schwinger j
TRANSCRIPT
QuantumKinematicsand Dynamics
This page intentionally left blank
QuantumKinematics
and Dynamics
JULIAN SCHWINGERUniversity of California, Los Angeles
This book was originally published as part of theFrontiers in Physics Series, edited by David Pines,
Advanced Book Program
A Member of the Perseus Books Group
WesternPRESS
Many of the designations used by manufacturers and sellers to distinguishthier products are claimed astrademarks. Where those designations appear in this book and Perseus Books Group was aware of a trademarkclaim, the designations have been printed in initial capital tetters.
Copyright © 1991, 2000 by Westview Press
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmit-ted! in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, withoutthe prior written permission of the publisher. Printed in the United States of Amenca.
A CIP catalog record for this book is available from the Library of Congress.ISBN 0-7382-0303-3
Westview Press is a Member of the Perseus Books Group
Perseus Publishing books are available at special discounts for bulk purchases in the U.S. by corporations,institutions, and other organizations. For more information, please contact the Special Markets Depart-ment at HarperCollins Publishers, 10 East 53rd Street, New York, NY 10022, or call 1-212-207-7528.
2 3 4 5 6 7 8 9 10
ADVANCED BOOK CLASSICS
David Pines, Series Editor
Anderson, P.W., Basic Notions of Condensed Matter PhysicsBethe H, and Jackiw, R,, Intermediate Quantum Mechanics, Third EditionCowan, G. and Pines, D., Compkxity: Metaphors, Models, and Realityde Germes, P.G., Superconductivity of Metals and Alloysd'Bspagnat, B., Conceptual Foundations of Quantum Mechanics, Second EditionFeynman, R., Photon-Hadron InteractionsFeynman, R., Quantum ElectrodynamicsFeynman, R., Statistical MechanicsFeynman, R., The Theory of Fundamental ProcessesGeH-Mann, M. and Ne'eman, Y., The Eightfold WayKhalatnikov, I. M. An Introduction to the Theory of SuperfluidityMa, S-K., Modern Theory of Critical PhenomenaMigdal, A. B., Qualitative Methods in Quantum TheoryNegele, J. W, and Orland, H,, Quantum Many-Particle SystemsNozieres, P., Theory of Interacting Fermi SystemsNozieres, P. and Pines, D., The Theory of Quantum LiquidsParisi, G., Statistical Field TheoryPines, D., Elementary Excitations in SolidsPines, D., The Many-Body ProblemQuigg, C, Gauge Theories of the Strong, Weak, and Electromagnetic InteractionsSchrieffer, J.R., Theory of Superconductivity, RevisedSchwinger, J-, Particles, Sources, and Fields, Volume ISchwinger, J., Particles, Sources, and Fields, Volume IISehwinger, J., Particles, Sources, and Fields, Volume IIISchwinger, J., Quantum Kinematics and DynamicsWyld, H.W., Mathematical Methods for Physics
This page intentionally left blank
Editor's Foreword
Perseus Publishing's Frontiers in Physics series has, since 1961, madeit possible for leading physicists to communicate in coherent fashiontheir views of recent developments in the most exciting and activefields of physics—without having to devote the time and energyrequired to prepare a formal review or monograph. Indeed,throughout its nearly forty year existence, the series has emphasizedinformality in both style and content, as well as pedagogical clarity.Over time, it was expected that these informal accounts would bereplaced by more formal counterparts—textbooks or monographs—as the cutting-edge topics they treated gradually became integratedinto the body of physics knowledge and reader interest dwindled.However, this has not proven to be the case for a number of thevolumes in the series: Many works have remained in print on an on-dernand basis, while others have such intrinsic value that the physicscommunity has urged us to extend their life span,
The Advanced Book Classics series has been designed to meet thisdemand. It will keep in print those volumes in Frontiers in Physicsthat continue to provide a unique account of a topic of lastinginterest And through a sizable printing, these classics will be madeavailable at a comparatively modest cost to the reader.
The late Nobel Laureate Julian Schwinger was not only one of thegreat theoretical physicists of our time, but also one of the greatpedagogues of the past century. His lectures were legendary for theiralmost unique combination of clarity and elegance. I am accordinglyvery pleased that the publication in Advanced Book Classics ofQuantum Kinematics and Dynamics will continue to make his lectureson this topic readily accessible to future generations of the scientificcommunity.
David PinesCambridge, EnglandMay, 2000
This page intentionally left blank
Vita
Julian SchwingerUniversity Professor, University of California, and Professor of Physics at the Universityof California, Los Angeles since 1972, was born in New York City on February 12,1918.Professor Schwinger obtained his Ph.D. in physics from Columbia University in 1939. Hehas also received honorary doctorates in science from five institutions: Purdue University(1961), Harvard University (1962), Brandeis University (1973), Gustavus AdolphusCollege (1975), and the University of Paris (1990). In addition to teaching at theUniversity of California, Professor Schwinger has taught at Purdue University (1941 -43),and at Harvard University (1945-72). Dr, Schwinger was a Research Associate at theUniversity of California, Berkeley, and a Staff Member of the Massachusetts Institute ofTechnology Radiation Laboratory. In 1965 Professor Schwinger became a co-recipient(with Richard Feynman and Sin Itiro Tomonaga) of the Nobel Prize in Physics for workin quantum electrodynamics. A National Research Foundation Fellow (1939-40) and aGuggenheim Fellow (1970), Professor Schwinger was also the recipient of the C. L.Mayer Nature of Light Award (1949); the First Einstein Prize Award (1951); a J. W. GibbsHonorary Lecturer of the American Mathematical Society (1960); the National Medal ofScience Award for Physics (1964); a HumboJdt Award (1981); the Premio Citta diCastiglione de Sicilia (1986); the Monie A. Ferst Sigma Xi Award (1986); and theAmerican Academy of Achievement Award (1987).
This page intentionally left blank
Special Preface
The first two chapters of this book are devoted to Quantum Kinematics. In 1985 I had theopportunity to review that development in connection with the celebration of the 100thanniversary of Hermann Weyl's birthday, (See the last footnote of Chapter 2.) Inpresenting my lecture (Hermann Weyl and Quantum Kinematics, in Exact Sciences andTheir Philosophical Foundations, Verlag Peter Lang, Frankfurt am Main, 1988, pp. 107-129), I felt the need to alter only one thing: the notation. Lest one think this rather trivial,recall that the ultimate abandonment, early in the 19th century, of Newton's method offluxions in favor of the Leibnitzian calculus, stemmed from the greater flexibility of thelatter's notation.
Instead of the symbol of measurement: M(a', b'), I now write: I a'b'l , combiningreference to what is selected and what is produced, with an indication that the act ofmeasurement has a beginning and an end. Then, with the conceptual analysis of I a'b' Iinto two stages, one of annihilation and one of creation, as symbolized by
the fictitious null state, and the symbols HP and <fy can be discarded.As for Quantum Dynamics, I have long regretted that these chapters did not contain
numerous examples of the practical use of the Quantum Action Principle in solvingphysical problems. Perhaps that can be remedied in another book, on Quantum Mechan-ics. There is, however, a cornucopia of action applications in the non-operator context ofSource Theory. (See Particles, Sources, and Fields, Vols. I, 2, 3, Advanced BookClassics.)
Los Angeles, California J. S.April 1991
This page intentionally left blank
Foreword
Early in 1955 I began to write an article on the Quantum Theory of Fields. The introductioncontained this description of its plan. "In part A of this article a general scheme of quantumkinematics and dynamics is developed within the nonrelativistic framework appropriateto systems with a finite number of dynamical variables. Apart from specific physicalconsequences of the rclativistic invariance requirement, the extension to fields in part Bintroduces relatively little that is novel, which permits the major mathematical features ofthe theory of fields to be discussed in the context of more elementary physical systems."
A preliminary and incomplete version of part A was used as the basis of lecturesdelivered in July, 1955 at the Les Houches Summer School of Theoretical Physics. Workon part A ceased later that year and part B was never begun. Several years after, I usedsome of the material in a series of notes published in the Proceedings of the NationalAcademy of Sciences. And there the matter rested until, quite recently, Robert Kohler(State University College at Buffalo) reminded me of the continuing utility of the LesHouches notes and suggested their publication. He also volunteered to assist in thisprocess. Here is the result. The main text is the original and still incomplete 1955manuscript, modified only by the addition of subheadings. To it is appended excerpts fromthe Proc. Nat. Acad. of Sciences articles that supplement the text, together with two papersthat illustrate and further develop its methods.
JULIAN SCHWINGERBelmont, Massachusetts1969
This page intentionally left blank
Contents
I The Algebra of Measurement 1
I.1 Measurement Symbols 21.2 Compatible Properties. Definition of State 51.3 Measurements that Change the State 71.4 Transformation Functions 91.5 The Trace 121.6 Statistical Interpretation 141.7 The Adjoint 171.8 Complex Conjugate Algebra 191.9 Matrices 19
1.10 Variations of Transformation Functions 221.11 Expectation Value 251.12 Addendum: Non-Selective Measurements 26
II The Geometry of States 29
2.1 The Null State 292.2 Reconstruction of the Measurement Algebra 322.3 Vector Algebra 352.4 Wave Functions 372.5 Unitary Transformations 402.6 Infinitesimal Unitary Transformations 442.7 Successive Unitary Transformations 462.8 Unitary Transformation Groups. Translation and Rotations 482.9 Reflections S3
2.10 Continuous Spectra 54
x.v
2
2.11 Addendum: Operator Space 562.12 Addendum: Unitary Operator Bases 62
III The Dynamical Principle 73
3.1 The Action Operator 743.2 Lagrangian Operator 763.3 Stationary Action Principle 773.4 The Hamiltonian Operator 793.5 Equations of Motion. Generators 803.6 Commutation Relations 833.7 The Two Classes of Dynamical Variables 863.8 Complementary Variables of the First Kind 973.9 Non-Hermitian Variables of the First Kind 103
3.10 Complementary Variables of the Second Kind 106
IV The Special Canonical Group 113
I. VARIABLES OF THE FIRST KIND 1144.1 Differential Operators 1154.2 Schrodinger Equations 1194.3 The q p Transformation Function 1204.4 Differential Statements of Completeness 1224.5 Non-Hermitian Canonical Variables 1254.6 Some Transformation Functions 1264.7 Physical Interpretation 1304.8 Composition by Contour Integration 1334.9 Measurements of Optimum Compatibility 140
II. VARIABLES OF THE SECOND KIND 1434.10 Rotation Group 1434.11 External Algebra 1454.12 Eigenvectors and Eigenvalues 148
III. UNIFICATION OF THE VARIABLES 1524.13 Constructive Use of the Special Canonical Group 1524.14 Transformation Functions 1564.15 Integration 1664.16 Differential Realizations 170
V Canonical Transformations 173
5 1 Group Properties and Superfluous Variables 1755.2 Infinitesimal Canonical Transformations 178
Contentsxvi
5.3 Rotations. Angular Momentum 1825.4 Translations. Linear Momentum 1855.5 Transformation Parameters 1875.6 Hamilton-Jacobi Transformation 1905.7 Path Dependence 1915.8 Path Independence 1945.9 Linear Transformations 195
VI Groups of Transformations 201
6.1 Integrability Conditions 2026.2 Finite Matrix Representation 2046.3 Subgroups 2076.4 Differential Forms and Composition Properties 2096.5 Canonical Parameters 2116.6 An Example, Special Canonical Group 2166.7 Other Parameters. Rotation Group 2196.8 Differential Operator Realizations 2266.9 Group Volume 228
6.10 Compact Groups 2316.11 Projection Operators and Invariants 2336.12 Differential Operators and the Rotation Group 2386.13 Non-Compact Group Integration 2436.14 Variables of the Second Kind 2476.15 Reflection Operator 2496.16 Finite Operator Basis 2506.17 Addendum: Derivation of the Action Principle 2546.18 Addendum Concerning the Special Canonical Group 2596.19 Addendum: Quantum Variables and the Action Principle 275
VII Canonical Transformation Functions 285
7.1 Ordered Action Operator 2857.2 Infinitesimal Canonical Transformation Functions 2877.3 Finite Canonical Transformation Functions 2937.4 Ordered Operators. The Use of Canonical
Transformation Functions 2977.5 An Example 2997.6 Ordered Operators and Perturbation Theory 3027.7 Use of The Special Canonical Group 3067.8 Variational Derivatives 3097.9 Interaction of Two Sub-Systems 317
7.10 Addendum: Exterior Algebra and the Action Principle 321
Contents xvii
xviii Contents
VIII Green's Functions 331
8.1 Incorporation of Initial Conditions 3318.2 Conservative Systems. Transforms 3358.3 Operator Function of a Complex Variable 3378.4 Singularities 3408.5 An Example 3418.6 Partial Green's Function 343
IX Some Applications And Further Developments 347
9.1 Brownian Motion of a Quantum Oscillator 3479.2 Coulomb Green's Function 374
CHAPTER ONETHE ALGEBRA OF
1.1 Measurement Symbols 21.2 Compatible Properties. Definition of 5
State1.3 Measurements that Change the State 71.4 Transformation Functions 91.5 The Trace 121.6 Statistical Interpretation 141.7 The Adjoint 171.8 Complex Conjugate Algebra 191.9 Matrices 191.10 Variations of Transformation Functions 221.11 Expectation Value 251.12 addendum: Non-Selective Measurements 26
The classical theory of measurement is built
upon the conception of an interaction between
the system of interest and the measuring apparatus
that can be made arbitrarily small, or at least
precisely compensated, so that one can speak mean-
ingfully of an idealized measurement that disturbs
1
MEASUREMENT
2 QUANTUM KINEMATICS AND DYNAMICS
no property of the system. But it Is character-
istic of atomic phenomena that the interaction
between system and instrument is not arbitarily
small. Nor can the disturbance produced by the
interaction be compensated precisely since to some
extent it is uncontrollable and unpredictable.
Accordingly, a measurement on one property can
produce unavoidable changes in the value previously
assigned to another property, and it is without
meaning to speak of a microscopic system possessing
precise values for all its attributes. This con-
tradicts the classical representation of all phys-
ical quantities by numbers. The laws of atomic
physics must be expressed, therefore, in a non-
classical mathematical language that constitutes
a symbolic expression of the properties of micro-
scopic measurement.
1.1 MEASUREMENT SYMBOLS
We shall develop the outlines of this math-
ematical structure by discussing simplified phys-
ical systems which are such that any physical quan-
tity A assumes only a finite number of distinct
THE ALGEBRA OF MEASUREMENT 3
values, a',..aM. In the most elementary type of
measurement, an ensemble of independent similar
systems is sorted by the apparatus into subensembles,
distinguished by definite values of the physical
quantity being measured. Let M(a') symbolize the
selective measurement that accepts systems possess-
ing the value a1 of property A and rejects all
others. We define the addition of such symbols to
signify less specific selective measurements that
produce a subensemble associated with any of the
values in the summation, none of these being dis-
tinguished by the measurement.
The multiplication of the measurement symbols
represents the successive performance of measure-
ments (read from right to left). It follows from
the physical meaning of these operations that add-
ition is commutative and associative, while multi-
plication is associative. With 1 and 0 symbol-
izing the measurements that, respectively, accept
and reject all systems, the properties of the
elementary selective measurements are expressed by
4 QUANTUM KINEMATICS AND DYNAMICS
Indeed, the measurement symbolized by M(a')
accepts every system produced by M(a') and rejects
every system produced by M(a") , a" a* , while
a selective measurement that does not distinguish
any of the possible values of a' is the measure-
ment that accepts all systems.
According to the significance of the measure-
ments denoted as 1 and 0 , these symbols have the
algeoraic properties
and
which justifies the notation. The various properties
of 0, M(a') and 1 are consistent, provided multi-
plication is distributive. Thus,
THE ALGEBRA OF MEASUREMENT
1.2 COMPATIBLE PROPERTIES. DEFINITION OF STATE
Two physical quantities h^ and A2 are
said to be compatible when the measurement of one
does not destroy the knowledge gained by prior
measurement of the other. The selective measure-
ments MCai) and Mia,) / performed in either
order, produce an ensemble of systems for which one
i
can simultaneously assign the values a to A^i
and a2 to A2 . The symbol of this compound
measurement is
5
The introduction of the numbers 1 and 0 as multi-
pliers, with evident definitions, permits the multi-
plication laws of measurement symbols to be combined
in the single statement
where
' '
QUANTUM KINEMATICS AND DYHAMICS
then describes a complete measurement, which is
such that the systems chosen possess definite values
for the maximum number of attributes; any attempt
to determine the value of still another independent
physical quantity will produce uncontrollable
changes in one or more of the previously assigned
values. Thus the optimum state of knowledge con-
cerning a given system is realized by subjecting
it to a complete selective measurement. The
systems admitted by the complete measurement
M(a') are said to be in the state a' . The
symbolic properties of complete measurements are
also given by (1.1), (1.2) and (1.3).
6
By a complete set of compatible physical quantities,
A , ...A , we mean that every pair of these quan-J. J fc
tities is compatible and that no other quantities
exist, apart from functions of the set A , that
are compatible with every member of this set. The
measurement symbol
THE ALGEBRA OF MEASURMENT
1.3 MEASUREMENTS THAT CHANGE THE STATE
A more general type of measurement incorporates
a disturbance that produces a change of state. The
symbol M(a', a") indicates a selective measure-
ment in which systems are accepted only in the
state a" and emerge in the state a1 , The meas-
urement process M(a') is the special case for
which no change of state occurs >
The properties of successive measurements of the
type M(a', a") are symbolized by
for, if a" a"1 , the second stage of the compound
apparatus accepts none of the systems that emerge
from the first stage, while if a" - a11*, all such
systems enter the second stage and the compound
measurement serves to select systems in the state
a'" and produce them in the state a1. Note that
if the two stages are reversed, we have
7
which differs in general from (1.12). Hence the
multiplication of measurement symbols is noncommu-
tative.
The physical quantities contained in one com-
plete set A do not comprise the totality of phy-
sical attributes of the system. One can form other
complete sets, B, C, ..., which are mutually in-
compatible, and for each choice of non-interfering
physical characteristics there is a set of selective
measurements referring to systems in the appropriate
states, M(b', b"), M(c', c"), ... . The most
general selective measurement involves two incompat-
ible sets of properties. We symbolize by M(a', b')
the measurement process that rejects all impinging
systems except those in the state b', and permits
only systems in the state a1 to emerge from the
apparatus. The compound measurement
M(a', b')M(c', d') serves to select systems in the
state d1 and produce them in the state a 1, which
is a selective measurement of the type M(a", d1} .
But, in addition, the first stage supplies systems
in the state c1 while the second stage accepts
8 QUANTUM KINEMATICS AND DYNAMICS
THE ALGEBRA. OF MEASUREMENT 9
only systems in the state b' . The examples of
compound measurements that we have already consid-
ered involve the passage of all systems or no sys-
tems between the two stages, as represented by the
multiplicative numbers 1 and 0. More generally,
measurements of properties B, performed on a sys-
tem in a state c' that refers to properties in-
compatible with B, will yield a statistical dis-
tribution of the possible values. Hence, only a
determinate fraction of the systems emerging from
the first stage will be accepted by the second
stage. We express this by the general multiplication
law
where <b' c')> is a number characterizing the
statistical relation between the states b1 and c'.
In particular,
1.4 TRANSFORMATION FUNCTIONS
Special examples of (1.14) are
10 QUANTUM KINEMATICS AND DYNAMICS
which shows that measurement symbols of one type can
be expressed as a linear combination of the measure-
ment symbols of another type. The general relation
is
and
We infer from the fundamental measurement symbol
property (1.3) that
and similarly
Prom its role in effecting such connections, the
THE ALGEBRA OF MEASUREMENT 11
totality of numbers (a1 b1) is called the trans-
formation function relating the a - and the
b-descriptions, where the phrase "a -description"
signifies the description of a system in terms of
the states produced by selective measurements of
the complete set of compatible physical quantities
A.
A fundamental composition property of trans-
formation functions is obtained on comparing
On identifying the a - and c -descriptions this
becomes
with
namely
12 QUANTUM KINEMATICS AND DYNAMICS
and similarly
which means that N> the total number of states ob-
tained in a complete measurement, is independent of
the particular choice of compatible physical quanta^
ties that are measured. Hence the total number of
can simultaneously assign the values a toA^
Arbitrary numerical multiples of measurement symbols
in additive combination thus form the elements of
a linear algebra of dimensionality N - the
algebra of measurement. The elements of the measure-
ment algebra are called operators.
1.5 THE TRACE
The number { a'|b') can be regarded as a
linear numerical function of the operator M(b'» a').
We call this linear correspondence between operators
and numbers the trace,
As a consequence, we observe that
which verifies the consistency of the definition
(1.27). In particular,
Hence, despite the noncommutativity of multiplication,
the trace of a product of two factors is independent
of the multiplication order. This applies to any
THE ALGEBRA OF MEASUREMENT 13
and observe from the general linear relation (1.20)
that
The trace of a measurement symbol product is
which can be compared with
where the numbers X(a') and X(b') can be given
arbitrary non-zero values. The elementary measure-
ment symbols M(a') and the transformation function
<a'|a"> are left unaltered. In view of this arbi-
trariness, a. transformation function <a'|b'> can-
not, of itself, possess a direct physical inter-
pretation but must enter in some combination that
remains invariant under the substitution (1.34).
14 QUANTUM KINEMATICS AND DYNAMICS
two elements X, Y, of the measurement algebra,
A special example of (1.30) is
1.6 STATISTICAL INTERPRETATION
It should be observed that the general multi-
plication law and the definition of the trace are
preserved if we make the substitutions
THE ALGEBRA OP MEASUREMENT 15
The appropriate basis for the statistical
interpretation of the traasformation function can
be inferred, by a consideration of the sequence
of selective measurements M(b')M{a')M(b'), which
differs from M(b') in virtue of the disturbance
attendant upon the intermediate A-measurement.
Only a fraction of the systems selected in the
initial B-measurenten-t is transmitted through the
complete apparatus. Correspondingly, we have the
symbolic equation
and, for the A-measurement that does not distinguish
among any of the states, there appears
is invariant under the transformation (1.34). If
we perform an A-measureinent that does not distin-
guish between two (or more) states, there is a
related additivity of the numbers p(a', b') ,
where the number
16 QUANTUM KINEMATICS AND DYNAMICS
whence
These properties qualify p(a', b1) for the role of
the probability that one observes the state a1 in a,
measurement performed on a system known to be in
the state b1. But a probability is a real, non-
negative number. Hence we shall impose an admissi-
ble restriction on the numbers appearing in the
measurement algebra, by requiring that <a' b"> and
<b'|a'> form a pair of complex conjugate numbers
To maintain the complex conjugate relation (1.40),
the numbers X(a') of (1.34) must obey
and therefore have the form
for then
THE ALGEBRA OP MEASUREMENT 17
in which the phases ¥»(a') can assume arbitrary
real values.
1.7 THE ADJOINT
Another satisfactory aspect of the probability
formula (1»36) is the symmetry property
Let us recall the arbitrary convention that accom-
panies the interpretation of the measurement sym-
bols and their products - the order of events is
read from right to left (sinistrally). But any
measurement symbol equation is equally valid if
interpreted in the opposite sense (dextrally).
and no physical result should depend upon which
convention is employed. On introducing the dextral
interpretation, <a'|b'> acquires the meaning
possessed by <b' a'> with the sinistral conven-
tion. We conclude that the probability connecting
states a1 and b' in given sequence must be con-
structed symmetrically from <a'|b'> and <b' a'> .
The introduction of the opposite convention for
measurement symbols will be termed the adjoint
operation, and is indicated by t . Thus,
18 QUANTUM KINEMATICS AND DYNAMICS
and
In particular,
which characterizes M(a') as a self-adjoint or
Hermitian operator. For measurement symbol pro-
ducts we have
in which X is an arbitrary number.
The significance of addition is uninfluenced by the
adjoint procedure, which permits us to extend
these properties to all elements of the measurement
algebra:
or equivalently.
THE ALGEBRA OF MEASUREMENT 19
1.8 COMPLEX CONJUGATE ALGEBRA
The use of complex numbers in the measurement
algebra Implies the existence of a dual algebra in
which all numbers are replaced^ by the complex con-
jugate numbers. No physical result can depend upon
which algebra is employed. If the operators of
the dual algebra are written X* , the correspon-
dence between the two algebras is governed by the
laws
It has the algebraic properties
1.9 MATRICES
The measurement symbols of a given descrip-
tion provide a basis for the representation of an
The formation of the adjoint within the complex
conjugate algebra is called transposition,
20 QUANTUM KINEMATICS AND DYNAMICS
2arbitrary operator by N numbers, and the abstract
properties of operators are realized by the combina-
torial laws of these arrays of numbers, which are
those of matrices. Thus
and in particular
shows that
The elements of the matrix that represents X can
be expressed as
The sum of the diagonal elements of the matrix is
the trace of the operator. The corresponding
defines the matrix of X in the a-description or
a-representation, and the product
If we set X or Y equal to 1 , we obtain
examples of the connection between the matrices of
THE ALGEBRA OF MEASUREMENT 21
basis in the dual algebra is M{a', a")* , and the
matrices that represent X* and X are the complex
conjugate and transpose, respectively, of the matrix
representing X . The operator X = XT* , an
element of the same algebra as X , is represented
by the transposed, complex conjugate, or adjoint
matrix.
The matrix of X is the mixed ab-representation
is defined by
where
The rule of multiplication for matrices in mixed
representations is
On placing X = Y = 1 we encounter the composition
property of transformation functions, since
22 QUANTUM KINEMATICS AND DYNAMICS
a given operator in various representations. The
general result can be derived from the linear rela-
tions among measurement symbols. Thus,
1.10 VARIATIONS OF TRANSFORMATION FUNCTIONS
As an application of mixed representations,
we present an operator equivalent of the fundamental
properties of transformation functions:
which is achieved by a differential characteriza-
tion of the transformation functions. If &<a'\b'>
The adjoint of an operator X , displayed in the
mixed ab-basis, appears in the ba-basis with the
matrix
THE ALGEBRA OP MEASUREMENT 23
and <S<b*|c'> are any conceivable infinitesmal
alteration of the corresponding transformation
functions, the implied variation of <a'|c'> is
which is the definition of an infinitesimal operator
<SW . . If infinitesimal operators 6W, and6W are defined similarly, the differential proper-acty (1,66) becomes the matrix equation
and also
One can regard the array of numbers o<a'|b'> as
the matrix of an operator in the ab-representation.
We therefore write
from which we infer the operator equation
24 QUANTUM KINEMATICS AND DYNAMICS
Thus the multiplicative composition law of trans-
formation functions is expressed by an additive compo-
sition law for the infinitesimal operators 6W.
On identifying the a- and b~ descriptions in
(1.70), we learn that
or
Indeed, the latter is not an independent condition
on transformation functions but is implied by the
composition property and the requirement that
transformation functions, as matrices, be nonsingu-
lar. If we identify the a- and c- descriptions we
are informed that
Now
which expresses the fixed numerical values of the.
transformation function
THE ALGEBRA OF MEASUREMENT 25
which must equal
and therefore
The complex conjugate property of transformation
functions is thus expressed by the statement that
the infinitesimal operators <§W are Hermitian.
1.11 EXPECTATION VALUE
The expectation value of property A for sys-
tems in the state b1 is the average of the possible
values of A, weighted by the probabilities of oc-
curence that are characteristic of the state b1
On using (1.33) to write the probability formula
as
the expectation value becomes
26 QUANTUM KINEMATICS AND DYNAMICS
where the operator A is
The correspondence thus obtained between operators
and physical quantities is such that a function
f(A) of the property A is assigned the operator
£{A) , and the operators associated with a complete
set of compatible physical quantities form a com-
plete set of commuting Hermitian operators. In
particular, the function of A that exhibits the
value unity in the state a' , and zero otherwise,
is characterized by the operator M(a')
1.12 ADDENDUM: NON-SELECTIVE MEASUREMENTS1"
tReproduced from the Proceedings of the National
Academy of Sciences Vol. 45, pp. 1552-1553 (1959).
1552 PHYSICS: J. SCHWINGEK PJROC. N. A, 8.
where the real phases w are independent, randomly distributed quantities. Theuncontrollable nature of the disturbance produced by a measurement thus finds itsmathematical expression in these random phase factors. Since a nonselectivemeasurement does not discard systems we must have
The physical operation symbolized by M(a') involves the functioning of an ap-paratus capable of separating an ensemble into subensembles that are distinguishedby the various values of a', together with the act of selecting one subenserable andrejecting the others. The measurement process prior to the stage of selection,which we call a nonselective measurement, will now be considered for the purpose offinding its symbolic counterpart. It is useful to recognize a general quantitativeinterpretation attached to the measurement symbols. Let a system in the statee' be subjected to the selective M(b') measurement and then to an A -measurement.The probability that the system will exhibit the value b' and then a', for the re-spective properties, is given by
There are examples of the relation between the symbol of any selective measure-ment and a corresponding probability,
Now let the intervening measurement be nonselective, which is to say that the ap-paratus functions but no selection of systems is performed. Accordingly,
which differs from
by the absence of interference terms between different br states. This indicatesthat the symbol to be associated with the nonselective ^-measurement is
If, in contrast, the intermediate B-measurement accepts all systems without dis-crimination, which is equivalent to performing no .B-measurenient, the relevantprobability is
each of which can also be extended to all types of selective measurements, and tononseleetive measurements (the adjoint form is essential here). The expectationvalue construction shows that a quantity which equals unity if the properties A,B,. . . S successively exhibit, in the sinistral sense, the values a', 6', . . . «', and iszero otherwise, is represented by the Hermitian8 operator (M(a'). . . M(s'))t-(M(a1) ,. M(«')).
Measurement is a dynamical process, and yet the only time concept that hasbeen used is the primitive relationship of order. A detailed formulation of quan-tum dynamics must satisfy the consistency requirement that its description of theinteractions that constitute measurement reproduces the symbolic characterizationsthat have emerged at this elementary stage. Such considerations make explicitreference to the fact that all measurement of atomic phenomena ultimately involvesthe amplification of microscopic effects to the level of macroscopic observation.
Further analysis of the measurement algebra leads to a geometry associated withthe states of systems.
1 Thia development has been presented In numerous lecture series since 1951, but is heretoforeunpublished.
* Here we bypass the question of the utility of the real number field. According to a comment inTHESE PROCMEBINOS, 44, 223 (I9S8), the appearance of complex numbers, or their real equivalents,may be an aspect of the fundamental matter-antimatter duality, which can hardly be discussed atthis stage.! Compare P. A. M. Dirae, Be», Mad. Phys, 17, 19S (1945), where n<»n-Hermitian operators
and complex "probabilities" are introduced.
and
Voi. 4S, 1989 ACKNOWLEDGMENT: I. OLKIN 1SS3
which corresponds to the unitary property of the Ms operators,
It should also be noted that, within this probability context, the symbols of theelementary selective measurements are derived from the nonseleetive symbol by re-placing all but one of the phases by positive infinite imaginary numbers, which is anabsorptive description of the process of rejecting subenaembles.
The general probability statement for successive measurements is
which is applicable to any type of observation by inserting the appropriate meas-urement symbol. Other versions are
CHAPTER TWO
THE GEOMETRY OF STATES
2.1 The Null State 292.2 Reconstruction of the Measurement 32
Algebra2.3 Vector Algebra 352.4 Wave Functions 372.5 Unitary Transformations 402.6 Infinitesimal Unitary Transformations 442.7 Successive Unitary Transformations 462.8 Unitary Transformation Groups- 48
Translations and Rotations2.9 Re2.10 Continuous Spectra 542.11 Addendum: Operator Space 562.12 Addendum: Unitary Operator Bases 62
2.1 THE NULL STATE
The uncontrollable disturbance attendant upon
a measurement implies that the act of measurement
is indivisible. That is to say, any attempt to
trace the history of a system during a measurement
process usually changes the nature of the measure-
29
53
30 QUANTUM KINEMATICS AND DYNAMICS
ment that is being performed. Hence, to conceive
of a given selective measurement M(a" , b 1) as a
compound measurement is without physical implica-
tion. It is only of significance that the first
stage selects systems in the state b' , and that
the last one produces them in the state a' ; the
interposed states are without meaning for the
measurement as a whole. Indeed, we can even invent
a non-physical state to serve as the intermediary.
We shall call this mental construct the null state
0 , and write
The measurement process that selects a system in
the state b1 and produces it in the null state,
can be described as the annihilation of a system
in the state b1 ; and the production of a system
in the state a1 following its selection from the
null state,
can be characterized as the creation of a system
THE GEOMETRY OF STATES 31
in the state a1 . Thus, the content of (2.1) is
the indiscernability of M(a" , b1} from the com-
pound process of the annihilation of a system in
the state b* followed by the creation of a system
in the state a' ,
and
whereas
and
The extension of the measurement algebra to
include the null state supplies the properties of
the f and * symbols. Thus
32 QUANTUM KINEMATICS AND DYNAMICS
Some properties of M{0) are
The fundamental arbitrariness of measurement sym-
bols expressed by the substitution (1.34),
2.2 RECONSTRUCTION OF THE MEASUREMENT ALGEBRA
and
Furthermore, in the extended measurement algebra,
implies the accompanying substitution
in which we have effectively removed ¥>(0) by
expressing all other phases relative to it.
THE GEOMETRY OF STATES 33
The characteristics of the measurement oper-
ators M(a" f b') can now be derived from those
of the *F and * symbols. Thus
In addition, the substitution (2.13) transforms the
measurement operators in accordance with (2.12).
