Heisenberg's Uncertainties and the Probabilistic Interpretation

of Wave Mechanics
with Critical Notes o/the Author
Fundamental Theories of Physics

An International Book Series on The Fundamental Theories of Physics:

Their Clarification, Development and Application

Editor: ALWYN VANDER MERWE

University of Denver, U.S.A.

Editorial Advisory Board:

AS 1M BARUT, University of Colorado, U.S.A.
HERMANN BONDI, University of Cambridge, U.K.
BRIAN D. JOSEPHSON, University of Cambridge, U.K.
CLIVE KILMISTER, University of London, U.K.
GUNTER LUDWIG, Philipps-Universitiit, Marburg, F.R.C.
NATHAN ROSEN, Israel Institute of Technology, Israel
MENDEL SACHS, State University of New York at Buffalo, U.SA.
ABDUS SALAM, International Centre for Theoretical Physics, Trieste, Italy
HANS-JURGEN TREDER, Zentralinstitutfur Astrophysik der Akademie der
Wissenschaften, CD.R.

Volume 40
Heisenberg's Uncertainties
and the Probabilistic Interpretation
of Wave Mechanics
with Critical Notes o/the Author

by

Louis de Broglie
Honorary Permanent Secretary 0/ the Academie des Sciences, Nobel Laureate

With a Foreword by
Asim Barnt
University o/Colorado,
Boulder, U.S.A.

Preface and Supplementary Notes by

Georges Lochak
Director 0/ the Fondation Louis de Broglie

Translated by
Alwyn van der Merwe
University 0/ Denver,
Denver, Colorado, U.S.A.

FONDATION

23 QUAI DE CONTI. 75006 PARIS

KLUWER ACADEMIC PUBLISHERS

DORDRECHT I BOSTON I LONDON
Library of Congress Cataloging in Publication Data
Broglie, Louis de, 1892-
(Incertitudes d'Heisenberg et l' interpretation probabi 1 iste de la
mecanique ondulatoire, Engl ish]
Heisenberg's uncertainties and the probabil istic interpretatlon of
wave mechanics with critical notes of the author I by Louis de
Brog1 ie ; with a foreword by Asim Barut ; preface and supplementary
notes by Georges Lochak ; translated by Alwyn van der Merwe.
p. em. -- (Fundamental theories of physics; v. 40.)
Translation of, Les incertitudes d'Heisenberg et l' interpretation
probabiliste de la mecanique ondulatoire.
Includes bibl iographic"l references and index.
ISBN-13: 978-94-010-7457-5 e-ISBN-13:978-94-009-2127-6
DOl: 10.1007/978-94-009-2127-6

1. Wave mechanics. 2. Heisenberg uncertainty principle.

1. Series.
OC174.2.B653513 1990
530.1 '24--dc20 90-48032

ISBN-J3: 978-94-010-7457-5

Published by Kluwer Academic Publishers,

P.O. Box 17, 3300 AA Dordrecht, The Netherlands.

Kluwer Academic Publishers incorporates

the publishing programmes of
D. Reidel, Martinus Nijhoff, Dr W. Junk and MTP Press.

Sold and distributed in the U.S.A. and Canada

by Kluwer Academic Publishers,
101 Philip Drive, Norwell, MA 02061, U.S.A.

In all other countries, sold and distributed

by Kluwer Academic Publishers Group,
P.O. Box 322, 3300 AH Dordrecht, The Netherlands.

Printed on acid-free paper

All Rights Reserved

© 1990 by Kluwer Academic Publishers
Softcover reprint of the hardcover 1st edition 1990

No part of the material protected by this copyright notice may be reproduced or

utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
 TABLE OF CONTENTS

Preface by Asim O. Barut Xl

Foreword by Louis de Broglie XV
The Problem of Hidden Determinism XIX
Preface by Georges Lochak XXlll
On the True Ideas Underlying Wave Mechanics xli

PART ONE (1950-1951)

ON HEISENBERG'S UNCERTAINTIES
AND THE PROBABILISTIC INTERPRETATION
OF WAVE MECHANICS

Chapter 1. Principles of Wave Mechanics

1. Classical Mechanics of the Point Mass. Theory of Jacobi 3
2. Wave Propagation in an Isotropic Medium 7
3. Transition from Classical Mechanics to Wave Mechanics 10
4. General Equation for the Wave Mechanics of a Point Mass 12
5. Automatic Procedure for Obtaining the Wave Equation 13

Chapter 2. Probabilistic Interpretation of Wave Mechanics

1. Interpretation of the 'ljJ Wave 15
2. Principle of Interference 16
3. Precise Statement of the Principle of Interference. Probability Fluid 17
4. The Uncertainty Relations of Heisenberg 18
5. The Principle of Spectral Decomposition (Born) 21
6. New Ideas Resulting from the Preceding Conceptions 22
7. Return of Wave Mechanics to Classical Mechanics.
The Theorem of Ehrenfest. Group Velocity 23

Chapter 9. Wave Mechanics of Systems of Particles

1. Old Dynamics of Systems of Point Masses 28
2. Wave Mechanics of Systems of Particles 30
3. Interpretation of Wave Mechanics for a System of Particles 31
vi Table of Contents

Chapter 4. General Formalism of Wave Mechanics

1. New Conception of the Quantities Attached to a Particle
(or to a System) 34
2. Eigenvalues and Eigenfunctions of a Linear Hermitian Operator 36
3. Continuous Spectrum of the Free-Particle Hamiltonian.
Dirac's Delta Function 39

Chapter 5. General Principles of the Probabilistic Interpretation

of Wave Mechanics
1. General Ideas 42
2. The Algebraic Matrices and Their Properties 46
3. Operators and Matrices in Wave Mechanics 48
4. Mean Values and Dispersions in Wave Mechanics 49
5. First Integrals in Wave Mechanics 51
6. Angular Momentum in Wave Mechanics 53

Chapter 6. Theory of the Commutation of Operators

in Wave Mechanics
1. General Theorems 55
2. Corollaries of the Foregoing Theorems 60
3. Simultaneous Measurement of Two Quantities in Wave Mechanics 63
4. Examples of Quantities That Are Not Simultaneously Measurable.
Distinction Between Two Kinds of Non-Commutation 66

Chapter 7. Physical Impossibility of Simultaneously Measuring

Canonically Conjugate Quantities
1. Necessity to Examine the Impossibility of Simultaneously and
Precisely Measuring Two Canonically Conjugate Quantities 68
2. The Heisenberg Microscope 69
3. Measurement of the Speed of an Electron by Means of
the Doppler Effect 71
4. Passage of a Particle through a Rectangular Aperture 73
5. Important Remark on the Measurement of Speed 76
6. The Case of Two Operators with a Nonzero Commutator 77
7. Bohr's Complementarity 79
8. Bohr's Calculation for Young's Slits 81

Chapter 8. Precise Form of the Uncertainty Relations

1. Theorem About the Dispersion of Non-Commuting Quantities 85
2. Optimal Nature of the Gaussian Wave Packet 90
 Table of Contents Vll

3. Comparison of the Theorem on Dispersions with the Qualitative

Uncertainty Relations of Heisenberg (Pauli and Robertson) 94
4. Various Considerations about Uncertainties.
Sharp-edged Uncertainties 95

Chapter 9. Heisenberg's Fourth Uncertainty Relation

1. The Absence of Symmetry between Space and Time
in Wave Mechanics 103
2. Correct Formulation of the Fourth Uncertainty Relation 105
3. Illustration of the Preceding Definition 105
4. Various Remarks about the Fourth Uncertainty Relation 108
5. Method of Variation of Constants. Transition Probability 110
6. Transition Probabilities 114
7. Uncertainty Relations and Relativity Theory 117
8. Formulas of Mandelstam and Tamm 121

Chapter 10. Examination of Some Difficult Points

in Wave Mechanics
1. Reduction of the Probability Packet by Measurement 124
2. Impossibility of Discovering the Anterior State of a Measurement
from its Posterior State. Effacement of Phases by Measurement 125
3. Possibility of Discovering the Past, Starting from a Measurement
Made at a Given Instant (Postdiction) 127
4. Interference of Probabilities 129
5. Some Consequences of the Disappearance of the Trajectory Concept 132
6. Discussions Concerning "Correlated" Systems 136
7. Complementary Remarks on the Einstein-Bohr Controversy 143

PART TWO (1951-1952)

ON THE PROBABILISTIC INTERPRETATION

OF WAVE MECHANICS
AND VARIOUS RELATED QUESTIONS

Chapter 11. Summary of Some General Concepts

of Probability Calculus
1. Probability Laws for One Variable. Distribution Function 153
2. Probability Laws for Two Variables 159
3. Very Important Remark Concerning the Foregoing Results 166
viii Table of Contents

Chapter 12. Recalling the General Concepts of Wave Mechanics

1. The Interference Principle. Theory of the Pilot Wave 173

Chapter 13. Introduction of the Characteristic Function into

the Probability Formalism of Wave Mechanics
1. Characteristic Function for a Single Quantity 180
2. Characteristic Function for Two Commuting Quantities 186
3. Correlation Coefficient. Marginal Laws 187
4. General Theorems of Wave Mechanics Considered from
the Characteristic-Function Point of View 190
5. Case of Two Noncommuting Quantities 197
(a) Reminder About the Reduction of the Probability Packet by
Measurement 197
(b) Interference of Probabilities 198
(c) The Distribution Function p( x, Px) of Wigner and Bass 201
(d) The Wigner-Bass Density p(x,Px) and the Hydrodynamic
Interpretation of Wave Mechanics 207
(e) Yvon's Theory 213

Chapter 14. Theory of Mixtures and von Neumann's Theory

of Measurement
1. Mixtures and Pure Cases 220
2. Von Neumann's Statistical Matrix for a Pure Case 224
3. The Statistical Matrix for a Mixture of Pure Cases 227
4. Irreducibility of Pure Cases 230
5. Impossibility of Reducing the Laws of Wave Mechanics
to a Hidden Determinism (von Neumann) 232

Chapter 15. Measurement Theory in Wave Mechanics

1. Generalities 238
2. Statistics of Two Interacting Systems 239
3. Correlation Coefficients in the Interaction between Two
Quantum Systems 243
4. Measurement of a Quantity 244
5. Example of a Measuring Experiment 249
6. Diverse Remarks on Measurement 253
7. Thermodynamical Considerations (von Neumann) 259
8. Reversible and Irreversible Evolutions 263
9. The Statistical Matrix Po 266
 Table of Contents ix

Chapter 16. The Role of Time in Wave Mechanics

1. Retrodiction According to Costa de Beauregard 268
2. Special Role of Time in Quantum Mechanics.
The Fourth Uncertainty Relation 273
3. Correct Statement of the Fourth Uncertainty Relation 275
4. The Fourth Uncertainty Relation and Perturbation Theory 276
5. The Operators H and (h/21ri)8/8t 279
6. Application of the Formalism of Arnous to the Operators
Acting on Time 280
7. MuItitemporal Formalism. Multiplicity Curves in Spacetime 283

Works of Louis de Broglie 285

Index 299
 PREFACE

LOUIS DE BROGLIE
AND
THE SINGLE QUANTUM PARTICLE

By A. O. Barut

We have abundant evidence and testimony that Louis de Broglie deeply cared
about the foundations, the meaning, and our understanding of quantum theory
in general and of wave mechanics in particular. So, too, did Erwin Schrodinger,
along with Einstein, Bohr, Dirac, and Heisenberg. For de Broglie and Schrodinger
this preoccupation meant not simply the acceptance of a novel set of rules, but a
constant struggle and a search for complete clarity about the way in which the new
theory fits into the great classical traditions of an objective physical world view.
We may call this a striving for "physical rigor," rigor in reasoning, or intellectual
rigor. There is not only mathematical rigor inside an axiomatic system with which
everybody agrees, but there is, and there should be, rigor also in our concepts and
methods. To this kind of rigor belongs the unity, the economy and simplicity, and
the consistency of physical theories; naturally along with as complete and as clear an
understanding of phenomena as possible. No loose ends, no proliferation of poorly
tested and phenomenological entities, no bending of logic and compromise, and no
handwaiving arguments can be tolerated. Unfortunately this kind of rigor seems
to be missing in today's forefront of fundamental physical theories, viz., particle
or high-energy physics. This branch of physics should go hand in hand with, in
fact ought to be the testing ground for, the elucidation of the two most important
theories of this century, relativity and quantum theory. Instead, particle physics
seems to have separated itself from the rest of physics, developed its own jargon
and logic, and does not appear to worry about its foundations. This turn of events
has its origin, I believe, in the rapid acceptance and success of the rules of quantum
theory without an accompanying full understanding of these rules. Which brings us
back to Louis de Broglie.
After the rules of quantum mechanics with Born's statistical interpretation
became mathematically axiomatized by J. von Neumann, it was no longer fash-
ionable to question the interpretation of quantum theory. My late colleague
Edward Condon used to tell the story of Bergen Davis, a distinguished older
Columbia University professor in the twenties, who, after long wrestling with quan-
tum mechanics, said in despair, "I don't believe you younger fellows understand
Xll Preface

it any better than I do, but you all stick together and repeat the same thing."
Condon called this attitude the conspiracy interpretation of quantum theory. It
takes courage, when the majority of physicists repeat the same thing, taking refuge
under the safety-in-numbers syndrome, for someone to say "I don't understand it."
Louis de Broglie, Schrodinger, Einstein, and later Dirac displayed such courage,
and paid the price-finding themselves isolated at the periphery of the so-called
"establishment." Having left quantum mechanics incomplete, we embarked upon
quantum-electrodynamics with the extension of quantum rules from particles to
fields. QED itself was left incomplete because of renormalization and convergence
problems. We then embarked on electroweak theory and quantum-chromodynamics
and the standard model, all of which are incomplete and appearing more and more
phenomenological and awkward. The tragedy of sinking deeper and deeper into
a quicksand formed of poorly justified assumptions, or building towers on card-
board foundations, is well known. "Human reason so delights in constructions that
it has several times built a tower and then razed it to examine the nature of its
foundations" wrote I. Kant in his Prolegomena. The number of untested building
blocks in a theory must not be allowed to grow too large, otherwise the building
becomes insecure. In modern particle physics, by contrast, the major components
of the theory are hypothetical or unobservable. I think physics should progress
by carefully and completely testing one hypothesis after another. Kant also knew
that "he who raises doubts must expect opposition on all sides." Thus, twice in
the history of modern physics, complacency caused physics to stand still, once in
the quantum theory of measurement and then again in the theory of renormal-
ization. Although people felt that something was terribly wrong, they held the
rules as sacred and violently opposed those who wanted to change them. For-
tunately there arc always some minds that strive for rigor in reasoning, and one
cannot by rules for long place limitations on human knowledge or decree that cer-
tain questions should not be asked. That is why Louis de Broglie is important even
today.
The central question in de Broglie's mind, from the beginning to the end, was
the nature of a single quantum particle. Precisely this was the theme of his eloquent,
succinct, and beautiful address in 1973 on the occasion of the fiftieth anniversary of
the discovery of wave mechanics,1 which is reproduced in the present volume, "One
should return to the idea" he says-and so he did himself after about twenty years
of hesitation-"that a particle is a very small object that is localized and moves
along a trajectory." And at the beginning, in his Thesis, he "tried to imagine a real
physical wave which transported minute and localized objects through space in the
course of time." He assumed that the particle possesses an internal vibration, and
hence does not trace a line trajectory. This is a crucial property, as we shall see. By
contrast, Max Born2 concentrated his attention on repeated events, on scattering
experiments, and on the regularities in such experiments. He was well aware of the
difference between single events and the probable or typical behavior of a single
particle in repeated experiments. He claimed that there were no internal properties
of atomic systems which would fix and determine a definite result of any collision

1. L. de Broglie, C. R. Acad. Sci. B 277, 71 (1973).

2. M. I3orn, Z. Phys. 37, 863 (1926); 38,803 (1926).
 Louis de Broglie and the Single Quantum Particle xiii

in a single event. De Broglie observes that "Schrodinger viewed his t/J wave as a
physical wave. But he abandoned completely any idea of localizing the particle in
this wave, so that the picture which he formed of the atom and, more generally,
of the t/J waves, made no provision for localized particles." Born's application of
Schrodinger's mechanics to collision problems "gives a very definite answer, but it
is not a causal answer. You don't get an answer to the question 'What is the state
after the collision?' but to the question 'What is the probability of a given effect
of the collision?'" According to de Broglie, Born, by introducing arbitrarily the
normalization of the amplitude of the wave, deprived it of all physical reality (as
an individual event). Born, however, left a small door open: "But it is of course
allowed to someone, who is not satisfied with this, to assume that there are ad-
ditional parameters that are not in the theory which would determine the single
event."
It is true that the equation, e.g., Schrodinger's equation, has no additional pa-
rameters, but these can occur in the solutions. For localized solutions there are
parameters that specify the location of the center of the solution, its shape pa-
rameters, and its velocity. We are not talking about wave packets that spread the
more the more localized they are, but about genuine non spreading localized so-
lutions with internal oscillations. Such solutions have no place in the axiomatic
formulation of standard quantum mechanics the way wave packets have. In fact,
the axiomatics is so constructed that it is impossible to ask questions about sin4
gle individual events. But de Broglie's vision was prophetic: Single events can
be observed individually, and localized nonspreading solutions of the wave equa-
tion that he pioneered3 can be constructed in their full generality. More impor-
tantly, these solutions lead to quantities of energy and momentum, calculated as
integrals over the localization space of the solution, which are proportional to the
frequency and the wave vector, respectively, of the oscillating lump.4 In regard
to energy and momentum, de Broglie writes: "But contrary to what is usually
admitted, (standard) quantum mechanics does not have the right to postulate
W = hv and p = hi A, because the energy Wand the momentum p of a parti-
cle are properties which are associated with the concept of a localized object that
moves through space along a trajectory. The reason that I was able to establish
these formulas was that I advanced the hypothesis of a particle localized inside a
wave."
The localized solutions referred to above explicitly realize other propositions of
de Broglie: the concept of "mass" as an expression of the internal oscillations, the
Lorentz transformation of the rest-frame solution to a moving frame, hence the
transformation of internal frequency and rest mass, and the relation between the
phase velocity u and the group velocity v, viz., uv = c2 ; all these results would not be
possible without the hypothesis of internal oscillations. It is not necessary to imagine

3. 1. de Broglie, C. R. Acad. Sci. B 180, 498 (1925); Ann. Phys. (Paris) 3, 22

(1925).
4. A. O. Barut, Phys. Lett. A 143, 349 (1990); A. O. Barut and A. Grant,
Found. Phys. Lett. 3, 303 (1990); A. O. Barut, Found. Phys. 20 (10), forthcoming
(1990).
XIV Preface

a particle guided or surrounded by a wave; a localized oscillating nonspreading

solution is a quantum particle. It is not difficult to see that a major part of the
confusion in discussions about quantum theory is due to the fact that we do not
clearly distinguish between Born's Ill, describing the typical behavior of a particle
in repeated experiments, and de Broglie's 1/;, describing an individual single event.
The former includes no "hidden" parameters, the latter does: the parameters of
localization. They answer to different types of experiments. When we try to apply
Born's III to a single event, we are forced to invoke the very unnatural notion of the
"collapse of the wave function" and other similar inconsistencies. To avoid all of this,
I proposed some time ago [A. O. Barut, Found. Phys. Lett. 1,47 (1988)] using two
different wave functions: Ill, for probability amplitudes, and 1/;(x, a), depending on
the localization parameters a, for an individual event. These functions are indeed
quite different. The incoherent sum of all individual 1/;(x, a) over the parameters
exhibits the regularities, the interference phenomena, easily accounted for by the
wave amplitudes Ill, precisely as an interference experiment evolves when one records
one event after another and collects the results in a memory.
In his preface, Georges Lochak recounts in very moving words the background,
the birth, and the historical significance of this book, which marked a turning point
for de Broglie, when he started to revert to his original ideas, which pictured a
quantum particle as an objective, localized, oscillating object. Lochak also predicts
that this late period of de Broglie, like the late work of a Michelangelo, might carry
the seed for future developments. Indeed, I wanted above to expand first on Louis
de Broglie's sentiment that we should look to the completion of quantum theory for
the solution of some of our present problems in particle and nuclear physics, and
then to outline that his vision for a theory of an objective, single quantum particle
is alive and in the process of being further developed and realized.

ACKNOWLEDGEMENT
The translator of this book gratefully acknowledges financial support by the Re-
search and Creative Work Fund of the University of Denver. He is also indebted to
Professor Stanley Gudder for his careful reading of the final manuscript

Camera-ready text for this book was prepared by Paula Spiegel Gudder, using
'lEX with Textures.
'lEX is a trademark of the American Mathematical Society.
Textures is a trademark of Blue Sky Research.
 FOREWORD

(The text placed here as a foreword was written by the author for the Annales de
la Fondation Louis de Broglie with a view to the commemoration of the Einstein
centenary. )
xvi The Need for Freedom in Scientific Research

.2:1J?'~~ko ,,1$> J,'",,,,, 41'1'~ ~... ;dt•.!rk d(>1o<A/~ )

;;, fk p<>w4 i'4'J';'".,.J II'I ~pf,;4IAIJ ,(14!JA<J-"
l' I , ... r "" 4 1"1 A V(/.~ ", d"y,,~£..r/i("",..,I~
a...,.r'",c." It-

f'u/:....r"", ., ',.fe-,#iP""--Vr ..j,' .?, (~/I'" ~H"".......{;

ole ;-du/f ,)",' ,..v Iff~ ;fo-'/J~I? A 4 b.,'h<u
Mo;Ia~ A'l/t1I"~ ~(I.J~lA'~~ A,bj 4J~,fU/~"'2A? 4
411:'V,~ ~.I1O / _ _ 2M-V" .0_ ".../krL~
~~Vif-ue; ~IfWlIIk'• ....rHe '~ h1A.I'>1~ """
~,"....t en k '~'J.'''dt.·rZe.r;R k ,p~d.-A' £-t.. .
I'" if> 4> .;i..ITil .6 6..M1t" II"'''' e,~1 IN/.. A-.t of
S)

-:fw<' tfM~ ~.;;..?~ ;; .&. f~"f"'l CJ')tf;;i..-? "I.. ~

1l0 ..... ''11 "'~ 1"'~./-.{iJ; .( ! ;'--f ',!...,"" ",6
i lU 4<1 r-f,

flvll'",~b; l.-fT)A 61$~ ..,tl/i~ Ayle j)~,~ tl.v4' #,h

.!{;;,'e1l1#J1 . ol#'J -v~,,~~ ~tn,,>.d 1141f1d(/~ ~~ .

o /
(jli/JtJ,/jW ::lti'>1j'~ d~(7d~1hlJ 1,1l'JLl.-i.
.i .I ~'1 LA) 11""1"'/~AI1c.t
I

cYk fI.ttt0e I AI;. ~I't(-Ir)r~f C"dI(iJTM "~"'V'(J~~ N t~ I~

 The Need for Freedom in Scientific Research xvii

e"
~~J.. ~t~It.Vp/ ~'14V;;'/h -d4> j,ulj '.IJr~ a.
t#'~" "f""l.... l"'" ~ .&< v# ..f. f !,..,''''-v;. 1"4 ~ •""./ii, .,...j1i,
~t'2-1'~ 4 ~)4~r-e I~~'J~, ~4. 2P&4}~ A~/~~~
~~I..'e / J ~ ~<U;:r,1. #'.-6 .......iJJ Iff/N....A ......
fir t'$,rr ~JJ ~~J',(lr .tiJ.,;.~
-r,~~1rK~ -1UI'-;-.,t A~itk~lh ~ (II 'JU'1 'r~' 7A.,~,
~ rfot"11Q#a....eA d'.J'M>, .J~~O--4
2fdvt~ ~,v-. J"'l- f' ~b~~ de:-c~:I~ ~/~
a'~ ~(:,!;,u-) ~~~c~y;;:j;.;~~ ...r~ ~~t'~~~
}~~,'Ja4e~ g~-t1I(,'l:'" 4~~"'~ ~ ~~/~~J,"'~
ei ~;,r"
4 ~-h...I./'L>.
'r VU~ J~,J~ /"'-ilu
I!-r-
'4-11
•___ J ,
/~.-~ "",W'.I/,j~~
YI#1f,-.I'J

?4te;.~t>vt" 1'1 A fu~v,V.A- Ii d:k.//~ ~ 4 ~N"e~

~ (,~r,::t.4tu.-
~I
tv ~ 4 "e:-
f'A -rlth.r "f1.1,jf/~ ~ e#~d~ ~/IoAIcIt
fa>, Ct'~, I~ ~ .!fit,. , / r.- ~ tr""
/ ';jJ~ .dJ1' e--ff"" ~,~~,...r?It-:P1cJI,f

.4 -",.....,'U) ~ fIw. rll'f~ ~J?"""; .J.~/';' ~

-fa ~;41te.
21J~{ 'rtf
~~j).~
XVlll The Need for Freedom in Scientific Research

THE NEED FOR FREEDOM IN SCIENTIFIC RESEARCH

The history of science teaches that the greatest advances in the scientific domain
have been achieved by bold thinkers who perceived new and fruitful approaches that
others failed to notice. If one had taken the ideas of those scientific geniuses who
have been the promoters of modern science and submitted them to committees of
specialists, there is no doubt that the latter would have viewed them as extravagant
and would have discarded them for the very reason of their originality and profun-
dity. As a matter of fact, the battles waged, for example, by Fresnel and by Pasteur
suffice to prove that some of these pioneers ran into a lack of understanding from
the side of eminent scholars which they had to fight with vigor before emerging as
the winners. More recently, in the domain of theoretical physics, of which I can
speak with knowledge, the magnificent novel conceptions of Lorentz, of Planck, and
particularly of Einstein also clashed with the incomprehension of eminent scientists.
The new ideas here triumphed; but, in proportion as the organization of research
becomes more rigid, the danger increases that new and fruitful ideas will be unable
to develop freely.
Let us state in a few words the conclusion to be drawn from the foregoing. While,
by the very force of circumstances, research and scientific teaching are weighed down
by administrative structures and financial concerns and by the heavy armature
of strict regulations and of planning, it becomes more indispensable than ever to
preserve the freedom of scientific research and the freedom of initiative for original
investigators, because these freedoms have always been and will always remain the
most fertile sources for the grand progress of science.

April 25, 1978

Louis de Broglie
 THE PROBLEM
OF
HIDDEN DETERMINISM

(Facsimile of one of the Notes inserted by Louis de Broglie into his manuscript; the
typed text appears as a Note at the end of Chapter 14 of the present book.)
xx m
The Problem of Hidden Determinis
 The Problem of Hidden Determinism xxi

p'-~ fa. ~ ~#, {fii4f"f,/.. ,t.;'-:I"PP;:;;:J~~J -;,.1~,~

7Cf ¥'~........., ""''''''uu ~ ~",.. .}P4t4'....... 4-.J,.
~/JIC/4
-r Ifvi./w...4 /~~/lI-- ,,*.. A'k .t1:J'~~ 4f r/,.·
- ~~P/tI~ ,( ~6/,'.J.,.£;:' "J;6'" ~, ..., . t'4' -P!IA.:-'/4-, /~
j"IU"';;;"'~ "..tz~ / CI //1" 4-'f Pi~-k J' hd,--.;..... ~ ....,. Alht-..
/' tI- - 21'h l"
JlU;"'u"Y aD'-. j.,,'/, rUil eU' , ....'hW ~= Z'r{l.J P 7:.,.
-: ill. I' 'r, 9t.-"' 4'''10'' "'" I')...t: -;. 't1f;.-- /f~.~a~
J ;; /1f1'4.
a __

~1
'11'.11.. de r"IJ.'Jfi.!i7t/-f"j/'l.. /Jllt'!' , (/p_~/,;~ ..,.v_..Jj. -;.
t -'~J
- t; ... Tl ~/,,....,, . /Jbtl A4c
~
(Q~eo/ ~'1"~_ ?- ':f..{--~ /' /:/'''~
(~.'tf'M· /"-J.
/1')

I c/flf'J!'Z d'/m. fj';;..f# 1""'''''60-4. l"t. t? 4'- -:e{A) ."" ... , /ij'J'
t?f"M.." fOA ~=- (It ", .::1.- f ; •. #r~:au. -.e11- ~ .h JU;Q
I'"' i/'- n'~f.4t I: ~vr.tJ~} ~7~ ~ ";'p~~r;;l~
r;.Ir'eI, o(e oq;:;..,s~ ~;;tf r..hJ,;~
~-
.:.r«u'~ ~ I;.J'.
, ~ .

":14~ hu(~,4t~-.b'~ ~y~ -L.-r ~ ,.",. ~?,,".q /'"1foJ!·

.,f'";UbM, 41.. ,N.'J... ~ rt ,," .. ~ .."'..... ~/4"F t: t..J d ;,....,1,.,•• 4tH-I.'•
•/Mct r.. ~ V;;;;, ~ /n<Jc ~$/9 fa> '-J'_"v"~ /I..,.
"'/lUi !fU, .....·r,:;:" /'~/... ~-4 "4u":~A' r~,f 'let'euJ
~ ~ "n.'''''' ;CI/; '-I,...~t-#A·r~,. .... -1;...." .. 4 ~J'';..4 e4ri"" ...'..
,: ,v-J;.r:i~ ,-1/. .
~"f,p, ~ {;..;;; 4'./,a ..4_ ,If; ;::~,(. jtrf4,'~
d. t",.~ (nli I-~ /:. ...1') - . . {f
Yo -%'A ",.. "..4 q,..... -t.. f';"" .),/:fZ: a...,j""..rd . . $;,.~-i
" .....".,t.,ri "*
",..,0'''''#..(;_ ,""-1'.
,J4d tll'/~-r.~"" ~ ~'k/4Li..J'..J;;... , 2....
~,,:: I,pU~
J..t.. ~,...'N,-It;;.. til. h. ,,,i,,,....--?.. ....r<,"'
/ 4~l.(
JI
 PREFACE

THE EVOLUTION OF THE IDEAS

OF LOUIS DE BROGLIE
ON THE INTERPRETATION
OF WAVE MECHANICS

By Georges Lochak

This book is quite exceptional, not only within the work of Louis de Broglie, but
also within scientific literature in general; for here one will capture, from life, as it
were, the thoughts of a great author as he suddenly turned back on his own work
and started to question the real meaning of a theory which he himself in part had
created.
The manuscript of this book is dated 1950-51, and Louis de Broglie waited thirty
years before allowing its publication: Indeed, he disowned it shortly after having
written it; for, in this manuscript he had exposed for the last time, without any
criticism and in a manner particularly brilliant and convincing, the interpretation
of wave mechanics according to the ideas of the Copenhagen school, which he still
supported at that time, but about which he began to have doubts the very moment
he re-read his own text. These doubts are expressed in this book in the form of Notes,
reproduced from notes that were stuck between the pages of de Broglie's manuscript
or in the form of deletions and significant emendations, which are mentioned each
time.
This dialog between the author and the person he was a few months earlier
arouses in the reader the stirring impression of invading the privacy of his thought,
all the more so since this is to be seen in the light of a later evolution which is
now known to us. For, in a reversal which seemed sudden to everybody, but which
was only the crystallization of prolonged reflection (as we shall later see), Louis
de Broglie was in fact to become an unrelenting scorner of the Copenhagen school,
whose ideas he had long adhered to. At the same time that he adopted this new
critical attitude, he took up again with youthful enthusiasm his theory of the double
solution, which he had abandoned earlier, but in which he once again invested all
his hopes. Whether this was right or wrong is a question that remains the subject
of a controversy which the future may one day resolve.
However, if the theory propounded by Louis de Broglie still remains debatable, I
think it is now possible, on the other hand, to contend that the events have proved
XXIV Preface

right those who, like him, have revived and developed the old questioning of the
interpretation of quantum theories.
Things have changed over the last thirty years, but in those days the sudden
turnaround of one of the most famous physicists of the century created a sensation,
not to say a scandal. People whispered about it in the corridors of the Institut
Henri Poincare, as if Louis de Broglie had suddenly been hit by a serious disease
from which it was wiser to keep one's distance.
Of course, this did not prevent a unanimously silent and respectful audience
from crowding to attend his lectures and seminars-while he passed by, imposing
and friendly as usual, pretending not to know anything. But was not the same
thing going on at Princeton, where Einstein, who obdurately refused to follow the
prevailing trend in physics, walked through the nimbus of devotion surrounding him
with an absent-minded, good-natured look but wrote to Max Born: "Here I am
considered as some kind of fossil which the years have made deaf and blind"?1
The debate developed and took on a world-wide dimension, almost like a religious
war: Most of the founders of quantum physics were seen to enter the lists followed by
their zealous followers, while a brilliant outsider such as David Bohm was suddenly
carried to the front of the scene because his great merit it was to have triggered the
whole business.
The debate has never ceased. It has become quite common to question the bases
of quantum mechanics, around which the beautiful and silent unanimity of earlier
days has been largely shattered-which is a good thing in the realm of science, since
total unanimity only exists to the detriment of inventiveness and can only lead to
the dead ends of ideology and scholastics.
In the present-day debate, Louis de Broglie's ideas are far from being accepted
by everyone; in fact, they are not only unknown to many but are also criticized by
some who know them (even more so by those who do not, but that is not unusual).
Indeed, his disciples and he himself recognize that they do not yet constitute a
coherent whole which could be regarded as a complete theory. However, these ideas
are beginning to come to the surface again, especially abroad.
Such is, for instance, the case for the recent works on nonlinear wave equations
and solitons, but only a few authors mention that Louis de Broglie and his disciples
were beyond doubt the initiators of these notions in the field of microphysics. I am
also thinking of ideas such as the primacy of position measurement over the mea-
surement of other physical parameters and of the ensuing primacy of the role of the
spectral analyzer over the measuring apparatus itself, which ideas were abundantly
developed by de Broglie and now are often given prominence.
After having written the manuscript which is only now being published, Louis
de Broglie developed his criticisms and defended his new ideas in twelve books and
over sixty articles, but I am sure the present book will retain a place of its own,
because it represents a turning point in his work and also because it is the book of
incipient doubts.
Indeed, in the works which followed this book, a double aspect is to be found:
criticism of the existing theories, on the one hand, and suggestions for an alter-

1. M. Born, ed., Einstein-Born Correspondance 1916-1955 (Seuil, Paris, 1972),

p.196.
 Evolution of The Ideas of Louis de Broglie xxv

nate solution, on the other. But a reader must be prepared to hear the criticism,
or at least the questioning, without however accepting the theory propounded by
de Broglie; so that, in the mind of such a reader, the answer offered may very well
conceal the relevance of the question. Here, on the contrary, there is no such prob-
lem, for the situation is reversed. The conception supported in the book is that of
the Copenhagen school, and in his footnotes de Broglie is only just beginning to
question this basis, which puts him on the same level as the reader who might also
be wondering about it. And yet these notes added to the manuscript, these "pap-
eroles," as Proust would have called them, already contain, in a form sometimes so
brief and elliptic and still full of doubts that 1 have had to expound them for the
reader, almost all the ideas that served as a basis for the work of another quarter of
a century which Louis de Broglie began at the age of sixty.
Whether the reader approves or disapproves of these ideas, it is to be hoped that
he will read them without any prejudice and that he will recognize genius wherever
it comes from, be it in the form of de Broglie or Heisenberg, of Bohr or Einstein.
Their glorious time is gone, unfortunately, and the founders of quantum physics are
all dead or very old. Those great figures of science are now part of history, and
so are their hatreds and rivalries. Naturally, rivalry still exists in our time and is
still just as fierce, for science is a combat area where men and ideas are forever
arranged in passionate confrontation, not a cold storage for well-established truths.
But let us at least fight between ourselves and no longer against them! Let us
admire their merit without spoiling the intellectual and artistic pleasure that their
teaching arouses in us with useless mental restrictions. Let us admire in this book
the subtle analyses of Bohr and Heisenberg, which are recorded here as much as
those of de Broglie himself. And let us admire the way this sexagenarian added
to his scientific genius enough strength of character, enough taste for danger and
disdain for people's opinions, to throw his own ideas back into question, in spite
of the fact that he had reached the apex of his career, and to stake his brilliant
reputation in the name of an idea which nobody yet wanted to know. And let us
also perhaps reflect upon the divine grace which enabled him to live long enough to
accomplish a new work in the evening of life and to say at the age of eighty: "1 have
often wondered if, from the intellectual point of view, the period which followed my
seventieth birthday was not the most beautiful one of my life." (Ref. III, 9y
However, in order to appreciate this transition book and the reversal of opinion
which it announces, it is necessary to place it precisely in the scientific itinerary of its
author, who had made the following text a conditio sine qua non for its publication.
1 think it may be interesting to begin with the history of this publication itself.
A few years ago Louis de Broglie told me he intended to entrust me with the
task of making use of his scientific papers in the way 1 thought best, at which time
he gave me several of them.
Most of these were loose sheets of small-size writing paper, with which all of
his assistants were well acquainted and on which he used to write short scientific
"poems," as it were, in black ink and with a penholder. Each of these "poems"
consisted of a clear and concise statement of a question, usually taking up no more

2. Roman numerals, followed by Arabic ones, identify references listed in various

sections of L. de Broglie's Bibliography appearing at the end of this book.
XXVI Preface

than one page, covered by his elegant, marginless, well-ordered writing, in which
everything would seem to indicate moderation, were it not for certain obvious details
in" the script, which pointed to an imperious, if always mastered, authority.
These sheets of paper were classified in envelopes, each of which bore a statement
in Louis de Broglie's handwriting on the degree of interest he himself attributed to
it.
Besides these envelopes, he also gave me several notebooks which corresponded
to what his younger collaborators called the "Thursday classes," because in the last
years of his academic life he gave two very different types of lectures at the Institut
Henri Poincare. On Mondays he would give the students classical university lec-
tures, which he repeated for the most part from year to year and which constituted
the basis of a theoretical teaching that was completed by other teachers and supervi-
sors. But on Thursday mornings, before a research audience, he gave a course whose
subject matter changed every year and which consisted of an original exposition on
a scientific problem of the moment, in accordance with a tradition belonging more
to the College de France than to the University. 3
These are the lectures which formed the basis for most of Louis de Broglie's
books, but some of them have not been published. This was the case for the few
texts he gave me. He immediately drew my attention to two of these notebooks
which stood out in any case because of their attractive cardboard binding covered
in beige, glazed cloth. Unfortunately, this was to warn me against their publication.
As a precaution, he had even written on the first page of each of them, in an energetic
hand and twice underscored: "Do not publish." He told me briefly his reasons for
this admonition, while at the same time encouraging me to read the manuscript
and kindly adding: "For someone like you, this could be very interesting." But
when I read his text the very next day, I was moved far more than out of mere
scientific curiosity, because I knew by heart the extension of each note and guessed
the meaning of the slightest penciled comment, because I could see that man, whom I
knew like my own father, behind the slightest deletion or the minutest emendation.
This convinced me immediately that, if he indeed could not publish this book,
someone else could; and soon I began trying to pressure and persuade him that in
a way his text no longer belonged to him but belonged to the history of science,
which certainly has a claim on it in one form or another. I also pointed to the fact
that the scientific public is only too accustomed to an antiseptic science, emerging
like Minerva in her shining armor from Jove's head, so that the presentation of
modern theories in formal, synthetic successions of axioms could lead us to think
(erroneously) that, unlike artists, men of science are able in some way to turn out
perfectly shaped results marked from the start with the seal of eternity. Here, by
contrast, was an opportunity to show science at work to the reader, as it hesitates
and painfully progresses before our eyes, as it does in real life. After all, everybody
contemplates with emotion the rough shapes of sculptors or the sketches of painters
in which, over and above the particular charm of each work-to-be, one is pleased to
discover the different styles of the artists and even projects abandoned and replaced
by other ones. Why then should one always conceal the sketches of physicists?

3. In reality, this distinction between the two courses was made only in 1954.
Up to then, both were of the same kind and different each year.
 Evolution of The Ideas of Louis de Broglie XXVll

Things did nDt go. so. easily at first, fDr LDuis de BrDglie is nDt a perSDn who.
changes his mind readily, but I did nDt drop the subject in the fDllDwing mDnths.
So, Dne day, with his apprDval, I returned the two. nDtebDDks. We examined them
tDgether, and then, with a feigned and slightly amused sDlemnity, we erased the
"Do. nDt publish," which was written in pencil, and I left with an authDrizatiDn to.
publish the manuscript ne varietur, Dn cDnditiDn that it WDuld be accompanied by
cDmplementary fDDtnDtes and an intrDductiDn which wDuld situate the text in the
tDtality Df his wDrk.
The main thing that must be explained is the dDuble reversal by which LDuis
de BrDglie first abandDned his Driginal ideas and then reverted to. them twenty five
years later. But, to. this end, it is necessary to. understand the very particular place
he Dccupies in the histDry Df quantum physics.
I think it is pDssible to. say that de Broglie was the first theDretician after Einstein
to. have believed in the existence Df light quanta (phDtDns) and the Dnly Dne to. nDt
believe in its dualism but, to. use his Dwn wDrds, in the coexistence of waves and
particles.
It is well-knDwn that the hypDthesis Df light quanta had great difficulty finding
acceptance and was regarded fDr a IDng time as a kind Df YDuthful errDr Dn the part
Df Einstein-an errDr fDr which he was fDrgiven Dn account Df the great fame he
had acquired in Dther fields. 4 Even the experimental cDnfirmatiDn by Millikan Df
the laws Df the phDtDelectric effect did nDt cDnvince anybDdy, nDt even Millikan,
and it is Dnly with the discDvery Df the CDmptDn effect in 1922 that peDple began
to. be impressed. But, at that date, de BrDglie had already been working Dn the
theory Df light quanta fDr a IDng time, and, in an article called "BlackbDdy radiatiDn
and light quanta"(Ref. I, 12), using this hYPDthesis, he derived the results Df the
thermDdynamics Df blackbDdy radiatiDn withDut using electrDmagnetism and with
the sDle help Df statistical mechanics and relativity. AmDng Dther things he fDund-
without the help of the wave theory-the expressiDn Df the Stefan-BDltzmann law and
(two. years befDre BDse!) the famDus 87rh/c3 CDnstant that appears in the fDrmula
fDr the energy density functiDn Df the radiatiDn. It is also. in this article that he first
advanced the hypDthesis Df the nDnzerD rest mass Df the phDtDn, which entailed that
its VelD city in vacuum wDuld depend Dn its frequency and that which we denDte by
C wDuld Dnly be a kind Df limiting velD city defined by relativity, which cDuld never
be reached, nDt Dnly by matter but also. by light.
In this article, he tried to. insist as much as pDssible Dn the cDrpuscular nature Df
light and to. use it in the best way, but, at the same time, he was careful to. keep the
cDrpuscular prDperties cDmpatible with thDse Df waves in a cDmmunicatiDn, entitled
"On interferences and the theDry Df light quanta" (Refs. I, 13; III, 9), in which he
put fDrward the idea Df the "existence Df cDnglDmeratiDns Df atDms Df light whDse
mDvements are nDt independent but cDherent," and he predicted that "Maxwell's
equatiDns wDuld prDbably appear as a cDntinuDus apprDximatiDn (valid in many
cases but nDt in all) Df the discDntinuDus structure Df radiating energy."
It is certain that this state Df mind dDminates all Df LDuis de BrDglie's wDrk,

4. Even in Einstein's letters there are references to. this general rejectiDn: "... I
no. IDnger hDld any dDubts Dn the reality Df quanta in radiatiDn, althDugh I am still
the Dnly Dne to. be cDnvinced Df it." (Letter to. Besso., 29th July 1918.)
XXVIll Preface

and this is doubtless the reason why he discovered wave mechanics: First of all,
because he was deeply convinced of the two-fold corpuscular and wave nature of
light and, secondly, because he did not conceive of corpuscular properties as a mere
"appearance," but as the actual existence of "true" particles, comparable on every
level with other material particles-possessing, like them, a rest mass and governed
like them by the laws of relativistic dynamics. It is revealing with regard to this to
note that de Broglie often wrote-and I believe he was the only one to do so-atoms
of light rather than quanta, and, when referring to their coherent conglomerations
in light waves, he spoke of these atoms as grouped in molecules.
Since he granted the same citizenship and a permanent and simultaneous exis-
tence to waves and particles in light, he was lead to look for the link which could
exist between them and establish a dependence between their motions. It is from
this question that wave mechanics arose, for it lead him to define an inner vibration
frequency of the particle by the equality moc 2 = hvo written in the proper reference
frame. This formula, which looks so simple, entails a major difficulty, since the mass
increases from the point of view of a moving observer, whereas the inner frequency
will seem smaller to him due to the slowing-down of clocks; accordingly the equality
thus written is not relativistically invariant. De Broglie noticed however that it
became invariant and that the relation of the quantum could be written

in all Galilean reference frames if, within the rest frame of the particle, one postulates
a stationary wave of frequency Vo identical to that of the inner vibration, for the
frequency v of this wave will then vary like the mass when there is a change in the
reference frame.
At the same time, Louis de Broglie established the formula V v = c2 , linking the
phase velocity of the wave with the velocity of the particle-which enabled him to
state the theorem of the harmony of phases, which constitutes for him the key to
the wave-particle duality: "The particle glides on its wave in such a way that the
internal vibration of the particle remains in phase with the vibration of the wave at
the point where it finds itself."
However, since he considered "atoms of light" as true particles, his reasoning
applied on a more general plane-as was in fact implied in the statement of it-to
any moving body, in particular to electrons, with which it was then necessary to
associate a wave, so that de Broglie was able to state as early as 1923 that "any
moving body could be diffracted in certain cases. An electron beam going through
a small enough aperture would exhibit diffraction phenomena." (Refs. 1,17; III, 9.)
Immediately afterwards he put forth the idea which is at the basis of his con-
ception of the physical world:
"Therefore we conceive of the phase wave as guiding the motions of energy,
and this is what may make possible a synthesis of waves and quanta. Wave
theory went too far, in that it denied the discontinuous structure of energy,
and not far enough in renouncing interference in dynamics."
Very soon after having written his famous Thesis in 1924, Louis de Broglie made
his thoughts clearer in a Note (Ref. I, 24), which was published one week before
 Evolution of The Ideas of Louis de Broglie XXIX

he defended his Thesis, and in which he introduced for the first time the notion of
singularity. One there reads especially:
"This property (we are concerned here with the theorem on group velocity) is
a direct consequence of the Hamiltonian equations and enables us to consider
the mass point as a singularity of the wave group whose motion is governed
by the Hamilton-Fermat principle."
This foreshadows the law which he was later to call the guidance law. He ended his
note with this sentence, which was a program in itself:
"However, the whole theory will become clear only if we succeed in defining the
structure of the light wave and the nature of the singularity as constituted by
the quantum, the motion of which we should be able to predict solely from
the point of view of waves."
Finally, a little later, on February 16, 1925, de Broglie made a first attempt
to develop his program in a Note, entitled "On the proper frequency of the elec-
tron"(Ref. I, 25), and he showed that, if the phase wave

'P(x, y, z, t) exp [2i11' V (t - ~ )]

satisfies d'Alembert's equation, its amplitude 'P obeys the equation

moc2)
( Vo = -h- .

Despite appearances, this is not the Klein-Gordon equation, for the sign is wrong,
and this is not due to an error in the calculations. However, de Broglie was not far
from having discovered the real wave equation. Therefore, as soon as Schrodinger's
work was first published, he amended his 1925 Note and was among the first sci-
entists to establish the relativistic scalar wave equation of the electron; more im-
portantly, he immediately introduced it in his ideas on wave groups and computed
the first singular solutions (Ref. I, 29): He had known for a long time that these
singular solutions should exist, and only they were able in his eyes to represent the
coexistence of waves and particles.
Soon afterwards, guided by his principle of phase harmony, de Broglie imagined
the kind of link which could exist between his singular waves and Schrodinger's
continuous waves, and he developed the theory of the double solution in a long
memoir in 1927 (Ref. 1,34).
This was not merely an interpretation of Schrodinger's equation-as was Made-
lung's, which was published at the same time. It was a continuation of the same
basic idea that had guided his research since he had started on this work a:ld which
had already resulted in the discovery of the wave properties of material particles.
And yet, several months later, he gave all this up. Louis de Broglie explained why
he did so in his "Personal memories on the beginning of wave mechanics" (Ref. III, 4),
often in very moving words. Naturally he disowned in part this argumentation when
he took up again his original ideas, but this article shows us at least what his reasons
were at the time.
These reasons can be summed up in one sentence: All of a sudden he felt he
was at a dead end, while physics was triumphantly moving ahead of him. And he
xxx Preface

immediately realized why he was being left behind in spite of his lightning start: It
was due to the difference between his point of view and that adopted by most of
the other theoreticians.
Louis de Broglie has an intuitive, concrete and realistic mind; he likes simple
physical images in three-dimensional space. He does not consider that mathematical
models have any ontological value, especially geometrical representations in abstract
spaces; he sees them as practical mathematical instruments among others and only
uses them as such: His physical intuition is not exercised at its best in the handling
of such abstract representations, and he always keeps in mind the idea that all
phenomena really take place in physical space, so that these mathematical reasonings
have a true meaning for him only insofar as he perceives at all times what physical
laws they correspond to in ordinary space.
Now he was confronted with the emergence of a completely different approach to
theoretical physics which was in the process of bearing fruit: This was a totally ab-
stract conception of physics, a description of natural laws which no longer employed
images in space and time but algebraic rules or even geometrical reasonings in ab-
stract representation spaces, which were generally complex and many-dimensional.
Theoreticians were developing a new kind of physical intuition on a different level,
an intuition based less and less directly on physical facts and exercised systemati-
cally in the field of mathematical analogies, algebraic rules, and laws of symmetry
or group transformations.
Theoreticians no longer tried to describe facts but to predict them. Their
premises and reasonings seemed purely mathematical, and it was becoming very
difficult, if not impossible, to discern any physical image behind them; and yet the
formulas they obtained were often miraculously confirmed by experiments. This
was a far cry from the theoretical physics of Fresnel, Maxwell, or Lorentz. The
remarkable thing is that Einstein, who was known for his keen physical intuition
and always remained very close to experiment, was also one of the main origina-
tors of this new state of mind, partly because of his work on relativity, but also
because of his 1917 memoir on quanta,S which had a great influence on de Broglie
and Schrodinger and in which a quantum problem was solved for the first time with
a geometrical reasoning in the configuration space of Hamiltonian dynamics.
It is obvious that Heisenberg's matrix mechanics and Dirac's q-numbers me-
chanics evolved from this abstract conception of physics, but the same holds for
Schrodinger's works, in which de Broglie's wave had already lost any kind of direct
physical existence, since Schrodinger saw it as propagating no longer in ordinary
space but in configuration space, where a single wave represented a whole system
of particles: It was to this abstract wave that Schrodinger now applied Huygens's
principle, through a generalization of de Broglie's ideas.
This tendency was soon to become the Copenhagen school's working principle-
as was stated clearly by Bohr and Heisenberg: 6 Their position can be neatly summed

5. A. Einstein, Verh. Deut. Phys. Ges. 19,82 (1917).

6. W. Heisenberg, Die physikalischen Prinzipien der Quantentheorie (Hirzel,
Leipzig, 1930); W. Heisenberg, Physics and Philosophy (Harper & Row, New York,
1959); N. Bohr, Physique atomique et connaissance humaine (Gauthier-Villars,
Paris, 1961).
 Evolution of The Ideas of Louis de Broglie XXXI

Table!

Classical theory Quantum theory

either or
Description of phenomena Description of Mathematical scheme
in time and space phenomena in corresponding neither
time and space to time nor to space

Causality Indeterminacy Causality

relations

up in Table I, which was proposed by Bohr. 7

Bohr and his disciples soon rejected as totally inadequate any causal and space-
time representation of quantum phenomena, on the basis of the uncertainty relations
and, on a more general scale, of the mathematical structure of quantum theories as
they emerged from the work of Heisenberg, Schrodinger, Dirac, Born, and von Neu-
mann. In other words, the left-hand column in Bohr's table had definitely been
rejected as outdated, and for them the future was to be found in the right-hand
column. There was no sense for them in speaking at the same time about the local-
ization of the electron and about its wave-properties, since these two aspects were
complementary, in the sense understood by Bohr, and there was no loss, according
to them, if one gave up the idea of their coexistence, since, thanks to sophisticated
physical analyses, it was known that no experiment could reveal both properties at
the same time.
This interpretation of the theory itself presented a double aspect:
1. First of all, it proceeded from philosophical choices linked with positivism
(since it proclaimed that only observable quantities were worthy of being included
in the theory), with idealism (for it recognized the existence of a fact or a property
only while it was being observed), and finally with indeterminism (because it gave
up describing individual microphysical processes in time and space from a causal
point of view).
2. But the interpretation also originated in an operational choice which was
probably essential for the success of this tendency, at a time when physics was
expanding rapidly. For, the attitude of the Copenhagen school could, in a way,
be characterized by Goethe's famous phrase: "Do not search for anything beyond
the facts: They themselves are the tenets." They might even have said: "Do not
search for anything beyond formulas: They themselves are reality." It seems quite
certain that, at a time when the theory was spreading rapidly and when people
were becoming acquainted with its essential mathematical foundations (such as the
principle of superposition of wave functions, the correspondence between physical
quantities and operators, and so on), such an attitude spared the mind from having

7. Quoted by Heisenberg in Die physikalischen Prinzipien de, Quantentheorie

(Hirzel, Leipzig, 1930), p. 53.
XXXll Preface

to look for underlying physical representations which would account for the facts
observed.
This undeniably facilitated the progress of the theory, and de Broglie was cruelly
affected by this at a time when he was having great difficulty in expressing mathe-
matically the wave-particle duality in the spacetime continuum. He mentions this
in Physique et microphysique (Ref. III, 4, p. 174):
"The more I tried to make the new material furnished by my ideas on wave
mechanics fit into this preexisting framework, the more obstacles I encoun-
tered, and this constantly rising sense of difficulty contributed to prevent me
from developing rapidly the construction I had begun during the whole of
the year 1925."
Let us not forget that 1925 was the year in which he got close to the wave equation,
barely missing it.
His first work on the singular solutions of Schrodinger's equation and on the rel-
ativistic wave equation gave him some new hope, but immediately afterwards great
obstacles began to pile up before him: viz., the general search for singular solutions
themselves, the strange and hardly believable behavior of singularities in stationary
states, and essentially the problem of describing particle systems in spacetime terms,
which is the approach he would have liked to substitute for Schrodinger's theory,
which remained confined to configuration space.
Louis de Broglie was invited by Lorentz to give a lecture at the famous Solvay
Congress of 1927, but as he was uneasy about the mathematical difficulties of the
double-solution theory, he presented only a very weakened version of it, which he
called the "pilot wave theory"; this version merely involved adding to Schrodinger's
continuous wave a hidden parameter consisting of a point representing the particle
and supposed to follow the flow lines of the wave.
In so doing, de Broglie evaded the mathematical difficulties of the double-solution
theory, but he also lost the logical coherence of this causal theory by viewing the
particle as "piloted" by a continuous wave whose probabilistic significance was unan-
imously recognized-as his adversaries did not fail to point out. Louis de Broglie's
communication was met by Pauli's penetrating criticism; in this clash de Broglie
was supported neither by Schrodinger, who did not believe in the existence of par-
ticles, nor by Lorentz, who was on his side but was too old, nor indeed by Einstein,
who merely encouraged him without really approving of his views, although he oth-
erwise attacked the Copenhagen school. On the other hand, de Broglie was faced
by the brilliant quintet consisting of Bohr, Heisenberg, Born, Pauli, and Dirac, who
presented, in an ostentatiously triumphant manner, their probabilistic interpreta-
tion and who were little inclined to compromise, although their theory offered a
number of conceptual defects, upon which Einstein did not fail to enlarge, because
it offered-and still offers-the most practical and heuristic interpretation that has
ever been proposed.
Thus, worried as he was by the discussions at the Solvay Congress, having lost
all hope of solving the problems he had set himself and having just been appointed
Professor at the Institut Henri Poincare, where he faced the embarrassing problem
of deciding which theory to teach, de Broglie soon resolved to join the dominant
trend and to adopt the views of the Copenhagen school.
 Evolution of The Ideas of Louis de Broglie xxxiii

However, rallying in this way implied a profound reversal, for Louis de Broglie
was not only giving up his interpretation of wave mechanics-which was in itself a
painful action-but he was also compelled to acquire new modes of thought that
were totally foreign to him and even contrary to his most profound instincts.
It followed from this that during five years, from 1927 to 1932, apart from a few
surveys, he did not publish a single memoir!
Then, all of a sudden, he emerged from this long silence and, within a few years,
he made his second great discovery: the wave mechanics of the photon, of which
Heisenberg8 later wrote that: "The idea expressed by Louis de Broglie in 1936,9
according to which light quanta must also be considered as compound entities,
leads to problems of principle as important as those raised by the famous discovery
of matter waves."
It is impossible to expound or even summarize this theory here, but it is necessary
at least to know in which way it fits into the totality of de Broglie's work, in order to
understand how he was lead to this second turnabout and the return to his original
ideas.
Wave mechanics originated in a generalization of Einstein's ideas on the wave-
particle duality of light to the totality of matter, but, strangely enough, the photon
did not obey the basic equations of wave mechanics: Schrodinger's equation was not
relativistic, and the Klein-Gordon equation was not able to account for polarization.
This was an enigma which could not leave de Broglie indifferent, and it fell to
him more than to anybody else to try and make light re-enter the bosom of wave
mechanics, to which it first gave rise. As soon as Dirac's theory appeared, he felt
that this was now possible. He knew this theory very well and in fact wrote a
brilliant treatise on it: L 'electron magnetique (Hermann, Paris, 1934). The theory
was relativistic, it contained an element which looked like polarization (the spin),
and, among some of the quantities it enabled one to define, there occurred an
antisymmetric tensor of rank two similar to Maxwell's. All this being so, the spin
1/2 is nevertheless not correct, and the ensuing statistics is that of Fermi, not of
Bose: A photon was certainly not a Dirac particle.
Louis de Broglie groped for several years before he found the key to the mystery.
He was guided by considerations of symmetry, by the necessity to account for the
annihilation of the photon in phenomena such as the photoelectric effect, and by
the analogy existing between this phenomenon and the annihilation of an electron-
positron pair in Dirac's theory: This gave him the idea that the photon cannot
be an elementary particle and that in fact it must be composed of such a pair of
Dirac particles with a very small mass-maybe of "neutrinos," which explains the
designation "neutrinian theory of light" which is sometimes used.
In 1934 de Broglie established the wave equations of this composite particle and
an algebraic transformation with which he showed that de Broglie's equations, con-
structed by a fusion of two Dirac's equations, could be separated into two distinct
systems of equations: One of them corresponds to a spin-O particle, for which an
experimental counterpart has not yet been found; the other is simply the system

8. W. Heisenberg, in L. de Broglie, physicien et penseur, A. George, ed. (Albin-

Michel, Paris, 1953).
9. In fact, Heisenberg's reference is incorrect: The date of the theory is 1934.
XXXIV Preface

of Maxwell's equations, supplemented however by complementary terms which in-

troduce the electromagnetic potentials. These terms are very small because they
contain the rest mass of the photon as a factor, but they cannot completely can-
cel one another, as a consequence of a circumstance by which the internal logic of
the theory matched de Broglie's profound conviction: Indeed, the coherence of the
formalism requires the rest mass of the photon not to cancel out.
Now, however small these complementary terms may be, they are sufficient to
break the gauge invariance of the theory, and it appears that the automatically
ensuing gauge condition is that of Lorentz.
Although de Broglie's theory is not free from difficulties, it is impossible to
suppress one's admiration for this grand synthesis of matter and light which was
achieved by wave mechanics and lies at the basis of much subsequent work. Louis
de Broglie himself worked at it for ten years, as did several of his assistants, and
he devoted twenty papers and six books to this theory and to its generalization to
particles of arbitrary spin.
It is interesting to note that in all these papers there are more traces of the
author's profound mind and spontaneous manner of reasoning than of his adhesion
to Bohr's ideas and to the more abstract and formal mode of thought he had imposed
upon himself in physics: A man can change his opinions but not his true nature.
The wave mechanics of the photon is well and truly constructed as a physical
model in ordinary space. Its reasonings always remain very close to classical images
and only move away from them insofar as they are transposed immediately into the
mathematical language of quantum mechanics and, more precisely, of Dirac's theory.
But the formal laws of group invariance or the use of group representations, for
instance, are not heuristic procedures for de Broglie, in contrast to other physicists
such as Heisenberg, Pauli, Jordan, or Dirac. For example, de Broglie himself did
not have the idea of linking the spin-particle equations to the finite-dimensional
representations of the Lorentz group. But it must be noticed that this idea was
propounded only after he had opened up the way, by constructing his theory of light
with intuitive arguments on the emission and absorption of photons, their possible
relation to Dirac's pairs, the properties of the center of mass of a pair of relativistic
particles, and so on. He analyzed these delicate relations between "Abstract theories
and concrete representations in modern physics" (Ref. III, 3, p. 91) in a text where
he demonstrates an awareness of the strength and rigor of abstract reasoning and
yet remains convinced that concrete representations, although they are vague and
fragile, always questionable and often abandoned because they are more or less
inaccurate, still remain the salt of the earth.
In his theory of light, de Broglie's adherence to the ideas of the Copenhagen
school is to be traced only in his use of those algorithms which had become common
in quantum mechanics, in particular the computations of transition probabilities
starting from wave functions and Hermitian operators representing physical quan-
tities. In short, he was using the same idiom as everybody else, which implied in
itself that he had given up his ideas on the permanent localization of particles, that
he had therefore abandoned his initial project, for the general reasons that Bohr
and Heisenberg had developed, but which he had adopted, ending up being "all the
more convinced because [he] had tried so long to avoid them in vain," as he put it
himself (Ref. II, 4, p. 166).
 Evolution of The Ideas of Louis de Broglie xxxv

It must be recognized that it seems rather unlikely in any case that he would
have completed the ambitious program he had set himself in the first place, for we
know how great were the difficulties that stopped him. On the other hand, maybe
it was because he relinquished this program that de Broglie was able to complete
the grand cycle he had himself begun and thus to unify the electron and the photon
in the world picture of wave mechanics.
However, once the cycle was completed (in the forties, during the war), he sud-
denly felt a void in himself, for what else could he do after all this? The synthesis
he had dreamt of had been achieved for the main part, and there seemed little
possibility of improving it within the framework of existing theory. The only other
project that might have matched his ambition was the problem of the atomic nu-
cleus, but, after having studied it carefully (he wrote a three volume treatise on it),
he remained dissatisfied with the existing theories, while recognizing that he could
not improve them. So he began to wonder if the inadequacies of the theory of the
nucleus were due to a temporary lack of savoir faire on the part of the theoreticians
or to a more deep-rooted inadequacy of the quantum theories themselves.
This is why he did not really devote all his energy to this problem. However,
he was fifty years old and felt perfectly fit from the intellectual point of view. He
had no other physical handicaps than those that the war years had brought! His
diet was poor and he was cold in winter, like everybody else. Tending to avoid
his freezing cold study, he would only occasionally sneak into it with his coat on
to find a book; otherwise he would work cloistered in his room, which was heated
with logs from the Foret de Chantilly (which belongs to the Institut de France).
But this was only a temporary inconvenience: The real hindrance to his work was
the number of obligations he had to face now that he was at the peak of his career,
although even this did not prevent him from writing thirteen books and thirty-three
original papers between 1941 and 1951 (the years we are concerned with), thanks
to a monastic way of life and unremitting labor.
Now, the interesting aspect of these papers is their extremely varied and even
desultory character, which points to the absence of a grand project in progress.
Indeed, in the space of those ten years, Louis de Broglie wrote about the photon
and particles with spin (but less and less so), the nucleus, wave guides, electronic
optics, adiabatic invariants in classical mechanics, the relativistic transformation
law of temperature, the structure of the probabilistic quantum scheme, the phase
wave and the proper frequency of the electron (for the first time in twenty years), the
experimental problem of spin measurement, quantum field theory, thermodynamic
analogies in classical mechanics and electrodynamics, and lastly he wrote the present
book, or rather, as we know, the text of the lectures he was to deliver in 1951 and
1952 at the Institute Henri Poincare; for, as always, he was drawn back to the
quanta.
Nevertheless, if one examines these books and papers closely, if one considers
together the remarks gleaned here and there, during years of conversations with
the author, as I was able to do, and, finally, if one thinks of the work he produced
afterwards, the list of themes quoted here falls readily into place.
Naturally some themes fitted the occasion; for instance, the book on wave guides,
which was commissioned by the State at the beginning of the war, or that on corpus-
cular optics, which had apparently been requested by his collaborators. But even
XXXVI Preface

those works, which seem very specialized, are full of fundamental remarks (and even
developments) on waves and optics, which Louis de Broglie subsequently reused and
extended. The papers on the photon and particles with spin need not be accounted
for, since they obviously represent the culmination of a whole period. Those con-
cerning the nucleus aimed, as we said, at exploring the ultimate possibilities of wave
mechanics in that field; they made Louis de Broglie reach the conclusion that the
limits of the present theory had probably been attained. The same can be said of his
work on quantum field theory, in which he never accepted the idea that tricks in cal-
culations, however ingenious, might resolve the problem of the "infinities." In fact,
even before 1950, he had begun to be persuaded that the difficulties surrounding
the theory of the nucleus, as well as those of quantum electrodynamics, could not
be solved within the accepted conceptual framework and revealed the fundamental
inability of the entire theory to describe spacetime structures. This growing con-
viction was to be expressed later, once it had crystallized in his mind, in 1952, and
he voiced it unambiguously: " ... today the explanatory power of wave mechanics,
such as it is taught, seems almost totally exhausted." (Ref. 111,6, p. 143.)
This state of mind accounts for all the other lines of thought enumerated above,
including those of the present book. Indeed, these themes can be divided into
two simple categories: On the one hand, he reexplored the deepest and remotest
foundations of the quantum theory, through a return to wave mechanics such as
he had conceived of it originally and to considerations in thermodynamics, classical
mechanics, and relativity, which he was later to develop at length. On the other
hand, he questioned the now classical and orthodox interpretation of wave mechanics
such as he taught it himself, and this is visible in his work on the probabilistic scheme
of the theory, on the measuring of spin and, most of all obviously, in the present
book. Was this book an attempt to convince himself? One may be inclined to think
so, but to tell the truth, I do not know if it is, and de Broglie himself probably never
really has known if it was. In fact, making this kind of evaluation is no better than
drawing a line on the unconscious, and therefore no value can be assigned to it. Let
us simply say that, since de Broglie was questioning himself about these problems,
he reexamined them carefully and wished to make his audience share the result of
his reflections. One thing is certain: the lectures were perfectly orthodox, sounding
conviction-filled and convincing enough! Only complex analyses can introduce any
doubt on these questions, and there were no such analyses in the initial text: The
critical Notes which we have reproduced are all posterior, even if some of them were
only added a few months afterwards (we do not know their dates exactly). It is also
sure that, when de Broglie wrote the book, he was not yet considering taking up
again the theory of the double solution and even less that of the pilot wave.
Now, in the summer of 1951, between the two parts of this sequence of lectures,
the manuscript being, as it seems, already completed in conception if not in writing,
de Broglie received from Princeton the preprint of a long paper by a young physicist,
who was still little known in France in spite of his work on plasmas and a remarkable
book on quantum mechanics, which had just been published. His name was David
Bohm: In his paper he took up again and developed the theory of the pilot wave-
and he only found out at the last moment, and just in time to include this fact in his
bibliography, that de Broglie had already constructed the same theory twenty-five
years earlier and almost immediately abandoned it.
 Evolution of The Ideas of Louis de Broglie XXXVll

Louis de Broglie's first reaction was negative (Ref. I, 93). He knew better than
anyone what arguments could be raised against the pilot wave, notably that one
cannot claim the causal movement of a particle, if one sees that as depending on a
wave that propagates in configuration space, rather than in physical space, and is
also subject to wave-packet collapse in any localization measurement.
And yet Bohm's paper worked in such a way on de Broglie as to break a spell
that had long held him captive. This spell, which he shared with most other physi-
cists, had been cast on him by the scientific-philosophic language with which the
Copenhagen school had enveloped the quantum formalism and which was perme-
ated by uncertainties and complementarity. This language had been perfected and
even perpetuated by von Neumann's famous theorem, which stated, on the basis
of an impressive mathematical construction, that it was impossible to account for
quantum laws with a causal theory involving hidden parameters.
When one considers this theorem, one realizes its incredible philosophical preten-
tiousness, for it tries to prove, from within a theory, that the principles on which the
theory is based are definitive and constitute the ultimate limit of human knowledge.
Here was a new form of the triumphant excesses of Laplace's determinism-which
was antinomic to it only on the surface. Now, Louis de Broglie had just presented
von Neumann's theorem in his lectures as irrefutable-and in fact no one had even
tried to attack it in the foregoing twenty-five years. Well, Louis de Broglie suddenly
had an intuition that the theory of the pilot wave, however imperfect, constituted a
counter example to this theorem, which forbade its existence in principle; he added
to his manuscript an essential note which begins with the following sentence, in
which a slight hesitancy still appears: "This existence of the theory of the pilot
wave seems however to point to some kind of flaw in Mr. von Neumann's theorem."
Subsequently he showed that this flaw resides in the implicit hypothesis accord-
ing to which the probability distributions predicted by quantum mechanics must
all be realized simultaneously, even when they are associated with quantities that
cannot be measured simultaneously. De Broglie was well prepared for this type of
analyses thanks to the very clear study of the probabilistic scheme which he had
recently published (Ref. V, 44; III, 9). Later on, he developed his argumentation
in several books (Refs. II, 27,29,33). It constitutes, in my opinion, the only valid
physical refutation of von Neumann's theorem. Other refutations were proposed
afterwards for, once the theorem had been attacked, other flaws were found, but
they all have to do with logical or formal problems. They are not uninteresting, but
I think a general significance can only be obtained by a physical reasoning based
on the statistical scheme of the theory and involving the fundamental problem of
the wave-particle duality. This is why this reasoning retains its validity also against
other theorems of the same type, even if they are presented in a very different form. lO
This refutation was essential to de Broglie, for, behind the veil he had just lifted
again, he rediscovered the image of the world that had once been his and that had
almost been erased from his mind.
This, then, was his second turnabout. The lost image reconstructed itself grad-

10. Let us also point out that Bohm, in his cited paper, rejected von Neumann's
theorem with an argument on measurements similar to that used by de Broglie but
not developed in full.
XXXVlll Preface

ually before his eyes, feature after feature, as we can see in those few Notes that
are still tinged with the prudence of the last doubts, an image he was beginning
to believe in once again and which he was to try to perfect and extend during his
entire old age.
No physicist can remain indifferent to this conception, whether he shares it or
not. It is another conception of microphysics, a conception completely different
from that which is normally taught, and everyone should know that it exists, for it
brought about the birth of wave mechanics.
But are the seeds of any new discoveries to be found in this theory as taken up
again belatedly by the same man? As for myself, I think there are, and I shall men-
tion them further on, but I believe this work of an illustrious old physicist should
above all be contemplated in the same way as Michelangelo's Milan Pieta, which
is also a late work; it may seem rough and unfinished when compared to the mag-
nificent Vatican Pieta, but is it not also a remote prefiguration of the art of our
time?
The allusion to art is just as natural in the case of this physicist, who was also
a man of letters, as in the case of Einstein, who was the scientist de Broglie most
passionately admired. Yet, whereas Einstein's career and thought immediately sug-
gests musical harmony, such a metaphor would be quite inappropriate for de Broglie:
Music is the only art that has remained foreign to him. But de Broglie's work seems
to have two keys.
The first is of course history: He studied it so closely that one day he told me he
believed he had read even more books on history than on physics. More particularly,
the history of ideas in physics, from the 16th century and especially since the 17th
century, played an essential part in his work: This was not merely an erudite hobby
or pastime, it was both a driving force for his synthetic mind and a nurturing soil
for his thought. The first word of his famous 1924 Thesis was "History," and this
was not a coincidence.
The second key to his work is visualization: It consists in a search for a world
image, which he often calls "Weltbild," citing Max Planck. For de Broglie, to
understand means to see. He likes concrete models, and therefore a model can
only be a visual representation in physical space. In his speech, which is in, itself
transparent, metaphors taken from optics and the visual world abound. When he
refers to a great idea, most of the time he calls it a "flash of light" or a "light
in the night." "The only real creators are visionaries," he likes to say; speak-
ing of our inner life, "the only object of our knowledge," he says that "all that
we know is filtered and refracted in it" (Ref. III, 6, p. 250); he described the dis-
covery of wave mechanics in this way: "A great light suddenly appeared in my
mind" (Ref. III, 6, p. 180). Joy for him consists in seeing: When remembering
the discovery of atomism and statistics, he exclaimed: "On that day, the veil
was torn away, and at last we glimpsed with relief the physical reality that had
been hidden behind the abstract forms of classical thermodynamics" (Ref. III, 3,
p.93).
His mind is fundamentally relativistic and his imagination moves in a kind of
four-dimensional continuum whose temporal dimension is borrowed from history,
and whose images are spatial. "Only the motions of elements localized in space and
 Evolution of The Ideas of Louis de Broglie XXXIX

time are really physical"ll he said. For de Broglie a great idea is one that, in a
single spatial image, accounts for a synthesis resulting from the sudden perception
of an analogy between two different physical laws or between two representations
which had long been termed contradictory. He always experiences creation as a
dazzling poetic vision, and he cannot help feeling sad when he sees it weaken and
fade as it is translated by himself or by others into a necessarily mathematical
language. His need for images is so strong that, in his evocation of "The scientist's
last moments"(Ref. III, 6), his ultimate dream is that, maybe, what we see in
spacetime is not a real vision yet. Maybe physicists are like craftsmen weaving a
tapestry on the back side: They will only be able to give an account of the real work
"when they can turn over that piece of work and view it from the front."
Thus, de Broglie took up again his work and rediscovered the world picture of
his youth. Within a few months he reexamined various aspects of it, began to talk
about it with more and more self-assurance, swept away the conceptions he had
allowed to be imposed on his mind, and undertook to criticize it and tackle the
technical difficulties involved in his own theory.
With the help of a new group of young assistants, he attacked the problem
regarding the guiding of singularities and of nondeformable wave packets in nonlinear
equations (i.e., what is now called solitons); he generalized his ideas to Dirac's
theory, to refringent media optics, and partly to systems of particles; he elaborated
a new quantum theory of measurement; he developed variable-rest-mass dynamics
and relativistic thermodynamics; lastly, he launched the idea of a thermodynamics
of the isolated particle. All this is to be found in about fifty papers and fourteen
books, including the one we are dealing with here.
If I were to pick out something in all of this, I would choose two ideas which
seem essential to me.
The first idea concerns the solitons, which we would call ondes a bosses (humped
waves) at the Institut Henri Poincare. This idea of de Broglie's used to be considered
as obsolete and too classical, but it is now quite well known, as I mentioned above,
and is likely to be developed in the future, but only provided we realize what the
obstacle is and has been for twenty-five years: It resides in the lack of a general
principle in the name of which we would be able to choose one nonlinear wave
equation from among the infinity of possible equations. If we succeed one day in
finding such an equation, a new microphysics will arise.
The second idea is the thermodynamics of the isolated particle, which de Broglie
based on the one hand on the analogy in relativistic variance between the frequency
of a clock and temperature,12 and, on the other hand, on a comparison between the

11. L. de Broglie, "13 remarques sur divers sujets de physique theorique," Ann.
Fond. L. de Broglie 1 (3), 116 (1976).
12. The variance of temperature still remains a subject of controversy, be-
cause the Planck-Einstein-Laue formula, which de Broglie used, has been attacked
[see H. Arzelies, Thermodynamique relativiste et quantique (Gauthier-Villars, Paris,
1968).]; but, to tell the truth, even the variance of the frequency of a clock is de-
batable [see L. Brillouin, Relativity Reexamined (Academic, New York, 1971)]~and
yet this did not prevent the birth of wave mechanics on the basis of the generally
accepted variance! The controversy around these problems is not over.
xl Preface

three extremum principles of physics, those of Fermat, Maupertuis, and Carnot.

Even this is not a real theory, it is only an insight into a synthesis we may one
day be able to achieve and use: Within a year or within a century? Who knows?
Let us not forget that in the days of Laplace people still doubted Fermat's principle
and that a century after Huygens's death his famous principle was still a dead letter.
Great ideas unfold by a slow process, and this is not easily admitted in our hectic
age where there is little space left for poetic meditations. This is probably one of the
reasons why, outside a small circle of disciples, de Broglie has remained for several
years so ignored, at the same time that he became covered by petrifying honors.
Maybe another reason is that, in the face of a dominating, self-assured school of
thought, locked into its dogma and unwilling to change its position, de Broglie
stood out as the uncompromising and austere incarnation of a certain tradition
in French science, that of Fermat, Laplace, Fresnel, and Poincare, which is often
rejected today in favor of more pragmatic or formal methods, which are also more
international and collective.
Now, whatever the necessity or the results of team work may be, we must not
forget that no large team has ever produced a great idea: Only individual men have
ideas, teams merely develop them. If individuals are crushed, there will be no more
ideas. In the same way, whatever the benefits of international cooperation may be,
we must remember that, although scientific results are valid on an international
plane (in fact, most of us hope they are universal!), the cast of mind, the style of
men's thought and work, remain national: This is why only the Americans can build
a really American science, only the Germans can build a really German science, and
so on. If the French were to neglect their tradition, no one would revive it for them.
Each tradition should indeed be enriched by other traditions and evolve, but it seems
to me that if French scientists too easily follow fashions from abroad, they diminish
their own destinies, even while possibly improving their international standing.
Louis de Broglie has often spoken out against the growing power of the scientific
community and the anonymous influence of committees of specialists, which he
perceived as tending to make thought too uniform. He rose up against the danger
of directed research and underlined the importance of freedom in the field of scientific
research and the necessity to reexamine commonly accepted theories and principles
without reservations.
This is the spirit in which the Fondation Louis de Broglie was created; he passed
on to it his pursuit of physical clarity and his search for simple, intuitive theoretical
images. Above all, however, he passed on his profound belief in the fact that no
theory and no hypothesis are established once and for all and that therefore no
criticism and no new idea can be rejected without first having been thoroughly
debated.
I confess it is with a trembling hand that I complete this preface for a master
whom I perceive as being at the same time so superior and so close to me. He who
has written so many prefaces! Nonagenarian, exhausted by work and by the years,
he was not going to write this one, but I would like to leave him the last word,
with the text of a Note that he presented before the Academie des Sciences on the
occasion of the fiftieth anniversary of wave mechanics and which we may regard as
his scientific testament.
 ON THE TRUE IDEAS
UNDERLYING WAVE MECHANICS 1

Note2 by Louis de Broglie, Member of the Academy

On the occasion of the fiftieth anniversary of the discovery of wave mechanics, the
author recalls the ideas that guided him at that time and set forth the reasons why
presently it appears necessary to return to those ideas, which nowadays are well-nigh
forgotten in the teaching of quantum mechanics.

I presented the first principles of wave mechanics in three Notes that ap-
peared in Comptes rendus of September-October 1923 and later gave a more
elaborate version in my Doctoral Thesis submitted on the 25th of November
1924. My essential idea was to generalize for all particles the coexistence of
waves and particles, which had been discovered by Einstein in 1905 for the case
of light and photons. In conformity with the clear ideas of classical physics,
I tried to imagine a real physical wave which transported minute and localized
objects through space in the course of time. Two ways of doing so then oc-
curred to me. The first is not taught anymore in the ordinary classroom, but
today I consider it by far the more profound. It was sketched in my 1923
Notes and further developed in the first chapter of my Thesis. It was based
on the difference between the relativistic transformations of the frequencies of
a wave and of a clock. Assuming that a particle possesses an internal vibra-
tion, which causes it to resemble a little clock, I supposed that this clock moved
in its wave in such a manner that its internal vibration remained always in
phase with the vibration of the wave. This is the postulate of "phase coinci-
dence." These hypotheses appeared to me necessary because the relation W =
hv, when applied to a particle, implies the existence of an internal frequency v
of the particle, while it is known since the work of Planck and Einstein that
v is also the frequency of the wave carrying the particle. Hence, the parti-
cle appears to be incorporated in the wave, such that it occupies a tiny re-
gion where the wave amplitude is very large. From this one can derive the
well-known formula p = h/ A. In the second chapter of my Thesis, I then
showed that, in the case where the wave propagation is studied in the approx-
imation of geometric optics, one is led to the identification of Fermat's princi-
1. The original, French version of this Note appeared in C. R. Acad. Sci. B 277,
71 (1973).
2. Presented at the June 25, 1973 meeting of the Academie des Sciences.
xlii On The True Ideas Underlying Wave Mechanics

pIe with Maupertuis' principle of least action and hence again to the formula
p = hi>"·
It is proper to emphasize the differences existing between these two modes of
reasoning. The first, viz., the postulate of phase coincidence, is of an essentially rela-
tivistic nature, considering that it is based on the difference between two relativistic
transformation formulas; whereas the second, the identification of the principles of
Fermat and Maupertuis, has nothing essentially relativistic about it, since these two
principles are equally valid in classical and relativistic theories. Another difference
separating the two methods is that the first reasoning is valid for all wave propaga-
tions, whereas the second makes sense only for those propagations that take place
in the approximation of geometric optics.
Following the appearance of my Thesis, my ideas have often been misinter-
preted by those saying that, according to me, the electron was a wave, which,
of course, did away with the particle itself. It seems that adoption of this idea
motivated Schrodinger in his very elegant 1926 papers to write down for the first
time the propagation equation of a wave, which he called the 'Ij; wave. This he
did for the electron, but only in the Newtonian approximation and without taking
the spin into account. In this manner, Schrodinger was able to compute exactly
the wave processes corresponding to the quantified states of an atomic system, as
conceived classically in the works of Bohr and his followers. It is quite certain that
Schrodinger then viewed his 'Ij; wave as a physical wave. But he abandoned com-
pletely any idea of localizing the particle in this wave, so that the picture which
he formed of the atom and, more generally, of the 'Ij; waves, made no provision for
localized particles. This had very grave consequences and made Schrodinger's use
of configuration space paradoxical in the case of particle systems. Soon afterwards,
Born introduced the normalization of the 'Ij; wave, by which he modified arbitrarily
the amplitude of the wave and hence deprived it of all physical reality. The normal-
ized 'Ij; wave was thus transformed into a simple probabilistic representation which,
while leading to a large number of exact predictions, provided no understandable
picture of the coexistence between waves and particles.
Schrodinger's work had the merit of making one see clearly that wave mechan-
ics, when applied to atomic systems, gives rise to problems where the approxi-
mation of geometric optics is no longer valid. It follows from this that Fermat's
principle ceases to be applicable and no longer allows us to define a ray which
could be interpreted as a particle trajectory. If, therefore, one refuses to invoke
the principle of phase coincidence, one is led to the conclusion that it is impossi-
ble to ascribe a trajectory to the particle within its wave and to the claim that
the particle can have only isolated localizations without intermediate positions.
But such a conception raises serious difficulties, in particular the one pointed out
by Einstein at the Solvay Congress of 1927. It can be summarized as follows:
Consider a source that emits a spherical wave transporting a particle. After a
moment, the particle manifests its presence at a point of the spherical wave by
its localized effect on a detector. It is evident that the emission of the parti-
cle by the source is the cause of its arrival at the detector. But the causal link
between the two phenomena implies the existence of a trajectory, and to deny
this existence is to renounce causality and to deprive oneself of any understand-
ing.
 On The True Ideas Underlying Wave Mechanics xliii

Let me now make an important observation. Since normalization, which arbi-

trarily modifies the wave's amplitude, does not alter its phase, customary quantum
mechanics is able to define the same frequency and the same wavelength as my
theory, and it is this feature that makes the former a powerful theory leading to a
large number of precise results. But, contrary to what is usually admitted, quantum
mechanics does not have the right to postulate W = hv and p = hi A, because the
energy Wand the momentum p of a particle are properties which are associated
with the concept of a localized object that moves through space along a trajectory.
The reason I was able to establish these formulas was that I advanced the hypothesis
of a particle localized inside its wave.
When, in 1928, I assumed the functions of a teacher, I presented the ideas then
current in quantum mechanics, and for many years I abstained from developing my
own original ideas.
But in about the last 20 years I have again convinced myself that one should
return to the idea that a particle is a very small object that is localized and moves
along a trajectory. As I have shown in a series of increasingly thorough studies, 3
this idea can be implemented, while conserving the statistical significance of the nor-
malized 'ljJ wave, with the aid of my concept of a particle guided by its wave, if one
completes this notion by a so-called hidden thermodynamics, whose development
opens up very novel perspectives. One particular consequence of this therrnody-
namics appears very important to me: The principle of least action would merely
be an aspect of the second principle of thermodynamics. 4
It is important to note the astonishing fact that, in light and particle op-
tics, one can predict with extreme precision an enormous number of phenom-
ena starting from wave propagation, without at all invoking the no less certain
corpuscular structure of the energy carried by the waves. In the case of in-
terference and diffraction phenomena, the statistical postulate of Born is suffi-
cient to explain the observational data. But, whereas in the usual quantum
theory one has arbitrarily to assume this postulate, I am able to supply a jus-
tification for it. There exists, however, a case where the postulate of phase
coincidence appears to provide an explanation which the usual theory is un-
able to give; it obtains when one considers the action of a IIertzian wave with
frequency v on an oscillating circuit or any other device adjusted to this fre-
quency. It is, in fact, natural to think that some of the photons carried by
the wave will transfer their energy to the oscillating circuit in the form of sud-
den impulses, which compensate the damping. But the energy thus transmitted
to the circuit cannot sustain its natural oscillation unless these impulses have
a rhythm corresponding to the natural frequency of the circuit, which is that
of the wave. This seems to prove that the incident photons possess an inter-
nal oscillation frequency equal to that of the wave, which in turn confirms the
principle of phase coincidence, while the ordinary theory provides no analogous
idea.

