7.4(b) where M was a plane in R3.
The case of a (d - 1)-dimensional affine subspace in Rd (d > 3) is not much different.
The submanifold M need not be flat but can be, for example, the surface of a sphere in
Rd (still with d > 3); for instance, KS2 = SS2, the "^-function" on S2 (the unit sphere
in R3)—which, mathematically, is simply the area (or Hausdorff) measure on S2—is a
smooth measure on R3.
Remark 13.7.18 (a) Let K — KM be as in Example 13.7.17 and for any bounded, Borel
measurable function g, g : Rd —> M, let
under U.S. or applicable copyright law.

Then for any B e B(Rd), we have

andboth of these measures are in GKd. Hence, fj, = /z+ — /z_ e 5 — GKj and Theorem
13.7.16 is applicable. (In fact, the comments made so far in this remark are not limited to
the EBSCO
special measures
Publishing K in
: eBook Example
Academic 13.7.17
Collection but hold
(EBSCOhost) for anyonK6/8/2017
- printed € GKd5:46
andPM any
via bounded
Borel measurable
UNIVERSIDAD function
DISTRITAL g.)JOSE DE CALDAS
FRANCISCO
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 369
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(b) A submanifold oflHd(d > 3) of codimension 1 is the focus of the discussion in

Example 13.7.17 and in part (a) of this remark. It is natural to ask if similar assertions
hold for submanifolds of Rd of codimension 2 or more. The answer is "No"—at least for
our present approach to this subject. The key is that such submanifalds have capacity 0
(see (3) of (13.5.6)) and so Wiener paths that start at a point f which is outside of the
submanifold almost surely never visit it (see (2) of (13.5.6)). The result is that the Wiener
integral which is involved in the Feynman-Kac formula (see 5'), namely

does not distinguish the measure K from the 0 measure or, equivalently, the 0 potential,
and so does not distinguish the additive functional AKt from the additive functional that
is identically 0.
(c) In Example 13.7.17 and in (a) of this remark, the interaction is localized to a set
which has Lebesgue measure 0 but positive capacity. (An example of such a set is S2, the
unit sphere in R3, as was seen above.) A considerable amount of work has been done
using Hamiltonian methods with interactions that are localized even further to such sets
as points, curved wires or line segments that share a common end point. A good brief
treatment of this subject including many references and a discussion of circumstances
under which the resulting models are physically reasonable can be found in [BkExH]
(see especially Section 14.6 in that book). An additional interesting feature of the work
just referred to is that in a number of cases, the associated generalized Schrodinger
equations are exactly solvable.
(d) The origin in R2, (0), is a set of capacity zero in R2. (This follows from property
(3) recalled in (13.5.6) because a point is of codimension 2 in M 2 .j Consequently, the
interaction £t(= /Cjoj) = <5{0), the Dime measure at the origin in R2, is too singular to
be a smooth measure (or a perturbation thereof) on #(R2). Hence, neither the extended
Feynman-Kac formula 5' nor Theorem 13.7.16 (about the analytic-in-time Feynman
integral) can be applied to this situation, when the measure is taken to be /it = S(o). How-
ever, recent physical work of Henderson and Rajeev [HenRajl,2]—based on Wilson's
renormalization scheme applied to quantum mechanics (see Remark 13.6.24(b) above)—
provides a (Euclidean or "imaginary time ")path integral representation for the solution
of the corresponding (suitably defined) "Schrodinger equation ". (See also the relevant
references in [HenRajl,2].) In fact, the path space measure involved is no longer Wiener
under U.S. or applicable copyright law.

measure, and relative to this new measure, the "potential" can be taken to be zero
when \JL =• <5(0) in M2 in the counterpart of the Feynman-Kac formula. Moreover, it is
expected—but, to our knowledge, not yet proved—that the new path space measure is
singular with respect to Wiener measure m (in the usual measure-theoretic sense). Note
that by contrast, all the results considered in the present section can be reinterpreted in
terms of functional measures that are absolutely continuous with respect to m. // would
be very interesting to rigorously investigate this example and to eventually develop a uni-
fied theory that would include interaction measures JJL which are of the type considered in
Theorem 13.7.16 (and hence correspond to suitable perturbations of smooth measures),
as well as singular interactions such as fj, = 5{o) in R2. (For the latter example, the ear-
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
lierUNIVERSIDAD
mathematical work
DISTRITAL in [Hugl,2]
FRANCISCO and the relevant references therein may be useful.)
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 370 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In particular, it would be important to be able to include in this extended framework the

model case of the attractive inverse-square potential (say, V(x) = — 1/r 2 in R3, with
r = \\u\\) discussed in some detail in case (3) of Example 13.6.18 from the point of view
of the various analytic-in-mass Feynman integrals considered in Sections 13.5 and 13.6.
The latter interaction, n(du) = V(u)du, has been studied at the physical level of rigor
in [GupRaj] from the point of view of renormalization theory (see Remark 13.6.24(b)
above) but, as far as we know, has not yet been treated successfully from the point of
view of the "imaginary time" path integals considered in [HenRajl,2].
The measures involved in the rest of our examples are all of the form V(u)du; that
is, the interactions are all given by R-valued potential energy functions V. The next
example is a simple one. It shows that a smooth measure need not be locally finite and
also illustrates how the compact sets { K n } from the definition of a smooth measure (see
Definition 13.7.8) can be adjusted to the particular case at hand.
Example 13.7.19 Let V(u) = \\u\\~a. Fix d > 2. We will see that the measure Vdu is
smooth for any a e R. Note that Vdu does not charge sets of capacity 0 for any such a.
Now, let a < d. Then V e Ljoc(R'/) and so Vdu is locally finite and hence smooth.
However, when a > d, V £ Lj oc (R rf ) and so Vdu is not locally finite; specifically,
fv V(u)du = +00 for any open set U containing the origin. Nevertheless, Vdu is still
smooth. In order to see this, take the set Kn from Definition 13.7.8 to be the closed
ball of radius n centered at 0, with the open ball of radius l/n centered at the origin
removed. Note that the set Kn is constructed with the location of the singularity of V in
mind.
Theorem 13.7.16 can, of course, be applied where the measure IJL is Vdu from
Example 13.7.19. In fact, Theorem 13.3.1 can be applied if it is supplemented by Remark
13.3.3(a) where the closed set G is just the singleton set {0}.
Our next example shows that it is possible to have extremely singular positive poten-
tials V such that the measure Vdu is smooth and so Theorem 13.7.16 is applicable. This
example was discussed in connection with the Feynman integral on page 286 of [AlJoMa]
but the potential appeared earlier in [AlMal]. A closely related example appeared still
earlier in [StlVo]. Example 10.3.21 above is somewhat similar in character and illumi-
nates the distinction between algebraic and form sums of operators.
Example 13.7.20 Fix d > 2. Let {uj} be a countable dense subset of Rd and let { a j }
under U.S. or applicable copyright law.

be (for now) an arbitrary sequence of real numbers. It was shown in Proposition 1.3 of
[AlMal] that there always exists a sequence [cj] of strictly positive numbers such that
if we define

thenEBSCO
VduPublishing
is a smooth measure on Rd. Now if we choose [ctj} in such a way that ctj < —d
: eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
for UNIVERSIDAD
all j greater than some
DISTRITAL jo, JOSE
FRANCISCO thenDEVdu
CALDASis nowhere locally finite; i.e. for any nonempty
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 371
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

open subset U of Kd, we have

If, even further, we let a;- = — j for j — 1,2,..., then the function V defined by
(13.7.18) is nowhere Lp-integrable for any p > 0; i.e. for any nonempty open subset U
of Rd and any p > 0,

Remark 13.7.21 (a) It is unlikely that the example just given is of any physical interest;
but it does tell us something about the boundaries of the mathematical theory. The
potential V is everywhere (i.e. on every nonempty open set) extremely singular and yet
the Hamiltonian H = Ho + V is defined on a dense subset of L2(R d ) and the Feynman
integral Jit (Fy) exists for every value it on the imaginary axis. This situation should
be contrasted with earlier results for positive potentials where arbitrary singularities
were allowed either on closed sets of Lebesgue measure 0 (see Remarks 11.4.3(b) and
13.3.3(a)) or on closed sets of capacity 0 [see Theorem 13.5.7 (along with (13.5.5)) as
well as Theorem 13.6.4] and where, in the case of analytic continuation in mass, the
existence of the Feynman integral was established only for Leb.-a.e. value of the mass
parameter.
(b) When a positive potential V belongs to L,1oc(Rd), the familiar space D — D(Rd)
(also denoted Coo00(Rd)) of infinitely differentiable functions with compact support, is a
form core (see Definition 10.3.3) for the quadratic form associated with the operator
HO + V. It is easy to see that when V is nowhere locally integrable as in Example
13.7.20, then the only function in TJ which belongs to the form domain of V, that is,
which belongs to

is the function that is identically 0. In fact, Q(V) cannot contain any nonzero continuous
function. It is natural to ask in this situation if it is possible to find a form core consisting
of functions with at least some nice properties. Theorem 5.7 of[AlMa2] implies that there
is a form core for Ho + V (with V as in Example 13.7.20) such that all thefunctions in the
under U.S. or applicable copyright law.

form core are bounded, compactly supported and quasi-continuous. /Note: A function
f on Rrf is said to be quasi-continuous (seepage 64 of[Fuk]) if and only if f is defined
except on a set of capacity 0 in Kd and for every e > 0, there exists an open subset U of
Rd such that Cap(t/) < e and the restriction of f to K d \t7 is continuous.] We remark
that one not only needs Theorem 5.7 of [AlMa2] for the above but also the fact that
our quadratic form is positive (and so certainly semibounded) and closed, from which it
follows that the quadratic form in [AlMa2, Theorem 5.7] coincides with ours.
We note that the existence of a dense form core for H0+V in the case of extremely
singular potentials, such as in Example 13.7.20, is relevant not only to the analytic-in-
timeEBSCO
(operator-valued) Feynman integral but also, for example, to the modified Feynman
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
integral; see DISTRITAL
UNIVERSIDAD Remark FRANCISCO
11.4.3(c).
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 372 ANALYTIC-IN-TIME OR -MASS FEYNMAN INTEGRALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Items l'-6' gave a theorem (restated later as Theorem 13.7.16) which insured the
existence of the analytic-in-time operator-valued Feynman integral Jit(F l i ) under the
assumptions that /u,+ e S and /u_ 6 GKj. As mentioned earlier in the section, this
hypothesis on fj,-, while convenient for comparison with the work in Section 13.3 (see
Theorem 13.3.1), is not the best that can be made. We illustrate this by simply stating
an example where J" (FM) exists even though /*_ is not locally finite and so cannot be
in GKd (see Remark 13.7.4(a)). We begin by defining a concept that plays a role in the
example below.
Definition 13.7.22 Let K and v be positive smooth measures with associated PCAFs AKS
and Avs, respectively. We say that K is compatible with respect to v if and only if there
exists t > 0 such that

where N is a set of capacity 0 containing the exceptional sets of both AK and Av.
Example 13.7.23 Let d > 2. We define for any k > 2,

If we let /LI = Vdu, then /z is a generalized signed smooth measure (see Definition
13.7.10) and we have

Neither fi+ nor /u_ is locally finite and so neither belongs to GKd- Since //,_ £ GKd,
Theorem 13.7.16 is not applicable. However, it was proved in [Sturl] that /z_ is compat-
ible with n+ and so it follows from Theorem 3.4.8, page 293 of [AUoMa] that 7"(FM)
exists and equals the unitary group e~l!Hc for every t e M. The rough idea is that HO
and V+ rather than just HO are used to control the singularities of the negative part V-.
[In the language of unbounded quadratic forms, one can say that V- is relatively form
bounded with respect to HO + V+. With this in mind, we see that Example 13.7.23 is
also relevant to the study of the modified Feynman integral (from Section 11.4).]
In the example just given, the rapid oscillations of V help us to deal with the strong
under U.S. or applicable copyright law.

singularity at the origin. A more extensive discussion of such highly singular oscillatory
potentials can be found in [Stur2].
In addition to Theorem 3.4.8 of [AUoMa] which was referred to just above, Theorems
3.4.4 and 3.4.5 from the same paper are directed towards extensions of Theorem 13.7.16.
The consequences of Theorem 3.4.4 for the Feynman integral have not been explored and
may well yield further interesting results. We recommend the paper [AlMa2], especially
Theorems 4.1 and 5.5, to the reader who may wish to consider this possibility. Finally,
we remark that q^ should be replaced by q^_ in (iv) of Theorem 3.4.4 of [AUoMa]. (In
the notation of [AUoMa], one must replace Q^ by QI]L_.)
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
We close DISTRITAL
UNIVERSIDAD this section withJOSE
FRANCISCO twoDEcomments.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 ANALYTIC-IN-TIME FEYNMAN INTEGRAL VIA ADDITIVE FUNCTIONALS 373
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 13.7.24 (a) Recently, Chang, Lim and Ryu [ChanLimRy] have established a
stability theoremfor the analytic-in-time operator-valued Feynman integral in the setting
of the present section. Part of their proof adapts techniques from Lapidus' dominated
convergence theorem [Lal2] (see Theorem 11.5.19 along with Theorem 11.5.7). How-
ever, the assumption that Vm -> V Leb.-a.e. in Theorem 11.5.19 is replaced by the
assumption that, for every B e B(M.d), fJ^m(B) and f j , m ^ ( B ) converge monotonically
downward to /i(fi) and n-(B), respectively, as m —> oo. In addition, the sequence of
measures {/^m}^=1 is assumed to "converge dominatedly" to fj. in the following sense:
Mm,+ < v and fJ-mi- < r\, where v e Sa and rj e GKj, with Sa denoting the space of
a-finite smooth measureson Rd. (Here, Mm,+ and nm,- denote the positive and negative
part of n,m, respectively. Further, the order relation involved is the usual one between
measures.) The changes noted above necessitate quite different techniques of proof in
parts of the paper [ChanLimRy].
We note that the assumption in [ChanLimRy] was that v e G Kj, but that it was more
recently shown in [Lim] that v e Sa suffices. This is in closer analogy with Theorem
11.5.19. However, to improve this result still further, it would still remain to eliminate the
"monotonicity assumptions " made in both [ChanLimRy] and [Lim], but not in Theorem
11.5.19 or in [Lal2J.
(b) If ^ = ^i+ — [A- € S — GKd (i.e. (f/u.+ is a smooth measure and /z_ is in the
generalized Kato class), as was assumed in the statement of Theorem 13.7.16, then the
self-adjoint operator associated with [i- is Ho-form bounded with relative bound less
than 1. It follows that the abstract product formula for imaginary resolvents (Theorem
11.3.1) can be applied to this situation, thereby potentially enlarging significantly the
class of examples which can be dealt with via the modified Feynman integral (of Section
11.4). However, the concreteform of the modified Feynman integralfor potentials belong-
ing to this enlarged class remains to be investigated.
If the problem suggested at the end of the last paragraph can be resolved, then the
stability results discussed in part (a) of this remark can be fully applied to the modified
Feynman integral. This would be a further example where (in the spirit of Section 13.4)
one of our approaches to the Feynman integral yields information about another.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

14
FEYNMAN'S OPERATIONAL CALCULUS FOR
NONCOMMUTING OPERATORS: AN INTRODUCTION

... just as a poet often has license from the rules of grammar and pronunciation, we should like
to ask for "physicists' license" from the rules of mathematics in order to express what we wish to
say in as simple a manner as possible.
Richard P. Feynman, 1951 [Fey8, p. 124]

The mathematics is not completely satisfactory. No attempt has been made to maintain mathemat-
ical rigor. The excuse is not that it is expected that rigorous demonstrations can be easily supplied.
Quite the contrary, it is believed that to put the present methods on a rigorous basis may be quite
a difficult task, beyond the abilities of the author.
Richard P. Feynman, 1951 [Fey8, p. 108]

In my paper, the fact that XY was not equal to YX was very disagreeable to me. I felt that this
was the only point of difficulty with the whole scheme.
Werner Heisenberg, reminiscing about his key 1925 paper on quantum mechanics ([Hei]).
(Quoted in [Han, p. 81].)

In this chapter, we give an elementary introduction to the general ideas of Feynman's

operational calculus for noncommuting operators and indicate the connection between
this calculus and the Wiener and Feynman integrals. It may appear at first that these path
integrals have little to do with the operational calculus of Feynman, but the topics are
intimately related as we will begin to see in Section 14.3 below.
The Wiener and Feynman integrals will be used in Chapters 15-18 to study
Feynman's operational calculus in the diffusion (or probabilistic) and quantum-
mechanical settings, respectively. We will return to a more general setting for the oper-
ational calculus in Chapter 19.
The last part of this introduction will concentrate on Chapters 15-18 and will stress
under U.S. or applicable copyright law.

those aspects of the path integrals that are connected with Feynman's operational calcu-
lus. However, many of the results in those chapters have an interest as contributions to
the Wiener or Feynman integrals apart from that connection.
Three final notes here: (1) Further introductory material on our general approach to
Feynman's operational calculus can be found at the beginning of Chapter 19.
(2) When the words "operational calculus" are used in Chapters 14-19, they will
refer to Feynman's operational calculus for noncommuting operators unless it is said or
is clear from the context that we have some other operational calculus in mind.
(3) Our main purpose in this chapter is to introduce the heuristic ideas of the oper-
ational calculus and to describe in broad terms some of the highlights of Chapters 15-19.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
We UNIVERSIDAD
leave the DISTRITAL
mathematically
FRANCISCO precise discussion for these later chapters. We settle here
JOSE DE CALDAS
for AN:
mentioning that our
98476 ; Johnson, standard
Gerald setting
W., Lapidus, willL..;
Michel beThe
a separable Hilbert
Feynman Integral and space Ji over C with
Feynman's
Operational Calculus
Account: ns000601
 FUNCTIONS OF OPERATORS 375
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

the operators either being (i) bounded, or else being (ii) unbounded operators which are
the generators of (Co) semigroups (see Chapter 9).
At the end of this chapter, we will give a list of references connected to Feynman's
operational calculus.

14.1 Functions of operators

It is useful in many areas of mathematics and its applications to form functions of
operators. One sees this even in elementary settings. Consider the initial-value problem
x' — Ax, x(0) = jco, where A is an n by n matrix of constants and x is an n vector
whose components are functions of time. The elegant solution to this problem is given
by*(0 = [exp(fA)]C*o)-
Since exponential functions play a central role in solving evolution equations, it is
especially important to be able to exponentiate operators. The theory of semigroups of
operators outlined in this book in Chapters 8 and 9 may be regarded as the theory of
exponentiating operators. (It is necessary to exercise some caution in adopting this point
of view; in fact, as was seen in Chapter 9, (Co) semigroups have some but not all of the
properties of the numerical exponential function.)
In quantum mechanics, where the basic observables are (possibly unbounded) self-
adjoint operators on an infinite dimensional Hilbert space, one often deals with functions
of such operators. Indeed, this is one of the recurring themes in Chapters 6-19 of this
book.
Let H be the self-adjoint energy operator or Hamiltonian for a quantum system and
let HO be the free Hamiltonian for the same system. One of the following three functions
of H or HO has either been directly involved in, or has been the main motivation for,
the central results in the last four chapters: e~"H (the unitary group), e~'H (the heat
semigroup), and [/ + i(t/n)Ho]~~] (the free resolvent). Various other functions of self-
adjoint operators have arisen as well; see, for example, the proof of Theorem 11.3.1
where a number of such functions are used.
The functional calculus for a single self-adjoint operator (bounded or unbounded)
is extremely rich, as we have seen in Chapters 10-13. The situation is unchanged for
functions of any finite number of self-adjoint operators provided these operators commute
with one another. However, as soon as commutativity fails, a functional calculus becomes
much more difficult, even in the presence of self-adjointness.
The question of defining functions of noncommuting self-adjoint operators is not the
under U.S. or applicable copyright law.

only difficulty. Even when this can be accomplished, the process may not be unique. The
lack of uniqueness gives rise to the "ambiguity of quantization." Before explaining how
this ambiguity arises, we describe briefly and in a special case the canonical quantization
of a classical mechanical system. We refer to [Mac, §2.4] for a much more complete
discussion of the quantization procedure.
Suppose that we have a single classical particle moving in K. A classical observable
is an R-valued function f(q,p) of the position q and momentum p of the particle. The
position observable q itself is quantized by replacing it by the position operator Q on
L2(K) defined by
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 376 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The momentum operator,

quantizes the classical momentum p. (Here, as in Chapter 6,h = h/2n denotes Planck's
constant divided by 2n.) Both Q and P are unbounded, self-adjoint operators. Note that
Q and P do not commute with one another; in fact,

The idea is to carry out the general quantization procedure by replacing the classical
observable f(q, p) by the quantum observable f(Q, P). If f(q, p) — q2 + p2, this
can be done unambiguously by taking f(Q, P) = Q2 + P2. However, ambiguities
arise in the procedure as soon as products of Q and P are involved. Consider the simple
example f(q, p) := qp. Now qp = pq, and thus both products give the same classical
observable. But QP / PQ by (14.1.3), and so QP and PQ are distinct quantum
observables. Which operator should be associated with f(Q, P)? Should it be QP or
PQ, or perhaps something else, say, \(QP + PQ)1
Remark 14.1.1 (a) Much of the focus so far in this section has been on self-adjoint
operators since these are the observables in quantum mechanics. However, the difficulties
and ambiguities involved in forming functions of operators are present whenever the
operators involved fail to commute.
(b) As we continue, we will see that the ambiguity "problem " can actually be an asset
insofar as it will allow us to model a wider variety of physical phenomena. This positive
aspect of the ambiguity will be involved in all of Chapters 15-19 but will perhaps be
seen most clearly in Chapters 17 and 19.
(c) The richness of the functional calculus for commuting self-adjoint operators car-
ries over to commuting normal operators. Indeed, we made use of thisfact in Sections 11.6
and 13.6.
(d) 'While it is true that a functional calculus for noncommuting self-adjoint opera-
tors is limited and difficult, interesting work along these lines has been done. See, for
example, the papers [And], [Tayl] and, for recent work andfurther references, thepaper
[KisRal,2].
under U.S. or applicable copyright law.

14.2 The rules for Feynman's operational calculus

Motivated by his work on path integration in nonrelativistic quantum mechanics [Fey2]
and on quantum electrodynamics [Fey 5-7], Feynman gave in his 1951 paper An operator
calculus having applications in quantum electrodynamics [Fey8], a heuristic formulation
for an operational calculus for noncommuting operators. Our purpose in this section and
in the first part of Section 14.3 is not to discuss rigorous mathematics but to illustrate
and discuss Feynman's heuristic ideas.
Talking about the unconventional use of rules and formulas in his time-ordered
operator calculus, Feynman writes [Fey8, p. 124]:
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
... UNIVERSIDAD
The physicist is very
DISTRITAL familiar
FRANCISCO with
JOSE such a situation and satisfied with it, especially since he
DE CALDAS
AN: 98476 that
is confident ; Johnson,
he canGerald
tell ifW.,theLapidus,
answerMichel L..; The Feynman
is physically Integral
reasonable. Butandmathematicians
Feynman's may be
Operational Calculus
Account: ns000601
 THE RULES FOR FEYNMAN' S OPERATIONAL CALCULUS 377
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

completely repelled by the liberties taken here. The liberties are taken not because the mathematical
problems are considered unimportant. On the contrary, this appendix is written to encourage the
study of these forms from a mathematical standpoint. In the meantime, just as a poet often has
license from the rules of grammar and pronunciation, we should like to ask for "physicists' license"
from the rules of mathematics in order to express what we wish to say in as simple a manner as
possible.

The position on the page is the standard way of keeping track of the order in which
products of noncommuting operators act. Feynman had a different way of doing this,
which is one of the keys to his operational calculus.

Feynman's time-ordering convention

Feynman used time indices to specify the order of operators in products, where it is
understood that operators with earlier time indices always act before operators with later
time indices. For example, given operators P and y,

Feynman's heuristic rules

Some of the "rules", roughly described, for the operational calculus are as follows:
(1) Attach time indices to the operators to specify the order of operators in products.
(2) With time indices attached, form functions of these operators by treating them as
though they were commuting.
(3) Finally, "disentangle " the resulting expressions; that is, restore the conventional
ordering of the operators.
Feynman says of the disentangling process [Fey8, p. 110], "The process is not always
easy to perform and, in fact, is the central problem of this operator calculus." Feynman
did not attempt to prove his results mathematically, and it is not always clear, even
heuristically, how his rules are to be applied.
How should one attach time indices to the operators as mentioned in rule (1)? First
of all, the operators may come with indices naturally attached. This happens with the
operators of multiplication by time-dependent potentials, for example, and also in con-
under U.S. or applicable copyright law.

nection with the Heisenberg (or interaction) representation in quantum mechanics (see
[Sud, Sections 3.4 and 3.5]). However, the operators that appear most frequently in the
literature of quantum mechanics, perhaps especially in the mathematical literature, are
time independent. Given such an operator a, almost without exception, Feynman attaches
time indices according to Lebesgue measure as follows:

where a(s) := a for 0 < s < t. Despite its artificial appearance, we will eventually see
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
thatUNIVERSIDAD
this device is extremely
DISTRITAL FRANCISCOuseful
JOSE DEinCALDAS
many situations. However, it is not always the right
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 378 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

thing to do. In the following chapters, a variety of measures will be used to assign time
indices. Physical or mathematical considerations may determine the appropriate choice.
We present two situations which are easily described and are motivated by physical
considerations.
An experimenter may wish to study the reaction of a system to a force which is
being turned on or off at certain times. In this case, the natural measure to use would be
H(ds) = g(s)ds, where g(s) alternates between 1 and 0 at the appropriate times. (Of
course, in a case such as this, one can equally well continue to think of the measure as
Lebesgue measure and regard g(s) as multiplying the operator. Indeed, Feynman adopts
this point of view in two places in his paper [Fey8, pp. 113 and 114].)
A system, perhaps again under the control of an experimenter, is subjected to sig-
nificant forces which act over certain very short time intervals. Such a situation can
often be usefully modeled by a linear combination of Dirac measures (or, heuristically,
"<5-functions"). We will frequently encounter such situations in the following chapters.
We now turn to two simple examples which illustrate Feynman's heuristic rules as
well as the roles played by the time-ordering convention and by the measures. We note
that systematic use of measures in connection with Feynman's operational calculus was
introduced by the authors in [JoLal] and will play a prominent role in Chapters 15-19.

Two elementary examples

Let a and /3 be operators on the separable Hilbert space Ti.. We let a(s) := a and
P(s) := f> for all s in the time interval in question.
Example 14.2.1 (Disentangling aft) Given numbers a and b, let

g(a,b)=ab. (14.2.3)

Also let t = 1 for convenience. We follow the most common procedure with a.
and assign time indices according to Lebesgue measure; i.e. a = f0 a(s)ds. On the
other hand, we take /6 = /0 f}(s)8\ (ds), where S\ is the Dirac measure with total mass 1
concentrated on {!}. We first write down the string of "equalities" below and then discuss
them:
under U.S. or applicable copyright law.

The first equality in (14.2.4) just restates the assignment of time indices, while the
second reflects the nature of the measures involved. The third "equality" time-orders the
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
expression;
UNIVERSIDADthat is, theFRANCISCO
DISTRITAL time indices
JOSE DE involved
CALDAS in the second factor on the right-hand side
of the third "equality" are all less than the timeThe
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; Feynman
index Integralinand
involved theFeynman's
first factor. We are
Operational Calculus
Account: ns000601
 THE RULES FOR FEYNMAN'S OPERATIONAL CALCULUS 379
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

done disentangling when the expression is time-ordered; that is, when the order of the
operators on the page corresponds to the order of the time indices. We are then free to
calculate insofar as possible any remaining integrals. Doing this, we obtain the fourth
equality in (14.2.4).
// is typical ofFeynman's use of his rules that not all of the "equalities" in a calcu-
lation are valid if they are interpreted in the usual way. If a and ft do not commute, the
third equality and also the final conclusion of (14.2.4) are certainly false if interpreted
literally.
How then should (14.2.4) be interpreted? The idea is that the product aft may have
more than one meaning if a and ft do not commute. (Compare (14.2.4) with (14.2.6) and
(14.2.8) below.) The sense of (14.2.4) is that with Lebesgue measure / and S\ associated
with a and ft, respectively, the product is defined as fta. To state it another way,

We did not use Feynman's second rule in the calculation in (14.2.4). This is not
typical. The second rule usually plays a crucial role. We will see an illustration of this
in Example 14.2.3 below.
Next, we replace &\ by So in (14.2.4) and see what changes result:

The first and second equality in (14.2.6) are much the same as before, but here the right-
hand side of the second equality is already time-ordered, and so we can go directly to
calculating the integrals.
It happens in this case that all three equalities in (14.2.6) are valid if interpreted in
the usual way. The outcome of (14.2.6) can be summarized by writing

What do we get if Lebesgue measure / is used to assign time indices to both a and ftl
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 380 INTRODUCTION TO FEYNMAN' S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 14.2.1. The unit square [0,1]2 in the heuristic derivation of formula (14.2.9)

In the second equality, we wrote the product of one-dimensional integrals as an

integral over the unit square [0, I]2. Now, there is no definite time-ordering in [0, I]2,
and so in the third equality we expressed the integral over [0, I]2 as the sum of integrals
over the upper (s\ < $2) and lower (52 < si) triangles, respectively. (See Figure 14.2.1.)
In the fourth equality, we made the ordering of the operators in each term consistent with
the time-ordering. (More specifically, following Feynman's time-ordering convention
(14.2.1), we wrote fi(s2)ot(s\) = flat for si < $2 and a(si)p(s2) = aft for $2 < s\.
See Remark 14.2.2(b) below for the case when s\ = si.) Note that the upper and lower
triangles in Figure 14.2.1 have the same area, equal to 1/2. As before, we could then
calculate the remaining integrals.
We see from the calculation in (14.2.8) that

We have an additional comment about the computation in (14.2.8), but first we make
under U.S. or applicable copyright law.

a remark that comes from comparing the three calculations above.

Remark 14.2.2 (a) The three functions gi,S[, gi,s0 and gij all agree when a and ft
commute, even though they are all distinct without this commutativity. What was one
function in the commutative case has become threefunctions. (In fact, there are additional
possible functions as well.) The three functions represent three different disentanglings
of the simple product function; the distinct disentanglings are associated with the three
pairs of measures (I, &\), (/, So] and (I, /).
Is it useful to have different definitions for functions which somehow involve products
ofnoncommuting
EBSCO Publishing :operators? We
eBook Academic will see (EBSCOhost)
Collection that it is so- in Chapters
printed 15 through
on 6/8/2017 19—perhaps
5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
most strikingly in Chapters 17 and 19 where exponential functions and associated evo-
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
lution equations
Operational play a prominent role.
Calculus
Account: ns000601
 THE RULES FOR FEYNMAN'S OPERATIONAL CALCULUS 381
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(b) In the third equality of our calculation in (14.2.8), we have ignored the diagonal
of the unit square. This seems reasonable since the diagonal C$i = $2) has measure 0
with respect to Lebesgue measure I x/ on [0, I]2v see Figure 14.2.1. Related steps will be
involved in two or more dimensions in many places as we continue. However, when the
measure we are dealing with has nonzero discrete part, ignoring the diagonal will not
be possible and complicated combinatorial issues will frequently arise. [See especially
Chapter 15 (Sections 15.3-15.6), as well as Chapters 17 and 19.]
(c) There are circumstances in which it is desirable (or at least customary) to write the
product of two noncommuting operators in one or another specific order as in (14.2.5) or
(14.2.7). It is standard to write differential operators so that each term is a kth derivative
(where k is a nonnegative integer) followed by a multiplication operator. However, in
the functional calculus ofKohn andNirenberg [KoNi] for pseudodifferential operators,
this order is exactly reversed; i.e. the multiplication operator acts first, Similarly, if
a and ft are, respectively, the annihilation and creation operators from quantum field
theory, then the ordering that results from (14.2.5) corresponds to Wick normal ordering
[GliJa, p. 16] whereas the reverse ordering (resulting from (14.2.7)) corresponds to
Wick anti-normal ordering. [Recall that the Wick (or normal) ordering is often used
(possibly without explicit mention) in many calculations occurring in quantum field
theory; see, for instance, ([Berl], [GliJa], [Wei,2]).] It is somewhat harder to explain,
but we mention that if a and ft are the momentum and positions operators, respectively,
then the disentangling associated with (14.2.9) is connected with the Weyl functional
calculus ([And], [Foil, p. 79], [Hdrl], [Ne3, §6/j.
We are now ready for our second elementary example. Let a and ft be operators on
the separable Hilbert space H, and let a(s) = a and ft(s) = ft for all s in the time
interval in question.
Example 14.2.3 (Disentangling e-'a+<^) Given parameters t > 0 and a> e C, as well
as numbers a and b, let

[Here, / (resp., w) can be thought of as a "time" (resp., "weight") parameter.]

Let T e [0, t]. We attach time indices to a and ft as follows: a = j /Qa(s)ds,
ft = ^ ft(s)&r(ds). Letting r e (0, t) for now and using rules, we write
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 382 INTRODUCTION TO FEYNMAN' S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The first equality in (14.2.11) restates the assignment of time indices. The second equality
is crucial; it follows Feynman's second rule by using the commutative exponential law
even though the operators a and B need not commute. The fourth equality time-orders
the product; that is, the order of the operators in the product is made to coincide with
the time-ordering. This is the situation which one works towards in the disentangling
process. Once this is done, we are free to calculate (wherever possible) any remaining
integrals.
Note that as in Example 14.2. 1, the "equalities" in (14.2.1 1) cannot all be interpreted
in the usual way if a and B do not commute; specifically, the second and fourth equalities
need not hold if interpreted literally. The calculation above is more typical of the disen-
tangling process than the one in Example 14.2.1 insofar as the second rule is invoked.
However, it is still much simpler than many of the calculations that will appear in later
chapters.
We can summarize the calculation in (14.2.11) by the following continuum of for-
mulas: With / given by (14.2.10), we have for every r € (0, t),

Assume for simplicity that a and ft are bounded operators. If a and B ft commute, then
the continuum of functions {fi,&r : 0 < rT < t) of a and B
ft reduce to a single function.
Remark 14.2.4 (a) When the operator —sa, s > 0, is being exponentiated, we will often
want to permit a to be unbounded. In that case, our usual assumption will be that —a
is the generator of a (Co) semigroup. We will continue to use the suggestive exponential
notation in that case (as we frequently did in Chapters 8-13) even though what we really
have mathematically is the semigroup generated by —a.
(b) Note that it is normalized Lebesgue measure 1/tds that is used when we write
a —1/t j /Q
/Qa(s)ds.
a(s)ds. Obviously,
Obviously, one
one can
can equivalently
equivalently write
write ta
ta == f^f^ a(s)ds,
a(s)ds, and
and itit isis this
this
that will usually be done as we continue.
(c) Naturally, if instead of choosing f as in (14.2.10), we let

and set
under U.S. or applicable copyright law.

(but keep a as before), then we obtain similarly that for every T e (0, t),

This choice of discrete measure (v = wSr, a Dime measure with total mass u> con-
centrated at the instant r) turns out to be better suited to our later developments in
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
Chapters 15-19.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TIME-ORDERED PERTURBATION SERIES 383
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Exercise 14.2.5 (The r = 0 and r = t cases of (14.2.12)) Using Feynman's heuristic

rules much as in (14.2.11), show that we arrive at the formulas

and

[Observe that the r = 0 and r — t cases of (14.2.12) agree with (14.2.13) and (14.2.14),
respectively.]
Remark 14.2.6 Although we are not prepared to discuss the details now, we mention
that formula (14.2.12) can be used to model the evolution of certain physical systems.
In fact, this formula represents related but distinct evolutions for distinct values of T.
The expression u>ft(s)ST(ds) is associated with an impulsive force acting sharply at
time r. These and related matters will come up many times as we continue, especially
in Chapters 15-17 and 19. (See, for instance, Section 15.3, formulas (15.5.18) and
(15.5.24), Corollary 17.2.9, and Example 16.2.9.)
We will continue to illustrate the use of Feynman's rules in the first part of the next
section by using them to calculate a perturbation series.

14.3 Time-ordered perturbation series

Time-ordered perturbation series will be one of the central themes of Chapters 15-19.
Perturbation series via Feynman 's operational calculus
We begin by using Feynman's heuristic ideas to calculate in a simple case the first
few terms of a perturbation series. Our purpose in this is both to further illustrate and
explain the use of Feynman's ideas and to see the connection between these ideas and
perturbation series.
Let H be a separable Hilbert space over C and let a be a fixed operator on H. It
will be convenient to define «(» := a for every s > 0. In the sequel we will typically
either have (Chapters 15-18) or require (Chapter 19) that —a is the generator of a (Co)
semigroup on H. We attach time indices to a in the following way: For every / > 0,
write
under U.S. or applicable copyright law.

We allow for the possibility that B be time-dependent, B : [0, +00) ->• C(ri). If
ft is constant, ft e £-(H), we write here tft = /0' ft(s)ds, where ft(s) := ft for every
5 > 0. We will frequently associate measures other than Lebesgue measure with ft, but
note that, for now, we are using Lebesgue measure / on [0, t]. We will usually require
(in Chapter 19) or have (in Chapters 15-18) that the operators {ft(s) : 0 < s < 00}
all EBSCO
commute with one another, but we will not require
Publishing : eBook Academic Collection (EBSCOhost)
that the Bs commute with the
- printed on 6/8/2017 5:46 PM via
sa
operators in DISTRITAL
UNIVERSIDAD the semigroup
FRANCISCO(T(s)
JOSE DE = e~ : 0 < s < 00} associated with a. We will
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 384 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

need a measurability assumption on the function B, but we omit that discussion here.
Finally, when Lebesgue measure is used, we will assume that for every t > 0,

This last condition is of course satisfied when B is a constant function.

We now wish to discuss the disentangling of the exponential expression

Exponential functions are central to the study of evolving physical systems and they
played a key role in [Fey8]. Their role is expanded still further in the remaining chapters
of this book.
We write

where the second "equality" comes from applying the commutative exponential law,
exp(A + B) = [exp(A)][exp(5)], in a noncommutative setting as we did in the second
"equality" in ( 14.2. 1 1 ). Just as in the earlier situation, the "equality" cannot be interpreted
in the usual way. The second step in (14.3.3) is, of course, inspired by Feynman's second
rule.
A further point concerning the third expression in (14.3.3): It is counterproductive
to calculate the integral — f0 a(s)ds at this stage. Leaving it as it is will facilitate the
disentangling process.
Continuing, we expand exp(/0 f)(s)ds) from (14.3.3) in a series and obtain
under U.S. or applicable copyright law.

Next we work out the m = 0,1,2 terms of the series.

m = 0: This term is just exp (- /Q a(s)ds\. Here, there are no operators present
which fail to commute with a. Thus no disentangling needs to be done, and hence nothing
is lost in calculating the integral. Doing so, we obtain the semigroup generated by —a:

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TIME-ORDERED PERTURBATION SERIES 385
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

m = 1: We begin with the formal calculations suggested by Feynman's heuristic

ideas and then comment on the steps involved.

In the first equality above, we relabel the dummy variable in the left-hand factor and
move the expression inside the other integral. The second equality is obtained by writing
the integral with respect to •$•] as the integral from 0 to s plus the integral from s to t.
This step looks ahead to the third "equality", where the time-ordering is carried out.
Once the time-ordering is complete, we can calculate the integrals in the exponentials
and obtain the final expression in (14.3.6). Note that the third equality in (14.3.6) cannot
be interpreted in the standard manner.
m = 2: We begin as before with the formal calculations.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 386 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Some of the steps in (14.3.7) are simple or are similar to earlier steps, and we will
not need to comment on them here. The sixth 'equality' is one key; the time-ordering is
completed there. This 'equality' cannot be interpreted in the usual way. Two issues are
involved in the third equality. The first is that the integrand is unchanged by a permutation
of the variables. The other is that the diagonal (i.e. {(s\, $2) € [0, r]2 : s\ = $2}) is a
set of Lebesgue measure zero in [0, t]2 and so can be ignored. (See Remark 14.2.2(b)
above.) Putting these two comments together, the integral over [0, t]2 can be reduced to
2! times the integral over the open triangle 0 < s\ < $2 < t. Using (14.3.4) through
(14.3.7), we now write

Remark 14.3.1 (a) The mth term of the series (14.3.4) will be disentangled in Chapter 19
via direct use of Feynman 's ideas as above. Further, in the next chapter, the disentangling
will be carried out in a setting where the Wiener and Feynman integrals can be employed.
(See the special case of Corollary 15.3.6 where u is taken to be Lebesgue measure I.)
(b) If a = — | A = i HQ and /6 (s) = —iV, the operator of multiplication by —i times
the potential energy function V, then (14.3.8) becomes
under U.S. or applicable copyright law.

which is the Dyson series expansion [Dysl] for the quantum-mechanical unitary group
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
exp(— it(Ho DISTRITAL
UNIVERSIDAD + V)). FRANCISCO
(See formula (15.3.22) below.) On the other hand, if we take
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TIME-ORDERED PERTURBATION SERIES 387
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

a = — 2 A = HO and B ( s ) = —V, we are in the diffusion (or probabilistic) setting,

and (14.3.8) yields

Formulas (14.3.9) and (14.3. 10) are time-ordered perturbation series expansions for the
unitary group e ~"(#o+V) and the heat semigroup e~t(-Ho+v), respectively. [Note that by
using the integral expressions obtained in Section 10.2 (Theorems 10.2.7 and 10.2.6)
for the free unitary group e~"H° and the free heat semigroup e~t H °, we can rewrite
(14.3.9) and (14.3.10) in a more concrete form.] We stress that in order to insure that
the perturbing operator B is bounded, we will assume in both cases immediately above
that the potential energy function V is essentially bounded (see Proposition 10.1.1).
(c) The series that is most often referred to as the "Dyson series " in the literature
([Dysl,2,4], [ReSi2, pp. 282-283], [Schwel, §11. f], [Sud, p. 119]) is more compact and
slightly different from the series in (14.3.9). Let V(t) := U ( t ) - l VU(t), where U(t) =
exp(-it H0). (The time-dependent operator V is the "Heisenberg representation" of the
operator V.) Note that \\V(t)\\ = \\ V\\ for every t since U(t) and U(t)~l = U(-t)are
unitary for all t. Form the series

This is the Dyson series in the "Interaction

"interaction representation " (or "interaction picture ")
of quantum mechanics; see, e.g., [Han, §11.1] or [Sud, §3.4 and §3.5]. It is not equal to
the series in (14.3.9), but the two series are related in a simple way:
under U.S. or applicable copyright law.

[In a much more general setting, we will use a closely related form of the "interaction
(orDirac) representation" ([Han, §11.1]) of such series in Section 17.6; this will enable
us, in particular, to obtain a product integrator "time-ordered chronological product")
representation of the solutions to the associated evolution equation (see [La 16]).]
Since exp(-t Ho) is a semigroup but not a group, the series (14.3.10) has no inter-
action representation. However, in Chapters 15-18, it will frequently be useful for us to
compare (14.3.9) with (14.3.10). Hence (except in Section 17.6), formula (14.3.9) will
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
suitUNIVERSIDAD
our needs betterFRANCISCO
DISTRITAL than theJOSEseries in the Heisenberg or interaction representation.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 388 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(As the reader will see in Chapter 15, it is usually generalizations of (14.3.9) and
(14.3.10) that will be compared.)
We indicated in Section 14.2 that a variety of measures can be used to assign time
indices to operators. Let u be a (Borel) measure on [0, +00), which is finite on compact
subsets of [0, +00). What happens to the perturbation series (14.3.8) if /Q P(s)ds is
replaced with /J B(s)[i(ds)? If we make the additional assumption that u is a continuous
measure, that is, every single point set is u-null, then the appearance of (14.3.8) is not
changed much. We obtain

where u x u denotes the measure on [0, +00) x [0, +00) obtained by taking the product
of u with itself. The largely heuristic 'derivation' that led to (14.3.8) needs to be changed
significantly at only one place to give (14.3.11) instead: In the third equality of (14.3.7),
we used the fact that the diagonal D of the square [0, t]2 has Lebesgue measure zero and
so can be ignored in the integral over [0, t]2. Similarly, it is true that (u x u ) ( D ) = 0.
We will examine this issue carefully for m > 2 in Lemma 15.2.7; we simply mention for
now that the equality (u x u ) ( D ) = 0 follows from Fubini's theorem and the continuity
of the measure u (that is, the fact that u({s}) = 0 for every s in [0, t]).
In spite of the similarity between formulas (14.3.11) and (14.3.8), different choices
for u can represent very different disentanglings and very different time evolutions. We
will examine various possibilities in later chapters, but the reader might wish to consider
briefly even now the following cases:
(a) The measure u is the zero measure until some time to and then is Lebesgue
under U.S. or applicable copyright law.

measure beyond to-

(b) The measure u, alternates between the zero measure and Lebesgue measure on
successive time intervals.
(c) The measure u(ds) = g(s)ds is approximately the Dirac measure at some time
to (see Example 16.2.9).
(d) The measure u is a continuous singular measure, such as the Cantor-Lebesgue
measure (see Example 15.5.3).
Suppose that the continuous measure u in (14.3.11) is replaced by a measure n
which is not continuous; that is, n({r}) > 0 for at least one instant T in the time inter-
val EBSCO
[0, t].Publishing
One can: eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
still obtain a perturbation series for exp ( — f a + /0' fi(s)r)(ds) J, but
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TIME-ORDERED PERTURBATION SERIES 389
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

the series will no longer have the simple form indicated in (14.3.1 1). See Section 15.3
for a discussion of the case where rj = n + 0)ST with u continuous (and r in (0, t)).
[As will be explained further on, the measure v :— a)Sr, with (constant) "weight" co
and <5r the Dirac measure concentrated at the instant r, is called the discrete part of
r;; see especially Section 15.2.F.] The situation can become extremely complicated as
the reader will see in Section 15.4. We can get some idea why complications arise for
discontinuous n [that is, for rj with a nonzero discrete part v, necessarily a (possibly
infinite) linear combination of Dirac measures concentrated at various instants] by con-
sidering the m = 2 terms of the series and realizing that the diagonal cannot be ignored
in integrals over [0, t]2 with respect to n x n. (See Figure 14.2.1 above, with [0, I]2
replaced with [0, t]2.)

Perturbation series via a path integral

We turn now to the calculation of the first three terms of a perturbation series using the
path integral of Wiener (introduced and discussed in Chapters 3 and 4). We will compare
this mathematically rigorous computation with the earlier one (14.3.8) which made use
of the heuristic ideas of Feynman's operational calculus. In particular, we will compare
our final result here with (14.3.10), the special case of (14.3.8) which involves the heat
semigroup.
We work in one space dimension for simplicity. Also, we assume that the potential
V : R -+ R belongs to L°°(R); that is, V is essentially bounded on R. [Actually, in our
work presented in Chapters 15-18, we will allow V to be time-dependent (and hence to
be defined on [0, t] x R), but we will leave such issues aside for now.] This assumption
makes it easy to verify the existence of the integrals involved as well as to justify the
interchange of limit processes.
Using the relatively simple case of the Feynman-Kac formula (Theorem
12.1.1) where V is essentially bounded, we can write for every <p e L 2 (R) and
Leb.-a.e. £ in R,
under U.S. or applicable copyright law.

[Recall from Chapter 3 that Wiener measure m is a probability measure acting on Wiener
space C'0 (the space of continuous paths x = x(s) : [0, t] -» M such that x(0) = 0).]
Note that there is some similarity between (14.3.12) and (14.3.4). While the semi-
group e~ t H 0 does not appear explicitly in the last three expressions in (14.3.12), it is
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
there implicitly
UNIVERSIDAD through
DISTRITAL the presence
FRANCISCO of Wiener measure m.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 390 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Next we calculate the m = 0, 1, 2 terms of the series (14.3.12). It is the m = 2 term

that will be the most instructive in relation to our later work.
m — 0: By applying the Feynman-Kac Formula, Theorem 12.1.1, with the potential
identically equal to zero, we obtain

The reader should compare this fact with (14.3.5) and with the m = 0 term on the right-
hand side of (14.3.10). We remark that (14.3.13) could also be calculated by using the
Wiener integration formula in Theorem 3.3.5 and then the explicit formula for the free
heat semigroup found in Theorem 10.2.6.
m = 1: By Fubini's theorem, we have

We will typically need to work at ordering the time indices (see the case m = 2 below)
and that will sometimes be combinatorially complicated. However, in (14.3.14) we have
0 < s\ < t , and since the set {s\ e [0, t] : s\ = 0 or si = t } has Lebesgue measure
zero, we can limit attention to the set where 0 < s\ < t.
Now applying Wiener's integration formula, Theorem 3.3.5, to (14.3.14) and then
making the simple substitution i>o = MO + £» v\ = u\ + £, we can write
under U.S. or applicable copyright law.