The various equivalent statements contained
in (2.6) show that the only significant products —
those not identically zero — are of the form
¥<f , <f¥ and X¥ , 4>X , in addition to XY ,
where the latin symbols are operators, elements of
the physical measurement algebra. According to
and
while
34 QUANTUM KINEMATICS AND DYNAMICS
the measurement operator construction (2.4), all
operators are linear combinations of products
ff
and the evaluation of the products Xf , $X , and
XY reduces to the ones contained in (2,7),
Hence, in any manipulation of operators leading to
a product *¥ , the latter is effectively equal
to a number,
Accordingly,
and in particular
It should also be observed that, in any application
of 1 as an operator we have, in effect,
THE GEOMETRY OF STATES 35
are designed to make this result an automatic
consequence of the notation (Dirac). In the brac-
ket notation various theorems, such as the law of
matrix multiplication (1.61), or the general for-
mula for change of matrix representation (1.63),
appear as simple applications of the expression
for the unit operator
2.3 VECTOR ALGEBRA
We have associated a ¥ and a * symbol
with each of the N physical states of a descrip-
tion. Now the symbols of one description are
linearly related to those of another description,
which shows that
The bracket symbols
and
which also implies the linear relation between
measurement operators of various types. Arbitrary
numerical multiples of f or $ symbols thus
form the elements of two mutually adjoint algebras
of dimensionality N , which are vector algebras
since there is no significant multiplication of
elements within each algebra. We are thereby pre-
sented with an N-dimensional geometry -- the geo-
metry of states — from which the measurement alge-
bra can be derived, with its properties character-
ized in geometrical language. This geometry is
metrical since the number $¥ defines a scalar
product. According to (2.20), the vectors *(a')
and ¥(a1) of the a-descriptioa provides an ortho-
normal vector basis or coordinate system, and thus
the vector transformation equations (2.26) and
(2.27) describe a change in coordinate system.
The product of an operator with a vector expresses
36 QUANTUM KINEMATICS AND DYNAMICS
which characterizes f(a1) and $(a') as the
right and left eigenvectors, respectively, of the
complete set of commuting operators A , with the
eigenvalues a' . Associated with each vector
algebra there is a dual algebra in which all num-
bers are replaced by their complex conjugates.
2.4 WAVE FUNCTIONS
The eigenvectors of a given description pro-
vide a basis for the representation of an arbitrary
vector by N numbers. The abstract properties of
THE GEOMETRY OF STATES 37
a mapping upon another vector in the same space,
The effect on the vectors of the a-coordinate sys-
tem of the operator symbolizing property A ,
is given by
38 QUANTUM KINEMATICS AND DYNAMICS
vectors are realized by these sets of numbers,
which are known as wave functions. We write
and
If $ and 41 are in adjoint relation, $ = f + ,the corresponding wave functions are connected by
The scalar product of two vectors is
and, in particular,
THE GEOMETRY OF STATES 39
which characterizes the geometry of states as a
unitary geometry. The operator ¥,$,, is repre-
sented by the matrix
and
On placing X = 1 , we obtain the relation between
the wave functions of a given vector in two dif-
ferent representations,
Note that the wave function representing ¥(b')
in the-a-description is
and wave functions that represent Xf and fX
are
40 QUANTUM KINEMATICS AND DYNAMICS
2,5 UNITARY TRANSFORMATIONS
The automophisms of the unitary geometry of
states are produced by the unitary transformations
From the viewpoint of the extended measurement
algebra, <j» and ifi wave functions are matrices
with but a single row, or column, respectively.
It is a convenient fiction to assert that
every Hermitian operator symbolizes a physical
quantity, and that every unit vector symbolizes a
state. Then the expectation value of property X
in the state ¥ is given by
In particular, the probability of observing the
values a1 in an A-measurement performed on sys-
tems in the state ¥ is
THE GEOMETRY OF STATES 41
applied to every vector and operator, where the
unitary operator U obeys
All algebraic relations and adjoint connections
among vectors and operators are preserved by this
transformation. Two successive unitary transfor-
mations form a unitary transformation, and the
inverse of a unitary transformation is unitary -
unitary transformations form a group. The appli-
cation of a unitary transformation to the ortho-
normal basis vectors of the a-description, which
are characterized by the eigenvector equation
yields orthonormal vectors
that obey the eigenvector equation
42 QUANTUM KINEMATICS AND DYNAMICS
Hence the <a'j are the states of a new descrip-
tion associated with quantities A that possess
the same eigenvalue spectrum as the properties A ,
Since all relations among operators and vectors
are preserved by the transformation, we have
where
The equivalent forms
exhibit the a-representatives of operators and
vectors as the a-representatives of associated
operators and vectors.
The basis vectors of any two descriptions,
with each set placed in a definite order, are
connected by a unitary operator. Thus
THE GEOMETRY OF STATES 43
obeys
The transformation function relating the a- and
b-representations can thereby be exhibited as a
matrix referring entirely to the a- or the b--
representations ,
and all quantities of the b-representation can be
expressed as a-representatives of associated oper-
ators and vectors,
If the two sets of properties A and B possess
the same spectrum of values, the operators A and
B are also connected by a unitary transformation.
With the ordering of basis vectors established by
44 QUANTUM KINEMATICS AND DYNAMICS
corresponding eigenvalues we have
2.6 INFINITESIMAL UNITARY TRANSFORMATIONS
The definition of a unitary operator, when
expressed as
shows that a unitary operator differing infinites!-
mally from unity has the general form
where G is an infinitesimal Rermitian operator.
The coordinate vector transformation described by
this operator is indicated by
Now, according to (2.49), the change of coordinate
THE GEOMETRY OF STATES 45
system, in its effect upon the representatives of
operators and vectors, is equivalent to a corres-
ponding change of the operators and vectors rela-
tive to the original coordinate system. Hence
and
where
and
The rectangular bracket represents the commutator
Since all algebraic relations are preserved, the
operator and vector variations are governed by
46 QUANTUM KINEMATICS AND DYNAMICS
rules of the type
the latter is the operator that exhibits the same
properties relative to the a-description that X
possesses in the a-description. Thus the basis
vectors <a'| are the eigenvectors of A - 6A
with the eigenvalues a1
2.7 SUCCESSIVE UNITARY TRANSFORMATIONS
In discussing successive unitary transforma-
tions, it must be recognized that a transformation
which is specified by an array of numerical coeffi-
cients is symbolized by a unitary operator that de-
pends upon the coordinate system to which it is
applied. Thus, let U, and U„ be the operators
describing two different transformations on the
same coordinate system. When the first transfor-
mation has been applied, the operator that symbol-
One must distinguish between X + <SX and
THE GEOMETRY OF STATES 47
izes the second transformation, in its effect, upon
the coordinate system that has resulted from the
initial transformation, is
Hence the operator that produces the complete
transformation is
The same form with the operators of successive
transformations multiplied from right to left,
applies to any number of transformations. In
particular, if one follows two transformations,
applied in one order, by the inverse of the suc-
cessive transformations in the opposite order, the
unitary operator for the resulting transformation
is
When both transformations are infinitesimal,
48 QUANTUM KINEMATICS AND DYNAMICS
the combined transformation described by
is infinitesimal to the first order in each of the
individual transformations,
The infinitesimal change that the latter transfor-
mation produces in an operator is
2.8 UNITARY TRANSFORMATION GROUPS. TRANSLATIONS
AND ROTATIONS
The continual repetition of an infinitesimal
unitary transformation generates a finite unitary
which, expressed in terms of commutators, yields
the operator identity (Jacobi)
THE GEOMETRY OF STATES 49
transformation. On writing the infinitesimal
Hermitian operator G , the generator of the
unitary transformation, as STG/,» , we find that
the application of the infinitesimal transformation
a number of times expressed by T/<ST yields, in
the limit St-M) ,
These operators form a one-parameter continuous
group of unitary transformations,
A number of finite Hermitian operators ^n\ /•••/
G,, , generates a k-parameter continuous groupIK)
of unitary transformations if they form a linear
basis for an operator ring that is closed under
the unitary transformations of the group. This
requires that all commutators fG/.j\ » G(')^ t>e
linear combinations of the generating operators.
There is a fundamental continuous group of
unitary transformations based upon the significance
50 QUANTUM KINEMATICS AND DYNAMICS
of measurements as physical operations in three-
dimensional space. A measurement apparatus defines
a spatial coordinate system with respect to which
physical properties are specified. We express the
uniformity of space by asserting that two coordi-
nate systems, differing only in location and orien-
tation, are intrinsically equivalent. In particu-
lar, physical quantities that are analogously de-
fined with respect to different coordinate systems
exhibit the same spectra of possible values, and
the associated operators must be related by a
unitary transformation. Since the totality of
translations and rotations of a coordinate system
form a six-parameter continuous group, we infer
the existence of an isomorphic group of unitary
operators.
An infinitesimal change of coordinate system
is specified by stating that a point with coordi-
nate vector x in the initial system is assigned
the coordinate vector x - fix in the new system,
where
The infinitesimal generator of the corresponding
THE GEOMETRY OF STATES 51
unitary transformation is written
with the usual association of axial vectors and
antisymmetrical tensors characteristic of three
dimensions. On comparing the two ways in which
a pair of infinitesimal coordinate changes can be
performed, in the manner of (2.68), we find that
which requires that the associated infinitesimal
generators obey the commutation relation (2.71),
Hence,
The six Hermitian operators comprised in P
and J , the generators of infinitesimal transla-
tions and rotations, respectively, are identified
as the operators of total linear momentum and
total angular momentum. These physical quantities
appear measured in certain natural units - pure
numbers for angular momentum, and inverse length
for linear momentum. The connection between such
atomic units and the conventional macroscopic
standards must be found empirically. If the
latter are to be employed, a conversion factor
should be introduced, which involves the replace-
ment of P and J with fe P and ti J, respec-
52 QUANTUM KINEMATICS AND DYNAMICS
where the last statement appears in three-dimen-
sional vector notation as
THE GEOMETRY OF STATES 53
tively. The constant ft possesses the dimensions
of action, and its measured value is
The natural units' are preferable for general theo-
retical investigations and will be used here.
2,9 REFLECTIONS
The continuous group of transformations
among kinematically equivalent coordinate systems
can be enlarged by the operation of reflecting
the positive sense of every spatial coordinate
axis. We associate with this change of descrip-
tion the unitary reflection operator R ,
A reflection, followed by the infinitesimal dis-
placement 6e , 5u is equivalent to first per-
forming the displacement -<Se , Sw , and then
the reflection. Accordingly,
54 QUANTUM KINEMATICS AND DYNAMICS
or
2.10 CONTINUOUS SPECTRA
The general mathematical structure of quan-
tum mechanics as the symbolic expression of the
laws of atomic measurement has been developed in
the context of physical systems possessing a
finite number of states. We shall comment only
briefly on the extension of these considerations
to systems with infinite numbers of states, and
properties exhibiting a continuous spectrum of
possible values. In any measurement of such a
property, systems displaying values within a cer-
tain range are selected, and the concept of state
now refers to the specification of a complete set
of compatible quantities within arbitrarily small
intervals about prescribed values. We symbolize
such states by la'>. , A<a* ' anc^ express
their completeness by
THE GEOMETRY OP STATES 55
The change of normalization,
This is the continuum analogue of the property
in which Aa1 is the product of eigenvalue inter-
vals for each continuous property, now yields, in
the limit Aa'-M) ,
if all members of set A have continuous spectra.
We deduce, for any vector ¥ represented by an
arbitrary wave function
that
which is the operational definition of the delta
function ,
and generally, in all formal relations referring
to continuous spectra, integrals replace summa-
tions. In particular, the probability that a
measurement on a system in the state f will
yield one of a set of states appears as an integral
over that set, / da" f^(a')| , so that
2.11 ADDENDUM: OPERATOR SPACE1"
tReproduced from the Proceedings of the National
Academy of Sciences, Vol. 46, pp 261-265 (I960).
56 QUANTUM KINEMATICS AND DYNAMICS
can be described as the probability of encounter-
ing a system with properties A in the infinitesi-
mal range da' about a1 .
Vox,. 46, I960 PHYSICS: J. SCHWINGEK 2t»t
The geometry of states provides the elements of the measurement algebra withthe geometrical interpretation of operators on a vector space. But operators con-
262 PHYSICS: J, SCHWWGER Pnoe. N. A, S,
sidered In themselves also form a vector space, for the totality of operators is closedunder addition and under multiplication by numbers. The dimensionality of thisoperator space is Nt according to the number of linearly independent measurementsymbols of any given type. A unitary scalar product is defined iu the operatorspace by the number
The probability relating two states appears as a particular type of operator spacetransformation function.
The connection with the M(a', V) basis is described by the transformation function
Let X(a), « = \ ,. N*t he the elements of an arbitrary orthonormal basis,
One can also verify the composition property of transformation functions,
and
which is such that
can now be viewed as the transformation connecting two orthonormal bases. Thischange of basis is described by the transformation function
and the general linear relation between measurement symbols,
characterizes the M ( < t , p ) basis as orthonormal.
The trace evaluation
which has the properties
Vol.. 46, WOT
We also have
PHYSICS: J. 8CHWINQER 2(iH
and the transformation function property
and
On altering the basis the components of a given operator change in accordance with
For measurement symbol bases this becomes the law of matrix transformation.
the elements of the ab-naatrix representation of X. The scalar product in operatorspace is evaluated as
For the basis M(a', b'), the components are
defines the components
acquires the matrix form
If we multiply the latter by the &-matrix of an arbitrary operator Y, the summationwith respect to ft' and b" yields the o-matrix representation of the operator equation
the validity of which for arbitrary Y is equivalent to the completeness of the oper-ator basis X(a). Since the operator set Z(a)t also forms an orthonormal basiswe must have
and the particular choice Y = I/AT gives
The expression of an arbitrary operator relative to the orthonormal basis X(a),
264 PHYSICS: J. 8CHWINOER PBOC, N. A. S.
There are two aspects of the operator space that have no counterpart in thestate spaces—the adjoint operation and the multiplication of elements are defined
*in the same space. Thus
where
Otherwise expressed, the set of N measurement symbols M(a', b'), for fixed b', orfixed a', are left and right ideals, respectively, of the operator space.
The possibility of introducing Hermitian ortbonormal operator bases is illustratedby the set
For any such basis
and
and
and
Some consequences are
which generalize the adjoint and multiplication properties of matrices. The ele-ments of the operator space appear in the dual role of operator and operand on de-fining matrices by
The measurement symbol bases are distinguished in this context by the completereducibility of such matrices, in the sense of
VOL. 46, 1960 PHYSICS: J. SCHWINGER 265
which implies that a Hermitian operator X has real components relative to aHermitian basis, and therefore
Thus the subspace of Hermitian operators is governed by Euclidean geometry, anda change of basis is a real orthogonal transformation,
When the unit operator (multiplied by N~l/a) is chosen as a member of such basesit defines an invariant subspace, and the freedom of orthogonal transformationrefers to the JV2 — 1 basis operators of zero trace.
Important examples of orfhonormal operator bases are obtained through thestudy of unitary operators,
1 Sehwinger, J., these PEOCBBDINOS, 45,1542 (1959).
62 QUANTUM KINEMATICS AND DYNAMICS
2.12 ADDENDUM: UNITARY OPERATOR BASES1"
fReproduced from the Proceedings of the National
Academy of Sciences, Vol. 46, pp. 570-579 (1960).
Hepriat«d from the Proceedings of the NATIONAL ACAWJSMV or ScieufcisisVol. 46, No. 4, pp. 5T0-S79. April, 1980,
UNITARY OPERATOR BASMK*
BY JULIAN SCHWINUJSK
HABVABB OMIVSRSOTf
Communicated February t, 1900
To qualify as the fundamental quantum variables of a physical system, a set ofoperators must suffice to construct all possible quantities of that system. Suchoperator? will therefore be identified as the generators of a complete operator basis.Unitary operator bases are the principal subject of this note.1
Two state vector space coordinate systems and a rule of correspondence define aunitary operator. Thus, given the two ordered sets of vectors (o*|, {&*[, k = I . . N,aad their adjoints, we construct
which are such that
and
for both Ua> and £/»„. If a third ordered coordinate system is given, (c*|, k — I . .N,we can similarly define the unitary operators U,f, [/»„, which obey the compositionproperty
implying the unitary property
obeys the same multiplication law as the X(a), and the F(a) t are given by the samelinear combination of the F(«) as are the X(«)f of the X(a) set. In particular, ifX{a) is a Hermitiari basis, so also is F(a),
We cannot refrain from illustrating these remarks for the simplest of the N*-dimensional operator spaces, the quaternion space associated with a physical systempossessing only two states. If a particular choice of these is arbitrarily designatedas + and —, we obtain the four measurement symbols M(±, ±), and can then in-troduce a Hermitian orthonormal operator basis
Voi. 46, 1960 PHYSICS: J. SCSWINGER 571
A unitary operator is also implied by two orthonormal operator bases in a givenspace, that have the same multiplication properties:
Let us define
and observe that
where the latter statement follows from the remark that the F(a)t form an ortho-normal basis and therefore
We also have the adjoint relation
and in consequence
according to the completeness of the X(a) basis. Thus the operator U is unitaryif we choose
and, to within the arbitrariness of a phase constant,
The converse theorems should be noted. For any unitary operator U, the ortho-normal basis
with
572 PHYSICS: J, SCHWINGSB, Pmoc. N. A, 8
such that
Accordingly, the three operators «% k — 1,2, 3, obey
and an explicit construction is given by
the coefficients of which constitute the well-known Pauli matrices. With thesedefinitions, the multiplication properties of the & operators can be expressed as
or, equivalently, by the additional trace
where <*&,, is the alternating symbol specified by em = +1. If we now introduceany other Hermitian orthonormal operator basis F(a) = 2~1/2S'a, with *« = 1,the resulting three-dimensional orthonormal basis transformation
is real and orthogonal,
The multiplication properties of the <r-basis assert that
where, characteristic of an orthogonal transformation, det r = ± 1. If the orthog-onal transformation is proper, the multiplication properties of the ff-basis coincidewith those of the o--basis, while with an improper transformation the opposite signof t is effectively employed in evaluating the <r»»j products. Hence only in the firstsituation, that of a pure rotation, does a unitary operator exist such that
The unitary operator is constructed explicitly as!
Voiu 48, 1080 PHYSICS: J SCHWINOER 573
Let us return to the definition of a unitary operator through the mapping of onecoordinate system or another, and remark that the two of vectors can be iden-tical, apart from their ordering. Thus, consider the definition of a unitary operatorF by the cvclic permutation
On multiplying M (»*) by (o'v| and using the defining property of F, we obtain
to within the ehoiee of the factor N~l, which is such that
permits the identification of the Hermitiau operator
have all the properties required of measurement symbols. Now let us observethat the factorization of the minimum equation for F that is given by
Unitary operators can be regarded as complex functions of Hermitian operators,and the entire spectral theory of Hermitian operators can be transferred to them.If the unitary operator F has JV distinct eigenvalues, its eigenvectors constitute anorthonormal coordinate system. The adjoint of a right eigenvector |»') is the lefteigenvector (v1 associated with the same eigenvalue, and the products
is the minimum equation, the polynomial equation of least degree obeyed by thisoperator, which we characterize as being of period N. The eigenvalues of F obeythe same equation and are given by the N distinct complex numbers
Thus
until we arrive at
which indicates the utility of designating the same state by any of the integers thatare congruent with respect to the modulus N, The repetition of F defines linearlyindependent unitary operators,
where
874 PHYSICS: J. 8CHWINQSR PROC, N. A, S.
from which follows
Then, with a convenient phase convention for (a*| v") we get
which is also expressed by
the elements of the transformation functions that connect the given coordinatesystem with the one supplied by the eigenvectors of the unitary operator thatcyclically permutes the vectors of the given system.
Turning to the new coordinate system {»*(, we define another unitary operatorby the cyclic permutation of this set. It is convenient to introduce U such that
which is equivalent to
This operator is also of period ff,
and has the same spectrum as V,
After using the property U"~l = U~l to write the corresponding measurementsymbol as
we follow the previous procedure to construct the eigenvectors of U,
Thus, the original coordinate system is regained and our results can now be statedas the reciprocal definition of two unitary operators and their eigenvectors,
The relation between the two coordinate systems is given by
and, supplementing, the periodic properties
Vox,. 48, 1960 PHYSICS: J, 8CHWINOBR 67S
we infer from the comparison
that
Each of the unitary operators U and F is a function of a Hermitian operator thatin itself forms a complete set of physical properties. It is natural lo transfer thisidentification directly to the unitary operators which are more accessible than theimplicit Hermitian operators. Accordingly, we now speak of the statistical rela-tion between the properties U and F, as described by the probability
As a consequence of the latter result, we also have
which are invariant under this substitution when combined with m -*• », n -*• — m.One proof of completeness for the operator basis generated by U and F depends
upon the following lemma: If an operator commutes with both U and F it is neces-sarily a multiple of the unit operator. Since U is complete in itself, such an opera-tor must be a function of U. Then, according to the hypothesis of eomniutatrvitywith V, for each k we have
The latter could be emphasized by choosing the elements of the operator basis as
and therefore together supply the foundation for a full description of a physicalsystem possessing N states. Both of these aspects are implied in speaking of U andF as a complementary pair of operators.* Incidentally, there is complete symmetrybetween U and F, as expressed by the invariance of all properties under the sub-stitution
for this asserts that the intervening non-selective w-measurement has destroyedall prior knowledge concerning w-states. Thus the properties U and F exhibit themaximum degree of incompatibility. We shall also show that U and F are thegenerators of a complete orthonormal operator basis, such as the set of Nf operators
The significance of this result can be emphasized by considering a measurement se-quence that includes a nonselective measurement, as in
576 PHYSICS: J. SCHWIN0MR PBOC. N. A. S,
and this function of U assumes the same value for every state, which identifies itwith that multiple of the unit operator. Now consider, for arbitrary F,
It is interesting to notice that a number of the powers C/*, k = 1 . , N — I , canhave the period N. This will occur whenever the integers fc and N have no commonfactor and thus the multiplicity of such operators equals $(AT), the number of in-
which is a kind of ergodic theorem, for it equates an average over all spectral trans-lations to an average over all states. The explicit reference to operators can beremoved if F(U, V) is constructed from terms which, like the individual operatorsX(mn), are ordered with U standing everywhere to the left of F, Then we canevaluate a matrix element of the operator equation, corresponding to the stateshf and I tf), which gives the numerical relation
which exhibits the unitary transformations that produce only cyclic spectral trans-lations. Now, if F is given as an arbitrary function of U and F, the completenessexpression of the operator basis reads
and observe that left and right hand multiplication with U and U~ , respectively,or with F and F~J, only produces a rearrangement of the summations. Accord-ingly, this operator commutes with U and F. On taking the trace of the resultingequation, the multiple of unity is identified with ir Y, and we have obtained
the statement of completeness for the ]V2-dimensional operator basis X(mn). Al-ternatively, we demonstrate that these JV2 operators are orthonormal by evaluating
The unit value for m = m', n — n' is evident. If m ^ TO', the difference TO' — mcan assume any value between N — I and — (N — 1), other than zero. When thetrace is computed in the v-representation, the operator [/«'-»> changes each vector(vk\ into the orthogonal vector {»*+»>-»' j and the trace vanishes. Similarly, if n 7*n' and the trace is computed in the «-representation, each vector {«* is convertedby Vn'~" into the orthogonal vector (t**'H1/~n| and the trace equals zero.
One application of the operator completeness property is worthy of attention.We first observe that
Vol., 48, I960 PHYSICS: J, SCHWINGSM 877
tegers less than and relatively prime to N. Furthermore, to every such power of Uthere can be associated a power F', also of period N, that obeys with 17* the sameoperator equation satisfied by U and V,
Furthermore,
so that I/i, Fi and Ut, Vt constitute two independent pairs of complementary oper-ators associated with the respective periods JVi and Nt. We also observe that theN = JViJVz independent powers of U can be obtained as
since all of these are distinct powers owing to the relatively prime nature of N\ andNt. With a similar treatment for F, we recognize that the members of the ortho-normal operator basis are given in some order by
It is seen that Ui, Yt are of period N\t while Ut, V% have the period Nt, and that thetwo pairs of operators are mutually commutative, as illustrated by
with
where the integers N\ and Nt are relatively prime, and define
This requires the relation
and the unique solution provided by the Fermat-Euler theorem is
The pair of operators U*, V1 also generate the operator basis X(mn), in some per-muted order,
We shall now proceed to replace the single pair of complementary operators U, Vby several such pairs, the individual members of which have smaller periods thanthe arbitrary integer N. This leads to a classification of quantum degrees of free-dom in relation to the various irreducible, prime periods. Let
578 PHYSICS; J, SCSWINGSR PEOC, N A. S.
together with the commutativity of any two operators carrying different sub-scripts. The orthonormality of the NfN<? — N* operators X'(?»i«»»j»z} can nowbe directly verified. Also, by an appropriate extension of the proceeding discus-sion, we obtain the transformation function
with
The continuation of the factorization terminates in
Equivatently, we replace
where
Another approach to this commutative factorization of the operator basis pro-ceeds through the construction of the eigenvalue index k = 0.. N — 1 with the aidof a pair of integers,
which gives
by
On identifying these vectors as the eigenvectors of sets of two commutative unitaryoperators, we can deine U\,t, Vi,» by the reciprocal relations
which reproduce the properties
We shall not carry out the necessary operations for v -*• «, which evidently yieldthe well-known pair of complementary properties with continuous spectra. Oneremark must be made, however. In this approach one does not encounter thesomewhat awkward situation in which the introduction of continuous spectra re-quires the construction of a new formalism, be it expressed in the language ofDirac's delta function, or of distributions. Rather, we are presented with the directproblem of finding the nature of the subspaees of physically meaningful states andoperators for which the limit v -*• <*> can be performed uniformly,
* Publication assisted by the Office of Scientific Research, United States Air Force.1 For the notation and concepts used here see these PROCEEDINGS, 45, 1542 (1959), and 46,287
(1«60).2 The absence in the available literature of an explicit statement of this simple, general result
is rather surprising. The inverse calculation giving the three-dimensional rotation matrix interms of the elements of the unitary matrix is very we!! known (rotation parametrizations ofBuler, Cay ley-Klein), and the construction of the unitary matrix with the aid of Euleriftn anglesis also quite familiar.
* Operators having the algebraic properties of U and F have long been known from the work ofWeyl, H., Theory of Groups and Quantum Mechanics (New York: B. P. Button Co., 1932), chap.4, sect. 14, but what has been lacking is an appreciation of these operators as generators of a com-plete operator basis for any If, said of their optimum incompatibility, as summarized in the at-tribute of complementarity. Nor bag it been clearly recognized that an a priori classification ofall possible typee of physical degrees of freedom emerges from these considerations.
and where
Vol. 46, 1960 PHYSICS: J, SCHWINGER 579
where / is the total number of prime factors in N, including repetitions. We callthis characteristic property of N the number of degrees of freedom for a systempossessing N states. The resulting eommutatively factored basis
is thus constructed from the operator bases individually associated with the / de-grees of freedom, and the pair of irreducible complementary quantities of each de-gree of freedom is classified by the value of the prime integer v — 2, 3, 5, . . . «.In particular, for v — 2 the complementary operators U and V are anticommutativeand of unit square. Hence, they can be identified with <n and <r», for example, andthe operator basis is completed by the product —iUV — <r\.
The characteristics of a degree of freedom exhibiting an infinite number of statescan be investigated by making explicit the Hermitian operators upon which U andV depend,
CHAPTER THREETHE DYNAMICAL PRINCIPLE
3.1 The Action Operator 743.2 Lagrangian Operator 763.3 Stationary Action Principle 773.4 The Hamiltonian Operator 793.5 Equations of Motion. Generators 803.6 Commutation Relations 833.7 The Two Classes of Dynamical 86
Variables3.8 Complementary Variables of the 97
First Kind3.9 Non-Hermitian Variables of the 103
First Kind3.10 Complementary Variables of the 106
Second Kind
A measurement is a physical operation in
space and in time. The properties of a system
are described in relation to measurements at a
given time, and no value of the time is intrin-
sically distinguished from another by the results
of measurements on an isolated physical system.
73
74 QUANTUM KINEMATICS AND DYNAMICS
Hence the operators symbolizing analogous proper-
ties at different times must be related by a uni-
tary transformation. The propagation in time of
the disturbance produced by a measurement implies
that physical quantities referring to different
times are incompatible, in general. Accordingly,
complete sets of compatible properties will per-
tain to a common time, and the characterization of
a state requires the specification of the values
of these quantities (together with a spatial
coordinate system) and of the time. The trans-
formation function relating two arbitrary des-1 H
criptions thus appears as <a_t.|a2t2> , special
cases of which are <a't[b't> , connecting two
different sets of quantities at the same time, and
<a't1|allt_> , which relates analogous properties
at different times. The connection between states
at two different times involves the entire dynami-
cal history of the system in the interval. Hence
the properties of specific systems must be con-
tained completely in a dynamical principle that
characterizes the general transformation function.
3.1 THE ACTION OPERATOR
for consecutive transformations. We now state our
fundamental dynamical postulate: There exists a
special class of infinitesimal alterations for
which the associated operators SW12 are obtained
by appropriate variation of a single operator,
the action operator W,,., »
and has the additive combinatorial property
where 5W is an infinitesimal Hermitian opera-J.1&
tor with the additivity property
THE DYNAMICAL PRINCIPLE 75
Any infinitesimal alteration of the trans-1 i "formation function <a,t,|a~ t_> can be expressed
as [Eq. (1.68)1
It is consistent with the properties of the infini-
tesimal operators to assert that the action opera-
tor is Hermitian,
76 QUANTUM KINEMATICS AND DYNAMICS
3.2 LAGRANGIAN OPERATOR
If one views the transformation from the a2t2
description to the a, t^ description-s>s occurring
continuously in time through an infinite succession
of infinitesimally differing descriptions, the
additivity property of action operators asserts that
where
since <a't|a"t> has fixed numerical values. On
writing
the action operator acquires the general form
in which L(t) , the Lagrangian operator, is a
Hermitian function of some fundamental dynamical
THE DYNAMICAL PRINCIPLE 77
variables x (t) in the infinitesimal neighbor-€*
hood of time t . There is no loss of generality
in taking the operators xa(t) to be Hermitian,
and, for our present purposes, we suppose their
number to be finite. The conceivable objects of
variation in the action operator are the terminal
times t, and t- , the dynamical variables, and
the structure of the Lagrangian operator.
3.3 STATIONARY ACTION PRINCIPLE
For a given dynamical system, changes in a
transformation function can be produced only by
explicit alteration of the states to which it
refers. Such variations of states arise from
changes of the physical properties or of the time
involved in the definition of state, and infini-
tesimal eigenvector transformations are generated
by Hermitian operators that depend only upon dyna-
mical variables at the stated time,
78 QUANTUM KINEMATICS AND DYNAMICS
Hence,
which is the operator principle of stationary ac-
tion for it asserts that the variation of the
action integral involves only dynamical variables
at the terminal tiroes. This principle implies
equations of motion for the dynamical variables
and yields specific forms for generators of infi-
nitesimal transformations. Note that the Lagran-
gian operator cannot be determined completely by
the dynamical nature of the system, since we must
be able to produce a variety of infinitesimal
transformations for a given system. Indeed, if
two Lagrangians differ by a time derivative,
the action operators are related by
and, with a given dynamical system,
THE DYNAMICAL PRINCIPLE 79
and
which also satisfies the stationary action require-
ment but implies new generators of infinitesimal
transformations at times t, and t~ , as given
by
3.4 THE HAMILTONIAN OPERATOR
The development of the fundamental dynamical
variables through an infinitesimal time interval
is described by an infinitesimal unitary transfor-
mation, which implies first order differential
equations of motion for these variables. A gener-
al form of Lagrangian operator that yields first
order equations of motion is
80 QUANTUM KINEMATICS AND DYNAMICS
in which A , is a numerical matrix. We shall
speak of the two parts of the Lagrangian operator
as the kinematical and the dynamical parts, re-
spectively. Note that the kinematical part has
been symmetrized with respect to transference of
the time derivative. The Hermitian requirement on
L applies to the two parts separately. Hence
H , the Hamiltonian operator, must be Hermitian,
and the finite matrix A must be skew - Hermitian.
For an isolated dynamical system, there is no ex-
plicit reference to time in H
3.5 EQUATIONS OF MOTION. GENERATORS
The action operator implied by the particular
structure of L in (3.17) is
THE DYNAMICAL PRINCIPLE 81
In this form the limits of integration are objects
of variation. However, one can think of intro-
ducing an auxiliary variable T , and producing
the variations 6t, , 6t~ by altering the func-
tional dependence of t = t(t) upon T , with
fixed limits T, , T- . This procedure places
the time variable on somewhat the same footing as
the dynamical variables. Now
since
and thus the stationary action principle asserts
that
or
82 QUANTUM KINEMATICS AND DYNAMICS
and yields the infinitesimal generators
at the terminal times.