3. 1. de Broglie, La reinterpretation de la Mecanique ondulatoire (Gauthier-

Villars, Paris, 1971).
4. L. de Broglie, La Thermodynamique de la particule isolee (Gauthier-Villars,
Paris, 1964).
xliv On The True Ideas Underlying Wave Mechanics

In conclusion, I think that my early ideas, which I have amended and fur-
ther developed over the last years, permit us to understand the real nature of
the coexistence between waves and particles, for which customary quantum me-
chanics and its extensions provide only an exact statistical view, without reveal-
ing their true nature. The principle of phase coincidence teaches us, in fact,
that there exists a corpuscular dynamics, having the character of a dynamics
with variable rest mass, which underlies all wave propagation phenomena, even
if these take place outside the approximation leading to geometric optics. And I
think this is a point which the present quantum mechanics has not allowed one to
see.
As at my age I cannot hope to continue my personal work for a long time, I
would like to express the hope that some young researchers will dedicate themselves
to the development, in the direction I have indicated these last years, of the ideas
which half a century ago gave rise to the birth of wave mechanics in France.
 PART ONE
(1950-1951)

ON
HEISENBERG'S UNCERTAINTIES
AND THE
PROBABILISTIC INTERPRETATION
OF WAVE MECHANICS
 CHAPTER 1

PRINCIPLES OF WAVE
MECHANICS

1. CLASSICAL MECHANICS OF THE POINT MASS.

THEORY OF JACOBI

We are going to summarize the general principles of wave mechanics for the case
of a given particle that is acted on by a force field derived from a known potential
function V(x, y, z, t). One should start by recalling some broad features of the
classical mechanics of a point mass.
Under the old conceptions, a particle has at any instant a well-determined po-
sition in space and, in the course of time, describes a certain path, the trajectory,
under the influence of the force field. One can therefore at any time assign to the
particle three rectangular coordinates x, y, z which fix its position. The classical
equations of motion are then

d2 x av
m dt2 = - ax ' ... , (1)

where m is a characteristic constant of the particle, called its mass. These three
second-order differential equations completely define the variation of the coordinates
of the particle in time, that is, its motion, if one assumes that six arbitrary constants,
representing the coordinates and the velocity components at a given instant, called
the initial time, are specified. The determinism of the old mechanics consisted
therein that, with the initial state of position and velocity known, the final states
were rigorously determined.
We go back to the treatises of rational mechanics for a demonstration of the
general theorems in the dynamics of a point mass and for the theory of the equations
of Lagrange, of Hamilton, etc. We shall limit ourselves to stating the fundamental
theorem of Jacobi, which will be useful in what follows.

Note G. L.* Among the numerous treatises, we cite J. M. Souriau, Structure des
systCmes dynamiques (Dunod, Paris, 1970) and H. Goldstein, Classical Mechanics,
2nd edn. (Addison-Wesley, Reading, Massachusetts, 1980). We note that it is not

*Remark: In the notes that follow, Louis de Broglie will be designated by "L. B."
and Georges Lochak by "G. L."
4 Principles of Wave Mechanics

for the sake of simplicity that Louis de Broglie enunciates the theorem of Jacobi in
1R3 , but because the optics-mechanics analogy has a real sense in his eyes only in
physical space, without which the analogy is only a formal one.

Theorem: If one manages to find a complete integral (that is, a solution contain-
ing three arbitrary non-additive constants) S( x, y, z, t, a, (3,,) of the partial differ-
ential equation (equation of Jacobi)

1
2m
[(OS)2
ox + (OS)2
oy + (OS)2]
oz + V(x,y,z,t) =
oS
Bt' (2)

then the equations

oS oS oS
00'. = a, 0{3 = b, 0, = c, (3)

where a, b, c are three new arbitrary constants, define one possible motion in the
force field; and the momentum components of the particle, when during this motion
it occupies at a time t the position x, y, z, are given by the relations
oS oS oS
px = mvx = - ox ' py = mvy = -8'
y
pz = mvz = - oz . (4)

We therefore see that, according to this theorem of Jacobi, the possible motions
of the particle are divided into classes, the motions of anyone class corresponding
to a complete integral sex, y, z, t, a, (3, ,) with given values of a, (3, ,. Each of these
classes contains an infinity of possible motions, everyone of which is characterized
by the values of the secondary constants a, b, c.
Let us also recall that one can obtain Jacobi's equation by starting with the
energy expressed as a function of the coordinates and their conjugate momenta,

H(x, y, Z,Px,Py,Pz, t) = 1 2
2m (Px + Py2 + pz)
2
+ Vex, y, z, t ) , (5)

replacing therein Px,Py,Pz by -oS/ox, -oS/oy, -oS/oz, respectively, and equat-

ing the resulting expression to oS/ ot.
In the case where the potential function V does not depend on the time, the
theorem of Jacobi takes a form that will be particularly useful to us. We know that
in this instance energy is conserved, which means that throughout the motion the
sum of the kinetic energy and the potential energy, (1/2)mv 2 + V, retains a constant
value E. This constant E then plays the role of one of the primary constants of
motion, for example ,. We accordingly set
S(x,y,z,t,O'.,{3,E) = Et - SI(x,y,z,O'.,{3,E), (6)
where SI no longer depends on time, and then seek a complete integral, dependent
on the constant E and on the two arbitrary constants a and {3, of the partial
differential equation (the shortened Jacobi equation)

1
2m
[(OS1)2
ox + (OS1)2 oz )2] + V(x,y,z) = E.
oy + (OSI (7)
 Classical Dynamics of Material Points 5

Applying the theorem of Jacobi to this particular case, we learn that, if such a
complete integral has been found, the motion defined by the equations
aSI _ aSI _ b
aCt - a, a{3 - ,
(8)
as aSI
aE = t - aE = c,
where a, b, c are three arbitrary constants, is one of the possible motions of the
particle in the constant force field and that the linear momentum of the particle, on
passing through the point x, y, z, is given by
aSI aSI aSI
Px = mvx = ax' Py = mvy = ay' pz = mvz = az .
The possible motions are divided into classes corresponding to the same values of
the energy E and of the two primary constants Ct and {3, and every class contains an
infinite number of motions, each characterized by the values of the three secondary
constants a, b, c.
The first two equations (8), not containing the time, define a curve in space
which is the trajectory of the particle. The third equation, which can be written
as aSI/aE = t - to, describes the motion along the trajectory (the time equation).
One thus sees that, in the case of constant fields, the study of the trajectory can
be pursued independently of the study of the motion: This does not happen in the
general situation of fields varying with time.
Another important theorem that is valuable in the case of constant fields is the
following: "Trajectories of the same class that correspond to the same complete
integral SI (x, y, z, Ct, {3, E) are orthogonal to the surfaces SI = const." This theorem
results immediately from the fact, expressed by the equations P = \7 SI, that the
velocity is proportional to the gradient of SI at each point.
The property of trajectories being normal to surfaces SI = const. permits one
to recover Maupertuis's principle of least action. Towards that end, let us consider
all those surfaces SI = const. that correspond to the values infinitesimally close to
the constant lying between the values C! and C2, and let us represent some of them
in a cross section.
Let AEB be a trajectory of the class corresponding to SI and AF B a curve
infinitesimally close to AEB.

If one denotes by dn an element of the normal to the surfaces SI const., the

6 Principles of Wave Mechanics

Jan
integral
aS1 ds

taken along AEB will equal C2 -Cl, since then ds = dn. Let us take the same integral
along the curve AF B. The contribution to the integral from a small element such
as FG is greater than, or at least equal to, the variation of Sl from F to G: As a
matter of fact, if FG is normal to the surfaces Sl = const. that pass through these
extremities, then FG = dn and

whereas, if FG is not normal to the surfaces Sl = const., one gets FG > dn and

results. But all elements of AF B cannot be normal to the surfaces Sl = const.,

unless AF B coincides with the trajectory AEB. Consequently the integral

rB
aS1 ds
JA an
is greater along AF B than along AEB.
According to the equation satisfied by Sl, one has

( as)2
ax1 + (as)2
ay1 + (as)2
az1 = J2m(E - V(x,y,z)). (9)

V'Ve arrive therefore at the following statement: "The particle trajectory passing
through two points A and B in space is characterized by the fact that the integral

L B
J2m(E - V) ds

is smaller for the trajectory than for any neighboring curve." This is the principle
of least action of Maupertuis.
[The above reasoning is flawed when the trajectories have an envelope and the
trajectory AEB touches this envelope between A and E. The integral of Maupertuis
must then be a maximum instead of a minimum, but it remains always stationary.]
A very simple example will allow us to illustrate the previous considerations.
Let us envisage the motion of a particle in the absence of a field. Then V = 0
and, because energy is conserved, one can write the abridged equation of Jacobi as
follows:
(10)
 Wave Propagation in an Isotropic Medium 7
A complete integral of this equation is obtained, for instance, on setting

(11 )

and, according to the theorem of Jacobi, one finds the trajectories

851 ;;:;---;:; [ a ]
800 = v2mE x- (1 _ a 2 _ (32)1/2 Z = a,
(12)
851 ;;:;---;:; [
>:\(3 = V 2mE Y - 2
(3 ]
2 1/2 Z = b.
u (1 - a - (3 )

These are lines, with direction cosines a, (3, (1 - 00 2 - (32)1/2, perpendicular to the
planes 51 = const. The motion along these lines is defined by the equation

85 ax + (3 y + y'1 - a 2 - (32]-
1 m [
8E -- (2mE)1/2 z - t - to· (13)

It is linear and uniform and takes place with a speed v = (2E / m) 1/2. Finally, one
may easily verify the relations

Px == mvx = mav = V2mE a = ~1 , etc.

The complete integral under consideration therefore defines the class of linear uni-
form motions with direction 00,(3" and speed (2E/m)1/2.
One would likewise define the class of linear uniform motions issuing from a point
o with coordinates xo, Yo, Zo by considering the complete integral of the equation
for 5,
m [ (x-xo) 2+(y-yo) 2+(z-zo) 2] .
5=-2t (14)

2. WAVE PROPAGATION IN AN ISOTROPIC MEDIUM

In preparing for the transition to wave mechanics, let us now briefly study
the propagation of monochromatic waves in an isotropic, refringent, and disper-
sive medium.
We shall assume that this propagation is governed by the equation

(15)

'lj; being the wave function and 'V a quantity that generally is a function of the
position coordinates x, y, z and the frequency II of the wave. 'V is the speed of
8 Principles of Wave Mechanics

propagation of the phase or simply the speed of propagation. We shall write the
monochromatic wave in the complex form
'1f; = u( x, y, z) exp (21fivt) (16)
and set
1 n(x,y,z,v)
0/- 'lIo
'lIo being the speed of propagation in a reference medium whose index of refraction
equals unity. One then has

(17)

Rigorously, the propagation of monochromatic waves in a dispersive medium should

be studied by searching for solutions of this equation, but it often happens in practice
that the problem can be solved by an approximate procedure which forms the basis
of geometrical optics.
To well understand the meaning of this approximation, let us first consider
the case where the index of refraction does not depend on x, y, z (a homogeneous
medium). One then obtains a rigorous solution of the foregoing equation by positing

'1f; = aexp {21fiv [t - nta) (ax +,8y + Jl - a 2 _,82 z)]}, (18)

where a is a constant called the amplitude of the plane wave. We shall use the
designation "wave phase" for the linear function

Cf! = v [t - %(ax + ,8y + Jl - a 2 - ,82 z) ]. (19)

The surfaces of equal phase, Cf! = const., also called "wave surfaces," are planes
perpendicular to the direction a,,8, 1= (1 - a 2 - ,82)1/2. In the course of time, the
values of the phase, and consequently of the function '1f;, propagate in this direction
with the speed
(20)

At a given time, one finds the same value of '1f; on planes of equal phase separated
from one another by the distance

(21)

called the "wavelength"; and, at a given point, one recovers the same values of'1f; at
time intervals equal to the period T = 1/ v.
Let us now consider a medium where the index n varies from one point to
another. It will always be possible to write a monochromatic wave in the form
'1f; = a(x, y, z) exp {21fi[vt - Cf!l(X, y, z)]}, (22)
 Wave Propagation in an Isotropic Medium 9

the functions a and 101 being real. One can always define a wavelength ), by the
formula), = %/nv, but this wavelength is "local," in the sense that it varies from
one point to another. If, in some region of space, the index n changes little from
one point to the next, on the scale of the wavelength, then it is readily seen that the
derivatives of a are negligible compared to those of \01; and, on substituting (22)
into the equation of propagation, one obtains an approximate equation, called the
"equation of geometrical optics,"

( 0\0 1)2 + (0\0 1)2 + (0 101 )2 = v 2n 2(x,y,z). (23)

ox oy oz 0/02
It allows us to determine the phase variations without having to worry about the
slow changes of the amplitude a.
Let \01 (x, y, z, v, a, (3) be a complete integral of the equation of geometrical optics.
The function
¢ = aexp {21ri[vt - \01 (x, y, z, v,a,(3))},
where a is a slowly varying function on a large scale, is an approximate solution
of the propagation equation. By definition, the curves orthogonal to the surfaces
101 = const. are the "rays" of the wave. Just as we justified above the principle of
least action of Maupertuis for the trajectories normal to the surfaces SI = const.,
so we could here demonstrate the following principle of Fermat: If the curve C is a
wave propagation ray passing through the points A and B of space, then the integral

{B 0\01 ds = (B nv ds
iA on iA 0/0
taken along the ray C is smaller than the same integral taken along a curve infinites-
imally close to C and joining A to B.
Geometrical optics is a valid approximation only if the index n varies little on
the scale of the wavelength. For vanishingly small wavelength, this approximation
tends to become rigorous.
The presence of the frequency v in the propagation equation (17) should attract
one's attention. Instead of considering a monochromatic wave, one may have to
study the more general case of a superposition of monochromatic waves, each of
which satisfies the propagation equation with a value of n corresponding to the
frequency involved. But then it is desirable to have a form of the propagation
equation that does not involve the frequency and which describes the wave function
even when it is composed of a superposition of monochromatic waves.
Suppose, to give an example, that the index is defined by the dispersion law

n(x,y,z,v) = 1- F(x,y,z) %2
41r 2 V 2 '

where F is some function of position. Then one can adopt, as a general equation of
propagation,
1 02¢
b.¢ - -2!i2 = F(x,y,z)¢,
% vt
10 Principles of Wave Mechanics

because, for a monochromatic wave, one has

and thus will recover Eq. (17). We shall encounter a case of this kind in wave
mechanics. [Note G. L. Concerning geometrical optics, one will profit from reading
references (1, 27) and (I, 29) of the author; see Bibliography at the end of this book.]

3. TRANSITION FROM CLASSICAL MECHANICS

TO WAVE MECHANICS

The striking analogy in form between the theory of Jacobi and that of waves,
already perceived by Hamilton a century ago, can today lead us to the synthesis
realized by wave mechanics.
Let us start by comparing the motion of a particle in the absence of a field
(V = 0) with the propagation of a wave in a homogeneous medium, where the index
n is independent of x, y, z. For a particle not acted on by a field, we found

(11 )
= m v[ax + (3y + \11 - a2- (32 z] .
On the other hand, for a monochromatic wave in a homogeneous medium, the
wavelength ,\ now being constant, one can write the equation of geometrical optics
in the form
'PI =~ [ax + (3y + VI - a 2 - (32 z]. (24)
The complete functions Sand 'P then become

(25)
'P = vt - ~ [ax + (3y + VI - a 2 - (32 z] ,
on making the direction of motion coincide with that in which the wave is prop-
agated. It is in the spirit of quantum theory to set E = hv, that is, to associate
with the motion of a particle of energy E the propagation of a wave possessing the
frequency v = E / h. This leads us to posit

(26)

Note G. L. At that time, Louis de Broglie, under surrounding pressures, deviated

from the strictly relativistic viewpoint of his Thesis, to which he was subsequently to
 Transition from Classical Mechanics to Wave Mechanics 11

return. In reality, only relativity determines v, on fixing the energy constant, which
de Broglie does not indicate, while however attaching a fundamental importance to
it.

If we take this relation as a hypothesis, the two formulas

E= hv, (27)

will result.
In other words, with the linear and uniform motion of a particle of energy E and
momentum mv, we associate the propagation, in its direction of motion, of a plane
monochromatic wave having the frequency E / h and the wavelength h/mv, a wave
expressed by
(a constant),
where S has the value given above.
This correspondence between wave and motion is generalized when one consid-
ers the motion of a particle in a constant field defined by the potential function
V (x, y, z). It then becomes necessary to compare the motion with the propagation
of a wave in a nonhomogeneous medium, where the index n and consequently the
wavelength). vary from one point to another.
The expressions for comparing Jacobi's function S and the phase r.p then are

S=Et- S l(X,y,Z),
(28)
r.p = vt - r.pl (x, y, z),
the functions SI and r.pl being, respectively, the complete integrals of the equations

( 8S
8x
1 )2
+ (8S1)2
8y + (8S
8z
1 )2
= 2m[E - V(x, y, z)]'
(29)
( 8r.p1) (Or.p1)2 (Or.pl)2
2 1
ox + oy + oz = ).2(x, y, z) .
It is quite natural here again to advance the hypothesis expressed by r.p = S / hand
consequently to set E = hv and SI = hr.pl. The second formula easily gives
1 h h
).------- . (30)
- 1\7r.pll - I\7S11 - (2m[E _ V(x,y,z)])1/2'

and, since at each point one must have E = (1/2)mv 2 + V(x, y, z), we find again
h
).= - ,
mv
but here v and), are variable from one point to another.
12 Principles of Wave Mechanics

How is the propagation equation for the wave corresponding to motion III a
constant field to be written? Let us write Eq. (17) in the form

47r 2
Ll'ljJ+ \2( )'ljJ=0 (31)
A X,Y,Z

and substitute herein the value of Ai the equation becomes

87r2 m
Ll'ljJ + ~ [E - V(x,y,z)]'ljJ = O. (32)

On putting V = 0, one recovers the equation that is valid in the absence of a field.
Whenever geometrical optics is valid for the propagation of the wave 'ljJ, we are
able to write

'ljJ = aexp ( h27ri S) = aexp {27ri

h [Et - SI(X,y,Z)] } ,

and the trajectories predicted by the old mechanics of a material point, normal to
the surfaces SI = const., will be nothing other than the rays of the wave 'ljJ normal
to the surface CPl = const.
We thus arrive at one of the essential ideas of the new mechanics. Whereas
the old mechanics attributed to its equations a rigorous character and viewed them
as always valid, the new mechanics assigns the essential role to the wave 'ljJ: It
regards the old mechanics as a valid approximation only when the geometrical optics
approximation suffices to describe the propagation of the wave 'ljJ.
Classical mechanics is therefore nothing but an approximation: It can be used
only when the index n for the wave 'ljJ varies little on the scale of the wavelength or,
which is the same, when the potential varies little on this scale. If the wavelength
of the wave 'ljJ were infinitely small, the old mechanics would be rigorously valid.
According to the formula (27) for A, it is seen that this would always be realized (for
nonzero v) if h were infinitely small: For h - t 0, classical mechanics must always
regain its validity.

4. GENERAL EQUATION FOR THE WAVE MECHANICS

OF A POINT MASS

Above we were led to substitute the propagation equation of a monochromatic

wave in the equations belonging to the classical mechanics of a point mass in a
constant field. But, as will soon emerge, we shall often be led to consider wave
trains 'ljJ formed by a superposition of monochromatic waves. It is therefore useful
to try and obtain a propagation equation which will describe the function 'ljJ when
it 'represents such a superposition of monochromatic waves. The equation

_I. _ 87r 2 m V( )_1. = 47rim f)'ljJ

at
A
iJ.'I-' h2 X,y,Z'I-' h (33)
 Automatic Procedure for Obtaining The Wave Equation 13

satisfies this requirement, because it recovers Eq. (32) for a plane monochromatic
wave of frequency Elh. [Note G. L. The author has the habit of using the complex
conjugate of the customary equation.) But this new form of equation frees us from
any restriction to plane monochromatic waves and allows one to consider a superpo-
sition of such waves. It moreover suggests the manner in which the new mechanics
is to be extended in the case of fields varying with time. Indeed, since we are no
longer confined to monochromatic waves, time ceases to playa special role, and it
is therefore natural to permit the form of the equation to remain unaltered when V
depends on time, i.e., to write

(34)

as the general form of the wave equation for a single particle in nonrelativistic wave
mechanics.

5. AUTOMATIC PROCEDURE FOR OBTAINING

THE WAVE EQUATION

We are going to present a formal method for automatically recovering the wave
equation.
In classical mechanics one gives the name "Hamiltonian function" to the function
that expresses energy in terms of the coordinates and the Lagrangean momenta.
Relative to rectangular coordinates, the well-known expression for this function is

1
H(x, y, Z,Px,Py,Pz, t) = 2m (Px
2
+ Py2 + pz)
2
+ V(x, y, z, t). (35)

If, in the second term of this expression, Px, Py, pz are replaced by

h fJ h fJ h fJ

respectively, one obtains an operator, the Hamiltonian operator

h . : '--2
H(x,y,z, - -2
~z ux
h . : '--2
~z uy
h.
~z uZ
,t) !
(36)

On applying this operator to the function 'lj; (that is, multiplying 'lj; from the left
by the Hamiltonian operator) and equating the result to
14 Principles of Wave Mechanics

one gets
1 ( h
2m 21ri
)2 /::"'lj; + V(x, y, z, t)'lj; = 21rih at'
a'lj;
(37)
an equation identical with the general equation (34) obtained above.
Vie thus see that the general propagation equation can be written in the form

(38)

where P x , P y, P z are, respectively, the operators

h a h a h a
- 21ri ax' - 21ri ay' - 21ri az '

which we have made to correspond to the components of the momentum.

It is important to note that the indicated automatic procedure for obtaining
the wave equation will not succeed in general if one uses curvilinear coordinates.
Thus, in spherical coordinates one will not find in this way the correct form of the
Laplacian operator /::,. appearing in the equation. This difficulty has its origin in the
fact that one cannot unequivocally deduce from the classical Hamiltonian function
the form of the Hamiltonian operator, because a term of the form qPq, for example,
in the classical function can give rise, depending on the order of its factors, to terms
qPq, Pqq, or
~(qPq + Pqq),
which are not equivalent expressions.
 CHAPTER 2

PROBABILISTIC INTERPRETATION
OF WAVE MECHANICS

1. INTERPRETATION OF THE 1/1 WAVE

We have obtained the general equations of wave mechanics. Now we must learn
how to use them and, in particular, what meaning to attribute to the function "p.
H one were guided by classical analogies, one would be led to consider the func-
tion "p as representing a physical quantity that is perhaps tied to the vibration of
some medium. One circumstance alerts us straight-away that such an interpreta-
tion is impossible: The general equation containing among its coefficients the factor
i = V-I, the wave function must be regarded as essentially a complex quantity,
contrary to what happened in the classical theory of waves and vibrations, where
the use of complex functions always appears as a simple mathematical artifice.
In wave mechanics, the wave function does not appear as giving the value of a
physical quantity but as constituting a "predictive element," with whose aid one is
able to evaluate the probability that particular results of measurement will occur.
The function "p is complex, but, as will be shown, one can from it construct real
quantities which do have a physical significance in terms of probabilities. The fact
that the function "p is essentially connected with probabilities explains why, as we
shall see, its value is never fully determined: In the first place, "p always contains
a phase factor exp (in), which disappears when one constructs the real probabilis-
tic quantities and consequently is of no importance. Furthermore, the modulus of
"p is determined only up to constant factor, and one takes advantage of this inde-
terminacy, as will be seen, to "normalize" the wave function, thereby allowing the
probabilities to be expressed "in absolute value." All this would be incomprehen-
sible if "p represented a vibration of a physical nature, because then its amplitude
and its phase would have well-determined values. We shall later return to some
particular characteristics of the function "p.

Note G. L. All of the foregoing paragraph the author would later have written
with much greater care. What de Broglie says here is true only for Schrodinger's
continuous and normalized wave function, but he subsequently again took up the
idea of the double solution, according to which one should associate with each
probabilistically significant continuous solution a singular solution of the same phase
but having an amplitude that allows for a singular region representing the particle.
16 Probabilistic Interpretation of Wave Mechanics

2. PRINCIPLE OF INTERFERENCE

In order to use our knowledge of the function 1/J, wave mechanics quickly came
to enunciate a basic principle which we shall call the "principle of interference" or
the "principle of localization." It asserts:
"The modulus squared of the function 1/J measures, at every point and at every
instant, the probability for observing at that time the presence of the particle at
that point."
The function 1/J, being a complex quantity, can be written in the form 1/J =
a exp (i~), where a and ~ denote its modulus and its argument, respectively, both
of which are real quantities that generally are functions of x, y, z, t. Let us des-
ignate by 1/J* the quantity a exp (-i~), i.e., the complex conjugate of 1/J; then one
has
(1)
It is this real quantity that appears in the formulation of the principle of interfer-
ence.
It is easy to relate the interference principle to some classical ideas in the the-
ory of light. In all theories of light one assumes that the intensity of a light wave
at any point and at any time measures the amount of energy that can at that lo-
cation be collected at that instant: It is this rule that allows an exact prediction
of interference phenomena. But today we know that, in the energy exchanges be-
tween matter and light, everything happens as if light were composed of corpuscles
having the energy hv. These are the "photons." If we imagine a light wave as
sweeping along a large number of photons, then the foregoing explanation of inter-
ference requires the wave intensity to measure the photon density at every point.
But this "statistical" interpretation does not suffice and must be transformed into
a "probabilistic" interpretation. Indeed, one has been able to obtain (in the ex-
periments of Taylor and of Dempster and Batho) interference patterns of the usual
kind even on using, over an extensive period of time, a very weak light intensity,
so weak in fact that there could never have been more than one photon at a time
in the interference apparatus. Moreover, we shall see that it is not really possible
to attribute to a particle a well-defined position in space. We are thus necessarily
led to say: The intensity of a light wave measures the probability for a photon to
produce an observable effect at the given point in space. In this way one will exactly
recover the classical expressions of interference even in the case of extremely weak
illumination.
An extension of the principle of interference from light to material particles
is justified by the fact that one can observe the phenomena of interference and
diffraction for material particles just as one can for light. For example, for electrons
that are easily employed in experiments (electrons with energy ranging from tens
to hundreds of thousands of electron volts), the associated wave has, according
to the formula A = h/mv, a wavelength of the order 10- 8 to 10- 9 cm. With such
electrons it must therefore be possible to obtain diffraction phenomena analogous to
those observed with x rays whose wavelength is of the same order. This is precisely
what was shown in 1927 by the famous experiments of Davisson and Germer and
shortly thereafter by the findings of G. P. Thomson, Rupp, Ponte, Kikuchi, and
 Precise Statement of the Principle of Interference 17
others.1 These experiments prove that a mono-energetic beam of electrons can, on
diffraction by a crystal, give rise to phenomena entirely analogous to those observed
with x rays (the Laue-Bragg experiments). Rupp was even able to obtain diffraction
of electrons with an ordinary grating under grazing incidence; and in 1940 Borsch,2
repeating a fundamental experiment of Fresnel on light, was successful in achieving
the diffraction of electrons at the edge of a screen. All these experiments allow
us to obtain an excellent confirmation of the general ideas of wave mechanics and
in particular of the formula A = h/mv. They also provide decisive support for
the notion that it is appropriate to extend the principle of interference to material
particles, since this principle underlies all interpretations in the realm of interference
and diffraction.

3. PRECISE STATEMENT OF THE PRINCIPLE

OF INTERFERENCE. PROBABILITY FLUID

To further clarify the principle of interference, we remark that the wave 'IjJ, which
is the solution of a partial differential equation and does not have the character of
a measurable physical property, is determined only up to a multiplicative constant,
a .quantity that can be complex. We may choose this factor such that

/ / / 'IjJ'IjJ*dr = 1, (2)

the integration extending over all of space. At the very least, the choice of the
arbitrary factor allows us to "normalize" 'IjJ by the foregoing relation at a given
instant to, and we are going to show that then this function will remain normalized
at all times t. One can thus state the principle of interference more precisely by
saying: "The probability that an observation will allow the localization of a particle
whose normalized wave function is 'IjJ(x, y, z, t) within a volume element dr at the
moment t is equal to

'!fJ(x,y,z,t)'IjJ*(x,y,z,t)dr = 1'!fJ(x,y,z,t)1 2 dr."

In order to visually represent the variations in time of the position probability
density 1'!fJ1 2, we may imagine a fictitious fluid 2 whose density, per definition, at
every point and at every instant is furnished by

p(x, y, z, t) = 'IjJ(x, y, z, t)'IjJ*(x, y, z, t). (3)

We define the motion of this fluid by postulating that its velocity at any point x, y, z
at the instant t is given by the formula

1. C. J. Davisson and L. H. Germer, Phys. Rev. 30,705 (1927); G. P. Thomson,

Nature 120, 802 (1927); E. Rupp, Naturwiss. 16, 556 (1928); M. Ponte, C. R.
Acad. Sci. 244, 909 (1929).
2. M. Borsch, Naturwiss. 28, 709 (1940).
18 Probabilistic Interpretation of Wave Mechanics

v = _1___ ~_ (1/;V1/;* -1/;*V1/;) = __h_V<p. (4)

1/;1/;* 4nm 27rm

Note G. L. One speaks of this fluid as "Madelung's fluid." Subsequently it will

playa major role in the causal interpretation of wave mechanics.

But the functions 1/; and 1/;* obey the complex conjugate equations

87r 2 m 47rim EN
b.1/; - ~ V(x,y,z,t)1/; = -h- at'
(5)
* 87r 2 m * 47rim a1/;*
b.1/; - ~ vex, y, z, t)1/; = --h- Tt,

from which one can easily deduce

1/;* b.1/; -1/;b.1/;* = 47rim !!... (1/;1/;*) = 47rim

ap (6)
h at hat'
which may be rewritten as

ap = ~(1/;* b.1/; -1/;b.1/;*) = __h_. " ~ (1/; a1/;* _1/;* a1/;) (7)
at 47rtm 47rzm X,y,Z
6 ax ax ax

or as
ap
at+V,(pv)=O. (8)

This equation, which is well known in hydrodynamics under the name of "equa-
tion of continuity," expresses the fact that the fictitious fluid of density p is conserved
in the course of time, i.e., that the integral

remains constant. The normalization of 1/; therefore has a permanent character.

4. THE UNCERTAINTY RELATIONS OF HEISENBERG

The old mechanics assumed that it is possible to attribute to each particle a

well-defined position and velocity: In other words, one assigned to the coordinates
x, y, z of any particle, just as one did for its energy E and its linear momentum
p = mv, values that are well determined at every instant. We shall see that the
same cannot be done in wave mechanics.
Let us study the simple case of uniform rectilinear motion outside any field. We
know that to a uniform rectilinear motion of energy E and momentum p taking place
 The Uncertainty Relations of Heisenberg 19

in a direction given by the direction cosines a, f3, VI - a 2 ,- f32 there corresponds a

plane monochromatic wave

tjJ = aexp {2~i [Et - J2mE( ax + f3y + VI - a 2 - f32 z)]} (9)

of frequency E/h and wavelength h/mv. This monochromatic wave thus corre-
sponds to a well-determined state of motion, but it gives no indication about the
position of the particle, because the wave is homogeneous, i.e., of the same ampli-
tude at every point in space, thereby causing the position probability density tjJtjJ*
to be the same for all points.
However, instead of being a plane monochromatic wave, the solution tjJ of the
wave equation appropriate to the state of a particle can be superposition of plane
waves, representing a wave train of limited extension. Then the density tjJtjJ* will
be different from zero only within a limited region of space, and thus, according to
the interference principle, the particle can be detected nowhere but in this region.
The uncertainty in position is therefore not as large as in the case of the plane
wave. On the other hand, if with each monochromatic component of frequency II
and wavelength ..\ we associate a state of motion defined by

(10)

one will no longer be able to attribute to the particle a well-defined state of motion.
In passing from the case of the plane monochromatic wave to that of a finite wave
train, we have thus diminished the uncertainty in the position, but we have also
augmented the uncertainty about the state of motion. We can now go on to the
limiting case of a wave train having infinitely small dimensions. In the analytical
representation of such a wave train it is necessary to introduce a superposition
of monochromatic waves that comprises all possible frequencies, wavelengths, and
directions of propagation. This limiting case, symmetrically opposed to that of the
plane monochromatic wave, corresponds to a well-determined localization of the
particle but to a complete ignorance concerning its state of motion.
In summary, the better the position of a particle is defined, the larger is the
uncertainty about its state of motion, and conversely. We thus arrive at a first qual-
itative statement of Heisenberg's uncertainty relations, which we shall now further
sharpen and clarify.
To this end, let us study the representation of a wave tjJ by a superposition of
plane monochromatic waves. If we put

E
II = h' J-lx = hPx = ha~
2mE, J-lz = P; = ~ J2mE,
(11 )
then the corresponding monochromatic plane wave can be written as

aexp [21l'i(lIt - J-lxX - J-lyY - J-lzz)]. (12)

20 Probabilistic Interpretation of Wave Mechanics
It is possible to represent the wave 1jJ by a Fourier integral

1jJ(x,y,z,t) = 111 a(J1-x,J1-y,J1-z)

(13)

wherein one must set

(14)
The coefficients a(J1-x, J1-y, J1-z) are in general complex, i.e., they contain a factor of
the form exp (ia), because the various monochromatic components occurring in the
integral expansion of 1jJ do not all have the same phase.
Let us now envisage the wave train 1jJ at an instant which may be taken as the
initial time t = o. The function

(15)
X exp [-27ri(J1-xx + J1-yY + J1-zz)] dJ1-x dJ1-y dJ1-z
must differ from zero only inside a limited region R. We designate by the symbols
~x, ~Y, ~z the maximum variations of the coordinates in R, i.e., the side lengths
parallel to the coordinate axes of the circumscribed parallelepiped of R. We may
locate the origin of our coordinate system at one of the vertices of this parallelepiped,
so that x,y,Z in R will vary within the ranges (O,~x), (O,.6y), (O,.6z).
The theory of Fourier integrals furnishes the relation

(16)
X exp [27ri(J1- x x + J1-yY + J1-zZ )]dx dy dz.
Since not all the a's in (16) can be infinitely small, there is at least a set of J1-
values, say J1-~, J1-g, J1-~, for which a(J1-~, J1-g, J1-~) has an appreciable value. Let us vary
/1x,f.t y, /1z by tiJ1-x, tiJ1-y, tiJ1-z, starting from J1-~, J1-~, J1-~, such that these variations are
not necessarily infinitesimally small. One gets

a(J1-~ + tiJ1-x, J1-~ + tiJ1-y, J1-~ + 6J1-z) - a(J1-~, J1-g, J1-~)

xexp [27ri(J1-~x+J1-~Y+J1-~z)]dxdydz.
The exponential function between braces in the foregoing integral can differ
sensibly from unity only if one or more of the products 6J1-x.6x, 6J1-y.6y, 6J1-z.6z
 The Principle of Spectral Decomposition (Born) 21
exceed a fraction ." that is not very small compared with unity. Therefore,
if simultaneously all three of these products are less than or equal to 'fJ, then
a({L~ + O{Lx, {Lg + OILy, {L~ + O{Lz) will differ little from a({L~, {Lg, {L~) and will thus,
by hypothesis, have an appreciable value. One can therefore say that the extent of
the domain of variation of the three parameters {Lx, {Ly, {Lz in the Fourier represen-
tation (15) of the wave train 'IjJ is measured by three quantities b..{Lx, b..{Ly, b..{Lz that
satisfy the inequalities

or, by definition of {Lx, {Ly, {Lz,

(17)
the inequalities being valid in regard to orders of magnitude. We have thus obtained
the uncertainty inequalities of Heisenberg: They teach us that the product of the
uncertainty in one coordinate and the uncertainty in the conjugate component of
the linear momentum is always at least of the order h.

5 .. THE PRINCIPLE OF SPECTRAL DECOMPOSITION (BORN)

In the arguments that have just been presented, we have implicitly assumed
a principle which it is now important to state clearly. This principle, which was
imposed during the development of wave mechanics and was first formulated by
Max Born, may be stated as follows: "If the wave 'IjJ is formed by a superposition
of a certain number of plane monochromatic waves, then each of these components
corresponds to a possible state of motion of the particle, that is, an observation or
measurement can allow the attribution of this state of motion to the particle." More
precisely, one may say with Born: "If the wave 'IjJ is constructed by a superposition
of plane monochromatic waves forming a discrete spectrum, that is, if one has

'IjJ = L a(a,{3, E) exp {2:i [Et - .../2mE (ax + {3y + ~!I - a 2 - {32 z)]}, (18)
OI,{3,E

then the probability that a measurement will lead one to attribute to a particle a
motion of energy E in a direction having direction cosines a, {3, VI -
a 2 - (32 is
a(a,{3,E)a*(a,{3,E) = la(a,{3,E)1 2 ." If the wave 'IjJ is constructed by a superposi-
tion of plane waves forming a continuous spectrum (which is the case for the usual
wave trains), that is, if one has

'IjJ = fff a(a,{3,E)

(19)
X exp {2:i [Et - .../2mE ( ax + {3y + VI - a 2 - (32 z) ]}da d{3 dE,
22 Probabilistic Interpretation of Wave Mechanics

then the probability that a measurement will assign to a particle a motion of energy
between E and E + flE, and taking place in a direction defined by the intervals
(0:, 0: + flo:) and ((3, (3 + fl(3), is equal to

J11
r[ [
D.cr,D.{3,D.E
la( 0:, (3, E) 12 do: d(3 dE.

One can therefore say that the probability for each state of motion is measured
by the intensity of the corresponding spectral component. The states of motion not
appearing in the Fourier development of the wave function thus have a probability
zero: Hereon, we shall see, is based the theory of quantum states in wave mechanics.
We have formulated the principle of spectral decomposition for the simple case
where a field is absent. Soon we shall get to know a general principle which is
applicable in all situations, the generalized principle of spectral decomposition, of
which Born's principle and even that of interference are only special cases.

6. NEW IDEAS RESULTING

FROM THE PRECEDING CONCEPTIONS

The foregoing considerations allow us to give a more precise meaning to the wave
'if; and to link it to completely novel ideas.
The wave 'if; is not a physical quantity in the classical sense: It is a predictive
tool. Its form results from earlier observations, providing us with some information
regarding the initial state of the particle, and from its evolution in accordance with
the wave equation since these last observations. Although the evolution of the wave
'if; is entirely determinate, we shall find that it does not provide a rigorous prediction
of future observations, because knowing the wave 'if; does not allow one to forecast
which particular value of a given quantity will occur in a new observation, but only
what the possible values of the quantity and their respective probabilities are.

Note G. L. It is precisely for this reason that de Broglie in 1927 introduced

the idea, whereto he was going to return in the year following this manuscript,
according to which there must exist two solutions of Schrodinger's equation which
are interrelated but not identical: the one physical, the other statistical.

Every time that fresh observations provide us with new knowledge about the
state of a particle, the form of the wave 'if; is thereby modified: This is easy to
understand if one clearly realizes that the wave 'if; is only a representation of our
present knowledge about the state of the particle, not a representation of an objec-
tive reality.
We shall see that observations carried out simultaneously in the course of the
same experiment can never allow us to gain a knowledge of the quantities belong-
ing to a particle more precise than that permitted by the Heisenberg uncertainty
inequalities. Some (one could say half) at least of the quantities characterizing the
particle are at any time affected by uncertainty. If we measure exactly the values
of certain quantities, then the values of the canonically conjugate quantities will
 Return of Wave Mechanics to Classical Mechanics 23

remain totally unknown. There exist therefore "maximal" measuring experiments

which furnish the greatest knowledge one could have about the state of a particle
without however letting us completely know it. If there existed experiments that
permitted us to know exactly all the quantities associated with a particle, then
Heisenberg's uncertainty relations would evidently no longer by satisfied; and it fol-
lows by the foregoing reasoning that, after an experiment of this kind, we could no
longer represent the state of our knowledge by a wave 'IjJ. But we shall see that
no such experiment can be realized, and that indeed because of the existence of
the quantum of action. All these considerations will be rendered clearer and more
precise in what follows.

7. RETURN OF WAVE MECHANICS TO CLASSICAL MECHANICS.

THE THEOREM OF EHRENFEST. GROUP VELOCITY

We now want to show how one can from the wave-mechanical point of view
justify the success of classical mechanics in the macroscopic domain. A basic method
introduces the theorem of Ehrenfest which we shall now explain.
Consider again the probability fluid of density p = 1'IjJ 12. For a wave train it
occupies a finite region R, and one can define its "center of mass" by the intuitive
formulas

(20)

More generally, we define the mean value of a function f(x, y, z) in the probability
fluid as the quantity
(21 )

With reference to these definitions, the theorem demonstrated not so long ago by
Ehrenfest now reads as follows:
"The center of mass of the probability fluid, with coordinates x, 'fl, z, moves in
the course of time in the same way as a material point of mass m would by the laws
of classical mechanics when subjected to a force 7."
Indeed, one finds, on employing the wave equation and performing some inte-
grations by parts (it being assumed that the function 'IjJ is sufficiently regular and
zero at the limits of R):

dx
dt
= r/J'IjJ'IjJ*
}R at
dr = __h_ { x ""' ~ ['IjJ o'IjJ* _ 'IjJ* o'IjJ] dr
47rim } R ox
~ ox ox
X,Y,Z
(22)
24 Probabilistic Interpretation of Wave Mechanics

arid thus

which, by virtue of the propagation equation, can be rewritten as

Two further integrations by parts eliminate two of the integrals appearing here on
the right:

k (8:fx* ~1jJ + ~: ~1jJ* )dT = k [8:fx* ~1jJ + 1jJ* ! (~1jJ)] dT

= k:x [1jJ* ~1jJldT = 0,

and there remains

m ~~ = k! V (1jJ1jJ*)dT = - k~~ 1jJ1jJ* dT,

that is,
(23)
Analogous equations being obtainable also for the y and z directions, the theorem
of Ehrenfest thus results.
Let us now consider a macroscopic experiment that permits us to observe the
motion of a particle, sayan electron. The wavelength of the wave 1jJ is always
extremely small on our own scale of observation, and one may thus envisage a
wave train whose dimensions are tiny on this scale (quasi-punctual wave train) but
nevertheless large in relation to the wavelength. The macroscopic force field to
which the particle is subjected will therefore always vary very little inside a wave
train, so that 7 will at all points of this region have a value approximately equal
that of the force at the center of the wave train.
Seeing then that one can macroscopically identify the wave train with its center
of mass and that the particle manifests its presence only inside the wave train, we can
describe the resulting situation, according to Ehrenfest's theorem, by saying that the
particle experiences a motion predicted by classical theory. Assuredly, a microscopic
experiment would show that the particle can have any position inside its wave
train, but macroscopically all these possible positions are indistinguishable from
one another, since the wave train appears pointlike on our scale of measurement.
 Return of Wave Mechanics to Classical Mechanics 25

We can treat the same question from another point of view by employing the
theorem of group velocity.
Let us first recall that a group of waves is a wave train that can be represented
as a superposition of plane monochromatic waves having frequencies, wavelengths,
and directions of propagation very close to one another. One can therefore approxi-
mately attribute to it one frequency, one wavelength, and one direction of propaga-
tion, although rigorously it is not equivalent to a monochromatic wave. The wave
group has limited dimensions, because the different waves that are superposed in
phase at the center of the wave train get to cancel each other through destructive
interference outside the limits of the group. It is easy to prove that the dimensions of
a wave group are always large in comparison with its mean wavelength >'0. Indeed,
if various components are in phase at the center of a wave group formed by the
superposition of waves having wavelengths lying in the interval (>'0 - 6.>', >'0 + 6.>'),
with 6.>' ~ >'0, such that the components cancel through destructive interference
outside the space occupied by the group, then it is necessary that the phase differ-
ence between the waves of wavelengths >'0 and >'0 ± 6.>' becomes at least 7r /2 when
one goes from the center of the group to any of its limits. If d denotes the distance
from the center to a limit, one must have
d d d6.>' 7r
i.e., q.e.d.
>'0 >'0 + 6.>' ~ >'5 IV 2'
Let us now derive Lord Rayleigh's formula for the group velocity. In a medium
of variable index, a monochromatic wave of frequency Vo could, in the geometrical-
optics approximation, be represented by aexp {27ri[vot - <PI(X,y,Z,vo)]}, the
function <PI being a complete integral of the equation of geometrical optics. A
group of waves will thus have the mathematical form

.,p = l Vo +ll. V

vo-ll.v
a(v)exp {27ri [vt - <PI(X,y,z,v)] }dv, (24)

Let us set v = Vo + 'f/, with 'f/ varying between -6.v and +6.v. One can then write
approximately

(25)

where (8<pl/8v)0 is the partial derivative of <PI with respect to v for v = Yo. The
integral in this formula being a function of the parameter t - (8<pl/8v)0, Eq. (25)
can be restated as

(26)
26 Probabilistic Interpretation of Wave Mechanics

The wave train accordingly behaves approximately like a monochromatic wave

whose amplitude is a function of t - (Ocpl / ov)o. One can see that this approximation
ceases to be valid for times that are very long.
On moving along a ray, i.e., a curve orthogonal to the surfaces CPl = const., in
such a manner that dt - (o2cpI/ avos )ds equals zero, one will always be accompanied
by the same value of the amplitude. We can therefore say that, during any time
which is not too long, the wave group moves as a whole along the path of the rays
with a velocity
u = ds = (o2CPl )-1. (27)
dt avas
But we have seen that ocpI/ as = !Vcpl! at every point is equal to the reciprocal of
the local wavelength A(X, y, z, v)j therefore

~ = ~ (~) = o(v/o/) = ~ a(nv). (28)

u av A ov 'lIo ov
This is the formula of Lord Rayleigh, which furnishes at every point the velocity of
the group. If the medium is homogeneous, U will be independent of x, y, z. If it
is moreover dispersionless (on/ov = 0), one gets U = 0/, i.e., the group velocity
becomes identical with the wave velocity.
Let us apply Lord Rayleigh's formula to the propagation of the waves 'IjJ in wave
mechanics. For a particle moving in a force field derivable from a potential function
V(x,y,z), we earlier [Chap. 1, Eq. (30)) found
h
A= 1/2' with E = hv,
[2m(E - V(x,y,z)))
whence
o(l/A) (1/h)oy'2m(E - V) m 1
a;;- = (l/h)oE = [2m(E _ V))1/2 v'
because y'2m(E - V) = mv. The formula of Rayleigh thus gives U = v.
From this follows an important theorem on group velocity in wave mechanics:
"The velocity of a wave group 'IjJ associated with a particle equals the particle
velocity corresponding to the central frequency of the wave group."
Let us now return to the connection between classical mechanics and wave me-
chanics in the microscopic domain. In this domain, the fields and, consequently,
the index of refraction of the wave 'IjJ vary little on the scale of the wavelength.
Furthermore, the wavelengths being very small, we may study wave groups that are
nearly pointlike on our scale. Consider then the propagation of a monochromatic
wave corresponding to the central frequency Vo of the group. One will have a set of
surfaces of equal phase CPl(X, y,z, vo) = const. and the rays, or orthogonal curves,
of these surfaces.
On the macroscopic scale, the wave group is analogous to a small globule that
slides along the length of a tube formed by rays. On the microscopic level of the
wavelength, the wave group would in its central part have the appearance of a
 Return of Wave Mechanics to Classical Mechanics 27
monochromatic wave, and it is only at the edges of the group that the interference
of its diverse components would cause its intensity to rapidly fall to zero. The
wave train propagates parallel to the rays with a velocity U which would be that
of a classical particle in the given field. Because we are unable on a macroscopic
scale to distinguish the different points of a wave group that appears pointlike to
us, and since the particle can manifest itself only inside the group, one gains the
impression of dealing with a point particle impelled by a classical motion. We see
that in this way one also recovers exactly the conclusions drawn earlier from Ehren-
fest's theorem. The theorem of Ehrenfest and that of group velocity are intimately
related, and either one allows us to establish a connection between wave mechan-
ics and classical mechanics in the case of macroscopic phenomena for which the
propagation of the wave 'Ij; can be described by the approximation of geometrical
optics.
 CHAPTER 3

WAVE MECHANICS
OF
SYSTEMS OF PARTICLES

1. OLD DYNAMICS OF SYSTEMS OF POINT MASSES

Up to now we have considered a single particle placed in a known force field.

How does one generalize the method presented above to the case of a system of
mutually interacting particles? To find out, we must first recall a few facts about
the classical mechanics of a system of material points.
Let us consider a system formed of N particles. The mass of the particle i is mi,
and its coordinates are Xi, Yi, Zi. The kinetic energy of the system is
N
~ (dZ i )2].
T =
2 Li m
1
I
[(dXi)2
dt +
(dYi)2
dl + dl
(1)

The momenta conjugate to the three coordinates are

dx; dYi dZi
Px; = miTt, PY. = mi Tt' Pz; = miTt· (2)

The potential energy V(Xl, ... ,ZN,t) of the system is composed of two kinds
of terms: (1) those that express the mutual interaction between the particles
and are assumed to depend only on there distances apart; they arc of the form
V;j ( J( Xi - Xj)2 + (Yi - Yj)2 + (Zj - Zj)2); (2) those that express the possible ac~
tion of an external field on everyone of the particles; they are of the form
V;(Xi, Yi, Zi, t).
The Hamiltonian, which furnishes the energy of the system as a function of its
coordinates and momenta, is
N
H(Xl'"'' ZN, t) = L z. -2m"
1
(Px
2
t
+ p2y . + pz·)
t
2
t
+ V(Xl,"" ZN, t). (3)
1 I

If the external field does not depend on the time (or is absent), V is independent of
t, and we know that H will then maintain a constant value E during the course of
the motion; the system is conservative.
 Ola Dynamics of Systems of Point Masses 29
The theory of Jacobi can be extended to systems of particles. Jacobi's equation
for a system is

N 1 [(as)2 (as)2 as 2] as
~i2mi aXj + aYi +(az) +V(XI, ... ,ZN,t)=(-mJ (4)

If one succeeds in finding a complete integral of this equation containing 3N ar-

bitrary nonadditive constants al, ... , a3N, one will obtain a possible motion on
writing
as(Xl' ... ' ZN,
a t, aI, .•. , a3N) = a,., i = 1,2, ... ,3N,
(5)
ai
where the ai are 3N new arbitrary constants and Lagrange's momenta are given by
the formulas
as as
PYi = --a' PZi = --a' i = 1,2, ... ,N. (6)
Yi Zi

In the particular case where the external forces are independent of the time
(or zero), V does not depend on t, and one can find solutions of the form S =
Et - Sl(Xl, ... ,ZN). We are then led back to a consideration of the "shortened"
Jacobi equation

(7)

and to a search for a complete integral containing 3N arbitrary nonadditive con-

stants E, aI, ... , a3N-l. The equations of motion then are
aS
l
--
aai
= ai, i = 1,2, ... , 3N - 1,

the equation of the trajectory pursued by the representative point through the
configuration space of xI, ... , ZN, and the "timetable equation"
aSl
aE = t - to;

and one has

aSl aSl aSl
PXi = -a
Xi
' PYi = -a
Yi ' PZi = -a
Zj
.
As in the case of a single point mass, the equation of Jacobi allows us to define
"classes" of motion of the representative point of the system in configuration space,
each class corresponding to a function Sl(XI, ... , ZN, E, al, ... , a3N-l) with given
values of the constants E, aI, ... , a3N -1, and the various motions of the same class
being characterized by the values of the constants aI, . .. , a3N -1 and to.
30 Wave Mechanics of Systems of Particles

2. WAVE MECHANICS OF SYSTEMS OF PARTICLES

To obtain a wave mechanics for any particle system, one must, as Schrodinger
showed, consider the propagation of a wave in the configuration space of this system;
and, in order to recover classical mechanics in a first approximation, the geometrical
optics of this propagation must lead us to the theory of Jacobi.

Note G. L. This is what de Broglie refused to admit in 1926 (see Ref. I, 29),
considering that the waves associated with different particles of the system "have a
physical reality and must be expressed by functions of the three spatial coordinates
and of time." In 1927 (Ref. I, 34) he made a first attempt to reconstruct the theory
of systems in physical space, an attempt which he was to resume 25 years later with
Andrade e Silva. But, at the time he wrote the present text, he was resigned to
adopting, without further criticizing it, the viewpoint that had become customary.

We assume that the propagation equation in configuration space can be discov-

ered by the same formal procedure that succeeded in the case of a single particle.
One starts from the classical Hamiltonian function H(XI, . .. , ZN ,PXI" .. ,PZN , t) ap-
propriate to the envisaged system and transforms this expression into an operator
through replacing the momenta PXk' PYk' PZk by

(8)

One therefore obtains the Hamiltonian operator

H(XI, ... 'ZN'-~aa

2n Xl
""'-~aa
27fZ Z N
,t)
and adopts as propagation equation

h 'aa ""'-~aa ,t)~=~aa~'

H(xr"",ZN'--2 (9)
7fZ Xl 2n ZN 2n t
Accordingly, one finds

For N = 1, the equation valid for a single particle is recovered.

For conservative systems (av/at = 0), one may consider monochromatic solu-
tions that depend on the time only through the factor exp [(27fi / h )Et] and for which
Eq. (10) can thus be written as

1
L
N 87f2
-6.k~ + -h2 [E - V(XI, ... ,ZN)]~ = O. (11 )
I k mk
 Interpretation of Wave Mechanics for a System of Particles 31

If in a region of configuration space the function V and, consequently, the index

vary only little on the scale of the local wavelength, then geometrical optics becomes
valid and the wave will have the approximate form

(12)

a being a slowly-varying function whose derivatives are very small in relation to

those of S1. On substituting this result into the propagation equation, one sees that
S1 must be a solution of Jacobi's equation for the system, whereby the link with
classical mechanics is established.
An interesting situation arises when the particles of the system do not interact
with one another. In this case one may also treat them as isolated even though they
form a system. The function V then reducing to a sum of terms Vi(xi, Yi, Zi, t), each
one of which defines the action of an external field on a single particle, the equation
describing the system simplifies to

(13)

On postulating 'IjJ(Xl, ... ,ZN, t) = 'ljJ1 (Xl. Yl,Z1, t) ... 'ljJN(XN, YN, ZN, t), we find
that the equation for the system decomposes into N equations of the type

(14)

from which one sees that any particular particle can be treated in isolation from all
others. Nevertheless, the propagation equation also admits solutions that are arbi-
trary linear combinations of the functions IT 'ljJk(Xk, Yk, Zk, t). These combinations
k
represent situations where the particles have earlier been interacting, so that their
present states are not independent. The solutions IT 'ljJk represent those cases where
k
the states of the particles are all independent.

3. INTERPRETATION OF WAVE MECHANICS

FOR A SYSTEM OF PARTICLES

It is easy to adapt the interference principle to systems of particles. One for-

mulates it then as follows: "If the state of a system of particles is represented in
configuration space by the wave function 'IjJ(X1' . .. , ZN, t), then the probability that
an experiment will allow the localization at time t of the system's representative
point within a volume element dr = dX1 ... dZN of configuration space is given by
32 Wave Mechanics of Systems of Particles

When dealing with a single particle, one obviously reverts to the form of the
interference principle studied earlier. For N particles that do not interact and never
have interacted with one another (independent states), one has

N
'¢ = Ilk '¢k(Xk, Yk, zk, t)
1

and therefore

I'¢ 12 dT = I'¢l (Xl, Yl, Zl, t) 12 dXl dYl dZl

(15)
X ••• x I'¢N(XN,YN,ZN,t)1 2dxNdy N dzN.

Consequently, the probability that the representative point of the system will lie
inside the volume element dXl ... dZN of configuration space is equal to the product
of the probabilities that the particles 1,2, ... , N will occupy the volume elements

respectively. This result is in agreement with the theorem of compound probabili-

ti~s, because the presence of individual particles in separate elements of space are
independent events. We thus see clearly why the wave function must then have the
N
form ilk '¢k.
1
In order that the quantity 1,¢1 2 dT shall give an absolute value for the probability
that the representative point will lie within the element dT of configuration space,
it is necessary to normalize the wave function by setting

which determines '¢ up to a phase constant of the form exp (io:).

We must demonstrate that the normalization effected at any instant t will persist
thereafter. To this end, we consider a fictitious probability fluid in configuration
space defined by the relations

p = 1'¢1 2 ,
(16)
PVk = - 4 ~ ['¢~k'¢* - '¢*~k'¢l,
1rzmk

Vk and ~ having the components

f) f) f)
and
f)xk' f)Yk' f)zk'

respectively.
 Interpretation of Wave Mechanics for a System of Particles 33

If one multiplies the propagation equation (13) by'IjJ* and the conjugate of (13)
by 'IjJ, subtraction of the last result from the first gives

that is,
~ 1 (.1.* .1.
L.Jk mk V"k 0/ V"ko/ -
.1. .1.*)
o/V"ko/ =
47ri
h at'
ap
1

or, if one invokes the definitions (16) of the fictitious probability fluid,

(17)

This equation is the generalization to 3N dimensions of the hydro dynamical equa-

tion of continuity

it asserts that the fictitious probability fluid is conserved in the course of its flow
through configuration space. The normalization of 'IjJ therefore has a permanent
character.
The principle of spectral decomposition is formulated as for a single particle: If
the system is conservative, the wave 'IjJ can always be represented as a superposition
of monochromatic waves, and the intensity of any spectral component gives the
probability that an experiment will permit the assignment of the corresponding
energy to the system.
On studying the representation of a wave train in configuration space by a Fourier
integral, one recovers uncertainty relations of the form

in order of magnitude. These relations have the same physical meaning as the
corresponding ones for a single particle.
In the foregoing theory we have supposed that the particles are free to move
throughout all of space (unconstrained systems), and rectangular Cartesian coordi-
nates of the particles have been employed to specify the configuration of the system.
If one wants to use curvilinear coordinates, which is normally the case where con-
straints exist, causing the degrees of freedom of the system to be fewer than 3N,
one must develop the preceding theory somewhat differently. We shall not here
pursue this point [see Ref. (II, 22)]. Likewise, if the system is made up of similar
particles, their indistinguishability will lead one to admit only particular solutions
of the propagation equation. Questions of this nature, too, will be left aside.
 CHAPTER 4

GENERAL FORMALISM
OF WAVE MECHANICS

We are now going to adopt a different point of view and develop the general
principles of wave mechanics from a more formal perspective. If our presentation
were to be given with extreme mathematical rigor, it would be necessary often to
introduce quite complex mathematical considerations, and some points will anyway
still remain in doubt.
In this way the theory would become more satisfying for people with rigorous
minds, but it would not differ much in its practical results from the more concise
theory that I am going to present; and since the latter theory meets the present
needs of theoretical physics, I shall adhere to it in my own exposition.

1. NEW CONCEPTION OF THE QUANTITIES ATTACHED

TO A PARTICLE (OR TO A SYSTEM)

We shall develop the general formalism of wave mechanics while restricting

ourselves to the case of a single particle in a known field of force. The general-
ization to systems of particles could be done easily by following similar lines of
reasoning.
In the automatic procedure that furnishes the Hamiltonian operator starting
from the Hamiltonian expression for the energy in the corresponding classical prob-
lem, one replaces the variables x, y, z by the operators x x, y x, z X and the variables
Px,Py,Pz by the operators

h () h () h ()

We thus see how the idea appears to substitute "operators" for, or make them
correspond to, the "quantities" of classical mechanics. This idea has been raised
to a general principle of wave mechanics in the course of its development: It
has come to be assumed that every measurable quantity (observable) defined in
the old mechanics or physics must correspond to an operator in the new me-
chanics. In order to succeed in constructing, from the classical expression for
any given observable quantity, an operator corresponding to it, one was led to
admit a rule which is a simple generalization of the one already assumed as
 New Conception of the Quantities Attached to a Particle 35
valid for the construction of the Hamiltonian operator and which symbolically
reads
h 0
q--+qx, p --+ - - -
27ri oq'
Whereas space variables are thus transformed into operators, the variable t re-
tains its character as a numerical quantity. This hypothesis, which breaks the
symmetry between variables of space and that of time, is the origin of the dif-
ficulties that one encounters in conciliating quantum theory and the theory of
relativity.
Since, in any problem of classical mechanics, every mechanical quantity pertain-
ing to a particle is expressed as a function of the canonical variables x, y, Z,Px,Py,Pz,
and the time t, one only has to replace in this function every canonical variable by
its corresponding operator in order to obtain the operator that is to be associated
with the mechanical quantity. This operator must moreover contain the time as
a parameter if time appears in the classical expression. If rectangular Cartesian
coordinates are employed, then the operator is completely determined, whatever
the order of the factors in the classical expression might be. If other coordinates
are used, this might not be the case; in order to find the correct operator, one
must then apply certain rules for "symmetrizing" the classical expression of the
quantity.
To give an example, let us apply our method to the construction of the operator
corresponding to the z component of the angular momentum (kinetic moment) of a
particle relative to the origin. One easily finds

(1)

r.p being the azimuthal angle about Oz.

Thus the operators which in wave mechanics correspond to measurable quantities
are the operators, complex in general, belonging to the category of linear operators,
i.e., operators A satisfying

A(cr.p) = cAr.p, (2)

with c denoting a complex constant. Moreover, these operators are Hermitian, that

in f* in
is, they obey
AgdT = g(Af)* dT, (3)

where f and g are two functions that are finite, uniform, and continuous through-
out an arbitrarily chosen domain of variation D of the variables. These func-
tions must cancel each other out at the limits of the domain D in such a man-
ner that the surface integrals that appear in the integration by parts, involved
in the verification of the preceding equation, will be zero. One can verify in
each particular case that the operators corresponding to observable quantities
are always Hermitian. The physical reason for this fact will appear further
on.
36 General Formalism of Wave Mechanics

Among the operators of wave mechanics, it will be useful to distinguish "com-

plete operators," which involve all variables of the domain D, from "incomplete
operators," which involve only some of these variables. [Note L. B. In the plane
xOy, the operator x(a/ay) - yea/ax) is not complete, since it equals a/a'P'] For
a particle that is free to move through all three dimensions of space, the operator
(Px)op is visibly incomplete, whereas the operator Hop is complete.
In summary, with each measurable quantity belonging to a particle, wave me-
chanics associates, starting from the classical expression of the quantity, an operator
that is linear and Hermitian and in general complex. But it is evident that, on car-
rying out the precise measurement of a quantity, one will obtain a real number.
Wave mechanics must therefore be able, starting from the operator, to predict the
possible, essentially real, values that can be furnished by a measurement of the
quantity.
From the linear Hermitian operator, which the new mechanics assign to each
measurable quantity, we must be able to deduce a list of real numbers representing
all the possible results of measuring this quantity. This is made possible precisely by
the fact that the linear Hermitian operators possess a sequence of real "eigenvalues."
Let us study this point in a general fashion.

2. EIGENVALUES AND EIGENFUNCTIONS

OF A LINEAR HERMITIAN OPERATOR

Let A be a linear Hermitian operator, and let us write the equation

where 'P is a function of the variables affected by A and a is a constant. The time
t can enter into A, 'P, and a as a numerical parameter. Per definition, we designate
as "eigenvalues of the operator A in a domain D" those values of the constant a for
which their exist at least one solution 'P( x, y, z, a) of the foregoing equation, called
the "eigenfunction," exhibiting the following properties: It is uniform and continu-
ous in the domain D, and the integral of its squared modulus over D is convergent;
if D is infinite, this last condition obviously implies that 'P must decrease rapidly
enough as infinity is approached. Finally, if D is finite, then 'P must furthermore be
zero at the limits of D.
It should be noted that we regard as distinct solutions of the eigenvalue equation
A'P = a'P only those solutions that are linearly independent.
We assume (and this is a delicate point in any rigorous approach) that eigenval-
ues of the operators used in wave mechanics do in fact exist. It can easily be shown
that they are real. Indeed, from the eigenvalue equation and its complex conjugate,
it follows that
L['P* A'P - 'P(A'P)*] dr = (a - a*) L 'P'P* dr.
The operator A being Hermitian, the integral on the left-hand side vanishes, whereas
the integral on the right is nonzero. One must conclude that a = a*, i.e., the
constant a is real.
 Eigenvalues and Eigenfunctions of a Linear Hermitian Operator 37

The totality of eigenvalues forms the "spectrum" of the equation Acp = acp (or
the spectrum of the operator A in the domain D). If the eigenvalues are isolated,
the spectrum is discrete; it is a line spectrum. If the eigenvalues form a continuous
sequence, one has a continuous spectrum, or a band spectrum. The spectrum can
furthermore be partly continuous and partly discrete. Continuous spectra appear
only for infinite D.
Let us consider discrete spectra. If a; denotes an isolated eigenvalue, then there
exists at least one eigenfunction CPi(X, y, z, t) corresponding to it.
We now demonstrate that the set of eigenfunctions belonging to a discrete spec-
trum forms an orthogonal system; that is, if CPi and CPj are two eigenfunctions
belonging to the distinct eigenvalues ai and aj =I ai, then

In cpicpj dT = O. (4)

Indeed, since the ai are real, one can write

The integral on the left being zero, as a consequence of the hermiticity of A, the
stated formula (4) ensues.
Nevertheless, the preceding demonstration is flawed for two eigenfunctions that
belong to the same eigenvalue. When such a case presents itself, one is said to be
dealing with a "multiple," or "degenerate," eigenvalue. Suppose ai is such an eigen-
value, to which belongs p linearly independent eigenfunctions CPil , CPi 2 , ••• ,CPip ' The
operator A being always linear, every linear combination of these p eigenfunctions
is again an eigenfunction. One may therefore replace CPi1 , ••• ,CPip by p linearly in-
dependent linear combinations of these functions, and it is possible to choose these
combinations in such a manner that they are orthogonal among themselves. One
may thus always suppose that the eigenfunctions of a linear Hermitian operator
constitute an orthogonal set.
The eigenfunctions are obviously determined only to within a multiplicative
complex factor, because, if CPi is a solution of Acp; = aiCPi, then Ccp; is likewise a
solution, due to the linear character of A. It is convenient always to choose the
modulus of the complex constant C such that

(5)

The function CPi is then said to be "normalized." It still will contain an arbitrary
"phase factor" exp (ia) of modulus unity.
If the functions cP; are both normalized and orthogonal (i.e., orthonormal), one
can write
L cpicpj dT = 8;j,

8ij being the "Kronecker symbol," defined to be 1, when i = j, and 0 otherwise.

38 General Formalism of Wave Mechanics

We have up to now studied the case of a discrete spectrum. If A possesses

a continuous spectrum, then an eigenfunction cp( x, y, z, a) will correspond to every
eigenvalue of this spectrum, where we have written a as a continuous variable rather
than as an index. One can easily demonstrate, as above, that every eigenfunction
belonging to a continuous spectrum is orthogonal to every eigenfunction of the
discrete spectrum, if there is one. To show that the eigenfunctions of the continuous
spectrum are normalized and orthogonal to each other, one must, in order to avoid
certain convergence difficulties, employ, instead of the eigenfunctions cp( x, y, z, a)
themselves, the expressions
l>+fl.l>
1l> cp(x, y, z, a) da,
called "eigen-differentials," the integration interval (a, a + ~a) being an extremely
small interval of the continuous spectrum. This substitution has a physical mean-
ing: It corresponds to the one carried out in the classical theory of waves when
one considers, in place of the plane monochromatic wave, which is an abstraction,
the "wave group" formed by the superposition of waves having very nearly equal
frequencies. One then expresses the orthonormality of the eigen-differentials by
writing
1
~aJDdT
f [1la'>'+tJ.l>cp(x,y,z,a)da]* [ll>"+fl.a
a" cp(x,y,z,a)da ] =Oa'a". (6)
The eigenfunctions of the complete operators of wave mechanics exhibit the
important property of forming a "complete" system. This means that, under a
certain wide set of conditions, a function that is defined in the domain D of the
variables affected by A can be expanded into a sum of eigenfunctions of this op-
erator. (For more rigor, this would be the place to introduce the notion of "con-
vergence in the mean," which in this brief exposition we shall however not do.) If,
for example, f(x,y,z) is a function of the variables x,y,z, it can quite generally be
expanded as follows in terms of the eigenfunctions of a complete Hermitian operator
A:
f(x,y,z) = ~CjCPi(X,y,Z) + jc(a)cp(x,y,Z,a) da,
,
the sum being extended over the discrete spectrum and the integral over the con-
tinuous spectrum. We can bring the eigen-differentials into our discussion by writ-
ing
a+tJ.a
f(x,y,z) = LCjCPi(X,y,Z) + Lc(a) l cp(x,y,z,a)da. (7)
i fl.a a
On invoking the formulas that express the orthonormality of the eigenfunctions
of the discrete spectrum and the eigen-differentials, one obtains the results

Cj= kcpU(x,y,Z)dT,
(8)
1 ( [ la +fl.acp(x, y, z, a)da] * f(x, y, z).
c(a) = ~a JD dr a
 Continuous Spectrum of the Free-Particle Hamiltonian. 39
The coefficients Ci and c( a) are often called the Fourier coefficients of the expansion
of the function f(x, y, z) in terms of the eigenfunctions of the operator A. The
Fourier series and the Fourier integrals are simple particular cases of this type of
development. It should be noted that the time may appear as a numerical parameter
in the expressions for Ci and c(o').
We also remark that, if 0'1, •.. , O'i are the eigenvalues of the linear operator A,
then O'i, ... ,O'f are the eigenvalues of An. The verification of this statement is a
trivial matter.

3. CONTINUOUS SPECTRUM OF THE FREE-PARTICLE

HAMILTONIAN. DIRAC'S DELTA FUNCTION

The eigenvalue equation for the Hamiltonian reads

Hr.p = Er.p, (9)
with E here replacing O'. For a free particle, V = 0, H = (_h2 /87r 2m )L\, and one
has
h2
- - 28 L\r.p = Er.p. (10)
7rm
Let the vector p denote the linear momentum of the particle. One then finds
the eigenfunctions

(11 )

wherein
1 (2 2 2) p2 E
2m px + py + pz = 2m = . (12)

We see therefore, first, that every positive value of E is an eigenvalue of Hand,

second, that to every positive value of E there corresponds an infinite number of
eigenfunctions ofthe type (11), obtained by assigning to Px,py,Pz all possible values
compatible with Eq. (12). Consequently one finds a continuous energy spectrum
ranging from 0 to +00, with degeneracy of infinite order for every value of E other
than O.
To each eigenfunction there corresponds a plane monochromatic solution of the
wave equation having the form

27ri) =aexp [27ri

'I/J(x,y,z,p,t)=r.p(z,y,z,p)exp ( TEt ]
T(Et-p.r). (13)

We thus recover some known results. One often sets

(14)

and then writes

'I/J(x,y, z, t, k) = aexp [i(kct - k· r)], (15)
40 General Formalism of Wave Mechanics

along with the relation

Ik 1 .
= .!::...
2
kc (16)
271' 2m
The vector k is called the "propagation vector" of the plane wave, which is
specified entirely by this quantity.
We note that it does not matter whether one takes the eigenfunctions of H to
be the 1/;k or the 'Pk, which differ from one another only by a factor exp (ikct), since
the eigenfunctions are defined anyway only up to a factor of modulus 1.
One can express the orthonormality of the plane waves by introducing eigen-
differentials. In the course of such a calculation, which we shall not reproduce, one
is led to introduce the "improper" or "singular" function 8(x) of Dirac, which is
defined by the following two properties:

(1) It is an even function of its argument x.

(2) For any continuous function f(x),

1x,
x2
f(x)8(x) dx =
{f(O), when
0, when
Xl

Xl
and

and
X2

X2
have opposite signs,

have the same sign.

It is possible to represent 8( x) as the singular Dirichlet function, by putting

8(x) = lim sin271'Nx. (17)

N-+oo 71'X

Finally, the normalization calculation in question establishes that the normal-

ized eigenfunctions belonging to the continuous spectrum of a free particle must be
written as
'P(x,y,z,k) = (271'~3/2 exp[-i(k.r)]'

+/2
(18)
1/;(x, y, z, t, k) = exp [i(kct - k· r)].
(271' )
The property of completeness possessed by this set of eigenfunctions is expressed
by the fact that, under very general conditions, a function f (x, y, z) can be expanded
in a Fourier integral of the form

f(x,y,z)=-(
1)3/2
271'
1 00

-00
c(k)exp(-ik.r)dk, (19)

wherein dk denotes the product dkxdkydk z . The c( k) are given by

c(k) = (271'~3/2 10 f(x, y, z) exp (ik . r) dr, (20)

dr signifying dx dy dz; this is the classical formula for the coefficients of the Fourier
integral.
 Continuous Spectrum of the Free-Particle Hamiltonian. 41

One can also write

f(x,y,z) = -1)3/2
(
271'
Joo c(k,t)'lj;(x,y,z,t,k)dk,
-00
(21)

with
c(k, t) = c(k) exp (-ikct). (22)
 CHAPTER 5

GENERAL PRINCIPLES
OF THE
PROBABILISTIC INTERPRETATION
OF WAVE MECHANICS

1. GENERAL IDEAS

Wave mechanics must enable us to calculate the eigenvalues of the measurable

quantities (or observables) belonging to a particle (or, by a natural generalization,
to a system). But it represents the state of a particle (or, more exactly, the state
of our knowledge about a particle) by a wave function ¢(x, y, z, t), a solution of
the propagation equation, a function which we always assume to be normalized.
Furthermore, it associates with every measurable quantity belonging to a particle a
linear Hermitian operator, which allows one to define an ensemble of real numbers,
its eigenvalues, and a complete system of functions, its eigenfunctions. We are thus
in a position to formulate the two fundamental principles for the physical formulation
of wave mechanics:

First Principle. The possible values of a measurable quantity, i.e., the different
possible results of a measurement of this quantity, are the eigenvalues of the linear
Hermitian operator corresponding to this quantity. (Quantification principle.)

Note G. L. One will notice that, contrary to many other authors, de Broglie does
not postulate these principles as a priori, but seeks to infer them, starting from the
theory of waves. He does not maintain that every Hermitian operator represents an
observable, but only assumes that, if we know an observable to exist, then it will
be thus represented. The present reader, nurtured on years of quantum mechanical
studies, would make a mistake to read these pages with only half an eye, because it
goes without saying: Here, in fact, it is that the origin of the quantum formalism
can truly be understood.

Second Principle. If the state of the particle is represented by a certain wave

function ¢(x, y, z, t), a solution of the propagation equation, then the probability
that a precise measurement of the measurable quantity, corresponding to a linear,
Hermitian, and complete operator A with nondegenerate eigenvalues, will furnish
at the time t a particular eigenvalue of A is equal to the squared modulus of the
 General Ideas 43

coefficient of the corresponding eigenfunction in the expansion of the wave function "p
in terms of the normalized eigenfunctions of A. (Generalized spectral decomposition
principle. )

More precisely, if the function "p allows an expansion in terms of the eigenfunc-
tions and eigen-differentials of A according to the formula

"p(x,y,z,t) = LCiipi + LC(o)

i Cl./l'
1 /l'
/l'+Cl./l'
ip(x,y,z,O) do, (1)

then the probability of the proper value OJ is Ic; 12 and the probability of a proper
value lying between 0 and 0 + ~o is Ic( 0) 12 ~o.
The wave function "p by hypothesis being normalized, one can easily verify that
the total probability of all possible eigenvalues is indeed equal to unity. Natu-
rally, the probabilities of the possible values separately must be functions of the
parameter t.
If the operator A has multiple eigenvalues, then the statement of the second
principle must be augmented. Let OJ be a multiple eigenvalue to which there be-
longs p linearly independent orthonormalized eigenfunctions ipil, ipi2, ... , ipip. The
probability of finding, in a measurement made at an instant t, the value 0i for the
quantity A is then equal to the sum of the squared moduli of the coefficients of
ipil, ... ,ipip, appearing in the expansion of "p, i.e., to

L. iciil
J
2•
1

It can be verified that this sum is, as it must be, independent of the manner,
arbitrary in part, in which the eigenfunctions ipil, ... ,ipip are chosen.
If the operator A is incomplete, the statement of the second principle must un-
dergo a modification. Then, indeed, the eigenfunctions of A do not depend on all
the variables x, y, z, and consequently the coefficients Ci and c( 0) become functions
of the variables that are not involved in A. Accordingly, the probability of a partic-
ular eigenvalue can then not be the corresponding ICi 12 , a quantity which now still
depends on certain variables. To obtain the correct probability, it is necessary to in-
tegrate the expressions indicated above with respect to the variables not involved in
A. One can verify that, after this modification, the total probability for all possible
eigenvalues is indeed equal to unity.
A simple application of our two principles is provided by the case of the operator
H, which is complete. If H is independent of the time, it will exhibit constant
eigenvalues Ei and eigenfunctions 'Pi. A measurement of the energy can furnish
only one of the values Ej, and, if"p = Ei Ci'Pi, the probability of Ej will be ICiI2.
One thus recovers the idea of the quantification of the atomic system and Born's
spectral decomposition principle. If the spectrum is complete, a discrete set of
stationary states with quantized energies is obtained.
As another example, let us consider the operator for the x coordinate of a par-
ticle, which corresponds to "multiplication by x." Here the eigenvalue equation is
44 General Principles of the Probabilistic Interpretation

x<p = a<p. One can view this equation as valid for all real values of x by positing
<pC x, a) = 5( x - a), the latter being the singular Dirac function, which is zero for all
x =f a. Therefore, according to our first principle, a measurement of x can furnish
any real value between -00 and +00. Furthermore, the eigen-differentials

1 <>
<>+6<>
5(x-a)da

form a complete orthonormal system. As one obviously has

'I/J(x,y,z,t)= 1 +00

-00 'I/J(a,y,z,t)5(x-a)da, (2)

the probability that a measurement of x will at the instant t furnish a value lying
between a and a + L'la is

and from this one easily infers the probability to be 1'I/J (x , y, z, t)1 2 dr that the particle
will at the time t manifest its presence in the element dr surrounding the point
x, y, z. The total probability for the particle to be present at any point of the space
D accessible to it is then indeed unity, since

Note L. B. This is the physical reason for normalizing 'I/J. We shall find that one
can deduce from general principles that two quantities A and B are simultaneously
measurable only if AB = B A. Thus the conjugate variables p and q cannot be
measured simultaneously.

Later on we shall investigate in a thorough manner the way in which Heisenberg's

uncertainties can be deduced from the general principles stated above.

Note L. B. Notion of superposition: Every eigenfunction <Pi of an operator A

describes a state of the system for which the quantity A for sure has the precise
value ai. In general the 'I/J of a system does not reduce to a single <Pi but is instead
equal to a sum of the <Pi: 'I/J = I:i Ci<Pi·
One then also says that 'I/J is a "superposition" of the <pi, this designation stem-
ming from the "superposition principle for small motions" in classical vibration
theories. But superposition here does not at all have the same meaning as in clas-
sical theories. It no longer relates to the vibration of a medium that is obtained by
adding several elementary vibrations but rather to the following assertion: If the
function 'I/J of a system has the form 'I/J = I:i Ci<Pi, and if one should try to attribute
to the system some state <Pi in measuring the quantity A, then the probability is
ICk 12 that one would be led to attribute to it the state <Pk. Therefore, prior to the
 General Ideas 45
measurement, the system occupying the state 'l/J = I:i Ci'Pi finds itself potentially
in several states 'Pi, each possessing a nonzero probability ICi 12. This is an entirely
novel idea, completely foreign to classical theories, in which the state of any system
is characterized by well-defined values of the properties of the system. This new
idea of superposition is perhaps the most important one emerging from the new
mechanics.
In the classical theory of vibrations, if one encounters a vibration described by

this means that the value of 'l/J at every instant and at every point is given by the
sum of the terms of the given series: The vibrational components add according to
their respective coefficients Ci. In wave mechanics, the postulated expansion of'l/J is
subject to the condition I:i ICil2 = 1, which is tied to the probabilistic interpretation
of 'l/J, and one can no longer regard 'l/J as furnished by the addition of terms having
predetermined amplitudes. Thus, in classical theory, two wave motions

on superposition give a wave 'l/J = 'l/JI +'l/J2 of amplitude CI +C2. In wave mechanics, by
contrast, the two states 'l/JI and'l/J2 considered individually must satisfy the conditions
ICII = v- 1/2 and IC21 = v- 1/2 ; and if these states are superimposed, one gets the
state 'l/J = 'l/JI + 'l/J2 but coupled with the condition ICJ + c21 = v- 1/2 , so that there is
no longer any addition of the amplitudes. This outcome demonstrates the gulf that
separates the notion of wave function in the classical theory of waves from that of
wave function in wave mechanics.

Note G. L. On the subject of the preceding note by the author: Louis de Broglie
envisages here only the continuous and normalized wave on which the probabilistic
interpretation of wave mechanics is founded. Placing himself behind the orthodox
point of view, he assumed at the time of writing these lines that this wave was the
only one possible; and in this note we see him insist on this view with as much
vigor as formerly he had been convinced of its contrary. As we know, it is to this
conviction that he was soon to return in developing his theory of the double solution
and in carefully distinguishing the wave 'l/J, possessing the properties that he here
describes, from the wave v (the regular part of the singular wave u) which has the
same phase as 'l/J but not the same amplitude, which is not normalized and not
subject to wave-packet reduction, and which, by contrast to 'l/J, obeys the ordinary
law for the addition of components used in the classical theory of vibrations. Louis
de Broglie would from then on regard v as the true physical wave, in opposition to
'l/J, which is only a predictive tool.
46 General Principles of the Probabilistic Interpretation

2. THE ALGEBRAIC MATRICES AND THEIR PROPERTIES

By a "matrix" is understood an array of numbers containing a finite or infinite

number of rows and columns. If this array has a finite dimensionality, we shall for
simplicity assume it to be square. Every number appearing in the array, or "matrix
element," can be located with the aid of two indices that specify respectively the row
and column of the array in which the element appears. We designate the element
situated at the intersection of the row i and column k by aik and the entire matrix
by A. The elements aii are the diagonal elements, and a matrix whose diagonal
elements alone are nonzero is called a "diagonal matrix." Two matrices A and B
are said to be equal (A = B) if all their corresponding elements are equal: aij = bij
for all i and j.
We encounter matrices in algebra when studying linear transformations. Indeed,
if the variablesxi are linear combinations of the variables Xi, one is dealing with
transformation formulas of the type xi = I:j aijXj, which can symbolically be writ-
ten as Xl = AX, with the convention (AX)i = I:j aijX j. One is thus naturally led
to define the sum and product respectively of two matrices by the following rules:

(1) The sum of two matrices A and B is per definition the matrix A +B with
elements aij + bij.