Finally, writing the inner integral in the last expression in (14.3.15) as an iterated integral
and then using the formula for the heat semigroup given in Theorem 10.2.6, we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TIME-ORDERED PERTURBATION SERIES 391
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The reader should compare the last expression in (14.3.16) with the last expression
in (14.3.6) and especially with the second term of the right-hand side of (14.3.10).
In (14.3.16), the integrand has the operators acting on <p € L2(M) and the resulting
function evaluated at £ e K whereas only the operators appear in the earlier formulas.
This difference is only superficial. A more important difference is that the time-ordering
of the operators in (14.3.16) is the reverse of the time-ordering in the earlier expressions.
Since the Wiener integral precisely reverses the earlier time-ordering, we will eventually
see (beginning with Remark 15.3.7) that it is not especially difficult to make an adjustment
which produces exact agreement.
m = 2: The ordering of time indices is crucial to Feynman's operational calculus
and to the use of Theorem 3.3.5, Wiener's integration formula. The second and third
equalities below prepare for the time-ordering and the fourth equality carries it out:

where the equality immediately above uses the fact that the diagonal D of [0, t]2 has
l x l measure zero, with / denoting Lebesgue measure on [0, t]. Note that in the first
part of the last expression in (14.3.17), we have 0 < s1 < s2 < t whereas we have
0 < s2 < s1 < t in the second part. Actually, the two parts are easily seen to be equal.
Using that fact, Fubini's theorem and the Wiener integration formula (Theorem 3.3.5),
we can write
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 392 INTRODUCTION TO FEYNMAN' S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We finish finding the desired formula for I2 by making the substitution ii0 = MO +
£, DI = MI + £, vz = U2 + £, writing the resulting integral over R3 as an iterated
integral and, finally, using formula (10.2.26) for the free heat semigroup e~sli° given in
Theorem 10.2.6:

The reader should compare the third term on the right-hand side of (14.3.10) with
the last expression above. As operators, they are the same except for the time-reversal.
The calculation in (14.3.7) should also be compared with the calculation of I2 above,
and the strong similarities should be noted. The computation of the Wiener integral I2
is, however, mathematically rigorous and can be done without ever explicitly invoking
Feynman 's rules.
Finally, we use the calculations of I0, I1, and I2 above to write the first three terms
of the series in (14.3. 12):
under U.S. or applicable copyright law.

[Eventually, we will want to analytically continue the expression (14.3.20) in a suit-

able (mass or diffusion) parameter A. When we do this (see, for example, Corollary 15.3.6
with n = 1), the form of the resulting perturbation series will essentially remain the
same.]
We have seen above that there is a connection between path integrals and Feynman's
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
operational
UNIVERSIDADcalculus. FeynmanJOSE
DISTRITAL FRANCISCO wasDE aware
CALDAS of this; indeed, it was a desire to extend his
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 TIME-ORDERED PERTURBATION SERIES 393
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

work on the path integral [Fey2] to quantum electrodynamics (QED) which led him to
his operational calculus.

The origins of Feynman's operational calculus

We give here a brief historical sketch of Feynman's invention of his "rules" for forming
functions of noncommuting operators. As a graduate student, Feynman was fascinated
by quantum electrodynamics (QED). The theoretical and computational difficulties with
QED were regarded as a central problem of the physics of the time. Recall from Chapter7
that Feynman's 1942 thesis [Fey 1] and the resulting 1948 paper [Fey2] developed a path
integral in the setting of nonrelativistic quantum mechanics. Feynman also wanted a
path integral for relativistic QED. He did not succeed in this, but he did invent methods
of calculation similar in certain situations to those for path integrals which allowed
him to compute perturbation (or Dyson) series for QED. These calculations were in
the spirit of (14.3.3)-(14.3.8) above but were more complicated. The famous Feynman
diagrams were a way of keeping track of the terms of the resulting perturbation series.
In Feynman's first two papers [Fey5,6] on this subject, he discussed the physics involved
but did not explain his methods of calculation. These methods as applied to QED were
explained in his 1950 paper [Fey7]. Finally, in 1951, he gave a largely mathematical, but
far from mathematically rigorous, explanation and extension of his formal methods of
calculation which he called the operator calculus (for noncommuting operators) [Fey8].
Although [Fey8] contained examples from quantum mechanics and QED, it is clear that
Feynman thought of it as primarily a general mathematical method for forming functions
of noncommuting operators.
The papers [Fey 2-8]—all of which appeared in the period 1948-51 —are the compon-
ents of a single extraordinarily rich and influential research development. This devel-
opment is traced in much more detail in Jagdish Mehra's fine biography of Richard
Feynman [Me, esp. Chapters 6 and 10-15] and in the valuable monograph [Schwe2] and
book [Schwe3] of Silvan Schweber.
Remark 14.3.2 (a) The Dyson series for nonrelativistic quantum mechanics (see
(14.3.9)) is not difficult to compute from Feynman's path integral [Fey2j; somewhat
surprisingly, it does not appear in [Fey2]. However, such Dyson series play a major role
in the later book of Feynman and Hibbs [FeyHi].
(b) Dyson's role was, in particular, to explain in [Dysl] the connections between
under U.S. or applicable copyright law.

the approaches to QED put forth by Tomonaga, Schwinger, and Feynman. (The latter
were jointly awarded the Nobel prize in physics in 1965 for their work on quantum
electrodynamics.) In his paper [Dysl], Dyson introduced the time-ordered perturbation
series that now bears his name. (We also refer to [Dys2,4], especially to Dyson's own
comments in [Dys4, pp. 9-16], as well as to Schweber's description of Dyson's work on
QED in [Schwe3, Chapter 9].)
Feynman's work on the operational calculus for noncommuting operators is not
mathematically rigorous. We will briefly discuss some ways of achieving rigor in the
next section and then will pursue this topic in Chapters 15-19.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 394 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

14.4 Making Feynman's operational calculus rigorous

The work of Feynman referenced at the end of this book has had a fundamental impact on
physics and a significant impact on a number of developments in mathematics. However,
this work is not rigorous in the sense in which mathematicians commonly understand
this word. How can it be made mathematically rigorous? We have given some answers
to this question earlier with regard to the Feynman integral, especially in Chapters 11
and 13. In this section, we will indicate three ways of making Feynman's operational
calculus rigorous.
Feynman himself recognized that lending rigor to his operational calculus was not
likely to be easy. The reader might be surprised to read what he had to say about this.
(See the second quote from [Fey8] given at the beginning of this chapter.)
We note that in spite of our emphasis in this section on the problem of making
Feynman's operational calculus rigorous, this is not our only concern in the remain-
ing chapters of this book. Our goal in Chapters 15-19 is also to interpret and extend
Feynman's ideas.

I. Rigor via path integrals

We indicated in Section 14.3 that Feynman's operational calculus can be regarded as
generalizing some aspects of path integrals. We quote Schweber [Schwe2, p. 502] on
this point of view (but the remarks in parentheses are ours): "In any case Feynman never
felt order-by-order (the time-ordered integrals involved in the perturbation expansions
obtained from applying Feynman's operational calculus) was anything but an approxi-
mation to the 'thing' and the 'thing' was the path integral." In the following paragraph
[Schwe2, p. 502] Schweber quotes Feynman (taped interview, November 1980): "A way
of saying what quantum electrodynamics was, was to say what the rule was for the dia-
gram (or Feynman graph)—although I really thought behind it was my action form"
(and so, presumably, an associated path integral of some sort).
Since the nonrigorous operational calculus of Feynman is a kind of generalization
of Feynman's nonrigorous path integral, it seems plausible that the rigorous Feynman
and Wiener integrals could be used to obtain a rigorous version of the operational cal-
culus in the quantum-mechanical and diffusion (or probabilistic) settings appropriate to
the Feynman and Wiener integrals, respectively. Indeed, this is exactly what is done in
Chapters 15-18. Those chapters—which are based on work by Johnson and Lapidus
in [JoLal-4] (for Chapters 15, 16 and 18) as well as of Lapidus in [Lal3-18] (for
under U.S. or applicable copyright law.

Chapter 17)—will be discussed briefly in Section 14.5. We refer the reader to that sec-
tion for further introductory material concerning this approach and, in particular, for a
discussion of the noncommutative operations * and + on "disentangling algebras" of
Wiener functional.
Remark 14.4.1 (a) We have discussed in rather simple cases the use of measures other
than Lebesgue measure in connection with Feynman's operational calculus in Exam-
ples 14.2.1 and 14.2.3, Exercise 14.2.5, Remark 14.2.2 and formula (14.3.11). The flexi-
bility provided by such measures will allow us to substantially extend the operational cal-
culus as Publishing
EBSCO the reader will see
: eBook repeatedly
Academic in(EBSCOhost)
Collection connection with approach
- printed on 6/8/2017I5:46
in Chapters
PM via 15-18
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 MAKING FEYNMAN'S OPERATIONAL CALCULUS RIGOROUS 395
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and approach II (described below) in Chapter 19. A similar statement holds for approach
III, but that approach will be discussed only briefly in this book; namely, in III of the
present section.
(b) We have been emphasizing above that the Wiener and Feynman path integrals
will be used in Chapters 15-18 to make the operational calculus rigorous. However,
as was already mentioned, many of the results in those chapters have an interest as
contributions to the Wiener and Feynman integrals apart from their connection with the
operational calculus.
(c) Both the Wiener and Feynman integrals are intimately associated with the Lapla-
cian, the "generator" of Brownian motion. (See Example 9.2.3, Remark 9.2.4, and The-
orems 12.1.1 and 13.3.1 for earlier material related to this statement.) It seems clear
that other path integrals associated with different generators and Markov processes (see
[Fuk, FukOT, L2, MaRo, Or, Sh, StroVa, Va, Wil]) could be used in appropriate settings
to make Feynman's operational calculus rigorous. The only work that we know of which
is explicitly in this direction is due to Riggs [Rig]; it extends to the Dirac operator
(in one space dimension) some of the results of [JoLal,2] and [Lal5,16] discussed in
Chapters 15 and 17, respectively. The stochastic process involved in [Rig] is the Poisson
process with generator and semigroup as indicated in Example 9.2.5. The related par-
tial differential equations are the telegrapher's equation and the Dirac equation (both in
one space dimension) rather than the heat and Schrodinger equations as in [JoLal,2]
and [Lal4-18]. (See Remark 17.6.32(c).)

II. Well-defined and useful formulas arrived at via Feynman's heuristic rules
A common mathematical strategy is to use heuristic ideas to arrive at mathematical
"objects" which can then be rigorously defined and are useful in themselves. This is
the point of view adopted in Chapter 19, based on the work of DeFacio, Johnson and
Lapidus in [dFJoLal,2].
Much of the motivation for [Fey8] as well as for the papers [Fey5-7] which led up to
it was to find a method of calculation akin to path integration which allowed Feynman
to "derive" formulas for evolving physical systems. As we indicated earlier, he was
especially interested in the perturbation series for quantum electrodynamics where there
was no path integral available. Although Feynman's operational calculus extends well
beyond exponential functions, exponentials of sums of noncommuting operators will
play a central role in Chapter 19 as they did in [Fey8]. In fact, in Chapter 19 we will
under U.S. or applicable copyright law.

be considering exponentials of sums of integrals of operators as in (19.1.2). The special

emphasis on exponentiation comes from its connection with evolution equations. A
major part of the idea in Chapter 19 is to develop mathematical models for complicated
physical systems which combine features of simpler systems but are not readily treatable
by familiar methods. In mathematical terms, the rigorously defined disentangled formulas
for (19.1.2) will provide the unique solution to the evolution equation (in integral form)
(19.5.1).
As will be further discussed in Chapter 19, one advantage of this abstract approach
[dFJoLa2] is that it allows us, for example, to deal with nonlocal interactions—such as
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
nonlocal potentials given by suitable integral operators (instead of the traditional local
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 396 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

potentials given by multiplication operators). This is of interest in studying models used

in phenomenological nuclear physics [Tab, ChSa, Mc].
III. A general theory of Feynman 's operational calculus with computations
which are rigorous at each stage
We describe briefly here a part of recent work of Jefferies and Johnson [JeJo]. Fix
bounded linear operators A1, . . . , An on a Banach space X over the complex numbers
C. It is not assumed that A1, . . . , An commute. We form a commutative Banach alge-
bra ID>(A1 , . . . , An) which consists of certain analytic functions f(A\, . . . , An), where
AI , . . . , An are regarded as purely formal commuting objects. Within O(Ai , . . . , A n ),
called the disentangling algebra, we can, in the spirit of Feynman's second rule (see (2)
of Section 14.2) form functions of the 'operators' as though they were commuting. In
fact, the actual elements of D(A1 , . . . , An) do commute.
Now let u1, . . . , un be (for convenience) Borel probability measures on the time
interval [0, 1]. The measures u1 , . . . , un will be used to assign time indices to specify
the order of the operators in products (see (1) of Feynman's rules in Section 14.2).
Saying it another way, u1, . . . , un will specify directions for disentangling functions
f ( A 1 , . . , , An) of A1 , . . . , An (see rule (3) in Section 14.2). Disentangling maps

are defined and the resulting theory may be thought of as providing a family of functional
calculi for the operators A1 , . . . , An . A simple example of a theorem from [JeJo] is that if
AI, ... , An commute, then the functional calculi specified byTu1,...,^,with u1, ..., un
as above, all agree with each other and with the usual analytic functional calculus for
commuting operators.
The discussion so far gives some idea of the content of the first several sections of
[JeJo]. Those sections study the algebraic properties of Feynman's operational calculi
and it is only the norm and not the spectrum of the individual operators that comes
into play. In later sections, with additional information about the n operators involved,
a "joint spectrum" can be defined and the functional calculus can be enriched. The
operator-valued version of the Paley-Wiener theorem is one of the keys to this. Also, a
connection with the Riesz-Dunford functional calculus for a single operator is made by
appealing to the Cauchy integral formula from Clifford analysis for functions of n + 1
under U.S. or applicable copyright law.

real variables taking values in a Clifford algebra.

In [JeJo], the emphasis is on the algebraic properties of Feynman's operational calculi
as well as on forming rather general functions of the not necessarily commuting operators
A1, . . . , An. On the other hand, in order to encompass evolution problems within the
theory and, in the process, to make connections with II above and Chapter 19 below, it
is necessary to adjust the framework somewhat and to emphasize exponential functions
and operator- valued functions of time. Such work is in progress in a paper by Jefferies,
Johnson and Nielsen [JeJoN].
We finish here by mentioning the thesis of Lance Nielsen [Nie] which is near-
ingEBSCO
completion. Nielsen proves several stability theorems in the settings of either
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
[JeJo] or [dFJoLa2].
UNIVERSIDAD For example,
DISTRITAL FRANCISCO the first theorem in [Nie] says that if a sequence
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN' S CALCULUS VIA WIENER AND FEYNMAN INTEGRALS 397
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

{(u1 m , . . . , un,m) : m = 1, 2, . . . } of n-tuples of probability measures on [0, 1] con-

verges weakly to the n-tuple (u1, . . . , u n ) , then the sequence of disentangling maps
{Tu1m,...,unm :m = 1,2, . . . } converges to the disentangling mapTu1,...,unin the strong
operator topology on £(D(A1, . . . , An), £(X)). This conclusion can also be rephrased
as asserting that the sequence of functional calculi associated with (u1 m , . . . , u n m ) con-
verges to the functional calculus associated with (u1, . . . , u n ) .

14.5 Feynman's operational calculus via Wiener and Feynman integrals:

Comments on Chapters 15-18
We will give here a brief description of some key parts of Chapters 15-18. The precise
statement of definitions and theorems will be postponed, but we will try to convey the
spirit of some of the results. Our discussion of Chapter 15 will be the fullest both because
it is the essential starting point for the later chapters but also so that it can serve as part
of the introduction to the next chapter.

Chapter 75
Recall that Ct = C([0, t], Rd), for t > 0, and let F : C' -> C. The operator-valued
function space integrals A"|(F), A e C+, which we will use throughout the next four
chapters will be defined for A e C+ just as in Section 13.5. (Here, as before, C+ =
{A e C : Re A > 0} and C+ = {X e C : X ^ 0 and Re X. > 0}.) In particular, K[(F)
will be defined for A > 0 by the Wiener integral (13.5.2) in Definition 13.5.1 (or see
(15.2.8) below) and then for A. e <C+ by operator-valued analytic continuation (in mass).
However, we will require much more for A e C^ \ C+ in Chapters 15-18 than we did in
Definitions 13.5.1 and 13.5.1', especially the latter. Specifically, Definition 15.2.1 will
demand that K'^(F) be strongly continuous in A throughout C+.
Fix t > 0 for the present. The functions F : C' -> C that we will consider are
infinite sums of finite products of functions of the form

where n is a C-valued measure of finite total variation on (0, t) and 6 : (0, t) x Rrf —> C
belongs to Looi;>; (see parts F and G of Section 15.2). For our present purposes, one does
not lose much by thinking of n as a finite (nonnegative, Borel) measure on (0, t) and
regarding 6 as bounded. (In order to insure convergence in an appropriate sense of the
under U.S. or applicable copyright law.

infinite sums mentioned above, we will require (15.4.19).)

The collection At of functions (equivalence classes, actually) which we have just
described is larger than we will need for any one of the concrete problems that we will
consider. However, At has the advantage of being an algebra and of being complete
under an appropriate norm. In fact, At is a commutative Banach algebra with identity
which we call the disentangling algebra (or the disentangling algebra associated with
time t). (See Section 15.7.) The family of disentangling algebras (A, : t > 0} will play
a prominent role in Chapter 18, as we will discuss further on.
Given F e At, the function of operators that is to be disentangled can be identified
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
fromUNIVERSIDAD
F; we DISTRITAL
will giveFRANCISCO
an example
JOSE DE of this below. The disentangling is carried out by
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 398 INTRODUCTION TO FEYNMAN' S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

calculating the path integral K[(F). [Recall that K[(F) is a Wiener integral for X > 0
and a Feynman integral for A, € C+ \ C+ (that is, for A ^ 0 purely imaginary).]
The disentangled operator K[(F) is expressed as a generalized Dyson series (GDS). If
F(x) = e\p[F-iv,i(x)] (see (14.5.1)), where / is Lebesgue measure on (0, t) and V is
R-valued and time independent, then—except for an exact reversal of the time-ordering
(see Remark 15.3.7)—the GDS forKt-1i(F)is just the Dyson series (14.3.9). We will
see that the GDS can be quite simple or extremely complicated (as in Section 15.4).
Generalized Feynman diagrams (see Section 15.6) are a visual aid in keeping track of
the individual terms of a GDS, and they help in giving physical interpretations where
that is appropriate.
Next we take a simple example of F e At and identify in the notation of Sections 14.2
and 14.3 the function of operators being formed when K[(F) is calculated. Let

where / is a complex-valued function of one complex variable. We assume for conveni-

ence that / is entire. When the GDS for K'^(F) is calculated, the function of operators
being computed is in fact

where 0(s) denotes here the operator of multiplication by the function &(s, •). In the
notation of Sections 14.2 and 14.3, a(s) = Ho/A for every s € [0, t], and so t ( H 0 / X ) =
/Q a(s)ds, while B(s) = 0(s) for 0 < s < t. Hence, (14.5.3) can be written as

One may well ask how exp[-t (Ho/A)] is obtained from (14.5.2). It actually comes out
of the path integral which starts for A. > 0 with a Wiener integral. Indeed, the exponential
factor exp[—t(Ho/A)] for some A e C- will be involved in all the disentanglings in
Chapters 15-18. (If some appropriate stochastic process other than the Wiener process
under U.S. or applicable copyright law.

were used, then the infinitesimal generator of that process would replace the infinitesimal
generator — HO of the Wiener process.)
Various examples discussed in Section 15.5 are helpful in understanding the theory
and its relationship to other subjects; we mention in particular Examples 15.5.1, 15.5.3,
15.5.5, and 15.5.9.

Chapter 16
Every F e At is defined in terms of "potentials" and measures. It is natural to ask for
conditions under which the operators K[(F) depend continuously on these objects. We
consider such questions with respect to potentials in Section 16.1 and with respect to
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
measures in Section
UNIVERSIDAD DISTRITAL 16.2.
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN' S CALCULUS VIA WIENER AND FEYNMAN INTEGRALS 399
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The results of Section 16.1 are more satisfying but those of Section 16.2 are perhaps
more interesting. Example 16.2.9 is especially illuminating; it illustrates how the GDS
for a sequence {nm } of absolutely continuous measures can pass to the GDS for a discrete
measure as nm converges "weakly" to the Dirac measure ST, where r e (0, t).
Chapter 1 7
The focus of Chapter 17 is on the work of Lapidus [La 14- 18] in which he established a
"Feynman-Kac formula with a Lebesgue-Stieltjes measure" (in short, FKLS) and related
it to the solution of various associated integral and differential (evolution) equations. In
these same papers, Lapidus made connections with the operational calculus of Feynman
through a variety of topics that are discussed in Chapter 17 and which are related to
material in Chapters 15, 16, and 18.
The classical Feynman-Kac formula was the subject of Chapter 12. Recall that the
functionals dealt with there had the form

where 6 = - V with V time independent and R-valued

R- valued (see Theorem 12.1.1).
1.2.1.1 ).In
Incontrast,
contrast,
the functional
functionals that will interest us in Chapter 17 will be of the form

where 90 : [0, t) x W Rd1 -> C is allowed to be time-dependent and C-valued and 77 is a

C-valued Borel measure on [0, t] t ] (see (17.1.5)).
( 1 7. 1 .5)).InInterms
termsofofthe
therestrictions
restrictionsmentioned
mentionedsoso
far, the functionals in (14.5.6) form a more general class than those in (14.5.5). However,
the potentials V (= —6) in (14.5.5) were considerably more general in one important
sense; they were permitted to have strong singularities (see Theorem 12.1.1). In contrast,
the functions 6 in (14.5.6) are required to be in Looi-^ (see (17.1.3)) which implies, in
particular, that 0(s,
9(s, •) is essentially bounded for n-a.e.s.
It will be necessary in Chapters 17 and 18 to treat time t as a variable whereas it will
suffice to fix one value of t in Chapters 15 and 16. Actually, it is not quite the functions
t (-*•
-> Kt/a(FK,n)
K'^FK,^) that will solve the evolution equations (in integral or differential form)
in Chapter 17, but rather
under U.S. or applicable copyright law.

(14.5.7)

where FK,n is the time-reversal of the functional FK,n (see Definition 15.7.5 and
Theorem 15.7.6).
The functional FK,n belongs to the disentangling algebra At for every t > 0 by
Theorem 15.4.1 (with m = 1) and Corollary 15.7.4. Hence, K [ ( F K n ) exists and
can be disentangled via a generalized Dyson series (GDS) for every >. e C+ (see
Theorem 15.7.1).
The presence of the Lebesgue-Stieltjes measure n will allow us to consider a variety
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
of interesting specialFRANCISCO
UNIVERSIDAD DISTRITAL cases andJOSE to
DE blend
CALDAS continuous and discrete phenomena. [In the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 400 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

following, we let n = u + v denote the unique decomposition of 77 into its continuous

part u. and its discrete part v (see Section 15.2.F).] The case n — u, with u absolutely
continuous with respect to Lebesgue measure /, is not much different mathematically
from the case n = l itself. However, even within the absolutely continuous case, the
physical interpretations of the evolving operator u(t) (see (14.5.7)) can be drastically
different.
The fact that the measure n = u+v is allowed to have a nonzero discrete part v affects
the physical interpretation of the evolving operators in both the diffusion and quantum-
mechanical cases (see especially Section 17.5). The presence of v = 0 introduces phe-
nomena that play an important role throughout all of the sections in Chapter 17. The
emphasis is on finitely supported v, say v = X)p=i Mp^rp with 0 < TI < • • • < TJ, < ?
and Wp e C, since that case is more tractable and seems to be the most reasonable
physically. (See Sections 17.2-17.5, which discuss the results of [Lal5].) However, the
general case, v — Y^Li (OP&TP with £^Li cop\ < oo, is treated in Section 17.6. [Actu-
ally, some of the results from [Lal8,16] discussed in Section 17.6 will provide new
information even in the special case (treated in Sections 17.2-17.5) when v is finitely
supported. In the quantum-mechanical case (i.e. 9 = —iV and k = —i), this is espe-
cially true regarding the distributional form of the differential equation and a product
integral representation of the solution to the evolution equation, which are obtained for
an arbitrary measure 77; see Theorems 17.6.15 and 17.6.10, respectively.]
Equation (17.5.12) (or (17.2.11)) gives a striking formula for K[(F), where F =
FK,T} is the Feynman-Kac functional (14.5.6). In addition, a comparison of the time-
ordered expression for the functional F in (17.5.11) with the last expression in (17.5.12)
will eventually show us that the noncommutative operations which will be introduced
in Chapter 18 fit well with both the operational calculus and the Feynman-Kac formula
with a Lebesgue-Stieltjes measure.
The study, in conjunction with the Feynman integral, of exponential (and other ana-
lytic) functions of functionals of the form

where n is an arbitrary C-valued Borel measure on [0, t], seems to have begun with
[ChanJoSk2, Jo4]. However, in those papers the approach to the Feynman integral and
the assumptions on 9 were different than in Chapters 15-18. The Feynman integral
under U.S. or applicable copyright law.

was scalar-valued rather than operator-valued and transform assumptions were placed
on 6. The main assumption on 9 was that for every s in [0, t], 0(s, •) was equal to the
Fourier transform of a C-valued Borel measure on Rd. The "Fresnel integral" [AlHol]
also involved transform assumptions but otherwise appeared quite different. However,
the two theories were shown to be essentially equivalent in [Jol] and a little later but
independently and in greater generality in [KalBr].
It had been shown [JoSk10] under transform assumptions that the scalar-valued
analytic Feynman integral of a functional much like exp(F1), except with n in (14.5.6)
equal to Lebesgue measure l, satisfies the Schrodinger equation in integral form. Johnson
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
asked a former
UNIVERSIDAD graduate
DISTRITAL student
FRANCISCO JOSE if
DE aCALDAS
similar result holds for a general C-valued Borel
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN'S CALCULUS VIA WIENER AND FEYNMAN INTEGRALS 401
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

measure n. He also observed that the result follows easily from [JoSkl0] if n is absolutely
continuous with respect to l since, in that case, the Radon-Nikodym derivativedn/dlcan be
used to absorb n into the potential. The same issue came up while Johnson and Lapidus
were working on [JoLal]]. Although the setting was different, earlier work [JoSk6, p — 2
case] again easily took care of the absolutely continuous situation. Lapidus went on
to establish the much more complicated case of a general C-valued Borel measure n
and to ask and resolve many further related questions. The operator-valued setting of
these papers of Lapidus [Lal4-18] [and of the paper [JoLal] (discussed in Chapters 15
and 16) that helped lead up to them] provided a natural framework in which to study the
relationship between the Feynman-Kac formula with a Lebesgue-Stieltjes measure and
the operational calculus of Feynman.
Chapter 18
In Chapter 18 — which discusses joint work of the authors in [JoLa3,4] — we study two
noncommutative operations on Wiener functionals and their relationship with the disen-
tangling algebras {At : t > 0} and the operational calculus of Feynman. We begin with
a description of the noncommutative operations avoiding the technical details; a more
precise treatment will be postponed until Sections 18.2 and 18.3. The reader may wish
to review our brief comments on the disentangling algebra At at the beginning of this
section (or else see Section 15.7).
Let F e At1 and G 6 At2. Then F and G are C-valued functions on Ct1 and C'2,
respectively. The noncommutative product F * G is to be a C-valued function on C't1+t2.
Given x e C t1+t2 , (F * G)(x) := F(x1)G(X2), where x\ is the restriction of x to [0, t\]
and X2 is obtained by first restricting x to [t1 , t1 + t2] and then translating the restriction
to [0, t2]. In a similar manner, the noncommutative addition + is defined by the formula
(F+G)(x) := F(x\) + G(x2), for x e C t1+t2 . We will show in Theorem 18.5.3 that
F * G and F+G belong to Att+t2. Note that the operations * and + are not internal to
At but rather map At1 x At2 to At1 +t2 . As the reader may have noticed, these operations
are not limited to the disentangling algebras but make sense for any functions F and G
on Cl1 and Ct2 , respectively. The operations have various pleasant algebraic properties
as we will see in Sections 18.3 and 18.5.
In order to illustrate how the operation * is related to Feynman's operational calcu-
lus, we will describe the problem which initially led the authors (in [JoLa3,4]) to the
definition of *. Let F and G belong to At1 and At2, respectively. We know from our
under U.S. or applicable copyright law.

earlier discussion of Chapter 15 that for every y e C+~, both Kt1y (F) and Kt2y(G) are
disentangled by their generalized Dyson series expansions (GDS). It is natural to ask if
the operator product Kt1y (F)Kt2y(G) can also be disentangled. The answer is "yes". In
fact, Kt1y (F)Kt2y(G) has a GDS since

(See Theorem 18.5.6.) Note that the definition of F * G and the formula (14.5.9) both
involve time-ordering.
We will see in Theorem 18.4.1 that the relationship (14.5.9) is not restricted to the
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM viat1
algebras [At :DISTRITAL
UNIVERSIDAD t > 0}FRANCISCO
but extends rather general functionals F and G on C and Ct2,
to CALDAS
JOSE DE
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 402 INTRODUCTION TO FEYNMAN'S OPERATIONAL CALCULUS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

respectively. However, the interpretation of (14.5.9) is different in that case since there
is no reason to think that any of the three operators involved can be disentangled via a
GDS. It is then reasonable to regard the right-hand side of (14.5.9) as providing a partial
disentangling of Kt1 +t2 (F * G).
The formula

is used in [Fey8] even when A and B do not commute. Following Feynman, we used
essentially the same relationship in the second "equality" in (14.2.11) and also in (14.3.3);
of course, these "equations" cannot be interpreted in the usual way. Such steps are in the
spirit of rule (2) of Feynman (see Section 14.2) and are used to disentangle expressions
which involve the exponential of sums of operators. Is there a useful way to rigorously
interpret (14.5.10) without changing its form? In fact, given F E At1 and G E At2, we
will show in Theorem 18.5.9 that

Note that (14.5.11) is a relationship between functionals whereas (14.5.12) gives the cor-
responding operator equation. The formula (14.5.11) looks like the standard exponential
formula but involves the noncommutative operations + and * introduced in [JoLa3,4]
(and in Section 18.3). The proof of the operator equation (14.5.12) follows easily from
(14.5.11) and (14.5.9).
Several other "paradoxical formulas" found in Feynman's paper [Fey8]—as well as
new ones—can be given a rigorous interpretation in this framework. (See, in particular,
Example 18.5.12.)
Remark 14.5.1 An attempt to capture the essence of the algebraic and analytical struc-
tures underlying the construction [JoLal-4] (carried out in Chapters 15 and 18) of
({At}t>0, +, *)—the family of "disentangling algebras" equipped with the noncommu-
under U.S. or applicable copyright law.

tative operations * and + defined above—was made by the second author in [Lal9],
where a possible set of axioms for (parts of) Feynman's operational calculus was pro-
posed. (See Appendix 18.6.) The counterpart of the map F H» Kt y (F) was then viewed as
a "quantization map" defined via a kind of generalized (Feynman) path integral. The dif-
ficulty in thisframework is, of course, to construct such a map in each concrete situation of
interest in the applications [as was done in [JoLal—4] (see Chapters 15 and 18) in a set-
ting corresponding to ordinary quantum mechanics]. In addition, it is likely (but has not
yet been established mathematically) that the work in [dFJoLa2] (see Chapter 19) aug-
mented by that in [JeJo] (described respectively in parts II and III of Section 14.4 above)
provides further illustrations of the axioms proposed in [Lal 9] for Feynman's operational
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
calculus via noncommutative
UNIVERSIDAD operations
DISTRITAL FRANCISCO JOSE DE CALDAS acting on suitable "disentangling algebras".
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 FEYNMAN'S CALCULUS VIA WIENER AND FEYNMAN INTEGRALS 403
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We close this chapter by giving a list (certainly not complete) of references connected
with Feynman's operational calculus. These references are divided into four categories
but the boundary between these categories is not always clear cut. The reader may be
surprised at the diversity of this work.

(1) References using path integrals to make Feynman's operational calculus rigorous:
[AhJo, ChaJo, JoLal-4, JoP, Lal3-18, 20, Rig].
(2) References of a more abstract nature: [Alb, And, Ar, dFJoLal,2, Gil1,2, GilZa,
JeJo, JeJoN, JoKal5, KisRal,2, Lal9, Masll, NazSS, Ne3, Nie, Rey, Tay1].
(3) References in the physical literature: [AgWo, dWmMN2, Fuji, Klal, KouN,
Lo, Ya].
(4) A sample of further work in some way related to the operator calcu-
lus: [BayFFLS1,2, Ber, Chen1-3, Con1,2, DoFri1-3, DouRotS, FlSter1,2, Fli, Foil,
GelKLLRT, GelRet1,2, GrsLouSte, Je3, JeMcIPic, H6rl-4, KoNi, Kon2,4,8, LM1,2,
Moy, Pins, Pry 1,2, PrySol, Ric, Riel,2, Ster, Stri2, Tay2, Un1,2, Wei].
In particular, the book Operational Methods by Maslov [Mas11] and the papers by
Nelson [Ne3] and Araki [Ar] were early and influential works. Maslov's work had a strong
impact on and was further developed in the recent book Methods of Noncommutative
Analysis by Nazaikinskii, Shatalov and Sternin [NazSS].
Finally, we mention that time-ordered integrals appeared in the literature well before
Feynman's paper [Fey8]. Examples can be found in the work of Vito Volterra (see [Vol,
VolHos] and the earlier references therein). Jagdish Mehra [Me, §14.4(a), p. 192] gives
a brief description of Volterra's work and points out its relevance to both the functional
calculus and functional integration.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

15
GENERALIZED DYSON SERIES, THE FEYNMAN INTEGRAL
AND FEYNMAN'S OPERATIONAL CALCULUS

15.1 Introduction
The reader will recall that the first part of Section 14.5 introduced some of the ideas of
this chapter. The present section continues this introduction but with an emphasis on the
role played by the discrete part of the measures n that are involved.
The present chapter (and the next one) discusses work of Johnson and Lapidus in their
memoir [JoLal]. One of its main goals is to lay some of the foundations for Feynman's
operational calculus, and particularly for the "disentangling process" discussed in Section
14.2. As was already briefly explained in Section 14.5, we will achieve this in a con-
crete setting by using the Wiener and Feynman path integrals of suitable functionals
to represent the associated operators via time-ordered perturbation expansions, called
generalized Dyson series (GDSs), valid in the diffusion (or probabilistic) as well as in
the quantum-mechanical case.
Further, we will show that the set of Wiener functionals giving rise to these GDS—
modulo a suitable equivalence relation and a suitable choice of norm—forms a commu-
tative Banach algebra, At, called the disentangling algebra.
As will be seen throughout this chapter, these perturbation series possess a rich
combinatorial structure, owing to the systematic use (introduced in [JoLa1,2]) of both
discrete and continuous measures in the time-integration involved in the definition of the
appropriate Wiener functionals.
Fix t > 0. Recall from Section 14.5 that the commutative Banach algebra At (which
will be constructed in Section 15.7 below) consists of infinite sums of finite products of
functions of the form
under U.S. or applicable copyright law.

where n is a C-valued Borel measure on (0, t). Recall further from our discus-
sion of Chapter 17 in Section 14.5 that the function (14.5.3), that is, the function
x i->-
\-> ex.p(F0^(x)),
exp(F0,n(x)), is a particularly useful example of an element of At- At.
The measures nr\ as above can be quite diverse. (See Section 15.2.F below for basic
definitions and facts regarding these measures.) Every such nr\ can be uniquely written as
r\ = IJL
n u + v, where nrj is continuous and and v is discrete. The
The additional flexibility
flexibility provided
by the use of Lebesgue-Stieltjes measures nrj has many implications which allowed us
in [JoLal^4,
[JoLal-4, Lal4-18] to broaden and unify known concepts and to introduce new
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
onesUNIVERSIDAD
having an interest
DISTRITAL in their
FRANCISCO JOSEown right. When r\
DE CALDAS n = un is a continuous measure, the
generalized
AN: 98476 ;Dyson
Johnson, series (GDS)
Gerald W., associated
Lapidus, Michel L..;with the functional
The Feynman exp(Fe
exp(Fo,u(x))
Integral and tM (x)) has the
Feynman's
Operational Calculus
Account: ns000601
 INTRODUCTION 405
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

same formal appearance as in the classical case (see (14.3.9)). However, even when 77 = u
is absolutely continuous with respect to Lebesgue measure, very different interpretations
may be suggested. Cases with 77 = u continuous but singular with respect to Lebesgue
measure / are interesting and quite different (see Example 15.5.3).
When n has a nonzero discrete part, the form of the GDS changes markedly and
strikingly different phenomena occur. The combinatorial structure of the series is much
more complicated even when v is finitely supported. For instance, additional summa-
tions appear as well as powers of the potential 9 evaluated at fixed times. Some of the
combinatorial complications and nearly all of the analytic difficulties are found in the
(relatively) simple case n = u + wdT, where dT is the Dirac measure at r. Accordingly,
we discuss this prototypical example in detail in Section 15.3 (and later in Section 17.3)
and use it as a conceptual aid to the general development.
Even the case when n = v is a purely discrete measure with finite support, i.e.

is of interest. By considering exp(F0,n) with F0,n Fo,n given by (15.1.1) and further specializ-
ing, we obtain apartial
a partial product (see (15.5.19)) as in the theory of the product integral. Still
more specializing yields the hth Trotter product. Connections with the (time-ordered)
product integral and the Trotter product formula will be discussed in Section 17.6 (espe-
cially in Theorems 17.6.10 and 17.6.12) and Example 16.2.7, respectively.
Our GDSs can be represented graphically by generalized Feynman diagrams. The
nth term of the classical Dyson series (see (14.3.9) and (15.3.22)) corresponds to a
single connected Feynman diagram. The same assertion is true if Lebesgue measure
/ in the classical Dyson series is replaced by any continuous measure n = u. (See
Figure 15.6.1 and the discussion which precedes it.) However, when n = u + v with
the discrete part v different from 0, then the nth term of the CDS GDS gives rise to many
disconnected components, one for each summand. Figures 15.6.2-15.6.4 and the related
discussion all deal with the case n = u + wdt, wdt, whereas Figure 15.6.5 deals with the case
n=m u +Eh Ehp=
p=11wpdtp.The complex combinatorial structure of the GDS is accurately
reflected in the generalized Feynman graphs, and the reader may find it helpful, after a
brief look at Section 15.6, to draw such graphs while following the proofs and examples
under U.S. or applicable copyright law.

in Sections 15.3-15.5.
A great variety of Feynman diagrams and perturbation expansions appear in the
physics literature. We should make it clear that we do not claim to be generalizing all of
these.
The combinatorial complications that are introduced when n has a nonzero dis-
crete part are associated with the inappropriateness of the third part of Feynman's time-
ordering convention (14.2.1). (See equation (14.2.11) for such measures.) For example,
if n ( { T } ) = 0 and 0 < r < t, then (n x n ) ( { ( Tr , r)}) = = n ® 22 ( { r } ) = 0 and so the
diagonal cannot be ignored in integrals over (0, t) x (0, t) with respect to n x n]. n. (See
Figure 14.2.1. Recall that the diagonal could be omitted in the earlier calculations (14.3.7)
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
andUNIVERSIDAD
(14.3.17).)DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 406 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A striking instance of the distinct roles played by the continuous and discrete parts of
the measure n will be found in Chapter 17 which deals with the Feynman-Kac formula
with a Lebesgue-Stieltjes measure [Lal4-18]. See, for example, Theorem 17.2.7 and
part (a) of the remark which follows it.
In our experience, Feynman's paper on the operational calculus [Fey8] is, at least
initially, difficult to read; perhaps more difficult for mathematicians than his celebrated
paper [Fey2] on the "Feynman path integral". The operational calculus paper is rich in
ideas, but is also in need of clarification and mathematical development. Indeed, Feynman
himself wrote [Fey8, p. 108] about his operational calculus: "The mathematics is not
completely satisfactory. No attempt has been made to maintain mathematical rigor. The
excuse is not that it is expected that rigorous demonstrations can be easily supplied.
Quite the contrary, it is believed that to put the present methods on a rigorous basis may
be quite a difficult task, beyond the abilities of the author."
The above quote may be surprising to the reader familiar with some of Feynman's
negative statements regarding mathematics as reported in the press or printed in some
of his own books or articles. However, it is in complete accord with many statements
made by Feynman during private conversations with the second author on the need for
the development of a mathematical theory of his operational calculus.
It is our hope that Chapters 14-19 of this book will contribute to the understanding and
rigorous development of Feynman's operational calculus for noncommuting operators.
We now briefly describe the organization of the remainder of this chapter:
The next section provides notation, facts and definitions that will be needed in
Chapters 15-18, with an emphasis on the operator-valued analytic (in mass) Feynman
integral. The section will close with three lemmas which will be used frequently in
Chapters 15-18.
In Section 15.3, we discuss the prototypical example n = u + wdt mentioned above;
our most detailed analytic proofs are given in this case.
Generalized Dyson series for the full class of functionals treated in this chapter are
obtained in Section 15.4. The reader should note the main results of this section but may
wish to skip the proofs, at least on a first reading.
Section 15.5 may be particularly helpful to the reader as it deals with a variety of
concrete examples of perturbation expansions. Much of the emphasis in this section is
on the combinatorics.
We present in Section 15.6 a graphical representation of our generalized Dyson series
under U.S. or applicable copyright law.

in terms of generalized Feynman diagrams.

In Section 15.7, we show that the general class of functionals treated in Section 15.4
forms a commutative Banach algebra, and we discuss the related functional calculus.
We also define the "time-reversal map" and study it in the context of this "disentangling
algebra". The disentangling provided by the path integrals precisely reverses the natural
physical ordering (of the GDS), which the time-reversal map restores. Finally, the last part
of this section discusses some of the connections with Feynman's operational calculus.
Possible physical interpretations are provided in various places throughout the
chapter.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE ANALYTIC OPERATOR-VALUED FEYNMAN INTEGRAL 407
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We point out that some of the material in this chapter is new (and, in particular, was not
contained in [JoLal]). The most essential addition is the definition of the time-reversal
map T (Definition 15.7.5) and a careful discussion of its properties (Theorem 15.7.6
and Corollary 15.7.8). (See also Remarks 15.3.7 and 15.4.4, as well as Example 15.3.8,
leading to the definition of T.) We also mention that an interesting explicit example of
the computation of the operator Kty(F), in the case where the measure 77 = u is purely
continuous but singular, is provided in Example 15.5.3.

15.2 The analytic operator-valued Feynman integral

In A-I below, we recall some facts and introduce much of the notation which we will
need in this chapter as well as in Chapters 16-18. With the exception of G and I, we
suggest that the reader go over the material quickly and then return to it if and when it
is necessary.
First we mention some general references: For the theory of the Wiener process and
applications of path integration that supplement our earlier discussion in Chapters 2-5
as well as 12, the reader may wish to consult [GliJa, Si9, Va]. For semigroup theory, we
refer to Chapters 8 and 9 for the basic developments, as well as to Chapters 10 and 11
for related facts. For the theory of the Bochner integral, we refer to the treatise of Hille
and Phillips [HilPh, Chapter III] where the essential facts about this analogue of the
Lebesgue integral for Banach space-valued functions are given in relatively few pages.
Finally, the basic facts of measure theory used in this chapter can be found in [ReSil,
§1.3 and §1.4, pp. 12-26] and [Cho3, Coh, Fol2, Roy, Ru2, WhZy],

Notation and definitions

A. C, C+, C+~ : These denote, respectively, the complex numbers, the complex num-
bers with positive real part, and the nonzero complex numbers with nonnegative real part.
B. L2(Rd): The space of Borel measurable, C-valued functions u on Rd such that
|u| is integrable with respect to Lebesgue (abbreviated Leb.) measure on Rd.
2

C. L°°(Rd): The space of Borel measurable, C-valued functions on Ra which are

essentially bounded.
[It is the presence of a wide variety of Borel measures on (0, t) that leads us to work
in B and C with Borel (rather than Lebesgue, as in the previous chapters) measurable
functions on Rd. Further, as usual, the elements of L2(Rd) and L°°(Rd) are equivalence
under U.S. or applicable copyright law.

classes of functions, with u1 and u2 said to be equivalent if they are equal almost
everywhere (a.e.) with respect to Lebesgue measure on R d .]
D. L(L 2 (R d )): The space of bounded linear operators from L2(Rd) to itself.
The notation || • || will be used both for the norm of vectors and for the norm of
operators; the meaning will be clear from the context.
E. The semigroup exp(—ZH0): We recall from Section 10.2, especially Theorems
10.2.5-10.2.7, some facts which we will use frequently concerning the holomorphic
semigroup {exp(—zH0)}zEc~+ generated by the "free Hamiltonian" H0 = — (1/2)A =
-(l/2)Eda=1 62/9x2aacting on L2(Rd). (Also, see [Kat8, §IX.1.8, pp. 495-97] or
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
[ReSi2, Example
UNIVERSIDAD 5, p.FRANCISCO
DISTRITAL 254].) We
JOSE use notation convenient for our purpose. The operators
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 408 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

{exp[-s(H0/y)] : s > 0, A E C~+} are all in £(L2(Rd)) and satisfy:

In fact, when y E C~+is purely imaginary, exp[—s(H0/y)l is a unitary operator. As a

function of y, exp[—s(Ho/y)] is analytic in C+ and continuous in the strong operator
topology (or strongly continuous) in C~+. (This follows, for example, from the trivial
case V = 0 of Theorem 13.3.1.) Recall from Remark 11.7.2(c) (or 13.2.2(c)) that
for operator-valued (or for vector-valued) functions, the natural notions of analyticity
coincide. See [HilPh, §3.10, esp. Theorem 3.10.1, p. 93].) Next, we state a familiar
explicit formula for the operator exp[-s(Ho/y)]. (See formulas (10.2.25), (10.2.26) and
(10.2.33).) Given V € L 2 (R d ) and X e C~+,

The integral in (15.2.2) exists as an ordinary Lebesgue integral for X e C+, but when
y is purely imaginary and U is not integrable, the integral should be interpreted in the
mean (equation (10.2.33)) just as in the theory of the Fourier-Plancherel transform. (See
Theorems 10.2.5-10.2.7.)
As is well known, the (negative) normalized Laplacian HO is the generator of the
Brownian motion on Rd: It follows, in particular, that the semigroup {exp(-s H0) : s >:
0} is intimately connected with Wiener measure m defined in Chapter 3 and used in I
below. (See, e.g., Chapters 3 and 12, as well as Example 9.2.3.)
F. The measure space M ( 0 , t): Let t > 0 be fixed. In the following, M(0, t) will
denote the space of complex Borel measures n on the open interval (0, t). For information
on such spaces of measures, see, for example, [Coh, Chapter 4] or [Ru2, Chapter 6].
Given a Borel subset B of (0, t), the total variation measure \n\ is defined by \n\(B) =
SUP{Enj=1 |n(Bj)|)' where the supremum is taken over all finite partitions of B by Borel
sets (see [Coh, p. 126] or [Ru2, p. 116]); of course, \n\ is a positive measure in M(0, t).
In fact, if n is complex-valued, the measure |n| is a finite measure. (See [Ru2, Theorems
6.2 and 6.4, pp. 117 and 118].) Under the natural operations, M(0, t) is a Banach space
when equipped with the norm
under U.S. or applicable copyright law.

the total variation of n.