The structure of the Hamiltonian operator is
as yet unspecified. If its variation is to possess
the form (3.23), with the 5x appearing only on
the left and on the right, these variations must
possess elementary operator properties character-
izing the special class of operator variations to
which the dynamical principle refers. Thus, we
should be able to displace each Sx entirely toSI
the left or to the right, in the structure of
5H , which defines the left and the right deriva-
tives Of H ,
In view of the complete symmetry between left and
right, we infer that the terms in (3.23) with 6x
on the left, and on the right, are identical. We
thus obtain
THE DYNAMICAL PRINCIPLE 83
and
and, generally, left and right hand forms are
connected by the adjoint operation. We shall
assume in our discussion that the matrix A is
nonsingular. This implies that every variation
<5x appears independently in G and that eachct X
variable x obeys an explicit equation of motion,ci
3.6 COMMUTATION RELATIONS
The infinitesimal generator Gt = - H fit
where the latter must be equivalent forms of the
equations of motion. Similarly, the infinitesimal
generator G possesses the equivalent forms3v
34 QUANTUM KINEMATICS AND DYNAMICS
evidently describes the unitary transformation
from a description at time t to the analogous
one at time t -f 6t , which identifies H as the
energy operator of the system. -If F is any func-
tion of the dynamical variables x(t) and of the
time, the operator F that plays the role of F
in the description referring to time t + 6t is
or
which is the general equation of motion. On plac-
ing F = H , we obtain
in agreement with (3.26), derived from the station-
ary action principle. Such consistency must also
since the numerical parameter t is not affected
by operator transformations. Hence
THE DYNAMICAL PRINCIPLE 85
appear in the equations of motion for the dynamical
variables, which, requires that
and
On writing these relations in the form
we can conclude from the equality of the right
sides, and from the two equivalent expressions for
Gx , that
We shall satisfy this consistency requirement by
demanding that each variation 6x commute witha
the Hamiltonian operator,
86 QUANTUM KINEMATICS AND DYNAMICS
for this enables the commutation relations (3.35)
to be stated as
and identifies the transformation generated by G2v
as the change of each dynamical variable x byOL
*s6xa . Accordingly, for an arbitrary function of
the dynamical variables at the stated time, we
have
3.7 THE TWO CLASSES OF DYNAMIC VARIABLES
The two equivalent versions of Gx , bilinear
in x and 5x , indicate that displacing a
variation <$x across any of the dynamical vari-Si
ables induces a linear transformation on these
variables,
where k_ designates a matrix. The adjoint state-a
ment is
THE DYNAMICAL PRINCIPLE 87
from which we conclude that
The commutativity of each <$x with the Bamil-a
tonian operator H(x) now appears as the set of
invariance properties
which shows that the set of linear transformations
k form a group of invariance transformations fora
H . From the fundamental significance of this
group we conclude that it must apply to the com-
plete structure of the Lagrangian operator. The
kinematical term is invariant under the linear
where the latter statement expresses the Hermitian
nature of H . On forming the commutator with a
second variation 6x, we learn thatb
88 QUANTUM KINEMATICS AND DYNAMICS
transformation k If3s.
Also, the two equivalent expressions for GX
imply that
which is consistent with (3.45). For this intrinsic
equivalence of the variables x and k x to bea
complete, the latter must be Hermitian operators.
Hence the matrices k are real, and obeya
The construction of the invariance group
described by the matrices k has been based upona
the particular operator variation Sx . ButC*
there must exist some freedom of linear transfor-
mation whereby new Hermitian variables are intro-
duced,
THE DYNAMICAL PRINCIPLE 89
For the new choice of variables, the invariance
properties of the Hamiltonian read
so that
On the other hand, these matrices should appear
directly from the commutation properties of the
variations <Sx ,ct
By referring this statement back to the character-
istics of the <Sx , we obtain
with accompanying redefinitions of the matrix A
and of the Hamiltonian operator.
Now this result cannot be valid for arbitrary £ ,
within a certain group of transformations, unless
the matrices k, are identical for all values ofb
b that can be connected by the linear transforma-
tion. Accordingly, the dynamical variables x
must decompose into classes such that linear trans-
formations within each class only are permissible,
with the different classes distinguished by speci-
fic matrices k , In view of the freedom of in-
dependent linear transformations within each class,
the matrices k^ must maintain the decompositiona
into classes and thus contain only submatrices
characteristic of each class-of variable. The
partitioning of the dynamical variables produced
by any of the k matrices,a
combined with the variation properties
90 QUANTUM KINEMATICS AND DYNAMICS
THE DYNAMICAL PRINCIPLE 91
Indicate further that the submatrices of k3.
appear merely as numbers kab, labelling the
various classes. According to (3.47) / these
numbers obey
Hence the two possibilities of k = ± 1 defineclct
two distinct classes of dynamical variables»
We shall see shortly that k, _ - k,,. , and,
since the identity transformation must appear in
the group of k-transformations, we have
If we distinguish the two classes of dynamical
variables as variables of the first kind, z, ,.K
and variables of the second kind, £K , the
operator properties of the variations <Sz. ,JC
6 ?; , summarized in
92 QUANTUM KINEMATICS AND DYNAMICS
are given explicitly by
requires that H be an even function of the vari-
ables of the second kind, but no restriction ap-
pears for the dependence upon the variables of the
first kind.
The property of the matrix A contained in
(3.45), together with the opposite signs of k«2
and k--, , shows that all elements of A connec-
ting the two classes of variables must be zero.
Hence A reduces completely into two submatrices
associated with the two kinds of variable, which
we shall designate as a and io , respectively.
where the curly bracket signifies the anticoKHnu-
tator
The Hamiltonian invariance transformation implied
by commutativity with the <5e ,
THE DYNAMICAL PRINCIPLE 93
It follows from (3,46) that a is antisymmetrical»
and hence real,
Two complete reduction of A into a and ia
implies a corresponding additive decomposition of
the generator G »«K.
with
and
This structure of Gx is an aspect of the additive
form assumed by the kinematical term in the
Lagrangian,
while a is symmetrical and real,
94 QUANTUM KINEMATICS AND DYNAMICS
which we express by calling the two sets of vari-
ables kinematieally independent. The equations of
motion for these kinematically independent sets of
variables are
We shall adopt a uniform notation to indicate the
characteristic symmetrization in the Lagrangian of
bilinear structures referring to variables of the
first kind, and the antisymnnetrization for bilinear
functions of variables of the second kind, namely
On displacing <$x to the left or to the
right in the general commutation relation (3,39),
and
THE DYNAMICAL PRINCIPLE 95
we obtain
The equations resulting from the special choice
F (x) = xh can be presented as
and, on interchanging a and b in one version,
we conclude from the other that
We have already made use of the only significant
statement contained here, k-,2 m kpi * Tne com~
mutation properties of the two classes of dynami-
cal variables now appear explicitly as
96 QUANTUM KINEMATICS AND DYNAMICS
It will be seen that the structure of the operators
reproduces that of the matrices, with the antisyra-
metrical, skew-Hermitian commutator appearing with
the antisymrrtetrical, imaginary matrix (l/i)a ,
while the symmetrical, Hermitian anticommutator is
related to the syymetrical, real matrix a . In
addition, the matrix a must be positive-definite
if the variables £ are to be linearly indepen-K.
dent. The explicit forms of the general commuta-
tion relation are
without distinction between left and right deriva-
tives, while, by distinguishing operators that are
even and odd functions of the variables of the
second kind, we find that
and
THE DYNAMICAL PRINCIPLE 97
3.8 COMPLEMENTARY VARIABLES OF THE FIRST KIND
The matrices a and a are nonsingular,
according to the assumption made about A . For
the antisymmetrical matrix a , associated with
variables of the first kind, the remark that
shows that the number of variables of the first
kind cannot be odd. We shall designate this num-
ber as 2n, , Now the matrix defined by
is a real, symmetrical, positive-definite function
of a , and on writing
we observe that X is a real, antisymmetrieal
matrix that obeys
98 QUANTUM KINEMATICS AND DYNAMICS
Furthermore, there exists a real, symmetrical
matrix p such that
The matrix p commutes with a. and anticommutes
with X . By an appropriate choice of Hermitian
variables z , all these 2n,-dimensional matrices
can be displayed in a partitioned form, with n,-
dimensional submatrices, as
pondingly partitioned sets of n., variables
z,: and z,, , k - l,...,n, . According toK K X
(3.75), these variables obey the commutation rela-
with a an n,-dimensional real, symmetrical, posi-
tive-definite matrix. We shall call the corres-
which establishes the possibility of attaining the
matrix forms
THE DYNAMICAL PRINCIPLE 99
tions
With the partitioned form of a , the kine-
matical term and the infinitesimal generator, re-
ferring to the variables of the first kind, ac-
quire the forms
and
G
where the latter generates changes in z and
z(2) of h&z^ and %6z(2) , respectively.
Now let us exploit the freedom to add a time deri-
vative term to the Lagrangian with a corresponding
alteration of the infinitesimal generator, in the
manner of (3.13) and (3.16). With the choice
100 QUANTUM KINEMATICS AND DYNAMICS
we obtain the new kinematical term
while the opposite sign for W gives the kinema-
tical term
and the generator
By comparison with (3,88) we recognize that G ,^%*
generates the transformation in which the operators
z are unaltered and the z are changed by
6z . The converse interpretation applies to
G ... . Thus we have divided the variables ofz k '
the first kind into two sets that are complemen-
tary, each set comprising the generators of infini-
tesimal variations of the other set. The interpre-
and the new generator
THE DYNAMICAL PRINCIPLE 101
tation of the generators G ,,, and G ._. isZ £i
expressed by
from which we regain the commutation relations
(3.86), and derive the equations of motion for
the complementary variables,
in agreement with the implications of the action
principle.
A real, symmetrical positive-definite matrix
can always be reduced to the unit matrix by a real
transformation, which, applied to a. , places the
description by complementary variables in a canon-
ical form. The required transformation is pro-
duced by introducing the canonical variables
for this converts the kinematical term (3.90) into
102 QUANTUM KINEMATICS AND DYNAMICS
Thus the canonical variables constitute n, com-
plementary pairs of kinematically independent dy-
namical variables of the first kind. Various
aspects of the canonical variables are: the
equations of motion,
the infinitesimal generators,
and the commutation properties of the canonical
variables,
the general commutation relations»
THE DYNAMICAL PRINCIPLE 103
3.9 NON-HEBMITIAN VARIABLES OF THE FIRST KIND
It is important to recognize that the dyna-
mical theory, which has been developed in terms of
Hermitian dynamical variables, permits the intro-
duction of non-Hermitian complementary variables.
Let us define
and observe that the kinematical term (3.87) can
also be written as
which still has the same structure, with iy
and y replacing z and z , respectively.
The latter form also persists under arbitrary com-
plex linear transformations of the non-Hermitian
variables y , with appropriate redefinitions of
a. as a positive-definite Hermitian matrix. The
formal application of the previous considerations
to the non-Hermitian variables now leads, for ex-
ample, to the commutation relations
104 QUANTUM KINEMATICS AND DYNAMICS
written for real a , which are precisely the re-
sults that would be obtained by combining the com-
mutation relations for the Hermitian variables, in
accordance with the definitions of the non-Hermi-
tian variables, together with
Regarded as the definition of differentiation with
respect to the non-Hermitian variables, these
equations imply that
which justifies the formal treatment in which y
THE DYNAMICAL PRINCIPLE 105
and y are subjected to independent variations.
Thus the formal theory employing non-Hermitian
variables produces correct equations of motion
and commutation relations. In particular, the
canonical version is applicable, although the
canonical variables q, and p, are not self-iC JC
adjoint but rather
and
in which the pairs of equations stand in adjoint
relation.
Accordingly, the canonical equations of motion
and commutation relations can be written
106 QUANTUM KINEMATICS AND DYNAMICS
3.10 COMPLEMENTARY VARIABLES OF THE SECOND KIND
The necessity of an even number of variables
and the possibility of dividing them into two com-
plementary sets also appears for variables of the
second kind. Let us observe that the real, symme-
trical, positive-definite matrix a can be reduced
to the unit matrix by an appropriate real trans-
formation of the variables, which is effectively
produced on introducing
The new canonical Bermitian variables obey
which means that two different £ -operators
anticoimnute, and that the square of each £ is
a common numerical multiple of the unit operator.
The accompanying canonical form of the equations
of motion is
while the general commutation relations become
We now want to emphasize, for a system described
by variables of the second kind, the requirement
that the complete measurement algebra be derived
from the fundamental dynamical variables, which is
the assumption that accompanies the introduction
of such variables. The linearly independent oper-
ators that can be constructed from the £;-variables
are enumerated as follows; the unit operator;
the v operators £ ; the 'sv(v-l) operatorsIs
C-^C, i K < A ; the i v ( v - l ) ( v - 2 ) operators
C 5,5 , t c < X < y ; and so forth, culminating in theK A \i
single operator £iCo »*••» € • The total num-
ber of independent operators so obtained, the
dimensionality of the £-algebra, is
But this number must also be the dimensionality of
the measurement algebra, which equals the square
of the integer N representing the total number
THE DYNAMICAL PRINCIPLE 107
108 QUANTUM KINEMATICS AND DYNAMICS
of states. The desired equivalence is possible
only if v is an even integer,
On dividing the Hermitian canonical variables
into two sets of equal number, £ and 5 ,
the kinematical term for the variables of the
second kind becomes
A description employing complementary variables
appears with the introduction of non-Herinitian
canonical variables of the second kind,
which converts the kinematical term into
This structure is applicable to either of the two
kinds of complementary dynamical variable, as are
THE DYNAMICAL PRINCIPiLE 109
the forms
obtained by adding suitable time derivatives.
Accordingly, the equation of motion
and the generators of infinitesimal changes in q
or p ,
can refer to either of the two types of variable.
The distinction between the two classes is impli-
cit in the relation of left and right derivatives,
and, generally, in the operator properties of the
variations <$q and 6p . Thus, for the variables
of the second kind,
110 QUANTUM KINEMATICS AND DYNAMICS
The general commutation relations that express the
significance of these generators are
together with
and these statements are identical with the results
obtained directly from (3.113), the commutation
properties of the Hermitian dynamical variables.
As applications of the general commutation
relations, we regain the canonical equations of
motion, and derive the commutation properties of
the canonical non-Hermitian variables of the
second kind,
THE DYNAMICAL PRINCIPLE 111
In virtue of the adjoint connection between the
canonical variables,
and
the equations of motion and commutation relations
can also be presented as
This page intentionally left blank
CHAPTER FOURTHE SPECIAL CANONICAL GROUP
I. VARIABLES OF THE FIRST KIND H4
4.1 Differential Operators H5
4.2 Schrodinger Equations H^4.3 The q p Transformation Function 1204.4 Differential Statements of 122
Completeness4.5 Non-Herroitian Canonical Variables 1254.6 Some Transformation Functions 1264.7 Physical Interpretation 1304.8 Composition by Contour Integration 1334.9 Measurements of Optimum Compatibility 1^0
II. VARIABLES OF THE SECOND KIND I43
4.10 Rotation Group 1434.11 External Algebra i45
4.12 Eigenvectors and Eigenvalues 148III. UNIFICATION OF THE VARIABLES i52
4.13 Constructive Use of the Special I52
Canonical Group4.14 Transformation Functions 1564.15 Integration 1664.16 Differential Realizations 170
The commutation properties of the infinites-
imal operator variations that are employed in the
fundamental dynamical principle are such as to
113
114 QUANTUM KINEMATICS AND DYNAMICS
maintain the commutation relations obeyed by the
dynamical variables. Accordingly, these special
variations possess a transformation group aspect,
which we now proceed to examine.
I. VARIABLES^ OF THE FIRST KIND
For variables of the first kind, the variations
5z commute with all operators and thus appear as
arbitrary infinitesimal real numbers. If one con-
siders the generators of two independent infinites-
imal variations
their commutator can be evaluated by enploying the
generator significance of either operator,
The equivalent canonical forms are
THE SPECIAL CANONICAL GROUP 115
We thus recognize that the totality of infinitesi-
mal generators G , or G , G , togetherz cj p
with infinitesimal multiples of the unit operator
are closed under the formation of commutators and
therefore constitute the infinitesimal generators
of a group, which we call the special canonical
group.
4,1 DIFFERENTIAL OPERATOR'S
The transformations of this group can be
usefully studied by their effect on the eigenvec-
tors of the complete set of commuting Hermitian
operators provided by the n, canonical variables
q , or p , at a given time t . We first
consider the interpretation of the transformation
generated by G on the eigenvectors of the q-
description, and similarly for the p -description,
as indicated by (2.58),
Now the vector <q't| + 6<q't| is an eigenvector
of the operators q - <Sq with the eigenvalues
q1 . But, since the <$q are numerical multiples
of the unit operator, the varied, vector is des-
cribed equivalently as an eigenvector of the oper-
ators q with the eigenvalues q1 -f 5q . This
shows that the spectra of the Bermitian operators
q, form a continuum, extending from -« to « ,
and that the variation <$• <q't can be ascribedq
to a change of the eigenvalues q' by Sq . A
similar argument applies to the complementary vari-
ables p . Accordingly,
and
116 QUANTUM KINEMATICS AND DYNAMICS
THE SPECIAL CANONICAL GROUP 117
The adjoint relations are
Twhere the symbol 3 is used to indicate that the
conventional sense of differentiation is reversed.
For functions F (q » p) thai, can be constructed
algebraically from the variables p and arbitrary
functions of the q , we establish by induction
that
and
which supplies a differential operator realization
118 QUANTUM KINEMATICS AND DYNAMICS
for the abstract operators. Similarly, for func-
tions derived by algebraic construction from the
q and. arbitrary functions of the p ,
The significance of the transformation gener-
ated by G on the eigenvectors of the q - des-
cription, and conversely, is indicated by
namely, these transformations multiply the vectors
by numerical phase factors. The transformations
of which infinitesimal real multiples of the unit
operator are the generators also multiply vectors
by phase factors, but without distinction between
the two descriptions. Hence, in response to the
operations of the special canonical group, any
and
THE SPECIAL CANONICAL GROUP 119
eigenvector in the q or p descriptions with
specified eigenvalues can be transformed into one
with any other set of eigenvalues, and multiplied
with an arbitrary phase factor.
4.2 SCHR5DINGER EQUATIONS
The infinitesimal operator G. = -H <5t gen-
erates the group of time translations, whereby a
description at any one time is transformed into
the analogous description at any other time.
Thus, for the infinitesimal transformation from
the q - description at time t to the q -des-
cription at time t + <St we have
or, as an application of the differential operator
realization (4.8), for a system described by vari-
ables of the first kind,
if H is an algebraic function of the p varia-
120 QUANTUM KINEMATICS AND DYNAMICS
bles. The adjoint statement can be written
The arbitrary state symbolized by f is represen-
ted by the complex conjugate pair of wave functions
4.3 THE q p TRANSFORMATION FUNCTION
The transformation function connecting the
q and p representations at a common time can be
constructed by integration of the differential
equation
and the variation of these functions in time is
thus described by the (Schrodinger) differential
equations
THE SPECIAL CANONICAL GROUP 121
in which we have made direct use of the operator
properties contained in (4.10) and (4.11). Hence
The magnitude of the constant C is fixed by the
composition property
while the phase of c is an intrinsically arbi-
trary constant which can be altered freely by
common phase factor transformations of the q
states relative to the p states. Hence, with
a conventional choice of phase, we have
and then
According to the structure of this transformation
function, eigenvectors in the q and. p repre-
sentations are related by the reciprocal Fourier
transformations
122 QUANTUM KINEMATICS AND DYNAMICS
4.4 DIFFERENTIAL STATEMENTS OF COMPLETENESS
Alternative forms, involving differentiation
rather than integration, can be given to these
latter expressions of completeness, either by di-
but also in mixed qp or pq representations,
With the aid of these connections, the unit opera-
tor can be exhibited, not only in terms of the
complete set of q , or p states,
THE SPECIAL CANONICAL GROUP 123
rect transformation, or from the following consi-
derations. An arbitrary vector ¥ is constructed
from the states q't> and the wave function
4>(q't) as (suppressing the reference to time)
in which we have introduced the operator ^(q) ,
and recognized the structure' of the state |p'>
with eigenvalues p'=0 . Employing a similar
procedure, we find
and thus the scalar product of two vectors can be
evaluated as
The use of the differential operator realizations
(4.8) or (4.9) now gives
124 QUANTUM KINEMATICS AND DYNAMICS
On abstracting from the arbitrary vectors, these
results become
and the adjoint operation produces the analogous
properties
or
THE SPECIAL CANONICAL GROUP 125
The practical utility of these forms depends, of
course, upon the differentiation operations appear-
ing in an algebraic manner. The basic examples of
this type occur in the differential composition
properties of the transformation functions <q'|p'>
and <p'Iq'> ,
and
4.5 NON-HERMITIAN CANONICAL VARIABLES
The significance of the special canonical
group will now be studied for the non-Hermitian
canonical variables y , iy , which are defined
by
126 QUANTUM KINEMATICS AND DYNAMICS
(In this discussion, we restrict the symbols q ,
p to represent Hermitian canonical variables).
The formal theory provides the infinitesimal gen-
erators
and these operators obey the commutation relations
To follow the pattern set by the Hermitian canoni-
cal variables, we must first construct the eigen-
vectors of the two complete sets of commuting
operators y and y
4.6 SOME TRANSFORMATION FUNCTIONS
The vectors |y't> are described in relation
THE SPECIAL CANONICAL GROUP 127
to the states <q't| by the transformation func-
tion <q't y't> , which obeys the differential
equation
Now the following forms of the generators,
and
permit all operators to be replaced by their eigen-
values ,
which gives the desired transformation function.
Since the adjoint of the right eigenvector equation
128 QUANTUM KINEMATICS AND DYNAMICS
we conclude that the complex conjugate of (4.41)
is
and this is a finite number. Hence the eigenvee-j. i
tors |y't> and <y t| exist for all complex
values of y' . It will be noted that the lengths
of these vectors depend upon the eigenvalues, the
is the left eigenvector equation for y ,
We can now compute the transformation function
which is the analogue of <p't q't> . In particu-
lar, the scalar product of a vector |y*t> with
its adjoint is given by
THE SPECIAL CANONICAL GROUP 129
eigenvector belonging to zero eigenvalues being
uniquely distinguished as the one of minimum
length. On normalizing the latter to unity, the
constant C is determined in magnitude, and, with
a conventional choice of phase, we have
for the substitution y •*• y , y -*• -y , while
formally preserving the commutation relation, con-
verts the non-negative operator y y into the non-
positive operator -yyt , Thus, unlike |y't>
A similar discussion for the transformation
i t *function <q't|y t> gives
i t'and no vector jy t> , or any linear combination
of these vectors, possesses a finite length. This
asymmetry between y and y , in striking con-trast with the situation for Hermitian canonical
variables is evident from the commutation relation
130 QUANTUM KINEMATICS AND DYNAMICS
and <y t| , the vectors |y t> and <y't| do
not exist. Yet we shall find it possible to define
vectors which, in a limited sense, are right
eigenvectors of y and left eigenvectors of y
4.7 PHYSICAL INTERPRETATION
The physical interpretation of the states
|y't> requires some comment. Since these vectors
are not normalized to unity, in general, we must
express the expectation value for such states as
Now
and, on writing
we infer that
THE SPECIAL CANONICAL GROUP 131
and
Thust the states |y't> are such that neither the
q nor the p variables have definite values but,
rather, the distribution about the expectation
values corresponds to an (optimum) compromise be-
tween the complementary aspects of the two incom-
patible sets of variables. We shall designate the
normalized vectors of this description as
in which notation the transformation function
(4.47) reads
The interpretation of the transformations
produced by G , , and G , on the states
<y t , and jy't> , respectively, is quite the
same as for the Hermitian variables, which the
construction of (4.41) implicitly assumes. Thus
+Y
132 QUANTUM KINEMATICS AND DYNAMICS
and generally,
Accordingly, the wave functions 4» (y t) and
(j)(y't) for a system described by non-Hermitian
canonical variables of the first kind obey the
Schrodinger equations
and
4. I
The transformations generated, by G on <y tj
and by Go. on |y't> are given byy
THE SPECIAL CANONICAL GROUP 133
These changes are infinitesimal multiples of the
original vectors, but unlike the situation with
Hermitian variables, the multiplicative factors
are not necessarily imaginary numbers. The same
comment applies to the infinitesimal multiples of
the unit operator produced by commuting two gener-
ators, as in (4,36). Hence the special canonical
group for non-Hermitian variables constitutes the
totality of transformations that change an eigen-
vector with specified complex eigenvalues into one
with any other set of eigenvalues, and which is
multiplied by arbitrary complex numerical factors.
4.8 COMPOSITION BY CONTOUR INTEGRATION
t'The transformation function <y t|y"t>
differs in form from <p'tjq't> only in the ab-
sence of the powers of 2 IT, and therefore obeys
the differential composition property
134 QUANTUM KINEMATICS AND DYNAMICS
We infer that
or, that scalar products of arbitrary vectors can
be computed as
Now one can give the evaluation at zero of these
complex eigenvalues a contour integral form (in
this discussion we place n=l , for simplicity;
the extension to arbitrary n is immediate).
Thus,
THE SPECIAL CANONICAL GROUP 135
where the integration paths encircle the origin in
the positive sense and enclose only the singularity
at the origin. Such paths can be drawn since the
existence of the limits in (4.64) requires that4, I
wave functions of the type $(y t) and $(y't)
be regular functions of the corresponding complex
variable in a neighborhood of the origin. Alterna-
tive forms obtained by partial integration are
where
These functions are regular in a- neighborhood
of the point at infinity. The integrations in
136 QUANTUM KINEMATICS AND DYNAMICS
(4.66) are to be extended through a common domain
of regularity of the two factors, with the enclosed
+'region containing all singularities of <j> (y t) ,j.»
or tfi(y't) » but no singularity of ty (y t) , or
cHy't) .4. I
If the wave functions ty(y't) and $(y t)
axe related to the vector bases <y't| and4-1
jy t> / given symbolically by
and
the scalar product evaluations can be presented as
the completeness properties
It is implied by these statements that
THE SPECIAL CANONICAL GROUP 137
and
where, according to the definition (4.68), and the
transformation function (4.47),
The indicated regularity domain corresponds to
that of wave functions of the types ^i(y') and
<Hy") . Indeed,
if the contour encloses y" but contains no sin-
gularity of 4>(y't) t while
if the integration path encloses a region continu-
ing all the singularities of 4>(y"t) but not the
point y1 . In the latter situation, the inte-
138 QUANTUM KINEMATICS AND DYNAMICS
gral is evaluated with the aid of a circle about
the point at infinity where the function i{i(y"t)
vanishes. A similar discussion applies to the
properties of the transformation function
The symbolic construction of <y't| is made
explicit on writing
and
With these forms we can examine to what extent
i t 1
<y't| and |y t> are left eigenvector of y(t) ,
where the integration path is extended to infinity
along any path that implies convergence of this,
and subsequent integrals. Thus
THE SPECIAL CANONICAL GROUP 139
and right eigenvector of y (t) , respectively.
Now
provided the contour encircles all the singulari-
ties of the integrand. The adjoint statement is
Hence, as our discussion of the asymmetry between
y and y would lead us to anticipate, <y't|
and y t> are not true eigenvectors , although
1The use of complex eigenvalues has been developedin a more formal way by P.A.M. Dirac, Cornm. Dub,Inst. for Adv. Studies, Ser. A, No. 1 (1943),without initial recognition of the asymmetry be-
or, as we can recognize from the symbolic form
(4.68)
+
+ 1
4.9 MEASUREMENTS OF OPTIMUM COMPATIBILITY
According to the eigenvector constructions
tween left and right eigenvectors of the non-Her-itdtian variables. With this procedure, the alter-native evaluations of <y*|y[y"> lead to resultsdiffering by unity and, generally, one is forcedto assume that two <|»(y') functions differingby an arbitrary non-negative power series in y1 ™describe the same state.
140 QUANTUM KINEMATICS AND DYNAMICS
the failure of this property appears in a very
simple form. The consistency of the theory, in
which the action of y on its left eigenvectors
contains the additional contour integral term of
(4.81) , can be verified from the alternative eval-
uations of <y llyly" > where, acting on the right
eigenvector,
whereas the properties of the left eigenvectors
give
THE SPECIAL CANONICAL GROUP 141
(4.78) and (4,79), the unit vector can be presented
as
which is the analogue of (4.24). Here the complex
t"variables y' and y are to be integrated along
orthogonal paths. If we write
the paths can be so deformed that q" and p'
are integrated independently from -« to » through
the domain of real values. This produces the form
The transformation function composition property
implied by this expression of completeness,
or, in the notation of (4.55),
can be verified by direct integration. Although
the vectors |q'p't> are complete, they are
certainly not linearly independent or the transfor-
mation function (4.56) would be a delta function.
Instead, it appears that displacing the eigenvalues
q'p' from q"p" f by amounts of the order of unity,
does not produce an essentially different state.
Changes in eigenvalues that are considerably in ex-
cess of unity do result in new states, however,
since the value of the transformation function
becomes very small. In a rough sense, there is one
state associated with each eigenvalue range
{ Aq' Ap1 )/2ir = "I . It is not necessary to con-
struct the linearly independent vectors that describe
distinct states if we are interested only in the
comparison with measurements on a classical level
(as the final stage of every measurement must be)
for then we are concerned with the probability
that the system be encountered in one of a large
number of states, corresponding to an eigenvalue
range ( Aq' Ap1 )/2ir » 1 . Within this context,
142 QUANTUM KINEMATICS AND DYNAMICS
THE SPECIAL CANONICAL GROUP 143
it can be asserted that
is the probability that q and p measurements,
performed with optimum compatibility on the state
f at the time t , will yield values q' and p1
lying in the intervals dq1 and dp1 , respectively.
II VA BLE j Hj ECONjp KIMD
4.10 ROTATION GROUP
Now we turn to the variables of the second
kind. The requirement that the variations Si,
antieommute with each variable of this type is
expressed most simply with the aid of the canonical
Hermitian variables £ . The enumeration of the^K
2 =4 distinct operators of the algebra shows
that there is only one operator with the property
of anticommuting with every £ • This is the
product
144 QUANTUM KINEMATICS JVND DYNAMICS
which is written as a Hermitian operator with unit
square. Thus the operator properties of the C
variables contained in (3.110) also apply to
I, - . We see that the variations 55 must be&.U * J» K
infinitesimal real multiples of the single operator
C2n+1 , say
acquires the form
On forming the commutator of two such generators
we obtain
where the
so that the infinitesimal generator
THE SPECIAL CANONICAL GROUP 145
comprise a set of %v(v-l) = n(2n-l) Hermitian
operators obeying
Thus the generators Gg and their commutators can
be constructed from the operator basis provided by -the
n(2n+l) Hermitian operators £ , , where K andK A
X range from 1 to 2n-fl . This basis is com-
plete according to the commutation property
and the totality of these transformations form a
group, which possesses the structure of the (proper)
rotation group in 2n+l dimensions [compare (2.80)] .
But this is not what we shall call the special
canonical group for variables of the second kind.
4.11 EXTERNAL ALGEBRA
We have been discussing a group of inner
automorphisms - transformations, constructed from
the elements of the algebra, that maintain all
algebraic relations and Hermitian properties. It
146 QUANTUM KINEMATICS AND DYNAMICS
is characteristic of the structure of the algebra
that the only operators anticommutative with every
| , K = l,...,2n , are numerical multiples of(C
£? , , and two variations formed in this way are
commutative. Hence the generator of one variation
does not commute with a second such variation and
two variations are not independent. To obtain
independent variations, the operators 6 C and
6 ' ' £ roust anticommute, and this is impossible for
inner automorphisms. But the algebraically desirable
introduction of independent variations can be
achieved - at the expense of Hermitian properties
by considering outer automorphisms, constructed
with the aid of a suitably defined external algebra.
Let e be a set of 2n completely anti-Is
commutative operators,
that commute with the elements of the physical
algebra. That anticommutativity property includes
and these operators cannot be Hermitian, nor is the
adjoint of any e - operator included in the set.
THE SPECIAL CANONICAL GROUP 147
We see that the 2n operator products e 5
are completely anticommutative among themselves,
and also antieommute with every £ , <— l,...,2n
Hence variations <SE defined as numerical multi-1C
pies of e^an-n wil1 obey
and
It is the latter property that unifies the generators
for the variables of the first and second kind, and
permits the commutator of two generators,3C
and G , to be evaluated aenerallv as
The canonical form is (4.3) which, as written, applies
to either type of variable. For the variables of
the second kind, 6 'x. A 5^2'x is linearly related
to the products e e.. , and these combinationsK A
commute with all operators of the physical algebra
1
148 QUANTUM KINEMATICS AND DYNAMICS
and of the external algebra. Hence, with either
kind of variable the generators of independent
variations together with their commutators form the
infinitesimal elements of a group, which is the
special canonical group,
4.12 EIGENVECTORS AND EIGENVALUES
The existence of this group for the variables
of the second kind enables one to define eigenvectors
of the complete set of anticommuting operators
q or p=iq , Let us observe first that only thealgebraic properties of the operators and variations
are involved in the infinitesimal transformation
equations
and
I 4. I
Furthermore, if q and q are quantitiesK, K
formed in the same way as the independent variations,
THE SPECIAL CANONICAL GROUP 149
anticommuting among themselves and with the operators
q and q , we can assert, that
Hence if vectors exist, obeying
and
so also do the vectors
exist, and these are similarly related to the
eigenvalues q + 6q and q 4- <Sq , respectively,We shall see that that the eigenvectors associated
with zero eigenvalues certainly exist, which implies
that eigenvectors of the type |q t> and <q tj
150 QUANTUM KINEMATICS AND DYNAMICS
can be constructed from the null eigenvalue states
by the operations of the special canonical group.
It should be remarked that these right and left
eigenvectors are not in adjoint-relationship, sinceI 4. I
there is no such connection between q and q
A vector obeying the equations
is an eigenvector of the Hermitian operators
q q , with zero eigenvalues.
The converse is also true, since
implies (4.110). The operators q q are commuta-fC rC
tive and indeed constitute a complete set of commut-
ing Hermitan operators for the variables of the
second kind. (These statements apply equally to the
non-Hermitian variables of the first kind.) And,
from the algebraic properties of the canonical
variables, we see that
THE SPECIAL CANONICAL GROUP 151
the spectrum of each operator q q contains onlyK K
the values 0 and 1 , Hence there is a state for
which all operators q q possess the value zeroIN K,
and the state is also described by a right eigenvector
of the non-Hermitian operators q , or the adjoint
left eigenvector of the q , belonging to the set
of zero eigenvalues. We should observe here that,
unlike the situation with non-Hermitian variables
of the first kind, there is complete symmetry between
the operators q and q . In particular, the
operators q q possess zero eigenvalues, whichK, 1C
are equivalent to the unit eigenvalues of q q ,
and the zero eigenvalue, right eigenvector of the
q and left eigenvector of the q also exist.