(2) The product of the matrix A with the matrix B is the matrix AB whose ik
element equals I:j aijbjk.

This last definition implies that in general the matrix AB does not equal the
matrix BA. If by exception AB = BA, A and B are said to commute. One often
introduces the matrix [A, Bl = AB - BA, called the "commutator" of the matrices
A and B, which, when not zero, serves as a measure of the lack of commutation
between A and B.
Sometimes one also introduces the matrix [A, Bl+ = AB + BA, or the "anti-
commutator" of A and B. If this matrix vanishes, AB = -BA, and A and Bare
said to anticommute; if not, [A, Bl+ measures the lack of anticommutation of A and
B.
Matrices are real or complex depending on whether their elements are real or
complex. We envisage the general case of complex matrices.
A matrix is called Hermitian if aik = aki for all i and k. A real Hermitian matrix
is therefore symmetric with respect to its diagonal, and all diagonal elements of any
Hermitian matrix are real.
A matrix is anti-Hermitian if aik = -aki for all i and k. The diagonal elements
of any anti-Hermitian matrix are thus purely imaginary.
The product of two Hermitian matrices A and B is itself Hermitian if and only if
the two matrices commute; the product is anti-Hermitian if A and B anticommute.
The matrix A is know as the "transpose" of A if (hi = aik; and one calls At
the "adjoint" of A if ark = aki' i.e., At = A*. If A is Hermitian, At = A, i.e., A is
self-adjoint.
One obviously has (At)t = A, and it is easy to demonstrate that (AB)t = BtAt.
 The Algebraic Matrices and Their Properties 47
A diagonal Hermitian matrix is necessarily real. In particular, the diagonal
Hermitian matrix aik = {iik is the "unit matrix," often denoted by l.
If, for a given matrix A, there exists a matrix A -1 such that AA -1 = A-I A = 1,
then the matrix A-I is called the inverse of A. If A has a finite number of rows
and columns, A-I will always exist if the determinant formed from the array of the
elements aik of A is different from zero. If A has an infinite number of rows and
columns, A-I mayor may not exist, depending on the particular case. Whenever
A-I and B- 1 exist, one has (AB)-1 = B- 1A-I.
Suppose A is a real matrix and its elements satisfy the equations

I>jiajk = {iik, 2: ajlak! = {ijk; (3)

j I
the matrix A is then said to be orthogonal. The linear transformation described by
such a matrix represents an orthogonal transformation in space that leaves the sum
xr
Ei invariant. One may generalize this definition to include complex matrices by
saying that a matrix A defines a complex orthogonal, or "unitary," transformation
if
2: a ji a jl = {iil, 2:aj/akl = {ijb (4)
j I
and it is easy to verify that the quantity Ei xixi will remain invariant under such
a transformation. The matrix A is then said to be "unitary," and one has

2: ataij = {ikj, 2: aj/ark = {ijb

which means that
(5)
Thus the adjoint of a unitary matrix is identical with its inverse.
By definition, the "trace" of a matrix A is the sum of its diagonal elements:
Tr A = Ei aii. One sees immediately that

Tr AB = Tr BA = 2: aikbki· (6)
i,k
Consider two square matrices A and S of the same dimensionality, such that A
is arbitrary and S unitary. Then the matrix B defined by
B = S-IAS (7)
is said to be obtained from A by a "canonical transformation." One can readily
verify that, if A is Hermitian, so is B. Canonical transformations thus conserve the
Hermitian character of matrices: It is easy to see that they likewise leave the trace
of any matrix unchanged. Moreover, if two square matrices A and A' become B
and B', respectively, under a canonical transformation, then their product AA' is
transformed into BB' by the same transformation, because
S-IAA'S = S-IASS- 1 A'S.
48 General Principles of the Probabilistic Interpretation

3. OPERATORS AND MATRICES IN WAVE MECHANICS

Suppose some set of orthonormal functions 'Pl, ... , 'Pi, ... is known in the do-
main of variation D of certain variables. We shall call them basis functions;
they may, for example, be the normalized eigenfunctions of a wave-mechanical
linear Hermitian operator. Using this basis set, one can associate a matrix with
each linear operator. Indeed, let A be such an operator; then the application
of A to anyone of the basis functions 'P will furnish a new function, which
can be expanded in terms of the basis functions, resulting in relations of the
form
A'Pi = Laji'Pj, (8)
j

with
(9)
D being the domain of variation of the arguments of the functions 'Pi. Per definition,
the aij are the elements of the matrix generated by the operator A in the basis set
of the 'Pi. We designate this matrix by the same symbol A as the operator or, if
the basis employed is to be specified, by A'P. It is easy to verify that the matrices
thus defined obey the rules of addition and multiplication for algebraic matrices
formulated earlier.
If the basis set is formed by eigenfunctions of a wave-mechanical operator and
if the operator A is itself a linear Hermitian operator of wave mechanics, then we
say A is a wave-mechanical matrix. It is obvious that such matrices are always
themselves Hermitian, because the definition (9) of ajj shows ajj = aji to be a
consequence of the Hermitian character of the operator A. Quite generally, one sees
that a necessary and sufficient condition for the matrix generated by an operator
A, with respect to some basis set, to be Hermitian is that the operator itself be
Hermitian. Hermiticity is therefore an intrinsic property of operators, in the sense
that any Hermitian operator will generate Hermitian matrices in arbitrary basis
sets. All wave-mechanical matrices are thus Hermitian.
Our definitions accordingly establishes an intimate correlation between oper-
ators and matrices. In particular, the necessary and sufficient condition for two
matrices to commute (or anti commute) is that the corresponding operators com-
mute (or anticommute), and conversely. This leads us to define the commutator
and anticommutator of two operators as

[A,B] = AB -BA, [A,B]+ = AB+ BA. (10)

A particularly important category of wave-mechanical matrices is obtained by
always choosing the basis functions to be the eigenfunctions of the Hamiltonian
operator belonging to the physical system under consideration. Let 'ljJl, . .. , 'ljJn, ...
be the eigenfunctions of the operator H. Then the matrices AtP generated by a
linear Hermitian operator A in the basis set of the 'ljJj, and whose elements thus are

(11 )
 Mean Values and Dispersions in Wave Mechanics 49
can be called "Heisenberg matrices," because they are the ones on which Heisenberg
based his quantum mechanics. If one includes the exponential factor

in the definition of the 'ifyk and puts

one gets
(12)

These elements define the Heisenberg matrix proper, which depends on the time.
Sometimes the exponential factor is suppressed in the expression for 'ifyk, and one
sets
ajk = In ajAak dr,

thereby defining a matrix A' of elements ajk that is independent of the time. This
is Schrodinger's matrix corresponding to the preceding matrix of Heisenberg. We
shall generally employ Heisenberg's matrices.
When Heisenberg matrices are used, the matrix H corresponding to the energy
is a diagonal matrix whose diagonal elements equal the eigenvalues of the energy
(i.e., the stationary quantized energies). Nevertheless, in the event that the operator
H possesses multiple eigenvalues, the foregoing property holds only if one has taken
care to choose the eigenfunctions corresponding to the multiple proper values in
such a manner that they are orthogonal to one another. The verification of this fact
is a trivial matter, because

The preceding result is moreover only a particular case of the following theorem,
the proof of which is immediate: "The matrix generated by an operator A in the
basis composed of its orthonormal eigenfunctions is diagonal, and the diagonal ma-
trix elements are equal to the eigenvalues of the operator A, multiple eigenvalues
appearing to the order of their multiplicity."

4. MEAN VALUES AND DISPERSIONS IN WAVE MECHANICS

For any state of a particle (or system of particles) characterized by some form of
the wave function 'ify, every quantity A has a set of possible values (possible results
of measuring A) capable of occurring with definite probabilities. One can there-
fore define the "mean value" A of the quantity A as the mathematical expectation
corresponding to a measurement of A.
50 General Principles oj the Probabilistic Interpretation

Therefore, if (Xi and Ji are the eigenvalues and eigenfunctions, respectively, of A,

the mean value A will, according to general principles, be defined by

-A " (Xi Ic;! 2 .

= 'L.,..

On replacing 'if; by L:i Ci<pi and taking into account the orthonormality of the <pi,
one is able to verify that A can also be written as

(13)

which allows the immediate calculation of A from a knowledge of 'if;.

Having defined the mean value of the random (aleatory) variable A, one can also
define the corresponding "dispersion" (in the sense of probability calculus), that is,
the square root of the mean square deviation of A. Denoting the dispersion by 0"A,
we have
= y'(A-A)2 = JA2 _2AA+A2 = JA2 _A 2,

r·
O"A

O"~ = in 'if; *A2'if;dr - (in

from where

'if;* A'if;dr (14)

N ow consider two quantities A and B belonging to the same particle. To A there

correspond the eigenvalues (Xi and eigenfunctions <Pi, to B the eigenvalues f3k and
eigenfunctions Xk. If the wave function 'if; is expanded in the form 'if; = L:i di<Pi,
substitution of this expansion into the integral expression for B gives

B = Ldidkb;k' (15)
i,k
where b;k is the ik element of the matrix generated by the operator B in the basis de-
fined by the <Pk. The mean value of B can thus always be expressed linearly in terms
of elements of the matrix generated by the operator B in the set of eigenfunctions
belonging to another operator A.
In particular, if the particle (or system of particles) finds itself in one of the
eigenstates of the operator A (a state which arises following a precise measurement
of the quantity A), then one has 'if; = dj<pj, with Idil = 1, and therefore
-
B = b'Pjj , (16)
whence the following theorem: "The ii diagonal element of the matrix generated by
the operator B in the basis provided by eigenfunctions of the operator A represents
the mean value of the quantity B when one knows that the quantity A has the
precise value (Xj."
This theorem gives a physical meaning to the diagonal elements of the matrices
of wave mechanics. Another theorem assigns a physical meaning to the nondiagonal
elements:
 First Integrals in Wave Mechanics 51

Let us suppose the particle is always in the state 'l/J = 'Pi. We have seen that
then b~ is the mean value of B in this state. The mean value of B2 is then

f *B2 'P,. dT -- (B2)'Pii -- (b 2 )'Pii'

B 2 -- JD'Pi (17)

But the rule of matrix multiplication gives

j j

#i
= (b~)2 +L IbiJ 12
j

(the last step being valid because B is Hermitian), whence

#i #i
i7~ = B2 - (B)2 = L IbiJI 2 = L Ibjil 2 . (18)
j j

Consequently we have the following theorem:

"If one constructs the matrix corresponding to a quantity B in a basis formed
by the eigenfunctions of a quantity A, then the sum of the squared moduli of the
nondiagonal elements appearing in the row i (or column i) of the matrix B is equal
to the square of the dispersion i7B of the quantity B when it is known that the
quantity A has the precise value ai." This statement imparts a physical meaning
to the nondiagonal elements of wave-mechanical matrices.
If the eigenfunctions Xi of B coincide with the eigenfunctions 'Pi of A (we shall
see that [A, B] = 0 is a necessary condition for this to happen), then the quantity
B will in the state 'l/J = 'Pi = Xi have the precise value f3i corresponding to Xi and
the matrix B'P will be diagonal. Therefore i7 B = O.

5. FIRST INTEGRALS IN WAVE MECHANICS

Let us consider the Heisenberg matrix A whose elements are defined by

ajk = in'l/Jj A'l/Jk dT .

The element ajk may depend on the time t via 'l/J; and 'l/Jk and also via A if this
operator depends explicitly on the time. Let us therefore derive the dependence of
ajk on t, keeping in mind the fact that both 'l/Jj and 'l/Jk obey the wave equation and
that the operator A is Hermitian. It is easily found that

dajk =
dt
f .I,~ [8A
JDo/)
27ri (AH _ H A)] .1, dT
8t + h o/k ,
(19)
52 Geneml Principles of the Probabilistic Interpretation
where aA/ at is the operator one obtains on formally differentiating A with respect to
the parameter t. We can interpret the foregoing formula by saying: The Heisenberg
matrix whose j k element is dajk/ dt is generated in the basis of the 1jJi by the symbolic
operator dA/ dt such that

dA = aA 21l'i [AH -HAl. (20)

dt at + h
It happens quite frequently that the operator A does not explicitly depend on the
time. Then aA/ at = 0 and
~~ = 2~i [A,Hl.
Note L. B. It is easy to deduce that, for A independent of the time,

By definition, in a problem where the Hamiltonian H is given, the observable

quantity corresponding to an operator A is called a "first integral" or a "constant
of motion" for the problem under consideration if dA/ dt is zero, that is, if

(21)

Thus, when A does not explicitly depend on the time, the quantity A is a first
integral if the operator A commutes with the Hamiltonian operator.
One can also define the first integrals in the following manner: A quantity with
operator A is a first integral if, 1jJ being any solution of the wave equation, A1jJ is
likewise a solution. Indeed, by hypothesis, a1jJ/at = (21l'i/h)H1jJ; consequently

A a1jJ = 21l'i AH1jJ (22a)

at h
and

(22b)

For A1jJ to be a solution of the wave equation, one must therefore require that

(23)

The necessary and sufficient condition for this equation to be satisfied, whatever the
solution 1jJ of the wave equation, is precisely the relation (21); q.e.d.
We give here a few classical examples of first integrals.
If the external field of force acting on a particle (or a system of particles) is
independent of the time, the operator H will not contain t, and thus, since H obvi-
ously commutes with itself, the energy becomes a first integral: We thus obtain the
analog of the conservation of energy for conservative systems in classical mechanics.
 Angular Momentum in Wave Mechanics 53

Similarly, if the x component of the external force is zero, the operator H will not
depend on x (f)V/f)x = 0), causing it to commute with (Px)op- The x component of
the linear momentum thus becomes a first integral, an outcome which is analogous
to a theorem of classical mechanics.
Finally, if the force field has a vanishing torque about the Oz axis, i.e., if the
potential energy V does not depend on the azimuth r.p about Oz, then the Hamil-
tonian is independent of r.p and, as a result, will commute with the operator for the
z component of the angular momentum, this operator being given by
h f)
(Mz)op = --2 '!l'
7rZ ur.p

The quantity Mz is thus a first integral, as in classical mechanics. If the force

field is central, the three components of the angular momentum M with respect
to axes passing through the force center are first integrals, and the same is true
for the quantity M2 = M; + M; + M; (square of the magnitude of the angular
momentum). This leads us to say a few things about the angular momentum.

6. ANGULAR MOMENTUM IN WAVE MECHANICS

In the present exposition, we shall leave aside the property of spin and confine
ourselves to the orbital angular momentum. For a particle, the orbital angular
momentum M with respect to a center 0 (taken as the origin of coordinates) is
defined as the moment of the linear momentum of the particle about the point 0;
that is,
M = l' xp. (24)
The components of M thus are Mx = YPz - ZPy, etc.
We have seen that, if the force acting on the particle has a zero torque with
respect to one of the axes, then the component of M along that axis is a first
integral.
The magnitude of the angular momentum is defined by its square, i.e.,

(25)
according to the identity of Lagrange. This quantity is a first integral in a central
field of force.
In wave mechanics one replaces M x, My, Mz by the operators

(Mx)op h . (Y ~
= --2 - z UY~) = --2
h . af) ,etc., (26)
7rZ uZ 7rZ r.px
with r.px, etc. denoting the azimuths about the axes Ox, etc. Anyone of the op-
erators Mk possessess the eigenvalues m( h/27r) and the normalized eigenfunctions
(27r)-1/2 exp (-imr.pk) (with m = 0, ±1, ±2, ... ), as may be easily verified.
According to the general principles of wave mechanics, one must therefore con-
clude that the exact measurement of any component of the angular momentum must
54 General Principles of the Probabilistic Interpretation

always furnish a multiple of h/27r. This quantity may therefore be regarded as the
quantum unit of angular momentum.
We notice then that the representation of the angular momentum by a vector
is misleading at the quantum level. Indeed, at this level the three components
of the angular momentum are not in general simultaneously measurable, because
the operators Mx, My, Mz fail to commute. If therefore one performs an exact
measurement of any particular rectangular component of M, the exact values of
the other components of M will remain unknown; there will exist only a probability
distribution for the possible values of these components. We are therefore unable
to construct the vector M exactly, since no more than one of its components can
ever be known precisely at any given time. It is moreover obvious that one cannot
assume the vector M to simultaneously possess three rectangular components that
are multiples of h/27r, whatever the orientation of the Cartesian axes about the
point O.
The noncommutation of the operators M x , My, Mz is easily proved. One finds
in fact that
h
[Mx, My] = - . M z , etc .. (27)
27rz
We shall make use of these relations.
With the property M2 of classical mechanics, wave mechanics associates the
operator

(28)
h2 1 [0 ( 0) 1 02 ]
= 47r 2 sin 0 00 sin 0 00 + sin 0 Or.p2
when spherical polar coordinates around the axis Oz are used. Aside from the factor
h 2 / 47r 2 , this operator is nothing but the Laplacian on the surface of a sphere with
unit radius.
The eigenvalue equation
(29)
allows as finite, uniform, and continuous solutions on the unit sphere only the Lapla-
cian functions Yk(O,r.p), the eigenvalue corresponding to the function 1/, where I is
a positive integer or zero, being (h 2 /47r 2 )1(1 + 1). We thus finally see that the
eigenvalues of M2 are

h2
M2= 47r21(l+1), 1=0,1,2,,, .. (30)

It is easy to verify that M2 commutes with Mx,My, M z . One can therefore simul-
taneously measure the magnitude of M and anyone of its components.
 CHAPTER 6

THEORY OF THE COMMUTATION

OF OPERATORS
IN WAVE MECHANICS

1. GENERAL THEOREMS
Consider two wave-mechanical operators A and B. In general they will not
commute and thus AB =f BA. By exception, one may have AB = BA. We are going
to show that the commutation of two operators corresponding to two measurable
quantities is a property of great importance in wave mechanics. This importance
rests essentially on the following theorem: "The necessary and sufficient condition
allowing two linear Hermitian operators A and B to have a common system of
eigenfunctions is AB = BA."
To prove this theorem rigorously, one must distinguish three cases: (1) Both
operators are complete. (2) One operator is complete, the other is not. (3) Both
operators are incomplete. The demonstration of the theorem and even its correct
statement differ slightly from case to case.

Case 1. Theorem: The necessary and sufficient condition permitting two com-
plete operators A and B to have the same system of eigenfunctions is that they
commute.

Indeed, suppose first that the two operators possess a common system of eigen-
functions 'PI, ... ,'Pi,· ... One then has A'Pi = ai'Pi and B'Pi = f3i'Pi for all i, whence
BA'Pi = aiB'Pi = aif3i'Pi and AB'Pi = f3iA'Pi = f3iai'Pi. Therefore, AB'Pi = BA'Pi
for all 'Pi. But, since the latter form a complete system, one gets ABf = BAf for
any function f that can be expanded in terms of the 'Pi. One thus concludes that
AB = BA, i.e., the stated condition is necessary.
Next we demonstrate its sufficiency. We thus start by assuming that AB = BA.
If 'Pi are the eigenfunctions of A and Xi those of B, one has A'Pi = ai'Pi and
BXi = f3iXi. It follows from the first equation that BA'Pi = aiB'Pi = AB'Pi (since
BA = AB). B'Pi is therefore an eigenfunction of A with the eigenvalue ai. Suppose
first that aj is not a multiple eigenvalue; then B'Pi will necessarily be porportional
to 'Pi, and one gets B'Pi = (const.) 'Pi = f3i'Pi. But 'Pi is a function that is finite,
uniform, and continuous throughout the domain D and zero at its limits. According
to the last equation, it is therefore an eigenfunction of B. All eigenfunctions of A
56 Theory of the Commutation of Operators

are thus eigenfunctions also of B if none of the ai is multiple. On applying the

operator A to the equation BXi = f3iXi, one could likewise demonstrate that every
eigenfunction of B is an eigenfunction of A if none of the f3i is multiple. Therefore,
if all ai and all f3i are simple eigenvalues, the set of CPi will coincide with the set of
Xi. Thus the stated condition is also sufficient.
The demonstration is flawed if some of the aj or f3j are multiple eigenvalues.
Suppose, for example, that a certain eigenvalue aj of A corresponds to p eigen-
functions CPil, ••. , cpip. Then one will have relations of the form ABcpij = BAcpij =
aiBcpij by the preceding line of reasoning. From this, one can only conclude that
BCPij is a linear combination of CPil. . .. , CPip, that is,
p

BCPij = Lk cjCPik,
1

the cj being complex constants. It must be possible to express the p functions BCPij
linearly in terms of the p eigenfunctions Xi of the operator B. The p functions BCPij
are in fact linearly independent, and they could not have this property if they were
expressible as linear combinations of less than p functions Xi; if, on the other hand,
they could be expressed only with the aid of more than p functions Xi, then the Xi
would not be linearly independent. One can thus linearly express the BCPij using p
functions Xi, and p functions only; conversely, the p functions Xi can be expressed
linearly with the aid of the p functions BCPij. As in the case of degeneracy, one can
replace the p eigenfunctions by p linearly independent linear combinations of them.
The BCPij can be replaced by the Xi in question, and these will simultaneously
be eigenfunctions of A and B for the eigenvalue ai. A similar reasoning can be
advanced if one of the f3i is multiple, and we thus arrive at the conclusion that it
is always possible, through a proper choice of eigenfunctions, to obtain a common
set of eigenfunctions for the operators A and B. Our theorem thus becomes fully
established in the Case 1.

Case 2. One of the operators is complete, the other is not.

Theorem: Let A be a complete operator and B an incomplete operator. If the

two operators commute, every eigenfunction of A is equal to a product between an
eigenfunction of B and a function of the variables that are absent in B. Conversely,
if this property is confirmed, then the operators will commute.

Direct Proposition. We assume AB = BA. Let x, etc. be the variables ap-

pearing in B, y, etc. those that do not. The eigenfunctions of A are the functions
CPi(X,y, ... ), those of Bare Xi(X, ... ). One has ACPi = aiCPi and, consequently,
BAcpi = aiBcpi = ABcpi (because AB = BA). If ai is not multiple, then BCPi must
be proportional to CPj, i.e., BCPi = f3cpj; thus cPj is an eigenfunction of B. But the
set of Xi being complete for the variables x, etc., CPi can be an eigenfunction of B
only if it is equal to a function Xk multiplied by a factor that depends only on y,
etc.; that is,
CPi(X"",y, ... ) = lik(Y,···)Xk(X, ... ), (1)
 General Theorems 57

which is the stated proposition.

The foregoing demonstration is at fault if A has multiple eigenvalues. Then the
eigenvalue aj will correspond to p linearly independent eigenfunctions <Pil, •.• , <pip.
Let A' be a complete operator having simple eigenvalues and commuting with
B. According to what has been demonstrated, every eigenfunction <p~(x, ... ,y, ... )
of A' can be written in the form <p~(x, ... ,y, ... ) = !;k(Y, ... )Xk(X, ... ). For the
reasons given above, the p eigenfunctions <Pij can be expressed linearly with the aid
of p, and only p, functions <Pi, and conversely. In the list of eigenfunctions of A,
one may therefore replace the p functions <Pij by p functions <p~. Consequently, if
there exist multiple eigenvalues ai, one can, exploiting the indeterminate nature of
the corresponding eigenfunctions, arrange to choose these eigenfunctions in such a
manner that the theorem under discussion will again be confirmed.

Converse Proposition. We suppose that every eigenfunction of A is of the form

<Pi(X, ... ,y, ... ) = !;k(Y,· . .)Xk(X, . .. ). (2)

From A<Pi = ai<pi, one can conclude

BA<pi = ajB<pi = a;!ik(Y, ... )BXk

= a;!ik(3kXk = ai(3k<Pi ;

likewise, from

one infers, first,

and then

Thus, in summary,
AB<pi = BA<pi. (3)
This result being true for all functions of a complete set, it implies that AB = BA,
and thus the converse theorem is established.
We should make an important remark about the formula

<Pi(X, ... ,y, . .. ) = !;k(Y,· . .)Xk( x, ... ). (2)

In general, to the same eigenfunction Xk of the incomplete operator B there corre-
sponds several eigenfunctions <Pi of the complete operator A. It is this circumstance
that obligates us to attach two indices to the function !;k(Y, .. .), since, in general,
for every given value of k, there are several values of i. In other words, there is
no one-to-one correspondence between the eigenfunctions of A and those of B, the
first set being much more numerous than the second. As an example, let us choose
the operator A to be the Hamiltonian operator of a system exhibiting spherical
symmetry, such as the hydrogen atom, and as operator B the incomplete operator
Mz•
58 Theory of the Commutation of Operators

On using the spherical polar coordinates around Oz, one has

and the eigenvalue equation of H will read

(4)

On setting CPi = f(r,O)x(cp), one readily finds that the eigenfunctions of A =

H are of the form CPnlm(r,O,cp) = fnl(r,O)exp(imcp), where n,l,m are quantum
numbers, with m in particular equal to a positive or negative integer or zero.
But the eigenfunctions of B = Mz are the uniform solutions of the eigenvalue
equation -(h/27ri)(8X/8cp) = (JX, that is, Xm(CP) = exp (-imcp) (m integral), the
corresponding eigenvalues being (Jm = m(h/27r). One thus sees that our theorem is
well confirmed. To a given value of the integer m there correspond an eigenfunction
Xm of Mz and a whole set of eigenfunctions of H that are the products of Xm with
the functions fnl(r,O) corresponding to various possible values of nand l.
We can add a remark which is in a way the converse of the foregoing. To every
eigenfunction Xk(X, ... ) of B corresponds at least one eigenfunction CPi(X, ... , y, ... )
of A that is proportional to it. Indeed, if in the CPi one assigns constant values to y,
then all the functions CPi that are proportional to anyone of the Xk are equivalent.
The set composed of the CPi, which is complete for the set of variables x, etc. and
y, etc., must also remain complete for the variables x, etc.; and, for that to be true,
obviously at least one of the functions CPi must reduce, up to a constant factor, to
one of the Xk when constant values are given to the y's. Therefore, to every function
Xk there corresponds at least one function CPi of the form lik(Y," .)Xk(X, .. .).

Case 3. Both operators are incomplete.

We divide the variables into four categories: (1) the variables x, etc. that appear
in A without being present in B; (2) the variables y, etc. that appear in both A
and B; (3) the variables z, etc. that appear only in B; (4) the variables u, etc. that
appear neither in A nor in B. We denote the eigenvalues and eigenfunctions of A
by Qj and CPi(X, ... ,y, ... ), respectively, and those of B by (Ji and Xi(Y, ... ,z, ... ).
The following theorem then holds:

Theorem: If the relation AB = BA is satisfied, there will exist between the CPi
and the Xk relations of the form

Wj( x, ... ,y, . .. ,z, ... ,u, ... ) = fji(z, ... ,u, ... )CPi( x, ... ,y, ... )
(5)
= gjk(X, ... , u, ... )Xk(Y, ... , z, .. . ),
 General Theorems 59
the functions Wj forming a complete set for the set of variables X, ••• , y, .. . ,
z, . .. ,U,. . .. Conversely, if the foregoing relations are satisfied, the operators A
and B will commute.

Direct Proposition. Suppose AB = BA, and let C be a Hermitian operator act-

ing only on the variables u, etc. This operator obviously commutes with both A
and B. Consider then the operator ABC: It is a complete operator that com-
mutes with C and also with A and B, since A by hypothesis commutes with B.
Moreover, ABC is Hermitian, being the product of commuting Hermitian opera-
tors. Applying the theorem of Case 2 to ABC and to A and on designating by
wj(x, . .. , y, . .. , z, . .. , u, ... ) the eigenfunctions of ABC, which form a complete set
for the set of variables, one has

Wj(X, ... ,y, ... ,z, ... ,u, ... ) = /ji(Z, ... ,u, ... )'Pi(X, ... ,y, ... ). (6)
Applying the same theorem to ABC and to B, one likewise obtains

Wj(X, . .. , y, . .. , z, .. . , u, ... ) = 9jk(X, ... , u, . . . )Xk(Y, . .. , z, . .. ). (7)

The validity of the relations (5) are therefore demonstrated. Recalling the remarks
made at the end of our investigation of Case 2, we see that every function 'Pi satisfies
at least one relation of the form (5) and that the same is true for every function n.

Converse Proposition. We assume relations (5) to hold, the Wj forming a com-

plete set. One then has

and consequently

Similarly, one finds

and consequently

Therefore, in summary,
ABWj = BAwjj
and, since the Wj form a complete set, one gets AB = BAj q.e.d.
A particularly interesting case arises when there is no variable of the type y, that
is, no variable appears in both A and B. We then say the operators A and Bare
independent, and naturally they commute. If now one designates by A/(U, ... ) the
eigenfunctions of the operator C introduced in the foregoing argument, all products

'Pi(X, ... )n(z, ... )A/(U, ... )

60 Theory of the Commutation of Operators

will be eigenfunctions of ABC, and therefore

Wj(X, ... , z, ... , u, ... ) = 'Pi(X, ... )Xk(Z, ... )A/(U, .. .), (8)
which means that, in (5),

hi(Z, ... ,U, ... ) = Xk(Z, ... )A/(U, ... )

and
9jk(X, ... , u, ... ) = 'Pi(X", .)A/(U, .. .).
We should also remark that the operator C introduced in the preceding demon-
stration is entirely arbitrary, apart from the fact that it must act exclusively on
the variables u, etc. Consequently, when A and B commute, there exist an infinite
number of ways of writing the relations (5), depending on the arbitrary choice of
C.

2. COROLLARIES OF THE FOREGOING THEOREMS

A first important corollary of the preceding theorems is the following: "If two
complete operators A and B commute, one can, in taking as basis functions their
common eigenfunctions 'Pi, simultaneously reduce the matrices A and B to diagonal
form."
Indeed, the two operators, commuting by hypothesis, possess a common set of
eigenfunctions 'Pi such that

The ik element of the matrix corresponding to the operator A in the basis defined

k
by the 'Pi is
aik = 'Pi A'Pk dr = Ok In 'Pi'Pk dr = ok {jik;

similarly, the ik element of the matrix corresponding to the operator B is

These formulas show immediately that the two matrices A and B are of diago-
nal form, their diagonal elements being the eigenvalues of the operators A and B,
respectively.
Conversely, if, for any particular choice of the basis functions 'Pi, the matrices
A and B corresponding to two complete operators A and B take a diagonal form,
then the operators will commute.
Indeed, by hypothesis, we have

In 'Pi A'Pk dr = ak {jib In 'Pi B'Pk dr = bk {jik' (9)

 Corollaries of the Foregoing Theorems 61

Therefore, all the Fourier components of the functions A'Pk and B'Pk in the basis
formed by the 'Pi are zero except the components with indices k, which are equal to
ak and bb respectively. One thus has
(10)
The functions 'Pk are therefore simultaneously eigenfunctions of A and B, and thus,
by virtue of the fundamental theorem, AB = BA.
The demonstrated corollary can be generalized to the case where at least one
of the two operators A and B is incomplete. The reasoning is easy to give. We
consider only the case of two incomplete operators. The statement of the corollary
then becomes: "If two incomplete operators A and B commute, then it is possible
to reduce the matrices A and B to diagonal form by a convenient choice of the basis
set."
Indeed, let us take as basis functions the complete set defined in our demon-
stration of Case 3 above with the aid of an operator C acting on the variables that
appear neither in A nor in B. As, by hypothesis, A and B commute, one has the
relations (5) and consequently

aik = kwiAWkdT = kWifkl(Z, ... ,u, ... )A'PldT

(11 )

similarly,
bik =k wi BWk dT = kwi9ki(x, ... ,u, ... )BXi dT
(12)
= (3j k WiWk dT = (3j Dib

and thus the proposition is demonstrated.

Conversely, if one can, through a proper choice of the basis functions, reduce
A and B simultaneously to diagonal form, the incomplete operators A and B will
commute.
Indeed, the hypothesis now asserts that it is possible to find a complete set of
functions Wi of all the variables x, ... ,y, ... ,z, ... ,u, .. ., such that

k wi AWk dT = ak Dib k wi BWk dT = bk Dik·

One then sees that all components of the functions AWk and BWk in the set of Wi
are zero except the components with indices k, whence
(13)
The functions Wk are therefore at the same time eigenfunctions of A and B;
consequently one can, exploiting the indeterminate nature of the degenerate eigen-
functions of A and B, regard them as being proportional to an eigenfunction of A
62 Theory of the Commutation of Operators

and also to an eigenfunction of B, which allows us to write

Wj(X, ... , y, ... , z, ... , u, ... ) = !Jk(Z, ... , u, ... )CPk(X, ... , y, ... )
= 9jl(X, ... , u, ... )n(y, ... , z, . .. ).

It then follows from Case 3 (converse proposition) that A and B commute.

From the foregoing one naturally infers that, if two operators A and B do not
commute with one another, then it is impossible to reduce the corresponding ma-
trices simultaneously to diagonal form. This result enables us to demonstrate the
following elegant theorem:

Theorem: "If two operators FI and F2 commute with a third complete operator
A but not with one another, then the operator A necessarily has multiple eigenval-
ues."

Indeed, since A and FI commute, it is possible to choose a set of eigenfunctions

of A as a basis for simultaneously reducing the matrices A and Fl to diagonal form;
let CPI, ... ,CPi be this eigenfunction set. Similarly, since A and F2 commute, one can
find a set of eigenfunctions of A, say cp~, . .. ,cpi, which, when taken as a basis, will
reduce the matrices A and F2 simultaneously to diagonal form. But if the operator
A does not have multiple eigenvalues, this set of eigenfunctions would be determined
unambiguously, causing the functions cpi to coincide with the functions CPi. It would
then be possible, by the same choice of basis functions, to simultaneously reduce the
matrices A, H, and F2 to diagonal form. But this cannot be correct, since Hand
F2 by hypothesis do not commute. We must therefore conclude that A has multiple
eigenvalues.
As an example where this theory is applied, consider a spherically symmetric
system whose Hamiltonian II depends only on the distance 1" from a central point
0, taken as the coordinate origin. We have seen that the operators Mx and My,
corresponding to the angular momenta about the axes Ox and Oy, respectively, do
not commute. On the other hand, both of these incomplete operators commute with
the operator H, as may easily be verified. One thus concludes that H has multiple
eigenvalues. The quantized states of a spherically symmetric system are therefore
degenerate, a result that is well known in the study of quantification.

Operators Possessing a Common Eigenfunction. Consider two operators A and

B having an eigenfunction cP in common; this means Acp = acp and Bcp = (3cp,
whence
(AB - BA)cp = [A, BJcp = (3Acp - aBcp = O. (14)
But this equation [A, Blcp = 0 can obtain only in two cases:

(1) [A, BJ = 0, i.e., A and B commute and thus have in common an entire
system of eigenfunctions, one of which is cpo

(2) The operator [A, BJ is not identically zero but has an eigenvalue O.
 Simultaneous Measurement of Two Quantities 63

In this case, A and B, though not commuting, will have in common at least one
eigenfunction c.p, if c.p denotes any eigenfunction of [A, B] belonging to the eigenvalue
o.
As an example of this second possibility, consider again the operators Mx and
My. We have
h
[Mx, My] = -. M z f:. o.
2n
The operators do not commute, but Mz has the set of eigenvalues m(h/27r), with
m = 0,1,2, ... , which includes the value o. Accordingly, [Mx, My] possesses an
eigenvalue 0, and the operators Mx and My share the eigenfunction c.p = const., as
can be verified immediately: c.p = (27r)-1/2 exp (-imc.px) = (27r)-1/2 for m = o.
An important situation arises when [A, B] is a constant times the unit operator
1. This is the case of canonically conjugate quantities, for which [A, B] = (h/27ri)1.
Since the unit matrix clearly does not have any eigenvalue 0, the operators A and
B are not permitted to have any common eigenfunction.
The foregoing considerations, we may note, are valuable in the study of contin-
uous spectra.

3. SIMULTANEOUS MEASUREMENT OF TWO QUANTITIES

IN WAVE MECHANICS

We are now going to use the theorems and corollaries demonstrated above to
study questions concerning the simultaneous measurement of two quantities.
In the new mechanics a Hermitian operator is associated with every mechanical
quantity. Given a quantity and the corresponding operator A, it is very important to
distinguish quantities whose operators commute with A from those whose operators
do not commute with A. The importance of this distinction stems from the fact
that two mechanical quantities can be measured simultaneously when their operators
commute, and only in that case. This is what we now wish to show.
We shall start from the essential postulate of wave mechanics that every state
of a particle or a system of particles must at every instant be represented by a
wave function 1/J, which in reality represents the state of our knowledge about this
particle or system at that instant. Every act of measurement or observation on
the microscopic elements modifies the state of our knowledge about the particle
or system and consequently abruptly modifies the form of 1/J. But also before the
act of measurement, as afterwards, one must be able to represent the state of the
particle by a wave 1/J; this is the fundamental postulate that makes wave mechanics
possible and of which it is essential to take note. Immediately after a measurement
or observation revealing something about the state of the atomic elements that are
inaccessible to our senses has been made, one can adopt a certain form of the wave 1/J
representing the state of one's knowledge. And from this moment on, as long as no
further observations or measurements are performed, the wave 1/J will evolve, starting
from this initial form, in accordance with the wave equation of wave mechanics, i.e.,
in a manner that is rigorously determined.
If at a subsequent time a new measurement or observation allows us to assign
64 Theory of the Commutation of Operators

a precise value (Xi to a quantity A, this value will, according to our general princi-
ples, be one of the eigenvalues of the operator A, and the wave functions 'IjJ after
the measurement must be proportional to an eigenfunction 'Pi belonging to (Xi. If,
immediately after having measured A, one performs a new measurement of A, one
will from general principles be sure to recover the value (Xi (repeatability of mea-
surements). Therefore, in order that one would be able to measure simultaneously
and accurately both the quantity A and another quantity B, with eigenvalue (3i and
eigenfunction Xi, the wave function 'IjJ must after the measurement be proportional
to one of the 'Pi as well as to one of the Xi, otherwise the representation of the state
of our knowledge after the measurement by a wave 'IjJ would not be possible.
Let us apply this idea first to two complete operators A and B. In order for
the corresponding quantities to be simultaneously measurable with full precision, it
is necessary that, after the measurement, one should have 'IjJ = ai'Pi = biXi, with
la; I = Ib, I = 1, a convenient unambiguous correspondence having been established
between the (Xi and the (3i (in such a way that corresponding quantities will have
the same index).
The set of 'Pi must therefore coincide with the set of Xi, and we know that the
necessary and sufficient condition for such an agreement is AB = BA: Precise and
simultaneous measurements of A and B is possible only if their operators commute.
The two measurements are then fully connected with each other: Knowledge of the
result of one entails knowledge of the result for the other, at least in the absence of
multiple eigenvalues.
Let us next consider the situation where A is complete and B incomplete (Case 2,
studied at the beginning of this chapter). We would like to be able after the mea-
surement to have
'IjJ = ai'Pi = fik(Y,·· ·)Xi(X, .. .),
with la; I = 1 and
j ... j Ifikl2dy = 1,
for every i. Now we know that the necessary and sufficient condition for this to
happen is again AB = BA. But here the two simultaneous measurements are never
fully linked. Indeed, we have seen that to one value of k there may correspond
several values of i. If therefore one knows the result (3k of a measurement B, this
knowledge does not in general imply knowledge of the value (Xi of A.
Let us also consider the situation where both A and B are incomplete, using the
notation of Case 3, studied earlier. For A and B to be simultaneously measurable
with full precision, one must after the measurement have
'IjJ = fji(Z, ... ,u, ... )'Pi(X, ... ,y, ... ) = gjk(x, ... ,U, ... )Xk(z, ... ,y, ... ), (15)
for every i, along with the conditions

j ... jlf;i(Z, ... ,u, ... )1 2dUdZ = 1, j ... jI9ik(X, ... ,U, ... ) 12 dUdX = 1. (16)

For this to be true in Case 3, it is necessary and sufficient that AB = BA. Here
the connection between the two measurement is weaker than in the preceding case,
 Simultaneous Measurement of Two Quantities 65
because several values of j may correspond to the same value of i, and several values
of j may correspond to the same value of k. Thus knowledge concerning the result
of one of the measurements does not in general follow from knowing the outcome of
the other measurement.
Finally, suppose that A and B are independent operators. They then will nec-
essarily commute, and if an operator C acting on the variables absent from A or
B is added to them, one will obtain a complete set of basis functions for all the
variables on using the eigenfunctions Wj of the complete operator ABC. After a
measurement, the function 'IjJ will be reduced to one of the wi> i.e., will have the
form
(17)
where lejl = 1, with i, k, I having some or other integral values. This means that it
is always possible to simultaneously measure the quantities A and B and that the
results of the two measurements are totally independent; knowing one of the results
teaches nothing about the other.
In summary, the necessary and sufficient condition for two quantities A and B
to be simultaneously measurable is that their operators commute. The results of a
simultaneous measurement of A and B, if possible, are more or less connected with
one another, depending on the complete or incomplete character of the operators.
Consideration of cases of independent operators leads to the notion of "maximal
measurement." Suppose that a particle or system of particles is defined by the n
coordinates Xl, . .• , x n . With each coordinate Xi we associate a measurable quantity
whose operator Ai involves only the variable Xi. Let o:~ and tp~ be the eigenvalues
and eigenfunctions, respectively, of Ai.
We obtain a complete operator on considering the product of all Ai, i.e.,
n

II Ai,
1

the eigenfunctions of which are the products

n
Wj = II tp~(Xi).
1

The operators Ai being independent, the corresponding quantities are simul-

taneously measurable. Suppose that we measured all of them in the same act of
measurement. After this measurement, the wave function 'IjJ will have the form
n
Wj = II tp~(Xi)'
1

assuming that the measurement of Ai has furnished the eigenvalue o:~ corresponding
to the eigenfunction tp~. One will then have performed a "maximal" observation or
measurement, which determines completely the function 'IjJ and, consequently, the
probability of every possible value of all the measurable quantities belonging to the
66 Theory of the Commutation of Operators

system. The measurement of another measurable quantity B at the same time as

the measurements of the Ai is either impossible, if B fails to commute with the
product of the Ai, or possible, in the contrary case, but of no interest, because then
this measurement will not teach us anything more about the state of the system,
for which the simultaneous measurement of all the Ai has already maximized our
knowledge.

4. EXAMPLES OF QUANTITIES THAT ARE NOT

SIMULTANEOUSLY MEASURABLE. DISTINCTION BETWEEN
TWO KINDS OF NON-COMMUTATION

The best-known example of quantities that cannot be measured at the same

time is provided by a coordinate q and the conjugate component p of the linear
momentum. If Q and P designate the corresponding operators, one has

Q =qx, P=---,
h a h
QP-PQ=-2., (18)
27ri aq 7rZ

because
q(-~) af(q) _ (_~) ~ [qf(q)] = ~ f(q).
27rZ aq 27rZ aq 2n
The quantities q and p are therefore not simultaneously measurable. If there are
several q's, say ql, .. . ,qi, . .. , then naturally QkPi = PiQk, and, since

we see that it is always possible to measure simultaneously two coordinates or two

components of momentum, as well as one coordinate and a non-conjugate compo-
nent of momentum. Only the simultaneous measurement of a coordinate and the
corresponding component of momentum is ruled out.
In the case of a particle described by three coordinates x, y, z and three conjugate
rectangular components of linear momentum Px,Py,Pz, one recovers the prohibition
against simultaneously knowing the conjugate quantities x and px, etc. which we
earlier managed to deduce from the representation of the waves 'IjI by Fourier inte-
grals. The uncertainty relations of Heisenberg are derived-we shall return to this
point-from the relations QP - PQ = h/27ri. Quite generally, if, in a problem of
mechanics, p and q are canonically conjugate variables, then QP - PQ = h/27ri
will always obtain. Thus the azimuthal angle 'Pz about a polar axis Oz is canon-
ically conjugate to the component Mz of the angular momentum along Oz. Since
the operator -(h/27ri)(a/a'Pz) corresponds to M z , the operators indeed satisfy the
equation 'PzMz - Mz'Pz = h/27ri.
In the case just studied, the two noncommuting operators are such that [A, B] =
c, c being a constant which here equals h/27ri. The canonically conjugate quantities
in question belong to the general category of quantities that are not simultaneously
measurable and for which the commutator equals a constant. But there exists
 Quantities That Are Not Simultaneously Measurable 67

another category of quantities defying simultaneous measurement: those for which

the commutator equals a nonzero operator.
This is for instance the case for the quantities Mx and My, whose operators obey
the rule
h
[Mx, My] = -2. Mz.
7rZ
The essential difference between this type of noncommuting operators and the pre-
ceding type results from the theorem stated on p. 62. Indeed, two noncommuting
operators A and B cannot have any eigenfunction in common if the commutator
equals a constant c, because the operator cl does not have a zero eigenvalue. Thus
the quantities corresponding to two such operators never are simultaneously mea-
surable.
By contrast, should the commutator of the two noncommuting operators A and
B equal an operator, then A and B will have some eigenfunctions in common pro-
vided [A, B] possesses some null eigenvalues. The simultaneous measurement of A
and B can then accidentally become possible and, if so, will furnish as values for A
and B some eigenvalues belonging to one of their common eigenfunctions. For ex-
ample, in the case of Mx and My, whose commutator is proportional to Mz and thus
allows the eigenvalue 0 for the eigenfunction <Po = const. = (27r )-1/2, a simultane-
ous measurement of Mx and My can by exception enable us to attribute to them the
values Mx = 0 and My = 0, corresponding also to the eigenfunction <po = (27r)-1/2.
But in general it is impossible to measure Mx and My simultaneously.

Note L. B. In reality the eigenfunction of Mz for the eigenvalue 0 is of the

form f(p,z) in cylindrical coordinates around Oz, and the eigenfunctions of Mx
and Mz for the eigenvalues 0 can be expressed analogously, so that the common
eigenfunction of Mx, My, Mz for the eigenvalue 0 will have the form F(r) = F(p2
+ z2): It represents a state of spherical symmetry.
When studying the theorem dealing with the dispersions of two noncommuting
quantities, we shall again see the importance of distinguishing noncommuting quan-
tities whose commutators are constant from those whose commutator equals some
operator.

Note. It should be noted that, if [A, B] = cl, the constant c will always be
proportional to h, because A and B must commute when h ~ 0, in order that one
would recover classical mechanics in this limit. The situation under consideration
can therefore always be reduced to the case of canonically conjugate quantities.
 CHAPTER 7

PHYSICAL IMPOSSIBILITY
OF SIMULTANEOUSLY MEASURING
CANONICALLY CONJUGATE
QUANTITIES

1. NECESSITY TO EXAMINE THE IMPOSSIBILITY

OF SIMULTANEOUSLY AND PRECISELY MEASURING
TWO CANONICALLY CONJUGATE QUANTITIES

We have shown that, according to wave mechanics, it must be impossible to

measure simultaneously with full precision two noncommuting quantities and, in
particular, two canonically conjugate quantities. This impossibility was deduced
from the fundamental postulate according to which it must be possible to represent
the state of our knowledge about a system by a wave 1/;, after as well as before an
act of measurement. If two canonically conjugate quantities were simultaneously
measurable with full precision, it would be impossible to represent the state of the
system after the measurement by a wave 'l/J, and one would be forced to abandon
wave mechanics.
But one may wonder if it is really impossible to carry out such a simultaneous
measurement of two conjugate variables and what the physical origin of this impos-
sibility is. Subtle analyses developed first by Bohr and Heisenberg have shown that
it actually does not appear possible to imagine observations allowing the simulta-
neous measurement of two conjugate quantities with an accuracy greater than that
permitted by the uncertainties of Heisenberg. The lengthy discussions that stirred
up the arguments of Bohr and Heisenberg have been turned to their advantage, and
today it appears that their thesis is agreed to by all physicists who have seriously
studied the question.
These arguments are moreover of interest because they demonstrate that the
impossibility of the precise simultaneous measurement of two conjugate quantities
has its origin in the existence itself of the quantum of action measured by Planck's
constant h.
As Planck's constant h is negligible from the macroscopic point of view, the
simultaneous measurement of two conjugate quantities is possible in practice for
macroscopic phenomena, because in this domain the imprecision of measurements
masks the quantum uncertainties. But at the level of elementary-particle phenom-
 The Heisenberg Microscope 69
ena, h is no longer negligible, and the quantum uncertainties will play an essential
role.
We shall now study one of the examples given by Bohr and Heisenberg.

2. THE HEISENBERG MICROSCOPE

Heisenberg developed a celebrated argument, known under the name of Heisen-

berg's microscope, by supposing that a microscope is used to observe an electron
placed on a slide. Obviously this experiment cannot be realized, but one may present
it in a form simulating more nearly what is achievable in practice.
Consider an optical or corpuscular (electronic) microscope, and suppose that
we examine with its aid an object of mass M sufficiently small to be regarded as
pointlike, which is placed on a slide resting on the microscope stage.

71

I
pi

The object is "illuminated" by particles of the same energy traveling parallel

to the axis of the microscope. Let p be their linear momentum; the associated
wavelength is then A = hlp. The incident particles are photons, if the microscope
is an optical instrument, or electrons (or perhaps protons) if the microscope is
corpuscular.
If the microscope were perfect, that is, if aberrations and diffraction effects were
negligible, then a point image P in the object plane 1[' would correspond to the
point object M. Aberrations can be made very weak (in an optical microscope by
a suitable choice of lenses, in a corpuscular microscope through the use of a very
small angular aperture 2c:). But one can never suppress the diffraction phenomenon
which is due to the passage of the wave 'IjJ, associated with the illuminating parti-
cles, through the limited aperture of the apparatus. The theory dealing with the
resolving power of microscopes teaches that any observation of the point image al-
lows us to determine the position of the point object on the x axis only with an
uncertainty equal to 8x = A/2 sin c:. If there were no diffraction (nor aberrations),
the arrival of a particle at P (a phenomenon observable in principle) would allow us
to assign a precise position to the point object M. But the unavoidable interven-
tion of diffraction has as a result that the arrival of a particle at P will permit the
localization of M on the x axis only with an uncertainty 8x = hI2sinc:. It is seen
that this uncertainty exists independently of the intensity of the illuminating beam,
since one can evaluate it on the assumption that only a single particle is scattered
70 Measuring Canonically Conjllgate Q'u.antities

by the punctual object.

The scattering of any particle by the object is the result of a brief interaction,
or impact, between the object and the particle. In the course of this interaction,
the exchange of linear momentum between the particle in motion and the object,
assumed to be initially at rest, must be minute, otherwise the wave associated with
the scattered particle would have a wavelength different from that of the incident
particle, and one would not any longer obtain a regular image. To treat the problem,
one would even need to consider the system formed by the object and the particle
and envisage the configuration space of this combined system. We may therefore
assume that the linear momentum p of the scattered particle has a magnitude
Ipi = hiA.
After the impact, the scattered particle has a linear momentum pi that makes an
angle a with the original direction of motion (i.e., direction of the microscope axis);
and since the scattered particle must enter the microscope in order to be involved
in any measurement, the inequality lal < e: holds. Let finally Px be the component
along Ox of the linear momentum of the object after the impact. Then one can
write the conservation of linear momentum along Ox for the system "object plus
incident particle" as

Px=p'sina~psina= (~)sina. (1)

P x therefore has a value lying between

-(~) sine: and + (~) sine:.

The uncertainty in the value of P x is therefore (2hl'\) sin e:. After having ascer-
tained the arrival of a particle at P, the observer will thus know the abscissa x and
the component Px of the linear momentum of the object only with the uncertainties
,\
8x = - - (2)
2 sin e:'
whence 8x 8Px ::::: h.
We write the "larger than or equal to" sign, because the equality sign assumes
all observations to be perfect, aberrations to be absent, etc.
We thus have recovered the uncertainty relation for the canonically conjugate
quantities x and P x ; and one sees, at least in the case here studied, that these
quantities can never be determined with full precision.
Our reasoning has shown that this circumstance is due to the finite value of the
constant h. Should one want to increase the precision of x, the wavelength ,\ must
be diminished by employing faster incident particles. But then (and it is here that
the finite value of the quantum of action enters) the linear momentum p of the
incident particles will increase in magnitude, since Ipi = hl'\ and h is finite. Thus
8Px will increase, and one is still left with the Heisenberg relation. Our inability
to break the link created between 8x and 8Px by the existence of the quantum of
action thus appears as the profound source of the Heisenberg relation, which leads
 Measurement of the Speed of an Electron 71

us to believe that one must be able to recover this relation whatever measuring
mechanism is employed.

3. MEASUREMENT OF THE SPEED OF AN ELECTRON

BY MEANS OF THE DOPPLER EFFECT

Let us now study the measurement of the speed of an electron by the Doppler
effect. Suppose that an electron is displaced with the speed v along the x axis in
the positive direction. We expose this electron to a train of light waves of mean
wavelength A propagating along the x axis in the negative direction. If scattering
takes place, the scattered photon will undergo a reversal of its velocity and return
along the x axis, traveling now in the positive direction. Suppose this happens and
that we are able to measure exactly the scattered frequency Vi. For simplicity, we
assume the speed of the electron to be much less than that of light and write the
equations expressing the conservation of energy and linear momentum as

h I hv hv l
hv + 2"1 mov2 = v + 2"1 mov 12
, mov - - = mov
I

c
+-,
c
(3)
Vi being the final speed of the electron. On eliminating Vi between these two equa-
tions, one gets

I
h(v - v) = -
1 "2 (v
[h2 + v) 2 - I
2mov -h iv)] .
(v + (4)
2mo c c
We now put Vi = V - 15 while noting that 15 is small and thus that cv I c and 15 2
are negligible; consequently
chv hv 10- 13
with
moc2 ' moc2 '" 10- 6 '
is also small. Finally, one thus finds
hv 2 v
15 = 2 - - -2-v,
moc 2 c
from where
Vi = V _ 15 = v [1 _moc
2hv 2 + 2V].
c
(5)

This equation summarizes in the case under study, to within the acknowledged
approximations, both the Doppler effect, represented by the term 2vlc, and the
Compton effect, represented by the term -2hvlmoc2 • The Compton effect perturbs
the velocity of the electron; thus, if one wants to measure the latter exactly, using
the Doppler effect, the Compton effect must be made negligible compared with the
Doppler effect, which requires
hvlmoc2 h
=
vic movA
72 Measuring Canonically Conjugate Quantities

to be very small. Then the Doppler effect alone will be significant, and we may
write
v' = v (1 + 2;) and >.' = >. (1 _ 2;).
But the wave train unavoidably has a finite length 1. Consequently it is not rigor-
ously monochromatic; and if we introduce the wave number 1/>', this number will
vary for the different monochromatic waves ofthe train by the amount 1i(1/ >'), with
1i(1/ >.) ~ 1/1 according to the theory describing the Fourier-integral representation
of wave trains. Therefore, even if >" is measured without any experimental error,
there will remain an uncertainty about the value of v, because the latter is given by

the uncertainty in >. consequently produces an uncertainty in v equal to

liv = -2c >. ,Ii (1)

-
>"
so that, after the measurement, the uncertainty in the linear momentum of the
electron along Ox becomes
me>.
lipz ~ 21. (6)

But the simultaneous measurement of the coordinate is also subject to an uncer-

tainty. Indeed, the Compton effect, although by hypothesis weak compared with
the Doppler effect, exists nevertheless and causes, as we have seen, a variation of
the velocity equal to v' - v ~ -2hv/moc = -2h/mo>'. Suppose, to take the most
favorable case, that the initial position of the particle is well known. A cause of
uncertainty about its position after the measurement will then exist due to the fol-
lowing circumstance: One does not know at what instant of the time interval 1/ c,
which is how long it takes the wave train to pass over the electron, scattering actu-
ally occurs; and this lack of knowledge gives rise to an uncertainty lix in the final
position of the electron equal to

, 1 2h 1
lix=(v-v)-=--. (7)
c mA c
In the most favorable case, one therefore gets

me>. 2h 1
lixlipz ~ -1- ~ - = h,
2 mA c

thus recovering the minimum Heisenberg uncertainty.

 Passage of a Particle Through a Rectangular Aperture 73
4. PASSAGE OF A PARTICLE THROUGH A RECTANGULAR
APERTURE

As another example, we consider again the determination of the position of a

particle, now achieved by letting the particle pass through a rectangular opening
of side lengths 2a and 2b cut in a plane screen. In trying to better define the
coordinates of the particle, one is led to use a very small opening; but the more the
sides 2a and 2b of the rectangular aperture are decreased, the more one will augment
the importance of the diffraction phenomena, which, according to the ideas of wave
mechanics, are engendered by the passage of the particle across this opening.
For the purpose of introducing here the fourth uncertainty relation, which so
far has not been mentioned but which will be studied more completely later on, we
imagine that one determines the instant at which any particle traverses the opening
with the help of a mobile shutter which is capable of instantaneously opening or
closing the aperture. The more one lowers, by quick operation of the shutter, the
time during which the aperture remains open, the better the instant of the particle's
passage can be determined. But, at the same time, the wave train associated with
the particle will be shortened proportionally; thus the monochromaticity of the wave
train will be diminished, causing the energy of the particle to become less and less
well defined. In this way the fourth uncertainty relation oW Or 2:: h will arise.
To examine the problem mathematically, choose the coordinate origin 0 at the
center of rectangular opening, the coordinate axes Ox and Oy parallel respectively
to the sides 2a and 2b of the opening, and the z axis perpendicular to its plane,
in such a way that the positive z direction and the direction of propagation of the
light will coincide. Let M, with coordinates X, Y, 0, be a point in the opening and
dX dY a small rectangle surrounding it. As we know, the Huygens principle allows
one to calculate the amplitude of the elementary wave emanating from the small
rectangle dX dY in a direction defined by the direction cosines a, /3" and making
a very small angle with Oz. If x, y, z denote the coordinates of a very remote point
in the direction a, /3", then the elementary wave in question has the amplitude

d.,pa{3 = K dX dYexp [27ri(vt _ a(x - X) +i(Y - Y) + Z)], (8)

where K is a coefficient that varies with the direction a, /3" but much more slowly
than exponentially; we assume that , ~ 1. The resulting wave traveling in the
direction a, /3" will therefore have the amplitude

/3y + Z)] '

.,pap = C exp [27rZ. ( vt - ax + A (9)

with

(10)
74 Measuring Canonically Conjugate Quantities

the integral being extended over the rectangular opening. The symmetry of the
aperture immediately requires that B = 0, so that C reduces to

C j
r{
= A = J]( cos 211"
aX +,8Y
,\ dX dY. (11 )

Herein the cosine can be replaced by a product of cosines plus a product of sines;
and, since the latter gives a zero integral, we are left with
r 211"aX {b ,8Y
A = 4]( Jo dX cos - , \ - Jo dY cos 211" T
(12)
[0 2 • 211"aa . 211",8b
= 1I" 2 a,8 sm -,\- sm -,\-,

whence
0/' _
'Pa/3 -
[0 2 • 211"aa . 211",8b
a,8 sm -,\- Sill - , \ - exp
1I" 2
[2'(vt -
1I"Z
ax +,8y +
,\
z)] . (13)

The amplitude 1/;",/3 is therefore zero for 211"(aa/>..) = h and for 211"(,8b/>..) = h,
with k integral; i.e., in the directions given by
k'\
a= - ,8=kA.
2a' 2b
By contrast, the function 1/;a/3 is a maximum in the directions for which
2k
a= - - - -
+1
(l = 2k + 1 ~
,\
2 2a' JJ 2 2b'
One thus obtains what is called a diffraction phenomenon localized at infinity. To
observe it, at least in the optical domain, one may place a lens behind the screen in
such a way that its optical axis coincides with Oz. If there were no diffraction, one
would only observe an image of the rectangular opening situated on the optical axis
in the focal plane of the lens. But, due to the existence of plane monochromatic
waves with normals inclined to the axis, one will also obtain a series of other images,
corresponding to the maxima of 1/;"'/3' The intensity of these images decreases as the
integer k increases, since a and ,8 appear in the denominator of the expression (13)
for 1/;a/3'
In summary, the plane wave falling on the screen has the form

1/; = a exp [211"i (vt - I) ], (14)

but passage of this wave through the rectangular opening transforms it into a set of
plane waves with normals slightly inclined to the z axis and of the form

" a( a,,8) exp [211"z. ( vi - ax

1/; = L.." + ,\,8y + z)] , (15)
01,/3
 Passage of a Particle Through a Rectangular Aperture 75

wherein the partial amplitudes a( a, /3) represent a function of a and /3 which exhibits
successive maxima and minima. Since the intensity of successive orders of these
extrema diminishes rapidly, the extension of the wave group with respect to the
variable a is measured by Sa = kl (>../2a) ~ A/2a, with kl designating a small
positive integer corresponding to the highest order of diffraction whose intensity is
still appreciable. Similarly, the extension of the wave group with respect to /3 will
be given by S/3 = k2(A/2b) ~ A/2b.
If now I' denotes the wave-number vector belonging to the wave diffracted in
the direction of propagation a, /3, [' i.e., a vector of magnitude (1/ A) pointing in the
direction a, /3, [' then one has
a /3 [1
ftx = X, fty = X, ftz = I ~ I' (16)
The maximal variation of ftx and fty are Sftx = Sa/A and Sfty = S/3/A, respec-
tively, whence
1 1
Sftx ~ 2a' Spy ~ 2b' (17)
But the position of the particle while passing through the opening is known with
the uncertainties Sx = 2a and Sy = 2b; consequently one gets
SpSx ~ 1, bftSy ~ 1.
Moreover, the fundamental relation ipi = h/ A corresponds to the vector equation
p = hI', so that one can finally write
Spx {jx ~ h, SPy by ~ h,
that is, once more the Heisenberg inequalities are recovered.
On the other hand, should we want to determine both the coordinate z of a
particle and the time t when it passes through the opening, a movable shutter
must be employed, as explained above. Let 7 be the time during which the shutter
is open. Obviously the uncertainty in t is then equal to 7 and that in 7 to U 7, U
being the group velocity of the wave 'I/J, which, as we know, equals the velocity of
the particle. Therefore,
bt = 7, Sz = U7.
But, in leaving the aperture open only for the time 7, one permits only a limited
wave train to pass through the opening, in other words, a train composed of waves
occupying a frequency range having a width at least of the order 1/7; i.e., Sy ~ 1/7.
One therefore gets .
S(.!.)
A
=o(I/A) by> _1 , since ~ = o(I/A)
Oy - U7 U OY
Practically one thus will have Sy ~ 1/7 and S(l/A) ~ I/U7. But, according
to the general principles of wave mechanics, the uncertainty in the final energy of
the particle will be hSy, and the uncertainty in the component pz of the linear
momentum is Mftz ~ M(1/ A). We therefore find
SW 8t ~ h, Spz bz ~ h.
These are the two other uncertainty relations of Heisenberg.
76 Measuring Canonically Conjugate Quantities

5. IMPORTANT REMARK ON THE MEASUREMENT OF SPEED

We have just established in some examples that the measurement procedures for
two canonical conjugate quantities lead to the inequalities of Heisenberg.
One may however be tempted to raise the following objection: Suppose some
experiment performed at the time t1 shows that the particle is situated in the
neighborhood of a point A of space, while another experiment, performed at a
later time t2, reveals that the particle then finds itself in the neighborhood of a
different point B. If the time t2 - t1 is sufficiently long, it will seem that the posit
v = AB / (t2 - t1) would give an excellent estimate of the particle's speed, and it will
seem permissible to say that one knows both the position and the linear momentum
of the particle in a precise manner, which assertion conflicts with the relations of
Heisenberg.
But this conflict is only apparent. We must, in fact, first remark that, if one
performed the envisaged measurement of speed on a large number of particles in the
same initial state, then different results would be found. Indeed, it can be demon-
strated that the very concentrated wave train t/J, characterizing the localization of
the particle close to A in the first experiment, will spread out rapidly while prop-
agating and consequently have a large extension at the end of the, by hypothesis
very long, time interval t2 - t1' Thus, according to the interference principle, the
particle can at the time t2 be present anywhere inside a large region of space, which
means that the ensemble of envisaged experiments would furnish a set of different
points B
Moreover-and this is the main point-we cannot claim that the envisaged ex-
periments would enable one to simultaneously know both the position and the linear
momentum of the particle. In fact, first of all, the speed v = AB / (t2 - tt) will ob-
viously be known only after the second observation: One can therefore not say that
simultaneous knowledge exists at the instant t1 of both the position and the speed.
Does this kind of knowledge exist at the instant t2? Well, the second observation
indeed furnishes the position of the moving particle and allows us, if we wish, to
attribute to it during the given time interval a rectilinear trajectory AB described
with the speed v = AB/(t2 - t1)' However, what is important is to know the linear
momentum of the moving particle after the second observation. But observation
of the position B disturbs completely the state of motion, in such a manner that
one can in no way assign the calculated speed v to the particle localized at B; thus
we are barred from using this speed in any prediction of the later evolution of its
motion. The speed v is known only at the moment when it does not represent
anything. Wave mechanics, like all other physical theories, has prediction as its
goal and is therefore always oriented towards the future. Thus what concerns it is
the state of our knowledge after every observation. But after the second observa-
tion, as after the first, if one knows exactly the position of the particle, one will
be completely ignorant of its velocity. Even the hypothesis attributing retrospec-
tively a value v to the speed in the interval (tl, t2) of time is arbitrary, because,
no observation having taken place inside this interval, any claim that the particle
has carried out a uniform motion along a straight line AB constitutes an arbitrary
assertion.
 The Case of Two Operators with a Nonzero Commutator 77
6. THE CASE OF TWO OPERATORS WITH A NONZERO
COMMUTATOR

We have examined attempts to simultaneously measure two canonically conju-

gate quantities, and we have found that the attainable precision is always limited
by the relations of Heisenberg. But these conjugate quantities belonged to the first
category of noncommuting quantities, i.e., those with constant commutators. Can
one arrive at some analogous conclusions for non commuting quantities of the sec-
ond kind, i.e., those with commutator equal to some or other nonzero operator?
We shall examine this question in the most important case from the physical point
of view, namely, that of the components Mx, My, Mz of the angular momentum
M.
To take a very simple situation, consider an electron (of charge e and mass m)
which is describing a circular orbit of radius R with speed v and thence is char-
acterized by a quantized magnetic moment of one Bohr magneton. The magnetic
moment M and angular momentum M associated with this orbital electronic cur-
rent are given, respectively, by

M = movR, (18)

assuming that v <t: c; from where results the well-known formula which, as Ein-
stein has shown, is valid under neglect of spin for any system of charges in mo-
tion:
M e
M---
eh
-=--, (19)
M 2moc - 47rmoc'
if we take M = h/27r.
We now wish to determine M by measuring M through the action of the
underlying electronic current on a magnetometer placed at a distance r from
it.
If Jl is the magnetic moment of the magnetometer needle, the field produced
by the magnetometer at the position of the electronic current is of the order fllr3.
On the other hand, in order to measure M exactly, we must with the aid of the
magnetometer exactly evaluate the magnetic field caused by the electronic cur-
rent at the position of the magnetometer, a field that is of the order M/r 3 . It
will therefore be necessary to know this field with an uncertainty tlH = (M/r 3)1]
such that 1] <t: 1, i.e., know the energy Jl' H of the magnetometer with an un-
certainty of the order tlE = fltlH = fl(M/r 3)1]. The fourth Heisenberg re-
lation then demands that the duration of the experiment be at least equal to
tlt = hi tlE = hr31 flM1]. But, during this period of time, the electronic cur-
rent, acted on by the magnetic field '" fl/r3 = H' of the magnetometer, will
precess, about this field with an angular speed equal to the Larmor frequency
w = eH'/2moc. Indeed, if the electronic current is represented by an equivalent
tiny magnet placed in the field H', one will have, according to the theorem for an-
gular motion, dM / dt = torque about 0 (see figure) of the force exerted by H'
on M. The torque being perpendicular to H', the angle () between M and H' is
78 Measuring Canonically Conjugate Quantities

constant, since d(M cos ())/dt = 0; and, projecting on the plane normal to H', we
get
!i
dt
[M sin () { C?S
sm rp
rp}] = H'M sin () { - cos
sin rp }
rp ,
(20)

that is,
M { - sin rp } w = H'M { - sin rp } (21 )
cos rp cos rp ,

w being the angular speed with which the vector M precesses about the direction
of H'. One therefore finds indeed

w-H,M _ eH'. q.e.d.

- M - 2moc'

The rotation around H' performed by the axis of the electronic current, i.e., by
the vector M , during the time taken up by the experiment will thus be

e p, hr 3 eh 1 1 21r
a=wt!.t ~ - - - - - = ----- = - ~ 21r.
2moc 1'3 p,MTf 2moc M Tf Tf
During the experiment, the axis of the electronic circuit accordingly undergoes
a large number of revolutions about the direction of H', and consequently only the
component along H I of M , and thus of M, can be measured. We therefore see
that the measurement of one component of M cannot be made at the same time
as that of another component, conforming to the fact that the components of M
do not commute. Nevertheless, if M and M are zero, one can verify with the help
of the magnetometer that the three components of each of these vectors are zero:
In this exceptional case, the simultaneous measurement of all components becomes
possible; and this is again in agreement with the predictions of the general theory.
 Bohr's Complementarity 79
7. BOHR'S COMPLEMENTARITY

We are now going to clearly state a point that is important to note if one is to
have a good understanding of certain expositions. In elementary optics texts, the
name "wave" is generally given to plane monochromatic waves, because the usual
light wave train, although limited, is long enough for its central part to appear like
a plane monochromatic wave. A "wave" thus defined will have a well-determined
frequency, wavelength, and direction of propagation: Wave mechanics associates
with it a vector p pointing in the direction of propagation and having a magnitude Ipi
given by the wavelength A = h/lpl, from which the frequency can be deduced. The
associated wave will therefore be defined by the vector p. This plane monochromatic
wave, which is homogeneous and does not permit any localization of the particle, is
an idealization of the idea of pure motion untinged by any feature of spatio-temporal
localization.
By contrast, the coordinates x, y, z of a particle correspond to the idea of spatio-
temporal localization at an instant t. The canonically conjugate variables Px,Py,Pz
and x, V, z pertain respectively to the wave aspect of the entity "particle," an as-
pect that is purely dynamical without localization, and to a granular aspect of the
particle, which in a certain sense excludes the idea of motion (Zeno of Elea). Re-
ferring then to the Heisenberg inequalities, one sees that the elementary particles
of physics can be described by either a plane wave or by a localized lump only in
extreme cases: In general, the aspects of wave and localized lump coexist, but both
features are a bit blurred, the associated wave tP being formed by a superposition
of a certain number of plane monochromatic waves and the localization remaining
uncertain within a region that is more or less extended in space.
Heisenberg's uncertainty relations also teach that the more an observation al-
lows the precise description of one aspect of a particle, the more blurred the
other aspect becomes. This fact enables us to explain how wave mechanics can
permit the simultaneous use of the two apparently contradictory notions of a
plane homogeneous wave of infinite extension and a localized lump: These two
pictures, so different in character, can never give rise to a flagrantly contradic-
tory situation, since anyone of them tends to fade away as the other asserts
itself. Here we have a very interesting aspect of the conceptions of modern mi-
crophysics. Bohr has expressed it by saying: "The wave and the particle are
'complementary aspects' of reality." Every time the behavior of the particle en-
tity is able to manifest itself by the propagation of a plane monochromatic wave,
its granular aspect disappears; and every time its behavior can be represented
by the displacement in space of a localized lump, its undulatory aspect will dis-
appear. The concept of "complementarity" thus introduced by Bohr is very cu-
rious: It could be, as Bohr himself indicates, that it has applications outside
physics.

Note G. L. Need we recall that the opinion of de Broglie on complementar-

ity subsequently underwent an evolution. One finds, for example, in Certitudes
et incertitudes de la science the following passage: "If the use of the word 'com-
plementarity' only serves to interpret the successive apparition of corpuscular and
undulatory manifestations in undeniable phenomena, then such a use is entirely le-
80 Measuring Canonically Conjugate Quantities

Crystal

Electron gun

~
Photographic plate

gitimate; but, on the other hand, it does in no way constitute a real explanation
of the duality of waves and particles. One may compare complementarity to the
'dormitive virtue' of opium, ridiculed by Moliere: It is perfectly legitimate to inter-
pret the soporific properties of opium by attributing to this substance a 'dormitive
virtue,' but it is well to guard against seeing in these words an explanation of these
properties." (Ref. III, 8, p. 20.)

To illustrate the idea of complementarity, let us take as a concrete example the

phenomenon of electron diffraction by a crystal. The electrons are produced by
an "electron gun," a device consisting of a hot filament, emitting electrons, and a
system of plates maintained at appropriate potentials, which impart to all electrons
the same acceleration in the same direction. One therefore obtains a cylindrical
beam of mono-energetic electrons. This beam is directed onto the surface of a
crystal, and the scattered electrons are allowed to strike a photographic plate, where
they produce punctual impressions on a sensitive layer.
The right-hand section of the electron gun is supposed to be infinitely large in
comparison with the wavelength of the wave associated with each electron. Every
electron leaving the gun therefore has a perfectly determined linear momentum,
since it is acted on by a known potential difference, but its position is entirely un-
known on the level of the wavelength. One can therefore represent it by a plane
monochromatic wave. This wave strikes the surface of the crystal and penetrates
the first layers of atoms. The regular atomic arrangement within the crystal then
gives rise to the diffraction phenomenon, due to which the probability for subse-
quently detecting the electron in one or other direction varies with the direction
under consideration and presents intense maxima in certain privileged directions
(theory of Laue and Bragg). This scattering process can be described only by em-
ploying the wave picture, because the latter assumes that all of an extended portion
of the crystal participates in the phenomenon, and it introduces phase differences
(essentially a wave notion!) between wavelets scattered by different atoms, regu-
larly distributed throughout the crystal. If one tried to represent the scattering of
electrons with the aid of a granular picture, it would be necessary to consider an
 Bohr's Calculation for Young's Slits 81

electronic trajectory that gets to meet the crystal in a point, such as the one indi-
cated by a solid line in the figure above. But then the reflection of the electron by
the crystal would depend only on the physical properties of the crystalline surface at
the point of impact and not at all on the regular structure of the crystal; moreover,
it would be impossible to explain with this corpuscular picture the occurrence of
phase differences.
Therefore, in electron diffraction by a crystal, it is the undulatory aspect of
the electron that manifests itself; its granular aspect disappears completely. But
afterwards it is the scattered electron which makes a well-localized impression
in the sensitive layer of the photographic plate, like a bullet puncturing its tar-
get in a specific point. In this second phenomenon, the electron manifests itself
like a localized particle, like a lump, nothing any more betraying its undulatory
facet.
Here then we have the same experiment producing successively two processes
whose explanation requires the invocation of the undulatory picture for the first
and the granular picture for the second. But for each process only one of the two
images obtains, so that a flagrant contradiction never occurs.

8. BOHR'S CALCULATION FOR YOUNG'S SLITS

Here is a slightly different illustration of complementarity given by Bohr. We

know the setup for Young's experiment. One normally allows a coherent monochro-
matic light beam to fall on one side of a screen containing two slits. On the other
side of the screen, the two slits, which are very close together, allow the transmission
of the light and function as two narrow sources of coherent light. The light waves
emitted behind the screen by these two sources will undergo superposition; and this
superposition, by the process of interference, will produce bright and dark interfer-
ence fringes. To determine the phase difference between waves arriving from both
slits at a given point, one must calculate the positions of the fringes, positions which
naturally depend on the wavelength of the wave employed. Observation completely
confirms the predictions of wave theory, and this is why Young's two-slit experiment
was one of those which about 150 years ago provided decisive proof in favor of the
wave theory of light.
Here we thus have an experiment where the undulatory aspect of light clearly
manifests itself. But, should we attempt to introduce into this experiment the
idea of a photon as a localized lump, we would encounter insurmountable difficul-
ties. The trajectory of the photon would have to pass through one slit or the
other, which would destroy the symmetric role of the two openings, a symme-
try which is indispensable for the interpretation of the phenomenon: How oth-
erwise would one explain that the trajectory of a photon passing through one of
the slits is influenced by the presence of the other slit? Yet such an influence
would be necessary to account for a phenomenon that depends on the relative po-
sitions of the two openings. How would one introduce in a purely granular picture
the phase difference arising from the separation between the slits, a difference of
phase without which one cannot arrive at a correct prediction of the observed phe-
nomenon?
82 Measuring Canonically Conjugate Quantities

-- A

-
Incident wave o ----D--+~-----y
B

These difficulties are created by Bohr's idea of complementarity. The interfer-

ences produced by Young's device constitute a phenomenon where the undulatory
aspect of light manifests itself: The granular aspect of light could not make an
appearance here without leading to contradictions. Pursuing the question further,
Bohr has shown that any device enabling us to decide through which of Young's
openings a photon has passed would necessarily make the phenomenon of interfer-
ence disappear: While enabling us to reveal the granular aspect of light in a precise
manner, such a device would necessarily cause its undulatory aspect to vanish. In
broad outline, Bohr's reasoning is as follows.
We suppose, to be precise, that the monochromatic light incident on the front of
Young's two-slit screen emanates from a single slit, functioning as a light source, cut
in another screen. Let us denote by a the distance between the slits in the two-slit
screen and choose the x and y axes as indicated in the figure above.
Designate by D the distance separating the screens, assumed to be parallel to
one another and perpendicular to the y axis, and by A the wavelength of the light
used.
Assume that the position on the x axis of the single slit is known with an
uncertainty of .6.x. Practically a and .6.x are always small compared with D.
The phase difference between the light waves arriving at the two slits in Young's
screen is equal to

.6.<p = 2; [j + G+
D2 .6. x ) 2 _ j D2 + (~ _ .6. x) 2 ], (22)

i.e., approximately to .6.<p = 21r( a.6.x/ AD).

In order to obtain clear fringes behind Young's screen, the phase difference be-
tween the light waves emanating from its two slits must be well determined, that
is, the uncertainty affecting the phase difference must be much less than 21ri this
requirement gives
AD
.6.x« - . (23)
a
 Bohr's Calculation for Young's Slits 83

On the other hand, so that one might be able to tell through which slit in
Young's screen any photon originating in the primary source slit would go, it is
necessary to know with sufficient precision the direction of its linear momentum
on leaving the source. If Px and Py are the components of this momentum, the
point reached by the photon on the two-slit screen will have an abscissa equal to
D{px/py); and, if Px is affected by an uncertainty /:).Px, this abscissa will be subject
to an uncertainty D(/:).px/py). Therefore, in order to be capable of verifying that
the photon has passed through a particular Young slit, one must clearly demand
that
(24)

But the light beam issuing from the primary slit is very nearly parallel to the y axis,
so that one approximately has Py = P = h/ Ai consequently the foregoing condition
can be written approximately

/:).Px
a~Dp =D/:).px
(A)
h . (25)

But we know that, whatever device is used for measuring the coordinate x of the
photon and the component Px of its linear momentum when it traverses the single-
slit screen, one will always be left with the Heisenberg inequality /:).x/:).px ~ h. The
condition (25) therefore a fortiori gives
AD
/:).x~-. (26)
a
Now, obviously, the inequalities (23) and (26) are contradictory. We thus
conclude that, if one can specify through which of Young's slits the photon has
passed, then it is impossible to observe an interference phenomenon; and, con-
versely, if it is possible to obtain an interference pattern, then one will not be able
to tell through which opening the photon has passed. The granular and undulatory
aspects of light here playa kind of hide-and-seek and never enter into direct conflict
with one another: This is the essential idea of Bohr's complementarity.

Note G. L. Louis de Broglie later criticized this line of reasoning and replaced
it by another (see Ref. II, 29, p. 65). He notably observes: "The manner in which
the uncertainty relation occurs is somewhat singular, because it assumes implic-
itly that one is able to measure the component Px of the particle's linear mo-
mentum through the recoil of the first screen along the x axis, which is impos-
sible, since this screen has a macroscopic mass and can be fixed solidly. More-
over, the reasoning does not take into account the slit size of the first screen
(which is not to be confused with the uncertainty /:).x), a size which plays an es-
sential role in the diffraction phenomenon which allows the wave, after its pas-
sage through the slit of the first screen, to reach the two openings of the sec-
ond." And de Broglie recovers the result of Bohr, by demonstrating that, in tar-
geting one of Young's slits, it would be necessary to increase the width of the
first slit, but this would have the effect of making the interference pattern disap-
pear.
84 Measuring Canonically Conjugate Quantities

The preceding reasoning for photons would equally well apply to electrons
and other material particles. In fact, the experiment with Young's slits must,
at least in principle, be reali7.able for all these particles. The same kind of
considerations could moreover be extended to interference devices other than
Young's.
 CHAPTERS

PRECISE FORM OF
THE UNCERTAINTY RELATIONS

1. THEOREM ABOUT THE DISPERSION OF NONCOMMUTING

QUANTITIES

In taking as our fundamental postulate the possibility of representing at every

instant the state of our knowledge about a particle by a wave function 'ljJ, we have
shown, by relying on the properties of Fourier expansions, that the uncertainties in
two conjugate quantities p and q obey the inequalities
f).qf).p ~ h, (1)
which are exact in regard to orders of magnitude. This is the "qualitative" statement
of Heisenberg's relations; and we have seen that no experiment can furnish values
for canonically conjugate quantities that are more exact than what these so-called
relations allow.
We have also shown that the simultaneous measurement of any two quantities
with noncommuting operators is in general impossible, even if these quantities are
not canonically conjugate (e.g., the components of the angular momentum); still,
in this case it can happen by way of exception that simultaneous measurement is
possible (namely, when these quantities have the value 0).
We shall recover these results in a more rigorous manner while demonstrating
a theorem, apparently due to Pauli, on the dispersion of noncommuting quantities.
We are going to give two slightly different demonstrations of this theorem.

First Demonstration. We start by introducing a new definition. Let F be a

linear operator defined for certain variables in a domain D. Also in D, we then
designate as "adjoint operator" the operator Ft which satisfies the equation

tf*Fgdr = t(Ftf)*gdr,
for all functions f and 9 that are finite, uniform, and continuous in D and that
cancel one another at the limits of D in such a manner that the surface integrals
arising in the integration by parts of the integral over D will be zero. If one compares

t t
the definition of Ft with the definition of a Hermitian operator,

f* Ag dr = g(Af)*dr,
86 Precise Form oj the Uncertainty Relations

then it follows that Ft = F if F is Hermitian, so that a Hermitian operator is its

own adjoint; Le., Hermitian is the same as self-adjoint.
Whether a linear operator is Hermitian or not, the wave mechanical mean value
of F Ft is always real and positive (or zero), because

Having noted this, we are now in a position to demonstrate the announced theorem,
which asserts:

Theorem: If two observable physical quantities correspond to two linear Hermi-

tian operators A and B, respectively, then

(3)

[A,B] being the commutator of A and Band O"A and O"B being the dispersions, or
standard deviations, defined by

(4)
To demonstrate this fundamental theorem, we consider the linear (non-
Hermitian) operator A+iAB, with A denoting a real constant; its adjoint is A-iAB,
and, on invoking the result (2), one sees that the mean value

(A + iAB)(A - iAB) = A2 + A2 B2 - iA[A, B]

is real and positive or zero. Consequently the function

is real and nonnegative, leading us to the conclusion that [A, B] is purely imaginary.
Furthermore, J(A) exhibits a minimum at

and there has the value

J(AO) = A2 + ~ ....>.(=[A::,=B~])_2
4 B2
As this value must be positive or zero, one infers that

(5)
 Theorem About the Dispersion of Noncommuting Quantities 87

Put
8A= A-A, 8B= B-B;
A and B being numbers, both 8A and 8B are operators. One then easily finds

[8A,8B] = [A - A,B - B] = [A,B].

We can apply the inequality (5) to the operators 8A and 8B; keeping in mind
the last equation, we get

As [A, B] is purely imaginary, it follows that

which is the stated theorem.

As an application, consider two canonically conjugate quantities for which
[A, B] = -h/27ri. One has

[A,B] = in 1jJ*( -2~i)1jJdT = -2~i'

a purely imaginary quantity, as it must be. The theorem thus gives

(6)

This formula provides an entirely exact form of Heisenberg's uncertainty relations.

In particular, it implies
(7)
We next apply the theorem to two noncommuting observable quantities whose
commutator [A, B) equals an operator C; one then has [A, B] = C and conse-
quently

OAO"B ~
ICI
-2-· (8)
In particular, for A = Mx and B = My,

h
[A,B] = -2. M z ,
7rZ

so that one gets

(9)
88 Precise Form of the Uncertainty Relations
Generally the product of the dispersions will exceed zero; it can nevertheless be zero
in the exceptional case when Mz = o.

