DYNAMICAL

symmcTry

Carl E Wulfman

World Scientific
This page is intentionally left blank
 Carl E Wulfman
University of the Pacific, USA

World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TA I P E I • CHENNAI
 Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

DYNAMICAL SYMMETRY
Copyright © 2011 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.

ISBN-13 978-981-4291-36-1
ISBN-10 981-4291-36-6

Typeset by Stallion Press

Email: enquiries@stallionpress.com

Printed in Singapore.

YeeSern - Dynamical Symmetry.pmd 1 12/29/2010, 2:44 PM

October 28, 2010 16:55 9in x 6in b943-fm

Dedication

To my fellow scientists who paused to encourage as they explored new

vistas — Leonello Paoloni, John Platt, Marcos Moshinsky, Brian Judd,
Brian Wybourne, Raphy Levine, and John Avery.

v
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-fm

Preface

In his Harmonices Mundi, published in 1619, Johannes Kepler exclaims,

“Finally I have brought to light and verified beyond all my hopes and
expectations that the whole Nature of Harmonies permeates to the fullest
extent, and in all its details, the motion of heavenly bodies; not, it is true,
in the manner in which I had earlier thought, but in a totally different,
altogether complete way.”
This book deals with concepts of symmetry which, nearly four cen-
turies after Kepler, continue to provide ever deepening, and often surpris-
ing, understandings of interrelations between phenomena associated with
moving bodies.
Kepler’s religious convictions and his intuition led him to believe that
geometric symmetries and musical harmonies are expressed in natural phe-
nomena. The development of the calculus and Newtonian physics in the lat-
ter half of the seventeenth century made it possible to deduce and extend
Kepler’s three laws of planetary motion, but they were seldom seen as
expressions of symmetries and harmonies in the natural world. Then, two
and a half centuries after we were given Harmonices Mundi, our concepts of
symmetry were profoundly deepened by Sophus Lie. Lie saw that any sys-
tem governed by differential equations exhibits symmetries in a heretofore
unrecognized sense. Because so many laws of nature can be formulated as
differential equations, Lie’s concept of symmetry makes it possible to estab-
lish a host of relationships between natural phenomena. By mid twentieth
century, physicists had come to realize that relationships expressing Lie
symmetries, though easily overlooked, are surprisingly useful. Now, much
modern physics has become a search, first for symmetries, then, secondly,
for the dynamics that produces them.
Readers of Lie’s works cannot help but realize that much of his
mathematics was guided by geometric intuitions. The extraordinary fertility

vii
October 28, 2010 16:55 9in x 6in b943-fm

viii Dynamical Symmetry

of his thought processes, beyond a shadow of doubt, arose in part from

the extraordinary visual capabilities of the human mind. Just as a picture
can be “worth a thousand words”, so also can a geometric illustration,
even a non-Euclidean one, suggest a host of mathematical and physical
relationships. For this reason, the first three chapters of this book use
interconnected geometrical, analytic, and physical concepts of symmetry
to stimulate questions as well as insights in the mind of the reader.
A more formal treatment of the subject begins in Chapter 4. Among
other things, Chapters 4 and 6 establish the manner in which Lie sym-
metries are defined by differential equations. Chapter 5 explains how one
determines the invariance transformations of functions, functionals, and
equations. Readers may wish to shift back and forth between the first three
chapters, and these chapters.
The remaining chapters of the book deal with physical and chemical con-
sequences of the symmetries implied in the dynamical equations of mechan-
ics, quantum mechanics, and electromagnetism. Invariance transformations
of partial differential equations are discussed in Chapter 9, the first chapter
dealing with Schrödinger equations.
A number of monographs currently in print deal with the invariance
properties of differential equations, and several recent and very fine ones
deal with the utilization of these properties in applied mathematics and
engineering. A list of these will be found in the Bibliography at the end of
this Preface. Helpful, but out of print, monographs are also listed.
The approach taken in this book contrasts with those to be found in
the books of the Bibliography: here, geometric concepts play a greater role.
It is hoped that the reader’s natural geometric understanding of ordinary
symmetry, will, like Lie’s, be extended into a partly intuitive, geometric
as well as analytic, understanding of symmetries not seen in the space of
our common experience. A good deal of emphasis is placed upon dynami-
cal symmetries of classical and quantum mechanical systems of interest to
both chemists and physicists. Readers interested in engineering or applied
mathematics may also find that the book provides viewpoints and analyses
which can enrich their studies.

References
[1] G. W. Bluman, S. Kumei, Symmetries and Differential Equations (Springer-
Verlag, N.Y., 1989).
[2] J. E. Campbell, Continuous Groups; reprint of the 1903 edition of Introduc-
tory Treatise on Lie’s Theory of Finite Continuous Transformation Groups
(Chelsea, Bronx, N.Y, 1966).
October 28, 2010 16:55 9in x 6in b943-fm

Preface ix

[3] B. J. Cantwell, Introduction to Symmetry Analysis (Cambridge University

Press, Cambridge, 2002).
[4] A. Cohen, The Lie Theory of One-Parameter Groups; reprint of the 1911
edition (Stechert, N.Y., 1931).
[5] N. H. Ibragimov, CRC Handbook of Lie Group Analysis of Differential Equa-
tions, Vol. 1, 1994–6 (CRC Press, N.Y. 1996).
[6] S. Lie, Theorie der Transformation Gruppen, I; translated by M. Acker-
man, entitled Sophus Lie’s 1880 Transformation Group Paper; comments
by R. Hermann (Math Sci Press, Brookline, Mass, 1975).
[7] Peter J. Olver, Applications of Lie Groups to Differential Equations
(Springer-Verlag, N.Y., 1986).
[8] J. M. Page, Ordinary differential Equations with an Introduction to Lie’s
Theory of the Group of One Parameter (Macmillan, London, 1897).
[9] H. Stephani, Differential Equations: Their Solution Using Symmetries
(Cambridge University Press, Cambridge, 1989).
[10] P. Hydon, Symmetry Methods for Differential Equations (Cambridge Uni-
versity Press, Cambridge, 2000).

AUDIENCE: Primarily physicists, physical chemists, applied mathemati-

cians and other natural scientists with a mathematical bent.

BACKGROUND ASSUMED: Differential and integral calculus, intro-

duction to ordinary differential equations. The familiarity with classical
mechanics and/or quantum mechanics attained by undergraduate students
of physics by their junior or senior year, or by first-year graduate students
of physical chemistry.
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-fm

Acknowledgments

It is a great pleasure to be able to acknowledge my indebtedness to those

who have provided so much of the mental stimulation leading to these pages:
to my parents Eugene and Jean, my wife Constance, and my brother, David;
to some very special teachers, Christian Rondestvedt, George Uhlenbeck,
and Michael Dewar; and to my students Fred Yarnell, Tai-ichi Shibuya,
Sukeyuki Kumei, Yutaka Kitigawara, Yuuzi Takahata, and Gerald Hyatt.
Jerry Blakefield, Rochelle Wolber, and Jack Bench provided invaluable
help checking and proofreading the manuscript, and Eve Wallis greatly
improved many of the figures.
Generous grants from the Research Corporation, the Lindbergh Foun-
dation, the University of the Pacific, and the Fulbright Commission, as well
as grants from the National Science Foundation, supported research that
was instrumental to the development of this book.

xi
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-fm

Contents

Dedication v

Preface vii

Acknowledgements xi

1. Introduction 1
1.1 On Geometric Symmetry and Invariance in the Sciences 1
1.2 Fock’s Discovery . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Keplerian Symmetry . . . . . . . . . . . . . . . . . . . . 8
1.4 Dynamical Symmetry . . . . . . . . . . . . . . . . . . . 8
1.5 Dynamical Symmetries Responsible for Degeneracies
and Their Physical Consequences . . . . . . . . . . . . . 10
1.6 Dynamical Symmetries When Energies Can Vary . . . . 12
1.7 The Need for Critical Reexamination of Concepts of
Physical Symmetry. Lie’s Discoveries . . . . . . . . . . . 13
Appendix A: Historical Note . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2. Physical Symmetry and Geometrical Symmetry 23

2.1 Geometrical Interpretation of the Invariance Group of
an Equation; Symmetry Groups . . . . . . . . . . . . . 23
2.2 On Geometric Interpretations of Equations . . . . . . . 29
2.3 Geometric Interpretations of Some Transformations in
the Euclidean Plane . . . . . . . . . . . . . . . . . . . . 30
2.4 The Group of Linear Transformations of Two Variables 36

xiii
October 28, 2010 16:55 9in x 6in b943-fm

xiv Dynamical Symmetry

2.5 Physical Interpretation of Rotations . . . . . . . . . . . 37

2.6 Intrinsic Symmetry of an Equation . . . . . . . . . . . . 39
2.7 Non-Euclidean Geometries . . . . . . . . . . . . . . . . 40
2.8 Invariance Group of a Geometry . . . . . . . . . . . . . 48
2.9 Symmetry in Euclidean Spaces . . . . . . . . . . . . . . 49
2.10 Symmetry in the Spacetime of Special Relativity . . . . 51
2.11 Geometrical Interpretations of Nonlinear Transforma-
tions: Stereographic Projections . . . . . . . . . . . . . 56
2.12 Continuous Groups that Leave Euclidean and Pseudo-
Euclidean Metrics Invariant . . . . . . . . . . . . . . . . 62
2.13 Geometry, Symmetry, and Invariance . . . . . . . . . . 65
Appendix A: Stereographic Projection of Circles . . . . . . . . 66
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3. On Symmetries Associated With Hamiltonian

Dynamics 71
3.1 Invariance of a Differential Equation . . . . . . . . . . . 71
3.2 Hamilton’s Equations . . . . . . . . . . . . . . . . . . . 73
3.3 Transformations that Convert Hamilton’s Equations
into Hamilton’s Equations; Symplectic Groups . . . . . 76
3.4 Invariance Transformations of Hamilton’s Equations of
Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.5 Geometrization of Hamiltonian Mechanics . . . . . . . . 80
3.6 Symmetry in Two-dimensional Symplectic Space . . . . 91
3.7 Symmetry in Two-dimensional Hamiltonian Phase Space 92
3.8 Symmetries Defined by Linear Symplectic Transformations 95
3.9 Nonlinear Transformations in Two-dimensional Phase
Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.10 Dynamical Symmetries as Intrinsic Symmetries of
Differential Equations and as Geometric Symmetries . . 102
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4. One-Parameter Transformation Groups 105

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 105
4.2 Finite Transformations of a Continuous Group Define
Infinitesimal Transformations and Vector Fields . . . . 108
4.3 Spaces in which Transformations will be Assumed to Act 112
October 28, 2010 16:55 9in x 6in b943-fm

Contents xv

4.4 The Defining Equations of One-Parameter Groups

of Infinitesimal Transformations. Group
Generators . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.5 The Differential Equations that Define Infinitesimal
Transformations Define Finite Transformation
Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.6 The Operator of Finite Transformations . . . . . . . . . 119
4.7 Changing Variables in Group Generators . . . . . . . . 122
4.8 The Rectification Theorem . . . . . . . . . . . . . . . . 124
4.9 Conversion of Non-autonomous ODEs to Autonomous
ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.10 N-th Order ODEs as Sets of First-order ODEs . . . . . 131
4.11 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 131
Appendix: Homeomorphisms, Diffeomorphisms,
and Topology . . . . . . . . . . . . . . . . . . . . . . . . 132
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

5. Everywhere-Local Invariance 135

5.1 Invariance under the Action of One-Parameter
Lie Transformation Groups . . . . . . . . . . . . . . . . 135
5.2 Transformation of Infinitesimal Displacements . . . . . 141
5.3 Transformations and Invariance of Work, Pfaffians,
and Metrics . . . . . . . . . . . . . . . . . . . . . . . . . 143
5.4 Point Transformations of Derivatives . . . . . . . . . . . 147
5.5 Contact Transformations . . . . . . . . . . . . . . . . . 149
5.6 Invariance of an Ordinary Differential Equation of
First-order under Point Transformations; Extended
Generators . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.7 Invariance of Second-order Ordinary Differential
Equations under Point Transformations. Harmonic
Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.8 The Commutator of Two Operators . . . . . . . . . . . 156
5.9 Invariance of Sets of ODEs. Constants of Motion . . . . 158
5.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 161
Appendix A. Relation between Symmetries and Integrating
Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
October 28, 2010 16:55 9in x 6in b943-fm

xvi Dynamical Symmetry

Appendix B. Proof That The Commutator of Lie Generators

is Invariant Under Diffeomorphisms . . . . . . . . . . . 163
Appendix C. Isolating and Non-isolating Integrals
of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

6. Lie Transformation Groups and Lie Algebras 169

6.1 Relation of Many-Parameter Lie Transformation Groups
to Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . 169
6.2 The Differential Equations that Define Many-Parameter
Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.3 Real Lie Algebras . . . . . . . . . . . . . . . . . . . . . 177
6.4 Relations between Commutation Relations and
the Action of Transformation Groups: Some
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.5 Transitivity . . . . . . . . . . . . . . . . . . . . . . . . . 186
6.6 Complex Lie Algebras . . . . . . . . . . . . . . . . . . . 187
6.7 The Cartan–Killing Form; Labeling and Shift
Operators . . . . . . . . . . . . . . . . . . . . . . . . . . 188
6.8 Casimir Operators . . . . . . . . . . . . . . . . . . . . . 191
6.9 Groups That Vary the Parameters of Transformation
Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
6.10 Lie Symmetries Induced from Observations . . . . . . . 193
6.11 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 195
Appendix A. Definition of Lie Groups by Partial
Differential Equations . . . . . . . . . . . . . . . . . . . 196
Appendix B. Classification of Lie Algebras
and Lie Groups . . . . . . . . . . . . . . . . . . . . . . . 202
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

7. Dynamical Symmetry in Hamiltonian Mechanics 207

7.1 General Invariance Properties of Newtonian
Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 207
7.2 Relationship of Phase Space to Abstract
Symplectic Space . . . . . . . . . . . . . . . . . . . . . . 210
7.3 Hamilton’s Equations in PQ Space. Constants
of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . 214
October 28, 2010 16:55 9in x 6in b943-fm

Contents xvii

7.4 Poisson Bracket Operators . . . . . . . . . . . . . . . . 217

7.5 Hamiltonian Dynamical Symmetries in PQ Space . . . 220
7.6 Hamilton’s Equations in Classical PQET Space;
Conservation Laws Arising From Galilei Invariance . . . 230
7.7 Time-dependent Constants of Motion; Dynamical
Groups That Act Transitively . . . . . . . . . . . . . . 234
7.8 The Symplectic Groups Sp(2n,r) . . . . . . . . . . . . . 237
7.9 Generalizations of Symplectic Groups That Have an
Infinite Number of One-Parameter Groups . . . . . . . 239
Appendix A. Lagranges Equation’s and the Definition
of Phase Space . . . . . . . . . . . . . . . . . . . . . . . 240
Appendix B. The Variable Conjugate to Time
in PQET Space . . . . . . . . . . . . . . . . . . . . . . 243
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

8. Symmetries of Classical Keplerian Motion 247

8.1 Newtonian Mechanics of Planetary Motion . . . . . . . 247
8.2 Hamiltonian Formulation of Keplerian Motions in Phase
Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
8.3 Symmetry Coordinates For Keplerian Motions . . . . . 257
8.4 Geometrical Symmetries of Bound Keplerian Systems in
Phase Space . . . . . . . . . . . . . . . . . . . . . . . . 262
8.5 Symmetries of Keplerian Systems with Non-negative
Energies . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
8.6 The SO(4,1) Dynamical Symmetry . . . . . . . . . . . . 270
8.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . 271
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273

9. Dynamical Symmetry in Schrödinger

Quantum Mechanics 275
9.1 Superposition Invariance . . . . . . . . . . . . . . . . . 275
9.2 The Correspondence Principle . . . . . . . . . . . . . . 276
9.3 Correspondence Between Quantum Mechanical
Operators and Functions of Classical Dynamical
Variables . . . . . . . . . . . . . . . . . . . . . . . . . . 280
9.4 Lie Algebraic Extension of the Correspondence
Principle . . . . . . . . . . . . . . . . . . . . . . . . . . 281
October 28, 2010 16:55 9in x 6in b943-fm

xviii Dynamical Symmetry

9.5 Some Properties of Invariance Transformations

of Partial Differential Equations Relevant to
Quantum Mechanics . . . . . . . . . . . . . . . . . . . . 285
9.6 Determination of Generators and Lie Algebra
of Invariance Transformations
of ((−1/2)∂ 2 /∂x2 − i∂/∂t)ψ(x, t) = 0 . . . . . . . 289
9.7 Eigenfunctions of the Constants of Motion
of a Free-Particle . . . . . . . . . . . . . . . . . . . . . . 294
9.8 Dynamical Symmetries of the Schrödinger
Equations of a Harmonic Oscillator . . . . . . . . . . . 297
9.9 Use of the Oscillator Group in Pertubation
Calculations . . . . . . . . . . . . . . . . . . . . . . . . 300
9.10 Concluding Observations . . . . . . . . . . . . . . . . . 301
Historical Note . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303

10. Spectrum-Generating Lie Algebras and Groups

Admitted by Schröedinger Equations 305
10.1 Lie Algebras That Generate Continuous Spectra . . . . 306
10.2 Lie Algebras That Generate Discrete Spectra . . . . . . 308
10.3 Dynamical Groups of N-Dimensional Harmonic
Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . 309
10.4 Linearization of Energy Spectra by Time Dilatation;
Spectrum-Generating Dynamical Group of Rigid
Rotators . . . . . . . . . . . . . . . . . . . . . . . . . . 311
10.5 The Angular Momentum Shift Algebra; Dynamical
Group of the Laplace Equation . . . . . . . . . . . . . . 315
10.6 Dynamical Groups of Systems with Both Discrete
and Continuous Spectra . . . . . . . . . . . . . . . . . . 319
10.7 Dynamical Group of the Bound States
of Morse Oscillators . . . . . . . . . . . . . . . . . . . . 320
10.8 Dynamical Group of the Bound States
of Hydrogen-Like Atoms . . . . . . . . . . . . . . . . . . 322
10.9 Matrix Representations of Generators
and Group Operators . . . . . . . . . . . . . . . . . . . 325
10.10 Invariant Scalar Products . . . . . . . . . . . . . . . . . 327
10.11 Direct-Products: SO(3) ⊗ SO(3) and the Coupling
of Angular Momenta . . . . . . . . . . . . . . . . . . . . 328
October 28, 2010 16:55 9in x 6in b943-fm

Contents xix

10.12 Degeneracy Groups of Non-interacting Systems.

Completions of Direct-Products . . . . . . . . . . . . . 329
10.13 Dynamical Groups of Time-dependent
Schrödinger Equations of Compound Systems.
Many-Electron Atoms . . . . . . . . . . . . . . . . . . . 331
Appendix A: References Providing Invariance Groups
of Schrödinger Equations . . . . . . . . . . . . . . . . . 334
Appendix B. Reference to Work Dealing with Dynamical
Symmetries in Nuclear Shell Theory . . . . . . . . . . . 334
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

11. Dynamical Symmetry of Regularized

Hydrogen-like Atoms 337
11.1 Position-space Realization of the Dynamical
Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . 337
11.2 The Momentum-space Representation . . . . . . . . . . 344
11.3 The Hyperspherical Harmonics Yklm . . . . . . . . . . . 349
11.4 Bases Provided by Eigenfunctions of J12 , J34 , J56 . . . 354
Appendix A. Matrix Elements of SO(4,2) Generators . . . . . 355
Appendix B. N-Shift Operators For the Hyperspherical
Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . 356
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359

12. Uncovering Approximate Dynamical Symmetries.

Examples From Atomic and Molecular Physics 361
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 361
12.2 The Stark Effect; One-Electron Diatomics . . . . . . . . 362
12.3 Correlation Diagrams and Level Crossings:
General Remarks . . . . . . . . . . . . . . . . . . . . . . 364
12.4 Coupling SO(4)1 ⊗ SO(4)2 to Produce SO(4)12 . . . . . 369
12.5 Coupling SO(4)1 ⊗ SO(4)2 to Produce SO(4)1−2 . . . . 371
12.6 Configuration Mixing in Doubly Excited States
of Helium-like Atoms . . . . . . . . . . . . . . . . . . . 373
12.7 Configuration Mixing Arising From Interactions
Within Valence Shells of Second and Third
Row Atoms . . . . . . . . . . . . . . . . . . . . . . . . . 375
12.8 Origin of the Period-Doubling Displayed
in Periodic Charts . . . . . . . . . . . . . . . . . . . . . 375
October 28, 2010 16:55 9in x 6in b943-fm

xx Dynamical Symmetry

12.9 Molecular Orbitals in Momentum Space.

The Hyperspherical Basis . . . . . . . . . . . . . . . . . 377
12.10 The Sturmian Ansatz of Avery, Aquilanti
and Goscinski . . . . . . . . . . . . . . . . . . . . . . . 380
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385

13. Rovibronic Systems 389

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 389
13.2 Algebraic Treatment of Anharmonic Oscillators
With a Finite Number of Bound States . . . . . . . . . 389
13.3 U(2) ⊗ U(2) Model of Vibron Coupling . . . . . . . . . 394
13.4 Spectrum Generating Groups of Rigid
Body Rotations . . . . . . . . . . . . . . . . . . . . . . 396
13.5 The U(4) Vibron Model of Rotating
Vibrating Diatomics . . . . . . . . . . . . . . . . . . . . 401
13.6 The U(4) ⊗ U(4) Model of Rotating
Vibrating Triatomics . . . . . . . . . . . . . . . . . . . 403
13.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . 406
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406

14. Dynamical Symmetry of Maxwell’s Equations 409

14.1 The Poincare Symmetry of Maxwell’s Equations . . . . 410
14.2 The Conformal and Inversion Symmetries
of Maxwell’s Equations; Their Physical
Interpretation . . . . . . . . . . . . . . . . . . . . . . . 413
14.3 Alteration of Wavelengths and Frequencies
by a Special Conformal Transformation:
Interpretation of Doppler Shifts in Stellar
Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431

Index 433
October 28, 2010 16:55 9in x 6in b943-ch01

CHAPTER 1

Introduction

Methinks that all of nature and the graceful sky

are set into symbols in geometrium.1

1.1 On Geometric Symmetry and Invariance

in the Sciences
The concept of symmetry that is commonly used in the physical and natural
sciences is that of geometric symmetry as usually conceived. A benzene
molecule is said to have the symmetry of a regular hexagon, a salt crystal
has a lattice with cubic symmetry, mammals have approximate bilateral
symmetry, and an ellipsoid has lower symmetry than a sphere.
The geometric symmetry of an object expresses an invariance of the
relation between an observer and the object — a relation that persists
after some particular alteration of the coordinates which determine this
relationship. The symmetry of a sphere provides an extreme example, since
the relationship between a sphere and an observer is unchanged by any
rotation of the sphere, or the observer, about the center of the sphere. The
symmetries of isolated atoms and molecules as well as of crystal lattices are
all defined by the particular rotations, translations, reflections, and inver-
sions that leave invariant the relationship between these physical objects
and the coordinate system of an observer. Associated with each geometric
object is the set of symmetry operations that leave invariant the relation
between the object and a coordinate system.
These conceptions of symmetry and operations of symmetry have their
basis in Euclidean geometry. Euclid defined two triangles as congruent if
they can be superposed by rotations and translations. All two- or three-
dimensional figures, polyhedra, surfaces, or solids that can be interconverted
by rotations and translations are congruent. In Euclidean geometry, objects

1
October 28, 2010 16:55 9in x 6in b943-ch01

2 Dynamical Symmetry

that can be superposed with the aid of translations, rotations, or uniform

dilatations, have the same symmetry.
At the beginning of the Renaissance, Euclid’s Elements appeared to
correctly and elegantly describe geometric relations here on earth. It was
natural to hope that geometric simplicity would find further expression in
the celestial works of God. Visualizing the sun at the center of the solar
system, Copernicus believed he had uncovered a wonderful role played by
the symmetry and simplicity of regular circular motion. Kepler thought
to extend this “Divine Harmony” by embedding the circular orbits of the
planets between concentric polyhedra. But intellectual honesty and a care-
ful analysis of the accuracy of Tyco Brahe’s observations forced Kepler to
recognize that the orbit of Mars is not circular. Thereafter, years of strug-
gle led him to the happy realization that nature expresses harmonies more
profound than those he, a mortal, at first anticipated. Kepler’s three laws,
and much of the intellectual journey that led to them, are all to be found
in his Harmonices Mundi.2
“Standing on the shoulders of giants,” Newton concluded that material
objects move in the space of Euclidean geometry, and that in the absence of
material objects, no point in the space is physically distinguished from any
other point. To maintain this translational invariance, he assumed that real
mass points are subject to a universal law of gravitation. He also assumed
that no point in time is fundamentally distinguished from any other point in
time: a time-line in his mechanics has a translational invariance analogous
to that of the infinite straight line. Newton’s laws of motion consequently
remain the same when the motion is viewed from coordinate systems with
different origins and timed by clocks with different initial settings. They are
also the same when objects are viewed by observers moving with constant
relative velocity, V . If Newton’s equations of motion are obeyed by an object
in an isolated system, the Euclidean symmetry of the object is the same for
all observers moving at constant velocities with respect to the object.
The development of Lie’s theory of groups in the latter half of the nine-
teenth century was followed by Noether’s realization that the conservation
laws of physics were consequences of the invariance properties of space
and time presupposed by Newton.3 The translational invariance of New-
ton’s laws of motion implies the law of conservation of momentum, their
rotational invariance implies the law of conservation of angular momen-
tum, and their invariance as time evolves implies the law of conservation
of mechanical energy. (A simple proof of these statements can be found in
Chapter 7.)
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 3

The invariance properties of Newton’s equations have many further

consequences. An outstanding early example of the exploitation of the
rotational and translational invariance inherent in Newtonian mechan-
ics is provided by Maxwell’s 1859 derivation of the Maxwell–Boltzmann
distribution law of statistical mechanics.4
In his early studies of electromagnetism, Maxwell visualized Faraday’s
lines of force as tubes of flow associated with the motion of incompress-
ible inviscid fluids.5a Such motions can conserve energy, momentum, and
angular momentum. To preserve these conservation laws of Newtonian
mechanics, Maxwell imagined that a displacement current flowed between
the plates of charging or discharging capacitors. This allowed him to consis-
tently suppose that Faraday’s electric and magnetic fields carry, but neither
create nor destroy, energy, momentum, and angular momentum. Preserv-
ing these conservation laws, Maxwell derived his equations governing elec-
tromagnetic fields.5b Maxwell’s discovery that light is a self-propagating
electromagnetic field, and the continual experimental verifications of his
theory, naturally led to the expectation that the theory was essentially
correct. In applying Maxwell’s theory, it was expected that electromag-
netic waves moved relative to the medium, aether, supposed to carry them.
When the experiments of Michelson and Morley showed that motion with
respect to such a medium is unobservable, it appeared that, as in New-
tonian mechanics, only relative motions of physical objects were physi-
cally significant. Galilean invariance survived, but only briefly, until the
Fitzgerald contraction of meter sticks and time dilation of clocks dealt it a
mortal blow.
Then, in 1905, Albert Einstein published his profound analysis of how
light signals are used in comparing time, distance, and mass measurements
made by observers in relative motion.6 Einstein concluded that physical
space is not the Euclidean space assumed by Newton. Realizing that the
invariance properties of physical measurements require the merging of space
and time into spacetime, Einstein replaced the separate concept of distance
between two points in space, and the concept of time interval between two
instants, by a single concept: the spacetime interval between two events.
As the spacetime interval is not a four-dimensional generalization of the
Euclidean distance in ordinary space, geometric relations between events in
spacetime are non-Euclidean.
These discoveries have, of course, required major modifications in the
laws and equations of physics — modifications that continue to occupy
physicists to this day. Here we wish only to call attention to a conceptual
October 28, 2010 16:55 9in x 6in b943-ch01

4 Dynamical Symmetry

(a) Unit Cube, Vz = 0 (b) Unit Cube, Vz =

2c
3

Fig. 1.1 A unit cube viewed by an observer in its rest frame and by an oberserver
moving with two thirds the velocity of light in its rest frame.

problem raised by the experimentally established validity of the theory of

special relativity. (We reach ahead into Sec. 2.10 to do so.) Figure 1.1a
depicts a cubic crystal in the spatial coordinate system of an observer at
rest with respect to the crystal. Figure 1.1b depicts the same crystal as
seen by a second observer moving relative to the first with a velocity 87% of
that of light. If both observers are to agree on the symmetry of the crystal,
their concepts of physical symmetry must be altered: standard concepts of
physical symmetry are incompatible with the standard interpretations of
relativistic effects.
But that is not all.
Eulenbeck and Goudsmit’s discovery of electron spin in 1925 provided
a hint that, even with the modifications required by relativity, accepted
concepts of symmetry might be inadequate for an understanding of phys-
ical phenomena.7 From a physical standpoint, the conceptual problems
were dealt with by allowing the magnetic moment vector of individual
electrons to express a symmetry not previously recognized in nature. As
explained in Sec. 2.5 this symmetry cannot exist everywhere in ordinary
three-dimensional Euclidean space.
In the decade following the invention of quantum mechanics in
1926, there appeared a further hint that accepted concepts of physical
symmetry might require further modification. Shortly after the inception of
quantum theory it was realized that quantum numbers could have a dual
interpretation: they were found to fix both the eigenvalues of observable
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 5

quantities and the symmetry properties of wavefunctions. The quantum

number m, of atomic spectroscopy, fixes both the component of an atom’s
angular momentum, m(h/2π), on some chosen axis, and the rotational sym-
metry of its wavefunction about that axis. Knowing the azimuthal quantum
number l, one knows both the square of the total angular momentum of the
atom, l (l + 1)(h/2π)2 , and the allowed values of m. Symmetry-based argu-
ments require that m can only take on the values −l, −l + 1, . . . , l − 1, l.
For a given value of l there are thus 2l + 1 wavefunctions with different
rotational symmetries about the chosen axis, and different angular momen-
tum components, m, along this axis. The connection between symmetries
and observables in quantum mechanics is so unique and so direct that the
connection is a major theme in Herman Weyl’s 1931 monograph, Group
Theory and Quantum Mechanics.8
By 1931, physicists had begun to thoroughly develop the consequences
of the fact that the potential energy of interaction of two charged particles
depends only upon the distance between them. In a one-electron atom the
electrostatic potential energy is the same as that of an electron in the field of
a nucleus, so it depends only on the radial distance between the electron and
nucleus. In a frame of reference centered on the nucleus, the electrostatic
environment in which the electron moves is invariant under rotations. As
the kinetic energy of the electron is also invariant under rotations, all states
having the same radial distribution for the electron must have the same
energy, provided one neglects the small energies associated with spin–orbit
coupling. Observable states with the same energy and radial distribution
need not have definite angular momentum, but they must be interconverted
by rotations. As noted above, arguments of symmetry require that the
number of these states is 2l + 1, where the azimuthal quantum number
l can have values 0, 1, 2 . . . . Spherical symmetry requires that the energy
levels of hydrogen, and other one-electron atoms, be (2l+1)-fold degenerate.
They are not.
As Pauli, Schrödinger, and Heisenberg found, Bohr’s expression En =
(Ze/n)2
− 2a0 for the energy of one-electron atoms with nuclear charge Ze applies
to all n2 eigenstates of principal quantum number n. The levels with energy
En are n2 -fold degenerate, and n is always greater than l.
By the early 1930’s it was realized that the spatial symmetry of a sys-
tem often completely accounted for any degeneracy of its energy levels. In
cases where spatial symmetry did not account for the degeneracy, it became
customary to speak of accidental degeneracy. The equality of the energy of
a 2s state with the energies of the 2p states of the hydrogen atom was an
October 28, 2010 16:55 9in x 6in b943-ch01

6 Dynamical Symmetry

accident as was the degeneracy of the 3s, 3p and 3d states (and so on). It is
natural to suspect that all these accidents afflicting every hydrogen atom
in the universe might have a cause.

1.2 Fock’s Discovery

In 1935 Vladimir Fock uncovered the cause: the electron in a one-electron
atom moves as though it were in an environment with the symmetry of a
hypersphere in four-space.9 Fock found a set of transformations which con-
vert wave functions of all bound states of the hydrogen atom into (hyper)-
spherical harmonics Ynlm in a four-dimensional Euclidean space. States of
the same n, but differing l, can be interconverted by rotations in this four-
dimensional space. The effective symmetry of the atom is that of an object,
a hypersphere, described by an equation such as

y12 + y22 + y32 + y42 = R2 , (1.2.1)

in which R is a real constant, and the y’s are Cartesian coordinates of a

point. It is this hyperspherical symmetry which causes all states of the same
principal quantum number n to be degenerate and requires them to be n2
in number.
In Sec. 1.4 below we will identify the space in which Fock’s hypersphere
exists. Fock’s work will be dealt with in detail in subsequent chapters. Here
we wish only to call attention to one of its key features. In his analysis,
Fock used a stereographic projection that connects points on a hypersphere
in a four-dimensional Euclidean space to points in a three-dimensional
Euclidean space. Figures 1.2 illustrate the analogous projection connect-
ing points on a sphere in three-dimensional space to points in its equatorial
plane. They illustrate the mapping of points from the northern and south-
ern hemispheres onto the equitorial plane. The other figures depict the
stereographic projection of spherical harmonics with l = 1. The projections
of the Px function and the Pz function onto the plane yield functions that
correspond to degenerate p and s type wave functions of the planar analog
of a hydrogen atom.10 The relationship between the projected functions
and the spherical harmonics is directly analogous to the relationship Fock
established for real hydrogen atoms.
Note that rotating the Px spherical harmonic through 90◦ around the z
axis converts it into a Py harmonic, and carries out the same rotation of its
projection in the plane. In contrast, the Pz harmonic, and its projection,
are invariant under rotations about the z axis. No rotation in the plane can
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 7

(a) (b)

(c) (d)

Fig. 1.2 (a) Stereographic projection of point on upper hemisphere. (b) Stereo-
graphic projection of point on lower hemisphere. (c) Stereographic projection of
circles of constant latitude onto an equatorial plane. (d) Stereographic projection
obtained when sphere in Fig. 1.2(c) is rotated 90◦ about the y axis.

interconvert the p and s type functions. However, they are indirectly inter-
converted by rotations in the three-space that carry a Px or Py function into
a Pz function. Such a rotation carries a p type function in the plane into an
sp type hybrid function, and then into a pure s type function. This illus-
trates a key feature underlying Fock’s analysis: stereographic projections
can interconvert functions not related by rotations in three-dimensional
Euclidean space, to functions related by rotations in a higher dimensional
Euclidean space.
Fock’s use of a stereographic projection to explain the degeneracy of
the hydrogen atom is subject to an evident objection: the projection could
be said to produce, rather than uncover, a symmetry. However, Bargmann
quickly gave a Lie-algebraic treatment which established the hyperspherical
symmetry without the use of any coordinate transformations.11 The work of
Fock and Bargmann provides an impressive demonstration that degenera-
cies can be due to symmetries that are definitely not symmetries in ordinary
space. For natural scientists it was an early hint that natural phenomena can
October 28, 2010 16:55 9in x 6in b943-ch01

8 Dynamical Symmetry

be governed by symmetries that are not at all self-evident. It is now realized

that such symmetries can govern a wide range of natural phenomena.

1.3 Keplerian Symmetry

Hyperspherical symmetry is a property of the dynamics of all Keplerian
motion — motion of a particle under an inverse square law force when
the speed of the motion is a negligible fraction of the speed of light. In
the absence of relativistic effects and tidal effects a single spherical planet
moving about a star also moves as though it were in an environment with
hyperspherical symmetry. As a consequence, planetary orbits of different
ellipticity can have the same energy.
Because the Bohr–Sommerfeld model of the atom inherits this hyper-
spherical symmetry from classical mechanics, it can correctly determine
the degeneracy and n-dependence of the spectrum of the hydrogen atom,
despite the fact that Bohr’s atom is flatter than a bicycle wheel.
The physical and chemical properties summarized in periodic charts
are determined by the reduction of the hyperspherical symmetry to ordi-
nary symmetry to ordinary spherical symmetry not by ordinary spherical
symmetry itself. The hyperspherical symmetry has a persistent effect in
organizing the energy levels of many electron atoms. It also affects the
organization of energy levels in molecules. In independent-electron mod-
els of polyatomics it produces level crossings that would not otherwise be
present. In real molecules, these level crossings causes inflections in poten-
tial energy curves, and as noted below, are instrumental in determining the
shapes of molecules. In short, the hyperspherical symmetry, by influencing
the organization of energy levels in atoms and molecules, affects much of
physics, chemistry, geology and biology.
The hyperspherical symmetry governing the motion of a single planet
has similar consequences; it does much to govern the response of the planet
to the perturbing influences of other planets and satellites.
How do these symmetries of objects in higher dimensional spaces get
into our three-dimensional world? How is that they govern motions in ordi-
nary three-dimensional Euclidean space? We begin now to address these
questions.

1.4 Dynamical Symmetry

The Hamiltonian expression whose value is the energy, E, of a particle
subject to gravitational or electrostatic force is
H = p2 /2m − K/|r|. (1.4.1)
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 9

The potential energy term, K/|r|, with K a constant, has ordinary rota-
tional symmetry; and the kinetic energy term, p2 /2m, with the mass, m,
constant, has ordinary rotational symmetry. This statement remains true in
Schrödinger quantum mechanics, where p is converted to −i
∇. The hyper-
spherical symmetry is not a symmetry of the potential energy term in the
Hamiltonian nor is it a symmetry of its kinetic energy term, it is a symmetry
of the entire Hamiltonian. Operations of a group of transformations inter-
convert the kinetic and potential energy terms of the Hamiltonian H while
leaving E, the value of H, unchanged. These operations transform the posi-
tion vector r to r = f(r, p), and the momentum vector p to p = g(r, p), so
as to leave only the entire expression p2 /2m−K/|r|, and hence E, invariant.
For a two-dimensional atom, the operation that interconverts the kinetic
energy and potential energy can be put in one-to-one correspondence with
the Px ↔ Pz rotations on the sphere of Fig. 1.2, i.e. the rotations that
interconvert the s type and p type projections in the plane.
The hyperspherical symmetry of Kepler Hamiltonians is a truly dynam-
ical symmetry — a symmetry present when there is motion. It is a symme-
try that exists only when motion is allowed. It is extensively investigated
in Chapter 8.
An analogous situation occurs with the harmonic oscillator Hamiltonian

H = p2 /2m + kr2 /2. (1.4.2)

Harmonic oscillators have energy degeneracies larger than expected. Their

energy is also unchanged by transformations of the form r → r = f(r, p),
and p → p = g(r, p). The shell model of nuclear structure based on
the Hamiltonian (1.4.2) develops extensive physical consequences of these
degeneracies and their removal by internuclear forces.12
These dynamical symmetries are symmetries of systems allowing
motions. The dynamical symmetry of a system may, or may not, be greater
than its evident geometrical symmetry. The phrase hidden symmetry is
sometimes used to signify the presence of dynamical symmetry greater than
ordinary geometrical symmetry. In general, no analysis of forces or poten-
tials alone will find dynamical symmetries. Methods for uncovering them
are discussed in Chapters 7 and 9.
In recent years it has been realized that in classical mechanics dynam-
ical symmetries are, in the simplest cases, geometrical symmetries in the
combined space of positions, q, and momenta, p, i.e. symmetries in the
phase space of the physicist. The observation, that the classical analog of
October 28, 2010 16:55 9in x 6in b943-ch01

10 Dynamical Symmetry

the hyperspherical symmetry of the hydrogen atom is a hyperspherical geo-

metric symmetry in phase space, was not at all evident in Fock’s paper, nor
in that of Bargmann.11 Such a connection was difficult to interpret at the
time, because the uncertainty principle states that one cannot know both
the position and the momentum of a particle with arbitrary accuracy. In
Chapter 9 a Lie-algebraic extension of Dirac’s correspondence principal is
stated and used to establish the relation between the dynamical symmetry
of quantum mechanical systems and that of the corresponding classical sys-
tems, which are geometrical symmetries in phase space. It enables one to
correctly exploit classical analogies when interpreting the dynamical sym-
metries in quantum mechanics.

1.5 Dynamical Symmetries Responsible for Degeneracies

and Their Physical Consequences
If no analysis of forces or potentials alone can expose the symmetry and
consequent degeneracy of the hydrogen atom, one would expect surprises to
continually emerge in attempts to uncover other organizing features in the
natural sciences, unless, of course, only atoms, nuclei, elementary particles,
or other comparatively simple systems are directly affected by dynamical
symmetries. Because most physical systems are much more complex, and
furthermore, exact or approximate spatial symmetries are rather rare in
nature, it has long been supposed that hidden symmetry is rarer still. How-
ever, direct physical evidence establishes that large systems can have large
degeneracy groups. In statistical mechanics, the entropy of a macroscopic
system is a measure of its degeneracy, for it is equal to Boltzmann’s con-
stant times the logarithm of the number of equally likely microscopic states
of the same energy. A great host of operations can redistribute the energy
among the constituents of the system, and the corresponding symmetry
groups are enormous. Dynamical symmetry is ubiquitous, but often well
hidden.
Whenever a quantum mechanical system is degenerate, an infinitesi-
mal perturbation can make a finite alteration in its wave function. This
has the physical consequence that minute forces and minute displacements
that involve a minute, even infinitesimal, amount of work can make finite,
even large, changes in the physical properties of initially degenerate sys-
tems. The connection between symmetries of a Hamiltonian and energy
degeneracies imparts additional importance to symmetries in the quantum
mechanics of bound states, for which the energy levels that differ in E are
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 11

separated by energy gaps. As a consequence, among the bound states only

the degenerate ones are exquisitely sensitive to perturbations. Degeneracies
not directly due to hyperspherical symmetry, and the subtleties and con-
sequences of their removal, also have many consequences in chemistry and
physics:
Forces between nucleons produce an average binding potential which
may be approximated by that in radially symmetric harmonic oscil-
lators. Removal of the accidental oscillator degeneracies organizes
much nuclear physics.13
In producing the resonance stabilization so ubiquitous in chemi-
cal compounds, at least two degenerate, or nearly degenerate, elec-
tronic wave functions represented by different valence-bond pictures
interact to produce wave functions, molecular orbitals, of lower
energy than either.14
In the superconductive materials dealt with in the theory of
Bardeen, Cooper, and Schrieffer, electrons with initially degener-
ate one-electron translational and spin wave functions have this
degeneracy removed, and a new degeneracy is produced by minute
velocity-dependent forces stemming from interactions between the
electrons and vibrating atomic nuclei.15
Minute forces which act on nearly degenerate translational wave
functions can also develop other coordinated, gross, electronic
motions in solids, motions such as Davydov excitons16 and charge-
density waves.17
Sudden polarization18 occurs in molecules when nuclear motions
cause electronic energy levels to cross, producing degeneracies that
allow small molecular asymmetries to produce very large polariza-
tions.
Iterative algorithms that approximate wave functions of mole-
cules and solids may fail to converge for some assumed nuclear
configurations. (This quite often occurs in Hartree–Fock calcula-
tions.) In attempting to choose an improved approximate wave
function, the computer cycles between several different electronic
wave functions whose energy cannot be distinguished. Calculations
on cuprate lattices exhibit this effect, and even Cu(OH2 )3+ has
states whose degeneracy is not due to spatial symmetry.19 Fail-
ure of Hartree–Fock calculations to converge can provide a very
useful signal that unexpected physical or chemical phenomena are
waiting to be discovered. Hartree–Fock symmetry breaking occurs
October 28, 2010 16:55 9in x 6in b943-ch01

12 Dynamical Symmetry

when such degeneracy leads to one or more nuclear configurations

with less energy and less symmetry in three-space than the initial
nuclear configuration.20
Approximate dynamical symmetries of rotating and vibrat-
ing molecules have been found to organize their rotational and
vibrational energy states.21 This has made it possible to express
rovibrational energies as functions of far fewer parameters than
spectroscopists had previously thought possible, cf. Chapter 13.

1.6 Dynamical Symmetries When Energies Can Vary

In the late 1960’s it was found that one-electron atoms and their gravi-
tational analogs possess more symmetry than that of the hypersphere of
(1.2.1). When energy, E, is considered variable, Keplerian systems behave
as though they have the symmetry of the surface defined by the equation22

y12 + y22 + y32 + y42 − y52 − y62 = 0. (1.6.1)

Here, as before, the y’s are Cartesian cordinates of a point.

What is the physical interpretation of this symmetry? In classical Hamil-
tonian mechanics, the operations of a group that alters the energy of a
system express relationships between the position, momenta, energy, and
time variables of the system. The dynamical symmetries defined by the
invariance group of (1.6.1) are symmetries in the extended phase space of
positions, momenta, energy, and time, an eight-dimensional PQET space.
In quantum mechanics the symmetry of Eq. (1.6.1) becomes a symmetry
of the time-dependent Schrödinger equation of hydrogen-like atoms, and
governs their dynamics.23 Operations of this dynamical group can be used
to change energy states. The group can be a spectrum-generating group for
the atom. Chapters 9–13 deal with such dynamical symmetries of quantum
mechanical systems.
Dynamical symmetries of time-dependent Schrödinger equations, Hψ =
i∂ψ/∂t, may have surprising consequences even for simple systems. Because
of a dynamical symmetry of their time-dependent Schrödinger equation,
hydrogenic atoms obey surprising selection rules that preserve l as a good
quantum number.24 This, together with the related hyperspherical sym-
metry of the atom, has the consequence that when a nucleus of charge Z
is (conceptually) pulled apart to form the protomolecule termed a nearly-
united atom, there is no mixing of wave functions of different orbital angular
momentum quantum number l. If one does not recognize the presence of a
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 13

hidden symmetry, the united atom–nearly united atom energy level corre-
lation diagrams may appear to violate the noncrossing rule of Wigner and
von Neumann.25 As shown in Chapter 12, this hidden symmetry ensures
that nearly-united atoms obey the same rules that govern the geometrical
symmetry of actual triatomic molecules.26 These rules are due to Walsh,
who first pointed out that the ordinary spatial symmetry of triatomic and
tetraatomic molecules can be anticipated by simply counting the number
of valence electrons in the molecules.27
As the distance between atoms varies during the course of chemical reac-
tions, degeneracies, or near degeneracies, are often produced. Though often
considered accidental, many can be predicted from energy-level correlation
diagrams. Profound and rapid changes may take place at such crossings.
This happens, for example, when a neutral alkali atom approaches a neutral
halogen atom from afar. An electron jumps from the metal to the halogen
during a change of distance on the order of 10−2 angstroms.28
Finally, we would note that even when the entropy and other equilibrium
properties of highly degenerate, weakly interacting, systems are unaffected
by weak interactions, these interactions can govern energy flow between
subsystems. The resulting dynamical symmetry breaking then becomes
relevant in nonequilibrium statistical mechanics.

1.7 The Need for Critical Reexamination of Concepts

of Physical Symmetry. Lie’s Discoveries
The observations made in the preceding paragraphs illustrate that simple
relations in complicated systems may be overlooked by familiar modes of
thought. Surprises develop for several reasons:
First of all, even in simple classical systems, dynamical symmetries
are symmetries in a phase space of many dimensions.
Secondly, phase space is not a Euclidean space. For this rea-
son, the concepts of physical symmetry appropriate to classical and
quantum mechanics, the concepts of dynamical symmetry, are pro-
foundly different than the usual concepts appropriate to Euclidean
space. Chapter 2 begins our discussion of the difference.
Thirdly, symmetries in phase space have no immediate obvious
interpretation as symmetries in quantum mechanics. Nevertheless,
as previously noted, dynamical symmetries in Schrödinger mechan-
ics and Hamiltonian mechanics are closely related, and the relation-
ship is established in Chapter 9.
October 28, 2010 16:55 9in x 6in b943-ch01

14 Dynamical Symmetry

Dynamical symmetries are, in general, so different from ordinary

symmetries that understanding them requires a critical reexamina-
tion of the concept of symmetry itself.

The beginnings of such a critical analysis originated in the mind of

the mathematician Sophus Lie, apparently during the winter of 1871–2.29
As noted in the Appendix at the end of this chapter, Lie was attempt-
ing to develop a unified theory of solution for ordinary differential equa-
tions. He discovered that the solutions of differential equations can be both
defined by, and interconverted by, operations of continuous groups of trans-
formations. Lie’s groups are termed continuous because their operations are
labeled by continuously variable parameters, for example the angles of rota-
tion. As will be seen in Chapters 4–6 of this book, differential equations have
group structure simply because

b
c
c
df + df = df. (1.7.1)
a b a

This observation, though at first sight trivial, profoundly connects the study
of differential equations to studies of groups of transformations. In fact, Lie
used first-order ordinary differential equations, and sets of first-order partial
differential equations, to define continuous groups. He devoted much of his
life to developing the extensive mathematical implications of the connection
between differential equations and continuous groups. In doing so, he gave
us much of what is now called the theory of Lie groups.
The operations that convert solutions of a differential equation into
themselves and into other solutions, define the invariance group of the dif-
ferential equation. These groups are primarily Lie groups, and they can
be of great utility. In fact, most of the differential equations of the sci-
ences have Lie group structure which neatly organizes and simplifies their
solution. Unfortunately, for much of the past century the utility of the con-
nection between Lie groups and differential equations went unrecognized,
and was seldom exploited by mathematicians interested in solving differ-
ential equations. Several monographs that are outstanding exceptions to
this statement have been listed at the end of the Preface. In this book, the
group structure of the equations of dynamics is used to relate dynamics to
geometry by relating dynamical symmetries to geometrical symmetries.
The relevance of Lie’s work to the natural sciences is direct and fun-
damental. Whenever a continuous succession of causes and effects governs
natural phenomena, it becomes possible to formulate differential equations
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 15

which govern the phenomena. Approximate or exact integration of the equa-

tions then produces solutions that can correlate and predict the results of
observations. And whenever differential equations describe nature, their
group structure describes relationships amongst natural phenomena. As
earlier sections of this chapter have pointed out, when the symmetries are
unrecognized, the relationships they imply may lie unrecognized — even
when the solutions are static in time, and some of the natural phenomenon
are known and much studied. An example is provided by the history of
superconductivity. Onnes discovered the phenomenon in 1911.30a Despite
intensive investigations carried out by brilliant investigators, it was not
until 1957, 31 years after the advent of quantum mechanics, that Bardeen,
Cooper, and Schreifer found the cause of superconductivity.30b As previ-
ously mentioned they thereby uncover a symmetry breaking that creates a
new symmetry.
In evolving systems, the consequences of group structure can also be pro-
found and the causes well hidden. Quantum mechanics provides many such
examples. The wave functions of quantum mechanical systems, evolving in
accord with equations such as the time-dependent Schrödinger equation,
may smoothly lead to degenerate states utterly unstable to perturbations.
At such points, symmetries may suddenly be enlarged, decreased, or sim-
ply changed. This behavior illustrates a general property of nature: stable
states may smoothly evolve into states so unstable that minute perturbations
can produce gross changes in the state of the system. A discontinuously pro-
duced state may then result, and become the initial state of a new continuous
causal chain. A few examples follow:
The temperature and pressure of a gas reach critical values and
crystals form: a system with continuous translational and rota-
tional symmetry suddenly produces systems with discrete transla-
tional symmetry. If the crystals are metastable, subsequent phase
changes may then occur when they smoothly evolve into a state
whose energy coincides with that of a state with different crystallo-
graphic symmetry. What will be the symmetry of the initial crystal
lattice? Of subsequent ones? Current science makes astonishingly
few predictions.
Diffusion of the components of oscillatory chemical reactions
can develop chemical concentration waves in initially homogeneous
media.31,32 In a spherical container, standing concentration waves
with s, p, d, . . . type spatial symmetries can develop. Such reactions
make it possible for a fertilized egg to produce a set of identical
October 28, 2010 16:55 9in x 6in b943-ch01

16 Dynamical Symmetry

daughter cells which in a coordinated manner then produce fur-

ther daughter cells of different types. Subsequent intertwined chains
of continuous causality, punctuated by symmetry breaking events,
make it possible for “a sphere to become a horse”.32 Even at
the few-cell stage where reaction–diffusion phenomena are likely
to dominate cell differentiation, the biochemistry involved remains
wondrously obscure as this is written.
The differential equations governing fluid flow describe the causal
chains that produce smooth streamline flow, and allow for the pul-
sating, swirling, chaotic, eddies and vortex rings that may break
the translational symmetry of a flow. Investigations of the invari-
ance groups of the differential equations, and the symmetries they
imply, have produced powerful tools of analysis utilized by aerody-
namic engineers.33 But some phenomenon defy detailed prediction;
it is still a good idea to determine the stall speed of a new wing
using a wind tunnel.
A homogenous fluid contained in a vertical circular cylinder and
heated from below by a cylindrical heat source may develop insta-
bilities which lead to stable Benard convection cells. Their verti-
cal boundaries appear at unpredictable locations, and break the
initial cylindrical symmetry of the system.34 The differential equa-
tions which govern motions of the earth’s atmosphere develop insta-
bilities that limit the reliability of weather predictions. The very
complex equations involved are elegantly modeled in the much sim-
pler equations of Lorentz. They predict circumstances that produce
unpredicability.35
Geologists believe that oceanic rifts and plate tectonic motions
are produced by convection cells in the magma below the earth’s
crust. Even minute perturbations in a sphere filled with fluid and
centrally heated with perfect spherical symmetry can develop sym-
metry breaking convection cells.36 Though plate tectonics appar-
ently rests on the ultimately unpredictable, its sudden general
acceptance in 1967–8 has enabled today’s geologists to accept,
and greatly extend, earlier correlations of global geology.37 This
most recent of scientific revolutions has established an ever-growing
body of causal connections between geological events both past
and present. And this is being accomplished in a branch of natural
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 17

science in which it can be very difficult to distinguish between the

causal and the adventitious.38
These are but a few examples of how the continuous causal chains
that organize so much of science typically develop instabilities, that in
turn produce versions of the punctuated evolution of biologist Steven Jay
Gould.39 Previously expressed relationships are broken, and new relation-
ships develop.
In many of the examples just given, symmetries in ordinary Euclidean
space have changed. However, as we have emphasized, by no means all
physically relevant symmetry is symmetry in ordinary Euclidean space. As
will be seen in latter chapters, it is the differential equations of the physical
sciences that determine the physically relevant spaces.
Early in the twentieth century, Lie’s mathematical insights began to
stimulate new physical insights. We have already mentioned Noether’s work
of 1918. Yet even earlier, in 1909, Cunningham and Bateman40 showed
that the transformation group which interconverts solutions of Maxwell’s
equations directly requires that the spacetime that propagates light is non-
Euclidean and has the properties discovered by Einstein. In Chapter 14
it will be pointed out that the group has other, surprising, properties as
well. Though, as we have seen, relativity theory requires a rethinking of the
meaning of physical symmetry, historically it is the 1935 work of Fock and
Bargmann which first established that physically consequential symmetries
may be symmetries in spaces other than spacetime.
The dynamical symmetries that are the subject of this book arise
because the differential equations governing classical and quantum dynam-
ics have solutions that can be defined, time-evolved, and interconverted by,
operations of groups. The operations of a dynamical group have direct phys-
ical interpretation, and geometric interpretation as operations of a symme-
try group. The resulting connection between geometry and dynamics is the
organizing principle of the book. However, the connection is more general
than can be dealt with herein. The general situation is as follows:
The equations describing a great variety of physical systems are differ-
ential equations. All have invariance groups and an associated geometry
in which transformations that interconvert solutions of the equations can
be interpreted as geometric operations. The space in which these geometric
operations act is usually a space in which geometric symmetries are not
just those possible in ordinary space or spacetime. And, when differential
October 28, 2010 16:55 9in x 6in b943-ch01

18 Dynamical Symmetry

equations describe the natural world their geometric symmetries may, or

may not, be physically stable, they may persist, or be altered in ways made
more understandable by Lie’s work.
The next two chapters informally explore fundamental conceptual rela-
tionships between the invariance groups of equations and their geometric
and physical interpretations. To follow the resulting connection between
geometry and physics one must consistently conceive of symmetries in
spaces of higher dimension than three, and one must consistently conceive
of symmetries in spaces that are not necessarily Euclidean. These introduc-
tory chapters are intended to develop the required conceptions in a manner
that will enable readers to expand, and continue to exploit, their intuitive
geometric insights. Subsequent chapters more formally develop and apply
concepts introduced in Chapters 1–3.

Appendix A. Historical Note

Sophus Lie was born in the Norwegian village of Nordfordeid, and spent his
childhood there as a son of the local Lutheran pastor. He was thirty years
old when he began to develop his theory of differential equations, a theory
he saw as analogous to Galois’ theory41 of solution of algebraic equations.
He quickly established that a first-order ordinary differential equation can
be integrated directly, or with the aid of an integrating factor, if and only
if there is a continuous group of transformations that leaves the differential
equation invariant.42 Knowledge of the group determines whether an inte-
grating factor is necessary, and if it is, the group determines the integrating
factor (and vice versa). Subsequently Lie expanded his theory of first-order
equations into a unified theory of solution of sets of ordinary differential
equations of arbitrary order. Several texts on the subject appeared at the
beginning of the twentieth century.43 Even before his death from perni-
cious anemia in 1899, Lie’s ideas began to have an extraordinary influence
in pure mathematics. However, two world wars, practical difficulties, and
several historical accidents delayed the use of Lie’s work by applied mathe-
maticians for some fifty years. There also appeared to be theoretical imped-
iments limiting extensions of Lie’s ideas to partial differential equations.44
The theoretical impediments were found to be largely illusory,45 and it is
now known that partial differential equations possess much more group
structure than was previously realized.
During the 1940’s Bargmann46 and Wigner47 developed the represen-
tation theory of the Poincare group that underpins relativistic quantum
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 19

kinematics. In the mid 1950’s attempts to generalize Gell-Mann’s eight-fold

way48 led physicists to a renewed interest in the symmetry of the hydrogen
atom and to an interest in Lie groups in general. The 1960’s and 70’s saw
the recognition of the dynamical symmetries of a number of systems gov-
erned by Schrödinger equations. These symmetries are discussed in several
chapters of this book. The final chapter of the book is devoted to symme-
tries of Maxwell’s equations. Special attention is given to the consequences
of inversion symmetry that enlarges the Poincare group.49
By the turn of the twentieth century, the development of symbolic
computation algorithms, and elegant applied mathematics due to Reid,50
removed much of the practical difficulty in applying Lie’s ideas. Computer
programs now make it possible to systematically uncover the symmetries
of increasingly complex systems governed by differential equations. And
studies of the symmetries of partial differential equations have led to new
methods for obtaining their solutions.42 Kumei51 initiated the use of Lie
methods to obtain solutions of nonlinear partial differential equations, a
field in which they are proving particularly useful.
The connection between dynamics and geometry that is emphasized in
this book has impressive antecedents. For example, Synge in his article
on dynamics in Handbuch der Physik long ago pointed out that Hamilto-
nian dynamics allows a complete geometrization of classical mechanics.52
This conception of a geometrodynamics, applied via tensor analysis rather
than Lie theory, is extensively developed in Meisner, Thorn, and Wheeler’s
monograph on general relativity.53

References
[1] J. Kepler, Tertius Interveniens, Gesammelte Werke, edited by W. van Dycke,
M. Caspar; translated by A. Koestler, The Watershed (Doubleday, Garden
City, N.Y. 1960), pp. 65, 226.
[2] J. Kepler, Mysterium Cosmographicum (Tuebingen, 1596).
[3] E. Noether, Nachr. Koenig. Gesell. Wissen. Goettigen, Math.-Phys. Kl 235
(1918).
[4] J. C. Maxwell, Phil. Mag., I, 48 (1860); cf. The Scientific Papers of James
Clerk Maxwell, ed. W. D. Niven, Vol. 1, p. 379 (Librairie Scientific J. Her-
mann, Paris, 1890).
[5] a) J. C. Maxwell, Cambridge Phil. Soc. Trans., X, pt.1, (1855–6) Scientific
Papers, 155.
b) J. C. Maxwell, Phil. Trans. Roy. Soc. (London) CLV (1864) Scientific
Papers, 526.
[6] a) A. Einstein, Ann. der Physik 17, 891 (1905); translated by R. W. Lawson,
Relativity, The Special and General Theory, (Henry Holt, New York, 1921).
October 28, 2010 16:55 9in x 6in b943-ch01

20 Dynamical Symmetry

b) A. S. Eddington, Space Time and Gravitation (Cambridge University

Press, 1920).
[7] G. E. Euhlenbeck, S. Goudsmit, Naturwiss. 13 (1925) 13.
[8] H. Weyl, Gruppenntheorie und Quantenmechanik, 1930; cf. Hermann Weyl,
Theory of Groups and Quantum Mechanics, translated by (H. P. Robertson,
Dover, N.Y., 1950).
[9] V. Fock, Z. Physik, 98 (1936) 145; translated by V. Griffing, J. Phys. Chem.
61, (1957) 11.
[10] T. Shibuya, C. E. Wulfman, Am. J. Phys. 33, (1965) 376.
[11] V. Bargmann, Z. Physik, 99, (1936) 576.
[12] M. Goeppert Mayer, J. H. D. Jensen, Elementary Theory of Nuclear Shell
Structure (Wiley, N.Y., 1955).
[13] a) G. A. Baker, Jr, Phys. Rev. 103, (1956) 1119.
b) J. P. Elliott, in Topics in Atomic and Nuclear Theory (University of
Canterbury, Christchurch, New Zealand, 1970); deals with the breaking of
the degeneracy group of isotropic oscillators.
[14] C. A. Coulson, Valence (Clarendon Press, Oxford, 1952), Ch. VI.
[15] J. R. Schrieffer, Theory of Superconductivity, revised edn. (Benjamin, N.Y.,
1983).
[16] A. S. Davydov, Theory of Molecular Excitons (Plenum, N.Y., 1977); cf. also
Physica Scripta 20 (1979) 387.
[17] R. M. White and T. H. Geballe, Long Range Order in Solids (Academic
Press, N.Y., 1979); analyzes the effects of many symmetry breaking interac-
tions.
[18] a) C. E. Wulfman, S. Kumei, Science 172 (1971) 1061;
b) L. Salem, Isr. J. Chem. 12 (1979) 8;
c) C. E. Wulfman and G. C. Hyatt Proc. Indian Acad. Sci. (Chem. Sci.)
107, No. 6, (1995).
[19] C. E. Wulfman, Larry Curtis, Int. J. Mod. Phys. B. 3 (1989) 1287.
[20] a) J. Paldus, A. Veillard, Chem. Phys. Lett. 50 (1977) 6.
b) M. Ozaki, Prog. Theor. Phys. 67 (1982) 83.
c) J. L. Stuber, J. Paldus, in Fundamental World of Quantum Chemistry
(Kluver, Dordrecht, 2003), pp. 67–139.
[21] F. Iachello, R. D. Levine, Algebraic Theory of Molecules (Oxford University
Press, Oxford, 1995).
[22] a) E. C. G. Sudarshan, N. Mukunda, L. O’Raifertaigh, Phys. Rev. Lett. 15
(1965) 1041; ibid. 19 (1965) 322;
b) A. O. Barut, P. Budini, C. Fronsdal, Proc. Roy. Soc. (London) A291, 106
(1966);
c) H. Bacry, Nuovo Cimento 41 (1966) 223;
d) A. Bohm, Nuovo Cimento 43 (1966) 665;
e) A. O. Barut, H. Kleinert, Phys. Rev. 156 (1967) 1541;
f) M. Bednar, Ann. Phys. (NY) 75 (1973) 305.
[23] S. Kumei, M.Sc. Thesis, University of the Pacific, Stockton, CA, 1972; R. L.
Anderson, S. Kumei, C. E. Wulfman, Phys. Rev. Lett. 28 (1972) 988.
[24] a) S. Pasternack, R. M. Sternheimer, J. Math. Phys. 3 (1962) 1280;
October 28, 2010 16:55 9in x 6in b943-ch01

Introduction 21

b) N. V. V. J. Swami, R. G. Kulkarni, L. C. Biedenharn, J. Math. Phys. 11

(1970) 1165.
[25] J. von Neumann, E. Wigner, Physik Z. 30 (1929) 467.
[26] C. Wulfman, J. Chem. Phys. 31 (1959) 381.
[27] A. D. Walsh, Disc. Faraday Soc., Series 2, 18 (1947); J. Chem. Soc. (1953)
2260, 2266, 2278, 2296, 2300, 2306, 2318, 2325, 2330.
[28] R. S. Berry, J. Chem. Phys. 27 (1957) 1288.
[29] E. Strom, Marius Sophus Lie, in Proceedings of the 1992 Sophus Lie Memo-
rial Conference, eds. O. A. Laudal, B. Jahren, (Scandinavian University
Press, Oslo, 1994).
[30] a) H. K. Onnes, Comm. Phys. Lab. Univ. Leiden, 119, 120, 122 (1911);
b) J. Bardeen, L. N. Cooper, J. R. Schrieffer, Phys. Rev. 108 (1957)
1175.
[31] a) A. M. Weinberg, Growth, 2 (1938) 81;
b) A. M. Weinberg, Bull. Math. Biophysics 1 (1939) 19.
c) A. M. Zabotinsky, Biophysics, 9 (1965) 329; (translated from Of Biofizika
2 (1964) 306).
[32] A. M. Turing, Phil. Trans. Roy. Soc. London, B237 (1952) 37.
[33] B. J. Cantwell, Introduction to Symmetry Analysis (Cambridge, 2002).
[34] H. L. Swinney, J. P. Golub, Hydrodynamic Instabilities and the Transition
to Turbulence, Springer-Verlag, N.Y. (1981).
[35] a) E. Lorenz, J. Atmospheric Sci. 20 (1963) 130–141;
b) V. I. Arnol’d, Catastrophe Theory; translated by G. S. Wassermann, R.
K. Thomas, (Springer-Verlag, Berlin, 1984, p. 23);
c) D. Ruelle, Strange Attractors, Mathematical Intelligencer, 2, no. 3 (1980)
126.
[36] F. H. Busse, Ann. Rev. Fluid. Mech. 32 (2000) 383.
[37] a) A. Wegener, Die Entstelbung der Kontinent und Ozeane (1915); cf. The
Origin of the Continents and Oceans, a translation of the 4-th edition by J.
Biram (Methuen, London);
b) A. Du Toit, Our Wandering Continents (Oliver and Boyd, London, 1937);
c) S. Warren Carey et al., Continental Drift, A Symposium, (University of
Tasmania, Hobart, 1956).
[38] D. Yuen, Chaotic Processes in the Geological Sciences, (Springer-Verlag,
1992).
[39] S. Jay Gould, The Structure of Evolutionary Theory (Harvard U. Press,
Cambridge, Massachusetts, 2002a). (Punctuated evolution is seen as a con-
sequence of punctuated biological equilibria.)
[40] a) E. Cunningham, Proc. Math. Soc. (London) Ser. 2 8 (1910) 77;
b) H. Bateman, ibid. pp. 228, 469.
[41] F. Cajori, Theory of Equations (Macmillan, N.Y., 1928), Chapters XIII–
XVI.
[42] a) G. W. Bluman, S. Kumei, Symmetries and Differential Equations
(Springer, 1989);
b) Brian J. Cantwell, Introduction to Symmetry Analysis (Cambridge U.
Press, 2002).
October 28, 2010 16:55 9in x 6in b943-ch01

22 Dynamical Symmetry

[43] a) J. M. Page, Ordinary Differential Equations with an Introduction to Lie’s

Theory of the Group of One Parameter (Macmillan, London, 1897);
b) A. Cohen, An Introduction To The Lie Theory of One-parameter Groups
(1911); reprinted by G. E. Stechert, New York, 1931.
[44] E. A. Muller, K. Matschat, Miszellanen der Angewandten Mechanik (Berlin,
1962), p. 190.
[45] a) S. Kumei, M.Sc. Thesis, University of the Pacific (Stockton, CA, 1972);
b) N. H. Ibragimov, R. L. Anderson, Dokl. Akad. Nauk SSSR 227 (1976)
539.
[46] V. Bargmann, Ann. of Math. 48 (1947) 568.
[47] E. Wigner, Ann. of Math. 40 (1939) 149.
[48] M. Gell Mann, Phys. Rev. 125 (1962) 1067.
[49] E. Cunningham, Proc. Math. Soc. (London) Ser. 2. 8 (1909) 77.
[50] G. J. Reid, J. Phys. A, 23 (1990) 2853; Eur. J. Appl. Math. 2 (1991) 293,
319.
[51] S. Kumei, J. Math. Phys. 16 (1975) 2461; ibid. 18 (1977) 256.
[52] J. L. Synge, Classical Dynamics, in Encyclopedia of Physics, Vol. III/1,
S. Flugge (Springer, Berlin, 1960).
[53] C. W. Meisner, K. S. Thorn, J. A. Wheeler, Gravitation (W.H. Freeman,
San Francisco, 1971).
October 28, 2010 16:55 9in x 6in b943-ch02

CHAPTER 2

Physical Symmetry and Geometrical Symmetry

In Newtonian physics it is supposed that r, dr and d(dr) are approximable

by measurements of r, ∆r and ∆(∆r) respectively in a local Cartesian ref-
erence frame in Euclidean space. Because forces are, in principle, defined
via measurements of displacements, Newtonian mechanics inherits conse-
quences of the presuppositions (axioms) of Euclidean geometry. Newton also
presupposed that the time, t, and its derivatives dt, d(dt), can be approx-
imated by measurements of ∆t and ∆(∆t) using local clocks. Quantum
mechanics inherits these presuppositions about space and time measure-
ments in a restricted way, under the supposition that the position vector
r of a particle of fixed mass and its velocity vector dr/dt cannot both be
approximated by increasingly refined measurements. However, the corre-
spondence principle enables one to carry over into quantum mechanics some
of the geometric intuition one develops from studying Euclidean geometry.
As suggested in the previous chapter, understanding by concepts of
symmetry the manner in which natural phenomena are interrelated might
require a reconception of relationships between geometry and measurements
of r and ∆r, and also t and ∆t. The rethinking required goes beyond that
suggested by the remarks in Chapter 1. With this in mind we shall proceed
with more care than what might at times seem necessary.

2.1 Geometrical Interpretation of the Invariance Group

of an Equation; Symmetry Groups
The equation G(z) = c, where c is a constant, is by definition invariant
under a transformation z → z
= f(z) if and only if G(z
) = c for all
values of z that satisfy the original equation. If substituting of z
for z
carries all solutions of an equation into solutions of that equation, it defines
an invariance transformation of the equation. Conversely, a transformation

23
October 28, 2010 16:55 9in x 6in b943-ch02

24 Dynamical Symmetry

z → z
= f(z) that leaves the equation invariant is a transformation that

converts solutions of the equation to solutions of the equation.
Let us investigate some common assumptions on geometric interpreta-
tion of invariance transformations by considering the invariance transfor-
mations of an algebraic equation. Consider, for example, the equation

w4 = 1. (2.1.1)

It is invariant under the transformation w → w
= f(w) if and only if

4
w
4 = 1, i.e. (f(w)) = 1, (2.1.2)

for all values of w satisfying (2.1.1). The substitution w → −w leaves w4

unchanged and transforms a solution of Eq. (2.1.1) into a solution. The
substitution which carries w → iw has the same property. Compounding
these two substitutions carries w → −w → −iw. As w = 1 is a solution
of (2.1.1), it follows that −1, i and −i, are also solutions. Figure 2.1, in
which w = x + iy, represents the four solutions by points with coordinates
(x, y) in a plane. For clarity, the points are connected by straight lines.
Though four points are sufficient to represent the solutions, more than four
geometric operations carry solutions into solutions — all such operations,

(0,1)

(− 1,0) (1,0)

(0,− 1)

Fig. 2.1 A plot of the roots of Eq. (2.1).

October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 25

Table 2.1. Operations that carry the square into itself and
interconvert solutions of w4 = 1, w = x + iy = 1, i, −1, −i.

A. Rotations:
R(0, 1) = (1, 0), R(1, 0) = (0, −1),
R(0, −1) = (−1, 0), R(−1, 0) = (0, 1).
B. Reflections:
S1 (x, y) = (x, −y), S2 (x, y) = (−x, y).
C. Inversions:
J1 (x, y) = (y, x), J2 (x, y) = (−y, −x).

and their action on a coordinate vector r = (x, y), are listed in Table 2.1.
Each operation corresponds to a substitution that leaves (2.1.1) invariant.
We wish to determine whether these operations give rise to a group
of operations. A set of operations comprise a group of operations iff the
following hold:
(a) The application of any operation Ta , followed by the application of any
operation Tb , is an operation in the set, say Tc . Symbolically,
Tb Ta = Tc . (2.1.3a)
(b) The set contains the identity operation, I, which satisfies
Ta I = I Ta = Ta . (2.1.3b)
(c) Every operation Ta has an inverse Ta
in the set:
Ta Ta
= I = Ta
Ta . (2.1.3c)
(d) The operations obey the associative law
Ta (Tb Tc ) = (Ta Tb ) Tc . (2.1.3d)
Here, the inverse of a compound operation (Ta Tb ) is Tb
Ta
.
The operations of Table 2.1 give rise to the operations A’ in the first
column of Table 2.2. Their action on the operators A of the top row of the
table is indicated at the intersection of the corresponding rows and columns.
With the aid of the table the reader may verify that, taken together, the
operations are those of a group. Note that S1 S1 = I, S2 S2 = I, J1 J1 = I
and J2 J2 = I. It follows that each of these operations is its own inverse,
e.g. Sinv
1 = S1 . Consequently the operations S1 and I are themselves those
of a group. As this group is contained in the larger group, it is termed a
October 28, 2010 16:55 9in x 6in b943-ch02

26 Dynamical Symmetry

Table 2.2. Composition of the Operations A
and

A to give A
A.

A
A\ A I R R2 R3 S1 S2 J1 J2
I I R R2 R3 S1 S2 J1 J2
R R R2 R3 I J1 J2 S2 S1
R2 R2 R3 I R S2 S1 J2 J1
R3 R3 I R R2 J2 J1 S1 S2
S1 S1 J2 S2 J1 I R2 R R3
S2 S2 J1 S1 J1 I R2 R R3
J1 J1 S1 J2 S2 R3 R I R2
J2 J2 S2 J1 S1 R R3 R2 I

subgroup of the later. Each of the operations of reflection and inversion are
members of a subgroup of the larger group. Because R4 = I, R3 so R = I,
R3 is the inverse of R. Similarly, R2 is its own inverse. These rotations also
comprise a subgroup of the full group.
Because the operations carry solutions of Eq. (2.1.1) into solutions of
the same equation, it is left unchanged, i.e. it is invariant under the opera-
tion of the group. Given our geometric assumptions, the transformations in
Table 2.2 are also those of the symmetry group of the square of Fig. 2.1.
There are geometrical operations that do not leave Eq. (2.1.1) in-
variant, but do leave the symmetry of the square unchanged. These include
arbitrary rotations of the square about any point, translations of the square
in any direction, and the uniform dilatations, which enlarge or shrink the
square.
Rotations about the origin carry a point with coordinates (x, y) to the
point with coordinates (x
, y
), where

x
= x cos(θ) − y sin(θ), y
= x sin(θ) + y cos(θ), (2.1.4)

and θ is the angle of rotation. Operations of the group will carry a circle
with center at the origin into itself. The group is termed a continuous group
because the operations of the group depend upon a parameter, θ, that can
vary continuously. The rotations that carry the square of Fig. 2.1 into itself
comprise a discrete subgroup of this continuous group of rotations.
Translations carry a point with coordinate (x, y) to the points at (x
, y
),
where

x
= x + α, y
= y + β. (2.1.5)

These transformations also comprise a continuous group.

October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 27

Composing the operations of (2.1.4) and (2.1.5) one obtains E(2), the
Euclidean group of the plane. If we let T(α)x represent the operation that
converts x to x + α, the application of T(α1 ) followed by T(α2 ) is a trans-
formation
T(α2 )T(α1 )x = T(α2 )(x + α1 ) = x + α1 + α2 = T(α3 ), (2.1.6)

with
α3 = α1 + α2 .
The group E(2) is termed a three-parameter group because its operations
depend on the parameter θ, the angle of rotation, the parameter α, the
distance a point is translated parallel to the x axis, and the parameter β,
the distance it is translated parallel to the y axis.
Note that the operations of E(2) do not alter the distances Dij separating
points i and j.
Uniform dilatations in the plane carry out the transformations
x → eµ x, y → eµ y, −∞ < µ < ∞. (2.1.7)
Together with the operations of E(2), they comprise the operations of a
continuous group termed the similitude group of the plane. The operations
of this four-parameter group may be used to superpose objects of the same
symmetry — bring them into coincidence; they are unable to bring objects
of different symmetry into coincidence.
Operations of the similitude group that do not carry the square into
itself, do however, leave its symmetry unchanged. They might be said to
alter the perspective of a viewer of the square, changing the relation of the
square to this viewer, by actively rotating, translating, or dilatating it —
altering its apparent size. Operations of the similitude group of the plane
do not alter the relative distances Dij separating points i and j, and Di
j

separating points i
and j
, because both are multiplied by eβ .
Now let us compare the effects of the discrete operations of inversion
and reflection applied to the vertices in Fig. 2.1.
The inversions through the origin transform r → −r. The same effect
can be obtained by rotating the vector r, with components x, y, through
180◦ . Inversions are operations contained in the similitude group.
However, the reflection x → −x, y → y, and the reflection y → −y,
x → x, may alter the symmetry of a “left-handed” or “right-handed” figure
in the plane, interconverting the one into the other. None of these reflec-
tions is an inversion. Also, reflections are not special cases of rotations.
October 28, 2010 16:55 9in x 6in b943-ch02

28 Dynamical Symmetry

Reflections, though they leave invariant the separation between points, can-
not be carried out by any operation of the continuous similitude group of
the plane because there is no continuous motion of the similitude group
that can move a point Px,y through all intermediate points P between Px,y
and P−x,y , or between Px,y and Px,−y . In short, it is impossible for the
similitude group acting in the plane to continuously deform a right-handed
object in the plane into a left-handed object.
A situation that is analogous to the Euclidean plane holds in three-
dimensional Euclidean space: operations of the corresponding continuous
similitude group may be used to rotate, translate, and enlarge or shrink
objects and thereby superpose geometrical objects of the same symmetry.
Extending the group by including the separate reflections, x → −x, y → −y,
z → −z, one obtains the group whose subgroups define symmetries in the
space. Again, the effect of the inversions r → −r can be obtained by opera-
tions of rotation and reflection. Right-handed and left-handed objects have
different symmetry, and reflections may interconvert them, i.e., alter their
symmetry. However, there is no continuous operation acting in the three-
dimensional space that will deform a right-handed object so as to convert
it to a left-handed object, while at the same time leaving the separation
between points in three-dimensional Euclidean space unchanged, uniformly
enlarged, or uniformly shrunk.
These observations on the symmetry of objects call attention to gen-
eral principles that are true in both two- and three-dimensional Euclidean
geometry:
1. The group defining the symmetry of any figure or lattice is a subgroup
of the group of all similitude and inversion transformations.
2. The symmetry of any figure or lattice is unaltered by any operation of
the continuous group of similitude operations acting on the entire figure
or lattice.
The human mind is able to recognize that two objects in space are
essentially identical if an operation of the similitude group of three-space
can convert the one into the other. Apparently our minds habitually and
rapidly rotate, translate, dilatate, and then compare memories of visual
images without us being conscious of this. When our minds impart such
structures in group theory, we may not notice that we have seen similar
objects from different perspectives.
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 29

2.2 On Geometric Interpretations of Equations

A number of arbitrary decisions were made in using Fig. 2.1 to give a geo-
metric meaning to the algebraic equation w4 = 1. These affect the geometric
interpretation of the operations that leave w4 = 1 invariant. In particular:

1. The real part of w was plotted on one axis, and the coefficient of its
imaginary part on the other. The coordinates of a point, (x, y), were
thus chosen to be real numbers.
2. The x and y axes were, by choice, made perpendicular to each other so as
to provide an orthogonal coordinate system. It is conceivable that there
might be relations between x and y which invalidate this supposition.
Gibbs phase diagrams provide an example of this.
3. Though four points do not necessarily lie in a Euclidean plane, the four
roots were plotted in a plane. One could have plotted them in any two-
dimensional space, such as that provided by the surface of a sphere
or a hyperboloid in three-dimensional Euclidean space. And on each
surface one could have chosen both x and y to be measured along the
axes of a variety of coordinate systems. Some examples are depicted in
Figs. 2.2. In each case one can imagine the points labeled a, b, c, d to

Fig. 2.2 Plots of roots of Eq. (2.1) on a hyperboloid (a) and on a sphere (b).
October 28, 2010 16:55 9in x 6in b943-ch02

30 Dynamical Symmetry

correspond to the solutions i, 1, −i, −1. Depending upon the physical (or
other) interpretations given to the roots, a plot on one of these surfaces
might be more appropriate than a plot in the plane.
4. One could have supposed the four points to be points in some space of
dimension greater than three.
5. We supposed a fixed coordinate system and moved points about it. The
relation between a point and a coordinate system is, however, a relative
one. We could just as well have considered the points to remain fixed and
carried out the inverse motion on the coordinate system. For example, we
rotated points by 90◦ about the point (0, 0) in the plane, but the resulting
relation of point to coordinate system is the same as that obtained by
rotating the coordinate system through −90◦ . We imagined uniformly
enlarging the square, but the resulting relation to the coordinate system
could have been obtained by uniformly shrinking units of measurement
on the axes. (Considering an operation to act on a geometrical figure
is the active interpretation of the operation; considering it to act on a
coordinate system is its passive interpretation.)
6. By plotting the roots of (2.1.1) in a Euclidean plane equipped with an
orthogonal coordinate system, we accepted a concept of measurement
in which the distance, ∆s, between a point with Cartesian coordinates
(x, y) and a point with Cartesian coordinates (x
, y
) is predetermined;
if ∆x = x
− x, and ∆y = y
− y are two sides of a right triangle with ∆s
as the hypotenuse, then ∆s satisfies

∆s2 = ∆x2 + ∆y2 . (2.2.1a)

This relation would not hold true on the curved surfaces indicated in
Figs. 2.2. The infinitesimal analog of (2.2.1a), obtained by allowing ∆x
and ∆y to become arbitrarily small but nonzero, is:

ds2 = dx2 + dy2 . (2.2.1b)

In the following sections some of the assumptions and observations in

the above list are examined in more detail.

2.3 Geometric Interpretations of Some Transformations

in the Euclidean Plane
In this section we begin by considering the relation between the active and
passive interpretations of the transformations of E(2).
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 31

Let a transformation T be defined by

x → x
= F(x, y), y → y
= G(x, y), (2.3.1a)

and let it have inverse, Tinv , defined by

x
→ x = Finv (x
, y
), y
→ y = Ginv (x
, y
). (2.3.1b)

The transformations we will consider in this section have the property that
the transformed variables may be inhomogeneous linear functions of the
original variables. That is, in (2.3.1a,b)

x
= a0 + a1 x + a2 y, y
= b0 + b1 x + b2 y,
(2.3.1c,d)
x = a
0 + a
1 x
+ a
2 y
, y = b
0 + b
1 x
+ b
2 y
.

Inhomogeneous linear transformations, together with homogeneous linear

transformations, are termed affine transformations. Not all affine transfor-
mations in the plane are operations of the similitude group of the plane.
Let us suppose that x,y and x
, y
are variables in a Euclidean plane with
Cartesian coordinates defined by orthogonal unit vectors xc = 1i, yc = 1j
satisfying the usual relations i
i = 1, j
j = 1, i
j = 0. If the transformation is
interpreted in the active sense, T may be considered to move points about
in a plane equipped with a single system of Cartesian coordinates. A point
is moved from (x, y) = xi + yj to (x
, y
) = x
i + y
j. Tinv is considered
to move the point back again. In Fig. 2.3.1a the transformation T is a
rotation, interpreted as an active transformation in which

x
= x cos(θ) − y sin(θ), y
= x sin(θ) + y cos(θ). (2.3.2a)

The point follows the indicated path as θ varies continuously from an inital
value of zero. For each value of θ the inverse transformation is obtained
by changing θ to −θ and applying the transformation to the final point.
That is,

x = x
cos(−θ) − y
sin(−θ), y = x
sin(−θ) + y
cos(−θ). (2.3.2b)

In Fig. 2.3.1b, the rotation has been followed by a translation.

Figure 2.3.1c depicts a uniform dilation of Fig. 2.3.1b by a factor of e1/2 .
In the passive interpretation of a rotation, the relation between the
point and the coordinate system in Fig. 2.3.1a is obtained by rotating the
coordinate system through an angle −θ. This is illustrated in Fig. 2.3.2a.
In it, the notation XX and YY denote the transformed axes X and Y. If we
October 28, 2010 16:55 9in x 6in b943-ch02

32 Dynamical Symmetry

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1

0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 1.0 1.2

(a) (b)

1.2

1.0

0.8

0.6

0.4

0.2

0.5 1.0 1.5 2.0

(c)

Fig. 2.3.1 Rotation followed by translation followed by dilatation.

Y YYY
YY YY

P P

X
XXX

XX XX

(a) (b)

Fig. 2.3.2 A passive rotation followed by a passive translation.

use the superscript c to denote points on the coordinate axes, the passive
interpretation of an active rotation through an angle θ is defined by
xxc = xc cos(−θ) − yc sin(−θ), yyc = xc sin(−θ) + yc cos(−θ).
(2.3.3a)
If xc , yc , are of unit length, then in vector notation, (2.3.3a) becomes
xxc = i cos(−θ) − j sin(−θ), yyc = i sin(−θ) + j cos(−θ). (2.3.3b)
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 33

It follows that xxc xxc = 1, yyc yyc = 1, xxc yyc = 0, so xxc and yyc are
orthogonal unit vectors, and we write

ii = i cos(−θ) − j sin(−θ), jj = i sin(−θ) + j cos(−θ). (2.3.3c)

Figure 2.3.2b illustrates the passive interpretation of the combined tran-

formations of Fig. 2.3.1b. Note the passive version of the transformations
of Fig. 2.3.1b is a transformation in which the final transformation, the
translation, is inverted first; then the inverse rotation is applied to the
result.
Figures 2.3.3 depicts the passive and the active interpretations of a
dilation. As the passive interpretation represents a change of units of length,
it would be confusing and inconsistent to superimpose Figs. 2.3.3a,b. On
the other hand, no confusion results from superimposing Fig. 2.3.3a and
the circle of Fig. 2.3.3c obtained by is active dilatation.
Finally, we consider an affine transformation which is not an operation
of the similitude group of the plane. It is a pseudo-rotation or hyperbolic
rotation, a transformation in which

x
= x cosh(γ) + y sinh(γ), y
= x sinh(γ) + y cosh(γ). (2.3.4a)

Its interpretation as an active transformation acting on a point is illus-

rated in Fig. 2.3.4a. The path illustrated is obtained by having γ run
continuously from 0 to 2/3. Figure 2.3.4b depicts the action of a hyper-
bolic rotation, with γ = 1/4, on a unit circle. The inverse of a transfor-
mation with any value of the parameter γ is obtained by changing that
value of γ to −γ in (2.3.4a), and applying the transformation to x
, y
,
rather than x, y:

x = x
cosh(−γ)+y
sinh(−γ), y = x
sinh(−γ)+y
cosh(−γ). (2.3.4b)

To obtain the passive interpretation of hyperbolic rotations we apply

the inverse transformation to the coordinate vectors xc and yc and obtain

xxc = xc cosh(−1/4) + yc sinh(−1/4),

(2.3.5a)
yyc = xc sinh(−1/4) + yc cosh(−1/4).
Figure 2.3.4c depicts the effect of this passive transformation on the unit
vectors i, j, and the corresponding alteration of a Cartesian coordinate
lattice, when γ = 1/4. Care must be exercised when interpolating the coor-
dinates of points in this system of coordinates because separations between
points with coordinates (x
1 , y1
) and (x
2 , y2
) do not obey the Pythagorean
October 28, 2010 16:55 9in x 6in b943-ch02

34 Dynamical Symmetry

1.0 2

0.5 1

−1.0 −0.5 0.5 1.0 −2 −1 1 2

−0.5 −1

−1.0 −2
(a) (b)

−2 −1 1 2

−1

−2

(c)

Fig. 2.3.3 Passive and active interpretations of a uniform diltation.

theorem. Though, ∆r2 = ∆x2 + ∆y2 , ∆r
2 = ∆x
2 + ∆y
2 . In fact, using
the relation cosh(γ)2 − sinh(γ)2 = 1 one finds it

∆s2 ≡ ∆x2 − ∆y2 = ∆x
2 − ∆y
2 , (2.3.6)

unchanged by pseudo-rotations.
When studying the illustrations of the previous figures one naturally
supposes the paper to be held fixed, and may unconsciously use this per-
ception in interpreting the transformations being illustrated. A geometric
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 35

Y
1.0

Y
2.0
0.5

1.5

1.0 X
− 1.0 − 0.5 0.5 1.0

0.5

− 0.5
0.0 X
0.0 0.5 1.0 1.5 2.0

− 1.0

(a) (b)
Action of Hyperbolic "Rotations"

1.0

0.5

− 1.0 − 0.5 0.5 1.0

− 0.5

− 1.0

(c)

Fig. 2.3.4 Transformation of Cartesian coordinates by a hyperbolic rotation.

interpretation of transformations which helps to avoid confusion may be

obtained by using a rule of transformation to relate points in one coor-
dinate system of choice to points in another freely chosen coordinate sys-
tem. If one wishes, one may consider the mapping to be from one space
to another, and one may consider the two spaces as surfaces in a higher
dimensional space. Points in one system are mapped into points in the other.
Figure 1.2 depicts a mapping that connects points in a two-dimensional
October 28, 2010 16:55 9in x 6in b943-ch02

36 Dynamical Symmetry

Euclidean space to points in another two-dimensional space. The plane and

this second space, the surface of a unit sphere, are considered to be imbed-
ded in a three-dimensional Euclidean space. In this case, the transformation,
a stereographic projection, relates the Cartesian coordinates (x, y, z) of a
point in the plane to the coordinates (θ, φ) of a point on the surface of the
sphere.
Let a transformation x → x
= F(x, y), y → y
= G(x, y) map points
with coordinates (x, y) in frame A to a points with coordinates (x
, y
)
in frame B. We shall require that the inverse transformation exists, and
express it by x
→ x = Finv (x
, y
), y
→ y = Ginv (x
, y
). This may be used
to develop the inverse mapping from frame B to frame A. As an example,
in Figs. 2.3.3, both Fig. 2.3.3b and Fig. 2.3.3c may be considered to
be mappings of the circle of Fig. 2.3.3a, Cartesian coordinates being used
in every case. Figure 2.3.4b may be considered to depict a mapping of
the unit circle from one Cartesian coordinate system to another, using the
mapping whose points have coordinates defined by (2.3.4a).
When interpreting a transformation as a geometric mapping, one neces-
sarily states the coordinate system used in each space, as well as the trans-
formation which relates points in one space to points in the other space.
This reduces the possible role hidden assumptions play when one gives geo-
metric and physical interpretations to a transformation. However, as we
will see in latter sections of this chapter, and in subsequent chapters, one
naturally comes equipped with further presuppositions. This must also be
recognized if one desires to understand the connection between dynamical
symmetries and observations.

2.4 The Group of Linear Transformations of Two Variables

Rotation and pseudo-rotation transformations are homogeneous linear
transformations, as are the dilation transformations x → x
= exp(β) x,
y → y
= exp(γ) y. Homogeneous linear transformations of two variables
x1 and x2 can be expressed as X
= MX, where X represents column
vector that is the transpose of the row vector (x1 , x2 ), X
is the trans-
pose of (x
1 , x
2 ) and M is a 2 × 2 matrix with elements m11 , m12 , m21 ,
m22 that may depend on parameters. The inhomogeneous transformations
arising from translations cannot be expressed in this same manner. Taken
together, translations and homogeneous linear transformations are those of
the affine group. The affine transformation
(x, y) → (x
, y
) = (x + a + bx + cy, y + d + ex + fy) (2.4.1)
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 37

converts the difference vectors

(∆x, ∆y) = (x2 , y2 ) − (x1 , y1 ) = (x2 − x1 , y2 − y1 )
to
(∆x
, ∆y
) = (∆x + b ∆x + c ∆y, ∆y + e ∆x + f ∆y). (2.4.2)
These are homogeneous linear transformations, independent of the param-
eters a and d. In matrix and vector notation (2.4.2) may be written
∆X
= M∆X. For the same reason one may write dX
= MdX.
Altogether, the rotation, pseudo-rotation, dilation, reflection, and inver-
sion transformations, form a group of transformations, the general linear
group in two real variables, denoted Gl(2, R). Any homogeneous linear
transformation of two real variables is a transformation of this group. It
contains a four-parameter subgroup, the continuous group Sl(2, R). If the
rotations and pseudo-rotations depend on the group parameters θ, and γ,
and the dilatations of x, y depend on α, β, then in the general transforma-
tion of the group, X
= MX, or equivalently
x
1 = m11 (α, β, γ, θ) x1 + m12 (α, β, γ, θ)x2 ,
(2.4.3)
x
2 = m21 (α, β, γ, θ) x1 + m22 (α, β, γ, θ)x2 .
The functional dependence of the matrix elements mij upon the group para-
meters α, β, γ, θ depends upon the order in which the transformations are
carried out. Sl(2,R) differs from the similitude group because it includes
pseudo-rotations; the similitude group is a subgroup of Sl(2, R).
Generalizing the observations of this section, one finds that the group
of affine transformations of N real variables xi yields the group Gl(N, R) of
linear transformations of the corresponding ∆xi and dxi .

2.5 Physical Interpretation of Rotations

Suppose that one has


h11 h12
X
= HX, H= , X = (x1 , x2 )T (2.5.1a)
h21 h22
Now let


cos(θ) − sin(θ)
R = R(θ) = (2.5.1b)
sin(θ) cos(θ)
actively rotate X
. To determine the relation between the operation of R
on X
and the operation of R on X, we use the equations
RX
= R(HX) = RHR−1 RX = H
RX (2.5.2a)
October 28, 2010 16:55 9in x 6in b943-ch02

38 Dynamical Symmetry

with

H
= RHR−1 . (2.5.2b)

Here R−1 , the inverse of R(θ), is R(−θ). Carrying out the matrix multipli-
cations in (2.5.2b) and using trigonometric identities to simplify the result,
one finds that the elements of H
are
h
11 (θ) = 1/2 h11 (1 + cos(2θ)) + 1/2 h22 (1 − cos(2θ))
− 1/2 (h12 + h21 )sin(2θ),
h
12 (θ) = 1/2 h12 (1 + cos(2θ)) − 1/2 h21 (1 − cos(2θ))
+ 1/2 (h11 − h22 )sin(2θ),
(2.5.3)
h
21 (θ) = 1/2 h21 (1 + cos(2θ)) − 1/2 h12 (1 − cos(2θ))
+ 1/2 (h11 − h22 )sin(2θ),
h
22 (θ) = 1/2 h22 (1 + cos(2θ)) + 1/2 h11 (1 − cos(2θ))
+ 1/2 (h12 + h21 )sin(2θ).

The components of the vector RX have the dependence on θ given by

R(θ)(x1 , x2 )T = (x1 cos(θ) − x2 sin(θ), x1 sin(θ) + x2 cos(θ))T . In contrast,
as θ varies continuously from 0 to π, the matrix elements h
ij (θ) return to
their initial values hij , while the vector RX
reverses direction!
A further 180◦ rotation cycles the matrix elements h
ij (θ) through their
full range a second time and brings the vector back to its original orienta-
tion. Alternatively, if φ = 2θ and we let H
= H
(φ), then H
(2π) = H
(0),
but RX(2π) = −X. If for physical or mathematical reasons θ has a range
of 2π, then X can be a vector in a Euclidean plane. A passive, as well as
an active, rotation through 360◦ will restore the coordinates of this vector
to its original value. If it is φ that has a range of 2π, then active and pas-
sive rotations of X establish different relations between it and a coordinate
system, so X cannot be a vector in a Euclidean plane.
A geometric illustration, of vectors that are carried into their negative
by a 360◦ rotation, is provided by vectors normal to the girdle of a Moebius
strip. Figure 2.5.1a depicts such a band. Figure 2.5.1b depicts a portion
of the surface swept out by a vector normal to the Moebius band when its
“root” moves around the circle formed by the intersection of the band
with the x–y plane. An active 360◦ rotation reverses the direction of the
normal. On the other hand, the passive rotation obtained by walking once
around the z axis of the strip while carrying a coordinate system, brings the
normal back to its original relation to the observer and coordinate system.
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 39

Fig. 2.5.1 (a) Moebius band. (b) Surfaces swept through by the normal to the
band.

As both bands in Fig. 2.5.1b are Moebius strips, vectors in both behave
similarly.

2.6 Intrinsic Symmetry of an Equation

In Sec. 2.1 the roots of the equation w4 = 1 were plotted as the four
vertices of figures on three different surfaces. The vertices can, in each
case, be converted into themselves by operations of a group determined
by the group composition relations summarized in Table 2.2. Though the
composition relations are the same in all three cases, in only one case is
the figure a square. The group composition laws were determined by the
equation w4 = 1, not by its geometric realizations.
The invariance group of an equation is an intrinsic property of the equa-
tion; its geometric interpretation involves arbitrary choices. Nevertheless,
October 28, 2010 16:55 9in x 6in b943-ch02

40 Dynamical Symmetry

perhaps because Euclidean geometry is so ingrained in our minds, it is com-

monly said that an equation (or function) has the symmetry defined by its
invariance group. As this usage of words can obscure the arbitrariness inher-
ent in geometric interpretations, it can easily lead to misunderstandings.
To avoid this problem, we will henceforth use the words intrinsic symmetry
and intrinsic symmetry groups when speaking of invariance properties of
equations and functions. If functions or equations are left invariant by the
operations of a group of transformations, the group will be said to be an
intrinsic symmetry group of the functions or equations. When the invari-
ance group is given a geometric interpretation in some space, the intrinsic
symmetry is given a geometric realization.
Knowing invariance transformations of an equation enables one to con-
vert a solution of the equation into itself or other solutions. If the equations
govern a physical system, the intrinsic symmetries of the equation have
physical significance. Geometric interpretation of intrinsic symmetries can
then suggest, and organize, hosts of observable relations. This is a marvelous
capability of the human mind. However, to fully exploit this capability one
must expand and generalize one’s intuitive understandings of Euclidean
geometry. This is our next task.

2.7 Non-Euclidean Geometries

In 1820 The Elector of Hanover asked Gauss to superintend a cartographic
survey of his kingdom. Gauss, being by nature as much a mathematician
as a civil servant, the Elector’s request stimulated a general investigation
of the metrical properties of curved surfaces.1
In Euclidean three-space, an infinitesimal displacement dr with Carte-
sian components dx, dy, and dz gives rise to points separated by a distance
ds, with
ds2 = dr2 = dx2 + dy2 + dz2 . (2.7.1)
Except in the Harz mountain region, Hanover has a fairly flat terrain and
for many purposes it is sufficient to suppose the earth is a sphere of fixed
radius r. One can then use the colatitude θ and longitude φ as coordinates
of points on the sphere. The infinitesimal displacements dθ, dφ lie in the
surface r = constant, and one has
dx = d(r sinθ cosφ) = r{cosθ cosφ dθ − sinθ sinφ dφ},
dy = d(r sinθ sinφ) = r{cosθ sinφ dθ + sinθ cosφ dφ}, (2.7.2)
dz = d(r cosθ) = −r sinθ dθ.
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 41

y
0. 0.5 1.
1.

ds
da
0.5z

db

0.

Fig. 2.7.1 Displacements that determines the metric on a sphere.

Substituting the right-hand side of relations (2.7.2) into (2.7.1) yields

ds2 = g11 dθ2 + g12 dθdφ + g22 dφ2 , (2.7.3a)
with
g11 = r2 , g12 = 0, g22 = r2 sin2 θ. (2.7.3b–d)
The geometric meaning of (2.7.3) is illustrated in Fig. 2.7.1.
In the figure, |da| = rdθ, |db| = r sin θ dφ, and ds2 = da2 + db2 . Because
da and db are orthogonal, g12 is zero.
Equations such as (2.7.3) express metrical properties of the coordinate
system θ, φ on the two-dimensional surfaces r = constant. Gauss generalized
the equations to other surfaces and became the first person to take the
further, highly imaginative step of considering that surfaces with defined
metrical properties are spaces in their own right. They may, but need not
be, thought of as surfaces embedded in spaces of higher dimension. As the
axioms and theorems of plane Euclidean geometry do not all hold true on
non-planar surfaces, the imaginative jump of Gauss has the consequence of
introducing spaces with non-Euclidean geometries. The right-hand side of
an equation such as (2.7.3a) is then considered to define the metric in the
geometry, the measure of the separation of points in the space.
October 28, 2010 16:55 9in x 6in b943-ch02

42 Dynamical Symmetry

Gauss did little to publicize his conception of non-Euclidean geometries,

and in the first half of the nineteenth century Lobachevsky, Bolyai, and
others independently came to the conclusion that Euclidean geometry was
only a special type of coherently conceivable geometry.1 Spaces in which
the separation between points is defined by metric functions gij (x1 , x2 , . . .)
and quadratic in components dxi , dxj , . . . were extensively investigated by
Riemann prior to 1854 and are now termed Riemannian spaces.2
From time immemorial, many young animals have learned to recognize
that an object presented to them at different angular orientations is the
same object. They have also learned that an object moved away and back is
the same object. For homo sapiens who are beneficiaries of Pythagoras and
Descartes the resulting unconscious, and enormous, mental simplification
becomes the precursor of the abstract relation ∆r2 = ∆x2 +∆y2 +∆z2 . How-
ever, the mental organization of sense data has not lead humans to intuit
and visualize the four-dimensional non-Euclidean spacetime of Einstein.
The minds of human infants do not have to cope with visual impressions
of parents zipping about them at speeds approaching the velocity of light.
In the 1860’s Helmholtz recognized that our brains make use of geo-
metric presuppositions of which we are not aware.3 In profound studies
of human visual perceptions he independently obtained some of the key
results of Riemann. However, prior to 1905, mankind’s Euclidean geomet-
ric presuppositions led physicists to interpret the results of experiments
such as those of Michelson and Morley4 by supposing that objects moving
at constant velocity with respect to an observer shrunk in the direction of
motion, and that the mechanism of any type of clock attached to the moving
object slowed down. It required Einstein’s inspired logical analysis of physi-
cal observations and presuppositions to recognize that this strange behavior
is not a property of the objects themselves, but a property of the relation-
ship between the observer and the observed.5 Einstein then demonstrated
that if one adopted an appropriate geometry of spacetime, the observed
precession of the orbit of Mercury followed.
Suppose one observes a pair of events taking place at times t1 and t2
at points with Cartesian coordinates (x1 , y1 , z1 ), and (x2 , y2 , z2 ). Then one
will be dealing with events separated by a time interval ∆t = dt measured
by one’s
 clock, and at points
 separated from each other by a displacement
6
∆r = dr = ( dx, dy, dz). With this understanding the metric of special
relativity can be written as
ds2 = ds2sp = (dx2 + dy2 + dz2 ) − c2 dt2 (2.7.4a)
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 43

or as
ds2 = ds2t = c2 dt2 − (dx2 + dy2 + dz2 ), (2.7.4b)
where c is the velocity of light in vacuo. The metric in (2.7.4a) is said to
be spacelike, and the metric in (2.7.4b) is said to be timelike. All observers
find for c a value of approximately 2.99776 × 108 meters/sec. For a point
at (x, y, z) moving to (x + dx, y + dy, z + dz) with the velocity of light,
dx2 + dy2 + dz2 = c2 dt2 , and so in both cases ds = 0. We use dssp
to denote the real-valued ds that is obtained when the spatial separation
dr = (dx2 + dy2 + dz2 )1/2 is greater than c dt, the time it takes for light to
travel between them; and dst to denote the real-valued ds that arises when
dr < c dt. Some authors use only one definition of the metric, allowing ds
to take on imaginary as well as real values. By preference we require ds
be real-valued, and shift between the spacelike dssp and the timelike dst to
keep ds real.
The Lorentz transformations of special relativity leave invariant the ds2
of Eqs. (2.7.4) and relate observations of the same pair of events perceived
by two observers moving at constant velocity relative to one another, but
using their own Cartesian spacetime coordinate systems, coordinate sys-
tems in which they are at rest. These observers will not agree on the spatial
distance dr separating events nor upon the time interval dt between them,
but they will agree upon the spacetime interval ds separating them.
In the simplest case, an observer A and another observer B at rest with
respect to one another arrange their clocks and meter sticks to agree as
closely as possible. They align their x, y, and z axes as closely as possible
with those of the other, and then set themselves, together with their mea-
suring systems, in relative motion with constant velocity vectors parallel
to their x axes. If vx is the velocity A measures for B’s system, then B
finds that A’s system has velocity −vx . A and B observe the same set of
events and each measures the spatial separation and time intervals between
the events, then transmits the results to the other observer. If one finds
that two events are separated by ∆x and ∆τ = c∆t, the other finds them
separated by ∆x
and ∆τ
= c∆t
, with
∆x
= ∆x cosh(α) − ∆τ sinh(α), ∆τ
= ∆τ cosh(α) − ∆x sinh(α),
tanh(α) = v/c, α = ln((1 + v/c)/(1 − v/c))1/2 , v = ±vx .
(2.7.5a,b,c)
Each observer could conclude that the other’s clock has slowed, and meter
stick contracted, in the x direction. If, however, they plot their observed
October 28, 2010 16:55 9in x 6in b943-ch02

44 Dynamical Symmetry

Fig. 2.7.2 (a) Projection of B’s coordinate system onto the cartersian system of
A; (b) Projection of A’s coordinate system onto the cartersian system of B.

positions and times of occurrence of the events using Cartesian coordinates

in a Euclidean plane, they may come to a less desirable situation. Figures
2.7.2a,b illustrate the relationship that develops between coordinate sys-
tems when observes attached to them have a relative speed of 0.46c. On
plotting other’s observations reported, each may think the other is using
non-orthogonal coordinates. The transformation (∆x
, ∆τ
) → (∆x, ∆τ ) is
the inverse of the transformation (∆x, ∆τ ) → (∆x
, ∆τ
), because changing
vx to −vx in (2.7.5c) changes α to −α. The figures may hence be thought
of as depicting projections onto each observer’s Euclidean plane — projec-
tions of the spacetime coordinate system of the other observer. One may
consistently consider that the two observers view the same events from dif-
ferent perspectives. The perspectives are such that the observers agree on
the spacetime intervals, ∆s, because equations a, b, of the Lorentz trans-
formations (2.7.5) imply that
∆x
2 − c2 ∆t
2 = ∆x2 − c2 ∆t2 . (2.7.6)
Figures. 2.7.3 and 2.7.4 display lines of constant ∆s in a Euclidean
plane with orthogonal ∆x and c∆t axes. Observers in uniform relative
motion who observe the same events will assign them points with different
∆x and ∆(ct) coordinates in the figures, but the points will lie on the same
curve ∆s = const. Each curve is carried into itself by the Lorentz pseudo-
rotation transformations (2.7.5). The point (∆x, ∆τ ) = (0, 0) is invariant
under the transformation. If both observers measure (∆x, ∆τ ) from their
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 45

Fig. 2.7.3 Curves of constant ∆s, ∆x < ∆ ct.