The possibility of defining the eigenvectors |q t>
and <q't] is then indicated by the differential
equations
Q+
=
+
+
+
+
152 QUANTUM KINEMATICS AND DYNAMICS
which possess interpretations analogous to those
of (4.1095.
Ill UNIFICATION OF THE VARIABLES
4.13 CONSTRUCTIVE USE OF THE SPECIAL CANONICALGROUP
In the progression from Hermitian canonical
variables of the first kind to the non-Hermitian
canonical variables of the second kind, the impor-
tance of the special canonical group has increased
to the point where one uses it explicitly to define
the eigenvectors of the canonical variables, rather
than merely investigating the effect of the trans-
formation group on independently constructed vectors
The former approach is universally applicable for,
with all type of variables, the zero eigenvalue,
right eigenvector of the canonical variables q
can be constructed and the general eigenvector de-
fined by a finite operation of the group,
The analogous general construction of left eigenvectors
is
The Hermitian variables of the first kind and the
non-Hermitian variables of the second kind permit,
in addition, the construction of the vectors
and the exceptional situation of the Hermitian
canonical variables of the first kind stems front the
unitary nature of the operator group for those vari-
ables, which deprives the zero eigenvalues of any
distinguished position. The significance of the
operations of the special canonical group on the
eigenvectors of the canonical variables is indicated
generally by
THE SPECIAL CANONICAL GROUP 153
they are the totality of transformations that alter
eigenvalues and multiply eigenvectors by commutative
factors.
The eigenvalues in the equations
and
are formed by multiplying an element of the external
algebra with the member of the physical algebra that
anticommutes with every C , or eguivalently, with1C
the totality of operators q and q . The1C (C
translation of (4.91) into the language of the
canonical variables presents the anticommutative
operators as
154 QUANTUM KINEMATICS AND DYNAMICS
and
+
THE SPECIAL CANONICAL GROUP 155
which we shall designate as p . This product,
being a function of the commuting operators q qK ** t
possesses definite values in the null eigenvalue
states,
Accordingly, in considering
the factor p that occurs in q can be replacedK.
by the number (~l)n and the result expressed by
iwhere the final q is entirely an element of the
external algebra. In a similar way
and
and the ad^oxnt statements are
156 QUANTUM KINEMATICS AND DYNAMICS
although the element of the external algebra that
appears in the latter equation lacks the numerical
factor (-l)n and therefore differs in sign from
that of (4.126) if n is odd. Hence it is only
for even n that complete symmetry between left
and right eigenvectors exists, which invites the
aesthetic judgment that no system described by an
odd number (of pairs) of dynamical variables of the
second kind exists in nature.
4.14 TRANSFORMATION FUNCTIONS
In constructing transformation functions, one
must eliminate explicit reference to the operators
of the physical system and express the transformation
function in terms of the eigenvalues which, for
variables of the second kind, are the elements of
the external algebra. The replacement of eigenvalues
that anticommute with the dynamical variables by
purely external quantities is accomplished generally
by equations of the type
and occurs automatically for products of eigenvalues.
With variables of the second kind, the transformation
function can be expressed more specifically as
THE SPECIAL CANONICAL GROUP 157
The transformation function <p'j«ll> is character-
ized for all variables by the differential expression
and differs in its integral form only by the numeri-
cal factors that express the normalization conven-
tions for the particular type of variable. To
achieve the universal form
we must remove the factor (2IT) that appears
for Hermitian variables of the first kind. This
will be done if all integrations are performed with
the differentials
ana delta functions correspondingly redefined:
158 QUANTUM KINEMATICS AND DYNAMICS
which is expressed symbolically by
Thus, for variables of the second kind, the scalar
product of two vectors can be computed from the
representative wave functions
by
since the square of any eigenvalue vanishes. The
<p'|q'> transformation function possesses the gen-
eral differential composition property
THE SPECIAL CANONCIAL GROUP
If non-Hermitian variables of the first kind
are excepted, one can construct the transformation
function
which, for variables of the second kind, becomes
We also have as the analogues of (4.137) and (4.138),
the wave functions
and the scalar product evaluation
159
160 QUANTUM KINEMATICS AND DYNAMICS
With the same exception, the transformation function
<q'|q"> and <p'|p"> are meaningful without quali-
fication. The differential equation
indicates that <q'|q"> is a function of the eigen-
value differences that vanishes on multiplication
with any of its variables. (For non-hermitian
variables of the first kind, the latter equation
would read
which is solved by <y'|y"> = ty'-y"5 .) With
Herraitian variables of the first kind these properties
define the delta function,
combined with
-1
THE SPECIAL CANONICAL GROUP
and we shall retain this notation for the correspond-
ing function referring to the second class of variable,
in which the product of n antieommuting factors
is arranged in some standard order, say 1,.. , ,n , as
read from left to right. For the similar transfor-
mation function referring to the variables q we
write
with the reversed sense of multiplication. The
consistency of these definitions follows from the
composition property
161
162 QUANTUM KINEMATICS AND DYNAMICS
for the opposite sense of multiplication in the two
products permits their combination without the inter-
vention of the sign changes that accompany the anti-
comrautativity of the eigenvalues.4. I f"
The transformation function <q |q > pro~
vides the connection between the wave functionst'
^Cq't) and i/i(q t) * ^e composition property
yields
and similarly
supplies the inverse relation
THE SPECIAL CANONICAL GROUP 163
We also have
and
To obtain an expression of completeness re-
ferring entirely to q - eigenvectors, we observe
that
If this symbolic form is realized by < wave
functions that are commutative with the eigenvalues,
or, if n is even, the scalar product evaluation
can be presented as
The integral notation is designed to evoke an analogy
and has no significance apart from its differential
definition. In a similar way we have
and
According to the definitions adopted for variables
of the second kind,
and
164 QUANTUM KINEMATICS AND DYNAMICS
THE SPECIAL CANONICAL GROUP 165
Thus the extension of the delta function notation to
tho variables of the second kind is not without
justification. Another related example of formulae
that are applicable to both Hermitian variables of
the first kind and the variables of the second kind
is obtained_from (4.153), written as
rith the variables q replaced by p = iq . On
inserting the integral representation (4.164) this
Futhermore, the composition property
appears as
and similarly
166 QUANTUM KINEMATICS AND DYNAMICS
becomes
For Hermitian variables of the first kind, these are
the reciprocal Fourier transformations stated in,
(4.22) .
4 .15 INTEGRATION
Although the integral notation is effective in
unifying some of the formal properties of the two
classes of variables, the nature of the operations
is quite distinct. Indeed, the symbol J has the
significance of differentiation for variables of the
second kind, and the inverse of differentiation for
variables of the first kind. This is emphasized by
the effect of subjecting the eigenvalues to a linear
transformation , q' -*• Xq' . For the Hermitian
variables, the differential element of volume in the
q' - space changes in accordance with
and (4.151) supplies the inverse formula
The latter result Is also expressed formally by
With the particular choice A=-l , we learn that
<5[g'] is an even function of the Hermitian variables,
but it possesses this property for variables of the
second kind only if n~ is even. An interesting
formal difference also appears on considering the
evaluation, by integration, of the trace of an oper-
ator. From the expression of completeness for
Hermitian variables of the first kind, (4.23), we
derive the integral formulae
THE SPECIAL CANONICAL GROUP 167
which implies that
But the delta function of the variables of the second
kind is defined as a product of anticommutative
factors, and therefore
168 QUANTUM KINEMATICS AND DYNAMICS
To obtain the analogues referring to variables of
the second kind, we first note the generally valid
differentiation formula, derived from (4.136), in
terms of the matrix representation <p'|x|q'> ,
and
Then, for variables of the second kind, we deduce
which restores somewhat the uniformity of the two
classes for it is the unit operator, commuting with
all variables, that plays the role of p for varia-
bles of the first kind. Since the trace is unaltered
on replacing X with pXp , one can aso write
pX in place of Xp , which is to say that odd
functions of the variables have vanishing trace.
The simplest example of a trace evaluation is
THE SPECIAL CANONICAL GROUP 169
With even n_ we have, for example
and these trace formulae can also be.expressed as
Of course, with a system requiring both types of
dynamical variables for'its description, the trace
operations referring to the two classes must be
superimposed. A formula of suitable generality for
170 QUANTUM KINEMATICS AND DYNAMICS
even n, , derived from (4.172), (4.174), and
(4.166) , is
4.16 DIFFERENTIAL REALIZATIONS
Finally, we note the universality of the
differential operator realizations
with their algebraic generalizations
which imply the Schrodinger differential equations
and
The <q't| and |p't> states do not permit such
general assertions, for, in addition to the asymmetry
characteristic of the non-Hermitian variables of the
first kind, we must distinguish between the two
classes of variables in the differential operator
realizations
as contrasted with
THE SPECIAL CANONICAL GROUP 171
where the upper sign refers to variables of the first
kind. This sign distinction originates from the
necessary change of multiplication order in the
differential expression
172 QUANTUM KINEMATICS AND DYNAMICS
Thus the following Schrodinger equations are appli-
cable to Hermitian variables of the first kind and
to the variables of the second kind,
CHAPTER FIVECANONICAL TRANSFORMATIONS
5.1 Group Properties and Superfluous 175Variables
5.2 Infinitesimal Canonical Transformations 1785.3 Rotations. Angular Momentum 1825.4 Translations. Linear Momentum 1855.5 Transformation Parameters 1875.6 Hamilton-Jacobi Transformation 1905.7 Path Dependence 1915.8 Path Independence 1945.9 Linear Transformations 195
yields the corresponding canonical form of the
generator describing infinitesimal transformations
at a specified time t ,
173
The use of a canonical version of the Lagran-
gian operator, such as
which implies the canonical commutation relations
and equations of motion. From a given generator
G other generators G can be obtained, in accor-
dance with
appears to be the result of subjecting the variables
in the action operator W to infinitesimal varia-
tions, which for q and 3 , are the special
operators variations. However, the action opera-
tor can also contain other, superfluous, canonical
variables v , and we infer the differential
equations
174 QUANTUM KINEMATICS AND DYNAMICS
and we now ask whether G can also be exhibited in
a canonical form referring to new dynamical variables
q(t) , pCt) , and a new Hamiltonian operator
H (qpt) . Such new variables would then obey
canonical commutation relations and equations of
motion, characterizing the transformation of dyna-
mical variables at the time t as a canonical
transformation. The differential form
CWQMIC&L TRANSFORMATIONS 175
A canonical transformation is obtained should these
implicit operator equations possess a solution for
q and p
5.1 GROUP PROPERTIES AND SUPERFLUOUS VARIABLES
Canonical transformations form a group. The
action operator that describes the transformation
inverse to qp -*• qp is
The latter form illustrates the concept of super-
fluous variable. If the individual operators
w(q t g) and w(q , q) contain just the indicated
variables, the sum (5.7) involves q, q and q
while, for two successive transformations,
qp ~** QP "* QP » ^ne generating action operator of
the composite transformation is
.
176 QUANTUM KINEMATICS AND DYNAMICS
But, according to the statements of the individual
canonical transformations,
and it must be possible to exhibit W(q , q) as a
function of the variables q , q only. It is not
always desirable, however, to eliminate the super-
fluous variables. This is particularly true when
the canonical transformation involves algebraic rela-
tions between the variables q and q , which
inhibit their independent variation. By retaining
the variables of a suitable intermediate transfor-
mation, one can deiive the desired transformation
by independent differentiation. An important ex-
ample is provided by the identity transformation.
Let us remark first on the now familiar transforma-
tion that interchanges the roles of the complementary
q and p variables, as described by
where
CANONICAL TRANSFORMATIONS
Since
which contains the p as superfluous variables.
To eliminate the latter we must impose the transfor-
mation equations q=q , which yields W=0 for the
identity transformation. But, with the superfluous
variables retained, the differential equations (5.5)
are applicable and generate the transformation.
A given canonical transformation can be derived
by differentiation with respect to either set of
complementary variables. Thus, from the action
operator W(qqt) , obeying the differential equations
this is the canonical transformation
On adding the action ope,ators for the transformation
q -* p and its inverse p •*• q = q , we obtain the
following characterization of the identity transfor-
mation ,
177
178 QUANTUM KINEMATICS AND DYNAMICS
{5.5}, we obtain
and these equations are equally suitable for des-
cribing the transformation qp •*• qp . The identity
transformation, for example, is derived from the
action operator - p q
5.2 INFINITESIMAL CANONICAL TRANSFORMATIONS
Transformations in the infinitesimal neighbor-
hood of the identity - infinitesimal canonical
transformations - must be described by an action
operator that differs infinitesimally from the one
producing the identity transformation. The appro-
priate form for the variables q , q is
in which the q appear as superfluous variables.
The differential properties of the new action opera-
tor W(pqt) are now deduced to be
.
where G is an infinitesimal function of the in-
dicated variables that should have an even depen-,
dence upon, the second class of dynamical variable,
but is otherwise arbitrary. On applying (5.5),
with the p as superfluous variables, we obtain-
the explicit equations of an infinitesimal canonical
transformation,
Now, according to (3.100) and (3.123), which applies
to even functions of the variables of the second
kind, we have
or, without specialization to canonical variables,
with
CANONICAL TRANSFORMATIONS 179
and if G is a Hermitian operator this is a unitary
transformation. The subgroup of canonical transfor-
mations that preserve the Hermiticity of dynamical
variables is equivalent to the group of unitary
transformations. Without reference to Hermitian
properties the transformation (5.21) maintains all
algebraic relations, and therefore
which makes explicit the functional form of the new
Hamiltonian. Infinitesimal canonical transformations
that do not change the form of the Hamiltonian
operator.have infinitesimal generators that are
constants of the motion.
180 QUANTUM KINEMATICS AND DYNAMICS
Hence
whence
CANONICAL TRANSFORMATIONS 181
We have already encountered examples of in-
finitesimal canonical transformations. The tran-
formation generated by G, = - H <5t ,
is one in which the dynamical variables at time t
are replaced by those at time t + <St ,
and the energy operator H is a constant of the
motifcn when H is not an explicit function of t ,
being the condition for the maintenance of the func-
tional form of the Hamiltonian operator under time
translation. The special operator variations, which
are distinguished by their elementary commutation
properties, appear as the canonical transformations
The new form of the Hamiltonian operator is
182 QUANTUM KINEMATICS AND DYNAMICS
and
in Which we have introduced a notational distinction
between the general infinitesimal canonical trans-
formations and those of the special canonical group.
For variables of the second kind, the consideration
of the special canonical group within the framework
of general canonical transformations implies a for-
mal extension of the latter through the introduc-
tion of the elements of the external algebra.
5.3 ROTATIONS. ANGULAR MOMENTUM
The change in description that accompies a
rotation of the spatial coordinate system is a
canonical transformation, with the infinitesimal
generator
but the form of this canonical transformation is not
yet known. Differently oriented coordinate systems
are intrinsically equivalent and we should expect
that the kinematical term in the Lagrangian operator
presents the same appearance in terms of the varia-
bles appropriate to any coordinate system. This
comment also applies to the dynamical term - the
Hamiltonian operator - of a physically isolated
system, for which the total angular momentum, opera-
tor J , as the generator of a transformation
that leaves the form of H invariant, is a constant
of the motion. In view of the bilinear structure
of the kinematical term, p f -T*. , the change in-volved in an arbitrary rotation of the spatial
coordinate system will be a linear transformation
among suitably chosen q variables, combined with
the contragredient transformation of the complemen-
tary variables. For an infinitesimal rotation,
then,
CANONICAL TRANSFORMATIONS 183
in which the components of the vector j are
matrices. For Hermitain variables j is an imag-
inary matrix, while with non Hermitian variables
in the relation p = iq , the matrix j is
Hermitian. The general form of the Hermitian angu-
lar momentum operator is thus obtained as
to which the two kinds of dynamical variables make
additive contributions. The symmetrization or anti-
symmetrization indicated here is actually unnecessary,
On applying the infinitesimal transformation
generated by G^ to the operator J we findw
according to the commutation relations (2.79). But
we also have
the matrices j obey the same commutation relations
as the angular momentum operator J . It is a
consequence of these commutation relations that the
trace of the matrix j vanishes and thus the
and therefore
184 QUANTUM KINEMATICS AND DYNAMICS
explicit symmetrization or antisyiraaetrization of
factors in J is unnecessary. The decompositon
of the matrix j into irreducible submatrices pro-
duces a partitioning of the dynamical variables in-
to kinematically independent sets that appear add-
itively in the structure of J . Each such set
defines a dynamical variable of several components,
the rotational transformation properties of which
are fixed by the number of components, for this
integer, the dimensionality of the corresponding
submatrix of j , essentially determines the
structure of these matrix representations of J
An irreducible set of three variables, for example,
necessarily has the rotational transformation proper-
ties of a three-dimensional space vector. The num-
ber of components possessed by a dynamical variable
of the second kind is presumably even, according to
a comment of the preceding section.
5.4 TRANSLATIONS. LINEAR MOMENTUM
The remarks concerning invariance with re-
spect to rotations of the coordinate system appJ._
equally to coordinate system translations, which
have the infinitesimal generator
CANONICAL TRANSFORMATIONS 185
QUANTUM KINEMATICS AND DYNAMICS
For systems described by a finite number of dynami-
cal variables, the appropriate transformation that
leaves the kinematical part of L invariant is the
addition of constants to suitably chosen canonical
variables q . Thus G has the structure of G
and different generators of this type do commute,
as the commutation properties of the total linear
momentum require. If we apply an infinitesimal
rotation to the operator
we recognize that the members of the class of dynam-
ical variables p that make a contribution to P
have the rotational transformation properties of
space vectors. Hence, only three-component variables
of the first kind can be affected by a translation
of the coordinate system. If the latter set of
variables is presented as the Hermitian vectors
in accordance with
186
r, » ... t T , with the complementary variables
p ,.. ., p , a suitable adjustment of the rela-
tive eigenvalue scales will guarantee that
and therefore
One exhibiting the contribution of the vector varia-
bles to the total angular momentum, we have
where the latter term contains all variables that
are uninfluenced by translations. These are evident-
ly the internal variables for a system of n part-
icles that are localized spatially by the position
vectors r,
5.5 TRANSFORMATION PARAMETERS
It is useful to regard a general infinitesimal
canonical transformation as the result of subjecting
certain parameters T , s = l,...,v , to
CANONICAL TRANSFORMATIONS 187
infinitesimal changes, say -dT , so that thes
infinitesimal generator has the form
in which the G, . may depend explicitly upon thevs /
parameters. This interpretation of the transforma-
tion is expressed by
or
Accordingly, the canonical variables obey equation
of motion,
which govern the evolution of the canonical trans-
formation. By repeated application of such infini-
tesimal transformations, a finite transformation is
generated in which the parameters t are altered
188 QUANTUM KINEMATICS AND DYNAMICS
from T, to T. along a definite path. The
action operator characterizing the finite transfor-
mation is the sum of those for the individual in-
finitesimal transformations
is the action operator generating the finite canon-
ical transformation, with the operators referring
to all values of T intermediate between t. and
i appearing as superfluous variables. Since
W,„ must be independent of these intermediate
variables, it is stationary with respect to infini-
tesimal special variations of all dynamical quanti-
ties that do not refer to the terminal values of
where, according to (5.16),
CANONICAL TRANSFORMATIONS 189
Hence
190
and thus the stationary requirement, applied to a
given parameter path (61 = 0) , again yields the
differential equations (5.44), and
5.6 HAMILTON - JACOBI TRANSFORMATION
With a single parameter t and generator
-H , we regain the original action principle, now
appearing as the characterization of a canonical
transformation - the Hamiltonian - Jacobi transfor-
QUANTUM KINEMATICS AND DYNAMICS
which in turn, equals
Now
mation - from a description at time t (=t.) to
the analogous one referring to another time
tfl (=t?) . With t_ held fixed, the action opera-tor W (=W,2) obeys
or
On comparison with {5.5} we recognize that the Ham-
ilton-Jacobi transformation is such that 5 = 0 ,
which expresses the lack of dependence on t of
the new dynamical variables x(t } , The new
Hamiltonian at time t thus differs from H eval-
uated at time t~ , which governs the dependence
of W upon the parameter t_ ,
5.7 PATH DEPENDENCE
For a. canonical transformation involving
several parameters, the last term of (5.49) displays
CANONICAL TRANSFORMATIONS 191
the effect of altering the path along which the
parameters evolve. According to the significance
of G, , as a. generator, we have\s i
is antisymraetrical with respect to the indices r
and s ,
The complete variation of W.„ is thus
With fixed terminal conditions, the consideration
of two independent path variations in the combinat-
ion
and therefore
192 QUANTUM KINEMATICS AND DYNAMICS
CANONICAL TRANSFORMATIONS 193
leads to the integrability condition of the differ-
ential form (5.5?) in its dependence upon the para-
meter path,
which is indeed satisfied by virtue of the operator
identity (2.73), for
and
5.8 PATH INDEPENDENCE
If the canonical transformation is to be in-
dependent of the integration path, it is necessary
that
As an example of this situation, consider a canoni-
cal transformation with two sets of parameters and
generators: t , -H ; X , G,,, . On referringI A;
to (5.55) we see that the condition for path inde-
pendence can be presented as
Hence the Hamiltonian operator must be an explicit
function of the parameter X , which is to say that
it changes its functional form under the infinitesimal
transformation G,,v dX « Since this change isI A;
identical with (5.23), characteristic of an infini-
tesimal canonical transformation, we learn that the
same resultant canonical transformation is obtained
whether the system evolves in time and a canonical
transformation is performed at the terminal time,
194 QUANTUM KINEMATICS AND DYNAMICS
or if the canonical transformation is applied
continuously in time, subject only to the fixed
endpoint. With the latter viewpoint, the superposi-
tion of the continuous change in description on the
dynamical development of the system is described
by the effective Hamiltonian operator
and the principle of stationary action includes the
numerical variable X(t) . It should be noted that
when G,,v is not an explicit function of time,(A I
the X dependence of H is such that dH/dX = 0.
5.9 LINEAR TRANSFORMATIONS
under some circumstances, this extension of
the action principle can be expressed as a widening
of the class of variations, without alteration of
the Hamiltonian. Thus, let
CANONICAL TRANSFORMATIONS 195
which produces the linear transformation described
by
the action operator is thereby expressed as that of
a pure Hamilton-Jacob! transformation. But the X
transformation can now be introduced by remarking
196 QUANTUM KINEMATICS AND DYNAMIC:
If g is a constant matrix, the explicit X trans-
formation, for constant t , is
If these linear operator relations are substituted
into the Lagrangian operator, we obtain
from which all reference to the X transformation
has disappeared. We have used the properties
for fixed X , and
that a. special variation of q(A5 and an infinit-
esimal change of A , in
implies a special variation of q , combined with
a linear transformation:
Together with the similar properties of p , this
yields an extended class of variations for the
action principle. To verify directly the correct-
ness of this extension, we observe that the latter
induces the following variation in L »
in which 3H/3A has been introduced to measure the
lack of invariance displayed by H under the A
transformation. The application of the stationary
action principle now properly yields (5.63) and
confirms the interpretation of Gi\\ ^A as ^e
generator of the infinitesimal A transformation.
CANONICAL TRANSFORMATIONS 197
We may well note here the special situation
of the linear Hamilton-Jacobi transformation,
corresponding to the bilinear Hamilton!an operator
and the equations of motion
Since
in virtue of the equations of motion, the action
operator W(qq0t} is identically zero, which in-
dicates the existence of algebraic relations between
the variables q and qQ . The transformation is
more conveniently described with the aid of
WCpq/vt) . According to (5,14), we must eliminate
the variables q , which is accomplished by the
explicit solution of the equations of motion,
(h constant) , and thus
198 QUANT0M KINEMATICS AND DYNAMICS
CANONICAL TRANSFORMATIONS 199
This page intentionally left blank
CHAPTER SIX
GROUPS OF TRANSFORMATIONS
6.1 Integrability Conditions 2026.2 Finite Matrix Representation 2046.3 Subgroups 2076.4 Differential Forms and Composition 209
Properties6.5 Canonical Parameters 2116.6 An Example, Special Canonical Group 2166.7 Other Parameters, Rotation Group 2196.8 Differential Operator Realizations 2266.9 Group Volume 2286.10 Compact Groups 2316.11 Projection Operators and Invariants 2336.12 Differential Operators and the 238
Rotation Group6.13 Non-Compact Group Integration 2436.14 Variables of the Second Kind 2476.15 Reflection Operator 2496.16 Finite Operator Basis 2506.17 Addendum: Derivation of the Action 254
Principle6.18 Addendum Concerning the Special 259
Canonical Group6.19 Addendum: Quantum Variables and the 275
Action Principle
201
6.1 INTEGRABILITY CONDITIONS
We will now examine the construction from its
infinitesimal elements of a. continuous group of
canonical transformations, where a transformation
must be completely specified by the values of the
parameters and thus is independent of the integra-
tion path. Apart from the elementary situation of
a completely commutative (Abelian) group, the gene-
rators G, . must be explicit funtions of the
parameters if the operators R are to be zero.JL 5
The group property is exploited in exhibiting the
operators G, , (x , t) , assumed finite in number,\ s}
as a linear combination of an equal number of
operators that do not depend explicity upon the
parameters,
202 QUANTUM KINEMATICS AND DYNAMICS
for the condition of path independence demands that
the commutators of the operators G, » (x) be line-\3.)
arly related to the same set. On writing
we obtain the following differential equations for
the functions C c(t)O.S
The numbers g , are antisymmetrical in the last
two indices, and they are imaginary if the operators
G, v are Hermitian. Other algebraic propertiesv3/
can be conveniently presented by introducing a ma-
trix notation for the array with fixed second index,
The latter establishes a correspondence between the
operator G,, , and the finite matrix g, . This
correspondence maintains commutation properties
according to the identity (2.73),
while writing the commutation relations (6.2) in
the form
GROUPS OF TRANSFORMATIONS 203
QUANTUM KINEMATICS AND DYNAMICS
and thus the g matrices also obey the commutation
relations (6.2). A second set of matrices withm
that property, -g , follows from the correspon-
dence
The quadratic connections among the g coefficients,
comprised in the commutation relations, are identi-
cal with the conditions of integral-ility for the
differential equations (6.3), which verifies the
consistency of the operator presentation (6.1).
6.2 FINITE MATRIX REPRESENTATION
The correspondence between the operators G, ,
and the matrices g persists under a change ofa
operator basis, to within the freedom of matrix
transformations that preserve algebraic relations;
the non-singular transformation
induces
204
GROUPS OF TRANSFORMATIONS
An important application of this ability to change
the operator basis occurs when the operators G
are Hermitian and possess a non-zero, linearly in-
dependent, finite-dimensional, Hermitian matrix re-
presentation. Then, with the trace computed from
the bounded matrix representation,
is a real, symmetric, positive-dafinite matrix.
Accordingly, there is a choice of Hermitian basis
for which Y is a multiple of the unit matrix, and
g . is completely antisyinmetrical. Relative to
this basis, which still has the freedom of orthogo-
nal transformations, the g matrices are antisyramet:
rical,
is completely antisymmetrical in a , b , and c ,
where
205
QUANTUM KINEMATICS AND DYNAMICS
and, being imaginary, are Hermitian matrices. Thus
the g matrices qualify as a finite-dimensional
matrix representation provided they are linearly
independent. A linear relation among the g matri-
ces will occur only if a linear combination of the
operators G, v , a = l,...,v, commutes with every\3.l
G . Should such a linear combination exist, let
it be labelled G, , by an appropriate orthogonal
basis transformation. Then
and g . = 0 , which states that G, , will never
appear in the expression for any commutator. This
procedure can be continued if there are several such
linear combinations and we reach the conclusion
that the group can be factored into an Abelian group,
and a non-commutative group with its structure
characterized by the property that the g matrices
constitute a finite dimensional representation of
the generating operators.
206
6.3 SUBGROUPS
Groups of the latter type are necessarily
semi-simple, by which is meant that they possess no
Abelian invariant subgroups (a. simple group has no
invariant subgroup). The significance of these
terms can be given within the framework of infini-
tesimal transformations. Let the generators be
divided into two sets, designated as 1 and 2 ,
of which the first refers to the subgroup. Then,
as the condition for the formation of commutators
to be closed within the subgroup, we have
The subgroup is invariant if the commutator of any
subgroup element with an outside operator is still
within the subgroup,
and, if the subgroup is Abelian,
invariant subgroup:
subgroup:
GROUPS OF TRANSFORMATIONS 20?
Abelian subgroup
Then, if the group possesses an Abelian invariant
subgroup, the only non-zero elements of a matrix
g, are of the form g , , and the matrixb l alblC2
g, cannot be antisymmetrical. Alternatively, weDlconclude from these attributes of an Abelian invar-
iant subgroup that
in contradiction with the positive-definiteness
this array of numbers should exhibit if the matrices
g constitute a finite-dimensional representation.alEvidently a group that contains an Abelian invariant
subgroup cannot possess finite dimensional matrix
representations. A fundamental example of this
situation is provided by the group of translations
and rotations in three-dimensional space. On re-
ferring to the commutation properties (2.80), we
recognize that translations form an Abelian invari-
ant subgroup, and the mathematical impossibility of
a finite dimensional representation corresponds to
the physical existence of an infinite number of
states that are connected by the operation of tran-
slation. In contrast, the subgroup of rotations,
considered by itself, is a simple group and every
208 QUANTUM KINEMATICS AND DYNAMICS
matrix representation, labelled by the value of
the total angular momentum, is of finite dimension-
ality,
6.4 DIFFERENTIAL FORMS AND COMPOSITION PROPERTIES
The transformation described by the infinitesi-
mal changes dT of the parameters is produced by
the operator
where the quantities
form a set of inexact differentials (Pfaffians).
The subscript A(eft) refers to the manner in which
this operator is combined with the operator U(T)
that produces the finite transformation from the
standard zero values of the parameters, namely
GROUPS OF TRANSFORMATIONS 209
when the dynamical variables are referred to the
standard values of the parameters with the aid of
the transformation
This illustrates the general composition property
of the group,
for infinitesimal t, . In addition to the infin-
itesimal transformation U(T+dT) U(t) one can
—Iconsider U(T) u(t+dt) r and there must exist a
second set of inexact differentials, <5 T suchXT ct
that
An infinitesimal change of T.. in the general
multiplication property induces a corresponding
change of t and,
210 QUANTUM KINEMATICS AND DYNAMICS
GROUPS OF TRANSFORMATIONS
serve to determine the composition properties of
the group parameters. The same function, as per-
formed by the second set of differentials, is ex-
pressed by the differential equations
6.5 CANONICAL PARAMETERS
The choice of parameters is arbitrary to
within non-singular transformations, t -* T ' ,
and the initial conditions
211
or
The ensuing differential equations (Maurer-Cartan)
together with the initial conditions
which do not affect the inexact differentials,
The differential equations (6.3) maintain their
form under parameter transformation. Through the
freedom of parameter and basis changes one could
require identity of the basis operators G, v with\a,l
the generators G, ,(T) , for T = 0 , This would
be expressed by adding the initial condition
to the 5 differential equations, A special set
of parameters, termed canonical, is defined as
follows. As the number X varies from 0 to 1
let a point in the T-parameter space move out from
the origin along the curve described by
and thus
212 QUANTUM KINEMATICS AND DYNAMICS
where the t are arbitrary constants. The pointa
in the t-space that is reached at X = 1 is deter-
mined by the numbers t , which constitute thect
new set of parameters. According to the invariance
of the differential forms the same path is described
in the t-parameter space by
Now if t is replaced by Yt t Y < 1 > thea a
point reached for X = 1 is identical with the
point attained at X = y along the curve character-
izing the point t , Hence the path appears in
the t-space as
a straight line, and (6,34) asserts that
or
GROUPS OF TRANSFORMATIONS 213
and thus the finite transformations of the group
have the exponential form
and, on applying this equation at the point At ,
we obtain
becomes
in terms of the canonical parameters.
The.differential equations for the functions
C »_(t) can be simplified with the aid of the
property (6.37). Indeed,
214 QUANTUM KINEMATICS AND DYNAMICS
The operator producing the infinitesimal transfor-
mation characterized by dA is
GROUPS OF TRANSFORMATIONS
and the formal solution is
Thus the differential forms 6,t are obtained
explicitly as
The substitution t •*• ~t~cft , dt -*• dt , converts
U(t+dt) U(t)"1 into U(t)~ U(t-t-dt) and therefore
In a matrix notation the latter reads
215
in which we have used an evident notation. The
structure of 6 t can now be inferred from ther
following property of the canonical parameters,
QUANTUM KINEMATICS AND DYNAMICS
The general expression of this connection between
the two differential forms, characteristic of the
canonical parameters, is the group composition
property
6.6 AN EXAMPLE. SPECIAL CANONICAL GROUP
A simple illustration of these considerations
is provided by a group of three parameters defined
by the commutation relations
The matrices g are conveniently presented in thecl
linear combination
which asymmetrical form shows that
216
GROUPS OF TRANSFORMATIONS 217
Thus the matrices g do not furnish a finite-di-a
mensional Hermitian matrix representation, which is
related to the existence of the one-parameter in-
variant subgroup generated by £*»,» . In virtue
of the algebraic property (6.51), we have
and the differential equations (6.26) read (primes
are now used to distinguish the various parameter
sets)
The solution of these equations subject to the
initial condition t' = 0 : t = t" , and the
canonical parameter composition law of the group,
is
218 QUANTUM KINEMATICS AND DYNAMICS
and specializations of this result, in which the
parameters are combined with the generators, can be
presented as
It will not have escaped attention that the
commutation properties (6.49) are realized by the
special canonical group. Accordingly, the operators
expressing the finite transformations of this group
possess the multiplication property
which illustrates the reflection property (6.48).