A measure u in M ( 0 , t) is said to be continuous if u ( { r } ) = 0 for every r in (0, t).
In contrast, v in M(0, t) is discrete (or is a "pure point measure" in the terminology of
Reed and Simon [ReSi1]) if and only if there is an at most countable subset {TP}8p=1 of
(0, t) and a summable sequence {w p }8p=1 from C such that

where dTp is the Dirac measure with total mass one concentrated at rp [Coh, p. 22]. Every
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
measure n e DISTRITAL
UNIVERSIDAD M(0, t) FRANCISCO
has a unique
JOSE DEdecomposition,
CALDAS n = u + v, into a continuous part u
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE ANALYTIC OPERATOR-VALUED FEYNMAN INTEGRAL 409
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and a discrete part v [ReSi1, Theorem I.13, p. 22]. We will make frequent use of such
decompositions.
Results for complex measures can usually be reduced to the case of positive measures
by using the formula ([Ru2, Theorem 6.12, p. 124]) n(dt) = e i g n ( t ) | n | ( d t ) , where gn(t)
is a real-valued measurable function of t and |n| is the positive total variation measure
defined above. The key to the proof is the fact that n is absolutely continuous with respect
to |n | so that a Radon-Nikodym theorem can be applied. The formula above can often
be used directly, but it is sometimes helpful to write

and then to decompose each of the cosine and sine functions into the difference of
their positive and negative parts. The result is that 77 is expressed as a (simple) linear
combination of four positive measures.
Unless otherwise specified, we work with the space M(0, t) throughout, but M[0, t)
could be treated without any essential complications. However, allowing n to have
nonzero mass at 0 introduces additional alternatives which, with only a few exceptions,
we have chosen to avoid.
Finally, we remark that the product of measures m1, . . . ,uk (whether in M(0, t) or
not) will be denoted u1 x . . . x uk or xku=1 uu. If u1 = . . . = uk = u, we will write
uxk instead. (See Remark 15.2.10 below for more details.)
G. The mixed norm space (Lool;n, || . ||oo1;n): Let n e M(0, t). A C-valued, Borel
measurable function 8 on (0, t) x Rd is said to belong to Loo1;n if

Note that if 9 e L-oo1;n, then 0(s, .) must be in L°°(Rd) for n-a.e. s in (0, t). If one
makes the usual identification of functions which are equal n x Leb — a.e., the mixed
norm space Loo1;n, equipped with the norm || . ||oo1;n, becomes a Banach space. Note
that all bounded, everywhere defined, Borel measurable functions on (0, t) x Rd are in
Loo1;n for every n in M(0, t). (Here, as before, "Leb." denotes Lebesgue measure on R d .)
The reader will see further on that the norm (15.2.5) appears in our estimates in a
natural way.
The functions 9 will be interpreted physically as potentials. The condition that 6 be
under U.S. or applicable copyright law.

in Loo1;n is rather minimal in most respects. No smoothness is required, and 9 is allowed

to be time-dependent and C-valued. The use of C-valued functions 0 will enable us, in
particular, to treat simultaneously the diffusion case and the quantum-mechanical case;
that is, in the language often used by physicists, the "imaginary time" and "real time"
case, respectively. (See Remark 15.2.6(b).) The importance of C-valued potentials in
the study of decay systems in quantum-mechanics is discussed thoroughly in the book
of Exner [Ex]. (See also Sections 11.6 as well as Sections 13.5 and 13.6.) Certainly,
the most serious restriction in our assumptions is that 0(s, .) be essentially bounded for
n-a.e. s. (See Remark 15.2.6(a).) However, even this condition seems quite reasonable
in light of our goal of obtaining rigorously justified perturbation series valid in the
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
quantum-mechanical case. JOSE DE CALDAS
UNIVERSIDAD DISTRITAL FRANCISCO
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 410 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

If 9 e Loo1;n and if n — u + v is decomposed into its continuous and discrete parts,

then it is not difficult to show that 0 e Loo1;u n Loo1;v and

H. The multiplication operators 0(s): We remind the reader that the operator of mul-
tiplication by a (C-valued) function in L°°(Rd) belongs to L(L2(Rd)) and has operator
norm equal to the essential supremum of the function. (See the extension of Propositions
10.1.1 and 10. 1.3 stated in Remark 10.1.4; see also, e.g., [Kat8, Example 2.11, p. 146].)
For us, the L°° -functions that arise will be of the form 0(s, .), where 0 E Loo1;n. It
will be convenient to let 0(s) denote the operator of multiplication by 0(s, .), acting on
L 2 (R d ). According to Remark 10.1.4, 9(s) is then a bounded (normal) operator and its
operator norm ||0(.s)|| satisfies

The analytic (in mass) operator-valued Feynman integral Kty(.)

Recall from Section 3.1 (and, e.g., Chapters 12 and 13) that Ct = C([0, t ] , R d )
denotes the space of Rd-valued continuous functions x on [0, t]. Further recall that
Ct0 = Co([0, t], Rd) consists of those functions x in Ct such that x(0) = 0. The space
Ct0is, as earlier, equipped with d-dimensional Wiener measure m, a probability measure
which is just the product of d one-dimensional Wiener measures (see Chapter 3).
I. The operator-valued function space integrals Kty(F), y e C~+:
Definition 15.2.1 Fix t > 0. Let F be a function from Ct to C. Given X > 0, u e
and E e Rd, we consider the expression

The operator- valued function space integral Kty(F) exists for y > 0 if (15.2.8) defines
Kty((F) as an element of L(L 2 (R d )). If, in addition, Kty(F), as a function of y, has an
extension (necessarily unique) to an analytic function on C+ and a strongly continuous
function on C~+, we say that Kty(F) exists for y e C~+.When X is purely imaginary,
Kty(F) is called the analytic (in mass) operator-valued Feynman integral of F.
under U.S. or applicable copyright law.

Remark 15.2.2 (a) Our notation for Kty(F) reflects the fact that later on, t will
be allowed to vary (see Chapters 17 and 18). Throughout the present chapter (and
Chapter 16), the time parameter t will be fixed.
(b) The analytic Feynman integrals considered throughout the rest of this book (specif-
ically, in Chapters 15-18) will be given as in Definition 15.2.1 and so we will always
omit the words "in mass" when referring to Kty(F). Note that the parameter y can be
expressed in terms of either the mass parameter or the diffusion constant. (See Section
13.5, especially Remark 13.5.2(c).)
We stress that the definition adopted here is much stronger than that used in
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
Section 13.5,DISTRITAL
UNIVERSIDAD where only a "nontangential
FRANCISCO JOSE DE CALDAS limit" of y i-> Kty(F) was assumed to exist
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE ANALYTIC OPERATOR-VALUED FEYNMAN INTEGRAL 411
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

along the imaginary axis. (Compare Definitions 13.5.1' and 15.2.1.) Hence, when Kty (F)
exists (for all y e C~+) in the present sense of Definition 15.2.1, then it will certainly
exist in the weaker sense of Definition 13.5.1'. (See the comments following Definition
13.5.1'.)
(c) The function F in Definition 15.2.1 (often referred to as a "functional" in the
physics literature), need not be everywhere defined; however, in order to have Kty(F)
defined for all A > 0, it must be the case that, for every y > 0, F (y - 1 / 2 x + £ ) is defined
for m x Leb.-a.e. (x, £) € Ct0 x Rd.
Definition 15.2.3 Given functions F and G defined (in the sense of Remark 15.2.2(c)
above) on Ct, we say that F is equivalent to G and write F ~ G if, for every y > 0,
F(y- l / 2 x + E) = G(y- 1 / 2 x+E) for m x Leb.-a.e. (x, £) e Ct0 x Rd.
We leave it to the reader to check the following two simple facts:
(i) The relation ~ just introduced above is an equivalence relation.
(ii) The algebraic operations of pointwise addition, pointwise multiplication, and
scalar multiplication are compatible with the relation ~. For example, if F1 ~ F
and G1 ~ G, then F1 + G1 ~ F + G.
This equivalence is necessitated by the pathology of Wiener measure under scale
change and the fact that infinitely many scale changes (corresponding to all y > 0) are
involved here. These matters were discussed in detail in Chapter 4 for the scalar-valued
analytic Feynman integral. Although the setting here is somewhat more complicated, the
essential difficulties are not increased by the presence of £ e Rd. We illustrate this by
converting Example 4.5.3 to the present operator- valued setting. Some of the notation
and facts from Chapter 4 will be useful to us here.
Example 15.2.4 Let G : Ct -> C be identically 0. Then for every y > 0 and u E
), we have

Thus Kty (G) is the 0 operator for every y > 0. Certainly thenKty(G) exists in the sense
of Definition 15.2.1 and equals the 0 operator for every y E C~+.
Next fix d0 > 0 such that d0 = 1 . Let F : Ct —> C be given by
under U.S. or applicable copyright law.

where £la is given by (4.2.1) for every a > 0 (with b — a = t in this case) and xna
denotes its characteristic function. The function F is Borel measurable since Qdo is a
Borel set by Proposition 4.2. l(i). For x € Ct0 and £ E Rd, we have F(x + £) = xndo (x).
Hence F(x + £) is a Borel measurable function of the two variables (x,E) €Ct0 x Rd.
(In fact, F ( x + E ) is independent of E.) Since m ( £ 1 ) = m1 ( £ 1 ) = 1 and Q1 n $d0 = 0
by Proposition 4.2.1, we see that F(x + i- ) = Xfi do (x) = 0 m1 x Leb.-a.e. (In fact, more
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
is true than we
UNIVERSIDAD need FRANCISCO
DISTRITAL here; forJOSE
every E e Rd, F(x + E) = 0 m1-a.e.) Thus F(x + E )
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 412 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and G(x + £) are equivalent functions on Ct0 x Rd, in the sense of equality m1 x Leb.-
a.e. Nevertheless, we claim that F fails to have an operator-valued analytic Feynman
integral.
For (x, £) E Ct0 x Rd, we have by Proposition 4.2.1(ii) that y-1/2x e fid0 if and
only if x e y1/2od0 = £y1/2 d 0 . Thus

Therefore, F(y- 1/2 x + £) = 1 for m1 x Leb.-a.e. (x, £) if and only if y1/2d0 = 1 or

= d 0 -2 .For y = d0-2, F(y- l/2 x + E) = 0 for m1 x Leb.-a.e. (x, £). Hence, for y > 0,

Recall from (9.2.15), a simple consequence of Wiener's integration formula (3.3.9),

that

for every y > 0 and every U e L 2 (R d ).

Hence, from (15.2.10) and (15.2.11), we see that for all y > 0,

Clearly then Kty(F) cannot be analytically continued to C+.

In summary, the function (x,E) i-»- F(x + £) is Borel measurable and equivalent to
0 in the sense that it is equal to 0 m1 x Leb.-a.e. on Ct0 x Rd. However, Kty(F) fails to
exist at every A 6 C~+ \ (0, oo). This concludes Example 15.2.4.
A family of examples much like the above is described in Corollary 31, page 172
of [JoSk7]. Further related information can be found on pages 171-174 of the same
reference.
It is easy to show that the more refined equivalence relation ~ given in Definition
under U.S. or applicable copyright law.

15.2.3 has the properties which we need. The simple theorem which we are about to state
is the analogue for the operator-valued analytic Feynman integral Kty(F) of our earlier
result, Theorem 4.5.7, 1/2
for the scalar-valued
-1/2x analytic Feynman integral.

Theorem 15.2.5 Let F and G be C-valued functions on Ct such that F ~ G; that is,
for every A > 0, F(y- x + £) = G(y + E) for m x Leb.-a.e. (x, E) E Ct0 x Rd.
Suppose that the operator-valued integral Kty(G) exists for every y E C~+. Then Kty (F)
exists for every y e C~+ and we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
for UNIVERSIDAD
every y EDISTRITAL
C~+. FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE ANALYTIC OPERATOR-VALUED FEYNMAN INTEGRAL 413
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The simple proof is much like that of Theorem 4.5.7 and is left to the reader.
Definition 15.2.1, given in [JoLa 1,2], is a variation of a definition given by Cameron
and Storvick [CaSt1] and earlier, in the exponential case which has traditionally been
of most interest, by Nelson [Ne1]. For A purely imaginary (the Feynman case), the
requirements in Definition 15.2.1 for the existence of Kty (F) are more stringent than
the requirements in either of [CaSt1] or [Ne1]. (See Section 13.5.) Hence, when Kty (F)
exists in our sense, it will certainly exist in the sense of [CaSt1]. The "integral" intro-
duced in [CaSt1] has been studied in several later papers including, for example,
[JoSk1,2, JoSk6]. The interested readers may wish to check some of the references
in [CaSt1, JoSk2, JoSk6]. They should also note the differences in notation between this
book, [JoLa1,2], and earlier papers such as [CaSt1, JoSk2, JoSk6].
A reader not familiar with the difficulties involved in defining the Feynman integral
may wonder why the Wiener integral in (15.2.8) is not used to define Kty (F) for all
y E C~+. We do not wish to go into this in detail but remark that formula (15.2.8) can
be rewritten with A appearing as the scaling m o y1/2 of the measure m rather than in
the argument of the functions in the integrand. (See the first equality in (4.3.1).) The
problem associated with using (15.2.8) to define Kty (F) for y E C~+\ R are then due
to the fact that scaled Wiener "measure" is not countably additive for nonreal scalings,
a fact discussed in some detail in Section 4.6 above that dealt with the nonexistence of
Feynman's "measure". (See especially Theorem 4.6.1.)
Remark 15.2.6 (a) In order to avoid possible misunderstandings, we mention that the
theory in Chapters 15-19 is different in spirit and in purpose from the approaches to
the Feynman integral discussed earlier in this book, especially in Chapter 11 (Sections
11.2 and 11.3-11.6) and Chapter 13. In particular, as observed in G above, no attempt
is made here to treat very singular potentials as was done in those earlier chapters. On
several occasions, however, we shall take advantage of some of the earlier results and
techniques.
(b) The physical interpretations that we will emphasize refer to the quantum-
mechanical case, i.e. to y purely imaginary. We stress that the standard quantum mechan-
ical case corresponds to 0 = —iV, with V real-valued, as well as A = —i (i := \/-1).
In contrast, the diffusion (or probabilistic) case corresponds to 9 = —V and y = 1. This
flexibility will enable us to treat the quantum-mechanical case and the diffusion case in
parallel.
under U.S. or applicable copyright law.

Preliminary results
We finish this section with three lemmas. We begin with a simple lemma that we will
frequently need.
Lemma 15.2.7 Let J be an interval in R and let K1, . . . , Kk be continuous
d-finite measures on (J, B(J)). Then the following sets have K1 x . . . x KK measure 0:
(i) The subset of Jk where two or more coordinates are equal.
EBSCO Publishing : eBookk Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
(ii) The subset of J where one or more coordinates has a fixed value.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 414 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof We just prove (i) since the proof of (ii) is similar and slightly easier. Clearly it
suffices to show that

If we section the set involved in (15.2.13) at an arbitrary point (S2, . . . , Sk) E J k - 1 ,

we obtain the single point set s\ = s2 in J. But this set has K1-measure 0 since K1 is
continuous. Now if we integrate the K1-measure of such sections over Jk-1 with respect
to K2 x . . . x Kk, we will get 0. Thus (15.2.13) follows from the Fubini theorem.
We continue with a somewhat technical measure-theoretic result which will be often
used throughout Chapters 15-18, most of the time without explicit mention. The reader
should note the result but may wish to skip the proof at least initially.
Lemma 15.2.8 Let n E M(0, t) and suppose that 9 e Loo1;n. Let

for any y e C' for which the integral exists. Then, for every y > 0, F1 (y -1/2 x + E) is
defined and satisfies

for m x Leb.-a.e. (x, E) e Ct0 x Rd.

Proof We first show: (*) For every y > 0 and m x Leb.-a.e. (x, E), 0(s, y - 1 / 2 x ( s ) + E)
is defined and satisfies |0(s, y-1/2x (s) + E)| < ||0(s, .)||oo for |n|-a.e.s.
Let Hy : (0, t) x Ct0x Rd -> (0, t) x Rd be defined by Hy (s, x, E) = (s, y - 1 / 2 x(s) +
E). The map Hy is everywhere defined and continuous and so 0 o Hy is certainly Borel
measurable though it need not be everywhere defined. Let

N :={(s, v) € (0, t) x Rd : 6(5, v) fails to be defined

or£(s, u)isdefinedbut|0(s, u)| > ||0(s, .)||oo}.

Since 0 E Loo1;n, a part of the Fubini theorem [Coh, Chapter 5] assures us that 9 is
under U.S. or applicable copyright law.

defined and satisfies 0 ( s , v)| < ||0(s, .)||oo for |n| x Leb.-a.e. (s, v); i.e. N is |n| x Leb.-
null. Let y > 0 be given. Note that to establish (*), it suffices to show that Hy-1 (N) is
|n| x mx Leb.-null. Accordingly, we section H-1y(N) at (s, v) € (0, t) x Rd:

where N(s) := [u e Rd : (s, u) e N}.

Now, since N is |n| x Leb.-null, it follows that
EBSCO Publishing : eBook Academic Collection (EBSCOhost)
for |n|-a.e.
- printed
s and every E, and so
d 1/2on 6/8/2017 5:46 PM via
certainly for DISTRITAL
UNIVERSIDAD |n| x Leb.-a.e. (x,JOSEe)DEeCALDAS
FRANCISCO Ct0 x R , the set y [N(s) - E] is Leb.-null. But
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE ANALYTIC OPERATOR-VALUED FEYNMAN INTEGRAL 415
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

it is well known and easily follows from Wiener's integration formula (3.3.9) that the
set of Wiener paths x whose value at a particular time s lies in a Leb.-null set is a set
of m-measure zero. (See Lemma 12.1.4.) Hence, by the Fubini theorem [Coh, Theorem
5.2.2, pp. 159-160], H-1y (N) is |n| x mx Leb.-null. Thus (*) is established.
It now follows from (*) that for every y > 0 and mx Leb.-a.e. (x, E),

Hence, for every y > 0 and m x Leb.-a.e. (x, E),

is defined and we have

This concludes the proof of Lemma 15.2.8.

We remark that even if 9 is everywhere defined, there may be continuous functions
y in C' for which F1 (y) fails to be defined. If 9 is everywhere defined and bounded,
then F1 (y) is defined for all y in C1. Note that even when 0 is essentially bounded, the
graph of the function s i-» y - 1 / 2 x ( s ) + E could lie in {(s, v) € (0, t) x Rd : |0(s, v)| >
||0(s, .)||oo) for some y > 0, x and E. In this case, (15.2.15) could fail since the second
inequality in (15.2.16) could fail.
Our last lemma does not seem to be explicitly stated in the standard references on
the Bochner integral. We include its easy proof.
Lemma 15.2.9 (Bochner integrals depending on a parameter) Let X be a complex
Banach space. Let (E, y) be a a-finite measure space and let T be a metric space.
Consider the function g : T x E —> X [or L ( X ) ] . Assume that/or all y in T, g ( y , y) is
under U.S. or applicable copyright law.

a strongly measurable function of y in E. Suppose further that there exists h in L1 (E, y)

such that ||g(y, y)|| < h(y) for y-a.e. y € E and all y e T. Set

(Note that G is well-defined by the basic Bochner integrability criterion; that is, Bochner
measurabitity of a function and integrability of its norm implies its Bochner integrability
[HilPh, Theorem 3.7.4, p. 80].)
(1) Assume that for y-a.e. y € E, g(y, y) is a strongly continuous function of y e T.
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
Then GDISTRITAL
UNIVERSIDAD is strongly continuous
FRANCISCO on T.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 416 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(2) Assume that T is an open subset of C and that for y-a.e. y E E, g ( y , y) is an

analytic function of y on T. Then G is analytic on T.
Proof If g(y, y) is operator- valued, we consider the vector-valued function
for fixed U in X; so that we may assume that g is X-valued.
Part (1) is a consequence of the dominated convergence theorem for Bochner integrals
[HilPh, Theorem 3.7.9, p. 83] which is the exact counterpart of the usual theorem for
scalar-valued functions.
Let X* denote the dual space of X and {•, •) the duality bracket between X and X*
(see, e.g., Definition 9.4.5). Fix U* in X*. Recalling our earlier remark in Section 15.2.E
about the equivalence of all the natural notions of analyticity, we see that (2) will follow
if we show that G1(y) := {U*, G(y)} is analytic on T. Let g1(y, y) = (U*, g(y, y)}.
Clearly, under our assumptions, g1(y, y) is an analytic function of y for y-a.e. y E E.
Further, by [HilPh, Equation (3.7.5), p. 80], G1(y) = fE g1(y, y)y(dy). Moreover, for
y-a.e. y, we have

By hypothesis, the dominating function in (15.2.17) lies in L 1 ( E , y). Part (2) now
follows from the corresponding result for Lebesgue integrals of scalar-valued functions
depending on a parameter.
When we apply Lemma 15.2.9 in Chapters 15-18, we shall always choose T — C~+
in part (1) and T = C+ in part (2).
Remark 15.2.10 (Notation for product measures) Product measures will appear
frequently in Chapters 15-19. When the product of the measures u1, . . . , uk
appears under the integral sign, we prefer to write (u1 x . . . x uk)(ds1, . . . , dsk)
or xku=1 uu (ds1, . . . , dsk), but, for the sake of brevity, we will sometimes write
xku=1 uu(dsu), When the measures are all the same, say u1 = . . . = uk = u, we
will use the briefer (tensor product) notation u® k (ds1, . . . , dsk) to denote the product
of k copies of u.

15.3 A simple generalized Dyson series (n = u + wdr)

We begin by treating a prototypical example and deriving the corresponding generalized
under U.S. or applicable copyright law.

Dyson series. The difficulties that we will encounter in this chapter are of two kinds,
analytic and combinatorial. Most of the analytic obstacles involved in Sections 15.3-
15.5 occur in the special case that is our concern in Theorem 15.3.1 and Corollary
15.3.4 below and will be dealt with in detail in this section. The combinatorial problems
encountered here provide a good introduction to Sections 15.4 and 15.5 where we will
concentrate on the combinatorial aspects. This division makes the exposition easier to
follow, and, once the reader has understood the present section, it will be quite clear how
to supply the analytical details omitted in Sections 15.4 and 15.5. (However, in Section
15.4—where the "generalized Dyson series" obtained is of much greater combinatorial
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
complexity—some
UNIVERSIDAD DISTRITALofFRANCISCO
the main analytic
JOSE DE CALDASarguments will also be provided.)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 A SIMPLE GENERALIZED DYSON SERIES (n = u + 417
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Let n = u + wdr, where u e M(0, t) is continuous, dT is the Dirac mass at r € (0, t)

and w e C. Let 0 € Loo1:n. Set

We will be interested in analytic functions of F1 (y), but we postpone that until Corollary
15.3.4. Most of the problems arise in dealing with the nth power Fn(y), as in (15.3.1),
and that is the subject of our first theorem.
Theorem 15.3.1 The operator Kty (Fn) exists for all y e C~+ and

where, for 0 < j < k < n,

The integral over Ak;j in (15.3.2) is a Bochner integral in the strong operator sense.
(That is, for every U E L 2 (R d ), Lj U is an L 2 (R d )-valuedBochner integrable function;
see [HilPh, Chapter III] and Remark 15.3.2(e) below.)
Moreover, for all E 6 C~+,we have

Before beginning the proof, we make some remarks intended to help the reader
understand the notation and follow the arguments.
under U.S. or applicable copyright law.

Remark 15.3.2 (a) Note that in the expression (15.3.3) for A k ; j , we have 0 < r < s1 <
• • • < Sk < t when j = 0 and 0 < s1 < . . . < Sk < T < t when j = k. Thus L0 begins
with e - r ( H 0 / y ) [ 0 ( r ) ] n - k and L k ends with [0(r)]n-k e- ( t - r ) ( H 0 / y ) .
(b) Figure 15.3.1 shows the regions Ak;j for k = 2, where

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
(c) The integrand Lj of (15.3.2) depends on k as well as j.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 418 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(d) The expression u®k in (15.3.2) denotes the product of k copies of the measure u,
in agreement with our conventions in Section 15.2.F (and Remark 15.2.10); alternatively,
we sometimes write the integral in iterated form.
(e) The sense of the equality (15.3.2) is that for every U € L2(Rd),

We caution the reader that the integral appearing in (15.3.2) may not make sense in the
operator norm topology (often called the uniform operator topology).
(f) The reader may now find it useful to consult Section 15.6 on Feynman diagrams.
In particular, Figures 15.6.1-15.6.3 are relevant to the present theorem.

Proof of Theorem 15.3.1 We will first emphasize the formal calculations leading to
(15.3.2). Later we will provide the justification for the numbered steps.
Let V 6 L2(Rd), E e Rd and y > 0 be given. Set

Since u is a continuous measure, one can see by a sectioning argument (analogous to

the one used in the proof of Lemma 15.2.7) that

except for a set of mxk-measure zero. Note for later use in the proof that, by (15.3.3),
the sets Ak;j in (15.3.7) are pairwise disjoint. The reader may find it helpful to consult
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
FIG. 15.3.1. The regions Ak;j, k = 2
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 A SIMPLE GENERALIZED DYSON SERIES (n = u + wdr) 419
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Figure 15.3.1 above. Now, in view of (15.2.8),

under U.S. or applicable copyright law.

Hence, (15.3.2), with Lj given by (15.3.4), is established (formally, at

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017
this point) for all
5:46 PM via
d
y >UNIVERSIDAD
0. (Of course, formula (15.3.8) only
DISTRITAL FRANCISCO JOSE DE CALDAS holds for a.e. E E R .)
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 420 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Before proceeding to the justification for the numbered steps, we note that Lj =
L j ( s 1 , . . . , Sk) as given by (15.3.4) is measurable in the strong operator topology [HilPh,
§3.5, esp. Corollary 2, p. 73].
Step (I) results from writing n as u + wdr and carrying out the integral with respect
to wdT . In (II), we separate the continuous part of n from the discrete part of n by means
of the binomial theorem. Step (III) follows from the "simplex trick" which works just as
well for an arbitrary continuous measure as for Lebesgue measure I as we now explain
briefly: the set (0, t)k is, except for boundaries, the union of k! simplexes, one being Ak,
the others differing from Ak only by permutations of the s-variables. Since m is a con-
tinuous measure, Lemma 15.2.7 implies that these boundaries have uxk-measure zero.
Furthermore, the integrand is invariant under permutations of the s-variables. Hence, the
integrals over the k! simplexes are all equal and (III) follows.
Equation (15.3.7) implies (IV). In Ak;j, the time indices are ordered in anticipation
of carrying out the Wiener integral. Step (V) follows from Fubini's theorem which will
be justified below in conjunction with the proof of the norm estimate (15.3.5). Finally,
(VI) is obtained by application of Wiener's integration formula (Theorem 3.3.5) for
finitely based functionals, followed by a simple change of variables for finite-dimensional
Lebesgue integrals. For more detail on this last step in a simpler situation, see the m = 1
and m =2 cases of the calculations which begin with equation (14.3.14) and end with
equation (14.3.19).
Replacing Lj U by ||LjU||, w by |w| and u by |u|, we obtain the norm estimate
(15.3.5) for y > 0 essentially by reversing the steps above. First,

Using this, the norm inequality (15.2.1) and the equalities (15.3.4), (15.3.7), (15.2.5),
(15.2.7) and (15.2.6), we can write:
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
Consequently (15.3.5)FRANCISCO
UNIVERSIDAD DISTRITAL follows.JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 A SIMPLE GENERALIZED DYSON SERIES (n = u + wdr) 421
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

As previously noted, Lj is strongly measurable; thus, in view of (15.3.9) and [HilPh,

Theorem 3.7.4, p. 80], Lj is Bochner integrable in the strong operator topology as a
function of (s1, . . . , sk).
Doing the calculation in (15.3.8) with U, 9 and w replaced by their absolute values
and u replaced by its total variation \u\ leads to the Wiener integral of the function

Since \ U ( y - 1 / 2 x ( t ) + E)| is Wiener integrable, by (15.2.11) with U replaced by |U| E

L2(Rd), the use of the Fubini theorem in Step (V) above is justified.
We assumed throughout that Kty (F n ) exists for A > 0. However, the strict logical
order would be as follows: First, establish the norm estimates (15.3.9) and deduce that
the right-hand side of (15.3.2) defines a bounded linear operator on L 2 (R d ); secondly,
reverse the steps in (15.3.8) and conclude that Kty (Fn) exists for A > 0 and is given by
(15.3.2).
We finish this proof by showing that the right-hand side of (15.3.2) is strongly con-
tinuous for A in C~+ and analytic for A in C+. This will establish the existence of Kty (Fn)
(in the sense of Definition 15.2.1) as well as the equality (15.3.2) for all A in C~+. (We
continue to write Kty (Fn) for the extension to C~+ of the function initially defined for
A > 0. Note that, for nonreal y e C~+, there is no claim that Kty (Fn) is given by a Wiener
integral.)
The operator Lj given by (15.3.4) depends on A; we indicate this dependence now
by writing Lj = Lj(y; s1, . . . , Sk). Using the norm inequality (15.2.1) and the norm
equality (15.2.7), we see that for all y e C~+ and u®k-a.e. (s1, . . . , sk),

Next, noting that the multiplication operators involved in (15.3.4) are independent of
A and recalling (from Section 15.2.E) the strong continuity and analyticity properties
of exp[—^(H0/y)], we see that, for uxk-a.e. (s1, . . . , sk), Lj(y; s1, . . . , sk) is strongly
continuous for A in C~+ and analytic for y e C+. Moreover, the right-hand side of
(15.3.11) is M®k-integrable since 0 e Loo1;n. (See (15.2.5) and (15.2,7).) It thus follows
from Lemma 15.2.9 on Bochner integrals depending on a parameter that
under U.S. or applicable copyright law.

is strongly continuous on C~+ and analytic on C+. We conclude that Kty (Fn) exists and
is given by (15.3.2) for all A e C~+. Further, the norm inequality (15.3.5) follows from
(15.3.9) which clearly continues to hold for A e C~+.
Remark 15.3.3 (a) The existence of Kty (Fn) and the norm estimate (15.3.5) can be
obtained more easily and more directly for A > 0, as someone familiar with both the
Wiener and Feynman integrals might
EBSCO Publishing : eBook Academic
guess. Simply use the bound (15.2.15) from Lemma
Collection (EBSCOhost) n- printed on 6/8/2017 5:46 PM via
2
15.2.8 and replace
UNIVERSIDAD F nFRANCISCO
DISTRITAL ( y 1 / x JOSE
+ E) by (||0||oo1;n) .
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 422 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(b) Note that, although we have not explicitly mentioned it, Lemma 15.2.8 has been
needed all along to assure us that, for every y > 0, F n (y-1/ 2x + E) is defined for
m x Leb.-a.e. (x, E) € Ct0 x Rd.
Now, let

be an analytic function with radius of convergence strictly greater than

Corollary 15.3.4 Consider the functional

with f as in (15.3.13). Then Kty (F) exists for all X e C~+and is given by the "generalized
Dyson series" (or "time-ordered" perturbation expansion)

where Fn is the functional defined by (15.3.1) and Kty (Fn) is given by (15.3.2). Moreover,
for y e C~+, the series in (15.3. 15) converges in operator norm and we have

Proof We could show that the convergence of the series in (15.3.15) is uniform in
y e C~+ and argue from there. Instead, we regard the series as an additional integral;
then the proof parallels that of Theorem 15.3.1. The application of the Fubini theorem,
the Bochner integrability, the analyticity and the strong continuity all follow from the
analogue of the norm estimate (15.3.9).
Remark 15.3.5 (a) A slight variation of the argument yielding (15.3.16) gives the fol-
lowing estimate for the norm of the tail of the series for Kty (F):
under U.S. or applicable copyright law.

for all nonnegative integers p. We leave the easy verification as an exercise for the reader.
(b) Even when the "potential" 9 is time independent, the family of operators {Kty (F)}
is not in general a semigroup in the time parameter t when f ( z ) is different from exp(z)
or n is different from Lebesgue measure l. (See Chapter 17, especially Sections 17.2,
17.3 and 17.6.) Nevertheless, the use of results connected with the theory of semigroups
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:46 PM via
(Chapters 8 and
UNIVERSIDAD 9, as FRANCISCO
DISTRITAL well as Section 10.2) is still helpful and allows us (as was the case in
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 A SIMPLE GENERALIZED DYSON SERIES (n = u + wdT) 423
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

[JoLa1]) to simplify considerably earlier arguments for measurability, strong continuity

and analyticity found in [CaSt1, JoSk2, JoSk6] and elsewhere for n = l, In addition,
the systematic use of operator-theoretic notation (also made in [JoLa1]) substantially
shortens all the expressions. These comments apply equally well to later sections in this
chapter (and to Chapters 16-18). The advantage of operator-theoretic methods in this
context was pointed out in [La12, pp. 58-60] in reference to [Jo3]. (See the introduction
to Section 11.5.)
Next, we examine the case w = 0; further specialization yields the classical Dyson
series (see Equation (15.3.22) below). We refer to Figure 15.6.1 for the Feynman dia-
grams associated with the series in formula (15.3.17) below.
Corollary 15.3.6 (Purely continuous measure: n = u) Let F be given by (15.3.14).
When CD = 0, the perturbation expansion for Kty (F) becomes

where

and, for (s1 , . . . , sn) € An,

Proof It suffices to consider Kty (Fn) as given by (15.3.2). When w = 0, the only nonzero
term on the right-hand side of (15.3.2) is obtained when k = n; thus

We observe that when k = n, Lj—as given by (15.3.4)—does not depend on j; in fact,

under U.S. or applicable copyright law.

since

(15.3.21)

Lj equals the integrand £ of (15.3.17). In view of (15.3.7), we have

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
as desired.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 424 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The classical Dyson series

The classical Dyson series—except for an exact reversal of the time-ordering which
will be discussed specifically in Remark 15.3.7 below—corresponds to the special case
of (15.3.17) obtained by letting (u = l = Lebesgue measure, f(z) = exp(z) (so that
n!an = 1), as well as A = —i and 6 = -i V (the "quantum-mechanical case", as in
Remark 15.2.6(b)), where the potential V belongs to Loo1;n. In fact, it is given by

where the simplex An is as in (15.3.18). (Usually, V is assumed to be time independent

and so for each j = 1, . . . , n, V(sj) is replaced by V in (15.3.22).)
It will be helpful to recall briefly the physical interpretation of the nth term of the
classical Dyson series (15.3.22) ([Dys 1,2,4], [Schu, pp. 67-68], [Schwel, § ll.f]). For
now, we think of a single quantum-mechanical particle moving in the potential V. The
integrand of the nth term may be described as follows: A free evolution between times 0
and s1, interaction with the potential V at time s1, free evolution between s1 and S2, and
so on up to an nth interaction with V at time sn followed by a free evolution between
sn and t. We now integrate (or "sum") over the simplex An to take into account all the
possible times of interaction; in the present case, the times receive equal weight since
n = l.
The form of the series in (15.3.17) is essentially the same for any continuous measure
n = u. However, even with n = u absolutely continuous, different physical interpret-
ations are suggested. For example, the weighting may force all the interactions to take
place in some short time interval. The absolutely continuous case amounts to multiplying
0(s, .) by a function of time which equals the density of u with respect to l. (With this
in mind, one can see that this situation is included in the p = p' = 2 case of [JoSk6].)
In general, n = u has a singular part. (See, for example, [Coh, p. 141], [ReSi1,
pp. 20-23] or [WhZy, pp. 35, 116 and 180].) This could be, for instance, the Lebesgue-
Stieltjes measure associated with the Cantor function. Cantor-like functions have been
used, for example, in connection with the notion of "fractal time" in the theory of errors in
under U.S. or applicable copyright law.

data transmission lines. (See [Mand, Chapter 8, esp. pp. 74-75 and 78-83].) The Cantor
function and its associated measure will be used in a quantum-mechanical setting in
Example 15.5.3 below, as well as in several places in Chapter 17.
When n has a discrete part, as in Corollary 15.3.4, even the formal appearance of
the generalized Dyson series changes markedly. This is most striking in the general case
to be examined in the following section, but can already be seen in formulas (15.3.2)-
(15.3.4) of Theorem 15.3.1 where n = u + wdr and w = 0. The single integral over An
in the nth term of (15.3.17) is replaced in (15.3.2) by a double sum of integrals over the
sets Ak;j. This double sum accounts for the various placements of the variables Sj with
respect to r. Note the special role played by the fixed time r; indeed, r always appears
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
andDISTRITAL
0 ( r ) appears toJOSE
FRANCISCO powers ranging from 0 to n. In the time intervals [0, r) and (r, t],
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 A SIMPLE GENERALIZED DYSON SERIES (n = u + wdr) 425
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

there is a succession of free evolutions and interactions weighted by u®k just as in the
case w = 0.
The reader can easily imagine that there will be substantial additional complications
in the next section where many different 9s and ns are involved and where the discrete
parts of the ns may have countable support.
Remark 15.3.7 Each summand in the generalized Dyson series (15.3.15) (with Kty (Fn)
as in (15.3.2) and Lj as in (15.3.4)) should, in the natural physical interpretation, be
read from time t to time 0 rather than from 0 to t, as you go from left to right on the
page. The change to the natural physical ordering can be brought about mathematically
by replacing (in the definition (15.3.14) or (15.3.1) of F or of Fn, respectively) the
"potential" 9 and the measure n by their "time-reversals" 9 and n~, respectively, where

and

the image measure of n under the "time-reversal" map s -> t — s from [0, t] to itself.
(We will in fact need to substitute the pair (0, n~) for the pair (0, n) in Chapter 17 where
we will be interested in the time evolution of Kty (F).)
Observe that our assumptions (in this section) are still satisfied by the new pair (0, n~)
since 0 E Loo1;n~ if and only if 0 e L001;n. Indeed,

since \n~\ = \n\ and by the abstract change of variable theorem (Theorem 3.3.2),

An alternative (which we will not use) would be to start Brownian paths at time t
and let them flow backwards to time 0.
The above comments in Remark 15.3.7 also apply to the simpler generalized Dyson
series (15.3.17) (with n = u and £ as in (15.3.19)) and will be extended in Remark
under U.S. or applicable copyright law.

15.4.4 to the general situation considered in Section 15.4 below as well as throughout
Chapters 15-18. For now, we illustrate it by a very simple example. (We leave the easy
verification to the reader.)
Example 15.3.8 Let u be a continuous measure and let 0 e Loo1;u (so that, by (15.3.24),
0 E Loo1-u~).
(a) Let

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 426 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then we have

[Note that (15.3.25) is just the n = 2 term of the series (15.3.17) in Corollary 15.3.6
where a2 = 1/2! and an = 0 for all n = 2.]
(b) Next, let ~F be the "time reversal" of F; namely,

where 9 and u~ are given as in (15.3.23).

Then, in contrast to (15.3.26), we have

Note that, in agreement with Remark 15.3.7, (15.3.28) follows from (15.3.26) (and
Corollary 15.3.6) by substituting the pair (~0, u~) for (6, u~). Indeed, it can be deduced
from (15.3.26) (and the abstract change of variable formula, Theorem 3.3.2) by means
of the substitution s' = t — s.

15.4 Generalized Dyson series: The general case

We now extend the results of Section 15.3 to a much broader setting. The main results
of this section, Theorem 15.4.1 and Corollary 15.4.3, show that for a large class of
functions F, the operator Kty (F) exists and can be disentangled by a (rather involved)
time-ordered perturbation expansion or generalized Dyson series (in short, GDS). This
will be the key to the construction in Section 15.7 of a commutative Banach algebra
of functionals, called the disentangling algebra, which will play a central role in later
under U.S. or applicable copyright law.

developments (see Chapter 18).

By contrast to Section 15.3, many measures and potentials may be involved; more-
over, the discrete part of each measure is unrestricted. The new complications are mainly
combinatorial in nature; otherwise, the proofs proceed much as in Section 15.3. For this
reason, we shall emphasize the combinatorial aspects.
Eventually, we wish to consider series of functions of the type we are about to
introduce, but for now, we let

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERALIZED DYSON SERIES: THE GENERAL CASE 427
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where nu e M(0, t) and 0u e Loo1;nu for u = 1, . . . , m. Note that by Lemma 15.2.8,

for every X > 0, F ( y - 1 / 2 x + E) is defined for m x Leb.-a.e. (x, £) in Ct0 x Rd. Let

be the unique decomposition of nu into its continuous part nu and its discrete part vu
(see Section 15.2.F). For each u = 1, . . . , m, we write

where {r p ; u }oop=1 is a sequence from (0, t) and {wp;u}?li is a sequence from C such that

We will write each of the factors in (15.4.1) as an absolutely convergent infinite

series. We will then find it advantageous to multiply these series together. With this is
mind, we introduce the following notation:
Given k between 0 and m, [k; m] will denote the collection of all subsets of size k
(or k-sets) of the set of integers {1, . . . , m}. If {a1, . . . , ak} e [k; m], we shall always
write

Now, in view of (15.4.2) and (15.4.3),

under U.S. or applicable copyright law.

Note that when k = 0, {ak+1, . . . , am] = [1, . . . , m] and the integral involving the
continuous measures does not appear in the corresponding term of the final expression.
On EBSCO
the other hand,: when
Publishing k = m, then
eBook Collection {a1, . .- .printed
(EBSCOhost) , ak} =on {1, . . . , 5:47
6/8/2017 m} and only
PM via the continuous
UNIVERSIDAD
measures
DISTRITALappear.
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 428 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Soon we will wish to calculate the Wiener integral defining Kty (F). For this purpose,
we will need to order the time variables. We begin by ordering the rs that appear within a
given term of the series in (15.4.4). For fixed k,{a1, . . . , ak} in [k; m] and pk+1, . . . , pm,
let a be a permutation of {k + 1, . . . , m} such that

(If the rS involved in (15.4.5) are distinct, then the permutation a is unique.)
We can now state the key result of this section ([JoLa1, Theorem 2.1, pp. 27-29]).
(See also Corollary 15.4.3 below [JoLa1, Corollary 2.1, pp. 33-34].)
Theorem 15.4.1 Let F be defined by (15.4.1). Then Kty (F) exists for y € C~+. In
particular, for y purely imaginary, the analytic operator-valued Feynman integral of F
is well defined. (See Definition 15.2.1.) Moreover, for all y e C~+,

where, for each fixed k E {0, . . . , m}, p ranges through the group Sk of permutations of
[1, . . . , k]and

In addition, for (s1, . . . , Sk) e Ak;j1, . . .,j m _ k+1 (p) and r = k, . . . ,m,
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
Here, a is aFRANCISCO
DISTRITAL permutation
JOSE DE of {k + 1, . . . , m} as defined in (15.4.5).
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERALIZED DYSON SERIES: THE GENERAL CASE 429
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The integrals in (15.4.6) are to be interpreted as Bochner integrals in the strong

operator topology and the series converges in the operator norm. Further, for all y € C~+,
we have

Remark 15.4.2 (a) In (15.4.7) and (15.4.8) above, we adopt the conventions
tpd(k);ad(k) = 0; tpa(m+1);ad(m+1) = t and 0(T pa(k);aa(k) ) = 1, the identity operator. Fur-
ther, we take jo = 0; then, when r = k, it is reasonable to interpret j\ + . . . + jr-k + 1
as 1, and we also get jr-k = j0 = 0. The s-values between two equal rs are omitted in
(15.4.7).
(b) If Tpa(r);ad(r) = rpd(r+1);aa(r+1), then jr-k+1 = 0, Lr in (15.4.8) reduces to
0ad(r) ( r AV(r);ad(r)) and the term L r L r +1 involves the product

In fact, more than two successive TS may be equal in (15.4.5), giving rise to a cor-
responding product of 0s evaluated at the same time and without intermediate semi-
groups.
(c) Note the simplicity of the norm estimate (15.4.9) in spite of the complexity of the
formula (15.4.6) for Kty (F).
Proof of Theorem 15.4.1. Except for the combinatorics, the proof parallels that of The-
orem 15.3.1.
We first show that Kty (F) is given by (15.4.6) for y > 0. Fix U € L2(Rd). For
notational simplicity, we let A = 1 and E = 0. (Actually, the calculations hold for any
y > 0 and almost every £ € Rd.) Also, we write 6(x(s)) instead of 0(s, x(s)). From
(15.2.8), (15.4.4) and (15.4.5), it follows that
under U.S. or applicable copyright law.

For p E Sk, define the simplex

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 430 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The continuity of the measures u au , u = 1, . . . , k, and a sectioning argument as in

Lemma 15.2.7 give the equality (0, t)k = Upesk Ak(p),up to a set of xku=1 uau -measure
zero. Consequently, since the above union is disjoint,

Again using Lemma 15.2.7 and the fact that the measures u1, . . . , um are continuous,
we see that

where the pairwise disjoint subsets Ak;j1, . . .,j m-k+1(p) are defined in (15.4.7) and the
equality holds up to a set of x ku=1 uap(u) -measure zero. In view of (15.4.12) and (15.4.13),
we thus have

By application of Fubini's theorem, we deduce from (15.4.10) and (15.4.14) that

under U.S. or applicable copyright law.

Observe that, within the integrand of the Wiener integral in (15.4.15), the time variables
are explicitly ordered according to (15.4.7). By Wiener's integration formula (3.3.9), it
thus follows that this Wiener integral is equal to ((Lk . . . Lr . . . L m )U)(0), with Lr given
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
by (15.4.8). We conclude
DISTRITAL FRANCISCO JOSE DE that (15.4.6) holds for all y > 0.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERALIZED DYSON SERIES: THE GENERAL CASE 431
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

It will be helpful to keep in mind that, in the above derivation, the string of inequalities

corresponds to the operator Lr.

The additional combinatorial complications of our present setting make the proof of
the norm estimates somewhat more involved. Once we have the norm estimates, however,
the rest of the proof proceeds much like that of Theorem 15.3.1.

Hence,
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 432 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where (15.2.7) and (15.2.6) are used to obtain the last two equalities.
By replacing U, 0u, w p;u by their absolute values and uu by |uu| in the calculation
leading to (15.4.15), we obtain, exactly as in the proof of (15.4.17), the Wiener integral
of the function

Since by (15.2.11), U ( x ( t ) ) is Wiener integrable, this justifies the use of Fubini's theorem
in the derivation of (15.4.15).
Analyticity and strong continuity of Kty (F) are argued just as in the proof of Theorem
15.3.1 by using Lemma 15.2.9 which deals with Bochner integrals depending on a
parameter.
Corollary 15.4.3 (Generalized Dyson series) Let{Fn}oon=0be a sequence of functionals
each given by an expression of the type (15.4.1):

where nn,u e M(0, t) and 0n,u € Loo1;nn,u. (Note that if mn = 0, we take Fn = 1.)
under U.S. or applicable copyright law.

Assume that

Then for all y > 0, the individual terms of the seriesEoon=0F n (y-1/ 2 x + E) are defined
and the series converges absolutely for m x Leb.-a.e. (x, E) e Ct0 x Rd. Let

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERALIZED DYSON SERIES: THE GENERAL CASE 433
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then for all y e C~+, Kty (F) exists and is given by the following time-ordered
perturbation expansion (GDS):

where Kty (Fn) is defined by (15.4.6) with the functional F from Theorem 15.4.1 replaced
by Fn as in (15.4.18).
The series in (15.4.21) converges in operator norm; furthermore, for all A 6 C~+, we
have the estimate

Proof The fact that for every A > 0 the terms of the series in (15.4.20) are defined and
that the series converges absolutely for m x Leb.-a.e. (x, E ) follows from Lemma 15.2.8
and the assumptions on the sequence {bn}, where

(See (15.4.19) according to whichEoon=0bn < oo.)

Let y > 0. A use of the integrable (by (15.2.11)), dominating function

allows us to interchange the order of the integral and sum and write
under U.S. or applicable copyright law.