Second Demonstration. We are going to give another demonstration of the for-

mula uxup ", ~ hj41'f for canonically conjugate quantities.
Let qk be a coordinate and Pk its conjugate momentum. One has

(19)

Let qk define the origin for qk, so that the mean value of qk will be zero; we then
get

On the other hand,

(11 )

and

Let us define
(12)

The function 'IjJ', like the function 'IjJ, is finite, continuous, and uniform in D and
equal to zero at the boundaries of D. One can write

(13)

on keeping in mind that 'IjJ'* 'IjJ' = 1'IjJ 12 , so that 'IjJ', like 'IjJ, is normalized; the second
formula readily follows when one replaces 'IjJ' by its definition (12).
As 'IjJ is zero at the boundaries of D, integration by parts allows us to write

We now claim that

(14)
 Theorem About the Dispersion of Noncommuting Quantities 89

To establish this inequality, consider two series of terms involving the sets of
complex quantities al, ... ,an and bI, .. . ,bn . One obviously has

lL::aibil
,
: :; L::, laibil :::; L::, la;llbil,
whence
lL::aibiI2:::; (L:: laillb;lf
i i
and, consequently,

L:: lail 2L:: Ibil 2-1L::aibiI2 ~ L:: lail 2 L:: Ibil 2- (L:: laillbil) 2.
i i i i i i

But the right-hand side of this inequality is found to be

L::(laillbjl-lb;llajI)2,
i>j

that is, a sum of squares. The inequality therefore reduces to

L:: lad 2L:: Ib;l2 ~ lL::aibiI2,

i i i

or, more explicitly, to

Now divide the domain D into small elements of size Llr and consider the two
sets of complex quantities

[88'1/;'] J Llr2,etc.,
qk AT2

wherein [qk'l/;'bTl denotes the mean value of qk'lf/ in the element Llrl' etc. Applying
formula (15) to these sets of quantities, one gets
90 Precise Form of the Uncertainty Relations

On increasing the number of elements flTj indefinitely, while letting each one of
them tend to zero, one obtains the limiting result

(16)

The function 'ljJ' being zero at the boundaries of D, integration by parts transforms
the left-hand side of (16) into

and the stated formula (14) becomes thereby verified.

If now the expression

given by Eq. (13), is combined with Eq. (14), one finds

since 'ljJ' is normalized in D. Therefore,

that is, we recover the dispersion theorem for canonically conjugate quantities.

2. OPTIMAL NATURE OF THE GAUSSIAN WAVE PACKET

The second demonstration, just given, of the dispersion theorem allows us to

establish that, in order to obtain the equal sign in the inequality CTqkCTPk ~ h/47r,
one must require i'ljJi to depend on qk solely via the Gaussian exponential

exp [
(qk -
2a 2
qki] •
We thus have a Gaussian wave train or wave packet. Such a wave packet corre-
sponds to a minimum of dispersion and in this sense is "optimal."
 Optimal Nature of the Gaussian Wave Packet 91

To demonstrate this feature, we recall the reasoning that led to Eq. (15),
i.e., to

2:)aiI 2L: Ibd2 ~ lL:aibiI2.

iii
We see that, in order to obtain the equal sign in this formula, it is necessary and
sufficient for the quotient lail/lbil to have the same value for all i.
The argument resulting in the inequality (14) then shows that, for the equality
sign to appear in (14), the quotients

must have the same value in all elements AT;.

From this it follows that, in order to get the equality sign in the formula

it is necessary for
1 81tP'l
qkWI 8qk

to have the same value for all qk.

The quantity ItP'l, considered as a function of qk, must therefore obey the differ-
ential equation

(17)

which has the general solution

-t .
ItP'l = C'exp ( Cq2)
Since ItP'l is required to be zero for qk = ±oo, C must be negative; thus we may
set C = _a- 2 and accordingly write

WI = C' exp (-2~2)' (18)

This result is obtained under the earlier assumption that qk is zero. If that is not
the case, one must write

(19)
92 Precise Form of the Uncertainty Relations

The constant C' can moreover depend on any of the coordinates other than qk.
One also easily finds that O'qk = a/2 1/ 2 • Indeed,

1: 00

+00
u 2 exp ( _u 2 / a2 ) du
=
a2
2'
(20)

100 exp (_u 2 /a 2 ) du

If the wave function is a Gaussian wave packet with respect to qk, then the
position probability in the interval (qk, qk + dqk) is given by

11/;'1 2 ()( exp [-( qk -;; qj;)2] ,

20'qk

i.e., the probability obeys a Gaussian law.

The state 1/; under consideration is one which results from a Gaussian distribution
of possible errors in the measurement of qk. It represents the most favorable case of
the Heisenberg relations, when the product of the dispersions equals h/47r, instead
of being larger than this number, and thus attains its minimum value.

Remark on Gaussian Packets. The property just mentioned demonstrates the

interest attached in wave mechanics to the use of Gaussian wave packets. These
packets possess another remarkable property, which can be stated as follows: A
wave packet that is Gaussian in qk is likewise Gaussian in Pk.
Indeed, suppose we have

11/;1 = Cexp (/ - qZ2) = Cexp (- 40'qk

2a
q\ )
(corresponding to qk = 0), where the factor C may depend on the q other than qk,
and thus
I1/; 12 = C 2 exp (- q\ ).
20'qk

1:
One can expand the wave function 1/; in a Fourier integral of the form

1/; = h- 1/ 2
00 C(Pk) exp ( - 2:i Pkqk) dpk, (21)

wherein 1/; and C(Pk) may also depend on the coordinates other than qk; we ne-
glect to exhibit these q's in our equations, since they will not play any role in our
reasoning.
 Optimal Nature of the Gaussian Wave Packet 93
The theory of Fourier integrals teaches that

C(Pk) = h- 1/ 2 [:00 'IjJ exp C:i Pkqk ) dqk' (22)

On substituting herein

one gets

_~ 1+-00
- h1/ 2
00

exp
(_ [~_
a 21/ 2
271'i(Pk - Pk)a] 2)
h 21/ 2

that is,
C(Pk) = C' exp [_ 4~2 (Pk _ Pk)2 ~2] ,
with C' defined in an obvious manner. It follows that

IC(Pk)1 2 = IC'I2 exp [_ 4~2 (Pk _ Pk)2a 2],

or
(23)

provided we define
h2
O';k = -71' 8
2 2'
a
Since, by Eq. (21), IC(Pk)1 2dpk gives the probability that Pk will lie in the interval
dpk, we see that the probability distribution for Pk is Gaussian with the dispersion
O'Pk' As a = 21 / 2 0'qk' it follows that

h
O'qkO'Pk = 471"

Thus the relation for the dispersion product with the equality sign is recovered,
and one sees that the wave packet is Gaussian in Pk as well as in qk.
94 Precise Form of the Uncertainty Relations

3. COMPARISON OF THE THEOREM ON DISPERSIONS WITH

THE QUALITATIVE UNCERTAINTY RELATIONS
OF HEISENBERG (PAULI AND ROBERTSON)

We are now in possession of two uncertainty statements. First of all, we have

demonstrated, by relying on the properties of Fourier expansions, that, in order of
magnitude,
opoq ~ h. (24)
Equation (24) is somewhat qualitative, because it is true only when orders of
magnitude are compared. One sees why this is so on recalling our discussions about
expansions and, perhaps more clearly, on recalling the argument that uses Heisen-
berg's microscope, where demonstration of the uncertainty relation involved the
definition of resolving power (two neighboring points of an object ceases to be sepa-
rable by an optical instrument when the center of the main diffraction maximum due
to one point coincides with the first diffraction minimum due to the other), and rec-
ognizing that this definition is somewhat arbitrary and only valid in an approximate
sense.
In the second place, we have obtained a rigorous statement

(7)

which originates from an application to the canonically conjugate quantities q and

p of the general relation
(3)
We have seen that Heisenberg's relations in their qualitative form signify that
at every moment, and in particular when one carries out a measurement, the values
of two canonically conjugate quantities are subject to uncertainties whose product
always is, in order of magnitude, larger than or equal to h. The same conclusion
is reached in a more precise fashion if one starts from the condition imposed on
the dispersions. At every instant, and in particular immediately after a measuring
operation, our knowledge about the state of a system is represented by a wave
function 7/J; and two conjugate quantities p and q have random values corresponding
to some probability distributions such that the product of their dispersions always
will be larger than or equal to h/47r.
The dispersion theorem therefore leads, just like the qualitative relations of
Heisenberg, to the conclusion that it is impossible in the same operation to measure
with complete precision two conjugate quantities p and q; since otherwise p and q
would after the measurement be accurately known, and one would have both O"q = 0
and O"p = 0, which conflicts with the relation O"qO"p ~ h/47r. Thus, after the mea-
surement, the state of our knowledge would no longer be representable by a wave
function 7/J, because such a presentation gives rise, as we have seen, to the relation
O"qO"p ~ h/47r.
If I stress this point, it is because one could have tried to reason as follows:
Consider a larger number of systems in the same state, that is, represented by the
 Considerations About Uncertainties. Sharp-Edged Uncertainties 95

same function 'IjJ, and let us try to measure simultaneously two conjugate quanti-
ties p and q. We know that one can for every system obtain different values with
different probabilities, which a knowledge of 'IjJ allows us to calculate. One could
then believe that the dispersion law always requires the dispersions of p and q to
be such that their product exceeds or equals h/41f but that the law would nev-
ertheless not prevent certain measurements from simultaneously furnishing precise
values for p and q. The error in such a reasoning comes from the fact that one
has considered only the state of the probabilities existing before the measurement
(represented by the wave function 'IjJ obtaining at that time). But it is moreover
necessary that also the state following the measurement be representable by a wave
function 'IjJ and that the dispersion relation then belongs to the corresponding prob-
ability distribution. It is this circumstance which allows us to deduce from the
dispersion relation that the simultaneous measurement of p and q is an impossibil-
ity.

4. VARIOUS CONSIDERATIONS ABOUT UNCERTAINTIES.

SHARP-EDGED UNCERTAINTIES

To further clarify the nature of the uncertainties Sp and Sq occurring in the

qualitative relations of Heisenberg, we now subject them to a closer look.
Let A be an observable quantity. For a system in a given state 'IjJ, the var-
ious values of A have well-defined probabilities, which can ·be calculated from 'IjJ
on invoking the general principles of wave mechanics. We now impart a precise
meaning to uncertainty in the Heisenberg sense by giving the name "uncertainty
of the quantity A in the state 'IjJ" to the small interval 8A that contains all the
A values for which the total probability is larger than 1 - e, where e is a very
small amount (e.g., e = 1/1000). Any measurement of A will then almost certainly
produce a value lying in the interval 8A. This definition of uncertainty depends
on the value chosen for ej but, once e has been chosen, the uncertainties are well
defined.
Having adopted the foregoing definition, one can use the procedure of Fourier
decomposition to demonstrate the following: If SA and 8B are the uncertainties of
two canonically conjugate quantities A and B in the state 'IjJ, then

SASB ~ a(e)h, (25)

a( e) being a number at least of the order unity whose exact value varies with ej
the smaller one chooses e, the larger a will be. With the small values of e assumed
(often implicitly) in practice, a(e) will have a value in the neighborhood of unity.
We thus recover the Heisenberg relations with some additional precision.
We can imagine that one of the intervals 8A and 8B defined as above is "sharp-
edged," that is, the probability for finding values of A (or B) outside 8A (or 8B) is
zeroj then e = 0 for the interval in question.
In this case it can be shown that a(e) = a(O) = 00 and 8A8B = 00. One can
justify this result by recalling the Fourier expansion procedurej we shall soon give
an example of this. Thus, if the wave 'IjJ is nonzero only inside the interval ~x of
96 Precise Form of the Uncertainty Relations

the variable x (a sharp-edged interval), the Fourier expansion of'l/J will involve all
values of Px, i.e., i:lpx = 00. Therefore i:lxi:lpx = i:lx x 00 = 00.
But in fact this result, perfectly rigorous from a mathematical standpoint, is
only of very limited practical importance, because generally the probability becomes
practically zero as soon as it falls below a sufficiently small value c. This is why
the Heisenberg relation DADB ~ exh, with ex in the neighborhood of unity, is always
verified in practice, even if one of the intervals DA or DB is sharp-edged.

I(v)

The question has an analogy in the theory of spectral line widths. There one
shows that, when the intensity of any spectral line is plotted against the frequency,
the ensuing curve has a profile as shown in the figure. Theoretically the line has
an infinite width, but this mathematically rigorous result is not real in a physi-
cal sense, because the intensity I(v) becomes unobservable, and thus practically
zero, as soon as it falls below a certain value. Practically, therefore, the spectral
lines do not extend to all values of the frequency but instead have well-defined
sIzes.

Examples of Sharp-Edged Uncertainties

To illustrate the preceding remarks, a simple example of sharp-edged uncertain-

ties will be studied. We consider a wave 'l/J such that
'l/J = 0 for x < a and x > b;
(26)
'l/J = (b - a)-1/2 exp (-ikox) for a < x < b,
where the factor (b - a)-1/2 assures that the wave 'l/J is normalized:

[:00 1'l/J1 2 dx = 1.
Between x = a and x = b, the wave has the form of a monochromatic wave;
outside this sharp-edged interval 'l/J = O. If one measures the x coordinate, it will
certainly be found to lie between a and b (Dx = b - a, with c = 0).
Let us put ko = (211'/ h )po and k = (211'/ h)p and investigate the Fourier expansion
of'l/J. One has

'l/J = (211')-1/2 1+00

-00 c(k)exp(-ikx)dk, (27a)

 Considerations About Uncertainties. Sharp-Edged Uncertainties 97

( k) = (2 )-1/21 b exp [i( k - ko)x ) d (27b)

c 7r a (b_a)1/2 x,

which gives

c( k) = [27r(b - a )rl/2[i( k - kO))-l [exp [i( k - ko)b) - exp [i( k - ko)a)]

and, consequently,
2 1 1 1
Ic(k)1 = 27r b _ a (k _ k o)2 2[1 - cos(k - ko)(b - a)]

_ ~ 4 sin2 [(k - ko)(b - a)]

- 27r (b - a)(k - ko)2 2'
that is,
Ic(k)f = b2- a sin: u, with u = (k - ko)(b - a) . (28)
7r U 2
We therefore see that Ic( k) 12 is rigorously zero only at infinity on the k axis.
Thus, to a sharp-edged interval in x there corresponds an infinite interval in k
(or p). If therefore one takes c = 0, one gets .6.x = b - a and .6.Px = 00, whence
.6.x.6.px = 00.

Note L. B. One may easily verify that

The function Ic( k) 12 diminishes rapidly as k - ko increases, but it is still notice-

able for u > 7r, i.e., for .6.k > 27r j(b - a). One therefore certainly has .6.x.6.k > 27r,
i.e., .6.x.6.p > h. However, if k deviates further from ko, the value of Ic(k)12 falls
off rapidly. Thus one may calculate that Ic( k) 12 ~ 1O-2Ic( ko) 12 for .6.p.6.x = 3h and
Ic(k)12 rv 1O-3Ic(ko)12 for .6.p.6.x = 9h. Therefore, although Ic(k)12 vanishes only at
k = 00, in practice a measurement of p leads always to a value of p lying in an inter-
val.6.p (around the value P = PO = (h/27r)k o) such that .6.p(b - a) = .6.p.6.x = ah,
with a rv 1; and this is all that matters for the practical application of Heisenberg's
relations.

Heisenberg's Microscope. Let us recall the example of Heisenberg's microscope.

Here .6.Px is sharp-edged because it is limited by the aperture of the microscope.
By contrast, .6.x is not sharp-edged but extends theoretically to infinity, because
it is determined by the phenomenon of diffraction introduced via the definition of
resolving power. In principle, the wave reaching a point pi in the image plane does
not originate uniquely in a point P of the object plane. Instead it can issue from
98 Precise Form of the Uncertainty Relations

every point in the object plane, and thus ~x is in principle limitless; but, in practice,
as is shown by the theory of resolving power, the uncertainty ~x in the position
of P is limited to the immediate neighborhood of the point in the object plane for
which pI is the geometrical image.

Screen With an Opening. Consider next the example of an opening cut in a

screen. Here the uncertainty ~x is determined by the contour of the opening and
therefore is sharp-edged. Behind the screen, light is diffracted in all directions,
so that ~Px is in principle infinite, but the observable bright fringes all occur in
the vicinity of the central bright fringe, and consequently ~Px is in practice very
limited.

Gaussian Wave Packet. For the Gaussian wave packet we have

(29)
,2
Ic(p)12 = G exp (_ p22) ,
2rJp

with rJqrJp = hi 47r. The Gaussian wave packet is therefore not sharp-edged, either
in q or in p.
Let us put 5q = mrJq and 5p = mrJp and take

In accordance with our definition of uncertainty, 5q and 5p will then be the uncer-
tainties corresponding to this value of E, E being a function of m, and conversely.
For given E, m is fixed, and one has

5q5p = m 2 rJqrJp = m 2 - It (30)

47r
If E -t 0, then m - t 00 and 5q5p - t 00. But in practice, it suffices to suppose
that E is very small. This will be realized already for m = (47r )1/2, since

as e- 6 '" 11350. In practice one will therefore have 5q5p ,.." h.

Summary. In order to precisely define the Heisenberg uncertainties, it is nec-

essary to define the uncertainty 5A of any quantity A as the interval of A values
 Considerations About Uncertainties. Sharp-Edged Uncertainties 99
such that the probability of finding a value outside SA is less than some small
quantity c. For two canonically conjugate quantities A and B one then finds
that SASB '" a(c)h, where a(c) depends on the choice of c. The function a(c)
is infinite for c = 0, so that then SASB = 00, from which it follows that SB
is infinite if SA is finite (case of a sharp-edged interval). However, in practice it
suffices to choose c very small but nonzero; and then the product SASB can in
the most favorable cases become as small as something of the order h but not
smaller. Practically we thus have SASB 2:: h in order of magnitude. The question
is analogous to one that is encountered in the study of diffraction and resolving
power.
The dispersion theorem stated for conjugate quantities in the form crA cr B 2:: h /47r
is more precise than the Heisenberg uncertainty relations. Like these relations, the
theorem leads to the finding that it is impossible in the same measurement operation
to obtain precise values for conjugate quantities.

Note L. B. Consider the following case, where the uncertainty relation applies
even though the dispersion product is infinite.
Let a wave train be defined by

'I/J(x) = 0, for x < 0,

'I/J(x) -,X) exp ( 27rikox),

= exp (-2- for x > 0 b > 0).

1
On setting
+00

'I/J(x) = -00 c(k)exp(27rikx)dk,

we then get

One thus finds

1'I/J(x)12 = exp (-,x),

If N is a very large number, we can define the uncertainties Sx and Sk by putting

whence
1
Sx = -logN,
I
One thus gets
1
SxSk = -Vii
47r
10gN,
100 Precise Form of the Uncertainty Relations
or, setting k = Pxlh,
h
8x8px = -v'N
411"
log N.

For N = 20, this result gives

By contrast, O"p", = 00, because

diverges; and for o"x one finds the value II,. Therefore, O"p.,O"x = 00, thus infinitely
exceeding h1411". Here the dispersion relation is useless, whereas the uncertainty
relation is still of value.

Case of Angular Momentum Components. We have just studied the case of

canonically conjugate quantities, which is a particular instance of non commuting
quantities with commutator equal to some constant. But, as we know, there ex-
ist actual noncommuting quantities of a different kind for which the commutator
equals some or other operator. This is the case for the components of the angular
momentum, which satisfy the relations

The dispersion theorem gives

O"M.,O"My 2: 2 [Mx, My] I=

'11 hM-z, etc.
411"

The question now is: Do uncertainty relations exist for M x , My, Mz ?

Naturally, one can, with the aid of a very small number c, define as above the
uncertainties 8Mx , etc. A priori, we know nothing about the value of a product such
as 8Mx8My. But, since the likelihood of a typical deviation is still large enough, it
will most often happen that 8Mx exceeds O"M" and 8My exceeds O"My ' so that

We now want to show in another manner that one cannot have M z =f 0 along
with 8Mx8My = o.
Indeed, in order to have Mz =f 0, an expansion of'r/J in terms of the eigenfunctions
of Mz must contain at least one eigenfunction with eigenvalue different from zero.
This eigenfunction of Mz cannot be an eigenfunction of Mx or of My, since the
only common eigenfunction of M z and Mx (or My) is f = o. Therefore, if one
expands 'r/J in terms of eigenfunctions of Mx (or My), this expansion will contain
at least two eigenfunctions of Mx (or My) belonging to different eigenvalues. As a
 Considerations About Uncertainties. Sharp-Edged Uncertainties 101

result, both SMx and SMy are different from zero. It is therefore impossible to have
simultaneously Mz =f 0 and SMxSMy = O.
Here are some further remarks demonstrating the difference between the first
and second kinds of noncommuting quantities.
Let a system which initially occupies the state 1 be represented by a wave func-
tion .,p1, and denote by A and B two observable quantities of this system.
Suppose that a certain measuring operation causes the system to make a tran-
sition to a state 2 represented by some wave function .,p2.
Before the measurement, one has

while afterwards
a(2)a(2) > ~ 1 [A ' B)21 .
A B-2
If A and B are canonically conjugate or, quite generally, if [A, B) is a multiple
of unity, [A, B) becomes a constant independent of the state, and thus the products
a(l) a(l) and a(2) a(2) will have the same lower limit
A B A B .
If, by contrast, [A, B) equals an operator C, then [A, B) will in general vary with
the state under consideration; and, if the state 2 is such that

1 [A,Blz 1 < 1 [A,Bh I,

then it can happen, even though a~)a~) > (1/2) 1[A, Blz I, that

(2) (2)
a A aB < 2"1 1 [A, Bh 1.

In other words, the lower limit of the product a~) a~) after the measurement
is determined by the state that exists after the measurement, not by the initial
state. Let us apply the foregoing to the case where A = Mx and B = My, with
[A, B) = (h/27ri)Mz • If M z =f 0 in the state 1, if follows that a~ a~ > O. But
a measurement can lead to a state 2 having Mz = 0 and a~ a~ = 0, that is, to
a state for which Mx and My have precise values different from Mx = My = O.
Herein lies a big difference with the case of two canonically conjugate quantities,
where no measurement can bring about a state in which both quantities have exact
values.
In the initial state 1, definite probability distributions exist for A and B, which
are derivable from a knowledge of .,pl. Let SA and SB be two arbitrarily chosen
intervals of A and B values. In general, these intervals will for the state 1 not be
the precise uncertainties defined above. But let us imagine that a measurement has
been made which enables us to assert that the probabilities for finding an A value
outside SA and a B value outside SB are in either case less than c. Then, in the
102 Precise Form of the Uncertainty Relations

state 'ifJ2 following the measurement, 8A and 8B will have become uncertainties in
the precise sense, and one will have

8A8B ~ ~ I [A,Bh Ia,

a being a number that depends on e and is at least of the order unity.
If A and B are canonically conjugate, one recovers the inequalities of Heisenberg,
and it is thus proven that no measurement can furnish values of A and B with a
precision exceeding that allowed by these inequalities; because, if it were otherwise,
the state 2 following the measurement would not have a representation in wave
mechanics.
If A and B are such that [A, B] = C, one sees that the lower limit of 8A8B can
vary with the state and that after the measurement this limit is determined by the
va1ue of [A, Bh.

Note G. L. Louis de Broglie has never subsequently published a study of the

uncertainty relations as exhaustive as the present one, and I doubt that one will
in the literature find any other exposition that recovers practically the complete
analysis of Heisenberg and Bohr, which here are augmented by the analysis of
de Broglie himself. Let us note the aspect peculiar to him: the care he took to
clearly separate information about the state of a system before any measurement
from that concerning its state after the measurement. This distinction prepared
the way for de Broglie's later interpretation of the uncertainty relations (see Ref. II,
27, 29, and 33), the main idea of which is the following: It being assumed that a
particle is always localized at a particular point of its wave, any localization mea-
surement would only reveal a pre-existing position of the particle, which means
that the uncertainty ~x exists in the initial state of the system, even before any
measurement (of localization or whatever) takes place; it is a presently existing
uncertainty. By contrast, the measurement of Px, necessitating a preparation of
the system, does not uncover a pre-existing value of the momentum but rather a
value created at the time of this preparation, so that ~Px is not a present un-
certainty; it is rather an "uncertainty anticipated in the initial state, on the ba-
sis of the value that px can have following the action of an apparatus capable of
measuring Px, at a time when one does not yet know the result of this measure-
ment."
 CHAPTER 9

HEISENBERG'S FOURTH
UNCERTAINTY RELATION

1. THE ABSENCE OF SYMMETRY BETWEEN SPACE AND TIME

IN WAVE MECHANICS

If one adopts a relativistic point of view, Heisenberg's fourth uncertainty relation

bWbt rv h appears as a natural complement of the first three relations bPi bXi rv h,
because relativity theory views energy as a quantity canonically conjugate to time,
in the same sense that the components Px,Py,Pz of the momentum are canonically
conjugate to the variables x, y, z, respectively. This approach is evident, for example,
in the observation that the element

W dt - Px dx - Py dy - pz dz

of the Hamiltonian action integral is a spacetime invariant.

But in wave mechanics, at least in the present form of this theory, the symmetry
between the fourth uncertainty relation and the first three relations is more appar-
ent than real. In fact, wave mechanics, even in Dirac's relativistic form, does not
establish a true symmetry between the variables of space and time: While the co-
ordinates x, y, z of a particle are observable quantities that correspond to operators
and have values obeying a definite probability distribution law for every state (de-
fined by a wave function 1jJ), the time t is always regarded as a parameter exhibiting
a well-defined value.

Note G. L. Let us point out that one finds a long exposition in L'electron
magnetique (Ref. II, 11) on the fourth uncertainty relation and on the problem
of time in relativity and wave mechanics. In Certitudes et incertitudes de la science
(Ref. III, 8), one can read an analysis of the possible connection (contested by the
author) between the fourth relation and the fifth, which links the phase of a light
wave to the occupation number: bN b<P 2: 1.

The situation may be clarified in the following manner. Let us suppose that
all measurements are carried out by a Galilean observer. He employs coordinates
x, y, z, t, which serves to locate events in the macroscopic frame of his experiences.
The variables x, y, z, t are numbers, or parameters, and it is these numbers that
feature in the wave equation and in the wave function. But to every atomic particle
104 Heisenberg's Fourth Uncertainty Relation
there correspond some "observable quantities" which constitute the particle coor-
dinates. The relationship between these observable quantities x, y, z and the space
frame of the x, y, z used by the Galilean observer under consideration is statistical,
since in this frame everyone of the quantities x, y, z can in general assume a set of
values obeying some probability distribution. By contrast, no "observable quantity"
t is in present-day wave mechanics associated with any particle: There exists only
a variable t which is defined with the aid of (essentially macroscopic) clocks by the
Galilean observer within his spacetime frame.
In wave mechanics it is necessary to have an "evolution variable" which enables
us to keep track of the variation of the state of any quantum system. But this
evolution of the state of a system or, more exactly, of our knowledge concerning this
state, necessarily occurs in relation to time as it exists in the consciousness of an
observer, i.e., time whose flow can be monitored only with the aid of macroscopic
clocks. It is in the frame of this consciousness-related time that especially the
abrupt modifications in the form of 'IjJ, caused by measurement operations and by
the information they convey to us, will take place. But the circumstance that we are
obliged to adopt the macroscopic time (the variable t of relativistic spacetime) as
evolution variable prevents us from attributing to particles or to quantum systems
an observable quantity t of a random nature, in the same way that one can place an
observable quantity obeying a probability distribution in correspondence with the
cqordinates q.
Such are some of the profound reasons opposing, in my opinion, the establish-
ment in wave mechanics of a symmetry between space and time analogous to the
symmetry postulated in the theory of relativity. These difficulties are intimately
related to the fact that quantum physics creates a link of a novel kind between the
object and the subject. In this new theory, "the state" of a quantum system does
not have an objective definition corresponding to a description of "what is." On the
contrary, it is defined uniquely as a function of "what we know"; it is a description
of our knowledge, and we cannot go beyond this representation. It is therefore in the
consciousness of the observer and, consequently, in the frame of macroscopic time,
that "the state" defined by the wave 'IjJ will evolve. And, if quantum theories do not
succeed in establishing a symmetry between space and time, this failure appears to
originate in the particular character of time as perceived by our consciousness, in
its continuous unfolding, and in its irreversibility.

Note L. B. Should one wish to treat the energy as corresponding to the operator
(h/27ri)8/8t, the eigenvalue equation

h 8cp
27ri Ft = Ecp,

would result; thus quantization of the energy would not occur, E being free to
assume all possible values between -00 and +00.
 Illustration of the Preceding Definition 105

2. CORRECT FORMULATION OF THE FOURTH

UNCERTAINTY RELATION

We are now in a position to correctly state the fourth uncertainty relation and,
in doing so, shall recover the form of this statement that we have already implicitly
been led to assume.
Let us recall the case of a wave train 'I/J of limited size that occupies a region
R of space. The value of the particle's x coordinate is uncertain; it can reveal
itself after a measurement as defining any position inside R, the corresponding
probability varying as 1'l/J12. Moreover, the probability of finding a given value of
Px is equal to Ic(Px)12, so that the value of Px, too, is uncertain. We know that
Heisenberg's uncertainty product {)Px {)x is of the order h. But, while a measurement
can determine the x coordinate of a particle, one cannot speak of the measurement
of its time, because in wave mechanics the time t, being the macroscopic time of
the observer, always exhibits a definite value.
What then does the relation {)E{)t '" h signify? It signifies that, in order to
attribute to a particle an energy E with an uncertainty {)E, one is required to make
an observation, a measurement operation, occupying at least the time {)t '" h/{)E.
Indeed, it follows from our analysis of the Fourier expansions of wave trains that the
time {)t it takes a wave train to pass through a point is at least ofthe order {jt = l/{)v.
To be able to assert that the uncertainty in the energy is at most {)E = h {)v, one
must observe at a point P the passage of both the front and rear ends of the wave
train, which requires an observation lasting at least for the time {jt '" l/{)v. In
particular, to affirm that {)E is zero, i.e., that the wave train is monochromatic,
one would have to make an observation of infinite duration, since the length of a
monochromatic wave is infinite.
Thus, while the first three uncertainty relations express the existence of a prob-
ability distribution for the quantities q and p, that is, the fact that these quantities
are "random variables," in the probability-calculus sense, the fourth uncertainty
relation must be interpreted differently: The time t is not a random variable, but
the measurement of E can only be carried out with the aid of observations of finite
duration; and the briefer the observation, the larger the uncertainty about the ex-
act value of E. Since t is not a random variable, there exists no relation between
the dispersions of E and t, so that here the qualitative uncertainty relation {)E{)t?h
does not coexist with a precise statement about dispersion of the kind (J'E(Jt 2: h/47r,
because, t being a variable of precise value, (Jt is rather meaningless (Jt would al-
ways be zero). It is clearly seen that these conclusions exist in opposition to the
relativistic symmetry of space and time.

3 .. ILLUSTRATION OF THE PRECEDING DEFINITION

To illustrate the meaning of the fourth uncertainty relation, we give an example

first put forth, in a slightly different form, by Lennuyer [Ann. Phys. (Paris) 20,
91-110 (1946)] in an article on optical resonance.
We consider an optical grating with a very large number N of lines, any adjacent
106 Heisenberg's Fourth Uncertainty Relation

two of which are a distance a apart, and allow a beam of light of wavelength'\ (= c/ v)
to impinge perpendicularly upon it.
The light scattered by two adjacent lines of the grating in a direction making
an angle 0 with the normal to the grating has a phase difference equal to u =
(27r /,\) a sin 0, so that the light scattered in the direction 0 at infinity will have an
amplitude proportional to

N-l ('N )
" (.) exp z u -1
~ exp znu = (. ) .
on exp zu -1

The corresponding intensity (squared modulus of the amplitude) is therefore pro-

portional to
exp (iNu) - 1 12 = 1 - cos Nu = sin 2 (Nu/2) .
1 (1)
exp(iu)-l l-cosu sin 2 (u/2)
This classical result shows that maxima will occur in directions for which u/2 = m7r,
where m is any integer, so that sinOm = m(,\/a) = (m/a)(c/v). One can thus
determine the frequency v of the photon (and therefore its energy E = hv) by
observing the angle Om corresponding to the mth maximum. But, on proceeding in
this manner, there remains an uncertainty in the value of v (or of E), because one
can never exactly determine Om. Let us examine this point.
The expression 1= sin 2(Nu/2)/ sin 2 (u/2) exhibits a maximum value N2 when
u/2 = m7r and becomes equal to sin 2 N(m7r + rJ)/ sin 2 rJ when u/2 = m7r + rJ. For
rJ = 7r / N, N being assumed large, one gets

1= sin2(Nm7r + 7r) = O.
sin 2 ( 7r / N)

I therefore decreases from its maximum value N 2 to the value 0 when u/2 changes
by 7r / N. The error which one may commit in measuring Om is therefore always such
that 5(u/2) will amount to a fraction of 7r/N. Since 5(u/2) = (7r/c)(asinO)5v, we
thus see that the uncertainty 5v in the value of v measured in this way can equal
some fraction of c/NasinO; i.e., 5v rv c/NasinO.
Let now 5t be the limiting duration of the experiment. In order that all the
lines of the grating will be involved in the scattering and that consequently the
preceding theory will be applicable, the light scattered in the direction 0 by the
Nth line of the grating must succeed in overtaking the plane of wave P. This
requires that
c NasinO
at ~ ,
C

whence
c c c NasinO 1
avat> = in order of magnitude
- NasinO c
and thus
5E 5t ~ h in order of magnitude.
 Illustration of the Preceding Definition 107
One here easily sees how the duration of the experiment enters into the measure-
ment of the photon's energy E = hv; and we notice that the meaning of the quantity
8t in the fourth uncertainty relation is essentially the duration of the experiment for
the observer who performs it.

Darwin's Example. Darwin [Proc. Roy. Soc. A 130, 632 (1931)] considers an
electron which is at rest but free to move along a straight line. At some point of
this line is located an electrometer which can determine the coordinate x of the
electron by measuring its electric field. In order to easily analyze the functioning
of this electrometer, Darwin supposes that it consists of an atom exhibiting the
Stark effect. This atom can emit a spectral line on making a transition from a
state with energy El + Ml£ to a state with energy E2 + M2£; here El and E2
define the energy levels in the absence of any electric field £, while Ml and M2
denote the electric moments of the atom respectively before and after the transi-
tion.
In order to measure the field £ with a precision 5£, one must be able to distin-
guish two frequencies differing by IMI - M21( 5£ / h), which requires a waiting period
5t such that 5v 5t '" 1, i.e.,
(2)

In this way the fourth uncertainty relation enters into the discussion.
Continuing our argument, at some unknown instant, the atomic electrometer
will jump from a state 1 to a state 2. This transition exerts an influence on the
electron, which is at first subject to the action of the electric moment Ml and then,
starting from the unknown instant, to the action of the electric moment M 2 • We
cannot find a better way to compensate for this change in action than by applying
a constant field to the electron equivalent to the field that would accompany an
electric moment (Ml + M 2)/2 placed at the position of the atomic electrometer.
However, even with this compensation, there still remains a residual moment of
size (Ml - M2)/2 initially and size (M2 - Ml)/2 finally. The true electric moment
of the atomic electrometer at any moment is therefore subject to an uncertainty
of the order 1M2 - MIl. Darwin adds: "The fact that there is no observation
without uncertainty is illustrated by the spectral lines for which Ml = M 2. For
them the action on the electron is exactly compensated, but precisely then the
line does not exhibit a Stark effect and cannot serve for an electromctric measure-
ment."
In order to determine the position x of an electron with the uncertainty 5x, one
must measure its electric field with an uncertainty 5(e/r 2 ) = e5x/r 3 • This, as we
have seen, requires the observation to occupy a time 8t such that

(3)

During this interval of time, the electron is subject to an action equivalent to the
action exerted by an electric moment M residing in the electrometer and having a
magnitude at least of the order IMI - M21; the electron is therefore acted on by a
108 Heisenberg's Fourth Uncertainty Relation

force e(M/r 3 ) ~ (e/r 3 )IMI - M 2 1. During the time U this force causes the linear
momentum Px of the particle to change by

(4)

which gives
DX Dpx ~ h in order of magnitude.
We thus recover the first uncertainty relation through the intervention of the fourth.

4. VARIOUS REMARKS ABOUT THE FOURTH

UNCERTAINTY RELATION

We are now going to make several remarks concerning the fourth uncertainty
relation, beginning with a very old one due to Bohr.
It is known that one can excite or even ionize an atomic system by bombard-
ing it with fast particles. But this phenomenon appears incomprehensible when
analyzed with classical ideas. Indeed, if the incident particle traverses an atom
of mean diameter a with a speed v, the time it requires to cross the atom is at
most of the order a/v. The incident particle therefore only has a time of this or-
der to act on the constituents of the atom and exchange enough energy with them
to bring about excitation or ionization. For an atomic constituent to absorb en-
ergy, it must undergo an appreciable displacement during the time T = a/v, which
requires T to be large in relation to the oscillation periods of the electrons inside
the atom. One sees this easily when studying a linear oscillator. Such an oscil-
lator has a well-determined natural period T, with which it will vibrate if set in
motion by an external force; and, for this force to succeed in imparting energy to
the oscillator, it must act on the latter during a time that is appreciably larger
than T. We therefore have a/v = T > T, i.e., a > vT and thus v < avo But,
for excitation or ionization to be possible, the kinetic energy of the incident par-
ticle must be of the order hv, which means that mv 2 '" hv if the particle's speed
v is small enough (as is usually the case) for relativistic corrections to be negligi-
ble.
Consider first an outer atomic electron, whose frequency will in order of magni-
tude equal that of light; for this case,

a '" 10-8 cm,

The condition v < av then gives v < 10 6 cm sec-I, while the condition mv 2 .-v hv
leads to v .-v (hll/m)I/2 = 2.7 x 10 7 cm sec-I. We are consequently left with a
contradiction.
. Consider in t~e sam~ wayan inner atomic electron, whose fre~uency agrees
III order of magmtude WIth that of x rays. We now have a .-v 10- cm and v '"
10 18 sec-I. The condition v < II then gives v < 10 9 cm sec-I, while mv 2 .-v hll
requires that v '" 2.7 X 10 9 cm sec-I. A contradiction thus again arises. One
 Various Remarks About the Fourth Uncertainty Relation 109

therefore sees that the phenomena of excitation and ionization by collision cannot
be explained with classical ideas.
It is not the same when the new concepts are used. In order to apply conservation
of energy to a collision, the kinetic energy of the incident particle must be known
with an uncertainty liE much smaller than its value E rv hv, whence liE ~ hv. But
then the wave train associated with the incident particle will be relatively long and
take the time lit 2: hi liE to traverse an atom. Since the instant when the incident
particle enters the atom can be fixed within the interval lit, one cannot attribute to
the transit time T a value less than lit. Consequently,
h h 1
T '" -
liE
> -E = - = T.
v
We may assume that the condition T > T is fulfilled, because one cannot suppose
that the duration of the interaction between the incident particle and the atomic
constituents is less than the time lit it takes the wave train associated with the
particle to pass through the atom.
Let us make another remark. Suppose a quantum system is given whose wave
function is
1/J = L Ckak exp C~i Ek t ).
k
As we have pointed out, any measurement of the energy at time t must produce
one of the eigenvalues Ej, and the probability is ICk 12 that this value would be Ek
in particular. But we now see that this statement is not quite correct, because any
energy measurement will always take a certain time (and the more time the more
precise the measurement). We can therefore not speak of a measurement made at an
instant t, but only of a measurement occupying some time interval hi surrounding
the instant t. (The situation is different for measurements of linear momentum or
position.)
Nevertheless, the restriction hereby introduced has little practical importance,
because we are going to establish that E will have the value Ek as soon as liE
becomes much smaller than the smallest of the differences IEk-l - Ek 1 and IEk -
Ek+ll. But these differences, even where states with loosely bound electrons are
concerned, correspond to transitions for which the emitted frequency at least equals
that of the far infrared region of the spectrum, i.e., is at least of the order 10 12 sec-I.
Consequently, the observation time needed to distinguish a given quantized state
from a neighboring one will at most be of the order

h h -12
lit '" liE '" h10 12 = 10 sec.

In practice the observation time will therefore always be very small, allowing us to
view the measurement of quantized energies as being practically instantaneous.
Here are some further remarks whose importance will soon emerge:
Let a system be given which has, among others, two quantized states i and k
of energies Ej and Ek. Suppose this system is perturbed by an external action and
a calculation predicts (how, we shall soon explain) that, under the action of this
110 Heisenberg's Fourth Uncertainty Relation

perturbation, the system will oscillate between the states i and k with the frequency
Vik = (Ei -Ek)jh. From this result, one cannot conclude that the system physically
makes a transition from the state i to the state k, and conversely; because such
an inference would be valid only if one could catch the system in one or other of
these states, that is, measure its energy in state i or state k. But, since the system
stays in anyone of these states only for a time tit less than Ijvik = h(Ei - E k ),
no procedure will allow the energy of either state to be measured with a precision
exceeding
h
tiE rv - '" Ei - Ek,
tit
so that one is unable to distinguish one state from the other. While the interaction
lasts, the energy of the system thus remains undetermined between Ei and Ek, and
it is possible to verify the conservation of energy only up to lEi - Ek I.
Again, suppose that a system possesses the energy E1 up to the time iI, that
from t1 to t2 it is subject to an external action, which disturbs the system without
finally supplying it with energy, and that the system occupies a state of energy
E2 for t > t2. Since we have the means to measure El and E2 at respectively all
times preceding t1 and all times following t2, these energies can be measured exactly,
and the conservation of energy requires E2 = E1. But, during the time t2 - t1 of
the perturbation, the system can pass through an intermediary state of energy E;
and, if t2 - t1 is small compared with hj(E - E1), it will be impossible to catch the
system in this state by measuring the energy E. One may say that the system passes
through the "virtual" state of energy E, and in reality the energy is undetermined
during the interaction up to R - E1. Conservation of energy is thus verified for the
overall transition E1 ---+ E2 but not necessarily for the virtual transitions El ---+ E
and E ---+ E2.
We are now going to amplify these considerations by using the method of vari-
ation of constants to quickly develop the theory of perturbations.

5. METHOD OF VARIATION OF CONSTANTS.

TRANSITION PROBABILITY

To illustrate the foregoing considerations, we briefly summarize the large features

of the variation-of-constants method and the notion of transition probability.
In this method one considers a time-independent unperturbed Hamiltonian H(O)
to which there corresponds the stationary states of the system that would obtain
in the absence of any perturbation. It is assumed that the eigenvalues ElO) and
the eigenfunctions 1fJiO) of this Hamiltonian are known. We further suppose that
the system is subject to perturbative actions that can depend on the time; they are
represented by a term V in the Hamiltonian, so that one can write H = H(O) +Vet).
The equation describing the evolution of the system in the presence of the per-
turbation accordingly reads

[H(O) + V(t)]1fJ = ~ 81fJ . (5)

27rl 8t
 Method of Variation of Constants. Transition Probability 111
At every instant t, the perturbed wave function tP can be expanded in terms of
the complete set of eigenfunctions tPkO) belonging to the unperturbed Hamiltonian
H(O):

(6)

with
(7)

According to the general principles of wave mechanics, the probability that, at

the instant t, the system will find itself in the state tPkO) is given by ICk( t) 12. Again it
is necessary, as we have remarked above, that the system stays in this state during
a sufficiently long time for us to recognize the value of its energy. The wave function
is always supposed to be normalized, from which it follows that

2:h1 2 = 1.
k

On substituting the expression (6) for 'I/J into the evolution equation (5) and
taking into account that the 'l/JiO) are eigenfunctions of H(O), one finds

" dCk .1.(0) (21ri E(O) ) _ 21ri" ()V( ).1.(0)

Lj: dt 'Pk exp h k t - h ~ Ck t t 'Pk exp (21ri E(O) )
h k t .
(8)

On multiplying this equation by 'I/J~O)* exp [-(21ri/h)E}0)tj and integrating the result
over D, while taking the orthonormality of the 'l/JiO) into account, one obtains the
fundamental equations

dCl = h
dt 21ri"
~ Vlk(t)Ck(t) exp
[(21ri)
h (Ek(0) - El(0)]
)t , (9)

where
(10)

For V = 0, the Cl are constants, and 'I/J in Eq. (6) becomes the sum of eigen-
functions of H(O) with constant coefficients, which is a known case. But, if the
perturbation V is nonzero, the coefficients CR. will vary with time, whence the name
"method of variation of constants" given to this mode of calculation. (One may
easily verify that

and thus 2: ICk 12 = const. = 1

k

at every instant.)
112 Heisenberg's Fourth Uncertainty Relation

Integration of the fundamental variation equations (9) is in general difficult to

perform. One can draw various conclusions from them, which will here not be
studied in detail. We shall restrict ourselves to the case where it is known at the
time t = 0 that the system is in the state n, so that cn(O) = 1 and cm(O) = 0 for
m =f n.
The perturbation represented in the Hamiltonian by the operator Vet) being
weak by hypothesis, we arrive at a solution of Eqs. (9) that will be valid for a
certain length of time by postulating

dC m
dt = h27ri Vmnexp [(27ri)
h (En - Em)t ], m =f n, (11)

where the superscript 0 on E has been suppressed. The solution satisfying the initial
condition em(O) = 0 is

() _ V. exp[(27ri/h)(En -Em)t]-l
Cm t - mn En - Em '

which gives
2 2IVmn 12 [
Icm(t) 1 = (En _ Em)2 1 - cos h27r (En - Em) t
]
(12)

=
4IVmn 12
(En _ EmF
. 2
sm
7r (
h En -
)
Em t.
This quantity can be regarded as furnishing the probability for the system being in
the state m at the moment t; it is proportional to IVmn 12 , which imparts a special
importance to the matrix elements Vmn . However, in accordance with a remark
made above, as long as the perturbation lasts, one cannot physically catch the
system in the state of energy Em, because this would require an energy measurement
carried out with an uncertainty less than Em - En, while the time during which the
system remains in the state m is less than h/(Em -En). Only if at the instant t the
perturbation Vet) should abruptly cease would the system be caught in the state m
(then constituting a permanent final state), the probability being Icm(t) 12 that this
eventuality will be realized. Since the energy of the system is not measurable while
the perturbation lasts, the notion of its conservation is inapplicable during this time;
energy conservation becomes verifiable for the system only at the termination of the
perturbing action.

Note L. B. In perturbation theory, several tricky questions arise in connection

with the perturbation potential V. Consider first a system exhibiting a discrete
spectrum; for it, V can be a potential associated with internal interactions (inde-
pendent of t) or a potential of external origin (which might depend on t). In the first
case, one cannot attribute either a beginning or an end to the interaction, and it is
impossible to measure the energies E~O), E}:{), etc., in the absence of interactions. If,
on the other hand, V describes an external interaction, one may suppose that this
 Method of Variation of Constants. Transition Probability 113

action has a beginning and an end (e.g., caused by the approach, followed by the
recession, of the system producing the external field). One can then measure the
energy respectively before and after the interaction. In general these two energies
will not be equal, but one can save the conservation of energy law by saying that
energy has been received from, or surrendered to, the external system producing
the field.
In the latter situation, it is natural to consider the system ~ formed by combin-
ing the system S under study with the system S' producing the field. But, in order
to apply the law of energy conservation, one must be able to bring the systems Sand
S' together and afterwards to separate them, which requires the system ~ to possess
a continuous spectrum. This brings us to the theory of the next section (Sec. 6),
which will demonstrate how the conservation of energy appears. In summary, every
time that the conservation of energy is physically verifiable, it seems that one is led
back to the theory of Sec. 6.

It can happen that a particular Vmn is zero, which will signify that the transition
n - t m cannot take place directly. But sometimes the same transition can still occur
indirectly via an intermediate state p, namely when Vmp and Vpn are nonzero. Quite
generally, several states p, p', etc., may exist that can serve as intermediate states in
accordance with the scheme
p

n --+ p' --+ m.

"\, p" /

On utilizing the approximation used in the calculations above, one can write

(13a)

(13b)

where the summation index can assume the values p,p',p", etc.
Integrating Eq. (13b), one gets

( ) _ V. exp [(27ri/ h )(En - Ep) t] - 1 (14)

cpt-pn E n - Ep '

which, on substitution into Eq. (13a), gives

~ dC m _ "'v. V. exp[(27ri/h)(En -Em)t]-exp[(27ri/h)(Ep -Em)t]

27ri dt - L..J mp pn En - Ep .
p
(15)
114 Heisenberg's Fourth Uncertainty Relation

Integration of this equation, subject to the initial condition cm(O) = 0, then leads
to

(16)

x [eX P [(27ri/h)(En -Em)t1-1 _ eXP [(27ri/h)(Ep -Em)t1 -1].

En -Em Ep -Em

On introducing the notation

one thus finds that the probability for the system being in the state m at the time
t is given by

. In general, the terms in (Ep - En) are effectively not involved in the application
of the formula (17), which is of great importance in the theory of the interaction
between matter and radiation.
Here again, while the perturbation lasts, one cannot physically catch the sys-
tem in the state p or apply the conservation of energy. Only if the perturbation
ceases at the instant t is it possible afterwards to find the system in the state
m, the probability for this eventuality to occur being lem(t) 12. But, even in this
case, the system can at no time be found in one of the states p,p',etc., and con-
sequently conservation of energy never applies to these intermediate states. This
is what the study of transition probabilities in the next section will more clearly
show.
It should be noted that analogous circumstances obtain when the transition
n --+ m can occur in successive steps via several intermediary states.

6. TRANSITION PROBABILITIES

Up to this point we have assumed that the quantum states form a discrete set.
But it sometimes happens-and this is an important case-that one has to deal
with states forming a continuous spectrum (in collision problems, for example). It
is then necessary to reconsider the foregoing calculations in assuming that a system
which initially occupies a known state n is able to make a transition to a final state
belonging to a continuous set. We shall accordingly suppose that the number of
possible final states whose energies are contained in an interval (E, E + dE) is given
by an expression p(E) dE, where p(E) is a continuous, slowly-varying function of
E.
 Transition Probabilities 115

We therefore intend to transform the preceding theory so that it will apply to

the case where the final state m belongs to a small interval !:lE of a continuous
spectrum. Let us first suppose that it is possible for the transition n ~ m to occur
directly (Vnm -::f 0). Then, according to the formula (12), the total probability that
a transition will take place in the time t from the initial state n to any of the states
m belonging to the interval (E, E + !:lE) will be

P ( ) = 471"2 rE+AE IV, 12 [sin(n"/h)(E - En) t] 2 (E) dE (18)

n,AE t h2 JE nm (n/h)(E-En) p .

One can easily verify that, if En falls outside the interval !:lE, the integral in (18)
will be very small whatever the value of t, so that P(t)/t will tend to zero when
t ~ 00: One may say that the transition probability per unit time is zero in this
situation.
The same is not true when En belongs to the interval !:lEo In this case, if t is
large enough, the integral in (18) will increase proportionally with t, causing P(t)/t,
the transition probability per unit time, to assume a finite value; the transition will
thus take place to a noticeable degree.
More precisely, a study of the integral in (18) reveals that, in order to assert
that the transition occurs between the initial state n and a state comprised in the
interval 8E, one must wait a time of the order 8t = h/2n 8E, which corresponds
to the fourth uncertainty relation 8E8t rv h. It is therefore only at the end of a
sufficiently long time that the conservation of energy, Em ~ En, will be verified; but
in practice this time will appear to be very short because of the minute value of h
on our level.
Therefore, at the end of the time 8t, in practice very soon after the perturbation
has started, the conservation of energy will have been established, and one will be
able to assert that the system has passed from its initial state n of energy En to
a final state m of nearly equal energy Em (since Em - En rv h/8t). One will then
very approximately have

(19)

because, the preponderant contribution to the integral being made by the element
at E = En, one can, without committing a significant error, replace p(E) by p(En)
and extend the integration interval to all values of E. After the substitution u =
(7r / h )(E - En) t, Eq. (19) becomes

Pn,AE(t) _ 47r IT7 12 (E )

t - h vmn p jn
1+ -00
00
sin 2u d _ 47r 2 IT7
u2 u - h
12 (E )
vmn p n' (20)

The transition probability per unit time for the transition from the state n to
the state m of a continuous spectrum is thus given by

47r 2 2
Pn-->m = h IVmnl p(En). (21)
116 Heisenberg's Fourth Uncertainty Relation

This is the fundamental formula (Wentzel's formula) which holds when the direct
transition n --t m is possible. Vnm , then different from zero, is the matrix element
corresponding to the transition from the state n to the state m, of energy Em, of
the continuous spectrum. The preceding analysis shows clearly how, in accordance
with the fourth uncertainty relation, the conservation of energy is established with
ever-increasing precision as time goes by.
If the transition n --t m is impossible as a direct process (Vnm = 0), the same
transition can sometimes take place indirectly via an intermediary state Pi and
any discussion will then have to start from the formula (17). One can verify that,
apart from some exceptional cases, where resonance between the states m and P
would exist, the terms in Ep - Em appearing in the formula (17) will not make
any significant contribution to the transition probability. The above calculation is
then again valid under the simple replacement of Vmn by V~n' leaving us with the
result
with v.'
mn
="" p
Vmp Vpn
~ E -E '
n p

wherein m denotes a state belonging to a continuous energy spectrum.

Again conservation of energy obtains in the overall process n --t m but not in
the intermediary state p, because Ep can be different from En and Em.
And also here one cannot trap the system in the intermediary state, so that the
departure from the conservation of energy is not noticeable. Such a state is called
a "virtual" state, since it cannot effectively manifest itself.
For transitions that take place via several intermediary steps, one would find
analogous but more complicated formulas.

Physical Example. We now give a physical example illustrating the preceding

theory.
Let us first consider the scattering of light by an atom. The atom has a nor-
mal quantum state of minimum energy Eo and a set of excited quantum states
of energies El, E2, etc., larger than Eo. When irradiated by a light wave of fre-
quency v, the atom scatters the incident light without changing its frequency. An
analysis of this phenomenon, resulting in the discovery of a "dispersion" law for
the atomic species under consideration, leads one to regard an atom exposed to
light as oscillating between the state (of energy) Eo and the virtual states El, etc.
More precisely, an energy-conserving transition occurs between the initial state "in-
cident photon of frequency v + atom in the state Eo" and the final state "photon
of frequency v scattered generally in a direction making an angle with the inci-
dent direction + atom restored to the state Eo," this transition taking place via
a stopover in one of the states E 1 , E2, etc., in accordance with the scheme indi-
cated in the theory above. Since, apart from the exceptional case of resonance
(which requires a special investigation), the energy differences Ei - Eo do not equal
hv, energy is not conserved in the intermediate state Ei; but that is of little conse-
quence, considering that we are dealing with an experimentally unsustainable virtual
state.
In the quantum theory of interactions between charged particles, we are similarly
led to employ transitions via virtual intermediate states that do not conserve energy;
 Uncertainty Relations and Relativity Theory 117
one speaks of the virtual exchange of photons.
The theory of spectral line width provides a beautiful illustration of the fourth
uncertainty relation. Consider a spectral line corresponding to the transition of
an atom from the excited state Ej to the ground state Eo. Experiment shows,
and quantum theory demonstrates, that the emitted line has a width and that the
intensity distribution over the breadth of the line is given by the formula

l( ) 10, (22)
v = 411'2(ViO _ v)2 + ,214 '
where V;o = (E; - Eo)lh is the (central) frequency of the line and, the "damping
coefficient" of the initial state of energy Ei; that is, the probability of finding an
atom in the excited state E; at a time t after excitation declines as exp (-,t).
One sees that lev), which has a maximum at v = ViO, diminishes rapidly with the
distance from the center of the line:
1
lev) = 2"l(ViO) when Iv - viol = ,/411'.
It is conventional to say that ,/411' measures the "width" of the line. But it is
evident from the definition of, that one can follow the atom in the state Ei only for
a time U of the order II,: It is therefore impossible to measure Ei with a precision
exceeding SE '" hlSt '" hi. The frequency of the line is accordingly subject to the
uncertainty Sv = 8Elh "-'" which agrees well with the width of the line.
For a deeper study of the questions lightly touched upon in this section, we refer
the reader to Une nouvelle theorie de la /umiere, Vol. II (Ref. II, 16).

7. UNCERTAINTY RELATIONS AND RELATIVITY THEORY

In the domain where relativistic corrections need to be taken into account (speeds
approaching light speed), one can encounter novel forms of the uncertainty relations,
which were pointed out by Landau and Peierls (Z. Phys. B 69, p. 5 et seq.), and
which have given rise to numerous enough discussions.
In the relativistic domain, the energy E must be defined under inclusion of the
internal mass energy. Thus, for a particle of rest mass mo, one should write

moc2
E = (1 -132)1/2 '
where, per definition, f3 = vic. If

denotes the x component of the linear momentum, then the Hamiltonian equation
{)EI{)px = Vx gives the relation
SE = Vx 8px, (23)
118 Heisenberg's Fourth Uncertainty Relation

as one may easily verify. Starting from a state of known energy Eo and linear mo-
m~ntum po, let us perform an energy measurement lasting a time Dt; the uncertainty
in energy will then be DE ;::: hlDt. But DE = Vx Dpx and Vx is necessarily less than
c, with the result that
(24)
This is an uncertainty relation of a new form, which connects an uncertainty in
a Lagrangian momentum with the duration of a measurement, independent of the
uncertainty in the corresponding coordinate x. Thus, if relativity is taken into
account, any measurement of linear momentum would require a certain time, if it
is to be precise to some degree.
The novel form (24) can also be found by the following reasoning: Suppose a
measurement localizes a particle in the immediate vicinity of a point O. After the
measurement, one has a corresponding wave train 'IjJ of vanishingly small extent;
and, as we have seen, Fourier analysis teaches that such a wave train will comprise
all frequencies. But the relativistic formula

shows that infinitely large frequencies correspond to speeds infinitely close to c.

Consequently, if the measurement lasts a time Dt, the front of the wave 'IjJ can at the
end of the measurement already be a distance CDt from the point 0, so that finally
DX = cDt. Thus the Heisenberg relation Dpx DX ;::: h gives Dpx ;::: hl(cM), and we
recover the uncertainty relation (24).

Note L. B. On combining Dt ;::: hiDE with the requirement DE < E, one gets

But, since in relativity theory wave fronts can attain the speed c, one must put
Dq = c Dt, whence
hc h r;-o:;
Dq ;::: - = - VI - (32 ,
E moc
or
Dq ;::: >'(3,
sInce
>. = _h_ \/1 - (32.
mo(3c
Thus Dq ;::: >. for light, but Dq can be less than>. for material particles.

In relativity theory, the energy E of a particle and its linear momentum pare
connected through the relation E2 I c2 = p2 + m5c2, whence E dE = c2p dp; and,
 Uncertainty Relations and Relativity Theory 119
since p = Ev I c2 , it follows that SE = v Sp, allowing us to write the uncertainty
relation
vh
SE = v Sp ~ Sq' or SE Sq ~ vh, (25)
which links the uncertainty in energy to the uncertainty in position. A physical ex-
ample of this connection is provided by the experiments of Rausch von Traubenberg,
which were theoretically analyzed by Schrodinger [Z. Phys. 78, 309 (1932)].

L1x x
I " f

I
~ \
/26\
-
I~ \

Consider atoms that have been excited to a state of energy Ei and are traveling
with a constant speed v along an axis Ox (see figure), at each point of which there
exists a non-homogeneous magnetic field H(x). Because of the dependence of H
on x, the value of Ei will likewise vary with x. On returning to their normal
states of energy Eo, the atoms must therefore emit a spectral line whose frequency
will vary according to the position of the emitting atoms at the time of their de-
excitation. Accordingly, one may try to find the variation of Ei with x by observing
the frequency of the line emitted from every element of the axis Ox. Unfortunately,
as Schrodinger has remarked, one is hampered by the following circumstances.
First, two points of the axis Ox are "separable" from one another only if their
distance apart Sx is larger than >,,(x)1 sine, where >"(x) denotes the wavelength of
the light emitted at the point x and e the half-angle of the aperture belonging to
the observational instrument, as shown in the figure. On the other hand, during the
period of observation St, the atomic emitter is traveling towards the observational
instrument with a speed that can be as high as v sin e. Due to the Doppler effect,
this motion will produce an uncertainty in the wavelength equal to S>.. = (vic)>" sine,
whence SII = (v I c) II sin e, because

IS:I 16:1; =
and, since 6x = AI sine, one finds

SE = hvCi~e) ~ ~; and thus 6E6x ~ hv, (26)

120 Heisenberg's Fourth Uncertainty Relation
thereby recovering the uncertainty relation (25).
Let us write once more
h hv
8x ~ 8px = 8E·

This implies that it is possible to localize a particle of well-defined energy E (8E ~

E) in the interval 8x only if
hv v
8x ~ - =-
E v
(because E = hv), and thus 8x ~ (32 A, since>. = Vlv = c2 lvv.
In the Newtonian approximation (particles with low velocities), (32 is negligible,
so that this inequality is non-restrictivej the localization can be carried out in a
domain that is small compared with the wavelength. The situation changes when
v --+ Cj then (3 --+ 1, and the inequality becomes 8x ~ >.. One can thus no longer
localize the particle within a distance of the order of the wavelength. Accordingly,
for particles of speeds approaching c, and for photons in particular, localization on
the scale of the wavelength is impossible.
The problems exposed above have given rise to numerous discussionsj these bear
on the question of negative-energy states, well known in Dirac's theory, and the
general theory of particles with spin, just as on the non-existence of a positive
position probability density for photons and paired-spin particles. But no clear
conclusions have emerged from these discussions, and the question remains in need
of elucidation.
Other considerations involving the comparison of uncertainty relations with rel-
ativistic ideas have been developed by various authors, notably by Schrodinger
[Ann. Inst. H. Poincare 2, 287 (1932)]. We shall recall only some points, re-
lating to the measurement of times and lengths, with a reasoning that is a lit-
tle shorter than Schrodinger's. Suppose that in a Galilean frame of reference,
for which synchronization has been established, one wishes to regulate a clock
of rest mass Mo. For that purpose, we imagine that the clock emits a pho-
ton which is received at the time t by an observer situated at the distance f
from the clock: The photon must then have been emitted at the time t - flc,
and one will be able to adjust the clock accordingly. But to prevent the clock
from suffering a recoil at the time of emission, which would disturb the mea-
surement, the energy of the emitted photon must be very small in comparison
with the rest energy of the clock, i.e., hv ~ Moc2 is required. But the time 8t
needed for the emission of the wave train associated with the photon is such that
8t ~ hi h8v, where 8v is the uncertainty in the value of v, which necessarily is
much less than Vj consequently, 8t ~ 1/v ~ hlMoc2 • Since the recording of
the photon's arrival at the observer can occur at any instant within the time 8t
it takes the wave train to pass over the observer, we see that the setting of the
clock cannot take place with a precision corresponding to any uncertainty less than
TO = hlMoc2 •
Similarly, if one wants to measure the length of a ruler in a Galilean frame, this
measurement can, because of the relation 8x ~ hl8px, be performed only with an
 Formulas of Mandelstam and Tamm 121
uncertainty

since
Movx Eo
Px = (1 _ /12)1/2 ~ iF Vx and thus

Because 8v x :::; c, one thus is left with

h h
8x>--=--.
- Eo/e Moe
Accordingly, the length of the ruler can never be known with an uncertainty less
than
h
AO = - - ,
Moe
Mo being the rest mass of the ruler.
For the details the reader is referred to Schrodinger's exposition, the entire rea-
soning of which in the end boils down to the fourth uncertainty relation and to the
relations given above, but which contains very penetrating remarks on the role of
the time in wave mechanics.

8. FORMULAS OF MANDELSTAM AND TAMM

Starting from considerations that are perhaps in part somewhat questionable,

Mandelstam and Tamm arrived at some interesting formulas connected with the
fourth uncertainty relation [J. Phys. IX (4),249 (1945)]. These we shall now study.
The authors remark that, if a system occupies a stationary state 1f; ex: exp (21!'ivt) ,
the probability distributions for all the dynamical variables are independent of time,
as may easily be verified. From this they conclude that there must exist a general
relation between the dispersion a E of the energy and the variation in time of the
coordinates, momenta, etc. To see this, let us start from the relation

(27)

which is valid for every pair of observables A and B. Moreover, by definition,

whence one easily infers (if A is independent of t, as we assume) that

~~ = 2:i J
1f;*(AH - H A)1f; dT = 2:i rA, H], (28)
122 Heisenberg's Fourth Uncertainty Relation

where H is the Hamiltonian of the system. On taking B = H in Eq. (27), one

obtains
h -
aHaA>- IdAI
- 411" dt '
(29)
with aH = aE denoting the energy dispersion, which is the relation we sought (the
Mandelstam-Tamm relation). For a stationary state of known energy, aH = 0 and
dAjdt = O.
One can write the obtained equation in a different form. The system being
isolated, a H is constant, but a A can vary. Let us consider a time interval 8t and
denote by (f ~ the time average of a A over this interval (this average is not of the
same kind as the mean values which up to now we have designated by a bar). If one
integrates over the time interval tit, while noting that the integral of the absolute
value of a function is always larger than or equal to the absolute value of the integral
of the function, the result is

tit ~ -h A(t + c5t) - A(t)

(30)
aH 5t .
411" a A
Mandelstam and Tamm then introduce a "standard time" tiTA, which is defined as
the briefest interval of time during which the mean value of A will vary by (fA. For
this interval, the formula (30) reads
h
aHtiTA ~ 411" • (31)
From the formula (29) one concludes that, if the mean value of a quantity A is
to vary, it is not enough for a H to be nonzero; in addition a A must not constantly
be zero. This fact is evident in the case where A has a discrete spectrum, but the
situation is different for a continuous spectrum. One also sees from the formula (30)
that, if a A vanishes at a particular moment without A ceasing to vary, then initially,
i.e., for 8t very small, a A must vary much faster than A.
An interesting illustration of the preceding formulas arises when one considers
the propagation of a wave train along the x axis and takes A = x. Then x will
be the x coordinate of the wave train's center of mass, while a A may be regarded
as the mean wavelength of the train and 8TA as the mean duration of its passage
through any point. The relation aHoTA ~ hj411" shows that this mean duration
is larger the smaller aH is. We thus recover a well-known conclusion; but while
our earlier reasoning furnished this conclusion only in the absence of a field, the
conclusion here is valid even in the presence of an external field, because the argu-
ments that led to the formulas above do not at all assume the absence of such a
field.
Here is another example given by Mandelstam and Tamm. Let t.pn be a wave
function representing a particular state of a system for which the energy dispersion
is aH. [Note L. B. The function t.pn can be an eigenfunction of a quantity A that
does not commute with H.] With 'ljJ designating any state of the system, we consider
the operator Ln such that Ln'ljJ = Cnt.pn, where

en = Jt.p~'ljJ dT.
 Formulas of Mandelstam and Tamm 123

Ln thus denotes an of erator which isolates the component Cn<pn of 'IjJ; it is a projec-
tor, and obviously Ln = Ln. Since

(32)

the Mandelstam-Tamm relation gives

(33)

This inequality, which contains only Ln, can easily be integrated. Suppose that in
the initial state Ln(O) = 1, which means that the system originally is for certain in
the state <pn; then, on integrating from 0 to t, one finds

h211" O"Ht ~ "211" - arcsin

JLn(t). (34)

For 0 < t < h/( 40"H), this gives

Ln(t) ~ cos 2 C: O"Ht).

For t > h/(40"H), the inequality (34) places no restriction on the value of Ln(t),
which always lies in the closed interval [0,1).
Let us designate by T the mean life of the state <Pn such that Ln (T) = 1/2 if
Ln(O) = 1. Then the last inequality gives
h
O"HT ~ 8' (35)

which is a bit more restrictive than O"HT ~ h/411".

The paper of Mandelstam and Tamm also contains a somewhat less clear appli-
cation to the case of perturbations, which we shall not stress.
 CHAPTER 10

EXAMINATION
OF SOME DIFFICULT POINTS
IN WAVE MECHANICS

1. REDUCTION OF THE PROBABILITY PACKET BY

MEASUREMENT

In the physical interpretation of wave mechanics, measurement plays an essential

role. It is this procedure that provides us with new information, changes the state
of our knowledge about the particle or system under investigation, and abruptly
alters the form of the function 'if; representing this knowledge. If, for example, the
measurement is a more or less precise determination of position, then the wave train
representing 'IjJ before the measurement will be "reduced" to a wave train that is
less extended and perhaps, when the measurement is very precise, nearly pointlike;
which explains the name "reduction of the probability packet" given by Heisenberg
to this abrupt modification of 'IjJ. If, by contrast, the measurement is designed to
determine the components of linear momentum, it is in momentum space, rather
than coordinate space, where the sudden reduction of the wave train will take place.
The reduction of the wave train gives rise to a novel situation which was un-
foreseeable, since only the probabilities of the various possible outcomes could be
calculated before the measurement. Herein lies the indeterminism of the new me-
chanics (and von Neumann showed that this indeterminism is without doubt of
an essential nature, because it is impossible to restore probabilities to the state of
ignorance we would be in about the exact values of certain hidden variables).

Note G. L. The remark in parentheses was later crossed out by Louis de Broglie
when he became convinced of the fallacious character of von Neumann's reasoning,
which will be discussed at length in the second part of this book. In his later works,
in his theory of the double solution, to be precise, Louis de Broglie developed a
hidden-variables theory which contradicted the prohibition expressed by von Neu-
mann's theorem and that featured, besides the probability wave 'IjJ, another wave v
which is not susceptible to wave packet reduction. It follows that the uncertainties
in the measurement results do not exclude an underlying determinism.

After a measurement, one that furnishes us with the maximum knowledge con-
sistent with the theory of noncommuting variables (experiment of maximal measure-
 Impossibility of Discovering the Anterior State from the Posterior State 125

ment, defined in Chap. 6, Sec. 3), we can construct the wave function 1/J, representing
our knowledge after the measurement, and then follow its evolution in the course
of time with the help of the wave equation until we learn the results of some new
measurement, which modifies the state of our knowledge and abruptly interrupts
the regular evolution of the wave 1/J. The evolution of the wave 1/J between two
measurements, governed by the wave equation, is entirely determined by the initial
state of 1/J, since the wave equation is of the first order with respect to t. Thus
determinism exists in the evolution of probabilities between two measurement but
not in the succession of observable phenomena.

2. IMPOSSIBILITY OF DISCOVERING THE ANTERIOR STATE

OF A MEASUREMENT FROM ITS POSTERIOR STATE.
EFFACEMENT OF PHASES BY MEASUREMENT

Suppose we have just made a measurement and know the wave function 1/Jl
describing the state of our knowledge after the measurement. The question now
is: Can we start from this wave function and discover the one that existed before
the measurement? One might imagine this to be possible by, for example, taking
1/J after the measurement as the initial form of the wave function and following the
evolution of the latter back into the past with the aid of the wave equation wherein
the sign of the time has been reversed.
The question merits close inspection:
Assuming that we are dealing with a single particle (or system), let

be the wave function of the particle at the beginning, before the measurement, as it
emerges from our knowledge at that instant, expanded in terms of the eigenfunctions
'Pk of the observable quantity A that is to be measured. The exact measurement
of A will give one of its eigenvalues % say, for which the probability before the
measurement was ICJ 12 . After the measurement, the wave function will then be
1/Jl = 'Pl. This knowledge of 1/Jl does not tell us anything about the values of the Ck
before the measurement, not even about the value of the coefficient CJ corresponding
to al (except that CJ was nonzero). It is therefore impossible to start from 1/Jl and
recover 1/Jo.
The situation is more favorable if one has a large number N of particles (or
systems) that all are initially in the state 1/Jo. The measurement of the quantity A
on the N systems will very approximately furnish the value al with the frequency
N IClI2, the value a2 with the frequency Nlc212, etc. One will therefore get to know
the ICk 12, but this knowledge is not equivalent to that of the

because the phases or arguments Dk remain unknown. As a result, the phase differ-
ences between the components Ck'Pk of 1/Jo remain unknown. But these differences, as
126 Some Difficult Points in Wave Mechanics
we shall soon see, play an essential role. One can therefore not even with statistical
measurements go from 'l/JI to 'l/Jo.
To make our reasoning more concrete, we consider the measurement of position
and of momentum, while restricting ourselves to a free particle undergoing one-
dimensional motion.
If the quantity A is the linear momentum P, one should write

and the measurement made on the particle will give P = PI, which eventuality had
the probability Icd2 prior to the measurement. After the measurement one will have

and there is no way one can obtain 'l/Jo from 'l/JI. A statistical measurement made
on a large number of particles that initially had the same wave function 'l/Jo would
furnish the ICk 12 but would not give any information on the phase relations between
the components of 'l/Jo.
Let us now consider the case A = x (position measurement). The measurement
of x will give a particular value Xi, whose probability before the measurement was
l'l/JO(Xi) 12. After the measurement, the wave function thus is 'l/JI = 8(x - Xi), but
this knowledge of 'l/JI will not teach us anything about 'l/Jo, other than the fact that
'l/JO(XI) is not zero. Statistical measurements made on a large number of particles
having initially the same wave function 'l/Jo will thus furnish l'l/Jo(X) 12 for every point x
but tell us nothing about the relative phases of the 'l/Jo(x) values in different points.
For example, if one should find that I'I/J(x) 1 has everywhere the same value AI,
then 'l/Jo could be a plane monochromatic wave of amplitude A having an arbitrary
wavelength and any direction of propagation; it could even be of the form 'l/Jo =
Aexp [i 8(x)), with 8(x) denoting an arbitrary function.
One sees therefore that every measurement has the effect of completely ef-
facing the phases (Bohr). This remark has served as the point of departure for
Dirac in his first exposition of second quantization; and his method of introduc-
ing second quantization remains the most instructive one from a physical point of
VIew.

Note G. L. For an understanding of the evolution of de Broglie's ideas, one

cannot recommend too highly the reading of his book La theorie de la mesure
en mecanique ondulatoire (Ref. II, 27), where he shows that all measurements in
the final analysis reduce to measurements of position, which assumes that one has
beforehand decomposed the initial wave into wave packets separated in space, so that
the mere recording of the presence of a particle in one of them allows the unequivocal
attribution of a particular value to the quantity one proposes to measure. Now, it is
this resolution of the initial wave train into separated wave trains and the hitching
up of the particle with one of these wave trains that are responsible for the breaking
of the phase relations.
 Possibility of Postdiciion 127
The obliteration of phases attending any measurement means that the act of
measurement creates a break in the evolution of 'I/J that is insurmountable equally
well in the future-past direction as in the past-future direction.
The phase differences between the components CPk of the wave function 'I/J are Of
crucial significance: Any knowledge about 'I/J which does not include information of
the phases is radically incomplete. The importance of phases is strongly emphasized
by the study, so decisive in wave mechanics, of the interference of probabilities.

3. POSSIBILITY OF DISCOVERING THE PAST, STARTING

FROM A MEASUREMENT MADE AT A GIVEN INSTANT
(POSTDICTION)

We have seen that it is impossible for an observer who is aware of a measurement

made at the instant t on a system to reconstruct the wave function that described
the state of the system before the measurement for an observer possessing anterior
information.
Suppose, for example, that an observer A measures the position of a particle at
the instant tl, finding it at a point MI. As the initial value of the wave function
he will then take 'l/JA(M,tI) = 8(M - MI). Starting from here and invoking the
wave equation, he will be able to calculate the form 'l/JA(M, t) of the wave function
at any subsequent moment and specify the probability I'I/J A(M, t2) 12 of finding the
particle in an arbitrary point M at the instant t2 > tl' Let us suppose that a
position measurement made at the instant t2 locates the particle at the point M 2•
An observer B who knows only about the second localization (and not about the
first) can in no way discover the form of the wave function 'l/JA(M, t) employed by
the observer A who knows the first location. If B had the same knowledge, he
would be able, by going backwards in time, to ascertain the localization in MI
at the instant tl; but in reality that is now not possible. Even if a statistical
experiment provided B with the amplitudes l'l/JA(M,t2) 1 at all points M of space,
he would still be ignorant of the corresponding phases and thus unable to follow
the backward evolution of the wave function 'l/JA. The same situation obtains in
the case of momentum measurements, that is, in regard to the localization of the
representative point in momentum space.
Nevertheless, if it is impossible for the observer B who knows only about the
second localization to reconstruct with certainty the first [by establishing the form
of 'l/JA(M, t)), he can still determine the relative probabilities of the various locations
at the time tl, starting from the known localization of the particle in the point
M2 at the instant t2' For this purpose he will have to consider the wave function
'l/JB(M, t) which would evolve, under reversal of the time, in accordance with the
wave equation, starting from the initial form

He would then find the form 'l/JB(M, tt) for the wave function 'l/JB at the past instant
tt < t2, and the quantity I'I/J B(M, tl) 12 would for him represent the probability
that the particle would have been localized in the point M at the time tl: This
128 Some Difficult Points in Wave Mechanics

probability exists for the observer B who knows about the localization in M2 at the
instant t2 but is ignorant of the localization that occurred in MI at the earlier time
tl. Consistency requires that \"pB(MI,tt)\2 be different from zero (otherwise one
would find a zero probability for an event that actually occurs!). If one looks at the
matter more closely, even more is seen to be required: The probability \"pA(M2, t2)\2
that at time t2 localization will take place in M2, if one knows that there has been a
localization in MI at the time tI, must equal the probability that there will have been
a localization in MI at the time tl if one knows that there has been a localization
in M2 at time t2; that is,

with the preceding definitions of "pA and "pB. This equality is directly linked to
the symmetry property of Green functions, and its physical meaning can be seen as
follows.
We suppose that a light source of unit intensity is placed at the point MI co-
inciding with the entrance of any optical apparatus R that is capable of producing
interference or diffraction phenomena. In a point M 2 , situated at the exit of the
interference apparatus, the light intensity will then be i, say. If one now places a
light source of unit intensity at M 2, the intensity i will be recorded at MI. This is a
necessary consequence of thermodynamics, because, if the entire apparatus is imag-
ined to be immersed in a thermostat of temperature T, and if one places at each
of MI and M2 a nearly pointlike blackbody emitting thermal radiation, then either
body must radiate just as much energy as it receives from the other, since otherwise
the temperature equality of the two blackbodies will be destroyed spontaneously,
which is an impossibility.
The observer B can therefore, with the aid of the wave function "pB and starting
from the knowledge of the localization in M2 at the time t2, find the probabilities
for localizations at times prior to tl; this problem has an analogy in the classical
probability calculus concerned with the probability of causes.
Let us remark that in a sense one can say that the "true" wave function in the
interval (tl, t2) is the function "pA, because it takes into account the precision with
which one can know MI at any instant t > tl' (This assertion would be subject to
restrictions if one adopts a relativistic point of view, by taking the finite speed of
propagation of light into account; but they will here be left aside.*) If the observer
B knew the function "pA, he could extrapolate with certainty to the first localization
by following the evolution of "p A under time reversal. But since, by hypothesis, he is
aware only of the localization in M2 at the instant t2, his "information" is imperfect,
seeing that he does not know everything that could be known at the instant t2: This
deficiency in his knowledge compels B to employ an "imperfect" function "pB, which
datum gives him only probabilities, not certainty about the localization at the prior
instant tl'

* Note L. B. If the first localization is represented in spacetime by a point event

(Mt, tt), the second localization will correspond to a point event (M2' t2) that nec-
essarily lies in the future light cone of the first point event. The person B who
observes the localization in M2 at the time t2 could therefore have known (through
 Interference of Probabilities 129
the agency of a light signal if necessary) of the localization at M!; if he does not,
that would constitute a gap in his information.

Let us reiterate that all of the above applies also to the determination of mo-
menta, that is, to localization of the representative point in momentum space.

4. INTERFERENCE OF PROBABILITIES

Let us consider two non-commuting observable quantities A and B. The eigen-

values and eigenfunctions of A shall be OJ and 'Pi, respectively; those of B, f3i and
Xi, respectively. The eigenfunction set of the 'Pi cannot be identical with that of the
Xi, since the quantities A and B do not commute. We may suppose that the initial
state of the system is represented by the wave function

"p = L Cj'Pi·
However, since the Xi form a complete set, one can express every 'Pi in terms of the
Xk by formulas of the type
'Pi = LSikXk, (1)
k
the Sik being elements of a unitary matrix. One therefore gets

"p = L Ci'Pi = L CiSikXk· (2)

i,k

If the quantity A is measured on the system in the state "p, one of its eigenvalues
0i will be found, the probability of any OJ being ICjI2. After the measurement of
A, the system will find itself in the state 'Pi; and, in this state, a measurement of B
will, by Eq. (1), produce the value 13k with a probability ISik12. As a result, the total
probability of finding the value 13k for B when one performs first the measurement
of A and then that of B is

But suppose now that we carried out the measurement of B directly on the
initial state"p. Then, in view of the expansion (2) of"p in terms of the functions Xk,
a general principle of wave mechanics teaches that the probability for finding the
value 13k of B would be 12:: CiSik 12. This expression differs fundamentally from the
i
probability obtained in the preceding paragraph, because it depends on the phases
of the coefficients Ci and Sik, whereas the preceding probability does not.
Let us illustrate the above reasoning with the simple example of a particle con-
fined in its motion to a segment of length L along the x axis. In this domain, the
normalized eigenfunctions of the linear momentum P = A are

'Pi = 1 [(27ri)]
..;L exp - h PiX •
130 Some Difficult Points in Wave Mechanics

Take
1/; = 0~ Vi
ci exp [(27ri)
h (Wit -PiX) ]
,
to be the wave function of the particle in its initial state. If one first measures P
and then the position X, the probability of the value x = Xo will be

because 2::, ICi 12 = 1; all positions in the domain of length L are therefore equally

probable.
If, by contrast, one measures x directly in the initial state, the probability of
x = Xo becomes

12:, )r exp [ C~i) (Wit - PiXO)] 12

and results from the interference of the planes waves that make up 1/). In particular,
if the initial1/; is a stationary wave formed by the superposition of two waves of the
same frequency and the same amplitude that propagate in opposite directions along
the x axis, one has

and the probability of finding x = Xu in this initial state becomes

(3)

= ±[1 + cos (~~ px + 82 - 81) ].

We in this way obtain a stationary wave, for which Eq. (3) explicitly shows that
the interference of probabilities depends essentially on the phases (whose role thus
emerges as central) and that it is these phases which allow the interpretation of
interference phenomena in the usual sense attaching to this concept in optics.

Note G. L. De Broglie subsequently abandoned the expression "interference of

probabilities," which had become customary in the theory, and returned to his
original idea according to which there truly exists a physical wave (his wave v, which
has the same phase as the wave 1/; but not the same amplitude) which interferes in
the usual sense of the word and guides the movement of the particle in accordance
with a law that takes into account the statistical results predicted with the help of
the wave 1/;, which obviously retains its statistical significance.
 Interference of Probabilities 131

The fact that the probability of the value 13k of B measured directly on the
initial state is 12: CiSik 12, and not ~ Ic;l21sik 12, can appear to be a priori in conflict
I I

with the theorem of compound probabilities. In reality it is not: The probability

2: ICi 121Sik 12 is indeed the proper expression to use when one makes a determination
i
first of A and then of B, since every term appearing in this sum is a product that
results from multiplying the probability to first obtain some value (Xi of A with the
probability to then (the value (Xi of A having been established) get the value 13k for
B. There is no reason why the probability defined in this manner should equal the
probability for finding 13k directly by a measurement of B on the initial state.
What creates a bit of confusion on this question is that in mathematical statistics
one most often assumes that the measurement of a random quantity (which always
is of a macroscopic nature in ordinary statistics), a procedure that statisticians
generally call a "trial," does not in any way modify the probabilities of other random
quantities. Thus if, in seeking to establish a correlation between height and chest
size in a batch of draftees, one measures these two quantities for all draftees, it
is assumed that the height measurement does not affect the size of the chest, and
conversely. Accordingly, if x denotes the height of a draftee and y the circumference
of his chest, one writes

(4)

Py,(Xk) denoting the probability of finding a height Xk for any draftee whose chest
measurement is Yi, and there is no need here to specify if the measurement of x
is made before or after that of y. These hypotheses, if natural, are true only for
macroscopic quantities. In the microscopic quantum domain, the measurement of
one random quantity (a trial) modifies the probabilities of the others. The probabil-
ity of A is thus not the same before as after the measurement of B: The probability
of finding a particular value of B, if its measurement is preceded by a measurement
of A, is correctly given by the theorem of compound probabilities (as ~ Ic; 121sik 12),
,
but it does not equal the probability (12;: Ci Sik 12) of the same value of B measured
I
directly on the initial state.