Fig. 2.7.4 Curves of constant ∆s.

own (0, 0) point, then both may replace (∆x, c ∆t) by (x, t). Points on the
straight lines in the figures represent intervals measured for motions taking
place at the speed of light in vacuo, c. These lines separate regions where
∆s = ∆st , from the top and bottom regions where ∆s = ∆ssp .
October 28, 2010 16:55 9in x 6in b943-ch02

46 Dynamical Symmetry

Fig. 2.7.5 Curves of σ = constant; they are orthogonal to the curves of Fig. 2.7.4.

It is useful to have coordinates that measure separations along curves

of constant s or ∆s. A set of such can be obtained by rotating Fig. 2.7.4
through 90◦ . The resulting curves, Fig. 2.7.5, obey the equation. σ =
constant, where σ ≡ x·τ . As illustrated in Fig. 2.7.6, the curves of constant
σ, together with the curves of constant ∆s or s, provide each observer with
a local orthogonal spacetime coordinate system (ds, dσ) at each point (s, σ).
Translations and rotations in x, ct space can convert each of these locally
orthogonal systems to a locally orthogonal (dx, cdt) system. Finally we note
an important property of Lorentz transformations that is expressed in the
figures. As either of ∆x or c ∆t becomes much larger than the other, these
locally orthogonal coordinates approach Cartesian coordinates on almost
parallel lines of constant ∆s.
Now, having laid out the relation between the spacetime observations
of observers A and B, we can begin to ponder the meaning of symmetry in
their spacetime. We start with two-dimensional spacetime.
Fig. 2.7.7a plots the results of three (x, ct) observations made at x = 0
by an observer, A, who has clocked flashes of light from two sources, both
observed at t = 0, one from x = 0, the other from x = 2. To this observer
the flashes have ∆x values of 0, +2, respectively, and have the spacelike ∆s
values of 0 and 2. The third observation is of a flash from x = 1, observed
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 47

Fig. 2.7.6 A locally orthogonal coordinate system: s = constant, σ = constant.

Fig. 2.7.7 (a) Triangle connecting three events, through observer A’s frame of
references; (b) triangle connecting same events, through observer B’s frame of
reference.
√ √
at time t = 3 in observer A’s system. In this
√ 2 case ∆x = √1, ∆t = 3, so
2 1/2
∆s is timelike, and ∆st has the value (( 3) − 1 ) = 2. Connecting
the observations by straight lines yields the equilateral triangle shown in
the figure.
October 28, 2010 16:55 9in x 6in b943-ch02

48 Dynamical Symmetry

The sides of the triangle in Fig. 2.7.7b connect the observations of

the same events seen by A, but as observed by B, moving with velocity
v/c = 0.46 with respect to A. The origin of both observer’s measurements
is their (0, 0) point, which is unchanged by the transformation (2.7.5). To
decide whether the triangles in the two figures have the same symmetry,
we shall, in the upcoming sections, use concepts of invariance to define the
meaning of symmetry in spaces that need not be Euclidean. This turns out
to be the key in developing consistent geometric interpretations of extensive
applications of group theory in modern science.

2.8 Invariance Group of a Geometry

The invariance groups of the dynamical equations of physics and chem-
istry are seldom the Euclidean group in ordinary space or its subgroups.
In fact, the space of the variables of differential equations is often more
than three-dimensional. This is not only true for the equations of relativis-
tic mechanics, it is true for the differential equations of classical mechanics
as well as quantum mechanics, electrodynamics, and many other branches
of physics. The geometries representing the equations governing much of
physical reality are also not Euclidean geometries. Though events in the
physical world appear to us to occur entirely within a Euclidean world, rela-
tions among the variables in the equations that govern events may not be
confined by Euclidean geometry. Spatial symmetries change during molecu-
lar rearrangments, phase changes of crystals, embryological morphogenesis,
and stellar explosions — these processes are formulated in multidimensional
spaces that are not necessarily Euclidean. Because of this, the processes
oft-times produce surprises in our mental Euclidean three-dimensional geo-
metric world.
The previously mentioned work of Helmholtz apparently intrigued Lie
and led him to realize that Riemannian spaces have inherent symmetric
properties which predetermine the possible geometric symmetries of objects
in them.7 Our understanding of the connection between symmetry and
geometry was further advanced by Lie’s friend and mentor, Klein.8
The metric of an n-dimensional Euclidean geometry is
ds2 = dx21 + · · · + dx2i + · · · + dx2n . (2.8.1)
The transformation groups that characterize Euclidean symmetry in an n-
dimensional space are transformations of real variables that leave invariant
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 49

these metrics. These groups have continuous subgroups, Euclidean groups

termed E(n). These contain a subgroup of translations T(α1 , . . . , αi , . . . , αn )
that carries points with cartesian coordinates xi to points with carte-
sian coordinates xi + αi ; and they contain a subgroup group of rotations
R(θ1,2 , . . . , θi,j , . . . , θn−1,n ) that rotate points through angles θi,j about the
origin in the two-planes which contain the xi and xj axes. The invariance
group of the Euclidean metric (2.8.1) also contains a discrete subgroup of n
reflections xj → −xj . A reflection of all of the x’s together produces an
inversion through the center of the coordinate system.
The geometry of the spacetime of special relativity is defined by the
group of transformations of real variables that leave invariant the met-
rics (2.7.4). The continuous subgroup of this invariance group is called the
Poincare group. The Poincare group contains the operations of translation
and rotation of the group E(3) acting in the subspace of x, y, z, together
with the operations of time translation, t → t + τ , and the subgroup of
hyperbolic pseudo-rotations
x
= x cosh(α) − ct sinh(α), ct
= ct cosh(α) − x sinh(α),
y
= y cosh(β) − ct sinh(β), ct
= ct cosh(β) − y sinh(β), (2.8.2)
z
= z cosh(γ) − ct sinh(γ), ct
= ct cosh(γ) − z sinh(γ).
The group parameters are functions of the component of the vector V =
(vx , vy , vz ) which expresses the relative velocity of the (x
, y
, z
, t
) frame
of reference to that of the unprimed frame. In the resulting Lorentz trans-
formations, one has, for example

1/2
(1 + |vz |/c)
γ = ln . (2.8.3)
(1 − |vz |/c)
The Poincare group of spacetime translations, rotations and Lorentz
pseudo-rotations is a Lie group whose operations depend upon ten param-
eters. Four parameters determine translations, three determine rotations,
and three determine hyperbolic rotations; none can have infinite values. The
spacetime metric of special relativity is also left invariant by the inversions
xj → −xj and t → −t. Maxwell’s equations are invariant under the actions
of the Poincare group and these space and time inversions. Maxwell’s equa-
tions are also left invariant by the larger group which includes uniform
dilatations of spacetime. And they are left invariant by a still larger group,
a conformal group, which will be discussed in Chapter 14.
October 28, 2010 16:55 9in x 6in b943-ch02

50 Dynamical Symmetry

2.9 Symmetry in Euclidean Spaces

We have previously noted that the symmetry of a particular figure or lattice
in a two- or three-dimensional Euclidean space is defined by a subgroup of
the group of all translations, rotations, reflections and inversions — the
subgroup that converts the object into itself. And we have noted that the
symmetry of the object is unchanged by all operations of the continuous
group of arbitrary translations and rotations when these are applied to the
entire figure or lattice. All these operations leave invariant
ds2 = dx21 + dx22 , (2.9.1a)
or
ds2 = dx21 + dx22 + dx23 . (2.9.1b)
We have also noted that the symmetry of an object in these Euclidean
spaces is left invariant by any operation of the continuous similitude groups
which leave (2.9.1a,b) invariant, or convert them to ds
2 = e2α ds2 . As the
ds2 of (2.1.9a,b) are special cases of the corresponding ds
2 , in which α = 0,
the continuous goups which leave invariant the metrics
ds
2 = e2α (dx21 + dx22 ), or ds
2 = e2α (dx21 + dx22 + dx23 ), (2.9.1c,d)
leave invariant the symmetries of objects in these spaces.
If in an n-dimensional space, displacements along n mutually orthogonal
axes are denoted dxj , the space is Euclidean if its metric is
ds2 = dx21 + · · · + dx2i + · · · + dx2n . (2.9.2)
In this space, the operations of the continuous group E(n) carry out n
linearly independent translations, one parallel to each of its n Cartesian
axes, and rotations in each of n(n − 1)/2 planes that contain two of these
axes. The operations of E(n) consequently depend on n(n + 1)/2 indepen-
dent parameters. The metric is also left invariant by the discrete group of
inversions and reflections.
The symmetry of any particular object in the space is defined by the
subgroup of the group that leaves invariant the metric. The symmetry of
the object is invariant under all the operations of the continuous similitude
group of the space. These leave invariant
ds
2 = e2α (dx21 + · · · + dx2i + · · · + dx2n ), −∞ < α < ∞. (2.9.3)
The operations of the continuous similitude group of n-dimensional
Euclidean space consequently depends on one more independent parameter.
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 51

2.10 Symmetry in the Spacetime of Special Relativity

We next consider what is to be meant by symmetry when the metric is the

ds2 = ±(dx2 + dy2 + dz2 − c2 dt2 ) (2.10.1a)

that is left invariant by all transformations of the Poincare group. We shall

call the continuous group that leaves (2.10.1a) invariant or converts it to

ds
2 = ±e2α (dx2 + dy2 + dz2 − c2 dt2 ), −∞ < α < ∞, (2.10.1b)

the Poincare similitude group.

The metrics are, first of all, left invariant by all transformations in ordi-
nary space which separately leave invariant the Euclidean metric dr2 =
dx2 + dy2 + dz2 , and the metric dt2 . The latter is unchanged by time trans-
lations and the inversion t → −t. Observers not moving relative to an
object in space can agree on both dt2 and dr2 , and hence agree on the
spatial symmetry of an object, and the time span over which it is observed.
However, as was illustrated in Fig. 1.1, observers moving relative to one
another may not agree on the Euclidean symmetry of an object in space-
time. Observers moving at different, but constant, velocity with respect to
an object will agree on the symmetry of the object if and only if they change
their concept of symmetry.
The alternative to making such a change is to believe, in the sense under-
stood prior to 1905, that Fitzgerald contractions and time dilatations are
physical effects, i.e., that objects moving at constant velocity compress in
the direction of their motion, and that clocks moving at constant velocity
tick at a different rate. If, on the other hand, one adopts Einstein’s view-
point, the unit cells of Figs. 1.1 and the two spacetime triangles exhibited
in Figs. 2.7.7 must be considered as the same physical objects viewed from
different spacetime perspectives.
Objects in spacetime are to be considered identical if their only apparent
difference is due to a change in spacetime perspective that can be produced
by an operation of the Poincare group. This produces a further general
conclusion: objects in spacetime that can be interconverted by operations
of the Poincare similitude group have the same spacetime symmetry. We
therefore make the definitions which follow.

Definition 2.1. Let a set of events determine a figure or lattice in space-

time with metric

ds2 = ±(dx2 + dy2 + dz2 − c2 dt2 ). (2.10.2)

October 28, 2010 16:55 9in x 6in b943-ch02

52 Dynamical Symmetry

The symmetry of the figure or lattice is invariant under the action of the
continuous group of transformations of the Poincare similitude group, which
leaves invariant the real-valued spacetime intervals defined by (2.10.2), or
multiplies these by a real number greater than zero.

Definition 2.2. The symmetry of any figure or lattice in spacetime is

defined by the subgroup of the group of all transformations which leave
spacetime intervals invariant — the subgroup of the Poincare group that
carries the spacetime figure or lattice into itself.

In the next few paragraphs we briefly consider implications of these

definitions.
The Euclidean similitude group of space is a subgroup of the Poincare
similitude group of spacetime. Its operations can therefore be used to define
symmetries in spacetime, but its operations only alter spatial perspectives.
Because Lorentz transformations act in spacetime they can define sym-
metries in spacetime, and alter spatial perspectives as well as spacetime
perspectives.
Figures 2.10.1–2.10.3 display examples of these observations.
Figure 2.10.1 plots the coordinates of four light flashes recorded by
observer A, with (x, y) = (0, 0), as having occured at ct = −1/2. The
points all lie on the on the circle x2 + y2 = r2 = 1/2 and the hyperboloid

Fig. 2.10.1 Four simultaneous events plotted on a hyperholoid in spacetime

x, y, ct.
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 53

X,Y Coordinates of Four Events Recorded by Different Observers

0.6
0.4

0.4

0.2
0.2

− 0.4 − 0.2 0.2 0.4 − 0.6 − 0.4 − 0.2 0.2 0.4 0.6

− 0.2
− 0.2

− 0.4

− 0.4

(a) (b)

0.6
0.4

0.4

0.2
0.2

− 0.8 − 0.6 − 0.4 − 0.2 0.2 −1 − 0.8 − 0.6 − 0.4 − 0.2 0.2 0.4

− 0.2
− 0.2

− 0.4

− 0.4
− 0.6

(c) (d)

Fig. 2.10.2 Events from the perspectives of A, B, C and D in (a), (b), (c) and
(d) respectively.

x2 + y2 − c2 t2 = s2 = 1/4. If A resets his clock to convert ct = 1/2

to ct = 0, then Figure 2.10.1 projects onto the (x, y, 0) plane to yield
Fig. 2.10.2a, for which r2 = s2 = 1/2. Figures 2.10.2b,c,d plot obser-
vations of the same flashes by three observers B,C,D who have clocks ini-
tially syncronized with A’s (say, in the manner described in Sec. 2.7). In
Figs. 2.10.3, θ is the angle of rotation and α, the “angle” of pseudo-
rotation, which relate the observer’s coordinate system to the coordinate
system of A. The value α = 1/2 arises when two observers are moving
October 28, 2010 16:55 9in x 6in b943-ch02

54 Dynamical Symmetry

−0.5
0.0 −0.5
0.5 0.0 0.5
1 1 1
[ ,− , ]
2 2 2
1 1 1
1 1 1
[− , , ] [− ,− , ]
2 2 2
2 2 2 0.5 0.5

0.0
0.0

1 1 1
[ ,− ,− ] −0.5
2 2 2
−0.5

−0.5
0.0 −0.5
0.0
0.5
0.5

(a) (b)

−0.5 0.0
0.5 −0.5
0.0
0.5
1.0

0.5
0.5

0.0
0.0

−0.5

−0.5
−1.0
−1.0
−0.5
0.0
−0.5
0.5
0.0
1.0
0.5

(c) (d)

Fig. 2.10.3 Relation between eight events in spacetime from perspectives of four
observers in x, y, ct coordinates. (a) Events in A’s frame; (b) events in B’s frame,
α = 0, θ = π2 ; (c) events in C’s frame, α = 12 , θ = 0; (d) events in D’s frame,
θ = 12 , α = 12 .

with relative velocity 0.46c. All observers find s2 = 1/2. Observers A, with
(θ = 0, α = 0), and B, with (θ = π/4, α = 0), find the flashes occur at the
corners of a square. Observer C moving relative to A with (θ = 0, α = 1/2)
observes the events at corners of a rectangle; and observer D, with (θ = π/4,
α = 1/2), finds them to be at the vertices of a rectangular parallelepiped.
An analogous situation occurs in Euclidean geometry. If a square in an x–y
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 55

plane through z = 1/2 is rotated about an axis in the plane, its projec-
tions on the x–y plane through z = 0 can be similar to the projections in
Figs. 2.10.2.
As the physical relationship defined by the four events depicted in
Fig. 2.10.2a has not itself changed, observers A,B,C,D are all recording
events defining a quadrilateral with all the spatial symmetry of the square
determined in the analysis of Fig. 2.1 and tabulated in Table 2.1. Stated in
another way, figures with the apparently different spatial symmetries pic-
tured in Figs. 2.10.2 have physical symmetries that are indistinguishable
in a spacetime with metric (2.10.2).
Figures 2.10.3 illustrate the spacetime perspectives of observers
A, B, C, D on two sets of four events, each defining quadrilaterals like those
in Figs. 2.10.2. For simplicity it is supposed that the clocks are so imagi-
natively set that for A the events occur at ct = −1/2 and ct = 1/2. All the
observers find s2 = 1/4. Again, all the figures are interconverted by actions
of the Poincare group, so their symmetries in spacetime are identical.
Figures 2.10.1 and 2.10.2 display a symmetry defined by an E(2) spa-
tial subgroup of the Poincare group: the E(2) operations act in x–y planes
with fixed t. Figures 2.10.3 display this same symmetry, and also rela-
tions defined in planes parallel to a t axis. To deal with these, we consider
briefly symmetries defined by operations of the Poincare group that are not
contained in its spatial subgroups.
Hyperbolic symmetry is defined by the invariance of curves of constant
∆s or s under the one-parameter subgroup of hyperbolic pseudo-rotations.
Hyperbolic symmetry can be thought of as a spacetime analog of rotational
symmetry: the Minkowski transformations z → ict, r2 → ±s2 convert circles
and spheres defined by

y 2 + z2 = r 2 , x2 + y2 + z2 = r2 , (2.10.3a,b)

to hyperbolae and hyperboloids defined by

y2 − c2 t2 = ±s2 , x2 + y2 − c2 t2 = ±s2 . (2.10.3c,d)

In Figs. 2.10.3, the Lorentz transformations move points along hyperbo-

lae on the hyperboloid x2 + y2 − c2 t2 = 1/4. All faces of the rhomboids
of Figs. 2.10.3c,d have symmetries defined by discrete subgroups of the
Poincare group. The symmetries of the faces lying in planes in which t
varies are defined by discrete subgroups of a continuous group of Lorentz
transformations. The operations of these subgroups move points along
October 28, 2010 16:55 9in x 6in b943-ch02

56 Dynamical Symmetry

hyperbolae rather than circles. The same is true of the discrete opera-
tions which act on the top and bottom faces of Figs. 2.10.3a,b to yield
Figs. 2.10.3c,d.
For a conclusion, we note that arguments of the form we have used to
define the meaning of geometric symmetry both in Euclidean spaces and in
spacetime may be used to define geometric symmetry in any space with a
metric. We shall be using them later. In each case geometrical interpreta-
tions will depend upon physical interpretations.

2.11 Geometrical Interpretations of Nonlinear

Transformations: Stereographic Projections
The geometrical and physical interpretation of nonlinear transformations
raises issues that we have not yet considered. In Sec. 2.3 we investigated
active and passive interpretations of linear transformations. Linear transfor-
mations were also considered as transformations from one space to another.
In this section we will quickly find that the interpretation of nonlinear trans-
formations introduces geometrical and physical considerations that make
the active and passive interpretations inequivalent and make the passive
interpretation extremely misleading.
Let (x, y) be the Cartesian coordinates of a point in a Euclidean plane.
A transformation of these variables defined by
x → x
= f(x, y), y → y
= g(x, y) (2.11.1)
is nonlinear if at least one of the functions f or g has a nonlinear dependence
on at least one of the variables x, y. It follows that in the equations
dx
= (∂f/∂x) dx + (∂f/∂y) dy, and dy
= (∂g/∂x) dx + (∂g/∂y) dy,
(2.11.2)
at least one of the derivatives is not a constant. However, if x
, y
are to be
Cartesian coordinates in a single Euclidean plane it is necessary that

dx
2 + dy
2 = k2 (dx2 + dy2 ), k real. (2.11.3)

In the case of uniform dilations, k is constant. Transformations of conformal
groups, similar to the group discussed in Chapter 14, can produce k’s that
are functions of x.
Given an n-dimensional Euclidean space of variables z, and a nonlinear
transformation of these variables, one may consider the transformation to
be an active one that moves points in the space, but it is seldom easy to
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 57

believe with consistency the transformation is a passive transformation of

coordinates in the same Euclidean space.
However, the coordinates of points moving in an n-dimensional non-
Euclidean space may be related by nonlinear one-to-one transformations
to coordinates of points moving in a Euclidean space of higher dimension.
This happens when the non-Euclidean space can be considered to be a
surface in the higher dimensional space. Such transformations, and their
inverse transformations, are said to be projections. In this section several
transformations of physical interest will be used to illustrate a number of
properties of projections.
The most common projections are those that map points on the sur-
face of the earth onto flat paper. As previously mentioned, one of these,
the stereographic projection, is also used in crystallography to map points
on the surfaces of a three-dimensional crystal onto a Euclidean plane. In
his discussion of the degeneracy of the hydrogen atom, Fock used a four-
dimensional generalization of this transformation to map points on the
surface of a hypersphere in a four-dimensional Euclidean space to points in
a three-dimensional Euclidean space.
The following relations define a stereographic projection connecting
points with Cartesian coordinates (x, y) in a Euclidean plane to points
with Cartesian coordinates (z1 , z2 , z3 ) in Euclidean three-space:
z1 = 2ax/(a2 + r2 ) = 2(x/a)/(1 + (r/a)2 ),
z2 = 2ay/(a2 + r2 ) = 2(y/a)/(1 + (r/a)2 ),
z3 = (a2 − r2 )/(a2 + r2 ) = (1 − (r/a)2 )/(1 + (r/a)2 ), (r2 = x2 + y2 ).
(2.11.4)
It follows from these equations that
z21 + z22 + z23 = 1. (2.11.5)
Thus the projection connects points at r in the Euclidean plane to points
at z on the surface of a unit sphere in the Euclidean space of the variables
zj with center at the origins z = 0, r = 0.
The transformation of coordinates inverse to (2.11.4) is
x/a = z1 /(1 + z3 ), y/a = z2 /(1 + z3 ). (2.11.6)
The variable a establishes the scale of the mapping. In particular, if z3 = 0,
then r = (x2 + y2 )1/2 = a, so in the x, y coordinate system of the plane,
a is the radius of the circle where the sphere intersects the plane. On this
October 28, 2010 16:55 9in x 6in b943-ch02

58 Dynamical Symmetry

circle z1 = x/a, z2 = y/a. Note that both r and the scale parameter a must
have identical units, as the variables zj satisfy (2.11.5). Equations (2.11.4)
define the stereographic projection illustrated in Figs. 1.2, the second set
of figures in Chapter 1.
Figures 1.2a,b depict the manner in which points on the Northern
and Southern hemispheres are mapped into points in the plane. Points in
the Northern hemisphere map into points that lie inside the sphere, and
points in the Southern hemisphere map to points outside the sphere. This
property is independent of the value of the scale factor a. In the illustrated
case, the scale factor is unity, so the intersection of the sphere with the x–y
plane is a unit circle in both the x and y coordinates, and the zj coordinate
systems. When a is not unity, the circle of intersection is defined by
z21 + z22 = 1 in the one coordinate system, and by x2 + y2 = a2 in the other.
Using spherical polar coordinates (θ, φ), with 0 ≤ θ ≤ π and 0 ≤ φ ≤ 2π,
one has on a unit sphere
z1 = sin(θ) cos(φ), z2 = sin(θ) sin(φ), z3 = cos(θ). (2.11.7)
Using (2.11.4), (2.11.7) one finds that stereographic projection connects the
x, y coordinates to angular coordinates on the sphere via the relations:
a sin(θ) cos(φ)
x= = a cos(φ) tan(θ/2) ≡ X(θ, φ),
(1 + cos(θ))
(2.11.8)
a sin(θ) sin(φ)
y= = a sin(φ) tan(θ/2) ≡ Y(θ, φ).
(1 + cos(θ))
It follows that
r/a = tan(θ/2). (2.11.9)
In the x–y plane, r = a, θ = π/2, and (2.11.8) requires
x = r cos(φ), y = r sin(φ). (2.11.10)
Stereographic projections relate measurements of coordinates x, y and
displacements dx, dy in a plane to measurements of coordinates θ, φ and
displacements dθ, dφ in the curved two-dimensional space, termed S(2),
that may be considered to be the surface of the sphere in three-space. In
the Euclidean space of (z1 , z2 , z3 ), the metric is
ds2 = dz21 + dz22 + dz23 . (2.11.11)
On the unit sphere this becomes
ds2 = dθ2 + sin2 (θ)dφ2 . (2.11.12)
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 59

(b)

Fig. 2.11.1 The projection shaded in each figure is onto the equatorial plane.

This metric is invariant under the group of rotations about the center
of the sphere, inversions z → −z, and reflections zj → −zj . The group,
denominated O(3), is termed an orthogonal group because it transforms
orthogonal vectors into orthogonal vectors. Both the metric and the sphere
are left invariant by the transformations of O(3).
Subgroups of the invariance group of the sphere will consequently define
the symmetry of a figure on the sphere. The symmetry is not altered by
rotations of the entire figure. Figures may be shrunk or enlarged without
October 28, 2010 16:55 9in x 6in b943-ch02

60 Dynamical Symmetry

changing their symmetry, but this dilatation invariance is limited because

no two points on the sphere can become separated by a distance greater than
πR, where R is the radius of the sphere.
We next consider the relation between symmetries in the plane
and symmetries on the sphere established by stereographic projection.
Figures 2.11.1a,b illustrate the stereographic projection of circles of
constant latitide. Not surprisingly, their projections are circles in the
plane which are centered at the origin. Figures 2.11.1c,d illustrate a
less-than-obvious property of stereographic projections that is familiar to
mineralogists and crystallographers.9 Any circle in the plane can be the
stereographic projection of a circle on the sphere. The converse is not true
in the ordinary sense: the stereographic projection of all circles on the
sphere, except great circles passing through the poles, are circles in the
plane. Great circles project as straight lines. A proof of these statements is
given in Appendix 2.A.
Figures on the sphere with lesser rotational symmetry about the polar
axis than a circle, for example that of a regular hexagon, project to fig-
ures in the plane that have the same symmetry. However, projections of a
figure with more than twofold rotational symmetry about any other axis
do not retain the same rotational symmetry in the plane. In such cases
stereographic projections restrict symmetries. The inverse projections con-
sequently create symmetries. The stereographic projections in Figs. 1.2
convert the level curves of atomic orbitals with s, p type symmetries that
differ in the plane, to level curves expressing identical, p type, symmetries
on the sphere.
We next turn to a consideration of the relationship that stereographic
projections establish between local metrical properties on the sphere and
local metrical properties in the plane. Using (2.11.8) one finds the following
relation between the metric ds2 on the sphere and the metric

dr2 = dx2 + dy2 (2.11.13)

in the plane:

ds2 = dr2 (1 + cos(θ))2 . (2.11.14)

As cos(θ) ranges in value from 1 at the North pole of the sphere, to −1

at its South pole, (1 + cos(θ))2 ranges from 4 to 0. As one expects from
the figures, separations in the Northern hemisphere project into separations
in the plane that are smaller than those obtained for points in the plane
that lie on the equator of the sphere, where cos(θ) = 0. Then, by making
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 61

60

40

20

0
1 2 3 4 5 6
θ
− 20

− 40

− 60

Fig. 2.11.2 The function tan( 21 θ).

increasingly large local-scale-transformations as r increases, the projection

fits all the remaining points in the plane onto the Southern hemisphere
of the sphere. In short, as θ approaches π, r approaches infinity, and ds
approaches zero.
Before concluding this discussion of stereographic projections we must
deal a little more extensively with the behaviour of x = a tan(θ/2) cos(φ),
and y = a tan(θ/2) sin(φ) as θ → π. Figure 2.11.2 illustrates a well known
behavior of the tangent function. Both x and y shift from +∞ to −∞, as
θ passes π, and the reverse shift in θ produces the reverse shift in x, y.
This happens with a minute shift in θ, and at a point where ds2 = 0 dr2 .
The stereographic mapping has the effect of treating −∞ as though it
differed minutely from +∞! In subsequent chapters we will deal with other
mappings which have this property, and with physical consequences that
result from it.

2.12 Continuous Groups that Leave Euclidean

and Pseudo-Euclidean Metrics Invariant
In this section we consider some generalizations of the metrics of ordinary
Euclidean space and spacetime considered in earlier sections.
All of the continuous groups that leave these metrics invariant are special
cases of Lie groups. Each is obtained by defining the manner in which a
Lie group transforms the variables of interest. A Lie group acting on a
particular set of variables is said to be a realization of an abstract Lie
group. Thus, for example, E(1) the transformation group which converts
a Cartesian coordinate x in Euclidean space to x + a, is a realization of
October 28, 2010 16:55 9in x 6in b943-ch02

62 Dynamical Symmetry

an abstract Lie group. It converts x2 to (x + a)2 . A different realization

of the same abstract Lie group might be defined to convert x2 to x2 + a.
Unfortunately, a Lie group and its most common realization are often given
the same name. One may tell usually from the context which meaning is
intended.
The group of all linear transformations that leave invariant
z21 + · · · + z2i + · · · + z2n (2.12.1)
is called an orthogonal group, and denoted by O(n). The group O(n) con-
tains, as subgroups, the continuous group of rotations, SO(n), and the dis-
crete group of inversions and reflections that arise from letting
zi → −zi . (2.12.2)
The continuous group of linear transformations that leaves invariant
dz21 + · · · + dz2i + · · · + dz2n (2.12.3)
is denoted by E(n) or ISO(n), which stands for Inhomogeneous Special
Orthogonal group. Its operations carry out translations zj → zj + aj , as
well as rotations. When the latter Lie group is extended to include the
continuous group of uniform dilatations, zj → eα zj , j = 1, . . . , n, the group
becomes the corresponding similitude group. When ISO(n) is extended to
include the discrete inversions zj → −zj , the group becomes the inhomoge-
nous orthogonal group denoted IO(n).
The discussion of these groups is greatly simplified if one makes use of
the concept of vector introduced by Gibbs.10
For an n-dimensional space there is a direct generalization of Cartesian
vectors in ordinary two- and three-dimensional space. Let z = (z1 , . . . , zn )
represent such a vector.
For z and z
to be vectors it is required that:
(i) There is a law of vector addition which states that z+z
= z

, meaning
(z1 , z2 , . . . , zN ) + (z
1 , z
2 , . . . , z
N )
= (z1 + z
1 , z2 + z
2 , . . . , zN + z
N ) = z

. (2.12.4)
where z

is also a vector.
(ii) There is a law of multiplication of z by a real number, c, (a scalar)
defined by
z → cz = (cz1 , cz2 , . . . , czN ) = z
, (2.12.5)
where z
is also a vector.
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 63

(iii) Vectors satisfy the distributive law

c(z + z
) = cz + cz
= z

. (2.12.6)

(iv) Finally, vectors obey an associative law

(z + z
) + z

= z + (z
+ z

) = z


. (2.12.7)

Vectors will henceforth be denoted by bold-face roman characters. A list

of variables contained within parentheses will be considered a vector if the
parentheses in boldface.
Vectors may be subject to further operations. Taking the scalar product
of two vectors z and z
is one such operation. In a Euclidean space it is
defined as

z · z
= z1 z
1 + · · · + zn z
n . (2.12.8a)

The magnitude of the vector z is then defined to be |z| = (z · z)1/2 . The

cosine of the angle α, between two vectors z and z
, is

cos(α) = z·z
/|z||z
|. (2.12.8b)

All operations of the orthogonal group O(n) leave this quantity invariant.
Orthogonal groups leave orthogonal vectors orthogonal, hence their name.
Unit vectors vj = (0, . . . , 0, vi , 0, . . . , 0), with vj = 1, j = 1, . . . , n, may be
attached to a set of n orthogonal axes. The component cj of the vector
r = Σcj vi on the axis j is then defined by cj = r·vi /|vi |. While unit vectors
remain so when dilatations are considered to be active transformations, they
change length when dilatation transformations are interpreted passively.
In the most common realization of the group SO(n), the n(n − 1)/2
group parameters are the angles that define rotation operations. These act
in n(n−1)/2 mutually orthogonal two-planes in an n-dimensional Euclidean
space. Rotation of a vector z through an angle θjm about the origin in the
zj , zm two-plane only affects the coordinates (zj , zm ). It is defined by

zj → z
j = zj cos(θjm ) − zm sin(θjm ),

(2.12.9)
zm → z
m = zj sin(θjm ) + zm cos(θjm ).
All the transformations preserving Euclidean symmetries preserve angles
between lines. One can erect a fixed set of n mutually orthogonal axes in
n-dimensional Euclidean spaces, and these orthogonal axes remain orthog-
onal when transformations of the groups O(n) and SO(n) are interpreted
passively or actively.
October 28, 2010 16:55 9in x 6in b943-ch02

64 Dynamical Symmetry

In spacetime the scalar product may be defined to be plus or minus

z1 z
1 + z2 z
2 + z3 z
3 − z4 z
4 , z4 = ct. (2.12.10)
It is invariant under operations of the continuous group SO(3,1). The oper-
ations of its SO(3) subgroup leave spatial angles invariant, while Lorentz
transformations do not. However, as noted in Sec. 2.7, there exist well
defined locally orthogonal spacetime coordinate systems.
The group of all linear transformations that leave invariant
(z21 + · · · + z2m ) − (z2m+1 + · · · + z2n ) (2.12.11)
is denoted O(m, n − m). The Lie group SO(m, n − m) is a subgroup of
O(m, n − m). (It is commonplace to suppose m ≥ n − m.) SO(m, n − m)
has SO(m) and SO(n − m) subgroups which operate in the subspace of
z1 · · · zm and zm+1 · · · zn , respectively. In O(m, n − m) the scalar product of
two vectors is defined by
(z1 z
1 + · · · + zm z
m ) − (zm+1 z
m+1 + · · · + zn z
n ). (2.12.12)
It is invariant under all the operations that leave (2.12.11) invariant.
Orthogonality, Cartesian coordinates, and unit vectors in these spaces are
defined using this scalar product.
Interpretating hyperbolic pseudo-rotations as mappings, one is allowed
to use Cartesian coordinates to interpret their action. This was done in
the figures of Sec. 2.3, and throughout Sec. 2.7. In the general case, the
operations of SO(m, n − m) can be then considered to carry out m(n − m)
active hyperbolic pseudo-rotations about mutually perpendicular axes. The
group has two subgroups SO(n) and SO(n− m), whose operations carry out
active rotations as described above. Passive interpretation of hyperbolic
pseudo-rotations destroys Euclidean orthornormality. One may interpret
all the operations of these groups as mappings that relate the coordinates
of points in one orthogonal coordinate system to those of points in another,
equivalent orthogonal coordinate system.
The transformations of the continuous group ISO(m, n − m) supplement
those of SO(m, n − m) with m+n translation operations zj → zj + αj and
leave invariant
ds2 = dz21 + · · · + dz2m − dz2m+1 − · · · − dz2n . (2.12.13)

On including the discrete inversions zj → −zj , the group becomes

IO(m, n − m). Supplementing these operations of ISO(m, n − m) with uni-
form continuous dilatations yields the corresponding similitude group that
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 65

multiplies the ds2 of (2.12.13) by a positive constant. Minkowsk’s trick z →

iz may be used to convert these non-Euclidean spaces to Euclidean spaces.
The transformation groups discussed in this section carry out linear
transformations on a set of variables zi , and leave metrics invariant or
multiply them by a constant. The sphere of action of the groups may be
greatly extended by considering nonlinear transformations of the Cartesian
variables z, transformations which do not preserve the ISO(n) or ISO(m, n)
metrics or multiply them by constants. As an example, one may observe that
the stereographic projection converts rotations on a sphere in Euclidean
three-space into operations in the plane, some of which are not rotations,
and do not leave Euclidean distances in the plane invariant.

2.13 Geometry, Symmetry, and Invariance

In the previous sections we have used mathematical properties of trans-
formations of position and time variables, and also invariance properties
of Euclidean space and physical spacetime, to demonstrate that for one to
understand the role of symmetry in the physical sciences, it is necessary
to first investigate the geometric interpretation imposed on observations.
This lead us into the realm of non-Euclidean geometries and symmetries.
These symmetries can be related to ordinary symmetries in a manner that
enhances, rather than undermines, one’s Euclidean insights. The greatest
surprises occur when nonlinear mappings relate one space to another.
We have seen that geometry in a Euclidean space may be characterized
by the group of invariance transformations of its metric, ds2 . The symme-
try of a geometric object in the space is characterized by the group that
transforms the object into itself. This is a subgroup of the invariance group
of the metric. Geometric objects are considered equivalent, i.e. congruent,
if operations of the group can superpose them. Objects also have the same
symmetry if they may be superposed by operations of an enlarged group
that contains dilations which can uniformly stretch or compress all distances
by multiplying ds2 with a positive constant.
Geometry and congruence in any Riemannian space may be similarly
characterized. However, as we have seen in the case of the spherical surface
S(2), the role of dilatations must be restricted if ∆s has a restricted range
of values.
Let us re-emphasize three other key points made in this chapter:
(1) Equations possess intrinsic symmetries, defined by the transformation
groups that leave them invariant. Even in simple cases, a variety of
October 28, 2010 16:55 9in x 6in b943-ch02

66 Dynamical Symmetry

assumptions are made when one interprets these intrinsic symmetries

as geometric symmetries.
(2) When equations describe physical relationships, intrinsic symmetries of
the equations are given physical interpretation. Geometric interpreta-
tion of these physical relationships involves assumptions about obser-
vations and measurements as well as their geometric interpretation.
Geometric interpretations of relations between observations of natural
phenomena are not dictated solely by the relations themselves.
(3) Because spacetime is non-Euclidean, Euclidean conceptions of geomet-
ric symmetry require revision when considering physical symmetries in
spacetime.
These observations are of central importance in much of what follows. In
Chapter 3 we introduce the conception of symmetry — symmetry in phase–
space — that is appropriate to systems governed by the ordinary differential
equations of Hamiltonian mechanics. In Chapter 9 we will describe the
manner in which this conception applies to systems governed by the partial
differential equations of Schrödinger mechanics.
Differential equations govern a great variety of physical and chemical
systems, and differential equations are left invariant by groups of transfor-
mations of their variables. For each system, the variables can be considered
to reside in a space which has an associated metric. The metric is itself
invariant under a group of transformations. This allows one to define the
space of the physical variables and the meaning of symmetry in the space.
Once this space has been defined, the transformations that leave the differ-
ential equation invariant can be given geometrical interpretations. Because
the operations of the group carry solutions of the equation into solutions,
the operations interrelate observations and have physical interpretation.
This provides the connection between geometry and the natural world when
natural phenomena are related by continuous causal chains.

Appendix A: Stereographic Projection of Circles

We begin by setting up a description of circles on a unit sphere. When a
plane, Ps , parallel to the z1 –z3 plane, intersects the unit sphere of Sec. 2.11,
the intersection is a circle of radius s whose points are defined by the vector
Z = (z1 , z2 , z3 ) = (s cos(τ ), (1 − s2 )1/2 , s sin(τ )), (2.A.1)
with s lying between 0 and 1. The polar axis of the sphere is z3 . The z2
coordinate in Z, (1 − s2 )1/2 , is the distance to the Ps plane, measured along
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 67

Fig. 2.A.1 The vector C and the plane normal to it.

a vector C, a normal to the plane from the origin. Figure 2.A.1 depicts the
relationship between C, and the plane normal to it when C has been rotated
about the z1 axis through an angle Θ, measured from the positive z3 axis.
Equation (2.A.1) applies when Θ = 90◦ . In this case, C = (0, (1 − s2 )1/2 , 0).
As τ ranges from 0 to 2π the terminus of the vector Z sweeps along the
circle at the intersection of the plane with the unit sphere. For all values of
s, the terminus of C is the center of the circles swept out by Z.
When the plane Ps is tilted by altering Θ, the normal from the origin
becomes
C(Θ) = (0, cos(Θ)(1 − s2 )1/2 , − sin(Θ)(1 − s2 )1/2 ). (2.A.2)
The vector Z of (2.A.1) then becomes a vector Z(Θ, s, τ ) with coordinates
z1 = s cos(τ ), z2 = cos(Θ)(1 − s2 )1/2 + sin(Θ)s sin(τ ),
(2.A.3)
z3 = cos(Θ)s sin(τ ) − sin(Θ)(1 − s2 )1/2 .
As τ ranges from 0 to 2π, the terminus of this vector describes circles at
the intersection of the tilted plane with the sphere, and center at C(Θ).
Consider, then, the projection of these circles onto the x–y plane. The
three-vector Z(Θ, s, τ ) is projected onto the x–y plane, i.e. z1 –z2 plane,
using the relations
x = az1 /(1 + z3 ), y = az2 /(1 + z3 ). (2.A.4)
This projection yields the two-vector (x, y) = r(Θ, s, τ ) whose compo-
nents are
x = a s cos(τ )/D,
y = a{cos(Θ)(1 − s2 )1/2 + sin(Θ)s sin(τ )}/D, (2.A.5)
October 28, 2010 16:55 9in x 6in b943-ch02

68 Dynamical Symmetry

where
D = 1 + cos(Θ)s sin(τ ) − sin(Θ)(1 − s2 )1/2 .
The centers of the circles on the sphere project to points defined by (2.A.4),
which have coordinates (x, y) given by
(x0 , y0 ) = (0, a cos(Θ)(1 − s2 )1/2 /{1 − sin(Θ)(1 − s2 )1/2 }). (2.A.6)
If the projections are themselves circles, (2.A.6) should fix their centers.
Expanding r(Θ, s, τ )·r(Θ, s, τ ), and using (2.A.6) and (2.A.5) to relate x, y
to τ , one finds that as τ ranges from 0 to 2π, the terminus of r(Θ, s, τ ) does
indeed sweep out a circle defined by
x2 + (y − y0 )2 = R2 ,
with radius
R = a s/{1 − sin(Θ)(1 − s2 )1/2 }, (2.A.7)
and
y0 = a cos(Θ)(1 − s2 )1/2 /{1 − sin(Θ)(1 − s2 )1/2 }.
Both y0 and R become infinite when {1 − sin(Θ)(1 − s2 )1/2 } becomes zero,
which happens for a great circle passing through the poles. This projection
of the circle is one that is indistinguishable from a straight line.
The circles defined by (2.A.3) lie in planes whose normal through the
origin lies in the y–z plane. Rotating this normal, C, about the z3 axis, and
varying Θ, allows one to obtain planes with arbitrary orientation. By vary-
ing s one can then obtain all possible circles produced by the intersection
of planes with the sphere. The rotation of C about the z3 axis through an
angle α produces a corresponding rotation of the vector (x0 , y0 ) and the
circles of (2.A.7). As all circles inscribed on the sphere can be produced by
varying s, Θ, and α, it follows from (2.A.7) that the stereographic projection
of any circle on the sphere is also a circle.

References
[1] B. A. Rosenfeld, A History of Non-Euclidean Geometry, translated by A.
Shenitzer, (Springer, New York, 1988).
[2] B. Riemann, Gesammelte mathematische Werke und wissenschaftlicher
Nachlass, Herausg. von H. Weber unter Mitwerkung von R. Dedekind
(Dover., N.Y., 1953).
[3] H. Helmholtz, Wissenschaftlich Abhandlungen, Bd. 2 (Barth, Leipzig, 1882).
[4] A. A. Michelson, E. W. Morley, Silliman’s. J., 34 (1887) 333.
October 28, 2010 16:55 9in x 6in b943-ch02

Physical Symmetry and Geometrical Symmetry 69

[5] A. Einstein, Annalen der Physik, 17 (1905) 891, cf. “On The Electrodynam-
ics of Moving Bodies”, in The Principle of Relativity, W. Perrett, translated
by G. B. Jeffery, (Dover, 1923).
[6] This involves a presupposition about the continuity of space and time which
is open to challenge. It also involves a presupposition that the infinitesi-
mal quantities can be integrated to give unique finite quantities. Cf. A. S.
Eddington, The Mathematical Theory of Relativity (Cambridge University
Press, 1924), pp. 198–9.
[7] S. Lie, Gesammelte Abhandlungen, Vol. 2, pp. 374–479 (Teubner, Leipzig,
1935) (Johnson Reprint, New York, 1973).
[8] F. Klein, Elementary Mathematics From An Advanced Standpoint, Vol. II,
Geometry (Dover, New York, 1939).
[9] E. S. Dana, A Textbook of Mineralogy, 4th edn. revised by W. E. Ford (Wiley,
New York, 1932) pp. 49–56.
[10] E. B. Wilson, Vector Analysis Founded upon the Lectures of J. Willard Gibbs
(1901).
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-ch03

CHAPTER 3

On Symmetries Associated With

Hamiltonian Dynamics

3.1 Invariance of a Differential Equation

The reader will find that the connections made in this chapter are guided
by the conceptions due to Sophus Lie. Properties of motions in one spatial
dimension will be used to clarify the manner in which invariance proper-
ties of Hamilton’s differential equations bring concepts of geometry and
symmetry into dynamics. Systems of higher dimension are considered in
Chapters 7 and 8. As Hamiltonian dynamics is the classical dynamics most
directly relevant to quantum mechanics, concepts developed in this chapter
will also underpin the discussions from the beginning of Chapter 9 to the
end of Chapter 14.
The differential equations of different Hamiltonian systems may have
different intrinsic symmetries. It will be shown that all have a common
intrinsic symmetry that determines an associated invariant metric. Investi-
gating the geometric and dynamical expressions of this metric will enable
us to define the meaning of dynamical symmetry in classical Hamiltonian
mechanics.
Any transformation that leaves a differential equation invariant must
also convert a solution of the differential equation into a solution that can
be the same solution or a different one. In this respect the intrinsic symme-
tries of differential equations have consequences similar to intrinsic symme-
tries of algebraic polynomial equations. However such algebraic equations
possess a discrete set of solutions, while differential equations typically pos-
sess families of solutions. Members of these families may be distinguished
by continuously variable parameters, or in the case of partial differential
equations, continuously variable functions. For ordinary differential equa-
tions, ODEs, changing the values of the parameters corresponds to changing
initial values of the variables, which for Hamilton’s equations are positions

71
October 28, 2010 16:55 9in x 6in b943-ch03

72 Dynamical Symmetry

and momenta. Because the solutions of differential equations can be labeled

by continuous variables, most operations that interconvert the solutions are
operations of continuous groups, rather than that of a discrete group such
as that admitted by the equation w4 = 1 of the previous chapter. The
differential equation
dx/dt = 1 (3.1.1a)
has, for example, the particular solution x = t. The continuous group of
translations in time,
t → t
= t + α, (3.1.1b)
converts the particular solution into the one-parameter family of solutions
x = t + α, (3.1.1c)
which is the general solution of the equation. The differential equation
d2 x/dt2 = −x, (3.1.2a)
has the particular solution x = cos(t). The continuous group of time-
translations converts this to the one-parameter family of solutions
x = cos(t + α). (3.1.2b)
The group of dilatations
x → x
= eβ x, (3.1.2c)
converts (3.1.2b) to the general solution
x = eβ cos(t + α). (3.1.2d)
In (3.1.1), α is the group parameter, while in (3.1.2) the group parameters
are α and β.
As noted in Chapter 1, it was Lie’s investigations of the (intrinsic) sym-
metries of differential equations that led him to develop what is now known
as the theory of Lie transformation groups. Lie’s theory of the intrinsic
symmetries of differential equations involves several conceptual subtleties.
In the usual terminology, one says that the differential equations (3.1.1a)
and (3.1.2a) involve one independent variable, the argument t, and one
dependent variable, f(t). This statement can be misleading because, when
f(t) denotes a solution of a differential equation it is a function whose value
only becomes fixed when one fixes the value of further variables, parameters
of integration. Thus (3.1.1a) itself only relates dx to dt; it does not define
a definite functional relation between x and t. The solution is a function
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 73

of two variables, the t and α of (3.1.1c). The solution of the second-order

differential equation (3.1.2a) establishes a functional relation between the
four variables, the x, t, α and β of (3.1.2c). However, in each case, only t
appears in the differential equation as an argument.
It is important to keep in mind the distinction between relations defined
by definite functions, say y = f(t), and relations defined by differential equa-
tions, such as dy = g(t)dt. In the latter, y and t remain independently
variable. This distinction is fundamental to Lie’s work, to d’Alembert’s prin-
ciple, to Hamiltonian mechanics, to all variational treatments of mechanics,
and to all of science that is governed by differential equations.
The relation between transformations of functions, arguments, deriva-
tives, and solutions of differential equations, will be more thoroughly dealt
with in Chapters 5–7. The connection is often subtle enough to make it
difficult to anticipate the intrinsic symmetry of differential equations. Very
fortunately, the intrinsic symmetries of Hamilton’s equations are easy to
characterize. These intrinsic symmetries become geometric symmetries in
the phase–space of physics — a non-Euclidean metric space. In the fol-
lowing pages the reader will find that the non-Euclidean geometry of this
phase-space allows free reign to one’s Euclidean geometric imagination.

3.2 Hamilton’s Equations

Hamilton’s mechanics treats position and momentum as initially indepen-
dent variables on an equal footing. It deals with variables that may, for
example, be linear combinations of ordinary position and momentum vari-
ables. For this reason it is convenient to use dimensionless variables. In
the following discussion ordinary type will be used to denote dimensionless
variables in equations. The italic letters l, m, t will denote the units in which
length, mass and time, respectively, are measured. Thus:

q = length/l, t = time/t, m = mass/m, p = momentum/lmt−1,

E = energy/ml 2 t−2 , L = angular momentum/l 2 mt−1 .

We begin with the relation

E = H(q, p), (3.2.1)

where H(q,p) is the Hamiltonian function of a particle with energy E, q is

the Cartesian position coordinate of the particle of mass m, and p = mdq/dt
October 28, 2010 16:55 9in x 6in b943-ch03

74 Dynamical Symmetry

represents the particle’s momentum.a Hamilton’s equations of motion are

dqj /dt = ∂H/∂pj , dpj /dt = −∂H/∂qj , (3.2.2)
with dt representing a time interval.b A familiar example is provided by a
mass of magnitude m oscillating under the influence of a potential (k/2) q2 .
(We suppose that k, like m, is dimensionless.) For this harmonic oscillator,
H = p2 /2m + (k/2) q2 . (3.2.3)
Equations (3.2.2) become
dq/dt = p/m, dp/dt = −kq. (3.2.4a,b)
The well known family of solutions of these equations can be written
Q(t) = Q(0) cos(ωt) + (P(0)/mω) sin(ωt),
(3.2.4c–e)
P(t) = P(0) cos(ωt) − mω Q(0) sin(ωt), ω = (k/m)1/2 .
To gain further insight into the implications of these equations, it is
helpful to simplify their appearance by setting ωt = τ, and then defining the
new variables x = q (km)1/4 , and y = p/(km)1/4 . This changes (3.2.3) to
H = (ω/2)(y2 + x2 ), (3.2.5)
and Eqs. (3.2.4a,b) become
dx/dτ = ωy, dy/dτ = −ωx. (3.2.6a,b)
Solving these last equations converts x to X(τ ), y to Y(τ ), and determines
the two-parameter family of solutions
X(τ ) = X(0) cos(ωτ ) + Y(0) sin(ωτ ), (3.2.6c)
Y(τ ) = Y(0) cos(ωτ ) − X(0) sin(ωτ ). (3.2.6d)
Several of these solutions, circles of radius (ω/2)1/2 , are plotted in
Figs. 3.2.1. In the figures, the distinguished points may be considered
to be points with coordinates (X(0), Y(0)). If one views X(0) and Y(0)
as arbitrary values of x and y, then on referring back to the discussion of
Fig. 2.2.1 one is led to consider Eqs. (3.2.6) as equations that define a
transformation group

a Thismomentum variable is said to be conjugate to the q. The general definition of the

conjugate momentum, p, that must be used in Hamilton’s equations, and the equations
themselves, are derived from Lagrange’s equations; cf. Chapter 7.
b Though dt is a time interval, identifying t with physical time can be misleading, even

in nonrelativistic mechanics. Here, for example, t is a cyclic variable (cf. Sec. 5.7).
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 75

y y

R R'

x x
0 0

(a) (b)

Fig. 3.2.1 (a) A solution of Hamilton’s equations (3.2.6). (b) A solution of Eqs.
(3.2.6) with different initial conditions.

x → x
= X(τ ) = x cos(τ ) + y sin(τ ),

(3.2.7a,b)
y → y
= Y(τ ) = y cos(τ ) − x sin(τ ),

or, equivalently,

q → q
= Q(τ ), p → p
= P(τ ).

The point moves through a space of dimensionless position and momentum

variables, developing a trajectory in the phase–space as the physical system
evolves. The inverse motion is obtained by changing t to −t. At every point
along a trajectory, the motion of q and p is restricted by the equation H(q,p)
= E. The fully evolved trajectory is sometimes termed a phase portrait of
the motion. In this case, the phase portrait can be sketched by rotating
the representing point at (q,p) about the origin (0,0), with the angle of
rotation, α, equal to ωt.
The transformations (3.2.7) convert Eqs. (3.2.6a,b) to

dx
/dτ = y
, dy
/dτ = −x
, (3.2.8a,b)

that is, they leave Hamilton’s equations invariant. They also leave invariant
the energy of the system:



E = (ω/2)(y2 + x2 ) = (ω/2)(y 2 + x 2 )
October 28, 2010 16:55 9in x 6in b943-ch03

76 Dynamical Symmetry

or, equivalently,
E = H(q, p) = H(q
, p
). (3.2.9)
In Chapter 7 it is shown that these observations can be greatly generalized:
The family of solutions obtained by solving Hamilton’s equations for a
given mechanical system may be considered to define a continuous group
of transformations of its positions and momenta. This group is termed an
evolution group.

3.3 Transformations that Convert Hamilton’s Equations

into Hamilton’s Equations; Symplectic Groups
We have just observed that solutions of Hamilton’s equations define a
transformation group. We now characterize the general class of transfor-
mations that convert Hamilton’s equations into Hamilton’s equations when
the equations govern the evolution of a particle with only one position
coordinate and one momentum coordinate.
Transformations that leave invariant the equation H(q,p) = E, even for a
fixed value of E, do not necessarily convert Hamilton’s equations of motion
into Hamilton’s equations of motion. The transformation utilized to convert
(3.2.4a,b) to (3.2.5a,b) does both. On the other hand, the transformation
q → q
= p, p→ p
= q, leaves (3.2.3) invariant, but it does not leave
(3.2.4a,b) invariant. If a transformation does not leave Hamilton’s equations
invariant, the interpretation of E as an energy need no longer be valid.
When investigating transformations that convert Hamilton’s equations
into Hamilton’s equations one is dealing with functions that must possess
derivatives. We shall use the now standard term C k function to denote
a function whose derivatives of k’th order are all themselves continuous
functions for all real values of their arguments. For k > 0, a Ck function
is necessarily Ck−1 ; e.g., if a function is C5 , then it is C4 , . . . , C0 . If k is
infinite, the function is said to be smooth. Let
q → q
= Q(q, p), p → p
= P(q, p), (3.3.2a)
and
q
→ q = Qinv (q
, p
), p
→ p = Pinv (q
, p
), (3.3.2b)
define a one-to-one transformation and its inverse. If Q and P are Ck func-
tions of q and p, then (3.3.2a) is said to define a Ck transformation; and if



Qinv and Pinv are Ck functions of q
, p
, then (3.3.2b) defines a Ck trans-
formation. Together, the four equations define a Ck diffeomorphism, with
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 77

k being the smaller of k and k
.c Diffeomorphisms transform (q + dq, p +

dp) to (q
+ dq
, p
+ dp
) with
dq
= (∂Q(q, p)/∂q) dq + (∂Q(q, p)/∂p) dp,
(3.3.2c)
dp
= (∂P(q, p)/∂q) dq + (∂P(q, p)/∂p) dp.
The diffeomorphisms also transform (q
+ dq
, p
+ dp
) to (q + dq,
p + dp) with
dq = (∂Qinv (q
, p
)/∂q
) dq
+ (∂Qinv (q
, p
)/∂p
) dp
,
(3.3.2d)
dp = (∂Pinv (q
, p
)/∂q
) dq
+ (∂Pinv (q
, p
)/∂p
) dp
.
For reasons to be explained in Chapter 7, we require transformations
of position and momentum variables be C2 diffeomorphisms. These convert
the Hamiltonian equations of motion
dqj /dt = ∂H/∂pj , dpj /dt = −∂H/∂qj , (3.3.3a)
into the Hamiltonian equations of motion
dq
/dt = ∂H
(q
, p
)/∂p
, dp
/dt = −∂H
(q
, p
)/∂q
, (3.3.3b)
iff
∂q
/∂q ∂p
/∂p − ∂p
/∂q ∂q
/∂p = 1. (3.3.4)
Mathematicians term a transformation of q and p that satisfies (3.3.4) a
symplectic transformation.
To ensure that Hamilton’s equations of motion produce evolution trans-
formations that are C2 , their Hamiltonian must be a C3 function of q and p.
Unless otherwise noted, we henceforth require that H(q,p) is C 3 .
Equations involving expressions similar to that in (3.3.4) appear so often
in Hamiltonian mechanics that it is convenient to introduce the Poisson
bracket of any two functions A(q,p), B(q,p), written {A,B}, and defined
by
{A, B} = Σj (∂A/∂qj ∂B/∂pj − ∂B/∂qj ∂A/∂pj ). (3.3.5)
In the exercises at the end of the chapter, the reader is asked to establish
that when q → q
= A, p → p
= B, then if {A,B} = 1, the variables q
and

c C0 transformations do not define diffeomorphisms, they can define homeomorphisms.

A homeomorphism is a transformation which carries adjacent points with coordinates
(q1 ,p1 ), (q2 ,p2 ) to adjacent points with coordinates (q
1 ,p
1 ), (q
2 ,p
2 ), and has an inverse
which carries out the reverse transformation. In a C0 homeomorphism, the transforma-
tion and its inverse are continuous functions. In the Euclidean plane, a homeomorphism
can interconvert a square and a circle. A diffeomorphism cannot do so, because the
derivatives required in (3.3.2e,f) do not exist at the corners of the square.
October 28, 2010 16:55 9in x 6in b943-ch03

78 Dynamical Symmetry

p
are also conjugate variables. When this happens, one may treat (q
,p
)
in the same way one treats (q,p) in Hamilton’s equations.
The following one-parameter groups of transformations are all groups
of smooth (i.e., C∞ ) symplectic diffeomorphisms, and hence canonical dif-
feomorphisms:

q → q
= q + a, p → p
= p + b, (3.3.6a, b)

β
−β
q → q = e q, p→p =e p, (3.3.6c)
q → q
= q cos(α) − p sin(α), p → p
= p cos(α) + q sin(α), (3.3.6d)
q → q
= q cosh(γ) + p sinh(γ), p → p
= p cosh(γ) + q sinh(γ).
(3.3.6e)

Figures 2.3.1a,c, 2.3.3c and 2.3.4a in Chapter 2 depict smooth sym-

plectic transformations if one considers one of the variables x,y to be a ‘q’,
and the other to be its conjugate, ‘p’.
The symbolism Sp(2n,R) is used to denote a group of homogeneous lin-
ear symplectic transformations of 2n real variables. Taken together, trans-
formations (3.3.6c,d,e) comprise the 3-parameter group Sp(2,R). Any real
homogeneous linear transformation of (q,p) that is symplectic may be
expressed as a succession of the three transformations (3.3.6c–e). If the
translations (3.3.6a,b) are included, the group becomes an inhomogeneous
linear symplectic group, ISp(2,R). Many useful symplectic transformations
are linear, and many are nonlinear.
The symplectic transformations of the mathematician are a more
restricted class of transformations than those often called canonical trans-
formations by physicists. This is because physicists, in interconverting units
of measurement, often use scale transformations q → q
= aq, p → p
= bp
with a,b not equal to 1. Then (3.3.4) becomes ab rather than 1.

3.4 Invariance Transformations of Hamilton’s Equations

of Motion
Symplectic transformations

q → q
= Q
(q, p), p → p
= P
(q, p), (3.4.1a,b)

that convert H(q,p) to H(q
,p
) and convert Hamilton’s equations into
Hamilton’s equations, do not in general leave the Hamiltonian function,
H(q,p), unchanged. To determine whether H(q,p) is unchanged, one first
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 79

substitutes Q
(q,p) for q
, and P
(q,p) for p
in H(q
,p
). Then

H(q, p) → H(q
, p
) = H(Q
(q, p), P
(q, p)) = H
(q, p). (3.4.1c)
A symplectic transformation is an invariance transformation of Hamilton’s
equations with Hamiltonian H(q,p) iff H
(q,p) = H(q,p) = E. Then
dq/dt = ∂H
(q, p)/∂p → dq/dt = ∂H(q, p)/∂p,
and
dp/dt = −∂H
(q, p)/∂q → dp/dt = −∂H(q, p)/∂q. (3.4.2)
As a consequence, Hamilton’s differential equations possess the wonderfully
simple property that they are left invariant by a transformation of q and
p iff the transformation is a symplectic one which leaves the Hamiltonian
function itself invariant:
If t is not transformed, the study of the invariance transformations of
Hamilton’s equations of motion reduces to a study of invariance transfor-
mations of the equation E = H(q,p) under symplectic transformations of
the variables q and p.
The invariance transformations may be different for different values of E.
The characterization of invariance when t is transformed will be considered
in Sec. 3.7, below, and in Chapter 7.
The evolution transformations produced by Hamilton’s equations are
themselves invariance transformations of the equations of motion. They
carry every point on a solution curve into a point on the same curve; they
carry solutions into themselves. Other invariance transformations of the
equations of motion may interconvert solution curves.
Before proceeding further let us consider a further example that illus-
trates some of the foregoing discussion. The Hamiltonian of a free particle
of mass m is H = p2 /2m. Hamilton’s equations of motion become
dq/dt = p/m, dp/dt = 0. (3.4.3a,b)
For each value of E > 0 they determine the two separate straight-line tra-
jectories in q,p space defined by q = q0 + (p/m)t, with p = (2mE)1/2 ,
and p = −(2mE)1/2 . The two trajectories, illustrated in Fig. 3.4.1, are
interconverted by the symplectic transformation p → −p, q → −q, which
corresponds to a 180◦ rotation in q,p space.
Letting q → q
= q + a, also does not alter dq or p∂/∂q, so the
free-particle Hamiltonian, H(p), the energy, and the equations of motion
are all invariant under this transformation. The symplectic transformation
October 28, 2010 16:55 9in x 6in b943-ch03

80 Dynamical Symmetry

p
1

0.5

q
−1 − 0.5 0.5 1

− 0.5

−1

Fig. 3.4.1 Trajectories of a free particle in phase space.

p → p
= eβ p, q → q
= e−β q, converts the Hamiltonian equations of motion

into Hamiltonian equations of motion, but it converts H(p) into H(p
) =


p 2 /2m = e2β p2 /2m. As the transformation does not leave H invariant it
is not an invariance transformation of the equations of motion for a free
particle of mass m. It is, however, a class of scaling transformations that it
is useful in investigations of isotope effects and effective masses in quantum
mechanics.

3.5 Geometrization of Hamiltonian Mechanics

In Newtonian mechanics one thinks of motions as taking place in a Eucli-
dean space of at most three dimensions. Position vectors, r, are defined in
the space of Newton, as are infinitesimal displacement vectors dr, velocity
vectors v = dr/dt, and momentum vectors p = mv. Symmetries are often
defined in this Euclidean space, or in its relativistic generalization, space–
time. However, the intrinsic symmetries of Hamilton’s equations of motion
are all symmetries defined by symplectic transformations. Assigning geo-
metric expression to these intrinsic symmetries requires the development of
the geometry of spaces whose points have coordinates (qj , pj ) that may be
altered by symplectic transformations.
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 81

If one considers symplectic transformations to define active transforma-

tions in a phase–space, one can erect a fixed cartesian coordinate system
in it. Let us denote vectors in p,q space by s. For the moment, consider the
case n = 3, and let s = (q1 , q2 , q3 , p1 , p2 , p3 ) be a vector with cartesian
components qj and pj in the space. Defining the unit vectors iq , jq , kq , one
has q = (q1 , q2 , q3 , 0, 0, 0) = q1 iq + q2 jq + q3 kq . Similarly, a momentum
vector p = (0, 0, 0, p1 , p2 , p3 ) = p1 ip + p2 jp + p3 kp . The components
of any vector in this space can be obtained by determining its dot product
with six orthogonal unit vectors which are iq , jq , kq , ip , jp and kp . From
a mathematical standpoint, a momentum vector in ordinary space may be
obtained from this momentum vector by a mapping, such as a rotation
through 90◦ , that projects ip to iq , etc.
For ordinary three-dimensional vectors one defines the cross product of
two vectors, of which L = r × p is a prime example. An inversion of Carte-
sian coordinate axes through their origin has the effect of negating all the
Cartesian components of an ordinary vector V, which in the inverted sys-
tem becomes expressed as −V. Both r and p behave in such a manner, and
as a consequence L does not do so. For this reason L is sometimes termed
a pseudo-vector. As the cross product is only defined in three-dimensional
space, in Hamiltonian mechanics we shall replace Lx , Ly , Lz = L1 , L2 , L3 by

Lk = εijk qi pj , i, j, k = 1, . . . , 3. (3.5.1)

Summation over repeated indices is assumed, and εijk = +1 for clockwise

cyclic permutations of i, j, k = 1, . . . , 3, and equals −1 for counterclockwise
cyclic permutations. For noncyclic permutations of the indices, εijk = 0.
Thus, e.g., L3 = r1 p2 −r2 p1 . The Lk should not be considered components
of a vector in the six-dimensional phase–space of positions and momenta.
(They are components of a tensor.)
We next turn to an investigation of the elements of a plane geometry
in which only symplectic transformations of s = [q,p] are allowed. We will
then settle on the concept of symmetry that this geometry entails. Many
other key points will emerge from a study of the properties of this two-
dimensional real space, and it will be the subject of the remainder of this
chapter. In a Cartesian system, s = qiq + pip , with iq · ip = 0, and both
q and p real numbers. A moving mass point will be represented by the
motion of a point at the terminus of s. Hamiltonian equations of motion
govern canonical evolutionary motions of points in this space. Figs. 3.2.1
and 3.4.1 have illustrated this.
October 28, 2010 16:55 9in x 6in b943-ch03

82 Dynamical Symmetry

p
1.0

0.8

0.6

0.4

0.2
θ=t
q
−0.2 0.2 0.4 0.6 0.8 1.0

Fig. 3.5.1 Trajectory of an oscillator in phase space if k/m = 1.

Because the transformations we are considering are characterized only

by being C2 symplectic diffeomorphisms, if our space of q,p contains a given
point, it must contain all others that can be obtained from the given point
by any C2 symplectic diffeomorphism. Mathematicians call a space whose
points are all connected by symplectic diffeomorphisms, a symplectic space.
In two-dimensional symplectic space, the relation
{Q(q, p), P(q, p)} = 1 (3.5.2a)
implies
{Qinv (q, p), Pinv (q, p)} = 1. (3.5.2b)
Symplectic space differs from the phase–space of most theoretical
mechanics texts because, as previously noted, when physicists compare
observations they often utilize scale transformations that are not symplec-
tic.
Figure 3.5.2a depicts unit vectors i, j with q lying along i and p lying
along j. In Fig. 3.5.2b, q and p have been transformed to q
and p
by a
symplectic transformation. Figure 3.5.2c illustrates the effect an inversion
through the origin has on vectors A, B, C that define a triangle traversed
in a clockwise direction. The inversion is a symplectic transformation. In
Fig. 3.5.2d, the vectors A, B, C are reflected through the p axis along j.
This transformation is not a symplectic one. Note that in Fig. 3.5.2c the
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 83

p= j
1.0

j 1.0

0.8
j
0.8
p′ q′
0.6 0.6

0.4
0.4

0.2

0.2
i
−0.4 −0.2 0.2 0.4 0.6 0.8 1.0
i
q= i
0.2 0.4 0.6 0.8 1.0

(a) (b)

p p

j j

−C A A′ A
0.5 0.5

i i
q q
−0.5 0.5 −0.5 0.5

−B B B B

−0.5 −0.5
−A C C′ C

(c) (d)

Fig. 3.5.2 (a), (b) A sympletic transformation of p and q to p
and q
. (c)

Inversion of A, B, C. (d) A reflection of A, B, C.

inverted vectors A, B, C define a triangle that is traversed in the same,

clockwise, sense as the triangle defined by A, B, C. On the other hand,
the reflected vectors in Fig. 3.5.2d define a triangle that is traversed in a
counterclockwise direction.
The successive application of any two symplectic transformations, car-
ries a point in symplectic space into a point in the space, and is necessarily a
C1 canonical transformation with a C1 inverse. As the identity transforma-
tion is a symplectic transformation, taken together, symplectic diffeomor-
phims comprise a group of transformations. All the transformations convert
solutions of the differential equation {Q(q,p), P(q,p)} = 1 into solutions of
October 28, 2010 16:55 9in x 6in b943-ch03

84 Dynamical Symmetry

the same equation. Thus, the group of symplectic transformations is an

invariance group of the differential equation

∂Q/∂q ∂P/∂p − ∂P/∂q ∂ Q/∂p = 1. (3.5.3)

For any given function Q(q,p) this equation defines P(q,p) to within
an additive constant. Because one has available an uncountable infinity
of differentiable functions Q, the group of all symplectic transformations
contains an uncountable infinity of transformations. The inhomogeneous
linear symplectic group, ISp(2,R), is a five-parameter subgroup of this
group.
The linear symplectic transformation (3.3.6e), a hyperbolic rotation,
does not preserve the orthogonality of vectors. From this alone it follows
that the space defined by canonical transformations is not a Euclidean
space. However, as in spacetime, one may consistently erect an orthogonal
coordinate system in the space of active canonical transformations of q
and p. We have used a such coordinate system in Figs. 3.5.2. For it, the
Euclidean distance between nearby points with coordinates (q,p) and (q
+ dq, p + dp) is ds2 = dq2 + dp2 . Figures 3.5.3 depict the effect of
linear symplectic transformations (3.3.6) on four vectors A, B, C, D that
describe the square in Fig. 3.5.3a. The reader should imagine that in
each of the subsequent figures these vectors are transformed into vectors
A
, B
, C
, D
. Only translation and rotation transformations, which are
transformations of the Euclidean group, leave the lengths and orthogonality
of the vectors, and the Euclidean symmetry of the square unchanged. The
transformations (3.3.6 a,b,d,e) leave invariant a metric dq2 − dp2 , but the
symplectic dilatation (3.3.6c) does not. And, though (3.3.6c,d) leaves dq dp
invariant, (3.3.6e) does not.
In short, symplectic transformations do not in general preserve a
Euclidean metric, nor a metric analogous to that of special relativity, nor dq
dp. However, there is a quantity that is left invariant by transformations of
ISp(2,r). We shall shortly use it to define a metric in symplectic p,q space.
The cross product of Euclidean three-vectors may be used to define the
oriented area of each of Figs. 3.5.3 as (A × B)/k = Aq Bp − Bq Ap where
k = i × j. The oriented area of the square is (A×B)/k = Aq Bp = 1. In the
general case, if θa→b is the angle between A and B measured clockwise from
A to B, then (A×B)/k = |A||B| sin(θa→b ). The dimensionless area defined
here can be obtained by dividing the corresponding physical quantity, an
action, by the units of action, ml2 t−1 . These are units of energy, ml2 t−2 ,
multiplied by the units, t, of time.
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 85

0.5
0.5

−1 − 0.5 0.5 1 − 0.5 0.5 1 1.5

− 0.5
− 0.5

−1

−1

(a) (b)

1.5

1
0.5

0.5
−1 − 0.5 0.5 1

− 0.5

−1 − 0.5 0.5

−1

− 0.5

(c) (d)

0.5 0.6
0.4
0.2

−1 − 0.5 0.5 1 − 1.5 −1 − 0.5 0.5 1 1.5

− 0.2
− 0.4
− 0.5 − 0.6

−1

(e) (f)

Fig. 3.5.3 Action of transformation of ISp, (R). (a) Figure in phase space. (b)
Translation along q axis. (c) Translation along p axis. (d) Rotation. (e) Hyperbolic
rotation. (f) Sympletic dilatation.
October 28, 2010 16:55 9in x 6in b943-ch03

86 Dynamical Symmetry

The reader will find from (3.3.6) that the oriented areas enclosed in each
of Figs. 3.5.3, are related to one another by symplectic transformations, is
the same. The two triangles in Fig. 3.5.2c have the same oriented area. The
oriented areas in the two triangles in Fig. 3.5.2d have the same magnitude,
but opposite sign.
Adjacent oriented areas defined by the cross product have a property
illustrated in Figs. 3.5.3a. If the clockwise circuit A → B is continued
along the dotted line, and the clockwise circuit −A → −B along the dotted
line, the diagonal is traversed in opposite directions in the two different
circuits. It is evident that the total oriented area is the sum of the adja-
cent areas. If one of the oriented areas were defined by a clockwise circuit,
the other by a counterclockwise circuit, the dotted lines separating the
areas would be traversed in the same direction; the oriented areas would
have opposite signs, and the total oriented area would be the difference
of the two adjacent oriented areas. This is a general property of oriented
areas. It follows from the properties of the vector cross product. To sum-
marize, one has the following rule: If the boundary separating two adja-
cent oriented areas is traversed in opposite directions in circuits around
each sub-area, the oriented areas have the same sign, and the total ori-
ented area of the two is the sum of their individual oriented areas. This
is the area contained within the closed path that circumscribes the two
sub-areas.
In Hamiltonian mechanics one needs a generalization of the cross
product, one that defines oriented areas and actions in spaces of 2n dimen-
sions, rather than three dimensions. If A and B are vectors in a two-
dimensional p,q space we will use the (now standard) symbolism A ∧ B
to denote |A||B| sin(θa→b ). It is called the wedge product of A and B.
Because three points determine a plane, while four may not, it is best to
interpret A ∧ B as two times the oriented area contained within the trian-
gle formed by connecting the head of A to the tail of B, as has been done
in Figs. 3.5.2c and 3.5.2d.
So far we have only considered particular cases in which canonical trans-
formations leave areas invariant. To understand why all canonical transfor-
mations leave areas invariant, one need only understand how they leave
infinitesimal areas invariant — for any finite area can be obtained by inte-
grating infinitesimal areas. Two arbitrary infinitesimal vectors δA and δB
with a common origin determine a local two-plane. The infinitesimal area
defined by δA ∧ δB lies in this plane. One may install a coordinate system
with orthogonal i and j axes in the plane, then determine the components
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 87

of δA and δB on these axes. Let us apply these observations to equations

(3.3.2). They state that if a diffeomorphism converts

q → Q(q, p), p → P(q, p), (3.5.4a,b)

then

dq → a(q, p) dq + b(q, p) dp, (3.5.4c)

dp → c(q, p) dq + d(q, p) dp, (3.5.4d)

with

a = ∂Q/∂q, b = ∂Q/∂p, c = ∂P/∂q, d = ∂P/∂p. (3.5.4e)

One may choose axes such that the untransformed dq, which we shall term
dqo , lies along i, and dpo , the untransformed dp, lies along j. The right-hand
side of (3.5.4c) can be considered to define a vector, dA, and the right-hand
side of (3.5.4d) can be considered to be dB. And neither of these need lie
along i or j. Equations (3.5.4c,d) can then be written

dqo → dq = a dqo + b dpo , dpo → dp = c dqo + d dpo , (3.5.5a)

with

dqo = dqo i, dpo = dpo j. (3.5.5b)

These equations imply that dq and dp are not, in general, orthogonal.

Figures 3.5.4 depict the active interpretation of such a transformation,
one in which, at the point of interest, the derivatives in (3.5.4c,d) have
numerical values such that

dq = (3/2)1/2dqo + (1/2)1/2 dpo ,

(3.5.6)
dp = (1/2)1/2dqo + (3/2)1/2 dpo .

The oriented area defined by dqo ∧dpo in Fig. 3.5.4a is a special case;
the area defined by dq ∧ dp in Fig. 3.5.4b represents the more general
case, for which dq and dp are not orthogonal. As the point at (q,p) is
continuously shifted, this dq ∧ dp will change continuously to dq
∧ dp
,
the transformation of dq and dp being defined by (3.5.4). The oriented
areas dq
∧ dp
and dq ∧ dp are related via (3.5.5) and (3.5.6), with the
October 28, 2010 16:55 9in x 6in b943-ch03

88 Dynamical Symmetry

dp
dp'

dq dq'
dq x dp dq'x dp'

Fig. 3.5.4 Sympletic transformation of dq × dp to dp
× dp
.

general relation
dq
∧dp
= (a dq + b dp) ∧ (c dq + d dp)

= (ad − bc)dq∧dp (3.5.7a)

= (∂Q/∂q ∂P/∂p − ∂P/∂q ∂Q/∂p)dq∧dp.

If the transformation is symplectic, ∂Q/∂q ∂P/∂p − ∂P/∂q ∂Q/∂p = 1, so
dq
∧dp
= dq∧dp. (3.5.7b)
Hence, symplectic diffeomorphisms do not alter oriented infinitesimal areas.
A symplectic diffeomorphism that interconverts dq ∧ dp and dq
∧
dp will interconvert a region R and a region R
, and each area element dq

∧ dp within R will be converted to a corresponding area element dq
∧

dp
within R
, and vice versa. From (3.5.7) and the definition of the sum
of oriented areas, it follows that





dq
∧dp
= dq∧dp. (3.5.8)
R
R
Thus, symplectic diffeomorphisms do not alter oriented finite areas.
The transformation from (q,p) to (r, θ) defined by q = r cos(θ), p = r
sin(θ), is not symplectic, since {r,θ} = 1/r, and the area element is r dr dθ.
√ √
On the other hand, the transformation defined by q = r cos(2θ), p = r
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 89

p 2

0.5
1

q
1 1 1 1
(a) (b)

p p
1 1

q q
1 1 1 1

1 1

(c) (d)

Fig. 3.5.5 (c) Positive area. (d) Negative area.

sin(2θ), has {r,θ} = 1, so it is a symplectic, and therefore area-preserving,

transformation.
As an application of (3.5.8), consider the action developed when the
points with coordinates (q,p) lie along a smooth curve. Figures 3.5.5a,b
display oriented areas below two phase trajectories. The gray areas may be
considered as defined by integrations. In Fig. 3.5.5b the area under the
trajectory must be assigned a negative value. The trajectory in Fig. 3.5.5a
may be extended to produce a clockwise elliptical trajectory containing pos-
itive area. The total oriented area within the ellipse will be determined by
the addition rule given above. The oriented area and action can be obtained
directly by evaluating the double integral over the enclosed region:



A= dq∧dp. (3.5.9a)
R
October 28, 2010 16:55 9in x 6in b943-ch03

90 Dynamical Symmetry

The evaluation of such integrals in phase–space may often be simplified by

making the transformation q = r sin(θ), p = r cos(θ). Then



2
A = dθ rdr = dθ d(r /2), (3.5.9b)

the integral is positive if the trajectory is clockwise, and negative if it is

counterclockwise. Let the trajectory be defined by r = g(θ). Holding θ fixed,
and integrating r from 0 to g(θ) yields g(θ)2 /2. One thereby obtains

A = dθ g(θ)2 /2. (3.5.9c)

Two such signed areas are indicated in Figs. 3.5.5c,d. Physically, they
represent the action developed in moving a mass point with initial position
qa and conjugate momentum pa around a closed trajectory in phase–space,
while imparting to it the action defined by the enclosed area. The trajec-
tories may be considered as those produced by a harmonic oscillator with
Hamiltonian p2 /2 + q2 /2, and energy E = 1/2. The clockwise trajectory in
Fig.
√ 3.5.5d will develop as time advances if both p0 and q0 have the value
1/ 2. The counterclockwise√ trajectory in Fig.
√ 3.5.5c can be developed as
time advances, if q0 = 1/ 2 and p0 = −1/ 2. Note that this change in the
sense of rotation has been brought about by changing an initial value, and
not by changing t to −t.
Now let us return to Eqs. (3.5.7), and consider a more general geo-
metric significance of the fact that symplectic transformations do not alter
the value of dq ∧ dp. The relationship between symplectic transforma-
tions and the invariant dq ∧ dp is analogous to the relationship between
Euclidean transformations in an x–y plane and the Euclidean metric ds ·
ds = dx2 + dy2 .
In fact, dq ∧ dp is the metric that characterizes geometry in the q,p
space termed a symplectic space. The geometry of this space is the geometry
of a space with an invariant metric that is a measure of the oriented area
that represents action.
This result establishes a direct connection between symplectic geometry,
and the classical mechanics of Hamilton. The concept of action took on
increased physical importance in 1913 when Paul Ehrenfest1 argued that
the magnitude of the action developed by any closed trajectory in phase–
space is quantized. It appeared that it must be an integer multiple of Plank’s
constant divided by 2π:



|dq∧dp| = n h/2π. (3.5.10)
R
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 91

This straightforward connection between physics and symplectic geome-

try persisted until 1927, when Heisenberg demonstrated the impossibility of
measuring both positions and momenta with arbitrary accuracy.2 The con-
ceptual foundations of the old quantum theory then had to be abandoned,
and were replaced by a quantum mechanics in which conjugate position
and momentum variables satisfying {q,p} = 1 are replaced by operators
qop and pop satisfying the operator relation

qop pop − pop qop = i h/2π. (3.5.11)

The correspondence principle establishes connections between quantum

dynamics and classical dynamics that exist when the classical variables
q and p are replaced by operators that satisfy this relation.3 As stated by
Dirac, the principle was only applicable to polynomial functions of p
s and
q
s. And as mentioned previously, in Chapter 9 we show that Dirac’s cor-
respondence principle can be generalized in a manner which enables one
to directly establish the relation between invariance groups of Schrödinger
equations and invariance groups of Hamilton’s equations.

3.6 Symmetry in Two-Dimensional Symplectic Space

Knowing that the metric of two-dimensional symplectic space is dq ∧ dp
enables one to deal with the questions: “how are symmetries in this space
defined? When do two geometric objects in the space have the same sym-
metry?” Let us first deal with the first question.
We have seen that the symmetry of an object in Euclidean space is
defined by the subgroup of the Euclidean group, E(2), that carries it
into itself. And we have seen that analogous statements hold for three-
dimensional space, and for spacetime. Applying the same program to Hamil-
tonian mechanics, we shall define the symmetry of an object in symplectic
space by that subgroup of all symplectic transformations which carries the
object into itself.
Now let us turn to the second question. Two objects in Euclidean space
have identical symmetry if operations of the Euclidean group, and/or uni-
form dilatations can superimpose the objects. These operations do not alter
the symmetry of an object, they simply change the relationship of the object
to the coordinate system of an observer. An analogous statement was found
to hold for spacetime. However, in contrast to the metric in a Euclidean
space, the spacetime metric could be both positive and negative — a prop-
erty of real physical significance.
October 28, 2010 16:55 9in x 6in b943-ch03

92 Dynamical Symmetry

We shall suppose that two objects must certainly have identical sym-
plectic symmetry if they can be superimposed by symplectic transforma-
tions. As translations and rotations of a rigid figure do not change the
area enclosed by it, they may be used to test for superimposibility of two
objects in symplectic space. However, this by no means exhausts the catalog
of operations that can be used to superpose figures in symplectic space. All
symplectic diffeomorphisms can be considered to change the relationship
of an object to the coordinate system of an observer, without altering the
symplectic symmetry of the object. Before going on to say anything about
the geometric roles of uniform dilatations of dq ∧ dp, and of transforma-
tions that change its sign, we will, in the next section, consider the physical
significance of these operations.

3.7 Symmetry in Two-Dimensional Hamiltonian Phase

Space
The phase space of mechanics differs from symplectic space in several res-
pects. As we have already noted, changing initial values p0 and q0 can con-
vert a closed clockwise trajectory into a counterclockwise one, and do so
without changing the magnitude of the action. Now p0 and q0 are special
values of p and q, similiar to p and q, they may take on a continuous range
of real values. We conclude that, for this reason alone, transformations that
change the sign of dq ∧ dp should be allowed in the p,q space of Hamilto-
nian mechanics. There is another reason for admitting the transformation
dq ∧ dp → − dq ∧ dp as an operation that moves points in Hamiltonian
phase–space. On inspection, one will find that reversing the direction of a
motion described by A → B requires that one reverse the direction of the
vectors, and also traverse them in the reverse order. This converts A × B to
−B × −A, that is to −A × B. For the same reason, reversing the direction
of motion converts dq ∧ dp to −dq ∧ dp. As reversing the direction of
a motion may be brought about by the transformation t → −t, this time
reversal also changes the sign of dq ∧ dp. Time is a parameter in evolution
transformations in (p,q) space, just as is an angle is a parameter in rotation
transformations. Time reversal is an operation that leaves kinetic energies,
potential energies, and H(p,q) invariant. Time reversal defines a symme-
try of every time-independent Hamiltonian that is a quadratic function of
velocities.
There is also another reason, mentioned previously, for distinguishing
physical phase–space from symplectic space. Observers necessarily choose
units of measurement when measuring positions and momenta. When one
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 93

converts measurements from one system to another, the transformation

can be considered to be a dilatation, but it is seldom a symplectic one. For
example, in the conversion between MKS and CGS systems, the numerical
relations are qCGS = 102 qMKS , mCGS = 103 mMKS , tCGS = tMKS , pCGS =
105 pMKS , and the relation between the corresponding units of action is
qCGS pCGS = 107 qMKS pMKS .
Suppose, then, that a change of units multiplies q by a and p by b, both
a and b being positive real numbers. Such a transformation may always
be considered as the composition of a symplectic dilation and a uniform
dilation of q and p. To see this, let a symplectic dilatation convert q to q
=
eα q, and convert p to p
= e−α p. Let a = eν eα , and b = eν e−α . Then e2ν =
ab, and a/b = e2α . Consequently, the change of units is the composition of a
uniform dilatation with parameter ν = ln(ab)/2, and a symplectic dilation
with parameter α = ln(a/b)/2. For example, in changing MKS units to
CGS units, ab = 107 , a/b = 10−2 , so ν = (7/2)ln(10), and α = −ln(10).
Though scale changes may not convert H(p,q) to H(p
q
), they are
allowed operations in Hamiltonian mechanics, and we conclude they must
be allowed operations in the phase–spaces of Hamiltonian mechanics. We
therefore allow uniform dilatations as well as symplectic dilatations to carry
points in the phase–space of physics into points in the phase–space.
With these observations in hand we are prepared to define our two-
dimensional Hamiltonian phase–space, and the meaning of symmetry in it.

Definition of Two-dimensional Hamiltonian Phase–Space:

The Hamiltonian phase–space of the canonically conjugate variables q,
and p, is the space containing all points that may be carried into one another
by uniform dilatations, and C2 diffeomorphisms which leave invariant the
metric
ds2 = |dq∧dp|. (3.7.1)
Henceforth, we will use canonical transformation to denote a symplectic
transformation which is C2 . We will use canonical diffeomorphism to denote
a symplectic transformation with an inverse, both of which are C2 .

Definition of Symmetry in Two-Dimensional Hamiltonian

Phase–Space:
The symmetry of an object in two-dimensional Hamiltonian phase–space is
defined by the group of canonical diffeomorphisms which leaves ds invariant,
and carries the object into itself.
October 28, 2010 16:55 9in x 6in b943-ch03

94 Dynamical Symmetry

Two objects have the same symmetry in two dimensional Hamiltonian

phase–space if they can be superposed by canonical diffeomorphisms that
leave the magnitude of dq ∧ dp invariant, or multiply it by a positive real
number.
The transformations dq → − dq and dp → −dp each preserve the
value of dp2 + dq2 and |dq ∧ dp|, but not dq ∧ dp itself. These local
reflections, and their counterparts q → −q, p → −p which define ordinary
reflections in Euclidean space, interconvert clockwise and counterclockwise
motions in Hamiltonian phase–space. The symplectic symmetries which
they interconvert are identical symmetries in the Hamiltonian phase space.
Circles of different area have different symmetries in symplectic space,
and cannot be interconverted by symplectic transformations. However, they
have the same symmetry in the Hamiltonian phase–space because they can
be interconverted by uniform dilatations of p,q.
Rotations, the inversion defined by q → −q, p → −p taken together,
the transformation q → p, p → −q, and the transformation q → −p,
p → q — they all preserve the value of both dp2 + dq2 and dq ∧ dp. These
transformations, which define symmetries in the Euclidean plane, can also
define symmetries in two-dimensional symplectic space. The two Euclidean
reflections q → −q, and p → −p, leave dp2 + dq2 and |dq ∧ dp| invariant.
Consequently, symmetries in a Euclidean space of (p,q) are also symmetries
in the Hamiltonian phase–space of (p,q).
As hyperbolic pseudo-rotations in p,q space are also symplectic trans-
formations, they too can be used to define symmetries in Hamiltonian
phase–space, in the same way as they define symmetries in two-dimensional
spacetime.
Treating transformations in Hamiltonian phase–space as passive trans-
formations, and figures in the space as fixed objects, provides a direct way
to investigate the manner in which the transformations alter the relation
of an object to the (q,p) coordinate system of an observer. The uniform
dilatations, along with the reflections, inversions, and continuous trans-
formations of E(2) leave the object undisturbed, and simply change the
perspective from which it is observed. Clockwise and counterclockwise
motions then become identical motions observed from right-handed and
left-handed coordinate systems. Symplectic transformations in phase space
can interconvert Euclidean symmetries, just as Lorentz transformations in
spacetime can interconvert Euclidean symmetries. One must adopt a non-
Euclidean perspective if one wishes to consider the symmetry of an object
to be unchanged by transformations that are linear and non-Euclidean,
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 95

or nonlinear and non-Euclidean. In the first case, it is simplest to adopt

an active viewpoint. In general, the choice between active and passive
viewpoints must be determined by the physical context, and when this
is not decisive, the transformation is best considered to be a mapping. As
neither the concept of symmetry in phase space, nor its physical impli-
cations, are intuitively obvious, the remainder of this chapter is devoted
to developing examples that establish connections between the geometry
and the physics.

3.8 Symmetries Defined by Linear Symplectic

Transformations
We begin by considering three systems whose Hamiltonians cannot be inter-
converted by symplectic transformations. Each has a different symmetry in
symplectic space and Hamiltonian phase–space. Define free-particle, oscilla-
tor, and repulsive oscillator Hamiltonians, by respectively:
Hfp = p2 /2, HOSC = p2 /2 + q2 /2, Hrep = p2 /2 − q2 /2. (3.8.1)
Hfp is invariant under the translations q → q + a. HOSC is invariant under
rotations about the origin (q,p) = (0,0), and Hrep is invariant under hyper-
bolic pseudo-rotations about this origin.
The three Hamiltonians are all invariant under symplectic inversions
and the non-symplectic reflections allowed in Hamiltonian phase–space.
Figure 3.8.1a depicts the direction of motion along trajectories deter-
mined by Hfp when E = 1/2. It is designed to be used by the reader to
determine the consequences of reflections in the q axis and the p axis.
Figures 3.8.1b,c similarly depict trajectories determined by Hrep for E
= 1/2 and E= −1/2, respectively. Figures 3.5.5c,d may be used to
determine the manner in which reflections affect trajectories of HOSC .
Note that in every case, reflection interconverts clockwise and counter-
clockwise motions, as one expects, since each reflection changes the sign of
dq ∧ dp.
For each Hamiltonian in (3.8.1), the equations H(q,p) = E directly
determine non-directional phase portraits. The phase portraits for Hfp have
Euclidean translational symmetry in q, these for HOSC have Euclidean cir-
cular symmetry, and those for Hrep have non-Euclidean hyperbolic symme-
try. The directional portraits are determined by the equations of motion.
The successive composition of any of the transformations of the lin-
ear symplectic group ISp(2,R) will transform Hamilton’s equations into
Hamilton’s equations. Any object obtained by action of a symplectic
October 28, 2010 16:55 9in x 6in b943-ch03

96 Dynamical Symmetry

p
1

0.5

q
2 1 1 2

0.5

(a)

p p
2

1.5

1 1

0.5

q q
1.5 1 0.5 0.5 1 1.5 2 1 1 2

0.5

1 1

1.5

E = 1/2 E = -1/2

(b) (c)

Fig. 3.8.1 (a) Trajectories of a free particle in phase space. (b), (c) Trajectories
of republic oscillator in phase space.

transformation on the equation Hfp = const. has the same symplectic sym-
metry as Hfp = const. Any object obtained by action of symplectic trans-
formations on the equations HOSC = const. or Hrep = const., has the same
symmetry as HOSC = const. or Hrep = const., respectively.
The effect of several linear symplectic transformations that alter the
phase portrait of a free particle with Hfp = E = 1/2 is displayed in
Figs. 3.8.2. All have the same symplectic symmetry as Fig. 3.8.2a.
Figures 3.8.3b,c illustrate the effect of two linear symplectic trans-
formations that alter the phase portrait of an oscillator with HOSC = E
= 1/2. All these portraits in Figs. 3.8.3 are considered as having the cir-
cular symplectic symmetry of Fig. 3.8.3a.
Linear symplectic transformations that alter the phase portrait of a
repulsive oscillator with Hrep = E = 1/2 are depicted in Figs. 3.8.4b,c.
Again, the portraits have the same symplectic symmetry as the first por-
trait, Fig. 3.8.4a.
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 97

1.0 2

0.5 1

−2 −1 1 2 −2 −1 1 2

− 0.5 −1

− 1.0 −2

(a) (b)

1.5

0.5

−2 −1 1 2
− 0.5

−1

− 1.5

(c)

Fig. 3.8.2 Sp(2) transformation of phase portaits of systems with E = 1/2.

(a) Free particle: Grid [{{a,b}}, Spacings → {6,0}]. (b) Rotation of 2a. (c) Hyper-
bolic rotation of 2a: Grid [{{c,d}}, Spacings → {6,0}].

In this section we have considered all transformations as active. The

reader interested in geometrodynamics may wish to also consider their pas-
sive interpretation in some detail.

3.9 Nonlinear Transformations in Two-Dimensional Phase

Space
Because Hamilton’s equations are differential equations, they directly relate
the properties of adjacent points in phase–space. This is reflected in the
Poisson criterion for a canonical transformation; it is a local criterion.
As a consequence, Hamiltonian mechanics allows local symplectic changes
in scale, changes that vary continuously throughout phase–space. These
canonical transformations involve nonlinear transformations of q and p.
The action of any symplectic diffeomorphism, linear or nonlinear, will trans-
form Hamilton’s equations into Hamilton’s equations, and will not alter the
symplectic symmetry of the Hamiltonian or its phase portraits.
To obtain some of the consequences of these observations, consider,
for example, the transformations
q → q
= q exp(αqp), p → p
= p exp(−αqp). (3.9.1a)
October 28, 2010 16:55 9in x 6in b943-ch03

98 Dynamical Symmetry

1.0

0.5

− 1.0 − 0.5 0.5 1.0

− 0.5

− 1.0

(a)

0.5
0.5

−1 − 0.5 0.5 1 − 0.8 − 0.6 − 0.2 0.2 0.4 0.6 0.8

− 0.5
− 0.5

−1

−1

(b) (c)

Fig. 3.8.3 (a) Harmonic oscillator. (b) Hyperbolic rotation of 3a. (c) Sympletic
dilatation of 3a.

For them,
{p
, q
} = 1, (3.9.1b)
and
q = q
exp( − αq
p
), p = p
exp(αq
p
). (3.9.1c)
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 99

2 2

1 1

− 1.5 −1 − 0.5 0.5 1 1.5 −2 −1 1 2

−1 −1

−2 −2

(a) (b)

Fig. 3.8.4 (a) Repulisive “oscillator”. (b) Rotation of 4a.

Note that

p
q
= p q. (3.9.1d)

The transformations are smooth canonical diffeomorphisms for all real

values of α. They convert the oscillator Hamiltonian of (3.8.1) to
2
H
(q, p) = (p /2 + q2 /2) cosh(2α q · p) − (p2 /2 + q2 /2) sinh(2α q · p).
(3.9.2)
Figures 3.9.1–3.9.4. illustrate the phase portraits that arise when E = 1,
and α takes on the values 0, 1/2, 3/2, and 2.
The symmetry of all these figures in phase–space is the same: it is that
of a circle. The action developed in each trajectory is the same. In the old
quantum theory, the energy of each system would be the same.
If one chose to interpret the transformation (3.9.1a) in the passive sense,
one would plot the orbits in a coordinate system Qs , Ps related to a Q,P
Cartesian system by the inverse of (3.9.1a), that is by (3.9.1c). An observer
using these coordinate systems would, for each value of a, find all the orbits
to be circles. From the standpoint of the Q,P system, the coordinates would
have unit vectors whose orientations and lengths vary from point to point.
From this standpoint, no symplectic diffeomorphism can alter the Euclidean
symmetry of the circles of Figs. 3.5.5 and 3.9.1. From this standpoint,
all canonical transformations will simply provide the coordinate systems
defining different perspective snapshots of the circles.
October 28, 2010 16:55 9in x 6in b943-ch03

100 Dynamical Symmetry

0.5

−1 − 0.5 0.5 1

− 0.5

−1

Fig. 3.9.1 q
= qeaqp , p
= pe−aqp , a = 0.

1.5

1.0

0.5

− 1.5 − 1.0 − 0.5 0.5 1.0 1.5

− 0.5

− 1.0

− 1.5

Fig. 3.9.2 q
= qeaqp , p
= pe−aqp , a = 1

4.
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 101

−2 −1 1 2

−1

−2

Fig. 3.9.3 q
= qeaqp , p
= pe−aqp , a = 3

4.

−3 −2 −1 1 2 3

−1

−2

−3

Fig. 3.9.4 Trajectories with a = 0 and a = 1.

October 28, 2010 16:55 9in x 6in b943-ch03

102 Dynamical Symmetry

In the light of this example, one might, at first sight, suspect that all
closed curves have the same symmetry in phase–space. In concluding this
section we wish to show that this is not the case. Consider, for example,
the equation
H(q, p) = E, H = p2 /2 + q4 /2, (3.9.3)
that defines closed curves. It may be obtained from the equation



p 2 /2 + q 2 /2 = E, (3.9.4)
by the transformation p
→ p, q
→ q2 . However, aside from the fact that
this is not a symplectic transformation, no diffeomorphism can convert q

to q2 because the inverse transformation must convert both +q and −q to
the same value of q
.
In subsequent chapters we will consider transformations that are nonlin-
ear symplectic diffeomorphisms in 2N dimensional symplectic spaces. Some
of these will turn out to possess marvelous properties.

3.10 Dynamical Symmetries as Intrinsic Symmetries of

Differential Equations and as Geometric Symmetries
In this section we summarize and generalize the observations of the previous
sections of the chapter.
Hamiltonian mechanics provide an illustration of the manner in which
intrinsic symmetries of differential equations are defined, and then realized
as geometrical symmetries. We have seen that Hamilton’s equations are
converted to Hamilton’s equations by diffeomorphisms that convert a pair
of variables p,q to another pair of variables p
= P(q,p), q
= Q(q,p), which
satisfy the Poisson bracket relation {q,p} = {q
,p
} = 1. Such a pair of posi-
tion and momentum variables is termed conjugate and the diffeomorphisms
that leave {q,p} invariant are termed symplectic.
Geometrization begins when positions and momenta are represented by
coordinates of a point in a space. Vectors q, p and dq, dp are introduced,
and considered to have components on orthogonal axes. It is then required
that all points in the space can be reached by symplectic diffeomorphisms,
transformations that leave invariant a metric, dq ∧ dp. The transforma-
tions that convert Hamilton’s equations to Hamilton’s equations leave this
metric invariant. The space of all points connected by symplectic diffeomor-
phisms is given the name symplectic space. The transformations of Hamil-
ton’s equations can be interpreted as geometric operations in this space. For
October 28, 2010 16:55 9in x 6in b943-ch03

On Symmetries Associated with Hamiltonian Dynamics 103

a given Hamiltonian, a symplectic transformation of Hamilton’s equations

either leaves them invariant, or transforms them into those corresponding
to a different Hamiltonian. Symplectic transformations need not preserve
the orthogonality of vectors, so symplectic space is non-Euclidean. However
the set of all symplectic diffeomorphisms contains a Euclidean subgroup.
In order to compare physically significant intrinsic symmetries of equa-
tions arising from different Hamiltonians, it is necessary to allow dilatations
and reflections, as well as symplectic diffeomorphisms, to alter positions and
momenta. In the resulting space, which we have termed Hamiltonian phase
space, intrinsic symmetries of Hamilton’s equations become geometric sym-
metries.
In Chapter 7 this geometrization of Hamiltonian mechanics will be
extended to equations involving n different q
s and their conjugate p
s, a
2n-dimensional PQ space. The transformation from Hamiltonian dynam-
ics to geometry will thereafter be completed by treating time itself as
a position-like variable. Energy (or rather, its negative) then becomes a
momentum-like variable. The enlarged space is termed PQET space. Evolv-
ing trajectories in PQ space become stationary curves in PQET space. In
this phase space all dynamics becomes geometry, and every intrinsic dynam-
ical symmetry can be given geometrical interpretation.
Both the meaning and role of symmetry in Hamiltonian mechanics is
evidently larger than physical experience might seem to suggest. This rich-
ness of meaning, and of implication, has its origin in a property of nature,
unknown until the time of Newton and Liebniz: the laws governing the
spatial and temporal development of natural processes can often be
expressed by differential equations. It was over two centuries after the time
of Newton and Leibniz when Lie first realized that differential equations
possess, and define, symmetries. The path taken through the subject mat-
ter of this chapter is a path guided by Lie’s conceptions.
The intrinsic symmetries of most differential equations governing phys-
ical systems are larger, and different, than would be expected from a
knowledge of symmetries in ordinary two- and three-dimensional space,
or four-dimensional spacetime. There are fundamental reasons for this:
(1) Differential equations may have more independent variables than argu-
ments because, without initial conditions or boundary conditions, dif-
ferential equations do not establish fixed functional relations. This has
the consequence that the spaces in which dynamical symmetries find
October 28, 2010 16:55 9in x 6in b943-ch03

104 Dynamical Symmetry

geometrical expression can be spaces of dimension greater than that of

ordinary space or spacetime.
(2) The intrinsic symmetries of differential equations are in part determined
by purely local relations, relations between dx’s, dt’s, etc., that do not
of themselves establish, or depend upon, fixed functional relations.
(3) As these local intrinsic symmetries often define metrics that are not
Euclidean metrics, the symmetries often find geometric realization in
non-Euclidean spaces.
The next three chapters lay the mathematical groundwork needed to
understand and exploit Lie’s discoveries in studies of the intrinsic symme-
tries of dynamical systems governed by differential equations.
Exercises:
1. Does the transformation q → p, p → −q, determine a symplectic diffeo-
morphism?
2. Find the phase–space trajectories determined by the Hamiltonian,
H = q p, and plot representative examples of them.
3. After inspecting phase–space trajectories, show that no diffeomorphism
in p q space can interconvert the Hamiltonian of the harmonic oscillator
and: (a) that of the previous exercise; (b) the free-particle.
4. Let r = (q2 + p2 )1/2 be a position variable. Show that if f(q,p) is the
corresponding momentum variable, it satisfies the differential equation
(1/r)(q ∂/∂p − p ∂/∂q)f = 1,
and find a function f(q,p) that satisfies this equation. What is its general
solution?
5. Let (3.3.2c) and (3.3.2d) arise from a canonical diffeomorphism relating
(q,p) and (q
,p
). Show that
∂Q/∂q ∂P/∂p − ∂P/∂q ∂Q/∂p = 1
implies
∂Q
/∂q
∂P
/∂p
− ∂P
/∂q
∂Q
/∂p
= 1.

References
[1] P. Ehrenfest, Verh. Deutsch Physikal. Ges. 15 (1913) 451; Physik. Zeitschr.
15 (1914) 657; Phil. Mag. 33 (1917) 500.
[2] W. Heisenberg, Z. Physik 43 (1927) 172; cf., W. Heisenberg, Physical Prin-
ciples of the Quantum Theory (University Chicago Press, 1930).
[3] P. A. M. Dirac, Proc. Roy. Soc. A 109 (1926) 642.
October 28, 2010 16:55 9in x 6in b943-ch04

CHAPTER 4

One-Parameter Transformation Groups

4.1 Introduction
Previous chapters have considered examples of continuous groups of trans-
formations — groups such as those of translation, rotation, hyperbolic
rotation, and dilatation, which act on a set of real variables x with transfor-
mations that depend upon a set of continuous real variables, termed group
parameters. The parameters are not members of the set of variables x. This
chapter deals with fundamental properties possessed by groups of trans-
formations whose operations are labeled by one continuously variable real
parameter. In this and the next six sections we will sometimes distinguish
between initial values, final values, and running values lying between the
initial and final values of variables. Primes will be used to denote running
values and asterisks to denote final values. Variables with neither primes nor
asterisks will denote initial values. We begin with examples that illustrate
key points.
Figure 4.1.1 depicts an active interpretation of the dilatation trans-
formation

x → x
= T(α
)x, with T(α
)x = x exp(α
). (4.1.1a)

As the running parameter α
increases from 0, a point P, with initial coor-

dinate x, takes on coordinates x
, and moves along curves to a final point
at which x
≡ x∗ , α
≡ α:

x∗ = T(α)x = x exp(α). (4.1.1b)

The single-valuedness of the function x exp(α
) has an important geometric

consequence: it ensures that only one curve passes through each point.
In Chapter 2 we assumed that the transformations (4.1.1) are those
of a continuous group. Let us prove that this is so. For T(α) to define a

105
October 28, 2010 16:55 9in x 6in b943-ch04

106 Dynamical Symmetry

χ
2

α
− 1.0 − 0.5 0.5 1.0

−1

−2

Fig. 4.1.1 Action of Dilatation Transformation x
= xeα .

continuous group of transformations of x that depends on a parameter, α,

it is first of all necessary that there be an identity transformation, I, for
some value of the parameter, α0 , such that

T(α0 )x = I x = x. (4.1.2a)

In this case, α0 = 0. Next, let the transformation T(α)x yield xα , and let
it be followed by a transformation of xα given by T(β) i.e.