Now, the inequality||kty(Fn)|| < bn from (15.4.9) in Theorem 15.4.1 assures us that
the series Eoon=0 Kty (Fn) converges in operator norm, uniformly for y e C~+. The ana-
lyticity and strong continuity ofEoon=0Kty(Fn) follow from this and the corresponding
assertions about Kty (Fn) which were given in Theorem 15.4.1. In the light of Definition
15.2.1, the existence of Kty (F) for y € C~+ and the formula (15.4.21) now follow from
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
(15.4.23).
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 434 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 15.4.4 (Time-reversal and natural physical ordering) The GDS in (15.4.21)
(with each Kty (Fn) as in (15.4.6)-(15.4.8)) is not written in the natural physical order. In
that order, consistent with Feynman's time-ordering convention (14.2.1), the operators
involving smaller time indices always act before those involving larger time indices.
(This same idea was expressed differently in Remark 15.3.7.) We can obtain the natural
physical order by replacing (in the definition (15.4.18) of Fn) each pair (0 n,u , n n , u ) by its
"time-reversal" (0n,u, nn,u), where the "potential" 0n,u and the measure n~ are defined
as in (15.3.23); namely,

and

the image measure of nn,u under the "time-reversal" map s |-> t — s of the interval
[0, t].
Observe that our assumptions (in Corollary 15.4.3) are still satisfied since by
(15.3.24), we have ||0n,u||oo1;n~n,u = ||0n,u||oo1;nn,u and so, by (15.4.19),

is also finite.
Hence our claim follows by simply applying Corollary 15.4.3 to the new functional

the "time-reversal" of F, and then using the abstract change of variable theorem (The-
orem 3.3.2) by making the substitution s' = t — s in each term of the series (15.4.21)
(with Kty (Fn) given by (15.4.6)), where (0n,u,n~n,u)has been substituted for (0n,u nn,u).
(For a simple illustration of this, we refer back to Example 15.3.8.)
The above comments in Remark 15.4.4 (which will be expanded upon in Section 15.7,
especially in Theorem 15.7.6) apply to the generalized Dyson series occurring throughout
Chapters 15-18 and are particularly relevant to the development in Chapter 17. We note
under U.S. or applicable copyright law.

that in Chapter 19, where we address Feynman's time-ordered calculus in a more general
setting, our perturbation expansions will be obtained directly in the natural physical order.

15.5 Disentangling via perturbation expansions: Examples

We now wish to consider various special cases. In each of these cases, the existence
of the operator Kty (F) for y E C~+ and corresponding norm estimates follow from
Theorem 15.4.1 and Corollary 15.4.3. In each example, we shall give the time-ordered
perturbation expansion (GDS) of Kty (F) for y e C~+. The series appearing in these
formulas all converge in operator norm and the integrals are taken as Bochner integrals
withEBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
respect to the strong operator topology.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DISENTANGLING VIA PERTURBATION EXPANSIONS: EXAMPLES 435
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

A single measure and potential

Let n e M (0, t) and 9 € Loo1;n. As usual, n = u + v will be the decomposition of n
into its continuous and discrete parts, and we will write

where {T P }oop=1 is a sequence from (0, t) and {w P }oop=1 is a sequence from C such that

be an analytic function with radius of convergence strictly greater than ||0 ||oo1;n. Consider
the functional

The case where v has finite support is perhaps most likely to be of physical interest;
this is the object of our first example.
Example 15.5.1 (Finitely supported v) Let u be a continuous measure in M(0, t) and
let

where we may as well assume that 0 < r1 < . . . < Th < t. Then
under U.S. or applicable copyright law.

where q0, . . . , qh, j1, . . . , jh+1 are nonnegative integers,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 436 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and, for (s1, . . . , sq0)EAq0;j1,. . .jh+1 and r E {0, . . . . h},

(We use the conventions r0 = 0, rh+1 = t and [ 0 ( t 0 ) ] q 0 = 1.) Moreover,

We briefly explain the structure of the GDS (15.5.5). Let

We abbreviate 0(s, x ( s ) ) to 9(x(s)). Using the multinomial formula, we have

where Aq0 is given as in (15.3.18). We now order the s-variables with respect to the TS,
thereby giving rise to the sum over the js. Finally, we calculate the Wiener integral to
obtain the nth term of the series (15.5.5).
While formula (15.5.5) is not especially simple, note that the arguments which led to
under U.S. or applicable copyright law.

it are much simpler than in the general case discussed in Theorem 15.4.1 and Corollary
15.4.3. The Feynman diagram corresponding to the nth term of the generalized Dyson
series (15.5.5) is given in Figure 15.6.5.
Observe that we recover formula (15.3.15) of Corollary 15.3.4 by letting h = I , T 1 =
T and w1 = w in Example 15.5.1 . By specializing even further (w1 = 0), we obtain the
case of a purely continuous n.
Example 15.5.2 (n = u purely continuous) The generalized Dyson series correspond-
ing to this case is given in Corollary 15.3.6. It is similar in form to the classical Dyson
series given
EBSCO in (15.3.22).
Publishing The latter
: eBook Collection is recovered
(EBSCOhost) - printedby
onletting
6/8/2017f (5:47
z ) PM= via
exp(z), n= u =l
UNIVERSIDAD
DISTRITAL
and 0 = -iV. FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DISENTANGLING VIA PERTURBATION EXPANSIONS: EXAMPLES 437
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In our next example, we consider a special case of Example 15.5.2 where u is the
Lebesgue-Stieltjes measure associated with the Cantor function C = C(t) defined on
[0, 1]. (See [Coh, p. 55 and pp. 22-24], [ReSi1, pp. 20-23] and [Mand, Chapter 8, esp.
Plate 83, pp. 82-83]. Recall that C is a nonnegative, nondecreasing continuous function
on [0, 1] such that C(0) = 0 and C(l) = 1; further, C is constant on each interval in the
complement of the (ternary) Cantor set in [0, 1].) It is natural in this setting to extend C
periodically to [0, +00), but we will (mostly) restrict our attention to the usual Cantor
function and to t between 0 and 1.
The calculation is not complicated but the resulting formula for Kty (F) is interesting,
especially in the quantum-mechanical case, and seems to be new.
Example 15.5.3 (n = u purely continuous and singular) The basic relationship between
the Cantor function C and the Cantor-Lebesgue measure u is given by the formula

The measure u is a continuous probability measure which is supported by the Cantor

set. Although the cardinality of the Cantor set is the same as the cardinality of [0, 1], it
has Lebesgue measure 0. Thus u is singular with respect to Lebesgue measure.
We now take 9 constant (say, 0 = K), and f(z) = exp(z). The function F is
independent of the path x in this case and is given by

Since an =1/n!,we see from (15.3.17) of Corollary 15.3.6 that for t E [0, 1],
under U.S. or applicable copyright law.

Hence, for all y E C~+,

In particular,
EBSCO Publishing :taking K = — 1(EBSCOhost)
eBook Collection and y = - 1, we have
printed in the5:47
on 6/8/2017 diffusion or probabilistic
PM via UNIVERSIDAD
case,
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 438 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

On the other hand, with K = —i and y = —i, we obtain in the quantum-mechanical

case,

Whether (15.5.15) models some real physical phenomenon is not clear to us; never-
theless, we make some comments about the situation described there. Note that for any
initial state U,

also; hence the computation of probabilities associated with the evolution of this isolated
"particle" is exactly the same as for the free evolution of that particle. The phase changes
induced by the factor e -ic(t) will have an influence on the calculation of probabilities if
the particle interacts with other particles since the changing phases will alter interference
effects. Because the Cantor function is constant on the countable union of intervals which
make up the complement of the Cantor set, the phase remains constant across those
intervals. The changes take place at the fractal set of times in the Cantor set itself.
Finally, we note that our results (15.5.13)-(15.5.15) extend in an obvious manner if
instead of the standard Cantor function C on [0, 1] as above, we consider its periodic
extension to [0, +00) or to R (as is done in a different context in [Mand, Chapter 8, esp.
pp. 74-75 and 78-79]).
Exercise 15.5.4 (a) Calculate the final expression for Kty (F) in (15.5.13) by beginning
for X > 0 with (15.2.8) in Definition 15.2.1 and then extending your result first to C+ by
analytic continuation and then to C~+by strong continuity.
(b) Pick some other positive Borel measure u on [0, +00) which has finite total
under U.S. or applicable copyright law.

variation on [0, t]for every t > 0, and let Du (t) be the associated function of bounded
variation [Coh, pp. 22-24]. Do a calculation for this situation as in the proof of (15.5.13)
and interpret your result in the quantum-mechanical case.
We will return to Example 15.5.3 in several places in Chapter 17 where we will
discuss the integral and differential equations associated with this evolution.
Next we single out the case when n from Example 15.5.1 is purely discrete. We will
use this result in Sections 16.2 and 17.3.
Example 15.5.5 (n purely discrete and finitely supported) Let n = v be given by(15.5.4)
with thePublishing
EBSCO rs ordered. WeCollection
: eBook write Kty(EBSCOhost)
(F) in two ways;onthe
- printed first emphasizes
6/8/2017 the connection
5:47 PM via UNIVERSIDAD
withDISTRITAL
formulaFRANCISCO JOSE the
(15.5.5), DE CALDAS
second, the similarity of the inner sum to the multinomial
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DISENTANGLING VIA PERTURBATION EXPANSIONS: EXAMPLES 439
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

formula:

If f ( z ) — z, (15.5.16) simply becomes

Alternatively, (15.5.16) and (15.5.17) can be obtained with relative ease in this case
by returning to Wiener's integration formula (3.3.9) followed by analytic continuation
and strong continuity.
Note that the formulas for Kty (F) in (15.5.16) do not involve integrals; it is the
continuous part u of the measure n which leads to the integrals in (15.5.5) as well as to
the sum over the js.
If f(z) = exp(z), then Wiener's integration formula (3.3.9) and the multiplicative
property of the exponential function yield

(See Exercise 15.5.6 below for a simple special case.)

Of course, one can also recover (15.5.18) from (15.5.16); indeed, since n!an = 1 in
(15.5.16), the multinomial formula can be applied to yield (15.5.18).
under U.S. or applicable copyright law.

Specializing further, we begin to see connections with the product integral ([DoFri3],
[Mas], [dWmMN2, §2]) and the Trotter product formula (Chapter 11, especially Section
11.1). Let w1 = T1, w2 = t2 — T1, . . . ,wh = th — th-1. Then (15.5.18) becomes

a "partial product" which plays a role in the theory of the product integral analogous to
that played by partial sums in Riemann integration. Product integration in this context
wasEBSCO
pursued by Lapidus
Publishing in ([La14,15,
: eBook Collection Lal8]
(EBSCOhost) and especially
- printed on 6/8/2017 [Lal6]).
5:47 PM viaItUNIVERSIDAD
is not one of the
main themesFRANCISCO
DISTRITAL of thisJOSE
book, however, although we will return to it briefly in Section 17.6.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 440 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

If the Tp's are equally spaced (i.e. if rp = p(t/h) for p = 1, . . . , h), and if 9 is time
independent, we obtain

which is a Trotter product. Observe that if y = 1 and 6 = — V, then

on the other hand, if y = —i and 9 = -iV, then

We shall use this link with the Trotter product formula in Example 16.2.7 of Chapter 16,
in connection with our discussion of "stability in the measures". Recall from Section 11.1
that this formula involves taking the limit as h -> oo in expressions such as (15.5.20)-
(15.5.22). (Caution: The integer h in (15.5.22) does not stand for Planck's constant.) For
convenience, our assumption all along has been that the rs were in (0, t). Consistent
with our remark at the end of Section 15.2.F, it is quite possible to let rh = t, and we
have taken advantage of that in (15.5.20)-(15.5.22).
Exercise 15.5.6 Given w E C, y E C~+, T € (0, t) and 9 essentially bounded, let
under U.S. or applicable copyright law.

The results of the next example will be used in Section 16.2 (Example 16.2.11) and
in Section 17.6.
Example 15.5.7 (n
EBSCO Publishing an arbitrary
: eBook Collection Borel measure)
(EBSCOhost) We
- printed on treat here5:47
6/8/2017 thePMcase of a single arbi-
via UNIVERSIDAD
trary n e M(0, t) and a single 0 e Loo1;n. Let v, the discrete part of n, be defined by
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DISENTANGLING VIA PERTURBATION EXPANSIONS: EXAMPLES 441
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(15.5.1). Then

where, for each h, a is the permutation of {1, . . . , h} such that

and

further, for (s1, . . . ,s q 0 ) e Aq0;j1,...,jh+1 and r e {0, . . . , h},

(We use the conventions rd(o) = 0, rd(h+1) = t and [0(td(0))] qd(0) = 1.)
Finally, the norm estimate (15.5.8) holds. In particular, if f(z) = exp(z),
under U.S. or applicable copyright law.

Note that if [rp] is an increasing sequence, then the permutation a that appears in
(15.5.25)-(15.5.28) can be omitted.
The key to the combinatorial structure of Equation (15.5.25) is the following version
of the "No-nomial formula":

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 442 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Equation (15.5.30) can be obtained as follows:

which equals the right-hand side of (15.5.30).

To use (15.5.30) in order to deduce (15.5.25), take bo — f ( 0 , t ) 0 ( x ( s ) ) u ( d s ) and
bp = w p 0 ( x ( T p ) ) for p > 1. For a given h, introduce the permutation a of {1, . . . , h]
to order T1, . . . , rh, as in (15.5.26). One may now finish much as in Example 15.5.1.

Several measures and potentials

We now proceed somewhat less formally than before. Our main intention is to suggest
the rich variety of possible examples within our framework.
The measures denoted by u, u1, u2, . . . , will always be continuous measures in
M(0, t) and T, T1, T2, . . . will lie in the interval (0, t).
Example 15.5.8 Let

where 0 e Loo1;u and 01 e L001;dr. (Note that 01(r, x(r)) = f(0,t)01(s, x(s))d T (ds).)
Then

where An and An;j are as in (15.3.6) and (15.3.3), respectively. Hence

under U.S. or applicable copyright law.

where

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DISENTANGLING VIA PERTURBATION EXPANSIONS: EXAMPLES 443
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

If

it then follows from (15.5.33) and Corollary 15.4.3 that

One can think of 01(t) appearing in (15.5.36) and (15.5.34) as corresponding to the
first order term of an external disturbance, at the fixed time T, of the physical system
determined by 6 and u. Recall from Remark 15.2.6(b) that taking X = -i and 0 = — i V
puts us in the quantum-mechanical setting. Such "instantaneous interactions" might
occur at a finite number of fixed times T1 , . . . , rh. We consider this situation in the
next example where we shall also highlight some of the connections with Feynman's
operational calculus.
Example 15.5.9 Let 0 < r1 < . . . < rh < t. Put

Then

where An;j1,...,jh+1 is as in (15.5.6).

under U.S. or applicable copyright law.

We write explicitly the case when n = 1:

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 444 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and so

The passage from the functional F1 written in the form (15.5.39) to the operator
Kty (F1) respects the ordering of the time indices. (Mathematically, this is due to Wiener's
integration formula (3.3.9).) Further, in (15.5.40), we have taken the operators not involv-
ing the variable s out from under the integral sign. We see here an explicit connection
with Feynman's time-ordering convention [Fey8, p. 109] recalled in formula (14.2.1).
This convention is central to the "disentangling" process which, as Feynman notes [Fey8,
p. 110], lies at the core of his operational calculus.
Next we consider the functional

Then, in view of (15.5.38),

where, for r = 0, . . . , h,
under U.S. or applicable copyright law.

here, r0 = 0, rh+1 = t and 00(ro) = 1.

The Feynman graph corresponding to the nth term of the generalized Dyson series
(15.5.42) is given in Figure 15.6.6.
Example 15.5.10 For 7 = 1, 2, let 8j e Loo1;uj. Put

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DISENTANGLING VIA PERTURBATION EXPANSIONS: EXAMPLES 445
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then

Note that within Am, r1 < . . . < rm, whereas within An, s1 < . . . < sn. In order to
obtain the formula for Kty (F m , n ), one needs to order the r and .s variables with respect
to each other much as in some earlier examples. (See also Example 19.7.2 and related
material in Chapter 19.) However, we will simply give the formula for Kty (F1,1):

By considering the functional

where u2 is a probability measure on (0, t), we would obtain a continuous analogue of

the perturbation series (15.5.36) in Example 15.5.8.
Exercise 15.5.11 Compute Kty (F) for F given by (15.5.47).
More generally, by using methods similar to those of the previous examples, we can treat
functionals of the form
under U.S. or applicable copyright law.

where nj e M(0, t), 0j € L0o1;nj for j = 1, . . . , q and where / is a function of q

complex variables which is analytic in a region containing

We have discussed several rather diverse examples, but we should emphasize that
much more is possible. Actually, not only any finite number, but even an infinite number
of distinct ns and 9s
EBSCO Publishing mayCollection
: eBook appear. We finish this
(EBSCOhost) section
- printed by giving,
on 6/8/2017 5:47 as
PM an
via exercise,
UNIVERSIDADa simple
DISTRITAL FRANCISCO
example of this type. JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 446 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Exercise 15.5.12 (Infinitely many ns and 8s) Suppose that nn E M(0, t) and 0n e
Loo1;nn for n = 1, 2, . . . . Further assume thatE8n=1||0n||0nllool;nn< oo. Let

Then show that

15.6 Generalized Feynman diagrams

At Cornell I was learning Richard Feynman's quite different way of calculating atomic
processes. . . . He had his own private way of doing calculations. His was based on things that
he called "Propagators", which were probability amplitudes for particles to propagate themselves
from one space-time point to another. He calculated the probabilities of physical processes by
adding up the propagators. He had rules for calculating the propagators. Each propagator was rep-
resented graphically by a collection of diagrams. Each diagram gave a pictorial view of particles
moving along straight lines and colliding with one another at points where the straight lines met.
When I learned this technique of drawing diagrams and calculating propagators from Feynman, I
found it completely baffling, because it always gave the right answers but did not seem to be based
on any solid mathematical foundation. Feynman called his way of calculating physical processes
"the space-time approach", because his diagrams represented events as occurring at particular
places and at particular times. . . .
Freeman J. Dyson, 1996 [Dys4, p. 12]

In this section, we introduce generalized Feynman diagrams corresponding to our gen-

eralized Dyson series. The diagrams contain most of the information conveyed by the
series and help to understand and visualize its various terms.
A great variety of Feynman diagrams associated with perturbation expansions appear
in the physics literature. Although our diagrams can be complicated in their own right,
they generalize the simple diagrams of nonrelativistic quantum mechanics but not those
of, for example, quantum electrodynamics.
We begin by considering the Feynman diagrams associated with the Dyson series
(15.3.17) of Corollary 15.3.6 where n = u and f ( z ) — exp(z). When n = u = l, we
under U.S. or applicable copyright law.

shall recover the classical Feynman diagrams associated with the classical Dyson series.
Figure 15.6.1 shows the graph corresponding to the nth term

of the series (15.3.17). This diagram describes a succession of free evolutions during
the time intervals (0, s1), (s1, S2), . . . , (sn, t) alternating with interactions which take
place at the times s1, S2, . . . , sn. The symbol u®n appearing at the bottom of the diagram
indicates integration over An with(EBSCOhost)
EBSCO Publishing : eBook Collection
respect to- u®n. In the classical Dyson series (15.3.22),
printed on 6/8/2017 5:47 PM via UNIVERSIDAD
we DISTRITAL
have u = l, but JOSE
FRANCISCO in general,
DE CALDASthe continuous measure u might be quite different from
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERALIZED FEYNMAN DIAGRAMS 447
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 15.6.1. The graph corresponding to the nth term of the GDS for n = u, continuous

Lebesgue measure. (See the discussion following the proof of Corollary 15.3.6, as well
as Example 15.5.3 above.) Note that in the setting of Example 15.5.3, where u is the
Cantor-Lebesgue measure, interactions could occur only at the points of the Cantor set.
A simple introduction to the concept of Feynman diagrams, as they appear in
the physics literature, can be found in [Mat, esp. §3.2.]; for the connections between
Feynman diagrams and Dyson series, we also mention ([Mi, Chapter 4], [Schu,
Chapter 10], [Si9, §20]).
We now consider generalized Feynman graphs corresponding to the nth term of the
generalized Dyson series (15.3.15) from Corollary 15.3.4. Recall that n = u + wdr
in this case. We assume for simplicity that f ( z ) = exp(z). Figure 15.6.2 gives three
representations for the graph corresponding to the term

To avoid the extreme cases for now, we assume that 0 < k < n and 0 < j < k.
We have n interactions just as in the example above but n - k of them are now
occurring at the fixed time T. We use a large dot at r to stress its special role. From the
graph-theoretic point of view, the number of vertices has been reduced by n — k — 1 from
n + 2 to k + 3. What we denoted in (a) by [ 0 ( r ) ] n - k is expressed in (b) and (c) by n—k—1
under U.S. or applicable copyright law.

loops attached to the vertex r. The segments connecting successive times represent edges
of the graph. The loops can be thought of as "infinitesimal edges" connecting r to itself
as we will discuss further on. The number of edges is reduced by n — k — \ without the
addition of the loops, but, with them, it remains unchanged and equal to n + 1.
In place of u®n in our earlier example, we now have u®k and wn-k /(n— k)! Here, u®k
represents integration with respect to the k variables s1, . . . , sk, whereas w n - k / ( n — k)!
arises from integration with respect to wdT of the remaining n — k variables. (We will
see eventually in Example 16.2.9 of the next chapter that the factorial comes from the
ordering of the variables.)
In the classical case, the term of the Dyson series corresponding to n interac-
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
tions is represented
DISTRITAL by aDE single
FRANCISCO JOSE CALDAS Feynman diagram (see Figure 15.6.1). Here, however,
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 448 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 15.6.2. Three representations of the graph corresponding to n = u + wdr for n, k

and j fixed

the generalized Feynman graph associated with n interactions has many disconnected
components, one corresponding to each term of the type (15.6.2) in the sum

(15.6.3)

Figure 15.6.3 shows these disconnected components for the case n = 3.

Remark 15.6.1 Clearly, there are k + 1 disconnected components associated with a
fixed k, k < n. If k = n, however, we have only one component rather than n + 1; in
fact, T does not appear, and we obtain the same graph as in Figure 15.6.1 where n = u.
(See Figure 15.6.3.)
under U.S. or applicable copyright law.

Note that the total number of disconnected components for a given n equals (n (n +

We wish to indicate an alternative way of thinking of Figure 15.6.2(b) drawing on

the intuition of infinitesimals. We adopt the language of [Ke]. In Figure 15.6.4, we
use an "infinitesimal microscope" to magnify part of the monad of T. The loops of
Figure 15.6.2(b) are replaced by the edges connecting n — k distinct vertices within the
monad of T. (Compare the methods of Example 16.2.9.)
We now consider generalized Feynman diagrams associated with the generalized
Dyson
EBSCO series (15.5.5)
Publishing : eBookobtained
Collection in Example- printed
(EBSCOhost) 15.5.1.onRecall
6/8/2017that
5:47 nPM =
viauUNIVERSIDAD
+ Ehp=1 wpd T p
withDISTRITAL
0 < T1 < . . JOSE
FRANCISCO . < DETh < t. Figure 15.6.5 gives the graph corresponding to the
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 GENERALIZED FEYNMAN DIAGRAMS 449
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 15.6.3. The four graphs corresponding to n = u + wdr with n = 3

under U.S. or applicable copyright law.

FIG.EBSCO
15.6.4. The :instant
Publishing T in Figure
eBook Collection 15.6.2- under
(EBSCOhost) printedan
on infinitesimal
6/8/2017 5:47 PMmicroscope
via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 450 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 15.6.5. A generic graph corresponding to n = u +

(generic) term
under U.S. or applicable copyright law.

We close this section by drawing the graph associated with the term

of the Dyson series (15.5.42) obtained in Example 15.5.9. This is given in Figure 15.6.6
and provides one illustration of a generalized Feynman graph involving several 9s.
The reader may have noticed that our Feynman diagrams are drawn to correspond
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
to the "natural
DISTRITAL physical
FRANCISCO JOSE order"
DE CALDASas discussed in Remark 15.3.7, Example 15.3.8, Remark
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS 451
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

FIG. 15.6.6. An example with multiple 9s

15.4.4, and below in Section 15.7; to obtain the order resulting directly from calculating
the Wiener integral of functionals F as in Section 15.4, simply read our diagrams from
top to bottom.

15.7 Commutative Banach algebras of functionals: The disentangling algebras

under U.S. or applicable copyright law.

The class of functionals considered in Section 15.4— equipped with a natural norm that
we shall soon define—forms, for every t > 0, a commutative Banach algebra At under
pointwise multiplication. This Banach algebra structure along with our earlier results will
help us to make rigorous certain parts of Feynman's operational calculus [Fey8]. Indeed,
we have made related comments in Section 15.1 and we shall make additional remarks in
this section. We will develop this topic much further in Chapter 18—where two auxiliary
noncommutative operations will play a key role. As we will see in Chapter 18, the fact
that we have at our disposal a one-parameter family of algebras (At : t > 0}, indexed
by time t, will be:crucial
EBSCO Publishing to our discussion
eBook Collection (EBSCOhost) - of theseonnoncommutative
printed 6/8/2017 5:47 PM via operations.
UNIVERSIDAD In the
present section,
DISTRITAL however,
FRANCISCO JOSE DE tCALDAS
will be fixed.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 452 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The disentangling algebras At

Given t > 0, the Banach algebra At will consist of equivalence classes of functions,
where the equivalence is that given in Definition 15.2.3. (At this point, the reader
may wish to review this definition as well as Example 15.2.4 and Theorem 15.2.5.)
The equivalence relation ~ is compatible with the algebraic operations as we observed
just after Definition 15.2.3. Following the common convention, we will often blur the dis-
tinction between equivalence classes and representatives. However, there will be places
in this section and in Chapter 18, specifically, Sections 18.2 and 18.3, where we will
need to keep the distinction in mind.
For the reader's convenience, we begin by recalling some facts and notation from
Section 15.4.
Let {Fn}8n=0 be a sequence of functionals each of which is given by an expression of
the following form:

with mn a nonnegative integer, nn,u e M(0, t) and 0n,u e L001;nn,u. Assume that

Then define a functional F as in Corollary 15.4.3 by

where y > 0 and (x, E) e Ct0 x Rd (so that y-1/2x + £ e Ct).

Recall that for every X > 0, the series in (15.7.3) converges absolutely for m x Leb.-
a.e. (x, £) e Ct0 x Rd.
Denote by At the set of all equivalence classes of functionals F obtained in this
manner. It is important to realize that the representation of functionals in (15.7.3) and
even in (15.7.1) is not unique. For F in At, we let ||F||t be the infimum of the left-hand
side of (15.7.2) over all representations of F of the form (15.7.3).
under U.S. or applicable copyright law.

We make two observations concerning the equivalence relation ~ and representations

of functionals as in (15.7.1)-(15.7.3). These facts follow trivially from the transitivity
property of ~; nevertheless, it is helpful to keep them in mind:
(i) If F ~ G, then F and G have exactly the same set of representatives as described
in (15.7.1)-(15.7.3).
(ii) Let the equivalence classes [F] and [G] belong to At. If there exist F1 6 [F]
and G1 e [G] such that F\ and G\ have in common even one representation as
described in (15.7.1)-(15.7.3), then [F] = [G].
NowPublishing
EBSCO we come: to theCollection
eBook main result of this- section
(EBSCOhost) ([JoLa1,
printed on 6/8/2017 Theorem
5:47 PM via 6.1, p. 70]). (Our
UNIVERSIDAD
proof of theFRANCISCO
DISTRITAL completeness will be more detailed than its counterpart in [JoLa1].)
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS 453
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 15.7.1 (The disentangling algebras {At}t>o) For each t > 0, the space
(At, || . ||t) is a commutative Banach algebra with identity.
Moreover, given F in At, Kty (F) exists for all y e C~+and satisfies the norm estimate

Further, for every choice of a representation of F, the operator Kty (F) is given (or
disentangled) by the corresponding generalized Dyson series (15.4.21).
Proof We begin by showing that if ||F||t = 0, then F is equivalent to 0. Let p be any
positive integer. Since \\F\\t = 0, there exists a representation for F given by (15.7.1)
and (15.7.3) such that the left-hand side of (15.7.2) is less than 1 / p . We see from the
inequality (15.2.15) in Lemma 15.2.8 that, for every y > 0, \ F ( y - 1 / 2 x + £)| < 1/P
for a.e. (x, £). Since p is arbitrary, it follows that, for every A > 0, F ( y - 1 / 2 x + £) = 0
for a.e. (x, £). Hence F is equivalent to 0.
Next suppose that F and G are in At. We claim that FG is in At and that

Given e > 0, take a representation for F defined by (15.7.1) and (15.7.3) such that the
left-hand side of (15.7.2) is less than ||F||t + e. Choose a similar representation for G.
Then, for every y > 0 and a.e. (x, E), we have

and

where the convergence is absolute. It follows that, for every A > 0 and a.e. (x, E), the
series, written by using the diagonalization procedure,
under U.S. or applicable copyright law.

converges absolutely and has the sum F(y - 1 / 2 x + £ ) G ( y - 1 / 2 x + E). Further, each
term F j ( y - 1 / 2 x + E ) G k ( y - 1 / 2 x + E) in (15.7.6) is of the type (15.7.1). Consequently,
FG e At and

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
Inequality
DISTRITAL (15.7.5)
FRANCISCO now follows
JOSE DE CALDAS since €was arbitrary.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 454 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Using arguments similar to some of those above, it is now easy to show that (At,||.||t)
is a normed linear space. To show completeness, it is convenient to use the fact that a
normed linear space is complete if and only if every absolutely summable series is
summable in the norm of the space [Roy, Proposition 6.5, p. 124]. Let {Fj} be a sequence
from At such that

Then for each j, we can choose representations of the form (15.7.3) and (15.7.1),

such that for each j, the left-hand side of (15.7.2) is less than || Fj||t + 2-j. The terms
of the series

are, of course, of the form (15.7.1), and the corresponding series of the form (15.7.2)
converges to a number less than £8j=1||Fj||t+ 1. Thus, by Corollary 15.4.3, the series
(15.7.8) with its terms evaluated at y-1/2x + § is absolutely convergent for every y > 0
and m x Leb.-a.e. (x, £) in Ct0 x Rd. Let
under U.S. or applicable copyright law.

We see from the discussion above that F e At.

What remains is to show that ||F - Enj=1 Fj||t -> 0 as N -> oo. Given
€ > 0, it follows from (15.7.7) that we can choose N0 large enough so that
E8j=N0+1 ||Fj||t + 2-j) < €. Using the definition of the norm || . ||t, it is now not
difficult to show that for N > N0, we have

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
We DISTRITAL
leave this last step
FRANCISCO JOSEtoDEthe reader.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS 455
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The fact that Kty (F) exists for all y e C~+ and is given by a generalized Dyson series
has already been shown in Corollary 15.4.3. (Also, recall from Theorem 15.2.5 that
Kty (F) does not depend on the choice of representative in the equivalence class of F.)
Finally, the estimate (15.7.4) follows from the fact that the norm estimate (15.4.22) in
Corollary 15.4.3 holds for any representation for F. This concludes the proof of Theorem
15.7.1.
Remark 15.7.2 Banach algebras smaller than but related to At were studied by Johnson
and Skoug in [JoSk2,6]. It is the p = 2 case of [JoSk6] which compares most directly
to [JoLaI, §67 and to our present discussion. There was an error made in [JoSk6] in
connection with the definition of the norm on the Banach algebra. This difficulty is easily
corrected as we pointed out in Remark 6.1(a) of [JoLa1], and it is then easy to see that
the Banach algebra At = A from [JoSk6] is a subalgebra of At. (More specifically, A,
is the subalgebra obtained by taking all measures nn,u equal to Lebesgue measure l in
the definition of At; see equations (15.7.1)-(15.7.3).)
Various formulas from [JoSk2,6] are, from the perspective of our present work, for-
mulas for disentangling. The authors of [JoSk2] and [JoSk6] did not think of their work
in these terms although they did suspect that it had some connection with Feynman's
time-ordering ideas. The 1 < p < 2 case of [JoSk6] as well as [JoSkS] show that
mathematically rigorous meaning can be given via functional integration to parts of
Feynman's operational calculus even when the space of functionals involved is not an
algebra. However, the presence of a Banach algebra structure permits a fuller develop-
ment of the ideas from [Fey8],
Work of Cameron and Storvick beginning with [CaSt1] influenced ([JoSk2,6],
[JoSk5]) and so had an influence on the memoir by Johnson and Lapidus [JoLa1] and
hence on the present chapter.
Our first corollary follows immediately from Theorem 15.7.1.
Corollary 15.7.3 For each fixed y 6 C~+, the mapping Kty : At -> L ( L 2 ( R d ) ) which
associates Kty (F) with F e At, is a bounded linear operator of norm at most 1, and
equal to 1 if y is purely imaginary.
Our second corollary follows from well known facts about the holomorphic func-
tional calculus in Banach algebras ([KadRi, §3.3] or [Nai, pp. 202-205]).
under U.S. or applicable copyright law.

Corollary 15.7.4 Let F E At and let f be a complex-valued function of a complex

variable which is analytic in a disk about the origin with radius strictly greater than
||F||t. Then the function f o F is also in At.

The time-reversal map on At and the natural physical ordering

In the following, we further address—within the context of the Banach algebra At—the
twin issues of the time-reversal operation and of the natural physical ordering of operators
that were already discussed in a closely related context in Remark 15.4.4 above (and,
in a much simpler situation, in Remark 15.3.7 and Example 15.3.8). We note that this
material was not included
EBSCO Publishing in [JoLal].
: eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 456 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Let F be an arbitrary element of At. Then the "time-reversal" F of F is defined as

follows. In each representation of F of the form (15.7.3), we substitute the pair (0n,u, ~nn,u)
for (0n,u nn,u), where 0n,u and~nn,uare the time-reversals of 0n,u and nn,u, respectively,
as defined by (15.4.24). More precisely, if F is given by (15.7.1)-(15.7.3), then F is
given by (15.4.26); namely,

where A > 0 and (x, £) e Ct0 x Rd. (This definition is compatible with the equivalence
relation ~, as we will soon see.) Moreover, by (15.4.25), the counterpart of (15.7.2) still
holds, so that F also belongs to At and ||F||t = ||F||t.
We can now formally define the "time-reversal" operation T as a map from the
Banach algebra At onto itself.
Definition 15.7.5 The time-reversal map T: At -> At is defined by

where F is given as in (15.7.10) above. The first assertion in Theorem 15.7.6 below will
assure us that T is well defined on the set of equivalence classes that make up At.
The following result explains our interest in the map T.
Theorem 15.7.6 (i) The time-reversal map T is well defined on At. Further, it is an
involution on At (i.e. ~F = F for all F e At) and is an isometric isomorphism of
the Banach algebra (At, || . ||t) onto itself.
(ii) For every F e At and X € C~+, we have

the Hilbert adjoint of the bounded linear operator Kt/y (F), where F (e At) denotes
the complex conjugate of F.
Moreover, by Remark 15.4.4, Kty (F) can be disentangled (as in Theorem 15.7.1)
via a GDS of the type (15.4.21), except that each term in the time-ordered perturbation
under U.S. or applicable copyright law.

series (15.4.21) is now written in the natural physical ordering; that is, in agreement
with Feynman's time-ordering convention (14.2.1), the operators involving smaller time
indices always act before those involving larger time indices.
Proof It follows easily from observation (ii) just before Theorem 15.7.1 that the map T
respects the equivalence relation ~; that is, F ~ G implies that F ~ G. In other words,
given an equivalence class [F] e At, it makes sense to define [F] as [F]. Thus T is a
well defined map on the set of equivalence classes that make up At •
The equation F = F follows from (15.7.10) and the fact that 0n,u = 0n,u and
nn,u EBSCO
= nPublishing
n,u for each n and
: eBook u. It is(EBSCOhost)
Collection clear that- the maponT6/8/2017
printed is linear
5:47and onto,
PM via and it follows
UNIVERSIDAD
fromDISTRITAL
(15.4.25) and (15.3.24)
FRANCISCO that it is an isometry.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS 457
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The last claim made in (ii) of the theorem was established in Remark 15.4.4.
(The combinatorial complications involved in (15.4.21) through the associated formula
(15.4.6) tend to obscure the simplicity of the idea. The reader may find it helpful to
compare formulas (15.3.26) and (15.3.28) of Example 15.3.8.)
The proof of the theorem will be complete if we establish the equality (15.7.12); to
do this, it suffices to show that

Recall from (15.4.21) that Kty (F) = E8n=0 Kty (Fn), where the series converges in oper-
ator norm. Since the adjoint operation on L(L 2 (R d )) is continuous (because it is a
conjugate linear isometry by Proposition 9.6.2), it suffices to establish the desired for-
mula for F = Fn. We will do this by using the formula (which will be discussed at the
end of this proof)

where n is a complex measure on the measurable space (Y, y) and g : Y -> L ( H ) is

Bochner integrable in the strong operator sense with respect to n. [Here, H is an abstract
complex separable Hilbert space (rather than L2 (Rd)).]
What happens when the formula above for the adjoint of the integral is applied
along with the conjugate linearity of the adjoint operation to the formula for Kty (Fn)
in (15.4.6) with the Lrs given by (15.4.8)? The (as and the us are conjugated and the
product Lk . . . Lm is replaced by L*m . . . L*k. Further, each L*r is the product in reverse
order of the adjoints of the operators appearing on the right-hand side of (15.4.8). Since
the adjoints of the multiplication operators 0(s, .) are given by 0(s, .)* = 0(s, .) (by
Theorem 10.1.11(i) or Remark 10.1.4) and since ( e - s ( H 0 / y ) * =e-s(H0/y)for y e C~+
and s > 0 (by Theorem 10.1.11(i) with g(x) := e - s x / y for x e d(H0) C R, and so
g(x) = e-sx/y), we obtain the formula Kty (F)* = Kt/y (F) and hence (15.7.12). (The
reader may find it helpful to calculate Kty (F)* in the simple but instructive case where
F is given by (15.3.25) in Example 15.3.8.)
We now discuss briefly the proof of the formula given above for the adjoint of the
integral. Since the integral is in the strong operator sense but the adjoint operation fails
under U.S. or applicable copyright law.

to be continuous in the strong operator topology [DunSc1, p. 513], the limiting argument
that may first come to mind fails. However, one can quite easily prove that the operator
fy g * ( y ) n ( d y ) acts as the adjoint of fY g ( y ) n ( d y ) should; that is, for every o, U e H,

where (., .)H denotes the inner product on H. Actually, it is enough to show this last
equality when n is a positive measure since the general case then follows from the
linearity (in the measure)
EBSCO Publishing of the (EBSCOhost)
: eBook Collection integral, the conjugate
- printed linearity
on 6/8/2017 ofvia
5:47 PM *, UNIVERSIDAD
and the fact that
DISTRITAL
every FRANCISCO
complex JOSE DEn CALDAS
measure can be written as n = n1 — n2 + i(n1 - n4), where nj is a
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 458 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

positive measure for j = 1, . . . , 4 [Coh, p. 126]. The equality of the two inner products
above follows from the theorem [HilPh, Theorem 3.7.12, p. 83] allowing one to take a
closed linear operator, and so certainly a continuous linear functional, under the integral
sign.
Remark 15.7.7 (a) When Kty (F) from Theorem 15.7.6 is expressed in terms of the
pairs (0n,u, n n , u ), the natural physical ordering is not achieved. Returning once again
to Example 15.3.8 for an illustration, we have

In contrast, when Kty (F) is written in terms of the pair (0, u), we do have the natural
physical order as we saw in equation (15.3.28) of Example 15.3.8:

(b) While we were writing [JoLa1], we knew from Feynman's ideas, from earlier work
such as [CaSt1, JoSk2, JoSk6], and especially from work then in progress [La14-16,
La18] that the time-ordering in the GDS for Kty (F) is exactly the reverse of the natural
physical order. We also realized that taking the adjoint would produce the desired order.
However, we thought mistakenly that taking the Banach space adjoint rather than the
Hilbert space adjoint would yield the right order (see [JoLa1, Remark 1.5, p. 25]) without
introducing complex conjugates on 0, n] and y. This is correct when 0, n and y are all
real-valued, but is not so otherwise. This error was carried on in [La14-18], but since
the GDS actually being used was the physically natural one, no serious harm was done.
We will study the material from [La14-18] in Chapter 17 below, but we will of course
use Kty (F) and not the Banach space adjoint.
Corollary 15.7.8 Let R : At -> A, be defined by
under U.S. or applicable copyright law.

[In other words, R is defined as T above except that each pair (0n,u nn,u) is replaced
by (0n,u, nn,u) instead of (O n,u , n n , u ).]
Then:
(i) (At, || . | | t ) , equipped with R, is a (commutative) complex Banach algebra
with involution, in the sense of [KadRi, §4.1, (i)-(iii), p. 236] (that is, F## =
F, ( F G ) # = G#F#(= F#G# here) and (aF + BG) # = aF# + BG#, for all
EBSCOF, G e At: eBook
Publishing and a, B € C).
Collection Further,
(EBSCOhost) R is an
- printed isometric
on 6/8/2017 5:47anti-isomorphism
PM via UNIVERSIDAD from
(At, FRANCISCO
DISTRITAL || . ||t) onto itself.
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS 459
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(ii) For every F e At, we have

the (Hilbert) adjoint of Kty (F). (In particular, for y > 0, the adjoint of Kty (F) is
given by Kty (F#).)
Consequently, Kty (F)* = Kty ( F # ) can be disentangled via a GDS of the form
(15.4.21), except with the operators in their natural physical ordering (and the potentials
0n,u and measures nn,u replaced by their complex conjugates).
Proof (i) follows easily from part (i) of Theorem 15.7.6. Note that by (15.7.13) and the
definition of the norm || . ||t,

for all F e At.

(ii) follows immediately from (15.7.13) and (15.7.12), as well as from Theorem
15.7.1.
Remark 15.7.9 Clearly, (At, || . ||t, R) is not a C*-algebra since, in general, ||FF#||t
does not equal ||F||2t for F e At, as the reader will easily verify. (See, for example,
[KadRi, § 4.1, (i)-(iv), p. 236] for the definition of a C* -algebra.)
Exercise 15.7.10 Show that the algebra At from [JoSk6] (and defined in Remark 15.7.2
above) is a subalgebra of At which is stable under both of the maps T and R.
Connections with Feynman's operational calculus
We close this section by comments that may help put our work in this chapter in a broader
context and connect it with material discussed in later chapters.
One of the purposes of a functional calculus is to allow one to form a rich class of
functions of the objects of interest. In the traditional cases, the Feynman-Kac formula
(y = 1) and the Feynman integral (y = —i), one considers the exponential of
under U.S. or applicable copyright law.

where / is Lebesgue measure. The present theory is more general in many respects. In
[JoSk6, pp. 121-123], one could already form rather general analytic functions of Gl.
Here, however, we are able to treat analytic functions of

where n e M(0, t) is arbitrary. The significance of this additional flexibility is seen

throughout the present chapter and much of the rest of the book.
We note
EBSCO that we
Publishing can Collection
: eBook actually deal with analytic
(EBSCOhost) functions
- printed on of a PMgeneral
6/8/2017 5:47 element of the
via UNIVERSIDAD
Banach algebra
DISTRITAL At as
FRANCISCO inDECorollary
JOSE CALDAS 15.7.4. Further, it follows from the spectral theory of
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 460 FEYNMAN'S OPERATIONAL CALCULUS AND DYSON SERIES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Banach algebras—more specifically, from the holomorphic functional calculus in that

context ([KadRi, §3.3] or [Nai, pp. 202-205])—that it suffices for the function / of
Corollary 15.7.4 to be analytic in a neighborhood of the spectrum of F. We can also take
functions of infinitely many variables as was done in Exercise 15.5.12.
As commented on earlier, Feynman [Fey8] observed that disentangling is the key
to his operational calculus. Within the setting of nonrelativistic quantum mechanics
(alternatively, the diffusion equation (12.1.5)) and multiplication operators as in the
present chapter, our generalized Dyson series (GDS) provide a method of disentangling
in a rather broad context. The general theory to be presented in Chapter 19 is in many
respects broader still but does not have the analytic power connected with the Wiener
integral. (See, in particular, Theorems 16.2.3 and 16.2.6 as well as Examples 16.2.7 and
16.2.11 in Chapter 16 below which illustrate the use of the dominated convergence for the
Wiener integral in this context.) In the present setting, even when X is not real, the explicit
nature of the GDS for all y e C~+ will have its advantages. We will see illustrations of this
in Chapters 16, 17 and 18. It will allow us, in particular, to make limiting arguments valid
for both the diffusion and quantum-mechanical case. [See Example 16.2.9 as well as the
"Feynman-Kac formula with a Lebesgue-Stieltjes measure" (FKLS) in the general case
(Section 17.6 and [Lal8]).] The generalized Dyson series will also be used to obtain the
results in Chapter 17 on the FKLS from [Lal4-16, Lal8] (see especially Sections 17.3
and 17.6) and further used in Chapter 18 in relation with the disentangling algebras At
[JoLa3,4] (see Section 18.5).
It is important to note that the process of disentangling is not unique. Example
15.5.5 provides a simple illustration of this. Compare formula (15.5.16) in the case
where f ( z ) = exp(z) (so that n!an = 1) with formula (15.5.18). These two time-
ordered expressions for the same operators Kty (F) are quite different from one another.
Examples 16.2.7 and 16.2.11 in the next chapter, as well as the results from [Lal4-16,
La18] discussed in Chapter 17, will provide further illustrations of nonuniqueness. When
there is more than one disentangling of the operator Kty (F), the most useful choice may
depend on the goal being pursued.
Even though F e At does not uniquely determine the disentangling of Kty (F), the
operator Kty (F) itself is unambiguously associated with F. This follows from the fact
that Kty (F) is defined by the Wiener integral (15.2.8) for y > 0, followed by analytic
continuation for X € C+ and strong continuity for y e C~+.
It is clear that the process of disentanglement, which might be viewed as the main
under U.S. or applicable copyright law.

theme of the present chapter, is intimately related to the noncommutativity of the opera-
tors involved. Since the algebra At is commutative whereas £(L 2 (R d )) is not, we would
not expect the linear mapping Kty defined in Corollary 15.7.3 to preserve multiplication.
Indeed, it does not as the following simple example shows.
Example 15.7.11 Take F = 1. Then F2 = F and, for all y e C~+,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS 461
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the last equality comes from (15.2.11). On the other hand, by the semigroup
property of e-tH0/y,

Thus, we see that (for t > 0)

We will introduce in Chapter 18 an auxiliary noncommutative operation * : At1 x

At2 -> At1+t2 which will fit well with Feynman's operational calculus and which will
allow us to write (for t1 , t2 > 0 and y e C~+)

The operation * is a type of multiplication but is rather different from the pointwise
product of functions involved in Example 15.7.11. Nevertheless, it is interesting to note
the formal contrast between (15.7.20) and Example 15.7.1 1.
As we will see in Chapter 18, this "noncommutative multiplication" * —along with
its companion operation +, a kind of "noncommutative addition"—will enable us to
take into account in a more complete way the noncommutativity underlying Feynman's
operational calculus.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

16

STABILITY RESULTS

The functionals introduced in the previous chapter and studied in Chapters 15-18 are
defined in terms of measures and potentials. It is natural to ask if the corresponding
operators are stable under perturbations of either of these objects. First, we consider in
Section 16.1 stability with respect to the potentials. We then consider in Section 16.2
stability with respect to the measures.
Apart from some extensions discussed at the end of Section 16.2, the results of
this chapter are due to Johnson and Lapidus in [JoLa1, §4]. Those in Section 16.1 are
motivated by and extend to a broader class of Wiener functionals the stability theorem
(for analytic operator-valued Feynman integrals) of Johnson in [Jo3] and [JoSk2, p = 2
case]. Further, the stability theorems with respect to the measures obtained in Section 16.2
will enable us to reinterpret in a new light and unify various phenomena connected with
discrete and continuous measures. (This theme will be pursued in parts of Chapter 17.)

16.1 Stability in the potentials

For much of the notation in our first theorem, we refer the reader to Corollary 15.4.3.
Theorem 16.1.1 Let nn,u e M(0, t) and let 0n,u e L 0 0 1 n , u . Assume that the
hypotheses of Corollary 15.4.3 are satisfied for the family of potentials ®n,u. Let
0n,u(m), m = 1 , 2 , . . . , be Borel measurable functions on (0, t) x Rd such that, n x Leb.
almost everywhere,

and
under U.S. or applicable copyright law.