The operator statement contained here is
GROUPS OF TRANSFORMATIONS
6.7 OTHER PARAMETERS . ROTATION GROUP
Canonical parameters are not always the most
useful parameter choice. This can be illustrated
by the three-dimensional rotation group. We combine
the three canonical parameters into the vector w ,
and observe that the three-dimensional Hermitian
matrices j(=g } , defined by the commutation re~a
lations (2.81), can also be presented as a vector
operation:
Hence
which is applicable to all types of dynamical
variables. Specializations analogous to (6.56) are
219
220 QUANTUM KINEMATICS AND DYNAMICS
and
The latter result shows that the eigenvalues of any
component of j are 1 , 0 , -1 , and therefore
On applying this result to the explicit construction
of $£**/ as given in (6.45) , we obtain
or, with a simple rearrangement.
GROUPS OF TRANSFORMATIONS 221
Thus, the new parameters
are such that
The substitution u> -* -w converts u into -u
while leaving un unaltered, and therefore the
analogous expression for the differential form 6 u
is
Although u is not an independent parameter,
it can largely be treated as such. This is indicated
by the structure of the differentials du , du«
that a given 60w or & u imply:J6 3T
222 QUANTUM KINEMATICS AND DYNAMICS
and
Thus, there is a unit vector in a four-dimensional
Euclidean space associated with every three-dimen-
sional rotation, and, to the composition of two three-
dimensional rotations there is associated a four-
dimensional rotation. The algebraic simplification
achieved by the u-parameters appears in the group
composition law. The invariance of the bilinear
form for the differentials 6ew, expressed by theX*
differential equations (6,26), implies a linear
relation between the parameters of the individual
transformations and those of the product transfor-
mation. One easily verifies that
for these changes maintain the normalization (6.67),
The u-parameters appear in another way on
recalling the existence of automorphisms - unitary
transformations - of the algebra defined by 2n
Hermitian canonical variables of the second kind,
£ , that have the structure of the Euclidean ro-K
tation group in 2n + 1 dimensions. Hence, a
representation of the three-dimensional rotation
group is generated by such unitary transformations
of the three anticommutative Hermitian operators
where e is the completely antisymmetrical func-
tion of its indices specified by £,„- = +1
The generators of the three independent infinitesimal
which have the following multiplication characteris
tics:
ar, of the operators
GROUPS OF TRANSFORMATIONS 223
224 QUANTUM KINEMATICS AND DYNAMICS
rotations are the operators %0 for, according to
(4.96) ,
and
Explicit matrix representations are obtained on
relating the operator basis 1 , a to measurement
symbols. The measurement symbols of the 0,,
representation are presented in the following array,
and (Pauli)
for example. Theus, the simplest not-trivial measur-
ment algebra, of dimensionality 2 2, provides an
angular momentum operator matrix representation
Now, apart from the freedom of multiplication by a
numerical phase factor, any unitary operator of
this algebra has the form
where the four numbers comprised in un , u are
real, and obey
The latter condition also states that U is unimo-
dular,
With any such unitary operator there is associated
a three-dimensional proper- rotation
GROUPS OF TRANSFORMATIONS 225
226 QUANTUM KINEMATICS AND DYNAMICS
which correspondence is 2:1 since U and -U
produces the same rotation, and successive unitary
transformations generate successive rotations. The
parameter composition law that emerges from the
product
where j signifies the three-dimensional angular
momentum matrix representation displayed in (6.60).
6.8 DIFFERENTIAL OPERATOR REALIZATIONS
The differentiable manifold of the group
parameters enables differential operator realiza-
tions of the infinitesimal generators of a group
to be constructed. Let us define for this purpose
two sets of functions rj (-T) , t, (T) , accordingas as
to
is just (6.73). Incidentally, the explicit form
of the three-dimensional rotation matrix r is
GROUPS OF TRANSFORMATIONS 227
and, using canonical parameters
Now, the infinitesimal composition properties
stated in (6.20) and (6.23) can be presented as
or, on writing
as
The two sets of differential operators defined here
are commutative,
Thus
228 QUANTUM KINEMATICS AND DYNAMICS
These are intrinsic properties of the differential
operators. Thus, one can verify that the differ-
ential equations implied for n(t) ,
are a direct consequence of (6.3) and the relation
(6.88).
6.9 GROUP VOLUME
An infinitesimal element of volume can be
defined on the group manifold with the aid of the
and each set obeys the G commutation relations
This volume element is independent of the choice of
parameters, and is unchanged by the parameter trans-
formation, T- •+ i , that expresses group multipli-
cation, T = T(T. , T2) • An alternative definition
of volume accompanies the differentials 6 i , and
that volume element is invariant under the parameter
transformation T, -»• T of group multiplication.£*
The relation
shows that the two definitions or volume are iden-
tical if
The latter property certainly holds if the group
possesses a finite-dimensional matrix representation,
for then all the g matrices are antisymmetrical
and its consequence
GROUPS OF TRANSFORMATIONS 229
inexact differentials (6.19)
230 QUANTUM KINEMATICS AND DYNAMICS
(or zero) to within the latitude of matrix transfor-
mations that do not alter the trace. But, as the
example of (6.50) indicates, this is by no means a
necessary condition. The invariance of the volume
element, stated by
is given a differential form on choosing the para-
meters T to be infinitesimal, in which circum-
stance the explicit transformation is
and we infer that
As an application of this result, let us observe that
GROUPS OF TRANSFORMATIONS 231
which transformation maintains the commutation
properties of the differential operators, and yields
a. formally Hermitian differential operator to repre-
sent a Hermitian generating operator Gf \ . The(a)
study of a v-parameter group of unitary transforma-
tions can thus be performed with the aid of an equi-
valent dynamical system described by v pairs of
complementary variables of the first kind which are
generally quasi-canonical for, unless the range of
the individual parameters is -°° to » , these
variables do not possess all the attributes of
canonical variables.
6.10 COMPACT GROUPS
The ability to integrate over the group mani-
fold is particularly valuable when the group is
compact, which is to say that any infinite sequence
of group elements possesses a limit point belonging
to the group manifold. Thus the manifold of a
compact group is bounded, and its volume can be
chosen as unity by including a suitable scale factor
in the volume element. We first notice that the
matrices g for a compact group are necessarilya
232 QUANTUM. KINEMATICS AND DYNAMICS
traceless and, accordingly, the two definitions of
volume element are identical. To prove this consi-
der a particular transformation U , as character-
ized by canonical parameters t , and the corres-
ponding finite matrix
of which (6.5) is the infinitesimal transformation
form. If the imaginary matrices g are not anti-
symmetrical,
may differ from unity. Then, to the sequence of
koperators U , k = ± l , ± 2 , . . . , there corresponds
ka sequence of matrices U , for which
increases without limit as k -»• +°° , if det U > 1 ,
or, as k -»• -•» , if det U < 1 . This contradicts
the requirement that an infinite sequence of group
which appears in the general relation
GROUPS OF TRANSFORMATIONS 233
elements possess a limit point on the group manifold
with its associated finite U matrix. Hence the
matrix U roust be unimodular, and every gQ has acL
vanishing trace. It may be noted here that a group
for which the g matrices supply a representation
of its infinitesimal Hermitian generators has a
bounded manifold. According to the explicit con-
struction of the volume element in terms of the
canonical parameters,
and the boundaries of the manifold are reached when
the weight factor in the element of volume vanishes.
The statement that the g matrices are Hermitian and
linearly independent implies that, for every t , gt
possesses non-zero, real eigenvalues. The numer-
ically largest of these eigenvalues equated to 2ir
then determines the finite points where the boundary
of the group manifold intercepts the ray, directed
from the origin of the parameter space, which is
specified by the relative values of the parameters
*a '
6.11 PROJECTION OPERATORS AND INVARIANTS
since intergration with respect to t , or -t ,
covers the group manifold and the volume element is
invariant under the transformation t ->• -t . Thus,
the Hermitian operator P_ is a measurement symbol
or, in geometrical language, a projection opera-
tor, for the subspace of states that are invari-
ant under all the transformations of the group.
These are also the states for which all the
generating operators can be simultaneously
234 QUANTUM KINEMATICS AND DYNAMICS
The group property and the invariance aspects
of the volume element for a compact group assert
that the operator
has the following characteristics
and, using the canonical parameters,
describes the construction of the subalgebra of
operators that are invariant under all transforma-
tions of the group. One can also apply a slight
modification of the latter procedure to the finite,
real matrices U{T} , with the result
This is a real symmetrical positive definite matrix
and therefore it can be expressed as the square of
a matrix of the same type, say X . Thus the con-
tent of (6.113) is
whish is to say that a basis for the Hermitian
generating operators of a compact group can be
found that implies real orthogonal, or unitary,
matrices U(T) , ?nd antisyiranetrical matrices
GROUPS OF TRANSFORMATIONS 235
assigned the value zero. In a similar way
236 QUANTUM KINEMATICS AND DYNAMICS
If the operator X in (6.112) is chosen as
an algebraic function of the generating operators
G , the integration process will produce those
algebraic functions that commute with every G and
therefore serve to classify, by their eigenvalues
the various matrix representations of the infini-
tesimal generators of the group. Now
in which it is supposed that the basis is suitable
to produce a unitary matrix U . The effect of
the integration is achieved by requiring that
and, since f(G) can be chosen as a symmetrical
homogeneous function of the various G , the opera-cl
tor nature of the latter is not relevant which per-
mits (6.116) to be replaced by the numerical invar-
iance requirement
referring to functions of a vector y in a v-dimen-
sional space. The infinitesimal version of this
GROUPS OF TRANSFORMATIONS 237
which are analogous to, but are less general than
the differential operators Q. and Q
of invariant functions can be constructed directly
if a finite dimensional matrix representation is
known. Let G and U symbolize a K-dimensional
matrix representation of the corresponding Hermitian
and unitary operators, Then
and
invariance property can be expressed as
where the differential operators
are realizations of the generating operators (for
a semi-simple group) f
supply invariant symmetrical functions of y, and
thereby of the operators G
6.12 DIFFERENTIAL OPERATORS AND THE ROTATION GROUP
The differential and integral group properties
we have been discussing can be illustrated with the
three-dimensional rotation group. On referring to
(6.70-1) we see that the two sets of differential
operators that realize the abstract angular momen-
tum operators are
is an invariant function. The coefficients of the
powers of X , or equivalently, the traces
238 QUANTUM KINEMATICS AND DYNAMICS
these differential operators appear as
together with the results of cyclically permuting
the indices 123 , The differential operators
§. , are such that
and are evidently associated with infinitesimal
rotations in a four-dimensional Euclidean space.
(They obey the four-dimensional extension of the
angular momentum commutation relations given in
(2.80). ) Of course, the group manifold is three-di-
mensional and, on restoring n as a function of
the independent parameters u , the term contain-
ing V3uQ is omitted in (6.124). The three-di-
mensional element of volume contains
(du) = du, du2 du., , together with the factor
or, if we introduce the notation
239GROUPS OF TRANSFORMATIONS
240 QUANTUM KINEMATICS AND DYNAMICS
|det 5 I which can be evaluated directly or inferred
from the relevant form of the differential equations
(6.102)
The unique solution of this equation states the
constancy of un det £ . Alternatively, an
employ the four variables u = u~ , u , and a
volume element proportional to (du) = duQ (du)
with the restriction on the variables u_ enforced*7 O
by a delta function factor 6 (u 4- u -l) . On
integrating over u» , one regains the factor
|un which, in the previous method, is supplie
by |det C| • Thus, the group manifold is the
three-dimensional surface of a four-dimensional
Euclidean unit sphere, and the volume element can
be described intrinsically, or in terms of the space
in which the manifold is imbedded,
Here the proper constants have been supplied to nor-
malize the total volume to unity, although in the
three-dimensional form one must also sum over the
GROUPS OF TRANSFORMATIONS 241
two pieces of the group that correspond to
u-. SB ± (l-u ) . The advantage of the four-dimen-2
sional form stems from the commutativity of 6(u -I)
with the differential operators J , which permits
the latter to be defined in the unbounded four-di-
mensional Euclidean space, thereby identifying the
four variables u with canonical variables of thea
first kind. Thus the general properties of a three-
dimensional angular momentum can be studied in terms
of an equivalent system consisting of a particle
in a four-dimensional Euclidean space, with the
correspondence between its orbital angular momentum
and the general three-dimensional angular momentum
described by (6.126).
There is only one independent operator that2
commutes with every component of J , namely, J ,
for there is only one independent rotationally in-
variant function of a three-dimensional vector. We
observe from (6.124-6) that
242 QUANTUM KINEMATICS AND DYNAMICS
where the square of the three-dimensional angular
momentum vector is represented by the four-dimen-
sional differential operator
Thus eigenfunctions of the total angular momentum
are obtained from four-dimensional spherical har-
monics ~ solutions of the four-dimensional Laplace
equation that are homogeneous of integral degree
n = 0 , 1, 2,..., - and the eigenvalues are
2Another procedure uses complex combinations
of the parameters, as comprised in the pair of com-plex numbers that form any row or column of the ma-trix
imply
Let y indicate either two compo row vector,with y1 similarly designating a column vector.The differential composition properties.
GROUPS OF TRANSFORMATIONS 243
The completeness of these eigenfunctions, and of
the three-dimensional angular momentum spectrum can
be inferred from the structure of the fundamental
solution, of degree -2 , which refers to the
inhomogeneous equation
6.13 NON-COMPACT GROUP INTEGRATION
The technique of group integration can also
be effective for groups that are not compact. We
shall illustrate this with the special canonical
group referring to Herinitian variables of the first
kind. As the analogue of the operator appearing in
(6.112), we consider
The squared angular momentum is thereby representedas, for example,
The corresponding differential operator realizationsof an angular momentum vector are:
(The transformations described by the parameter X
are without effect here), The integrations are ex-
tended over the infinite spectral range of Heraitian
canonical variables. As we can recognize directly
from the group multiplication law (6,58), the opera-/%
tor X commutes with every unitary operator U(q'p')
Hence it commutes with both sets of canonical vari-
ables , q and p , and for a system described byA
variables of the first kind, X must be a multiple
244
where, in the notation of (6.57)
QUANTUM KINEMATICS AND DYNAMICS
~ •
and any function of the y that is homogeneousof integral degree n provides an eigenfunction,which is associated with the eigenvaluej(j+1) , j = n/2 . One recognizes that the differ-ential operators refer to an equivalent system des-cribed by two complementary pairs of non-Hermitianvariables of the first kind, and
This equivalence was used for a systematic develop-ment of the theory of angular momentum in a paper
GROUPS OP TRANSFORMATIONS 245
of the unit operator. This result can also be de-
rived through an explicit construction of the ma-^
trix representing X in some canonical representa-
tion. We first observe that
where the right side is a multiple of the unit opera-
tor. If X is chosen as a Measurement symbol,
M(a') , this equation reads
that was written in 1951, but remained unpublished.It is now available in the collection "QuantumTheory of Angular Momentum" edited by L.C. Biedenharnand H. Van Dam, Academic Press, 1965.
We have thus shown that
and therefore
The statements contained in (6.140) and (6,143) can
be regarded as assertions of the completeness and
orthonormality of the operator basis provided by
246 QUANTUM KINEMATICS AND DYNAMICS
and multiplication on the right by XM(a') ,
followed by summation over a1 , converts our re-
sult into
which is the explicit exhibition of any operator as
a function of the fundamental dynamical variables
of the first kind. If the operator X of (6.138)
is replaced by such a function, F(q , p) , that
formula becomes
or, according to the multiplication property
As an application of the latter form, we have
GROUPS OF TRANSFORMATIONS 247
the continuous set of unitary operator functions of
the variables of the first kindf U(q'p')
6.14 VARIABLES OF THE SECOND KIND
The general formal analogy between the two
types of dynamical variables, together with the
specific difference associated with the evaluation
of traces, correctly suggests that the statements
of Eqs. (6,138), (6.140-3) are applicable to vari-
ables of the second kind if the operation tr ...
is replaced by tr p.., . But a word of caution is
needed. The eigenvalues that appear in U(q'p')
are anticoiranutative with the variables of the sec-
ond kind and thus contain the factor p , whereas
the eigenvalues involved in the integrations and
in the delta functions are entirely elements of the
external algebra. Incidentally, the general trace
formula (4.179) emerges from (6.138) and its analogue
for variables of the second kind on forming the
<p'=0| q'=0> matrix element. It is also worth
noting that our various results could be freed from
explicit reference to complementary variables. We
shall make limited use of this possibility to trans-
248 QUANTUM KINEMATICS AND DYNAMICS
form the statement for variables of the second kind
from their non-Hermitian canonical versions to forms
appropriate for Hermitian canonical variables. This
will be accomplished by using the transformation
(3.116) for operators, together with the analogous
one for eigenvalues, which yields, for example,
or
Similarly
and
When using this notation one roust not confuse the
GROUPS OF TRANSFORMATIONS 249
I i n
symbol 5 (6 -O) with an eigenvalue of theEv Kl
Hermitian operator £
and
which is such that
6.15 REFLECTION OPERATOR
If the integration process (6,109) is applied
•bo the operators U(q'p'X) , referring to variables
of the first kind, we obtain zero as a result of
the X integration. Hence there is no state in-
variant under all the operations of the special
canonical group. But, if the X integration is
omitted, we are led to consider the Hermitian opera-
tor
250 QUANTUM KINEMATICS AND DYNAMICS
Thus, R anticoiramutes with each variable of the first
kind.
The now familiar operator possessing these proper-
ties for the variables of the second kind is simi-
larly produced by
6.16 FINITE OPERATOR BASIS
The formal expressions of completeness and
orthogonality for the operator basis of the variables
of the second kind, which are comprised in the
various aspects of the U(lp ,
and correspondigly
can be freed of explicit reference to the special
canonical group. We first recognize that U(C')
is the generating function for the 2 n distinct
elements of the operator algebra. These we define
more precisely as
If we use the notation «'{v} to designate the
similar products formed from the eigenvalues £" iK,
we have
GROUPS OF TRANSFORMATIONS 251
where each v assumes the value 0 or 1 , andis
With this definition the Hermitian operators a{v)
possess unit squares and, in particular,
252 QUANTUM KINEMATICS AND DYNAMICS
The product of two operators, with indices {v}
and (1 - v} equals the operator p , to within
a phase factor,
The integration symbol stands for differentia-
tion with respect to each C * * and thus the terms
that contribute to the integrals are of the form
But one must also recall that the anticommutative
operator p is to be separated from the eigenvalues
prior to integration. This has no explicit effect
GROUPS OF TRANSFORMATIONS 253
if the number of eigenvalues in the individual terms
is even, corresponding to v = £ v an even integer,H»
For odd v , however, the additional factor of
p multiplying a{v} induces
The net result has the same form in either circum-
stance, leading to the following expression of
completeness for the 4 dimensional operator basis
a{v}
The orthonormality property
can be inferred from the completeness and linear
independence of the a{v} or obtained directly
from (6.154) on remarking that
254 QUANTUM KINEMATICS AND DYNAMICS
t6.17 ADDENDUM: DERIVATION OP THE ACTION PRINCIPLE
tReproduced from the Proceedings of the National
Academy of Sciences, Vol. 46, pp. 893 - 897 (1960).
VOL. 46, I960 PHYSICS: J. SCHWINGEM 893
We are now going to examine the construction of finite unitary transformationsfrom infinitesimal ones for a physical system of n continuous degrees of freedom.Thus, all operators are functions of the n pairs of complementary variables qk, pt,which we deaote collectively by x. Let us consider a continuous set of unitaryoperators labeled by a single parameter, U(r). The change from r to r + rfr is theinfinitesimal transformation
which includes a possible explicit T dependence of the generator, and
where
are the fundamental quantum variables of the system for the description producedby the transformation V(r). The accompanying state transformations are indi-cated by
A useful representation of the unitary transformation is given by the transformationfunction
where the omission of the labels a', 6' emphasizes the absence of explicit referenceto these states. Yet some variation of the states must be introduced if a sufficientlycomplete characterization of the transformation function is to be obtained. Forthis purpose we use the infinitesimal transformations of the special canonical group,performed independently on the states associated with parameters r and T + <J-r[S'l.nnu,,^.
894 PHYSICS: J. SCHWINGEB PBOC. N. A. S.
which includes
the matrix of V(r) in the arbitrary ab representation. The relation between in-finitesimally neighboring values of r is indicated by
The general discussion of transformation functions indicates that the most com-pact characterization is a differential one. Accordingly, we replace this explicitstatement of the transformation function (a'-r + dr\b'r} by a differential descrip-tion in which the guiding principle will be the maintenance of generality by avoidingconsiderations that refer to specific choices of the states o' and &', Wenote first that the transformation function depends upon the parameters T, T +AT and upon the form of the generator 0(x, T), Infinitesimal changes in theseaspects [8*] induce the alteration
in which the infinitesimal generators are constructed from the operators appropriateto the description employed for the corresponding vectors, namely x(r + dr) andX(T). It is convenient to use the symmetrical generator <?,, „ which produceschanges of the variables x by '/jte. Then
which, with the similar expression for 0,{r + dr), gives
wfoer**
and the Sx(r), &x(r -f dr) are independent arbitrary infinitesimal numbers uponwhich we impose the requirement of continuity in r.
The infinitesimal unitary transformation that relates X(T) and x(r + dr) is ob-tained from
Our result is a specialization or tne general diflerential characterization 01 trans-formation functions whereby, for a class of alterations, the infinitesimal operatorSW is derived as the variation of a single operator W. This is a quantum actionprinciple8 and W is the action operator associated with the transformation.
We can now proceed directly to the action principle that describes a finite unl-tarv transformation.
for multiplicative composition of the individual infinitesimal transformation func-tions is expressed by addition of the corresponding action operators
As written, this action operator depends upon all operators X(T) in the r intervalbetween r\ and r». But the transformations of the special canonical group, appliedto (n | ra), give
which is to say that d'Wi» does not contain operators referring to values of r in theopen interval between r\ and n, or that W-a, is stationary with respect to the specialvariations of x(r) in that interval. Indeed, this principle of stationary action,the condition that a finite unitary transformation emerge from the infinitesimalones, asserts of q(r), p(r) that
where
m which & is used here to describe the change of q, p by Sq and Sp, occurring inde-pendently but continuously at r and r + dr. The two species of variation cannow be united: J = &' + &*, and
You 46, 1960
as
Accordingly, one can write
PHYSICS: J, SCHWINGBR, 895
or
89« PHYSICS: J. SCMWINOBK PBOC. N, A. S.
which are immediate implications of the various infinitesimal generators.The use of a single parameter in this discussion is not restrictive. We have only
to write
where
The vanishing of each of these operators is demanded if the transformation ia to beindependent of path. When the operators Gt(x, r) can be expressed as a linear com-bination of an equal number of operators that are not explicit functions of the param-eters, Oa(x)f the requirement of path independence yields the previously con-sidered conditions for the formation of a group.
We now have the foundations for a general theory of quantum dynamics andcanonical transformations, at least for systems with continuous degrees of freedom.The question is thus posed whether other types of quantum variables can also beemployed in a quantum action principle.
* Publication assisted by the Office of Scientific Besearch, United States Air Force, ynder con-tract number AF49(638)-589.
« These PBOCBBMNGS, 45,1S42 (1959); 48, 257 (1960); and 46, 570 (1960),* The group can be obtained directly as the limit of the finite order group associated with each
r. That group, of order »•*, is generated by U, ¥, and the »th root of unity given by WV~lU~l.Some aspects of the latter group are worthy of note. There are »' + r — 1 classes, the commu-tator subgroup is of order e, and the order of the corresponding quotient group, F', is the numberof inequivatent one-dimensional representations. The remaining » — 1 matrix representationsmust be of dimensionality », «•' ~ »* + (» ~ I)"1) and differ only in the choice of the generating»th root of unity. That choice is already made in the statement of operator properties for U andF and there can be only one irreducible matrix representation of these operators, to within thefreedom of unitary transformation.
* In earlier work of the author, for example Phyt, Km., 91, 713 (1953), the quantum actionprinciple has been postulated rather than derived.
with each drK/dr given as an arbitrary function of T, and then regard the transforma-tion as one with p parameters, conducted along a particular path in the parameterspace that is specified by the p functions of a path parameter, T>(T). Now
is the action operator lor a traasiormation referring to a prescribed path and gener-ally depends upon that path. If we consider an infinitesimal path variation withfixed end points we find that
GROUPS OF TRANSFORMATIONS 259
6.18 ADDENDUM CONCERNING THE SPECIAL CANONICAL
GROUP1"
tReproduced from the Proceedings of the National
Academy of Sciences, Vol. 46, pp. 1401-1415 (I960)
Reprinted from the Proceedings of the NATIONAL ACABEMTT op SCIENCESVol. 48, No. JO, pp. 1401-1416. October, IBM.
THE SPECIAL CANONICAL GROUP*
BY JULIAN SCHWINGER
HABVABD uNivxnuurr
Communicated August SO, 1960
This note is concerned with the further development and application of anoperator group described ia a previous paper,' It is associated with the quantumdegree of freedom labeled v = » which is characterized by the complementarypair of operators q, p with continuous spectra.8 The properties of such a degreeof freedom are obtained as the limit of one with a finite number of states, spe-cifically given by a prime integer v. We recall that unitary operators U and
define two orthonormal coordinate systems (w*| and (0*1, where
and
For any prime v > 2 we can choose the integers k and I to range from — l/t(v — 1)to l/t(v — I ) , rather than from 0 to v — J. An arbitrary state * can be representedalternatively by the wave functions
where
and the two wave functions are reciprocally related by
We now shift our attention to the Hermitian operators g, p defined by
and the spectra
Furthermore, we redefine the wave functions so that
where e = Ag' = Ap', the interval between adjacent eigenvalues. Then we have
V obeying
1402
and
PHYSICS: J, SCHWINGSB Pmoc. N. A. S.
and
This is what is implied by the symbolic notation (Dirac)
We shall also use the notation that is modeled on the discrete situation, withintegrals reolacine summations, as in
with
which must be an identity for wave functions of the physical class when theoperations are performed as indicated. There will also be a class of functionsK(p', «), such that
We shall not attempt here to delimit more precisely the physical class of states.Note however that the reciprocal relation between wave functions can be combinedinto
and
As v increases without limit the spectra of q and p become arbitrarily dense andthe eigenvalues of largest magnitude increase indefinitely. Accordingly we mustrestrict all further considerations to that physical class of states, or physical sub-space of vectors, for which the wave functions $(q') and ${p') are sufficiently wellbehaved with regard to continuity and the approach of the variable to infinitythat a uniform transition to the limit v = » can be performed, with the result
In the limit i» = oo there is a class of functions f(q'p') such, that
Vou 46, I960 PHYSICS: J. SCHWINGER 1403
There are other applications of the limit v —* <*>. The reciprocal property of theoperators U and F is expressed by
or
with an exception when q' or p' is the greatest eigenvalue, for then q' + t or p' + eis identified with the least eigenvalue, —q' or —p'. We write these relations aswave function statements, of the form
and
In the transition to the limit v = «, the subspaee of physical vectors * is dis-tinguished by such properties of continuity and behavior at infinity of the wavefunctions that the left-hand limits. « -+• 0, exist as derivatives of the correspondingwave function. We conclude, for the physical class of states, that
and
It will also be evident from this application of the limiting process to the unitaryoperators exp(feg), exp(icp) that a restriction to a physical subspaee is needed forthe validity of the commutation relation
Fhe elements of an orthonormal operator basis are given bj
or v-^Uia'-D'), with
Thus, since »-' = Ag'Ap'/2-r, we have, for an.arbitrary function f(q'p') of thediscrete variables q', p',
1404 PHYSICS: J. SCHWINGER PBOC. N. A. S.
which we express symbolically by
In particular,
where g and p on the right-hand side are numerical integration variables.The completeness of the operator basis U(g'p') is expressed by
The properties of the U(g'p') basis are also described, with respect to an arbitradiscrete operator basis X(d), by
and
When X is given as F(q, p), the operations of the special canonical group can beutilized to bring the completeness expression into the form
This operator relation implies a numerical one if it is possible to order F(q, p),so that all g operators stand to the left, for example, of the operator p: F(§; p).Then the evaluation of the (g' = 0 p' = 0} matrix element gives
which result also applies to a system with « continuous degrees of freedom if it isunderstood that
As an example of this ordering process other than the one already given by{r U(g'p'), we remark that
where
VOL. 46, I960 PHYSICS; J. SCSWINGER 1405
and thwefori
which reproduces the well-known non-degenerate spectrum of the operatorw + p5)-
Now we shall consider the construction of finite special canonical transformationsfrom a succession of infinitesimal ones, as represented by the variation of a pa-rameter T. Let the generator of the transformation associated with r -*• r + dr be
where Q(T) and P(r) are arbitrary numerical functions of r. That is, the in-finitesimal transformation is
which implies the finite transformation
Some associated transformation functions are easily constructed. We have
or
and therefore
which, in conjunction with the initial condition
gives
with
Accordingly, if we also multiply by e x p ( — i p ' f f ) and Integrate with respect todp'/Zr, what emerges is {q'n \ q"Tt)gr, as we can verify directly.
and then indicate the effect of the additional localized transformations by theequivalent unitary operators, which gives
It is important to recognize that the trace, which is much more symmetricalthan any individual transformation function, also implies specific transformationfunctions. Thus, let us make the substitutions
where, in view of the delta function factors, f+(r — T') can be replaced by otherequivalent functions, such as q+ — !/j = */»«, or */je(r ~ T'^ — (T — r')/Tt withf = T, — Tj. The latter choice has the property of giving a zero value to thedouble integral whenever Q(r) or P(r) is a constant. As an operator statement,the trace formula is the known result
We can compute the trace of a transformation function, regarded as a matrix, andthis will equal the trace of the associated unitary operator provided the otherwisearbitrary representation is not an explicit function of r. Thus
These transformation functions can also be viewed as matrix elements of theunitary operator, an element of the special canonical group, that produces thecomplete transformation. That operator, incidentally, is
where «(r — r') is the odd step function
1406 PHYSICS: J. SCHWINOEB PBOO, N. A, S.
From this result we derive
or, alternatively
on using tee tact tnat
You 46, I960 PHYSICS; J. SCHWINGSR 1407
We shall find it useful to give an altogether different derivation of the traceformula. First note that
and similarly
which is a property of periodicity over the interval T — r\ — rt. Let us, therefore,represent the operators g(r), J>(T) by the Fourier series
where the coefficients are so chosen that the action operator for an arbitrary specialcanonical transformation
acquires the form
Here the dash indicates that the term n = 0 is omitted, and
The action principle for the trace is
and the principle of stationary action asserts that
together with
The first of these results implies that the trace contains the factors &(Qo) andJ(Po). The dependence upon Qn, Pn, n 4= 0, is then given by the action principle as
First, let us observe how an associated transformation function {n| r^)0^p depends
upon the arbitrary functions Q(r), P(r). The action principle asserts that
1408 PHYSICS: J, 8CHWINOER PBOC. N. A. S.
Therefore
where the factor of 2-jr is supplied by reference to the elementary situation withconstant Q(r) and P(r). We note, for comparison with previous results, that
and
The new formula for the trace can be given a uniform integral expression by usingthe representation
for now we can write
where
and W[q, pi is the numerical function formed in the same way as the action operator,
Alternatively, we can use the Fourier series to define the numerical functionsq(r), pM. Then
f T*
and d[q, p] appears as a measure in the quantum phase space of the functions«W, P(T).
It is the great advantage of the special canonical group that these considerationscan be fully utilized in discussing arbitrary additional unitary transformations,as described by the action operator
Here ( )0 is an ordered product that corresponds to the sense of progression fromTJ to n. If T follows T' in sequence, the operator function of r stands to the left,while if r precedes T' the associated operator appears on the right. This descriptioncovers the two algebraic situations: n > r», where we call the ordering positive,( )+, and n < TI, which produces negative ordering, ( )_. For the momenttake TI > rj and compare
with
PHYSICS: J. SCHWINGER 1409
which we express by the notation
More generally, if F(T') is an operator function of x(r') but not of Q, P we have
where [ r') X (T' symbolizes the summation over a complete set of states, andtherefore
The difference of these expressions,
refers on one side to the noncommutativity of the complementary variables q andp, and on the other to the "equation of motion" of the operator g(r). Accordingto the action principle
and therefore
which yields the expected result,
Vol. 46,1960
1410 PHYSICS: J. SCHWINGER PBOO, N. A. S.
Thus, through the application of the special 'canonical group, we obtain functionaldifferential operator representations for all the dynamical variables. The generalstatement is
where F(q, p)a is an ordered function of the q(r), p(r) throughout the interval be-tween TS and n, and, as the simple example of q(T)p(r) and p(r)q(r) indicates,the particular order of multiplication for operators with a common value of tmust be reproduced by a suitable limiting process from different r values.