T(β)xα = T(β)T(α)x. (4.1.2b)

For the transformations to comprise a group it is necessary that

T(β)T(α)x = T(γ)x, γ = φ(β, α). (4.1.2c)

Since (4.1.2a) and (4.1.2b) hold for a range of x, they state a more abstract
property of the group which may be expressed as

T(β)T(α) = T(γ), γ = φ(β, α). (4.1.2d)

The expression γ = φ(β, α) is said to define the composition law of the

group.
The requirement that every transformation of a group have an inverse
is the requirement that for every value of α there must be a value, β such
that

T(α)T(β) = I; φ(α, β) = α0 . (4.1.2e)

The associative law

T(γ)(T(β)T(α)) = (T(γ)T(β))T(α), (4.1.2f)

October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 107

requires that
φ(γ, φ(β, α)) = φ(φ(γ, β), α). (4.1.2g)
For the transformation (4.1.1), with
x → T(α)x = x exp(α) = x∗ , (4.1.3)
the identity transformation is obtained by setting α = 0. For the composi-
tion of two transformations one has
T(γ)x = T(β)(T(α)x) = T(β)x exp(α) = x exp(β) exp(α) = x exp(α + β).
(4.1.4)
The composition law is γ = α + β. The identity transformation is produced
if β = −α; consequently the inverse of T(α) is T(−α). Note that this inverse
transformation, from x∗ to x in (4.1.3) may also be obtained by solving
equation (4.1.3) for x.
The reader can directly verify that all of the relations (4.1.2) hold true
for a continuous range of the parameters α, β, γ. Once the reader also
verifies that the associative law, Eq. (4.1.2f) holds, it will become evident
that Eq. (4.1.3) defines a continuous transformation group.
So far, in Eq. (4.1.2), we have not considered the range allowed for the
parameters α, β, and γ. A continuous transformation group acting on real
variables x is termed global if the group parameter may run through the full
range of the reals, −∞ < α
< ∞, without carrying real variables x outside
of the range of the reals. In global groups, α
and α
mod some number, e.g.
2π, may give rise to the same transformation. The reader may verify that
the transformation group defined in this example is a global one-parameter
continuous group.
Figure 4.1.2 illustrates curves determined by the transformations x
=
T(α
)x with
x
= x/(1 − α
x). (4.1.5)
The identity transformation is obtained when α
→ 0. A transformation
with parameter α, followed by one with parameter β, gives
x

= T(β)T(α
)x = T(β)x
, (4.1.6a)
so
x

= x
/(1 − βx
), with x
= x/(1 − α x). (4.1.6b)
Thus
{x/(1 − αx)}
x

= . (4.1.6c)
(1 − β{x/(1 − αx)})
October 28, 2010 16:55 9in x 6in b943-ch04

108 Dynamical Symmetry

χ
30

20

10

α
− 0.5 0.5 1.0 1.5
− 10

− 20

− 30

− 40

Fig. 4.1.2 Action of the projective transformation x
= 1

1−α
x .

On rationalizing the denominators one finds

x

= x/(1 − γx) with γ = α + β. (4.1.6d)

It follows that the identity transformation is obtained if β
= −α
. The

reader may verify that all the group requirements are met if x and the
group parameter, say α
, do not take on values such that 1 − α
x = 0.
A local continuous transformation group of real variables is a transfor-
mation group whose group parameter may have to be restricted in range
to ensure that it transforms real variables into real variables. The transfor-
mation x → x
= x/(1 − α
x) defines a local transformation group, but not
a global one.
These concepts of real local and global groups may be extended by
allowing the variables x and/or the parameters α to become imaginary or
complex.

4.2 Finite Transformations of a Continuous Group Define

Infinitesimal Transformations and Vector Fields
Let the Greek letter α denote a real group parameter that yields the identity
transformation when it equals 0. It may take on running values α
between
0 and α, in a range lying between −∞ and +∞. Let x stand for a set of
real variables xj that may also take on values x
j between −∞ and +∞.
Consider all variables to be running ones, and write

x

j = Fj (x
, α
). (4.2.1a)

October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 109

The requirement that the Fj (x
, α
) are continuous functions of their argu-
ments implies that as α
→ 0,
x

j = Fj (x
, α
) → x
j = Fj (x
, 0). (4.2.1b)
Let us also require that the derivatives
dx

j /dα
= dFj (x
, α
)/dα
, j = 1, 2, . . . , n, (4.2.2a)
are continuous real functions of x
, α
. Define
fj (x
) = dFj (x
, α
)/dα
)|α
=0 , (4.2.2b)
then, as α
→ 0, Eqs. (4.2.2a) become
dx
j /dα
= fj (x
), j = 1, 2, . . . , n. (4.2.2c)
Given these requirements, the transformation group determines a set of
first-order ordinary differential equations in which −∞ < fj (x
) < + ∞.
One may suppose that (4.2.2c) relates infinitesimal changes, δα
in the
parameter α
and infinitesimal changes δx
in the x
j , via the n equations
δx
j = fj (x
)δα
, j = 1, 2, . . . , n. (4.2.3a)
From a physical standpoint one may consider that, in some system of
measurement, these equations extrapolate an observed relationship
∆x
j = fj (x
)∆α
(4.2.3b)
that continues to hold within experimental error as the observer makes
increasingly refined observations. As an example of this, consider the group
of mathematically defined functional relations
x∗ = x cos(α) − y sin(α), y∗ = y cos(α) + x sin(α). (4.2.4)
For running values of the variables one has
(dx

/dα
)|α
=0 = (−x
sin(α
) − y
cos(α
))|α
=0 = −y
, (4.2.5)
and
(dy

/dα
)|α
=0 = (−y
sin(α
) + x
cos(α
))|α
=0 = x
.
Also, when α
→ 0, the variables x

→ x
, and y

→ y
, so one may write
dx
/dα
= −y
, dy
/dα
= x
. (4.2.6)
An experimentalist studying rotations in the plane might make measure-
ments of increasingly minute values of ∆x
, ∆y
, and a change in angle, ∆α
.
October 28, 2010 16:55 9in x 6in b943-ch04

110 Dynamical Symmetry

After taking account of experimental errors in the observations, the exper-

imentalist might extrapolate the results to
δx
/δα
= −y
, δy
/δα
= x
, or δx
= −y δα
, δy
= x
δα
, (4.2.7)
and imagine δx
, δy
, δα
to represent arbitrarily small, but nonzero,
infinitesimal quantities. In obtaining these relations, the experimentalist, in
contrast to the mathematician, has made no assumptions about functional
relations and mathematical limits.
However, after obtaining relations (4.2.7) for many repeated experi-
ments, the experimentalist would conclude that functional relations exist
between x
, y
, α
. For a point Px
,y
that is rotated about the origin, either
set of equations, (4.2.6) or (4.2.7) can define infinitesimal transformations.
Keisler has established the conditions under which infinitesimal quantities
such as δx are well defined, without necessarily being obtained by the lim-
iting process mathematicians use to define differentiation of a function.1
Physicists often switch between the two views. We shall do so as well.
The geometric consequences of (4.2.6) and (4.2.7) are suggested by
Figs. 4.2.1a,b. In them, the infinitesimal increments δx
, δy
, δα
are
replaced by small finite increments ∆x
, ∆y
, ∆α
. The vector fields illus-
trated in Figs. 4.2.1a,b, may be considered to be composed of vectors
tangent to circles and spirals such as those depicted in Figs. 4.2.1c,d. The
figures could just as well represent electric or magnetic force fields deter-
mined by solving Maxwell’s equations, or determined by measurements such
as those Maxwell describes in his monograph Electricity and Magnetism.2
Taking a mathematical approach, one finds by differentiating (4.1.3)
that the infinitesimal transformation associated with the dilatation trans-
formation can be expressed either as
dx
/dα
= x
, or as δx
= x
δα
. (4.2.8)
In the same manner, one finds that the infinitesimal transformation asso-
ciated with the finite transformation (4.1.5) is
δx
= (x
)2 δα
. (4.2.9)
A further geometric example is provided by the transformations
depicted by the paths and vector fields of Figs. 4.2.2. The infinitesimal
transformations that define the vector fields are:
δx
= δα
, δy
= (y
2 − x
)δα
. (4.2.10)
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 111

1.0
1

0.5

0
− 1.0 − 0.5 0.5 1.0

1
− 0.5
−1
0 y
−1
0
− 1.0 −1
x
1

(a) (b)

1.0

0.5

− 1.0 − 0.5 0.5 1.0

− 0.5

− 1.0

(c) (d)

Fig. 4.2.1 (a) Vector field ∆r = [∆x, ∆y] defined by infinitesimal rotations;
(b) vector field [∆x, ∆y, ∆α] defined by infinitesimal rotations; (c) path curves
of successive finite rotations in the x–y plane; (d) path curve in the space of
(x, y, α).

The finite transformations that move points along the indicated lines can be
approximated by connecting the arrows. For the figure, this has been done
by numerically integrating Eqs. (4.2.10), beginning at an initial point on
each line. In this last example we have not produced a general equation that
defines the finite transformations. The reason for this will become apparent
in Sec. 4.8 below.
October 28, 2010 16:55 9in x 6in b943-ch04

112 Dynamical Symmetry

y
y
0.0 x
3 0.5 1.0 1.5 2.0 2.5 3.0

2
− 0.5

− 1.0

x
− 0.5 0.5 1.0 1.5 2.0 2.5

− 1.5

−1

Fig. 4.2.2 Vector field and finite transformation determined by infinitesimal

transformations of Eq. (4.2.10).

4.3 Spaces in which Transformations will be Assumed

to Act on
We will shortly begin a general investigation of the use of differential
equations to define groups of finite transformations. Before giving geomet-
ric realizations to the transformations, it is necessary to be clear about
the spaces in which the transformations act/move points about. Let, i.e.
x = (x1 , x2 , . . . , xn ) represent a collection of n real variables xj , and let
a be a real parameter. Consider transformations that convert x → x∗ =
(x∗1 , x∗2 , . . . , x∗n ) = X(x, α), i.e., for j = 1, 2, . . . , n, one has x∗j = Xj (x, α).
As before, we will write x∗ = T(α)x when we are concerned only with
the relation between initial and final values of x and the parameter, and
use the notation x
= T(α
)x when we wish to emphasize that x
and α

can vary continuously. The parameters α, α
will always be allowed to vary
over all the reals.
We will deal with transformations that may be nonlinear. As the trans-
formation (4.1.5) demonstrates, such transformations have properties which
make it necessary to proceed with mathematical caution. Unless otherwise
noted, throughout this book we will be dealing with the set RN of N
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 113

continuous real variables zi that may be independently varied over the full
open range of the reals, i.e. −∞ < zi < ∞. Variations of zi that are arbi-
trarily small but nonzero, that is, infinitesimal, will hereafter be denoted
by dzi or δzi . Except when otherwise noted, the variables zi will represent
Cartesian coordinates of points PZ in a Euclidean space with metric

δs2 = δz21 + δz22 + · · · + δz2n . (4.3.1)

This space is termed EN . Real variables that do not necessarily have all
the properties of the Cartesian variables zi, will be generically denoted as
xi ’s. Often the x’s will be geometrically interpreted as defining points on
a manifold in a Euclidean space. Thus the angular coordinates (θ, φ), of a
point on the unit sphere defined by z21 + z22 + z23 = 1, are x’s, as are the z’s,
but θ, φ, are not z’s. The time coordinate in spacetime will be considered
an ‘x’, except when otherwise stated.
We follow current mathematical usage and require that, by definition, a
function is single-valued. Thus, for example, the relation z3 = ±(z21 − z22 )1/2
does not define a function; it defines two functions which have misleadingly
been denoted by the same symbol, z3 .
As previously noted we will be dealing with transformations which, like
stereographic projections and canonical transformations, are more general
than the linear transformations of vector algebra. Letting x represent a set
of variables, x1 , . . . , xn , their transformation from x to x
will be defined by
sets of functions:

x
1 = F1 (x), x
2 = F2 (x), . . . , x
n = Fn (x). (4.3.2a)

The inverse transformation will be denoted by

x1 = Finv

1 (x ), x2 = Finv
inv

2 (x ), . . . , xn = Fn (x ). (4.3.2b)

The transformation x → x
= x/(1 − αx) has as inverse the transformation

x
→ x = x
/(1 + α
x
). These equations define one diffeomorphism when
1 − αx < 0, or 1 − α
x
< 0, and another when 1 − αx > 0, or 1 − α
x
> 0.
They do not define a diffeomorphism when 1 ± αx or 1 ± α
x = 0.
Lines, surfaces, and hypersurfaces in EN may be defined by sets of diffeo-
morphisms stating functional relations amongst Cartesian coordinates zi .
The number of independent coordinates required to uniquely fix a point
on a manifold is the dimension of the manifold. A manifold is a smooth
manifold if the functions defining it possess partial derivatives of all orders
at it. The two-dimensional surface of a sphere of fixed finite radius is, for
October 28, 2010 16:55 9in x 6in b943-ch04

114 Dynamical Symmetry

example, related to the Cartesian coordinates in Euclidean 3-space by fixing

the value of r in the equations
z1 = r sin(θ) cos(φ), z2 = r sin(θ) sin(φ), z3 = r cos(θ). (4.3.3)
Derivatives of the z’s with respect to θ and φ are well defined functions
however high their order. The surface of the sphere is thus a smooth man-
ifold. Theorems due to Whitney ensure that any smooth n-dimensional
manifold may be embedded in a Euclidean space of no more than 2n
dimensions.3 In this case, a two-dimensional manifold has been embedded in
three-dimensional space. Whitney’s embedding theorems ensure that one’s
natural tendency to use Cartesian coordinates in Euclidean spaces need not
necessarily lead one astray.
The equations defining a transformation may, or may not, establish a
one-to-one correspondence between points Pz and Pz
in EN , or between
points in EN and points on a manifold of the same or lesser dimension.
For example, the stereographic projections illustrated in Chapters 2 and
3, though differentiable, establish a one-to-one transformation between all
points in the infinite plane and points on the surface of a sphere, except
the point at the south pole. Objects that can be interconverted by a dif-
feomorphism are said to be diffeomorphic. In the Euclidean plane, a circle
and an ellipse are diffeomorphic; a circle and a figure eight are not. Some
further concepts related to the concept of diffeomorphism are discussed in
the Appendix.

4.4 The Defining Equations of One-Parameter Groups

of Infinitesimal Transformations. Group Generators
Now, let us return to the problem of determining the relation between
infinitesimal and finite transformations. We first note that the set of equa-
tions
δx
j = δα
ξj (x
), j = 1, . . . , n, (4.4.1a)
or, equivalently, the set of equations
dx
j /dα
= ξj (x
), j = 1, . . . , n, (4.4.1b)
define a one-parameter group of infinitesimal transformations T(δα). This
follows from the following observations. One has:
1. T(δα)T(δβ) = T(δγ), δγ = δα + δβ; (a law of composition)
2. T(0) = I; (an identity operation)
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 115

3. T(−δα)T(δα) = I; (an inverse for every

transformation)
4. T(δα)(T(δβ)T(δγ)) = (T(δα)T(δβ))T(δγ). (associativity)
(4.4.2a–d)
Associated with the infinitesimal transformation defined by Eqs. (4.4.1)
is the group generator, the operator
U = Σn1 ξj (x
)∂/∂x
j . (4.4.3a)
The generator, U, of rotations, determined from (4.2.6), is x
∂/∂y
−y
∂/∂x
.
As each x
signifies a variable which may take on all values from x
through x∗ . For U to act on x means that Ux ≡ Ux
|x
=x . This is equivalent
to letting
U → Σni ξj (x)∂/∂xj . (4.4.3b)
One has
(1 + δα U)xi = xi + δα ξi (x) = x
i ; x
i = xi + δxi . (4.4.4)
The operator (1 + δα U) carries out infinitesimal transformations: it is an
expression for T(δα).
Taken in the active sense, (1 + δαU) moves a point with coordinates x
to the point with coordinates x
= x + δx = x + δαUx. A point so moved
is termed an ordinary point. However, if all the functions ξi (x) vanish at
a point, Px , the point is not moved by the transformation, and is termed
an invariant point or critical point. The point at the origin x = 0 is an
invariant point of the transformation determining Fig. 4.4.1, and of two
of the transformations used as examples in Sec. 4.3.
If one supposes the variables xi to be Cartesian variables zi in a
Euclidean space, then
Uzi = ξi (z), (4.4.5a)
and one may write
U = ξ · ∇, ∇ = (∂/∂z1 , ∂/∂z2 , . . . , ∂/∂zn ). (4.4.5b)
At each point Pz the vector ξ points in the direction that the infinitesimal
transformation moves Pz . The vector fields of ∆r’s in previous figures are
depictions of vectors, ξ. The path curves in Fig. 4.4.1 represent trans-
formations in Euclidean space generated when ξ = (y
z
, −x
z
, x
y
). The
tangent vectors to the path curves at the points Px
,y
,z
are ∆α times the
ξ at the respective points.
October 28, 2010 16:55 9in x 6in b943-ch04

116 Dynamical Symmetry

y
0.5 0 −0.5

0.5

0
z

−0.5

0.5
0
−0.5 x

Fig. 4.4.1 Two finite transformations of a one-parameter group with generators

V = X·∆, X = (zy, −zx, xy).

Because the generators U are first-order differential operators they

possess the property of derivatives
Uf(x)g(x) = g(x)Uf(x) + f(x)Ug(x). (4.4.6)

Making use of (4.4.4), the composition of two successive transformations

of x with parameters δα1 and δα2 may be expressed as
(1 + δα2 U)(1 + δα1 U)x = (1 + (δα2 + δα1 )U)x + δα2 δα1 UUx. (4.4.7a)

Thus, if UUx is finite,

(1 + δα2 U)(1 + δα1 U)x = (1 + (δα2 + δα1 )U)x + O(δ 2 ). (4.4.7b)
This equation suggests a number of relations that will be investigated in
the remainder of the chapter.

4.5 The Differential Equations that Define Infinitesimal

Transformations Define Finite Transformation Groups
In this section we establish the conditions required for the set of equations

δα
= δy1
/ξ1 (y
) = δy2
/ξ2 (y
) · · · = δyn
/ξn (y
), (4.5.1)
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 117

which define infinitesimal transformations of the yj
, to also define a group

of finite transformations. We begin by observing that for each value of j
from 1 to n, Eqs. (4.5.1) imply that

b
yb
δα
= ∆ba = b − a = δyj
/ξj (y
). (4.5.2a)
a ya

Similarly,

c
yc
δα
= ∆cb = c − b = δyj
/ξj (y
). (4.5.2b)
b yb

Path curves developed by successive integrations in a three-dimensional

space are illustrated in Figs. 4.5.1a,b.
Because the process of integration possesses the composition property

c
b
c
dg + dg = dg, (4.5.3)
b a a

it follows that

b
c
c
δα
+ δα
= δα
, (4.5.4a)
a b a

just as

(b − a) + (c − b) = c − a. (4.5.4b)

Using this observation one easily establishes that integration can be con-
sidered an operation that defines a group in which the law of composition
is addition.
Because each of the terms δyj
/ξj (y
) in (4.5.1) satisfy the relation
δyj
/ξj (y
) = δα
, one also has

yc
yb
yc
δyj
/ξj (y
) = δyj
/ξj (y
) + δyj
/ξj (y
). (4.5.4c)
ya ya yb

From the geometric standpoint illustrated in Fig. 4.5.1c, the transforma-

tion in which α
ranges from a to b carries the coordinates of the point
Py from y = ya to y = yb . This transformation may be expressed as
yb = T(∆ba )ya . The transformation in which α
ranges from b to c carries
y from yb to yc and may be expressed as yc = T(∆cb )yb . The transfor-
mation in which α
ranges from a to c carries y from ya to yc and may be
October 28, 2010 16:55 9in x 6in b943-ch04

118 Dynamical Symmetry

y y
0.5 0 0.5 0
−0.5 −0.5

0.5 0.5

0 z 0 z

−0.5 −0.5

0.5 0.5
0 0
−0.5 −0.5
x x

(a) (b)
y
0.5 0 −0.5

0.5

0 z

−0.5

0.5
0
−0.5 x

(c)

Fig. 4.5.1 Addition of the finite transformation of a one-parameter group.

expressed as yc = T(∆ca )ya . In terms of the notation of Sec. 4.1, we have,

from (4.5.4):
T(∆cb )T(∆ba )y = T(∆ca )y, ∆ca = ∆cb + ∆ba . (4.5.5)
From (4.5.4) that it also follows that
T(0) = I, (4.5.6)
the identity transformation. One easily also finds from (4.5.4) that
T(∆)T(−∆) = I, (4.5.7)
and that
T(∆3 )(T(∆2 )T(∆1 )) = (T(∆3 )T(∆2 ))T(∆1 ). (4.5.8)
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 119

Consequently, if the differential equations (4.5.1) that define a group of

infinitesimal transformations are, in principle, integrable, they also define
a group of finite transformations.
The transformations above may be put in standard form by replacing
the parameter changes ∆ by a parameter, signified by a lower case Greek
letter, that yields the identity transformation when it is set equal to zero.
The composition law (4.5.5) then becomes, e.g.,

T(α)T(β)y = T(γ)y, γ = β + α. (4.5.9)

An example of this replacement is provided by the transformation x∗ = ax.

Replacing a by exp(α), one obtains the transformation x∗ = exp(α)x, which
becomes the identity transformation when α approaches 0.
These results lead to the following observations:

1. The group property of the operation of integration imposes finite group

structure on sets of first-order ordinary differential equations if the equa-
tions possess solutions that may be expressed as functions of the “inde-
pendent variable,” which becomes the group parameter.

The solution curves of the set of differential equations (4.5.1) are path curves
of a transformation group T(α). The operations of the group, integrations,
evolve solutions from initial data. Acting at any point on a given solution
curve, they carry the point along the solution curve. They may be said to
convert the solution curve into itself. Consequently:

2. The differential equations define a group structure, and an intrinsic sym-

metry, which they themselves possess.

This intrinsic symmetry becomes a geometric symmetry when, as in the pre-

vious figures, the variables are given a geometric interpretation. The intrin-
sic symmetry, and the geometric symmetry, becomes a dynamical symmetry
when the variables represent dynamical variables.

4.6 The Operator of Finite Transformations

The function ew has the well known power series expansion

ew = Σ∞ n
0 w /n!. (4.6.1)

If one formally replaces w by αU in it one obtains

exp(αU) = Σ∞ n
0 (αU) /n!. (4.6.2)
October 28, 2010 16:55 9in x 6in b943-ch04

120 Dynamical Symmetry

It is also true that

exp(w) = lim (1 + w/n)n , (4.6.3a)
n→∞
and one expects that
exp(αU) = lim (1 + αU/n)n . (4.6.3b)
n→∞

If U = Σnj ξj (x
)∂/∂x
j , then when exp(αU) acts on functions of the x
j ,
(4.6.3b) represents its action as the limit of a continuous succession of
infinitesimal transformations each with operator (1 + δαU), in which δα =
α/n. With this observation as a motivation, let us explore the idea that the
finite transformation
x → x∗ = T(α)x, x = (x1 , . . . , xn ), (4.6.4)
defined by
xj → x∗j = Xj (x, α), j = 1, . . . , n,
can be expressed as
xj → x∗j = exp(αU)xj , (4.6.5a)
with
Xj (x, α) = exp(αU)xj . (4.6.5b)
Because the transformation depends continuously upon x and α, it is nec-
essary that Eq. (4.6.5) hold true for all x
ranging from x to x∗ . Because of
the differentiabilty requirement on (4.6.4), one must have
dx
j /dα = dXj (x
, α)/dα (4.6.6a)
so one must also have
dx
j /dα = d(exp(αU)x
j )/dα,
(4.6.6b)
= exp(αU)Ux
j .
The group property requires that these equations must hold in the limit
α → 0. Now
(dXj (x
, α)/dα)|α=0 = ξj (x
), (4.6.7a)
and
(exp(αU)Ux
j )|α=0 = Ux
j . (4.6.7b)
Thus if
U = Σn1 ξj (x
)∂/∂x
j , (4.6.7c)
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 121

then exp(αU)xj satisfies the differential equations satisfied by the Xj (x, α).
Since both exp(αU)xj and Xj (x, α) yield the identity transformation when
α → 0, it follows that, as conjectured,
Xj (x, α) = exp(αU)xj . (4.6.8)
Thus, a transformation with generator U can be expressed as
T(α)x = exp(αU)x. (4.6.9)
Putting these last two equations together, and recalling that each
Xj (x, α) is defined by integrating Eqs. (4.4.1), one concludes that in (4.6.9)
the operation of integration has been converted into one of exponentiation.
It is important to note, however, that for the series and product expan-
sions of exp(αU)x to be valid, the functions ξj (x
) in (4.6.7c) must be
infinitely differentiable, and the expansion must converge.
When the group parameter α is so related to physical time, t, that
dα = dt, the group becomes the evolution group of the dynamical system
governed by the differential equations that define the group. As an example
of this, consider Hamiltonian’s differential equations (3.3.3), viz.,
dqj /dt = ∂H(q, p)/∂pj , dpj /dt = −∂H(q, p)/∂qj . (4.6.10a)
They can be rewritten in the form
dt = dqj /(∂H(q, p)/∂pj ) = dpj /(−∂H(q, p)/∂qj), (4.6.10b)
and define the group generator
U = −(∂H/∂qi ∂/∂pi − ∂H/∂pi ∂/∂qi ). (4.6.11)
This generator is often written as {−H·}, where, for any function A(q, p),
the operator
{A·} = Σi (∂A/∂qi ∂/∂pi − ∂A/∂pi ∂/∂qi ), (4.6.12)
is a “Poisson Bracket waiting to happen”. The corresponding evolution
operator exp(t{−H·}) acts on q, p to give q∗ = Q(q, p, t), p∗ = P(q, p, t),
the values of the position and momentum vectors at t, given their initial
values q, p at t = 0. If F(q, p) is an analytic function of q, p, then its value
evolves to F(Q(q, p, t), P(q, p, t)) = exp(t{−H·})F(q, p).
As the motions of a system of particles express the action of a one-
parameter group upon the positions and momenta of the particles, they
express a dynamical symmetry of the system. In subsequent chapters, oper-
ators exp(α{A·}) that convert solutions of the equations of motion into
October 28, 2010 16:55 9in x 6in b943-ch04

122 Dynamical Symmetry

different solutions will be shown to express dynamical symmetries of the

system in addition to the symmetry expressed by exp(t{−H·}).
When the functions ξj (x
) are sufficiently differentiable, approximat-
ing exp(αU) by a finite product (1 + αU/n)n can provide a conve-
nient way of programming computers to approximately integrate sets
of autonomous first-order ordinary differential equations. The expansion
exp(αU) = Σ0 (αU)n /n!, which is also used to develop approximate series
expansions of the solutions of differential equations, is sometimes termed a
Lie series expansion. Consider for example the equations

dx
/dt
= y
, dy
/dt
= x
. (4.6.13a)

Here

U = y
∂/∂x
+ x
/∂y
. (4.6.13b)

One has

Ux
= y
, Uy
= x
; U2 x
= x
, U2 y
= y
, . . . . (4.6.14)

If x, y are the values of x
, y
at t
= 0, the Lie series expression for the final
value of x
at t
= t, is

x∗ = exp(t
U)x
= x
(1 + t
2 /2! + t
4 /4! + · · · ) + y
(t
+ t
3 /3! + · · · )

→ x cosh(t
) + y sinh(t
). (4.6.15a)

The corresponding Lie series expansion of exp(t
U)y
produces

y∗ = y cosh(t
) + x sinh(t
). (4.6.15b)

In most cases the series expansion one obtains is not that of easily rec-
ognized functions, and one simply retains a finite number of terms in the
expansion and so obtains approximate solutions of differential equations.

4.7 Changing Variables in Group Generators

It is often useful to change variables in the generator U. Let the diffeomor-
phism

x
j = fj (x), xj = fjin (x
), (4.7.1)

determine a change of coordinates x ↔ x
. Using the chain rule one sees

that U remains a first-order differential operator, and one may write

U = Σj ξj (x)∂/∂xj → Σ
k ξk
(x
)∂/∂x
k . (4.7.2)

October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 123

To determine the form U takes in the x
coordinate system one need only
determine the functions ξk
(x
). In the x
system one has
ξk
(x
) = Ux
k , (4.7.3)
with x
j = fj (x). Hence
Ux
k = U fk (x) = Σj ξj (x)∂fk (x)/∂xj . (4.7.4)
Denoting ∂fk (x)/∂xj by fk,j (x), one uses the inverse transformation
x = f inv (x
), to convert the right-hand side of (4.7.4) to a function of x
,
obtaining
ξk
(x
) = Ux
k = Σj ξj (f inv (x
))fk,j (f inv (x
)). (4.7.5)
As an example, let us express the generator of rotations in the z1 –z2 plane,
viz.,
U = z1 ∂/∂z2 − z2 ∂/∂z1 (4.7.6)
in the polar coordinates defined by
x1 = r = (z21 + z22 )1/2 , x2 = θ = arctan(z2 /z1 ). (4.7.7a)
The inverse transformation functions, f inv , are given by
z1 = r cos(θ), z2 = r sin(θ). (4.7.7b)
In the left-hand side of (4.7.2) one has ξ1 = −z2 , ξ2 = z1 , so the fundamental
Eqs. (4.4.1) take on the form
dz
j /dα
= ξ(z
)j , (4.7.8)
and require
dz
1 /dα
= −z
2 , dz
2 /dα
= z
1 . (4.7.9)
Applying U to r and θ, one finds
ξr
= Ur = 0, ξθ
= Uθ = 1. (4.7.10)
Consequently, in the polar coordinate system, U = ∂/∂θ, and Eqs. (4.7.9)
are replaced by
dθ/dα = 1, dr/dα = 0. (4.7.11)
In this example, because the ξ
turned out to be constants rather than
functions of the original variables, it is not necessary to use the inverse
transformation to express them as functions of the new variables. The exer-
cises contain examples that require the reader to use the inverse transfor-
mation to obtain the final results.
October 28, 2010 16:55 9in x 6in b943-ch04

124 Dynamical Symmetry

In the example just given, the change of variables converted the differ-
ential equations (4.7.9) into the much simpler form (4.7.11). This simplifi-
cation is a special case of a very general result to be described in the next
section. (In it we will cease using primes to distinguish running values of
variables.)

4.8 The Rectification Theorem

The following theorem has been termed by Arnold the fundamental theorem
of the theory of ordinary differential equations.4

Theorem 4.8.1. Let x represent a set of real variables. If the functions

ξj (x) in the equations
dxj /dα = ξj (x), j = 1, . . . , n < ∞, (4.8.1)
are differentiable functions of class Ck , k > 0, then in the neighborhood of
every point Px that is not a critical point, there exist open regions in which
local diffeomorphisms x → y = φ(x1 , x2 , . . .) convert these equations to
dy1 /dα = 1, dyj /dα = 0, j = 2, . . . , n. (4.8.2)

The adjective ‘local’ implies that different diffeomorphisms may be required

in different regions. The adjective ‘open’ means that each xj and each yj lies
in an open region between two real numbers aj and bj , e.g., aj < xj < bj .
From the theorem it follows that the generator U = Σn1 ξj (x)∂/∂xj of the
group defined by (4.8.1) can be converted by one or more diffeomorphisms to
the form U = ∂/∂y1 , except in the region of a critical point, where U = 0.
Canonical coordinates are coordinates y = (y1 , y2 , . . .) in which U takes
on the form ∂/∂y1 .
Figures 4.8.1 illustrate the result of converting Eqs. (4.7.9) to the cor-
responding equations in canonical coordinates, (4.7.11). They are designed
to emphasize that canonical coordinates are not necessarily Cartesian coor-
dinates in a Euclidean space, but can, for example, be locally orthogonal
coordinates of the sort discussed in Sec. 2.7.
The most direct general method for converting a group generator U =
Σn1 ξj (x)∂/∂xj to the rectified form U = ∂/∂y1 is to integrate Eqs. (4.8.1)
in the form
dx1 /ξ1 (x) = dx2 /ξ2 (x) . . . = dxN /ξN (x) = dα. (4.8.3)
This method was illustrated in the previous section. It often requires mak-
ing a number of substitutions and finding a number of integrating factors
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 125

1.0

0.5

x
− 1.0 − 0.5 0.5 1.0

− 0.5

− 1.0

(a)

r = y2
1.0
1.0

0.5
0.5

θ = y1 − 1.0 − 0.5 0.5 1.0

− 1.0 − 0.5 0.5 1.0

− 0.5
− 0.5

− 1.0
− 1.0

(b) (c)

Fig. 4.8.1 Local effect of rectification of rotations.

before one arrives at the final Eqs. (4.8.2). However, some equations cannot
be integrated analytically. This is the case for Eqs. (4.2.10), which allow no
integrating factor. Nevertheless, the theorem guarantees that if the func-
tions are at least C1 , there exist diffeomorphisms which convert the x’s to
local or global canonical coordinates y.
In physical applications, the most common functions that are not at
least C1 are step functions, which are discontinuous and have discontinu-
ous first derivatives, and sawtooth functions, which have discontinuous first
derivatives. However in physically relevant equations, such as those of elec-
tric circuit theory, these functions approximate to smoother functions. They
October 28, 2010 16:55 9in x 6in b943-ch04

126 Dynamical Symmetry

are themselves commonly approximated by truncated Fourier expansions,

i.e. by analytic functions.
On integrating Eqs. (4.8.2) from α
= 0 to α, one obtains the finite
transformation
y1 → y1 + α, yj → yj , j = 2, . . . , N, (4.8.4)
valid within each range of x = (x1 , . . . , xN ), for which x and y = (yj , . . . , yN )
are related by a particular diffeomorphism. The most common natural
effects limiting the range of a rectifying diffeomorphism are those that pro-
duce closed solution curves and open solution curves in different regions,
and those that produce a chaotic or turbulent mixture of closed and open
curves in some regions. No diffeomorphism can transform closed curves into
open ones, or vice versa (see the appendix of this chapter).
An example of a set of ODEs that requires two different diffeomorphisms
is provided by the set of equations
dx/dt = x(1 − x2 − y2 ) − y, dy/dt = y(1 − x2 − y2 ) + x. (4.8.5)
This pair of equations occurs in the theory of predator–prey relations. For
example, x(t), may represent a deviation from the mean concentration of
arctic foxes, and y(t) may represent a corresponding deviation of the con-
centration of arctic hares from their mean value. In the abstract polar
coordinates defined by x = r cos(θ), y = r sin(θ), Eqs. (4.8.5) become
dθ/dt = 1, dr/dt = r − r3 . (4.8.6)
Interpreting t as a group parameter, the corresponding Lie generator is
U = ∂/∂θ + (r − r3 )∂/∂r, (4.8.7)
and exp(tU) acts as an evolution operator. It evolves r and θ, and hence
the species concentrations x, y, from their initial values at t0 = 0 to their
values at time t. Inspecting these equations one observes that dr vanishes
when r = 0, 1. In the latter case, as θ varies from −π to + π, r describes a
circle of unit radius. In both cases dr/dt = 0, dθ/dt = 1, and the equation
is rectified. On the other hand, if initially r lies between 0 and 1, then in
(4.8.6) dr/dt is positive, and as t increases, r increases — approaching, but
never reaching, 1. Conversely, if r is initially greater than 1, then r decreases
as t increases, and r approaches the unit circle. If r initially = 1, it remains
so, and r rotates with constant velocity in the space of arctic foxes and
rabbits. Solution curves of each type are illustrated in Figs. 4.8.2 in which
x and y are treated as Cartesian variables in a Euclidean plane. Note that
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 127

hares hares

1.0 1.0

0.5 0.5

foxes foxes
− 1.0 − 0.5 0.5 1.0 − 1.0 − 0.5 0.5 1.0

− 0.5 − 0.5

− 1.0 − 1.0

(a) (b)

hares

1.0

0.5

foxes
− 1.0 − 0.5 0.5 1.0

− 0.5

− 1.0

(c)

Fig. 4.8.2 Local transformation and limit cycle behavior determined by

Eqs. (4.8.5).

for the circular solution, θ and θ + 2π are equivalent, while on the spiral
solutions θ must range through all the reals to describe the system.
Let us determine local canonical coordinates of the system when r = 1.
Different transformations are required when r < 1 and when r > 1. Equa-
tions (4.8.6) imply that

dr/dθ = r − r3 . (4.8.8a)
October 28, 2010 16:55 9in x 6in b943-ch04

128 Dynamical Symmetry

Hence

3 dln(r/(r2 − 1)1/2 ), r > 1,
dθ = dr/(r − r ) = (4.8.8b)
dln(r/(1 − r2 )1/2 ), 0 < r < 1.
Multiplying these equations through by two, one can replace them by
d(2θ) = dln(r2 /(r2 − 1)), r > 1, (4.8.8c)
2 2
d(2θ) = dln(r /(1 − r )), 0 < r < 1. (4.8.8d)
Integrating these, one obtains the solutions in a form valid over both ranges:
exp(−2θ)(r2 − 1)/r2 = k, (4.8.8e)
where
k = exp(−2θ0 )(r20 − 1)/r20 . (4.8.8f)
If the initial value of r is < 1, k is negative and r must remain less than 1.
If the initial value of r is > 1, k is positive and r must remain greater
than 1. If we define s = exp(−2θ)(r2 − 1)/r2 , then in each case s is constant
and ds = 0. If r is initially 1, k = 0, and s = 0, as is ds. In this case, in
(4.8.8g), exp(−2θ) is indeterminate.
Returning to (4.8.8a–d) we see that transforming to θ and s variables
converts these equations to the rectified form
dθ/dt = 1, ds/dt = 0. (4.8.9)
However we must no longer suppose that θ and θ + 2π are necessarily
equivalent, because, even though s remains constant, when s is nonzero, r
varies with s in such a manner that r(θ) and r(θ + 2π) do not have the same
value. Thus when r0 is neither one nor zero, θ must be considered to be a
noncyclic coordinate.
Equations (4.8.5) have limit cycle solutions that, in general, approach
but do not reach the cyclic solution which acts as a stable attractor. The
practical utilization of stable attractors has a long history. Since the days of
De Forest, electrical engineers have utilized negative feedback to stabilize
electrical oscillations. And, much before that, Watt designed a governor
which utilized centrifugal force to ensure that his steam engines would settle
down to a constant running speed.
Long before the arrival of Europeans, native American hunters were
aware that populations of predator and prey species oscillate in near syn-
chrony, the peaks of each population being separated from the other by
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 129

a few years. Eighteenth and nineteenth century fur trade records of the
Hudson’s Bay Company appear to have led to the realization that these
populations obey rate equations with limit cycle solutions.5
A variety of chemical kinetic equations that produce concentration oscil-
lations can have the same property — a property which can also be respon-
sible for stable oscillatory behavior in biological systems.
If limit cycle oscillations become very small, the attractor approaches a
point, and the corresponding biological system exhibits homeostasis. Oscil-
latory chemical concentrations can develop when an autocatalytic reac-
tion and an autodepressive one are chemically coupled. They are of central
importance in Turing’s reaction–diffusion theory in which chemical concen-
tration waves produce cell differentiation and biological morphogenesis.6
Equations (4.8.5) were chosen to provide an instructive example of the
workings of the rectification theorem when the required change of variables
is different for different ranges of the variables x, y, or r, θ.
Differential equations with solution curves of differing topologies are of
common occurrence. For example, objects moving under gravitational or
electrical attractions typically have both closed orbits and open orbits. In
latter chapters it will be shown that this presence of both open and closed
orbits in classical Kepler systems affects the dynamical groups of quantum
mechanical Kepler systems as well as their classical counterparts.

4.9 Conversion of Non-autonomous ODEs

to Autonomous ODEs
The set of non-autonomous equations
dxj /dt = fj (x, t), j = 1, . . . , m, (4.9.1)
can always be converted to a set of autonomous equations with the same
solutions. To accomplish this, one supplements the equations with the fur-
ther equation dt/dτ = 1, and then replaces them with the autononomous
equations
dt/dτ = 1, dxj /dτ = fj (x, t), j = 1, . . . , m. (4.9.2)
When using this autonomous extension of the non-autonomous Eqs. (4.9.1),
one needs to keep in mind that though dt = dτ , the variables t and τ may
not only differ by a constant; one may also be confined within a closed
interval, while the other is not.
October 28, 2010 16:55 9in x 6in b943-ch04

130 Dynamical Symmetry

Equations (4.9.2), being autonomous, define a group generator

V = ∂/∂t + Σ fj (x, t)∂/∂xj . (4.9.3)
Exp(τ V) acts on initial values of x and t to yield solution curves of
Eqs. (4.9.1), defined by the m+1 equations
t → t
= t + τ, xj → x
j = Xj (x, t; β). (4.9.4)
If one considers dt and dτ to represent physical time intervals, it does
not follow that t or τ represent physical time; in physics, as currently under-
stood, it is necessary that the variable representing time be noncyclic. If t
happens to be a cyclic variable, one may choose τ to be noncyclic. The phys-
ical system obeying Eqs. (4.9.2) is then a dynamical system, and its symme-
tries may be said to be dynamical symmetries. The solutions of (4.9.2) then
define an evolution curve in space of variables τ, t, x that includes time. V
is the generator of evolution, and exp(τ V) is the evolution operator.
Equations (4.9.1) may be expressed in canonical coordinates as
dy1 /dt
= 1, dyj /dt
= 0, j = 2, . . . , m. (4.9.5)


The variables y, t are related to the variables x, t by diffeomorphisms of
the form
t → t
= τ (t, x), xj → yj = ψj (x, t);



(4.9.6)
t → t = τinv (t , y), yj → xj = ψjinv (y, t
).
Supplementing Eqs. (4.9.5) with the equation dt
/dβ = 1, they may be
rewritten as
dt
/dβ = 1, dy1 /dβ = 1, dyj /dβ = 0, j = 2, . . . , m. (4.9.7)



The evolution generator is then V = ∂/∂y1 + ∂/∂t and the group parame-
ter is β. The solutions of the equations are defined by t
− β = a, y1 − β = b,
yj = cj , j = 2, . . . , m. One may use the first equation to eliminate β from
the second, obtaining y1 − t
= c1 , and let the solutions involving y be
denoted by the n equations φj (y, t
) = cj , j = 1, . . . , n. The discussions of
this section lead to the following result:
The solutions of the equations
dxj /dt = fj (x, t), j = 1, . . . , m, (4.9.8)
1
with fj that are C functions may be put in the form
ϕj (x, t) = cj , j = 1, . . . , m. (4.9.9)
This statement follows from the form of the solutions to (4.9.5), which is
obtained in the previous paragraph, and the diffeomorphisms (4.9.6); these
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 131

together yield
ϕj (x, t) = φj (ψ(x, t), τ (t, x)). (4.9.10)

4.10 N-th Order ODEs as Sets of First-Order ODEs

A second-order ODE such as
d2 y1 /dx2 = f(y1 , dy1 /dx, x), (4.10.1)
may always be expressed as the pair of first-order ODEs. One simply lets
dy1 /dx = y2 , dy2 /dx = f(y1 , y2 , x). (4.10.2a,b)
In a similar manner, any m-th order ODE which can be put in the form
dm y1 /dxm = f(y1 , . . . , ym ; x), (4.10.3)
can, on writing dyr /dx = yr,x , be expressed as a set of first-order ODEs:
y1,x = y2 , y2,x = y3 , ym−1,x = ym , ym,x = f(y, x). (4.10.4)
This is a special case of the set of first-order ODEs
dyj /dx − fj (y, x) = 0, j = 1, . . . , m. (4.10.5)

4.11 Conclusion
Any finite set of ordinary differential equations defined by sufficiently
smooth functions may be converted to a set of first-order autonomous equa-
tions
δyj
/δα
= ξj (y
), j = 1, . . . , n, (4.11.1a)
and expressed as
δy1
/ξ1 (y
) = δy2
/ξ2 (y
) = · · · = δyn
/ξn (y
) = δα
. (4.11.1b)
The rectification theorem guarantees that if the ξj (y
) are at least C1 func-
tions, then these equations at the very least define local one-parameter
groups acting within open intervals of the y’s, and the equations may define
a global transformation group. The generator of a group defined by (4.11.1)
is U = Σ ξj (y)∂/∂yj .
When Eqs. (4.11.1) are integrated over finite intervals, as the inte-
gration proceeds, the variables yj
change from their initial values yj =
Yj (y, 0) to their final values yj∗ = Yj (y, γ). In each successive inter-
val, ∆α, ∆β, . . . , the integration process defines a finite transformation
October 28, 2010 16:55 9in x 6in b943-ch04

132 Dynamical Symmetry

∆y = T(∆α)y, ∆y = T(∆β)y, . . . , of the variables yj . The group com-

position law, T(∆β)T(∆α)y = T(∆β + ∆α)y, follows from the fact that
integration over two successive intervals gives the same result as integration
over the total interval.
The finite transformations of the group, T(α)yj , may be expressed as
exp(αU)yj , an expression which formally satisfies the differential equations
(4.11.1). The expression can provide a convenient means to evaluate, or
approximate, the effect of the transformation.
The operations of the group defined by (4.11.1) develop a solution of the
differential equations through any given initial point; they carry each solu-
tion of the equation into itself. The equations are thus invariant under the
operations of the transformation of the group they define; they possess the
intrinsic symmetry so defined. It will be shown in the next chapter that such
systems of ordinary differential equations possess further intrinsic symme-
tries.

Appendix A. Homeomorphisms, Diffeomorphisms,

and Topology
As previously noted, a homeomorphism is a one-to-one invertible transfor-
mation
xj → x
j = fj (x), x
j → xj = fjin (x)
with transformation functions f and f in that depend continuously upon
all their variables in the range allowed for these variables. The equation
x → y = x2 has two roots x = ±y, so the inverse transformation y → x is
not one-to-one. In this case, the transformation is not a homeomorphism
if y is allowed to have both positive and negative values. Any polynomial
equation y = P(x) that defines a transformation over a range of y in which
the polynomial has roots with different values, will fail to define a homeo-
morphism.
Considered in an active sense, because it is defined by continuous func-
tions, a homeomorphism carries contiguous points Pa and Pb into contigu-
ous points P
a and P
b . Pairs of geometric objects such as lines, surfaces, and
hypersurfaces, that are transformed into each other by a homeomorphism
are termed homeomorphic. By definition, homeomorphic geometric objects
have the same topology. A square and a circle are homeomorphic, but a
circle and a straight line are not.
Diffeomorphisms map noncompact regions into noncompact regions. If
a diffeomorphism acts within a compact subregion of the open region over
October 28, 2010 16:55 9in x 6in b943-ch04

One-Parameter Transformation Groups 133

which it is defined, the diffeomorphism will map the compact region into
a compact region. A square and a circle in the Euclidean plane are not
diffeomorphic because components of the vector r = (z1 , z2 ) from the origin
to points on the boundary of the square do not change smoothly at the
corners of the square.
The following equations define the C∞ diffeomorphisms indicated:
y = exp(x), x = ln(y), −∞ < x < ∞ → 0 < y < ∞;
y = x/(1 − x), x = y/(1 + y), −∞ < x < 1 → −1 < y < ∞;
(x, y) → (x
, y
) = (xea , yeb ), with a, b real.
The following do not define diffeomorphisms for the reasons indicated:
i. y = exp(|x|), |x| = ln(y), −∞ < x < ∞ → 0 < y < ∞;
dy/dx is discontinuous at x = 0.
ii. Any transformation from angular coordinates of points on a sphere
to angular coordinates on a torus; the topologies of the surfaces are
different.
A C∞ diffeomorphism becomes an analytic transformation iff the func-
tions fj (x), fj
(x) can be expressed by convergent power series expansions
over the range allowed to x. The expansions need not be about a single
point, but may be about a set of points. (If the power series expansion of a
function is convergent about all points, the function is said to be an entire
function.)

Exercises
1. Determine the infinitesimal transformations and generators of the fol-
lowing finite transformations:
a. x
= (x2 + 2α)1/2 , y
= (y2 − α)1/2 ;
b. x
= x/(1 − αx), y
= y/(1 − αx);
c. x
= eα (x cos(α) − y sin(α)), y
= eα (x sin(α) + y cos(α)).
2. Let a be a continuously variable parameter. Can the transformation
x
= 1/(x + a), be that of a Lie group with group parameter a, or
function of a?
3. Determine the finite transformations produced by the groups with gen-
erators
a. x(x∂/∂y − y∂/∂x);
b. x∂/∂y − y∂/∂x + k∂/∂z, where k is a constant.
4. Does the transformation defined by Eq. (4.2.10) have any critical points?
October 28, 2010 16:55 9in x 6in b943-ch04

134 Dynamical Symmetry

5. What are the ordinary differential equations that define each of the fol-
lowing group generators?
U1 = x∂/∂y − y∂/∂x, U2 = x∂/∂(ct) + (ct)∂/∂x, U3 = xU1 ,
U4 = xU2 , U5 = ctU2 , U6 = x∂/∂x + y∂/∂y + t∂/∂t,
U7 = (x2 + y2 − c2 t2 )∂/∂x − 2xU6 ,
U8 = (x2 + y2 − c2 t2 )∂/∂(ct) + 2ctU6 .
6. Plot the two-dimensional vector fields associated with the Lie generators
U1 through U5 in problem (5).
7. Plot the three-dimensional vector fields associated with the generators
U6 , U7 , U8 , and plot the corresponding fields in the three two-planes in
which x, or y, or ct, is zero.
8. Consider the transformation defined by:
s1 = kx/(x2 +y2 −c2 t2 ), s2 = ky/(x2 +y2 −c2 t2 ), s3 = kct/(x2 +y2 −c2 t2 ),
where k is a positive constant and c is the velocity of light. Calculate
s21 +s22 −s23 , and determine the corresponding inverse transformation from
s = (s1 , s2 , s3 ) to (x, y, ct).
9. Transform the U’s of problem 5 to Lie generators of the form W =
Σξj (s)∂/∂sj .
Note: Problems 5 through 9 all relate to material that will be discussed in
subsequent chapters.

References
[1] H. Jerome Keisler, Foundations of Infinitesimal Calculus (Prindle, Weber &
Schmidt, Boston, 1976).
[2] J. C. Maxwell, Electricity and Magnetism (Cambridge, 1873).
[3] H. Whitney, Collected Papers, eds. J. Eels, D. Toledo, Vol. II (Birkhauser,
Boston, 1992).
[4] V. I. Arnold, Ordinary Differential Equations, translated by R. A. Silverman,
(MIT Press, Cambridge, 1973).
[5] A. J. Lotka, Elements of Physical Biology (Williams and Wilkins, 1925).
[6] A. M. Turing, Phil. Trans. Roy. Soc. (London) 237 (1952) 37.
October 28, 2010 16:55 9in x 6in b943-ch05

CHAPTER 5

Everywhere-Local Invariance

5.1 Invariance under the Action of One-Parameter Lie

Transformation Groups
In this chapter we characterize the invariance of functions, equations, met-
rics, and ordinary differential equations under the action of Lie transforma-
tion groups. All the discussions will deal with invariance that is independent
of the value taken on by a group parameter. We will not consider discrete
invariance transformations, or invariance that holds only for discrete values
of a group parameter, e.g. transformations such as

exp(2π∂/∂x) sin(x) = sin(x + 2π) = sin(x). (5.1.1)

A function f(x) is invariant under the actions of a Lie transformation

group with operator exp(αU), U = Σ ξj (x) ∂/∂xj , iff for all allowed values
of x and α,

f(x) = f(x
), x
= exp(αU)x. (5.1.2)

If f(x) is differentiable, then invariance requires that ∂f(x
)/∂α = 0, along

each path curve x
= exp(αU)x, regardless of its initial point Px . Now

∂f(x
)/∂α = ∂f(exp(αU)x)/∂α = f(U(exp(αU)x)) = Uf(x
). (5.1.3)

Because of the group property, every point on each path curve may be
considered an initial point of the group action as well as a point obtained
by an infinitesimal or finite transformation of the group, so one may let
α → 0 in (5.1.3) after taking the derivative. Then x
→ x, and Ux
→ Ux.
The right-hand side of (5.1.3) then becomes Uf(x), and

∂f(x
)/∂α|α=0 = Uf(x). (5.1.4)

135
October 28, 2010 16:55 9in x 6in b943-ch05

136 Dynamical Symmetry

Thus one obtains:

Theorem 5.1.1. A differentiable function f(x) is invariant under the

action of a one-parameter Lie group with generator U iff

Uf(x) = 0 (5.1.5)

for all allowed values of x.

This criterion for invariance is purely local. It is equivalent to requiring that

for arbitrary δα,

f(x
) = (1 + δα U)f(x
) (5.1.6)

for all allowed values of x
. The group property of the transformation,

together with the assumed differentiabilty of f(x) for all values of x, assures
that the local invariance holds around each successive point reached by the
action of exp(αU) on x. Thus the global invariance of a function under the
action of a continuous group is a consequence of local invariance everywhere.
We next characterize the invariance of equations under a one-parameter
group of transformations. Given an equation of the form g(x) = c, where
c is a constant, and g is everywhere differentiable, let f(x) = g(x) − c.
The resulting equation f(x) = 0, will be invariant under the transformation
x → x
= exp(αU)x iff

f(x
) = 0 (5.1.7a)

when

f(x) = 0, (5.1.7b)

for all values of α, and for all values of x such that f(x) = 0. Let it be sup-
posed that f satisfies no other independently valid equations. In particular,
suppose that f(x) is not of form g(x)2 h(x), that contains repeated factors
g(x) which themselves vanish. With these restrictions in force, there are
only two ways in which exp(αU) can leave the equation invariant. Either:

i) the transformation leaves the function f(x) invariant for all values of x,
or,
ii) it leaves the equation f(x) = 0 invariant for values of x satisfying
f(x) = 0. (This can only happen if Uf contains f(x) as a factor.)
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 137

Both possibilities are subsumed in:

Theorem 5.1.2. If f(x) is a differentiable function containing no repeated

factors, satisfying an equation f(x) = 0 and no other independent equation,
then the equation f(x) = 0 is invariant under the action of a one-parameter
transformation group with generator U iff

0 = Uf(x)|f(x)=0 . (5.1.8)

That is to say Uf(x) = 0 for all values of x such that f(x) = 0.

Just as in the previous theorem, global invariance under the action of

a continuous group is determined by a purely local invariance that is valid
for all allowed values of the coordinates x. The group property of the trans-
formation allows x to be considered x
, and vice versa.
There is a hidden aspect to the argument just given: the operator
(1 + δαU) varies all the variables x, which are thus initially considered
to vary independently. The functional relation among the variables x, and
the functional relation among the running variables x
is thereafter estab-
lished by a restriction on the form of U. This is imposed by the condition
that Uf(x) = 0 for all values of x that satisfy f(x) = 0.
In using Theorem 5.1.2, the reader is reminded to keep in mind the
trivial example that develops if U itself contains f as a factor, for then, as
mentioned above, Uf|f=0 will necessarily vanish. Such U’s are generally of
little interest.
The following application of Theorem 5.1.2 demonstrates that a unit
circle is left invariant by a transformation that is not generally a rotation.
Let

f(x) = x21 + x22 − 1 = 0, (5.1.9a)

and let

T(α) = exp(αU), U = x1 x2 ∂/∂x1 + (x22 − 1)∂/∂x2 . (5.1.9b,c)

Then, though U itself does not contain x21 + x22 − 1 as a factor,

Uf(x) = 2x1 x2 x1 + 2(x22 − 1)x2 = 2x2 (x21 + x22 − 1). (5.1.9d)

This expression vanishes on the unit circle defined by x21 + x22 = 1. Conse-
quently

0 = Uf(x)|f(x)=0 . (5.1.9e)
October 28, 2010 16:55 9in x 6in b943-ch05

138 Dynamical Symmetry

x2
1.5

1.0

0.5

x1
− 1.5 − 1.0 − 0.5 0.5 1.0 1.5

− 0.5

− 1.0

− 1.5

Fig. 5.1.1 Transformation (5.1.9) defines displacement vectors tangent to a unit

circle.

Thus, though the transformation (x1 , x2 ) → (x
1 , x
2 ) = T(α)(x1 , x2 ) does

not leave invariant x21 + x22 , it carries the unit circle into itself for all values
of α. Figure 5.1.1 illustrates the action of the transformation on points
on the unit circle and points off it.
We next turn to a discussion of the invariance of families of curves. The
equation

f(x, y) = c (5.1.10a)

with c a parameter rather than a constant, defines a one-parameter family of

curves. These families might represent generalized solutions of a differential
equation

M(x, y)dx + N(x, y)dy = 0, (5.1.10b)

i.e. of

dy/dx = −M(x, y)/N(x, y) (5.1.10c)

or

dx/dy = −N(x, y)/M(x, y). (5.1.10d)

October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 139

The same family of curves defined by (5.1.10a) may also be defined by

F(f(x, y)) = C, C = F(c), (5.1.10e)

where F is an arbitrary differentiable function.

Equations (5.1.10a) and (5.1.10e) represent the solutions of the differ-
ential Eqs. (5.1.10b–d) in a way which, like Eq. (5.1.10b), does not distin-
guish independent and dependent variables. (Planetary motion provides an
example where this representation is important: the radial position, r, of
a planet at time t is not a simple function of t, but t is a simple function
of r). Figure 5.1.2 illustrates the action of a group that interconverts the
members of a family of curves that comprise a flow, as defined in Chapter 3.
In Fig. 5.1.2 the transformations of the group are active ones that convert
curves of the flow into other curves of the flow. The figure calls attention
to the group defining a second flow that intersects the path curves of the
original differential equation.
Now let us consider the requirements the generator of an invariance
group of Eqs. (5.1.10) must satisfy. A Lie group with generator U =
ξ(x, y)∂/∂x + η(x, y)∂/∂y will leave Eqs. (5.1.10) invariant if 0 = Uf|f=c .

1.0

0.5

0.2 0.4 0.6 0.8 1.0

− 0.5

− 1.0

dx
Fig. 5.1.2 Transformation that interconvert curves of a flow defined by da = x.
October 28, 2010 16:55 9in x 6in b943-ch05

140 Dynamical Symmetry

Now if f is analytic, and if

Uf = 0 (5.1.11a)
identically, then f(exp(αU)x, exp(αU)y) = exp(αU)f = f; so if f(x, y) = c,
and
f(exp(αU)x, exp(αU)y) = c. (5.1.11b)
Each curve of the family is thus carried into itself. For a transformation to
carry solutions of (5.1.10) into different solutions it is necessary that
f(exp(αU)x, exp(αU)y) = f(x, y) − c(α), (5.1.11c)
with c(α) = c
, c
= c. The infinitesimal transformation of f must then be
such that (1 + δαU)f(x, y) can be expressed as f(x, y) − c(δα). This will be
true, for example, if
f(x, y) + δαUf = f(x, y) + δα const., (5.1.12a)
i.e. if
Uf = const. (5.1.12b)
If U satisfies (5.1.12b), then so does U/const. One thus obtains the simple
and useful result, that if f is differentiable, and
Uf(x, y) = 1, (5.1.13)
then exp(α U) converts f to exp(α)f, and converts a curve defined by
f(x, y) = c, (5.1.14b)
into the family of curves defined by
f(x, y) = c exp(−α). (5.1.14c)
If one knows a U such that Uf = G(f) rather than a constant, then one
may use the freedom indicated

in (5.1.10e) to redefine f. The change from
f to F(f), defined by F(f) = df/G(f), will then yield a F(f) such that
U F(f) = 1. (5.1.15)
The two families of solutions of the differential Eqs. (5.1.10b–d) defined
by f(x, y) = c, c any real number, and f(x, y) = exp(−α)c, coincide if the
parameter c is a parameter that may take on any real value. When this is
possible, exp(αU) is the operator of a global Lie invariance group of the
family of solutions of the differential equations. If the range of c or α is
restricted, then exp(αU) is the operator of a local Lie invariance group of
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 141

the family. This must be the case when the solution curves of (5.1.10b–d)
do not all have the same topology, for the diffeomorphism carried out by
exp(αU) cannot interconvert curves of different topology. An example of
this latter case is provided by the equations of motion of Kepler systems,
for they allow both open, hyperbolic and parabolic trajectories, and closed,
elliptical and circular trajectories.

5.2 Transformation of Infinitesimal Displacements

The previous paragraph investigated the effect of operators exp(αU) on
the family of solutions of a differential equation. It did not investigate the
effect of the exp(αU) on the differential equation itself. To further char-
acterize Lie transformation groups that act on differential equations and
their solutions it is necessary to investigate the effect of the transformation
groups on infinitesimal displacements and derivatives. The key to an under-
standing of transformations of derivatives, metrics, and other expressions
involving infinitesimal changes, is an understanding of transformations of
infinitesimal displacements.
We begin by considering finite transformations of infinitesimal dis-
placements carried out by one-parameter Lie groups. We will then deter-
mine the corresponding infinitesimal transformations of these infinitesimal
displacements.
Let x = (x1 , x2 , . . . , xn ) and x
= (x
1 , x
2 , . . . , x
n ) be coordinates of two
points in EN , and let x
− x become arbitrarily small, that is for all j, let
x
j − xj → dxj . We wish to obtain an expression for exp(αU)(x
− x) in the
limit x
− x → dx. One has
T(α)(x
j − xj ) = Xj (x
, α) − Xj (x, α). (5.2.1a)


As x → x + dx, this becomes T(α)(xj + dx) − T(α)xj , and one has
Xj (x + dx, α) − Xj (x, α) = Σk (∂Xj (x, α)/∂xk ) dxk . (5.2.1b)
Thus for a fixed value of α the finite transformation converts
dx = (dx1 , dx2 , . . . , dxn )
to
dx∗ = d(T(α)x) = (dx∗1 , dx∗2 , . . . , dx∗n ) (5.2.2)
with
dx∗j = d(T(α)xj ) = Σk (∂Xj (x, α)/∂xk )dxk .
October 28, 2010 16:55 9in x 6in b943-ch05

142 Dynamical Symmetry

δ(dx)
1.0

0.8 ∆r'

0.6

∆r

0.4

0.2
dx

0.0
0.0 0.2 0.4 0.6 0.8 1.0

(a) (b)

Fig. 5.2.1 (a) Finite transformation of a small displacement. (b) Infinitesimal

transformation of an infinitesimal displacement.

This action of a finite transformation on a displacement vector, dx =

(dx1 , dx2 ) is illustrated in Fig. 5.2.1a.
When the group parameter α becomes arbitrarily small, T(α) → T(δα),
and
d(T(α)xj ) → d(T(δα)xj ) = d(T(0)xj + δαUxj ). (5.2.3a)
On letting U = Σj ξj (x)∂/∂xj , and noting that T(0)xj is just xj , Eq. (5.2.3a)
becomes
dxj + δα d(ξj (x)) = dxj + δα Σk (∂ξj (x)/∂xk ) dxk . (5.2.3b)
Consequently, under the infinitesimal transformation with generator U, the
dx∗j of (5.2.2) are given by
dx∗j = dxj + δαΣk (∂ξj (x)/∂xk ) dxk . (5.2.4)
Let δ(dxj ) ≡ dx∗j −dxj
be the infinitesimal change in dxj due to the infinites-
imal transformation. On then obtains the key result:
An infinitesimal transformation with operator exp(δα U) and generator
U = Σj ξj (x)∂/∂xj (5.2.5a)
alters the infinitesimal displacements dxj by the infinitesimal amounts
δ(dxj ) = δα Σk (∂ ξj (x)/∂xk )dxk . (5.2.5b)
Figure 5.2.1b suggests the manner in which a dx = (dx1 , dx2 ), is
converted to δ(dx) = (δ(dx1 ), δ(dx2 )) by an infinitesimal transformation.
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 143

In this case the group generator is U = x1 (−x2 ∂/∂x1 + x1 ∂/∂x2), so ξ1 =

−x1 x2 and ξ2 = x21 , whence δ(dx) = δα( − (x1 dx2 + x2 dx1 ), 2x1 dx1 ).

5.3 Transformations and Invariance of Work, Pfaffians,

and Metrics
A Pfaffian form is an expression such as

δv − Σi yi δxi , i = 1, . . . , n, (5.3.1a)

or

Σi yi δxi , i = 1, . . . , n, (5.3.1b)

that is linear in all its variables δv, δxi , yi . The variables are a priori inde-
pendent. A Pfaffian equation is an equation in which a Pfaffian form is set
equal to zero. An example of a Pfaffian equation is provided by the expres-
sion for the element of work, δw, done by a force F acting on a particle
that undergoes a displacement δr, viz.,

δw = F · δr. (5.3.2)

When n=1 in (5.3.1a), the Pfaffian equation may be written dv−ydx = 0

or dv/dx = y, and dv/dx may be interpreted as the derivative of v with
respect to x. When n is greater than 1 the situation is not so simple. As is
familiar from discussions of work and potential energy, it need not follow
from Eq. (5.3.1a) that the yi are partial derivatives ∂v/∂xi . However, if for
all i, j = 1, . . . , n, one has

∂yi /∂xj = ∂yj /∂xi , (5.3.3)

then, and only then, each yi is the partial derivative ∂v/∂xi of a differen-
tiable function v(x), and

dv = Σi (∂v/∂xi )dxi = d(v(x)). (5.3.4)

In this case dv is an exact differential and the Pfaffian equation is integrable

as it stands.
For a Lie group with generator U to act on a Pfaffian form, it must be
considered to act on the variables y, x and v, all initially considered to be
independent. Let T(α) = exp(αU) with

U = Σnl ξj (x, y, v)∂/∂xj + Σnl πj (x, y, v)∂/∂yj + ζ(x, y, v)∂/∂v. (5.3.5)

October 28, 2010 16:55 9in x 6in b943-ch05

144 Dynamical Symmetry

For clarity of notation we will, in the next few lines, use the symbol δ

to signify variations carried out by the infinitesimal transformation with
generator U. The infinitesimal transformation of the yi is then

δ
(yi ) = δ
α πi (x, y, v). (5.3.6)

The infinitesimal transformation of the δxi and δv, in (5.3.1a) is given by

(5.2.5) in a form that does not assume integrability of the infinitesimal
displacements, i.e. by

δ
(δxj ) = δα Σk (∂ξj (x)/∂xk ) δxk . (5.3.7)

Along with the xi , the v and the yi are variables whose values may change
infinitesimally. Consequently

δ
(δxi ) = δ
α{Σk (∂ξi (x, y, v)/∂xk )δxk + Σk (∂ξi (x, y, v)/∂yk )δyk
+ (∂ξi (x, y, v)/∂v)δv}, (5.3.8)
δ (δv) = δ
α{(∂ζ(x, y, v)/∂v)δv + Σk (∂ζ(x, y, v)/∂xk )δxk

+ Σk (∂ζ(x, y, v)/∂yk )δyk }.

As an example of the foregoing results, consider the element of work dv,

defined by

dv − (y1 dx1 + y2 dx2 ) = 0, (5.3.9a)

where the yi are understood to be arbitrary functions of the x
s. As a test

case, suppose that the generator of the transformation T(α) acting on the
variables x,y,v is

U = x1 ∂/∂x1 + x2 ∂/∂x2 + y1 ∂/∂y1 + y2 ∂/∂y2 + v ∂/∂v. (5.3.9b)

The reader is invited to use (5.3.8) to show that this is the generator of a
transformation that leaves (5.3.9a) invariant.
We next turn to an investigation of the action of a continuous trans-
formation group on a metric. Not all metrics define separations between
points that are independent of the path taken between the points.1 Path-
dependent metrics are termed non-integrable. However the metrics of pri-
mary interest here are integrable and the separation between points may
be written as ds, with ds defined by

ds2 = ΣΣgij (x)dxi dxj . (5.3.10)

October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 145

An infinitesimal transformation of infinitesimal displacements dxj will

leave invariant the Euclidean metric

ds2 = Σdx2j (5.3.11a)

iff

Σ(dx
j )2 = Σ(dxj + δ(dxj ))2 = Σ(dxj )2 , (5.3.11b)

i.e. iff

Σ2dxj δ(dxj ) + δ(dxj )2 = 0. (5.3.11c)

For δ(dxj )2 to vanish one must have

Σ dxj δ(dxj ) = 0. (5.3.11d)

If the generator of the transformation is

U = Σi ξi (x)∂/∂xi , (5.3.12a)

then

δ(dxj ) = δα Σk (∂ξj (x)/∂xk ) dxk . (5.3.12b)

Inserting this into (5.3.11d) and requiring that the result vanish for all
values of δα, one determines that the metric is left invariant by the trans-
formation with generator U if

Σj Σk dxj dxk ∂ξj (x)/∂xk = 0. (5.3.13)

The case of two variables, when ds2 = dx21 + dx22 , is instructive. One
then has

Σj Σk dxj dxk ∂ξj (x)/∂xk

= dx1 dx1 ∂ξ1 (x)/∂x1 + dx1 dx2 ∂ξ1 (x)/∂x2
+ dx2 dx1 ∂ξ2 (x)/∂x1 + dx2 dx2 ∂ξ2 (x)/∂x2 . (5.3.14a)

The invariance condition (5.3.13) thus requires that

dx21 ∂ξ1 (x)/∂x1 + dx1 dx2 (∂ξ1 (x)/∂x2 + ∂ξ2 (x)/∂x1 )

+ dx22 ∂ξ2 (x)/∂x2 = 0. (5.3.14b)

Because dx1 and dx2 are independently variable, dx21 , dx1 dx2 , and dx22 are
linearly independent. Consequently, (5.3.14) can vanish for all choices of
October 28, 2010 16:55 9in x 6in b943-ch05

146 Dynamical Symmetry

dx1 and dx2 iff

∂ξ1 (x)/∂x1 = 0, ∂ξ1 (x)/∂x2 + ∂ξ2 (x)/∂x1 = 0, ∂ξ2 (x)/∂x2 = 0.

(5.3.15)
The first of these equations requires that ξ1 = ξ1 (x2 ), and the last requires
that ξ2 = ξ2 (x1 ). Consequently, the middle equation can be written

∂ξ1 (x2 )/∂x2 + ∂ξ2 (x1 )/∂x1 = 0. (5.3.16)

Since the first term in this is a function of x2 only, while the second is a
function of x1 only, and x1 and x2 are independently variable, one must
have

∂ξ1 (x2 )/∂x2 = c = −∂ξ2 (x1 )/∂x1 , (5.3.17)

where c is an arbitrary constant. Solving these two differential equations

one obtains

ξ1 = a + cx2 , ξ2 = b − cx1 , (5.3.18)

where a and b are also arbitrary constants. It follows that the generator of
the group that leaves the metric dx21 + dx22 invariant, is

U = ξ1 ∂/∂x1 + ξ2 ∂/∂x2 = (a + c x2 )∂/∂x1 + (b − c x1 )∂/∂x2

= a ∂/∂x1 + b ∂/∂x2 + c(x2 ∂/∂x1 − x1 ∂/∂x2 ). (5.3.19)

As a, b, and c are arbitrary constants this generator is an arbitrary linear

combination of three generators

Ua = ∂/∂x1 , Ub = ∂/∂x2 , Uc = (x2 ∂/∂x1 − x1 ∂/∂x2 ). (5.3.20)

The first of these generates translations in the x1 direction, the second

generates translations in the x2 direction, and the third generates rota-
tions in the x1 –x2 plane. It will be noted that this analysis implies that
the Euclidean metric in the plane of the Cartesian coordinates (z1, z2 ) is
invariant under three one-parameter Lie transformation groups acting on
(z1, z2 ) — those carrying out translations and rotations in the plane. The
resulting three-parameter group here is of course the Euclidean group E(2)
investigated in Chapter 2.
A non-Euclidean metric

ds2 = ΣΣ gij (x) dxi dxj , (5.3.21)

October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 147

is invariant under the infinitesimal transformation with generator

U = Σnl ξj (x)∂/∂xj , (5.3.22a)

such that

δ(dxj ) = δα Σk (∂ξj (x)/∂xk )dxk , (5.3.22b)

if, to first-order in δα one has ds
2 = ds2 , where ds
2 is given by

Σij ((1 + δαU)gij (x))(dxi + δαΣk ∂ξi /∂xk dxk )(dxj + δαΣm ∂ξj /∂xm dxm ).
(5.3.22c)

For this to vanish to first-order in δα, the following expression must vanish:

Σi Σj {(Ugij (x))dxi dxj +gij(x)(dxi Σm ∂ξj /∂xm dxm +dxj Σk ∂ξi /∂xk dxk )}.
(5.3.22d)

Examples of the application of this result are developed in the exercises at

the end of the chapter.

5.4 Point Transformations of Derivatives

It was pointed out in Chapter 3 that a function f(t) and its derivative
df/dt may be varied independently. A variation in f determines a variation
in df. Because dt can be varied independently of variations in f and df,
the ratio df/dt can be varied independently of f. Thus one can investigate
variations in the derivative df/dt by investigating the effect of independent
variations of df and dt. We may, in particular, use the results of the previous
section to investigate infinitesimal transformations that convert δx1 /δx2 to
δx
1 /δx
2 , ratios which we may also treat as derivatives dx1 /dx2 ∼ df/dx2
and dx
1 /dx
2 ∼ df/dx
2 . Transformations of a derivative such as df/dx,
which depend only upon variables x that define the position of a point are
termed point transformations. They move the points along curves, without
any reference to the slope of the curve other than that resulting from the
requirement df = 0.
An infinitesimal point transformation with generator

U = ξ1 (x)∂/∂x1 + ξ2 (x)∂/∂x2 (5.4.1)

produces variations in dx1 and dx2 , given by

δ(dx1 ) = dx∗1 − dx1 = δα ((∂ξ1 /∂x1 ) dx1 + (∂ξ1 /∂x2 ) dx2 ),

(5.4.2)
δ(dx2 ) = dx∗2 − dx2 = δα ((∂ξ2 /∂x1 ) dx1 + (∂ξ2 /∂x2 ) dx2 ).
October 28, 2010 16:55 9in x 6in b943-ch05

148 Dynamical Symmetry

This implies
dx∗1 = dx1 + δα ((∂ξ1 /∂x1 ) dx1 + (∂ξ1 /∂x2 ) dx2 ),
(5.4.3a)
dx∗2 = dx2 + δα ((∂ξ2 /∂x1 ) dx1 + (∂ξ2 /∂x2 ) dx2 ).
As the group properties of the transformation are governed by terms first-
order in the group parameter, it is sufficient to expand this through first-
order in δα. To this order,
dx∗1 /dx∗2 = dx1 /dx2 + δα ξ1,2 ,
with
ξ1,2 = ∂ξ1 /∂x1 dx1 /dx2 + ∂ξ1 /∂x2
−(dx1 /dx2 )(∂ξ2 /∂x1 dx1 /dx2 + ∂ξ2 /∂x2 ). (5.4.3b)
When dx1 /dx2 is replaced by df/dx2 ,
ξ1,2 = ∂ξ1 /∂x1 df/dx2 + ∂ξ1 /∂x2 − (df/dx2 )(∂ξ2 /∂x1 df/dx2 + ∂ξ2 /∂x2 ).
(5.4.3c)
Note that ξ1,2 contains terms that depend upon the zero, first, and second
power of df/dx2 .
One may use (5.4.3) to determine the one-parameter Lie groups of point
transformations whose action will leave invariant a derivative dx1 /dx2 . It
will be invariant iff ξ1,2 vanishes for all values of dx1 /dx2 . As (dx1 /dx2 )2 ,
(dx1 /dx2 ), and (dx1 /dx2 )0 are linearly independent, this requires
∂ξ1 /∂x2 = 0,
∂ξ1 /∂x1 − ∂ξ2 /∂x2 = 0, (5.4.4)
∂ξ2 /∂x1 = 0.
The first of these equations implies that ξ1 must be a function of x1 only,
and the last implies that ξ2 must be a function of x2 only. The middle
equation therefore requires that
∂ξ1 (x1 )/∂x1 = ∂ξ2 (x2 )/∂x2 . (5.4.5)
Because x1 and x2 can be varied independently, this is only possible if each
of the two derivatives is equal to the same constant, say a. It follows that
ξ1 = a x1 + b, ξ2 = a x2 + c. (5.4.6)
The corresponding generator of the transformations of x1 and x2 is
U = (a x1 + b)∂/∂x1 + (a x2 + c)∂/∂x2 . (5.4.7)
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 149

Thus U is an arbitrary linear combination of the three generators

Ua = x1 ∂/∂x1 + x2 ∂/∂x2, Ub = ∂/∂x1 , Uc = ∂/∂x2 . (5.4.8)

They generate one-parameter groups of uniform scalings and translations:

(x1 , x2 ) → (eα x1 , eα x2 ), x1 → x1 + β, x2 → x2 + γ. (5.4.9)

As noted above, when x1 is considered as a function, f(x2 ), the trans-

formation yielding (5.4.3) is said to be a point transformation. Then

∂ξk /∂x1 dx1 /dx2 + ∂ξk /∂x2 → ∂ξk /∂f df/dx2 + ∂ξk /∂x2 = Dξk /Dx2 ,
(5.4.10)
the total derivative of ξk with respect to x2 . Consequently,

ξ1,2 → ∂ξ1 /∂f df/dx2 + ∂ξ1 /∂x2 − (df/dx2 )(∂ξ2 /∂f df/dx2 + ∂ξ2 /∂x2 ),
(5.4.11)
i.e.

ξ1,2 = Dξ1 /Dx2 − (df/dx2 )(Dξ2 /Dx2 ).

5.5 Contact Transformations

Contact transformations appear in mechanics when one investigates the
motion of an object in contact with a surface on which it is able to roll
without slipping. In Fig. 5.5.1 the circular wheel of radius r rotates through
an angle dθ about its axis as it rolls along a straight wire. If the center
of the wheel advances a distance δx = rδθ, no slipping has occurred. A
contact transformation preserves the form of this condition, converting it
to δx
= rδθ
, and ultimately into δx∗ = rδθ ∗ . The equivalent equation

δx − rδθ = 0 (5.5.1a)

Fig. 5.5.1 ∆x = r∆θ.

October 28, 2010 16:55 9in x 6in b943-ch05

150 Dynamical Symmetry

is an example of a Pfaffian equation

δv − Σi yi δxi = 0. (5.5.1b)
∗
A transformation that converts the variables in (5.5.1b) to v , yi∗ , x∗i , will
leave the equation invariant iff
δv∗ − Σi yi∗ δx∗i = ρ(v, yi , xi )(δv − Σi yi δxi ), (5.5.2)
with ρ = 0 identically. A contact transformation is a transformation that
transforms v and other variables in the Pfaffian equation (5.5.1b), while
leaving the equation invariant.
Equation (5.5.1a) requires that δx/δθ = r, a constant. It is a special
case of the more general equations
δx1 − x3 δx2 = 0,
(5.5.3a,b)
dx1 − x3 dx2 = 0.
The latter equation is integrable, and establishes that x1 = f(x2 ) plus a
constant. In this case
δx1 /δx2 → dx1 /dx2 → df(x2 )/dx2 . (5.5.3c)
A contact transformation will convert the derivative df(x2 )/dx2 into a
derivative df ∗ (x∗2 )/dx∗2 , and depend upon df(x2 )/dx2 .
To determine the form of the generator of a contact transformation of
Eqs. (5.5.3), let the generator be
U = ξ1 (x1 , x2 , x3 )∂/∂x1 + ξ2 (x1 , x2 , x3 )∂/∂x2 + ξ3 (x1 , x2 , x3 )∂/∂x3 (5.5.4)
It generates variations in dx1 , dx2 and x3 given by
δ(dx1 ) = δα Σk (∂ξ1 (x)/∂xk ) dxk ,
δ(dx2 ) = δα Σk (∂ξ2 (x)/∂xk ) dxk , (5.5.5a–c)
δ(x3 ) = δα ξ3 (x).
To first-order in δα this infinitesimal transformation induces the variation
δ(dx1 − x3 dx2 ) ≡ δαS, with
S = Σk (∂ξ1 (x)/∂xk )dxk − x3 Σk (∂ξ2 (x)/∂xk )dxk − ξ3 (x) dx2
(5.5.6a,b)
= A1 dx1 + A2 dx2 + A3 dx3 .
In this last expression
A1 = ∂ξ1 (x)/∂x1 − x3 ∂ξ2 (x)/∂x1 ,
A2 = ∂ξ1 (x)/∂x2 − x3 ∂ξ2 (x)/∂x2 − ξ3 (x), (5.5.6c–e)
A3 = ∂ξ1 (x)/∂x3 − x3 ∂ξ2 (x)/∂x3 .
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 151

Invariance requires that

S = ρ(x)(dx1 − x3 dx2 ), (5.5.7)
because the dx
s are independent, invariance will ensue if
A1 → ρ(x), A2 → −x3 ρ(x), A3 → 0. (5.5.8a–c)
From (5.5.6) one thus obtains
∂ξ1 (x)/∂x1 − x3 ∂ξ2 (x)/∂x1 = ρ(x),
ξ3 (x) = ∂ξ1 (x)/∂x2 − x3 ∂ξ2 (x)/∂x2 + x3 ρ(x), (5.5.9a–c)
∂ξ1 (x)/∂x3 − x3 ∂ξ2 (x)/∂x3 = 0.
As these equations are compatible with the assumption that ρ(x) is an
arbitrary function, they have an uncountable infinity of solutions. Thus
an uncountable infinity of Lie transformation groups will leave a derivative
dx1 /dx2 invariant if the transformations are allowed to functionally depend
upon the derivative itself. This result contrasts with the result of Sec. 5.4
above, where it was found that a derivative dx1 /dx2 is left invariant by just
three one-parameter groups that carry out point transformations.
Additional properties of contact transformations are uncovered in the
exercises at the end of the chapter.
In concluding the discussion in this section we would note that
derivative-dependent transformations play a more important, and simpler,
role in studies of partial differential equations than they do in studies
of ordinary differential equations. This will become evident in subsequent
chapters.

5.6 Invariance of an Ordinary Differential Equation

of First-Order under Point Transformations;
Extended Generators
For notational convenience, let dy/dx = yx and let the Lie generator that
acts on x and y be written
U = ξ(x, y)∂/∂x + η(x, y)∂/∂y. (5.6.1)
It determines an infinitesimal point transformation
x → x + δα ξ(x, y), y → y + δα η(x, y), (5.6.2)
and an infinitesimal transformation that acts on yx . Let the latter be
yx → yx + δα η x . (5.6.3)
October 28, 2010 16:55 9in x 6in b943-ch05

152 Dynamical Symmetry

One can define an extended generator

U(1) = U + η x ∂/∂yx . (5.6.4)
It is the generator of a Lie group with operator exp(αU(1) ) which acts in
the space of yx , x, y. One then has, from (5.4.11),
η x = Dη/Dx − yx Dξ/Dx. (5.6.5)
= ∂η/∂x + yx (∂η/∂y − ∂ξ/∂x) − yx2 ∂ξ/∂y.
(1)
Using Theorem 5.1.2 with U interpreted as U , one sees that the
equation
yx − f(x, y) = 0 (5.6.6)
is invariant under the action of a Lie group with extended generator U(1)
iff
U(1) (yx − f(x, y)) = 0 when yx − f(x, y) = 0. (5.6.7a)
This requires that
ηx − (ξ(x, y)∂f/∂x + η(x, y)∂f/∂y) = 0 when yx − f(x, y) = 0.
(5.6.7b)
Consequently, for invariance of (5.6.6), one must have
∂η/∂x + f(∂η/∂y − ∂ξ/∂x) − f 2 ∂ξ/∂y − Uf = 0. (5.6.8)
One may use this analysis to determine the general form of the ODEs
yx = f(x, y) that are left invariant by the transformations of any one-
parameter group of physical interest. Consider, for example, the group of
rotations about the origin in the x–y plane. Let U be the rotation generator
U = x∂/∂y − y∂/∂x. In this case ξ = −y, η = x. Using (5.6.5) one finds
ηx = (1 + yx2 ). (5.6.9)
(1)
If U is to generate an invariance transformation of the equation yx =
f(x, y), then (5.6.7) requires that (5.6.8) becomes
(1 + f 2 ) = Uf. (5.6.10)
√
To solve Eq. (5.6.10) we may change from x, y to r = (x2 + y2 ), and
θ = arctan(y/x), the canonical coordinates for U. Then U = ∂/∂θ. Letting
f(x, y) = f(r cos(θ), r sin(θ)) = F(r, θ), (5.6.11)
Eq. (5.6.10) becomes
∂F/∂θ = (1 + F2 ). (5.6.12)
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 153

Its solution is
F = −cotan(θ + g); g an arbitrary function of r. (5.6.13)
To express F(r, θ) as f(x, y) one may use the relations
cotan(θ + g) = cos(θ + g)/ sin(θ + g),
cos(θ + g) = cos(θ) cos(g) − sin(θ) sin(g), (5.6.14)
sin(θ + g) = sin(θ) cos(g) + cos(θ) sin(g).
On setting cos(θ) = x/r, sin(θ) = y/r, one finds
f = −(ax − by)/(ay + bx), (5.6.15a)
with a, b functions of r such that
a(r)2 + b(r)2 = 1. (5.6.15b)
With this restriction, the equation
dy/dx = −(ax − by)/(ay + bx) (5.6.16)
states the form of the most general ODE that is invariant under the group
of rotations carried out by exp(αU), U = x∂/∂y − y∂/∂x. U itself generates
the path curves defined locally by
dx/(−y) = dy/x = dα, (5.6.17a)
or equivalently, by
dθ = dα, r = const. (5.6.17b)
It follows that along any path curve generated by U,
dy/dx = −x/y, (5.6.18a)
and
xdx + ydy = 0 = d(r2 /2). (5.6.18b)
Equation (5.6.18a) coincides with Eq. (5.6.16) if b = 0. In all other cases
the circular path curves, evolved by the action of exp(αU) on r = (x, y), do
not coincide with the solution curves of the differential equation (5.6.16).
Thus, when b = 0, in Eq. (5.6.16), rotations interconvert its solution curves.
Lie1 showed that in such cases, the functions ξ and η in the generator
U = ξ(x, y)∂/∂x + η(x, y)∂/∂y determine an integrating factor for the dif-
ferential equation yx = f(x, y). This discovery is discussed in Appendix A
of this chapter.
October 28, 2010 16:55 9in x 6in b943-ch05

154 Dynamical Symmetry

5.7 Invariance of Second-Order Ordinary Differential

Equations under Point Transformations; Harmonic
Oscillators
In the previous section we saw that an infinitesimal point transformation
with parameter α and generator

U = ξ(x, y)∂/∂x + η(x, y)∂/∂y, (5.7.1)

induces a transformation of the derivative yx to yx + δα η x . It also induces

a transformation of all higher order derivatives. The second derivative yxx
is altered to yxx + δαη xx . An argument parallel to that of Sec. 5.4 gives

ηxx = Dη x /Dx − yxx Dξ/Dx,

= ∂ηx /∂x + yx ∂η x /∂y + yxx ∂η x /∂yx − yxx (∂ξ/∂x + yx ∂ξ/∂y),
= ηxx + yx (2ηxy − ξxx ) + yx2 (ηyy − 2ξxy ) − yx3 ξyy
+ (ηy − 2ξx )yxx − 3yx yxx ξy . (5.7.2a–c)

The corresponding twice extended generator is

U(2) = U + η x ∂/∂yx + ηxx ∂/∂yxx . (5.7.3)

A second-order differential equation

g(x, y, yx , yxx ) = 0 (5.7.4)

is invariant under the transformation (x, y) → (exp(αU)x, exp(αU)y) iff

0 = (U(2) g)|g=0 . (5.7.5)

As an application of (5.7.5), let us determine the point transformations

that leave invariant Newton’s equation of motion for a harmonic oscillator.
Let this equation of motion be

d2 y/dt2 + y = 0. (5.7.6)

Here y represents the position of an oscillating mass and dt corresponds

to an infinitesimal time interval. On setting U = ξ(t, y)∂/∂t + η(t, y)∂/∂y,
one requires

U(2) (d2 y/dt2 + y) (5.7.7)

October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 155

to vanish identically, when (d2 y/dt2 + y) vanishes. Equations (5.7.2c) and

(5.7.6) then yield the determining equation
η + ηtt + yt (2ηty − ξtt ) + yt2 (ηyy − 2ξty ) − yt3 ξyy − (ηy − 2ξt )y + 3yty ξy = 0.
(5.7.8)
Here, as required by (5.7.5), we have used (5.7.6) to replace ytt by −y. In
(5.7.8), ξ and η are functions of y, t, which with yt are independently vari-
able. Because different powers of yt are linearly independent, the coefficient
of each power of yt in (5.7.8) must vanish. Consequently,
ξyy = 0, ηyy − 2ξty = 0,
(5.7.9a–d)
2ηyt − ξtt + 3yξy = 0, η + ηtt − y(ηy − 2ξt ) = 0.
To solve these equations, begin with (5.7.9a), from which follows
ξ = ξ0 (t) + yξ1 (t). (5.7.9e)
It then follows from (5.7.9b) that
ηyy = 2∂ξ1 (t)/∂t. (5.7.9f)
Inserting these results in the remaining two equations one finds that the
solutions of the equations are linear combinations of eight independent solu-
tions:
η = Σcj ηj , ξ = Σcj ξj , (5.7.10)
the cj being arbitrary constants of integration. The solutions may be given
the form
ξ = c1 + c3 (−y cos(t)) + c4 y cos(t) + c5 y sin(t)
+ c6 (−y sin(t)) + c7 sin(2t) + c8 cos(2t)
(5.7.11)
η = c2 y + c3 (1 + y2 ) sin(t) + c4 (1 − y2 ) sin(t)
+ c5 (1 + y2 ) cos(t) + c6 (1 − y2 ) cos(t)c7 y cos(2t) + c8 (−y sin(2t)).
The resulting generator U = ξ∂/∂t + η∂/∂y is thus of the form Σcj Uj . As
there is no choice for the eight constants cµ such that Σcµ = 0, there are
eight linearly independent generators Uj :
U1 = ∂/∂t,
U2 = y∂/∂y,
U3 = (1 + y2 ) sin(t)∂/∂y − y cos(t) ∂/∂t,
October 28, 2010 16:55 9in x 6in b943-ch05

156 Dynamical Symmetry

U4 = (1 − y2 ) sin(t)∂/∂y + y cos(t) ∂/∂t,

U5 = (1 + y2 ) cos(t)∂/∂y + y sin(t) ∂/∂t,
U6 = (1 − y2 ) cos(t)∂/∂y − y sin(t) ∂/∂t,
U7 = y cos(2t)∂/∂y + sin(2t) ∂/∂t,
U8 = −y sin(2t)∂/∂y + cos(2t) ∂/∂t.
(5.7.12)
In our discussion of point transformations that leave a first-order ODE
invariant we found the generators contained arbitrary functions. In contrast,
we have just found that our second-order ODE admits point transforma-
tions with just eight linearly independent generators, U, none of which
contains an arbitrary function. Note however, that we did not obtain the
evolution generator of the oscillator given in Chapter 3; it is an explicit func-
tion of the momentum, p, which for (5.7.6), is just yt . If we had allowed
our generators in this section to depend upon yt as well as y and t, that
is, to be generators of contact transformations, Eq. (5.7.8) would not have
separated into four equations. We would not have obtained a finite number
of linearly independent generators each with no free functions. In the gen-
eral case, a second-order ODE is invariant under contact transformations
whose generators may depend upon arbitrary functions.
The harmonic oscillator is one of a small class of second-order ODEs
that are invariant under eight different one-parameter groups of point trans-
formations. This is the maximum number allowed any second-order ODE.3

5.8 The Commutator of Two Operators

The commutator
[U, V] ≡ UV − VU (5.8.1)
of two operators U and V plays a central role in the study of Lie groups.
If [U, V] = 0, then UV = VU, and by definition, U and V commute.
From the definition, it follows that
[U, V] = −[V, U] (5.8.2)
and
[U1 + U2 , V] = [U1 , V] + [U2 , V]. (5.8.3)
A variable c is a parameter if it commutes with U and V; then
[cU, V] = c[U, V] = [U, cV] = [U, V]c. (5.8.4)
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 157

The operator identity

[[U1 , U2 ], U3 ] + [[U3 , U1 ], U2 ] + [[U2 , U3 ], U1 ] = 0 (5.8.5)
may be established by writing the left-hand side out in full and collecting
terms. It is known as Jacobi’s relation.
The following lemma states an important property of the commutator
of two first-order differential operators.:
Lemma 5.8.1. If U and V are first-order differential operators
U = Σ ξi (x)∂/∂xi , V = Σ υj (x)∂/∂xj , (5.8.6)
defined by differentiable ξi (x) and υj (x), and f(x) is a twice differentiable
function of its arguments, then
[U, V]f(x) = Wf(x), (5.8.7a)
where
W = Σ ωj (x)∂/∂xj (5.8.7b)
is also a first-order differential operator.
The proof is direct. One has
[U, V]f(x)
= (Σi ξi (x)∂/∂xi Σj υj (x)∂/∂xj − Σj υj (x)∂/∂xj Σi ξi (x)∂/∂xi )f(x)
= Σi Σj (ξi (x)∂/∂xi υj (x)∂/∂xj − υj (x)∂/∂xj ξi (x)∂/∂xi )f(x)
= Σi Σj (ξi (x) ∂υj (x)/∂xi ∂/∂xj − υj (x) ∂ξi (x)/∂xj ∂/∂xi)f(x)
+ Σi Σj (ξi (x) υj (x)∂/∂xi ∂/∂xj − υj (x) ξi (x)∂/∂xj ∂/∂xi)f(x).
Because f(x) is twice differentiable the last line vanishes. On interchanging
summation indices in the second term of the previous equality one obtains
[U, V]f = Wf, (5.8.8)
with
ωj (x) = Σi (ξi (x) ∂υj (x)/∂xi − υj (x) ∂ξi (x)/∂xi ).
This result is often written
[U, V] = W, (5.8.9)
presupposing that the differential operators U, V, W are to be applied to
functions that are at least twice continuously differentiable.
In the Appendix B to this chapter it will be shown that a diffeomorphism
that transforms the U, V and W, of (5.8.9) to U
, V
, W
also transforms
[U
, V
] to W
. In short, diffeomorphisms do not alter the commutation rela-
tions of first-order differential operators.
October 28, 2010 16:55 9in x 6in b943-ch05

158 Dynamical Symmetry

5.9 Invariance of Sets of ODEs. Constants of Motion

In this section we use the commutator of differential operators to develop
an alternative criterion for determining the generators of groups that leave
sets of first-order ODEs invariant. The commutator criterion will be used to
characterize the constants of motion of dynamical systems whose evolution
is governed by ordinary differential equations.
We begin with the set of autonomous ODEs

dxj /dt = fj (x), j = 1, . . . , n, (5.9.1a)

or, the equivalent set

dt = dx1 /f1 (x) = dx2 /f2 (x) · · · = dxn /fn (x). (5.9.1b)

Then, if

V = Σj fj (x)∂/∂xj , (5.9.2)

the evolution operator exp(tV) acts on the xj to convert them from their
values at t = 0, to their values at t = t
.
If all the f’s are at least C1 functions, then, except in regions where
(5.9.1) has singular points, local diffeomorphisms

x → y = Y(x), with yi = Yi (x),

(5.9.3)
y → x = X(y), xi = Xi (y) = Yiinv (x),

will convert V to V
= ∂/∂y1 , and (5.9.1b) to the canonical form

dy1 /dt = 1, dyj /dt = 0, j = 2, . . . , m, (5.9.4)

with solutions

y1 − t = c1 , yj = cj , j = 2, . . . , m. (5.9.5)

As yj = Yj (x), the first relation in (5.9.5) establishes that Y1 (x) = t + c1 .

For j > 1, (5.9.5) establishes the relations Yj (x) = cj . If Eqs. (5.9.1) govern
the time evolution of a system of bodies in motion, these Yj (x) are constants
of the motion or integrals of the motion.a

a Y1 (x) − t is sometimes called a time-dependent constant of motion. Constants of

motion Yj (x) are divided into two classes: isolating integrals and non-isolating integrals.
The distinction is addressed in Appendix C.
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 159

To prepare for the statement of the theorem which follows, we append

to Eqs. (5.9.2) the equation dt/dτ = 1, and use the relation dt = dτ to
replace (5.9.1b) by

dτ = dt/1 = dx1 /f1 (x) = dx2 /f2 (x) · · · = dxn /fn (x). (5.9.6)

The evolution operator of these equations will be taken to be exp(τ Vt ),

V = ∂/∂t + V = ∂/∂t + Σj fj (x)∂/∂xj . (5.9.7)

Because the diffeomorphism that converts (5.9.1) to (5.9.4) is independent

of t, it converts (5.9.6) to

dt/dτ = 1, dy1 /dτ = 1, dyj /dτ = 0, j = 2, . . . , m. (5.9.8)

Thus Vt is converted to

Vt
= ∂/∂t + ∂/∂y1 . (5.9.9)

Vt
acts on the solution y1 − t = c1 to give zero, and it acts on all the
other yj to give zero. It follows that if φ(y, t) = c represents any solution
of (5.9.4), with or without explicit time dependence, then

Vt
φ(y, t) = 0, (5.9.10a)

and

exp(τ Vt
) φ(y, t) = φ(y, t). (5.9.10b)

On transforming back to the x system, φ(y, t) → Φ(x, t), and one obtains
the corresponding relations

Vt Φ(x, t) = 0, (5.9.11a)

and

exp(τ Vt )Φ(x, t) = Φ(x, t), (5.9.11b)

where Φ1 (x, t) = c defines the solution of (5.9.1) corresponding to y1 − t =

c1 , and the other Φj correspond to yj = cj . Let us proceed:

Theorem 5.9.1. Let the solutions Φ(x, t) = c and Φ(x) = c, of the equa-
tions

dxj /dt = fj (x), j = 1, . . . , n, x = (x1 , . . . , xn ), (5.9.12)

October 28, 2010 16:55 9in x 6in b943-ch05

160 Dynamical Symmetry

be at least twice differentiable. Let

Vt = ∂/∂t + Σj fj (x)∂/∂xj , j = 1, . . . , n. (5.9.13)

Require that

U = ξ0 (x, t)∂/∂t + Σk ξk (x, t)∂/∂xk , k = 1, . . . , n (5.9.14)

satisfy the operator equation

[Vt , U] = w(x, t)Vt , (5.9.15)

where w(x, t) is an arbitrary function. Then U is the generator of an invari-

ance group of Eqs. (5.9.12).

Proof. Applying both sides of (5.9.15) to a function Φ(x, t) or Φ(x) which

defines a solution Φ = c, and using the definition of the commutator,
Eq. (5.9.16) becomes

(Vt U − Ut V)Φ = w(x, t)Vt Φ. (5.9.16)

The fact that for any solution, Φ = c, Vt Φ = 0, requires that the right-hand
side of the equation, as well as UVt Φ, vanish, whence

Vt (UΦ) = 0. (5.9.17)

Thus UΦ is a solution of Eqs. (5.9.12). Consequently, the one-

parameter family of functions exp(αU)Φ satisfies Eqs. (5.9.12).
Q.E.D.
To exemplify the previous discussion, suppose n = 3 and that one wishes
to have transformations that will convert a given solution of (5.9.4), and
hence the differential Eqs. (5.9.1) from which it originated, into any other
solution. The solutions of (5.9.4) are y1 − t = c1 , y2 = c2 , y3 = c3 . To
convert a particular solution to a general solution one must be able to
change all three initial values cj through the complete allowed range. If
all three constants may range over all the reals, then transformations with
the generators exp(aj ∂/∂yj ) will have the effect of converting each cj to
c
j = cj + aj , and thereby yield all other solutions. In this case one or more
of the yk might be an angular variable. If one of the constants, say c2 , is
restricted to be nonnegative, the dilatation operator exp(αy2 ∂/∂y2 ) will
convert the solution y2 = c2 to exp(α)x2 = c2 , i.e. to y2
= exp(−α)c2 ,
giving all allowed nonzero solutions if c2 is nonzero. Such a situation would
arise if y2 were to be a radial distance, r.
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 161

One can distinguish the various possible cases if one knows the trans-
formation that puts the original differential equations into canonical form.
One can avoid this by solving the determining equations that arise from
the original differential equations themselves. The availability of computer
programs that set up and solve determining equations often make this the
most practical procedure. A basic mathematical discovery of Reid enables
some of these programs to produce series approximations to generators.4
This is particularly useful when one does not expect exact results in closed
form.5
Jacobi’s relation ensures that if U1 and U2 satisfy

[V, U1 ] = c1 V, [V, U2 ] = c2 V, (5.9.18)

then

U3 = [U1 , U2 ] (5.9.19a)

satisfies

[V, U3 ] = 0. (5.9.19b)

As a consequence, if V (or Vt ) is the evolution operator of a set of ODEs,

knowing a U1 and a U2 , which satisfy (5.9.18) and do not commute, allows
one to know the generator of another one-parameter invariance group of
the ODEs. This procedure may often be repeated. A set of operators
U1 , U2 , U3 , . . . obtained in this way generates groups that often provide
a great deal of information about the physical system whose evolution is
governed by V and U. These groups with multiple generators are the subject
of the next chapter.

5.10 Conclusion
This chapter has developed basic methods for uncovering and using invari-
ance properties of ordinary functions and equations, and of expressions
involving infinitesimal changes — metrics, Pfaffians and derivatives. Par-
ticular attention has been paid to invariance properties of systems governed
by ordinary differential equations.
Trajectories or solutions defined by a given set of ODEs may be con-
verted into one another by one-parameter transformation groups if the tra-
jectories or solutions are diffeomorphic. The groups are invariance groups
of the differential equations; they define intrinsic symmetries possessed by
the equations.
October 28, 2010 16:55 9in x 6in b943-ch05

162 Dynamical Symmetry

Ordinary differential equations are left invariant by an uncountable

infinity of one-parameter transformation groups. These establish a host of
relations between solutions. One may choose among the groups according to
the purpose at hand. A common choice is to provide a set of one-parameter
groups sufficient to convert any given particular solution into all other solu-
tions that have diffeomorphic trajectories. Especially useful information
comes to light if a set of one-parameter groups comprise a many-parameter
group. Many-parameter groups are defined and explored in the next chapter.
Their mathematical properties will be found to restrict geometric interpre-
tations of the intrinsic symmetries of the differential equations.
For further information on the use of systematic group theoretic meth-
ods to solve ordinary differential equations the reader is referred to refer-
ences listed at the end of this chapter. The recent monographs of Bluman
and Kumei,6 and Cantwell,7 provide clear, useful, and extensive treatments
of the invariance properties of many differential equations important in sci-
ence and engineering. Further helpful monographs are also listed at the end
of the chapter.

Appendix A. Relation between Symmetries

and Intergrating Factors
As noted earlier, a first-order ODE dy/dx = f(x, y) may be written as
M(x, y) dx + N(x, y) dy = 0, (5.A.1)

and its family of solutions may be put in the form

φ(x, y) = c, (5.A.2)

where c is a parameter. Both expressions avoid a distinction between inde-

pendent and dependent variables. From (5.A.2) it follows that

dφ = ∂φ/∂x dx + ∂φ/∂y dy = 0. (5.A.3)

For both (5.A.1) and (5.A.3) to hold true it is necessary that

∂φ/∂x = A(x, y) M(x, y), ∂φ/∂y = A(x, y) N(x, y). (5.A.4)

A(x,y) is an integrating factor for Eq. (5.A.1), for multiplying (5.A.1) by

A(x,y) yields
A(x, y)(M(x, y)dx + N(x, y)dy) (5.A.5a)
= ∂φ/∂x dx + ∂φ/∂y dy = dφ. (5.A.5b)
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 163

It was shown in (5.1.13) that, if U = ξ(x, y)∂/∂x + η(x, y)∂/∂y is such

that

Uφ(x, y) = ξ(x, y)∂φ/∂x + η(x, y)∂φ/∂y = 1, (5.A.6)

then exp(αU) has the effect of converting a solution φ(x, y) = c, into a

different solution φ(x, y) = c
with c
= c exp(−α). We next show that
knowing such a U allows one to determine the integrating factor A(x,y).
Using (5.A.4) to eliminate the derivatives in (5.A.6) one obtains

A(x, y)(ξ(x, y)M(x, y) + η(x, y)N(x, y)) = 1. (5.A.7)

It follows that

A(x, y) = 1/(ξM + ηN). (5.A.8)

Thus, 1/(ξM + ηN) is an integrating factor of differential equation (5.A.1).

This connection between integrating factors for differential equations
and transformations that interconvert solutions of differential equations is
of considerable utility. Cohen’s monograph, The Lie Theory of One Param-
eter Groups,8 contains an extensive tabulation of the invariance groups of
many types of first-order ordinary differential equations, together with their
integrating factors.

Appendix B. Proof That the Commutator of Lie Generators

is Invariant under Diffeomorphisms
Let T symbolize the change of variables defined by a diffeomorphism

xj → yj = Txj = fj (x), (5.B.1a)

for example,

x → y = Tx = x2 , 0 < x < ∞. (5.B.1b)

The inverse transformation, symbolized by Tinv , is defined by

yj → xj = Tinv yj = fjinv (y), (5.B.1c)

e.g.

y → x = Tivn y = y1/2 , 0 < y < ∞. (5.B.1d)

When the transformations of a diffeomorphism act on a function of the

x or y variables of (5.B.1), they, by definition, act on the variables x,y
October 28, 2010 16:55 9in x 6in b943-ch05

164 Dynamical Symmetry

themselves. That is,

TG(x) = G(Tx) = G(y) = G(f(x)), (5.B.2a)

inv ivn inv
T G(y) = G(T y) = G(x) = G(f (y)). (5.B.2b)

Note that the transformations of a Lie group acting on an analytic function

g(x) obey (5.B.2a) and (5.B.2b) becausea

exp(αU)g(x) = T(α)g(x) = g(T(α)x). (5.B.3)

In an expression such as Tg1 (x)g2 (x), T is construed to act on the

compound expression, that is

Tg1 (x)g2 (x) = T(g1 (x)g2 (x)) = g1 (Tx)g2 (Tx). (5.B.4)

As a consequence one may use the identity Tinv T = I to write

Tg1 (x)g2 (x) = Tg1 (x)Tinv Tg2 (x) = (Tg1 (x)Tinv )Tg2 (x),
(5.B.5)
= (Tg1 (x)Tinv )g2 (Tx).

To determine which form

U = Σξj (x)∂/∂xj (5.B.6)

takes when acting on the y variables of (5.B.1), we begin with the action of U
on xk , viz. Uxk , and use T to transform it to an action on the corresponding
y variable. Then

T(Uxk ) = (TUTinv )Txk (5.B.7a)

inv inv
= (TUT )yk = (TUT )fk (x). (5.B.7b)

The first term in (5.B.7a) is evidently just Tξk (x) = ξk (Tx). The first term
in (5.B.7b) represents the action of the transformed generator on yk . In
Sec. 5.7, we investigated the effect of a change of variables on a group
generator. Setting yk = x
k in Eqs. (4.7.3)–(4.7.5) gives

ξk
(y) = Uyk |x=f inv (y) = (Σj ξj (x)fk,j (x))|x=f inv (y) , (5.B.8a)

a Some authors require (5.B.3) to be true, by definition, for any g(x). We have not done

so because, in physical applications, doing so can easily lead to error.

October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 165

where
fk,j (x) = ∂fk (x)/∂xj . (5.B.8b)
This establishes that the diffeomorphism x → y = Tx transforms
U = Σ ξj (x)∂/∂xj to
TUTinv = Σ ξk
(y)∂/∂yk . (5.B.9)
We are now ready to prove that the commutator of two Lie generators
is invariant under diffeomorphisms.
Theorem 5.B.1. Let
W = [U, V], (5.B.10a)
with
U = Σ ξi (x)∂/∂xi , V = Σ υj (x)∂/∂xj , W = Σ ωk (x)∂/∂xk . (5.B.10b)
Let
U
= TUTinv , V
= TVTinv , W
= TWTinv . (5.B.11)
Then
W
= [U
, V
]. (5.B.12)
Proof.
TWTinv = T[U, V]Tinv = T(UV − VU)Tinv
= TUTinv TVTinv − TVTinv TUTinv (5.B.13a–c)
in in
= [TUT , TVT ].
That is to say,
W
= [U
, V
]. (5.B.14)
Q.E.D.