Note L. B. Consider a hand of 32 cards. Each card has a suit (heart, spade,
club, or diamond) and a rank (ace, king, etc.). One may note in succession the
suit and the rank. Let us begin with the suit. The hypothesis implicitly adopted
by statisticians is that the recording of a suit does not modify the probability of
a rank. Thus the probability of recording a king after one has noted the suit is
1/8, and the probability of finding a king before noting the suit is the same, viz.,
4 x 1/4 x 1/8 = 1/8. But logically nothing prevents us from imagining that the
observance of the suit modifies the probability of the rank. One can, for example,
assume that the probability for recording a king becomes zero if the observed suit
is red (heart or diamond), while, if the observed suit is black (club or spade), the
probability for finding a king becomes unity. According to the theorem of compound
132 Some Difficult Points in Wave Mechanics

probabilities, the probability for finding a king after the suit has been noted then
becomes: 1/2 x 0 + 1/2 x 1 = 1/2.

Note G. L. The author's preceding note briefly calls to mind an example which
he developed a little earlier in his article on "La statistique des cas purs en mecanique
ondulatoire" (Refs. V, 44; III, 8). De Broglie reports (Ref. II, 29, p. 60) that the
conclusions he reached in this article caused "some uneasiness" in his mind, "which
has without doubt contributed to incline it towards another interpretation." The
latter would be based for a large part on the deepening of the schema of quantum
statistics and on the idea that is in fact compatible with the existence of a hidden
classical statistical schema. These problems are treated in La theorie de la mesure
and in Etude critique (Refs. II, 27 and 29).

5. SOME CONSEQUENCES OF THE DISAPPEARANCE

OF THE TRAJECTORY CONCEPT

Note G. L. It is clear that Louis de Broglie would no longer write such a heading!
He would at least add the adjective "observable" to the word "trajectory."

We have seen that the notion of a trajectory disappears in wave mechanics, at

least every time one leaves the domain of validity of geometric optics for the propa-
gation of the 'ljJ waves. When this approximation is valid, one can retain the concept
of trajectory and consider the nearly pointlike wave trains as describing ray trajecto-
ries; but as soon as, for example, interference and diffraction phenomena intervene,
the idea of rays and consequently that of trajectories become inappropriate.
A particle in ordinary space (or the representative point of a system in its con-
figuration space) can be located only every now and then by a measurement; and
between localizations no trajectory is in principle attributable to it (or to its repre-
sentative point). From this fact, important differences result in the application of
the notion of probability to classical mechanics and to wave mechanics, respectively.
Let us, employing the ideas of classical mechanics, consider an ensemble of
possible motions corresponding to the same Jacobi function S. The theory of
Jacobi instructs us to treat the envisaged trajectories as rays of wave propaga-
tion perpendicular to the surfaces of constant S as wave fronts. If we were deal-
ing with an infinity of particles describing all the trajectories of the class under
consideration, the spatial density of these particles would be expressed, as we
have seen, by the formula p = Ilj! 1 2 , where lj! is the wave defined by the the-
ory of Jacobi. With only one particle present, only one trajectory is described,
and 1lj!12 then will represent the probability density for finding that particle in
a given point at a given instant. Probability therefore enters here as a conse-
quence of our ignorance of the trajectory that actually is described. In princi-
ple, the actual trajectory and the particle motion along it could be calculated
by starting from the initial position and velocity of the particle. But if part
of these initial conditions is missing, one would only know which are the possi-
ble trajectories; and only a probability, not a certainty, would be given to find
the particle in a point M at a time t. If an observation allows one to de-
 Disappearance of the Trajectory Concept 133

tect the presence of the particle in the point M at the instant t, the parti-
cle will be known to describe a trajectory passing through M; and thenceforth
one will for sure be unable to detect the particle anywhere but on this trajec-
tory. The probability 1~12, which was different from zero in an extended region
of space, thus merely expressed our ignorance of the trajectory that was actu-
ally being followed: It loses all significance from the moment we get to know
the trajectory. Such is the classical point of view, which is consistent with
our intuitive and traditional conceptions in classical science. In particular, this
view allows deterministic motion, and probability intrudes only as a consequence
of our ignorance about the factors required for the determination of the actual
motions.
Completely different is the [present] viewpoint of wave mechanics. [Note
G. L. The restrictive adjective "present" was inserted afterwards by the author,
who no longer firmly regarded as evident the point of view expressed in this para-
graph.] For it, the notion of trajectory is merely a first approximation, valid only
when geometrical optics is applicable to the propagation of the ~ wave. As soon as
this is no longer the case, in particular every time that interference or diffraction of
the wave ~ occurs, the trajectory concept becomes useless, and one can speak only
of successive localizations of a particle in space (or of the representative point in
configuration space) by measurement operations, in the general sense of these words.
Probability thus intervenes in a novel way: It is not any more an expression of our
ignorance concerning a trajectory (since there no longer is any), and the arguments
of von Neumann, to which I have already alluded, show that this change is not due
to the existence of additional variables whose values elude us (hidden variables).
Determinism no longer exists, only probabilities do. In other words, nothing allows
us to predict exactly (except in very particular cases) the result of a measurement;
we can only assign a probability to each of its possible outcomes.
What remains when one goes from classical mechanics to wave mechanics, that
is, from geometrical optics to the wave optics of the ~ wave, is the expression I~ 12 for
the position probability, but without this expression giving rise in wave mechanics
to the existence of a hidden trajectory.
At the Solvay Conference in 1927, where the new ideas were the subject of
closely argued discussions, Einstein, who opposed these ideas, underlined some of
their consequences that very poorly conform with our intuition. In particular, he
developed the following example: Suppose a particle is perpendicularly incident on
a screen that is pierced by a circular opening. Behind this screen is situated a
photographic film bent in the shape of a large hemisphere.
If the opening is very small, the wave ~ associated with the particle will be
diffracted on traversing it, and the amplitude of ~ will be nearly constant over the
hemispheric film, since the aperture acts as a tiny source. If, at a given instant t, an
observation allows the detection (by a photographic impression) of the impact of a
particle at a point A of the film, the interpretation of this fact will be very different
depending on whether one reasons according to the classical ideas or the new ideas.
Under the classical rules, one would say: Any particle that travels through the
opening of the screen has a trajectory which necessarily must strike the film in a
definite point (dotted line in the figure). But as long as this impact has not been
recorded, one will not know which trajectory the particle actually followed. It is for
134 Some Difficult Points in Wave Mechanics

this reason that we attribute a nonzero probability (given by 10/12) to the arrival of
the particle at every point of the film. But as soon as the particle's presence at a
spot A on the film has been registered, we are acquainted with its trajectory, and
the probability for finding the particle at any other point B of the film will drop to
zero.
But with the new ideas, we are forced to admit that there are no trajectories: The
propagation of the wave to the right of the screen results in a diffraction effect.
As long as the localization at A does not take place, the particle is in some way
potentially present over the entire photographic film with the probability 10/12. As
soon as the particle manifests itself in A, the probability for finding it at any other
point B becomes zero, since only one particle is associated with the wave 0/. The
interpretation of this fact, quite simple when one admits the existence of a trajectory,
becomes here more mysterious. It is indeed impossible to understand with our
classical ideas on space and time (or even with relativistic ideas of spacetime) how
the mere fact that a photographic effect has taken place at A can instantaneously
prevent the production of a similar effect at every other point B of the film, unless
it is assumed that the particle is localized in space at every instant and describes a
trajectory in the course of time. If, in the new mechanics, the notion of trajectory
is· abandoned, one is led to imagine that the particle, while being an indivisible
and localizable unit, is not actually in general localized in space and time: It is in a
sense virtually present over the entire extent of the wave train, and this is what Bohr
clearly expressed in saying that "particles are entities defined in a blurry fashion
over extended regions of spacetime."

Note G. L. It is precisely this phrase which less than two years later made
de Broglie say that Bohr "is a bit like the Rembrandt of modern physics, but he
sometimes displays a certain taste for the clair-obscure" (Refs. II, 25, p. 14; III, 6,
p. 132). He soon had to abandon Bohr's point of view, return to that of Einstein,
and consider that one must, whatever the difficulties, recover the clear images of
classical physics.

In Einstein's example, the particle would in some way be scattered into a virtual
 Disappearance of the Trajectory Concept 135
state throughout the entire space behind the screen; and at the moment when the
photographic effect occurs in A, the particle would condense, so to speak, in a
point where it produces an observable effect. No mechanism in accord with the
old, or even the relativistic, ideas of space and time can, it seems, interpret such
an instantaneous contraction, which is moreover intimately linked to the indivisible
character of the particle. Einstein, assuming relativistic ideas on space and time,
viewed this conclusion as an objection against the ideas of the new mechanics.
Today, the latter appearing to be well established, it seems one must rather regard
the foregoing conclusions as proof of the insufficiency of our conceptions about space
and time, even when they are amended by the theory of relativity.
One must moreover not forget that the "localization" approach corresponding
to the corpuscular aspect is an extreme point of view: In practice it is always com-
bined with the complementary viewpoint of "dynamical state," which corresponds
to the undulatory aspect (/I = Elh, ..\ = hlp). But this second aspect obviously
"transcends" the spacetime frame (the monochromatic plane wave corresponding to
a well-defined state of motion occupies all of space and time!),* and it is the only one
that allows us to state the conservation of energy and the conservation of momentum
principles. [* Note G. L. This same remark will later lead de Broglie to deny any
physically real existence to the plane wave: "A plane wave is an abstraction: Ex-
perimentally one always deals with a wave train that is limited in space and whose
time of passage through a point is limited in time." (Ref. II, 26, p. 238.)] It is in
this fashion that, according to Bohr "the individuality of the particles transcending
the spacetime frame satisfies the requirements of causality."

Note L. B. He should rather say "the requirements of the law" (in Meyerson's
sense).

Note G. L. The author returned to this analysis (Ref. II, 29, p. 37) and showed
that one can, on the contrary, from it draw an argument in favor of a wave associated
with a particle propagating in physical space and "composed of a very small region
of strong concentration."

To further illustrate the concepts of the new mechanics, let us also consider
the following example cited by Heisenberg. Suppose a beam of light is incident on
a half-silvered mirror M (shown in the figure below), which splits the beam into a
reflected wave and a transmitted wave. One cannot say that the particle, on arriving
at the mirror, "chooses" between the reflected and the transmitted wave, because
the arrival of the particle at the mirror is not an observable event: As long as the
particle is not localized, it exists potentially in both the reflected and the transmitted
waves. If, at a given instant, one manages to localize the particle in either the
reflected or the transmitted beam, the other beam even ceases to exist, because it
corresponds to a possibility that has not been realized; and this well demonstrates
the "non-objective" character of the waves '1/;. But if, instead of trying to localize
the particle, one places a second mirror at M', he will in the region ABCD observe
interferences corresponding to the variation of the localization probability in this
region; which shows that it is necessary, until localization occurs, to consider both
the reflected and transmitted beams.
136 Some Difficult Points in Wave Mechanics

6. DISCUSSIONS CONCERNING "CORRELATED" SYSTEMS

Lively and interesting discussions, participated in by qualified scientists, have

taken place on the subject of "correlated" systems, that is, systems which, hav-
ing been allowed to interact, afterwards find themselves in states for which their
probabilities are not independent.
These controversies have been energized by a paper of Einstein, Podolsky, and
Rosen [Phys. Rev. 47, 777 (1935)], commented on in an exposition by Schrodinger
[Naturwiss. 23, 887, 823, 844 (1935)], to which an article by Bohr [Phys. Rev.
48, 696 (1935)] responded. Other remarks on the same topic appear in a paper by
Furry [Phys. Rev. 49,393 (1936)].
Let us expound the indicated difficulty first by an abstract reasoning. We con-
sider in a general way two systems 1 and 2, designating by Xl the set of coordinates
belonging to system 1 and by X2 the set of coordinates for system 2. Initially these
systems are "separated," Le., not interacting; the wave function for the combined
system then has the form 1}>(XI,X2) = 1}>1(XI)1}>2(X2) (see Chap. 3, Sec. 2).
Subsequently the systems are allowed to interact, before they separate again, so
that finally all interaction ceases. In order to expand 1}>, we consider two quantities
A(l) and A(2) relating to systems 1 and 2, respectively, with associated complete
operators. Let ap) and c,oP) denote the eigenvalues and eigenfunctions, respectively,
of A(1); and let a~2) and c,o)2) be the corresponding quantities for A(2). At every
instant during the interaction, one can then write

1}>(Xl, X2, t) = I>ik(t)c,o~1)c,o12), (5)

i,k
where the equations for the variation of constants enable us to follow the variation
of the Cik(t) with time. When the interaction ceases, one gets

1}> = L Cikc,o~1)c,o~2), (6)

i,k
 Discussions Concerning "Correlated" Systems 137

wherein the coefficients cik now are constants.

It often happens that the preceding expansion assumes the form

t/J = 2>ii'<PP)<P~~)' (7)

i,i'

such that to every i there corresponds one and only one i', and conversely. If, after
the interaction, one measures the quantity A(2) on the system 2, and if the value
a~~) is found, the general principles of wave mechanics allows us to assert that the
system 1 is in a state for which the quantity A(1) has the value ap). Owing to the
original interaction, there thus exists a "correlation" between the quantities A(1)
and A (2), although the systems 1 and 2 are separated.
Something in this conclusion seems outrageous. Up till now, when a system exists
in a state for which the quantity A does not have a well-determined value but rather
a set of possible values, we have said: If one carries out a precise measurement of A,
the system will undergo a transition to a state where A has a well-determined value;
this moreover happens because the measurement itself of A produces in the system
an uncontrollable perturbation, which causes us to lose all knowledge of the values
of those quantities that do not commute with A. And we explained this fact by
remarking that, in measuring A, one necessarily exerts an action on the system, an
action which the existence of a quantum does not permit us to diminish indefinitely.
But for the correlated system the situation becomes paradoxical: The measurement
experiment is performed on the system 2, which is supposedly separated from the
system 1, and still this act changes the state of system 1. As Schrodinger puts it,
"that would be magic" ("Das ware Magie").
One could evidently try to escape from this dilemma by saying that the exper-
iment done on 2 does not change the state of 1 but only our knowledge of this
state. But then one would have to admit that after the interaction the quantity
A(1) possesses a definite value, and that thus the presence of several terms in the
expansion of t/J only represents our ignorance of this exact value of A (1). The prob-
abilities that are introduced in wave mechanics would then simply result from our
ignorance of the actual values of the quantities and not from an indetermination
of these values, which view would lead to a "classical" interpretation of wave me-
chanics. Now, as has already emerged from several results presented above, in
particular from the interference of probabilities, and as will be shown in more de-
tail further on, this classical interpretation of wave mechanics is not admissible.
Observable quantities do not in general have values before measurement occurs:
It is the measurement which, in a certain sense, creates the value of a quantity.
This claim might appear very plausible if the measurement is performed on the
very system to which the quantity belongs, but it appears inadmissible when the
measurement is made on another system that is completely separated from the
first.
Einstein, Podolsky, and Rosen (EPR) further clarified the problem by stressing
the physicist's freedom to "choose" the type of measurement he is going to perform.
Instead of the quantities A(1) and A(2), one may consider other quantities B(l) arid
B(2), with eigenvalues and eigenfunctions .a!1), X~l) and .a~2), X~2), respectively. It
138 Some Difficult Points in Wave Mechanics
can then happen (EPR gave an example of this, which we shall discuss later) that,
after the measurement, 'IjJ has the form

'IjJ = 2: Cjj' cpP)cp~;) = 2: d kk , X11>X1~)' (8)

i,i' k,k'

with a one-to-one correspondence between the i and the i' as well as between the k
and k'. Then, if after the interaction has ceased, one measures A (2) on the system
2 and finds that it has the value a~;), he would know that A(1) has the value ap)
for system 1; but if, on the other hand, one measures B(2) and finds the value
11k;), he would know that B(l) has the value I1k1) for system 1. Now, after the
interaction has ended, the physicist is free to measure either A (2) or B(2) on system
2, and this is true even if A(2) and B(2) are noncommuting quantities and thus
not simultaneously measurable. Therefore, without in any way interfering with
system 1, which is separated from system 2, one can be led, after the measurement
on system 2 has been carried out, to attribute precise values to A(1) or to B(1).
This conclusion is again very paradoxical, appearing even to be in conflict with the
uncertainty relations of Heisenberg when A (1) and B(l) are the canonically conjugate
quantities x and px.
Einstein, Podolsky, and Rosen employed the following definition of the "physical
reality of a quantity": "If, without in any way disturbing a system, we can predict
with certainty ... the value of a physical quantity, then there exists an element
of physical reality corresponding to this physical quantity." In the case envisaged
above, the quantities A(1) and B(1) thus would have physical reality after the inter-
action; and, since wave mechanics does not permit the simultaneous specification of
exact values for A(l) and B(l) when these quantities are canonically conjugate, it is
necessary, according to EPR, to conclude that "the quantum-mechanical description
of physical reality given by the wave function is not complete."
To examine the question more closely, it is convenient to present the example
that was given by EPR and subsequently commented on by Schrodinger.
Consider two particles that are constrained to move along the axis Ox. We thus
have two systems, each of which is defined by a single coordinate, Xl for the first
system and X2 for the second one. Suppose that the wave function for the combined
system is

(9)
= j 8(X1 - a)8(x2 - a) da.
Since
8(X1 -a) = jexP [27rik(X 1 -a)]dk, (10)
 Discussions Concerning "Correlated" Systems 139

one can also write

(11)

From these different expressions for '1/;, we conclude: (1) If one measures X2, one
gets Xl = X2; (2) if one measures k 2, one gets kl = -k2 (up to a factor Ilh, the
quantities kl and k2 here are the linear momenta of the two particles).

Note L. B. In all, Xl - X2 = 0 and kl + k2 = o.

One can have a state 'if; giving precise values to Xl - X2 and to PI + P2 <X kl + k 2,
because

[(Xl - X2), (PI + P2)] = [XI,PI] + [X}'P2] - [X2,PI] - [X2,P2] = 0,

in view of [XI,P2] = [X2,PI] = 0 and [Xl,ptl = [X2,P2] = hI27ri.
What does the function 'if; physically represent? To find out, consider a screen
containing a very narrow straight slit. If, in the plane of the screen, we choose the
X axis perpendicular to the slit, the position of the slit will be defined by X = a.
Let the 'if; waves of particles 1 and 2, both of which are assumed to exist in a
state of uniform motion, be perpendicularly incident on the front of the screen. On
the back of the screen, the wave function 'if; of the two-particle system will have the
form given above, because it is zero everywhere except on the slit at X = a, and
the two particles must pass through the slit in order to enter the space behind the
screen. Suppose we know exactly the state of motion of the screen in the direction
Ox; then we may ignore the position of the slit, and all values of a will become
equally probable. This is why we have, in (9),

'if; = J
6(XI - a)6(x2 - a) da,

which expresses the presence of the particles in a slit, the position of which is entirely
indeterminate.
But since we know exactly the linear momentum of the screen, which is assumed
not to vary as a result of the passage of the particles through the slit [see Remark
surrounding Eq. (13)], conservation of the X component of the linear momentum
gives kl + k2 = 0, all values of kl being equally probable; which explains the form
of 'if; given in the second expansion (11), viz.,
140 Some Difficult Points in Wave Mechanics

However, say Einstein, Podolsky, and Rosen, we are free to measure X2 or k 2,

which leads us to attribute either the value Xl = X2 to the position of the particle 1
or the value kl = -k2 to its linear momentum (divided by h). And, since neither of
these measurements affects particle 1, we can attribute either a position or a linear
momentum to this particle without acting on it in any way; from where arise the
difficulties indicated above.
Now, we must confess that, presenting it as we have just done, the foregoing
example is frankly bad. In fact, even the form of 'I/J implies that the position of
particle 1 coincides with that of 2. One can therefore not say that the systems here
are separated; and if they no more are separated, there no longer exists a paradox,
since every measurement performed on one system will evidently also act on the
other. To avoid this objection, EPR studied not the 'I/J that we have considered
above, but the following wave function:

'I/J = IIO(XI - a)o(x2 - b - d)o(a - b) da db

= IO(xl - a)o(x2 - a - d) da,

(12)
'I/J = IIO(k l + k2 ) exp [21ri(klXI + k2X2)) exp (-21rik2 d) dk 1 dk2

= lexP[21rikl(xl - X2 + d)) dkl,

where d is a nonzero constant. With this 'I/J, we see that the measurement of k2 on
particle 2 always implies that kl = -k2, but that the measurement of X2 on the same
particle gives rise to Xl = X2 - d. One can therefore here recover the paradoxically
appearing consequences developed in the preceding example; but, since one no longer
has Xl = X2, it appears possible to claim that the systems are separated, thus leaving
the paradox unresolved.
In reality, not much is gained by the introduction of the parameter d. This is
seen clearly when one tries to physically interpret the new form of 'I/J, as Bohr did
in his article. For this purpose, we need to consider a screen pierced not by a single
narrow slit, but by two very narrow parallel slits a distance d apart. The waves of
the two particles, initially plane monochromatic waves, are allowed to be normally
incident on the front face of this screen.
If in the initial state we know exactly the motion of the screen along Ox (i.e.,
its Px), then the position of the screen along Ox (and thus the abscissa a of the first
slit) will be unknown, all values of X being equally probable. The value of 'I/J for the
two-particle system is then correctly given on the back face of the screen by

lo(xl - a)o(x2 - a - d) da,

which expresses the presence of particle 1 in the first slit and that of particle 2 in
the second slit, the positions of the two slits being indeterminate.
 Discussions Concerning "Correlated" Systems 141

r
X2

11 X2 = Xl +d
Xl

If the linear momentum of the screen does not vary, one will have kl + k2 = 0,
in agreement with the following expression for 'IjJ:

'IjJ = JJ8(k + k2) exp [21ri(klXl + k2X2)] exp (-21rik2d) dk1dk2.

1

Remark. It might appear arbitrary to assume here (as we did two pages back)
that the screen does not exchange linear momentum with the particles. But this
objection can easily be removed: On denoting the abscissa of the first slit by xo,
the initially given momentum of the screen by Ko, and its final momentum by Kl,
we can take the motion of screen into account by writing

(13)
'IjJ = JJJJexp[ -21ri( kl + k2 + Kl - Ko)a] . .. dkl dk2 dKl da,

'IjJ = J8(Xl - a)8(x2 - a - d)8(xo - a) da.

These forms show that, if one measures the abscissa Xl of the first slit, he will get
Xl = Xo and X2 = xo+d, and that, if one measures Kl and kl ,he will find k2 = Ko-
Kl - kl' a relation which expresses the conservation of momentum. The preceding
reasoning implicitly assumes that one has verified Kl = Ko by measurement.

What is relevant in this physical illustration of the significance of the 'IjJ un-
der investigation is its showing, since the exchange of momentum between the two
142 Some Difficult Points in Wave Mechanics

particles occur by way of the screen strip of width d, that these particles must be
viewed as interacting when X2 = Xl + d. They are therefore not "separated," and
it again appears that our problem arises without doubt, in part at least, from the
exact sense of the word "separated," which is all the more difficult to define in wave
mechanics, because the localization of the particles in a given state is in general not
a precise one.
In his paper responding to that of EPR, Bohr stressed the fact that in wave
mechanics one must never divorce the mathematical formalism from the experi-
mental apparatuses that permit the measurements. This is why he immediately
gave a physical interpretation (with the aid of the two-slit screen) of the 'Ij; stud-
ied by EPR. One then sees that the two possible measurements, that of position
and that of momentum, correspond to different experimental setups. The posi-
tion measurements assume that one fixes the screen relative to a macroscopic sup-
port which defines our space coordinates. Then the first slit will have a known
abscissa Xo = a, and one will get Xl = Xo and X2 = Xl + d. But knowl-
edge of the linear momenta will be completely lost, because, the screen being
rigidly tied to the support, the linear momenta absorbed by the screen will get
lost in the support. Conversely, if one wants to measure the linear momenta,
one must measure the initial and final momenta of the screen, which assumes
that the screen is allowed to remain mobile; as a result, one cannot exactly know
the abscissas of the slits. In this case, the variation Ko - Kl of the linear mo-
mentum of the screen being known, the measurement of k2 will give the value
kl = -k2 + Ko - Kl of kl. One cannot treat the problem in the abstract,
without referring to the experimental apparatus, because one must, at the start
of the experiment, arrange a setup that corresponds to one or the other of the
possible measurements. Bohr expresses this by saying: ". .. we are not deal-
ing with an incomplete description, characterized by the arbitrary picking out
of different elements of physical reality at the cost of sacrificing other such el-
ements, but with a rational discrimination between essentially different experi-
mental arrangements and procedures which are suited either for an unambigu-
ous use of the idea of space location or for a legitimate application of the con-
servation theorem of momentum." On the same occasion, Bohr also made the
important remark that, in the final analysis, one always measures momentum
by transferring it to a macroscopic body, which can be treated with the ideas
of classical mechanics and whose momentum one can consequently determine via
two consecutive position measurements separated by a sufficiently long interval of
time.

Note G. L. De Broglie later crossed out the word "important," replacing it by

"questionable enough," and added a question mark after "always." In commenting
on this remark of Bohr in Ref. II, 26, he qualified "questionable" and added: "On
the contrary, it seems to us that one never measures the momentum of a particle
in this manner but rather infer it from the observed localization of another parti-
cle under possible application of the conservation of momentum." He could have
added: or by the direct localization of the first particle, after its passage through
a device capable of separating wave trains according to the values of momentum.
Wherever Bohr speaks of interchanges between the particle and the measurement
 Complementary Remarks on the Einstein-Bohr Controversy 143

device, de Broglie will introduce the separation of wave trains, permitting one to
reduce the measurement of any quantity to a measurement of localization.

To us, Bohr's "experimental" point of view seems moreover to be in accord with

the remark that the two systems are not in reality "separated" at the moment when
one has the choice between two possible measurements. When, having made this
choice, one sets up the corresponding measuring device, the state for the example
under consideration is represented by the "p specified above, and one cannot say that
the two systems are separated. * Referring to the definition of physical reality given
by EPR, Bohr characterizes as ambiguous therein the phrase "without in any way
disturbing a system." About this point he says: "Of course there is ... no question
of a mechanical disturbance of the system under investigation during the last criti-
cal stage of the measuring procedure. But even at this stage there is essentially the
question of an influence on the very conditions which define the possible types of
predictions regarding the future behavior of the system." These conditions consti-
tuting an element inherent in the description of all phenomena to which the term
"physical reality" can be applied, Bohr rejects the conclusion that the description
given by wave mechanics is incomplete. On the contrary, this description appears
to him as characterized by "a rational utilization of all possibilities of unambiguous
interpretation of measurements, compatible with the finite and uncontrollable inter-
action between the objects and the measuring instruments in the field of quantum
theory." In this characterization Bohr sees one of the essential aspects of comple-
mentarity.

*Note G. L. De Broglie has adhered to this criticism, and it is in reference thereto

that he nowhere has taken the EPR paradox into account (Refs. 11,26, p. 77; 11,33,
p. 169). Curiously, he never appears interested in the version Bohm gave of the
problem [in Quantum Theory (Prentice-Hall, Englewood-Cliffs, New Jersey, 1951)],
which is the version nowadays cited by everyone.

7. COMPLEMENTARY REMARKS ON THE EINSTEIN-BOHR

CONTROVERSY

In 1949, on the occasion of Einstein's seventieth birthday, there appeared in

the United States a thick jubilee volume (Vol. 7 of the Library of Living Philoso-
phers, published by Open Court, La Salle, Illinois, 1949) dedicated to the founder
of relativity theory, which contained articles contributed by scholars from eleven
countries. In this book, a large number of physicists, such as Born, Pauli, Heitler,
etc., expressed, sometimes in lively fashion, their disappointment in seeing Einstein
persist in his negative attitude towards the current interpretation of quantum the-
ory.
The most interesting of the studies contained in this book is without doubt the
chapter due to Bohr (pp. 201-241). In these pages, Bohr sets out to explain how
his discussions with Einstein on the theory of quanta and his efforts at refuting the
very ingenious and very subtle arguments of his contradictor have led him to clarify
144 Some Difficult Points in Wave Mechanics

his point of view.

Bohr begins by recalling the history of the old quantum theory: He summarizes
the discovery by Planck of the existence of the quantum of action and its application
by Einstein to radiation (theory of light quanta, or "photons"). In regard to the
difficulties encountered by Einstein's photon hypothesis in explaining interferences
and diffraction, Bohr recalls that Einstein, from the time of his first papers on this
subject, had indeed strongly realized that the reconciliation of the discontinuous
structure of light with the wave theory of light required the introduction of proba-
bilities. And Bohr stresses that here the introduction of probabilities does not result
from our ignorance about a hidden mechanism but instead from the existence itself
of the quantum discontinuities. "In fact," writes Bohr, "in quantum physics we
are presented not with intricacies of this kind, but with the inability of the classical
frame of concepts to comprise the peculiar feature of indivisibility, or 'individuality,'
characterizing the elementary processes."
Bohr next summarizes his quantum conception of the atom and its confirmation
by experiments of the Franck and Hertz type on the excitation and ionization of
atoms by collisions with electrons. He emphasizes the fact that the concepts of
quantum states and the abrupt transitions between them, which lie at the basis of
his atomic theory, do not allow us to give any deterministic description of atomic
phenomena but only to introduce transition probabilities and from them to infer
the relative frequencies of observable processes. He recalls how he systematically
introduced the idea of transition probability in his studies on the correspondence
principle, in such a way that the global observable phenomena would be represented
statistically in an exact manner by the laws of classical mechanics. He also comments
on the major paper of 1917 wherein Einstein analyzed the interaction between an
atom and blackbody radiation in equilibrium thermodynamics, while introducing
absorption probabilities and the probabilities governing spontaneous and stimulated
emissions of radiation by atoms.
After having analyzed the discovery of the Compton effect, the development
of the formalisms of the new mechanics (wave mechanics and quantum mechan-
ics), their unification by Schrodinger, and the statement of the uncertainty re-
lations by Heisenberg, Bohr presented his ideas on "complementarity" which he
developed in September 1927 at the reunion of physicists held at Como in com-
memoration of the work of Volta. The concept of complementarity has as its goal,
he says, "to embrace the characteristic features of individuality of quantum phe-
nomena, and at the same time to clarify the peculiar aspects of the observational
problem in this field of experience." And he adds: "For this purpose, it is deci-
sive to recognize that, however far the phenomena transcend the scope of classi-
cal physical explanation, the account of all evidence must be expressed in classi-
cal terms." This requirement is indispensable in enabling us to describe and ex-
plain our experimental arrangements to others. The crucial point, says Bohr, is
then "the impossibility of any sharp separation between the behavior of atomic ob-
jects and the interaction with the measuring instruments which serve to define the
conditions under which the phenomena appear. In fact, the individuality of the
typical quantum effects finds its proper expression in the circumstance that any
attempt of subdividing the phenomena will demand a change in the experimen-
tal arrangement introducing new possibilities of interaction between objects and
 Complementary Remarks on the Einstein-Bohr Controversy 145

measuring instruments which in principle cannot be controlled. Consequently, ev-

idence obtained under different experimental conditions cannot be comprehended
within a single picture, but must be regarded as complementary in the sense that
only the totality of the phenomena exhausts the possible information about the
objects."
From this it follows that the partial images obtained in utilizing, for example,
the concept of particle and that of wave are incomplete and reciprocally limiting.
But the formalism of the new quantum theory allows us under all circumstances to
obtain an exact description of the observable facts in experimentally determinate
situations.
Bohr next analyzes the discussions that took place in Brussels on the occasion of
the Fifth Solvay Conference (October 1927). He discusses the objection of Einstein
that we presented earlier, in Sec. 5, as well as the question, raised further on in
the same section, of the semi-reflecting mirror. In the latter case, he underlines
the importance of clearly specifying the experimental arrangement: If only a semi-
reflecting mirror M is used, one will be able to localize the photon either in the
transmitted beam or the reflected beam; but if a second (perfect) mirror M' is
added to this experimental setup, one will be able to create interferences in the region
where the reflected and transmitted beams intersect. Bohr then gives the reasoning
by which he had proven that, if the usual device of Young's slits is modified in such
a way that one can say through which slit the photon has passed, the interference
phenomenon can be made to disappear.
Bohr goes on to emphasize the utility of representing in a very detailed and
almost naive manner ("in a semi-serious style") the experimental arrangement cor-
responding to the particular measurement one wants to perform, because only by
this method can the quantum formalism be applied in a manner that is unambiguous
and free of contradictions.
Continuing the history of his discussions with Einstein, Bohr explains how new
controversies arose between them on the occasion of the Sixth Solvay Conference
(Brussels, 1930). Einstein thought he had found a way to exactly measure the energy
E of a particle at the instant of its emission, contrary to the relation 6E6t 2:: h,
starting from the relativistic relation E = moc2 and the equality of inertial mass
and gravitational mass. Suppose, he said, that a certain amount of radiation is
enclosed in a box furnished with a hole in its side, which can be opened or closed
by a shutter. A clockwork mechanism could open the shutter at a precise moment
and thus allow a photon to escape from the box at a well-determined time. Now, by
carefully weighing the box before and after the release of the photon, it would be
possible also to find the exact energy of this photon by using the relation E = moc2 •
After a closely-argued discussion, Bohr succeeded in finding an answer to Einstein's
argument, starting from the remark that one must, at the time of weighing the box,
take into account the influence of the gravitational field on the rate of a clock, which
effect manifests itself in the red shift of radiation emitted by stellar surfaces.
We give here the argumentation of Bohr. Assume that one weighs the box by
suspending it from a spring, as shown in the figure below. However, the position of
the box along a vertical scale is known with an uncertainty /1q '" hi /1p, where /1p
denotes the uncertainty in the linear momentum of the box. But it is evident that
/1p must be less than the linear momentum acquired in total during the time T it
146 Some Difficult Points in Wave Mechanics

takes to weigh a body of mass t.m, where t.m denotes the mass lost by the box;
whence
h
t.p '" t.q < Tg t.m. (14)

Thus, for a given value of t.m, the more accurately the position q of the box
is determined, the longer the interval T must be. Now, according to the theory of
general relativity, when the clock contained in the mechanism that opens the shutter
of the box gets displaced vertically by an amount t.q, its rate will change in such a
way that at the end of the time T the clock reading will be off by an amount

(15)

(according to the expression for ds 2 in a gravitation field). On eliminating t.q

 Complementary Remarks on the Einstein-Bohr Controversy 147

between Eqs. (14) and (15), one finds

h
D.T > ~, or D.TD.E> h, (16)
c urn

in agreement with the fourth uncertainty relation. Here again a careful analysis
of the measuring device has removed a seeming difficulty. The procedure itself for
measuring the change in weight of the box at the moment of the photon's release,
which permits the determination of E with an uncertainty D.E, generates the un-
certainty D.T in the time of the photon's escape which is required by Heisenberg's
fourth uncertainty relation.
Bohr again remarks that the essential thing is always to consider the totality of
the experimental setup in such a way that a well-defined application of the formalism
of quantum mechanics will be obtained.
Bohr next writes about the EPR paper, which he earlier analyzed. He remarks
that, although [qi, Pi] =I 0, if qI and q2 are the coordinates of the two particles form-
ing a systcm and PI and P2 the conjugate momenta, one has [( qI -q2), (PI +P2)] = 0,
as can be verified immediately, since [% Pk] = 0 for i =I k. It follows that
qI - q2 and PI + P2 are simultaneously measurable and that accordingly noth-
ing prevents the prediction of the value of qI or of PI if one measures either q2
or P2, respectively. The situation is therefore exactly as in considering, for ex-
ample, the passage of a particle through a diaphragm: One has in principle, af-
ter the particle has traversed the diaphragm, the possibility of measuring either
the position or the linear momentum of the diaphragm and then in each case to
make predictions regarding later observations on the particle. But the essential
remark here is that such determinations require mutually exclusive experimental
arrangements. In studying this kind of problem, one must never forget that it
is useless to view a qucstion in the abstract: It must be remembered that ev-
ery measuring instrument is macroscopic and that a measuring device consisting
of macroscopic bodies always intervenes in every observable phenomenon. It is
the conditions imposed by the experimentalist on these macroscopic bodies that
fix the information which the measuring device can furnish about the atomic-
scale entities involved in the measuring procedures. The mathematical formalism
of quantum mechanics automatically covers all the measurement procedures one
can imagine, but in every measurement that is actually made, only one of these
procedures is realized. Let me moreover add that in discussions of this kind of
problem we are constantly hampered by our intuitive ideas about space and time,
even when they are amended by relativity theory. The existence of the uncertainty
relations by itself suffices to show that this spacetime frame, which is evidently
applicable to the description of measuring devices and to the statement of mea-
surement results, is not valid for the exact description of entities on the atomic
level.

Note G. L. De Broglie later placed a question mark in the margin in front of this
passage. In his personal code, this sign does not convey a question but a criticism,
as in chess commentaries. It is clear that he, having come back to spatio-temporal
representations, could no longer accept this conclusion.
148 Some Difficult Points in Wave Mechanics

Bohr ends his presentation by recalling his attempts to generalize the idea of
complementarity outside the domain of physics: I shall not stress this point. He
emphasizes the difficulty of finding terms in our language that are entirely adequate
for the expression of circumstances so far removed from our intuition as those here
encountered. Our language, being the product of our macroscopic experiences, is
very poorly suited for the task of expressing the subtle concepts that are required in
the interpretation of processes on the atomic level. It seemed to Bohr that phrases
such as "disturbing of phenomena by observation" or "creating physical attributes
of atomic objects by measurement" are justly apt to create false ideas. Even when
one speaks about the impossibility of simultaneously measuring the position and the
momentum of a particle, one risks giving the impression that the position and the
momentum exist before the measurement. In reality, every happening that can be
characterized as an "observable phenomenon" is linked to a well-defined set of ex-
perimental arrangements; and the theories applicable to entities on the atomic level
have as their goal only the creation of a connection between those phenomena that
are successively observed under these conditions, this connection being moreover of
a statistical nature. Every attempt to attribute physically objective characteristics
to entities on the atomic level must be abandoned.

Note G. L. It is hardly necessary to stress that de Broglie later forcefully con-

demned this statement. Such passages are interesting because they, having been
written less than two years before de Broglie's change of mind, testify to the force
with which he returned to and held on to his convictions, even if, deep down and
hidden from view, opposing tendencies were already at work, as other indications
lead us to suppose.

At the conclusion of the jubilee volume, Einstein replies to the criticism di-
rected at him in regard to his attitude towards the current interpretation of
quantum theory. He declares himself unable to admit that the function tP of
wave mechanics can give the complete description of the state of an atomic sys-
tem. For him the wave function provides the description not of an individual
system but of an ideal ensemble of identical systems. Einstein's essential ar-
gument is that it must be possible to obtain an image of reality that is inde-
pendent of the measurement procedure. He appears to be certain that, if one
admits the existence of an objective reality independent of measurement proce-
dures, then the viewpoint belonging to the current interpretation of quantum the-
ory must be abandoned. But it seems that one could justifiably respond to Ein-
stein that his point of view constitutes an a priori metaphysical hypothesis and
that he is more justified in setting up theoretical physics as before to establish,
with the aid of well-defined observational procedures, a connection between phe-
nomena that are actually observed. [Note G. L. Disapproving later of his own
response to Einstein, the author placed a question mark next to it in the mar-
gin.]
Einstein recognizes moreover that the present formalism of quantum mechanics
describes observable phenomena and the wave-particle duality in a perfect man-
ner. But, writes he, "I am, in fact, firmly convinced that the essentially statistical
character of contemporary quantum theory is solely to be ascribed to the fact that
 Complementary Remarks on the Einstein-Bohr Controversy 149

this theory operates with an incomplete description of physical systems." As an

example, Einstein studies Gamow's theory of radioactive disintegration, where one
represents the probability of an alpha decay of a radioactive nucleus by the fact that
the 'ljJ wave of the alpha particle is capable of progressively escaping from the nu-
cleus in the form of an expanding spherical wave that tunnels through the potential
barrier surrounding the nucleus.
This picture, says Einstein, is perfect if one simply intends to study the sta-
tistical properties of an ensemble of radioactive nuclei, but it fails to give a truly
complete description of an individual nucleus, because it does not specify a disinte-
gration time and since one must evidently suppose that each nucleus disintegrates
at a well-defined instant. Einstein thus gives a reply that without doubt charac-
terizes him as a partisan of the new interpretation of quantum physics. This reply
consists essentially of the remark that the disintegration time is not known a priori
and that it can be known only as a result of an observation, an observation which
would change the state of our knowledge of the system. Although this remark may
make it seem that Einstein injects pliilosophical ideas, the correctness of which he
cannot assume, into "physical reality," he recognizes that it can appear satisfac-
tory when only a system on the microscopic level, such as a radioactive nucleus, is
concerned.
However, he adds that one can, with Schrodinger, consider not an isolated ra-
dioactive nucleus but a system comprising, in addition to this nucleus, a macroscopic
measuring device, such as a Geiger counter, equipped with an automatic recording
mechanism. This mechanism may consist of a paper strip which is uncoild evenly
by a clockwork, and on which a mark is made every time the counter is triggered.
One is thus faced with a very complex system whose configuration space has a
large number of dimensions, but logically there is no objection to discussing it from
the quantum mechanical point of view. If one considers all possible configurations,
then, after a time that is long in relation to the decay time of the radioactive atom,
there will be at most one mark on the recording strip. To each configuration there
will correspond a definite position of the mark on the paper strip. But, since the
present theory gives only the relative probabilities of the conceivable configurations,
we can calculate only relative probabilities for the positions of the marks on the
recording strip. Now, says Einstein, the position of the mark on the strip is a fact
that belongs to macroscopic physics, which is not the case for the determination
of the moment of disintegration. Therefore, if we regard present quantum theory
as giving a complete description of the individual system, one is forced to admit
that the position of the mark on the paper strip is not something which belongs to
the system as such, but that the existence of this position depends essentially on
the execution of an observation made on the recording strip. Einstein admits that
such an interpretation is possible, but he views it as one that is highly improba-
ble.
[To examine in depth this new objection of Einstein, it would without doubt be
necessary to engage in a rather complicated analysis; and I have the impression that
it would be helpful to invoke the ideas of von Neumann on pure states and mixtures in
making the remarks on this subject which I hope to be able to develop in the coming
year. I moreover do not think that Einstein's objection contains anything essentially
new: Indeed, it always is a question of the microscopic entities whose manifestations
150 Some Difficult Points in Wave Mechanics
one observes with the aid of a macroscopic measuring device (here the Geiger counter
+ recording instrument), for which the usual spacetime concepts are satisfactory.
It seems that it always is the transition from a microscopic reality, for which the
concepts of space and time are no longer valid, to macroscopic appearances, which
we perceive in our frame of space and time, that introduces the probabilities and
the uncertainties: The present formalism of quantum theory expresses, in a manner
that appears satisfactory, those circumstances that poorly conform to our intuition,
which, originating in our sense perceptions, is unable without reference to the frame
of space and time to describe observable phenomena and, consequently, to state the
results of measurement.]

Note G. L. The author later crossed out the paragraph between brackets. He
did that without doubt when he made this objection of Einstein and Schrodinger
his own (Ref. II, 26).

Speaking of his article with Podolsky and Rosen, Einstein writes that Bohr's
reply is very clear. It signifies, according to him, that one has a choice between the
following two assertions:
(1) The description of a system by its wave 1jJ is complete.
(2) The real states of two spatially separated objects are independent of one
ariother.
The viewpoint of Bohr consists in making the first choice and rejecting the
second.
Einstein prefers to adopt the second postulate and regard the wave 1jJ as pro-
viding only a statistical description of an ensemble of systems occupying the same
state.
Against the standpoint of Einstein, I think one can argue that the definition of
the "spatial separation" of two systems is not a simple matter when the localizations
of the two systems is incomplete, and that one can there have localization of the
two systems in the same region of space. This is a difficult question which should
be gone into deeper.
One sees how very difficult these interpretational questions of current quantum
theory are: The greatest minds of our time have managed to hold opposing views
on this subject. It is therefore most useful to look at them from different angles.
This is what I shall try to do in the coming year in deepening the conceptions of
von Neumann on the role of measurement in the quantum formalism.
 ONTRE
PROBABILISTIC INTERPRETATION
OF WAVE MECHANICS
AND
VARIOUS RELATED QUESTIONS
 CHAPl'ER 11

SUMMARY OF
SOME GENERAL CONCEPTS
OF PROBABILITY CALCULUS

We intend in this chapter to give a non-rigorous and incomplete summary of a

number of notions commonly used in probability calculus, while referring to spe-
cial works those readers who would want to deepen their understanding of these
questions or are looking for more rigorous expositions.

1. PROBABILITY LAWS FOR ONE VARIABLE.

DISTRIBUTION FUNCTION

Let us consider a variable whose value is not exactly known but can perhaps
be determined by an experiment or observation (an operation called a "trial" by
statisticians). Without discussing here the nature of the probability notion, which
continues to be debated among specialists, it will be assumed that we understand the
meaning of the expression "the random variable X has a probability for possessing
a value less than x." The probability in question will be expressed by a function
F(x), the distribution function, which is zero for x = -00 and increases from 0 to
1 as x increases from -00 to +00.
The function F can increase abruptly for a certain value of x, if this value has a
finite probability for being realized, or increase continuously with x, if an infinitely
small probability p( x) dx is associated with the interval dx. In the first case, F( x) is
a step function, while in the second case it is a function with a continuous derivative.
Moreover, anyone of the two types of increase can occur in different segments of
the interval (-00, +00).
We may give some kind of physical picture of this variation by viewing F( x) as
the sum up to the abscissa x of mass elements distributed along the x axis in such
a manner that some of them are concentrated in points of the x axis, while others
are distributed continuously throughout certain segments of the axis. If F(x) is
continuous for all x, then

F(x) = l~ p(x) dx,

154 Some General Concepts of Probability Calculus

wherein the function p(x), called the probability density, can be studied in isolation
from F(x). But, at any points of the x axis where the probability density undergoes
an abrupt change, it is necessary to consider F(x)j one can then write, in the Stieltjes
integral notation
F(x) = [~ dF(x)j
herein dF(x) reduces to p(x) dx over segments where the distribution is continuous
but assumes a finite value in all points marked by a sudden increase of the probability
density.

Moments. In probability calculus, the mean values of positive integral powers of

x, that is, the quantities
mk = [:00 xk dF(x),
are called "moments." (One should not forget that

[:00 dF(x) = F(+oo) _ F(-oo) = 1.)

The moments used most frequently are the first two,

ml = 1-00+00 xdF(x), (1)

which in the continuous case reduce to

ml = 1-00+00 xp(x) dx, (2)

The first moment is the "mean value" of x or its "mathematical expectation."

One may also write
ml =x, m2 = x 2 , etc.

The name "deviation" is often given to the quantity x - ml, that is, to the
difference between the value of x under consideration and its mean value. The
 Probability Laws for One Variable. Distribution Function 155

standard deviation, or dispersion, is defined as the square root of the mean squared
deviation:

(3)

One then finds

a 2 = m2 + m 2I - 2 mImI = m2 - mI2 = 2x - (-)2

x , (4)

which is an often-used formula.

We should note that not all distribution laws define finite moments, because the
integrals furnishing the moments can diverge. This happens, for example, for the
so-called Cauchy's probability law

1 1
p( x) = ;- 1 + x2 ' (5)

whose dispersion and all moments are infinite.

Characteristic Function. The characteristic function, introduced by Laplace, is

nowadays used a great deal in probability calculus, in particular in statistical prob-
lems. It can be stated in various forms; we shall here adopt the following version:
The characteristic function 'P(u) corresponding to the distribution law F(x) is
given by the formula

'P(u) = j_=+OO exp (iux) dF(x)

(6)
= exp(iux) = mathematical expectation of exp(iux).

1:=
It assumes the form
'P(u) = exp (iux)p(x) dx, (7)

for a continuous distribution and the form

'P( u) = 2: Pn exp (iux n ) (8)

n

for any distribution that is completely discontinuous, Pn denoting the finite proba-
bility for the value x = x n .
In the case of a continuous distribution, the formula for the inversion of Fourier
integrals gives
p(x) = -1
211"
j+=
_= 'P(u)exp(-iux)du. (ga)
156 Some General Concepts of Probability Calculus

[In the general case, one obtains the formula

· j+N exp(-iuxo) -exp(-iux) ()d
x - F()
F() Xo = 11m . r.p u u, (9b)
N--+oo -N zu
of which the preceding is a special case. To see this, it suffices to note that

1 ) [exp (-iuxo) - exp (-iux)] r.p( u) =

( -:--
zu -00
l x

Xo
de j+oo exp [iu(c; - 0] dF(c;)
and then to introduce the new integration variable 'f/ = c; - e.]

Characteristic Function and Moments. Second Characteristic Function. The

characteristic function is closely linked to the moments. In fact, if r.p( u) can be
expanded in a MacLaurin series, one has
u 2
r.p(u) = r.p(0) + ur.pI (0) + 2" r.p
/I
(0) + ... + un n
n! r.p (0) + ... ; (10)
and, on expanding exp (iux) in the definition (6) of r.p( u ),-;mefinds

r.p(u) = l+iu j +OO xdF+--

-00
(iu)2j+00
2 -00
(iu)nj+oo
x dF(x)+ ... +-,- 2
n. -00
xndF(x)+ ... ,
(11)
whence the identification mn = xn = i-nr.pn(o) and, in particular,
iml = iX = r.p'(0), (12)
Accordingly, if the moments exist, one can calculate them starting from the
characteristic function. The inversion formulas (9) demonstrated above show that
knowing r.p(u) is equivalent to knowing F(x) (and thus p(x) in the continuous case).
A knowledge of the moments, when they exist, allowing us to obtain r.p( u) by using
MacLaurin's formula, we then see that knowing all the moments is equivalent to
knowing the probability distribution law.
In place of the characteristic function, one often employs a second characteristic
function which is the logarithm of the first, that is,

<p(u) = log r.p(u) = log 1: 00

The expansion of <p(u) in a MacLaurin series gives (since r.p(O) = 1)

exp (iux) dF(x). (13)

r.p'(0) u2 (r.p1l(0) r.p12(0))

<p(u) = log r.p(0) + u r.p(O) + 2" r.p(0) - r.p2(0) + ...
. (iu)2 2
= zuml + -2- (m2 - ml) + ... (14)

. (iu)2 2 u 2
2
= zuml + -2- a + ... =
.
zuml - 2" a + ....
 Probability Laws for One Variable. Distribution Function 157

The advantage of this series is that it features the dispersion 0-.

Examples

(1) Continuous Laplace-Gauss Law. This law is defined by the probability den-

1
sity
+00
p(x) = (21[")11/20- exp [_ (x 2~;)2], -00 p(x) dx = 1. (15)

The corresponding characteristic function is

(16)
0-2u2)
= exp (iux) exp ( --2- .

If x is taken as the coordinate origin (one then says the probability law is stated
with respect to its center of mass), the formula (16) simplifies to

(17)

whence m1 = 0 (naturally!) and m2 = 0- 2• The quantity 0- in the formula (15) for

p(x) thus is the dispersion corresponding to this law. The continuous Laplace-Gauss
law is the fundamental law of chance resulting from a multitude of irregular small
causes.

(2) Discontinuous Poisson Law. This law, which one encounters in a number of
applications of probability calculus, applies to a random variable X that is capable
of assuming only the non-negative integer values 0,1,2, ... , n, .. . , the probability of
the value x = n being
Pen) = ane-a
n!
Clearly, LPn = 1, as it should.
n
Here the characteristic function is
. an
c,o(u) = '"'" e,un e- a - ,
L...J
n
n!
that is,
c,o(u) = exp [a(e iu -1)]. (18)
Expansion of c,o( u) gives
. a(a + 1) 2
c,o( u) = 1 + zau - 2 u +"', (19)
158 Some General Concepts of Probability Calculus

whence
m2 = 0:(0: + 1). (20)
One can easily verify these values by writing

00 n 00 n-l
ml =n= 2:: e- a ;
n.
n = o:e- a 2:: (n
0:_ 1 )1 = 0:,
.
n=O n=l

00 n 00 n-l n (21)
2:: e- 2:: (n0:_ 1.)1 n = o:e- a 2::;
00

m2 = n2 = a ; n 2 = o:e- a (n + 1),
n. n.
n=O n=l n=O

We therefore get
0'2 = m2 -my = 0:(0:+ 1) _0: 2 = 0:,
a formula which one could also readily obtain on starting with the second charac-
teristic function

. (il1?
<1>(11) = log <,0(11) = o:(e'U -1) = iO:l1 + -2-0: + ....

It is seen that the mean value x equals the dispersion (around this mean value) and
that both are equal to the constant 0: of Poisson's law.

(3) Cauchy's Law. This is another frequently-encountered law, which is defined

by the probability density
1 1
p(x) - - - -2
- 7r 1 + x .

We have already remarked that the moments corresponding to this law are infinite.
The characteristic function

() 11+00 exp (il1X) d

<,011=- X (22)
7r-00 1 + x 2

can be calculated by application of the calculus of residues. One finds

<,0(11) = exp (-1111). (23)

This function, being nonanalytic, cannot be expanded in a MacLaurin series in the

vicinity of the origin, which fact implies the non-existence of the moments.
One could naturally study an infinity of other probability laws; we shall not
pursue the matter here.
 Probability Laws for Two Variables 159
2. PROBABILITY LAWS FOR TWO VARIABLES

We now consider two random variables X and Y and suppose (which is essen-
tial for the future application of our considerations to wave mechanics) that the
measurement of anyone of these variables will in no way disturb the value of the
other and that it is possible to measure both variables simultaneously. One can then
define a distribution function F(x, y) such that the probability of a simultaneous
measurement of X and Y (or two consecutive measurements, one of X and the other
of Y, carried out in any order) producing values that are at most equal to X = x
and Y = y is given by F(x, y). As before, we must have

F( -00, y) = F(x, -00) = F( -00, -00) = 0

and F(+oo, +00) = 1; F must always increase with x and with y. Here again this
increase can take place continuously or discontinuously. In the plane of the variables
x and y, the function F(x,y) can be viewed as representing the sum of appropriate
masses that are distributed over the plane and have coordinates less than x and y,
respectively; these masses may be concentrated in certain points or along particular
lines of the plane, or they may be distributed continuously over some areas of the
plane. If F(x, y) is continuous everywhere, it is possible to introduce a "probability
density" p(x,y) such that

F(x,y) = loX dx loY dyp(x,y), (24)

and everything can be expressed in terms of this p(x,y). By contrast, if points

or lines of discontinuity of F exist in the xy plane, one must write the Stieltjes
integral
F(x,y) = JJdF(x,y),
wherein dF(x,y) reduces to p(x,y)dxdy in regions where the distribution is con-
tinuous but has a finite value at points or along lines of discontinuity of F(x, y).

Moments. One can here again define moments, but it is necessary in the notation
to take account of the existence of two variables. We write

In particular, mx = X, my = y, m x2 = x 2 , etc.
In the continuous case, one will have
160 Some General Concepts of Probability Calculus
Characteristic Function. The definition of the characteristic function is an im-
mediate generalization of the definition valid for one variable:

cp(u, v) = 11 exp[i(ux+vy)]dF(x,y), (26)

which in the purely continuous case reduces to

cp(u, v) = 11 exp[i(ux + vy)]p(x,y)dxdy

and in the purely discontinuous case assumes the form

cp( u, v) = L Pnm exp [i( uX n + VYm)],

n,m

P nm denoting the probability of the value pair X = X n , Y = Ym.

Inversion formulas again exist; here they read

p(x,y) = -1
2
47r
J1+ -00
00
exp [-i(ux + vy)] cp(u, v) du dv, (27)

for the purely continuous case (Fourier inversion formula), and

F( X,y ) _ rr;,( xo,YO ) -_ 1·1m j+U j+V exp (-iuxo) .- exp (-iux)
~:: -u -V zu
(28)
exp (-ivyo) - exp (-ivy) ( )d d
x . cp U,V U v
tv

in the general case. These formulas show that knowledge of the characteristic func-
tion is equivalent to knowledge of the distribution function.
As in the case of a single variable, one can find the moments, if they exist, that
is, provided the integrals defining them are convergent, with the help of a function
cp( u, v) whose expansion in a MacLaurin series is given:

Indeed, comparison of (29) with the expansion

·2
cp(u, v) = 1 + i(mxu + myv) + 22 (mx2u 2 + 2mxyuv + my2v2) + ... , (30)

resulting from the definition (26) of cp( u, v), allows the identification

imx = cp~(O,O), imy = cp~(O,O), -m x2 = CP~2(0,0), -mxy = cp~v(O,O),

 Probability Laws for Two Variables 161
and, in general,
(31)
These equations express the moments in terms of the derivatives of the function rpj
accordingly, if these moments exist, knowing them is tantamount to knowing rp( u, v)
and, therefore, the probability distribution.
One can obviously define standard deviations 17, or dispersions, for the variables
X and Y such that
(32)

<p( u, v) = log rp( u, v) = log

Its expansion in a MacLaurin series gives

1:
Let us now introduce the second characteristic function:

J 00
exp [i( ux + vy)) dF(x, y). (33)

·2
<p(u, v) = i(um x + vmy) +~ «(T;U 2 + 2r(Tx(Tyuv + (T;v2) + ... , (34)

where the coefficient mxy - mxmy of 2uv has been written as r(Tx(Ty. In this way
the "correlation coefficient" r is introduced as

(35)

We shall see that this coefficient is connected with the degree of independence
of X and Y. Computation reveals that the mean value of the quantity [(x - x)
+A(y - y))2 is
17; + 2A (x - x)(y - y) + A2(T;.
By its definition, this quadratic expression in A is obviously nonnegative, which
means that its discriminant, that is,

must be negative or zero. But

(x - x)(y - y) = xy - xy = mxy - mxmy = r(Tx O'y, (36)

with the result that (r2 - 1) 17; (T~ ~ 0, whence Irl ~ Ij the correlation coefficient
thus is in absolute value less than or at most equal to unity.

Marginal and Conditional Distribution Laws. If the value of X alone is of inter-

est, one is led to introduce the distribution law

Fx(x) = 1 +00

-00 dyF(x, y),

162 Some General Concepts of Probability Calculus

the integration being carried out over the line x = const. Similarly, if one is inter-
ested only in the value of Y, the distribution law

Fy(y) = 1 +00

-00 dxF(x, v),

is suggested, the integration now extending over the line y = const.

The functions Fx(x) and Fy(y) define "marginal distribution laws." The corre-
sponding characteristic functions are

<px(u) = 1 +00
-00 exp (iux) dFx(x) =
jrf+oo
J-oo exp(iux)dF(x,y) = <p(u, 0),
(37)
<Py(V) = <p(O,v).
In the case of purely continuous distributions, where dF( x, y) = p( x, y) dx dy,
one is thus led to define the marginal probability densities

px(x) = 1 +00

-00 p(x,y)dy, Py(y) = 1 +00

-00 p(x,y)dx. (38)

For purely discontinuous distributions, one defines the marginal probabilities

Px(xn) = L P(Xn, Ym), (39)

m n

In the case of independent variables X and Y, the theorem of compound

probabilities (the proof of which I assume to be known) allows one to write
F(x,y} = Fx(x)Fy(y), or
p(x, y) = px(x)Py(y), P(xn' Ym) = Px(xn)pY(Ym}
for the purely continuous case and the purely discontinuous case, respectively. But in
the general situation when X and Yare not independent, the theorem of compound
probabilities leads one to write

dF(x,y) = dFx(x) dF}X) (x, y) and dF(x,y) = dFy(y) dF¥)(x,y), (40)

where dF}X)(x,y) defines the probability law for Y when X is known to have the
value x and dFSP
(x, y) defines the probability law for X when Y is known to have
the value y. These functions establish, respectively, the distribution law for Y, given
X, and the distribution law for X, given Y; alternatively they are said to define the
"conditional probabilities."
In the purely continuous case, the conditional probabilities are represented by
the probability densities /{) (x) and P?,") (y) such that
(X) (Y)
p(x,y) = px(x)py (x,y), p(x,y) = Py(y)px (x,y).
 Probability Laws for Two Variables 163

In the purely discontinuous situation, one introduces the probabilities pf)(x n )

and p~X)(Ym) such that

P(Xn,Ym) = PX(Xn)p~X)(Xn'Ym) and P(xn,Ym) = PY(Ym)pf)(xn,Ym).

The introduction of conditional probability distributions naturally allows us to
define moments and a dispersion corresponding to these laws. One can, for example,
write

(41)
(of)r = £:00 (Y _'y{X) r pC;)(x, y) dy = g(x),
thus obtaining quantities dependent on x.
The functions f( x) and g( x) define, respectively, the "regression curve" of Y for
given X and the dispersion equation of Y for given X. The liaison by X is called
"homostochastic" if 9 is independent of x (i.e., dispersion (T~X) = const.) Analogous
definitions obtain when the roles of X and Yare interchanged.
We have seen that, from the standpoint of probability calculus, independence
(stochastic independence) of X and Y is characterized by the formulas

F(x, y) = Fx{x)Fy{y), p(x, y) = Px{x)py(y), P(xn, Ym) = Px(xn)pY(Ym).

(42)
One can easily verify that the conditional probability laws then are identical with
the marginal probability laws [Ff)(x) = Fx(x), etc.]. The characteristic function
of two stochastically independent variables equals the product of their marginal
characteristic functions, because

if'(u,v) = J£~exp[i{uX+VY)]dF(X,y)

= 1-00+00
exp (iux) dFx(x)
[+00
Loo exp (ivy) dFy(y)

Conversely, if the characteristic function equals the product of two functions,

each one of which depends on only one of two variables, then the two variables are
stochastically independent. This result follows from an application of the inversion
formulas that allow the computation of probability laws from a knowledge of the
corresponding characteristic functions.
If the variables x and yare stochastically independent, the mean value mxy
of xy obviously equals the product mxmy. According to its definition (35), the
164 Some General Concepts of Probability Calculus
correlation coefficient r then is zero. The vanishing of r therefore is a necessary
condition for stochastic independence. It is however not a sufficient condition:
The coefficient r can be zero even when the variables are not stochastically in-
dependent. Consider, for example, two random variables X and Y that are con-
nected with one another by the completely strict relation Y = X2. To represent
this connection, we introduce the Dirac function <5(z), which is defined to be zero
everywhere except at z = 0 and to become infinite at z = 0 in such a manner
that
1: b
f(z) <5(z) dz = f(O)
if a and b are any two positive numbers. It has been difficult to justify the use
of the Dirac function in a mathematically rigorous manner,l but in practice its
application causes no error. [Note G. L. This text was written at a time when the
theory of distributions was being developed. Without doubt, de Broglie ignored
this theory, but later he added in pencil: "1. Schwartz."] Therefore, employing
~he Dirac function, we can say that the probability density in the present case

11:
IS

p(x,y) = <5(y -x 2)p(x), such that

00
pdxdy = 1,
because

pf{)(x,y) = <5(y-x 2) and 1f:

Suppose that p(x) is an even function of
p(x)<5(y-x2)dxdy=

Xj one then has

1: 00
p(x)dx=1.

mxy=xy= 11: 00
xyp(x)<5(y-x2)dxdy= 1: 00
x 3 p(x)dx=0, (43)

mx = x= J"[+oo
1-00 xp(x) <5(y - x 2) dx dy =
1+00
-00 x p(x) dx = 0, (44)

Jr[+oo
my=fj= 1-00 yp(x)<5(y-x 2)dxdy= -00 x 2 p(x)dx:f0.
1+00
(45)

Consequently, m xy - mxmy = 0 and r = O. Therefore, if Y = X2 and p(x) is

even, the correlation coefficient is zero, despite the strict connection existing between
Y and X.

Note L. B. The correlation coefficient r of two random variables can therefore

be zero even if they depend on one another. By contrast, the two variables must be
dependent if r is nonzero.

Example: Gauss's Law for Two Variables. We have seen that the second char-
acteristic function <J> ( u, v) has the MacLaurin expansion
·2
<J>(u,v) = i(um x + vmy) + t2 (0"1u 2 + 2rO"xO"yuv + 0"}v 2 ) + ... , (46)
 Probability Laws for Two Variables 165
r being the correlation coefficient. If, by an appropriate choice of the origin for
x and y, the probability law is referred to its center of mass, Eq. (46) reduces
to
<I> ( u, v) = -~ (O"iu 2 + 2rO"xO"yuv + 0"}v 2) + ....
The simplest situation obtains when all terms of order higher than the second in
this series are zero. <I> then becomes a homogeneous function of the second degree
in u and v:

<I>(u, v) = _~(O"iu2+2rO"xO"yuv+0"}v2) (47)

and, consequently,

r,o( u, v) = exp ( -~ (O"iu 2 + 2rO"xO"yuv + 0"}v 2)). (48)

By analogy with the case of one variable, where it was shown [Eq.(17)] that
<I>(u) has the form _(0"2/2)u 2 for Gauss's law, it is clear that we are now faced with
Gauss's law (the normal law) for two variables.
Use of the inversion formula generates the probability density corresponding to
the characteristic function (48):

47r2p(x,y) = 1[:00 r,o(u,v) exp (-i(ux + vy))dudv.

A simple calculation based on the formula

gives
p(x,y)

= [27rO"xO"y(1 - r 2)r 1 exp [

(x/O"X)2 - 2r(x/O"x)(Y/O"y) + (y/O"y?] (49)
2(1 - r2) .
With the aid of this result, it is easy to deduce the conditional distribution laws,
which are defined by the formulas
(X)( ) _ p(x,y) _
Py x,Y - P (x) -1+00p(x,y) '
X
-00 p(x,y) dy

(Y)
Px (x,y) = ....
One easily finds

(X)( ) _ [(2 )1/2 (1 _ 2)1/2]-1 [_ ((y/O"y) - r(x/O"x))2] (50)

Py x,y - 7r O"y r exp 2(1 _ r2) .
166 Some General Concepts of Probability Calculus

If r = 0, the distribution pV")

(x, y) is Gaussian with respect to the variable y,
p(x,y)has the form p(x)p(y), and stochastic independence obtains.
The conditional mean value is

(51)

and one readily gets

y(X) = f(x) = rO'y(~), (52)

so that here the regression curve is simply a straight line.

Similarly,

or
O'~X) = ~ O' y = g(x).
Thus g(x) is constant; the liaison is homostochastic. If r = 0, O'~X) = 0'1;'
From these results, one sees that a zero value of the correlation coefficIent implies
stochastic independence, but this outcome is peculiar to Gauss's law.

Note L. B. These results can otherwise be obtained without calculation by

noticing that pV")
(y) has the form of a Gaussian law with standard deviation
a y(l - r 2)1/2.

One could study probability distributions for several variables, which would lead
to generalizations of the preceding results; but this will not be done here.

3. VERY IMPORTANT REMARK CONCERNING

THE FOREGOING RESULTS

In the foregoing we have assumed the existence, for two variables, of a distribu-
tion function F(x,y) or, in the continuous case, a probability density p(x,y), and
from that we inferred a whole set of consequences. All these conclusions implicitly
presuppose that the measurement of anyone of the quantities X and Y does not
modify the state of affairs existing before the measurement, which is a simple noting
of the value of the variable in question. It is important to give a more thorough dis-
cussion of this point, which plays a major role in wave mechanics. For this purpose,
we are going to study a continuous distribution; the extension of our arguments to
the purely discontinuous case and the general case will be straightforward.
To make our reasoning concrete, suppose one wants to investigate whether any
correlation exists between the height and the chest size of individuals. As random
variables X and Y one may now take the height and chest size, respectively, and
 Very Important Remark Concerning the Foregoing Results 167
suppose that these variables are measured on a large number of individuals, military
draftees, for example. It is then possible to define a probability density p(x,y) such
that the probability for the height lying between x and x + dx and the chest size
between y and y + dy will be

p(x,y)dxdy, with 11: pdxdy = 1, of course.

But here it is obvious that the measurement has the character of a simple recording,
because it is unimaginable that measurement of a draftee's height would modify his
chest size or the height or chest size of any other draftee. Consequently,

px(x) = j +OO p(x,y)dy,

-00 Py(y) = j +OO p(x,y)dx
-00 (53)

obviously represent the probability densities for the height and the chest size con-
sidered separately. The formulas

(Y)( )_p(x,y) (X)( ) _ p(x, y)

Px x,y - py ()' Py x, Y - () (54)
y Px x
define the conditional probabilities as having the following meaning: Among the
draftees with chest size y, the probability that the height will lie between x
and x + dx is given by lJ)(x, y) dx; and an analogous interpretation holds for
(X)
Py (x,y).
But let us imagine, however implausible the idea, that the measurement of a
draftee's height will cause his chest size to vary, and conversely; then the preceding
formalism would lose its validity. The simultaneous measurement of X and Y would
become impossible by reason of the variation of one quantity when the other is
being measured. One could obviously measure the two quantities in succession,
but then the order in which the two measurements are performed would become
important. One could, in measuring only X, define a probability density px(x)
and, in measuring only Y, define a probability density py(y). However, if one
were to measure Y only after having measured X, one would be led to define the
probability density

I
py(y) =
j+oo px(x)py
-00
(X)
(x,y)dx, (55a)

where p?,")(x,y) is the probability density of Y for the category of draftees ex-
hibiting X = x in the first measurement. This formula is demanded by the the-
orems on total and compound probabilities; but nothing guarantees that p'y(y)
will equal Py(y): There is no reason why the probability distribution for Y after
measurement of X should equal the probability distribution corresponding to the
direct measurement of Y, since the measurement of X no longer amounts to a mere
recording but entirely disturbs the state of the probabilities. Similar remarks ap-
ply if one measures first Y and then X; the probability of anyone value of X is
168 Some General Concepts of Probability Calculus

then given, according to the general theorems on total and compound probabilities,
by
I
Px(x)
roo (Y)
= 1-00 Py(y)px (x,y)dy, (55b)

where /{)(x,y) is the probability density for the height within the category of
draftees for whom the first measurement, that of the chest size, produced the value
y = y. But there is no reason why p~(x) should equal Px(x).
Can a function p(x,y) exist under these circumstances? Such a function can at
any rate no longer have the meaning that a simultaneous measurement of X and Y
will with probability p(x, y) dx dy furnish a value of X lying between x and x + dx
and a value Y lying between y and y + dy, since a simultaneous measurement of
X and Y is impossible. By contrast, nothing prevents the existence of a function
p(x,y) such that
[+00
Loo roo
p(x,y)dx = py(y), Loo p(x,y)dy = Px(x), (56)

Py(y) and Px(x) being the probability densities corresponding to the directly per-
formed measurements of X and Y. But the probability densities associated with
the measurement of Y performed after that of X and with the measurement of X
performed after that of Yare p'y(y) and p~(x), respectively, and will be different
from py(y) and Px(x). The probabilities pf{)(x,y) and /{)(x,y), which appear in
the formulas (55) for p'y(y) and p~(x) are thus no longer given by Eq. (54), because
if they were, one would have

px(x) = p~(x), Py(y) = p'y(y), (57)

while in reality there is now no reason for these equalities to exist, since the mea-
surements of X and Y no longer are mere recordings but instead operations that
disturb the states of probability. As a result, even if one can in the present case
find a function p(x, y) obeying Eq. (56), this function no longer has the well-defined
meaning of the preceding theory, and, in particular, it does not any more permit
the calculation of the functions

pX(Y)( x, y,
) (X)( x, y) ,
py p'y(y).
But, one might say, what interest is there in considering this strange situation
where the measurement of a draftee's height would modify his chest size? In ordi-
nary practice, that is, for all probability problems on the macroscopic level, such
a situation apparently never occurs. To take a different example, in the classi-
cal problem where a certain number of gold and silver coins are distributed over
two drawers, the hypothesis we have just examined amounts to assuming that the
opening of one drawer changes the distribution of the coins over the two draw-
ers, and that again is unimaginable. However, what appears inconceivable for the
operations of macroscopic physics (to which in general statistical calculations are
 Very Important Remark Concerning the Foregoing Results 169
applied) in microphysics becomes the rule when the variables X and Yare (non-
commuting) canonical variables; and then one must guard against introducing a
function p(x,y) and assuming implicitly that the measurement is a simple record-
ing.

Note C. L. The author here returns to an analysis which already appeared in his
article on "La statistique des cas pur" (Refs. V, 44; III, 9). It is in developing this
theme that he arrived at the idea, quite decisive for the theories of hidden variables,
that the statistical distributions for such variables (which must necessarily obey the
classical statistical schema) must themselves be hidden and thus cannot be identical
with the statistical distributions which in quantum mechanics govern the results of
measurement. One sees this idea emerge in the mentioned work at the end of a note
on the theorem of von Neumann, where it still appears only beneath the surface, as
we shall point out in due course. The author later amply developed his idea in a
book entitled La theorie de la mesure (Ref. II, 27; see also Ref. II, 33).
 CHAPTER 12

RECALLING
THE GENERAL CONCEPTS
OF WAVE MECHANICS

Notes of the editors (S. D., D. F., and G. L. The author would certainly have
rewritten this chapter, which contains numerous repetitions of aspects dealt with in
the first part of the present book because the corresponding lectures were delivered
in separate years. We have taken it upon ourselves to suppress certain passages:
pp. 34-42 (of the manuscript), which would have repeated the chapter "General For-
malism of Wave Mechanics"; pp. 42-47, which would have taken us back to "General
Principles of the Probabilistic Interpretation of Wave Mechanics"; pp. 48 and 49,
which would have summarized the uncertainty relations; pp. 50-59 which would
have duplicated the section "The Algebraic Matrices and their Properties"; pp. 60-
64, which would again have given the "Precise Form of the Uncertainty Relations"
and the "Angular Momentum in Wave Mechanics." We have almost completely
suppressed pp. 65-70, which would have restated the chapter "Wave Mechanics of
a System of Particles," but we have extracted from it a remark concerning config-
uration space (p. 69 of the manuscript), which we shall reintroduce further on in
the form of a note, while specifying its origin. Finally, the deleted pp. 78 and 79
would have returned us to the commutation problem for operators, which was al-
ready discussed above. What we have retained of the original chapter is not free
of repetitions, but it complements the first part of the present book and especially
traces the evolution of the author's thoughts on the subjects there introduced.

Wave mechanics started from the notion that it was necessary to associate with
each particle a wave represented by a certain function 1jJ(x, y, z, t). In other words,
to the particle idea it added the concept of a "field," as this word is understood in
the physics of fields, the field 1jJ.
The function 1jJ must satisfy a certain propagation equation, which in a sense
has to replace the equations of classical mechanics. We shall occupy ourselves here
only with the nonrelativistic form of wave mechanics; the generalization of the ideas
we shall encounter to the relativistic case can be made by introducing the theory of
Dirac and analogous theories for particles of spin exceeding 1/2. This matter is left
aside for the moment.
Consider a particle of mass m moving in a field of force derivable from a potential
energy function U(x, y, z, t). Let p be the linear momentum of the particle and E
 Recalling the General Concepts of Wave Mechanics 171

its energy, defined as the sum of its kinetic and potential energies:
1 1
E = "2mv
2
+ U(X,y,z,t) = 2m
2
(px + Py2 + pz)
2
+ U(x,y,z,t). (1)

By definition, the name Hamiltonian is given to the function H(x, y, Z,Px,Py,Pz, t)

that is obtained when the energy E is expressed as a function of the coordinates,
the components of the linear momentum, and the timej here, therefore,
1 2
H(x, y, Z,Px,Py,Pz, t) = 2m (Px + Py2 + pz)
2
+ U(x, y, z, t). (2)
The development of wave mechanics has shown that one can, starting from
the Hamiltonian function H, obtain the propagation equation for the '¢ waves by
the following procedure: First, replace in H every momentum pq by the operator
-(hj27ri)(8j8q) and every coordinate q by the operator qXj this gives rise to an
operator
Hop(x,y,z,t, -2h . 88 '--2
h . 88 '-2 h . 88)
7rZ x 7rZ Y 7rZ Z
called the "Hamiltonian operator" or simply the "Hamiltonian." One then writes
h 8'¢
27ri 8t = Hop'¢,
where '¢ is the wave function, and thereby obtains the propagation equation of the
wave '¢ for the particle under consideration.
Now, since
h2 82 82 82
Hop = - 87r2 m (8x 2 + 8y2 + 8z2 ) + U(x, y, z, t), (3)

the preceding equation can be written more explicitly to give the wave equation in
either one of the forms

(4a)

(4b)
This partial differential equation, being of the first order with respect to the time,
allows one in principle to calculate the form of the wave '¢(x, y, z, t) at any instant
t provided its form at an initial moment to is known.
The coefficients of the wave equation are complex. The function '¢ is therefore
essentially complex and thus cannot represent the vibration of a real medium, as is
always assumed to be the case for wave functions in classical physics.
From now on we shall always designate the complex conjugate of F by F*. The
wave '¢* then naturally obeys the propagation equation

l:!..'¢* _ 87r 2m U'¢* = _ 47rim 8'¢* (5)

h2 h 8t
172 Recalling the General Concepts of Wave Mechanics

conjugate to that of'1f;.

Let us, by definition, set

f = -~ ('1f;*\1'1f; - '1f;\1'1f;*).
4nm
These are obviously real quantities (because f * = f). From the propagation equa-
tions (4) and (5), one concludes

(6)
This equation has the form of the classical "continuity equation" for a fluid which
in each point has the density p and a flux density pv equal to f. It asserts that the
amount of fluid is conserved in the course of time and that therefore the integral

k k pdr = 1'1f;1 2 dr,

D being the finite or infinite domain occupied by the wave '1f;, is constant in time.
Because the function '1f;, a solution of a linear equation, is defined only up to a
multiplicative constant, one may choose this factor in such a manner that, at all
times,
(7)
One then says that the function '1f; is "normalized." We shall assume all wave
functions to be normalized, which hypothesis will justify the physical interpretation
given later on of the quantity 1'1f;1 2 • Let us moreover note that the function '1f;, even
when normalized, still contains an arbitrary factor exp (ia) of unit modulus.
It can be demonstrated that wave mechanics recovers the old mechanics in the
first, geometrical-optics, approximation. We shall not here dwell on this otherwise
fundamental point, which we developed at length in other places.
A particularly important case occurs when the function U does not depend on
the time (a permanent field). The equation for '1f; then allows "monochromatic"
solutions, which involve the time only through an exponential factor of the form
exp(27ri/h)Et. Such a monochromatic solution satisfies the equation
87r m
2
f:.'1f; + --;;:r- [E - U(x,y,z)]'1f; = O. (8)
In the special case U = 0 (no field present), the wave equation possesses the
plane-wave solutions

'1f;(x,y,z,t) = Aexp ( h ~ (ax

27ri [Et - v2mE + {3y + /Z)] ) , (9)

where A is a complex constant and where a, {3, ,,{, with squares adding up to unity, are
the direction cosines defining the direction of propagation. This solution represents
a plane wave, of frequency v = E / h and wavelength

A= h =~=~
(2mE)1/2 p mv'
 The Interference Principle. Theory of the Pilot Wave 173
propagating in the direction (a,p,,). Quantum mechanics originated in the study
of this kind of wave, which is associated with uniform rectilinear motion, in the
absence of a field, of a particle possessing a mass m, an energy E, and a linear
momentump.

1. THE INTERFERENCE PRINCIPLE. THEORY OF THE PILOT

WAVE

When wave mechanics started to develop, one of the first facts recognized by
theoretical physicists was the necessity to view the quantity 'IjJ'IjJ* = 1'ljJ12 as a measure
of the probability that the particle will be present at a point. This principle of
interference, or principle of localization, is moreover only a generalization to wave
mechanics of a principle that is always admitted in the wave theories of light, because
in the latter theories one always assumes that the energy density at any point is
given by the wave intensity obtaining there, this intensity per definition being the
square of the amplitude. Now it is easy, in the case of plane monochromatic waves
or in the case of superimposed plane waves of the same frequency, to account for
1'ljJ12 precisely equaling the intensity: If waves are regarded as formed by photons,
one is led to say that in a light wave the density of photons in every point of space
is proportional to 1'IjJ 12 , the function 'IjJ being the light variable written in a complex
form; and thus it is seen that the interference principle of wave mechanics is only an
immediate generalization to the case of arbitrary particles (electrons, for example)
of the principle valid for photons of light.
Moreover, particle diffraction phenomena (diffraction of electrons by a crystal,
Davisson and Germer, 1927; diffraction of protons and other particles by a crystal,
Stern and Estermann; electron diffraction at a screen edge, Borsch, 1940; neutron
diffraction, Amaldi, etc., 1945-47) can only be interpreted by assuming the inter-
ference principle, which is thereby experimentally verified.

Note G. L. C. J. Davisson and L. H. Germer, Phys. Rev. 30, 705 {1927};

I. Estermann and O. Stern, Z. Phys. 61, 95 (1930); H. Borsch, Naturwiss. 28, 709
(1940). Let us add the more recent findings of J. Faget, Revue d'Optique 40, 347
(1961) and C. Jonsson, Z. Phys. 161,454 (1961). One may consult on this subject
the collective work Fifty Years of Electron Diffraction, P. Goodman, ed. (Reidel,
Dordrecht, 1981).

But there is more. Our argument above assumes that we are dealing with a wave,
a wave of light or the wave 'IjJ, with which is associated a large number of particles,
photons or material particles; this allows us to speak of the density of particles.
But very important experiments, those of Taylor and of Dempster and Batho, prove
that the interference of light always have the same appearance, whether one employs
intense light and a short time of exposure or very weak light and a long time of expo-
sure. [Note G. L. G. J. Taylor, Pmc. Camb. Phil. Soc. 15,2 (1908); A. J. Dempster
and H. F. Batho, Phys. Rev. 30,644 (1927).] Although analogous experiments have
not until now been done with electrons or other material particles, there seems to
be no doubt that they would give the same result. [Note G. L. In fact, the ex-
174 Recalling the General Concepts of Wave Mechanics

periment came to be done, with the result predicted by de Broglie: 1. Biderman,

N. Sowchkine, and V. Fabrikant, Dokl. Akad. Nauk. SSSR 66, 165 (1949).] In such
experiments, the photons arrive only one by one at the interference apparatus, and
thus it must be imagined that a light wave train is associated with each photon.
One can therefore here no longer speak of a photon density belonging to the inci-
dent wave; and, to retain the interference principle and the interpretations resulting
therefrom, it becomes necessary to assume that, in the wave train associated with
a photon, the quantity I~ 12 represents the probability that a photon will, by a local
action on matter, manifest its presence at the corresponding point of space.
Experiments of the Taylor type force us to assume, as a generalization, that
an extended wave train is associated with every particle and that 1~12 gives the
probability of experimentally observing this particle at the space point under con-
sideration. Since the wave train is in general extended and I~ 12 generally is different
from zero in the entire region occupied by the wave train at a given instant, we
conclude that the position of the particle in the wave is in some way a priori un-
determined and that one must simply attribute a probability to everyone of the
possible positions.
Then arises a question which is of great general and philosophical importance: Is
this indeterminism only an apparent one, due to our ignorance of the real trajectory
of the particle, or is it on the contrary a real indeterminism, the particle, so to
speak, assuming a position in space only at the moment when, as a result of an
interaction, it manifests its presence by an observable action? The first point of view
conforms with the old ideas of physics, probability appearing only as a consequence
of our ignorance about the exact determinism governing phenomena; by contrast, the
second view, which implies a limitation on the deterministic nature of phenomena,
is one which quite rapidly forced itself upon physicists and nowadays seems to be
well established. [Note G. L. The author later put a question mark in the margin
next to this assertion.]
At the time (1926-27) when this crucial question was being debated, people nat-
urally tried to maintain classical ideas by imagining that particles have trajectories
such that the meaning of 1~12 as a "position probability" resulted therefrom. An
attempt of this kind, outlined by Madelung and called the hydrodynamical theory
of wave mechanics, was developed by myself at that time under the name "theory of
the pilot wave. [Note G. L. The author later added in pencil after the words "pilot
wave": "To develop. Hypothesis of the double solution."]
We have seen that one can associate with the wave ~ = a exp (icp) a fictitious
fluid of density p = 1~12 and flux density

f = pv -h . (~*'V~ -
= -4nm ~'V~*)

and that this fluid is conserved. One may then, in the usual hydro dynamical man-
ner, define the trajectories of the molecules making up this imaginary fluid by the
equation

(10)
 The Interference Principle. Theory of the Pilot Wave 175
which gives the flow velocity at every point. In the case of a monochromatic wave
(not necessarily plane), corresponding to a permanent field,

27ri
'I/J cxexp ( TEt ) ,

which makes v independent of the time. In this situation the trajectories of the
fluid molecules will coincide with the flow lines (lines which at any given instant are
in everyone of their points tangential to the local vector v ), a result well known in
hydrodynamics. It is therefore proper to suppose that, if a large number of particles
is associated with a wave 'I/J, these particles will describe the hydrodynamical tra-
jectories we have just specified. It would immediately follow that i'I/Ji 2 is the density
of the particles in the wave and that the relation

expresses the conservation of the number of particles during the propagation of the
wave.