Then0n,u,0(m)n,uall belong to
Further, let Fn (resp.,Fn(m)) be the functional associated with the potentials
&n,u (resp., 0n,u(m)) and the measure nn,u as in equation (15.4.18) of Corollary 15.4.3.
Moreover, let F (resp., F(m)) be the functional defined in terms of the sequence {F n }8 n=0
(resp., {Fn(m)}£n=0) via equation (15.4.20).
Then, for all y € C~+, Kty (F), Kty (F(m)) exist and

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: is,
(That 98476
the; convergence
Johnson, Gerald in
W.,(16.1.1)
Lapidus, Michel
holds L..;
whenTheapplied
Feynman Integral
to any and U in L2(Rd).)
Feynman's
vector
Operational Calculus
Account: ns000601
 STABILITY IN THE POTENTIALS 463
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Further, the form of the generalized Dyson series is preserved. More precisely, for
each n, the combinatorial form of the nth term can be taken to be the same for all of the
GDSs involved.
We omit the proof of Theorem 16.1.1 since it is, except for notational complications,
essentially like the proof of our next result. In the following theorem, we treat the special
case considered in Theorem 15.3.1 and Corollary 15.3.4 and we use our earlier notation.
Theorem 16.1.2 Let n = u + wdT as in Theorem 15.3.1 and let ® € Loo1;n-
Let e (m) , m = 1, 2, . . . , be Borel measurable functions on (0, t) x Rd such that,
n x Leb.-a.e.,

and

Then 0, 0 (m) belong to Loo1;n. Moreover, let

where f is given by (15.3.13) and has radius of convergence strictly greater than
||®||oo1;n. Let F(m) be defined as in (16.1.3) except with 0 replaced by 8 (m) .
Then, for all X e C~+, Kty (F), Kty (F(m)) exist and

Further, theform of the generalized Dyson series is preserved in thefollowing precise

sense:
under U.S. or applicable copyright law.

strongly as m -» oo. Here, Lj is defined by (15.3.4) and Cj(m) is given as in (15.3.4)

except with 0 replaced by 0(m).
Proof By (16.1.2), ||0(m)||oo1;n < ||O||oo1;n, form = 1, 2, . . . and 0(m), 0 lie in Loo1;n.
Using (16.1.2) and the Lebesgue dominated convergence theorem, we see that 0(m)(s) ->
0 ( s ) strongly as m -» oo, for n-a.e. s in (0, t). It follows that
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 464 STABILITY RESULTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Now, by (16.1.6) and the norm estimate (15.3.16), we have

Hence the conclusion follows from equations (15.3.1), (15.3.2), (15.3.15) and the domi-
nated convergence theorem for Bochner integrals [HilPh, Theorem 3.7.9, p. 83].
Remark 16.1.3 (a) The above proof shows that the form of each summand in (16.1.5)
is preserved. We should stress that, in all cases, for instance, in the examples of Section
15.5, the form of the generalized Dyson series is preserved just as in Theorem 16.1.2.
(b) In both of Theorems 16.1.1 and 16.1.2, if U (m) -> V in L2(Rd), then
in
It has been notoriously difficult to establish pleasant analytic properties of the
Feynman integral. As far as we know, the first reasonably satisfactory stability the-
orem for the Feynman integral appeared in Johnson's paper [Jo3]. The results in [Jo3]
and [JoSk11, p = 2 case] are special cases of Theorem 16.1.2. The paper [JoSk11] also
includes results where the "Feynman integral" was interpreted as a bounded linear oper-
ator from L P ( R d ) to Lp ( R d ) , 1 < p < 2, where1/p+ 1/p' = 1. In [JoSk11], restrictions
are placed on the dimension d which become more and more serious as p gets closer
and closer to 1. The p = 1 case was covered in [Cha].
Recall that Theorems 11.5.13 and 11.5.19 above gave a stability theorem for the
modified Feynman integral which permitted highly singular potentials. (See Section
11.5.) This result is due to Lapidus [La 12]. In [Lal2, pp. 58-59] (or [La9]), one can also
find an operator-theoretic proof of the result of [Jo3].

16.2 Stability in the measures

We give a simple theorem establishing stability in the measures and use it, in particular,
to make connections with the Trotter product formula (Example 16.2.7) and to explain
the link between the absolutely continuous case and the discrete case (Example 16.2.9).
We do not strive here for maximum generality; indeed, additional study of the topics in
this section might make a good subject for further research.
Our next theorem involves the concept of weak convergence of measures.
under U.S. or applicable copyright law.

Definition 16.2.1 Let n, nm, m — 1, 2, . . . be in M(0, t). We say that nm converges

weakly to n and write nm -> n, provided that

for every bounded continuous function b on (0, t ).

Remark 16.2.2 The term "vague convergence" is sometimes used instead of weak con-
vergence. Weak convergence is weak* convergence from the functional analysis point of
view since the space M(0, t) is a closed subspace of the dual of the Banach space of all
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
bounded continuous
DISTRITAL functions
FRANCISCO JOSE DE CALDASon (0, t). The concept of weak convergence of measures
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 STABILITY IN THE MEASURES 465
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

makes sense well beyond the real line; a common setting for its study is a separable met-
ric space. This notion of convergence is important in the probability literature where it is
restricted to probability measures and often referred to as "convergence in law". Most
of the results in the probability literature carry over easily to finite, positive measures.
Hence, many things carry over as well to complex measures of finite total variation since
each such measure is a linear combination of four positive measures [Coh, p. 126]. Two
good references for weak convergence of measures are the books by Billingsley [Bil] and
Dudley [Du, esp. Chapters 9 and 11].
Theorem 16.2.3 Let 0 be a continuous function bounded by a finite constant c on all of
(0, t) x Rd. Let n, nm, m = 1, 2, . . . , be in M(0, t). Assume that

Put

where f is an entire function given by (15.3.13). Let Fm be defined as in (16.2.3) except

with n replaced by nm. Then

uniformly in y on all compact subsets of C+.

Proof Fix U E L 2 (R d ) and E e Rd. Let X > 0. In view of (15.2.8),

and similarly for F except with nm replaced by n. Given x e Ct0, the function
0 ( s , y - 1 / 2 x ( s ) + £) is bounded by c and is continuous as a function of s. Hence, by
hypothesis (16.2.2),
under U.S. or applicable copyright law.

Since / is continuous,

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 466 STABILITY RESULTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Note that

where c1 = c supp ||np||. Since a weakly convergent sequence of measures is bounded

in total variation norm, we have c1 < oo. Thus,

where c2 = sup |z| < Cl |f(z)| < oo.

Recall from (15.2.11) that |U(y 1 / 2 ;t(f) + £)| is Wiener integrable and

In view of (16.2.5), (16.2.7) and (16.2.9), the dominated convergence theorem yields

as m ^ oo for a.e. £ e Rd. Moreover, by (16.2.9) and (16.2.10),

A second application of the dominated convergence theorem yields

in L2(Rd) as m ->• oo for all y > 0.

Now, Kty (Fm) U is analytic for y e C+ and, by (16,2.12) and (15.2.1), we have

Hence, by (16.2.13) and (16.2.14), Vitali's classical theorem (Theorem 11.7.l(i) and
Remark 11.7.2(c)) gives the result for y e C+.
Our most interesting application of Theorem 16.2.3 will come in Example 16.2.9.
There we will see how the Dyson series associated with certain absolutely continuous
under U.S. or applicable copyright law.

measures nm converges to the combinatorially very different Dyson series associated

with the Dirac measure dr as nm converges weakly to dT.
It may well be that deeper facts about weak convergence of measures can be used
to extend Theorem 16.2.3 substantially. At the end of this section we will state a slight
adaptation of part of a result in [Bil, Theorem 5.2, p. 31] which will allow us to make a
modest start in that direction.
Observe that the conclusion of Theorem 16.2.3 is stated only for X 6 C+. The
next proposition gives some information for y purely imaginary (much as was done in
Appendix 11.7, especially in the proof of Proposition 11.7.4, in the related context of
unitary groups of operators). For notational convenience, we write y = iy with y e R.
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
2 d
In the statement
DISTRITAL of Proposition
FRANCISCO JOSE DE CALDAS 16.2.4, (., .) denotes the inner product in L (R ).
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 STABILITY IN THE MEASURES 467
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proposition 16.2.4 Assume that the hypotheses of Theorem 16.2.3 hold. Let U1, U2 be
in L2(Rd). Then for all o e L1(R.),

Proof The functions Kty (Fm) and Kty (F) are analytic in C+ and strongly continuous
in C~+. Moreover, by (16.2.14), \\Kty (F m )U\\ < c2. By the extended Vitali theorem dis-
cussed in Appendix 11.7 (see Theorem 11.7.l(ii) and Remark 11.7.2(c)), the conclusion
follows. (Note that, even though Kty (F) is not defined for y = 0, it is in H°°, the space
of (essentially) bounded analytic functions on the right half-plane. In view of [Dure,
Theorem 1.1, p. 2], the argument in the proof of Theorem 11.7.1(ii) still applies to this
situation.)
The convergence in (16.2.15) is in too weak a sense to be widely useful. However,
Proposition 16.2.4 does yield the following corollary via standard functional analysis
arguments. We omit the proof.
Corollary 16.2.5 Let the assumptions of Theorem 16.2.3 hold. If the strong limit of
K t i y (F m ) exists as m ->• oo and equals A(iy) (in £(L2(Rd)), then A(iy) = Ktiy (F)
for Leb.-a.e. y in R.
If we further assume that the limit function A(i y) is strongly continuous on R as a
function o f y , then A(iy) — Ktiy (F) for every nonzero y in R.
If we make the much stronger assumption that n,n —» n in norm, we can show without
any continuity assumption on 9 that Kty (Fm) —»• Kty (F) in operator norm.
Theorem 16.2.6 Let 9 : (0, t) x Rd -> C be everywhere defined, Borel measurable
and bounded by B and suppose that f is analytic at least in a circle of center at 0 with
radius greater than B \\n\\. Finally, suppose that \\nm — n\\ -> 0 as m —>• oo. Then

uniformly in A on all compact subsets 0f C+.

Moreover, for all X > 0, we have the norm estimate
under U.S. or applicable copyright law.

where

with

Proof
EBSCOSince || nm: —
Publishing n || Collection
eBook -» 0 as m(EBSCOhost)
-»• oo, we have that,
- printed at least
on 6/8/2017 5:47for
PM large m, f is analytic
via UNIVERSIDAD
in aDISTRITAL
circle centered at 0 DE
FRANCISCO JOSE with radius greater than B \\ nm \\. We will assume throughout the
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 468 STABILITY RESULTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

rest of the proof that m is large enough so that this is so. Now for any x 6 Ct, we clearly
have

where Mm is given by (16.2.19); thus, for U e L 2 (R d ), a.e. £ € Rd and X > 0,

where cm is defined by (16.2.18). Now by (16.2.5), (16.2.21), and (16.2.10), we have,

for X > 0,

and thus, since \\e~ t(Ho/y) || < 1,

Since U is arbitrary in L2(Rd), (16.2.17) follows for A > 0. Now since Mm -» 0 as

m ->• oo and / is uniformly continuous on the closed disk \z\ < B\\n\\, we see that
cm -» 0 as m —»• oo and so (16.2.16) follows for y > 0. Since strong analyticity
is equivalent to analyticity with respect to the operator norm [HilPh, Theorem 3.10.1,
p. 93] and since ||Kty((Fm)|| < B||nm|| < B[\\n\\ + 1] (at least for larger m), Vitali's
theorem (Theorem 11.7.1(1)) yields (16.2.16) for all X 6 C+.
We now present some applications of Theorem 16.2.3. We assume the conditions of
that theorem and use the corresponding notation insofar as possible. Our first example
relates our point of view to the Trotter product formula and the classical Feynman-Kac
formula.
under U.S. or applicable copyright law.

Example 16.2.7 (Trotter product formula and classical Feynman-Kac formula) Let
/(z) = exp(z), y = 1, and suppose that 6 — —V, with V time independent. Fur-
ther, let

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 STABILITY IN THE MEASURES 469
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Observe that

since, for every bounded continuous function b on (0, t), we have the Riemann sum
approximation:

as m -» oo. Thus, by Theorem 16.2.3,

strongly, where F(x) := exp(- f(0,t) V(x(s))ds).

If we assume the Trotter product formula (Theorem 11.1.4), that is,

strongly, then from (16.2.27) we have

We thus recover the classical Feynman-Kac formula (Theorem 12.1.1)

where U € L2(Rd) and £ e Rd.

Actually, we do not need the full strength of the Trotter product formula to obtain
(16.2.29); in fact, we know from Theorem 16.2.3 that the limit in (16.2.29) exists. By
[Cher2, Theorem 3.1, p. 34], the limit in (16.2.27) is a semigroup and is generated by
Ho + V; thus (16.2.29) follows.
As an alternative to the above, we can assume the Feynman-Kac formula (16.2.29)
and recover this concrete case of the Trotter product formula from (16.2.27).
under U.S. or applicable copyright law.

Of course, we know from our work in Chapters 11 and 12 that both the Trotter product
formula and the Feynman-Kac formula are valid under conditions on V that are far more
general than in this example. Further, in Chapter 12 we saw the Trotter product formula
used in the proof of the Feynman-Kac formula. (References to some of the earlier work
on these formulas and their relationship can be found in Chapters 11 and 12.) In spite of all
this, there is something to be gained from the present example. Our framework provides
a unifying point of view and a broader perspective. Both the approximating product and
the limit in (16.2.27) are Wiener integrals of expressions involving Lebesgue-Stieltjes
measures. Moreover, the approximating procedure—that is, essentially, the role played
by EBSCO
Lebesgue-Stieltjes measures in our theory—is reminiscent in some ways of that
Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
played by such
DISTRITAL measures
FRANCISCO JOSE DE in the context of one-dimensional integrals.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 470 STABILITY RESULTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 16.2.8 (a) We have restricted attention to X = 1 so that our formulas coincide
with the Feynman-Kac and Trotter product formulas as they are most often stated.
Actually, the argument works just as wellfor X € C+. For X purely imaginary, our method
gives only the rather weak result obtained from Proposition 16.2.4. Note, however, that
there is no problem in applying the Trotter product formula in this case since V is
bounded. The difficulty of going beyond Proposition 16.2.4 in this manner is closely
connected with the problem of passing from the Wiener integral to the Feynman integral
and of establishing the Trotter product formula for unitary groups in a general setting
(see Problem 11.3.9 above and [La6], as well as the discussion at the end of Appendix
11.7).
(b) A " Feynman-Kac formula with a Lebesgue-Stieltjes measure" [La14-18] will
be treated in detail in the next chapter. There the functional will have the form

The traditional Feynman-Kac formula is, of course, the special case where n = l. One
of the interesting phenomena that will be discussed in Chapter 17 is the distinct roles
played by the discrete and continuous parts of the measure n.
It is known that, for bounded time independent potentials, the Trotter product formula
is related to the product integral (see, for example, [DoFri3, §4.3, pp. 135-137]). A
product integral representation for Kty (F) is given in [La16] and will be discussed briefly
in Section 17.6; see especially Theorems 17.6.10 and 17.6.12. Here, the functional F is
the time-reversal of F (see Definition 15.7.5 and part (b) of Remark 15.7.7) and F is
given by (16.2.31).
Example 16.2.9 (Absolutely continuous measures approximating dT) Let [pm] be a 8-
sequence [Sta, pp. 106-117] centered at the point T in (0, t). For instance, let

Let nm be the absolutely continuous measure on (0, t) with density pm. Then, it is not
hard to show that
under U.S. or applicable copyright law.

Let y e C+ and f(z) = exp(z). Then Theorem 16.2.3 and formula (15.3.17) from
Corollary 15.3.6 yield:

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
strongly as m
DISTRITAL ->• oo,JOSE
FRANCISCO where An and £ are given by (15.3.18) and (15.3.19), respectively.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 STABILITY IN THE MEASURES 471
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Next, with pm given by (16.2.32), we discuss a direct derivation of (16.2.34) using

the generalized Dyson series for Kty (Fm). One purpose for this is to show that (16.2.34)
holds even for imaginary values of y. The (rough) principle seems to be that, when one
can work directly with the Dyson series, all y e C+ can be included. Note that, in the
proof of Theorem 16.2.3, Dyson series do not appear; the key step in that proof is the
use of the dominated convergence theorem for the Wiener integral. A second reason for
including the discussion to follow is that it may help the reader, as it did the writers, to
develop further intuition, not only for such limiting arguments, but also for some of the
new phenomena arising in this theory, such as the appearance of powers of the potential
evaluated at fixed times.
Because of the norm estimate given in Remark 15.3.5 for the tail of a Dyson series,
it suffices to show that the nth term of the Dyson series for Kty (Fm) converges strongly
to the nth term of the Dyson series for Kty (F). Let U e L2(Rd). In view of (16.2.34),
we must verify that, as m —»• oo,

Now,

as (s1 , . . . , sn) -»• (T, . . . , T), with (s1, . . . ,sn) e An. Hence, given e > 0, for large
enough m,(s1, . . . , sn) E An n [T — ( 1 / m ) , r + ( l / m ) ] n implies that
under U.S. or applicable copyright law.

For m so large that (16.2.37) holds, we can now use properties of pm and write:

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 472 STABILITY RESULTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Hence (16.2.35) is established and (16.2.34) holds for all y e C~+.

Remark 16.2.10 The proof above reveals rather nicely some of the intuitive ideas under-
lying the present theory. Taking m large forces all n interactions represented in the
expression for LU in (16.2.36) to take place very close to T and thus leads to [0(T)]n
in the limit. All the intermediate free evolutions are forced close to the identity operator.
The n! in the first expression of (16.2.38) may appear at first glance to be troublesome.
However, using the fact that the Lebesgue measure of [r — 1/m,r + 1/m]n is n! times the
Lebesgue measure of [T —1/m,r+1/m]nn An, one sees further on in (16.2.38) that the
initial n! is just what is needed.
By considering more elaborate examples in a like manner, we could presumably
acquire a deeper understanding of the expressions giving our generalized Dyson series
in the previous chapter and, in particular, of the appearance of powers of the potentials
as well as of the values of the combinatorial coefficients.
We next indicate how Theorem 16.2.6 can be used to obtain a new expression for
Kty (F) in the case when n is an arbitrary Borel measure (compare Example 15.5.7).
Example 16.2.11 (Approximation of an arbitrary n by measures with finitely supported
discrete part) Let n e M(0, t) and let 6 satisfy the hypotheses of Theorem 16.2.6; thus
8 is everywhere defined, Borel measurable and bounded by B. As in Example 15.5.7,
we may write n in the form
under U.S. or applicable copyright law.

Note that the times rp in (16.2.39) are not necessarily ordered.

Let nm in M(0, t) be defined by

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 STABILITY IN THE MEASURES 473
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

For our purpose, it is convenient to rewrite vm in the form vm =

where for a given m, ii is the permutation of {1, . . . , m} such that rpi(o) := 0 <

Observe that if {tp}8p=1 is an increasing sequence, then TT is the identity for any
positive integer m.
Clearly, nm converges to n in norm since || nm — n || =£8p=m+1|wp| —> O as m -*• oo.
It thus follows from Theorem 16.2.6 and from (15.5.5)-(15.5.7) in Example 15.5.1 (with
h replaced by m) that

as m -»• oo, uniformly in y on compact subsets of C+. Here,

with

and, for (s1, . . . , sqo) e A go; j 1 ,...,j m+1 and r = 0, . . . , m,

Moreover, for all y > 0, (16.2.17)-(16.2.19) of Theorem 16.2.6 as well as the

computation of \\nm - n|| above give the norm estimate
under U.S. or applicable copyright law.

where we have set

with

We can see directly from the above or alternately, from the proof of Theorem 16.2.6,
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
thatDISTRITAL
cm -> 0FRANCISCO
as m —> JOSEoo.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 474 STABILITY RESULTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 16.2.12 (a) If we add the assumption that 9 in Example 16.2.11 is continuous,
then equation (16.2.15) of Proposition 16.2.4 holds since ||nm — n|| —> 0 certainly
implies that nm —*• n.
(b) The nonuniqueness of the disentangling process was discussed in Section 15.7,
and examples were given or referred to. (Compare, for example, formulas (15.5.16)
and (15.5.18).) Example 16.2.11 in conjunction with Example 15.5.7 illustrates this
nonuniqueness once again.
Example 16.2.11 above was based on Theorem 16.2.16. However, it is possible to
treat that example under weaker assumptions than are made in Theorem 16.2.6 as the
exercise below shows.
Exercise 16.2.13 Let n and nm be as in Example 16.2.11 and suppose that 9 e L001;n.
Further assume that / is analytic in a circle centered at 0 with radius greater than

(a) Show that for X > 0, (16.2.45) and (16.2.46) hold but with Mm given by

(b) Show that \\Kty (F m ) - Kty ( F ) | | -> 0 as m -*• oo uniformly in y on all compact
subsets of C+.
[Hint: Review the proof of Theorem 16.2.6 and then begin by showing thatfor any x e Ct,

Actually, we will see in Theorem 17.6.27 ([La18, Theorem 4.2, p. 180]) that the
stability result of Example 16.2.11 can be considerably improved, not only by merely
assuming that 9 e Loo1;n (as in Exercise 16.2.13 above), but more significantly, by also
obtaining an estimate of the type of (16.2.45) (with cm —> 0) which is valid for all X in
C~+and hence, in particular, in the diffusion as well as in the quantum-mechanical case.
(See Theorem 17.6.27 and Remark 17.6.29(a).)

We finish this section by discussing how our weak convergence result can be strength-
under U.S. or applicable copyright law.

ened in ways which are relevant to the two examples which made use of it.
In Theorem 16.2.3 and its applications, namely Examples 16.2.7 and 16.2.9, we have
used the fact that nm -± n implies that f(0,t) b(s)nm(ds) -> f(0,t) b(s)n(ds) for every
bounded continuous function b on (0, t). This implication comes immediately from the
definition of weak (or vague) convergence (see Definition 16.2.1). However, stronger
implications of this kind are known. We adapt one part of [Bil, Theorem 5.2, p. 31] to
our setting: If nm -^ n and b is a C-valued, bounded, Borel measurable function on (0, t)
with \n\(D(b)) — 0, then f(0,t) b(s)nm(ds) ->• f(0,t) b(s)n(ds), where D(b) denotes the
discontinuity set of b.
We now give a result which is more restrictive in some respects than Theorem 16.2.3
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
but DISTRITAL
which makes a less
FRANCISCO restrictive
JOSE DE CALDAS continuity assumption. The proposition to follow is
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 STABILITY IN THE MEASURES 475
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

motivated by Example 16.2.7 where the potential was time independent and the limiting
measure was Lebesgue measure.
Proposition 16.2.14 Let 0 : Rd ->• C be Borel measurable, bounded by c and contin-
uous Leb.-a.e. on Rd. Let n, nm, m — 1, 2, . . . , be in M ( 0 , t) and suppose that |n| is
absolutely continuous with respect to Lebesgue measure l on (0, t). (We write|n| « /.)
Assume that

Put

where f is an entirefunction given by (15.3.13). Let Fm be defined as in (16.2.49) except

with n replaced by nm. Then

uniformly in A on all compact subsets of C+.

Proof The only step in the proof of Theorem 16.2.3 which does not carry over immedi-
ately to the present situation is the limit taken in (16.2.6). There the definition of weak
convergence was invoked and the limit existed for every x e Ct0. We will use the weak
convergence result adapted from [Bil] and stated above and establish convergence for
m-a.e. x e Ct0. This slightly weaker form of the limit causes no problems in finishing
the proof as before.
We will show that for every y > 0 and every £ € Rd,

for m-a.e. x e Ct0.This will follow immediately from the adapted weak convergence
result if we can show that for m-a.e. x 6 Ct0, the function s >- 0 ( y - 1 / 2 x ( s ) + E) is
continuous |n|-a.e. But this last function is continuous except at values of s such that
y - 1 / 2 x ( s ) + E € D(0), where D(0) denotes the discontinuity set of 9 (a subset of
under U.S. or applicable copyright law.

Rd). But y-1/2x(s) + £ e D(9) if and only if x(s) e X 1 / 2 [ D ( 6 ) - £]. Now D(0) has
Lebesgue measure 0 by assumption, and since every translate and every scaling of a set
of Lebesgue measure 0 has the same measure, we see that the set y1/ 2 [D(0) — E] also
has Lebesgue measure 0. It follows then from Corollary 3.3.4 that for every 5 6 (0, t),
m({x e Ct0 : x(s) E y1/ 2 [D(0) - £]}) = 0. Using this and applying the Fubini theorem
to the l x m-measure of the set {(s, x) e (0, t) x Cto : x(s) E y 1 / 2 [ D ( 0 ) - E ]}, we see
that for m-a.e. x e Ct0, l([s e (0, t) : x(s) e y 1 / 2 [D(0) -£]})= 0. (Also see Corollary
12.1.5.) But |n| «l by assumption and so for m-a.e. x e Ct0, |n|({s e (0, t : x(s) €
A 1/2EBSCO
[D(0) - £]}): =eBook
Publishing 0. Hence, for(EBSCOhost)
Collection m-a.e. x -eprinted
Ct0> the
on function s K>
6/8/2017 5:47 0(y-1/2x(s)
PM via UNIVERSIDAD + £) is
continuous |n|-a.e. This
DISTRITAL FRANCISCO concludes
JOSE DE CALDAS the proof as we observed above.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 476 STABILITY RESULTS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We will not state a formal proposition related to Example 16.2.9, but we will indicate
briefly how Theorem 16.2.3 can be adjusted so that much weaker continuity assumptions
on 9 are sufficient. The key as in Proposition 16.2.14 is to find a replacement for the
limiting argument justifying (16.2.6). Recall that the limiting measure in Example 16.2.9
is the Dirac measure dT. Accordingly, for every y > 0 and £ e Rd, we need to have
the function s i-> 0(s, y - 1 / 2 x ( s ) + E) continuous at r for m-a.e. x € Ct0. A condition
which is sufficient for making this argument in a manner similar to that in the proof of
Proposition 16.2.14 is that the r-section of the discontinuity set of 0(., .) has Lebesgue
measure 0 in Rd.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

17
THE FEYNMAN-KAC FORMULA WITH A
LEBESGUE-STIELTJES MEASURE AND FEYNMAN'S
OPERATIONAL CALCULUS

17.1 Introduction
Our goal in this chapter is to describe (and partially prove) results of Lapidus in [Lai 4-18]
in which he obtained a natural extension of the Feynman-Kac formula (from Chapter 12)
to the framework of the authors" memoir [JoLa1] studied in Chapters 15 and 16; namely,
a "Feynman-Kac formula with a Lebesgue-Stieltjes measure" (FKLS, in short).
As was discussed in Section 7.6 and in much greater detail in Chapter 12, the classical
Feynman-Kac formula expresses the solution of the heat (or diffusion) equation (with
potential V) as a functional integral of

with respect to the Wiener process; here, 6(x(s)) stands for the potential 6 (= —V, in
the notation of Chapter 12) evaluated along the path x = x(s) between times 0 and t.
Following [Lal4-18], we now investigate what happens if in the Feynman-Kac
functional (17.1.1), one performs the time integration with respect to a Lebesgue-Stieltjes
measure n(ds) rather than ordinary Lebesgue measure l(ds) — ds.
More precisely, we study the time evolution of the operator u(t) associated through
path integration (followed by analytic continuation and passage to the limit) with the
exponential functional
under U.S. or applicable copyright law.

where n = u+ v is an arbitrary (C-valued and thus bounded) Borel measure on a bounded

interval containing [0, t], with continuous part u and discrete part v. (In our present con-
text, FK,n, can be called the "Feynman-Kac functional with Lebesgue-Stieltjes measure
n".) Reversing the original procedure, we look for an evolution equation (in integrated
or differential form) satisfied by the function u = u(t). [For mathematical reasons, the
integrated form of the evolution equation will be the most suitable for our purpose.
Further, the assumptions on 6 will be of the same nature as in Chapters 15 and 16.
Consequently, they will be far more restrictive than in Chapter 12; on the other hand, 6
willEBSCO
be allowed to be time-dependent and C-valued. In addition, (most of) the results of
Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
[Lal4-l 8] presented
DISTRITAL here
FRANCISCO JOSE DE will be valid both in the diffusion and the quantum-mechanical
CALDAS
AN:
case.] 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 478 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Actually, we show that u obeys a certain Volterra-Stieltjes integral equation (of

which u = u(t) is the unique bounded solution). We then deduce from this key fact a
number of related results, some of which will serve as a bridge between Chapters 15
and 16, on the one hand, and Chapter 18 on the other hand, which will take into account
more explicitly the noncommutative aspects of Feynman's operational calculus within
the context of path integration. [In Chapter 19 we will investigate in a more abstract
setting the relationships between Feynman's operational calculus and certain evolution
equations. Part of the latter work is similar in spirit but quite different in detail from that
presented in the present chapter. (In particular, compare Sections 19.5 and Section 17.3.)]
New phenomena will appear in this setting, connected in particular with the discrete part
v of n; for example, u = u(t) will usually have a (multiplicative) time-discontinuity at
every point in the support of v. [Of course, when n = u = Lebesgue measure = l and
thus v = 0, the (operator-valued) function t \-* u(t) will be (strongly) continuous and
the above integral equation yields—or rather, is "formally equivalent to" (in the sense
of [CaSt1,2] and [JoSk2,6])—the heat or the Schrodinger equation in the diffusion or
quantum-mechanical case, respectively.] We note that when specialized to the quantum-
mechanical case, this will enable us, in particular, to study and write in terms of simpler
expressions (as in formula (17.2.11) below) the [analytic (in mass) operator-valued]
Feynman integral associated with a broader class of exponential functionals than in the
standard situation where n = l.
For the simplicity of exposition, we will present these results (and their possible
physical interpretation) in two steps: First, in Sections 17.2-17.5, in the case when v is
finitely supported (say, v = Ehp=1 wPdT P , with wp e C and a < r1 < . . . < rh, < b), as
in [Lal4,15]. Then, in Section 17.6, in the general case when n = u + v is arbitrary (say,
v = E8p=1 WpdTp, with£8p=1|wp|< oo), as in [Lal6,18]. More specifically, we will
first obtain in Section 17.6 a form of the integral equation that is valid for an arbitrary
measure n, as in [La 18]. Then, as in [La 16], we will obtain (in the quantum-mechanical
case) a product integral (or "time-ordered exponential product") representation of the
solution to this evolution equation, as well as a corresponding distributional differential
equation valid for an arbitrary n (possibly with a nontrivial singular part).
We will derive (in Section 17.3) the integral equation in the prototypical case when
n = u + wdr (as in Section 15.3) and refer to [Lal5] for the general case when v is
finitely supported, studied in Section 17.2. Furthermore, we will (mostly) refer to [Lal6,
18] for the proof of the results stated in Section 17.6. We stress, however, that both when
under U.S. or applicable copyright law.

v is finitely or infinitely supported, the generalized Dyson series (GDSs) from [JoLa1]
obtained in Chapter 15 play a crucial role in deriving the results of [Lal4-l 8], especially
in the quantum-mechanical case.

Notation and hypotheses

Let a, b be real numbers such that a < b. [In the following, we adapt in the obvi-
ous manner the notation from Chapter 15 (particularly, Sections 15.2.F-15.2.I).] Let
M([a, b]) denote the space of complex Borel measures on [a, b]. [Note that in contrast
to Section 15.2.F (where a — 0), we allow the measures to have mass at the endpoints
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 6/8/2017 5:47 PM via UNIVERSIDAD
of the interval.]
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 INTRODUCTION 479
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Let 0 = 0(s, £) be a complex-valued Borel measurable function on [a, b] x Rd.

Given n in M = M([a, b]), we say that 9 belongs to Lool;n = Lool;n(a, b)) if

where |n| denotes the total variation measure of n. It follows from (17.1.3) that for n-a.e.
(almost every) sin[a, b], 0(s, •) is in L°°(Rd), and that 0(s), the multiplication operator
with the function 6(s, .), belongs to £(L 2 (R d )) and has norm ||0(s)|| = ||0(s, .)||oo.
Recall that if 0 = 0(s, £) is bounded or if 6 — d(s) is strongly continuous on [a, b],
then 0 € Lool;n. Finally, C+ and C~+ denote, respectively, the complex numbers with
positive real part and the nonzero complex numbers with nonnegative real part.
Given t e [a, b], let Ct = Ca,t = C([a, t], Rd) be the space of continuous functions
from [a, t] to Rd. Then the Wiener space Ct0 = C 0 ( [ a , t], Rd) is the set of paths x e Ct
such that x(a) = 0, equipped with Wiener measure m. (See Chapter 3.)
Fix t € (a, b). Given a function F : Ct -> C, U e L 2 (R d ), and y > 0, we consider
(as in formula (15.2.8)) the expression

In Chapter 15 (Theorem 15.4.1), we have established that the operator-valued func-

tion space integral Kty (F) exists for all y e C~+, in the sense of Definition 15.2.1, and is
"disentangled" by a time-ordered perturbation expansion (or GDS). [Recall from Defini-
tion 15.2.1 that Kty (F) e L(L 2 (R d )) is given by (17.1.4) for y > 0, defined by analytic
continuation to y e C+ and then extended by strong continuity to A e C~+. Further, when
y is purely imaginary, Kty (F) is called the analytic (in mass) operator-valued Feynman
integral of F.] These results (from [JoLa1]) are valid for a broad class of functionals,
including those we are about to introduce (in (17.1.5) and (17.1.7)).
Let n e M([a, b]) be an arbitrary complex Borel measure on [a, b] and let 9 6
£oo1;n([a, b)), as above. We consider the exponential functional
under U.S. or applicable copyright law.

associated with the "potential" 0 and the "Lebesgue-Stieltjes measure" n.

It is mathematically convenient and physically natural to work with the Hilbert adjoint
Kt/y (F)* ofKt/y(F) , rather than with Kty (F) itself (see Remark 17.3.5 below); here, I
(resp., F) denotes the complex conjugate of y (resp., F). Note that Kt/y (F)* depends on
time t both through the definition of F in (17.1.5) and that ofKt/y(F) in (17.1.4). We thus
set, for y e C~+ and t € (a, b],

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
It isUNIVERSIDAD
understood that u(a)
DISTRITAL = I,JOSE
FRANCISCO theDEidentity
CALDAS operator.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 480 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Equivalently, sinceKt/y(F)* = Kty (F) by Theorem 15.7.6 (formula (15.7.12)), we

can also define u(t) in a simpler manner by

where F denotes the "time-reversal" of F (see Definition 15.7.5); namely, for x e C',

where 9 and n denote the "time-reversal" of the "potential" 9 and the measure n, respect-
ively. Recall from (15.4.24) (with [0, t] replaced by [a, t]) that 0(s) = 0(t - s + a) for
s € [a, t] and that n is the image measure of n under the time-reversal map s i-> t — s + a
from the interval [a, t] onto itself. (We know from Remark 15.3.7 that 9 € Loo1;n implies
that 9 € Loo1;n Further, note that in (17.1.6), we have slightly modified the definition of
u(t) given in [Lal4-18], for reasons discussed in Remarks 17.3.5(b) and 15.7.7(b). How-
ever, this will change neither the proofs nor the statements of the results of [Lal4-18].)
Finally, as in Chapters 15 and 16, we will refer to the case when y = 1 and 9 = — V
(resp., y = —(' and 6 = —iV), with V e Loo1;n, as the probabilistic or diffusion (resp.,
quantum-mechanical) case.

17.2 The Feynman-Kac formula with a Lebesgue-Stieltjes measure: Finitely

supported discrete part v
Whereas the study in Chapters 15 and 16 [JoLa1] was conducted for fixed time t, we
are interested here in the behavior of u(t) as time varies. One of our main goals is to
extend the Feynman-Kac formula to the present setting, as well as to obtain results that
are valid both in the diffusion and quantum-mechanical cases.
Let n = u + v be the unique decomposition of n into its continuous part u and its
discrete part v. This decomposition arises naturally in the study of the generalized Dyson
series (Chapter 15). We assume in the present section and in Sections 17.3-17.5 that v
is finitely supported; specifically, we suppose that
under U.S. or applicable copyright law.

Integral equation (integrated form of the evolution equation)

We first state our central result in this case ([Lal5, Theorem 2.1, p. 98]).
[In the following, unless otherwise specified, we always assume that y e C~+ is arbitrary.
Hence, for example, the integral equation (17.2.2) and the differential equation (17.2.3)
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
below hold for
UNIVERSIDAD all y eFRANCISCO
DISTRITAL C~+.] JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 MEASURE WITH FINITELY SUPPORTED DISCRETE PART v 481
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 17.2.1 For each fixed p = 0, . . . , h, the operator-valued function u satisfies

the Volterra-Stieltjes integral equation

for all t € (Tp,T p+1 ].

The integral in (17.2.2) converges in the strong operator topology (or is a strong
Bochner integral ).
Moreover, there is a unique bounded (strongly measurable) function u : [a, b] —>
£(L2(Rd)) which is a solution of the integral equation (17.2,2) on each interval
(rp, Tp+1], with p = 0, . . . , h.
In Section 17.3, we will establish this integral equation in a simple but prototypical
case (77 = u + wdr) and refer to [Lal5, §4] for its proof in the general case stated above.
[The last statement of Theorem 17.2.1 (concerning the uniqueness of the solution) will
be proved in Section 17.5.]
Differential equation (differential form of the evolution equation)
The next two statements follow easily from Theorem 17.2.1. However, the formulation
of Theorem 17.2.2 and Corollary 17.2.3 will perhaps seem more familiar to the reader
than that of Theorem 17.2.1.
Theorem 17.2.2 Let p e {0, . . . , h}. Then for Leb.-a.e. t E (TP, T P +1),

wheredu/ds(t) denotes the Radon-Nikodym derivative of u with respect to Lebesgue meas-

ure l. (That is, for each U in D(H0), the domain of HO, u(t)U is differentiable for Leb-a.e.
t and the differential equation (17.2.3) holds when applied to U.)
Consequently, in the diffusion or quantum-mechanical case, respectively, u satisfies
the heat or Schrodinger equation with potential (du/ds)(t) V in each open interval
(TP, Tp+1), p = 0, . . . , h, as is stated in the following corollary.
Corollary 17.2.3 In the probabilistic case (i.e. y = 1 and 6 — —V with V e Loo1;n)>
under U.S. or applicable copyright law.

(17.2.3) becomes the heat equation

with potential V\ := ( d u / d s ) ( t ) V .
On the other hand, in the quantum-mechanical case (i.e. A = — i and & = —iV with
V e Loo1;n), (17.2.3) yields the Schrodinger equation

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
with potentialDISTRITAL
UNIVERSIDAD V\ defined as above;
FRANCISCO note that V1 e
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 482 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

If, in addition, /z = / and 9 = 0(s) is strongly continuous (for instance, if V is time

independent and bounded), then V[ = V and the differential equation(17. 2.4) or (17. 2.5)
holds everywhere (not just l-almost everywhere) in each open i n t e r v a l ( r p , r p interval.

Warning: The logically correct form of the differential equation is of a distribu-

tional nature; for simplicity, however, we postpone to Section 17.6 (specifically, to The-
orem 17.6.15) a precise discussion of this point (in the quantum-mechanical case) and
refer to [Lal6, §2.B] for the details. (See also Remark 17.2.8(d) below.) For this reason,
the integrated form of the evolution equation (Theorem 17.2.1)— although possibly less
intuitive to the reader— is to be preferred in the mathematical exposition of the theory.

Discontinuities (in time) of the solution

We next see that the solution of the integral equation (17.2.2) (or of the differential
equation (17.2.3)) will usually have a discontinuity at a point rp of the support of v.
(This will be further discussed in Section 17.4.)
Theorem 17.2.4 The function u is strongly left continuous in (a, b]. Moreover, for p =
0, . . . ,h, u is strongly continuous in (rp, t p +1] and

where U(TP+) denotes the strong limit of u(t) as t —> rp, t > TP.
Remark 17.2.5 In (17.2.2), (17.2.6), and thererafter, we set a)0 = Qifr0:=a< r\
and (DO = a>\ if TO = T\.
It is worth while singling out the special case where r] is a continuous measure (i.e.

Corollary 17.2.6 (Purely continuous measure: rj = /j.)

(i) Assume that r] = /u, is a continuous measure. Then, for all t 6 [a, b],

Moreover, u is strongly continuous in all of [a, b] and satisfies the differential equation
under U.S. or applicable copyright law.

(17.2.3) l-almost everywhere in (a, b).

(ii) Suppose further that r/ — (j. = I (so that d/j,/ds)(t) = 1). Then, in the prob-
abilistic case, u satisfies the heat equation

while in the quantum-mechanical case, u obeys the Schrodinger equation

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 MEASURE WITH FINITELY SUPPORTED DISCRETE PART v 483
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

When 77 = [i — I as in part (ii) of Corollary 17.2.6—or, more generally, as com-

mented upon in Section 14.5, in the special case of part (i) when r/ = /x is absolutely
continuous with respect to Lebesgue measure /—the integral equation (17.2.7) follows
from that obtained in the paper by Johnson and Skoug [JoSk6] (the p — p1 = 2 case).
(Also see the earlier related work by Cameron and Storvick, beginning with [CaStl], as
well as the relevant comments in Section 14.5 above.)
Equation (17.2.8) yields the classical Feynman-Kac formula (Chapter 12, especially
Theorem 12.1.1 and Corollary 12.1.2), which expresses the solution of the diffusion
equation in terms of a Wiener integral. (Note that in Chapter 12, the Feynman-Kac
formula was stated for a time independent potential V, but under much more general
assumptions on V.) On the other hand, equation (17.2.9) leads to one interpretation of the
Feynman path integral (see Chapter 15, especially Definition 15.2.1 and the comments
following it). Strictly speaking, as was previously stressed, it would be more appropriate
in the present context to refer to the probabilistic (resp., quantum-mechanical) case, of
the evolution equation in its integrated form (17.2.7) rather than in its differential form
(17.2.8) (resp., (17.2.9)).
In view of (17.2.2)-(17.2.9), it is thus justified to refer to the above results from
[Lal5]—and, strictly speaking, to the probabilistic case of Theorem 17.2.1—as a
"Feynman-Kac formula with a Lebesgue-Stieltjes measure" (in short, FKLS). We
emphasize that the procedure used in [Lal5] (or [Lal4]) is the reverse of the usual
one, since, starting with a Wiener functional [given here by formula (17.1.7)], we find
the corresponding differential or integral equation.
We note that the approach to evolution equations (via Feynman's operational calculus
[Fey8]) which we will use in Chapter 19 (based on [dFJoLal,2]) is in the same spirit as
in the present chapter. One begins in Chapter 19 with an exponential function of sums of
noncommuting operators. After "disentangling", one then obtains an expression which
can be shown (see Section 19.5) to satisfy an evolution equation.

Propagator and explicit solution

The fact—stated at the end of Theorem 17.2.1 and established in Section 17.5—that
the integral equation (17.2.2), and a fortiori (17.2.7), has a unique bounded solution,
enables us to introduce the following operator. Given a < t\ < ?2 £ b, let P f a , t\)
be the propagator for the integral equation (17.2.7); that is, P(t2,t1) is the operator in
under U.S. or applicable copyright law.

£(L 2 (R d )) that maps the solution u(t\) at time t\ of (17.2.7) onto the solution u(t2) at
time ti. Clearly, for a < t\ < 12 < (3 < b,

Note that, by definition, u(t) = P(t, a) is the solution at time t of (17.2.7) when rj = /j.;
further, we stress that by definition of (17.2.7), P(ti, t\) depends on IJL but not on v.
We are now able to describe the combined effects of the continuous part n, and the
discrete part v of n. (See [Lal5, Theorem 2.4, p. 99].)
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 484 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 17.2.7 Let rj be as above. Then

Remark 17.2.8 (a) Equations (17.2.2), (17.2.6) and (17.2.11) suggest that the Dirac
mass opSrp in the discrete part of n corresponds to a "shock" or "instantaneous inter-
action" occurring at time rp. We will elaborate this interpretation in Section 17.5.
For now, we simply mention that it is physically reasonable to assume that v is finitely
supported, since only a finite number of events (for instance, shocks or scatterings) should
occur within the time interval [a, b]. Moreover, from a mathematical point of view, we
shall find it useful in Section 17.6 to approximate —as was done in [La 18] —an arbitrary
measure r\ by measures with finitely supported discrete part. (See Theorem 17.6.27
below.)
(b) The adjective "strong" in the statements of Theorems 17.2.1 and 17.2.4 means
that the integrability, or limit, holds when the corresponding expression is applied to an
arbitrary vector i/r e L2(Rd).
(c) When u — l, (17.2.7) is nothing but the standard Duhamel integral equation
(see, for example, [JoSk6, §57 in conjunction with (Lal2, pp. 58-60] or [La9]). If, in
addition, the potential is time independent, then

in the diffusion or quantum-mechanical case, respectively. Note that since V must then
be (essentially) bounded, the Hamiltonian HQ + V has the same domain as HQ; namely,

(d) We stress that the measure /u, need not be absolutely continuous. It could be, for
instance, the singular Lebesgue-Stieltjes measure associated with the Cantor function,
as was considered in Example 15.5.3; we will return to this situation in Section 17.4. As
we will see, the integral equation (17.2.2) is then more informative than the differential
equation (17.2.3) [since dfi/dl = 0 l-a.e.]. Alternatively, one can replace (17.2.3) by
under U.S. or applicable copyright law.

a distributional differential equation, established in [La16, Theorem 2.5, p. 286] and

discussed in Section 17.6 below. (See Theorem 17.6.15.)
We conclude by considering the special case when r\ is discrete (i.e. n = 0).
Corollary 17.2.9 (Purely discrete measure: n = v) (i) Assume that r\ = v is a discrete
measure given by (17.2.1). Then, for p € {0, . . . , h] and t € (TP, Tp+i],

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 MEASURE WITH FINITELY SUPPORTED DISCRETE PART v 485
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(ii) Suppose in addition that a = 0, b = t, rp = p(t/h), and u>p = TP — TP+\ =

t/h, for p = 1,..., h; so that r\ = (t/h) Yjp=i &p(t/h)- Then, for time independent
potential, we have in the probabilistic case (i.e. A. = 1 and 9 = — V)

and in the quantum-mechanical case (i.e. X = —i and 6 = —i V)

We recognize in ( 1 7.2. 14) or ( 1 7.2. 1 5) the familiar hth Trotter product for semigroups
or unitary groups, respectively. The link with the Trotter product formula (Section 11.1)
is discussed in Example 15.5.5 and especially in Example 16.2.7 above. More gener-
ally, the connections with product (or time-ordered) integrals (pointed out in [La 14])
are developed in [Lal6]. We will briefly describe them in Section 17.6. See especially
Theorems 17.6.10, 17.6.12 and 17.6.20, which deal with the quantum-mechanical case
and refer to the counterpart of the operator u (t) in the so-called Dirac (or interaction)
representation.
Remark 17.2.10 Note that in (17.2.13)— and in (17.2.11)— the expression for u(t) =
K ( ( F ) appears in the "natural physical ordering" (as discussed in Remarks 15.3.7 and
15.4.4, as well as in Section 15.7). Of course, (17. 2.1 3) follows from (17. 2.11) [by letting
H = 0 and hence P(ti, t\) = e -('2-n)flbWjf] while (17,2.14) and (17.2.15) follow from
(17.2.13) and Theorem 15.7.6 [or more simply, by replacing 9 and r\ = u by their
"time-reversals" 6 and rj, as defined by (15.3.23)].
Between the two extreme cases of a purely continuous and a purely discrete measure
treated in Corollaries 17.2.6 and 17.2.9 lies a whole range of possibilities described by
Theorems 17.2.1, 17.2.2, and 17.2.7.
Exercise 17.2.11 Rewrite the (classical) Feynman-Kac formula obtained in Chapter 12
(Theorem 12.1.1 and Corollary 12.1.2) in the form of an integral equation (rather than
of a differential equation, as in (12.1.5)).
[Hint: Your answer should be analogous to (17.2.7). Recall that in Chapter 12, Y\ =
//, = /, A. = 1, and 0 = — V, with V time independent and satisfying the hypotheses of
Theorem 12.1.1.]
under U.S. or applicable copyright law.