Appendix C. Isolating and Non-isolating Integrals of Motion

Roughly speaking, small errors in measurements lead to small errors in the
value of isolating integrals, and small errors in estimates of the value of
isolating integrals result in small errors in predicted motions. In contrast,
small errors in calculating or estimating non-isolating integrals can lead
to large errors in predicted motions. However, the difference in meaning
of small and large in these statements is subtle. We can here only briefly
October 28, 2010 16:55 9in x 6in b943-ch05

166 Dynamical Symmetry

address this subtlety. In Eqs. (5.9.4), fixing the value of c1 , and each of
the constants of motion, yk = Yk (x), k > 1, at a time t = t0 would com-
pletely fix the coordinates of the bodies in the space of the y’s, and in
the space of the x’s, at t0 . As t varies, the point representing the state
of the system traces out a line, which is the trajectory of the system in
the space of the x’s or y’s. Each successive point on this line determines
a state of the system at successive points of time. Because all the yk are
related to the x’s by diffeomorphisms, the Yk (x) are single-valued functions.
As a consequence trajectories evolving from different points cannot cross.
Furthermore, except in the region of singular points of (5.9.1), infinitesi-
mal changes δxj in initial values xj |t=0 , will produce infinitesimal changes
in the constants of motion Yk (x). However, small finite changes ∆xj |t=0
may produce changes ∆Yk |t=0 that grow exponentially with the ∆xj |t=0 ,
and small finite changes ∆yj |t=0 may produce changes in the ∆xk |t=0 that
grow exponentially with the ∆yj |t=0 . Under these circumstances, some tra-
jectories initially well separated may closely approach one another, then
recede — and in general wander about so much that they will pass through
any finite hypervolume of the space allowed for them by conservation of
energy, momentum, angular momentum, and any other isolating integrals.
An example is provided by the Henon and Heiles system,11 in which the
potential function is (x2 +y2 )/2+x2y−y3 /3. In these cases knowing a value
of the corresponding constant of motion can tell little about the evolution
of the system. The constants so affected are the ones that are said to be
non-isolating.
The study of the stability and reliability of predictions based upon con-
stants of motion of systems involving three or more gravitationally interact-
ing bodies has a long history. For a discussion of the contributions made by
Jacobi, Bruns, Painleve, and Poincare in the nineteenth and early twentieth
century, the reader is referred to Whittaker’s monograph.12 These workers
established fundamental mathematical results and showed that constants
of motion other than those of energy, momentum, and angular momentum
would generally have limited use in making long range predictions, and
retrodictions, of planetary and lunar motions. In the mid twentieth century
further fundamental advances were made by Kolmogorov,13 Arnold,14 and
Moser.15 Treating the sun and planets as point masses it was shown, for
example, that for two planetary orbits whose planes intersect at low angles
there are stable constants of motion which predict the constancy of the
inclination.
October 28, 2010 16:55 9in x 6in b943-ch05

Everywhere-Local Invariance 167

Exercises
1. Is the equation dv − ydx = dc, with dc arbitrary, left invariant by a
non-denumerable, or a denumerable, infinity of one-parameter groups?
2. What equations of the form (5.6.16) have path curves that are themselves
circles?
3. Determine the generators of the transformations that leave invariant the
space metrics obtained from (2.7.4) by setting dz = 0.
4. What is the action of each of the generators (5.7.12a–d) on the general
solution of the equation of motion: y − (A sin(t) + B cos(t)) = 0? Which
U’s generate a transformation that leaves a particular solution invariant?
For further information see Ref. 2.
5. Complete the solution of Eqs. (5.7.9a–d) and compare your result with
(5.7.11).
6. Write down a set of first-order ODEs corresponding to the equation
d2 x/dt2 = 0. Obtain the determining equation for the point transfor-
mations that leave this equation invariant. Then use (5.9.16) to obtain
the same determining equation. What choice of the function w must
be made? Solve the determining equations. Show that the solution is a
linear combination of eight linearly independent U’s.

References
[1] S. Lie, Christ. Forh. Aar., 1874 (1875) 242–254; Collected Works, Vol. III,
No. XIII, p. 176.
[2] C. E. Wulfman and B. G. Wybourne, J. Phys. A 9 (1976) 507–518.
[3] S. Lie, Vorlesungen uber Continuierliche Gruppen (Chelsea, Bronx, N.Y.,
1971).
[4] G. Reid, J. Phys. A 23 (1990) L853; G. J. Reid, Eur. J. Appl. Math. 2
(1991) 293; loc. cit., 319.
[5] W. Hereman, Euromath Bull. 1 (1994) 45.
[6] G. W. Bluman and S. Kumei, Symmetries and Differential Equations
(Springer, N.Y., 1989).
[7] B. J. Cantwell, Introduction to Symmetry Analysis (Cambridge University
Press, Cambridge, 2002).
[8] A. Cohen, An Introduction to The Lie Theory of One-parameter Groups
(D. C. Heath, 1911), reprinted by (G. E. Stechert, N.Y., 1931).
[9] P. J. Olver, Application of Lie Groups to Differential Equations (Springer-
Verlag, N.Y., 1986).
[10] R. Sheshadri and T. Y. Na, Group Invariance in Engineering Boundary
Value Problems (Springer-Verlag, N.Y., 1985).
[11] M. Henon and C. Heiles, Astron. J. 69 (1964) 73–9.
October 28, 2010 16:55 9in x 6in b943-ch05

168 Dynamical Symmetry

[12] E. T. Whittaker, A Treatise on the Analytical Dynamics of Particles and

Rigid Bodies, 4th edn., Ch. XII - XV (Cambridge University Press, Cam-
bridge 1959).
[13] A. N. Kolmogorov, Proc. Int. Congress Math. (Amsterdam, 1957); translated
in Appendix D of R. Abraham’s Foundation of Mechanics (Benjamin, N. Y.,
1967).
[14] V. I. Arnold, Russian Math. Surveys 18 (1963) 5,6.
[15] J. Moser, Stable and Random Motions in Dynamical Systems (Princeton
University Press, Princeton, N.J, 1973).
October 28, 2010 16:55 9in x 6in b943-ch06

CHAPTER 6

Lie Transformation Groups and Lie Algebras

In this chapter many-parameter groups are defined, and the concepts of Lie
group and Lie algebra are then developed along the lines initiated by Lie.1
The further topics selected are those most often needed in applied work.

6.1 Relation of Many-Parameter Lie Transformation Groups

to Lie Algebras
Suppose a succession of r different one-parameter transformation groups
with group parameters αk act on a set of variables x = (x1 , . . . , xn ). This
action may be expressed as
x → x
= exp(αr Ur ) exp(αr−1 Ur−1 ) · · · exp(α2 U2 ) exp(α1 U1 )x.
= T(αr , αr−1 , . . . , α2 , α1 )x. (6.1.1a)
Here each
Uk = Σj ξj,k (x)∂/∂xj (6.1.1b)
is the generator of a one-parameter transformation group in which
T(αk )x = exp(αk Uk )x. Let us label the ordered set of group parameters by
ω = (αr , αr−1 , . . . , α2 , α1 ), (6.1.2a)
so T(αr , αr−1 , . . . , α2 , α1 ) becomes T(ω). Then
x
= T(ω)x, (6.1.2b)
and for j = 1, . . . , n,
T(ω)xj = x
j
with
x
j = Φj (x; ω). (6.1.2c)

169
October 28, 2010 16:55 9in x 6in b943-ch06

170 Dynamical Symmetry

We will write ω = 0 iff all αk = 0. If the operation of T(ω) on x is that

of an r-parameter group, then for each xj the functions Φj (x; ω) yield the
identity transformation, T(0), when every group parameter αk vanishes,
and one must have

xj = Φj (x; ω)|ω=0 . (6.1.2d)

For T(ω) to define a group, it is necessary that it has an inverse. Refer-

ring to (6.1.1) one sees that its inverse, Tinv (ω), is

T(ω) = T(−α1 , −α2 , . . . , −αr−1 , −αr ) (6.1.3a)

(the parameters are listed in reverse order because the operations must be
carried out in that order). Then

T(ω in )T(ω)x = T(ω)T(ω in )x = T(0)x = I x = x. (6.1.3b)

For T(ω) to be the operator of a group, it is also necessary that the

action of two successive operations can be expressed as the operation of the
group, that is

T(ω
)T(ω)x = T(ω

)x. (6.1.4a)

It requires that, for all k = 1, . . . , r, one must have α

k = Ωk (ω
, ω), with

α

k → α
k as ω → 0, α
k → αk as ω
→ 0. (6.1.4b)

Definition 6.1.1. A group is said to be a Lie group iff its composition

functions Ωk (ω
, ω) are analytic functions in finite regions containing the
values of ω
and ω that define identity transformations.
It follows that for Lie groups, the functions Ωk can be expanded as a
convergent power series in the group parameters α
j and αj within finite
regions in which the parameters are sufficiently small.
To begin to develop an understanding of what is entailed when a group
is a Lie group, consider the case of a two-parameter transformation group
in which

x
= T(α2 , α1 )x = exp(α2 U2 ) exp(α1 U1 )x, (6.1.5)

with x = (x1 , x2 ). As the group-parameters approach zero,

T(α2 , α1 ) → T(δα2 , δα1 ),

October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 171

z2 / e P'
T1
P

T2

P''

z1 / e

Fig. 6.1.1a Action of two successive infinitesimal transformations.

and
x
→ (1 + δα2 U2 )(1 + δα1 U1 )x,
= (1 + δα2 U2 + δα1 U1 + δα1 δα2 U2 U1 )x. (6.1.6a,b)
In Fig. 6.1.1a this transformation moves a point P with coordinates x =
(z1 , z2 ) to P
and then to P

with coordinates x
= (z1 + δz1 , z2 + δz2 ). Note
that (6.1.6c) contains all terms in the expansion of (6.1.6a) that are linear in
both δα1 and δα2 . In the same approximation, the inverse transformation is
T(−δα1 , −δα2 ) = (1 − δα1 U1 )(1 − δα2 U2 )
= (1 − δα2 U2 − δα1 U1 + δα1 δα2 U1 U2 ). (6.1.7)
In Fig. 6.1.1a, the inverse transformation carries P

x to P
x and then to P.
Expanding T(−δα1 , −δα2 )T(δα2 , δα1 ) one obtains
1 + O(δα21 , δα22 ). (6.1.8)
If one carries out the general transformation T(β2 , βi )T(α2 , α1 )x, the
resulting operator must be able to be expressed as
T(γ2 , γ1 ) = exp(γ2 U2 ) exp(γ1 U1 ), (6.1.9a)
with
γ1 = Ω1 (α2 , α1 , β2 , βi ), γ2 = Ω2 (α2 , α1 , β2 , βi ). (6.1.9b)
A fundamental property of Lie groups is exposed if, instead of following
the transformation (6.1.6) by its inverse, it is followed by
T(−δα2 , −δα1 ) → (1 − δα2 U2 )(1 − δα1 U1 )
= (1 − δα2 U2 − δα1 U1 + δα1 δα2 U2 U1 ). (6.1.10)
October 28, 2010 16:55 9in x 6in b943-ch06

172 Dynamical Symmetry

z2 / e

T1 P'

P''''
T2

T4 P''

T3
P'''

z1 / e

Fig. 6.1.1b Action of two successive infinitesimal transformations followed by

the reverse of their inverse.

Figure 6.1.1b illustrates effect of the resulting transformation

T(−δα2 , −δα1 )T(δα2 , δα1 )x

→ (1 − δα2 U2 )(1 − δα1 U1 )(1 + δα2 U2 )(1 + δα1 U1 )x. (6.1.11a)

Expanding the RHS of this, one finds that through terms first order in
δα1 and/or δα2 ,

T(−δα2 , −δα1 )T(δα2 , δα1 ) → 1 + δα1 δα2 (U2 U1 − U1 U2 ). (6.1.11b)

For this to be expressed as an approximation to exp(γ2 U2 ) exp(γ1 U1 ), the

commutator, [U2 , U1 ] = (U2 U1 − U1 U2 ), must be expressible as a linear
combination of U1 and U2 :

[U2 , U1 ] = c121 U1 + c221 U2 . (6.1.12)

Then (6.1.11) becomes an approximation to

exp(γ2 U2 ) exp(γ1 U1 ), (6.1.13)

in which

γ2 → δα1 δα2 c221 , γ1 → δα1 δα2 c121 .

The result obtained in this example is a special case of the following

October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 173

Theorem 6.1.1. The generators

Uκ = Σj ξjκ (x) ∂/∂xj , κ = 1, . . . , r, (6.1.14a)

of r one-parameter groups are the generators of an r-parameter Lie group

iff they obey the commutation relations

[Uκ , Uλ ] = Σµ Cµκλ Uµ , κ, λ, µ = 1, . . . , r, (6.1.14b)

in which the cµκλ are neither functions of the variables x nor of the group
parameters ω.

The generators satisfy a further condition

[[Uκ , Uλ ], Uµ ] + [[Uµ , Uκ ], Uλ ] + [[Uλ , Uµ ], Uκ ] = 0. (6.1.14c)

This condition, a Jacobi relation, is automatically satisfied because the Lie

generators U are first-derivative operators, and for this reason it is also
called the derivative condition.
Let us consider examples of operators that satisfy (6.1.14). In Euclidean
three-space, the generators of rotations may be taken to be

Rx = y ∂/∂z − z ∂/∂y, Ry = z ∂/∂x − x ∂/∂z, Rz = x ∂/∂y − y ∂/∂x.

(6.1.15a)
They satisfy the commutation relations

[Rx , Ry ] = −Rz , [Rz , Rx ] = −Ry , [Ry , Rz ] = −Rx . (6.1.15b)

The nonzero structure constants are

czxy = −1, cyzx = −1, cxyz = −1 (6.1.15c)

and

czyx = −czxy = 1, cyxz = −cyzx = 1, cxzy = −cxyz = 1.

We have previously seen that each of the generators R annihilates the func-
tion x2 +y2 +z2 , and that the operations of the corresponding one-parameter
groups carry out rotations and leave invariant the Euclidean metric in three-
space.
October 28, 2010 16:55 9in x 6in b943-ch06

174 Dynamical Symmetry

The generators of the group SO(2,1) which leaves invariant the value
x + y2 − z2 and the metric (dx2 + dy2 − dz2 ) are the Rz given above, and
2

Sx = y ∂/∂z + z ∂/∂y, Sy = z ∂/∂x + x ∂/∂z. (6.1.16a)

Their commutation relations are
[Sx , Sy ] = −Rz , [Rz , Sx ] = Sy , [Sy , Rz ] = Sx . (6.1.16b)
The nonzero structure constants are
czxy = −1, cyzx = 1, cxyz = 1 (6.1.16c)
and
czyx = −czxy = 1, cyxz = −cyzx = −1, cxzy = −cxyz = −1.
Note how these differ from the commutation relations obeyed by the gen-
erators of the rotation group.
Lie used the results stated in (6.1.14) to determine all possible many-
parameter local Lie groups acting in spaces of one, two, and three variables.2
Consider, for example, the following set of generators of one-parameter
transformation groups acting on a single variable, x:
∂/∂x, x ∂/∂x, x2 ∂/∂x, x3 ∂/∂x. (6.1.17)
By direct calculation,
[xa ∂/∂x, xb ∂/∂x]f(x) = (a − b)x(a+b−1) ∂/∂xf(x), (6.1.18a)
i.e.
[xa ∂/∂x, xb ∂/∂x] = (a − b)x(a+b−1) ∂/∂x. (6.1.18b)
Consequently
[x2 ∂/∂x, x3∂/∂x] = −x4 ∂/∂x, (6.1.18c)
and
[x ∂/∂x, x3 ∂/∂x] = −2x3 ∂/∂x. (6.1.18d)
It follows that x∂/∂x, x3 ∂/∂x are generators of a two-parameter group,
but that x2 ∂/∂x, x3 ∂/∂x are not. Relations (6.1.18) make it evident that
there is no r-parameter group which contains amongst its generators both
x2 ∂/∂x and x3 ∂/∂x; this is simply because their commutator yields x4 ∂/∂x
and this then gives rise to a commutator producing x5 ∂/∂x, ad infinitum.
However, the operators ∂/∂x, and xn ∂/∂x are generators of a two-parameter
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 175

group. Note also that the operators ∂/∂x, x∂/∂x, x2 ∂/∂x are generators of
a three-parameter group.
These results have an immediate consequence. Suppose that the mem-
bers of a set of generators are operators U(a) = ξa (x)∂/∂x whose ξa (x)
have power series expansions beginning with xa . The previous discussion
establishes that there is no r-parameter transformation group which con-
tains as generators both U(2) and U(3) , and so forth. Lie used extensions
from this observation to construct a wide variety of many-parameter groups
transforming many variables.3
In the previous examples, the commutator of two group generators is
itself a group generator, rather than the more general linear combination
of generators allowed by the theorem. However, it is easily shown that if a
set of r generators U is closed under commutation, then any set of r gener-
ators U
obtained by taking independent linear combinations of U, is also
closed under commutation. As a consequence, a set of linearly independent
linear combinations of generators of a group may themselves be considered
generators of a group.

6.2 The Differential Equations That Define

Many-Parameter Groups
In Chapter 4 it was shown that a one-parameter transformation group
acting on n variables xi is defined by a set of n first-order ODEs,

∂xi /∂α = ξi (x), i = 1, 2, . . . , n, (6.2.1a)

and that the ξi (x) define a group generator

U = Σi ξi (x)∂/∂xi . (6.2.1b)

At first glance, one might suspect that an analogous set

∂xi /∂αj = ξij (x), i = 1, 2, . . . , n, j = 1, 2, . . . (6.2.1c)

would define a many-parameter group with generators

Uj = Σi ξij (x) ∂/∂xi. (6.2.1d)

However, these equations are not sufficiently restrictive; they impose no

connection between the one-parameter groups defined for each value of j.
Many-parameter groups can be defined by coupled first-order equations

∂xi /∂αj = ζi,j (x, α1 , α2 , . . .) (6.2.2a)

October 28, 2010 16:55 9in x 6in b943-ch06

176 Dynamical Symmetry

As shown in the appendix, for such equations to define a many-parameter

group they must be further restricted. First of all, one must have
ζi,j (x, ω) = Σk uik (x)λkj (ω), (6.2.2b)
so that the defining differential equations are of the form
∂xi /∂αj = Σk uik (x)λkj (ω). (6.2.2c)
For this set of partial differential equations to be integrable, one must have
∂ 2 xi /∂αj ∂αm = ∂ 2 xi /∂αm ∂αj . (6.2.2d)
This requires that
Σk ∂(uik (x)λkj (ω))/∂αm = Σk ∂(uik (x) λkm (ω))/∂αj . (6.2.2e)
These requirements are certainly satisfied if the ζi,j (x, ω) are analytic func-
tions of their arguments. Using them Lie established his First Fundamental
Theorem4 :

Theorem 6.2.1. If the uik (x) are linearly independent analytic functions,
the equations
∂xi /∂αj = Σk uik (x)λkj (ω), i = 1, . . . , n, j, k = 1, . . . , r, (6.2.3a)
and
Σk ∂(uik (x) λkj (ω))/∂αm = Σk ∂(uik (x) λkm (ω))/∂αj , (6.2.3b)
determine an r-parameter group whose generators are
Uk = Σj ujk (x) ∂/∂xj. (6.2.3c)
The transformations of the group xj → x
j
are defined by = Φj (x; α) x
j
in which the Φj (x; α) are analytic functions of their arguments in regions
surrounding (x; 0).

We have previously dealt with a case in which the region does not include
all of the space: the transformation x
= x/(1 − αx) which is not defined
when αx = 1. Its generator is U = u(x) ∂/∂x, where u(x) is the analytic
function x2 .
Because the functions ξ(x) defining the generators of Lie groups are
assumed to be analytic functions they may always be approximated by
power series. The commutation relations of the generators may be deter-
mined from a knowledge of terms in the series that are of low order. This
property of Lie transformation groups is, as expected, a great help in applied
work.5
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 177

The operator of an r-parameter Lie group with generators Uκ can be

put in the form

S(α1 , αr ) = exp(Σκ ακ Uκ ) (6.2.4a)

as well in the form

T(α1 , αr ) = exp(α1 U1 ) exp(α2 U2 ) · · · exp(αr Ur )

= Πκ exp(ακ Uκ ), (6.2.4b)

not to mention a variety of other forms. The first form is particularly useful
when we deal with minuscule values of the parameters.
If the generators of a group do not commute, different orderings of the
operations in (6.2.4b) will have different actions. A number of formulae,
known as Baker–Campbell–Hausdorff relations, use commutation relations
among the group generators to interrelate these.6−8 Commutation relations
of the group generators may also, in some cases, be used to express the oper-
ators S(α1 , αr ), or T(α1 , αr ), as products of one-parameter group operators
exp(ακ Uκ ), in which not all the Uκ are different. Operators of the rotation
group are, for example, commonly put in the form

T(γ, β, α) = exp(γU3 ) exp(βU2 ) exp(αU3 ), (6.2.5)

where the γ, β, α are Euler’s angles of rotation.9

Group operators of the form exp(Σκ ακ Uκ ) avoid complications pro-
duced by different orderings in the form Πκ exp(ακ Uκ ).

6.3 Real Lie Algebras

In this section we investigate properties of group generators that make it
possible to simplify and systematize much of the study of Lie groups that
is important in physical applications.
We begin with a set of generators Uκ that satisfy the condition

[Uκ , Uλ ] = Σµ Cµκλ Uµ , κ, λ, µ = 1, . . . , r, (6.3.1a)

and the Jacobi relation

[[Uκ , Uλ ], Uµ ] + [[Uµ , Uκ ], Uλ ] + [[Uλ , Uµ ], Uκ ] = 0. (6.3.1b)

As before, the cµλκ are neither functions of the variables x, nor of the group
parameters ω. The r different U’s are required to be linearly independent,
October 28, 2010 16:55 9in x 6in b943-ch06

178 Dynamical Symmetry

that is, they do not satisfy any condition of the form

c1 U1 + c2 U2 + · · · + cr Ur = 0. (6.3.1c)
One may imagine the Uκ to be represented by basis vectors uκ of a linear
algebra with, e.g.
U1 ∼ u1 (1, 0, 0, . . . ), U2 ∼ u2 = (0, 1, 0, . . . ), etc. (6.3.2a)
The general element of the algebra would then be U = Σaκ uκ ∼ Σaκ Uκ ,
with the aκ being the scalars of the algebra. Commutation is the compositon
operator of the algebra. The scalars cµκλ are restricted by the requirement
cµλκ = −cµκλ, (6.3.2b)
which is a direct consequence of the fact that
[Uλ , Uκ ] = −[Uκ , Uλ ]. (6.3.2c)
The reader is invited to show that the Jacobi relation implies that the cµκλ
obey the further restrictions
Σµ (cµαβ cνµλ + cµβλ cνµα + cµλα cνµβ ) = 0. (6.3.2d)
Matrix multiplication with a square matrix M will convert a vector
U = Σaκ uκ to
U
= MU, (6.3.3)
which will be a vector of the Lie algebra if the elements mij of M are scalars.
The vector U
will represent a linearly independent set of generators U
κ
if the determinant of M is nonzero. It is then easy to show that (6.3.1a)
implies
[U
κ , U
λ ] = Σµ (cµκλ )
U
µ , κ, λ, µ = 1, . . . , r. (6.3.4)
The structure constants are not, in general, identical to those in (6.3.1a).
The U’s satisfy all the laws of a linear algebra — if one considers the
0 vector to be a U as well; henceforth it will be denoted U0 . We shall use
bold face type to denote Lie generators when we wish to emphasize their
linear-algebraic, vectorial, properties.
A linear algebra whose elements satisfy (6.3.1) and (6.3.2) is said to be
a Lie algebra. It is a real Lie algebra if no element of the algebra is allowed
to take on complex values. The cµκλ are termed the structure constants of
the Lie algebra.
The rank of a Lie algebra is the maximum number of its generators that
commute amongst themselves. If no generator other than U0 commute with
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 179

another, the rank is 1. A Lie algebra, all generators of which commute with
one another, is termed Abelian. All the structure constants of an Abelian
Lie algebra are zero.
Lie algebras with identical structure constants are said to be isomorphic.
The freedom represented by the transformation from (6.3.1a) to (6.3.4)
indicates that two Lie algebras of the same dimension, r, may be isomorphic
though their structure constants are different. A method for determining
whether Lie algebras are isomorphic, devised by Killing, is discussed in
Sec. 6.6. It is extensively developed by Cartan and Dynkin and used to
classify Lie algebras. The methodology is outlined in the Sec. 6.7, and the
classification of Lie algebras and groups is outlined in the appendix.

6.4 Relations between Commutation Relations and the

Action of Transformation Groups: Some Examples
We begin this section with a brief study of the effect of transformations of
variables upon the action of a Lie transformation group. Suppose one has
a set of generators

Uk = Σj, ξj,k (z) ∂/∂zj , k = 1, . . . , r, (6.4.1)

in which the ξj,k are Ck functions. If one subjects the Uk to a Ck diffeo-

morphism

zj → z
j = fj (z), z
j → zj = fjinv (z
), (6.4.2a,b)

then

Uk → U
k = Σj ζj,k


(z
) ∂/∂z
j . (6.4.3)


In this, as usual, the ζj,k (z
) are obtained by using the inverse transforma-
tion, (6.4.2b), to replace z by z
in Uk z
j = Uk fj (z). Now let us consider the
transformation from z to z
passively, that is let us consider it to define
a change from one local coordinate system z to another z
. Such a pas-
sive change in coordinates affects only the coordinate labels in Fig. 6.4.1.
Consequently, one expects that

if [Uκ , Uλ ] = Σµ Cµκλ Uµ , then [U
κ , U
λ ] = Σµ Cµκλ U
µ , (6.4.4)

then and the structure constants remaining unchanged. In short, diffeomor-

phisms do not alter commutation relations.
October 28, 2010 16:55 9in x 6in b943-ch06

180 Dynamical Symmetry

y/ e

2.0
P'

1.5

P''''

1.0
P''
P'''

y' / e
0.5

x/ e
0.0
0.5 1.0 1.5 2.0
x' / e

Fig. 6.4.1 Commutation diagram with second local coordinate.

This fact profoundly influences the physical interpretation of symmetry

groups. Consider, for example, the commutation relations
[Rx , Ry ] = −Rz , [Rz , Rx ] = −Ry , [Ry , Rz ] = −Rx (6.4.5)
which are satisfied by the generators (6.1.15a) of the rotation group in
Euclidean three-space. Let
x → x/a, y → y/b, z → z/c. (6.4.6)
This transformation converts spheres defined by
x2 + y2 + z2 = r2 (6.4.7)
to ellipsoids defined by
(x/a)2 + (y/b)2 + (z/c)2 = r2 . (6.4.8)
It converts the generators R to generators such as
R
x = (y/b) ∂/∂(z/c) − (z/c) ∂/∂(y/b) = (c/b)y∂/∂z − (b/c)z∂/∂y.
(6.4.9)
The reader can verify that if a, b, c are all nonzero, the commutation rela-
tions (6.4.5) are also satisfied by the R
. The R
are generators of a group
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 181

whose operator carries points on each ellipsoid into points on the ellipsoid,
in exactly the same manner that the operator generated by the R carries
points on a sphere into points on a sphere.
Lie transformation groups generated by two different sets of gener-
ators with isomorphic commutation relations are said to be locally iso-
morphic because, for some (finite) range of the group parameters, the
actions T(α1 , αr )x and T
(α1 , αr )y of the two groups can be put in one-
to-one correspondence. Locally isomorphic groups are globally isomorphic
if the one-to-one correspondence persists for all allowed values of the group
parameters. SO(3) and SU(2) are locally isomorphic, but not globally iso-
morphic. Different realizations of a group are produced when different group
generators produce globally isomorphic groups. The R and the R
of the
previous paragraph are generators of different realizations of SO(3). Both
SO(3) and SU(2) groups appear in a variety of realizations in physics and
chemistry, some quite surprising.
All group parameters are allowed, in principle, to range through all the
reals, but, as is the case for one-parameter groups, it may happen that, for
any one-parameter subgroup, a group parameter and the same parameter
mod some number, e.g. 2π or 4π, will give rise to the same transformation.
One may, for convenience, then restrict the range of the group parameter so
that the effect of the group operation is not repeated. The group parameter,
say α, can then be contained within a compact region, e.g. −π ≤ α ≤ π.
Such a one-parameter group is termed a compact one-parameter group.
A one-parameter group that is not compact is termed noncompact. The
group parameter of a noncompact group ranges over an interval which may
or may not contain all the real numbers, examples being −∞ <α <∞ and
−π <α <π. The group SO(2,1) contains a compact one-parameter subgroup,
and two noncompact one-parameter subgroups. In the realization given
above, exp(αRz ) is the operator of a compact subgroup whose operation is
the same for α = π as it is for α = −π. On the other hand, the operators
exp(βSx ) and exp(γSz ) do not repeat their action as their group parameters
range through all the real numbers. The group SO(3) contains only compact
one-parameter subgroups.
If all one-parameter groups contained in a many-parameter group
are compact, the many-parameter group is termed compact. If a many-
parameter group contains one or more noncompact one-parameter groups,
the many-parameter group is noncompact. The groups SO(n) are, for all
values of n, compact. The groups SO(p, q) are noncompact for nonzero
p and q.
October 28, 2010 16:55 9in x 6in b943-ch06

182 Dynamical Symmetry

The structure constants of the Lie algebra of SO(3) are different from
the structure constants of the Lie algebra of SO(2,1). This is an exam-
ple of a general truth: the commutation relations of a real Lie algebra
determine whether the corresponding Lie group is compact or noncom-
pact. Thus, rather surprisingly, the compactness or noncompactness of a
Lie group is determined by relationships among its infinitesimal trans-
formations. Subtleties may however be involved, as the following example
demonstrates.
The projective group in one variable, say t, has generators
∂/∂t, t∂/∂t, t2 ∂/∂t. (6.4.10)
The general transformation of the group can be given the form
t → t
= (t + a1 )/(a2 t + a3 ). (6.4.11)
The identity transformation occurs when α1 = −a1 /a3 , α2 = −a2 /a3 , and
α3 = 1/a3 . The linear combinations of the generators (6.4.10),
U1 = (1/2)(1 − t2 ) ∂/∂t, U3 = (1/2)(1 + t2 ) ∂/∂t, (6.4.12)
together with U2 = t∂/∂t, obey the commutation relations
[U1 , U2 ] = U3 , [U2 , U3 ] = −U1 , [U3 , U1 ] = −U2 . (6.4.13)
The group is therefore locally isomorphic to SO(2,1), and the generator of
the compact subgroup is U3 . As the real line −∞ < t < ∞ is an open
interval it is rather surprising to find that a group that carries it into itself
contains a compact subgroup, which, as we shall see, can carry any point
on the line to any other point on the line.
The transformation (6.4.11), like that generated by t2 ∂/∂t, undergoes
a sudden change in sign from ±∞ to ±(−∞) when a2 t + a3 passes through
zero. In Chapter 2 we saw that a stereographic projection from the plane
onto the sphere projects onto the South pole of the sphere all points whose
(x, y) coordinates in the plane are ±∞. Let us consider the corresponding
projection of the line of t onto a unit circle. Set
x = 2t/(t2 + 1), y = (1 − t2 )/(t2 + 1). (6.4.14)
Then x2 + y2 = 1, and one may set x = sin(θ), y = cos(θ). Consequently
t = tan(θ/2), dt = (dθ/2)sec2 (θ/2). (6.4.15)
Figure 6.4.2 depicts the relation between t and θ. For −∞ < t < ∞ one
has a diffeomorphism that maps t, dt to θ, dθ respectively with θ lying in
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 183

1.0

0.5
θ
t
−2 −1 1 2

− 0.5

− 1.0

Fig. 6.4.2 Stereographic projection of t from the real line onto a unit circle.

the open interval −π ≤ θ ≤ π. However θ also remains real over the closed
interval −π ≤ θ ≤ π with θ = −π representing the same point as θ = π.
The projection converts the Uj to the Vj defined below:

U1 → y(y∂/∂x − x∂/∂y) → cos(θ)∂/∂θ = V1 ,

U2 → −x(y∂/∂x − x∂/∂y) → − sin(θ)∂/∂θ = V2 , (6.4.16)
U3 → −(y∂/∂x − x∂/∂y) → −∂/∂θ = V3 .

V1 vanishes when θ = ±π/2, and V2 vanishes when θ = 0, ±π. This has

the consequence (c.f. the exercises) that the finite transformations generated
by V1 and V2 cannot carry θ out of the open interval −π < θ < π. How-
ever, V3 has no singular point, and exp(γV3 ) carries θ to θ − γ, which can
become ±π.
The V’s obey commutation relations isomorphic to those of the U’s,
but they obey them for all real values of θ, a range which includes values
of θ in the compact interval π ≤ θ ≤ π. This interval corresponds to the
interval −∞ ≤ t ≤ ∞, with −∞ being identified with +∞. (Extending the
interval by adding the point at θ = ±π is termed a one-point compacti-
fication of the interval). Commutation relations express relations between
infinitesimal transformations that move points over infinitesimal regions.
Crudely speaking, an infinitesimal interval contains a continuum of points,
and removing or adding a point is unnoticeable in so far as an infinitesi-
mal transformation is concerned. In this case, the point added at θ = ±π
corresponds to t = ±∞. As the commutation relations are insensitive to
such trickery, the three-parameter group generated by ∂/∂t, t∂/∂t, t2 ∂/∂t.
contains a compact subgroup, isomorphic to that in the group generated
by U1 , U2 , U3 .
October 28, 2010 16:55 9in x 6in b943-ch06

184 Dynamical Symmetry

The commutation relations that determine the Lie algebra of a com-

pact Lie group do not determine the length of the compact interval over
which the group parameters range before causing the action of the group
to repeat. This allows different Lie groups to arise from Lie algebras that
have the same structure constants. The groups SO(3) and SU(2) provide an
example.
The generators of the ordinary rotation group and those of the spin 1/2
rotation group have isomorphic commutation relations. For the ordinary
rotation group, the Euler angles in the rotation operator9

T(γ, β, α) = exp(γU3 ) exp(βU2 ) exp(αU3 ) (6.4.17)

range over the intervals 0 ≤ α < 2π, 0 ≤ β < π, 0 ≤ γ < 2π without

bringing a point back to itself. The group is a realization of the abstract
Lie group SO(3). However, if half-integer spin vectors are being rotated,
the action of exp(γU3 ) only repeats when the group parameter γ moves
over a range of 4π or greater. The standard range of the parameter is 0 ≤
γ < 4π. The spin 1/2 rotation group is a realization of the abstract group
SU(2), rather than the abstract group SO(3). SO(3) and SU(2) are locally
isomorphic, but not globally isomorphic Lie groups. In short, isomorphic
Lie algebras yield locally isomorphic Lie groups, but do not necessarily yield
globally isomorphic Lie groups.
We conclude this section with several examples of physical interest. The
linear symplectic group Sp(2R) has a realization with generators

U1 = (q ∂/∂q − p ∂/∂p)/2,
U2 = (q ∂/∂p + p ∂/∂q)/2, (6.4.18)
U3 = (p ∂/∂q − q ∂/∂p)/2.

The three operators exp(αj Uj ) carry out canonical transformations in the

phase space of (p, q). The generators satisfy the commutation relations

[U1 , U2 ] = −U3 , [U2 , U3 ] = U1 , [U3 , U1 ] = U2 (6.4.19)

Now refer back to Eqs. (6.1.16a) and set up the correspondences U1 ∼ Sx ,

U2 ∼ Sy , U3 ∼ Rz . On comparing the commutation relations (6.1.16b)
with (6.4.19) one sees that the structure constants of the two Lie alge-
bras are identical. The Lie groups are consequently locally isomorphic.
However

U3 = (p ∂/∂q − q ∂/∂p)/2, (6.4.20a)

October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 185

while
Rz = (x ∂/∂y − y ∂/∂x), (6.4.20b)
so
exp(αU3 ) = exp(α
U3 ), α
= α + 4n π, (6.4.20c)
while
exp(αRz ) = exp(α
Rz ), α
= α + 2n π, (6.4.20d)
where n is an integer. Thus the groups are not globally isomorphic.
It is usually assumed that physical time runs continuously from an indef-
inite past through the present to an indefinite future: −∞ < t < ∞, or
even −∞ ≤ t ≤ ∞. One expects that when exp(α ∂/∂t) acts on t to give
t
= t + α, one must allow α the full range −∞ < α < ∞. However, even
in Newtonian mechanics, if one is not careful this view can lead to incon-
sistencies. In Chapter 5 we found eight generators of one-parameter groups
that leave invariant Newton’s equation of motion for a harmonic oscillator.
The commutation relations of three of these generators,
U1 = ∂/∂t, (6.4.21a)
U2 = (1 + y2 ) cos(t) ∂/∂y + y sin(t) ∂/∂t, (6.4.21b)

U3 = (1 + y2 ) sin(t) ∂/∂y − y cos(t) ∂/∂t, (6.4.21c)

are
[U3 , U1 ] = −U2 , [U2 , U3 ] = −U1 , [U1 , U2 ] = −U3 . (6.4.21d)
These are the commutation relations of the compact groups SO(3) and
SU(2). Each solution x = A sin(t) + B cos(t) is converted into itself by the
transformation t → t
= t + 2π. Thus, ∂/∂t is the generator of a compact
group, which is in fact, SO(3).10 In so far as the oscillator is concerned, t
and t + 2π are indistinguishable.
In quantum mechanics one deals with the operations of matrices M
that act on the coefficients cj of a linear combination of wave functions,
cj ψj (x), where the x are electronic coordinates. The matrix elements mjk
may be functions of parameters, R, such as nuclear coordinates. One may,
for example, be dealing with an eigenvalue problem Hop ψ = Eψ, ψ =
Σj cj ψj (x). If H is a matrix whose elements, Hij (R) =< ψi (x)|Hop |ψj (x) >,
are functions of a set of parameters R, one obtains an eigenvalue equation
of the form
(H(R) − E)C = 0, C = (c1 , c2 , cn ). (6.4.22)
October 28, 2010 16:55 9in x 6in b943-ch06

186 Dynamical Symmetry

As one varies the parameters, the coefficients cj vary. It is a general property

of such linear equations, that E may be left invariant by transformations of
the general real linear group Gl(n, R) of real transformations, if the cj and
the matrix elements Hjk (R) are real variables. This group contains both
SO(n) and SU(n) subgroups. If complex variables are involved the group
is GL(n, C), which also has SO(n) and SU(n)subgroups. In both cases the
SU(n) subgroups have operators that vary some cj ’s at half the rate the
Hjk (R) are varied.11 This has the consequence that varying the R through
a cycle that returns the matrix elements to their initial values can cause
the c’s to return to the negative of their initial values! Aharonov and Bohm
were the first to notice this phenomenon in a physical context.12 In chem-
ical systems with degenerate molecular wavefunctions, similar phase shifts
are found to be produced by cyclic nuclear motions. These can produce
surprising effects when potential energy curves or surfaces E(R) and E
(R)
become degenerate at crossings or intersections.13 Berry14 discovered that
similar, but more varied, phase shifts occur in systems governed by the
time-dependent analog of (6.4.22),
(H(R) − i ∂/∂t)C = 0. (6.4.23)

6.5 Transitivity
A transformation group acts transitively in a space or on a manifold if every
point of the space or manifold can be carried into every other point. A group
that fails this criterion, acts intransitively in the space or on the manifold.
Transitivity is a property that is not determined by commutation relations
alone, it depends upon the realization of the group, the functional form of
the ξj,k (z) in the group generators Uk = Σj ξj,k (z) ∂/∂zj . The translation
subgroup of the Euclidean group E(3) in three-dimensional Euclidean space
acts transitively in the space because it can move any point in the space
to any other point in the space. The SO(3) group that carries out rota-
tions about the origin acts transitively on the surface of a sphere centered
at the origin. However this group acts intransitively in three-dimensional
Euclidean space because it cannot move points from one sphere centered at
the origin to points on a concentric sphere of different radius. This exempli-
fies a general observation: a group that acts transitively in a space S cannot
simply move points about in a subspace of S. If S is a space whose points
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 187

are determined by assigning values to n independent variables, xj , a group

that acts transitively in the space is consequently able to change the value
of any function f(x1 , x2 , . . . , xn ) defined at each point P in the space. Such
a group cannot leave invariant any function or manifold defined within S by
such functions. It can leave metrics and coordinate differences invariant. As
an example, the group of translations x → x + a, y → y + b, z → z + c acts
transitively in the space of (x, y, z), and it leaves invariant the Euclidean
distance between two points P1 and P2 or any function depending only
upon x1 − x2 , y1 − y2 , z1 − z2 .
The most useful dynamical groups and Lie algebras contain operations
that act transitively on the space of solutions of dynamical equations. They
often act intransitively on the space of variables in which the solutions are
defined. Thus, for example, solutions of Hamilton’s equations are defined in
the phase space of the variables p, q, E and t. Transformations that carry
solutions of the equations into solutions of the same energy leave invariant
the expression H(p, q) − E. This expression defines a submanifold in the
space of p, q, E and t. In Chapters 7–9 many-parameter groups that trans-
form solutions of ODEs into solutions will be used to uncover many proper-
ties of the equations of classical dynamics. This will provide the foundation
for our subsequent investigations of Schrödinger’s and Maxwell’s equations.
The following sections of this chapter are primarily devoted to providing
mathematical tools that help uncover consequences of group structure.

6.6 Complex Lie Algebras

In quantum mechanics one deals with complex extensions of real Lie alge-
bras. The generators of the complex extension of a real Lie algebra are
allowed to be functions of complex variables. In the linear combinations
Σ Cκ Uk , the cκ are allowed to be complex numbers, as are structure con-
stants. In the Lie groups generated by the operators of a complex Lie alge-
bra, the group parameters are also allowed to be complex variables.
As a first example of the consequences of the complexification of a Lie
algebra, consider the effect on the rotation generators, (6.1.15a), by replac-
ing z with i t. The operators then become
R
x = −i(y∂/∂t+t∂/∂y), R
y = i(t∂/∂x+x ∂/∂t), Rz = x ∂/∂y−y ∂/∂x.
(6.6.1)
October 28, 2010 16:55 9in x 6in b943-ch06

188 Dynamical Symmetry

Note the similarity to the SO(2,1) generators of (6.1.16), viz

Sx = y ∂/∂z + z ∂/∂y, Rz = x ∂/∂y − y ∂/∂x.

Sy = z ∂/∂x + x ∂/∂z,
(6.6.2)
The operators of (6.6.1) generate one-parameter groups which leave invari-
ant x2 + y2 − t2 ; the operators of (6.6.2) generate one-parameter groups
which leave x2 + y2 − z2 invariant.
The key commutation relations among the generators in (6.6.1) are

[R
x , R
y ] = −Rz , [Rz , R
x ] = −R
y , [R
y , Rz ] = −R
x . (6.6.3)

These are isomorphic to the commutation relations obeyed by the original

rotation generators, and all the structure constants associated with (6.1.15)
are the same as those associated with (6.6.3). They are not the same as
the structure constants associated with the SO(2,1) generators in (6.1.16).
When complexification is allowed, the commutation relations of Lie algebra
of SO(2,1) and that of SO(3) can be made isomorphic, to have the same
structure constants. The same may be said of the commutation relations
of SO(p, q) and SO(p + q). The corresponding inhomogenous groups are
similarly affected.

6.7 The Cartan–Killing Form; Labeling and Shift Operators

Because both complexification and changing the basis of a Lie algebra can
change its commutation relations, it is helpful to have a way of determining
properties of the algebra that are basis independent. A tool that does this
was invented by Killing.15 The Cartan–Killing form is a matrix (and second
rank tensor) with elements

gjk = gkj = Σµ Σν Cµjν cνkµ . (6.7.1a)

Using the Einstein summation convention this is

gjk = gkj = cµjν cνkµ . (6.7.1b)

Cartan16 proved that if the determinant of this matrix is nonzero, the

algebra contains operators that satisfy commutation relations of the form

[U0 , U+ ] = k U+ , [U0 , U− ] = −k U− , (6.7.2)

where k is a constant. If the labeling operator U0 has eigenfunctions, the

eigenvalues of the labeling operator may be used to label the eigenfunctions.
The shift operators then shift this eigenvalue by ±k.
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 189

As a first example of these observations, consider the Lie algebra of the

groups SO(3) and SU(2). We have seen that their commutation relations
may be chosen to be

[Rx , Ry ] = −Rz , [Rz , Rx ] = −Ry , [Ry , Rz ] = −Rx . (6.7.3a)

Renaming Rx , Ry , Rz as U1 , U2 , U3 , respectively, the nonzero structure

constants are

c312 = −1, c231 = −1, c123 = −1, (6.7.3b)

c321 = −c312 = 1, c213 = −c231 = 1, c132 = −c123 = 1.

and Consequently,

g11 = cµ1ν cν1µ = c312 c213 + c213 c312 = (−1)(1) + (1)(−1) = −2

g12 = cµ1ν cν2µ = c213 c322 + c312 c223 = 0, (6.7.4)
g22 = cµ2ν cν2µ = c321 c123 + c123 c321 = −2.

Similarly, one finds that g33 = −2, g13 = g23 = 0. The Killing form is thus




−2 0 0





0 −2 0
. (6.7.5)




0 0 −2

The corresponding determinant has value −8, so the three generators U

may be used to construct a labeling operator and a pair of shift operators.
The inhomogeneous group ISO(3) has in addition to the rotation gen-
erators U1 , U2 , U3 , the translation generators U4 = ∂x, U5 = ∂y, U6 = ∂z,
which commute among themselves. Their commutation relations with the
rotation generators give rise to the nonzero structure constants:

c624 = 1 = −c642 , c534 = −1 = −c543 , c615 = −1 = −c651 ,

(6.7.6)
c435 = 1 = −c453 , c516 = 1 = −c561 , c426 = −1 = −c462 .

Using these one finds that the only nonzero elements of the 6 × 6 dimen-
sional Killing form for ISO(3) are those of its SO(3) subgroup, the diagonal
elements in (6.7.5). The determinant of the form is thus zero.
The class of groups termed semi-simple all have Killing forms whose
determinant is nonzero. This class, defined in Appendix B, includes the
groups SO(p,q), SO(p + q), and groups locally isomorphic to them, but not
ISO(p, q), ISO(p + q).
October 28, 2010 16:55 9in x 6in b943-ch06

190 Dynamical Symmetry

The structure constants of the Lie algebra of the angular momentum

operators Lj are −ih/2π times those of the generators Rj of the rotation
group SO(3). The determinant of the Killing form is consequently 8(h/2π)3 .
If we let Lz be the labeling operator, the shift operators are
L+ = Lx + iLy = −i(Rx + iRy ),
(6.7.7a,b)
L− = Lx − iLy = −i(Rx − iRy ),
and
[Lz , L+ ] = L+ , [Lz , L− ] = −L− . (6.7.7c,d)
It follows that
Lz L+ = L+ (Lz + 1), Lz L− = L− (Lz − 1). (6.7.7e,f)
The spherical harmonics Ylm (θ, φ) satisfy
Lz Ylm = m(h/2π)Ylm , (6.7.8)
with l a nonnegative integer, and m an integer. Equations (6.6.7c,d) imply
that
Lz (L+ Ym ) = (m + 1)(h/2π)(L+ Ylm ), (6.7.9)
Lz (L− Ylm ) = (m − 1)(h/2π)(L− Ylm ).
Consequently, if L+ Ylm does not vanish, it has eigenvalue (m +1)(h/2π); on
the other hand, if L− Ylm does not vanish, it has eigenvalue (m −1)(h/2π). If
the Ylm are those defined by Condon and Shortley,17 an elegant argument18
establishes
L+ Ylm = ν+lm Ylm+1 = (l (l + 1) − m(m + 1))1/2 (h/2π)Ylm+1 , (6.7.10a)

L− Ylm = ν−lm Ylm−1 = (l (l + 1) − m(m − 1))1/2 (h/2π)Ylm−1 . (6.7.10b)

These relations imply that

L+ Yl,l = 0, L− Yl,l = 0. (6.7.10c)
As a consequence, m can only take on the 2l + 1 integer values that begin
with −l and run through to +l .
In the general case, the allowed range of the eigenvalues of a labeling
operator may be finite or infinite, and the eigenvalues need not be integers.
The generators of the group SU(2) obey the same commutation relations
as those of SO(3), but in contrast to SO(3) the labeling operator of SU(2)
may have half-integer, as well as integer eigenvalues.
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 191

6.8 Casimir Operators

Though no generator of a group may commute with all generators of the
group, there may be polynomial functions of the generators that commute
with all of them. Since the importance of this fact was first recognized by
the Dutch physicist Casimir,19 these polynomial functions of generators are
called Casimir operators. The best example known is the Casimir operator
of SO(3), the total angular momentum operator, L2 , of quantum mechanics,
whose eigenvalues are l (l + 1)(h/2π).2
The rotation group in four-space has a quadratic Casimir operator and
a quartic one. Let the six generators of the group be U12 , U23 , U31 , U14 ,
U24 , and U34 with
Uκλ = xκ ∂/∂xλ − xλ ∂/∂xκ . (6.8.1a)
Define
R = (U23 , U31 , U12 ) = (Rx , Ry , Rz ), (6.8.1b)
A = (U14 , U24 , U34 ) = (Ax , Ay , Az ). (6.8.1c)
The Casimir operators may then be taken to be
C2 = −(A · A + R·R), C4 = (−A·R)2 . (6.8.2)
These Casimir operators of SO(4) play prominent roles in later chapters.
If one lets the generators of the Lorentz group be components of R and
T = (V14 , V24 , V34 ) = (Tx , Ty , Tz ),
Vκλ = xκ ∂/∂xλ +xλ ∂/∂xκ, x4 = ct,
(6.8.3)
then the quadratic and quartic Casimir operators are
C2 = (T·T − R·R), C4 = (−T·R)2 . (6.8.4)
A semi-simple group of rank r possesses r independent Casimir operators.20
The spherical harmonics are perhaps the most used among the special
functions of mathematical physics. If a Lie group, or its subgroups, has
labeling and shift operators as well as Casimir operators, these operators
may be used to greatly simplify investigations of any function that can
be made an eigenfunction of the labeling and Casimir operators. Sharp21
and Wigner22 established that the theory of a wide variety of special
functions of mathematical physics can be simplifed in this manner. The
economy of thought that results is quite marvelous. In this connection
the reader is invited to compare Watson’s Theory of Bessel Functions23
with Sharp’s discussion of Bessel functions.21 Vilenkin’s monograph on the
October 28, 2010 16:55 9in x 6in b943-ch06

192 Dynamical Symmetry

special functions24 approaches the subject from a similar group theoretic

standpoint.
The generators of a Lie group need not be first-order differential opera-
tors. They may be differential operators of order greater than one, opera-
tors of integration, matrices, etc. All continuous groups whose composition
functions satisfy Definition 6.1 are Lie groups.

6.9 Groups That Vary the Parameters

of Transformation Groups
The part of the theory of Lie groups that deals only with the action of the
operators upon themselves is known as the theory of abstract Lie groups.
During these actions of the abstract group only the group parameters α
vary. Operators of these abstract groups may be considered to act on the
group parameters. If α1 , α2 are group parameters, and b is not contained in
the space of the α’s, then exp(b(α1 ∂/∂α2 − α2 ∂/∂α1 )) is a such operator.
Many properties of a transformation group depend only upon properties
of its parameter space, the space the parameters sweep through as they
vary. There are several different ways of defining such groups.25 The theory
of topological groups is based on an analysis of parameter groups.26 If a
group has no generators which commute with all its generators, the adjoint
group can be specially revealing. Adjoint groups also play a central role in
the classification of Lie algebras. The generators of the adjoint group are
defined to be
Γs = Σ Σ ajk αj ∂/∂αk , ajk = ckjs . (6.9.1)
As an example, consider the transformations of the rotation group SO(3)
with operator of the form
exp(α3 R3 ) exp(α2 R2 ) exp(α1 R1 ) (6.9.2)
the R’s being those defined in Eqs. (6.7.3). Using the structure constants
in (6.7.3) to evaluate (6.9.1), one finds that the generators of the adjoint
group are given by
Γ1 = α2 ∂/∂α3 − α3 ∂/∂α2 , Γ2 = α3 ∂/∂α1 − α1 ∂/∂α3 ,
Γ3 = α1 ∂/∂α2 − α2 ∂/∂α1 . (6.9.3)
Though the adjoint group of the rotation group acts in a space of three
dimensions, in most cases adjoint groups act in spaces of higher dimen-
sions than does the group upon which they are based. Rotations in a four-
dimensional space with coordinates (x1 , x2 , x3 , x4 ) lead to a six-dimensional
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 193

space of group parameters: the group SO(4) is a six-parameter group. The

adjoint group of SO(4) has the generators (6.9.3), and three additional
generators:
Γ4 = α5 ∂/∂α6 − α6 ∂/∂α5 ,
Γ5 = α6 ∂/∂α4 − α4 ∂/∂α6 , (6.9.4)
Γ6 = α4 ∂/∂α5 − α5 ∂/∂α4 .
The SO(3,1) Lorentz group of spacetime also has a six-dimensional param-
eter space and an adjoint group with six generators. The generators of the
Poincare group, an ISO(3) group, are those of the Lorentz group supple-
mented with the generators of four translations in spacetime. The parameter
space of the adjoint group is therefore ten-dimensional.

6.10 Lie Symmetries Induced from Observations

We have seen that differential equations define Lie groups of one or more
parameters, that Lie groups interconvert solutions of the differential equa-
tions of the natural and physical sciences, and that Lie groups define intrin-
sic symmetries of these differential equations. Relationships uncovered by
Lie theory are found to be expressed in corresponding relationships between
physical observations. Intrinsic symmetries thereby become symmetries in
the natural world. These symmetries may not even be possible in three-
dimensional Euclidean space. Their geometric interpretation involves fur-
ther considerations, and is not necessarily implied by a differential equation
itself or its physical interpretation.
What can be said about the reverse process of inducing mathematical
connections from observations which establish group theoretic connections
between observables? One must face the fact that the operations of a Lie
transformation group, for example, SO(3), have an infinite number of real-
izations as differential operators. The space of continuous variables upon
which they act is far from fully determined by the knowledge that observa-
tions are connected by operations of the group.
The operations of a Lie group may be operations of the group upon a
variety of variables, or operations on a variety of differential (and integral)
operators that operate on such variables. The operations of a Lie group can
also be represented by matrices acting on vectors — vectors that may be
conceived of in a wide variety of ways. And, as we have seen, associated
with each r-parameter Lie transformation group, T(ω)x are groups whose
operators, like those of the adjoint group, act on the group parameters
October 28, 2010 16:55 9in x 6in b943-ch06

194 Dynamical Symmetry

of the transformation group. The dimension of the parameter space of an

r-parameter group is r, and r is commonly greater than the dimension of
the physical space in which the transformation group is conceived to act.
Thus, e.g. the Poincare group acts on the four variables of spacetime, or the
eight variables of the corresponding phase space, but its parameter group
is ten-dimensional.
Russell once, and famously, said, “Mathematicians do not know what
they are talking about, nor whether what they are saying is true.”27 As
coauthor with Whitehead of Principia Mathematica,28 Russell presumably
knew what he was talking about, but was not explicitly talking about the
theory of Lie groups, so Eddington added the following, “and a group the-
oretician is a supermathematician who does not know what he is doing.”29
Natural scientists are supposed to know what they are doing — but pre-
suppositions, sometimes unrecognized, occur beneath every theory. Newton
supposed that mass is an intrinsic property of the object, independent of
its velocity or acceleration. He presupposed that physical space was con-
tinuous as well as three-dimensional. And he presupposed that time flowed
continuously from a distant past, through the present, and into the future.
These presuppositions allowed him to define arbitrarily small changes in
spacetime, and instantaneous velocities dx/dt, accelerations d(dx/dt)/dt,
momenta p = mdx/dt, and their rate of change dp/dt. Observers equipped
with clocks and measuring sticks were able to approximately measure each
of these and give physical meaning. Classical mechanics developed as a the-
ory based on the physical existence of continuous chains of causes f(x, t)dt,
and effects dq, dp, evolving in spacetime. Several centuries after Newton,
Noether realized that these presuppositions implied the conservation laws
mentioned in Chapter 1, and their connection to Lie symmetries became
evident.
Measurement errors always swamp any attempt to measure arbitrar-
ily small changes such as dxj and dt, so the continuity of space and time
remains a hypothesis. And even in classical mechanics, simple, even separa-
ble, systems can smoothly evolve into unstable states, whose further evolu-
tion can be greatly altered by small perturbations. A tossed claw hammer,
or any asymmetric top rotating about its intermediate axis, provides an
intriguing example of such behavior. It is now well known that the classical
equations of motion for systems of three or more bodies can develop a vari-
ety of instabilities, some of which lead to effectively unpredictable motions.
Electromagnetic theory, relativity theory, and quantum theory, all retain
the presupposition of continuous spacetime: the dynamical equations of
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 195

all are differential equations involving operators such as ∂/∂xj, and ∂/∂t.
However, in quantum theory the continuous evolution of causal chains in a
continuous spacetime is replaced by the conception of the continuous space-
time evolution of wave functions, ψ, ψ∗ , which determine the probabilities
that an observable will be found to have a given eigenvalue.
There are times in the evolution of any science when presuppositions
are challenged for good reason, and conjecture is required. After Gell-
Mann proposed SU(3) as a symmetry group interrelating the properties
of baryons, there was a period when theoretical physicists conjectured that
properties of the (infinitely) stable elementary particles, the short lived
elementary particles observed in high-energy collisions, as well as the prop-
erties assumed for quarks are the physical expressions of structure found
in a great variety of groups. It is however one thing to find group struc-
ture, and another to find the underlying space of variables upon which the
group acts, or to distinguish transformation groups from their parameter
groups, which as we have seen often have higher dimensionality. Current
attempts to provide unifying theories of physical phenomena have led to
proposals that the observed relationships are consequences of symmetries
expressed in multidimensional spaces with discontinuities such as strings
and wormholes.30

6.11 Conclusion
Lie established the result that every many-parameter transformation group
whose generators are first-order differential operators is defined by a
restricted class of first-order partial differential equations. He showed that
first-order differential operators are those of a many-parameter transforma-
tion group iff they are closed under commutation.
The commutator of two operators is the noncommutative multiplica-
tion operation used to define Lie algebras: operators that are closed under
commutation are, together with scalars, the elementary objects of a Lie
algebra.
Individual Lie groups are characterized by Lie-algebraic properties of
their generators, and the intervals over which the group parameters range —
the topology of their parameter space. If one is only concerned with the
interaction of the operators upon each other, their characterization as
abstract topological groups is complete.
Each differential equation of the sciences is invariant under the action of
a Lie transformation group; its solutions are interconverted by the group. It
October 28, 2010 16:55 9in x 6in b943-ch06

196 Dynamical Symmetry

is now known that partial differential equations can be invariant under the
action of continuous transformation groups whose generators are differential
operators of order greater than one.a Each particular transformation group
can also be realized in a great number of ways. These truths have the con-
sequence that Lie groups have greater relevance to the natural sciences
than expected. When relations between observations are expressions of
Lie symmetries, this freedom of interpretation also allows one to consis-
tently conceive the Lie symmetries as geometric symmetries in a variety of
spaces — Euclidean three-space, Newtonian spacetime, Hamiltonian phase
spaces, Lorentzian spacetime, Einsteinian spacetimes with mass dependent
metrics, the spaces of string theories, etc.

Appendix A. Definition of Lie Groups by

Partial Differential Equations
In this appendix we give proofs of two of Lie’s basic theorems. The starting
point is more general than that in Sec. 6.1, because it is not assumed that
a many-parameter group can be expressed as a product of one-parameter
groups.
We first establish the conditions that a set of first-order partial differ-
ential equations
∂x
i /∂ακ = ξiκ (x
, α), (6.A.1)
must satisfy, to define a many-parameter Lie transformation group of vari-
ables x = (x1 , x2 , . . . , xn ). Let
xi → x
i = fi (x; α), i = 1, . . . , n (6.A.2a)
define a transformation T(α)x depending upon r independent, continuously
variable, parameters
α = (αr , αr−1 , . . . , α2 , α1 ). (6.A.2b)
Let the identity transformation be given, by setting α = 0. Denoting the
succession of two transformations by T(β) T(α), we wish to establish the
form of the differential Eqs. (6.A.1) that will ensure
T(β) T(α) = T(γ), (6.A.3a)
with γ = γ(α, β), i.e.
γκ = γκ (α, β), κ = 1, . . . , r. (6.A.3b)

a These are introduced in Chapter 9.

October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 197

If the first of two transformations is x
= f(x; α), and the second is x

=
f(x
; β), then

x

= f(f(x; α), β) = f(x, γ). (6.A.3c)

The first key question is What restrictions do these requirements impose

on differential equations that define infinitesimal transformations and group
generators? In his paper of 1880,1 Lie showed that this question could be
answered by considering one of the transformations to be infinitesimal. The
reader will find it rewarding to consult this paper: in it Lie first considers
a number of examples involving transformations of one or two variables.
He then derives his general theorems by explicitly considering the effects of
variations of every pair of the group parameters. In the derivation below,
this restriction is removed.25
Let

T(a)x = x
= f(x; α), (6.A.4a)

and

T(δa)T(a)x = T(a + da)x. (6.A.4b)

Define

T(δa)T(a)x − T(a)x = δx
, (6.A.4c)

and

T(a + da)x − T(a)x = dx
. (6.A.4d)

The composition stated in (6.A.4b) requires that δx
= dx
. Given (6.A.3b),
this can only occur if

da = γ(α, δα), (6.A.5)

i.e.

daκ = γκ (α, δα).

Now Eq. (6.A.4a) requires that

dx
i = Σκ (∂fi (x
; α)/∂ακ ) dακ , and (6.A.6a)

Equation (6.A.5) requires that

δx
i = Σλ ((∂fi (x
; α)/∂αλ )|α=0 )δαλ = Σλ uiλ (x
)δαλ . (6.A.6b)
October 28, 2010 16:55 9in x 6in b943-ch06

198 Dynamical Symmetry

Because δx
i = dx
i , it must also be true that

Σλ uiλ (x
)δαλ = Σκ (∂fi (x
; α)/∂ακ ) dακ , (6.A.7)
a relation in which the δαλ and dακ are related by (6.A.5). The latter
requires that
ακ = γκ (α, 0), (6.A.8a)
and hence, that
ακ + dακ = γκ (α, δα). (6.A.8b)
Let
νκλ (α) = (∂γκ (α, β)/∂βλ )|β=0 . (6.A.9a)
Then
ακ + dακ = ακ + Σλ νκλ (α) δαλ , (6.A.9b)
and consequently
dακ = Σλ νκλ (α) δαλ . (6.A.10)
This linear relation between the dα’s and the δα’s can be solved for the
latter, if the determinant of the νκλ does not vanish. If this were not true
the r group parameters would not all be independent. Thus the requirement
that the group parameters are independent requires that the determinant,
det(νκλ (α)), be nonzero. One then has
δαλ = Σκ ωλκ (α)dακ , (6.A.11a)
where the ωλκ satisfy
Σκ ωλκ νκλ
= δλ,λ
. (6.A.11b)
Here δλ,λ
is the Kronecker delta. If the ωλκ and the νκλ are considered to
be elements of r × r matrices, ω and ν, then (6.A.11b) states that ων = I.
The existence of ω implies the existence of its inverse, ν, and vice versa.
Using (6.A.11) one can express (6.A.6a) as
dx
i = Σλ uiλ (x) ωλκ (α) dακ , (6.A.12a)
so
∂x
i /∂ακ = Σλ uiλ (x) ωλκ (α). (6.A.12b)
Consequently, in the differential Eqs. (6.A.1) one must have
ξiκ (x
, α) = Σλ uiλ (x
) ωλκ (α). (6.A.13a)
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 199

This implies:
The differential equations defining a many-parameter group must be of
the form

∂x
i /∂ακ = Σλ uiλ (x
) ωλκ (α), i = 1, . . . , n, λ, κ = 1, . . . , r. (6.A.13b)

This result states a necessary, but not sufficient, condition that must be
satisfied if the differential equations are to define a many-parameter group.
The problem is that it may or may not be possible to integrate them, and
so, for each value of κ, determine the transformations of a one-parameter
group with parameter ακ . For Eqs. (6.A.13b) to be integrable it is necessary
and sufficient that

∂ 2 x
i /∂ακ ∂αµ = ∂ 2 x
i /∂αµ ∂ακ . (6.A.14)

This requires that

∂ ξiκ (x
, α)/∂αµ − ∂ ξiµ (x
, α)/∂ακ = 0, (6.A.15a)

with

ξiν (x
, α) = Σλ uiλ (x
) ωλν (α). (6.A.15b)

When these conditions are satisfied, one obtains:

Lie’s First Fundamental Theorem. A set of integrable differential equa-
tions of the form

∂xi /∂ακ = Σρ uiρ (x)ωρκ (α), i = 1, . . . , n, ρ, κ = 1, . . . , r, (6.A.16a)

defines an r-parameter continuous group with generators

Uρ = Σi uiρ (x)xi , (6.A.16b)

where the ωρκ (α) are elements of an invertible r × r matrix.

The following paragraphs show that this theorem implies a further the-
orem which states that the generators of a many-parameter transformation
group are closed under commutation. It is derived by applying the integra-
bilty conditions to the right-hand side of (6.A.16a).
For each pair of values of µ and κ, the integrabily conditions yield the
equation

Σρ (∂uiρ (x
)/∂αµ ωρκ (α) + uiρ (x
)∂ωρκ (α)/∂αµ )

− Σρ (∂uiρ (x
)/∂ακ ωρµ (α) + uiρ (x
)∂ωρµ (α)/∂ακ ) = 0. (6.A.17a)
October 28, 2010 16:55 9in x 6in b943-ch06

200 Dynamical Symmetry

This may be rearranged to read

Σρ ∂uiρ (x
)/∂αµ ωρκ (α) − Σρ ∂uiρ (x
)/∂ακ ωρµ (α)
= Σρ uiρ (x
)(∂ωρµ (α)/∂ακ − ∂ωρκ (α)/∂αµ ). (6.A.17b)

Let us first deal with the derivatives in the first line of (6.A.17b). As the
α’s vary and produce variations in the x’s, the resulting local changes in
the uiρ (x
) are given by

∂uiρ (x
)/∂ατ = Σj ∂uiρ (x
)/∂x
j ∂x
j /∂ατ . (6.A.18a)

As (6.A.16a) requires

∂x
j /∂ατ = Σσ ujσ (x
)ωστ (α),

one has

∂uiρ (x
)/∂ατ = Σj Σσ ∂uiρ (x
)/∂x
j ujσ (x
) ωστ (α). (6.A.18b)

The appearance of both uiρ (x
) and ujσ (x
) suggests that (6.A.18b) may be
used to establish relations between several generators. Applying (6.A.18b)
to the first sum in (6.A.17b), by setting τ = µ, yields

Σρ ∂uiρ (x
)/∂αµ ωρκ (α) = Σρ Σσ Σj ujσ (x
) ∂uiρ (x
)/∂x
j ωσµ (α) ωρκ (α).
(6.A.19a)
Applying (6.A.18b) to the second sum in (6.A.17b), by setting τ = κ, yields

Σρ ∂uiρ (x
)/∂ακ ωρµ (α) = Σρ Σσ Σj ujσ (x
) ∂uiρ (x
)/∂x
j ωσκ (α) ωρµ (α).
(6.A.19b)
Much of the following analysis involves reorganizing terms in multiple sums.
This may be accomplished by altering summation indices. We begin this
process by interchanging the summation indices ρ and σ in (6.A.19b),
obtaining

Σσ Σρ {(Σj ujρ (x
) ∂uiσ (x
)/∂x
j ) ωρκ (α) ωσµ (α)}. (6.A.19c)

Subtracting it from (6.A.19a) yields the following expression for the first
line in (6.A.17b):

Σρ Σσ {Σj (ujσ (x
) ∂uiρ (x
)/∂x
j − ujρ (x
) ∂uiσ (x
)/∂x
j )}ωσµ (α) ωρκ (α).
(6.A.19d)
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 201

Referring to Eq. (5.8.9) of Sec. 5.8, the reader will find that the expression
enclosed in curly brackets is the coefficient of x
i defined by the commutator
of two generators Uσ = Σj ujσ (x
)x
j , and Uρ = Σi uiρ (x
)x
i . That is, one can
write

[Uσ , Uρ ] = {Σj (ujσ (x
) ∂uiρ (x
)/∂x
j − ujρ (x
) ∂uiσ (x
)/∂x
j )}x
i . (6.A.20)

With this observation in mind, let us proceed.

We first rename the summation indices ρ and σ in (6.A.19d) by attach-
ing primes to them. In the second line of (6.A.17b) we also change the
summation index, from ρ to τ . Equation (6.A.17b) then becomes

Σρ
Σσ
{Σj (ujσ
(x
) ∂uiρ
(x
)/∂x
j − ujρ
(x
) ∂uiσ
(x
)/∂x
j ) ωσ
µ (α) ωρ
κ (α)}
= Στ uiτ (x
)(∂ωτ µ (α)/∂ακ − ∂ωτ κ (α)/∂αµ ). (6.A.21)

The terms ωσ
µ (α)ωρ
κ (α) on the left-hand side of this expression may be
removed with the aid of the general relation (6.A.11b): multiplying (6.A.21)
by νµσ and νκρ , and then summing over µ and κ converts (6.A.21) to

Σj (ujσ (x
)x
j uiρ (x
) − ujρ (x
)x
j uiσ (x
)) = Στ uiτ (x
) Cτσρ (α), (6.A.22a)

where

Cτσρ (α) = Σµ Σκ (∂ωτ µ (α)/∂ακ − ∂ωτ κ(α)/∂αµ )νµσ (α)νκρ (α). (6.A.22b)

In (6.A.22a) we have a relation in which the x’s and the α’s can in
principle be varied independently. The reader is invited to use this freedom
and a power series expansion of the Cτσρ (α), to show that for (6.A.22a) to
hold true for any choice of the x’s and α’s, each Cτσρ (α) must be independent
of the α’s, and may be replaced by a constant, cτσρ .
Having established this, one may replace (6.A.22) by

Σj (ujσ (x
)x
j uiρ (x
) − ujρ (x
)x
j uiσ (x
)) = Στ uiτ (x
)cτσρ , (6.A.23a)

with each

cτσρ = Σµ Σκ (∂ωτ µ (α)/∂ακ − ∂ωτ κ (α)/∂αµ )νµσ (α) νκρ (α). (6.A.23b)

a constant.
As the right-hand side of (6.A.23a) is a sum of the coefficient func-
tions uiτ (x
) of generators Uτ = Σi uiτ (x
)x
i , and the left-hand side is
the corresponding coefficient of the xi defined by the commutator of
October 28, 2010 16:55 9in x 6in b943-ch06

202 Dynamical Symmetry

Uσ = Σj ujσ (x
)x
j , and Uρ = Σi uiρ (x
)x
i , we have obtained:

Lie’s Second Fundamental Theorem. The generators Uγ of an

r-parameter group satisfy the relations

[Uσ , Uρ ] = Στ cτσρ Uτ , σ, ρ, τ = 1, . . . , r. (6.A.24)

Lie’s Third Fundamental Theorem, the proof of which is suggested in

Sec. 6.3, states that:
The structure constants cτσρ satisfy the relation

Σµ (cµαβ cνµγ + cµβγ cνµα + cµγα cνµβ ) = 0. (6.A.25)

The converse of the Second Fundamental Theorem was stated in

Theorem 6.1.1 above. For derivations of the converse of each of the the-
orems, the reader is referred to Eisenhart’s book.31

Appendix B. Classification of Lie Algebras and Lie Groups31

Two Lie algebras with the same set of structure constants are, from an
abstract algebraic standpoint, identical, and are classified as such. Lie alge-
bras may be classified in a variety of ways, and each classification has a
counterpart in the classification of Lie groups. It is customary in classi-
fying Lie algebras to eliminate the zero operator, U0 , from consideration.
This should be kept in mind in reading the material in this section.
A Lie algebra and its corresponding group are said to be Abelian if all its
generators commute, i.e. [Uκ , Uλ ] = 0 for the generators U in the algebra.
If a subset of the generators of an algebra commute among themselves
they constitute a Abelian gruop, subalgebra. The generators of an Abelian
subalgebra constitute the center of the whole algebra.
We move on to a consideration of further properties of Lie algebras
important in their classification. Each of the properties defined will be illus-
trated in the exercises.
If the operator A of a Lie algebra can be separated into two subalgebras
B, C such that none of the generators in B are contained in C and vice
versa, and that the generators in B commute with those in C, then the Lie
algebra A is a direct sum of the ideals B and C. This is written

A = B ⊕ C. (6.B.1)

With some abuse of notation one may write the conditions as

[B, B] = B, [C, C] = C, [B, C] = 0. (6.B.2)

October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 203

The corresponding group operator, TA , is then a direct product of the

subgroup operators TB and TC . This relationship is expressed as
TA = TB ⊗ TC . (6.B.3)
An example of two such commuting algebras is provided by the operators
y1 ∂/∂z1 − z1 ∂/∂y1 and y2 ∂/∂z2 − z2 ∂/∂y2 , which may be considered to be
generators of a two-parameter group that rotates two different particles in
the y–z plane.
A Lie algebra A is a semidirect sum of the subalgebras B and C if
[B, B] = B, [C, C] = C, [B, C] = B. (6.B.4)
This may be written
A = B ⊕s C, (6.B.5)
with the subalgebra B placed to the left of ⊕s . The corresponding group
operator, TA , is termed a semidirect product of the subgroup operators TB
and TC . This can be expressed as
TA = TB ⊗S TC . (6.B.6)
The subalgebra B is termed a ideal or normal subalgebra. If an ideal does
not contain U0 , the ideal is a proper ideal.
A Lie algebra is said to be simple if it contains no proper ideals. A Lie
algebra is semi-simple if it contains no Abelian ideals other than U0 . A Lie
algebra is semi-simple iff the determinant of its Killing form is nonzero.
A Lie algebra Ar of dimension r is a solvable Lie algebra if it possesses
a chain of subgroups Ak such that
A1 ⊃ A2 ⊃ Ar−1 ⊃ Ar . (6.B.7)
The term solvable was introduced by Lie. In his investigations of ordi-
nary differential equations he established that if an ordinary differential
equation of order R is left invariant by a solvable group of order r whose
generators are first-order differential operators U, then the order of the
differential equation may be reduced to R − r without carrying out any
integrations.33
The Levi–Malcev theorem shows that every Lie algebra A may be
expressed as a semidirect sum of a solvable Lie algebra So and a semi-
simple Lie algebra Se
A = So ⊕s Se. (6.B.8)
Every Lie group may be expressed as the corresponding semidirect product.
October 28, 2010 16:55 9in x 6in b943-ch06

204 Dynamical Symmetry

Further information on the classification of Lie algebras and groups will

be found in Gilmore’s monograph, which contains extensive tables of groups
and their subgroups.34
Exercises.

1. Let Rx , Ry , Rz be generators of rotations on the sphere. Find the commu-

tation relations of the three operators Rp = (Rx + Ry ), Rm = (Rx − Ry ),
Rz . Determine the Killing form.
2. Let U = t2 ∂/∂t, which, as we have seen, generates finite transformations
that have singularities. Determine the action of U on θ = 2arctan(t), the
form of the corresponding V = v∂/∂θ, and the finite transformations of
θ that it generates. Is V the generator of a compact group?
3. Let Rjk = xj ∂/∂xk − xk ∂/∂xj, and define U1 = (R23 + R14 )/2, U2 =
(R31 + R24 )/2, U3 = (R12 + R34 )/2.
(a) Show that the U’s obey the commutation relations of SO(3) and
SU(2).
(b) Prove that the U’s generate a transformation group whoes opera-
tions leave invariant the hyperspheres defined by x21 + x22 + x23 + x24 =
r2 .
(c) Show that, acting on this sphere, the group is SU(2).
(d) Change the Rj4 to Sj4 = xj ∂/∂x4 + x4 ∂/∂xj , then show that
operators of the form V1 = (R23 + R14 )/2, V2 = (R31 + R24 )/2,
V3 = (R12 + R34 )/2 do not generate SU(2) groups.
4. Use the invariant ds2 of the Lorentz realization of SO(3,1) to establish
the form of the manifolds in spacetime that are left invariant by trans-
formations of the group.
5. Construct the generators of an SO(3,1) group that leaves invariant a
unit sphere. Hint: cf. Eqs. (5.1.9).
6. Let L = (Lx , Ly , Lz ), p = −i∇. Do p and L commute with L2 + p2 ?
With L · p?
7. Let
U1 = ∂/∂x, U2 = x∂/∂x, U3 = (x2 /2) ∂/∂x.
Show that
[U1 , U2 ] = U1 , [U2 , U3 ] = 2U3 , [U3 , U1 ] = −U2 .
Then establish that
c112 = 1 = −c121 , c323 = 2 = −c332 , c231 = −1 = −c213 ,
October 28, 2010 16:55 9in x 6in b943-ch06

Lie Transformation Groups and Lie Algebras 205

and that all other c’s vanish. Show that in the Killing form,
g11 = 0, g12 = 0, g13 = −3, g22 = 5, g23 = 0, g33 = 0.
Evaluate the determinant of the form, and determine whether the group
is semi-simple.
8. Acting in two-dimensional Euclidean space, the Euclidean group E(2)
has generators
U1 = ∂/∂x, U2 = ∂/∂y, U3 = x∂/∂y − y∂/∂x.
Show that
c312 = 0, c123 = −1 = −c132 , c231 = −1 = −c213 ,
and that all other c’s vanish. Show that the Killing form is




0 0 0





0 0 0
.




0 0 −2

9. Let
dx/da = Ax, dx/db = Bx.
Let A, B respectively be linear and quadratic functions of a, b. In what
cases can the equations define a two-parameter group?

References
[1] (a) S. Lie, Math. Ann. 16 (1880) 451–528; Coll. Works, Vol. 6, pp. 1–94,
(Johnson Reprint, NY, 1973).
(b) Sophus Lie’s 1880 Transformation Group Paper, translated by M. Ack-
erman and R. Hermann (Math. Sci. Press, Broookline. Massachusetts).
[2] Ref. 1(b) above, pp. 161–169; S. Lie, Vorlesungen uber Continuirliche Grup-
pen, reprint (Chelsea, Bronx, N.Y. (1971)).
[3] S. Lie, Transformations Gruppen III, reprint (Chelsea, Bronx, N.Y, 1970).
[4] For a proof of the theorem, see S. Pontriagin, Topological Groups (Princeton,
1946).
[5] G. J. Reid, Eur. J. Appl. Math. 2 (1991) 293, 319.
[6] H. F. Baker, Proc. Lond. Math. Soc. 34(1) (1903) 347–360; 35(1) (1903)
333–374; 2(2) (1904) 293–296.
[7] J. E. Campbell, Proc. Lond. Math. Soc. (1) 29 (1898) 14–32.
[8] F. Hausdorff, Ber. Sach. Akad. Wiss. (Math. Phys. Klasse) Leipzig 58 (1906)
19–48.
[9] A. R. Edmonds, Angular Momentum in Quantum Mechanics (Princeton,
1957), pp. 5–7.
[10] C. E. Wulfman and B. G. Wybourne, J. Phys. A 9 (1976) 507–518.
October 28, 2010 16:55 9in x 6in b943-ch06

206 Dynamical Symmetry

[11] C. Wulfman, Int. J. Quantum Chem. 49 (1994) 185–195.

[12] Y. Aharonov and D. Bohm, Phys. Rev. 115 (1959) 485–491.
[13] (a) G. Herzberg and H. C. Longuet-Higgins, Discuss. Faraday Soc. 35 (1963)
77–82;
(b) H. C. Longuett-Higgins, Proc. Roy. Soc. (London), A 344 (1975) 147–
156;
(c) C. A. Mead, J. Chem. Phys. 49 (1980); 23–32. 33–38.
(d) C. A. Mead and D. G. Truhlar, J. Chem. Phys. 70 (1979) 2284–2296.
[14] M. V. Berry, Proc. Roy. Soc. (London), A 392 (1984) 45–57.
[15] W. Killing, Math. Ann. 31 (1888) 252; 33 (1889) 1; 34 (1889) 55; 36 (1890)
161.
[16] E. Cartan, Oeuvres Completes I, pt. 1,2 (Gauthier-Villers, Paris, 1952).
[17] E. U. Condon and G. H. Shortley, Theory of Atomic Spectra (Cambridge,
1935).
[18] A. R. Edmonds, Angular Momentum in Quantum Mechanics (Princeton,
1957).
[19] H. B. G. Casimir, Rotation of a Rigid Body in Quantum Mechanics, thesis,
(University of Leiden Wolters, Gronigen, 1931).
[20] B. G. Wybourne, Classical Groups for Physicists, Chap. XV (Wiley-
Interscience, NY, 1974).
[21] W. T. Sharp, Thesis (Princeton University, Princeton, N.J., 1957).
[22] J. D. Talman, Special Functions. A Group Theoretic Approach, Based on
Lectures by Eugene P. Wigner (Benjamin, NY 1968).
[23] G. N. Watson, Theory of Bessel Functions, 2nd edn. (Cambridge, 1966).
[24] N. J. Vilenkin, Special Functions and the Theory of Group Representations,
Am. Math. Soc., translation by V. N. Singh (Providence, R.I, 1968).
[25] G. Racah, Group Theory and Spectroscopy, Ergeb. Ex. Naturwiss, Vol. 37,
(Springer-Verlag, Berlin, 1965), pp. 28–84.
[26] S. Pontriagin, Topological Groups (Princeton, 1942).
[27] B. Russell, quoted in A.S. Eddington, New Pathways in Science (Cambridge,
1935), p. 257.
[28] B. Russell and A. N. Whitehead, Principia Mathematica, I–III (Cambridge
1910–1913).
[29] A. S. Eddington, loc. cit.
[30] For a popular account, see B. Greene, The Elegant Universe (Vintage Books,
N.Y., 2000).
[31] L. P. Eisenhart, Continuous Groups of Transformations (Princeton Univer-
sity Press, Princeton, N. J. 1933). Note: there is considerable variation in
the literature as to the nameing of Lie’s theorems and their converses.
[32] B. G. Wybourne, Lie Groups for Physicists (Wiley, NY, 1974), pp. 43–55.
[33] For an excellent exposition of the use of solvable Lie groups see
G. W. Bluman and S. Kumei, Symmetries and Differential Equations
(Springer-Verlag, NY, 1989).
[34] R. Gilmore, Lie Groups, Lie Algebras, and Some of Their Applications
(Wiley, NY, 1974).
October 28, 2010 16:55 9in x 6in b943-ch07

CHAPTER 7

Dynamical Symmetry in Hamiltonian Mechanics

This chapter generalizes and extends the discussion of the Hamiltonian

mechanics introduced in Chapter 3. Because Hamiltonian mechanics inher-
its certain general invariance properties from Newtonian mechanics, we
begin with a brief discussion of the latter.

7.1 General Invariance Properties of Newtonian Mechanics

Let a set of point masses mj , j = 1,. . . ,n, have position vectors rj =
(xj , yj , zj ) in a Cartesian frame of reference, X,Y,Z. If their velocity vec-
tors are vj = (vxj , vyj , vz ), then their momentum vectors are pj = (pxj ,
pyj , pzj ) = (mj vxj , mj vyj , mj vzj ). Newton’s first law states that if no force
is acting on the masses, then for each mass, mj ,

dpj /dt = 0. (7.1.1)

If the masses are constant, (7.1.1) reduces to

mj dvj /dt = 0. (7.1.2)

Newton’s second law requires the change of momenta, pj , to be due the

action of quantifiable forces, Fj = Fxj , Fyj , Fzj , such that

dpj /dt = d2 (mj rj )/dt2 = Fj . (7.1.3)

Newton’s third law states that if particle i exerts a force Fij on particle j,
then particle j exerts a force Fji on particle i, with

Fji = −Fij . (7.1.4)

In an isolated system of interacting particles, the total force, Fj , acting on

particle j is Σi Fij , so the total force, Σi Σj Fij , acting between all particles

207
October 28, 2010 16:55 9in x 6in b943-ch07

208 Dynamical Symmetry

is zero. From this it follows that the total momentum of the system is
conserved:
Σj dpj /dt = d2 (Σj mj rj )/dt2 = 0. (7.1.5)
Newton proved that spherically symmetric masses of total mass mj behave
as point masses of mass mj when acted on by external inverse square law
forces, such as his force of gravity.
Equations (7.1.3) are second-order ordinary differential equations. A
knowledge of the Fj , and initial position and initial velocity vectors, is
therefore sufficient to predict the motions in a classical mechanical system.
This quite amazing feature of Newtonian mechanics withstood all experi-
mental tests and challenges until the advent of quantum theory.
Newton’s equations of motion for a system of interacting particles, are
left invariant by a group of transformations that relates measurements made
by two observers moving with constant relative velocity using clocks that
agree on time intervals. The equations are invariant under the passive trans-
formation t → t
= t − (±) t0 . They are invariant under passive coordinate
transformations that convert one Cartesian system with axes (X, Y, Z) to
another with axes (X
, Y
, Z
), if the two origins are moving with relative
velocity −(±) (vx0 , vy0 , vz0 ), and at time t = 0, have origins whose posi-
tion vectors differ by −(±) x0 , y0 , z0 . The two systems of axes may differ
in orientation. Measurements of positions and velocities in the two systems
are related by the active transformations
t → t
= t ± t 0 , (7.1.6a)
rj → r
j = rj ± (x0 , y0 , z0 ), (7.1.6b)
T
r
j = ±R(xj , yj , zj ) , (7.1.6c)
where R is a rotation matrix. The transformation
vj → vj
= vj ± (vx0 , vy0 , vz0 ), (7.1.6d)
also converts rj to
r
j = rj ± (tvx0 , vy0 , vz0 ). (7.1.6e)
The transformations (7.1.4a–e) are those of the Lie transformation
group termed the Galilei group.1 From a purely mathematical standpoint,
the transformations of the group may be active or passive ones. From a
physical standpoint, the transformations must relate measurements made
in one reference system to those in another by passive transformations of
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 209

coordinates. If one were to use an active physical interpretation of the

group’s action, one would need to apply the transformations to physical
particles moving with respect to a single set of position and time coor-
dinates. This would require the application of forces to accelerate and/or
decelerate the particles. The particles, together with the required measuring
sticks and clocks, would no longer constitute an isolated system. To turn the
system into an isolated one requires making the experimenter part of the
system — a problem which still vexes quantum theory. However, because of
conceptual simplicity, one commonly uses active transformations to math-
ematically interconvert solutions of a given equation of motion.
The Galilei group is a ten-parameter group whose passive operations on
a (t, r) coordinate system may be summarized by
t → t
= t + τ, r → r
= R(α, β, γ)r + v0 t + r0 , (7.1.7)
where R(α, β, γ) is a rotation operator. The group parameters may be τ , α,
β, γ, and the six components of v0 and r0 . The generators of the subgroup
that does not depend on v0 may be
T0 = ∂/∂t, T1 = ∂/∂x, T2 = ∂/∂y, T3 = ∂/∂z, (7.1.8)

R1 = y∂/∂z − z∂/∂y, R2 = z∂/∂x − x∂/∂z, R3 = x∂/∂y − y∂/∂x.

R1 , R2 , R3 , and the T’s generate an E(3) group of point transformations
that acts in the Euclidean space of (X, Y, Z). T0 generates an E(1) group
that shifts the value of t, the independent variable in Newton’s equations.
Together, the seven operators generate E(3) × E(1). The generators of
the passive transformations corresponding to (7.1.6d,e), act on both r and
v = dr/dt, and are
K1 = t∂/∂x + ∂/∂vx , K2 = t∂/∂y + ∂/∂vy , K3 = t∂/∂z + ∂/∂vz .
(7.1.9)
These generate one-parameter groups whose parameters are, respectively,
the vxo , vyo , vyo , components of v0 .
A space whose points are carried into one another by the group of
translations and rotations is continuous, and is said to be homogeneous
and isotropic. Newtonian time is continuous, homogeneous, and isotropic.
These continuity properties make the calculus physically relevant, and make
it possible for Newtonian physical causality to be continuous.
There are also discrete transformations that leave Newton’s equations
invariant. For isolated systems, Eqs. (7.1.1) and (7.1.2), are invariant under
October 28, 2010 16:55 9in x 6in b943-ch07

210 Dynamical Symmetry

the combined passive time-reversal and space-inversion whose active form is

t → −t, rj → −rj . (7.1.10)

If the forces in Eq. (7.1.3) are time-independent, the equation is time-

reversal invariant. If Fj is an odd function of the r’s, Eq. (7.1.3) is also
space-inversion invariant. Transformations such as

X → Y, Y → X, Z → −Z, (7.1.11)

which convert a right-handed coordinate system into a left-handed one,

leave all of Newton’s equations invariant if the forces Fj (r1 , . . . , rn ) are
invariant under the corresponding active transformations of the ri .
The conversion of the space, time, and velocity transformations of the
Galilei group to transformations in phase space will be discussed in Sec. 7.6
below.

7.2 Relationship of Phase Space to Abstract

Symplectic Space
This section further develops and generalizes Sec. 3.5, which introduced a
geometric view of the Hamiltonian mechanics of systems with one q and
one p. Such systems are, for historical reasons, said to have one degree of
freedom, which they do have in position space.
We begin the treatment of Hamiltonian systems with n degrees of free-
dom by classifying the coordinates into two sets of n variables

q = (q1 , . . . , qn ), p = (p1 , . . . , pn ). (7.2.1)

At this point, the q’s and p’s are considered to be mathematical symbols
whose physical interpretations have yet to be supplied. To generalize the
discussion of Hamiltonian mechanics in Chapter 3, we consider transforma-
tions of the combined sets of variables (q, p) to sets (q
, p
) in which

qi → q
i = Qj (q, p), pj → p
j = Pj (q, p). (7.2.2)

These give rise to the active local transformations

dqi → dq
i = Σk (∂Qi /∂qk dqk + ∂Qi /∂pk dpk ) = Σk (aik dqk + bik dpk ),
dpj → dp
j = Σk (∂Pj /∂qk dqk + ∂Pj /∂pk dpk ) = Σk (cjk dqk + ejk dpk ),
(7.2.3)
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 211

with

∂Qi /∂qk = aik , ∂Qi /∂pk = bik , ∂Pj /∂qk = cjk , ∂Pj /∂pk = ejk .

A point Ss is considered to lie at the terminus of a Cartesian vector

s = (q1 , . . . , qn , p1 , . . . , pn ), (7.2.4a)

the vector sum of

q = (q1 , . . . , qn , 0, . . . , 0) (7.2.4b)

and

p = (0, . . . , 0, p1 , . . . , pn ).

The original dq’s and dp’s are components of 2n-dimensional displacement

vectors,

ds = (dq1 , . . . , dqn , dp1 , . . . , dpn ) = dq + dp. (7.2.4c)

The pair of orthogonal unit vectors of Sec. 3.5 may generalized to a set of 2n
orthogonal unit vectors which we denote i1 ,. . . ,in and j1 ,. . . ,jn . Eq. (7.2.3)
then state

dqi → dq
i = Σk (aik dqk + bik dpk ) = ds
i , (7.2.5a)

dpj → dp
j = Σk (cjk dqk + ejk dpk ) = ds
j .

The coefficients aik and bik represent the components of the local displace-
ment vectors on the unit vectors of the reference system:

aik = dq
i · ik , bik = dp
i · jk ,

(7.2.6)
cjk = dq
j · ik , ejk = dp
j · jk .

The vectors dq and dp have a common origin at the point Ss and together
define a two-plane. Figure 7.2.1a is an attempt to depict two such planes
that are produced as Ss moves in a space of dimension greater than two.
In Chapter 3, transformations which convert Hamilton’s equations of
motion into Hamilton’s equations of motion were characterized analyti-
cally with the aid of Poisson brackets, and geometrically by oriented areas
defined with the aid of the wedge product analog of the vector cross prod-
uct. To generalize these concepts to 2n dimensions, one first of all defines
October 28, 2010 16:55 9in x 6in b943-ch07

212 Dynamical Symmetry

dp
dp
dq dq

Fig. 7.2.1 Motion of the plane containing dq and dp.

the Poisson bracket {A(q, p), B(q, p)} of functions A(q,p) and B(q,p) by

{A, B} = Σj (∂A/∂qj ∂B/∂pj − ∂B/∂qj ∂A/∂pj ). (7.2.7)

As will be explained in Sec. 7.4 below, it is necessary to evaluate multiple

Poisson brackets involving three functions, e.g. {{A(q, p), B(q, p)}, C(q, p)}.
This requires that the functions possess second derivatives. As it will be
necessary that the resulting derivatives are themselves continuous functions,
we henceforth require all functions of the q’s and p’s to be C2 .
The transformations (7.2.2) are defined to be symplectic if

Σk (∂Qi /∂qk ∂Pj /∂pk − ∂Pj/∂qk ∂Qi /∂pk ) = δij ,

Σk (∂Qi /∂qk ∂Qj /∂pk − ∂Qj /∂qk ∂Qi /∂pk ) = 0, (7.2.8)

Σk (∂Pi /∂qk ∂Pj /∂pk − ∂Pj /∂qk ∂Pi /∂pk ) = 0.

In the Poisson bracket notation these equations become

{q
i , p
j } = δij , {q
i , q
j } = 0, {p
i , p
j } = 0. (7.2.9)

One gives a geometric interpretation to the equations with the aid of the
wedge product defined in Chapter 3. At the points Ss and S
s the vectors
dq and dp, and the vectors dq
and dp
define local wedge products whose
magnitude is twice the area of the plane triangle obtained by connecting
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 213

their termini. The transformation of q and p moves Ss to S
s , and in the

process

dq ∧ dp → dq
∧ dp
. (7.2.10)

If dqk ∧ dpk is the wedge product in the local qk –pk subplane at s, and
dqk ∧ dpk is the wedge product in the local q
k –p
k subplane at s
, then (cf.
the exercises), Eq. (7.2.9) imply that


dq ∧ dp = (dqk ∧ dpk ), (7.2.11a)
k


and that at s ,


dq
∧ dp
= (dqk ∧ dpk ). (7.2.11b)
k

In the coordinate system at, say s, Eq. (7.2.9) imply

dq ∧ dp = dq
∧ dp
. (7.2.11c)

The most direct physical interpretations of symplectic transformations

develop when one is dealing with Cartesian variables xj , and momentum
variables defined by pxj = mj dxj /dt = mj vj . The Poisson bracket of two
functions A, B is then defined using Eqs. (7.2.7) and (7.2.9), with qj = xj
and pj = pxj as the independent variables in (7.2.7). When generalized
coordinates are used, one begins with the Cartesian position coordinates,
makes transformations to new coordinates qj , and then determines their
conjugate momenta, pj . If the system has potential energy function V(q),
the determination of the new conjugate momenta is most simply done if
one uses the transformation from x’s to q’s to transform the kinetic energy,
T, of the system from T(mj , dxj /dt) to T(mj , qj , q
j ), where q
j = dqj /dt.
The Lagrangian, L, is then T − V(q), and Lagrange’s equations are

∂L(mj , qj , q
j )/∂qj = q
j , (7.2.12a)

pj ≡ ∂L(mj , qj , q
j )/∂q
j . (7.2.12b)

For the assumed form of L, Eq. (7.2.12b) determines the momenta pj cor-
responding to the qj . In Appendix A it is proved that the pj so defined are
canonical conjugates of the qj . The proof is prefaced by, and depends upon,
Thomson and Tait’s direct derivation of Lagrange’s equations of motion
from Newton’s.2
In the discussion of the Hamiltonian mechanics in Sec. 3.7, the phase
space of Hamiltonian mechanics differs from symplectic space, because in
October 28, 2010 16:55 9in x 6in b943-ch07

214 Dynamical Symmetry

physics one must consider scale changes q → aq, p → bp, that need not
be the scale changes q → aq, p → a−1p, which leave the wedge product
dq ∧ dp invariant. The same arguments that were applied to shifts from
clockwise directed, to counterclockwise directed, two-dimensional wedge
products, also apply to (7.2.11a). Thus, we adopt the following.

Definition of the Hamiltonian Phase Space

The phase space of 2n canonically conjugate variables q, and p, is the space
containing all points that are carried into one another by diffeomorphisms
that leave invariant
|dq ∧ dp|, (7.2.13)
or multiply it by a positive constant.

7.3 Hamilton’s Equations in PQ Space. Constants

of Motion
Hamilton’s equations of motion determine the evolution of congugate posi-
tion and momentum variables in a system with a known Hamiltonian func-
tion. In this section it is assumed that the Hamiltonian, H(p,q), is not a
function of time. Systems with time dependent Hamiltonians will be con-
sidered in Sec. 7.5 below.
We begin by deriving Hamilton’s equations from Lagrange’s equations
of the second kind (cf. Appendix A). For fixed values of the masses, mi , L is
assumed to be a function only of the qj and q
j and not an explicit function
of t. Consequently


dL = (∂L/∂qj dqj + ∂L/∂q
j dq
j ). (7.3.1)
j

Using Eq. (7.2.12) one obtains

dL = (p
j dqj + pj dq
j ). (7.3.2)
j

The second term on the rhs of this relation may be expressed as

 



d pj q
j  − q
j dpj . (7.3.3)
j j

Using this, one converts (7.3.2) to

 



d L − pj q
j  = (p
j dqj − q
j dpj ). (7.3.4)
j j
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 215

On defining
 


H = − L − pj q
j  , (7.3.5)
j

Equation (7.3.4) can be written as

dH = (q
j dpj − p
j dqj ). (7.3.6)
j

The fact that dH is a function of the dp’s and dq’s suggests that it might
be useful to convert H in (7.3.5) to a function of q’s and p’s. This is done
by defining Hamilton’s function, H, by


H(q, p) ≡ pj vj − L |v=g(q,p) . (7.3.7)

It has


dH = (∂H(q, p)/∂pj dpj + ∂H(q, p)/∂qj dqj ). (7.3.8)
j

Comparing this with (7.3.6) one observes that dH and dH become identi-
cal if

∂H(q, p)/∂pj = q
j , ∂H(q, p)/∂qj = −p
j . (7.3.9a)

Choosing this to be so, one obtains Hamilton’s equations of motion

dqj /dt = ∂H(q, p)/∂pj , dpj /dt = −∂H(q, p)/∂qj . (7.3.9b)

As H is a function only of q,p,

dH = (∂H(q, p)/∂pj dpj + ∂H(q, p)/∂qj dqj ), (7.3.10a)

and so


dH/dt = (dqj /dt dpj /dt − dqj /dt dpj /dt) = 0. (7.3.10b)

Now (7.3.7) can be expressed as

H(q, p) = (2T − (T − V))|v=g(q,p) = T(p, q) + V(q). (7.3.11)

The value of H(q,p) is thus the mechanical energy, E, of the system.

Equation (7.3.10b) then states that the energy of the system with the
Hamiltonian of (7.3.11) does not change with time.
October 28, 2010 16:55 9in x 6in b943-ch07

216 Dynamical Symmetry

Using Poisson bracket notation, Hamilton’s equations (7.3.9b) become

dqj /dt = {qj , H}, dpj /dt = {pj , H}, (7.3.12)
or, equivalently
dqj /dt = −{H, qj }, dpj /dt = −{H, pj }.
These last relations imply that one can identify {−H, . }, a first-order differ-
ential operator, as the generator of a one-parameter Lie group whose group
parameter is t. It follows directly that
q
j = exp(t{−H, ·})qj = Q
j (t), p
j = exp(t{−H, ·})pj = P
j (t). (7.3.13)
Differentiating q
j and p
j with respect to t, one checks that they satisfy
Hamilton’s equations, that is, Q
j (t) and P
j (t) are the solutions of the
equations of motion that develop as qj and pj evolve. One may term
exp(t{−H, ·}) the evolution operator of the system with Hamiltonian H,
and call {−H, ·} its evolution generator. Then t is the group parameter of
the evolution group.
Equations (7.3.12) may be considered to be special cases of the relation
df(q, p, t)/dt = {f, H} + ∂f/∂t, (7.3.14)
which applies when the evolution of q and p is governed by Hamilton’s
equations with Hamiltonian H. If f(q,p) is not an explicit function of time,
and
{H, f(q, p)} = 0, (7.3.15a)
it follows that
df/dt = 0. (7.3.15b)
The function f(q,p) is then said to be a constant of the motion of the
system with Hamiltonian H. The constants of motion of a Hamiltonian
may be different for different values of E, or different ranges of E. Stated
definitively, f(q, p) is a constant of motion when H(q, p) = E, iff
0 = {−H, ·}f|H=E . (7.3.15c)
It follows that, if f(q,p) is a constant of motion, it is invariant under the
one-parameter evolution group generated by {−H, ·}. Because (7.4.15c) can
also be expressed as
0 = ({f(q, p), ·}H(q, p))|H=E , (7.3.15d)
it also follows that {f(q, p), ·} generates a one-parameter Lie group which
leaves H(q,p) invariant for this value of E.
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 217

The function q1 (H–E) provides an example of a trivial constant of

motion. For this,

{f, H} = q1 {(H–E), H} + {q1 , H}(H–E) = p1 (H–E). (7.3.16)

This vanishes for all values of E, simply because H–E is a factor of f. This
is of little import. The physically interesting cases arise when f does not
directly contain H–E, but contains a factor that substitutions convert to
H–E. The situation is then similar to that illustrated in Fig. 5.1.1.

7.4 Poisson Bracket Operators

From the previous remarks it is apparent that, because Poisson bracket
operators generate one-parameter Lie groups, they play a fundamental
role in associating Lie transformation groups with symmetries in Hamil-
tonian mechanics. In fact, because they will be so ubiquitous throughout
the book, we will often abbreviate Poisson bracket as PB. This section
will deal, first of all, with the connection of PB operators to operators in
position space. It will then be shown that PB operators are Lie genera-
tors of one-parameter groups of canonical transformations in phase space.
Thereafter, PB Lie algebras will be defined and related to Lie algebras of
PB operators.
To begin, we note that one has the operator relations

∂/∂qk = −{pk , ·}, ∂/∂pk = {qk , ·}. (7.4.1)

As an example of the application of these correspondences, consider the

generator of rotations in the qi –qj plane of position space

Rij = (qi ∂/∂qj − qj ∂/∂qi ). (7.4.2a)

On replacing each ∂/∂qk in Rij by −pk , one finds

Rij → −(qi pj − qj pi ) = −Lij , (7.4.2b)

where Lij = qi pj − qj pi represents the angular momentum of a particle

moving in the qi –qj plane. The full PB operator obtained from −Lij is

{−Lij, ·} = (qi ∂/∂qj − qj ∂/∂qi ) + (pi ∂/∂pj − pj ∂/∂pi ). (7.4.2c)

It is a Lie generator of rotations in the two orthogonal two-planes qi –qj and

pi –pj in phase space. On identifying pk with mvk , one sees that (7.4.2c) is
also the first extension of (qi ∂/∂qj − qj ∂/∂qi ).
October 28, 2010 16:55 9in x 6in b943-ch07

218 Dynamical Symmetry


In the general case, to convert a Lie generator, ξk (q)∂/∂qk of point
transformations in position space, Q, to a PB operator that acts in the
phase space PQ, one first associates the position space operator with a
function of q,p by letting




ξk (q)∂/∂qk → ξk (q)(−pk ) = − ξk (q)pk . (7.4.3a)

One then associates this function with a Lie operator in phase space by
letting



− ξk (q) pk → {− ξk (q)pk , ·}, (7.4.3b)

The correspondences setup in (7.4.3) are also two-way correspondences


in the following sense: a function f(q,p) of the form − ξk (q) pk gives

rise to the PB operator {− ξk (q)pk , ·} and to a position space operator

ξk (q)∂/∂qk .
Only Lie generators of point transformations q → q
= g(q) yield
Poisson bracket operators whose position space part is of the form

ξk (q)∂/∂qk ; they correspond to functions f(q, p) that are linear in the
momenta pk . If f(q, p) is not linear in the p’s, then {f(q, p), ·} will contain
terms whose action on the q’s is a function of the p’s. As will be seen in
Chapter 8, some of the generators of transformations that convert Kepler
orbits into Kepler orbits have this property. The following paragraphs deal
with properties of PB operators of the general form {f(q, p), ·}. Each func-
tion of positions and momenta determines a PB operator that has both
physical and geometric interpretations.
One may show by direct calculation that if A, B and X are C2 functions
of the q’s and p’s, then Poisson brackets obey the Jacobi relation

{A, {B, X}} + {X, {A, B}} + {B, {X, A}} = 0. (7.4.4)

This enables one to quickly establish that PB operators are generators of

canonical transformations. Let

qi → q
i = exp(α{A, ·})qi , pj → p
j = exp(α{A, ·})pj . (7.4.5a)

Expanding {q
i ,p
j } in powers of α one obtains

{q
i , p
j } = {qi , pj } + α{{A, qi }, pj } + α{qi , {A, pj }} + O(α2 ). (7.4.5b)

The sum of the terms multiplying α can be rewriten as

{qi , {A, pj }} − {pj , {A, qi }} = {qi , {A, pj }} + {pj , {qi , A}}. (7.4.5c)
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 219

From Jacobi’s relation one sees that adding {A, {pj , qi }} to this gives zero.
As {pj , qi } is a number, {A, {pj , qi }} has the value zero. Consequently
{qi , {A, pj }} + {pj , {qi , A}} = 0, (7.4.5d)
and
{q
i , p
j } = {qi , pj } = δij . (7.4.5e)
Using the properties of Lie transformation groups established in previ-
ous chapters, one sees that (7.4.5e) holds to all orders in α. Applying the
same argument to {q
i , q
j } and {p
i , p
j }, one finds that they vanish. Thus, if
A(q, p) is a C2 function, {A, ·} generates canonical transformations.
We next prove the following relation between the PB’s of C2 functions
and PB operators:
If
{A, B} = C, (7.4.6a)
then
[{A, ·}, {B, ·}]X(q, p) = {C, ·}X(q, p). (7.4.6b)
This is a direct consequence of Jacobi’s relation amongst A,B and X, since
the relation can be rewritten as
{A, {B, X}} − {B, {A, X}} = −{X, {A, B}} = {{A, B}, X}. (7.4.7)
As the lhs is [{A, ·}, {B, ·}]X, and the rhs is {{A, B}, ·}X, the relationship
of (7.4.6) follows.
The reader may prove that if A, B and C are functions that satisfy the
Jacobi relation
{A, {B, C}} + {C, {A, B}} + {B, {C, A}} = 0, (7.4.8a)
then {A, ·}, {B, ·} and {C, ·} satisfy the Jacobi relation
[[{A, ·}, {B, ·}], {C, ·}] + [[{C, ·}, {A, ·}], {B, ·}]
+ [[{B, ·}, {C, ·}], {A, ·}] = 0. (7.4.8b)
If a finite set of C2 functions fi (q,p) satisfies (7.4.8a) and


{fi , fj } = ckij fk , (7.4.8c)
then


[{fi , ·}, {fj , ·}] = ckij {fk , ·}. (7.4.8d)
October 28, 2010 16:55 9in x 6in b943-ch07

220 Dynamical Symmetry

It follows from the general definition of Lie algebras (cf. Chapter 6,

Appendix C), that

1. If r linearly independent operators, {fi , ·}, satisfy (7.4.8d), and the fi are
C2 functions, then the {fi , ·} are elements of a Lie algebra with composi-
tion operation [ , ], and they generate an r-parameter Lie transformation
group.
2. The fi are elements of a Poisson bracket Lie algebra whose composition
operation is { , }, each giving rise to a function exp(αi fi ). The group
which the fi generate is termed a function group.
3. The commutator and PB Lie algebras are formally isomorphic.

However, if any fk is a constant, then {fi , ·} = 0, and ckij {fk , ·} = 0, a rela-

tion that also results if ckij = 0. An example can be found in Sec. 7.7 below.
It will be further investigated in the discussion of the quantum mechanical
Correspondence Principle in Chapter 9.
In this and the following chapters we will make use of Lie algebras
defined by (7.4.8c), as well as those defined by (7.4.8d).

7.5 Hamiltonian Dynamical Symmetries in PQ Space

For a system with n Cartesian q’s and n conjugate p’s, the equation

H(p, q) − E = 0, (7.5.1)
defines a manifold in a 2n-dimensional phase space, PQ space. For each
value of E it requires 2n − 1 numbers to fix a point in the manifold.
Hamilton’s equations of motion define evolution trajectories on these
(2n − 1)-dimensional hypersurfaces. The manifold defined by (7.5.1) pos-
sesses Euclidean geometric symmetries. Furthermore it possesses symplec-
tic geometric symmetries expressed by a dynamical group. The operations
of the latter leave Hamilton’s equations invariant, and define dynamical
symmetries.
We have seen that each constant of motion u(q, p) determines a Poisson
bracket operator U = {u, ·} that satisfies
0 = UH(q, p)|H=E , (7.5.2a)
and that, if UH can not contain H – E as a factor, then U also satisfies
UH(q, p) = 0. (7.5.2b)
Each U generates a one-parameter Lie group that transforms evolution
trajectories into one another or into themselves. All such transformations
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 221

carry points on the hypersurface of fixed energy into points on the same
hypersurface.
If one knows two constants of the motion, u1 and u2 , then U1 = {u1 , ·},
and U2 = {u2 , ·} satisfy
U1 H(q, p)|H=E = 0, U2 H(q, p)|H=E = 0. (7.5.3)
Then (cf. the exercises) [U1 ,U2 ] satisfies
[U1 , U2 ]H(q, p)|H=E = 0. (7.5.4)
If [U1 , U2 ] is not identically zero, then u3 = {u1 , u2 } is a further constant
of the motion. If a set of r linearly independent constants of motion, uj , is
closed under the PB operation, they generate an r-parameter Lie function
group.
Each u, defines a PB operator U. If a set of r constants of motion,
uj , provides a set of PB operators, Uj , that are linearly independent and
closed under commutation, they generate an r-parameter Lie transforma-
tion group. For the given value of E, the operations of this dynamical group
convert solutions of Hamilton’s equations into solutions. The group is the
classical mechanical analog of a quantum mechanical degeneracy group,
whose operations interconvert wave functions of the same energy.
As illustrated below, knowing the constants of motion, u, of a system
enables one to solve problems involving systems with Hamiltonians H
=
H+f(u). Constants of motion and the groups they determine can also imply
a variety of revealing connections between the dynamics of a system and
its geometrical properties in phase space.
Let
s = (s1 , . . . , sn , sn+1, . . . , s2n ) = (q1 , . . . , qn , p1 , . . . , pn ), (7.5.5)
be a vector whose components are conjugate q,p variables in this phase–
space. The components may be, but are not necessarily, Cartesian. Suppose
that s is a vector from the origin to a point ρs in the Hamiltonian hyper-
surface defined by H(p, q) = E. If uk = µk (s) is a constant of motion, then
at ρs each group generator, Uk = {uk , ·}, and infinitesimal parameter, δαk ,
determines an infinitesimal displacement, δsk = δαk Uk s, defined by
s → s + δαk Uk s + O(δα2k ) (7.5.6)
This may be imagined to lie in the hypersurface defined by H(q,p) =E. If
one lets ∆sk = Uk s, the components of ∆sk are just the functions ξkm (s)

in the generator Uk = m ξkm (s) ∂/∂sm :
∆sk = (ξk1 , . . . , ξkn , ξkn+1 , . . . , ξk2n ). (7.5.7)
The finite vector ∆sk is tangent to the hypersurface at ρs .
October 28, 2010 16:55 9in x 6in b943-ch07

222 Dynamical Symmetry

Because the generators are linearly independent, the ∆sk are linearly
independent. They may be used to establish a local coordinate system at
ρs . If, for a fixed value of s, the tangent vectors ∆sk associated with each
generator, including s itself, all transform as Cartesian vectors, then the
∆sk may be transferred to the origin of s and used to set up a Cartesian
coordinate system there. Evaluating these vectors at a particular time, t0 ,
provides a set of fixed vectors to which motions of ρs may be referred.
When the constants of motion uj transform as components of a Carte-
sian vector, they can also provide a fixed Cartesian coordinate system.
When the coordinate system provided by these fixed vectors, or by the
tangent vectors, is Cartesian, one is best able to use one’s geometric intu-
ition to guide investigations of the dynamics governed by the equations of
motion.
Harmonic oscillators with Hamiltonian

H = (p21 + q21 + p22 + q22 )/2 = H1 (q1, p1 ) + H2 (q2 , p2 ) (7.5.8a)

provide examples of the foregoing general remarks. The physical system

with this Hamiltonian may be considered to be a mass moving in the q1 q2
Euclidean plane, or two masses moving on a line. In either case, Hamilton’s
equations of motion are

dq1 /dt = p1 , dp1 /dt = −q1 , (7.5.8b)

dq2 /dt = p2 , dp2 /dt = −q2 . (7.5.8c)

One may assume q1 , q2 , p1 , p2 to be coordinates in a four-dimensional

phase space, and define Cartesian vectors

s = (q1 , q2 , p1 , p2 ) = (s1 , s2 , s3 , s4 ) (7.5.9a)

in the space. If s is a radius vector that stretches from the origin to a point
defined by H = E, then, for the Hamiltonian (7.5.8), all such points satisfy
the relation

(s1 , s2 , s3 , s4 )·(s1 , s2 , s3 , s4 ) = 2E, (7.5.9b)

√
and the length of s is 2E. For each √ value of E, the equation H = E then
defines an S(3) hypersphere of radius 2E. The surface of each such hyper-
sphere is a three-dimensional manifold, a space allowing three orthonor-
mal local Cartesian coordinates at each of its points. Though the manifold
defined by H − E = 0 has O(4) symmetry, we have yet to determine the
dynamical symmetries of the system.
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 223

The solutions of (7.5.8b,c) depend upon the four initial values of the q’s
and p’s at t0 = 0; q1 (0), p1 (0), q2 (0), and p2 (0). They are

q1 = q1 (0)cos(t) + p1 (0)sin(t), p1 = p1 (0)cos(t) − q1 (0)sin(t),

q2 = q2 (0)cos(t) + p2 (0)sin(t), p2 = p2 (0)cos(t) − q2 (0)sin(t).
(7.5.10)

The simple relations of (7.5.8) and (7.5.10) imply a host of geometrical

and dynamical connections. These are the subject of the remainder of this
section.
In the four-dimensional Euclidean space of (q1 , q2 , p1 , p2 ), there are
six mutually orthogonal two-planes that pass through the origin — the
planes of (q1 , q2 ), (p1 , p2 ), (q1 , p1 ), (q2 , p2 ), (p1 , q2 ), and (p2 , q1 ). For
given initial values, Eqs. (7.5.10) determine the motion of points in each
of these two-planes. One may project the motion in (q1 , p1 ) and (q2 , p2 )
onto a common q–p plane. Figure 7.5.1 depicts the result if q1 (0) = 1,
p1 (0) = 4, q2 (0) = 2, and p2 (0) = 3.
To determine the constants of motion u(p, q) of the oscillator and their
implications, we seek functions u(p, q) annihilated by the PB operator of the
Hamiltonian (7.5.8a). In this case, one procedure is to solve the equations
that result when u(q, p) is expanded in power series in the p’s and q’s.

q2

q1
−4 −2 2 4

−1

−2

−3

Fig. 7.5.1 Projection to trajectories of isotropic oscillator onto position space.

October 28, 2010 16:55 9in x 6in b943-ch07

224 Dynamical Symmetry

One lets


u= cm 1 n 1 m 2 n 2 qm 1 n 1 m2 n 2
1 p1 q2 p2 , (7.5.11a)
and requires


cm1 n1 m2 n2 {H, qm n1 m2 n2
1 p1 q2 p2 } = 0.
1
(7.5.11b)

The solving of (7.5.11b) is investigated in the exercises. One finds it has

no solutions that are linear combinations of the p’s and q’s. It has three
independent solutions that are quadratic functions of the positions and
momenta, functions that, together with H, are constants of motion. Letting
S1 = (q1 , p1 ), and S2 = (q2 , p2 ), these constants of the motion may be
chosen to be
(H1 − H2 ) = (S1 , ·S1 − S2 , ·S2 )/2, (7.5.12a)
(p1 p2 + q1 q2 ) = S1 · S2 , (7.5.12b)
and the vector S1 × S2 , with value (q1 p2 − p1 q2 ). If the coordinates are
those of a single particle, then
S1 × S2 = L12 , (7.5.12c)
i.e. its angular momentum.
In determining the Lie algebras defined by the constants of motion, it
is convenient to begin by multiplying the functions in (7.5.12) by constants
a, b and c respectively. The constants can then be chosen so as to put the
Lie algebra in an easily recognizable form. In this way, one finds that the
three functions
u1 = (p1 p2 + q1 q2 )/2, u2 = (H1 − H2 )/2, u3 = (q1 p2 − q2 p1 )/2,
(7.5.13a)
satisfy the Poisson Bracket relations
{u1 , u2 } = −u3 , {u3 , u1 } = −u2 , {u2 , u3 } = −u1 . (7.5.13b)
One also finds that
u21 + u22 + u23 = (H/2)2 . (7.5.13c)
The initial values of these constants of motion are
(p1 (0)p2 (0) + q1 (0)q2 (0))/2, (E1 − E2 )/2, L12 (0)/2, E2 /4. (7.5.14)
Adding any of the constants of motion to H gives a new Hamiltonian
function H
. The motions determined by H
may be obtained from those
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 225

determined by the original Hamiltonian. In the simplest case,

H
= H + cu1 = c1 H1 + c2 H2 , (7.5.15a)
with
c1 = 1 + c/2, c2 = 1 − c/2. (7.5.15b)
This eliminates the energy degeneracy E1 = E2 , without altering the form
of the solutions of (7.5.11). One may (see below) determine the motions
when H
is H + cu2 or H + cu3 by making use of transformations generated
by the PB operators of u1 , u2 and u3. These are the linearly independent
operators
U1 = {u1 , ·} = (−1/2)(p2 ∂/∂q1 + p1 ∂/∂q2 − q2 ∂/∂p1 − q1 ∂/∂p2 ),
U2 = {u2 , ·} = (−1/2)(p1 ∂/∂q1 − p2 ∂/∂q2 − q1 ∂/∂p1 + q2 ∂/∂p2 ),
U3 = {u3 , ·} = (−1/2)(q1 ∂/∂q2 − q2 ∂/∂q1 + p1 ∂/∂p2 − p2 ∂/∂p1 ).
(7.5.16)
They satisfy the commutation relations of the commutator Lie algebra
[U1 , U2 ] = −U3 , [U3 , U1 ] = −U2 , [U2 , U3 ] = −U1 , (7.5.17a)
with Casimir invariant
U21 + U22 + U23 . (7.5.17b)
The operators U1 , U2 , U3 each generate a one-parameter group of trans-
formations, exp(αk Uk ), which produces invariance transformations of the
equations of motion, each necessarily a one-parameter group of symplec-
tic transformations. The commutation relations establish that the resulting
three-parameter group is a linear symplectic group Sp(2).
If an isotropic oscillator with Hamiltonian H is affected by an outside
source that gives it angular momentum, c (q1 p2 − q2 p1 ), then H
can be
expressed as
H
= H + 2cu3 . (7.5.18a)
The resulting trajectories in the q1 p1 and q2 p2 subspaces are no longer
independent. However, acting on H
with the group operator exp(α2 U2 )
one finds
H
= H + 2c(u3 cos(α2 ) + u1 sin(α2 )). (7.5.18b)
These become Hamiltonians of the form (7.5.15a) if α2 = π.
October 28, 2010 16:55 9in x 6in b943-ch07

226 Dynamical Symmetry

All of the {U, ·}’s necessarily commute with the evolution generator
{−H, ·} = {2u4 , ·} = p1 ∂/∂q1 + p2 ∂/∂q2 − q1 ∂/∂p1 − q2 ∂/∂p2 . (7.5.19)
The group generated by {−H, ·} could, in the abstract, be any one-
parameter Lie group, but as we know, it generates symplectic transfor-
mations, the group is always a realization of Sp(1). In analogy to U1 , U2
and U3 , we defined U4 to be {−H, ·}/2. The group generated by the four
U’s is then Sp(2) × Sp(1). The quadratic Casimir invariant of the Sp(2)
groups is equal to the square of the generator of the Sp(1) group. It is the
symplectic group Sp(2) × Sp(1) that defines dynamical symmetries of the
oscillator.
As noted earlier, the Hamiltonian (7.5.8a) has evident four-dimensional
rotational symmetry. Let us investigate the relationship between its SO(4)
symmetry group and the Sp(2) × Sp(1) symplectic group uncovered to this
point. To this end, we let
Jij = si ∂/∂sj − sj ∂/∂si , (7.5.20)
be the generator of rotations in the i–j plane of the four-space in which the
Hamiltonian is defined. Six of the Jij are linearly independent. Together they
generate rotations of the SO(4) group that leaves invariant the Hamiltonian
(7.5.8a). Inspecting (7.5.16) and (7.5.18) one finds that
U1 = −(J14 + J23 )/2, U2 = −(J24 + J31 )/2, U3 = (J34 + J12 )/2,
(7.5.21a)
and
U4 = (J31 − J24 )/2. (7.5.21b)
Note that each U is composed of two rotation generators that commute.
Equations (7.5.21) allow one to connect the operations of the symplectic
group Sp(2) × Sp(1) generated by the U’s, to operations of the group that
leaves H invariant.
In addition to these U’s, there are two linear combinations of the rota-
tion generators (7.5.20), which annihilate H but do not generate canonical
transformations. These are
V2 = (J14 − J23 )/2, V3 = (J34 − J12 )/2. (7.5.22)
Let us re-label U4 , which equals {−H, ·}/2, and call it V1 . Then
[V1 , V2 ] = −V3 , [V3 , V1 ] = −V2 , [V2 , V3 ] = −V1 . (7.5.23)
Though V2 H = 0 and V3 H = 0, the operators V2 , V3 generate neither
symplectic transformations nor transformations that convert solutions of
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 227

the equations of motion (7.5.8b,c) into solutions. However, they do commute

with the three U’s in (7.5.21a) that generate the invariance group of these
equations!
As each U is a linear combination of two commuting rotation generators,
each generates finite transformations that are the product of two rotation
transformations, and one has

exp(α1 U1 ) = exp((−α1 /2)J23 ) exp(−(α1 /2)J14 ), (7.5.24a)

exp(α2 U2 ) = exp(−(α2 /2)J31 ) exp(−(α2 /2)J24 ), (7.5.24b)

and

exp(α3 U3 ) = exp((α3 /2)J12 ) exp((α3 /2)J34 ), (7.5.24c)

where each αk is a group parameter. The evolution operator is

exp(t{−H, ·}) = exp(2tU4 ) = exp(tJ31 ) exp(−tJ24 ). (7.5.24d)

In SO(4), the operator of each one-parameter rotation subgroup with gen-

erator Jij and group parameter γij is expressed as exp(γij Jij ), and in each
case the parameter ranges over an interval of 2π before the transformation
repeats. In contrast, the finite transformations generated by the Sp(2) gen-
erators U1 , U2 , and U3 return a point in the phase space to itself only after
the group parameters αk range through 4π. For the same reason, t in the
operator exp(tU4 ) must range through 4π to return a point in the phase
space to itself.
The four generators U have the following action on s:
U1 S = ( − p2 /2, −p1 /2, q2 /2, q1 /2) = ( − s4 /2, −s3 /2, s2 /2, s1 /2) ≡ ∆s2 ,
U2 S = ( − p1 /2, p2 /2, q1 /2, −q2 /2) = ( − s3 /2, s4 /2, s1 /2, −s2 /2) ≡ ∆s1 ,
U3 S = (q2 /2, −q1 /2, p2 /2, −p1 /2) = (s2 /2, −s1/2, s4 /2, −s3 /2) ≡ ∆s3 ,
U4 S = ( − p1 /2, −p2 /2, q1 /2, q2 /2) = ( − s3 /2, −s4 /2, s1 /2, s2 /2) ≡ ∆s4 .
(7.5.25a–d)
At a point ρs on the three-dimensional hypersurface of the hypersphere,
the linearly independent vectors δsk = εUk s can provide axes for a local
coordinate system determined by the constants of motion. The correspond-
ing finite tangent vectors ∆sk , k = 1, 2, 3, provide local systems of coordi-
nates in the four-dimensional space. The infinitesimal version of ∆s4 points
in the direction in which ρs will evolve. √ √
The length of each of the the vectors ∆s1 , ∆s2 and ∆s3 is E/ 2.
The vector dot products with each other are all zero, so they are mutually
October 28, 2010 16:55 9in x 6in b943-ch07

228 Dynamical Symmetry

orthogonal. The rotation generator in the phase space is L12 = {L12 , ·} =

{2u3 , ·}. As L12 ∆s1 = ∆s2 and L12 ∆s2 = −∆s1 , the operator exp(α3 L12 )
carries out rotations in the plane defined by ∆s1 , ∆s2 , rotations about an
axis along
√ ∆s √ at√every point ρs the unit
√ 3 . Consequently, √ tangent
√ vectors
i(s) = ( 2/ E)∆s1 , j(s) = ( 2/ E)∆s2 , and k(s) = ( 2/ E)∆s3 , pro-
vide the basis for a local Cartesian coordinate system which can be used to
describe infinitesimal displacements in the hypersurface. The vector ∆s4 ,
has the following components on these axes:
√ √ √
on i : S1 · S2 /( 2E), on j : (E1 − E2 )/( 2E), on k : L12 /( 2E).
(7.5.26)
It follows that the orientation of trajectories on the hypersurface is deter-
mined by the three constants of motion whose value is fixed by the initial
values of the positions and momenta as per (7.5.6). For a given value of
s, say s0 , unit vectors I0 , J0 and K0 that define a set of axes parallel
to i(s0 ), j(s0 ) and k(s0 ) can be set up at the center of the
√ hypersphere.
The initial position vector, s0 , when divided by its length, 2E, provides a
fourth unit vector, h0 , orthogonal to all the others. To sum up, the three
constants of motion have determined PB operators whose tangent vectors,
together with s0 , define a natural global Cartesian coordinate system in
phase space.
Figure 7.5.2a depicts the unit vectors i, j of a local axis system tan-
gent to the surface of an ordinary sphere at s0 the terminus of the radius
vector s0 . This set of axes is analogous to the axes at the surface of our
hypersphere. The unit vector h lies along the extension of the radius vector
to the point. In Fig. 7.5.2b, I, J and h axes that are parallel to the i, j
and h axes of Fig. 7.5.2a have been set up at the center of the sphere.
The components of any radius vector s on the I, J and h axes are clearly
identical with its components on the h, i and j axes.
The operator exp(α1 U1 ) moves a point ρso on the hypersphere defined
by (7.5.9b) to a one-parameter family of points, each of which may be
considered a new initial point. Evolving these with exp(α4 U4 ) one obtains a
one-parameter family of trajectories in phase space, trajectories that depict
solutions to the equations of motion, (7.5.8b,c). Because [U4 ,U1 ] = 0, the
result is the same as that obtained if the initial point is first evolved, then
transformed by exp(α1 U1 ). The operators exp(α2 U2 ) and exp(α3 U3 ) act
analogously, altogether generating a three-parameter family of trajectories
and solutions. Because the three Uk commute with H, all of the solutions
have the same energy. It can be proved that this three-parameter family
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 229

Fig. 7.5.2 (a) Radius vector Sa and unit tangent vectors. (b) Tangent vectors
translated to origin of Sa .

contains all the solutions of the equations of motion which have a common
energy.
In the foregoing we have used the Lie generators {uj , ·} to define coor-
dinate systems and motions in the four-dimensional phase space. An alter-
native approach is suggested by the fact that {u3 , ·} is a rotation generator
divided by two. Relations (7.5.8) imply that

{u3 , u1 } = u2 , {u3 , u2 } = −u1 , (7.5.27)

as would be expected if {u3 , ·} generates rotations of a Cartesian two-vector

[u1 , u2 ]. However,

exp(α3 {u3 , ·})u1 = cos(α3 /2)u1 + sin(α3 /2)u2 ,

(7.5.28)
exp(α3 {u3 , ·})u2 = cos(α3 /2)u2 − sin(α3 /2)u1 .

Because the group parameter α3 must range through 4π in order to cycle

u1 and u2 back to their initial point, the operation of exp(α3 {u3 , ·}) may
be considered to move a point with coordinates u1 , u2 , around the girdle
of a Möbius band. The same argument can be applied to the action of
exp(α2 {u2 , ·}) on u3 , u1 , and that of exp(α1 {u1 , ·}) on u2 , u3 . The space
of u1 , u2 , u3 is not Euclidean, and u1 , u2 , u3 cannot be considered to be
October 28, 2010 16:55 9in x 6in b943-ch07

230 Dynamical Symmetry

components of a vector in a Euclidean three-space. Furthermore, though

u21 + u22 + u23 = (H/2)2 = (E/2)2 , (7.5.29)

the manifold defined by this equation is not a sphere. To sum up: the
constants of motion u1 , u2 , and u3 do not provide a coordinate system
as intuitively useful as that provided by the tangent vectors of the PB
operators they determine. In Chapter 8, we will find that just the reverse
is true in Kepler systems.

7.6 Hamilton’s Equations in Classical PQET Space;

Conservation Laws Arising From Galilei Invariance
In this section, time t will be considered to be a coordinate, rather than
a parameter. It is shown in Appendix B that the momentum variable con-
jugate to t is −E, the negative of an energy. The resulting phase space
is termed PQET space. When Hamiltonian mechanics is extended to this
space one replaces t by an evolution parameter, which we shall denote by τ .
It is an independent variable that ranges through all the real numbers, and
is used to define the Lagrangian function that determines Hamiltonian func-
tions in PQET space. In nonrelativistic Hamiltonian mechanics in PQET
space it is presupposed that, though different particles have different P, Q
coordinates, the coordinates pj and qj , may all be observed simultaneously
at time t. The corresponding variable E is represents the total energy of a
system. With these presuppositions, there is only one t coordinate and one
E coordinate. The Poisson bracket, {A, B}, of two functions A(P,Q,E,t),
B(P,Q,E,t) is therefore defined by
n
1 (∂A/∂qj ∂B/∂pj −∂B/∂qj ∂A/∂pj ) − (∂A/∂t ∂B/∂E − ∂B/∂t ∂A/∂E).
(7.6.1a)

It is often convenient to set t = q0 and −E = p0 . When this is done,

n
{A, B} = 0 (∂A/∂qj ∂B/∂pj −∂B/∂qj ∂A/∂pj ). (7.6.1b)

For a Hamiltonian system defined in PQET space, the Hamiltonian func-

tion, H, may, as in Appendix B, be derived from a time-dependent
Lagrangian. Let

W(P, Q, E, t) ≡ H(P, Q, t) − E. (7.6.2a)

October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 231

Then Hamilton’s equations of motion in PQET space are

dqj /dτ = ∂W/∂pj , dpj /dτ = −∂W/∂qj ,

(7.6.2b–e)
dt/dτ = ∂W/∂(−E), d(−E)/dτ = −∂W/∂t.

If f = F(P,Q,E,t), one has the general relation

df/dτ = {f, W}. (7.6.3)

Applying this to a system with a time-independent Hamiltonian, one

obtains
dqj /dτ = ∂W/∂pj , dpj /dτ = −∂W/∂qj
(7.6.4a–d)
dt/dτ = 1, dE/dτ = 0.

It follows that ∆t = ∆τ , so if the motion is not periodic

t = τ + const (7.6.4e)

and the motion is periodic,

t − const = (τ )modT , (7.6.4f)

where T is the period. This case was studied in Sec. 3.3.

Lagrange’s equations of motion inherit the Galilei invariance of
Newton’s equations of motion for isolated systems of particles. Let us inves-
tigate the manner in which this invariance is passed on to Hamilton’s equa-
tions of motion in PQET space. Using (7.4.3), one finds that the generators
of the passive group given in (7.1.8) become the following PB operators in
PQET space:

T0 = ∂/∂t → {−E, ·}, (7.6.5a)

Tx = ∂/∂qx → {−px , ·}, etc., (7.6.5b)

and

Rxy = qx ∂/∂qy − qy ∂/∂qx

→ {−(qx py − qy px ), ·} = (qx ∂/∂qy − qy ∂/∂qx )
+ (px ∂/∂py − py ∂/∂px ), etc. (7.6.5c)

To express the active transformations determined by these operators for

a system with n q’s, and n p’s, define the position and momentum vectors
October 28, 2010 16:55 9in x 6in b943-ch07

232 Dynamical Symmetry

of particle i by

qi = (qxi , qyi , qzi ),

pi = (pxi , pyi , pzi ). (7.6.6)

The total momentum vector is p = pi . The operators

∇qi = (∂/∂qxi, ∂/∂qyi , ∂/∂qzi), ∇pi = (∂/∂pxi /∂pyi /∂pzi ), (7.6.7)

generate active translations of the position and momentum of particle i in

PQET space. The corresponding many-particle operators are



∇q = ∇qi , ∇p = ∇pi . (7.6.8)

These, together with p, enable one to express the effects of active Galilei
transformations on a many-particle system.

a. It follows immediately from (7.6.5a) that

{H − E, E} = 0 if ∂H/∂t = 0, (7.6.9a)

the energy of a system is conserved if its Hamiltonian is time-translation

invariant.
b. Equation (7.6.5b) implies that

{H, p} = 0 if ∇q H = 0, (7.6.9b)

i.e. the total momentum of a system is conserved if its Hamiltonian is

space-translation invariant.

c. The total angular momentum of the system is L = qi x pi , and the
vector operator that generates the active rotations that result from rota-
tions of the coordinate system is R = {L, ·}. Equation (7.6.5c) requires
that

{H, L} = 0 if RH = 0, (7.6.9c)

i.e. the angular momentum of a system is conserved if its Hamiltonian

is rotation invariant.

Though Newton’s equations are invariant under the passive transforma-

tion v → v + v0 , the same is not true of Hamilton’s equations in PQ space
because kinetic energies are not invariant under the transformation. As the
operators Kx , Ky , Kz depend upon t, they determine operators that act in
PQET space. For this reason one must formulate the Hamiltonian analog
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 233

of Newtonian Galilei transformations as transformations in PQET space.

For

Kx = t∂/∂qx + ∂/∂vx ,

one finds

Kx → {(−tpx + mqx ), ·} = t ∂/∂qx + px ∂/∂E + m ∂/∂px , (7.6.10a)

etc. These components of the vector operator K do not annihilate p · p/2m.

However, H – E is invariant under transformations generated by an operator
of active transformations closely related to K. To see this, let

Ki = −tpi + mi qi , (7.6.10b)

and


u=− Ki . (7.6.10c)
 
Now, let m = mi , then ρ ≡ mi qi /m is the vector from the origin to

the center of mass of the system. Also let tpi =tP, where P is the total
momentum vector of the system. Then

u = −mρ + tP. (7.6.10d)

The PB operator determined by u is

U = {u, ·} = (pi ∂/∂E + t∇qi + mi ∇pi ). (7.6.10e)

The components of U, PB operators that one may denote by Ux , Uy and

Uz , are the active forms of the generators Kx , Ky and Kz , of the Galilei
group. U generates the transformations

exp(v0 · U)qi = qi + v0 t = exp(v0 · K)qi , (7.6.11a)

and

exp(v0 · U)pi = pi + v0 mi . (7.6.11b)

Thus exp(v0 ·U) has the same effect as exp(v0 ·K) on q and v. One finds
that



exp(v0 · U)E = E + v0 · pi + (1/2) mi v0 · v0 , (7.6.11c)

and




exp(v0 · U) pi ·pi /2mi = v0 · pi + (1/2) mi v0 ·v0 . (7.6.11d)
October 28, 2010 16:55 9in x 6in b943-ch07

234 Dynamical Symmetry

Consequently, if




H= pi · pi /2mi + v(|qi − qj |), (7.6.12a)
then
U(H − E) = 0. (7.6.12b)
Thus, though observers in systems moving at constant velocity with respect
to each other will assign different values to E and H, they will agree that
H = E.

7.7 Time-dependent Constants of Motion;

Dynamical Groups That Act Transitively
A function G(p,q,E,t) defined in PQET space, is a time-dependent con-
stant of motion in a system with Hamiltonian H(p,q) if
0 = {H − E, G}H−E=0 . (7.7.1a)
Now
{E, ·} = ∂/∂t, (7.7.1b)
so (7.7.1a ) is equivalent to requiring that
{H, ·}G − ∂G/∂t = 0, (7.7.1c)
on the manifold in PQET space defined by H(q, p) − E = 0. The function u
of the previous paragraph is an example of a time-dependent vector constant
of motion. If H–E in (7.7.1a) is replaced by (H–E)f, with f any C2 function
of the dynamical variables, then a function G that satisfies (7.7.1a) will also
satisfy
0 = {(H − E)f, G}|H−E=0 . (7.7.1d)
This turns out to be a useful property of Hamiltonian systems and constants
of motion in PQET space.
Harmonic oscillators provide one of the simplest examples of physically
important time-dependent constants of motion. To investigate them for a
one-dimensional oscillator, suppose that H = (p2 + q2 )/2. Then
{H, ·} = q∂/∂p − p∂/∂q. (7.7.2)
Let u(p,q,E,t) be a prospective constant of motion, and expand it in a power
series. Setting


u= qa pb gab (t, E), (7.7.3)
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 235

one requires that

{H − E, u} = (q∂/∂p − p∂/∂q)u + ∂u/∂t = 0 (7.7.4)
when H = E. The first term in the expansion of u that gives a nontrivial
result is u1 = q g10 (t,E) + p g01 (t,E). The equation that u1 must satisfy is
(q∂/∂p − p∂/∂q)u1 − ∂u1 /∂t
= (g01 − ∂g10 /∂t)q − (g10 + ∂g01 /∂t)p = 0. (7.7.5a)
Because q, p are independently variable, (7.7.5a) requires that
g01 = ∂g10 /∂t, ∂g01 /∂t = −g10 . (7.7.5b)
These equations have two linearly independent solutions. To simplify the
subsequent transformation of our results to quantum mechanics, we allow
these solutions to be complex, and choose them to be
g10+ = exp(it), g10− = exp(−it). (7.7.6a)
Then
g01+ = iexp(it), g01− = −iexp(−it). (7.7.6b)
These yield the constants of motion
(q + ip)exp(it) ≡ u+ (7.7.7a)
and
(q − ip)exp(−it) ≡ u− . (7.7.7b)
One has
{H, ·}u+ = ∂u+ /∂t = iu+ , (7.7.8a)
{H, ·}u− = ∂u− /∂t = −iu− . (7.7.8b)
The two time-dependent constants of motion satisfy
{u+ , u− } = 2i. (7.7.9)
The PB operators obtained from them are
U+ = {u+, ·} = exp(it)(∂/∂p − i∂/∂q − i(q + ip)∂/∂E), (7.7.10a)
and
U− = {u− , ·} = exp(−it)∂/∂p + i∂/∂q + i(q − ip)∂/∂E). (7.7.10b)
When acting only in PQ space,
U+ → exp(it)(∂/∂p − i∂/∂q), U− → exp(−it)(∂/∂p + i∂/∂q).
(7.7.11a,b)
October 28, 2010 16:55 9in x 6in b943-ch07

236 Dynamical Symmetry

In both PQ and PQET space,

[U+ , U− ] = 0. (7.7.12)

Thus U+ and U− commute. Since {1, ·} = 0, this result is compatible with

relations (7.4.8).
Integrating the infinitesimal transformations defined by the generators
U+ and U− , (cf. Sec. 4.5), one finds the finite transformations produced by
the operators exp(α± U± ):

p → p + α+ exp(it), q → q − iα+ exp(it),

(7.7.13a)
E → E − iα+ u+ , H → H − iα+ u+ ;

p → p + α− exp(−it), q → q + iα− exp(−it),

(7.7.13b)
E → E + iα− u− , H → H + iα− u− .

At t = 0, the transformations of p, q and E become

p0 → p0 + α+ , q0 → q0 − iα+ , E0 → E0 − iα+ (q0 + ip0 ); (7.7.14a)

p0 → p0 + α− , q0 → q0 + iα− , E0 → E0 + iα− (q0 − ip0 ). (7.7.14b)

Equation (7.7.12) states that U+ and U− commute, so

exp(α+ U+ )exp(α− U− ) = exp(α+ U+ + α− U− ). (7.7.15)

To express the right-hand side of (7.7.15) in terms of real variables we let

the linear combination of PB generators be re-expressed as

α+ {(q + ip)exp(it), ·} + α− {(q − ip)exp(−it), ·}. (7.7.16)

Define

α = α+ + α− , β = i(α+ − α− ), (7.7.17a)

uα = q cos(t) − p sin(t), uβ = q sin(t) + p cos(t), (7.7.17b)

and

Uα = {uα , ·}, Uβ = {uβ , ·}. (7.7.17c)

Then

exp(α+ U+ + α− U− ) = exp(αUα + βUβ )

(7.7.18)
= exp(αUα )exp(βUβ ) = T(α, β).
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 237

T(α, β) carries out real transformations that may be obtained from the
composition of the two transformations (7.7.13). Evaluating the action of
T(α, β) at t = 0, one finds that it carries out the conversions
p0 → p0 + α, q0 → q0 + β, E0 → E0 + αp0 + βq0 + (α2 β 2 )/2. (7.7.18)
Using T(α, 0) and T(0, β), one is able to independently shift the two
initial values q0 and p0 , so the operations of the group can convert any
solution of the equations of motion into any other solution. Using the defi-
nition of transitivity in Chapter 5, the group is seen to act transitively on
the space of solutions.
The generators Uα and Uβ satisfy the commutation relations
[{H, ·}, Uα ] = Uβ , [{H, ·}, Uβ ] = Uα , [Uα , Uβ ] = 0. (7.7.19)
The corresponding group, denoted N(3), is called the Heisenberg group.
To apply the results of this section to the oscillator of Sec. 7.6, with
H = H1 + H2 , as defined in (7.5.8a), one uses operators T(α1 , β1 ) and
T(α2 ,β2 ) to carry out the operations of (7.7.19) in the subspaces of p1 ,
q1 , E1 and p2 , q2 , E2 . This enables one to independently change each of
the four initial values p1 (0), q1 (0), p2 (0), q2 (0), and thereby change E1 ,
E2 and E. The operations of the two Heisenberg groups carried out by
T(α1 ,β1 ),T(α2 ,β2 ), combined with the operations of the degeneracy group,
produce a subgroup of Sp(3) that acts transitively on the space of solutions
of the oscillator equations of motion (7.5.8b,c).
For classical harmonic oscillators, knowing a dynamical group that acts
transitively is of little practical value. For more complex systems, knowing
analogous groups can be very revealing, and of great utility — even if the
required group generators are only approximately known. For this reason,
as well as conceptual reasons previously set forth, we will often be working
with Hamiltonian dynamics and symmetries in PQET space, or with their
quantum mechanical analogs.
Subsequent chapters will also make much use of the PB operators
defined in this chapter, the commutator Lie algebras they define, and the
corresponding Poisson bracket Lie algebras. Chapter 9 establishes quantum
mechanical versions of a number of results contained in this chapter.

7.8 The Symplectic Groups Sp(2n,r)

The PB operators determined by quadratic functions of conjugate q’s and
p’s generate symplectic transformation groups that act linearly in phase
space. If there are n real q’s and n real p’s, these groups are realizations of
October 28, 2010 16:55 9in x 6in b943-ch07

238 Dynamical Symmetry

the group Sp(2n,r). The following Lie generators provide a basis for the Lie
algebra of Sp(2n,r) when i and j run from 1 to n:
{q2i /2, ·} = qi ∂/∂pi , {p2i /2, ·} = −pi ∂/∂qi ,
{qi pi , ·} = pi ∂/∂pi − qi ∂/∂qi , {qi pj , ·} = pj ∂/∂pi − qi ∂/∂qj ,
(7.8.1)
{qj pi , ·} = pi ∂/∂pj − qj ∂/∂qi , {qi qj , ·} = qi ∂/∂pj + qj ∂/∂pi ,
{pi pj , ·} = −(pi ∂/∂qj + pj ∂/∂qi ).
The commutator and PB Lie algebras obtained from the functions above
are isomorphic. Table 7.1 below gives Poisson bracket relations that are
sufficient to determine the Lie algebras of both types of symplectic groups.
Sp(2n,r) has n(2n + 1) generators and possesses the following subgroup
chains:2
Sp(2n, r) ⊃ U(n) ⊃ SU(n) ⊃ SO(n),
(7.8.2)
Sp(2n, r) ⊃ SL(n, r) × R ⊃ SL(n, r) ⊃ SO(n).
Writing 2n = 2a + 2b, with a and b integers, one also has
Sp(2a + 2b, r) ⊃ U(a, b) ⊃ SU(a, b) ⊃ SO(a, b),
(7.8.3)
Sp(2a + 2b, r) ⊃ Sp(2a, r) × Sp(2b, r).
Numerous group–subgroup chains of SO(n) and SO(a,b) are tabulated in
Ref. 2.
The group N(3) of the previous section is a symplectic group though the
functions determining that its PB generators are not quadratic in p,q,E,t.
Nonlinear transformations can convert the transformations of an
Sp(2n,r) group that acts linearly in phase space into those of Sp(2n,r)
groups that act nonlinearly. This enables Sp(2n,r) to take on a particu-
larly important role in Hamiltonian dynamics. An example of a nonlinear
realization of Sp(2n,r) will be found in Chapter 8.

Table 7.1. The Poisson Brackets {A, B}

A\B q2i /2 p2i /2 qi pi qi pj qj pi qi qj pi pj

q2i /2 0 qi pi q2i 0 qi q j 0 qi pj
p2i /2 0 −p2i −pi pj 0 −qj pi 0
qi pi 0 −qi pj qj pi −qi qj pi pj
qi pj 0 qj pj − qi pi −q2i p2j
qj pi 0 −q2j p2i
qi qj 0 qj pj +qi pi
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 239

7.9 Generalizations of Symplectic Groups That Have

an Infinite Number of One-parameter Groups
If one generalizes the concept of closure used in defining Lie groups so as
to allow an infinite set of generators, two different generalizations may be
used. One may allow either a countable or uncountable infinite set.
For transformation groups of the first type one requires
∞


[Ui , Uj ] = ckij Uk . (7.9.1)
k

An example is provided when one begins to compose transformations from

one-parameter groups generated by PB operators of polynomial functions
of the q’s and p’s of order greater than three.
Generalizations of the second type arise when generators U of one-
parameter groups depend upon a continuous parameter, say α. Varying the
parameter develops a one-parameter family of transformation groups. If

[U(α), U(α
)] = Φ(α, α
, α

)U(α

), α

= A(α, α
), (7.9.2)

the transformations are said to belong to an infinite group. If the U(α) are
Poisson bracket operators, the resulting group is a symplectic one. As an
example of this generalization, let u(α) = p exp(αq). Then

U(α) = exp(αq)(αp∂/∂p − ∂/∂q), (7.9.3)

and

[U(α), U(α
)] = (α − α
)exp((α α
)q)((α + α
)p∂/∂p − ∂/∂q)

= (α − α
)U(α α
).

In this case, the function A is (α + α
) and the function Φ is (α − α
).

The commutation relations (7.9.2) must be generalized when one is com-
posing one-parameter groups whose generators differ by more than a single
continuous parameter. This is necessary, for example, when one wishes to
compose the one-parameter groups generated by (7.9.3a) with those gener-
ated by {−p, ·} = ∂/∂q.
Though infinite dimensional symplectic groups will not be utilized in
subsequent chapters, infinite dimensional groups in general, play an impor-
tant role in the study of partial differential equations whose solutions
depend upon arbitrary functions rather than arbitrary constants.
October 28, 2010 16:55 9in x 6in b943-ch07

240 Dynamical Symmetry

Appendix A. Lagrange’s Equations and the Definition

of Phase Space
The equations of Lagrange and of Hamilton directly formulate the evolution
of kinetic energies, potential energies and work, rather than the evolution
of positions and momenta alone. Both treat Cartesian coordinates xi on
the same footing as generalized coordinates qj , related to the Cartesian
coordinates by C2 transformations

xi = Xi (q), q = (q1 , q2 , . . . , qn ). (7.A.1)

Using Cartesian coordinates, (x1 , x2 , x3 ), and velocity components vi =

dxi /dt, the kinetic energy of a mass point in an inertial Cartesian coordinate
system is m(v21 + v22 + v32 )/2. For a system with n particles and 3n, or 2n,
Cartesian coordinates, it is convenient to associate a mass mi with each
degree of freedom, keeping in mind that some of the mi have the same
value. If the masses are constant, and fi is the component of the force
acting on mi , Newton’s equations of motion are then

mi (dvi /dt) = fi , i = 1, 2 . . . . (7.A.2)

The total kinetic energy of the system is

T= mi vi2 /2. (7.A.3a)

To investigate the evolution of the kinetic energy under the action of forces,
one considers dT/dt. As


dT/dt = mi vi dvi /dt, (7.A.3b)

Newton’s equations imply

dT/dt = fi vi . (7.A.3c)

One then has the well known relation,

dT = fi dxi = dW, (7.A.3d)

where dW is the element of work done by all the forces during the displace-
ments dx.
Lagrange’s equations are obtained by determining the way in which
relations (7.A.3b–d) are altered when one replaces Cartesian coordinates
and velocities by generalized coordinates and their time derivatives.
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 241

The simplest derivation of Lagrange’s equations from Newton’s equa-

tions, appears to be that provided by Thompson and Tait.3 It begins with
the transformation of coordinates defined by (7.A.1), which implies that



dxi = (∂Xi /∂qj )dqj = (dxi /dqj ) dqj . (7.A.4a)
j j

This is used to express the element of work in (7A.3d) as

dW = fi (∂Xi /∂qj )dqj = Fj dqj , (7.A.4b)
i j j

where



Fj = fi ∂Xi /∂qj = fi dxi /dqj . (7.A.4c)
i i
Here, and hereafter, each of the generalized coordinates qj are considered
independently variable, so each Fj becomes the local component of the force
that acts in the direction of dqj .
The connection between the Fj and the fi is put to use in converting
the left-hand side of (7.A.2) to an expression analogous to (7.A.3c), one
in which the qj and their time derivatives replace the xi and their time
derivatives. To begin, for each value of i in Newton’s equations (7.A.2),
multiply both sides by dxi /dqj , to obtain
mi (dvi /dt)dxi /dqj = fi dxi /dqj . (7.A.5)
The term multiplying mi on the left-hand side of this may be re-expressed
as
d/dt(vi dxi /dqj ) − vi (d/dt)dxi /dqj . (7.A.6a)
Now (cf. the exercises) it can be shown that
dxi /dqj = (dxi /dt)/(dqj /dt). (7.A.6b)
On writing q
j for dqj /dt, (7A.6a) can thus be re-expressed as
(d/dt)(vi dvi /dq
j ) − vi dvi /dqj , (7.A.6c)
and as
(d/dt)(d(vi2 /2)/dq
j ) − (d(vi2 /2)/dqj ). (7.A.6d)
Using (7.A.6d) to replace the coefficient of mi in the left-hand side of
(7.A.5), Eq. (7.A.5) becomes

 
d(vi2 /2) 2
mi (d/dt) − d(vi /2)/dqj = fi dxi /dqj . (7.A.7)
dq
j
October 28, 2010 16:55 9in x 6in b943-ch07

242 Dynamical Symmetry

Summing this over i yields



(d/dt)(d(mi v2 /2)) d(mi v2 /2)
i i
− = Fj . (7.A.8a)
i
dq
j dqj

Thus,
(d/dt)∂T/∂q
j − ∂T/∂qj = Fj . (7.A.8b)
These equations, Lagrange’s equations of the first kind, are valid for all
coordinate systems that can be defined by Eqs. (7.A.1). If the forces are
derivable from a potential, V(q), then
Fj = −∂V(q)/∂qj , (7.A.9a)
and, on defining
L(q, q
) = T(q, q
) − V(q), (7.A.9b)
(7.A.8b) becomes
d(∂L/∂q
j )/dt − ∂L/∂qj = 0. (7.A.9c)
These are Lagrange’s equations of the second kind.
To utilize Lagrange’s equations it is necessary to express T as a function
of the q’s and their time derivatives. Using (7.A.1) one finds



T = (1/2) gjk (q)q
j q
k , (7.A.10)
j k

with


gjk (q) = mi ∂Xi /∂qj ∂Xi /∂qk .
i

The variables,
pj ≡ ∂T/∂q
j = ∂L/∂q
j , (7.A.11)
are the canonical momenta of the system.
We next show that the pj and qj are, in fact, members of canonically
conjugate sets of variables. Let the Cartesian positions and momenta be
denoted by χk and πk , with πk = mk vk . Using Poisson brackets defined
using derivatives with respect to χk and πk , one has
{qi , pj } = {qi , ∂T/∂q
j}, (7.A.12a)



= {qi , ∂ (mk vk2 /2)/∂q
j } = {qi , mk vk dvk /dq
j }. (7.A.12b)
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 243

Hence


{qi , pj } = {qi , πk dvk /dq
j }. (7.A.12c)

As in (7.A.6b), one has

dvk /dq
j = dχk /dqj . (7.A.12d)

Consequently,


{qi , pj } = {qi , πk dχk /dqj }. (7.A.12e)

Because dχk /dqj can be expressed as a function of the χ’s only, for the
Poisson bracket on the right-hand side of (7.A.12e) one has



{qi , πk dχk /dqj } = {qi , πk }dχk /dqj
k k



= ( ∂qi /∂χm ∂πk /∂πm ) dχk /dqj . (7.A.12f)
k m

Only the terms ∂πk /∂πm with k = m are nonvanishing, so

{qi , pj } = ∂qi /∂χm dχm /dqj = dqi /dχm dχm /dqj = dqi /dqj .
m m
(7.A.12g)
Thus

{qi , pj } = δij . (7.A.13a)

Similar, but shorter, analyses establish that

{qi , qj } = 0 = {pi , pj }. (7.A.13b,c)

To sum up, for a transformation from Cartesian coordinates to gener-

alized coordinates, qj , the momenta pj conjugate to the qj are given by
(7.A.11), as claimed.

Appendix B. The Variable Conjugate to Time

in PQET Space
We wish to establish that in PQET space, the variable conjugate to t is −E.
Following Lanczos,4 we begin the argument with Hamilton’s variational
derivation of Lagrange’s equations. Let L(mj ,qj ,q’j ,t) be the Lagrangian
October 28, 2010 16:55 9in x 6in b943-ch07

244 Dynamical Symmetry

function of a system of masses mj , with position coordinates qj , and velocity

coordinates q
j = dqj /dt. The integral
t2
S≡ L(mj , qj , q
, t) dt, (7.B.1)
t1
has the dimensions of an action, and represents a total action developed
as the masses move between times t1 and t2 . Hamilton established that
Lagrange’s equations result if one requires that, as t evolves, the point
with coordinates qj , q
j , describes a path through the space of positions and
velocities which makes S an extremum. Put more precisely, the motions
are such that δS = 0 if the position functions q
j (t), and velocity func-
tions q
j (t) = dqj (t)/dt, are subjected to infinitesimal variations, δqj (t),
δq
j (t) = d(δqj (t))/dt, at all values of t between t1 and t2 . The variations
are restricted by the requirement that they vanish at t1 and t2 :
δqj (t1 ) = δqj (t2 ) = 0 = δq
j (t1 ) = δq
j (t2 ). (7.B.2)
Thus, one requires
t2
δS = (L(mj , qj + δq
j , q
j + δqj , t) − L(mj , qj , q
, t)) dt, (7.B.3a)
t1
t2

= ((∂L/∂qj )δqj + (∂L/∂q
j )δq
j ) dt = 0. (7.B.3b)
t1
Integrating the second term by parts enables one to express this as

t2

δS = (∂L/∂q
j )δqj |tt21 + ((∂L/∂qj ) − (d/dt)(∂L/∂q
j ))δqj dt = 0.
t1
(7.B.4)
As (7B.2) requires that the variation vanishes at t1 and t2 , the first term
must vanish, one has
t2

δS = ((∂L/∂qj ) − (d/dt)(∂L/∂q
j ))δqj dt = 0. (7.B.5)
t1
The integral vanishes iff for each value of j, Lagrange’s equations
(d/dt)(∂L/∂q
j ) ∂L/∂qj = 0 (7.B.6)
are satisfied.
The momentum conjugate to each qj is pj = ∂L/∂q
j . If L is not a func-
tion of t, the Hamiltonian, H, corresponding to it is obtained by expressing
the function


H= q
j pj − L(mj , qj , q
j , t), (7.B.7)
as a function of the q’s and p’s.
October 28, 2010 16:55 9in x 6in b943-ch07

Dynamical Symmetry in Hamiltonian Mechanics 245

To determine the applications of this derivation to motions in PQET

space, we change the variable of integration in (7.B.1). The evolu-
tion parameter t is replaced by an evolution parameter τ , and t is
replaced by t
, a new independent dynamical variable. To this end we
let dt → (dt/dτ )dτ ≡ t
dτ , and replace d/dt by (1/t
)d/dτ . If we write
q

j = dqj /dτ , then q
j = q

j /t
. Supposing that L does not depend explicitly
upon t, and writing

L
(mj , qj , q

j /t
) = t
L(mj , qj , q

j /t
), (7.B.8a)

(7.B.1) becomes
τ2


S ≡ L
(mj , qj , q

j /t
)dτ. (7.B.8b)
τ1

The Lagrangian L
depends upon the generalized coordinates qj t
, and the

generalized velocities q

j . As L
is not a function of t, the variable conjugate
to t is a constant of motion. Using the S
and L
of (7.B.8), one finds
Eq. (7.B.6) replaced by
(d/dτ )(∂L
/∂q

j ) − ∂L
/∂qj = 0. (7.B.9)
The momenta conjugate to the q
j are p
j ≡ ∂L
/∂q

j , and the momentum
conjugate to t
is

p
t
≡ ∂L
/∂t
. (7.B.10)

Thus


p
t = L − t
((q

j /t
2 )∂L/∂q
j ) (7.B.11a)


=L− ((q

j /t
)∂L/∂q
j) (7.B.11b)


=L− q
j ∂L/∂q
j (7.B.11c)


=L− q
j pj (7.B.11d)

= −H → −H(q, p). (7.B.11e)

Consequently

p
t → −E, (7.B.11f)

and −E is the variable conjugate to t.

To define functions W(p, q, E, t) that are generalizations of H(p, q) −E,
one begins by defining a generalized kinetic energy that is allowed to be a
October 28, 2010 16:55 9in x 6in b943-ch07

246 Dynamical Symmetry

function of generalized coordinates q, t, and generalized velocities dq/dτ

and dt/dτ . One then defines a generalized Lagrangian T − V(q) and pro-
ceeds as in (7.B.10) to determine the generalized momenta conjugate to the
coordinates q and t. W(p, q, E, t) is then defined by the same process that
converts (7.B.11a) to (7.B.11e).

Exercises
1. In matrix notation, (7.2.3) may be expressed as ds
= M dsT , with
M composed from n x n submatrices A, B, C and E, with the upper
left sub-matrix, A, having elements aij , the upper right submatrix, B,
having elements bij , and the lower left and right sub-matrices, C, E
having elements cij , eij respectively. What relations between the matrix
elements are required by Eqs. (7.2.8)?
2. a) Produce plots, analogous to Fig. 7.5.1, of trajectories in a common
q–q plane and a common p–p plane. b) Plot motions implied by (7.5.10)
in each of the six two-planes.
3. Show that u1 ,u2 ,u3 satisfy Eq. (7.5.11b).
4. Expand (7.5.11b), collect terms that multiply the same powers of each
q and p, and determine the relations that the coefficients c must satisfy
when H is given by (7.5.8a).
5. Derive (7.2.11) from (7.2.9).
6. Derive (7.7.5a) from (7.7.4).

References
[1] J.-M. Levy-Leblond, Galilei Group and Galilei Invariance, in Group Theory
and it Applications, II, ed. E. M. Loebl (Academic Press, N.Y., 1971).
[2] R. Gilmore, Lie Groups, Lie Algebras, and Some of Their Applications
(Wiley-Interscience, N.Y., 1974).
[3] Sir W. Thomson and P. G. Tait, Treatise on Natural Philosophy (Cambridge
University Press, Cambridge, 1879), p. 302.
[4] C. Lanczos, The Variational Principles of Mechanics (University Toronto
Press, Toronto, 1949), p. 133.
October 28, 2010 16:55 9in x 6in b943-ch08

CHAPTER 8

Symmetries of Classical Keplerian Motion

In recent years it has become possible to turn Kepler’s intuitions of the

“music of the spheres” into a mathematically and physically sound under-
standing of the dynamical symmetries of classical Keplerian motion. A few
of their consequences were pointed out in Chapter 1. This chapter develops
their conceptual underpinnings.

8.1 Newtonian Mechanics of Planetary Motion

Let GmM/r2 be the magnitude of the gravitational force attracting a spher-
ically symmetrical planet of mass m, and the sun of mass M, with G
being the universal constant of gravitation, and r the distance between
the centers of mass of the two bodies. The reduced mass of the planet is
µ = Mm/(M + m). If one defines κ = GmM/µ = G(m + M), and r is the
vector from the center of mass of the sun to the center of mass of the planet,
then Newton’s equation for the motion of the center of mass of the planet
reduces to
dv/dt = −f(r)r, (8.1.1)
with
v = dr/dt, f(r) = κ/r3 .
Kepler’s three laws of motion are direct consequences of dynamical sym-
metries of this equation.
We first deduce the scaling symmetry that underlies Kepler’s law of
periodic times. Let T represent a measure of time, and R a position vector
of length R. Suppose t → T
= aT, R → R
= bR, and κ → κ
= cκ.
Equation (8.1.1) with T = t, R = r, is invariant under such a transforma-
tion if a2 c/b3 = 1. Because the masses of the planets are small compared

247
October 28, 2010 16:55 9in x 6in b943-ch08

248 Dynamical Symmetry

to the mass of the sun, κ is nearly the same for all, so one may in good
approximation set c = 1. For such systems, a2 /b3 = 1, and the time and dis-
tance scalings must produce the approximate relation (T
/T)2 = (R
/R)3 .
As the planetary orbits are all ellipses of small eccentricity, they may be
approximated by circles of radii R and R
. Letting T and T
be the period
of time it takes for two planets with orbital radii R and R
, to traverse such
a circle, one obtains Kepler’s law of periodic times:
(T
/T)2 = (R
/R)3 . (8.1.2)
A discussion of the accuracy of this relation will be found in Ref. 1 at the
end of the chapter. The exact law will be derived below.
We next consider time-independent properties of planetary orbits that
are consequences of two different vector constants of motion of Keplerian
systems. Subsequent sections of the chapter will investigate the way in
which these constants of motion are related to dynamical symmetries.
Taking the cross product of (8.1.1) with r, one obtains
r × dv/dt = −f(r) r × r = 0. (8.1.3)
Now
d(r × v)/dt = (v × v) + (r × dv/dt) = (r × dv/dt). (8.1.4a)
It then follows from (8.1.3) that
d(r × v)/dt = 0. (8.1.4b)
Hence,
h ≡ r×v (8.1.4c)
is a vector that remains constant as r and v evolve. It has a length, h,
given by
h2 = (h · h) = r2 v2 − (r · v)2 . (8.1.4d)
From (8.1.4c) it follows that
r × dr = h dt, (8.1.5a)
so r and dr lie in a fixed plane perpendicular to h. This plane contains
the center

of mass of the planet and that of the sun. In every time interval
∆t = dt, the area swept out by the radius vector r has the same value,

(r(t) × dr(t))/2 = h∆t/2. (8.1.5b)

This implies Kepler’s law of equal areas: a line from the sun to the planet
sweeps out equal areas in equal times. This law may also be considered to
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 249

arise from the constancy of the angular momentum vector of the planet,

L = r × p, p = µv. (8.1.6)

Integrating (8.1.5b) over an entire period of length T, the area swept

out is an ellipse of area πΛa Λb , where 2Λa is length of its major axis, and
2Λb is the length of its minor axis. Consequently, for an elliptical orbit,

T = 2πΛa Λb /h. (8.1.7)

Kepler systems possess a further vector constant of motion.2 We shall

see that it and L together determine the dynamical symmetry of every
Keplerian manifold defined by H = E, and provide the initial conditions
that determine motion upon it. To uncover this vector, take the vector
cross-product of h with the equation of motion, (8.1.1), to obtain

h × dv/dt = −(κ/r3 ) h × r. (8.1.8a)

Because h is constant, this implies that

d(h × v)/dt = −(κ/r3 ) h × r. (8.1.8b)

Using the identity

(a × b) × c = (c · a)b − a(c · b)

to expand

h × r = (r × dr/dt) × r,

one obtains

h × r = r2 dr/dt − r(r · dr/dt) = r3 d(r/r)/dt. (8.1.8c)

Thus (8.1.8b) becomes

d(h × v)/dt = −κd(r/r)/dt. (8.1.8d)

Consequently

d((h × v) + κr/r)/dt = 0, (8.1.8e)

and the vector

b ≡ (h × v) + κr/r, (8.1.9a)

is a constant of the motion.

By expanding h × v one may obtain an alternative expression for b.
October 28, 2010 16:55 9in x 6in b943-ch08

250 Dynamical Symmetry

Letting v = |v|, one finds

b = (r · v)v + (κ/r − v2 )r. (8.1.9b)
This vector is perpendicular to h (and L = µh) because both v · h and
r · h vanish. The constancy of its components was known to Laplace and to
Hamilton.3 Soon after Gibbs4 introduced the concept of vector they were
assembled into one. Today, b and vectors proportional to it, are commonly
called Runge–Lenz vectors.5
From initial measurements of a planet’s, or comet’s, position and veloc-
ity vectors, observers can determine h and b in their frame of reference
and make them the basis of a new right-handed frame of reference that is
particularly adapted to describing the motion of the body. To this end, one
translates and rotates one’s original coordinate system, putting the origin of
the new system at the center of mass, and supposing the plane of the orbit
is the x–y plane. Letting the x axis and unit vector i lie on the major axis of
the orbit, one may choose its positive direction to be that from the center
of mass to aphelion. Then one may set the z axis along h = r × v. If the
motion is counter-clockwise, one chooses k = h/h, if clockwise, k = −h/h.
Then if c = h×b, one has j = c/c. It is also convenient to shift the origin
of clock time so that at t = 0 the planet is at perihelion. This frame of ref-
erence is used in Figs. 8.1.1. The planet’s initial coordinates are x = x0 ,
y = 0 and z = 0. Its initial velocities are vx = 0, vy = v0 and vz = 0. These
values define the further initial values
r = |x0 |, E = µ(v02 /2 − κ/|x0 |); (8.1.10a)
and
h = (0, 0, hz), hz = x0 v0 ; (8.1.10b)
b = (bx , 0, 0), bx = x0 (κ/|x0 | − v02 ); (8.1.10c)
c = (0, cy , 0), cy = bx hz . (8.1.10d)
At perihelion, v20
> κ/|x0 |, so bx is positive, and b points toward aphelion.
In the Exercises one finds that for an orbit with eccentricity ε, and semilatus
rectum λ,
|b| = εκ, (8.1.10e)
1/2
|h| = (λκ) . (8.1.10f)
Figure 8.1.1 illustrate three orbits obtained when the eccentricity is
nonzero. It is assumed that x0 is negative and that the motion is coun-
terclockwise, so v0 is also negative. The perihelion distance is |x0 |.
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 251

0.4

0.2

b
0.5 1.0 1.5 2.0

− 0.2

− 0.4

(a)

2 3

2
1
1
b b
1 2 3 4 1 2 3
−1
−1
−2

−2 −3

(b) (c)

Fig. 8.1.1 Runge–Lenz vectors for elliptical, parabolic, and hyperbolic orbits,
(a) |b| = 0.8, |b| = 0.71, (b) |b| = 1, 0, |b| = 0.71, (c) |b| = 1.2, |b| = 0.71.

All the orbits available to bodies moving under an inverse square law
of force can be determined by considering the scalar product of r with b.6
One has

b · r ≡ (h × v) · r + κr = −h2 + κr. (8.1.11a)

Letting θ be the angle between b and r, one may express this equation as

b r cos(θ) = −h2 + κr, (8.1.11b)

so the dependence of r upon θ is expressed by

(h2 /κ)
r= (8.1.11c)
(1 − (b/κ) cos(θ)).

This is the equation of a conic section with eccentricity ε = b/κ, and

semilatus rectum λ = h2 /κ. For planets the conic sections are, as stated in
October 28, 2010 16:55 9in x 6in b943-ch08

252 Dynamical Symmetry

Kepler’s First Law, ellipses. For ellipses

Λa = λ/(1 − ε2 ), Λb = λ/(1 − ε2 )1/2 . (8.1.12a,b)

The dynamical quantities h = |h| and b = |b| are then related to the
geometrical variables Λa and Λb by

b/κ = (1 − (Λb /Λa )2 )1/2 ,

(8.1.13a,b)
h(b/κ)1/2 = Λa .

Using (8.1.12a) to eliminate h from (8.1.7), one finds that the period T
satisfies

T2 = (4π 2 /κ)Λ3a . (8.1.14)

This is the exact relation approximated by Kepler’s Law of Periodic Times.

Circular, elliptical, parabolic, and hyperbolic orbits are obtained when
ε = 0, 0 < ε < 1, ε = 1, and ε > 1, respectively. The constancy of b reflects
the fact that neither the direction of the major axis, nor the eccentricity of
an orbit, changes with time.
The energy of the system,

E = µv2 /2 − GMm/r = µ(v2 /2 − κ/r), (8.1.15)

is a scalar constant of the motion. It is shown in standard monographs on

theoretical mechanics7 that for 0 ≤ ε < 1,

E = −µ/2Λa.. (8.1.16a)

For ε > 1,

E = µ/2Λa. . (8.1.16b)

As these energies do not depend upon the length of the minor axis, for given
values of µ and a, varying Λb produces a family of degenerate orbits —
orbits of the same energy. The dynamical symmetries responsible for this
degeneracy are investigated in the upcoming sections of the chapter.
A development analogous to that of (8.1.10) may be used to obtain the
relation among velocity components that defines the hodograph of Keplerian
motion.6 The component of v on the j axis is vy , so (8.1.10) implies

vy = (bh)−1 (h x b) · v = −(bh)−1 h2 (v2 − κ/r) = −(h/b)(v2 /2 + E /µ).

(8.1.17a)
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 253

Because v has no component on the k axis, (8.1.17a) can be expressed as

vx2 + vy2 + 2vy b/h + 2E /µ = 0, (8.1.17b)
so
vx2 + (vy + b/h)2 = (b/h)2 − 2E /µ. (8.1.17c)
As the right-hand side of this equation is only a function of constants of
the motion, for every Keplerian motion the hodograph is a circle, of radius
(b/h)2 −2E /µ, with its center at a point −b/h on the j axis. Figures 8.1.2

vc

0.5

va
− 1.0 − 0.5 0.5 1.0

− 0.5

− 1.0

− 1.5

− 2.0

− 2.5

(a)

vc vc
va va
− 1.0 − 0.5 0.5 1.0 − 1.5 − 1.0 − 0.5 0.5 1.0 1.5

− 0.5
− 0.5

− 1.0
− 1.0

− 1.5

− 1.5
− 2.0

− 2.0
− 2.5

− 2.5 − 3.0

(b) (c)

Fig. 8.1.2 Hodographs corresponding to orbits of Figs. 8.1.1, for (a) |b| =
0.8, |h| = 0.71, (b) |b| = 1.0, |h| = 0.7, and (c) |b| = 1.2, |h| = 0.71.
October 28, 2010 16:55 9in x 6in b943-ch08

254 Dynamical Symmetry

illustrate the hodographs corresponding to the orbits of Figs. 8.1.1. For

them, the value of the radius defined by the right-hand side of (8.1.17c) is
κ2 /(x20 v02 ).

8.2 Hamiltonian Formulation of Keplerian Motions

in Phase Space
Hamilton’s mechanics describes how constants of motion determine dynam-
ical symmetries in a phase space. We therefore move from a Newtonian to
a Hamiltonian description of Keplerian motions.
Setting q
= r, q
= |r|, p
= µv, p
= |p
|, the Hamiltonian with value
E is
H
(p
, q
) = p
2 /2µ − µκ/q
. (8.2.1a)
Here, as before, µ is the reduced mass of the planet, and µκ = GMm. As
previously, all variables are supposed to be dimensionless. If t represents
ordinary clock time, the CT’s
E → E = E /(µ3 κ2 ), t
→ t = (µ3 κ2 )t
, (8.2.1b)
and
q
→ q = (µ2 κ)q
, p
→ p = p
/(µ2 κ), (8.2.1c)
convert (8.2.1a) to
H
(p
, q
)/(µ3 κ2 ) ≡ H(q, p). (8.2.1d)
Here q = |q|, p = |p|, and
H(q, p) = p2 /2 − 1/q = E. (8.2.1e)

Hamilton’s equations governing motions in the phase space of p and q

are then
dq/dt = ∂H/∂p = p, dp/dt = −∂H/∂q = −q/q3 . (8.2.2)
We will be dealing with vectors s that have nonzero components in both
the q and p subspaces of PQ space. The evolution parameter is t, and E
can be considered a parameter. The evolution operator exp(t{−H, . }) acts
on q and p to evolve trajectories s(t) = (q(t), p(t)) in the phase space. The
orbits, f(q) = constant, are orthogonal projections of s onto the q subspace
of PQ space, and the hodographs, g(p) = constant, are projections onto
its p subspace. If one considers planetary motions in a fixed plane, both
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 255

q-space and p-space become two-dimensional, and the PQ space is reduced

to a four-dimensional phase space.
For a function X(q, p, E) to be a constant of motion in PQ space, it
must satisfy

{X, H}|H=E = 0. (8.2.3)

Two vectors whose components satisfy this requirement are

L = µh, (8.2.4a)
A ≡ −(L × p + q/q) = −b/κ ≡ a/κ. (8.2.4b)

The negative sign has been introduced in the definition of A and a, because
doing so substantially simplifies the discussion in subsequent sections. In
position space, the vectors A and a point from the center of mass toward
perhelion. After expanding the L × p term in A to (q · p)p − qp2 , one can
use (8.2.1d,e) to make the conversions

A → A(H) = q(p2 + 2H)/2 − (q · p)p, (8.2.5a)

and

A → A(E) = q(p2 + 2E)/2 − (q · p)p, (8.2.5b)

both forms of A satisfy (8.2.3) with X = A. All the expressions for A

change sign when q → −q and p → −p, so A, unlike L, is a proper vector
in PQ space.
From the components of L and A one obtains the components of the
PB operators L ≡ {L, ·} and A ≡ {A, ·} that generate invariance transfor-
mations of the equation H = E. One has

{Lij , ·} = −(qi ∂/∂qj − qj ∂/∂qi ) − (pi ∂/∂pj − pj ∂/∂pi ) ≡ Lij . (8.2.5)

The form of the components of A is more complicated. In Sec. 8.3 we will

use a CT to greatly simplify them.
Here we wish to consider the Lie algebras generated by the these PB
operators. In expressing the commutation relations it is simplest to make
use of the notation in which, e.g. Lx = L1 rather than L23 . We will also
make use of the conventional symbol εijk , which has the value +1 when the
indices are in clockwise order, and −1 when they are not. Then one finds
October 28, 2010 16:55 9in x 6in b943-ch08

256 Dynamical Symmetry

that the components of L and A obey the commutation relations:

[Li , Lj ] = εijk Lk , [Li , Aj ] = εijk Ak , (8.2.6a,b)

[Ai , Aj ] = (−2H)εijk Lk , A = A(H),

(8.2.6c,d)
[Ai , Aj ] = (−2E)εijkLk , A = A(E).
Because of the presence of H in (8.2.6c), Eqs. (8.2.6a–c) do not define a
finite-dimensional Lie algebra. However, because A and L commute with
H, equation c may be replaced by equation d, and E may be treated as a
parameter.
For bound states E is negative. In the subspace of phase space that
contains only bound trajectories

A< (E) ≡ A(E)/(−2E)1/2 , (8.2.7a)

defines PB operators A< = {A< (E), ·}, that given (8.2.6d), satisfy the
relations
[Li , Lj ] = εijk Lk , (8.2.7b)
[Li , A<j ] = εijk A<k , (8.2.7c)
[A<i , A<j ] = εijk Lk . (8.2.7d)
These are the commutation relations of an SO(4) Lie algebra. The quadratic
Casimir operator is
A2< + L2 = Σ(A2<j + L2j ). (8.2.8a)

Substituting H for E in (8.2.7a) one obtains A< (H). Expanding

A< (H)·A< (H) and L · L one finds that

(2H)(A< (H)2 + L2 ) = −(q(p2 − 2H)/2)2 . (8.2.8b)

Now q(p2 − 2H)/2 = 1, so the right-hand side of this equation is equal to

−1. Consequently

H = −(1/2)/(A2< + L2 ). (8.2.9)
The Lie algebra is of rank 2, and consequently possesses a quartic Casimir
operator, which is (L · A< )2 . Because L is perpendicular to A, for all values
of E, (L · A> )2 vanishes.
The operators exp(αi A<i ) and exp(βi Li ) may be shown to be those of the
group SO(4). They alter the components of A<j , and Lj , j = i, while leaving
(A2< +L2 ) unaltered. Because of this, when E is negative, the equation H = E
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 257

is SO(4) invariant and the dynamical symmetry of the system is that of a

hypersphere S(3).
For Keplerian systems with positive energy, one may proceed analo-
gously. Let
A> (E) ≡ A(E)/(2E)1/2 = (q(p2 + 2E)/2 − (q · p)p)/(2E)1/2 .
(8.2.10a)
The PB operators, A> = {A> (E), ·} satisfy the commutation relations
[Li , Lj ] = εijk Lk , (8.2.10b)
[Li , A>j ] = εijk A>k , (8.2.10c)
[A>i , A>j ] = −εijk Lk . (8.2.10d)
These are the commutation relations of the Lie algebra of an SO(3, 1)
group. The quadratic Casimir operator is L2 − A2> . In this case one finds
that
H = (1/2)/(L2 − A2< ). (8.2.11)
The dynamical symmetry is that of a hyperboloid of revolution in four-
space.
When E = 0, H → 0, and
A → A0 (E) ≡ {qp2 /2 − (q · p)p, ·}, (8.2.12a)
the resulting commutation relations are those of the Lie algebra of E(3):
[Li , Lj ] = εijk Lk , (8.2.12b)
[Li , A>j ] = εijk A>k , (8.2.12c)
[A>i , A>j ] = 0. (8.2.12d)
The dynamical symmetry is that of a three-hyperplane in four-space.
The same question arises for all of these three dynamical systems: what
are the geometric objects which possess these symmetries in a phase space?
The next two sections are devoted to providing the answer to this question.

8.3 Symmetry Coordinates For Keplerian Motions

In this section we will be introducing a pair of canonical diffeomorphisms
that produce coordinate systems in which the dynamical symmetries of
Keplerian systems become self-evident geometric symmetries.
To this end, because E is a variable, we consider the motions to be in
PQET space, letting t = q4 , and −E = p4 . The 1/q singularity in H can
October 28, 2010 16:55 9in x 6in b943-ch08

258 Dynamical Symmetry

be eliminated by multiplying the equation H − E = 0 through by q. This

replaces it by the regularized equation
W = 1, (8.3.1)
with
W = q(p2 − 2E)/2.
For all values of the dynamical variables, this equation defines a six-
dimensional manifold in a seven-dimensional PQE subspace of the PQET
space. Let τ be an evolution parameter that ranges through all the real
numbers, let t = q4 , −E = p4 , and define the PB of functions F and G to be
{F, G} = Σ4j=1 (∂F/∂qj ∂G/∂pj − ∂F/∂pj ∂G/∂qj ). (8.3.2)
Hamilton’s equations of motion in PQET space are then special cases of
the general equation
dF/dτ = {F, W}. (8.3.3)
One has
dL/dτ = {L, W} = 0, dA/dτ = {A, W} = 0, (8.3.4)
so L and A remain constants of motion.
We next take advantage of a useful property of Runge–Lenz vectors.
Because
A(E) = q(p2 + 2E)/2 − (q · p)p, (8.3.5a)
one has
(A(E) − A(−E))/2E = q. (8.3.5b)
When E is negative, E = −|E|, and
q = |−2E|−1 (A(−|E|) − A(|E|)) = (−2E)−1/2 (A< (E) − A< (−E)).
(8.3.5c)
It follows that for each negative value of E, the motion of r can be described
by the motion of A< (−E) with respect to the constant vector A< (E).
Figure 8.3.1 illustrates this relation.
When E is positive, one has
q = (1/(2E))(A(|E|) − A(−|E|)) = (2E)−1/2 (A> (E) − A> (−E).
(8.3.5d)
The motion of r can thus be described by the motion of A> (−E) with
respect to the stationary vector A> (E).
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 259

r
− a(− E)

a(E)

Fig. 8.3.1 Relation of position vector to Runge–Lenz vectors a(E), a(−E) when
E = 12 .

We now describe the first canonical diffeomorphism. Equation (8.3.3)

implies that
dt/dτ = {t, W} = r, (8.3.6)
and so dt = r dτ . For states with E ≥ 0, both r and t range over an
open interval, but for bound states, r is cyclic, so mathematically, t is
cyclic, and confined within a compact interval. For these reasons we use
different coordinates for different ranges of E in the (−E, t) subspace of
PQET space. A canonical transformation will be used to define these. For
E < 0, E > 0, and 0− ≤ E ≤ 0+ , define coordinates p
0 = (p
0< , p
0> , p
00 )
and q
0 = (q
0< , q
0> , q
00 ) by
p
0< ≡ (−2E)1/2 cos(t), q
0< ≡ (−2E)1/2 sin(t), E < 0; (8.3.7a)

p
0> ≡ −(2E)1/2 cosh(t), q
0> ≡ (2E)1/2 sinh(t), E > 0; (8.3.7b)

p
00 ≡ −E, q
00 ≡ t, 0− ≤ E ≤ 0+ . (8.3.7c)
8
The transformation (8.3.7a) is due to Poincare . For all real values of E
and t, the transformation (−E, t) → (p
0 , q
0 ) is a one-to-one map from the
full space of (E, t) to that of (p
0 , q
0 ) and is C2 in each interval. It converts
t = q4 and −E = p4 in the of PB (8.3.2) to q
0 and p
0 . If one denotes
negative values of E by E< , and positive values of E by E> , then when
t → 0, one has q
0 → 0 in all three cases, and
p
0< → (−2E<)1/2 ≡ p0< , E< = −p20< /2; (8.3.8a)
p
0> → −(2E>) 1/2
≡ p0> , E> = p20> /2; (8.3.8b)
and
p
0 → 0± ≡ p00 . (8.3.8c)
October 28, 2010 16:55 9in x 6in b943-ch08

260 Dynamical Symmetry

When E is negative, a(E) = a(−p20 /2) is given by

a(E< ) = (p2 − p20< )r/2 − (r · p)p, (8.3.9a)

and

A< (E) = p−1 2 2

0< ((p − p0< )r/2 − (r · p)p). (8.3.9b)

The moving counterpart of a(E< ) is

a(−E< ) = (p2 + p20< )r/2 − (r · p)p. (8.3.9c)

For E > 0, one has the stationary vectors

a(E> ) = (p2 + p20> )r/2 − (r · p)p, (8.3.9d)

A> (E) = p−1 2 2
0> ((p + p0> )r/2 − (r · p)p), (8.3.9e)

and the moving vector

a(−E> ) = (p2 − p20> )r/2 − (r · p)p. (8.3.9f)

The second canonical transformation in PQET space is based on an

inversion in the momentum subspace of p0 , p1 , p2 , p3 . Set

p
4 = (p
0 , p1 , p2 , p3 ), (8.3.10a)

and define

(p
(4) · p
(4) )± = p2 ± p
2

0. (8.3.10b)

Then, letting

(4)
p
(4) → P± = 2p
(4) /(p
(4) · p
(4) )± , (8.3.10c)

and setting

q
(4) = (q
0 , q1 , q2 , q3 ), (q
(4) · q
(4) )± = q2 ± q
2

0, (8.3.10d)

the canonically conjugate transformation is

(4)
q
(4) → Q± = (p
(4) · p
(4) )± q
(4) /2 − (q
(4) · p
(4) )p
(4) . (8.3.10e)

Here,

q
(4) · p
(4) = q
0 p
0 + q · p, (8.3.10f)

October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 261

is unaffected by the choice of ± signs. The transformations defined by

(8.3.10) are generalizations of a transformation due to Levi-Civita.9 The
inverse transformations are obtained by interchanging the roles of the vari-
ables represented by lower case letters with those represented by capital
letters. They are diffeomorphisms in every region not containing the origin
in PE space. Under them

q
(4) · p
(4) → −Q
(4) · P
(4) , (8.3.11a)

thus orthogonal q
(4) and p
(4) produce orthogonal Q
(4) and P
(4) .
When the positive signs in (8.3.10) are chosen,

Lij = qi pj − qj pi = Q4i P4j − Q4j P4i , i < j = 0, 1, 2, 3. (8.3.11b)

When the negative signs are chosen,

Lij = qi pj − qj pi → Q4i P4j − Q4j P4i , i < j = 1, 2, 3;
(8.3.11c,d)
K0j = q0 pj + qj p0 → Q40 P4i + Q4i P40 , i = 1, 2, 3.
To use these coordinate systems as reference systems in PQE space, we
set t = q0 = 0. The resulting momentum vectors are

p(4) = (p0 , p1 , p2 , p3 ) = (p0 , p), p = (p1 , p2 , p3 ); (8.3.12a)

and
(4)
P± = 2p(4) /(p(4) · p(4) )± = (P0± , P± ); (8.3.12b)

with

P0± = 2p0 /(p2 ± p20 ); (8.3.12c)

P± = 2p/(p2 ± p20 ). (8.3.12d)

The corresponding position vectors are

q(4) = (0, q1 , q2 , q3 ) = (0, q) = (0, r); (8.3.13a)

(4) (4) (4) (4) (4) (4) (4)
Q± = (p ·p )± q /2 − (q ·p )p = (Q0± , Q± ); (8.3.13b)

with

Q0± = q · pp0 ; (8.3.13c)

Q± = (p2 ± p20 )q/2 − (q · p)p = a(±p20 /2). (8.3.13d)

To summarize: the combined canonical transformations have converted

ordinary position and momentum coordinates in PQET space into a new
October 28, 2010 16:55 9in x 6in b943-ch08

262 Dynamical Symmetry

set of coordinates, diffeomorphic to the first in regions that do not contain

the origin. Setting t = 0 has turned the original coordinates into a set of
coordinates in PQE space: three of the resulting position coordinates are
components of Runge–Lenz vectors, and three of the resulting momentum
coordinates are proportional to Fock’s projective momentum coordinates.10
The role of the fourth position and momentum coordinates will appear in
the following section.

8.4 Geometrical Symmetries of Bound Keplerian Systems

in Phase Space
In this section the coordinate systems of the previous section are used to
determine the manifolds in phase space upon which bound motions evolve,
and the dependence of these motions on the evolution parameter.
We first consider bound motion and begin by renaming the moving Qj
as follows:
ρj< = (p2 + p20< )qj /2 − (q · p)pj , j = 1, 2, 3, (8.4.1a)
and
ρ4< = (q · p)p0< . (8.4.1b)
One has
a(E< ) = (p2 − p20< )r/2 − (r · p)p, A< = a(E< )/p0< . (8.4.1c)
The four ρj< are functionally independent in PQE space. The SO(4) group
operators exp(αj {Aj< , ·}) and exp(λij {Lij , ·}) move points at the terminus
of ρ< = (ρ1< , ρ2< , ρ3< , ρ4< ). The action of {Ai< , ·} and {Lij , ·} upon the
ρj< is as follows:
{Ai< , ·}ρj< = δij ρ4< , {Ai< , ·}ρ4< = −ρi<,
(8.4.2)
{Lij , ·}ρj< = −ρj< .
Consequently, the SO(4) generators {Ai< , ·} and {Lij , ·} generate rotations
of the four-vector ρ< . One finds that
ρ2< ≡ ρ< · ρ< = Σ41 ρ2j< = (r/2)(p2 + p20< )2 = W(E< )2 . (8.4.3)
As W has unit value, the rotations move points on the surface defined by
ρ< = 1. (8.4.4)
Thus, in the coordinate system of the ρj , Eqs. (8.4.1a,b) define a unit
hypersphere.
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 263

We next turn our attention to the Pj (E) with E = p20< . Define

πi< = Pi (p20< ) = 2pi /(p2 + p20< ), i = 1, 2, 3. (8.4.5)
The action of the {Ai< , ·} on these πj< produces a further momentum
coordinate:
{Ai< , ·}πj< = δij π4< . (8.4.6a)
2
π4< ≡ (1/p0< )(p − p20< )/(p2 + p20< ). (8.4.6b)
Also
{Ai<, ·}π4< = −πi< , {Lij , ·}ρj< = −ρj< , i = 1, 2, 3. (8.4.6c)
Equations (8.4.6) imply that the {Ai< , ·} and {Lij , ·} are generators of rota-
tions of the locally cartesian four-vectors π< ≡ (π1< , π2< , π3< , π4< ), as well
as rotations of the vectors ρ< ≡ (ρ1< , ρ2< , ρ3< , ρ4< ). Equations (8.4.2) and
(8.4.6) together imply that the generators have the realization
{Ai< , ·} = −(ρi< ∂/∂ρ4< − ρ4< ∂/∂ρi< + πi< ∂/∂π4< − π4< ∂/∂πi< ),
i = 1, 2, 3; (8.4.7a)
{Lij , . } = −(ρi< ∂/∂ρj< − ρj< ∂/∂ρi< + πi< ∂/∂πj< − πj< ∂/∂πi< ),
i < j = 1, 2, 3. (8.4.7b)
The variable π4< is related to the coordinate
P0+ = 2p0< /(p2 + p20< ), (8.4.8a)
by
π4< = P0+ − 1/p0< . (8.4.8b)
The four functions πi< are Fock’s projective momentum coordinates divided
by p<0 . They are linearly independent in the PE subspace of PQE space,
but satisfy the identity
 2 2 2 2

E< = (−1/2)/ π1< + π2< + π3< + π4< . (8.4.9a)
Consequently
(p0< π < ) · (p0< π < ) = 1. (8.4.9b)
Direct calculation shows that vectors ρ< and π < are orthogonal:
Σ4 ρj< πj< ≡ ρ< · π < = 0. (8.4.10)
Consequently, for each value of E, bound state Keplerian motions in ordi-
nary six-dimensional phase space evolve on an SO(4)-invariant manifold
October 28, 2010 16:55 9in x 6in b943-ch08

264 Dynamical Symmetry

defined by three equations

ρ< · ρ< = 1, (p0< π < ) · (p0< π < ) = 1, ρ< · (p0< π < ) = 0. (8.4.11)
For any fixed value of the energy, the vector S< ≡ ρ< + p0< π < terminates
on an SO(4)-invariant manifold in six-dimensional PQ space. It has length
21/2 , and the vectors ρ< and p0< π < may be considered to be projections of
S< onto locally orthogonal subspaces.
The geometric interpretation of these statements is most transparent
for one-dimensional regularized Keplerian motion, which exhibits SO(2)
symmetry. The generator of the SO(2) group is
{A< , ·} = q1 p1 ∂/∂q1 − ((p20< + p21 )/2)∂/∂p1
→ −(ρ1< ∂/∂ρ4< − ρ4< ∂/∂ρ1< + π1< ∂/∂π4< − π4< ∂/∂π1< ),
(8.4.12)
and the unit four-vectors of the three-dimensional system are replaced by
unit two-vectors. The vector
p0< π < = (2p0< p1 /(p20< + p21 ), (p21 − p20< )/(p20< + p21 )) (8.4.13a)
≡ ( sin(θ), cos(θ)), (8.4.13b)
expresses the stereographic projection of p1 onto a unit circle. In
Fig. 8.4.1a, p1 is depicted lying on the y axis, and as p1 changes, p0< π <
rotates about the center of the unit circle in the y–z plane. In the figure
p1 is denoted by p, and the point at the terminus of p0< π < is denoted by
p0 π. When the group operator exp(α{A, ·}) rotates p0< π < , the angle θ is
converted to θ + α. The vector
ρ< = ((p21 + p20< )(q1 /2) − q1 p21 , q1 p1 p0< ) (8.4.14a)
ρ< = ((p20< − p21 )(q1 /2), q1 p1 p0< ), (8.4.14b)
ρ< ≡ ( sin(φ), cos(φ)), (8.4.14c)
is a function of p1 as well as q1 . When p1 → 0,
ρ< → q1 (p20< /2, 0) = q1 (−E1 , 0), (8.4.14d)
and when p1 → ± p0< ,
ρ< → q1 (0, −E). (8.4.14e)
The variables q1 and p1 are related by the equation
ρ< · ρ< = (q1 /2)2 (p21 + p20< )2 = W2 = 1. (8.4.15a)
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 265

z
z
1.0
1.0

P0 π
0.5
0.5

ρ x'
y q
− 1.0 − 0.5 0.5 1.0 − 1.0 − 0.5 0.5 1.0
ρ

− 0.5
− 0.5

− 1.0
− 1.0
(a) (b)

P0 π

y'
x'
ρ

(c) (d)

Fig. 8.4.1 (a) Projection connecting P0 π and ρ; (b) projection connecting ρ and
q; (c) relation between the ρ and p0 π vectors; (d) S< , the sum of the two vectors
in the previous figure, terminates on the surface of the sphere.

It requires that

q1 = ±2/(p21 + p20< ). (8.4.15b)

The potential function, 1/|q1 |, becomes infinite at the origin, so for the
kinetic energy to remain finite, q1 cannot go to zero, and the sign on the
right-hand side of (8.4.15b) cannot change during the motion. We will
assume that the positive sign remains. Then, on substituting (8.4.15b)
October 28, 2010 16:55 9in x 6in b943-ch08

266 Dynamical Symmetry

into (8.4.14b),

ρ< → ((p20< − p21 )/(p21 + p20< ), 2p1 p0< /(p21 + p20< )). (8.4.16)

This can be interpreted as a stereographic projection in the plane of p0< π < ,

orthogonal to p0< π < . However, making the substitution only in the second
term of (8.4.14b),

ρ< (q1 , p1 ) = ((p20< − p21 )(q1 /2), 2p1 p0< /(p21 + p20< )). (8.4.17a)

Then

ρ< (q1 , 0) = (q1 , 0)p20< /2, (8.4.17b)

and

ρ< (0, p1 ) = (0, 2p1 p0< /(p21 + p20< )). (8.4.17c)

The resulting mapping of q1 is illustrated in Fig. 8.4.1b. In this figure,

q1 is denoted by q, and the terminus of ρ< (q1 , p1 ) is denoted by q
. In
Fig. 8.4.1c both ρ< (q1 , p1 ) and p0< π < can be projected onto the same
plane, mapping ρ< and p0< π < onto the same unit circle.√Figure 8.4.1d
depicts the two-vector S< = ρ< + p0< π < . It has length 2 and its stere-
ographic projection describes a circle in the x–y plane. This has the result
that the phase space available to the motion is one-dimensional and has
SO(2) symmetry.
The case of ordinary Keplerian motion in planar position-space is dealt
with in the exercises. Here we wish only to note a few parallels with the lower
dimensional system. The four-dimensional phase space is mapped onto the
SO(3) invariant surface of two unit spheres by the orthogonal three-vectors
ρ < and p 0< π < . The generators of the SO(3) invariance group are the PB
operators defined by the angular momentum and the components of the
Runge–Lenz two-vector. As the analog of Eq. (8.4.15) only relates r and
p2 , while the connection between ρ < and p 0< π < is less direct than in the
lower dimensional case.
We return to the discussion of the system with SO(4) invariance. The
evolution of ρ< is determined by the equation

dρ< /dτ = {ρ< , ρ< } = p20< Wπ < = p20< π < , (8.4.18a)

and that of the π < is determined by

dπ < /dτ = {π< , ρ< } = −ρ< /W = −ρ< . (8.4.18b)

October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 267

Consequently,
d2 ρ< /dτ 2 = −p20< ρ< , (8.4.18c)
d2 π < /dτ 2 = −p20< π < . (8.4.18d)
If the initial values of ρi< and πi< are ρi< (0) and πi< (0) respectively, then
integration of each component of these vector equation yields
ρi< (τ ) = ρi< (0) cos(p0< τ ) + p0< πi< (0) sin(p0< τ ),
(8.4.19)
πi< (τ ) = πi< (0) cos(p0< τ ) − (ρi< (0)/p0< ) sin(p0< τ ).
These describe uniform circular motions of the vectors with period 2π/p0< .
The relation between τ and t is considered in the exercises.

8.5 Symmetries of Keplerian Systems with Non-negative

Energies
We conclude this section with a brief discussion of Keplerian systems with
fixed non-negative energies. When the energy is positive, E = E> = p20> /2.
The constant reference three-vector, A> ≡ a> /|p0> | has components
Aj> = ((r · p)pj − (p2 + p20> )qj /2)/|p0> |, (8.5.1)
and the moving four-vector ρ> has components
ρj> = (p2 − p20> )qj /2 − (r · p)pj , i = 1, 2, 3; (8.5.2a)
ρ4> ≡ (r · p)p0> . (8.5.2b)
The corresponding components of the vector π > are
πj> = 2pj /(p2 − p20> ), π4> = (1/p0> )(p2 + p20> )/(p20< − p2 ). (8.5.3)
One finds
ρ21> + ρ22> + ρ23> − ρ24> = W2 , (8.5.4a)
(1/2)
E> = 2 − (π 2 + π 2 + π 2 )) , (8.5.4b)
(π4> 1> 2> 3>

Σ31 ρj> πj> + ρ4> π4> = 0. (8.5.4c)

The action of the PB operators of Ai> and Lij on the coordinates ρj> and
πj is expressed by
{Ai< , ·} = −(ρi< ∂/∂ρ4< + ρ4< ∂/∂ρi< + πi< ∂/∂π4< + π4< ∂/∂πi<),
i = 1, 2, 3,
{Lij , ·} = −(ρi< ∂/∂ρj< − ρj< ∂/∂ρi< + πi< ∂/∂πj< − πj< ∂/∂πi<),
i < j = 1, 2, 3,
(8.5.5)
October 28, 2010 16:55 9in x 6in b943-ch08

268 Dynamical Symmetry

and the evolution of ρ> and π> is governed by the equations

dρ> /dτ = {ρ> , ρ> } = −2E> π > , (8.5.6a)
dπ > /dτ = {π > , ρ> } = −ρ> , (8.5.6b)
and
d2 ρ> /dτ 2 = p20> ρ> , (8.5.6c)
2 2
d π > /dt = p20> π > . (8.5.6d)
Equations (8.5.4) determine SO(3,1)-invariant manifolds in phase space
upon which the the termini of the vectors ρ> and π> evolve.
When E = 0, we consider the vector π 0 to move with respect to the
fixed Runge–Lenz vector. For the latter, one has
aj0 = (r · p)pj − p2 qj /2, j = 1, 2, 3; (8.5.7a)
a40 = (r · p)p0 . (8.5.7b)
Consequently,
rj0 = −aj0 , (8.5.7c)
r40 = (r · p)p00 , (8.5.7d)
with 0− ≤ p00 ≤ 0+. The components of the vector π 0 are
πj0 = 2pj /p2 , j = 1, 2, 3, (8.5.8a)
while
p0 π40 = 1. (8.5.8b)
It follows that
Σ41 ρ2j0 = (rp2 /2)2 = W2 , (8.5.9a)
Σ41 2
πj0 2
= 4/p , (8.5.9b)
Σ41 ρj0 πj0 = 0. (8.5.9c)
One finds
{aj0 , ·} = ∂/∂πj0 , i = 1, 2, 3; (8.5.9d)
{Lij , ·} = −(ρi0 ∂/∂ρj0 − ρj0 ∂/∂ρi0 + πi0 ∂/∂πj0 − πj0 ∂/∂πi0 ),
i < j = 1, 2, 3. (8.5.9e)
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 269

Because a(0) = a(E) + a(−E), a Lie algebra that contains a(E) and a(−E)
also contains a(0). The same can be said of the corresponding PB operators.
The evolution of the vectors ρ0 and π 0 is determined by the equations

dρ0 /dτ = {ρ0 , ρ0 } = 0, (8.5.10a)

dπ 0 /dτ = {π 0 , ρ0 } = 0. (8.5.10b)

Consequently,

d2 ρ0 /dτ 2 = 0, (8.5.10c)
2 2
d π 0 , /dτ = 0. (8.5.10d)

To summarize, as E varies, all a(E), r(E) = a(E) − a(−E), and p(E),

adjust themselves so that the relation r(H − E) = 0, i.e., r(p2 /2 − E) = 1,
determines manifolds in PQET phase space defined by
 1/2
ρ21 + ρ22 + ρ23 + g44 ρ24 = K, (8.5.11a)
 2 
E> = (−1/2)/ π1 + π22 + π32 + g44 π42 , (8.5.11b)
 
ρ1 π1 + ρ2 π2 + ρ3 π3 + g44 ρ4 π4 = 0, (8.5.11c)

which for the case when E < 0,

g44 = g44 = g44 = 1, (8.5.11d)

and when E = 0,

g44 = g44 = g44 = 0, (8.5.11e)

and when E > 0,

g44 = g44 = −1, g44 = 1. (8.5.11f)

For each range of E, the points that represent the system evolve in phase
space on a three-dimensional manifold determined by these equations.
These manifolds have everywhere-local geometric symmetry determined by
the Lie generators of the corresponding degeneracy group, all of which are
first-order differential operators that are of degree one or zero in the vari-
ables ρj and πj .
October 28, 2010 16:55 9in x 6in b943-ch08

270 Dynamical Symmetry

8.6 The SO(4,1) Dynamical Symmetry

In this section, the fourth component of the vectors ρ is used to provide
the group generator
D = {(q · p), . } = p · ∇p − q · ∇q . (8.6.1a)
The operator,
S(α) = exp(αD), (8.6.1b)
of the corresponding one-parameter group carries out the dilatations
S(α)p = exp(α)p, S(α)q = exp(−α)q. (8.6.1c,d)
This property enables one to, indirectly change energies because
S(α)W(E) = S(α)q(p2 − 2E)/2
= (exp(α)qp2 − exp(−α)q 2E)/2
= exp(α)(qp2 − 2E
q)/2 = exp(α)W(E
), (8.6.2)
with
E
= exp(−2α)E.
In the two preceding sections dynamical symmetries were considered to be
geometrical symmetries in PQ space, and the energy, E, was treated as a
parameter. In this section E will be treated as a dynamical variable, and
dynamical symmetries will be considered to be geometrical symmetries in
PQE space.
We begin by using the finite dilatations to convert the vectors a(E) with
p0 = 1 and E = ±1/2, to the general vectors A(E). If exp(α) = p0 , then
S(−α)a(−1/2) = S(−α)((q p2 /2 − q · p p) − q/2)
= p−1 2
0 (q p /2 − q · pp) − p0 q/2

= p−1 2 2 2
0 (q(q p /2 − q · p p) − p0 q/2) = A< (−p0 /2).
(8.6.3a)
Similarly
S(−α)a(1/2) = A> (p20 /2). (8.6.3b)
(Because a(0) = (a(−1/2)+a(1/2))/2, it is not necessary to treat the E = 0
case separately.) The general effect of S(α) on a(−1/2) and a(1/2) is
S(α){a(1/2), ·} = cosh(α){a(1/2), ·} + sinh(α){a(−1/2), ·}, (8.6.4a)
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 271

and
S(α){a(−1/2), ·} = cosh(α){a(−1/2), ·} + sinh(α){a(1/2), ·}. (8.6.4b)
These relations enable one to use D and the components of a(−1/2) and
a(1/2) to provide PB operators that constitute a basis for the Lie algebras
and groups that act in the E < 0, 0− ≤ E ≤ 0+ , and E > 0, subspaces of
PQET space.11 One may set
D = K45 ≡ K54 , (8.6.5a)
and for i < j, j = 1, 2, 3, define the operators
Jij = {Lij , ·} = −Jji , (8.6.5b)
Jj4 = {aj(−1/2), ·} ≡ −J4j , (8.6.5c)
Kj5 = {aj(1/2), ·} ≡ K5j . (8.6.5d)
These Jij , Jj4 satisfy the commutation relations of the SO(4) bound-state
degeneracy group, and the Jij , Kj5 satisfy the commutation relations of the
SO(3,1), E > 0, degeneracy group. For E = 0, the generators of the E(3)
degeneracy group may be taken to be Jij and (Jj4 + Kj5 )/2. Together, the
Jij , Jj4 , Kj5 , and K45 generate an SO(4,1) invariance group of W(E) and
the equations of motion.

8.7 Concluding Remarks

The following general features of the remarkable symmetries of Kepler sys-
tems are worth emphasizing:
1. All the dynamical symmetries of Keplerian systems can be realized as
geometrical symmetries in PQET phase space or its subspaces. Kepler’s
laws, and many further unifying relations, may be considered to be con-
sequences of these symmetries. The bound motions in unbounded PQE
space map into motions on the compact surfaces of hyperspheres. Oper-
ations of the SO(4) degeneracy group that interconvert Keplerian tra-
jectories and hodographs carry out rotations on these hyperspheres. The
time-evolution of points on the trajectories and hodographs also become
rotations on the hyperspheres.
2. The dynamical groups of all systems governed by Hamilton’s equations
of motion may be interpreted as geometrical symmetry groups in the
appropriate phase spaces.
3. There are systematic ways to determine the generators of the dynamical
groups of these systems.
October 28, 2010 16:55 9in x 6in b943-ch08

272 Dynamical Symmetry

4. However, currently, there is no known general method for finding a

canonical transformation that will convert these generators, which may
be of the generalized Lie form, to first-order differential operators.
5. The SO(4, 1) dynamical symmetry of Keplerian systems can be enlarged
to an SO(4, 2) symmetry. Because it is particularly relevant to atomic
and molecular physics, this symmetry will be considered in Chapter 11.
6. As is evident from the citations, the treatment of Keplerian systems
given here is based on a long history of discoveries in classical mechanics,
followed by the discoveries of Fock10 and of Bargmann,12 that exposed
the hidden symmetries of hydrogen-like atoms in the 1930s. In the early
1960s, interest in the role of symmetries in elementary particle theory
led to a great extension of the work of Fock and Bargmann, and to
Barut’s introduction of the concept of dynamical symmetry.13 In addi-
tion to works already cited we would call attention to a review,14 and
to an article by Macintosh,15 published during this period. The trans-
formations that convert ordinary position and momentum variables to
Runge–Lenz vectors and Fock variables are nonrelativistic versions of
transformations discovered in 2008.16

Exercises:
1. Is q · p invariant under the rotations generated by qj ∂/∂pj − qj ∂/∂pj , or
transformations generated by qj ∂/∂pj + qj ∂/∂pj ?
2. Let the elliptical orbit of a planet lying in the q1 –q2 plane have its
Hamilton–Runge–Lenz vector b on the q1 axis. Determine the effect on
q of infinitesimal transformations with the PB generators determined by
Lij , Ai . Average the effect over a circular orbit. What do these infinitesi-
mal transformations do to p? Average their effect over the corresponding
hodograph.
3. Let s = p0 τ , and use the equations of motion (8.3.3) to determine the
relation between physical time, t, and s. Set both t and τ equal to zero
when a planet or comet is closest to the sun. For the planetary case verify
that your periodic time agrees with the exact form of Kepler’s Law.
4. Determine the stereographic projection of Keplerian hodographs onto a
unit sphere (cf. Am. J. Phys. 33 (1965) 570). Generalize your results by
considering rotations of the plane of the hodograph, and compare with
the work of Fock in Ref. 10.
October 28, 2010 16:55 9in x 6in b943-ch08

Symmetries of Classical Keplerian Motion 273

5. Letting E = 0, carry through the developments analogous to those of

sections three through six.
6. Verify that the PB relations of the components of the four-vectors Q4 ,
P4 introduced in Sec. 8.3 are those of canonically conjugate variables.

References
[1] B. J. Cantwell, Introduction to Symmetry Analysis (Cambridge University
Press, Cambridge, 2002).
[2] E. A. Milne, Vectorial Mechanics (Methuen, London, Interscience, N.Y.,
1948), pp. 236–7.
[3] H. Goldstein, Am. J. Phys. 43 (1975) 737.
[4] E. B. Wilson, Vector Analysis Founded upon the Lectures of J. Willard Gibbs
(Scribners, N.Y., 1901).
[5] W. Pauli, Z. Phys. 36 336.
[6] E. A. Milne, ibid. p. 238.
[7] A. L. Fetter, J. D. Walecka, Theoretical Mechanics of Particles and Continua
(McGraw-Hill, N.Y, 1980); H. Goldstein, Classical Mechanics (Addison-
Wesley, Reading, Mass, 1980).
[8] H. Poincare, New Methods of Celestial Mechanics I, NASA Technical Trans-
lation TTF- 450 (Washington, D.C., 1967) pp. 14, 24.
[9] C. Caratheodory, Calculus of Variations and Partial Differential Equations
of the First Order, translated by R. B. Dean, J. J. Brandstatter (Holden-Day,
San Francisco, 1965).
[10] V. Fock, Z. Phys. 98 (1935) 145.
[11] The Lie generators in equations (8.6.5) are classical mechanical versions of
operators introduced by M. Bednar in Ann. Phys. (N.Y.) 75 (1973) 305;
see also A. O. Barut and H. Kleinert, Phys. Rev. 156 (1967) 1541, and 157
(1967) 1180.
[12] V. Bargmann, Z. Physik 99 (1936) 576.
[13] A. O. Barut, Dynamical Groups and Generalized Symmetries in Quantum
Theory (University of Canterbury Press, Christchurch, N. Z. 1972).
[14] M. Bander, C. Itzykson, Rev. Mod. Phys. 38 (1966) 330, 346.
[15] H. V. Macintosh, Symmetry and Degeneracy, Group Theory and its Appli-
cations, Vol. 2, ed. E. M. Loebl (Academic Press, N.Y., 1971).
[16] C. E. Wulfman, J. Phys. A 42 (2009) 185301.
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-ch09

CHAPTER 9

Dynamical Symmetry in Schrödinger

Quantum Mechanics

In this chapter we set out the fundamental principles needed to uncover

the intrinsic symmetries of systems obeying Schrödinger equations, and to
interpret these as geometric symmetries. The chapter begins with a treat-
ment of Dirac’s correspondence principle connecting classical and quantal
systems. A Lie algebraic extension of the principle is then established. It is
used to connect intrinsic local symmetries of Schrödinger equations to local
geometric symmetries of the corresponding classical systems in phase space.
These connections will be shown to exist, despite the fact that Heisenberg’s
uncertainty principle establishes that the positions and momenta of a par-
ticle cannot, even in principle, both be measured with arbitrary accuracy.
Once the connection between quantal and classical symmetries has been
established, the method used to determine invariance transformations of
Schrödinger equations is presented. Finally, dynamical symmetry algebras
of two simple quantal systems are determined, and their physical interpre-
tations and applications are noted.

9.1 Superposition Invariance

Schrödinger’s equations and all the usual equations governing quantum
mechanical systems are linear partial differential equations. If a and b are
constants, and ψ and ψ
are solutions of a linear equation, then aψ + bψ

are also solutions. As a consequence, every such equation is invariant
under the one-parameter groups with generators ψ∂/∂ψ, ψ
∂/∂ψ, ψ∂/∂ψ
,
ψ
∂/∂ψ
. If ψ = Σck ψk is a general solution of the Schrödinger equation
of interest, one may choose to transfer the variations in Ψ to variations
of the ck carried out by one-parameter groups with generators ck ∂/∂ck ,
cj ∂/∂ck ± ck ∂/∂cj . When there are N linearly independent real ψ’s, the
resulting group of superposition transformations is GL(N,R). It the ψ’s are

275
October 28, 2010 16:55 9in x 6in b943-ch09

276 Dynamical Symmetry

complex variables, the corresponding group is GL(N,C). Such superposition

groups, in and of themselves, carry no information that is dependent upon
the Hamiltonian operator that defines a particular quantum mechanical sys-
tem. Hamiltonian-dependent information is carried by groups whose gen-
erators are functions of position and momentum operators. It is these later
groups which can correspond to the dynamical groups of classical Hamilto-
nian systems.