The connection with the previous considerations emerges on supplying G witha variable factor X. For states at n and n that do not depend explicitly upon Xwe use the Action principle to evaluate
The formal expression that gives the result of integrating this differential equationfrom X = 0 to X = 1 is
where the latter transformation function is that for X = 0 and therefore refersonly to the special canonical group. An intermediate formula, corresponding to0 = Gi + Gt, contains the functional differential operator constructed from G\acting on the transformation function associated with (?j. The same structureapplies to the traces of the transformation functions. If we use the integralrepresentation for the trace of the special canonical transformation function, andperform the differentiations under the integration sign, we obtain the generalintegral formula'
Here the action functional W[q, p] is
which is formed in essentially the same way as the action operator PFM, the multi-plication order of noncommutative operator factors in G being replaced by suitableinfinitesimal displacements of the parameter r. The Hermitian operator G canalways be constructed from symmetrized products of Hermitian functions of qand of p, and the corresponding numerical function is real Thus the operatorVtf/ifato, MvM] is represented by 'AC/jCfCr + «)) + /,(<?(r - t)))MpM),for example. One will expect to find that this averaged limit, * -*• 0, implies nomore than the direct use of /I(#(T)/J(P(T)), although the same statement is certainlynot true of either term containing e. Incidentally, it is quite sufficient to construct
Another property of the example is that the contributions to the trace of allFourier coefficients except n = 0 tend to unity for sufficiently small T at ft. Thisis also true for a class of operators of the form G = !/»pz + /(<?)• By a suitabletranslation of the Fourier coefficients for p(X) we can write w[q, p] as
This example also illustrates the class of Hermitian operators with spectra thatare bounded below and for which the trace of exp (iTG) continues to exist on givingT a positive imaginary component, including the substitution T -*• iff, ft > 0.The trace formula can be reatated for the latter situation on remarking that theFourier series depend only upon the variable (r — T$)/T = A, which varies from0 to 1, and therefore
VOL. 46, 1960 PHYSICS: J. SCHWINGER 1411
the action from pdq, for example, rather than the more symmetrical version, invirtue of the periodicity,
As a specialization of this trace formula, we place Q — P = 0 and consider theclass of operators G that do not depend explicitly upon T, Now we are computing
in which we have used the possibility of setting ra = 0. A simple example isprovided by G = '/»(?* + <?*)» where
&nfL
For sufficiently small B, the trem involing dq/dy to which all the Fourier coffi-
cients of q(y) contribute except q0, will effectively suppress these Fouier coefficientPovides appropiate restric are imposed concering sigular pointes in the neighborhood of which f(q) acquires arge negative values. Then f(q (y)f (q0))and we can reduce the integrations to just the contribution of qo and po, as experssed
by
1412 PHYSICS: J. SCBWINGSR PEOC. N. A. S.
Comparison with the previously obtained trace formula involving the ordering ofoperators shows that the noneommutativity of g and p is not significant in this limit,Thus we have entered the classical domain, where the incompatibility of physicalproperties at the microscopic level is ao longer detectible. Incidentally, a, firstcorrection to the classical trace evaluation, stated explicitly for one degree offreedom, is
Now introduce the functional differential operator that is derived from thenumerical action function W0[q, p] which refers only to the transformationsrcnerated bv G. namelv
These equations are valid for any such transformation function. The trace isspecifically distinguished by the property of periodicity,
which gives the exact value when /(g) is a positive multiple of q2.Another treatment of the general problem can be given on remarking that the
equations of motion implied by the stationary action principle,
can be represented by functional differential equations*
which asserts that the trace depends upon Q(r), P(r) only through the Fouriercoefficients Qn, Pn, and that the functional derivatives can be interpreted by meansof ordinary derivatives:
and observe that the differential equations are given by
VOL. 46, 1960 PHYSICS; J. SCHWINGER 1413
>r by
The latter form follows from the general expansion
on noting that [Wo, Q(r)], for example, is constructed entirely from differentialoperators and is commutative with the differential operator WG- Accordingly,
which asserts that exp(—iW0)(tr) vanishes when multiplied by any of the Fouriercoefficients Qn, Pn and therefore contains a delta function factor for each of thesevariables. We conclude that
where, anticipating the proper normalization constants,
A verification of these factors can be given by placing G = 0, which returns usto the consideration of the special canonical transformations. In this procedurewe encounter the typical term
the proof of which follows from the remarks that
and
The result is just the known form of the transformation function trace
When integral representations are inserted for each of the delta function factorsin 8[Q, P], we obtain
and the consequence of performing the differentiations in Wa under the integrationsigns is*
where the action function W[q, p] now includes the special canonical transformationdescribed by Q and P.
1414 PHYSICS: J. SCHWINOER PBOC. N. A, 8.
We shall also write this general integral formula as
in order to emphasize the reciprocity between the trace, as a function of Qn, Pn orfunctional of Q(r), P(r) and exp(«fPG[g, p]) as a function of qn, pn or functional of?(T), p(r). Indeed,
where
is such that
A verification of the reciprocal formula follows from the latter property on insertingfor the trace the formal differential operator construction involving 5[Q, F]. Thereality of Wg[q, p] now implies that
f d[Q, P]d[Q', F']eVTW(p-F')-«0-e')) (tr)0e* (&•)«'/•' = \
or, equivalently,
f d[Q, P](fr)x+*,*(fr)*+*i - «[Qi - Qt, Pi - Pt]
where X combines Q and P.
The trace possesses the composition property
in view of the completeness of the operator basis formed from f(-ri), p(n),No special relation has been assumed among TO, n and T». If n and TS are
equated, one transformation function in the product is the complex conjugate ofthe other. We must do more than this, however, to get a useful result. The mostgeneral procedure would be to choose the special canonical transformation inft" {raj Ti)e«p = tr (n\ Tt)g<>r* arbitrarily different from that in tr (n TJ)OW. Thecalculations! advantages that appear in this way will be explored elsewhere.
The operation involved is the replacement of Q(r), P(r) in the respective factorsby Q(r) =F q'S(r — n), F(r) =F p'S(r — n) followed by integration with respectto dq'dj>'/2ir. The 'explicit form of the left side is therefore
VOL. 46, 1960 PHYSICS: /. SCHWINOER 1415
Here we shall be content to make the special canonical transformations differonly at r». The corresponding theorem is
where
and similarly for Q"{r), P*(r). This statement follows immediately from theorthonormality of the U(q'p') operator basis on evaluating the left-hand side as
* Supported by the Air Force Office of Scientific Research (ARDC).1 These PBOOBBOIMGS, 46, 883 (1960).' These PEOCBBDINOS, 46, 570 (1960),' This formulation is closely related to the algorithms of Feynman, Pkys. Bee., 84, 108 (1951),
Res. Mod, Phys., 20, 36 (1948), It differs from the latter in the absence of ambiguity associatedwith noncommutative factors, but primarily in the measure that is used. See Footnote 5.
4 These are directly useful as differential equations only when 0(fp) is a sufficiently simplealgebraic function of q and p. The kinematical, group foundation for the representation ofequations of motion by functional differential equations is to be contrasted with the dynamicallanguage used in these PmocmEDisas, 37, 4S2 (1851).
* In this procedure, q(r) and p(T) are continuous functions of the parameter r and the Fouriercoefficients that represent them are a denumerably infinite set of integration variables. Analternative approach is the replacement of the continuous parameter r by a discrete index whileinterpreting the derivative with respect to r as a finite difference and constructing S[Q, PI as sproduct of delta functions for each discrete r value. With the latter, essentially the Peynman-Wiener formulation, the measure d[q, p] is the product of dg(r)<ip(r)/2*r for each value of T,periodicity is explicitly imposed at the boundaries, and the limit is eventually taken of an infinitelyfine partitioning of the interval T = n — TJ. The second method is doubtless more intuitive,since it is also the result of directly compounding successive infinitesimal transformations but it igmore awkward as a mathematical technique.
GROUPS OF TRANSFORMATIONS 275
6.19 ADDENDUM: QUANTUM VARIABLES AND THE ACTIONPRINCIPLE
tReproduced from the Proceedings of the National
Academy of Sciences, Vol. 47, pp. 1075 -1083 (1961)
Hpprhited from the Proeeetlmgs of the NATSOKAZ, ACAOJKMY or SCIENCESVol. ff, No. 7, pp. 1078-1083, July, W6I.
QUANTUM VARIABLES AND THE ACTION PRINCIPLE
BY JULIAN SCHWIJJGER
HiBVARD UNIVERSITY AND I'XIVEHSITY OF CALIFORNIA AT I,OH ANGBIJU5
Communicated Alaij 89, lUfii
In previous communications, a classification of quantum degrees of freedom by aprime integer v has been given,1 and a quantum action principle has been con-structed2 for v=<*>. Can a quantum action principle be devised for other types ofquantum variables? We shall examine this question for the simplest quantum de-gree of freedom, »• = 2.
Let us consider first a single degree of freedom of this type. The operator basisis generated by the complementary pair of Hermitian operators
is the three-dimensional infinitesimal rotation
Accordingly, an infinitesimal transformation that varies £» and not & can only be arotation about the second axis,
The corresponding infinitesimal generator is
Jimilarly,
is generated by
and the generator for the combination of these elementary transformations is (k =1,2)
which obey
The basis is completed by the unit operator and the product
We have remarked1 upon the well-known connection between the 0%, k = 1,2, 3,and three-dimensional rotations. In particular, the most general unitary operatorthat differs inftnitesimally from unity is, apart from a phase factor, of the form
and the cosTesponding operator transformation
1075
1078 PHYSICS: J. SCUWINOER PBOC. N. A. S.
It is evident that G» and ft must be supplemented by
to form the infinitesimal generators of a unitary group, which is isomorphie to thethree-dimensional rotation group. The transformation induced by 63 is
and
Thus, the concentration on the complementary pair of operators fi and fe does notgive a symmetrical expression to the underlying three-dimensional rotation group.This is rectified somewhat by using, for those special transformations in which |jand fe are changed independently, the generator of the variations VtSft, k =» 1,2,
n>
An arbitrary infinitesimal unitary transformation is described by the transforma-tion function
Infinitesimal variations in r, r + dr, and the structure of G induce
To this we add S', the transformations generated by Gt, performed independentlybut continuously in r, on the states (r + dr\ and I r).
where S', in its effect upon operators, refers to the special variations 8ft, k = 1, 2,performed independently but continuously at r and r + dr.
It is only if the last term is zero that one obtains the quantum action principleis = r + «1
with
Since the special variation is such that 8|i and Jfe are arbitrary multiples of &, it
Here
or
VOL. 47, 1961 PHYSICS: J. SCMWINGKS 1077
is necessary that [6, |3] commute with & and fe. Hence this commutator must bea multiple of the unit operator, which multiple can only be zero, since the trace ofthe commutator vanishes or, alternately, as required by [G, Is2] = 0. For an actionprinciple formulation to be feasible, it is thus necessary and sufficient that
Terms in & and S-t are thereby excluded from G, which restriction is also conveyedby the statement that a permissible <? must be an even function of the {», k = 1, 2.Apart from multiples of the unit operator, generating phase transformations, theonly allowed generator is fife, which geometrically is a rotation about the thirdaxis.
It should be noted that the class of variations &' can be extended to include theone generated by (?», without reference to the structure of G. Thus,
where the latter term equals
and therefore, for this kind of $' variation,
The action operator for a finite unitary transformation is
or, more symmetrically,
since Wn is only defined to within an additive constant, and
The principle of stationary action
which refers to a fixed form of the operator G({(r), r), expresses the requirementthat a finite transformation emerge from the succession of infinitesimal transforma-tions. It will be instructive to see how the properties of the quantum variablesare conversely implied by this principle. The discussion will be given withoutexplicit reference to the single pair of variables associated with one degree of free-dom, since it is of greater generality.
The bilinear concomitant
The generator term G&T evidently restates the transformation significance of theoperator G, The effect upon operators is conveyed by the infinitesimal unitarytransformation
which is the general equation of motion,
Let us take Sf* to be a special variation, which we characterize by the followingproperties: (I) Each 8& anticommutes with every &,
(2) the &^(T) have no implicit T dependence; and (3) every Sft is an arbitrary in-finitesimal numerical multiple of a common nonsingular operator, which does notvary with r. The second basic property asserts that
which restricts 0 to be an even function of the &, in virtue of the anticommutativityof the special variations with each member of this set. Furthermore, the r deriva-tives of the special variations are also antieommutative with the {», according toproperty (3), and therefore,
Then, if we write
which defines the left and right derivatives of G with respect to the f»,- we get
which is consistent with the general equation of motion., and
1078
shows that
PHYSICS: J,
and gives
PRoc. N. A. S.
Vol.. 47, 1981 PHYSICS: J. SCHWINOBR 1079
The nonsingular operator contained in every special variation can be cancelledfrom the latter equation, and the arbitrary numerical factor in each SfctM impliesthat
In the opposite signs of left and right derivatives, we recognize the even property ofthe function G.
On comparing the two forms of the equation of motion for f* we see that
If we reintroduce the special variations, this reads
The left side is just the change induced in (? by (?{, the generator of the special vari-ations, which also appears in {?», t, while the right side gives the result in terms ofchanges of the {» by Vifi£». Since (? is an arbitrary even function of the f*, we acceptthis as the general interpretation of the transformation generated by (?$, whichthen asserts of any operator F that
The implication for an odd function of the ft is
and the particular choice F = f j gives the basic operator properties of these quantumvariables
In this way, we verify that the quantum action principle gives a consistent accountof all the characteristics of the given type of quantum variable.
The operator basis of a single degree of freedom is used in a different mannerwhen the object of study is the three-dimensional rotation group rather than trans-foimations of the pair of complementary physical properties. The generator ofthe infinitesimal rotation Sir = Su X e is
On applying this transformation independently but continuously to the states ofthe transformation function (r 4- drl T). we encounter
The identity,
1080 PHYSICS: J. SCHWINGIBR Pnoc. N. A, S.
The most general form for G, describing phase transformations and rotations, is
The action principle asserts that
The extension to n degrees of freedom of type v = 2 requires some discussion.At first sight, the procedure would seem to be straightforward. Operators asso-ciated with different degrees of freedom are commutative, and the infinitesimalgenerators of independent transformations are additive, which implies an actionoperator of the previous form with the summation extended over the n pairs ofcomplementary variables. But we should also conclude that G must be an even
which appear in the anticipated form on remarking that the left-hand side of thelatter equation equals
where the operator variations are arbitrary infinitesimal rotations, So- = fo X <r.Hence the equations of motion are
We shall be content to verify that the principle of stationary action reproduces "theequations of motion that also follow directly from the significance of G,
combined with the equivalence of the two left-hand terms, both of which equal
then gives
where 8' describes the independent operator variations S«r(r) = &i(r) X a(r) atr and T + dr. If we add the effect of independent variations of r, T + dr, and thestructure of (7 (8"], we obtain an action principle3 without restrictions on the formof G.
The action operator for a finite transformation is
Vol. 47, 1961 PHYSICS; J. SCHWINGSR 1081.
function of the complementary variables associated with each degree of freedomseparately, and this is an unnecessarily strong restriction.
To loosen the stringency of the condition for the validity of the action principle,we replace the relationship of commutativity between different degrees of freedomby one of anticommutativity. Let {<a>i,»,i be the operators associated with the adegree of freedom. We define
and this set of 2n Hermitian operators obeys
The inverse construction is
and ia
we have an operator that extends by one the set of anticommuting Hermitianoperators with squares equal to Vs. In particular this operator anticommuteswith every {*, k = 1 ... 2n.
An infinitesimal transformation that alters only f* must be such that
which identifies SJ» as an infinitesimal numerical multiple of {»„+.). We shall write
and the generator of all these special variations is
The latter can also be written as
On forming the commutator of two such generators, we get
where the Hermitian operators
for the condition that permits an action principle formulation. Each special varia-tion is proportional to the single operator $s*+i. It is necessary, therefore, that[G, £**+i] commute with every £*, k = 1 . . . 2«, or equivalently, with each comple-mentary pair of operators fi, j(a>, a — I ... n. Such an operator can only be a mul-tiple of the unit operator and that multiple must be zero. Hence the infinitesimalgenerators of transformations that can be described by an action principle mustcommute with f»«+i. Considered as a function of the anticomrnutative operatorset Jt, k = 1 . . . In which generates the 22" dimensional operator basis, an admis-sible operator (?({) must be even. This single condition replaces the set of n condi-tions that appear when commutativity is the relationship between different degreesof fieedom. Of course, if the class of transformations under consideration is suchthat Q is also an even function of some even-dimensional subsets of the {*, one can
1082 PHYSICS: J. SCHWINOSR Pnoc. N. A. S.
are n(2n — 1} in number, and obey (k ?£ I)
The generators G( and their commutators can thus be constructed from the basisprovided by the n(2n + 1) operators {*;, with k and I ranging from 1 to 2n + 1.And, since
these operators are the generators of a unitary transformation group, which has thestructure of the Euclidean rotation group in 2n + 1 dimensions. For n > 1, theoperators {«, k, I = 1 .. 2n + 1 can also be combined with the linearly independentset
to form the (n + 1) (2n + 1) generators of a similar group associated with rotationsin 2n -f- 2 dimensions.
It should also be Doted that
induces the linear transformation
and one can writ
These generators have the structure of the rotation group in 2n dimensions.The discussion of the change induced in a transformation function (r -j- dr\ T) by
the special vanations, applied independently but continuously at r and T + dr,proceeds as in the special example n = 1, and leads to
and the earlier discussion can be transferred intact. It may be useful, however, toemphasize that the special variations anticommute, not only with each &>(r), butalso with dit/dr, since this property is independent of r. Then it follows directlyfrom the implication of the principle of stationary action,
The action principle is still severely restricted in a practical sense, foi the genera-tors of the special variations cannot be included in G since the operators &fc»+i areodd functions of the 2w fundamental variables. It is for the purpose of circum-venting this difficulty, and thereby of converting the action principle into an ef-fective computation device, that we shall extend the number system by adjoiningan exterior or Grassmann algebra,
1 These PROCBSDINGS, 46, 570 (1960),'Ibid., 46, 883(1960).* The possibility of using the components of an angular momentum vector as variables in an
action principle was pointed out to me by I), Volkov during the 1959 Conference on High EnergyPhysics held at Kiev, U.8.S.R.
that the equations of motion an
for this type of 6' variation.The action operator associated with a finite transformation generalizes the form
already encountered for n = 1,
which gives
VOL. 47, 1961 PHYSICS: J. SCHWINGER 1083
consistently adopt eommutativity as the relationship between the various subsets.Incidentally, the eommutativity of G with &»+I(T) asserts that the latter operatordoes not vary with T. Accordingly, the special variations S&(T) have no implicitoperator dependence upon T and one concludes that the special variations are anti-commutative with the 2w fundamental variables {», without reference to the as-sociated r values.
The action principle is also valid for the linear variations induced by (?„, withoutregard to the structure of G. Thus,
and the last term equals
CHAPTERCANONICAL TRANSFORMATION FUNCTIONS
7.1 Ordered Action Operator 2857.2 Infinitesimal Canonical Transformation 287
Functions7.3 Finite Canonical Transformation 293
Functions7.4 Ordered Operators. The Use of 297
Canonical Transformation Functions7.5 An Example 2997.6 Ordered Operators and Pertubation 302
Theory7.7 Use of The Special Canonical Group 3067.8 Variational Derivatives 3097.9 Interaction of Two Sub-Systems 3177.10 Addendum: Exterior Algebra and the 321
Action Principle
7.1 ORDERED ACTION OPERATOR
According to the significance of the action
operator W(q , q , t) that defines a canonical
transformation at time t , infinitesimal altera-
tions of the eigenvalues and of t produce a change
in the canonical transformation function <q't|q't>
286 QUANTUM KINEMATICS AND DYNAMICS
given by
For some transformations, the commutation proper-
ties of the q and q variables can be used to
rearrange the operator SW so that the q's every-
where stand to the left of the q's . This ordered
differential expression will be denoted by
6W(q ; q , t) and
From the manner of construction of the latter opera-
tor, the variables q and q act directly on
their eigenvectors in (7.1) and this equation be-
comes
Hence 8W must be an exact differential, and in-
tegration yields
in which a multiplicative integration constant is
incorporated additively in W . This constant is
CANONICAL TRANSFORMATION FUNCTIONS 287
fixed, in part, by the composition properties of
the transformation function. It is to be emphasized
that the ordered operator W does not equal W
and, indeed, is not Hermitian, should W possess
this property. We have in effect already illustrat-
ed the ordering method by the construction of the
transformation function <q'jp'> . For Hermitian
variables of the first kind, for example,
7.2 INFINITESIMAL CANONICAL TRANSFORMATIONFUNCTIONS
The transformation function for an arbitrary
infinitesimal canonical transformation is easily
constructed in this way, if one uses the action
operator
whereas
for
is brought into the ordered form by performing the
required operation on the infinitesimal quantity
SG with the aid of the known commutation relations
between q = q and p . It is convenient to
supplement G with a numerical factor X so that
we obtain a one-parameter family of transformations
that includes the desired one (A = 1} and the
identity transformation (X=0) . Then
and the ordering operation is to be applied to G
as well as to 5G . From the former we obtain an
equivalent operator which we call G(p ; q) and thus
288 QUANTUM KINEMATICS AND DYNAMICS
CANONICAL TRANSFORMATION FUNCTIONS 289
The integrability of the ordered operator <5W now
demands that the ordered version of 6G be simply
the variation of G(p;g) and thus (A=l)
in which the additive constant is that appropriate
to the identity transformation, W(p;q) , and de-
pends upon the kinds of variables employed. Hence
or
which simply repeats the significance of G as the
generator of the infinitesimal transformation. When
the transformation corresponds to an infinitesimal
change of parameters, (7.12) reads
and in view of the infinitesimal nature of G
290 QUANTUM KINEMATICS AND DYNAMICS
The composition properties of transformation
functions can now be used to introduce other choices
of canonical variables. For example, with Hermitian
variables of the first kind,
according to {4.21} and- (7.12). The integrations
are easily performed if G is a linear function of
the p variables, or a non-singular quadratic
function. In the first situation
and, in particular,
CANONICAL TRANSFORMATION FUNCTIONS 291
and
For the quadratic functions or p , it is conven-
ient to move the origin of the p' variables to
the point defined by
Then
where
On eliminating p with the aid of (7.20) , we find
292 QUANTUM KINEMATICS AND DYNAMICS
2in which M(q") is the matrix inverse to 3 G/3p3p
In the limit as the latter approaches zero the delta
function form (7.19) is obtained. Both results are
also easily derived directly, without reference to
the intermediate p representation. For the example
of an infinitesimal Hamiltonian-Jacobi transformation
in which H meets the requirements of a non-singu-
lar quadratic dependence on p , we have
2where rn is the matrix inverse to 3 H/3p3p and
An equivalent form for the latter is
CANONICAL TRANSFORMATION FUNCTIONS 293
in which pn is defined by
7.3 FINITE CANONICAL TRANSFORMATION FUNCTIONS
The transformation function for a finite
canonical transformation can be constructed by the
repeated composition of that for an infinitesimal
transformation. To evaluate <T,JT2> along a
particular path in the parameter space, we choose
N intermediate points on the path, and compute
in which x signifies the application of a method
of composition appropiate to the canonical variables
employed. In the limit N -»• « the required path
is traced as an infinite sequence of infinitesimal
transformations. With Hermitian variables of the
first kind and composition by integration, for
example, we begin with
294 QUANTUM KINEMATICS AND DYNAMICS
and obtain
in which
and the delta function factor enforces the restric-
tion q - q" . For the Hamacobi
transformation with a Hamiltonian that depends
quadratically on p, this general form reduces to
in which w is given by (7.25) or (7.26). The
generally applicable technique of composition by
differentiation can be applied directly to the
infinitesimal transformation function (7.15) and we
obtain
CANONICAL TRANSFORMATION FUNCTIONS 295
296 QUANTUM KINEMATICS AND DYNAMICS
apart from numerical factors of (2-n) for Hermi-
tian variables of the first kind. The boundary
values are
and the sense of multiplication is that of (7.28).
CANONICAL TRANSFORMATION FUNCTIONS 297
7.4 ORDERED OPERATORS. THE USE OF CANONICALTRANSFORMATION FUNCTIONS
the transformation function is obtained as a matrix
in which the states and canonical variables do not
depend on T . The result of composing successive
transformations then appears as the matrix of the
product of the associated operators and thus
which defines the ordered exponential operator. The
+ subscript refers to the manner of multiplication,
in which the positions of the operators correspond
to the path in parameter space as deformed into a
On writing
298 QUANTUM KINEMATICS AND DYNAMICS
straight line, with T- on the right and T, on
the left. We speak of negative rather than positive
ordering if the sense of multiplication is reversed,
so that
in analogy with the more specific group structure
(6,39). Should the operators G, , commute, thus
generating an Abelian group, the exponential form
(7.38) is valid without reference to the integration
path. Let us suppose that the generators constitute,
or can be extended to be a complete set of commuting
Hermitian operators. Then a G representation exists
and a canonical transformation function describing
When the generators are independent of T , and
the path is a straight line, the ordering is no
longer significant and
CANONICAL TRANSFORMATION FUNCTIONS 299
the transformations of the group, say <p'T1|q*T?> ,
can be exhibited as
which shows how the canonical transformation function
serves to determine the eigenvalues of the operators
G as well as wave functions representing the G
states.
7.5 AN EXAMPLE
An elementary example for n = 1 is provided
by the Hermitian operator
in which the non-Hermitian canonical variables may
be of either type. A direct construction of the
canonical transformation function can be obtained
from (7.33) by repeated application of
300 QUANTUM KINEMATICS AND DYNAMICS
namely
The eigenvalues thus obtained are
and wave functions of these states are given by
The derivation applies generally to both types of
variables. However, for variables of the second
CANONICAL TRANSFORMATION FUNCTIONS 301
*5 «t» | o
kind (q1) = (q ) = 0 , and the summation in
(7.42) terminates after n = 0 , 1 , which are in-
deed the only eigenvalues of q, The result(7.42) is reached more quickly, however, along the
lines of (5,78). For non-Hermitian variables
syranetrization or antisymmetrization (,) is un-
necessary and it is omitted in (7.40). No ordering
is required to obtain
which immediately gives (7.42). An alternative
procedure is
according to the solution of the equations of motion
which gives (7.42) on integration when combined with
the initial condition provided by
302 QUANTUM KINEMATICS AND DYNAMICS
7.6 ORDERED OPERATORS AND PERTURBATION THEORY
A more general ordered operator form appears
on decomposing the generators into two additive parts
in which the subscripts indicate that the infinites-
imal parameter change is governed by the generators
G or G , respectively. The resulting finite
transformation function is
Then (7,15) can be written
CANONICAL TRANSFORMATION FUNCTIONS 303
Another derivation of the latter result pro-
.ceeds directly from the fundamental dynamical prin-
ciple, now applied to a change in the dynamical
characteristics of the system. Thus, for a system
with the Hamiltonian
an infintesimal change of the parameter X produces
a change in Hamiltonian, and hence of a canonical
transformation function <tii't2> ' *ascri^e<^ y
in which the dynamical variables and states that
appear on the right vary along the integration path
in accordance with the generatorexample of the Hamilton-Jacobi transformation we
havfi
304 QUANTUM KINEMATICS AND DYNAMICS
In the latter form, the dependence upon X appears
in the two transformation functions <t,|t> and
<t|t2> « A second differentiation yields
in which the two terms are equal, the limits of
integration being alternative ways of expressing
the ordered nature of the operator product. One
can also write
CANONICAL TRANSFORMATION FUNCTIONS 305
and the general statement is
Hence the transformation function for the system
0 iwith Barailtonian H = H + H (X=l) can be obtained
formally as a power series expansion about A = 0 ,
and is thus expressed in terms of properties of the
The identification of this result with (7.52)
supplies the expansion of the ordered exponential
0system with Hamiltonian H
306 QUANTUM KINEMATICS AND DYNAMICS
The transformation function construction given in
(7.52) is the foundation of perturbation theory,
whereby the properties of a dynamical system are
inferred from the known characteristics of another
system. The expansion (7.58) is the basis of re-
lated approximation procedures.
7.7 USE OF THE SPECIAL CANONICAL GROUP
The properties of the special canonical group
can be exploited as the foundation of a technique
for obtaining canonical transformation functions.
If we are interested in the Kami1ton-Jacobi transfor-
mation, with infinitesimal generator - H fit we
CANONICAL TRANSFORMATION FUNCTIONS 307
consider the extended transformation characterized
by
which includes infinitesimal displacements of the
canonical variables. On supposing such displacements
to be performed independently within each infinitesi-
mal time interval, as described by
and the extended transformation appears as a Hamil-
ton-Jacobi transformation with an effective Hamil-
tonian. The corresponding equations of motions are
which exhibits the independent changes in the
canonical variables that occur in a small time
interval. If the displacements Q and P area a
localized at a time t which means that
the infinitesimal generator becomes
308 QUANTUM KINEMATICS AND DYNAMICS
the equations of motion imply a finite discontinuity
in the canonical variables on passing through the
time tQ ,
As an application of the latter result, the
arbitrary eigenvalues of canonical variables that
specify states at a certain time can be replaced by
convenient standard values, provided the compensat-
ing canonical displacement is included. Thus with
tx > t2
where the displacements localized at the terminal
tiroes,
do indeed convert the standard states into the
desired ones at times t.(-0) and t2(+0) , A
proof of equivalence can also be obtained from
CANONICAL TRANSFORMATION FUNCTIONS 309
(7.52), applied to produce the transformation
function of the system with Hamiltonian
H + I (p Q -Pq) from that with Hamiltonian H <
With the localized displacement (7.66), the ordering
of the exponential operator is immediate and
since the two exponential operators produce the
required canonical transformations at times t, and
7.8 VARIATIGNAL DERIVATIVES
A more significant use of the special canoni-
cal group appears on considering arbitrary displace-
ments throughout the interval between t. and t_
t23
310 QUANTUM KINEMATICS AND DYNAMICS
It is now convenient to unify the canonical varia-
bles and write
An infinitesimal variation of X (t) produces thect
corresponding transformation function change
which defines the left variational derivative ofi
the transformation function with respect to X (t)
The operator structure of the displacements that
is used here,
CANONICAL TRANSFORMATION FUNCTIONS 311
refers explicitly to variables of the second kindi
where the X are anticommuting exterior algebract
elements and p is the operator that anticommutes
with all dynamical variables, but it also covers
variables of the first kind on replacing "anticommut-
ing" by "commuting". In particular, p •> 1 . The
effect of a second variation is obtained from (7.55)
as
and
312 QUANTUM KINEMATICS AND DYNAMICS
where
It follows from the properties of the 6X1 , or ex-
plicitly from (7.74), that variational derivatives
are commutative, with the exception of those referr-
ing entirely to variables of the second kind, which
are anticommutative.
In the limit as t •* t1 in a definite sense,
we obtain operator products referring to a common
time in which the order of multiplication is still
determined by the time order, Thus
whereas
CANONICAL TRANSFORMATION FUNCTIONS 313
in which the minus sign appears only for pairs of
second kind variables. The difference between the
two limits is related to the commutation properties
of the fundamental dynamical variables
where the ambiguous bracket implies the commutator,
for variables of the first kind or a single variable
of the second kind, and becomes the anticommutator
for variables of the second kind. Now according to
the equations of motion for the system with Hamil-
tonian
the change of the dynamical variables in a small
time interval, as determined by the displacement, is
314 QUANTUM KINEMATICS AND DYNAMICS
Hence
and
which combines the commutation properties of all
the fundamental variables.
Products of three operators at a common time
are expressed, by a formula of the type (7.76), as
and more generally, if p(x(t)} is any algebraic
function of the dynamical variables at time t , re-
stricted only to be an even function of variables
of the second kind, we have
CANONICAL TRANSFORMATION FUNCTIONS 315
in which the variational derivatives refer to times
differing infinitesimally from t as implied by
the particular multiplication order of the operators,
If the Hamiltonian operator is an algebraic function
of the dynamical variables, we can utilize this
differential operator representation through the
device of considering a related system with the
Hamiltonian AH , and examining the effect of an
infinitesimal change in the parameter A ,
This differential equation is analogous in structure
to a Schrodinger equation, with the transformation
function, in its dependence upon the infinite num-i
ber of variables X (t) , t, > t > t- , appearinga L ~ ~ e.
as the wave function, On integrating from A = 0
to A-l , the formal solution,
316 QUANTUM KINEMATICS AND DYNAMICS
exhibits the transformation function as obtained by
a process of differentiation from the elementary
transformation function that refers only to the
special canonical transformations. A more general
form emerges from the decomposition
It must be noted that, since H and H are
necessarily even functions of the variables of the
second kind, the associated differential operators
are commutative. On applying first the exponential
operator constructed from H , we obtain
which could also be derived directly by appropriate
modification of (7.52). This produces a basis for
perturbation theory, by which the desired transfer-
CANONICAL TRANSFORMATION FUNCTIONS 317
mation function for Hamiltonian H is obtained from
a simpler one for Hamiltonian H by
differentiation.
7.9 INTERACTION OF TWO SUBSYSTEMS
The general dynamical situation of a system
formed of two sub-systems in interaction is des-
cribed by
where x1 and x_ refer to the dynamical variables
of the respective sub-systems. Correspondingly the
displacements X decompose into two sets X, and
X~ . In the non-interacting system described by
the Hamiltonian H , the two sub-systems are
dynamically independent and, in accordance with
the additive structure of the action operator, the
transformation function appears as the product of
that for the separate sub-systems,
318 QUANTUM KINEMATICS AND DYNAMICS
The transformation function for the interacting
systems is therefore given by
In this construction, both sub-systems appear quite
symmetrically. It is often convenient, however, to
introduce an asymmetry in viewpoint whereby one
part of the system is tnought of as moving under
the influence of the other. The Hamiltonian
^l^xl' "*" ^12 xl ' o(t}) describes the first system
only, as influenced by the external disturbance
originating in the second system, with its variables
regarded as prescribed but arbitrary functions of
the time. The transformation function for this
incomplete system will be given by
CANONICAL TRANSFORMATION FUNCTIONS 319
and (7.91) asserts that the complete transformation
function can be obtained by replacing the prescribed
variables of the second system by differential
operators that act on the transformation function
referring to the second system without interaction,
The displacements have served their purpose, after
the differentiations are performed, and will be
placed equal to zero throughout the interval
between t. and t_ . They will still be needed
at the terminal times if the transformation functions
refer to standard eigenvalues, but will be set equal
320 QUANTUM KINEMATICS AND DYNAMICS
to zero everywhere if this device is not employed.