We now outline the contents of the remainder of this chapter. We begin in Section 17.3
by deriving the integral equation (Theorem 17.2.1) in the relatively simple case when
v = a>8r. We then sketch (in Problem 17.3.6) the proof of Theorem 17.2.1 for an arbi-
trary finitely supported v. [This general proof of Theorem 17.2.1 (detailed in [Lal5, §4])
is interesting in its own right because of its rich combinatorial structure and because it
shows that the generalized Dyson series introduced in Chapter 15 [JoLal] can be very
useful computational tools. We have chosen, however, to omit it here.] In Section 17.4, we
analyze the discontinuities due to the discrete part v (Theorem 17.2.4) and explain heuris-
tically
EBSCOhow they are
Publishing connected
: eBook Academic with a change
Collection of initial
(EBSCOhost) condition
- printed in the
on 6/8/2017 differential
5:49 PM via equa-
tion.UNIVERSIDAD
In Section 17.5, FRANCISCO
DISTRITAL we establish
JOSE DEthe "explicit" expression obtained in Theorem 17.2.7.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 486 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We also provide physical interpretations of our results in both the quantum-mechanical

and probabilistic cases as well as suggest further connections between Chapters 15-16
[JoLal], the present work [Lal5], and Feynman's operational calculus.
Finally, in Section 17.6, we consider the general case of the FKLS (derived in [Lal8]
and [La 16]) when r\ is an arbitrary Borel measure [and hence u may have countably
many (possibly unordered) points in its support]. We state, in particular, an integral
equation [Lal8], a corresponding distributional differential equation [Lal6], as well as
(in the quantum-mechanical case) a product integral representation [La 16] of the solution
u(t) = K[(F) (and of the associated "scattering operator"). We will refer the reader to
[La 18, La1 6] for the detailed proofs of these results. The basic idea of the proof of the
integral equation, however, consists in applying the FKLS when v is finitely supported
(Theorem 17.2.1), rewriting it in a way that does not depend on the ordering of the times
TP, and then passing to the limit in a suitable way; a sketch of this proof will be provided
at the end of this chapter.

17.3 Derivation of the integral equation in a simple case (n = n + w5 r )

We give a detailed proof of Theorem 17.2.1 (apart from the uniqueness statement which
will be established in Section 17.5) in the prototypical case n = u + co8T, as in [Lal5,
§3]. Hence we assume that

where u is a continuous measure, w E C, and r e (a, b].

At the end of this section, we shall then briefly explain how to deal with the additional
combinatorial complications encountered in the case when the discrete part of n is finitely
supported (see Problem 17.3.6). (The complete proof in that situation can be found in
[Lal5, §4].)
We stress that whether in the case treated here or in the general case when v is finitely
supported, the results from [JoLal] on generalized Dyson series (GDSs) obtained in
Chapter 15 are crucial to the proof of the integral equation (see Theorem 17.3.1 below
and part (a) of Problem 17.3.6).
Our starting point is the following result, which in view of (17.1.6), (17.1.7) and
Theorem 15.7.6, is a consequence of Theorem 15.3.1 as well as Corollaries 15.3.4 and
15.3.6. We purposely use more precise notation than in Chapter 15. [Observe that each
under U.S. or applicable copyright law.

CDS below is written according to the "natural physical ordering", as will be further
explained in Remark 17.3.5 below. This is precisely where we use the fact that u(t) =
K[(F) (rather than K [ ( F ) ) , where F denotes the time-reversal of F. See Remarks 15.3.7
and 15.4.4, as well as Theorem 15.7.6.]
Theorem 17.3.1 For all A e C+, u(t) = K[(F) is disentangled (time-ordered) by the
following perturbation expansion or CDS:
For t e (T, b],

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DERIVATION OF THE INTEGRAL EQUATION WHEN n = fjt + toS, 487
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where for 0 < j < k < n,

Here,

and for (si,...,Sk) € &k;j(t),

On the other hand, for t e [a, T],

where

Here,

and for(si,...,sn) € A n (f),

under U.S. or applicable copyright law.

In (17.3.2) or (17.3.6), the series is summable in £(L 2 (R d )) and converges in

operator norm, uniformly for t in (T, b] or (a, T], respectively; further, the integrals in
(17.3.3) or (17.3.7) are strong Bochner integrals.
Moreover, for all t € [a, b], we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 488 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 17.3.2 (a) For j = k, Cn,k-j in (17.3.5) begins with e-(t-T)(Ho^[e(T)]n-k,

whereas for j = 0 it ends with [e(T)]n-ke~(r~a)(Ho/^; in fact, in (17.3.4), we have
by convention a < s\ < •• • < Sk < T or T < s\ < • • • < SK < t for j = k or
0, respectively. Further, in (17.3.3), the notation /u,®* stands for the product measure
of k copies of n, as in Chapters 15 and 16; of course, the integral in (17.3.3) could
equivalently be written in iterated form. (See Remark 15.2.10.)
(b) Assume that r\ = ^ = I. Then, in the quantum-mechanical case, (17.3.6) yields
the classical Dyson series (see formula (15.3.22)). Further, if 9 is time independent, then
K [ ( F ) = K[(F)[=KL(F)*, by (15.7.12)]. (Actually, u(t) = <r'(//(>+v> ore-it(Hf>+V)
in the diffusion or the quantum-mechanical case, respectively.) This simple (but possibly
confusing) fact can also be deduced directly from the Dyson series by observing that
the "time-reversal" map s i-> t — s + a from (a, t) to itself leaves Lebesgue measure /
invariant. (Consider Example 15.3.8 with v = l for a special case.)
(c) For a physical interpretation of the above GDS, we refer to the comments following
formula (15.3.22). For instance, in the quantum-mechanical case (A. = —i and 0 =
—iV), (17.3.9) corresponds to a particle propagating freely between 0 ands\, interacting
with the potential V at time s\, propagating freely between s\ and si, and so on up to
time t. In (17.3.7) and (17.3.6), one then sums over all "past histories" of the quantum-
mechanical particle. The powers of 6(r) appearing in (17.3.5) are more difficult to
interpret. (See, however, Example 16.2.9, especially Remark 16.2.10.) Nevertheless, the
results of [Lal 5] discussed in the present chapter show that in the case of an exponential
functional, they globally correspond to an "instantaneous interaction " of the particle
with the potential V, occurring at time T. (See the discussion in Section 17.5 below.)
Generalized Feynman diagrams associated with the series (17.3.6) or (17.3.2) were
provided in Figures 15.6.1 or 15.6.2-15.6.4, respectively. The reader may find it helpful
to draw such diagrams while following the proof of Theorem 17.2.1.
We note that it suffices to prove Theorem 17.2.1 for t e (T, b], since the case when
t e [a, T] can then be deduced by letting at = 0. In the present situation, we thus have
to show that for t 6 (T, b],

(As before, unless otherwise specified, we assume throughout that A. 6 C+ is fixed, but
under U.S. or applicable copyright law.

arbitrary.)
In the course of the proof of (17.3.11), we shall use the following key result ([Lal5,
Proposition 3.1, p. 103]):

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DERIVATION OF THE INTEGRAL EQUATION WHEN r, = /x + euSr 489
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Clearly,

Now, for sk+i 6 (r, t) and (s\,..., sk) e A* ; ;(s k +i), we have

In fact, by (17.3.5) with t replaced by SK+I,

Hence, if we observe that Q<j<k+l<n + l and

equations (17.3.16) and (17.3.5) yield successively

This proves (17.3.15). (See (17.3.19) below.)

Next, for Sk+i € (T, t), we have
under U.S. or applicable copyright law.

In the first equality of (17.3.18), we have used (17.3.3) with f replaced by sjt+i. In
the EBSCO
second one, we
Publishing have
: eBook takenCollection
Academic the given bounded
(EBSCOhost) linearonoperator
- printed inside
6/8/2017 5:49 the integral,
PM via
according to aDISTRITAL
UNIVERSIDAD specialFRANCISCO
case ofJOSE
[HilPh, Theorem 3.7.12, p. 83].
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 490 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We note that, by (17.3.4) and Remark 17.3.2(a),

for 0 < j < k. (Observe that j ^ k + 1 implies that sk+\ > r if (s\,..., sk, s^+i) e
Aft+i ; y(0-) Consequently,

by (17.3.3). In view of (17.3.13), this yields (17.3.12), as desired. (For notational sim-
plicity, we write jU®*+1 rather than jtt®(*+1) for the product of k + 1 copies of the
measure ^.)
We now explain the main steps leading to (17.3.20). In the first two equalities of
(17.3.20), we have used (17.3.14) and (17.3.18), respectively. In the third one, we have
combined Fubini's theorem and (17.3.19). Note that the Fubini theorem for (strong)
Bochner integrals [HilPh, Theorem 3.7.13, p. 84] applies here; in fact, for V € L2(R.d),

Further, by (Lemma 15.2.7 and) the "simplex trick" for continuous measures discussed
shortly after formula (15.3.8), we have (with A^+j (t) defined as in (17.3.8) above),
under U.S. or applicable copyright law.

since 6 ePublishing
EBSCO Lcci;?; implies that 6 Collection
: eBook Academic e £001-,;*(EBSCOhost)
(see equation (15.2.6)).
- printed on 6/8/2017 5:49 PM via
This completes
UNIVERSIDAD theFRANCISCO
DISTRITAL proof ofJOSEProposition
DE CALDAS 17.3.3.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DERIVATION OF THE INTEGRAL EQUATION WHEN r\ = fj. + (o8T 491
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We can now give the proof of Theorem 17.2.1 in the case when rj = & + co8T.
(We focus here on the derivation of the integral equation; the fact—stated at the end
of Theorem 17.2.1—that this integral equation has a unique bounded solution will be
established towards the beginning of Section 17.5.)
Proof of Theorem 17.2.1 For t e (T, b], set

As was noted previously, it suffices to show that (17.3.11) holds for t e (T, b], or
equivalently,

We first see that

where wn,k\j is defined as in (17.3.13). Indeed, by (17.3.2),

in view of (17.3.13), this yields (17.3.24). The interchange of the order of integral and
sum in the last equality of (17.3.25) will be justified at the end of the proof.
By Proposition 17.3.3,

Now, by making the change of indices m = n+l and A" = fc + 1 and noting once again
under U.S. or applicable copyright law.

that m — K = n — k, we obtain

Observe that we may as well assume that the indices m and K start at 0 instead of 1 in
(17.3.27). Hence

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 492 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

It follows from (17.3.2) and (17.3.28) that

Further, we claim that for t e (T, b],

with uk given as in (17.3.7).

In fact, by (17.3.4) and (17.3.8),

for all t e (T, b], and by (17.3.5), (17.3.9) and Remark 17.3.2(a), we have for
(si,...,sk) e Ajt ; jfc(r),

Thus by (17.3.3), (17.3.31) and (17.3.7),

This proves (17.3.30).

In view of (17.3.29) and (17.3.30),
under U.S. or applicable copyright law.

where
EBSCOwe have used
Publishing (17.3.6)
: eBook Academicin the last(EBSCOhost)
Collection equality; -moreover, in the 5:49
printed on 6/8/2017 second equality of
PM via
(17.3.32), weDISTRITAL
UNIVERSIDAD have recognized theDECauchy
FRANCISCO JOSE CALDAS product of two absolutely summable series
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DERIVATION OF THE INTEGRAL EQUATION WHEN rj = n + a>Sr 493
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

in £(L2(Rd)). Since (17.3.32) yields (17.3.23), we obtain (the integral equation stated
in) Theorem 17.2.1 in the present case where rj = n, + a>ST.
We conclude this proof by justifying the interchange of sum and integral in (17.3.25).
Indeed, for 5 e (r, f),

since A^ = IJ 7=0 &k;j UP to a set of I Ml®* -measure zero (see equation (15.3.7)),

Hence, for ^ € L2(Rd),

and so
under U.S. or applicable copyright law.

In the last equality of (17.3.33), we have used the fact that (see equation (15.2.6))

We may therefore apply Fubini's theorem in (17.3.25).

FromPublishing
EBSCO a strict :logical point of
eBook Academic view, the
Collection following
(EBSCOhost) comment
- printed should
on 6/8/2017 5:49have
PM viabeen made
at the beginning
UNIVERSIDAD of this
DISTRITAL section.
FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 494 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Remark 17.3.4 We justify the existence of the strong Bochner integral in (17.3.11) or
(17.3.12) by appealing to Bochner's integrability criterion [HilPh, Theorem 3.7.4, p. 80],
The (strong) Bochner measurability of the integrand in (17.3.11) or (17.3.12) follows
from a simple measurability lemma for Bochner integrals depending on a parameter,
analogous to Lemma 15.2.9. (It can also be obtained by applying the recent abstract
measurability results in [Jo8, BadJoY] which will be briefly discussed at the beginning of
Section 19.4 or else by showing directly, as in [Lal4, §2.2, pp. 11-12], that u or un k-j is
continuous on (a, b], except at T, where it has one-sided limits.) Moreover, let us denote
by Aa(s) the integrand in (17.3.11) or (17.3.12), with a = 1 or 2, respectively. Then,
fora given iff & L2(Wi), the integrability of\\J\fa(s)i/\\ with respect to \/j.\(ds) follows
from the norm estimate (17.3.10) or (17.3.21), respectively.
We next explain why it is advantageous in this context to work with K[(F) rather
than with K[(F) itself.

Remark 17.3.5 (a) The GDS for K[(F) and K[(F) differ only by the time-ordering of
each summand. (Compare equations (17.3.2) and (17.3.3) with equations (15.3.2) and
(15.3.15).) More precisely, according to Theorem 15.7.6, each summand in the series for
K[(F) (resp., K[(F)) is written according to increasing (resp., decreasing) time indices
as you read from right to left on the page. (Compare equation (17.3.5) with equation
(15.3.4).) (This distinction is blurred in the classical case where rj = /u, = I because of the
invariance of Lebesgue measure under time-inversion; see Remark 17.3.2(b).) Clearly,
the ordering according to increasing time indices in the series for K^(F) corresponds
to the natural physical interpretation (see Remarks 15.3.7 and 15.4.4). Further, it is
in agreement with Feynman 's time-ordering convention for noncommuting operators,
according to which the operator with the smallest time index acts first (see formula
(14.2.1)).
As we know from Section 15.7, this comment applies to all the generalized Dyson
series (and associated Feynman diagrams) treated in Chapters 15 and 16, and in par-
ticular to that considered in Problem 17.3.6(a) below. Moreover, in the present case of
an exponential functional, the use of F, the time-reversal of F, also enables us to obtain
for u(t) = K[(F) a differential or integral equation in a convenient form.
(b) In [Lai4-18], the second author had erroneously defined u(t) by K'^(F)', the
(Banach) adjoint of K((F), rather than by K[(F) [ = KL(F)*, the (Hilbert) adjoint of
under U.S. or applicable copyright law.

Ki-(F), by(15.7.12)] as in (17.1.6). (This is correct in case the parameterX, the potential
6 and the measure r\ are all real-valued, but not in general; see formula (17.1.6a) and
Remark 15.7.7(b).) The reason for introducing the adjoint was to handle the situation
associated with the "time-reversal" and hence to obtain the GDS in the "natural physical
ordering ", as explained in (a) above. [We have already commented in Remark 15.7.7(b)
on this problem which led us to introduce (in Definition 15.7.5) the "time-reversal" map
F M» F in order to formalize the notion of "time-reversal" for the GDS disentangling
K[(F).] However, since u(t) was then precisely given by the physically natural GDS
[for example, in the setting of the present section (where r) = ^ + a)Sr), by the GDS in
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
Theorem 17.3.1], neither the statement of the results in [Lal4~18] nor their proof need
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DERIVATION OF THE INTEGRAL EQUATION WHEN r\ = n + a>8r 495
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

to be changed in any way [once, of course, the original definition of u(t) is modified
according to (17.1.6)].
Sketch of the proof when v is finitely supported
We close this section by briefly commenting on the proof of the integral equation (17.2.2)
from Theorem 17.2.1 in the case when the discrete part v is finitely supported. Because
of the greater combinatorial complexity of the GDS for u (t ) , the proof of Theorem 17.2.1
is technically much more involved than in the special case studied above; the previous
proof should serve as a helpful guide, however.
The following problem — which should take longer to solve than most exercises in
this book —provides an outline of the proof of Theorem 17.2.1 for finitely supported
v. [The interested reader can find a complete proof of the integral equation in [Lal5,
§4]; more precisely, the proof of part (b) (resp., (c)) below is given in [Lai 5, Proof of
Proposition 4.1, pp. 113-114] (resp., [Lal5, Proof of Theorem 2.1, pp. 114-118]).]
Problem 17.3.6 Let n = n + v, with discrete part v given by (17.2.1).
(a) Deduce from Corollary 15.4.3 and Example 15.5.1 (as well as Theorem 15.7.6)
that for all A. € C+ and all t e (r/,, b], the operator u(t) = K[(F) is disentangled by
the following GDS:

where for nonnegative integers qo, q\,..., qh and j\,..., jh+i such that q^-i Vqh =
n and j\ + . . . + jn + jh+i = qo,

(17.3.35)

Here, &q0;j\,...,jt,+\ — Aqo'Ji.—Jh+i^ is given by (15.5.6) (with 0 replaced by a),

while for (s\ Sq0) e A^,,...,^^), we have £n,go;j\,...,JH+\ = £n,90;h JH+\
under U.S. or applicable copyright law.

(s\,..., sqo; t) := Lh ... L\L$, with Lr defined by (15.5.7) for r =0,... ,h.
Further, show that the series (17.3.34) is summable in £(L 2 (R d )) and converges
in operator norm, uniformly in t e (r/,, b), while the integral in (17.3.35) is a strong
Bochner integral. Moreover, prove that estimate (17.3.10) holds for all t € [a, b].
(b) For t e (T/J, b] and nonnegative integers qo, q\,..., qh and j\,..., jh+\ such
that qo H \-qh = n and j\ H h jh + jh+i — qo, set

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 496 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Then establish the exact counterpart of Proposition 17.3.3; namely, show that for all
t 6 (r/,, b], we have

(17.3.37)

[In order to understand (17.3.37), it is useful to note that j\ + - • • + jh+ Jh+\ = <1Q and
qo-\-----\-qh = n implies that ji-\-----K/ft+C/A+l + l) = qoand(q0+l)-\ ----- \-qh - n+l.
Further, if we were to indicate explicitly the dependence on q\, . . . , qh in the left-hand
side of (17.3.37), then the same h-tuple (q\, ... , qh) would appear on both sides of
(17.3.37).]
(c) Use (b) and (a) to show that for p := h, the integral equation (17.2.2) holds for
allt € (rh,b].
[Warning: The proof of part (c) is combinatorially more involved and requires several
additional steps than its counterpart in the simpler case when r\ = IL + u>Sr.]
(d) Conclude from (c) that for every p e {0, . . . . A}, the integral equation (17.2.2)
holds for all t e (TP, rp+i], as desired.
[Hint: If I < p < h and t € (TP, Tp+i], simply replace the discrete part of r\ by

We close this section with a very simple exercise that may, however, be helpful to
some readers.
Exercise 17.3.7 (a) Write the integral equation (17.2.2) in a more concrete form (valid
for any X € C+ and when applied to an arbitrary \jf e L2(R.d) ), not involving explicitly
the operator notation. Specialize your result to the diffusion case and then to the quantum-
mechanical case.
[Hint: Use formula (15.2.2) or, more precisely, use the integral expressions for the free
semigroup (A e C+) given in Theorems 10.2.5-10.2.7.]
(b) Looking ahead to Section 17.6, answer the same question as in part (a) for the
more general integral equation (17.6.2), valid for an arbitrary Borel measure r].
[Advice: Use in addition formula (17.6.5').]

17.4 Discontinuities of the solution to the evolution equation

We establish here Theorem 17.2.4 and discuss, in particular, its relationships with the
associated evolution equation. Unless otherwise specified, we assume throughout this
under U.S. or applicable copyright law.

Section and the next that rj = ^ + v has finitely supported discrete part v given by
(17.2.1), just as in Section 17.2.
The time discontinuities
We first show that the discrete part v gives rise to discontinuities in the solution u = u(t)
of the evolution equation (17.2.2).
Proof of Theorem 17.2.4. Fix X € C+ and p with 0 < p < h. By (17.2.2), we have
for all t e (TP,-CP+\\,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 DISCONTINUITIES OF THE SOLUTION TO THE EVOLUTION EQUATION 497
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where we have set

For i < t, we also set

so that by (17.4.2),

Note that because ^ is a continuous measure, we may integrate over (TP, t) instead of
[TP, t). Further, since g-O-^XtfoA) js strongly continuous in t, so is f ( s ; t). Clearly, by
(17.3.10),

Since the right-hand side of (17.4.3) is |?j|-integrable and a fortiori |/i|-integrable, the
Lebesgue dominated convergence theorem for Bochner integrals [HilPh, Theorem 3.7.9,
p. 83] easily yields that w is strongly continuous on (rp, TP+\]. By (17.4.1), it follows
that u is strongly continuous on (TP, rp+1 ] and, since W(TP+) — w(rp) = 0,that u(r p +)
exists and is given by

This establishes Theorem 17.2.4.

Differential equation and change of initial condition
It is easy to deduce (at least formally) the differential equation (17.2.3) in Theorem 17.2.2
from the integral equation (17.2.2) in Theorem 17.2.1. We now briefly explain how, from
the point of view of the differential equation, the discontinuities of u = u(t) established
above can be viewed as a change of initial condition occurring at every point TP in the
support of v. Our discussion will be more informal than in the rest of this section.
First we observe that Theorems 17.2.2 and 17.2.4 can be restated as follows. Fix
under U.S. or applicable copyright law.

p e {0,..., h} and \jf e D(HQ); then for /-a.e. t e (TP, rp+\), the function u(t)\fr is
differentiable in the norm topology of L 2 (JR rf ) and satisfies the differential equation

with initial condition

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
(Note that (17.4.5b)
UNIVERSIDAD DISTRITALholds forJOSE
FRANCISCO all w
DE e L 2 (R d ).)
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 498 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Actually, it is of interest to interpret this change of initial condition from ^.probabilis-

tic point of view. Consider the Wiener integral (17.1.4) that defines K[(F)\{s for A. > 0
and iff 6 L2(Rd); here, the functional F is given by (17.1.5). For notational simplicity,
we let | = 0 and A, = 1 in (17.1.4); further, we use F rather than its time-reversal F
given by (17.1.7). For t e (TP, TP+I] and x e C', we write

As t —»• TP+ and under appropriate assumptions, the expression between brackets
tends to

Consequently, as t -*• TP+, F(x)$(x(t)) converges to

where we have set

We thus see that at time rp, the "initial condition" 1/r is replaced by the new initial
condition <j) given by (17.4.6).
We conclude that under suitable hypotheses and for A, > 0, the analogue of (17.4.4)
for K[(F) (or rather, for u(t) = K^(F)) can be obtained in this manner. However,
technical difficulties—only partly overcome by use of Vitali's theorem for the conver-
gence of a sequence of analytic functions (Theorem 11.7.1 and Remark 11.7.2(c))—arise
when we attempt to establish this result for A. purely imaginary. (This is analogous to the
difficulties described at the end of Appendix 11.7.) Recall that the Wiener integral rep-
resentation of K'^(F) given in (17.1.4) no longer holds for nonreal A. (See, for example,
under U.S. or applicable copyright law.

the discussion preceding Remark 15.2.6.)

In contrast, the following exercise illustrates the fact that our perturbation expansions
(or GDSs) in Chapters 15 and 16 allow us to work directly for any A. in C+ and, in
particular, to obtain results valid both in the probabilistic and quantum-mechanical cases.
(Of course, we already know that Theorem 17.2.4 is valid for all A e C^ since we have
deduced it from Theorem 17.2.1 which was proved by means of these same GDSs.)
Exercise 17.4.1 Fix an arbitrary A. in C+. Assume for simplicity that r/ = /u, + a>ST, as
in Section 17.3. Then establish directly—by using the GDSs in (17.3.2) and (17.3.6)—
Theorem 17.2.4, :and
EBSCO Publishing inAcademic
eBook particular, the (EBSCOhost)
Collection time discontinuities expressed
- printed on 6/8/2017 5:49 by (17.2.6) (or
PM via
(17.4.4)).
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 EXPLICIT SOLUTION AND PHYSICAL INTERPRETATIONS 499
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We now comment on the relationships between Theorem 17.2.4 and the various
forms of the evolution equation. As we know, the integral equation (17.2.2) obtained
in Theorem 17.2.1 provides the mathematically correct form of the evolution equation.
For instance, u, could be a singular measure, in which case (dp,/ds)(t) = 0 for /-a.e.
t e [a, b]', then, clearly, a great deal of information is lost (in (17.2.3)) and hence (17.2.3)
combined with (17.2.6) can not imply (17.2.2). (Of course, this is so even if v = 0.)
In order to take into account, in particular, both the singular part of /j, and the time
discontinuities caused by v, we would have to replace equations (17.2.3) and (17.2.6)
by a single distributional differential equation on [ a , b ] .
We refrain from doing this in a rigorous manner at this stage and instead refer to
[Lal6, Theorem 2.5, p. 286] which will be stated and briefly discussed in Section 17.6
(see Theorem 17.6.15). However, we mention that we can approximate v by absolutely
continuous measures /% with density pm := ]S£/>=i u>pPm;p, where (Pm:p}^=i is an
appropriate <5-sequence centered at TP. If we let um(t) = K[(Fm), where Fm is defined
as in (17.1.5) except with n replaced by r\m := \L + (j,m, then methods similar to those of
Example 16.2.9 show that um(t) —>• u(t) as m —> oo, uniformly for all t € [a, b]. Since
by Theorem 17.2.2 (or Corollary 17.2.6), we have for /-a.e. t € (a, b)

one may expect to obtain in the limit a distributional differential equation for u.
The method just outlined would probably seem quite natural to a physicist having to
deal with a similar situation. Actually, when presented by the second-named author with
some of these results, Richard P. Feynman pointed out that (in the quantum-mechanical
case) strong interactions between elementary particles could be modeled in a similar
manner, although, in practice, it would usually not be necessary to pass to the limit.
Exercise 17.4.2 (a) Make the above limiting argument more precise and guess the form
of the distributional differential equation obtained in the limit.
(b) Verify your guess by taking for u the Cantor-Lebesgue (singular) measure ([Coh,
p. 55 and pp. 22—24] and [Mand, pp. 82-83]) and by approximating it by absolutely
continuous measures associated with (suitable) step functions.
(c) In the quantum-mechanical case, compare your results with Theorem 17.6.15
below ([Lal6, Theorem 2.5, p. 286]), where is provided a suitable distributional dif-
under U.S. or applicable copyright law.

ferential equation. Further, compare directly your answer with the result obtained in
Example 15.5.3; see especially formula (15.5.15).

17.5 Explicit solution and physical interpretations

Let u = u(t) = K[(F) be given by (17.1.6), with the functional F as in (17.1.7). In
this section, we show that the integral equation of Theorem 17.2.1 has a unique bounded
solution, and we establish Theorem 17.2.7, which yields an "explicit" expression for u (t)
in terms of simpler operators. We also provide physical interpretations for our results both
in the quantum-mechanical
EBSCO andCollection
Publishing : eBook Academic probabilistic case, -asprinted
(EBSCOhost) well as
on suggest further
6/8/2017 5:49 connections
PM via
withUNIVERSIDAD
Feynman's operational
DISTRITAL FRANCISCOcalculus. (The latter subject will be pursued in Section 17.6
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 500 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

during our brief discussion of the relationships [La 16] between the present work and
product integration.)
We begin by proving Theorem 17.2.7 for the case of a continuous measure and then
treat the more general case where v is finitely supported.

Continuous measure: Uniqueness of the solution

Recall from Corollary 17.2.6 that when q = /* is a continuous measure, u = u(t) satisfies
the integral equation

for all? e [ a , b ] .
It is easy to see that (17.5.1) has a unique bounded (strongly measurable) solution,
which must therefore equal u in [a, b]. In fact, let v be a solution of (17.5.1) such that
||u(OII < M for all? e [a,£]. Aftern- 1 iterations of (17.5.1), we obtain for t e [a,b]:

where Aa and Ca are given as in (17.3.8) and (17.3.9), respectively. It follows that

Hence
under U.S. or applicable copyright law.

and, by (17.3.6) (and (17.3.10)), v = u on [a, b].

For a < t\ < t% < b, we may therefore consider, as was done in Section 17.2,
the propagator P(ti,t\) associated with the integral equation (17.5.1). Recall that
P(t2, t\)u(t\) = ufo) and that, in particular,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD
is the unique DISTRITAL
boundedFRANCISCO
(stronglyJOSEmeasurable)
DE CALDAS solution of (17.5.1) at time t e [a, b].
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 EXPLICIT SOLUTION AND PHYSICAL INTERPRETATIONS 501
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Measure with finitely supported discrete part: Propagator and explicit solution
Let r] = n + v, with v given by ( 17.2. 1 ). With the obvious change of notation, it follows
from the above discussion that the integral equation

has a unique bounded solution on the interval (rp, TP+I], given by

(We use here implicitly the fact that equation (17.5. 1) holds when applied to an arbitrary
vector \lr e L 2 (R rf ).)
By applying this result successively on the intervals [TO, r\], [r\, TZ], . . • , [r/,, r/,+i],
we conclude that there is a unique bounded function v from [a , b] to C (L 2 (M.d ) ) satisfying
(17.5.2) on (TP, Tp+i] for p = 0, . . . , h. (Recall that by convention, TO = a and T/,+I =
b.) Moreover, v is given by (17.5.3) in (rp, rp+\]

Since, by (17.2.2) in Theorem 17.2.1 and by (17.3.10), u is bounded and satisfies

(17.5.2) on each interval (TP, TP+\], it follows that u = u on [a, b]. In view of (17.5.3)
and (17.5.4), we thus obtain for A. 6 C+ and p e {0, . . . , h},

for all t e (TP, TP+I]', here, we use for O>Q the same convention as in Remark 17.2.5.
Since (17.5.5) is clearly equivalent to (17.2.11), we have established Theorem 17.2.7.
The following comment may help the reader gain some further intuitive understand-
ing of (17.5.5) from a probabilistic viewpoint.

Remark 17.5.1 Note that the Dirac mass a)p8Tp in the discrete part of ij acts in the
immediate present and, according to (17.2.2), (17.2.6) and (17.5.5), influences only the
future, whereas, due to the Markov property [Hid, §2.47, the increments of the Wiener
process forget the past. (See also Chapter 3.) Recall that for X > 0, K^(F) is given by
under U.S. or applicable copyright law.

the Wiener integral (17.1.4) and the exponential function in the definition (17.1.7) of F
is multiplicative.
Since we have proved that the integral equation (17.5.2) had a unique bounded
solution, we may now introduce the associated propagator U. Recall that by definition,
U(t2,ti)u(t\) = «(?2) for a < t] < t2 < b. It follows from the above results that
U(ti,t\) = I and fora < t\ < ti < (3 < b,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 502 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Moreover, for

where Ct (resp. C'0 ) is replaced by Ct2, t1 = C([t2, t1], Rd) (resp., Cgt2,t1) and F (resp.,
K t 2 , t 1 ( F ) ) ) is given as in (17.1.7) (resp., (17.1.4)), except with a replaced by t\ and t
replaced by t2. In particular, for all t e [a,b],

is the solution of the integral equation (17.5.2) at time t.

We point out a special case of interest.
Proposition 17.5.2 (Unitary propagator) Assume that u = l, v is real (i.e. wp is real
for all p) and V is real-valued. Then in the quantum-mechanical case, U(t2, t1) is a
unitary operator for a - t1 - t2 - b.
Proof Since ul — l and in view of Corollary 17.2.6, P(t2,t1) is the standard uni-
tary propagator for the Schrodinger equation (17.2.9) with potential V (or rather, for
the corresponding integral equation (17.2.2)); see, for example, [Schwel]. Further, our
assumptions imply that w p V ( t p ) is self-adjoint and hence that e®?o(tp) = e-iwPv(Tp)
is unitary. Since a finite product of unitary operators is unitary, the conclusion follows
from Theorem 17.2.7.
Remark 17.5.3 Recall that for time independent potentials, the condition 9 € Lool;^
implies that 0 is (essentially) bounded. In this respect, the spirit of the present work is
quite different from the previous work in [La6, Lall, BivLa] presented in Sections 11.3-
11.6 and which focused on defining the "modified Feynman integral" for very general
potentials. Nevertheless, it would be interesting to know whether the results from [La12]
discussed in Section 11.5—or related perturbation results in Section 12.2—can be used
in conjunction with Theorem 17.2.7 above (in the case where u = l) to extend some of
the present results to unbounded potentials. The following exercise asks the reader to
investigate this open-ended question.
Exercise 17.5.4 Suppose that u — l and v is given by (17.2.1). Attempt to determine
a class of unbounded (time independent) potentials to which some of the results of
under U.S. or applicable copyright law.

Section 17.2 (including Theorems 17.2.1 and 17.2.7) can be extended in the diffusion or
in the quantum-mechanical case.
Remark 17.5.5 For the discussion that follows, it is important to observe that since

0 € Lool;n if and only 9 e Lool;u and 6 ( t p , •) is well defined and bounded for each
p e {0,..., h}. Consequently, the values of the potentials w p 6 ( t p ) are completely
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
independent
UNIVERSIDAD of one another
DISTRITAL FRANCISCO and
JOSE of that of 0(s), for (u-a.e.) s e [a, b] \ {t1, ..., Th}.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 EXPLICIT SOLUTION AND PHYSICAL INTERPRETATIONS 503
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In fact, we could just as well have given different names to the potentials ( o p 0 ( t p ) and
e(s)forse[a,b]\{Tl,...,Th}.
We now provide physical interpretations of our results in the quantum-mechanical
and probabilistic cases. For simplicity, we assume throughout our discussion that u = l
and v = Ehn= w) n S T n , with to = a < TJ < • • • < T/, < b = TA+I .

Physical interpretations in the quantum-mechanical case

Suppose that A = —i and 9 = — i V with V € Lcci;^- Then P(t2, ?i) is the propagator
for the Schrodinger equation (17.2.9) with potential V (or rather, for the corresponding
integral equation (17.2.2)). Recall from Proposition 17.5.2 that if V is R-valued, then
P (t\, t-i) is a unitary propagator. In particular, if V is also assumed to be time independent,
then P(t\, t\ + T) = e~'T<-H°+V) is the unitary group generated by the Hamiltonian
H := Ho + V.

Consider a single nonrelativistic quantum-mechanical particle. We propose to inter-

pret (17.5.8) as follows: The particle moves in the potential V between times a and TJ,
interacts instantaneously with the potential w V(T\) at time t1, moves (or propagates)
in the potential V between times t1 and T2, interacts instantaneously with (or is scattered
by) the potential w2 V(t2) at time T2, and so on until a final instantaneous interaction at
time tp followed by a propagation between TP and t.
If the particle is in the initial stated e L 2 (K d )attimea, we interpret \jf(s) := U(s)tjf
as representing its state at time s e [a,b]. According to Theorems 17.2.1, 17.2.2,
17.2.4 and 17.2.7, \js(s) obeys the Schrodinger equation (17.2.9) with potential V dur-
ing each time interval (r ? , Tq+\), 0 - q - p; for t\, t-i inside this interval, we have
i/r(t2) = P(t2,t\)\jr(t\). However, immediately after time rq (in reality, a very short
time afterwards), the particle is in the state
under U.S. or applicable copyright law.

As was seen above, the factor e~ lwq v(tq) can be interpreted as the result of a change
of phase due to an instantaneous interaction with (or a scattering by) the potential
wqV(rq) at time Tq. Similarly, it can also be thought of as being created by a "shock"
or a "hit" of some kind. (Strictly speaking, we should use the expression "change of
phase" only in the most common case where wq and V(tq) are real; see Proposition
17.5.2 above which then implies that U(t2, t1) is a unitary propagator.) The reason why
we may use the term "instantaneous interaction" should be clear by now; we refer, for
instance, to the discussion at the end of Section 17.4 (as well as in Example 16.2.9),
where
EBSCOwe approximated
Publishing the Dirac
: eBook Academic mass (EBSCOhost)
Collection (wq&rq by -absolutely continuous
printed on 6/8/2017 5:49 PMmeasures
via with
density sharply
UNIVERSIDAD concentrated
DISTRITAL about
FRANCISCO JOSE DE time
CALDAS rq .
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 504 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

[Caution: The expression "change of phase " used above solely refers to the operators
involved, and not to the wave functions (considered as functions of the space variable).
Hence it corresponds to an observable physical effect.]
The results from [Lal5] described in this chapter discussed here could possibly
provide a simplified model for multiple scattering, as we next further illustrate. (We
caution the reader that there are several defects with this model, one of which is mentioned
below.) Imagine that a particle is projected onto a thick target; it is then subjected to
multiple scattering when it collides with the atoms of the target, say at times T\ ,..., TH •
When the projectile leaves the target, it is in the state ty(t) = u(t)\lf, with u(t) as in
(17.5.8) and with p = h. An obvious problem here is that one does not know how many
scatterings occur or at what times they take place; in order to obtain a more realistic
model, it may therefore be necessary to use some kind of averaging (or randomization)
process. (See Problem 17.6.31(c) and Remark 17.6.32(a) below.) Actually, because of
the difficulties involved, experimental physicists try to avoid multiple scattering by using
a very thin target; ideally, a single scattering occurs in that case, and from our point of
view, we are in the simple situation where n — u + w S T :

Remark 17.5.6 A special case of interest is obtained when V = 0 in [a, b ] \ { t 1 , . . . , th).

Then P (t2, t1) = e~l<-t2~'^H°, and the particle propagates freely during the time interval
(rq, Tq+\), 0 - q - h. Since V = 0 l-a.e. in [a, b] in this case, everything happens as
if n] were purely discrete and equal to v. (See Corollary 17.2.9.) The analogue of this
remark holds in the probabilistic case, where P(t2,t1) = e ~ (t2 ~t1 )Ho represents free
diffusion.
We now turn to the probabilistic or diffusion case; we shall be more concise in
our explanations, since, once the proper identifications have been made, the discussion
parallels that of the quantum-mechanical case.

Physical interpretations in the diffusion case

Assume that A = 1 and 9 = —V, with V e Looi.ij- Then P(t2, t1) is the propagator
for the heat equation (17.2.8) with potential V (or rather, for the corresponding integral
equation (17.2.2)). In particular, if V is time independent, P(t1, t1 + T) = e - T ( H ° + V )
under U.S. or applicable copyright law.

is the heat (or diffusion) semigroup generated by H = HO + V.

Since it may not be as well known as in the quantum-mechanical case, we briefly
recall the physical meaning of P (t2, t1). (See [Klul, pp. 161-163 and p. 177] for an
elegant exposition.) Imagine a solvent which fills the entire space Rd. We suppose that
"a substance reacting with [this] solvent spreads through it according to the laws of
diffusion" [Klul, p. 16]. The rate of reaction and diffusion is proportional to the amount
of substance present, the coefficient of proportionality at time s and location E being
equal to V (s, E); with our sign convention, V(s,%) - 0 or - 0 means that the substance
is being destroyed or created, respectively. Now, the propagator P(t2, t1) corresponds to
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
the UNIVERSIDAD
phenomena of reaction
DISTRITAL andJOSE
FRANCISCO diffusion
DE CALDAStaking place simultaneously.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 EXPLICIT SOLUTION AND PHYSICAL INTERPRETATIONS 505
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

If the substance has T]S e Ll(E.d) n L2(E.d) (ijr > 0) as its initial mass distribution
at time a, we interpret ^(*) := M(*)^ as being its mass distribution at time £ e [a, b].
(Note that since V(rg, •) is bounded, by Remark 17.5.5, it follows from (17.5.10) that
V^(s) e L1 n L2 and yf(s) > 0; a more natural assumption to obtain this result would
bethato> 9 V(r 9 , •) > 0.) According to Theorems 17.2.1, 17.2.2,17.2.4and 17.2.7, VCO
satisfies the heat equation (17.2.8) with potential V during each time interval (rp, r p+ i);
however, immediately after tq, the mass distribution of the substance is given by

This suggests that at time r?, an "instantaneous reaction" with rate characterized by
a>qV(Tg, •), occurs throughout space. No diffusion is then taking place. Somehow, the
diffusion is "frozen" at time rq.
In the following discussion, we assume that V > 0 and coqV(Tq, •) > 0 for all
q e ( l , . . . , / z ) . We now adopt an alternative interpretation of Pfo, t\) in probabilistic
terms (see, e.g., [Klul, p. 177]). Consider a Brownian particle which is exposed to a
risk of destruction characterized by V. (If at time s the particle is located at £ e Rd, the
probability that it will be destroyed within the short time period AT is approximately
equal to V(s, £)AT.) Now, in view of (17.5.10), we may interpret the contribution of
the discrete part v of of as creating an additional risk of destruction, characterized by
wq wq, •) and occurring throughout space at precisely each instant rq.

Further connections with Feynman 's operational calculus

An interesting aspect of the above results and of the theory developed in [JoLal] (Chap-
ters 15 and 16 above) is that they allow us to blend discrete and continuous structures
by use of Lebesgue-Stieltjes measures, as well as to unify concepts which were not
previously thought to be directly related.
This work can be extended in various directions. In particular, by making more
under U.S. or applicable copyright law.

explicit use of the rules of Feynman's operational calculus in place of path integral
methods, some of the results of this chapter will be extended to a more general setting in
Chapter 19 (see Theorems 19.5.1 and 19.6.1). The techniques of Chapter 19 have a good
deal in common with some of those of the present chapter and of Chapter 15. This is not
so surprising (after the fact) if one recalls from Sections 14.3 and 14.4 that Feynman's
operational calculus for noncommuting operators can be thought of as generalizing some
aspects of path integration.
There doubtless exist other derivations of the "Feynman-Kac formula with a
Lebesgue-Stieltjes measure". This is especially true in the probabilistic case, where
we EBSCO
are aware of several
Publishing alternative
: eBook Academic proofs.
Collection (The quantum-mechanical
(EBSCOhost) case
- printed on 6/8/2017 5:49 PM could
via then be
handled by analytic
UNIVERSIDAD DISTRITAL continuation followed
FRANCISCO JOSE DE CALDAS by a suitable limiting argument.) However,
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 506 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

we have chosen here to emphasize an approach that is very close to the one that led the
second author to discover this formula in [Lal4,15]. It possesses a rich combinatorial
structure and makes explicit connections with the disentangling process, which—as was
discussed in Chapter 14—is central to Feynman's operational calculus [Fey8] and is
one of the main themes of Chapter 15. In fact, it yields a very concrete example of the
usefulness of our generalized Dyson series in [JoLal] (Chapters 15 and 16) which, as
we recall, provide a way of disentangling the operator u (t).
It is natural to wonder whether the "miraculous cancellations" that occur in the course
of the proof of Theorem 17.2.1 (especially in the general case omitted here but studied
in detail in [Lal5, §4]) do not suggest the existence of an underlying mathematical
structure. This is indeed the case. Actually, motivated by the present work and that in
[JoLal] (Chapters 15 and 16 above), one can define a new noncommutative and time-
ordered multiplication on the space of Wiener functionals. It explains, in particular,
the striking similarity between the expression for the functional F in (17.1.5) (with
1 = V- + Ej=i WA, as in (17.2.1) and t € (rp, r p +i]),

and the formula (17.5.5) obtained in the threorem 17.27 for he operotor as in (17.1.6):
under U.S. or applicable copyright law.

In the quantum-mechanical case, the passage from (17.5.11) to (17.5.12) could be

considered as a kind of quantization procedure. In Chapter 18—which discusses the
authors' work in [JoLa4]—we will develop this deeper interpretation of a rigorous and
extended form of Feynman's operational calculus for noncommuting operators.
Remark 17.5.7 In his 1948paper [Fey2, pp. 381-382], Feynman noted somewhat as we
did just above the correspondence between time-ordered functional and time-ordered
operators obtained by path integrating the functional. It seems likely that such obser-
vations played a key role in his later formulation of the perturbation series in quantum
electrodynamics [Fey5—7] and in his development of the operational calculus for non-
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
commuting operators
UNIVERSIDAD [Fey8]. JOSE DE CALDAS
DISTRITAL FRANCISCO
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 507
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

17.6 The Feynman-Kac formula with a Lebesgue-Stieltjes measure:

The general case (arbitrary measure 77)
In this section, we give an overview of the main results obtained by Lapidus in ([Lal8],
[Lal6]). These results extend to an arbitrary measure rj the corresponding ones from
[Lal5] discussed in Section 17.2, where v was assumed to be finitely supported. (See
[Lal8].) In the quantum-mechanical case, they also provide information—such as a
product integral representation of the solution and a distributional differential equation—-
that are new even in the setting of Sections 17.2-17.5. (See [Lal6].)
We now briefly recall our assumptions and notation from Section 17.1. Let n €
M([a, b]) be an arbitrary (complex-valued and hence bounded) Borel measure on [a, b].
Let n = n. + v be the unique decomposition of n into its continuous part u, and its discrete
part v. We write

where {rp}™=l is a sequence from [a, b} and {cap}^=l is a sequence from C such that
Y^L\ \U>P\ < °°- (Without loss of generality, we may assume that all the rp are distinct.)
Note that in (17.6.1), the times rp need not be naturally ordered; they could be,
for instance, all the rational numbers in the interval [a, b]. This is, of course, in sharp
contrast with the situation of Sections 17.2-17.5 where the definition of v involved only
finitely many times TP. (Compare (17.6.1) with (17.2.1).)
Finally, given 9 e Looi;r/ = Looi;ij([a, b)), t e [a, b] and A. e C+, we let u(t) =
K((F) = KL(F)*, as in (17.1.6), where F and its time-reversal F are given by (17.1.5)
and (17.1.7), respectively.

Integral equation (integrated form of the evolution equation)

The first main result ([Lal8, Theorem 4.3, p. 185]) provides, in this general setting,
the integrated form of the evolution equation. We will refer to it as the "Feynman-Kac
formula with a Lebesgue-Stieltjes measure" (briefly, FKLS) in the general case.

Theorem 17.6.1 (FKLS in the general case) For all X e C+, the operator-valued func-
under U.S. or applicable copyright law.

tion u satisfies the following Volterra-Stieltjes integral equation on [a, b]:

for all t 6 [a, b], where 0M is given by (17.6.5) below.

The integral in (17.6.2) is a strong Bochner integral; so that, for all \{r e L2(K.d)
and t 6 [a, b],

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 508 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Further, for all t 6 [a, b],

Here, we have set

where rj(s) := n([s]) and <p is the entire Junction defined by for all
that is,

Note that #M is defined by means of the holomorphic functional calculus in

the (complex) Banach algebra £(L2(E.d)). (See, e.g., [KadRi, §3.3].) We will see in
Lemma 17.6.24 ([Lal8, Lemma 2.1, p. 170]) that since 0 e Looi;n, so does 9v^-t hence,
0<p^(s) € £(Z,2(Rd)) for Tj-a.e. s in [a,b]. (In the following, for notational simpli-
city, we will do as though #M(.5) were in £(L 2 (R rf )) for every s e [a, b]. We refer to
Lemma 17.6.24 below and the comments preceding it for some other useful properties
of# M .)
Moreover, observe that 6Vttl is the (bounded) multiplication operator in £(L2(M''))
by the (essentially bounded) Borel measurable function, still denoted 0M, defined by

As will be further discussed below, the replacement of 9 by 0M in formula (17.6.3)

was motivated in part by the work of Dollard and Friedman ([DoFri2], [DoFri3,
Chapter 5]) in the related context of product integrals of matrices with respect to
measures.
Of course, if s is not a "pure point" of n (i.e. if n(s) = 0), then Ov,n(s) = 0(s).
(In view of (17.6.5), this follows since 6(0) = 1. Here, we use the term "pure point
of or "atom of 77" to mean any point 5 in [a, b] such that n(s) = 0, or equivalently,
any point rp in the support of v; that is, such that a)p ± 0.) In particular, if /? = u
is a purely continuous measure, as in Corollary 17.2.6, then (17.6.2) becomes (17.2.7).
under U.S. or applicable copyright law.