9.2 The Correspondence Principle

In their first paper on the new quantum mechanics, Born and Jordan showed
that Heisenberg’s recently published paper on the subject could be inter-
preted in terms of the operations of matrix algebra, with physical observ-
ables being represented by Hermitian matrices.1,2 In the general case, if
M is a matrix with elements Mab , the corresponding elements of MT , the
transpose of M, are Mba . The matrix whose elements are the complex con-
jugate of those of M is denoted by M*. The adjoint of M, denoted M† , is
the complex conjugate of MT , and its elements have the value M∗ba . Born
and Jordan required that if H is the Heisenberg matrix representing a clas-
sical Hamiltonian, then H must be Hermitian (self-adjoint), that is must
satisfy the relation H = H† .
After the publication of Schrödinger’s first paper it was realized that
Schrödinger’s wave functions, ψ, could be considered to correspond to com-
ponents of Heisenberg vectors. The overall correspondence is as follows: Let
ψa and ψb be general components of a wave-vector, ψ, and let ψ † repre-
sent the transpose of the complex conjugate of ψ. Then the elements of
the matrix M representing the effect of the Schrödinger operator µ on this
vector are


Mab = dv ψa∗ (µ ψb ). (9.2.1)

Here dv is the volume element in position space, and ψk∗ is the complex
conjugate of the Schrödinger wave function ψk , which is in general a func-
tion of t as well as the position variables x. Dirac proposed the bra and
ket notation that is commonly used to represent these integrals, and matrix
elements in quantum theory. Because Dirac’s notation can lead to confusion
when one is dealing with operators that are not self-adjoint, we prefer to
use the mathematician’s notation (f, g), or
f, g, to denote a general scalar
product of f and g. Using it, one would express (9.2.1) as
Mab = (ψa∗ , µ ψb ), or Mab =
ψa∗ , µ ψb . (9.2.2)
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 277

The reader may wish to establish that if µ is self-adjoint and ψa and ψb are
eigenfunctions of it with eigenvalues, ma , mb , then the eigenvalues must be
real numbers.
Because adjoints of Schrödinger operators are defined by the scalar
products, (9.2.1), in which there is no integration over t, in Schrödinger
mechanics the adjoint of i
∂/∂t and of t, is not defined. However, in
time dependent-Schrödinger mechanics, energies are eigenvalues of the self-
adjoint operator H(q, pop ) and the operator i
∂/∂t. If
H(q, pop )φk (q) = Ek φk (q), (9.2.3a)
then
ψk (q, t) = exp(−iEk t/
)φk (q) (9.2.3b)
is the time-dependent Schrödinger function that is a solution of (9.2.3a)
and
H(q, pop )ψk (q, t) = i
∂ψk (q, t)/∂t. (9.2.4a)
Though the adjoint of i
∂/∂t is not defined, it is evident from (9.2.3) and
(9.2.4a) that ψk (q, t)∗ satisfies
H(q, pop )ψk (q, t)∗ = (i
∂/∂t)∗ ψk (q, t)∗ = −i
∂ψk (q, t)/∂t∗ . (9.2.4b)
The adjoint of the product of two operators, AB, is (AB)∗T = B† A† ; so
the adjoint of the commutator of two operators A and B is
[A, B]† = [B† , A† ]. (9.2.5a)
If A and B are self-adjoint, their commutator is skew-adjoint, that is
[A, B]† = [B, A] = −[A, B]. (9.2.5b)
If A and B are self-adjoint, and c is a real constant, either of the equations
C = ic[A, B] or C = −ic[A, B] (9.2.6)
must produce from A and B a self-adjoint operator C.
Born and Jordan observed that in Cartesian coordinates the commuta-
tion relations
[qi , −i
∂/∂qj ] = i
δij , [−i
∂/∂qi, −i
∂/∂qj] = 0,
[qi , qj ] = 0,
(9.2.7a)
are in one-to-one correspondence with the Poisson bracket relations
{qi , pj } = δij , {qi , qj } = 0, {pi , pj } = 0. (9.2.7b)
October 28, 2010 16:55 9in x 6in b943-ch09

278 Dynamical Symmetry

For Cartesian q’s and p’s, this sets up a one-to-one correspondence

[q, pop ] ↔ i
{q, p} (9.2.8)
q = (q1 , q2 , . . . ), p = (p1 , p2 , . . . ), (9.2.9)
pop = ( − i
∂/∂q1 , −i
∂/∂q2, . . . ).
Since we are interested in establishing connections between classical
symmetries and quantal symmetries, we require operations with the quan-
tum mechanical operators that represent observables to correspond to oper-
ations that are on real functions of real classical variables. We thus prefer
to express the correspondence (9.2.8) as
(−i/
)[qj , pkop ] ↔ {qj , pk } = δij ; (9.2.10a)
the left-hand side is then self adjoint and the right-hand side is real. Defining
Eop to be i∂/∂t one has for the classically conjugate variables q4 = t,
p4 = −E,
(−i/
)[t, −Eop ] ↔ {t, −E} = 1, (9.2.10b)
a special case of (9.2.9a), though as mentioned previously the adjoint of
i∂/∂t is not defined in Schrödinger mechanics.
With these observations in mind, we note that for each value of j in
(9.2.9a), one has a quantal three-parameter Lie algebra, in which the oper-
ation of composition is via the commutator, and
(−i/
)[qj , pjop ] = 1, (−i/
)[qj , 1] = 0, (−i/
)[pjop , 1] = 0. (9.2.11a)
In the corresponding classical Lie algebra, composition is via the Poisson
bracket, and
{qj , pj } = 1, {qj , 1} = 0, {pj , 1} = 0. (9.2.11b)
These two isomorphic Lie algebras are termed Heisenberg algebras.
The Born-Jordan correspondence produces a revealing correspondence
between transformation groups. The commutation relation −i[qk , pkop ] =

implies Weyl’s relation
exp(iapkop ) exp(ibqk ) = exp(iab
) exp(ibqk ) exp(iapkop ). (9.2.12a)
The corresponding classical relation is, of course,
exp(iapk ) exp(ibqk ) = exp(ibqk ) exp(iapk ). (9.2.12b)
The relation between (9.2.11a) and (9.2.11b), may be summarized as
exp(iab
) ↔ 1, q k ↔ qk , pkop = −i
∂/∂qj ↔ pk . (9.2.12c)
The function exp(iab
) only takes on unit value if its argument is zero. For
this reason the spaces related by the correspondence (9.2.11), when
= 0,
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 279

are not the spaces one needs to relate classical and quantum mechanical
symmetries in which the group parameters a,b can vary.
There is another way of viewing group-theoretic consequences of the
Born–Jordan correspondence. Let g(q) be an analytic function and consider
the expression
(−i/
)[exp(iapop ), g(q)] = (−i/
)[exp(a
∂/∂q), g(q)]
= (−i/
)(g(q + a
) − g(q)) exp(a
∂/∂q).
(9.2.13a)
Its Poisson bracket analog is
{exp(ap), g(q)} = −(∂ exp(ap)/∂p) ∂g(q)/∂q
(9.2.13b)
= −a ∂g(q)/∂q exp(ap).
A Maclaurin series expansion of the term g(q+ a
)− g(q) in (9.2.12a) yields
−i(a∂g(q)/∂q + (a2
)(∂ 2 g(q)/∂q2 )/2 + · · · ) exp(a
∂/∂q). (9.2.13c)
Comparing this with the corresponding term, −a ∂g(q)/∂q, in (9.2.12b) one
sees that relations (9.2.12a) and (9.2.12b) are in one-to-one correspondence
only to first order in
.
Both these observations indicate that a change in viewpoint is required if
one is to understand why it is that some, but not all, classical and quantum
mechanical systems have identical local symmetry groups.
Intrinsic local symmetries of Schrödinger equations are determined
by Lie algebras whose generators are functions of q’s and pop ’s, t and
−Eop = −i
∂/∂t. On the other hand, as we have seen, the local dynam-
ical symmetries of classical Hamiltonian systems are determined by Lie
algebras whose generators are Poisson bracket operators. In Schrödinger
mechanics
[exp(iapkop ), g(q)] = [exp(a
∂/∂qk), g(q)]. (9.2.13)
In classical mechanics
[exp(−a{pk , · }), g(q)] = [exp(a∂/∂qk ), g(q)]. (9.2.14)
The group operators in these expressions are in one-to-one correspon-
dence to all orders in
As a consequence, correspondences between the
group actions of Poisson bracket operators in phase space and the group
actions of quantum mechanical operators in position or momentum sub-
spaces of phase space are valid to all orders in
.
It will be shown in Secs. 9.3 and 9.4, below, that these observations lead
to a change in viewpoint that does much to clarify the relation between
October 28, 2010 16:55 9in x 6in b943-ch09

280 Dynamical Symmetry

quantal dynamical symmetries and classical dynamical symmetries. These

sections develop the correspondence that relates Lie algebras of Schrödinger
operators of a more general form than q, pop , to Lie algebras of their corre-
sponding Poisson bracket operators. As in Chapter 3, position and momen-
tum variables will be assumed dimensionless. Because the relations that will
be set forth are valid independent of the value of
, it simplifies matters if we
require
to be dimensionless, and use units that have been so chosen that
its value is 1. This puts a constraint on the units of time being used. If one
uses atomic units with the Bohr radius, ao = 0.5292 × 10−8 cm, as the unit
of length, and mo , the mass of the electron (or mor , its reduced mass in the
H atom, 9.102 × 10−28 g), as the unit of mass, then
will have unit value if
t is measured in units of t0 = 2.41776 × 10−17 s (or tor = 2.41782 × 10−17 s).
Using these units the absolute value, e, of the charge of the electron has
unit value, and c, the velocity of light, has the value 136.958 (or 136.962).

9.3 Correspondence Between Quantum Mechanical Operators

and Functions of Classical Dynamical Variables
The connection between classical functions of positions and momenta and
Schrödinger operators, or their momentum space analogs, is ambiguous if
one does not specify an ordering of the non-commuting quantum mechanical
representatives of the p’s and q’s. These ordering ambiguities have been
termed impediments to quantization of classical mechanics, and are still a
subject of interest to theoreticians.3 The ambiguity is reduced by requiring
that Schrödinger operators are either self-adjoint or skew-adjoint. To further
reduce the ambiguity: one may require either that

i) a quantum mechanical operator A which contains a product of qj ’s and

∂/∂qj ’s must also contain the adjoint of this product,

or

ii) the self-adjoint or skew-adjoint operator, A
can be obtained from an

operator of form i) by a similarity transformation A → A
= S−1 AS,
with a corresponding classical analog that is a canonical transformation.

Examples of Schrödinger operators of the first kind and their corresponding

classical functions are:

A1 = (−i∂/∂x)2 /2 − i∂/∂t = p2op /2 − Eop ↔ A1 = p2 /2 − E (9.3.1a)

October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 281

and

A2 = −it(x∂/∂x + ∂/∂xx)/2 − x2 /2 − i(t2 ∂/∂t + ∂/∂tt2 )/2

(9.3.1b)
↔ tqp − q2 /2 − t2 E.

The self-adjoint Schrödinger operator A2 generates invariance transforma-

tions of the time-dependent free-particle Schrödinger equation A1 ψ = 0,
which will then be investigated in Sec. 9.6. One might use commutation
relations to replace the symmetrized operators in (9.3.1) by −i(x∂/∂x+1/2)
and i(t2 ∂/∂t + t), but we will not do so.
The operators in (9.3.1) can be considered to be special cases of a more
general class of operators each of whose members are sums of symmetrized
products of the form

Γ= (Fj (q, t)gj (pop , Eop ) + (Fj (q, t)gj (pop , Eop ))† ). (9.3.2)

Operators of this form will be used throughout this chapter. It will be

assumed that these operators have a well-defined action on the space of
wave functions that is of interest.
An example of the second type is provided by the inversion transfor-
mation of Eqs. (8.3.10). Classically, it interconverts two sets of canonically
conjugate variables. Its quantum mechanical analog produces self-adjoint
Schrödinger operators and momentum space operators that are not neces-
sarily of the form (9.3.2), but satisfy the relations (9.2.10a). In any given
case it may not be easy to recognize whether an operator which contains
terms not of the symmetrized form can be obtained by a similarity trans-
formation from operators that are of this form. Finally, it should be noted
that the considerations of this section apply, ipso facto, to momentum space
realizations of quantum mechanics, as well as to position space realiza-
tions.

9.4 Lie Algebraic Extension of the Correspondence Principle

In this section we investigate the circumstances under which one can have
the isomorphic relations

(−i/
)[A(q, pop ), B(q, pop )] = C(q, pop ) (9.4.1a)

{A(q, p), B(q, p)} = C(q, p), (9.4.1b)
[{A(q, p), · }, {B(q, p), · }] = {C(q, p), · }. (9.4.1c)
October 28, 2010 16:55 9in x 6in b943-ch09

282 Dynamical Symmetry

We first consider the classical correspondence between (9.4.1b) and

(9.4.1c) when a is not a function of q, p, E nor t. Then

{a, · } = 0. (9.4.3)

Thus

{(A(q, p, t, E) + a), · } = {A(q, p, t, E), · } (9.4.4a)

and the two-way correspondence

{Ak (q, p, t, E), · } ↔ Ak (q, p, t, E) (9.4.4b)

becomes ambiguous. This can have particularly confusing consequences if

one does not adhere to the use of symmetrized self-adjoint, or skew-adjoint,
quantum-mechanical operators introduced in the previous section.
In Chapter 7 it was shown that if a set of C2 functions Gm (q, p) satisfy

{Gm (q, p), Gn (q, p)} = Σk ckmn Gk (q, p), (9.4.5)

then the corresponding Poisson bracket operators satisfy

[{Gm (q, p), · }, {Gn (q, p), · }] = Σk ckmn {Gk (q, p), · }. (9.4.6)

It was also noted that the apparent isomorphism between (9.4.5) and (9.4.6)
was misleading if any Gk (q,p) could be a constant. If this occurs, a {Gk , · }
will vanish, and the Lie algebras determined by (9.4.5) and (9.4.6) will not
be isomorphic, the vanishing of a {Gk , ·} being equivalent to the vanishing
of every ckmn with the same value of k.
This is illustrated by the Lie algebra that is put in correspondence with
the Heisenberg algebras of (9.2.10) by using (9.4.1). In this algebra

{qj , · } = ∂/∂pj , {pj , · } = −∂/∂qj, (9.4.7a)

and

[{qj , · }, {pj , · }] = {1, · } = 0, [{qj , · }, {1, · }] = 0,

[{pj , · }, {1, · }] = 0).
(9.4.7b)
The algebra is Abelian, but its quantum mechanical version is not, because

[qj , pj ] = i
. (9.4.8)

It is not difficult to determine the circumstances under which the general

correspondences of (9.4.1) give rise to Poisson bracket Lie algebras that
contain this Abelian algebra as a subalgebra. Let two functions F(q, p) and
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 283

G(q, p) each be expandable in a Maclaurin series with fixed coefficients,

and a remainder term, R, so that
F = f00 + f01 q + f10 p + R(fab qa pb ), a + b > 1, (9.4.9a)
G = g00 + g01 q + g10 p + R(gab qc pd ), c + d > 1. (9.4.9b)
Then

{F, G} = (f01 g10 − g01 f10 ) + (hmn qm pn ), m + n > 1. (9.4.9c)

Consequently, {F,G} can be a nonzero constant iff (f01 g10 − g01 f10 ) is

nonzero and R vanishes. This is the only case in which F and G can be
members of a Heisenberg algebra, and hence produce an Abelian subalge-
bra containing {qj , · } and {pj , · }.
These observations lead to the following statement4 :
Let a set of operators Γk (q, pop , t, Eop ) of the form (9.3.2) be genera-
tors of invariance transformations of a Schrödinger equation, and members
of a Lie algebra whose basis cannot contain a Heisenberg subalgebra or a
generator that is a constant. The two-way correspondences
Γk (q, pop , t, Eop ) ↔ Gk (q, p, t, E) ↔ {Gk (q, p, t, E), · }, (9.4.10a)
then produce the three isomorphic Lie algebras
−i[Γj , Γk ] =
cm
jk Γm , (9.4.10b)
{Gj , Gk } = cm
jk Gm , (9.4.10c)
[{Gj , · }, {Gk , · }] = cm
jk {Gm , · }. (9.4.10d)
When the Γ’s are generators of invariance transformations of a Schrödinger
equation, the isomorphisms establish a one-to-one correspondence between
local intrinsic symmetries of Schrödinger equations, and local geometrical
symmetries in classical phase space.

It should be noted that if [Aj , Ak ] =
cR R

jk AR , and K{Aj , Ak } =
cjk AR ,
with K a nonzero constant, then K can be absorbed into the structure con-
stants without altering the Killing–Cartan classification of the Lie algebra.

When these conditions are satisfied one may say that Eqs. (9.4.10) define
a faithful Lie-algebraic extension of the correspondence principle. In this
case, the Lie algebra of quantum mechanical operators and the Lie alge-
bra of PB operators define locally isomorphic dynamical groups. Heisen-
berg’s uncertainty principle, a consequence of the commutation relation
October 28, 2010 16:55 9in x 6in b943-ch09

284 Dynamical Symmetry

[qop , pop ] = i
, does not affect this correspondence. The SO(4,2) local Lie
groups of quantum-mechanical and classical Kepler systems are an exam-
ple of this isomorphism which may be said to exist despite the uncertainty
principle.
Harmonic oscillators provide a contrasting example. The Lie algebra of
their Dirac energy shift operators is a Heisenberg algebra. Because of this,
there is no one-to-one correspondence between the consequent quantum
mechanical and classical dynamical groups of harmonic oscillators.
When a Poisson bracket Lie algebra contains a pair of operators −∂/∂qj ,
and ∂/∂pj , they generate translations in the q and p subspaces of classical
phase space. Thus translation invariance in phase space produces systems
with quantum mechanical and classical symmetries that need not be iso-
morphic. In Schrödinger mechanics, −∂/∂qj innocuously becomes −i∂/∂qj .
On the other hand, ∂/∂pj becomes −iqj and generates gauge transforma-
tions which multiply ψ(q) by exp(iaqj ).
To determine whether corresponding quantal and classical systems with
locally isomorphic symmetry groups have globally isomorphic groups, one
must determine the interval over which their group parameters can range
without producing an identity operation. In the classical mechanical cases
the group moves a point in a phase space, so one determines the range which
the group parameters can span before restoring points in the phase space to
their initial positions. In Schrödinger quantum mechanics the wave function
Ψ (q, t) defines points with coordinates (ψ, q, t), with ψ having the value
Ψ(q, t). One determines the range which group parameters can span before
restoring q,t to their original values, and ψ to its original value. When
the correspondence between classical and quantum mechanical global Lie
groups exists, intrinsic quantum mechanical global symmetries correspond
to global geometrical symmetries in phase space.
An example is provided by the correspondence between quantum angu-
lar momentum operations and rotational symmetry. For L12 , the component
of the angular momentum operator on the q3 axis, one has

L12 = −i(q1 ∂/∂q2 − q2 ∂/∂q1) = (q1 p2op − q2 p1op )

and

q1 p2 − q2 p1 ↔ {(q1 p2 − q2 p1 ), · } = −R12 . (9.4.11)

Note that

R12 = (q1 ∂/∂q2 − q2 ∂/∂q1 ) + (p1 ∂/∂p2 − p2 ∂/∂p1 ).

October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 285

Applying the correspondence to the other angular momentum components,

−i(q2 ∂/∂q3 − q3 ∂/∂q2 ) and −i(q3 ∂/∂q1 − q1 ∂/∂q3 ), one obtains three Rij
corresponding to three Lij . The L’s obey the commutation relations
√
−1[Lij , Ljk ] = Lki , i, j, k cyclic. (9.4.12a)

The −R’s obey the isomorphic commutation relations

[−Rij , −Rjk ] = −Rki , i, j, k cyclic. (9.4.12b)

Allowing the operators exp(iαjk Ljk ) to act upon the spherical harmonics
Ylm (θ, φ) in order to reproduce the identity operation, the group parameters
must have the same ranges that are required of exp(−αjk Rjk ) when it acts in
phase space — the group in both cases is SO(3).

9.5 Some Properties of Invariance Transformations

of Partial Differential Equations Relevant
to Quantum Mechanics
In this section we lay the ground work required for determining the intrinsic
symmetries of Schrödinger equations. For the time being, it will not be
assumed that group generators are self-adjoint.
Let Gop = G(q, pop , t, Eop ) = G(x, −i∂/∂x, t, i∂/∂t) be a Schrödinger
operator that acts on ψ(x, t) and suppose that ψ(x, t) satisfies the
Schrödinger equation

(H(x, −i∂/∂x) − i∂/∂t)ψ(x, t) = 0. (9.5.1)

Suppose also that, for arbitrary infinitesimal values of , the infinitesimal

transformation ψ(x, t) → ψ
(x, t) = (1 +  Gop )ψ(x, t) yields a ψ
(x, t) that
is a solution of (9.5.1). This requires that

(H − i∂/∂t)Gop ψ(x, t) = 0 (9.5.2)

when

(H − i∂/∂t)ψ(x, t) = 0.

Note that ψ
(x, t) may or may not be a solution of the equation

(H − E
)Gop ψ(x, t) = 0, (9.5.3a)

when

(H − E)ψ(q) = 0. (9.5.3b)
October 28, 2010 16:55 9in x 6in b943-ch09

286 Dynamical Symmetry

Also, if it is, E
may, or may not, equal E. We shall, for example, see in
Sec. 9.6 that if
H = −(1/2)∂ 2/∂x2 , (9.5.4a)
then,
Gop = −i(x∂/∂x + 2t∂/∂t) (9.5.4b)
satisfies (9.5.2). It converts solutions of the time-dependent equation into
solutions, but does not convert an eigenfunction of H into an eigenfunction.
Similarly, if
H = −(1/2)∂ 2/∂x2 + (1/2)kx2 , (9.5.5a)
the shift operator
exp(−ikt)(∂/∂x − ikx) (9.5.5b)
satisfies (9.5.2), and converts eigenfunctions of energy E into eigenfunctions
with E
not equal to E.
If G(q; p) = G(x, t; px , −E) is an analytic function of the components
of p in a region about p = 0, we may expand it in a Maclaurin series in
the components of p, E. We suppose this is the case, and let
G(q, p) = Σgijk··· (p1 )i (p2 )j (p3 )k · · · , (9.5.6)
where the g’s are functions of the q’s or constants. We use the form (9.5.6) in
determining Lie generators, then subsequently convert them to self-adjoint
operators A(G + G† )/2 or their symmetrized counterparts.
Before proceeding further let us make a change in notation to conform
to usage in much literature on invariance transformations of Schrödinger
equations. Let us express the Schrödinger operator corresponding to the
power series expansion (9.5.6) as
Gop = Q(q, ∂/∂q) = Σgijk···(∂/∂q1 )i (∂/∂q2 )j (∂/∂q3 )k · · · , (9.5.7)
ijk···
the g being functions of the position variables, q = x, and/or t. The
operators may have a term with all indices equal to zero, which contains
no derivative operator, as well as terms containing derivatives of arbitrary
order. If the coefficient of a term linear in a ∂/∂x is real, one may wish
to multiply Q by i. Call the result Qc . Then, if Q generates invariance
transformations of a Schrödinger equation, (Qc ± Q†c )/2 can be made self-
adjoint or skew-adjoint. Doing this for each of the generators Qk of interest,
one can construct a set of operators with definite adjointness and definite
commutation relations.
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 287

The Q of (9.5.7) appear different from the generators introduced by

Lie, all of which were first-order differential operators. The Q’s may have
operators with terms free of derivatives, as well as terms with higher-order
derivatives. To understand how to put the derivative-free terms in Lie form
it is sufficient to consider the action of a simple Q of the form
Q = q0 (x) + q1 (x)∂/∂x. (9.5.8)
This acts upon solutions ψ(x) to give
Qψ(x) = q0 (x)ψ(x) + q1 (x)∂ψ(x)/∂x. (9.5.9)
We have seen in Chapter 4 that it is always possible to vary solutions of
differential equations without varying what are usually termed their inde-
pendent variables. In this case ψ(x) can be varied without varying x, so
∂/∂ψ can act upon it to give a nonzero result even if x is fixed. Thus Q has
the same effect as the Lie generator
U = q0 (x)ψ∂/∂ψ + q1 (x)∂/∂x. (9.5.10)
This relation between U and Q can be maintained as long as the wave-
function ψ(x) is not completely fixed. If a Schrödinger equation had only
one solution, fixed in phase and normalization, (9.5.10) could not replace
(9.5.9). This never happens.
In the light of these remarks, let us consider Q’s that involve second-
and higher-order derivatives, operators of the form
Q = q1 (x)∂/∂x + q2 (x)∂ 2 /∂x2 + q3 (x)∂ 3 /∂x3 + · · · (9.5.11)
As ψ and x can be independently varied, this Q has the same effect on Ψ(x)
as the action of the operator
U = q1 (x)∂/∂x + (q2 (x)∂ 2 Ψ(x)/∂x2 + q3 (x)∂ 3 Ψ(x)/∂x3 + · · · )ψ∂/∂ψ
(9.5.12)
on ψ. This U is that portion of the Lie generator of a contact transformation
that acts in the space of x, Ψ(x), and these derivatives. These considerations
apply to the more general Q of (9.5.11), which can be expressed as a first-
order differential operator U by converting all of its terms that are not those
of a first-order differential operator to an analogous coefficent of ψ∂/∂ψ.
In applying (9.5.7) one finds a fundamental distinction between invari-
ance transformations of ordinary differential equations and invariance trans-
formations of partial differential equations: PDEs can be invariant under
transformations with generators that depend upon derivatives of order
higher than those in the PDE.
October 28, 2010 16:55 9in x 6in b943-ch09

288 Dynamical Symmetry

We now turn to a discussion of this difference. Consider an ordinary

differential equation such as the Schrödinger equation for which the kinetic
energy operator T is −(1/2)∂ 2Ψ(x)/∂x2 and the potential function is v(x).
Let us rearrange the equation to isolate T, and write it as
∂ 2 Ψ(x)/∂x2 = 2(v(x) − E)Ψ(x). (9.5.13)
Taking the derivative of both sides of this equation one obtains the identity
∂ 3 Ψ/∂x3 = 2(v∂Ψ/∂x + (∂v/∂x − E)Ψ). (9.5.14)
This equation is said to be a differential consequence of (9.5.13). Inspecting
it, one observes that at any point with coordinate x, ∂ 3 Ψ/∂x3 cannot be
varied independently of Ψ and ∂Ψ/∂x. A further differentiation then estab-
lishes that ∂ 4 Ψ/∂x4 is not independent of Ψ, ∂Ψ/∂x, and ∂ 2 Ψ(x)/∂x2 , and
so forth. In short, because (9.5.13) is a second-order ODE, only Ψ and
∂Ψ/∂x can be independently varied at x. This has the consequence that
the generators of invariance transformations of (9.5.13) need only depend
upon, x, Ψ and ∂Ψ/∂x. The result is not affected by the presence of the
potential v(x), and a similar statement holds for all second-order ODEs of
a form such that the highest derivative can be isolated, or more generally,
expressed as a function of lower-order derivatives. An analogous argument
establishes that for an n-th order ODE with isolable highest derivative, all
derivatives of order n or greater, are functionally dependent on those of
lower order. In Lie’s theory of differential equations the dependencies on
these derivatives are treated as contact transformations, and expressed by
the use of extended generators. (cf., Chapter 5.)
Now consider the time-dependent free-particle Schrödinger equation
i∂Ψ(x, t)/∂t = −(1/2)∂ 2Ψ(x, t)/∂x2 , (9.5.15a)
and write it as
2iΨt = −Ψxx . (9.5.15b)
It has the differential consequences:
2iΨtt = −Ψxxt , 2iΨtx = −Ψxxx , . . . . (9.5.16)
One may consider all the functions on one side of these equations to be
independently variable. As a result generators of invariance transforma-
tions of (9.5.15) can depend upon derivatives of arbitrarily large order.
In contrast to ordinary differential equations, partial differential equations
have invariance transformations whose generators U and Q may depend
upon derivatives of arbitrary order.5
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 289

Thus, when one is dealing with partial differential equations, Q must, in

principle, be allowed to be a differential operator of arbitrarily high order.
However, as will be seen in the following pages, for many purposes one
may only require generators of point transformations, or generators Q that
depend only on a few higher order derivatives.
From the standpoint of the correspondence principle, the fact that a
Schrödinger operator can depend upon derivatives of arbitrary order should
be expected, because the corresponding functions in a classical Poisson
bracket operator can depend upon arbitrary powers of p. If one expands
the functions in PB operators {G(q,p),·} in a power series in p, one confronts
the classical analog of the discussion of this section. The reader may suspect
that, because kinetic energies involve only first and second powers of p’s,
differential consequences of Schrödinger equations may not prove overly
troublesome. This suspicion will be verified in subsequent pages, which
provide a number of illustrations of the concepts introduced in this section.

9.6 Determination of Generators and Lie Algebra

of Invariance Transformations of
((−1/2)∂ 2 /∂x2 − i∂/∂t)ψ(x, t) = 0
The equation may be considered as the time-dependent Schrödinger equa-
tion of a free-particle of unit mass moving in one dimension. To simplify
matters we multiply it by −2, set ψ = y and obtain the equivalent equation

Sop y(x, t) = 0, Sop = ∂ 2 /∂x2 + 2i∂/∂t. (9.6.1a)

Among the transformations which leave this equation invariant, some leave
invariant the corresponding time-independent equation

(∂ 2 /∂x2 + 2E)y(x, t) = 0. (9.6.1b)

We will seek invariance generators of (9.6.1) that have the simple form

Q = q00 (x, t) + q10 (x, t)∂/∂x + q01 (x, t)∂/∂t. (9.6.2)

This requires that

(∂ 2 /∂x2 + 2i∂/∂t)Qy = 0 (9.6.3)

for all y satisfying (9.6.1a) and all of its relevant differential consequences.
Given the assumed form of Q, (9.6.3) contains a third-order derivative,
October 28, 2010 16:55 9in x 6in b943-ch09

290 Dynamical Symmetry

and these differential consequences involve derivatives up to third-order.

One has
∂Q/∂x = Qx = q00 10 01
x (x, t) + qx (x, t)∂/∂x + qx (x, t)∂/∂t,

∂Q/∂t = Qt = q00 10 01
t (x, t) + qt (x, t)∂/∂x + qt (x, t)∂/∂t, (9.6.4)
∂ 2 Q/∂x2 = Qxx = q00 10 01
xx (x, t) + qxx (x, t)∂/∂x + qxx (x, t)∂/∂t.

Using the same convention to denote derivatives of y, one has

∂Qy/∂t = Qt y + Qyt , ∂Qy/∂x = Qx y + Qyx ,
(9.6.5)
∂ 2 Qy/∂x2 = Qxx y + 2Qx yx + Qyxx .
The invariance condition (9.6.3) requires
Qxx y + 2Qx yx + 2iQt y = 0, (9.6.6a)
and is subject to the differential consequences of (9.6.1a), viz.
yxx + 2iyt = 0. (9.6.6b)
These are
yxxx + 2iytx = 0, yxxx + 2iytt = 0. (9.6.6c,d)
Expanding (9.6.6a) yields
(q00 10 01 00 10 01
xx y + qxx yx + qxx yt ) + 2(qx yx + qx yxx + qx yxt )

+ 2i(q00 10 01
t y + qt y x + q t y t ) = 0 (9.6.7)
We choose y, yx , yt , ytx to be independently variables, and use (9.6.6b) to
eliminate yxx from (9.6.7). This converts (9.6.7) to
(q00 00 10 00 10
xx + 2i qt )y + (qxx + 2 qx + 2i qt )yx

+ q01 01 10 01
xx yt + 2i(qt − 2 qx )yt + 2qx yxt = 0. (9.6.8)
As y, yx , yt , ytx can be assigned arbitrary values, for (9.6.8) to hold, the
coefficients of these functions in the equation must themselves vanish. We
thus obtain the following determining equations for Q:
q00 00
xx + 2iqt = 0, (9.6.9a)
q10 00 10
xx + 2qx + 2iqt = 0, (9.6.9b)
q01 01 10
xx + 2i(qt − 2qx ) = 0, (9.6.9c)
q01
x = 0. (9.6.9d)
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 291

The last equation is the simplest, so we start out with it and work our way
upward. For q01
x to vanish, q
01
can not be a function of x, so one may set

q01 = A(t). (9.6.10)

This causes the first term in (9.6.9c) to vanish — with the consequence that

q10 01
x = (1/2)qt = (1/2)∂A(t)/∂t. (9.6.11)

Hence

q10 = (x/2)∂A(t)/∂t + B(t). (9.6.12)

It follows that in (9.6.9b), q10

xx = 0, and

q10 2 2
t = (x/2)∂ A(t)/∂t + ∂B(t)/∂t. (9.6.13)

Thus (9.6.9b) requires

q00 10 2 2
x = −iqt = −i(x/2)∂ A(t)/∂t − i∂B(t)/∂t, (9.6.14)

so

q00 = −i(x2 /4)∂ 2 A(t)/∂t2 − ix∂B(t)/∂t + C(t). (9.6.15)

Inserting (9.6.15) into (9.6.9a) yields

(i/2)∂ 2 A(t)/∂t2 + (x2 /2)∂ 3 A(t)/∂t3 + 2x∂ 2 B(t)/∂t2 + 2i∂C(t)/∂t = 0.

(9.6.16)

As x and x2 are linearly independent this requires

∂ 3 A(t)/∂t3 = 0, (9.6.17a)
2 2
∂ B(t)/∂t = 0, (9.6.17b)

and

−(i/2)∂ 2 A(t)/∂t2 + 2i∂C(t)/∂t = 0. (9.6.17c)

The first two equations imply that

A(t) = c1 t2 + c2 t + c3 , B(t) = c4 t + c5 . (9.6.18a,b)

Inserting these results into (9.6.17c) gives

∂C(t)/∂t = c1 /2, C(t) = c1 t/2 + c6 . (9.6.18c,d)

October 28, 2010 16:55 9in x 6in b943-ch09

292 Dynamical Symmetry

This completes the determination of q01 , q10 , and q00 . Inserting

(9.6.18a,b,d) into (9.6.10), (9.6.12) and (9.6.15) one obtains

q01 = A(t) = c1 t2 + c2 t + c3 , (9.6.19a)

q10 = (x/2)∂A(t)/∂t + B(t) = (x/2)(2c1 t + c2 ) + c4 t + c5 , (9.6.19b)

q00 = −i(x2 /4)∂ 2 A(t)/∂t2 − ix ∂B(t)/∂t + C(t), (9.6.19c)

so

q00 = −i(x2 /2)c1 − ix c4 + c1 t/2 + c6 . (9.6.19d)

Inserting these results in (9.6.2), that is in Q = q00 + q10 ∂/∂x + q01 ∂/∂t,
yields

Q = −ix2 c1 /2 − ix c4 + c1 t/2 + c6 + (c1 xt + c2 x/2 + c4 t + c5 )∂/∂x

+ (c1 t2 + c2 t + c3 )∂/∂t. (9.6.20)

Collecting terms that are multiplied by the same cj , one may express this as

Q = c1 Q 1 + c 2 Q 2 + c 3 Q 3 + c 4 Q 4 + c 5 Q 5 + c 6 Q 6 . (9.6.21)

The coefficients cj are arbitrary constants of integration.

The six Q’s, their corresponding U’s, and the transformations the U’s
generate are as follows:
Q1 = xt∂/∂x + t2 ∂/∂t + 1/2(t − ix2 ),

U1 = xt∂/∂x + t2 ∂/∂t + 1/2(t − ix2 )u∂/∂u,

(9.6.22a)
x → x/(1 − a1 t), t → t
= t/(1 − a1 t),
u → u
= u(1 − a1 t)−1/2 exp(−i(a1 /2)x2 /(1 − a1 t)).

Q2 = x∂/∂x + 2t∂/∂t : x → x
= exp(a2 )x, t → t
= exp(2a2 )t.

(9.6.22b)
Q3 = ∂/∂t : t → t
= t + a3 . (9.6.22c)

Q4 = t∂/∂x − ix, U4 = t∂/∂x − ixu∂/∂u :

x → x = x + ta4 , u → u
= u exp(−ia4 (x + a4 t/2)). (9.6.22d)


Q5 = ∂/∂x : x → x = x + a5 . (9.6.22e)
Q6 = 1, U6 = u∂/∂u : u → u
= exp(a6 )u. (9.6.22f)
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 293

The operators so far obtained do not correspond to quantum mechani-

cal observables. However, as the c’s of (9.6.18)–(9.6.20) are arbitrary con-
stants of integration, the Q’s are determined only up to a multiplicative
constant. (We used this freedom when defining Q2 as a standard genera-
tor of dilatations.) The constants c may be chosen to be real, complex, or
pure imaginary. Let us multiply the Q’s by −i, so that wherever it appears,
∂/∂x is converted into the self-adjoint operator −i∂/∂x that represents pop .
Symmetrizing products of x and −i∂/∂x one converts the Q operators to
self-adjoint operators, Q
. The resulting Q
and their classical analogs, are:
Q
1 = −it(x∂/∂x + ∂/∂x x)/2 − it2 ∂/∂t−x2/2 → t x px − t2 E − x2 /2;
(9.6.23a)

Q
2 = −i(x∂/∂x + ∂/∂x x)/2 − 2it∂/∂t → x px − 2tE; (9.6.23b)

Q
3 = i∂/∂t → E; (9.6.23c)

Q
4 = −it∂/∂x − x → tpx − x; (9.6.23d)

Q
5 = −i∂/∂x → px ; (9.6.23e)

Q
6 = −i → 1. (9.6.23f)
The commutation relations of the Q
are given in Table 9.1 below.
Q
1 , Q
2 , Q
3 , Q
4 ,
and Q
5 are the quantal analogs of classical operators which
generate a five-parameter subgroup of the projective group in the two vari-
ables x, t. On inspecting the table one observes that Q
4 and Q
5 are members
of a Heisenberg subalgebra, so the classical and quantal subgroups are not
isomorphic. In this subgroup, Q
6 generates transformations which multiply
a wave function by the phase factor eiα . The generators Q
1 , Q
2 , Q
3 , are
members of a Lie algebra that has no Abelian or Heisenberg subalgebra.

Table 9.1. Commutators of self-adjoint invariance

generators of Eq. (9.6.1).

[Q
A , Q
B ]

Q
A \Q
B Q
1 , Q
2 , Q
3 , Q
4 , Q
5 , Q
6

Q
1 0 2iQ
1 −iQ
2 , 0 iQ
4 0
Q
2 0 2iQ
3 −iQ
4 iQ
5 0
Q
3 0 iQ
5 0 0
Q
4 0 −i 0
Q
5 0 0
Q
6 0
October 28, 2010 16:55 9in x 6in b943-ch09

294 Dynamical Symmetry

The corresponding classical and quantal three-parameter groups are locally

isomorphic. The transformations of x and t generated by Q
3 , Q
4 , Q
5 are
also those of a subgroup. This is the Galilei group if, in the case of Q
4 one
considers the group parameter to be the relative velocity of two classical
frames of reference used to measure time and position coordinates. Q
3 , Q
5
commute with Hop , but not Q
4 which commutes with Sop . Q
1 , Q
2 commute
with Sop only on the surface defined by y(x, t) − u = 0 when Sop y(x, t) = 0.

9.7 Eigenfunctions of the Constants of Motion

of a Free-Particle
The eigenfunctions of the constants of motion of a quantal system are
analogs of classical trajectories labeled by the values of constants of motion.
The eigenvalues represent the values of these constants of motion. The
classical trajectories of Hamiltonian systems may have definite values of
every constant of motion of the system, but the wave functions of the
corresponding quantal system can, in general, only have eigenvalues for
commuting constants of motion. In the quantal, as in the classical case,
transformations that leave invariant the equations of motion may be used
to convert solutions of unpertubed equations of motion to solutions of
equations perturbed by functions of the group generators.
The solutions of the time-dependent Schrödinger equation, (9.6.1a), that
are eigenfunctions of the commuting operators pop = −i∂/∂x = Q
5 and
i∂/∂t = Q
3 , are necessarily eigenfunctions of H. They may be chosen as
√
yk (x, t) = (1/ 2π) exp(ikx) exp(−ik2 t/2). (9.7.1)

One has

Q
3 yk (x, t) = (k2 /2)yk (x, t) = E yk (x, t) = H yk (x, t), (9.7.2)

and

pop yk (x, t) = p yk (x, t), p = k.

Using the Schrödinger scalar product, these yk are orthonormal in the gen-
eralized sense of Dirac’s delta function:

∞

yk , yk
 = dx yk∗ (x, t)yk
(x, t) = δ(k
− k). (9.7.3)
−∞

In this and subsequent relations, we use the standard notation where y∗

represents the complex conjugate of y.
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 295

As k may take on a continuous range of values, the general solution

y(x,t) of (9.6.1) is a continuous linear combination of energy eigenfunctions:

∞
y(x, t) = dkf(k)yk (x, t), (9.7.4)
−∞

a Fourier transform of y(k).

If Q is the self-adjoint Schrödinger operator corresponding to a classical
constant of motion q(x,t), and if y(x,t) is the normalized wave function of
a system, the mean value of measurements of the value of q(x, t) is

∞
qy
y, Qy = dx y∗ (x, t)Qy(x, t)
−∞

∞
= dx(Qy(x, t))∗ Qy(x, t) =
Qy, y. (9.7.5)
∞

The time translation operator exp(iaQ
3 ) = exp(−a∂/∂t) multiplies y by a

phase factor:
exp(−a∂/∂t)y(x, t) = y
(x, t) = exp(iak2 /2)y(x, t). (9.7.6)
The space translation operator exp(b∂/∂x) multiplies y(x,t) by the phase
factor exp(ißk). In either case, the other Schrödinger operators Q
will not
alter these factors of the general form exp(iq). Writing y
= exp(iqy )y,
one has
Q
y
= Q
exp(iqy )y = exp(iqy )Q
y. (9.7.7a)

If y is an eigenfunction of Q
, then so also is y
, and both have the same

eigenvalues. Whether y is or not an eigenfunction of Q
,

y
∗ Q
y
= y∗ exp(−iqy )Q
exp(iqy )Q
exp(iqy )y = y∗ Q
y. (9.7.7b)

If y is normalizable then one has

y
, Q
y
 =
y, Q
y. (9.7.7c)
For the one-dimensional motion of a single particle, the classical Galilei
transformations that relate observations made in coordinate systems mov-
ing at constant relative velocity have, as their generator, the PB operator
of the constant of motion tpx − x. The corresponding Schrödinger operator
is Q
4 . As Q
4 does not commute with either Q
3 or Q
5 , one cannot expect to
find sets of common eigenfunctions of all three generators. One concludes
that, in general, it is the expectation values of Schrödinger operators, rather
than their eigenvalues, that are invariant under the totality of operations
October 28, 2010 16:55 9in x 6in b943-ch09

296 Dynamical Symmetry

of the Galilei group. The previous statement allows exceptions. For exam-
ple, the time and space translation invariant eigenfunction y0 (x, t) = (1)
satisfies Q
4 y0 = 0.
The eigenfunctions of Q
4 with eigenvalue λ are of the general form
Φλ (x, t) = N(t) exp((i/t)(x2 /2 + λx)), (9.7.8)
where N(t)√is an arbitrary normalizable function of t. Thus one may choose
N(t) = (1/ 2π) exp(if(t)), with f being an arbitrary function of t. Observers
whose clocks and meter sticks agree at t = 0, will agree on the value of λ if
they are moving with uniform relative velocity.
Now, because Q
1 commutes with Q
4 , one can replace N(t) with a func-
tion of t which makes Φλ (x,t) an eigenfuntion of Q
1 . Letting C be a con-
stant, one finds the resulting functions
Φnl (x, t) = C t−1/2 exp(−in/t) exp((i/t)(x2 /2 + λx))
(9.7.9)
= C t−1/2 exp((i/t)(x2 /2 + λx − n)).
One finds that these functions satisfy Schrödinger’s equation if n = −λ2 /2.
The resulting functions,
yλ (x, t) = C t−1/2 exp((i/t)(x + λ)2 /2), (9.7.9)
also satisfy the relations
Q
1 yλ (x, t) = −(λ2 /2)yλ(x, t), Q
4 yλ (x, t) = λyλ (x, t). (9.7.10)
The probability function is
yλ (x, t)∗ yλ (x, t) = C∗ C/t, (9.7.11)
so



yλ
, yλ  = t−1 dxC∗ C = t−1 C∗ Cδ(λ
− λ).

Consequently, yλ is not normalizable for all times. The t−1/2 factor ensures
that the resulting wave functions can only be normalized for fixed nonzero
values of t. These observations lead to the conclusions that:
a. Time-dependent Schrödinger equations may be left invariant by trans-
formations that do not leave the Schrödinger scalar product invariant.
b. Time-dependent Schrödinger equations may have solutions, yn (x, t), with
constents of motion whose eigenvalues, n, do not change with time, but
the probability of finding a system in the eigenstate yn (x,t) can vary with
time, even though the eigenstate remains the same. This can be true
even when the eigenvalues correspond to values of classical constants of
motion.
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 297

In the case involved here, it will be noticed that this unpleasantness

did not appear for wavefunctions that are eigenstates of Q
4 , but it did
appear when they became eigenstates of Q
1 . Using the analysis and meth-
ods of Sec. 5.3, the reader will find that transformations with generator
U1 do not leave dx invariant, and that those with generator U2 also do
not leave dx invariant. As a result, the Schrödinger scalar product is not
left invariant by transformations with either generator. On the other hand,
the generators U3 , U4 , U5 , generate transformations that leave invariant
the Schrödinger scalar product, and Q
3 , Q
4 , Q
5 leave invariant the time-
dependent Schrödinger equation. These generators correspond to the clas-
sical generators of the Galilei group. One concludes that the Schrödinger
Eq. (9.6.1a) and the Schrödinger scalar product (9.7.3) are both invariant
under the transformations of the Galilei group.
This concludes our discussion of the quantal free-particle carried out
to illustrate some of the general connections established in Secs. 9.1–7,
their physical interpretations and consequences. The next section illustrates
further connections.

9.8 Dynamical Symmetries of the Schrödinger Equations

of a Harmonic Oscillator
For a system with potential energy function v(x), the determining equations
corresponding to (9.6.9) are
q00 00 00 10
xx + 2i qt + 4v qx + 2∂v/∂xqx = 0, (9.8.1a)
q10 00 10
xx + 2qx + 2i qt = 0, (9.8.1b)
q01
xx + 2i q01
t − 4i q10
x = 0, (9.8.1c)
q01
x = 0. (9.8.1d)
All of these equations, except the top one, are the same as the corresponding
equations in (9.6.9); the strategy for solving both sets is the same. For a
one-dimensional harmonic oscillator with
H = (p2op + kx2 )/2, pop = −i∂/∂x, (9.8.2)
one finds the following solutions of these determining equations:
√ √
Q1 = 1, Q3 = exp(i kt)(∂/∂x + kx),
√ √ (9.8.3)
Q2 = i∂/∂t, Q4 = exp(−i kt)(∂/∂x − kx).
If one allows second-order differential operators, the resulting determining
equations also produce Q23 , Q24 , Q3 Q4 , and of course, Hop itself.
October 28, 2010 16:55 9in x 6in b943-ch09

298 Dynamical Symmetry

√
Setting ω = k, define
Q3 → bt = (2ω)−1/2 exp(iωt)(ωx + ∂/∂x),
(9.8.4)
Q4 → b+
t = (ω)
−1/2
exp(−iωt)(ωx − ∂/∂x).
Here b+t is the adjoint of bt , and the operators are those introduced by
Dirac. They obey the following commutation relations:
[i∂/∂t, bt ] = −ωbt , [i∂/∂t, b+ +
t ] = ω bt ,
(9.8.5a–c)
[bt , b+
t ] = 1.

The first two equations imply that

[H, bt ] = −ω bt , [H, b+ +
t ] = ω bt . (9.8.5d,e)
One also finds
H = (ω/2)(b+ +
t bt + bt bt ). (9.8.6)
Much of the material described in this section will be found in many
textbooks on quantum mechanics. It is not as well known that the
Lie-algebraic properties of the oscillator are closely related to those of a
variety of systems. In Chapter 11, for example, it will be shown that they
are but a special case of Lie-algebraic properties of Gegenbauer’s equation,
which plays a central role in the theory of the orthogonal groups O(n).
Let us continue with this example to elucidate a general pattern of
thought. If yE has energy eigenvalue E, then (9.8.5e) implies
H b+ + + +
t yE = bt (H + ω)yE = bt (E + 1)yE = (E + ω)bt yE . (9.8.7)
Thus b+ t yE is an eigenfunction of H with eigenvalue E + ω. This shows that
H (and hence i∂/∂t) has a discrete energy spectrum.
The lowering operator bt has an important invariant function; that is,
there is a function y0
such that
(2ω)−1/2 exp(iωt)(∂/∂x + ωx)y0
= 0. (9.8.8a)
Solving this ODE yields
y0
= A
exp(−ωx2 /2). (9.8.8b)
Operating on this with H one finds
Hy0
= E0 y0
, E0 = (ω/2). (9.8.9)
As the lowering operator bt annihilates y0
there is no solution of the time
independent Schrödinger equation with eigenvalue E less than E0 : the Lie
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 299

algebraic properties of the system establish that its energy spectrum has a
bottom level.
An integration gives the result that the normalized wave function of this
bottom level is

y0 = (ω 1/2 /p1/4 ) exp(−ωx2 /2) exp(−iωt/2). (9.8.12)

The raising operator b+

t acts on y0 to give

y1
= 21/2 ωx y0 . (9.8.13)

The commutation relations (9.8.3) require that this is a solution of the

Schrödinger equations with eigenvalue 3ω/2. They furthermore require

(b+ n

t ) y0 = yn (9.8.14)

to be a solution with eigenvalue (n + 1/2)ω. These yn
are not normalized.

Using the relations obeyed by b+ t and bt one can show that





dx yn∗ yn
= n! dx y0∗ y0 = n!. (9.8.15)

Consequently the functions

yn = (1/(n!)1/2 )(b+ n
t ) y0 . (9.8.16)

are normalized. These energy eigenfunctions can be conveniently con-

structed using a computer math program to evaluate (9.8.16). The operator
b+
t has been so defined that the functions obtained do not introduce any
phase factors into the usual harmonic oscillator functions,

yn = Nn Hn (ω 1/2 x) exp(−ωx2 /2) exp(−iωt/2). (9.8.17)

The raising and lowering operators act upon the normalized functions as
follows:

b+
t yn = (n + 1)
1/2
yn+1 ,
(9.8.18)
bt yn = n 1/2 yn−1 .

If we replace x by q, and ∂/∂x by ip, in bt and b+

t we obtain the classi-
cal functions whose Poisson bracket operators are standard Lie generators
October 28, 2010 16:55 9in x 6in b943-ch09

300 Dynamical Symmetry

corresponding to bt and b+
t . One finds

bt → Bt = exp(iωt)(i∂/∂p − ∂/∂q + (q − ip)∂/∂E)/21/2,

(9.8.19)
b+ +
t → Bt = exp(−iωt)(−i∂/∂p − ∂/∂q + (q + ip)∂/∂E)/2
1/2
.

Evaluating the commutator of these operators, one obtains, as expected

from Sec. 9.4, that the Poisson bracket operator analog of the quantal rela-
tion [bt , b+
t ] = 1 is

[Bt , B+
t ] = 0. (9.8.20)

The quantal group and the group generated by the PB operators are not
isomorphic. The group generated by bt , b+
t and 1 has no isomorphic ana-
log in the phase space PQET. In contrast, the Lie algebra of the quan-
tum mechanical double-jump operators Qm = (bt )2 , Qp = (b+ 2
t ) , and the
+ +
labeling operator Qpm = bt bt + bt bt is isomorphic to the corresponding
classical Lie algebra of PB operators.

9.9 Use of the Oscillator Group in Pertubation Calculations

When one can express a Hamiltonian in the form H = H0 + v, where
H0 = T + V, and v is in some sense “small”, much can be learned if
v can be expressed as a function of operators of the invariance group of
the Schrödinger equations involving H. The situation is analogous to that
occuring in classical Hamiltonian mechanics, and once again the simplest
example is provided by the harmonic oscillator. On setting t = 0 in the
oscillator bt , b+
t ,

bt → b = (2ω)−1/2 (ωx + ∂/∂x),

(9.9.1a,b)
b+ +
t → b = (2ω)
−1/2
(ωx − ∂/∂x).

Hence

x = (2/ω)1/2 (b+ + b), ∂/∂x = (2ω)1/2 (b+ − b). (9.9.2a,b)

If v = v(x) can be expanded in a power series in x, it can be expressed as a

power series in (b+ + b). One can then use the expressions to determine the
levels that interact under ther influence of the perturbation, and evaluate
matrix elements of v. If v is velocity dependent, one can make similar use
of (b+ − b). Examples can be found in the exercises.
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 301

9.10 Concluding Observations

In this chapter we have dealt with fundamental principles needed to uncover
and interpret the intrinsic symmetries of systems obeying Schrödinger equa-
tions or their momentum-space analogs. This required a treatment of cor-
respondences between classical and quantal systems, and the development
of a Lie algebraic extension of the correspondence principle. The extension
enables one to determine whether the intrinsic symmetries of the quantum
mechanical equations correspond to geometric symmetries in classical phase
space.
A method for determining the generators of invariance transforma-
tions of Schrödinger equations has been described. Several important and
straightforward, but less than obvious, aspects of the mathematics were
noted. The resulting determining equations were used to determine the
generators of transformations which leave invariant the time-dependent
Schrödinger equation governing a free-particle moving in one spatial
dimension, and the generators of transformations that leave invariant
the time-dependent Schrödinger equation for a one-dimensional harmonic
oscillator.
The free-particle equation was found to admit a six-parameter dynami-
cal group isomorphic to the corresponding group of Hamilton’s equations in
PQET space. The intrinsic symmetry of the Schrödinger equation does not,
however, correspond to the geometric symmetry of the classical free parti-
cle in phase space. The six-parameter group contains a three-parameter
subgroup of Galilei transformations. The Schrödinger scalar product is
invariant under the transformations of this subgroup. The generators of
its time translation and space translation groups commute. The remain-
ing generator of the Galilei group does not commute with these. The
Schrödinger scalar product is not invariant under all transformations of the
six-parameter group that leave invariant the time-dependent Schrödinger
equation of the free-particle. It follows from these observations that the
quantum mechanical consequences of the classical dynamical symmetries
are more surprising than one might expect.
The time-dependent Schrödinger equation for the oscillator admits a
three-parameter invariance group. Two of its generators are the well known
shift operators that generate the energy spectrum of the system, and con-
vert eigenfunctions Ψn to Ψn+1 and Ψn−1 . The form of its lowering opera-
tor requires that the system has a lowest energy state, and determines the
functional form of this ground state. Together with the energy operator,
October 28, 2010 16:55 9in x 6in b943-ch09

302 Dynamical Symmetry

the operators generate a three-parameter Lie group. This group has no

isomorphic classical analog: it defines a dynamical symmetry that is not
isomorphic to the geometric symmetry in phase space defined by corre-
sponding PB operators.
Historical Note
Lie did not develop a general theory of the invariance transformations of
partial differential equations of order greater than one. Noether, in her
treatment of the variational principles and conservation laws of mechan-
ics, introduced group generators that were second-order partial deriva-
tive operators.5a Johnson was the first to generalize Lie’s theory of ODEs
to PDEs of arbitrary order.5b The lack of a theory of integration for
the resulting infinite set of differential equations defining finite trans-
formations stymied further developments of Johnson’s theory (personal
communication). In Ref. 5c it is argued that the generators of invariance
transformations of second-order PDEs could not, in general, be second-
order differential operators. Unaware of these mathematical developments,
but knowing that second-order PDE’s of physics admitted invariance gen-
erators that are second-order differential operators, several physicists in
the early 70’s independently began to use second- and higher-order differ-
ential operators as group generators.5d,5e The theory of invariance trans-
formations of Schrödinger equations was first elaborated in Ref. 5e by
Kumei, who then extended the theory to nonlinear PDEs in Ref. 5f. In
1978, Ibragimov and Anderson5g found the error in Ref. 5c, the implicit
assumption of a finite dimensional space of derivatives, a “finite dimen-
sional jet space”. They introduced the name Lie–Backlund transformations
for the group operations generated by differential operators of arbitrary
order. (The transformations connect conceptions originating with Lie and
Backlund, but because neither Lie nor Backlund used such transformations,
some authors prefer to call them generalized Lie transformations.) There
are now many articles that exploit these transformations, and by chang-
ing the definition of the base variables, integral-dependent transformations
have been developed.5h

Exercises
(1,2)
1. Compute the commutator of QPj with qj and with p2jop .
2. Establish the condition under which the Q1 and Q2 of Sec. 9.6 can be
expressed in terms of products of Q3 , Q4 , Q5 .
October 28, 2010 16:55 9in x 6in b943-ch09

Dynamical Symmetry in Schrödinger Quantum Mechanics 303

3. Does using the symmetrized self-adjoint form of it2 ∂/∂t in Q
1 produce
an operator that leaves invariant the surface defined by Sop y(x, t) = 0
in Eq. (9.6.1a)?
4. Solve the determining Eqs. (9.8.1) when the potential v = kx or k/x2 ,
or simply a constant, k.
5. Using the Bt , B+ t given in Eqs. (9.8.19), determine the finite transfor-
mations of p, q, E, t carried out by the groups generated by (Bt + B+ t )/2
and i(Bt − B+ t )/2.
6. Obtain real linear combinations of the oscillator shift operators and com-
pute their commutation relations. Do the same with the corresponding
Poisson bracket operators. Determine the finite transformations of p, q,
E, t generated by the Poisson bracket operators.
7. Use the scaling properties of the harmonic oscillator Hamiltonian to
prove that

Y(x), Top Y(x) =
Y(x), VY(x).

References
[1] W. Heisenberg, Z. Physik, 33 (1925) 879.
[2] M. Born and P. Jordan, Z. Physik 34 (1925) 858. For translations, of this and
the previous reference, see Sources of Quantum Mechanics, ed. B. L. Van Der
Warden (North-Holland, Amsterdam, 1967).
[3] a) H. J. Groenewald, Physica 12 (1946) 405–460.
b) L. Van Hove, Acad Roy. Belgique Bull. Cl. Sci. (5) 37 (1951) 610–620.
c) For a recent mathematical review see, Mark J. Gotay, in Mechanics: From
theory to computation, J. Nonlinear Science (2000) 271–316.
[4] C. Wulfman, J. Phys. Chem. A 102 (1998) 9542-8; J. Phys. A 42 (2009)
185301.
[5] a) E. Noether, Nachr. Konig. Gesell. Wissen. Gottingen, Math.-Phys. p. 23;
English translation: Transport Theory Stat. Phys. I, Vol. 186 (1971).
b) H. H. Johnson, Proc. Am. Math. Soc. 15 (1964) 432, 675 .
c) E. A. Muller and K. Matschat, Miszellaneen der Angewandten Mechanik
(1962) 190.
d) Z. Khukhunashvili, Izvestiy Fizika Vyss. Ucheb. Zavadenii 3 1971 95–103.
e) S. Kumei, M. Sc. Thesis, University of the Pacific (1972); R. L. Anderson,
S. Kumei, C. Wulfman, Phys. Rev. Lett. 28 (1972) 988.
f) S. Kumei, J. Math. Phys. 16 (1975) 2461; 18 (1977) 256.
g) N. Ibragimov and R. L. Anderson, Dokl. Akad. Nauk. SSSR 227 (1976)
539.
h) G. W. Bluman and S. Kumei, Symmetries and Differential Equations
(Springer, N.Y., 1989).
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-ch10

CHAPTER 10

Spectrum-Generating Lie Algebras and

Groups Admitted by Schrödinger Equations

Introduction
The generators of invariance transformations of time-dependent Schrödin-
ger equations provide operators that can change the eigenstates and
eigenenergies determined by the corresponding time-independent equa-
tions. The previous chapter considered two examples of time-dependent
Schrödinger equations whose invariance groups provide operators that com-
muted with H, along with those that did not commute with H. The eigen-
values of the group generators that commuted with H were used to label the
energy eigenstates, and the generators that did not commute with H were
used to change eigenstates with a given energy into states with a different
energy. Both group generators and group operators were used for the lat-
ter purpose. Groups whose generators and/or operators convert an energy
eigenstate into such sets of eigenstates are said to be spectrum generating.
Infeld and Hull developed a method which uncovered spectrum gen-
erating groups,1 long before the extension of Lie’s theory of differential
equations to Schrödinger equations was available.2 Their method, the fac-
torization method, depends upon factoring a Hamiltonian into a product of
raising and lowering operators, and determines the generators of a three-
parameter group. A generalization produces Hamiltonians termed super-
symmetric, along with the Lie algebra of their three-parameter groups.3
In this chapter we develop a characterization of spectrum-generating Lie
algebras and use the generalized Lie theory of differential equations to
determine generators of the dynamical groups of a number of quantum
mechanical systems. Appendix 10.A lists references in which the dynamical
groups of other Schrödinger equations are displayed.

305
October 28, 2010 16:55 9in x 6in b943-ch10

306 Dynamical Symmetry

As a Hamiltonian operator, H, commutes with all the generators of its

invariance group, it commutes with the Casimir operator(s) of its invari-
ance group. Consequently, the eigenfunctions of H can be chosen to be
eigenfunctions of both Casimir operators and a commuting set of gener-
ators of the group. An example is provided by the wave functions ψnlm
of a hydrogen atom. The quantum number n fixes the eigenvalue of the
quadratic Casimir operator of the O(4) invariance group of H, the quan-
tum number l fixes the eigenvalue of the Casimir operator L2 of its SO(3)
subgroup, and m is the eigenvalue of a generator of this SO(3). Operators of
SO(4) can alter the value of l and m, but not n. Operators of the invariance
group of the time-dependent Schrödinger equation can alter n, as well as
l , m. Hydrogen atoms, and most quantal systems, have both discrete and
continuous spectra. The forthcoming sections show how the operators that
generate these two types of spectra are characterized by their commutation
relations.

10.1 Lie Algebras That Generate Continuous Spectra

In Chapter 9 it was shown that the time-dependent Schrödinger equation
governing a free particle moving along the x axis admits a Lie generator
Q2 = x∂/∂x + 2t∂/∂t. Q2 generated transformations that convert a given
solution of the Schrödinger equations into a continuous family of solutions,
each member of the family having a different energy. This implied the free
particle possessed a continuous energy spectrum. The generator Q2 satisfies
the commutation relation
[H, Q2 ] = −2 H, (10.1.1a)
and hence, also,
[i∂/∂t, Q2 ] = −2 i∂/∂t. (10.1.1b)
In the next two paragraphs we establish that commutation relations of this
form imply that H and i∂/∂t have a continuous spectrum.4
Let Q be the generator of an invariance transformation of a time-
dependent Schrödinger equation in any number of spatial variables qj . Sup-
pose Q satisfies the relations
[Q, H] = b H, [Q, i∂/∂t] = b i∂/∂t, −∞ < b < ∞. (10.1.2a,b)
Let Ψ(q, t), q = (q1 , q2 , . . .), be any eigenstate of H, and i∂/∂t, satisfying
the equations
HΨ = i∂Ψ/∂t, HΨ = EΨ, (10.1.3a,b)
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 307

and physically appropriate boundary conditions. Let Ψ0 represent the eigen-

state with energy E0 . We shall shortly prove that if
Ψαb,0 = exp(−αQ)Ψ0 , (10.1.4a)
then
HΨαb,0 = i∂Ψαb,0 /∂t = E(αb, 0)Ψαb,0 , (10.1.4b)
with
E(αb, 0) = exp(αb)E0 . (10.1.4c)
Thus if E0 is the energy eigenvalue for Ψ0 , then E(αb, 0), as a continu-
ous function of the group parameter α, is the corresponding eigenvalue for
Ψαb,0 = exp(αQ)Ψ0 . It follows, that if Ψαb,0 satisfies the required bound-
ary conditions, H must possess a continuous energy spectrum, and Ψαb,0
will represent the eigenstate with energy E(αb, 0).
To prove this claim, we make use of the general expansion:
exp(αQ)B exp(−αQ) = B + [αQ, B] + (1/2!)[αQ, [αQ, B]] + · · · (10.1.5)
in which the n-th term is 1/n times the commutator of αQ with the previous
term. If [Q, B] = bB, then (10.1.5) implies
exp(αQ)Bexp(−αQ) = B + αbB + ((αb)2 /2!)B + · · ·
(10.1.6)
= exp(αb)B.
Applying this result to (10.1.3a), and using (10.1.2a), one has
exp(−αQ)HΨ0 = exp(−αQ)EΨ0 = E0 exp(−αQ)Ψ0 ,
and
exp(−αQ) H exp(αQ) exp(−αQ)Ψ0 = E0 exp(−αQ)Ψ0 , (10.1.7)
so
exp(−αb) H exp(αQ)Ψ0 = E0 exp(−αQ)Ψ0 .
Hence
HΨαb,0 = exp(αb)E0 Ψαb,0 = E(αb, 0)Ψαb,0 .
Applying the same process with H replaced by i∂/∂t one obtains the ana-
log of Eq. (9.6.29) in which H is replaced by i∂/∂t. It follows that if the
Schrödinger equation (H − i∂/∂t)Ψ = 0 has an invariance generator Q
which satisfies Eq. (10.1.2), and Ψαb,0 satisfies the proper boundary condi-
tions, then H and i∂/∂t have a continuous spectrum.
October 28, 2010 16:55 9in x 6in b943-ch10

308 Dynamical Symmetry

10.2 Lie Algebras That Generate Discrete Spectra

Suppose that a time-dependent Schrödinger equation admits a set of gen-
erators Nj and Qj± which on the solution space Ψ(q, t) satisfy
[(H − i∂/∂t), Nj ] = 0, [(H − i∂/∂t), Qj±] = 0, (10.2.1a,b)
and which have the following commutation relations
[H, Nj ] = 0 = [i∂/∂t, Nj ], [Nj , Qj± ] = ±bj Qj , j = 1, 2, . . . ,
(10.2.2a,b)
where the bj are finite constants. Equation (10.2.2a) indicates that the
eigenfunctions of H may be chosen to be eigenfunctions of the operators
Nj . Equation (10.2.2b) implies that if Ψnj , is an eigenfunction of Nj with
eigenvalue nj , then
Nj Qj± Ψnj = Qj± (Nj ± bj )Ψnj = (nj ± bj )Qj± Ψnj . (10.2.3)
Thus Qj± Ψnj is an eigenfunction of Nj with eigenvalue (nj ± bj ), and Nj has
a discrete spectrum.
In the general case, one may choose a set of commuting Nj , and label
energy eigenfunctions, Ψ, with the eigenvalues nj of the Nj , so
Nj Ψnj = nj Ψnj , (10.2.4a)
and
HΨn = E(n)Ψn , HΨn = i∂Ψn /∂t, n = (n1 , n2 , . . . , nj , . . . , nm ).
(10.2.4b,c)
Then
Qj± Ψn = Ψn
, n
= (n1 , n2 , . . . , nj ± bj , . . . , nm ), (10.2.4d)
where Ψn
is also a solution of both Schrödinger’s equations. As the oper-
ators need not convert normalized solutions of Schrödinger equations into
normalized solutions, it is convenient to replace (10.2.4d) by
Qj± Ψn = c± (nj )Ψn
(10.2.5)
in which Ψn and Ψn
are normalized, and the c± (nj ) are normalizing factors.
If for some value of nj , c+ (nj ) = 0, the spectrum of Nj has an upper
bound; if for some value of nj , c− (nj ) = 0, its spectrum has a lower bound.
If the Qj± do not commute with H, then different values of nj produce
different values of E. If the Qj± commute with H, then E is not changed by
the shifting of the values of nj , and H has a spectrum that is degenerate.
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 309

Equations (10.2.1c) are equations of the type used by Cartan to clas-

sify the Lie algebras of semisimple groups. The symplectic groups Sp(2N)
are semisimple groups. The Lie-Algebraic Correspondence Principle, and
the fact that the invariance groups of Hamilton’s equations are symplectic
groups, gives (10.2.1c) a wide range of applications in quantum mechanics.
One often wishes to label eigenfunctions with the eigenvalues of Casimir
operators rather than group generators. To alter the eigenvalues of the
Casimir operator of a group, one must use operators that do not commute
with it — operators not contained in the group. The next two sections deal
with examples in which the label l in functions ψlm can be changed by
operators which do not commute with the generators of SO(3).

10.3 Dynamical Groups of N-Dimensional

Harmonic Oscillators
Let the Schrödinger Hamiltonian of a one-dimensional oscillator be
expressed in terms of Cartesian coordinates by

Hj = −(∂/∂qj )2 /2 + (kj /2)q2j . (10.3.1a)

√ √
Setting ωj = kj , and substituting yj = qj ωj into (10.3.1a), converts it to

Hj = (ωj /2)(−∂ 2 /∂yj2 + yj2 ). (10.3.1b)

The Dirac energy raising and lowering operators, Hj are, respectively,

b†j = exp(−iωj t)(yj − ∂/∂yj ), (10.3.1c)

bj = exp(iωj t)(yj + ∂/∂yj). (10.3.1d)

Expressed in terms of them,

Hj = (ωj /2)(b†j bj + bj b†j ). (10.3.1e)

The Hamiltonian for the corresponding N-dimensional oscillator is

H = ΣN
1 Hj . (10.3.2)

The oscillator is isotropic if all the ωj have the same value, ω. The ground
state of such an oscillator has energy

E0 = (N/2)ω. (10.3.3)

In 1956, Baker5 showed that because this H is invariant under opera-

tions of the unitary group, U(N), the n-th energy level of the oscillator,
October 28, 2010 16:55 9in x 6in b943-ch10

310 Dynamical Symmetry

with energy

En = E0 + nω, (10.3.4)

has degeneracy
(N + n − 1)!
D= . (10.3.5)
n!(N − 1)!
Fowler had previously determined D by working out the way in which n
quanta can be assigned N degrees of freedom.6 The operator that shifts a
quantum of energy from degree of freedom j to degree of freedom k is

b†k bj = (1/2)(yk − ∂/∂yk )(yj + ∂/∂yj). (10.3.6)

Its adjoint, b†j bk , carries out the inverse transfer. These N(N − 1) opera-
tors together with the N operators Hj comprise a set of N2 operators that
generate the degeneracy group U(N).
Several different types of dynamical groups are known for N-dimensional
isotropic harmonic oscillators. One of them, denoted NU(N) ,7 is generated
by these operators, together with the 2N Dirac shift operators and the
identity operator of the Heisenberg group “N”.
A dynamical group of three-dimensional isotropic harmonic oscillators
that is particularly useful is the group Sp(6), whose Lie algebra contains
the double-jump operators mentioned above. It may be used to interconvert
states containing odd numbers of excitation quanta, or interconvert states
containing even numbers of these quanta. For a more extensive discussion,
the reader is directed to Chapter 20 of Wybourne’s book,13 and references
therein.
The dynamical group of non-isotropic oscillators is the same as that
of the isotropic ones, but the generators b†k bj involving degrees of freedom
with different ω’s become time-dependent operators which alter energies,
and cease being generators of the degeneracy group. However, if the ω’s in
two different degrees of freedom are so related that there are two integers
Ia and Ib such that ωa /ωb = (Ia /Ib ), then degeneracies will appear when
na /nb = (Ib /Ia ).
The dynamical symmetries of harmonic oscillators have come to play
a large role in nuclear shell theory. This begins with the assumption that
nucleons move in approximate harmonic oscillator potential wells. It then
develops the consequences of j–j coupling between spins of the nucleons.
Appendix 10.B lists a number of references which will introduce the reader
to this application of dynamical symmetries which is now very extensive.
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 311

10.4 Linearization of Energy Spectra by Time Dilatation;

Spectrum-Generating Dynamical Group
of Rigid Rotators
The spectrum of eigenvalues of a Cartan labeling operator Nj is linear:
nj → nj
= nj + bj ,
(10.4.1)
nj
→ nj

= nj
+ bj = nj + 2bj ,
and so forth. When the energy of a system is not a linear function of its
quantum numbers, requiring that Nj satisfy
[H, Nj ] = 0, [i∂/∂t, Nj ] = 0, (10.4.2a,b)
does not ensure that it satisfies
[Nj , Qj± ] = dbj Qj± , d = 1, 2, . . . . (10.4.2c)
When one knows the dependence of E upon a quantum number n, methods
developed by Kumei8 enable one to circumvent this difficulty. They produce
determining equations for Cartan labeling and shift operators analogous to
those obtained for the harmonic oscillator in Sec. 9.8.
Given the Schrödinger equations
HΨ = i∂Ψ/∂t, HΦn = E(n)Φn , (10.4.3a,b)
and let
Ψ = Σcn Φn (q) exp(−iE(n)t). (10.4.3c)
One begins by making a transformation that converts exp(−iE(n)t) to
exp(−int). To accomplish this, let E = λ(n), and let N be an operator
N(q, pop ) with eigenvalues n, so H = λ(N), and N commutes with H. The
dilatation operator
D = exp(αt∂/∂t), α = ln(N/λ(N)), (10.4.4a)
then converts
N
t → t
= eα t = t, (10.4.4b)
λ(N)
and
λ(N)
∂/∂t → ∂/∂t
= e−α ∂/∂t = ∂/∂t. (10.4.4c)
N
In these expressions N must not, of course, be a function of t or ∂/∂t. As
N commutes with H, the operator (10.4.4a) converts
(H − i∂/∂t)Σcn Φn (q) exp(−iE(n)t)
October 28, 2010 16:55 9in x 6in b943-ch10

312 Dynamical Symmetry

to
D(H − i∂/∂t) Σcn Φn (q) exp(−iE(n)t)
= (H − D i∂/∂tD−1 )DΣcn Φn (q) exp(−iE(n)t)
= (H − (λ(N)/N)i∂/∂t)Σcn Φn (q) exp(−int). (10.4.5)
As

N Φn (q) exp(−int) = i∂/∂t Φn (q) exp(−int) = n Φn (q) exp(−int),

(10.4.6a)
on writing

Ψ
= Σcn Φn (q) exp(−int), (10.4.6b)

one obtains a transformed version of (10.4.3a,b), which has been con-
verted to
(H − λ(N))Ψ
= 0, (10.4.6c)
and, equivalently to

(H − λ(i∂/∂t))Ψ
= 0. (10.4.6d)
As both N and i∂/∂t have linear spectra, both may be considered
to be Cartan labeling operators. Consequently, on seeking operators
Q(q, t, pop , i∂/∂t) that generate transformations which leave equation
(10.4.6d) invariant, one expects to find Cartan shift operators Q± for N
and i∂/∂t.
The determination of the spectrum-generating group of a rigid rotor
provides an example of the foregoing general discussion. Because, of the
rotor’s assumed rigidity, its Schrödinger wave function does not have a
radial dependence, and the Schrödinger equations for a rotor with moment
of inertia I, are
(L2 /2I − i∂/∂t)Ψ(t, θ, φ) = 0, (L2 /2I − El )Ψl (t, θ, φ) = 0,
(10.4.7a,b)
with

El = l (l + 1)
2 /2 I. (10.4.7c)
Here, L2 is the total angular momentum operator. The operator Lz =
−i
∂/∂φ commutes with L2 , and has eigenvalues m
. The two operators,
together have eigenstates Φlm (t, θ, φ), with m an integer constrained by
the condition |m| ≤ l . The Ψ(t, θ, φ) can be a linear combination of such
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 313

eigenstates, and the general solution can be put in the form

Ψ(t, θ, φ) = Σl Σm clm Φlm (t, θ, φ) (10.4.7d)
the clm being constants.
Multiplying Eqs. (10.4.7a,b,c) by 2I/
2 gives
(L2 − i∂/∂t )Ψ = 0, (L2 − El )Ψ = 0, El = l (l + 1),
with
L2 = (L2 /
2 ) = −{(sin(θ)−1 ∂/∂θ(sin(θ)∂/∂θ) + sin(θ−2 ∂ 2 /∂φ2 },
t = (
2 /2I)t, El = (2I/
2 )El .
(10.4.8a–c)
Expressed in terms of spherical harmonics, Ylm (θ, φ), the general solution
of (10.4.8a) is
Ψ = Σl Σm clm ψlm (x, φ, t), (10.4.9)
with
ψlm (x, φ, t) = exp(−il (l + 1)t)Ylm (θ, φ).
Given (10.4.9) one seeks operators Q that satisfy
(L2 − i∂/∂t )QΨ = 0. (10.4.10)
The operator that depends only upon q, pop and has eigenvalues l is
γ = (−1 + sqrt(1 + 4L2 ))/2. (10.4.11a)
As L2 → γ(γ + 1), Eqs. (10.4.4) require that
Dt = exp(ln(γ + l )t ∂/∂t ). (10.4.11b)
Dt converts exp(−il (l +1)t) to exp(−il t) and carries out the transformation
i∂/∂t → (i∂/∂t)(i∂/∂t + 1). (10.4.12)
It converts Eq. (10.4.10) to a determining equation for the Q
:
AQ
f = 0, (10.4.13a)
with
A = L2 − (i∂/∂t )(i∂/∂t + 1), (10.4.13b)
f ≡ Dt Ψ = Σclm exp(−il t)Ylm (θ, φ), (10.4.13c)


Q = Dt QD−1
t . (10.4.13d)
It is convenient to make the change of variables x = cos(θ), y = sin(θ)
whence
L2 = −y2 ∂ 2 /∂x2 + 2x∂/∂x − y−2 ∂ 2 /∂φ2 , y 2 = 1 − x2 . (10.4.14)
October 28, 2010 16:55 9in x 6in b943-ch10

314 Dynamical Symmetry

Seeking Q
that contain derivatives no higher than the second, Kumei set

Q
= Q
φφ ∂ 2 /∂φ2 + Q
xφ ∂ 2 /∂φ∂x + Q
x ∂/∂x

+ Q
φ ∂/∂φ + Q
t ∂/∂t + Q
0 . (10.4.15)

The derivatives in the determining Eq. (10.4.13) produce 13 linearly inde-

pendent derivatives of f. This provides 14 linearly independent functions
which one may choose to be:

f, ft , fφ , fx , fφφ , fxφ , ftφ , ftt , fxt , fφφφ , fxφφ , fttφ , fxtφ , ftφφ (10.4.16)

Inserting (10.4.13b) and (10.4.15) into (10.4.13a) and collecting the terms
that multiply each of the functions in (10.4.16) he used the linear indepen-
dence of the functions to obtain the following 14 equations:

Q
φφ
t = 0, Q
xφ
t = 0, Q
φφ
x −y
−4
xφ
Qφ = 0, Q
xφ
x + xy
−2
xφ
Q = 0,
Q
φφ
φ − xy
−2
xφ
Q = 0, Q
t
x −y
−2
x
Qt = 0, Q
t
x
t − Qx − xy
−2
x
Q = 0,
Q
φ
t −y
−2
t
Qφ = 0, AQ
xφ − 2Q
xφ − 2y−2 Q
x 2
φ
φ − 2y Qx = 0,

AQ
φφ + 2y−2Q
x
x − 4x y Q − 2y2 Q
φ
2 −4
x
φ = 0, (10.4.17)

x
x
AQ − 2Q − 4xQ
x
x − 4x y 2 −2
x
Q − 2y 2
Q
0
x = 0,

t
AQ + 2Q
0
t + 2iQ
x
x + 2ixy −2
x
Q = 0,

φ −2
AQ − 2y Q
0
φ = 0, AQ = 0.
0

Solving these equations yields 14 linearly independent operators for Q
,

three of which are products of the others. The remaining 11 operators may
be chosen to be

Q
0 = 1, Q
1,2 = e±iφ (y ∂/∂x − i(±x/y)∂/∂φ),

Q
3 = ∂/∂φ, Q
4 = e−it (iy2 ∂/∂x + x ∂/∂t − ix),
(10.4.18)
Q
5 = eit (−iy2 ∂/∂x + x∂/∂t), Q
6 = ∂/∂t,
Q
7,8 = eit e±iφ (ixy∂/∂x ± y−1 ∂/∂φ + y∂/∂t),
Q
9,10 = e−it e±iφ (−ixy∂/∂x ± (−y−1 )∂/∂φ + y∂/∂t − iy).

The corresponding operators that act on the functions Ψ of (10.4.9)

are then

Qj = D−1

t Qj Dt . (10.4.19)

The Qj and Q
j , with j = 1, . . . , 10, generate SO(3,2) groups.

October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 315

10.5 The Angular Momentum Shift Algebra; Dynamical

Group of the Laplace Equation
The results of the previous section may be used to construct shift and
labeling operators for the angular momentum eigenfunctions, and skew- or
self-adjoint operators of the SO(3, 2) group that acts upon these functions.
We shall suppose that units have been chosen in which
has unit value. The
simplest way to proceed is to makes use of the functions having the form
given in (10.4.13c). The Schrödinger scalar product of two such functions,
f and f’, is

2π
π
f, f
 = dφ sin(θ)dθ f ∗ f
. (10.5.1a)
0 0
Using the substitution x = cos(θ), sin(θ)dθ = −d(cos(θ)) = −dx, yields

2π
1
f, f
 = dφ dx f ∗ f
. (10.5.1b)
0 −1

Thus ∂/∂x = −(1/sin(θ))∂/∂θ is skew-adjoint, while i∂/∂x is self-adjoint:

f, (d/dx)f
 = (−d/dx)f, f
 = −(d/dx)f, f
,
and
f, (id/dx)f
 = if, (d/dx)f

= (id/dx)f, f
 = (i∗ )(−d/dx)f, f
. (10.5.2)
We shall suppose that the spherical harmonics Ylm (θ, φ) are normalized
to unity and that they have Condon and Shortley phase factors, so that
Ylm (θ, φ)∗ = (−1)m Yl−m (θ, φ). (10.5.3)
The m labeling and m shift operators for the functions exp(−il t)Ylm that
correspond to the standard operators for the Ylm are 9 :
M0 = −iQ
3 = −i∂/∂φ, (10.5.4a)
M+ = −Q
2 =e +iφ
(−y∂/∂x + ixy −1
∂/∂φ)
(10.5.4b)
+iφ
=e (∂/∂θ + i cot(θ)∂/∂φ),
M− = Q
3 = e−iφ (y∂/∂x + ixy−1 ∂/∂φ)
(10.5.4c)
= e−iφ (−∂/∂θ + i cot(θ)∂/∂φ).
They have long been known to satisfy the standard commutation relations
of the A2 Lie algebra of SO(3):
[M0 , M+ ] = M+ , [M0 , M− ] = −M− , [M+ , M− ] = 2M0 . (10.5.5)
October 28, 2010 16:55 9in x 6in b943-ch10

316 Dynamical Symmetry

Their actions on normalized functions ψlm are

M0 ψlm = mψlm ,
M+ ψlm = ((l − m)(l + m + 1))1/2 ψlm+1 , (10.5.6)
M− ψlm = ((l + m)(l − m + 1))1/2 ψlm−1 .
The M± are skew-adjoint:
< ψlm
, M± ψlm >= − < M± ψlm , ψlm
> . (10.5.7)
In terms of Cartesian coordinates,
z1 = r sin(θ) sin(φ), z2 = r cos(θ) sin(φ), z3 = r cos(φ) :
L12 = M0 = −i(z1 ∂/∂z2 − z2 ∂/∂z1 ),
L23 = (M+ + M− )/2 = −i(z2 ∂/∂z3 − z3 ∂/∂z2 ), (10.5.8)
L31 = −i(M+ + M− )/2 = −i(z3 ∂/∂z1 − z1 ∂/∂z3 ).
The Lij = −Lji obey standard quantum mechanical SO(3) commutation
relations
[Lab , Lbc] = −i Lac . (10.5.8)
The Casimir operator of the group is L2 , and for each value of l the set of
functions ψlm provide the basis for a (2l + 1) dimensional UIR of the group
SO(3) with the generators (10.5.8)
The determining equations for the Q’s also yield an l labeling operator,
L
0 = i∂/∂t
and l shift operators,
L
+ = iQ4 , and L
− = −iQ5 ,
with
L
+ = e−it (−y2 ∂/∂x + ix ∂/∂t + x) = e−it (sin(θ)∂/∂θ + cos(θ)(i∂/∂t + 1)),
L
− = eit (y2 ∂/∂x + ix ∂/∂t) = eit (−sin(θ)∂/∂θ + cos(θ)(i∂/∂t)).
(10.5.10a, b)
It is convenient to define the l labeling operator to be
L
0 = i∂/∂t + 1/2, (10.5.10c)
The l shift operators have the following action on the ψlm = exp(−i l t)Ylm :
L
± ψlm=c± (l,m) ψl±1m
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 317

with
1/2
c+ (l , m) = {(2 l + 3)/(2 l + 1)} ((l + 1 + m)(l + 1 − m))1/2 ,

and

c− (l , m) = {(2l + 1)/(2l − 1)}1/2 ((l + m)(l − m))1/2 . (10.5.11)

The L
s satisfy the commutation relations

[L
0 , L
+ ] = L
+ , [L
0 , L
− ] = −L
− , [L
+ , L
− ] = 2L
0 . (10.5.12)

The operators

L
45 = −L
54 = L
0 , K
34 = K
43 = i(Q
4 + Q
5 )/2,

and

K
35 = K
53 = (Q
4 − Q
5 )/2. (10.5.13)

obey the quantum mechanical SO(2,1) commutation relations

[L
ab , K
bc ] = −i K
ac , [K
ab , K
bc ] = −i L
ac . (10.5.14)

No linear combination of L
+ and L
− is self-adjoint under the scalar

product (10.5.1a). The similarity transformation



1/2
−1/2
L
± → L
± = L0 L± L0 (10.5.15)

corrects this, and removes the factors in curly brackets from the expressions
for the c± (l , m) in (10.5.11), so that

L
+ ψlm = ((l + 1 + m)(l + 1 − m))1/2 ψl+1m (10.5.16a)

and

L
− ψlm = ((l + m)(l − m))1/2 ψl−1m . (10.5.16b)

The transformation does not affect the other group generators L’. All may
be subjected to the inverse of the spectrum-linearizing transformation,
which does not change the relations (10.5.6) and (10.5.16).
When written in spherical polar coordinates, Laplace’s equation,

∆Ψ = 0, (10.5.17a)

becomes

{∂ 2 /∂r2 + (2/r)∂/∂r − L2 /r2 }Ψ(r, θ, φ) = 0. (10.5.17b)

October 28, 2010 16:55 9in x 6in b943-ch10

318 Dynamical Symmetry

Separating it yields solutions of the form

Ψ = Σl Σm clm rl Ylm (θ, φ). (10.5.17c)

In this case, the spectrum may be linearized by the dilation operator Dr =

exp(ln(r∂/∂r)). Now

∆rl Ylm (θ, φ) = 0, (10.5.18)

consequently, Dr converts (10.5.15) to

{l (l + 1)/r2 − L2 /r2 }Σl Σm clm rl Ylm (θ, φ) = 0. (10.5.19)

In this case it has been the radial variable r, rather than time, t, which has
been used in the linearization of the spectrum of an operator.
In Cartesian coordinates, the invariance generators of the Laplace equa-
tion are:
the generators of the Euclidean group E(3):

Tj = ∂/∂xj , Rjk = xj ∂/∂xk − xk ∂/∂xj; (10.5.20a,b)

the generator of dilations

D = Σ xj ∂/∂xj + 1/2, (10.5.20c)

and the three generators of special conformal transformations:

Sj = 2xj D − (x21 + x22 + x23 )∂/∂xj . (10.5.20d)

They satisfy commutation relations isomorphic to those of SO(4,1). This

is most easily seen in quantum mechanical applications if one makes the
identifications

Ljk = −i Rjk , Lj4 = −iSj , K45 = −iD,

Kj5 = −iTj , j = 1, 2, 3.
(10.5.21)
The operator exp(α · S) has the following action on functions f(x):

exp(α · S)f(x) = (1 − 2α · x + α2 x2 )−1/2 f(x
), (10.5.22)

with

x
= (x − αx2 )/(1 − 2α·xx2 ).

Solutions of the Laplace equation are also converted to solutions by

the reflections Rj xk = −δjk xk , and by the inversion transformation, U,
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 319

defined by

Uf(x) = (x−2 )f(x/x2 ). (10.5.23)
The reflections and the inversion are their own inverse, and
UTj U−1 = −S, URjk U−1 = Rjk , UDU−1 = −D. (10.5.24)
On the space of solutions, Ψ, of the Laplace equation, because ∆Ψ = 0,
one has
T · T Ψ = 0, S · S Ψ = 0, T · R Ψ = 0, (10.5.25a–c)
and
(R · R+D2 )Ψ = 1/4Ψ. (10.5.25d)
In Chapter 11 the reader will find that the invariance group of Laplace’s
equation and its generalization to Euclidean four-space becomes relevant to
the theory of the hydrogen atom. In Chapter 14, the generalization of the
group to Minkowski space provides invariance transformations of Maxwell’s
equations.

10.6 Dynamical Groups of Systems with Both Discrete

and Continuous Spectra
The invariance groups of quantum mechanical systems with both discrete
and continuous spectra are noncompact. These noncompact groups have at
least a one-parameter compact subgroup, and also at least a one-parameter
noncompact subgroup. Generators of the compact subgroup have discrete
spectra, in contrast to those of the noncompact subgroup that have contin-
uous spectra.
Time-dependent Schrödinger equations invariant under any of the
locally isomorphic three-parameter groups, SO(2, 1), SU(1, 1), Sp(2, r),
exemplify the general principles involved. When their three generators Rk
are functions of real variables and derivatives, they obey the commutation
relations
[R1 , R2 ] = −R3 , [R2 , R3 ] = R1 , [R3 , R1 ] = R2 . (10.6.1a)
The corresponding quantum mechanical generators Qk = −i Rk satisfy
[Q1 , Q2 ] = −iQ3 , [Q2 , Q3 ] = iQ1 , [Q3 , Q1 ] = iQ2 . (10.6.1b)
The Casimir invariant can be chosen to be either
Q23 − Q21 − Q22 , (10.6.2)
October 28, 2010 16:55 9in x 6in b943-ch10

320 Dynamical Symmetry

or its negative. Q3 is the generator of the compact one-parameter subgroup

of the groups; Q1 and Q2 are the generators of their noncompact subgroups.
The operators Q± = Q1 ±iQ2 obey the following commutation relations:
[Q3 , Q± ] = ±Q± , [Q+ , Q−] = 2Q3. (10.6.3)
The Q± thus shift the eigenvalues of Q3 by one unit, so Q3 has a discrete
spectrum. An operator that is a linear combination of Q3 and a generator
of the noncompact subgroup, e.g., an operator of the form
Q = c 1 Q 1 + c3 Q 3 , (10.6.4)
can be converted to a constant times Q3 if |c3 | > |c1 |, and to a constant
times Q1 if |c1 | > | c3 |. This can be accomplished by the action of a finite
transformation of the group. In this case one makes use of the relations
exp(iαQ2 )Q1 exp(−iαQ2 ) = Q1 cosh(α) + Q3 sinh(α), (10.6.5a)
exp(iαQ2 )Q3 exp(−iαQ2 ) = Q3 cosh(α) + Q1 sinh(α), (10.6.5b)
which have the consequence
exp(iαQ2 )Q exp(−iαQ2 )
= Q1 (c1 cosh(α) + c3 sinh(α)) + Q3 (c3 cosh(α) + c1 sinh(α)). (10.6.5c)
If, for example, c3 > c1 , one may require that α is such that
c1 cosh(α) + c3 sinh(α) = 0, (10.6.5d)
whence
tanh(α) = −c1 /c3 , α = (1/2)ln{(c3 − c1 )/(c3 + c1 )}. (10.6.5e,f)

10.7 Dynamical Group of the Bound States

of Morse Oscillators
Morse’s potential,
V(r) = d(exp(−2a(r − ro )) − 2 exp(−a(r − ro ))), (10.7.1a)
gives rise to Schrödinger equations that can be solved exactly for systems
with no angular momentum. Illustrated in Fig. 10.7.1, it provides rea-
sonable approximations to the potential governing vibrational motions in
diatomic molecules. Assuming eigenstates ψn of energy En and the time-
dependent Schrödinger equation
(H − i∂/∂t)Ψ = 0, (10.7.1b)
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 321

0.5

0
1 2 3 4 5

−0.5

−1

Fig. 10.7.1 Morse potential.

Morse showed that for the bound states of H,

En = −(a2 /2m)(n + 1/2 − k)2 , k = (2md)1/2 /a. (10.7.1c)

The operator

D = exp(ln((nop + 1/2 − k)/H)t∂/∂t), (10.7.2)

linearizes the spectrum of − i∂/∂t in (10.7.1b) and converts exp(−iEn t)

to exp(−int). Kumei8 set up and solved the determining equations for the
generators Q
of the dynamical group that leaves invariant the resulting
analog of (10.7.1), viz.

(−1/2m)(∂ 2 /∂r2 + r−1 ∂/∂r) + V(r)
 (10.7.3)
−(a2 /2m)(∂/∂t)2 Σcn exp(−int)ψn = 0.

In solving his Schrödinger equation, Morse10 introduced the change of vari-

able

r → z = 2d exp(−a(r − ro )). (10.7.4)

October 28, 2010 16:55 9in x 6in b943-ch10

322 Dynamical Symmetry

When Kumei’s operators are expressed in terms of z,t one obtains opera-
tors Q = DND−1 , in which the N are the n-labeling and n-shifting operators

N0 = i∂/∂t, (10.7.5a)
N+ = exp(−it)(i∂/∂z − N0 /z)(2N0 + 1) + d, (10.7.5b)
N− = exp(it)(i∂/∂z + N0 /z)(2N0 − 1) − d. (10.7.5c)

The operators

N1 = (N+ + N− ), N2 = −i(N+ − N− ), N3 = ∂/∂t, (10.7.5d–f)

satisfy the commutation relations:

[N1 , N2 ] = −N3 , [N2 , N3 ] = N1 , [N3 , N1 ] = N2 . (10.7.6)

The function with n = 0 is not normalizable, so n is required to be a

semi-integer:

n = 1/2, 3/2, . . . , nmax ≤ d − 1/2. (10.7.7)

The dynamical group is SU(1,1).11

10.8 Dynamical Group of the Bound States

of Hydrogen-Like Atoms
In spherical polar coordinates the time-independent Schrödinger equation
with H = p2 /2m − Z/r has solutions which we write as ψnlm (p0n r, θ, φ).
The time-dependent equation can then be expressed as

(−1/2){∂ 2/∂r2 +r−1 ∂/∂r−L2/r2 +2Z/r−2i∂/∂t}Σcnlm exp(−iEn t)ψnlm = 0,

(10.8.1a)
where

p0n = (−2En )1/2 , En = (−1/2)Z2 /n 2 , (10.8.1b,c)

and, as in (10.4.14),

L2 = −y2 ∂ 2 /∂x2 + 2x ∂/∂x − y−2 ∂ 2 /∂φ2 . (10.8.1d)

Defining

P0op = (−2H)1/2 , (10.8.2)

the operator that linearizes the spectrum,

Dt = exp{ln 2Z (P0op )−3 t∂/∂t}, (10.8.3)

October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 323

converts the operator in (10.8.1) to

(−1/2){∂ 2/∂r2 + r−1 ∂/∂r − L2 /r2 + 2Z/r − 2Z (P0op )−3 2i∂/∂t},
(10.8.4)
and converts the exp(−iEn t) to exp(int), but does not alter ψnlm (2p0n r,
θ, φ).
The operator identity P0op = iZ/(∂/∂t), which holds on the space of
solutions of (10.8.4), can be used to convert (10.8.4) to
(−1/2){∂ 2/∂r2 + r−1 ∂/∂r − L2 /r2 + 2Z/r + Z2 (∂/∂t)−2 }. (10.8.5)
To eliminate the (∂/∂t)−2 term Kumei8 acted on (10.8.4), and the functions
exp(int)ψnlm (p0n r, θ, φ), with the operator
Dr = exp{ln(ai∂/∂t)r∂/∂r}, (10.8.6)
in which a = −1/2Z. Dr following the action of Dt converts (10.8.1) to
(∂ 2 /∂t2 /8Z2 ) times the standard equation11
Ωf(r, θ, φ, t) = 0,
with
Ω = ∂ 2 /∂r2 + r−1 ∂/∂r − L2 /r2 + 2i∂/∂t/r − 1/4 (10.8.7)
and
f(r, θ, φ, t) = Σcnlm exp(−int)ψnlm (r/2, θ, φ).
The ground state solution, ψ100 , is exp(−(1/2)r) exp(it), and in all these
ψnlm the radial functions are sturmians with the the same coefficient in the
exponent, cf., Table 11.3.1 on page 351.
If Q is to represent generators of the Lie group that leaves (10.8.7)
invariant, one must have
ΩQf(r, θ, φ, t)I = 0. (10.8.8)
The generators that act on the solutions of the original time-dependent
Schrödinger equation
exp(−iEn t)ψnlm (2p0n r, θ, φ) = D−1 −1
t Dr exp(−int)ψnlm (r/2, θ, φ) (10.8.9)

are then
Q
= D−1 −1
t Dr Q Dr Dt . (10.8.10)
Allowing Q to contain derivative operators of zero, first, and second order in
all the variables, Kumei obtained and solved a set of 22 coupled determining
October 28, 2010 16:55 9in x 6in b943-ch10

324 Dynamical Symmetry

equations. The 16 generators Q that are not products of others, are Q16 = 1,
and Q1 , . . . , Q15 , which satisfy SO(4,2) commutation relations.
Of these generators, Q, three are the standard angular momentum m
labeling operator M0 = L12 , and m shift operators M± = L23 ± iL31 .
Another three represent the components of Az = L34 , Ax ±iAy = L14 ±iL24 ,
that act on the functions ψnlm (r/2, θ, φ). These change the value of the
angular momentum quantum number, l . One of the operators, −i∂/∂t, is
the n labeling operator for the functions exp(−int)ψnlm (r/2, θ, φ). The
remaining eight operators all change the value of n, and hence, E, and are
consequently explicit functions of t. In the expressions for the generators
given below, x = cos(θ), y = sin(θ). Generators of compact subgroups are
labeled Lab (= −Lba ). Generators of non-compact subgroups are labeled
Kab (= +Kba). The generators satisfy the commutation relations
[Lab , Lbc ] = i Lac , [Lab , Kbc ] = i Kac , [Kab , Kbc ] = i Lac . (10.8.11)
Using these commutation relations, one is able to obtain the explicit form
of all the generators from the following:
L12 = −i∂/∂φ, (10.8.12)
−1
M± = exp(±iφ)(ixy ∂/∂φ − ±y∂/∂x) = (L23 ± iL31 ),

L34 = 2 −y2 ∂ 2 /∂r∂x + r−1 xy2 ∂ 2 /∂x2 + r−1 xy−2 ∂ 2 /∂φ2 + x∂/∂r

−2r−1x2 ∂/∂x − (i/2)x∂/∂t , (10.8.12a–e)
K45 = cos(t)(−ir∂/∂r − i) − sin(t)(i∂/∂t + r/2), (10.8.13)
L56 = i∂/∂t.
The generators not listed above are L14 , L24 , K15 , K25 , K35 , K16 , K26 , K36 ,
and K46 . The generator K46 may, for example, be determined from the
relation
[L56 , K45 ] = −iK46 , (10.8.12f)
which implies that
K46 = − sin(t)(−ir∂/∂r − i) − cos(t)(i∂/∂t + r/2). (10.8.12g)
When acting on the functions in (10.8.7), i∂/∂t multiplies them by n,
so in the generators (10.8.12c–e,g) it may be replaced by Nop = Z(P0op )−1 .
If one is primarily interested in the action of the generators on the position-
space wave functions ψnlm (r/2, θ, φ), one may set t = 0 in the operators that
result from this replacement of i∂/∂t. This process carries out conversions
such as
K45 → K
45 = −ir∂/∂r − i, K46 → K
46 = Nop + r/2. (10.8.13a,b)
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 325

The generators are self-adjoint if one replaces the r2 dr in the Schrödinger

scalar product by rdr. (cf. Secs. 10.8, 11.1 and 11.3).
The generators L
34 and K
45 have the following action on the functions
ψnlm (r, θ, φ):
L
34 ψnlm (r, θ, φ) = a+ −
nlm ψnl+1m (r, θ, φ) + anlm ψnl−1m (r, θ, φ) (10.8.14a)
with
a+ 2 2 2 2 1/2
nlm = {(n − (l + 1) )((l + 1) − m )} {(2l + 1)(2l − 1)}−1/2
(10.8.14b)
a− 2 2 2 2 1/2
nlm = {(n − l )(l − m )} {(2l + 3)(2l + 1)}−1/2 (10.8.14c)
and
K
45 ψnlm (r, θ, φ) = d+ −
nlm ψn+1lm (r, θ, φ) + dnlm ψn−1lm (r, θ, φ) (10.8.15a)
with
d±
nlm = ±{n(n ± 1) − (l (l + 1)}
1/2
(10.8.15b)
The action of the generators on the functions exp(−iEn t) ψnlm (p0n r,θ, φ)
may be obtained by acting with the operator D−1 t Dr
−1
on the functions



exp(−in t) ψn
l
m (r, θ, φ), where in (10.8.15), n = n ± 1.
In Chapter 11 we will more extensively investigate the isomorphic SO(4,
2) dynamical group of the regularized Schrödinger hydrogen-like atoms.

10.9 Matrix Representations of Generators

and Group Operators
Because Schrödinger’s differential equations are linear, the differential oper-
ators in them may be represented by matrices that operate on vectors
defined on the space of solutions of the equations. All operators that act
on these wave functions may then be represented by matrices.
As an example, suppose that Ψ is a linear combination of p−1 , p0 , p+1
atomic wave functions
Ψ = c−1 ψ−1 + c0 ψ0 + c+1 ψ+1 , (10.9.1)
and that one seeks the matrix representation of the operator
L0 = −i(x∂/∂y − y∂/∂x). (10.9.2)
One has
Ψ
≡ L0 Ψ = (−1)c−1 ψ−1 + (0)c0 ψ0 + (+1)c+1 ψ+1 . (10.9.3)
October 28, 2010 16:55 9in x 6in b943-ch10

326 Dynamical Symmetry

If one considers the ψm to be unit vectors, then (10.9.1) and (10.9.3) may
be represented, respectively, by the vectors
C = (c−1 , c0 , c+1 ), C
= ( − c−1 , 0, c+1 ). (10.9.4)
L0 is then represented by a 3 × 3 matrix, L0 , whose off-diagonal elements
are zero and diagonal elements are −1, 0, +1. The functions ψm are said
to provide the basis for the matrix representation of the operator.
The analogous matrix representations of L+ = Lx+iLy, and L− = Lx−i
Ly, are defined by their action on ψm , for which one finds
L+ ψm = ψm+1 ,
L− ψm = ψm−1 . (10.9.5)
The reader may verify that the resulting matrix representations, L+ and
L− , satisfy the commutation relations
L0 L+ − L+ L0 = L+ ,
L0 L− − L− L0 = −L, (10.9.6)
L+ L− − L− L+ = 2L0 ,
which are isomorphic to the commutation relations obeyed by the corre-
sponding differential operators. As there is a one-to-one correspondence
between the matrices and the differential operators and a one to one cor-
respondence between their Lie algebras, it is said that the basis provides a
faithful representation of the Lie algebra. When such one-to-one correspon-
dence does not exist, the commutation relations of the matrices and that
of the differential operators are merely homomorphic.
The reader will find that all the matrices M in (10.9.6) are unitary,
that is
M−1 = M∗T . (10.9.7)
It can also be proven that these matrices cannot be reduced to the form
 
A B

(10.9.8)
B D
in which the elements of the submatrices B and/or B
are all zero. Matrices
that cannot be put into this form are said to be irreducible. (When both
B and B
are zero matrices, the matrix (10.9.8) is the product of the two
matrices A and D). All these conditions being fulfilled, the basis ψ−1 , ψ0 ,
ψ+1 , of the vector in (10.9.1) is said to be that of a unitary irreducible
representation, abbreviated UIR.
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 327

The eigenfunctions of L2 with quantum number l provide the bases for

the (2l + 1)-dimensional UIR of SO(3), and of a B1 Lie algebra. They also
provide the basis for infinite dimensional UIR of the SO(3,2) groups of
Secs. 10.4 and 10.5.
A straightforward introductory treatment of the properties of group rep-
resentations will be found in Chapter 2 of Wybourne’s monograph, Classical
Groups for Physicists.13 An extensive treatment is provided in Barut and
Raczka’s Theory of Group Representations and Applications.14 Work of
Gelfand and Zeitlin provides a very useful scheme for describing and deter-
mining UIR of the orthogonal and unitary groups.15
We will summarize only a few basic facts here.
1. Compact groups and noncompact groups are not distinguishable if one
allows the generators to have complex or imaginary eigenvalues, and/or
the group parameters to be complex or imaginary. When the distinction
is possible:
1a. The UIR of compact groups are finite dimensional.
1b. The UIR of noncompact groups are either one-dimensional or infinite
dimensional. The latter may in some cases be labeled by quantum num-
bers that take on discrete values, and in other cases they must be
labeled by quantum numbers that take on continuously variable val-
ues.
2. If X is a matrix representation of a group generator Q that is a real
function of real variables and real derivative operators, (such as that
of Rz = x∂/∂y − y∂/∂x on the basis provided by x, y, z) and α is a
real parameter, the expansion of exp(αX) in powers of α can only be
convergent only if the absolute values of the matrix elements in X have
an upper bound. (In the example given in (10.9.2), one can replace Rz
by Lz , and α by iα.)

10.10 Invariant Scalar Products

In using the functions in (10.9.1) to define the vectors (10.9.4) of the pre-
vious section it was implicitly assumed that the functions ψm could be
treated as unit vectors. This assumption presupposes that a scalar product
is available so that one can define the meaning of unit vector. As the time-
independent parts of the ψm in the example are the spherical harmonics,
Ylm (θ,φ), the scalar product that defines their normalization, is

2π
π
f, g = dφ sin(θ)dθ f ∗ g. (10.10.1)
0 0
October 28, 2010 16:55 9in x 6in b943-ch10

328 Dynamical Symmetry

In this integral, the value of sin(θ)dθdφ is invariant under rotations. Because

of this, the definition of the unit vectors in (10.9.1) is unambiguous. How-
ever, in other cases this is not necessarily so. A similar problem arises when
one attempts to define the adjoint of an operator: the adjoint cannot be
defined if the scalar product is not defined.
In short, when using a basis to associate matrices with differential oper-
ators and their adjoints it becomes necessary to determine what is an appro-
priate scalar product. Thus, for example in the case of the SO(4) group of
the hydrogen atom, it is necessary to determine whether one’s definition
of Runge–Lenz operators A generates transformations that leave invariant
the usual scalar product

∞
2π
π
2
f, g = r dr dφ sin(θ)dθ f ∗ g. (10.10.2)
0 0 0
In Fock’s realization of the SO(4) group the volume element involved is
the area element on a unit hypersphere in four-space. Fock’s stereographic
projection transforms it to a modified volume element in momentum three-
space, and Fourier transformation converts it to a volume element in posi-
tion three-space which is r−1 times the volume element in (10.10.2)16 .
Consequently, self-adjoint rotation operators on the Fock hypersphere, pro-
duce Runge–Lenz operators in momentum space which transform into
position-space Runge–Lenz operators that are self-adjoint only if a r−1
factor is inserted into the Schrödinger scalar product. Another example
is provided by the Morse oscillator. A number of ways of dealing with the
invariant scalar product of its SO(2, 1) group will be found in Ref. 17.
In the general case one may use the extended generators of a group to
determine the volume element, dv’, that is left invariant by the group.