Note L. B. In its motion, the particle would be subject not only to the external
potential V(x, y, z, t) but also to a "quantum potential"

which would represent some kind of reaction of the wave on the particle.

However, we know that one must be able to associate an individual wave train
with every particle. It is then possible to define for the wave train an infinity of
hydro dynamical trajectories. But, since, by hypothesis, only one particle is present
in the wave, only one of these trajectories will actually be described, the others
remaining virtual. It is easy to see that i'I/Ji 2 will not because of this lose all signif-
icance. It will evidently no longer represent the local particle density, but, if one
is completely ignorant of which trajectory is actually described, it is easily realized
that at any moment the quantity pdT = i'I/Ji 2dT will represent the probability for
finding the particle at the instant under consideration in the volume element dT.
This interpretation is of a classical nature, because it presupposes that the particle
actually has a trajectory, that the intervention of probability results only from our
ignorance about this trajectory, and that, if at the time t the presence of the particle
is detected at a point P of space, this is because at that instant the particle was
passing through the point P along its trajectory.
It can then be said that the wave guides the motion of the particle, since this
motion, assumed to be perfectly determined, is inferred from a knowledge of the wave
'I/J, whence the name "theory of the pilot wave" proposed by me. Unfortunately this
theory, quite seductive at first glance, runs into insurmountable difficulties on being
examined in detail.
176 Recalling the General Concepts of Wave Mechanics

Note G. L. It is interesting to follow here the "successive layers" of the author's

thinking. Indeed, first of all, the mere citing of the pilot-wave theory, even if only
to criticize it, is new with reference to the first part of this work (written in the
preceding year), where de Broglie did not even allude to his attempt of 1926-27 to
"maintain classical ideas," as he puts it above. Next there appear some notes of
significant corrections added in a different ink: This is the case for the introduction
of the phase of the wave in the speed of the Madelung fluid, where de Broglie returns
to the notations of the pilot-wave theory. It also is the case for his short note on
the quantum potential; and even here he crossed out the word "insurmountable"
and replaced it with: "which now to me appear insurmountable" (considering the
precision of de Broglie's style, the nuance is important). But it should be noted
that he still speaks only of the pilot-wave theory which he later had to qualify as
a "truncated, degenerate form of the theory of the double solution" (Ref. II, 26,
p. 107), "a somewhat hybrid theory" he would moreover say. Now, remembering
his failure at the Solvay Congress of 1927, he still seems to mix up the two theories
and not to speak of the double solution; but, after another lecture, he alludes to it
in penciled notes. This is the second step of his reflections.

First, the theory is in reality much less classical than it appears to be, because the
motion of the particle is defined by the function '¢(x, y, z, t), which is derived from
the initial form '¢(x, y, z, to) with the aid of the wave equation. Now I'¢(x, y, z, to) 12
represents the probability for finding the particle in the point x, y, z at the initial
time to. The motion of the particle at time to would therefore depend not only
on its position and velocity at that time, as in classical mechanics, but also on the
probability that at to the particle will find itself at any point x, y, z, which is a
strange state of affairs, entirely in opposition to classical ideas!
Quite generally, the motion of the particle would depend on the global form of
the wave '¢, which is determined by everything that happens to the latter on its
travel through space. Let us consider, for example, Young's double-slit experiment.
In the pilot-wave theory, the photon, having a determinate trajectory, would pass
through one of the slits; but its motion on leaving this slit would be influenced
by the existence of the other slit, since the form of the wave '¢ behind the screen
depends symmetrically on the existence of both slits. One thus is left with conclu-
sions that are paradoxical and far removed from the old ideas. [Note G. L. In the
theory of the double solution, this defect becomes a virtue: The motion of the wave
singularity, influenced by the boundary conditions, provides an excellent description
of diffraction and interference effects.] Moreover, the actual construction of the hy-
drodynamical trajectories, which is always possible, leads one, for example, in the
case of interference in the vicinity of a mirror (Wiener's experiment), to attribute
complicated and unlikely trajectories to the particle, sometimes corresponding to
speeds exceeding that of light. [Note G. L. These difficulties, on the other hand,
still remain in the theory of the double solution; an analysis of this question is given
in Ref. II, 26.]
But still more decisive reasons oppose the adoption of the pilot-wave theory.
[Note G. L. Using the same ink as for the preceding corrections, the author later
crossed out the words "still oppose" and replaced them by "seem to oppose."] We
soon shall see that the development of wave mechanics led to a (generalized) prin-
 The Interference Principle. Theory of the Pilot Wave 177
ciple of spectral decomposition, for which the principle of interference is only a
particular case, a principle which is necessary for the interpretation of the totality
of quantum phenomena. This principle leads one to assign to all quantities belong-
ing to a particle (and not only to its coordinates) a series of possible values governed
by probabilities. In other words, to every quantity there corresponds a distribution
function (as understood in probability calculus) which depends on the function tf;.
This statement is applicable particularly to the components of linear momentum;
and one can conclude from it that at every instant the values of one coordinate x
and the corresponding component Px of the linear momentum are subject to uncer-
tainties such that !:1x!:1px ;::: h (Heisenberg's uncertainty relations). These results,
which the coherence and experimental verification ofthe new mechanics have placed
beyond any doubt, can in no way be reconciled with the pilot-wave theory. The lat-
ter leads in particular to a well-defined value of the linear momentum and does not
allow us to obtain the uncertainty relations.

Note G. L. This is entirely wrong! This objection shows that Louis de Broglie,
who had distanced himself from his old ideas, did not reflect again on the problem
of measurement in a casual theory, which soon he would do under the influence of
Bohm's paper. He would then show that one must not confuse the hidden values of
a quantity, such as momentum, with its measured values. The latter are those which
one obtains after having decomposed the initial wave train into a set of separate
wave trains each corresponding to a very precise value of the momentum. The
measurement then consists in attributing to the momentum the value belonging to
the wave train in which the presence of the particle was recorded. One sees that
in this way a modification of the possible values of the momentum is introduced by
the decomposition of the initial wave, as a random element, due to the transfer of
the particle to one or another of the wave trains resulting from the decomposition
of the initial wave. These two circumstances lead to the Heisenberg relations, but
with an interpretation that is different from the usual one (see Ref. II, 27, 29, 33).

Let us further add that the extension of the pilot-wave theory to the wave me-
chanics of a system of particles gives rise to other difficulties that seem insurmount-
able.
A detailed study of these difficulties led me in 1927-28 to completely abandon
the pilot-wave theory and to adopt the new point of view of Bohr and Heisenberg,
wherein any notion involving the trajectories of elementary particles and the de-
terminism of their movement is abandoned. From current efforts on my part to
conserve the old ideas, I have gained the impression that every attempt to reach
this goal would founder on the same difficulties. If I developed this outdated point
of view it is precisely because in the most recent attempts the same objections have
surfaced again. So it happens that, in the ideas of Bass, soon to be considered,
one again encounters the formulas of the pilot-wave theory and the difficulties they
raIse.

Note L. B. In a recent work, "A suggested new interpretation of quantum the-

ory" (1951), Bohm of Princeton has returned exactly to myoid theory of the pilot
wave. The only truly new thing introduced by Bohm's work appear to me to be
178 Recalling the General Concepts of Wave Mechanics

an analysis of the measurement process, which is more detailed than that which I
published 25 years ago.
The theory of Bohm naturally runs into the same objections as mine, and they
to me appear always insurmountable. As Bohm has noted, the theory makes sense
(in particular where the introduction of the quantum potential is concerned) only if
the wave 'IjJ is a "physical reality." But this is exactly what seems to me impossible
to admit. First, the wave 'IjJ is represented by a function that is essentially complex,
and, in the general case, it propagates through a space which is clearly abstract and
fictitious, the configuration space. Already this makes it quite difficult to view 'IjJ
as a physical reality in the old sense of this phrase in classical mechanics. Further-
more, every localization experiment abruptly reduces the extension of the wave 'IjJ
in space and thereby changes its form (reduction of the probability packet), such
that a measurement performed in one region of space will modify entirely the form
of'IjJ in other regions remote from the former; and this fact, too, appears to argue
against the characterization of 'IjJ as a physical reality.
Bohm believes that the pilot-wave theory completely resolves the difficulties
pointed out by Einstein, Podolsky, and Rosen. But I am not sure of that, because
after a collision the wave 'IjJ is represented by a series of wave packets separated
from one another in configuration space, and a localization within anyone of these
wave packets will cause the other wave packets to vanish. There are no difficulties
left that concern particles which, in the pilot-wave theory, have trajectories and for
which the correlation of motions are explained quite naturally as in classical theory.
But they occur again for the wave 'IjJ, since the latter is here a physical reality, and
we can no more understand how a measurement made in one part of space manages
to modify this physical reality in another part of space. In short, the difficulty is
transferred from particle to wave.

Note G. L. It is hardly necessary to underline the importance of this note, in

which the author tells us that he had just read Bohm's paper, which soon was going
to trigger his own change of mind. His first reaction is altogether negative, and he
would confirm that publicly in a Note to Comptes rendus de ['Academie des Sciences
(see Refs. I, 93; II, 25, p. 65); but it must be strongly emphasized that what he at-
tacks is essentially the theory of the pilot wave and not that of the double solution,
about which he at this time is not thinking at all. It is Vigier who, a little later,
would anew draw de Broglie's attention to the latter theory, which here comes to
light in the form of a note scribbled later in pencil below the previous note: "Vigier
and the double solution."
Louis de Broglie's criticism of the pilot-wave theory is aimed entirely at the
subjective character of the wave 'IjJ, to which this theory, taken up again by Bohm,
attempts to give a direct physical significance. The fact that the wave is complex
constitutes a minor inconvenience, because its amplitude and phase are not defined.
On the other hand, the argument involving configuration space seems crippling and
by itself would condemn the pilot-wave theory as a causal theory, if it was not no
longer that, perhaps, because of its obvious subjective character, which confers on
the wave 'IjJ the property of probability-packet reduction by measurement. This is
why, as soon as de Broglie came back to his work on the causal interpretation of
wave mechanics, he definitively abandoned the pilot wave and turned to the double-
 The Interference Principle. Theory of the Pilot Wave 179
solution theory, wherein two waves occur, one of which is the usual wave tP, while
the other, which is not normalized and not subject to wave-packet reduction by
measurement, can actually play the role of a physical wave. Unfortunately, the
double-solution theory does not for all that remove the difficulties raised by config-
uration space. As early as 1926-27 (Refs. I, 29, 34; II, 25) de Broglie attempted to
construct a theory of particle systems in physical space and assumed Schrodinger's
theory in configuration space only for lack of anything better. Later, in the years
1952-60, he again made big efforts in the same direction with the collaboration of
Andrade e Silva, who took this topic as the subject of his thesis (see Ref. II, 26,
27, 29, and Andrade e Silva's thesis). Despite undeniable successes, one cannot
speak of an accomplished theory which can be substituted for the usual theory;
in particular, the problem of the EPR paradox remains unresolved. To illuminate
de Broglie's state of mind in 1951, we here reproduce a remark that appeared in
a review of particle systems, which belonged to the present chapter but which we
have omitted, as we said above, because it would have duplicated some material of
Part One:
"In classical mechanics, the representative point has in the course of time a
well-determined trajectory in configuration space, corresponding to the fact that
the particles of the system are supposed to be constantly well localized in space.
To preserve this idea and save classical determinism, one could attempt to develop
here again a theory of the pilot wave, the propagation of the wave tP through con-
figuration space guiding the motion of the representative point in this space. For
this purpose we supposed that the representative point pursued a hydro dynamical
trajectory defined in configuration space by the 3N velocity components V Xk , v 1lk , V Zk
defined above. But the theory thus obtained is even more fictitious and encounters
still bigger problems than those found for a single particle. One must accordingly
renounce it and adopt a general probabilistic interpretation analogous to the one
developed above for a particle."
Let us note that a subsequent correction, added in the same ink as the few pre-
ceding corrections, subtly weakened the last phrase, replacing "one must" by "one
had to."
 CHAPTER 13

INTRODUCTION OF THE
CHARACTERISTIC FUNCTION
INTO THE PROBABILITY FORMALISM
OF WAVE MECHANICS

To his very great merit, E. Arnous in 1944 introduced the characteristic-function

concept into the probability formalism of wave mechanics. This function leads to
elegant expressions and furnishes new methods of calculation for tackling certain
problems, as Arnous notably showed in his Thesis (1946).1

1. CHARACTERISTIC FUNCTION FOR A SINGLE QUANTITY

Arnous noticed that the probability distribution belonging to a quantity A for

a state of a system represented by the wave function 'IjJ( q, t), where q denotes the
set of variables of the system varying within a domain D, can be derived from the
characteristic function

cp(u) = k 'IjJ*(q,t)exp(iuA)'IjJ(q,t)dr, (1)

where the operator exp (iuA) is defined by

. A) . A (iu)2 A2 (iu)n An
exp (w = 1 +w + -2- + ... + -,- + .... (2)
n.
Indeed, consider the case of a discrete spectrum for which the operator A has
the eigenvalues ak and the eigenfunctions CPk, and let

One then finds

cp(u) = 1:E4CPk
D k
exp (iuA) :Ecjcpjdr
j
= :ECkCj
j,k
rCPk
iD
exp (iuA)cpj dr. (3)

1. E. Arnous, These C.D.D., Paris, 1946; J. Phys. (Paris) Ser. VIII VIII, 87
(1947).
 Characteristic Function for a Single Quantity 181
But

(4)

whence

and thus
tp(u) = I: I kl
C 2 exp (iuak) = exp (iua). (5)
k
This corresponds clearly to the definition of the characteristic function as the math-
ematical expectation of exp (iua).
In the case of a continuous spectrum, one finds (as can be justified completely
with the aid of eigendifferentials)

tp(u) = j 1c(a) 12 ex p (iua) da = exp(iua). (6)

The introduction of the characteristic function, defining both the possible values
of. the quantity A and their respective probabilities, can be viewed as replacing the
two fundamental principles of quantification and spectral decomposition.

Note G. L. In a note added afterwards, the author carries further the calculation
given in the text and verifies that the definitions of Arnous remain valid for time-
dependent operators A, provided these do not operate on the time.

Arnous indicated a simple method for calculating the characteristic function

tp(u). Consider the equation
1 aX
-:-a
z u
= Ax, (7)
which defines the evolution, in its dependence on the variable u, of the function
X( q, u) of u and the coordinates q. If X( q, 0) is the value of X for u = 0, then one
has (because A is independent of u)

x(q,u) = exp(iuA)X(q,O). (8)

One may indeed verify immediately that X( q, u) satisfies Eq. (7) and, moreover, for
u = 0 has the initial form X(q, 0). Let us now require the initial form X(q, 0) to be
identical with the wave function 'IjJ of the system at the instant t of interest; that is,
we put X(q,O) = 'IjJ(q, t). One then gets X(q, u) = exp (iuA)'IjJ(q, t) and

tp(u) = k 'IjJ*(q,t)exp(iuA)'IjJ(q,t)dr = kx*(q,O)x(q,u)dr.

182 The Chamcteristic Function

In the language of Hilbert space, it can then be said that 4?( u) is the scalar
product of X(q,O) with X(q,u). Therefore, to construct the characteristic function,
it will be simplest to look for a solution of the equation (l/i)(oX/ou) = AX which
°
reduces to tP for u = and then to form the scalar product

4?(u) = kx*(q,O)x(q,u)dr.

If it is possible to expand X(q,u) in terms of a complete set of orthonormal

eigenfunctions, such that

x(q, u) = L Cm (U)4?m(q),
m

then one will have

4?(u) = Lc:;'(O)cm(u). (9)
m

Examples

(1) Case of Energy. On taking the quantity A to be the energy, one has A = H
and thus
4?(u) = k tP* exp {iuH)tP dr.
In: a situation where H is independent of the time, the equation (1/ i) (OX / ou) = H X
will reduce to the wave equation if one sets u = (211"/ h )t, and thus X can be identified
with '¢. We then have

tP(q,t) = exp C:i tH ),¢(q, 0),

(10)
X(q,u) = exp(iuH)x(q,O) = exp(iuH)tP(q,O),

in
and therefore
4?(u) = tP*(q, 0) exp (iuH)tP(q, 0) dr. (11 )
If herein one substitutes the expansion

(12)
n

with the tPn denoting eigenfunctions of H, the outcome is

4?( u) = L c;"cn { tP:n exp (iuH)tPn dr

m,n JD
(13)
= L c;"cn exp (iuEn) ( tP:n tPn dr = L ICn 12 exp (iuEn) = exp (iuE).
m,n JD n
 Chamcteristic Function Jor a Single Quantity 183
(2) Case oj a Linear Momentum Component. Here one has, for example,
h 8
A = -21ri 8x·
After the substitution s = (h/21r)u, the equation (l/i)(8X/8u) = AX therefore
becomes
8x 8x (14)
8s = - 8x'
whence X(x,s) = X(x - s).
Setting X(x,O) = 'Ij;(x,t), we thus find

'P(u) = Iv 'Ij;*(x, t)'Ij;(x - s, t) dr.

The operator exp (iuA) applied to any function J(x) gives

exp (iuA)J(x) = exp (-8 :JJ(x)

8 s2 82 sn an ] (15)
= [l-s 8x +"2 8x2 + ... +(-It n! 8x n + ... J(x)
=J(X-8),
by virtue of Maclaurin's formula.
This means that exp (iuA) is a translation operator, which demonstrates the
connection existing between the translation group and the operators belonging to
the components of linear momentum.

(3) Case oj the z Component oj Angular Momentum. On taking spherical coor-

dinates r, (), 'P around Oz, one gets
h 8
A = - 21ri 8'P '
and, with 8 = (h/21r)u, the equation (l/i)(8X/8u) = AX gives
8X 8x
-=--,
8s 8'P
whence X( 'P, 8) = xC'P - 8). Putting xC'P, 0) = 'Ij;('P, t), we thus obtain
'P( u) = Iv 'Ij;*('P, t)'Ij;('P - 8, t) dr.

The operator exp (iuA) applied to the function J('P) gives

exp (iuA)J('P) = exp (-8 :'P)J('P) = J('P - 8). (16)

184 The Characteristic Function

It therefore generates a rotation about Oz, which establishes the connection between
the operator belonging to one of the components of angular momentum and the
group of rotations about the corresponding axis.

(4) Case of the Operator p;. In this instance

h2 8 2
A = 411'2 8x2 '

If one puts s = (211'/h)u, the equation (1/i)(8X/8u) = AX will read

which resembles the heat flow equation for an imaginary thermal conductivity.
The solution which at s = 0 assumes the value 'IjJ(x, t) is known since Fourier
and can be written as

1
X(x, s) =
-00
+00
'IjJ(x - v, t)
exp (_1I'hv2 /2 is)
(. / )1/2
2zs h
dv, (17)

1 1
whence
+00 exp (_h 2 v 2 /4 iu)
cp(u) = 'IjJ(x,t) 'IjJ(x-v,t) 1/2 dvdr. (18)
D -00 [(41I'i/h)u]
The same formula applies to the operator M; when x is changed to cpo
Application to Darwin's Particle. Amous gave detailed examples of his general
method. In particular, he studied the case of Darwin's particle. We shall give some
of these results, confining ourselves to one-dimensional motion.
Darwin's particle is a free particle whose wave 'IjJ in one dimension is a solution
of
h 8'IjJ h2 8 2 'IjJ
211'i at = - 811' 2m 8x 2 '
the initial form of 'IjJ at the instant t = 0 being

1 )1/2 2 2 .
'IjJ(x,O) = ( 0"(211')1/2 exp ( - 4: 2 ) exp (- ~z mvxx). (19)

At the instant t, one then has, on putting a = h/411'm0"2,

1 ]1/2
'IjJ(x, t) =
[
0"(211')1/2 (1 + a2t2)
(20)
1 (x-v xt)2] exp [1I'im
X exp [- - - - (x-v xt)2 - x 2)]
40"2 1 + a2t 2 ht 1 + a2t 2 .
 Characteristic Function for a Single Quantity 185

The position probability density is therefore given by a Gaussian law (the func-
tion t/J having been chosen precisely such that this would be the case):
1 [ 1 (x-vxt)2]
p(x) = 0"(271")1/2(1 + a2t2)1/2 exp - 20"2 1 + a2t2 . (21)

This is a Gaussian distribution around a mean position x = vxt which moves uni-
formly in the course of time. The dispersion thus has the value
o"x = 0")1 + a2t 2 ,
which increases with time, meaning that the wave train spreads out while it propa-
gates.
The calculation of the probability law for Px furnishes an interesting example of
the application of Arnous's method. The characteristic function here is (see Case 2
above)
cp(u) = Iv t/J*(x,t)t/J(x -s,t)dx, with s= 2~ u.
But Px now being a first integral, its probability law does not depend on the time
(because all the moments
An = Iv t/J*Ant/JdT
are constants), as will be seen further on. One can therefore write

cp(u) = Iv t/J*(x,O)t/J(x -s,O)dx,

which gives

cp(u) = 1 +00
-00
1 (X2
0"(271")1/2 exp -20"2
8X
+ 20"2
s2
- 40"2
271"
+ Tmwx s
. )
dx. (22)

).
The integration, which reduces to a known type, easily yields

cp( u) = exp ( - 8 8;2) exp (2~i mvxs) = exp (iumvx) exp ( - 3~:~:2 (23)

But, for a variable Px obeying Gauss's law, one must have [see Chap. 11, Eq. (16)]

cp(u)=exp(iupx)exp(-O";., ~). (24)

The probability law for Px is therefore also of the Gaussian type with
h
px = mvx , O"p ., =471"0"
-- . (25)
As o"x = 0" at the instant zero, it follows that O"xO"p., = h/471". The minimum product
of the dispersions is thus realized in this case--and, one can demonstrate, only in
this case. As time goes on, the wave packet will spread, causing o"x to exceed 0", so
that O"xO"p,. > h/471".
186 The Characteristic Function

2. CHARACTERISTIC FUNCTION FOR TWO COMMUTING

QUANTITIES

Nothing prevents the generalization of Arnous's formalism to two, or even more,

commuting quantities. If A and B are two quantities to which there correspond the
commuting operators A and B, i.e., [A, B] = 0, then we can set

cp( u, v) = L 1/;* exp (iuA + ivB)1/; dT. (26)

From (26) one can deduce a distribution function F(a, (3) for the values a and (3
of A and B [or, in the continuous case, a probability density p(a, (3)]. No difficulty
is encountered in doing this, since the quantities A and B are simultaneously mea-
surable. The generalization to the case of n commuting quantities A, B, C, etc. is
immediate.
To clarify the application of this probability scheme, it is useful to recall the fol-
lowing theorem: "A necessary and sufficient condition for the existence of a common
set of eigenfunctions of the operators A and B is that they commute."
Without dwelling on the various forms this theorem can assume in the different
cases that one may consider (I have done that previously), I shall illustrate the
theorem with a simple example. Let a free particle (zero field) be given, and consider
the two operators
h2 h fJ
Hop = - -7rm
28 6. and (Px)op = --2.-::l'
7rZ uX
Since they commute, these operators must share a common system of eigenfunctions.
Indeed, the complete operator H has the (unnormalized) plane-wave eigenfunctions

1/;(px,Py,Px) = exp (- 2~i (pxx + pyy + Pzz)), (27)

and the eigenfunctions of the operator (px )op are

exp ( _ 2~i Px x ).
In agreement with the theorem, an eigenfunction of H always is an eigenfunction of
Px multiplied by a function of the variables y and z not appearing in PX' Moreover,
as we know, the eigenvalues of Hop are of the form

1 2 2 2
E = 2m (Px + Py + pz)·
Therefore, if p;/2m < E, there is, for a given value of E and Px, an infinite number
of possible values of Py and pz. The eigenvalue of E thus is degenerate to an infinite
order, and there is an infinite number of corresponding eigenfunctions, all of which
are proportional to the eigenfunction
27ri )
exp ( -TPxx
 Correlation Coefficient. Marginal Laws 187
of (Px)op.
Suppose therefore that we are dealing with two commuting quantities A and B,
the operator A being complete. Let 'Pi be the eigenfunctions of A and Xk the eigen-
functions of B. We have 'Pik(X, ... , y, ... ) = fik(Y, ... )Xk(x", .)j and, on writing
the wave function 'IjJ as an expansion of these functions, one gets

'IjJ(X, ... y, ... ) = 2: Ci'Pi(X""y",,) = 2:C;!ik(Y"")xk(X",,)

i,k
(28)
= 2: bk(Y," ,)Xk(x", .).
k
One therefore obtains

'P(u,v)= ( 'IjJ*exp(iuA+ivB)'ljJdr= LickI2exp(iuak)exp(iv!1k)

iD k (29)
= exp (iua + iv!1).
The hypothesis requiring A to be a complete operator is moreover not essential,
and one can easily discard it at the price of some writing.
In the case where A and B both are complete operators, it follows from the
preceding theorem that there exists a complete system of eigenfunctions, depending
on the set of variables that appear in A and B, that is common to A and B. If
A and B both exhibit simple eigenvalues, then each eigenfunction of the system
corresponds to one eigenvalue of A and one eigenvalue of Bj which means that these
eigenvalues have a one-to-one correspondence such that knowledge of the value of A
implies that of B, and conversely. The eigenvalue correspondence is not one-to-one
if one of the operators has multiple eigenvalues.
If the operators A and B are independent, that is, operating respectively on
different variables x and Y, then one obtains all the simultaneous eigenfunctions of
A and B by multiplying each eigenfunction of A by all the eigenfunctions of B.

3. CORRELATION COEFFICIENT. MARGINAL LAWS

Our ability to define a characteristic function for commuting quantities permits

us to introduce the notions of dispersion, correlation coefficient, marginal laws,
regression curves, and so on.
The correlation coefficient is defined by
AB-AB
r=---- (30)
O"AO"B

In general r is not zero. Consider, for example, the case where A and B are complete
and have simple eigenvalues. Then they have the same system of eigenfunctions 'Pi;
and, if
188 The Characteristic Function
one obviously has

(31)

This expression is zero only by exception, for example, when all the Ck are zero
barring one, which then has the modulus 1; this is the case where the quantities A
and B have known values.
As another example, consider two independent quantities A and B, such that
the operator A acts only on the variables x, etc. and the operator B only on
the variables y, etc. The eigenvalues of A are ai, its eigenfunctions <pj(x, .. .); the
eigenvalues of Bare (3kl its eigenfunctions Xk(Y'" .). The expansion of"p then has
the form
"p = LCik<Pi(X, .. ')xk(Y'" .),
i,k
the Cik being some constants, if it is assumed that the set x, etc., and the set Y, etc.,
exhaust the totality of variables on which "p depends. One then has

(32)

Although the operators A and B are independent in the sense that they act on
different variables, the quantities A and B are not in general stochastically indepen-
dent, because generally there is no reason why the expression (32) should be zero.
This stems from the fact that the probabilities for the possible values of A and B
are interrelated by the form of the expansion of "p, i.e., by the values of the Cik.
We now proceed to examine the marginal laws. To vary the exposition somewhat
and to prepare ourselves for later considerations, we consider the case of a contin-
uous probability distribution (operator with a continuous spectrum). Suppose that
one of the operators, say A, is complete and let it have eigenvalues a, forming a
continuous sequence, and eigenfunctions <p( a, x, ... , Y, ... ). The operator B is taken
to be incomplete; let it have the eigenvalues (3, forming a continuous sequence, and
eigenfunctions X((3, .. .). Since, by hypothesis, A and B always commute, one has

<p(a,(3,x, ... ,y, ... ) = d(a,(3,y, ... )x((3,x, .. .),

on applying to the continuous case formulas demonstrated previously for the discrete
situation. Let us expand "p as follows:

"p(x, ... ,y, ... ) = jc(a,(3)<p(a,(3,x, ... ,y, ... )dad(3

(33)
= j c( a, (3) d( a, (3, Y, ... )x((3, x, .. .) da d(3.
 Correlation Coefficient. Marginal Laws 189
The probability for finding the value a for A is then clearly given by the probability
density
(34)

The probability that B will be found to have the value (3 is given by the prob-
ability density obtained when the squared modulus of the coefficient of X((3, x, ... )
in the development (33) of'lj; is integrated over the variables y:

(35)

Obviously, the probability for finding simultaneously the value a for A and the value
(3 for B is
p( a, (3) = Ic( a, (3) 12. (36)
The formula
/ p(a, (3) d(3 = PA(a)
follows immediately from Eq. (34). As for the formula

we can verify it by writing

PB((3) = /dy ... /c*(a,(3)d*(a,(3,y, ... )da /c(a',(3)d(a',(3,y, ... )da'

(37)
= / / da da' c*(a, (3)c(a', (3) / dy d*(a, (3,y, ... ) d(a', (3, y, ... )

and by noting that one must have

/f 'P*(a, (3, x, ... , y, .. .)<p(a', (3, x, ... , y, ... ) dx dy = <5(a - a') (38)

(orthonormality of the 'P), whence

/ /d*( a, (3, y, . .. ) d( a', (3, y, ... )X* ((3, x, ... )X((3, x, ... ) dx dy
(39)
= /d y d*(a,(3,y, ... )d(a',(3,y, ... ) = <5(a - a').

From this it follows that

PB((3) = / / da da' c*(a, (3)c( a', (3) <5(a - a')

( 40)
= /lc(a,(3)1 2 da= /p(a,(3)da; q.e.d.
190 The Characteristic Function
We have argued under the assumption that A is complete and B incomplete. An
analogous reasoning could be given for A and B incomplete or even independent.
In all cases it is equally easy to define the laws for the conditional probabilities

(A) ( (3) = p( a, (3) (41)

PB a, ( ) .
PA a

When the quantities A and B commute, all the above concepts have the meanings
commonly assigned to them in probability calculus. Further on we shall see that,
when A and B do not commute, it is impossible to bluntly apply the usual formulas.
This point will be the subject of a deep examination.
In the case of a free Darwin particle (whose position probability is a Gaussian
distribution), Arnous in his Thesis studied the pairs of quantities H, Px and H, M z ,
each of which commutes for a free particle. For the first pair he calculated the
regression curve and the dispersion equation. Although these calculations are inter-
esting as illustrations of the theory, they do not appear to produce any new results
that are important from the physical point of view; thus we shall not present them.

4. GENERAL THEOREMS OF WAVE MECHANICS

CONSIDERED FROM THE CHARACTERISTIC-FUNCTION
POINT OF VIEW

Arnous linked together the general theorems of wave mechanics from the point of
view of the characteristic function he had introduced. We here give some examples
of his findings:

(a) First Integrals. We defined the first integrals (see Chap. 5, Sec. 5) by requir-

k
ing all the matrix elements
aij = 'l/JiA'l/Jjdr

to be constants in time and obtained for A the condition

~~ + 2~i (AH -HA) = O.

We now want to show that this condition leads to the constancy in time of the
probability law for A. To this end, let us first note that, 'I/J being a linear combination
L ci'I/Ji of the 'l/Ji, one has
i

(42)

so that, if A is a first integral, A, that is, the first moment of A given by its
probability law, is constant in time.
 General Theorems of Wave Mechanics 191
We now note that, if

8Ak
8t + 27ri
h (AkH· _ HAk) = 0 J:
lOr
k 1 2 ... , n,
="

then it is also true that

Indeed, multiplying the nth equation by A from the left, one obtains

(43)

But
h 8A
27rZ'-8
AH=HA-- t '
whence

and, since

we conclude that
8A n+1 2'
(A n+1 H - = o·
--
8t +~
h
HAn+!)
'
q.e.d.

From this result it follows by recursion that, if

8A 27ri (AH _ HA) = 0

8t + h '
then also
8An +(AnH-HAn)=o
8t
for all integral values of n. Consequently: "If A is a first integral, An will likewise
be a first integral." Therefore, finally: "If A is a first integral, then all moments

of the probability law of A will be constants in time." A knowledge of all the

moments being equivalent to knowing the law of probability, one may conclude that
the latter is independent of the time. This is the statement made at the start.
Moreover, the constancy in time of the probability law for A simultaneously implies
the independence of the eigenvalues of A and their respective probabilities.
192 The Characteristic Function

Since the probability law for A is invariable if and only if A is a first integral,
the corresponding characteristic function

<p(u) = k 'Ij;*exp(iuA)'Ij;dr

must also be invariable. But 'Ij;, being a solution of the wave equation

J!:.- a'lj; - H.I.

27ri at - 'f"

is determined by its initial form 'lj;o, and there exists a unitary function U(t) of the
time that transforms 'lj;o into 'Ij;:

'Ij;( q, t) = U( t)'Ij;( q, 0), ( 44)

q denoting the set of variables of the domain D, where U-1(t) = ut(t).

If the operator H does not depend on the time, one has

27ri
U(t) = exp ( htH ) . ( 45)

If H depends on the time, U has a more complicated form. The constancy of

the characteristic function in the course of time is expressed by

L 'Ij;*(q,t)exp[iuA(t)]'Ij;(q,t)dr = L 'Ij;*(q,O)exp [iuA(O)]'Ij;(q, 0) dr, (46)

that is,
k[U( t)'Ij;( q, 0)]* exp [i71A(t)]U( t)'Ij;( q, 0) dr

=k 'Ij;*(q, O)ut exp [i71A(t)]U'Ij;(q,O)dr ( 47)

=k 'Ij;*(q, 0) exp [iuA(O)]'Ij;(q, 0) dr,

whence
L 'Ij;*(q, 0) [ut exp [iuA(t)]U - exp [iuA(O)]] 'Ij;( q, 0) dr = O. (48)

This equation being true for any 'Ij;(q, 0), one can conclude

ut exp [iuA(t)]U = exp [iuA(O)], or exp [iuA(t)]U = U exp [iuA(O)]. (49)

If H is independent of t,

U 27rit)
= exp ( -h- .
H = exp (zvH), with
 General Theorems of Wave Mechanics 193
and the condition

exp [iuA(t)] exp (ivH) = exp (ivH) exp [iuA(O)]

obtains. If A, too, is independent of the time, A(t) = A(O) = A and

exp (iuA) exp (ivH) = exp (ivH) exp (iuA).

This commutation condition can be realized (as one may readily verify by writing
the exponential function as a series) only if A and H commute. The condition for
the first integral therefore becomes AH = H A, as we already know.
Consider now the general case where both A and H may depend on the time.
The condition for a first integral can now be written

exp [iuA(t)] = U exp [iuA(0)]U- 1 ,

and it is easy to see that this is equivalent to

A(t) = U(t)A(O)U-l(t).

But, by its very definition, the operator U(t) must obey the equation

~ f)U =HU
27ri f)t '
while the operator U- 1 satisifies the equation

_~ f)U- 1 = U- 1 H
27ri f)t '
as one readily sees. Therefore

2:i f)~;t) = [f)~;t) A(O)U-l(t) + U(t)A(O) f)U;;(t)] 2:i

(50)
= HU A(O)U-l - U A(O)U-l H = H A - AH,

whence
f)Af)(t) + -2
h . [A(t)H - HA(t)] = 0, (51 )
t 7rZ
a known condition.
[One immediately obtains the formula used above for the time variation of u- 1
by writing
1/J(q,O) = U- 1 (t)1/J(q, t)
and differentiating this equation with respect to the time, which gives

f)U- 1 .1. U- 1 27ri H.I.

0=--0/+ - 0/
f)t h
194 The Characteristic Function

for any 1JI.]

(b) Interference Principle. This principle is found immediately starting from

the characteristic function: On taking A = x, one gets

<p(u) = k 1JI* exp (iux)1JIdr = 1 + iux + ... ,

k
with
x = x11JI1 2dr,

which proves that the probability density for the coordinate x is 11JI12.
(c) Probability Laws for px, M z , p~ and M;. Amous studied the probability
laws for these quantities from the perspective of the characteristic function. That
gave him the opportunity to make various interesting remarks from the mathemat-
ical point of view. But we shall pass over them, seeing that the essential results for
physicists are those which we already know.

(d) Theorems Relative to the Center of Mass of a Particle System. Consider a

system of masses mI, ... , mn, and let xl, YI, Zl, ... , Xn , Yn, Zn be their coordinates.
One can (at the very least in the nonrelativistic wave mechanics we are studying
here) define the coordinates of the center of mass of the system by
n n
Xg = L miX; / L mi,
I 1
n n
Yg = LmjYi /Lmj, (52)
I I

and then introduce the relative coordinates

ek = Xk - Xg, 'TJk = Yk - Yg, (k = Zk - Zg (53)

of the kth particle. Obviously
n n n
Lmkek = Lmkxk - LmkXg = 0,
I I I
n n
(54)
Lmk'TJk = Lmk(k = O.
I I
 General Theorems of Wave Mechanics 195
By detailed arguments, which can be found for example in my book on wave-
mechanical systems, one can then establish theorems for systems in wave mechanics
that are analogous to theorems obtaining in the mechanics of classical systems.
Arnous gave synthetic demonstrations of this, starting from the characteristic func-
tion, which we are here going to reproduce.
Consider the quantity

whose corresponding operator is

h n a
-21ri ~ OXk •
The characteristic function has the form

'Pp
z
(u) = JDf 'I/J*exp (-~u
21r
Lk -a
O )'l/J dr
Xk
(55)

where s = hu/21r. But

a
(-s -)f(x) s-axa + -2 -ox2
a + .. ·)f(x)
s2 2
exp
ax = (1 -
(56)
=f(x-s),
according to Taylor's formula, so that

'Pp
z
(u) = f
JD
'I/J*( ... ,Xk, •.. )'I/J( ... ,Xk -s, ... )drj (57)

thus, after the change of variables (Xl, • .• , zn) -+ (x g , • •• , (n),

'Pp
z
(u) =
JD
f 'I/J*( ... ,ek + x g, ••• )'I/J( ... ,ek + Xg - s, ... ) dr. (58)

But
'I/J( ... ,ek + Xg - s, ... ) = eXPC~i SPXg )'I/J( ... ,ek + Xg , ••• )
(59)
= exp (iupxg)'I/J( . .. ,ek + xg , . •. ),
and consequently

'Pp
z
(u) =
JD
f 'I/J*( ... ,ek + Xg , ... ) exp (iupxg)'I/J( ... , ek + xg,- .. ) dr. (60)
196 The Characteristic Function

The characteristic function for Px thus equals that of PXg. From this follows a
theorem analogous to the classical theorem on the linear momentum of the center
of mass (first theorem of Krenig):

Theorem: The total linear momentum of a system equals the linear momentum
of its center of mass.

Let us now study the component Mz of the angular momentum of a system:

n
(Mz)op = - Lk 27ri
h (0 Xk OYk - Yk OXk
0) . (61)
1

With s = hu/27r, the corresponding characteristic function reads

(62)
x "p( ... ,Xk cos s - Yk sin S,Yk cos S + xksin S,Zk, ... ) dr.

Under the change of variables (Xl, .. . ,zn) --+ (x g , •• • ,(n), one gets
Xk cos S - Yk sin S --+ (xg + ek) cos S - (Yg + '17k) sin S
(63)
= Xg cos S - Yg sin S + ek cos S - '17k sin s,
which gives

with
Mg. = -2h . (xg -:- - Yg fClO ),
7rZ uYg uXg
(65)

It follows that
Mz = Mg. +M,.
From this outcome a theorem analogous to Krenig's second classical theorem can
be inferred:

Theorem: The total angular momentum of a system equals the angular momen-
tum of its center of mass augmented by the angular momentum of the relative
motion around the center of mass.

Arnous's Thesis contains a large number of other interesting demonstrations, to

which we shall return.
 Case of Two Noncommuting Quantities 197
We are now going to examine in detail the question of extending the usual
formalism of probability calculus to the case of two noncommuting variables, which
exercise will lead us to be more specific about some very important points.

5. CASE OF TWO NONCOMMUTING QUANTITIES

From the discussions above, it follows that in the case of two commuting
variables it is not difficult to define the characteristic function and the proba-
bility distribution in the manner that is customary in statistics. On the other
hand, for variables that are noncommuting, and thus not simultaneously mea-
surable, one must expect, according to the remarks made in Chap. XI, Sec. 3,
et seq., to encounter difficulties in the application of the usual formalism. That
is exactly what happens, and one must examine the ensuing problem with care.
To take up this study, we must begin by carefully clarifying some important
points.

(a) Reminder About the Reduction of the Probability Packet by Measurement.

We know that measurement plays an essential role in the physical interpreta-
tion of wave mechanics: It provides us with new information, changes our state
of knowledge about the system studied, and consequently forces us to abruptly
modify the form of the wave 1/J representing our knowledge about the system. If,
for example, the measurement is a more or less precise measurement of position,
the wave train representing 1/J before the measurement will afterwards find itself
"reduced" to a less-extended wave train or even, if the measurement is precise,
to' a train that is nearly pointlike, since the region wherein the position proba-
bility 11/J12 is different from zero will have diminished in size, whence the name
"reduction of the probability packet" given by Heisenberg to this sudden mod-
ification of 1/J. If, on the other hand, the measurement consists in a determi-
nation of one of the momentum components, it is in momentum space, not in
configuration space, that the reduction of the probability wave-train packet will
occur.
The reduction of the wave train gives rise to a new situation, characterized by a
new 1/J, a situation which was unforeseeable, since only the probabilities of the vari-
ous measurement results could be calculated in advance of the measurement. In the
last part of this course, we shall analyze more closely the manner in which a mea-
surement operation isolates one of the possibilities contained in the state preceding
the measurement. And we shall show, with von Neumann, that the indeterminism
thus introduced into atomic physics is without doubt of an essential nature, be-
cause it is impossible to explain 2 the intervention of probabilities as being due to
us not knowing the exact values of some variables that would elude us and conse-
quently to interpret this intervention in the manner that is customary in classical
physics.

Note G. L. A correction, made in a different ink and certainly after the text,
plainly weakens this statement, which became, in its new version: " ... and we will
have to wonder, with von Neumann if the indeterminism which thus enters into
198 The Characteristic Function

atomic physics is of an essential nature and if it is impossible to explain ... " The
italics are those of the author; one will note the disappearance of words "without
doubt."

Following a measurement providing us with the maximal knowledge about a

system compatible with the theory of non commuting variables, we are in a position
to construct the wave function 'ljJ, representing our knowledge of the system after
the measurement, and follow its evolution in the course of time with the aid of the
wave equation. We shall thus at any time be able to calculate the probabilities
belonging to all possible results of various measurements that one could perform at
that moment. This will remain the case until we would get to know the results of
new measurements, modifying the state of our knowledge and suddenly interrupting
the regular evolution of the wave 'ljJ. The evolution of 'ljJ between two measurements,
which is governed by the wave equation, is entirely determined by the initial form of
'ljJ, since the wave equation is of the first order with respect to the time. Determinism
thus exists with regard to the evolution of'ljJ between two measurements but not in
relation to the observable phenomena.
The knowledge of 'ljJ after a measurement does not permit us in any way to
establish the value 'ljJ before the measurement, since the measurement introduces
a discontinuity in the evolution of 'ljJ. Let us consider a large number of systems
finding themselves in the same state 'ljJ and assume that we measure for each of
these systems a certain quantity A with eigenfunctions 'PI and eigenvalues al. The
fraction of systems for which one will find that A has the value al is furnished by
the squared modulus of the coefficient q appearing in the expansion

of'ljJ before the measurement. Therefore, if we know 'ljJ for the ensemble of systems
after the measurement, this amounts to knowing the values of the Iq 12 and thus of
the moduli Iq I. But a lack of knowledge of the phases of the of the q will always
prevent us from knowing 'ljJ before the measurement.
We thus see that every measurement causes a complete effacement of the phases
(Bohr). This obliteration of the phases by measurement implies that the act of mea-
surement constitutes a break in the evolution of'ljJ which is insurmountable, in the
past-future as well as in the future-past directions. Moreover, the phase differences
between the components 'PI of 'ljJ have a basic significance: Any knowledge of 'ljJ
which does not include a knowledge of the phase differences is radically incomplete.
This importance of the phases is well illuminated by a study of the fundamental
question concerning the interference of probabilities.

(b) Interference of Probabilities. Consider two noncommuting quantities A and

B. Let the eigenvalues and eigenfunctions of A be ai and 'Pi, respectively; and let
the corresponding properties of B be f3i and Xi. The set of 'Pi and the set of Xi
cannot be identical, since the operators A and B do not commute; see the theorem
in Chap. VI, Sec. 1. As the Xk form a complete set, one may express the 'Pi in terms
 Case of Two Noncommuting Quantities 199
of the Xk by formulas of the type

<Pi =L SikXk' (66)

k
the Sik being the elements of a unitary matrix S. In this expansion several terms
will in general appear on the right-hand side, since the set of <Pi and that of Xk do
not coincide. Suppose then that the initial state of the system is represented by the
wave function

One then has

.,p = L Ci<Pi = L CisikXk·
i,k
If now the quantity A is measured on a system in this state, one of the eigenvalues
(Xiwill be found, the probability for any (Xj being ICj 12. After this measurement, the
system will find itself in the state <Pi ; and, in this new state, a measurement of B will
lead to a value (3k with the probability ISik 12. Consequently, the total probability
that one will find the value (3k of B when performing first the measurement of A
and then that of B is

But, suppose now that one carries out the measurement of B directly on the
initial state.,p. Then, in view of the above expansion of .,p in terms of the functions
xi." the general principles of quantum mechanics teach that one will find the value
(3k for the quantity B with the probability

This expression differs fundamentally from the preceding one, because it depends
on the phases of the coefficients Ci and Sik, while the preceding expression does not.
Let us illustrate this with a simple example in a one-dimensional domain of
length L. In this domain, the normalized eigenfunctions of the linear momentum
p=Aare

Let
.,p = ~
~ Vi exp h
Ci (27ri (Wit - PiX)
)
I

be the wave function of the particle in its initial state. If now one measures first P
and then the position X, the probability of the position X = Xo will be

~
~ ICil Vi exp (27ri
2/1 -I: PiXO ) /2 = L'
1
I
200 The Characteristic Function

because

Thus all positions in the domain L are equally probable. If, by contrast, one mea-
sures x directly in the initial state, the probability for x = Xo will be

12;: Jr
c' (21l'i
exp -T PiXO
) 12
,
I

and it results from the interference of plane waves constituting 'IjJ, an outcome that
is necessary for the explanation of interference in optics and in electron diffraction.
It is therefore seen that the interference of probabilities, whose existence is required
for the interpretation of experimental facts, depends basically on the phases, whose
role is thus seen to be fundamental.
The fact that the probability of the value 13k of B measured directly in the initial
state is
II:CisikI2, and not I:ICiI2IsikI2,
,
seems to be a priori contrary to the theorem of compound probabilities. But, in
reality, it is not: The probability

is the probability one must adopt when performing first the measurement of A,
then that of Bj but there is no reason why this probability should equal that of
finding the value 13k for B in a direct measurement of B in the initial state, since
the measurement of A, incompatible with that of B, entirely changes the state of
probabilities.
What creates a certain confusion in this question is, as we have seen, that,
in the usual statistical applications of probability calculus, one assumes that the
measurement of a stochastic variable (the "trial" of statisticians) does not in any way
modify the relative probabilities of the other stochastic variables. This hypothesis,
which is obviously correct in questions of macroscopic physics which usually occupy
statisticians, is no longer accurate in microphysics.
The foregoing circumstance explains why it is impossible, in dealing with the
noncommuting variables of microphysics, to construct the probability densities
p(a,f3), p1A )(a,f3), and p~B)(a,f3) corresponding to the scheme usually employed
in statistics. What happens is well illustrated by an attempt (also otherwise inter-
esting) in this direction due to Bass,2 which we are going to study next.

2. Jean Bass, "Les equations d'evolution et de transfert en mecanique alE~atoire.

Exemples mathematiques. Applications a. la mecanique quantique," These, Paris,
1948.
 Case of Two Noncommuting Quantities 201

(c) The Distribution Function p(x,Px) of Wigner 3 and Bass. Consider two
canonically conjugate quantities x and Px' We know that they do not commute
and are not simultaneously measurable. A priori, a probability density p(x, Px)
defined in the usual manner must be viewed as nonexistent, since p(x,Px) dx dpx
would have to represent the probability of simultaneously finding a value of x in the
interval dx and a value of Px in the interval dpx, when in fact it is impossible to
simultaneously determine x and Px; and, when one measures these two quantities
in succession, the first measurement will completely disturb the previously existing
state of probabilities.
Bass was however able very ingeniously to employ a function p(x,Px) such that
the conditions

J p(a, (3) da = Pn((3) and J p(a, (3) d(3 = PA(a)

would be satisfied. But we shall see that this function cannot be considered as giving
the simultaneous probability for the variables a and (3.
To find Bass's function, we reason as follows. Starting from the formula of
Amous for two commuting quantities A and B,

'P(u, v) = Iv 1f*exp (iuA)exp (ivB)1fdr, (67)

we shall try (as Amous had the prudence not to do) to apply it to two noncommut-
ing variables. We immediately run into a difficulty: While, with two commuting
variables, one can write indifferently under the integral sign

exp (iuA) exp (ivB) or exp (i ¥B) exp (iuA) exp (i ¥B), etc.,
because the exponential functions obviously commute, the situation might here seem
to be different. But one can verify that this is not the case and that the two
expressions are still equal if A and B fail to commute. For A = x and B = Px one
can therefore set

'1"( u, v) = 1:00 1f*(x) exp (i ¥(Px)op) exp (iux) exp (i ¥(Px)op )1f(x) dx
(on disregarding the variables y and z). Since the operator
(68)

exp (i ¥(px )op )

is Hermitian, one can write

(69)
= [:00 1f*(x+~)exp(iux)1f(X- ~;)dx,
3. E. P. Wigner, Phys. Rev. 40, 749 (1932).
202 The Characteristic Function
in accordance with the properties of the operators

exp ( ± i ~ (Px )op ) .

This expression one could call the "Wigner-Bass characteristic function," but from
what follows it will become clear that (69) is not a true characteristic function.
With the aid of Fourier's inversion formula, let us go over to the probability

f1+
density
00
p(x,px) = -1
2 exp(-iux)exp(-ivpx)cp(u,v)dudv, (70)
411" -00
where, in the second factor of the integrand, Px denotes the value of Px and not the
operator (Px)op. On substituting into this formula the expression for cp(u, v), one
finds
p(x,px) = 411"2
1 fr [[+00
J J-oo 1j!* (w
hv
+ 411") exp [iu(w - x)]
(71)
.
x exp (-zvPx)1j! hv) du dv dw.
( w - 411"

But herein

l +ooexP[iu(w-x)]du = 211"h'(w-x) =211" lim sin

N-+oo
211"~(W-)x),
11"

I:
-00 W - X

whence
p(x,px) = 2~ 1j!* (x + ~:) exp (-ivpx)1j!(X - ~:) dv
or, if we set z = hv/411",

(72)

This is the Wigner-Bass distribution law.

Note L. B. It goes without saying that p in general is a function of the time,

since 1j! depends on t, so that one should strictly write p(x,Px, t).

We now verify that this expression, even if obtained in a debatable fashion,

nonetheless satisfies the two conditions

PA(a) = jp(a,f3)df3 and PB(f3) = jp(a, f3) da.

Here one has PA (a) = 11j!( x) 12 according to the interference principle. The first
equation to be verified therefore is
 Case of Two Noncommuting Quantities 203
Now, we have in fact

1-00+00 p(x,Px) dpx = -h21+00

-00
[ l~oo --2
h
1T
41T
sin -h Px
p", Z
Z
]
'I/;(x - z)'I/;*(x + z) dz. (73)

And, since
C( U ) -_ 1.1m sin 21Tpx U ,
Q
p.,--+oo 1TU
one finds, on using the properties of the delta function (take 2z / h as the integration
variable), that

-00 8 (2Z)
1-00+00 p(x,px) dpx = h21+00 h 'I/;(x - z)'I/;*(x + z) dz = 1'I/;(x)12j q.e.d. (74)

The second equation to be verified is

c(px) being the coefficient of

in the expansion
.t. -1+00 c(Px )exp [-(21Ti / h )pxx] dPx
'f/ -
-00 vh
Ii:

of '1/;, which is a superposition of plane normalized waves. Now Fourier's inversion

formula gives

c(px) = 1 1+
..Jh -00 00 exp (21Ti
-;;: PxX ) 'I/;(x) dx,

whence
Ic(Px)12 = ~ 1£:00 exp C~i px(e - T/) )'I/;(O'l/;*(T/) de dT/. (75)

But, from (72), one has

1-00+00 p(x,px) dx =
2J~ [+00 exp (41Ti
h Loo --;;: px Z
)
'I/;(x - z)'I/;*(x + z) dz dx,

which, subsequent to the transformation x - z = e, x + z = T/, can be written as

(76)
204 The Characteristic Function

Despite the success of the above two verifications, the Wigner-Bass function
p(x,Px) cannot be regarded as giving the simultaneous probability for x and Px:
Apart from the fact that the precise simultaneous measurement of x and Px is
impossible, as Bohr and Heisenberg's analyses of the measurement procedures for
these quantities have shown, we may note, with Arnous, that the function p(x,Px)
is not positive-definite; it can assume negative values, which opposes its adoption
as a probability density.
Moreover, the functions

and

do not have the significance of joint probabilities in the usual sense. Indeed, p( 0:,13)
is defined by 7/J before any measurement, and the same is true for the quantities PA (0:)
and PE(j3). The quantities Pit) and piE) are thus defined in the state preceding any
measurement. Thus they cannot represent the probability of any value of B when
one knows the value of A and the probability of any value of A when the value of B
is known, since the first measurement one performs will completely change the state
of probabilities relative to the unmeasured quantity. This kind of situation never
occurs in macroscopic statistics, where it would correspond to the measurement of
the height of a draftee influencing his chest size! But here, in microphysics, this
state of affairs occurs in a basic manner for noncommuting quantities.
Consider, for example, the simple case where 7/J for the original state is a, plane
monochromatic wave
1/J = c exp ( _ 2~i p~O) x) ,
corresponding to the value p~O) of px. Then a measurement of px performed on
the initial state will certainly give the value p~O). By contrast, a measurement of x
made on the initial state will produce any value of x with equal probability, since
17/J12 = const. This is a well-known result: When a plane monochromatic wave is
associated with a particle, the linear momentum of the particle has a definite value,
while its position is completely indeterminate. Calculation of Bass's density p(x,Px)
for the initial state gives

p(x,Px) = const. 1+00

-00 exp
(47ri
-T Px Z ) exp (27ri
T Px(0) (x + z) )
27ri (0)
x exp ( -T px (x - z) dz
)

(77)

= const. 1+00
-00 exp
(hi
T (0) )
(Px - Px)z dz

(0)
=const.li(px -Px).
 Case of Two Noncommuting Quantities 205
This form of p(x,Px) is a natural one: The delta factor expresses the fact that only
a single value of Px is possible, and the absence of x in the expression conveys the
fact that all values of x are equally probable. However, on constructing

(X)( ) _ p(x,Px)
pp
x
px - ()'
Px x

one finds const. 8(p~O) - Px). Now this result is wrong, because, if one first measures
x so as to know its exact value Xo, then this measurement will completely modify
the function 'lj;, which becomes 8(x - xo), and in this new state all values of Px
are equally probable. The true P~:)(Px), giving the probability of the value of px
when one knows that x has the value Xo, is therefore a constant, and one sees that
the joint probabilities cannot be deduced from p(x,Px) with the aid of the usual
statistical formulas.
In the usual statistical formalism, if one knows the function p( a, 13) for two
quantities A and B, then all quantities of the type f( a, 13) have a mean value

f(a, 13) = jjf(a, j3)p(a, 13) dadj3, (78)

which means, to be precise, that, if one measures the quantity f( a, 13) for an infinity
of systems all having the probability distribution p( a, 13), then one must find the
mean value of f as written.
In wave mechanics, a measurable quantity f is defined not by a numerical func-
tion but by an operator. Let accordingly j[x, (Px)op] be a linear Hermitian operator
corresponding to a certain measurable quantity; its mean value will then be
+00
7= 1-00 'lj;*(x)j[x(px)op]'lj;(x) dx.
We can then ask if this mean value equals the mean value which the Wigner-Bass
density p( x, Px) allows one to define by the relation

7B = ji:oof(x,px)p(x,px)dXdPx, (79)

This is in fact what happens if f depends only on x or only on Px. Indeed, in

the first case, the quantum mean value is
+00
7= 1 -00 f(x)J'lj;(x)J 2 dx,

while Bass's mean value is

7B = 1+00f(x) dx 1+00p(X,Px) dpx

-00 -00

=1
+00
-00 f(x) J'lj;(x) J2 dx = 7,
206 The Characteristic Function
according to the properties of p(x,Px). In the second case, the quantum mean value
is

-f -1+ood 1+ood I ( ' ) exp[(21ri/h)P~x]1+ood f( ) ( ) exp[-(21ri/h)pxx]

- x Px C Px r;: Px Px c Px r;: ,
-00 -00 yh -00 yh

sInce
J[(px)op] exp ( - 2:i Pxx) = f(px) exp ( _ 2:i Pxx) ,
as one could see on developing J[(px)op] in a Taylor series. The orthogonality of the
plane waves then gives
(SO)

As for Bass's mean value, it is given by

(Sl)

in accordance with the properties of the p(x,Px)j q.e.d.

On the other hand, if f[x, (Px)op] depends on both x and Px, it does not appear
possible to show that 1B
= 1, the measurement of f(x,px) being in general incom-
patible at the same time with the measurement of x and with the measurement of
px. [Note L. B. See the work of Yvon further on.]
In summary, we see that the Wigner-Bass function p(x,px) possesses some in-
teresting properties but that it cannot be classed as a density p( a, (3) of the type
commonly employed in macroscopic statistics. To further clarify the matter, we
again take up the question.
Let two noncommuting quantities A and B be associated with a quantum system,
and assume <p(a,q) and X((3,q) to be their respective eigenfunctions. We have

<p(a,q) = 1+00

-00 d(a,(3)x((3,q)d(3. (S2)

If the wave function 'ljJ of the system can be expanded in the form

'ljJ(q) = 1+00
-00 c(a)<p(a,q)da =
Jrf+oo
i-oo c(a)d(a,(3)x((3,q)dad(3, (S3)

the probability densities for A and B in the state 'ljJ will be

PB((3) = \ 1 +00

-00
2
c(a)d(a,(3)da\ ' (S4)

the second equation of which exhibits the interference of probabilities.

 Case of Two Noncommuting Quantities 207
However, if one performs successively a measurement of A and then of B, the
theorem of compound probabilities predicts that the B values will occur with the

I:
probabilities

p~«(J) = I:00Ic(a)12Id(a, (J)1 2 da = oo

PA(a)p1A)(a,(J)da, (85)

but here p~«(J) does not equal PB«(J), and p1A)(a,(J) differs from p(a, (J)!PA(a).
In the case A = x and B = Pz, on introducing the Wigner-Bass density p(x,Pz),
one easily sees that

but the probability of pz after consecutive measurements of x and of pz is not equal

to Pp.,(Pz), and the true conditional probability p~:)(x'Pz) is not calculable from
the formula
(x)( ) _ p(x,Pz)
Pp X,Pz - ().
'" Px x
Therefore, if one wants to use the Wigner-Bass function p(x,Pz), it is necessary
to proceed with great caution, since otherwise one would risk, as we are going to
show, falling into difficulties analogous to those besetting the pilot-wave theory.

(d) The Wigner-Bass Density p(x,Pz) and the Hydrodynamic Interpretation of

Wave Mechanics. The existence of the Wigner-Bass density p( x, Pz), enjoying some
interesting properties, could revive hopes for hydrodynamic interpretation of wave
mechanics.
Consider a fluid conceived of in the classical sense as consisting of an enor-
mous number of particles, each of which at every instant has a position and a
well-determined velocity. To simplify matters, we assume a linear fluid, which as-
sumption does not restrict the generality of our discussion, since it amounts to
considering only the projection on a straight line of the positions and velocities of
the fluid particles. One is then able to define a function px(x, t), giving the prob-
ability that at any instant t a particle will have an abscissa lying between x and
x + dx, and a function Pp", (Pz, t), supplying the probability that, at any instant t,
pz will have a value lying between pz and pz + dpz. It is also possible to define a
function p( x, Pz, t) such that p( x, Pz, t) dx dpz will furnish the probability for x and
pz to be found simultaneously within the respective intervals dx and dpz. One then

1 1
has
+00 +00
px(x, t) = -00 p(X,Pz, t) dpz, Pp,.{Pz, t) = -00 p(X,Pz, t) dx. (86)

One may also introduce the functions

(P)( t) _ p(X,Pz, t) and (X)( t) _ p(x,Pz, t)

Px X,Pz, - ( ) Pp X,Pz, - ()'
Px x '" Pp pz
208 The Characteristic Function

gIvmg, respectively, the distribution of abscissas for those particles which at the
instant t have a known Px and the distribution of linear momenta for those particles
which at the instant t passes through a known point x. In the case where an
enormous number of particles interact among themselves in such a manner that the
position and velocity of every particle at any moment depend on the positions and
motions of the other particles, one can envisage the value of the function p(x, Px, t)
at a given point X,Px as depending on the function's values in other points of the
space of x and Px' By contrast, this state of affairs is no longer conceivable if the
particles are noninteracting and affected only by external forces.
Now, instead of studying a uniform fluid consisting of a vast number of molecules,
let us consider a single particle undergoing stochastic motion. We can again define
the various p functions envisaged above, but this time it appears inadmissible for
the local value, in regard to x and Px, of the function p( x, Px, t) to depend on this
function's values in other points of the x, Px space, since these values, for x and
Px values other than those specifying the local instantaneous state of the particle,
correspond to hypotheses that are not realized and thus cannot influence the state
of the particle.
We know that the formalism of wave mechanics must apply to the case of a
single particle. In order for this fact to be compatible with the idea that the particle
describes a trajectory and consequently at every instant possess a definite position
and velocity, it would be necessary for the corresponding p(x,Px, t) not to depend
on the probabilities at the instant t of the values of x and px other than those under
consideration. But the Wigner-Bass density p(x,Px,t)

p(x,Px, t) = 21+
h
00
-00 exp (41l"i
-h Px Z ) 'ifJ*(x + z)'ifJ(x - z) dz, (87)

because of the integration over z occurring here, depends on all values of 'ifJ and
thus of 1'ifJ12. The value of p(x,Px,t) at any given point x therefore depends on the
position probability of the particle at all points of Ox.
This circumstance, which is quite analogous to the difficulty we encountered in
developing the pilot-wave theory (the motion of a particle depended on its initial
position probability at all points of space), appears to absolutely bar us from viewing
the Wigner-Bass function as a means for returning to the classical conception of
particle motion; and this is so independently of all the other obstacles that we
already know.

Note G. L. It is certain that Louis de Broglie would significantly have revised his
position, because these points of criticism extend in reality only to the pilot-wave
theory and not to the theory of the double solution. Indeed, it is only in con-
structing p(x,Px,t) with the help of a purely probabilistic 'ifJ that one arrives at the
inadmissible conclusion that "the local instantaneous state" of the particle depends
on "hypotheses that are not realized"; things would be different if one reasoned
about a singular physical wave. One could then say that the influences affecting
the region of singularity are quite simply those of the wave v, whose propagation
is modified by the different force fields and the obstacles acting on it and which
modifies, through the intervention of the wave, the guidance of the singular region.
 Case of Two Noncommuting Quantities 209
The double-solution theory thus replaces the inadmissable picture of a particle that
would be "guided" by a set of probabilities for the realization of events by a picture
of a singularity joined to a physical wave which in some way "palpates" the sur-
rounding space and transmits this information to the singularity in modifying its
guidance. Thus it happens, for example, that a singularity passing through one of
Young's slits "knows" about the existence of the other slit and, responding to this
fact, directs itself with greater probability towards one of the bright fringes than
towards a fringe that is dark, which explains interferences involving isolated quanta
(see Refs. II, 26, 29, and the article by F. Fer in Louis de Broglie, sa conception
du monde physique (ouvrage collectif) (Gauthier-Villars, Paris, 1973). Likewise,
when de Broglie complains that the pilot-wave theory makes the motion of a parti-
cle depend on "the initial position probability," this criticism looses its value if the
particle is incorporated into a wave that is physical and no longer probabilistic: The
particle's motion now depends on its own initial conditions, at the same time that
it depends on the initial conditions of the wave guiding its motion, and no longer
on a set of probabilities.

Some years ago Wehrle and Dedebant developed a "stochastic mechanics" based
on the properties of stochastic functions. The principles of this theory were presented
notably by J. Bass in an exquisite article entitled "Les fonctions aleatoires et leur
interpretation mecanique" [Revue scientifique 83, 3-20 (1945)].
In stochastic mechanics, one does not explicitly introduce the idea that a
"stochastic" particle at every moment has a position and a velocity; one introduces
only the probability densities for the coordinates and the velocity components in
viewing them as stochastic variables and assuming that velocity is a conveniently
defined stochastic derivative of position. This very interesting theory must certainly
be capable of application to the motion of fluids, when studied on the macroscopic
level, without one having to introduce any hypothesis that the molecular coordi-
nates are differentiable functions of the time, allowing velocities to be defined by
the operations of ordinary differentiation. But the theory does not seem applicable
to wave mechanics, because it assumes the validity of the ordinary statistical scheme
and, in particular, the existence of a simultaneous probability density p(x, v x ) per-
mitting the definition of conditional distributions, whereas we have seen that such
a scheme breaks down in wave mechanics.
At most one can apply the laws of stochastic mechanics to the motion of the
fictitious probability fluid of wave mechanics. But we have seen that the study
of this fictitious fluid, which allows one to follow the variations of the position
probability in the course of time, is not at all equivalent to wave mechanics, because
it explains neither the generalized principle of spectral decomposition nor the role
of measurement in wave mechanics.

Note G. L. We already know that de Broglie was able afterwards to reply to

this objection by reconstructing the theory of measurement and by distinguishing
in wave mechanics three kinds of probability-present, predicted, and hidden.

In his article, Bass defines the concept of a conditional velocity, given a point of
space, which, in the usual statistical scheme, is the mean particle velocity when the
210 The Characteristic Function
particle is known to occupy this particular position. Writing all formulas for the
linear case, one therefore postulates that Bass's conditional velocity in the point x
at the instant t is furnished by

vx(x,t) = jvxp<C)(vx,x,t)dvx, (88)

Since, in the usual statistical scheme,
(X)( )_p(vx,x,t)
PV Vx , x, t - ( )'
Px x,t
one also finds

(89)

If each particle were assumed to have a trajectory, vx(x, t) would be the mean
velocity of all the particles which at the instant t pass through the point x.

Note L. B. Denoting by Vg the guidance velocity and by Pg the corresponding

linear momentum, one has

j Pxp(x,px) dx dpx
pg = 1~12 '

Pg = jl~12pgdx= jdx jdpxpxp(x,px) = jPxlc(px)1 2 dPx,

Vg = jVxic(vx)12dvx,
The mean linear guidance momentum thus equals the mean value of the Px when
weighted by the factors Ic(px) 12. Consequently, if the c(px) are independent of time,
Pg will be conserved in guidance momentum space.
Note C. L. This note of the author, scribbled in pencil, is more important than
it appears: Indeed, formally it seems to do nothing other than repeat his own text,
but in reality we see here appearing for the first time the word "guidance," i.e., the
language belonging to the double-solution theory. In other words, for de Broglie the
wave is not fictitious, but rather physical. A significant conceptual step thus exists
between the text and this note.

Let us try to apply this idea to the probability fluid of wave mechanics, assuming
that Vx = Px/m and that p(x,Px, t) is given by Bass's formula. One then has
Px(x, t) = I~(x, t)12,

r+ (47ri
2 oo
p(x,Px, t) = h J-oo exp - } ; Px Z
)
~*(x + z)~(x - z) dz,
(90)
 Case of Two Noncommuting Quantities 211

which, by Eq. (89), leads to

~ ( )
Vx x,t = 1'l/J12
1 2
h Loojr
{+oo Px
m
(41!'i
exp -TPx Z
)_,.*( x + )'I/J(x - )dzdpx.
'I' Z Z (91)

But
Px exp ( - 4:i Pxz) = - d~ [4~i exp ( - 4:i Pxz) ] ,
whence it follows, through integration by parts ('I/J being zero at infinity), that

1 {+oo (41!'iJr
) d
pvx(x,t)=21!'im i_oo exp -TPx z d),¢*(x+z)'I/J(x-z)]dzdpx

1
= 21!'im
jri-oo
{+oo (41!'i
exp -T Px
)
Z
(92)

x ['I/J(x - z) d~ 'I/J*(x + z) - 'I/J*(x + z) ~ 'I/J(x - z)] dpx dz.

But

1
-00
+00 41!'i
exp (-TPxz)dpx =-4.
h
1!'Z
1+
-00
00 h h
ex p (-iuz)du=-4 21!'8(z)=-2 8(z),
1!'

and thus finally

(93)

We thus see that here Bass's velocity is the velocity of the fictitious probability
fluid (see the definition of the flux vector /, following Eq. (5), Chap. XII). It is this
velocity which the pilot-wave theory attempted to attribute to the particle itself.
Having introduced the conditional velocity V, Bass showed that, in a three-
dimensional fluid described by stochastic mechanics, one obtains the conservation
equation
~ + V . (pv) = O.
Applied to the probability fluid of wave mechanics, this relation expresses the
v
conservation of the probability fluid, because equals the velocity, as defined, of
this fictitious fluid. Stochastic mechanics appears to be adapted to the description
of the probability fluid of wave mechanics; but that, we know, does not remotely
suffice to account for the novel conceptions of wave mechanics. All these attempts
at classically interpreting wave mechanics always succeed in restricting the hydro-
dynamic aspect of this theory (corresponding to the motion of the probability fluid),
but by the same token they fail to capture the essence of the novel ideas of wave
mechanics.
212 The Chamcteristic Function
Note G. L. This is nothing! But de Broglie was still unaware that soon he would
demonstrate this fact himself, when he would come to understand that the entire
scheme under discussion must be constructed not on the probabilistic wave t/J but
on the physical wave v, and that he would then realize how to recover in a new form
(thanks to his theory of measurement) that which he here calls "the essence of the
novel ideas of wave mechanics," that is, the influence of the measuring apparatus
on the values of the observed quantities (Ref. II, 27).

Let us make a further remark. The formula defining vx ,

pVx = 1-00
+00 p
2.. p(x,Px,t)dpx,
m
(94)

with p = 1t/J12 and where p(x,Px, t) is the Wigner-Bass function, allows one to write

1-00+00pVx dx = 11+00 P
-00 2..m p(X,Px, t) dpx dx. (95)

But, as Px/m depends only on Px, the right-hand side of this equation is, ac-
cording to a property of the Wigner-Bass function demonstrated above, equal to
the quantum mean value of Px / m:

-Px = 1+00 t/J*(x,t)(PX)

m -00 m op
t/J(x,t)dx.

As a result, one gets

1+00
-00 pvx dx = 1+00 1+00 P 1+00 (p )
-00 dx [-00 ~ p(x,px, t) dpx] = -00 [t/J* ~ op t/J] dx.

We are therefore led to write (although this is not a rigorous consequence of the
preceding equation)
pvx(x, t) = t/J*(x, t) (Px)op t/J(x, t). (96)
m
In wave mechanics, the expression

t/J*(x, t) (Px)op t/J(x, t)

m
is often called the density of the mean value of Px/m, since one obtains the mean
value by integrating this quantity over space. According to Eq. (96), one would
have
~ h ./.* at/J
pvx = --2-'- 0/
1rzm x
-a '
which is not in agreement with the formula (93). But let us note that one could
equally well set

pvx(x,t) = t/J*(x,t) (Px)op t/J(x,t) + aa f(x), (97)

m x
 Case of Two Noncommuting Quantities 213

if f(±oo) 0, and that the two expressions obtained for pVx differ only by the
quantity
~~(1jJ1jJ*).
47rzm ox
They are therefore integrally equivalent, that is, they give the same result when
integrated over space.
In wave mechanics, what really enjoys physical significance are the integrals
defining mean values; the densities for these integrals are defined only up to a
divergence and have a fictitious character. This is a fact which I have stressed many
times in my expositions of wave mechanics. It is

which in wave mechanics has physical meaning, and not 1jJ* A1jJ (except in the case
A = 1, where 11jJ1 2 has a physical meaning). The components of the flux vector for
the probability fluid are defined by densities, which is why this fluid is only fictitious.

Note G. L. With the theory of the double solution, Louis de Broglie would soon
change his opinion and attribute a physical meaning to this fluid when defining it
with the aid of the wave v (Ref. 11,26). As often happens in theoretical physics, the
author here presents theories which he regards as "mathematically empty," but to
which he will give physical substance thanks to a novel physical interpretation. We
may recall how Faraday likewise introduced the field notion: He started by drawing
the force lines, but the theory was born the day he began to believe that these lines
depicted an underlying physical reality.

(e) Yvon's Theory. Jacques Yvon has done interesting work somewhat along the
same lines as that of Bass [Cahiers de physique theoriquei C. R. Acad. Sci. (Paris)
223, 311, 347 (1946)], which is very elegant from a mathematical point of view. I
have not studied this work in detail and shall quote from it only the following result,
which Yvon established no doubt in the secret hope of reducing the formalism of
wave mechanics to the usual statistical formalism:
Designate by () the operator

-0-
2

- -oxoPx'
()-

and let G denote a linear Hermitian operator corresponding to any measurable

quantity G. We then define

(98)

While herein G is expressed in terms of Xop and (Px)op, the quantity ,(x,Px) is a
function of the ordinary variables x and Px. Using Eq. (98), one can accordingly
associate a function ,(x,Px) with every operator G and thus with every observable.
Yvon's theorem then reads as follows:
214 The Characteristic Function
Theorem: The quantum mean value

G= 1 +00

-00 tjJ*(x)GtjJ(x) dx

can be obtained by constructing the mean value ofthe function ,(x,Px) correspond-
ing to G with the aid of the Wigner-Bass probability density. In other words,

We are going to directly verify this result, which Yvon obtained by skillful
transformations. G being defined by the foregoing formula, one gets

roo exp (ih)

-G = JrLoo 41r () [exp (21ri)
h pxX G exp (21ri)] 2
- h PxX h
(99)
x 1+00
-00
(41ri )
exp - h Px Z tjJ*(x + z)tjJ(x - z) dx dpx dz.

Integrations by parts with respect to x and Px then furnishes

rroo h2 exp (21ri

-G = Jr}}-oo ) [ (21ri
h Pxx G exp -hPxx
)]

(100)
- h Px Z) tjJ *(x
ih () ) [exp (41ri
x exp ( 41r + z)tjJ(x - z) ] dx dpx dz.

But herein

as one may easily verify by expanding the exponential operator in a series. Therefore,

- h Px Z) tjJ *(x
ih () ) [exp (41ri
exp ( 41r + z)tjJ(x - z) ] = exp (41ri)
- h PxZ tjJ(x - 2z)tjJ*(x)

and, consequently,

(102)
x exp [2:i Px(x - 2z)] tjJ(x - 2z)tjJ*(x) dx dpx dz.
 Case of Two Noncommuting Quantities 215

On introducing the new variables x, Px, and u = x - 2z, corresponding to

D(x,px, z) 1
=-,
D(x,px,u) 2

one can rewrite Eq. (102) as

G= (
h1J~i i-oo
roo 1jJ*(x) [G exp (27ri)]
-h PxX exp (27ri
h Px U )1jJ(u) dx dpx duo (103)

Now, if 1jJ(u) has the Fourier expansion

Fourier's inversion formula will give

C(Px) = h1 1+ 00
-00 exp (27ri
h Px ) 1jJ(u) duo
U

On substituting these formulas into Eq. (103), we are left with

G = j rLoo
roo 1jJ*(x)G exp (27ri)
-h PxX C(Px) dpx dx

1
(104)
+00

= -00 1jJ*(x)G1jJ(x) dXj q.e.d.

Yvon's result is very elegant and perhaps useful, but one cannot view it as
reducing wave mechanics to an interpretation of a classical kind. That G is calculable
if one knows 1jJ is already clear from the definition

G= k 1jJ*G1jJdr.

Yvon's formula furnishes another method for calculating this mean value, a method
that resembles the classical way of calculation when a true simultaneous probability
density p(x,px, t) is available. The function whose classical mean value needs to be
computed is not the function G(x, Px), classically corresponding to the quantity G,
but rather a function ,(x,Px) which one can calculate starting from the quantum
operator G[x, (Px)op).
Let us make a few remarks about this subject. The operator
216 The Characteristic Function

applied to a function only of x or only of Px simply reproduces the function itself,

because, for example,

[1 + -
ih -a2 + ... + (ih)n
- a2n + ... ] f( x) = f( x).
47r ax Bpx 47r ax n ap~
Therefore, if G is of the form f(x), a case which can occur in wave mechanics, or of
the form f(px), which never happens in wave mechanics, where px in the expression
for G is always an operator, one has

,(x,Px) = exp 47r ()

( ih ) [exp (27ri)
h Pxx G exp (27ri)]
-h pxx
(105)
f(x)
-G- {
f(px)
In these cases, ,(x,Px) becomes identical with G(x,px), and Yvon's mean takes
on a completely classical appearance. Things are different if G is a function both of x
and of px. Even if one may treat px in the expression for G as an ordinary variable
(which is not the case in wave mechanics), ,(x,Px) does not reduce to G(x,px),
except when G = f(x) + ~(Px). The disagreement between G and, gets worse if
px in the expression for G is an operator, as actually happens in wave mechanics.
One then gets once more involved in the discussions we had in subsection (c) III
connection with the Wigner-Bass density.
It is moreover easy to see that, if

then one does not in general find

as one had to if p( x, Px, t) were a true probability density for simultaneous values of
the variables x and PX' Consider, for example, the operator

h (a a) h (a
G= "21 [x(Px)op + (Px)opx] = - 47ri x ax + ax x = - 27rix ax +"21) ' (106)

which one obtains on symmetrizing the classical expression xpX' The operator G
thus defined is Hermitian and can pertain to a measurable quantity. We are now
going to calculate the function ,(x,Px) belonging to this G. First we note that

exp (h27ri) (27ri)

pxx G exp -T Pxx = xPx - 47ri
h
;
 Case of Two Noncommuting Quantities 217

consequently

(107)

ih h
= XPx - 411" - 411"i = xpx·
Consider now the operator G2 (which also is Hermitian). It is given by

G2 = -~
411"2
[x ~x ~ + x~
ox ox ox
+ ~];
4
(108)

hence it follows that

211"i ) 2 (211"i ) 2 2 h h2 (109)

exp ( -,;: Pxx G exp --,;: Pxx = Pxx - 2 211"i pxx - 1611"2

and, as a result,

2 2 ih 4 (ih) 2 h h ih h2 (110)
=p x --4pxx+- - -2-Pxx+2-----
x 411" 2 411" 211"i 211"i 411" 1611"2

2 2 h2
= Pxx + 1611"2·
Clearly, therefore, ,2(X,px) does not equal [,(x,Px)]2.
In brief, elegant and interesting as the researches we have just discussed are
from a mathematical point of view, they do not cause the fundamental differences
existing between the formalisms of wave mechanics and classical mechanics to dis-
appear. This difference in the formalisms stems from some very remarkable facts
which our study of quantum theory has revealed, such as the impossibility of simul-
taneously measuring two canonically conjugate (or, more generally, noncommuting)
quantities, the creation of a new state of probabilities and the effacement of phases
accompanying the act of measurement, and the interference of probabilities. These
items are without analogs in classical theories, and one cannot describe them by
adopting the usual statistical formalism, which views measurement (the trial) as a
simple recording.
[Any return to a classical description of microscopic phenomena seems impos-
sible. This is what the beautiful analyses of von Neumann, notably his theory of
m,easurement, have shown. We shall present these matters next.]
218 The Characteristic Function

Note G. L. For obvious reasons, the passage between brackets was later crossed
out by the author, who noted in pencil: "Bodiou's Thesis." The allusion here is to
Georges Bodiou; his Thesis (of 1949), for which Louis de Broglie was the rapporteur,
was entitled "Recherches sur les fondements du calcul quantique des probabilites
dans les cas purs." In regard to the text, there appears the following note, added no
doubt much later, which looks a bit sibylline, but which we shall actually explain
quite easily.

Note L. B.
Application of the Usual Statistical Scheme
to the Pilot- Wave Theory:

px(X) = 1"p(x)1 2 , "p = a exp (i<p );

h
pl;;)(vx) = 8(vx + 2:m :~), v = ---Vcp;
21rm

p(X,vx) = Px(x)p~)(vx) = 1"p(x) 12 8(v x + 2:m :~).

It is easily verified that

Moreover, one must require

Xi being the solution of

and

vVe also have

-vx(x, t) = f (+
Vx 8 Vx
h ~
-2-
1rm
OCP) dv x
uX
= ---h ~.
ocp
21rm uX
 Case of Two Noncommuting Quantities 219
Note G. L. Let us first point out that, except for a detail in writing, these
formulas are studied on pp. 88 and 93 of a work by the author entitled La theorie
de la mesure en mecanique ondulatoire (Ref. II, 27) and that they apply equally
well to the theory of the double solution as to that of the pilot wave. There it is a
question of hidden statistical scheme based on two postulates;

(1) The particle is supposed to be localized at every instant in a definite point

of the wave not known to us, but one assumes that the probability density for
registering the particle in a point x (i.e., to find the value x for the quantity X) is
given by j1,b(x) 12.

(2) If the particle occupies a point x, its velocity is supposed to equal

(-h/27rm)'V<p, where <p is the phase of the wave. This postulate is none other
than the guidance formula; in other words, the conditional probability, when X is
known to have the value x, that the velocity will be given by the guidance formula
is equal to unity.