Considering the latter for simplicity of notation,
the transformation function of the system with
Hamiltonian H is obtained as
which has the form of a scalar product evaluated by
the differential composition of wave functions.
Accordingly, from the known equivalent evaluations
of such products, other forms can be given to (7.94) ,
such as
and, for variables of the first kind, one can sub-
stitute composition by integration for the differ-
CANONICAL TRANSFORMATION FUNCTIONS 321
ential method here employed.
7.10 ADDENDUM: INTERIOR ALGEBRA AND THE ACTION
PRINCIPLE1"
tReproduced from the Proceedings of the National
Academy of Sciences, Vol. 48 pp. 603-611 (1962).
Reprinted from the Proceedings of the NATIONAL ACA&EM* OF SCIENCESVol. 48, No. 4, pp. 803-611, April, 1962.
EXTERIOR ALGEBRA AND THE ACTION PRINCIPLE, /*
BY- JULIAN SCHWINOBB
HAKVABD UNIVEBSITY
Communicated February %7, 19&%
The quantum action principle1 that has been devised for quantum variables oftype v = 2 lacks one decisive feature that would enable it to function as an instru-ment of calculation. To overeome_ this difficulty we shall enlarge the numbersystem by adjoining an exterior or Grassmann algebra.2
An exterior algebra is generated by N elements «„ « = 1 • • .V, that obey
which includes «,J = 0.
A basis for the exterior algebra is supplied by the unit element and the homo-geneous products of degree d
for d = 1 • • JV. The total number of linearly independent elements is counted as
The algebraic properties of the generators are unaltered by arbitrary nonsingularlinear transformations.
To suggest what can be achieved in this way, we consider a new class of special
604 PHYSICS: J. SCBWINGSR PBOC. N. A, 8.
variations, constructed as the product of £2n+1 with an arbitrary real infinitesimallinear combination of the exterior algebra generators,
The members of this class antieommute with the & variables, owing to the factorf2«-f.!, and among themselves,
since they are linear combinations of the exterior algebra generators. Accordingly,the generator of a special variation,
commutes with any such variation,
If we consider the commutator of two generators we get
where the right-hand side is proportional to the unit operator, and to a bilinearfunction of the exterior algebra generating elements,
The latter structure commutes with all operators and with, all exterior algebraelements. The commutator therefore commutes with any generator 0(, and thetotality of the new special variations has a group structure which is isomorphicto that of the special canonical group for the degrees of freedom of type » » = * » .This, rather than the rotation groups previously discussed, is the special canonicalgroup for the v = 2 variables.
We are thus led to reconsider the action principle, now using the infinitesimalvariations of the special canonical group. The class of transformation generators0 that obey
includes not only all even operator functions of the In variables {», but also evenfunctions of the ft that are multiplied by even functions of the N exterior algebragenerators e,, and odd functions of the J* multiplied by odd functions of the ««.The generators of the special canonical transformations are included in the lastcategory.
The concept of a Hermitian operator requires generalization to accommodatethe noncommutative numerical elements. The reversal in sense of multiplicationthat is associated with the adjoint operation now implies that
where X is a number, and we have continued to use the same notation and language
VOL. 48,1962 PHYSICS: J. SCMWINGSR 805
despite the enlargement of the aumber system. Complex conjugation thus hasthe algebraic property
The anticonimutativity of the generating elements is maintained by this operation,and we therefore regard complex conjugation in the exterior algebra as a linearmapping of the JV-dimensional subspaee of generators,
The matrix R obeys
which is not a statement of unitarity. Nevertheless, a Cayley parameterizationexists,
provided det(l + R) * 0. Then,
obeys
which asserts that a basis can always be chosen with real generators. There stillremains the freedom of real nonsingular linear transformations.
This conclusion is not altered when R has the eigenvalue — 1. In. that circum-stance, we can construct p(R), & polynomial in R that has the properties
The matrix
also obeys
while
since the contrary would imply the existence of a nontrivial vector » such that
or, equivaleatly,
which is impossible since p(—1) = 1. Now,
and if we use the Cayley construction for R' in terms of a real matrix rft which is
a function of R with the property r'(—1) *= 0, we get
600 PHYSICS: J, SCHWINGER PBOC. N. A. S,
This establishes the generality of the representation
where the nonsingular matrices a and p* are commutative, and thereby provesthe reality of the generator set
When real generators are chosen, the other elements of a real basis are suppliedby the nonvanishing products
for
A Hermitian operator, in the extended sense, is produced by linear combinationsof conventional Hermitian operators multiplied by real elements of the exterioralgebra. The generators Gt are Hermitian, as are the commutators i[Cf^, (?'!*].
The transformation function (n | rz) associated with a generalized unitary trans-formation is an element of the exterior algebra. It possesses the properties
and
where X symbolizes the summation over a complete set of states which, for themoment at least, are to be understood in the conventional sense. These attributesare consistent with the nature of complex conjugation, since
For an infinitesimal transformation, we have
where 0 is Hermitian, which implies that its matrix array of exterior algebraelements obeys
The subsequent discussion of the action principle requires no explicit referenceto the structure of the special variations, and the action principle thus acquires adual significance, depending upon the nature of the number system.
We shall need some properties of differentiation in an exterior algebra. Since«x2 = 0, any given function of the generators can be displayed uniquely in thealternative forms
where /a, /i, /, do not contain «x- By definition, /, and ft are the right and leftderivatives, respectively, of /(«) with respect to «x,
Vou 48, 1962 PBY8ICS: J. SCHWINGMR mi
If /(«) is homogeneous of degree d, the derivatives are homogeneous of degreed — I , and left and right derivatives are equal for odd d, but of opposite sign whend is even. For the particular odd function «w we have
so that
Since a derivative is independent of the element involved, repetition of the opera-tion annihilates any function,
To define more general second derivatives we write, for X <& it,
where the coefficients are independent of sx and «„. The last term has the alterna-tive forms
with
Then,
and
which shows that different derivatives are anticommutative,
A similar statement applies to right derivatives.The definition of the derivative has been given in purely algebraic terms. We
now consider /(« + $«), where St\ signifies a linear combination of the exterioralgebra elements with arbitrary infinitesimal numerical coefficients, and concludethat
to the first order in the infinitesimal numerical parameters.If this differential
property is used to identify dervatives, it must be supplemented by the require-ment that the derivative be of lower degree than the function, for any numbericalmultple of E1E2..EN can be added to a derivative without changing the differential
608 PHYSICS: J. SCHWINGMR PBOC. N. A. S.
form. Let us also note the possibility of using arbitrary nonsingular linear combi-nations of the «x in differentiation, as expressed by the matrix formula
where the If* are variations constructed from the exterior algebra elements. Wecannot entirely conclude that
This apparent incompleteness of the action principle is removed on stating theobvious requirement that the transformation function (n| r») be an ordinary numberwhen all Xt(r) vanish. Thus, the elements of the exterior algebra enter onlythrough the products of the X»(T) with &(r), as the latter are obtained by inte-grating the equations of motion, and terms in these equations containing «!••«»are completely without effect,
Let us examine how the transformation function (rijrj}cx depends upon the
X»(r). It is well to keep in mind the two distinct factors that compose XI,(T},
which follows directly from the differential expression,We shall now apply the extended action principle to the superposition of two
transformations, one produced by a conventional Hermitian operator G, an evenfunction of the ft, while the other is a special canonical transformation performedarbitrarily, but continuously in T, The effective generator is
where —Xt(r)dr is the special variation induced in |jt(r) during the interval dr.The objects -3T*(r) are constructed by multiplying %in+i with a linear combinationof the «x containing real numerical coefficients that are arbitrary continuous func-tions of r. To use the stationary action principle, we observe that each XI(T),as a special variation, is commutative with a generator of special variations andthus
The action principle then asserts that
since there remains an arbitrariness associated with multiples of «> • • ew> as in theidentification of derivatives from a differential form. No such term appears,however, on evaluating
Here,
VOL, 48, 1962 PHYSICS: J. SCHWINGER 809
while Xt'(r) is entirely an element of the exterior algebra. Thus, it is really theJT'(r) upon which the transformation function depends. An infinitesimal changeof the latter induces
where the summation index k has been suppressed. The repetition of such varia-tions gives the ordered products
which form is specific to an even number of variations. Complex conjugation isapplied to the exterior algebra elements in order to reverse the of multi-plication. In arriving at the latter form, we have exploited the fact that specialcanonical variations are not implicit functions of r and therefore anticommutewith the £* without regard to the r values. Thus, one can bring together the1m quantities &X(rl), • -SX^™), and this product is a multiple of the unit operatorsince pl =» 1. The multiple is the corresponding product of the exterior algebraelements MT'Cr1), • •&Z'(r2at), which, as an even function, is completely commuta-tive with all elements of the exterior algebra and therefore can be withdrawn fromthe matrix element. The reversal in multiplication sense of the exterior algebraelements gives a complete account of the sign factors associated with antieom-mutativity.
The notation of functional differentiation can be used to express the result.With left derivatives, we have
where the antieontmutativity of exterior algebra derivatives implies that
Here,
according to whether an even or odd permutation is required to bring r1, • • r2"1
into the ordered sequence. This notation ignores one vital point, however. Inan exterior algebra with N generating elements, no derivative higher than theNth exists. If we wish to evaluate unlimited numbers of derivatives, in order toconstruct correspondingly general functions of the dynamical variables, we mustchoose N = <*>; the exterior algebra is of infinite dimensionality. Then we canassert of arbitrary even ordered functions of the |»(T) that
In the alternative right derivative form, &T signifies that successive differentiationsare performed from left to right rather than in the conventional sense. We alsonote that the particular order of multiplication for operator products at a commontime is to be produced by limiting processes from unequal r values. As an appliea-
610 PHYSICS: 3. SCHWISGER Pnoc, N. A. S,
tion to the transformation function (n \ Tt)gx, we supply the even operator function
0 with a variable factor X and compute
which gives the formal construction
where the latter transformation function refers entirely to the special canonicalgroup.
The corresponding theorems for odd ordered functions of the f*(r) are
and
provided the states are conventional ones. Since the variations SX(r) have noimplicit T dependence, the factors pe can be referred to any r value. If this ischosen as n or T% there will remain one p operator while the odd product of exterioralgebra elements can be withdrawn from the matrix element if, as we assume, theproduct commutes with the states (n| and TJ). The sign difference between theleft and right derivative forms stems directly from the property {AST = —iXf.
As an example, let F(Q be the odd function that appears on the left-hand sidein the equation of motion
and
Some properties that distinguish the trace of the transformation function canalso be derived from the statement about odd functions. Thus,
The corresponding functional derivatives are evaluated as
since the X(r) are special canonical displacements, and this yields the functionaldifferential equations obeyed by a transformation function {n | Tt)0
x, namely,
since either side is evaluated as
YOL. 48, 1962 PHYSICS: J. SCHWINGMB 611
Accordingly,
which shows that the trace is an even function of the X'(r). The nature of thetrace is involved again in the statement
which is an assertion of effective antiperiodicity for the operators {(T) over theinterval T = n — T*. The equivalent restriction on the trace of the transforma-tion function is
or
* Publication assisted by the Air Force Office of Scientific Research.» These PBQCBKBINQS, 47, 1075 (1961).* A brief mathematical description can be found in the publication The Construction, and Study
o/ Certain Important Algebras, by C. Cfaevalley (1955, the Mathematical Society of Japan).Although such an extension of the number system has long been employed in quantum fieldtheory (see, for example, these PROCEEDINGS, 37, 482 (1951)}, there has been an obvious needfor an exposition of the general algebraic and group theoretical basis of the device.
and, in particular,
CHAPTER EIGHTGREEN'S FUNCTIONS
8.1 Incorporation of Initial Conditions 3318.2 Conservative Systems. Transforms 3358.3 Operator Function of a Complex 337
Variable8.4 Singularities 3408.5 An Example 3418.6 Partial Green's Function 343
8.1 INCORPORATION OF INITIAL CONDITIONS
The most elementary method for the construc-
tion of canonical transformation functions associat-
ed with parameterized transformations is the direct
solution of the differential equations that govern
the dependence upon the parameters. For time de-
velopment these are the Schrodinger equations
332 QUANTUM KINEMATICS AND DYNAMICS
where H here refers to the differential operator
representatives of the Hamiltonian» at times
t or t0 , which depend upon the particularJ~ <&
choice of canonical representation. The desired
transformation function is distinguished among the
solutions of these Schrodinger equations by the
initial condition referring to equal times,
which means that the canonical transformation func-
tion is independent of the common time and is deter-
mined only by the relation between the descriptions.
This formalism is given an operator basis on writing
where the unitary time development operator
is to be constructed as a function of dynamical
variables that do not depend upon time, by solving
GREEN'S FUNCTIONS 333
the differential equations
-it is usetux to incorporate tne initial con-*
ditions that characteri-ze transformation functions,
or the time development operator, into the differ-
ential equations. This is accomplished by intro-
ducing related discontinuous functions of time -
the Green's functions or operators. The retarded
and advanced Green's functions are the matrices, in
some representation, of the retarded and advanced
Green's operators defined, respectively, by
where
together with the initial condition
The discontinuities of the functions n and n,
are expressed in differential form by
334
md
QUANTUM KINEMATICS AND DYNAMICS
and therefor, in consequence of the differtial
equations (8.5) and the initial conditions (8.6)
both the retarded and advanced green,s operator
obey the inhomogeneous equations
The two Green's soperators are distingushed as the
solutions of these equations that by
GREENS'S FUNCTIONS 335
8.2 CONSERVATIVE SYSTEMS . TRANSFORMS
For a conservative system, in which t does
not appear in the Hamiltonian operator, the time
development operator and the Green's operators can
depend only upon the relative time,
and the defining properties of the Green's operators
appear as
which is evidently consistent with the adjoint rela-
tion
From these operators the unitary time developmen-
operator is constructed as
336 QUANTUM KINEMATICS AND DYNAMICS
It is now possible to eliminate the time dependence
in the Green's operators by defining the transform
operators
As we have already indicated, since the time
integrations are extended only over semi-infinite
intervals these operators exist for complex values
of the energy parameter E , when restricted, to
the appropriate half-plane. The application of the
transformation to the differential equations (8.16)
yields
GREEN'S FUNCTIONS 337
for both Green's operators which, as functions of the
complex variable E, are now defined by the respective
domains of regularity indicated in (8.17). Corres-
pondingly the adjoint connection now appears as
where the integration path, extended parallel to
the real axis, is drawn in the domain of regularity
appropriate to the Green's operator under considera-
tion.
8.3 OPERATOR FUNCTION OF A COMPLEX VARIABLE
Both Green's operators are given formally by
and therefore form together a single operator fun-
ction of the complex variable E , defined every-
where except perhaps on the common boundary of the
The inversions of (8.17) are comprised in
338 QUANTUM KINEMATICS AND DYNAMICS
two half-planes, the real axis. Indeed, the form
(8.21), expressed in terms of the eigenvectors of
the operator H and a supplementary set of constants
of the motion y , shows that the singularities of
G(E) are simple poles on the real E axis, coin-
ciding with the spectrum of energy values for the
system. The construction of the Green's function
in some convenient representation, and an investi-
gation of its singularities will thus supply the
entire energy spectrum of the system together with
automatically normalized and complete sets of wave
functions for the energy states. For a system with
the Hamiltonian H(q , q) , for example, the
Green's function G(q , q' , E) could be obtained
by solving the inhomogeneous differential equation
On exhibiting the solution as
+
+
GREEN'S FUNCTIONS 339
the desired information concerning energy values
and wavefunctions is disclosed. We should also
note the possibility of a partial Green's function
construction, which supplies information about a
selected group of states. Thus if we place the
eigenvalues q' equal to zero in (8.22), the differ-
ential equation reads
and the solution of this equation will yield the
energy values only for those states with
<E"Y*|0> 0 * &H these states are still repre-
sented in the further specialized Green's function
O
where the coefficients |<E'Y*|O>| obey
and give the probabilities for realizing the various
energy states in a measurement on the zero eigen-
value state 0> .
340 QUANTUM KINEMATICS AND DYNAMICS
8.4 SINGULARITIES
The singularities of a Green's function can
be determined from the discontinuities encountered
on crossing the real E axis. Thus, for real E ,
according to the delta function construction
Note that this is also a measure of the non-Hermi-
tian part of G (E) for real E ,
There is no discontinuity and G is Herraitian at
any point on the real axis that does not belong to
the energy spectrum of the system, whereas a dis-
crete energy value is recognized by the correspond-
ing localized discontinuity. If the energy spectrum
GREEN'S FUNCTIONS 341
forms a continuum beginning at EQ » the discontin-
uity for E > EO is
The existence of such a finite discontinuity for
every E > EQ implies that G(E) possesses a
branch point singularity at E = EQ , the contin-
uous line of poles extending from En to infinity
supplying the cut in the E plane. Thus the
precise nature of the energy spectrum for a system
is implied by the character of the singularities
exhibited by G(E) as a function of a complex
variable.
8.5 AN EXAMPLE
The elementary example of a single free
particle in space is described non-relativistically
in the r representation by the Green's function
equation
342 QUANTUM KINEMATICS AND DYNAMICS
The solution
J«involves the double-valued function E of the
complex energy parameter, which must be interpreted
kas +i|E| for E < 0 in order that the Green's
function remain bounded as |r-r'| -> «• . Accor-ie i-
dingly, for E > 0 we must have E = -f-|E| 2 on
the upper half of the real axis and E = -|E|
on the lower half. Hence there is a discontinuity
across the real axis for E > 0 which constitutes
the entire energy spectrum, since the Green's func-
tion is always bounded as a function of E . The
discontinuity is given by
GREEN'S FPNCTIONS 343
where the integral in the latter form is extended
over all directions of the vector p = p n . On
comparison with (8.30) we see that the various states
of a given energy can be labelled by the unit vector
n , specified within an infinitesimal solid angle
dw . The corresponding wave functions of the r
representation are
where (dp) is the element of volume in the p space.
8.6 PARTIAL GREEN'S FUNCTION
The utility of a partial Green's function
construction appears in two general situations,
which can overlap. One or more compatible constants
of the motion may be apparent from symmetry conside-
rations and it is desired to investigate only states
with specific values of these quantities, or, one
may be interested for classification purposes in
constructing the states of a perturbed system which
correspond most closely to certain states of a re-
lated unperturbed system. Both situations can be
344 QUANTUM KINEMATICS AND DYNAMICS
characterized as a decomposition of the complete
set of states into two parts» or subspaces» as
symbolized by
in which the measurement symbols , or projection
operators, obey
and where one seeks to construct the projected
Green's operator referring only to the subspace
M, ,
From the equation
one obtains
and
GREEN'S FUNCTIONS 345
and we obtain as the determining equation for
where
If the two subspaces refer to distinct values of
constants of the motion for the complete Hamiltonian,
there will be no matrix elements connecting the sub-
spaces, , H_ = ?H, = 0 » and (8.42) reduces to the
fundamental form of the Green's operator equation,
which we write as
The second equation is formally solved, within the
subspace ML , by
346 QUANTUM KINEMATICS AND DYNAMICS
(8.18), now defined entirely within the space ML
CHAPTER NINESOME APPLICATIONS AND FURTHER DEVELOPMENTS
9.1 Brownian Motion of a Quantum Oscillator 3479.2 Coulomb Green's Function 374
9.1 BROWNJAN MOTION OF A QUANTUM OSCILLATORt
tReproduced from the Journal of MathematicalPhysics, Vol. 2, pp. 407 -432 (1961).
347
Reprinted from the JOURNAL OF MATHEMATICAL Pavsicss Vol. 2, Mo. 3, 407-432, Jtinted S» V- S, A,
lay-June, 1961
Browaian Motion of a Quantum Oscillator
JCUAS SdffiWIUQES*Elt&wfd Uniserstfy, Csm^fi4ge9 Massachuseiis
(Received November 28,I960)
Aa action principle tec&nkjue for the direct coasputatioR of esEp^etatioti values is described aad illustratedm detail by a special physical example, tfe« effect on an oscillator of another physics! system. This simpleproblem has the advaatage of orobiaiiisg immediate physical applicability (e,f,, resistive damping ormaser amplification of a single electromagnetic cavity mode) with a slgniieaat idealisation of the complexproblems encountered in many-particle and relativiatic fieM titeory. Successive sections eemtain discussionsof the oscillator subjected to external forces, the oscillator loosely coupled io the external systemt aaimproved treatment of this problem and, finally, there is a brief account of a general fonrnsktkta.
IfTOGOTCTIG!?
HE title of this paper refers to an elementaryphysical example that we shall as© to illustrate,
at some length, a solution of the loHowimg methodologicalproblem, The quantum action principle1 is a differentialcharacterization of transformation functions, (o'fc 1 &%}»and thus is ideally suited to the practical computationof transition probabilities (which includes the deter-mination o£ stationary states). Many physical questionsdo not pertain to individual transition probabilities,however^ btit rather to expectation vaiyes of a physicalproperty for a specified initial state,
or, more generally, a mixture of states. Can one deviseais aciiSoa principle technique that is adapted to thedirect computation of such expeefcati&B values, withoutrequiring knowledge of the individual transformationfunctions?
The action principle asserts that {^^ 1}>
which expresses the fked numerical value of
But now, imagine that the positive and negative sensesof time development are governed by differtot dy-namics. Then the transformation function for the closedcircuit will be described by the action principle
ami
in which we shall take ^>la. These mutually complex-conjugate forms correspond to the two viewpointswhereby states at different times can be compared,either by progressing forward from the earlier time* orbackward from the later time. The relation betweenthe pair of transformation functions is such that
in which abbreviated notation the multiplication signsymbolizes the composition of transformation functionsby summation over a complete set of states, Ilf inparticular, the Lagraagiaa operators Z± contain thedynamical term X±(<)^T(<), we have
and, therefore^
* Supported by the Mi Force Office of Scientific Research(ARDCV1 Some reierenees are: Julias Scfewinger, Fhys. Rev. , 9M(!»51}j 91, ?J3 (!9»)j Phil Mag. 44, H?J (WSJ). The first twopapers also appear io Selected Papers OK Quantum Ele£tf®dyn®mfo$(Dover Pttbffcallons, New fork, »S8), A recent discussion iscontained in Julian Scbwinger, Proc. JMatl. Acad. Sd. U. S. 46,883 (I960).
where Xj. can BOW be identified. Accordingly, if asystem is suitably perturbed2 in a manner that dependsupon the time sense, a knowledge of the transformationfunction referring to a dosed time path determines theexpectation value of any desired physical quantity fora specified initial state or state mixture.
OSCILLATOR
To illustrate this remark we first consider an oscillatorsubjected to an arbitrary external force, as described bythe Lagrangian operator
a ESespite this dynamical language, a change in the Hamiltoroan
ratot of a system can be kineiratical in character, arising fromconsideration of another transformation along with (lie
dynamical aw generated by the Hamiltonkn. See Use last paperquoted in footnote I, sod Julias Schwinger, Proc Natl Acad.Sd. U. S. «, 1401 (1«0).
407
7
408 J . S C H W I N G E R
We shall begin by constructing the transformationfunction referring to the lowest energy state of the andunperturbed oscillator, (GfelQfe)-**, This state can becharacterized by
in which the complementary pair of nor*-Hermiiian The choice of initial state implies effective boundaryoperators y, iy>, are constructed from Hermitian conditions that supplement the equations of motion,
The equations of motion implied by the action principleare
and solutions are given by
together with the adjoint equation. Since we nowdistinguish between the forces encountered, in thepositive time sense, K+(t), K+*(i), and in the reversetime direction, K~.(t), K~*{t), the Integral must betaken along the appropriate path. Thus, when I isreached first in the time evolution from 4, we have
together with the similar adjoint equations obtained byinterchanging the ± labels. For convenience, stepfunctions have been introduced:
or, equivalently, by the eigenvector equations
Since the transformation function simply equalsunity if K+***K~ and K+*~K-*, we must examine theeffect of independent changes in K+ and JC_, and ofJC+* and K-", as described by tie action principle
The requirement that the transformation functionreduce to unity on identifying X+ with £_, K+* withK~.*, is satisfied by the null sum of all elements of Ge,as assured by the property %.+>!_= 1.
An operator interpretation of Gt is given by thesecond variation
Generally, on performing two distinct variations ia thestructure of L that refer to parameters upon which
operatorss q,p, by
Hence in effect we have
while on the subseqent return to time t,
The solution of the resulting integrable differentialexpression for log(ot2/ ot2)k is given by
We shall also havcasion to use the odd function
Note
in a matrix notation with
and
B R O W N 1 A N M O T I O N O F A Q U A N T U M O S C I L L A T O R 409
If we are specifically interested in {afef^s)'8'*, whichsupplies all expectation values referring to the initialstate nt we myst extract the coefficient of (3t*'y'l)*/nlfrom aa exponential of the form
All the terms that contribute to the required coefficient
s A discussion of non-Hermitjan representations is given mLectures on Quantum Mec&snies (Les HoycheSj 1955) unpublished.
iims, unaer the influence oi these lorces, me states|%) and (0^1 become, at the time fe-f-O, the statesiy%) &&d (yf%], which are right an<l left eigenvectors!respectively, of the operators y(fe) and yf(J»). OH takiaginto account arbitrary additional forces* the transforma-tion function for the closed time path can be expressed
where the expectation values and operators refer to thelowest state and the dynamical variables of the un~perturbed oscillator. The property of Go that the sumof rows and columns vanishes is here a consequence ofthe algebraic property
The choice ol oscillator ground state Is no essentialrestriction since we eas now derive the analogousresults for any initial oscillator state. Consider, forthis purpose, the impulse forces
the effects of which are described by
Tfae eigenvectors of the Bon-Hermitiaa canonicalvariables are complete and have an intrinsic physicalInterpretation in terms of g and p measurements ofoptimum compalibility.3 For otir immediate purposes.,however, we are roore interested in the unperturbedoscillator energy states. The connection between the twodescriptions can be obtained by considering theunperturbed oscillator transformation function
in which the multiplication order follows the sense oftime development. Accordingly*
the dynamical variables at a gives time are not explicitly &nddependent, we have
We infer She norincgative integer spectrum o! yry( andthe corresponding wave functions
since
Accordingly^ a non-Hermitian canonical variable trans-formation function can serve as a generator for thetransformation (unction referring to unperturbedoscillator energy statesr
in which
Now
W h c h Y i e l s d s
a s
410 J . S C H W I N G E R
are contained m
and since the elements of G& are also given by un-perturbed oscillator thermal expectation values
inhere we have written
where the latter version is obtained from
andthe designation {n}$ is consistent with its identificationas (yW*
The thermal forms can also be derived directly bysolving the equations of motion, in the manner used tofind (0^ I Qi%}K*. On replacing the single diagonal element
in which the nth Lagaerre polynomial has been, intro-duced on observing that
we find the following relation
One obtains a much neater form, however, from whichthese results can be recovered, on considering anInitial mixture of oscillator energy states for whichthe nth state is assigned the probability
instead of the effective initial condition {^^O. Thisis obtained by combining
s*ith the property of the trace
and
can be interpreted as a temperature. Then, since
We also hav;
and therefore} effectively,
Hence, to the previously determined y^(t) is to beadded the term
we obtain
with
and in whichand correspondingly
which rej>reduces the earlier result.
bby the statistical avera
T h u s
B R O W N 1 A N M O T I O N O F A Q U A N T U M O S C I L L A T O R 411
As an elementary application let us evaluate the there appearsexpectation value of the oscillator energy at time t\tot a system that was in thermal equilibrium at time (5and is subsequently disturbed by an arbitrarily time-varying force. This can be computed as
It may be worth remarking, in connection with theseresults, that the attention to expectation values doesThe derivative i/SK+*(k) supplies the factor * j • * .1 ».•!• .,. ,, ,
not (Jepnve uj of ti,e ability to compute individualprobabilities. Indeed, if probabilities for specific oscil-lator energy states are of interest, we have only toexhibit, as functions of y and y1, the projection operatorsfor these states, the expectation values of which are the
the subsequent variation with respect to K.(h) gives required probabaitics. frow
is represented by the matrix
and the required enetgy expectation value equals *""> therefore,
u{n)0+u
More generally, the expectatian values of all functions of y((.) and y«(fc) are known from that of
exp{ - i[X/{h)+MyOi)]}, n whicl, we vs jntro<iuced a notation to indicate thisand this quantity is obtained on supplementing K+ ordered multiplication of operators. A convenientand K-t* by the impulsive forces (note that in this use generating function for these projection operators isof the formalism a literal eoraplex-conjugat© relationshipis not required)
K+(t)-\t(t h), g(ves tj,e probability of finding the oscillator in theK^.*(t)"iit{l—lt+0), nth energy state after an arbitrary time-varying force
Then
and we observe that
Accordingly,
which involves the special step-function value
JtsmaHvelv. if Wf* rh*msf>
412 j . S C H W 1 N G E R
has acted, il it was initially m a thermal mixture ofstates,
To evaIlia If?
rhls ferai; and the implied transition probabilities, haveilready been derived in another connection,4 and weshall only state the general resell here;
in which ra> and ^< represent the larger and smaller ofthe two integers n and nf.
Another kind of probability is also easily identified,that referring to the continuous spectrum erf theMermitian operator
from which foilo^
we first remark that
If we multiply this result by exp(—»#Y) and integratewith respect to p'/ln: from — « to «, we obtain theexpectation value of Qf {**) ""* $1 which is the probabilityof realizing a value of q(t\) in a unit interval about (/i
Still another derivation ojf the lormuk giving thermalexpectation values merits attention. Now we let thereturn path terminate al a different time l/=4~"^»and on regarding the resulting transformation functionas a matrix, compute the traces or rather the tracerattn
which reduces to unity in the absence of externalforces. The action principle again describes the dqsend-enee upon JC±*(/}, K±(t) through the opera-tore y±0)»y^(f) which are related to the forces by the solutionsof the equations of motion, and, is particular*
and <m referring 10 the previously used Lagiier:polynomial sum formula, we obtain
as one shows with a similar procedure^ or by directseries expari&iors, Tfeeref&re.,
Here
or
In addition to describing the physical situation o(initial thermal equiliibrisimt this result provides agenerating function for the Individual transitionprobabilities between oscillator energy states,
Next we recognize that the structure ol the trace impliesthe effective boundary condition
w i t h
w h r e r
julan Schwinger Phy, Rev 91,728,1953
For this purpose,we place
abd obtain
B R O W N I A N M O T I O N O F A Q U A N T U M O S C I L L A T O R 413
Let us consider
ixW\y-(t*')\ti-Z« <aVb-(V)k%),where we require of the a representation only that ithave no explicit time dependence. Then
a&e
which is the slated result.T%e ffffrfiw? initial rnrtHitmn nnw aranp&rs asi
in which /,#,({ characterizes the external system andQ(i) is a Hermltlan operator of that system.
We begin our treatment with a discussion of thetransformation function {(»1*^*9*** that refers initiallyto a thermal mixture at temperature $ lor the externalsystem, and to an Independent thermal mixture attemperature$9 for the oscillator. The latter temperaturecan be interpreted literally, or as a convenient para-metric device lor obtaining expectation values referringto oscillator energy states. To study the effect of thecoupling between the oscillator and the externalsystem we supply the coupling term with a variableparameter \ and compute
and the action principle supplies the following evalua-tion of the trace ratio: where the distinction between the forward and return
paths arises only from the application of differentexternal forces &&(!} OR the two segments of the closedtime contour. The characterization of the externalsystem as essentially macroscopic now enters throughthe assumption that this large system is only slightlyaffected by the coupling to the oscillator. In a corre-sponding first approximation, we would replace theo|>erat0rs Q&(t) by the effective aumerical quantity(CCO)*- The phenomena that appear in this order ofaccuracy are comparatively trivial, however^ and weshall suppose that
where the time variable in K+ and £„ ranges from£g to It and from I/ to l\t respectively. To solve the givenphysical problem we require that &.({) vanish le theinterval between // and t$ so that all time integrations3l"P *ixt**nrfw'! HpfWfPfTi to anH f.t TThj^n einff
which forces us to proceed to the next approximation.A second differentiation with respect to X giveswfaat has been evaluated equals
aad by addmg the remark that this ratio continues Uexist on making the complex substitution
the desire<l formula emerges asThe introduction of an approximation based upon theslight disturbance at the macroscopic system convertsthis into
EXraRHAt SYSTEM
This concludes our preliminary survey of the oscil-lator and we turn to the specific physical problem &iinterest: An oscillator subjected to prescribed externalforces and loosely coupled to an essentially macroscopicexternal system. All oscillator interactions are linearin the oscillator variables, as described by the Lagran-glan operator
whre
414 J . S C H W I N G E R
and we have also discarded all terms containingy(l)y(t') and /(«}/(<')• The latter approximationrefers to the assumed weakness of the coupling of theoscillator to the external system, for, durmg the manyperiods that are needed tor the effect of the coupling toaccumulate, quantities with the time dependencee±«0e(H-i'> ^rjj} Ijecome suppressed in comparison withthose varying as g^***"™1'). At this point we ask whateffective term in an action operator that refers to thedosed time path of the oscillator would reproduce thisapproximate value of (9/dX)*(fs\t,} at A=>0. The
complete action that satisfies this requirement, withXs set equal to unity, is gives by
The nonlocal character of these equations is not verymarked if, for example, the correlation between Q(t)and Q(f} in the macroscopic system disappears when\t—lf\ is still small compared with the period of theoscillator, Then, since the behavior of y(t) over a shorttime interval is given approximately by «"**', the matrixA (<—;') is effectively replaced by
The application of the principle of stationary actionto this action operator yields equations of motion thatare nonlocal in time, namely, *
and the equations of motion read
The latter set is also obtained by combitning the formaladjomt operation with the interchange of the + and~ labels attached to the operators and K{t). Anothersignificant form is conveyed by the pair of equations
One consequence
and
ana
it should be noted that A+,(a) and 4_+(a) are reaKjsitive Quantities since
It also follows from
that
is real. Furthermore
where
whers
and
together with
and
B R O W N I A N M O T I O N O F A Q U A N T U M O S C I L L A T O R 415
and the required solution is given by
and, for the choice at an initial thermal mixture,
We now find that
The net result of this part of the discussion is toremove al! explicit reference to the external system asa dynamical entity. We are given elective equations ofmotion for y+ and y^ that contain the prescribedexternal forces and three parameters, the angularfrequency »(£»»»), 7, and <t, the latter pair beingrelated by the temperature of the macroscopic system.
We have not yet made direct reference to the natureof the expectation value for the macroscopic system,which is now taken as the thermal average;
where S is the energy operator of the external system.The implication for the structure of the expectationvalues is contained in
The corresponding solutions for y^(tj are obtained byinterchanging the ± labels in the formal adjointequation.
The differential dependence of the transformationfunction (fe \ ti)»t>tK* upon the external forces is describedby these results, and the explicit formula obtained onintegration is
which employs the formal property
On introducing the "time Fourier transforms, howevithis becomes the explicit relation
and we conclude that
which is a positive even function of . As a consequentwe have
which can also be written as
so that
snd « emerges as the real quantity
although the simplest description of G is supplied by
The accompanying boundary conditions areA n o t h e r w a y o f p e r e a e n t i n g t h i s r e s u l t s i s
w h i c h s u p p l i t e s t h e i n i t i a l c o n d i t i r o n f o r t h e s
e q u a t i o n o f m o t i o n
416 J . S C H W I N G E R
the differential equation
(where f indicates differentiation to the left) hconjunction with the initial value
We note the vanishing sum of ail G elements, and thatthe rote of complex conjugation in exchanging the twosegments of the dosed time path is expressed by
which is to say that
It will be observed that only when
is Gfg»(<— f t , f—1%) independent of k and a function of(—('. This dearly refers to the initial physical situationof thermal equilibrium between the oscillator and theexternal system at the common temperature <>»=$> 0,which equilibrium persists ia the absence of externalforces. If the initial circumstances do not constitutethermal equilibrium, that will be established in thecourse of time at the macroscopic temperature tf>0.Thus, all reference to the initial oscillator temperaturedisappears from G»e>(t— Is, (—fe) when, for fixed t—t.
from which a variety of probability distributions andexpectation values can be obtained.
The latter calculation Illustrates a general character-istic of the matrix G(t,(), which is implied by thelack of dependence on the time ti. Indeed, such aterminal time need not appear explicitly in the structureof the transformation function {k\i%)K* and all timeintegrations can range from it to +<*>. Then ti isimplicit as the time beyond which K+ and K~ areidentified, and the structure of G must be such as toremove any reference to a time greater than t\. In thepresent situation, the use of an impulsive force at hproduces, for example, the term
in which K+ and K~ are set equal. Hence it is necessarythat
which says that adding the columns of G(/,f) givesretarded functions of/ £—/, while the sum of rowssupplies a vector that is an advanced function oft—t'. In each instance, the two components must have asero sum- These statements are immediately verified forthe explicitly calculated C?^Q>?(£—la, ?—t%) and followmore generally from the operator construction
and the boundary conditions
A more symmetrical version of this differential equatio!is given by
The previously employed technique of impulsive force;applied at the time t\ gives the more general result
and is expressed by
md similarly that
The thermal relaxation of the oscillator energy is
for, as we have already noted in connection with Qproducts,
d e r v i e d f o r m
B R O W N I A H M O T I O N O F A Q U A N T U M O S C I L L A T O R
and
Our results show^ incidentally, that
This is a formula for the direct computation, of expecta-tion values of general functions of y(t) and yfO). A lessexplicit but simpler result can also be givesi by means ofexpectation values for functions of the operators
Another general property can be illustrated by ourcalculation, the posttlveness of -~iG(t/)+~t
Let us recognize at once that
and therefore that the fluctuations of y(t)t yr(/) can beascribed to the effect of: the forces Kft jf/, which appearas the qtiantam analogs of the random forces In theclassical l.angevin approach to the theory of theBrownian motion. The change in viewpoint is accom-plished by introducing
and it is clearly necessary that each term obey sep-arately the positiveness requirement. The first term is
where we assume, just for simplicity f that the functionsw(l), ?(/} vanish at the time boundaries* Then, partialtime integrations will replace the operators % y* withKt, K,t.
To carry this out, however, we need the followinglemma on time-ordered products:
tnd the required property of the second term, followsrom the formula
which involves the unessential assumption that B(i)vanishes at the time terminals; and the hypothesis that[X(0,#(0] and [dB(t}/dt, B(tYj are commutativewith all the other operators. The proof is obtained byreplacing B(t) with XJS(0 aad diiferentlatJBg withrespect to X,
All the information that has been obtained aboutthe oscillator is displayed on considering the forces
and making explicit the effects ofequivalent time-ordered operators:
Tfees, a partial integration, yields
417
418 j . S C H W I N G E R
according to the hypothesis, and the stated resultfellows &n integrating this differential equation,
Tbe structure of the lemma Is given; by the rearrange-ment
Such expectation values are to be understood as effectiveevaluations that serve to describe the properties of theoscillator under the circumstances that validate thevarious approximations that have bees used,
It will be observed that when » is sufficiently largeto permit the neglect of all other terms,
for then
and the seEse ol operator multiplication is no longersignificant. This is the classical limit, for which
wb ere we have placed % —• -«_—tt% &$, ™ tr_ ~ i f . Onintroducing real components of the raadom force
Kfiri^+iKd, ir/'-z-Htfi-flCt),the classical limiting result reads
The fluctuations at different times are independent. IIwe consider time-averaged forces,
which is the Gaussian distribution giving the probabilitythat the force averaged over a time interval &£ willhave a vatae within a small neighborhood of the pointK'. In this classical limit the fluctuation constant a isrelated to the damping or dissipation constant 7 andthe macroscopic temperature $ by
ine last term involves aiscaromg a touu time oenvauvethat will not contribute to the final result. To evaluate[4 3] we must refer to the mean-teg of K/ and K/ thatis supplied by the actual equations of motion.
which is also proportional to the unit operator. Ac-cordingly,
where
we find by Fourier transformation thatand complex conjugation yields the analogous result fornegatively time^-ordered products.
With the aid of the differential equation obeyed byG, we now get
The elements of this matrix are also expressed b;
Our simplified equations can also he applied tosituations io which the external, system is not at thermalequilibrium. To see this possibility let us return to thereal positive functions A,+(ta)t A+~(<a) that describe theexternal system arscl remark that, generally,
and we immediately find a commutator that is amultiple of the unit operator.
B R O W H 1 A H M O T I O N O F & Q U A H T 0 M O S C I L L A T O R 419
These properties can be expressed by writing
where 0(w) is a real eves function that cas range from— «* to 4" °°. When only one value of « is of interest^all conceivable situations for the external system can bedescribed by the single parameter & the reciprocal ofwhich appears as an effective temperature of themacroscopic system. A new physical domain thatappears m this way is characterised by negativetemperatarej $<G. Since & is an iatrmsicaJly positiveconstant, it is y that will reverse sign
aad the effect of the external system on the oscillatorchanges from damping to amplification,
We shall discuss the following physical sequence. Attime Ig the oscillator, in a thermal mixture of states attemperature $#, is acted on by external forces whichare present lor a time, short in comparison with 1/J7J,Altar a sufficiently extended interval "-(/i—fc) suchthat the amplification factor or gain is very large,
measurements are made in the neighborhood of timet\. A prediction of all such measurements is contained IBthe general expectation value formula. Approximationsthat convey the physical situation under considerationare given by
with
and, on defining
-exp[- ««>.,+ (l-«-i«i")->)wJ
which Implies that
From the appearance of the combinations #,f—/*_**/*,Xf.~-)u™X only, we recognize that noaoesmmtitativity ofoperator multiplication Is no longer significant, aadthus the motion of the oscillator has been amplified tothe classical level. To express the consequences mostsimply, we write
Thus, the oscillator coordinate y(l) is the amplifiedsuperposition of two harmonic terms, one of definiteamplitude and phase (signal), the other with random,amplitude and phase (noise), governed by a two-dimensional Gaussian probability distribution.
These considerations with regard to amplification canbe viewed as a primitive model of a ntaser device,6
with the oscillator corresponding to a slogte mode of aresonant electromagnetic cavity, aod the externalsystem to an atomic ensemble wherein, for a selectedpair of levels, the thermal population inequality isreversed by some meaas such as physical separation orelectromagnetic pumping,
m IMFEO?E0 TEEA1MEHT
In this section we seek to remove some of the limita-tions ot the preceding discussion. To aid in dealingsuccessfully with the Ronlocal time behavior of theoscillator, it is conveaient to replace the non-Hermitianoperator description with one employing Hermitlanoperatois. Accordingly^ we begin the developmentagain, now using the Lagrangian operator
where Q has altered its meanirsg by a omstaat factor.One could also include a& external prescribed ferce thatjs coupled to p* We repeat the previous approximateconstructioii of the transformation function (lsifeVfl*JF*which proceeds by the introduction of an effectiveaction operator that retains only the simplest correlationaspects of the external system^ as comprised in
s A sinsilar nwtei has been discussed recently by R. Ser!>erand C. E, TOWMS, Symposium an Qusntum Ele&sfmm (ColumbiaUmwts&y Press, New York, I960).
we obtain the time"-mQepeadent result
and
J . SCH W! M G E R420
The action operator, with no other approximations, is The accompanying boundary conditions are
and the implied equations of motion,, presented assecond-order differentia! equations after eliminating
or> more conveniently expressed,
which replace the nom-Herrmtiaa relations
anc
Note that il is the intrinsic oscillator frequency we thatappears here since the initial condition refers to athermal raisture oi unperturbed oscillator states,
The required solution of the equation lor q~~~g-$,eats be written as
It will be seen that the adjoint operation is equivalentto the interchange of the ±labels.
We define
Implicit is the time /i as one beyond which F~.~F+
equals zero. The initial conditions Cor the secondequation, which this solytiou supplies, are
arm
together witl
and
which enables us to present the integro-differenlialequations as
and
The Green's function that is appropriate lor theequation obeyed by q.++•§-- is defined by
&nd
are
and
B R O W N I A N M O T I O H O F A Q U A N T U M O S C I L L A T O R 4 2 1
and the two real factions are related by The latter obeys
The desired solution of the second differential equationis
and its elements are given by
whereWe note the identifications
It is also seen that the mm of the columns of G isproportional to (*>(£—/)? while the sum of the towscontains only <?«(*--0-
We shall suppose that Gr(t~~^} can have iw> more thanexponential growth, ^g«^~^^ as |—|r--» «?. Thee thecomplex Fourier transformis a real symmetrical mnctioa ol its two arguments.
The differential description of the transformatikmfunction that these solutions imply is indicated by
exists in the upper faalf-plaae
and is given explicitly by
and the result oi integration isHere
'bb can also be displayed m the matrix form sr, since (A~,+-~A^~)(&) is an odd function ol <&}
We have already remarked on the generality of therepresentation
with
and thus we shall write
and
422 J. S C H W I N G E R
axis of fs. As to the negative real axis, GKI")"1 Is araonotordcaliy decreasing function of 1"*= that beginsat -I- * for £= — <» and will therefore have no zero onthe negative real axis if it is still positive at a;=0,The corresponding condition is
which gives
Since IMS is art even function of f, it also representsthe Fourier transform of Ga in the lower half-planeIm£<-«.
If the effective temperature is positive and finiteat all frequencies, $(&?)>0, <?(!*} can have no complexpoles as a function of the variable £*. A complex poleat ft^x+iy, y+0, is a zero of G(f)~l and requires that
tinder these circumstances a—0, for G(%), qua- functionof f!, has no singularity other than the branch line onthe positive real axis, and the f singularities are thereforeconfined entirely to the real axis. This is indicated by
which is impossible since the (jusntity in bracketsexceeds unity. On letting y approach zero, we see thata pole of G(£) can occur at a point £=w'2>Q only ifs(t/)~0. If the external system responds through theoscillator coupling to any impressed frequency, <j(a»)>Gfor all a and no pole can appear on the positive real *nd B(t*>2) is the positive quantity
Some integral relations are easily obtained by of regularity. Accordingly,comparison of asymptotic forms. Thus
The integral relations mentioned previously can beexpressed in terms of these Green's functions. Thus,
while, in the limit of small positive r,
while setting {=0 yields
The Green's functions are recovered on using thiinverse Fourier transformation
which indicates the initial effect of the coupling to theexternal system.
The function S(u*} is bounded, and the Green'sfunctions mast therefore approach sero as 11~~ t1 \ ~~t «s.Accordingly, all reference to the initial oscillator condi-tion and to the time k must eventually disappear.For sufficiently large t—It, f—h, the function »(«— />,where the path of integration is drawn in the half-plane
and
B R O W N I A N M O T I O N OF A Q U A N T U M O S C I L L A T O R 423
with»(»)- («M«">-!)->,
which describes the oscillator in equilibrium at eachfrequency with the external system. When the tempera-ture is frequency independent, this is thermal equilib-rium. Note also that at zero temperature »(w)~0, andG(t~ <%+ is characterized by the temporal outgoingwave boundary condition—positive (negative) fre-quencies (or positive (negative) time difference. Thesituation is similar for G(t~~t'}~~ as a function of £'—t,
It can no longer be maintained that placing ft>=tfremoves all reference to the initial time. An intervalmust elapse before thermal equilibrium is establishedat the common temperature. This can be seen byevaluating the fe derivative of w(<~fe, t'—t^j:
for if this is to vanish, the integrals involving Gn say,must be expressible as linear combinations of Gf(t—t$)and its time derivative, which returns us to the approx-imate treatment of the preceding section, including theapproximate identification of o^ with the electiveoscillator frequency. Hence $s»$ does not representthe initial condition of thermal equilibrium betweenoscillator and external system. While it is perfectlyclear that the latter situation is described by the matrix
G»(<—(*)» a derivation that employs thermal equilib-rium as an initial condition would be desirable,
The required derivation is produced by the device ofcomputing the trace of the transformation function(fe'Ha)^ in which the return path terminates at thedifferent time t,'**f,~T, and the external force F_{»)is zero in the interval between (j and It. The particularsignificance of the trace appears on varying the param-eter \ that measures the coupling between oscillatorand external system:
The operators ft are needed to generate infinitesimaltransformations of the individual states at the corre-sponding times, if these states are defined by physicalquantities that depend upon X, such as the total energy.Th^re is no analogous contribution to the trace,however, for the trace is independent of the representa-tion, which is understood to be defined similarly at t>and tj, and one could use a complete set that does notrefer to X, More generally, we observe that G\{lt)bears the satn« relation to the (VI states as does 6n(<j)to the states at time % and therefore
Accordingly, the construction of an effective actionoperator can proceed as before, with appropriatelymodified ranges of time integration, and, for theexternal system, with
and, therefore,
The corresponding asymptotic form of the matrixG(t~(t, t'—tt) is given by
Ms trace structure implies that
or, since these correlation functions depend only on
Butt1-t2) reduces to
424 J , S C H W I N G E R
WMCB nas seen written tor external torees tftat areaero until the m&m&nt t% has passed.
Perhaps the simplest procedure at this point is toask for the dependence o£ the latter solution upon t$,for fixed T, We find that
time pjiiereiKesi that
ivhicfa if -"'so expressed by
The equations of motion for ^> it &ra given by
on using the relations
These are supplemented by the eciuation lor f_(/) inthe interval from I/ to fe:
Therefore,
smce, with positive time argument;
The utility of this result depends upon the approach ofthe Green's functions to sera with Increasing magnitudeol the time argume&t, which is assured, after making thesubstitution r -* ijff, under the circumstances we haveindicated. Then we can let *,-+ — » and obtain
and the effective bcnmaary condition
The equation lor $-—#$. is selved as before.
wit!whereas
as anticipated.Our results determiae the trace ratio
where H is the Hamiltonian op^aU>r of the completeSystran, and the substitution T~~»-i$ yidds the trans-formation ftiEtction
B R O W H I A N M O T I O N O F A Q U A N T U M O S C I L L A T O R 425
In addition to the trace ratio, which determines thethermal average transformation function {414}sF*with its attendant physical information, it is possibleto compute the trace
which describes the complete energy spectrum andthereby the thermostatic properties of the oscillator inequilibrium with the external system. For tMs purposewe set F±—0 lor i">t% and apply an arbitrary externalforce F-{%) in the interval from £/ to 4. Moreover, thecoupling term between oscillator and external system"m the effective action operator is supplied with thevariable factor X (formerly Xs). Then we have
ana
with
We can also write
where
What is asserted here about expectation values in thepresence of ari external field F($ becomes explicit onwriting
with the accompanying boundary condition
where ?-(£) obeys the equation of motion
and indicating the effect of j±(t} by equivalent time-ordered operators,
Thin
and the properties of $~~{g}$F, which are mdeperideatof PS are given by setting F—0 m the geaeral result.In particular, we recover the matrix identity
and the requirement of periodicity. We em now placeF~_™0 is the differential equation for the tracef andobtain
which is a statement 01 periodicity for the intervalT«l/—Ig, The solution of this equation is
The relation between w and <?.—G? can then bedisplayed as a connection between symmetrical prodactand commutator expectation values
The periodic Green's function. Is given by the Fourierseries
with
where the Green's function obeys
J . S C H W I N G E R
vanishes at infinity in this cat plane. Hence
426
and
where, it is to be recalled,
where the value
•eproduces the pole of G-'EO/t-P) at ?2=0, We also•ecogmse, on relating the two forms,
so that the iategrand has BO singularities at wT~ 2w [ n \.Now we have
that
which, together with the initial conditionThe positive value of the right-hand side as i*? —* 0shows that <f(a) approaches the zero frequency limitingvalue o! ?r from below, and the assumption that a(&s)>0for all w implies
where the lower limit is approached as &; —* .A comparison of asymptotic forms for G~l{$) shows
that
yields
We have already introduced the function
and examined some of its properties for real andpositive jl_j,(w}, A+-(a), This situation is recoveredon makine the substitution T—*i8, and thus
The introduction of the phase derivative can also beperformed directly in the structure &i G~~l({),
the existence oi which for all ft>Q requires thatG^l($) remain positive at every value comprised inj"*— — (2rn/ft)1, which is to say the entire negative faxis including the origin. The condition
and equating the two values for G~"(Q) gives
is thereby identified as a stability criterion. To evaluatethe summation over » most conveniently we shall givean alternative construction for the function lag(G~l(()/—I1), which, as a function of fl, has all its singularitieslocated on the branch line extending from 0 to <* arid
We now have the representation
We shall suppose that the stability requirement iscomfortably satisfied, so that the right-hand side ofthe equation for eot$>(&t} is an appreciable fraction ofwe2 at sufficiently low frequencies. Then tan^u is verysmall at such frequencies, or ?(iu}"v*v and this persistsuntil we teach the immediate neighborhood o! thefrequency &Ji<wt such that
That the function in question, ReG^^-HD), has azero, follows from its positive value at w^O and itsssyraptotic approach to — <# with indefinitely inereasmgfrequency. Under1 the conditions we have described.,with the major contribution to the integral comingfrom high frequencies^ the gero point is given approxi-mately as
ind somewhat more accurately byThe second factor can be ascribed to the oscillator,with its properties modified by interaction with theexternal system, The average energy of the oscillatorat temperature $~0rl is therefore gives by
B R O W N 1 A N M O T I O N O F A Q U A N T U M O S C I L L A T O R 427
aaa the summation tormula derived worn the pnxtuciform of the hyperbolic sine function,
gives us the desired result
in which the temperature dependence of the phase^(w) is not to be overlooked. In an extreme faigh-temperatyre limit, such that &i$<sCl for all significantfrecitiencks, we have
As we shall see, B Is less than, unity, but only slightlyso under the circumstances assumed,
In the neighborhood of the frequency w(, the equationthat determines ^(<y) can be approximated by
where
with the definitionand the simple classical result £—$ appears when{Q% is proportional to $„ The oscillator energy alxero temperature is given by
Hence, as w rises through the frequency &»i} ^ decreasesabruptly from a value close to w to one near zero. Thesubsequent variations of the phase are comparativelygradual, and <p eventually approaches aero as <y —* *>.A simple evaluation of the average oscillator energycan be given when the frequency range «>wi overwhich e(w) is appreciable in magnitude is such thatj3w^j, Ttiere will be n& significant temperature varia-tion in the latter domain a&d in particular &n shouldbe essentially temperature independent Thes, since— (I/-$}(&%>/$&} in the neighborhood of on closelyre^mbles ft(w—wi), we haw approxiisately
and the osctllaior coEtrtbution to the specific heatvanishes.
The following physical situation has consequencesthat resemble the simple model of the previous section.For values o? to<w0J a(«) tanh(|t^)4Cw«Si and *(w)differs significantly from zero until one attains fre-quencies that are large in comparison with &j». Themagnitudes that a{&?) can assume at frequencies greaterthan &}g is limited only by the assumed absence of rapidvariations and by the requirement of stability. Thelatter is generally assured if
which describes a simple oscillator of frequency wj,with a displaced origin of eaergy,
Note that with $>(«} wry small at a frequencyslightly greater than w and zero at infinite frequency,eye have
428
Related integrals are
J . SCH W I N G E R
The farther concentration on the immediate vicinityof &fi, 1 &f—^i j -^^ gives
and
The latter result confirms that B< i. A somewhat moreaccurate formula lor ,8 fa
If the major contributions to all these integrals comefrom the general vicinity of a frequency «»««, wecan mate toe crude estimates
which clearly identifies B<1 with the contribution tothe integral J"AaiJS(in'') that comes from the vicinity ofthis resonance of width t at frequency «i, although thesame result is obtained without the last approximation.The remainder of the integral, t—B, arises from fre-quencies considerably higher than an according to ourassumptions.
There is a similar decomposition of the expressionsfor the Green's functions. Thus, with Of,
The second high-frequency term will decrease veryquickly on the time scale set by I/MI,. Accordingly, inusing this Green's function, say in the evaluation of
Then neither the energy shift nor the deviation of thefactor B from unity are particularly significant effects.
The approximation of Re<3{aj-HQ} as B™1^!*"™^}evidently holds from zero frequency up to a frequencyconsiderably in excess of &n. Throughout this frequencyrartere we }IK\TP
or
wito
If in particular /9»i<Cl, the frequencies under considera-tion are in the classical domain and 7 is the frequencyindependent constant
for an external force that does not vary rapidly inrelation to &si, the contribution of the high-frequencyterm is essentially given by
To regard j as constant for a quantum oscillator requiresa suitable frequency restriction to the vicinity of on,The function B(ws) can be computed from
and the response to such an external force is adequatelydescribed by the low-frequency part of the Green'sfunction. We can represent this situation by an equi-valent differential equation
which needs no further qualification when the oscilla*lions are classical but implies a restriction to a frequencyinterval within which y is constant, for quantum oscilla-tions. We note the reduction in the effectiveness of theexternal force by the factor B. Under the circumstances
and accordingly is given by
But
B R O W N I A N M O T I O N O F A O U A H T U M O S C I L L A T O R 429
mtlined this effect is not important an4 we shall place5 equal to unity.
One can ntafee a general replacement of the Green'sunctions by their low "frequency parts:
which here asserts that
if one limits the time focahzability ©f measurements sothat only time averages of q{t) are of physical Interest.This is represented in the expectation value formula byconsidering only functions f±{t) that do not vary tooaukklv. The corresmwirfin? replacement iotwit—i*} is
with
and the entire matrix G&(t—tf) obtained in this wayofsevs the. differential ^filiation
The latter matrix can also be identified as
where s^o^Wi),The simplest presentation of results is again to be
found m the Laogevin. viewpoint, which directs theemphasis fom the coordinate operator g(t) to thefluctuating force defined by
and
In the classical limit
If a comparison is made with the similar results o! theprevious section it cars be appreciated that the f reqtsencyrange has been extended and the jrestriction «tscs&j»removed.
We return from these extended considerations onthermal equilibrium and consider am extreme exampleof negative temperature for the external system, itiisis described by
which Is to say
This change is introduced by the substitution
and the necessary partial integrations involve thepreviously established Jemima oa tiffie-ordered operatorsj
With the definition
and
430
we have
J . S C H W I N G E R
sfithin the range dg i, 4g% Is
where g» and «? ate the amplitude and phase of f,(i).Despite rather different assumptions about the externalsystem, these are the same conclusions as before, apartfrom a factor of f in the formula (or the gain.
6IHERAL THE0M
The whole of the preceding discussion assumes anexternal system that is only slightly influenced by thepresence of the oscillator. Now we must attempt toplace this simplification within the framework ol ageneral formulation, A more thorough treatment isalso a practical necessity in situations such as thoseproducing amplification of the oscillator motion, for asizeable reaction in the external system must eventuallyappear, unless a counter mechanism is provided,
It is useful to supplement the previous Lagrangianoperator with the term f'('X?, in which q'{i) is anarbitrary numerical function of time, and also, toimagine the coupling term gQ supplied with a variablefactor X. Then
We shall suppose, for simplicity, that wt^wa andsfXsat. Then the poles of
Its a function of £*t G(£) now has complex poles if
After the larger time intervals j»(«—i,}, f(l'—%)»!,we have
with
When I is in the vicinity of a time li, such that th<mptification factor
the oscillator is described by the classical coordins provided that the states to which the transformationfunction refers do not depend upon the couplingbetween the systems, or that the trace of the trans-formation functisn is being evaluated. A similarstatement would apply to a transformation fusctioiswith different terminal times. This differential equationimplies an integrated form, in which the transformationfunction for the fully coupled system (X= 1) is expressedin terms of the transformation function for the un-coupled system (X=0), The tatter is the product oftransformation functions for the independent escalatorand external system. The relation is
where g-t and ^ are characterised by the expectationvalue formula
Here
and
in which
AceordJKgly, the probability of observing ft and j»
are located at andAcoordingly is regulat outside a stirip of wich
The associated Greeen's Futcation are given by
and the function compoued fior
431
On moving P&($ to the left ol the exponential, thisa Such positive and negative time-ordered products occur in a
recent paper bv it. Svnmnak Q, Math. Phys* 1, 249 (1960)1which appeared after this paper had beem written and its contentsssed as a basis for lectures delivered at the Brandos SummerSefeaol, July, I960.
B E O W N I A H M O T I O N O F A Q U A N T U M O S C I L L A T O R
and we have indicated that q^ is finally set equal tozero If we are eonceraed OBly with nseasarements onthe oscillator.
Let us consider for the moment just the exteraalsystem with the perturbation gfQ, the effect 0! which Isindicated by8
become
Bui
We shall define
and similarlywhich leads us to the following functional differentialequation lor the transformation function (lali^*, inwhich a knowledge is assumed of the external systemsreaction to the perturbation ?±'(Q:
and farthermore,
When q±(l)**tf(t)t we haye
which is the expectation value ol Q{$ in the presence ofthe perturbation described by ?'(^). This is assumed tobe aero for §'00—0 and depends generally upon thehistory of /(/} between % and the given time.
The operators §&(!} are produced withlm the trans-formation function by the functional differentialoperators (±1/0 V^±(*)» &n^ ^nce the equation ofmotion for the uncoupled oscillator is
Throughout this discussion, one must distinguishbetween the ± signs attached to particular compoHeaf saud those involved in the listing of complete sets ofvariables.
The differential equations for time development aresupplemented by boundary conditions which assert, ata time h beyond which F^(I)~F_(0, that
we have
fhe previous treatment caa sow be identified as the
while, for the exai^ple of the transformation function(4f4Ve^*» ^e have the initial conditions
432 J . S C H W I M G E R
approximation of the Q±(ttq±) oy linear functions of converts the functional differential equations into
%&>
wherein the linear equations tor the operators q±(t) anatheir meaning in terms of variations of the P& havebeen united in one pair of functional differentialenuafJons. This relation becomes clearer if one writes
Tne bouiitoaiy <3>uaitiOBS EOW appear as
When the Q$, are linear functions of g±( the functionaldifferential operators disappear7 amcl we regam thelinear eolations for q±(t)t which in turn imply thequadratic form of $T{F&) that characterizes thepreceding discussion,
TThe degeneration of th« lunctioriml ^jtiation? iato or^nafyEliffeiwitial equations a^e» oeeurs when the RHJ!ion of the oscillatoris classical aad free of Iuct«atioR.
and, with the defiuition
and
374 QUANTUM KINEMATICS AND DYNAMICS
9.2 COULOMB GREEN'S FUNCTION
tReproduced from the Journal of Mathematical Physics,
Vol. 5, pp. 1606-1608 (1964).
+
Reprinted liom
JOUKKAL OF M A T H E M A T I C A L PHYSICS
Printed in VSA.
N O V E M B E R I 9 6 <
r has long been known that the degenwasy ofthe bound states in the nonrektrristio Coulomb
problem can be described by a four-dimensionalEuclidean rotation group, and that the momentumrepresentation is most eonvenient for realizing theconnection. It seems not to have been recofpiraed,however, that the same approach cat* b§ used toobtain an exploit construction for the Green'sfunction of this problem. The derivation1 is givenhere.
The momentum representation equation for the* It was wolfed cm$ to preseat at a Harvard quaatum
mechanics eourae give® in ttte late 1940*8. I h&ve beenstimulated to rescue it from tfa® quiet d«ath oj lectee aotesby recent publications is this Jouro&l, which give alternativeforms of the Green's faaetwm: B. H. Wielimaaa aad C. H.Woo, J. Mutt. Phys. Z, 178 (1W1): JU Ho«tter, iW. S, S91(J964).
Coulomb Green's Function
JyiaAN ScHwxKasitffarmrd £/nwers%, Cambridge, MasmekmeM*
(Beceirad 19 June 1964)
A one-parameter istegml Eeprestutation is given for tlis momestucft 8[>a«e Green's fuoctioahe nasrelativistie Coulomb problem.
VOLOKB 5, N U M B E R 11
Sreen's function a (ft = 1)
We shall solve this equation by assuming, at first,that
ia real and negative. The general result is inferredby analytic continuation.
The parameters
define the surface of a unit four-dimensional By-
r
COBLOMB GEBBN'S FUNCTION 1607
elidean sphere,
the points of which are in, one to one eorrespondencwith the momentum spaee. The element of areon the sphere is
if one keeps JB mind that p ^. Pa corresponds to thetwo semispheres Is » ^(1 — £*)** As another form ofthis relation, we write the delta function connect-ing two points on the uait sphere as
whei
The Green's function D is exhibited as
conveys the oorroaliwtion and completeness of thesurface harmonies. One can verify that D has theradial discontinuity implied by the delta functioninhomogeneity of the differential equation,
The function D is used in the integral equation forF with p « = / » ' • » 1. The equation is solved byNext, observe thi
The singularities of this function at f-n —1,2, • • •give the expected negative energy eigenvalues.The residues of Q at the corresponding poles inthe B plane provide the normalized wavefune-tions, whieh are
Then, if we define
that function obeys a four-dimensional auelideaEsurface integral equation,
One can exhibit r(tt, 0') in essentially closedform with the end of the expansion for D. We usethe following version of this expansion:
where
and
The function D that is defined similarly through-out the Euclidean space is the Green's funetiocof the four-dimensional Poisson equation,
where { and £' are of unit length and 0 < p < I.Note, incidentally, that if we set { ** J* and inte-grate over the unit sphere, of area 2ir3, we get
It can be constructed in terms of a eomplete setof four-dimensional solid harmonics. In the sphericalcoordinates indicated by />, Q, these are where m* is the multiplicity of the quantum number
n. This eoRfirms that m* *= ns,The identitywhere the quantum numbers J, m provide a three-
dimensional harmonic classification of the four-dimensional harmonics. The largest vslue of i con-tained in the EQisogeneous polynomial fli>~tY*tm($l)is the degree of the polynomial, n — 1. Thus,
together with the integral representation
label the n distinct harmonies that have a commonvalue of n. ?alid for » < 1, gives
1608 J U t l A N S C B W I N G E R
anil
One would have found the same asymptotic formfor any potential that decreases more rapidly thaathe Coulomb potential at large distances, but with<J*(p) = (g - I*)"*. The factors GV) and G°(p)describe the propagation of the particle before andafter the collision, respectively, and / is identifiedas the scattering amplitude. The same interprets-"tion is applicable here since the modified 6° justincorporates the long-range effect of the Coulombpotential. This is most evident from the asymptoticbehavior of the corresponding spatial function,which is a distorted spherical wave.
The scattering amplitude obtained in this waycoincides with the known result,
The path C begme »t p — 1 + Qz, where the phaseof p is Bero and terminates at p « 1 — Gi, afterencircling the origin within the unit circle,
The Green's function expressions implied by (1),f2>. and C31 are
and
where
The restrietion p < 1 ean be removed by re-placing the real integrals with contour integrals,
The $eeond of the three forms given for G is mostconvenient here. The asymptotic behavior is domi-nated by small f values, and one immediatelyobtains
Equivalent forms, produced by partial integra-tions, are
U>
The Green's function is regular everywhere inthe complex E plane with the exception of thephysical energy spectrum. This consists of thenegative-energy eigenvalues already identified andthe positive-energy eontitmam. The integral repre-sentations (!'), ('2'), &sd (30 are not completelygeneral sinee it is required that
C2)As we have indicated, this restriction ean be re-moved. It is not necessary to do 80, however, ifone is interested in the limit of real k. Tliese repre-sentations can therefore be applied directly to thephysical scattering problem.
The asymptotie conditions that characterise finiteangle deflections are
am
which uses the limiting relatio
Note that i is a function of a single variable,
where