10.11 Direct-Products: SO(3)⊗SO(3) and the Coupling

of Angular Momenta
The product of spherical harmonics Yl1 m1 (θ1 , φ1 )Yl2 m2 (θ2 , φ2 ) is an eigen-
state of Lz1 +Lz2 with eigenvalue M = m1 +m2 . From a set of such products,
with the same values of M , one may build an eigenstate of L2 = (L1 + L2 )2
with eigenvalue L(L + 1). Because L2 commutes with L21 and L22 , it can also
be an eigenstate of L21 with eigenvalue l1 (l1 + 1), and of L22 with eigenvalue,
l2 (l2 + 1). If ψLM (l1 l2 ) is such an eigenstate, one has
ψLM (l1 l2 ) = ΣC(l1 l2 LM ; m1 m2 )Yl1 m1 (θ1 , φ1 )Yl2 m2 (θ2 , φ2 ), (10.11.1)
where the sum is over the values of m1 and m2 whose sum is M. The
coefficients C(l1 l2 LM;m1 m2 ) are known as Clebsch–Gordon, or Wigner,
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 329

coefficients. For normalized spherical harmonics Ylm satisfying a given

phase convention they may be completely determined using the commu-
tation relations satisfied by the components of the vectors L1 and L2 . A
discussion of the process will be found in Edmond’s monograph.9
If L1 and L2 represent the angular momentum vectors of particles 1
and 2, respectively, and their sum, L, represents the total angular momen-
tum vector of a two-particle system, then ψLM (l1 l2 ) represents a state of
the system with definite total angular momentum |L2 |1/2 = |L(L + 1)|1/2 .
Because
L2 = L21 + L22 + 2L1 · L2 , (10.11.2)
ψLM (l1 l2 ) is also an eigenfunction of L1 ·L2 :
(L1 · L2 )ψLM (l1 l2 ) = B2 ψLM (l1 l2 ) ,
(10.11.3)
B2 = {L(L + 1) − l1 (l1 + 1) − l2 (l2 + 1)}/2.
Physically speaking, one has coupled motions allowed to L1 and L2 by
establishing a fixed angle, β, between them, such that
cos(β) = |L1 · L2 |/|L1 ||L2 | = B2 /{l1 (l1 + 1)l2 (l2 + 1)}1/2 . (10.11.4)
In Dirac notation, Eq. (10.11.1) may be expressed as
|LM (l1 l2 ) = ΣC(l1 l2 LM ; m1 m2 )|l1 m1 |l2 m2 . (10.11.5)
2
In this form it holds true for all eigenstates |lm, of L , Lz that satisfy
Condon and Shortley’s phase conventions. As (10.11.1) is derived using only
SO(3) commutation relations, and these are the same as those of SU(2), it
also applies to the coupling of spin angular momenta. Thus (10.11.5) may
be replaced by the more general equation
|JM (j1 j2 ) = ΣC(j1 j2 JM ; m1 m2 )|j1 m1 |j2 m2 . (10.11.6)
The coupling of the direct-product of representations generalizes to all
Lie groups.

10.12 Degeneracy Groups of Non-interacting Systems;

Completions of Direct-Products
We begin this section with an example. Suppose that the components of
La = (L12 , L23 , L31 ) and Lb = (L45 , L56 , L64 ) (10.12.1)
are the generators of the two SO(3) subgroups of the direct-product group
SO(3)a ⊗ SO(3)b , that K is a constant and that the Hamiltonian of a
system is
H = K(L2a + L2b ). (10.12.2)
October 28, 2010 16:55 9in x 6in b943-ch10

330 Dynamical Symmetry

Then SO(3)a ⊗ SO(3)b is a degeneracy group of the system with Schrödinger

equation
Hψ = Eψ, E = Ea + Eb = K(la (la + 1) + lb (lb + 1)). (10.12.3a)
If
Hψ = i∂ψ/∂t,
then
ψ = Σma ,mb c(la lb ma mm )Yla ma (θa , φa )Ylb mb (θb , φb ) exp(−i(Ea + Eb )t).
(10.12.3b)
Particular sets of solutions are obtained when
c(la lb ma mb ) = C(la lb LM ; ma mb ), (10.12.3c)
and the sum is over ma , mb such that ma + mb = M . Then ψ = ψLM (la lb ) .
Now SO(3)a ⊗ SO(3)b is contained in an SO(6) group, whose Lie algebra
contains the components of La and Lb , and nine further rotation generators
L14 , L15 , L16 , L24 , L25 , L26 , L34 , L35 , L36 . (10.12.4)
If x1 ,x2 ,x3 are Cartesian coordinates of particle a and x4 ,x5 ,x6 are those
of particle b, then the operators in (10.12.4) are those of generators of
rotations acting on a basis which is provided by a product of single-particle
states, regardless of whether it is coupled or uncoupled. As an example one
might have a six-dimensional basis provided by products of s and p orbitals
of two electrons. The rotation operators with generators (10.12.4) would
then mix (hybridize) these orbitals in the manner of valence-bond theory.
No matter what the choice of two-particle basis is, SO(6) provides operators
that can mimic the effects of a much wider variety of interactions than the
operators provided by SO(3) ⊗ SO(3).
One might term SO(6) the completion of the direct-product group
SO(3) ⊗ SO(3). Though its Casimir operators commute with the gener-
ators of SO(3) ⊗ SO(3), it is not a degeneracy group of the original sys-
tem. It can become a dynamical group of the modified system if all the
generators Ljk are considered to be time-independent versions of invariance
generators, Q, of an appropriate time-dependent Schrödinger equation. The
discussion of Eqs. (10.8.12) and (10.8.13) provides an example of this rela-
tionship between time-independent generators and invariance generators of
a time-dependent Schrödinger equations.
Completions of direct-product groups have general properties that arise
from relationships between the Casimir operators of the groups. The case
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 331

at hand illustrates some of these. If one defines L2ab to be the sum of the
squares of the generators in (10.12.4), then the quadratic Casimir operator
of the group is

C2 = L2a + L2b + L2ab . (10.12.5)

As the components of La and Lb commute with the Casimir operator, C2 ,

and with L2a and L2b , they necessarily commute with L2ab . The components
of L2ab of course commute with C2 , however, they do not commute with the
components of La , Lb , nor with those of La + Lb . Properties such as these
have several consequences:

a. Adding a constant times L2ab to the Hamiltonian (10.12.2) leaves L2a and
L2b , and hence (La + Lb )2 as constants of motion, so it alters the energy
of the unpertubed states by an amount determined by the eigenvalues
of C2 .
b. As SO(6) is of rank 3, it also has two further Casimir operators. Adding
these to the Hamiltonian (10.17.2) also leaves la , lb and L as labeling
eigenvalues of constants of motion.

In both cases, adding the additional Casimir operators to the original

Hamiltonian does not alter its eigenstates, it simply alters their energies.
In the general case, the completion of a direct-product group has two
useful properties:

i) The Casimir operators of the group obtained by completion of a direct-

product group necessarily commute with the generators of the direct-
product group and so provide labeling operators whose physical rele-
vance can be directly assessed.
ii) The generators of the enlarged group will have known matrix elements
which can be used as a guide in developing approximate treatments of
actual physical systems. In the next section, and in Chapter 12, we will
discuss several other direct-product groups and their completions.

10.13 Dynamical Groups of Time-dependent Schrödinger

Equations of Compound Systems; Many-Electron
Atoms
Let Ha = H(ra ) be the Hamiltonian of system ā, and let Hb = H(rb ) be that
of system b̄, and suppose that both Hamiltonians have the same functional
October 28, 2010 16:55 9in x 6in b943-ch10

332 Dynamical Symmetry

form. Let k, k’ represent sets of quantum numbers or labels, and let

φa,k = φk (ra ), φb,k
= φk
(rb ), ψa = ψ(ra , t), ψb
= ψ
(rb , t).
(10.13.1a–d)
Require
Ha φa,k = Ea,k φa,k , Hb φb,k
= Eb,k
φb,k
, (10.13.1e,f)
and
Ha ψa, = i∂ψa, /∂t, Hb ψa, = i∂ψb /∂t. (10.13.1g,h)
The time-dependent functions ψ can be expressed as sums (or integrals)
over the functions c(k)exp(−iEk t)φk . We shall suppose that one has avail-
able a spectrum-generating algebra of operators
Qa = Q(ra , ∂/∂ra , ∂/∂t), Qb = Q(rb , ∂/∂rb , ∂/∂t), (10.13.2)
for Eqs. (10.13.1e,f), and that these Q satisfy the determining equations
arising from the time-dependent Schrödinger equations (10.13.1g,h).
Next consider the compound system with
H = Ha + Hb , (10.13.3a)
and let


Φa,b = Φk,k
= φa,k φb,k
, and Ψa,b = Ψk,k
= ψa,k ψb,k
. (10.13.3b,c)
Then
H Φk,k
= (Ea,k + Eb,k
)Φk,k
, (10.13.3d)
and
H Ψa,b = ψb
Ha ψa + ψa Hb ψb

(10.13.3e)
= i(ψb
∂ψa /∂t + ψa ψb /∂t).
This last equation becomes separable if its right-hand side is replaced by
i(ψb
∂ψa /∂ta + ψa ∂ψb /∂tb ). (10.13.3f)
To make this replacement one may set ta + tb = t, which implies that the
physical time-evolution operator is
∂/∂t = ∂/∂ta + ∂/∂tb . (10.13.4)
The dynamical group then becomes a constrained direct-product of the
dynamical groups Ga and Gb . In it ta and tb , as well as ∂/∂ta and ∂/∂tb ,
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 333

can be considered to be mathematically independent until one imposes the

physical requirement in (10.13.4).
Completion of such direct-products yields groups which can be
the dynamical groups of systems whose Hamiltonian is obtained from
H(ra ) + H(rb ) by adding interaction terms, or by introducing parameters
that produce new potential and/or kinetic energy terms in it. For the group
to be a dynamical group of the altered system, the changes in the Hamil-
tonian must be expressible by functions of the generators of the group,
and the solutions Φ and Ψ, of the resulting Schrödinger
 equations, must be
expressible as convergent linear combinations, (Σ + ), of the solutions of
(10.13.3):


Φ= Σ+ (c(k, k
)Φk,k
), (10.13.5a)


Ψ= Σ+ (c(k, k
)Ψk,k
). (10.13.5b)

If these conditions are satisfied the completed group will be a dynamical

group of the altered system.
In the case of two-electron atoms, one can begin with Ha = p2a /2 − Z/ra,
Hb = p2b /2 − Z/rb , with eigenstates ψa , ψb . The basis states of the direct-
product SO(4, 2)a ⊗ SO(4, 2)b and its completion, SO(8,4), can then be
chosen to be ψa ψb . One may add the electron repulsion potential 1/rab
to Ha + Hb , and use operations available within SO(4, 2)a ⊗ SO(4, 2)b or
SO(8,4) to deal with the effects of the added potential, as well as those of
other interactions of interest. For atoms with N electrons, the analogous
group is SO(4N, 2N).
Kato investigated the question of the convergence of the series (10.13.5a)
for two-electron atoms of nuclear charge Z when the added term is the
electron repulsion potential.18 He established that the series converges for Z
> 4.1, but did not establish that this value is the lowest possible for Z. Even
in cases where convergence of many-particle analogs of the series (10.13.5)
is problematic, the groups can prove useful if they lead to discoveries of new
approximate (or exact) degeneracies and an understanding of their physical
consequences.
Chapter 12 contains a discussion of SO(4)⊗SO(4) degeneracies and sym-
metry breaking in several-electron atoms. It also contains brief discussions
of the relevance of its SO(8,4) completion to studies of systems containing
more than one electron.
October 28, 2010 16:55 9in x 6in b943-ch10

334 Dynamical Symmetry

Appendix A. References Providing Invariance Groups

of Schrödinger Equations
L. Infeld and T. E. Hull, The Factorization Method, Rev. Mod. Phys., 23
(1951) 21.
B. L. Dunlap and L. Armstrong Jr., SO(2,1) and the Hulthe’n Potential,
Phys. Rev. A 6 (1972) 1370.
A. O. Barut and G. L. Bornzin, Algebraic solution of the Schrödinger equa-
tion for a class of velocity-dependent potentials, Lett. Nuovo Cimento 6
(1973) 177.
G. Bluman, S. Kumei and G. J. Reid, New classes of symmetries for partial
differential equations, J. Math. Phys. 29 (1988) 806.
N. Ibragimov, CRC Handbook of Lie Analysis of Differential Equations,
I, II, III, (CRC Press 1994–1996).
F. Cooper, A. Khare and U. Sukhatme, Supersymmetry in Quantum
Mechanics (World Scientific, Sinagapore, 2001).

Appendix B. References to Work Dealing with Dynamical

Symmetries in Nuclear Shell Theory
M. Moshinsky, Group Theory and the Many-body Problem (Gordon and
Breach, N.Y., 1968).
K. Helmers, Symplectic invariants and Flowers’ classification of shell model
states, Nucl. Phys. 23 (1961) 594.
G. Racah, Group theory and spectroscopy, Erg. D. Exakt. Naturwiss 37
(1965) 30.
D. J. Rowe, Dynamical symmetries of nuclear collective models, Prog. Part.
Nucl. Phys. 37 (1996).

Exercises
1a. Show that the quantum mechanical version, -iD, of the dilation operator
D defined in (10.5.18), is not self-adjoint under the Schrödinger scalar
product, but becomes self-adjoint if one replaces r2 dr by rdr.
1b. Construct symmetrized quantum-mechanical versions of all the other
generators in (10.5.18) that are self-adjoint under this scalar product.
2a. Determine the commutation relations of the generators defined in
(10.5.18).
2b. Prove that they are isomorphic to commutation relations of the genera-
tors of SO(4,1), and establish the relation of the generators in (10.5.18)
October 28, 2010 16:55 9in x 6in b943-ch10

Spectrum-Generating Lie Algebras and Groups 335

to the generators Lab = −Lba of the SO(4) subgroup of SO(4, 1), and
those, Kab = Kba , of its noncompact subgroup.
3a. Completion of the direct-product group SO(2)⊗SO(2) produces SO(4).
Consider the circles defined by the one-parameter groups generated by
L12 and L34 , and suppose that these are contained in orthogonal planes,
so that (x1 , x2 )·(x3 , x4 ) = 0. How are these circles affected by rotations
generated by L23 , L14 ?
3b. Determine the small rotations in the SO(4) group that leave invariant
the scalar product (x1 , x2 ) · (x3 , x4 ).

References
[1] L. Infeld and T. E. Hull, Rev. Mod. Phys. 23 (1951) 21.
[2] Chap. IX Ref. 5.
[3] F. Cooper, A. Khare and U. Sukhatme, Supersymmetry in Quantum
Mechanics (World Scientific Publishers, Sinagapore, 2001).
[4] R. L. Anderson, S. Kumei and C. E. Wulfman, Rev. Mex. de Fisica 21 (1972)
1–33.
[5] G. A. Baker Jr., Phys. Rev. 103 (1956) 1119.
[6] R. H. Fowler, Statistical Mechanics (Cambridge University Press, N. Y.,
1955), Sec. 2.21.
[7] A. O. Barut and A. Bohm, Phys. Rev. B 139 (1965) 1107.
[8] S. Kumei, Group Theoretic Properties of Schrödinger Equations– Systematic
Derivation, M.Sc. Thesis, Department of Physics, University of the Pacific
(Stockton, CA, 1972).
[9] A. R. Edmonds, Angular Momentum in Quantum Mechanics (Princeton
University Press, Princeton, 1957).
[10] P. M. Morse, Phys. Rev. 34 (1929) 57.
[11] P. Cordero and S. Hojman, Lett. Nuovo Cimento 4 (1970) 1123.
[12] A. Messiah, Quantum Mechanics (John Wiley, N.Y., 1962), Vol. I, Chap. XI.
[13] B. G. Wybourne, Classical Groups for Physicists (Wiley-Interscience, N.Y.,
1974).
[14] A. O. Barut and R. Raczka, Theory of Group Representations and Applica-
tions, 2nd edn., (World Scientific, Singapore, 1986).
[15] I. M. Gelfand and M. L. Tsetlein, Dokl. Akad. Nauk. SSSR 71 (1950) 25;
ibid. 1017.
[16] T. Shibuya, Fock’s Representation of Molecular Orbitals, M.Sc. Thesis,
Department of Physics, University of the Pacific (Stockton, CA, 1965);
cf., T. Shibuya and C. E. Wulfman, Proc. Roy. Soc. A 286 (1965) 376.
[17] C. E. Wulfman, Phys. Rev. A 54 (1996) R987.
[18] T. Kato, Perturbation Theory for Linear Operators (Springer, N. Y. 1966),
pp. 410–413.
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-ch11

CHAPTER 11

Dynamical Symmetry of Regularized

Hydrogen-like Atoms

Introduction
This chapter relates the dynamical symmetries of one-electron atoms to
the dynamical symmetries of regularized classical Keplerian systems. It
first describes the quantum mechanical position-space formulation of the
regularized system and the generators of its SO(4,2) group, then sets forth
the corresponding momentum-space formulation and Fock’s projection of
its momentum space onto the hypersphere S(3). The chapter concludes
with a description of the SU(2) ⊗ SU(2) structure of the SO(4) degeneracy
group. To simplify the presentation in this and the subsequent chapter, we
will use units in which the charge of the electron, e, has unit value, and as
in the two previous chapters, will use units in which both the reduced mass
of the electron and
= h/(2π) also have unit value.

11.1 Position-space Realization of the Dynamical Symmetries

The self-adjoint Schrödinger analogs of classical PB operators are not, in
general, those appropriate for the regularized system in position-space. The
reason is that the regularized equation which defines classical Keplerian
motion when the Hamiltonian is p2 /2 − 1/r has, as its simplest quantum
mechanical analog, the equation

(Wop − 1)ψ = 0, (11.1.2a)

in which Wop is obtained from Schrödinger’s time-independent equation by

multiplying it from the left with r:

Wop = r(p2op − 2E)/2, p2op = −∇2 . (11.1.2b)

337
October 28, 2010 16:55 9in x 6in b943-ch11

338 Dynamical Symmetry

When one inserts r(Hop − E) into the Schrödinger scalar product

∗  ∗ 
ψ ψ dv = ψ ψ dxdydz = ψ ∗ ψ  r2 drsin(θ)dθdφ, (11.1.3a)

one obtains


ψ ∗ r(Hop − E)ψ  dv. (11.1.3b)

For Wop to be self-adjoint it is necessary to alter the definition of the scalar

product. If X is an operator, let


ψ, Xψ   ≡ ψ ∗ (1/r)Xψ  dv. (11.1.4a)

This implies



ψ, r(Hop − E)ψ  = ψ ∗ (Hop − E)ψ  dv, (11.1.4b)

and


ψ, rψ   = ψ ∗ ψ  dv. (11.1.4c)

Though r and −r∇2 are self-adjoint under the scalar product (11.1.4a),
the operator −i∂/∂x is not. However, operators, such as ordinary angular
momentum generators, which commute with r and are self-adjoint under
the Schrödinger scalar product, remain self-adjoint under (11.1.4a).
The dynamical symmetry of the classical Keplerian systems investigated
in Chapter 8 can be enlarged to an SO(4,2) symmetry.1−3 The simplest way
to do this is to add to the SO(4,1) generators of Sec. 8.6, the five operators:

Kj6 = {rpj , ·} ≡ K6j , j = 1, 2, 3; K46 = {(r/2)(p2 − 1), ·} ≡ K64 ,

J56 = {−(r/2)(p2 + 1), ·} ≡ −J65 . (11.1.1)

The Lie-algebraic extension of the correspondence principle ensures that

this SO(4,2) dynamical symmetry is also a dynamical symmetry of
hydrogen-like atoms.
We will use Bednar’s form of the appropriate self-adjoint quan-
tum mechanical position-space SO(4,2) generators.4 These are the fifteen
operators:
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 339

Jjk = −i(qj ∂/∂qk − qk ∂/∂qj) = −Jkj , j < k = 1, 2, 3;

Jk4 = −(xk ∇2 − 2∂/∂xk − 2r · ∇∂/∂xk + xk )/2 ≡ −J4k , k = 1, 2, 3;
2
J56 = −(r/2)(∇ − 1) ≡ −J65 ; K45 = −i(1 + r · ∇) ≡ K54 ;
2
Kk5 = −(xk ∇ − 2∂/∂xk − 2r · ∇∂/∂xk − xk )/2 ≡ K5k , k = 1, 2, 3;
2
K46 = −(r/2)(∇ + 1) ≡ K64 ; Kk6 = −ir∂/∂xk ≡ K6k , k = 1, 2, 3.
(11.1.5)

The relations

[Jab , Jbc ] = iJac , [Jab , Kbc ] = iKac , [Kab , Kbc] = iJac , (11.1.6)

determine all the nonzero commutators of the generators. It is convenient

to collect twelve of the classical generators into the vectors

L = (J12 , J23 , J31 ) ∼ {(r × p), ·}, A− = (J14 , J24, J34 ) ∼ {A(−1/2), ·},
A+ = (K15 , K25 , K35 ) ∼ {A(1/2), ·}, Γ = (K16 , K26 , K36 ) ∼ {(rp), ·}.
(11.1.7a)

The components of each vector are transformed amongst themselves by

rotations generated by the components of L. We will use L, A− , A+ , and
Γ to denote their quantum mechanical counterparts. The three further
generators

K45 ∼ {(r · p), ·}, K46 ∼ {W(1/2), ·}, J56 ∼ {W(−1/2), ·}, (11.1.7b)

and their quantum-mechanical counterparts, K45 , K46 , J56 , transform as

scalars under rotations.
SO(4,2) possesses the semisimple subgroups listed in Table 11.1,
together with their generators, and nonvanishing Casimir operators.

Table 11.1.

Subgroup Generators Casimir Operators, Eigenvalues

SO(3) L L2 → l (l + 1)
SO(2,1) K45 , K46 , J56 (J56 )2 − (K45 )2 − (K46 )2 → l (l + 1)
SO(4) L, A− L2 + A2− → n 2 − 1; (A− · L)2 ≡ 0
SO(3,1) L, A+ L2 − A2+ → ν 2 − 1; (A+ · L)2 ≡ 0
SO(4,1) L, A− , A+ , K45 A2+ + (K45 )2 − A2− − L2 ≡ 2
SO(4,1) L, A− , Γ, K46 Γ2 + (K46 )2 − A2− − L2 ≡ 2
October 28, 2010 16:55 9in x 6in b943-ch11

340 Dynamical Symmetry

As SO(4,2) has rank three, it has three functionally independent Casimir

operators. The quadratic Casimir operator
[2]
CSO(4,2) = A2− + L2 + (J56 )2 − (A2+ + Γ2 + (K45 )2 + (K46 )2 ). (11.1.8)

has unit value. Both higher order SO(4,2) Casimir operators vanish. The
SO(4,2) basis that is defined by the values of these Casimir operators is
countably-infinite discrete. Because A is perpendicular to L, the quar-
tic Casimir operator of SO(4) vanishes identically. As a result the SO(4)
representations contained in the SO(4,2) have dimension n 2 . The n 2 -fold
degeneracy of hydrogen atom energy levels is a physical expression of this
group-theoretic property. The basis states of the SO(4) representations,
with a given value of n, are most commonly chosen to be those of n different
SO(3) representations, labeled by l, which takes on the values 0, 1, . . . , n −1,
and by the integer m which takes on the 2l + 1 values for which |m| ≤ l.
The resulting basis for the representation of SO(4,2) is provided by the
functions which satisfy the equations
[2]
CSO(4,2) χnlm = χnlm , J56 χnlm = nχnlm ,
J12 χnlm = mχnlm , L2 χnlm = l (l + 1)χnlm , (11.1.9)

An important alternative basis will be discussed in Sec. 11.4 below.

Utilizing the particular form of the generators given in (11.1.5), these
abstract group theoretic considerations establish analytic relationships. One
finds that the χnlm can be realized as functions, which when normalized
under the scalar product (11.1.3a), are

χnlm (r, θ, φ) = Rnl (r)Ylm (θ, φ), (11.1.10a)

with

Rnl (r) = Nnl (2r)l exp(−r)F(l + 1 − n; 2l + 2; 2r). (11.1.10b)

Here F is the confluent hypergeometric function; the normalizing factor

Nnl = [(−i)l (23/2 )/(2l + 1)!][(l + n)!/n(n − l − 1)!]1/2 , (11.1.10c)

has been assigned the phase factor, (−i)l .5 The functions χnlm , now com-
monly called Sturmians, differ from those introduced in Sec. 10.8, and are
not the same as the usual normalized bound-state electronic wave functions,
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 341

Ψnlm (r, θ, φ), which satisfy the Schrödinger equation

(−∇2 /2 − 1/r)Ψnlm = En Ψnlm . (11.1.11a)
If the phase choice is the same, the two types of wave functions are related
by
Ψnlm (r, θ, φ) = n 1/2 χnlm (r/n, θ, φ). (11.1.11b)
The basis provided by the χnlm is discrete, and complete. That is, any
normalizable piecewise continuous function f(r, θ, φ) can be expressed as a
convergent series of these harmonics. The action of the group generators on
this basis is given in Appendix A. The discreteness and completeness of the
Sturmian basis stands in marked contrast to the usual complete hydrogenic
bases, which contain a continuum of scattering states as well as a count-
able infinity of bound states. Because of this, working with the regularized
system has the advantage that it is much simpler to expand a given bound-
state function as a linear combination of Sturmian eigenfunctions than as
a linear combination of hydrogenic eigenfunctions. Chapter 12 deals with
the exploitation of these observations in studies of systems more complex
than hydrogen-like atoms.
As in the classical system, SO(4) transformations interconvert bound-
state solutions — wave functions here — of the same energy. When the
energy is positive, SO(3,1) transformations interconvert degenerate solu-
tions. The additional transformations available in the SO(4,1) and SO(4,2)
groups superpose eigenfunctions of differing energies, but cannot superpose
functions whose energies differ in sign. In the quantum mechanical systems
the states of the discrete spectrum respond to perturbations differently than
those of the continuous spectrum, and those of classical trajectories — only
a few states of the discrete spectrum, the n 2 degenerate ones, mix appre-
ciably when subjected to small perturbations. Consequences of this have
been pointed out in Chapter 1, and they organize much of the discussion
in Chapter 12.
The equation governing regularized hydrogen-like systems with nuclear
charge of magnitude Z is
(Wop − Z)ψ = 0, (11.1.12a)
Wop = r(p2op − 2E)/2, (11.1.12b)
2
p2op = −∇ . (11.1.12c)
It can be obtained by the action of the unitary dilatation operator
exp(iαJ45 ) on Eq. (11.1.2). We will suppose that polar coordinates are being
October 28, 2010 16:55 9in x 6in b943-ch11

342 Dynamical Symmetry

used, and set ψ = Ψ(r, θ, φ). One has

exp(iαJ45 )(r(p2op − 2E)/2 − 1)Ψ(r, θ, φ) = 0, (11.1.13a)
so
(exp(iαJ45 )(r(p2op − 2E)/2 − 1) exp(−iαJ45 )) exp(iαJ45 )Ψ(r, θ, φ) = 0.
(11.1.13b)
Carrying out the dilations of r and p converts this equation to
((e−α rp2op − 2Eeα r)/2 − 1)Ψ (r, θ, φ) = 0, (11.1.13c)
with
Ψ (r, θ, φ) = eα Ψ(eα r, θ, φ). (11.1.13d)
Thus
e−α (r(p2op − 2Ee2α )/2 − eα )Ψ (r, θ, φ) = 0. (11.1.13e)
α α
If one multiplies this last equation by e and then sets e = Z, it can be
rewritten as
(r(p2op − 2E )/2 − Z)Ψ (r, θ, φ) = 0, (11.1.14a)
with
E = Z2 E, Ψ (r, θ, φ) = ZΨ(Zr, θ, φ). (11.1.14b,c)
We shall henceforth drop the prime on E, assume n, l , m are the appropriate
quantum numbers, and express (11.1.14a) as
(r(p2op − 2E)/2 − Z)|znlm = 0, (11.1.15a)
or as
(Wop − Z)|znlm = 0. (11.1.15b)
Having established that the operator of a one-parameter subgroup of
SO(4,2) can be used to directly change the nuclear charge, we will for the
time being concentrate attention on the Z = 1 hydrogen atom system, for
which
χnlm → |1nlm ≡ |nlm. (11.1.15c)
Using the definitions of J56 and K46 , one finds that
r = J56 − K46 , rp2op = J56 + K46 . (11.1.16)
Thus if E = −1/2, Wop = J56 , and if E = +1/2, Wop = K46 , and in general
Wop = (1/2 − E)J56 + (1/2 + E)K46 . (11.1.17)
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 343

When E is negative, the coefficient of J56 is greater than that of K46 , and
when E is positive the reverse is true. The unitary dilatation operators
exp(±iαJ45 ) can also be used to relate a system with arbitrary E to the
systems with the E values ±1/2. For E < 0 one uses the relation
exp(−iαJ45 )J56 exp(iαJ45 ) = cosh(α)J56 + sinh(α)K46 . (11.1.18)
Letting k to be a constant, one may in (11.1.17) set
(1/2 − E) = k cosh(α), (1/2 + E) = k sinh(α). (11.1.19a)
This requires
(1/2 − E)2 /k2 − (1/2 + E)2 /k2 = 1, (11.1.19b)
whence
k = (−2E)1/2 , (11.1.19c)
and
cosh(α) − sinh(α) = e−α = (−2E)1/2. (11.1.19d)
It follows that, when E < 0, the equation
(Wop − 1)ψ = 0 (11.1.20a)
can be expressed as
(exp(−iαJ45 )J56 exp(iαJ45 ) − 1)/(−2E)1/2)ψ = 0. (11.1.20b)
Multiplying this through by exp(iαJ45 ) gives
(J56 − 1/(−2E)1/2 )ψ  = 0, ψ  = exp(iαJ45 )ψ. (11.1.20c)
Thus ψ  is an eigenfunction of J56 , which has eigenvalues n, and the eigen-
functions χnlm defined Lie-algebraically in (11.1.9). Consequently,
n − 1/(−2E)1/2 = 0, e−α = 1/n, E = −1/(2n 2 ). (11.1.21)
One can thus write

ψnlm = exp(−iαJ45 )χnlm (r, θ, φ) = e−α χnlm (e−α r, θ, φ), (11.1.22a)
and

ψnlm = (1/n)χnlm (r/n, θ, φ). (11.1.22b)
The Schrödinger eigenfunctions Ψnlm (r, θ, φ) defined in (11.1.11) differ from

these ψnlm only because of the difference in normalization that arises from
October 28, 2010 16:55 9in x 6in b943-ch11

344 Dynamical Symmetry

the difference in scalar products. That is,

Ψnlm (r, θ, φ) = n −1/2 ψnlm


. (11.1.22c)

When E is positive, one replaces (11.1.18) with

exp(−iαJ45 )K46 exp(iαJ45 ) = cosh(α)K46 + sinh(α)J56 , (11.1.23)

and relates the general solution of (11.1.20a) to the eigenfunctions and

eigenvalues of the particular solution that is an eigenfunction of K46 . This
can be obtained from the eigenfunction of J56 because the relation between
these solutions can be established with the aid of a non-unitary dilatation.
On setting α = −iπ/2 in (11.1.18), one finds that

exp(−(π/2)J45 )J56 exp((π/2)J45 ) = cos(π/2)J46 − i sin(π/2)K46 = −iK46 .

Multiplying this through from the right by exp(−(π/2)J45 ) yields

K46 exp(−(π/2)J45 ) = i exp(−(π/2)J45 )J56 . (11.1.24)

Thus

K46 (exp(−(π/2)J45 )χnlm ) = i exp(−(π/2)J45 )J56 χnlm

= in(exp(−(π/2)J45 )χnlm ),

and the functions exp(−(π/2)J45 )χnlm are eigenfunctions of K46 with eigen-
values in. Let us set in equal to ν, and denote these eigenfunctions by χνlm .
Using (11.1.22a) establishes then that

χνlm = e−α χnlm (e−α r, θ, φ) = iχnlm (ir, θ, φ). (11.1.25)

11.2 The Momentum-space Representation

In the discussion of the SO(4) symmetry of classical Keplerian motions in
Chapter 8 it was noted that the hodographs of all trajectories are circles.
Then it was shown that the PB operators which generate the motions in
phase space produce rotations in an invariant momentum subspace — a
space in which the generators do not depend upon functions of the posi-
tion variables. These properties of Keplerian motion have the consequence
that in quantum mechanics the realization of the dynamical symmetry of
hydrogen-like atoms is both simpler and more direct in momentum space
than it is in position space. In this and the following section, we lay the
mathematical groundwork required to exploit this fact.
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 345

The momentum-space analogs of Schrödinger functions ψ(r) are their

Fourier transforms


φ(p) = (2π)−3/2 exp(−ip · r)ψ(r)d3 r, d3 r = dxdydz. (11.2.1)

Dirac’s introduction of his delta function, δ(x), greatly simplified the

theory of Fourier transforms. It has the properties6

∞
f(x)δ(x − x )dx = f(x ), (11.2.2a)
−∞

and

∞
= (2π)−1 exp(ipx)dpδ(x). (11.2.2b)
−∞

Letting δ(r) represent δ(x)δ(y)δ(z) one has

−3
δ(r) = (2π) exp(ip · r)d3 p (11.2.3a)

and


f(r)δ(r − r )d3 r = f(r ). (11.2.3b)

Using (11.2.3) one may derive that the inverse of transformation (11.2.1) is


ψ(r) = (2π)−3/2 exp(ip · r)φ(p)d3 p, d3 p = dpx dpy dpz . (11.2.4)

The position-space analog of i∂/∂px φ(p) is

−3/2
i∂/∂px (2π) exp(−ip · r)ψ(r)d r = (2π)−3/2 exp(−ip · r) xk ψ(r)d3 r,
3

(11.2.5a)

and the p-space analog of −i∂/∂x ψ(r) is

−i∂/∂x (2π)−3/2 exp(ip · r)φ(p)d3 p = (2π)−3/2 exp(ip · r)px φ(p)d3 p.

(11.2.5b)

These connections are those expected from the correspondence principle.

However the product of the Fourier transforms of two functions is not
the Fourier transform of the product of the functions. Symbolically

FT(f · g) = FT(f) · FT(g). (11.2.6a)

October 28, 2010 16:55 9in x 6in b943-ch11

346 Dynamical Symmetry

For two functions of r one has

FT(f(r) · g(r)) = (2π)−3/2 exp(−ip · r)(f(r) · g(r))d3 r. (11.2.6b)

By expressing g(r) as the Fourier transform of G(p ), and f(r) as that of

F(p), the right-hand side can be written as



(2π)−3/2 exp(−ip · r)f(r) (2π)−3/2 exp(ip · r)G(p )d3 p d3 r



−3/2 −3/2   3 
= (2π) (2π) exp(−i(p − p ) · r)f(p)G(p )d p d3 r



= (2π)−3/2 (2π)−3/2 exp(−i(p − p ) · r)f(p)d3 r G(p )d3 p .
(11.2.6c)
Thus


FT(f(r) · g(r)) = (2π)−3/2 F(p − p )G(p )d3 p .

An analogous string of connections establish the following relation

between position-space and momentum-space scalar products:



φ∗ (p)φ (p)d3 p = ψ ∗ (r)ψ  (r)d3 r. (11.2.7a)

To prove this relation one may apply (11.2.1), to its left-hand side, and
obtain


∗ 

(2π)−3 exp exp(−ip · r)Ψ(r)d3 r exp(−ip · r )Ψ (r )d3 r d3 p




−3/2 −3/2
= (2π) (2π) exp(ip · (r − r )) d3 p Ψ(r)∗ d3 r ψ  (r ) d3 r




= δ(r − r )ψ(r)∗ d3 rψ  (r )d3 r = ψ ∗ (r)ψ  (r)d3 r,
(11.2.7b)
and thus Eq. (11.2.7a) is established.
The momentum-space analog of the Schrödinger equation
(p2op /2 + v(r) − E)φ = 0, (11.2.8)
is

(p2 /2 + v(rop ) − E)φ = 0, rop = −∇2 . (11.2.9)
Because working with the function of rop is often very difficult, one pro-
ceeds differently. Taking the Fourier transform of the original Schrödinger
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 347

equation, one has

(2π)−3/2 {exp(−ip · r)(p2op /2 + v(r) − E)ψ}d3 r = 0, (11.2.10)

which becomes


(p2 /2 − E)φ(p) = (2π)−3/2 exp(−ip · r)v(r)ψ(r)d3 r


= φ(p )V(|p − p|)d3 p , (11.2.11)

where V(p) is the Fourier transform of v(r). This equation provides an

alternative to (11.2.9).
For hydrogen-like systems in which v(r) = −Z/r, the Fourier transform,
V(p) is (2/π)1/2 (−Z/|p|2 ). Setting E = −p20 /2, Eq. (11.2.11) then becomes


(1/2)(p0 + p )φ(p) = (2/π) Z φ(p )/|p − p|2 d3 p .
2 2 1/2
(11.2.12)

To solve this equation, Fock7 began by using a stereographic projection

to transform the integrand. His stereographic projection maps points with
coordinates px , py , pz in momentum-space to points with coordinates x1 ,
x2 , x3 , x4 on a unit hypersphere. The transformation, defined by θ → θ,
φ → φ, and

(1 − cos(α))
p/p0 → tan(α/2) = , (11.2.13)
(1 + cos(α))
sets x = (x1 , x2 , x3 , x4 ), with
x4 = (p20 − p2 )/(p20 + p2 ) = cos(α), (11.2.14a)
x3 = 2p0 pz /(p2 + p20 ) = sin(α) cos(θ), (11.2.14b)
x2 = 2p0 py /(p2 + p20 ) = sin(α) sin(θ) sin(φ), (11.2.14c)
2
x1 = 2p0 px /(p + p20 ) = sin(α) sin(θ) cos(φ). (11.2.14d)
It converts 1/(p − p)2 to
(2p20 /(p20 + p2 ))(1/|x − x |2 )(2p20 /(p20 + p2 )), (11.2.15a)
3 
and converts d p to
((p20 + p2 )/2p0 )3 dΩ . (11.2.15b)
Here,
dΩ = sin2 (α ) sin(θ ) dα dθ dφ (11.2.15c)
is the area element on the unit hypersphere defined by |x |2 = 1. Both x
and x have the same origin and terminate on this hypersphere. Inserting
October 28, 2010 16:55 9in x 6in b943-ch11

348 Dynamical Symmetry

these relations into the integrand, and multiplying both sides of (11.2.12)
by (p2 + p20 ), converts the equation to


(p2 + p20 )2 φ(p) = (Z/2p0 )(1/π)2 (p20 + p2 )2 φ(p )/|x − x |2 dΩ .
(11.2.16a)
On setting
(p2 + p20 )2 φ(p) = Φ(x), (11.2.16b)
and multiplying (11.2.16a) through by p0 , it takes on the simpler form


p0 Φ(x) = (Z/π ) Φ(x )/|x − x |2 dΩ .
2
(11.2.17)

Fock recognized that this equation is invariant under the rotations of

the group SO(4). To gain insight into how this invariance is expressed by
the equation itself, we begin with the much used expansion of the Coulomb
potential 1/|r − r |:
1/|r − r | = Σj (rj< /rj>+1 )(2j + 1)−1 Yjm
∗
(θ , φ )Yjm (θ, φ). (11.2.18a)
When both r and r terminate on a unit sphere, this becomes
1/|r − r | = Σj (2j + 1)−1 Yjm
∗
(θ , φ )Yjm (θ, φ). (11.2.18b)
The analog of (11.2.18b) in Euclidean four-space is the equation8
1/|x − x |2 = Σk (k + 1)−1 Ykjm
∗
(α , θ , φ )Yklm (α, θ, φ), (11.2.18c)
in which the functions Yklm are hyperspherical harmonics on S(3). Like the
spherical harmonics on S(2), these hyperspherical harmonics are members
of an orthornormal set. That is, one has


∗
Ykjm (α, θ, φ)Yk
l
m
(α, θ, φ)dΩ = δk ,k
, δl,l
δm,m
. (11.2.19)

On inserting (11.2.18c) into (11.2.17) one obtains


p0 Φ(x) = (Z/π 2 ) Φ(x ) Σk (k + 1)−1 Ykjm
∗
(α , θ, φ )Yklm (α, θ, φ) dΩ


= (Z/π )Σk (k + 1) Yklm (α, θ, φ) Φ(x )Ykjm
2 −1 ∗
(α , θ, φ )dΩ .
(11.2.20)
If one sets Φ(x ) = Yk
j
m
(x ) in the integral


Φ(x )Ykjm
∗
(α , θ, φ )dΩ , (11.2.21)
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 349

the orthonormality of the Y’s requires that it becomes unity if Φ is Ykjm (x).
Equation (11.2.20) then states that

p0 Φ(x) = (Z/(k + 1))Yk jm (x). (11.2.22a)

If both Φ(x) and Yk
j
m
(x) are normalized, this relation requires that

p0 = Z/(k + 1), Φ(x) = Yk jm (x). (11.2.22b,c)

Identifying k + 1 with the quantum number n, one obtains the energy,

E = −Z2 /2n 2 , and the associated wave function Yk
j
m
(x).
For a fixed value of n = k + 1, there are n 2 Yklm with different values of
l , m. One may insert any one of them, or any normalized linear combination
of them, into (11.2.2a), without changing the relation p0 = Z/(k + 1). Fock
used this fact to explain the n 2 -fold degeneracy of the energy spectrum of
hydrogen atoms. Shortly thereafter Bargmann showed that Pauli’s Runge–
Lenz operators,9 together with the angular momentum operators, obeyed
SO(4) commutation relations.10 This removed the possibility that Fock’s
remarkable result was an artifact produced by his stereographic projection.
Bargmann’s six operators are position-space realizations of the generators
which act in the S(3) projective momentum space of x. These are the six
operators
√ √
Jij = −1(xi ∂/∂xj − xj ∂/∂xi ) = −1(πi< ∂/∂πj< − πj< ∂/∂πi< ),
(11.2.23)
which are the quantum mechanical versions of the classical SO(4) generators
introduced in Sec. 8.4.

11.3 The Hyperspherical Harmonics Yklm

The relations
(p2 + p20 )2 φ(p) = Φ(x),
(11.3.1)
Φ(x) = Yklm (x),
of the previous section, together with the equations that define Fock’s
sterographic projection, enable one to establish the relation between
the hyperspherical harmonics Yn−1lm (α, θ, φ), and the momentum–space
eigenfunctions φnlm (p, θ, φ) that satisfy (11.2.17). Tracing through the con-
nections, the reader will find that the relation is
−3/2
φnlm (p/p0 , θ, φ) = Np0
−2 Yklm (α, θ, φ), (11.3.2a)
October 28, 2010 16:55 9in x 6in b943-ch11

350 Dynamical Symmetry

where

= (p2 + p20 )/2p20 = ((p/p0 )2 + 1)/2,
and N is a normalization factor that may also contain a phase factor.
Conversely,
3/2
Yklm (α, θ, φ) = N−1 p0
2 φnlm (p/p0 , θ, φ), (11.3.2b)
The momentum-space functions φnlm , are of the form
ρnl (p/p0 )Ylm (θ, φ). (11.3.2c)
The hyperspherical harmonics are directly defined in terms of Gegen-
bauer polynomials Cλν with ν = n − (l + 1) = k − l , and λ = l + 1. One
has11
Yklm (α, θ, φ) = Nl Cλν (cos(α)) sinl (α)Ylm (θ, φ), (11.3.3a)
with
Nl = (−i)l (2π 2 )−1/2 . (11.3.3b)
The phase factor, (−i)l , which makes the normalization factor l dependent,
has been chosen so that the hyperspherical harmonics satisfy a standard
phase convention5
∗
Yklm = (i)l−m Ykl−m . (11.3.3c)
The term sinl (α)Ylm (θ, φ) in (11.3.3), is a function of x1 , x2 , x3 that
is obtained from the usual Ylm (θ, φ) when one replaces x, y, z defined in
momentum-space by the x1 , x2 , x3 defined above. This leads us to define a
unit vector y in p-space, analogous to the unit vector x in the projective
space. Then one of the effects of Fock’s projection is to convert Ylm (y) to
Ylm (x).
−3/2
Its other effect is to convert the radial function ρnl (p/p0 ) to Nl p0
times
((p/p0 )2 + 1)−2 Cλν (cos(α)) = (1 + cos(α))2 Cλν (cos(α)) = (1 + x4 )2 Cλν (x4 )
= (1 + x4 )2 Cλν (x4 ).
(11.3.4)
Variations in the length of the momentum vector are mapped into variations
of x4 , and the consequent variations in the radial function ρnl (p/p0 ) are
expressed through variations in the Gegenbauer functions and the factor
(1 + x4 )2 . The color figures in Chapter 1 depict the stereographic projection
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 351

of the planar analogs of 2s and 2p momentum-space functions onto an

ordinary sphere.12
The Cλν satisfy Gegenbauer’s differential equation

{(1 − x2 )(∂/∂x)2 + (2λ + 1)x∂/∂x + ν(ν + 2λ)}Cλν (x) = 0. (11.3.5)

The Gegenbauer polynomials Cλν (x) play the same role in the theory of
the rotation groups SO(N), N > 3, that Legendre polynomials play in the
theory of SO(3). The Cλν (x) have a generating function, G, defined by

G ≡ (1 − 2hx + h2 )−λ = Σhν Cλν (x), (11.3.6)

a relation that is easily exploited using computer mathematics systems.

Two recent books by Avery provide clear and extensive treatments of the
properties of Gegenbauer functions and hyperspherical harmonics in spaces
of arbitrary dimension.8,13 Operators that shift the ν in Gegenbauer func-
tions are discussed in Appendix B. Table 11.3.1 displays a few of the Yklm ,
along with their corresponding Sturmian position-space functions. A more
extensive tabulation is contained in Judd’s Angular Momentum Theory for
Diatomic Molecules,14 which deals with many of the topics in this chapter.
The volume element in the four-dimensional Euclidean space of the x’s is

dx1 dx2 dx3 dx4 = ρ3 dρ dΩ, ρ2 = x21 + x22 + x23 + x24 , (11.3.7)

the area element being

dΩ = sin2 (α) sin(θ) dα dθ dφ.

Table 11.3.1. Hyperspherical harmonics and sturmians.

Harmonic (2π 2 )1/2 Yklm (α, θ, φ) (2π 2 )1/2 Yklm (x) Radial Sturmian

Y000 1 1 2e−r
√
Y100 −2 cos(α) −2x4 2 2(1 − r)e−r
√ √
Y110 i 2 sin(α) cos(θ) i 2x3 (2 2/ 3)re−r
√ √
Y111 i 2 sin(α) sin(θ)eiφ −i 2(x1 + ix2 )/21/2 (2 2/ 3)re−r
√ √
Y11−1 2 sin(α) sin(θ)e−iφ i2(x1 − ix2 )/21/2 (2 2/ 3)re−r
√
Y200 4 cos2 (α) −1 4x24 − 1 2 3(1 − 2r + 2r2 /3)e−r
√ √
Y210 −i61/2 sin2 (α) cos(θ) −i61/2 x4 x3 (4 2/ 3)r(1 − r/2)e−r
√ √
Y21±1 ±(3)1/2 sin(2α) sin(θ)e±iφ ±i(3)1/2 2x4 (x1 ± ix2 ) (4 2/ 3)r(1 − r/2)e−r
√
Y220 −21/2 sin2 (α)(3 cos2 (θ) − 1) −21/2 (3x23 + x24 − 1) (4/( 30))r2 e−r
√
Y22±1 ±(3)1/2 sin2 (α) sin(2θ)e±iφ ±i(3)1/2 2x3 (x1 ± ix2 ) (4/( 30))r2 e−r
√
Y22±2 −(3)1/2 sin2 (α) sin2 (θ)e±2iφ −(3)1/2 (x1 ± ix2 )2 (4/( 30))r2 e−r
October 28, 2010 16:55 9in x 6in b943-ch11

352 Dynamical Symmetry

The area element bears the following relation to the volume element d3 p in
momentum space:

dΩ =
−3 p−3 3
0 d p. (11.3.8)

The scalar product introduced in (11.2.19), that is,

π
π
2π
dα dθ dφ sin2 (α) sin(θ) Yklm
∗
Yk
l
m
(11.3.9)
0 0 0

∗
will in the following be expressed as Yklm Yk
l
m
dΩ, or (Yklm , Yk
l
m
).
From the previous discussion it follows that



∗
Yklm Yk
l
m
dΩ = φ∗nlm
φn
l
m
d3 p. (11.3.10)

The weight factor,

, in the momentum-space scalar product corresponds
to the factor 1/r in the position-space scalar product,15 for one has

(p2 + p20 )/2 = p2 /2 − E = 1/r. (11.3.11)

Thus Fock’s realization of hydrogen-like systems is the momentum–space

analog of the realization of the regularized Schrödinger system that was
discussed in Sec. 11.1. The two realizations provide isomorphic realizations
of the dynamical symmetries of regularized hydrogen-like systems.
The radial Sturmians are normalized using the rdr scalar product. Nor-
√
malization with the r2 dr scalar product multiplies them by n, n = k + 1.
At this point, let us summarize some general relationships. The cor-
respondence principle sets up an operator correspondence Xq → Xp and
careful attention to phase and normalization factors has ensured that the
position-space scalar products


χ∗nlm (1/r)Xq χn
l
m
dv = χnlm , Xq χn
l
m
, (11.3.12a)

and their momentum-space counterparts,

d3 p φ∗nlm
Xp φn
l
m
≡ φnlm , Xp φn
l
m
, (11.3.12b)

have the same values. To extend the correspondence to operators Ξ that

act in the hyperspherical momentum-space, we required that the scalar
products (11.3.12a,b) and


∗
Yklm Ξ Yk
l
m
dΩ ≡ (Yklm , ΞYk
l
m
), (11.3.12c)
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 353

also have the same values. The phase factors have been so chosen that


d3 p p−3
0

−2 ∗
Yklm
Xp
−2 Yk
l
m


= dΩ
−3
−2 Yklm ∗

Xp
−2 Yk
l
m


∗
= dΩ Yklm
2 Xp
−2 Yk
l
m . (11.3.24)

Thus if
Xp → Ξ =
2 Xp
−2 , (11.3.25a)
then
(Yklm , Ξ Yk
l
m
) = φnlm , Xp φn
l
m
. (11.3.25b)
The inverse of the similarity transformation defined by (11.3.25a), that is
Ξ → Xp =
−2 Ξ
2 , (11.3.25c)
connects the SO(4) rotation generators Ji4 , to momentum-space versions of
Runge–Lenz operators.
In Fock’s treatment, the dynamics of the atom is governed by the equa-
tion,


p0 Φ(x) = (Z/π ) Φ(x )/|x − x |2 dΩ ,
2
(11.3.26a)

which replaces the regularized Schrödinger equation

(Wop − Z)ψ = 0. (11.3.26b)
The equation


i∂Φ(x,τ )/∂τ = (Z/π 2 ) Φ(x , τ )/|x − x |2 dΩ , (11.3.27)

is an analog of Schrödinger’s time-dependent equation in which, i∂/∂t can

be considered equivalent to (−p0 /2) i∂/∂τ . The time-dependent equation
associated with (11.3.26b) is the analog of the corresponding classical equa-
tion of Chapter 8.
There are several important aspects of regularized quantum mechan-
ical treatments that make them particularly relevant to investigations of
more complicated atomic, molecular, and solid state systems. At this point
we wish only to call attention to a consequence of the fact that the hyper-
spherical harmonics Yklm comprise a discrete and complete set of functions:
any piecewise continuous function ϕ(x) defined on the hypersphere can be
October 28, 2010 16:55 9in x 6in b943-ch11

354 Dynamical Symmetry

expanded as series of these harmonics. When such a piecewise continuous

function is stereographically projected onto the surface of a hypersphere,
the projection is necessarily piecewise continuous, except, at the “South
pole”.
Finally, it should be noted that the reciprocal-space functions of solid
state physics have a natural projection onto Fock’s hypersphere, where they
become superpositions of hyperspherical harmonics. Several depictions of
molecular reciprocal-space functions will be found in Avery’s book, Hyper-
spherical Harmonics.13

11.4 Bases Provided by Eigenfunctions of J12 , J34 , J56

The SO(4) group whose generators, Jab = −Jba , satisfy the classical com-
mutation relations

[Jab , Jbc ] = Jac , (11.4.1)

has two commuting generators, which may be choosen to be J12 and J34 .
The quantum mechanical position-space operators corresponding to the Jab
satisfy commutation relations in which the negative sign on the right-hand
side is replaced by its square root, and the same may be said of the corre-
sponding momentum-space operators. In the case of the SO(4) degeneracy
group of hydrogen-like atoms, one may label degenerate eigenstates | by
the quantum number n, and by the eigenvalue m12 of J12 and the eigenvalue
m34 of J34 . Then one has the n2 states defined by

J56 |m12 , m34 , n = n|m12 , m34 , n, (11.4.2a)

and

J12 |m12 , m34 , n = m12 |m12 , m34 , n, (11.4.2b)

J34 |m12 , m34 , n = m34 |m12 , m34 , n. (11.4.2c)

An alternative labeling turns out to be revealing.16 Define

M1 = (J23 + J14 )/2, M2 = (J31 + J24 )/2, M3 = (J12 + J34 )/2,

(11.4.3)
N1 = (J23 − J14 )/2, N2 = (J31 − J24 )/2, N3 = (J12 − J34 )/2,

one finds that all the M operators commute with all the N operators. It
follows that SO(4) is the direct product of two three-parameter groups.
The quantum mechanical version of the operators satisfy

[Ma , Mb ] = i eabc Mc , [Na , Nb ] = i eabc Mc , [Mi , Nj ] = 0. (11.4.4)

October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 355

One may choose M3 , N3 , and J56 to be a commuting set of operators defining

the n 2 states that satisfy

J56 |m3 , n3 , n = n|m3 , n3 , n,

M3 |m3 , n3 , n = m3 |m3 , n3 , n, (11.4.5)
N3 |m3 , n3 , n = n3 |m3 , n3 , n.

The labeling schemes (11.4.2) and (11.4.5) are connected by the relations

m3 = (m12 + m34 )/2, n3 = (m12 − m34 )/2. (11.4.6)

As m3 and n3 can have half-integer values, SO(4) = SU(2) ⊗ SU(2). As will

be seen in the next chapter, this fact can be quite useful.

Appendix A. Matrix Elements of SO(4,2) Generators3,16

In the previous pages only the matrix elements of labeling operators have
been presented. To represent the action of the remaining group generators
and other important operators on the |nlm = |1nlm basis, it is convenient
to define the raising and lowering operators:

L± = J23 ± iJ31 , N± = K46 ± iK45 , Γ± = K16 ± iK26 ,

(11.A.1)
A±
− = J14 ± iJ24 , A±
+ = K15 ± iK25 .

It is also useful to define the functions

c0 (a, b) ≡ ((a2 − b2 )/(4b2 − 1))1/2 , d0 (a, b) ≡ (a2 − b2 )1/2 ,
(11.A.2)
d±1 (a, b) ≡ ((a ± b)(a + 1 ± b))1/2 .

One has

N± |nlm = (n(n + 1) − l (l + 1))1/2 |n ± 1, l , m, (11.A.3a)

J34 |nlm = c0 (n, l + 1), d0 (l + 1, m)|n, l + 1, m + c0 (n, l ),
(11.A.3b)
d0 (l , m)|n, l − 1, m
A±
− |nlm = ±{−c0 (n, l + 1, m) d±1 (l , m)|n, l − 1, m ± 1

− c0 (n, l ) d±1 (l − m, −m)|n, l + 1, m ± 1} (11.A.3c)

K35 |nlm = (1/2)c0 (n, l + 1) d0 (l + 1, m)
{[(n + l + 2)/(n − l − 1)]1/2 (11.A.3d)
|n + 1, l + 1, m
+ [(n − l − 2)/(n + l + 1)1/2 ]|n − 1, l + 1, m}
October 28, 2010 16:55 9in x 6in b943-ch11

356 Dynamical Symmetry

+(1/2) c0 (n, l ) d0 (l , m){[(n − l + 1)/(n + l )]1/2 |n +1, l − 1, m

+ [(n + l + 2)/(n−l −1)]1/2 [(n + l − 1)/(n − l )]1/2

× |n − 1, l − 1, m}

K36 |nlm = c0 (n, l + 1) d0 (l + 1, m){(−1/2)[(n − l − 2)/(n + l + 1)]1/2

× |n − 1, l + 1, m + (1/2)[(n + l + 2)/(n − l − 1)]1/2

× |n + 1, l + 1, m}

+ c0 (n, l ) d0 (l , m){(−1/2)[(n − l − 2)/(n + l + 1)]1/2

× |n − 1, l − 1, m

+ (1/2)[(n + l + 2)/(n − l − 1)]1/2 |n + 1, l − 1, m} (11.A.3e)

Using the relations r = (J56 − K46 ), and z = (K35 − J34 ), one obtains:

r|nlm = (1/2)[(n − l − 1)(n + l )]1/2 |n − 1, l , m + n|nlm

+ (1/2)[(n + l + 1)(n − l )]1/2 |n + 1, l , m.

and

z|nlm = c0 (n, l + 1)d0 (l +, m){(1/2)[(n − l − 1)(n + l + 1)]1/2

× |n − 1, l + 1, m + |n, l + 1, m
+ (1/2)(n − l − 1)1/2 |n + 1, l + 1, m}
+ c0 (n, l ) d0 (l , m){(1/2)[(n − l − 1)(n + l +1)]1/2 |n − 1, l − 1, m
+ |n, l − 1, m + (1/2)(n − l − 1)1/2 |n + 1, l − 1, m}

Appendix B. N-Shift Operators For the Hyperspherical

Harmonics
Kumei17 discovered that Gegenbauer’s general differential equation,

{(1 − x24 )(∂/∂x4 )2 + (2λ + 1)x4 ∂/∂x4 + ν(ν + 2λ)}Cλν (x4 ) = 0.

(11.B.1a)

has an SO(2,1) invariance group which contains ν-labeling and ν-shift

operators.
In the special case of interest here, ν(ν +2λ) = n 2 −l 2 , and the operators
shift the princial quantum number n. Kumei linearized the spectrum of ν
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 357

by converting (11.B.1a) to

{(1 − x24 )(∂/∂x4 )2 + (2λ + 1)x4 ∂/∂x4 + i∂/∂τ (i∂/∂τ + 2λ)} (11.4.1)
× exp(−iντ )Cλν (x4 ) = 0. (11.B.1b)

The determining equations for the generators of the invariance group of this
equation yield the required labeling and v-shift operators for the Cλν when
one allows Q’s that correspond to first-order differential operators, U. The
solutions of the determining equations are defined to within a multiplicative
constant. He chose them to be

Q1 = i∂/∂τ, Q2 = 1, (11.B.2a)
Q− = exp(iτ ){(1 − x24 )∂/∂x4 + i x4 ∂/∂τ }, (11.B.2b)
Q+ = exp(−iτ ){(1 − x24 )∂/∂x4 − i x4 ∂/∂τ − 2λ x4 }. (11.B.2c)

Q1 is the ν-labeling operator. The n-labeling operator of interest in the

present context is

Q0 = Q1 + λQ2 . (11.B.2d)

The operators Q0 , Q+ and Q− satisfy

[Q+ , Q− ] = 2Q0 , [Q± , Q0 ] = ±Q± . (11.B.2e,f)

Consequently

Q± exp(−iντ )Cλν (x4 ) = c± (n, l ) exp(−i(ν ± 1)τ )Cλν±1 (x4 ). (11.B.2g)

They shift ν by ±1 without shifting l , and hence shift n by ±1. The Casimir
operator is

C2 = Q20 + (Q+ Q− + Q− Q+ )/2. (11.B.2h)

The functions exp(−iντ )Cλν (x4 ) are eigenfunctions of C2 with eigenvalue

λ(λ − 1), i.e., l (l + 1).
To obtain n-labeling and shift operators that act on the functions

exp(−iντ )Yklm = exp(−iντ ) Nl sin(α)l Cλν (x4 )

(11.B.3)
= exp(−iντ ) Nl (1 − x24 )l/2 Cλν (x4 ),
we use a similarity transformation to exploit Kumei’s general result. Let

Q exp(−iντ )Cλν (x4 ) → N exp(−iντ )(1 − x24 )l/2 Cλν (x4 ), (11.B.4a)
October 28, 2010 16:55 9in x 6in b943-ch11

358 Dynamical Symmetry

with
N = (1 − x24 )l/2 Q(1 − x24 )−l/2 . (11.B.4b)
Q0 = i∂/∂τ + λ ≡ N0 is not altered by the transformation. If we let
|nlm = exp(−iντ )Yklm (11.B.5a)
then
N0 |nlm = n|nlm. (11.B.5b)
Also let
N± |nlm = N± exp(−iντ )Yklm = {(1 − x24 )l/2 Q± (1 − x24 )−l/2 }
× exp(−iντ |Nl (1 − x24 )l/2 Cλν (x4 )
= (1 − x24 )l/2 Q± exp(−iντ ) Nl Cλν (x4 ) (11.B.5c)
= (1 − x24 )l/2 c± (n, l ){exp(−i(ν ± 1)τ ) Nl Cλν±1 (x4 )}
= c± (n, l )|n ± 1, l , m
One finds
N− = exp(+iτ ){(1 − x24 )∂/∂x4 + x4 (N0 − 1)}, (11.B.5d)
N+ = exp(+iτ ){(1 − x24 )∂/∂x4 − x4 (N0 + 1)}. (11.B.5e)
Because they have been obtained by similarity transformation, these Q’
operators satisfy the same commutation relations as the original Q opera-
tors. The corresponding Casimir operator has eigenvalues l (l + 1).
If N+ and N− are matrices with elements n  lm, N± |nlm, then one
finds that N− = −N†+ : the operator N− is the negative of the adjoint of
N+ . In the general case,

N± |nlm = c± (n, l )|n ± 1, l , m (11.B.6)
with
c± (n, l ) = ±((l + 1 ± n)(±n − l ))1/2 .
The operators Q also yield operators Q = exp(iντ )Q exp(−iντ ) that can
be used to produce operators which act directly on the Yklm . After removing
the time-dependent exponents from the Q , the operator Q0 → N0 , and
one obtains operators Θ± that convert Yklm to c± (k + 1 , l ) Yk ±1lm . These
operators are
Θ− = (1 − x24 )∂/∂x4 + (k − l )x4 ,
(11.B.7)
Θ+ = (1 − x24 )∂/∂x4 + (k − l )x4 − 2(l + 1)x4 .
October 28, 2010 16:55 9in x 6in b943-ch11

Dynamical Symmetry of Regularized Hydrogen-like Atoms 359

References
[1] A. O. Barut and H. Kleinert, Phys. Rev. 156 (1967) 1541.
[2] C. Fronsdal, Phys. Rev. 156 (1967) 1665.
[3] Y. Nambu, Phys. Rev. 160 (1967) 1171.
[4] M. Bednar, Ann. Phys. (N.Y.) 75 (1973) 305.
[5] B. R. Judd, Angular Momentum Theory for Diatomic Molecules (Academic
Press, N. Y., 1975), pp. 32–34.
[6] P. A. M. Dirac, Quantum Mechanics, 3rd edn. (Oxford, 1947), p. 53.
[7] V. Fock, Z. Phys. 98 (1935) 145.
[8] J. Avery, Hyperspherical Harmonics and Generalized Sturmians (Kluwer,
Dordrecht, 2000), p. 63.
[9] W. Pauli, Z. Phys. 36 (1926) 336.
[10] V. Bargmann Z. Phys. 99 (1936) 576.
[11] J. Avery, ibid. (Ref. 8) p. 25.
[12] T. Shibuya and C. E. Wulfman, Amer. J. Phys. 33 (1965) 570.
[13] J. Avery, Hyperspherical Harmonics (Kluwer, Dordrecht, 1989).
[14] B. R. Judd, ibid. (Ref. 5) p. 222.
[15] T. Shibuya, C. E. Wulfman, Proc. Roy. Soc. (London) A 286 (1965) 376.
[16] L. C. Biedenharn, J. Math. Phys. 2 (1961) 433.
[17] S. Kumei, M. Sc. Thesis (University of the Pacific, 1972).
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-ch12

CHAPTER 12

Uncovering Approximate Dynamical Symmetries.

Examples From Atomic and Molecular Physics

12.1 Introduction
In 1885, Annalen der Physik published a paper on the physical relevance of
certain small integers.1 In the paper, the Basel secondary-school teacher,
J. J. Balmer, pointed out that the observed wavelengths, λ, of the Hα , Hβ ,
Hγ , and Hδ lines in the sun’s Fraunhofer spectrum2 obeyed, with amazing
accuracy, a relation

λ = Gn 2 /(n 2 − 4), (12.1.1)

in which G is a constant, and n = 3, 4, 5, 6. It was soon recognized that

if one sets n1 = n and n2 = 2, the corresponding frequencies, ν, of these
Balmer lines satisfy the general equation

ν = R(1/n22 − 1/n12 ), (12.1.2)

which is obeyed by subsequently related lines in the spectrum of hydrogen.

In this equation, R, the Rydberg constant, is 109667.581 cm−1 .
When Bohr introduced his theory of the atom,3 the integers n
became the principal quantum numbers of the Bohr and Bohr–Sommerfeld
theory.3,4 Ehrenfest then showed that n, and many other quantum num-
bers, defined closed areas of magnitude n h/2π in phase space. And then
in 1935 Fock discovered that n defines an SO(4) symmetry.
The advent of quantum mechanics in 1926, together with extensive con-
tributions from experimentalists, led to rapid strides in many branches of
physics. The immediate consequences in atomic physics are evident in Con-
don and Shortley’s 1935 monograph,5 written shortly before Fock’s dis-
covery. By 1931 Wigner and Eckart had already produced, and demon-
strated the utility of SO(3) coupling coefficients.6 In the early 1940’s this
work was greatly extended by Racah, who also introduced the concept of

361
October 28, 2010 16:55 9in x 6in b943-ch12

362 Dynamical Symmetry

seniority.7 Many subsequent developments of the group theory relevant to

atomic physics are expounded and reviewed in Edmond’s monograph,8 and
in monographs by Biedenharn and Louck,9 Wybourne,10 and Judd.11
The development of methods for approximating the solutions of deter-
mining equations makes it possible to systematically investigate the dynam-
ical groups of Schrödinger equations more complex than those considered
in Chapter 11, but it does not obviate the need to use concepts, such as
those of the theory of angular momenta, to simplify equations as complex
as those governing many-particle systems, such as those arising in atomic
and molecular physics. The development of relevant simplifications is the
subject of the next few sections of this chapter, and much of Chapter 13.
In recent years it has been shown that the use of rather small Stur-
mian basis sets, or equivalent sets of hyperspherical harmonics, can lead to
surprisingly accurate calculations of atomic energies.12 Limited basis sets
are able to produce much better approximate energy and potential energy
curves than produced by LCAO or Gaussian bases of similar size. This
approach will be outlined in Secs. 12.10 and 12.11. At this point we would
note that in previous chapters it has become apparent that the genera-
tors of the dynamical groups SO(4), SO(4,1), and SO(4,2), are most sim-
ply expressed in these bases. This suggests that as the use of these bases
increase, previously unrecognized dynamical symmetries, and their conse-
quences, will be uncovered. The topics in this chapter can only hint at the
possibilities.

12.2 The Stark Effect; One-Electron Diatomics

The regularization of the Hamiltonian for a hydrogen atom immersed in a
uniform electric field of strength ε in the z direction gives
r(H − E) = r(p2 /2 − E − εz − 1/r) = W0 − εz r − 1. (12.2.1)
To understand how the field affects the degeneracy of the states |nlm
one
need only examine the matrix elements of states of the same n;
nlm, r z nlm
. (12.2.2)
Now,
z = Bz − Az (12.2.3)
and B only has nonzero matrix elements between states with different n
values, so the removal of the degeneracy is due to Az = J34 . The SU(2) ⊗
SU(2) states |m12 m34 n
defined in Eq. (11.4.2) are eigenstates of Lz and Az
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 363

with eigenvalues m12 and m34 , respectively. These then become the zero-
order eigenstates of the Stark effect. The |nlm
can be considered as being
obtained from them by the coupling defined by


|nlm
= (j1 m1 j2 m2 |j1 j2 lm)|j1 m1, j2 m2
, (12.2.4a)
m1 m2

for the special case in which j1 = j2 . As the coupling transformation

|j1 m1, j2 m2
= (j1 m1 j2 m2 |j1 j2, lm)|nlm
(12.2.4b)
m

is unitary, and the energy shift defined by (12.2.2) may be calculated by

evaluating the radial integral

χnl
m
(r)(1/r)r χnlm (r) r2 dr sin(θ)dθ dφ
 (12.2.5)
2
= χnl
m
(r) z χnlm (r) r dr sin(θ)dθ dφ.

The integral can be evaluated by relating the Sturmians to the ordinary

hydrogen-atom functions Ψnlm . Referring back to (11.1.11b), one has
Ψnlm (r, θ, φ) = il n −3/2 χnlm (r/n), (12.2.6a)
and
exp(iαK45 ) χnlm (r) = χnlm (r/n). (12.2.6b)
Using these relations one can easily evaluate

χnl
m
(r) z χnlm (r) r2 dr sin(θ)dθ dφ. (12.2.7)

A most enlightening discussion of the Stark effect is contained in Sommer-

feld’s Atomic Structure and Spectral Lines.4
In 1964 Redmond13 found that the Stark effect Hamiltonian
p2 /2m − Ze2 /r + eE, (12.2.8)
has an exact constant of motion, C, that is obtained by adding a term
−(r x E) x r/2Ze to the Runge–Lenz vector operator:
C = r/r + (L x p − p x L)/(2Ze2 m) − (r x E) x r/2Ze. (12.2.9)
14
In 1949 Erikson and Hill showed that one-electron diatomic molecule–
ions possess a constant of motion in addition to the component, m
of the
electronic angular momentum on their axis. If the molecular axis is the
z axis, and internuclear distance is 2a, the Hamiltonian, whose value is
October 28, 2010 16:55 9in x 6in b943-ch12

364 Dynamical Symmetry

E(a), is
H = p2 /2 − (Za /|q − ak| + Zb /|q + ak|), (12.2.10)
and the constant of motion can be expressed as
F = L2 + a2 p2z + 2az(Za /|q−ak| − Zb /|q + ak|). (12.2.11)
The corresponding Poisson bracket operator is the generator of a one-
parameter symmetry group in classical phase space. As F commutes with
Lz , the degeneracy group is Abelian.
The non-crossing rule applies to sets of energy level curves such as E(a).
Wigner and von Neumann proved that varying a single parameter cannot
produce crossings of levels whose wave functions have the same symmetry;
if they appear to do so, a symmetry has been overlooked.15 One-electron
diatomics provide one of the simplest such examples. Some levels of the
same spatial symmetry cross because they have different dynamical symme-
try. In diatomics with two or more electrons this symmetry is removed, and
levels that previously crossed approach, but avoid, each other. The reader
is referred to Judd’s monograph11 for discussions of a number of interest-
ing aspects of the dynamical symmetry of one-electron diatomics. Included
within it is a discussion of the simple solutions that arise for special sets
of nuclear charges. In some later calculations on systems with Za = 1, and
Zb = 3, . . . , 8, Ponomarev and Puzynia observed pseudo-intersections.16
This raises a suspicion that as the bond between the proton and the other
nucleus weakens, such systems may have a useful approximate dynamical
symmetry larger than that of homonuclear systems.

12.3 Correlation Diagrams and Level Crossings:

General Remarks
Whenever a change in a Hamiltonian produces a new degeneracy, new phe-
nomena may develop. The symmetry involved can be revealing, and may
have practical importance.
If the change in a Hamiltonian can be produced by changing a param-
eter, much can be learned from a simple plot of the energy level diagram
arising from varying the parameter. Let λ be a parameter, and write
H = H0 + λH1 , E = Em,0 + λEm,1 . (12.3.1)
As a simple example, suppose that H is the Hamiltonian of a molecule
with one or more double bonds, such as the molecules depicted in
Figs. 12.3a–d. In these figures, λ is the angle θ that measures the twist
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 365

B1 A2

B1 B2

A1
A2
A1
B2

Fig. 12.3a Twisting of a substituted ethylene molecule.

Fig. 12.3b The p-orbitals attached to atoms C1 and C2 .

Fig. 12.3c Relation between the p-orbitals after a twist through 45◦ .

of the A2 C2 B2 plane with respect to the A1 C1 B1 plane. In (b) the two

p orbitals that produce the π bond are depicted. As the twist increases to
90◦ , the overlap between them decreases to zero, and the bond is destroyed.
Further increase of θ to 180◦ produces a state that has the same energy
October 28, 2010 16:55 9in x 6in b943-ch12

366 Dynamical Symmetry

Fig. 12.3d Relation between the p-orbitals after a twist through 180◦ .

as the excited state, which may be produced when the molecule in (c)
absorbs a photon. Such photo-excited molecules often undergo cis-trans
isomerism — that is twist through 180◦ to produce the ground state
of the molecule, (b). If A1 and B1 are identical to A2 and B2 respec-
tively, there must be a level crossing that produces a degeneracy at the
90◦ twisted configuration.17 In systems with less geometrical symmetry,
degeneracies develop when the p orbitals are approximately perpendic-
ular to each other. The resulting degenerate states are very sensitive to
the presence of polarizing fields with components in the C1 –C2 direction,
and the molecules begin to develop large dipole moments when the twist-
ing becomes a few degrees less than 90◦ . In the case of ethylene, a pro-
ton situated along this axis and a C–C bond length away produces the
dipole moment obtained by moving a charge of 0.9e from one C atom
to the other.17 In molecules with a conjugated system of double bonds,
asymmetries due to substituents may lead to sudden polarizations in the
twisted states that produce very large dipole moments. Bruckner and Salem
proposed that just such a sudden photo-polarization of retinal is respon-
sible for initiating nerve impulses in the retina.18 For further informa-
tion the reader is referred to the proceedings of a conference on sudden
polarization.19
When variation of a parameter changes the spatial point-group sym-
metry of a nonlinear molecular system, the theorem of Jahn and Teller
can come into play.20 It dictates that such a variation creates a degener-
acy of energy levels, when in the region of the crossing the energies are
linear functions of the parameter. It follows that if the degeneracy were to
occur naturally, the molecule would undergo a distortion that removes the
symmetry.
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 367

Fig. 12.3e Correlation between the united-atom and separated-atom energy lev-
els of one-electron linear homonuclear triatomic molecules with identical bond
lengths.

If variation of a parameter bends a linear system, an extension of the

Jahn–Teller Theorem becomes relevant. In the linear system, rotational
symmetry about the axis of linearity allows one to assign to each level
a rotational quantum number m. As the linear system bends, the ener-
gies of all levels must depend quadratically upon the bending angle in
the vicinity of the linearity. The bending produces a mixing of pairs of
degenerate levels whose m values are the negative of one another. One of
the new levels then exhibits a quadratically increasing energy, the other a
quadratically decreasing energy. If the degeneracy develops naturally, the
linear system bends on its own accord. A similar extension of the Jahn–
Teller Theorem applies when planar systems are bent to produce nonpla-
nar ones.
October 28, 2010 16:55 9in x 6in b943-ch12

368 Dynamical Symmetry

In 1953 the molecular spectroscopist Walsh published a set of rules

which state that the shape of triatomic and tetraatomic molecules depends
only upon small integers — the total number of valence electrons in the
molecule.21 Figure 12.3e is a correlation diagram which will be used to
illustrate the manner in which the Wigner–von Neumann rule and the
extended Jahn–Teller Theorem help to imply one of Walsh’s rules. The
rule under consideration is “All non-hydride triatomic molecules contain-
ing 15 or 16 valence electrons are linear, while those containing 17, 18, 19,
or 20 valence electrons are nonlinear”. These rules were the first to lead
to a revelation of the role dynamical symmetries and can play in deter-
mining point-group symmetries in position space.22,23 In the figure, the
one-electron energy levels of linear homonuclear triatomic systems, such
as N− 3 , are correlated with their united-atom and separated-atom limits
by varying a scaling parameter which uniformly alters all internuclear dis-
tances. The D∞h spatial symmetry of each wave function is indicated by
the labels σ, π, g and u. The SO(4) united-atom N symmetry is broken,
but the L symmetry remains amazingly good because the integrals
 ∞
Rnl1 (Zr/n)(rk /r)Rnl2 (Zr/n)r2 dr, l2 = l1 , (12.3.2)
0

vanish for a wide range of k values.24a When Neumann’s expansion of

1/|r − r | is applied to the integrals
 ∞
Ψnl1m1 (Zr/n)(1/|r − R|)Ψnl2m2 (Zr/n)dv, (12.3.3)
0

some, as expected, vanish as a consequence of SO(3) symmetry. Oth-

ers, however, vanish unexpectedly as a consequence of SO(4) symmetry.22
Swamy, Kulkarni, and Biedenharn subsequently proved that whenever they
vanish it is because of SO(4,2) symmetry.24b
In developing Fig. 12.3e, the center of nuclear charge of the system was
assumed to be the origin of coordinates. This eliminates terms depending
linearly on the nuclear displacements. The forbidden crossings that develop
are those involving πu orbitals. When electron-repulsion effects are taken
into account, one finds that they take place in the united-atom. As a conse-
quence, no matter what the size of the system is, the degenerate πu orbitals
always begin to be occupied when the system contain 17 valence electrons.
This remains true when one mathematically shifts nuclei about to create
non-hydride systems, such as NCO, that are no longer homonuclear. The
extended Jahn–Teller Theorem then requires that molecules with 17, 18
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 369

or 19 valence electrons are nonlinear. As the systems are bent, the sym-
metry becomes C2v . However, levels of the same C2v symmetry, but with
different l, cross, and the πu orbitals correlate with orbitals of much lower
energy. Systems with 15 or 16 valence electrons remain linear, but those
with 17 to 19 valence electrons are bent. In 20 electron systems, bend-
ing increases the electron density at the center of nuclear charge, and
this stabilizes the bent system. Extensive analyses have fully established
the connection between all of Walsh’s rules and the extended Jahn–Teller
Theorem.25
The examples in this section have been chosen to illustrate a general
strategy that is available whenever one is dealing with a system governed
by differential equations. If one begins to attack the problem by seeking the
dynamical symmetry that, when broken, can produce the phenomenon of
interest, several advantages result:

a. The simplest form of perturbation theory (in quantum mechanics, degen-

erate perturbation theory) usually enables one to determine with cer-
tainty the physical interactions that are involved.
b. The extensive and well organized mathematics of the theory of Lie alge-
bras and groups is available to guide and aid the investigation.

The strategy enables one to treat physical phenomena as displays of the

consequences of symmetries of differential equations and their symmetry
breaking.
For a Coulomb system of 3N degrees of freedom in position-space, we
have seen that the relevant dynamical group may be as large as a con-
strained SO(4N,2N). However, in this chapter we can only adumbrate
studies of the physical relevance of a few subgroups of the two-particle
dynamical group SO(8,4). The goal is to gain a better understanding of the
consequences of electronic repulsions. To apply the general strategy just
outlined, several of the following sections are devoted to investigations of
SO(4) ⊗ SO(4). An example of the direct physical relevance of the larger
group is described in Sec. 12.8.

12.4 Coupling SO(4)1 ⊗ SO(4)2 to Produce SO(4)12

The mathematics described in this section applies to atoms containing two
non-interacting electrons, each having associated Runge–Lenz and angular
momentum vectors, Aj and Lj . The components of A = A1 +A2 and the
components of L = L1 +L2 close under commutation, and generate an
October 28, 2010 16:55 9in x 6in b943-ch12

370 Dynamical Symmetry

SO(4) group which is sometimes designated as SO(4)12 . In the absence

of other interactions this becomes a degeneracy group of the two-electron
system. As the generators of this group are the sum of the generators of two
SO(4) groups denoted by SO(4)1 and SO(4)2 , the group SO(4)12 is their
direct product, SO(4)1 ⊗ SO(4)2 . The situation is closely analogous to the
SO(3) ⊗ SO(3) system discussed in Chapter 10.
Here we wish to produce functions that are eigenfunctions of the two-
particle SO(4) Casimir operators C2 = A2 + L2 , and C2 = A · L and also of
L2 and Lz . The one-particle Casimir operators A21 + L21 , A22 + L22 , commute
with C2 and C2 , as of course do all 12 generators of SO(4)⊗SO(4). However,
A1 does not commute with L21 , and A2 does not commute with L22 . It follows
that the SO(4)12 eigenfunctions which we seek must be expected to be linear
combinations of products of one-particle states of differing l and m values.
The eigenvalues of C2 and C4 are determined by two quantum numbers
P and Q. These observations may be summarized with the statement that
the desired SO(4)12 eigenfunctions could be labeled |PQLM (n1 , n2 )
. These
functions are a linear combination of products, |n1 l1 m1
|n2 l2 m2
, of single
particle functions:

|PQLM (n1 , n2 )



= B(n1 l1 m1 , n2 l2 , m2 |n1 n2 , PQLM )|n1 l1 m1
|n2 l2 m2
. (12.4.1)
l1 l2 m1 m2

The coefficients B(n1 l1 m1 , n2 l2 , m2 |n1 n2 , PQLM ) are SO(4) ⊗ SO(4) cou-

pling coefficients analogous to the (l1 m1 l2 , m2 |l1 l2 , LM ) coupling coeffi-
cients of SO(3) ⊗ SO(3). In 1961 Biedenharn determined their dependence
upon the quantum numbers involved.26 In doing so he imposed the
phase convention of Condon and Shortley,27 and the mathematics
he used required him to make the assumption that A, like L, had the
derivative property possessed by first-order derivative operators, DOp ,
which is DOp (f ∗ g) = f ∗ (DOp (g)) + (DOp (f))∗ g. A determination of the
B(n1 l1 m1 , n2 l2 , m2 |n1 n2 , PQLM ) due to Wybourne does not require this
last assumption.28 Biedenharn also established that

C2 |PQLM (n1, n2 )
= (P (P + 2 ) + Q 2 )|PQLM (n1, n2 )
, (12.4.2)

A · L|PQLM (n1 , n2 )
= Q(P + 1)|PQLM (n1, n2 )
. (12.4.3)

Q may take on both positive and negative values. Because A·L is

not invariant under the parity operation x → −x, the SO(4) functions
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 371

|PQLM (n1 , n2 )
may not have definite parity when n1 = n2 . Eigenfunc-

tions of the O(4) Casimir operator (A·L)2 , with eigenvalue Q 2 (P + 1)2 ,
and definite parity, may be constructed by taking linear combinations of
the functions obtained by interchanging the two quantum numbers.
The two-electron wave functions |PQLM (n1 , n2 )
may be thought of as
obtained by a recoupling of the already coupled angular momentum eigen-
states |LM (l1 , l2 , n1 , n2 )
. In other words, the states |PQLM (n1 , n2 )
can
be considered to be linear combinations of states |LM (l1 , l2 , n1 , n2 )
with
differing l values:


|PQLM (n1 , n2 )
= C(LM (l1 , l2 , n1 , n2 ), PQLM )|LM (l1 , l2 , n1 , n2 )
.
l1 ,l2
(12.4.4)
Multiplying these by the appropriate spin functions, |σ
, produces states
that are antisymmetric for particle exchange. When n1 = n2 = 2, for exam-
ple, the two singlet |LM (l1 , l2 , n1 , n2 )
|1 σ
states are

1
S(2s2 ) = |00(0, 0, 2, 2)
|1σ
, 1
S(2p2 ) = |00(1, 1, 2, 2)
|1σ
. (12.4.5)

The first state is simply a product of two 2s functions and a spin func-
tion, |1 σ
, that is antisymmetric under exchange. In the second state, prod-
ucts of two 2p functions are coupled to produce a state of zero angular
momentum that is symmetric under particle exchange. Multiplying this
by |1 σ
produces the 1 S(2p2 ) wave function. Together, these produce the
|PQLM (n1 , n2 )
states29

|0000(2, 2)
|1σ
= 0.524 1 S(2s2 ) − 0.851 1 S(2p2 ),

(12.4.6)
|2000(2, 2)
|1σ
= 0.851 1 S(2s2 ) + 0.524 1 S(2p2 ).

In this case, Q = 0 for both states, and A1 +A2 is orthogonal to L.

12.5 Coupling SO(4)1 ⊗ SO(4)2 to Produce SO(4)1−2

The group SO(4) has a most surprising property30 : two different SO(4)12
groups can be obtained by the coupling of SO(4)1 and SO(4)2 . The alter-
native coupling is the subject of this section.
The components of A− = A1 − A2 and L = L1 + L2 close under
commutation and generate an SO(4) group which we designate SO(4)1−2 .
October 28, 2010 16:55 9in x 6in b943-ch12

372 Dynamical Symmetry

The Casimir operators of the group are

C2 = A2− + L2 and C2 = A− · L. (12.5.1)

The eigenstates of these operators and the operators

L2 , Lz , A21 + L21 , A22 + L22 ,

are of the form

|PQ LM (n1 , n2 )
= D(LM (l1 , l2 , n1 , n2 ), PQ LM )|LM (l1 , l2 , n1 , n2 )
.
l1 ,l2
(12.5.2a)
They are eigenfunctions of C2 and C2 :

(A2− + L2 )|PQ LM (n1 , n2 )
= (P (P + 2) + Q 2 )|PQ LM (n1 , n2 )
,

(12.5.2b)
(A− · L)|PQ LM (n1 , n2 )
= Q (P + 1)|PQLM (n1 , n2 )
. (12.5.2c)


When n1 = n2 , O(4) eigenfunctions of C22 and definite parity may, as
before, be obtained as linear combinations of the functions obtained by
interchanging the two quantum numbers n1 , n2 .
The functions D are determined in Ref. 30. The |PQ LM (n1 , n2 )
states
analogous to the |PQLM (n1 , n2 )
states of Sec. 12.5 are

|0000(2, 2)|1σ
= −0.500 1 S(2s2 ) + 0.866 1 S(2p2 ),

(12.5.3)
|2000(2, 2)|1σ
= −0.866 1 S(2s2 ) − 0.500 1 S(2p2 ).

The coupling process that produces these eigenfunctions of A2− + L2

and (A− · L)2 , and the coupling process of Biedenharn which produces
the eigenstates of A2 + L2 and (A · L)2 , begin with the same product of
one-particle SU(2) ⊗ SU(2) states:

|ja1 ma1 , jb1 mb1
|ja2 ma2 , jb2 mb2
.

For them
ja1 = (L1 + A1 )/2, jb1 = (L1 − A1 )/2,
(12.5.4)
ja2 = (L2 + A2 )/2, jb2 = (L2 − A2 )/2,
and

L1 = ja1 + jb1 , A1 = ja1 − jb1 ; L2 = ja2 + jb2 , A2 = ja2 − jb2 .

October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 373

Biedenharn’s result may be obtained by determining the couplings aris-

ing from:
i) adding ja1 and ja2 to obtain Ja+ = (L1 + L2 + A1 + A2 )/2 and deter-
mining the eigenfunctions of J2a+ ,
ii) adding jb1 and jb2 to obtain Jb+ = (L1 + L2 − A1 − A2 )/2 and to
determining the eigenfunctions of J2b+ ,
iii) adding Ja+ and Jb+ to obtain L = L1 + L2 and determining the eigen-
functions of L2 .
P = Ja+ + Jb+ and Q = Ja+ − Jb+ . (12.5.5)
The result in (12.5.2) is obtained by determining the couplings arising
from:
i) adding ja1 and jb2 to obtain Ja− = (L1 + L2 + A1 − A2 )/2 and deter-
mining the eigenfunctions of J2a− ,
ii) adding ja2 and jb1 to obtain Jb− = (L1 + L2 − A1 + A2 )/2 and deter-
mining the eigenfunctions of J2b− ,
iii) adding Ja− and Jb− to obtain L = L1 + L2 and determining the eigen-
functions of L2 .
Then P = Ja− + Jb− and Q = Ja− − Jb− . (12.5.6)
From the couplings (12.5.5) it follows that
2(J2a+ + J2b+ ) = A2 + L2 , J2a+ − J2b+ = A · L, (12.5.7a,b)
and those of (12.5.6) imply
2(J2a− + J2b− ) = A2− + L2 , J2a− − J2b− = A− · L. (12.5.7c,d)
The next several sections investigate the utility of these two couplings.

12.6 Configuration Mixing in Doubly Excited States

of Helium-like Atoms
If one uses degenerate perturbation theory to determine the effect of the
1/r12 repulsions between electrons in the initially degenerate states of the
previous two sections, one finds the configurations with definite l values,
1
S(2s2 ) and 1 S(2p2 ), mix to produce the following two 1 S wave functions:
Ψa = −0.476 1 S(2s2 ) + 0.880 1 S(2p2 ),
(12.6.1)
Ψb = −0.880 1 S(2s2 ) − 0.476 1 S(2p2 ).
These are approximations to wave functions of an excited helium atom
with two electrons in the n = 2 level. The overlap of these states with the
October 28, 2010 16:55 9in x 6in b943-ch12

374 Dynamical Symmetry

corresponding SO(4)12 states of Sec. 12.4 is, to three significant figures,

0.982.29 To the same accuracy, their overlap with the SO(4)1−2 states of
Sec. 12.5 is 1.000.
The states given in (12.6.1) are defined by the eigenvectors of the inter-
action matrix
 
S(2p2 ), r−1
12 S(2p 2
)
S(2p2
), r −1
12 S(2s 2
)

S(2s2 ), r−1 2
12 S(2p )
S(2s2 ), r−1 2
12 S(2s )

 √  (12.6.2)
1 111
√ −15 3
= .
512 −15 3 77
In the corresponding SO(4)1−2 matrix that defines the states in (12.5.3),
the lower right-hand entry, 77, is replaced by 81, with an increase of approx-
imately 5%. The remarkable agreement of the SO(4)1−2 configuration mix-
ing of degenerate two-electron states and the mixing caused by the electron
repulsion operator 1/r12 persists for a wide variety of two-electron n, n
states.30 It arises from the fact that in the regularized systems
r1 − r2 = (B1 − A1 ) − (B2 − A2 ). (12.6.3)
This causes in-shell interactions to produce approximate eigenstates of
|A1 − A2 |2 .31 To estimate energy changes produced by 1/r12 when one
is using the regularized system defined by W = 0, with
W = r1 r2 (p21 /2 + p22 /2 − Z/r1 − Z/r2 + 1/r12 − 2E), (12.6.4)
one may consider all wave functions to be built from Sturmian func-
tions. To simplify notation, let us for the moment denote the collection
of quantum numbers LM l1 , l2 , n1 , n2 , σ, by k, and write Ψk (r1 , r2 ) for
|LM (l1 , l2 , n1 , n2 )
|σ
. The matrix elements of 1/r12 are then

Ψ∗k (r1 , r2 ) Ψk
(r1 , r2 ) (r1 r2 /r12 )dr1 dr2 dω1 dω2 . (12.6.5)

These integrals vanish if k and k do not have identical L, M, and σ. They

are approximated in Ref. 31, and may be evaluated exactly with the aid of
formulas such as those given in Ref. 32.
Sinanoglu and Herrick found that the SO(4)1−2 coupling formula
(12.5.2) quite accurately predicts configuration mixing when n1 = n2 .33
The Bj terms in (12.6.3) contribute to this mixing.
In 1980, Herrick and Kellman pointed out that the first two couplings
outlined in (12.5.6) are couplings that determine internal motions which
have not been coupled to produce rotational states of definite angular
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 375

momentum. They made and developed the intriguing observation that

these internal motions are analogous to the internal vibrations in triatomic
molecules.34

12.7 Configuration Mixing Arising From Interactions

Within Valence Shells of Second and Third Row
Atoms
In the 2s-2p shell, the number of electrons, N, varies from 1 to 8. It turns
out that in every case where the in-shell 1/r12 interactions act on initially
degenerate 2sD 2pN−D configurations, the resulting states are very similar

to eigenstates of A2− ≡ i<j |Ai − Aj |2 . In every case, the overlap between
the corresponding states is 0.999.35 As in the two-electron case, a slight
difference in the configuration mixing arises because the actual 2s-2s diag-
onal element is slightly less than that required for eigenstates of A2− . As
the number of electrons increases, the difference decreases, becoming less
than 1% when N reaches 8. The coupling (12.4.1) produces eigenstates of
 2
i<j |Ai + Aj | which in all cases give poorer approximations to the actual
configuration mixing.29
The effects of Coulomb repulsions within the 3s-3p shell are very similar
to those within the 2s-2p shell. The overlap between the calculated states
produced by configuration mixing and the eigenstates of A2− varies between
0.987 and 0.997.36a
In-shell configuration interaction between 3s, 3p, 3d electrons is well
approximated by two-electron eigenstates of A2− , and three-electron states
if the orbital and spin angular momenta are fairly high, as in 4 S, 4 P, 4 F
or 2 D states. However when d electrons are involved in 2 S,2 P or 2 D states,
the overlap between the eigenstates of A2− and those of 1/r12 can become
as small as 0.657.36b

12.8 Origin of the Period-Doubling Displayed

in Periodic Charts
Contributions to a recent monograph, The Mathematics of the Periodic
Table, well depict current understanding of the manner in which dynam-
ical symmetries organize the properties of the chemical elements — and
therefore much of the natural sciences.37
Barut was the first to use the dynamical group SO(4,2) for this
purpose.38 He pointed out that Madelung’s rule, which describes the filling
in of the periods, reflects a property of an SO(3,1) subgroup of SO(4,2).
October 28, 2010 16:55 9in x 6in b943-ch12

376 Dynamical Symmetry

The rule states that in the neutral atoms in a given period, orbitals of
lowest n + l are filled first, and that orbitals having the same n + l value
are filled in order of increasing n.39 Unfortunately, as Ostrovsky notes, the
SO(3,1) subgroups do not contain the orbitals required to deal with period
doubling.42
Advances that follow Barut’s are described in Ref. 37 in contributions
from Novaro,40 Kibler,41 and Ostrovsky, who provides a very extensive list
of references to the earlier literature.42 These authors explore the relevance
of the groups SU(2) ⊗ SU(2) ⊗ SU(2), SO(4, 2) ⊗ SU(2), and SO(4, 2) ⊗
SU(2) ⊗ Z2 . (In each of these direct-product groups the SU(2) subgroup is
not due to electron spin).
Most periodic charts portray one period of length two, Period 1. Then,
Periods 2 and 3 both have the same length, eight; Periods 4 and 5, both
have length eighteen; Periods 6 and 7 both have length thirty-two.
Though SO(4) has unitary irreducible representations of the required
dimensions, 1, 4, 9, and 16, each of these representations occurs only once
in SO(4,2). Thus a definitive feature of the periodic chart can have no
explanation within SO(4,2). A dynamical explanation of the observed order
of filling of orbitals with nodal properties of hydrogen-like n, l, m orbitals
must deal with this fundamental problem. Each of the authors mentioned
above has approached it in a different way, but much remains to be done
to develop a definitive deduction of period doubling.
The discussions in previous chapters make it evident that a dynamical
explanation of period doubling will also provide a knowledge of the relevant
subgroup(s) of the dynamical group of n-electron atoms.
In this connection, a key group-theoretic aspect of period doubling
emerges from an investigation carried through by Kitigawara.43 He estab-
lished that the smallest special orthogonal groups with UIR of the required
dimensions are the rank three group SO(7) and its noncompact general-
izations SO(5,2), etc. No known single-particle system appears to have the
required dynamical symmetry.
Fortunately, Kitigawara’s analysis provides further insight into what is
required to derive period doubling from the dynamics — and hence dynam-
ical symmetries — encoded in many-electron Schrödinger equations. The
length of each period is a reflection of the fact that the energy gaps between
two successive periods is greater than the step-by-step energy changes that
develop as each periods fills. One might conjecture that in the simplest
of approximations the dynamical symmetry responsible for this could be a
degeneracy group which is a subgroup of SO(8,4). It could be that of systems
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 377

with a single outer electron moving in an approximate self-consistent field

potential provided by a nucleus surrounded by inner-shell electrons. The
inner shells must then have the effect of shifting the energy of an n−d state
to that of an n+1 s or p state, and must have the effect of shifting the energy
of an n − f state to that of an n + 2 s or p state. If η is the quantum number
required to give a common label to all the states of a period, then states
whose n, l values assign the same value to the expression n + l + δl,0 − 1
must have the same value of η. The appropriate UIR’s of the group must
contain only the wave functions whose n, l quantum numbers produce the
same value of η. The dimensions of these UIR’s must be η2 , so that when
spin is taken into account, the number of initially degenerate outer-electron
states with quantum number η will be 2η2 . The quantum number η must, in
all the periods, play much the same role as the principal quantum number
n does in Periods 1, 2 and 3. It should be able to label the Period, and one
should be able to express one-electron energies in the outer shell of neutral
atoms by
E = −ζ 2 /2η 2 , (12.8.1)
where the effective nuclear charge, ζ, is a function of η. Further, quantum
numbers must be able to label orbitals in such a way that those filled in
each period have the correct number of nodes. As the period fills, ζ can be
expected to become a function of these additional quantum numbers.
To summarize:
Though well developed inductions of possible symmetries respon-
sible for period doubling are available, there is not yet avail-
able a definitive deduction of period doubling from the relevant
Schrödinger equations.

12.9 Molecular Orbitals in Momentum-Space;

The Hyperspherical Basis
To make contact with current literature, in this and subsequent sections of
the chapter we will replace the quantum number k of Chapter 11, by p, and
use p = n −1 to replace the label n in both position-space and momentum-
space hydrogen-like wave functions. In the units we have been using the
operator of translations that carries r to r + R, that is, exp(R · ∇q ), can
be expressed as exp(iR · pop ). The corresponding operator that acts on
momentum space functions and their associated hyperspherical harmonics
is the operator of multiplication exp(ip · R). If, for example, Yplm (Ω), p =
October 28, 2010 16:55 9in x 6in b943-ch12

378 Dynamical Symmetry

n − 1, are hyperspherical harmonics corresponding to Sturmian position-

space functions, χnlm (r), and Schrödinger functions Ψnlm (r), then

exp(ip · R)Ynlm (Ω) (12.9.1a)

corresponds to

exp(−iR · pop )χnlm (r) = χnlm (r − R), (12.9.1b)

and

exp(−iR · pop )Ψnlm (r) = Ψnlm (r − R). (12.9.1c)

To determine the effect of the translation one may expand the product
exp(ip · R)Yplm (Ω) as a sum of hyperspherical harmonics, writing,44


exp(ip · R)Yplm (Ω) = Snlm
n
l
m
(p0 R) Yp
l
m
(Ω). (12.9.2a)

Then

Snlm
n
l
m
(p0 R) = Yp
l
m
(Ω)∗ exp(ip · R)Ynlm (Ω)dΩ. (12.9.2b)

If X is a vector from the origin to a point on Fock’s unit hypersphere, then

his equation governing the motion of an electron in the field of a nucleus of
charge Z is

p0 Ψ(Ω) = (Z/|X − X |2 ) Ψ(Ω ) dΩ . (12.9.3)

Translations convert this into an equation governing the momentum of an

electron moving in the electric field due to a set of nuclei of charge Zj at
positions Rj . The resulting generalization of Fock’s equation is44



p0 Ψ(Ω) = Zj |X − X |−2 exp(−i(p − p ) · Rj Ψ(Ω ) dΩ .
j
(12.9.4a)
It defines the equation

(P − p0 I)C = 0, (12.9.4b)

in which C is the vector of coefficients in the expansion

Ψ(Ω) = Cplm Yplm (Ω). (12.9.4c)

The matrix



P= Zj W(p0 Rj ) = Zj S(p0 Rj )−1 ΠS(p0 Rj ). (12.9.4d)
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 379

The elements of the matrix S(p0 Rj ) are Snlm

n
l
m
(p0 Rj ), and

S(p0 Rj )−1 = S(−p0 Rj ). (12.9.4e)

The matrix Π is diagonal with elements (1/n)δnlm,n
l
m
. Its inverse, N,
has elements n δnlm,n
l
m
, and
N j ≡ S(p0 Rj )N S(p0 Rj )−1 . (12.9.5a)
Thus
W(p0 Rj ) = N −1
j = (S(p0 Rj )(N 2 )S(p0 Rj )−1 )1/2 . (12.9.5b)
The analog of (12.9.4b) that corresponds to Schrödinger’s time-
dependent equation is
(P − I i∂/∂τ )C = 0. (12.9.6)
The expansions in the past few paragraphs can be translated into cor-
responding expansions in a Sturmian basis. Here we wish to call attention
to two properties that are common to both expansions:
(i) The operator P becomes that of a united-atom as the Rj approach
zero and its eigenvalues approach n. As the Rj increase from zero the
degeneracies associated with each value of n are broken, and thereafter
each wave function and energy undergoes adiabatic change until any
further crossings develop. One can think of these altered n values as
effective quantum numbers, and they enter into the Sturmian functions
as such.
(ii) In Eq. (12.9.5), the independently variable quantities in the matrix
elements of P are the components of the variables sj = p0 Rj . For each
set of chosen values of the sj , the equation determines the eigenvalue
p0 ; then one can determine the Rj . Though at first sight a nuisance,
this property has several important consequences:
(a) It introduces a scale factor that multiplies the radial argument r of
Sturmian basis functions in a manner which ensures that united-
atom and separated-atom hydrogenic levels are properly correlated.
(b) This helps enable limited basis sets to produce much better approx-
imate energies and potential energy curves than are produced by
small LCAO or Gaussian basis sets.45,46
(c) Near nuclear configurations that make p0 a degenerate eigenvalue,
the rank of the matrix P becomes unstable to small displacements.
This provides a method for uncovering crossings and degeneracies
at which new chemical and physical phenomena may develop.
October 28, 2010 16:55 9in x 6in b943-ch12

380 Dynamical Symmetry

12.10 The Sturmian Ansatz of Avery, Aquilanti and

Goscinski
If one neglects center of mass motions and regularizes the Schrödinger equa-
tion for a helium atom by multiplying it by the radial coordinates r1 , r2 , of
electrons number 1 and 2, one obtains an equation which may be written as
r2 {r1 (p21 /2 − E1 ) − Z} + r1 {r2 (p22 /2 − E2 ) − Z} + r1 r2 /r12 = 0, (12.10.1)
with
E1 + E2 = E.
If the r1 r2 /r12 term were not present, the solutions of the Schrödinger
equation would be a product of the scaled Sturmians that are solutions
of the single-particle Schrödinger equations whose regularized Hamiltoni-
ans are the operators contained in the curly brackets. The Sturmian ansatz
of Avery, Aquilanti and Goscinski may be thought of as a method, a very
clever one, for parceling out portions of the r1 r2 /r12 term between the curly
brackets. It is now a well developed system for approximating solutions to
the regularized Schrödinger equations of multi-electron systems of atomic
and molecular physics.
The following outline of the ansatz closely follows that of Avery and
uses his symbolism.45,46 He begins with the general Schrödinger equation
(−∆/2 + V(x) − E)ψ = 0, (12.10.2a,b)
in which
d


∆= ∂ 2 /∂x2j , x = (x1 , x2 , . . . , xd ), (12.10.2c)
1

and the total potential V(x) is the sum of the attractive potentials, V0 (x),
the total electrostatic repulsion potentials, and any other relevant poten-
tials.
Let us for the moment ignore the Pauli principle and spin, and expand
ψ as a linear combination of many-particle Sturmian functions


ψ= Bν Φν (x), (12.10.3)
in which the Φν are required to satisfy the equations
(−∆/2 + βν V0 (x) − E)Φν (x) = 0. (12.10.4)
Here βν is chosen in such a way that the energry E in these equation does
not depend on ν. The members of the set of solutions to Eq. (12.10.4)
all correspond to the same energy. This energy is chosen according to the
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 381

state that is being represented. It can be shown that the set of isoenergetic
solutions to (12.10.4) obeys a potential-weighted orthogonality relation.
It is convenient to normalize the configurations in such a way that these
orthogonality relations take the form

dx Φ∗ν
(x) V0 (x) Φν (x) = (2E/βν )δν,ν
, (12.10.5)

where dx represents the d-dimensional volume element of the system. Equa-

tions (12.10.4) and (12.10.5) assign a portion of the attractive poten-
tial to each φν (x) in a manner which, when supplemented with side
conditions given below, produces energies and wave functions that closely
approximate self-consistent field results. Taken together, (12.10.4) and
(12.10.5) require that



(−∆/2 + V(x) − E) Bν Φν (x) = (−βν V0 (x) + V(X)) Bν Φν (x) = 0.
ν ν
(12.10.6)
Multiplying these relations from the left by Φ∗ν
(x)
and using (12.10.5) one
obtains the simple set of equations


 
dx Φ∗ν
(x) V(x)Φν (x) − 2Eδν
ν , Bν = 0. (12.10.7)
ν

Next, suppose that the functions Φν (x) are products of single particle
Sturmians φη (x), η = nlm, and spin functions σ± , and then write
χµ (x1 ) = φη (x)σ± , µ = η ± 1/2. (12.10.8a)
Then let
φν (x) = χµ (x1 ) χµ
(x1 ) χµ

(x1 ) . . . (12.10.8b)
and require:
(A) The functions χµ (xj ) are solutions of the equations
(−∆j + ku2 − 2nk u /rj ) χµ (xj ) = 0 (12.10.9a)
and
(B)

d3 xj χµ
(xj )∗ (1/rj ) χµ (xj ) = (kµ /n)δµ,µ
, (12.10.9b)

(C)

d3 xj |χµ (xj )|2 = 1. (12.10.9c)

The ku are required to be related by the constraint

October 28, 2010 16:55 9in x 6in b943-ch12

382 Dynamical Symmetry

(D)

kµ2 + kµ2
+ kµ2

+ · · · = p20 = −2E (12.10.9d)

and a further constraint, which for an atom with nuclear charge Z, is

(E)

nµ kµ = nµ
kµ
, = · · · = Zβν . (12.10.9e)

Equations (A) and (D) taken together assign to the argument of each χµ (xj )
a scale factor kµ , and hence eigenvalues

poµ = kµ = (−2Eµ )1/2 . (12.10.9f)

Then equation E requires

kµ = Zβν /nµ . (12.10.9g)

Thus Zβν is an effective nuclear charge that becomes assigned to the Stur-
mian with single-particle energy

−(1/2)kµ2 = −(1/2)(Zβν )2 /nµ2 . (12.10.10)

With the substitution given by (12.10.9g) the orbitals χµ become solutions

to the one-electron hydrogen-like wave equation for an atom with effective
nuclear charge Z βν . All the orbitals belonging to the configuration Φν have
the same effective nuclear charge.
Equation C defines a normalization of the scaled Sturmian functions
that is obtained by multiplying the Sturmians of Table 11.3.1 by a factor
3/2
of kµ after their radial argument has been multiplied by kµ . Thus, for
example, the normalized radial function for a 2p function is the Sturmian
3/2
kµ 2(kµ r) exp(−kµ r).
Now let us return to Eq. (12.10.7), that is to the equations


 
dxΦ∗ν
(x) V(x) Φν (x) − 2Eδν
ν Bν = 0.
ν

Let

V(x) = V0 (x) + V1 (x), (12.10.11a)

with V0 (x) being the sum of nuclear–electronic attractive potentials, and

V1 (x) = 1/rij . (12.10.11b)
i<j
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 383

Next define

dx Φ∗ν
(x) V0 (x) Φν (x) = −p0 T0ν
ν , (12.10.11c)

dx Φ∗ν
(x) V1 (x) Φν (x) = −p0 T1ν
ν . (12.10.11d)

The T1ν
ν are negative dimensionless quantities and are independent of p0 .