Recall that, in general, u has both an absolutely continuous and a singular part. If we
assume in addition that n = u = I, the ordinary Lebesgue measure on [a, b], then as
was discussed in Section 17.2, we obtain the classical Feynman-Kac formula in the
probabilistic case (A = 1 and 6 = — V) and an interpretation of the Feynman path
integral in the quantum-mechanical case (A = — i and 0 = —iV). (See, e.g., [CaStl,2]
and [JoSk6].)
We stress that if v is finitely supported, as in Section 17.2, then the integral equation
(17.6.2) in Theorem 17.6.1 is equivalent to that provided by (17.2.2) (fbr p = 0 , . . . , h)in
Theorem 17.2.1. Actually, as we shall see at the end of this section, this basic fact ([Lai 8,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
Theorem 3.2 DISTRITAL
UNIVERSIDAD (and Proposition 5.1),
FRANCISCO JOSE pp. 173 and 198])—which may not seem entirely
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE rf) 509
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

obvious —constitutes the first step of the proof of Theorem 17.6.1. (See Theorem 17.6.23
below.) The second step of the proof of Theorem 17.6.1 consists in approximating an
arbitrary Borel measure rj (with discrete part given by (17.6.1)) by measures ijm with
finitely supported discrete part of the form: rjm = n + Y^"^={ CDpSrp,farm = 1,2 .....
(See Theorem 17.6.27 below [Lai 8, Theorem 4.2, p. 180].)
In the next result ([La 1 8, Theorem 5. 1 , p. 1 92]), we establish in particular the unique-
ness of the solution to the evolution equation (17.6.2).

Basic properties of the solution to the integral equation

Theorem 17.6.2 Fix A. e C+. Then, under the same assumptions as Theorem 17.6.1,
the integral equation (17.6.2) (with u replaced by v) has a unique bounded (strongly
measurable) solution v : [a, b] —> £(L 2 (R d )), necessarily equal to the function u of
Theorem 17.6.1. Moreover, for all t e [a, b], u(t) = K[(F) is given by the following
time-ordered exponential series:

where

and for

with defined as in (17.65).

In (17.6.9), the series is summable in £(L 2 (R )) and the integral is a strong Bochner
integral.
Remark 17.6.3 It follows from Theorems 17.6.1 and 17.6.2 that for each t e [a, b], the
time-ordered exponential series in (17.6.7) is equal to the time-reversal (in the sense of
under U.S. or applicable copyright law.

Theorem 15.7.6(ii)), of the CDS given by formulas (15.5.25)-( 15.5.28) (withn\an = lj

of Example 15.5.7. (Indeed, by Example 15.5.7 and Theorem 15.7.6, u(t) = K[(F) is
equal to the time-reversal of the series in (15.5.25); further, by the uniqueness statement
in Theorem 17.6.2, u(t) is also equal to the series in (17.6.7).) This fact was certainly
not obvious a priori. Note that the "disentangling process" (in the sense of Feynman's
operational calculus as developed in Chapter 15) is carried out much further in the
generalized Dyson series in (15.5.25) than in the series in (17.6.7). However, the latter
series is written in a rather concise form.
We next give further properties of the solution to the integral equation (17.6.2). (See
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
[Lal8, Theorems 5.2 and 5.3, pp. 204-205].)
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 510 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 17.6.4 (a) (Time discontinuities) Fix A. e C+ and fef u(f) = K((F) be the
unique solution of the evolution equation (17.6.2), as in Theorem 17.6.1. Then u = u(t)
is strongly left-continuous on (a, b] and is strongly continuous at every point of [a, b]
which is not a pure point of n. Moreover, for every r 6 [a, b) such that w := n ( r ) 7^ 0,
we have

where M(T+) denotes the strong limit of u(t) as t —» r, t > i (i.e. for every i/r 6
L 2 (R d ), limt-t,t > u(r)tfr = e w0(r) u(r)VO.
ffcj The function z defined by z(t) - u(t) - e -(t-a)(H0) /5 strongly of bounded
variation on [a, b].
Exercise 17.6.5 (a) Prove part (a) of Theorem 17.6.4 in two different ways:
(i) By using the integral equation (17.6.3) obtained in Theorem 17.6.1.
(ii) By using the corresponding result (Theorem 17.2.4) obtained in Section 17.2 for
measures with a finitely supported discrete part, and then by approximating 77 by
rim •= M + Ep=i WPSV
[Hint: In case (i), use the identity

which follows from the definitions (17.6.6) and (17.6.5) of w andOy^, respectively. (See
formula (17.6.38) in Lemma 17.6.24(b) below.)
In case (ii), use Theorem 17.6.27, the (strengthened) stability theorem relative to
measures ([Lai 8, Theorem 4.2, p. 180]) stated below; see [La 8, pp. 204-205] for mo re
details.]
(b) Show directly (that is, without using Theorem 17.6.2) that the series in (17.6.9)
and the time-reversals of the GDSs in (15.5.25) coincide.
[Advice: For combinatorial simplicity, you may assume at first that v is finitely supported,
as in Section 17.2; then a direct computation shows that the series in (17.6.7) and the
GDS in (17.3.34) are equal.]
Remark 17.6.6 (a) Theorems 17.6.1, 17.6.2, as well as (when v is finitely supported)
under U.S. or applicable copyright law.

Theorem 17.2.7 and Corollary 17.2.9, all suggest the existence of a (strong) product
integral in this context. For instance, equation (17.6.2) looks like Duhamel's formula or
the "sum rule " [DoFriS, pp. 33, 106 and 1 72]; further, formula (1 7.6. 7) yields the corres-
ponding time-ordered perturbation expansion, which is often taken to be the definition
of the product integral (see the discussion in the introduction of [DoFril]). Finally,
formula (17.2.11) shows that, when v is finitely supported, it suffices to establish the
existence of the product integral for a continuous measure. For the sake of brevity and
simplicity, however, we choose not to consider the general case when A. e C^ here but
to focus instead on the quantum-mechanical case in the rest of this chapter. Actually, the
corresponding abstract theory of product integration (developed in Section 3 of [Lal6])
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
is more elegant
UNIVERSIDAD in thisFRANCISCO
DISTRITAL setting.JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE t)) 511
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(b) (Discreteness in space vs. discreteness in time.) We have previously discussed

how to reinterpret a simple case of the Trotter product formula in our context by letting
weighted sums of Dirac measures (in time) converge to Lebesgue measure I. (See, e.g.,
Corollary 17.2.9 and especially Example 16.2.7.) When we assume that /u = /, it would
be interesting (but probably difficult) to investigate whether the work in [Hug 1,2]—
which was partially motivated by [La6,ll]—could be adapted to this situation in order
to show the existence of the product integral for more singular potentials than those
considered below, such as, for example a Dirac measure in the space (rather than in the
time) variable.
We close this subsection by considering the "propagator" {/(•, •) associated with
the integral equation (17.6.2); that is, for each t\,ti e [a, b] with t1 - 12, Ufa, t1) e
£(L 2 (Rd)) sends the solution of (17.6.2) at time t\ onto the solution at time t2- (In
particular, with this notation, we have u(t) = U(t, a) for all t in [a, b].) Clearly, for
A. > 0, U (t2,t1) is also given by the right-hand side of (17.1.6), but with the time
interval [a, t) replaced by [t\, t2) in the definition of the functional F given by (17.1.5).
(Note that the operator Ufa, t\) was denoted P (t2, t1) in Section 17.2.)
Theorem 17.6.7 (a) (Propagator) For all t1,t2, t3 in [a, b] with t\ <t2< t3, we have

(b) (Unitary propagator) Assume that we are in the quantum-mechanical case (A, =
—i and 9 = —iV, with V e LOO,;/^), as we did in (17.5.6). If in addition, n = l, the
ordinary Lebesgue measure on [a, b], v is real (i.e. a>p is real for all p > 1) and V is
real-valued, then Ufa, t\) is a unitary propagator (with inverse identified with U(t\, 12))
for all t1, t2 in [a,b]). (Note that this implies in particular that u(t) = U(t,a) is unitary
for all t in [a,b].)
Theorem 17.6.7 is proved much as is suggested in part (ii) of Exercise 17.6.5(a),
by deducing the stated property from the corresponding fact (equation (17.2.10) and
Proposition 17.5.2, for part (a) and (b), respectively) for measures with finitely supported
discrete part. (Note that a norm limit of unitary operators is still unitary.)
The remainder of this chapter will be largely devoted to the study of the unitary
propagator Ufa, t\) (and, in particular, of u(t) = U(t, a)) in the quantum-mechanical
case (but under more general assumptions and with different notation, to be introduced
under U.S. or applicable copyright law.

below.)
Quantum-mechanical case: Reformulation in the interaction (or Dirac) picture
We now focus exclusively on the quantum-mechanical case. This is the situation of great-
est physical interest from our present perspective since it leads to a suitable interpretation
of the Feynman path integral in this context. Thus we assume throughout that A. = —/
and 6 = —iV, with V e Loo1;n, b)). We continue to use our earlier notation; in
particular, u(t) = K'^F) is defined as in (17.1.6), with F as in (17.1.5).
It will
EBSCO be convenient
Publishing to begin
: eBook Academic by reformulating
Collection our basic
(EBSCOhost) - printed results5:49
on 6/8/2017 in the "interaction
PM via
(or UNIVERSIDAD
Dirac) representation".
DISTRITAL FRANCISCO(For anCALDAS
JOSE DE introduction to this subject, closely related to the
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 512 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

"Dirac or interaction picture" of quantum mechanics, we refer, for instance, to [Han,

§11.1], [Sud, §3.5], or [Berl, Appendix on Wick's theorem, pp. 203-222].)
Fix V e L2(Rd). For s € [a, b], set

and for w

Since u(a) = I, we have Q(a) = I and ^(a) = T/T. Clearly, Q(s) € £(L 2 (M d ))
and $(s) e L2(R.d) for all s e [a, b]; further, W(s) € £(L 2 (K d )) for n-a.e. s € [a, b].
[Note how the "observables" (operators) W(s) as well as the "states" (vectors) ur evolve
with time s in the "interaction (or Dirac) representation"; see (17.6.12c) and (17.6.12b).
(On the other hand, in the closely related "Heisenberg picture" of quantum mechanics—
which can be viewed as a special case of the "interaction picture" [Han, §11.1 ]—the states
remain constant while the observables evolve with time according to (17.6.12c); see, e.g.,
[Han, §11.1], [Sa, § 1.3, pp. 4-5], or [Sud, §3.4 and §3.5]. We remark that in the literature,
the time-dependent operator W is often called the "Heisenberg representation" of the
operator W. Of course, the symbol ^ does no^stand here for the Fourier transform.)]
We shall simply write WM instead of (W) M . Indeed, in view of (17.6.5) and
(17.6.6), it is easy to check that

(Even though W does not necessarily belong to LQO^, it is still possible to define (W) M
by (17.6.5); see also [La 16, §3].)
We can now reformulate Theorems 17.6.1 and 17.6.2 in the quantum-mechanical
case in (the analogue in our context of) the "interaction representation".
Theorem 17.6.8 The function Q = Q(t) satisfies the following Volterra-Stieltjes inte-
gral equation on [a, b]:
under U.S. or applicable copyright law.

Further, the integral equation (17.6.14) (with Q replacedby R) has a unique bounded
(strongly measurable) solution R : [a, b] -> C(L2(E.d)), necessarily equal to Q.
Moreover, for all t e [ a , b ] , Q(t) is given by the following time-ordered exponential
series:

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 513
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where A^(0 is as in (17.6.8) and for (s\,... , sn) e Aj,(f),

The series in (17.6.15) is (absolutely) summable in £(L2(Kd)) and the integrals in

(17.6.14) and (17.6.15) are strong Bochner integrals; so that, in particular, we have for
all r 6 [a, b],

In addition, for all t e [a, b],

Proof This follows easily from Theorems 17.6.1 and 17.6.2. We provide the proof,
however, in order to explain how the expressions of the operators involved are signifi-
cantly simplified in the interaction representation. (Compare, for example, (17.6.16) and
(17.6.9).) Note that since by (17.6.2) (applied with 6 := -iV),

we have, in view of [HilPh, Theorem 3.7.12, p. 83] and (17.6.12)-(17.6.13),

this establishes (17.6.14).

We deduce similarly (17.6.15) from (17.6.7).
under U.S. or applicable copyright law.

Remark 17.6.9 (a) If ^, the continuous part of r), is equal to Lebesgue measure I, we
have seen in pan (b) of Theorem 17.6.7 that u(t) is unitary, for n and V real; in view of
(17.6.12a), Q(t) is also a unitary operator in this case. In particular, we have \\Q(t)\\ = 1
and not merely \\Q(t)\\ -exp(|| VHoo,-,,) as statedin (17.6.18).
(b) In general, (—i V)v^ is not equal to — i (V)v<r). However, if n = u is a continuous
measure, then it follows from (17.6.12c) and (17.6.13) that

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
TheUNIVERSIDAD
formulas DISTRITAL
(17.6.14)—(17.6.17) can
FRANCISCO JOSE DE thus be simplified accordingly in this case.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 514 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(c) It is noteworthy that, in the quantum-mechanical case, all the time-ordered

perturbation expansions (GDSs) obtained (or discussed) in Chapters 15-16 (as well as in
the previous sections of the present chapter) can be rewritten in the interaction represen-
tation. This provides a much more concise (and possibly more natural) expression/or our
generalized Dyson series. As an exercise, some readers may find it helpful to check this
fact concretely in a few examples; for instance, for the GDS defined by (17. 3. 2)-(17.3. 5)
(or (17.3.6)-(17.3.9)) or, more generally, for the GDS given by formulas (17.3.34)-
(17.3.35) (and corresponding to the "time-reversal" of that in (15.5.5)-(15.5.7) from
Example 15.5.1).

Product integral representation of the solution

We still suppose that we are in the quantum-mechanical case and continue to discuss
the results from [Lal6]. We stress that all the results provided through to the end of this
chapter (with the exception of Theorem 17.6.14) will be new and of interest even in the
context of Sections 17.2-17.5, where the discrete part of n was supposed to be finitely
supported.
The following Theorem ([Lal6, Theorem 2.2, pp. 282-283])— which is obtained
by combining Theorem 17.6.8 above ([Lal6, Theorem 2.1]) with Theorems 3.1, 3.3,
and 3.4, pages 293, 299, and 304 in [Lal6] —provides the key link between the gen-
eral "Feynman-Kac formula with a Lebesgue-Stieltjes measure" (as presented in The-
orem 17.6.8) and the theory of strong product integration of measures (as elaborated
in [Lal6, §3]). The theory of (strong) product integrals (for general Borel measures)—
also called "time-ordered multiplicative (or chronological product) integrals" — has been
developed in the present context by Lapidus in [La 16], building on earlier work by
Dollard and Friedman [DoFril-3] and others (and going all the way back to early work
(on the composition of) integral operators by Volterra ([VolHos], [Vol]) which can be
thought of in some respects as a precursor of some of Feynman's ideas on his oper-
ational calculus [Fey8]); see Remark 17.6. ll(b) below. For conciseness, however, we
will not provide here all the necessary definitions and details of the theory, for which
the interested reader is referred to [Lal6, esp. §3] (as well as to the book [DoFri3] for
the general framework in more particular situations). We note that for many readers,
Theorem 17.6.12 below ([Lal6, Theorem 2.3, p. 285]) may provide a suitable substitute
for the general definition ([La 16, Theorem 3. 1, p. 293]) of the product integral appearing
in formula (17.6.19). (Later on, we will use indifferently n[a,t) or nj, in the expression
under U.S. or applicable copyright law.

of the product integral.)

Theorem 17.6.10 Let n be an arbitrary Borel measure on [a, b] and let V € LOO,;,,.
Then the strong product integral of—iV with respect to the measure n exists over [a, b)
[and hence over every subinterval of [a, b)].
Moreover, for all t € [a , b], we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
where V is defined
UNIVERSIDAD byFRANCISCO
DISTRITAL (17.6.1 2c).
JOSE (In particular, Q(t) € C(L2 (Rd), for every t E [a, b].)
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 515
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The product integral in (17.6. 19a) converges in the strong operator topology; so that,
for every faced w e L 2 (R d ),

wherep(t)isgivenby(17.6.12b).(Recallthatp(a)=p.)
Here is a sketch of the proof of Theorem 17.6.10: By [Lal6, Theorem 3.1, p. 293],
the product integral on the right-hand side of (17.6.19) exists; thus, for every fixed
t e [a, b], formula (17.6.19a) defines a bounded linear operator in £(L 2 (R af )), denoted
by w(t). Next, by [Lal6, Theorem 3.3, p. 299], w = w(t) is a bounded solution of
the integral equation (17.6.14) on the entire time interval [a, b]. Since—according to
Theorem 17.6.8—such a solution is unique and necessarily equal to Q(t), as defined by
formula (17.6.12a) and by the time-ordered exponential series in (17.6.15), we conclude
that (17.6.19a) holds; that is, for all t e [a, b], w(t) = Q(t) and hence Q(t) is given by
the product integral in (17.6.19a), as desired.
For a detailed proof of Theorem 17.6.10 ([Lal6, Theorem 2.2, pp. 282-283]), we
refer the interested reader to [Lal6, pp. 283-284].
Remark 17.6.11 (a) Roughly speaking, product integrals are multiplicative and non-
commutative analogues of the usual (additive) integrals. In the physics literature (and
in the classical case when n = u = I), the product integral in (17.6.19a) is often called
a "chronological" or "time-ordered" (exponential,) integral and is denoted by

(with n(ds) = ds), where T stands for the time-ordering symbol ([Dyl,2,4], [Fey8]).
For instance,

(See, for example, [Sa, p. 5]. Note the complete agreement between (17.6.20b) and
Feynman's time-ordering convention (14.2.1) discussed in Section 14.2.) With a similar
under U.S. or applicable copyright law.

notation and in the general case of Theorem 17.6.10, Equation (17.6.19b) yields

(We use here Dirac's "bra and ket" notation [Dir2].) Of course, the product integral
used in Theorem 17.6.10 and throughout the end of this Section are not merely formal
expressions but are precisely defined mathematical objects, as is explained at some length
in Section 3 of [La16].
(b) A rigorous treatment of strong product integrals, in the case when n — I—as well
as of product
EBSCO integrals
Publishing : eBookofAcademic
measures in the(EBSCOhost)
Collection case of finite-dimensional
- printed on 6/8/2017matrices—is
5:49 PM via given in
[DoFriS], Chapters
UNIVERSIDAD DISTRITAL3FRANCISCO
and 5, respectively.
JOSE DE CALDAS (See also the research articles [DoFril,2] and
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 516 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

the relevant references therein.) In [Lal6], Section 3, the second author has unified and
(partially) extended both approaches to strong product integrals of measures, in order to
make sense, in particular, of the product integral appearing in (17.6.19) and to establish
some of its main properties [such as the associated evolution equation in integrated or
(distributional) differential form]. Note that in (17.6.19), an arbitrary Borel measure
occurs and that, in addition to V, the unbounded operator HQ acting in the infinite
dimensional space L 2 (R d ) appears in disguised form (see (17.6.12c) above).
(c) The theory of (strong) product integration of measures developed in [Lai 6] is
sufficiently general to handle considerably more abstract situations than the present one.
Although this has not yet been carried out explicitly, it should help provide a product
integral representation of the solution to the evolution equation later obtained in our
(joint) paper [dFJoLa2J and discussed in Chapter 19 below. [In [dFJoLal,2], the set-
ting for "Feynman 's operational calculus" involves suitable (noncommuting) abstract
bounded linear operators acting in a Hilbert space, as well as (unbounded) generators
of (Co) contraction semigroups; see Chapter 19.] Moreover, it could possibly help estab-
lish connections between aspects of the present work and the notion of "expansionals"
in Banach algebras introduced by Araki in [Ar].
(d) An informal introduction to the notion of product integrals, primarily directed
towards physicists, is provided in [dWmMN2, §2, pp. 284-293], in the simple case
of finite-dimensional matrices and of ordinary Lebesgue measure. We have to disagree,
however, with the statement in [dWmMN2, p. 284] according to which "product integrals
are a simple and rigorous vehicle for Feynman's operator calculus." In fact, both on the
basis of his paper [Fey8] and of the second author's private conversations with him, we
think that Feynman had a much more ambitious program in mind, extending well beyond
the mere use of the (admittedly very important) exponential function. It would perhaps
be more accurate to state that close relationships exist between product integration and
Feynman's operational calculus.
Fix t e [a,b] and let P be an arbitrary partition of [ a , t ) . HP is of the form
S0 := a < s1 < • • • < t =: sn, we set Ik = [sk-1,Sk)> for k = 1,... ,n; further, as
usual, we define the mesh of P by mesh(P) = max1<k<H |sk — Sk-1|.
Now, in view of Theorem 17.6.10, the next result ([Lal6, Theorem 2.3, p. 285])
follows from Theorem 3.1 and Proposition 3.1 in [Lal6, p. 293]. We stress that all the
remaining theorems in this section (with the exception of Theorem 17.6.20) have exactly
under U.S. or applicable copyright law.

the same hypotheses as Theorems 17.6.8 and 17.6.10 above.

Theorem 17.6.12 With the notation introduced just above, the following (strong) limit
exists as the mesh of the partition P tends to 0, and is equal to the (strong) product
integral appearing in (17.6.19a); namely, for all t e [a, b],

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 517
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The integrals in (17.6.21) are strong Bochner integrals and the limit holds in the
strong operator topology, uniformly for t € [a,b]; so that, in particular, for every
^ e L2(E.d), w ( t ) = Q(t)t/f ^(appearing in (17.6.19b)) is obtained by applying the
right-hand side of'(17'.6.21) to w ( a ) = w,
Remark 17.6.13 We briefly comment on the rigorous definition—given in [Lal6, §37—
of the productintegral occurring in (17.6.19a) and (17.6.21). Fort e [a,b], the(strong)
product integral, ^ \ ^ a t ) e x p { B ( s ) n ( d s ) } , of an operator-valued function B : [a,b] -
£(L 2 (R d ) (with respect to the Borel measure n on[a, b]) is first defined for a step func-
tion (and then denoted Y[g (t)), much as in the theory of Riemann-Stieltjes integrals, but
in a multiplicative rather than additive way (and also by taking into account the non-
commutativity of the exponential function in a Banach algebra). (See [Lal6, Definition
3.4, p. 292].) Then, for a general (suitable) function B, it is defined as the (strong) limit
of the operators \\Bn(t) (uniformly in t e [a, b]) associated with a sequence {Bn}^1
of step functions converging (in an appropriate sense) to B. (See [Lal6, Theorem 3.1,
p. 293].) Of course, it can be shown that for the class of functions B considered in
[La}6], there always exists a sequence of step functions converging to B, and that, in
addition, the (strong) limit of{\\B (t)}%Li always exists and is independent of the choice
of the sequence of step functions (suitably) converging to B. (See [Lai6, Proposition
3.1 and Corollary 3.1, p. 293].) [We note that the setting of [Lal6, §37 is more general
than indicated here, as it allows for functions B with values in a strongly closed Banach
subalgebra of C(X), where X is an arbitrary Banach space.]
The next result is just a restatement of Theorem 17.6.4 in the interaction repre-
sentation. (In the light of Theorem 17.6.10, it also follows directly from the prop-
erties of the product integral obtained in [Lal6, Theorem 3.2(d), p. 297]. Note that
e«?(*) = eisH0eaV(s)e-isH0 for a e C and ^.a.e. s e [a< b),)

Theorem 17.6.14 The (operator-valued) function Q = Q(t) is of bounded variation

on [a, b]. Moreover, it is left-continuous on all of (a, b] and is continuous at every point
of [a, b] which is not a pure point of n. Further, if r is a pure point of rj, then Q(r+)
exists and
under U.S. or applicable copyright law.

(i.e. u(r+) = e "°V^U(T)), where w := n(T) ^ 0. These statements must be under-

stood in the strong operator topology.

Distributional differential equation (true differential form of the evolution equation)

Heuristically, one may expect that the integral equation (17.6.14) can be rewritten in
the form of a distributional differential equation. (Refer to the comments concluding
Section 17.4 above.) We now see that this intuition is correct, as was shown in [Lal6,
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
Theorem 2.5,DISTRITAL
UNIVERSIDAD p. 286].FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 518 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 17.6.15 The following distributional differential equation holds on (a,b):

in the sense of distributions on the open interval (a, b). (Recall that the function 1/r
satisfies the initial condition w ( a ) = w.) Specifically, this means that for all infinitely
differentiable complex-valued functions p with compact support contained in (a, b), we
have

where p' denotes the (ordinary) derivative of the "test function" p.

Theorem 17.6.15 follows by combining Theorem 17.6.10 above and [Lal6, The-
orem 3.5, p. 305].
The interest of equation (17.6.23) is twofold:
(i) Firstly, the measure n (or rather, its continuous part u) may have a nontrivial
singular part. (See, for instance, Example 15.5.3 and Exercise 17.4.2, as well as Exercise
17.6.16 below.)
(ii) Secondly, the measure n may have a nonzero discrete part v, the action of which
(as seen, for example, in formula (17.6.22) of Theorem 17.6.14) is taken into account
by the (modified) "potential" (—z V) .
(The comments in (i) and (ii) are of interest even in the context of Sections 17.2-17.5,
where v was assumed to be finitely supported.)
Note that in the special case when n = u is a continuous measure, (17.6.23b)
becomes, in view of Remark 17.6.9(b):
under U.S. or applicable copyright law.

Exercise 17.6.16 Let n — ul be the Cantor-Lebesgue measure, as in Example 15.5.3.

Hence n is a singular continuous measure and so ( d n / d s ) = 0 l-a.e. Rewrite the inte-
gral equation (17.6.14) and the distributional differential equation (17.6.23) or formula
(17.6.24) in this case. Compare these equations with each other. Finally, check that their
solution is consistent with the result obtained in formula (15.5.15) of Example 15.5.3
(where a = 0).
[Hint: You may use (17.6.25) since n = u. Further, recall that Q (t) is given by(l 7.6.12a),
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
withu(t) = K'_i(F).]FRANCISCO JOSE DE CALDAS
UNIVERSIDAD DISTRITAL
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 519
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Unitary propagators
For a - t1 - t2 - b, we will frequently use the symbol f|^ instead of Yl[tt t2y ^
follows from results of [Lal6, §3] that ["[^ exp{— i V (s)n(ds)} is invertible in the Banach
algebra £(L2(Md)); further, its inverse, denoted by f]^ exp{-/ V(s)n(ds)}, is defined
just like the above product integral, except with a time-ordering with respect to increasing
instead of decreasing time indices as you read from left to right on the page. (See [La 16,
Lemma 3.2, p. 295] and the discussion preceding it.)
For any t1 , 12 6 [a, b], we now set

Note that, in particular, Q(t ) = Q(t , a) for t e [a, b].

Our next result follows from Theorems 17.6.10 combined with [Lal6, Theorems
3.2(c), 3.3, and 3.4, pp. 296, 299, and 304],
Theorem 17.6.17 For any t\ , ti, t2 e [a, b], we have Q(t1 , t1) = I and

Q(t3, t1) = Q(h, t2)Q(t2, ti). (17.6.27)

Moreover, for a < t\ < t2 < b,Q(t2,t1) is the "propagator" for the integral
equation (17.6.14) (i.e. tyfa) = Q (t2, t 1 ) w ( t } ) ) and is given by (17.6.15), with &'n(t)
and Ln^ as in (17.6.8) and (17.6.16) except with a, t replaced by t\, ti, respectively. [For
ti < ?2, Q(ti,t\) = Q(t\,t2)~l is the "propagator" for the "time-reversal" of'(17.6.14)
(or equivalently, of(17.6.14') below); that is, for the integral equation (17.6.14') stated
below.]
We refer the reader to [Lal6, Corollaries 3.2 and 3.4, pp. 303-304] for a more com-
plete statementofTheorem 17.6.17.There,itisshownthatforanya, B e [ a , b ] , Q(a, ft)
and its inverse, Q(B, a) = Q(a, B ) ~ l , are given by a time-ordered perturbation expan-
sion analogous to (17.6.15), and satisfy a Volterra-Stieltjes integral equation similar
to (17.6.14). In particular, for fixed ft (resp., a), Q(a, ft) satisfies a kind of "forward"
(resp., "backward") integral equation; see [Lal6, equations (3.23) and (3.24), p. 303]
with A := —iV. More precisely, for arbitrary a, ft 6 [a, b], Q(a, ft) is the unique
under U.S. or applicable copyright law.

bounded solution of each of the Volterra-Stieltjes integral equations

and

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
where, for zDISTRITAL
UNIVERSIDAD e C, o(z) is given
FRANCISCO byCALDAS
JOSE DE (17.6.6) and p (z) := p(-z) = e-2p(z).
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 520 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We note that given Theorem 17.6.10, the remaining results (Theorems 17.6.12,
17.6.14,17.6.15 and 17.6.17) follow easily from the corresponding results about abstract
product integrals established in [La 16, §3].
Remark 17.6.18 It follows from Remark 17.6.9(b) that for any a, B in [a, b], Q(a, B)
is a unitary operator when u, = l and the potential V and the measure v are both real-
valued, with n = u + v as before. (For a - ft, this is obvious, while for a > B, we simply
use the fact that Q ( a , ft) is the inverse of the unitary operator Q(B, a).) By abuse of
language, we still refer to Q(a, ft) as a "unitary propagator" even when this assumption
is not satisfied.

Scattering matrix and improper product integral

In order to state our last result, we shall need to slightly modify our hypotheses. Assume
that n is a (complex-valued) set-function defined on (the bounded Borel subsets of) R
and whose restriction to every compact subinterval [a, b] c R lies in M([a, b]) (the
set of complex Borel measures on [a, b]). (See Remark 17.6.19 below.) Suppose further
that V : R x Rd -> C is a Borel measurable function which belongs to Looj^R) (i.e.
b\\V(s,-)\\00\rl\(ds)<oo).
Remark 17.6.19 (a) Since we are now working on all of R instead of the fixed compact
interval [a, b], we cannot assume that n is a Borel complex measure on R. Otherwise,
n would have to be of bounded variation on all of R, which would exclude even the
traditional case when r\ = l, Lebesgue measure on R = (—00, +00).
(b) The assumption made above about n is that it is a local complex measure on R,
in the sense of [DoFr3, Definition 6.1, p. 179]. (In particular, if n takes its values in
[0, +00], then it is a standard positive Borel measure on R which is finite on bounded
Borel subsets of R.)
In view of (17.6.12a), Theorem 17.6.20 ([Lal6, Theorem 2.7, p. 288]) follows from
[Lal6, Theorem 3.6, p. 306]. (For a precise definition of the product integral appearing
in (17.6.28) below, we refer to [Lal6, §3.E] and, in the special case of either finite-
dimensional matrices or n = /, to [DoFri3, §5.5 or §1.6].)
Theorem 17.6.20 (Existence of the scattering operator defined via a product
integral) Under the above assumptions, the strong improper product integral
Y[t™exp{—iV(s)n(ds)} exists and is an invertible element of £(L2(Rd)). Moreover,
under U.S. or applicable copyright law.

we have

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 521
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where u(b) = Kb_t (F) is given by (17.1.6), with F as in (17.1.7). The limits in (17.6.28)
hold in the strong operator topology.
If r) = I, then S is often called the "scattering operator" or "S'-matrix" and is denoted
by

in the physics literature. (See [Sa, p. 5] and Remark 17.6.1 l(a) above.) Therefore, we
may think of Theorem 17.6.20 as establishing the existence of the scattering operator
in our context. When v is finitely supported, this is of particular interest in view of
Theorem 17.2.7 [see equation (17,5.8) with t := b, which provides (in the quantum-
mechanical case) an "explicit expression" for u(b) as a function of b (and a) in R] and
of the physical interpretations in terms of multiple scatterings provided in Section 17.5
above [seethe discussion following equation (17.5.8), which also deals with the quantum-
mechanical case].
We close our exposition of the results of [La 16, 18] by giving a natural sufficient
condition under which S is unitary. (We write /z instead of the continuous part of the
restriction of n to an arbitrary compact interval [a, b] C R.)
Corollary 17.6.21 Suppose further that u = l and that v and V are real-valued but
otherwise arbitrary. Then S, as defined by (17.6.28), is a unitary operator.
Exercise 17.6.22 Deduce Corollary 17.6.21 from Theorem 17.6.20and Remark 17.6.18.
[Hint: With the obvious notation, it follows from the comment preceding equation

Sketch of the proof of the integral equation

We conclude this section by providing a sketch of the proof of the integral equation
(17.6.2) stated in Theorem 17.6.1 ([Lal8, Theorem 4.3, p. 185]) for an arbitrary Borel
measure n on [a, b].
We will closely follow the outline of the proof given in [La 18], and (except for the
main part of the proof of Step 1) we will refer to [Lai 8] for most of the details. More
under U.S. or applicable copyright law.

precisely, in Step 1 below, we assume that v, the discrete part of r;, is finitely supported
(as in Section 17.2) and then reformulate the integral equation obtained in Section 17.2
(Theorem 17.2.1, recalled in (17.6.29) below) in the more invariant form of (17.6.2)
(or (17.6.30) below). (Compare equation (17.6.29) with equation (17.6.30), which no
longer depends on the specific definition of v.) Then we treat the case of an arbitrary
Borel measure r\ = u + v in Steps 2 and 3. In Step 2, we use in a crucial way the
time-ordered perturbation expansions (GDS) derived in Chapter 15 to approximate the
operator u(t) = K((F) associated with 77 = u. + v by operators um(t) = K[(Fm)
associated with the measures r)m = u + vm (m — 1, 2, . . . ), with vm finitely supported.
Finally, in Step 3, we use Step 1 along with the approximation theorem from Step 2 to
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
deduce that uDISTRITAL
UNIVERSIDAD = u(t) FRANCISCO
satisfiesJOSE
the DE
integral
CALDAS equation (17.6.2), as desired.
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 522 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We now proceed to give a more detailed outline of the proof of Theorem 17.6.1.
Step 1: Reformulation of the integral equation when v is finitely supported.
Fix A 6 C+. Assume that v is finitely supported; namely, v = £p=i oP5rp, with
(Op e C and TO := a < T\ < • • • < rm < b =: tm+\ (exactly as in (17.2.1), except with
the subscript h replaced by m). Then by Theorem 17.2.1, u = u(t) satisfies the integral
equation (17.2.2) on each interval (TP, TP+\]; namely, for p = 0 , . . . , m,

The next result ([Lal8, Theorem 3.2, p. 173]) shows that u satisfies the integral
equation (17.6.2) on the entire interval [a, b]. (Some of the necessary technical facts will
be provided in Lemma 17.6.24 below.)
Theorem 17.6.23 (Finitely supported v) Under the above assumptions,

Proof of Theorem 17.6.23 . Fix t e [a, b] and A e C+. We will show that u(t) is given
by the right-hand side of (17.6.30). If t = a, this is obvious since u(a) = I. We now
assume that t e (a, b]; let p be the unique integer in {0,... , m} such that t e (rp, TP+i ].
According to Theorem 17.2.1 (see equation (17.6.29) above),

Hence, since u is continuous and since in shorthand notation,

under U.S. or applicable copyright law.

where a = 1 if n(a) = 0 and a = 2 otherwise (recall from Remark 17.2.5 that &>o = (a\
and TO = TI = a if n(a) ^ 0), we have:

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE 77) 523
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

By inserting a telescoping sum, we obtain

[Here, we have again used the convention specified in Remark 17.2.5 and the fact that
W(TO) = u(d) = I to deduce that (for a — 1,2)

Now, by Theorem 17.2.1—more precisely, by equation (17.2.2) or (17.6.31) applied

to Tq e (Tq_I , T?] for # e {1,... , m]—we see that

Thus, by [HilPh, Theorem 3.7.12, p. 83] and by the semigroup property of

>o, it follows that for q e {1,... , p],
under U.S. or applicable copyright law.

Combining (17.6.32) and (17.6.33), we obtain

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 524 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

By Lemma 17.6.24(a) below,

0w,n(5) = 6(s) |u|-a.e. in (a, t), (17.6.35)

and since for q e {0, . . . , m), n(T?) = a>q, it follows from formula (17.6.38) in
Lemma 17.6.24(b) that

wq0<p.u(r 9 ) = ^0(Tq} ~ I (17.6.36)

and hence, in particular,

Consequently, in view of Remark 17.2.5, equation (17.6.34) becomes successively:

This proves equation (17.6.30), as desired.

Let 9 € Looi;i? and define as before Qv<n by (17.6.5) via the holomorphic functional
calculus in £(L 2 (R d )), with n(s) := n(M) and (p given by (17.6.6). [Strictly speaking,
since 6 e Loo1;n and by (17.1.3), dw,n(s) is only defined (as an element of £(L2(Rd))
for |n|-a.e. s in [a, b]; in particular, 9w,n(s) is well defined for every "pure point" s in
[a, b) (that is, every s such that r](s) = 0) and for |u|-a.e. s in [a, b), where u denotes
the continuous part of 77.]
under U.S. or applicable copyright law.

Throughout Section 17.6 and, in particular, in the proof of Theorem 17.6.23 above,
we have used (most often implicitly) the following simple result [La18, Lemma 2.1, p.
170] (compare with [DoFri3, pp. 159-161]):
Lemma 17.6.24 (a)If s is not a pure point of n) (i.e. if n ) ( s ) = 0), then

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE 17) 525
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(c) Under the above hypotheses. 0n,n e loo;n and

Exercise 17.6.25 Prove Lemma 17.6.24.

Exercise 17.6.26 Establish the converse of Theorem 17.6.23; namely (still for v finitely
supported), show that if u satisfies (17.6.30) on [a, b], then it also satisfies (17.6.29) on
(Tp,Tp+\]forp = 0 , . . . , m.
[Hint: The proof—which can be found in [Lal8, Proposition 5 1, pp. 198-199]—is
shorter than that of Theorem 17.6.23 and makes use of Lemma 17.6.24.]
Step 2: Approximation by measures with finitely supported discrete part.
Let n = u + v be an arbitrary Borel measure on [a, b], with v = ELi w^-ip an^
Xi^Li \o>p\ < °o (as in (17.6.1)). Without loss of generality, we may assume that the
times tp are pairwise distinct (but we may not assume that they are linearly ordered).
For m > 1, set nm = u + vm, where vm := £)p=i w p & r p . (Note that the measures 7?m
and n have the same continuous part u.)
Let F be defined by (17.1.5), and let Fm be defined similarly, except with n replaced
by nm. Given t € [a, b) and A. e C+, let u(t) = K((F) (as in (17.1.6)) and let um(t) =
K [ ( F m ) f o r m > 1.
The following result ([Lai8, Theorem 4.2, p. 180]) provides a strengthening in this
specific context of some of the stability theorems (from [JoLa 1 ]) with respect to measures
discussed in Chapter 16 (Section 16.2). (See Remark 17.6.29(a) below.)
Theorem 17.6.27 (Stability theorem: approximation ofv by vm) With the above nota-
tion, we have:

so that, in particular, {um}^_^ converges to u in (the norm operator topology of)

under U.S. or applicable copyright law.

£(L2(E.d)), uniformly in t on all of [a, b].

We next introduce some cumbersome but convenient notation for use in the solution
of Problem 17.6.28 below.
Given t e [a, b], we let (for p > 1)

so that the restriction of r\ to [a, t) is equal to the measure n, := u + Yl^Li Op.t&Tp

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
(where u, nowDISTRITAL
UNIVERSIDAD denotesFRANCISCO
the restriction of the continuous part of r] to [a, t)).
JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 526 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

so that the restriction of r)m to [a, t) is equal to the measure n m > , := /j, + £!^Li <dp<fiip •
(Of course, u>'p_, afao depends on m but we do not indicate it explicitly, for notational
simplicity. Further, note that even though only finitely many of the weights co'p_, are
nonzero, we will not need here to reorder them linearly.)
The following problem—the solution of which can be found in [La 18], Section 4. A
(for part (a)) and Section 4.B (for parts (b) and especially (c))—is not really difficult but
involves somewhat lengthy computations (particularly for part (c)).
Problem 17.6.28 (Proof of Theorem 17.6.27) (a) Write down a suitable generalized
Dyson series (CDS, as discussed in Chapter 15) for the operator u(t) = K[(F).
[Hint: Use formula (17.6.42) above. According to Example 15.5.7 and Theorem 15.7.6,
the natural GDS for u(t) in this context is given by the "time-reversal" (in the sense
of Theorem 15.7.6(ii)) of the GDS in (15.5.25); see [Lal8], equations (4.3) and (4.4),
pages 176-177.]
(b) Write down a GDS for um (t) — K[(Fm), much as in part (a).
[Advice: Use formula (17.6.43) and specialize the result of part (a) to this situation; do
not use the much simpler looking GDS obtained by taking the "time reversal" of the GDS
in formula (15.5.5) of Example 15.5.1 (where the discrete part of the measure involved
was also assumed to have finite support).]
(c) Use parts (a) and (b) to derive an estimate of the desired form:

where cm —> 0 as m —> oo. This proves Theorem 17.6.27.

[Note: More precisely, the positive number cm in (17.6.44) is explicitly computable and
can be chosen to be the tail of a specific convergent series; see [Lai 8], equation (4.22),
page 183.]
Remark 17.6.29 (a) In Example 16.2.11, we used our earlier stability Theorem (The-
orem 16.2.6) to obtain a conclusion analogous to that of Theorem 17.6.27. We point
under U.S. or applicable copyright law.

out, however, a significant difference with the present result: Estimate (16.2.45)—the
counterpart in Example 16.2.11 of estimate (17.6.41) or (17.6.44)—was valid only for
A > 0 (rather than for all A. e C+, as in Theorem 17.6.27). This is so because the proof
of Theorem 16.2.6 relied on the dominated convergence theorem for functional integrals
whereas that of Theorem 17.6.27 makes use of our generalized Dyson series, which are
well defined and have the sameform for all A. € C+. This provides one more illustration of
the usefulness of these time-ordered perturbation expansions (GDSs) to establish results
which are valid in the diffusion as well as in the quantum-mechanical case. [See also,
e.g.,EBSCO
Example 16.2.9 or the proof of the integral equation (Theorem 17.2.1) discussed in
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
Section 17.3.]DISTRITAL FRANCISCO JOSE DE CALDAS
UNIVERSIDAD
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 527
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(b) It is the uniformity in t e [a,b] of the limit in (17.6.41) that permits us to

interchange the limits involved in a variety of situations and to deduce results for general
Borel measures/mm the corresponding onesfor measures with finitely supported discrete
part. (See, for example, part (ii) of Exercise 17.6.5(a) as well as Step 3 below.)
Step 3: Passage to the limit in the integral equation.
Fix A. e C+. With the notation and hypotheses of Step 2, we know from Step 1
(Theorem 17.6.23) that for each fixed m > 1 , um = um(t) satisfies the integral equation
(17.6.30) on [a, b]; namely, for all t e [a, b],

The present step consists in justifying the passage to the limit (as m —> oo) in (17.6.45);
namely, we show that, for all t e [a, b],

Fix t e [a, b]. Clearly, the left-hand side of (17.6.45) converges to that of (17.6.46),
according to the stability result from Step 2 (Theorem 17.6.27). Moreover, by means
of Lemma 17.6.24 and Theorem 17.6.27 (including the uniformity of the convergence
of um(s) to u(s) for s e [a, b], as stressed in Remark 17.6.29(b)), we deduce after
somewhat lengthy computations that the right-hand side of (17.6.45) also converges
to that of (17.6.46). This proves that the integral equation (17.6.46) (or equivalently,
(17.6.2)) holds for all t e [a, b] and thus completes the proof of Theorem 17.6.1, as
required.
The following problem asks the reader to provide the missing details in the above
argument. (The full proof can be found in [Lal8, §4.C, pp. 185-191].)
Problem 17.6.30 Complete the proof of Step 3.
[Hint: It should be useful, in particular, to note that &<f,ri(s} — Q<f,rim (s) is equal to Ofor
s € [a, b]\{Tm+[,...} and is given by a simple expression for s = TP, with p > m + l.]
Our final problem is rather open-ended and, especially for parts (b) and (c), may not
have a simple answer. Indeed, a satisfactory (interpretation and) resolution of parts (b)
under U.S. or applicable copyright law.

and (c) may well lead to new results (not presently available in this form in either the
probabilistic or mathematical physics literature), with interesting physical applications,
as suggested by the discussion provided towards the end of Section 17.5 above. [(The
second author wishes to thank D. Kannan for several conversations related to part (b).)
We note that the more recent works in [ChunZh] or [benACas] (and the appropriate
references therein) may be relevant to part (b) or (c). The paper [JoKalS]—which extends
some of the results of Chapter 19 [dFJoLa2] below to a stochastic setting—may also be
useful in this context.]
Problem 17.6.31 (a)In the diffusion case (i.e. for X > Qand9 = -V.withV e Looi;r;J>
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
findUNIVERSIDAD
(under suitable
DISTRITAL assumptions)
FRANCISCO JOSE DEaCALDAS
direct probabilistic proof of the integral equation
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 528 FEYNMAN-KAC FORMULAS WITH STIELTJES MEASURES
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(17.6.2). Then (possibly under a different set of hypotheses), deduce in two steps that
(17.6.2) holds for all A, e C+; namely, by analytic continuation (A, e C+) followed by
passage to the limit along the imaginary axis (A. e C+).
[Advice: For simplicity, you may wish at first to assume that v is finitely supported.
Further, only when A, > 0, you may look for "optimal" hypotheses on the potential V
under which the FKLS holds.]
(b) (Stochastic version of the FKLS?) Suppose again that we are in the diffusion
case (say A. = 1 and 0 — —V). Try to generalize the "Feynman-Kac formula with
a Lebesgue-Stieltjes measure" (FKLS) to the case when the measure n (as well as,
possibly, the potential V = V(s, •)) is allowed to be "random" (hence, for instance,
the atoms of n are no longer located at predetermined times TP, but at "random " times,
which can be interpreted probabilistically by using the notion of "stopping time"). That
is, obtain a suitable stochastic version of th- integral equation (17.6.2) (or (17.6.46)).
(c) Can you extend to the quantum-mechanical case (when A. ^ 0 is purely imaginary
and 9 = —iV) the probabilistic results which you have obtained in part (b)? More
specifically, can you find—by means, for example, of a suitable analytic continuation
procedure followed by a passage to the limit—a counterpart in the quantum-mechanical
case of the stochastic (integral) equation obtained in part (b) in the probabilistic case?
Under which assumptions does your extension hold?
[Advice: In part (c), you may have to restrict considerably your hypotheses on the
potential V.]

Remark 17.6.32 (a) We note that a stochastic version of the FKLS in the diffusion (part
(b)) or, especially, in the quantum-mechanical case (part (c)) would broaden the scope of
the physical interpretations and possible applications discussed in Section 17.5 above.
For instance, in the quantum-mechanical case, it might improve the simplified model
of "multiple scattering" (within a target) discussed just before equation (17.5.9), by
allowing the impacts (or scatterings) of the particles to occur at random (rather than
at fixed predetermined) times. (For example, the corresponding probability distribution
could be that associated with a Poisson process or with a more complicated "jump
process".)
(b) In part (a) of this remark, we have discussed a possible "randomization" of
the discrete part of the measure. In some sense, recent work [JoKalS] of Johnson and
Kallianpur on Stochastic Dyson series and the solution to stochastic evolution equations
under U.S. or applicable copyright law.

can be thought of as providing a suitable "randomization" of the continuous part of the

measure; namely, heuristically, u,(ds) is replaced by %(s)ds, where t-(s) is "white noise"
(roughly speaking, the "derivative" of Brownian motion).
(c) Another type of extension of some of the results of this chapter would consist
in replacing Brownian motion by a more general (Markov) stochastic process; that
is, in allowing HQ to denote the generator of (the heat semigroup associated with) a
Markov process other than the Wiener process. (See Remark 14.4. l(c).) For instance,
for d = 1 and under suitable assumptions, the FKLS of Theorem 17.2.1 was extended
by Riggs [Rig] to the case when HQ denotes the generator of the Poisson process (as in
EBSCO Publishing
Example : eBook in
9.2.5). Then, Academic Collection (EBSCOhost)
the (analogue - printedor
of the) diffusion on quantum-mechanical
6/8/2017 5:49 PM via case,
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE GENERAL CASE (ARBITRARY MEASURE n) 529
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

respectively, we obtain the (generalized) telegrapher's equation or the Dirac equation

(both in one space dimension) rather than the (generalized) heat or Schrodinger equa-
tion, respectively. In this context, the (counterpart of the) Dirac equation describes the
time evolution of a relativistic quantum-mechanical particle subjected to "shocks " or
"scatterings " occurring at deterministic times. Naturally, in the spirit of Problem 17.6.31
and of part (a) of this remark, one may still wonder what happens if those times are ran-
domized. We leave the investigation of these and related questions to the interested
reader.
under U.S. or applicable copyright law.