These postulates render obvious the first two probability densities Px (x) and
A~r)(vx) written down by de Broglie; from them the other densities can be deduced
in accordance with the general scheme described earlier in the section entitled "Re-
calling some general concepts of probability calculus."
But let us not forget that we still are dealing only with a hidden scheme, because,
if the position probability is always directly verifiable (it is a present probability),
the probability density pertaining to the velocity, on the other hand, is in general
hidden; because, apart from special states, the guidance velocity is not observable
(it is a hidden parameter) and does not correspond to measurement results. It there-
fore remains to explain how one can deduce from this scheme the usual probabilistic
scheme of quantum mechanics for a measurement of momentum. This is what Louis
de Broglie later would do on the above-cited pages of his book on measurement.
 CHAPTER 14

THEORY OF MIXTURES
AND VON NEUMANN'S THEORY
OF MEASUREMENT l

1. MIXTURES AND PURE CASES

Let us first again take up some considerations regarding the interference of proba-
bilities. Let a large number.N of systems, all occupying the same state t/J, be given,
and let A be a measurable physical quantity having the eigenvalues ak and the
eigenfunctions <Pk. Then, if one has

any measurement of A will lead one to find that let 1

2.N systems has the eigenvalue
2
aI, IC21 .N systems has the eigenvalue a2, and so on.
The mean value of A will therefore be

Imagine now that instead of.N systems in the same state, we have let 1 2.N systems
2
in the state <PI, IC21 .N systems in the state <P2, etc., then measurement of A will give
1. See J. von Neumann, Mathematische Grundlagen der Quantenmechanik
(Springer, Berlin, 1932); American translation: Mathematical Foundations of Quan-
tum Mechanics (Princeton University Press, Princeton, 1955); French transla-
tion: Les fondements mathematiques de la mecanique quantique (Alcan, Paris,
1947). See also F. London and E. Bauer, La theorie de l'observation en mecanique
quantique (Actualites scientifiques et industrielles) (Hermann, Paris, 1939).
Finally, recall once more the work of Louis de Broglie on measurement (Ref. II,
27), where von Neumann's theory is presented and then carefully criticized, which is
not done here, where the presentation is still orthodox. However, we shall see some
criticism appear in some notes added afterwards; and, even in the initial text, the
outlines already emerge of some personal interpretations of the author, foreshadow-
ing his future theory. We shall point them out in passing.
 Mixtures and Pure Cases 221
the same results and the same mean value as in the first case. One could therefore
believe that the two situations are equivalent.
But they are not. Indeed, consider a measurable physical quantity B which does
not commute with A. The eigenfunctions of B will then not coincide with those of
Aj and, if Pie and Xle are the eigenvalues and eigenfunctions, respectively, of B, one
will have
'Pie = I:: dleIXI,
I
this expansion in general containing several terms.
Consider now the first case, where all N systems are in the same state

tP = I:: Ck'Pk = I:: CledklXI·

Ie 1e,I

Measurement of B on all these systems will

times produce the value PI; the mean value of B is thus given by

which can also be written as

since this equals

Consider now, by contrast, the second case, where Nicl 12 systems are in the
state 'PI, etc. Measurement of B on the first Nlct 12 systems will then give the
value PI for a fraction Idlll2 of these systems, etc. In total, the value PI of B is thus
obtained
NI:: ICle1 2 1dkl l2
Ie
times; consequently, the mean value of B is now furnished by

L kl dkll 2 PI =
IC 2 1 L hl Brie 2
k,1 k
222 Mixtures and Theory of Measurement

One thus sees that, for every quantity B not commuting with A, the two envis-
aged situations are entirely different: In the first case, there is interference between
the probabilities; in the second case such interference does not occur. One must
therefore not argue that the ./II systems in the state 1/; form what in probability
calculus is called a collective, comprising ./IIlcll2 systems having the value al for A,
etc. It is moreover evident that it would be just as legitimate to consider the ./II
systems as forming a collective consisting of ./IIld l l2 systems having the value (31 for
B, with
dl = LCkdkl,
k
etc.; and this second collective would not coincide with the first. It is therefore im-
possible to reduce the ensemble of./ll systems in the state 1/; to a well-determined col-
lective, since this collective would vary with the quantity A envisaged. As von Neu-
mann puts it, the state 1/; constitutes, a "pure case," which cannot be decomposed
into a "mixture" forming a collective in the usual sense of probability calculus.
On the other hand, one can very well imagine ./Ill systems having the wave
function 1/;(1), ./112 systems having the wave function 1/;(2), etc. The ensemble of
these systems forms a mixture of ./II pure cases corresponding to 1/;(1), ./112 pure
cases corresponding to 1/;(2), etc. We immediately recover our second case by taking
./Ill = ./IIlq 12, etc.
Let us put M/./II = Pi. Our mixture is then defined by the set of Pi, and
obviously

The Pi may be taken as the "statistical weights" of the different pure states 1/;(i).
On setting
Ck = v'Pk exp (iak)
in the example studied above, we see that the Pk = ICk 12 thus defined are the
statistical weights of the mixture equivalent, as far as the measurement of A is
concerned, to the pure state 1/;. But the mixture equivalent to the pure case 1/;
for measurements of a quantity B that does not commute with A would involve
statistical weights different from the preceding ones, and that is why one cannot
truly reduce a pure case to a mixture.
According to the example treated above, the mean value of B in a pure state

IS

with
 Mixtures and Pure Cases 223
If one replaces this pure case by the collective pertaining to the measurement of A,
the mean value of B is given by the expression

It is easy to see wherein these two mean values differ: If one sets Ck = ICk Iexp (iak),
with ak denoting the argument of the "phase" of Ck, the first mean value becomes

L ickllcd exp [i(a, - ak)]Bkl.

k,l

In a situation where the phases ak are completely unknown, one obtains the
probable value of the foregoing expression by taking its mean over the values of the
ak all assumed to be equally probable. One thus finds

i.e., the expression appropriate to the second case. In other words, one passes from
the first case (of pure 1jJ) to the second case (collective belonging to the measurement
of A) by supposing that any knowledge of the phases ak has been completely lost.
This fact is easily understood in light of the general ideas on the interpretation of
wave mechanics. Indeed, the first case corresponds to the direct measurement of
the quantity B on the pure state 1jJ; the second one corresponds to the measurement
of the quantity A followed by that of B, the measurement of A transforming the
state 1jJ into a mixture defined by the ICk 12. But we have seen that any measurement
of A completely wipes out the phase differences between the components C{)k of the
initial1jJ (i.e., the phase differences ak -a,), thereby causing any knowledge of these
phases to be lost entirely. This ties in well with the result obtained above.
We have thus obtained a clear idea of the difference existing between the pure
case, defined for a wave function 1jJ, and a mixture of pure cases, corresponding
to the wave functions 1jJ(1), 1jJ(2), etc., and defined by the statistical weights PI, P2,
etc. We have also seen that in general the measurement of a quantity on a pure
case transforms this pure case into a mixture of pure cases. While the notion of
a mixture, corresponding to the conception of a collective, is a classical one in
probability calculus, the pure case, which contains an infinity of collectives blended
together in a completely novel fashion, is a new idea, ignored by classical probability
calculus.
Let us add a word about the expression "pure case." It is sometimes said that
the state defined by a wave function 1jJ is a pure case for the quantity A if this 1jJ
reduces to an eigenfunction of A (1jJ = djC{)i, with Idil = 1); the quantity A then has
a definite value (viz., ai). But above we have spoken of a "pure case" as obtaining
when the probability situation is defined by a single function 1jJ without appeal to a
mixture of given statistical weights. The two definitions of "pure case" may appear
to be different, but in reality they can be reduced to one another. Indeed, when
the wave function of a system is 1jJ, there always exists a linear Hermitian operator
224 Mixtures and Theory of Measurement

A (and even an infinity of such operators) having '¢ as an eigenfunction; to this

operator there corresponds a measurable physical quantity whose value is fixed in
the state '¢ under consideration. One can see this by viewing '¢ as a vector in
Hilbert space and by noting that this vector can always be treated as a member of
an orthonormal basis set (even of an infinity of such sets). Every state defined by a
'¢ may therefore be viewed as a pure case for some suitably chosen quantity A; and
in this manner the two definitions of a "pure case" can be reconciled.

Note G. L. Today the author would undoubtedly have expressed himself more
cautiously: To every measurable quantity there certainly corresponds a Hermitian
operator; but it is difficult to maintain, as moreover everyone did, that conversely
a measurable quantity can be associated with every arbitrarily chosen operator.

2. VON NEUMANN'S STATISTICAL MATRIX FOR A PURE CASE

Let us first envisage a pure case defined by a given wave function '¢. One may
treat this function as a vector in Hilbert's function space. Moreover, if CPI, ••• , CPn, ..•
form a complete orthonormal set of basis functions (e.g., the eigenfunctions of a
linear Hermitian operator A), the CPi can be regarded as forming an orthogonal set
of unitary vectors in Hilbert space, and the expression

'¢ = LqCPk
k

will be analogous to the expansion of a vector in terms of its components in the

orthogonal directions defined by the unitary vectors. The Ck will be the components
of '¢ relative to the basis set of CPk. The Hilbert space under consideration is a
complex space, and the components Ck are in general complex.
Consider now two vectors in Hilbert space,

'¢ = LCkCPk
k
By definition, their scalar product is

('¢ . X) = f '¢*X dr = L Ck dl f CPkCPk dr

~ kJ ~
(1)
= L4
k,l
dl bkl = L4
k
d k,

where the third step follows from the orthonormality of the CPk. One has here the
generalization to the complex domain of the classical expression for a scalar product.
Note that ('¢. X) = (X· ,¢)*. The scalar product of a vector '¢ with itself, the analog
of the squared length of an ordinary vector, is called the norm N(,¢) of '¢:

(2)
 Von Neumann's Statistical Matrix for a Pure Case 225
If "p is normalized, N ("p) = 1 and

We already know all that.

In Hilbert space, an "operator" corresponds to an operation that transforms one
vector into another; thus X = A"p defines the operation which changes "p into X.
Accordingly
L dl 'PI = A L Ck'Pk,
I k
whence, on multiplication by 'Pj and integration over D, one gets

dj = LCk [ 'PjA'PkdT = Lajkck. (3)

k iD k

The matrix elements generated by A within the set of 'Pk are the coefficients of the
linear transformation which allows us to pass from the components of"p to those of
x·
Again taking "p to denote the wave function for the pure case, let us envisage
in Hilbert space the operation "projection on the vector "p." If the corresponding
operator is designated by P.p, it is evident that pJ
= P.p, and quite generally
PJ = P.p. All its powers being identical, the operator P.p is called "idempotent."
Let a complete set of orthonormal basis functions 'PI, ... ,'Pn, ... be given. Then
one can write an expansion of "p in the form

with

"p being normalized. We already noted that there exists an infinity of orthonormal
basis sets having "p as one of its vectors. In one of these sets, the function 'Pk will
have an expansion of the form

'Pk = d"p + "', with d= in "p*'Pk dT = ci·

The operator P.p, which is the "projector" on "p, is defined by the property
P.p'Pk = d"p = ck"p for all 'Pk.
The matrix generated by the operator Pt/J within the basis system of the 'Pk has
the mn element

(4)
226 Mixtures and Theory of Measurement

Thus the matrix Pt/J belonging to the pure case under consideration can be expressed
with the aid of the coefficients that characterize the expansion of 1/> in terms of
the chosen basis set. In this way we have defined what von Neumann called the
"statistical matrix" associated with the pure case 1/>. It is obviously Hermitian.
The statistical matrix possesses two fundamental properties:

(1) Its trace equals unity. Indeed

(5)
n n n

(2) It is "idempotent," that is, P; = Pt/J for integral n. Indeed, one has, for
example,
(PJ)mn = I:: cmc;cpc~ = cmc~ = (Pt/J)mn,
p
(6)

whence PJ = Pt/J and thus, by recurrence, P; = Pt/J.

Let now A denote a measurable quantity of the system under consideration. The
'Pk being arbitrary orthonormal basis functions (which here no longer necessarily are
eigenfunctions of A), we have seen that the mean value of A is given by

with Arl denoting the elements of the matrix generated by the operator A within
the set 'Pk and Ck signifying the components of 1/> with respect to the 'Pk. One can
also write
A = I::(pt/J)lkArl = Tr(Pt/JA) = Tr(APt/J). (7)
k,1

Consequently, knowledge of the statistical matrix furnishes a simple method for

calculating A.
The statistical matrix for a pure case is often called the "elementary" statistical
matrix (Einzelmatrix), in opposition to the more general statistical matrices which
we are going to define later for mixtures of pure cases.
An elementary statistical matrix can easily be reduced to diagonal form. To
show this, it suffices to choose our basis set such that the given 1/> is itself one of
the basis functions, e.g., such that 'PI = 1/>. The elementary matrix then takes the

1
form
o 0 o
o 0 o
o 0 o
 The Statistical Matrix for a Mixture of Pure Cases 227
Thus, all the diagonal elements are zero except one, which is unity, since (P",)nm =
cnc;;., whence
(P",)nm = 0 for n -f m,
(P",)nn = 0 for n -f 1,
(P",)l1 = ciq = 1,
since 1j; = q 'PI with ICII = 1. The trace, being an invariant under any change of the
basis system, must always equal 1; this is what here strikes one immediately. We
also see that pn = P.

3. THE STATISTICAL MATRIX FOR A MIXTURE OF PURE

CASES

We are now going to study a mixture of pure cases. Such a mixture we defined
above by considering N systems, of which N PI were in the state 1j;(1), N pz were in
the state 1j;(2), etc., with
LPk = 1.
k

But the idea of a mixture can also be introduced for a single system, namely if one
ignores the exact form of its 1j; and assumes as known only that the system has the
probabilities PI to be in the state 1j;(1), pz to be in the state 1j;(Z), ... , and Pn to be
in the state 1j;(n), with, naturally,

The unique state of the system is then represented by a mixture of pure cases
1j;(1), 1j;(2), . .. ,1j;(n), defined by the fractions PI, pz, .. . ,Pn'
Every pure case of the mixture has its own statistical matrix P",(k). We accord-
ingly attribute to the mixture a Hermitian statistical matrix
n
'"' (k) (k)* (8)
so that Pl m = L..k Pk c1 Cm ,
1

where the statistical weights Pk are positive numbers lying between 0 and 1, such
that n
LkPk = 1.
1

The c(k) are the components of the various 1j;(k) relative to the same basis system
'PI, ... , 'Pi, .... The statistical matrix thus appears as a superposition of elementary
statistical matrices.
228 Mixtures and Theory of Measurement

Note L. B. If one chooses as basis functions <pi the set of position eigenfunctions,
i.e., the 8(q - q'), then the c~k) will have the values 'IjJ{k) (q', t), as one sees from the
formula
'IjJ{k)(q, t) = j'IjJ{k)(q', t) 8(q - q') dq',

and thus n
P(q',q") = Lk 'IjJ{k)(q')'IjJ{k)*(q").
1

This is "Diraes statistical matrix."

The mean value of a measurable quantity A for the system is

n
A = Lk PkA.p(k) , (9)
1

A.p(k) being the mean value A would have if the system were in the pure state 'IjJ{k):

Since
(10)
one finds

(11 )

= Tr(PA) = Tr(AP),
i.e., the same formula as in the pure case.
As in the pure case, the statistical matrix for a mixture always has a trace equal
to unity:
n n
Tr P = LPmm = L LkPkC}!)C}!)* = LkPk L Ic}!) 12 = 1, (12)
m m I l m

sInce

is normalized.
On the other hand, while the elementary statistical matrix is idempotent, the
same is not true for the statistical matrix of a mixture. Indeed, we can demonstrate
that every idempotent statistical matrix is elementary. To that end, we assume the
equality p2 = P and write P in diagonal form, which is always possible. With Pi
 The Statistical Matrix for a Mixture of Pure Cases 229
denoting the ith diagonal element of P, the relation p2 = P requires that Pi = pr.
Thus the Pi are zero or equal to 1. The relation Tr P = 1, satisfied by every
statistical matrix, then shows that only one of the Pi is different from zero and
equal to 1. The system thus has a unique 'IjJ, which merges with one of the basis
functions of the representation that reduces P to diagonal form. Accordingly, the
necessary and sufficient condition for a statistical matrix to be idempotent is that
it be elementary.
Consider now the non-elementary statistical matrix for a mixture. If the
'IjJ(I) , 'IjJ(2), • •. ,'IjJ(n) defining the pure cases involved in the mixture were orthogonal
(which can happen by exception), one could take them as the first n basis functions
of an orthonormal set. Then c~) = Skm, since

reduces to 'Pk, and Pmn is zero for m =f n, while Pkk = Pk for k ~ n. The statistical
matrix therefore takes the diagonal form

PI 0 0 0 .... 0 ........ ...... ....

0 P2 0 0 ...... ...... .... 0 .........

0 0 P3 0 .............. 0 .........

0 .... .... ........................... (13)

.................... 0 Pn 0 0
.................... 0 0 0 0
.0 ................. 0 0 0 0

where the first n diagonal elements are equal to PI, ... ,Pn, while all the other ones
are zeros.
But this is an exceptional situation: In general, the 'IjJ(I) , 'IjJ(2) , ••• , 'IjJ(n) are not
orthogonal. One can nevertheless reduce the Hermitian matrix P to a diagonal
form, but the diagonal elements pi,' p~, etc., do not necessarily equal PbP2, etc.:

pi 0 0 o
0 p~ 0 o
P= 0 0 o (14)
p~

0 0 0

The matrix P being Hermitian, the Pk are real numbers. Moreover

230 Mixtures and Theory of Measurement

since the trace of the matrix equals 1. We now are going to show that the P~ cannot
be negative. For this purpose, let ek denote the components of a vector 3 in Hilbert
space, and consider the scalar product of 3 with P3. It has the value
n n
Le;;. LkPkC~)c~k)*en = L k Pk l(3 . 1jJ(k)) 12.
m,n 1 1

Since the squared modulus appearing here on the right is necessarily non-negative
and the same holds for the Pk, we see that (3. P3) is necessarily positive or zero.
But if P is put in diagonal form, this scalar product assumes the form

whence
(3. P3) = LP~leml2 ~ 0, (15)
m

and this must be true for any vector 3. Thus all the p'r,. are positive or zero. Since
moreover their sum equals 1, one has

O~p~~1.

From this we conclude that p'r,. _p/~ ~ 0, whence one finds, for any vector in Hilbert
space,
(16)
m

4. IRREDUCIBILITY OF PURE CASES

We now come to a very important theorem, which plays a big role in von Neu-
mann's theory (to be given further on), where he showed that it is impossible to
reduce the probabilistic character of wave mechanics to a hidden determinism. [Note
G. L. In a correction added in a different ink, the author replaced "showed" by
"tried to show."]
The important theorem in question establishes the truly special character of pure
cases. It reads:

Theorem: "It is impossible to represent a pure case in the form of a mixture."

Or, if one prefers: "A pure case is never reducible to a sum of pure cases."

Indeed, if this theorem were not true, it would be possible, at least in certain
cases, to have a relation of the form
 Irreducibility of Pure Cases 231
with P and the Qi denoting elementary statistical matrices, i.e., Hermitian idem-
potent matrices of trace 1, and where the ai are positive numbers such that

It would then follow that

(17)

because
L.aj = l - a i·
J
#i
Thus one would have

(18)

But p2 =P and Qr = Qi, whence

~
~ ..•
a'a'(Q'
J ' - Q.)2
J = 0 (19)
I,J
i>j

and therefore, since the ai are all positive,

(20)

But the square of a Hermitian matrix can be zero only if the matrix itself is zero;
because, if A is a Hermitian matrix, the elements of A 2 will be

(a 2 )ik = Lailalk = Lai/akl,

I I
232 Mixtures and Theory of Measurement

and thus, if the (a 2 )ii are zero, one must have

which requires that ail = 0 and thus A = O.

Q;-Qj being a Hermitian matrix, the condition (Qi-Qj)2 = 0 implies Qi = Qj,
i.e., all the Qi would be the same. Thus one would have

SInce
La; = 1.
Consequently, P is not truly a sum of elementary statistical matrices, contrary to
our initial hypothesis.
We have therefore proven that pure cases are irreducible and never can be re-
duced to a mixture of pure cases. The pure case of wave mechanics thus enjoys the
following two properties: (1) It is represented by an elementary statistical matrix,
whereas every mixture possesses a statistical matrix that is not elementary; (2) it
cannot by any means be reduced to a mixture of pure cases.

5. IMPOSSIBILITY OF REDUCING THE LAWS

OF WAVE MECHANICS TO A HIDDEN DETERMINISM
(VON NEUMANN)

We now come to the celebrated reasoning by which J. von Neumann [showed]

the impossibility of interpreting the probability laws of wave mechanics by assuming
the existence of an underlying determinism that escapes us. [Note G. L. De Broglie
later weakened this phrase, as he did above, by replacing "showed" by "tried to
show."] In classical physics, every time it becomes necessary to introduce prob-
abilities in place of rigorous laws, one always supposes that the phenomena are
subject to determinism, but that this determinism is very complicated or too subtle
for us to follow in detail; the only observable phenomena are those that are of a
statistical order and for this reason expressible by probabilities. One thus denies
the existence of a contingency in the philosophic sense, that is, of any absence of
determinism: Determinism is always assumed to reign in the final analysis, at the
deepest levels of reality. Only the phenomena at the macroscopic level directly per-
ceptible to our senses can appear to be governed by the laws of probability, chance
intervening uniquely as a result of our inability to follow the real determinism, which
is too fine and too complicated. This is the definition of chance one finds in the
writings of all pre-quantum philosophers of science, notably Henri Poincare.
The simplest example of a pseudo-statistical theory in classical physics is the
kinetic theory of gases. There one assumes that the gas molecules, and thus their
mutual collisions, are governed by the rigorous laws of classical mechanics, so that
 Impossibility of a Hidden Determinism 233
there is an underlying determinism. But the molecules are so numerous and their
motions so complicated that one can in no way follow this elementary determinism.
Moreover, the molecular motions themselves completely evade our senses; we can
perceive only their macroscopic effects, such as the pressure and temperature of the
gas, certain local density fluctuations, the Brownian motion of visible granules, etc.
These macroscopic phenomena, resulting from an enormous number of very com-
plicated elementary events, appear to us as arising from a statistical theory which
involves the intervention of probabilities. But this intervention of chance exists only
in appearance. Thus, for example, the disorganized motions of a Brownian gran-
ule would appear to us as governed by a rigorous determinism if we knew how to
calculate the motion of the molecules and their collisions with the granule.
Since this elimination of true chance has been successful in classical physics,
it is tempting to try the same approach in quantum physics. In wave mechanics,
one is faced with probability laws: Could we not suppose that they result from our
ignorance of a hidden determinism? If one were to succeed in this enterprise, one
would simultaneously have eliminated indeterminism and true contingency and have
maintained the classical conception of chance. If, on the contrary, one were to fail,
it would be necessary to assume indeterminism and contingency. In the first case,
wave mechanics would become a pseudo-statistical theory of the classical kind; in the
second case, it would, in the words of von Neumann, be a "truly statistical" theory.
Now, von Neumann developed a reasoning [which seems to show the impossibility] of
reducing the probability laws of wave mechanics to an underlying determinism: [The
question therefore appears to be settled in the second sense, and wave mechanics
appears as a truly statistical theory, irreconcilable with classical determinism.]

Note G. L. Louis de Broglie later weakened this statement by two correc-

tions: He replaced (in pencil) "which seems to show" by "which he believed able
to deduce"; he furthermore twice struck through the conclusion which we placed
between square brackets, once in ink and once in pencil. Viewing the manuscript as
a whole, it seems very probable that the corrections in pencil were done after those
in ink and that thus at least two successive readings were made; their order can be
guessed from the fact that all allusions about the double solution are in pencil. It
is therefore certain that this manuscript, which Louis de Broglie for many years did
not intend to publish, serves nevertheless as a reference on his reflections. Particular
passages of the manuscript have definitely been used in some of his other works.

To provide his demonstration, von Neumann starts from the following re-
marks: To assume an underlying determinism is to assume the existence of variables
whose exact values we do not know (hidden variables), such as the positions and ve-
locities of gas molecules, and of probabilities introduced as a result of our ignorance
regarding the exact values of these hidden parameters. In a deterministic theory of
hidden parameters, the real state of a gas, for example, is at every moment entirely
determined: All the molecules of the gas have well-determined positions and veloci-
ties; and if we knew the values of all these parameters, we would be able to represent
the state of the gas by a point in phase space. But we do not know the exact values
of the hidden parameters, and, to represent the global appearances accessible to
our senses, we envisage a "mixture" of elementary states with conveniently chosen
234 Mixtures and Theory of Measurement

statistical weights. A deterministic theory of hidden variables, as in the kinetic the-

ory of gases, therefore envisages a statistical mixture of elementary states, where all
quantities have perfectly determined values, these mixtures occurring because one
cannot observe the elementary states of perfectly determined quantities nor follow
their evolution in time. The elementary states forming the mixture are naturally
indecomposable; and they are also "dispersionless," because, for such a state~very
quantity A has a well-determined value such that A equals its mean value A and
that
(J'2 = {A - A)2 = A2 - { A)2
is zero, as are moreover all other differences An - {A)n.
Briefly, for a statistical theory to be reducible to a deterministic scheme of hid-
den variables, it is necessary for all the statistical distributions appearing in this
theory to reduce to mixtures of elementary states, indecomposable and dispersion-
less. Von Neumann showed that this is not the case for wave mechanics. [Note
G. L. Here, again the author later wrote "tried to show" in place of "showed."] For
that demonstration he relied on the following fundamental theorem:

Theorem: The states encountered in wave mechanics are never dispersionless.

To put it differently, for every realizable state in wave mechanics, one cannot for
every measurable quantity have A2 = {A)2.
To see this, we shall rely on the fact, demonstrated above, that in wave mechanics
every state (whether a mixture or a pure case) is characterized by a Hermitian
statistical matrix P of trace 1 such that the mean value of every quantity in the
state considered is given by
A = Tr (P A) = Tr (AP). (21 )
Therefore, if in wave mechanics a state were to be dispersionless, it would be nec-
essary for every quantity A that A2 = (A)2, that is,
Tr (P A2) = [Tr (P A)t (22)
Let then 'PI, . .. ,'Pi, ... be a set of orthonormal basis functions, and consider in
Hilbert space the operator which projects every vector of this space on a vector 'Pi.
This projector P<Pi is a Hermitian linear operator, and we may take A = P<Pi' If the
state is dispersionless, it will in particular be necessary that
Tr(PP;,) = [Tr pp<p,]2. (23)
However, P<Pi being a projector, P~i = P<Pi' whence
Tr PP<Pi = (Tr pp<py. (24)
But

(25)
 Impossibility of a Hidden Determinism 235
or
(26)

Thus, by Eq. (24), Pi; equals its square, which implies Pii = 1 or o. And this must
be true for all indices i, because one can reason in the same manner for all the P<pi.
But one might suppose that some Pi; are equal to 0 and some to 1. One could then
satisfy the relation
Z=Pii = 1

by assuming that all Pi; are zero except one.

The latter hypothesis must however be rejected, because one can in Hilbert space
vary in a continuous manner the set of orthonormal basis functions by an operation
corresponding to a rotation of the axes in this functional space. We can thus, by
a continuous operation, successively cause each of the primary axes to change into
one of the other axes. In the course of this continuous operation, each of the Pii
must vary continuously; which means, as they can have only the values 0 and 1,
that they will keep their initial values. Therefore, the Pii are all either equal to 1
or to O. But neither of these two solutions is acceptable, since the trace L:i Pi; of
P must equall, whereas it would be 0 for one solution and infinity for the other.
Thus there exists no acceptable statistical matrix P corresponding to the ab-
sence of dispersion for all quantities. This result was moreover to be foreseen, since
we already know that for a pure case (system having a well-determined wave func-
tion "p) the dispersions U x and up., of two canonically conjugate quantities cannot
simultaneously be zero (by reason of the theorem on dispersion which we learned as
uxupx ;::: (h/47r)).
We can therefore not reduce the probability distributions of wave mechanics
to mixtures of indecomposable states without dispersion. In quantum mechanics
there indeed exist indecomposable states (the pure cases), but these states are not
dispersionless. From this one can conclude, with von Neumann, that it is impossi-
ble to interpret the probability distributions of wave mechanics by the hypothesis
of an underlying determinism with hidden parameters. It is possible to reach this
conclusion by the study of pure cases only, but the general analysis of von Neu-
mann allows a more exact comparison with the probabilistic theories of classical
physics.

[As a result of his reflections on the foundations of the new mechanics, von Neu-
mann expressed himself very categorically against every possibility of returning to
the conceptions of classical determinism. For the information of the reader, we cite
the following sentences from his book Formalisme mathematique de la mecanique
quanti que.
"This status of the problem of determinism (von Neumann writes 'causality') in
modern physics can be summarized as follows: In macroscopic physics, no experi-
ment can prove its existence, because the causal order apparent in the macroscopic
world originates nowhere but in the law of large numbers; and it is completely inde-
pendent of whether or not the elementary processes, which are the true processes of
physics, follow causal laws or not. That macroscopically similar objects behave in
236 Mixtures and Theory of Measurement
the same way has little to do with determinism. These objects are in fact not really
identical, since the coordinates which fix the state of their atoms almost never co-
incide exactly, and the macroscopically observable phenomena result from averaging
over these coordinates.
It is only on the atomic level, in the elementary processes themselves, that the
question of determinism can really be put to the test. But on this level everything
speaks against determinism in the present state of our knowledge, because the only
formal theory near enough in accord with our experiences, and summarizing them, is
quantum mechanics, and it is in logical conflict with determinism . . " There remains
today no reason allowing us to claim the existence of determinism in nature: No
experience can furnish a proof of determinism, since macroscopic phenomena are,
by their very nature, incapable of supplying that, and the only theory compatible with
our knowledge of elementary phenomena leads us to reject determinism."
It is very difficult to maintain that this condemnation is entirely final, but one
must recognize that a return to the deterministic ideas of classical physics has become
extremely improbable.
It may seem that one could undermine these conceptions of von Neumann only
by demonstrating that the probability distributions predicted by the principles of wave
mechanics do not correspond to the facts. But today the precision of these probability
distributions has been proven by the study of innumerable phenomena on the atomic
level, so that it appears very difficult to question them.]

Note G. L. Probably at the time of a first rereading, Louis de Broglie crossed

out (in ink) his own conclusion, in which he accepted, although in modern terms,
the categorical conclusion of von Neumann. But he (later?) crossed out in pencil
all of the long passage in which he reported the conclusion of von Neumann himself.
It is opposite this suppressed text that he inserted quite a large leaf of paper, on
which appeared the following important note, constituting the first refutation of
von Neumann's theorem, which was subsequently developed in Refs. II, 27 and 29.
Let us add that the long citation from von Neumann does not follow Proca's French
translation (p. 223 of the edition mentioned above); it is rather Louis de Broglie's
personal translation.

Note L. B. (In ink, on a sheet inserted later into the text.) The existence
of the pilot-wave theory seems however to reveal a kind of flaw in the reasoning of
von Neumann. Indeed, the pilot-wave theory furnishes a causal interpretation of the
wave-mechanical probability laws by hidden variables, which, despite the difficulties
that it raises, has the merit that it exists, whereas the reasoning of von Neumann
claims to forbid such an interpretation. This reasoning therefore contains a weak
point, which I think is the following: Von Neumann assumes that all the probability
distributions exist in the same sense at every instant. It is different in the pilot-
wave theory: The particle would at each instant have a well-determined position
and momentum, which for us would be "hidden variables" with values that are
inaccessible to our measurements. Our ignorance of the position would then in a
totally classical manner introduce localization probabilities identical with the wave-
mechanical probabilities 11/J12. On the other hand, the components of the linear
momentum would have possible values whose variation in general would be very
 Impossibility of a Hidden Determinism 237
complicated and whose probability laws would be of no practical interest, since
these values are not measurable. The probability distribution for the components of
momentum given by wave mechanics (principle of spectral decomposition) would be
utilizable only after a measurement of linear momentum when one does not know
the result of the measurement. These two probability distributions, for localization
and for linear momentum, would therefore not simultaneously be valid, which fact
would cause the failure of von Neumann's reasoning.
Let us clarify this point. Assume an initial state

t/;(x) = L: c(Px) exp ( - 2:i Pxx) = a exp (irp)

p..,

is given. In the pilot-wave theory the particle has a real position x' with probability
1t/;(x')1 2 and a linear momentum

(orp)
h -
p x -- - -211" ox x=x' .

We have a classical collective (mixture) defined by the weights It/;(x')12 ,and for every
quantity A one gets A2 = (A)2 in the state defined by x = x'. If a measurement of Px
is carried out which has the effect of isolating the various components of the Fourier
expansion, one obtains another collective, formed by wave packets corresponding to
the various values of Px, with the weights Ic(Px)12. Von Neumann's error of reasoning
would then be that he wants to consider at the same time these two collectives, one
of which is valid in the initial state, the other only after the measurement of Px' The
same flaw would likewise prevent the recovery of the classical scheme if one adopts
for all quantities the wave-mechanical probability laws (see above).
The viewpoint of the pilot-wave theory leads one to attribute more importance to
the q representation than to that of p, which gives rise to interesting considerations
on the old question of the pre-existence of colors in a complex wave.
 CHAPTER 15

MEASUREMENT THEORY
IN WAVE MECHANICS

1. GENERALITIES

As we have seen, it is measurement that plays a fundamental role in quan-

tum indeterminism. According to present ideas, this role is in fact entirely dif-
ferent in microphysics from what it was in macrophysics according to the old
conceptions. In classical physics, measurement, at least when it is performed
with suitable care, is a simple "ascertainment," which gives precise form to our
knowledge about a real object without disturbing the latter in any way. The
real elementary states being supposedly perfectly determined, all ignorance on
our part is expressed by probabilities based on a mixture, with conveniently
chosen statistical weights for various elementary states; and measurements are
then capable of diminishing our ignorance or even eliminating it, in narrow-
ing down the mixture or even reducing it to a precise elementary statc. Thus,
in classical kinetic theory, an elementary state is defined by definite valucs of
the coordinates and velocities of all the molecules. But, as we cannot observe
these quantities, their exact values are unknown, and we are forced to argue
in terms of collectives corresponding to mixtures characterized by given statisti-
cal weights of the elementary states. However, if we were able at any given in-
stant to measure the position and velocity of every gas molecule (a measurement
which classical physics assumed in principle to be entirely possible, while recog-
nizing that in practice it was not realizable), then such a measurement would
sharpen our knowledge and reduce the mixture to an indecomposable elementary
state.
The probability distributions occurring in classical physics therefore always have
the character of mixtures, and a measurement (the trial of statisticians) augments
our knowledge in revealing the precise true value of a quantity as it would exist
objectively at the moment of measurement and without modifying it appreciably (if
the measurement is carried out properly).
Things are quite different in quantum theory. Here our maximum knowledge
about a system is realized when we can treat it as a pure case, that is, attribute to
it a well-determined 1f;. In this state of maximal knowledge, it is impossible for us
to specify the values of all quantities of the system. Indeed, these quantities have
only some possible values, that is, values capable of revealing themselves in an act of
measurement. And, if certain quantities can have only one possible value (which is
 Statistics of Two Interacting Systems 239

therefore well known), this can never happen simultaneously for all quantities; the
pure case never is dispersion-free for every quantity. Here measurement can there-
fore never furnish, for the purpose of representing the state of our system, anything
more precise than a new pure case, itself comprising dispersions for some quanti-
ties: Measurement augments our knowledge of certain quantities, but correlatively
it diminishes our knowledge of other quantities, in such a manner that our optimal
knowledge of the system always remains represented by a pure case with dispersion.
Moreover, measurement does not in any way increase our knowledge of the state of
a system prior to the measurement, but it creates a new state.
To properly fathom the role of measurement, it is now necessary to analyze how
this act is carried out. Measurement assumes essentially an interaction between
the system on which one performs a measurement and some measuring apparatus.
Moreover, the reading of this apparatus must express itself through a macroscopic
phenomenon, directly observable by our senses, such as the deflection of a needle
across a dial. Let us begin by analyzing more closely what occurs in the interaction
between system and measurement apparatus.

Note G. L. It would be a betrayal of Louis de Broglie's thoughts not immediately

to quote the following passage taken from his Etude critique (Ref. II, 29, p. 5):
"Finally, I should like again to emphasize the certainly exaggerated role which, in
analyses of the observation of microphysical entities, the measurement apparatus is
frequently made to play. Quite often, it is not the measuring apparatus in the proper
sense of the term that intervenes. When the impact of a photon or an electron on a
photographic plate produces a local blackening and one examines this darkening by
eye, where is the measuring instrument? ... One can nevertheless in certain cases
introduce a measuring apparatus: For example, one measures the blackening with
the aid of an instrument measuring the local opacity of the plate .... But in every
case the measuring instrument will not intervene until the end of the observable
localization process, when the chain reaction will sufficiently have amplified the
phenomenon for it to become measurable by a measuring instrument in the ordinary
sense of the term."
The "chain reaction" alludes to the analysis of the localization process that
appears at the very beginning of the cited book and commences as follows:
"The microscopic world, that is, physical reality at the atomic and molecular
level, eludes our direct observations. How then do we know this world? It seems
that one would have to respond to this question by saying: We know it only through
the mediation of 'observable corpuscular localizations,' that is, of phenomena where
a particle, acting on the microphysical level, by way of a chain reaction, triggers an
observable effect."

2. STATISTICS OF TWO INTERACTING SYSTEMS

Let two systems be given, one of which is the system being studied and the other
the measuring apparatus. Designate by x the set of variables characterizing the first
system, by y the set of variables pertaining to the second. Let furthermore Uk(X)
be the complete set of orthonormal eigenfunctions for the first system and vp(y) a
240 Measurement Theory in Wave Mechanics

similar set for the second system.

When the systems are isolated from one another (the initial state), their wave
functions '1/;1 and 'l/;n will evolve separately in conformity with the corresponding
wave equations, and one can set

'1/;1 = LCk(t)Uk(X), 'l/;n = Ldp(t)vp(y). (1)

k p

In particular, the system I under investigation, being assumed initially to be in a

pure state, will remain in a pure state. The total system formed by the two systems,
whose Hamiltonian is then the sum HI + HI! of the individual Hamiltonians, has
the wave function
W(X,y,t) = 'l/;I(X, t)'I/;n(Y, t) = LCk(t)dp(t)Uk(X)Vp(y),
k,p
It represents a pure case of the combined system, which will last as long as the
interaction between the parts of this system has not started.
The measurement will commence at the same time as does the interaction of the
two systems. At this moment the Hamiltonian HI + Hn of the overall system will
be augmented by an interaction term Hi which depends on the coordinates x and y
of the two systems in a form that is not solely additive.
The function W of the total system will now cease to be a sum of products of
any CkUk(X) with any dpvp(y). But, since the products Uk(X)Vp(Y) form a complete
orthonormal basis set for the variables x and y, one can write

W(X,y,t) = LCkp(t)Uk(X)Vp(y), (2)

k,p
where the Ckp are however no longer of the form Ckdp. We keep on having for the
total system a wave W that evolves in agreement with a wave equation: The state
of this system thus constantly remains a pure case. The corresponding elementary
statistical matrix is
(3)
Note that two indices are required to represent a state of the overall system.
We now turn our attention to the state of the system I and envisage a particular
quantity A belonging to this system such that

Akl = r uk(x)Avj(x) dx.

JD 1

The mean value of A during the interaction is

A = jw* Aw dx dy = L CkpC/u jU'kAUldx jv;VUdy

kp,iu
(4)
= L CkpCluAklbpu = L CkpCipAkl.
kp,/r> kip
 Statistics of Two Intemcting Systems 241
But the statistical matrix of the system I must during the interaction be such that

(5)
which implies that
(PI)lk = L ClpCiw (6)
p

Similarly, the statistical matrix of the system II is given by

(PII)"p = L Ck"Ciw (7)

k

The statistical matrix P of the total system is Hermitian, of unit trace, and
idempotent, as one may easily verify by taking the orthonormal character of the
product Uk(X)Vp(Y) into account: It is an elementary matrix. The situation is dif-
ferent for PI and PIli these matrices are Hermitian of trace 1, but they are not
idempotent. The statistics of systems I and II considered separately are there-
fore no longer those of pure cases but rather those of mixtures. (It is important
to note that the definition of these mixtures is obtained by considering mean val-
ues.)
To specify the composition of these mixtures, let us return to the formula

W(X,y, t) =L CkpUk(X)Vp(Y). (8)

k,p

For a given value of p, the coefficient of vp(y) in this expansion is

L CkpUk(X), with L ICkp l 2 = l.

k k~

One may therefore say that, for a given value of p, i.e., a given state of the system II,
the system I has a probability proportional to ICkpl2 to be in the state k. In absolute
value this probability will be proportional to Ick p) 12 if one puts

(9)

so that

One can then write

(qTJ) lk = ~ C(p)C(p)*
~Pp k I ' (10)
p
242 Measurement Theory in Wave Mechanics

with

Likewise, one finds

(11)

on putting
and (12)

The matrices PI and Pn are thus seen to define mixtures with the statistical
weights Pk and Pp, respectively.
Therefore, while the total system remains a pure case despite the interaction,
each of the systems considered separately is transformed by the interaction from
a pure case into a mixture. Thus, while knowledge of the overall system remains
maximal, that of the two component systems ceases to be maximal, a circumstance
which has no analog in classical physics. The system can be viewed as finding itself
in a pure case unknown to us; the mixture represents this ignorance. A simple
ascertainment will then suffice to remove this ignorance by informing us which case
is actually realized.
Let us further clarify matters by studying the form of the statistical matrices
PI and Pn . One sees that for each system the mixture is determined by the states
of the other system; this is expressed in Eq. (10) by the fact that the sum on the
right is taken over an index p pertaining to the second system. It is therefore in
ascertaining the state of the second system (i.e., the value of p actually realized)
that we are able to say which pure case is to be attributed to the first system. But it
must be possible to ascertain the state of the second system without disturbing this
state, that is, by a macroscopic ascertainment analogous to the one whose existence
is postulated in classical physics. This is an important point to which we shall
return.
It is obvious that a knowledge of W for the combined system I + II would tell
us more about this system than would the partial statistics relative to one or the
other of the component systems, expressed by PI and Pn . The function W, or the
corresponding elementary matrix P, contains statistical correlations between sys-
tems I and II that the matrices PI and Pn, even when considered simultaneously,
do not contain. In particular, one sees from the expressions for PI and Pn that
knowing these quantities is not equivalent to knowing the individual Ckp' Thus, in
being content to consider PI and Pn , we surrender part of the information that a
knowledge of W would furnish us. This loss of knowledge corresponds to the sum-
mation over the index p in the expression for PI, a summation which disregards
what one might know about the state p of the system II and about its connec-
tion with the system 1. The loss of knowledge thus suffered manifests itself in
the appearance of probabilities in the classical physics sense, that is, probabilities
stemming from a lack of knowledge and expressing themselves by way of a mix-
ture.
 Correlation Coefficients in Interaction 243

3. CORRELATION COEFFICIENTS IN THE INTERACTION

BETWEEN TWO QUANTUM SYSTEMS

Assume again that I and II are two interacting systems, and let A(1) and B(2)
be two quantities attached respectively to each of these systems. Denote further by
o:~1) and <,O~1) the eigenvalues and eigenfunctions, respectively, of A(1) and by ~?)
and X~2) the corresponding properties of B(2).
Before the interaction, one has
IT! = 2:>k<,O~1) 2: dIX~2) = 2: ckdl<'oi1\~2), (13)
k I k,l
with

The quantity A(1)B(2) _A(1) .B(2), which appears in the numerator ofthe expression
defining the correlation coefficient

of A(1) and B(2), here has the value

2: hdd2o:i1)~?) - (2: I kl 2o:i1) 2: Idd2) (2: Idd2~i2) 2: h1 2)

C (14)
k,l k I I k
or
2: I k1 2O:11) 2: Idd2~f2) - 2: ICk1 2O:11) 2: IdI12~f2) = o.
C (15)
k I k I
The correlation coefficient is zero, as was to be expected.
After the interaction, the wave function of the overall system I + II is
IT! = 2: Ckl<,O~1)X~2), (16)
k,l
with

k,l
One accordingly finds:

A(1) = jlT!* A(1) IT! dX1dx2

= 2: CkPk'l' j<,011)* A(1)<'o1~)dx1 jx?)*xf;)dx 2

(17)
kl,k'l'

= 2: ICki 12 o:i1) ,
k,l
244 Measurement Theory in Wave Mechanics

(18)
=L I C kl1 213;2) ,

k,l

A(1)B(2) = jW*A(1)B(2)WdXldX2

= L CkPk'I' jcp~l)* A(1)cp~~)dxl jX~2)* B{2)X~;)dx2 (19)

kl,k'l'

= L ICkIl2a~1) f3j2),
k,l

and therefore

A{1)B(2) - A{1)· B(2) = L ICkd2 [a~l)f3?) - a~1) L ICmnI2f3~2)], (20)

k,l m,n

a quantity which in general is different from 0, so that r #- 0; a correlation has thus

been established by the interaction.

4. MEASUREMENT OF A QUANTITY

We have studied the interaction of two systems, the second of which served as a
measuring apparatus. But to be able to measure a quantity of the first system, it is
necessary that the result of the interaction be of a particular kind. In other words,
an arbitrary interaction will not serve to measure a well-determined quantity of the
first system. Indeed, we have seen that, in (macroscopically) ascertaining the state
of the second system after the measurement, one can infer that the first system finds
itself in a particular pure state. But, since in a pure case a physical quantity does
not in general have a precise value, we shall not generally succeed in measuring the
quantity that interests us.
Suppose A is a physical quantity of the first system that we would like to measure.
As the basis functions Uk (x) of the first system, let us take the eigenfunctions of
A themselves. For the measuring device to be capable of measuring A, there must
exist a quantity B of the second system (e.g., the pointer reading of the measuring
instrument) such that, with vp(Y) denoting the eigenfunctions of B, W of the total
system will after the interaction have the form

W= L CkpUk(X)Vp(y), with Ckp = Ckfikp, (21)

k,p
 Measurement of a Quantity 245
that is,
\II = L CkUk(X)Vk(Y). (22)
k

A one-to-one relationship can then be established between the vp and the Uk or, if
one wishes, between the pointer reading of the measuring instrument and the value
of A. Let us calculate Pr under this hypothesis. We have

(23)

(whence C~k) = 1 and C~p) = 0 if P =f k) and

Pp = L IClpl2 = L 181pCI1 2 = ICpI2.

I I

It follows that

(24)

so that Pr is a diagonal matrix whose diagonal elements are the ICk 12. It is easily
seen that Pn is identical with Pr.
One will therefore have a mixture of states, each one of which is characterized by
a particular value of ak and a value of f3k related to it in a one-to-one manner, the
probability of any pair of corresponding values ak, f3k being ICk 12. The ascertain-
ment of the value f3k of B (a pointer reading, say) then allows us to attribute the
value ak to A; there really is a measurement. This ascertainment, which we assume
to be macroscopically possible (by direct perception or by recording), sharpens our
knowledge of A by revealing which value of A is actually realized in the mixture
produced by the interaction.
Let us examine under which conditions the hypothesis advanced about the form
of \II can actually be realized. Suppose that, before the measurement, system II is
in the state vo(y) and system I in the state Uk(X); then the wave function of the
total system will have the initial form

\II (x, y) = VO(Y)Uk(X), (25)

The hypothesis put forth on the final form of \II will be fulfilled if, at the end of
the interaction process, one finds, whatever eigenfunction Uk( x) is initially realized,
that
(26)
246 Measurement Theory in Wave Mechanics

Vk(Y) being an eigenfunction of B that stands in a one-to-one relation to Uk(X).

Indeed, because of the linear character of the wave equation, if the initial state,
instead of being presented by W = VO(Y)Uk(X), is given by the superposition

W(X,y) = LCkVO(Y)Uk(X),
k

t.hen the wave function will at the end of the interaction have the form

W(X,y) = L CkUk(X)Vk(Y), (27)

k

and the measurement of A will be possible. We shall further on develop an example

where this is indeed what happens.
It is essential that one be able to ascertain the final state of the measuring ap-
paratus by a macroscopic procedure, such as the direct reading of a pointer position
on a dial, the arrival of an electron at an electron counter, triggering a macroscopic
electric phenomenon, etc. This will also be seen in the example to be studied.
During a measurement, the evolution of the W for the overall system unfolds
continuously, this system remaining in a pure state, while the state of each partial
system becomes a well-defined mixture. A break occurs in the continuity of this
evolution, and a new situation is created, when the observer ascertaining the state
of the system II is able to attribute to system I a wave function corresponding to
a well-defined value of the quantity A. One sees that it is the consciousness of the
observer which, in ascertaining the state of the measuring apparatus, allows the
reduction of the mixture belonging to the state of the system under investigation
and resulting from the interaction, to one of these terms.
This very tricky question concerning the intervention of the observer's conscious-
ness was discussed notably by von Neumann. I reproduce here a summary of this
discussion given by F. London and E. Bauer in their fascicle of Actualites scien-
tifiques et industrielles (No. 775; Hermann, Paris, 1939).
Consider the set of three systems, viz., the object under study (x), the measuring
apparatus (y), and the observer (z), envisaged as forming a unique combined system,
which we describe by a global wave [unction

(28)

If the combined system is taken as the object of our study, we have a persistent
pure case for this overall system and a mixture for each of its constituent systems.
The function I}! provides the "maximal" description of the overall system without
one knowing exactly the state of the object x.
But the observer has an entirely different point of view, because for him it is
only the object x and the apparatus Y that belong to the external objective world.
From his perspective, he enjoys a completely special position, because he possesses
consciousness, or the faculty of introspection, which enables him to directly know
his state. In virtue of this immediate knowledge, the observer believes he has the
right to create his own objectivity-by severing the chain of statistical correlations
 Measurement of a Quantity 247
expressed by the \][ and by declaring: "I am in the state Wk, therefore the measuring
apparatus is in the state Vk and the object in the state Uk, which leads me to attribute
a well-determined value to the quantity A for which Uk is an eigenfunction, and thus
to a measurement of A."
"It is therefore not," write Bauer and London, "a mysterious interaction between
the apparatus and the object that causes a new \][ for the system to appear during
the measurement. It is only the consciousness of an 'I' who separates himself from
the earlier function \][ (x, y, z) and, by virtue of his conscious observation, constructs
a new objectivity in henceforth attributing a new wave function Uk(X) to the object."

Note G. L. This is how, some years later, Louis de Broglie, commented on the
same quote:
"I have cited this text, but I do not understand it very well: 'This I who separates
himself from the wave function' seems to me much more mysterious than would be
an interaction between the object and the measuring apparatus. One understands
why Schrodinger was able to say with an ironic play on words: 'The theory of the
wave 'Ij! has become psychological.' It does not help much to add that these con-
siderations lend support to Bohr's opinion, according to which in quantum physics
one cannot any more draw an exact boundary between the object and the subject,
because this statement, too, is rather incomprehensible and clarifies nothing. The
more one reflects on the matter, the more one gets the impression that this entire
interpretation is to be reconsidered on another basis."

It should be noted that the quantum uncertainty does not derive from an uncer-
tainty about the state of the observer: Even if one assumes that the observer can in
a perfectly precise manner keep track of his own state, not commiting, for example,
any personal error in the estimation of the pointer reading, the quantum uncertain-
ties, which are connected with the impossibility of finding a set of eigenfunctions
Uk{X) common to two noncommuting quantities, will always remain. We clearly see
wherein lies the difference between the quantum uncertainty and a simple error of
observation.
One can easily explain the profound and difficult character of the interpreta-
tion of the measurement process thus obtained. The interpretation raises all sort
of problems of a philosophical nature which have not at all been studied well: It
demonstrates notably the fact, underlined many times by Bohr, how difficult it is
in quantum theory to exactly draw the boundary between the objective and sub-
jective domains, since the object is in a way created by a conscious act of the
observer.
One of the fine points of this interpretation is the following: If objective reality
is created by a conscious act of the observer, will this reality not vary from one
observer to the next? This is certainly not the case, since otherwise all collective
science, all science common to man would be impossible. It should be remarked in
this connection that the ascertainment permitting a measurement is a macroscopic
procedure which does not modify the state of the observed object: Thus reading the
position of a pointer on a dial is a macroscopic act, and the observer's glancing at
the dial in order to make this reading does not influence the system. Consequently
nothing forbids other observers from repeating the same reading; and it is an em-
248 Measurement Theory in Wave Mechanics

pirical fact that, within experimental error, all observers get the same reading. It is
this finding that allows us to disregard the personality of the observer and to create
a science preserving an objective character. In summary, in the mixture resulting
from the interaction of the studied object with the measuring apparatus there is one
possibility and only one that reveals itself as realized for all observers.
Let us again follow London and Bauer. In classical physics, one imagines a sys-
tem as possessing at every instant, in an unequivocal and continuous manner, a
set of properties defined by measurable quantities having well-defined values. It is
precisely the possibility of establishing this continuity and this precision in the defi-
nition of the quantities characterizing the system that has in general been regarded
as a proof that physics studies a "real object" whose properties are independent of
the observer. In quantum physics, by contrast, an object is the carrier not of well-
defined quantities but only of a set of potential statistical distributions belonging to
measurable quantities; and each of these distributions comes into effect only at the
time that the corresponding quantity is measured. If one leaves all measurements
aside, it makes no sense to regard the properties attached to an object as having
definite values: We have seen, notably by the general arguments of von Neumann,
that the mathematical form of the distributions does not permit such a view (since
no state of the object can be dispersion-free for all quantities).
But all this does not prevent us from predicting or interpreting the results of
experiments. Quantum theory teaches us how to obtain a pure state for any quan-
tity, that is, a state where repeated measurements of the quantity in question will
give the same result. It also tells how one must make the measurements to verify
the theoretical predictions. It adapts itself marvelously to questions posed by ex-
periment, and it remains silent on those so-called questions that experiment cannot
settle.
In quantum microphysics, the notion of objectivity is less strict than in classical
physics. One can no longer define it by the possibility of constantly attributing to
quantities belonging to an object values that are well-determined, but only by the
fact that the corresponding properties are present at the time of the experiment,
that one can measure the values of the quantities in agreement with the statistical
predictions of the theory, and that, at the time of a measurement, the ascertainments
are the same for all observers. In the transition to macroscopic phenomena, one
naturally recovers the classical conception of objectivity as a limiting notion on the
large scale.

Note L. B. In all of the preceding, we have followed the customary ideas adopted
for more than 20 years on the interpretation of wave mechanics, following notably the
work of von Neumann and the exposition of London and Bauer, and we have therein
pushed the development to its extreme limits. It is certain that all these ideas, so
strange to old physicists and so difficult to properly clarify because of their often
too "philosophical" appearance, impose themselves the moment one admits that
knowledge of the wave '1/1, whose importance as an element of statistical prediction
is not in doubt, exhausts all one can know about physical reality, and that the
probability distributions deducible from '1/1, by application of the general principles
governing the probabilistic interpretation of wave mechanics, all exist potentially in
the state defined by this function. In the present state of our knowledge, one thus
 Example of a Measuring Experiment 249

obtains a theory whose observable consequences all appear to be rigorously exact

and capable of answering all factual questions that physicists could pose. But, in
reducing the description of reality to the use of a quantity 1jJ which only provides
a representation of probabilities that vary according to the state of knowledge of
the physicist employing it, one arrives inevitably at a "subjective" interpretation
that rejects more or less implicitly the objectivity of the physical world, lending
itself in a somewhat regrettable manner to philosophical verbalism, and is forced to
consider the agreement of observations among various observers as a "pre-established
harmony" whose origin remains mysterious.
Certain scientists, notably Einstein, have refused to consider the present attitude
of quantum physics as final. With the ClIid of often disturbing arguments, Einstein
has maintained that the use of the function 1jJ with a statistical character, although
leading to results that are certainly exact, does not constitute a complete description
of physical reality and provides only the statistical aspect of an objective reality
which we do not yet know how to clearly specify. A more complete description, of
the kind desired by Einstein, could perhaps be furnished by our pilot-wave theory (in
the form of the double solution), thus leading quantum physics back to conceptions
that are more objective and more precise. The future will tell.

Without dwelling any further on the study of these questions, which could involve
big developments, we are going to make clear, with the aid of an example, the nature
of experiments that allow the measurement of a quantity.

5. EXAMPLE OF A MEASURING EXPERIMENT

We have seen that any measuring process must satisfy the following three con-
ditions: (1) It must include an interaction between the object and a measuring
apparatus. (2) It must, after its completion, leave the system "object + measuring
instrument" in a final state where a one-to-one correlation exists between the values
of the quantity to be measured and the states of the measuring instrument. (3) It
must permit an observer to become aware, through a macroscopic ascertainment, of
the final state of the measuring instrument.
The type of measurement analyzed notably by von Neumann is the determi-
nation of the magnetic moment of an atom by the method of Stern and Gerlach.
This measurement consists in observing the result of passing an atom through a
nonuniform magnetic field. One seeks to determine the value of the component of
the magnetic moment in the field direction. Here the coordinates y of the center of
mass of the atom play the role of the pointer reading in the general discussion, if
these coordinates can still be determined by a macroscopic ascertainment.

Note G. L. One sees that Louis de Broglie still retains the traditional language
which he would criticize some years later, where everything is expressed in terms
of the measuring apparatus, in a form that is quite vague and artificial, because
it focuses too much on the act of recording, whereas he would later show that
the essential act of the experimenter, the one that truly disturbs the system, is
the prepamtion, which consists in splitting up the initial wave into separated wave
250 Measurement Theory in Wave Mechanics

trains with the help of a spectral analyzer, chosen to suit the quantity that one is
seeking to measure. The measurement proper then reduces to the ascertainment of
the particle's presence in one of the wave trains issuing from the analyzer.

The internal coordinates of the atom constitute the coordinates x of the ob-
ject. Let the magnetic moment of the atom in the field direction be represented
by an operator M(x,px); then the wave equation for the problem will have the
form

herein Ho denotes the Hamiltonian of the atom in the absence of a field after the
center-or-mass variables have been separated off; -(h 2 /87r 2 m)6. y corresponds to the
kinetic energy of the center of mass; and (M, F) stands for the energy due to the
interaction of the magnetic field with the magnetic moment of the atom. As long
as the magnetic field F is constant, i.e., independent of y, the motion of the center
of mass will separate from the internal state of the atom, and the latter is given by
the eigenvalue equation

(29)

Suppose that the atom is in its state of minimum energy. If this state is degen-
erate, it follows from the quantum theory of the Zeeman effect that

kf-l
Ek = Eo + -. F, (30)
J

where Eo denotes the energy of the ground state in the absence of a field, f-l the
magnetic moment of the atom, j its total angular momentum (spin included) in
units of h/27r, and k the magnetic quantum number, which is capable of assuming
the values -j, -j + 1, ... ,j -l,j.
If the magnetic field F is not constant, the energy eigenvalues are

(31)

They depend on the local value of F, and the same is true for the corresponding
eigenfunctions, which should be written as Uk (x, y). But practically the perturbation
of the Uk by the local value of the field is so weak that one can continue to write
Uk(X).
Viewed as a function of x, the wave function can be expanded in terms of the
Uk(X), which allows us to write

1/;(x, y, t) = 2.:>k(y, t)Uk(X). (32)

k
 Example of a Measuring Experiment 251
On introducing this expansion into the wave equation and noticing that [Ho +
(M, F)]Uk = EkUk, one finds
h2 ] h aVk(Y, t)
[- 81l"2m !:!..y + Ek(Y) Vk(Y, t) = 21l"i at . (33)

This wave equation describes the motion only of the center of mass, where Ek(Y)
plays the role of a potential. From the formula (31) for Ek(Y), it is seen that, for
each value of k, a different deflection of the electron beam will result, where this
deflection can occur in the same direction as the gradient of the magnetic field or in
the opposite direction, depending on the sign of k; and it may even be zero if k can
take the value 0 (for integral j). If, for example (for j = 1), k can assume the values
-1,0, +1, the initial beam of electrons will end up splitting into three branches, as
shown in the figure below.

-d-k=-l

----......"""
k=O

"'k=+l

Let vo(y) be the wave function that represents the motion of the center of mass
before the measurement and

L Icd;;: 1,
k

the internal wave function of the atom. Initially the 1/J of the system will therefore
have the form
1/J(x,y) = vo(y) 2..:
CkUk(X)
k
as it enters the field. We have just seen that, if the initial 1/J were of the form
VO(Y)Uk(X), then the function Vk(Y, t) would during the process of measurement
evolve, starting from vo(y), in a manner that is determined by a knowledge of the
index k. By virtue of the linearity of the wave equation, the wave of the system
will, on starting from the initial form

have the final form

Il1(X,y) = L CkVk(Y)Uk(X),
k
so that a one-to-one correspondence is realized between the values of the quantity
M and the final motion of the atomic center of mass.
252 Measurement Theory in Wave Mechanics

Now we have seen that the center of mass of the atom, on exiting from the
field, finds itself in distinct regions of space depending on the value of k. It is
therefore possible to ascertain macroscopically which value of k is realized for the
center of mass of the atom, and from this datum one can infer the value of the
atom's magnetic moment. Suppose, for example, that k = 0, ±1; on leaving the
field, the atom then has three possible directions, as shown in the figure above. If
now a screen is erected perpendicular to the initial direction of the atomic beam,
one may pierce three holes in it, corresponding to the three possible points of impact
of the atoms. By placing behind each opening a device that can be triggered by
the arrival of an atom (for example, a charge counter if the atoms bear a charge),
one becomes able to macroscopically ascertain which counter is being activated
and thus which of the three Vk is realized when the atom exits from the magnetic
field; from this knowledge, in turn, the magnetic moment of the atom after the
measurement can be deduced. One may naturally replace the counting instruments
by a simple photographic plate placed on the screen, because each sensitive region
of the photographic plate can be considered as a counting device capable of giving
rise to a macroscopic ascertainment.
Staying with the case of the three values 0, ±1 of k, we may note that, before
the measurement, the coefficients Ckp of the wave function for the overall system
x + y have the values

(34)

(k numbering the columns and p the lines of the array), while after the measurement
orie has

(35)

Before the measurement, the statistical matrices PI and Pn correspond to pure

cases and have the forms
(Pr)lk = L ClpCk p,
p

(Pn)PO" = L CkO"Ckp,
k
so that

Pr =
C-lC~l
COC~l
C-lC~
CoC~
C-lCi
CoCi
1,
ClC~l clq ClCi
 Diverse Remarks on Measurement 253
After passage through the nonhomogeneous magnetic field, they take the form

(37)

which clearly shows the one-to-one connection established between the Uk and the
Vk by the interaction between the variables x and y.
In regard to the measurement example above, the following interesting remark
can be made. If one sends through the Stern-Gerlach type of apparatus not one atom
but a flood of identical particles, it becomes possible to collect behind each of the
openings in the device described above a host of atoms all having magnetic moments
of the same orientation. Thus the apparatus now serves to prepare pure cases for
the component of the magnetic moment in the field direction, and that without any
ascertainment on the part of the observer being called for. This fact may appear
contradictory after what has been said about the importance of ascertainment by
the observer. But it must be noted that here it is not a question of attributing a
value of the magnetic moment to a particular atom, so that "measurement" of the
magnetic moment of an individual atom now does not occur. One obtains fluxes of
atoms having the same magnetic orientation: The Stern-Gerlach type of apparatus,
playing the role of a filter, gives pure cases in an anonymous form; whereas any
measurement made on one specific object requires a macroscopic ascertainment.

6. DIVERSE REMARKS ON MEASUREMENT

We have seen that, in order for an apparatus to serve as a measuring device, it

is necessary for the wave function of the system "object + measuring device" after
the interaction to be of the form
tjJ(X,y) = LCkUk(X)Vk(Y).
k

We now resume the reasoning that allowed us to define the mixture resulting
from an interaction (Sec. 2). Let A be an operator corresponding to a quantity
attached to some object. If this object were isolated and if

tjJ(X) = LCkUk(X),
k

then A would have the mean value

A= rtjJ*AtjJdx = Lcicl
JD k,l
ruiAu,dx = LciclAkl,
JD k,l
(38)

corresponding to the elementary statistical matrix

P'k = ckC" so that A = Tr (PA). (39)
254 Measurement Theory in Wave Mechanics

One also has Tr P = 1 and pn = p.

However, in the case under consideration, following the interaction, one finds

(40)

= 2::>kCiAk/<5kl = L Ick\2 Akk,

k,1 k

corresponding to the non-elementary statistical matrix

(41 )

We again have Tr Pr = 1 but not pl

= Pr.
If one compares the two expressions above for Jr, viz.,

LCkCiAkl and
k,l
then it is seen that the second sum results from the first when only the diagonal
terms of the first are retained. It follows that, while the first expression depends on
the phase differences between the q, the second does not; or, the second expression
is obtained if one starts from the first and forms its average over all phase values,
assuming these values to be equally probable. It is thus again found that knowledge
of the phases is lost in measurement; and we clearly see how this loss is related to
the transformation of a pure case into a mixture by the interaction involved in a
measurement.

Note G. L. The author returns at length to this process in his work on mea-
surement (Ref. II, 27), where he shows that the effacement of phases results from
the recording of the particle's presence in one of the wave trains issuing from the
spectral analyzer.

In discussing the Stern-Gerlach type of measurement, we remarked that it does

not seem precise to say (as London and Bauer did) that the position of the atom's
center of mass plays the same role as the pointer of a measuring instrument, because
the location of the center of mass is not directly ascertainable by our senses: What
is ascertainable is a macroscopic effect (discharge current through a counter, black-
ening of a photographic plate, etc.) caused by the localization of the atom. In
a more general fashion, every measuring process appears in the final analysis to
comprise a macroscopic effect due to the localization of a microphysical element.
For example, if we consider the collision of a particle 1 with a particle 2, each of
 Diverse Remarks on Measurement 255

which initially belongs to a narrow beam of waves "p, there will be an entire series
of possible outcomes of this process, corresponding to various exchanges of energy
and linear momentum, and the motion of the two particles in their final state are
"correlated." [Note G. L. These remarks are fundamental, because they foreshadow
de Broglie's subsequent theory of measurement.]
If, after the collision, particle 1 finds itself in beam 1, particle 2 will be present in
beam 2; if particle 1 finds itself in beam I', particle 2 will be present in beam 2'; etc.
If particle 2 gets to be localized in a region R of space by causing a macroscopically
ascertainable phenomenon, one can from this event infer the state of motion of
particle 1, a state that will be "correlated" with the state that the observation has
permitted one to attribute to particle 2. Here particle 2 serves indeed to measure
the motion of 1, the localization of particle 2 in R completely changing the state
of 2 without reacting on the state of 1 other than in allowing us to say which, in
the mixture representing the final state of 2, is the constituent of the mixture that
reveals itself as being realized. But what plays the role of the pointer in a measuring
instrument is not the position of the particle 2, which itself is unobservable; it
rather is the macroscopic phenomenon, ascertainable by our senses, that produces
the localization of particle 2 in the region R.
A complete theory of measurement in microphysics must therefore not be content
to analyze the evolution of the system of microphysical particles that get to interact
during the measurement; it must also relate this evolution to some phenomenon that
is directly ascertainable through our senses, a phenomenon which in some sense plays
the role of an instrument pointer. How is measurement theory able to do this? We
shall dwell on this point, which generally is left aside in the usual expositions of the
theory.
Let us start by posing the following question: How can one go from the micro-
physical quantities envisaged by wave mechanics to quantities directly observable
through our senses? The answer is that this transition takes place through the in-
tervention of mean values. Let S be a microphysical system and A a measurable
quantity attached to it. A can have a whole series of quantized values whose mean
value (mathematical expectation) is A. But the value of A is not directly ascer-
tainable by our senses. However, if, instead of dealing with a single system S, we
consider an enormous number .N of identical systems S, we are able to directly
observe (at least in favorable cases) the macroscopic quantity Amacr = .N A. The
256 Measurement Theory in Wave Mechanics

localization of one microphysical particle, unobservable by itself, can be accompa-

nied by an interaction with a very large number of elementary systems, causing the
appearance of a macroscopic observable quantity of the preceding kind.
Consider, for example, the wave 'IjJ associated with an electron striking the sur-
face of a plane photographic plate. A priori, the electron can be found with equal
probability (if the incident wave is plane) at every point of the photographic plate.
Suddenly it will produce a macroscopically observable photographic impression at a
particular point of the sensitive layer: The electron [will be localized] at this point
of the plate. In reality, the "point" in question of the sensitive layer is a very
small region containing enormously many atoms of the emulsion, and the [localiza-
tion] of the electron within this tiny region is accompanied by the ionization of a
very large number of atoms populating this region, which process initiates a local
chemical reaction that leaves a local impression in the sensitive layer. This con-
stitutes the macroscopically ascertainable event caused by the localization of the
electron.

Note G. L. The author later replaced "will be localized" by "will manifest its
presence." This is a correction heavy with meaning, because the first expression
implies that the electron, potentially present in every point of the wave, became
localized by the act of observation (Bohr's theory), whereas the second expression
connotes that the particle was already situated at a certain point of the wave and
that the act of observation only served to reveal its presence (theory of de Broglie).

Note G. L. The author later replaced "localization" by "presence," with evi-

dently the same intention as in the preceding correction; the two corrections are in
pencil and probably made at the time of the same rereading.

We shall now try to analyze the problem by the wave-mechanical method.

In the figure, let P be the plane of the photographic plate, and let the arrow
indicate the direction in which the variables increase in value. For the incident
electron, the variable will be denoted by x. Imagine that the sensitive layer is
subdivided into very small regions numbered 1,2, etc., and let Yi be the variable
defining the position of the ith region in question. It may seem that an infinity of
regions should be employed and that Yi would thus be a continuous variable; but
 Diverse Remarks on Measurement 257

physically one perhaps approaches reality more closely by treating the variables Yi
as discontinuous; in any event, this choice facilitates our calculations. Every region
located in the sensitive layer by the value of its Yi contains an enormous number
of atoms of the emulsion. We designate by the symbol Zi the set of coordinates,
enormous in number, describing these atoms and assume that the normal state of
the region i is defined by a wave function v~i)(Zi). The wave function of the global
system "incident electron + photographic plate" will then have the initial form

tPo(x, Zi) = uo(x)val)(ZI)V~2)(Z2)'" van)(zn),

if there are n regions, where uo( x) designates the wave function describing the initial
state of the electron (or rather the values of this wave function on the plate); it will
therefore be a constant if the incident wave tP of the electron is planar and normally
incident on the plate; and it will equal exp (-ikr)/r if the electrons are emitted
with a given energy and uniformly in all directions by a point source, r denoting
the distance of points on the plate from this source.
The interaction of the electron with the sensitive layer causes the wave function
of the system to assume the form
.1. _
0/ -
~
L.J Uo (y,. ) u{;( X - y,.) VI(i)( ZS.) Vo(1)()
zl· .. Vo(i-l)(.)
Z,-l Vo(1+1)(.) zn· (42)
z.+I· .. Vo(n)()

Indeed, if for example the electron becomes localized in the region i (a supposition
leading us to attribute the individual wave function D(X - yd to it), the electron
will produce complex ionization phenomena in the atoms of this region, which will
give to the wave function of the region i the form v~i)(Zi) corresponding to this state
of ionization. But the same event can occur in anyone of the n regions, with a
probability which for the region i is given by IUO(Yi) 12 , according to the interference
principle
J ItPI2dxdZI ... dZn = 2;::I U O(Yi)1 2 = 1.
•
Let now Bj be a measurable quantity (but not one that is directly accessible to
our senses) belonging to one of the microphysical elements of the jth region, such
as the component PVx of the current density vector associated with one of those
electrons of the jth region that have been detached from atoms by the process of
ionization.
By its general definition, the mean value of Bj is

Bj = JtP* BjtP dxdz 1 ... dZn = luo(Yj)1 2 (Bj)11 + Eluo(Yi)I\Bj)oo, (43)

if.j

as one easily sees on taking the normalization of the vail into account and employing
the usual definitions
258 Measurement Theory in Wave Mechanics

The observable effect (the ionization current in the x direction if Bj = pV x , etc.)

will be proportional to B i' Indeed, since everyone of the regions 1, 2, ... , n contains
an abundance of similar microphysical systems (atoms, electrons, etc.), the effect of
the localization of the atom in the jth region will be proportional to Bj, and this
mean value will therefore be macroscopically ascertainable.
One may suppose that (Bj )00, the mean value of Bj in the jth region when
the latter is in its initial state, is zero. By contrast, (Bj)11, relating to the ionized
state caused by the action of the electron, is different from zero. Accordingly, if it
is ascertained macroscopically that the mean value Bj =I- 0, one will know that the
localization of the electron has taken place in the jth region. One will thus measure
the x coordinate of the electron with the aid of a macroscopically ascertainable
impression occurring in a small region of the photographic plate.
If we take the index i to correspond to the function h(x - Yi) and the index
nj (which can assume the values 0 and 1) to the wave function v(i)(Zj) of the jth
region, the coefficient of the product h(x - Yi)v~l) (zI) ... v~i) (Zi) ... in the expansion
of 1/1 is seen to be
(45)
whence
1

Lnl, ... ,nn ICinl ... nn 12 = IUO(Yi) 12. (46)

o
This is the weight of the state h(x - Yi) relating to the electron in the mixture that
corresponds to the given 1/1 after interaction of the electron with the photographic
plate. The transformation, due to the interaction, of the initial pure case for the
electron into a mixture with weights IUO(Yi)12 corresponds to the possibility of fixing
the localization of the particle by a macroscopic ascertainment that concerns the
mean value of a quantity Bj; one thus better understands why the definition of a
mixture realized by interaction with a measuring device must be closely linked to
that of the mean values.

Note L. B. (scribbled in pencil). The importance of localization. Superiority of

the q representation.

Note L. B. (on a separate leaf). The considerations we have just developed

appear to prove that every measurement amounts to a localization, that is, to an
ascertainment of the presence of a particle in a small region of space. It follows
that, while a localization is a direct measurement, every other measurement (e.g.,
that of a linear momentum) is an indirect one resulting from the ascertainment of
some localization.
To us this appears to show that, in agreement with our conceptions about the
pilot-wave theory, the position probability density 11/1(x)12 has a more immediate sig-
nificance than the other probability densities introduced by wave mechanics, which
amounts to saying that the p representation, for example, has a less direct physical
significance than the q representation.
 Thermodynamical Considerations (von Neumann) 259

Note G. L. The second of the foregoing notes clearly seems to come after the
first and indicates at least a third rereading of the text. The idea expressed here
would be developed at length by the author: See Refs. II, 27, p. 81; II, 29, p. 51; II,
33, pp. 21 and 164. The fact that he speaks here of the pilot wave and not of the
double solution is without importance, because the reasoning is the same except for
the interpretation of the wave.

7. THERMODYNAMICAL CONSIDERATIONS (VON NEUMANN)

With the aid of an operator P, we have succeeded in representing the statistical

aspect of an ensemble of identical systems distributed (in the classical sense) over
various states (in the wave mechanical sense). This makes it probable that there
exists a connection between the matrix P and the quantities envisaged by statistical
thermodynamics. John von Neumann studied this type of question in the last part
of his celebrated book. This study is difficult and would without doubt require
deeper examination. Watanabe, in his Thesis, made an effort in this direction, but
here is certainly a subject which lends itself to further research. We shall be content
to givc, with Bauer and London, a very summary idea of the question.
We first recall some points of classical statistical mechanics. Boltzmann estab-
lished the relation
S = k log P, (47)
betwecn the entropy of an ensemble and the probability of its state under consider-
ation, where k, Boltzmann's constant, has the value 1.37 x 10- 16 in c.g.s. units.
This fundamental relation has been confirmed by an enormous number of veri-
fications of its consequences. The value of k can be deduced from the fact that, in
setting up the theory of an ideal gas with the aid of Boltzmann's formula, one dis-
covers that k must numerically equal the gas constant R for one gram-mole divided
by A vagadro's number; thus k turns out to be the gas constant per molecule:
R 8.31 x 10 7 -1
k = N = 6 X 10 23 erg deg .

If we now consider an ensemble of N systems distributed over a given number

of states (in the classical sense), in such a manner that ni systems occupy the state
i, one easily finds that the probability of this distribution is
N! N!
P= I I =--.
n1· n 2···· Uni!

Since N and the ni are assumed to be very large, Stirling's formula gives for N! and
nil the approximate values
N! ~ e-NN N ,

from which one concludes that

log N! ~ N log N - N, log nil ~ ni log ni - ni,
260 Measurement Theory in Wave Mechanics
which gives

log P = log N! - L log nil ~ N log N - Lni log ni. (48)

On defining Pi = ni/N, so that Pi specifies the fraction of systems in the state i

(or the weight of the state i in the statistical distribution), one obtains

log P =-N L Pi log Pi + N log N - N L Pi log N

(49)
= -NLP; log Pi,

sInce

Consequently, according to Boltzmann's formula (47),

(50)

which is the classical entropy formula of statistical thermodynamics.

If we now want to construct a quantum thermodynamics, it becomes necessary
to modify the definition of entropy and employ the phrase "state of a system" in
its wave mechanical sense, that is, define every state by a wave function. If the
various states of the N systems correspond to wave functions tpl,' .. ,tpj, ... forming
a complete orthonormal set (eigenfunctions of a measurable quantity, for example),
then the matrix P will reduce to its diagonal form when the tpi are taken as basis
functions, that is, one will have Pkl = Pklikl with

Moreover, the matrix log P, whose elements are the logarithms of the elements of P,
will then also have a diagonal form, with the diagonal elements (log P)kk = log Pk.
It is therefore suggested that we define entropy, starting from the statistical matrix
P, by setting
S = -kN Tr (P log P), (51)
because this expression, which has a value independent of the choice of basis func-
tions (as a result of the invariance of the trace under a change of basis), assumes
the form
S = -kNLPk log Pk, (52)
k

when written in the basis that diagonalizes P and log P, so that the earlier definition
(50) is recovered.
 Thermodynamical Considerations (von Neumann) 261
We now look for the maximum value of the entropy when the number N of
systems and the total energy E are held fixed. For this purpose, let us first recall
the corresponding classical calculation. We ask for the maximum P while keeping
both Nand E constant. One must therefore demand
SP = 0, with oN = °
and oE = 0,
which, by the method of Lagrange multipliers, leads one to write
O(P - aN - (3E) = 0, (53)
that is,
(54)

for all variations of Pi. Consequently,

Pi = exp (-1 - a - (3Ei), (55)
which is the classical law of Boltzmann and Gibbs. The factor exp -a can be
eliminated by an appeal to

resulting in
exp (-(3Ei)
(56)
Pi = 'L,exp(-(3Ek)'
k
If one employs this expression for Pi in the theory of an ideal gas, it will be
found that (3 = 1/ kT, with k denoting Boltzmann's constant and T the absolute
temperature, which is assumed to be well defined for the ensemble of N systems.
One then easily gets
S = -kN LPi log Pi
i

[
'L, Ei exp ( - E;f kT) 1 (57)
= kN log ~ exp ( - E;f kT) + kT ---=-=~:-e-x-p
"'" 1 i -:-(--E--i-:-:/k:--T:-:-)-

for the entropy of the most probable distribution. On defining

Z((3) = L exp (-(3Ek),
k
Planck's "sum over states," we can then write, among other things,

S = kN [log Z _ (3 a l;~ Z] , (58)

"'"
E = N ~PiEi = -N
a log Z
a(3 , (59)
I

F = E - T S = -kNT log Z. (60)

262 Measurement Theory in Wave Mechanics
In quantum thermodynamics, the calculations proceed completely in parallel.
One must require that -Nk Tr (P log P) be maximal subject to the conditions
Tr P = 1 and E = N E = N Tr (P H), with H denoting the Hamiltonian matrix of
anyone of the systems. It is therefore necessary to write

8LPkk log Pkk = 0, with 8LPkk = 0 and 8 LPk/Hlk = 0, (61)

k k k,l
from where, on introducing the Lagrange multipliers a and 13, one obtains

8 L Pkk log Pkk + aD L Pkk + 138 L PklHlk = 0, (62)

k k k,l
that is
L 8Pkk[log Pkk + 1 + a + f3 Hkkl + Lf38Pk/ Hlk = 0 (63)
k k,l
for all 8Pk/. It is therefore necessary that the systems occupy energy eigenstates
(8Pk/ = 0 for k =f 1) and, moreover, that
(64)
with
exp (-1 - a) L exp (-f3Hkk) = 1,
k

since

One concludes that

P = exp (-f3H)
Z(f3) , (65)

on setting
Z = Tr exp (-f3H).
As in classical theory, it can be demonstrated that 13 = 1/ kTj and the entropy,
energy, and free energy of the most probable distribution are found to be:

_ kN -PH
S - Z(f3) Tr [e (f3H + log Z)l

= kN [log Z _ 13 8 l;~ Z], (66)

E = N Tr (PH) = -N 8 log Z (67)

813 '
F = E - T S = -kNT log Z. (68)
 Reversible and Irreversible Evolutions 263
One thus recovers the same formulas as in classical statistical thermodynamics, but
with a different definition of Z. The form (65) of P moreover reveals that the weight
of the stationary state 'ljJk, of energy Ek = Hkk> in a mixture is

PH = exp (-f3 E k) ,
2::exp (-f3EL)

which is exactly the Boltzmann-Gibbs canonical law (56).

8. REVERSIBLE AND IRREVERSIBLE EVOLUTIONS

The foregoing considerations lead us to distinguish two types of evolution on the

microphysical level: the reversible ones, which occur between two measurements,
and the irreversible ones, which result from measurements.
The reversible evolution of a system or of an ensemble of systems is represented
by the evolution, in an entirely deterministic manner, of the wave function belonging
to the system or of the wave functions belonging to the systems of the ensemble. If
one is dealing with a pure case, and if 'ljJo(O) is the initial form of its wave function,
the latter will evolve in accordance with the law
h a'ljJ
27riat = H'ljJ, (69)

where II is the Hamiltonian operator of the system, which, the system being assumed
isolated, is independent of the time. One therefore has

(70)

with
27ri ) ' " 1 (27ri ) n
exp ( TtH = ~n! TtH

The pure initial case accordingly remains a pure case. Furthermore, as can easily be
verified, the adjoint of the operator exp (27rij h )tH is exp -(27rij h )tH, i.e., identical
with its inverse. The exponential operator in (70) thus is a unitary operator, which
has the property of conserving the norm of 'ljJ, as it should. Since unitary transfor-
mations also conserve the trace of any matrix, the entropy S = -Nk Tr (P log P)
of an ensemble of N systems in the state 'ljJ remains the same in the course of the
evolution; that is, the process is reversible.
Consider now a mixture of pure cases. Everyone of the wave functions 'ljJ(k)(t)
characterizing its constituent pure cases evolves in accordance with the wave equa-
tions
~ a'ljJ(k) _ (k)
27ri at - H'ljJ ,
264 Measurement Theory in Wave Mechanics

H being the Hamiltonian operator of the identical systems under consideration.

This entirely determinate evolution is therefore represented by

(71 )

that is, by a unitary transformation. The evolution of the statistical matrix is

consequently given by
P(t) = ~PkP..p(k)(tl (72)
k
or, if
(73)

by
(P(t))lm = ~Pkc~k)(t)c~l*(t). (74)
k

The trace of P, furnished by

P = ~Pk ~ c}kl(t)c}kl*(t),
k I

is conserved by the unitary transformation of !fJk, which conserves

1!fJ(k)(t) 12 = ~lcikl(t)12.
I

The entropy S of the mixture therefore remains the same, i.e., the evolution of the
mixture is reversible.
The irreversible transformations corresponding to processes not subject to de-
terminism take place at the moment when the previously analyzed measurement
interactions occur. The interaction of the system 1 under study with the measuring
apparatus 2 corresponds to a determinate and reversible evolution of the overall
state of the combined system 1 plus 2, until the macroscopic ascertainment by the
observer of the individual state of system 2 comes to interrupt this evolution by
attributing to system 1 a new state by way of a process which is neither measurable
nor [even] causal [in the present interpretation].

Note G. L. The words "even" and "in the present interpretation" were added
afterwards.

The entropy of the system 1 in its initial state is

S = -k Tr(P log P) = -k ~Pj log Pi. (75)

 Reversible and Irreversible Evolutions 265

If this state is a pure case, all the Pi are zero except one, which is unity. S is
therefore zero and remains zero while the system 1 is isolated and evolves reversibly.
If afterwards the interaction with system 2 transforms the state of system 1 into
a mixture, all the Pi will become less than 1, causing all the log Pi to be negative
and the entropy to clearly become positive; that is, the process is irreversible, the
measurement being accompanied by an augmentation of the entropy. The chain of
reversible evolution is broken: As have already been indicated, one cannot by any
means go back from the state following the measurement to the one that preceded
it.
If the initial state is already a mixture, it can be demonstrated that every mea-
surement which actually modifies this mixture has the effect of augmenting the
entropy. The demonstration is quite long; it can be found in von Neumann's book
(p. 260 et seq. of the French edition).

Note L. B. (after the text). Interpretation of the increase in von Neumann's

entropy due to the loss of our knowledge regarding phase differences that accom-
panies measurement, a loss which constitutes a decrease of information and thus of
negentropy.

Note G. L. This note dates from the time of the full development of the work of
Shannon and of Brillouin, which Louis de Broglie followed very closely. He returned
to the idea expressed here in La theorie de la mesure en mecanique ondulatoire
(Ref. II, 27, p. 119).

Therefore, every measurement that modifies the existing situation has the effect
of augmenting the entropy. This conclusion is satisfying, because it corresponds to
the "irreversible" character of the measurement, which has the effect of creating a
novel situation such that one cannot by any means recover the previous situation.
Since the notion of time flow in a definite direction itself, an experimental fact of
the first rank, is connected with the irreversibility of physical phenomena and the
concomitant augmentation of entropy, one sees that time must play a particular
role in wave mechanics by reason of the role played therein by measurement, a
role ignored by classical theories and in particular by the theory of relativity. This
circumstance can throw some light on the fact that wave mechanics and relativ-
ity theory do not treat time in the same manner: Relativity theory assigns to the
time coordinate a role that is totally symmetric with that of the space coordinates,
without explicitly taking the well-defined direction of time flow into account; while
wave mechanics treats time as a parameter which is not linked to any statistics
analogous to the statistics governing the spatial coordinates, and it makes the mea-
surement performed at a particular instant playa role that is in an essential manner
connected with the irreversibility of time. We shall not further dwell on these ques-
tions, which are most interesting but very tricky [and which may in future take on
different appearances, depending on the evolution of the ideas regarding causality in
wave mechanics.]

Note G. L. The phrase in brackets was added afterwards.

266 Measurement Theory in Wave Mechanics

The somewhat microscopic definition of entropy at which we arrived is quite

different from the one employed in classical thermodynamics: Indeed, it makes the
increase of entropy result from the accomplishment of the measuring process, while
measurement plays no particular role in the classical theory of entropy. It is therefore
necessary to see how these two conceptions of entropy can be reconciled. This was
done by von Neumann in his book (p. 273 et seq. of the French edition), where he
introduced the operators, all commuting among themselves, that correspond to the
macroscopic observations. This theory, which was deepened notably by Watanabe
in his doctoral thesis of 1935, is quite subtle and contains some obscure points: New
researches are thus still called for. I shall not dwell on the matter here.

9. THE STATISTICAL MATRIX Po

To finish, let me say a few words about a statistical matrix endowed with re-
markable properties to which von Neumann has drawn the attention.
It is known that in classical statistical thermodynamics one regards all the mi-
croscopic states (complexions) of a system to be a priori equally probable-that is,
equally probable when one does not possess any information about the state of the
system, such as the value of the total energy or a definite temperature imposed on
the system through its thermal contact with a heat reservoir.
The analog of this idea in wave mechanics is to assume as a priori equally prob-
able all the states of the system defined by the different members of a complete set
of orthonormal basis functions. Let 'Ph ... ,'Pk, ... be such a set. Knowing that the
physical system is represented by a mixture of states 'Pk, one will have to postulate,
in the absence of any other information, that the statistical matrix of the system
IS

Po = LPPCPk' with LP=l, (76)

k k
that is, that Po is the statistical matrix of a mixture for which all the statistical
weights are equal. If the states k are infinite in number, there exists a mathematical
difficulty, as p must then be infinitesimally small. This difficulty can be removed
by convenient artifices. But we shall brush the latter aside by supposing that the
number of states is limited. In taking the 'Pk as basis functions for representing the
matrix Po, one finds
(PO)kl = PDkl· (77)
If a function f has the expansion

f = LCk'Pk
k

in terms of the 'Pk, the operator Po will, on application to f, give

Pof = p LPcpJ = p LCk'Pk = pf. (78)

k k

We shall now prove the following theorem:

 The Statistical Matrix Po 267
If the statistical state of an ensemble of systems is initially represented by a
matrix Po, and if one performs on all systems of this ensemble a measurement
of the same quantity A, then the statistical state of the final ensemble is again
represented by a matrix Po.

If one measures on all systems of the ensemble the same quantity A, whose
eigenfunctions are Xl. ... ,Xn, . .. , a new mixture is obtained. But one can write

(79)

the d nm being the elements of a unitary matrix, for which

(SO)
n n

Consequently, when one measures A on a system in the state 'Pn, a mixture of

states Xk is generated wherein the state Xm occurs with the weight Id nm 12. In the
final mixture, obtained after measurement of A on all systems of the ensemble, the
weight of the state Xm will thus be

L pldkm 12 = P L dkmdkm = P Dmm = p.

k k

This means that the final mixture is one in which all states Xk appear with the same
statistical weight Pi it accordingly is represented by the statistical matrix

q.e.d. (S1)
 CHAPTER 16

THE ROLE OF TIME

IN WAVE MECHANICS 1

1. RETRODICTION ACCORDING TO COSTA DE BEAUREGARD 2

We have already seen that it is impossible for an observer who knows the out-
come of a measurement made at an instant t on a system to reconstruct the wave
function which described the state of the system for an observer possessing anterior
information.
Let us take the simple case of the localization of a particle and suppose that at
the time tl an observer measures the location of a particle, finding it to be at the
point M l . He will then adopt

as the initial value of the wave function. Starting from this value, he can, with the
aid of the wave equation, calculate the form 'l/JA(M, t) of the wave function at any
subsequent time t and specify the value of the probability l'l/JA(M,t2)1 2 for finding
the particle in any point M at the time t2 > tl. If then a new position measurement
at the instant t2 localizes the particle in a point M 2 , an observer B who knows only
of this second localization in M 2 , and not of the first in Mb would in no way be able
to discover the form of the function 'if;A (M, t) that was employed by the observer A
acquainted with the first localization. If he knew 'l/JA' observer B could, by mentally
going back in time, recover the localization in Ml at the time tl; but that for him
is now impossible. Even if a statistical experiment were to furnish observer B with
the amplitudes l'if;A (M, t2) I for all points M of space, he would not always know
the phases and consequently would be unable to follow the evolution of the wave
function 'l/JA backwards in time.
One would be led to the same conclusion on determining the states of motion by
momentum measurements, that is, if a localization were made in momentum space.
1. Editorial note. We have suppressed four pages that repeat the beginning of
Chap. 10 ("Examination of some difficult points of wave mechanics"), compelling
us to slightly modify the first sentence.
2. See Louis de Broglie, physicien et penseur (Albin-Michel, Paris, 1953),
pp. 409-410; O. Costa de Beauregard, C. R. Acad. Bc. (Paris) 236, 277, 1632
(1953).
 Retrodiction According to Costa de Beauregard 269
Nevertheless, the mental procedure whereby the observer follows the evolution
of 'IjJ backwards in time, starting from the initial form

has a meaning: This was pointed out several years ago by Fock and Costa de Beaure-
gard, who emphasized the importance of this point and gave to the mental operation
in question the name "retrodiction." Indeed, if it is impossible for the observer B,
who knows only of the second localization in M2 at the time t2, to return with
certainty to the first localization in Ml at the time tl by recovering the form of
'ljJA (M, t), he can nevertheless determine the relative probabilities of the various lo-
calizations at the instant tl, starting from the known localization of the particle in
M2 at the time t2. In other words, B can phrase the problem of the "probability
of causes" as follows: "What is the probability that the localization of the particle
in M2 at the instant t2 had as its cause the localization of this particle in Ml at
the time it?" This well underscores the fact that, if in wave mechanics there no
longer exists a rigorous deterministic connection between cause and effect, there still
remains a stochastic link between cause and effect, equally well in the future-past
as in the past-future sense.

Note G. L. The author would today have said more precisely that there exists
no rigorous and observable deterministic connection between cause and effect, but
that this fact does not preclude the existence of an underlying deterministic link
which has remained hidden from the experiments thus far performed.

To accomplish this retrodiction, the observer B must consider the wave function
'ljJB(M, t) that one obtains with the aid of the wave equation on starting from the
initial form
'ljJB(M, t2) = 6(M - M2)
and reversing the flow of time. He will then find this wave function to have the
form 'ljJB(M, tt} at the past instant tl < t2; and it is the quantity I'ljJB(M, tl)1 2 which
for him must define the probability that a localization took place in M at time tl.
This probability exists for the observer B who is familiar with the localization of
the particle in M2 at the time t2 but who is ignorant of its earlier localization in Ml
at the instant tl.
. For the procedure of retrodiction to be acceptable, it is evidently necessary
for I'ljJB(Ml,tl)1 2 to be different from zero; otherwise, one would find a vanishing
probability for an event that actually occurred! Looking at things more closely, we
see that even more is required: The probability l'ljJA (M2, t2) 12 that there will be a
localization in M2 at the time t2, if one knows that a localization took place in Ml
at the time tt, must equal the probability that there had been a localization in Ml
at the time tl if it is known that a localization occurred in M2 at time t2. With the
help of the above definitions of 'ljJA and 'ljJB' this condition can be written as

(12)
270 The Role of Time in Wave Mechanics
This symmetry property of 'l/JA and 'l/JB is directly related to the classic property of
Green's functions, as one can see by noting that, on expressing the solutions 'l/JA and
'l/JB of the wave equation with the aid of a symmetric kernel K(M', t'; M, t) (see Note
below), one gets

'l/JA(M2,t2) = jK(M2,t2; Ml,tl) o(M - MddM = K(M2,t2;MI,tl),

(13)
'l/JB(MI, tl) = j K(Ml, tl; M2, t2)* o(M - M2) dM = K*(Ml, tl; M2, t2),

and that
K(M2' t2; Mb td = K(Mb il; M2, i2)j
whence follows Eq. (12).

Note L. B. Let us introduce a kernel K(M, tj P, T), depending on two spacetime

points (M,t) and (P,r), and set

so that
'l/JA(M2,t2) = K(M2,t2j MI,tl)' (2)
Substituting this equality into the wave equation for 'l/JA' one finds

h a
-2. -a K(M,tjP,T) = HMK(M,tjP,T) (3)
'In t
and, consequently,

-2h . aa K*(M,tjP,T) = HMK*(M,t;P,T), (4)

1rZ t
with the initial condition 'l/JA(M, tI) = o(M - M I ) satisfied if

[K(M, tj P, T)]t=r = o(M - P) = [K*(M, tj P, T)]t=r' (5)

which proves that K and K* are symmetric.
Likewise, for retrodiction, we may introduce a kernel KI(M, tj P, T) such that

'l/JB(M, t) = j KI(M, t; P, T) o(P - M2)O(T - t2)dPdT

(6)
= Kl(M,tj M2,t2),
whence
(7)
 Retrodiction According to Costa de Beauregard 271
As the evolution of '!/JB takes place under time reversal, one finds

h ()
- - . ""QKI(M,t;P,T) = HMKI(M,t;P,T), (8)
21rt ut

with the initial condition '!/JB(M, t2) = h(M - M 2) satisfied if

(9)
which shows that KI is a symmetric kernel.
Equations (4), (5) and (8), (9) show that K*(M,t;P,T) and KI(M,t;P,T) obey
the same partial differential equation, subject to the same initial condition, and
therefore
(10)
Taking into account the symmetry of the kernels K and KI, one concludes from
(10) that

(11)

One can moreover readily see the physical meaning of the symmetry property
(12) in the following way. Consider any optical device that is capable of producing
interference or diffraction effects.

[_ _ _ _ _R_ _ _ _ _ _ _ J

Let a light source of unit intensity be placed at a point MI ncar the entrance
of this device, and suppose that at the point M2, near the exit of this device, the
light intensity is found to be i. If now the source of unit intensity were placed at
M2, instead of at MI, one would obtain the intensity i in MI. That this is the case
can be demonstrated with an appeal to the equation governing the propagation of
light, but the same conclusion appears also as necessary from the thermodynamic
point of view; because, if we imagine that the entire device is immersed in a heat
bath of temperature T, and if two pointlike blackbodies emitting thermal radiation
is placed in MI and M2, then it is necessary for each of these bodies to transmit to
the other just as much energy as it receives, otherwise the temperature equality of
the two blackbodies would be destroyed spontaneously, which is an impossibility.
One sees therefore that the observer B can, with the aid of the wave function
'!/JB' and starting from the knowledge of the localization in M2 at the instant tz, find
the probability of an anterior localization in MI at the instant tl' We repeat that it
here concerns a problem analogous to a classical problem in the probability calculus
272 The Role of Time in Wave Mechanics

of causes and that we have thus brought to the fore the stochastic and reciprocal
link between cause and effect.
It is moreover easy to recover the foregoing results by application of Bayes's
formula for the probability of causes. According to this formula, the probability
that an ascertained event k has an event i as its cause is given by
p.(k) = WiPik , (14)
1 EWjPik
i
where Wj denotes the a priori probability of the cause j and Pjk the probability
that, the cause j having occurred, the effect k will result. In accordance with this
formula, we here conclude that the probability for the localization of the particle in
MI at the time it, its localization in M2 at the time t2 > tl having been ascertained,
IS
p(M2h) _
M1,tl - J W(Ml,tl)II«MI,it;M2,t2)1 2
W(MI' tl)II«MI, tl; M2, t2)1 2 dMI
(15)

It is natural to assume that all positions MI are a priori equally probable, so that
W(MI, tl) = A for all MI. Moreover,
(16)
accordingly, since the functions tP are normalized,

q.e.d. (17)
Let us add some remarks. In a certain sense, one could say that the "true" wave
function in the time interval (t2, it) is the function tPA' since it contains information
which the observer B would have if his knowledge were complete. The instant tl
being anterior to the instant t2, the observer B could at the time t2 know the form
of the wave function tPA' But as, by hypothesis, he is aware only of the localization
in M2 at the time t2, his knowledge is imperfect, since he does not know all he could
know at the instant t2. This inadequacy in the observer's knowledge forces him to
employ the "imperfect" function tPB for his reasoning backwards in time, but this
function can furnish him only with probabilities, not with a certainty, concerning
localization at the time i}.
The whole question, it is seen, can be envisaged by adopting the information-
theoretic point of view, which is the fashion at present; but naturally one must
take into account the intervention, in a very novel manner, of quantum-mechanical
probabilities, characterized by the interference property and the essential role of
the phases. If it is desired to treat the problem rigorously, it becomes neces-
sary also to pay attention to relativistic ideas and consider all speeds of propa-
gation, always less than c. If the first localization is represented in Einsteinian
spacetime by the event MI, tl, the second localization will be represented by a
point event M2, t2 that is necessarily situated in the future light cone of the event
M I , t l . Accordingly, the observer B, who ascertains the localization in M2 at
 Special Role of Time in Quantum Mechanics 273

the time t2, could conceivably know (for example, via a light signal) of the lo-
calization that occurred in MI at the time tl' If he does not, there will be a
gap in his knowledge, our considerations showing in particular that the symme-
try between past and future in the relation (12) is only an apparent one and
that it is impossible to ignore the fact that the same time sense exists in the
consciousness of all people. We shall have the opportunity to return to this
point.

2. SPECIAL ROLE OF TIME IN QUANTUM MECHANICS.

THE FOURTH UNCERTAINTY RELATION

If one formally takes the viewpoint resulting from the general conceptions of
relativity theory, the fourth uncertainty relation, oW St '" h, appears as a natural
complement to the first three, 0Pi OXj '" h, because relativity theory views energy
as a quantity canonically conjugate to time, in the same way that the components
p;,py,PZ of the linear momentum are conjugate respectively to the variables x, y, z.
One sees this, for example, in noticing that the element W dt - Px dx - py dy - pz dz
of Hamilton's action integral is a spacetime invariant.
However, in quantum mechanics, the fourth uncertainty relation does not appear
in a really symmetric manner with the three first ones. Indeed, wave mechanics,
even in the relativistic form given to it by Dirac, does not establish a true symmetry
between the variables of space and time. [Note G. L. This point was studied by
the author in L 'electron magnetique (Ref. II, 11 ).] For, while the coordinates x, y, z
of a particle are "observables," each associated with an operator exhibiting values
which, for every pure case defined by a given function t/J, obey a definite probabil-
ity distribution, the time t is always defined as a parameter having a well-defined
value.
To be more specific about this matter, let the measurements be performed by
a Galilean observer who employs the coordinates x, y, z, t to locate events in the
macroscopic frame of his experiments. The variables x, y, z, t are numbers, or pa-
rameters, and these are the numbers that feature in the propagation equations and
in the wave function. But to every particle of atomic physics there correspond
"observable quantities" that are the coordinates of the particle. The correspon-
dence between the observable quantities x, y, z and the space frame of the Galilean
observer's x, y, z is stochastic, every observable quantity x, y, z being in general ca-
pable of assuming in this frame a whole series of values governed by a probability
distribution. By contrast, present-day quantum mechanics does not attach any "ob-
servable quantity t" to a particle; the theory contains only a variable t that is one
of the variables used in the spacetime frame of the observer, one which is defined
by the (essentially macroscopic) clocks employed by this observer.

Note L. B. (later scribbled in pencil). Pilot wave. The coordinates of the par-
ticles exist at every instant, but measurement can establish only a statistical link
between the coordinates and the spacetime frame.

Note L. B. (on a separate leaf and obviously written later than the preceding
274 The Role of Time in Wave Mechanics

note). In the pilot-wave theory, a distinction was drawn between the viewpoint of
the super-observer and that of the observer. For the super-observer, the particles
have positions and trajectories, and the description of the particle motions is given
in the spacetime frame with symmetry between space and time and with reversibility
in time. But for the human observer, who can know the positions and velocities
only through measurements made at a given moment of his time and subject to
uncertainty relations, the time which is "the observer's time," recorded by his clocks,
does not play the same role as the space coordinates. And, for this human observer,
the time is irreversible, because the events of which he is aware evolve in a fixed
direction; and, being endowed with a memory, he knows the past but cannot know
the future.

Note G. L. It is interesting to observe that in this note, where he takes a further

step towards his original ideas, de Broglie still remains under the influence of the
Copenhagen School and links the irreversibility of time to the consciousness and
memory of the observer. We know that later he would tie the irreversibility of the
measurement process to a "switching" of the particle, when the latter, on leaving
a spectral analyzer "clings" to one of the separate wave packets resulting from the
initial wave (Ref. II, 27, p. 119). It should also be noted that, following his work
on thermodynamics, de Broglie would adopt the idea of a time that is intrinsically
irreversible, even on the microscopic level, and unrelated to observation, which would
lead him to argue that the notion of spacetime, the utility of which in relativity he
obviously recognized, nevertheless has a fallacious aspect, because it establishes a
symmetry between space and time, whereas the notion of the irreversible passage of
time has no analog for the spatial coordinates. [Ann. Fond. L. de Broglie I(3), 116
(1976)].
It is necessary in wave mechanics to have an "evolution parameter" allowing
us to follow the variation of the state of any quantum system. But this evolution
of the state of a system or, more exactly, of our knowledge about it, necessarily
takes place in the time existing in the consciousness of the observer, time whose
flow one can follow only through the functioning of macroscopic clocks. It is in this
consciousness-related time frame where notably those sudden modifications in the
form of 'IjJ take place that are due to the measuring operations one can perform and to
the information they furnish us. But the fact that we are forced to adopt as evolution
variable the macroscopic time, i.e., the variable t of relativistic spacetime, prevents
us from attributing to particles or quantum systems an "observable quantity" t of
an aleatory nature, in the same way that we attribute to any space coordinate q an
observable quantity characterized by a probability distribution.
Such are some of the profound reasons opposing [at least from the human ob-
server's point of view] the establishment in wave mechanics of a symmetry between
space and time analogous to that postulated in the theory of relativity [if one accepts
the present theory]. These difficulties are intimately related to the fact that quan-
tum physics creates a link of a novel kind between the object and the subject. In
the new theory, the state of a quantum system no longer has an objective definition
corresponding to a description of "what is." On the contrary, any state is defined
uniquely as a function of "what we know"; it is a representation of our knowledge,
 The Fourth Uncertainty Relation 275
and beyond this representation we cannot go. It is therefore in the consciousness of
the observer, and consequently in the frame of macroscopic time, that the "state"
defined by the wave IjJ evolves; and, if quantum theories do not succeed in estab-
lishing a true symmetry between space and time, this failure seems to be due to the
particular character of the time perceived by our consciousness, to its continuous
unfolding and to its irreversibility.

Note G. L. The words between brackets, in the paragraph above, were added
afterwards, in two different inks.

Let us moreover point out that in relativity theory itself there is no complete
symmetry between space and time: first, because of the irreversibility of time
and, second, because it is At, and not t, that is symmetric with X,y,z (which
fact is connected with the opposite signs of the time and space derivatives in the
d'Alembertian). But this lack of symmetry becomes of course more profound in
quantum theories.

3. CORRECT STATEMENT OF THE FOURTH UNCERTAINTY

RELATION

Consider the case of a wave train 'Ij; that occupies a limited region of space.
The value of the coordinate x of the particle involved will then be uncertain: As
a result of an appropriate measurement, it can reveal itself as corresponding, with
a probability 1'Ij;(M)12, to any position M inside R. Moreover, the probability for
finding a given value of Px is equal to Ic(px) 12 , so that the value of Px, too, is
uncertain. We know that Heisenberg's uncertainty product lix lipx is of the or-
der h. But, though the particle coordinate x can be determined by a measure-
ment, one cannot speak of a measurement of its time t, because in wave mechan-
ics t denotes the macroscopic time of the observer, which always has a precise
value.
What then does the fourth uncertainty relation liE lit ;:::: h mean? It asserts that,
before one may attribute an energy E with an uncertainty liE to a particle, it is
necessary to perform an observation, a measuring operation, lasting at least the time
lit = h/ liE. Indeed, it follows from an analysis of the Fourier-integral representation
of the wave train that the time lit it takes a wave train to pass through a point P is
at least of the order lit = 1/ liv, where liv is the frequency interval effectively involved
in the Fourier expansion. To be able to affirm that the uncertainty in the energy is
at most liE = h liv, one must observe the passage through the point P of the front
end of the wave train and then of its rear end, which requires the observation to
take at least the time lit '" h/liE. Thus, before one could assert, for example, that
liE is zero, i.e., that the wave train is monochromatic, one would, strictly speaking,
have to make an observation of infinite duration.
Accordingly, while the first three uncertainty relations represent the existence
of a probability distribution for the conjugate variables q and p, that is, the fact
that these quantities are "aleatory variables" of probability calculus, the fourth
276 The Role of Time in Wave Mechanics

uncertainty relation must be interpreted differently: The time t is not an aleatory

variable, but the measurement of E can be carried out only with the aid of an
observation of infinite duration; and the more the time of observation is dimin-
ished, the more the uncertainty in the value of E will increase. Since the vari-
able t is not aleatory, there cannot exist precise relations between the disper-
sions of t and E of the kind (Jx(Jpx :::: h/47r, which characterizes the case of x
and Px. Phrased differently, the qualitative uncertainty relation fiEfit :::: h is
not paired with a relation between dispersions of the type (JE(Jt :::: h/47r, be-
cause (Jt is always zero. Here again we see the opposition that exists between the
foregoing conclusions and the relativistic symmetry obtaining between space and
time.

4. THE FOURTH UNCERTAINTY RELATION AND

PERTURBATION THEORY

Consider a system which has, among others, two quantum states i and k of
energies Ei and Ek, respectively. Calculation shows (as will soon be clarified) that
this system, when subjected to an external perturbation, will under the influence of
this action, oscillate between the states i and k with the frequency

However, from this fact it cannot be concluded that the system physically passes
from the state i to the state k, and conversely. One would be entitled to draw this
inference if it were possible to catch the system in either of these states, i.e., to
measure its energy in the state i or the state k. But, since the system stays in any
one of the states during a time ot
less than

1 h
-- ,
l/ik Ej - Ek
no energy measurement will allow us to evaluate the energy of the state i or k with
a precision superior to
h
oE Eo" - Ek
rv -
ot '"
'
and we can therefore not distinguish the two states from one another. During the
interaction, the energy of the system remains indeterminate between Ei and Ek,
and one is therefore unable to verify the conservation of energy with an uncertainty
less than lEi - Ek I.
Suppose now that the system possesses the energy El up to the time tl, that
between tl and t2 it is acted on by an external perturbation, and that it remains in
the state E2 for all t > t 2 • As one is able to measure El for all times preceding tl and
E2 for all times following t 2, the energies El and E2 can be very precisely known;
and the conservation of energy requires E2 = E l . However, in the time interval
t2 - tl during which the perturbation is present, the system can pass through an
 The Fourth Uncertainty and Perturbation Theory 277
intermediate state of energy E. If t2 - tl is small compared with hj(E2 - El), it is
impossible to catch the system in this state by measuring its energy. One may say
that the system passes through the "virtual" state of energy E; and, in reality, for
as long as the interaction lasts, the energy will remain indeterminate up to E2 - E l .
The conservation of energy can thus be verified for the overall transitions El --+ E2
but not necessarily for the virtual transitions El --+ E and E --+ E 2.
One may more rigorously arrive at the same conclusions by pursuing the pro-
cedure known as the "variation of constants." In this method, due to Dirac,
one supposes that, prior to the perturbation, the envisaged system possesses
the unperturbed time-independent Hamiltonian H(O), corresponding to the sta-
tionary states of the system, whose eigenvalues EiO) and eigenfunctions "piO) are
presumed to be known. For a limited duration 0 --+ T the system is sub-
jected to the action of a perturbation that may depend on the time. Such
a perturbation is represented in the Hamiltonian H by a term V(t), so that
now
H = H(O) + V(t);
and the equation describing the evolution of the perturbed system will be

[H(O) + V(t)]"p = ~ {}"p . (18)

2n {}t
At every instant t, the wave function of the system can be written as an ex-
pansion employing the complete system of eigenfunctions "piO) of the unperturbed
Hamiltonian:
"p(t) = ~Ck(t)"pkO)exp C~i EkO)t), (19)

wherein always

since "p is normalized.

According to the general principles of wave mechanics, the probability that the
system will at the time t find itself in the state "piO) is given by ICk(t) 12. Again it is
necessary, as we have remarked above, that the system remains in this state during
a time long enough for us to be able to know the value of its energy.
On substituting this expression for "p into the evolution equation, one easily finds
that the variation of the Ck is given by the equations

(20)

wherein
278 The Role of Time in Wave Mechanics

For V = 0, the q are constants, and .,p becomes a sum with cmlstant coefficients
of eigenfunctions of H(O)-a familiar case. If the perturbation V Is not zero, the
coefficients q will vary with time, whence the name "variation of constants" given
to this method of calculation. (One verifies easily that

whence L I kl C 2 = const. = 1
k

at every instant.)
In general the integration of Eqs. (20) is difficult, but in the simple case where it
Ei
is known that the system is in the state of energy O) at the start of the integration
(t = 0), and where consequently one must set cn(O) = 1 and cm(O) = 0 for m =f n,
an approximate integration furnishes the result

. (0) (0)
( ) _ V. exp[(2n/h)(En - Em )t]-l (21)
Cm t - mn (0) (0) ,
En -Em

whence

(22)

This quantity can be considered as giving the probability that the system will at
the instant t be in the state m; it is proportional to IVmn 12 , which assigns a special
importance to the matrix elements Vmn . However, in accordance with the remarks
made above, as long as the perturbation lasts, one cannot physically catch the
system in the state of energy Em, because that would require the energy to be
measured with an uncertainty less than Em - En, while the time during which the
system stays in the state m would nevertheless be shorJer than h/(Em - En). Only
if the perturbation Vet) terminates at an instant T (Le., Vet) = 0 for t > T) can
the system be caught in the state m (which then becomes a permanent final state),
the probability for this outcome being given by Icm(T) 12. As the energy of a system
undergoing a perturbation is not measurable, one cannot to such a system apply
the conservation of energy; this law becomes verifiable only after the perturbation
has ceased.
The preceding conclusions are confirmed by a more thorough studyof.perturba-
tion theory, which notably leads to the introduction of the notion of probability per
unit time. For a discussion of these questions, we refer to last year's course (first
part of the present book).
 The Operators Hand (h/27ri)8/at 279
5. THE OPERATORS HAND (h/21ri)8/8t

The time playing a special role in wave mechanics, one must expect also to find
something special for the conjugate quantity, the energy. Indeed, since the operator

h 8

corresponds to the quantity px, it would be natural from the relativistic point of
view to assign the operator
h 8
27ri 8t
to the energy. But it is the Hamiltonian operator H, involving the partial derivatives
with respect to x, y, z, which one must associate with the energy. The wave equation

clearly shows this duality of the operator as well as the distinctive role played in
wave mechanics by the quantity conjugate to time.
The necessity for adopting H as the operator corresponding to the energy
emerges from the fact that the energy eigenvalues of a quantum system are defined
by the equation H'Ij; = E'Ij; along with the boundary conditions for some spatial
domain. The equation
h 8'1j;
27ri at= E'Ij;,
whose solution would be
'Ij; = a expC:i Et),
is not suitable for quantification: It would be incapable of satisfying boundary con-
ditions in time, and such conditions would moreover not furnish eigenvalues in
agreement with experiment. This clearly demonstrates the necessity in quantum
mechanics not to treat the coordinates of space and time in a symmetric fashion.
Here is still another important remark: The principle of spectral decomposition
teaches that, if <Xk and 'f'k are the eigenvalues and eigenfunctions, respectively, of
the operator A in wave mechanics, and if one writes the expansion

then the probability that a measurement will furnish at the instant t the value <Xk for
the quantity A is given by Iq(t) 12. Applying this fact to the energy, whose operator
H has the eigenvalues Ek and eigenfunctions 'lj;k, we are thus led to say: If the wave
function can be written as
280 The Role of Time in Wave Mechanics
then the probability that a measurement made at the instant t will cause one to
attribute the energy Ek to the system is given by ICk(t) 12. But here this assertion
contains an important restriction: It is truly applicable only if all the q(t) vary
slowly, so that one can make a precise measurement of the energy without coming
into conflict with the fourth uncertainty relation. When this condition is not fulfilled
(as happened in the example above), it is evident that the probability assertion loses
all physical meaning. This circumstance, which is not encountered for quantities
other than energy, completes our discussion of the special role played by energy,
the quantity canonically conjugate to the time variable, in the formalism of wave
mechanics, a role that is particularly related to the fact that in wave mechanics time
is not an aleatory variable but a numerical evolution parameter.

6. APPLICATION OF THE FORMALISM OF ARNOUS TO

THE OPERATORS ACTING ON TIME

In the hope (not justified in our opinion) of establishing a symmetry between

the spatial variables and the time variable, Costa de Beauregard had the interesting
idea of applying the formalism of Arnous to operators acting on the time.
To present the problem, let us first summarize the formalism of Arnous. As basic
postulate he assumes that the probability distribution corresponding to a quantity
whose operator is A has the characteristic function

FA(s) = in 1jJ*exp(isA)1jJdr. (23)

From the definition itself of the characteristic function as the mathematical ex-
pectation of exp (isA), i.e.,

FA(S) = 1 +00

-00 P(a')exp(isa')da',

one concludes, on invoking the Fourier inversion formula, that

P(a',t) = 21
7r
1+
-00
00
FA(s)exp(-isa')ds
(24)
= -1
27r
1+
-00
00
ds exp (-isa') 1D
1jJ* exp (isA)1jJdr.

With a and <p designating respectively the eigenvalues and eigenfunctions of A, one
has

1jJ = 1
+00

-00 c(a, t)<p(a, r) da, exp (isA)1jJ = 1 +00

-00 c(a, t) exp (isA)<p(a, r) da,

(25)
 Arnous's Formalism Applied to Operators Acting on Time 281

whence

1
p(ci,t) = -2 1+ 00 ds exp(-isa' ) f dT1+00 e*((3,t)cp*((3,r)d(3
1r -00 JD-oo
X 1-00+00 e(a,t)exp(isa)cp(a,r)da
1
= 21r 1+ -0000 e*((3, t) d(3
-0000 ds exp (-isa' ) 1+
x 1+ 00 e( a, t) exp (isa) da f cp* ((3, r )cp( a, r) dT (26)
-00 JD
= 1+
1
21r-0000 ds exp(-isa' ) 1+
-0000 le(a,t)1 2 exp(isa)da
1 1+00
= 1 +00 le(a,t)1 2 da.- dsexp(is(a-a' ))
-00 21r -00
+00
= 1-00 le(a,t)1 28(a-a' )da
where in the third step 8((3 - a) was substituted for the integral over T. Finally,
therefore
P(al,t) = le(al ,t)1 2 , (27)
the expected result.
The foregoing formalism can be applied without difficulty to the case where the
operator A acts only on the space variables, the time t possibly appearing but only
as a numerical parameter. Costa de Beauregard proposed to extend the formalism
to operators acting on the time.
Consider the case of the energy operator when a quantum system is subjected
to an external perturbation for a limited interval 0 -4 T of time. If during this time
the wave function of the system can be expanded in the form

1jJ = Je(E,t)1jJ~)(r)dE,
where the 1jJ~) denote the eigenfunctions of the Hamiltonian H(O) for the unper-
turbed system, Arnous's formalism leads one, as we have just seen, to regard
le(E, T)12 as the probability for finding the system in the state of energy E at
the time t. But our study of the significance of the fourth uncertainty relation has
shown that this assertion does not in general have a physical meaning, no measure-
ment being able to catch the system in the state of energy E. One must be content
282 The Role of Time in Wave Mechanics

to say that, at the end ofthe perturbation (t ~ T), when the coefficients c(E, t) have
assumed their final constant values c(E, T), the probability for finding the system
in the state of energy E is equal to Ic(E,T)12.
All of this agrees with what we have seen earlier. Let us now attempt, with
Costa de Beauregard, to apply Arnous's formalism to the operator
h a
A = 27l"i at'

1:
assuming that the corresponding probability is given by the formula

P(E,t) = 2~ 00
ds exp(-isE) L~*exp C!i :t)~dr, (28)

which must be the case if it makes sense to introduce the given operator A into
the definition of Arnous's characteristic function. Since, for t ~ T, i.e., after the
perturbation has ceased,

C:i :t}p C:i H(O»)~.

one also has
exp = exp

We then see that, for t ~ T, the probability P(E, t) takes the constant value
P(E,T) = Ic(E,T)1 2,
i.e., the value already found earlier; and, in the domain that is actually observable,
the substitution of the operator (hj27l"i)(ajat) for the operator H in the formalism
of Arnous changes nothing.
But let us now apply Arnous's formalism to the operator A = tx, which would
correspond to the time, if the time were an aleatory variable like x, y, z. One would

2.1+
then have
P{t', t) =
21T -00
00
ds exp (-ist') r~*
JD
exp (ist)~ dr
(29)
= 2.1+
21T -00
00
ds exp (is(t - t')) = 8(t - t').

This result signifies that, at the instant t indicated on our clocks, the aleatory vari-
able t belonging to a system (or particle) necessarily has the value t' = t. Conse-
quently, this time variable, having a definite value, is not really an aleatory variable,
and we are driven back to the conclusion that in wave mechanics there is no statistics
governing the time.
The introduction, into the formalism of Arnous, of operators acting on the time
therefore does not at all suppress the profound difference existing in wave mechanics
between the time variable and energy, on the one hand, and the space variables and
their conjugate quantities, on the other.
 Multitemporal Formalism. Multiplicity Curves 283
7. MULTITEMPORAL FORMALISM.
MULTIPLICITY CURVES IN SPACETIME

In order to verify the covariant character of the equations employed in the for-
mulas of quantum mechanics, people (Dirac, Fock, and Podolski) have been led to
use the following method, known under the name of "multi temporal theory":
Let a system of particles 1,2, ... ,k, . .. ,N be given. The localization of the kth
particle in spacetime enables one to attribute to this "event" the four coordinates
Xk, Yk, Zk, tk, that is, to assign to every particle a time tk. When equations are
rewritten in terms of the individual times tk, their relativistic covariance can more
easily be verified. But if (as is inevitable) one wants to go on from here and make
predictions regarding the results of observations and measurements performed by
an observer, it becomes necessary to equate all these individual times tk to the same
time t, the macroscopic time of a Galilean observer as noted on his clocks. This
leads one to consider plane sections of constant t in spacetime for the observer under
consideration~the foliation of spacetime by hyperplanes. One thus gets back to all
the preceding conclusions regarding the special role played by the variable t, the
macroscopic time of the observer, which, unlike the space coordinates Xk, Yk, Zk of
the particles, is not governed by any statistics.
The work carried out in the last years by Tomonaga, Schwinger, and others, has
given rise to another approach, whose importance Costa de Beauregard has also
underlined, which notably allows one to better bring to the fore the covariance of
quantum field theory. This method consists therein that, after attributing to every
particle an individual time tk, one foliates spacetime not any longer by parallel
hyperplanes corresponding each to a constant value of a Galilean observer's time
but by a family of hypersurfaces of the spatial kind: The points of these three-
dimensional hypersurfaces can be located by three spatial curvilinear coordinates
U, v, W; and one may define a fourth, temporal variable T, which is counted in a
direction normal to the hypersurfaces, so that for each hypersurface of the family one
has T = const. The earlier considered hyperplanes t = const. are evidently recovered
as a particular case of the present scheme. Having thus foliated spacetime, one can
in the equations obtained for the assembly of particles set all the tk equal to the same
value T, that is, in short, group together all events Xk, Yk, Zk, tk lying in the same
surface T = const. One thus obtains a generalization, called the "supermultitemporal
theory," of the ordinary multi temporal theory; and this method has advantages
notably for the study of covariance questions. Costa de Beauregard has proposed
to use this foliation of spacetime by surfaces S of the spatial kind to write Arnous's
characteristic function in the form

FA(S) = /s'1f;*ex P (iSA)'1f;d(J",

wherein the integration is extended over a surface S. In this way one obtains a theory
that in certain respects is more elegant and more general than that of the preceding
section. But it does not seem to us that this theory can serve to eliminate the
difference in nature between the temporal variable and the spatial variables; because
this difference, existing in the particular cases where the hyperplanes are taken as the
284 The Role of Time in Wave Mechanics
surfaces S, cannot disappear in the general case, which includes this special case.
It follows moreover from the remarks made in his notes by Costa de Beauregard
himself that the temporal variable 7 plays a role very different from that of the
spatial variables u, v, w. To me it therefore appears impossible to eliminate the
singular role played in quantum theory by the temporal variable.
 WORKS OF LOUIS DE BROGLIE

I. NOTES AND PAPERS

[1] Sur Ie calcul des frequences limites d'absorption K et L des elements lourds,
C. R. Acad. Sci. (Paris) 170, 585 (1920).
[2] Sur l'absorption des rayons X par la matiere, C. R. Acad. Sci. (Paris) 171,
1137 (1920).
[3] Sur Ie modele d'atome de Bohr et les spectres corpusculaires, C. R. Acad. Sci.
(Paris) 172, 746 (1921) [in collaboration with Maurice de Broglie].
[4] Sur la structure electronique des atomes lourds, C. R. Acad. Sci. (Paris) 172,
1650 (1921) [in collaboration with A. Dauvillier].
[5] Sur la distribution des electrons dans les atomes lourds, C. R. Acad. Sci. (Paris)
173, 137 (1921) [in collaboration with A. Dauvillier].
[6] Sur Ie spectre corpusculaire des elements, C. R. Acad. Sci. (Paris) 173, 527
(1921) [in collaboration with M. de Broglie]. •
[7] Sur la degradation du quantum dans les transformations successives des radia-
tions de haute frequence, C. R. Acad. Sci. (Paris) 173, 1160 (1921).
[8] Sur la theorie de l'absorption des rayons X par la matiere et Ie principe de
correspondance, C. R. Acad. Sci. (Paris) 173, 1456 (1921).
[9] Rayons X et equilibre thermodynamique, J. Phys. (VI) III, 33-45 (1922).
[10] Sur Ie systeme spectral des rayons Rrentgen, C. R. Acad. Sci. (Paris) 175, 685
(1922) [in collaboration with A. Dauvillier].
[11] Sur les analogies de structure entre les series optiques et les series Rrentgen,
C. R. Acad. Sci. (Paris) 175,755 (1922) [in collaboration with A. Dauvillier].
[12] Rayonnement noir et quanta de lumiere, J. Phys. (VI) III, 422-428 (1922).
[13] Sur les interferences et la theorie des quanta de lumiere, C. R. Acad. Sci. (Paris)
175, 811 (1922).
[14] Remarques sur les spectres corpusculaires et l'effet photoelectrique, C. R. Acad.
Sci. (Paris) 175, 1139 (1922) [in collaboration with M. de Broglie].
[15] Remarques sur Ie travail de E. Hjalmar concernant la serie M des clements,
C. R. Acad. Sci. (Paris) 175, 1198 (1922) [in collaboration with A. Dauvillier].
[16] Ondes et quanta, C. R. Acad. Sci. (Paris) 177,517 (1923).
[17] Quanta de lumiere, diffraction et interferences, C. R. Acad. Sci. (Paris) 177,
548 (1923).
[18] Les quanta, la theorie cinetique des gaz et Ie principe de Fermat, C. R. Acad.
Sci. (Paris) 177, 630 (1923).
286 Works of Louis de Broglie

[19] Sur Ie systeme spectral des rayons X et la structure de l'atome, J. Phys. (VI)
V, 119 (1924) [in collaboration with A. Dauvillier].
[20] Sur la verification experiment ale des projections d'eIectrons prevues lors de la
diffusion des rayons X par les considerations de Compton et Debye, C. R. Acad.
Sci. (Paris) 178, 383 (1924) [in collaboration with M. de Broglie].
[21] A tentative theory of light quanta, Philos. Mag. XLVII, 446-458 (1924).
[22] Sur la definition generale de la correspondance entre onde et mouvement, C. R.
Acad. Sci. (Paris) 179, 39 (1924).
[23] Sur un theoreme de Bohr, C. R. Acad. Sci. (Paris) 179, 676 (1924).
[24] Sur la dynamique du quantum de lumiere et les interferences, C. R. Acad. Sci.
(Paris) 179, 1039 (1924).
[25] Sur la frequence propre de l'electron, C. R. Acad. Sci. (Paris) 180,498 (1925).
[26] Sur l'interpretation physique des spectres X d'acides gras, C. R. Acad. Sci.
(Paris) 180, 1485 (1925) [in collaboration with J.-J. Trillat].
[27] Sur Ie parallelisme entre la dynamique du point materiel ell'optique goometri-
que, J. Phys. (VI) VII (I), 1-6 (1926).
[28] Remarques sur la nouvelle mecanique ondulatoire, C. R. Acad. Sci. (Paris)
183, 272 (1926).
[29] Les principes de la nouvelle mecanique ondulatoire, J. Phys. (VI) VII (11),
321, 337 (1926).
[30] Sur la possibilite de relier les phenomenes d'interferences et de diffraction It la
theorie des quanta de lumiere, C. R. Acad. Sci. (Paris) 183,447 (1926).
[31] Sur la possibilite de mettre en accord la theorie electromagnetique avec la nou-
velle mecanique ondulatoire, C. R. Acad. Sci. (Paris) 184, 81 (1927).
[32] La structure atomique de la matiere et du rayonnement et la mecanique ondu-
latoire, C. R. Acad. Sci. (Paris) 184, 273 (1927).
[33] L'Univers It cinq dimensions et la mecanique ondulatoire, J. Phys. (VI) VIII
(2), 65-73 (1927).
[34] La Mecanique ondulatoire et la structure atomique de la matiere et du rayon-
nement, J. Phys. (VI) VIII (5), 225-241 (1927).
[35] Sur Ie role des ondes continues en mecanique ondulatoire, C. R. Acad. Sci.
(Paris) 185, 380 (1927).
[36] Rapport au Ve Conseil de Physique Solvay sur la "nouvelle dynamique des
quanta," in Electrons et photons, rapports et discussions du Ve Conseil de
Physique Solvay (Gauthier-Villars, Paris, 1922), p. 105.
[37] Corpuscule et ondes W, C. R. Acad. Sci. (Paris) 185, 1180 (1927).
[38] Sur les equations et les conceptions generales de la mecanique ondulatoire, Bull.
Soc. Math. France, May 1930.
[39] Remarques sur les integrales premieres en mecanique ondulatoire, C. R. Acad.
Sci. (Paris) 194, 693 (1932).
[40] Sur les densites des valeurs moyennes dans la theorie de Dirac, C. R. Acad. Sci.
(Paris) 194, 1062 (1932).
[41] Sur une analogie entre l'eIectron de Dirac et I'onde electromagnetique, C. R.
Acad. Sci. (Paris) 195, 536 (1932).
[42] Remarques sur Ie moment magnetique et Ie moment de rotation de l'electron,
C. R. Acad. Sci. (Paris) 195,577 (1932).
 Notes and Papers 287

[43] Sur Ie champ electromagnetique de l'onde lumineuse, C. R. Acad. Sci. (Paris)

195, 862 (1932).
[44] Sur la densite de l'energie dans la theorie de la lumiere, C. R. Acad. Sci. (Paris)
197, 1377 (1933).
[45] Sur la nature du photon, C. R. Acad. Sci. (Paris) 198, 135 (1934).
[46] Quelques remarques sur la thoorie de l'electron magnetique de Dirac, Arch. Sci.
Phys. Nat. (5) XV, 465 (1933).
[47] Remarques sur la theorie de la lumiere, Mem. Acad. Roy. Sci. Liege (3) XIX
(1934).
[48] L'equation d'ondes du photon, C. R. Acad. Sci. (Paris) 199, 445 (1934).
[49] Sur Ie spin du photon, C. R. Acad. Sci. (Paris) 199,813 (1934) [in collaboration
with J. Winter].
[50] Sur l'expression de la densite dans la nouvelle theorie des photons, C. R. Acad.
Sci. (Paris) 199, 1165 (1934).
[51] Une remarque sur l'interaction entre la matiere et Ie champ electromagnetique,
C. R. Acad. Sci. (Paris) 200, 361 (1935).
[52] Sur Ie theoreme de Krenig en mecanique ondulatoire, C. R. Acad. Sci. (Paris)
201, 369 (1935) [in collaboration with J.-L. DestouchesJ.
[53] La variance relativiste du moment cinetique d'un corps en rotation, J. Math.
Pur. Appl. XV, 89 (1936).
[54] La thoorie du photon et la mecanique ondulatoire relativiste des systemes, C. R.
Acad. Sci. (Paris) 203, 473 (1936).
[55] Les recentes conceptions theoriques sur la lumiere, Ann. Soc. Sci. Bruxelles (1)
CLVII, 99-119 (1937).
[56] La quantification des champs en thoorie du photon, C. R. Acad. Sci. (Paris)
205, 345 (1937).
[57] Sur un cas de reductibilite en mecanique ondulatoire des particules de spin 1,
C. R. Acad. Sci. (Paris) 208, 1697 (1939).
[58] Sur les particules de spin quelconque, C. R. Acad. Sci. (Paris) 209, 265 (1939).
[59] Champs reels et champs complexes en thoorie eIectromagnetique quantique du
rayonnement, C. R. Acad. Sci. (Paris) 211,41 (1940).
[60J Sur l'interpretation de certaines equations dans la theorie des particules de spin
2, C. R. Acad. Sci. (Paris) 212,657 (1941).
[61] Sur la propagation de l'energie lumineuse dans les milieux aniostropes, C. R.
Acad. Sci. (Paris) 215, 153 (1942).
[62] Sur la representation des grandeurs electromagnetiques en thoorie quantique
des champs et en mecanique ondulatoire du photon, C. R. Acad. Sci. (Paris)
217, 89 (1943).
[63] L'introduction des constantes de Coulomb et de Newton en mecanique ondu-
latoire, C. R. Acad. Sci. (Paris) 218, 373 (1944) [in collaboration with M.-
A. Tonnelat].
[64] Remarques sur quelques difficultes de la theorie du photon liees it l'emploi
d'une solution d'annihilation, C. R. Acad. Sci. (Paris) 218, 889 (1944) [in
collaboration with M.-A. Tonnelat].
[65] Sur un effet limit ant les possibilites du microscope corpusculaire, C. R. Acad.
Sci. (Paris) 222, 1017 (1946).
288 Works of Louis de Broglie
[66] Remarques sur la formule de Boltzmann relative aux systemes periodiques,
C. R. Acad. Sci. {Paris} 223, 298 (1946).
[67] Sur l'etude des tres petites structures au microscope corpusculaire, C. R. Acad.
Sci. {Paris} 223, 490 (1946).
[68] Sur l'application du tMoreme des probabilites compo sees en mecanique ondu-
latoire, C. R. Acad. Sci. {Paris} 223, 874 (1946).
[69] Sur les electrinos de M. Thibaud et l'existence eventuelle d'une tres petit charge
du neutron, C. R. Acad. Sci. {Paris} 224, 615 (1947).
[70] La diffusion coMrente et Ie microscope corpusculaire, C. R. Acad. Sci. {Paris}
224,1743 (1947).
[71] Le principe d'inertie de l'energie et la notion d'energie potentielle, C. R. Acad.
Sci. {Paris} 225, 163 (1947).
[72] Sur la frequence et la vitesse de phase des ondes planes monochromatiques en
mecanique ondulatoire, C. R. Acad. Sci. {Paris} 225,361 (1947).
[73] Sur la variance relativiste de la temperature, Cahiers Phys., January 1948,
pp. I-II.
[74] Sur la statistique des cas purs en mecanique ondulatoire, C. R. Acad. Sci.
{Paris} 226, 1056 (1948).
[74 bis] La statistique des cas purs en mecanique ondulatoire et l'interference des
probabilites, Revue Sci., 87th year, 1948, p. 259.
[75] Sur la possibilite de mettre en evidence Ie moment magnetique pro pre des
particules a spin, C. R. Acad. Sci. {Paris} 226, 1765 (1948).
[76] Sur la possibilte de mettre en evidence Ie moment magnetique propre des par-
ticules de spin 1/2, J. Phys. {VIII} IX, 265 (1948).
[77] Sur Ie calcul classique de l'energie et de la quantite de mouvement d'un electron
purement electromagnetique, C. R. Acad. Sci. {Paris} 228, 1265 (1949).
[78] Energie libre et fonction de Lagrange. Application a l'electrodynamique et a
l'interaction entre courants et aimants permanents, Portugal. Phys. III, 1-20
(1949).
[79] Penetration d'une onde electromagnetique dans un milieu ou la constante
dielectrique varie lineairement (preliminary note of the National Laboratory
of Radio-electricity, No. 129, 1949).
[80] Sur une forme nouvelle de l'interaction entre les charges eIectriques et Ie champ
electromagnetique, C. R. Acad. Sci. {Paris} 229, 157 (1949).
[81] Nouvelles remarques sur l'interaction entre une charge elect rique et Ie champ
electromagnetique, C. R. Acad. Sci. {Paris} 229, 269 (1949).
[82] Sur la theorie du champ soustractif, C. R. Acad. Sci. {Paris} 229,401 (1949).
[83] Sur les champs crees par Ie proton et par Ie neutron, C. R. Acad. Sci. {Paris}
229, 640 (1949).
[84] Une conception nouvelle de l'interaction entre les charges eIectriques et Ie champ
electromagnetique, Portugal. Math. 8, 37-58 (1949).
[85] Sur les champs mesoniques lies a l'electron dans la nouvelle theorie du champ
soustractif, C. R. Acad. Sci. {Paris} 230, 1009 (1950) [in collaboration with
Rene Reulos].
[86] Sur la possibilite d'une structure complexe pour les particules de spin 1, C. R.
Acad. Sci. {Paris} 230, 1329 (1950) [in collaboration with M.-A. Tonnelat].
 Notes and Papers 289

[87] Remarques complement aires sur la structure complexe des particules de spin
1, C. R. Acad. Sci. {Paris} 230, 1434 (1950).
[88] Sur la convergence des integrales dans Ie probleme de polarisation du vide,
C. R. Acad. Sci. {Paris} 230,2061 (1950).
[89] Sur une forme nouvelle de la theorie du champ soustractif, J. Phys. XI, 481
(1950).
[90] Schema lagrangien de la theorie du champ soustractif, C. R. Acad. Sci. {Paris}
232, 1269 (1951).
[91] Sur la possibilite d'une structure complexe des particules de spin 1, J. Phys.
XII, 509 (1951).
[92] Quelques considerations sur les transformations de jauge et la definition des
tenseurs de Hertz en theorie du corpuscule maxwellien de spin 1, C. R. Acad.
Sci. {Paris} 232,2056 (1951) [in collaboration with Bernard KwaiJ.
[93] Remarques sur la theorie de l'onde pilote, C. R. Acad. Sci. {Paris} 233, 641
(1951 ).
[94] Sur Ie tenseur energie-impulsion dans la theorie du champ soustractif, C. R.
Acad. Sci. {Paris} 234, 20 (1952).
[95J Sur la possibilite d'une interpretation causale et objective de la mecanique
ondulatoire, C. R. Acad. Sci. {Paris} 234, 265 (1952).
[96] Sur les relations entre les coefficients de charge et de masse dans la theorie du
champ soustractif, C. R. Acad. Sci. {Paris} 234, 1505 (1952).
[97] Sur l'introduction des idees d'onde pilote et de double solution dans la theorie
de l'eIectron de Dirac, C. R. Acad. Sci. {Paris} 235, 557 (1952).
[98] Sur l'interpretation de la mecanique ondulatoire des systemes de corpuscules
dans l'espace de configuration par la theorie de la double solution, C. R. Acad.
Sci. {Paris} 235, 1345 (1952).
[99] La mecanique ondulatoire des systemes de particules de meme nature et la
theorie de la double solution, C. R. Acad. Sci. {Paris} 235, 1453 (1952).
[100] Sur l'interpretation de la mecanique ondulatoire it l'aide d'ondes it region sin-
guliere, C. R. Acad. Sci. {Paris} 236, 1459 (1953).
[101] Sur l'interpretation causale et non lineaire de la mecanique ondulatoire, C. R.
Acad. Sci. {Paris} 237, 441 (1953).
[102] Considerations de mecanique classique preparant la justification de la mecani-
que ondulatoire des systemes dans la theorie de la double solution, C. R. Acad.
Sci. {Paris} 239, 521 (1954).
[103] Justification du point de la double solution de la mecanique ondulatoire des
systemes dans l'espace de configuration, C. R. Acad. Sci. {Paris} 239, 565
(1954).
[104] Une nouvelle demonstration de la formule du guidage dans la theorie de la
double solution, C. R. Acad. Sci. {Paris} 239, 737 (1954).
[105] Ondes regulieres et ondes it region singuliere en mecanique ondulatoire, C. R.
Acad. Sci. {Paris} 241, 345 (1955).
[106] Illustration par un exemple de la forme des fonctions d'onde singulieres de la
theorie de la double solution, C. R. Acad. Sci. {Paris} 243, 617 (1956).
[107] La significant du 1\Ii 12 pour les etats stationnaires pour l'interpretation causale
de la mecanique ondulatoire, C. R. Acad. Sci. {Paris} 243, 689 (1956).
 290 Works of Louis de Broglie

[108] Idees nouvelles concernant les systemes de corpuscules dans l'interpretation

causale de la mecanique ondulatoire, C. R. Acad. Sci. (Paris) 244, 529 (1957)
[in collaboration with Joao Luis Andrade e Silva].
[109] Tentative de raccord entre l'equation de Heisenberg et l'equation de l'onde u
en thCorie de la double solution, C. R. Acad. Sci. (Paris) 246, 2077 (1958).
[110] Sur la nomenclature des particules, C. R. Acad. Sci. (Paris) 247, 1069 (1958).
[111] Deux remarques en relation avec Ie probleme du disque tournant en theorie de
la relativite, C. R. Acad. Sci. (Paris) 249, 1426 (1959).
[112] Problemes classiques et representation bilocale du rotateur de Nakano, C. R.
Acad. Sci. (Paris) 249, 2255 (1959) [in collaboration with Pierre Hillion and
Jean- Pierre Vigier].
[113] L'interpretation de la mecaniqlle ondulatoire, J. Phys. 20, 963 (1959).
[114] La thermodynamique de la particule isolee, C. R. Acad. Sci. (Paris) 253, 1078
(1961 ).
[115] Nouvelle presentation de la thermodynamique de la particule isolce, C. R. Acad.
Sci. (Paris) 255, 87 (1962).
[116] Quelques consequences de la thermodynamique de la particule isolee, C. R.
Acad. Sci. (Paris) 255, 1052 (1962).
[117] Remarques sur l'interpretation de la dualite des ondes et des corpuscules,
Cahiers Phys., No. 147, October 1962, pp. 425~445.
[118] Application de la theorie de la fusion au nouveau modele etendu des particules
element aires , C. R. Acad. Sci. (Paris) 256, 3390 (1963) [in collaboration with
J.-P. Vigier].
[119] Table des particules elementaires associees au nouveau modele des particules
element aires , C. R. Acad. Sci. (Paris) 256, 3551 (1963) [in collaboration with
J.-P. Vigier].
[120] Sur l'introdudion de l'energie libre dans la therrnodynamique cachee des par-
ticules, C. R. Acad. Sci. (Paris) 257, 1430 (1963).
[121] Sur la theorie des foyers cinetiques dans la thermodynamique de la particllie
isolee, C. R. Acad. Sci. (Paris) 257, 1822 (1963).
[122J Ondes eIectromagnetiques et photons, C. R. Acad. Sci. (Paris) 258, 6345
(1964).
[123] Sur un point de la thoorie des lasers (communication to the Academy of Sciences
of Lisbon, November 21, 1963).
[124] La thermodynamique cachee des particules, Ann. [nst. H. Poincare 1 (1), 1~ 19
(1964).
[125] Sur la relation d'incertitude bnbcp ;:::: 271", C. R. Acad. Sci. (Paris) 260,6041
(1965).
[126] Sur la transformation relativiste de la quantite de chaleur et de la temperature
ct la thermodynamique cac1ee de la particule, C. R. Acad. Sci. B 262, 1235
(1966).
[127] Sur Ie deplacement des raics cmises par un objet astronomique lointain,
C. R. Acad. Sci. B 263, 589 (1966).
[128] Sur l'interpretation de l'operateur hamiltonien Hop et de l'operateur "carre du
moment angulaire" 11.{;p de la mecanique quantique, C. R. Acad. Sci. A 263,
645 (1966) [in collaboration with J. Andrade e Silva].
 Opuscules and Comprehensive Works 291

[129] Sur la formule Q = Qo J1 - (32 et les bases de la mecanique ondulatoire,

C. R. Acad. Sci. B 264, 1041 (1967).
[130] Le mouvement brownien d'une particule dans son onde, C. R. Acad. Sci. B
264, 1041 (1967).
[131] Sur la dynamique des corps It masse propre variable et la formule de transfor-
mation relativiste de la chaleur, C. R. Acad. Sci. B 264, 1173 (1967).
[132] La dynamique du guidage dans un milieu refringent et dispersif et la theorie
des antiparticules, J. Phys. (Paris) 28, 481 (1967).
[133] Sur l'equation .6. W = .6.Q + .6.£ en thermodynamique relativiste, C. R. Acad.
Sci. B 265, 437 (1967).
[134] ~r les discussions relatives It la formule Q = QoJl - (32 et la definition de la
pression en thermodynamique relativiste, C. R. Acad. Sci. B 265, 889 (1967).
[135] Sur l'application de la mecanique ondulatoire It la theorie des guides d'ondes,
C. R. Acad. Sci. B 266, 1233 (1968).
[136] La thermodynamique relativiste et la thermodynamique cachee des particules,
Int. J. Theor. Phys. 1, 1-23 (1968).
[137] Thermodynamique relativiste et mecanique ondulatoire, Ann. Inst. H. Poincare
IX (2), 89-108 (1968).
[138] Interpretation of a recent experiment on interferences of photons beams, Phys.
Rev. 172 (5), 1284 (1968).
[139] Sur l'interpretation des relations d'incertitude, C. R. Acad. Sci. B 268, 277
(1969).
[140] Sur Ie choc des particules en mecanique ondulatoire, C. R. Acad. Sci. B 268,
1449 (1969) [in collaboration with J. Andrade e Silva].
[141] Sur une nouvelle presentation des formules de la mecanique ondulatoire, C. R.
Acad. Sci. B 271, 559 (1970).
[142] Spins et moments de quantite de mouvement, C. R. Acad. Sci. B 272, 349
(1971).
[143] Sur un probleme du mouvement d'une particule dans un milieu refringent,
C. R. Acad. Sci. B 272, 1333 (1971).
[144] Masse du photon. Effet Imbert et effet Goos-Hiinchen en lumiere incidente po-
larisee, C. R. Acad. Sci. B 273, 1069 (1972) [in collaboration with J.-P. Vigier].
[145] Sur la repartition des potentiels d'interaction entre les particules d'un systeme,
C. R. Acad. Sci. B 275,899 (1972).
[146] Sur les veritables idees de base de la mecanique ondulatoire, C. R. Acad. Sci.
B 277, 71 (1973).
[147] Sur la refutation du thCoreme de Bell, C. R. Acad. Sci. B 278, 721 (1974).
[148] L'invariance adiabatique et la thermodynamique cachCe des particules, Ann.
Fond. L. de Broglie 1 (1), 1 (1976).

II. OPUSCULES AND COMPREHENSIVE WORKS

[1] Recherches sur la theorie des quanta, doctoral thesis defended in Paris on
November 25, 1924; Ann. Phys. (Paris) (10) III, 22-128 (1925); German
translation (Akademische Verlagsgesellschaft, Leipzig, 1927).
292 Works of Louis de Broglie

[2] Ondes et mouvements (Collection de Physique mathematique, fascicule I)

(Gauthier-Villars, Paris, 1926).
[3] Introduction a la physique des rayons X et I (Gauthier-Villars, Paris, 1928) [in
collaboration with M. de Broglie]; German translation (J. A. Barth, Leipzig,
1930).
[4] La mecanique ondulatoire (Memorial des Sciences physiques, fascicule I)
( Gauthier-Villars, Paris, 1928).
[5] Selected Papers on Wave Mecanics (Blackie & Son, Glasgow, 1928) [in collab-
oration with Loon Brillouin].
[6] On des et corpuscules (Hermann, Paris, 1930).
a
[7] Introduction l'etude de la mecanique ondulatoire (Hermann, Paris, 1930); En-
glish translation (Methuen, London); German translation (Akademische Ver-
lagsgesellschaft, Leipzig).
[8] Theorie de la quantification dans la nouvelle mecanique (Hermann, Paris,
1932).
[9] Sur une forme plus restrictive des relations d'incertitude (Collection des Ex-
poses de physique theorique, fascicule I) (Hermann, Paris, 1932).
[10] Le passage des corpuscules electrises a travers les barrie res de potentiel, Ann.
Inst. H. Poincare III, 339-446 (1932-33).
[11] L'electron magnetique (theorie de Dirac) (Hermann, Paris, 1934).
[12] Une nouvelle conception de la lumiere (Exposes de physique theorique, fascicule
XIII) (Hermann, Paris, 1934).
[13] Nouvelles recherches sur la lumiere (Exposes de physique theorique, fascicule
XX) (Hermann, Paris, 1936).
[14] Le principe de correspondance et les interactions entre la matiere et le rayon-
nement (Collection des Exposes de physique theorique) (Hermann, Paris, 1938),
p.704.
[15] La mecanique ondulatoire des systemes de corpuscules (Collection de Physique
mathematique, fascicule V) (Gauthier-Villars, Paris, 1939).
[16] Une nouvelle theorie de la lumiere, la mecanique ondulatoire du photon (Her-
mann, Paris, 1940), Vol. I: La lumiere dans Ie vide (Hermann, Paris, 1942);
Vol. II: L'interaction entre les photons et la matiere.
[17] Problemes de propagation guidee des ondes electromagnetiques (Gauthier-
Villars, Paris, 1941)
[18] Theorie generale des particules a spin (Gauthier-Villars, Paris, 1943).
a
[19] De la mecanique ondulatoire la theorie du noyau, Vol. I (Hermann, Paris,
1943); Vol. II (Hermann, Paris, 1954); Vol. III (Hermann, Paris, 1946).
[20] Corpuscules, ondes et mecanique ondulatoire (conferences a l'Ecole superieure
d'Electricite) (Centre de documentation universitaire, Paris, 1943; Albin-
Michel, Collection Sciences d 'a ujo urd 'hui, Paris, 1945); Spanish and Italian
translations.
[21] Mecanique ondulatoire du photon et theorie quantique des champs (Gauthier-
Villars, Paris, 1949).
[22] Optique ondulatoire et corpusculaire (Hermann, Paris, 1950).
[23] La Theone des particules de spin 1/2 (Electrons de Dirac) (Gauthier-Villars,
Paris, 1951).
 Works on the Philosophy of Science 293
[24] Elements de theorie des quanta et de mecanique ondulatoire (Gauthier-Villars,
Paris, 1953).
[25] La physique quantique restera-t-elle indeterministe? (Gauthier-Villars, Paris,
1953) [in collaboration with J.-P. Vigier].
[26] Une tentative d'interpretation causale et non lineaire de la mecanique ondula-
toire: la theorie de la double solution (Gauthier-Villars, Paris, 1956); English
translation (Elsevier, Amsterdam, 1960).
[27] La theorie de la mesure en mecanique ondulatoire (interpretation usuelle et
interpretation causale) (Gauthier-Villars, Paris, 1957).
[28] La nouvelle theorie des particules de MM. Jean-Pierre Vigier et de ses collabo-
rateurs (Gauthier-Villars, Paris, 1961); English translation (Elsevier, Amster-
dam).
[29] Etude critique des bases de l'interpretation actuelle de la mecanique ondulatoire
(Gauthier-Villars, Paris, 1963).
[30] La thermodynamique de la particule isolee (thermodynamique cachee des par-
ticules) ( Gauthier-Villars, Paris, 1964).
[31] Ondes electromagnetiques et photons ( Gauthier-Villars, Paris, 1968).
[32] La reinterpretation de la mecanique ondulatoire. 1re partie: Principes generaux
(Gauthier-Villars, Paris, 1971).
[33] Jalons pour une nouvelle microphysique (Gauthier-Villars, Paris, 1978).

III. WORKS ON THE PHILOSOPHY OF SCIENCE

[1] La physique nouvelle et les quanta (BibliotMque de philosophie scientifique,

edited by Paul Gaultier) (Flammarion, Paris, 1937); Italian translation.
[2] Matiere et lumiere (Collection Sciences d'aujourd'hui, edited by Andre George)
(Albin-Michel, Paris, 1937); English, American, German, Italian, Japanese,
Spanish, and Dutch translations.
[3] Continu et discontinu en physique moderne (Collection Sciences d'aujourd'hui,
edited by Andre George) (Albin-Michel, Paris, 1941); German, Dutch, and
Italian translations.
[4] Physique et microphysique (Collection Sciences d'aujourd'hui, edited by Andre
George) (Albin-Michel, Paris, 1947); Italian, Spanish, and German translations.
[5] Savants et decouvertes (Collection Les savants et Ie monde, edited by Andre
George) (Albin-Michel, Paris, 1951); Spanish translation.
[6] Perspectives nouvelles en microphysique (Collection Sciences d 'a ujo urd 'hui,
edited by Andre George) (Albin-Michel, Paris, 1956); English translation (Basic
Books, New York, 1962).
[7] Sur les sentiers de la science (Albin-Michel, Paris, 1960); Italian translation
(Paolo Boringhieri, Turin, 1962).
[8] Certitudes et incertitudes de la science (Albin-Michel, Paris, 1966).
[9] Recherches d'un demi-siecle (Albin-Michel, Paris, 1976).
294 Works of Louis de Broglie

IV. NOTES AND ACADEMIC SPEECHES

[1] La vie et l'reuvre d'Emile Picard, read in public, December 21, 1942.
[2] La vie et l'reuvre d' Andre Blondel, read in public, December 18, 1944.
[3] Discours de reception a l'Academie franc;aise, acceptance speech on becoming
a member of the Academie franc;aise, delivered under the Dome, May 31, 1945;
de luxe edition (Albin-Michel, Paris, 1945).
[4] Le realite des molecules et l'reuvre de Jean Perrin, read in public, Decem-
ber 17, 1945.
[5] Rapports sur les prix de vertus, read in public, January 10, 1946.
[6J La vie et l'reuvre de Charles Fabry, read in public, December 16, 1946.
[7] La vie et l'reuvre de Paul Langevin, read in public, December 15, 1947.
[8] La physique contemporaine et l'reuvre d' Albert Einstein, read in public, De-
cember 19, 1949.
[9] La vie et l'reuvre de Hendrik Antoon Lorentz, read in public, December 10, 1951.
[10] La vie et l'reuvre de Aime Cotton, read in public, December 14, 1953.
[11] Le dualisme des ondes et des corpuscules dans l'reuvre d' Albert Einstein, read
in public, December 5, 1955.
[12] Notice sur la vie et l'reuvre d'Emile Borel, read in public, December 9, 1957.
[13] Notice sur la vie et l'reuvre de Frederic Joliot, read in public, December 14, 1960.
[14] Notice sur la vie et l'reuvre de Georges Darmois, read in public, Decem-
ber 9, 1962.
[15] Notice sur la vie et l'reuvre de Jean Becquerel, read in public, December 9, 1964.
[16] Notice sur la vie et l'reuvre de Camille Gutton, read in public, Decem-
ber 11,1965.
[17] Notice sur la vie et l'reuvre d'Albert Perard, read in public, December 11, 1967.
[18] Notice sur la vie et l'reuvre de Bernard Lyot, read in public, December 8, 1969.
[19] Notice sur la vie et l'reuvre d'Andre Danjon, read in public, December 13, 1971.

V. LECTURES AND GENERAL ARTICLES

[1] La theorie des quanta, synthese de la dynamique et de l'optique, Revue Gen.

Sci., 35th year, No. 22, 629 (1925).
[2] Deux conceptions ad verses de la lumiere et leur synthese possible, Scientia,
September 1926, p. 128.
[3] La physique moderne et l'reuvre de Fresnel, Revue Metaphys. Morale XXXIV
(4),421 (1927).
[4] L'reuvre de Fresnel et l'evolution actuelle de la physique (lecture given on the
occasion of the Fresnel centenary, October 29, 1927), Revue Opt. Theor. Exp.
6, 493 (1927).
[5] Continuite et individualite dans la physique moderne, Cahier de la nouvelle
journee, No. 15, p. 60.
 Lectures and General Articles 295

[6] La crise recente de l'optique ondulatoire (lecture given at the Conservatoire des
Arts et Metiers, April 17, 1929), Revue Sci., 67th year, No. 12, 353 (1929).
[7] Determinisme et causalite dans la physique contemporaine, Revue Metaphys.
Morale XXXVII (4),433 (1929).
[8] Sur la nature ondulatoire de l'electron (lecture given in Stockholm on accepting
the Nobel prize, December 11, 1929), Revue Sci., 68th year, No.1, 1 (1930).
[9] Ondes et corpuscules dans la physique moderne (lecture given at the Con-
servatoire des Arts et Metiers under the chairmanship of Paul Painleve, Jan-
uary 26, 1930), Revue Gen. Sci. XLI (4),101 (1930).
[10) La representation simultanee des possibilites dans la nouvelle physique, Revue
Metaphys. Morale XXXIV (2), 141 (1932).
[11] L'etat actuel de la theorie electromagnetique, report to the International
Congress on Electricity, Paris, 1932.
[12] Relativite et quanta, Revue Metaphys. Morale XL, 269~272 (1929).
[13] Quelques considerations sur les notions d'ondes et de corpuscule, Scientia 40,
177 (1934) [in collaboration with M. de Broglie].
[14] Les idees nouvelles introduites par la mecanique quantique, L 'enseignement
Math., 32nd year, 137 (1932).
[15) Sur la representation des phenomenes dans la nouvelle physique, Revue Univ.
Bruxelles, No.3, 277 (1934).
[16] Voies anciennes et perspectives nouvelles en theorie de al lumiere, Revue
Metaphys. Morale XLI, 445 (1934).
[17] Rcalite et idealisation, Revue de Synthese VIII, 125 (1934).
[18] Les progres de la physique contemporaine, Revue Frant;aise de Prague, No. 68,
93 (1935).
[19] Coup d'reil sur l'histoire de l'optique, Thales, 1st year, 3 (1934).
[20] Reflexions sur les deux sortes d'eIectricite, Revue Metaphys. Morale XLIII,
173~185 (1936).
[21] L'evolution de l'eIectron, in Livre du Centenaire d'Ampere (Lyon, March 1936).
[22] Un exemple des syntheses successives de la physique: les theories de la lumiere,
Thales, 2nd year, 9~22 (1935); reproduced in Revue des questions scientifiques,
May 20,1937, pp. 361~381.
[23] Individualite et interaction dans Ie monde physique, Revue Metaphys. Morale
XLIX, 353~368 (1937).
[24] L'invention dans les sciences theoriques, Sciences, 2nd year, No. 14, June 1937.
[25] Etat actuel de nos connaissances sur la structure de l'electricite (communication
at the Congres Galvani de Bologne, read October 19, 1937), Nuovo Cimento
XIV (9), November 1937.
[26] Physique ponctuelle et physique du champ, Revue Metaphys. Morale L, 325~
338 (1938).
[27] La theorie quantique du rayonnement, Revue Metaphys. Morale LI, 199~210
(1939).
[28] Recents progres dans la theorie des photons et autres particules, Revue Meta-
phys. Morale LII, 1~16 (1940).
[29] L'avenir de la physique dans l'avenir de la science (Collection Presences) (PIon,
Paris, 1941).
296 Works of Louis de Broglie

[30] L 'ceuvre d' Andre Blondel en physique generale, in Commemoration de l'reuvre

d'Andre-Eugene Blondel (Gauthier-Villars, Paris, 1942), pp. 7-19.
[31] Les particules Clementaires de la Matiere et les nouvelles theories du noyau de
l'atome, L'Astronomie, 56th year, 1-6, 27-34, 51-57 (1942).
[32] Souvenirs personnels sur les debuts de la mecanique ondulatoire, Revue Meta-
phys. Morale LIlI, 1-23 (1941).
[33] L'essor de la physique en France de 1815 a 1825 (Collection Hier et Demain,
No. IV, 1943), pp. 80-108
[34] Recherche scientifique et recherche technique, Revue generale du Caoutchouc
20, 45-49 (1943).
[35] Les conceptions de la physique contemporaine et les idees de Bergson sur Ie
temps et sur Ie mouvement, Revue Metaphys. Morale, October 1941, p. 261.
[36] Sur les notions de lois rigoureuses et de lois statistiques, Revue de l'Economie
contemporaine, 3rd year, No. 25, 1-3 (1944).
[37] Grandeur et valeur morale de la science N. R. F., March 1945, p. 41.
[38] Hasard et contingence en physique quantique, Revue Metaphys. Morale LV,
241-252 (1945).
[39] L'activite du Centre de physique theorique de l'Institut Henri Poincare pendant
les dernieres annees, Experientia II (1) (1946).
[40] La lumiere dans Ie monde physique, Cahiers du monde nouveau, 2nd year,
No.3, March 1946.
[41] Le microscope Clectronique et la dualite des ondes et des corpuscules, Revue
Metaphys. Morale LVII, 1 (1947).
[42] Le role des mathematiques dans Ie developpement de la physique theorique
contemporaine, in Les grands courants de la pensee mathematique (Cahiers du
Sud, 1948).
[43] Sur la complCmentarite des idees d'individu et de systeme, Dialectica, 1948,
p.325.
[44] La statistique des cas purs en mecanique ondulatoire et l'interference des prob-
abilites, Revue Scientifique, 87th year, 259-264 (1949).
[45] L'enseignement de la physique, Revue de Metrologie pratique et ligale, 27th
year, 5 (1949).
[46] Sur la relation d'incertitude de la seconde quantification Revue Int. Philos.,
No.8, April 1949.
[47] L'espace et Ie temps dans la physique quantique, Revue Metaphys. Morale
LVIII, 119-125 (1949).
[48] L'avenir influe-t-il sur Ie present?, L'orientation medicale 16 (2),11 (1950).
[49] Un nouveau venu en physique: Ie champ nucleaire, Revue Metaphys. Morale,
56th year, 117 (1951).
[50] Sens philosophique et portee pratique de la cybernetique (lecture given at the
Conservatoire des Arts et Metiers, October 5, 1951).
[51] La physique quantique restera-t-elle indeterministe? (lecture given at the Cen-
tre de Synthese, October 31, 1952), Revue Hist. Sci. V, 289 (1952).
[52] Les particules de la microphysique (lecture given at the Palais de la Decouverte,
December 20, 1952), published by the Palais de la Decouverte in Ser. A, No. 174.
 Lectures and General Articles 297
[53] Une interpretation nouvelle de la mecanique ondulatoire est-elle possible?
Nuovo Gimento, 1.37.50, January 1, 1955; published also by the Palais de la
Decouverte in Ser. A, No. 201 (lecture given October 16, 1954).
[54] La lumiere, les quanta et la technique de l'eclairage (lecture given at the Palais
de la Decouverte, January 16, 1960), published by the Palais de la Decouverte
in Ser. A, No. 255.
[55] La coexistence des photons et des ondes dans les rayonnements eIectromagneti-
ques et la theorie de la double solution, Energie nucleaire 7 (3) (1965).
[56] La reinterpretation de la mecanique ondulatoire, Phys. Bull. 19, 133 (1968).
[57] Quelques vues personnelles sur l'evolution de la physique theorique, Chronicle
of contemporary philosophy, La Nuova [talia, Florence, 1968, pp. 217-222.
[58] The reinterpretation of wave mechanics, Found. Phys. 1 (1),5 (1970).
[59] Vue d'ensemble sur l'histoire et l'interpretation de la mecanique ondulatoire,
Revue Jrant;aise de l'Electricite, 43rd year, No. 228, 13 (1970).
[60] Ondes electromagnetiqes et photons en radioelectricite, On de electrique, Sep-
tember 1970, p. 657.
[61] Waves and particle, Phys. Bull. 22 (1971).
[62] L'interpretation de la mecanique ondulatoire par la theorie de la double solu-
tion, in Foundations oj Quantum Mechanics (Course IL, International School
of Physics "Enrico Fermi," Varenna, July 1970) (Academic, New York, 1971).
[63] Discours prononce a la premiere seance de la Foundation Louis de Broglie, Ann.
Fond. L. de Broglie, special issue, 1975.
[64] Refiexions sur la physique contemporaine, Ann. Fond. L. de Broglie 1 (2), 49
(1976).
[65] Treize remarques sur divers sujets de physique thoorique, Ann. Fond. L.
de Broglie 1 (3), 116 (1976).
[66] Refiexions sur la causaliM, Ann. Fond. L. de Broglie 2 (2), 69 (1977).
[67] Toute description complete de la realite implique l'intervention de la causalite,
Ann. Fond. L. de Broglie 2 (2), 133 (1977).
[68] Necessite de la liberte dans la recherche scientifique, Ann. Fond. L. de Broglie
4 (1), 62 (1979).
INDEX
adjoint
matrix 46
operator 85
aleatory (random) variable 50, 153
Amaldi 173
Andrade e Silva 30, 179
angular momentum operators 35, 53,
100
anticommutator 46,48
anti-Hermitian matrix 46
Arnous, E. 180, 190, 194, 195
ascertainment 238
basis function 48
Bass 200, 201, 209
function of 201
Batho, H. F. 16, 173
Bauer, E. 220, 246, 248, 259
Bayes's formula 272
Biderman, L. 174
Bodiou, G. 218
Bohm 143, 177, 178
Bohr 136, 142, 144
Boltzmann-Gibbs canonical law
263
Born 21
Borsch H. 17, 173
Bragg 80
Brillouin 265
Brownian motion 233
canonical transformation 47
Cauchy's law 155, 158
causality 235
characteristic function 155, 156,
160-162, 180
collective 222, 238
commutator 46,48
complementarity 79, 144, 148
complete description 249
compound probabilities 131, 207
Compton effect 71, 144
conditional probabilities 162
conditional velocity 210
configuration space 178
consciousness 104, 246, 275
constant of motion 52
contingency 232
continuity equation 18, 33, 172
correlated systems 136
correlation 137
coefficient 161, 187, 243
correspondence principle 144
Costa de Beauregard, O. 269, 280, 281
damping coefficient 117
Darwin particle 107, 184, 190
Davisson, C. J. 17,173
Dedebant 209
Dempster, A. J. 16, 173
determinism 125, 235, 236
deviation 154
Dirac 283
function 40, 164
261, statistical matrix 228
theory 170
dispersion 50, 86, 155, 161
equation 163
theorem 86
dispersionless states 234
distribution
function 153
most probable 261
Doppler effect 71
double solution (theory) 174, 176,
208-210, 213, 249
double-slit experiment 81-84
Ehrenfest theorem 23
eigen-differential 38
eigenfunction 36
eigenvalue 36, 37
300 Index

Einstein 136, 143, 249 hydrodynamic interpretation of wave

Einstein, Podolsky, and Rosen (EPR) mechanics 18,32, 174, 176, 207
136, 137
electron diffraction 80
idempotent 225
entropy, statistical 259-261
indeterminism 174, 238
Estermann, I. 173
evolution 263 interacting systems
irreversible 263 correlation coefficients for 243
parameter 274 statistics of 239
reversible 263 interference
of probabilities 130, 198, 200
principle 16,17,173,194
Fabrikant, V. 174
inverse matrix 47
Faget, J. 173
Fermat, principle of 9
first integral 52 Jacobi equation 4
Fock 269, 283 Jonsson, C. 173
Fourier
coefficients 39
Kcenig's theorems 196
inversion formula 160
Franck-Hertz experiments 144 Kikuchi 16
Furry 136
Landau 117
Gauss's law for two variables 164 Laplace-Gauss law 157
geometrical optics, equation of 9 Laue 80
Germer, L. H. 17, 173 Lennuyer 105
Goodman, P. 173 line width 117
granular picture 81 localization 135
Green functions 128, 270 London, F. 220, 246, 248, 259
group velocity 25
guidance 210
Madelung's probability fluid
formula 219
velocity of 18, 32
velocity 219
continuity equation of 18, 33
Mandelstam-Tamm relation 122
Hamilton's action integral 273 marginal
Hamiltonian 171 distribution law 162, 187
Heisenberg probability 162
fourth uncertainty relation 73, 103,
matrix
105, 108, 147
Heisenberg 49
matrix 49
Schrodinger 49
microscope 69
wave-mechanical 48
uncertainties (relations) 19, 21, 85,
87, 95, 98 Maupertuis's least action principle 6
Hermitian measurement
matrix 46 in wave mechanics 238
operator 35, 85 maximal 65
hidden variables 124, 133, 169, 233, of electron speed 71
236 simultaneous 63, 68
Hilbert space 182, 224 mean value 49, 154
Huygens principle 73 mixture 222, 238
 Index 301

moments 154 Schrodinger matrix: 49

monochromatic solutions 172 Schwartz, L. 164
multitemporal theory 283 Schwinger 283
Shannon 265
non commuting quantities 77 Sowchkine, N. 174
norm of vector 224 spatial separation 150
normalized function 37, 172 spectral decompositions 21, 42, 176
spectrum
objective reality 249 eignenvalue 37
operator continuous 37
complete 36 discrete 37
in Hilbert space 225 Stark effect 107
linear 35 statistical matrix:
orthogonal matrix: 47 Dirac 228
orthonormal function 37 elementary 226
in thermodynamics 259
Peierls 117 statistical weights 222
perturbation theory 276 Stern, O. 173
phases, effacement of 125, 127, 198 Stern-Gerlach measurement 249, 254
pilot wave (theory) 173-179,208,218, stochastic mechanics 209
236,237,273,274 subjective interpretation 249
Podolski 283 sum over states 261
Poisson law 157 supermultitemporal theory 283
Ponte 17
superposition principle 44
position probability density 17
preparation 249
probability Tamm 121
density 154, 159 Taylor, G. J. 16, 173
fluid 17, 18 Thomson, G. P. 17
of causes 269 time
projector 225 flow 265
propagation vector 40
role in wave mechanics 268
pure case 222, 238
Tomonaga 283
trace 47
quantification principle 42 transpose matrix: 46
quantum potential 178 transition probability 115, 144
trial 131, 153, 200, 238
random (aleatory) variable 50, 153
Rayleigh formula 26
uncertainty
reality, physical 138
anticipated 102
reduction of wave packet 124, 197
of a quantity 95
regression curve 163, 187
relativistic covariance 283 presently existing 102
repeatability of measurements 64 sharp-edged 95
retrodiction 269, 270 undulatory picture 81
Rupp 17 unitary matrix: 47

Schrodinger 136 variation of constants 110, 111, 277

302 Index

Vigier 178 VVehrIe 209

virtual VVentzel's formula 116
photon exchange 117 VViener's experiment 176
state 110, 116 VVigner-Bass
transitions 110 characteristic function 202
von Neumann (theorem) 169,220 distribution law 201,202,207
von Traubenberg, R. 119
Young's double-slit experiment 81-84,
VVatanabe 259, 266 176
Yvon's theorem 213
wave equation 13
wave-packet reduction 124
wave-number vector 75 Zeeman effect 250