The T0ν
ν are positive, also dimensionless, and in the atomic case, indepen-
dent of p0 .
In the atomic case, conditions (12.10.9a–d) imply that

Tν
ν = −(1/p0 ) dx Φ∗ν
(x) V0 (x) Φν (x) = (−2E/p0 βν )δν
ν
0

= (p0 /βν )δν
ν = (1/βν )( ku2 )1/2 δν
ν . (12.10.12)
If there are Nj electrons with principle quantum number nj , it becomes
convenient to define

1/2
Rν = Nj /nj2 , (12.10.13)
a function whose value only changes when orbital occupancies change. Then
(12.10.9e) implies
T0ν
ν = ZRν δν
ν . (12.10.14)
Inserting (12.10.11b) into (12.10.7), and dividing through by p0 produces a
set of secular equations. If one lets T0 = T0 I be the diagonal matrix with
elements ZRν , and T1 be the matrix with elements T1ν
ν , then the secular
equations take on the matrix-vector form:
(T1 − (p0 − T0 )I)B = 0. (12.10.15)
Some of the key consequences of Eqs. (12.10.9) are revealed when the
methodology is applied to neutral several-electron atoms while configura-
tion interactions are neglected. If the many-electron states obtained by
solving (12.10.15) are labeled by an index κ, and their momenta p0 are
labeled p0κ , then
p0κ = ZRν − T1κκ , (12.10.16a)
and
Eκ = −(1/2)(ZRν − T1κκ )2 . (12.10.16b)
For the ground states of neutral atoms with 2 to 9 electrons, as Z runs from
2 through 9, T1κκ changes from 0.441942 to 2.41491, an average increase
October 28, 2010 16:55 9in x 6in b943-ch12

384 Dynamical Symmetry

of ∼0.281/electron. For Z > 3, the calculated energies are very close to

the Hartree–Fock self-consistent field values.45,46 However, with marked
contrast to SCF calculations, the orbitals produced have a simpler form:
they are all one-electron Sturmians. Avery and Avery have used this fact
and a 1/Z expansion to neatly isolate and study the large configuration
mixing effects that arise from the interaction of degenerate one-electron
functions the same principal quantum number.47
When configuration interactions are included, the resulting many- par-
ticle functions can still be considered to be Sturmians whose momentum-
space analogs correspond to functions defined on hyperspheres in spaces
of dimension greater than 4. Avery, Aquilanti and others have exten-
sively developed the connection between hyperspherical bases defined on n-
dimensional hyperspheres, and corresponding multiparticle Sturmians.12,46
Correlations between particle motions are introduced by both the radial and
angular functions that multiply these terms. In the hyperspherical bases,
these correlations are all expressed as angular correlations.
The theory has been developed and extensively applied when V(x) is the
potential function appropriate to many-electron atoms.12 It has been used
to turn the treatment in Sec. 12.9 into that of many-electron molecules.46
It has also been extended to relativistic systems.12
In concluding this section, we would call attention to the fact that
Eq. (12.10.16), and the previous deductions, require the development of an
entirely new analysis of effective nuclear charges for regularized systems.
Exercises
1. Draw a united-atom versus separated-atom correlation diagram for the
levels of H+2 that includes united-atom levels through n = 3. Identify
the “forbidden crossings”.
2. Express the integral (12.2.7) as a function of n, and use the result to
determine the dependency of (12.2.2) on all relevant quantum numbers.
Using these, determine the dependency of the wavenumber of Stark effect
lines of hydrogen upon these quantum numbers (cf. Ref. 4).
3. Use Sturmians for the ground state of He to exemplify Eq. (12.10.2)
through (12.10.14). Set Λ = 0.4419 and calculate the contribution of
1/r12
to p0 and E.
4. Use (12.9.1) to express the hyperspherical harmonic analog of the bond-
ing molecular orbital of H+ 2 . Do the same for the corresponding anti-
bonding orbital.
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 385

5. It has long been known that, in solutions, benzoquinone can exhibit

a dipole moment — despite the fact that its molecules have a center
of symmetry. Use extended Huckel or similar calculations to develop a
correlation diagram for the energy levels of its hydrocarbon analog as
the molecule undergoes a B3 u out-of-plane bending. (The corresponding
mode in benzoquinone has a very low frequency, ∼ 100 cm−1 ). Identify
crossings involving the highest occupied π level, if any, and estimate the
direction of polarization that might result. Carry out analogous calcula-
tions using an SCF or a better program.

References
[1] J. J. Balmer, Ann. Phys. 25 (1885) 80.
[2] J. V. Fraunhofer, Encylopedia Britannica, 11th edn. (Cambridge, 1910).
[3] N. Bohr, Phil. Mag. 26 (1913) 1.
[4] Cf. A. Sommerfeld, Atomic Structure and Spectral Lines, 3rd German edn.,
translated by H. L. Brose (Methuen, London, 1923).
[5] E. U. Condon and G. H. Shortley, The Theory of Atomic Spectra (Cambridge
University Press, 1935, 1951–1967).
[6] (a) C. Eckart, Rev. Mod. Phys. 2 (1930) 305;
(b) E. Wigner, Gruppentheorie (Viewig, 1931);
(c) E. Wigner, Group Theory and its Application to the Quantum Mechanics
of Atomic Spectra, (Academic Press, N.Y., 1959).
[7] G. Racah, Phys. Rev. 61 (1942) 186; Phys. Rev. 62 (1942) 438; Phys. Rev.
63 (1943) 367; Phys. Rev. 76 (1949) 1352.
[8] A. R. Edmonds, Angular Momentum in Quantum Mechanics (Princeton,
1957).
[9] (a) L. C. Biedenharn and J. D. Louck, Angular Momentum in Quantum
Physics (Addison Wesley, Reading, Mass, 1981);
(b) L. C. Biedenharn and J. D. Louck, The Racah-Wigner Algebra in Quan-
tum Theory (Addison Wesley, Reading, Mass, 1981).
[10] B. G. Wybourne, Symmetry Principles and Atomic Spectroscopy (Wiley-
Interscience, N.Y., 1970).
[11] B. R. Judd, Atomic Shell Theory Recast, Phys. Rev. 162 (1967) 28.
[12] J. Avery, Generalized Sturmians and Atomic Spectra, (World Scientific, Sin-
gapore, 2006).
[13] P. J. Redmond, Phys. Rev. 133 (1964) B1352.
[14] H. A. Erikson and E. L. Hill, Phys. Rev. 75 (1949) 29.
[15] J. von Neumann and E. Wigner, Z. Phys. 30 (1929) 467.
[16] L. I. Ponomarev and T. P. Puzynina, Soviet Physics JETP 25 (1967) 846.
[17] C. E. Wulfman and S. Kumei, Science 172 (1971) 1061.
[18] L. Salem, Isr. J. Chem. 12 (1979) 8.
October 28, 2010 16:55 9in x 6in b943-ch12

386 Dynamical Symmetry

[19] Proc. Indian Acad. Sci. (Chem. Sci.) 107(6) (1995).

[20] H. A. Jahn and E. Teller, Proc. Roy. Soc. (London) A161 (1937) 220.
[21] A. D. Walsh, Disc. Faraday Soc., Series 2, (1947) 18; J. Chem. Soc. (1953),
2260, 2266, 2278, 2296, 2300, 2306, 2318, 2325, 2330.
[22] C. E. Wulfman, J. Chem. Phys. 31 (1959) 381.
[23] C. Wulfman, Memoria do Primeiro Congresso Latinamericano De Fisica,
Vol. 23 (Estadio pr la Sociedad Mexicana de Fisica Mexico, D. F, 1969),
p. 105.
[24] (a) S. Pasternack and R. M. Sternheimer, J. Math. Phys. 3 (1962) 1280;
(b) N. V. V. J. Swamy, R. G. Kulkarni and L. C. Biedenharn, J. Math. Phys.
11 (1970) 1165.
[25] (a) R. F. W. Bader, Can. J. Chem. 40 (1962) 1164;
(b) R. G. Pearson, J. Am. Chem. Soc. 91 (1969) 1252; 91 (1969) 4947;
(c) R. G. Pearson, J. Chem. Phys. 52 (1970) 2167.
[26] L. C. Biedenharn, J. Math. Phys. 2 (1961) 433.
[27] E. U. Condon and G. H. Shortley, The Theory of Atomic Spectra (Cambridge
University Press, 1967), p. 48.
[28] B. G. Wybourne, Lie Groups for Physicists (Wiley-Interscience, New York,
1974), pp. 236–240.
[29] E. Chacon, et al., Phys. Rev. A 3 (1971) 166.
[30] C. Wulfman, Chem. Phys. Lett. 23 (1973) 370.
[31] C. Wulfman and S. Kumei, Chem. Phys. Lett. 23 (1973) 367.
[32] J. Avery, Hyperspherical Harmonics and Generalized Sturmians (Kluwer,
Dordrecht, 2000), p. 14.
[33] (a) O. Sinanoglu and D. Herrick, J. Chem. Phys. 62 (1975) 886;
(b) O. Sinanoglu and D. Herrick, J. Chem. Phys. 65 (1976) 850.
[34] D. Herrick and M. Kellman, Phys. Rev. A 21 (1980) 418.
[35] C. Wulfman, Phys. Rev. Lett. 51 (1983) 1159.
[36] (a) Y. Ho and C. Wulfman, Chem. Phys. Lett. 103 (1983) 35.
(b) Y. Ho and M.Sc. Thesis, Dept. of Phys., University of the Pacific, Stock-
ton, CA. (1985).
[37] D. H. Rouvray and R. B. King, eds., The Mathematics of the Periodic Table
(Nova Science Pub., N. Y., 2006).
[38] A. O. Barut, in The Structure of Matter, Proceedings of the Rutherford Cen-
tennial Symposium, edn. B. G. Wybourne (Christchurch, N. Z., 1972), p. 126.
[39] E. Madelung, Math. Hilfsmittel des Physikers, 6th edn. (Springer, Berlin,
1950), p. 611.
[40] O. Novaro, in The Mathematics of the Periodic Table (2006), p. 217.
[41] M. R. Kibler, ibid., p. 237.
[42] V. N. Ostrovsky, ibid., p. 265.
[43] Y. Kitigawara, M.Sc. Thesis, Dept. of Phys., University of the Pacific, Stock-
ton, CA (1977); J. C. Donini (ed.), Recent Advances in Group Theory and
Their Application to Spectroscopy, (Plenum Press, New York, 1978).
[44] T. Shibuya and C. E. Wulfman, Proc. Roy. Soc. A 286 (1965) 376.
October 28, 2010 16:55 9in x 6in b943-ch12

Uncovering Approximate Dynamical Symmetries 387

[45] J. Avery, Hyperspherical Harmonics and Generalized Sturmians (Kluwer,

Dordrecht, 2000), Ch. 1.
[46] Cf. J. Avery, Hyperspherical Harmonics (Kluwer, Dordrecht, 1989), and the
extensive list of references contained in it, and in Ref. 45.
[47] J. Avery and J. Avery, Generalized Sturmians and Atomic Spectra (World
Scientific, Singapore, 2006).
This page is intentionally left blank
October 28, 2010 16:55 9in x 6in b943-ch13

CHAPTER 13

Rovibronic Systems

13.1 Introduction
Lie-algebraic methods were introduced into the study of the spectra of
rotating and vibrating molecules by physicists who had contibuted to
the development of the liquid drop model of nuclei and its connections
to models composed of discrete nucleons. Their early success in nuclear
physics led to several thoroughly developed mathematical models of rovi-
brational motion: the harmonic oscillator model, the symplectic model and
the vibron model. For information about the first two, the reader is referred
to reviews by Moshinsky and Rowe.1,2 A U(6) vibron model was intro-
duced in nuclear physics by Arima and Iachello,3 and Iachello introduced
a U(4) vibron model into molecular physics.4 They have been applied to
diatomics,5 triatomics,6 tetra-atomics,7 and to molecules at least as com-
plex as benzene.8 Several monographs have been devoted to applications of
the model in nuclear physics and molecular physics.9–11
The vibron models are semi-empirical ones, and a major criticism leveled
against their application to molecular physics has been that there does not
appear to be a definitive connection between their Lie-algebraic structure
and that associated with Schrödinger equations with definite potentials.
Nevertheless, they represent an outstanding example of the inductive use
of the mathematics of Lie algebras in the sciences. We will concentrate
attention on the connection they establish between molecular spectroscopy
and Lie algebras.

13.2 Algebraic Treatment of Anharmonic Oscillators

With a Finite Number of Bound States
The Morse oscillator was one of the systems with both a discrete and a
continuous spectrum investigated in Chapter 10. It was found to have a
non-compact three-parameter dynamical group, SU(1,1).

389
October 28, 2010 16:55 9in x 6in b943-ch13

390 Dynamical Symmetry

Here we wish to deal only with bound states of anharmonic oscilla-

tors. The simplest way is to compactify the noncompact groups that are
the dynamical groups of these systems with both continuous and discrete
spectra. Compactifying the Lie algebras of the dynamical groups, SO(2,1)
of SU(1,1), of one-dimensional systems, produces the semi-simple A1 Lie
algebra of the groups SU(2) and SO(3). It provides a useful spectrum-
generating algebra for the Morse oscillator.12 Its potential energy function
can be expressed as
V = De (1 − exp(−a(r − re ))2 , (13.2.1)
where re is the value of the internuclear distance, r, at which the poten-
tial takes on its minimum value, De . This function is plotted in
Fig. 13.2.1a. The corresponding total energy surface, E(r,p), is depicted
in Fig. 13.2.1b.
Let three Lie generators Sk , of SU(2) and SO(3), obey the commutation
relations
[S1 , S2 ] = iS3 , [S2 , S3 ] = iS1 , [S3 , S1 ] = iS2 . (13.2.2)
In this notation, the shift operators for the labeling operator S3 are
S+ = S1 + iS2 , S− = S1 − iS2 . (13.2.3)
The mathematical eigenstates of S3 may be denoted |jm
. They satisfy the
relations
S+ |jm
= (j(j + 1) − m(m + 1))1/2 |jm + 1
,
(13.2.4)
S− |jm
= (j(j + 1) − m(m − 1)))1/2 |jm − 1
.
When the resulting Lie algebra and Lie group are those of the bound states
of an oscillator, it is assumed that only the absolute value of m is a physical
observable.
As the mathematical states |jm
and |j − m
correspond to the same
physical state, it is the value of |m| that is physically relevant. As
S− |j, −j
= 0, S+ |j, j
= 0, (13.2.5a)
one has
|m| = 1/2, 3/2, . . . , j (13.2.5b)
or
|m| = 0, 1, 2, 3, . . . , j. (13.2.5c)
In the first case the oscillator has 2j−1 energy levels and the group is SU(2).
In the second case it has j + 1 levels and the group is SO(3). However, as
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 391

(a)

(b)

Fig. 13.2.1 (a) Morse potential function; (b) Morse energy surface.

seen in Sec. 10.7, for the Morse oscillator, the m = 0 case is excluded, so in
the second case there are j energy levels.
To make contact with the vibron model which originated in nuclear
shell theory, one expresses the group generators as binary products of
boson creation and annihilation operators b†1 b1 , b†2 b2 , b†1 b2 , and b†2 b1 .
October 28, 2010 16:55 9in x 6in b943-ch13

392 Dynamical Symmetry

These operators are mathematically characterized by the commutation

relations
[bα , b†β ] = δαβ . (13.2.6)
These are relations obeyed by operators that create and annihilate particles
that have the interchange symmetry of bosons. The compound operators
b†j bk generate unitary groups and special orthogonal groups. For SU(2) and
SO(3), letting S → S
one has
S
1 = (b†1 b2 + b†2 b1 )/2, S
2 = −i(b†1 b2 − b†2 b1 )/2, (13.2.7a,b)

S
3 = (b†1 b1 − b†2 b2 ), S
+ = b†1 b2 , S− = b†2 b1 (13.2.7c–e)

The algebra of SU(2) becomes that of U(2) if one adds to these the further
generator
Nop = b†1 b1 + b†2 b2 . (13.2.7f)
In the Morse oscillator realization of the Lie algebra, one can establish
the dependence of the boson creation and anhillation operators upon the
dynamical variables q,t and the operators −i∂/∂q, i∂/∂t. To do so we
slightly modify a mathematical development due to Iachello and Levine.13
To simplify matters for the reader we first observe that the notation of Sec.
10.7, is related to that of Iachello and Levine by the changes d → V0 , a →
β, m → µ.
If one sets ξ = β(r − r0 ), the Schrödinger equation with Morse potential
V0 (exp(−2ξ) − 2 exp(−ξ)) and reduced mass µ can be written as
{−∂ 2 /∂ξ 2 + (2µ/(
β)2 )V0 (exp(−2ξ) − 2 exp(−ξ))}ΨE = (2µ/(
β)2 )EΨE .
(13.2.8a)
2 −1
If one defines a unit-free energy, ε = (2µ/(
β) ) , the results of Morse
require that
E = −ε(M/2)2 , M/2 = n + 1/2 − k, k = (
β)−1 (2µV0 )1/2 , (13.2.8b)
with M /2 being an integer. The relationship of this expression to observed
spectral lines associated with photons of frequencies ω is supplied by com-
paring it to the standard Dunham expansion,14 which can be put in the
form
ω = ωe (ν + 1/2) − ωe χe (ν + 1/2)2 + . . . . (13.2.8c)
After a time dilation similar to that introduced in Sec. 10.7, one can
write
ΨE = ΦE (ξ) exp(iM t). (13.2.8d)
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 393

Letting N be an eigenvalue of Nop , and setting

(2µ/(
β)2 )V0 )1/2 = (N + 1)/2, and ΦE (ξ) = RN,M (ξ), (13.2.9a)
converts (13.2.8a) to the following equation:
{−∂ 2 /∂ξ 2 + ((N + 1)/2)2 (exp(−2ξ) − 2 exp(−ξ)) + (M/2)2 }
× RN,M (ξ) exp(iM t) = 0. (13.2.9b)
To express this in terms of the boson operators one may define
x
= ρ cos(t), y
= ρ sin(t) (13.2.9c)
and, a la Dirac, define the shift operators
√ √
b1 = (x
+ ∂/∂x
)/ 2, b†1 = (x
− ∂/∂x
)/ 2,
√ √ (13.2.9d–g)
b2 = (y
+ ∂/∂y
)/ 2, b†2 = (y
− ∂/∂y
)/ 2.
On setting
exp(−ξ) = ρ2 /(N + 1), (13.2.9h)
and letting
RN,M (ξ) exp(iM t) → |N, M
, (13.2.9i)
the result of Iachello and Levine establishes that the Morse Schrödinger
equation is equivalent to the relations
Nop |N, M
= N |N, M
,
(13.2.10a,b)
S23 |N, M
= M 2 |N, M
.
These two last equations may be considered to be the defining equations
of an abstract one-dimensional vibron — an oscillator with U(2) dynamical
symmetry that is vibrating in one spatial dimension and possesses only a
finite number of bound states. In this particular case the relations (13.2.8)
and (13.2.9) define the vibron to be a Morse oscillator.
An unlimited number of realizations of the vibron algebra with spec-
trum (13.2.4) can be obtained by unitary transformations of the Morse
generators and Hamiltonian, yielding operators S
= TST−1 . These will
not alter the abstract group and they will not alter the spectrum of the
oscillator, but can alter transition probabilities. The spectrum itself can be
most easily altered by changing the parameters that determine N and D0 ,
and by adding functions of S
3 to the Hamiltonian. A vibron Hamiltonian
H = F(S
3 ) will have eigenstates T |jm
with energy E = F(m).
October 28, 2010 16:55 9in x 6in b943-ch13

394 Dynamical Symmetry

13.3 U(2) ⊗ U(2) Model of Vibron Coupling

In this model, a pair of non-rotating, initially independent, vibron oscilla-
tors are coupled.15 The pair of uncoupled oscillators has as its spectrum
generating group U(2)a ⊗ U(2)b . The U(2) groups are as defined in the pre-
vious section. This two-vibron group contains several subgroup chains, of
which we list the following three:

(i) U(1)ab ⊗ SU(2)a ⊗ SU(2)b ⊃ U(1)ab ⊗ SU(2)ab ⊃ SO(2)ab ,

(ii) U(2)ab ⊃ U(2)ab ⊃ SU(2)ab ⊃ SO(2)ab , (13.3.1)
(iii) SO(2)a ⊗ SO(2)b ⊃ SO(2)ab .

The generator of U(1)ab is

Nop ≡ Nopa + Nopb = b†1a b1a + b†2a b2a + b†1b b1b + b†2b b2b , (13.3.2a)

and the generators of SU(2)ab are the three corresponding operators

Sjab = Sja + Sjb . (13.3.2b)
The group SO(2)ab is usually chosen to have Sjab as its generator.
The coupling of the direct product SU(2)a ⊗SU(2)b produces the second-
order Casimir invariant
C2ab = (Sa + Sb ) · (Sa + Sb ) = S2a + S2b + 2Sa · Sb . (13.3.3)
It can be used to impose a correlation of the motions of the two vibrons.
The coupled states are those defined by (10.11.6), viz.
|JM (ja , jb )
= ΣC(ja jb JM ; ma mb )|ja ma
|jb mb
. (13.3.4)
They are eigenstates of Sa ·Sb with eigenvalue {J(J +1)−ja (ja +1)−ja (ja +
1)}/2. In the vibron model, Sa ·Sb couples the normal modes of two vibrons.
Using the notation of the previous section, all two-vibron Hamiltonians of
the form


H = F(C1,a , C
1,b , C
2,a , C
2,b , C
2ab , J
ab , S
3a , S
3b ), (13.3.5a)
will have eigenstates that can be linear combinations of T (|ja ma
|jb mb
)
and T|JM (ja , jb )
, with eigenvalues
E = F(ja , jb , ja (ja + 1), jb (jb + 1), J(J + 1), M, ma , mb ). (13.3.5b)
Strong couplings of normal modes take place when the modes are
degenerate, or very nearly so. As this specificity exerts the greatest strain
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 395

on coupled U(2) ⊗ U(2) vibron models, it is worth considering them in

some detail. A seminal study of the bond vibrations of Y–X–Y molecules
by van Roosmalen, Benjamin and Levine did just this.16 They found
the most relevant group chains to be (i) and (iii) The states of chain
(i) are the T|JM (ja , jb )
and those of chain (iii) are local mode states
T(|ja ma
|jb mb
). For their most general vibron Hamiltonian with SO(2)ab
symmetry the authors choose one with three parameters, A, B and λ, and of
the form
H = E0 + A(S23a + S23b ) + B(S3a + S3b )2 + λ(Na Nb /2 − 2Sa · Sb ). (13.3.6a)
Its eigenvalues are
E0 +A(m2a +m2b )+B(ma +mb )2 +λ(ja (ja +1)+jb (jb +1)+2ja jb −J(J +1)).
(13.3.6b)
H commutes with an operator, Π, which is the sum of the two operators
ΠJ = (b†1j b†1j − b†2j b†2j )(b1j b1j − b2j b2j ). (13.3.7)
Here, the parameters are determined from the vibrational spectra of the
system.
The model was tested using least-squares fits to the vibrational spectra
of H2 O, O3 and SO2 to determine the values of A, B and λ for each molecule.
The integer, N, was set to yield the correct number of bound vibrational
states of the molecule. The root mean square deviations from vibron ener-
gies were found to be 3.99 cm−1 , 23.7 cm−1 and 1.23 cm−1 , respectively.
The model was also applied to the C–H and C–D bond stretching vibrations
of acetylene and deuteroacetylene, for which root mean square deviations
were found to be 3.91 cm−1 and 4.85 cm−1 . These results show that the
model is able to predict vibrational energies with accuracies better than
1%, even in molecules with almost degenerate vibrational modes.
The comparisons with experimental data noted above turn out to be
similar to those obtained in a number of subsequent investigations of n-
vibron models of vibrations in polyatomic molecules with n modes of
vibration. It has even proved possible to include terms in the vibron Hamil-
tonian that deal properly with Fermi resonances17 and overtone vibrations
in linear18 and bent19 triatomics. To obtain results of such accuracy the
model usually requires fewer parameters than are required in the previ-
ously standard Dunham expansions. This implies that the vibron models
express group structure that must be very similar to that of the dynamical
groups of the physical systems.
October 28, 2010 16:55 9in x 6in b943-ch13

396 Dynamical Symmetry

13.4 Spectrum Generating Groups of Rigid Body Rotations

Edmond’s description and notation for Euler’s angular coordinates α, β and
γ will be used to describe the orientation of the principal axes of inertia of
a rigid body with respect to a space-fixed axes.20 Setting x = cos(β), the
Hamiltonian of a symmetric top is
(−1/2I1 ){(1 − x2 )∂ 2 /∂x2 − 2x∂/∂x + (I1 /I3 + x2 (1 − x2 )−1 )∂ 2 /∂γ 2
− 2x(1 − x2 )−1 ∂ 2 /∂α∂γ}.
The solutions of
(H − i∂/∂t)Ψ = 0 (13.4.1)
may be expressed in terms of Wigner’s D functions.20 We set


Ψ= Cjmn Ψjmn (α, β, γ) exp(−iEjn t) (13.4.2a)
with
Ψjmn (α, β, γ) = (2j + 1)1/2 Djmn (α, β, γ)/4π, (13.4.2b)
and
Ejn = j(j + 1)/2I1 + n2 (I1 − I3 )/2I1 I3 , (13.4.2c)
with
j = 0, 1/2, 1, 3/2, . . . , |m| < j, |n| < j. (13.4.2d)
The ψjmn (α, β, γ) are normalized D functions, and are orthonormal under
the scalar product
 2π  2π  2π
(f, g) = dα sin(β)dβ dγ f ∗ g. (13.4.3)
−2π 0 0

Bopp and Haag have proposed that the D functions with half-integer j
values could provide spatial realizations of the wave functions of particles
with half-integer spin.21
To determine the dynamical group of (13.4.1), Kumei22a began by using
Dop = exp{t∂/∂t ln((Jop + 1/2)/H)} (13.4.4a)
to linearize the spectrum of j. In (13.4.4a),
Jop + 1/2 ≡ (1/2){1 + 8I1 [H + ((I1 − I3 )/2I1 I3 )∂ 2 /∂γ 2 ]}1/2 . (13.4.4b)
The dilatation operator Dop converts (13.4.1) to
{(1 − x2 )∂ 2 /∂x2 − 2x∂/∂x + (1 − x2 )−1 (∂ 2 /∂α2 + ∂ 2 /∂γ 2 )
−2x(1 − x2 )−1 )∂ 2 /∂α∂γ + (i∂/∂t)2 − 1/4}Φ = 0.
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 397

with


Φ= Cjmn φjmn , φjmn = Ψjmn (α, β, γ) exp(−i(j + 1/2)t).
(13.4.5a,b)
Setting up and solving the 14 determining equations required to obtain
first- and second-order differential operators yields 16 generators Q
of the
invariance group of (13.4.5). One of these is the generator iΦ∂/∂Φ of the


group that can change the phase of Φ. The other 15 were denoted Qjk . They
obey the SL(4,R) commutation relations





[Q ij , Q k
m]
i
= δm Q k
j − δjk Q i
m. (13.4.6)
Here we summarize some of the general properties of this dynamical group.
For a fuller discussion the reader is referred to the original paper.22b


Converting the Q kj to skew-adjoint or self-adjoint generators of invari-
ance transformations of the original Schrödinger equation produces the gen-
erators of its dynamical group. The action of the resulting labeling operators
−i∂/∂α, −i∂/∂γ and i∂/∂t, on the eigenfunctions
φjmn = Ψjmn (α, β, γ) exp(−i(j + 1/2)t)
(13.4.7)
= ((2j + 1)1/2 /4π)Djmn (α, β, γ) exp(−i(j + 1/2)
is given by
−i∂/∂α φjmn= mφjmn , −i∂/∂γ φjmn= n φjmn ,
i∂/∂t φjmn= (j + 1/2)φjmn .
Kumei found that all of the generators which shift j shift it by 1/2 unit.
They must consequently shift both m and n by 1/2 unit. The Lie algebra is
consequently that of the group SU(2,2). The group contains operators that
mix states with integral and half-integral values of j. The physical interpre-
tation of these mixed states present problems that have been addressed by
Bohm and Teese.23 Double-jump operators for j may be constructed from
products of the single-jump operators. However they do not close under
commutation with the other SL(4,R) generators.
We next consider the degeneracy groups of the systems. When all three
moments of inertia are equal, the resulting spherical top has SU(2) ⊗ SU(2)
as its degeneracy group. In the notation of Ref. 22 the shift and labeling
operators of the group are
Q9 = Q†10 = exp(iα){(1 − x2 )1/2 ∂/∂x − ix(1 − x2 )−1/2 ∂/∂α
+ i(1 − x2 )−1/2 ∂/∂γ}, (13.4.8a)
Q11 = Q†12 2 1/2 2 −1/2
= exp(iγ){(1 − x ) ∂/∂x − ix(1 − x ) ∂/∂γ
+ i(1 − x2 )−1/2 ∂/∂α}. (13.4.8b)
October 28, 2010 16:55 9in x 6in b943-ch13

398 Dynamical Symmetry

These generators have the following action on the φjmn :

Q9 φjmn = (j(j + 1) − m(m + 1))1/2 φjm+1n ,
Q10 φjmn = −(j(j + 1) − m(m − 1))1/2 φjm−1n ,
(13.4.9a–d)
Q11 φjmn = −(j(j + 1) − n(n + 1))1/2 φjmn+1 ,
Q12 φjmn = (j(j + 1) − n(n − 1))1/2 φjmn−1 .
For the symmetric top with I1 = I2 = I3 the generators Q11 and Q12
become time-dependent, but remain invariance generators of Eq. (13.4.1).
The degeneracy group is reduced to SU(2) ⊗ U(1).
This change in the degeneracy group is an example of a general
phenomenon with interesting consequences.24 Suppose that Ψa (q,t) is a
general solution of the time-dependent Schrödinger equation with Hamil-
tonian Ha , and that Ψb (q,t) is a general solution of the time-dependent
Schrödinger equation with Hamiltonian Hb . Next, expand these Ψ(q,t) as
a sum of solutions of their time-dependent eigenstates, Ψan (q) exp(−iEan t)
and Ψbn
(q) exp(−iEbn t), respectively. Now, suppose also that all the func-
tions Ψan (q) and Ψbn
(q) are contained in the same Hilbert space of square-
integrable functions f(q). Then, if Qa is an operator satisfying the equation
[(Ha − i∂/∂t), Qa ]Ψa (q, t)| = 0, (13.4.10a)
and if
Sab = exp(−i(Hb − Ha )t), (13.4.10b)
is well defined,24 then
Qab = Sab Qa S−1
ab , (13.4.10c)
will satisfy
[(Hb − i∂/∂t), Qab ]Ψb (q, t)| = 0. (13.4.10d)
When Qa is a time-independent constant of motion of Ha , the operator
Qab will be a time-dependent constant of motion if Qa does not commute
with Hb . This is the case when a spherical top is converted to a symmetric
top. One then has
Hb − Ha = (−1/2)(I2 − I1 )∂ 2 /∂γ 2 . (13.4.11a)
Thus
Sab = exp(i(t/2))(I2 − I1 )∂ 2 /∂γ 2 ≡ exp(at∂ 2 /∂γ 2 ). (13.4.11b)
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 399

It acts on the term exp(±iγ) in Q11 and Q12 , converting both generators
to functions of time.
The transformation from the generators Qa to Qab is a similarity trans-
formation. Application to all the generators, Qa , of the Lie algebra asso-
ciated with Hamiltonian Ha does not change their commutation relations:
the corresponding set of generators Qab are those of an isomorphic Lie
algebra. Its generators are the quantum mechanical representatives of time-
dependent constants of motion of system b. An analogous change occurs
when the symmetry of the top is reduced to that of the asymmetric top
whose degeneracy group is simply SO(3).
The inverse of the transformation Sab is the transformation S−1
ab , which
can convert time-dependent generators of the dynamical group of the sym-
metric top, to the generators Q11 and Q12 of the degeneracy group of
the spherical top. Usually one expects perturbations to shrink degeneracy
groups. Here one has a perturbation that enlarges a degeneracy group, and
does so in a most obvious manner. As pointed out in Chapter 1, if one does
not know the dynamical group of a system, the perturbations that produce
increases or changes in observed symmetries are very difficult to detect.
In 1984 Harter and Patterson introduced the concept of rotational
energy surface. Such surfaces depict the relation between the energy of
a rotating object and the orientation of its angular momentum vector in
a body fixed system. The angular momentum vector is assumed to have a
given direction and magnitude in a space-fixed system. For a classical top
with angular momentum of magnitude L2 and principal moments of inertia
Ix = 1/2A, Iy = 1/2B, Iz = 1/2C,

H = A L2x + B L2y + C L2z . (13.4.12)

If the Euler coordinates of L in the body-fixed frame are −α, −β, −γ, then
the energy, E, of the top is given by the expression25

E = J2 {(1/3)(A + B + C) + (1/6)(2C − A − B)(3 cos2 (β) − 1)

(13.4.13)
+ (1/2)(A − B) sin2 (β) cos(2γ)}.

Figures 13.4 illustrate the resulting constant energy surfaces for prolate,
oblate and spherical tops. They are obtained by setting β and γ equal to the
polar coordinates −θ and −φ, respectively, then considering E = r(θ, φ).
Because the classical and quantum mechanical dynamical groups of these
October 28, 2010 16:55 9in x 6in b943-ch13

400 Dynamical Symmetry

Fig. 13.4.1 Momental ellipsoid of a prolate symmetric top in the body-fixed

system. The angular momentum vector is stationary in the space-fixed system.
I z > I x = Iy .

Fig. 13.4.2 Momental ellipsoid of an oblate symmetric top in the body-fixed

system. Iz > Ix = Iy .

Fig. 13.4.3 Momental ellipsoid of a spherical top in the body-fixed system. Iz =

Ix = I y .
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 401

systems are locally isomorphic, the relation between classical trajectories

of constant E on these surfaces can be used to graphically illustrate their
dynamical symmetries. The energy surfaces of the prolate and oblate sym-
metric tops express an SO(2) rotational symmetry about an axis; that of
the asymmetric top has no such symmetry. The reader is referred to Har-
ter’s article for a discussion of the classical trajectories of constant energy
on the surface of these tops.

13.5 The U(4) Vibron Model of Rotating Vibrating

Diatomics
If there were no coupling between the rotational motion and vibrational
motion of a diatomic molecule with a spectrum generating U(2) or SO(3)
group, one would expect the dynamical group of the vibrating and rotating
system to be U(2) ⊗ SO(3) or SO(3) ⊗ SO(3). However, one knows that
the moment of inertia of the molecule will increase when centrifugal forces
stretch the bond, so one anticipates interaction between the two groups. To
deal with this interaction Iachello introduced a U(4) model.4 Its Hamilto-
nian must be invariant under rotations of a coordinate system with origin
at the center of mass of the molecule. There are two group chains starting
with U(4) that meet this requirement:

(a) : U(4) ⊃ U(3) ⊃ U(2) ⊃ O(2),

(13.5.1a,b)
(b) : U(4) ⊃ O(4) ⊃ O(3) ⊃ O(2).

For chain (a), Iachello set

H = Ha = E0 + α1 C1 (U(3)) + α2 C2 (U(2)) + λL2 , (13.5.2a)

and for chain (b), set

H = Hb = E0 + A C2 (SO(4)) + BL2 . (13.5.2b)

The variables E0 , α1 , α2 , λ, A, B, are parameters whose value is determined

from experimental data.
4 †
The eigenvalue, N , of the U(4) number operator i bj bj , fixes the
dimension of the U(4) UIR, and thus determines the total number of bound
states the system can support. Let n be the eigenvalue of the U(3) num-
3
ber operator, i b†j bj , and let l, m be standard SO(3), SO(2) quantum
October 28, 2010 16:55 9in x 6in b943-ch13

402 Dynamical Symmetry

numbers. Then the rovibrational state |N nlm
will have

Ea = E0 + α1 n + α2 n(n + 3) + βl(l + 1),

(13.5.3)
n = N, N − 1, . . . , 0, l = n, n − 2, . . . , 1 or 0.

If w is the SO(4) quantum number with values 1, 2,. . . , then the rovibra-
tional state |N wlm
will have

Eb = E0 + Aw(w + 2) + Bl(l + 1),

(13.5.4)
w = N, N − 2, . . . , 1 or 0, l = w, w − 1, . . . , 0.

To identify the l = 0 case of Ha with a Morse Hamiltonian one must

have

Ea = E0 + α1 n + α2 n(n + 3),
(13.5.5)
E0 = −(1/4)(β 2 /2µ), α1 = (5/2)(β 2 /2µ), α2 = −(β 2 /2µ).

If one wishes to generalize to the l = 0 case, one must also take into account
the fact that n is an integer which jumps by unity from 0 up to N , while |m|
is half integer and jumps from 1/2 to j. This can only be arranged by setting
|m| = n + 1/2, and N = j + 1/2. This, however, requires l to be shifted by 2
units, so the change in l
is that of a two-photon transition. This indicates
that model (a) should be interpreted as applying to systems other than a
Morse oscillator. For further discussion of this algebraic Hamiltonian the
reader is referred to the review by Oss.26
To identify Hb with a Morse Hamilton when l = 0, one must take into
account that w is an integer which jumps by two units from 0 to N . This
requires that |m| = w/2 + k, where k = 0 if N is odd and k = 1/2 if N is
even. One then finds that A = −(1/4)(β 2 /2µ) = −(k/2)(β 2 /2µ), so k must
be chosen as 1/2. It follows that N must be an even number. Also, E0 must
have the same value it has under case (a). Consequently

Eb = E0 + (1/4)(β 2 /2µ)(w + 1)2 + Bl(l + 1). (13.5.6)

As in the atomic case considered in previous chapters, the SO(4) Lie algebra
contains operators that shift l by one unit.
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 403

In both cases (a) and (b), the energy expression is the sum of a kinetic
energy of rotation and an internal vibrational energy. As the internal energy
of the vibration motion increases, the number of rotational states allowed to
the bound system must decrease. If ν is the vibrational quantum number,
as it increases n and w must decrease. This is accommodated in model (a)
by setting n = j − ν − 1/2. In model (b) one sets (w + 1)/2 = j − ν.
The generators of U(4) and SO(4) that couple U(2) ⊗ U(2) relate the
vibrational and rotational motions, while converting product wave functions
or the form R(r)Ylm (θ, φ) to functions Z(r, θ, φ) that could be expressed as
a sum of functions of the general form Rl (r)Ylm (θ, φ). The existing work
on the U(4) model is examined in the monograph of Iachello and Levine,
who conclude that chain (b) “provides a reasonably good description of the
spectra of rigid diatomics”, but “quite often the acual situation deviates
somewhat from that of a simple Morse oscillator.” They find that some
general expansions which are polynomial functions of the Casimir operators
of model (b) converge more rapidly than standard Dunham expansions.

13.6 The U(4) ⊗ U(4) Model of Rotating Vibrating

Triatomics
Conceptually, mathematically and physically, the study of the coupling of
rotational and vibrational motions in systems with more than a few degrees
of freedom is nontrivial. For molecular physics, the correlation diagrams
of Kellman, Amar and Berry that connect the energy levels of rigid and
floppy triatomics are of significant assistance.27 From them, and the pre-
vious discussion of tops, one would expect that the coupling of vibrational
and rotational motions might be modeled as a coupling of the generators of
the vibron model with those generators of the dynamical groups of the tops
that affect the relation between the space-fixed total angular momentum
vector and the principle axes of inertia.
In their treatment of the rovibronic motion of triatomics, van Roos-
malen, Iachello, Levine and Diepernik have introduced a model in which
a U(4) algebra is used to model each vectorial degree of freedom in the
molecule.29 In a triatomic molecule, A–B–C, there is one vectorial degree
of freedom along bond A–B, and another along B–C. The total number of
internal degrees of freedom is then six, of which three describe the internal
configuration of the molecule, and three describe its orientation in an iner-
tial system. With this as a starting point, they investigate the consequences
October 28, 2010 16:55 9in x 6in b943-ch13

404 Dynamical Symmetry

of assuming that the Hamiltonian can be expressed in terms of generators

of U(4)1 ⊗ U(4)2 .
Each of the U(4) groups can be reduced using the chains (a), or (b)
above. Couplings between the chains can also be introduced, starting with
the coupling of U(4)1 ⊗ U(4)2 to U(4)12 , or at subsequent points along the
chains. They established that U(4)1 ⊗ U(4)2 can be reduced to SO(3) by
ten different intermediate chains, and determined those that are physically
relevant. These chains and their rough interpetations are:
Ia: U(3)1 ⊗ U(3)2 ⊃ O(3)1 ⊗ O(3)2 One rigid bond
Ib: U(3)1 ⊗ O(4)2 ⊃ O(3)1 ⊗ O(3)2 One rigid bond
Ic: O(4)1 ⊗ O(4)2 ⊃ O(3)1 ⊗ O(3)2 One rigid bond
IIa: U(3)1 ⊗ U(3)2 ⊃ U(3)12 No rigid bond (13.6.1a–g)
IIb: U(4)12 ⊃ U(3)12 No rigid bond
IIIa: O(4)1 ⊗ O(4)2 ⊃ O(4)12 Two rigid bonds
IIIb: U(4)12 ⊃ O(4)12 Two rigid bonds
There are three further chains in which, according to the notation of Chap-
ter 12.5, reductions such as O(4)1 ⊗ O(4)2 ⊃ O(4)1−2 are used. These are
IIIc: U(4)1−2 ⊃ O(4)1−2 Two rigid bonds
IVa: U(3)1 ⊗ U(3)2 ⊃ (3)1−2 No rigid bond (13.6.1h–j)
IVb: U(4)1−2 ⊃ U(3)1−2 No rigid bond
The authors first point out that there is not enough group structure
in the type I cases to substantially reduce the size of the matrices that
must be diagonalized when calculating the eigenvalues and eigenvectors of
a Hamiltonian with the assumed group structure. They then turn to type
II models. Quantum numbers associated with each group in the chain are
used to lable eigenstates of a Hamiltonian that is a linear combination
of the Casimir operators of each group. This gives an expression for the
rovibrational energy as a function of the quantum numbers, which can be
compared with a Dunham expansion. It was found that chain IIIa is the
most useful. In this case the Lie-algebraic Hamiltonian is
H = E0 + A1 C2 (SO(4)1 ) + A2 C2 (SO(4)2 ) + B C2 (SO(4)12 )
+ B
C
2 (|SO(4)12 |) + C C2 (SO(3)12 ). (13.6.2)
Here C
2 (|SO(4)12 |) is the square root of the quartic Casimir operator of
SO(4)12 . The eigenvalues of the Casimir operators and their relationship to
standard spectroscopic vibrational normal-mode quantum numbers are as
follows:
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 405

C2 (SO(3)12 ): J(J+1), the measure of the total angular momentum of the

molecule
C2 (SO(4)1 ): w1 (w1 + 2), w1 = N1 − 2v1
C2 (SO(4)2 ): w2 (w2 + 2), w2 = N2 − 2v2 (13.6.3a–e)
C2 (SO(4)12 ): τ1 (τ1 + 2) + (τ2 )2
C
2 (|SO(4)12 |): τ2 (τ2 + 1)
The quantum numbers vj are those of Morse-like normal vibrational modes.
For bent molecules,
τ1 = N1 + N2 − 2(v1 + v2 + v3 ) − K, K = τ2 . (13.6.3f)
For linear molecules,
τ1 = N1 + N2 − 2(v1 + v2 ) − K2 , K 2 = τ2 . (13.6.3g)
The quantum number K is that of a Poeschl–Teller oscillator28 that has
potential function
V = −d0 p(p + 1)/cosh(aθ)2 ,
with p an integer. Its energy levels are
E(ν) = −d0 (p − ν)2 , ν = 0, 1, 2, . . . , p.
K and K2 are the projection of J onto the corresponding principal axis of
the body-fixed system.
The coefficient B
is zero for linear molecules. When B
= 0, the H
of (13.6.2) has four fewer empirical parameters than the corresponding
Dunham expansion. For this case, van Roosmalen, Iachello, Levine, and
Diepernik29 use a least-squares method to fit the parameters in this H to
the values required to best fit the spectra of HCN and CO2 , which run from
667 to 18,377 cm−1 . They obtained fits with root-mean-square deviation of
9 to 12 cm−1 . Clearly, the dynamical symmetries of the model express real
and exploitable dynamical symmetries of molecular motions. It is particu-
larly noteworthy that the model deals well with known Fermi resonances
in CO2 .
The authors attempted an analogous fit to calculated spectra of the
nonlinear ion H+ 3 . This uncovered the inability of the Hamitonian to deal
with degeneracies removed by Darling–Dennison couplings.30
In a more recent work Iachello and Oss used a vibron Hamiltonian
closely related to that in (13.4.12) to deal with a class of Darling–Dennison
couplings in bent XY2 molecules. In it they consider C2 (SO(4)1 ) and
C2 (SO(4)2 ) to refer to local vibrational modes rather than normal modes.
October 28, 2010 16:55 9in x 6in b943-ch13

406 Dynamical Symmetry

They then add to their Hamiltonian the coupling term in the quadratic
Casimir operator of U(4)12 . This removes the degeneracies of the local-
modes that it requires only minor degenerate perturbation calculations to
accomodate. This is because the term only couples states with identical
SO(4)12 quantum numbers, i.e., with τ1 = τ2 . They find that with this
sophistication, the algebraic Hamiltonian of chain IIIa determines overtone
frequencies of the nonlinear molecules H2 O16 , H2 O18 , D2 O16 , H2 S32 and
S32 O16
2 , with root mean square deviations of 1 to 5 cm
−1 16c
. A similar, but
slightly more elaborate treatment of the spectra of N2 O, C12 O2 , C13 O2 ,
OCS and HCN, produced results of the same accuracy.16d
Harter and his coworkers developed extensions to semi-rigid rotors of
the energy surfaces of rigid rotors noted above. This produced amazing
insights that led to a number of useful simplifications in treatments of the
rotation of systems at least as complex as SF6 .3

13.7 Concluding Remarks

The remarkable successes of vibron models strongly suggest that for a
wide range of molecules, they and their classical counterparts express real
dynamical symmetries defined by differential operators as yet unknown.
The ability of some of the models to deal with the rotational stretching of
bonds suggests that even in the case of diatomics they may reflect dynam-
ical symmetries that exist when the Born–Oppenheimer approximation is
not imposed. If this guess is validated, these models may also find applica-
tion in technologically important studies of vibrational–electronic coupling
in solid state systems.

Exercises
1. Hold a claw hammer by its handle, with its head–claw axis in a horizon-
tal plane, and note the direction in which the head points. Throw the
hammer up in the air, giving it a rotation about its intermediate axis of
inertia. Catch it by its handle as it descends after one revolution about
this axis. In what direction does the head point? The figures on page
327 of Harter’s article, Ref. 25, can help in developing an explanation of
your observations.

References
[1] M. Moshinsky, Group Theory and the Many-Body Problem (Gordon &
Breach, New York, 1967).
October 28, 2010 16:55 9in x 6in b943-ch13

Rovibronic Systems 407

[2] D. J. Rowe, in Dynamical Groups and Spectrum Generating Algebras, Vol. I,

eds. A. Bohm, Y. Ne’eman and A. O. Barut (World Scientific, Singapore,
1988).
[3] A. Arima and F. Iachello, Phys. Rev. Lett. 35 (1975) 1069.
[4] F. Iachello, Chem. Phys. Lett. 78 (1981) 581.
[5] F. Iachello and R. D. Levine, J. Chem. Phys. 77 (1982) 3046; also cf. A.
Frank and R. Lemus, J. Chem. Phys. 84 (1986) 2698.
[6] O. S. van Roosmalen, F. Iachello, R. D. Levine and A. E. L. Dieperink, J.
Chem. Phys. 79 (1983) 2515.
[7] (a) F. Iachello, S. Oss and R. Lemus, J. Mol. Spect. 149 (1991) 132;
(b) F. Iachello, N. Maniniu and S. Oss, J. Mol. Spect. 156 (1992) 190;
(c) T. A. Holme and R. D. Levine, J. Chem. Phys. 131 (1989) 169;
[8] (a) F. Iachello and S. Oss, Chem. Phys. Lett. 187 (1991) 500;
(b) F. Iachello and S. Oss, J. Mol. Spect. 153 (1993) 225;
(c) F. Iachello and S. Oss, J. Chem. Phys. 99 (1993) 7337;
(d) F. Iachello and S. Oss, Chem. Phys. Lett. 205 (1993) 285.
[9] F. Iachello and A. Arima, The Interacting Boson Model (Cambridge Univer-
sity Press, Cambridge, 1991).
[10] F. Iachello and R. D. Levine, Algebraic Theory of Molecules (Oxford Uni-
versity Press, Oxford, 1994).
[11] A. Frank and P. Van Isaker, Algebraic Methods in Molecular and Nuclear
Structure Physics (Wiley, New York, 1994).
[12] R. D. Levine and C. E. Wulfman, Chem. Phys. Lett. 13 (1979) 372.
[13] F. Iachello and R. D. Levine, Algebraic Theory of Molecules, pp. 32–34.
[14] G. Herzberg, Molecular Spectra and Molecular Structure I. Spectra of
Diatomic Molecules (D. Van Nostrand, Princeton, 1950) pp. 106–109.
[15] (a) F. Iachello and R. D. Levine, J. Chem. Phys. 77 (1982) 3046;
(b) I. Benjamin and R. D. Levine, J. Mol. Spect. 126 (1987) 486;
(c) F. Iachello and S. Oss, Phys. Rev. Lett. 66 (1991) 2776;
(d) F. Iachello and S. Oss, J. Mol. Spect. 153 (1992) 225;
(e) A. Frank and R. Lemus, Phys. Rev. Lett. 68 (1992) 413.
(f) A. Mengoni and T. Shirai, J. Mol. Spect. 162 (1993) 246.
[16] O. S. van Roosmalen, I. Benjamin and R. D. Levine, J. Chem. Phys. 81
(1984) 5986.
[17] F. Iachello and R. D. Levine, Algebraic Theory of Molecules, pp. 96–98.
[18] F. Iachello and R. D. Levine, Algebraic Theory of Molecules, pp. 104–106.
[19] F. Iachello and R. D. Levine, Algebraic Theory of Molecules, pp. 106-108.
[20] A. R. Edmonds, Angular Momentum in Quantum Mechanics (Princeton Uni-
versity Press, Princeton, 1957), pp. 64–67.
[21] F. Bopp, R. Haag and Z. Naturforsch 5a (1950) 644.
[22] (a) S. Kumei, M. Sc. Thesis (University of the Pacific, 1972);
(b) Robert L. Anderson, S. Kumei and C. E. Wulfman, J. Math. Phys. 14
(1972) 1527.
[23] A. Bohm and R. B. Teese, J. Math. Phys. 17 (1976) 94.
[24] R. L. Anderson, T. Shibuya and C. E. Wulfman, Revista Mexicana de Fisica
23 (1974) 257.
October 28, 2010 16:55 9in x 6in b943-ch13

408 Dynamical Symmetry

[25] W. G. Harter, Comp. Phys. Rep. 8 (1958) 321.

[26] S. Oss, Algebraic Models in Molecular Spectroscopy, Adv. Chem. Phys. XCIII,
eds. S. Prigogine, A. Rice (Wiley, New York, 1996).
[27] M. E. Kellman, F. Amar and R. S. Berry, J. Chem. Phys. 73 (1980) 2387.
[28] G. Poeschl and E. Teller, Z. Phys. 83 (1933) 143.
[29] O. S. van Roosmalen, F. Iachello, R. D. Levine and A. E. L. Diepernik, J.
Chem. Phys. 79 (1983) 2515.
[30] B. T. Darling and D. M. Dennison, Phys. Rev. 15 (1940) 128.
[31] W. G. Harter, ibid. (Ref. 25), Section 3.
October 28, 2010 16:55 9in x 6in b943-ch14

CHAPTER 14

Dynamical Symmetry of Maxwell’s Equations

In his 1905 paper introducing the theory of relativity, Einstein prefaced his
arguments with the observation, “. . . the unsuccesful attempts to discover
any motion of the earth relative to the (aether), suggest that the phenomena
of electrodynamics as well as mechanics possess no property corresponding
to the idea of absolute rest.”1 He then made four key assumptions in devel-
oping his theory of special relativity. These are:

i. Sets of rigid rods and clocks may be used to define sets of local coordi-
nate systems that assign identical space and time intervals to successive
events.
ii. The rods and clocks are not affected by motions at constant velocity
with respect to one another.
iii. Measurements made using these comoving coordinate systems are com-
pared via electromagnetic radiation.
iv. Electromagnetic radiation moves with a constant velocity, c, with
respect to all observers.

Einstein established that these assumptions imply the transformation equa-

tions of Lorentz,2 which relate measurements of time intervals and spatial
displacements made by observers in relative motion with a constant velocity.
Four years later Cunningham and Bateman3a−c showed that Maxwell’s
equations are invariant under Lorentz transformations, and that the validity
of Einstein’s special theory is a necessary consequence of this invariance.
However, Bateman and Cunnningham also proved that Maxwell’s equa-
tions are invariant under the transformations of a larger group, which is
a conformal group. This preserves the angle between spacetime vectors
in Minkowski space, but contains inversion transformations that intercon-
vert small and large spacetime separations. To quote Cunningham, “The

409
October 28, 2010 16:55 9in x 6in b943-ch14

410 Dynamical Symmetry

question arises whether the theory of relativity also holds for the types of
motion of an electromagnetic system derived from one another by such a
transformation.”
The following pages briefly describe the transformations of Maxwell’s
equations carried out by their Poincare invariance group, and their physical
interpretation. The action of the Bateman–Cunningham inversion transfor-
mation is then described, and questions associated with its physical inter-
pretation are considered. The primary concern throughout this chapter
is the physical interpretation of invariance transformations of Maxwell’s
equations — none of which can change the velocity of light in vacuo.

14.1 The Poincare Symmetry of Maxwell’s Equations

Maxwell, in his treatise Electricity and Magnetism,4 provides an extensive
and enlightening discussion of the experimental measurement of electrical
and magnetic forces and fields. In discussing these it is necessary to use
variables whose units must be carefully defined. For our purposes one may
consider the concept of electrical charge, the measurement of the forces
between charges, and the measurement of forces between current carrying
wires, as fundamentals. Experiments beginning with Coulomb5a found that
charge, like mass, can be quantified, and thereafter established what became
known as Coulomb’s law. In it the Newtonian concept of point mass, M, is
replaced by that of point charge, Q. The magnitude of the forces between
two stationary point charges a and b, measured experimentally, is found to
be related to the distance rab between their centers by the law
Fel = κQa Qb /r2ab .
If Q is measured in coulombs C, and force is measured in newtons, i.e.,
kgms−2, then κ is measured in newton m2 /C2 = kg m3 s−2 C−2 . It turns
out that setting κ = 1/4πεo is quite revealing. Here εo ∼= 10−9/36π kg−1
−3 2 2
m s C , is termed the electrical permitivity of the space. Electrical cur-
rent is so defined that a charge of one coulomb flowing by a point in one
second generates a current of one ampere. Experiments culminating with
those of Ampere5b established that the magnetic force between two current-
carrying parallel wires with centers r meters apart in empty space, has
magnitude
Fm = µo Ia Ib /2πr.
If the current measured in wire a, Ia , and that in wire b, Ib , are both mea-
sured in amperes, then the constant, µo , the magnetic permeability of empty
space, is found experimentally to have the value 4π 10−7 kg m/C2 .
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 411

Readers of Maxwell’s collected works6 can gain a good deal of under-

standing of his thought processes, and of the way in which he utilized
Faraday’s graphic idea of lines of force in guiding the development of his
mathematical conceptions of electromagnetic fields. Maxwell wrote down
his equations shortly before Gibbs’ concept of a vector became known, but
quickly became a supporter of Gibbs.7 In a vector notation8 now commonly
used, his equations interrelate the following quantities.

Symbol Name Units

E(r, t), electric field strength, volts/meter (ms−2 kg/C),
H(r, t), magnetic field strength, amperes/meter (m−1 s−1 C),
B(r, t), magnetic flux density, webers/meter2 (s−1 kg/C), (14.1.1)
2 −2
D(r, t), electric displacement, coulombs/meter (m C),
2 −2 −1
J(r, t), electric current density, amperes/meter (m s C),
3 −3
ρ(r, t), electric charge density, coulombs/meter (m C).

Maxwell’s original set of 12 equations can be written as

∇ × E(r, t) + ∂B(r, t)/∂t = 0,

∇ × H(r, t) − ∂D(r, t)/∂t = J(r, t),
(14.1.2a–d)
∇ · B(r, t) = 0,
∇ · D(r, t) = ρ(r, t).

In regions of space in which there is no electric charge, there may,

however, be a displacement current, so J(r, t) need not vanish. In vacuo,
B = µo H, and D = εo E. Equations (14.1.2) then become

∇ × E(r, t) + µo ∂H(r, t)/∂t = 0,

∇ × H(r, t) − εo ∂E(r, t)/∂t = 0,
(14.1.3a–d)
∇ · H(r, t) = 0,
∇ · E(r, t) = 0.

This set of first-order PDEs implies the second-order PDE

∇2 E(r, t) = µo εo ∂ 2 E(r, t)/∂t2 , (14.1.4a)

which governs electric fields. Observing that the velocity of propagation

of the resulting electric field from r to r
is 1/(µo εo )1/2 , and knowing the
results of measurements of µo and εo , Maxwell established that this is c,
the velocity of light.9 Thus in vacuo, electromagnetic radiation obeys the
October 28, 2010 16:55 9in x 6in b943-ch14

412 Dynamical Symmetry

equation

∇2 E(r, t) − ∂ 2 E(r, t)/∂(ct)2 = 0. (14.1.4b)

Also,

H · E = 0. (14.1.4c)

The density of energy in the field is, in units of Jm−3 , given by

(εo E2 + µo H2 )/2. (14.1.5a)

√ √
Formulae are simplified if one absorbs εo into E, and µo into H. We
shall do this and denote the resulting field vectors by Ē and H̄. Then, for
example, (14.1.5a) becomes

(Ē2 + H̄2 )/2. (14.1.5b)

Defining x = x1 , y = x2 , z = x3 , and ct = x4 , the generators of the

Lorentz invariance group of Maxwell’s equations are

Lij = xi ∂/∂xj − xj ∂/∂xi = −Lji , i < j = 1, 2, 3; (14.1.6a)

Kj4 = xj ∂/∂x4 + x4 ∂/∂xj = K4j , j = 1, 2, 3. (14.1.6b)

Their Poincare invariance group is generated by these, and the further

operators,

Tpj = ∂/∂xj, j = 1, . . . , 4. (14.1.6c)

The commutation relations satisfied by these generators are

[Lab , Lbc ] = Lac , [Lab , Kb4 ] = Ka4 , [Ka4 , Kb4 ] = Lab ,

(14.1.7)
[Lab , Tpb ] = Tpa , [Ka4 , Tpb ] = Tpa , [Tpa , Tpb ] = 0.

The physical interpretation of the transformations of the group, dealt with

in Chapter 2, is summarized in the next paragraph.
Before beginning to deal with the action of Lorentz transformations on
electromagnetic fields we must define their action on points in spacetime.
Let observer b have velocity Vt with respect to observer a. Let x, y, z and
t be the spacetime coordinates of an event as measured in the coordinate
system of observer a, and let x
,y
,z
and t
be the coordinates that observer
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 413

b assigns to the same event. If β ≡ (1 − (v/c)2 )−1/2 , then the observations

of b are related to those of a by the Lorentz transformations

x
= β(x − vt), y
= y, z
= z, t
= β(t − vx/c2 ),

i.e. by (14.1.8)

x
= β(x − (v/c)ct), y
= y, z
= z, t
= β(ct − (v/c)x)/c.

The operators exp(αk Tpk ) translate points in spacetime, and leave

∆x, ∆y, ∆z and ∆t, unchanged.
Now suppose that observer a determines the components of E, H at
a point x, y, z and t in spacetime to be Ex , Ey , Ez and Hx , Hy , Hz , and
that both observers are using right-handed coordinate systems. Then the
analysis of Cunningham and Bateman establishes that if observer b is mov-
ing at velocity v = vx with respect to a, then observer b will find these
components to be

E
x = Ex , E
y = β(Ey − (v/c)Hz ), E
z = β(Ez + (v/c)Hy ),

(14.1.9)
H
x = Hx , H
y = β(Hy + (v/c)Ez ), H
z = β(Hz − (v/c)Ey ).

Cunningham and Bateman showed that Eqs. (14.1.4) and (14.1.5) are
invariant under Lorentz transformations, and that the same is true of
Eq. (14.1.2). Figures. 14.1.1a,b depict, in the frame of a, the electric
field E = Ey k, between the plates of a capacitor, and the magnetic field
H generated by a current running through a wire on the x axis a. Fig-
ures. 14.1.1c,d depict the fields E
= and H
that are observed by b,
moving at velocity v = 0.7ci with respect to a.

14.2 The Conformal and Inversion Symmetries

of Maxwell’s Equations; Their Physical Interpretation
In the late 1860’s and early 1870’s Lie investigated the continuous groups of
point transformations that leave metrics invariant, and he established con-
nections between the continuous group of conformal transformations that
leaves angles invariant in euclidean spaces and conformal transformations of
the Grupe der reciproken Radien that may alter their metrics.10 In Sec. 14.4
of Bateman’s 1909 paper,3b he refers to this work, and using Minkowski’s
transformation to spacetime coordinates, he and Cunningham generalize
Lie’s results. Cunningham3a establishes that if k is a real constant, the
October 28, 2010 16:55 9in x 6in b943-ch14

414 Dynamical Symmetry

1.0 1.0

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.4 0.5 0.6 0.7 0.8 0.9 1.0

(a) (b)

1.0 1.0

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.4 0.5 0.6 0.7 0.8 0.9 1.0

(c) (d)

Fig. 14.1.1 (a) Electric field in the y–z plane between plates of a capacitor;
(b) magnetic field surrounding a wire along the y-axis; (c) the electric field of
Fig. 14.1.1a observed in the reference frame of a moving observer; (d) the magnetic
field of Fig. 14.1.1.b observed in the reference frame of a moving observer.

spacetime inversion, IST , defined by

(x, y, z, ct) → (x
, y
, z
, ct
) = k2 (x, y, z, ct)/(x2 + y2 + z2 − (ct)2 ),

(14.2.1a)

converts solutions of Maxwell’s equations into solutions. For all real values
of k, one may define the dimensionless four-vector

s = (x, y, z, ct)/k. (14.2.1b)

October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 415

Then k can be considered to be a scale parameter with the dimension of

length. To further simplify notation, we define the scalar product of s with
itself by
s · s = Σj gjj s2j = s21 + s22 + s23 − s24 ≡ s2 . (14.2.1c)
With this change to dimensionless variables, (14.2.1a) becomes
s → s
= s/s2 . (14.2.1d)
It then follows that
s
2 = s2 /(s2 )2 , s
2 s2 = 1, (14.2.2a,b)
and that I−1
ST is defined by

s = s
/s
2 (14.2.2c)
Thus, IST is its own inverse.
In discussing the effect of the inversion (14.2.1) on electromagnetic fields,
it is useful to define
σ = (s1 , s2 , s3 ) = (x, y, z)/k = r/k, (14.2.3a)
τ = s4 , (14.2.3b)
2
u = s21 + s22 + s23 + s24 2
=σ +τ . 2
(14.2.3c)
Cunningham3a showed that the inversion converts solutions E and H of
Maxwell’s equations, to solutions E
and H
, which may be defined by
E
= (−s2 ){u2 E − 2(σ · E)σ + 2τ (σ × H)}, (14.2.3d)
and
H
= (s2 ){u2 H − 2(σ · H)σ − 2τ (σ × E)}. (14.2.3e)
This result holds true for all choices of the origin of the spacetime coordi-
nates for which s does not vanish.
Figures 14.2.1 illustrate the effect of inversions on the fields E and H
of Figs. 14.1.1a,b. Figures 14.2.2 depict the corresponding solutions, E

and H
, of Maxwell’s equations produced by the relations (14.2.3).
In dealing with the effect of the Cunningham–Bateman inversion on the
Lorentz metric
dx2 + dy2 + dz2 − d(ct)2 = dx21 + dx22 + dx23 − dx24 , (14.2.4a)
we will usually use the dimension free metric
ds2 = (dx2 + dy2 + dz2 − d(ct)2 )/k2 . (14.2.4b)
October 28, 2010 16:55 9in x 6in b943-ch14

416 Dynamical Symmetry

1.0 1.0

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.4 0.5 0.6 0.7 0.8 0.9 1.0

(a) (b)

1.0 1.0

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.4 0.5 0.6 0.7 0.8 0.9 1.0

(c) (d)

Fig. 14.2.1 (a) Electric field in the y–z plane between plates of a capacitor;
(b) effect of the inversion on the electric field at t = 0; (c) effect of the inversion
on the electric field at t = 0.5c; (d) effect of the inverson on the electric field at
t = 0.9c.

Inversion converts (14.2.4b) to

ds
2 = (1/s2 )2 ds2 . (14.2.4c)
As both s2 and ds2 are Lorentz invariant, for all nonvanising s, ds
2 is also
invariant under the operations of the Lorentz group. However, the inversion
does not leave invariant the Lorentz invariant metrics (14.2.4a) or (14.2.4b)
if ds2 = 0. And, though ds2 is invariant under Poincare transformations of s,
ds
2 is not when ds2 = 0. This has the consequence that Maxwell’s equations
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 417

1.0 1.0

0.9
0.8

0.8
0.6

0.7
y

y
0.4
0.6

0.2
0.5

0.4 0.0

0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x
(a) (b)

1.0 1.0

0.9 0.9

0.8 0.8

0.7 0.7
y

0.6 0.6

0.5 0.5

0.4 0.4

0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x x
(c) (d)

Fig. 14.2.2 (a) Effect of spacetime inversion on the electric field between plates
of a capacitor, ct = 0; (b) effect of spacetime inversion on the electric field between
plates of capacitor, ct = 0; (c) effect of spacetime inversion on the magnetic field
surrounding a current-carrying wire, ct = 0; (d) effect of spacetime inversion on
the magnetic field surrounding a current-carrying wire, ct = 1.

are invariant under transformations that rescale a Lorentz metric. At each

origin the metric takes on a radial position dependence.
Using (14.2.2b), relation (14.2.4c) may be expressed in the more sym-
metrical form,

ds
2 /s
2 = ds2 /s2 . (14.2.4d)

October 28, 2010 16:55 9in x 6in b943-ch14

418 Dynamical Symmetry

If dr = (dx, dy, dz) is a displacement vector defining the time-varying posi-

tion of a point, and dr ≡ |dr| = cdt, then the point is moving with the
velocity of light in vacuo. In this case, ds2 = 0, and (14.2.4c) requires, for
finite values of s and s
, that ds
2 = 0.
The Bateman–Cunningham group of conformal transformations con-
tains a 15 parameter continuous subgroup of conformal transformations.
To begin our discussion of this Lie group we use the method of Sec. 4.7 to
convert the generators Tpj = ∂/∂xj in (14.1.7) to operators acting on the
variables s
j . If
Tj ≡ k∂/∂xj = ∂/∂sj , D
≡ s
1 ∂/∂s
1 + s
2 ∂/∂s
2 + s
3 ∂/∂s
3 + s
4 ∂/∂s
4 ,
(14.2.5a,b)
then
D ≡ s1 ∂/∂s1 + s2 ∂/∂s2 + s3 ∂/∂s3 + s4 ∂/∂s4 , (14.2.5c)
Tj = ∂/∂sj → T
j =s
2
∂/∂s
j − 2s
j D
, j = 1, 2, 3, (14.2.5d)
and
T4 = ∂/∂s4 → T
4 = s
2 ∂/∂s
4 + 2s
4 D
, (14.2.5e)
T
4 = (s
2
2
2
2








1 + s2 + s3 + s4 ) ∂/∂s4 + 2s4 (s1 ∂/∂s1 + s2 ∂/∂s2 + s3 ∂/∂s3 ).
(14.2.5f)
The action of the finite transformations generated by T
1 and T
4 is deter-
mined below.
Using (14.2.5) one obtains, for the generators Lij , Kj4 of (14.1.6),
Lij = si ∂/∂sj − sj ∂/∂si → L
ij = s
i ∂/∂s
j − s
j ∂/∂s
i , (14.2.6a)
and
Kj4 = sj ∂/∂s4 + s4 ∂/∂sj → K
j4 = s
j ∂/∂s
4 + s
4 ∂/∂s
j . (14.2.6b)
Supplementing the operators Tj , Lij , Kij , with the further five operators,
P
j ≡ ∂/∂s
j , (14.2.6.c)


D ≡ s
1 ∂/∂s
1 + s
2 ∂/∂s
2 + s
3 ∂/∂s
3 + s
4 ∂/∂s
4 (14.2.6d)
yields the 15 generators of a Lie group that acts in the space of the s
j .
Replacing s
j by sj in all these operators, one obtains the generators of the
isomorphic conformal group acting in the space of the sj . These are
Kj4 , Lij , Pj = ∂/∂sj and D = Σ41 sj ∂/∂sj , (14.2.7a–d)
and
Cj ≡ s2 ∂/∂sj − 2sj D, j = 1, 2, 3. (14.2.7e)
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 419

They satisfy the commutation relations of the Poincare group, cf. (14.1.7),
and the following commutation relations:

[D, Lab ] = 0, [D, Kbc ] = 0, [D, Pa ] = −Pa , [D, Ca ] = Ca ,

[Lab , Cb ] = Ca , b = 1, 2, 3; [Lab , C4 ] = 0,
[Ka4 , Pb ] = −δa,b P4 , b = 1, 2, 3; [Ka4 , P4 ] = −Pa ,
[Ka4 , Cb ] = −δa,b C4 , b = 1, 2, 3; [Ka4 , C4 ] = −Ca ,
[Pa , Cb ] = 2(Lab − δa,b D), b = 1, 2, 3; [Pa , C4 ] = 2Ka4 , a = 1, 2, 3;
[P4 , C4 ] = 2D, [Ca , Cb ] = 0.
(14.2.8)

The Cj are known as generators of the special conformal group.11 They,

and D, inherit from the inversion transformation the ability to change
Lorentz metrics.
The action of the group operators exp(αi Ci ), can, a la Lie, be obtained
from the translation operators, exp(αi Pi ), by inversion. Noting that

exp(αi Ci ) = (Ist )−1 exp(αi Pi )Ist , (14.2.9a)

one finds

exp(αi Ci ) sj = (sj + δi,j αi s̄2 )/(1 + 2giiai si + a2i s̄2 ). (14.2.9b)

With this result in hand, one can easily obtain all the finite transformations
of the full 15 parameter conformal group. The operations of this continuous
group do not, in general, leave the Lorentz metric invariant. The group is
(see the exercises) locally isomorphic to a SO(4,2) group.
Cunningham argued that, just as in the case of the Poincare trans-
formations of special relativity, the inversion and special conformal trans-
formations of Maxwell’s equations could not make it possible “for an
observer to discriminate between the sequence of electromagnetic phenom-
ena as he knows them and the sequence obtained by this transforma-
tion.” Cunningham arrived at this conclusion before the advent of quantum
mechanics. A fundamental contribution by Barut and Haugen12 implies
that it can only continue to be correct if one replaces the usual concept of
(Poincare invariant) mass by a concept of mass that is invariant under the
subgroup of special conformal transformations. Barut and Raczka point
out that the operations of this subgroup of the Bateman–Cunningham
invariance group of Maxwell’s equations also leave invariant relativistic
wave equations for massless particles of spin 0 and 1/2.12 However, wave
October 28, 2010 16:55 9in x 6in b943-ch14

420 Dynamical Symmetry

equations for massive particles are only invariant under its actions if the
mass m is also transformed by a dilation operator that also carries out the
transformation

m2 → m
2 = e2β m2 , (14.2.10a)

and the special modified conformal transformations must convert

m2 → m
2 = (1 + 2α · s + α2 s2 )m2 , (14.2.10b)

and so yield a mass that evolves in spacetime in a manner that could be

difficult to detect if the group parameters are small.13
To carry out these transformations, the authors added an additional
term to the expressions for the group generators Cj and D. Several other
attempts to use conformal transformations to extend special relativity will
be found in papers by Fulton, Rohrlich and Witten,13a and by Hill13b−d .
Taking a nonrelativistic limit of the generators of their conformal
group, and converting them to Schrödinger operators, Barut and Raczka
obtain the time-independent and time-dependent constants of motion of
the Schrödinger equation of a free-particle, and of the Schrödinger equa-
tion of an isotropic oscillator. The author has shown how a related lim-
iting process can be used to convert generators of the special conformal
group to the Runge–Lenz operators of bound and unbound states of regular-
ized hydrogen-like atoms, and produce the conjugate Fock variables.14 The
transformation converts relativistic free-particle motion to classical Kepler
motion. Many other applications of conformal groups are discussed in an
engaging and extensive historical review by Kastrup which also contains a
critical analysis of Ref. 13.15

14.3 Alteration of Wavelengths and Frequencies

by a Special Conformal Transformation:
Interpretation of Doppler Shifts in Stellar Spectra
In this section we will consider further macroscopic consequences of the
invariance of Maxwell’s equations under the Bateman–Cunningham confor-
mal group. As will be shown below, transformations of the special conformal
group alter the frequencies and wavelengths of electromagnetic waves with-
out altering their velocity. This can produce effects similar to the Doppler
effect, and as will appear below, these may have been detected in radar
signals received from spacecraft.
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 421

When a source emitting radiation with wavelength λ has velocity V
with

respect to a spectrometer, the wavelength, λ
, measured by a spectrometer
or interferometer is given by the equation16

1/2
1 + V
/c
λ
/λ = . (14.3.1)
1 − (V
/c)
Expanding this in powers of V
/c and keeping only the first two terms, one
obtains the non-relativistic Doppler equation
λ
= (1 + V
/c)λ. (14.3.2a)
Rearranging this produces the usual Doppler relation
V
= c(λ
/λ − 1). (14.3.2b)
The special conformal transformations that alter λ to λ
can produce the
same effect as a relative velocity, V
, of a source with respect to an observer.
Because the velocities of distant heavenly bodies are estimated using
measured Doppler shifts of wavelengths in the spectra of radiation emit-
ted by them, special conformal transformations can, in principle, intro-
duce ambiguities into astronomy, astrophysics, and cosmology. Beginning
in the 1920’s Hubble made an extensive series of astronomical observations
of Cepheid variable stars and the Doppler shifts of spectral lines in light
received from them.17 He concluded that his observations were consistent
with the supposition that in a galaxy the stars at a distance R from the
earth are receding from earth with a mean radial velocity V
proportional
to R. This relation may be expressed as
V
= H0 R. (14.3.3a)
If V represents the radial velocity of star with respect to this mean, this
becomes Hubble’s equation
V
= H0 R + V. (14.3.3b)
In the mid 1940’s, it was estimated that Hubble’s constant, H0 had a value
approximately equal to 1.8 × 10−17 s−1. The currently accepted value of H0
is (2.19 ± 0.56)× 10−18s−1 .18a The estimated values of V, R, and H0 depend
in part upon observations and empirical relations that supplement Doppler
measurements — most notably, relations between the period and brightness
of Cepheid variables.18b The analyses and correlations of observations that
support Hubble’s theory and equation (14.3.3b) are now very extensive.
They have led to the conclusion that throughout the universe objects not
gravitationally bound to each other are uniformly receding from each other.
October 28, 2010 16:55 9in x 6in b943-ch14

422 Dynamical Symmetry

Thus distances between most heavenly objects appear to slowly increase in

much the way distance between points on the surface of a balloon increase
as the balloon is inflated.
To take this inflation into account, cosmologists often alter Minkowski’s
metric. It is supposed that in regions free of strong local gravitational effects
the metric governing heavenly motions is a Robertson–Walker metric, a
metric of the form18c
ds2 = dx24 − a(x4 )2 {dr2 /(1 − (r/ρ)2 ) + r2 (dθ2 + sin2 (θ)dφ2 )}. (14.3.4)
In this expression, ρ is a constant with the dimension of length, and the
function a(x4 ) expresses the time dependence of the radius ρa(x4 ) of the
universe.
In 1945, Hill19 noticed an unexpected property of the special conformal
transformations generated by
C4 = (r2 + x24 )∂/∂x4 + 2x4 r∂/∂r. (14.3.5)
The group operator exp(βC4 ) carries out the transformations
r → r
= γr, x4 → x
4 = γ{x4 − βs2 }, (14.3.6a,b)
in which
s2 = (x24 − r2 ), γ = γ(β, x4 , r) = (1 − 2βx4 + β 2 s2 )−1 (14.3.6c,d)
To the first-order in the group parameter:
r
= (1 + 2βx4 )r, x
4 = x
4 + β(r2 + x24 )/2. (14.3.7a,b)
Hill wrote these equations as
r
= (1 + αt)r, t
= t + α(r2 /c2 + t2 )/2, (14.3.7c,d)
with α = 2βc. From them he derived the change in velocity from v = dr/dt,
to v
= dr
/dt
:
v
= v + α(r − v(r · v)/c2 ). (14.3.8)
On neglecting terms of order v2 /c2 this gives
v
= v + αr. (14.3.9)
Hill observed that this equation could be interpreted as Hubble’s relation
if one sets α equal to H0 . However, in startling contrast to Hubble’s relation,
Hill’s relation arises solely from a change in spacetime coordinate systems.
Its implications have been overlooked for nearly 65 years. Here, we will
concentrate our attention on one of the questions it raises, the question:
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 423

“What is the spacetime metric governing the motion of electromagnetic

waves?”20
To address this question it is necessary to develop additional conse-
quences of the transformation generated by C4 . Using (14.3.6) one finds
that the transformation carries

dr → dr
= γ 2 (A dr + B dx4 ), (14.3.10a)

and

dx4 → dx
4 = γ 2 (B dr + Adx4 ), (14.3.10b)

with

A = 1 − 2βx4 + β 2 (r2 + x24 ), B = 2βr(1 − βx4 ). (14.3.10c,d)

Thus

dr
/dx
4 = (A dr/dx4 + B)/(B dr/dx4 + A). (14.3.10e)

Using (14.3.10) one finds that the Minkowski metric

ds2 = |(dr)2 − (dx4 )2 | (14.3.11)

is transformed to the conformal metric

ds
2 = |(dr
)2 − (dx
4 )2 | = γ 2 |(dr)2 − (dx4 )2 |. (14.3.12a)

Thus

ds
2 = γ(α/2c, x4 , r)2 ds2 , ds2 = γ(β, x
4 , r
)2 ds
2 . (14.3.12b,c)

Note that the conformal metric smoothly reduces to the Minkowski metric
as the group parameter α approaches zero. This metric is specific to C4 . As
it is the only special conformal metric we will deal with in the remainder
of the chapter, we will sometimes, simply call it the conformal metric.
The function γ introduces a relation between coordinates that destroys
the Poincare invariance of spacetime with metric ds2 . In this respect the
conformal metric is qualitatively similar to the Robertson–Walker metric.
However, the conformal metric produces a local multiplicative scaling of
both dr and dx4 , while the Robertson–Walker metric only applies such a
rescaling to dr. Most importantly, in the conformal metric the choice of
spacetime origin is arbitrary. However because the origin r = 0, t = 0, is
an invariant point of the conformal transformation, there one also has the
special property r
= 0, t
= 0.
October 28, 2010 16:55 9in x 6in b943-ch14

424 Dynamical Symmetry

Fig. 14.3.1 Radial and axial coordinates of points on a light cone.

Though the conformal metric is not Poincare-invariant, exp(β4 C4 )

defines similarity transformations that convert relations between group gen-
erators that subsist in Minkowski spacetime into isomorphic relations in the
conformal spacetime. Thus the Poincare group reappears with a different
realization as an invariance group of spacetime with conformal metric ds
2 .
We will henceforth consider that Eqs. (14.3.6), together with
Eqs. (14.3.10), define mappings that relate the two sets of coordinates,
(r, t, v = dr/dt) and (r
, t
, v
= dr
/dt
), of any point, P. As will perhaps
become evident in the following paragraphs, thinking in terms of map-
pings can help one avoid the insertion of unrecognized presuppositions into
descriptions of physical phenomena.
Let a point P on a trigonometric wave have ordinary coordinates (r,t),
and conformal coordinates (r
, t
). If it is an electromagnetic wave, the point
will have radial velocities

dr/dt = vEM = dr
/dt
, |vEM | = c, (14.3.13)

the same in both coordinate systems. Suppose P1 = P+dP and P2 = P−dP

are two such points on an electromagnetic wave, and suppose they are
separated by one wavelength. Denote this wavelength “at P,” as measured
in the two systems, by λ and λ
, and denote the corresponding frequencies
by ν and ν
. (As a consequence of Eq. (14.3.12), one has λν = c = λ
ν
.)
The frequency of the wave in the region centered at P, as measured in the
two coordinate systems, is the inverse of the time interval that it takes for
this one wavelength to pass through a point with fixed position coordinates
r, r
. Let δt and δt
be the interval as measured in the two systems. Equation
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 425

(14.3.10b) then implies that

δt
/δt = ν/ν

2
1 − 2βx4 + β 2 (s2 + 2r2 )
= . (14.3.14)
(1 − 2βx4 + β 2 s2 )
Thus as a point P on the wave propagates with it, (14.3.14) requires that the
conformal wavelength, λ
, of the wave at P must vary with the coordinates
(r, x4 ) according to the law

λ
/λ = {1 − 2βx4 + β 2 (s2 + 2r2 )}/(1 − 2βx4 + β 2 s2 )2 . (14.3.15a)

Consequently,

λ
/λ = 1 + 2βx4 + 3β 2 (x24 + r2 ) + O(β 3 ). (14.3.15b)

Because frequencies and wavelengths measured in Minkowski space do not

vary with position or time, λ may be considered constant in this equation.
It follows that the wavelength in the conformal coordinate system increases
or decreases with t as βct is positive or negative. The two wavelengths are
equal at the common origin of coordinates, (0, 0). As there is no evidence
for supposing that λ
differs from λ in regions near a wave source, we shall
choose the origin (0, 0) to be at the source of interest.
When a wave with wavelengths λ, λ
at (r, x4 ), (r
, x
4 ), travels from
its source to the observer, the change in r must equal the change in x4 . If
the source is fixed at the origin, r remains equal to x4 , and substituting
β = α/2c into (14.3.15b) yields

λ
/λ = 1 + αr/c + 3/2(αr/c)2 + O((α/c)3 ). (14.3.16)

This implies that when r reaches R,

c(λ
/λ − 1) = c{αR/c + 3/2(αR/c)2 + O((α/c)3 ). (14.3.17)

Interpreting the left-hand side of this equation as a Doppler shift, one

obtains to O(αR/c),

c(λ
/λ − 1) = V
= αR. (14.3.18)

This Doppler shift will not be observed if the propagation of the wave
has taken place in a space with Minkowski metric: it may be considered
to have arisen because the conformal wavelength, λ
, changes as the wave
propagates.
October 28, 2010 16:55 9in x 6in b943-ch14

426 Dynamical Symmetry

When the observer has velocity V = dR/dT relative to the source, then
the wavelength λ in the (R,T) system undergoes the additional Doppler
shift given by the usual formula

1/2
1 + V/c
∆λ/λ = − 1 = V/c + O(V2 /c). (14.3.19)
1 − V/c
If the sign of V/c is positive when the source is receding from the observer,
a reshift is produced. On neglecting terms of O((αR)2 /c) and O(V2 /c),
(14.3.18) then yields Hill’s relation

V
= αR + V. (14.3.20)

For values of α as small as H0 , the group parameter α could, in principle,

be determined by experimental measurements of the radial positions and
velocities of distant sources producing the measured Doppler shifts.
For sources as distant as the nearest star, only Doppler shifts can be
directly measured. However, the Pioneer spacecraft program established
that it is possible to measure, with an accuracy of one part in 1012 , Doppler
shifts of S band radar frequencies that are developed over distances up to
70 AU.21 This precision is sufficient to measure α values to the order of
10−18 sec−1 with an accuracy better than 1%.
Suppose, for simplicity of thought, that an observer on earth arranges to
have a clock and interferometer with no accelerations. Let these have coor-
dinates R = R
= 0. Suppose that at T = T
= 0, the observer sends a radar
signal to a receding spacecraft. The radar signal will take a time T, T
to
reach the spacecraft, and if it is immediately returned, when it reaches the
observer the clock times will be 2T, 2T
respectively. The observer will con-
clude that when the signal reached the spacecraft the distance from it was
R = cT, R
= cT
. Equations (14.3.7c,d) imply that to the first-order in α,

R
= (1 + αT)R, R = (1 − αT
)R
,

(14.3.21a–d)
T
= (1 + αT)T, T = (1 − αT
)T
.

If V, V
is the radial velocity of the spacecraft with respect to the origin,

then to the first order in α,

V
= V + αR, V = V
− αR
. (14.3.21e,f)

In the Pioneer experiments systematic differences were found between

the observed Doppler shifts and the shifts that were expected. These anoma-
lous results have since been denoted the “Pioneer Anomaly.”
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 427

To understand the origin of this anomaly it is necessary to rather care-

fully investigate how it arose. This first become possible in early 2010, when
an extensive report on the Pioneer program was put on the internet22 . The
following discussion benefits from additional information kindly supplied by
Toth, one of its authors.23 In the Pioneer program, pulses of radar waves
of fixed frequency were sent from sources on earth to the receding Pioneer
10 and 11 spacecraft, which transmitted back to earth an amplified signal
that had been phase locked at the spacecraft to the incoming signal it had
received. When these wave pulses arrived back at their source they were
mixed with a wave of fixed frequency. The beat frequency this produced
was the frequency shift developed in the wave during its journey.
At the time the experiments were carried out it was not possible to
directly measure the distance to the spacecraft at the moment it receives
the pulse sent back to earth. The time and distance had to be estimated by
comparing the measured Doppler frequency with that produced by a sophis-
ticated dynamical model solving the equations of motion for the spacecraft,
given the forces acting upon it. The model was first supplied with initial
values, t0 , r0 , v0 . It then predicted the positions r(t) and velocities v(t)
of the spacecraft, and the Doppler shift that would result at time t. The
observed frequency shift was carefully matched with the predicted Doppler
frequency to obtain the best estimate to the modeled value of v, and hence
t, and r. The new values of r(t), v(t) were then evolved by the equations
of motion, which were thereby expected to predict values of subsequent
Doppler shifts that would better match the next observed Doppler shift.
This matching and revision was repeated until the spacecraft reached a
distance of about 70 AU and the signals became too weak to use. It was
found that, as the distance, r, to the spacecraft increased beyond 20 AU,
the Doppler shift corrections required by the matching process were small
frequency shifts ∆ν. These frequency shifts appeared to correspond to the
blueshift that one would obtain from anomalous velocity contributions

v = −(2.80 ± 0.42) × 10−18 sec−1 r. (14.3.22)

However, in developing the dynamical model, it was presumed that when

strong gravitational fields were not present, both the spacecraft and the
radar waves moved in a space with Minkowski metric, a space in which
the coordinates r, t, v are those that apply. The Doppler matching pro-
cedure used to modify the model, while enabling it to continue supposing
that the measurements were made in a Minkowski space, transferred to
the model the functional dependence on the r
, t
, v
coordinates that arises
October 28, 2010 16:55 9in x 6in b943-ch14

428 Dynamical Symmetry

if the metric in which the waves move is the special conformal metric of
(14.3.12). To understand the consequences it is helpful to reorganize Hill’s
approximate equation, which for V  c, can be written as


VC − VM = αR. (14.3.23a)

As the distance R
traversed by the radar waves was twice the distance r
in (14.3.22), in Eq. (14.3.23c) one has −2αr = −(2.8 ± 0.42) × 10−18 sec−1
r, so α = (1.4 ± 0.42) × 10−18 sec−1 . This converts (14.3.23a) to



VM − VC = αR
, i.e. VC


− VM = −αR
. (14.3.23b,c)

As R
differs from R only by terms of O(α2 ) in these relations R
may

be replaced by R. This has the consequence that sets −α to have the
anomalous reported value −(2.80 ± 0.42) × 10−18 sec−1 . Thus the anoma-
lous Pioneer blueshift corrections arise from a physical redshift of radar
wavelengths. This shift has the magnitude, and the dependence on R that
develop if the waves move in a space with special conformal metric that
has the group parameter α approximately equal to 2/3 Hubble’s constant
H0 . This explanation of the Pioneer anomaly was anticipated by Tomilchik
before a detailed description of the modeling process was available.22 How-
ever, he concluded that the group parameter is actually Hubble’s constant.
Though the value of α deduced from the Pioneer observations is cur-
rently close to 2/3 the value (2.19 ± 0.56)× 10−18sec−1 assigned to Hubble’s
constant, this relation between α and H0 should not be assumed because
the Pioneer matching algorithms have altered several parameters in the
dynamical model, and at least one fixed parameter may have been assigned
an incorrect value.23 Moreover, in the actual matching process other param-
eters in the dynamical model were also altered. The whole process that
uncovered the Pioneer anomaly is currently the subject of a series of fur-
ther investigations which could have the effect of significantly altering the
numerical value of α.23,24
It appears that spacecraft similar to Pioneer 10 and Pioneer 11 could
provide quite accurate values of all the variables required to directly deter-
mine the group parameter α, and so determine the spacetime metric in
whatever distant regions they might visit.20 To do so, the spacecraft should,
like the Pioneers, be equipped with a system to receive radar pulses from
ground stations and return them as amplified pulses that have been phase-
locked to the received pulses. If both the Doppler shifts, and time-delays
between emission and receipt of successive pulses are measured, the neces-
sary data are provided. Repeaters installed on the outer planets and their
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 429

moons might also be able to provide useful restrictions on the value of α,

and hence the metric.
If it is established that α has a nonzero value, it will be necessary to
reckon with the fact that measurements of wavelengths are measurements of
special conformal wavelengths λ
, and that the currently accepted metrics
must be changed to special conformal ones. A few of its consequences are
suggested by considering the way in which it would affect the interpretations
of Hubble’s equation (14.3.3b). In doing so it is helpful to replace V by V
,
R by R
, and set H0 = h0 + α. The value of h0 will then determine the uni-
versal rate of expansion previously determined by H0 . V
and R
will become
the special conformal analogs of V and R.

Conclusion
Few physical implications of Lie’s discoveries were noticed prior to the dis-
coveries of Bateman and Cunningham. Hill seems to have been one of the
first to investigate possible physical consequences of their conformal group.
The increasing interest in the role of group theory in quantum mechan-
ics that began in the 1960’s led to an increasing recognition of possible
consequences of the conformal symmetry of Maxwell’s equations at the
submicroscopic level. Some applications of the conformal group in quan-
tum theory abandon the requirement of Lorentz invariance on the submi-
croscopic scale. Hill’s work and the Pioneer results suggest that Poincare
invariance of electromagnetic phenomena on the supermacroscopic scale
should not be assumed. Though no conclusions as startling as that of Hill
have emerged, a number of workers have investigated relativistic and non-
relativistic consequences of conformal invariance.25 If further determina-
tions of the conformal group parameter α show that it is nonzero, much
physics will have to be revised. If it is found that α has a value of the same
order of magnitude as Hubble’s constant, the consequences for physics and
cosmology will be even more extensive.
Over a century has passed since Bateman and Cunningham established
the conformal invariance group of Maxwell’s equations. However, as has
become evident in this chapter, many fundamental physical implications of
this invariance have yet to be established.

Exercises
1. Use the extended generators of the Poincare group to show that the
group leaves the Lorentz metric (14.2.4a) invariant.
October 28, 2010 16:55 9in x 6in b943-ch14

430 Dynamical Symmetry

2. Determine whether or not the Lie subgroup of the Bateman–

Cunningham conformal group leaves invariant the metric (14.2.4c).
3. Use the methods of Chapter 4 to convert the generators uj = ∂/∂xj , of
translations in spacetime, to Lie generators Uj = Σk ξjk (X)∂/∂Xk that
act on functions of the inverted variables X.
4. Do the same for the six generators of Lorentz transformations.
5. Show that the conformal Lie group with generators (14.2.5) is locally
isomorphic to SO(4,2).
6. Determine the effect of finite special conformal transformations on
straight-lines in spacetime, and suppose that the lines represent tra-
jectories of a particle. Compare the physical interpretations of your
results when you have independent evidence that a force is acting, and
when no such evidence exists.
7. Integrate the infinitesimal transformation of x4 = ct and r generated
by Hill’s X14 , and determine its invariant function. Use Hill’s physical
interpretation of the transformation to give a physical interpretation to
this function.
8. In a series of articles26 Hill developed a generalization of special relativ-
ity by considering that coordinate systems moving at constant relative
velocity are special cases of coordinate systems moving at constant
relative acceleration. Can you detect any errors in Hill’s analysis in
Ref. 26a? What conclusions can be drawn about interpretations of New-
ton’s laws? Cf. Kastrup’s review.15
9. The action of exp(βC4 ) on a Hamiltonian H converts it to
H
= H + 2(α/c)[C4 , H] + O((α/c)2 ).
(a) Show that to the first-order in α, a radial potential r−n in H under-
goes the dilatation
r−n → (1 − 4(α/c)n)r−n .
(b) Show that to the first-order in α, the Laplacian operator ∆ under-
goes the conversion
∆ → (1 − 8αt)∆ − (4α/c2 )(3 + 2r∂/∂r)∂/∂t,
and that
∂/∂t → ∂/∂t − 4α(t∂/∂t + r∂/∂r).
10. Determine, to the first-order in β, the effect of exp(βC4 ) on the ener-
gies, transition frequencies of H atoms, and the evolution of its wave
functions.
October 28, 2010 16:55 9in x 6in b943-ch14

Dynamical Symmetry of Maxwell’s Equations 431

11. Generalize C4 to a C
4 that contains an additional term which acts on

the mass M of the proton in a hydrogen atom such that exp(βC
4 ) will
leave invariant the Schrödinger equations governing the motion of an
electron with reduced mass µ.

References
[1] A. Einstein, Ann. Phys., 17 (1905) 891.
[2] H. Lorentz, Proc. Amstrdam Acad. 6 (1904) 809.
[3] (a) E. Cunningham, Proc. London Math Soc. 8(2) (1909) 77;
(b) H. Bateman, ibid. 223;
(c) H. Bateman, Proc. London Math Soc. 7(2) (1909) 70.
[4] J. C. Maxwell, Electricity and Magnetism (Cambridge University Press,
Cambridge, 1873).
[5] a) W. F. Magie, Source Book in Physics (McGraw-Hill, N.Y., 1935), p. 408;
(b) E. Whittaker, A History of the Theories of Aether and Electricity
(Thomas Nelson and Sons, Edinburgh, 1951), p. 83.
[6] W. D. Niven, ed., The Scientific Papers of James Clerk Maxwell (Cambridge
University Press, Cambridge, 1880); photographic reprint (J. Hermann,
Paris), (Steichert, N.Y., 1930).
[7] L. P. Wheeler, Josiah Willard Gibbs (Yale University Press, New Haven,
1952).
[8] J. A. Kong, Electromagnetic Wave Theory, 2nd edn. (Wiley Interscience,
New York, 1990).
[9] J. C. Maxwell, Trans. Royal. Soc. (London) CLV, (1864), Collected works,
XXV, A dynamical theory of the electromagnetic field, Part I, Sec. (20).
[10] S. Lie, Transformations Grupen, III (Leipzig, 1893), p. 351; reprinted
(Chelsea, New York, N.Y., 1970).
[11] B. G. Wybourne, Classical Groups for Physicists, (Wiley, New York, 1974),
pp. 345–7.
[12] (a) A. O. Barut and R. Raczka, Theory of Group Representations and Appli-
cations, 2nd revised edn. (World Scientific, Singapore, 1986);
(b) A. O. Barut and R. B. Haugen, Ann. Phys. 71 (1972) 519.
[13] T. Fulton, F. Rohrlich and L. Witten, Rev. Mod. Phys. 34 (1962) 442.
[14] C. Wulfman, J. Phys. A: Math. Theor. 42 (2009) 185301.
[15] H. A. Kastrup, Ann. Phys. 17 (2008) 631.
[16] M. Born, Einstein’s Theory of Relativity, translated by H. Brose (Methuen,
London, 1924), p. 240.
[17] E. Hubble, The Realm of the Nebulae (Yale University Press, New Haven,
1936).
[18] (a) P. J. E. Peebles, Principles of Physical Cosmology (Princeton, 1993),
pp. 24, 71, 82;
(b) ibid., pp. 20, 106–7;
(c) ibid., pp. 73, 4.
[19] E. L. Hill, Phys. Rev. 68 (1945), 235.
[20] C. Wulfman, [arXiv: 1003.0507]; [arXiv: 1010.2139] (2010).
October 28, 2010 16:55 9in x 6in b943-ch14

432 Dynamical Symmetry

[21] J. D. Anderson et al., Phys. Rev. D 65 (2005) 082004.

[22] L. M. Tomilchik, AIP Conf. Proc. 1205 (2010) 177.
[23] S. G. Turyshev and V. T. Toth, [arXiv:1001.3686v].
[24] V. T. Toth, private communication.
[25] (a) F. Hoyle, G. Burbidge and N. Narlikar, A Different Approach to
Cosmology (Cambridge University Press, Cambridge, 2000).
(b) E. F. Bunn and D. W. Hogg, Am. J. Phys. 77 (2009).
[26] (a) E. L. Hill, Phys. Rev. 67 (1945) 358;
(b) E. L. Hill, Phys. Rev. 72 (1947) 143;
(c) E. L. Hill, Phys. Rev. 94 (1951) 1165.
October 28, 2010 16:55 9in x 6in b943-Index

Index

Abelian group, 202 commutation relations

subalgebra, 179 Bateman–Cunningham
abstract Lie groups, 170, 193–195 conformal group, 409
accidental degeneracy, 5, 11 defining continuous spectra, 306
action, 84 defining discrete spectra, 308
active interpretation of invariance group of the
transformations, 30–35 free-particle Schrödinger
adjoint group, 192 equation, 293
adjoint matrix, 276 invariant under diffeomorphisms,
affine transformation, 31 49
Aharonov–Bohm phase, 186 PB operators of Kepler
angular momentum constants of motion, 257
shift algebra, 315 projective group, 182
angular momentum in phase space, SO(2,1) angular momentum
217 shift algebra, 317
attractor, 128 SO(3,1), 257
autonomous extension, 129 SO(4), 256, 354
SO(4,2) H atom dynamical
Baker–Campbell–Hausdorff relations, group, 324, 335
177 Sp(2R), 184
Balmer formula, 361 SU(1,1) Morse oscillator shift
Bateman–Cunningham conformal algebra, 322
group, 413 compact Lie group, 181
benzoquinone completion of direct-product
anomalous dipole moment, 385 group SO(2) ⊗ SO(2), 335
Berry phase, 186 of SO(4,2) ⊗ SO(4,2), 376
complex Lie algebra, 187
canonical transformation, 93 configuration mixing within valence
Cartan–Killing form, 188 shells of electrons, 375
Casimir operators conformal finite transformations, 419
SO(3), 191 conformal group of
SO(4), 191 euclidean four space, 413
SO(4,2), 340 spacetime, 413
center of a Lie algebra, 202 conformal spacetime metric, 417, 423

433
October 28, 2010 16:55 9in x 6in b943-Index

434 Dynamical Symmetry

congruent, 1 electromagnetic field

correspondence principle, 276 action of transformation,
Lie algebraic extension, 281 413–417
cosmology, 421, 422, 429 energy density, 412
coupling of angular momenta vectors, 411
effect on L1 · L2 , 329 entropy and degeneracy, 10
crossings, 364, 367 ethylenes
twisted, 365
Davydov excitons, 11 Euclidean geometry, 2, 48
degeneracy, 10–12 Euclidean symmetry, 26–28, 50
degeneracy group Eulenbeck and Goudsmit’s discovery
of classical Kepler motion, of spin, 4
256–258, 263 extended generators, 152, 154
of Kepler motions, 269
degeneracy group of hydrogen, Fock variables, 258, 263, 272, 346
340–342, 349, 354, 355 Fock’s Discovery, 6
degeneracy of isotropic harmonic Bargmann’s Lie algebraic
oscillators, 309, 310 explanation of, 7
derivative property, 116 forbidden crossings, 368
determining equations, 19
Fourier transforms, 345
for Schrödinger equations for
functional dependence in the theory
rigid rotor, 314
of ODEs, 72, 73
for Schrödinger equations of
functional dependence in the theory
free-particle, 290
of PDEs, 288
for Schrödinger equations of
harmonic oscillator, 297
Galilei invariance, 208, 230
for the Laplace equation, 315
Gegenbauer polynomials, 351
degeneracy group of hydrogen, 349
shift operators, 356
diatomic molecules
global isomorphism, 181
one electron, 364
diffeomorphism, 74, 76
differential consequence, 288 H-like atoms
direct sum, 202 regularized, 337, 338, 340–344
direct-product harmonic oscillator
completion, 329 isotropic two-dimensional, 222
constrained, 332 Hartree–Fock
two different SO(4), 369–373 failure of calculations, 11
Divine Harmony, 2 symmetry breaking, 11
Doppler equation, 421 Heisenberg algebras, 278
doubly excited states of He-like Helmholtz’s studies of human vision,
atoms, 373 42
Henon–Heiles system, 166
E(n), 62 Hermitian matrix, 276
Einstein’s theory of special relativity, hidden symmetry, 9, 10
3, 42, 409 Hill’s discovery, 422
October 28, 2010 16:55 9in x 6in b943-Index

Index 435

hodograph Levi–Malcev theorem, 203

Kepler, 252 Lie
homeomorphism, 77 fundamental theorems, 198–202
Hubble’s relations, 421, 422, 428, 429 Lie algebra
hyperbolic rotations, 33 definition, 173, 178
hyperspherical harmonics proper ideal, 203
N-shift operators, 356 rank, 178
semi-simple, 203
ideal semidirection sum, 203
Lie algebraic, 203 solvable, 203
independent variables in differential Lie group
equations, 72, 73 definition, 170
integrating factors, 134 generators, 173
intrinsic symmetry, 102 global, 181, 184, 185
invariance local, 181, 184, 185
differential forms, 143 light cone, 424
equations, 137
limit cycle, 128
functions, 135
locally isomorphic, 181
metric, 144
Lorentz group, 412
ODE, 143–150
Lorentz transformations, 42–44, 46,
PDE, 283, 285–289
47, 413
work, 143
invariant point, 115
many-electron atoms, 370–377
inversion
mapping, 35, 36
spacetime, 414–416
matrix representations
irreducible representation, 326
irreducible, 326
ISO(m,n−m), 64
ISO(n), 62 of group generators, 325
of group operators, 325
Jacobi relation, 173 unitary, 326
Jahn–Teller Theorem, 366 Maxwell’s Equations, 3, 409–418
extended, 367 action of Lorentz
transformation, 413
Kepler’s laws, 248 effects of inverse transformation,
Keplerian motion 415
classical, 247 Maxwell–Boltzmann distribution law,
quantum mechanical, 322–325, 3
337 metric, 40, 48, 50, 51, 61
Killing form, 204, 205 metric governing propagation of EM
waves
labeling operators, 188, 311 determination, 428, 429
Lagrangian function Michelson–Morley experiments, 3
of classical mechanics, 242, Minkowski metric, 423–425
244–246 Minkowski’s trick, 65, 413
Laplacian Moebius band, 39
conformal invariance, 318 molecular orbitals, 11
October 28, 2010 16:55 9in x 6in b943-Index

436 Dynamical Symmetry

momentum-space orbitals of real Lie algebra, 177, 178

molecules, 377–383 realizations of a group, 181
momentum-space representation for resonance, 11
atoms, 344–349 Robertson–Walker metric, 422, 423
Morse oscillator, 320, 390 rotation
physical interpretations of, 37–39
non-crossing rule rotation groups, 62, 63
Wigner–von Neumann, 364, 368 Runge–Lenz vector, 250
noncompact Lie group, 181
nonlinear transformations, 56–62, 97 scalar product
Schrödinger, 276
O(m,n−m), 64 spacetime, 64
O(n), 62 self-adjoint matrix, 276
one-point compactification, 183 semi-simple Lie group, 203
ordering ambiguities in semidirect product, 203
correspondence relations, 280 semidirect sum, 203
ordinary differential equations, shifts in stellar spectra, 420–429
fundamental theorem of, 124 skew-adjoint matrix, 277
orthogonal group, 62 SO(2,1), 174
SO(3) ⊗ SO(3), 328–334
passive interpretation of SO(3), 173, 174, 285
transformations, 30–35 SO(3,1), 257
period-doubling in periodic charts, SO(3,2), 314
375–377 SO(4) ⊗ SO(4), 369
Pfaffian equation, 143 SO(4), 257, 263, 354
Pfaffian form, 143 SO(4,1), 271
phase shift, 186 SO(4,2), 338–340, 355
phase space SO(m,n−m), 64
definition, 8, 215 SO(n), 62
Pioneer anomaly, 427, 428 solvable Lie group, 203
Pioneer spacecraft experiments, Sophus Lie
426–428 development of his ideas, 18
Poeschl–Teller oscillator, 405 Sp(2n, 2) subgroup chains, 237, 238
Poincare group, 49, 51–56 special conformal metric of
point transformation, 147 spacetime, 423
Poisson bracket, 77 special functions, 191
definition, 211 special relativity
Lie algebras, 217 symmetry, 51–56
operators, 217 square
PQ space, 220 symmetry group of, 24–28
PQET space, 230 stability of planetary motion, 166
predator–prey relations, 126–128 stereographic projection, 6, 7, 56–62,
projective transformation, 99, 100 66–68, 182, 183, 347–349
propagation of EM waves, 423, 425 strategy for dealing with
proper ideal, 203 perturbations, 369
punctuated evolution, 17 structure constants, 178
October 28, 2010 16:55 9in x 6in b943-Index

Index 437

Sturmians, 340, 349, 350, 380–384 tangent vector, 111, 112, 115, 119
sudden polarization, 366 topology, 132
superconductivity, 15 triatomic molecules, 367
superposition invariance, 275
symmetry breaking and creation U(4) ⊗ U(4) subgroup chains, 404
Bernard convection, 16 U(4) subgroup chains, 401
cell differentiation, 15 unit circle
crystallography, 15 invariance transformation, 137
crystals, 15 united atom and separated atom
fluid flow, 16 correlation diagram, 367
geology, 16
Hartree–Fock calculations, 11 vector, 62, 63
reaction–diffusion equation, 16 vector fields, 110–112
superconductivity, 11 vibron models, 389
symmetry in phase space, 92–95
symplectic diffeomorphism, 76 Walsh’s rules, 368
symplectic space, 82, 91 wavelength
symplectic transformations, 80–84, conformal, 425
86–90, 95–97, 210–214 wedge product, 86
 Whenever systems are governed by continuous chains of causes
and effects, their behavior exhibits the consequences of dynamical
symmetries, many of them far from obvious. Dynamical Symmetry
introduces the reader to Sophus Lie’s discoveries of the connections
between differential equations and continuous groups that underlie
this observation. It develops and applies the mathematical relations
between dynamics and geometry that result. Systematic methods
for uncovering dynamical symmetries are described, and put to
use. Much material in the book is new and some has only recently
appeared in research journals.

Though Lie groups play a key role in elementary particle physics,

their connection with differential equations is more often exploited
in applied mathematics and engineering. Dynamical Symmetry
bridges this gap in a novel manner designed to help readers
establish new connections in their own areas of interest. Emphasis
is placed on applications to physics and chemistry. Applications to
many of the other sciences illustrate both general principles and
the ubiquitousness of dynamical symmetries.
Wu l f m a n
Carl E Wulfman

World Scientific
ISBN-13 978-981-4291-36-1
ISBN-10 981-4291-36-6

www.worldscientific.com ,!7IJ8B4-cjbdgb! World Scientific

7548 hc

7548.Yee Sern.CC5.indd 1 11/1/10 3:13 PM