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

18
NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS,
DISENTANGLING ALGEBRAS AND FEYNMAN'S
OPERATIONAL CALCULUS

18.1 Introduction
A family {At '• t > 0} of commutative Banach algebras of functionals on Wiener space
was introduced in Chapter 15, and it was shown (see Theorem 15.7.1) that, for every
F e At, the functional integrals K[ (F) exist and are given by a time-ordered perturbation
expansion which serves to disentangle, in the sense of Feynman's operational calculus,
the operator K[(F).
In this chapter, noncommutative operations * and + on Wiener functionals are defined
which will enable us to provide a precise and rigorous interpretation of certain aspects
of Feynman's operational calculus for noncommuting operators. The noncommutative
operations, the "disentangling algebras" {At '• t > 0} and the functional integrals provide
a rich interlocking structure.
The present chapter is based on the joint paper [JoLa4] of Johnson and Lapidus
(announced in part in [JoLa3]). The paper [JoLa4] in turn relied heavily on [JoLal],
which was the main subject of Chapter 15.
The last part of Section 14.5 discussed Chapter 18 and can serve as part of the
introduction to this chapter. For the convenience of the reader, however, we recall some
of the most important elements and make a few additional comments.
In Section 14.5, we discussed the fact that if F 6 Atl and G 6 At2, then both
F * G and F+G are in Att+t2. Now that we have defined the norms || • ||(, t > 0 (see
Section 15.7), we can state associated norm inequalities (see Theorem 18.5.3):
under U.S. or applicable copyright law.

Recall from Section 15.7 that for a fixed t > 0, the Banach algebra (At, II • ||/) is
a suitable set of (equivalence classes of) functions [often called "Wiener functionals"]
defined on C' - C([0, t], Rd), the space of continuous paths x : [0, t] - Rd. (Also,
At is equipped with the natural pointwise addition and multiplication.)
Further,
EBSCO given: FeBook
Publishing e AAcademic
tl and Collection
G e At-,(EBSCOhost)
with t1, t2- > 0, Fon* 6/8/2017
printed G e At15:49
+t2PMis via
defined by
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 INTRODUCTION 531
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

where the paths x\ e C'1 and X2 € C'2 are given by

and

Similarly, with x\ and xi as in (18.1.4), F+G £ A,+r 2 1 S defined by

We leave aside for now the measure-theoretical technicalities associated with the
definition of the algebras At (due mainly to the properties of scale-change in Wiener
space), and hence of the operations * and +. We refer the interested reader to the earlier
material in Sections 15.2.1 and 15.7. The precise definitions of the "noncommutative
multiplication" * and the "noncommutative addition" + will be given in Section 18.2
below.
Our central result (Theorem 18.5.6)—established below in the broader context of the
analytic operator-valued Wiener and Feynman integrals (see Theorem 18.4.1)—can be
formulated as follows in the present setting of the disentangling algebras {A, : t > 0}.
If F 6 At} and G e At2, then for all A e C+, we have the operator identity (in
r i 1 2/in></\\.
L\L (IK )).

[Observe that if we replace F * G by G * F on the left-hand side of (18.1.6), then the

order of the operators on the right-hand side must be reversed.]
We deduce in particular from (18.1.6) that the operator product K^(F)K^(G) can
be disentangled via a generalized Dyson series (GDS) because (by Theorem 15.7.1) the
operator K'^+'2 (F * G) itself can be disentangled in this manner. (See Corollary 18.5.7.)
This answers a very natural question motivated in part by our earlier work presented in
Chapter 15 ([JoLal]) and in Chapter 17 ([Lal4-18]). (See, in particular, the comments
at the end of Sections 15.7 and 17.5 above.)
under U.S. or applicable copyright law.

[Note that although the time parameter t was kept fixed in Chapter 15, we now allow /
to vary in Chapter 18, much as in the study of the evolution equations in Chapter 17. This
fact is crucial in interpreting and deriving formula (18.1.6) and its corollaries discussed
below.]
There are a variety of equalities relating * and + which have corresponding operator
equalities. Formulas (14.5.11) and (14.5.12) are one such pair. More generally (see
Example 18.5.14), given arbitrary tj > 0 and Fj e Atj (for j = 1 , . . . , h), we have the
identity between elements of Atl+...+th'•

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 532 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

and os, by (18.1.6) above (theorem 18.5.6),

[We stress that in (18.1.8), for instance, exp(FH ----- i-F/,) is defined by means of
the analytic functional calculus in the Banach algebra At1-\_____\-th', further, for j =
1, ...,/!, exp(F;-) is similarly defined within the Banach algebra Atj.]
Under the same assumptions as for (18.1.7) and (18.1.8), another example is given
by a type of multinomial formula and a related operator equation (see (18.5.35) and
(18.5.36)):

and hence

Finally, it makes sense to consider the commutator [F, G] := F * G - G * F of

F € Atl and G 6 At2 and to ask if the functional integrals K[ preserve commutators.
They do (see Corollary 18.5.8), and we have

The rest of Chapter 18 is organized as follows. The main part of the chapter (Sections
18.1-18.5) deals with the authors' work ([JoLa3,4]) discussed above. Moreover, in the
appendix (Section 18.6), we discuss some work of the second author ([Lai 9]) in which
was proposed an axiomatic approach to Feynman's operational calculus, motivated in
large part by the results of [JoLa4].

18.2 Preliminaries: maps, measures and measurability

Throughout this chapter, we will need various maps between spaces of continuous func-
tions and related results involving measures and measurability. Proposition 18.2.1, which
under U.S. or applicable copyright law.

concludes this section, will help us to show in Section 3 that the operations * and + are
well defined. At least on first reading, one may wish to skip the proof of Proposition
18.2.1.
Let Ca<b denote the space of all Revalued continuous functions on the interval [a, b].
Let CQ' denote the associated Wiener space; that is,

In the following, ma,b will denote Wiener measure on CQ' & . Frequently, we will have
EBSCO Publishing : eBook Academicb Collection (EBSCOhost) - printed!bon 6/8/2017 5:49 PM via
a =UNIVERSIDAD
0 and then we willFRANCISCO
DISTRITAL write CJOSE
, CQDE and m^ rather than C° , CQ 'b and mo, b, respectively.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 PRELIMINARIES: MAPS, MEASURES AND MEASURABILITY 533
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

In particular, with this notation, we have m, = m, Wiener measure on C'0 = C0'r, as

defined in Chapter 3. (Caution: Throughout the present chapter, mt is defined as above
and hence does not denote scaled Wiener measure, as defined in Chapter 4.)
We begin by considering two restriction maps and a translation map all of which will
be involved in the definitions of the operations * and + • Suppose that a < b < c and
let R\ : Ca'c ->• Ca'b be the map of restriction to the first part of the interval. If we want
to emphasize that R\ depends on a, b and c, we will write R"' 'c rather than simply 7?i.
So

Similarly, /?2 : Ca'c -> C b,c will denote the map of restriction to the second part of the
interval; that is,

In this chapter, we will need certain variations on the maps and results which we
are describing. When these differ only slightly from items being discussed here, we will
usually treat them briefly. For example, if [a, b] is partitioned into n subintervals, Rj
will denote the map which restricts x in Ca-b to the jth subinterval, j = 1, 2 , . . . , n.
Let T : Ca'b ->• Cb~a be the translation map

The restriction TO of T to CQ ' has range CQ a. Further, the stationarity of the increments
of the Wiener process (see Proposition 3.3.17) yields

Equation (18.2.5) along with (18.2.7) below will be instrumental in the proof of formula
(18.1.6); the latter equation will be used in turn in the proof of formula (18.1.8).
There are three bijective maps, p\, p2, p3, onto product spaces which we will find
useful. Given a < b < c, the map pi : CQ'C -> CQ' x C0'c is defined by

Since, by Proposition 3.3.18, the Wiener process has independent increments, we have
under U.S. or applicable copyright law.

We will often regard p\, pi and p$ as identifying the spaces involved. For example, given
x in CQ'C we will often write (y, z) in place of x, where y = R\x and z = RZX — x(b).
It is then natural to write m a]C = ma,b x Tnj,,c rather than (18.2.7).
The map p^ : Ca'b ->• Rd x C^b is defined by

We will frequently think of R4 x CQ 'b or, under the "identification" pi, Ca-b, as equipped
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:49 PM via
withUNIVERSIDAD
the measure Leb.FRANCISCO
DISTRITAL xm a ,ft, JOSE
where as before, Leb. denotes Lebesgue measure on Rd.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 534 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Finally, Given a < b < c, the map p3: is defined by

We will sometimes think of Rd x Ca,b x CQ'C or, under the "identification" p3, Ca-c,
as equipped with the measure Leb.xm fli £ x rrifoiC. Given (£, y, z) in W* x CQ' x C0'c,
we will often write x = (£,y, z) rather than the more precisely correct equality x =
pj1 (£, y, z). Similarly, given (£, y) in E.d x Ca b ', we will often write x = (£, y) rather
than* = /^~ 1 (£,y).
The spaces of continuous functions above are equipped with the sup norm topology.
Under these topologies, R\ and R2 are continuous maps and T, p\, p2, and p^ are all
homeomorphisms.
Let F and F1 be C-valued Borel measurable functions on the space Ca'b which is
identified, via the map p2 above, with Rd x CQ'*. Experience with standard measure-
theoretic settings suggests that we require functions to be defined Leb. x ma,t-a.e. and that
we regard F and F\ as equivalent provided that F(*+£) = Fi(jt+f)forLeb.xm a ,fo-a.e.
(|, x) in Rd x CQ'*. However, due to the pathology of Wiener measure under change
of scale (see Chapter 4) and the fact that a continuum of positive scalings enter into
the definition of the analytic operator-valued Feynman integral (Definition 15.2.1), we
need as in several places before to use more refined equivalence classes. (See Sections
14.2-14.4 and especially Sections 15.2.I and 15.7.) We will say that F is equivalent to
F1, and write F ~ F1, provided that, for every p > 0, F(px + £) = F\(px + £) (or
F(£ + px) = FI(£, px)) for Leb. x ma,b-a.e. (£, x) in Rd x C^b.
We conclude this section with a rather technical proposition which will be needed as
we continue; in particular, it will enable us to show that the operations * and +, which
will be introduced in the next section, are well defined when thought of as acting on the
appropriate equivalence classes of functions.
Proposition 18.2.1 Let a < b < c.
(i) Suppose F, F\ : Ca'b —> C are Borel measurable and that F ~ FI. Then
F o RI, F, o #1 : Ca-c ->- C and F o R\ ~ FI o RI. Similarly, if H, H\ : Cb-c ->• C are
Borel measurable and H ~ H\,thenHoR2, H\oR2 : Ca'c -»• CandHoR2 ~ H\oR2.
(ii) Suppose G, G\ : Cc~b ->• C are Borel measurable and that G ~ G\. Then
G o Tb~c o R2, G\ o Tb'c o R2 : Ca-c -^C and Go Tb>c oR2~G\o Tb*c o R2.
under U.S. or applicable copyright law.

Proof (i) Given x = (|, y, z) € C a ' c = Rd x CQ'* x Cb,c R1x = R\($,y,z) =

(£,y) e R d xCo' f c = C a 'b.Letp 0 befixed.By assumption,F(py+§) = F{(py+%)
or F(|, py) = Fi(f, py) for Leb.xma,fc-a.e. (§, y). Then (F o /?!)(£, py, pz) = (F1 o
R)(t, py, pz); that is, F(£, py) = F,(|, py) for Leb.xma,b-a.e. (§, y) e Rd x C^'fo
and every z 6 CQ' C . Certainly then (F o fli)(£, py, pz) = (Fj o R\)(%, py, pz) for
Leb.xma,fe x mfc,c-a.e. (|, y, z) 6 Rd x CQ'* x CQ'C.
In the proof just given, after R\ is applied to (£, py, pz), pz is no longer present.
TheEBSCO
corresponding statement is not quite true when R2 is oninvolved
Publishing : eBook Academic Collection (EBSCOhost) - printed
and this makes the
6/8/2017 5:49 PM via
nextUNIVERSIDAD
proof more complicated.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE NONCOMMUTATIVE OPERATIONS * AND + 535
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Given x = (£, y, z) s Ca'c = Rd x Ca,b x C0b'c, R2* = R2(£, y, z) = (I +

Y(b), z) eR d xC0' c = Cb, c .Let/o > 0 befixed.Since H ~ Hi,H($,pz) = H1(£, pz)
for Leb.xm&iC-a.e. (£, z) e Rd x C0b'C. Let M C Rd x Cg'c be the Leb.xmb,c,-null
exceptional set; that is, M = {(E, z) e Rd x Cp'e : H(f, pz) #,(£, pz)}. We wish
to show that (H o R2) Ay, pz) = (H1 ° R2)(f, Py, pz), that is, H(£ + py(b), pz) =
H1(? + py(b), pz) for Leb.xma,b x mb,c-a.e. (£, y, z) e Rd x Ca,b x Cb'C. In fact,
we will show more. We will prove that for every y e CQ' , H(£ + py(b), pz) =
H1(f + py(b), pz) for Leb.xmb,c-a.e. (c, z) € Rd x Cp'c.
Fix y 6 Ca,b. To complete the proof, we will show that M' := {(c, z) e Rd x Cg'c :
(f + Py(b), z) € M) is a Leb.xmfo>c-null set. Since .M is Leb-xm^^-null, the section
M.(z) is Leb.-null formb,c-a.e.z € CQ'C. Take any z such that M(z) is Leb. null. Sectioning
M' at z, we obtain

But ./M^' is null and Lebesgue measure is translation invariant on Rd and so (M')^ =
M{z) - py(b) is null. Since (M')(z) is Leb.-null for mb,c-a.e. z e Cb'c, M' is null as
we wished to show.
(ii) Let T = Tb'c : Cb<c -»• Cc~b be the translation map as in (18.2.4) (except that
the interval is different). Of course, we can think of T as follows; T : Rd x CQ'C ->
R.d x C^~b. Note that T(£, z) = (£, T0z) = (te, T0z),where t denotes the identity map
on Rd and T0 is the restriction of T to CQ'C. Hence

(Leb. x m b,c ) o T-1 = (Leb. o r1) x (mb,c o T - 1 ) = Leb. x m c _ fe , (18.2.10)

where the last equality follows from (18.2.5).
Because of the second part of (i), we can establish (ii) by showing that G o T ~ G1 o T.
Let p > 0 be fixed. We wish to show that (G o T)(%, pz) = (G| o T)(E, pz) or
G(£,p7bz) = Gi(f,p7bz) for Leb.xmi>c-a.e. (^,z) 6 Rrf x CQ' C . Since G ~ G1,
under U.S. or applicable copyright law.

G(E, Px) = GI(§, pjr) except for a Leb.xmc_6-null exceptional set M. Thus .M =
{(£,*) e Erf x Co~b : G($,px) ± GI(|,PJC)} is Leb. x m c _ fe -null. Let M' :=
{(I, z) € Rd x CQ'C : (£, T0z) e >f}. Note that X' = T- 1(M) . From this, (18.2.10),
and the fact that M. is Leb. x mc_(,-null we can write (Leb. x mb,c)(M) = (Leb. x
m b , C ) ( T - l ( M ) ) = (Leb. x mc_b)(M) = 0. It follows that G(E, pTbz) = G1(E, pF0z)
for Leb. x mb,C-a.e. (§, z) e Rd x Cb,C.

18.3 The noncommutative operations * and +

In this section, we introduce the noncommutative multiplication * and the noncommu-
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
tative additionDISTRITAL
UNIVERSIDAD + andFRANCISCO
presentJOSE
some of their algebraic properties.
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 536 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Throughout the rest of this chapter, we will assume, usually without explicit mention,
that thefunctions introduced are Borel measurable. (Actually, except for Theorem 18.3.2,
the results of the present section can be thought of purely algebraically.) Also, as we
continue, t , t 1 , t 2 , . . . will denote positive real numbers.
Let F and G be functions on C'1 and C'2, respectively. Both F * G and F + G are
to act on Ct1+t2. Given x e Ct1+t2, we define x1 e Ct1 and x2 € Ct2 by

and

Note that x1 is the restriction of x to [0, t1 ] and X2 is the restriction of x to [t1,t\ +

t2] followed by translation to [0, t2]. In terms of the brief form of the notation from
Section 18.2,

We now define F * G and F + G as C-valued functions on C'l+'2 via the formulas

and

Alternatively, we can write formulas (18.3.2) and (18.3.3) as

and

Remark 18.3.1 (a) In the definitions of * and +, one can begin to see connections with
Feynman 's time-ordering ideas [Fey8] discussed in Chapter 14. Both equations (18.3.2)
under U.S. or applicable copyright law.

and (18.3.3) involve time-ordering. Indeed, note that F(x1) depends on the part of the
path x over the interval [0, t\], whereas G(x2) depends on the part of x over the later
time interval [t1, t1 + t2].
(b) It may well be natural for some purposes (such as the study of an evolving physical
system) to work with the abutting intervals [0,t1 ] and [t1, t1 + t2] rather than with [0,t1]
and [0, t2\. If this alternative were followed, given x 6 Ct1l+t2, we would define x\ as
before but would take x2 := R2X as there would be no need for the translation T. Given
functions F and G on Ct1 and Ct1,t1+t2,respectively, we would define F * G and F + G
by (18.3.2) and (18.3.3) as before except that x2 would now be simpler. We will not
pursue this line of thought in the present work even though it would be easy to make the
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
necessary adjustments
UNIVERSIDAD throughout
DISTRITAL FRANCISCO JOSE DEthe rest of the chapter.
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE NONCOMMUTATIVE OPERATIONS * AND + 537
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

We will need to work with equivalence classes of functions rather than with the
functions themselves. Our first result shows that the equivalence relation ~, introduced
in Definition 15.2.3 and discussed further in Section 18.2, is compatible with the oper-
ations * and +. Once we have this result in hand, we will usually follow the standard
measure-theoretic convention and blur the distinction between equivalence classes and
their representatives.
Theorem 18.3.2 Let F, F] : Ct1 -> C and let G, G1 : Ct2 -» C. Further suppose
that F ~ F1 and G ~ G1. Then

Proof Proposition 18.2.1 assures us that F o R1 ~ FI o R1 and that G o T o R2 ~

GI o T o R2. The result now follows from the compatibility of ~ with the usual product
and sum (see Section 15.7) and from the definition of * and + as Equations (18.3.2')
and (18.3.3') make particularly clear. D
It is rather evident that * and + are noncommutative operations (even if t\ = t2);
however, these operations do have a variety of pleasant algebraic properties as the next
three theorems show.
Theorem 18.3.3 (Algebraic properties of*)
(1) Let F, FI : C" -> C and let G, G1 : Ct2 -> C; further, let a, B e C. Then

and

(2) Let F : C"1 -» C, G : C'2 -+ C, and H : C'3 -> C. Then (F * G) * H and

F * (G * //) are C-valuedjunctions on C'l+'2+'3 and

(3) Lef 1 fee the Junction identically equal to one on C[0, 0] = C({0}). Then for all
F : C' -»• C,
under U.S. or applicable copyright law.

Proof The bilinearity (18.3.5) of * is easily verified as is (18.3.7). The key to establishing
the "associativity" of * is to show that, for any x e Ct1+t2+t3, both sides of (18.3.6),
when applied to x, yield F(x\)G(x2)H(x3,), where x\ and x2 are given by (18.3.1) and

The two arguments are similar. We carry out the proof that
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 538 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

By definition of *,

where x'1(s) = x(s), s e [0, t\ + t2], and x'2(s) — x(t\ + t2 + s), s e [0, t3]. Note
that x'2 — XT, so that the second factor on the right-hand side of (18.3.10) is just H(XI).
Hence, it remains to show that (F * G)(x() = F(x\)G(x2). But, by definition,

where x'{(s) = x [ ( s ) , s e [0, ti], and x^(s) = x[(t\ + s), s e [0, t2\- Hence, it remains
only to show that x" = x\ and x£ — X2-
But t\ + s e [t\,t\ + t2\ since ^ € [0, ^1 and so x((t\ + s) = x(t\ + s). Hence
x%(s) = x\ (t\ + s) = x(t\ + s) for ^ e [0, t2\; that is, x% = X2 as desired.
Finally, x\(s) — x(s) since ^ e [0, t\] and so *"(s) = x(s), s e [0,t\]. Hence
x" = x\ as desired.
Theorem 18.3.4 (Algebraic properties of +)
(1) Let F, FI : C"1 -> C and let G, G{ : C'2 -* C; let a, ft 6 C. Then

(3) Let 0 be the function identically equal to zero on C[0,0] = C({0}). Then for all
F : C' ->• C,

Proof The "associativity" of + is obtained just as in the proof of (18.3.6) by showing

thatbothsides of (18.3.13), when applied to* € C'[+t2+'\ yield F(xi) + G(x2) + H(x3).
under U.S. or applicable copyright law.

As for the linearity of +, we note that (18.3.11), when applied to x e ct1+t2+t}, is

equivalent to the equality

as required.

We conclude this section by giving two properties relating * and 4-. Two further
examples of such properties will be provided by equations (18.5.30), (18.5.32), (18.5.35)
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
and UNIVERSIDAD
(18.5.37) DISTRITAL
of Section 18.5.JOSE DE CALDAS
FRANCISCO
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE NONCOMMUTATIVE OPERATIONS * AND + 539
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 18.3.5 Let F : Cf' -*• C and G : C* -> C.

(1) Then F + G, exp (F + G) and exp(F) * exp(G) all map C"1 +'2 to C and we have

(2) Let n be a positive integer. With the conventions indicated in Remark 18.3.6(b)
below, we have

Remark 18.3.6 (a) By (18.3.15), we also have exp(G-j-F) = exp(G) * exp(F), but
due to the noncommutativity involved, these quantities are not equal to exp (F + G). An
analogous comment applies to (18.3.16).
(b) In (18.3.16), we let F° = l t ] , where lf| is the function which is identically one
on [0, t}]; further, G° is interpreted as It2, with It2 similarly defined.
Proof of Theorem 18.3.5 We first establish the exponential formula (18.3.15). For* e
C" 1+ ' 2 ,wehave

We next derive the binomial formula (18.3.16). Let x e C'1+'2. Then

Note that in the last equality of (18.3.17) we have made use of Remark 18.3.6(b)
according to which
under U.S. or applicable copyright law.

and

Hence the result is established.

The reader should keep in mind that the results of this section hold for equivalence
classes of functionals; however, except for Theorem 18.3.2, they clearly hold for point-
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
wiseUNIVERSIDAD
operations as well.
DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 540 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

18.4 The functional integrals K[(-) and the operations * and +

In this section, we present results relating the noncommutative operations introduced in
Section 18.3 and the functional integrals K[(F) defined in Section 15.2.1.
Let F and G be C-valued functions on C'] and C'2, respectively. We originally
conjectured the formula

for F € Ati and G e At2 • (The definition of the Banach algebras At, t > 0, introduced
in Section 15.7, will be briefly reviewed in Section 18.5.) It is still in the framework
of the Banach algebras that we will provide in the next section an affirmative answer
to our question, discussed in the Section 18.1, about the possibility of disentangling
K^ (F)K'^(G). However, it turns out that equation (18.4.1) itself is true in much greater
generality; we establish it here for the functional integrals K [ ( F ) , t > 0, A. e C+ and,
in particular, for the Feynman integral (A. e C^ \ C+). The case where A > 0 is treated
in Theorem 18.4.1, whereas the case A e C+ is dealt with in Corollary 18.4.4. (See
[JoLa4, Theorem 4.1 and Corollary 4.1, pp. 85 and 88].)
Theorem 18.4.1 Let X > 0 be given and suppose that F : C'1 -» C and G : C'2 -> C
are such that K'^(F) and K^(\G\) exist. Then K[I +'2(F*G) exists and formula (18.4.1)
holds.
Remark 18.4.2 Let A. > 0 be given. If we assumed in Theorem 18.4.1 only that K'£ (G)
exists, this would mean that, for any <p e L2(M.d) and for Leb.-a.e. t-,

exists as an absolutely convergent integral and defines a function of E which is in L2(Rd).

[See equation (15.2.8) from Definition 15.2.1; further, recall that with our present nota-
tion, mt2 (= m) denotes Wiener measure on CQ.] It is not required by the definition
that
under U.S. or applicable copyright law.

is an L2-function of. However, our use of the Fubini theorem in the proof below does
require this. It is in order to justify the use of the Fubini theorem that we have assumed
the existence of K£ (\ G \).
Proof of Theorem 18.4.1 The arguments will depend heavily on the notation and results
given in Section 18.2.
LetA. > Oandi/f e L 2 (R d )be given. The first equality in(18.4.4)belowfollows from
the definition of *. (See (18.3.2').) The second, third and fourth equalities in (18.4.4)
are the main steps in the proof. The second equality follows from (18.2.7), which relies
on the independent increment property of the Wiener process; the fourth results from the
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
application ofDISTRITAL
UNIVERSIDAD the translation
FRANCISCO map
JOSE given in (18.2.4) and from (18.2.5), which rests on the
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE FUNCTIONAL INTEGRALS K((-) 541
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

stationarity of the increments of the Wiener process; the third equality follows from the
Fubini theorem. The use of the Fubini theorem will be justified below.
In (18.4.4), we make implicit use of the maps p1, pi and p3 given in (18.2.6), (18.2.8)
and (18.2.9), respectively, which identify appropriate spaces of continuous functions. We
remind the reader that we write C'0 rather than C®''.

Note that in (1), (5) and (6) we have made use of the definition of K'^(F) given in
under U.S. or applicable copyright law.

(15.2.8).
To justify the use of the Fubini theorem, it will be convenient to work backward
beginning with expression (5) above. That expression exists with F, G and ty replaced
by | F|, \G\ and |u | as we next explain. Since we have assumed that K'£ (\G\) exists, the
function

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 542 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

belongs to L2(Wd) as a function of f. Then, since K^(F) exists, we know that for
Leb.-a.e. £ e Rd,

By combining (18.4.5) and (18.4.6), we obtain the existence (for Leb.-a.e. £) of expres-
sions (5) and (4) in (18.4.4) with absolute values on F, G and i/r. Finally, to go from
(4) to (3), we use (18.2.5) and the abstract change of variable theorem (Theorem 3.3.2),
but, this time, with \F\, \G\ and |^| involved. This shows that the Fubini theorem can be
applied to (3) (without absolute values) to yield (2). Hence, the proof of Theorem 18.4.1
is now complete. d
Remark 18.4.3 (a) It is not our present concern, but it appears that the proof given
above carries over to the setting of a certain class of stochastic processes which includes
the Wiener process as one example.
(b) Assume that GO : C'2 -*• [0, oo) dominates G in the sense that, for X > 0 and
forLeb.xmt2-a.e. (f, w) e Rd x C%, \G(X~^2w + |)| < GD(l.-l/2w+$). Then, the
existence of K£(GD) for X > 0 implies that of K^(\G\) (and of K'*(G))for A > 0.
This simple observation will be of use to us in Section 18.5 (see, in particular, Lemma
18.5.5 and Theorem 18.5.6) and should be kept in mind by the reader throughout the rest
of this section.
Next we proceed to Corollary 18.4.4 and the case A. e C+. We wish to make assump-
tions on F and G which will insure the existence of ^1+/2(f * G) and the validity of
formula (18.4.1) for all X e C+. A minimal set of assumptions would seem to be that
K^(F) and K^(G) exist for all A. e C+. In fact, we will make one mild additional
assumption: the existence, for all A. > 0, of A^ 2 (|G|). We remind the reader that the
case A 6 C+ \ C+ (i.e. A. purely imaginary and nonzero) corresponds to the analytic
operator-valued Feynman integral. (See Definition 15.2.1.)
Corollary 18.4.4 Suppose that F : C'[ -+ C and G : C'2 -+ C are such that K^ (F)
and K%(G) exist for all \ € C+ and that K%(|G|) exists for all A > 0. Then K1J+'2(F *
G) exists for all A € C+ and
under U.S. or applicable copyright law.

Proof According to Theorem 18.4.1, equation (18.4.7) holds for all A > 0; the cases
A e C+ and A 6 C^ can now be dealt with by analytic continuation and strong continuity,
respectively. Q
Remark 18.4.5 Recall that, in Remark 18.3.1, we pointed out the connection between
time-ordering and the operation * and +. Equation (18.4.7) is intimately related to this
time-ordering. The function F in the formula (F * G)(x) = F(x\)G(x2) depends only
on EBSCO
x\, the pan of the path x over [0, t\], whereas G depends only on X2, the part of x
Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
overUNIVERSIDAD
[t\, t\ +DISTRITAL
t2\; correspondingly,
FRANCISCO JOSE DE the operators K'^ (F) and K'^ (G) appear first and
CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE FUNCTIONAL INTEGRALS K{(-) 543
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

second, respectively, on the right-hand side of (18.4.7). [Since the operator K£ (G) acts
first, formula (18.4.7) actually correponds to precisely reversing the time-ordering. This
can be rectified by working with the "time-reversal" of the functionals involved, as is
discussed carefully in Remark 15.4.4 and Section 15.7 (as well as in Remark 15.3.7 and
Example 15.3.8).]
Note that in order for equation (18.4.7) to hold, it is important that the functional
integrals K[(F), defined in (15.2.8), act at the appropriate times.
There is a natural definition of "commutator" relative to the operation *. Given
F : Ct1 ->• C and G : C'2 -»• C, the commutator [F, G] of F and G is a C-valued
function on C'1 +'2 which we define by

The next corollary shows that the functional integrals K'^(-) preserve commutators,
a simple but possibly useful fact.
Corollary 18.4.6 Let F : C" ->• C and G : C'2 -» C be given.
(1) FixK>Q. Suppose that K^ ( \ F \ ) and K**(\G\) exist. Then K[l+'2([F, G]) exists
and

where the bracket on the right-hand side of (18.4.9) denotes the usual commutators
of bounded operators on L2(Rd); namely, [K'^(F), K'2(G)] := K[l(F)K^(G) -
K'2(G)K^(F).
(2) Suppose that K^ (F) and K%(G) exist for all A. e C+ and that K^ ( \ F \ ) and
K%(\G\) exist for all A > 0. Then, for all A e C~, Kt1 + t 2 ([F, G]) exists and (18.4.9)
holds.
Proof The existence of K!X[+'2(F * G) and K[l+'2(G * F)—and hence of the operator
K[[+t2([F, G])—is insured by Theorem 18.4.1 and Corollary 18.4.4. A straightforward
calculation now yields (18.4.9):
under U.S. or applicable copyright law.

as desired.

The following result is easily obtained by combining Theorem 18.3.5 with Theorem
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
18.4.1 and Corollary
UNIVERSIDAD 18.4.4. JOSE DE CALDAS
DISTRITAL FRANCISCO
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 544 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Theorem 18.4.7 Let F : Ctl -> C and G : C 2 t 2 - C be given.

(1) Fix A > 0. Suppose that A^1 (exp(F)) and K^2 (exp(Re G)) exist. Then
Kl+'2(exp(F + G)) exists and

(2) Suppose that Kt1(exp(F)) and Kt2(exp(G)) exist for all A. 6 C+ and that
Kt2(exp(Re G)) exist for all X > 0. Then, for all A e C+, ^1+'2(exp(F + G)) exists
and (18.4.10) holds.
Proof According to Theorem 18.4.1 (respectively, Corollary 18.4.4), we know that for
A. > 0 (respectively, A e C+), K'^+'2 (exp(F) * exp(G)) exists and equals the right-hand
side of (18.4.10). Moreover, by equation (18.3.15) and Theorem 18.3.5, exp(F + G) =
exp(F) * exp(G). The present theorem now follows by combining these facts.
Just as the exponential formula (18.3.15) was used to obtain equation (18.4.10), we
could use the binomial formula (18.3.16) to obtain a related equation. We forego this
here, but we will give such a formula in the Banach algebra setting, namely, equation
(18.5.34) of Section 18.5. In fact, it is possible to give a variety of further formulas. (See,
for instance, Examples 18.5.12-18.5.14 below.) However, we will limit ourselves in this
section to one additional result; although it is a simple consequence of Corollary 18.4.6
and Theorem 18.4.7, the nature of the formula is perhaps slightly surprising.
Corollary 18.4.8 Let F : C" -» C and G : C'2 -» C be given.
(1) Fix X > 0. Suppose that /s^'(exp(Re F)) and /sT^(exp(Re G)) emf. TTzew
K^ (exp(F + G)) and Kt +'2 (exp(G + F)) exist and
+t2

(2) Suppose that K1(exp(F)) and KT(2(exp(G)) exist for all X <= C+ and that
AT('(exp(Re F)) and A^2(exp(Re G)) exist for all A. > 0. Then, for all X e C+,
^'+'2 (exp(F + G)) anrf ^'+tl (exp(G + F)) exwr and (18.4.11) toWs.
Proof The existence of the expression on the right-hand side of (18.4.11) is guaranteed
by Theorem 18.4.7. Further, by (18.4.8) and (18.3.15),
under U.S. or applicable copyright law.

Equation (18.4.11) now follows by appling K^+'2 to both sides of (18.4.12) and using
Corollary 18.4.6. D

18.5 The disentangling algebras At, the operations * and +, and the
disentangling process
We begin this section by reviewing facts about the Banach algebra At introduced in
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
Section 15.7. We show, in Theorem 18.5.3, that the family of Banach algebra (A '• t > 0}
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS A, 545
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

is closed under * and 4- by proving that, if F € Atl and G e At2, then F * G and F + G
are in Atl+tr Several formulas given in the functional integration framework of the last
section are seen to hold in the Banach algebra setting; in particular, we show that, if F €
AH and G 6 A,2, then (without further hypotheses) K^+'2(F * G) = Kt^(F)Kt)2(G)
for all X e C+. An immediate corollary of this result and of our earlier work in [JoLal]
(see Chapter 15, especially Theorem 15.4.1 and Corollary 15.4.3) is that K^ (F)K^(G)
can be disentangled via a generalized Dyson series, thus resolving in the affirmative the
question raised in Section 18.1 and which initially motivated our work in [JoLa3,4].
We will see in this section that the noncommutative operations * and + combine with
the Banach algebra framework and functional integration to provide a rich interlocking
structure.
Let t > 0 be fixed for now. We recall briefly some definitions and notation from
Chapter 15. More explanation and related references can be found in Sections 15.2,15.4
and 15.7. Let M[0, t) denote the space of C-valued Borel measures on [0, t). Given
n e M[0, t), a C-valued Borel measurable function 9 on [0, t) x Rd is said to belong to
Looi;n - £<x>i;>n[0, t)if

where |n| in (18.5.1) denotes the usual total variation measure associated with n) (see
Section 15.2.G). Recall from Section 15.7 that functions (often called functionals) in
At are formed as follows: take a sequence {Fn}^L0 of functions on C' each of which is
given by an expression of the form

with mn a nonnegative integer, Nn,u e M[0, t), and 0n,u e Loo1;Nn,u• Assume that

Define a functional F by
under U.S. or applicable copyright law.

It is shown in Corollary 15.4.3 that for every k > 0, the series in (18.5.4) converges
absolutely for Leb.xm/-a.e. (|, x) € Rd x C'0; then At is defined as the set of all
equivalence classes of functionals F obtained in this manner. The representation of
functionals in (18.5.4) and even in (18.5.2) is not unique. For F € At, we let \\F\\t be
the infimum of the left-hand side of (18.5.3) for all representations of F of the form
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
(18.5.4). In Theorem
UNIVERSIDAD 15.7.1, we
DISTRITAL FRANCISCO JOSE have shown that (At, \\ • \\t) is a commutative Banach
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 546 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

algebra under pointwise multiplication and addition such that, given F € At, K^(F)
exists for all A. e C+ and satisfies

The inequality (18.5.5) shows that the linear operator K( from At to £(L 2 (R d )) is
bounded. (See Corollary 15.7.3.)
One of the main results of Chapter 15 shows that each operator K[(F), F e At, is
given in disentangled form by a generalized Dyson series (GDS). See Corollary 15.4.3
(in conjunction with Theorem 15.4.1). As in Section 15.7, we will often refer to the
Banach algebra At as the "disentangling algebra". Roughly speaking, a GDS is a time-
ordered perturbation expansion of operators. One operator K[(F) may have many GDSs
correponding, in particular, to various representations of the form (18.5.4), (18.5.2).
Remark 18.5.1 (a) As was explained in various places in Chapters 15-17, the use of
Lebesgue-Stieltjes measures in the definition of At enabled us to blend continuous and
discrete structures.
We used the space Af (0, t) throughout Chapter 15 with only a few exceptions; how-
ever, the use of the open interval (0, t) was merely a matter of convenience as was
essentially noted in Section 15.2.F. (Also see Chapter 17.)
(b) If the measures nn:U in (18.5.2) are all taken to be Lebesgue measure I, we obtain
a Banach subalgebra At of the disentangling algebra At- (See Remark 15.7.2.) This
Banach subalgebra was introduced and studied in [JoSk6, esp. pp. 121-124] except that,
in [JoSk6], the functions on,u were taken to be Lebesgue measurable rather than Borel
measurable. A Banach algebra closely related to At was introduced earlier in [JoSk2].
In [JoSk6], a Dyson series was given for each F e At- These perturbation series are
rather general in certain respects but possess a much less complicated combinatorial
structure than the GDSs from [JoLal] and studied in Chapter 15. The papers [JoSk2,
6] are relevant to Feynman 's operational calculus. However, the connection, although
suspected by the authors of [JoSk2,6], was not understood at the time.
Our first theorem provides the key step in showing that if F is in At, and G E At2,
then F * G and F + G are in Atl+t2-
Theorem 18.5.2 Let t > 0 and suppose that H e As-r, where 0 - < s < t. Let
under U.S. or applicable copyright law.

R : C' —* Cr's be the restriction map (i.e. the map that restrictsx in C' to the subinterval
[r, s]) and let T : Cr's -> Cs~r be the translation map. Then H o T o R e At and

Remark. Note that T = Tr-s as in (18.2.4) but that R is slightly different than R\ and
R2 from (18.2.2) and (18.2.3), respectively, since the restriction is allowed to be to any
subinterval.
Proof of Theorem 18.5.2. Let H e As-r- We take an arbitrary representation of H
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
in terms of measures
UNIVERSIDAD <$„,„ e JOSE
DISTRITAL FRANCISCO M[0, 5 - r) and functions finiU e £ooi;/3 nu [0, s - r),
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS A, 547
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

n = 0, 1, 2 , . . . , u = 1,2, ...,mn, such that

where

Translate £2n,u and fin,u to [r, s). Call the resulting functions and measures &n\ and
Pn,u> respectively. Mure precisely,

and, given B 6 B([r, s)), the Borel class of [r, s), we have

In view of (18.5.9), we have

Next we extend Q^u and fi^l to [0, t) by letting both be zero off the interval [r, s).
Call the resulting functions and measures £2J^ and /3^, respectively. In particular,
Pn2l(B) = Pn]l(B n [r, s)), for B e B([0, t)) . Note that A® € M[0, t) and that
under U.S. or applicable copyright law.

where we have used (18.5.10) in the last equality. In view of (18.5.11), £2^, €
L EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
v[0,t).
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
°°l;Bn,u
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 548 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(18.5.12)

It now follows from (18.5.1), (18.5.11), (18.5.12) and the definition of the norm as the
infimum of the left-hand side of (18.5.3) taken over all representations that

Since the representation of H in (18.5.7) was arbitrary, it follows from (18.5.13) that
H oT oR e At and \\HoToR\\t < \\H\\s-r, as desired.
The following theorem will enable us to study the operations * and 4- in the context
of the Banach algebras [At '• t > 0}. It will also serve as a connecting link with the
results of the previous sections.
Theorem 18.5.3 If F e Atl and G e At2, then F * G and F + G are in At1 +t2- Further,
we have

and
(18.5.15)

Proof Let x 6 C'1+'2. By definition of the operations * and + (see (18.3.2') and
under U.S. or applicable copyright law.

(18.3.3')),

where R1 : C'1+'2 -> Cfl and R2 : C'1+f2 ->• Ct1,t1+'2 are restriction maps and T :
C'i,ti+t2 _^ c'2 is the translation map. The map R of Theorem 18.5.2 corresponds to
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
RI UNIVERSIDAD
and R2, respectively, for the
DISTRITAL FRANCISCO JOSE special
DE CALDAScases (i) r = 0, s = t1 and t = t\ + t1 and
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS A, 549
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

(ii) r = t1 , s = t = t\ + t2- Further, the translation map of Theorem 18.5.2 corresponds

to T in case (ii), whereas it is equal to the identity in case (i). We deduce from Theorem
18.5.2 that F o R1 and G o T o R2 belong to At1+t2 and that

It now follows from (18.5. 16) and (18.5.17) that F * G and F 4- G are in A,t+,2 and that,
by (18.5.18),

and

as desired. Note that we have used here the fact that Atl+t2 is a normed algebra (Theorem
15.7.1).

We emphasize to the reader that each disentangling algebra At has its own opera-
tions, pointwise addition and commutative multiplication. The operations * and + are
additional operations which act on a pair of algebras Atl x A,2 and produce an element
of the algebra At1 +t2 .
The next result shows that * and + are compatible with the structure of the Banach
algebras (A, II • II/), ' > 0.
Proposition 18.5.4 The operation * (resp., + ) is C-bilinear (resp., C-linear) and
continuous from At{ x At2 to A^+tr Further, * and + are "associative" in the sense
that if F € Ati, G € -^2 an^ H 6 At3, then the following equalities hold in Atl+t2+t3'-

and
under U.S. or applicable copyright law.

Proof The algebraic properties follow from their counterparts in Theorems 18.3.3 and
18.3.4 as well as from Theorem 18.5.3. The continuity of* (resp., + ) is a consequence of
the bilinearity (resp., linearity) and the norm estimate (18.5.14) (resp. (18.5.15)).

Lemma 18.5.5, to follow, will be useful in the proof of Theorem 18.5.6 below and is
of some independent interest.
Lemma 18.5.5 Let G e At . Then there exists GD € At which dominates G. (Compare
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
(18.5.21) and (18.5.22) below.) Further, K((\G\) exists for all X > 0.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 550 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

Proof Consider a representation of G of the form

where

Define

The functional GQ dominates G in the sense that

for all X > 0 and Leb.xm,-a.e. (£,*) € Rd x CQ. Clearly, GD € A- Consequently,

by Theorem 15.7.1, K t ( G D ) exists for all A e C^ and, in particular, for all A, > 0 as
we recalled just preceding (18.5.5). It now follows from Remark 18.4.3(b) that K[(\G\)
exists for all A. > 0.
Note that Lemma 18.5.5 does not assert that |G| € At but only that Kt(\G\) exists
for all X > 0.
The nexttheorem ([JoLa4, Theorem 5.3, p. 96]), which is now an easy consequence of
our earlier results, is one of the main theorems of this chapter. As an immediate corollary,
we obtain a solution to the problem concerning disentangling which was stated in the
introduction and that initially motivated this work.
Theorem 18.5.6 Let F e At1 and G e At2- Then, for all X e C+, the analytic operator-
valued Feynman integrals K[{ (F), K[2 (G) and Kt1 +'2 (F * G) exist and we have

Proof Recall again that if H € At, K[(H) exists for all A > 0. The result now
under U.S. or applicable copyright law.

follows from Theorem 18.5.3 according to which F * G 6 Atl+t2, Corollary 18.4.4 on

functional integrals, as well as Lemma 18.5.5. Note that this lemma justifies the fact that
the hypothesis of Corollary 18.4.4, stating that K^(\G\) exists for all A > 0, is satisfied.
D

Corollary 18.5.7 If F e A, and G e A,2, then, for all X e C+, K^ (F)K^(G) can be
disentangled via a generalized Dyson series (GDS) for Kt1 +t2 (F * G).
Proof Recall that if H 6 At, then, for all X € C^, K{(H) can be disentangled via a
GDS (Theorem 15.7.1). But F * G is in the disentangling algebra At1 +t2 and so the result
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
follows from equation (18.5.23) of Theorem 18.5.6.
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 THE DISENTANGLING ALGEBRAS A, 551
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

The next result is the counterpart of Corollary 18.4.6 and is an immediate consequence
of Theorem 18.5.6.
Corollary 18.5.8 Let F e Atl and G e At2. Then [F, G] := F * G - G * F is in A<l+t2
and, for all k € C+,

In reference to the next result, the counterpart of Theorem 18.4.7, we mention that
the holomorphic functional calculus for Banach algebras ([Nai, pp. 202-205], [KadRi,
§3.3]) assures us that exp(H) e At if H e At. Since the Banach algebra operations are
just pointwise addition and multiplication, exp(H), in the sense of the Banach algebra
At, coincides with exp(H), which was defined pointwise in Section 18.3.
Theorem 18.5.9 Let F e Atl and G e At2. Then the following equality holds in A,l+t2:

Proof Equation (18.5.25) is just equation (18.3.15) of Theorem 18.3.5, interpreted in

At1+t2- Equation (18.5.26) now follows from Theorem 18.5.6. Note that the fact that
H e At implies that K^ (exp(H)) exists for all A. e C+.
Remark 18.5.10 (a) The exponential formula (18.5.26) is the counterpart of
Theorem 18.4.7. Observe that this result, along with Theorem 18.5.6, Corollary 18.5.8,
and Corollary 18.5.11 below, takes a simpler form in the Banach algebra setting than in
Section 18.4.
(b) It is well known that if A and B are noncommuting operators, then the formula

may fail. Nevertheless, Feynman gives examples which show that equation (18.5.27),
properly understood using his time-ordering convention and "disentangling ", may also
under U.S. or applicable copyright law.

"hold" in a certain sense and be useful when A and B are noncommuting [Fey8, equa-
tion (4), p. 110 and equation (8), p. 111]. There are various ways of making Feynman's
ideas precise in certain cases [Ar, Gil, GilZa, JoLal, Masll, NazSS, Ne3, ...].
In our present setting, the form of the paradoxical formula (18.5.27) is preserved
in formula (18.5.25), but of course we have changed the ordinary algebraic operations
+ and • to the noncommutative operations + and *, respectively.
Note that we are not working directly with operators but with functionals; we first
establish (18.5.25) for functionals and then derive the corresponding formula (18.5.26)
for the operator-valued functional integrals.
(c) In formulas (18.5.25) and(18.5.26), one sees particularly clearly the relationships
EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
between aspects
UNIVERSIDAD of theFRANCISCO
DISTRITAL presentJOSE
work and that of Lapidus on the "Feynman-Kac formula
DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 552 NONCOMMUTATIVE OPERATIONS ON WIENER FUNCTIONALS
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted

with a Lebesgue-Stieltjes measure" [Lal4-18] discussed in Chapter 17. (See especially

Theorem 1 7.2. 7 [Lai 5, Theorem 2.4, p. 99] and the comments preceding Remark 1 7.5. 7.)
(d) In [Lal9], an axiomatic formulation of key aspects of Feynman's operational
calculus is developed in which, for instance, the abstract analogue of formula (18.5.25)
holds, as will be discussed in Appendix 18.6 below.
The last corollary which we will formally state is the counterpart of Corollary 1 8.4.8
in the setting of the Banach algebras [At : t > 0}.
Corollary 18.5.11 Let F e At1 and G € At2. Then the following equality holds in

further, for all h e C=,

Proof The result follows immediately from Corollary 18.4.8 with the aid of
Lemma 18.5.5.

Examples: Trigonometric, binomial and exponential formulas

Many further formulas hold in the setting of the Banach algebras (At : t > 0}. We
conclude this section by simply stating three examples.
Example 18.5.12 (Trigonometric formulas)
(i) Let F € At] and G € At2, then the following equality holds in Atl+t2'.
under U.S. or applicable copyright law.

(ii) Similarly, under the same assumptions as in (i), we have the following equality
in Atl+t2-

fu rther ,foralh e C+,

EBSCO Publishing : eBook Academic Collection (EBSCOhost) - printed on 6/8/2017 5:50 PM via
UNIVERSIDAD DISTRITAL FRANCISCO JOSE DE CALDAS
AN: 98476 ; Johnson, Gerald W., Lapidus, Michel L..; The Feynman Integral and Feynman's
Operational Calculus
Account: ns000601
 APPENDIX: AXIOMATIC FEYNMAN'S OPERATIONAL CALCULUS 553
Copyright © 2000. Oxford University Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted